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>
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
39 struct request_queue *brd_queue;
40 struct gendisk *brd_disk;
41 struct list_head brd_list;
44 * Backing store of pages and lock to protect it. This is the contents
45 * of the block device.
48 struct radix_tree_root brd_pages;
52 * Look up and return a brd's page for a given sector.
54 static DEFINE_MUTEX(brd_mutex);
55 static struct page *brd_lookup_page(struct brd_device *brd, sector_t sector)
61 * The page lifetime is protected by the fact that we have opened the
62 * device node -- brd pages will never be deleted under us, so we
63 * don't need any further locking or refcounting.
65 * This is strictly true for the radix-tree nodes as well (ie. we
66 * don't actually need the rcu_read_lock()), however that is not a
67 * documented feature of the radix-tree API so it is better to be
68 * safe here (we don't have total exclusion from radix tree updates
69 * here, only deletes).
72 idx = sector >> PAGE_SECTORS_SHIFT; /* sector to page index */
73 page = radix_tree_lookup(&brd->brd_pages, idx);
76 BUG_ON(page && page->index != idx);
82 * Look up and return a brd's page for a given sector.
83 * If one does not exist, allocate an empty page, and insert that. Then
86 static struct page *brd_insert_page(struct brd_device *brd, sector_t sector)
92 page = brd_lookup_page(brd, sector);
97 * Must use NOIO because we don't want to recurse back into the
98 * block or filesystem layers from page reclaim.
100 * Cannot support DAX and highmem, because our ->direct_access
101 * routine for DAX must return memory that is always addressable.
102 * If DAX was reworked to use pfns and kmap throughout, this
103 * restriction might be able to be lifted.
105 gfp_flags = GFP_NOIO | __GFP_ZERO;
106 #ifndef CONFIG_BLK_DEV_RAM_DAX
107 gfp_flags |= __GFP_HIGHMEM;
109 page = alloc_page(gfp_flags);
113 if (radix_tree_preload(GFP_NOIO)) {
118 spin_lock(&brd->brd_lock);
119 idx = sector >> PAGE_SECTORS_SHIFT;
121 if (radix_tree_insert(&brd->brd_pages, idx, page)) {
123 page = radix_tree_lookup(&brd->brd_pages, idx);
125 BUG_ON(page->index != idx);
127 spin_unlock(&brd->brd_lock);
129 radix_tree_preload_end();
134 static void brd_free_page(struct brd_device *brd, sector_t sector)
139 spin_lock(&brd->brd_lock);
140 idx = sector >> PAGE_SECTORS_SHIFT;
141 page = radix_tree_delete(&brd->brd_pages, idx);
142 spin_unlock(&brd->brd_lock);
147 static void brd_zero_page(struct brd_device *brd, sector_t sector)
151 page = brd_lookup_page(brd, sector);
153 clear_highpage(page);
157 * Free all backing store pages and radix tree. This must only be called when
158 * there are no other users of the device.
160 #define FREE_BATCH 16
161 static void brd_free_pages(struct brd_device *brd)
163 unsigned long pos = 0;
164 struct page *pages[FREE_BATCH];
170 nr_pages = radix_tree_gang_lookup(&brd->brd_pages,
171 (void **)pages, pos, FREE_BATCH);
173 for (i = 0; i < nr_pages; i++) {
176 BUG_ON(pages[i]->index < pos);
177 pos = pages[i]->index;
178 ret = radix_tree_delete(&brd->brd_pages, pos);
179 BUG_ON(!ret || ret != pages[i]);
180 __free_page(pages[i]);
186 * This assumes radix_tree_gang_lookup always returns as
187 * many pages as possible. If the radix-tree code changes,
188 * so will this have to.
190 } while (nr_pages == FREE_BATCH);
194 * copy_to_brd_setup must be called before copy_to_brd. It may sleep.
196 static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n)
198 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
201 copy = min_t(size_t, n, PAGE_SIZE - offset);
202 if (!brd_insert_page(brd, sector))
205 sector += copy >> SECTOR_SHIFT;
206 if (!brd_insert_page(brd, sector))
212 static void discard_from_brd(struct brd_device *brd,
213 sector_t sector, size_t n)
215 while (n >= PAGE_SIZE) {
217 * Don't want to actually discard pages here because
218 * re-allocating the pages can result in writeback
219 * deadlocks under heavy load.
222 brd_free_page(brd, sector);
224 brd_zero_page(brd, sector);
225 sector += PAGE_SIZE >> SECTOR_SHIFT;
231 * Copy n bytes from src to the brd starting at sector. Does not sleep.
233 static void copy_to_brd(struct brd_device *brd, const void *src,
234 sector_t sector, size_t n)
238 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
241 copy = min_t(size_t, n, PAGE_SIZE - offset);
242 page = brd_lookup_page(brd, sector);
245 dst = kmap_atomic(page);
246 memcpy(dst + offset, src, copy);
251 sector += copy >> SECTOR_SHIFT;
253 page = brd_lookup_page(brd, sector);
256 dst = kmap_atomic(page);
257 memcpy(dst, src, copy);
263 * Copy n bytes to dst from the brd starting at sector. Does not sleep.
265 static void copy_from_brd(void *dst, struct brd_device *brd,
266 sector_t sector, size_t n)
270 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
273 copy = min_t(size_t, n, PAGE_SIZE - offset);
274 page = brd_lookup_page(brd, sector);
276 src = kmap_atomic(page);
277 memcpy(dst, src + offset, copy);
280 memset(dst, 0, copy);
284 sector += copy >> SECTOR_SHIFT;
286 page = brd_lookup_page(brd, sector);
288 src = kmap_atomic(page);
289 memcpy(dst, src, copy);
292 memset(dst, 0, copy);
297 * Process a single bvec of a bio.
299 static int brd_do_bvec(struct brd_device *brd, struct page *page,
300 unsigned int len, unsigned int off, int rw,
307 err = copy_to_brd_setup(brd, sector, len);
312 mem = kmap_atomic(page);
314 copy_from_brd(mem + off, brd, sector, len);
315 flush_dcache_page(page);
317 flush_dcache_page(page);
318 copy_to_brd(brd, mem + off, sector, len);
326 static void brd_make_request(struct request_queue *q, struct bio *bio)
328 struct block_device *bdev = bio->bi_bdev;
329 struct brd_device *brd = bdev->bd_disk->private_data;
333 struct bvec_iter iter;
335 sector = bio->bi_iter.bi_sector;
336 if (bio_end_sector(bio) > get_capacity(bdev->bd_disk))
339 if (unlikely(bio->bi_rw & REQ_DISCARD)) {
340 discard_from_brd(brd, sector, bio->bi_iter.bi_size);
348 bio_for_each_segment(bvec, bio, iter) {
349 unsigned int len = bvec.bv_len;
352 err = brd_do_bvec(brd, bvec.bv_page, len,
353 bvec.bv_offset, rw, sector);
356 sector += len >> SECTOR_SHIFT;
366 static int brd_rw_page(struct block_device *bdev, sector_t sector,
367 struct page *page, int rw)
369 struct brd_device *brd = bdev->bd_disk->private_data;
370 int err = brd_do_bvec(brd, page, PAGE_CACHE_SIZE, 0, rw, sector);
371 page_endio(page, rw & WRITE, err);
375 #ifdef CONFIG_BLK_DEV_RAM_DAX
376 static long brd_direct_access(struct block_device *bdev, sector_t sector,
377 void __pmem **kaddr, unsigned long *pfn)
379 struct brd_device *brd = bdev->bd_disk->private_data;
384 page = brd_insert_page(brd, sector);
387 *kaddr = (void __pmem *)page_address(page);
388 *pfn = page_to_pfn(page);
393 #define brd_direct_access NULL
396 static int brd_ioctl(struct block_device *bdev, fmode_t mode,
397 unsigned int cmd, unsigned long arg)
400 struct brd_device *brd = bdev->bd_disk->private_data;
402 if (cmd != BLKFLSBUF)
406 * ram device BLKFLSBUF has special semantics, we want to actually
407 * release and destroy the ramdisk data.
409 mutex_lock(&brd_mutex);
410 mutex_lock(&bdev->bd_mutex);
412 if (bdev->bd_openers <= 1) {
414 * Kill the cache first, so it isn't written back to the
417 * Another thread might instantiate more buffercache here,
418 * but there is not much we can do to close that race.
424 mutex_unlock(&bdev->bd_mutex);
425 mutex_unlock(&brd_mutex);
430 static const struct block_device_operations brd_fops = {
431 .owner = THIS_MODULE,
432 .rw_page = brd_rw_page,
434 .direct_access = brd_direct_access,
438 * And now the modules code and kernel interface.
440 static int rd_nr = CONFIG_BLK_DEV_RAM_COUNT;
441 module_param(rd_nr, int, S_IRUGO);
442 MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
444 int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
445 module_param(rd_size, int, S_IRUGO);
446 MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
448 static int max_part = 1;
449 module_param(max_part, int, S_IRUGO);
450 MODULE_PARM_DESC(max_part, "Num Minors to reserve between devices");
452 MODULE_LICENSE("GPL");
453 MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
457 /* Legacy boot options - nonmodular */
458 static int __init ramdisk_size(char *str)
460 rd_size = simple_strtol(str, NULL, 0);
463 __setup("ramdisk_size=", ramdisk_size);
467 * The device scheme is derived from loop.c. Keep them in synch where possible
468 * (should share code eventually).
470 static LIST_HEAD(brd_devices);
471 static DEFINE_MUTEX(brd_devices_mutex);
473 static struct brd_device *brd_alloc(int i)
475 struct brd_device *brd;
476 struct gendisk *disk;
478 brd = kzalloc(sizeof(*brd), GFP_KERNEL);
482 spin_lock_init(&brd->brd_lock);
483 INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC);
485 brd->brd_queue = blk_alloc_queue(GFP_KERNEL);
489 blk_queue_make_request(brd->brd_queue, brd_make_request);
490 blk_queue_max_hw_sectors(brd->brd_queue, 1024);
491 blk_queue_bounce_limit(brd->brd_queue, BLK_BOUNCE_ANY);
493 /* This is so fdisk will align partitions on 4k, because of
494 * direct_access API needing 4k alignment, returning a PFN
495 * (This is only a problem on very small devices <= 4M,
496 * otherwise fdisk will align on 1M. Regardless this call
499 blk_queue_physical_block_size(brd->brd_queue, PAGE_SIZE);
501 brd->brd_queue->limits.discard_granularity = PAGE_SIZE;
502 blk_queue_max_discard_sectors(brd->brd_queue, UINT_MAX);
503 brd->brd_queue->limits.discard_zeroes_data = 1;
504 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, brd->brd_queue);
506 disk = brd->brd_disk = alloc_disk(max_part);
509 disk->major = RAMDISK_MAJOR;
510 disk->first_minor = i * max_part;
511 disk->fops = &brd_fops;
512 disk->private_data = brd;
513 disk->queue = brd->brd_queue;
514 disk->flags = GENHD_FL_EXT_DEVT;
515 sprintf(disk->disk_name, "ram%d", i);
516 set_capacity(disk, rd_size * 2);
521 blk_cleanup_queue(brd->brd_queue);
528 static void brd_free(struct brd_device *brd)
530 put_disk(brd->brd_disk);
531 blk_cleanup_queue(brd->brd_queue);
536 static struct brd_device *brd_init_one(int i, bool *new)
538 struct brd_device *brd;
541 list_for_each_entry(brd, &brd_devices, brd_list) {
542 if (brd->brd_number == i)
548 add_disk(brd->brd_disk);
549 list_add_tail(&brd->brd_list, &brd_devices);
556 static void brd_del_one(struct brd_device *brd)
558 list_del(&brd->brd_list);
559 del_gendisk(brd->brd_disk);
563 static struct kobject *brd_probe(dev_t dev, int *part, void *data)
565 struct brd_device *brd;
566 struct kobject *kobj;
569 mutex_lock(&brd_devices_mutex);
570 brd = brd_init_one(MINOR(dev) / max_part, &new);
571 kobj = brd ? get_disk(brd->brd_disk) : NULL;
572 mutex_unlock(&brd_devices_mutex);
580 static int __init brd_init(void)
582 struct brd_device *brd, *next;
586 * brd module now has a feature to instantiate underlying device
587 * structure on-demand, provided that there is an access dev node.
589 * (1) if rd_nr is specified, create that many upfront. else
590 * it defaults to CONFIG_BLK_DEV_RAM_COUNT
591 * (2) User can further extend brd devices by create dev node themselves
592 * and have kernel automatically instantiate actual device
593 * on-demand. Example:
594 * mknod /path/devnod_name b 1 X # 1 is the rd major
595 * fdisk -l /path/devnod_name
596 * If (X / max_part) was not already created it will be created
600 if (register_blkdev(RAMDISK_MAJOR, "ramdisk"))
603 if (unlikely(!max_part))
606 for (i = 0; i < rd_nr; i++) {
610 list_add_tail(&brd->brd_list, &brd_devices);
613 /* point of no return */
615 list_for_each_entry(brd, &brd_devices, brd_list)
616 add_disk(brd->brd_disk);
618 blk_register_region(MKDEV(RAMDISK_MAJOR, 0), 1UL << MINORBITS,
619 THIS_MODULE, brd_probe, NULL, NULL);
621 pr_info("brd: module loaded\n");
625 list_for_each_entry_safe(brd, next, &brd_devices, brd_list) {
626 list_del(&brd->brd_list);
629 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
631 pr_info("brd: module NOT loaded !!!\n");
635 static void __exit brd_exit(void)
637 struct brd_device *brd, *next;
639 list_for_each_entry_safe(brd, next, &brd_devices, brd_list)
642 blk_unregister_region(MKDEV(RAMDISK_MAJOR, 0), 1UL << MINORBITS);
643 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
645 pr_info("brd: module unloaded\n");
648 module_init(brd_init);
649 module_exit(brd_exit);