LDD3源码分析之ioctl操作

转载自http://blog.csdn.net/liuhaoyutz/article/details/7386254

作者：刘昊昱 

博客：http://blog.csdn.net/liuhaoyutz

编译环境：Ubuntu 10.10

内核版本：2.6.32-38-generic-pae

LDD3源码路径：examples/scull/main.c

 

本文分析LDD3第六章中关于ioctl操作的代码，并编写测试程序对ioctl功能进行测试。

 

一、ioctl操作

驱动程序中ioctl函数的函数原型如下：

int (*ioctl)(struct inode *inode, struct file *filp,

             unsigned int cmd, unsigned long arg);

其中cmd和arg参数是ioctl与其它驱动程序函数不同的地方。cmd是预先定义好的一些命令编号，对应要求ioctl执行的命令。arg是与cmd配合使用的参数。

ioctl函数的功能比较繁琐，从函数名可以看出，它一般是实现对设备的各种控制操作。可以这样理解，通过常规的read，write，lseek等等函数实现不合理的功能，就交给ioctl来实现。例如：要求设备锁门，弹出介质，改变波特率，甚至执行自我破坏，等等。

ioctl的实现一般是通过一个大的switch语句，根据cmd参数执行不同的操作。所以，在实现ioctl函数之前，要先定义好cmd对应的命令编号。为了防止发生混淆，命令编号应该在系统范围内是唯一的。为此，Linux内核将命令编号分为4个部分，即4个位段，分别是：

type: 幻数(magic number)，它占8位。个人理解幻数就是一个标志，代表一个(类)对象。后面我们会看到，scull使用字符’k’作为幻数。

number：序数，即顺序编号，它也占8位。

direction：如果相关命令涉及到数据的传输，则这个位段表示数据传输的方向，可用的值包括_IOC_NONE(没有数据传输)，_IOC_READ(读)、_IOC_WRITE(写)、_IOC_READ | _IOC_WRITE(双向传输数据)。注意，数据传输方向是从应用程序的角度看的，也就是说_IOC_READ意味着从设备中读数据，所以驱动程序必须向用户空间写数据。

size：所涉及的用户数据大小。这个位段的宽度与体系结构有关，通常是13或14位。

<linux/ioctl.h>中包含的<asm/ioctl.h>头文件定义了一些构造命令编号的宏：

_IO(type, nr)，用于构造无数据传输的命令编号。

_IOR(type, nr, datatype)，用于构造从驱动程序中读取数据的命令编号。

_IOW(type, nr, datatype)，用于构造向设备写入数据的命令编号。

_IOWR(type, nr, datatype)，用于双向传输命令编号。

其中，type和number位段从以上宏的参数中传入，size位段通过对datatype参数取sizeof获得。

另外，<asm/ioctl.h>头文件中还定义了一些用于解析命令编号的宏，如_IOC_DIR(cmd)，_IOC_TYPE(cmd)，_IOC_NR(cmd)，_IOC_SIZE(cmd)。

首先我们来看一下scull是如何定义命令编号的，理解scull的ioctl函数的实现，关键是理解这些命令是什么含义，即要求完成什么工作。在scull.h中有如下定义：

 135/*
 136 * Ioctl definitions
 137 */
 138
 139/* Use 'k' as magic number */
 140#define SCULL_IOC_MAGIC  'k'
 141/* Please use a different 8-bit number in your code */
 142
 143#define SCULL_IOCRESET    _IO(SCULL_IOC_MAGIC, 0)
 144
 145/*
 146 * S means "Set" through a ptr,
 147 * T means "Tell" directly with the argument value
 148 * G means "Get": reply by setting through a pointer
 149 * Q means "Query": response is on the return value
 150 * X means "eXchange": switch G and S atomically
 151 * H means "sHift": switch T and Q atomically
 152 */
 153#define SCULL_IOCSQUANTUM _IOW(SCULL_IOC_MAGIC,  1, int)
 154#define SCULL_IOCSQSET    _IOW(SCULL_IOC_MAGIC,  2, int)
 155#define SCULL_IOCTQUANTUM _IO(SCULL_IOC_MAGIC,   3)
 156#define SCULL_IOCTQSET    _IO(SCULL_IOC_MAGIC,   4)
 157#define SCULL_IOCGQUANTUM _IOR(SCULL_IOC_MAGIC,  5, int)
 158#define SCULL_IOCGQSET    _IOR(SCULL_IOC_MAGIC,  6, int)
 159#define SCULL_IOCQQUANTUM _IO(SCULL_IOC_MAGIC,   7)
 160#define SCULL_IOCQQSET    _IO(SCULL_IOC_MAGIC,   8)
 161#define SCULL_IOCXQUANTUM _IOWR(SCULL_IOC_MAGIC, 9, int)
 162#define SCULL_IOCXQSET    _IOWR(SCULL_IOC_MAGIC,10, int)
 163#define SCULL_IOCHQUANTUM _IO(SCULL_IOC_MAGIC,  11)
 164#define SCULL_IOCHQSET    _IO(SCULL_IOC_MAGIC,  12)
 165
 166/*
 167 * The other entities only have "Tell" and "Query", because they're
 168 * not printed in the book, and there's no need to have all six.
 169 * (The previous stuff was only there to show different ways to do it.
 170 */
 171#define SCULL_P_IOCTSIZE _IO(SCULL_IOC_MAGIC,   13)
 172#define SCULL_P_IOCQSIZE _IO(SCULL_IOC_MAGIC,   14)
 173/* ... more to come */
 174
 175#define SCULL_IOC_MAXNR 14

140行，定义scull的幻数是字符'k'

146行，’S’代表通过参数arg指向的内容设置。

147行，’T’代表直接通过参数arg的值设置。

148行，’G’代表通过参数arg指向的地址返回请求的值。

149行，’Q’代表通过ioctl函数的返回值返回请求的值。

150行，’X’代表通过参数arg指向的内容设置，再把原来的值通过arg指向的地址返回。即’S’与’G’两个操作合为一步。

151行，’H’代表通过参数arg的值直接设置，再通过ioctl函数的返回值将原来的值返回。即’T’和’Q’两个操作合为一步。

153行，定义命令SCULL_IOCSQUANTUM，该命令表示通过参数arg指向的内容设置quantum。

154行，定义命令SCULL_IOCSQSET，该命令表示通过参数arg指向的内容设置qset。

155行，定义命令SCULL_IOCTQUANTUM，该命令表示通过参数arg的值直接设置quantum。

156行，定义命令SCULL_IOCTQSET，该命令表示通过参数arg的值直接设置qset。

157行，定义命令SCULL_IOCGQUANTUM，该命令表示通过参数arg指向的地址返回quantum。

158行，定义命令SCULL_IOCGQSET，该命令表示通过参数arg指向的地址返回qset。

159行，定义命令SCULL_IOCQQUANTUM，该命令表示通过ioctl的返回值返回quantum。

160行，定义命令SCULL_IOCQQSET，该命令表示通过ioctl的返回值返回qset。

161行，定义命令SCULL_IOCXQUANTUM，该命令表示通过参数arg指向的内容设置quantum，然后，再把quantum原来的值写入arg指向的地址返回。

162行，定义命令SCULL_IOCXQSET，该命令表示通过参数arg指向的内容设置qset，然后，再把qset原来的值写入arg指向的地址返回。

163行，定义命令SCULL_IOCHQUANTUM，该命令表示通过参数arg的值直接设置quantum，然后，再通过ioctl的返回值返回quantum原来的值。

164行，定义命令SCULL_IOCHQSET，该命令表示通过参数arg的值直接设置qset，然后，再通过ioctl的返回值返回qset原来的值。

171行，定义命令SCULL_P_IOCTSIZE，该命令表示通过参数arg的值直接设置scull_p_buffer。

172行，定义命令SCULL_P_IOCQSIZE，该命令表示通过ioctl的返回值返回scull_p_buffer。

175定义SCULL_IOC_MAXNR为14，代表一共有14个命令。

理解了scull的ioctl命令的含义，我们就可以看ioctl的代码了，下面列出scull的ioctl函数代码如下：

 389/*
 390 * The ioctl() implementation
 391 */
 392
 393int scull_ioctl(struct inode *inode, struct file *filp,
 394                 unsigned int cmd, unsigned long arg)
 395{
 396
 397    int err = 0, tmp;
 398    int retval = 0;
 399
 400    /*
 401     * extract the type and number bitfields, and don't decode
 402     * wrong cmds: return ENOTTY (inappropriate ioctl) before access_ok()
 403     */
 404    if (_IOC_TYPE(cmd) != SCULL_IOC_MAGIC) return -ENOTTY;
 405    if (_IOC_NR(cmd) > SCULL_IOC_MAXNR) return -ENOTTY;
 406
 407    /*
 408     * the direction is a bitmask, and VERIFY_WRITE catches R/W
 409     * transfers. `Type' is user-oriented, while
 410     * access_ok is kernel-oriented, so the concept of "read" and
 411     * "write" is reversed
 412     */
 413    if (_IOC_DIR(cmd) & _IOC_READ)
 414        err = !access_ok(VERIFY_WRITE, (void __user *)arg, _IOC_SIZE(cmd));
 415    else if (_IOC_DIR(cmd) & _IOC_WRITE)
 416        err =  !access_ok(VERIFY_READ, (void __user *)arg, _IOC_SIZE(cmd));
 417    if (err) return -EFAULT;
 418
 419    switch(cmd) {
 420
 421      case SCULL_IOCRESET:
 422        scull_quantum = SCULL_QUANTUM;
 423        scull_qset = SCULL_QSET;
 424        break;
 425
 426      case SCULL_IOCSQUANTUM: /* Set: arg points to the value */
 427        if (! capable (CAP_SYS_ADMIN))
 428            return -EPERM;
 429        retval = __get_user(scull_quantum, (int __user *)arg);
 430        break;
 431
 432      case SCULL_IOCTQUANTUM: /* Tell: arg is the value */
 433        if (! capable (CAP_SYS_ADMIN))
 434            return -EPERM;
 435        scull_quantum = arg;
 436        break;
 437
 438      case SCULL_IOCGQUANTUM: /* Get: arg is pointer to result */
 439        retval = __put_user(scull_quantum, (int __user *)arg);
 440        break;
 441
 442      case SCULL_IOCQQUANTUM: /* Query: return it (it's positive) */
 443        return scull_quantum;
 444
 445      case SCULL_IOCXQUANTUM: /* eXchange: use arg as pointer */
 446        if (! capable (CAP_SYS_ADMIN))
 447            return -EPERM;
 448        tmp = scull_quantum;
 449        retval = __get_user(scull_quantum, (int __user *)arg);
 450        if (retval == 0)
 451            retval = __put_user(tmp, (int __user *)arg);
 452        break;
 453
 454      case SCULL_IOCHQUANTUM: /* sHift: like Tell + Query */
 455        if (! capable (CAP_SYS_ADMIN))
 456            return -EPERM;
 457        tmp = scull_quantum;
 458        scull_quantum = arg;
 459        return tmp;
 460
 461      case SCULL_IOCSQSET:
 462        if (! capable (CAP_SYS_ADMIN))
 463            return -EPERM;
 464        retval = __get_user(scull_qset, (int __user *)arg);
 465        break;
 466
 467      case SCULL_IOCTQSET:
 468        if (! capable (CAP_SYS_ADMIN))
 469            return -EPERM;
 470        scull_qset = arg;
 471        break;
 472
 473      case SCULL_IOCGQSET:
 474        retval = __put_user(scull_qset, (int __user *)arg);
 475        break;
 476
 477      case SCULL_IOCQQSET:
 478        return scull_qset;
 479
 480      case SCULL_IOCXQSET:
 481        if (! capable (CAP_SYS_ADMIN))
 482            return -EPERM;
 483        tmp = scull_qset;
 484        retval = __get_user(scull_qset, (int __user *)arg);
 485        if (retval == 0)
 486            retval = put_user(tmp, (int __user *)arg);
 487        break;
 488
 489      case SCULL_IOCHQSET:
 490        if (! capable (CAP_SYS_ADMIN))
 491            return -EPERM;
 492        tmp = scull_qset;
 493        scull_qset = arg;
 494        return tmp;
 495
 496        /*
 497         * The following two change the buffer size for scullpipe.
 498         * The scullpipe device uses this same ioctl method, just to
 499         * write less code. Actually, it's the same driver, isn't it?
 500         */
 501
 502      case SCULL_P_IOCTSIZE:
 503        scull_p_buffer = arg;
 504        break;
 505
 506      case SCULL_P_IOCQSIZE:
 507        return scull_p_buffer;
 508
 509
 510      default:  /* redundant, as cmd was checked against MAXNR */
 511        return -ENOTTY;
 512    }
 513    return retval;
 514
 515}

404行，如果_IOC_TYPE(cmd) != SCULL_IOC_MAGIC，即cmd的幻数不是’k’，则退出。

405行，如果_IOC_NR(cmd) > SCULL_IOC_MAXNR，即cmd的序数大于14，则退出。

413 - 417行，如果要使用arg指向的地址进行数据的读或写，必须保证对该地址的访问是合法的，这可通过access_ok函数来验证，如果访问不合法，则退出。

419行，进入switch语句块。根据传入的cmd值，进入不同的分支执行。

420 - 512行，是个各cmd的处理分支，只要我们理解了各个cmd的含义，就很容易实现这些命令要求完成的工作。如果有不理解的地方，回到前面的各个cmd的定义处再研究一下。值得一提的是，驱动程序与用户空间传递数据，采用的是__put_user和__get_user函数，相比copy_to_user和copy_from_user来说，这些函数在处理1、2、4、8个字节的数据传输时，效率更高。另外，scull允许任何用户查询quantum和qset的大小，但只允许被授权的用户修改quantum和qset的值。这种权能的检查是通过capable()函数实现的。

 

二、测试ioctl

要测试scull驱动中ioctl函数是否实现了我们要求的功能，需要编写用户空间程序对scull模块进行测试。下面是我写的一个比较简单的测试程序：

首先是头文件scull_ioctl.h：

 #ifndef _SCULL_IOCTL_H_
 #define _SCULL_IOCTL_H_

 #include <linux/ioctl.h> /* needed for the _IOW etc stuff used later */

 /*
  * Ioctl definitions
  */

 /* Use 'k' as magic number */
 #define SCULL_IOC_MAGIC  'k'
 /* Please use a different 8-bit number in your code */

 #define SCULL_IOCRESET    _IO(SCULL_IOC_MAGIC, 0)

 /*
  * S means "Set" through a ptr,
  * T means "Tell" directly with the argument value
  * G means "Get": reply by setting through a pointer
  * Q means "Query": response is on the return value
  * X means "eXchange": switch G and S atomically
  * H means "sHift": switch T and Q atomically
  */
 #define SCULL_IOCSQUANTUM _IOW(SCULL_IOC_MAGIC,  1, int)
 #define SCULL_IOCSQSET    _IOW(SCULL_IOC_MAGIC,  2, int)
 #define SCULL_IOCTQUANTUM _IO(SCULL_IOC_MAGIC,   3)
 #define SCULL_IOCTQSET    _IO(SCULL_IOC_MAGIC,   4)
 #define SCULL_IOCGQUANTUM _IOR(SCULL_IOC_MAGIC,  5, int)
 #define SCULL_IOCGQSET    _IOR(SCULL_IOC_MAGIC,  6, int)
 #define SCULL_IOCQQUANTUM _IO(SCULL_IOC_MAGIC,   7)
 #define SCULL_IOCQQSET    _IO(SCULL_IOC_MAGIC,   8)
 #define SCULL_IOCXQUANTUM _IOWR(SCULL_IOC_MAGIC, 9, int)
 #define SCULL_IOCXQSET    _IOWR(SCULL_IOC_MAGIC,10, int)
 #define SCULL_IOCHQUANTUM _IO(SCULL_IOC_MAGIC,  11)
 #define SCULL_IOCHQSET    _IO(SCULL_IOC_MAGIC,  12)

 /*
  * The other entities only have "Tell" and "Query", because they're
  * not printed in the book, and there's no need to have all six.
  * (The previous stuff was only there to show different ways to do it.
  */
 #define SCULL_P_IOCTSIZE _IO(SCULL_IOC_MAGIC,   13)
 #define SCULL_P_IOCQSIZE _IO(SCULL_IOC_MAGIC,   14)
 /* ... more to come */

 #define SCULL_IOC_MAXNR 14

 #endif /* _SCULL_IOCTL_H_ */

下面是测试程序scull_ioctl_test.c的代码：

 #include <sys/types.h>
 #include <sys/stat.h>
 #include <sys/ioctl.h>
 #include <fcntl.h>
 #include <stdio.h>
 #include "scull_ioctl.h"

 #define SCULL_DEVICE "/dev/scull0"

 int main(int argc, char *argv[])
 {
     int fd = 0;
     int quantum = 8000;
     int quantum_old = 0;
     int qset = 2000;
     int qset_old = 0;

     fd = open(SCULL_DEVICE, O_RDWR);
     if(fd < 0)
     {
         printf("open scull device error!\n");
         return 0;
     }

     printf("SCULL_IOCSQUANTUM: quantum = %d\n", quantum);
     ioctl(fd, SCULL_IOCSQUANTUM, &quantum);
     quantum -= 500;
     printf("SCULL_IOCTQUANTUM: quantum = %d\n", quantum);
     ioctl(fd, SCULL_IOCTQUANTUM, quantum);

     ioctl(fd, SCULL_IOCGQUANTUM, &quantum);
     printf("SCULL_IOCGQUANTUM: quantum = %d\n", quantum);
     quantum = ioctl(fd, SCULL_IOCQQUANTUM);
     printf("SCULL_IOCQQUANTUM: quantum = %d\n", quantum);

     quantum -= 500;
     quantum_old = ioctl(fd, SCULL_IOCHQUANTUM, quantum);
     printf("SCULL_IOCHQUANTUM: quantum = %d, quantum_old = %d\n", quantum, quantum_old);
     quantum -= 500;
     printf("SCULL_IOCXQUANTUM: quantum = %d\n", quantum);
     ioctl(fd, SCULL_IOCXQUANTUM, &quantum);
     printf("SCULL_IOCXQUANTUM: old quantum = %d\n", quantum);

     printf("SCULL_IOCSQSET: qset = %d\n", qset);
     ioctl(fd, SCULL_IOCSQSET, &qset);
     qset += 500;
     printf("SCULL_IOCTQSET: qset = %d\n", qset);
     ioctl(fd, SCULL_IOCTQSET, qset);

     ioctl(fd, SCULL_IOCGQSET, &qset);
     printf("SCULL_IOCGQSET: qset = %d\n", qset);
     qset = ioctl(fd, SCULL_IOCQQSET);
     printf("SCULL_IOCQQSET: qset = %d\n", qset);

     qset += 500;
     qset_old = ioctl(fd, SCULL_IOCHQSET, qset);
     printf("SCULL_IOCHQSET: qset = %d, qset_old = %d\n", qset, qset_old);
     qset += 500;
     printf("SCULL_IOCXQSET: qset = %d\n", qset);
     ioctl(fd, SCULL_IOCXQSET, &qset);
     printf("SCULL_IOCHQSET: old qset = %d\n", qset);

     return 0;
 }

为了能看到测试效果，在修改驱动程序中的ioctl函数，打印一些语句。下面直接列出修改后的ioctl函数的实现：

 /*
  * The ioctl() implementation
  */

 int scull_ioctl(struct inode *inode, struct file *filp,
                  unsigned int cmd, unsigned long arg)
 {

     int err = 0, tmp;
     int retval = 0;

     /*
      * extract the type and number bitfields, and don't decode
      * wrong cmds: return ENOTTY (inappropriate ioctl) before access_ok()
      */
     if (_IOC_TYPE(cmd) != SCULL_IOC_MAGIC) return -ENOTTY;
     if (_IOC_NR(cmd) > SCULL_IOC_MAXNR) return -ENOTTY;

     /*
      * the direction is a bitmask, and VERIFY_WRITE catches R/W
      * transfers. `Type' is user-oriented, while
      * access_ok is kernel-oriented, so the concept of "read" and
      * "write" is reversed
      */
     if (_IOC_DIR(cmd) & _IOC_READ)
         err = !access_ok(VERIFY_WRITE, (void __user *)arg, _IOC_SIZE(cmd));
     else if (_IOC_DIR(cmd) & _IOC_WRITE)
         err =  !access_ok(VERIFY_READ, (void __user *)arg, _IOC_SIZE(cmd));
     if (err) return -EFAULT;

     switch(cmd) {

       case SCULL_IOCRESET:
         scull_quantum = SCULL_QUANTUM;
         scull_qset = SCULL_QSET;
         printk("SCULL_IOCRESET: scull_quantum = %d, scull_qset = %d\n", scull_quantum, scull_qset);
         break;

       case SCULL_IOCSQUANTUM: /* Set: arg points to the value */
         if (! capable (CAP_SYS_ADMIN))
             return -EPERM;
         retval = __get_user(scull_quantum, (int __user *)arg);
         printk("SCULL_IOCSQUANTUM: scull_quantum = %d\n", scull_quantum);
         break;

       case SCULL_IOCTQUANTUM: /* Tell: arg is the value */
         if (! capable (CAP_SYS_ADMIN))
             return -EPERM;
         scull_quantum = arg;
         printk("SCULL_IOCTQUANTUM: scull_quantum = %d\n", scull_quantum);
         break;

       case SCULL_IOCGQUANTUM: /* Get: arg is pointer to result */
         retval = __put_user(scull_quantum, (int __user *)arg);
         printk("SCULL_IOCGQUANTUM: use arg return scull_quantum = %d\n", scull_quantum);
         break;

       case SCULL_IOCQQUANTUM: /* Query: return it (it's positive) */
         printk("SCULL_IOCQQUANTUM: return scull_quantum = %d\n", scull_quantum);
         return scull_quantum;

       case SCULL_IOCXQUANTUM: /* eXchange: use arg as pointer */
         if (! capable (CAP_SYS_ADMIN))
             return -EPERM;
         tmp = scull_quantum;
         retval = __get_user(scull_quantum, (int __user *)arg);
         if (retval == 0)
             retval = __put_user(tmp, (int __user *)arg);
         printk("SCULL_IOCXQUANTUM: scull_quantum = %d, and use arg return old scull_quantum = %d\n", scull_quantum, tmp);
         break;

       case SCULL_IOCHQUANTUM: /* sHift: like Tell + Query */
         if (! capable (CAP_SYS_ADMIN))
             return -EPERM;
         tmp = scull_quantum;
         scull_quantum = arg;
         printk("SCULL_IOCHQUANTUM: scull_quantum = %d, and return old scull_quantum = %d\n", scull_quantum, tmp);
         return tmp;

       case SCULL_IOCSQSET:
         if (! capable (CAP_SYS_ADMIN))
             return -EPERM;
         retval = __get_user(scull_qset, (int __user *)arg);
         printk("SCULL_IOCSQSET: scull_qset = %d\n", scull_qset);
         break;

       case SCULL_IOCTQSET:
         if (! capable (CAP_SYS_ADMIN))
             return -EPERM;
         scull_qset = arg;
         printk("SCULL_IOCTQSET: scull_qset = %d\n", scull_qset);
         break;

       case SCULL_IOCGQSET:
         retval = __put_user(scull_qset, (int __user *)arg);
         printk("SCULL_IOCGQSET: use arg return scull_qset = %d\n", scull_qset);
         break;

       case SCULL_IOCQQSET:
         printk("SCULL_IOCQQSET: return scull_qset = %d\n", scull_qset);
         return scull_qset;

       case SCULL_IOCXQSET:
         if (! capable (CAP_SYS_ADMIN))
             return -EPERM;
         tmp = scull_qset;
         retval = __get_user(scull_qset, (int __user *)arg);
         if (retval == 0)
             retval = put_user(tmp, (int __user *)arg);
         printk("SCULL_IOCXQSET: scull_qset = %d, and use arg return old scull_qset = %d\n", scull_qset, tmp);
         break;

       case SCULL_IOCHQSET:
         if (! capable (CAP_SYS_ADMIN))
             return -EPERM;
         tmp = scull_qset;
         scull_qset = arg;
         printk("SCULL_IOCHQSET: scull_qet = %d, and return old scull_qset = %d\n", scull_qset, tmp);
         return tmp;

         /*
          * The following two change the buffer size for scullpipe.
          * The scullpipe device uses this same ioctl method, just to
          * write less code. Actually, it's the same driver, isn't it?
          */

       case SCULL_P_IOCTSIZE:
         scull_p_buffer = arg;
         break;

       case SCULL_P_IOCQSIZE:
         return scull_p_buffer;


       default:  /* redundant, as cmd was checked against MAXNR */
         return -ENOTTY;
     }
     return retval;

 }

在我的系统上，测试过程如图所示。需要注意的是测试程序必须以root权限运行，因为普通用户只能读quantum和qset的值，只有root用户才能修改。