sftl/sftl.c

/*
 * Simple log-structured translation layer for FTLed flash drives
 * like memory cards or USB sticks.
 *
 * (C) 2013 Vitaliy Filippov <vitalif at mail d0t ru>
 * Redistributable under the terms of the GNU GPL 3.0+.
 */

#include <linux/module.h>
#include <linux/moduleparam.h>
#include <linux/init.h>

#include <linux/kernel.h> /* printk() */
#include <linux/fs.h>     /* everything... */
#include <linux/errno.h>  /* error codes */
#include <linux/types.h>  /* size_t */
#include <linux/vmalloc.h>
#include <linux/genhd.h>
#include <linux/idr.h>
#include <linux/blkdev.h>
#include <linux/hdreg.h>

#define KERNEL_SECTOR_SIZE 512

#define ERROR(fmt, args...) printk(KERN_ERR "sftl: " fmt "\n" , ## args)
#define INFO(fmt, args...) printk(KERN_INFO "sftl: " fmt "\n" , ## args)

MODULE_LICENSE("GPL");
MODULE_AUTHOR("Vitaliy Filippov <vitalif@mail.ru>");
MODULE_DESCRIPTION("Log-structured translation layer for USB sticks and memory cards");
MODULE_VERSION("0.1");

static int major_num = 0;

const u32 magic = 0x4C544673; // Magic: sFTL
const int phy_sz = 512;
const int clust_sz = 4096;
const int clust_blocks = 4096/512;
const int seg_sz = 512/16; /* (32) clusters */

/* Mapping element */
struct sftl_map {
	u32 magic, block, ver, checksum;
};

/* Trivial checksum using some prime number */
#define sftl_map_checksum(m) ((u32)((1+(m).block+((m).magic>>16))*(1+(m).ver+((m).magic&0xFFFF))*0xC4489EF5))

/* The internal representation of our device */
struct sftl_dev {
	u32 size;       // device size in blocks
	u32 segs;       // device size in segments
	u32 reserved_segs;
	u32 *map, *ver; // block mappings and versions
	u32 nextblk;    // next free block pointer
	u32 freeclust, freesegs; // free block count, free segment count
	struct gendisk *gd;
	struct block_device *blkdev;
	struct request_queue *queue;
	spinlock_t spinlock;
	struct mutex write_mutex;
	struct list_head list;
};

/* Index allocator */
static DEFINE_SPINLOCK(sftl_index_lock);
static DEFINE_IDA(sftl_index_ida);

/* Our block device list, used in cleanup_module */
static LIST_HEAD(sftl_device_list);

static void sftl_complete_seg(struct bio *bio, int err)
{
	bio_endio((struct bio *)bio->bi_private, err);
}

static void sftl_make_request(struct request_queue *q, struct bio *bio)
{
	struct sftl_dev *sftl = (struct sftl_dev*)q->queuedata;
	BUG_ON(bio->bi_vcnt > 1);
	BUG_ON(bio->bi_sector % clust_blocks);
	BUG_ON(bio->bi_size != clust_sz);
	if (bio->bi_sector > sftl->size)
	{
		INFO("Beyond-end i/o (starting sector = %ld)", (long)bio->bi_sector);
		bio_endio(bio, -EIO);
	}
	else if (!bio_rw(bio))
	{
		if (!sftl->ver[bio->bi_sector/clust_blocks])
		{
			// version=0 => unallocated cluster
			zero_fill_bio(bio);
			bio_endio(bio, 0);
		}
		else
		{
			struct block_device *bdev = sftl->blkdev;
			struct request_queue *q = bdev_get_queue(bdev);
			struct bio *bb = bio_alloc(GFP_KERNEL, 1);
			u32 m;
			if (IS_ERR(bb))
				return;
			bio_add_pc_page(q, bb, bio_page(bio), bio->bi_size, bio_offset(bio));
			m = sftl->map[bio->bi_sector/clust_blocks];
			bb->bi_sector = m/seg_sz * (seg_sz*clust_blocks + 1) + (m%seg_sz)*clust_blocks;
			bb->bi_bdev = bdev;
			bb->bi_private = bio;
			bb->bi_end_io = sftl_complete_seg;
			submit_bio(READ, bb);
			if (!(bb->bi_flags & (1 << BIO_UPTODATE)))
			{
				bio_endio(bio, -EIO);
				bio_put(bb);
			}
		}
	}
	else
	{
		INFO("Write request (starting sector = %ld, count = %ld)", (long)bio->bi_sector, (long)bio_sectors(bio));
		bio_endio(bio, -EIO);
	}
}

/*
 * The HDIO_GETGEO ioctl is handled in blkdev_ioctl(), which
 * calls this. We need to implement getgeo, since we can't
 * use tools such as fdisk to partition the drive otherwise.
 */
int sftl_getgeo(struct block_device * block_device, struct hd_geometry * geo)
{
	geo->cylinders = (block_device->bd_block_size & ~0x3f) >> 6;
	geo->heads = 4;
	geo->sectors = 16;
	geo->start = 0;
	return 0;
}

/*
 * The device operations structure.
 */
static struct block_device_operations sftl_ops = {
	.owner  = THIS_MODULE,
	.getgeo = sftl_getgeo
};

static int sftl_reg_major(void)
{
	if (!major_num)
	{
		/* Register major number */
		major_num = register_blkdev(major_num, "sftl");
		if (major_num < 0)
		{
			printk(KERN_WARNING "sftl: unable to get major number\n");
			return -1;
		}
	}
	return 0;
}

static int __init sftl_init(void)
{
	return sftl_reg_major();
}

static void sftl_free_device(struct sftl_dev *dev)
{
	if (!dev)
		return;
	if (dev->gd)
	{
		del_gendisk(dev->gd);
		put_disk(dev->gd);
		blk_cleanup_queue(dev->queue);
	}
	if (dev->blkdev)
	{
		invalidate_mapping_pages(dev->blkdev->bd_inode->i_mapping, 0, -1);
		blkdev_put(dev->blkdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
	}
	if (dev->map)
		vfree(dev->map);
	if (dev->ver)
		vfree(dev->ver);
	kfree(dev);
}

static void __exit sftl_exit(void)
{
	struct list_head *pos, *next;
	/* Remove all devices */
	list_for_each_safe(pos, next, &sftl_device_list)
	{
		struct sftl_dev *dev = list_entry(pos, typeof(*dev), list);
		sync_blockdev(dev->blkdev);
		list_del(&dev->list);
		sftl_free_device(dev);
		INFO("%s: removed", dev->gd->disk_name);
	}
	unregister_blkdev(major_num, "sftl");
}

module_init(sftl_init);
module_exit(sftl_exit);

static void endFunc_tryKM2(struct bio *bb, int err)
{
	if (bb->bi_private)
	{
		complete((struct completion*)(bb->bi_private));
	}
}

static void sync_read(struct block_device *bdev, unsigned sector, void *buf, unsigned len)
{
	struct bio *bb;
	struct request_queue *q;
	DECLARE_COMPLETION_ONSTACK(waithandle);
	q = bdev_get_queue(bdev);
	bb = bio_map_kern(q, buf, len, GFP_KERNEL);
	if (IS_ERR(bb))
		return;
	bb->bi_sector = sector;
	bb->bi_bdev = bdev;
	bb->bi_private = &waithandle;
	bb->bi_end_io = endFunc_tryKM2;
	submit_bio(READ, bb);
	if (!(bb->bi_flags & (1 << BIO_UPTODATE)))
	{
		bio_put(bb);
		return;
	}
	wait_for_completion(&waithandle);
	bio_put(bb);
}

static void read_maps(struct sftl_dev *dev)
{
	struct sftl_map *buf = kmalloc(phy_sz, GFP_KERNEL);
	int i, j, seg;
	INFO("reading translation maps");
	for (i = 0; i < dev->segs; i++)
	{
		sync_read(dev->blkdev, i*(seg_sz*clust_blocks+1), buf, phy_sz);
		for (seg = 1, j = 0; j < seg_sz; j++)
		{
			if (buf[j].magic == magic && buf[j].checksum == sftl_map_checksum(buf[j]) &&
				dev->ver[i*seg_sz+j] < buf[j].ver)
			{
				dev->map[i*seg_sz+j] = buf[j].block;
				dev->ver[i*seg_sz+j] = buf[j].ver;
				seg = 0;
			}
			else
			{
				dev->ver[i*seg_sz+j] = 0;
				dev->freeclust++;
			}
		}
		dev->freesegs += seg;
	}
	kfree(buf);
}

static struct sftl_dev *add_device(char *devname)
{
	const fmode_t mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
	struct block_device *bdev;
	struct sftl_dev *dev;
	int error, index;
	uint64_t t;

	if (!major_num)
		sftl_reg_major();

	if (!devname)
		return NULL;

	dev = kzalloc(sizeof(struct sftl_dev), GFP_KERNEL);
	if (!dev)
		return NULL;

	/* Get a handle on the underlying device */
	bdev = blkdev_get_by_path(devname, mode, dev);
	if (IS_ERR(bdev))
	{
		ERROR("cannot open device %s", devname);
		goto devinit_err;
	}
	dev->blkdev = bdev;

/*	if (MAJOR(bdev->bd_dev) == major_num)
	{
		ERROR("attempting to translate an SFTL device");
		goto devinit_err;
	}*/

	mutex_init(&dev->write_mutex);

	t = dev->blkdev->bd_inode->i_size;
	do_div(t, clust_sz*seg_sz + phy_sz);
	dev->segs = t;
	dev->size = dev->segs * seg_sz * clust_blocks;
	dev->reserved_segs = seg_sz * (seg_sz+1);
	dev->map = vmalloc(sizeof(u32) * dev->size);
	dev->ver = vmalloc(sizeof(u32) * dev->size);
	memset(dev->ver, 0, sizeof(u32) * dev->size);

	/* Get a request queue */
	spin_lock_init(&dev->spinlock);
	dev->queue = blk_alloc_queue(GFP_KERNEL);
	if (!dev->queue)
		goto devinit_err;
	blk_queue_make_request(dev->queue, sftl_make_request);
	dev->queue->queuedata = dev;
	/* FIXME: It's OK when PAGE_SIZE==clust_sz==4096
	   but we should ALWAYS support bio of size==PAGE_SIZE */
	blk_queue_max_hw_sectors(dev->queue, clust_blocks);
	blk_queue_logical_block_size(dev->queue, clust_sz);

	/* Allocate index for the new disk */
	do {
		if (!ida_pre_get(&sftl_index_ida, GFP_KERNEL))
			goto devinit_err;

		spin_lock(&sftl_index_lock);
		error = ida_get_new(&sftl_index_ida, &index);
		spin_unlock(&sftl_index_lock);
	} while (error == -EAGAIN);

	/* Initialise gendisk structure */
	dev->gd = alloc_disk(16);
	if (!dev->gd)
		goto devinit_err;
	dev->gd->major = major_num;
	dev->gd->first_minor = index*16;
	dev->gd->fops = &sftl_ops;
	dev->gd->private_data = dev;
	snprintf(dev->gd->disk_name, 32, "sftl%d", index);
	set_capacity(dev->gd, dev->size);
	dev->gd->queue = dev->queue;

	/* Read maps from the device */
	read_maps(dev);

	/* Add disk */
	add_disk(dev->gd);

	list_add(&dev->list, &sftl_device_list);
	INFO("%s: translating %s; %d sectors; %d free", dev->gd->disk_name, devname, dev->size, dev->freeclust*clust_blocks);
	return dev;

devinit_err:
	sftl_free_device(dev);
	return NULL;
}

static inline void kill_final_newline(char *str)
{
	char *newline = strrchr(str, '\n');
	if (newline && !newline[1])
		*newline = 0;
}

static int sftl_setup(const char *val, struct kernel_param *kp)
{
	char buf[80 + 12];
	char *str = buf;
	char *token[2];
	char *name;
	int i;

	if (strnlen(val, sizeof(buf)) >= sizeof(buf))
	{
		ERROR("parameter too long");
		return 0;
	}

	strcpy(str, val);
	kill_final_newline(str);

	for (i = 0; i < 1; i++)
		token[i] = strsep(&str, ",");

	if (str)
	{
		ERROR("too much parameters");
		return 0;
	}

	if (!token[0])
	{
		ERROR("underlying block device name missing");
		return 0;
	}

	name = token[0];
	if (strlen(name) + 1 > 80)
	{
		ERROR("device name too long");
		return 0;
	}

	add_device(name);

	return 0;
}

module_param_call(dev, sftl_setup, NULL, NULL, 0200);
MODULE_PARM_DESC(dev, "Block device to translate. \"dev=<dev>\"");