hanzoai/visor
2
0
Fork 0
mirror of https://github.com/hanzoai/visor.git synced 2026-09-24 17:04:14 +00:00
Code Issues Packages Projects Releases Wiki Activity
Files
master
visor/test/syscalls/linux/socket_netlink_route.cc
T
Danny Hershko Shemesh c619177102 Refuse a veth peer attribute shorter than its ifinfomsg 
The peer payload is sliced at the size of a struct ifinfomsg before anything measures it, so 4 bytes panic the Sentry and take the sandbox with it. Any user reaches it after unshare -Ur -n.

Check the length instead of re-testing a lookup that already succeeded. A short payload now answers EINVAL.

Fixes #14516
2026-08-31 16:08:32 +03:00

3159 lines
106 KiB
C++
Raw Permalink Blame History

// Copyright 2018 The gVisor Authors.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include <arpa/inet.h>
#include <fcntl.h>
#include <ifaddrs.h>
#include <linux/fib_rules.h>
#include <linux/if_ether.h>
#include <linux/netlink.h>
#include <linux/rtnetlink.h>
#include <linux/veth.h>
#include <net/if.h>
#include <poll.h>
#include <string.h>
#include <sys/socket.h>
#include <sys/types.h>
#include <unistd.h>
#include <algorithm>
#include <cerrno>
#include <cstdint>
#include <functional>
#include <iostream>
#include <iterator>
#include <map>
#include <sstream>
#include <string>
#include <tuple>
#include <vector>
#include "gmock/gmock.h"
#include "gtest/gtest.h"
#include "absl/strings/str_format.h"
#include "test/syscalls/linux/socket_netlink_route_util.h"
#include "test/syscalls/linux/socket_netlink_util.h"
#include "test/util/cleanup.h"
#include "test/util/file_descriptor.h"
#include "test/util/fs_util.h"
#include "test/util/linux_capability_util.h"
#include "test/util/logging.h"
#include "test/util/multiprocess_util.h"
#include "test/util/posix_error.h"
#include "test/util/save_util.h"
#include "test/util/socket_util.h"
#include "test/util/test_util.h"
// Tests for NETLINK_ROUTE sockets.
namespace gvisor {
namespace testing {
namespace {
constexpr uint32_t kSeq = 12345;
using ::testing::_;
using ::testing::AnyOf;
using ::testing::Eq;
// Parameters for SockOptTest. They are:
// 0: Socket option to query.
// 1: A predicate to run on the returned sockopt value. Should return true if
// the value is considered ok.
// 2: A description of what the sockopt value is expected to be. Should complete
// the sentence "<value> was unexpected, expected <description>"
using SockOptTest = ::testing::TestWithParam<
std::tuple<int, std::function<bool(int)>, std::string>>;
TEST_P(SockOptTest, GetSockOpt) {
int sockopt = std::get<0>(GetParam());
auto verifier = std::get<1>(GetParam());
std::string verifier_description = std::get<2>(GetParam());
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE));
int res;
socklen_t len = sizeof(res);
EXPECT_THAT(getsockopt(fd.get(), SOL_SOCKET, sockopt, &res, &len),
SyscallSucceeds());
EXPECT_EQ(len, sizeof(res));
EXPECT_TRUE(verifier(res)) << absl::StrFormat(
"getsockopt(%d, SOL_SOCKET, %d, &res, &len) => res=%d was unexpected, "
"expected %s",
fd.get(), sockopt, res, verifier_description);
}
std::function<bool(int)> IsPositive() {
return [](int val) { return val > 0; };
}
std::function<bool(int)> IsEqual(int target) {
return [target](int val) { return val == target; };
}
INSTANTIATE_TEST_SUITE_P(
NetlinkRouteTest, SockOptTest,
::testing::Values(
std::make_tuple(SO_SNDBUF, IsPositive(), "positive send buffer size"),
std::make_tuple(SO_RCVBUF, IsPositive(),
"positive receive buffer size"),
std::make_tuple(SO_TYPE, IsEqual(SOCK_RAW),
absl::StrFormat("SOCK_RAW (%d)", SOCK_RAW)),
std::make_tuple(SO_DOMAIN, IsEqual(AF_NETLINK),
absl::StrFormat("AF_NETLINK (%d)", AF_NETLINK)),
std::make_tuple(SO_PROTOCOL, IsEqual(NETLINK_ROUTE),
absl::StrFormat("NETLINK_ROUTE (%d)", NETLINK_ROUTE)),
std::make_tuple(SO_PASSCRED, IsEqual(0), "0")));
PosixErrorOr<FileDescriptor> netlinkRouteSocketInLinkGroup() {
ASSIGN_OR_RETURN_ERRNO(FileDescriptor fd, NetlinkBoundSocket(NETLINK_ROUTE));
const unsigned int group = RTNLGRP_LINK;
RETURN_ERROR_IF_SYSCALL_FAIL(setsockopt(
fd.get(), SOL_NETLINK, NETLINK_ADD_MEMBERSHIP, &group, sizeof(group)));
return fd;
}
// NETLINK_LIST_MEMBERSHIPS returns the output length when optval is not
// specified.
TEST(NetlinkRouteTest, ListMembershipsSizeProbe) {
SKIP_IF(IsRunningWithHostinet());
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(netlinkRouteSocketInLinkGroup());
socklen_t len = 0;
EXPECT_THAT(getsockopt(fd.get(), SOL_NETLINK, NETLINK_LIST_MEMBERSHIPS,
nullptr, &len),
SyscallSucceeds());
EXPECT_GT(len, 0);
}
// A buffer larger than the value succeeds; optlen is updated to the actual
// size, which fits within the provided buffer.
TEST(NetlinkRouteTest, ListMembershipsLargeBuffer) {
SKIP_IF(IsRunningWithHostinet());
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(netlinkRouteSocketInLinkGroup());
// Fetch the real length
socklen_t len = 0;
EXPECT_THAT(getsockopt(fd.get(), SOL_NETLINK, NETLINK_LIST_MEMBERSHIPS,
nullptr, &len),
SyscallSucceeds());
EXPECT_GT(len, 0);
// And make a buffer larger than that
socklen_t new_len = len + 10;
std::vector<char> buf(new_len, 0);
EXPECT_THAT(getsockopt(fd.get(), SOL_NETLINK, NETLINK_LIST_MEMBERSHIPS,
buf.data(), &new_len),
SyscallSucceeds());
EXPECT_EQ(len, new_len);
// First len elements were set (some may be zero, so all we can do is test
// that they're not *all* 0).
EXPECT_FALSE(
std::all_of(buf.begin(), buf.end() - 10, [](char c) { return c == 0; }));
// Last 10 elements were not touched
EXPECT_TRUE(
std::all_of(buf.begin() + len, buf.end(), [](char c) { return c == 0; }));
}
// A buffer smaller than the value is truncated to optlen bytes, but optlen is
// still updated to the full size.
TEST(NetlinkRouteTest, ListMembershipsTruncated) {
SKIP_IF(IsRunningWithHostinet());
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(netlinkRouteSocketInLinkGroup());
// Fetch the real length
socklen_t len = 0;
EXPECT_THAT(getsockopt(fd.get(), SOL_NETLINK, NETLINK_LIST_MEMBERSHIPS,
nullptr, &len),
SyscallSucceeds());
// Make a smaller buffer
ASSERT_GE(len, 2);
socklen_t new_len = len / 2;
std::vector<char> buf(new_len, 0);
int ret = getsockopt(fd.get(), SOL_NETLINK, NETLINK_LIST_MEMBERSHIPS,
buf.data(), &new_len);
const bool was_efault = Value(ret, SyscallFailsWithErrno(EFAULT));
// Strangely, this EFAULTs on native, but only on our testing infrastructure.
// Seems like something is intercepting the getsockopt() call. Skip it for
// now.
SKIP_IF(was_efault && !IsRunningOnGvisor());
EXPECT_THAT(ret, SyscallSucceeds());
EXPECT_EQ(new_len, len);
// First len / 2 elements were set (some may be zero, so all we can do is test
// that they're not *all* 0).
EXPECT_FALSE(!buf.empty() && std::all_of(buf.begin(), buf.end(),
[](char c) { return c == 0; }));
}
// Validates the responses to RTM_GETLINK + NLM_F_DUMP.
void CheckGetLinkResponse(const struct nlmsghdr* hdr, int seq, int port) {
EXPECT_THAT(hdr->nlmsg_type, AnyOf(Eq(RTM_NEWLINK), Eq(NLMSG_DONE)));
EXPECT_TRUE((hdr->nlmsg_flags & NLM_F_MULTI) == NLM_F_MULTI)
<< std::hex << hdr->nlmsg_flags;
EXPECT_EQ(hdr->nlmsg_seq, seq);
EXPECT_EQ(hdr->nlmsg_pid, port);
if (hdr->nlmsg_type != RTM_NEWLINK) {
return;
}
// RTM_NEWLINK contains at least the header and ifinfomsg.
EXPECT_GE(hdr->nlmsg_len, NLMSG_SPACE(sizeof(struct ifinfomsg)));
// TODO(mpratt): Check ifinfomsg contents and following attrs.
}
TEST(NetlinkRouteTest, GetLinkDump) {
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get()));
// Loopback is common among all tests, check that it's found.
bool loopbackFound = false;
ASSERT_NO_ERRNO(DumpLinks(fd, kSeq, [&](const struct nlmsghdr* hdr) {
CheckGetLinkResponse(hdr, kSeq, port);
if (hdr->nlmsg_type != RTM_NEWLINK) {
return;
}
ASSERT_GE(hdr->nlmsg_len, NLMSG_SPACE(sizeof(struct ifinfomsg)));
const struct ifinfomsg* msg =
reinterpret_cast<const struct ifinfomsg*>(NLMSG_DATA(hdr));
std::cout << "Found interface idx=" << msg->ifi_index
<< ", type=" << std::hex << msg->ifi_type << std::endl;
if (msg->ifi_type == ARPHRD_LOOPBACK) {
loopbackFound = true;
EXPECT_NE(msg->ifi_flags & IFF_LOOPBACK, 0);
}
}));
EXPECT_TRUE(loopbackFound);
}
// CheckLinkMsg checks a netlink message against an expected link.
void CheckLinkMsg(const struct nlmsghdr* hdr, const Link& link) {
ASSERT_THAT(hdr->nlmsg_type, Eq(RTM_NEWLINK));
ASSERT_GE(hdr->nlmsg_len, NLMSG_SPACE(sizeof(struct ifinfomsg)));
const struct ifinfomsg* msg =
reinterpret_cast<const struct ifinfomsg*>(NLMSG_DATA(hdr));
EXPECT_EQ(msg->ifi_index, link.index);
const struct rtattr* rta_name = FindRtAttr(hdr, msg, IFLA_IFNAME);
EXPECT_NE(nullptr, rta_name) << "IFLA_IFNAME not found in message.";
if (rta_name != nullptr) {
std::string name(reinterpret_cast<const char*>(RTA_DATA(rta_name)));
EXPECT_EQ(name, link.name);
}
const struct rtattr* rta_mtu = FindRtAttr(hdr, msg, IFLA_MTU);
EXPECT_NE(nullptr, rta_mtu) << "IFLA_MTU not found in message.";
if (rta_mtu != nullptr) {
const auto mtu = *(uint32_t*)(RTA_DATA(rta_mtu));
EXPECT_EQ(mtu, link.mtu);
}
const struct rtattr* rta_address = FindRtAttr(hdr, msg, IFLA_ADDRESS);
EXPECT_NE(nullptr, rta_address) << "IFLA_ADDRESS not found in message.";
if (rta_address != nullptr) {
std::string address(reinterpret_cast<const char*>(RTA_DATA(rta_address)));
EXPECT_EQ(address, link.address);
}
EXPECT_EQ(ASSERT_NO_ERRNO_AND_VALUE(LinkKind(hdr, msg)), link.kind);
}
TEST(NetlinkRouteTest, GetLinkByIndex) {
Link loopback_link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETLINK;
req.hdr.nlmsg_flags = NLM_F_REQUEST;
req.hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
req.ifm.ifi_index = loopback_link.index;
bool found = false;
ASSERT_NO_ERRNO(NetlinkRequestResponse(
fd, &req, sizeof(req),
[&](const struct nlmsghdr* hdr) {
CheckLinkMsg(hdr, loopback_link);
found = true;
},
false));
EXPECT_TRUE(found) << "Netlink response does not contain any links.";
}
// NetlinkRouteTest with a single parameter that must be RTM_NEWLINK or
// RTM_SETLINK.
using NetlinkSetLinkTest = ::testing::TestWithParam<int>;
INSTANTIATE_TEST_SUITE_P(_, NetlinkSetLinkTest,
::testing::Values(RTM_NEWLINK, RTM_SETLINK));
TEST_P(NetlinkSetLinkTest, ChangeLinkName) {
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
SKIP_IF(IsRunningWithHostinet());
// Hosts that run with old kernel allow renaming only when
// the interface is down. The restriction has been removed.
SKIP_IF(!IsRunningOnGvisor());
Link loopback_link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
struct rtattr rtattr;
char ifname[IFNAMSIZ];
char pad[NLMSG_ALIGNTO + RTA_ALIGNTO];
};
const std::string new_linkname = "notloopback";
// Change the link name.
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = GetParam();
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
req.hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
req.ifm.ifi_index = loopback_link.index;
req.rtattr.rta_type = IFLA_IFNAME;
req.rtattr.rta_len = RTA_LENGTH(new_linkname.size() + 1);
strncpy(req.ifname, new_linkname.c_str(), sizeof(req.ifname));
req.hdr.nlmsg_len =
NLMSG_LENGTH(sizeof(req.ifm)) + NLMSG_ALIGN(req.rtattr.rta_len);
EXPECT_NO_ERRNO(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)));
// Search the link by the new name.
loopback_link.name = new_linkname;
req.hdr.nlmsg_type = RTM_GETLINK;
req.ifm.ifi_index = 0;
req.hdr.nlmsg_flags = NLM_F_REQUEST;
req.rtattr.rta_type = IFLA_IFNAME;
req.rtattr.rta_len = RTA_LENGTH(new_linkname.size() + 1);
strncpy(req.ifname, new_linkname.c_str(), sizeof(req.ifname));
req.hdr.nlmsg_len =
NLMSG_LENGTH(sizeof(req.ifm)) + NLMSG_ALIGN(req.rtattr.rta_len);
bool found = false;
ASSERT_NO_ERRNO(NetlinkRequestResponse(
fd, &req, sizeof(req),
[&](const struct nlmsghdr* hdr) {
CheckLinkMsg(hdr, loopback_link);
found = true;
},
false));
EXPECT_TRUE(found) << "Netlink response does not contain any links.";
}
struct MtuRequest {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
struct rtattr rtattr;
uint32_t mtu;
};
MtuRequest GetMtuRequest(const Link& link, uint16_t nlmsg_type, uint32_t mtu) {
MtuRequest req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = nlmsg_type;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
req.hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
req.ifm.ifi_index = link.index;
req.rtattr.rta_type = IFLA_MTU;
req.rtattr.rta_len = RTA_LENGTH(sizeof(uint32_t));
req.mtu = mtu;
return req;
}
TEST_P(NetlinkSetLinkTest, ChangeMTU) {
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
SKIP_IF(IsRunningWithHostinet());
SKIP_IF(!IsRunningOnGvisor());
Link loopback_link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
// Change the MTU.
uint16_t nlmsg_type = GetParam();
MtuRequest req =
GetMtuRequest(loopback_link, nlmsg_type, loopback_link.mtu + 10);
EXPECT_NO_ERRNO(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)));
// Update the local loopback_link's MTU to the requested value.
loopback_link.mtu = req.mtu - ETH_HLEN;
// Verify the new MTU.
struct searchrequest {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
} search = {};
search.hdr.nlmsg_len = sizeof(search);
search.hdr.nlmsg_type = RTM_GETLINK;
search.hdr.nlmsg_flags = NLM_F_REQUEST;
search.hdr.nlmsg_seq = kSeq;
search.ifm.ifi_family = AF_UNSPEC;
search.ifm.ifi_index = loopback_link.index;
bool found = false;
ASSERT_NO_ERRNO(NetlinkRequestResponse(
fd, &search, sizeof(search),
[&](const struct nlmsghdr* hdr) {
CheckLinkMsg(hdr, loopback_link);
found = true;
},
false));
EXPECT_TRUE(found) << "Netlink response does not contain any links.";
}
TEST_P(NetlinkSetLinkTest, ChangeMACAddress) {
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
SKIP_IF(!IsRunningOnGvisor());
SKIP_IF(IsRunningWithHostinet());
Link loopback_link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
struct rtattr rtattr;
char address[1024];
};
const int address_size = 6;
const char address[address_size + 1] = {static_cast<char>(0xa1),
static_cast<char>(0xa2),
static_cast<char>(0xa3),
static_cast<char>(0xa4),
static_cast<char>(0xa5),
static_cast<char>(0xa6),
'\0'};
// Change the link MAC address.
struct request req = {};
req.hdr.nlmsg_type = GetParam();
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
req.hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
req.ifm.ifi_index = loopback_link.index;
req.rtattr.rta_type = IFLA_ADDRESS;
req.rtattr.rta_len = RTA_LENGTH(address_size);
strncpy(req.address, address, sizeof(req.address));
req.hdr.nlmsg_len =
NLMSG_LENGTH(sizeof(req.ifm)) + NLMSG_ALIGN(req.rtattr.rta_len);
EXPECT_NO_ERRNO(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)));
loopback_link.address = std::string(address);
// Search the link by its index.
struct searchrequest {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
} search = {};
search.hdr.nlmsg_len = sizeof(search);
search.hdr.nlmsg_type = RTM_GETLINK;
search.hdr.nlmsg_flags = NLM_F_REQUEST;
search.hdr.nlmsg_seq = kSeq;
search.ifm.ifi_family = AF_UNSPEC;
search.ifm.ifi_index = loopback_link.index;
bool found = false;
ASSERT_NO_ERRNO(NetlinkRequestResponse(
fd, &search, sizeof(search),
[&](const struct nlmsghdr* hdr) {
CheckLinkMsg(hdr, loopback_link);
found = true;
},
false));
EXPECT_TRUE(found) << "Netlink response does not contain any links.";
}
TEST(NetlinkRouteTest, LinkUp) {
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
SKIP_IF(IsRunningWithHostinet());
Link loopback_link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_NEWLINK;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
req.hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
req.ifm.ifi_index = loopback_link.index;
req.ifm.ifi_change = IFF_UP;
req.ifm.ifi_flags = IFF_UP;
EXPECT_NO_ERRNO(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)));
}
TEST(NetlinkRouteTest, GetLinkByName) {
Link loopback_link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
struct rtattr rtattr;
char ifname[IFNAMSIZ];
char pad[NLMSG_ALIGNTO + RTA_ALIGNTO];
};
struct request req = {};
req.hdr.nlmsg_type = RTM_GETLINK;
req.hdr.nlmsg_flags = NLM_F_REQUEST;
req.hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
req.rtattr.rta_type = IFLA_IFNAME;
req.rtattr.rta_len = RTA_LENGTH(loopback_link.name.size() + 1);
strncpy(req.ifname, loopback_link.name.c_str(), sizeof(req.ifname));
req.hdr.nlmsg_len =
NLMSG_LENGTH(sizeof(req.ifm)) + NLMSG_ALIGN(req.rtattr.rta_len);
bool found = false;
ASSERT_NO_ERRNO(NetlinkRequestResponse(
fd, &req, sizeof(req),
[&](const struct nlmsghdr* hdr) {
CheckLinkMsg(hdr, loopback_link);
found = true;
},
false));
EXPECT_TRUE(found) << "Netlink response does not contain any links.";
}
TEST(NetlinkRouteTest, GetLinkByIndexNotFound) {
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETLINK;
req.hdr.nlmsg_flags = NLM_F_REQUEST;
req.hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
req.ifm.ifi_index = 1234590;
EXPECT_THAT(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)),
PosixErrorIs(ENODEV, _));
}
TEST(NetlinkRouteTest, GetLinkByNameNotFound) {
const std::string name = "nodevice?!";
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
struct rtattr rtattr;
char ifname[IFNAMSIZ];
char pad[NLMSG_ALIGNTO + RTA_ALIGNTO];
};
struct request req = {};
req.hdr.nlmsg_type = RTM_GETLINK;
req.hdr.nlmsg_flags = NLM_F_REQUEST;
req.hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
req.rtattr.rta_type = IFLA_IFNAME;
req.rtattr.rta_len = RTA_LENGTH(name.size() + 1);
strncpy(req.ifname, name.c_str(), sizeof(req.ifname));
req.hdr.nlmsg_len =
NLMSG_LENGTH(sizeof(req.ifm)) + NLMSG_ALIGN(req.rtattr.rta_len);
EXPECT_THAT(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)),
PosixErrorIs(ENODEV, _));
}
TEST(NetlinkRouteTest, RemoveLoopbackByName) {
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
Link loopback_link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
struct rtattr rtattr;
char ifname[IFNAMSIZ];
char pad[NLMSG_ALIGNTO + RTA_ALIGNTO];
};
struct request req = {};
req.hdr.nlmsg_type = RTM_DELLINK;
req.hdr.nlmsg_flags = NLM_F_REQUEST;
req.hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
req.rtattr.rta_type = IFLA_IFNAME;
req.rtattr.rta_len = RTA_LENGTH(loopback_link.name.size() + 1);
strncpy(req.ifname, loopback_link.name.c_str(), sizeof(req.ifname));
req.hdr.nlmsg_len =
NLMSG_LENGTH(sizeof(req.ifm)) + NLMSG_ALIGN(req.rtattr.rta_len);
EXPECT_THAT(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)),
PosixErrorIs(ENOTSUP, _));
}
TEST(NetlinkRouteTest, RemoveLoopbackByIndex) {
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
Link loopback_link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_DELLINK;
req.hdr.nlmsg_flags = NLM_F_REQUEST;
req.hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
req.ifm.ifi_index = loopback_link.index;
EXPECT_THAT(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)),
PosixErrorIs(ENOTSUP, _));
}
TEST(NetlinkRouteTest, RemoveLinkByIndexNotFound) {
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETLINK;
req.hdr.nlmsg_flags = NLM_F_REQUEST;
req.hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
req.ifm.ifi_index = 1234590;
EXPECT_THAT(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)),
PosixErrorIs(ENODEV, _));
}
TEST(NetlinkRouteTest, RemoveLinkByNameNotFound) {
const std::string name = "nodevice?!";
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
struct rtattr rtattr;
char ifname[IFNAMSIZ];
char pad[NLMSG_ALIGNTO + RTA_ALIGNTO];
};
struct request req = {};
req.hdr.nlmsg_type = RTM_DELLINK;
req.hdr.nlmsg_flags = NLM_F_REQUEST;
req.hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
req.rtattr.rta_type = IFLA_IFNAME;
req.rtattr.rta_len = RTA_LENGTH(name.size() + 1);
strncpy(req.ifname, name.c_str(), sizeof(req.ifname));
req.hdr.nlmsg_len =
NLMSG_LENGTH(sizeof(req.ifm)) + NLMSG_ALIGN(req.rtattr.rta_len);
EXPECT_THAT(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)),
PosixErrorIs(ENODEV, _));
}
TEST(NetlinkRouteTest, MsgHdrMsgUnsuppType) {
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
// If type & 0x3 is equal to 0x2, this means a get request
// which doesn't require CAP_SYS_ADMIN.
req.hdr.nlmsg_type = ((__RTM_MAX + 1024) & (~0x3)) | 0x2;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
req.hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
EXPECT_THAT(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)),
PosixErrorIs(EOPNOTSUPP, _));
}
TEST(NetlinkRouteTest, MsgHdrMsgTrunc) {
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETLINK;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
req.hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
struct iovec iov = {};
iov.iov_base = &req;
iov.iov_len = sizeof(req);
struct msghdr msg = {};
msg.msg_iov = &iov;
msg.msg_iovlen = 1;
// No destination required; it defaults to pid 0, the kernel.
ASSERT_THAT(RetryEINTR(sendmsg)(fd.get(), &msg, 0), SyscallSucceeds());
// Small enough to ensure that the response doesn't fit.
constexpr size_t kBufferSize = 10;
std::vector<char> buf(kBufferSize);
iov.iov_base = buf.data();
iov.iov_len = buf.size();
ASSERT_THAT(RetryEINTR(recvmsg)(fd.get(), &msg, 0),
SyscallSucceedsWithValue(kBufferSize));
EXPECT_EQ((msg.msg_flags & MSG_TRUNC), MSG_TRUNC);
}
TEST(NetlinkRouteTest, SpliceFromPipe) {
Link loopback_link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
int fds[2];
ASSERT_THAT(pipe(fds), SyscallSucceeds());
FileDescriptor rfd(fds[0]);
FileDescriptor wfd(fds[1]);
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETLINK;
req.hdr.nlmsg_flags = NLM_F_REQUEST;
req.hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
req.ifm.ifi_index = loopback_link.index;
ASSERT_THAT(write(wfd.get(), &req, sizeof(req)),
SyscallSucceedsWithValue(sizeof(req)));
EXPECT_THAT(splice(rfd.get(), nullptr, fd.get(), nullptr, sizeof(req) + 1, 0),
SyscallSucceedsWithValue(sizeof(req)));
close(wfd.release());
EXPECT_THAT(splice(rfd.get(), nullptr, fd.get(), nullptr, sizeof(req) + 1, 0),
SyscallSucceedsWithValue(0));
bool found = false;
ASSERT_NO_ERRNO(NetlinkResponse(
fd,
[&](const struct nlmsghdr* hdr) {
CheckLinkMsg(hdr, loopback_link);
found = true;
},
false));
EXPECT_TRUE(found) << "Netlink response does not contain any links.";
}
TEST(NetlinkRouteTest, MsgTruncMsgHdrMsgTrunc) {
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETLINK;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
req.hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
struct iovec iov = {};
iov.iov_base = &req;
iov.iov_len = sizeof(req);
struct msghdr msg = {};
msg.msg_iov = &iov;
msg.msg_iovlen = 1;
// No destination required; it defaults to pid 0, the kernel.
ASSERT_THAT(RetryEINTR(sendmsg)(fd.get(), &msg, 0), SyscallSucceeds());
// Small enough to ensure that the response doesn't fit.
constexpr size_t kBufferSize = 10;
std::vector<char> buf(kBufferSize);
iov.iov_base = buf.data();
iov.iov_len = buf.size();
int res = 0;
ASSERT_THAT(res = RetryEINTR(recvmsg)(fd.get(), &msg, MSG_TRUNC),
SyscallSucceeds());
EXPECT_GT(res, kBufferSize);
EXPECT_EQ((msg.msg_flags & MSG_TRUNC), MSG_TRUNC);
}
TEST(NetlinkRouteTest, ControlMessageIgnored) {
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get()));
struct request {
struct nlmsghdr control_hdr;
struct nlmsghdr message_hdr;
struct ifinfomsg ifm;
};
struct request req = {};
// This control message is ignored. We still receive a response for the
// following RTM_GETLINK.
req.control_hdr.nlmsg_len = sizeof(req.control_hdr);
req.control_hdr.nlmsg_type = NLMSG_DONE;
req.control_hdr.nlmsg_seq = kSeq;
req.message_hdr.nlmsg_len = sizeof(req.message_hdr) + sizeof(req.ifm);
req.message_hdr.nlmsg_type = RTM_GETLINK;
req.message_hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
req.message_hdr.nlmsg_seq = kSeq;
req.ifm.ifi_family = AF_UNSPEC;
ASSERT_NO_ERRNO(NetlinkRequestResponse(
fd, &req, sizeof(req),
[&](const struct nlmsghdr* hdr) {
CheckGetLinkResponse(hdr, kSeq, port);
},
false));
}
TEST(NetlinkRouteTest, GetAddrDump) {
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get()));
struct request {
struct nlmsghdr hdr;
struct rtgenmsg rgm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETADDR;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
req.hdr.nlmsg_seq = kSeq;
req.rgm.rtgen_family = AF_UNSPEC;
ASSERT_NO_ERRNO(NetlinkRequestResponse(
fd, &req, sizeof(req),
[&](const struct nlmsghdr* hdr) {
EXPECT_THAT(hdr->nlmsg_type, AnyOf(Eq(RTM_NEWADDR), Eq(NLMSG_DONE)));
EXPECT_TRUE((hdr->nlmsg_flags & NLM_F_MULTI) == NLM_F_MULTI)
<< std::hex << hdr->nlmsg_flags;
EXPECT_EQ(hdr->nlmsg_seq, kSeq);
EXPECT_EQ(hdr->nlmsg_pid, port);
if (hdr->nlmsg_type != RTM_NEWADDR) {
return;
}
// RTM_NEWADDR contains at least the header and ifaddrmsg.
EXPECT_GE(hdr->nlmsg_len, sizeof(*hdr) + sizeof(struct ifaddrmsg));
// TODO(mpratt): Check ifaddrmsg contents and following attrs.
},
false));
}
TEST(NetlinkRouteTest, GetAddrRoot) {
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get()));
struct request {
struct nlmsghdr hdr;
struct rtgenmsg rgm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETADDR;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_ROOT;
req.hdr.nlmsg_seq = kSeq;
req.rgm.rtgen_family = AF_UNSPEC;
bool loopbackFound = false;
ASSERT_NO_ERRNO(NetlinkRequestResponse(
fd, &req, sizeof(req),
[&](const struct nlmsghdr* hdr) {
EXPECT_THAT(hdr->nlmsg_type, AnyOf(Eq(RTM_NEWADDR), Eq(NLMSG_DONE)));
EXPECT_TRUE((hdr->nlmsg_flags & NLM_F_MULTI) == NLM_F_MULTI)
<< std::hex << hdr->nlmsg_flags;
EXPECT_EQ(hdr->nlmsg_seq, kSeq);
EXPECT_EQ(hdr->nlmsg_pid, port);
if (hdr->nlmsg_type != RTM_NEWADDR) {
return;
}
EXPECT_GE(hdr->nlmsg_len, sizeof(*hdr) + sizeof(struct ifaddrmsg));
const struct ifaddrmsg* msg =
reinterpret_cast<const struct ifaddrmsg*>(NLMSG_DATA(hdr));
if (msg->ifa_family == AF_INET && msg->ifa_scope == RT_SCOPE_HOST) {
int len = IFA_PAYLOAD(hdr);
for (struct rtattr* attr = IFA_RTA(msg); RTA_OK(attr, len);
attr = RTA_NEXT(attr, len)) {
if (attr->rta_type == IFA_LABEL) {
std::string label(reinterpret_cast<const char*>(RTA_DATA(attr)));
if (label == "lo") {
loopbackFound = true;
}
}
}
}
},
false));
EXPECT_TRUE(loopbackFound);
}
TEST(NetlinkRouteTest, LookupAll) {
struct ifaddrs* if_addr_list = nullptr;
auto cleanup = Cleanup([&if_addr_list]() { freeifaddrs(if_addr_list); });
// Not a syscall but we can use the syscall matcher as glibc sets errno.
ASSERT_THAT(getifaddrs(&if_addr_list), SyscallSucceeds());
int count = 0;
for (struct ifaddrs* i = if_addr_list; i; i = i->ifa_next) {
if (!i->ifa_addr || (i->ifa_addr->sa_family != AF_INET &&
i->ifa_addr->sa_family != AF_INET6)) {
continue;
}
count++;
}
ASSERT_GT(count, 0);
}
TEST(NetlinkRouteTest, AddAndRemoveAddr) {
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
// Don't do cooperative save/restore because netstack state is not restored.
// TODO(gvisor.dev/issue/4595): enable cooperative save tests.
const DisableSave ds;
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
Link loopback_link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
struct in_addr addr;
ASSERT_EQ(inet_pton(AF_INET, "10.0.0.1", &addr), 1);
// Create should succeed, as no such address in kernel.
ASSERT_NO_ERRNO(LinkAddLocalAddr(fd, loopback_link.index, AF_INET,
/*prefixlen=*/24, &addr, sizeof(addr)));
Cleanup defer_addr_removal = Cleanup([&] {
// First delete should succeed, as address exists.
EXPECT_NO_ERRNO(LinkDelLocalAddr(fd, loopback_link.index, AF_INET,
/*prefixlen=*/24, &addr, sizeof(addr)));
// Second delete should fail, as address no longer exists.
EXPECT_THAT(LinkDelLocalAddr(fd, loopback_link.index, AF_INET,
/*prefixlen=*/24, &addr, sizeof(addr)),
PosixErrorIs(EADDRNOTAVAIL, _));
});
// Replace an existing address should succeed.
ASSERT_NO_ERRNO(LinkReplaceLocalAddr(fd, loopback_link.index, AF_INET,
/*prefixlen=*/24, &addr, sizeof(addr)));
// Create exclusive should fail, as we created the address above.
EXPECT_THAT(LinkAddExclusiveLocalAddr(fd, loopback_link.index, AF_INET,
/*prefixlen=*/24, &addr, sizeof(addr)),
PosixErrorIs(EEXIST, _));
}
TEST(NetlinkRouteTest, LinkedToNetns) {
SKIP_IF(IsRunningWithHostinet());
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
// Don't do cooperative save/restore because netstack state is not restored.
// TODO(gvisor.dev/issue/4595): enable cooperative save tests.
const DisableSave ds;
FileDescriptor root_netns_nlsk =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
const FileDescriptor nsfd =
ASSERT_NO_ERRNO_AND_VALUE(Open("/proc/thread-self/ns/net", O_RDONLY));
Cleanup defer_netns = Cleanup([&] {
ASSERT_THAT(setns(nsfd.get(), CLONE_NEWNET), SyscallSucceedsWithValue(0));
});
ASSERT_THAT(unshare(CLONE_NEWNET), SyscallSucceedsWithValue(0));
FileDescriptor nlsk =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
Link loopback_link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
struct in_addr addr;
ASSERT_EQ(inet_pton(AF_INET, "10.0.0.1", &addr), 1);
// Create should succeed, as no such address in kernel.
ASSERT_NO_ERRNO(LinkAddLocalAddr(nlsk, loopback_link.index, AF_INET,
/*prefixlen=*/24, &addr, sizeof(addr)));
// No such address in the root network namespace.
EXPECT_THAT(LinkDelLocalAddr(root_netns_nlsk, loopback_link.index, AF_INET,
/*prefixlen=*/24, &addr, sizeof(addr)),
PosixErrorIs(EADDRNOTAVAIL, _));
// The address exists in the current namespace.
EXPECT_NO_ERRNO(LinkDelLocalAddr(nlsk, loopback_link.index, AF_INET,
/*prefixlen=*/24, &addr, sizeof(addr)));
}
// GetRouteDump tests a RTM_GETROUTE + NLM_F_DUMP request.
TEST(NetlinkRouteTest, GetRouteDump) {
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get()));
struct request {
struct nlmsghdr hdr;
struct rtmsg rtm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETROUTE;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
req.hdr.nlmsg_seq = kSeq;
req.rtm.rtm_family = AF_UNSPEC;
bool routeFound = false;
bool dstFound = true;
bool defaultRouteDstFound = false;
ASSERT_NO_ERRNO(NetlinkRequestResponse(
fd, &req, sizeof(req),
[&](const struct nlmsghdr* hdr) {
// Validate the response to RTM_GETROUTE + NLM_F_DUMP.
EXPECT_THAT(hdr->nlmsg_type, AnyOf(Eq(RTM_NEWROUTE), Eq(NLMSG_DONE)));
EXPECT_TRUE((hdr->nlmsg_flags & NLM_F_MULTI) == NLM_F_MULTI)
<< std::hex << hdr->nlmsg_flags;
EXPECT_EQ(hdr->nlmsg_seq, kSeq);
EXPECT_EQ(hdr->nlmsg_pid, port);
// The test should not proceed if it's not a RTM_NEWROUTE message.
if (hdr->nlmsg_type != RTM_NEWROUTE) {
return;
}
// RTM_NEWROUTE contains at least the header and rtmsg.
ASSERT_GE(hdr->nlmsg_len, NLMSG_SPACE(sizeof(struct rtmsg)));
const struct rtmsg* msg =
reinterpret_cast<const struct rtmsg*>(NLMSG_DATA(hdr));
// NOTE: rtmsg fields are char fields.
std::cout << "Found route table=" << static_cast<int>(msg->rtm_table)
<< ", protocol=" << static_cast<int>(msg->rtm_protocol)
<< ", scope=" << static_cast<int>(msg->rtm_scope)
<< ", type=" << static_cast<int>(msg->rtm_type)
<< ", prefixlen=" << static_cast<int>(msg->rtm_dst_len);
int len = RTM_PAYLOAD(hdr);
bool rtDstFound = false;
for (struct rtattr* attr = RTM_RTA(msg); RTA_OK(attr, len);
attr = RTA_NEXT(attr, len)) {
if (attr->rta_type == RTA_DST) {
char address[INET_ADDRSTRLEN] = {};
inet_ntop(AF_INET, RTA_DATA(attr), address, sizeof(address));
std::cout << ", dst=" << address;
rtDstFound = true;
}
}
std::cout << std::endl;
// If the test is running in a new network namespace, it will have only
// the local route table.
if (msg->rtm_table == RT_TABLE_MAIN ||
(!IsRunningOnGvisor() && msg->rtm_table == RT_TABLE_LOCAL)) {
routeFound = true;
if (msg->rtm_dst_len) {
dstFound = rtDstFound && dstFound;
} else {
defaultRouteDstFound = rtDstFound || defaultRouteDstFound;
}
}
},
false));
// At least one route found in main route table.
EXPECT_TRUE(routeFound);
// Found RTA_DST for each route in main table.
EXPECT_TRUE(dstFound);
// RTA_DST should not be present for default routes.
EXPECT_FALSE(defaultRouteDstFound);
}
// GetRouteRequest tests a RTM_GETROUTE request with RTM_F_LOOKUP_TABLE flag.
TEST(NetlinkRouteTest, GetRouteRequest) {
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get()));
struct request {
struct nlmsghdr hdr;
struct rtmsg rtm;
struct nlattr nla;
struct in_addr sin_addr;
};
constexpr uint32_t kSeq = 12345;
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETROUTE;
req.hdr.nlmsg_flags = NLM_F_REQUEST;
req.hdr.nlmsg_seq = kSeq;
req.rtm.rtm_family = AF_INET;
req.rtm.rtm_dst_len = 32;
req.rtm.rtm_src_len = 0;
req.rtm.rtm_tos = 0;
req.rtm.rtm_table = RT_TABLE_UNSPEC;
req.rtm.rtm_protocol = RTPROT_UNSPEC;
req.rtm.rtm_scope = RT_SCOPE_UNIVERSE;
req.rtm.rtm_type = RTN_UNSPEC;
req.rtm.rtm_flags = RTM_F_LOOKUP_TABLE;
req.nla.nla_len = 8;
req.nla.nla_type = RTA_DST;
inet_aton("127.0.0.2", &req.sin_addr);
bool rtDstFound = false;
ASSERT_NO_ERRNO(NetlinkRequestResponseSingle(
fd, &req, sizeof(req), [&](const struct nlmsghdr* hdr) {
// Validate the response to RTM_GETROUTE request with RTM_F_LOOKUP_TABLE
// flag.
EXPECT_THAT(hdr->nlmsg_type, RTM_NEWROUTE);
EXPECT_TRUE(hdr->nlmsg_flags == 0) << std::hex << hdr->nlmsg_flags;
EXPECT_EQ(hdr->nlmsg_seq, kSeq);
EXPECT_EQ(hdr->nlmsg_pid, port);
// RTM_NEWROUTE contains at least the header and rtmsg.
ASSERT_GE(hdr->nlmsg_len, NLMSG_SPACE(sizeof(struct rtmsg)));
const struct rtmsg* msg =
reinterpret_cast<const struct rtmsg*>(NLMSG_DATA(hdr));
// NOTE: rtmsg fields are char fields.
std::cout << "Found route table=" << static_cast<int>(msg->rtm_table)
<< ", protocol=" << static_cast<int>(msg->rtm_protocol)
<< ", scope=" << static_cast<int>(msg->rtm_scope)
<< ", type=" << static_cast<int>(msg->rtm_type);
EXPECT_EQ(msg->rtm_family, AF_INET);
EXPECT_EQ(msg->rtm_dst_len, 32);
EXPECT_TRUE((msg->rtm_flags & RTM_F_CLONED) == RTM_F_CLONED)
<< std::hex << msg->rtm_flags;
int len = RTM_PAYLOAD(hdr);
std::cout << ", len=" << len;
for (struct rtattr* attr = RTM_RTA(msg); RTA_OK(attr, len);
attr = RTA_NEXT(attr, len)) {
if (attr->rta_type == RTA_DST) {
char address[INET_ADDRSTRLEN] = {};
inet_ntop(AF_INET, RTA_DATA(attr), address, sizeof(address));
std::cout << ", dst=" << address;
rtDstFound = true;
} else if (attr->rta_type == RTA_OIF) {
const char* oif = reinterpret_cast<const char*>(RTA_DATA(attr));
std::cout << ", oif=" << oif;
}
}
std::cout << std::endl;
}));
// Found RTA_DST for RTM_F_LOOKUP_TABLE.
EXPECT_TRUE(rtDstFound);
}
// GetRouteRoot tests a RTM_GETROUTE + NLM_F_ROOT request.
TEST(NetlinkRouteTest, GetRouteRoot) {
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get()));
struct request {
struct nlmsghdr hdr;
struct rtmsg rtm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETROUTE;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_ROOT;
req.hdr.nlmsg_seq = kSeq;
req.rtm.rtm_family = AF_UNSPEC;
bool routeFound = false;
ASSERT_NO_ERRNO(NetlinkRequestResponse(
fd, &req, sizeof(req),
[&](const struct nlmsghdr* hdr) {
EXPECT_THAT(hdr->nlmsg_type, AnyOf(Eq(RTM_NEWROUTE), Eq(NLMSG_DONE)));
EXPECT_TRUE((hdr->nlmsg_flags & NLM_F_MULTI) == NLM_F_MULTI)
<< std::hex << hdr->nlmsg_flags;
EXPECT_EQ(hdr->nlmsg_seq, kSeq);
EXPECT_EQ(hdr->nlmsg_pid, port);
if (hdr->nlmsg_type != RTM_NEWROUTE) {
return;
}
ASSERT_GE(hdr->nlmsg_len, NLMSG_SPACE(sizeof(struct rtmsg)));
routeFound = true;
},
false));
EXPECT_TRUE(routeFound);
}
// GetRoutePrefSrc tests that RTM_GETROUTE returns RTA_PREFSRC attribute.
TEST(NetlinkRouteTest, GetRoutePrefSrc) {
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get()));
struct request {
struct nlmsghdr hdr;
struct rtmsg rtm;
struct nlattr nla;
struct in_addr sin_addr;
};
constexpr uint32_t kSeq = 12345;
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETROUTE;
req.hdr.nlmsg_flags = NLM_F_REQUEST;
req.hdr.nlmsg_seq = kSeq;
req.rtm.rtm_family = AF_INET;
req.rtm.rtm_dst_len = 32;
req.rtm.rtm_src_len = 0;
req.rtm.rtm_tos = 0;
req.rtm.rtm_table = RT_TABLE_UNSPEC;
req.rtm.rtm_protocol = RTPROT_UNSPEC;
req.rtm.rtm_scope = RT_SCOPE_UNIVERSE;
req.rtm.rtm_type = RTN_UNSPEC;
req.nla.nla_len = 8;
req.nla.nla_type = RTA_DST;
inet_aton("127.0.0.1", &req.sin_addr);
bool prefSrcFound = false;
ASSERT_NO_ERRNO(NetlinkRequestResponseSingle(
fd, &req, sizeof(req), [&](const struct nlmsghdr* hdr) {
EXPECT_THAT(hdr->nlmsg_type, RTM_NEWROUTE);
EXPECT_EQ(hdr->nlmsg_seq, kSeq);
EXPECT_EQ(hdr->nlmsg_pid, port);
ASSERT_GE(hdr->nlmsg_len, NLMSG_SPACE(sizeof(struct rtmsg)));
const struct rtmsg* msg =
reinterpret_cast<const struct rtmsg*>(NLMSG_DATA(hdr));
EXPECT_EQ(msg->rtm_family, AF_INET);
int len = RTM_PAYLOAD(hdr);
for (struct rtattr* attr = RTM_RTA(msg); RTA_OK(attr, len);
attr = RTA_NEXT(attr, len)) {
if (attr->rta_type == RTA_PREFSRC) {
char address[INET_ADDRSTRLEN] = {};
inet_ntop(AF_INET, RTA_DATA(attr), address, sizeof(address));
EXPECT_STREQ(address, "127.0.0.1");
prefSrcFound = true;
}
}
}));
EXPECT_TRUE(prefSrcFound) << "RTA_PREFSRC not found in RTM_GETROUTE response";
}
// GetRouteUnreachable tests that RTM_GETROUTE for an unreachable destination
// returns NLMSG_ERROR with -ENETUNREACH.
TEST(NetlinkRouteTest, GetRouteUnreachable) {
SKIP_IF(IsRunningWithHostinet());
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
const DisableSave ds;
const FileDescriptor nsfd =
ASSERT_NO_ERRNO_AND_VALUE(Open("/proc/thread-self/ns/net", O_RDONLY));
Cleanup defer_netns = Cleanup([&] {
ASSERT_THAT(setns(nsfd.get(), CLONE_NEWNET), SyscallSucceedsWithValue(0));
});
ASSERT_THAT(unshare(CLONE_NEWNET), SyscallSucceedsWithValue(0));
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get()));
struct request {
struct nlmsghdr hdr;
struct rtmsg rtm;
struct nlattr nla;
struct in_addr sin_addr;
};
constexpr uint32_t kSeq = 12345;
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETROUTE;
req.hdr.nlmsg_flags = NLM_F_REQUEST;
req.hdr.nlmsg_seq = kSeq;
req.rtm.rtm_family = AF_INET;
req.rtm.rtm_dst_len = 32;
req.rtm.rtm_src_len = 0;
req.rtm.rtm_tos = 0;
req.rtm.rtm_table = RT_TABLE_UNSPEC;
req.rtm.rtm_protocol = RTPROT_UNSPEC;
req.rtm.rtm_scope = RT_SCOPE_UNIVERSE;
req.rtm.rtm_type = RTN_UNSPEC;
req.nla.nla_len = 8;
req.nla.nla_type = RTA_DST;
inet_aton("198.51.100.1", &req.sin_addr);
bool errorFound = false;
ASSERT_NO_ERRNO(NetlinkRequestResponseSingle(
fd, &req, sizeof(req), [&](const struct nlmsghdr* hdr) {
EXPECT_EQ(hdr->nlmsg_type, NLMSG_ERROR);
EXPECT_EQ(hdr->nlmsg_seq, kSeq);
EXPECT_EQ(hdr->nlmsg_pid, port);
ASSERT_GE(hdr->nlmsg_len, sizeof(*hdr) + sizeof(struct nlmsgerr));
const struct nlmsgerr* msg =
reinterpret_cast<const struct nlmsgerr*>(NLMSG_DATA(hdr));
EXPECT_EQ(msg->error, -ENETUNREACH);
errorFound = true;
}));
EXPECT_TRUE(errorFound);
}
// NetlinkRouteTest with a single parameter that must be AF_INET or AF_INET6.
using NetlinkRouteIpInvariantTest = ::testing::TestWithParam<int>;
INSTANTIATE_TEST_SUITE_P(NetlinkRouteIpv4AndIpv6Tests,
NetlinkRouteIpInvariantTest,
::testing::Values(AF_INET, AF_INET6));
TEST_P(NetlinkRouteIpInvariantTest, NewRoute) {
// CAP_NET_ADMIN is required to modify the routing table.
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
SKIP_IF(!IsRunningOnGvisor());
// Routes are not savable.
DisableSave ds;
const std::string dst_v4_address = "192.0.2.0";
const std::string dst_v6_address = "2001:db8::";
// Based on the test parameter, build an IPv4 or IPv6 destination subnet.
int family = GetParam();
void* dst = nullptr;
int dst_len;
int prefixlen;
struct in_addr dst_v4;
struct in6_addr dst_v6;
switch (family) {
case AF_INET:
ASSERT_EQ(inet_pton(family, dst_v4_address.c_str(), &dst_v4), 1);
prefixlen = 24;
dst = &dst_v4;
dst_len = sizeof(dst_v4);
break;
case AF_INET6:
ASSERT_EQ(inet_pton(family, dst_v6_address.c_str(), &dst_v6), 1);
prefixlen = 64;
dst = &dst_v6;
dst_len = sizeof(dst_v6);
break;
default:
FAIL() << "address family must be AF_INET or AF_INET6";
}
Link loopback_link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
ASSERT_NO_ERRNO(
AddUnicastRoute(loopback_link.index, family, prefixlen, dst, dst_len));
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct rtmsg rtm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETROUTE;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
req.hdr.nlmsg_seq = kSeq;
req.rtm.rtm_family = AF_UNSPEC;
bool routeDstFound = false;
ASSERT_NO_ERRNO(NetlinkRequestResponse(
fd, &req, sizeof(req),
[&](const struct nlmsghdr* hdr) {
// Validate the response to RTM_GETROUTE + NLM_F_DUMP.
EXPECT_THAT(hdr->nlmsg_type, AnyOf(Eq(RTM_NEWROUTE), Eq(NLMSG_DONE)));
// The test should not proceed if it's not a RTM_NEWROUTE message.
if (hdr->nlmsg_type != RTM_NEWROUTE) {
return;
}
const struct rtmsg* msg =
reinterpret_cast<const struct rtmsg*>(NLMSG_DATA(hdr));
int len = RTM_PAYLOAD(hdr);
for (struct rtattr* attr = RTM_RTA(msg); RTA_OK(attr, len);
attr = RTA_NEXT(attr, len)) {
if (attr->rta_type == RTA_DST) {
char v4_address[INET_ADDRSTRLEN] = {};
char v6_address[INET6_ADDRSTRLEN] = {};
switch (family) {
case AF_INET:
inet_ntop(AF_INET, RTA_DATA(attr), v4_address,
sizeof(v4_address));
if (strcmp(v4_address, dst_v4_address.c_str()) == 0) {
routeDstFound = true;
return;
}
break;
case AF_INET6:
inet_ntop(AF_INET6, RTA_DATA(attr), v6_address,
sizeof(v6_address));
if (strcmp(v6_address, dst_v6_address.c_str()) == 0) {
routeDstFound = true;
return;
}
break;
}
}
}
},
false));
EXPECT_TRUE(routeDstFound);
}
TEST_P(NetlinkRouteIpInvariantTest, DeleteRoute) {
// CAP_NET_ADMIN is required to modify the routing table.
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
SKIP_IF(!IsRunningOnGvisor());
// Routes are not savable.
DisableSave ds;
const std::string dst_v4_address = "192.0.3.0";
const std::string dst_v6_address = "2011:db8::";
// Based on the test parameter, build an IPv4 or IPv6 destination subnet.
int family = GetParam();
void* dst = nullptr;
int dst_len;
int prefixlen;
struct in_addr dst_v4;
struct in6_addr dst_v6;
switch (family) {
case AF_INET:
ASSERT_EQ(inet_pton(family, dst_v4_address.c_str(), &dst_v4), 1);
prefixlen = 24;
dst = &dst_v4;
dst_len = sizeof(dst_v4);
break;
case AF_INET6:
ASSERT_EQ(inet_pton(family, dst_v6_address.c_str(), &dst_v6), 1);
prefixlen = 64;
dst = &dst_v6;
dst_len = sizeof(dst_v6);
break;
default:
FAIL() << "address family must be AF_INET or AF_INET6";
}
Link loopback_link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
ASSERT_NO_ERRNO(
AddUnicastRoute(loopback_link.index, family, prefixlen, dst, dst_len));
ASSERT_NO_ERRNO(
DelUnicastRoute(loopback_link.index, family, prefixlen, dst, dst_len));
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct rtmsg rtm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETROUTE;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
req.hdr.nlmsg_seq = kSeq;
req.rtm.rtm_family = AF_UNSPEC;
bool routeDstFound = false;
ASSERT_NO_ERRNO(NetlinkRequestResponse(
fd, &req, sizeof(req),
[&](const struct nlmsghdr* hdr) {
// Validate the response to RTM_GETROUTE.
EXPECT_THAT(hdr->nlmsg_type, AnyOf(Eq(RTM_NEWROUTE), Eq(NLMSG_DONE)));
// The test should not proceed if it's not a RTM_NEWROUTE message.
if (hdr->nlmsg_type != RTM_NEWROUTE) {
return;
}
const struct rtmsg* msg =
reinterpret_cast<const struct rtmsg*>(NLMSG_DATA(hdr));
int len = RTM_PAYLOAD(hdr);
for (struct rtattr* attr = RTM_RTA(msg); RTA_OK(attr, len);
attr = RTA_NEXT(attr, len)) {
if (attr->rta_type == RTA_DST) {
char v4_address[INET_ADDRSTRLEN] = {};
char v6_address[INET6_ADDRSTRLEN] = {};
switch (family) {
case AF_INET:
inet_ntop(AF_INET, RTA_DATA(attr), v4_address,
sizeof(v4_address));
if (strcmp(v4_address, dst_v4_address.c_str()) == 0) {
routeDstFound = true;
return;
}
break;
case AF_INET6:
inet_ntop(AF_INET6, RTA_DATA(attr), v6_address,
sizeof(v6_address));
if (strcmp(v6_address, dst_v6_address.c_str()) == 0) {
routeDstFound = true;
return;
}
break;
}
}
}
},
false));
// No route that matches the given destination address can be found.
EXPECT_FALSE(routeDstFound);
// Removing a route that doesn't exist returns an error.
EXPECT_THAT(
DelUnicastRoute(loopback_link.index, family, prefixlen, dst, dst_len),
PosixErrorIs(ESRCH, _));
}
TEST_P(NetlinkRouteIpInvariantTest, AddAndRemoveRoute) {
// Gvisor does not support `RTM_NEWROUTE` or `RTM_DELROUTE`.
SKIP_IF(IsRunningOnGvisor() && GvisorPlatform() != Platform::kStarnix);
// CAP_NET_ADMIN is required to modify the routing table.
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
// Based on the test parameter, build an IPv4 or IPv6 destination subnet.
int family = GetParam();
struct in_addr dst_v4;
struct in6_addr dst_v6;
void* dst = nullptr;
int dst_len;
int prefixlen;
switch (family) {
case AF_INET:
ASSERT_EQ(inet_pton(family, "192.0.2.0", &dst_v4), 1);
prefixlen = 24;
dst = &dst_v4;
dst_len = sizeof(dst_v4);
break;
case AF_INET6:
ASSERT_EQ(inet_pton(family, "2001:db8::", &dst_v6), 1);
prefixlen = 64;
dst = &dst_v6;
dst_len = sizeof(dst_v6);
break;
default:
FAIL() << "address family must be AF_INET or AF_INET6";
}
Link loopback_link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
// Create should succeed, as no such route in kernel.
ASSERT_NO_ERRNO(
AddUnicastRoute(loopback_link.index, family, prefixlen, dst, dst_len));
// Second create should fail, as we already created the route above.
EXPECT_THAT(
AddUnicastRoute(loopback_link.index, family, prefixlen, dst, dst_len),
PosixErrorIs(EEXIST, _));
// First delete should succeed, as route exists.
EXPECT_NO_ERRNO(
DelUnicastRoute(loopback_link.index, family, prefixlen, dst, dst_len));
// Second delete should fail, as route no longer exists.
EXPECT_THAT(
DelUnicastRoute(loopback_link.index, family, prefixlen, dst, dst_len),
PosixErrorIs(ESRCH, _));
}
// GetRuleDump tests a RTM_GETRULE + NLM_F_DUMP request.
TEST(NetlinkRouteTest, GetRuleDump) {
// Gvisor does not support `RTM_GETRULE`
SKIP_IF(IsRunningOnGvisor() && GvisorPlatform() != Platform::kStarnix);
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get()));
struct request {
struct nlmsghdr hdr;
struct rtmsg rtm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETRULE;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
req.hdr.nlmsg_seq = kSeq;
req.rtm.rtm_family = AF_UNSPEC;
bool ruleFound = false;
ASSERT_NO_ERRNO(NetlinkRequestResponse(
fd, &req, sizeof(req),
[&](const struct nlmsghdr* hdr) {
// Validate the response to RTM_GETRULE + NLM_F_DUMP.
EXPECT_THAT(hdr->nlmsg_type, AnyOf(Eq(RTM_NEWRULE), Eq(NLMSG_DONE)));
EXPECT_TRUE((hdr->nlmsg_flags & NLM_F_MULTI) == NLM_F_MULTI)
<< std::hex << hdr->nlmsg_flags;
EXPECT_EQ(hdr->nlmsg_seq, kSeq);
EXPECT_EQ(hdr->nlmsg_pid, port);
// The test should not proceed if the multipart message is done.
if (hdr->nlmsg_type == NLMSG_DONE) {
return;
}
// RTM_NEWRULE contains at least the header and rule.
ASSERT_GE(hdr->nlmsg_len, NLMSG_SPACE(sizeof(struct fib_rule_hdr)));
const struct fib_rule_hdr* rule =
reinterpret_cast<const struct fib_rule_hdr*>(NLMSG_DATA(hdr));
std::cout << std::dec << "Found rule"
<< ": family=" << static_cast<int>(rule->family)
<< ", table=" << static_cast<int>(rule->table)
<< ", action=" << static_cast<int>(rule->action) << std::endl;
// All rules should have a non-zero action.
EXPECT_NE(rule->action, 0);
ruleFound = true;
},
false));
// At least one rule found.
EXPECT_TRUE(ruleFound);
}
TEST_P(NetlinkRouteIpInvariantTest, AddAndRemoveRule) {
// Gvisor does not support `RTM_NEWRULE` or `RTM_DELRULE`.
SKIP_IF(IsRunningOnGvisor() && GvisorPlatform() != Platform::kStarnix);
// CAP_NET_ADMIN is required to modify the rule table.
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
// Based on the test parameter, build an IPv4 or IPv6 destination subnet.
int family = GetParam();
struct in_addr dst_v4;
struct in6_addr dst_v6;
void* dst = nullptr;
int dst_len;
int prefixlen;
switch (family) {
case AF_INET:
ASSERT_EQ(inet_pton(family, "192.0.2.0", &dst_v4), 1);
prefixlen = 24;
dst = &dst_v4;
dst_len = sizeof(dst_v4);
break;
case AF_INET6:
ASSERT_EQ(inet_pton(family, "2001:db8::", &dst_v6), 1);
prefixlen = 64;
dst = &dst_v6;
dst_len = sizeof(dst_v6);
break;
default:
FAIL() << "address family must be AF_INET or AF_INET6";
}
// Unique values for table and priority fields to ensure the new rule does not
// collide with any of the default rules installed by Linux.
const int kTable = 99;
const int kPriority = 12345;
Link loopback_link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
// Create should succeed, as no such rule in the kernel.
ASSERT_NO_ERRNO(AddExclusiveLookupInTableRule(family, kTable, kPriority,
prefixlen, dst, dst_len));
// Second create should fail, as we already created the rule above.
EXPECT_THAT(AddExclusiveLookupInTableRule(family, kTable, kPriority,
prefixlen, dst, dst_len),
PosixErrorIs(EEXIST, _));
// First delete should succeed, as rule exists.
EXPECT_NO_ERRNO(
DelLookupInTableRule(family, kTable, kPriority, prefixlen, dst, dst_len));
// Second delete should fail, as rule no longer exists.
EXPECT_THAT(
DelLookupInTableRule(family, kTable, kPriority, prefixlen, dst, dst_len),
PosixErrorIs(ENOENT, _));
}
// RecvmsgTrunc tests the recvmsg MSG_TRUNC flag with zero length output
// buffer. MSG_TRUNC with a zero length buffer should consume subsequent
// messages off the socket.
TEST(NetlinkRouteTest, RecvmsgTrunc) {
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct rtgenmsg rgm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETADDR;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
req.hdr.nlmsg_seq = kSeq;
req.rgm.rtgen_family = AF_UNSPEC;
struct iovec iov = {};
iov.iov_base = &req;
iov.iov_len = sizeof(req);
struct msghdr msg = {};
msg.msg_iov = &iov;
msg.msg_iovlen = 1;
ASSERT_THAT(RetryEINTR(sendmsg)(fd.get(), &msg, 0), SyscallSucceeds());
iov.iov_base = NULL;
iov.iov_len = 0;
int trunclen, trunclen2;
// Note: This test assumes at least two messages are returned by the
// RTM_GETADDR request. That means at least one RTM_NEWLINK message and one
// NLMSG_DONE message. We cannot read all the messages without blocking
// because we would need to read the message into a buffer and check the
// nlmsg_type for NLMSG_DONE. However, the test depends on reading into a
// zero-length buffer.
// First, call recvmsg with MSG_TRUNC. This will read the full message from
// the socket and return it's full length. Subsequent calls to recvmsg will
// read the next messages from the socket.
ASSERT_THAT(trunclen = RetryEINTR(recvmsg)(fd.get(), &msg, MSG_TRUNC),
SyscallSucceeds());
// Message should always be truncated. However, While the destination iov is
// zero length, MSG_TRUNC returns the size of the next message so it should
// not be zero.
ASSERT_EQ(msg.msg_flags & MSG_TRUNC, MSG_TRUNC);
ASSERT_NE(trunclen, 0);
// Returned length is at least the header and ifaddrmsg.
EXPECT_GE(trunclen, sizeof(struct nlmsghdr) + sizeof(struct ifaddrmsg));
// Reset the msg_flags to make sure that the recvmsg call is setting them
// properly.
msg.msg_flags = 0;
// Make a second recvvmsg call to get the next message.
ASSERT_THAT(trunclen2 = RetryEINTR(recvmsg)(fd.get(), &msg, MSG_TRUNC),
SyscallSucceeds());
ASSERT_EQ(msg.msg_flags & MSG_TRUNC, MSG_TRUNC);
ASSERT_NE(trunclen2, 0);
// Assert that the received messages are not the same.
//
// We are calling recvmsg with a zero length buffer so we have no way to
// inspect the messages to make sure they are not equal in value. The best
// we can do is to compare their lengths.
ASSERT_NE(trunclen, trunclen2);
}
// RecvmsgTruncPeek tests recvmsg with the combination of the MSG_TRUNC and
// MSG_PEEK flags and a zero length output buffer. This is normally used to
// read the full length of the next message on the socket without consuming
// it, so a properly sized buffer can be allocated to store the message. This
// test tests that scenario.
TEST(NetlinkRouteTest, RecvmsgTruncPeek) {
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct request {
struct nlmsghdr hdr;
struct rtgenmsg rgm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETADDR;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
req.hdr.nlmsg_seq = kSeq;
req.rgm.rtgen_family = AF_UNSPEC;
struct iovec iov = {};
iov.iov_base = &req;
iov.iov_len = sizeof(req);
struct msghdr msg = {};
msg.msg_iov = &iov;
msg.msg_iovlen = 1;
ASSERT_THAT(RetryEINTR(sendmsg)(fd.get(), &msg, 0), SyscallSucceeds());
int type = -1;
do {
int peeklen;
int len;
iov.iov_base = NULL;
iov.iov_len = 0;
// Call recvmsg with MSG_PEEK and MSG_TRUNC. This will peek at the message
// and return it's full length.
// See: MSG_TRUNC http://man7.org/linux/man-pages/man2/recv.2.html
ASSERT_THAT(
peeklen = RetryEINTR(recvmsg)(fd.get(), &msg, MSG_PEEK | MSG_TRUNC),
SyscallSucceeds());
// Message should always be truncated.
ASSERT_EQ(msg.msg_flags & MSG_TRUNC, MSG_TRUNC);
ASSERT_NE(peeklen, 0);
// Reset the message flags for the next call.
msg.msg_flags = 0;
// Make the actual call to recvmsg to get the actual data. We will use
// the length returned from the peek call for the allocated buffer size..
std::vector<char> buf(peeklen);
iov.iov_base = buf.data();
iov.iov_len = buf.size();
ASSERT_THAT(len = RetryEINTR(recvmsg)(fd.get(), &msg, 0),
SyscallSucceeds());
// Message should not be truncated since we allocated the correct buffer
// size.
EXPECT_NE(msg.msg_flags & MSG_TRUNC, MSG_TRUNC);
// MSG_PEEK should have left data on the socket and the subsequent call
// with should have retrieved the same data. Both calls should have
// returned the message's full length so they should be equal.
ASSERT_NE(len, 0);
ASSERT_EQ(peeklen, len);
for (struct nlmsghdr* hdr = reinterpret_cast<struct nlmsghdr*>(buf.data());
NLMSG_OK(hdr, len); hdr = NLMSG_NEXT(hdr, len)) {
type = hdr->nlmsg_type;
}
} while (type != NLMSG_DONE && type != NLMSG_ERROR);
}
// No SCM_CREDENTIALS are received without SO_PASSCRED set.
TEST(NetlinkRouteTest, NoPasscredNoCreds) {
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
ASSERT_THAT(setsockopt(fd.get(), SOL_SOCKET, SO_PASSCRED, &kSockOptOff,
sizeof(kSockOptOff)),
SyscallSucceeds());
struct request {
struct nlmsghdr hdr;
struct rtgenmsg rgm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETADDR;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
req.hdr.nlmsg_seq = kSeq;
req.rgm.rtgen_family = AF_UNSPEC;
struct iovec iov = {};
iov.iov_base = &req;
iov.iov_len = sizeof(req);
struct msghdr msg = {};
msg.msg_iov = &iov;
msg.msg_iovlen = 1;
ASSERT_THAT(RetryEINTR(sendmsg)(fd.get(), &msg, 0), SyscallSucceeds());
iov.iov_base = NULL;
iov.iov_len = 0;
char control[CMSG_SPACE(sizeof(struct ucred))] = {};
msg.msg_control = control;
msg.msg_controllen = sizeof(control);
// Note: This test assumes at least one message is returned by the
// RTM_GETADDR request.
ASSERT_THAT(RetryEINTR(recvmsg)(fd.get(), &msg, 0), SyscallSucceeds());
// No control messages.
EXPECT_EQ(CMSG_FIRSTHDR(&msg), nullptr);
}
// SCM_CREDENTIALS are received with SO_PASSCRED set.
TEST(NetlinkRouteTest, PasscredCreds) {
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
ASSERT_THAT(setsockopt(fd.get(), SOL_SOCKET, SO_PASSCRED, &kSockOptOn,
sizeof(kSockOptOn)),
SyscallSucceeds());
struct request {
struct nlmsghdr hdr;
struct rtgenmsg rgm;
};
struct request req = {};
req.hdr.nlmsg_len = sizeof(req);
req.hdr.nlmsg_type = RTM_GETADDR;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
req.hdr.nlmsg_seq = kSeq;
req.rgm.rtgen_family = AF_UNSPEC;
struct iovec iov = {};
iov.iov_base = &req;
iov.iov_len = sizeof(req);
struct msghdr msg = {};
msg.msg_iov = &iov;
msg.msg_iovlen = 1;
ASSERT_THAT(RetryEINTR(sendmsg)(fd.get(), &msg, 0), SyscallSucceeds());
iov.iov_base = NULL;
iov.iov_len = 0;
char control[CMSG_SPACE(sizeof(struct ucred))] = {};
msg.msg_control = control;
msg.msg_controllen = sizeof(control);
// Note: This test assumes at least one message is returned by the
// RTM_GETADDR request.
ASSERT_THAT(RetryEINTR(recvmsg)(fd.get(), &msg, 0), SyscallSucceeds());
struct ucred creds;
struct cmsghdr* cmsg = CMSG_FIRSTHDR(&msg);
ASSERT_NE(cmsg, nullptr);
ASSERT_EQ(cmsg->cmsg_len, CMSG_LEN(sizeof(creds)));
ASSERT_EQ(cmsg->cmsg_level, SOL_SOCKET);
ASSERT_EQ(cmsg->cmsg_type, SCM_CREDENTIALS);
memcpy(&creds, CMSG_DATA(cmsg), sizeof(creds));
// The peer is the kernel, which is "PID" 0.
EXPECT_EQ(creds.pid, 0);
// The kernel identifies as root. Also allow nobody in case this test is
// running in a userns without root mapped.
EXPECT_THAT(creds.uid, AnyOf(Eq(0), Eq(65534)));
EXPECT_THAT(creds.gid, AnyOf(Eq(0), Eq(65534)));
}
#ifndef NLMSG_TAIL
#define NLMSG_TAIL(nmsg) \
((struct rtattr*)(((char*)(nmsg)) + NLMSG_ALIGN((nmsg)->nlmsg_len)))
#endif
void addattr(struct nlmsghdr* n, int maxlen, int type, const void* data,
int alen) {
int len = NLA_HDRLEN + alen;
struct rtattr* rta;
ASSERT_LE(NLMSG_ALIGN(n->nlmsg_len) + RTA_ALIGN(len), maxlen);
rta = NLMSG_TAIL(n);
rta->rta_type = type;
rta->rta_len = len;
memcpy(RTA_DATA(rta), data, alen);
n->nlmsg_len = NLMSG_ALIGN(n->nlmsg_len) + RTA_ALIGN(len);
}
struct VethRequest {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
char buf[1024];
};
struct VethRequest GetVethRequest(uint32_t seq, const char* ifname_first,
const char* ifname_second) {
struct VethRequest req = {};
req.hdr.nlmsg_len = NLMSG_LENGTH(sizeof(struct ifinfomsg));
req.hdr.nlmsg_type = RTM_NEWLINK;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE;
req.hdr.nlmsg_seq = seq;
req.ifm.ifi_family = AF_UNSPEC;
req.ifm.ifi_index = 0;
addattr(&req.hdr, sizeof(req), IFLA_IFNAME, ifname_first,
strlen(ifname_first));
struct rtattr* linkinfo = NLMSG_TAIL(&req.hdr);
{
addattr(&req.hdr, sizeof(req), IFLA_LINKINFO, nullptr, 0);
addattr(&req.hdr, sizeof(req), IFLA_INFO_KIND, "veth", 4);
struct rtattr* veth_data = NLMSG_TAIL(&req.hdr);
{
addattr(&req.hdr, sizeof(req), IFLA_INFO_DATA, NULL, 0);
struct rtattr* peer_data = NLMSG_TAIL(&req.hdr);
{
struct ifinfomsg ifm = {};
addattr(&req.hdr, sizeof(req), VETH_INFO_PEER, &ifm, sizeof(ifm));
addattr(&req.hdr, sizeof(req), IFLA_IFNAME, ifname_second,
strlen(ifname_second));
}
peer_data->rta_len = (uint64_t)NLMSG_TAIL(&req.hdr) - (uint64_t)peer_data;
}
veth_data->rta_len = (uint64_t)NLMSG_TAIL(&req.hdr) - (uint64_t)veth_data;
}
linkinfo->rta_len = (uint64_t)NLMSG_TAIL(&req.hdr) - (uint64_t)linkinfo;
return req;
}
// Builds a veth create whose VETH_INFO_PEER payload is too short to hold the
// peer's ifinfomsg.
struct VethRequest GetShortPeerVethRequest(uint32_t seq, const char* ifname) {
struct VethRequest req = {};
req.hdr.nlmsg_len = NLMSG_LENGTH(sizeof(struct ifinfomsg));
req.hdr.nlmsg_type = RTM_NEWLINK;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE;
req.hdr.nlmsg_seq = seq;
req.ifm.ifi_family = AF_UNSPEC;
req.ifm.ifi_index = 0;
addattr(&req.hdr, sizeof(req), IFLA_IFNAME, ifname, strlen(ifname));
struct rtattr* linkinfo = NLMSG_TAIL(&req.hdr);
{
addattr(&req.hdr, sizeof(req), IFLA_LINKINFO, nullptr, 0);
addattr(&req.hdr, sizeof(req), IFLA_INFO_KIND, "veth", 4);
struct rtattr* veth_data = NLMSG_TAIL(&req.hdr);
{
addattr(&req.hdr, sizeof(req), IFLA_INFO_DATA, NULL, 0);
uint32_t truncated = 0;
addattr(&req.hdr, sizeof(req), VETH_INFO_PEER, &truncated,
sizeof(truncated));
}
veth_data->rta_len = (uint64_t)NLMSG_TAIL(&req.hdr) - (uint64_t)veth_data;
}
linkinfo->rta_len = (uint64_t)NLMSG_TAIL(&req.hdr) - (uint64_t)linkinfo;
return req;
}
struct BridgeRequest {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
char buf[1024];
};
struct BridgeRequest GetBridgeRequest(uint32_t seq, const char* ifname) {
struct BridgeRequest req = {};
req.hdr.nlmsg_len = NLMSG_LENGTH(sizeof(struct ifinfomsg));
req.hdr.nlmsg_type = RTM_NEWLINK;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE;
req.hdr.nlmsg_seq = seq;
req.ifm.ifi_family = AF_UNSPEC;
req.ifm.ifi_index = 0;
addattr(&req.hdr, sizeof(req), IFLA_IFNAME, ifname, strlen(ifname));
struct rtattr* linkinfo = NLMSG_TAIL(&req.hdr);
{
addattr(&req.hdr, sizeof(req), IFLA_LINKINFO, nullptr, 0);
addattr(&req.hdr, sizeof(req), IFLA_INFO_KIND, "bridge", 6);
}
linkinfo->rta_len = (uint64_t)NLMSG_TAIL(&req.hdr) - (uint64_t)linkinfo;
return req;
}
void DumpInetAddressSequence(std::vector<std::string>* sequence) {
struct ifaddrs* if_addr_list = nullptr;
ASSERT_THAT(getifaddrs(&if_addr_list), SyscallSucceeds());
auto cleanup = Cleanup([&if_addr_list]() { freeifaddrs(if_addr_list); });
sequence->clear();
for (struct ifaddrs* i = if_addr_list; i; i = i->ifa_next) {
if (!i->ifa_addr || (i->ifa_addr->sa_family != AF_INET &&
i->ifa_addr->sa_family != AF_INET6)) {
continue;
}
char buf[INET6_ADDRSTRLEN] = {};
switch (i->ifa_addr->sa_family) {
case AF_INET:
inet_ntop(AF_INET,
&(reinterpret_cast<const struct sockaddr_in*>(i->ifa_addr)
->sin_addr),
buf, sizeof(buf));
break;
case AF_INET6:
inet_ntop(AF_INET6,
&(reinterpret_cast<const struct sockaddr_in6*>(i->ifa_addr)
->sin6_addr),
buf, sizeof(buf));
break;
}
sequence->push_back(
absl::StrFormat("%s/%d/%s", i->ifa_name, i->ifa_addr->sa_family, buf));
}
}
void ReadProcNetDevInterfaceNames(std::vector<std::string>* names) {
names->clear();
const std::string contents =
ASSERT_NO_ERRNO_AND_VALUE(GetContents("/proc/net/dev"));
std::istringstream input(contents);
std::string line;
for (int line_no = 0; std::getline(input, line); ++line_no) {
if (line_no < 2) {
continue;
}
const size_t colon = line.find(':');
if (colon == std::string::npos) {
continue;
}
std::string name = line.substr(0, colon);
const size_t first = name.find_first_not_of(" \t");
if (first == std::string::npos) {
continue;
}
const size_t last = name.find_last_not_of(" \t");
names->push_back(name.substr(first, last - first + 1));
}
}
TEST(NetlinkRouteTest, VethAdd) {
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
SKIP_IF(IsRunningWithHostinet());
// Running the test in an ephemeral netns allows it to not interfere with
// other tests that also want to create veth pairs with the same names.
const FileDescriptor curr_nsfd =
ASSERT_NO_ERRNO_AND_VALUE(Open("/proc/thread-self/ns/net", O_RDONLY));
Cleanup restore_netns = Cleanup([&] {
ASSERT_THAT(setns(curr_nsfd.get(), CLONE_NEWNET),
SyscallSucceedsWithValue(0));
});
ASSERT_THAT(unshare(CLONE_NEWNET), SyscallSucceedsWithValue(0));
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
VethRequest req = GetVethRequest(kSeq, "veth1", "veth2");
EXPECT_NO_ERRNO(NetlinkRequestAckOrError(fd, kSeq, &req, req.hdr.nlmsg_len));
}
TEST(NetlinkRouteTest, VethAddShortPeerIfInfoMsg) {
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
SKIP_IF(IsRunningWithHostinet());
const FileDescriptor curr_nsfd =
ASSERT_NO_ERRNO_AND_VALUE(Open("/proc/thread-self/ns/net", O_RDONLY));
Cleanup restore_netns = Cleanup([&] {
ASSERT_THAT(setns(curr_nsfd.get(), CLONE_NEWNET),
SyscallSucceedsWithValue(0));
});
ASSERT_THAT(unshare(CLONE_NEWNET), SyscallSucceedsWithValue(0));
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
VethRequest req = GetShortPeerVethRequest(kSeq, "veth1");
EXPECT_THAT(NetlinkRequestAckOrError(fd, kSeq, &req, req.hdr.nlmsg_len),
PosixErrorIs(AnyOf(EINVAL, ERANGE), _));
std::vector<Link> links = ASSERT_NO_ERRNO_AND_VALUE(DumpLinks(fd));
for (const Link& link : links) {
EXPECT_NE(link.name, "veth1");
}
}
TEST(NetlinkRouteTest, LinkInfoKind) {
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
SKIP_IF(IsRunningWithHostinet());
const FileDescriptor curr_nsfd =
ASSERT_NO_ERRNO_AND_VALUE(Open("/proc/thread-self/ns/net", O_RDONLY));
Cleanup restore_netns = Cleanup([&] {
ASSERT_THAT(setns(curr_nsfd.get(), CLONE_NEWNET),
SyscallSucceedsWithValue(0));
});
ASSERT_THAT(unshare(CLONE_NEWNET), SyscallSucceedsWithValue(0));
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
VethRequest veth_req = GetVethRequest(kSeq, "veth1", "veth2");
ASSERT_NO_ERRNO(
NetlinkRequestAckOrError(fd, kSeq, &veth_req, veth_req.hdr.nlmsg_len));
BridgeRequest br_req = GetBridgeRequest(kSeq + 1, "br0");
ASSERT_NO_ERRNO(
NetlinkRequestAckOrError(fd, kSeq + 1, &br_req, br_req.hdr.nlmsg_len));
std::vector<Link> links = ASSERT_NO_ERRNO_AND_VALUE(DumpLinks(fd));
std::map<std::string, std::string> link_kinds;
for (const Link& link : links) {
link_kinds[link.name] = link.kind;
}
EXPECT_EQ(link_kinds["veth1"], "veth");
EXPECT_EQ(link_kinds["veth2"], "veth");
EXPECT_EQ(link_kinds["br0"], "bridge");
EXPECT_EQ(link_kinds["lo"], "");
}
TEST(NetlinkRouteTest, LookupAllAddrOrder) {
// Run the test multiple times to identify any flakiness with the order of
// addresses returned. The order should be IPv4(AF_INET = 2) addresses
// followed by IPv6(AF_INET6 = 10).
for (int i = 0; i < 10; i++) {
struct ifaddrs* if_addr_list = nullptr;
// Not a syscall but we can use the syscall matcher as glibc sets errno.
ASSERT_THAT(getifaddrs(&if_addr_list), SyscallSucceeds());
std::vector<int> addrs_family;
for (struct ifaddrs* i = if_addr_list; i; i = i->ifa_next) {
if (!i->ifa_addr || (i->ifa_addr->sa_family != AF_INET &&
i->ifa_addr->sa_family != AF_INET6)) {
continue;
}
addrs_family.push_back(i->ifa_addr->sa_family);
}
ASSERT_TRUE(std::is_sorted(addrs_family.begin(), addrs_family.end()));
freeifaddrs(if_addr_list);
}
}
TEST(NetlinkRouteTest, LookupAllAddrStableAcrossCalls) {
std::vector<std::string> baseline;
DumpInetAddressSequence(&baseline);
ASSERT_FALSE(baseline.empty());
for (int i = 0; i < 10; ++i) {
std::vector<std::string> current;
DumpInetAddressSequence(&current);
EXPECT_EQ(current, baseline);
}
}
TEST(NetlinkRouteTest, LinIndexOrder) {
// This test verifies that links are sorted by index in ascending order.
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
SKIP_IF(IsRunningWithHostinet());
const FileDescriptor curr_nsfd =
ASSERT_NO_ERRNO_AND_VALUE(Open("/proc/thread-self/ns/net", O_RDONLY));
Cleanup restore_netns = Cleanup([&] {
ASSERT_THAT(setns(curr_nsfd.get(), CLONE_NEWNET),
SyscallSucceedsWithValue(0));
});
ASSERT_THAT(unshare(CLONE_NEWNET), SyscallSucceedsWithValue(0));
// Add some interfaces
FileDescriptor fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
VethRequest req3 = GetVethRequest(kSeq, "veth3a", "veth3b");
ASSERT_NO_ERRNO(
NetlinkRequestAckOrError(fd, kSeq, &req3, req3.hdr.nlmsg_len));
VethRequest req1 = GetVethRequest(kSeq, "veth1a", "veth1b");
ASSERT_NO_ERRNO(
NetlinkRequestAckOrError(fd, kSeq, &req1, req1.hdr.nlmsg_len));
VethRequest req2 = GetVethRequest(kSeq, "veth2a", "veth2b");
ASSERT_NO_ERRNO(
NetlinkRequestAckOrError(fd, kSeq, &req2, req2.hdr.nlmsg_len));
// Dump links and verify they come back in ascending order by index.
std::vector<int> found_indexes;
ASSERT_NO_ERRNO(DumpLinks(fd, kSeq, [&](const struct nlmsghdr* hdr) {
if (hdr->nlmsg_type != RTM_NEWLINK) {
return;
}
if (hdr->nlmsg_len < NLMSG_SPACE(sizeof(struct ifinfomsg))) {
return;
}
const struct ifinfomsg* msg =
reinterpret_cast<const struct ifinfomsg*>(NLMSG_DATA(hdr));
found_indexes.push_back(msg->ifi_index);
}));
ASSERT_GE(found_indexes.size(), 6)
<< "Expected at least 3 veth pairs to be found";
ASSERT_TRUE(std::is_sorted(found_indexes.begin(), found_indexes.end()))
<< "Link indexes are not in ascending order";
}
TEST(NetlinkRouteTest, ProcNetDevStableAcrossCalls) {
SKIP_IF(IsRunningWithHostinet());
// Avoid a temporary netns here: in gVisor `/proc/net/*` is not yet
// guaranteed to follow `unshare(CLONE_NEWNET)` (gvisor.dev/issue/1833).
// Native Linux may also see concurrent device churn, so only check that
// baseline interfaces keep the same relative order.
std::vector<std::string> baseline;
ReadProcNetDevInterfaceNames(&baseline);
ASSERT_FALSE(baseline.empty());
for (int i = 0; i < 10; ++i) {
std::vector<std::string> current;
ReadProcNetDevInterfaceNames(&current);
std::vector<size_t> baseline_positions;
baseline_positions.reserve(current.size());
for (const std::string& name : current) {
auto it = std::find(baseline.begin(), baseline.end(), name);
if (it == baseline.end()) {
continue;
}
baseline_positions.push_back(std::distance(baseline.begin(), it));
}
EXPECT_TRUE(
std::is_sorted(baseline_positions.begin(), baseline_positions.end()));
}
}
struct NetNSRequest {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
char buf[1024];
};
struct NetNSRequest GetNetNSRequest(uint32_t seq, int if_index, int ns_fd) {
struct NetNSRequest req = {};
req.hdr.nlmsg_len = NLMSG_LENGTH(sizeof(struct ifinfomsg));
req.hdr.nlmsg_type = RTM_NEWLINK;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
req.hdr.nlmsg_seq = seq;
req.ifm.ifi_family = AF_UNSPEC;
req.ifm.ifi_index = if_index;
addattr(&req.hdr, sizeof(req), IFLA_NET_NS_FD, &ns_fd, sizeof(ns_fd));
return req;
}
// Guard against b/456552490, a bug where gvisor'd try to modify the given
// attributes in the wrong stack when the request also directs it to change the
// netns.
TEST(NetlinkRouteTest, ChangeNetnsAndOtherAttrsTogether) {
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
SKIP_IF(IsRunningWithHostinet());
// TODO(gvisor.dev/issue/4595): enable cooperative save tests.
const DisableSave ds;
// Running the test in an ephemeral netns allows it to not interfere with
// other tests that also want to create veth pairs with the same names.
const FileDescriptor curr_nsfd =
ASSERT_NO_ERRNO_AND_VALUE(Open("/proc/thread-self/ns/net", O_RDONLY));
Cleanup restore_netns = Cleanup([&] {
ASSERT_THAT(setns(curr_nsfd.get(), CLONE_NEWNET),
SyscallSucceedsWithValue(0));
});
ASSERT_THAT(unshare(CLONE_NEWNET), SyscallSucceedsWithValue(0));
FileDescriptor nlsk =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
VethRequest req = GetVethRequest(kSeq, "veth1", "veth2");
EXPECT_NO_ERRNO(
NetlinkRequestAckOrError(nlsk, kSeq, &req, req.hdr.nlmsg_len));
int inner_veth_idx = if_nametoindex("veth2");
ASSERT_NE(inner_veth_idx, 0);
// Enter a new network namespace and move veth2 into it.
ASSERT_THAT(unshare(CLONE_NEWNET), SyscallSucceedsWithValue(0));
const FileDescriptor inner_nsfd =
ASSERT_NO_ERRNO_AND_VALUE(Open("/proc/thread-self/ns/net", O_RDONLY));
NetNSRequest set_netns_req =
GetNetNSRequest(kSeq, inner_veth_idx, inner_nsfd.get());
// Request a name change along with the netns change.
std::string new_name = "veth2_new";
addattr(&set_netns_req.hdr, sizeof(set_netns_req), IFLA_IFNAME,
new_name.c_str(), new_name.size());
EXPECT_NO_ERRNO(NetlinkRequestAckOrError(nlsk, kSeq, &set_netns_req,
set_netns_req.hdr.nlmsg_len));
// Verify that an interface with the new name exists in the inner netns.
bool found_name_inner = false;
std::vector<Link> links = ASSERT_NO_ERRNO_AND_VALUE(DumpLinks());
for (const Link& link : links) {
if (link.name == new_name) {
found_name_inner = true;
break;
}
}
EXPECT_TRUE(found_name_inner)
<< "Did not find interface with name " << new_name;
// And verify also that the outer netns does not have the new name.
links = ASSERT_NO_ERRNO_AND_VALUE(DumpLinks(nlsk));
for (const Link& link : links) {
ASSERT_NE(link.name, new_name)
<< "Found interface with name " << new_name << " in outer netns";
}
}
struct NameRequest {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
struct rtattr rtattr;
char name[IFNAMSIZ];
};
NameRequest GetNameRequest(const Link& link, const char* name, uint32_t seq) {
NameRequest req = {};
req.hdr.nlmsg_type = RTM_SETLINK;
req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
req.hdr.nlmsg_seq = seq;
req.ifm.ifi_family = AF_UNSPEC;
req.ifm.ifi_index = link.index;
const size_t payload_len = strlen(name) + 1;
req.rtattr.rta_type = IFLA_IFNAME;
req.rtattr.rta_len = RTA_LENGTH(payload_len);
memcpy(req.name, name, payload_len);
req.hdr.nlmsg_len =
NLMSG_LENGTH(sizeof(struct ifinfomsg)) + RTA_SPACE(payload_len);
return req;
};
TEST(NetlinkRouteTest, LinkMulticastGroupBasic) {
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
SKIP_IF(IsRunningWithHostinet());
// TODO(gvisor.dev/issue/4595): enable cooperative save tests.
const DisableSave ds;
// nlsk_bound_group joins RTMGRP_LINK via bind().
struct sockaddr_nl addr = {};
addr.nl_family = AF_NETLINK;
addr.nl_groups = RTMGRP_LINK;
FileDescriptor nlsk_bound_group =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE, &addr));
// nlsk_sockopt_group joins RTMGRP_LINK via setsockopt().
addr = {};
addr.nl_family = AF_NETLINK;
FileDescriptor nlsk_sockopt_group =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE, &addr));
unsigned int group = RTMGRP_LINK;
ASSERT_THAT(setsockopt(nlsk_sockopt_group.get(), SOL_NETLINK,
NETLINK_ADD_MEMBERSHIP, &group, sizeof(group)),
SyscallSucceeds());
int64_t res_groups;
socklen_t res_groups_len = sizeof(res_groups);
EXPECT_THAT(
getsockopt(nlsk_sockopt_group.get(), SOL_NETLINK,
NETLINK_LIST_MEMBERSHIPS, &res_groups, &res_groups_len),
SyscallSucceeds());
EXPECT_EQ(res_groups_len, sizeof(res_groups));
EXPECT_EQ(res_groups, RTMGRP_LINK);
FileDescriptor control_fd =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
// Change the name of the loopback interface.
const Link link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
std::string old_loopback_name = link.name;
NameRequest name_request = GetNameRequest(link, "lo_test", kSeq);
ASSERT_NO_ERRNO(NetlinkRequestAckOrError(control_fd, kSeq, &name_request,
name_request.hdr.nlmsg_len));
auto restore_loopback_name = Cleanup([&]() {
NameRequest name_request =
GetNameRequest(link, old_loopback_name.c_str(), kSeq);
ASSERT_NO_ERRNO(NetlinkRequestAckOrError(control_fd, kSeq, &name_request,
name_request.hdr.nlmsg_len));
});
const Link link_newname = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
// Change the MTU of the loopback interface.
MtuRequest mtu_request = GetMtuRequest(link, RTM_SETLINK, link.mtu + 10);
ASSERT_NO_ERRNO(NetlinkRequestAckOrError(control_fd, kSeq, &mtu_request,
sizeof(mtu_request)));
auto restore_mtu = Cleanup([&]() {
MtuRequest mtu_request = GetMtuRequest(link, RTM_SETLINK, link.mtu);
ASSERT_NO_ERRNO(NetlinkRequestAckOrError(control_fd, kSeq, &mtu_request,
sizeof(mtu_request)));
});
const Link link_newmtu = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
struct TestCase {
const char* name;
FileDescriptor* nlsk;
const char* event_name;
const Link& link;
};
std::vector<TestCase> test_cases = {
{
.name = "bound_group",
.nlsk = &nlsk_bound_group,
.event_name = "name_change",
.link = link_newname,
},
{
.name = "sockopt_group",
.nlsk = &nlsk_sockopt_group,
.event_name = "name_change",
.link = link_newname,
},
{
.name = "bound_group",
.nlsk = &nlsk_bound_group,
.event_name = "mtu_change",
.link = link_newmtu,
},
{
.name = "sockopt_group",
.nlsk = &nlsk_sockopt_group,
.event_name = "mtu_change",
.link = link_newmtu,
},
};
for (const auto& tc : test_cases) {
SCOPED_TRACE(std::string(tc.name) + " " + tc.event_name);
struct pollfd pfd = {.fd = tc.nlsk->get(), .events = POLLIN};
constexpr int kPollTimeoutMs = 1000;
int poll_ret = RetryEINTR(poll)(&pfd, 1, kPollTimeoutMs);
ASSERT_EQ(poll_ret, 1);
bool got_msg = false;
ASSERT_NO_ERRNO(NetlinkResponse(
*tc.nlsk,
[&](const struct nlmsghdr* hdr) {
const struct ifinfomsg* msg =
reinterpret_cast<const struct ifinfomsg*>(NLMSG_DATA(hdr));
if (msg->ifi_index != tc.link.index) {
return;
}
EXPECT_EQ(hdr->nlmsg_pid, 0);
CheckLinkMsg(hdr, tc.link);
got_msg = true;
},
/*expect_nlmsgerr=*/false));
EXPECT_TRUE(got_msg);
}
}
// To verify the namespaced nature of the netlink multicast groups.
TEST(NetlinkRouteTest, LinkMulticastGroupNamespaced) {
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
SKIP_IF(IsRunningWithHostinet());
// TODO(gvisor.dev/issue/4595): enable cooperative save tests.
const DisableSave ds;
// Running the test in an ephemeral netns allows it to not interfere with
// other tests that also want to create veth pairs with the same names.
const FileDescriptor curr_nsfd =
ASSERT_NO_ERRNO_AND_VALUE(Open("/proc/thread-self/ns/net", O_RDONLY));
Cleanup restore_netns = Cleanup([&] {
ASSERT_THAT(setns(curr_nsfd.get(), CLONE_NEWNET),
SyscallSucceedsWithValue(0));
});
ASSERT_THAT(unshare(CLONE_NEWNET), SyscallSucceedsWithValue(0));
FileDescriptor control_nlsk =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
VethRequest req = GetVethRequest(kSeq, "veth1", "veth2");
EXPECT_NO_ERRNO(
NetlinkRequestAckOrError(control_nlsk, kSeq, &req, req.hdr.nlmsg_len));
int inner_veth_idx = if_nametoindex("veth2");
ASSERT_NE(inner_veth_idx, 0);
struct sockaddr_nl mcast_addr = {};
mcast_addr.nl_family = AF_NETLINK;
mcast_addr.nl_groups = RTMGRP_LINK;
FileDescriptor outer_nlsk =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE, &mcast_addr));
// Enter a new network namespace.
ASSERT_THAT(unshare(CLONE_NEWNET), SyscallSucceedsWithValue(0));
FileDescriptor inner_nlsk =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE, &mcast_addr));
// And move veth2 into it.
const FileDescriptor inner_nsfd =
ASSERT_NO_ERRNO_AND_VALUE(Open("/proc/thread-self/ns/net", O_RDONLY));
NetNSRequest set_netns_req =
GetNetNSRequest(kSeq, inner_veth_idx, inner_nsfd.get());
EXPECT_NO_ERRNO(NetlinkRequestAckOrError(control_nlsk, kSeq, &set_netns_req,
set_netns_req.hdr.nlmsg_len));
constexpr int kPollTimeoutMs = 1000;
bool got_msg = false;
// We expect an RTM_DELINK message for veth2 in the outer netns socket.
// But an RTM_NEWLINK is also expected for veth1 because its peer was moved.
// Hence the two attempts. N.B. gVisor does not send the RTM_NEWLINK because
// IFLA_LINK_NETNSID is not yet supported.
for (int i = 0; i < 2; ++i) {
struct pollfd pfd = {.fd = outer_nlsk.get(), .events = POLLIN};
ASSERT_EQ(RetryEINTR(poll)(&pfd, 1, kPollTimeoutMs), 1)
<< "outer_nlsk: Did not get veth2 DELLINK";
ASSERT_NO_ERRNO(NetlinkResponse(
outer_nlsk,
[&](const struct nlmsghdr* hdr) {
const struct ifinfomsg* msg =
reinterpret_cast<const struct ifinfomsg*>(NLMSG_DATA(hdr));
if (hdr->nlmsg_type != RTM_DELLINK) return;
if (msg->ifi_index != inner_veth_idx) return;
EXPECT_EQ(hdr->nlmsg_pid, 0);
got_msg = true;
},
/*expect_nlmsgerr=*/false));
if (got_msg) break;
}
EXPECT_TRUE(got_msg) << "outer_nlsk: Did not get veth2 DELLINK";
// We expect an RTM_NEWLINK message for veth2 in the inner netns socket.
{
struct pollfd pfd = {.fd = inner_nlsk.get(), .events = POLLIN};
ASSERT_EQ(RetryEINTR(poll)(&pfd, 1, kPollTimeoutMs), 1)
<< "inner_nlsk: Did not get veth2 NEWLINK";
bool got_msg = false;
ASSERT_NO_ERRNO(NetlinkResponse(
inner_nlsk,
[&](const struct nlmsghdr* hdr) {
const struct ifinfomsg* msg =
reinterpret_cast<const struct ifinfomsg*>(NLMSG_DATA(hdr));
ASSERT_EQ(hdr->nlmsg_type, RTM_NEWLINK);
if (msg->ifi_index == 1) return; // Ignore the loopback interface.
char ifname[IF_NAMESIZE];
EXPECT_NE(if_indextoname(msg->ifi_index, ifname), nullptr);
EXPECT_STREQ(ifname, "veth2");
EXPECT_EQ(hdr->nlmsg_pid, 0);
got_msg = true;
},
/*expect_nlmsgerr=*/false));
EXPECT_TRUE(got_msg) << "inner_nlsk: Did not get veth2 NEWLINK";
}
}
// NOOP requests should not result in any netlink multicast messages.
TEST(NetlinkRouteTest, LinkMulticastGroupNoop) {
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
SKIP_IF(IsRunningWithHostinet());
// TODO(gvisor.dev/issue/4595): enable cooperative save tests.
const DisableSave ds;
struct sockaddr_nl mcast_addr = {};
mcast_addr.nl_family = AF_NETLINK;
mcast_addr.nl_groups = RTMGRP_LINK;
FileDescriptor nlsk =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE, &mcast_addr));
// Issue a request to set the name of the loopback interface to the same name.
const Link link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
NameRequest name_request = GetNameRequest(link, link.name.c_str(), kSeq);
FileDescriptor control_nlsk =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
ASSERT_NO_ERRNO(NetlinkRequestAckOrError(control_nlsk, kSeq, &name_request,
name_request.hdr.nlmsg_len));
// We expect no RTM_NEWLINK message for the loopback interface.
struct pollfd pfd = {.fd = nlsk.get(), .events = POLLIN};
constexpr int kPollTimeoutMs = 500;
bool got_msg = false;
if (RetryEINTR(poll)(&pfd, 1, kPollTimeoutMs) >= 1) {
ASSERT_NO_ERRNO(NetlinkResponse(
nlsk,
[&](const struct nlmsghdr* hdr) {
const struct ifinfomsg* msg =
reinterpret_cast<const struct ifinfomsg*>(NLMSG_DATA(hdr));
if (hdr->nlmsg_type != RTM_NEWLINK) return;
if (msg->ifi_index != link.index) return;
got_msg = true;
},
/*expect_nlmsgerr=*/false));
}
EXPECT_FALSE(got_msg)
<< "Should not get a newlink event for the loopback interface.";
}
// Userspace should know that it failed to keep up with its recvmsg()s, and the
// kernel alerts it to this by having recvmsg() return ENOBUFS.
TEST(NetlinkRouteTest, LinkMulticastGroupEnobufs) {
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
SKIP_IF(IsRunningWithHostinet());
// TODO(gvisor.dev/issue/4595): enable cooperative save tests.
const DisableSave ds;
// TODO(b/456238795): enable this test once gVisor returns ENOBUFS.
if (IsRunningOnGvisor()) {
GTEST_SKIP() << "gVisor never returns ENOBUFS.";
}
struct sockaddr_nl mcast_addr = {};
mcast_addr.nl_family = AF_NETLINK;
mcast_addr.nl_groups = RTMGRP_LINK;
FileDescriptor nlsk =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE, &mcast_addr));
// N.B. gvisor ignores the SO_RCVBUF value.
constexpr int kSmallRcvBufSize = 512;
ASSERT_THAT(setsockopt(nlsk.get(), SOL_SOCKET, SO_RCVBUF, &kSmallRcvBufSize,
sizeof(kSmallRcvBufSize)),
SyscallSucceeds());
int recv_buf_size;
socklen_t rec_buf_size_len = sizeof(recv_buf_size);
ASSERT_THAT(getsockopt(nlsk.get(), SOL_SOCKET, SO_RCVBUF, &recv_buf_size,
&rec_buf_size_len),
SyscallSucceeds());
// Generate enough link events to overflow poor nlsk's receive buffer.
FileDescriptor control_nlsk =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
Link link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
constexpr int kMinimumNewlinkMsgSize = 32;
const int num_msgs = recv_buf_size / kMinimumNewlinkMsgSize;
for (int i = 0; i < num_msgs || link.name != "lo"; ++i) {
std::string name = link.name == "lo" ? "lo_test" : "lo";
NameRequest name_request = GetNameRequest(link, name.c_str(), kSeq);
ASSERT_NO_ERRNO(NetlinkRequestAckOrError(control_nlsk, kSeq, &name_request,
name_request.hdr.nlmsg_len));
link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
}
std::vector<char> buf(kSmallRcvBufSize);
struct iovec iov = {};
iov.iov_base = buf.data();
iov.iov_len = buf.size();
struct msghdr msg = {};
msg.msg_iov = &iov;
msg.msg_iovlen = 1;
EXPECT_THAT(RetryEINTR(recvmsg)(nlsk.get(), &msg, 0),
SyscallFailsWithErrno(ENOBUFS));
}
// Tests the ability of a userspace process to simultaneously send an
// RTM_SETLINK message to both the kernel and the userland RTMGRP_LINK
// subscribers.
TEST(NetlinkRouteTest, LinkMulticastGroupUserToUserSend) {
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
SKIP_IF(IsRunningWithHostinet());
// TODO(gvisor.dev/issue/4595): enable cooperative save tests.
const DisableSave ds;
const struct sockaddr_nl recv_addr = {
.nl_family = AF_NETLINK,
.nl_groups = RTMGRP_LINK,
};
FileDescriptor nlsk_recv =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE, &recv_addr));
int passcred_val = 1;
ASSERT_THAT(setsockopt(nlsk_recv.get(), SOL_SOCKET, SO_PASSCRED,
&passcred_val, sizeof(passcred_val)),
SyscallSucceeds());
FileDescriptor nlsk_send =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct sockaddr_nl nlsk_send_addr = {};
socklen_t addr_len = sizeof(nlsk_send_addr);
ASSERT_THAT(getsockname(nlsk_send.get(), (struct sockaddr*)&nlsk_send_addr,
&addr_len),
SyscallSucceeds());
// Send an RTM_SETLINK message to change the MTU of the loopback
// interface.
struct sockaddr_nl send_addr = {
.nl_family = AF_NETLINK,
.nl_pid = 0, // Send to the kernel.
.nl_groups = RTMGRP_LINK, // and also send to userland subscribers.
};
const Link link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
MtuRequest mtu_request = GetMtuRequest(link, RTM_SETLINK, link.mtu + 10);
mtu_request.hdr.nlmsg_pid = getpid();
struct iovec iov = {
.iov_base = &mtu_request,
.iov_len = sizeof(mtu_request),
};
struct msghdr msg = {
.msg_name = &send_addr,
.msg_namelen = sizeof(send_addr),
.msg_iov = &iov,
.msg_iovlen = 1,
};
ASSERT_THAT(RetryEINTR(sendmsg)(nlsk_send.get(), &msg, 0), SyscallSucceeds());
// And verify we received a response from the kernel.
constexpr int kPollTimeoutMs = 1000;
struct pollfd pfd = {.fd = nlsk_send.get(), .events = POLLIN};
ASSERT_EQ(RetryEINTR(poll)(&pfd, 1, kPollTimeoutMs), 1)
<< "nlsk_send: Did not get the expected unicast from the kernel.";
bool got_msg = false;
ASSERT_NO_ERRNO(NetlinkResponse(
nlsk_send,
[&](const struct nlmsghdr* hdr) {
EXPECT_EQ(NLMSG_ERROR, hdr->nlmsg_type);
EXPECT_EQ(hdr->nlmsg_seq, kSeq);
EXPECT_GE(hdr->nlmsg_len, sizeof(*hdr) + sizeof(struct nlmsgerr));
const struct nlmsgerr* msg =
reinterpret_cast<const struct nlmsgerr*>(NLMSG_DATA(hdr));
ASSERT_EQ(msg->error, 0);
EXPECT_EQ(hdr->nlmsg_pid,
ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(nlsk_send.get())));
got_msg = true;
},
/*expect_nlmsgerr=*/true));
ASSERT_TRUE(got_msg) << "Did not get a response from the kernel.";
// Now that we know the kernel has processed the message, setup a cleanup.
auto restore_mtu = Cleanup([&]() {
MtuRequest mtu_request = GetMtuRequest(link, RTM_SETLINK, link.mtu);
ASSERT_NO_ERRNO(NetlinkRequestAckOrError(nlsk_send, kSeq, &mtu_request,
sizeof(mtu_request)));
});
// Verify that the RTM_SETLINK we sent was also received by the RTMGRP_LINK
// subscriber. We also expect an RTM_NEWLINK message from the kernel. Hence
// the two attempts.
bool got_user_msg = false;
for (int i = 0; i < 2; ++i) {
struct pollfd pfd2 = {.fd = nlsk_recv.get(), .events = POLLIN};
if (RetryEINTR(poll)(&pfd2, 1, kPollTimeoutMs) < 1) break;
char control[CMSG_SPACE(sizeof(struct ucred))] = {};
constexpr size_t kBufferSize = 4096;
std::vector<char> buf(kBufferSize);
iov = {
.iov_base = buf.data(),
.iov_len = buf.size(),
};
struct sockaddr_nl from_addr = {};
msg = {
.msg_name = &from_addr,
.msg_namelen = sizeof(from_addr),
.msg_iov = &iov,
.msg_iovlen = 1,
.msg_control = control,
.msg_controllen = sizeof(control),
};
int len;
ASSERT_THAT(len = RetryEINTR(recvmsg)(nlsk_recv.get(), &msg, 0),
SyscallSucceeds());
ASSERT_NE(msg.msg_flags & MSG_TRUNC,
MSG_TRUNC); // The buf we gave was big.
struct nlmsghdr* hdr = reinterpret_cast<struct nlmsghdr*>(buf.data());
for (; NLMSG_OK(hdr, len); hdr = NLMSG_NEXT(hdr, len)) {
ASSERT_TRUE(NLMSG_OK(hdr, len));
// Ignore the kernel's message.
if (hdr->nlmsg_type == RTM_NEWLINK) {
EXPECT_EQ(from_addr.nl_pid, 0);
continue;
}
ASSERT_EQ(hdr->nlmsg_type, RTM_SETLINK);
ASSERT_EQ(hdr->nlmsg_pid, getpid());
EXPECT_EQ(from_addr.nl_family, AF_NETLINK);
EXPECT_EQ(from_addr.nl_pid, nlsk_send_addr.nl_pid);
struct ucred creds;
struct cmsghdr* cmsg = CMSG_FIRSTHDR(&msg);
ASSERT_NE(cmsg, nullptr);
ASSERT_EQ(cmsg->cmsg_len, CMSG_LEN(sizeof(creds)));
ASSERT_EQ(cmsg->cmsg_level, SOL_SOCKET);
ASSERT_EQ(cmsg->cmsg_type, SCM_CREDENTIALS);
memcpy(&creds, CMSG_DATA(cmsg), sizeof(creds));
EXPECT_EQ(creds.pid, getpid());
got_user_msg = true;
}
}
EXPECT_TRUE(got_user_msg) << "RTMGRP_LINK subscriber did not receive the "
"expected RTM_SETLINK message.";
}
TEST(NetlinkRouteTest, LinkMulticastGroupCapNetAdmin) {
SKIP_IF(IsRunningWithHostinet());
// TODO(gvisor.dev/issue/4595): enable cooperative save tests.
const DisableSave ds;
AutoCapability cap(CAP_NET_ADMIN, false);
FileDescriptor nlsk =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
struct sockaddr_nl send_addr = {
.nl_family = AF_NETLINK,
.nl_pid = 0,
.nl_groups = RTMGRP_LINK,
};
const Link link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink());
MtuRequest mtu_request = GetMtuRequest(link, RTM_SETLINK, link.mtu + 10);
struct iovec iov = {
.iov_base = &mtu_request,
.iov_len = sizeof(mtu_request),
};
struct msghdr msg = {
.msg_name = &send_addr,
.msg_namelen = sizeof(send_addr),
.msg_iov = &iov,
.msg_iovlen = 1,
};
EXPECT_THAT(RetryEINTR(sendmsg)(nlsk.get(), &msg, 0),
SyscallFailsWithErrno(EPERM));
// But joining a multicast group should not require CAP_NET_ADMIN.
struct sockaddr_nl mcast_addr = {
.nl_family = AF_NETLINK,
.nl_groups = RTMGRP_LINK,
};
ASSERT_NO_ERRNO(NetlinkBoundSocket(NETLINK_ROUTE, &mcast_addr));
}
TEST(NetlinkRouteTest, LinkMulticastGroupNoSelfBroadcast) {
SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN)));
SKIP_IF(IsRunningWithHostinet());
// TODO(gvisor.dev/issue/4595): enable cooperative save tests.
const DisableSave ds;
const struct sockaddr_nl linkgrp_addr = {
.nl_family = AF_NETLINK,
.nl_groups = RTMGRP_LINK,
};
FileDescriptor nlsk_send = ASSERT_NO_ERRNO_AND_VALUE(
NetlinkBoundSocket(NETLINK_ROUTE, &linkgrp_addr));
FileDescriptor nlsk_recv = ASSERT_NO_ERRNO_AND_VALUE(
NetlinkBoundSocket(NETLINK_ROUTE, &linkgrp_addr));
// Create an RTM_GETLINK request.
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
};
request req = {};
req.hdr = {
.nlmsg_len = sizeof(req),
.nlmsg_type = RTM_GETLINK,
.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP,
.nlmsg_seq = kSeq,
};
req.ifm.ifi_family = AF_UNSPEC;
// And send it to both the kernel and the RTMGRP_LINK group.
struct iovec iov = {
.iov_base = &req,
.iov_len = sizeof(req),
};
struct sockaddr_nl send_addr = linkgrp_addr;
ASSERT_EQ(send_addr.nl_pid, 0);
struct msghdr msg = {
.msg_name = &send_addr,
.msg_namelen = sizeof(send_addr),
.msg_iov = &iov,
.msg_iovlen = 1,
};
ASSERT_THAT(RetryEINTR(sendmsg)(nlsk_send.get(), &msg, 0), SyscallSucceeds());
// The sender should only receive the kernel's response, not its own
// broadcast.
constexpr int kPollTimeoutMs = 500;
bool got_response = false;
for (int i = 0; i < 2; ++i) {
struct pollfd pfd = {.fd = nlsk_send.get(), .events = POLLIN};
if (RetryEINTR(poll)(&pfd, 1, kPollTimeoutMs) != 1) {
break;
}
ASSERT_NO_ERRNO(NetlinkResponse(
nlsk_send,
[&](const struct nlmsghdr* hdr) {
ASSERT_NE(hdr->nlmsg_type, RTM_GETLINK); // No self-broadcast.
if (hdr->nlmsg_type == RTM_NEWLINK) {
got_response = true;
}
},
/*expect_nlmsgerr=*/false));
}
EXPECT_TRUE(got_response) << "Did not get a response from the kernel.";
// Whereas the bystander RTMGRP_LINK subscriber should get the broadcast.
bool got_broadcast = false;
{
struct pollfd pfd = {.fd = nlsk_recv.get(), .events = POLLIN};
ASSERT_EQ(RetryEINTR(poll)(&pfd, 1, kPollTimeoutMs), 1)
<< "RTMGRP_LINK subscriber did not get the broadcast.";
ASSERT_NO_ERRNO(NetlinkResponse(
nlsk_recv,
[&](const struct nlmsghdr* hdr) {
ASSERT_EQ(hdr->nlmsg_type, RTM_GETLINK);
got_broadcast = true;
},
/*expect_nlmsgerr=*/false));
}
EXPECT_TRUE(got_broadcast) << "Did not get the broadcast.";
}
TEST(NetlinkRouteTest, NonGetRequestsRequireCapNetAdmin) {
SKIP_IF(IsRunningWithHostinet());
AutoCapability cap(CAP_NET_ADMIN, false);
FileDescriptor nlsk =
ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE));
const int fd_raw = nlsk.get();
struct request {
struct nlmsghdr hdr;
struct ifinfomsg ifm;
};
request req = {};
req.hdr = {
.nlmsg_len = sizeof(req),
.nlmsg_type = RTM_NEWLINK,
.nlmsg_flags = NLM_F_REQUEST,
.nlmsg_seq = kSeq,
};
req.ifm.ifi_family = AF_UNSPEC;
struct iovec iov = {
.iov_base = &req,
.iov_len = sizeof(req),
};
struct msghdr msg = {
.msg_iov = &iov,
.msg_iovlen = 1,
};
ASSERT_THAT(RetryEINTR(sendmsg)(nlsk.get(), &msg, 0), SyscallSucceeds());
bool got_eperm = false;
ASSERT_NO_ERRNO(NetlinkResponse(
nlsk,
[&](const struct nlmsghdr* hdr) {
EXPECT_EQ(NLMSG_ERROR, hdr->nlmsg_type);
EXPECT_EQ(hdr->nlmsg_seq, kSeq);
EXPECT_GE(hdr->nlmsg_len, sizeof(*hdr) + sizeof(struct nlmsgerr));
const struct nlmsgerr* msg =
reinterpret_cast<const struct nlmsgerr*>(NLMSG_DATA(hdr));
ASSERT_EQ(msg->error, -EPERM);
got_eperm = true;
},
/*expect_nlmsgerr=*/true));
ASSERT_TRUE(got_eperm) << "Did not get an EPERM from the kernel.";
// Try to beat the EPERM with an unshare().
ASSERT_THAT(InForkedProcess([&] {
// Enter a new user namespace in which we'd have all caps.
TEST_PCHECK(syscall(SYS_unshare, CLONE_NEWUSER) == 0);
// And then send the RTM_NEWLINK again.
TEST_CHECK_SUCCESS(syscall(SYS_sendmsg, fd_raw, &msg, 0));
_exit(0);
}),
IsPosixErrorOkAndHolds(0));
// Even with the unshare, the RTM_NEWLINK should still be rejected.
got_eperm = false;
ASSERT_NO_ERRNO(NetlinkResponse(
nlsk,
[&](const struct nlmsghdr* hdr) {
EXPECT_EQ(NLMSG_ERROR, hdr->nlmsg_type);
EXPECT_EQ(hdr->nlmsg_seq, kSeq);
EXPECT_GE(hdr->nlmsg_len, sizeof(*hdr) + sizeof(struct nlmsgerr));
const struct nlmsgerr* msg =
reinterpret_cast<const struct nlmsgerr*>(NLMSG_DATA(hdr));
ASSERT_EQ(msg->error, -EPERM);
got_eperm = true;
},
/*expect_nlmsgerr=*/true));
EXPECT_TRUE(got_eperm) << "Did not get an EPERM from the kernel.";
}
} // namespace
} // namespace testing
} // namespace gvisor
Reference in New Issue View Git Blame Copy Permalink