Go 捕获网卡流量使用最多的库为 github.com/google/gopacket，需要依赖 libpcap 导致必须开启 CGO 才能够进行编译。
为了减少对环境的依赖可以使用原始套接字捕获网卡流量，然后使用 gopacket 的协议解析功能，这样就省去了解析这部分的工作量，正确性也可以得到保证，同时 CGO 也可以关闭。
cilium 里有一个原始套接字打开的测试用例：
// Both openRawSock and htons are available in
// https://github.com/cilium/ebpf/blob/master/example_sock_elf_test.go.
// MIT license.
func OpenRawSocket(index int) (int, error) {
sock, err := syscall.Socket(syscall.AF_PACKET, syscall.SOCK_RAW|syscall.SOCK_NONBLOCK|syscall.SOCK_CLOEXEC, int(htons(syscall.ETH_P_ALL)))
if err != nil {
return 0, err
}
sll := syscall.SockaddrLinklayer{Ifindex: index, Protocol: htons(syscall.ETH_P_ALL)}
if err := syscall.Bind(sock, &sll); err != nil {
syscall.Close(sock)
return 0, err
}
return sock, nil
}
// htons converts the unsigned short integer hostshort from host byte order to network byte order.
func htons(i uint16) uint16 {
b := make([]byte, 2)
binary.BigEndian.PutUint16(b, i)
return *(*uint16)(unsafe.Pointer(&b[0]))
}
但是这个示例有一个问题，只能拿到本机流量。
通过 tcpdump 是可以抓到经过网桥的转发流量的，我们使用 strace 对 tcpdump 进行跟踪分析
root@localhost:~# strace -f tcpdump -i b_2_0 arp -nne
...
socket(AF_PACKET, SOCK_RAW, htons(0 /* ETH_P_??? */)) = 4
ioctl(4, SIOCGIFINDEX, {ifr_name="lo", ifr_ifindex=1}) = 0
ioctl(4, SIOCGIFHWADDR, {ifr_name="b_2_0", ifr_hwaddr={sa_family=ARPHRD_ETHER, sa_data=4e:59:d6:32:f6:42}}) = 0
newfstatat(AT_FDCWD, "/sys/class/net/b_2_0/wireless", 0x7ffdf063bc50, 0) = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/sys/class/net/b_2_0/dsa/tagging", O_RDONLY) = -1 ENOENT (No such file or directory)
ioctl(4, SIOCGIFINDEX, {ifr_name="b_2_0", ifr_ifindex=6053}) = 0
bind(4, {sa_family=AF_PACKET, sll_protocol=htons(0 /* ETH_P_??? */), sll_ifindex=if_nametoindex("b_2_0"), sll_hatype=ARPHRD_NETROM, sll_pkttype=PACKET_HOST, sll_halen=0}, 20) = 0
getsockopt(4, SOL_SOCKET, SO_ERROR, [0], [4]) = 0
setsockopt(4, SOL_PACKET, PACKET_ADD_MEMBERSHIP, {mr_ifindex=if_nametoindex("b_2_0"), mr_type=PACKET_MR_PROMISC, mr_alen=0, mr_address=4e:59:d6:32:f6:42}, 16) = 0
getsockopt(4, SOL_SOCKET, SO_BPF_EXTENSIONS, [64], [4]) = 0
mmap(NULL, 266240, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7fec47cbe000
看到有一个 setsockopt(PACKET_MR_PROMISC) 设置，看起来是开启的混杂模式，查看资料看到这是一个针对套接字级别的混杂模式。
由于之前看过 suricata 的代码，看看它是怎么做的，直接在 suricata 的仓库里面搜索 PACKET_MR_PROMISC 关键字，出现代码
memset(&sock_params, 0, sizeof(sock_params));
sock_params.mr_type = PACKET_MR_PROMISC;
sock_params.mr_ifindex = bind_address.sll_ifindex;
r = setsockopt(ptv->socket, SOL_PACKET, PACKET_ADD_MEMBERSHIP,(void *)&sock_params, sizeof(sock_params));
if (r < 0) {
SCLogError("%s: failed to set promisc mode: %s", devname, strerror(errno));
goto socket_err;
}
套接字设置混杂模式的 Go 实现如下
// Set socket level PROMISC mode
err = unix.SetsockoptPacketMreq(sock, syscall.SOL_PACKET, syscall.PACKET_ADD_MEMBERSHIP, &unix.PacketMreq{Type: unix.PACKET_MR_PROMISC, Ifindex: int32(index)})
if err != nil {
syscall.Close(sock)
return 0, err
}
目前只能拿到普通的以太网流量，如果还需要拿到 VLAN Id 的话，需要设置 PACKET_AUXDATA，参考 man packet
PACKET_AUXDATA (since Linux 2.6.21)
If this binary option is enabled, the packet socket passes
a metadata structure along with each packet in the
recvmsg(2) control field. The structure can be read with
cmsg(3). It is defined as
struct tpacket_auxdata {
__u32 tp_status;
__u32 tp_len; /* packet length */
__u32 tp_snaplen; /* captured length */
__u16 tp_mac;
__u16 tp_net;
__u16 tp_vlan_tci;
__u16 tp_vlan_tpid; /* Since Linux 3.14; earlier, these
were unused padding bytes */
};
Go 的实现如下
// Enable PACKET_AUXDATA option for VLAN
if err := syscall.SetsockoptInt(sock, syscall.SOL_PACKET, unix.PACKET_AUXDATA, 1); err != nil {
syscall.Close(sock)
return 0, err
}
完整的实现如下
func OpenRawSocket(index int) (int, error) {
sock, err := syscall.Socket(syscall.AF_PACKET, syscall.SOCK_RAW|syscall.SOCK_NONBLOCK|syscall.SOCK_CLOEXEC, int(htons(syscall.ETH_P_ALL)))
if err != nil {
return 0, err
}
// Enable PACKET_AUXDATA option for VLAN
if err := syscall.SetsockoptInt(sock, syscall.SOL_PACKET, unix.PACKET_AUXDATA, 1); err != nil {
syscall.Close(sock)
return 0, err
}
// Set socket level PROMISC mode
err = unix.SetsockoptPacketMreq(sock, syscall.SOL_PACKET, syscall.PACKET_ADD_MEMBERSHIP, &unix.PacketMreq{Type: unix.PACKET_MR_PROMISC, Ifindex: int32(index)})
if err != nil {
syscall.Close(sock)
return 0, err
}
sll := syscall.SockaddrLinklayer{Ifindex: index, Protocol: htons(syscall.ETH_P_ALL)}
if err := syscall.Bind(sock, &sll); err != nil {
syscall.Close(sock)
return 0, err
}
return sock, nil
}
这里使用 select(2) 简单地对 fd 进行监听，使用 recvmsg(2) 来读取数据，包括 VLAN tag。
实现如下
package pcap
import (
"context"
"syscall"
)
func FD_SET(fd int, p *syscall.FdSet) { p.Bits[fd/64] |= 1 << (uint(fd) % 64) }
func FD_CLR(fd int, p *syscall.FdSet) { p.Bits[fd/64] &^= 1 << (uint(fd) % 64) }
func FD_ISSET(fd int, p *syscall.FdSet) bool { return p.Bits[fd/64]&(1<<(uint(fd)%64)) != 0 }
func FD_ZERO(p *syscall.FdSet) {
for i := range p.Bits {
p.Bits[i] = 0
}
}
type RecvmsgHandler func(buf []byte, n int, oob []byte, oobn int, err error) error
func RecvmsgLoop(ctx context.Context, sockfd int, fn RecvmsgHandler) error {
buf := make([]byte, 1024*64)
oob := make([]byte, syscall.CmsgSpace(1024))
readfds := syscall.FdSet{}
for {
select {
case <-ctx.Done():
return ctx.Err()
default:
}
FD_ZERO(&readfds)
FD_SET(sockfd, &readfds)
tv := syscall.Timeval{Sec: 0, Usec: 100000} // 100ms
nfds, err := syscall.Select(sockfd+1, &readfds, nil, nil, &tv)
if err != nil {
continue
}
if nfds > 0 && FD_ISSET(sockfd, &readfds) {
n, oobn, _, _, err := syscall.Recvmsg(sockfd, buf, oob, 0)
err = fn(buf, n, oob, oobn, err)
if err != nil {
return err
}
}
}
}
VLAN 数据的解析逻辑如下
func decodeVlanIdByAuxData(oob []byte) (uint16, error) {
msgs, err := syscall.ParseSocketControlMessage(oob)
if err != nil {
return 0, err
}
for _, m := range msgs {
if m.Header.Level == syscall.SOL_PACKET && m.Header.Type == 8 && len(m.Data) >= 20 {
auxdata := unix.TpacketAuxdata{
Status: binary.LittleEndian.Uint32(m.Data[0:4]),
Vlan_tci: binary.LittleEndian.Uint16(m.Data[16:18]),
}
if auxdata.Status&unix.TP_STATUS_VLAN_VALID != 0 {
return auxdata.Vlan_tci, nil
}
}
}
return 0, nil
}
以上代码都在实际的场景中使用，只是稍微修改了一点细节以及使用 epoll(2) 来监听，结合 sync.Pool 和精简了解析逻辑，性能尚可能够满足要求。