blob: 6039d9a05a5c08b89542a8b3c6b2828067172e04 [file] [log] [blame]
Daniel Drown0da73fc2012-06-20 16:51:39 -05001/*
2 * Copyright (C) 2008 The Android Open Source Project
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 */
Bernie Innocenti51a0e0f2018-10-05 20:24:06 +090016
Lorenzo Colittiac7fefc2014-10-20 17:14:13 +090017#include <map>
18#include <string>
19
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +090020#include <arpa/inet.h>
Daniel Drown0da73fc2012-06-20 16:51:39 -050021#include <errno.h>
Maciej Żenczykowski1c086e52019-03-29 23:13:49 -070022#include <linux/if_arp.h>
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -070023#include <linux/if_tun.h>
24#include <linux/ioctl.h>
Maciej Żenczykowskic8c38aa2019-03-29 01:24:51 -070025#include <net/if.h>
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +090026#include <netinet/in.h>
Luke Huang94c43a12019-02-14 19:51:38 +080027#include <spawn.h>
Daniel Drown0da73fc2012-06-20 16:51:39 -050028#include <sys/types.h>
29#include <sys/wait.h>
Luke Huang94c43a12019-02-14 19:51:38 +080030#include <unistd.h>
Daniel Drown0da73fc2012-06-20 16:51:39 -050031
32#define LOG_TAG "ClatdController"
Logan Chien3f461482018-04-23 14:31:32 +080033#include <log/log.h>
Daniel Drown0da73fc2012-06-20 16:51:39 -050034
Maciej Żenczykowski1c086e52019-03-29 23:13:49 -070035#include "ClatdController.h"
Maciej Żenczykowskia1699952019-05-11 17:07:44 -070036#include "InterfaceController.h"
Maciej Żenczykowski1c086e52019-03-29 23:13:49 -070037
38#include "android-base/properties.h"
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -070039#include "android-base/scopeguard.h"
Lorenzo Colitti91fd5802019-06-28 19:22:01 +090040#include "android-base/stringprintf.h"
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +090041#include "android-base/unique_fd.h"
Maciej Żenczykowski55262712019-03-29 23:44:56 -070042#include "bpf/BpfMap.h"
43#include "netdbpf/bpf_shared.h"
44#include "netdutils/DumpWriter.h"
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +090045
46extern "C" {
47#include "netutils/checksum.h"
48}
49
Maciej Żenczykowski55262712019-03-29 23:44:56 -070050#include "ClatUtils.h"
Lorenzo Colitti45d3dd02014-06-09 14:09:20 +090051#include "Fwmark.h"
Paul Jensen84c1d032014-05-30 13:29:41 -040052#include "NetdConstants.h"
53#include "NetworkController.h"
Bernie Innocenti189eb502018-10-01 23:10:18 +090054#include "netid_client.h"
Daniel Drown0da73fc2012-06-20 16:51:39 -050055
Lorenzo Colittiac7fefc2014-10-20 17:14:13 +090056static const char* kClatdPath = "/system/bin/clatd";
57
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +090058// For historical reasons, start with 192.0.0.4, and after that, use all subsequent addresses in
59// 192.0.0.0/29 (RFC 7335).
60static const char* kV4AddrString = "192.0.0.4";
61static const in_addr kV4Addr = {inet_addr(kV4AddrString)};
62static const int kV4AddrLen = 29;
63
Lorenzo Colitti91fd5802019-06-28 19:22:01 +090064using android::base::StringPrintf;
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +090065using android::base::unique_fd;
Maciej Żenczykowski55262712019-03-29 23:44:56 -070066using android::bpf::BpfMap;
67using android::netdutils::DumpWriter;
68using android::netdutils::ScopedIndent;
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +090069
Lorenzo Colitti7035f222017-02-13 18:29:00 +090070namespace android {
71namespace net {
72
Maciej Żenczykowski56280272019-03-30 03:32:51 -070073void ClatdController::init(void) {
74 std::lock_guard guard(mutex);
75
Maciej Żenczykowski1c086e52019-03-29 23:13:49 -070076 // TODO: should refactor into separate function for testability
77 if (bpf::getBpfSupportLevel() == bpf::BpfLevel::NONE) {
78 ALOGI("Pre-4.9 kernel or pre-P api shipping level - disabling clat ebpf.");
79 mClatEbpfMode = ClatEbpfDisabled;
80 return;
81 }
82
83 // We know the device initially shipped with at least P...,
84 // but did it ship with at least Q?
85
86 uint64_t api_level = base::GetUintProperty<uint64_t>("ro.product.first_api_level", 0);
87 if (api_level == 0) {
88 ALOGE("Cannot determine initial API level of the device.");
89 api_level = base::GetUintProperty<uint64_t>("ro.build.version.sdk", 0);
90 }
91
92 // Note: MINIMUM_API_REQUIRED is for eBPF as a whole and is thus P
93 if (api_level > bpf::MINIMUM_API_REQUIRED) {
94 ALOGI("4.9+ kernel and device shipped with Q+ - clat ebpf should work.");
95 mClatEbpfMode = ClatEbpfEnabled;
96 } else {
97 // We cannot guarantee that 4.9-P kernels will include NET_CLS_BPF support.
98 ALOGI("4.9+ kernel and device shipped with P - clat ebpf might work.");
99 mClatEbpfMode = ClatEbpfMaybe;
100 }
101
102 int rv = openNetlinkSocket();
103 if (rv < 0) {
104 ALOGE("openNetlinkSocket() failure: %s", strerror(-rv));
105 mClatEbpfMode = ClatEbpfDisabled;
106 return;
107 }
108 mNetlinkFd.reset(rv);
109
110 rv = getClatIngressMapFd();
111 if (rv < 0) {
112 ALOGE("getClatIngressMapFd() failure: %s", strerror(-rv));
113 mClatEbpfMode = ClatEbpfDisabled;
114 mNetlinkFd.reset(-1);
115 return;
116 }
117 mClatIngressMap.reset(rv);
118
119 int netlinkFd = mNetlinkFd.get();
120
121 // TODO: perhaps this initial cleanup should be in its own function?
122 const auto del = [&netlinkFd](const ClatIngressKey& key,
123 const BpfMap<ClatIngressKey, ClatIngressValue>&) {
124 ALOGW("Removing stale clat config on interface %d.", key.iif);
125 int rv = tcQdiscDelDevClsact(netlinkFd, key.iif);
126 if (rv < 0) ALOGE("tcQdiscDelDevClsact() failure: %s", strerror(-rv));
127 return netdutils::status::ok; // keep on going regardless
128 };
129 auto ret = mClatIngressMap.iterate(del);
130 if (!isOk(ret)) ALOGE("mClatIngressMap.iterate() failure: %s", strerror(ret.code()));
131 ret = mClatIngressMap.clear();
132 if (!isOk(ret)) ALOGE("mClatIngressMap.clear() failure: %s", strerror(ret.code()));
133}
134
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900135bool ClatdController::isIpv4AddressFree(in_addr_t addr) {
136 int s = socket(AF_INET, SOCK_DGRAM | SOCK_CLOEXEC, 0);
137 if (s == -1) {
138 return 0;
139 }
140
141 // Attempt to connect to the address. If the connection succeeds and getsockname returns the
142 // same then the address is already assigned to the system and we can't use it.
143 struct sockaddr_in sin = {.sin_family = AF_INET, .sin_addr = {addr}, .sin_port = 53};
144 socklen_t len = sizeof(sin);
145 bool inuse = connect(s, (struct sockaddr*)&sin, sizeof(sin)) == 0 &&
146 getsockname(s, (struct sockaddr*)&sin, &len) == 0 && (size_t)len >= sizeof(sin) &&
147 sin.sin_addr.s_addr == addr;
148
149 close(s);
150 return !inuse;
Lorenzo Colittiac7fefc2014-10-20 17:14:13 +0900151}
Daniel Drown0da73fc2012-06-20 16:51:39 -0500152
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900153// Picks a free IPv4 address, starting from ip and trying all addresses in the prefix in order.
154// ip - the IP address from the configuration file
155// prefixlen - the length of the prefix from which addresses may be selected.
156// returns: the IPv4 address, or INADDR_NONE if no addresses were available
157in_addr_t ClatdController::selectIpv4Address(const in_addr ip, int16_t prefixlen) {
158 // Don't accept prefixes that are too large because we scan addresses one by one.
159 if (prefixlen < 16 || prefixlen > 32) {
160 return INADDR_NONE;
161 }
Lorenzo Colittiac7fefc2014-10-20 17:14:13 +0900162
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900163 // All these are in host byte order.
164 in_addr_t mask = 0xffffffff >> (32 - prefixlen) << (32 - prefixlen);
165 in_addr_t ipv4 = ntohl(ip.s_addr);
166 in_addr_t first_ipv4 = ipv4;
167 in_addr_t prefix = ipv4 & mask;
168
169 // Pick the first IPv4 address in the pool, wrapping around if necessary.
170 // So, for example, 192.0.0.4 -> 192.0.0.5 -> 192.0.0.6 -> 192.0.0.7 -> 192.0.0.0.
171 do {
172 if (isIpv4AddressFreeFunc(htonl(ipv4))) {
173 return htonl(ipv4);
174 }
175 ipv4 = prefix | ((ipv4 + 1) & ~mask);
176 } while (ipv4 != first_ipv4);
177
178 return INADDR_NONE;
179}
180
181// Alters the bits in the IPv6 address to make them checksum neutral with v4 and nat64Prefix.
182void ClatdController::makeChecksumNeutral(in6_addr* v6, const in_addr v4,
183 const in6_addr& nat64Prefix) {
184 // Fill last 8 bytes of IPv6 address with random bits.
185 arc4random_buf(&v6->s6_addr[8], 8);
186
187 // Make the IID checksum-neutral. That is, make it so that:
188 // checksum(Local IPv4 | Remote IPv4) = checksum(Local IPv6 | Remote IPv6)
189 // in other words (because remote IPv6 = NAT64 prefix | Remote IPv4):
190 // checksum(Local IPv4) = checksum(Local IPv6 | NAT64 prefix)
191 // Do this by adjusting the two bytes in the middle of the IID.
192
193 uint16_t middlebytes = (v6->s6_addr[11] << 8) + v6->s6_addr[12];
194
195 uint32_t c1 = ip_checksum_add(0, &v4, sizeof(v4));
196 uint32_t c2 = ip_checksum_add(0, &nat64Prefix, sizeof(nat64Prefix)) +
197 ip_checksum_add(0, v6, sizeof(*v6));
198
199 uint16_t delta = ip_checksum_adjust(middlebytes, c1, c2);
200 v6->s6_addr[11] = delta >> 8;
201 v6->s6_addr[12] = delta & 0xff;
202}
203
204// Picks a random interface ID that is checksum neutral with the IPv4 address and the NAT64 prefix.
205int ClatdController::generateIpv6Address(const char* iface, const in_addr v4,
206 const in6_addr& nat64Prefix, in6_addr* v6) {
207 unique_fd s(socket(AF_INET6, SOCK_DGRAM | SOCK_CLOEXEC, 0));
208 if (s == -1) return -errno;
209
210 if (setsockopt(s, SOL_SOCKET, SO_BINDTODEVICE, iface, strlen(iface) + 1) == -1) {
Luke Huang6d301232018-08-01 14:05:18 +0800211 return -errno;
Daniel Drown0da73fc2012-06-20 16:51:39 -0500212 }
213
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900214 sockaddr_in6 sin6 = {.sin6_family = AF_INET6, .sin6_addr = nat64Prefix};
215 if (connect(s, reinterpret_cast<struct sockaddr*>(&sin6), sizeof(sin6)) == -1) {
216 return -errno;
217 }
218
219 socklen_t len = sizeof(sin6);
220 if (getsockname(s, reinterpret_cast<struct sockaddr*>(&sin6), &len) == -1) {
221 return -errno;
222 }
223
224 *v6 = sin6.sin6_addr;
225
226 if (IN6_IS_ADDR_UNSPECIFIED(v6) || IN6_IS_ADDR_LOOPBACK(v6) || IN6_IS_ADDR_LINKLOCAL(v6) ||
227 IN6_IS_ADDR_SITELOCAL(v6) || IN6_IS_ADDR_ULA(v6)) {
228 return -ENETUNREACH;
229 }
230
231 makeChecksumNeutral(v6, v4, nat64Prefix);
232
233 return 0;
234}
235
Maciej Żenczykowski1c086e52019-03-29 23:13:49 -0700236void ClatdController::maybeStartBpf(const ClatdTracker& tracker) {
237 if (mClatEbpfMode == ClatEbpfDisabled) return;
238
239 int rv = hardwareAddressType(tracker.iface);
240 if (rv < 0) {
241 ALOGE("hardwareAddressType(%s[%d]) failure: %s", tracker.iface, tracker.ifIndex,
242 strerror(-rv));
243 return;
244 }
245
246 bool isEthernet;
247 switch (rv) {
248 case ARPHRD_ETHER:
249 isEthernet = true;
250 break;
251 case ARPHRD_RAWIP: // in Linux 4.14+ rmnet support was upstreamed and this is 519
252 case 530: // this is ARPHRD_RAWIP on some Android 4.9 kernels with rmnet
253 isEthernet = false;
254 break;
255 default:
256 ALOGE("hardwareAddressType(%s[%d]) returned unknown type %d.", tracker.iface,
257 tracker.ifIndex, rv);
258 return;
259 }
260
261 rv = getClatIngressProgFd(isEthernet);
262 if (rv < 0) {
263 ALOGE("getClatIngressProgFd(%d) failure: %s", isEthernet, strerror(-rv));
264 return;
265 }
266 unique_fd progFd(rv);
267
268 ClatIngressKey key = {
269 .iif = tracker.ifIndex,
270 .pfx96 = tracker.pfx96,
271 .local6 = tracker.v6,
272 };
273 ClatIngressValue value = {
Maciej Żenczykowski1c086e52019-03-29 23:13:49 -0700274 // TODO: move all the clat code to eBPF and remove the tun interface entirely.
Maciej Żenczykowski39b0b902019-05-08 00:36:30 -0700275 .oif = tracker.v4ifIndex,
Maciej Żenczykowski1c086e52019-03-29 23:13:49 -0700276 .local4 = tracker.v4,
277 };
278
279 auto ret = mClatIngressMap.writeValue(key, value, BPF_ANY);
280 if (!isOk(ret)) {
281 ALOGE("mClatIngress.Map.writeValue failure: %s", strerror(ret.code()));
282 return;
283 }
284
285 // We do tc setup *after* populating map, so scanning through map
286 // can always be used to tell us what needs cleanup.
287
288 rv = tcQdiscAddDevClsact(mNetlinkFd, tracker.ifIndex);
289 if (rv) {
290 ALOGE("tcQdiscAddDevClsact(%d[%s]) failure: %s", tracker.ifIndex, tracker.iface,
291 strerror(-rv));
292 ret = mClatIngressMap.deleteValue(key);
293 if (!isOk(ret)) ALOGE("mClatIngressMap.deleteValue failure: %s", strerror(ret.code()));
294 return;
295 }
296
297 rv = tcFilterAddDevBpf(mNetlinkFd, tracker.ifIndex, progFd, isEthernet);
298 if (rv) {
299 if ((rv == -ENOENT) && (mClatEbpfMode == ClatEbpfMaybe)) {
300 ALOGI("tcFilterAddDevBpf(%d[%s], %d): %s", tracker.ifIndex, tracker.iface, isEthernet,
301 strerror(-rv));
302 } else {
303 ALOGE("tcFilterAddDevBpf(%d[%s], %d) failure: %s", tracker.ifIndex, tracker.iface,
304 isEthernet, strerror(-rv));
305 }
306 rv = tcQdiscDelDevClsact(mNetlinkFd, tracker.ifIndex);
307 if (rv)
308 ALOGE("tcQdiscDelDevClsact(%d[%s]) failure: %s", tracker.ifIndex, tracker.iface,
309 strerror(-rv));
310 ret = mClatIngressMap.deleteValue(key);
311 if (!isOk(ret)) ALOGE("mClatIngressMap.deleteValue failure: %s", strerror(ret.code()));
312 return;
313 }
314
315 // success
316}
317
Lorenzo Colitti91fd5802019-06-28 19:22:01 +0900318void ClatdController::maybeSetIptablesDropRule(bool add, const char* pfx96Str, const char* v6Str) {
319 if (mClatEbpfMode == ClatEbpfDisabled) return;
320
321 std::string cmd = StringPrintf(
322 "*raw\n"
323 "%s %s -s %s/96 -d %s -j DROP\n"
324 "COMMIT\n",
325 (add ? "-A" : "-D"), LOCAL_RAW_PREROUTING, pfx96Str, v6Str);
326
327 iptablesRestoreFunction(V6, cmd);
328}
329
Maciej Żenczykowski1c086e52019-03-29 23:13:49 -0700330void ClatdController::maybeStopBpf(const ClatdTracker& tracker) {
331 if (mClatEbpfMode == ClatEbpfDisabled) return;
332
333 // No need to remove filter, since we remove qdisc it is attached to,
334 // which automatically removes everything attached to the qdisc.
335 int rv = tcQdiscDelDevClsact(mNetlinkFd, tracker.ifIndex);
336 if (rv < 0)
337 ALOGE("tcQdiscDelDevClsact(%d[%s]) failure: %s", tracker.ifIndex, tracker.iface,
338 strerror(-rv));
339
340 // We cleanup map last, so scanning through map can be used to
341 // determine what still needs cleanup.
342
343 ClatIngressKey key = {
344 .iif = tracker.ifIndex,
345 .pfx96 = tracker.pfx96,
346 .local6 = tracker.v6,
347 };
348
349 auto ret = mClatIngressMap.deleteValue(key);
350 if (!isOk(ret)) ALOGE("mClatIngressMap.deleteValue failure: %s", strerror(ret.code()));
351}
352
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900353// Finds the tracker of the clatd running on interface |interface|, or nullptr if clatd has not been
354// started on |interface|.
355ClatdController::ClatdTracker* ClatdController::getClatdTracker(const std::string& interface) {
356 auto it = mClatdTrackers.find(interface);
357 return (it == mClatdTrackers.end() ? nullptr : &it->second);
358}
359
360// Initializes a ClatdTracker for the specified interface.
Maciej Żenczykowskia56b2e62019-04-24 13:17:18 -0700361int ClatdController::ClatdTracker::init(unsigned networkId, const std::string& interface,
Maciej Żenczykowski4657e512019-05-08 06:35:08 -0700362 const std::string& v4interface,
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900363 const std::string& nat64Prefix) {
Maciej Żenczykowskia56b2e62019-04-24 13:17:18 -0700364 netId = networkId;
Lorenzo Colitti32b2e792015-01-07 15:11:30 +0900365
Lorenzo Colitti32b2e792015-01-07 15:11:30 +0900366 fwmark.netId = netId;
367 fwmark.explicitlySelected = true;
368 fwmark.protectedFromVpn = true;
369 fwmark.permission = PERMISSION_SYSTEM;
370
Lorenzo Colitti32b2e792015-01-07 15:11:30 +0900371 snprintf(fwmarkString, sizeof(fwmarkString), "0x%x", fwmark.intValue);
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900372 snprintf(netIdString, sizeof(netIdString), "%u", netId);
373 strlcpy(iface, interface.c_str(), sizeof(iface));
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700374 ifIndex = if_nametoindex(iface);
Maciej Żenczykowski4657e512019-05-08 06:35:08 -0700375 strlcpy(v4iface, v4interface.c_str(), sizeof(v4iface));
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700376 v4ifIndex = if_nametoindex(v4iface);
Lorenzo Colitti32b2e792015-01-07 15:11:30 +0900377
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900378 // Pass in everything that clatd needs: interface, a netid to use for DNS lookups, a fwmark for
379 // outgoing packets, the NAT64 prefix, and the IPv4 and IPv6 addresses.
380 // Validate the prefix and strip off the prefix length.
381 uint8_t family;
382 uint8_t prefixLen;
Maciej Żenczykowski1c06f9c2019-03-29 23:19:19 -0700383 int res = parsePrefix(nat64Prefix.c_str(), &family, &pfx96, sizeof(pfx96), &prefixLen);
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900384 // clatd only supports /96 prefixes.
Maciej Żenczykowski1c06f9c2019-03-29 23:19:19 -0700385 if (res != sizeof(pfx96)) return res;
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900386 if (family != AF_INET6) return -EAFNOSUPPORT;
387 if (prefixLen != 96) return -EINVAL;
Maciej Żenczykowski1c06f9c2019-03-29 23:19:19 -0700388 if (!inet_ntop(AF_INET6, &pfx96, pfx96String, sizeof(pfx96String))) return -errno;
Luke Huang6d301232018-08-01 14:05:18 +0800389
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900390 // Pick an IPv4 address.
391 // TODO: this picks the address based on other addresses that are assigned to interfaces, but
392 // the address is only actually assigned to an interface once clatd starts up. So we could end
393 // up with two clatd instances with the same IPv4 address.
394 // Stop doing this and instead pick a free one from the kV4Addr pool.
Maciej Żenczykowski55cacfb2019-03-30 02:01:35 -0700395 v4 = {selectIpv4Address(kV4Addr, kV4AddrLen)};
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900396 if (v4.s_addr == INADDR_NONE) {
397 ALOGE("No free IPv4 address in %s/%d", kV4AddrString, kV4AddrLen);
398 return -EADDRNOTAVAIL;
399 }
400 if (!inet_ntop(AF_INET, &v4, v4Str, sizeof(v4Str))) return -errno;
401
402 // Generate a checksum-neutral IID.
Maciej Żenczykowski1c06f9c2019-03-29 23:19:19 -0700403 if (generateIpv6Address(iface, v4, pfx96, &v6)) {
404 ALOGE("Unable to find global source address on %s for %s", iface, pfx96String);
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900405 return -EADDRNOTAVAIL;
406 }
407 if (!inet_ntop(AF_INET6, &v6, v6Str, sizeof(v6Str))) return -errno;
408
Maciej Żenczykowski1c06f9c2019-03-29 23:19:19 -0700409 ALOGD("starting clatd on %s v4=%s v6=%s pfx96=%s", iface, v4Str, v6Str, pfx96String);
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900410 return 0;
411}
412
413int ClatdController::startClatd(const std::string& interface, const std::string& nat64Prefix,
414 std::string* v6Str) {
Maciej Żenczykowski56280272019-03-30 03:32:51 -0700415 std::lock_guard guard(mutex);
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700416
417 // 1. fail if pre-existing tracker already exists
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900418 ClatdTracker* existing = getClatdTracker(interface);
419 if (existing != nullptr) {
420 ALOGE("clatd pid=%d already started on %s", existing->pid, interface.c_str());
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700421 return -EBUSY;
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900422 }
423
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700424 // 2. get network id associated with this external interface
Maciej Żenczykowskia56b2e62019-04-24 13:17:18 -0700425 unsigned networkId = mNetCtrl->getNetworkForInterface(interface.c_str());
426 if (networkId == NETID_UNSET) {
427 ALOGE("Interface %s not assigned to any netId", interface.c_str());
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700428 return -ENODEV;
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900429 }
JP Abgrall69261cb2014-06-19 18:35:24 -0700430
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700431 // 3. open the tun device in non blocking mode as required by clatd
432 int res = open("/dev/tun", O_RDWR | O_NONBLOCK | O_CLOEXEC);
433 if (res == -1) {
434 res = errno;
435 ALOGE("open of tun device failed (%s)", strerror(res));
436 return -res;
437 }
438 unique_fd tmpTunFd(res);
439
440 // 4. create the v4-... tun interface
Maciej Żenczykowski4657e512019-05-08 06:35:08 -0700441 std::string v4interface("v4-");
442 v4interface += interface;
443
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700444 struct ifreq ifr = {
445 .ifr_flags = IFF_TUN,
446 };
Maciej Żenczykowski4657e512019-05-08 06:35:08 -0700447 strlcpy(ifr.ifr_name, v4interface.c_str(), sizeof(ifr.ifr_name));
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700448
449 res = ioctl(tmpTunFd, TUNSETIFF, &ifr, sizeof(ifr));
450 if (res == -1) {
451 res = errno;
452 ALOGE("ioctl(TUNSETIFF) failed (%s)", strerror(res));
453 return -res;
454 }
455
Maciej Żenczykowskia1699952019-05-11 17:07:44 -0700456 // disable IPv6 on it - failing to do so is not a critical error
457 res = InterfaceController::setEnableIPv6(v4interface.c_str(), 0);
458 if (res) ALOGE("setEnableIPv6 %s failed (%s)", v4interface.c_str(), strerror(res));
459
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700460 // 5. initialize tracker object
Maciej Żenczykowskia56b2e62019-04-24 13:17:18 -0700461 ClatdTracker tracker;
Maciej Żenczykowski4657e512019-05-08 06:35:08 -0700462 int ret = tracker.init(networkId, interface, v4interface, nat64Prefix);
Maciej Żenczykowskia56b2e62019-04-24 13:17:18 -0700463 if (ret) return ret;
464
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700465 // 6. create a throwaway socket to reserve a file descriptor number
466 res = socket(AF_INET6, SOCK_DGRAM | SOCK_CLOEXEC, 0);
467 if (res == -1) {
468 res = errno;
469 ALOGE("socket(ipv6/udp) failed (%s)", strerror(res));
470 return -res;
471 }
472 unique_fd passedTunFd(res);
473
474 // 7. this is the FD we'll pass to clatd on the cli, so need it as a string
475 char passedTunFdStr[INT32_STRLEN];
476 snprintf(passedTunFdStr, sizeof(passedTunFdStr), "%d", passedTunFd.get());
477
478 // 8. we're going to use this as argv[0] to clatd to make ps output more useful
Lorenzo Colittiac7fefc2014-10-20 17:14:13 +0900479 std::string progname("clatd-");
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900480 progname += tracker.iface;
Daniel Drown0da73fc2012-06-20 16:51:39 -0500481
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900482 // clang-format off
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700483 const char* args[] = {progname.c_str(),
484 "-i", tracker.iface,
485 "-n", tracker.netIdString,
486 "-m", tracker.fwmarkString,
487 "-p", tracker.pfx96String,
488 "-4", tracker.v4Str,
489 "-6", tracker.v6Str,
490 "-t", passedTunFdStr,
491 nullptr};
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900492 // clang-format on
493
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700494 // 9. register vfork requirement
495 posix_spawnattr_t attr;
496 res = posix_spawnattr_init(&attr);
497 if (res) {
498 ALOGE("posix_spawnattr_init failed (%s)", strerror(res));
499 return -res;
500 }
501 const android::base::ScopeGuard attrGuard = [&] { posix_spawnattr_destroy(&attr); };
502 res = posix_spawnattr_setflags(&attr, POSIX_SPAWN_USEVFORK);
503 if (res) {
504 ALOGE("posix_spawnattr_setflags failed (%s)", strerror(res));
505 return -res;
506 }
507
508 // 10. register dup2() action: this is what 'clears' the CLOEXEC flag
509 // on the tun fd that we want the child clatd process to inherit
510 // (this will happen after the vfork, and before the execve)
511 posix_spawn_file_actions_t fa;
512 res = posix_spawn_file_actions_init(&fa);
513 if (res) {
514 ALOGE("posix_spawn_file_actions_init failed (%s)", strerror(res));
515 return -res;
516 }
517 const android::base::ScopeGuard faGuard = [&] { posix_spawn_file_actions_destroy(&fa); };
518 res = posix_spawn_file_actions_adddup2(&fa, tmpTunFd, passedTunFd);
519 if (res) {
520 ALOGE("posix_spawn_file_actions_adddup2 failed (%s)", strerror(res));
521 return -res;
522 }
523
Lorenzo Colitti91fd5802019-06-28 19:22:01 +0900524 // 11. If necessary, add the drop rule for iptables.
525 maybeSetIptablesDropRule(true, tracker.pfx96String, tracker.v6Str);
526
527 // 12. actually perform vfork/dup2/execve
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700528 res = posix_spawn(&tracker.pid, kClatdPath, &fa, &attr, (char* const*)args, nullptr);
Luke Huang40962442019-02-26 11:46:10 +0800529 if (res) {
530 ALOGE("posix_spawn failed (%s)", strerror(res));
Luke Huang6d301232018-08-01 14:05:18 +0800531 return -res;
Daniel Drown0da73fc2012-06-20 16:51:39 -0500532 }
533
Lorenzo Colitti91fd5802019-06-28 19:22:01 +0900534 // 13. configure eBPF offload - if possible
Maciej Żenczykowski1c086e52019-03-29 23:13:49 -0700535 maybeStartBpf(tracker);
536
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900537 mClatdTrackers[interface] = tracker;
538 ALOGD("clatd started on %s", interface.c_str());
Daniel Drown0da73fc2012-06-20 16:51:39 -0500539
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900540 *v6Str = tracker.v6Str;
Daniel Drown0da73fc2012-06-20 16:51:39 -0500541 return 0;
542}
543
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900544int ClatdController::stopClatd(const std::string& interface) {
Maciej Żenczykowski56280272019-03-30 03:32:51 -0700545 std::lock_guard guard(mutex);
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900546 ClatdTracker* tracker = getClatdTracker(interface);
Lorenzo Colittiac7fefc2014-10-20 17:14:13 +0900547
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900548 if (tracker == nullptr) {
Daniel Drown0da73fc2012-06-20 16:51:39 -0500549 ALOGE("clatd already stopped");
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900550 return -ENODEV;
Daniel Drown0da73fc2012-06-20 16:51:39 -0500551 }
552
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900553 ALOGD("Stopping clatd pid=%d on %s", tracker->pid, interface.c_str());
Daniel Drown0da73fc2012-06-20 16:51:39 -0500554
Maciej Żenczykowski1c086e52019-03-29 23:13:49 -0700555 maybeStopBpf(*tracker);
556
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900557 kill(tracker->pid, SIGTERM);
558 waitpid(tracker->pid, nullptr, 0);
Lorenzo Colitti91fd5802019-06-28 19:22:01 +0900559
560 maybeSetIptablesDropRule(false, tracker->pfx96String, tracker->v6Str);
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900561 mClatdTrackers.erase(interface);
Daniel Drown0da73fc2012-06-20 16:51:39 -0500562
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900563 ALOGD("clatd on %s stopped", interface.c_str());
Daniel Drown0da73fc2012-06-20 16:51:39 -0500564
565 return 0;
566}
567
Maciej Żenczykowski55262712019-03-29 23:44:56 -0700568void ClatdController::dump(DumpWriter& dw) {
569 std::lock_guard guard(mutex);
570
571 ScopedIndent clatdIndent(dw);
572 dw.println("ClatdController");
573
574 {
575 ScopedIndent trackerIndent(dw);
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700576 dw.println("Trackers: iif[iface] nat64Prefix v6Addr -> v4Addr v4iif[v4iface] [netId]");
Maciej Żenczykowski55262712019-03-29 23:44:56 -0700577
578 ScopedIndent trackerDetailIndent(dw);
579 for (const auto& pair : mClatdTrackers) {
580 const ClatdTracker& tracker = pair.second;
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700581 dw.println("%u[%s] %s/96 %s -> %s %u[%s] [%u]", tracker.ifIndex, tracker.iface,
582 tracker.pfx96String, tracker.v6Str, tracker.v4Str, tracker.v4ifIndex,
583 tracker.v4iface, tracker.netId);
Maciej Żenczykowski55262712019-03-29 23:44:56 -0700584 }
585 }
586
Maciej Żenczykowskiabb2cbf2019-04-01 01:29:29 -0700587 int mapFd = getClatIngressMapFd();
588 if (mapFd < 0) return; // if unsupported just don't dump anything
589 BpfMap<ClatIngressKey, ClatIngressValue> configMap(mapFd);
Maciej Żenczykowski55262712019-03-29 23:44:56 -0700590
591 ScopedIndent bpfIndent(dw);
592 dw.println("BPF ingress map: iif(iface) nat64Prefix v6Addr -> v4Addr oif(iface)");
593
594 ScopedIndent bpfDetailIndent(dw);
Maciej Żenczykowski55262712019-03-29 23:44:56 -0700595 const auto printClatMap = [&dw](const ClatIngressKey& key, const ClatIngressValue& value,
596 const BpfMap<ClatIngressKey, ClatIngressValue>&) {
597 char iifStr[IFNAMSIZ] = "?";
598 char pfx96Str[INET6_ADDRSTRLEN] = "?";
599 char local6Str[INET6_ADDRSTRLEN] = "?";
600 char local4Str[INET_ADDRSTRLEN] = "?";
601 char oifStr[IFNAMSIZ] = "?";
602
603 if_indextoname(key.iif, iifStr);
604 inet_ntop(AF_INET6, &key.pfx96, pfx96Str, sizeof(pfx96Str));
605 inet_ntop(AF_INET6, &key.local6, local6Str, sizeof(local6Str));
606 inet_ntop(AF_INET, &value.local4, local4Str, sizeof(local4Str));
607 if_indextoname(value.oif, oifStr);
608
609 dw.println("%u(%s) %s/96 %s -> %s %u(%s)", key.iif, iifStr, pfx96Str, local6Str, local4Str,
610 value.oif, oifStr);
611 return netdutils::status::ok;
612 };
613 auto res = configMap.iterateWithValue(printClatMap);
614 if (!isOk(res)) {
615 dw.println("Error printing BPF map: %s", res.msg().c_str());
616 }
617}
618
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900619auto ClatdController::isIpv4AddressFreeFunc = isIpv4AddressFree;
Lorenzo Colitti91fd5802019-06-28 19:22:01 +0900620auto ClatdController::iptablesRestoreFunction = execIptablesRestore;
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900621
Lorenzo Colitti7035f222017-02-13 18:29:00 +0900622} // namespace net
623} // namespace android