1/*
2 * Ram backed block device driver.
3 *
4 * Copyright (C) 2007 Nick Piggin
5 * Copyright (C) 2007 Novell Inc.
6 *
7 * Parts derived from drivers/block/rd.c, and drivers/block/loop.c, copyright
8 * of their respective owners.
9 */
10
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>
20#include <linux/fs.h>
21#include <linux/slab.h>
22
23#include <asm/uaccess.h>
24
25#define SECTOR_SHIFT		9
26#define PAGE_SECTORS_SHIFT	(PAGE_SHIFT - SECTOR_SHIFT)
27#define PAGE_SECTORS		(1 << PAGE_SECTORS_SHIFT)
28
29/*
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
34 * device).
35 */
36struct brd_device {
37	int		brd_number;
38
39	struct request_queue	*brd_queue;
40	struct gendisk		*brd_disk;
41	struct list_head	brd_list;
42
43	/*
44	 * Backing store of pages and lock to protect it. This is the contents
45	 * of the block device.
46	 */
47	spinlock_t		brd_lock;
48	struct radix_tree_root	brd_pages;
49};
50
51/*
52 * Look up and return a brd's page for a given sector.
53 */
54static DEFINE_MUTEX(brd_mutex);
55static struct page *brd_lookup_page(struct brd_device *brd, sector_t sector)
56{
57	pgoff_t idx;
58	struct page *page;
59
60	/*
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.
64	 *
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).
70	 */
71	rcu_read_lock();
72	idx = sector >> PAGE_SECTORS_SHIFT; /* sector to page index */
73	page = radix_tree_lookup(&brd->brd_pages, idx);
74	rcu_read_unlock();
75
76	BUG_ON(page && page->index != idx);
77
78	return page;
79}
80
81/*
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
84 * return it.
85 */
86static struct page *brd_insert_page(struct brd_device *brd, sector_t sector)
87{
88	pgoff_t idx;
89	struct page *page;
90	gfp_t gfp_flags;
91
92	page = brd_lookup_page(brd, sector);
93	if (page)
94		return page;
95
96	/*
97	 * Must use NOIO because we don't want to recurse back into the
98	 * block or filesystem layers from page reclaim.
99	 *
100	 * Cannot support XIP and highmem, because our ->direct_access
101	 * routine for XIP must return memory that is always addressable.
102	 * If XIP was reworked to use pfns and kmap throughout, this
103	 * restriction might be able to be lifted.
104	 */
105	gfp_flags = GFP_NOIO | __GFP_ZERO;
106#ifndef CONFIG_BLK_DEV_XIP
107	gfp_flags |= __GFP_HIGHMEM;
108#endif
109	page = alloc_page(gfp_flags);
110	if (!page)
111		return NULL;
112
113	if (radix_tree_preload(GFP_NOIO)) {
114		__free_page(page);
115		return NULL;
116	}
117
118	spin_lock(&brd->brd_lock);
119	idx = sector >> PAGE_SECTORS_SHIFT;
120	if (radix_tree_insert(&brd->brd_pages, idx, page)) {
121		__free_page(page);
122		page = radix_tree_lookup(&brd->brd_pages, idx);
123		BUG_ON(!page);
124		BUG_ON(page->index != idx);
125	} else
126		page->index = idx;
127	spin_unlock(&brd->brd_lock);
128
129	radix_tree_preload_end();
130
131	return page;
132}
133
134static void brd_free_page(struct brd_device *brd, sector_t sector)
135{
136	struct page *page;
137	pgoff_t idx;
138
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);
143	if (page)
144		__free_page(page);
145}
146
147static void brd_zero_page(struct brd_device *brd, sector_t sector)
148{
149	struct page *page;
150
151	page = brd_lookup_page(brd, sector);
152	if (page)
153		clear_highpage(page);
154}
155
156/*
157 * Free all backing store pages and radix tree. This must only be called when
158 * there are no other users of the device.
159 */
160#define FREE_BATCH 16
161static void brd_free_pages(struct brd_device *brd)
162{
163	unsigned long pos = 0;
164	struct page *pages[FREE_BATCH];
165	int nr_pages;
166
167	do {
168		int i;
169
170		nr_pages = radix_tree_gang_lookup(&brd->brd_pages,
171				(void **)pages, pos, FREE_BATCH);
172
173		for (i = 0; i < nr_pages; i++) {
174			void *ret;
175
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]);
181		}
182
183		pos++;
184
185		/*
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.
189		 */
190	} while (nr_pages == FREE_BATCH);
191}
192
193/*
194 * copy_to_brd_setup must be called before copy_to_brd. It may sleep.
195 */
196static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n)
197{
198	unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
199	size_t copy;
200
201	copy = min_t(size_t, n, PAGE_SIZE - offset);
202	if (!brd_insert_page(brd, sector))
203		return -ENOMEM;
204	if (copy < n) {
205		sector += copy >> SECTOR_SHIFT;
206		if (!brd_insert_page(brd, sector))
207			return -ENOMEM;
208	}
209	return 0;
210}
211
212static void discard_from_brd(struct brd_device *brd,
213			sector_t sector, size_t n)
214{
215	while (n >= PAGE_SIZE) {
216		/*
217		 * Don't want to actually discard pages here because
218		 * re-allocating the pages can result in writeback
219		 * deadlocks under heavy load.
220		 */
221		if (0)
222			brd_free_page(brd, sector);
223		else
224			brd_zero_page(brd, sector);
225		sector += PAGE_SIZE >> SECTOR_SHIFT;
226		n -= PAGE_SIZE;
227	}
228}
229
230/*
231 * Copy n bytes from src to the brd starting at sector. Does not sleep.
232 */
233static void copy_to_brd(struct brd_device *brd, const void *src,
234			sector_t sector, size_t n)
235{
236	struct page *page;
237	void *dst;
238	unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
239	size_t copy;
240
241	copy = min_t(size_t, n, PAGE_SIZE - offset);
242	page = brd_lookup_page(brd, sector);
243	BUG_ON(!page);
244
245	dst = kmap_atomic(page);
246	memcpy(dst + offset, src, copy);
247	kunmap_atomic(dst);
248
249	if (copy < n) {
250		src += copy;
251		sector += copy >> SECTOR_SHIFT;
252		copy = n - copy;
253		page = brd_lookup_page(brd, sector);
254		BUG_ON(!page);
255
256		dst = kmap_atomic(page);
257		memcpy(dst, src, copy);
258		kunmap_atomic(dst);
259	}
260}
261
262/*
263 * Copy n bytes to dst from the brd starting at sector. Does not sleep.
264 */
265static void copy_from_brd(void *dst, struct brd_device *brd,
266			sector_t sector, size_t n)
267{
268	struct page *page;
269	void *src;
270	unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
271	size_t copy;
272
273	copy = min_t(size_t, n, PAGE_SIZE - offset);
274	page = brd_lookup_page(brd, sector);
275	if (page) {
276		src = kmap_atomic(page);
277		memcpy(dst, src + offset, copy);
278		kunmap_atomic(src);
279	} else
280		memset(dst, 0, copy);
281
282	if (copy < n) {
283		dst += copy;
284		sector += copy >> SECTOR_SHIFT;
285		copy = n - copy;
286		page = brd_lookup_page(brd, sector);
287		if (page) {
288			src = kmap_atomic(page);
289			memcpy(dst, src, copy);
290			kunmap_atomic(src);
291		} else
292			memset(dst, 0, copy);
293	}
294}
295
296/*
297 * Process a single bvec of a bio.
298 */
299static int brd_do_bvec(struct brd_device *brd, struct page *page,
300			unsigned int len, unsigned int off, int rw,
301			sector_t sector)
302{
303	void *mem;
304	int err = 0;
305
306	if (rw != READ) {
307		err = copy_to_brd_setup(brd, sector, len);
308		if (err)
309			goto out;
310	}
311
312	mem = kmap_atomic(page);
313	if (rw == READ) {
314		copy_from_brd(mem + off, brd, sector, len);
315		flush_dcache_page(page);
316	} else {
317		flush_dcache_page(page);
318		copy_to_brd(brd, mem + off, sector, len);
319	}
320	kunmap_atomic(mem);
321
322out:
323	return err;
324}
325
326static void brd_make_request(struct request_queue *q, struct bio *bio)
327{
328	struct block_device *bdev = bio->bi_bdev;
329	struct brd_device *brd = bdev->bd_disk->private_data;
330	int rw;
331	struct bio_vec *bvec;
332	sector_t sector;
333	int i;
334	int err = -EIO;
335
336	sector = bio->bi_sector;
337	if (sector + (bio->bi_size >> SECTOR_SHIFT) >
338						get_capacity(bdev->bd_disk))
339		goto out;
340
341	if (unlikely(bio->bi_rw & REQ_DISCARD)) {
342		err = 0;
343		discard_from_brd(brd, sector, bio->bi_size);
344		goto out;
345	}
346
347	rw = bio_rw(bio);
348	if (rw == READA)
349		rw = READ;
350
351	bio_for_each_segment(bvec, bio, i) {
352		unsigned int len = bvec->bv_len;
353		err = brd_do_bvec(brd, bvec->bv_page, len,
354					bvec->bv_offset, rw, sector);
355		if (err)
356			break;
357		sector += len >> SECTOR_SHIFT;
358	}
359
360out:
361	bio_endio(bio, err);
362}
363
364#ifdef CONFIG_BLK_DEV_XIP
365static int brd_direct_access(struct block_device *bdev, sector_t sector,
366			void **kaddr, unsigned long *pfn)
367{
368	struct brd_device *brd = bdev->bd_disk->private_data;
369	struct page *page;
370
371	if (!brd)
372		return -ENODEV;
373	if (sector & (PAGE_SECTORS-1))
374		return -EINVAL;
375	if (sector + PAGE_SECTORS > get_capacity(bdev->bd_disk))
376		return -ERANGE;
377	page = brd_insert_page(brd, sector);
378	if (!page)
379		return -ENOMEM;
380	*kaddr = page_address(page);
381	*pfn = page_to_pfn(page);
382
383	return 0;
384}
385#endif
386
387static int brd_ioctl(struct block_device *bdev, fmode_t mode,
388			unsigned int cmd, unsigned long arg)
389{
390	int error;
391	struct brd_device *brd = bdev->bd_disk->private_data;
392
393	if (cmd != BLKFLSBUF)
394		return -ENOTTY;
395
396	/*
397	 * ram device BLKFLSBUF has special semantics, we want to actually
398	 * release and destroy the ramdisk data.
399	 */
400	mutex_lock(&brd_mutex);
401	mutex_lock(&bdev->bd_mutex);
402	error = -EBUSY;
403	if (bdev->bd_openers <= 1) {
404		/*
405		 * Kill the cache first, so it isn't written back to the
406		 * device.
407		 *
408		 * Another thread might instantiate more buffercache here,
409		 * but there is not much we can do to close that race.
410		 */
411		kill_bdev(bdev);
412		brd_free_pages(brd);
413		error = 0;
414	}
415	mutex_unlock(&bdev->bd_mutex);
416	mutex_unlock(&brd_mutex);
417
418	return error;
419}
420
421static const struct block_device_operations brd_fops = {
422	.owner =		THIS_MODULE,
423	.ioctl =		brd_ioctl,
424#ifdef CONFIG_BLK_DEV_XIP
425	.direct_access =	brd_direct_access,
426#endif
427};
428
429/*
430 * And now the modules code and kernel interface.
431 */
432static int rd_nr;
433int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
434static int max_part;
435static int part_shift;
436module_param(rd_nr, int, S_IRUGO);
437MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
438module_param(rd_size, int, S_IRUGO);
439MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
440module_param(max_part, int, S_IRUGO);
441MODULE_PARM_DESC(max_part, "Maximum number of partitions per RAM disk");
442MODULE_LICENSE("GPL");
443MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
444MODULE_ALIAS("rd");
445
446#ifndef MODULE
447/* Legacy boot options - nonmodular */
448static int __init ramdisk_size(char *str)
449{
450	rd_size = simple_strtol(str, NULL, 0);
451	return 1;
452}
453__setup("ramdisk_size=", ramdisk_size);
454#endif
455
456/*
457 * The device scheme is derived from loop.c. Keep them in synch where possible
458 * (should share code eventually).
459 */
460static LIST_HEAD(brd_devices);
461static DEFINE_MUTEX(brd_devices_mutex);
462
463static struct brd_device *brd_alloc(int i)
464{
465	struct brd_device *brd;
466	struct gendisk *disk;
467
468	brd = kzalloc(sizeof(*brd), GFP_KERNEL);
469	if (!brd)
470		goto out;
471	brd->brd_number		= i;
472	spin_lock_init(&brd->brd_lock);
473	INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC);
474
475	brd->brd_queue = blk_alloc_queue(GFP_KERNEL);
476	if (!brd->brd_queue)
477		goto out_free_dev;
478	blk_queue_make_request(brd->brd_queue, brd_make_request);
479	blk_queue_max_hw_sectors(brd->brd_queue, 1024);
480	blk_queue_bounce_limit(brd->brd_queue, BLK_BOUNCE_ANY);
481
482	brd->brd_queue->limits.discard_granularity = PAGE_SIZE;
483	brd->brd_queue->limits.max_discard_sectors = UINT_MAX;
484	brd->brd_queue->limits.discard_zeroes_data = 1;
485	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, brd->brd_queue);
486
487	disk = brd->brd_disk = alloc_disk(1 << part_shift);
488	if (!disk)
489		goto out_free_queue;
490	disk->major		= RAMDISK_MAJOR;
491	disk->first_minor	= i << part_shift;
492	disk->fops		= &brd_fops;
493	disk->private_data	= brd;
494	disk->queue		= brd->brd_queue;
495	disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
496	sprintf(disk->disk_name, "ram%d", i);
497	set_capacity(disk, rd_size * 2);
498
499	return brd;
500
501out_free_queue:
502	blk_cleanup_queue(brd->brd_queue);
503out_free_dev:
504	kfree(brd);
505out:
506	return NULL;
507}
508
509static void brd_free(struct brd_device *brd)
510{
511	put_disk(brd->brd_disk);
512	blk_cleanup_queue(brd->brd_queue);
513	brd_free_pages(brd);
514	kfree(brd);
515}
516
517static struct brd_device *brd_init_one(int i)
518{
519	struct brd_device *brd;
520
521	list_for_each_entry(brd, &brd_devices, brd_list) {
522		if (brd->brd_number == i)
523			goto out;
524	}
525
526	brd = brd_alloc(i);
527	if (brd) {
528		add_disk(brd->brd_disk);
529		list_add_tail(&brd->brd_list, &brd_devices);
530	}
531out:
532	return brd;
533}
534
535static void brd_del_one(struct brd_device *brd)
536{
537	list_del(&brd->brd_list);
538	del_gendisk(brd->brd_disk);
539	brd_free(brd);
540}
541
542static struct kobject *brd_probe(dev_t dev, int *part, void *data)
543{
544	struct brd_device *brd;
545	struct kobject *kobj;
546
547	mutex_lock(&brd_devices_mutex);
548	brd = brd_init_one(MINOR(dev) >> part_shift);
549	kobj = brd ? get_disk(brd->brd_disk) : ERR_PTR(-ENOMEM);
550	mutex_unlock(&brd_devices_mutex);
551
552	*part = 0;
553	return kobj;
554}
555
556static int __init brd_init(void)
557{
558	int i, nr;
559	unsigned long range;
560	struct brd_device *brd, *next;
561
562	/*
563	 * brd module now has a feature to instantiate underlying device
564	 * structure on-demand, provided that there is an access dev node.
565	 * However, this will not work well with user space tool that doesn't
566	 * know about such "feature".  In order to not break any existing
567	 * tool, we do the following:
568	 *
569	 * (1) if rd_nr is specified, create that many upfront, and this
570	 *     also becomes a hard limit.
571	 * (2) if rd_nr is not specified, create CONFIG_BLK_DEV_RAM_COUNT
572	 *     (default 16) rd device on module load, user can further
573	 *     extend brd device by create dev node themselves and have
574	 *     kernel automatically instantiate actual device on-demand.
575	 */
576
577	part_shift = 0;
578	if (max_part > 0) {
579		part_shift = fls(max_part);
580
581		/*
582		 * Adjust max_part according to part_shift as it is exported
583		 * to user space so that user can decide correct minor number
584		 * if [s]he want to create more devices.
585		 *
586		 * Note that -1 is required because partition 0 is reserved
587		 * for the whole disk.
588		 */
589		max_part = (1UL << part_shift) - 1;
590	}
591
592	if ((1UL << part_shift) > DISK_MAX_PARTS)
593		return -EINVAL;
594
595	if (rd_nr > 1UL << (MINORBITS - part_shift))
596		return -EINVAL;
597
598	if (rd_nr) {
599		nr = rd_nr;
600		range = rd_nr << part_shift;
601	} else {
602		nr = CONFIG_BLK_DEV_RAM_COUNT;
603		range = 1UL << MINORBITS;
604	}
605
606	if (register_blkdev(RAMDISK_MAJOR, "ramdisk"))
607		return -EIO;
608
609	for (i = 0; i < nr; i++) {
610		brd = brd_alloc(i);
611		if (!brd)
612			goto out_free;
613		list_add_tail(&brd->brd_list, &brd_devices);
614	}
615
616	/* point of no return */
617
618	list_for_each_entry(brd, &brd_devices, brd_list)
619		add_disk(brd->brd_disk);
620
621	blk_register_region(MKDEV(RAMDISK_MAJOR, 0), range,
622				  THIS_MODULE, brd_probe, NULL, NULL);
623
624	printk(KERN_INFO "brd: module loaded\n");
625	return 0;
626
627out_free:
628	list_for_each_entry_safe(brd, next, &brd_devices, brd_list) {
629		list_del(&brd->brd_list);
630		brd_free(brd);
631	}
632	unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
633
634	return -ENOMEM;
635}
636
637static void __exit brd_exit(void)
638{
639	unsigned long range;
640	struct brd_device *brd, *next;
641
642	range = rd_nr ? rd_nr << part_shift : 1UL << MINORBITS;
643
644	list_for_each_entry_safe(brd, next, &brd_devices, brd_list)
645		brd_del_one(brd);
646
647	blk_unregister_region(MKDEV(RAMDISK_MAJOR, 0), range);
648	unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
649}
650
651module_init(brd_init);
652module_exit(brd_exit);
653
654