Linux: When sending Ethernet frames the ethertype is being re-written
I have written a C program that is writing Ethernet frames directly onto the wire (which runs in two modes, sender or receiver). The sender is sending the frames with two VLAN tags on them (QinQ) but strangely when the frame reaches the receiver the ethertype has changed to that of a standard (single) VLAN encapsulated frame. Is it possible the NIC is doing this, or does Linux not allow this? Wireshark shows the same behaviour as tcpdump.
To explain the image below the sender is sending frames to the Ethernet broadcast address FF:FF:FF:FF:FF:FF to find the receiver (these are two test machines connected via a crossover cable but the result below is the same with a switch or hub). As you can see the frames are coming in with two VLAN tags on them, the outer tag has an ethertype of 0x8100 and a VLAN ID of 40, the inner VLAN has ethertype 0x8100 and a VLAN ID of 20. As we all know though, with QinQ frames the outer frame should have an ethertype of 0x88a8!
When the frames are sent from the sender in my application they do have the outer ethertype of 0x88a8 but as per the picture below they are received with 0x8100 on both the inner and outer ethertypes. The highlighted text is the receiver sending back a reply, as you can see the frames have 0x88a8 on the outer frame and 0x8100 on the inner. tcpdump on the other machine shows the same (it's the same code! Frames are sent with 0x88a8 outer 0x8100 inner but always received as 0x8100 outer and 0x8100 inner).
void BuildHeaders(char* &txBuffer, unsigned char (&destMAC)[6],
unsigned char (&sourceMAC)[6], short &PCP, short &vlanID,
short &qinqID, short &qinqPCP, int &headersLength)
{
int offset = 0;
short TPI = 0;
short TCI = 0;
short *p = &TPI;
short *c = &TCI;
short vlanIDtemp;
// Copy the destination and source MAC addresses
memcpy((void*)txBuffer, (void*)destMAC, ETH_ALEN);
memcpy((void*)(txBuffer+ETH_ALEN), (void*)sourceMAC, ETH_ALEN);
offset = (ETH_ALEN*2);
// Add on the QinQ Tag Protocol Identifier
vlanIDtemp = qinq
TPI = htons(0x88a8); //0x88a8 == IEEE802.1ad, 0x9100 == older IEEE802.1QinQ
memcpy((void*)(txBuffer+offset), p, 2);
offset+=2;
// Now build the QinQ Tag Control Identifier:
TCI = (qinqPCP & 0x07) << 5;
qinqID = qinqID >> 8;
TCI = TCI | (qinqID & 0x0f);
qinqID = vlanIDtemp;
qinqID = qinqID << 8;
TCI = TCI | (qinqID & 0xffff);
memcpy((void*)(txBuffer+offset), c, 2);
offset+=2;
// VLAN headers
vlanIDtemp = vlanID;
TPI = htons(0x8100);
memcpy((void*)(txBuffer+offset), p, 2);
offset+=2;
TCI = (PCP & 0x07) << 5;
vlanID = vlanID >> 8;
TCI = TCI | (vlanID & 0x0f);
vlanID = vlanIDtemp;
vlanID = vlanID << 8;
TCI = TCI | (vlanID & 0xffff);
memcpy((void*)(txBuffer+offset), c, 2);
offset+=2;
// Push on the Ethertype (IPv4) for the payload
TPI = htons(0x0800);
memcpy((void*)(txBuffer+offset), p, 2);
offset+=2;
headersLength = offset;
}
sendResult = sendto(sockFD, txBuffer, fSizeTotal, 0, (struct sockaddr*)&socket_address, sizeof(socket_address));
(Fully rewritten to simplify the answer. I also fixed quite a few bugs in my C header and source files listed below.)
There was a discussion about exactly this on the linux-netdev mailing list in April 2014, subject "802.1AD packets - Kernel changes ether type from 88A8 to 8100 on all packets".
It turns out that the kernel does not change the ether type, it simply consumes it on receiving a packet. I show below that it is correctly used for VLAN routing (including separate rules for 802.1AD and 802.1Q VLANs), given a recent enough kernel. Even if the VLAN tag is not used for routing (say, if there are no VLANs configured, or if the 8021q kernel module is not loaded), the VLAN tag is consumed by the kernel.
Thus, the original question, "When sending Ethernet frames the ethertype is being re-written", is incorrect: the ethertype is not being re-written. It is consumed by the kernel.
Because the VLAN tag is consumed by the kernel, libpcap -- which is the packet capture library used by tcpdump, wireshark et al. -- tries to reintroduce it back into the packet headers. Unfortunately, it always uses the 802.1Q VLAN header (8100).
There is a suggested change to libpcap that fixes exactly this problem in libpcap, but as of this writing, it does not seem to have been included yet; you can still see htons(ETH_P_8021Q) hardcoded in several places in the libpcap source file for Linux.
I cannot assume you'll take my word for this, so let me show you how you can ascertain this for yourself.
Let's write a simple packet sender and receiver, that uses the kernel interfaces directly, without the assistance of libpcap.
rawpacket.h:
#ifndef RAWPACKET_H
#define RAWPACKET_H
#include <unistd.h>
#include <sys/socket.h>
#include <sys/ioctl.h>
#include <netpacket/packet.h>
#include <net/ethernet.h>
#include <net/if.h>
#include <arpa/inet.h>
#include <linux/if_ether.h>
#include <string.h>
#include <errno.h>
#include <stdio.h>
static int rawpacket_socket(const int protocol,
const char *const interface,
void *const hwaddr)
{
struct ifreq iface;
struct sockaddr_ll addr;
int socketfd, result;
int ifindex = 0;
if (!interface || !*interface) {
errno = EINVAL;
return -1;
}
socketfd = socket(AF_PACKET, SOCK_RAW, htons(protocol));
if (socketfd == -1)
return -1;
do {
memset(&iface, 0, sizeof iface);
strncpy((char *)&iface.ifr_name, interface, IFNAMSIZ);
result = ioctl(socketfd, SIOCGIFINDEX, &iface);
if (result == -1)
break;
ifindex = iface.ifr_ifindex;
memset(&iface, 0, sizeof iface);
strncpy((char *)&iface.ifr_name, interface, IFNAMSIZ);
result = ioctl(socketfd, SIOCGIFFLAGS, &iface);
if (result == -1)
break;
iface.ifr_flags |= IFF_PROMISC;
result = ioctl(socketfd, SIOCSIFFLAGS, &iface);
if (result == -1)
break;
memset(&iface, 0, sizeof iface);
strncpy((char *)&iface.ifr_name, interface, IFNAMSIZ);
result = ioctl(socketfd, SIOCGIFHWADDR, &iface);
if (result == -1)
break;
memset(&addr, 0, sizeof addr);
addr.sll_family = AF_PACKET;
addr.sll_protocol = htons(protocol);
addr.sll_ifindex = ifindex;
addr.sll_hatype = 0;
addr.sll_pkttype = 0;
addr.sll_halen = ETH_ALEN; /* Assume ethernet! */
memcpy(&addr.sll_addr, &iface.ifr_hwaddr.sa_data, addr.sll_halen);
if (hwaddr)
memcpy(hwaddr, &iface.ifr_hwaddr.sa_data, ETH_ALEN);
if (bind(socketfd, (struct sockaddr *)&addr, sizeof addr))
break;
errno = 0;
return socketfd;
} while (0);
{
const int saved_errno = errno;
close(socketfd);
errno = saved_errno;
return -1;
}
}
static unsigned int tci(const unsigned int priority,
const unsigned int drop,
const unsigned int vlan)
{
return (vlan & 0xFFFU)
| ((!!drop) << 12U)
| ((priority & 7U) << 13U);
}
static size_t rawpacket_qinq(unsigned char *const buffer, size_t const length,
const unsigned char *const srcaddr,
const unsigned char *const dstaddr,
const unsigned int service_tci,
const unsigned int customer_tci,
const unsigned int ethertype)
{
unsigned char *ptr = buffer;
uint32_t tag;
uint16_t type;
if (length < 2 * ETH_ALEN + 4 + 4 + 2) {
errno = ENOSPC;
return (size_t)0;
}
memcpy(ptr, dstaddr, ETH_ALEN);
ptr += ETH_ALEN;
memcpy(ptr, srcaddr, ETH_ALEN);
ptr += ETH_ALEN;
/* Service 802.1AD tag. */
tag = htonl( ((uint32_t)(ETH_P_8021AD) << 16U)
| ((uint32_t)service_tci & 0xFFFFU) );
memcpy(ptr, &tag, 4);
ptr += 4;
/* Client 802.1Q tag. */
tag = htonl( ((uint32_t)(ETH_P_8021Q) << 16U)
| ((uint32_t)customer_tci & 0xFFFFU) );
memcpy(ptr, &tag, 4);
ptr += 4;
/* Ethertype tag. */
type = htons((uint16_t)ethertype);
memcpy(ptr, &type, 2);
ptr += 2;
return (size_t)(ptr - buffer);
}
#endif /* RAWPACKET_H */
sender.c:
#include <string.h>
#include <errno.h>
#include <stdio.h>
#include "rawpacket.h"
static size_t parse_data(unsigned char *const data, const size_t size,
const char *const string)
{
char *ends = strncpy((char *)data, string, size);
return (size_t)(ends - (char *)data);
}
static int parse_hwaddr(const char *const string,
void *const hwaddr)
{
unsigned int addr[6];
char dummy;
if (sscanf(string, " %02x:%02x:%02x:%02x:%02x:%02x %c",
&addr[0], &addr[1], &addr[2],
&addr[3], &addr[4], &addr[5],
&dummy) == 6 ||
sscanf(string, " %02x%02x%02x%02x%02x%02x %c",
&addr[0], &addr[1], &addr[2],
&addr[3], &addr[4], &addr[5],
&dummy) == 6) {
if (hwaddr) {
((unsigned char *)hwaddr)[0] = addr[0];
((unsigned char *)hwaddr)[1] = addr[1];
((unsigned char *)hwaddr)[2] = addr[2];
((unsigned char *)hwaddr)[3] = addr[3];
((unsigned char *)hwaddr)[4] = addr[4];
((unsigned char *)hwaddr)[5] = addr[5];
}
return 0;
}
errno = EINVAL;
return -1;
}
int main(int argc, char *argv[])
{
unsigned char packet[ETH_FRAME_LEN + ETH_FCS_LEN];
unsigned char srcaddr[6], dstaddr[6];
int socketfd;
size_t size, i;
ssize_t n;
if (argc < 3 || argc > 4 || !strcmp(argv[1], "-h") || !strcmp(argv[1], "--help")) {
fprintf(stderr, "\n");
fprintf(stderr, "Usage: %s [ -h | --help ]\n", argv[0]);
fprintf(stderr, " %s interface hwaddr [message]\n", argv[0]);
fprintf(stderr, "\n");
return 1;
}
if (parse_hwaddr(argv[2], &dstaddr)) {
fprintf(stderr, "%s: Invalid destination hardware address.\n", argv[2]);
return 1;
}
socketfd = rawpacket_socket(ETH_P_ALL, argv[1], &srcaddr);
if (socketfd == -1) {
fprintf(stderr, "%s: %s.\n", argv[1], strerror(errno));
return 1;
}
memset(packet, 0, sizeof packet);
/* Construct a QinQ header for a fake Ethernet packet type. */
size = rawpacket_qinq(packet, sizeof packet, srcaddr, dstaddr,
tci(7, 0, 1U), tci(7, 0, 2U),
ETH_P_IP);
if (!size) {
fprintf(stderr, "Failed to construct QinQ headers: %s.\n", strerror(errno));
close(socketfd);
return 1;
}
/* Add packet payload. */
if (argc > 3)
size += parse_data(packet + size, sizeof packet - size, argv[3]);
else
size += parse_data(packet + size, sizeof packet - size, "Hello!");
/* Pad with zeroes to minimum 64 octet length. */
if (size < 64)
size = 64;
/* Send it. */
n = send(socketfd, packet, size, 0);
if (n == -1) {
fprintf(stderr, "Failed to send packet: %s.\n", strerror(errno));
shutdown(socketfd, SHUT_RDWR);
close(socketfd);
return 1;
}
fprintf(stderr, "Sent %ld bytes:", (long)n);
for (i = 0; i < size; i++)
fprintf(stderr, " %02x", packet[i]);
fprintf(stderr, "\n");
fflush(stderr);
shutdown(socketfd, SHUT_RDWR);
if (close(socketfd)) {
fprintf(stderr, "Error closing socket: %s.\n", strerror(errno));
return 1;
}
return 0;
}
receiver.c:
#include <sys/types.h>
#include <sys/socket.h>
#include <string.h>
#include <signal.h>
#include <errno.h>
#include <stdio.h>
#include "rawpacket.h"
static volatile sig_atomic_t done = 0;
static void handle_done(int signum)
{
done = signum;
}
static int install_done(const int signum)
{
struct sigaction act;
sigemptyset(&act.sa_mask);
act.sa_handler = handle_done;
act.sa_flags = 0;
if (sigaction(signum, &act, NULL))
return errno;
return 0;
}
static const char *protocol_name(const unsigned int protocol)
{
static char buffer[16];
switch (protocol & 0xFFFFU) {
case 0x0001: return "ETH_P_802_3";
case 0x0002: return "ETH_P_AX25";
case 0x0003: return "ETH_P_ALL";
case 0x0060: return "ETH_P_LOOP";
case 0x0800: return "ETH_P_IP";
case 0x0806: return "ETH_P_ARP";
case 0x8100: return "ETH_P_8021Q (802.1Q VLAN)";
case 0x88A8: return "ETH_P_8021AD (802.1AD VLAN)";
default:
snprintf(buffer, sizeof buffer, "0x%04x", protocol & 0xFFFFU);
return (const char *)buffer;
}
}
static const char *header_type(const unsigned int hatype)
{
static char buffer[16];
switch (hatype) {
case 1: return "ARPHRD_ETHER: Ethernet 10Mbps";
case 2: return "ARPHRD_EETHER: Experimental Ethernet";
case 768: return "ARPHRD_TUNNEL: IP Tunnel";
case 772: return "ARPHRD_LOOP: Loopback";
default:
snprintf(buffer, sizeof buffer, "0x%04x", hatype);
return buffer;
}
}
static const char *packet_type(const unsigned int pkttype)
{
static char buffer[16];
switch (pkttype) {
case PACKET_HOST: return "PACKET_HOST";
case PACKET_BROADCAST: return "PACKET_BROADCAST";
case PACKET_MULTICAST: return "PACKET_MULTICAST";
case PACKET_OTHERHOST: return "PACKET_OTHERHOST";
case PACKET_OUTGOING: return "PACKET_OUTGOING";
default:
snprintf(buffer, sizeof buffer, "0x%02x", pkttype);
return (const char *)buffer;
}
}
static void fhex(FILE *const out,
const char *const before,
const char *const after,
const void *const src, const size_t len)
{
const unsigned char *const data = src;
size_t i;
if (len < 1)
return;
if (before)
fputs(before, out);
for (i = 0; i < len; i++)
fprintf(out, " %02x", data[i]);
if (after)
fputs(after, out);
}
int main(int argc, char *argv[])
{
struct sockaddr_ll addr;
socklen_t addrlen;
unsigned char data[2048];
ssize_t n;
int socketfd, flag;
if (argc != 2 || !strcmp(argv[1], "-h") || !strcmp(argv[1], "--help")) {
fprintf(stderr, "\n");
fprintf(stderr, "Usage: %s [ -h | --help ]\n", argv[0]);
fprintf(stderr, " %s interface\n", argv[0]);
fprintf(stderr, "\n");
return 1;
}
if (install_done(SIGINT) ||
install_done(SIGHUP) ||
install_done(SIGTERM)) {
fprintf(stderr, "Cannot install signal handlers: %s.\n", strerror(errno));
return 1;
}
socketfd = rawpacket_socket(ETH_P_ALL, argv[1], NULL);
if (socketfd == -1) {
fprintf(stderr, "%s: %s.\n", argv[1], strerror(errno));
return 1;
}
flag = 1;
if (setsockopt(socketfd, SOL_SOCKET, SO_REUSEADDR, &flag, sizeof flag)) {
fprintf(stderr, "Cannot set REUSEADDR socket option: %s.\n", strerror(errno));
close(socketfd);
return 1;
}
if (setsockopt(socketfd, SOL_SOCKET, SO_BINDTODEVICE, argv[1], strlen(argv[1]) + 1)) {
fprintf(stderr, "Cannot bind to device %s: %s.\n", argv[1], strerror(errno));
close(socketfd);
return 1;
}
while (!done) {
memset(data, 0, sizeof data);
memset(&addr, 0, sizeof addr);
addrlen = sizeof addr;
n = recvfrom(socketfd, &data, sizeof data, 0,
(struct sockaddr *)&addr, &addrlen);
if (n == -1) {
if (errno == EINTR)
continue;
fprintf(stderr, "Receive error: %s.\n", strerror(errno));
break;
}
printf("Received %d bytes:\n", (int)n);
printf("\t Protocol: %s\n", protocol_name(htons(addr.sll_protocol)));
printf("\t Interface: %d\n", (int)addr.sll_ifindex);
printf("\t Header type: %s\n", header_type(addr.sll_hatype));
printf("\t Packet type: %s\n", packet_type(addr.sll_pkttype));
fhex(stdout, "\t Address:", "\n", addr.sll_addr, addr.sll_halen);
fhex(stdout, "\t Data:", "\n", data, n);
printf("\n");
fflush(stdout);
}
shutdown(socketfd, SHUT_RDWR);
close(socketfd);
return 0;
}
To compile, you can use
gcc -O2 receiver.c -o receiver
gcc -O2 sender.c -o sender
Run without parameters, or with -h, to see the usage for either one. sender sends only one packet. receiver listens on the specified interface (in promiscuous mode), until you interrupt it (Ctrl+C) or send it a TERM signal.
Start receiver in one virtual terminal on the loopback interface:
sudo ./receiver lo
In another virtual terminal on the same machine, running
sudo ./sender lo FF:FF:FF:FF:FF:FF '_The contents of a 64-byte Ethernet frame_'
will output (newlines and indentation added for ease of understanding)
Sent 64 bytes: ff ff ff ff ff ff
00 00 00 00 00 00
88 a8 e0 01
81 00 e0 02
08 00
5f 54 68 65 20 63 6f 6e
74 65 6e 74 73 20 6f 66
20 61 20 36 34 2d 62 79
74 65 20 45 74 68 65 72
6e 65 74 20 66 72 61 6d
65 5f
In the receiver terminal, however, we see (newlines and indentation added):
Received 64 bytes:
Protocol: ETH_P_ALL
Interface: 1
Header type: ATPHRD_LOOP: Loopback
Packet type: PACKET_OUTGOING
Address: 00 00 00 00 00 00
Data: ff ff ff ff ff ff
00 00 00 00 00 00
88 a8 e0 01
81 00 e0 02
08 00
5f 54 68 65 20 63 6f 6e
74 65 6e 74 73 20 6f 66
20 61 20 36 34 2d 62 79
74 65 20 45 74 68 65 72
6e 65 74 20 66 72 61 6d
65 5f
Received 60 bytes:
Protocol: ETH_P_8021Q (802.1Q VLAN)
Interface: 1
Header type: ATPHRD_LOOP: Loopback
Packet type: PACKET_MULTICAST
Address: 00 00 00 00 00 00
Data: ff ff ff ff ff ff
00 00 00 00 00 00
81 00 e0 02
08 00
5f 54 68 65 20 63 6f 6e
74 65 6e 74 73 20 6f 66
20 61 20 36 34 2d 62 79
74 65 20 45 74 68 65 72
6e 65 74 20 66 72 61 6d
65 5f
The first one, PACKET_OUTGOING, was captured as outgoing; it shows that the kernel did not consume any headers when the packet was sent.
The second one, PACKET_MULTICAST, was captured as it arrived. (Since the Ethernet address was FF:FF:FF:FF:FF:FF, it is a multicast packet.)
As you can see, the latter packet has only the 802.1Q VLAN header -- the client VLAN --, the kernel having consumed the 802.1AD service VLAN tag.
The above confirms the scenario for the loopback interface, at least. Using the raw packet interface, the kernel consumes the 802.1AD VLAN header (the one immediately following the recipient address). If you use tcpdump -i eth0 alongside the receiver, you can see that libpcap is reinserting the consumed header back to the packet!
Loopback interface is a bit special, so let's redo the test using virtual machines. I happen to be running up-to-date Xubuntu 14.04 (all updates installed as of 2014-06-28, Ubuntu 3.13.0-29-generic #53 x86_64 kernel). Sender HW address is 08 00 00 00 00 02, receivers is 08 00 00 00 00 01, and the two are connected to an internal network without anybody else present.
(Again, I add newlines and indentation to the output to make it easier to read.)
Sender, on virtual machine 2:
sudo ./sender eth0 08:00:00:00:00:01 '_The contents of a 64-byte Ethernet frame_'
Sent 64 bytes: 08 00 00 00 00 01
08 00 00 00 00 02
88 a8 e0 01
81 00 e0 02
08 00
5f 54 68 65 20 63 6f 6e
74 65 6e 74 73 20 6f 66
20 61 20 36 34 2d 62 79
74 65 20 45 74 68 65 72
6e 65 74 20 66 72 61 6d
65 5f
Receiver, on virtual machine 1:
sudo ./receiver eth0
Received 60 bytes:
Protocol: ETH_P_8021Q (802.1Q VLAN)
Interface: 2
Header type: ARPHRD_ETHER: Ethernet 10Mbps
Packet type: PACKET_HOST
Address: 08 00 00 00 00 02
Data: 08 00 00 00 00 01
08 00 00 00 00 02
81 00 e0 02
08 00
5f 54 68 65 20 63 6f 6e
74 65 6e 74 73 20 6f 66
20 61 20 36 34 2d 62 79
74 65 20 45 74 68 65 72
6e 65 74 20 66 72 61 6d
65 5f
As you can see, the results are basically the same as for the loopback case. In particular, the 802.1AD service VLAN tag was consumed on receive. (You can use tcpdump or wireshark, to compare the received packets: libpcap is obviously reinserting the consumed VLAN tag pack into the packet.)
If you have recent enough kernel (support was added in April 2013), then you can configure a 802.1AD VLAN(s) on the recipient using:
sudo modprobe 8021q
sudo ip link add link eth0 eth0.service1 type vlan proto 802.1ad id 1
Receiving on eth0 will receive all packets, but on eth0.service1 only those packets with an 802.1AD VLAN tag, with VLAN ID 1. It does not capture frames with 802.1Q VLAN tags with the same VLAN ID, meaning that you can do full routing on receive for both 802.1AD and 802.1Q VLANs.
I didn't trust just the above test, myself; I created a number of 802.1AD and 802.1Q VLANs, with separate receive instances on each one, and changed the packet headers (not only service (first) tci() and client (second) tci() in the rawpacket_qinq() call in sender.c to change the service and client VLAN IDs, but also changing rawpacket.h, to verify that 802.1AD (88a8) and 802.1Q (8100) VLAN headers are routed correctly on receive). It all worked beautifully, without any hiccups.
In summary:
Given a recent enough Linux kernel version, Ethernet frames are correctly routed by the Linux kernel (by the 8021q module) on receive, including separate VLAN interfaces for 802.1AD and 802.1Q with the same VLAN IDs. The kernel does consume the VLAN header used for routing, even if no VLANs are configured.
Questions?
- How portable is code with #pragma optimize?
- Make warn_unused_result applied to all function with GCC
- Multiple threads and CPU cache
- Linking error vs. compilation error
- how to compile multiple instances of a source file linking to it different files.o
- Background compile in VS2010 and C/C++
- What Turing complete macro languages are available for C/C++
- How to use Cython to compile Python 3 into C
- Are forward declarations needed when the typedef declaration is done?
- How to assign real and imaginary parts of complex variables in C
- Strict aliasing between Complex numbers and real numbers in C
- How to CORRECTLY install gsl library in Linux?
- Using = operator on atomic variable?
- Unknown conversion types in this program I'm meant to compile
- Iterator in C language
- Treating __func__ as a string literal instead of a predefined identifier
- c - strcpy with struct string returning error
- Segmentation fault when executing a Python script in a C program?
- Questions about C Function Prototypes and Compilation
- What is the purpose of "-Wa,option" in GCC?
- Do we have c99 subflags
- netcat gcc compile option so that IDA pro can show function name
- What is the correct way to free output buffer after call to openssl::i2d_X509?
- How can I link C++ files to a C program?
- Why doesn't this program in C compile? Error: undefined reference to `i2c_smbus_read_byte_data'
- integer promotion isn't taken into account
- Why do I get a SEGFAULT when I try to sort an array of integers using K & R code examples?
- Are C preprocessor statements a part of the C language?
- FFMPEG Api conversion from YUV420P to RGB produces strange output
- Why is the tree constructed this way incorrect?