LINUX TC:HTB相关源码

 

3、             什么时候调用qdisc_create函数

在函数tc_modify_qdisc中，会调用qdisc_create函数，而tc_modify_qdisc这个函数在pktsched_init(void)中，被注册到系统中

CODE

static int __init pktsched_init(void) {        int err;          err = register_pernet_subsys(&psched_net_ops);        if (err) {               printk(KERN_ERR "pktsched_init: "                      "cannot initialize per netns operations/n");               return err;        } /* *在系统默认的情况下， 系统内只注册了pfif0，bfifo，mq_qdisc，pfifl_head_drop *这几种qdisc */        register_qdisc(&pfifo_qdisc_ops);        register_qdisc(&bfifo_qdisc_ops);        register_qdisc(&pfifo_head_drop_qdisc_ops);        register_qdisc(&mq_qdisc_ops);          rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL);        rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL);        rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc);        rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL);        rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL);        rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass);          return 0; }

下面看看rtnl_register函数具体做了哪些事情.

/** *rtnl_register函数会调用__rtnl_register函数来完成注册（undone） */ void rtnl_register(int protocol, int msgtype,                  rtnl_doit_func doit, rtnl_dumpit_func dumpit) {        if (__rtnl_register(protocol, msgtype, doit, dumpit) < 0)               panic("Unable to register rtnetlink message handler, "                     "protocol = %d, message type = %d/n",                     protocol, msgtype); } ------------------------------------------------------------------------------------------------------- /**  * __rtnl_register - Register a rtnetlink message type  * @protocol: Protocol family or PF_UNSPEC  * @msgtype: rtnetlink message type  * @doit: Function pointer called for each request message  * @dumpit: Function pointer called for each dump request (NLM_F_DUMP) message  *  * Registers the specified function pointers (at least one of them has  * to be non-NULL) to be called whenever a request message for the  * specified protocol family and message type is received.  *  * The special protocol family PF_UNSPEC may be used to define fallback  * function pointers for the case when no entry for the specific protocol  * family exists.  *  * Returns 0 on success or a negative error code.  */ int __rtnl_register(int protocol, int msgtype,                   rtnl_doit_func doit, rtnl_dumpit_func dumpit) {        struct rtnl_link *tab;        int msgindex;          BUG_ON(protocol < 0 || protocol > RTNL_FAMILY_MAX);        msgindex = rtm_msgindex(msgtype);          tab = rtnl_msg_handlers[protocol];        if (tab == NULL) {               tab = kcalloc(RTM_NR_MSGTYPES, sizeof(*tab), GFP_KERNEL);               if (tab == NULL)                      return -ENOBUFS;                 rtnl_msg_handlers[protocol] = tab;        }          if (doit)               tab[msgindex].doit = doit;          if (dumpit)               tab[msgindex].dumpit = dumpit;          return 0; } EXPORT_SYMBOL_GPL(__rtnl_register);

 

static struct rtnl_link *rtnl_msg_handlers[RTNL_FAMILY_MAX + 1];

 

看了这么的代码，得到tc_modify_qdisc 会被注册到rtnl_msg_handlers中。

rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL);

首先通过PF_UNSPEC 在rtnl_msg_handlers找到相应的rtnl_link，注册到rtnl_link[RTM_NEWQDISC].doit中。即：

rtnl_msg_handlers[PF_UNSPEC][RTM_NEWQDISC].doit=tc_modify_qdisc

尽管知道了tc_modify_qdisc会被注册到rtnl_msg_handler中，但是不知道何时调用它？

其实在系统中：

rtnetlink_init会调用函数register_pernet_subsys(&rtnetlink_net_ops) rtnetlink_net_ops： static struct pernet_operations rtnetlink_net_ops = {        .init = rtnetlink_net_init,        .exit = rtnetlink_net_exit, }; 而在函数register_pernet_subsys最终会调用rtnetlink_net_init

下面看看rtnetlink_net_init函数

static int __net_init rtnetlink_net_init(struct net *net) {        struct sock *sk;        sk = netlink_kernel_create(net, NETLINK_ROUTE, RTNLGRP_MAX,                                rtnetlink_rcv, &rtnl_mutex, THIS_MODULE);        if (!sk)               return -ENOMEM;        net->rtnl = sk;        return 0; }

在rtnetlink_net_init函数中，调用netlink_kernel_create函数，这个函数会把rtnetlink_rcv函数赋值给创建的netlink_sock->netlink_rcv,如果在netlink上接受数据时，会调用rtnetlink_rcv函数来处理。

再看看rtnetlink_rcv函数

static void rtnetlink_rcv(struct sk_buff *skb) {        rtnl_lock();        netlink_rcv_skb(skb, &rtnetlink_rcv_msg);        rtnl_unlock(); }

在rtnetlink_rcv函数中，调用netlink_rcv_skb函数，这个函数会回调函数rtnetlink_rcv_msg。rtnetlink_rcv_msg正是我们找寻的函数。

Ok，下面看看rtnetlink_rcv_msg函数是如何找到相应的消息处理函数的。

static int rtnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh)

rtnetlink_rcv_msg函数的有sk_buff和nlmsghdr。Sk_buff为要处理的数据，nlmsghdr则包含了这个消息的类型等参数。

Ø  通过nlmsghdr找到对应的family和type。

Ø  （如果是需要调用doit函数）调用rtnl_get_doit(family, type)来得到相应的doit函数。如果是要调用tc_modify_qdisc函数，那么family=PF_UNSPACE, type=RTM_NEWQDISC

Ø  最后调用得到的doit函数，来处理数据

return doit(skb, nlh, (void *)&rta_buf[0])

 

CODE:rtnl_get_doit

static rtnl_doit_func rtnl_get_doit(int protocol, int msgindex) {        struct rtnl_link *tab;          if (protocol <= RTNL_FAMILY_MAX)               tab = rtnl_msg_handlers[protocol];        else               tab = NULL;          if (tab == NULL || tab[msgindex].doit == NULL)               tab = rtnl_msg_handlers[PF_UNSPEC];          return tab ? tab[msgindex].doit : NULL; }

 

使用strace tc qdisc add dev eth0 root handle 1: default 12 的结果如下：

（先列在吧，我也不知道是怎么回事L）

execve("/sbin/tc", ["tc", "qdisc", "add", "dev", "eth0", "root", "handle", "1:", "htb", "default", "12"], [/* 21 vars */]) = 0 …………………………………………………………………… open("/proc/net/psched", O_RDONLY)      = 3 fstat64(3, {st_mode=S_IFREG|0444, st_size=0, ...}) = 0 mmap2(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xb7fdb000 read(3, "00000001 00000001 000f4240 00000"..., 4096) = 36 close(3)                                = 0 munmap(0xb7fdb000, 4096)                = 0 socket(PF_NETLINK, SOCK_RAW, 0)         = 3 setsockopt(3, SOL_SOCKET, SO_SNDBUF, [32768], 4) = 0 setsockopt(3, SOL_SOCKET, SO_RCVBUF, [32768], 4) = 0 bind(3, {sa_family=AF_NETLINK, pid=0, groups=00000000}, 12) = 0 getsockname(3, {sa_family=AF_NETLINK, pid=23391, groups=00000000}, [12]) = 0 time(NULL)                              = 1306133974 open("/usr/lib/tc/q_htb.so", O_RDONLY)  = -1 ENOENT (No such file or directory) sendto(3, "/24/0/0/0/22/0/1/3/327/5/332M/0/0/0/0/0/0/0/0", 20, 0, {sa_family=AF_NETLINK, pid=0, groups=00000000}, 12) = 20 recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=00000000}, msg_iov(1)=[{"/364/0/0/0/20/0/2/0/327/5/332M_[/0/0/0/0/4/3/1/0/0/0I/0/1/0/0/0/0/0"..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 736 recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=00000000}, msg_iov(1)=[{"/24/0/0/0/3/0/2/0/327/5/332M_[/0/0/0/0/0/0/1/0/0/0I/0/1/0/0/0/0/0"..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 20 sendmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=00000000}, msg_iov(1)=[{"H/0/0/0$/0/5/6/330/5/332M/0/0/0/0/0/0/0/0/2/0/0/0/0/0/1/0/377/377/377/377"..., 72}], msg_controllen=0, msg_flags=0}, 0) = 72 recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=00000000}, msg_iov(1)=[{"$/0/0/0/2/0/0/0/330/5/332M_[/0/0/0/0/0/0H/0/0/0$/0/5/6/330/5/332M"..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 36 close(3)                                = 0 exit_group(0)