emo/ndhc
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
6804be2277
ndhc/src/sockd.c

634 lines
21 KiB
C
Raw Normal View History

Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
/* sockd.c - privsep socket creation daemon
*
* Copyright (c) 2014 Nicholas J. Kain <njkain at gmail dot com>
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* - Redistributions of source code must retain the above copyright notice,
* this list of conditions and the following disclaimer.
*
* - Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*/
#include <stdbool.h>
#include <stdint.h>
#include <unistd.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <errno.h>
#include <signal.h>
#include <fcntl.h>
Have sockd apply BPF filters to ARP sockets.
2014-04-06 07:09:27 +05:30
#include <assert.h>
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
#include <sys/socket.h>
#include <sys/signalfd.h>
#include <sys/epoll.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/prctl.h>
#include <arpa/inet.h>
#include <netinet/ip.h>
#include <netpacket/packet.h>
#include <net/ethernet.h>
Have sockd apply BPF filters to ARP sockets.
2014-04-06 07:09:27 +05:30
#include <netinet/if_ether.h>
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
#include <linux/filter.h>
#include <pwd.h>
#include <grp.h>
#include "nk/log.h"
#include "nk/io.h"
#include "nk/privilege.h"
#include "sockd.h"
#include "ndhc-defines.h"
#include "ndhc.h"
#include "dhcp.h"
#include "sys.h"
#include "seccomp.h"
static int epollfd, signalFd;
/* Slots are for signalFd and the ndhc -> ifchd pipe. */
static struct epoll_event events[2];
uid_t sockd_uid = 0;
gid_t sockd_gid = 0;
// Interface to make requests of sockd. Called from ndhc process.
int request_sockd_fd(char *buf, size_t buflen, char *response)
{
if (!buflen)
return -1;
Switch to using a socket for ndhc/sockd IPC so that fd passing works.
2014-04-05 14:55:56 +05:30
ssize_t r = safe_write(sockdSock[0], buf, buflen);
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
if (r < 0 || (size_t)r != buflen)
suicide("%s: (%s) write failed: %d", client_config.interface,
__func__, r);
char data[MAX_BUF], control[MAX_BUF];
struct iovec iov = {
.iov_base = data,
.iov_len = sizeof data - 1,
};
struct msghdr msg = {
.msg_iov = &iov,
.msg_iovlen = 1,
.msg_control = control,
.msg_controllen = sizeof control
};
retry:
Switch to using a socket for ndhc/sockd IPC so that fd passing works.
2014-04-05 14:55:56 +05:30
r = recvmsg(sockdSock[0], &msg, 0);
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
if (r == 0) {
suicide("%s: (%s) recvmsg received EOF", client_config.interface,
__func__);
} else if (r < 0) {
if (errno == EINTR)
goto retry;
suicide("%s: (%s) recvmsg failed: %s", client_config.interface,
__func__, strerror(errno));
}
data[iov.iov_len] = '\0';
char repc = data[0];
for (struct cmsghdr *cmsg = CMSG_FIRSTHDR(&msg); cmsg;
cmsg = CMSG_NXTHDR(&msg, cmsg)) {
if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
if (response)
*response = repc;
else if (repc != buf[0])
suicide("%s: (%s) expected %c sockd reply but got %c",
client_config.interface, __func__, buf[0], repc);
int *fd = (int *)CMSG_DATA(cmsg);
return *fd;
}
}
suicide("%s: (%s) sockd reply did not include a fd",
client_config.interface, __func__);
}
static int create_arp_socket(void)
{
int fd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ARP));
Change most error comparisons from == -1 to < 0. Some were not changed, as the different negative values equate to different errors. Tests against syscall returns and fds are very common and mostly fit the pattern of this change. The gain is increased range-exclusion.
2014-04-06 16:03:14 +05:30
if (fd < 0) {
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
log_error("%s: (%s) socket failed: %s", client_config.interface,
__func__, strerror(errno));
goto out;
}
int opt = 1;
Change most error comparisons from == -1 to < 0. Some were not changed, as the different negative values equate to different errors. Tests against syscall returns and fds are very common and mostly fit the pattern of this change. The gain is increased range-exclusion.
2014-04-06 16:03:14 +05:30
if (setsockopt(fd, SOL_SOCKET, SO_BROADCAST, &opt, sizeof opt) < 0) {
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
log_error("%s: (%s) setsockopt failed: %s", client_config.interface,
__func__, strerror(errno));
goto out_fd;
}
Change most error comparisons from == -1 to < 0. Some were not changed, as the different negative values equate to different errors. Tests against syscall returns and fds are very common and mostly fit the pattern of this change. The gain is increased range-exclusion.
2014-04-06 16:03:14 +05:30
if (fcntl(fd, F_SETFL, fcntl(fd, F_GETFL) | O_NONBLOCK) < 0) {
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
log_error("%s: (%s) fcntl failed: %s", client_config.interface,
__func__, strerror(errno));
goto out_fd;
}
struct sockaddr_ll saddr = {
.sll_family = AF_PACKET,
.sll_protocol = htons(ETH_P_ARP),
.sll_ifindex = client_config.ifindex,
};
if (bind(fd, (struct sockaddr *)&saddr, sizeof(struct sockaddr_ll)) < 0) {
log_error("%s: (%s) bind failed: %s", client_config.interface,
__func__, strerror(errno));
goto out_fd;
}
return fd;
out_fd:
close(fd);
out:
return -1;
}
// Returns fd of new udp socket bound on success, or -1 on failure.
static int create_udp_socket(uint32_t ip, uint16_t port, char *iface)
{
int fd;
if ((fd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP)) < 0) {
log_error("%s: (%s) socket failed: %s",
client_config.interface, __func__, strerror(errno));
goto out;
}
int opt = 1;
Change most error comparisons from == -1 to < 0. Some were not changed, as the different negative values equate to different errors. Tests against syscall returns and fds are very common and mostly fit the pattern of this change. The gain is increased range-exclusion.
2014-04-06 16:03:14 +05:30
if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof opt) < 0) {
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
log_error("%s: (%s) Set reuse addr failed: %s",
client_config.interface, __func__, strerror(errno));
goto out_fd;
}
Change most error comparisons from == -1 to < 0. Some were not changed, as the different negative values equate to different errors. Tests against syscall returns and fds are very common and mostly fit the pattern of this change. The gain is increased range-exclusion.
2014-04-06 16:03:14 +05:30
if (setsockopt(fd, SOL_SOCKET, SO_DONTROUTE, &opt, sizeof opt) < 0) {
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
log_error("%s: (%s) Set don't route failed: %s",
client_config.interface, __func__, strerror(errno));
goto out_fd;
}
struct ifreq ifr;
memset(&ifr, 0, sizeof ifr);
ssize_t sl = snprintf(ifr.ifr_name, sizeof ifr.ifr_name, "%s", iface);
if (sl < 0 || (size_t)sl >= sizeof ifr.ifr_name) {
log_error("%s: (%s) Set interface name failed.",
client_config.interface, __func__);
goto out_fd;
}
if (setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, &ifr, sizeof ifr) < 0) {
log_error("%s: (%s) Set bind to device failed: %s",
client_config.interface, __func__, strerror(errno));
goto out_fd;
}
Change most error comparisons from == -1 to < 0. Some were not changed, as the different negative values equate to different errors. Tests against syscall returns and fds are very common and mostly fit the pattern of this change. The gain is increased range-exclusion.
2014-04-06 16:03:14 +05:30
if (fcntl(fd, F_SETFL, fcntl(fd, F_GETFL) | O_NONBLOCK) < 0) {
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
log_error("%s: (%s) Set non-blocking failed: %s",
client_config.interface, __func__, strerror(errno));
goto out_fd;
}
struct sockaddr_in sa = {
.sin_family = AF_INET,
.sin_port = htons(port),
.sin_addr.s_addr = ip,
};
Change most error comparisons from == -1 to < 0. Some were not changed, as the different negative values equate to different errors. Tests against syscall returns and fds are very common and mostly fit the pattern of this change. The gain is increased range-exclusion.
2014-04-06 16:03:14 +05:30
if (bind(fd, (struct sockaddr *)&sa, sizeof sa) < 0)
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
goto out_fd;
return fd;
out_fd:
close(fd);
out:
return -1;
}
static int create_raw_socket(struct sockaddr_ll *sa, bool *using_bpf,
const struct sock_fprog *filter_prog)
{
int fd;
if ((fd = socket(AF_PACKET, SOCK_DGRAM, htons(ETH_P_IP))) < 0) {
log_error("create_raw_socket: socket failed: %s", strerror(errno));
goto out;
}
if (filter_prog) {
int r = setsockopt(fd, SOL_SOCKET, SO_ATTACH_FILTER, filter_prog,
sizeof *filter_prog);
if (using_bpf)
*using_bpf = !r;
}
int opt = 1;
Change most error comparisons from == -1 to < 0. Some were not changed, as the different negative values equate to different errors. Tests against syscall returns and fds are very common and mostly fit the pattern of this change. The gain is increased range-exclusion.
2014-04-06 16:03:14 +05:30
if (setsockopt(fd, SOL_SOCKET, SO_DONTROUTE, &opt, sizeof opt) < 0) {
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
log_error("create_raw_socket: Failed to set don't route: %s",
strerror(errno));
goto out_fd;
}
Change most error comparisons from == -1 to < 0. Some were not changed, as the different negative values equate to different errors. Tests against syscall returns and fds are very common and mostly fit the pattern of this change. The gain is increased range-exclusion.
2014-04-06 16:03:14 +05:30
if (fcntl(fd, F_SETFL, fcntl(fd, F_GETFL) | O_NONBLOCK) < 0) {
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
log_error("create_raw_socket: Set non-blocking failed: %s",
strerror(errno));
goto out_fd;
}
if (bind(fd, (struct sockaddr *)sa, sizeof *sa) < 0) {
log_error("create_raw_socket: bind failed: %s", strerror(errno));
goto out_fd;
}
return fd;
out_fd:
close(fd);
out:
return -1;
}
// Returns fd of new listen socket bound to 0.0.0.0:@68 on interface @inf
// on success, or -1 on failure.
static int create_udp_listen_socket(void)
{
int fd = create_udp_socket(INADDR_ANY, DHCP_CLIENT_PORT,
client_config.interface);
Change most error comparisons from == -1 to < 0. Some were not changed, as the different negative values equate to different errors. Tests against syscall returns and fds are very common and mostly fit the pattern of this change. The gain is increased range-exclusion.
2014-04-06 16:03:14 +05:30
if (fd < 0)
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
return -1;
int opt = 1;
Change most error comparisons from == -1 to < 0. Some were not changed, as the different negative values equate to different errors. Tests against syscall returns and fds are very common and mostly fit the pattern of this change. The gain is increased range-exclusion.
2014-04-06 16:03:14 +05:30
if (setsockopt(fd, SOL_SOCKET, SO_BROADCAST, &opt, sizeof opt) < 0) {
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
log_error("%s: (%s) Set broadcast failed: %s",
client_config.interface, __func__, strerror(errno));
close(fd);
return -1;
}
return fd;
}
static int create_udp_send_socket(uint32_t client_addr)
{
return create_udp_socket(client_addr, DHCP_CLIENT_PORT,
client_config.interface);
}
static int create_raw_listen_socket(bool *using_bpf)
{
static const struct sock_filter sf_dhcp[] = {
// Verify that the packet has a valid IPv4 version nibble and
// that no IP options are defined.
BPF_STMT(BPF_LD + BPF_B + BPF_ABS, 0),
BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, 0x45, 1, 0),
BPF_STMT(BPF_RET + BPF_K, 0),
// Verify that the IP header has a protocol number indicating UDP.
BPF_STMT(BPF_LD + BPF_B + BPF_ABS, 9),
BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, IPPROTO_UDP, 1, 0),
BPF_STMT(BPF_RET + BPF_K, 0),
// Make certain that the packet is not a fragment. All bits in
// the flag and fragment offset field must be set to zero except
// for the Evil and DF bits (0,1).
BPF_STMT(BPF_LD + BPF_H + BPF_ABS, 6),
BPF_JUMP(BPF_JMP + BPF_JSET + BPF_K, 0x3fff, 0, 1),
BPF_STMT(BPF_RET + BPF_K, 0),
// Packet is UDP. Advance X past the IP header.
BPF_STMT(BPF_LDX + BPF_B + BPF_MSH, 0),
// Verify that the UDP client and server ports match that of the
// IANA-assigned DHCP ports.
BPF_STMT(BPF_LD + BPF_W + BPF_IND, 0),
BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K,
(DHCP_SERVER_PORT << 16) + DHCP_CLIENT_PORT, 1, 0),
BPF_STMT(BPF_RET + BPF_K, 0),
// Get the UDP length field and store it in X.
BPF_STMT(BPF_LD + BPF_H + BPF_IND, 4),
BPF_STMT(BPF_MISC + BPF_TAX, 0),
// Get the IPv4 length field and store it in A and M[0].
BPF_STMT(BPF_LD + BPF_H + BPF_ABS, 2),
BPF_STMT(BPF_ST, 0),
// Verify that UDP length = IP length - IP header size
BPF_STMT(BPF_ALU + BPF_SUB + BPF_K, 20),
BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_X, 0, 1, 0),
BPF_STMT(BPF_RET + BPF_K, 0),
// Pass the number of octets that are specified in the IPv4 header.
BPF_STMT(BPF_LD + BPF_MEM, 0),
BPF_STMT(BPF_RET + BPF_A, 0),
};
static const struct sock_fprog sfp_dhcp = {
.len = sizeof sf_dhcp / sizeof sf_dhcp[0],
.filter = (struct sock_filter *)sf_dhcp,
};
struct sockaddr_ll sa = {
.sll_family = AF_PACKET,
.sll_protocol = htons(ETH_P_IP),
.sll_ifindex = client_config.ifindex,
};
return create_raw_socket(&sa, using_bpf, &sfp_dhcp);
}
static int create_raw_broadcast_socket(void)
{
struct sockaddr_ll da = {
.sll_family = AF_PACKET,
.sll_protocol = htons(ETH_P_IP),
.sll_pkttype = PACKET_BROADCAST,
.sll_ifindex = client_config.ifindex,
.sll_halen = 6,
};
memcpy(da.sll_addr, "\xff\xff\xff\xff\xff\xff", 6);
return create_raw_socket(&da, NULL, NULL);
}
Have sockd apply BPF filters to ARP sockets.
2014-04-06 07:09:27 +05:30
static bool arp_set_bpf_basic(int fd)
{
static struct sock_filter sf_arp[] = {
// Verify that the frame has ethernet protocol type of ARP
// and that the ARP hardware type field indicates Ethernet.
BPF_STMT(BPF_LD + BPF_W + BPF_ABS, 12),
BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, (ETH_P_ARP << 16) | ARPHRD_ETHER,
1, 0),
BPF_STMT(BPF_RET + BPF_K, 0),
// Verify that the ARP protocol type field indicates IP, the ARP
// hardware address length field is 6, and the ARP protocol address
// length field is 4.
BPF_STMT(BPF_LD + BPF_W + BPF_ABS, 16),
BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, (ETH_P_IP << 16) | 0x0604, 1, 0),
BPF_STMT(BPF_RET + BPF_K, 0),
// Sanity tests passed, so send all possible data.
BPF_STMT(BPF_RET + BPF_K, 0x7fffffff),
};
static const struct sock_fprog sfp_arp = {
.len = sizeof sf_arp / sizeof sf_arp[0],
.filter = sf_arp,
};
int ret = setsockopt(fd, SOL_SOCKET, SO_ATTACH_FILTER, &sfp_arp,
sizeof sfp_arp) != -1;
if (ret < 0)
log_warning("%s: Failed to set BPF for basic ARP socket: %s",
client_config.interface, strerror(errno));
return ret == 0;
}
static bool arp_set_bpf_defense(int fd, uint32_t client_addr,
uint8_t client_mac[6])
{
uint32_t mac4b;
uint16_t mac2b;
memcpy(&mac4b, client_mac, 4);
memcpy(&mac2b, client_mac + 4, 2);
struct sock_filter sf_arp[] = {
// Verify that the frame has ethernet protocol type of ARP
// and that the ARP hardware type field indicates Ethernet.
BPF_STMT(BPF_LD + BPF_W + BPF_ABS, 12),
BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, (ETH_P_ARP << 16) | ARPHRD_ETHER,
1, 0),
BPF_STMT(BPF_RET + BPF_K, 0),
// Verify that the ARP protocol type field indicates IP, the ARP
// hardware address length field is 6, and the ARP protocol address
// length field is 4.
BPF_STMT(BPF_LD + BPF_W + BPF_ABS, 16),
BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, (ETH_P_IP << 16) | 0x0604, 1, 0),
BPF_STMT(BPF_RET + BPF_K, 0),
// If the ARP packet source IP does not match our IP address, then
// it can be ignored.
BPF_STMT(BPF_LD + BPF_W + BPF_ABS, 28),
BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, client_addr, 1, 0),
BPF_STMT(BPF_RET + BPF_K, 0),
// If the first four bytes of the ARP packet source hardware address
// does not equal our hardware address, then it's a conflict and should
// be passed along.
BPF_STMT(BPF_LD + BPF_W + BPF_ABS, 22),
BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, mac4b, 1, 0),
BPF_STMT(BPF_RET + BPF_K, 0x7fffffff),
// If the last two bytes of the ARP packet source hardware address
// do not equal our hardware address, then it's a conflict and should
// be passed along.
BPF_STMT(BPF_LD + BPF_H + BPF_ABS, 26),
BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, mac2b, 1, 0),
BPF_STMT(BPF_RET + BPF_K, 0x7fffffff),
// Packet announces our IP address and hardware address, so it requires
// no action.
BPF_STMT(BPF_RET + BPF_K, 0),
};
struct sock_fprog sfp_arp = {
.len = sizeof sf_arp / sizeof sf_arp[0],
.filter = (struct sock_filter *)sf_arp,
};
int ret = setsockopt(fd, SOL_SOCKET, SO_ATTACH_FILTER, &sfp_arp,
sizeof sfp_arp) != -1;
if (ret < 0)
log_warning("%s: Failed to set BPF for defense ARP socket: %s",
client_config.interface, strerror(errno));
return ret == 0;
}
static int create_arp_defense_socket(uint32_t client_addr,
uint8_t client_mac[6], bool *using_bpf)
{
assert(using_bpf);
int fd = create_arp_socket();
*using_bpf = arp_set_bpf_defense(fd, client_addr, client_mac);
return fd;
}
static int create_arp_basic_socket(bool *using_bpf)
{
assert(using_bpf);
int fd = create_arp_socket();
*using_bpf = arp_set_bpf_basic(fd);
return fd;
}
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
// XXX: Can share with ifch
static void setup_signals_sockd(void)
{
sigset_t mask;
sigemptyset(&mask);
sigaddset(&mask, SIGPIPE);
sigaddset(&mask, SIGUSR1);
sigaddset(&mask, SIGUSR2);
sigaddset(&mask, SIGTSTP);
sigaddset(&mask, SIGTTIN);
sigaddset(&mask, SIGCHLD);
sigaddset(&mask, SIGHUP);
sigaddset(&mask, SIGINT);
sigaddset(&mask, SIGTERM);
if (sigprocmask(SIG_BLOCK, &mask, NULL) < 0)
suicide("sigprocmask failed");
signalFd = signalfd(-1, &mask, SFD_NONBLOCK);
if (signalFd < 0)
suicide("signalfd failed");
}
// XXX: Can share with ifch
static void signal_dispatch(void)
{
int t;
size_t off = 0;
struct signalfd_siginfo si = {0};
again:
t = read(signalFd, (char *)&si + off, sizeof si - off);
if (t < 0) {
if (t == EAGAIN || t == EWOULDBLOCK || t == EINTR)
goto again;
else
suicide("signalfd read error");
}
if (off + (unsigned)t < sizeof si)
off += t;
switch (si.ssi_signo) {
case SIGINT:
case SIGTERM:
exit(EXIT_SUCCESS);
break;
case SIGPIPE:
log_line("ndhc-sockd: IPC pipe closed. Exiting.");
exit(EXIT_SUCCESS);
break;
default:
break;
}
}
static void xfer_fd(int fd, char cmd)
{
char control[sizeof(struct cmsghdr) + 10];
struct iovec iov = {
.iov_base = &cmd,
.iov_len = 1,
};
struct msghdr msg = {
.msg_iov = &iov,
.msg_iovlen = 1,
.msg_control = control,
.msg_controllen = sizeof control,
};
struct cmsghdr *cmsg = CMSG_FIRSTHDR(&msg);
cmsg->cmsg_level = SOL_SOCKET;
cmsg->cmsg_type = SCM_RIGHTS;
cmsg->cmsg_len = CMSG_LEN(sizeof fd);
int *cmsg_fd = (int *)CMSG_DATA(cmsg);
*cmsg_fd = fd;
msg.msg_controllen = cmsg->cmsg_len;
retry:
Switch to using a socket for ndhc/sockd IPC so that fd passing works.
2014-04-05 14:55:56 +05:30
if (sendmsg(sockdSock[1], &msg, 0) < 0) {
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
if (errno == EINTR)
goto retry;
suicide("%s: (%s) sendmsg failed: %s", client_config.interface,
__func__, strerror(errno));
}
}
static size_t execute_sockd(char *buf, size_t buflen)
{
if (!buflen)
return 0;
char c = buf[0];
switch (c) {
case 'L': {
bool using_bpf;
int fd = create_raw_listen_socket(&using_bpf);
xfer_fd(fd, using_bpf ? 'L' : 'l');
return 1;
}
case 'U': xfer_fd(create_udp_listen_socket(), 'U'); return 1;
Have sockd apply BPF filters to ARP sockets.
2014-04-06 07:09:27 +05:30
case 'a': {
bool using_bpf;
int fd = create_arp_basic_socket(&using_bpf);
xfer_fd(fd, using_bpf ? 'A' : 'a'); return 1;
}
case 'd': {
uint32_t client_addr;
uint8_t client_mac[6];
bool using_bpf;
if (buflen < 1 + sizeof client_addr + 6)
return 0;
memcpy(&client_addr, buf + 1, sizeof client_addr);
memcpy(client_mac, buf + 1 + sizeof client_addr, 6);
int fd = create_arp_defense_socket(client_addr, client_mac,
&using_bpf);
xfer_fd(fd, using_bpf ? 'D' : 'd');
return 11;
}
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
case 's': xfer_fd(create_raw_broadcast_socket(), 's'); return 1;
case 'u': {
uint32_t client_addr;
if (buflen < 1 + sizeof client_addr)
return 0;
memcpy(&client_addr, buf + 1, sizeof client_addr);
xfer_fd(create_udp_send_socket(client_addr), 'u');
return 5;
}
default: suicide("%s: (%s) received invalid commands: '%c'",
client_config.interface, __func__, c);
}
}
static void process_client_pipe(void)
{
static char buf[MAX_BUF];
static size_t buflen;
if (buflen == MAX_BUF)
suicide("%s: (%s) receive buffer exhausted", client_config.interface,
__func__);
Switch to using a socket for ndhc/sockd IPC so that fd passing works.
2014-04-05 14:55:56 +05:30
int r = safe_recv(sockdSock[1], buf + buflen, sizeof buf - buflen,
MSG_DONTWAIT);
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
if (r == 0) {
// Remote end hung up.
exit(EXIT_SUCCESS);
} else if (r < 0) {
if (errno == EAGAIN || errno == EWOULDBLOCK)
return;
suicide("%s: (%s) error reading from ndhc -> sockd pipe: %s",
client_config.interface, __func__, strerror(errno));
}
buflen += (size_t)r;
buflen -= execute_sockd(buf, buflen);
}
static void do_sockd_work(void)
{
epollfd = epoll_create1(0);
Change most error comparisons from == -1 to < 0. Some were not changed, as the different negative values equate to different errors. Tests against syscall returns and fds are very common and mostly fit the pattern of this change. The gain is increased range-exclusion.
2014-04-06 16:03:14 +05:30
if (epollfd < 0)
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
suicide("epoll_create1 failed");
if (enforce_seccomp_sockd())
log_line("sockd seccomp filter cannot be installed");
Switch to using a socket for ndhc/sockd IPC so that fd passing works.
2014-04-05 14:55:56 +05:30
epoll_add(epollfd, sockdSock[1]);
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
epoll_add(epollfd, signalFd);
for (;;) {
int r = epoll_wait(epollfd, events, 2, -1);
Change most error comparisons from == -1 to < 0. Some were not changed, as the different negative values equate to different errors. Tests against syscall returns and fds are very common and mostly fit the pattern of this change. The gain is increased range-exclusion.
2014-04-06 16:03:14 +05:30
if (r < 0) {
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
if (errno == EINTR)
continue;
else
suicide("epoll_wait failed");
}
for (int i = 0; i < r; ++i) {
int fd = events[i].data.fd;
Switch to using a socket for ndhc/sockd IPC so that fd passing works.
2014-04-05 14:55:56 +05:30
if (fd == sockdSock[1])
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
process_client_pipe();
else if (fd == signalFd)
signal_dispatch();
else
suicide("sockd: unexpected fd while performing epoll");
}
}
}
void sockd_main(void)
{
prctl(PR_SET_NAME, "ndhc: sockd");
umask(077);
setup_signals_sockd();
nk_set_chroot(chroot_dir);
memset(chroot_dir, 0, sizeof chroot_dir);
Update to latest ncmlib privilege.[ch].
2014-04-05 14:53:18 +05:30
nk_set_uidgid(sockd_uid, sockd_gid,
"cap_net_bind_service,cap_net_broadcast,cap_net_raw=ep");
Introduce a ndhc-sockd daemon that separates out the remaining elevated capabilities from the ndhc master process. Privsep is now complete. The only notable improvement from before is that exploitation of ndhc would only allow an attacker to open raw sockets, bind sockets to ports < port 1024, and create broadcast sockets on the interface that ndhc is performing dhcp on rather than on all interfaces. However, this seems like a worthwhile change; note that it was already impossible for an attacker to sniff packets on any interfaces (as that requires CAP_NET_ADMIN, which was always separated to ifch).
2014-04-04 13:42:25 +05:30
do_sockd_work();
}
Reference in New Issue Copy Permalink