浅析ethx网卡控制函数ioctl实现具体流程【转】

关于介绍ioctl的文章，对无线网卡wireless ioctl实现有启发，转载于此。

作者：飞镖 luther (gliethttp) gliethttp

http://blog.chinaunix.net/u1/38994/showart_1107738.html

浅析ethx网卡控制函数ioctl实现具体流程

====================
1.应用层程序iwpriv
wireless tools网络配置应用程序iwpriv命令格式:
iwpriv ethX private-command [parameters]

iwpriv部分实现源码如下:
int main(int argc, char *argv[])
{
    ...
    sockfd = socket(AF_INET, SOCK_STREAM, 0);
    ...
    ioctl(sockfd, ioctl_val, &iwr);//将控制命令通过ioctl发送到无线网卡
    ...
}
====================
2.系统调用sys_ioctl
应用层通过ioctl(sockfd, ioctl_val, &iwr);触发sys_ioctl系统调用,实际流程:
sys_ioctl=>vfs_ioctl=>do_ioctl=最后调用
filp->f_op->unlocked_ioctl执行具体的ioctl操作,该操作就是sock_ioctl,至于为什么是sock_ioctl,后边作了进一步分析
sock_ioctl=>
{
    ...
    #ifdef CONFIG_WIRELESS_EXT
        if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
            err = dev_ioctl(net, cmd, argp);//

        } else
    #endif
    ...
}
dev_ioctl=>wext_handle_ioctl
{
    ...
/* Take care of Wireless Extensions */
    if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
        return wext_handle_ioctl(net, &ifr, cmd, arg);
    ...
}
wext_handle_ioctl=>wireless_process_ioctl=>
然后通过if ((dev = __dev_get_by_name(net, ifr->ifr_name)) == NULL)函数,
从系统管理的net链表中,把ioctl指定的ethX对应的struct net_device摘出来,
最后调用ioctl_private_call(handler)或者调用dev->do_ioctl(dev, ifr, cmd)来处理该ioctl,
这两个函数分别指向wlan_handler_def和wlan_do_ioctl
====================
3.wifi网卡是怎么登记到kernel上的
wlan_probe()=>wlan_add_card()=>alloc_etherdev()=>
之后将操作方法添加到struct net_device *dev=alloc_etherdev()申请的dev上去,其中包括:
    ...
    /* Setup the OS Interface to our functions */
    dev->open = wlan_open;
    dev->hard_start_xmit = wlan_hard_start_xmit;
    dev->stop = wlan_close;
    dev->do_ioctl = wlan_do_ioctl;
    dev->set_mac_address = wlan_set_mac_address;

    dev->tx_timeout = wlan_tx_timeout;
    dev->get_stats = wlan_get_stats;
    dev->watchdog_timeo = MRVDRV_DEFAULT_WATCHDOG_TIMEOUT;
    dev->wireless_handlers = (struct iw_handler_def *) &wlan_handler_def;
    dev->set_multicast_list = wlan_set_multicast_list;
    ...
4.socket系统调用如何关联上ioctl和ethX设备

asmlinkage long sys_socket(int family, int type, int protocol);

sys_socket=>sock_create=>__sock_create=>sock = sock_alloc();通过sock_mnt->mnt_sb从socket文件系统的超级块上申请一个inode节点,这样也就同时获得了由该inode描述的一个sock结构体单元,所以sokcet和dentry目录项等效,
接下来从net_families全局管理结构体中找到当前family对应的ops操作集,
net_proto_family *pf=net_families[family];
pf->create(net, sock, protocol);//核心调用,对于ipv4,就是inet_create
以ipv4为例
static struct net_proto_family inet_family_ops = {
    .family = PF_INET,
    .create = inet_create,
    .owner    = THIS_MODULE,
};
还记得上面应用层创建sokcet的函数吧,
sockfd = socket(AF_INET, SOCK_STREAM, 0);//AF_INET虽然等于PF_INET,但是因为种种原因我们提倡使用PF_INET
可见family等于AF_INET,type等于SOCK_STREAM,协议protocol为0,也就是采用IP协议,
inet_create=>inetsw[sock->type]也就是inetsw[SOCK_STREAM],
从inetsw[sock->type]中找到已经登记的protocol网络协议处理函数,
inetsw[]是怎么填充的呢?inet_init()=>inet_register_protosw(inetsw_array)=>这样inetsw_array中的所有protocol处理模块都将登记到inetsw中了,
static struct inet_protosw inetsw_array[] =
{
    {
        .type = SOCK_STREAM,
        .protocol = IPPROTO_TCP,
        .prot = &tcp_prot,
        .ops = &inet_stream_ops,
        .capability = -1,
        .no_check = 0,
        .flags = INET_PROTOSW_PERMANENT | INET_PROTOSW_ICSK,
    },

    {
        .type = SOCK_DGRAM,
        .protocol = IPPROTO_UDP,
        .prot = &udp_prot,
        .ops = &inet_dgram_ops,
        .capability = -1,
        .no_check = UDP_CSUM_DEFAULT,
        .flags = INET_PROTOSW_PERMANENT,
    },


    {
        .type = SOCK_RAW,
        .protocol = IPPROTO_IP,    /* wild card */
        .prot = &raw_prot,
        .ops = &inet_sockraw_ops,
        .capability = CAP_NET_RAW,
        .no_check = UDP_CSUM_DEFAULT,
        .flags = INET_PROTOSW_REUSE,
    }
};
至于inet_init,则是以fs_initcall(inet_init)方式,以5号优先级被build in到了内核中,当kernel启动时会在start_kernel=>rest_init=>kernel_init=>do_basic_setup=>do_initcalls中依据优先级号优先于其他module驱动被调用.
这样sock->ops = answer->ops;对于ipv4也就等于inet_stream_ops,
接下来就是将ops填充到file操作指针中了,
sys_socket=>sock_map_fd=>sock_attach_fd=>
dentry->d_op = &sockfs_dentry_operations;
init_file(file, sock_mnt, dentry, FMODE_READ | FMODE_WRITE, &socket_file_ops);
file->private_data = sock;
其中init_file=>file->f_op = fop;也就是file->f_op = socket_file_ops;
所以read(),wirte(),poll()和ioctl()应用程序调用的file->f_op就是socket_file_ops了,
比如:
read()对应sock_aio_read网络异步读
write()对应sock_aio_write网络异步写
ioctl()对应sock_ioctl

socket_file_ops结构体具体实现如下:
static const struct file_operations socket_file_ops = {
    .owner =    THIS_MODULE,
    .llseek =    no_llseek,
    .aio_read =    sock_aio_read,
    .aio_write =    sock_aio_write,
    .poll =        sock_poll,
    .unlocked_ioctl = sock_ioctl,
#ifdef CONFIG_COMPAT
    .compat_ioctl = compat_sock_ioctl,
#endif
    .mmap =        sock_mmap,
    .open =        sock_no_open,    /* special open code to disallow open via /proc */
    .release =    sock_close,
    .fasync =    sock_fasync,
    .sendpage =    sock_sendpage,
    .splice_write = generic_splice_sendpage,
};
网卡控制因为涉及到的知识点比较多,上面只是从宏观上对数据流程做了一个简单的介绍,深入到其中的每个知识点,都会牵扯出一系列文章,读者需要自己去一个个的慢慢深入,希望本文能够对刚刚接触网络驱动的读者有所帮助和启发【gliethttp.Leith】

wireless extention扩展接口

Blog作者的回复: wlan_add_card=> wlan_create_thread(wlan_service_main_thread, &priv->MainThread, "wlan_main_service"); =>wlan_service_main_thread=>wlan_exec_next_cmd=> 将调用wlan_enter_ps和wlan_exit_ps sbi_interrupt=>从sdio口上传来的中断数据,sdio_irq_thread=>process_sdio_pending_irqs=>调用func->irq_handler(func);即本. 在mmc_signal_sdio_irq=>将调用wake_up_process(host->sdio_irq_thread);来唤醒该irq处理线程,可能还有其他命令需要处理wlan_exec_next_cmd 这个pxamci_irq就是mmc的物理irq中断了,pxamci_irq=>mmc_signal_sdio_irq(host->mmc); wlan_exec_next_cmd=>只要cmd链表上CmdNode还存在, 那么就会执行wlan_dnld_cmd_to_fw(wlan_private * priv, CmdCtrlNode * CmdNode)将CmdNode中的数据下发下去, 然后重新触发wlan_mod_timer(&Adapter->MrvDrvCommandTimer, MRVDRV_TIMER_5S); 也就是wlan_cmd_timeout_func命令超时处理函数, 在cmd已经有了恢复之后,在主线程中调用wlan_process_cmdresp,立即调用wlan_cancel_timer(&Adapter->MrvDrvCommandTimer);来删除定时器 wlan_service_main_thread=>每次唤醒都会检查 ====     /* Execute the next command */     if (!priv->wlan_dev.cmd_sent && !Adapter->CurCmd)         wlan_exec_next_cmd(priv); ==== wlan_prepare_cmd=> wlan_hostcmd_ioctl=> 获取一个空闲的CmdNode节点wlan_get_cmd_node,当完成赋值之后,执行如下语句,将CmdNode节点添加到处理队列中: wlan_insert_cmd_to_pending_q(Adapter, CmdNode, TRUE); wake_up_interruptible(&priv->MainThread.waitQ); 另外在数组中 /*  * iwconfig settable callbacks   */ static const iw_handler wlan_handler[]这个数组中全部是回调函数, /** wlan_handler_def */ struct iw_handler_def wlan_handler_def = {   num_standard:sizeof(wlan_handler) / sizeof(iw_handler),   num_private:sizeof(wlan_private_handler) / sizeof(iw_handler),   num_private_args:sizeof(wlan_private_args) / sizeof(struct iw_priv_args),   standard:(iw_handler *) wlan_handler,   private:(iw_handler *) wlan_private_handler,   private_args:(struct iw_priv_args *) wlan_private_args, #if WIRELESS_EXT > 20   get_wireless_stats:wlan_get_wireless_stats, #endif }; 在wlan_add_card函数中 dev->wireless_handlers = (struct iw_handler_def *) &wlan_handler_def; ===============在kernel的net中使用wireless extention扩展接口 static iw_handler get_handler(struct net_device *dev, unsigned int cmd) {     /* Don't "optimise" the following variable, it will crash */     unsigned int    index;        /* *MUST* be unsigned */     /* Check if we have some wireless handlers defined */     if (dev->wireless_handlers == NULL)         return NULL;     /* Try as a standard command */     index = cmd - SIOCIWFIRST;     if (index < dev->wireless_handlers->num_standard)         return dev->wireless_handlers->standard[index];     /* Try as a private command */     index = cmd - SIOCIWFIRSTPRIV;     if (index < dev->wireless_handlers->num_private)         return dev->wireless_handlers->private[index];     /* Not found */     return NULL; } =>sock_ioctl =>dev_ioctl +++/* Take care of Wireless Extensions */ +++if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) +++return wext_handle_ioctl(net, &ifr, cmd, arg); =>wext_handle_ioctl =>wireless_process_ioctl(struct net *net, struct ifreq *ifr, unsigned int cmd) =>get_handler(dev, cmd);如果没有实现该cmd,那么将调用dev->do_ioctl来处理, wlan_reassoc_timer_func=> wmm_start_queue=> wlan_tx_packet=> wlan_tx_timeout=> wlan_remove_card=> wlan_hostcmd_ioctl=> wlan_auto_deep_sleep=> wlan_set_deep_sleep=> wlan_prepare_cmd=> wlan_cmd_timeout_func=> 将调用wake_up_interruptible(&priv->MainThread.waitQ);唤醒wlan_service_main_thread主处理线程. wlan_hard_start_xmit=>wlan_tx_packet发送数据包 dev->tx_timeout = wlan_tx_timeout; wlan_initialize_timer(&Adapter->MrvDrvCommandTimer, wlan_cmd_timeout_func, priv); int wlan_do_ioctl(struct net_device *dev, struct ifreq *req, int cmd) {     ...         case WLAN_WAKEUP_MT:             if (wrq->u.data.length > 0)                 Adapter->IntCounter++;             wake_up_interruptible(&priv->MainThread.waitQ);             break;     ... } 在wlan_process_cmdresp()处理完该cmd之后,调用 wlan_insert_cmd_to_free_q=>wlan_clean_cmd_noder,从命令链表上删除已经处理完成的cmd_node, wlan_clean_cmd_noder然后pTempNode->CmdWaitQWoken = TRUE;同时如果该cmd_node是一个被阻塞等待的,那么唤醒等待的程序. wake_up_interruptible(&pTempNode->cmdwait_q);