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