RouteController.cpp revision 53ea9cadf6cc5f8be1c16b5b6b660cd7366fd3f0
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 <map> 27 28#include "Fwmark.h" 29#include "UidRanges.h" 30 31#define LOG_TAG "Netd" 32#include "log/log.h" 33#include "logwrap/logwrap.h" 34#include "resolv_netid.h" 35 36namespace { 37 38// BEGIN CONSTANTS -------------------------------------------------------------------------------- 39 40const uint32_t RULE_PRIORITY_VPN_OVERRIDE_SYSTEM = 10000; 41const uint32_t RULE_PRIORITY_VPN_OUTPUT_TO_LOCAL = 11000; 42const uint32_t RULE_PRIORITY_SECURE_VPN = 12000; 43const uint32_t RULE_PRIORITY_EXPLICIT_NETWORK = 13000; 44const uint32_t RULE_PRIORITY_OUTPUT_INTERFACE = 14000; 45const uint32_t RULE_PRIORITY_LEGACY_SYSTEM = 15000; 46const uint32_t RULE_PRIORITY_LEGACY_NETWORK = 16000; 47const uint32_t RULE_PRIORITY_LOCAL_NETWORK = 17000; 48const uint32_t RULE_PRIORITY_TETHERING = 18000; 49const uint32_t RULE_PRIORITY_IMPLICIT_NETWORK = 19000; 50const uint32_t RULE_PRIORITY_BYPASSABLE_VPN = 20000; 51const uint32_t RULE_PRIORITY_VPN_FALLTHROUGH = 21000; 52const uint32_t RULE_PRIORITY_DEFAULT_NETWORK = 22000; 53const uint32_t RULE_PRIORITY_DIRECTLY_CONNECTED = 23000; 54const uint32_t RULE_PRIORITY_UNREACHABLE = 32000; 55 56const uint32_t ROUTE_TABLE_LOCAL_NETWORK = 97; 57const uint32_t ROUTE_TABLE_LEGACY_NETWORK = 98; 58const uint32_t ROUTE_TABLE_LEGACY_SYSTEM = 99; 59 60const char* const ROUTE_TABLE_NAME_LOCAL_NETWORK = "local_network"; 61const char* const ROUTE_TABLE_NAME_LEGACY_NETWORK = "legacy_network"; 62const char* const ROUTE_TABLE_NAME_LEGACY_SYSTEM = "legacy_system"; 63 64const char* const ROUTE_TABLE_NAME_LOCAL = "local"; 65const char* const ROUTE_TABLE_NAME_MAIN = "main"; 66 67// TODO: These values aren't defined by the Linux kernel, because our UID routing changes are not 68// upstream (yet?), so we can't just pick them up from kernel headers. When (if?) the changes make 69// it upstream, we'll remove this and rely on the kernel header values. For now, add a static assert 70// that will warn us if upstream has given these values some other meaning. 71const uint16_t FRA_UID_START = 18; 72const uint16_t FRA_UID_END = 19; 73static_assert(FRA_UID_START > FRA_MAX, 74 "Android-specific FRA_UID_{START,END} values also assigned in Linux uapi. " 75 "Check that these values match what the kernel does and then update this assertion."); 76 77const uint16_t NETLINK_REQUEST_FLAGS = NLM_F_REQUEST | NLM_F_ACK; 78const uint16_t NETLINK_CREATE_REQUEST_FLAGS = NETLINK_REQUEST_FLAGS | NLM_F_CREATE | NLM_F_EXCL; 79 80const sockaddr_nl NETLINK_ADDRESS = {AF_NETLINK, 0, 0, 0}; 81 82const uint8_t AF_FAMILIES[] = {AF_INET, AF_INET6}; 83 84const char* const IP_VERSIONS[] = {"-4", "-6"}; 85 86const uid_t UID_ROOT = 0; 87const char* const IIF_NONE = NULL; 88const char* const OIF_NONE = NULL; 89const bool ACTION_ADD = true; 90const bool ACTION_DEL = false; 91const bool MODIFY_NON_UID_BASED_RULES = true; 92 93const char* const RT_TABLES_PATH = "/data/misc/net/rt_tables"; 94const int RT_TABLES_FLAGS = O_CREAT | O_TRUNC | O_WRONLY | O_NOFOLLOW | O_CLOEXEC; 95const mode_t RT_TABLES_MODE = S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH; // mode 0644, rw-r--r-- 96 97const unsigned ROUTE_FLUSH_ATTEMPTS = 2; 98 99// Avoids "non-constant-expression cannot be narrowed from type 'unsigned int' to 'unsigned short'" 100// warnings when using RTA_LENGTH(x) inside static initializers (even when x is already uint16_t). 101constexpr uint16_t U16_RTA_LENGTH(uint16_t x) { 102 return RTA_LENGTH(x); 103} 104 105// These are practically const, but can't be declared so, because they are used to initialize 106// non-const pointers ("void* iov_base") in iovec arrays. 107rtattr FRATTR_PRIORITY = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_PRIORITY }; 108rtattr FRATTR_TABLE = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_TABLE }; 109rtattr FRATTR_FWMARK = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_FWMARK }; 110rtattr FRATTR_FWMASK = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_FWMASK }; 111rtattr FRATTR_UID_START = { U16_RTA_LENGTH(sizeof(uid_t)), FRA_UID_START }; 112rtattr FRATTR_UID_END = { U16_RTA_LENGTH(sizeof(uid_t)), FRA_UID_END }; 113 114rtattr RTATTR_TABLE = { U16_RTA_LENGTH(sizeof(uint32_t)), RTA_TABLE }; 115rtattr RTATTR_OIF = { U16_RTA_LENGTH(sizeof(uint32_t)), RTA_OIF }; 116 117uint8_t PADDING_BUFFER[RTA_ALIGNTO] = {0, 0, 0, 0}; 118 119// END CONSTANTS ---------------------------------------------------------------------------------- 120 121// No locks needed because RouteController is accessed only from one thread (in CommandListener). 122std::map<std::string, uint32_t> interfaceToTable; 123 124uint32_t getRouteTableForInterface(const char* interface) { 125 uint32_t index = if_nametoindex(interface); 126 if (index) { 127 index += RouteController::ROUTE_TABLE_OFFSET_FROM_INDEX; 128 interfaceToTable[interface] = index; 129 return index; 130 } 131 // If the interface goes away if_nametoindex() will return 0 but we still need to know 132 // the index so we can remove the rules and routes. 133 auto iter = interfaceToTable.find(interface); 134 if (iter == interfaceToTable.end()) { 135 ALOGE("cannot find interface %s", interface); 136 return RT_TABLE_UNSPEC; 137 } 138 return iter->second; 139} 140 141void addTableName(uint32_t table, const std::string& name, std::string* contents) { 142 char tableString[UINT32_STRLEN]; 143 snprintf(tableString, sizeof(tableString), "%u", table); 144 *contents += tableString; 145 *contents += " "; 146 *contents += name; 147 *contents += "\n"; 148} 149 150// Doesn't return success/failure as the file is optional; it's okay if we fail to update it. 151void updateTableNamesFile() { 152 std::string contents; 153 154 addTableName(RT_TABLE_LOCAL, ROUTE_TABLE_NAME_LOCAL, &contents); 155 addTableName(RT_TABLE_MAIN, ROUTE_TABLE_NAME_MAIN, &contents); 156 157 addTableName(ROUTE_TABLE_LOCAL_NETWORK, ROUTE_TABLE_NAME_LOCAL_NETWORK, &contents); 158 addTableName(ROUTE_TABLE_LEGACY_NETWORK, ROUTE_TABLE_NAME_LEGACY_NETWORK, &contents); 159 addTableName(ROUTE_TABLE_LEGACY_SYSTEM, ROUTE_TABLE_NAME_LEGACY_SYSTEM, &contents); 160 161 for (const auto& entry : interfaceToTable) { 162 addTableName(entry.second, entry.first, &contents); 163 } 164 165 int fd = open(RT_TABLES_PATH, RT_TABLES_FLAGS, RT_TABLES_MODE); 166 if (fd == -1) { 167 ALOGE("failed to create %s (%s)", RT_TABLES_PATH, strerror(errno)); 168 return; 169 } 170 // File creation is affected by umask, so make sure the right mode bits are set. 171 if (fchmod(fd, RT_TABLES_MODE) == -1) { 172 ALOGE("failed to set mode 0%o on %s (%s)", RT_TABLES_MODE, RT_TABLES_PATH, strerror(errno)); 173 } 174 ssize_t bytesWritten = write(fd, contents.data(), contents.size()); 175 if (bytesWritten != static_cast<ssize_t>(contents.size())) { 176 ALOGE("failed to write to %s (%zd vs %zu bytes) (%s)", RT_TABLES_PATH, bytesWritten, 177 contents.size(), strerror(errno)); 178 } 179 close(fd); 180} 181 182// Sends a netlink request and expects an ack. 183// |iov| is an array of struct iovec that contains the netlink message payload. 184// The netlink header is generated by this function based on |action| and |flags|. 185// Returns -errno if there was an error or if the kernel reported an error. 186WARN_UNUSED_RESULT int sendNetlinkRequest(uint16_t action, uint16_t flags, iovec* iov, int iovlen) { 187 nlmsghdr nlmsg = { 188 .nlmsg_type = action, 189 .nlmsg_flags = flags, 190 }; 191 iov[0].iov_base = &nlmsg; 192 iov[0].iov_len = sizeof(nlmsg); 193 for (int i = 0; i < iovlen; ++i) { 194 nlmsg.nlmsg_len += iov[i].iov_len; 195 } 196 197 int ret; 198 struct { 199 nlmsghdr msg; 200 nlmsgerr err; 201 } response; 202 203 int sock = socket(AF_NETLINK, SOCK_DGRAM | SOCK_CLOEXEC, NETLINK_ROUTE); 204 if (sock != -1 && 205 connect(sock, reinterpret_cast<const sockaddr*>(&NETLINK_ADDRESS), 206 sizeof(NETLINK_ADDRESS)) != -1 && 207 writev(sock, iov, iovlen) != -1 && 208 (ret = recv(sock, &response, sizeof(response), 0)) != -1) { 209 if (ret == sizeof(response)) { 210 ret = response.err.error; // Netlink errors are negative errno. 211 if (ret) { 212 ALOGE("netlink response contains error (%s)", strerror(-ret)); 213 } 214 } else { 215 ALOGE("bad netlink response message size (%d != %zu)", ret, sizeof(response)); 216 ret = -EBADMSG; 217 } 218 } else { 219 ALOGE("netlink socket/connect/writev/recv failed (%s)", strerror(errno)); 220 ret = -errno; 221 } 222 223 if (sock != -1) { 224 close(sock); 225 } 226 227 return ret; 228} 229 230// Returns 0 on success or negative errno on failure. 231int padInterfaceName(const char* input, char* name, size_t* length, uint16_t* padding) { 232 if (!input) { 233 *length = 0; 234 *padding = 0; 235 return 0; 236 } 237 *length = strlcpy(name, input, IFNAMSIZ) + 1; 238 if (*length > IFNAMSIZ) { 239 ALOGE("interface name too long (%zu > %u)", *length, IFNAMSIZ); 240 return -ENAMETOOLONG; 241 } 242 *padding = RTA_SPACE(*length) - RTA_LENGTH(*length); 243 return 0; 244} 245 246// Adds or removes a routing rule for IPv4 and IPv6. 247// 248// + If |table| is non-zero, the rule points at the specified routing table. Otherwise, the rule 249// returns ENETUNREACH. 250// + If |mask| is non-zero, the rule matches the specified fwmark and mask. Otherwise, |fwmark| is 251// ignored. 252// + If |iif| is non-NULL, the rule matches the specified incoming interface. 253// + If |oif| is non-NULL, the rule matches the specified outgoing interface. 254// + If |uidStart| and |uidEnd| are not INVALID_UID, the rule matches packets from UIDs in that 255// range (inclusive). Otherwise, the rule matches packets from all UIDs. 256// 257// Returns 0 on success or negative errno on failure. 258WARN_UNUSED_RESULT int modifyIpRule(uint16_t action, uint32_t priority, uint32_t table, 259 uint32_t fwmark, uint32_t mask, const char* iif, 260 const char* oif, uid_t uidStart, uid_t uidEnd) { 261 // Ensure that if you set a bit in the fwmark, it's not being ignored by the mask. 262 if (fwmark & ~mask) { 263 ALOGE("mask 0x%x does not select all the bits set in fwmark 0x%x", mask, fwmark); 264 return -ERANGE; 265 } 266 267 // Interface names must include exactly one terminating NULL and be properly padded, or older 268 // kernels will refuse to delete rules. 269 char iifName[IFNAMSIZ], oifName[IFNAMSIZ]; 270 size_t iifLength, oifLength; 271 uint16_t iifPadding, oifPadding; 272 if (int ret = padInterfaceName(iif, iifName, &iifLength, &iifPadding)) { 273 return ret; 274 } 275 if (int ret = padInterfaceName(oif, oifName, &oifLength, &oifPadding)) { 276 return ret; 277 } 278 279 // Either both start and end UID must be specified, or neither. 280 if ((uidStart == INVALID_UID) != (uidEnd == INVALID_UID)) { 281 ALOGE("incompatible start and end UIDs (%u vs %u)", uidStart, uidEnd); 282 return -EUSERS; 283 } 284 bool isUidRule = (uidStart != INVALID_UID); 285 286 // Assemble a rule request and put it in an array of iovec structures. 287 fib_rule_hdr rule = { 288 .action = static_cast<uint8_t>(table != RT_TABLE_UNSPEC ? FR_ACT_TO_TBL : 289 FR_ACT_UNREACHABLE), 290 }; 291 292 rtattr fraIifName = { U16_RTA_LENGTH(iifLength), FRA_IIFNAME }; 293 rtattr fraOifName = { U16_RTA_LENGTH(oifLength), FRA_OIFNAME }; 294 295 iovec iov[] = { 296 { NULL, 0 }, 297 { &rule, sizeof(rule) }, 298 { &FRATTR_PRIORITY, sizeof(FRATTR_PRIORITY) }, 299 { &priority, sizeof(priority) }, 300 { &FRATTR_TABLE, table != RT_TABLE_UNSPEC ? sizeof(FRATTR_TABLE) : 0 }, 301 { &table, table != RT_TABLE_UNSPEC ? sizeof(table) : 0 }, 302 { &FRATTR_FWMARK, mask ? sizeof(FRATTR_FWMARK) : 0 }, 303 { &fwmark, mask ? sizeof(fwmark) : 0 }, 304 { &FRATTR_FWMASK, mask ? sizeof(FRATTR_FWMASK) : 0 }, 305 { &mask, mask ? sizeof(mask) : 0 }, 306 { &FRATTR_UID_START, isUidRule ? sizeof(FRATTR_UID_START) : 0 }, 307 { &uidStart, isUidRule ? sizeof(uidStart) : 0 }, 308 { &FRATTR_UID_END, isUidRule ? sizeof(FRATTR_UID_END) : 0 }, 309 { &uidEnd, isUidRule ? sizeof(uidEnd) : 0 }, 310 { &fraIifName, iif != IIF_NONE ? sizeof(fraIifName) : 0 }, 311 { iifName, iifLength }, 312 { PADDING_BUFFER, iifPadding }, 313 { &fraOifName, oif != OIF_NONE ? sizeof(fraOifName) : 0 }, 314 { oifName, oifLength }, 315 { PADDING_BUFFER, oifPadding }, 316 }; 317 318 uint16_t flags = (action == RTM_NEWRULE) ? NETLINK_CREATE_REQUEST_FLAGS : NETLINK_REQUEST_FLAGS; 319 for (size_t i = 0; i < ARRAY_SIZE(AF_FAMILIES); ++i) { 320 rule.family = AF_FAMILIES[i]; 321 if (int ret = sendNetlinkRequest(action, flags, iov, ARRAY_SIZE(iov))) { 322 return ret; 323 } 324 } 325 326 return 0; 327} 328 329WARN_UNUSED_RESULT int modifyIpRule(uint16_t action, uint32_t priority, uint32_t table, 330 uint32_t fwmark, uint32_t mask) { 331 return modifyIpRule(action, priority, table, fwmark, mask, IIF_NONE, OIF_NONE, INVALID_UID, 332 INVALID_UID); 333} 334 335// Adds or deletes an IPv4 or IPv6 route. 336// Returns 0 on success or negative errno on failure. 337WARN_UNUSED_RESULT int modifyIpRoute(uint16_t action, uint32_t table, const char* interface, 338 const char* destination, const char* nexthop) { 339 // At least the destination must be non-null. 340 if (!destination) { 341 ALOGE("null destination"); 342 return -EFAULT; 343 } 344 345 // Parse the prefix. 346 uint8_t rawAddress[sizeof(in6_addr)]; 347 uint8_t family; 348 uint8_t prefixLength; 349 int rawLength = parsePrefix(destination, &family, rawAddress, sizeof(rawAddress), 350 &prefixLength); 351 if (rawLength < 0) { 352 ALOGE("parsePrefix failed for destination %s (%s)", destination, strerror(-rawLength)); 353 return rawLength; 354 } 355 356 if (static_cast<size_t>(rawLength) > sizeof(rawAddress)) { 357 ALOGE("impossible! address too long (%d vs %zu)", rawLength, sizeof(rawAddress)); 358 return -ENOBUFS; // Cannot happen; parsePrefix only supports IPv4 and IPv6. 359 } 360 361 uint8_t type = RTN_UNICAST; 362 uint32_t ifindex; 363 uint8_t rawNexthop[sizeof(in6_addr)]; 364 365 if (nexthop && !strcmp(nexthop, "unreachable")) { 366 type = RTN_UNREACHABLE; 367 // 'interface' is likely non-NULL, as the caller (modifyRoute()) likely used it to lookup 368 // the table number. But it's an error to specify an interface ("dev ...") or a nexthop for 369 // unreachable routes, so nuke them. (IPv6 allows them to be specified; IPv4 doesn't.) 370 interface = OIF_NONE; 371 nexthop = NULL; 372 } else if (nexthop && !strcmp(nexthop, "throw")) { 373 type = RTN_THROW; 374 interface = OIF_NONE; 375 nexthop = NULL; 376 } else { 377 // If an interface was specified, find the ifindex. 378 if (interface != OIF_NONE) { 379 ifindex = if_nametoindex(interface); 380 if (!ifindex) { 381 ALOGE("cannot find interface %s", interface); 382 return -ENODEV; 383 } 384 } 385 386 // If a nexthop was specified, parse it as the same family as the prefix. 387 if (nexthop && inet_pton(family, nexthop, rawNexthop) <= 0) { 388 ALOGE("inet_pton failed for nexthop %s", nexthop); 389 return -EINVAL; 390 } 391 } 392 393 // Assemble a rtmsg and put it in an array of iovec structures. 394 rtmsg route = { 395 .rtm_protocol = RTPROT_STATIC, 396 .rtm_type = type, 397 .rtm_family = family, 398 .rtm_dst_len = prefixLength, 399 .rtm_scope = static_cast<uint8_t>(nexthop ? RT_SCOPE_UNIVERSE : RT_SCOPE_LINK), 400 }; 401 402 rtattr rtaDst = { U16_RTA_LENGTH(rawLength), RTA_DST }; 403 rtattr rtaGateway = { U16_RTA_LENGTH(rawLength), RTA_GATEWAY }; 404 405 iovec iov[] = { 406 { NULL, 0 }, 407 { &route, sizeof(route) }, 408 { &RTATTR_TABLE, sizeof(RTATTR_TABLE) }, 409 { &table, sizeof(table) }, 410 { &rtaDst, sizeof(rtaDst) }, 411 { rawAddress, static_cast<size_t>(rawLength) }, 412 { &RTATTR_OIF, interface != OIF_NONE ? sizeof(RTATTR_OIF) : 0 }, 413 { &ifindex, interface != OIF_NONE ? sizeof(ifindex) : 0 }, 414 { &rtaGateway, nexthop ? sizeof(rtaGateway) : 0 }, 415 { rawNexthop, nexthop ? static_cast<size_t>(rawLength) : 0 }, 416 }; 417 418 uint16_t flags = (action == RTM_NEWROUTE) ? NETLINK_CREATE_REQUEST_FLAGS : 419 NETLINK_REQUEST_FLAGS; 420 return sendNetlinkRequest(action, flags, iov, ARRAY_SIZE(iov)); 421} 422 423// An iptables rule to mark incoming packets on a network with the netId of the network. 424// 425// This is so that the kernel can: 426// + Use the right fwmark for (and thus correctly route) replies (e.g.: TCP RST, ICMP errors, ping 427// replies, SYN-ACKs, etc). 428// + Mark sockets that accept connections from this interface so that the connection stays on the 429// same interface. 430WARN_UNUSED_RESULT int modifyIncomingPacketMark(unsigned netId, const char* interface, 431 Permission permission, bool add) { 432 Fwmark fwmark; 433 434 fwmark.netId = netId; 435 fwmark.explicitlySelected = true; 436 fwmark.protectedFromVpn = true; 437 fwmark.permission = permission; 438 439 char markString[UINT32_HEX_STRLEN]; 440 snprintf(markString, sizeof(markString), "0x%x", fwmark.intValue); 441 442 if (execIptables(V4V6, "-t", "mangle", add ? "-A" : "-D", "INPUT", "-i", interface, "-j", 443 "MARK", "--set-mark", markString, NULL)) { 444 ALOGE("failed to change iptables rule that sets incoming packet mark"); 445 return -EREMOTEIO; 446 } 447 448 return 0; 449} 450 451// A rule to route responses to the local network forwarded via the VPN. 452// 453// When a VPN is in effect, packets from the local network to upstream networks are forwarded into 454// the VPN's tunnel interface. When the VPN forwards the responses, they emerge out of the tunnel. 455WARN_UNUSED_RESULT int modifyVpnOutputToLocalRule(const char* vpnInterface, bool add) { 456 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_VPN_OUTPUT_TO_LOCAL, 457 ROUTE_TABLE_LOCAL_NETWORK, MARK_UNSET, MARK_UNSET, vpnInterface, OIF_NONE, 458 INVALID_UID, INVALID_UID); 459} 460 461// A rule to route all traffic from a given set of UIDs to go over the VPN. 462// 463// Notice that this rule doesn't use the netId. I.e., no matter what netId the user's socket may 464// have, if they are subject to this VPN, their traffic has to go through it. Allows the traffic to 465// bypass the VPN if the protectedFromVpn bit is set. 466WARN_UNUSED_RESULT int modifyVpnUidRangeRule(uint32_t table, uid_t uidStart, uid_t uidEnd, 467 bool secure, bool add) { 468 Fwmark fwmark; 469 Fwmark mask; 470 471 fwmark.protectedFromVpn = false; 472 mask.protectedFromVpn = true; 473 474 uint32_t priority; 475 476 if (secure) { 477 priority = RULE_PRIORITY_SECURE_VPN; 478 } else { 479 priority = RULE_PRIORITY_BYPASSABLE_VPN; 480 481 fwmark.explicitlySelected = false; 482 mask.explicitlySelected = true; 483 } 484 485 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, priority, table, fwmark.intValue, 486 mask.intValue, IIF_NONE, OIF_NONE, uidStart, uidEnd); 487} 488 489// A rule to allow system apps to send traffic over this VPN even if they are not part of the target 490// set of UIDs. 491// 492// This is needed for DnsProxyListener to correctly resolve a request for a user who is in the 493// target set, but where the DnsProxyListener itself is not. 494WARN_UNUSED_RESULT int modifyVpnSystemPermissionRule(unsigned netId, uint32_t table, bool secure, 495 bool add) { 496 Fwmark fwmark; 497 Fwmark mask; 498 499 fwmark.netId = netId; 500 mask.netId = FWMARK_NET_ID_MASK; 501 502 fwmark.permission = PERMISSION_SYSTEM; 503 mask.permission = PERMISSION_SYSTEM; 504 505 uint32_t priority = secure ? RULE_PRIORITY_SECURE_VPN : RULE_PRIORITY_BYPASSABLE_VPN; 506 507 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, priority, table, fwmark.intValue, 508 mask.intValue); 509} 510 511// A rule to route traffic based on an explicitly chosen network. 512// 513// Supports apps that use the multinetwork APIs to restrict their traffic to a network. 514// 515// Even though we check permissions at the time we set a netId into the fwmark of a socket, we need 516// to check it again in the rules here, because a network's permissions may have been updated via 517// modifyNetworkPermission(). 518WARN_UNUSED_RESULT int modifyExplicitNetworkRule(unsigned netId, uint32_t table, 519 Permission permission, uid_t uidStart, 520 uid_t uidEnd, bool add) { 521 Fwmark fwmark; 522 Fwmark mask; 523 524 fwmark.netId = netId; 525 mask.netId = FWMARK_NET_ID_MASK; 526 527 fwmark.explicitlySelected = true; 528 mask.explicitlySelected = true; 529 530 fwmark.permission = permission; 531 mask.permission = permission; 532 533 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_EXPLICIT_NETWORK, table, 534 fwmark.intValue, mask.intValue, IIF_NONE, OIF_NONE, uidStart, uidEnd); 535} 536 537// A rule to route traffic based on a chosen outgoing interface. 538// 539// Supports apps that use SO_BINDTODEVICE or IP_PKTINFO options and the kernel that already knows 540// the outgoing interface (typically for link-local communications). 541WARN_UNUSED_RESULT int modifyOutputInterfaceRule(const char* interface, uint32_t table, 542 Permission permission, uid_t uidStart, 543 uid_t uidEnd, bool add) { 544 Fwmark fwmark; 545 Fwmark mask; 546 547 fwmark.permission = permission; 548 mask.permission = permission; 549 550 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_OUTPUT_INTERFACE, table, 551 fwmark.intValue, mask.intValue, IIF_NONE, interface, uidStart, uidEnd); 552} 553 554// A rule to route traffic based on the chosen network. 555// 556// This is for sockets that have not explicitly requested a particular network, but have been 557// bound to one when they called connect(). This ensures that sockets connected on a particular 558// network stay on that network even if the default network changes. 559WARN_UNUSED_RESULT int modifyImplicitNetworkRule(unsigned netId, uint32_t table, 560 Permission permission, bool add) { 561 Fwmark fwmark; 562 Fwmark mask; 563 564 fwmark.netId = netId; 565 mask.netId = FWMARK_NET_ID_MASK; 566 567 fwmark.explicitlySelected = false; 568 mask.explicitlySelected = true; 569 570 fwmark.permission = permission; 571 mask.permission = permission; 572 573 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_IMPLICIT_NETWORK, table, 574 fwmark.intValue, mask.intValue); 575} 576 577// A rule to enable split tunnel VPNs. 578// 579// If a packet with a VPN's netId doesn't find a route in the VPN's routing table, it's allowed to 580// go over the default network, provided it wasn't explicitly restricted to the VPN and has the 581// permissions required by the default network. 582WARN_UNUSED_RESULT int modifyVpnFallthroughRule(uint16_t action, unsigned vpnNetId, 583 const char* physicalInterface, 584 Permission permission) { 585 uint32_t table = getRouteTableForInterface(physicalInterface); 586 if (table == RT_TABLE_UNSPEC) { 587 return -ESRCH; 588 } 589 590 Fwmark fwmark; 591 Fwmark mask; 592 593 fwmark.netId = vpnNetId; 594 mask.netId = FWMARK_NET_ID_MASK; 595 596 fwmark.explicitlySelected = false; 597 mask.explicitlySelected = true; 598 599 fwmark.permission = permission; 600 mask.permission = permission; 601 602 return modifyIpRule(action, RULE_PRIORITY_VPN_FALLTHROUGH, table, fwmark.intValue, 603 mask.intValue); 604} 605 606// Add rules to allow legacy routes added through the requestRouteToHost() API. 607WARN_UNUSED_RESULT int addLegacyRouteRules() { 608 Fwmark fwmark; 609 Fwmark mask; 610 611 fwmark.explicitlySelected = false; 612 mask.explicitlySelected = true; 613 614 // Rules to allow legacy routes to override the default network. 615 if (int ret = modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LEGACY_SYSTEM, ROUTE_TABLE_LEGACY_SYSTEM, 616 fwmark.intValue, mask.intValue)) { 617 return ret; 618 } 619 if (int ret = modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LEGACY_NETWORK, 620 ROUTE_TABLE_LEGACY_NETWORK, fwmark.intValue, mask.intValue)) { 621 return ret; 622 } 623 624 fwmark.permission = PERMISSION_SYSTEM; 625 mask.permission = PERMISSION_SYSTEM; 626 627 // A rule to allow legacy routes from system apps to override VPNs. 628 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_VPN_OVERRIDE_SYSTEM, ROUTE_TABLE_LEGACY_SYSTEM, 629 fwmark.intValue, mask.intValue); 630} 631 632// Add rules to lookup the local network when specified explicitly or otherwise. 633WARN_UNUSED_RESULT int addLocalNetworkRules(unsigned localNetId) { 634 if (int ret = modifyExplicitNetworkRule(localNetId, ROUTE_TABLE_LOCAL_NETWORK, PERMISSION_NONE, 635 INVALID_UID, INVALID_UID, ACTION_ADD)) { 636 return ret; 637 } 638 639 Fwmark fwmark; 640 Fwmark mask; 641 642 fwmark.explicitlySelected = false; 643 mask.explicitlySelected = true; 644 645 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LOCAL_NETWORK, ROUTE_TABLE_LOCAL_NETWORK, 646 fwmark.intValue, mask.intValue); 647} 648 649// Add a new rule to look up the 'main' table, with the same selectors as the "default network" 650// rule, but with a lower priority. We will never create routes in the main table; it should only be 651// used for directly-connected routes implicitly created by the kernel when adding IP addresses. 652// This is necessary, for example, when adding a route through a directly-connected gateway: in 653// order to add the route, there must already be a directly-connected route that covers the gateway. 654WARN_UNUSED_RESULT int addDirectlyConnectedRule() { 655 Fwmark fwmark; 656 Fwmark mask; 657 658 fwmark.netId = NETID_UNSET; 659 mask.netId = FWMARK_NET_ID_MASK; 660 661 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_DIRECTLY_CONNECTED, RT_TABLE_MAIN, 662 fwmark.intValue, mask.intValue, IIF_NONE, OIF_NONE, UID_ROOT, UID_ROOT); 663} 664 665// Add an explicit unreachable rule close to the end of the prioriy list to make it clear that 666// relying on the kernel-default "from all lookup main" rule at priority 32766 is not intended 667// behaviour. We do flush the kernel-default rules at startup, but having an explicit unreachable 668// rule will hopefully make things even clearer. 669WARN_UNUSED_RESULT int addUnreachableRule() { 670 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_UNREACHABLE, RT_TABLE_UNSPEC, MARK_UNSET, 671 MARK_UNSET); 672} 673 674WARN_UNUSED_RESULT int modifyLocalNetwork(unsigned netId, const char* interface, bool add) { 675 if (int ret = modifyIncomingPacketMark(netId, interface, PERMISSION_NONE, add)) { 676 return ret; 677 } 678 return modifyOutputInterfaceRule(interface, ROUTE_TABLE_LOCAL_NETWORK, PERMISSION_NONE, 679 INVALID_UID, INVALID_UID, add); 680} 681 682WARN_UNUSED_RESULT int modifyPhysicalNetwork(unsigned netId, const char* interface, 683 Permission permission, bool add) { 684 uint32_t table = getRouteTableForInterface(interface); 685 if (table == RT_TABLE_UNSPEC) { 686 return -ESRCH; 687 } 688 689 if (int ret = modifyIncomingPacketMark(netId, interface, permission, add)) { 690 return ret; 691 } 692 if (int ret = modifyExplicitNetworkRule(netId, table, permission, INVALID_UID, INVALID_UID, 693 add)) { 694 return ret; 695 } 696 if (int ret = modifyOutputInterfaceRule(interface, table, permission, INVALID_UID, INVALID_UID, 697 add)) { 698 return ret; 699 } 700 return modifyImplicitNetworkRule(netId, table, permission, add); 701} 702 703WARN_UNUSED_RESULT int modifyVirtualNetwork(unsigned netId, const char* interface, 704 const UidRanges& uidRanges, bool secure, bool add, 705 bool modifyNonUidBasedRules) { 706 uint32_t table = getRouteTableForInterface(interface); 707 if (table == RT_TABLE_UNSPEC) { 708 return -ESRCH; 709 } 710 711 for (const UidRanges::Range& range : uidRanges.getRanges()) { 712 if (int ret = modifyVpnUidRangeRule(table, range.first, range.second, secure, add)) { 713 return ret; 714 } 715 if (int ret = modifyExplicitNetworkRule(netId, table, PERMISSION_NONE, range.first, 716 range.second, add)) { 717 return ret; 718 } 719 if (int ret = modifyOutputInterfaceRule(interface, table, PERMISSION_NONE, range.first, 720 range.second, add)) { 721 return ret; 722 } 723 } 724 725 if (modifyNonUidBasedRules) { 726 if (int ret = modifyIncomingPacketMark(netId, interface, PERMISSION_NONE, add)) { 727 return ret; 728 } 729 if (int ret = modifyVpnOutputToLocalRule(interface, add)) { 730 return ret; 731 } 732 if (int ret = modifyVpnSystemPermissionRule(netId, table, secure, add)) { 733 return ret; 734 } 735 return modifyExplicitNetworkRule(netId, table, PERMISSION_NONE, UID_ROOT, UID_ROOT, add); 736 } 737 738 return 0; 739} 740 741WARN_UNUSED_RESULT int modifyDefaultNetwork(uint16_t action, const char* interface, 742 Permission permission) { 743 uint32_t table = getRouteTableForInterface(interface); 744 if (table == RT_TABLE_UNSPEC) { 745 return -ESRCH; 746 } 747 748 Fwmark fwmark; 749 Fwmark mask; 750 751 fwmark.netId = NETID_UNSET; 752 mask.netId = FWMARK_NET_ID_MASK; 753 754 fwmark.permission = permission; 755 mask.permission = permission; 756 757 return modifyIpRule(action, RULE_PRIORITY_DEFAULT_NETWORK, table, fwmark.intValue, 758 mask.intValue); 759} 760 761WARN_UNUSED_RESULT int modifyTetheredNetwork(uint16_t action, const char* inputInterface, 762 const char* outputInterface) { 763 uint32_t table = getRouteTableForInterface(outputInterface); 764 if (table == RT_TABLE_UNSPEC) { 765 return -ESRCH; 766 } 767 768 return modifyIpRule(action, RULE_PRIORITY_TETHERING, table, MARK_UNSET, MARK_UNSET, 769 inputInterface, OIF_NONE, INVALID_UID, INVALID_UID); 770} 771 772// Returns 0 on success or negative errno on failure. 773WARN_UNUSED_RESULT int flushRules() { 774 for (size_t i = 0; i < ARRAY_SIZE(IP_VERSIONS); ++i) { 775 const char* argv[] = { 776 IP_PATH, 777 IP_VERSIONS[i], 778 "rule", 779 "flush", 780 }; 781 if (android_fork_execvp(ARRAY_SIZE(argv), const_cast<char**>(argv), NULL, false, false)) { 782 ALOGE("failed to flush rules"); 783 return -EREMOTEIO; 784 } 785 } 786 return 0; 787} 788 789// Adds or removes an IPv4 or IPv6 route to the specified table and, if it's a directly-connected 790// route, to the main table as well. 791// Returns 0 on success or negative errno on failure. 792WARN_UNUSED_RESULT int modifyRoute(uint16_t action, const char* interface, const char* destination, 793 const char* nexthop, RouteController::TableType tableType) { 794 uint32_t table; 795 switch (tableType) { 796 case RouteController::INTERFACE: { 797 table = getRouteTableForInterface(interface); 798 if (table == RT_TABLE_UNSPEC) { 799 return -ESRCH; 800 } 801 break; 802 } 803 case RouteController::LOCAL_NETWORK: { 804 table = ROUTE_TABLE_LOCAL_NETWORK; 805 break; 806 } 807 case RouteController::LEGACY_NETWORK: { 808 table = ROUTE_TABLE_LEGACY_NETWORK; 809 break; 810 } 811 case RouteController::LEGACY_SYSTEM: { 812 table = ROUTE_TABLE_LEGACY_SYSTEM; 813 break; 814 } 815 } 816 817 int ret = modifyIpRoute(action, table, interface, destination, nexthop); 818 // Trying to add a route that already exists shouldn't cause an error. 819 if (ret && !(action == RTM_NEWROUTE && ret == -EEXIST)) { 820 return ret; 821 } 822 823 return 0; 824} 825 826// Returns 0 on success or negative errno on failure. 827WARN_UNUSED_RESULT int flushRoutes(const char* interface) { 828 uint32_t table = getRouteTableForInterface(interface); 829 if (table == RT_TABLE_UNSPEC) { 830 return -ESRCH; 831 } 832 833 char tableString[UINT32_STRLEN]; 834 snprintf(tableString, sizeof(tableString), "%u", table); 835 836 int ret = 0; 837 for (size_t i = 0; i < ARRAY_SIZE(IP_VERSIONS); ++i) { 838 const char* argv[] = { 839 IP_PATH, 840 IP_VERSIONS[i], 841 "route", 842 "flush", 843 "table", 844 tableString, 845 }; 846 847 // A flush works by dumping routes and deleting each route as it's returned, and it can 848 // fail if something else deletes the route between the dump and the delete. This can 849 // happen, for example, if an interface goes down while we're trying to flush its routes. 850 // So try multiple times and only return an error if the last attempt fails. 851 // 852 // TODO: replace this with our own netlink code. 853 unsigned attempts = 0; 854 int err; 855 do { 856 err = android_fork_execvp(ARRAY_SIZE(argv), const_cast<char**>(argv), 857 NULL, false, false); 858 ++attempts; 859 } while (err != 0 && attempts < ROUTE_FLUSH_ATTEMPTS); 860 if (err) { 861 ALOGE("failed to flush %s routes in table %s after %d attempts", 862 IP_VERSIONS[i], tableString, attempts); 863 ret = -EREMOTEIO; 864 } 865 } 866 867 // If we failed to flush routes, the caller may elect to keep this interface around, so keep 868 // track of its name. 869 if (!ret) { 870 interfaceToTable.erase(interface); 871 } 872 873 return ret; 874} 875 876} // namespace 877 878int RouteController::Init(unsigned localNetId) { 879 if (int ret = flushRules()) { 880 return ret; 881 } 882 if (int ret = addLegacyRouteRules()) { 883 return ret; 884 } 885 if (int ret = addLocalNetworkRules(localNetId)) { 886 return ret; 887 } 888 if (int ret = addDirectlyConnectedRule()) { 889 return ret; 890 } 891 if (int ret = addUnreachableRule()) { 892 return ret; 893 } 894 updateTableNamesFile(); 895 return 0; 896} 897 898int RouteController::addInterfaceToLocalNetwork(unsigned netId, const char* interface) { 899 return modifyLocalNetwork(netId, interface, ACTION_ADD); 900} 901 902int RouteController::removeInterfaceFromLocalNetwork(unsigned netId, const char* interface) { 903 return modifyLocalNetwork(netId, interface, ACTION_DEL); 904} 905 906int RouteController::addInterfaceToPhysicalNetwork(unsigned netId, const char* interface, 907 Permission permission) { 908 if (int ret = modifyPhysicalNetwork(netId, interface, permission, ACTION_ADD)) { 909 return ret; 910 } 911 updateTableNamesFile(); 912 return 0; 913} 914 915int RouteController::removeInterfaceFromPhysicalNetwork(unsigned netId, const char* interface, 916 Permission permission) { 917 if (int ret = modifyPhysicalNetwork(netId, interface, permission, ACTION_DEL)) { 918 return ret; 919 } 920 if (int ret = flushRoutes(interface)) { 921 return ret; 922 } 923 updateTableNamesFile(); 924 return 0; 925} 926 927int RouteController::addInterfaceToVirtualNetwork(unsigned netId, const char* interface, 928 bool secure, const UidRanges& uidRanges) { 929 if (int ret = modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_ADD, 930 MODIFY_NON_UID_BASED_RULES)) { 931 return ret; 932 } 933 updateTableNamesFile(); 934 return 0; 935} 936 937int RouteController::removeInterfaceFromVirtualNetwork(unsigned netId, const char* interface, 938 bool secure, const UidRanges& uidRanges) { 939 if (int ret = modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_DEL, 940 MODIFY_NON_UID_BASED_RULES)) { 941 return ret; 942 } 943 if (int ret = flushRoutes(interface)) { 944 return ret; 945 } 946 updateTableNamesFile(); 947 return 0; 948} 949 950int RouteController::modifyPhysicalNetworkPermission(unsigned netId, const char* interface, 951 Permission oldPermission, 952 Permission newPermission) { 953 // Add the new rules before deleting the old ones, to avoid race conditions. 954 if (int ret = modifyPhysicalNetwork(netId, interface, newPermission, ACTION_ADD)) { 955 return ret; 956 } 957 return modifyPhysicalNetwork(netId, interface, oldPermission, ACTION_DEL); 958} 959 960int RouteController::addUsersToVirtualNetwork(unsigned netId, const char* interface, bool secure, 961 const UidRanges& uidRanges) { 962 return modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_ADD, 963 !MODIFY_NON_UID_BASED_RULES); 964} 965 966int RouteController::removeUsersFromVirtualNetwork(unsigned netId, const char* interface, 967 bool secure, const UidRanges& uidRanges) { 968 return modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_DEL, 969 !MODIFY_NON_UID_BASED_RULES); 970} 971 972int RouteController::addInterfaceToDefaultNetwork(const char* interface, Permission permission) { 973 return modifyDefaultNetwork(RTM_NEWRULE, interface, permission); 974} 975 976int RouteController::removeInterfaceFromDefaultNetwork(const char* interface, 977 Permission permission) { 978 return modifyDefaultNetwork(RTM_DELRULE, interface, permission); 979} 980 981int RouteController::addRoute(const char* interface, const char* destination, const char* nexthop, 982 TableType tableType) { 983 return modifyRoute(RTM_NEWROUTE, interface, destination, nexthop, tableType); 984} 985 986int RouteController::removeRoute(const char* interface, const char* destination, 987 const char* nexthop, TableType tableType) { 988 return modifyRoute(RTM_DELROUTE, interface, destination, nexthop, tableType); 989} 990 991int RouteController::enableTethering(const char* inputInterface, const char* outputInterface) { 992 return modifyTetheredNetwork(RTM_NEWRULE, inputInterface, outputInterface); 993} 994 995int RouteController::disableTethering(const char* inputInterface, const char* outputInterface) { 996 return modifyTetheredNetwork(RTM_DELRULE, inputInterface, outputInterface); 997} 998 999int RouteController::addVirtualNetworkFallthrough(unsigned vpnNetId, const char* physicalInterface, 1000 Permission permission) { 1001 return modifyVpnFallthroughRule(RTM_NEWRULE, vpnNetId, physicalInterface, permission); 1002} 1003 1004int RouteController::removeVirtualNetworkFallthrough(unsigned vpnNetId, 1005 const char* physicalInterface, 1006 Permission permission) { 1007 return modifyVpnFallthroughRule(RTM_DELRULE, vpnNetId, physicalInterface, permission); 1008} 1009