2 * Copyright (C) 2014 The Android Open Source Project
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
8 * http://www.apache.org/licenses/LICENSE-2.0
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.
17 #include "RouteController.h"
19 #include <arpa/inet.h>
22 #include <linux/fib_rules.h>
26 #include <private/android_filesystem_config.h>
31 #include "UidRanges.h"
33 #define LOG_TAG "Netd"
35 #include "logwrap/logwrap.h"
36 #include "utils/file.h"
37 #include "resolv_netid.h"
41 // BEGIN CONSTANTS --------------------------------------------------------------------------------
43 const uint32_t RULE_PRIORITY_VPN_OVERRIDE_SYSTEM = 10000;
44 const uint32_t RULE_PRIORITY_VPN_OUTPUT_TO_LOCAL = 11000;
45 const uint32_t RULE_PRIORITY_SECURE_VPN = 12000;
46 const uint32_t RULE_PRIORITY_EXPLICIT_NETWORK = 13000;
47 const uint32_t RULE_PRIORITY_OUTPUT_INTERFACE = 14000;
48 const uint32_t RULE_PRIORITY_LEGACY_SYSTEM = 15000;
49 const uint32_t RULE_PRIORITY_LEGACY_NETWORK = 16000;
50 const uint32_t RULE_PRIORITY_LOCAL_NETWORK = 17000;
51 const uint32_t RULE_PRIORITY_TETHERING = 18000;
52 const uint32_t RULE_PRIORITY_IMPLICIT_NETWORK = 19000;
53 const uint32_t RULE_PRIORITY_BYPASSABLE_VPN = 20000;
54 const uint32_t RULE_PRIORITY_VPN_FALLTHROUGH = 21000;
55 const uint32_t RULE_PRIORITY_DEFAULT_NETWORK = 22000;
56 const uint32_t RULE_PRIORITY_DIRECTLY_CONNECTED = 23000;
57 const uint32_t RULE_PRIORITY_UNREACHABLE = 32000;
59 const uint32_t ROUTE_TABLE_LOCAL_NETWORK = 97;
60 const uint32_t ROUTE_TABLE_LEGACY_NETWORK = 98;
61 const uint32_t ROUTE_TABLE_LEGACY_SYSTEM = 99;
63 const char* const ROUTE_TABLE_NAME_LOCAL_NETWORK = "local_network";
64 const char* const ROUTE_TABLE_NAME_LEGACY_NETWORK = "legacy_network";
65 const char* const ROUTE_TABLE_NAME_LEGACY_SYSTEM = "legacy_system";
67 const char* const ROUTE_TABLE_NAME_LOCAL = "local";
68 const char* const ROUTE_TABLE_NAME_MAIN = "main";
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.
74 const uint16_t FRA_UID_START = 18;
75 const uint16_t FRA_UID_END = 19;
76 static_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.");
80 const uint16_t NETLINK_REQUEST_FLAGS = NLM_F_REQUEST | NLM_F_ACK;
81 const uint16_t NETLINK_CREATE_REQUEST_FLAGS = NETLINK_REQUEST_FLAGS | NLM_F_CREATE | NLM_F_EXCL;
83 const sockaddr_nl NETLINK_ADDRESS = {AF_NETLINK, 0, 0, 0};
85 const uint8_t AF_FAMILIES[] = {AF_INET, AF_INET6};
87 const char* const IP_VERSIONS[] = {"-4", "-6"};
89 const uid_t UID_ROOT = 0;
90 const char* const IIF_NONE = NULL;
91 const char* const OIF_NONE = NULL;
92 const bool ACTION_ADD = true;
93 const bool ACTION_DEL = false;
94 const bool MODIFY_NON_UID_BASED_RULES = true;
96 const char* const RT_TABLES_PATH = "/data/misc/net/rt_tables";
97 const mode_t RT_TABLES_MODE = S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH; // mode 0644, rw-r--r--
99 const unsigned ROUTE_FLUSH_ATTEMPTS = 2;
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).
103 constexpr uint16_t U16_RTA_LENGTH(uint16_t x) {
104 return RTA_LENGTH(x);
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.
109 rtattr FRATTR_PRIORITY = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_PRIORITY };
110 rtattr FRATTR_TABLE = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_TABLE };
111 rtattr FRATTR_FWMARK = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_FWMARK };
112 rtattr FRATTR_FWMASK = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_FWMASK };
113 rtattr FRATTR_UID_START = { U16_RTA_LENGTH(sizeof(uid_t)), FRA_UID_START };
114 rtattr FRATTR_UID_END = { U16_RTA_LENGTH(sizeof(uid_t)), FRA_UID_END };
116 rtattr RTATTR_TABLE = { U16_RTA_LENGTH(sizeof(uint32_t)), RTA_TABLE };
117 rtattr RTATTR_OIF = { U16_RTA_LENGTH(sizeof(uint32_t)), RTA_OIF };
119 uint8_t PADDING_BUFFER[RTA_ALIGNTO] = {0, 0, 0, 0};
121 // END CONSTANTS ----------------------------------------------------------------------------------
123 // No locks needed because RouteController is accessed only from one thread (in CommandListener).
124 std::map<std::string, uint32_t> interfaceToTable;
126 uint32_t getRouteTableForInterface(const char* interface) {
127 uint32_t index = if_nametoindex(interface);
129 index += RouteController::ROUTE_TABLE_OFFSET_FROM_INDEX;
130 interfaceToTable[interface] = index;
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;
143 void 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;
152 // Doesn't return success/failure as the file is optional; it's okay if we fail to update it.
153 void updateTableNamesFile() {
154 std::string contents;
156 addTableName(RT_TABLE_LOCAL, ROUTE_TABLE_NAME_LOCAL, &contents);
157 addTableName(RT_TABLE_MAIN, ROUTE_TABLE_NAME_MAIN, &contents);
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);
163 for (const auto& entry : interfaceToTable) {
164 addTableName(entry.second, entry.first, &contents);
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));
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.
177 WARN_UNUSED_RESULT int sendNetlinkRequest(uint16_t action, uint16_t flags, iovec* iov, int iovlen) {
179 .nlmsg_type = action,
180 .nlmsg_flags = flags,
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;
194 int sock = socket(AF_NETLINK, SOCK_DGRAM | SOCK_CLOEXEC, NETLINK_ROUTE);
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.
203 ALOGE("netlink response contains error (%s)", strerror(-ret));
206 ALOGE("bad netlink response message size (%d != %zu)", ret, sizeof(response));
210 ALOGE("netlink socket/connect/writev/recv failed (%s)", strerror(errno));
221 // Returns 0 on success or negative errno on failure.
222 int padInterfaceName(const char* input, char* name, size_t* length, uint16_t* padding) {
228 *length = strlcpy(name, input, IFNAMSIZ) + 1;
229 if (*length > IFNAMSIZ) {
230 ALOGE("interface name too long (%zu > %u)", *length, IFNAMSIZ);
231 return -ENAMETOOLONG;
233 *padding = RTA_SPACE(*length) - RTA_LENGTH(*length);
237 // Adds or removes a routing rule for IPv4 and IPv6.
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
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.
248 // Returns 0 on success or negative errno on failure.
249 WARN_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);
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)) {
266 if (int ret = padInterfaceName(oif, oifName, &oifLength, &oifPadding)) {
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);
275 bool isUidRule = (uidStart != INVALID_UID);
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 :
283 rtattr fraIifName = { U16_RTA_LENGTH(iifLength), FRA_IIFNAME };
284 rtattr fraOifName = { U16_RTA_LENGTH(oifLength), FRA_OIFNAME };
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 },
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))) {
320 WARN_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,
326 // Adds or deletes an IPv4 or IPv6 route.
327 // Returns 0 on success or negative errno on failure.
328 WARN_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.
332 ALOGE("null destination");
337 uint8_t rawAddress[sizeof(in6_addr)];
339 uint8_t prefixLength;
340 int rawLength = parsePrefix(destination, &family, rawAddress, sizeof(rawAddress),
343 ALOGE("parsePrefix failed for destination %s (%s)", destination, strerror(-rawLength));
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.
352 uint8_t type = RTN_UNICAST;
354 uint8_t rawNexthop[sizeof(in6_addr)];
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;
363 } else if (nexthop && !strcmp(nexthop, "throw")) {
365 interface = OIF_NONE;
368 // If an interface was specified, find the ifindex.
369 if (interface != OIF_NONE) {
370 ifindex = if_nametoindex(interface);
372 ALOGE("cannot find interface %s", interface);
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);
384 // Assemble a rtmsg and put it in an array of iovec structures.
386 .rtm_protocol = RTPROT_STATIC,
388 .rtm_family = family,
389 .rtm_dst_len = prefixLength,
390 .rtm_scope = static_cast<uint8_t>(nexthop ? RT_SCOPE_UNIVERSE : RT_SCOPE_LINK),
393 rtattr rtaDst = { U16_RTA_LENGTH(rawLength), RTA_DST };
394 rtattr rtaGateway = { U16_RTA_LENGTH(rawLength), RTA_GATEWAY };
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 },
409 uint16_t flags = (action == RTM_NEWROUTE) ? NETLINK_CREATE_REQUEST_FLAGS :
410 NETLINK_REQUEST_FLAGS;
411 return sendNetlinkRequest(action, flags, iov, ARRAY_SIZE(iov));
414 // An iptables rule to mark incoming packets on a network with the netId of the network.
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
421 WARN_UNUSED_RESULT int modifyIncomingPacketMark(unsigned netId, const char* interface,
422 Permission permission, bool add) {
425 fwmark.netId = netId;
426 fwmark.explicitlySelected = true;
427 fwmark.protectedFromVpn = true;
428 fwmark.permission = permission;
430 char markString[UINT32_HEX_STRLEN];
431 snprintf(markString, sizeof(markString), "0x%x", fwmark.intValue);
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");
442 // A rule to route responses to the local network forwarded via the VPN.
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.
446 WARN_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);
452 // A rule to route all traffic from a given set of UIDs to go over the VPN.
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.
457 WARN_UNUSED_RESULT int modifyVpnUidRangeRule(uint32_t table, uid_t uidStart, uid_t uidEnd,
458 bool secure, bool add) {
462 fwmark.protectedFromVpn = false;
463 mask.protectedFromVpn = true;
468 priority = RULE_PRIORITY_SECURE_VPN;
470 priority = RULE_PRIORITY_BYPASSABLE_VPN;
472 fwmark.explicitlySelected = false;
473 mask.explicitlySelected = true;
476 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, priority, table, fwmark.intValue,
477 mask.intValue, IIF_NONE, OIF_NONE, uidStart, uidEnd);
480 // A rule to allow system apps to send traffic over this VPN even if they are not part of the target
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.
485 WARN_UNUSED_RESULT int modifyVpnSystemPermissionRule(unsigned netId, uint32_t table, bool secure,
490 fwmark.netId = netId;
491 mask.netId = FWMARK_NET_ID_MASK;
493 fwmark.permission = PERMISSION_SYSTEM;
494 mask.permission = PERMISSION_SYSTEM;
496 uint32_t priority = secure ? RULE_PRIORITY_SECURE_VPN : RULE_PRIORITY_BYPASSABLE_VPN;
498 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, priority, table, fwmark.intValue,
502 // A rule to route traffic based on an explicitly chosen network.
504 // Supports apps that use the multinetwork APIs to restrict their traffic to a network.
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().
509 WARN_UNUSED_RESULT int modifyExplicitNetworkRule(unsigned netId, uint32_t table,
510 Permission permission, uid_t uidStart,
511 uid_t uidEnd, bool add) {
515 fwmark.netId = netId;
516 mask.netId = FWMARK_NET_ID_MASK;
518 fwmark.explicitlySelected = true;
519 mask.explicitlySelected = true;
521 fwmark.permission = permission;
522 mask.permission = permission;
524 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_EXPLICIT_NETWORK, table,
525 fwmark.intValue, mask.intValue, IIF_NONE, OIF_NONE, uidStart, uidEnd);
528 // A rule to route traffic based on a chosen outgoing interface.
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).
532 WARN_UNUSED_RESULT int modifyOutputInterfaceRule(const char* interface, uint32_t table,
533 Permission permission, uid_t uidStart,
534 uid_t uidEnd, bool add) {
538 fwmark.permission = permission;
539 mask.permission = permission;
541 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_OUTPUT_INTERFACE, table,
542 fwmark.intValue, mask.intValue, IIF_NONE, interface, uidStart, uidEnd);
545 // A rule to route traffic based on the chosen network.
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.
550 WARN_UNUSED_RESULT int modifyImplicitNetworkRule(unsigned netId, uint32_t table,
551 Permission permission, bool add) {
555 fwmark.netId = netId;
556 mask.netId = FWMARK_NET_ID_MASK;
558 fwmark.explicitlySelected = false;
559 mask.explicitlySelected = true;
561 fwmark.permission = permission;
562 mask.permission = permission;
564 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_IMPLICIT_NETWORK, table,
565 fwmark.intValue, mask.intValue);
568 // A rule to enable split tunnel VPNs.
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.
573 WARN_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) {
584 fwmark.netId = vpnNetId;
585 mask.netId = FWMARK_NET_ID_MASK;
587 fwmark.explicitlySelected = false;
588 mask.explicitlySelected = true;
590 fwmark.permission = permission;
591 mask.permission = permission;
593 return modifyIpRule(action, RULE_PRIORITY_VPN_FALLTHROUGH, table, fwmark.intValue,
597 // Add rules to allow legacy routes added through the requestRouteToHost() API.
598 WARN_UNUSED_RESULT int addLegacyRouteRules() {
602 fwmark.explicitlySelected = false;
603 mask.explicitlySelected = true;
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)) {
610 if (int ret = modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LEGACY_NETWORK,
611 ROUTE_TABLE_LEGACY_NETWORK, fwmark.intValue, mask.intValue)) {
615 fwmark.permission = PERMISSION_SYSTEM;
616 mask.permission = PERMISSION_SYSTEM;
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);
623 // Add rules to lookup the local network when specified explicitly or otherwise.
624 WARN_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)) {
633 fwmark.explicitlySelected = false;
634 mask.explicitlySelected = true;
636 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LOCAL_NETWORK, ROUTE_TABLE_LOCAL_NETWORK,
637 fwmark.intValue, mask.intValue);
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.
645 WARN_UNUSED_RESULT int addDirectlyConnectedRule() {
649 fwmark.netId = NETID_UNSET;
650 mask.netId = FWMARK_NET_ID_MASK;
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);
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.
660 WARN_UNUSED_RESULT int addUnreachableRule() {
661 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_UNREACHABLE, RT_TABLE_UNSPEC, MARK_UNSET,
665 WARN_UNUSED_RESULT int modifyLocalNetwork(unsigned netId, const char* interface, bool add) {
666 if (int ret = modifyIncomingPacketMark(netId, interface, PERMISSION_NONE, add)) {
669 return modifyOutputInterfaceRule(interface, ROUTE_TABLE_LOCAL_NETWORK, PERMISSION_NONE,
670 INVALID_UID, INVALID_UID, add);
673 WARN_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) {
680 if (int ret = modifyIncomingPacketMark(netId, interface, permission, add)) {
683 if (int ret = modifyExplicitNetworkRule(netId, table, permission, INVALID_UID, INVALID_UID,
687 if (int ret = modifyOutputInterfaceRule(interface, table, permission, INVALID_UID, INVALID_UID,
691 return modifyImplicitNetworkRule(netId, table, permission, add);
694 WARN_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) {
702 for (const UidRanges::Range& range : uidRanges.getRanges()) {
703 if (int ret = modifyVpnUidRangeRule(table, range.first, range.second, secure, add)) {
706 if (int ret = modifyExplicitNetworkRule(netId, table, PERMISSION_NONE, range.first,
707 range.second, add)) {
710 if (int ret = modifyOutputInterfaceRule(interface, table, PERMISSION_NONE, range.first,
711 range.second, add)) {
716 if (modifyNonUidBasedRules) {
717 if (int ret = modifyIncomingPacketMark(netId, interface, PERMISSION_NONE, add)) {
720 if (int ret = modifyVpnOutputToLocalRule(interface, add)) {
723 if (int ret = modifyVpnSystemPermissionRule(netId, table, secure, add)) {
726 return modifyExplicitNetworkRule(netId, table, PERMISSION_NONE, UID_ROOT, UID_ROOT, add);
732 WARN_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) {
742 fwmark.netId = NETID_UNSET;
743 mask.netId = FWMARK_NET_ID_MASK;
745 fwmark.permission = permission;
746 mask.permission = permission;
748 return modifyIpRule(action, RULE_PRIORITY_DEFAULT_NETWORK, table, fwmark.intValue,
752 WARN_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) {
759 return modifyIpRule(action, RULE_PRIORITY_TETHERING, table, MARK_UNSET, MARK_UNSET,
760 inputInterface, OIF_NONE, INVALID_UID, INVALID_UID);
763 // Returns 0 on success or negative errno on failure.
764 WARN_UNUSED_RESULT int flushRules() {
765 for (size_t i = 0; i < ARRAY_SIZE(IP_VERSIONS); ++i) {
766 const char* argv[] = {
772 if (android_fork_execvp(ARRAY_SIZE(argv), const_cast<char**>(argv), NULL, false, false)) {
773 ALOGE("failed to flush rules");
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.
783 WARN_UNUSED_RESULT int modifyRoute(uint16_t action, const char* interface, const char* destination,
784 const char* nexthop, RouteController::TableType tableType) {
787 case RouteController::INTERFACE: {
788 table = getRouteTableForInterface(interface);
789 if (table == RT_TABLE_UNSPEC) {
794 case RouteController::LOCAL_NETWORK: {
795 table = ROUTE_TABLE_LOCAL_NETWORK;
798 case RouteController::LEGACY_NETWORK: {
799 table = ROUTE_TABLE_LEGACY_NETWORK;
802 case RouteController::LEGACY_SYSTEM: {
803 table = ROUTE_TABLE_LEGACY_SYSTEM;
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)) {
817 // Returns 0 on success or negative errno on failure.
818 WARN_UNUSED_RESULT int flushRoutes(const char* interface) {
819 uint32_t table = getRouteTableForInterface(interface);
820 if (table == RT_TABLE_UNSPEC) {
824 char tableString[UINT32_STRLEN];
825 snprintf(tableString, sizeof(tableString), "%u", table);
828 for (size_t i = 0; i < ARRAY_SIZE(IP_VERSIONS); ++i) {
829 const char* argv[] = {
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.
843 // TODO: replace this with our own netlink code.
844 unsigned attempts = 0;
847 err = android_fork_execvp(ARRAY_SIZE(argv), const_cast<char**>(argv),
850 } while (err != 0 && attempts < ROUTE_FLUSH_ATTEMPTS);
852 ALOGE("failed to flush %s routes in table %s after %d attempts",
853 IP_VERSIONS[i], tableString, attempts);
858 // If we failed to flush routes, the caller may elect to keep this interface around, so keep
859 // track of its name.
861 interfaceToTable.erase(interface);
869 int RouteController::Init(unsigned localNetId) {
870 if (int ret = flushRules()) {
873 if (int ret = addLegacyRouteRules()) {
876 if (int ret = addLocalNetworkRules(localNetId)) {
879 if (int ret = addDirectlyConnectedRule()) {
882 if (int ret = addUnreachableRule()) {
885 updateTableNamesFile();
889 int RouteController::addInterfaceToLocalNetwork(unsigned netId, const char* interface) {
890 return modifyLocalNetwork(netId, interface, ACTION_ADD);
893 int RouteController::removeInterfaceFromLocalNetwork(unsigned netId, const char* interface) {
894 return modifyLocalNetwork(netId, interface, ACTION_DEL);
897 int RouteController::addInterfaceToPhysicalNetwork(unsigned netId, const char* interface,
898 Permission permission) {
899 if (int ret = modifyPhysicalNetwork(netId, interface, permission, ACTION_ADD)) {
902 updateTableNamesFile();
906 int RouteController::removeInterfaceFromPhysicalNetwork(unsigned netId, const char* interface,
907 Permission permission) {
908 if (int ret = modifyPhysicalNetwork(netId, interface, permission, ACTION_DEL)) {
911 if (int ret = flushRoutes(interface)) {
914 updateTableNamesFile();
918 int 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)) {
924 updateTableNamesFile();
928 int 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)) {
934 if (int ret = flushRoutes(interface)) {
937 updateTableNamesFile();
941 int 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)) {
948 return modifyPhysicalNetwork(netId, interface, oldPermission, ACTION_DEL);
951 int 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);
957 int 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);
963 int RouteController::addInterfaceToDefaultNetwork(const char* interface, Permission permission) {
964 return modifyDefaultNetwork(RTM_NEWRULE, interface, permission);
967 int RouteController::removeInterfaceFromDefaultNetwork(const char* interface,
968 Permission permission) {
969 return modifyDefaultNetwork(RTM_DELRULE, interface, permission);
972 int RouteController::addRoute(const char* interface, const char* destination, const char* nexthop,
973 TableType tableType) {
974 return modifyRoute(RTM_NEWROUTE, interface, destination, nexthop, tableType);
977 int RouteController::removeRoute(const char* interface, const char* destination,
978 const char* nexthop, TableType tableType) {
979 return modifyRoute(RTM_DELROUTE, interface, destination, nexthop, tableType);
982 int RouteController::enableTethering(const char* inputInterface, const char* outputInterface) {
983 return modifyTetheredNetwork(RTM_NEWRULE, inputInterface, outputInterface);
986 int RouteController::disableTethering(const char* inputInterface, const char* outputInterface) {
987 return modifyTetheredNetwork(RTM_DELRULE, inputInterface, outputInterface);
990 int RouteController::addVirtualNetworkFallthrough(unsigned vpnNetId, const char* physicalInterface,
991 Permission permission) {
992 return modifyVpnFallthroughRule(RTM_NEWRULE, vpnNetId, physicalInterface, permission);
995 int RouteController::removeVirtualNetworkFallthrough(unsigned vpnNetId,
996 const char* physicalInterface,
997 Permission permission) {
998 return modifyVpnFallthroughRule(RTM_DELRULE, vpnNetId, physicalInterface, permission);