RouteController.cpp revision bd37832f1843ed78f64604e5627cf952ac9614ba
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 <arpa/inet.h> 20#include <errno.h> 21#include <fcntl.h> 22#include <linux/fib_rules.h> 23#include <net/if.h> 24#include <sys/stat.h> 25 26#include <private/android_filesystem_config.h> 27 28#include <map> 29 30#include "Fwmark.h" 31#include "UidRanges.h" 32 33#define LOG_TAG "Netd" 34#include "log/log.h" 35#include "logwrap/logwrap.h" 36#include "utils/file.h" 37#include "resolv_netid.h" 38 39namespace { 40 41// BEGIN CONSTANTS -------------------------------------------------------------------------------- 42 43const uint32_t RULE_PRIORITY_VPN_OVERRIDE_SYSTEM = 10000; 44const uint32_t RULE_PRIORITY_VPN_OUTPUT_TO_LOCAL = 11000; 45const uint32_t RULE_PRIORITY_SECURE_VPN = 12000; 46const uint32_t RULE_PRIORITY_EXPLICIT_NETWORK = 13000; 47const uint32_t RULE_PRIORITY_OUTPUT_INTERFACE = 14000; 48const uint32_t RULE_PRIORITY_LEGACY_SYSTEM = 15000; 49const uint32_t RULE_PRIORITY_LEGACY_NETWORK = 16000; 50const uint32_t RULE_PRIORITY_LOCAL_NETWORK = 17000; 51const uint32_t RULE_PRIORITY_TETHERING = 18000; 52const uint32_t RULE_PRIORITY_IMPLICIT_NETWORK = 19000; 53const uint32_t RULE_PRIORITY_BYPASSABLE_VPN = 20000; 54const uint32_t RULE_PRIORITY_VPN_FALLTHROUGH = 21000; 55const uint32_t RULE_PRIORITY_DEFAULT_NETWORK = 22000; 56const uint32_t RULE_PRIORITY_DIRECTLY_CONNECTED = 23000; 57const uint32_t RULE_PRIORITY_UNREACHABLE = 32000; 58 59const uint32_t ROUTE_TABLE_LOCAL_NETWORK = 97; 60const uint32_t ROUTE_TABLE_LEGACY_NETWORK = 98; 61const uint32_t ROUTE_TABLE_LEGACY_SYSTEM = 99; 62 63const char* const ROUTE_TABLE_NAME_LOCAL_NETWORK = "local_network"; 64const char* const ROUTE_TABLE_NAME_LEGACY_NETWORK = "legacy_network"; 65const char* const ROUTE_TABLE_NAME_LEGACY_SYSTEM = "legacy_system"; 66 67const char* const ROUTE_TABLE_NAME_LOCAL = "local"; 68const char* const ROUTE_TABLE_NAME_MAIN = "main"; 69 70// TODO: These values aren't defined by the Linux kernel, because our UID routing changes are not 71// upstream (yet?), so we can't just pick them up from kernel headers. When (if?) the changes make 72// it upstream, we'll remove this and rely on the kernel header values. For now, add a static assert 73// that will warn us if upstream has given these values some other meaning. 74const uint16_t FRA_UID_START = 18; 75const uint16_t FRA_UID_END = 19; 76static_assert(FRA_UID_START > FRA_MAX, 77 "Android-specific FRA_UID_{START,END} values also assigned in Linux uapi. " 78 "Check that these values match what the kernel does and then update this assertion."); 79 80const uint16_t NETLINK_REQUEST_FLAGS = NLM_F_REQUEST | NLM_F_ACK; 81const uint16_t NETLINK_CREATE_REQUEST_FLAGS = NETLINK_REQUEST_FLAGS | NLM_F_CREATE | NLM_F_EXCL; 82 83const sockaddr_nl NETLINK_ADDRESS = {AF_NETLINK, 0, 0, 0}; 84 85const uint8_t AF_FAMILIES[] = {AF_INET, AF_INET6}; 86 87const char* const IP_VERSIONS[] = {"-4", "-6"}; 88 89const uid_t UID_ROOT = 0; 90const char* const IIF_NONE = NULL; 91const char* const OIF_NONE = NULL; 92const bool ACTION_ADD = true; 93const bool ACTION_DEL = false; 94const bool MODIFY_NON_UID_BASED_RULES = true; 95 96const char* const RT_TABLES_PATH = "/data/misc/net/rt_tables"; 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 if (!android::WriteStringToFile(contents, RT_TABLES_PATH, RT_TABLES_MODE, AID_SYSTEM, AID_WIFI)) { 168 ALOGE("failed to write to %s (%s)", RT_TABLES_PATH, strerror(errno)); 169 return; 170 } 171} 172 173// Sends a netlink request and expects an ack. 174// |iov| is an array of struct iovec that contains the netlink message payload. 175// The netlink header is generated by this function based on |action| and |flags|. 176// Returns -errno if there was an error or if the kernel reported an error. 177WARN_UNUSED_RESULT int sendNetlinkRequest(uint16_t action, uint16_t flags, iovec* iov, int iovlen) { 178 nlmsghdr nlmsg = { 179 .nlmsg_type = action, 180 .nlmsg_flags = flags, 181 }; 182 iov[0].iov_base = &nlmsg; 183 iov[0].iov_len = sizeof(nlmsg); 184 for (int i = 0; i < iovlen; ++i) { 185 nlmsg.nlmsg_len += iov[i].iov_len; 186 } 187 188 int ret; 189 struct { 190 nlmsghdr msg; 191 nlmsgerr err; 192 } response; 193 194 int sock = socket(AF_NETLINK, SOCK_DGRAM | SOCK_CLOEXEC, NETLINK_ROUTE); 195 if (sock != -1 && 196 connect(sock, reinterpret_cast<const sockaddr*>(&NETLINK_ADDRESS), 197 sizeof(NETLINK_ADDRESS)) != -1 && 198 writev(sock, iov, iovlen) != -1 && 199 (ret = recv(sock, &response, sizeof(response), 0)) != -1) { 200 if (ret == sizeof(response)) { 201 ret = response.err.error; // Netlink errors are negative errno. 202 if (ret) { 203 ALOGE("netlink response contains error (%s)", strerror(-ret)); 204 } 205 } else { 206 ALOGE("bad netlink response message size (%d != %zu)", ret, sizeof(response)); 207 ret = -EBADMSG; 208 } 209 } else { 210 ALOGE("netlink socket/connect/writev/recv failed (%s)", strerror(errno)); 211 ret = -errno; 212 } 213 214 if (sock != -1) { 215 close(sock); 216 } 217 218 return ret; 219} 220 221// Returns 0 on success or negative errno on failure. 222int padInterfaceName(const char* input, char* name, size_t* length, uint16_t* padding) { 223 if (!input) { 224 *length = 0; 225 *padding = 0; 226 return 0; 227 } 228 *length = strlcpy(name, input, IFNAMSIZ) + 1; 229 if (*length > IFNAMSIZ) { 230 ALOGE("interface name too long (%zu > %u)", *length, IFNAMSIZ); 231 return -ENAMETOOLONG; 232 } 233 *padding = RTA_SPACE(*length) - RTA_LENGTH(*length); 234 return 0; 235} 236 237// Adds or removes a routing rule for IPv4 and IPv6. 238// 239// + If |table| is non-zero, the rule points at the specified routing table. Otherwise, the rule 240// returns ENETUNREACH. 241// + If |mask| is non-zero, the rule matches the specified fwmark and mask. Otherwise, |fwmark| is 242// ignored. 243// + If |iif| is non-NULL, the rule matches the specified incoming interface. 244// + If |oif| is non-NULL, the rule matches the specified outgoing interface. 245// + If |uidStart| and |uidEnd| are not INVALID_UID, the rule matches packets from UIDs in that 246// range (inclusive). Otherwise, the rule matches packets from all UIDs. 247// 248// Returns 0 on success or negative errno on failure. 249WARN_UNUSED_RESULT int modifyIpRule(uint16_t action, uint32_t priority, uint32_t table, 250 uint32_t fwmark, uint32_t mask, const char* iif, 251 const char* oif, uid_t uidStart, uid_t uidEnd) { 252 // Ensure that if you set a bit in the fwmark, it's not being ignored by the mask. 253 if (fwmark & ~mask) { 254 ALOGE("mask 0x%x does not select all the bits set in fwmark 0x%x", mask, fwmark); 255 return -ERANGE; 256 } 257 258 // Interface names must include exactly one terminating NULL and be properly padded, or older 259 // kernels will refuse to delete rules. 260 char iifName[IFNAMSIZ], oifName[IFNAMSIZ]; 261 size_t iifLength, oifLength; 262 uint16_t iifPadding, oifPadding; 263 if (int ret = padInterfaceName(iif, iifName, &iifLength, &iifPadding)) { 264 return ret; 265 } 266 if (int ret = padInterfaceName(oif, oifName, &oifLength, &oifPadding)) { 267 return ret; 268 } 269 270 // Either both start and end UID must be specified, or neither. 271 if ((uidStart == INVALID_UID) != (uidEnd == INVALID_UID)) { 272 ALOGE("incompatible start and end UIDs (%u vs %u)", uidStart, uidEnd); 273 return -EUSERS; 274 } 275 bool isUidRule = (uidStart != INVALID_UID); 276 277 // Assemble a rule request and put it in an array of iovec structures. 278 fib_rule_hdr rule = { 279 .action = static_cast<uint8_t>(table != RT_TABLE_UNSPEC ? FR_ACT_TO_TBL : 280 FR_ACT_UNREACHABLE), 281 }; 282 283 rtattr fraIifName = { U16_RTA_LENGTH(iifLength), FRA_IIFNAME }; 284 rtattr fraOifName = { U16_RTA_LENGTH(oifLength), FRA_OIFNAME }; 285 286 iovec iov[] = { 287 { NULL, 0 }, 288 { &rule, sizeof(rule) }, 289 { &FRATTR_PRIORITY, sizeof(FRATTR_PRIORITY) }, 290 { &priority, sizeof(priority) }, 291 { &FRATTR_TABLE, table != RT_TABLE_UNSPEC ? sizeof(FRATTR_TABLE) : 0 }, 292 { &table, table != RT_TABLE_UNSPEC ? sizeof(table) : 0 }, 293 { &FRATTR_FWMARK, mask ? sizeof(FRATTR_FWMARK) : 0 }, 294 { &fwmark, mask ? sizeof(fwmark) : 0 }, 295 { &FRATTR_FWMASK, mask ? sizeof(FRATTR_FWMASK) : 0 }, 296 { &mask, mask ? sizeof(mask) : 0 }, 297 { &FRATTR_UID_START, isUidRule ? sizeof(FRATTR_UID_START) : 0 }, 298 { &uidStart, isUidRule ? sizeof(uidStart) : 0 }, 299 { &FRATTR_UID_END, isUidRule ? sizeof(FRATTR_UID_END) : 0 }, 300 { &uidEnd, isUidRule ? sizeof(uidEnd) : 0 }, 301 { &fraIifName, iif != IIF_NONE ? sizeof(fraIifName) : 0 }, 302 { iifName, iifLength }, 303 { PADDING_BUFFER, iifPadding }, 304 { &fraOifName, oif != OIF_NONE ? sizeof(fraOifName) : 0 }, 305 { oifName, oifLength }, 306 { PADDING_BUFFER, oifPadding }, 307 }; 308 309 uint16_t flags = (action == RTM_NEWRULE) ? NETLINK_CREATE_REQUEST_FLAGS : NETLINK_REQUEST_FLAGS; 310 for (size_t i = 0; i < ARRAY_SIZE(AF_FAMILIES); ++i) { 311 rule.family = AF_FAMILIES[i]; 312 if (int ret = sendNetlinkRequest(action, flags, iov, ARRAY_SIZE(iov))) { 313 return ret; 314 } 315 } 316 317 return 0; 318} 319 320WARN_UNUSED_RESULT int modifyIpRule(uint16_t action, uint32_t priority, uint32_t table, 321 uint32_t fwmark, uint32_t mask) { 322 return modifyIpRule(action, priority, table, fwmark, mask, IIF_NONE, OIF_NONE, INVALID_UID, 323 INVALID_UID); 324} 325 326// Adds or deletes an IPv4 or IPv6 route. 327// Returns 0 on success or negative errno on failure. 328WARN_UNUSED_RESULT int modifyIpRoute(uint16_t action, uint32_t table, const char* interface, 329 const char* destination, const char* nexthop) { 330 // At least the destination must be non-null. 331 if (!destination) { 332 ALOGE("null destination"); 333 return -EFAULT; 334 } 335 336 // Parse the prefix. 337 uint8_t rawAddress[sizeof(in6_addr)]; 338 uint8_t family; 339 uint8_t prefixLength; 340 int rawLength = parsePrefix(destination, &family, rawAddress, sizeof(rawAddress), 341 &prefixLength); 342 if (rawLength < 0) { 343 ALOGE("parsePrefix failed for destination %s (%s)", destination, strerror(-rawLength)); 344 return rawLength; 345 } 346 347 if (static_cast<size_t>(rawLength) > sizeof(rawAddress)) { 348 ALOGE("impossible! address too long (%d vs %zu)", rawLength, sizeof(rawAddress)); 349 return -ENOBUFS; // Cannot happen; parsePrefix only supports IPv4 and IPv6. 350 } 351 352 uint8_t type = RTN_UNICAST; 353 uint32_t ifindex; 354 uint8_t rawNexthop[sizeof(in6_addr)]; 355 356 if (nexthop && !strcmp(nexthop, "unreachable")) { 357 type = RTN_UNREACHABLE; 358 // 'interface' is likely non-NULL, as the caller (modifyRoute()) likely used it to lookup 359 // the table number. But it's an error to specify an interface ("dev ...") or a nexthop for 360 // unreachable routes, so nuke them. (IPv6 allows them to be specified; IPv4 doesn't.) 361 interface = OIF_NONE; 362 nexthop = NULL; 363 } else if (nexthop && !strcmp(nexthop, "throw")) { 364 type = RTN_THROW; 365 interface = OIF_NONE; 366 nexthop = NULL; 367 } else { 368 // If an interface was specified, find the ifindex. 369 if (interface != OIF_NONE) { 370 ifindex = if_nametoindex(interface); 371 if (!ifindex) { 372 ALOGE("cannot find interface %s", interface); 373 return -ENODEV; 374 } 375 } 376 377 // If a nexthop was specified, parse it as the same family as the prefix. 378 if (nexthop && inet_pton(family, nexthop, rawNexthop) <= 0) { 379 ALOGE("inet_pton failed for nexthop %s", nexthop); 380 return -EINVAL; 381 } 382 } 383 384 // Assemble a rtmsg and put it in an array of iovec structures. 385 rtmsg route = { 386 .rtm_protocol = RTPROT_STATIC, 387 .rtm_type = type, 388 .rtm_family = family, 389 .rtm_dst_len = prefixLength, 390 .rtm_scope = static_cast<uint8_t>(nexthop ? RT_SCOPE_UNIVERSE : RT_SCOPE_LINK), 391 }; 392 393 rtattr rtaDst = { U16_RTA_LENGTH(rawLength), RTA_DST }; 394 rtattr rtaGateway = { U16_RTA_LENGTH(rawLength), RTA_GATEWAY }; 395 396 iovec iov[] = { 397 { NULL, 0 }, 398 { &route, sizeof(route) }, 399 { &RTATTR_TABLE, sizeof(RTATTR_TABLE) }, 400 { &table, sizeof(table) }, 401 { &rtaDst, sizeof(rtaDst) }, 402 { rawAddress, static_cast<size_t>(rawLength) }, 403 { &RTATTR_OIF, interface != OIF_NONE ? sizeof(RTATTR_OIF) : 0 }, 404 { &ifindex, interface != OIF_NONE ? sizeof(ifindex) : 0 }, 405 { &rtaGateway, nexthop ? sizeof(rtaGateway) : 0 }, 406 { rawNexthop, nexthop ? static_cast<size_t>(rawLength) : 0 }, 407 }; 408 409 uint16_t flags = (action == RTM_NEWROUTE) ? NETLINK_CREATE_REQUEST_FLAGS : 410 NETLINK_REQUEST_FLAGS; 411 return sendNetlinkRequest(action, flags, iov, ARRAY_SIZE(iov)); 412} 413 414// An iptables rule to mark incoming packets on a network with the netId of the network. 415// 416// This is so that the kernel can: 417// + Use the right fwmark for (and thus correctly route) replies (e.g.: TCP RST, ICMP errors, ping 418// replies, SYN-ACKs, etc). 419// + Mark sockets that accept connections from this interface so that the connection stays on the 420// same interface. 421WARN_UNUSED_RESULT int modifyIncomingPacketMark(unsigned netId, const char* interface, 422 Permission permission, bool add) { 423 Fwmark fwmark; 424 425 fwmark.netId = netId; 426 fwmark.explicitlySelected = true; 427 fwmark.protectedFromVpn = true; 428 fwmark.permission = permission; 429 430 char markString[UINT32_HEX_STRLEN]; 431 snprintf(markString, sizeof(markString), "0x%x", fwmark.intValue); 432 433 if (execIptables(V4V6, "-t", "mangle", add ? "-A" : "-D", "INPUT", "-i", interface, "-j", 434 "MARK", "--set-mark", markString, NULL)) { 435 ALOGE("failed to change iptables rule that sets incoming packet mark"); 436 return -EREMOTEIO; 437 } 438 439 return 0; 440} 441 442// A rule to route responses to the local network forwarded via the VPN. 443// 444// When a VPN is in effect, packets from the local network to upstream networks are forwarded into 445// the VPN's tunnel interface. When the VPN forwards the responses, they emerge out of the tunnel. 446WARN_UNUSED_RESULT int modifyVpnOutputToLocalRule(const char* vpnInterface, bool add) { 447 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_VPN_OUTPUT_TO_LOCAL, 448 ROUTE_TABLE_LOCAL_NETWORK, MARK_UNSET, MARK_UNSET, vpnInterface, OIF_NONE, 449 INVALID_UID, INVALID_UID); 450} 451 452// A rule to route all traffic from a given set of UIDs to go over the VPN. 453// 454// Notice that this rule doesn't use the netId. I.e., no matter what netId the user's socket may 455// have, if they are subject to this VPN, their traffic has to go through it. Allows the traffic to 456// bypass the VPN if the protectedFromVpn bit is set. 457WARN_UNUSED_RESULT int modifyVpnUidRangeRule(uint32_t table, uid_t uidStart, uid_t uidEnd, 458 bool secure, bool add) { 459 Fwmark fwmark; 460 Fwmark mask; 461 462 fwmark.protectedFromVpn = false; 463 mask.protectedFromVpn = true; 464 465 uint32_t priority; 466 467 if (secure) { 468 priority = RULE_PRIORITY_SECURE_VPN; 469 } else { 470 priority = RULE_PRIORITY_BYPASSABLE_VPN; 471 472 fwmark.explicitlySelected = false; 473 mask.explicitlySelected = true; 474 } 475 476 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, priority, table, fwmark.intValue, 477 mask.intValue, IIF_NONE, OIF_NONE, uidStart, uidEnd); 478} 479 480// A rule to allow system apps to send traffic over this VPN even if they are not part of the target 481// set of UIDs. 482// 483// This is needed for DnsProxyListener to correctly resolve a request for a user who is in the 484// target set, but where the DnsProxyListener itself is not. 485WARN_UNUSED_RESULT int modifyVpnSystemPermissionRule(unsigned netId, uint32_t table, bool secure, 486 bool add) { 487 Fwmark fwmark; 488 Fwmark mask; 489 490 fwmark.netId = netId; 491 mask.netId = FWMARK_NET_ID_MASK; 492 493 fwmark.permission = PERMISSION_SYSTEM; 494 mask.permission = PERMISSION_SYSTEM; 495 496 uint32_t priority = secure ? RULE_PRIORITY_SECURE_VPN : RULE_PRIORITY_BYPASSABLE_VPN; 497 498 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, priority, table, fwmark.intValue, 499 mask.intValue); 500} 501 502// A rule to route traffic based on an explicitly chosen network. 503// 504// Supports apps that use the multinetwork APIs to restrict their traffic to a network. 505// 506// Even though we check permissions at the time we set a netId into the fwmark of a socket, we need 507// to check it again in the rules here, because a network's permissions may have been updated via 508// modifyNetworkPermission(). 509WARN_UNUSED_RESULT int modifyExplicitNetworkRule(unsigned netId, uint32_t table, 510 Permission permission, uid_t uidStart, 511 uid_t uidEnd, bool add) { 512 Fwmark fwmark; 513 Fwmark mask; 514 515 fwmark.netId = netId; 516 mask.netId = FWMARK_NET_ID_MASK; 517 518 fwmark.explicitlySelected = true; 519 mask.explicitlySelected = true; 520 521 fwmark.permission = permission; 522 mask.permission = permission; 523 524 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_EXPLICIT_NETWORK, table, 525 fwmark.intValue, mask.intValue, IIF_NONE, OIF_NONE, uidStart, uidEnd); 526} 527 528// A rule to route traffic based on a chosen outgoing interface. 529// 530// Supports apps that use SO_BINDTODEVICE or IP_PKTINFO options and the kernel that already knows 531// the outgoing interface (typically for link-local communications). 532WARN_UNUSED_RESULT int modifyOutputInterfaceRule(const char* interface, uint32_t table, 533 Permission permission, uid_t uidStart, 534 uid_t uidEnd, bool add) { 535 Fwmark fwmark; 536 Fwmark mask; 537 538 fwmark.permission = permission; 539 mask.permission = permission; 540 541 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_OUTPUT_INTERFACE, table, 542 fwmark.intValue, mask.intValue, IIF_NONE, interface, uidStart, uidEnd); 543} 544 545// A rule to route traffic based on the chosen network. 546// 547// This is for sockets that have not explicitly requested a particular network, but have been 548// bound to one when they called connect(). This ensures that sockets connected on a particular 549// network stay on that network even if the default network changes. 550WARN_UNUSED_RESULT int modifyImplicitNetworkRule(unsigned netId, uint32_t table, 551 Permission permission, bool add) { 552 Fwmark fwmark; 553 Fwmark mask; 554 555 fwmark.netId = netId; 556 mask.netId = FWMARK_NET_ID_MASK; 557 558 fwmark.explicitlySelected = false; 559 mask.explicitlySelected = true; 560 561 fwmark.permission = permission; 562 mask.permission = permission; 563 564 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_IMPLICIT_NETWORK, table, 565 fwmark.intValue, mask.intValue); 566} 567 568// A rule to enable split tunnel VPNs. 569// 570// If a packet with a VPN's netId doesn't find a route in the VPN's routing table, it's allowed to 571// go over the default network, provided it wasn't explicitly restricted to the VPN and has the 572// permissions required by the default network. 573WARN_UNUSED_RESULT int modifyVpnFallthroughRule(uint16_t action, unsigned vpnNetId, 574 const char* physicalInterface, 575 Permission permission) { 576 uint32_t table = getRouteTableForInterface(physicalInterface); 577 if (table == RT_TABLE_UNSPEC) { 578 return -ESRCH; 579 } 580 581 Fwmark fwmark; 582 Fwmark mask; 583 584 fwmark.netId = vpnNetId; 585 mask.netId = FWMARK_NET_ID_MASK; 586 587 fwmark.explicitlySelected = false; 588 mask.explicitlySelected = true; 589 590 fwmark.permission = permission; 591 mask.permission = permission; 592 593 return modifyIpRule(action, RULE_PRIORITY_VPN_FALLTHROUGH, table, fwmark.intValue, 594 mask.intValue); 595} 596 597// Add rules to allow legacy routes added through the requestRouteToHost() API. 598WARN_UNUSED_RESULT int addLegacyRouteRules() { 599 Fwmark fwmark; 600 Fwmark mask; 601 602 fwmark.explicitlySelected = false; 603 mask.explicitlySelected = true; 604 605 // Rules to allow legacy routes to override the default network. 606 if (int ret = modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LEGACY_SYSTEM, ROUTE_TABLE_LEGACY_SYSTEM, 607 fwmark.intValue, mask.intValue)) { 608 return ret; 609 } 610 if (int ret = modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LEGACY_NETWORK, 611 ROUTE_TABLE_LEGACY_NETWORK, fwmark.intValue, mask.intValue)) { 612 return ret; 613 } 614 615 fwmark.permission = PERMISSION_SYSTEM; 616 mask.permission = PERMISSION_SYSTEM; 617 618 // A rule to allow legacy routes from system apps to override VPNs. 619 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_VPN_OVERRIDE_SYSTEM, ROUTE_TABLE_LEGACY_SYSTEM, 620 fwmark.intValue, mask.intValue); 621} 622 623// Add rules to lookup the local network when specified explicitly or otherwise. 624WARN_UNUSED_RESULT int addLocalNetworkRules(unsigned localNetId) { 625 if (int ret = modifyExplicitNetworkRule(localNetId, ROUTE_TABLE_LOCAL_NETWORK, PERMISSION_NONE, 626 INVALID_UID, INVALID_UID, ACTION_ADD)) { 627 return ret; 628 } 629 630 Fwmark fwmark; 631 Fwmark mask; 632 633 fwmark.explicitlySelected = false; 634 mask.explicitlySelected = true; 635 636 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LOCAL_NETWORK, ROUTE_TABLE_LOCAL_NETWORK, 637 fwmark.intValue, mask.intValue); 638} 639 640// Add a new rule to look up the 'main' table, with the same selectors as the "default network" 641// rule, but with a lower priority. We will never create routes in the main table; it should only be 642// used for directly-connected routes implicitly created by the kernel when adding IP addresses. 643// This is necessary, for example, when adding a route through a directly-connected gateway: in 644// order to add the route, there must already be a directly-connected route that covers the gateway. 645WARN_UNUSED_RESULT int addDirectlyConnectedRule() { 646 Fwmark fwmark; 647 Fwmark mask; 648 649 fwmark.netId = NETID_UNSET; 650 mask.netId = FWMARK_NET_ID_MASK; 651 652 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_DIRECTLY_CONNECTED, RT_TABLE_MAIN, 653 fwmark.intValue, mask.intValue, IIF_NONE, OIF_NONE, UID_ROOT, UID_ROOT); 654} 655 656// Add an explicit unreachable rule close to the end of the prioriy list to make it clear that 657// relying on the kernel-default "from all lookup main" rule at priority 32766 is not intended 658// behaviour. We do flush the kernel-default rules at startup, but having an explicit unreachable 659// rule will hopefully make things even clearer. 660WARN_UNUSED_RESULT int addUnreachableRule() { 661 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_UNREACHABLE, RT_TABLE_UNSPEC, MARK_UNSET, 662 MARK_UNSET); 663} 664 665WARN_UNUSED_RESULT int modifyLocalNetwork(unsigned netId, const char* interface, bool add) { 666 if (int ret = modifyIncomingPacketMark(netId, interface, PERMISSION_NONE, add)) { 667 return ret; 668 } 669 return modifyOutputInterfaceRule(interface, ROUTE_TABLE_LOCAL_NETWORK, PERMISSION_NONE, 670 INVALID_UID, INVALID_UID, add); 671} 672 673WARN_UNUSED_RESULT int modifyPhysicalNetwork(unsigned netId, const char* interface, 674 Permission permission, bool add) { 675 uint32_t table = getRouteTableForInterface(interface); 676 if (table == RT_TABLE_UNSPEC) { 677 return -ESRCH; 678 } 679 680 if (int ret = modifyIncomingPacketMark(netId, interface, permission, add)) { 681 return ret; 682 } 683 if (int ret = modifyExplicitNetworkRule(netId, table, permission, INVALID_UID, INVALID_UID, 684 add)) { 685 return ret; 686 } 687 if (int ret = modifyOutputInterfaceRule(interface, table, permission, INVALID_UID, INVALID_UID, 688 add)) { 689 return ret; 690 } 691 return modifyImplicitNetworkRule(netId, table, permission, add); 692} 693 694WARN_UNUSED_RESULT int modifyVirtualNetwork(unsigned netId, const char* interface, 695 const UidRanges& uidRanges, bool secure, bool add, 696 bool modifyNonUidBasedRules) { 697 uint32_t table = getRouteTableForInterface(interface); 698 if (table == RT_TABLE_UNSPEC) { 699 return -ESRCH; 700 } 701 702 for (const UidRanges::Range& range : uidRanges.getRanges()) { 703 if (int ret = modifyVpnUidRangeRule(table, range.first, range.second, secure, add)) { 704 return ret; 705 } 706 if (int ret = modifyExplicitNetworkRule(netId, table, PERMISSION_NONE, range.first, 707 range.second, add)) { 708 return ret; 709 } 710 if (int ret = modifyOutputInterfaceRule(interface, table, PERMISSION_NONE, range.first, 711 range.second, add)) { 712 return ret; 713 } 714 } 715 716 if (modifyNonUidBasedRules) { 717 if (int ret = modifyIncomingPacketMark(netId, interface, PERMISSION_NONE, add)) { 718 return ret; 719 } 720 if (int ret = modifyVpnOutputToLocalRule(interface, add)) { 721 return ret; 722 } 723 if (int ret = modifyVpnSystemPermissionRule(netId, table, secure, add)) { 724 return ret; 725 } 726 return modifyExplicitNetworkRule(netId, table, PERMISSION_NONE, UID_ROOT, UID_ROOT, add); 727 } 728 729 return 0; 730} 731 732WARN_UNUSED_RESULT int modifyDefaultNetwork(uint16_t action, const char* interface, 733 Permission permission) { 734 uint32_t table = getRouteTableForInterface(interface); 735 if (table == RT_TABLE_UNSPEC) { 736 return -ESRCH; 737 } 738 739 Fwmark fwmark; 740 Fwmark mask; 741 742 fwmark.netId = NETID_UNSET; 743 mask.netId = FWMARK_NET_ID_MASK; 744 745 fwmark.permission = permission; 746 mask.permission = permission; 747 748 return modifyIpRule(action, RULE_PRIORITY_DEFAULT_NETWORK, table, fwmark.intValue, 749 mask.intValue); 750} 751 752WARN_UNUSED_RESULT int modifyTetheredNetwork(uint16_t action, const char* inputInterface, 753 const char* outputInterface) { 754 uint32_t table = getRouteTableForInterface(outputInterface); 755 if (table == RT_TABLE_UNSPEC) { 756 return -ESRCH; 757 } 758 759 return modifyIpRule(action, RULE_PRIORITY_TETHERING, table, MARK_UNSET, MARK_UNSET, 760 inputInterface, OIF_NONE, INVALID_UID, INVALID_UID); 761} 762 763// Returns 0 on success or negative errno on failure. 764WARN_UNUSED_RESULT int flushRules() { 765 for (size_t i = 0; i < ARRAY_SIZE(IP_VERSIONS); ++i) { 766 const char* argv[] = { 767 IP_PATH, 768 IP_VERSIONS[i], 769 "rule", 770 "flush", 771 }; 772 if (android_fork_execvp(ARRAY_SIZE(argv), const_cast<char**>(argv), NULL, false, false)) { 773 ALOGE("failed to flush rules"); 774 return -EREMOTEIO; 775 } 776 } 777 return 0; 778} 779 780// Adds or removes an IPv4 or IPv6 route to the specified table and, if it's a directly-connected 781// route, to the main table as well. 782// Returns 0 on success or negative errno on failure. 783WARN_UNUSED_RESULT int modifyRoute(uint16_t action, const char* interface, const char* destination, 784 const char* nexthop, RouteController::TableType tableType) { 785 uint32_t table; 786 switch (tableType) { 787 case RouteController::INTERFACE: { 788 table = getRouteTableForInterface(interface); 789 if (table == RT_TABLE_UNSPEC) { 790 return -ESRCH; 791 } 792 break; 793 } 794 case RouteController::LOCAL_NETWORK: { 795 table = ROUTE_TABLE_LOCAL_NETWORK; 796 break; 797 } 798 case RouteController::LEGACY_NETWORK: { 799 table = ROUTE_TABLE_LEGACY_NETWORK; 800 break; 801 } 802 case RouteController::LEGACY_SYSTEM: { 803 table = ROUTE_TABLE_LEGACY_SYSTEM; 804 break; 805 } 806 } 807 808 int ret = modifyIpRoute(action, table, interface, destination, nexthop); 809 // Trying to add a route that already exists shouldn't cause an error. 810 if (ret && !(action == RTM_NEWROUTE && ret == -EEXIST)) { 811 return ret; 812 } 813 814 return 0; 815} 816 817// Returns 0 on success or negative errno on failure. 818WARN_UNUSED_RESULT int flushRoutes(const char* interface) { 819 uint32_t table = getRouteTableForInterface(interface); 820 if (table == RT_TABLE_UNSPEC) { 821 return -ESRCH; 822 } 823 824 char tableString[UINT32_STRLEN]; 825 snprintf(tableString, sizeof(tableString), "%u", table); 826 827 int ret = 0; 828 for (size_t i = 0; i < ARRAY_SIZE(IP_VERSIONS); ++i) { 829 const char* argv[] = { 830 IP_PATH, 831 IP_VERSIONS[i], 832 "route", 833 "flush", 834 "table", 835 tableString, 836 }; 837 838 // A flush works by dumping routes and deleting each route as it's returned, and it can 839 // fail if something else deletes the route between the dump and the delete. This can 840 // happen, for example, if an interface goes down while we're trying to flush its routes. 841 // So try multiple times and only return an error if the last attempt fails. 842 // 843 // TODO: replace this with our own netlink code. 844 unsigned attempts = 0; 845 int err; 846 do { 847 err = android_fork_execvp(ARRAY_SIZE(argv), const_cast<char**>(argv), 848 NULL, false, false); 849 ++attempts; 850 } while (err != 0 && attempts < ROUTE_FLUSH_ATTEMPTS); 851 if (err) { 852 ALOGE("failed to flush %s routes in table %s after %d attempts", 853 IP_VERSIONS[i], tableString, attempts); 854 ret = -EREMOTEIO; 855 } 856 } 857 858 // If we failed to flush routes, the caller may elect to keep this interface around, so keep 859 // track of its name. 860 if (!ret) { 861 interfaceToTable.erase(interface); 862 } 863 864 return ret; 865} 866 867} // namespace 868 869int RouteController::Init(unsigned localNetId) { 870 if (int ret = flushRules()) { 871 return ret; 872 } 873 if (int ret = addLegacyRouteRules()) { 874 return ret; 875 } 876 if (int ret = addLocalNetworkRules(localNetId)) { 877 return ret; 878 } 879 if (int ret = addDirectlyConnectedRule()) { 880 return ret; 881 } 882 if (int ret = addUnreachableRule()) { 883 return ret; 884 } 885 updateTableNamesFile(); 886 return 0; 887} 888 889int RouteController::addInterfaceToLocalNetwork(unsigned netId, const char* interface) { 890 return modifyLocalNetwork(netId, interface, ACTION_ADD); 891} 892 893int RouteController::removeInterfaceFromLocalNetwork(unsigned netId, const char* interface) { 894 return modifyLocalNetwork(netId, interface, ACTION_DEL); 895} 896 897int RouteController::addInterfaceToPhysicalNetwork(unsigned netId, const char* interface, 898 Permission permission) { 899 if (int ret = modifyPhysicalNetwork(netId, interface, permission, ACTION_ADD)) { 900 return ret; 901 } 902 updateTableNamesFile(); 903 return 0; 904} 905 906int RouteController::removeInterfaceFromPhysicalNetwork(unsigned netId, const char* interface, 907 Permission permission) { 908 if (int ret = modifyPhysicalNetwork(netId, interface, permission, ACTION_DEL)) { 909 return ret; 910 } 911 if (int ret = flushRoutes(interface)) { 912 return ret; 913 } 914 updateTableNamesFile(); 915 return 0; 916} 917 918int RouteController::addInterfaceToVirtualNetwork(unsigned netId, const char* interface, 919 bool secure, const UidRanges& uidRanges) { 920 if (int ret = modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_ADD, 921 MODIFY_NON_UID_BASED_RULES)) { 922 return ret; 923 } 924 updateTableNamesFile(); 925 return 0; 926} 927 928int RouteController::removeInterfaceFromVirtualNetwork(unsigned netId, const char* interface, 929 bool secure, const UidRanges& uidRanges) { 930 if (int ret = modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_DEL, 931 MODIFY_NON_UID_BASED_RULES)) { 932 return ret; 933 } 934 if (int ret = flushRoutes(interface)) { 935 return ret; 936 } 937 updateTableNamesFile(); 938 return 0; 939} 940 941int RouteController::modifyPhysicalNetworkPermission(unsigned netId, const char* interface, 942 Permission oldPermission, 943 Permission newPermission) { 944 // Add the new rules before deleting the old ones, to avoid race conditions. 945 if (int ret = modifyPhysicalNetwork(netId, interface, newPermission, ACTION_ADD)) { 946 return ret; 947 } 948 return modifyPhysicalNetwork(netId, interface, oldPermission, ACTION_DEL); 949} 950 951int RouteController::addUsersToVirtualNetwork(unsigned netId, const char* interface, bool secure, 952 const UidRanges& uidRanges) { 953 return modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_ADD, 954 !MODIFY_NON_UID_BASED_RULES); 955} 956 957int RouteController::removeUsersFromVirtualNetwork(unsigned netId, const char* interface, 958 bool secure, const UidRanges& uidRanges) { 959 return modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_DEL, 960 !MODIFY_NON_UID_BASED_RULES); 961} 962 963int RouteController::addInterfaceToDefaultNetwork(const char* interface, Permission permission) { 964 return modifyDefaultNetwork(RTM_NEWRULE, interface, permission); 965} 966 967int RouteController::removeInterfaceFromDefaultNetwork(const char* interface, 968 Permission permission) { 969 return modifyDefaultNetwork(RTM_DELRULE, interface, permission); 970} 971 972int RouteController::addRoute(const char* interface, const char* destination, const char* nexthop, 973 TableType tableType) { 974 return modifyRoute(RTM_NEWROUTE, interface, destination, nexthop, tableType); 975} 976 977int RouteController::removeRoute(const char* interface, const char* destination, 978 const char* nexthop, TableType tableType) { 979 return modifyRoute(RTM_DELROUTE, interface, destination, nexthop, tableType); 980} 981 982int RouteController::enableTethering(const char* inputInterface, const char* outputInterface) { 983 return modifyTetheredNetwork(RTM_NEWRULE, inputInterface, outputInterface); 984} 985 986int RouteController::disableTethering(const char* inputInterface, const char* outputInterface) { 987 return modifyTetheredNetwork(RTM_DELRULE, inputInterface, outputInterface); 988} 989 990int RouteController::addVirtualNetworkFallthrough(unsigned vpnNetId, const char* physicalInterface, 991 Permission permission) { 992 return modifyVpnFallthroughRule(RTM_NEWRULE, vpnNetId, physicalInterface, permission); 993} 994 995int RouteController::removeVirtualNetworkFallthrough(unsigned vpnNetId, 996 const char* physicalInterface, 997 Permission permission) { 998 return modifyVpnFallthroughRule(RTM_DELRULE, vpnNetId, physicalInterface, permission); 999} 1000