Linux内核bridge中的数据包处理流程

本文档的Copyleft归yfydz所有，使用GPL发布，可以自由拷贝，转载，转载时请保持文档的完整性，严禁用于任何商业用途。
msn: [email protected]
来源：http://yfydz.cublog.cn

1. 前言
本文简要介绍数据包在进入桥网卡后在Linux网络协议栈的处理流程，并描述netfilter的hook点的挂接处理情况，具体各部分的详细处理待后续文章中说明。

以下内核代码版本为2.6.19.2.

2. 函数处理流程

bridge入口点handle_bridge()

/* net/core/dev.c */

int netif_receive_skb(struct sk_buff *skb)
{
......
if (handle_bridge(&skb, &pt_prev, &ret, orig_dev))
   goto out;
......
}

bridge基本挂接点处理函数:br_handle_frame_hook()

static __inline__ int handle_bridge(struct sk_buff **pskb,
        struct packet_type **pt_prev, int *ret,
        struct net_device *orig_dev)
{
struct net_bridge_port *port;

if ((*pskb)->pkt_type == PACKET_LOOPBACK ||
     (port = rcu_dereference((*pskb)->dev->br_port)) == NULL)
   return 0;

if (*pt_prev) {
   *ret = deliver_skb(*pskb, *pt_prev, orig_dev);
   *pt_prev = NULL;
}

return br_handle_frame_hook(port, pskb);
}

bridge_handle_frame_hook()的实际实现:

/* net/bridge/br.c */

static int __init br_init(void)
{
......
br_handle_frame_hook = br_handle_frame;
......
}

br_handle_frame: PF_BEIDGE的prerouting点

/* net/bridge/br_input.c */

int br_handle_frame(struct net_bridge_port *p, struct sk_buff **pskb)
{
struct sk_buff *skb = *pskb;
const unsigned char *dest = eth_hdr(skb)->h_dest;

if (!is_valid_ether_addr(eth_hdr(skb)->h_source))
   goto err;

if (unlikely(is_link_local(dest))) {
// 自身包进入PF_BEIDGE的INPUT点, 一般处理的包数不多
   skb->pkt_type = PACKET_HOST;
// 正常是返回1的, 然后就返回1, 表示桥模块全权处理该包了
   return NF_HOOK(PF_BRIDGE, NF_BR_LOCAL_IN, skb, skb->dev,
          NULL, br_handle_local_finish) != 0;
}

if (p->state == BR_STATE_FORWARDING || p->state == BR_STATE_LEARNING) {
// br_should_route_hook函数一般没定义
   if (br_should_route_hook) {
    if (br_should_route_hook(pskb))
     return 0;
    skb = *pskb;
    dest = eth_hdr(skb)->h_dest;
   }

   if (!compare_ether_addr(p->br->dev->dev_addr, dest))
    skb->pkt_type = PACKET_HOST;

// PF_BRIDGE的prerouting处理结束后进入br_handle_frame_finish
   NF_HOOK(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
    br_handle_frame_finish);
// 处理后始终返回1, 表示不再进行其他协议族处理,该数据包已经完全由bridge处理完毕
   return 1;
}

err:
kfree_skb(skb);
// 处理后始终返回1, 表示不再进行其他协议族处理,该数据包已经完全由bridge处理完毕
return 1;
}

通过br_handle_frame_finish进入bridge的转发:

/* note: already called with rcu_read_lock (preempt_disabled) */
int br_handle_frame_finish(struct sk_buff *skb)
{
const unsigned char *dest = eth_hdr(skb)->h_dest;
struct net_bridge_port *p = rcu_dereference(skb->dev->br_port);
struct net_bridge *br;
struct net_bridge_fdb_entry *dst;
int passedup = 0;

if (!p || p->state == BR_STATE_DISABLED)
   goto drop;

/* insert into forwarding database after filtering to avoid spoofing */
br = p->br;
br_fdb_update(br, p, eth_hdr(skb)->h_source);

if (p->state == BR_STATE_LEARNING)
   goto drop;

if (br->dev->flags & IFF_PROMISC) {
   struct sk_buff *skb2;

   skb2 = skb_clone(skb, GFP_ATOMIC);
   if (skb2 != NULL) {
    passedup = 1;
    br_pass_frame_up(br, skb2);
   }
}

if (is_multicast_ether_addr(dest)) {
// 多播转发,也是调用广播处理
   br->statistics.multicast++;
   br_flood_forward(br, skb, !passedup);
   if (!passedup)
    br_pass_frame_up(br, skb);
   goto out;
}
// 根据目的MAC找目的出口
dst = __br_fdb_get(br, dest);
if (dst != NULL && dst->is_local) {
   if (!passedup)
    br_pass_frame_up(br, skb);
   else
    kfree_skb(skb);
   goto out;
}

if (dst != NULL) {
// 单播转发
   br_forward(dst->dst, skb);
   goto out;
}
// 广播转发
br_flood_forward(br, skb, 0);

out:
return 0;
drop:
kfree_skb(skb);
goto out;
}

广播/多播转发: br_flood_forward/br_flood

/* called under bridge lock */
void br_flood_forward(struct net_bridge *br, struct sk_buff *skb, int clone)
{
br_flood(br, skb, clone, __br_forward);
}

/* called under bridge lock */
static void br_flood(struct net_bridge *br, struct sk_buff *skb, int clone,
void (*__packet_hook)(const struct net_bridge_port *p,
         struct sk_buff *skb))
{
struct net_bridge_port *p;
struct net_bridge_port *prev;

if (clone) {
   struct sk_buff *skb2;

   if ((skb2 = skb_clone(skb, GFP_ATOMIC)) == NULL) {
    br->statistics.tx_dropped++;
    return;
   }

   skb = skb2;
}

prev = NULL;

list_for_each_entry_rcu(p, &br->port_list, list) {
   if (should_deliver(p, skb)) {
    if (prev != NULL) {
     struct sk_buff *skb2;

     if ((skb2 = skb_clone(skb, GFP_ATOMIC)) == NULL) {
      br->statistics.tx_dropped++;
      kfree_skb(skb);
      return;
     }
// 这里实际是__br_forward
     __packet_hook(prev, skb2);
    }

    prev = p;
   }
}

if (prev != NULL) {
// 这里实际是__br_forward
   __packet_hook(prev, skb);
   return;
}

kfree_skb(skb);
}

单播转发: br_forward

/* net/bridge/br_forward.c */
/* called with rcu_read_lock */
void br_forward(const struct net_bridge_port *to, struct sk_buff *skb)
{
if (should_deliver(to, skb)) {
// 也是调用__br_forward
   __br_forward(to, skb);
   return;
}

kfree_skb(skb);
}

FORWARD点:
static void __br_forward(const struct net_bridge_port *to, struct sk_buff *skb)
{
struct net_device *indev;

indev = skb->dev;
skb->dev = to->dev;
skb->ip_summed = CHECKSUM_NONE;
// 进入PF_BRIDGE的forward hook, 结束后进入br_forward_finish()
NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, indev, skb->dev,
    br_forward_finish);
}

POSTROUTING点:
// 从FORWARD点处理后直接进入POSTROUTING点处理
int br_forward_finish(struct sk_buff *skb)
{
// 进入PF_BRIDGE的postrouting hook, 结束后进入br_dev_queue_push_xmit()
return NF_HOOK(PF_BRIDGE, NF_BR_POST_ROUTING, skb, NULL, skb->dev,
         br_dev_queue_push_xmit);

}

数据包发出:

int br_dev_queue_push_xmit(struct sk_buff *skb)
{
/* drop mtu oversized packets except gso */
if (packet_length(skb) > skb->dev->mtu && !skb_is_gso(skb))
   kfree_skb(skb);
else {
   /* ip_refrag calls ip_fragment, doesn't copy the MAC header. */
   if (nf_bridge_maybe_copy_header(skb))
    kfree_skb(skb);
   else {
    skb_push(skb, ETH_HLEN);
// 此处调用dev设备的hard_start_xmit()函数
    dev_queue_xmit(skb);
   }
}

return 0;
}

桥网卡设备的hard_start_xmit()函数定义为:
/* net/bridge/br_device.c */

void br_dev_setup(struct net_device *dev)
{
......
dev->hard_start_xmit = br_dev_xmit;
......
}

/* net device transmit always called with no BH (preempt_disabled) */
int br_dev_xmit(struct sk_buff *skb, struct net_device *dev)
{
struct net_bridge *br = netdev_priv(dev);
const unsigned char *dest = skb->data;
struct net_bridge_fdb_entry *dst;

br->statistics.tx_packets++;
br->statistics.tx_bytes += skb->len;

skb->mac.raw = skb->data;
skb_pull(skb, ETH_HLEN);

if (dest[0] & 1)
// 多播发送
   br_flood_deliver(br, skb, 0);
else if ((dst = __br_fdb_get(br, dest)) != NULL)
// 单播发送
   br_deliver(dst->dst, skb);
else
// 广播发送
   br_flood_deliver(br, skb, 0);
// 这些发送函数最终都会调用__br_deliver()函数
return 0;
}

/* net/bridge/br_forward.c */
static void __br_deliver(const struct net_bridge_port *to, struct sk_buff *skb)
{
skb->dev = to->dev;
// 此处是PF_BRIDGE的OUTPUT点
NF_HOOK(PF_BRIDGE, NF_BR_LOCAL_OUT, skb, NULL, skb->dev,
    br_forward_finish);
}

总结: PF_BRIDGE中的各个hook点和PF_INET不同, 可用下面的图表示:

PREROUTING --+--FORWARD-----POSTROUTING------+----OUTPUT
               |                               |
               |                               |
              INPUT

3. BF_BRIDGE的hook点

在net/bridge/br_netfilter.c中定义了以下hook点,注意这些hook点主要是PF_BRIDGE协议族的。

/* net/bridge/br_netfilter.c */

/* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
* PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
* ip_refrag() can return NF_STOLEN. */
static struct nf_hook_ops br_nf_ops[] = {
// PF_BRIDGE的挂接点
// PREROUTING点
{ .hook = br_nf_pre_routing,
   .owner = THIS_MODULE,
   .pf = PF_BRIDGE,
   .hooknum = NF_BR_PRE_ROUTING,
   .priority = NF_BR_PRI_BRNF, },
// INPUT点
{ .hook = br_nf_local_in,
   .owner = THIS_MODULE,
   .pf = PF_BRIDGE,
   .hooknum = NF_BR_LOCAL_IN,
   .priority = NF_BR_PRI_BRNF, },
// FORWARD点
{ .hook = br_nf_forward_ip,
   .owner = THIS_MODULE,
   .pf = PF_BRIDGE,
   .hooknum = NF_BR_FORWARD,
   .priority = NF_BR_PRI_BRNF - 1, },
// FORWARD点
{ .hook = br_nf_forward_arp,
   .owner = THIS_MODULE,
   .pf = PF_BRIDGE,
   .hooknum = NF_BR_FORWARD,
   .priority = NF_BR_PRI_BRNF, },
// OUTPUT点
{ .hook = br_nf_local_out,
   .owner = THIS_MODULE,
   .pf = PF_BRIDGE,
   .hooknum = NF_BR_LOCAL_OUT,
   .priority = NF_BR_PRI_FIRST, },
// POSTROUTING点
{ .hook = br_nf_post_routing,
   .owner = THIS_MODULE,
   .pf = PF_BRIDGE,
   .hooknum = NF_BR_POST_ROUTING,
   .priority = NF_BR_PRI_LAST, },

// 后面是PF_INET/PF_INET6的挂接点, 其实也没进行什么数据包操作,
// 就是自身的输入输出包不通过桥处理,要短路掉
{ .hook = ip_sabotage_in,
   .owner = THIS_MODULE,
   .pf = PF_INET,
   .hooknum = NF_IP_PRE_ROUTING,
   .priority = NF_IP_PRI_FIRST, },
{ .hook = ip_sabotage_in,
   .owner = THIS_MODULE,
   .pf = PF_INET6,
   .hooknum = NF_IP6_PRE_ROUTING,
   .priority = NF_IP6_PRI_FIRST, },
{ .hook = ip_sabotage_out,
   .owner = THIS_MODULE,
   .pf = PF_INET,
   .hooknum = NF_IP_FORWARD,
   .priority = NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD, },
{ .hook = ip_sabotage_out,
   .owner = THIS_MODULE,
   .pf = PF_INET6,
   .hooknum = NF_IP6_FORWARD,
   .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_FORWARD, },
{ .hook = ip_sabotage_out,
   .owner = THIS_MODULE,
   .pf = PF_INET,
   .hooknum = NF_IP_LOCAL_OUT,
   .priority = NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
{ .hook = ip_sabotage_out,
   .owner = THIS_MODULE,
   .pf = PF_INET6,
   .hooknum = NF_IP6_LOCAL_OUT,
   .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
{ .hook = ip_sabotage_out,
   .owner = THIS_MODULE,
   .pf = PF_INET,
   .hooknum = NF_IP_POST_ROUTING,
   .priority = NF_IP_PRI_FIRST, },
{ .hook = ip_sabotage_out,
   .owner = THIS_MODULE,
   .pf = PF_INET6,
   .hooknum = NF_IP6_POST_ROUTING,
   .priority = NF_IP6_PRI_FIRST, },
};

// PF_BRIDGE的PRROUTING点处理函数
static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff **pskb,
          const struct net_device *in,
          const struct net_device *out,
          int (*okfn)(struct sk_buff *))
{
......

// 此处继续调用PF_INET族的PREROUTING点的hook处理
NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, skb->dev, NULL,
   br_nf_pre_routing_finish);

return NF_STOLEN;

inhdr_error:
//      IP_INC_STATS_BH(IpInHdrErrors);
out:
return NF_DROP;
}

// PF_BRIDGE的FORWARD点处理
static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff **pskb,
         const struct net_device *in,
         const struct net_device *out,
         int (*okfn)(struct sk_buff *))
{
......
// 此处继续调用PF_INET/PF_INET6族的FORWARD点的hook处理
NF_HOOK(pf, NF_IP_FORWARD, skb, bridge_parent(in), parent,
   br_nf_forward_finish);

return NF_STOLEN;
}

// PF_BRIDGE的OUTPUT点处理
static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff **pskb,
        const struct net_device *in,
        const struct net_device *out,
        int (*okfn)(struct sk_buff *))
{
......
/* IP forwarded traffic has a physindev, locally
* generated traffic hasn't. */
if (realindev != NULL) {
   if (!(nf_bridge->mask & BRNF_DONT_TAKE_PARENT)) {
    struct net_device *parent = bridge_parent(realindev);
    if (parent)
     realindev = parent;
   }
// 此处继续调用PF_INET/PF_INET6族的FORWARD点的hook处理, 不过优先权值要在//

NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD + 1以上
   NF_HOOK_THRESH(pf, NF_IP_FORWARD, skb, realindev,
          realoutdev, br_nf_local_out_finish,
          NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD + 1);
} else {
// 此处继续调用PF_INET/PF_INET6族的FORWARD点的hook处理, 不过优先权值要在
// NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT + 1以上
   NF_HOOK_THRESH(pf, NF_IP_LOCAL_OUT, skb, realindev,
          realoutdev, br_nf_local_out_finish,
          NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT + 1);
}

out:
return NF_STOLEN;
}

// PF_BRIDGE的POSTROUTING点
static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff **pskb,
           const struct net_device *in,
           const struct net_device *out,
           int (*okfn)(struct sk_buff *))
{
......
// 此处继续调用PF_INET/PF_INET6族的POSTROUTING点的hook处理
NF_HOOK(pf, NF_IP_POST_ROUTING, skb, NULL, realoutdev,
   br_nf_dev_queue_xmit);

return NF_STOLEN;

#ifdef CONFIG_NETFILTER_DEBUG
print_error:
if (skb->dev != NULL) {
   printk("[%s]", skb->dev->name);
   if (realoutdev)
    printk("[%s]", realoutdev->name);
}
printk(" head:%p, raw:%p, data:%p/n", skb->head, skb->mac.raw,
        skb->data);
dump_stack();
return NF_ACCEPT;
#endif
}

由此可见, PF_INET的各个hook点也被PF_BRIDGE的各个hook点调用，因此可以在桥网卡中进行过滤或NAT等操作。

4. 结论

BRIDGE的数据处理流程是是一个独立的处理过程, 如果处理正常的话就不再返回到其他协议处理。
在桥的处理层次也和IP协议一样，可以挂接多个PF_BRIDGE的挂接点，这些挂接点中又调用了PF_INET族的挂接点，从而实现了桥下的过滤、NAT等功能。