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>
24 #include <linux/dax.h>
27 #include <linux/uaccess.h>
29 #define SECTOR_SHIFT 9
30 #define PAGE_SECTORS_SHIFT (PAGE_SHIFT - SECTOR_SHIFT)
31 #define PAGE_SECTORS (1 << PAGE_SECTORS_SHIFT)
34 * Each block ramdisk device has a radix_tree brd_pages of pages that stores
35 * the pages containing the block device's contents. A brd page's ->index is
36 * its offset in PAGE_SIZE units. This is similar to, but in no way connected
37 * with, the kernel's pagecache or buffer cache (which sit above our block
43 struct request_queue *brd_queue;
44 struct gendisk *brd_disk;
45 #ifdef CONFIG_BLK_DEV_RAM_DAX
46 struct dax_device *dax_dev;
48 struct list_head brd_list;
51 * Backing store of pages and lock to protect it. This is the contents
52 * of the block device.
55 struct radix_tree_root brd_pages;
59 * Look up and return a brd's page for a given sector.
61 static DEFINE_MUTEX(brd_mutex);
62 static struct page *brd_lookup_page(struct brd_device *brd, sector_t sector)
68 * The page lifetime is protected by the fact that we have opened the
69 * device node -- brd pages will never be deleted under us, so we
70 * don't need any further locking or refcounting.
72 * This is strictly true for the radix-tree nodes as well (ie. we
73 * don't actually need the rcu_read_lock()), however that is not a
74 * documented feature of the radix-tree API so it is better to be
75 * safe here (we don't have total exclusion from radix tree updates
76 * here, only deletes).
79 idx = sector >> PAGE_SECTORS_SHIFT; /* sector to page index */
80 page = radix_tree_lookup(&brd->brd_pages, idx);
83 BUG_ON(page && page->index != idx);
89 * Look up and return a brd's page for a given sector.
90 * If one does not exist, allocate an empty page, and insert that. Then
93 static struct page *brd_insert_page(struct brd_device *brd, sector_t sector)
99 page = brd_lookup_page(brd, sector);
104 * Must use NOIO because we don't want to recurse back into the
105 * block or filesystem layers from page reclaim.
107 * Cannot support DAX and highmem, because our ->direct_access
108 * routine for DAX must return memory that is always addressable.
109 * If DAX was reworked to use pfns and kmap throughout, this
110 * restriction might be able to be lifted.
112 gfp_flags = GFP_NOIO | __GFP_ZERO;
113 #ifndef CONFIG_BLK_DEV_RAM_DAX
114 gfp_flags |= __GFP_HIGHMEM;
116 page = alloc_page(gfp_flags);
120 if (radix_tree_preload(GFP_NOIO)) {
125 spin_lock(&brd->brd_lock);
126 idx = sector >> PAGE_SECTORS_SHIFT;
128 if (radix_tree_insert(&brd->brd_pages, idx, page)) {
130 page = radix_tree_lookup(&brd->brd_pages, idx);
132 BUG_ON(page->index != idx);
134 spin_unlock(&brd->brd_lock);
136 radix_tree_preload_end();
142 * Free all backing store pages and radix tree. This must only be called when
143 * there are no other users of the device.
145 #define FREE_BATCH 16
146 static void brd_free_pages(struct brd_device *brd)
148 unsigned long pos = 0;
149 struct page *pages[FREE_BATCH];
155 nr_pages = radix_tree_gang_lookup(&brd->brd_pages,
156 (void **)pages, pos, FREE_BATCH);
158 for (i = 0; i < nr_pages; i++) {
161 BUG_ON(pages[i]->index < pos);
162 pos = pages[i]->index;
163 ret = radix_tree_delete(&brd->brd_pages, pos);
164 BUG_ON(!ret || ret != pages[i]);
165 __free_page(pages[i]);
171 * This assumes radix_tree_gang_lookup always returns as
172 * many pages as possible. If the radix-tree code changes,
173 * so will this have to.
175 } while (nr_pages == FREE_BATCH);
179 * copy_to_brd_setup must be called before copy_to_brd. It may sleep.
181 static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n)
183 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
186 copy = min_t(size_t, n, PAGE_SIZE - offset);
187 if (!brd_insert_page(brd, sector))
190 sector += copy >> SECTOR_SHIFT;
191 if (!brd_insert_page(brd, sector))
198 * Copy n bytes from src to the brd starting at sector. Does not sleep.
200 static void copy_to_brd(struct brd_device *brd, const void *src,
201 sector_t sector, size_t n)
205 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
208 copy = min_t(size_t, n, PAGE_SIZE - offset);
209 page = brd_lookup_page(brd, sector);
212 dst = kmap_atomic(page);
213 memcpy(dst + offset, src, copy);
218 sector += copy >> SECTOR_SHIFT;
220 page = brd_lookup_page(brd, sector);
223 dst = kmap_atomic(page);
224 memcpy(dst, src, copy);
230 * Copy n bytes to dst from the brd starting at sector. Does not sleep.
232 static void copy_from_brd(void *dst, struct brd_device *brd,
233 sector_t sector, size_t n)
237 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
240 copy = min_t(size_t, n, PAGE_SIZE - offset);
241 page = brd_lookup_page(brd, sector);
243 src = kmap_atomic(page);
244 memcpy(dst, src + offset, copy);
247 memset(dst, 0, copy);
251 sector += copy >> SECTOR_SHIFT;
253 page = brd_lookup_page(brd, sector);
255 src = kmap_atomic(page);
256 memcpy(dst, src, copy);
259 memset(dst, 0, copy);
264 * Process a single bvec of a bio.
266 static int brd_do_bvec(struct brd_device *brd, struct page *page,
267 unsigned int len, unsigned int off, bool is_write,
274 err = copy_to_brd_setup(brd, sector, len);
279 mem = kmap_atomic(page);
281 copy_from_brd(mem + off, brd, sector, len);
282 flush_dcache_page(page);
284 flush_dcache_page(page);
285 copy_to_brd(brd, mem + off, sector, len);
293 static blk_qc_t brd_make_request(struct request_queue *q, struct bio *bio)
295 struct block_device *bdev = bio->bi_bdev;
296 struct brd_device *brd = bdev->bd_disk->private_data;
299 struct bvec_iter iter;
301 sector = bio->bi_iter.bi_sector;
302 if (bio_end_sector(bio) > get_capacity(bdev->bd_disk))
305 bio_for_each_segment(bvec, bio, iter) {
306 unsigned int len = bvec.bv_len;
309 err = brd_do_bvec(brd, bvec.bv_page, len, bvec.bv_offset,
310 op_is_write(bio_op(bio)), sector);
313 sector += len >> SECTOR_SHIFT;
317 return BLK_QC_T_NONE;
320 return BLK_QC_T_NONE;
323 static int brd_rw_page(struct block_device *bdev, sector_t sector,
324 struct page *page, bool is_write)
326 struct brd_device *brd = bdev->bd_disk->private_data;
327 int err = brd_do_bvec(brd, page, PAGE_SIZE, 0, is_write, sector);
328 page_endio(page, is_write, err);
332 #ifdef CONFIG_BLK_DEV_RAM_DAX
333 static long __brd_direct_access(struct brd_device *brd, pgoff_t pgoff,
334 long nr_pages, void **kaddr, pfn_t *pfn)
340 page = brd_insert_page(brd, PFN_PHYS(pgoff) / 512);
343 *kaddr = page_address(page);
344 *pfn = page_to_pfn_t(page);
349 static long brd_dax_direct_access(struct dax_device *dax_dev,
350 pgoff_t pgoff, long nr_pages, void **kaddr, pfn_t *pfn)
352 struct brd_device *brd = dax_get_private(dax_dev);
354 return __brd_direct_access(brd, pgoff, nr_pages, kaddr, pfn);
357 static const struct dax_operations brd_dax_ops = {
358 .direct_access = brd_dax_direct_access,
362 static const struct block_device_operations brd_fops = {
363 .owner = THIS_MODULE,
364 .rw_page = brd_rw_page,
368 * And now the modules code and kernel interface.
370 static int rd_nr = CONFIG_BLK_DEV_RAM_COUNT;
371 module_param(rd_nr, int, S_IRUGO);
372 MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
374 unsigned long rd_size = CONFIG_BLK_DEV_RAM_SIZE;
375 module_param(rd_size, ulong, S_IRUGO);
376 MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
378 static int max_part = 1;
379 module_param(max_part, int, S_IRUGO);
380 MODULE_PARM_DESC(max_part, "Num Minors to reserve between devices");
382 MODULE_LICENSE("GPL");
383 MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
387 /* Legacy boot options - nonmodular */
388 static int __init ramdisk_size(char *str)
390 rd_size = simple_strtol(str, NULL, 0);
393 __setup("ramdisk_size=", ramdisk_size);
397 * The device scheme is derived from loop.c. Keep them in synch where possible
398 * (should share code eventually).
400 static LIST_HEAD(brd_devices);
401 static DEFINE_MUTEX(brd_devices_mutex);
403 static struct brd_device *brd_alloc(int i)
405 struct brd_device *brd;
406 struct gendisk *disk;
408 brd = kzalloc(sizeof(*brd), GFP_KERNEL);
412 spin_lock_init(&brd->brd_lock);
413 INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC);
415 brd->brd_queue = blk_alloc_queue(GFP_KERNEL);
419 blk_queue_make_request(brd->brd_queue, brd_make_request);
420 blk_queue_max_hw_sectors(brd->brd_queue, 1024);
421 blk_queue_bounce_limit(brd->brd_queue, BLK_BOUNCE_ANY);
423 /* This is so fdisk will align partitions on 4k, because of
424 * direct_access API needing 4k alignment, returning a PFN
425 * (This is only a problem on very small devices <= 4M,
426 * otherwise fdisk will align on 1M. Regardless this call
429 blk_queue_physical_block_size(brd->brd_queue, PAGE_SIZE);
430 disk = brd->brd_disk = alloc_disk(max_part);
433 disk->major = RAMDISK_MAJOR;
434 disk->first_minor = i * max_part;
435 disk->fops = &brd_fops;
436 disk->private_data = brd;
437 disk->queue = brd->brd_queue;
438 disk->flags = GENHD_FL_EXT_DEVT;
439 sprintf(disk->disk_name, "ram%d", i);
440 set_capacity(disk, rd_size * 2);
442 #ifdef CONFIG_BLK_DEV_RAM_DAX
443 queue_flag_set_unlocked(QUEUE_FLAG_DAX, brd->brd_queue);
444 brd->dax_dev = alloc_dax(brd, disk->disk_name, &brd_dax_ops);
452 #ifdef CONFIG_BLK_DEV_RAM_DAX
454 kill_dax(brd->dax_dev);
455 put_dax(brd->dax_dev);
458 blk_cleanup_queue(brd->brd_queue);
465 static void brd_free(struct brd_device *brd)
467 put_disk(brd->brd_disk);
468 blk_cleanup_queue(brd->brd_queue);
473 static struct brd_device *brd_init_one(int i, bool *new)
475 struct brd_device *brd;
478 list_for_each_entry(brd, &brd_devices, brd_list) {
479 if (brd->brd_number == i)
485 add_disk(brd->brd_disk);
486 list_add_tail(&brd->brd_list, &brd_devices);
493 static void brd_del_one(struct brd_device *brd)
495 list_del(&brd->brd_list);
496 #ifdef CONFIG_BLK_DEV_RAM_DAX
497 kill_dax(brd->dax_dev);
498 put_dax(brd->dax_dev);
500 del_gendisk(brd->brd_disk);
504 static struct kobject *brd_probe(dev_t dev, int *part, void *data)
506 struct brd_device *brd;
507 struct kobject *kobj;
510 mutex_lock(&brd_devices_mutex);
511 brd = brd_init_one(MINOR(dev) / max_part, &new);
512 kobj = brd ? get_disk(brd->brd_disk) : NULL;
513 mutex_unlock(&brd_devices_mutex);
521 static int __init brd_init(void)
523 struct brd_device *brd, *next;
527 * brd module now has a feature to instantiate underlying device
528 * structure on-demand, provided that there is an access dev node.
530 * (1) if rd_nr is specified, create that many upfront. else
531 * it defaults to CONFIG_BLK_DEV_RAM_COUNT
532 * (2) User can further extend brd devices by create dev node themselves
533 * and have kernel automatically instantiate actual device
534 * on-demand. Example:
535 * mknod /path/devnod_name b 1 X # 1 is the rd major
536 * fdisk -l /path/devnod_name
537 * If (X / max_part) was not already created it will be created
541 if (register_blkdev(RAMDISK_MAJOR, "ramdisk"))
544 if (unlikely(!max_part))
547 for (i = 0; i < rd_nr; i++) {
551 list_add_tail(&brd->brd_list, &brd_devices);
554 /* point of no return */
556 list_for_each_entry(brd, &brd_devices, brd_list)
557 add_disk(brd->brd_disk);
559 blk_register_region(MKDEV(RAMDISK_MAJOR, 0), 1UL << MINORBITS,
560 THIS_MODULE, brd_probe, NULL, NULL);
562 pr_info("brd: module loaded\n");
566 list_for_each_entry_safe(brd, next, &brd_devices, brd_list) {
567 list_del(&brd->brd_list);
570 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
572 pr_info("brd: module NOT loaded !!!\n");
576 static void __exit brd_exit(void)
578 struct brd_device *brd, *next;
580 list_for_each_entry_safe(brd, next, &brd_devices, brd_list)
583 blk_unregister_region(MKDEV(RAMDISK_MAJOR, 0), 1UL << MINORBITS);
584 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
586 pr_info("brd: module unloaded\n");
589 module_init(brd_init);
590 module_exit(brd_exit);