When an acceleration connection forwards the data packet, SNAT and DNAT will be performed on the packet; that is, the source and destination addresses of the data packet will be modified. The packet source address seen by the origin server will be the forwarding IP address of the acceleration connection, rather than the real client IP. To pass the client IP to the server, the acceleration connection will include the client IP and port in the custom tcp option
field when forwarding the packet, as shown below:
/*
* insert client ip in tcp option.
* must be 4 bytes alignment.
*/
struct ip_vs_tcpo_addr{
__u8 opcode;
__u8 opsize;
__u16 port;
__u32 addr;
};
After the Linux kernel has received the ACK packet of three-way handshake while listening the socket, its status is changed from SYN_REVC to TCP_ESTABLISHED. In this case, the kernel calls the tcp_v4_syn_recv_sock
function. The Hook function tcp_v4_syn_recv_sock_toa
calls the original tcp_v4_syn_recv_sock
function, then extracts TOA OPTION
from the TCP OPTION
by calling the get_toa_data
function, and saves it in the sk_user_data
field. After the above call is completed, the kernel calls inet_getname_toa hook inet_getname
to obtain the source IP and port. It first calls the original inet_getname
, and check whether the sk_user_data
field is empty. If the real IP and port can be extracted from this field, then replace the returned values of inet_getname
with these two values.
The server program calls getpeername
in the user mode, and the client's original IP and port are returned.
Was this page helpful?