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"
20 #include "UidRanges.h"
22 #define LOG_TAG "Netd"
24 #include "logwrap/logwrap.h"
25 #include "resolv_netid.h"
27 #include <arpa/inet.h>
29 #include <linux/fib_rules.h>
36 // BEGIN CONSTANTS --------------------------------------------------------------------------------
38 const uint32_t RULE_PRIORITY_VPN_OVERRIDE_SYSTEM = 10000;
39 const uint32_t RULE_PRIORITY_VPN_OUTPUT_TO_LOCAL = 11000;
40 const uint32_t RULE_PRIORITY_SECURE_VPN = 12000;
41 const uint32_t RULE_PRIORITY_EXPLICIT_NETWORK = 13000;
42 const uint32_t RULE_PRIORITY_OUTPUT_INTERFACE = 14000;
43 const uint32_t RULE_PRIORITY_LEGACY_SYSTEM = 15000;
44 const uint32_t RULE_PRIORITY_LEGACY_NETWORK = 16000;
45 const uint32_t RULE_PRIORITY_LOCAL_NETWORK = 17000;
46 const uint32_t RULE_PRIORITY_TETHERING = 18000;
47 const 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;
50 const uint32_t RULE_PRIORITY_DEFAULT_NETWORK = 22000;
51 const uint32_t RULE_PRIORITY_DIRECTLY_CONNECTED = 23000;
52 const uint32_t RULE_PRIORITY_UNREACHABLE = 32000;
54 const uint32_t ROUTE_TABLE_LOCAL_NETWORK = 97;
55 const uint32_t ROUTE_TABLE_LEGACY_NETWORK = 98;
56 const uint32_t ROUTE_TABLE_LEGACY_SYSTEM = 99;
58 const char* const ROUTE_TABLE_NAME_LOCAL_NETWORK = "local_network";
59 const char* const ROUTE_TABLE_NAME_LEGACY_NETWORK = "legacy_network";
60 const char* const ROUTE_TABLE_NAME_LEGACY_SYSTEM = "legacy_system";
62 const char* const ROUTE_TABLE_NAME_LOCAL = "local";
63 const char* const ROUTE_TABLE_NAME_MAIN = "main";
65 // TODO: These values aren't defined by the Linux kernel, because our UID routing changes are not
66 // upstream (yet?), so we can't just pick them up from kernel headers. When (if?) the changes make
67 // it upstream, we'll remove this and rely on the kernel header values. For now, add a static assert
68 // that will warn us if upstream has given these values some other meaning.
69 const uint16_t FRA_UID_START = 18;
70 const uint16_t FRA_UID_END = 19;
71 static_assert(FRA_UID_START > FRA_MAX,
72 "Android-specific FRA_UID_{START,END} values also assigned in Linux uapi. "
73 "Check that these values match what the kernel does and then update this assertion.");
75 const uint16_t NETLINK_REQUEST_FLAGS = NLM_F_REQUEST | NLM_F_ACK;
76 const uint16_t NETLINK_CREATE_REQUEST_FLAGS = NETLINK_REQUEST_FLAGS | NLM_F_CREATE | NLM_F_EXCL;
78 const sockaddr_nl NETLINK_ADDRESS = {AF_NETLINK, 0, 0, 0};
80 const uint8_t AF_FAMILIES[] = {AF_INET, AF_INET6};
82 const char* const IP_VERSIONS[] = {"-4", "-6"};
84 const uid_t UID_ROOT = 0;
85 const char* const IIF_NONE = NULL;
86 const char* const OIF_NONE = NULL;
87 const bool ACTION_ADD = true;
88 const bool ACTION_DEL = false;
89 const bool MODIFY_NON_UID_BASED_RULES = true;
91 const char* const RT_TABLES_PATH = "/data/misc/net/rt_tables";
92 const int RT_TABLES_FLAGS = O_CREAT | O_TRUNC | O_WRONLY | O_NOFOLLOW | O_CLOEXEC;
93 const mode_t RT_TABLES_MODE = S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH; // mode 0644, rw-r--r--
95 const unsigned ROUTE_FLUSH_ATTEMPTS = 2;
97 // Avoids "non-constant-expression cannot be narrowed from type 'unsigned int' to 'unsigned short'"
98 // warnings when using RTA_LENGTH(x) inside static initializers (even when x is already uint16_t).
99 constexpr uint16_t U16_RTA_LENGTH(uint16_t x) {
100 return RTA_LENGTH(x);
103 // These are practically const, but can't be declared so, because they are used to initialize
104 // non-const pointers ("void* iov_base") in iovec arrays.
105 rtattr FRATTR_PRIORITY = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_PRIORITY };
106 rtattr FRATTR_TABLE = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_TABLE };
107 rtattr FRATTR_FWMARK = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_FWMARK };
108 rtattr FRATTR_FWMASK = { U16_RTA_LENGTH(sizeof(uint32_t)), FRA_FWMASK };
109 rtattr FRATTR_UID_START = { U16_RTA_LENGTH(sizeof(uid_t)), FRA_UID_START };
110 rtattr FRATTR_UID_END = { U16_RTA_LENGTH(sizeof(uid_t)), FRA_UID_END };
112 rtattr RTATTR_TABLE = { U16_RTA_LENGTH(sizeof(uint32_t)), RTA_TABLE };
113 rtattr RTATTR_OIF = { U16_RTA_LENGTH(sizeof(uint32_t)), RTA_OIF };
115 uint8_t PADDING_BUFFER[RTA_ALIGNTO] = {0, 0, 0, 0};
117 // END CONSTANTS ----------------------------------------------------------------------------------
119 // No locks needed because RouteController is accessed only from one thread (in CommandListener).
120 std::map<std::string, uint32_t> interfaceToTable;
122 uint32_t getRouteTableForInterface(const char* interface) {
123 uint32_t index = if_nametoindex(interface);
125 index += RouteController::ROUTE_TABLE_OFFSET_FROM_INDEX;
126 interfaceToTable[interface] = index;
129 // If the interface goes away if_nametoindex() will return 0 but we still need to know
130 // the index so we can remove the rules and routes.
131 auto iter = interfaceToTable.find(interface);
132 if (iter == interfaceToTable.end()) {
133 ALOGE("cannot find interface %s", interface);
134 return RT_TABLE_UNSPEC;
139 void addTableName(uint32_t table, const std::string& name, std::string* contents) {
140 char tableString[UINT32_STRLEN];
141 snprintf(tableString, sizeof(tableString), "%u", table);
142 *contents += tableString;
148 // Doesn't return success/failure as the file is optional; it's okay if we fail to update it.
149 void updateTableNamesFile() {
150 std::string contents;
152 addTableName(RT_TABLE_LOCAL, ROUTE_TABLE_NAME_LOCAL, &contents);
153 addTableName(RT_TABLE_MAIN, ROUTE_TABLE_NAME_MAIN, &contents);
155 addTableName(ROUTE_TABLE_LOCAL_NETWORK, ROUTE_TABLE_NAME_LOCAL_NETWORK, &contents);
156 addTableName(ROUTE_TABLE_LEGACY_NETWORK, ROUTE_TABLE_NAME_LEGACY_NETWORK, &contents);
157 addTableName(ROUTE_TABLE_LEGACY_SYSTEM, ROUTE_TABLE_NAME_LEGACY_SYSTEM, &contents);
159 for (const auto& entry : interfaceToTable) {
160 addTableName(entry.second, entry.first, &contents);
163 int fd = open(RT_TABLES_PATH, RT_TABLES_FLAGS, RT_TABLES_MODE);
165 ALOGE("failed to create %s (%s)", RT_TABLES_PATH, strerror(errno));
168 // File creation is affected by umask, so make sure the right mode bits are set.
169 if (fchmod(fd, RT_TABLES_MODE) == -1) {
170 ALOGE("failed to set mode 0%o on %s (%s)", RT_TABLES_MODE, RT_TABLES_PATH, strerror(errno));
172 ssize_t bytesWritten = write(fd, contents.data(), contents.size());
173 if (bytesWritten != static_cast<ssize_t>(contents.size())) {
174 ALOGE("failed to write to %s (%zd vs %zu bytes) (%s)", RT_TABLES_PATH, bytesWritten,
175 contents.size(), strerror(errno));
180 // Sends a netlink request and expects an ack.
181 // |iov| is an array of struct iovec that contains the netlink message payload.
182 // The netlink header is generated by this function based on |action| and |flags|.
183 // Returns -errno if there was an error or if the kernel reported an error.
184 WARN_UNUSED_RESULT int sendNetlinkRequest(uint16_t action, uint16_t flags, iovec* iov, int iovlen) {
186 .nlmsg_type = action,
187 .nlmsg_flags = flags,
189 iov[0].iov_base = &nlmsg;
190 iov[0].iov_len = sizeof(nlmsg);
191 for (int i = 0; i < iovlen; ++i) {
192 nlmsg.nlmsg_len += iov[i].iov_len;
201 int sock = socket(AF_NETLINK, SOCK_DGRAM, NETLINK_ROUTE);
203 connect(sock, reinterpret_cast<const sockaddr*>(&NETLINK_ADDRESS),
204 sizeof(NETLINK_ADDRESS)) != -1 &&
205 writev(sock, iov, iovlen) != -1 &&
206 (ret = recv(sock, &response, sizeof(response), 0)) != -1) {
207 if (ret == sizeof(response)) {
208 ret = response.err.error; // Netlink errors are negative errno.
210 ALOGE("netlink response contains error (%s)", strerror(-ret));
213 ALOGE("bad netlink response message size (%d != %zu)", ret, sizeof(response));
217 ALOGE("netlink socket/connect/writev/recv failed (%s)", strerror(errno));
228 // Returns 0 on success or negative errno on failure.
229 int padInterfaceName(const char* input, char* name, size_t* length, uint16_t* padding) {
235 *length = strlcpy(name, input, IFNAMSIZ) + 1;
236 if (*length > IFNAMSIZ) {
237 ALOGE("interface name too long (%zu > %u)", *length, IFNAMSIZ);
238 return -ENAMETOOLONG;
240 *padding = RTA_SPACE(*length) - RTA_LENGTH(*length);
244 // Adds or removes a routing rule for IPv4 and IPv6.
246 // + If |table| is non-zero, the rule points at the specified routing table. Otherwise, the rule
247 // returns ENETUNREACH.
248 // + If |mask| is non-zero, the rule matches the specified fwmark and mask. Otherwise, |fwmark| is
250 // + If |iif| is non-NULL, the rule matches the specified incoming interface.
251 // + If |oif| is non-NULL, the rule matches the specified outgoing interface.
252 // + If |uidStart| and |uidEnd| are not INVALID_UID, the rule matches packets from UIDs in that
253 // range (inclusive). Otherwise, the rule matches packets from all UIDs.
255 // Returns 0 on success or negative errno on failure.
256 WARN_UNUSED_RESULT int modifyIpRule(uint16_t action, uint32_t priority, uint32_t table,
257 uint32_t fwmark, uint32_t mask, const char* iif,
258 const char* oif, uid_t uidStart, uid_t uidEnd) {
259 // Ensure that if you set a bit in the fwmark, it's not being ignored by the mask.
260 if (fwmark & ~mask) {
261 ALOGE("mask 0x%x does not select all the bits set in fwmark 0x%x", mask, fwmark);
265 // Interface names must include exactly one terminating NULL and be properly padded, or older
266 // kernels will refuse to delete rules.
267 char iifName[IFNAMSIZ], oifName[IFNAMSIZ];
268 size_t iifLength, oifLength;
269 uint16_t iifPadding, oifPadding;
270 if (int ret = padInterfaceName(iif, iifName, &iifLength, &iifPadding)) {
273 if (int ret = padInterfaceName(oif, oifName, &oifLength, &oifPadding)) {
277 // Either both start and end UID must be specified, or neither.
278 if ((uidStart == INVALID_UID) != (uidEnd == INVALID_UID)) {
279 ALOGE("incompatible start and end UIDs (%u vs %u)", uidStart, uidEnd);
282 bool isUidRule = (uidStart != INVALID_UID);
284 // Assemble a rule request and put it in an array of iovec structures.
285 fib_rule_hdr rule = {
286 .action = static_cast<uint8_t>(table != RT_TABLE_UNSPEC ? FR_ACT_TO_TBL :
290 rtattr fraIifName = { U16_RTA_LENGTH(iifLength), FRA_IIFNAME };
291 rtattr fraOifName = { U16_RTA_LENGTH(oifLength), FRA_OIFNAME };
295 { &rule, sizeof(rule) },
296 { &FRATTR_PRIORITY, sizeof(FRATTR_PRIORITY) },
297 { &priority, sizeof(priority) },
298 { &FRATTR_TABLE, table != RT_TABLE_UNSPEC ? sizeof(FRATTR_TABLE) : 0 },
299 { &table, table != RT_TABLE_UNSPEC ? sizeof(table) : 0 },
300 { &FRATTR_FWMARK, mask ? sizeof(FRATTR_FWMARK) : 0 },
301 { &fwmark, mask ? sizeof(fwmark) : 0 },
302 { &FRATTR_FWMASK, mask ? sizeof(FRATTR_FWMASK) : 0 },
303 { &mask, mask ? sizeof(mask) : 0 },
304 { &FRATTR_UID_START, isUidRule ? sizeof(FRATTR_UID_START) : 0 },
305 { &uidStart, isUidRule ? sizeof(uidStart) : 0 },
306 { &FRATTR_UID_END, isUidRule ? sizeof(FRATTR_UID_END) : 0 },
307 { &uidEnd, isUidRule ? sizeof(uidEnd) : 0 },
308 { &fraIifName, iif != IIF_NONE ? sizeof(fraIifName) : 0 },
309 { iifName, iifLength },
310 { PADDING_BUFFER, iifPadding },
311 { &fraOifName, oif != OIF_NONE ? sizeof(fraOifName) : 0 },
312 { oifName, oifLength },
313 { PADDING_BUFFER, oifPadding },
316 uint16_t flags = (action == RTM_NEWRULE) ? NETLINK_CREATE_REQUEST_FLAGS : NETLINK_REQUEST_FLAGS;
317 for (size_t i = 0; i < ARRAY_SIZE(AF_FAMILIES); ++i) {
318 rule.family = AF_FAMILIES[i];
319 if (int ret = sendNetlinkRequest(action, flags, iov, ARRAY_SIZE(iov))) {
327 WARN_UNUSED_RESULT int modifyIpRule(uint16_t action, uint32_t priority, uint32_t table,
328 uint32_t fwmark, uint32_t mask) {
329 return modifyIpRule(action, priority, table, fwmark, mask, IIF_NONE, OIF_NONE, INVALID_UID,
333 // Adds or deletes an IPv4 or IPv6 route.
334 // Returns 0 on success or negative errno on failure.
335 WARN_UNUSED_RESULT int modifyIpRoute(uint16_t action, uint32_t table, const char* interface,
336 const char* destination, const char* nexthop) {
337 // At least the destination must be non-null.
339 ALOGE("null destination");
344 uint8_t rawAddress[sizeof(in6_addr)];
346 uint8_t prefixLength;
347 int rawLength = parsePrefix(destination, &family, rawAddress, sizeof(rawAddress),
350 ALOGE("parsePrefix failed for destination %s (%s)", destination, strerror(-rawLength));
354 if (static_cast<size_t>(rawLength) > sizeof(rawAddress)) {
355 ALOGE("impossible! address too long (%d vs %zu)", rawLength, sizeof(rawAddress));
356 return -ENOBUFS; // Cannot happen; parsePrefix only supports IPv4 and IPv6.
359 uint8_t type = RTN_UNICAST;
361 uint8_t rawNexthop[sizeof(in6_addr)];
363 if (nexthop && !strcmp(nexthop, "unreachable")) {
364 type = RTN_UNREACHABLE;
365 // 'interface' is likely non-NULL, as the caller (modifyRoute()) likely used it to lookup
366 // the table number. But it's an error to specify an interface ("dev ...") or a nexthop for
367 // unreachable routes, so nuke them. (IPv6 allows them to be specified; IPv4 doesn't.)
368 interface = OIF_NONE;
371 // If an interface was specified, find the ifindex.
372 if (interface != OIF_NONE) {
373 ifindex = if_nametoindex(interface);
375 ALOGE("cannot find interface %s", interface);
380 // If a nexthop was specified, parse it as the same family as the prefix.
381 if (nexthop && inet_pton(family, nexthop, rawNexthop) <= 0) {
382 ALOGE("inet_pton failed for nexthop %s", nexthop);
387 // Assemble a rtmsg and put it in an array of iovec structures.
389 .rtm_protocol = RTPROT_STATIC,
391 .rtm_family = family,
392 .rtm_dst_len = prefixLength,
393 .rtm_scope = static_cast<uint8_t>(nexthop ? RT_SCOPE_UNIVERSE : RT_SCOPE_LINK),
396 rtattr rtaDst = { U16_RTA_LENGTH(rawLength), RTA_DST };
397 rtattr rtaGateway = { U16_RTA_LENGTH(rawLength), RTA_GATEWAY };
401 { &route, sizeof(route) },
402 { &RTATTR_TABLE, sizeof(RTATTR_TABLE) },
403 { &table, sizeof(table) },
404 { &rtaDst, sizeof(rtaDst) },
405 { rawAddress, static_cast<size_t>(rawLength) },
406 { &RTATTR_OIF, interface != OIF_NONE ? sizeof(RTATTR_OIF) : 0 },
407 { &ifindex, interface != OIF_NONE ? sizeof(ifindex) : 0 },
408 { &rtaGateway, nexthop ? sizeof(rtaGateway) : 0 },
409 { rawNexthop, nexthop ? static_cast<size_t>(rawLength) : 0 },
412 uint16_t flags = (action == RTM_NEWROUTE) ? NETLINK_CREATE_REQUEST_FLAGS :
413 NETLINK_REQUEST_FLAGS;
414 return sendNetlinkRequest(action, flags, iov, ARRAY_SIZE(iov));
417 // An iptables rule to mark incoming packets on a network with the netId of the network.
419 // This is so that the kernel can:
420 // + Use the right fwmark for (and thus correctly route) replies (e.g.: TCP RST, ICMP errors, ping
421 // replies, SYN-ACKs, etc).
422 // + Mark sockets that accept connections from this interface so that the connection stays on the
424 WARN_UNUSED_RESULT int modifyIncomingPacketMark(unsigned netId, const char* interface,
425 Permission permission, bool add) {
428 fwmark.netId = netId;
429 fwmark.explicitlySelected = true;
430 fwmark.protectedFromVpn = true;
431 fwmark.permission = permission;
433 char markString[UINT32_HEX_STRLEN];
434 snprintf(markString, sizeof(markString), "0x%x", fwmark.intValue);
436 if (execIptables(V4V6, "-t", "mangle", add ? "-A" : "-D", "INPUT", "-i", interface, "-j",
437 "MARK", "--set-mark", markString, NULL)) {
438 ALOGE("failed to change iptables rule that sets incoming packet mark");
445 // A rule to route responses to the local network forwarded via the VPN.
447 // When a VPN is in effect, packets from the local network to upstream networks are forwarded into
448 // the VPN's tunnel interface. When the VPN forwards the responses, they emerge out of the tunnel.
449 WARN_UNUSED_RESULT int modifyVpnOutputToLocalRule(const char* vpnInterface, bool add) {
450 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_VPN_OUTPUT_TO_LOCAL,
451 ROUTE_TABLE_LOCAL_NETWORK, MARK_UNSET, MARK_UNSET, vpnInterface, OIF_NONE,
452 INVALID_UID, INVALID_UID);
455 // A rule to route all traffic from a given set of UIDs to go over the VPN.
457 // Notice that this rule doesn't use the netId. I.e., no matter what netId the user's socket may
458 // have, if they are subject to this VPN, their traffic has to go through it. Allows the traffic to
459 // bypass the VPN if the protectedFromVpn bit is set.
460 WARN_UNUSED_RESULT int modifyVpnUidRangeRule(uint32_t table, uid_t uidStart, uid_t uidEnd,
461 bool secure, bool add) {
465 fwmark.protectedFromVpn = false;
466 mask.protectedFromVpn = true;
471 priority = RULE_PRIORITY_SECURE_VPN;
473 priority = RULE_PRIORITY_BYPASSABLE_VPN;
475 fwmark.explicitlySelected = false;
476 mask.explicitlySelected = true;
479 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, priority, table, fwmark.intValue,
480 mask.intValue, IIF_NONE, OIF_NONE, uidStart, uidEnd);
483 // A rule to allow system apps to send traffic over this VPN even if they are not part of the target
486 // This is needed for DnsProxyListener to correctly resolve a request for a user who is in the
487 // target set, but where the DnsProxyListener itself is not.
488 WARN_UNUSED_RESULT int modifyVpnSystemPermissionRule(unsigned netId, uint32_t table, bool secure,
493 fwmark.netId = netId;
494 mask.netId = FWMARK_NET_ID_MASK;
496 fwmark.permission = PERMISSION_SYSTEM;
497 mask.permission = PERMISSION_SYSTEM;
499 uint32_t priority = secure ? RULE_PRIORITY_SECURE_VPN : RULE_PRIORITY_BYPASSABLE_VPN;
501 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, priority, table, fwmark.intValue,
505 // A rule to route traffic based on an explicitly chosen network.
507 // Supports apps that use the multinetwork APIs to restrict their traffic to a network.
509 // Even though we check permissions at the time we set a netId into the fwmark of a socket, we need
510 // to check it again in the rules here, because a network's permissions may have been updated via
511 // modifyNetworkPermission().
512 WARN_UNUSED_RESULT int modifyExplicitNetworkRule(unsigned netId, uint32_t table,
513 Permission permission, uid_t uidStart,
514 uid_t uidEnd, bool add) {
518 fwmark.netId = netId;
519 mask.netId = FWMARK_NET_ID_MASK;
521 fwmark.explicitlySelected = true;
522 mask.explicitlySelected = true;
524 fwmark.permission = permission;
525 mask.permission = permission;
527 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_EXPLICIT_NETWORK, table,
528 fwmark.intValue, mask.intValue, IIF_NONE, OIF_NONE, uidStart, uidEnd);
531 // A rule to route traffic based on a chosen outgoing interface.
533 // Supports apps that use SO_BINDTODEVICE or IP_PKTINFO options and the kernel that already knows
534 // the outgoing interface (typically for link-local communications).
535 WARN_UNUSED_RESULT int modifyOutputInterfaceRule(const char* interface, uint32_t table,
536 Permission permission, uid_t uidStart,
537 uid_t uidEnd, bool add) {
541 fwmark.permission = permission;
542 mask.permission = permission;
544 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_OUTPUT_INTERFACE, table,
545 fwmark.intValue, mask.intValue, IIF_NONE, interface, uidStart, uidEnd);
548 // A rule to route traffic based on the chosen network.
550 // This is for sockets that have not explicitly requested a particular network, but have been
551 // bound to one when they called connect(). This ensures that sockets connected on a particular
552 // network stay on that network even if the default network changes.
553 WARN_UNUSED_RESULT int modifyImplicitNetworkRule(unsigned netId, uint32_t table,
554 Permission permission, bool add) {
558 fwmark.netId = netId;
559 mask.netId = FWMARK_NET_ID_MASK;
561 fwmark.explicitlySelected = false;
562 mask.explicitlySelected = true;
564 fwmark.permission = permission;
565 mask.permission = permission;
567 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_IMPLICIT_NETWORK, table,
568 fwmark.intValue, mask.intValue);
571 // A rule to enable split tunnel VPNs.
573 // If a packet with a VPN's netId doesn't find a route in the VPN's routing table, it's allowed to
574 // go over the default network, provided it wasn't explicitly restricted to the VPN and has the
575 // permissions required by the default network.
576 WARN_UNUSED_RESULT int modifyVpnFallthroughRule(uint16_t action, unsigned vpnNetId,
577 const char* physicalInterface,
578 Permission permission) {
579 uint32_t table = getRouteTableForInterface(physicalInterface);
580 if (table == RT_TABLE_UNSPEC) {
587 fwmark.netId = vpnNetId;
588 mask.netId = FWMARK_NET_ID_MASK;
590 fwmark.explicitlySelected = false;
591 mask.explicitlySelected = true;
593 fwmark.permission = permission;
594 mask.permission = permission;
596 return modifyIpRule(action, RULE_PRIORITY_VPN_FALLTHROUGH, table, fwmark.intValue,
600 // Add rules to allow legacy routes added through the requestRouteToHost() API.
601 WARN_UNUSED_RESULT int addLegacyRouteRules() {
605 fwmark.explicitlySelected = false;
606 mask.explicitlySelected = true;
608 // Rules to allow legacy routes to override the default network.
609 if (int ret = modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LEGACY_SYSTEM, ROUTE_TABLE_LEGACY_SYSTEM,
610 fwmark.intValue, mask.intValue)) {
613 if (int ret = modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LEGACY_NETWORK,
614 ROUTE_TABLE_LEGACY_NETWORK, fwmark.intValue, mask.intValue)) {
618 fwmark.permission = PERMISSION_SYSTEM;
619 mask.permission = PERMISSION_SYSTEM;
621 // A rule to allow legacy routes from system apps to override VPNs.
622 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_VPN_OVERRIDE_SYSTEM, ROUTE_TABLE_LEGACY_SYSTEM,
623 fwmark.intValue, mask.intValue);
626 // Add rules to lookup the local network when specified explicitly or otherwise.
627 WARN_UNUSED_RESULT int addLocalNetworkRules(unsigned localNetId) {
628 if (int ret = modifyExplicitNetworkRule(localNetId, ROUTE_TABLE_LOCAL_NETWORK, PERMISSION_NONE,
629 INVALID_UID, INVALID_UID, ACTION_ADD)) {
636 fwmark.explicitlySelected = false;
637 mask.explicitlySelected = true;
639 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LOCAL_NETWORK, ROUTE_TABLE_LOCAL_NETWORK,
640 fwmark.intValue, mask.intValue);
643 // Add a new rule to look up the 'main' table, with the same selectors as the "default network"
644 // rule, but with a lower priority. We will never create routes in the main table; it should only be
645 // used for directly-connected routes implicitly created by the kernel when adding IP addresses.
646 // This is necessary, for example, when adding a route through a directly-connected gateway: in
647 // order to add the route, there must already be a directly-connected route that covers the gateway.
648 WARN_UNUSED_RESULT int addDirectlyConnectedRule() {
652 fwmark.netId = NETID_UNSET;
653 mask.netId = FWMARK_NET_ID_MASK;
655 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_DIRECTLY_CONNECTED, RT_TABLE_MAIN,
656 fwmark.intValue, mask.intValue, IIF_NONE, OIF_NONE, UID_ROOT, UID_ROOT);
659 // Add an explicit unreachable rule close to the end of the prioriy list to make it clear that
660 // relying on the kernel-default "from all lookup main" rule at priority 32766 is not intended
661 // behaviour. We do flush the kernel-default rules at startup, but having an explicit unreachable
662 // rule will hopefully make things even clearer.
663 WARN_UNUSED_RESULT int addUnreachableRule() {
664 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_UNREACHABLE, RT_TABLE_UNSPEC, MARK_UNSET,
668 WARN_UNUSED_RESULT int modifyLocalNetwork(unsigned netId, const char* interface, bool add) {
669 if (int ret = modifyIncomingPacketMark(netId, interface, PERMISSION_NONE, add)) {
672 return modifyOutputInterfaceRule(interface, ROUTE_TABLE_LOCAL_NETWORK, PERMISSION_NONE,
673 INVALID_UID, INVALID_UID, add);
676 WARN_UNUSED_RESULT int modifyPhysicalNetwork(unsigned netId, const char* interface,
677 Permission permission, bool add) {
678 uint32_t table = getRouteTableForInterface(interface);
679 if (table == RT_TABLE_UNSPEC) {
683 if (int ret = modifyIncomingPacketMark(netId, interface, permission, add)) {
686 if (int ret = modifyExplicitNetworkRule(netId, table, permission, INVALID_UID, INVALID_UID,
690 if (int ret = modifyOutputInterfaceRule(interface, table, permission, INVALID_UID, INVALID_UID,
694 return modifyImplicitNetworkRule(netId, table, permission, add);
697 WARN_UNUSED_RESULT int modifyVirtualNetwork(unsigned netId, const char* interface,
698 const UidRanges& uidRanges, bool secure, bool add,
699 bool modifyNonUidBasedRules) {
700 uint32_t table = getRouteTableForInterface(interface);
701 if (table == RT_TABLE_UNSPEC) {
705 for (const UidRanges::Range& range : uidRanges.getRanges()) {
706 if (int ret = modifyVpnUidRangeRule(table, range.first, range.second, secure, add)) {
709 if (int ret = modifyExplicitNetworkRule(netId, table, PERMISSION_NONE, range.first,
710 range.second, add)) {
713 if (int ret = modifyOutputInterfaceRule(interface, table, PERMISSION_NONE, range.first,
714 range.second, add)) {
719 if (modifyNonUidBasedRules) {
720 if (int ret = modifyIncomingPacketMark(netId, interface, PERMISSION_NONE, add)) {
723 if (int ret = modifyVpnOutputToLocalRule(interface, add)) {
726 if (int ret = modifyVpnSystemPermissionRule(netId, table, secure, add)) {
729 return modifyExplicitNetworkRule(netId, table, PERMISSION_NONE, UID_ROOT, UID_ROOT, add);
735 WARN_UNUSED_RESULT int modifyDefaultNetwork(uint16_t action, const char* interface,
736 Permission permission) {
737 uint32_t table = getRouteTableForInterface(interface);
738 if (table == RT_TABLE_UNSPEC) {
745 fwmark.netId = NETID_UNSET;
746 mask.netId = FWMARK_NET_ID_MASK;
748 fwmark.permission = permission;
749 mask.permission = permission;
751 return modifyIpRule(action, RULE_PRIORITY_DEFAULT_NETWORK, table, fwmark.intValue,
755 WARN_UNUSED_RESULT int modifyTetheredNetwork(uint16_t action, const char* inputInterface,
756 const char* outputInterface) {
757 uint32_t table = getRouteTableForInterface(outputInterface);
758 if (table == RT_TABLE_UNSPEC) {
762 return modifyIpRule(action, RULE_PRIORITY_TETHERING, table, MARK_UNSET, MARK_UNSET,
763 inputInterface, OIF_NONE, INVALID_UID, INVALID_UID);
766 // Returns 0 on success or negative errno on failure.
767 WARN_UNUSED_RESULT int flushRules() {
768 for (size_t i = 0; i < ARRAY_SIZE(IP_VERSIONS); ++i) {
769 const char* argv[] = {
775 if (android_fork_execvp(ARRAY_SIZE(argv), const_cast<char**>(argv), NULL, false, false)) {
776 ALOGE("failed to flush rules");
783 // Adds or removes an IPv4 or IPv6 route to the specified table and, if it's a directly-connected
784 // route, to the main table as well.
785 // Returns 0 on success or negative errno on failure.
786 WARN_UNUSED_RESULT int modifyRoute(uint16_t action, const char* interface, const char* destination,
787 const char* nexthop, RouteController::TableType tableType) {
790 case RouteController::INTERFACE: {
791 table = getRouteTableForInterface(interface);
792 if (table == RT_TABLE_UNSPEC) {
797 case RouteController::LOCAL_NETWORK: {
798 table = ROUTE_TABLE_LOCAL_NETWORK;
801 case RouteController::LEGACY_NETWORK: {
802 table = ROUTE_TABLE_LEGACY_NETWORK;
805 case RouteController::LEGACY_SYSTEM: {
806 table = ROUTE_TABLE_LEGACY_SYSTEM;
811 int ret = modifyIpRoute(action, table, interface, destination, nexthop);
812 // We allow apps to call requestRouteToHost() multiple times with the same route, so ignore
813 // EEXIST failures when adding routes to legacy tables.
814 if (ret && !(action == RTM_NEWROUTE && ret == -EEXIST &&
815 (tableType == RouteController::LEGACY_NETWORK ||
816 tableType == RouteController::LEGACY_SYSTEM))) {
823 // Returns 0 on success or negative errno on failure.
824 WARN_UNUSED_RESULT int flushRoutes(const char* interface) {
825 uint32_t table = getRouteTableForInterface(interface);
826 if (table == RT_TABLE_UNSPEC) {
830 char tableString[UINT32_STRLEN];
831 snprintf(tableString, sizeof(tableString), "%u", table);
834 for (size_t i = 0; i < ARRAY_SIZE(IP_VERSIONS); ++i) {
835 const char* argv[] = {
844 // A flush works by dumping routes and deleting each route as it's returned, and it can
845 // fail if something else deletes the route between the dump and the delete. This can
846 // happen, for example, if an interface goes down while we're trying to flush its routes.
847 // So try multiple times and only return an error if the last attempt fails.
849 // TODO: replace this with our own netlink code.
850 unsigned attempts = 0;
853 err = android_fork_execvp(ARRAY_SIZE(argv), const_cast<char**>(argv),
856 } while (err != 0 && attempts < ROUTE_FLUSH_ATTEMPTS);
858 ALOGE("failed to flush %s routes in table %s after %d attempts",
859 IP_VERSIONS[i], tableString, attempts);
864 // If we failed to flush routes, the caller may elect to keep this interface around, so keep
865 // track of its name.
867 interfaceToTable.erase(interface);
875 int RouteController::Init(unsigned localNetId) {
876 if (int ret = flushRules()) {
879 if (int ret = addLegacyRouteRules()) {
882 if (int ret = addLocalNetworkRules(localNetId)) {
885 if (int ret = addDirectlyConnectedRule()) {
888 if (int ret = addUnreachableRule()) {
891 updateTableNamesFile();
895 int RouteController::addInterfaceToLocalNetwork(unsigned netId, const char* interface) {
896 return modifyLocalNetwork(netId, interface, ACTION_ADD);
899 int RouteController::removeInterfaceFromLocalNetwork(unsigned netId, const char* interface) {
900 return modifyLocalNetwork(netId, interface, ACTION_DEL);
903 int RouteController::addInterfaceToPhysicalNetwork(unsigned netId, const char* interface,
904 Permission permission) {
905 if (int ret = modifyPhysicalNetwork(netId, interface, permission, ACTION_ADD)) {
908 updateTableNamesFile();
912 int RouteController::removeInterfaceFromPhysicalNetwork(unsigned netId, const char* interface,
913 Permission permission) {
914 if (int ret = modifyPhysicalNetwork(netId, interface, permission, ACTION_DEL)) {
917 if (int ret = flushRoutes(interface)) {
920 updateTableNamesFile();
924 int RouteController::addInterfaceToVirtualNetwork(unsigned netId, const char* interface,
925 bool secure, const UidRanges& uidRanges) {
926 if (int ret = modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_ADD,
927 MODIFY_NON_UID_BASED_RULES)) {
930 updateTableNamesFile();
934 int RouteController::removeInterfaceFromVirtualNetwork(unsigned netId, const char* interface,
935 bool secure, const UidRanges& uidRanges) {
936 if (int ret = modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_DEL,
937 MODIFY_NON_UID_BASED_RULES)) {
940 if (int ret = flushRoutes(interface)) {
943 updateTableNamesFile();
947 int RouteController::modifyPhysicalNetworkPermission(unsigned netId, const char* interface,
948 Permission oldPermission,
949 Permission newPermission) {
950 // Add the new rules before deleting the old ones, to avoid race conditions.
951 if (int ret = modifyPhysicalNetwork(netId, interface, newPermission, ACTION_ADD)) {
954 return modifyPhysicalNetwork(netId, interface, oldPermission, ACTION_DEL);
957 int RouteController::addUsersToVirtualNetwork(unsigned netId, const char* interface, bool secure,
958 const UidRanges& uidRanges) {
959 return modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_ADD,
960 !MODIFY_NON_UID_BASED_RULES);
963 int RouteController::removeUsersFromVirtualNetwork(unsigned netId, const char* interface,
964 bool secure, const UidRanges& uidRanges) {
965 return modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_DEL,
966 !MODIFY_NON_UID_BASED_RULES);
969 int RouteController::addInterfaceToDefaultNetwork(const char* interface, Permission permission) {
970 return modifyDefaultNetwork(RTM_NEWRULE, interface, permission);
973 int RouteController::removeInterfaceFromDefaultNetwork(const char* interface,
974 Permission permission) {
975 return modifyDefaultNetwork(RTM_DELRULE, interface, permission);
978 int RouteController::addRoute(const char* interface, const char* destination, const char* nexthop,
979 TableType tableType) {
980 return modifyRoute(RTM_NEWROUTE, interface, destination, nexthop, tableType);
983 int RouteController::removeRoute(const char* interface, const char* destination,
984 const char* nexthop, TableType tableType) {
985 return modifyRoute(RTM_DELROUTE, interface, destination, nexthop, tableType);
988 int RouteController::enableTethering(const char* inputInterface, const char* outputInterface) {
989 return modifyTetheredNetwork(RTM_NEWRULE, inputInterface, outputInterface);
992 int RouteController::disableTethering(const char* inputInterface, const char* outputInterface) {
993 return modifyTetheredNetwork(RTM_DELRULE, inputInterface, outputInterface);
996 int RouteController::addVirtualNetworkFallthrough(unsigned vpnNetId, const char* physicalInterface,
997 Permission permission) {
998 return modifyVpnFallthroughRule(RTM_NEWRULE, vpnNetId, physicalInterface, permission);
1001 int RouteController::removeVirtualNetworkFallthrough(unsigned vpnNetId,
1002 const char* physicalInterface,
1003 Permission permission) {
1004 return modifyVpnFallthroughRule(RTM_DELRULE, vpnNetId, physicalInterface, permission);