1 /*
2 * Copyright 2012 Daniel Drown
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 *
16 * clatd.c - tun interface setup and main event loop
17 */
18 #include <arpa/inet.h>
19 #include <errno.h>
20 #include <fcntl.h>
21 #include <poll.h>
22 #include <signal.h>
23 #include <stdbool.h>
24 #include <stdio.h>
25 #include <stdlib.h>
26 #include <string.h>
27 #include <sys/ioctl.h>
28 #include <sys/prctl.h>
29 #include <sys/stat.h>
30 #include <sys/types.h>
31 #include <time.h>
32 #include <unistd.h>
33
34 #include <linux/filter.h>
35 #include <linux/if.h>
36 #include <linux/if_ether.h>
37 #include <linux/if_packet.h>
38 #include <linux/if_tun.h>
39 #include <linux/virtio_net.h>
40 #include <net/if.h>
41 #include <sys/uio.h>
42
43 #include "clatd.h"
44 #include "checksum.h"
45 #include "config.h"
46 #include "dump.h"
47 #include "logging.h"
48 #include "translate.h"
49
50 struct clat_config Global_Clatd_Config;
51
52 volatile sig_atomic_t sigterm = 0;
53 bool running = true;
54
55 // reads IPv6 packet from AF_PACKET socket, translates to IPv4, writes to tun
process_packet_6_to_4(struct tun_data * tunnel)56 void process_packet_6_to_4(struct tun_data *tunnel) {
57 // ethernet header is 14 bytes, plus 4 for a normal VLAN tag or 8 for Q-in-Q
58 // we don't really support vlans (or especially Q-in-Q)...
59 // but a few bytes of extra buffer space doesn't hurt...
60 struct {
61 struct virtio_net_hdr vnet;
62 uint8_t payload[22 + MAXMTU];
63 char pad; // +1 to make packet truncation obvious
64 } buf;
65 struct iovec iov = {
66 .iov_base = &buf,
67 .iov_len = sizeof(buf),
68 };
69 char cmsg_buf[CMSG_SPACE(sizeof(struct tpacket_auxdata))];
70 struct msghdr msgh = {
71 .msg_iov = &iov,
72 .msg_iovlen = 1,
73 .msg_control = cmsg_buf,
74 .msg_controllen = sizeof(cmsg_buf),
75 };
76 ssize_t readlen = recvmsg(tunnel->read_fd6, &msgh, /*flags*/ 0);
77
78 if (readlen < 0) {
79 if (errno != EAGAIN) {
80 logmsg(ANDROID_LOG_WARN, "%s: read error: %s", __func__, strerror(errno));
81 }
82 if (errno == ENETDOWN) running = false;
83 return;
84 } else if (readlen == 0) {
85 logmsg(ANDROID_LOG_WARN, "%s: packet socket removed?", __func__);
86 running = false;
87 return;
88 } else if (readlen >= sizeof(buf)) {
89 logmsg(ANDROID_LOG_WARN, "%s: read truncation - ignoring pkt", __func__);
90 return;
91 }
92
93 bool ok = false;
94 __u32 tp_status = 0;
95 __u16 tp_net = 0;
96
97 for (struct cmsghdr *cmsg = CMSG_FIRSTHDR(&msgh); cmsg != NULL; cmsg = CMSG_NXTHDR(&msgh,cmsg)) {
98 if (cmsg->cmsg_level == SOL_PACKET && cmsg->cmsg_type == PACKET_AUXDATA) {
99 struct tpacket_auxdata *aux = (struct tpacket_auxdata *)CMSG_DATA(cmsg);
100 ok = true;
101 tp_status = aux->tp_status;
102 tp_net = aux->tp_net;
103 break;
104 }
105 }
106
107 if (!ok) {
108 // theoretically this should not happen...
109 static bool logged = false;
110 if (!logged) {
111 logmsg(ANDROID_LOG_ERROR, "%s: failed to fetch tpacket_auxdata cmsg", __func__);
112 logged = true;
113 }
114 }
115
116 const int payload_offset = offsetof(typeof(buf), payload);
117 if (readlen < payload_offset + tp_net) {
118 logmsg(ANDROID_LOG_WARN, "%s: ignoring %zd byte pkt shorter than %d+%u L2 header",
119 __func__, readlen, payload_offset, tp_net);
120 return;
121 }
122
123 const int pkt_len = readlen - payload_offset;
124
125 // This will detect a skb->ip_summed == CHECKSUM_PARTIAL packet with non-final L4 checksum
126 if (tp_status & TP_STATUS_CSUMNOTREADY) {
127 static bool logged = false;
128 if (!logged) {
129 logmsg(ANDROID_LOG_WARN, "%s: L4 checksum calculation required", __func__);
130 logged = true;
131 }
132
133 // These are non-negative by virtue of csum_start/offset being u16
134 const int cs_start = buf.vnet.csum_start;
135 const int cs_offset = cs_start + buf.vnet.csum_offset;
136 if (cs_start > pkt_len) {
137 logmsg(ANDROID_LOG_ERROR, "%s: out of range - checksum start %d > %d",
138 __func__, cs_start, pkt_len);
139 } else if (cs_offset + 1 >= pkt_len) {
140 logmsg(ANDROID_LOG_ERROR, "%s: out of range - checksum offset %d + 1 >= %d",
141 __func__, cs_offset, pkt_len);
142 } else {
143 uint16_t csum = ip_checksum(buf.payload + cs_start, pkt_len - cs_start);
144 if (!csum) csum = 0xFFFF; // required fixup for UDP, TCP must live with it
145 buf.payload[cs_offset] = csum & 0xFF;
146 buf.payload[cs_offset + 1] = csum >> 8;
147 }
148 }
149
150 translate_packet(tunnel->fd4, 0 /* to_ipv6 */, buf.payload + tp_net, pkt_len - tp_net);
151 }
152
153 // reads TUN_PI + L3 IPv4 packet from tun, translates to IPv6, writes to AF_INET6/RAW socket
process_packet_4_to_6(struct tun_data * tunnel)154 void process_packet_4_to_6(struct tun_data *tunnel) {
155 struct {
156 struct tun_pi pi;
157 uint8_t payload[MAXMTU];
158 char pad; // +1 byte to make packet truncation obvious
159 } buf;
160 ssize_t readlen = read(tunnel->fd4, &buf, sizeof(buf));
161
162 if (readlen < 0) {
163 if (errno != EAGAIN) {
164 logmsg(ANDROID_LOG_WARN, "%s: read error: %s", __func__, strerror(errno));
165 }
166 if (errno == ENETDOWN) running = false; // not sure if this can happen
167 return;
168 } else if (readlen == 0) {
169 logmsg(ANDROID_LOG_WARN, "%s: tun interface removed", __func__);
170 running = false;
171 return;
172 } else if (readlen >= sizeof(buf)) {
173 logmsg(ANDROID_LOG_WARN, "%s: read truncation - ignoring pkt", __func__);
174 return;
175 }
176
177 const int payload_offset = offsetof(typeof(buf), payload);
178
179 if (readlen < payload_offset) {
180 logmsg(ANDROID_LOG_WARN, "%s: short read: got %ld bytes", __func__, readlen);
181 return;
182 }
183
184 const int pkt_len = readlen - payload_offset;
185
186 uint16_t proto = ntohs(buf.pi.proto);
187 if (proto != ETH_P_IP) {
188 logmsg(ANDROID_LOG_WARN, "%s: unknown packet type = 0x%x", __func__, proto);
189 return;
190 }
191
192 if (buf.pi.flags != 0) {
193 logmsg(ANDROID_LOG_WARN, "%s: unexpected flags = %d", __func__, buf.pi.flags);
194 }
195
196 translate_packet(tunnel->write_fd6, 1 /* to_ipv6 */, buf.payload, pkt_len);
197 }
198
199 // IPv6 DAD packet format:
200 // Ethernet header (if needed) will be added by the kernel:
201 // u8[6] src_mac; u8[6] dst_mac '33:33:ff:XX:XX:XX'; be16 ethertype '0x86DD'
202 // IPv6 header:
203 // be32 0x60000000 - ipv6, tclass 0, flowlabel 0
204 // be16 payload_length '32'; u8 nxt_hdr ICMPv6 '58'; u8 hop limit '255'
205 // u128 src_ip6 '::'
206 // u128 dst_ip6 'ff02::1:ffXX:XXXX'
207 // ICMPv6 header:
208 // u8 type '135'; u8 code '0'; u16 icmp6 checksum; u32 reserved '0'
209 // ICMPv6 neighbour solicitation payload:
210 // u128 tgt_ip6
211 // ICMPv6 ND options:
212 // u8 opt nr '14'; u8 length '1'; u8[6] nonce '6 random bytes'
send_dad(int fd,const struct in6_addr * tgt)213 void send_dad(int fd, const struct in6_addr* tgt) {
214 struct {
215 struct ip6_hdr ip6h;
216 struct nd_neighbor_solicit ns;
217 uint8_t ns_opt_nr;
218 uint8_t ns_opt_len;
219 uint8_t ns_opt_nonce[6];
220 } dad_pkt = {
221 .ip6h = {
222 .ip6_flow = htonl(6 << 28), // v6, 0 tclass, 0 flowlabel
223 .ip6_plen = htons(sizeof(dad_pkt) - sizeof(struct ip6_hdr)), // payload length, ie. 32
224 .ip6_nxt = IPPROTO_ICMPV6, // 58
225 .ip6_hlim = 255,
226 .ip6_src = {}, // ::
227 .ip6_dst.s6_addr = {
228 0xFF, 0x02, 0, 0,
229 0, 0, 0, 0,
230 0, 0, 0, 1,
231 0xFF, tgt->s6_addr[13], tgt->s6_addr[14], tgt->s6_addr[15],
232 }, // ff02::1:ffXX:XXXX - multicast group address derived from bottom 24-bits of tgt
233 },
234 .ns = {
235 .nd_ns_type = ND_NEIGHBOR_SOLICIT, // 135
236 .nd_ns_code = 0,
237 .nd_ns_cksum = 0, // will be calculated later
238 .nd_ns_reserved = 0,
239 .nd_ns_target = *tgt,
240 },
241 .ns_opt_nr = 14, // icmp6 option 'nonce' from RFC3971
242 .ns_opt_len = 1, // in units of 8 bytes, including option nr and len
243 .ns_opt_nonce = {}, // opt_len *8 - sizeof u8(opt_nr) - sizeof u8(opt_len) = 6 ranodmized bytes
244 };
245 arc4random_buf(&dad_pkt.ns_opt_nonce, sizeof(dad_pkt.ns_opt_nonce));
246
247 // 40 byte IPv6 header + 8 byte ICMPv6 header + 16 byte ipv6 target address + 8 byte nonce option
248 _Static_assert(sizeof(dad_pkt) == 40 + 8 + 16 + 8, "sizeof dad packet != 72");
249
250 // IPv6 header checksum is standard negated 16-bit one's complement sum over the icmpv6 pseudo
251 // header (which includes payload length, nextheader, and src/dst ip) and the icmpv6 payload.
252 //
253 // Src/dst ip immediately prefix the icmpv6 header itself, so can be handled along
254 // with the payload. We thus only need to manually account for payload len & next header.
255 //
256 // The magic '8' is simply the offset of the ip6_src field in the ipv6 header,
257 // ie. we're skipping over the ipv6 version, tclass, flowlabel, payload length, next header
258 // and hop limit fields, because they're not quite where we want them to be.
259 //
260 // ip6_plen is already in network order, while ip6_nxt is a single byte and thus needs htons().
261 uint32_t csum = dad_pkt.ip6h.ip6_plen + htons(dad_pkt.ip6h.ip6_nxt);
262 csum = ip_checksum_add(csum, &dad_pkt.ip6h.ip6_src, sizeof(dad_pkt) - 8);
263 dad_pkt.ns.nd_ns_cksum = ip_checksum_finish(csum);
264
265 const struct sockaddr_in6 dst = {
266 .sin6_family = AF_INET6,
267 .sin6_addr = dad_pkt.ip6h.ip6_dst,
268 .sin6_scope_id = if_nametoindex(Global_Clatd_Config.native_ipv6_interface),
269 };
270
271 sendto(fd, &dad_pkt, sizeof(dad_pkt), 0 /*flags*/, (const struct sockaddr *)&dst, sizeof(dst));
272 }
273
274 /* function: event_loop
275 * reads packets from the tun network interface and passes them down the stack
276 * tunnel - tun device data
277 */
event_loop(struct tun_data * tunnel)278 void event_loop(struct tun_data *tunnel) {
279 struct pollfd wait_fd[] = {
280 { tunnel->read_fd6, POLLIN, 0 },
281 { tunnel->fd4, POLLIN, 0 },
282 };
283
284 while (running && !sigterm) {
285 if (poll(wait_fd, ARRAY_SIZE(wait_fd), -1) == -1) {
286 if (errno != EINTR) {
287 logmsg(ANDROID_LOG_WARN, "event_loop/poll returned an error: %s", strerror(errno));
288 }
289 } else {
290 // Call process_packet if the socket has data to be read, but also if an
291 // error is waiting. If we don't call read() after getting POLLERR, a
292 // subsequent poll() will return immediately with POLLERR again,
293 // causing this code to spin in a loop. Calling read() will clear the
294 // socket error flag instead.
295 if (wait_fd[0].revents) process_packet_6_to_4(tunnel);
296 if (wait_fd[1].revents) process_packet_4_to_6(tunnel);
297 }
298 }
299 }
300