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/smp_lock.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 struct page *brd_lookup_page(struct brd_device *brd, sector_t sector)
64 * The page lifetime is protected by the fact that we have opened the
65 * device node -- brd pages will never be deleted under us, so we
66 * don't need any further locking or refcounting.
68 * This is strictly true for the radix-tree nodes as well (ie. we
69 * don't actually need the rcu_read_lock()), however that is not a
70 * documented feature of the radix-tree API so it is better to be
71 * safe here (we don't have total exclusion from radix tree updates
72 * here, only deletes).
75 idx = sector >> PAGE_SECTORS_SHIFT; /* sector to page index */
76 page = radix_tree_lookup(&brd->brd_pages, idx);
79 BUG_ON(page && page->index != idx);
85 * Look up and return a brd's page for a given sector.
86 * If one does not exist, allocate an empty page, and insert that. Then
89 static struct page *brd_insert_page(struct brd_device *brd, sector_t sector)
95 page = brd_lookup_page(brd, sector);
100 * Must use NOIO because we don't want to recurse back into the
101 * block or filesystem layers from page reclaim.
103 * Cannot support XIP and highmem, because our ->direct_access
104 * routine for XIP must return memory that is always addressable.
105 * If XIP was reworked to use pfns and kmap throughout, this
106 * restriction might be able to be lifted.
108 gfp_flags = GFP_NOIO | __GFP_ZERO;
109 #ifndef CONFIG_BLK_DEV_XIP
110 gfp_flags |= __GFP_HIGHMEM;
112 page = alloc_page(gfp_flags);
116 if (radix_tree_preload(GFP_NOIO)) {
121 spin_lock(&brd->brd_lock);
122 idx = sector >> PAGE_SECTORS_SHIFT;
123 if (radix_tree_insert(&brd->brd_pages, idx, page)) {
125 page = radix_tree_lookup(&brd->brd_pages, idx);
127 BUG_ON(page->index != idx);
130 spin_unlock(&brd->brd_lock);
132 radix_tree_preload_end();
137 static void brd_free_page(struct brd_device *brd, sector_t sector)
142 spin_lock(&brd->brd_lock);
143 idx = sector >> PAGE_SECTORS_SHIFT;
144 page = radix_tree_delete(&brd->brd_pages, idx);
145 spin_unlock(&brd->brd_lock);
150 static void brd_zero_page(struct brd_device *brd, sector_t sector)
154 page = brd_lookup_page(brd, sector);
156 clear_highpage(page);
160 * Free all backing store pages and radix tree. This must only be called when
161 * there are no other users of the device.
163 #define FREE_BATCH 16
164 static void brd_free_pages(struct brd_device *brd)
166 unsigned long pos = 0;
167 struct page *pages[FREE_BATCH];
173 nr_pages = radix_tree_gang_lookup(&brd->brd_pages,
174 (void **)pages, pos, FREE_BATCH);
176 for (i = 0; i < nr_pages; i++) {
179 BUG_ON(pages[i]->index < pos);
180 pos = pages[i]->index;
181 ret = radix_tree_delete(&brd->brd_pages, pos);
182 BUG_ON(!ret || ret != pages[i]);
183 __free_page(pages[i]);
189 * This assumes radix_tree_gang_lookup always returns as
190 * many pages as possible. If the radix-tree code changes,
191 * so will this have to.
193 } while (nr_pages == FREE_BATCH);
197 * copy_to_brd_setup must be called before copy_to_brd. It may sleep.
199 static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n)
201 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
204 copy = min_t(size_t, n, PAGE_SIZE - offset);
205 if (!brd_insert_page(brd, sector))
208 sector += copy >> SECTOR_SHIFT;
209 if (!brd_insert_page(brd, sector))
215 static void discard_from_brd(struct brd_device *brd,
216 sector_t sector, size_t n)
218 while (n >= PAGE_SIZE) {
220 * Don't want to actually discard pages here because
221 * re-allocating the pages can result in writeback
222 * deadlocks under heavy load.
225 brd_free_page(brd, sector);
227 brd_zero_page(brd, sector);
228 sector += PAGE_SIZE >> SECTOR_SHIFT;
234 * Copy n bytes from src to the brd starting at sector. Does not sleep.
236 static void copy_to_brd(struct brd_device *brd, const void *src,
237 sector_t sector, size_t n)
241 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
244 copy = min_t(size_t, n, PAGE_SIZE - offset);
245 page = brd_lookup_page(brd, sector);
248 dst = kmap_atomic(page, KM_USER1);
249 memcpy(dst + offset, src, copy);
250 kunmap_atomic(dst, KM_USER1);
254 sector += copy >> SECTOR_SHIFT;
256 page = brd_lookup_page(brd, sector);
259 dst = kmap_atomic(page, KM_USER1);
260 memcpy(dst, src, copy);
261 kunmap_atomic(dst, KM_USER1);
266 * Copy n bytes to dst from the brd starting at sector. Does not sleep.
268 static void copy_from_brd(void *dst, struct brd_device *brd,
269 sector_t sector, size_t n)
273 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
276 copy = min_t(size_t, n, PAGE_SIZE - offset);
277 page = brd_lookup_page(brd, sector);
279 src = kmap_atomic(page, KM_USER1);
280 memcpy(dst, src + offset, copy);
281 kunmap_atomic(src, KM_USER1);
283 memset(dst, 0, copy);
287 sector += copy >> SECTOR_SHIFT;
289 page = brd_lookup_page(brd, sector);
291 src = kmap_atomic(page, KM_USER1);
292 memcpy(dst, src, copy);
293 kunmap_atomic(src, KM_USER1);
295 memset(dst, 0, copy);
300 * Process a single bvec of a bio.
302 static int brd_do_bvec(struct brd_device *brd, struct page *page,
303 unsigned int len, unsigned int off, int rw,
310 err = copy_to_brd_setup(brd, sector, len);
315 mem = kmap_atomic(page, KM_USER0);
317 copy_from_brd(mem + off, brd, sector, len);
318 flush_dcache_page(page);
320 flush_dcache_page(page);
321 copy_to_brd(brd, mem + off, sector, len);
323 kunmap_atomic(mem, KM_USER0);
329 static int brd_make_request(struct request_queue *q, struct bio *bio)
331 struct block_device *bdev = bio->bi_bdev;
332 struct brd_device *brd = bdev->bd_disk->private_data;
334 struct bio_vec *bvec;
339 sector = bio->bi_sector;
340 if (sector + (bio->bi_size >> SECTOR_SHIFT) >
341 get_capacity(bdev->bd_disk))
344 if (unlikely(bio->bi_rw & REQ_DISCARD)) {
346 discard_from_brd(brd, sector, bio->bi_size);
354 bio_for_each_segment(bvec, bio, i) {
355 unsigned int len = bvec->bv_len;
356 err = brd_do_bvec(brd, bvec->bv_page, len,
357 bvec->bv_offset, rw, sector);
360 sector += len >> SECTOR_SHIFT;
369 #ifdef CONFIG_BLK_DEV_XIP
370 static int brd_direct_access(struct block_device *bdev, sector_t sector,
371 void **kaddr, unsigned long *pfn)
373 struct brd_device *brd = bdev->bd_disk->private_data;
378 if (sector & (PAGE_SECTORS-1))
380 if (sector + PAGE_SECTORS > get_capacity(bdev->bd_disk))
382 page = brd_insert_page(brd, sector);
385 *kaddr = page_address(page);
386 *pfn = page_to_pfn(page);
392 static int brd_ioctl(struct block_device *bdev, fmode_t mode,
393 unsigned int cmd, unsigned long arg)
396 struct brd_device *brd = bdev->bd_disk->private_data;
398 if (cmd != BLKFLSBUF)
402 * ram device BLKFLSBUF has special semantics, we want to actually
403 * release and destroy the ramdisk data.
406 mutex_lock(&bdev->bd_mutex);
408 if (bdev->bd_openers <= 1) {
410 * Invalidate the cache first, so it isn't written
411 * back to the device.
413 * Another thread might instantiate more buffercache here,
414 * but there is not much we can do to close that race.
416 invalidate_bh_lrus();
417 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
421 mutex_unlock(&bdev->bd_mutex);
427 static const struct block_device_operations brd_fops = {
428 .owner = THIS_MODULE,
430 #ifdef CONFIG_BLK_DEV_XIP
431 .direct_access = brd_direct_access,
436 * And now the modules code and kernel interface.
439 int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
441 static int part_shift;
442 module_param(rd_nr, int, 0);
443 MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
444 module_param(rd_size, int, 0);
445 MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
446 module_param(max_part, int, 0);
447 MODULE_PARM_DESC(max_part, "Maximum number of partitions per RAM disk");
448 MODULE_LICENSE("GPL");
449 MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
453 /* Legacy boot options - nonmodular */
454 static int __init ramdisk_size(char *str)
456 rd_size = simple_strtol(str, NULL, 0);
459 __setup("ramdisk_size=", ramdisk_size);
463 * The device scheme is derived from loop.c. Keep them in synch where possible
464 * (should share code eventually).
466 static LIST_HEAD(brd_devices);
467 static DEFINE_MUTEX(brd_devices_mutex);
469 static struct brd_device *brd_alloc(int i)
471 struct brd_device *brd;
472 struct gendisk *disk;
474 brd = kzalloc(sizeof(*brd), GFP_KERNEL);
478 spin_lock_init(&brd->brd_lock);
479 INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC);
481 brd->brd_queue = blk_alloc_queue(GFP_KERNEL);
484 blk_queue_make_request(brd->brd_queue, brd_make_request);
485 blk_queue_max_hw_sectors(brd->brd_queue, 1024);
486 blk_queue_bounce_limit(brd->brd_queue, BLK_BOUNCE_ANY);
488 brd->brd_queue->limits.discard_granularity = PAGE_SIZE;
489 brd->brd_queue->limits.max_discard_sectors = UINT_MAX;
490 brd->brd_queue->limits.discard_zeroes_data = 1;
491 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, brd->brd_queue);
493 disk = brd->brd_disk = alloc_disk(1 << part_shift);
496 disk->major = RAMDISK_MAJOR;
497 disk->first_minor = i << part_shift;
498 disk->fops = &brd_fops;
499 disk->private_data = brd;
500 disk->queue = brd->brd_queue;
501 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
502 sprintf(disk->disk_name, "ram%d", i);
503 set_capacity(disk, rd_size * 2);
508 blk_cleanup_queue(brd->brd_queue);
515 static void brd_free(struct brd_device *brd)
517 put_disk(brd->brd_disk);
518 blk_cleanup_queue(brd->brd_queue);
523 static struct brd_device *brd_init_one(int i)
525 struct brd_device *brd;
527 list_for_each_entry(brd, &brd_devices, brd_list) {
528 if (brd->brd_number == i)
534 add_disk(brd->brd_disk);
535 list_add_tail(&brd->brd_list, &brd_devices);
541 static void brd_del_one(struct brd_device *brd)
543 list_del(&brd->brd_list);
544 del_gendisk(brd->brd_disk);
548 static struct kobject *brd_probe(dev_t dev, int *part, void *data)
550 struct brd_device *brd;
551 struct kobject *kobj;
553 mutex_lock(&brd_devices_mutex);
554 brd = brd_init_one(dev & MINORMASK);
555 kobj = brd ? get_disk(brd->brd_disk) : ERR_PTR(-ENOMEM);
556 mutex_unlock(&brd_devices_mutex);
562 static int __init brd_init(void)
566 struct brd_device *brd, *next;
569 * brd module now has a feature to instantiate underlying device
570 * structure on-demand, provided that there is an access dev node.
571 * However, this will not work well with user space tool that doesn't
572 * know about such "feature". In order to not break any existing
573 * tool, we do the following:
575 * (1) if rd_nr is specified, create that many upfront, and this
576 * also becomes a hard limit.
577 * (2) if rd_nr is not specified, create 1 rd device on module
578 * load, user can further extend brd device by create dev node
579 * themselves and have kernel automatically instantiate actual
585 part_shift = fls(max_part);
587 if (rd_nr > 1UL << (MINORBITS - part_shift))
594 nr = CONFIG_BLK_DEV_RAM_COUNT;
595 range = 1UL << (MINORBITS - part_shift);
598 if (register_blkdev(RAMDISK_MAJOR, "ramdisk"))
601 for (i = 0; i < nr; i++) {
605 list_add_tail(&brd->brd_list, &brd_devices);
608 /* point of no return */
610 list_for_each_entry(brd, &brd_devices, brd_list)
611 add_disk(brd->brd_disk);
613 blk_register_region(MKDEV(RAMDISK_MAJOR, 0), range,
614 THIS_MODULE, brd_probe, NULL, NULL);
616 printk(KERN_INFO "brd: module loaded\n");
620 list_for_each_entry_safe(brd, next, &brd_devices, brd_list) {
621 list_del(&brd->brd_list);
624 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
629 static void __exit brd_exit(void)
632 struct brd_device *brd, *next;
634 range = rd_nr ? rd_nr : 1UL << (MINORBITS - part_shift);
636 list_for_each_entry_safe(brd, next, &brd_devices, brd_list)
639 blk_unregister_region(MKDEV(RAMDISK_MAJOR, 0), range);
640 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
643 module_init(brd_init);
644 module_exit(brd_exit);