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>
22 #ifdef CONFIG_BLK_DEV_RAM_DAX
23 #include <linux/pfn_t.h>
26 #include <linux/uaccess.h>
28 #define SECTOR_SHIFT 9
29 #define PAGE_SECTORS_SHIFT (PAGE_SHIFT - SECTOR_SHIFT)
30 #define PAGE_SECTORS (1 << PAGE_SECTORS_SHIFT)
33 * Each block ramdisk device has a radix_tree brd_pages of pages that stores
34 * the pages containing the block device's contents. A brd page's ->index is
35 * its offset in PAGE_SIZE units. This is similar to, but in no way connected
36 * with, the kernel's pagecache or buffer cache (which sit above our block
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 DEFINE_MUTEX(brd_mutex);
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 DAX and highmem, because our ->direct_access
104 * routine for DAX must return memory that is always addressable.
105 * If DAX 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_RAM_DAX
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;
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);
130 spin_unlock(&brd->brd_lock);
132 radix_tree_preload_end();
138 * Free all backing store pages and radix tree. This must only be called when
139 * there are no other users of the device.
141 #define FREE_BATCH 16
142 static void brd_free_pages(struct brd_device *brd)
144 unsigned long pos = 0;
145 struct page *pages[FREE_BATCH];
151 nr_pages = radix_tree_gang_lookup(&brd->brd_pages,
152 (void **)pages, pos, FREE_BATCH);
154 for (i = 0; i < nr_pages; i++) {
157 BUG_ON(pages[i]->index < pos);
158 pos = pages[i]->index;
159 ret = radix_tree_delete(&brd->brd_pages, pos);
160 BUG_ON(!ret || ret != pages[i]);
161 __free_page(pages[i]);
167 * This assumes radix_tree_gang_lookup always returns as
168 * many pages as possible. If the radix-tree code changes,
169 * so will this have to.
171 } while (nr_pages == FREE_BATCH);
175 * copy_to_brd_setup must be called before copy_to_brd. It may sleep.
177 static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n)
179 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
182 copy = min_t(size_t, n, PAGE_SIZE - offset);
183 if (!brd_insert_page(brd, sector))
186 sector += copy >> SECTOR_SHIFT;
187 if (!brd_insert_page(brd, sector))
194 * Copy n bytes from src to the brd starting at sector. Does not sleep.
196 static void copy_to_brd(struct brd_device *brd, const void *src,
197 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 page = brd_lookup_page(brd, sector);
208 dst = kmap_atomic(page);
209 memcpy(dst + offset, src, copy);
214 sector += copy >> SECTOR_SHIFT;
216 page = brd_lookup_page(brd, sector);
219 dst = kmap_atomic(page);
220 memcpy(dst, src, copy);
226 * Copy n bytes to dst from the brd starting at sector. Does not sleep.
228 static void copy_from_brd(void *dst, struct brd_device *brd,
229 sector_t sector, size_t n)
233 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
236 copy = min_t(size_t, n, PAGE_SIZE - offset);
237 page = brd_lookup_page(brd, sector);
239 src = kmap_atomic(page);
240 memcpy(dst, src + offset, copy);
243 memset(dst, 0, copy);
247 sector += copy >> SECTOR_SHIFT;
249 page = brd_lookup_page(brd, sector);
251 src = kmap_atomic(page);
252 memcpy(dst, src, copy);
255 memset(dst, 0, copy);
260 * Process a single bvec of a bio.
262 static int brd_do_bvec(struct brd_device *brd, struct page *page,
263 unsigned int len, unsigned int off, bool is_write,
270 err = copy_to_brd_setup(brd, sector, len);
275 mem = kmap_atomic(page);
277 copy_from_brd(mem + off, brd, sector, len);
278 flush_dcache_page(page);
280 flush_dcache_page(page);
281 copy_to_brd(brd, mem + off, sector, len);
289 static blk_qc_t brd_make_request(struct request_queue *q, struct bio *bio)
291 struct block_device *bdev = bio->bi_bdev;
292 struct brd_device *brd = bdev->bd_disk->private_data;
295 struct bvec_iter iter;
297 sector = bio->bi_iter.bi_sector;
298 if (bio_end_sector(bio) > get_capacity(bdev->bd_disk))
301 bio_for_each_segment(bvec, bio, iter) {
302 unsigned int len = bvec.bv_len;
305 err = brd_do_bvec(brd, bvec.bv_page, len, bvec.bv_offset,
306 op_is_write(bio_op(bio)), sector);
309 sector += len >> SECTOR_SHIFT;
313 return BLK_QC_T_NONE;
316 return BLK_QC_T_NONE;
319 static int brd_rw_page(struct block_device *bdev, sector_t sector,
320 struct page *page, bool is_write)
322 struct brd_device *brd = bdev->bd_disk->private_data;
323 int err = brd_do_bvec(brd, page, PAGE_SIZE, 0, is_write, sector);
324 page_endio(page, is_write, err);
328 #ifdef CONFIG_BLK_DEV_RAM_DAX
329 static long brd_direct_access(struct block_device *bdev, sector_t sector,
330 void **kaddr, pfn_t *pfn, long size)
332 struct brd_device *brd = bdev->bd_disk->private_data;
337 page = brd_insert_page(brd, sector);
340 *kaddr = page_address(page);
341 *pfn = page_to_pfn_t(page);
346 #define brd_direct_access NULL
349 static const struct block_device_operations brd_fops = {
350 .owner = THIS_MODULE,
351 .rw_page = brd_rw_page,
352 .direct_access = brd_direct_access,
356 * And now the modules code and kernel interface.
358 static int rd_nr = CONFIG_BLK_DEV_RAM_COUNT;
359 module_param(rd_nr, int, S_IRUGO);
360 MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
362 unsigned long rd_size = CONFIG_BLK_DEV_RAM_SIZE;
363 module_param(rd_size, ulong, S_IRUGO);
364 MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
366 static int max_part = 1;
367 module_param(max_part, int, S_IRUGO);
368 MODULE_PARM_DESC(max_part, "Num Minors to reserve between devices");
370 MODULE_LICENSE("GPL");
371 MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
375 /* Legacy boot options - nonmodular */
376 static int __init ramdisk_size(char *str)
378 rd_size = simple_strtol(str, NULL, 0);
381 __setup("ramdisk_size=", ramdisk_size);
385 * The device scheme is derived from loop.c. Keep them in synch where possible
386 * (should share code eventually).
388 static LIST_HEAD(brd_devices);
389 static DEFINE_MUTEX(brd_devices_mutex);
391 static struct brd_device *brd_alloc(int i)
393 struct brd_device *brd;
394 struct gendisk *disk;
396 brd = kzalloc(sizeof(*brd), GFP_KERNEL);
400 spin_lock_init(&brd->brd_lock);
401 INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC);
403 brd->brd_queue = blk_alloc_queue(GFP_KERNEL);
407 blk_queue_make_request(brd->brd_queue, brd_make_request);
408 blk_queue_max_hw_sectors(brd->brd_queue, 1024);
409 blk_queue_bounce_limit(brd->brd_queue, BLK_BOUNCE_ANY);
411 /* This is so fdisk will align partitions on 4k, because of
412 * direct_access API needing 4k alignment, returning a PFN
413 * (This is only a problem on very small devices <= 4M,
414 * otherwise fdisk will align on 1M. Regardless this call
417 blk_queue_physical_block_size(brd->brd_queue, PAGE_SIZE);
418 #ifdef CONFIG_BLK_DEV_RAM_DAX
419 queue_flag_set_unlocked(QUEUE_FLAG_DAX, brd->brd_queue);
421 disk = brd->brd_disk = alloc_disk(max_part);
424 disk->major = RAMDISK_MAJOR;
425 disk->first_minor = i * max_part;
426 disk->fops = &brd_fops;
427 disk->private_data = brd;
428 disk->queue = brd->brd_queue;
429 disk->flags = GENHD_FL_EXT_DEVT;
430 sprintf(disk->disk_name, "ram%d", i);
431 set_capacity(disk, rd_size * 2);
436 blk_cleanup_queue(brd->brd_queue);
443 static void brd_free(struct brd_device *brd)
445 put_disk(brd->brd_disk);
446 blk_cleanup_queue(brd->brd_queue);
451 static struct brd_device *brd_init_one(int i, bool *new)
453 struct brd_device *brd;
456 list_for_each_entry(brd, &brd_devices, brd_list) {
457 if (brd->brd_number == i)
463 add_disk(brd->brd_disk);
464 list_add_tail(&brd->brd_list, &brd_devices);
471 static void brd_del_one(struct brd_device *brd)
473 list_del(&brd->brd_list);
474 del_gendisk(brd->brd_disk);
478 static struct kobject *brd_probe(dev_t dev, int *part, void *data)
480 struct brd_device *brd;
481 struct kobject *kobj;
484 mutex_lock(&brd_devices_mutex);
485 brd = brd_init_one(MINOR(dev) / max_part, &new);
486 kobj = brd ? get_disk(brd->brd_disk) : NULL;
487 mutex_unlock(&brd_devices_mutex);
495 static int __init brd_init(void)
497 struct brd_device *brd, *next;
501 * brd module now has a feature to instantiate underlying device
502 * structure on-demand, provided that there is an access dev node.
504 * (1) if rd_nr is specified, create that many upfront. else
505 * it defaults to CONFIG_BLK_DEV_RAM_COUNT
506 * (2) User can further extend brd devices by create dev node themselves
507 * and have kernel automatically instantiate actual device
508 * on-demand. Example:
509 * mknod /path/devnod_name b 1 X # 1 is the rd major
510 * fdisk -l /path/devnod_name
511 * If (X / max_part) was not already created it will be created
515 if (register_blkdev(RAMDISK_MAJOR, "ramdisk"))
518 if (unlikely(!max_part))
521 for (i = 0; i < rd_nr; i++) {
525 list_add_tail(&brd->brd_list, &brd_devices);
528 /* point of no return */
530 list_for_each_entry(brd, &brd_devices, brd_list)
531 add_disk(brd->brd_disk);
533 blk_register_region(MKDEV(RAMDISK_MAJOR, 0), 1UL << MINORBITS,
534 THIS_MODULE, brd_probe, NULL, NULL);
536 pr_info("brd: module loaded\n");
540 list_for_each_entry_safe(brd, next, &brd_devices, brd_list) {
541 list_del(&brd->brd_list);
544 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
546 pr_info("brd: module NOT loaded !!!\n");
550 static void __exit brd_exit(void)
552 struct brd_device *brd, *next;
554 list_for_each_entry_safe(brd, next, &brd_devices, brd_list)
557 blk_unregister_region(MKDEV(RAMDISK_MAJOR, 0), 1UL << MINORBITS);
558 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
560 pr_info("brd: module unloaded\n");
563 module_init(brd_init);
564 module_exit(brd_exit);