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/mutex.h>
19 #include <linux/radix-tree.h>
20 #include <linux/buffer_head.h> /* invalidate_bh_lrus() */
21 #include <linux/slab.h>
23 #include <asm/uaccess.h>
25 #define SECTOR_SHIFT 9
26 #define PAGE_SECTORS_SHIFT (PAGE_SHIFT - SECTOR_SHIFT)
27 #define PAGE_SECTORS (1 << PAGE_SECTORS_SHIFT)
30 * Each block ramdisk device has a radix_tree brd_pages of pages that stores
31 * the pages containing the block device's contents. A brd page's ->index is
32 * its offset in PAGE_SIZE units. This is similar to, but in no way connected
33 * with, the kernel's pagecache or buffer cache (which sit above our block
41 unsigned brd_blocksize;
43 struct request_queue *brd_queue;
44 struct gendisk *brd_disk;
45 struct list_head brd_list;
48 * Backing store of pages and lock to protect it. This is the contents
49 * of the block device.
52 struct radix_tree_root brd_pages;
56 * Look up and return a brd's page for a given sector.
58 static DEFINE_MUTEX(brd_mutex);
59 static struct page *brd_lookup_page(struct brd_device *brd, sector_t sector)
65 * The page lifetime is protected by the fact that we have opened the
66 * device node -- brd pages will never be deleted under us, so we
67 * don't need any further locking or refcounting.
69 * This is strictly true for the radix-tree nodes as well (ie. we
70 * don't actually need the rcu_read_lock()), however that is not a
71 * documented feature of the radix-tree API so it is better to be
72 * safe here (we don't have total exclusion from radix tree updates
73 * here, only deletes).
76 idx = sector >> PAGE_SECTORS_SHIFT; /* sector to page index */
77 page = radix_tree_lookup(&brd->brd_pages, idx);
80 BUG_ON(page && page->index != idx);
86 * Look up and return a brd's page for a given sector.
87 * If one does not exist, allocate an empty page, and insert that. Then
90 static struct page *brd_insert_page(struct brd_device *brd, sector_t sector)
96 page = brd_lookup_page(brd, sector);
101 * Must use NOIO because we don't want to recurse back into the
102 * block or filesystem layers from page reclaim.
104 * Cannot support XIP and highmem, because our ->direct_access
105 * routine for XIP must return memory that is always addressable.
106 * If XIP was reworked to use pfns and kmap throughout, this
107 * restriction might be able to be lifted.
109 gfp_flags = GFP_NOIO | __GFP_ZERO;
110 #ifndef CONFIG_BLK_DEV_XIP
111 gfp_flags |= __GFP_HIGHMEM;
113 page = alloc_page(gfp_flags);
117 if (radix_tree_preload(GFP_NOIO)) {
122 spin_lock(&brd->brd_lock);
123 idx = sector >> PAGE_SECTORS_SHIFT;
124 if (radix_tree_insert(&brd->brd_pages, idx, page)) {
126 page = radix_tree_lookup(&brd->brd_pages, idx);
128 BUG_ON(page->index != idx);
131 spin_unlock(&brd->brd_lock);
133 radix_tree_preload_end();
138 static void brd_free_page(struct brd_device *brd, sector_t sector)
143 spin_lock(&brd->brd_lock);
144 idx = sector >> PAGE_SECTORS_SHIFT;
145 page = radix_tree_delete(&brd->brd_pages, idx);
146 spin_unlock(&brd->brd_lock);
151 static void brd_zero_page(struct brd_device *brd, sector_t sector)
155 page = brd_lookup_page(brd, sector);
157 clear_highpage(page);
161 * Free all backing store pages and radix tree. This must only be called when
162 * there are no other users of the device.
164 #define FREE_BATCH 16
165 static void brd_free_pages(struct brd_device *brd)
167 unsigned long pos = 0;
168 struct page *pages[FREE_BATCH];
174 nr_pages = radix_tree_gang_lookup(&brd->brd_pages,
175 (void **)pages, pos, FREE_BATCH);
177 for (i = 0; i < nr_pages; i++) {
180 BUG_ON(pages[i]->index < pos);
181 pos = pages[i]->index;
182 ret = radix_tree_delete(&brd->brd_pages, pos);
183 BUG_ON(!ret || ret != pages[i]);
184 __free_page(pages[i]);
190 * This assumes radix_tree_gang_lookup always returns as
191 * many pages as possible. If the radix-tree code changes,
192 * so will this have to.
194 } while (nr_pages == FREE_BATCH);
198 * copy_to_brd_setup must be called before copy_to_brd. It may sleep.
200 static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n)
202 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
205 copy = min_t(size_t, n, PAGE_SIZE - offset);
206 if (!brd_insert_page(brd, sector))
209 sector += copy >> SECTOR_SHIFT;
210 if (!brd_insert_page(brd, sector))
216 static void discard_from_brd(struct brd_device *brd,
217 sector_t sector, size_t n)
219 while (n >= PAGE_SIZE) {
221 * Don't want to actually discard pages here because
222 * re-allocating the pages can result in writeback
223 * deadlocks under heavy load.
226 brd_free_page(brd, sector);
228 brd_zero_page(brd, sector);
229 sector += PAGE_SIZE >> SECTOR_SHIFT;
235 * Copy n bytes from src to the brd starting at sector. Does not sleep.
237 static void copy_to_brd(struct brd_device *brd, const void *src,
238 sector_t sector, size_t n)
242 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
245 copy = min_t(size_t, n, PAGE_SIZE - offset);
246 page = brd_lookup_page(brd, sector);
249 dst = kmap_atomic(page, KM_USER1);
250 memcpy(dst + offset, src, copy);
251 kunmap_atomic(dst, KM_USER1);
255 sector += copy >> SECTOR_SHIFT;
257 page = brd_lookup_page(brd, sector);
260 dst = kmap_atomic(page, KM_USER1);
261 memcpy(dst, src, copy);
262 kunmap_atomic(dst, KM_USER1);
267 * Copy n bytes to dst from the brd starting at sector. Does not sleep.
269 static void copy_from_brd(void *dst, struct brd_device *brd,
270 sector_t sector, size_t n)
274 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
277 copy = min_t(size_t, n, PAGE_SIZE - offset);
278 page = brd_lookup_page(brd, sector);
280 src = kmap_atomic(page, KM_USER1);
281 memcpy(dst, src + offset, copy);
282 kunmap_atomic(src, KM_USER1);
284 memset(dst, 0, copy);
288 sector += copy >> SECTOR_SHIFT;
290 page = brd_lookup_page(brd, sector);
292 src = kmap_atomic(page, KM_USER1);
293 memcpy(dst, src, copy);
294 kunmap_atomic(src, KM_USER1);
296 memset(dst, 0, copy);
301 * Process a single bvec of a bio.
303 static int brd_do_bvec(struct brd_device *brd, struct page *page,
304 unsigned int len, unsigned int off, int rw,
311 err = copy_to_brd_setup(brd, sector, len);
316 mem = kmap_atomic(page, KM_USER0);
318 copy_from_brd(mem + off, brd, sector, len);
319 flush_dcache_page(page);
321 flush_dcache_page(page);
322 copy_to_brd(brd, mem + off, sector, len);
324 kunmap_atomic(mem, KM_USER0);
330 static int brd_make_request(struct request_queue *q, struct bio *bio)
332 struct block_device *bdev = bio->bi_bdev;
333 struct brd_device *brd = bdev->bd_disk->private_data;
335 struct bio_vec *bvec;
340 sector = bio->bi_sector;
341 if (sector + (bio->bi_size >> SECTOR_SHIFT) >
342 get_capacity(bdev->bd_disk))
345 if (unlikely(bio->bi_rw & REQ_DISCARD)) {
347 discard_from_brd(brd, sector, bio->bi_size);
355 bio_for_each_segment(bvec, bio, i) {
356 unsigned int len = bvec->bv_len;
357 err = brd_do_bvec(brd, bvec->bv_page, len,
358 bvec->bv_offset, rw, sector);
361 sector += len >> SECTOR_SHIFT;
370 #ifdef CONFIG_BLK_DEV_XIP
371 static int brd_direct_access(struct block_device *bdev, sector_t sector,
372 void **kaddr, unsigned long *pfn)
374 struct brd_device *brd = bdev->bd_disk->private_data;
379 if (sector & (PAGE_SECTORS-1))
381 if (sector + PAGE_SECTORS > get_capacity(bdev->bd_disk))
383 page = brd_insert_page(brd, sector);
386 *kaddr = page_address(page);
387 *pfn = page_to_pfn(page);
393 static int brd_ioctl(struct block_device *bdev, fmode_t mode,
394 unsigned int cmd, unsigned long arg)
397 struct brd_device *brd = bdev->bd_disk->private_data;
399 if (cmd != BLKFLSBUF)
403 * ram device BLKFLSBUF has special semantics, we want to actually
404 * release and destroy the ramdisk data.
406 mutex_lock(&brd_mutex);
407 mutex_lock(&bdev->bd_mutex);
409 if (bdev->bd_openers <= 1) {
411 * Invalidate the cache first, so it isn't written
412 * back to the device.
414 * Another thread might instantiate more buffercache here,
415 * but there is not much we can do to close that race.
417 invalidate_bh_lrus();
418 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
422 mutex_unlock(&bdev->bd_mutex);
423 mutex_unlock(&brd_mutex);
428 static const struct block_device_operations brd_fops = {
429 .owner = THIS_MODULE,
431 #ifdef CONFIG_BLK_DEV_XIP
432 .direct_access = brd_direct_access,
437 * And now the modules code and kernel interface.
440 int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
442 static int part_shift;
443 module_param(rd_nr, int, 0);
444 MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
445 module_param(rd_size, int, 0);
446 MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
447 module_param(max_part, int, 0);
448 MODULE_PARM_DESC(max_part, "Maximum number of partitions per RAM disk");
449 MODULE_LICENSE("GPL");
450 MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
454 /* Legacy boot options - nonmodular */
455 static int __init ramdisk_size(char *str)
457 rd_size = simple_strtol(str, NULL, 0);
460 __setup("ramdisk_size=", ramdisk_size);
464 * The device scheme is derived from loop.c. Keep them in synch where possible
465 * (should share code eventually).
467 static LIST_HEAD(brd_devices);
468 static DEFINE_MUTEX(brd_devices_mutex);
470 static struct brd_device *brd_alloc(int i)
472 struct brd_device *brd;
473 struct gendisk *disk;
475 brd = kzalloc(sizeof(*brd), GFP_KERNEL);
479 spin_lock_init(&brd->brd_lock);
480 INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC);
482 brd->brd_queue = blk_alloc_queue(GFP_KERNEL);
485 blk_queue_make_request(brd->brd_queue, brd_make_request);
486 blk_queue_max_hw_sectors(brd->brd_queue, 1024);
487 blk_queue_bounce_limit(brd->brd_queue, BLK_BOUNCE_ANY);
489 brd->brd_queue->limits.discard_granularity = PAGE_SIZE;
490 brd->brd_queue->limits.max_discard_sectors = UINT_MAX;
491 brd->brd_queue->limits.discard_zeroes_data = 1;
492 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, brd->brd_queue);
494 disk = brd->brd_disk = alloc_disk(1 << part_shift);
497 disk->major = RAMDISK_MAJOR;
498 disk->first_minor = i << part_shift;
499 disk->fops = &brd_fops;
500 disk->private_data = brd;
501 disk->queue = brd->brd_queue;
502 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
503 sprintf(disk->disk_name, "ram%d", i);
504 set_capacity(disk, rd_size * 2);
509 blk_cleanup_queue(brd->brd_queue);
516 static void brd_free(struct brd_device *brd)
518 put_disk(brd->brd_disk);
519 blk_cleanup_queue(brd->brd_queue);
524 static struct brd_device *brd_init_one(int i)
526 struct brd_device *brd;
528 list_for_each_entry(brd, &brd_devices, brd_list) {
529 if (brd->brd_number == i)
535 add_disk(brd->brd_disk);
536 list_add_tail(&brd->brd_list, &brd_devices);
542 static void brd_del_one(struct brd_device *brd)
544 list_del(&brd->brd_list);
545 del_gendisk(brd->brd_disk);
549 static struct kobject *brd_probe(dev_t dev, int *part, void *data)
551 struct brd_device *brd;
552 struct kobject *kobj;
554 mutex_lock(&brd_devices_mutex);
555 brd = brd_init_one(dev & MINORMASK);
556 kobj = brd ? get_disk(brd->brd_disk) : ERR_PTR(-ENOMEM);
557 mutex_unlock(&brd_devices_mutex);
563 static int __init brd_init(void)
567 struct brd_device *brd, *next;
570 * brd module now has a feature to instantiate underlying device
571 * structure on-demand, provided that there is an access dev node.
572 * However, this will not work well with user space tool that doesn't
573 * know about such "feature". In order to not break any existing
574 * tool, we do the following:
576 * (1) if rd_nr is specified, create that many upfront, and this
577 * also becomes a hard limit.
578 * (2) if rd_nr is not specified, create 1 rd device on module
579 * load, user can further extend brd device by create dev node
580 * themselves and have kernel automatically instantiate actual
586 part_shift = fls(max_part);
588 if (rd_nr > 1UL << (MINORBITS - part_shift))
595 nr = CONFIG_BLK_DEV_RAM_COUNT;
596 range = 1UL << (MINORBITS - part_shift);
599 if (register_blkdev(RAMDISK_MAJOR, "ramdisk"))
602 for (i = 0; i < nr; i++) {
606 list_add_tail(&brd->brd_list, &brd_devices);
609 /* point of no return */
611 list_for_each_entry(brd, &brd_devices, brd_list)
612 add_disk(brd->brd_disk);
614 blk_register_region(MKDEV(RAMDISK_MAJOR, 0), range,
615 THIS_MODULE, brd_probe, NULL, NULL);
617 printk(KERN_INFO "brd: module loaded\n");
621 list_for_each_entry_safe(brd, next, &brd_devices, brd_list) {
622 list_del(&brd->brd_list);
625 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
630 static void __exit brd_exit(void)
633 struct brd_device *brd, *next;
635 range = rd_nr ? rd_nr : 1UL << (MINORBITS - part_shift);
637 list_for_each_entry_safe(brd, next, &brd_devices, brd_list)
640 blk_unregister_region(MKDEV(RAMDISK_MAJOR, 0), range);
641 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
644 module_init(brd_init);
645 module_exit(brd_exit);