blob: c2b3449ecc134a53b77e1081bdafdcc07b00ecec [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 Colitti7ef8c0f2019-01-11 22:34:58 +090040#include "android-base/unique_fd.h"
Maciej Żenczykowski55262712019-03-29 23:44:56 -070041#include "bpf/BpfMap.h"
42#include "netdbpf/bpf_shared.h"
43#include "netdutils/DumpWriter.h"
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +090044
45extern "C" {
46#include "netutils/checksum.h"
47}
48
Maciej Żenczykowski55262712019-03-29 23:44:56 -070049#include "ClatUtils.h"
Lorenzo Colitti45d3dd02014-06-09 14:09:20 +090050#include "Fwmark.h"
Paul Jensen84c1d032014-05-30 13:29:41 -040051#include "NetdConstants.h"
52#include "NetworkController.h"
Bernie Innocenti189eb502018-10-01 23:10:18 +090053#include "netid_client.h"
Daniel Drown0da73fc2012-06-20 16:51:39 -050054
Lorenzo Colittiac7fefc2014-10-20 17:14:13 +090055static const char* kClatdPath = "/system/bin/clatd";
56
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +090057// For historical reasons, start with 192.0.0.4, and after that, use all subsequent addresses in
58// 192.0.0.0/29 (RFC 7335).
59static const char* kV4AddrString = "192.0.0.4";
60static const in_addr kV4Addr = {inet_addr(kV4AddrString)};
61static const int kV4AddrLen = 29;
62
63using android::base::unique_fd;
Maciej Żenczykowski55262712019-03-29 23:44:56 -070064using android::bpf::BpfMap;
65using android::netdutils::DumpWriter;
66using android::netdutils::ScopedIndent;
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +090067
Lorenzo Colitti7035f222017-02-13 18:29:00 +090068namespace android {
69namespace net {
70
Maciej Żenczykowski56280272019-03-30 03:32:51 -070071void ClatdController::init(void) {
72 std::lock_guard guard(mutex);
73
Maciej Żenczykowski1c086e52019-03-29 23:13:49 -070074 // TODO: should refactor into separate function for testability
75 if (bpf::getBpfSupportLevel() == bpf::BpfLevel::NONE) {
76 ALOGI("Pre-4.9 kernel or pre-P api shipping level - disabling clat ebpf.");
77 mClatEbpfMode = ClatEbpfDisabled;
78 return;
79 }
80
81 // We know the device initially shipped with at least P...,
82 // but did it ship with at least Q?
83
84 uint64_t api_level = base::GetUintProperty<uint64_t>("ro.product.first_api_level", 0);
85 if (api_level == 0) {
86 ALOGE("Cannot determine initial API level of the device.");
87 api_level = base::GetUintProperty<uint64_t>("ro.build.version.sdk", 0);
88 }
89
90 // Note: MINIMUM_API_REQUIRED is for eBPF as a whole and is thus P
91 if (api_level > bpf::MINIMUM_API_REQUIRED) {
92 ALOGI("4.9+ kernel and device shipped with Q+ - clat ebpf should work.");
93 mClatEbpfMode = ClatEbpfEnabled;
94 } else {
95 // We cannot guarantee that 4.9-P kernels will include NET_CLS_BPF support.
96 ALOGI("4.9+ kernel and device shipped with P - clat ebpf might work.");
97 mClatEbpfMode = ClatEbpfMaybe;
98 }
99
100 int rv = openNetlinkSocket();
101 if (rv < 0) {
102 ALOGE("openNetlinkSocket() failure: %s", strerror(-rv));
103 mClatEbpfMode = ClatEbpfDisabled;
104 return;
105 }
106 mNetlinkFd.reset(rv);
107
108 rv = getClatIngressMapFd();
109 if (rv < 0) {
110 ALOGE("getClatIngressMapFd() failure: %s", strerror(-rv));
111 mClatEbpfMode = ClatEbpfDisabled;
112 mNetlinkFd.reset(-1);
113 return;
114 }
115 mClatIngressMap.reset(rv);
116
117 int netlinkFd = mNetlinkFd.get();
118
119 // TODO: perhaps this initial cleanup should be in its own function?
120 const auto del = [&netlinkFd](const ClatIngressKey& key,
121 const BpfMap<ClatIngressKey, ClatIngressValue>&) {
122 ALOGW("Removing stale clat config on interface %d.", key.iif);
123 int rv = tcQdiscDelDevClsact(netlinkFd, key.iif);
124 if (rv < 0) ALOGE("tcQdiscDelDevClsact() failure: %s", strerror(-rv));
125 return netdutils::status::ok; // keep on going regardless
126 };
127 auto ret = mClatIngressMap.iterate(del);
128 if (!isOk(ret)) ALOGE("mClatIngressMap.iterate() failure: %s", strerror(ret.code()));
129 ret = mClatIngressMap.clear();
130 if (!isOk(ret)) ALOGE("mClatIngressMap.clear() failure: %s", strerror(ret.code()));
131}
132
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900133bool ClatdController::isIpv4AddressFree(in_addr_t addr) {
134 int s = socket(AF_INET, SOCK_DGRAM | SOCK_CLOEXEC, 0);
135 if (s == -1) {
136 return 0;
137 }
138
139 // Attempt to connect to the address. If the connection succeeds and getsockname returns the
140 // same then the address is already assigned to the system and we can't use it.
141 struct sockaddr_in sin = {.sin_family = AF_INET, .sin_addr = {addr}, .sin_port = 53};
142 socklen_t len = sizeof(sin);
143 bool inuse = connect(s, (struct sockaddr*)&sin, sizeof(sin)) == 0 &&
144 getsockname(s, (struct sockaddr*)&sin, &len) == 0 && (size_t)len >= sizeof(sin) &&
145 sin.sin_addr.s_addr == addr;
146
147 close(s);
148 return !inuse;
Lorenzo Colittiac7fefc2014-10-20 17:14:13 +0900149}
Daniel Drown0da73fc2012-06-20 16:51:39 -0500150
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900151// Picks a free IPv4 address, starting from ip and trying all addresses in the prefix in order.
152// ip - the IP address from the configuration file
153// prefixlen - the length of the prefix from which addresses may be selected.
154// returns: the IPv4 address, or INADDR_NONE if no addresses were available
155in_addr_t ClatdController::selectIpv4Address(const in_addr ip, int16_t prefixlen) {
156 // Don't accept prefixes that are too large because we scan addresses one by one.
157 if (prefixlen < 16 || prefixlen > 32) {
158 return INADDR_NONE;
159 }
Lorenzo Colittiac7fefc2014-10-20 17:14:13 +0900160
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900161 // All these are in host byte order.
162 in_addr_t mask = 0xffffffff >> (32 - prefixlen) << (32 - prefixlen);
163 in_addr_t ipv4 = ntohl(ip.s_addr);
164 in_addr_t first_ipv4 = ipv4;
165 in_addr_t prefix = ipv4 & mask;
166
167 // Pick the first IPv4 address in the pool, wrapping around if necessary.
168 // So, for example, 192.0.0.4 -> 192.0.0.5 -> 192.0.0.6 -> 192.0.0.7 -> 192.0.0.0.
169 do {
170 if (isIpv4AddressFreeFunc(htonl(ipv4))) {
171 return htonl(ipv4);
172 }
173 ipv4 = prefix | ((ipv4 + 1) & ~mask);
174 } while (ipv4 != first_ipv4);
175
176 return INADDR_NONE;
177}
178
179// Alters the bits in the IPv6 address to make them checksum neutral with v4 and nat64Prefix.
180void ClatdController::makeChecksumNeutral(in6_addr* v6, const in_addr v4,
181 const in6_addr& nat64Prefix) {
182 // Fill last 8 bytes of IPv6 address with random bits.
183 arc4random_buf(&v6->s6_addr[8], 8);
184
185 // Make the IID checksum-neutral. That is, make it so that:
186 // checksum(Local IPv4 | Remote IPv4) = checksum(Local IPv6 | Remote IPv6)
187 // in other words (because remote IPv6 = NAT64 prefix | Remote IPv4):
188 // checksum(Local IPv4) = checksum(Local IPv6 | NAT64 prefix)
189 // Do this by adjusting the two bytes in the middle of the IID.
190
191 uint16_t middlebytes = (v6->s6_addr[11] << 8) + v6->s6_addr[12];
192
193 uint32_t c1 = ip_checksum_add(0, &v4, sizeof(v4));
194 uint32_t c2 = ip_checksum_add(0, &nat64Prefix, sizeof(nat64Prefix)) +
195 ip_checksum_add(0, v6, sizeof(*v6));
196
197 uint16_t delta = ip_checksum_adjust(middlebytes, c1, c2);
198 v6->s6_addr[11] = delta >> 8;
199 v6->s6_addr[12] = delta & 0xff;
200}
201
202// Picks a random interface ID that is checksum neutral with the IPv4 address and the NAT64 prefix.
203int ClatdController::generateIpv6Address(const char* iface, const in_addr v4,
204 const in6_addr& nat64Prefix, in6_addr* v6) {
205 unique_fd s(socket(AF_INET6, SOCK_DGRAM | SOCK_CLOEXEC, 0));
206 if (s == -1) return -errno;
207
208 if (setsockopt(s, SOL_SOCKET, SO_BINDTODEVICE, iface, strlen(iface) + 1) == -1) {
Luke Huang6d301232018-08-01 14:05:18 +0800209 return -errno;
Daniel Drown0da73fc2012-06-20 16:51:39 -0500210 }
211
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900212 sockaddr_in6 sin6 = {.sin6_family = AF_INET6, .sin6_addr = nat64Prefix};
213 if (connect(s, reinterpret_cast<struct sockaddr*>(&sin6), sizeof(sin6)) == -1) {
214 return -errno;
215 }
216
217 socklen_t len = sizeof(sin6);
218 if (getsockname(s, reinterpret_cast<struct sockaddr*>(&sin6), &len) == -1) {
219 return -errno;
220 }
221
222 *v6 = sin6.sin6_addr;
223
224 if (IN6_IS_ADDR_UNSPECIFIED(v6) || IN6_IS_ADDR_LOOPBACK(v6) || IN6_IS_ADDR_LINKLOCAL(v6) ||
225 IN6_IS_ADDR_SITELOCAL(v6) || IN6_IS_ADDR_ULA(v6)) {
226 return -ENETUNREACH;
227 }
228
229 makeChecksumNeutral(v6, v4, nat64Prefix);
230
231 return 0;
232}
233
Maciej Żenczykowski1c086e52019-03-29 23:13:49 -0700234void ClatdController::maybeStartBpf(const ClatdTracker& tracker) {
235 if (mClatEbpfMode == ClatEbpfDisabled) return;
236
237 int rv = hardwareAddressType(tracker.iface);
238 if (rv < 0) {
239 ALOGE("hardwareAddressType(%s[%d]) failure: %s", tracker.iface, tracker.ifIndex,
240 strerror(-rv));
241 return;
242 }
243
244 bool isEthernet;
245 switch (rv) {
246 case ARPHRD_ETHER:
247 isEthernet = true;
248 break;
249 case ARPHRD_RAWIP: // in Linux 4.14+ rmnet support was upstreamed and this is 519
250 case 530: // this is ARPHRD_RAWIP on some Android 4.9 kernels with rmnet
251 isEthernet = false;
252 break;
253 default:
254 ALOGE("hardwareAddressType(%s[%d]) returned unknown type %d.", tracker.iface,
255 tracker.ifIndex, rv);
256 return;
257 }
258
259 rv = getClatIngressProgFd(isEthernet);
260 if (rv < 0) {
261 ALOGE("getClatIngressProgFd(%d) failure: %s", isEthernet, strerror(-rv));
262 return;
263 }
264 unique_fd progFd(rv);
265
266 ClatIngressKey key = {
267 .iif = tracker.ifIndex,
268 .pfx96 = tracker.pfx96,
269 .local6 = tracker.v6,
270 };
271 ClatIngressValue value = {
Maciej Żenczykowski1c086e52019-03-29 23:13:49 -0700272 // TODO: move all the clat code to eBPF and remove the tun interface entirely.
Maciej Żenczykowski39b0b902019-05-08 00:36:30 -0700273 .oif = tracker.v4ifIndex,
Maciej Żenczykowski1c086e52019-03-29 23:13:49 -0700274 .local4 = tracker.v4,
275 };
276
277 auto ret = mClatIngressMap.writeValue(key, value, BPF_ANY);
278 if (!isOk(ret)) {
279 ALOGE("mClatIngress.Map.writeValue failure: %s", strerror(ret.code()));
280 return;
281 }
282
283 // We do tc setup *after* populating map, so scanning through map
284 // can always be used to tell us what needs cleanup.
285
286 rv = tcQdiscAddDevClsact(mNetlinkFd, tracker.ifIndex);
287 if (rv) {
288 ALOGE("tcQdiscAddDevClsact(%d[%s]) failure: %s", tracker.ifIndex, tracker.iface,
289 strerror(-rv));
290 ret = mClatIngressMap.deleteValue(key);
291 if (!isOk(ret)) ALOGE("mClatIngressMap.deleteValue failure: %s", strerror(ret.code()));
292 return;
293 }
294
295 rv = tcFilterAddDevBpf(mNetlinkFd, tracker.ifIndex, progFd, isEthernet);
296 if (rv) {
297 if ((rv == -ENOENT) && (mClatEbpfMode == ClatEbpfMaybe)) {
298 ALOGI("tcFilterAddDevBpf(%d[%s], %d): %s", tracker.ifIndex, tracker.iface, isEthernet,
299 strerror(-rv));
300 } else {
301 ALOGE("tcFilterAddDevBpf(%d[%s], %d) failure: %s", tracker.ifIndex, tracker.iface,
302 isEthernet, strerror(-rv));
303 }
304 rv = tcQdiscDelDevClsact(mNetlinkFd, tracker.ifIndex);
305 if (rv)
306 ALOGE("tcQdiscDelDevClsact(%d[%s]) failure: %s", tracker.ifIndex, tracker.iface,
307 strerror(-rv));
308 ret = mClatIngressMap.deleteValue(key);
309 if (!isOk(ret)) ALOGE("mClatIngressMap.deleteValue failure: %s", strerror(ret.code()));
310 return;
311 }
312
313 // success
314}
315
316void ClatdController::maybeStopBpf(const ClatdTracker& tracker) {
317 if (mClatEbpfMode == ClatEbpfDisabled) return;
318
319 // No need to remove filter, since we remove qdisc it is attached to,
320 // which automatically removes everything attached to the qdisc.
321 int rv = tcQdiscDelDevClsact(mNetlinkFd, tracker.ifIndex);
322 if (rv < 0)
323 ALOGE("tcQdiscDelDevClsact(%d[%s]) failure: %s", tracker.ifIndex, tracker.iface,
324 strerror(-rv));
325
326 // We cleanup map last, so scanning through map can be used to
327 // determine what still needs cleanup.
328
329 ClatIngressKey key = {
330 .iif = tracker.ifIndex,
331 .pfx96 = tracker.pfx96,
332 .local6 = tracker.v6,
333 };
334
335 auto ret = mClatIngressMap.deleteValue(key);
336 if (!isOk(ret)) ALOGE("mClatIngressMap.deleteValue failure: %s", strerror(ret.code()));
337}
338
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900339// Finds the tracker of the clatd running on interface |interface|, or nullptr if clatd has not been
340// started on |interface|.
341ClatdController::ClatdTracker* ClatdController::getClatdTracker(const std::string& interface) {
342 auto it = mClatdTrackers.find(interface);
343 return (it == mClatdTrackers.end() ? nullptr : &it->second);
344}
345
346// Initializes a ClatdTracker for the specified interface.
Maciej Żenczykowskia56b2e62019-04-24 13:17:18 -0700347int ClatdController::ClatdTracker::init(unsigned networkId, const std::string& interface,
Maciej Żenczykowski4657e512019-05-08 06:35:08 -0700348 const std::string& v4interface,
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900349 const std::string& nat64Prefix) {
Maciej Żenczykowskia56b2e62019-04-24 13:17:18 -0700350 netId = networkId;
Lorenzo Colitti32b2e792015-01-07 15:11:30 +0900351
Lorenzo Colitti32b2e792015-01-07 15:11:30 +0900352 fwmark.netId = netId;
353 fwmark.explicitlySelected = true;
354 fwmark.protectedFromVpn = true;
355 fwmark.permission = PERMISSION_SYSTEM;
356
Lorenzo Colitti32b2e792015-01-07 15:11:30 +0900357 snprintf(fwmarkString, sizeof(fwmarkString), "0x%x", fwmark.intValue);
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900358 snprintf(netIdString, sizeof(netIdString), "%u", netId);
359 strlcpy(iface, interface.c_str(), sizeof(iface));
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700360 ifIndex = if_nametoindex(iface);
Maciej Żenczykowski4657e512019-05-08 06:35:08 -0700361 strlcpy(v4iface, v4interface.c_str(), sizeof(v4iface));
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700362 v4ifIndex = if_nametoindex(v4iface);
Lorenzo Colitti32b2e792015-01-07 15:11:30 +0900363
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900364 // Pass in everything that clatd needs: interface, a netid to use for DNS lookups, a fwmark for
365 // outgoing packets, the NAT64 prefix, and the IPv4 and IPv6 addresses.
366 // Validate the prefix and strip off the prefix length.
367 uint8_t family;
368 uint8_t prefixLen;
Maciej Żenczykowski1c06f9c2019-03-29 23:19:19 -0700369 int res = parsePrefix(nat64Prefix.c_str(), &family, &pfx96, sizeof(pfx96), &prefixLen);
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900370 // clatd only supports /96 prefixes.
Maciej Żenczykowski1c06f9c2019-03-29 23:19:19 -0700371 if (res != sizeof(pfx96)) return res;
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900372 if (family != AF_INET6) return -EAFNOSUPPORT;
373 if (prefixLen != 96) return -EINVAL;
Maciej Żenczykowski1c06f9c2019-03-29 23:19:19 -0700374 if (!inet_ntop(AF_INET6, &pfx96, pfx96String, sizeof(pfx96String))) return -errno;
Luke Huang6d301232018-08-01 14:05:18 +0800375
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900376 // Pick an IPv4 address.
377 // TODO: this picks the address based on other addresses that are assigned to interfaces, but
378 // the address is only actually assigned to an interface once clatd starts up. So we could end
379 // up with two clatd instances with the same IPv4 address.
380 // Stop doing this and instead pick a free one from the kV4Addr pool.
Maciej Żenczykowski55cacfb2019-03-30 02:01:35 -0700381 v4 = {selectIpv4Address(kV4Addr, kV4AddrLen)};
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900382 if (v4.s_addr == INADDR_NONE) {
383 ALOGE("No free IPv4 address in %s/%d", kV4AddrString, kV4AddrLen);
384 return -EADDRNOTAVAIL;
385 }
386 if (!inet_ntop(AF_INET, &v4, v4Str, sizeof(v4Str))) return -errno;
387
388 // Generate a checksum-neutral IID.
Maciej Żenczykowski1c06f9c2019-03-29 23:19:19 -0700389 if (generateIpv6Address(iface, v4, pfx96, &v6)) {
390 ALOGE("Unable to find global source address on %s for %s", iface, pfx96String);
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900391 return -EADDRNOTAVAIL;
392 }
393 if (!inet_ntop(AF_INET6, &v6, v6Str, sizeof(v6Str))) return -errno;
394
Maciej Żenczykowski1c06f9c2019-03-29 23:19:19 -0700395 ALOGD("starting clatd on %s v4=%s v6=%s pfx96=%s", iface, v4Str, v6Str, pfx96String);
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900396 return 0;
397}
398
399int ClatdController::startClatd(const std::string& interface, const std::string& nat64Prefix,
400 std::string* v6Str) {
Maciej Żenczykowski56280272019-03-30 03:32:51 -0700401 std::lock_guard guard(mutex);
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700402
403 // 1. fail if pre-existing tracker already exists
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900404 ClatdTracker* existing = getClatdTracker(interface);
405 if (existing != nullptr) {
406 ALOGE("clatd pid=%d already started on %s", existing->pid, interface.c_str());
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700407 return -EBUSY;
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900408 }
409
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700410 // 2. get network id associated with this external interface
Maciej Żenczykowskia56b2e62019-04-24 13:17:18 -0700411 unsigned networkId = mNetCtrl->getNetworkForInterface(interface.c_str());
412 if (networkId == NETID_UNSET) {
413 ALOGE("Interface %s not assigned to any netId", interface.c_str());
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700414 return -ENODEV;
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900415 }
JP Abgrall69261cb2014-06-19 18:35:24 -0700416
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700417 // 3. open the tun device in non blocking mode as required by clatd
418 int res = open("/dev/tun", O_RDWR | O_NONBLOCK | O_CLOEXEC);
419 if (res == -1) {
420 res = errno;
421 ALOGE("open of tun device failed (%s)", strerror(res));
422 return -res;
423 }
424 unique_fd tmpTunFd(res);
425
426 // 4. create the v4-... tun interface
Maciej Żenczykowski4657e512019-05-08 06:35:08 -0700427 std::string v4interface("v4-");
428 v4interface += interface;
429
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700430 struct ifreq ifr = {
431 .ifr_flags = IFF_TUN,
432 };
Maciej Żenczykowski4657e512019-05-08 06:35:08 -0700433 strlcpy(ifr.ifr_name, v4interface.c_str(), sizeof(ifr.ifr_name));
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700434
435 res = ioctl(tmpTunFd, TUNSETIFF, &ifr, sizeof(ifr));
436 if (res == -1) {
437 res = errno;
438 ALOGE("ioctl(TUNSETIFF) failed (%s)", strerror(res));
439 return -res;
440 }
441
Maciej Żenczykowskia1699952019-05-11 17:07:44 -0700442 // disable IPv6 on it - failing to do so is not a critical error
443 res = InterfaceController::setEnableIPv6(v4interface.c_str(), 0);
444 if (res) ALOGE("setEnableIPv6 %s failed (%s)", v4interface.c_str(), strerror(res));
445
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700446 // 5. initialize tracker object
Maciej Żenczykowskia56b2e62019-04-24 13:17:18 -0700447 ClatdTracker tracker;
Maciej Żenczykowski4657e512019-05-08 06:35:08 -0700448 int ret = tracker.init(networkId, interface, v4interface, nat64Prefix);
Maciej Żenczykowskia56b2e62019-04-24 13:17:18 -0700449 if (ret) return ret;
450
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700451 // 6. create a throwaway socket to reserve a file descriptor number
452 res = socket(AF_INET6, SOCK_DGRAM | SOCK_CLOEXEC, 0);
453 if (res == -1) {
454 res = errno;
455 ALOGE("socket(ipv6/udp) failed (%s)", strerror(res));
456 return -res;
457 }
458 unique_fd passedTunFd(res);
459
460 // 7. this is the FD we'll pass to clatd on the cli, so need it as a string
461 char passedTunFdStr[INT32_STRLEN];
462 snprintf(passedTunFdStr, sizeof(passedTunFdStr), "%d", passedTunFd.get());
463
464 // 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 +0900465 std::string progname("clatd-");
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900466 progname += tracker.iface;
Daniel Drown0da73fc2012-06-20 16:51:39 -0500467
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900468 // clang-format off
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700469 const char* args[] = {progname.c_str(),
470 "-i", tracker.iface,
471 "-n", tracker.netIdString,
472 "-m", tracker.fwmarkString,
473 "-p", tracker.pfx96String,
474 "-4", tracker.v4Str,
475 "-6", tracker.v6Str,
476 "-t", passedTunFdStr,
477 nullptr};
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900478 // clang-format on
479
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700480 // 9. register vfork requirement
481 posix_spawnattr_t attr;
482 res = posix_spawnattr_init(&attr);
483 if (res) {
484 ALOGE("posix_spawnattr_init failed (%s)", strerror(res));
485 return -res;
486 }
487 const android::base::ScopeGuard attrGuard = [&] { posix_spawnattr_destroy(&attr); };
488 res = posix_spawnattr_setflags(&attr, POSIX_SPAWN_USEVFORK);
489 if (res) {
490 ALOGE("posix_spawnattr_setflags failed (%s)", strerror(res));
491 return -res;
492 }
493
494 // 10. register dup2() action: this is what 'clears' the CLOEXEC flag
495 // on the tun fd that we want the child clatd process to inherit
496 // (this will happen after the vfork, and before the execve)
497 posix_spawn_file_actions_t fa;
498 res = posix_spawn_file_actions_init(&fa);
499 if (res) {
500 ALOGE("posix_spawn_file_actions_init failed (%s)", strerror(res));
501 return -res;
502 }
503 const android::base::ScopeGuard faGuard = [&] { posix_spawn_file_actions_destroy(&fa); };
504 res = posix_spawn_file_actions_adddup2(&fa, tmpTunFd, passedTunFd);
505 if (res) {
506 ALOGE("posix_spawn_file_actions_adddup2 failed (%s)", strerror(res));
507 return -res;
508 }
509
510 // 11. actually perform vfork/dup2/execve
511 res = posix_spawn(&tracker.pid, kClatdPath, &fa, &attr, (char* const*)args, nullptr);
Luke Huang40962442019-02-26 11:46:10 +0800512 if (res) {
513 ALOGE("posix_spawn failed (%s)", strerror(res));
Luke Huang6d301232018-08-01 14:05:18 +0800514 return -res;
Daniel Drown0da73fc2012-06-20 16:51:39 -0500515 }
516
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700517 // 12. configure eBPF offload - if possible
Maciej Żenczykowski1c086e52019-03-29 23:13:49 -0700518 maybeStartBpf(tracker);
519
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900520 mClatdTrackers[interface] = tracker;
521 ALOGD("clatd started on %s", interface.c_str());
Daniel Drown0da73fc2012-06-20 16:51:39 -0500522
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900523 *v6Str = tracker.v6Str;
Daniel Drown0da73fc2012-06-20 16:51:39 -0500524 return 0;
525}
526
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900527int ClatdController::stopClatd(const std::string& interface) {
Maciej Żenczykowski56280272019-03-30 03:32:51 -0700528 std::lock_guard guard(mutex);
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900529 ClatdTracker* tracker = getClatdTracker(interface);
Lorenzo Colittiac7fefc2014-10-20 17:14:13 +0900530
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900531 if (tracker == nullptr) {
Daniel Drown0da73fc2012-06-20 16:51:39 -0500532 ALOGE("clatd already stopped");
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900533 return -ENODEV;
Daniel Drown0da73fc2012-06-20 16:51:39 -0500534 }
535
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900536 ALOGD("Stopping clatd pid=%d on %s", tracker->pid, interface.c_str());
Daniel Drown0da73fc2012-06-20 16:51:39 -0500537
Maciej Żenczykowski1c086e52019-03-29 23:13:49 -0700538 maybeStopBpf(*tracker);
539
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900540 kill(tracker->pid, SIGTERM);
541 waitpid(tracker->pid, nullptr, 0);
542 mClatdTrackers.erase(interface);
Daniel Drown0da73fc2012-06-20 16:51:39 -0500543
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900544 ALOGD("clatd on %s stopped", interface.c_str());
Daniel Drown0da73fc2012-06-20 16:51:39 -0500545
546 return 0;
547}
548
Maciej Żenczykowski55262712019-03-29 23:44:56 -0700549void ClatdController::dump(DumpWriter& dw) {
550 std::lock_guard guard(mutex);
551
552 ScopedIndent clatdIndent(dw);
553 dw.println("ClatdController");
554
555 {
556 ScopedIndent trackerIndent(dw);
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700557 dw.println("Trackers: iif[iface] nat64Prefix v6Addr -> v4Addr v4iif[v4iface] [netId]");
Maciej Żenczykowski55262712019-03-29 23:44:56 -0700558
559 ScopedIndent trackerDetailIndent(dw);
560 for (const auto& pair : mClatdTrackers) {
561 const ClatdTracker& tracker = pair.second;
Maciej Żenczykowskif4b44fe2019-04-08 16:18:50 -0700562 dw.println("%u[%s] %s/96 %s -> %s %u[%s] [%u]", tracker.ifIndex, tracker.iface,
563 tracker.pfx96String, tracker.v6Str, tracker.v4Str, tracker.v4ifIndex,
564 tracker.v4iface, tracker.netId);
Maciej Żenczykowski55262712019-03-29 23:44:56 -0700565 }
566 }
567
Maciej Żenczykowskiabb2cbf2019-04-01 01:29:29 -0700568 int mapFd = getClatIngressMapFd();
569 if (mapFd < 0) return; // if unsupported just don't dump anything
570 BpfMap<ClatIngressKey, ClatIngressValue> configMap(mapFd);
Maciej Żenczykowski55262712019-03-29 23:44:56 -0700571
572 ScopedIndent bpfIndent(dw);
573 dw.println("BPF ingress map: iif(iface) nat64Prefix v6Addr -> v4Addr oif(iface)");
574
575 ScopedIndent bpfDetailIndent(dw);
Maciej Żenczykowski55262712019-03-29 23:44:56 -0700576 const auto printClatMap = [&dw](const ClatIngressKey& key, const ClatIngressValue& value,
577 const BpfMap<ClatIngressKey, ClatIngressValue>&) {
578 char iifStr[IFNAMSIZ] = "?";
579 char pfx96Str[INET6_ADDRSTRLEN] = "?";
580 char local6Str[INET6_ADDRSTRLEN] = "?";
581 char local4Str[INET_ADDRSTRLEN] = "?";
582 char oifStr[IFNAMSIZ] = "?";
583
584 if_indextoname(key.iif, iifStr);
585 inet_ntop(AF_INET6, &key.pfx96, pfx96Str, sizeof(pfx96Str));
586 inet_ntop(AF_INET6, &key.local6, local6Str, sizeof(local6Str));
587 inet_ntop(AF_INET, &value.local4, local4Str, sizeof(local4Str));
588 if_indextoname(value.oif, oifStr);
589
590 dw.println("%u(%s) %s/96 %s -> %s %u(%s)", key.iif, iifStr, pfx96Str, local6Str, local4Str,
591 value.oif, oifStr);
592 return netdutils::status::ok;
593 };
594 auto res = configMap.iterateWithValue(printClatMap);
595 if (!isOk(res)) {
596 dw.println("Error printing BPF map: %s", res.msg().c_str());
597 }
598}
599
Lorenzo Colitti7ef8c0f2019-01-11 22:34:58 +0900600auto ClatdController::isIpv4AddressFreeFunc = isIpv4AddressFree;
601
Lorenzo Colitti7035f222017-02-13 18:29:00 +0900602} // namespace net
603} // namespace android