RouteController.cpp revision bb40d5198943df0e2a88514cbcd82a23a5b0cd97
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; 39// const uint32_t RULE_PRIORITY_VPN_OVERRIDE_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; 45// const uint32_t RULE_PRIORITY_LOCAL_NETWORK = 17000; 46// const uint32_t RULE_PRIORITY_TETHERING = 18000; 47const uint32_t RULE_PRIORITY_IMPLICIT_NETWORK = 19000; 48// const uint32_t RULE_PRIORITY_BYPASSABLE_VPN = 20000; 49// const 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 = 24000; 53 54const uint32_t ROUTE_TABLE_LEGACY_NETWORK = 98; 55const uint32_t ROUTE_TABLE_LEGACY_SYSTEM = 99; 56 57const char* const ROUTE_TABLE_NAME_LEGACY_NETWORK = "legacy_network"; 58const char* const ROUTE_TABLE_NAME_LEGACY_SYSTEM = "legacy_system"; 59 60const char* const ROUTE_TABLE_NAME_LOCAL = "local"; 61const char* const ROUTE_TABLE_NAME_MAIN = "main"; 62 63// TODO: These values aren't defined by the Linux kernel, because our UID routing changes are not 64// upstream (yet?), so we can't just pick them up from kernel headers. When (if?) the changes make 65// it upstream, we'll remove this and rely on the kernel header values. For now, add a static assert 66// that will warn us if upstream has given these values some other meaning. 67const uint16_t FRA_UID_START = 18; 68const uint16_t FRA_UID_END = 19; 69static_assert(FRA_UID_START > FRA_MAX, 70 "Android-specific FRA_UID_{START,END} values also assigned in Linux uapi. " 71 "Check that these values match what the kernel does and then update this assertion."); 72 73const uint16_t NETLINK_REQUEST_FLAGS = NLM_F_REQUEST | NLM_F_ACK; 74const uint16_t NETLINK_CREATE_REQUEST_FLAGS = NETLINK_REQUEST_FLAGS | NLM_F_CREATE | NLM_F_EXCL; 75 76const sockaddr_nl NETLINK_ADDRESS = {AF_NETLINK, 0, 0, 0}; 77 78const uint8_t AF_FAMILIES[] = {AF_INET, AF_INET6}; 79 80const char* const IP_VERSIONS[] = {"-4", "-6"}; 81 82const uid_t UID_ROOT = 0; 83const char* const OIF_NONE = NULL; 84const bool ACTION_ADD = true; 85const bool ACTION_DEL = false; 86const bool MODIFY_NON_UID_BASED_RULES = true; 87 88const char* const RT_TABLES_PATH = "/data/misc/net/rt_tables"; 89const int RT_TABLES_FLAGS = O_CREAT | O_TRUNC | O_WRONLY | O_NOFOLLOW | O_CLOEXEC; 90const mode_t RT_TABLES_MODE = S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH; // mode 0644, rw-r--r-- 91 92// Avoids "non-constant-expression cannot be narrowed from type 'unsigned int' to 'unsigned short'" 93// warnings when using RTA_LENGTH(x) inside static initializers (even when x is already uint16_t). 94constexpr uint16_t U16_RTA_LENGTH(uint16_t x) { 95 return RTA_LENGTH(x); 96} 97 98// These are practically const, but can't be declared so, because they are used to initialize 99// non-const pointers ("void* iov_base") in iovec arrays. 100rtattr FRATTR_PRIORITY = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_PRIORITY }; 101rtattr FRATTR_TABLE = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_TABLE }; 102rtattr FRATTR_FWMARK = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_FWMARK }; 103rtattr FRATTR_FWMASK = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_FWMASK }; 104rtattr FRATTR_UID_START = { U16_RTA_LENGTH(sizeof(uid_t)), FRA_UID_START }; 105rtattr FRATTR_UID_END = { U16_RTA_LENGTH(sizeof(uid_t)), FRA_UID_END }; 106 107rtattr RTATTR_TABLE = { U16_RTA_LENGTH(sizeof(uint32_t)), RTA_TABLE }; 108rtattr RTATTR_OIF = { U16_RTA_LENGTH(sizeof(uint32_t)), RTA_OIF }; 109 110uint8_t PADDING_BUFFER[RTA_ALIGNTO] = {0, 0, 0, 0}; 111 112// END CONSTANTS ---------------------------------------------------------------------------------- 113 114// No locks needed because RouteController is accessed only from one thread (in CommandListener). 115std::map<std::string, uint32_t> interfaceToTable; 116 117uint32_t getRouteTableForInterface(const char* interface) { 118 uint32_t index = if_nametoindex(interface); 119 if (index) { 120 index += RouteController::ROUTE_TABLE_OFFSET_FROM_INDEX; 121 interfaceToTable[interface] = index; 122 return index; 123 } 124 // If the interface goes away if_nametoindex() will return 0 but we still need to know 125 // the index so we can remove the rules and routes. 126 auto iter = interfaceToTable.find(interface); 127 if (iter == interfaceToTable.end()) { 128 ALOGE("cannot find interface %s", interface); 129 return RT_TABLE_UNSPEC; 130 } 131 return iter->second; 132} 133 134void addTableName(uint32_t table, const std::string& name, std::string* contents) { 135 char tableString[UINT32_STRLEN]; 136 snprintf(tableString, sizeof(tableString), "%u", table); 137 *contents += tableString; 138 *contents += " "; 139 *contents += name; 140 *contents += "\n"; 141} 142 143// Doesn't return success/failure as the file is optional; it's okay if we fail to update it. 144void updateTableNamesFile() { 145 std::string contents; 146 147 addTableName(RT_TABLE_LOCAL, ROUTE_TABLE_NAME_LOCAL, &contents); 148 addTableName(RT_TABLE_MAIN, ROUTE_TABLE_NAME_MAIN, &contents); 149 150 addTableName(ROUTE_TABLE_LEGACY_NETWORK, ROUTE_TABLE_NAME_LEGACY_NETWORK, &contents); 151 addTableName(ROUTE_TABLE_LEGACY_SYSTEM, ROUTE_TABLE_NAME_LEGACY_SYSTEM, &contents); 152 153 for (const auto& entry : interfaceToTable) { 154 addTableName(entry.second, entry.first, &contents); 155 } 156 157 int fd = open(RT_TABLES_PATH, RT_TABLES_FLAGS, RT_TABLES_MODE); 158 if (fd == -1) { 159 ALOGE("failed to create %s (%s)", RT_TABLES_PATH, strerror(errno)); 160 return; 161 } 162 // File creation is affected by umask, so make sure the right mode bits are set. 163 if (fchmod(fd, RT_TABLES_MODE) == -1) { 164 ALOGE("failed to set mode 0%o on %s (%s)", RT_TABLES_MODE, RT_TABLES_PATH, strerror(errno)); 165 } 166 ssize_t bytesWritten = write(fd, contents.data(), contents.size()); 167 if (bytesWritten != static_cast<ssize_t>(contents.size())) { 168 ALOGE("failed to write to %s (%zd vs %zu bytes) (%s)", RT_TABLES_PATH, bytesWritten, 169 contents.size(), strerror(errno)); 170 } 171 close(fd); 172} 173 174// Sends a netlink request and expects an ack. 175// |iov| is an array of struct iovec that contains the netlink message payload. 176// The netlink header is generated by this function based on |action| and |flags|. 177// Returns -errno if there was an error or if the kernel reported an error. 178WARN_UNUSED_RESULT int sendNetlinkRequest(uint16_t action, uint16_t flags, iovec* iov, int iovlen) { 179 nlmsghdr nlmsg = { 180 .nlmsg_type = action, 181 .nlmsg_flags = flags, 182 }; 183 iov[0].iov_base = &nlmsg; 184 iov[0].iov_len = sizeof(nlmsg); 185 for (int i = 0; i < iovlen; ++i) { 186 nlmsg.nlmsg_len += iov[i].iov_len; 187 } 188 189 int ret; 190 struct { 191 nlmsghdr msg; 192 nlmsgerr err; 193 } response; 194 195 int sock = socket(AF_NETLINK, SOCK_DGRAM, NETLINK_ROUTE); 196 if (sock != -1 && 197 connect(sock, reinterpret_cast<const sockaddr*>(&NETLINK_ADDRESS), 198 sizeof(NETLINK_ADDRESS)) != -1 && 199 writev(sock, iov, iovlen) != -1 && 200 (ret = recv(sock, &response, sizeof(response), 0)) != -1) { 201 if (ret == sizeof(response)) { 202 ret = response.err.error; // Netlink errors are negative errno. 203 if (ret) { 204 ALOGE("netlink response contains error (%s)", strerror(-ret)); 205 } 206 } else { 207 ALOGE("bad netlink response message size (%d != %zu)", ret, sizeof(response)); 208 ret = -EBADMSG; 209 } 210 } else { 211 ALOGE("netlink socket/connect/writev/recv failed (%s)", strerror(errno)); 212 ret = -errno; 213 } 214 215 if (sock != -1) { 216 close(sock); 217 } 218 219 return ret; 220} 221 222// Adds or removes a routing rule for IPv4 and IPv6. 223// 224// + If |table| is non-zero, the rule points at the specified routing table. Otherwise, the rule 225// returns ENETUNREACH. 226// + If |mask| is non-zero, the rule matches the specified fwmark and mask. Otherwise, |fwmark| is 227// ignored. 228// + If |interface| is non-NULL, the rule matches the specified outgoing interface. 229// 230// Returns 0 on success or negative errno on failure. 231WARN_UNUSED_RESULT int modifyIpRule(uint16_t action, uint32_t priority, uint32_t table, 232 uint32_t fwmark, uint32_t mask, const char* interface, 233 uid_t uidStart, uid_t uidEnd) { 234 // Ensure that if you set a bit in the fwmark, it's not being ignored by the mask. 235 if (fwmark & ~mask) { 236 ALOGE("mask 0x%x does not select all the bits set in fwmark 0x%x", mask, fwmark); 237 return -ERANGE; 238 } 239 240 // The interface name must include exactly one terminating NULL and be properly padded, or older 241 // kernels will refuse to delete rules. 242 uint16_t paddingLength = 0; 243 size_t interfaceLength = 0; 244 char oifname[IFNAMSIZ]; 245 if (interface != OIF_NONE) { 246 interfaceLength = strlcpy(oifname, interface, IFNAMSIZ) + 1; 247 if (interfaceLength > IFNAMSIZ) { 248 ALOGE("interface name too long (%zu > %u)", interfaceLength, IFNAMSIZ); 249 return -ENAMETOOLONG; 250 } 251 paddingLength = RTA_SPACE(interfaceLength) - RTA_LENGTH(interfaceLength); 252 } 253 254 // Either both start and end UID must be specified, or neither. 255 if ((uidStart == INVALID_UID) != (uidEnd == INVALID_UID)) { 256 ALOGE("incompatible start and end UIDs (%u vs %u)", uidStart, uidEnd); 257 return -EUSERS; 258 } 259 bool isUidRule = (uidStart != INVALID_UID); 260 261 // Assemble a rule request and put it in an array of iovec structures. 262 fib_rule_hdr rule = { 263 .action = static_cast<uint8_t>(table != RT_TABLE_UNSPEC ? FR_ACT_TO_TBL : 264 FR_ACT_UNREACHABLE), 265 }; 266 267 rtattr fraOifname = { U16_RTA_LENGTH(interfaceLength), FRA_OIFNAME }; 268 269 iovec iov[] = { 270 { NULL, 0 }, 271 { &rule, sizeof(rule) }, 272 { &FRATTR_PRIORITY, sizeof(FRATTR_PRIORITY) }, 273 { &priority, sizeof(priority) }, 274 { &FRATTR_TABLE, table != RT_TABLE_UNSPEC ? sizeof(FRATTR_TABLE) : 0 }, 275 { &table, table != RT_TABLE_UNSPEC ? sizeof(table) : 0 }, 276 { &FRATTR_FWMARK, mask ? sizeof(FRATTR_FWMARK) : 0 }, 277 { &fwmark, mask ? sizeof(fwmark) : 0 }, 278 { &FRATTR_FWMASK, mask ? sizeof(FRATTR_FWMASK) : 0 }, 279 { &mask, mask ? sizeof(mask) : 0 }, 280 { &FRATTR_UID_START, isUidRule ? sizeof(FRATTR_UID_START) : 0 }, 281 { &uidStart, isUidRule ? sizeof(uidStart) : 0 }, 282 { &FRATTR_UID_END, isUidRule ? sizeof(FRATTR_UID_END) : 0 }, 283 { &uidEnd, isUidRule ? sizeof(uidEnd) : 0 }, 284 { &fraOifname, interface != OIF_NONE ? sizeof(fraOifname) : 0 }, 285 { oifname, interfaceLength }, 286 { PADDING_BUFFER, paddingLength }, 287 }; 288 289 uint16_t flags = (action == RTM_NEWRULE) ? NETLINK_CREATE_REQUEST_FLAGS : NETLINK_REQUEST_FLAGS; 290 for (size_t i = 0; i < ARRAY_SIZE(AF_FAMILIES); ++i) { 291 rule.family = AF_FAMILIES[i]; 292 if (int ret = sendNetlinkRequest(action, flags, iov, ARRAY_SIZE(iov))) { 293 return ret; 294 } 295 } 296 297 return 0; 298} 299 300// Adds or deletes an IPv4 or IPv6 route. 301// Returns 0 on success or negative errno on failure. 302WARN_UNUSED_RESULT int modifyIpRoute(uint16_t action, uint32_t table, const char* interface, 303 const char* destination, const char* nexthop) { 304 // At least the destination must be non-null. 305 if (!destination) { 306 ALOGE("null destination"); 307 return -EFAULT; 308 } 309 310 // Parse the prefix. 311 uint8_t rawAddress[sizeof(in6_addr)]; 312 uint8_t family; 313 uint8_t prefixLength; 314 int rawLength = parsePrefix(destination, &family, rawAddress, sizeof(rawAddress), 315 &prefixLength); 316 if (rawLength < 0) { 317 ALOGE("parsePrefix failed for destination %s (%s)", destination, strerror(-rawLength)); 318 return rawLength; 319 } 320 321 if (static_cast<size_t>(rawLength) > sizeof(rawAddress)) { 322 ALOGE("impossible! address too long (%d vs %zu)", rawLength, sizeof(rawAddress)); 323 return -ENOBUFS; // Cannot happen; parsePrefix only supports IPv4 and IPv6. 324 } 325 326 // If an interface was specified, find the ifindex. 327 uint32_t ifindex; 328 if (interface != OIF_NONE) { 329 ifindex = if_nametoindex(interface); 330 if (!ifindex) { 331 ALOGE("cannot find interface %s", interface); 332 return -ENODEV; 333 } 334 } 335 336 // If a nexthop was specified, parse it as the same family as the prefix. 337 uint8_t rawNexthop[sizeof(in6_addr)]; 338 if (nexthop && inet_pton(family, nexthop, rawNexthop) <= 0) { 339 ALOGE("inet_pton failed for nexthop %s", nexthop); 340 return -EINVAL; 341 } 342 343 // Assemble a rtmsg and put it in an array of iovec structures. 344 rtmsg route = { 345 .rtm_protocol = RTPROT_STATIC, 346 .rtm_type = RTN_UNICAST, 347 .rtm_family = family, 348 .rtm_dst_len = prefixLength, 349 }; 350 351 rtattr rtaDst = { U16_RTA_LENGTH(rawLength), RTA_DST }; 352 rtattr rtaGateway = { U16_RTA_LENGTH(rawLength), RTA_GATEWAY }; 353 354 iovec iov[] = { 355 { NULL, 0 }, 356 { &route, sizeof(route) }, 357 { &RTATTR_TABLE, sizeof(RTATTR_TABLE) }, 358 { &table, sizeof(table) }, 359 { &rtaDst, sizeof(rtaDst) }, 360 { rawAddress, static_cast<size_t>(rawLength) }, 361 { &RTATTR_OIF, interface != OIF_NONE ? sizeof(RTATTR_OIF) : 0 }, 362 { &ifindex, interface != OIF_NONE ? sizeof(ifindex) : 0 }, 363 { &rtaGateway, nexthop ? sizeof(rtaGateway) : 0 }, 364 { rawNexthop, nexthop ? static_cast<size_t>(rawLength) : 0 }, 365 }; 366 367 uint16_t flags = (action == RTM_NEWROUTE) ? NETLINK_CREATE_REQUEST_FLAGS : 368 NETLINK_REQUEST_FLAGS; 369 return sendNetlinkRequest(action, flags, iov, ARRAY_SIZE(iov)); 370} 371 372// Add rules to allow legacy routes added through the requestRouteToHost() API. 373WARN_UNUSED_RESULT int AddLegacyRouteRules() { 374 Fwmark fwmark; 375 Fwmark mask; 376 377 fwmark.explicitlySelected = false; 378 mask.explicitlySelected = true; 379 380 // Rules to allow legacy routes to override the default network. 381 if (int ret = modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LEGACY_SYSTEM, ROUTE_TABLE_LEGACY_SYSTEM, 382 fwmark.intValue, mask.intValue, OIF_NONE, INVALID_UID, 383 INVALID_UID)) { 384 return ret; 385 } 386 if (int ret = modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LEGACY_NETWORK, 387 ROUTE_TABLE_LEGACY_NETWORK, fwmark.intValue, 388 mask.intValue, OIF_NONE, INVALID_UID, INVALID_UID)) { 389 return ret; 390 } 391 392 fwmark.permission = PERMISSION_SYSTEM; 393 mask.permission = PERMISSION_SYSTEM; 394 395 // A rule to allow legacy routes from system apps to override VPNs. 396 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_VPN_OVERRIDE_SYSTEM, ROUTE_TABLE_LEGACY_SYSTEM, 397 fwmark.intValue, mask.intValue, OIF_NONE, INVALID_UID, INVALID_UID); 398} 399 400// Add a new rule to look up the 'main' table, with the same selectors as the "default network" 401// rule, but with a lower priority. Since the default network rule points to a table with a default 402// route, the rule we're adding will never be used for normal routing lookups. However, the kernel 403// may fall-through to it to find directly-connected routes when it validates that a nexthop (in a 404// route being added) is reachable. 405WARN_UNUSED_RESULT int AddDirectlyConnectedRule() { 406 Fwmark fwmark; 407 Fwmark mask; 408 409 fwmark.netId = NETID_UNSET; 410 mask.netId = FWMARK_NET_ID_MASK; 411 412 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_DIRECTLY_CONNECTED, RT_TABLE_MAIN, 413 fwmark.intValue, mask.intValue, OIF_NONE, UID_ROOT, UID_ROOT); 414} 415 416// Add a rule to preempt the pre-defined "from all lookup main" rule. Packets that reach this rule 417// will be null-routed, and won't fall-through to the main table. 418WARN_UNUSED_RESULT int AddUnreachableRule() { 419 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_UNREACHABLE, RT_TABLE_UNSPEC, MARK_UNSET, 420 MARK_UNSET, OIF_NONE, INVALID_UID, INVALID_UID); 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 traffic based on an explicitly chosen network. 452// 453// Supports apps that use the multinetwork APIs to restrict their traffic to a network. 454// 455// Even though we check permissions at the time we set a netId into the fwmark of a socket, we need 456// to check it again in the rules here, because a network's permissions may have been updated via 457// modifyNetworkPermission(). 458WARN_UNUSED_RESULT int modifyExplicitNetworkRule(unsigned netId, uint32_t table, 459 Permission permission, uid_t uidStart, 460 uid_t uidEnd, bool add) { 461 Fwmark fwmark; 462 Fwmark mask; 463 464 fwmark.netId = netId; 465 mask.netId = FWMARK_NET_ID_MASK; 466 467 fwmark.explicitlySelected = true; 468 mask.explicitlySelected = true; 469 470 fwmark.permission = permission; 471 mask.permission = permission; 472 473 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_EXPLICIT_NETWORK, table, 474 fwmark.intValue, mask.intValue, OIF_NONE, uidStart, uidEnd); 475} 476 477// A rule to route traffic based on a chosen outgoing interface. 478// 479// Supports apps that use SO_BINDTODEVICE or IP_PKTINFO options and the kernel that already knows 480// the outgoing interface (typically for link-local communications). 481WARN_UNUSED_RESULT int modifyOutputInterfaceRule(const char* interface, uint32_t table, 482 Permission permission, uid_t uidStart, 483 uid_t uidEnd, bool add) { 484 Fwmark fwmark; 485 Fwmark mask; 486 487 fwmark.permission = permission; 488 mask.permission = permission; 489 490 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_OUTPUT_INTERFACE, table, 491 fwmark.intValue, mask.intValue, interface, uidStart, uidEnd); 492} 493 494// A rule to route traffic based on the chosen network. 495// 496// This is for sockets that have not explicitly requested a particular network, but have been 497// bound to one when they called connect(). This ensures that sockets connected on a particular 498// network stay on that network even if the default network changes. 499WARN_UNUSED_RESULT int modifyImplicitNetworkRule(unsigned netId, uint32_t table, 500 Permission permission, bool add) { 501 Fwmark fwmark; 502 Fwmark mask; 503 504 fwmark.netId = netId; 505 mask.netId = FWMARK_NET_ID_MASK; 506 507 fwmark.explicitlySelected = false; 508 mask.explicitlySelected = true; 509 510 fwmark.permission = permission; 511 mask.permission = permission; 512 513 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_IMPLICIT_NETWORK, table, 514 fwmark.intValue, mask.intValue, OIF_NONE, INVALID_UID, INVALID_UID); 515} 516 517// A rule to route all traffic from a given set of UIDs to go over the VPN. 518// 519// Notice that this rule doesn't use the netId. I.e., no matter what netId the user's socket may 520// have, if they are subject to this VPN, their traffic has to go through it. Allows the traffic to 521// bypass the VPN if the protectedFromVpn bit is set. 522WARN_UNUSED_RESULT int modifyVpnUidRangeRule(uint32_t table, uid_t uidStart, uid_t uidEnd, 523 bool add) { 524 Fwmark fwmark; 525 Fwmark mask; 526 527 fwmark.protectedFromVpn = false; 528 mask.protectedFromVpn = true; 529 530 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_SECURE_VPN, table, 531 fwmark.intValue, mask.intValue, OIF_NONE, uidStart, uidEnd); 532} 533 534// A rule to allow system apps to send traffic over this VPN even if they are not part of the target 535// set of UIDs. 536// 537// This is needed for DnsProxyListener to correctly resolve a request for a user who is in the 538// target set, but where the DnsProxyListener itself is not. 539WARN_UNUSED_RESULT int modifyVpnSystemPermissionRule(unsigned netId, uint32_t table, bool add) { 540 Fwmark fwmark; 541 Fwmark mask; 542 543 fwmark.netId = netId; 544 mask.netId = FWMARK_NET_ID_MASK; 545 546 fwmark.permission = PERMISSION_SYSTEM; 547 mask.permission = PERMISSION_SYSTEM; 548 549 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_SECURE_VPN, table, 550 fwmark.intValue, mask.intValue, OIF_NONE, INVALID_UID, INVALID_UID); 551} 552 553WARN_UNUSED_RESULT int modifyPhysicalNetwork(unsigned netId, const char* interface, 554 Permission permission, bool add) { 555 uint32_t table = getRouteTableForInterface(interface); 556 if (table == RT_TABLE_UNSPEC) { 557 return -ESRCH; 558 } 559 560 if (int ret = modifyIncomingPacketMark(netId, interface, permission, add)) { 561 return ret; 562 } 563 if (int ret = modifyExplicitNetworkRule(netId, table, permission, INVALID_UID, INVALID_UID, 564 add)) { 565 return ret; 566 } 567 if (int ret = modifyOutputInterfaceRule(interface, table, permission, INVALID_UID, INVALID_UID, 568 add)) { 569 return ret; 570 } 571 return modifyImplicitNetworkRule(netId, table, permission, add); 572} 573 574WARN_UNUSED_RESULT int modifyVirtualNetwork(unsigned netId, const char* interface, 575 const UidRanges& uidRanges, bool add, 576 bool modifyNonUidBasedRules) { 577 uint32_t table = getRouteTableForInterface(interface); 578 if (table == RT_TABLE_UNSPEC) { 579 return -ESRCH; 580 } 581 582 for (const UidRanges::Range& range : uidRanges.getRanges()) { 583 if (int ret = modifyExplicitNetworkRule(netId, table, PERMISSION_NONE, range.first, 584 range.second, add)) { 585 return ret; 586 } 587 if (int ret = modifyOutputInterfaceRule(interface, table, PERMISSION_NONE, range.first, 588 range.second, add)) { 589 return ret; 590 } 591 if (int ret = modifyVpnUidRangeRule(table, range.first, range.second, add)) { 592 return ret; 593 } 594 } 595 596 if (modifyNonUidBasedRules) { 597 if (int ret = modifyIncomingPacketMark(netId, interface, PERMISSION_NONE, add)) { 598 return ret; 599 } 600 if (int ret = modifyExplicitNetworkRule(netId, table, PERMISSION_NONE, UID_ROOT, UID_ROOT, 601 add)) { 602 return ret; 603 } 604 return modifyVpnSystemPermissionRule(netId, table, add); 605 } 606 607 return 0; 608} 609 610WARN_UNUSED_RESULT int modifyDefaultNetwork(uint16_t action, const char* interface, 611 Permission permission) { 612 uint32_t table = getRouteTableForInterface(interface); 613 if (table == RT_TABLE_UNSPEC) { 614 return -ESRCH; 615 } 616 617 Fwmark fwmark; 618 Fwmark mask; 619 620 fwmark.netId = NETID_UNSET; 621 mask.netId = FWMARK_NET_ID_MASK; 622 623 fwmark.permission = permission; 624 mask.permission = permission; 625 626 return modifyIpRule(action, RULE_PRIORITY_DEFAULT_NETWORK, table, fwmark.intValue, 627 mask.intValue, OIF_NONE, INVALID_UID, INVALID_UID); 628} 629 630// Adds or removes an IPv4 or IPv6 route to the specified table and, if it's a directly-connected 631// route, to the main table as well. 632// Returns 0 on success or negative errno on failure. 633WARN_UNUSED_RESULT int modifyRoute(uint16_t action, const char* interface, const char* destination, 634 const char* nexthop, RouteController::TableType tableType) { 635 uint32_t table; 636 switch (tableType) { 637 case RouteController::INTERFACE: { 638 table = getRouteTableForInterface(interface); 639 if (table == RT_TABLE_UNSPEC) { 640 return -ESRCH; 641 } 642 break; 643 } 644 case RouteController::LEGACY_NETWORK: { 645 table = ROUTE_TABLE_LEGACY_NETWORK; 646 break; 647 } 648 case RouteController::LEGACY_SYSTEM: { 649 table = ROUTE_TABLE_LEGACY_SYSTEM; 650 break; 651 } 652 } 653 654 int ret = modifyIpRoute(action, table, interface, destination, nexthop); 655 // We allow apps to call requestRouteToHost() multiple times with the same route, so ignore 656 // EEXIST failures when adding routes to legacy tables. 657 if (ret && !(action == RTM_NEWROUTE && ret == -EEXIST && 658 tableType != RouteController::INTERFACE)) { 659 return ret; 660 } 661 662 // If there's no nexthop, this is a directly connected route. Add it to the main table also, to 663 // let the kernel find it when validating nexthops when global routes are added. 664 if (!nexthop) { 665 ret = modifyIpRoute(action, RT_TABLE_MAIN, interface, destination, NULL); 666 // A failure with action == ADD && errno == EEXIST means that the route already exists in 667 // the main table, perhaps because the kernel added it automatically as part of adding the 668 // IP address to the interface. Ignore this, but complain about everything else. 669 if (ret && !(action == RTM_NEWROUTE && ret == -EEXIST)) { 670 return ret; 671 } 672 } 673 674 return 0; 675} 676 677// Returns 0 on success or negative errno on failure. 678WARN_UNUSED_RESULT int flushRoutes(const char* interface) { 679 uint32_t table = getRouteTableForInterface(interface); 680 if (table == RT_TABLE_UNSPEC) { 681 return -ESRCH; 682 } 683 684 char tableString[UINT32_STRLEN]; 685 snprintf(tableString, sizeof(tableString), "%u", table); 686 687 for (size_t i = 0; i < ARRAY_SIZE(IP_VERSIONS); ++i) { 688 const char* argv[] = { 689 IP_PATH, 690 IP_VERSIONS[i], 691 "route", 692 "flush", 693 "table", 694 tableString, 695 }; 696 if (android_fork_execvp(ARRAY_SIZE(argv), const_cast<char**>(argv), NULL, false, false)) { 697 ALOGE("failed to flush routes"); 698 return -EREMOTEIO; 699 } 700 } 701 702 interfaceToTable.erase(interface); 703 return 0; 704} 705 706} // namespace 707 708int RouteController::Init() { 709 if (int ret = AddDirectlyConnectedRule()) { 710 return ret; 711 } 712 if (int ret = AddLegacyRouteRules()) { 713 return ret; 714 } 715 // TODO: Enable once we are sure everything works. 716 if (false) { 717 if (int ret = AddUnreachableRule()) { 718 return ret; 719 } 720 } 721 updateTableNamesFile(); 722 return 0; 723} 724 725int RouteController::addInterfaceToPhysicalNetwork(unsigned netId, const char* interface, 726 Permission permission) { 727 if (int ret = modifyPhysicalNetwork(netId, interface, permission, ACTION_ADD)) { 728 return ret; 729 } 730 updateTableNamesFile(); 731 return 0; 732} 733 734int RouteController::removeInterfaceFromPhysicalNetwork(unsigned netId, const char* interface, 735 Permission permission) { 736 if (int ret = modifyPhysicalNetwork(netId, interface, permission, ACTION_DEL)) { 737 return ret; 738 } 739 if (int ret = flushRoutes(interface)) { 740 return ret; 741 } 742 updateTableNamesFile(); 743 return 0; 744} 745 746int RouteController::addInterfaceToVirtualNetwork(unsigned netId, const char* interface, 747 const UidRanges& uidRanges) { 748 if (int ret = modifyVirtualNetwork(netId, interface, uidRanges, ACTION_ADD, 749 MODIFY_NON_UID_BASED_RULES)) { 750 return ret; 751 } 752 updateTableNamesFile(); 753 return 0; 754} 755 756int RouteController::removeInterfaceFromVirtualNetwork(unsigned netId, const char* interface, 757 const UidRanges& uidRanges) { 758 if (int ret = modifyVirtualNetwork(netId, interface, uidRanges, ACTION_DEL, 759 MODIFY_NON_UID_BASED_RULES)) { 760 return ret; 761 } 762 if (int ret = flushRoutes(interface)) { 763 return ret; 764 } 765 updateTableNamesFile(); 766 return 0; 767} 768 769int RouteController::modifyPhysicalNetworkPermission(unsigned netId, const char* interface, 770 Permission oldPermission, 771 Permission newPermission) { 772 // Add the new rules before deleting the old ones, to avoid race conditions. 773 if (int ret = modifyPhysicalNetwork(netId, interface, newPermission, ACTION_ADD)) { 774 return ret; 775 } 776 return modifyPhysicalNetwork(netId, interface, oldPermission, ACTION_DEL); 777} 778 779int RouteController::addUsersToVirtualNetwork(unsigned netId, const char* interface, 780 const UidRanges& uidRanges) { 781 return modifyVirtualNetwork(netId, interface, uidRanges, ACTION_ADD, 782 !MODIFY_NON_UID_BASED_RULES); 783} 784 785int RouteController::removeUsersFromVirtualNetwork(unsigned netId, const char* interface, 786 const UidRanges& uidRanges) { 787 return modifyVirtualNetwork(netId, interface, uidRanges, ACTION_DEL, 788 !MODIFY_NON_UID_BASED_RULES); 789} 790 791int RouteController::addInterfaceToDefaultNetwork(const char* interface, Permission permission) { 792 return modifyDefaultNetwork(RTM_NEWRULE, interface, permission); 793} 794 795int RouteController::removeInterfaceFromDefaultNetwork(const char* interface, 796 Permission permission) { 797 return modifyDefaultNetwork(RTM_DELRULE, interface, permission); 798} 799 800int RouteController::addRoute(const char* interface, const char* destination, const char* nexthop, 801 TableType tableType) { 802 return modifyRoute(RTM_NEWROUTE, interface, destination, nexthop, tableType); 803} 804 805int RouteController::removeRoute(const char* interface, const char* destination, 806 const char* nexthop, TableType tableType) { 807 return modifyRoute(RTM_DELROUTE, interface, destination, nexthop, tableType); 808} 809