Linux 块设备驱动 实例

Linux 块设备驱动 实例

任务：用一片虚拟地址连续的内存空间模拟一个块设备，并为其写一个驱动

/*
 * Sample disk driver, from the beginning.
 */

#include <linux/autoconf.h>
#include <linux/module.h>
#include <linux/moduleparam.h>
#include <linux/init.h>
#include <linux/sched.h>
#include <linux/kernel.h>    /* printk() */
#include <linux/slab.h>        /* kmalloc() */
#include <linux/fs.h>        /* everything... */
#include <linux/errno.h>    /* error codes */
#include <linux/timer.h>
#include <linux/types.h>    /* size_t */
#include <linux/fcntl.h>    /* O_ACCMODE */
#include <linux/hdreg.h>    /* HDIO_GETGEO */
#include <linux/kdev_t.h>
#include <linux/vmalloc.h>
#include <linux/genhd.h>
#include <linux/blkdev.h>
#include <linux/buffer_head.h>    /* invalidate_bdev */
#include <linux/bio.h>

MODULE_LICENSE("Dual BSD/GPL");

static int sbull_major = 0;
module_param(sbull_major, int, 0);
static int hardsect_size = 512;
module_param(hardsect_size, int, 0);
static int nsectors = 25600;    /* How big the drive is */
module_param(nsectors, int, 0);
static int ndevices = 1;
module_param(ndevices, int, 0);

/*
 * The different "request modes" we can use.
 */
enum {
    RM_SIMPLE  = 0,    /* The extra-simple request function */
    RM_FULL    = 1,    /* The full-blown version */
    RM_NOQUEUE = 2,    /* Use make_request */
};
//static int request_mode = RM_FULL;
static int request_mode = RM_SIMPLE;
//static int request_mode = RM_SIMPLE;
module_param(request_mode, int, 0);

/*
 * Minor number and partition management.
 */
#define SBULL_MINORS    16
#define MINOR_SHIFT    4
#define DEVNUM(kdevnum)    (MINOR(kdev_t_to_nr(kdevnum)) >> MINOR_SHIFT

/*
 * We can tweak our hardware sector size, but the kernel talks to us
 * in terms of small sectors, always.
 */
#define KERNEL_SECTOR_SIZE    512

/*
 * After this much idle time, the driver will simulate a media change.
 */
#define INVALIDATE_DELAY    60*HZ

/*
 * The internal representation of our device.
 */
struct sbull_dev {
        int size;                       /* Device size in sectors */
        // data 是本程序模拟的块设备，是一片连续的虚拟空间
        // 在初始化函数里分配的虚拟地址连续的内存空间
        u8 *data;                       /* The data array */
        short users;                    /* How many users */
        short media_change;             /* Flag a media change? */
        spinlock_t lock;                /* For mutual exclusion */
        struct request_queue *queue;    /* The device request queue */
        struct gendisk *gd;             /* The gendisk structure */
        struct timer_list timer;        /* For simulated media changes */
};

static struct sbull_dev *Devices = NULL;

/*
 * Handle an I/O request.
 */
static void sbull_transfer(struct sbull_dev *dev, unsigned long sector,
        unsigned long nsect, char *buffer, int write)
{
    unsigned long offset = sector*KERNEL_SECTOR_SIZE;     // 需要读写的扇区的偏移地址
    unsigned long nbytes = nsect*KERNEL_SECTOR_SIZE;        // 需要读写的字节数
   
    if ((offset + nbytes) > dev->size) {      // 判断输入参数是否合法，是否超出边界
        printk (KERN_NOTICE "Beyond-end write (%ld %ld)\n", offset, nbytes);
        return;
    }
    // 实际的读写操作
    // 由于本程序是用一片连续的内存空间模拟块设备
    // 所以这里对硬件(内存空间)的读写操作，就是复制内存
    // 在具体点，就是下面的memcpy
    // 具体的项目，需修改为具体的接口函数
    if (write)
        // 写
        memcpy(dev->data + offset, buffer, nbytes);
    else
        // 读
        memcpy(buffer, dev->data + offset, nbytes);
}

/*The simple form of the request function.*/

static void sbull_request(struct request_queue *q)
{
    struct request *req;

    // 服务完队列上的所有请求
    while ((req = elv_next_request(q)) != NULL) {  // elv_next_request :从队列上去一个下来
        struct sbull_dev *dev = req->rq_disk->private_data;
        if (! blk_fs_request(req)) {
            printk (KERN_NOTICE "Skip non-fs request\n");
            end_request(req, 0);
            continue;
        }
        sbull_transfer(dev, req->sector, req->current_nr_sectors,
                req->buffer, rq_data_dir(req));
        end_request(req, 1);
    }
}

/*
 * Transfer a single BIO.
 */
static int sbull_xfer_bio(struct sbull_dev *dev, struct bio *bio)
{
    int i;
    struct bio_vec *bvec;
    sector_t sector = bio->bi_sector;

    /* Do each segment independently. */
    bio_for_each_segment(bvec, bio, i) {
        char *buffer = __bio_kmap_atomic(bio, i, KM_USER0);
        sbull_transfer(dev, sector, bio_cur_sectors(bio),
                buffer, bio_data_dir(bio) == WRITE);
        sector += bio_cur_sectors(bio);
        __bio_kunmap_atomic(bio, KM_USER0);
    }
    return 0; /* Always "succeed" */
}

/*
 * Transfer a full request.
 */
static int sbull_xfer_request(struct sbull_dev *dev, struct request *req)
{
    struct bio *bio;
    int nsect = 0;

    // steps through each bio that makes up a request.
    // 遍历
    __rq_for_each_bio(bio, req) {
        sbull_xfer_bio(dev, bio);
        nsect += bio->bi_size/KERNEL_SECTOR_SIZE;
    }
    return nsect;
}

/*
 * Smarter request function that "handles clustering".
 */
static void sbull_full_request(struct request_queue *q)
{
    struct request *req;
    int sectors_xferred;
    struct sbull_dev *dev = q->queuedata;

    printk("<0>""in %s\n",__FUNCTION__);
    while ((req = elv_next_request(q)) != NULL) {
        if (! blk_fs_request(req)) {
            printk (KERN_NOTICE "Skip non-fs request\n");
            end_request(req, 0);
            continue;
        }
        sectors_xferred = sbull_xfer_request(dev, req);
        __blk_end_request(req,0,sectors_xferred<<9);//add by lht for 2.6.27
    }
}

 

//The direct make request version
static int sbull_make_request(struct request_queue *q, struct bio *bio)
{
    struct sbull_dev *dev = q->queuedata;
    int status;

    status = sbull_xfer_bio(dev, bio);
    //bio_endio(bio, bio->bi_size, status);
    bio_endio(bio, status);
    return 0;
}

/*
 * Open and close.
 */

static int sbull_open(struct inode *inode, struct file *filp)
{
    struct sbull_dev *dev = inode->i_bdev->bd_disk->private_data;
    //printk("<0>" "fdfjdlksjfdlkj\n");   
    del_timer_sync(&dev->timer);
    filp->private_data = dev;
    spin_lock(&dev->lock);
    if (! dev->users)
        check_disk_change(inode->i_bdev);
    dev->users++;
    spin_unlock(&dev->lock);
    return 0;
}

static int sbull_release(struct inode *inode, struct file *filp)
{
    struct sbull_dev *dev = inode->i_bdev->bd_disk->private_data;

    spin_lock(&dev->lock);
    dev->users--;

    if (!dev->users) {
        dev->timer.expires = jiffies + INVALIDATE_DELAY;
        add_timer(&dev->timer);
    }
    spin_unlock(&dev->lock);

    return 0;
}

/*
 * Look for a (simulated) media change.
 */
int sbull_media_changed(struct gendisk *gd)
{
    struct sbull_dev *dev = gd->private_data;
   
    return dev->media_change;
}

/*
 * Revalidate.  WE DO NOT TAKE THE LOCK HERE, for fear of deadlocking
 * with open.  That needs to be reevaluated.
 */
int sbull_revalidate(struct gendisk *gd)
{
    struct sbull_dev *dev = gd->private_data;
   
    if (dev->media_change) {
        dev->media_change = 0;
        memset (dev->data, 0, dev->size);
    }
    return 0;
}

/*
 * The "invalidate" function runs out of the device timer; it sets
 * a flag to simulate the removal of the media.
 */
void sbull_invalidate(unsigned long ldev)
{
    struct sbull_dev *dev = (struct sbull_dev *) ldev;

    spin_lock(&dev->lock);
    if (dev->users || !dev->data)
        printk (KERN_WARNING "sbull: timer sanity check failed\n");
    else
        dev->media_change = 1;
    spin_unlock(&dev->lock);
}

/*
 * The ioctl() implementation
 */

int sbull_ioctl (struct inode *inode, struct file *filp,
                 unsigned int cmd, unsigned long arg)
{
    long size;
    struct hd_geometry geo;
    struct sbull_dev *dev = filp->private_data;

    switch(cmd) {
        case HDIO_GETGEO:
            /*
         * Get geometry: since we are a virtual device, we have to make
         * up something plausible.  So we claim 16 sectors, four heads,
         * and calculate the corresponding number of cylinders.  We set the
         * start of data at sector four.
         */
        //printk("<0>""-------------size=%d\n",size);
        /****************for early version************/
        //size = dev->size*(hardsect_size/KERNEL_SECTOR_SIZE);
        //printk("<0>""-------------size=%d\n",size);
        //geo.cylinders = (size & ~0x3f) >> 6;
        //geo.cylinders=2000;
        //geo.heads = 4;
        //geo.sectors = 16;
        //geo.sectors=2560;
        //geo.start = 0;
        //if (copy_to_user((void __user *) arg, &geo, sizeof(geo)))
        //    return -EFAULT;
        return 0;
    }

    return -ENOTTY; /* unknown command */
}

static int sbull_getgeo(struct block_device *bdev, struct hd_geometry *geo)
{
    unsigned long size;
    struct sbull_dev *pdev = bdev->bd_disk->private_data;

    size = pdev->size;
    geo->cylinders = (size & ~0x3f) >> 6;
    geo->heads    = 4;
    geo->sectors = 16;
    geo->start = 0;
    return 0;
}

/*
 * The device operations structure.
 */
static struct block_device_operations sbull_ops = {
    .owner           = THIS_MODULE,
    .open              = sbull_open,
    .release      = sbull_release,
    .media_changed   = sbull_media_changed,
    .revalidate_disk = sbull_revalidate,
    .ioctl             = sbull_ioctl,
    .getgeo            = sbull_getgeo,
};

/*
 * Set up our internal device.
 */
// 初始化设备结构体 static struct sbull_dev *Devices中的成员
static void setup_device(struct sbull_dev *dev, int which)
{
    /*
     * Get some memory.
     */
    memset (dev, 0, sizeof (struct sbull_dev));
    dev->size = nsectors*hardsect_size;
    // 分配一片虚拟地址连续的内存空间，作为块设备。
    dev->data = vmalloc(dev->size);  
    if (dev->data == NULL) {
        printk (KERN_NOTICE "vmalloc failure.\n");
        return;
    }
    spin_lock_init(&dev->lock);
   
    /*
     * The timer which "invalidates" the device.
     */
    init_timer(&dev->timer);
    dev->timer.data = (unsigned long) dev;
    dev->timer.function = sbull_invalidate;
   
    /*
     * The I/O queue, depending on whether we are using our own
     * make_request function or not.
     */
    switch (request_mode) {
        case RM_NOQUEUE:
        dev->queue = blk_alloc_queue(GFP_KERNEL);
        if (dev->queue == NULL)
            goto out_vfree;
        blk_queue_make_request(dev->queue, sbull_make_request);
        break;

        case RM_FULL:
        dev->queue = blk_init_queue(sbull_full_request, &dev->lock);
        if (dev->queue == NULL)
            goto out_vfree;
        break;

        default:
        printk(KERN_NOTICE "Bad request mode %d, using simple\n", request_mode);
            /* fall into.. */
   
        case RM_SIMPLE:
        dev->queue = blk_init_queue(sbull_request, &dev->lock);
        if (dev->queue == NULL)
            goto out_vfree;
        break;
    }
    blk_queue_hardsect_size(dev->queue, hardsect_size);
    dev->queue->queuedata = dev;
    /*
     * And the gendisk structure.
     */
    dev->gd = alloc_disk(SBULL_MINORS);
    if (! dev->gd) {
        printk (KERN_NOTICE "alloc_disk failure\n");
        goto out_vfree;
    }
    dev->gd->major = sbull_major;
    dev->gd->first_minor = which*SBULL_MINORS;
    dev->gd->fops = &sbull_ops;
    dev->gd->queue = dev->queue;
    dev->gd->private_data = dev;
    snprintf (dev->gd->disk_name, 32, "sbull%c", which + 'a');
    set_capacity(dev->gd, nsectors*(hardsect_size/KERNEL_SECTOR_SIZE));
    add_disk(dev->gd);
    return;

  out_vfree:
    if (dev->data)
        vfree(dev->data);
}

 

static int __init sbull_init(void)
{
    int i;
    /*
     * Get registered.
     */
    //    printk("<0>" "add by lht\n");
    sbull_major = register_blkdev(sbull_major, "sbull");
    if (sbull_major <= 0) {
        printk(KERN_WARNING "sbull: unable to get major number\n");
        return -EBUSY;
    }
    /*
     * Allocate the device array, and initialize each one.
     */
    Devices = kmalloc(ndevices*sizeof (struct sbull_dev), GFP_KERNEL);
    if (Devices == NULL)
        goto out_unregister;
    for (i = 0; i < ndevices; i++)
        setup_device(Devices + i, i);
   
    return 0;

  out_unregister:
    unregister_blkdev(sbull_major, "sbd");
    return -ENOMEM;
}

static void sbull_exit(void)
{
    int i;

    for (i = 0; i < ndevices; i++) {
        struct sbull_dev *dev = Devices + i;

        del_timer_sync(&dev->timer);
        if (dev->gd) {
            del_gendisk(dev->gd);
            put_disk(dev->gd);
        }
        if (dev->queue) {
            if (request_mode == RM_NOQUEUE)
            //    blk_put_queue(dev->queue);
            kobject_put(&(dev->queue)->kobj);
            else
                blk_cleanup_queue(dev->queue);
        }
        if (dev->data)
            vfree(dev->data);
    }
    unregister_blkdev(sbull_major, "sbull");
    kfree(Devices);
}
   
module_init(sbull_init);
module_exit(sbull_exit);

 

测试方法：

# Makefile
ifeq ($(KERNELRELEASE),)

#KERNELDIR ?= /home/lht/kernel2.6/linux-2.6.14

KERNELDIR ?= /lib/modules/$(shell uname -r)/build M=$(PWD) modules
PWD := $(shell pwd)

modules:
        $(MAKE) -C $(KERNELDIR) M=$(PWD) modules

modules_install:
        $(MAKE) -C $(KERNELDIR) M=$(PWD) modules_install

clean:
        rm -rf *.o *~ core .depend .*.cmd *.ko *.mod.c .tmp_versions

.PHONY: modules modules_install clean

else
    obj-m := sbull.o
endif

将模块插入内核（2.6.27）
root@bkjia:/source/workplace/test/sbull_bkjia# insmod sbull.ko
用lsmod查看模块是否成功插入内核
root@bkjia:/source/workplace/test/sbull_bkjia# lsmod | grep sbu
sbull                  13452  0
出现上面结果，说明成功了
用ls查看/dev下是否有sbull设备
root@bkjia:/source/workplace/test/sbull_bkjia# ls /dev | grep sbu
sbulla
出现上面结果，说明有了，如果没有，用命令
mknod /dev/sbulla b 254 0
手动创建
至此，已经有一个块设备了
下面用fdisk对虚拟块设备分区
root@bkjia:/source/workplace/test/sbull_bkjia# fdisk /dev/sbulla
Device contains neither a valid DOS partition table, nor Sun, SGI or OSF disklabel
Building a new DOS disklabel with disk identifier 0x14d0973f.
Changes will remain in memory only, until you decide to write them.
After that, of course, the previous content won't be recoverable.

Warning: invalid flag 0x0000 of partition table 4 will be corrected by w(rite)

Command (m for help): n        这里选择n，新建
Command action
   e   extended
   p   primary partition (1-4)  这里选p，主分区
p
Partition number (1-4): 1   这里选1，第一个分区
First cylinder (1-400, default 1): 1
Last cylinder, +cylinders or +size{K,M,G} (1-400, default 400):
Using default value 400

Command (m for help): w  这里选w，保存并推出
The partition table has been altered!

Calling ioctl() to re-read partition table.
Syncing disks.
接着将其格式化为ext2

root@bkjia:/source/workplace/test/sbull_bkjia# mkfs.ext2 /dev/sbulla1
mke2fs 1.41.3 (12-Oct-2008)
Filesystem label=
OS type: Linux
Block size=1024 (log=0)
Fragment size=1024 (log=0)
3200 inodes, 12792 blocks
639 blocks (5.00%) reserved for the super user
First data block=1
Maximum filesystem blocks=13107200
2 block groups
8192 blocks per group, 8192 fragments per group
1600 inodes per group
Superblock backups stored on blocks:
        8193

Writing inode tables: done                           
Writing superblocks and filesystem accounting information: done

This filesystem will be automatically checked every 29 mounts or
180 days, whichever comes first.  Use tune2fs -c or -i to override.
新建一个文件夹，作为此处模拟的块设备的挂载点
root@bkjia:/source/workplace/test/sbull_bkjia# ls /mnt/
hgfs  initrd
root@bkjia:/source/workplace/test/sbull_bkjia# mkdir /mnt/sbulla1
挂载
root@bkjia:/source/workplace/test/sbull_bkjia# mount /dev/sbulla1 /mnt/sbulla1
进入目录，新建一个文件，测试一下
root@bkjia:/source/workplace/test/sbull_bkjia# cd /mnt/sbulla1/
root@bkjia:/mnt/sbulla1# ls
lost+found
root@bkjia:/mnt/sbulla1# echo hi > hello.c
root@bkjia:/mnt/sbulla1# ls
hello.c  lost+found
root@bkjia:/mnt/sbulla1# cat hello.c
hi
root@bkjia:/mnt/sbulla1#