Linux Audio

Check our new training course

Loading...
v3.15
 
  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	page->index = idx;
121	if (radix_tree_insert(&brd->brd_pages, idx, page)) {
122		__free_page(page);
123		page = radix_tree_lookup(&brd->brd_pages, idx);
124		BUG_ON(!page);
125		BUG_ON(page->index != idx);
126	}
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	struct bvec_iter iter;
334	int err = -EIO;
335
336	sector = bio->bi_iter.bi_sector;
337	if (bio_end_sector(bio) > get_capacity(bdev->bd_disk))
338		goto out;
339
340	if (unlikely(bio->bi_rw & REQ_DISCARD)) {
341		err = 0;
342		discard_from_brd(brd, sector, bio->bi_iter.bi_size);
343		goto out;
344	}
345
346	rw = bio_rw(bio);
347	if (rw == READA)
348		rw = READ;
349
350	bio_for_each_segment(bvec, bio, iter) {
351		unsigned int len = bvec.bv_len;
352		err = brd_do_bvec(brd, bvec.bv_page, len,
353					bvec.bv_offset, rw, sector);
354		if (err)
355			break;
356		sector += len >> SECTOR_SHIFT;
 
 
 
357	}
358
359out:
360	bio_endio(bio, err);
361}
362
363#ifdef CONFIG_BLK_DEV_XIP
364static int brd_direct_access(struct block_device *bdev, sector_t sector,
365			void **kaddr, unsigned long *pfn)
366{
367	struct brd_device *brd = bdev->bd_disk->private_data;
368	struct page *page;
369
370	if (!brd)
371		return -ENODEV;
372	if (sector & (PAGE_SECTORS-1))
373		return -EINVAL;
374	if (sector + PAGE_SECTORS > get_capacity(bdev->bd_disk))
375		return -ERANGE;
376	page = brd_insert_page(brd, sector);
377	if (!page)
378		return -ENOMEM;
379	*kaddr = page_address(page);
380	*pfn = page_to_pfn(page);
381
382	return 0;
383}
384#endif
385
386static int brd_ioctl(struct block_device *bdev, fmode_t mode,
387			unsigned int cmd, unsigned long arg)
388{
389	int error;
390	struct brd_device *brd = bdev->bd_disk->private_data;
391
392	if (cmd != BLKFLSBUF)
393		return -ENOTTY;
 
394
395	/*
396	 * ram device BLKFLSBUF has special semantics, we want to actually
397	 * release and destroy the ramdisk data.
398	 */
399	mutex_lock(&brd_mutex);
400	mutex_lock(&bdev->bd_mutex);
401	error = -EBUSY;
402	if (bdev->bd_openers <= 1) {
403		/*
404		 * Kill the cache first, so it isn't written back to the
405		 * device.
406		 *
407		 * Another thread might instantiate more buffercache here,
408		 * but there is not much we can do to close that race.
409		 */
410		kill_bdev(bdev);
411		brd_free_pages(brd);
412		error = 0;
413	}
414	mutex_unlock(&bdev->bd_mutex);
415	mutex_unlock(&brd_mutex);
416
417	return error;
418}
419
420static const struct block_device_operations brd_fops = {
421	.owner =		THIS_MODULE,
422	.ioctl =		brd_ioctl,
423#ifdef CONFIG_BLK_DEV_XIP
424	.direct_access =	brd_direct_access,
425#endif
426};
427
428/*
429 * And now the modules code and kernel interface.
430 */
431static int rd_nr;
432int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
433static int max_part;
434static int part_shift;
435module_param(rd_nr, int, S_IRUGO);
436MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
437module_param(rd_size, int, S_IRUGO);
 
 
438MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
439module_param(max_part, int, S_IRUGO);
440MODULE_PARM_DESC(max_part, "Maximum number of partitions per RAM disk");
 
 
 
 
441MODULE_LICENSE("GPL");
442MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
443MODULE_ALIAS("rd");
444
445#ifndef MODULE
446/* Legacy boot options - nonmodular */
447static int __init ramdisk_size(char *str)
448{
449	rd_size = simple_strtol(str, NULL, 0);
450	return 1;
451}
452__setup("ramdisk_size=", ramdisk_size);
453#endif
454
455/*
456 * The device scheme is derived from loop.c. Keep them in synch where possible
457 * (should share code eventually).
458 */
459static LIST_HEAD(brd_devices);
460static DEFINE_MUTEX(brd_devices_mutex);
 
461
462static struct brd_device *brd_alloc(int i)
463{
464	struct brd_device *brd;
465	struct gendisk *disk;
 
 
 
 
 
 
 
466
467	brd = kzalloc(sizeof(*brd), GFP_KERNEL);
468	if (!brd)
469		goto out;
470	brd->brd_number		= i;
471	spin_lock_init(&brd->brd_lock);
472	INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC);
 
 
 
 
 
 
 
 
 
 
 
 
473
474	brd->brd_queue = blk_alloc_queue(GFP_KERNEL);
475	if (!brd->brd_queue)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
476		goto out_free_dev;
477	blk_queue_make_request(brd->brd_queue, brd_make_request);
478	blk_queue_max_hw_sectors(brd->brd_queue, 1024);
479	blk_queue_bounce_limit(brd->brd_queue, BLK_BOUNCE_ANY);
480
481	brd->brd_queue->limits.discard_granularity = PAGE_SIZE;
482	brd->brd_queue->limits.max_discard_sectors = UINT_MAX;
483	brd->brd_queue->limits.discard_zeroes_data = 1;
484	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, brd->brd_queue);
485
486	disk = brd->brd_disk = alloc_disk(1 << part_shift);
487	if (!disk)
488		goto out_free_queue;
489	disk->major		= RAMDISK_MAJOR;
490	disk->first_minor	= i << part_shift;
 
491	disk->fops		= &brd_fops;
492	disk->private_data	= brd;
493	disk->queue		= brd->brd_queue;
494	disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
495	sprintf(disk->disk_name, "ram%d", i);
496	set_capacity(disk, rd_size * 2);
 
 
 
 
497
498	return brd;
499
500out_free_queue:
501	blk_cleanup_queue(brd->brd_queue);
502out_free_dev:
503	kfree(brd);
504out:
505	return NULL;
506}
507
508static void brd_free(struct brd_device *brd)
509{
510	put_disk(brd->brd_disk);
511	blk_cleanup_queue(brd->brd_queue);
512	brd_free_pages(brd);
513	kfree(brd);
514}
515
516static struct brd_device *brd_init_one(int i)
517{
518	struct brd_device *brd;
519
520	list_for_each_entry(brd, &brd_devices, brd_list) {
521		if (brd->brd_number == i)
522			goto out;
523	}
524
525	brd = brd_alloc(i);
526	if (brd) {
527		add_disk(brd->brd_disk);
528		list_add_tail(&brd->brd_list, &brd_devices);
 
529	}
530out:
531	return brd;
532}
533
534static void brd_del_one(struct brd_device *brd)
535{
536	list_del(&brd->brd_list);
537	del_gendisk(brd->brd_disk);
538	brd_free(brd);
539}
540
541static struct kobject *brd_probe(dev_t dev, int *part, void *data)
542{
543	struct brd_device *brd;
544	struct kobject *kobj;
545
546	mutex_lock(&brd_devices_mutex);
547	brd = brd_init_one(MINOR(dev) >> part_shift);
548	kobj = brd ? get_disk(brd->brd_disk) : NULL;
549	mutex_unlock(&brd_devices_mutex);
 
 
550
551	*part = 0;
552	return kobj;
 
 
 
553}
554
555static int __init brd_init(void)
556{
557	int i, nr;
558	unsigned long range;
559	struct brd_device *brd, *next;
560
561	/*
562	 * brd module now has a feature to instantiate underlying device
563	 * structure on-demand, provided that there is an access dev node.
564	 * However, this will not work well with user space tool that doesn't
565	 * know about such "feature".  In order to not break any existing
566	 * tool, we do the following:
567	 *
568	 * (1) if rd_nr is specified, create that many upfront, and this
569	 *     also becomes a hard limit.
570	 * (2) if rd_nr is not specified, create CONFIG_BLK_DEV_RAM_COUNT
571	 *     (default 16) rd device on module load, user can further
572	 *     extend brd device by create dev node themselves and have
573	 *     kernel automatically instantiate actual device on-demand.
 
 
 
574	 */
575
576	part_shift = 0;
577	if (max_part > 0) {
578		part_shift = fls(max_part);
579
580		/*
581		 * Adjust max_part according to part_shift as it is exported
582		 * to user space so that user can decide correct minor number
583		 * if [s]he want to create more devices.
584		 *
585		 * Note that -1 is required because partition 0 is reserved
586		 * for the whole disk.
587		 */
588		max_part = (1UL << part_shift) - 1;
589	}
590
591	if ((1UL << part_shift) > DISK_MAX_PARTS)
592		return -EINVAL;
593
594	if (rd_nr > 1UL << (MINORBITS - part_shift))
595		return -EINVAL;
596
597	if (rd_nr) {
598		nr = rd_nr;
599		range = rd_nr << part_shift;
600	} else {
601		nr = CONFIG_BLK_DEV_RAM_COUNT;
602		range = 1UL << MINORBITS;
603	}
604
605	if (register_blkdev(RAMDISK_MAJOR, "ramdisk"))
606		return -EIO;
607
608	for (i = 0; i < nr; i++) {
609		brd = brd_alloc(i);
610		if (!brd)
611			goto out_free;
612		list_add_tail(&brd->brd_list, &brd_devices);
613	}
614
615	/* point of no return */
 
616
617	list_for_each_entry(brd, &brd_devices, brd_list)
618		add_disk(brd->brd_disk);
619
620	blk_register_region(MKDEV(RAMDISK_MAJOR, 0), range,
621				  THIS_MODULE, brd_probe, NULL, NULL);
622
623	printk(KERN_INFO "brd: module loaded\n");
624	return 0;
625
626out_free:
627	list_for_each_entry_safe(brd, next, &brd_devices, brd_list) {
628		list_del(&brd->brd_list);
629		brd_free(brd);
630	}
631	unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
632
633	return -ENOMEM;
 
634}
635
636static void __exit brd_exit(void)
637{
638	unsigned long range;
639	struct brd_device *brd, *next;
640
641	range = rd_nr ? rd_nr << part_shift : 1UL << MINORBITS;
642
643	list_for_each_entry_safe(brd, next, &brd_devices, brd_list)
644		brd_del_one(brd);
645
646	blk_unregister_region(MKDEV(RAMDISK_MAJOR, 0), range);
647	unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
 
 
 
648}
649
650module_init(brd_init);
651module_exit(brd_exit);
652
v6.13.7
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Ram backed block device driver.
  4 *
  5 * Copyright (C) 2007 Nick Piggin
  6 * Copyright (C) 2007 Novell Inc.
  7 *
  8 * Parts derived from drivers/block/rd.c, and drivers/block/loop.c, copyright
  9 * of their respective owners.
 10 */
 11
 12#include <linux/init.h>
 13#include <linux/initrd.h>
 14#include <linux/module.h>
 15#include <linux/moduleparam.h>
 16#include <linux/major.h>
 17#include <linux/blkdev.h>
 18#include <linux/bio.h>
 19#include <linux/highmem.h>
 20#include <linux/mutex.h>
 21#include <linux/pagemap.h>
 22#include <linux/xarray.h>
 23#include <linux/fs.h>
 24#include <linux/slab.h>
 25#include <linux/backing-dev.h>
 26#include <linux/debugfs.h>
 27
 28#include <linux/uaccess.h>
 
 
 
 
 29
 30/*
 31 * Each block ramdisk device has a xarray brd_pages of pages that stores
 32 * the pages containing the block device's contents.
 
 
 
 33 */
 34struct brd_device {
 35	int			brd_number;
 
 
 36	struct gendisk		*brd_disk;
 37	struct list_head	brd_list;
 38
 39	/*
 40	 * Backing store of pages. This is the contents of the block device.
 
 41	 */
 42	struct xarray	        brd_pages;
 43	u64			brd_nr_pages;
 44};
 45
 46/*
 47 * Look up and return a brd's page for a given sector.
 48 */
 
 49static struct page *brd_lookup_page(struct brd_device *brd, sector_t sector)
 50{
 51	return xa_load(&brd->brd_pages, sector >> PAGE_SECTORS_SHIFT);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 52}
 53
 54/*
 55 * Insert a new page for a given sector, if one does not already exist.
 
 
 56 */
 57static int brd_insert_page(struct brd_device *brd, sector_t sector, gfp_t gfp)
 58{
 59	pgoff_t idx = sector >> PAGE_SECTORS_SHIFT;
 60	struct page *page;
 61	int ret = 0;
 62
 63	page = brd_lookup_page(brd, sector);
 64	if (page)
 65		return 0;
 66
 67	page = alloc_page(gfp | __GFP_ZERO | __GFP_HIGHMEM);
 
 
 
 
 
 
 
 
 
 
 
 
 
 68	if (!page)
 69		return -ENOMEM;
 70
 71	xa_lock(&brd->brd_pages);
 72	ret = __xa_insert(&brd->brd_pages, idx, page, gfp);
 73	if (!ret)
 74		brd->brd_nr_pages++;
 75	xa_unlock(&brd->brd_pages);
 76
 77	if (ret < 0) {
 
 
 
 78		__free_page(page);
 79		if (ret == -EBUSY)
 80			ret = 0;
 
 81	}
 82	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 83}
 84
 85/*
 86 * Free all backing store pages and xarray. This must only be called when
 87 * there are no other users of the device.
 88 */
 
 89static void brd_free_pages(struct brd_device *brd)
 90{
 91	struct page *page;
 92	pgoff_t idx;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 93
 94	xa_for_each(&brd->brd_pages, idx, page) {
 95		__free_page(page);
 96		cond_resched();
 97	}
 98
 99	xa_destroy(&brd->brd_pages);
 
 
 
 
 
100}
101
102/*
103 * copy_to_brd_setup must be called before copy_to_brd. It may sleep.
104 */
105static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n,
106			     gfp_t gfp)
107{
108	unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
109	size_t copy;
110	int ret;
111
112	copy = min_t(size_t, n, PAGE_SIZE - offset);
113	ret = brd_insert_page(brd, sector, gfp);
114	if (ret)
115		return ret;
116	if (copy < n) {
117		sector += copy >> SECTOR_SHIFT;
118		ret = brd_insert_page(brd, sector, gfp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
119	}
120	return ret;
121}
122
123/*
124 * Copy n bytes from src to the brd starting at sector. Does not sleep.
125 */
126static void copy_to_brd(struct brd_device *brd, const void *src,
127			sector_t sector, size_t n)
128{
129	struct page *page;
130	void *dst;
131	unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
132	size_t copy;
133
134	copy = min_t(size_t, n, PAGE_SIZE - offset);
135	page = brd_lookup_page(brd, sector);
136	BUG_ON(!page);
137
138	dst = kmap_atomic(page);
139	memcpy(dst + offset, src, copy);
140	kunmap_atomic(dst);
141
142	if (copy < n) {
143		src += copy;
144		sector += copy >> SECTOR_SHIFT;
145		copy = n - copy;
146		page = brd_lookup_page(brd, sector);
147		BUG_ON(!page);
148
149		dst = kmap_atomic(page);
150		memcpy(dst, src, copy);
151		kunmap_atomic(dst);
152	}
153}
154
155/*
156 * Copy n bytes to dst from the brd starting at sector. Does not sleep.
157 */
158static void copy_from_brd(void *dst, struct brd_device *brd,
159			sector_t sector, size_t n)
160{
161	struct page *page;
162	void *src;
163	unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
164	size_t copy;
165
166	copy = min_t(size_t, n, PAGE_SIZE - offset);
167	page = brd_lookup_page(brd, sector);
168	if (page) {
169		src = kmap_atomic(page);
170		memcpy(dst, src + offset, copy);
171		kunmap_atomic(src);
172	} else
173		memset(dst, 0, copy);
174
175	if (copy < n) {
176		dst += copy;
177		sector += copy >> SECTOR_SHIFT;
178		copy = n - copy;
179		page = brd_lookup_page(brd, sector);
180		if (page) {
181			src = kmap_atomic(page);
182			memcpy(dst, src, copy);
183			kunmap_atomic(src);
184		} else
185			memset(dst, 0, copy);
186	}
187}
188
189/*
190 * Process a single bvec of a bio.
191 */
192static int brd_do_bvec(struct brd_device *brd, struct page *page,
193			unsigned int len, unsigned int off, blk_opf_t opf,
194			sector_t sector)
195{
196	void *mem;
197	int err = 0;
198
199	if (op_is_write(opf)) {
200		/*
201		 * Must use NOIO because we don't want to recurse back into the
202		 * block or filesystem layers from page reclaim.
203		 */
204		gfp_t gfp = opf & REQ_NOWAIT ? GFP_NOWAIT : GFP_NOIO;
205
206		err = copy_to_brd_setup(brd, sector, len, gfp);
207		if (err)
208			goto out;
209	}
210
211	mem = kmap_atomic(page);
212	if (!op_is_write(opf)) {
213		copy_from_brd(mem + off, brd, sector, len);
214		flush_dcache_page(page);
215	} else {
216		flush_dcache_page(page);
217		copy_to_brd(brd, mem + off, sector, len);
218	}
219	kunmap_atomic(mem);
220
221out:
222	return err;
223}
224
225static void brd_do_discard(struct brd_device *brd, sector_t sector, u32 size)
226{
227	sector_t aligned_sector = (sector + PAGE_SECTORS) & ~PAGE_SECTORS;
228	struct page *page;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
229
230	size -= (aligned_sector - sector) * SECTOR_SIZE;
231	xa_lock(&brd->brd_pages);
232	while (size >= PAGE_SIZE && aligned_sector < rd_size * 2) {
233		page = __xa_erase(&brd->brd_pages, aligned_sector >> PAGE_SECTORS_SHIFT);
234		if (page) {
235			__free_page(page);
236			brd->brd_nr_pages--;
237		}
238		aligned_sector += PAGE_SECTORS;
239		size -= PAGE_SIZE;
240	}
241	xa_unlock(&brd->brd_pages);
 
 
242}
243
244static void brd_submit_bio(struct bio *bio)
 
 
245{
246	struct brd_device *brd = bio->bi_bdev->bd_disk->private_data;
247	sector_t sector = bio->bi_iter.bi_sector;
248	struct bio_vec bvec;
249	struct bvec_iter iter;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
250
251	if (unlikely(op_is_discard(bio->bi_opf))) {
252		brd_do_discard(brd, sector, bio->bi_iter.bi_size);
253		bio_endio(bio);
254		return;
255	}
256
257	bio_for_each_segment(bvec, bio, iter) {
258		unsigned int len = bvec.bv_len;
259		int err;
260
261		/* Don't support un-aligned buffer */
262		WARN_ON_ONCE((bvec.bv_offset & (SECTOR_SIZE - 1)) ||
263				(len & (SECTOR_SIZE - 1)));
264
265		err = brd_do_bvec(brd, bvec.bv_page, len, bvec.bv_offset,
266				  bio->bi_opf, sector);
267		if (err) {
268			if (err == -ENOMEM && bio->bi_opf & REQ_NOWAIT) {
269				bio_wouldblock_error(bio);
270				return;
271			}
272			bio_io_error(bio);
273			return;
274		}
275		sector += len >> SECTOR_SHIFT;
 
 
 
276	}
 
 
277
278	bio_endio(bio);
279}
280
281static const struct block_device_operations brd_fops = {
282	.owner =		THIS_MODULE,
283	.submit_bio =		brd_submit_bio,
 
 
 
284};
285
286/*
287 * And now the modules code and kernel interface.
288 */
289static int rd_nr = CONFIG_BLK_DEV_RAM_COUNT;
290module_param(rd_nr, int, 0444);
 
 
 
291MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
292
293unsigned long rd_size = CONFIG_BLK_DEV_RAM_SIZE;
294module_param(rd_size, ulong, 0444);
295MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
296
297static int max_part = 1;
298module_param(max_part, int, 0444);
299MODULE_PARM_DESC(max_part, "Num Minors to reserve between devices");
300
301MODULE_DESCRIPTION("Ram backed block device driver");
302MODULE_LICENSE("GPL");
303MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
304MODULE_ALIAS("rd");
305
306#ifndef MODULE
307/* Legacy boot options - nonmodular */
308static int __init ramdisk_size(char *str)
309{
310	rd_size = simple_strtol(str, NULL, 0);
311	return 1;
312}
313__setup("ramdisk_size=", ramdisk_size);
314#endif
315
316/*
317 * The device scheme is derived from loop.c. Keep them in synch where possible
318 * (should share code eventually).
319 */
320static LIST_HEAD(brd_devices);
321static DEFINE_MUTEX(brd_devices_mutex);
322static struct dentry *brd_debugfs_dir;
323
324static struct brd_device *brd_find_or_alloc_device(int i)
325{
326	struct brd_device *brd;
327
328	mutex_lock(&brd_devices_mutex);
329	list_for_each_entry(brd, &brd_devices, brd_list) {
330		if (brd->brd_number == i) {
331			mutex_unlock(&brd_devices_mutex);
332			return ERR_PTR(-EEXIST);
333		}
334	}
335
336	brd = kzalloc(sizeof(*brd), GFP_KERNEL);
337	if (!brd) {
338		mutex_unlock(&brd_devices_mutex);
339		return ERR_PTR(-ENOMEM);
340	}
341	brd->brd_number	= i;
342	list_add_tail(&brd->brd_list, &brd_devices);
343	mutex_unlock(&brd_devices_mutex);
344	return brd;
345}
346
347static void brd_free_device(struct brd_device *brd)
348{
349	mutex_lock(&brd_devices_mutex);
350	list_del(&brd->brd_list);
351	mutex_unlock(&brd_devices_mutex);
352	kfree(brd);
353}
354
355static int brd_alloc(int i)
356{
357	struct brd_device *brd;
358	struct gendisk *disk;
359	char buf[DISK_NAME_LEN];
360	int err = -ENOMEM;
361	struct queue_limits lim = {
362		/*
363		 * This is so fdisk will align partitions on 4k, because of
364		 * direct_access API needing 4k alignment, returning a PFN
365		 * (This is only a problem on very small devices <= 4M,
366		 *  otherwise fdisk will align on 1M. Regardless this call
367		 *  is harmless)
368		 */
369		.physical_block_size	= PAGE_SIZE,
370		.max_hw_discard_sectors	= UINT_MAX,
371		.max_discard_segments	= 1,
372		.discard_granularity	= PAGE_SIZE,
373		.features		= BLK_FEAT_SYNCHRONOUS |
374					  BLK_FEAT_NOWAIT,
375	};
376
377	brd = brd_find_or_alloc_device(i);
378	if (IS_ERR(brd))
379		return PTR_ERR(brd);
380
381	xa_init(&brd->brd_pages);
382
383	snprintf(buf, DISK_NAME_LEN, "ram%d", i);
384	if (!IS_ERR_OR_NULL(brd_debugfs_dir))
385		debugfs_create_u64(buf, 0444, brd_debugfs_dir,
386				&brd->brd_nr_pages);
387
388	disk = brd->brd_disk = blk_alloc_disk(&lim, NUMA_NO_NODE);
389	if (IS_ERR(disk)) {
390		err = PTR_ERR(disk);
391		goto out_free_dev;
392	}
 
 
 
 
 
 
 
 
 
 
 
393	disk->major		= RAMDISK_MAJOR;
394	disk->first_minor	= i * max_part;
395	disk->minors		= max_part;
396	disk->fops		= &brd_fops;
397	disk->private_data	= brd;
398	strscpy(disk->disk_name, buf, DISK_NAME_LEN);
 
 
399	set_capacity(disk, rd_size * 2);
400	
401	err = add_disk(disk);
402	if (err)
403		goto out_cleanup_disk;
404
405	return 0;
406
407out_cleanup_disk:
408	put_disk(disk);
409out_free_dev:
410	brd_free_device(brd);
411	return err;
 
412}
413
414static void brd_probe(dev_t dev)
415{
416	brd_alloc(MINOR(dev) / max_part);
 
 
 
417}
418
419static void brd_cleanup(void)
420{
421	struct brd_device *brd, *next;
422
423	debugfs_remove_recursive(brd_debugfs_dir);
 
 
 
424
425	list_for_each_entry_safe(brd, next, &brd_devices, brd_list) {
426		del_gendisk(brd->brd_disk);
427		put_disk(brd->brd_disk);
428		brd_free_pages(brd);
429		brd_free_device(brd);
430	}
 
 
431}
432
433static inline void brd_check_and_reset_par(void)
434{
435	if (unlikely(!max_part))
436		max_part = 1;
 
 
 
 
 
 
 
437
438	/*
439	 * make sure 'max_part' can be divided exactly by (1U << MINORBITS),
440	 * otherwise, it is possiable to get same dev_t when adding partitions.
441	 */
442	if ((1U << MINORBITS) % max_part != 0)
443		max_part = 1UL << fls(max_part);
444
445	if (max_part > DISK_MAX_PARTS) {
446		pr_info("brd: max_part can't be larger than %d, reset max_part = %d.\n",
447			DISK_MAX_PARTS, DISK_MAX_PARTS);
448		max_part = DISK_MAX_PARTS;
449	}
450}
451
452static int __init brd_init(void)
453{
454	int err, i;
 
 
455
456	/*
457	 * brd module now has a feature to instantiate underlying device
458	 * structure on-demand, provided that there is an access dev node.
 
 
 
459	 *
460	 * (1) if rd_nr is specified, create that many upfront. else
461	 *     it defaults to CONFIG_BLK_DEV_RAM_COUNT
462	 * (2) User can further extend brd devices by create dev node themselves
463	 *     and have kernel automatically instantiate actual device
464	 *     on-demand. Example:
465	 *		mknod /path/devnod_name b 1 X	# 1 is the rd major
466	 *		fdisk -l /path/devnod_name
467	 *	If (X / max_part) was not already created it will be created
468	 *	dynamically.
469	 */
470
471	brd_check_and_reset_par();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
472
473	brd_debugfs_dir = debugfs_create_dir("ramdisk_pages", NULL);
 
474
475	if (__register_blkdev(RAMDISK_MAJOR, "ramdisk", brd_probe)) {
476		err = -EIO;
477		goto out_free;
 
 
478	}
479
480	for (i = 0; i < rd_nr; i++)
481		brd_alloc(i);
482
483	pr_info("brd: module loaded\n");
 
 
 
 
 
 
484	return 0;
485
486out_free:
487	brd_cleanup();
 
 
 
 
488
489	pr_info("brd: module NOT loaded !!!\n");
490	return err;
491}
492
493static void __exit brd_exit(void)
494{
 
 
 
 
495
 
 
 
 
496	unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
497	brd_cleanup();
498
499	pr_info("brd: module unloaded\n");
500}
501
502module_init(brd_init);
503module_exit(brd_exit);
504