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)
93 page = brd_lookup_page(brd, sector);
98 * Must use NOIO because we don't want to recurse back into the
99 * block or filesystem layers from page reclaim.
101 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM | __GFP_ZERO);
105 if (radix_tree_preload(GFP_NOIO)) {
110 spin_lock(&brd->brd_lock);
111 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);
119 spin_unlock(&brd->brd_lock);
121 radix_tree_preload_end();
127 * Free all backing store pages and radix tree. This must only be called when
128 * there are no other users of the device.
130 #define FREE_BATCH 16
131 static void brd_free_pages(struct brd_device *brd)
133 unsigned long pos = 0;
134 struct page *pages[FREE_BATCH];
140 nr_pages = radix_tree_gang_lookup(&brd->brd_pages,
141 (void **)pages, pos, FREE_BATCH);
143 for (i = 0; i < nr_pages; i++) {
146 BUG_ON(pages[i]->index < pos);
147 pos = pages[i]->index;
148 ret = radix_tree_delete(&brd->brd_pages, pos);
149 BUG_ON(!ret || ret != pages[i]);
150 __free_page(pages[i]);
156 * This assumes radix_tree_gang_lookup always returns as
157 * many pages as possible. If the radix-tree code changes,
158 * so will this have to.
160 } while (nr_pages == FREE_BATCH);
164 * copy_to_brd_setup must be called before copy_to_brd. It may sleep.
166 static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n)
168 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
171 copy = min_t(size_t, n, PAGE_SIZE - offset);
172 if (!brd_insert_page(brd, sector))
175 sector += copy >> SECTOR_SHIFT;
176 if (!brd_insert_page(brd, sector))
183 * Copy n bytes from src to the brd starting at sector. Does not sleep.
185 static void copy_to_brd(struct brd_device *brd, const void *src,
186 sector_t sector, size_t n)
190 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
193 copy = min_t(size_t, n, PAGE_SIZE - offset);
194 page = brd_lookup_page(brd, sector);
197 dst = kmap_atomic(page, KM_USER1);
198 memcpy(dst + offset, src, copy);
199 kunmap_atomic(dst, KM_USER1);
203 sector += copy >> SECTOR_SHIFT;
205 page = brd_lookup_page(brd, sector);
208 dst = kmap_atomic(page, KM_USER1);
209 memcpy(dst, src, copy);
210 kunmap_atomic(dst, KM_USER1);
215 * Copy n bytes to dst from the brd starting at sector. Does not sleep.
217 static void copy_from_brd(void *dst, struct brd_device *brd,
218 sector_t sector, size_t n)
222 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
225 copy = min_t(size_t, n, PAGE_SIZE - offset);
226 page = brd_lookup_page(brd, sector);
228 src = kmap_atomic(page, KM_USER1);
229 memcpy(dst, src + offset, copy);
230 kunmap_atomic(src, KM_USER1);
232 memset(dst, 0, copy);
236 sector += copy >> SECTOR_SHIFT;
238 page = brd_lookup_page(brd, sector);
240 src = kmap_atomic(page, KM_USER1);
241 memcpy(dst, src, copy);
242 kunmap_atomic(src, KM_USER1);
244 memset(dst, 0, copy);
249 * Process a single bvec of a bio.
251 static int brd_do_bvec(struct brd_device *brd, struct page *page,
252 unsigned int len, unsigned int off, int rw,
259 err = copy_to_brd_setup(brd, sector, len);
264 mem = kmap_atomic(page, KM_USER0);
266 copy_from_brd(mem + off, brd, sector, len);
267 flush_dcache_page(page);
269 copy_to_brd(brd, mem + off, sector, len);
270 kunmap_atomic(mem, KM_USER0);
276 static int brd_make_request(struct request_queue *q, struct bio *bio)
278 struct block_device *bdev = bio->bi_bdev;
279 struct brd_device *brd = bdev->bd_disk->private_data;
281 struct bio_vec *bvec;
286 sector = bio->bi_sector;
287 if (sector + (bio->bi_size >> SECTOR_SHIFT) >
288 get_capacity(bdev->bd_disk))
295 bio_for_each_segment(bvec, bio, i) {
296 unsigned int len = bvec->bv_len;
297 err = brd_do_bvec(brd, bvec->bv_page, len,
298 bvec->bv_offset, rw, sector);
301 sector += len >> SECTOR_SHIFT;
310 static int brd_ioctl(struct inode *inode, struct file *file,
311 unsigned int cmd, unsigned long arg)
314 struct block_device *bdev = inode->i_bdev;
315 struct brd_device *brd = bdev->bd_disk->private_data;
317 if (cmd != BLKFLSBUF)
321 * ram device BLKFLSBUF has special semantics, we want to actually
322 * release and destroy the ramdisk data.
324 mutex_lock(&bdev->bd_mutex);
326 if (bdev->bd_openers <= 1) {
328 * Invalidate the cache first, so it isn't written
329 * back to the device.
331 * Another thread might instantiate more buffercache here,
332 * but there is not much we can do to close that race.
334 invalidate_bh_lrus();
335 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
339 mutex_unlock(&bdev->bd_mutex);
344 static struct block_device_operations brd_fops = {
345 .owner = THIS_MODULE,
350 * And now the modules code and kernel interface.
353 int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
354 module_param(rd_nr, int, 0);
355 MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
356 module_param(rd_size, int, 0);
357 MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
358 MODULE_LICENSE("GPL");
359 MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
362 /* Legacy boot options - nonmodular */
363 static int __init ramdisk_size(char *str)
365 rd_size = simple_strtol(str, NULL, 0);
368 static int __init ramdisk_size2(char *str)
370 return ramdisk_size(str);
372 __setup("ramdisk=", ramdisk_size);
373 __setup("ramdisk_size=", ramdisk_size2);
377 * The device scheme is derived from loop.c. Keep them in synch where possible
378 * (should share code eventually).
380 static LIST_HEAD(brd_devices);
381 static DEFINE_MUTEX(brd_devices_mutex);
383 static struct brd_device *brd_alloc(int i)
385 struct brd_device *brd;
386 struct gendisk *disk;
388 brd = kzalloc(sizeof(*brd), GFP_KERNEL);
392 spin_lock_init(&brd->brd_lock);
393 INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC);
395 brd->brd_queue = blk_alloc_queue(GFP_KERNEL);
398 blk_queue_make_request(brd->brd_queue, brd_make_request);
399 blk_queue_max_sectors(brd->brd_queue, 1024);
400 blk_queue_bounce_limit(brd->brd_queue, BLK_BOUNCE_ANY);
402 disk = brd->brd_disk = alloc_disk(1);
405 disk->major = RAMDISK_MAJOR;
406 disk->first_minor = i;
407 disk->fops = &brd_fops;
408 disk->private_data = brd;
409 disk->queue = brd->brd_queue;
410 sprintf(disk->disk_name, "ram%d", i);
411 set_capacity(disk, rd_size * 2);
416 blk_cleanup_queue(brd->brd_queue);
423 static void brd_free(struct brd_device *brd)
425 put_disk(brd->brd_disk);
426 blk_cleanup_queue(brd->brd_queue);
431 static struct brd_device *brd_init_one(int i)
433 struct brd_device *brd;
435 list_for_each_entry(brd, &brd_devices, brd_list) {
436 if (brd->brd_number == i)
442 add_disk(brd->brd_disk);
443 list_add_tail(&brd->brd_list, &brd_devices);
449 static void brd_del_one(struct brd_device *brd)
451 list_del(&brd->brd_list);
452 del_gendisk(brd->brd_disk);
456 static struct kobject *brd_probe(dev_t dev, int *part, void *data)
458 struct brd_device *brd;
459 struct kobject *kobj;
461 mutex_lock(&brd_devices_mutex);
462 brd = brd_init_one(dev & MINORMASK);
463 kobj = brd ? get_disk(brd->brd_disk) : ERR_PTR(-ENOMEM);
464 mutex_unlock(&brd_devices_mutex);
470 static int __init brd_init(void)
474 struct brd_device *brd, *next;
477 * brd module now has a feature to instantiate underlying device
478 * structure on-demand, provided that there is an access dev node.
479 * However, this will not work well with user space tool that doesn't
480 * know about such "feature". In order to not break any existing
481 * tool, we do the following:
483 * (1) if rd_nr is specified, create that many upfront, and this
484 * also becomes a hard limit.
485 * (2) if rd_nr is not specified, create 1 rd device on module
486 * load, user can further extend brd device by create dev node
487 * themselves and have kernel automatically instantiate actual
490 if (rd_nr > 1UL << MINORBITS)
497 nr = CONFIG_BLK_DEV_RAM_COUNT;
498 range = 1UL << MINORBITS;
501 if (register_blkdev(RAMDISK_MAJOR, "ramdisk"))
504 for (i = 0; i < nr; i++) {
508 list_add_tail(&brd->brd_list, &brd_devices);
511 /* point of no return */
513 list_for_each_entry(brd, &brd_devices, brd_list)
514 add_disk(brd->brd_disk);
516 blk_register_region(MKDEV(RAMDISK_MAJOR, 0), range,
517 THIS_MODULE, brd_probe, NULL, NULL);
519 printk(KERN_INFO "brd: module loaded\n");
523 list_for_each_entry_safe(brd, next, &brd_devices, brd_list) {
524 list_del(&brd->brd_list);
528 unregister_blkdev(RAMDISK_MAJOR, "brd");
532 static void __exit brd_exit(void)
535 struct brd_device *brd, *next;
537 range = rd_nr ? rd_nr : 1UL << MINORBITS;
539 list_for_each_entry_safe(brd, next, &brd_devices, brd_list)
542 blk_unregister_region(MKDEV(RAMDISK_MAJOR, 0), range);
543 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
546 module_init(brd_init);
547 module_exit(brd_exit);