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/initrd.h>
13 #include <linux/module.h>
14 #include <linux/moduleparam.h>
15 #include <linux/major.h>
16 #include <linux/blkdev.h>
17 #include <linux/bio.h>
18 #include <linux/highmem.h>
19 #include <linux/mutex.h>
20 #include <linux/radix-tree.h>
22 #include <linux/slab.h>
23 #ifdef CONFIG_BLK_DEV_RAM_DAX
24 #include <linux/pfn_t.h>
25 #include <linux/dax.h>
26 #include <linux/uio.h>
29 #include <linux/uaccess.h>
31 #define SECTOR_SHIFT 9
32 #define PAGE_SECTORS_SHIFT (PAGE_SHIFT - SECTOR_SHIFT)
33 #define PAGE_SECTORS (1 << PAGE_SECTORS_SHIFT)
36 * Each block ramdisk device has a radix_tree brd_pages of pages that stores
37 * the pages containing the block device's contents. A brd page's ->index is
38 * its offset in PAGE_SIZE units. This is similar to, but in no way connected
39 * with, the kernel's pagecache or buffer cache (which sit above our block
45 struct request_queue *brd_queue;
46 struct gendisk *brd_disk;
47 #ifdef CONFIG_BLK_DEV_RAM_DAX
48 struct dax_device *dax_dev;
50 struct list_head brd_list;
53 * Backing store of pages and lock to protect it. This is the contents
54 * of the block device.
57 struct radix_tree_root brd_pages;
61 * Look up and return a brd's page for a given sector.
63 static struct page *brd_lookup_page(struct brd_device *brd, sector_t sector)
69 * The page lifetime is protected by the fact that we have opened the
70 * device node -- brd pages will never be deleted under us, so we
71 * don't need any further locking or refcounting.
73 * This is strictly true for the radix-tree nodes as well (ie. we
74 * don't actually need the rcu_read_lock()), however that is not a
75 * documented feature of the radix-tree API so it is better to be
76 * safe here (we don't have total exclusion from radix tree updates
77 * here, only deletes).
80 idx = sector >> PAGE_SECTORS_SHIFT; /* sector to page index */
81 page = radix_tree_lookup(&brd->brd_pages, idx);
84 BUG_ON(page && page->index != idx);
90 * Look up and return a brd's page for a given sector.
91 * If one does not exist, allocate an empty page, and insert that. Then
94 static struct page *brd_insert_page(struct brd_device *brd, sector_t sector)
100 page = brd_lookup_page(brd, sector);
105 * Must use NOIO because we don't want to recurse back into the
106 * block or filesystem layers from page reclaim.
108 * Cannot support DAX and highmem, because our ->direct_access
109 * routine for DAX must return memory that is always addressable.
110 * If DAX was reworked to use pfns and kmap throughout, this
111 * restriction might be able to be lifted.
113 gfp_flags = GFP_NOIO | __GFP_ZERO;
114 #ifndef CONFIG_BLK_DEV_RAM_DAX
115 gfp_flags |= __GFP_HIGHMEM;
117 page = alloc_page(gfp_flags);
121 if (radix_tree_preload(GFP_NOIO)) {
126 spin_lock(&brd->brd_lock);
127 idx = sector >> PAGE_SECTORS_SHIFT;
129 if (radix_tree_insert(&brd->brd_pages, idx, page)) {
131 page = radix_tree_lookup(&brd->brd_pages, idx);
133 BUG_ON(page->index != idx);
135 spin_unlock(&brd->brd_lock);
137 radix_tree_preload_end();
143 * Free all backing store pages and radix tree. This must only be called when
144 * there are no other users of the device.
146 #define FREE_BATCH 16
147 static void brd_free_pages(struct brd_device *brd)
149 unsigned long pos = 0;
150 struct page *pages[FREE_BATCH];
156 nr_pages = radix_tree_gang_lookup(&brd->brd_pages,
157 (void **)pages, pos, FREE_BATCH);
159 for (i = 0; i < nr_pages; i++) {
162 BUG_ON(pages[i]->index < pos);
163 pos = pages[i]->index;
164 ret = radix_tree_delete(&brd->brd_pages, pos);
165 BUG_ON(!ret || ret != pages[i]);
166 __free_page(pages[i]);
172 * This assumes radix_tree_gang_lookup always returns as
173 * many pages as possible. If the radix-tree code changes,
174 * so will this have to.
176 } while (nr_pages == FREE_BATCH);
180 * copy_to_brd_setup must be called before copy_to_brd. It may sleep.
182 static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n)
184 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
187 copy = min_t(size_t, n, PAGE_SIZE - offset);
188 if (!brd_insert_page(brd, sector))
191 sector += copy >> SECTOR_SHIFT;
192 if (!brd_insert_page(brd, sector))
199 * Copy n bytes from src to the brd starting at sector. Does not sleep.
201 static void copy_to_brd(struct brd_device *brd, const void *src,
202 sector_t sector, size_t n)
206 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
209 copy = min_t(size_t, n, PAGE_SIZE - offset);
210 page = brd_lookup_page(brd, sector);
213 dst = kmap_atomic(page);
214 memcpy(dst + offset, src, copy);
219 sector += copy >> SECTOR_SHIFT;
221 page = brd_lookup_page(brd, sector);
224 dst = kmap_atomic(page);
225 memcpy(dst, src, copy);
231 * Copy n bytes to dst from the brd starting at sector. Does not sleep.
233 static void copy_from_brd(void *dst, struct brd_device *brd,
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);
244 src = kmap_atomic(page);
245 memcpy(dst, src + offset, copy);
248 memset(dst, 0, copy);
252 sector += copy >> SECTOR_SHIFT;
254 page = brd_lookup_page(brd, sector);
256 src = kmap_atomic(page);
257 memcpy(dst, src, copy);
260 memset(dst, 0, copy);
265 * Process a single bvec of a bio.
267 static int brd_do_bvec(struct brd_device *brd, struct page *page,
268 unsigned int len, unsigned int off, bool is_write,
275 err = copy_to_brd_setup(brd, sector, len);
280 mem = kmap_atomic(page);
282 copy_from_brd(mem + off, brd, sector, len);
283 flush_dcache_page(page);
285 flush_dcache_page(page);
286 copy_to_brd(brd, mem + off, sector, len);
294 static blk_qc_t brd_make_request(struct request_queue *q, struct bio *bio)
296 struct brd_device *brd = bio->bi_disk->private_data;
299 struct bvec_iter iter;
301 sector = bio->bi_iter.bi_sector;
302 if (bio_end_sector(bio) > get_capacity(bio->bi_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;
329 if (PageTransHuge(page))
331 err = brd_do_bvec(brd, page, PAGE_SIZE, 0, is_write, sector);
332 page_endio(page, is_write, err);
336 #ifdef CONFIG_BLK_DEV_RAM_DAX
337 static long __brd_direct_access(struct brd_device *brd, pgoff_t pgoff,
338 long nr_pages, void **kaddr, pfn_t *pfn)
344 page = brd_insert_page(brd, (sector_t)pgoff << PAGE_SECTORS_SHIFT);
347 *kaddr = page_address(page);
348 *pfn = page_to_pfn_t(page);
353 static long brd_dax_direct_access(struct dax_device *dax_dev,
354 pgoff_t pgoff, long nr_pages, void **kaddr, pfn_t *pfn)
356 struct brd_device *brd = dax_get_private(dax_dev);
358 return __brd_direct_access(brd, pgoff, nr_pages, kaddr, pfn);
361 static size_t brd_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
362 void *addr, size_t bytes, struct iov_iter *i)
364 return copy_from_iter(addr, bytes, i);
367 static const struct dax_operations brd_dax_ops = {
368 .direct_access = brd_dax_direct_access,
369 .copy_from_iter = brd_dax_copy_from_iter,
373 static const struct block_device_operations brd_fops = {
374 .owner = THIS_MODULE,
375 .rw_page = brd_rw_page,
379 * And now the modules code and kernel interface.
381 static int rd_nr = CONFIG_BLK_DEV_RAM_COUNT;
382 module_param(rd_nr, int, S_IRUGO);
383 MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
385 unsigned long rd_size = CONFIG_BLK_DEV_RAM_SIZE;
386 module_param(rd_size, ulong, S_IRUGO);
387 MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
389 static int max_part = 1;
390 module_param(max_part, int, S_IRUGO);
391 MODULE_PARM_DESC(max_part, "Num Minors to reserve between devices");
393 MODULE_LICENSE("GPL");
394 MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
398 /* Legacy boot options - nonmodular */
399 static int __init ramdisk_size(char *str)
401 rd_size = simple_strtol(str, NULL, 0);
404 __setup("ramdisk_size=", ramdisk_size);
408 * The device scheme is derived from loop.c. Keep them in synch where possible
409 * (should share code eventually).
411 static LIST_HEAD(brd_devices);
412 static DEFINE_MUTEX(brd_devices_mutex);
414 static struct brd_device *brd_alloc(int i)
416 struct brd_device *brd;
417 struct gendisk *disk;
419 brd = kzalloc(sizeof(*brd), GFP_KERNEL);
423 spin_lock_init(&brd->brd_lock);
424 INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC);
426 brd->brd_queue = blk_alloc_queue(GFP_KERNEL);
430 blk_queue_make_request(brd->brd_queue, brd_make_request);
431 blk_queue_max_hw_sectors(brd->brd_queue, 1024);
433 /* This is so fdisk will align partitions on 4k, because of
434 * direct_access API needing 4k alignment, returning a PFN
435 * (This is only a problem on very small devices <= 4M,
436 * otherwise fdisk will align on 1M. Regardless this call
439 blk_queue_physical_block_size(brd->brd_queue, PAGE_SIZE);
440 disk = brd->brd_disk = alloc_disk(max_part);
443 disk->major = RAMDISK_MAJOR;
444 disk->first_minor = i * max_part;
445 disk->fops = &brd_fops;
446 disk->private_data = brd;
447 disk->queue = brd->brd_queue;
448 disk->flags = GENHD_FL_EXT_DEVT;
449 sprintf(disk->disk_name, "ram%d", i);
450 set_capacity(disk, rd_size * 2);
452 #ifdef CONFIG_BLK_DEV_RAM_DAX
453 queue_flag_set_unlocked(QUEUE_FLAG_DAX, brd->brd_queue);
454 brd->dax_dev = alloc_dax(brd, disk->disk_name, &brd_dax_ops);
462 #ifdef CONFIG_BLK_DEV_RAM_DAX
464 kill_dax(brd->dax_dev);
465 put_dax(brd->dax_dev);
468 blk_cleanup_queue(brd->brd_queue);
475 static void brd_free(struct brd_device *brd)
477 put_disk(brd->brd_disk);
478 blk_cleanup_queue(brd->brd_queue);
483 static struct brd_device *brd_init_one(int i, bool *new)
485 struct brd_device *brd;
488 list_for_each_entry(brd, &brd_devices, brd_list) {
489 if (brd->brd_number == i)
495 add_disk(brd->brd_disk);
496 list_add_tail(&brd->brd_list, &brd_devices);
503 static void brd_del_one(struct brd_device *brd)
505 list_del(&brd->brd_list);
506 #ifdef CONFIG_BLK_DEV_RAM_DAX
507 kill_dax(brd->dax_dev);
508 put_dax(brd->dax_dev);
510 del_gendisk(brd->brd_disk);
514 static struct kobject *brd_probe(dev_t dev, int *part, void *data)
516 struct brd_device *brd;
517 struct kobject *kobj;
520 mutex_lock(&brd_devices_mutex);
521 brd = brd_init_one(MINOR(dev) / max_part, &new);
522 kobj = brd ? get_disk(brd->brd_disk) : NULL;
523 mutex_unlock(&brd_devices_mutex);
531 static int __init brd_init(void)
533 struct brd_device *brd, *next;
537 * brd module now has a feature to instantiate underlying device
538 * structure on-demand, provided that there is an access dev node.
540 * (1) if rd_nr is specified, create that many upfront. else
541 * it defaults to CONFIG_BLK_DEV_RAM_COUNT
542 * (2) User can further extend brd devices by create dev node themselves
543 * and have kernel automatically instantiate actual device
544 * on-demand. Example:
545 * mknod /path/devnod_name b 1 X # 1 is the rd major
546 * fdisk -l /path/devnod_name
547 * If (X / max_part) was not already created it will be created
551 if (register_blkdev(RAMDISK_MAJOR, "ramdisk"))
554 if (unlikely(!max_part))
557 for (i = 0; i < rd_nr; i++) {
561 list_add_tail(&brd->brd_list, &brd_devices);
564 /* point of no return */
566 list_for_each_entry(brd, &brd_devices, brd_list)
567 add_disk(brd->brd_disk);
569 blk_register_region(MKDEV(RAMDISK_MAJOR, 0), 1UL << MINORBITS,
570 THIS_MODULE, brd_probe, NULL, NULL);
572 pr_info("brd: module loaded\n");
576 list_for_each_entry_safe(brd, next, &brd_devices, brd_list) {
577 list_del(&brd->brd_list);
580 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
582 pr_info("brd: module NOT loaded !!!\n");
586 static void __exit brd_exit(void)
588 struct brd_device *brd, *next;
590 list_for_each_entry_safe(brd, next, &brd_devices, brd_list)
593 blk_unregister_region(MKDEV(RAMDISK_MAJOR, 0), 1UL << MINORBITS);
594 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
596 pr_info("brd: module unloaded\n");
599 module_init(brd_init);
600 module_exit(brd_exit);