RouteController.cpp revision 57947f02c00bb03651e3f9427c880211c689db7f
1/* 2 * Copyright (C) 2014 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 */ 16 17#include "RouteController.h" 18 19#include "Fwmark.h" 20#include "UidRanges.h" 21#include "DummyNetwork.h" 22 23#define LOG_TAG "Netd" 24#include "log/log.h" 25#include "logwrap/logwrap.h" 26#include "netutils/ifc.h" 27#include "resolv_netid.h" 28 29#include <arpa/inet.h> 30#include <fcntl.h> 31#include <linux/fib_rules.h> 32#include <map> 33#include <net/if.h> 34#include <sys/stat.h> 35 36namespace { 37 38// BEGIN CONSTANTS -------------------------------------------------------------------------------- 39 40const uint32_t RULE_PRIORITY_VPN_OVERRIDE_SYSTEM = 10000; 41const uint32_t RULE_PRIORITY_VPN_OVERRIDE_OIF = 10500; 42const uint32_t RULE_PRIORITY_VPN_OUTPUT_TO_LOCAL = 11000; 43const uint32_t RULE_PRIORITY_SECURE_VPN = 12000; 44const uint32_t RULE_PRIORITY_EXPLICIT_NETWORK = 13000; 45const uint32_t RULE_PRIORITY_OUTPUT_INTERFACE = 14000; 46const uint32_t RULE_PRIORITY_LEGACY_SYSTEM = 15000; 47const uint32_t RULE_PRIORITY_LEGACY_NETWORK = 16000; 48const uint32_t RULE_PRIORITY_LOCAL_NETWORK = 17000; 49const uint32_t RULE_PRIORITY_TETHERING = 18000; 50const uint32_t RULE_PRIORITY_IMPLICIT_NETWORK = 19000; 51const uint32_t RULE_PRIORITY_BYPASSABLE_VPN = 20000; 52const uint32_t RULE_PRIORITY_VPN_FALLTHROUGH = 21000; 53const uint32_t RULE_PRIORITY_DEFAULT_NETWORK = 22000; 54const uint32_t RULE_PRIORITY_DIRECTLY_CONNECTED = 23000; 55const uint32_t RULE_PRIORITY_UNREACHABLE = 32000; 56 57const uint32_t ROUTE_TABLE_LOCAL_NETWORK = 97; 58const uint32_t ROUTE_TABLE_LEGACY_NETWORK = 98; 59const uint32_t ROUTE_TABLE_LEGACY_SYSTEM = 99; 60 61const char* const ROUTE_TABLE_NAME_LOCAL_NETWORK = "local_network"; 62const char* const ROUTE_TABLE_NAME_LEGACY_NETWORK = "legacy_network"; 63const char* const ROUTE_TABLE_NAME_LEGACY_SYSTEM = "legacy_system"; 64 65const char* const ROUTE_TABLE_NAME_LOCAL = "local"; 66const char* const ROUTE_TABLE_NAME_MAIN = "main"; 67 68// TODO: These values aren't defined by the Linux kernel, because our UID routing changes are not 69// upstream (yet?), so we can't just pick them up from kernel headers. When (if?) the changes make 70// it upstream, we'll remove this and rely on the kernel header values. For now, add a static assert 71// that will warn us if upstream has given these values some other meaning. 72const uint16_t FRA_UID_START = 18; 73const uint16_t FRA_UID_END = 19; 74static_assert(FRA_UID_START > FRA_MAX, 75 "Android-specific FRA_UID_{START,END} values also assigned in Linux uapi. " 76 "Check that these values match what the kernel does and then update this assertion."); 77 78const uint16_t NETLINK_REQUEST_FLAGS = NLM_F_REQUEST | NLM_F_ACK; 79const uint16_t NETLINK_CREATE_REQUEST_FLAGS = NETLINK_REQUEST_FLAGS | NLM_F_CREATE | NLM_F_EXCL; 80 81const sockaddr_nl NETLINK_ADDRESS = {AF_NETLINK, 0, 0, 0}; 82 83const uint8_t AF_FAMILIES[] = {AF_INET, AF_INET6}; 84 85const char* const IP_VERSIONS[] = {"-4", "-6"}; 86 87const uid_t UID_ROOT = 0; 88const char* const IIF_LOOPBACK = "lo"; 89const char* const IIF_NONE = NULL; 90const char* const OIF_NONE = NULL; 91const bool ACTION_ADD = true; 92const bool ACTION_DEL = false; 93const bool MODIFY_NON_UID_BASED_RULES = true; 94 95const char* const RT_TABLES_PATH = "/data/misc/net/rt_tables"; 96const int RT_TABLES_FLAGS = O_CREAT | O_TRUNC | O_WRONLY | O_NOFOLLOW | O_CLOEXEC; 97const mode_t RT_TABLES_MODE = S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH; // mode 0644, rw-r--r-- 98 99const unsigned ROUTE_FLUSH_ATTEMPTS = 2; 100 101// Avoids "non-constant-expression cannot be narrowed from type 'unsigned int' to 'unsigned short'" 102// warnings when using RTA_LENGTH(x) inside static initializers (even when x is already uint16_t). 103constexpr uint16_t U16_RTA_LENGTH(uint16_t x) { 104 return RTA_LENGTH(x); 105} 106 107// These are practically const, but can't be declared so, because they are used to initialize 108// non-const pointers ("void* iov_base") in iovec arrays. 109rtattr FRATTR_PRIORITY = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_PRIORITY }; 110rtattr FRATTR_TABLE = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_TABLE }; 111rtattr FRATTR_FWMARK = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_FWMARK }; 112rtattr FRATTR_FWMASK = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_FWMASK }; 113rtattr FRATTR_UID_START = { U16_RTA_LENGTH(sizeof(uid_t)), FRA_UID_START }; 114rtattr FRATTR_UID_END = { U16_RTA_LENGTH(sizeof(uid_t)), FRA_UID_END }; 115 116rtattr RTATTR_TABLE = { U16_RTA_LENGTH(sizeof(uint32_t)), RTA_TABLE }; 117rtattr RTATTR_OIF = { U16_RTA_LENGTH(sizeof(uint32_t)), RTA_OIF }; 118 119uint8_t PADDING_BUFFER[RTA_ALIGNTO] = {0, 0, 0, 0}; 120 121// END CONSTANTS ---------------------------------------------------------------------------------- 122 123// No locks needed because RouteController is accessed only from one thread (in CommandListener). 124std::map<std::string, uint32_t> interfaceToTable; 125 126uint32_t getRouteTableForInterface(const char* interface) { 127 uint32_t index = if_nametoindex(interface); 128 if (index) { 129 index += RouteController::ROUTE_TABLE_OFFSET_FROM_INDEX; 130 interfaceToTable[interface] = index; 131 return index; 132 } 133 // If the interface goes away if_nametoindex() will return 0 but we still need to know 134 // the index so we can remove the rules and routes. 135 auto iter = interfaceToTable.find(interface); 136 if (iter == interfaceToTable.end()) { 137 ALOGE("cannot find interface %s", interface); 138 return RT_TABLE_UNSPEC; 139 } 140 return iter->second; 141} 142 143void addTableName(uint32_t table, const std::string& name, std::string* contents) { 144 char tableString[UINT32_STRLEN]; 145 snprintf(tableString, sizeof(tableString), "%u", table); 146 *contents += tableString; 147 *contents += " "; 148 *contents += name; 149 *contents += "\n"; 150} 151 152// Doesn't return success/failure as the file is optional; it's okay if we fail to update it. 153void updateTableNamesFile() { 154 std::string contents; 155 156 addTableName(RT_TABLE_LOCAL, ROUTE_TABLE_NAME_LOCAL, &contents); 157 addTableName(RT_TABLE_MAIN, ROUTE_TABLE_NAME_MAIN, &contents); 158 159 addTableName(ROUTE_TABLE_LOCAL_NETWORK, ROUTE_TABLE_NAME_LOCAL_NETWORK, &contents); 160 addTableName(ROUTE_TABLE_LEGACY_NETWORK, ROUTE_TABLE_NAME_LEGACY_NETWORK, &contents); 161 addTableName(ROUTE_TABLE_LEGACY_SYSTEM, ROUTE_TABLE_NAME_LEGACY_SYSTEM, &contents); 162 163 for (const auto& entry : interfaceToTable) { 164 addTableName(entry.second, entry.first, &contents); 165 } 166 167 int fd = open(RT_TABLES_PATH, RT_TABLES_FLAGS, RT_TABLES_MODE); 168 if (fd == -1) { 169 ALOGE("failed to create %s (%s)", RT_TABLES_PATH, strerror(errno)); 170 return; 171 } 172 // File creation is affected by umask, so make sure the right mode bits are set. 173 if (fchmod(fd, RT_TABLES_MODE) == -1) { 174 ALOGE("failed to set mode 0%o on %s (%s)", RT_TABLES_MODE, RT_TABLES_PATH, strerror(errno)); 175 } 176 ssize_t bytesWritten = write(fd, contents.data(), contents.size()); 177 if (bytesWritten != static_cast<ssize_t>(contents.size())) { 178 ALOGE("failed to write to %s (%zd vs %zu bytes) (%s)", RT_TABLES_PATH, bytesWritten, 179 contents.size(), strerror(errno)); 180 } 181 close(fd); 182} 183 184// Sends a netlink request and expects an ack. 185// |iov| is an array of struct iovec that contains the netlink message payload. 186// The netlink header is generated by this function based on |action| and |flags|. 187// Returns -errno if there was an error or if the kernel reported an error. 188WARN_UNUSED_RESULT int sendNetlinkRequest(uint16_t action, uint16_t flags, iovec* iov, int iovlen) { 189 nlmsghdr nlmsg = { 190 .nlmsg_type = action, 191 .nlmsg_flags = flags, 192 }; 193 iov[0].iov_base = &nlmsg; 194 iov[0].iov_len = sizeof(nlmsg); 195 for (int i = 0; i < iovlen; ++i) { 196 nlmsg.nlmsg_len += iov[i].iov_len; 197 } 198 199 int ret; 200 struct { 201 nlmsghdr msg; 202 nlmsgerr err; 203 } response; 204 205 int sock = socket(AF_NETLINK, SOCK_DGRAM, NETLINK_ROUTE); 206 if (sock != -1 && 207 connect(sock, reinterpret_cast<const sockaddr*>(&NETLINK_ADDRESS), 208 sizeof(NETLINK_ADDRESS)) != -1 && 209 writev(sock, iov, iovlen) != -1 && 210 (ret = recv(sock, &response, sizeof(response), 0)) != -1) { 211 if (ret == sizeof(response)) { 212 ret = response.err.error; // Netlink errors are negative errno. 213 if (ret) { 214 ALOGE("netlink response contains error (%s)", strerror(-ret)); 215 } 216 } else { 217 ALOGE("bad netlink response message size (%d != %zu)", ret, sizeof(response)); 218 ret = -EBADMSG; 219 } 220 } else { 221 ALOGE("netlink socket/connect/writev/recv failed (%s)", strerror(errno)); 222 ret = -errno; 223 } 224 225 if (sock != -1) { 226 close(sock); 227 } 228 229 return ret; 230} 231 232// Returns 0 on success or negative errno on failure. 233int padInterfaceName(const char* input, char* name, size_t* length, uint16_t* padding) { 234 if (!input) { 235 *length = 0; 236 *padding = 0; 237 return 0; 238 } 239 *length = strlcpy(name, input, IFNAMSIZ) + 1; 240 if (*length > IFNAMSIZ) { 241 ALOGE("interface name too long (%zu > %u)", *length, IFNAMSIZ); 242 return -ENAMETOOLONG; 243 } 244 *padding = RTA_SPACE(*length) - RTA_LENGTH(*length); 245 return 0; 246} 247 248// Adds or removes a routing rule for IPv4 and IPv6. 249// 250// + If |table| is non-zero, the rule points at the specified routing table. Otherwise, the rule 251// returns ENETUNREACH. 252// + If |mask| is non-zero, the rule matches the specified fwmark and mask. Otherwise, |fwmark| is 253// ignored. 254// + If |iif| is non-NULL, the rule matches the specified incoming interface. 255// + If |oif| is non-NULL, the rule matches the specified outgoing interface. 256// + If |uidStart| and |uidEnd| are not INVALID_UID, the rule matches packets from UIDs in that 257// range (inclusive). Otherwise, the rule matches packets from all UIDs. 258// 259// Returns 0 on success or negative errno on failure. 260WARN_UNUSED_RESULT int modifyIpRule(uint16_t action, uint32_t priority, uint32_t table, 261 uint32_t fwmark, uint32_t mask, const char* iif, 262 const char* oif, uid_t uidStart, uid_t uidEnd) { 263 // Ensure that if you set a bit in the fwmark, it's not being ignored by the mask. 264 if (fwmark & ~mask) { 265 ALOGE("mask 0x%x does not select all the bits set in fwmark 0x%x", mask, fwmark); 266 return -ERANGE; 267 } 268 269 // Interface names must include exactly one terminating NULL and be properly padded, or older 270 // kernels will refuse to delete rules. 271 char iifName[IFNAMSIZ], oifName[IFNAMSIZ]; 272 size_t iifLength, oifLength; 273 uint16_t iifPadding, oifPadding; 274 if (int ret = padInterfaceName(iif, iifName, &iifLength, &iifPadding)) { 275 return ret; 276 } 277 if (int ret = padInterfaceName(oif, oifName, &oifLength, &oifPadding)) { 278 return ret; 279 } 280 281 // Either both start and end UID must be specified, or neither. 282 if ((uidStart == INVALID_UID) != (uidEnd == INVALID_UID)) { 283 ALOGE("incompatible start and end UIDs (%u vs %u)", uidStart, uidEnd); 284 return -EUSERS; 285 } 286 bool isUidRule = (uidStart != INVALID_UID); 287 288 // Assemble a rule request and put it in an array of iovec structures. 289 fib_rule_hdr rule = { 290 .action = static_cast<uint8_t>(table != RT_TABLE_UNSPEC ? FR_ACT_TO_TBL : 291 FR_ACT_UNREACHABLE), 292 }; 293 294 rtattr fraIifName = { U16_RTA_LENGTH(iifLength), FRA_IIFNAME }; 295 rtattr fraOifName = { U16_RTA_LENGTH(oifLength), FRA_OIFNAME }; 296 297 iovec iov[] = { 298 { NULL, 0 }, 299 { &rule, sizeof(rule) }, 300 { &FRATTR_PRIORITY, sizeof(FRATTR_PRIORITY) }, 301 { &priority, sizeof(priority) }, 302 { &FRATTR_TABLE, table != RT_TABLE_UNSPEC ? sizeof(FRATTR_TABLE) : 0 }, 303 { &table, table != RT_TABLE_UNSPEC ? sizeof(table) : 0 }, 304 { &FRATTR_FWMARK, mask ? sizeof(FRATTR_FWMARK) : 0 }, 305 { &fwmark, mask ? sizeof(fwmark) : 0 }, 306 { &FRATTR_FWMASK, mask ? sizeof(FRATTR_FWMASK) : 0 }, 307 { &mask, mask ? sizeof(mask) : 0 }, 308 { &FRATTR_UID_START, isUidRule ? sizeof(FRATTR_UID_START) : 0 }, 309 { &uidStart, isUidRule ? sizeof(uidStart) : 0 }, 310 { &FRATTR_UID_END, isUidRule ? sizeof(FRATTR_UID_END) : 0 }, 311 { &uidEnd, isUidRule ? sizeof(uidEnd) : 0 }, 312 { &fraIifName, iif != IIF_NONE ? sizeof(fraIifName) : 0 }, 313 { iifName, iifLength }, 314 { PADDING_BUFFER, iifPadding }, 315 { &fraOifName, oif != OIF_NONE ? sizeof(fraOifName) : 0 }, 316 { oifName, oifLength }, 317 { PADDING_BUFFER, oifPadding }, 318 }; 319 320 uint16_t flags = (action == RTM_NEWRULE) ? NETLINK_CREATE_REQUEST_FLAGS : NETLINK_REQUEST_FLAGS; 321 for (size_t i = 0; i < ARRAY_SIZE(AF_FAMILIES); ++i) { 322 rule.family = AF_FAMILIES[i]; 323 if (int ret = sendNetlinkRequest(action, flags, iov, ARRAY_SIZE(iov))) { 324 return ret; 325 } 326 } 327 328 return 0; 329} 330 331WARN_UNUSED_RESULT int modifyIpRule(uint16_t action, uint32_t priority, uint32_t table, 332 uint32_t fwmark, uint32_t mask) { 333 return modifyIpRule(action, priority, table, fwmark, mask, IIF_NONE, OIF_NONE, INVALID_UID, 334 INVALID_UID); 335} 336 337// Adds or deletes an IPv4 or IPv6 route. 338// Returns 0 on success or negative errno on failure. 339WARN_UNUSED_RESULT int modifyIpRoute(uint16_t action, uint32_t table, const char* interface, 340 const char* destination, const char* nexthop) { 341 // At least the destination must be non-null. 342 if (!destination) { 343 ALOGE("null destination"); 344 return -EFAULT; 345 } 346 347 // Parse the prefix. 348 uint8_t rawAddress[sizeof(in6_addr)]; 349 uint8_t family; 350 uint8_t prefixLength; 351 int rawLength = parsePrefix(destination, &family, rawAddress, sizeof(rawAddress), 352 &prefixLength); 353 if (rawLength < 0) { 354 ALOGE("parsePrefix failed for destination %s (%s)", destination, strerror(-rawLength)); 355 return rawLength; 356 } 357 358 if (static_cast<size_t>(rawLength) > sizeof(rawAddress)) { 359 ALOGE("impossible! address too long (%d vs %zu)", rawLength, sizeof(rawAddress)); 360 return -ENOBUFS; // Cannot happen; parsePrefix only supports IPv4 and IPv6. 361 } 362 363 uint8_t type = RTN_UNICAST; 364 uint32_t ifindex; 365 uint8_t rawNexthop[sizeof(in6_addr)]; 366 367 if (nexthop && !strcmp(nexthop, "unreachable")) { 368 type = RTN_UNREACHABLE; 369 // 'interface' is likely non-NULL, as the caller (modifyRoute()) likely used it to lookup 370 // the table number. But it's an error to specify an interface ("dev ...") or a nexthop for 371 // unreachable routes, so nuke them. (IPv6 allows them to be specified; IPv4 doesn't.) 372 interface = OIF_NONE; 373 nexthop = NULL; 374 } else if (nexthop && !strcmp(nexthop, "throw")) { 375 type = RTN_THROW; 376 interface = OIF_NONE; 377 nexthop = NULL; 378 } else { 379 // If an interface was specified, find the ifindex. 380 if (interface != OIF_NONE) { 381 ifindex = if_nametoindex(interface); 382 if (!ifindex) { 383 ALOGE("cannot find interface %s", interface); 384 return -ENODEV; 385 } 386 } 387 388 // If a nexthop was specified, parse it as the same family as the prefix. 389 if (nexthop && inet_pton(family, nexthop, rawNexthop) <= 0) { 390 ALOGE("inet_pton failed for nexthop %s", nexthop); 391 return -EINVAL; 392 } 393 } 394 395 // Assemble a rtmsg and put it in an array of iovec structures. 396 rtmsg route = { 397 .rtm_protocol = RTPROT_STATIC, 398 .rtm_type = type, 399 .rtm_family = family, 400 .rtm_dst_len = prefixLength, 401 .rtm_scope = static_cast<uint8_t>(nexthop ? RT_SCOPE_UNIVERSE : RT_SCOPE_LINK), 402 }; 403 404 rtattr rtaDst = { U16_RTA_LENGTH(rawLength), RTA_DST }; 405 rtattr rtaGateway = { U16_RTA_LENGTH(rawLength), RTA_GATEWAY }; 406 407 iovec iov[] = { 408 { NULL, 0 }, 409 { &route, sizeof(route) }, 410 { &RTATTR_TABLE, sizeof(RTATTR_TABLE) }, 411 { &table, sizeof(table) }, 412 { &rtaDst, sizeof(rtaDst) }, 413 { rawAddress, static_cast<size_t>(rawLength) }, 414 { &RTATTR_OIF, interface != OIF_NONE ? sizeof(RTATTR_OIF) : 0 }, 415 { &ifindex, interface != OIF_NONE ? sizeof(ifindex) : 0 }, 416 { &rtaGateway, nexthop ? sizeof(rtaGateway) : 0 }, 417 { rawNexthop, nexthop ? static_cast<size_t>(rawLength) : 0 }, 418 }; 419 420 uint16_t flags = (action == RTM_NEWROUTE) ? NETLINK_CREATE_REQUEST_FLAGS : 421 NETLINK_REQUEST_FLAGS; 422 return sendNetlinkRequest(action, flags, iov, ARRAY_SIZE(iov)); 423} 424 425// An iptables rule to mark incoming packets on a network with the netId of the network. 426// 427// This is so that the kernel can: 428// + Use the right fwmark for (and thus correctly route) replies (e.g.: TCP RST, ICMP errors, ping 429// replies, SYN-ACKs, etc). 430// + Mark sockets that accept connections from this interface so that the connection stays on the 431// same interface. 432WARN_UNUSED_RESULT int modifyIncomingPacketMark(unsigned netId, const char* interface, 433 Permission permission, bool add) { 434 Fwmark fwmark; 435 436 fwmark.netId = netId; 437 fwmark.explicitlySelected = true; 438 fwmark.protectedFromVpn = true; 439 fwmark.permission = permission; 440 441 char markString[UINT32_HEX_STRLEN]; 442 snprintf(markString, sizeof(markString), "0x%x", fwmark.intValue); 443 444 if (execIptables(V4V6, "-t", "mangle", add ? "-A" : "-D", "INPUT", "-i", interface, "-j", 445 "MARK", "--set-mark", markString, NULL)) { 446 ALOGE("failed to change iptables rule that sets incoming packet mark"); 447 return -EREMOTEIO; 448 } 449 450 return 0; 451} 452 453// A rule to route responses to the local network forwarded via the VPN. 454// 455// When a VPN is in effect, packets from the local network to upstream networks are forwarded into 456// the VPN's tunnel interface. When the VPN forwards the responses, they emerge out of the tunnel. 457WARN_UNUSED_RESULT int modifyVpnOutputToLocalRule(const char* vpnInterface, bool add) { 458 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_VPN_OUTPUT_TO_LOCAL, 459 ROUTE_TABLE_LOCAL_NETWORK, MARK_UNSET, MARK_UNSET, vpnInterface, OIF_NONE, 460 INVALID_UID, INVALID_UID); 461} 462 463// A rule to route all traffic from a given set of UIDs to go over the VPN. 464// 465// Notice that this rule doesn't use the netId. I.e., no matter what netId the user's socket may 466// have, if they are subject to this VPN, their traffic has to go through it. Allows the traffic to 467// bypass the VPN if the protectedFromVpn bit is set. 468WARN_UNUSED_RESULT int modifyVpnUidRangeRule(uint32_t table, uid_t uidStart, uid_t uidEnd, 469 bool secure, bool add) { 470 Fwmark fwmark; 471 Fwmark mask; 472 473 fwmark.protectedFromVpn = false; 474 mask.protectedFromVpn = true; 475 476 uint32_t priority; 477 478 if (secure) { 479 priority = RULE_PRIORITY_SECURE_VPN; 480 } else { 481 priority = RULE_PRIORITY_BYPASSABLE_VPN; 482 483 fwmark.explicitlySelected = false; 484 mask.explicitlySelected = true; 485 } 486 487 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, priority, table, fwmark.intValue, 488 mask.intValue, IIF_LOOPBACK, OIF_NONE, uidStart, uidEnd); 489} 490 491// A rule to allow system apps to send traffic over this VPN even if they are not part of the target 492// set of UIDs. 493// 494// This is needed for DnsProxyListener to correctly resolve a request for a user who is in the 495// target set, but where the DnsProxyListener itself is not. 496WARN_UNUSED_RESULT int modifyVpnSystemPermissionRule(unsigned netId, uint32_t table, bool secure, 497 bool add) { 498 Fwmark fwmark; 499 Fwmark mask; 500 501 fwmark.netId = netId; 502 mask.netId = FWMARK_NET_ID_MASK; 503 504 fwmark.permission = PERMISSION_SYSTEM; 505 mask.permission = PERMISSION_SYSTEM; 506 507 uint32_t priority = secure ? RULE_PRIORITY_SECURE_VPN : RULE_PRIORITY_BYPASSABLE_VPN; 508 509 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, priority, table, fwmark.intValue, 510 mask.intValue); 511} 512 513// A rule to route traffic based on an explicitly chosen network. 514// 515// Supports apps that use the multinetwork APIs to restrict their traffic to a network. 516// 517// Even though we check permissions at the time we set a netId into the fwmark of a socket, we need 518// to check it again in the rules here, because a network's permissions may have been updated via 519// modifyNetworkPermission(). 520WARN_UNUSED_RESULT int modifyExplicitNetworkRule(unsigned netId, uint32_t table, 521 Permission permission, uid_t uidStart, 522 uid_t uidEnd, bool add) { 523 Fwmark fwmark; 524 Fwmark mask; 525 526 fwmark.netId = netId; 527 mask.netId = FWMARK_NET_ID_MASK; 528 529 fwmark.explicitlySelected = true; 530 mask.explicitlySelected = true; 531 532 fwmark.permission = permission; 533 mask.permission = permission; 534 535 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_EXPLICIT_NETWORK, table, 536 fwmark.intValue, mask.intValue, IIF_NONE, OIF_NONE, uidStart, uidEnd); 537} 538 539// A rule to route traffic based on a chosen outgoing interface. 540// 541// Supports apps that use SO_BINDTODEVICE or IP_PKTINFO options and the kernel that already knows 542// the outgoing interface (typically for link-local communications). 543WARN_UNUSED_RESULT int modifyOutputInterfaceRules(const char* interface, uint32_t table, 544 Permission permission, uid_t uidStart, 545 uid_t uidEnd, bool add) { 546 Fwmark fwmark; 547 Fwmark mask; 548 549 fwmark.permission = permission; 550 mask.permission = permission; 551 552 // If this rule does not specify a UID range, then also add a corresponding high-priority rule 553 // for UID. This covers forwarded packets and system daemons such as the tethering DHCP server. 554 if (uidStart == INVALID_UID && uidEnd == INVALID_UID) { 555 if (int ret = modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_VPN_OVERRIDE_OIF, 556 table, fwmark.intValue, mask.intValue, IIF_NONE, interface, 557 UID_ROOT, UID_ROOT)) { 558 return ret; 559 } 560 } 561 562 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_OUTPUT_INTERFACE, table, 563 fwmark.intValue, mask.intValue, IIF_NONE, interface, uidStart, uidEnd); 564} 565 566// A rule to route traffic based on the chosen network. 567// 568// This is for sockets that have not explicitly requested a particular network, but have been 569// bound to one when they called connect(). This ensures that sockets connected on a particular 570// network stay on that network even if the default network changes. 571WARN_UNUSED_RESULT int modifyImplicitNetworkRule(unsigned netId, uint32_t table, 572 Permission permission, bool add) { 573 Fwmark fwmark; 574 Fwmark mask; 575 576 fwmark.netId = netId; 577 mask.netId = FWMARK_NET_ID_MASK; 578 579 fwmark.explicitlySelected = false; 580 mask.explicitlySelected = true; 581 582 fwmark.permission = permission; 583 mask.permission = permission; 584 585 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_IMPLICIT_NETWORK, table, 586 fwmark.intValue, mask.intValue); 587} 588 589// A rule to enable split tunnel VPNs. 590// 591// If a packet with a VPN's netId doesn't find a route in the VPN's routing table, it's allowed to 592// go over the default network, provided it wasn't explicitly restricted to the VPN and has the 593// permissions required by the default network. 594WARN_UNUSED_RESULT int modifyVpnFallthroughRule(uint16_t action, unsigned vpnNetId, 595 const char* physicalInterface, 596 Permission permission) { 597 uint32_t table = getRouteTableForInterface(physicalInterface); 598 if (table == RT_TABLE_UNSPEC) { 599 return -ESRCH; 600 } 601 602 Fwmark fwmark; 603 Fwmark mask; 604 605 fwmark.netId = vpnNetId; 606 mask.netId = FWMARK_NET_ID_MASK; 607 608 fwmark.explicitlySelected = false; 609 mask.explicitlySelected = true; 610 611 fwmark.permission = permission; 612 mask.permission = permission; 613 614 return modifyIpRule(action, RULE_PRIORITY_VPN_FALLTHROUGH, table, fwmark.intValue, 615 mask.intValue); 616} 617 618// Add rules to allow legacy routes added through the requestRouteToHost() API. 619WARN_UNUSED_RESULT int addLegacyRouteRules() { 620 Fwmark fwmark; 621 Fwmark mask; 622 623 fwmark.explicitlySelected = false; 624 mask.explicitlySelected = true; 625 626 // Rules to allow legacy routes to override the default network. 627 if (int ret = modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LEGACY_SYSTEM, ROUTE_TABLE_LEGACY_SYSTEM, 628 fwmark.intValue, mask.intValue)) { 629 return ret; 630 } 631 if (int ret = modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LEGACY_NETWORK, 632 ROUTE_TABLE_LEGACY_NETWORK, fwmark.intValue, mask.intValue)) { 633 return ret; 634 } 635 636 fwmark.permission = PERMISSION_SYSTEM; 637 mask.permission = PERMISSION_SYSTEM; 638 639 // A rule to allow legacy routes from system apps to override VPNs. 640 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_VPN_OVERRIDE_SYSTEM, ROUTE_TABLE_LEGACY_SYSTEM, 641 fwmark.intValue, mask.intValue); 642} 643 644// Add rules to lookup the local network when specified explicitly or otherwise. 645WARN_UNUSED_RESULT int addLocalNetworkRules(unsigned localNetId) { 646 if (int ret = modifyExplicitNetworkRule(localNetId, ROUTE_TABLE_LOCAL_NETWORK, PERMISSION_NONE, 647 INVALID_UID, INVALID_UID, ACTION_ADD)) { 648 return ret; 649 } 650 651 Fwmark fwmark; 652 Fwmark mask; 653 654 fwmark.explicitlySelected = false; 655 mask.explicitlySelected = true; 656 657 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LOCAL_NETWORK, ROUTE_TABLE_LOCAL_NETWORK, 658 fwmark.intValue, mask.intValue); 659} 660 661int configureDummyNetwork() { 662 const char *interface = DummyNetwork::INTERFACE_NAME; 663 uint32_t table = getRouteTableForInterface(interface); 664 if (table == RT_TABLE_UNSPEC) { 665 // getRouteTableForInterface has already looged an error. 666 return -ESRCH; 667 } 668 669 ifc_init(); 670 int ret = ifc_up(interface); 671 ifc_close(); 672 if (ret) { 673 ALOGE("Can't bring up %s: %s", interface, strerror(errno)); 674 return -errno; 675 } 676 677 if ((ret = modifyOutputInterfaceRules(interface, table, PERMISSION_NONE, 678 INVALID_UID, INVALID_UID, ACTION_ADD))) { 679 ALOGE("Can't create oif rules for %s: %s", interface, strerror(-ret)); 680 return ret; 681 } 682 683 if ((ret = modifyIpRoute(RTM_NEWROUTE, table, interface, "0.0.0.0/0", NULL))) { 684 ALOGE("Can't add IPv4 default route to %s: %s", interface, strerror(-ret)); 685 return ret; 686 } 687 688 if ((ret = modifyIpRoute(RTM_NEWROUTE, table, interface, "::/0", NULL))) { 689 ALOGE("Can't add IPv6 default route to %s: %s", interface, strerror(-ret)); 690 return ret; 691 } 692 693 return 0; 694} 695 696// Add a new rule to look up the 'main' table, with the same selectors as the "default network" 697// rule, but with a lower priority. We will never create routes in the main table; it should only be 698// used for directly-connected routes implicitly created by the kernel when adding IP addresses. 699// This is necessary, for example, when adding a route through a directly-connected gateway: in 700// order to add the route, there must already be a directly-connected route that covers the gateway. 701WARN_UNUSED_RESULT int addDirectlyConnectedRule() { 702 Fwmark fwmark; 703 Fwmark mask; 704 705 fwmark.netId = NETID_UNSET; 706 mask.netId = FWMARK_NET_ID_MASK; 707 708 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_DIRECTLY_CONNECTED, RT_TABLE_MAIN, 709 fwmark.intValue, mask.intValue, IIF_NONE, OIF_NONE, UID_ROOT, UID_ROOT); 710} 711 712// Add an explicit unreachable rule close to the end of the prioriy list to make it clear that 713// relying on the kernel-default "from all lookup main" rule at priority 32766 is not intended 714// behaviour. We do flush the kernel-default rules at startup, but having an explicit unreachable 715// rule will hopefully make things even clearer. 716WARN_UNUSED_RESULT int addUnreachableRule() { 717 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_UNREACHABLE, RT_TABLE_UNSPEC, MARK_UNSET, 718 MARK_UNSET); 719} 720 721WARN_UNUSED_RESULT int modifyLocalNetwork(unsigned netId, const char* interface, bool add) { 722 if (int ret = modifyIncomingPacketMark(netId, interface, PERMISSION_NONE, add)) { 723 return ret; 724 } 725 return modifyOutputInterfaceRules(interface, ROUTE_TABLE_LOCAL_NETWORK, PERMISSION_NONE, 726 INVALID_UID, INVALID_UID, add); 727} 728 729WARN_UNUSED_RESULT int modifyPhysicalNetwork(unsigned netId, const char* interface, 730 Permission permission, bool add) { 731 uint32_t table = getRouteTableForInterface(interface); 732 if (table == RT_TABLE_UNSPEC) { 733 return -ESRCH; 734 } 735 736 if (int ret = modifyIncomingPacketMark(netId, interface, permission, add)) { 737 return ret; 738 } 739 if (int ret = modifyExplicitNetworkRule(netId, table, permission, INVALID_UID, INVALID_UID, 740 add)) { 741 return ret; 742 } 743 if (int ret = modifyOutputInterfaceRules(interface, table, permission, INVALID_UID, INVALID_UID, 744 add)) { 745 return ret; 746 } 747 return modifyImplicitNetworkRule(netId, table, permission, add); 748} 749 750WARN_UNUSED_RESULT int modifyVirtualNetwork(unsigned netId, const char* interface, 751 const UidRanges& uidRanges, bool secure, bool add, 752 bool modifyNonUidBasedRules) { 753 uint32_t table = getRouteTableForInterface(interface); 754 if (table == RT_TABLE_UNSPEC) { 755 return -ESRCH; 756 } 757 758 for (const UidRanges::Range& range : uidRanges.getRanges()) { 759 if (int ret = modifyVpnUidRangeRule(table, range.first, range.second, secure, add)) { 760 return ret; 761 } 762 if (int ret = modifyExplicitNetworkRule(netId, table, PERMISSION_NONE, range.first, 763 range.second, add)) { 764 return ret; 765 } 766 if (int ret = modifyOutputInterfaceRules(interface, table, PERMISSION_NONE, range.first, 767 range.second, add)) { 768 return ret; 769 } 770 } 771 772 if (modifyNonUidBasedRules) { 773 if (int ret = modifyIncomingPacketMark(netId, interface, PERMISSION_NONE, add)) { 774 return ret; 775 } 776 if (int ret = modifyVpnOutputToLocalRule(interface, add)) { 777 return ret; 778 } 779 if (int ret = modifyVpnSystemPermissionRule(netId, table, secure, add)) { 780 return ret; 781 } 782 return modifyExplicitNetworkRule(netId, table, PERMISSION_NONE, UID_ROOT, UID_ROOT, add); 783 } 784 785 return 0; 786} 787 788WARN_UNUSED_RESULT int modifyDefaultNetwork(uint16_t action, const char* interface, 789 Permission permission) { 790 uint32_t table = getRouteTableForInterface(interface); 791 if (table == RT_TABLE_UNSPEC) { 792 return -ESRCH; 793 } 794 795 Fwmark fwmark; 796 Fwmark mask; 797 798 fwmark.netId = NETID_UNSET; 799 mask.netId = FWMARK_NET_ID_MASK; 800 801 fwmark.permission = permission; 802 mask.permission = permission; 803 804 return modifyIpRule(action, RULE_PRIORITY_DEFAULT_NETWORK, table, fwmark.intValue, 805 mask.intValue); 806} 807 808WARN_UNUSED_RESULT int modifyTetheredNetwork(uint16_t action, const char* inputInterface, 809 const char* outputInterface) { 810 uint32_t table = getRouteTableForInterface(outputInterface); 811 if (table == RT_TABLE_UNSPEC) { 812 return -ESRCH; 813 } 814 815 return modifyIpRule(action, RULE_PRIORITY_TETHERING, table, MARK_UNSET, MARK_UNSET, 816 inputInterface, OIF_NONE, INVALID_UID, INVALID_UID); 817} 818 819// Returns 0 on success or negative errno on failure. 820WARN_UNUSED_RESULT int flushRules() { 821 for (size_t i = 0; i < ARRAY_SIZE(IP_VERSIONS); ++i) { 822 const char* argv[] = { 823 IP_PATH, 824 IP_VERSIONS[i], 825 "rule", 826 "flush", 827 }; 828 if (android_fork_execvp(ARRAY_SIZE(argv), const_cast<char**>(argv), NULL, false, false)) { 829 ALOGE("failed to flush rules"); 830 return -EREMOTEIO; 831 } 832 } 833 return 0; 834} 835 836// Adds or removes an IPv4 or IPv6 route to the specified table and, if it's a directly-connected 837// route, to the main table as well. 838// Returns 0 on success or negative errno on failure. 839WARN_UNUSED_RESULT int modifyRoute(uint16_t action, const char* interface, const char* destination, 840 const char* nexthop, RouteController::TableType tableType) { 841 uint32_t table; 842 switch (tableType) { 843 case RouteController::INTERFACE: { 844 table = getRouteTableForInterface(interface); 845 if (table == RT_TABLE_UNSPEC) { 846 return -ESRCH; 847 } 848 break; 849 } 850 case RouteController::LOCAL_NETWORK: { 851 table = ROUTE_TABLE_LOCAL_NETWORK; 852 break; 853 } 854 case RouteController::LEGACY_NETWORK: { 855 table = ROUTE_TABLE_LEGACY_NETWORK; 856 break; 857 } 858 case RouteController::LEGACY_SYSTEM: { 859 table = ROUTE_TABLE_LEGACY_SYSTEM; 860 break; 861 } 862 } 863 864 int ret = modifyIpRoute(action, table, interface, destination, nexthop); 865 // Trying to add a route that already exists shouldn't cause an error. 866 if (ret && !(action == RTM_NEWROUTE && ret == -EEXIST)) { 867 return ret; 868 } 869 870 return 0; 871} 872 873// Returns 0 on success or negative errno on failure. 874WARN_UNUSED_RESULT int flushRoutes(const char* interface) { 875 uint32_t table = getRouteTableForInterface(interface); 876 if (table == RT_TABLE_UNSPEC) { 877 return -ESRCH; 878 } 879 880 char tableString[UINT32_STRLEN]; 881 snprintf(tableString, sizeof(tableString), "%u", table); 882 883 int ret = 0; 884 for (size_t i = 0; i < ARRAY_SIZE(IP_VERSIONS); ++i) { 885 const char* argv[] = { 886 IP_PATH, 887 IP_VERSIONS[i], 888 "route", 889 "flush", 890 "table", 891 tableString, 892 }; 893 894 // A flush works by dumping routes and deleting each route as it's returned, and it can 895 // fail if something else deletes the route between the dump and the delete. This can 896 // happen, for example, if an interface goes down while we're trying to flush its routes. 897 // So try multiple times and only return an error if the last attempt fails. 898 // 899 // TODO: replace this with our own netlink code. 900 unsigned attempts = 0; 901 int err; 902 do { 903 err = android_fork_execvp(ARRAY_SIZE(argv), const_cast<char**>(argv), 904 NULL, false, false); 905 ++attempts; 906 } while (err != 0 && attempts < ROUTE_FLUSH_ATTEMPTS); 907 if (err) { 908 ALOGE("failed to flush %s routes in table %s after %d attempts", 909 IP_VERSIONS[i], tableString, attempts); 910 ret = -EREMOTEIO; 911 } 912 } 913 914 // If we failed to flush routes, the caller may elect to keep this interface around, so keep 915 // track of its name. 916 if (!ret) { 917 interfaceToTable.erase(interface); 918 } 919 920 return ret; 921} 922 923} // namespace 924 925int RouteController::Init(unsigned localNetId) { 926 if (int ret = flushRules()) { 927 return ret; 928 } 929 if (int ret = addLegacyRouteRules()) { 930 return ret; 931 } 932 if (int ret = addLocalNetworkRules(localNetId)) { 933 return ret; 934 } 935 if (int ret = addDirectlyConnectedRule()) { 936 return ret; 937 } 938 if (int ret = addUnreachableRule()) { 939 return ret; 940 } 941 // Don't complain if we can't add the dummy network, since not all devices support it. 942 configureDummyNetwork(); 943 944 updateTableNamesFile(); 945 return 0; 946} 947 948int RouteController::addInterfaceToLocalNetwork(unsigned netId, const char* interface) { 949 return modifyLocalNetwork(netId, interface, ACTION_ADD); 950} 951 952int RouteController::removeInterfaceFromLocalNetwork(unsigned netId, const char* interface) { 953 return modifyLocalNetwork(netId, interface, ACTION_DEL); 954} 955 956int RouteController::addInterfaceToPhysicalNetwork(unsigned netId, const char* interface, 957 Permission permission) { 958 if (int ret = modifyPhysicalNetwork(netId, interface, permission, ACTION_ADD)) { 959 return ret; 960 } 961 updateTableNamesFile(); 962 return 0; 963} 964 965int RouteController::removeInterfaceFromPhysicalNetwork(unsigned netId, const char* interface, 966 Permission permission) { 967 if (int ret = modifyPhysicalNetwork(netId, interface, permission, ACTION_DEL)) { 968 return ret; 969 } 970 if (int ret = flushRoutes(interface)) { 971 return ret; 972 } 973 updateTableNamesFile(); 974 return 0; 975} 976 977int RouteController::addInterfaceToVirtualNetwork(unsigned netId, const char* interface, 978 bool secure, const UidRanges& uidRanges) { 979 if (int ret = modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_ADD, 980 MODIFY_NON_UID_BASED_RULES)) { 981 return ret; 982 } 983 updateTableNamesFile(); 984 return 0; 985} 986 987int RouteController::removeInterfaceFromVirtualNetwork(unsigned netId, const char* interface, 988 bool secure, const UidRanges& uidRanges) { 989 if (int ret = modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_DEL, 990 MODIFY_NON_UID_BASED_RULES)) { 991 return ret; 992 } 993 if (int ret = flushRoutes(interface)) { 994 return ret; 995 } 996 updateTableNamesFile(); 997 return 0; 998} 999 1000int RouteController::modifyPhysicalNetworkPermission(unsigned netId, const char* interface, 1001 Permission oldPermission, 1002 Permission newPermission) { 1003 // Add the new rules before deleting the old ones, to avoid race conditions. 1004 if (int ret = modifyPhysicalNetwork(netId, interface, newPermission, ACTION_ADD)) { 1005 return ret; 1006 } 1007 return modifyPhysicalNetwork(netId, interface, oldPermission, ACTION_DEL); 1008} 1009 1010int RouteController::addUsersToVirtualNetwork(unsigned netId, const char* interface, bool secure, 1011 const UidRanges& uidRanges) { 1012 return modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_ADD, 1013 !MODIFY_NON_UID_BASED_RULES); 1014} 1015 1016int RouteController::removeUsersFromVirtualNetwork(unsigned netId, const char* interface, 1017 bool secure, const UidRanges& uidRanges) { 1018 return modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_DEL, 1019 !MODIFY_NON_UID_BASED_RULES); 1020} 1021 1022int RouteController::addInterfaceToDefaultNetwork(const char* interface, Permission permission) { 1023 return modifyDefaultNetwork(RTM_NEWRULE, interface, permission); 1024} 1025 1026int RouteController::removeInterfaceFromDefaultNetwork(const char* interface, 1027 Permission permission) { 1028 return modifyDefaultNetwork(RTM_DELRULE, interface, permission); 1029} 1030 1031int RouteController::addRoute(const char* interface, const char* destination, const char* nexthop, 1032 TableType tableType) { 1033 return modifyRoute(RTM_NEWROUTE, interface, destination, nexthop, tableType); 1034} 1035 1036int RouteController::removeRoute(const char* interface, const char* destination, 1037 const char* nexthop, TableType tableType) { 1038 return modifyRoute(RTM_DELROUTE, interface, destination, nexthop, tableType); 1039} 1040 1041int RouteController::enableTethering(const char* inputInterface, const char* outputInterface) { 1042 return modifyTetheredNetwork(RTM_NEWRULE, inputInterface, outputInterface); 1043} 1044 1045int RouteController::disableTethering(const char* inputInterface, const char* outputInterface) { 1046 return modifyTetheredNetwork(RTM_DELRULE, inputInterface, outputInterface); 1047} 1048 1049int RouteController::addVirtualNetworkFallthrough(unsigned vpnNetId, const char* physicalInterface, 1050 Permission permission) { 1051 return modifyVpnFallthroughRule(RTM_NEWRULE, vpnNetId, physicalInterface, permission); 1052} 1053 1054int RouteController::removeVirtualNetworkFallthrough(unsigned vpnNetId, 1055 const char* physicalInterface, 1056 Permission permission) { 1057 return modifyVpnFallthroughRule(RTM_DELRULE, vpnNetId, physicalInterface, permission); 1058} 1059