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"
21 #include "DummyNetwork.h"
23 #define LOG_TAG "Netd"
25 #include "logwrap/logwrap.h"
26 #include "netutils/ifc.h"
27 #include "resolv_netid.h"
29 #include <arpa/inet.h>
31 #include <linux/fib_rules.h>
38 // BEGIN CONSTANTS --------------------------------------------------------------------------------
40 const uint32_t RULE_PRIORITY_VPN_OVERRIDE_SYSTEM = 10000;
41 const uint32_t RULE_PRIORITY_VPN_OVERRIDE_OIF = 10500;
42 const uint32_t RULE_PRIORITY_VPN_OUTPUT_TO_LOCAL = 11000;
43 const uint32_t RULE_PRIORITY_SECURE_VPN = 12000;
44 const uint32_t RULE_PRIORITY_EXPLICIT_NETWORK = 13000;
45 const uint32_t RULE_PRIORITY_OUTPUT_INTERFACE = 14000;
46 const uint32_t RULE_PRIORITY_LEGACY_SYSTEM = 15000;
47 const uint32_t RULE_PRIORITY_LEGACY_NETWORK = 16000;
48 const uint32_t RULE_PRIORITY_LOCAL_NETWORK = 17000;
49 const uint32_t RULE_PRIORITY_TETHERING = 18000;
50 const uint32_t RULE_PRIORITY_IMPLICIT_NETWORK = 19000;
51 const uint32_t RULE_PRIORITY_BYPASSABLE_VPN = 20000;
52 const uint32_t RULE_PRIORITY_VPN_FALLTHROUGH = 21000;
53 const uint32_t RULE_PRIORITY_DEFAULT_NETWORK = 22000;
54 const uint32_t RULE_PRIORITY_DIRECTLY_CONNECTED = 23000;
55 const uint32_t RULE_PRIORITY_UNREACHABLE = 32000;
57 const uint32_t ROUTE_TABLE_LOCAL_NETWORK = 97;
58 const uint32_t ROUTE_TABLE_LEGACY_NETWORK = 98;
59 const uint32_t ROUTE_TABLE_LEGACY_SYSTEM = 99;
61 const char* const ROUTE_TABLE_NAME_LOCAL_NETWORK = "local_network";
62 const char* const ROUTE_TABLE_NAME_LEGACY_NETWORK = "legacy_network";
63 const char* const ROUTE_TABLE_NAME_LEGACY_SYSTEM = "legacy_system";
65 const char* const ROUTE_TABLE_NAME_LOCAL = "local";
66 const char* const ROUTE_TABLE_NAME_MAIN = "main";
68 // TODO: These values aren't defined by the Linux kernel, because our UID routing changes are not
69 // upstream (yet?), so we can't just pick them up from kernel headers. When (if?) the changes make
70 // it upstream, we'll remove this and rely on the kernel header values. For now, add a static assert
71 // that will warn us if upstream has given these values some other meaning.
72 const uint16_t FRA_UID_START = 18;
73 const uint16_t FRA_UID_END = 19;
74 static_assert(FRA_UID_START > FRA_MAX,
75 "Android-specific FRA_UID_{START,END} values also assigned in Linux uapi. "
76 "Check that these values match what the kernel does and then update this assertion.");
78 const uint16_t NETLINK_REQUEST_FLAGS = NLM_F_REQUEST | NLM_F_ACK;
79 const uint16_t NETLINK_CREATE_REQUEST_FLAGS = NETLINK_REQUEST_FLAGS | NLM_F_CREATE | NLM_F_EXCL;
81 const sockaddr_nl NETLINK_ADDRESS = {AF_NETLINK, 0, 0, 0};
83 const uint8_t AF_FAMILIES[] = {AF_INET, AF_INET6};
85 const char* const IP_VERSIONS[] = {"-4", "-6"};
87 const uid_t UID_ROOT = 0;
88 const char* const IIF_LOOPBACK = "lo";
89 const char* const IIF_NONE = NULL;
90 const char* const OIF_NONE = NULL;
91 const bool ACTION_ADD = true;
92 const bool ACTION_DEL = false;
93 const bool MODIFY_NON_UID_BASED_RULES = true;
95 const char* const RT_TABLES_PATH = "/data/misc/net/rt_tables";
96 const int RT_TABLES_FLAGS = O_CREAT | O_TRUNC | O_WRONLY | O_NOFOLLOW | O_CLOEXEC;
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 int fd = open(RT_TABLES_PATH, RT_TABLES_FLAGS, RT_TABLES_MODE);
169 ALOGE("failed to create %s (%s)", RT_TABLES_PATH, strerror(errno));
172 // File creation is affected by umask, so make sure the right mode bits are set.
173 if (fchmod(fd, RT_TABLES_MODE) == -1) {
174 ALOGE("failed to set mode 0%o on %s (%s)", RT_TABLES_MODE, RT_TABLES_PATH, strerror(errno));
176 ssize_t bytesWritten = write(fd, contents.data(), contents.size());
177 if (bytesWritten != static_cast<ssize_t>(contents.size())) {
178 ALOGE("failed to write to %s (%zd vs %zu bytes) (%s)", RT_TABLES_PATH, bytesWritten,
179 contents.size(), strerror(errno));
184 // Sends a netlink request and expects an ack.
185 // |iov| is an array of struct iovec that contains the netlink message payload.
186 // The netlink header is generated by this function based on |action| and |flags|.
187 // Returns -errno if there was an error or if the kernel reported an error.
188 WARN_UNUSED_RESULT int sendNetlinkRequest(uint16_t action, uint16_t flags, iovec* iov, int iovlen) {
190 .nlmsg_type = action,
191 .nlmsg_flags = flags,
193 iov[0].iov_base = &nlmsg;
194 iov[0].iov_len = sizeof(nlmsg);
195 for (int i = 0; i < iovlen; ++i) {
196 nlmsg.nlmsg_len += iov[i].iov_len;
205 int sock = socket(AF_NETLINK, SOCK_DGRAM, NETLINK_ROUTE);
207 connect(sock, reinterpret_cast<const sockaddr*>(&NETLINK_ADDRESS),
208 sizeof(NETLINK_ADDRESS)) != -1 &&
209 writev(sock, iov, iovlen) != -1 &&
210 (ret = recv(sock, &response, sizeof(response), 0)) != -1) {
211 if (ret == sizeof(response)) {
212 ret = response.err.error; // Netlink errors are negative errno.
214 ALOGE("netlink response contains error (%s)", strerror(-ret));
217 ALOGE("bad netlink response message size (%d != %zu)", ret, sizeof(response));
221 ALOGE("netlink socket/connect/writev/recv failed (%s)", strerror(errno));
232 // Returns 0 on success or negative errno on failure.
233 int padInterfaceName(const char* input, char* name, size_t* length, uint16_t* padding) {
239 *length = strlcpy(name, input, IFNAMSIZ) + 1;
240 if (*length > IFNAMSIZ) {
241 ALOGE("interface name too long (%zu > %u)", *length, IFNAMSIZ);
242 return -ENAMETOOLONG;
244 *padding = RTA_SPACE(*length) - RTA_LENGTH(*length);
248 // Adds or removes a routing rule for IPv4 and IPv6.
250 // + If |priority| is RULE_PRIORITY_UNREACHABLE, the rule returns ENETUNREACH (i.e., specifies an
251 // action of FR_ACT_UNREACHABLE). Otherwise, the rule specifies an action of FR_ACT_TO_TBL.
252 // + If |table| is non-zero, the rule points at the specified routing table. Otherwise, the table is
253 // unspecified. An unspecified table is only allowed when deleting a rule.
254 // + If |mask| is non-zero, the rule matches the specified fwmark and mask. Otherwise, |fwmark| is
256 // + If |iif| is non-NULL, the rule matches the specified incoming interface.
257 // + If |oif| is non-NULL, the rule matches the specified outgoing interface.
258 // + If |uidStart| and |uidEnd| are not INVALID_UID, the rule matches packets from UIDs in that
259 // range (inclusive). Otherwise, the rule matches packets from all UIDs.
261 // Returns 0 on success or negative errno on failure.
262 WARN_UNUSED_RESULT int modifyIpRule(uint16_t action, uint32_t priority, uint32_t table,
263 uint32_t fwmark, uint32_t mask, const char* iif,
264 const char* oif, uid_t uidStart, uid_t uidEnd) {
265 // Ensure that if you set a bit in the fwmark, it's not being ignored by the mask.
266 if (fwmark & ~mask) {
267 ALOGE("mask 0x%x does not select all the bits set in fwmark 0x%x", mask, fwmark);
271 // Interface names must include exactly one terminating NULL and be properly padded, or older
272 // kernels will refuse to delete rules.
273 char iifName[IFNAMSIZ], oifName[IFNAMSIZ];
274 size_t iifLength, oifLength;
275 uint16_t iifPadding, oifPadding;
276 if (int ret = padInterfaceName(iif, iifName, &iifLength, &iifPadding)) {
279 if (int ret = padInterfaceName(oif, oifName, &oifLength, &oifPadding)) {
283 // Either both start and end UID must be specified, or neither.
284 if ((uidStart == INVALID_UID) != (uidEnd == INVALID_UID)) {
285 ALOGE("incompatible start and end UIDs (%u vs %u)", uidStart, uidEnd);
288 bool isUidRule = (uidStart != INVALID_UID);
290 // Assemble a rule request and put it in an array of iovec structures.
291 fib_rule_hdr rule = {
292 .action = static_cast<uint8_t>(priority != RULE_PRIORITY_UNREACHABLE ? FR_ACT_TO_TBL :
294 // Note that here we're implicitly setting rule.table to 0. When we want to specify a
295 // non-zero table, we do this via the FRATTR_TABLE attribute.
298 // Don't ever create a rule that looks up table 0, because table 0 is the local table.
299 // It's OK to specify a table ID of 0 when deleting a rule, because that doesn't actually select
300 // table 0, it's a wildcard that matches anything.
301 if (table == RT_TABLE_UNSPEC && rule.action == FR_ACT_TO_TBL && action != RTM_DELRULE) {
302 ALOGE("RT_TABLE_UNSPEC only allowed when deleting rules");
306 rtattr fraIifName = { U16_RTA_LENGTH(iifLength), FRA_IIFNAME };
307 rtattr fraOifName = { U16_RTA_LENGTH(oifLength), FRA_OIFNAME };
311 { &rule, sizeof(rule) },
312 { &FRATTR_PRIORITY, sizeof(FRATTR_PRIORITY) },
313 { &priority, sizeof(priority) },
314 { &FRATTR_TABLE, table != RT_TABLE_UNSPEC ? sizeof(FRATTR_TABLE) : 0 },
315 { &table, table != RT_TABLE_UNSPEC ? sizeof(table) : 0 },
316 { &FRATTR_FWMARK, mask ? sizeof(FRATTR_FWMARK) : 0 },
317 { &fwmark, mask ? sizeof(fwmark) : 0 },
318 { &FRATTR_FWMASK, mask ? sizeof(FRATTR_FWMASK) : 0 },
319 { &mask, mask ? sizeof(mask) : 0 },
320 { &FRATTR_UID_START, isUidRule ? sizeof(FRATTR_UID_START) : 0 },
321 { &uidStart, isUidRule ? sizeof(uidStart) : 0 },
322 { &FRATTR_UID_END, isUidRule ? sizeof(FRATTR_UID_END) : 0 },
323 { &uidEnd, isUidRule ? sizeof(uidEnd) : 0 },
324 { &fraIifName, iif != IIF_NONE ? sizeof(fraIifName) : 0 },
325 { iifName, iifLength },
326 { PADDING_BUFFER, iifPadding },
327 { &fraOifName, oif != OIF_NONE ? sizeof(fraOifName) : 0 },
328 { oifName, oifLength },
329 { PADDING_BUFFER, oifPadding },
332 uint16_t flags = (action == RTM_NEWRULE) ? NETLINK_CREATE_REQUEST_FLAGS : NETLINK_REQUEST_FLAGS;
333 for (size_t i = 0; i < ARRAY_SIZE(AF_FAMILIES); ++i) {
334 rule.family = AF_FAMILIES[i];
335 if (int ret = sendNetlinkRequest(action, flags, iov, ARRAY_SIZE(iov))) {
343 WARN_UNUSED_RESULT int modifyIpRule(uint16_t action, uint32_t priority, uint32_t table,
344 uint32_t fwmark, uint32_t mask) {
345 return modifyIpRule(action, priority, table, fwmark, mask, IIF_NONE, OIF_NONE, INVALID_UID,
349 // Adds or deletes an IPv4 or IPv6 route.
350 // Returns 0 on success or negative errno on failure.
351 WARN_UNUSED_RESULT int modifyIpRoute(uint16_t action, uint32_t table, const char* interface,
352 const char* destination, const char* nexthop) {
353 // At least the destination must be non-null.
355 ALOGE("null destination");
360 uint8_t rawAddress[sizeof(in6_addr)];
362 uint8_t prefixLength;
363 int rawLength = parsePrefix(destination, &family, rawAddress, sizeof(rawAddress),
366 ALOGE("parsePrefix failed for destination %s (%s)", destination, strerror(-rawLength));
370 if (static_cast<size_t>(rawLength) > sizeof(rawAddress)) {
371 ALOGE("impossible! address too long (%d vs %zu)", rawLength, sizeof(rawAddress));
372 return -ENOBUFS; // Cannot happen; parsePrefix only supports IPv4 and IPv6.
375 uint8_t type = RTN_UNICAST;
377 uint8_t rawNexthop[sizeof(in6_addr)];
379 if (nexthop && !strcmp(nexthop, "unreachable")) {
380 type = RTN_UNREACHABLE;
381 // 'interface' is likely non-NULL, as the caller (modifyRoute()) likely used it to lookup
382 // the table number. But it's an error to specify an interface ("dev ...") or a nexthop for
383 // unreachable routes, so nuke them. (IPv6 allows them to be specified; IPv4 doesn't.)
384 interface = OIF_NONE;
386 } else if (nexthop && !strcmp(nexthop, "throw")) {
388 interface = OIF_NONE;
391 // If an interface was specified, find the ifindex.
392 if (interface != OIF_NONE) {
393 ifindex = if_nametoindex(interface);
395 ALOGE("cannot find interface %s", interface);
400 // If a nexthop was specified, parse it as the same family as the prefix.
401 if (nexthop && inet_pton(family, nexthop, rawNexthop) <= 0) {
402 ALOGE("inet_pton failed for nexthop %s", nexthop);
407 // Assemble a rtmsg and put it in an array of iovec structures.
409 .rtm_protocol = RTPROT_STATIC,
411 .rtm_family = family,
412 .rtm_dst_len = prefixLength,
413 .rtm_scope = static_cast<uint8_t>(nexthop ? RT_SCOPE_UNIVERSE : RT_SCOPE_LINK),
416 rtattr rtaDst = { U16_RTA_LENGTH(rawLength), RTA_DST };
417 rtattr rtaGateway = { U16_RTA_LENGTH(rawLength), RTA_GATEWAY };
421 { &route, sizeof(route) },
422 { &RTATTR_TABLE, sizeof(RTATTR_TABLE) },
423 { &table, sizeof(table) },
424 { &rtaDst, sizeof(rtaDst) },
425 { rawAddress, static_cast<size_t>(rawLength) },
426 { &RTATTR_OIF, interface != OIF_NONE ? sizeof(RTATTR_OIF) : 0 },
427 { &ifindex, interface != OIF_NONE ? sizeof(ifindex) : 0 },
428 { &rtaGateway, nexthop ? sizeof(rtaGateway) : 0 },
429 { rawNexthop, nexthop ? static_cast<size_t>(rawLength) : 0 },
432 uint16_t flags = (action == RTM_NEWROUTE) ? NETLINK_CREATE_REQUEST_FLAGS :
433 NETLINK_REQUEST_FLAGS;
434 return sendNetlinkRequest(action, flags, iov, ARRAY_SIZE(iov));
437 // An iptables rule to mark incoming packets on a network with the netId of the network.
439 // This is so that the kernel can:
440 // + Use the right fwmark for (and thus correctly route) replies (e.g.: TCP RST, ICMP errors, ping
441 // replies, SYN-ACKs, etc).
442 // + Mark sockets that accept connections from this interface so that the connection stays on the
444 WARN_UNUSED_RESULT int modifyIncomingPacketMark(unsigned netId, const char* interface,
445 Permission permission, bool add) {
448 fwmark.netId = netId;
449 fwmark.explicitlySelected = true;
450 fwmark.protectedFromVpn = true;
451 fwmark.permission = permission;
453 char markString[UINT32_HEX_STRLEN];
454 snprintf(markString, sizeof(markString), "0x%x", fwmark.intValue);
456 if (execIptables(V4V6, "-t", "mangle", add ? "-A" : "-D", "INPUT", "-i", interface, "-j",
457 "MARK", "--set-mark", markString, NULL)) {
458 ALOGE("failed to change iptables rule that sets incoming packet mark");
465 // A rule to route responses to the local network forwarded via the VPN.
467 // When a VPN is in effect, packets from the local network to upstream networks are forwarded into
468 // the VPN's tunnel interface. When the VPN forwards the responses, they emerge out of the tunnel.
469 WARN_UNUSED_RESULT int modifyVpnOutputToLocalRule(const char* vpnInterface, bool add) {
470 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_VPN_OUTPUT_TO_LOCAL,
471 ROUTE_TABLE_LOCAL_NETWORK, MARK_UNSET, MARK_UNSET, vpnInterface, OIF_NONE,
472 INVALID_UID, INVALID_UID);
475 // A rule to route all traffic from a given set of UIDs to go over the VPN.
477 // Notice that this rule doesn't use the netId. I.e., no matter what netId the user's socket may
478 // have, if they are subject to this VPN, their traffic has to go through it. Allows the traffic to
479 // bypass the VPN if the protectedFromVpn bit is set.
480 WARN_UNUSED_RESULT int modifyVpnUidRangeRule(uint32_t table, uid_t uidStart, uid_t uidEnd,
481 bool secure, bool add) {
485 fwmark.protectedFromVpn = false;
486 mask.protectedFromVpn = true;
491 priority = RULE_PRIORITY_SECURE_VPN;
493 priority = RULE_PRIORITY_BYPASSABLE_VPN;
495 fwmark.explicitlySelected = false;
496 mask.explicitlySelected = true;
499 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, priority, table, fwmark.intValue,
500 mask.intValue, IIF_LOOPBACK, OIF_NONE, uidStart, uidEnd);
503 // A rule to allow system apps to send traffic over this VPN even if they are not part of the target
506 // This is needed for DnsProxyListener to correctly resolve a request for a user who is in the
507 // target set, but where the DnsProxyListener itself is not.
508 WARN_UNUSED_RESULT int modifyVpnSystemPermissionRule(unsigned netId, uint32_t table, bool secure,
513 fwmark.netId = netId;
514 mask.netId = FWMARK_NET_ID_MASK;
516 fwmark.permission = PERMISSION_SYSTEM;
517 mask.permission = PERMISSION_SYSTEM;
519 uint32_t priority = secure ? RULE_PRIORITY_SECURE_VPN : RULE_PRIORITY_BYPASSABLE_VPN;
521 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, priority, table, fwmark.intValue,
525 // A rule to route traffic based on an explicitly chosen network.
527 // Supports apps that use the multinetwork APIs to restrict their traffic to a network.
529 // Even though we check permissions at the time we set a netId into the fwmark of a socket, we need
530 // to check it again in the rules here, because a network's permissions may have been updated via
531 // modifyNetworkPermission().
532 WARN_UNUSED_RESULT int modifyExplicitNetworkRule(unsigned netId, uint32_t table,
533 Permission permission, uid_t uidStart,
534 uid_t uidEnd, bool add) {
538 fwmark.netId = netId;
539 mask.netId = FWMARK_NET_ID_MASK;
541 fwmark.explicitlySelected = true;
542 mask.explicitlySelected = true;
544 fwmark.permission = permission;
545 mask.permission = permission;
547 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_EXPLICIT_NETWORK, table,
548 fwmark.intValue, mask.intValue, IIF_NONE, OIF_NONE, uidStart, uidEnd);
551 // A rule to route traffic based on a chosen outgoing interface.
553 // Supports apps that use SO_BINDTODEVICE or IP_PKTINFO options and the kernel that already knows
554 // the outgoing interface (typically for link-local communications).
555 WARN_UNUSED_RESULT int modifyOutputInterfaceRules(const char* interface, uint32_t table,
556 Permission permission, uid_t uidStart,
557 uid_t uidEnd, bool add) {
561 fwmark.permission = permission;
562 mask.permission = permission;
564 // If this rule does not specify a UID range, then also add a corresponding high-priority rule
565 // for UID. This covers forwarded packets and system daemons such as the tethering DHCP server.
566 if (uidStart == INVALID_UID && uidEnd == INVALID_UID) {
567 if (int ret = modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_VPN_OVERRIDE_OIF,
568 table, fwmark.intValue, mask.intValue, IIF_NONE, interface,
569 UID_ROOT, UID_ROOT)) {
574 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_OUTPUT_INTERFACE, table,
575 fwmark.intValue, mask.intValue, IIF_NONE, interface, uidStart, uidEnd);
578 // A rule to route traffic based on the chosen network.
580 // This is for sockets that have not explicitly requested a particular network, but have been
581 // bound to one when they called connect(). This ensures that sockets connected on a particular
582 // network stay on that network even if the default network changes.
583 WARN_UNUSED_RESULT int modifyImplicitNetworkRule(unsigned netId, uint32_t table,
584 Permission permission, bool add) {
588 fwmark.netId = netId;
589 mask.netId = FWMARK_NET_ID_MASK;
591 fwmark.explicitlySelected = false;
592 mask.explicitlySelected = true;
594 fwmark.permission = permission;
595 mask.permission = permission;
597 return modifyIpRule(add ? RTM_NEWRULE : RTM_DELRULE, RULE_PRIORITY_IMPLICIT_NETWORK, table,
598 fwmark.intValue, mask.intValue);
601 // A rule to enable split tunnel VPNs.
603 // If a packet with a VPN's netId doesn't find a route in the VPN's routing table, it's allowed to
604 // go over the default network, provided it wasn't explicitly restricted to the VPN and has the
605 // permissions required by the default network.
606 WARN_UNUSED_RESULT int modifyVpnFallthroughRule(uint16_t action, unsigned vpnNetId,
607 const char* physicalInterface,
608 Permission permission) {
609 uint32_t table = getRouteTableForInterface(physicalInterface);
610 if (table == RT_TABLE_UNSPEC) {
617 fwmark.netId = vpnNetId;
618 mask.netId = FWMARK_NET_ID_MASK;
620 fwmark.explicitlySelected = false;
621 mask.explicitlySelected = true;
623 fwmark.permission = permission;
624 mask.permission = permission;
626 return modifyIpRule(action, RULE_PRIORITY_VPN_FALLTHROUGH, table, fwmark.intValue,
630 // Add rules to allow legacy routes added through the requestRouteToHost() API.
631 WARN_UNUSED_RESULT int addLegacyRouteRules() {
635 fwmark.explicitlySelected = false;
636 mask.explicitlySelected = true;
638 // Rules to allow legacy routes to override the default network.
639 if (int ret = modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LEGACY_SYSTEM, ROUTE_TABLE_LEGACY_SYSTEM,
640 fwmark.intValue, mask.intValue)) {
643 if (int ret = modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LEGACY_NETWORK,
644 ROUTE_TABLE_LEGACY_NETWORK, fwmark.intValue, mask.intValue)) {
648 fwmark.permission = PERMISSION_SYSTEM;
649 mask.permission = PERMISSION_SYSTEM;
651 // A rule to allow legacy routes from system apps to override VPNs.
652 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_VPN_OVERRIDE_SYSTEM, ROUTE_TABLE_LEGACY_SYSTEM,
653 fwmark.intValue, mask.intValue);
656 // Add rules to lookup the local network when specified explicitly or otherwise.
657 WARN_UNUSED_RESULT int addLocalNetworkRules(unsigned localNetId) {
658 if (int ret = modifyExplicitNetworkRule(localNetId, ROUTE_TABLE_LOCAL_NETWORK, PERMISSION_NONE,
659 INVALID_UID, INVALID_UID, ACTION_ADD)) {
666 fwmark.explicitlySelected = false;
667 mask.explicitlySelected = true;
669 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_LOCAL_NETWORK, ROUTE_TABLE_LOCAL_NETWORK,
670 fwmark.intValue, mask.intValue);
673 int configureDummyNetwork() {
674 const char *interface = DummyNetwork::INTERFACE_NAME;
675 uint32_t table = getRouteTableForInterface(interface);
676 if (table == RT_TABLE_UNSPEC) {
677 // getRouteTableForInterface has already looged an error.
682 int ret = ifc_up(interface);
685 ALOGE("Can't bring up %s: %s", interface, strerror(errno));
689 if ((ret = modifyOutputInterfaceRules(interface, table, PERMISSION_NONE,
690 INVALID_UID, INVALID_UID, ACTION_ADD))) {
691 ALOGE("Can't create oif rules for %s: %s", interface, strerror(-ret));
695 if ((ret = modifyIpRoute(RTM_NEWROUTE, table, interface, "0.0.0.0/0", NULL))) {
696 ALOGE("Can't add IPv4 default route to %s: %s", interface, strerror(-ret));
700 if ((ret = modifyIpRoute(RTM_NEWROUTE, table, interface, "::/0", NULL))) {
701 ALOGE("Can't add IPv6 default route to %s: %s", interface, strerror(-ret));
708 // Add a new rule to look up the 'main' table, with the same selectors as the "default network"
709 // rule, but with a lower priority. We will never create routes in the main table; it should only be
710 // used for directly-connected routes implicitly created by the kernel when adding IP addresses.
711 // This is necessary, for example, when adding a route through a directly-connected gateway: in
712 // order to add the route, there must already be a directly-connected route that covers the gateway.
713 WARN_UNUSED_RESULT int addDirectlyConnectedRule() {
717 fwmark.netId = NETID_UNSET;
718 mask.netId = FWMARK_NET_ID_MASK;
720 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_DIRECTLY_CONNECTED, RT_TABLE_MAIN,
721 fwmark.intValue, mask.intValue, IIF_NONE, OIF_NONE, UID_ROOT, UID_ROOT);
724 // Add an explicit unreachable rule close to the end of the prioriy list to make it clear that
725 // relying on the kernel-default "from all lookup main" rule at priority 32766 is not intended
726 // behaviour. We do flush the kernel-default rules at startup, but having an explicit unreachable
727 // rule will hopefully make things even clearer.
728 WARN_UNUSED_RESULT int addUnreachableRule() {
729 return modifyIpRule(RTM_NEWRULE, RULE_PRIORITY_UNREACHABLE, RT_TABLE_UNSPEC, MARK_UNSET,
733 WARN_UNUSED_RESULT int modifyLocalNetwork(unsigned netId, const char* interface, bool add) {
734 if (int ret = modifyIncomingPacketMark(netId, interface, PERMISSION_NONE, add)) {
737 return modifyOutputInterfaceRules(interface, ROUTE_TABLE_LOCAL_NETWORK, PERMISSION_NONE,
738 INVALID_UID, INVALID_UID, add);
741 WARN_UNUSED_RESULT int modifyPhysicalNetwork(unsigned netId, const char* interface,
742 Permission permission, bool add) {
743 uint32_t table = getRouteTableForInterface(interface);
744 if (table == RT_TABLE_UNSPEC) {
748 if (int ret = modifyIncomingPacketMark(netId, interface, permission, add)) {
751 if (int ret = modifyExplicitNetworkRule(netId, table, permission, INVALID_UID, INVALID_UID,
755 if (int ret = modifyOutputInterfaceRules(interface, table, permission, INVALID_UID, INVALID_UID,
759 return modifyImplicitNetworkRule(netId, table, permission, add);
762 WARN_UNUSED_RESULT int modifyVirtualNetwork(unsigned netId, const char* interface,
763 const UidRanges& uidRanges, bool secure, bool add,
764 bool modifyNonUidBasedRules) {
765 uint32_t table = getRouteTableForInterface(interface);
766 if (table == RT_TABLE_UNSPEC) {
770 for (const UidRanges::Range& range : uidRanges.getRanges()) {
771 if (int ret = modifyVpnUidRangeRule(table, range.first, range.second, secure, add)) {
774 if (int ret = modifyExplicitNetworkRule(netId, table, PERMISSION_NONE, range.first,
775 range.second, add)) {
778 if (int ret = modifyOutputInterfaceRules(interface, table, PERMISSION_NONE, range.first,
779 range.second, add)) {
784 if (modifyNonUidBasedRules) {
785 if (int ret = modifyIncomingPacketMark(netId, interface, PERMISSION_NONE, add)) {
788 if (int ret = modifyVpnOutputToLocalRule(interface, add)) {
791 if (int ret = modifyVpnSystemPermissionRule(netId, table, secure, add)) {
794 return modifyExplicitNetworkRule(netId, table, PERMISSION_NONE, UID_ROOT, UID_ROOT, add);
800 WARN_UNUSED_RESULT int modifyDefaultNetwork(uint16_t action, const char* interface,
801 Permission permission) {
802 uint32_t table = getRouteTableForInterface(interface);
803 if (table == RT_TABLE_UNSPEC) {
810 fwmark.netId = NETID_UNSET;
811 mask.netId = FWMARK_NET_ID_MASK;
813 fwmark.permission = permission;
814 mask.permission = permission;
816 return modifyIpRule(action, RULE_PRIORITY_DEFAULT_NETWORK, table, fwmark.intValue,
820 WARN_UNUSED_RESULT int modifyTetheredNetwork(uint16_t action, const char* inputInterface,
821 const char* outputInterface) {
822 uint32_t table = getRouteTableForInterface(outputInterface);
823 if (table == RT_TABLE_UNSPEC) {
827 return modifyIpRule(action, RULE_PRIORITY_TETHERING, table, MARK_UNSET, MARK_UNSET,
828 inputInterface, OIF_NONE, INVALID_UID, INVALID_UID);
831 // Returns 0 on success or negative errno on failure.
832 WARN_UNUSED_RESULT int flushRules() {
833 for (size_t i = 0; i < ARRAY_SIZE(IP_VERSIONS); ++i) {
834 const char* argv[] = {
840 if (android_fork_execvp(ARRAY_SIZE(argv), const_cast<char**>(argv), NULL, false, false)) {
841 ALOGE("failed to flush rules");
848 // Adds or removes an IPv4 or IPv6 route to the specified table and, if it's a directly-connected
849 // route, to the main table as well.
850 // Returns 0 on success or negative errno on failure.
851 WARN_UNUSED_RESULT int modifyRoute(uint16_t action, const char* interface, const char* destination,
852 const char* nexthop, RouteController::TableType tableType) {
855 case RouteController::INTERFACE: {
856 table = getRouteTableForInterface(interface);
857 if (table == RT_TABLE_UNSPEC) {
862 case RouteController::LOCAL_NETWORK: {
863 table = ROUTE_TABLE_LOCAL_NETWORK;
866 case RouteController::LEGACY_NETWORK: {
867 table = ROUTE_TABLE_LEGACY_NETWORK;
870 case RouteController::LEGACY_SYSTEM: {
871 table = ROUTE_TABLE_LEGACY_SYSTEM;
876 int ret = modifyIpRoute(action, table, interface, destination, nexthop);
877 // Trying to add a route that already exists shouldn't cause an error.
878 if (ret && !(action == RTM_NEWROUTE && ret == -EEXIST)) {
885 // Returns 0 on success or negative errno on failure.
886 WARN_UNUSED_RESULT int flushRoutes(const char* interface) {
887 uint32_t table = getRouteTableForInterface(interface);
888 if (table == RT_TABLE_UNSPEC) {
892 char tableString[UINT32_STRLEN];
893 snprintf(tableString, sizeof(tableString), "%u", table);
896 for (size_t i = 0; i < ARRAY_SIZE(IP_VERSIONS); ++i) {
897 const char* argv[] = {
906 // A flush works by dumping routes and deleting each route as it's returned, and it can
907 // fail if something else deletes the route between the dump and the delete. This can
908 // happen, for example, if an interface goes down while we're trying to flush its routes.
909 // So try multiple times and only return an error if the last attempt fails.
911 // TODO: replace this with our own netlink code.
912 unsigned attempts = 0;
915 err = android_fork_execvp(ARRAY_SIZE(argv), const_cast<char**>(argv),
918 } while (err != 0 && attempts < ROUTE_FLUSH_ATTEMPTS);
920 ALOGE("failed to flush %s routes in table %s after %d attempts",
921 IP_VERSIONS[i], tableString, attempts);
926 // If we failed to flush routes, the caller may elect to keep this interface around, so keep
927 // track of its name.
929 interfaceToTable.erase(interface);
935 WARN_UNUSED_RESULT int clearTetheringRules(const char* inputInterface) {
938 ret = modifyIpRule(RTM_DELRULE, RULE_PRIORITY_TETHERING, 0, MARK_UNSET, MARK_UNSET,
939 inputInterface, OIF_NONE, INVALID_UID, INVALID_UID);
942 if (ret == -ENOENT) {
951 int RouteController::Init(unsigned localNetId) {
952 if (int ret = flushRules()) {
955 if (int ret = addLegacyRouteRules()) {
958 if (int ret = addLocalNetworkRules(localNetId)) {
961 if (int ret = addDirectlyConnectedRule()) {
964 if (int ret = addUnreachableRule()) {
967 // Don't complain if we can't add the dummy network, since not all devices support it.
968 configureDummyNetwork();
970 updateTableNamesFile();
974 int RouteController::addInterfaceToLocalNetwork(unsigned netId, const char* interface) {
975 return modifyLocalNetwork(netId, interface, ACTION_ADD);
978 int RouteController::removeInterfaceFromLocalNetwork(unsigned netId, const char* interface) {
979 return modifyLocalNetwork(netId, interface, ACTION_DEL);
982 int RouteController::addInterfaceToPhysicalNetwork(unsigned netId, const char* interface,
983 Permission permission) {
984 if (int ret = modifyPhysicalNetwork(netId, interface, permission, ACTION_ADD)) {
987 updateTableNamesFile();
991 int RouteController::removeInterfaceFromPhysicalNetwork(unsigned netId, const char* interface,
992 Permission permission) {
993 if (int ret = modifyPhysicalNetwork(netId, interface, permission, ACTION_DEL)) {
996 if (int ret = flushRoutes(interface)) {
999 if (int ret = clearTetheringRules(interface)) {
1002 updateTableNamesFile();
1006 int RouteController::addInterfaceToVirtualNetwork(unsigned netId, const char* interface,
1007 bool secure, const UidRanges& uidRanges) {
1008 if (int ret = modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_ADD,
1009 MODIFY_NON_UID_BASED_RULES)) {
1012 updateTableNamesFile();
1016 int RouteController::removeInterfaceFromVirtualNetwork(unsigned netId, const char* interface,
1017 bool secure, const UidRanges& uidRanges) {
1018 if (int ret = modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_DEL,
1019 MODIFY_NON_UID_BASED_RULES)) {
1022 if (int ret = flushRoutes(interface)) {
1025 updateTableNamesFile();
1029 int RouteController::modifyPhysicalNetworkPermission(unsigned netId, const char* interface,
1030 Permission oldPermission,
1031 Permission newPermission) {
1032 // Add the new rules before deleting the old ones, to avoid race conditions.
1033 if (int ret = modifyPhysicalNetwork(netId, interface, newPermission, ACTION_ADD)) {
1036 return modifyPhysicalNetwork(netId, interface, oldPermission, ACTION_DEL);
1039 int RouteController::addUsersToVirtualNetwork(unsigned netId, const char* interface, bool secure,
1040 const UidRanges& uidRanges) {
1041 return modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_ADD,
1042 !MODIFY_NON_UID_BASED_RULES);
1045 int RouteController::removeUsersFromVirtualNetwork(unsigned netId, const char* interface,
1046 bool secure, const UidRanges& uidRanges) {
1047 return modifyVirtualNetwork(netId, interface, uidRanges, secure, ACTION_DEL,
1048 !MODIFY_NON_UID_BASED_RULES);
1051 int RouteController::addInterfaceToDefaultNetwork(const char* interface, Permission permission) {
1052 return modifyDefaultNetwork(RTM_NEWRULE, interface, permission);
1055 int RouteController::removeInterfaceFromDefaultNetwork(const char* interface,
1056 Permission permission) {
1057 return modifyDefaultNetwork(RTM_DELRULE, interface, permission);
1060 int RouteController::addRoute(const char* interface, const char* destination, const char* nexthop,
1061 TableType tableType) {
1062 return modifyRoute(RTM_NEWROUTE, interface, destination, nexthop, tableType);
1065 int RouteController::removeRoute(const char* interface, const char* destination,
1066 const char* nexthop, TableType tableType) {
1067 return modifyRoute(RTM_DELROUTE, interface, destination, nexthop, tableType);
1070 int RouteController::enableTethering(const char* inputInterface, const char* outputInterface) {
1071 return modifyTetheredNetwork(RTM_NEWRULE, inputInterface, outputInterface);
1074 int RouteController::disableTethering(const char* inputInterface, const char* outputInterface) {
1075 return modifyTetheredNetwork(RTM_DELRULE, inputInterface, outputInterface);
1078 int RouteController::addVirtualNetworkFallthrough(unsigned vpnNetId, const char* physicalInterface,
1079 Permission permission) {
1080 return modifyVpnFallthroughRule(RTM_NEWRULE, vpnNetId, physicalInterface, permission);
1083 int RouteController::removeVirtualNetworkFallthrough(unsigned vpnNetId,
1084 const char* physicalInterface,
1085 Permission permission) {
1086 return modifyVpnFallthroughRule(RTM_DELRULE, vpnNetId, physicalInterface, permission);