diff options
Diffstat (limited to 'test')
22 files changed, 1522 insertions, 149 deletions
diff --git a/test/e2e/BUILD b/test/e2e/BUILD index 99442cffb..4fe03a220 100644 --- a/test/e2e/BUILD +++ b/test/e2e/BUILD @@ -19,7 +19,9 @@ go_test( visibility = ["//:sandbox"], deps = [ "//pkg/abi/linux", + "//pkg/bits", "//runsc/dockerutil", + "//runsc/specutils", "//runsc/testutil", ], ) diff --git a/test/e2e/exec_test.go b/test/e2e/exec_test.go index 7238c2afe..88d26e865 100644 --- a/test/e2e/exec_test.go +++ b/test/e2e/exec_test.go @@ -30,14 +30,17 @@ import ( "time" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/bits" "gvisor.dev/gvisor/runsc/dockerutil" + "gvisor.dev/gvisor/runsc/specutils" ) +// Test that exec uses the exact same capability set as the container. func TestExecCapabilities(t *testing.T) { if err := dockerutil.Pull("alpine"); err != nil { t.Fatalf("docker pull failed: %v", err) } - d := dockerutil.MakeDocker("exec-test") + d := dockerutil.MakeDocker("exec-capabilities-test") // Start the container. if err := d.Run("alpine", "sh", "-c", "cat /proc/self/status; sleep 100"); err != nil { @@ -52,27 +55,59 @@ func TestExecCapabilities(t *testing.T) { if len(matches) != 2 { t.Fatalf("There should be a match for the whole line and the capability bitmask") } - capString := matches[1] - t.Log("Root capabilities:", capString) + want := fmt.Sprintf("CapEff:\t%s\n", matches[1]) + t.Log("Root capabilities:", want) - // CAP_NET_RAW was in the capability set for the container, but was - // removed. However, `exec` does not remove it. Verify that it's not - // set in the container, then re-add it for comparison. - caps, err := strconv.ParseUint(capString, 16, 64) + // Now check that exec'd process capabilities match the root. + got, err := d.Exec("grep", "CapEff:", "/proc/self/status") if err != nil { - t.Fatalf("failed to convert capabilities %q: %v", capString, err) + t.Fatalf("docker exec failed: %v", err) } - if caps&(1<<uint64(linux.CAP_NET_RAW)) != 0 { - t.Fatalf("CAP_NET_RAW should be filtered, but is set in the container: %x", caps) + t.Logf("CapEff: %v", got) + if got != want { + t.Errorf("wrong capabilities, got: %q, want: %q", got, want) } - caps |= 1 << uint64(linux.CAP_NET_RAW) - want := fmt.Sprintf("CapEff:\t%016x\n", caps) +} - // Now check that exec'd process capabilities match the root. - got, err := d.Exec("grep", "CapEff:", "/proc/self/status") +// Test that 'exec --privileged' adds all capabilities, except for CAP_NET_RAW +// which is removed from the container when --net-raw=false. +func TestExecPrivileged(t *testing.T) { + if err := dockerutil.Pull("alpine"); err != nil { + t.Fatalf("docker pull failed: %v", err) + } + d := dockerutil.MakeDocker("exec-privileged-test") + + // Start the container with all capabilities dropped. + if err := d.Run("--cap-drop=all", "alpine", "sh", "-c", "cat /proc/self/status; sleep 100"); err != nil { + t.Fatalf("docker run failed: %v", err) + } + defer d.CleanUp() + + // Check that all capabilities where dropped from container. + matches, err := d.WaitForOutputSubmatch("CapEff:\t([0-9a-f]+)\n", 5*time.Second) + if err != nil { + t.Fatalf("WaitForOutputSubmatch() timeout: %v", err) + } + if len(matches) != 2 { + t.Fatalf("There should be a match for the whole line and the capability bitmask") + } + containerCaps, err := strconv.ParseUint(matches[1], 16, 64) + if err != nil { + t.Fatalf("failed to convert capabilities %q: %v", matches[1], err) + } + t.Logf("Container capabilities: %#x", containerCaps) + if containerCaps != 0 { + t.Fatalf("Container should have no capabilities: %x", containerCaps) + } + + // Check that 'exec --privileged' adds all capabilities, except + // for CAP_NET_RAW. + got, err := d.ExecWithFlags([]string{"--privileged"}, "grep", "CapEff:", "/proc/self/status") if err != nil { t.Fatalf("docker exec failed: %v", err) } + t.Logf("Exec CapEff: %v", got) + want := fmt.Sprintf("CapEff:\t%016x\n", specutils.AllCapabilitiesUint64()&^bits.MaskOf64(int(linux.CAP_NET_RAW))) if got != want { t.Errorf("wrong capabilities, got: %q, want: %q", got, want) } @@ -184,7 +219,7 @@ func TestExecEnvHasHome(t *testing.T) { if err := dockerutil.Pull("alpine"); err != nil { t.Fatalf("docker pull failed: %v", err) } - d := dockerutil.MakeDocker("exec-env-test") + d := dockerutil.MakeDocker("exec-env-home-test") // We will check that HOME is set for root user, and also for a new // non-root user we will create. diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 28b23ce58..d5a2b7725 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -2464,6 +2464,63 @@ cc_binary( ) cc_binary( + name = "socket_bind_to_device_test", + testonly = 1, + srcs = [ + "socket_bind_to_device.cc", + ], + linkstatic = 1, + deps = [ + ":ip_socket_test_util", + ":socket_bind_to_device_util", + ":socket_test_util", + "//test/util:capability_util", + "//test/util:test_main", + "//test/util:test_util", + "//test/util:thread_util", + "@com_google_googletest//:gtest", + ], +) + +cc_binary( + name = "socket_bind_to_device_sequence_test", + testonly = 1, + srcs = [ + "socket_bind_to_device_sequence.cc", + ], + linkstatic = 1, + deps = [ + ":ip_socket_test_util", + ":socket_bind_to_device_util", + ":socket_test_util", + "//test/util:capability_util", + "//test/util:test_main", + "//test/util:test_util", + "//test/util:thread_util", + "@com_google_googletest//:gtest", + ], +) + +cc_binary( + name = "socket_bind_to_device_distribution_test", + testonly = 1, + srcs = [ + "socket_bind_to_device_distribution.cc", + ], + linkstatic = 1, + deps = [ + ":ip_socket_test_util", + ":socket_bind_to_device_util", + ":socket_test_util", + "//test/util:capability_util", + "//test/util:test_main", + "//test/util:test_util", + "//test/util:thread_util", + "@com_google_googletest//:gtest", + ], +) + +cc_binary( name = "socket_ip_udp_loopback_non_blocking_test", testonly = 1, srcs = [ @@ -2740,6 +2797,23 @@ cc_library( alwayslink = 1, ) +cc_library( + name = "socket_bind_to_device_util", + testonly = 1, + srcs = [ + "socket_bind_to_device_util.cc", + ], + hdrs = [ + "socket_bind_to_device_util.h", + ], + deps = [ + "//test/util:test_util", + "@com_google_absl//absl/memory", + "@com_google_absl//absl/strings", + ], + alwayslink = 1, +) + cc_binary( name = "socket_stream_local_test", testonly = 1, @@ -3163,8 +3237,6 @@ cc_binary( testonly = 1, srcs = ["timers.cc"], linkstatic = 1, - # FIXME(b/136599201) - tags = ["flaky"], deps = [ "//test/util:cleanup", "//test/util:logging", @@ -3253,6 +3325,7 @@ cc_binary( "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", + "//test/util:uid_util", "@com_google_absl//absl/flags:flag", "@com_google_absl//absl/strings", "@com_google_googletest//:gtest", diff --git a/test/syscalls/linux/clock_nanosleep.cc b/test/syscalls/linux/clock_nanosleep.cc index 52a69d230..b55cddc52 100644 --- a/test/syscalls/linux/clock_nanosleep.cc +++ b/test/syscalls/linux/clock_nanosleep.cc @@ -43,7 +43,7 @@ int sys_clock_nanosleep(clockid_t clkid, int flags, PosixErrorOr<absl::Time> GetTime(clockid_t clk) { struct timespec ts = {}; - int rc = clock_gettime(clk, &ts); + const int rc = clock_gettime(clk, &ts); MaybeSave(); if (rc < 0) { return PosixError(errno, "clock_gettime"); @@ -67,31 +67,32 @@ TEST_P(WallClockNanosleepTest, InvalidValues) { } TEST_P(WallClockNanosleepTest, SleepOneSecond) { - absl::Duration const duration = absl::Seconds(1); - struct timespec dur = absl::ToTimespec(duration); + constexpr absl::Duration kSleepDuration = absl::Seconds(1); + struct timespec duration = absl::ToTimespec(kSleepDuration); - absl::Time const before = ASSERT_NO_ERRNO_AND_VALUE(GetTime(GetParam())); - EXPECT_THAT(RetryEINTR(sys_clock_nanosleep)(GetParam(), 0, &dur, &dur), - SyscallSucceeds()); - absl::Time const after = ASSERT_NO_ERRNO_AND_VALUE(GetTime(GetParam())); + const absl::Time before = ASSERT_NO_ERRNO_AND_VALUE(GetTime(GetParam())); + EXPECT_THAT( + RetryEINTR(sys_clock_nanosleep)(GetParam(), 0, &duration, &duration), + SyscallSucceeds()); + const absl::Time after = ASSERT_NO_ERRNO_AND_VALUE(GetTime(GetParam())); - EXPECT_GE(after - before, duration); + EXPECT_GE(after - before, kSleepDuration); } TEST_P(WallClockNanosleepTest, InterruptedNanosleep) { - absl::Duration const duration = absl::Seconds(60); - struct timespec dur = absl::ToTimespec(duration); + constexpr absl::Duration kSleepDuration = absl::Seconds(60); + struct timespec duration = absl::ToTimespec(kSleepDuration); // Install no-op signal handler for SIGALRM. struct sigaction sa = {}; sigfillset(&sa.sa_mask); sa.sa_handler = +[](int signo) {}; - auto const cleanup_sa = + const auto cleanup_sa = ASSERT_NO_ERRNO_AND_VALUE(ScopedSigaction(SIGALRM, sa)); // Measure time since setting the alarm, since the alarm will interrupt the // sleep and hence determine how long we sleep. - absl::Time const before = ASSERT_NO_ERRNO_AND_VALUE(GetTime(GetParam())); + const absl::Time before = ASSERT_NO_ERRNO_AND_VALUE(GetTime(GetParam())); // Set an alarm to go off while sleeping. struct itimerval timer = {}; @@ -99,26 +100,51 @@ TEST_P(WallClockNanosleepTest, InterruptedNanosleep) { timer.it_value.tv_usec = 0; timer.it_interval.tv_sec = 1; timer.it_interval.tv_usec = 0; - auto const cleanup = + const auto cleanup = ASSERT_NO_ERRNO_AND_VALUE(ScopedItimer(ITIMER_REAL, timer)); - EXPECT_THAT(sys_clock_nanosleep(GetParam(), 0, &dur, &dur), + EXPECT_THAT(sys_clock_nanosleep(GetParam(), 0, &duration, &duration), SyscallFailsWithErrno(EINTR)); - absl::Time const after = ASSERT_NO_ERRNO_AND_VALUE(GetTime(GetParam())); + const absl::Time after = ASSERT_NO_ERRNO_AND_VALUE(GetTime(GetParam())); - absl::Duration const remaining = absl::DurationFromTimespec(dur); - EXPECT_GE(after - before + remaining, duration); + // Remaining time updated. + const absl::Duration remaining = absl::DurationFromTimespec(duration); + EXPECT_GE(after - before + remaining, kSleepDuration); +} + +// Remaining time is *not* updated if nanosleep completes uninterrupted. +TEST_P(WallClockNanosleepTest, UninterruptedNanosleep) { + constexpr absl::Duration kSleepDuration = absl::Milliseconds(10); + const struct timespec duration = absl::ToTimespec(kSleepDuration); + + while (true) { + constexpr int kRemainingMagic = 42; + struct timespec remaining; + remaining.tv_sec = kRemainingMagic; + remaining.tv_nsec = kRemainingMagic; + + int ret = sys_clock_nanosleep(GetParam(), 0, &duration, &remaining); + if (ret == EINTR) { + // Retry from beginning. We want a single uninterrupted call. + continue; + } + + EXPECT_THAT(ret, SyscallSucceeds()); + EXPECT_EQ(remaining.tv_sec, kRemainingMagic); + EXPECT_EQ(remaining.tv_nsec, kRemainingMagic); + break; + } } TEST_P(WallClockNanosleepTest, SleepUntil) { - absl::Time const now = ASSERT_NO_ERRNO_AND_VALUE(GetTime(GetParam())); - absl::Time const until = now + absl::Seconds(2); - struct timespec ts = absl::ToTimespec(until); + const absl::Time now = ASSERT_NO_ERRNO_AND_VALUE(GetTime(GetParam())); + const absl::Time until = now + absl::Seconds(2); + const struct timespec ts = absl::ToTimespec(until); EXPECT_THAT( RetryEINTR(sys_clock_nanosleep)(GetParam(), TIMER_ABSTIME, &ts, nullptr), SyscallSucceeds()); - absl::Time const after = ASSERT_NO_ERRNO_AND_VALUE(GetTime(GetParam())); + const absl::Time after = ASSERT_NO_ERRNO_AND_VALUE(GetTime(GetParam())); EXPECT_GE(after, until); } @@ -127,8 +153,8 @@ INSTANTIATE_TEST_SUITE_P(Sleepers, WallClockNanosleepTest, ::testing::Values(CLOCK_REALTIME, CLOCK_MONOTONIC)); TEST(ClockNanosleepProcessTest, SleepFiveSeconds) { - absl::Duration const kDuration = absl::Seconds(5); - struct timespec dur = absl::ToTimespec(kDuration); + const absl::Duration kSleepDuration = absl::Seconds(5); + struct timespec duration = absl::ToTimespec(kSleepDuration); // Ensure that CLOCK_PROCESS_CPUTIME_ID advances. std::atomic<bool> done(false); @@ -136,16 +162,16 @@ TEST(ClockNanosleepProcessTest, SleepFiveSeconds) { while (!done.load()) { } }); - auto const cleanup_done = Cleanup([&] { done.store(true); }); + const auto cleanup_done = Cleanup([&] { done.store(true); }); - absl::Time const before = + const absl::Time before = ASSERT_NO_ERRNO_AND_VALUE(GetTime(CLOCK_PROCESS_CPUTIME_ID)); - EXPECT_THAT( - RetryEINTR(sys_clock_nanosleep)(CLOCK_PROCESS_CPUTIME_ID, 0, &dur, &dur), - SyscallSucceeds()); - absl::Time const after = + EXPECT_THAT(RetryEINTR(sys_clock_nanosleep)(CLOCK_PROCESS_CPUTIME_ID, 0, + &duration, &duration), + SyscallSucceeds()); + const absl::Time after = ASSERT_NO_ERRNO_AND_VALUE(GetTime(CLOCK_PROCESS_CPUTIME_ID)); - EXPECT_GE(after - before, kDuration); + EXPECT_GE(after - before, kSleepDuration); } } // namespace diff --git a/test/syscalls/linux/exec_binary.cc b/test/syscalls/linux/exec_binary.cc index 91b55015c..68af882bb 100644 --- a/test/syscalls/linux/exec_binary.cc +++ b/test/syscalls/linux/exec_binary.cc @@ -401,12 +401,17 @@ TEST(ElfTest, DataSegment) { }))); } -// Additonal pages beyond filesz are always RW. +// Additonal pages beyond filesz honor (only) execute protections. // -// N.B. Linux uses set_brk -> vm_brk to additional pages beyond filesz (even -// though start_brk itself will always be beyond memsz). As a result, the -// segment permissions don't apply; the mapping is always RW. +// N.B. Linux changed this in 4.11 (16e72e9b30986 "powerpc: do not make the +// entire heap executable"). Previously, extra pages were always RW. TEST(ElfTest, ExtraMemPages) { + // gVisor has the newer behavior. + if (!IsRunningOnGvisor()) { + auto version = ASSERT_NO_ERRNO_AND_VALUE(GetKernelVersion()); + SKIP_IF(version.major < 4 || (version.major == 4 && version.minor < 11)); + } + ElfBinary<64> elf = StandardElf(); // Create a standard ELF, but extend to 1.5 pages. The second page will be the @@ -415,7 +420,7 @@ TEST(ElfTest, ExtraMemPages) { decltype(elf)::ElfPhdr phdr = {}; phdr.p_type = PT_LOAD; - // RWX segment. The extra anon page will be RW anyways. + // RWX segment. The extra anon page will also be RWX. // // N.B. Linux uses clear_user to clear the end of the file-mapped page, which // respects the mapping protections. Thus if we map this RO with memsz > @@ -454,7 +459,7 @@ TEST(ElfTest, ExtraMemPages) { {0x41000, 0x42000, true, true, true, true, kPageSize, 0, 0, 0, file.path().c_str()}, // extra page from anon. - {0x42000, 0x43000, true, true, false, true, 0, 0, 0, 0, ""}, + {0x42000, 0x43000, true, true, true, true, 0, 0, 0, 0, ""}, }))); } @@ -469,7 +474,7 @@ TEST(ElfTest, AnonOnlySegment) { phdr.p_offset = 0; phdr.p_vaddr = 0x41000; phdr.p_filesz = 0; - phdr.p_memsz = kPageSize - 0xe8; + phdr.p_memsz = kPageSize; elf.phdrs.push_back(phdr); elf.UpdateOffsets(); diff --git a/test/syscalls/linux/packet_socket.cc b/test/syscalls/linux/packet_socket.cc index 7a3379b9e..37b4e6575 100644 --- a/test/syscalls/linux/packet_socket.cc +++ b/test/syscalls/linux/packet_socket.cc @@ -83,9 +83,15 @@ void SendUDPMessage(int sock) { // Send an IP packet and make sure ETH_P_<something else> doesn't pick it up. TEST(BasicCookedPacketTest, WrongType) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + // (b/129292371): Remove once we support packet sockets. SKIP_IF(IsRunningOnGvisor()); + if (!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) { + ASSERT_THAT(socket(AF_PACKET, SOCK_DGRAM, ETH_P_PUP), + SyscallFailsWithErrno(EPERM)); + GTEST_SKIP(); + } + FileDescriptor sock = ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_PACKET, SOCK_DGRAM, ETH_P_PUP)); @@ -118,18 +124,27 @@ class CookedPacketTest : public ::testing::TestWithParam<int> { }; void CookedPacketTest::SetUp() { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + // (b/129292371): Remove once we support packet sockets. SKIP_IF(IsRunningOnGvisor()); + if (!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) { + ASSERT_THAT(socket(AF_PACKET, SOCK_DGRAM, htons(GetParam())), + SyscallFailsWithErrno(EPERM)); + GTEST_SKIP(); + } + ASSERT_THAT(socket_ = socket(AF_PACKET, SOCK_DGRAM, htons(GetParam())), SyscallSucceeds()); } void CookedPacketTest::TearDown() { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + // (b/129292371): Remove once we support packet sockets. SKIP_IF(IsRunningOnGvisor()); - EXPECT_THAT(close(socket_), SyscallSucceeds()); + // TearDown will be run even if we skip the test. + if (ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) { + EXPECT_THAT(close(socket_), SyscallSucceeds()); + } } int CookedPacketTest::GetLoopbackIndex() { @@ -142,9 +157,6 @@ int CookedPacketTest::GetLoopbackIndex() { // Receive via a packet socket. TEST_P(CookedPacketTest, Receive) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - SKIP_IF(IsRunningOnGvisor()); - // Let's use a simple IP payload: a UDP datagram. FileDescriptor udp_sock = ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); @@ -201,9 +213,6 @@ TEST_P(CookedPacketTest, Receive) { // Send via a packet socket. TEST_P(CookedPacketTest, Send) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - SKIP_IF(IsRunningOnGvisor()); - // Let's send a UDP packet and receive it using a regular UDP socket. FileDescriptor udp_sock = ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); diff --git a/test/syscalls/linux/packet_socket_raw.cc b/test/syscalls/linux/packet_socket_raw.cc index 9e96460ee..6491453b6 100644 --- a/test/syscalls/linux/packet_socket_raw.cc +++ b/test/syscalls/linux/packet_socket_raw.cc @@ -97,9 +97,15 @@ class RawPacketTest : public ::testing::TestWithParam<int> { }; void RawPacketTest::SetUp() { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + // (b/129292371): Remove once we support packet sockets. SKIP_IF(IsRunningOnGvisor()); + if (!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) { + ASSERT_THAT(socket(AF_PACKET, SOCK_RAW, htons(GetParam())), + SyscallFailsWithErrno(EPERM)); + GTEST_SKIP(); + } + if (!IsRunningOnGvisor()) { FileDescriptor acceptLocal = ASSERT_NO_ERRNO_AND_VALUE( Open("/proc/sys/net/ipv4/conf/lo/accept_local", O_RDONLY)); @@ -119,10 +125,13 @@ void RawPacketTest::SetUp() { } void RawPacketTest::TearDown() { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + // (b/129292371): Remove once we support packet sockets. SKIP_IF(IsRunningOnGvisor()); - EXPECT_THAT(close(socket_), SyscallSucceeds()); + // TearDown will be run even if we skip the test. + if (ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) { + EXPECT_THAT(close(socket_), SyscallSucceeds()); + } } int RawPacketTest::GetLoopbackIndex() { @@ -135,9 +144,6 @@ int RawPacketTest::GetLoopbackIndex() { // Receive via a packet socket. TEST_P(RawPacketTest, Receive) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - SKIP_IF(IsRunningOnGvisor()); - // Let's use a simple IP payload: a UDP datagram. FileDescriptor udp_sock = ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); @@ -208,9 +214,6 @@ TEST_P(RawPacketTest, Receive) { // Send via a packet socket. TEST_P(RawPacketTest, Send) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - SKIP_IF(IsRunningOnGvisor()); - // Let's send a UDP packet and receive it using a regular UDP socket. FileDescriptor udp_sock = ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); diff --git a/test/syscalls/linux/pty.cc b/test/syscalls/linux/pty.cc index bf32efe1e..99a0df235 100644 --- a/test/syscalls/linux/pty.cc +++ b/test/syscalls/linux/pty.cc @@ -1527,27 +1527,36 @@ TEST_F(JobControlTest, SetForegroundProcessGroupEmptyProcessGroup) { TEST_F(JobControlTest, SetForegroundProcessGroupDifferentSession) { ASSERT_THAT(ioctl(slave_.get(), TIOCSCTTY, 0), SyscallSucceeds()); + int sync_setsid[2]; + int sync_exit[2]; + ASSERT_THAT(pipe(sync_setsid), SyscallSucceeds()); + ASSERT_THAT(pipe(sync_exit), SyscallSucceeds()); + // Create a new process and put it in a new session. pid_t child = fork(); if (!child) { TEST_PCHECK(setsid() >= 0); // Tell the parent we're in a new session. - TEST_PCHECK(!raise(SIGSTOP)); - TEST_PCHECK(!pause()); - _exit(1); + char c = 'c'; + TEST_PCHECK(WriteFd(sync_setsid[1], &c, 1) == 1); + TEST_PCHECK(ReadFd(sync_exit[0], &c, 1) == 1); + _exit(0); } // Wait for the child to tell us it's in a new session. - int wstatus; - EXPECT_THAT(waitpid(child, &wstatus, WUNTRACED), - SyscallSucceedsWithValue(child)); - EXPECT_TRUE(WSTOPSIG(wstatus)); + char c = 'c'; + ASSERT_THAT(ReadFd(sync_setsid[0], &c, 1), SyscallSucceedsWithValue(1)); // Child is in a new session, so we can't make it the foregroup process group. EXPECT_THAT(ioctl(slave_.get(), TIOCSPGRP, &child), SyscallFailsWithErrno(EPERM)); - EXPECT_THAT(kill(child, SIGKILL), SyscallSucceeds()); + EXPECT_THAT(WriteFd(sync_exit[1], &c, 1), SyscallSucceedsWithValue(1)); + + int wstatus; + EXPECT_THAT(waitpid(child, &wstatus, 0), SyscallSucceedsWithValue(child)); + EXPECT_TRUE(WIFEXITED(wstatus)); + EXPECT_EQ(WEXITSTATUS(wstatus), 0); } // Verify that we don't hang when creating a new session from an orphaned diff --git a/test/syscalls/linux/raw_socket_hdrincl.cc b/test/syscalls/linux/raw_socket_hdrincl.cc index a070817eb..0a27506aa 100644 --- a/test/syscalls/linux/raw_socket_hdrincl.cc +++ b/test/syscalls/linux/raw_socket_hdrincl.cc @@ -63,7 +63,11 @@ class RawHDRINCL : public ::testing::Test { }; void RawHDRINCL::SetUp() { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + if (!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) { + ASSERT_THAT(socket(AF_INET, SOCK_RAW, IPPROTO_RAW), + SyscallFailsWithErrno(EPERM)); + GTEST_SKIP(); + } ASSERT_THAT(socket_ = socket(AF_INET, SOCK_RAW, IPPROTO_RAW), SyscallSucceeds()); @@ -76,9 +80,10 @@ void RawHDRINCL::SetUp() { } void RawHDRINCL::TearDown() { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - EXPECT_THAT(close(socket_), SyscallSucceeds()); + // TearDown will be run even if we skip the test. + if (ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) { + EXPECT_THAT(close(socket_), SyscallSucceeds()); + } } struct iphdr RawHDRINCL::LoopbackHeader() { @@ -123,8 +128,6 @@ bool RawHDRINCL::FillPacket(char* buf, size_t buf_size, int port, // We should be able to create multiple IPPROTO_RAW sockets. RawHDRINCL::Setup // creates the first one, so we only have to create one more here. TEST_F(RawHDRINCL, MultipleCreation) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - int s2; ASSERT_THAT(s2 = socket(AF_INET, SOCK_RAW, IPPROTO_RAW), SyscallSucceeds()); @@ -133,23 +136,17 @@ TEST_F(RawHDRINCL, MultipleCreation) { // Test that shutting down an unconnected socket fails. TEST_F(RawHDRINCL, FailShutdownWithoutConnect) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - ASSERT_THAT(shutdown(socket_, SHUT_WR), SyscallFailsWithErrno(ENOTCONN)); ASSERT_THAT(shutdown(socket_, SHUT_RD), SyscallFailsWithErrno(ENOTCONN)); } // Test that listen() fails. TEST_F(RawHDRINCL, FailListen) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - ASSERT_THAT(listen(socket_, 1), SyscallFailsWithErrno(ENOTSUP)); } // Test that accept() fails. TEST_F(RawHDRINCL, FailAccept) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - struct sockaddr saddr; socklen_t addrlen; ASSERT_THAT(accept(socket_, &saddr, &addrlen), @@ -158,8 +155,6 @@ TEST_F(RawHDRINCL, FailAccept) { // Test that the socket is writable immediately. TEST_F(RawHDRINCL, PollWritableImmediately) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - struct pollfd pfd = {}; pfd.fd = socket_; pfd.events = POLLOUT; @@ -168,8 +163,6 @@ TEST_F(RawHDRINCL, PollWritableImmediately) { // Test that the socket isn't readable. TEST_F(RawHDRINCL, NotReadable) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - // Try to receive data with MSG_DONTWAIT, which returns immediately if there's // nothing to be read. char buf[117]; @@ -179,16 +172,12 @@ TEST_F(RawHDRINCL, NotReadable) { // Test that we can connect() to a valid IP (loopback). TEST_F(RawHDRINCL, ConnectToLoopback) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - ASSERT_THAT(connect(socket_, reinterpret_cast<struct sockaddr*>(&addr_), sizeof(addr_)), SyscallSucceeds()); } TEST_F(RawHDRINCL, SendWithoutConnectSucceeds) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - struct iphdr hdr = LoopbackHeader(); ASSERT_THAT(send(socket_, &hdr, sizeof(hdr), 0), SyscallSucceedsWithValue(sizeof(hdr))); @@ -197,8 +186,6 @@ TEST_F(RawHDRINCL, SendWithoutConnectSucceeds) { // HDRINCL implies write-only. Verify that we can't read a packet sent to // loopback. TEST_F(RawHDRINCL, NotReadableAfterWrite) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - ASSERT_THAT(connect(socket_, reinterpret_cast<struct sockaddr*>(&addr_), sizeof(addr_)), SyscallSucceeds()); @@ -221,8 +208,6 @@ TEST_F(RawHDRINCL, NotReadableAfterWrite) { } TEST_F(RawHDRINCL, WriteTooSmall) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - ASSERT_THAT(connect(socket_, reinterpret_cast<struct sockaddr*>(&addr_), sizeof(addr_)), SyscallSucceeds()); @@ -235,8 +220,6 @@ TEST_F(RawHDRINCL, WriteTooSmall) { // Bind to localhost. TEST_F(RawHDRINCL, BindToLocalhost) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - ASSERT_THAT( bind(socket_, reinterpret_cast<struct sockaddr*>(&addr_), sizeof(addr_)), SyscallSucceeds()); @@ -244,8 +227,6 @@ TEST_F(RawHDRINCL, BindToLocalhost) { // Bind to a different address. TEST_F(RawHDRINCL, BindToInvalid) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - struct sockaddr_in bind_addr = {}; bind_addr.sin_family = AF_INET; bind_addr.sin_addr = {1}; // 1.0.0.0 - An address that we can't bind to. @@ -256,8 +237,6 @@ TEST_F(RawHDRINCL, BindToInvalid) { // Send and receive a packet. TEST_F(RawHDRINCL, SendAndReceive) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - int port = 40000; if (!IsRunningOnGvisor()) { port = static_cast<short>(ASSERT_NO_ERRNO_AND_VALUE( @@ -302,8 +281,6 @@ TEST_F(RawHDRINCL, SendAndReceive) { // Send and receive a packet with nonzero IP ID. TEST_F(RawHDRINCL, SendAndReceiveNonzeroID) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - int port = 40000; if (!IsRunningOnGvisor()) { port = static_cast<short>(ASSERT_NO_ERRNO_AND_VALUE( @@ -349,8 +326,6 @@ TEST_F(RawHDRINCL, SendAndReceiveNonzeroID) { // Send and receive a packet where the sendto address is not the same as the // provided destination. TEST_F(RawHDRINCL, SendAndReceiveDifferentAddress) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - int port = 40000; if (!IsRunningOnGvisor()) { port = static_cast<short>(ASSERT_NO_ERRNO_AND_VALUE( diff --git a/test/syscalls/linux/raw_socket_icmp.cc b/test/syscalls/linux/raw_socket_icmp.cc index 971592d7d..8bcaba6f1 100644 --- a/test/syscalls/linux/raw_socket_icmp.cc +++ b/test/syscalls/linux/raw_socket_icmp.cc @@ -77,7 +77,11 @@ class RawSocketICMPTest : public ::testing::Test { }; void RawSocketICMPTest::SetUp() { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + if (!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) { + ASSERT_THAT(socket(AF_INET, SOCK_RAW, IPPROTO_ICMP), + SyscallFailsWithErrno(EPERM)); + GTEST_SKIP(); + } ASSERT_THAT(s_ = socket(AF_INET, SOCK_RAW, IPPROTO_ICMP), SyscallSucceeds()); @@ -90,9 +94,10 @@ void RawSocketICMPTest::SetUp() { } void RawSocketICMPTest::TearDown() { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - EXPECT_THAT(close(s_), SyscallSucceeds()); + // TearDown will be run even if we skip the test. + if (ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) { + EXPECT_THAT(close(s_), SyscallSucceeds()); + } } // We'll only read an echo in this case, as the kernel won't respond to the diff --git a/test/syscalls/linux/raw_socket_ipv4.cc b/test/syscalls/linux/raw_socket_ipv4.cc index 352037c88..cde2f07c9 100644 --- a/test/syscalls/linux/raw_socket_ipv4.cc +++ b/test/syscalls/linux/raw_socket_ipv4.cc @@ -67,7 +67,11 @@ class RawSocketTest : public ::testing::TestWithParam<int> { }; void RawSocketTest::SetUp() { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + if (!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) { + ASSERT_THAT(socket(AF_INET, SOCK_RAW, Protocol()), + SyscallFailsWithErrno(EPERM)); + GTEST_SKIP(); + } ASSERT_THAT(s_ = socket(AF_INET, SOCK_RAW, Protocol()), SyscallSucceeds()); @@ -79,9 +83,10 @@ void RawSocketTest::SetUp() { } void RawSocketTest::TearDown() { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - EXPECT_THAT(close(s_), SyscallSucceeds()); + // TearDown will be run even if we skip the test. + if (ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) { + EXPECT_THAT(close(s_), SyscallSucceeds()); + } } // We should be able to create multiple raw sockets for the same protocol. diff --git a/test/syscalls/linux/socket_bind_to_device.cc b/test/syscalls/linux/socket_bind_to_device.cc new file mode 100644 index 000000000..d20821cac --- /dev/null +++ b/test/syscalls/linux/socket_bind_to_device.cc @@ -0,0 +1,314 @@ +// Copyright 2019 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 <linux/if_tun.h> +#include <net/if.h> +#include <netinet/in.h> +#include <sys/ioctl.h> +#include <sys/socket.h> +#include <sys/types.h> +#include <sys/un.h> + +#include <cstdio> +#include <cstring> +#include <map> +#include <memory> +#include <unordered_map> +#include <unordered_set> +#include <utility> +#include <vector> + +#include "gmock/gmock.h" +#include "gtest/gtest.h" +#include "gtest/gtest.h" +#include "test/syscalls/linux/ip_socket_test_util.h" +#include "test/syscalls/linux/socket_bind_to_device_util.h" +#include "test/syscalls/linux/socket_test_util.h" +#include "test/util/capability_util.h" +#include "test/util/test_util.h" +#include "test/util/thread_util.h" + +namespace gvisor { +namespace testing { + +using std::string; + +// Test fixture for SO_BINDTODEVICE tests. +class BindToDeviceTest : public ::testing::TestWithParam<SocketKind> { + protected: + void SetUp() override { + printf("Testing case: %s\n", GetParam().description.c_str()); + ASSERT_TRUE(ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) + << "CAP_NET_RAW is required to use SO_BINDTODEVICE"; + + interface_name_ = "eth1"; + auto interface_names = GetInterfaceNames(); + if (interface_names.find(interface_name_) == interface_names.end()) { + // Need a tunnel. + tunnel_ = ASSERT_NO_ERRNO_AND_VALUE(Tunnel::New()); + interface_name_ = tunnel_->GetName(); + ASSERT_FALSE(interface_name_.empty()); + } + socket_ = ASSERT_NO_ERRNO_AND_VALUE(GetParam().Create()); + } + + string interface_name() const { return interface_name_; } + + int socket_fd() const { return socket_->get(); } + + private: + std::unique_ptr<Tunnel> tunnel_; + string interface_name_; + std::unique_ptr<FileDescriptor> socket_; +}; + +constexpr char kIllegalIfnameChar = '/'; + +// Tests getsockopt of the default value. +TEST_P(BindToDeviceTest, GetsockoptDefault) { + char name_buffer[IFNAMSIZ * 2]; + char original_name_buffer[IFNAMSIZ * 2]; + socklen_t name_buffer_size; + + // Read the default SO_BINDTODEVICE. + memset(original_name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + for (size_t i = 0; i <= sizeof(name_buffer); i++) { + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer_size = i; + EXPECT_THAT(getsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, + name_buffer, &name_buffer_size), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(name_buffer_size, 0); + EXPECT_EQ(memcmp(name_buffer, original_name_buffer, sizeof(name_buffer)), + 0); + } +} + +// Tests setsockopt of invalid device name. +TEST_P(BindToDeviceTest, SetsockoptInvalidDeviceName) { + char name_buffer[IFNAMSIZ * 2]; + socklen_t name_buffer_size; + + // Set an invalid device name. + memset(name_buffer, kIllegalIfnameChar, 5); + name_buffer_size = 5; + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + name_buffer_size), + SyscallFailsWithErrno(ENODEV)); +} + +// Tests setsockopt of a buffer with a valid device name but not +// null-terminated, with different sizes of buffer. +TEST_P(BindToDeviceTest, SetsockoptValidDeviceNameWithoutNullTermination) { + char name_buffer[IFNAMSIZ * 2]; + socklen_t name_buffer_size; + + strncpy(name_buffer, interface_name().c_str(), interface_name().size() + 1); + // Intentionally overwrite the null at the end. + memset(name_buffer + interface_name().size(), kIllegalIfnameChar, + sizeof(name_buffer) - interface_name().size()); + for (size_t i = 1; i <= sizeof(name_buffer); i++) { + name_buffer_size = i; + SCOPED_TRACE(absl::StrCat("Buffer size: ", i)); + // It should only work if the size provided is exactly right. + if (name_buffer_size == interface_name().size()) { + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, + name_buffer, name_buffer_size), + SyscallSucceeds()); + } else { + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, + name_buffer, name_buffer_size), + SyscallFailsWithErrno(ENODEV)); + } + } +} + +// Tests setsockopt of a buffer with a valid device name and null-terminated, +// with different sizes of buffer. +TEST_P(BindToDeviceTest, SetsockoptValidDeviceNameWithNullTermination) { + char name_buffer[IFNAMSIZ * 2]; + socklen_t name_buffer_size; + + strncpy(name_buffer, interface_name().c_str(), interface_name().size() + 1); + // Don't overwrite the null at the end. + memset(name_buffer + interface_name().size() + 1, kIllegalIfnameChar, + sizeof(name_buffer) - interface_name().size() - 1); + for (size_t i = 1; i <= sizeof(name_buffer); i++) { + name_buffer_size = i; + SCOPED_TRACE(absl::StrCat("Buffer size: ", i)); + // It should only work if the size provided is at least the right size. + if (name_buffer_size >= interface_name().size()) { + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, + name_buffer, name_buffer_size), + SyscallSucceeds()); + } else { + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, + name_buffer, name_buffer_size), + SyscallFailsWithErrno(ENODEV)); + } + } +} + +// Tests that setsockopt of an invalid device name doesn't unset the previous +// valid setsockopt. +TEST_P(BindToDeviceTest, SetsockoptValidThenInvalid) { + char name_buffer[IFNAMSIZ * 2]; + socklen_t name_buffer_size; + + // Write successfully. + strncpy(name_buffer, interface_name().c_str(), sizeof(name_buffer)); + ASSERT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + sizeof(name_buffer)), + SyscallSucceeds()); + + // Read it back successfully. + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer_size = sizeof(name_buffer); + EXPECT_THAT(getsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + &name_buffer_size), + SyscallSucceeds()); + EXPECT_EQ(name_buffer_size, interface_name().size() + 1); + EXPECT_STREQ(name_buffer, interface_name().c_str()); + + // Write unsuccessfully. + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer_size = 5; + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + sizeof(name_buffer)), + SyscallFailsWithErrno(ENODEV)); + + // Read it back successfully, it's unchanged. + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer_size = sizeof(name_buffer); + EXPECT_THAT(getsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + &name_buffer_size), + SyscallSucceeds()); + EXPECT_EQ(name_buffer_size, interface_name().size() + 1); + EXPECT_STREQ(name_buffer, interface_name().c_str()); +} + +// Tests that setsockopt of zero-length string correctly unsets the previous +// value. +TEST_P(BindToDeviceTest, SetsockoptValidThenClear) { + char name_buffer[IFNAMSIZ * 2]; + socklen_t name_buffer_size; + + // Write successfully. + strncpy(name_buffer, interface_name().c_str(), sizeof(name_buffer)); + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + sizeof(name_buffer)), + SyscallSucceeds()); + + // Read it back successfully. + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer_size = sizeof(name_buffer); + EXPECT_THAT(getsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + &name_buffer_size), + SyscallSucceeds()); + EXPECT_EQ(name_buffer_size, interface_name().size() + 1); + EXPECT_STREQ(name_buffer, interface_name().c_str()); + + // Clear it successfully. + name_buffer_size = 0; + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + name_buffer_size), + SyscallSucceeds()); + + // Read it back successfully, it's cleared. + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer_size = sizeof(name_buffer); + EXPECT_THAT(getsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + &name_buffer_size), + SyscallSucceeds()); + EXPECT_EQ(name_buffer_size, 0); +} + +// Tests that setsockopt of empty string correctly unsets the previous +// value. +TEST_P(BindToDeviceTest, SetsockoptValidThenClearWithNull) { + char name_buffer[IFNAMSIZ * 2]; + socklen_t name_buffer_size; + + // Write successfully. + strncpy(name_buffer, interface_name().c_str(), sizeof(name_buffer)); + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + sizeof(name_buffer)), + SyscallSucceeds()); + + // Read it back successfully. + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer_size = sizeof(name_buffer); + EXPECT_THAT(getsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + &name_buffer_size), + SyscallSucceeds()); + EXPECT_EQ(name_buffer_size, interface_name().size() + 1); + EXPECT_STREQ(name_buffer, interface_name().c_str()); + + // Clear it successfully. + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer[0] = 0; + name_buffer_size = sizeof(name_buffer); + EXPECT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + name_buffer_size), + SyscallSucceeds()); + + // Read it back successfully, it's cleared. + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer_size = sizeof(name_buffer); + EXPECT_THAT(getsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + &name_buffer_size), + SyscallSucceeds()); + EXPECT_EQ(name_buffer_size, 0); +} + +// Tests getsockopt with different buffer sizes. +TEST_P(BindToDeviceTest, GetsockoptDevice) { + char name_buffer[IFNAMSIZ * 2]; + socklen_t name_buffer_size; + + // Write successfully. + strncpy(name_buffer, interface_name().c_str(), sizeof(name_buffer)); + ASSERT_THAT(setsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, name_buffer, + sizeof(name_buffer)), + SyscallSucceeds()); + + // Read it back at various buffer sizes. + for (size_t i = 0; i <= sizeof(name_buffer); i++) { + memset(name_buffer, kIllegalIfnameChar, sizeof(name_buffer)); + name_buffer_size = i; + SCOPED_TRACE(absl::StrCat("Buffer size: ", i)); + // Linux only allows a buffer at least IFNAMSIZ, even if less would suffice + // for this interface name. + if (name_buffer_size >= IFNAMSIZ) { + EXPECT_THAT(getsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, + name_buffer, &name_buffer_size), + SyscallSucceeds()); + EXPECT_EQ(name_buffer_size, interface_name().size() + 1); + EXPECT_STREQ(name_buffer, interface_name().c_str()); + } else { + EXPECT_THAT(getsockopt(socket_fd(), SOL_SOCKET, SO_BINDTODEVICE, + name_buffer, &name_buffer_size), + SyscallFailsWithErrno(EINVAL)); + EXPECT_EQ(name_buffer_size, i); + } + } +} + +INSTANTIATE_TEST_SUITE_P(BindToDeviceTest, BindToDeviceTest, + ::testing::Values(IPv4UDPUnboundSocket(0), + IPv4TCPUnboundSocket(0))); + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/socket_bind_to_device_distribution.cc b/test/syscalls/linux/socket_bind_to_device_distribution.cc new file mode 100644 index 000000000..4d2400328 --- /dev/null +++ b/test/syscalls/linux/socket_bind_to_device_distribution.cc @@ -0,0 +1,381 @@ +// Copyright 2019 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 <linux/if_tun.h> +#include <net/if.h> +#include <netinet/in.h> +#include <sys/ioctl.h> +#include <sys/socket.h> +#include <sys/types.h> +#include <sys/un.h> + +#include <atomic> +#include <cstdio> +#include <cstring> +#include <map> +#include <memory> +#include <unordered_map> +#include <unordered_set> +#include <utility> +#include <vector> + +#include "gmock/gmock.h" +#include "gtest/gtest.h" +#include "gtest/gtest.h" +#include "test/syscalls/linux/ip_socket_test_util.h" +#include "test/syscalls/linux/socket_bind_to_device_util.h" +#include "test/syscalls/linux/socket_test_util.h" +#include "test/util/capability_util.h" +#include "test/util/test_util.h" +#include "test/util/thread_util.h" + +namespace gvisor { +namespace testing { + +using std::string; +using std::vector; + +struct EndpointConfig { + std::string bind_to_device; + double expected_ratio; +}; + +struct DistributionTestCase { + std::string name; + std::vector<EndpointConfig> endpoints; +}; + +struct ListenerConnector { + TestAddress listener; + TestAddress connector; +}; + +// Test fixture for SO_BINDTODEVICE tests the distribution of packets received +// with varying SO_BINDTODEVICE settings. +class BindToDeviceDistributionTest + : public ::testing::TestWithParam< + ::testing::tuple<ListenerConnector, DistributionTestCase>> { + protected: + void SetUp() override { + printf("Testing case: %s, listener=%s, connector=%s\n", + ::testing::get<1>(GetParam()).name.c_str(), + ::testing::get<0>(GetParam()).listener.description.c_str(), + ::testing::get<0>(GetParam()).connector.description.c_str()); + ASSERT_TRUE(ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) + << "CAP_NET_RAW is required to use SO_BINDTODEVICE"; + } +}; + +PosixErrorOr<uint16_t> AddrPort(int family, sockaddr_storage const& addr) { + switch (family) { + case AF_INET: + return static_cast<uint16_t>( + reinterpret_cast<sockaddr_in const*>(&addr)->sin_port); + case AF_INET6: + return static_cast<uint16_t>( + reinterpret_cast<sockaddr_in6 const*>(&addr)->sin6_port); + default: + return PosixError(EINVAL, + absl::StrCat("unknown socket family: ", family)); + } +} + +PosixError SetAddrPort(int family, sockaddr_storage* addr, uint16_t port) { + switch (family) { + case AF_INET: + reinterpret_cast<sockaddr_in*>(addr)->sin_port = port; + return NoError(); + case AF_INET6: + reinterpret_cast<sockaddr_in6*>(addr)->sin6_port = port; + return NoError(); + default: + return PosixError(EINVAL, + absl::StrCat("unknown socket family: ", family)); + } +} + +// Binds sockets to different devices and then creates many TCP connections. +// Checks that the distribution of connections received on the sockets matches +// the expectation. +TEST_P(BindToDeviceDistributionTest, Tcp) { + auto const& [listener_connector, test] = GetParam(); + + TestAddress const& listener = listener_connector.listener; + TestAddress const& connector = listener_connector.connector; + sockaddr_storage listen_addr = listener.addr; + sockaddr_storage conn_addr = connector.addr; + + auto interface_names = GetInterfaceNames(); + + // Create the listening sockets. + std::vector<FileDescriptor> listener_fds; + std::vector<std::unique_ptr<Tunnel>> all_tunnels; + for (auto const& endpoint : test.endpoints) { + if (!endpoint.bind_to_device.empty() && + interface_names.find(endpoint.bind_to_device) == + interface_names.end()) { + all_tunnels.push_back( + ASSERT_NO_ERRNO_AND_VALUE(Tunnel::New(endpoint.bind_to_device))); + interface_names.insert(endpoint.bind_to_device); + } + + listener_fds.push_back(ASSERT_NO_ERRNO_AND_VALUE( + Socket(listener.family(), SOCK_STREAM, IPPROTO_TCP))); + int fd = listener_fds.back().get(); + + ASSERT_THAT(setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); + ASSERT_THAT(setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, + endpoint.bind_to_device.c_str(), + endpoint.bind_to_device.size() + 1), + SyscallSucceeds()); + ASSERT_THAT( + bind(fd, reinterpret_cast<sockaddr*>(&listen_addr), listener.addr_len), + SyscallSucceeds()); + ASSERT_THAT(listen(fd, 40), SyscallSucceeds()); + + // On the first bind we need to determine which port was bound. + if (listener_fds.size() > 1) { + continue; + } + + // Get the port bound by the listening socket. + socklen_t addrlen = listener.addr_len; + ASSERT_THAT( + getsockname(listener_fds[0].get(), + reinterpret_cast<sockaddr*>(&listen_addr), &addrlen), + SyscallSucceeds()); + uint16_t const port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(listener.family(), listen_addr)); + ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); + } + + constexpr int kConnectAttempts = 10000; + std::atomic<int> connects_received = ATOMIC_VAR_INIT(0); + std::vector<int> accept_counts(listener_fds.size(), 0); + std::vector<std::unique_ptr<ScopedThread>> listen_threads( + listener_fds.size()); + + for (int i = 0; i < listener_fds.size(); i++) { + listen_threads[i] = absl::make_unique<ScopedThread>( + [&listener_fds, &accept_counts, &connects_received, i, + kConnectAttempts]() { + do { + auto fd = Accept(listener_fds[i].get(), nullptr, nullptr); + if (!fd.ok()) { + // Another thread has shutdown our read side causing the accept to + // fail. + ASSERT_GE(connects_received, kConnectAttempts) + << "errno = " << fd.error(); + return; + } + // Receive some data from a socket to be sure that the connect() + // system call has been completed on another side. + int data; + EXPECT_THAT( + RetryEINTR(recv)(fd.ValueOrDie().get(), &data, sizeof(data), 0), + SyscallSucceedsWithValue(sizeof(data))); + accept_counts[i]++; + } while (++connects_received < kConnectAttempts); + + // Shutdown all sockets to wake up other threads. + for (auto const& listener_fd : listener_fds) { + shutdown(listener_fd.get(), SHUT_RDWR); + } + }); + } + + for (int i = 0; i < kConnectAttempts; i++) { + FileDescriptor const fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP)); + ASSERT_THAT( + RetryEINTR(connect)(fd.get(), reinterpret_cast<sockaddr*>(&conn_addr), + connector.addr_len), + SyscallSucceeds()); + + EXPECT_THAT(RetryEINTR(send)(fd.get(), &i, sizeof(i), 0), + SyscallSucceedsWithValue(sizeof(i))); + } + + // Join threads to be sure that all connections have been counted. + for (auto const& listen_thread : listen_threads) { + listen_thread->Join(); + } + // Check that connections are distributed correctly among listening sockets. + for (int i = 0; i < accept_counts.size(); i++) { + EXPECT_THAT( + accept_counts[i], + EquivalentWithin(static_cast<int>(kConnectAttempts * + test.endpoints[i].expected_ratio), + 0.10)) + << "endpoint " << i << " got the wrong number of packets"; + } +} + +// Binds sockets to different devices and then sends many UDP packets. Checks +// that the distribution of packets received on the sockets matches the +// expectation. +TEST_P(BindToDeviceDistributionTest, Udp) { + auto const& [listener_connector, test] = GetParam(); + + TestAddress const& listener = listener_connector.listener; + TestAddress const& connector = listener_connector.connector; + sockaddr_storage listen_addr = listener.addr; + sockaddr_storage conn_addr = connector.addr; + + auto interface_names = GetInterfaceNames(); + + // Create the listening socket. + std::vector<FileDescriptor> listener_fds; + std::vector<std::unique_ptr<Tunnel>> all_tunnels; + for (auto const& endpoint : test.endpoints) { + if (!endpoint.bind_to_device.empty() && + interface_names.find(endpoint.bind_to_device) == + interface_names.end()) { + all_tunnels.push_back( + ASSERT_NO_ERRNO_AND_VALUE(Tunnel::New(endpoint.bind_to_device))); + interface_names.insert(endpoint.bind_to_device); + } + + listener_fds.push_back( + ASSERT_NO_ERRNO_AND_VALUE(Socket(listener.family(), SOCK_DGRAM, 0))); + int fd = listener_fds.back().get(); + + ASSERT_THAT(setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); + ASSERT_THAT(setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, + endpoint.bind_to_device.c_str(), + endpoint.bind_to_device.size() + 1), + SyscallSucceeds()); + ASSERT_THAT( + bind(fd, reinterpret_cast<sockaddr*>(&listen_addr), listener.addr_len), + SyscallSucceeds()); + + // On the first bind we need to determine which port was bound. + if (listener_fds.size() > 1) { + continue; + } + + // Get the port bound by the listening socket. + socklen_t addrlen = listener.addr_len; + ASSERT_THAT( + getsockname(listener_fds[0].get(), + reinterpret_cast<sockaddr*>(&listen_addr), &addrlen), + SyscallSucceeds()); + uint16_t const port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(listener.family(), listen_addr)); + ASSERT_NO_ERRNO(SetAddrPort(listener.family(), &listen_addr, port)); + ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); + } + + constexpr int kConnectAttempts = 10000; + std::atomic<int> packets_received = ATOMIC_VAR_INIT(0); + std::vector<int> packets_per_socket(listener_fds.size(), 0); + std::vector<std::unique_ptr<ScopedThread>> receiver_threads( + listener_fds.size()); + + for (int i = 0; i < listener_fds.size(); i++) { + receiver_threads[i] = absl::make_unique<ScopedThread>( + [&listener_fds, &packets_per_socket, &packets_received, i]() { + do { + struct sockaddr_storage addr = {}; + socklen_t addrlen = sizeof(addr); + int data; + + auto ret = RetryEINTR(recvfrom)( + listener_fds[i].get(), &data, sizeof(data), 0, + reinterpret_cast<struct sockaddr*>(&addr), &addrlen); + + if (packets_received < kConnectAttempts) { + ASSERT_THAT(ret, SyscallSucceedsWithValue(sizeof(data))); + } + + if (ret != sizeof(data)) { + // Another thread may have shutdown our read side causing the + // recvfrom to fail. + break; + } + + packets_received++; + packets_per_socket[i]++; + + // A response is required to synchronize with the main thread, + // otherwise the main thread can send more than can fit into receive + // queues. + EXPECT_THAT(RetryEINTR(sendto)( + listener_fds[i].get(), &data, sizeof(data), 0, + reinterpret_cast<sockaddr*>(&addr), addrlen), + SyscallSucceedsWithValue(sizeof(data))); + } while (packets_received < kConnectAttempts); + + // Shutdown all sockets to wake up other threads. + for (auto const& listener_fd : listener_fds) { + shutdown(listener_fd.get(), SHUT_RDWR); + } + }); + } + + for (int i = 0; i < kConnectAttempts; i++) { + FileDescriptor const fd = + ASSERT_NO_ERRNO_AND_VALUE(Socket(connector.family(), SOCK_DGRAM, 0)); + EXPECT_THAT(RetryEINTR(sendto)(fd.get(), &i, sizeof(i), 0, + reinterpret_cast<sockaddr*>(&conn_addr), + connector.addr_len), + SyscallSucceedsWithValue(sizeof(i))); + int data; + EXPECT_THAT(RetryEINTR(recv)(fd.get(), &data, sizeof(data), 0), + SyscallSucceedsWithValue(sizeof(data))); + } + + // Join threads to be sure that all connections have been counted. + for (auto const& receiver_thread : receiver_threads) { + receiver_thread->Join(); + } + // Check that packets are distributed correctly among listening sockets. + for (int i = 0; i < packets_per_socket.size(); i++) { + EXPECT_THAT( + packets_per_socket[i], + EquivalentWithin(static_cast<int>(kConnectAttempts * + test.endpoints[i].expected_ratio), + 0.10)) + << "endpoint " << i << " got the wrong number of packets"; + } +} + +std::vector<DistributionTestCase> GetDistributionTestCases() { + return std::vector<DistributionTestCase>{ + {"Even distribution among sockets not bound to device", + {{"", 1. / 3}, {"", 1. / 3}, {"", 1. / 3}}}, + {"Sockets bound to other interfaces get no packets", + {{"eth1", 0}, {"", 1. / 2}, {"", 1. / 2}}}, + {"Bound has priority over unbound", {{"eth1", 0}, {"", 0}, {"lo", 1}}}, + {"Even distribution among sockets bound to device", + {{"eth1", 0}, {"lo", 1. / 2}, {"lo", 1. / 2}}}, + }; +} + +INSTANTIATE_TEST_SUITE_P( + BindToDeviceTest, BindToDeviceDistributionTest, + ::testing::Combine(::testing::Values( + // Listeners bound to IPv4 addresses refuse + // connections using IPv6 addresses. + ListenerConnector{V4Any(), V4Loopback()}, + ListenerConnector{V4Loopback(), V4MappedLoopback()}), + ::testing::ValuesIn(GetDistributionTestCases()))); + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/socket_bind_to_device_sequence.cc b/test/syscalls/linux/socket_bind_to_device_sequence.cc new file mode 100644 index 000000000..a7365d139 --- /dev/null +++ b/test/syscalls/linux/socket_bind_to_device_sequence.cc @@ -0,0 +1,316 @@ +// Copyright 2019 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 <linux/capability.h> +#include <linux/if_tun.h> +#include <net/if.h> +#include <netinet/in.h> +#include <sys/ioctl.h> +#include <sys/socket.h> +#include <sys/types.h> +#include <sys/un.h> + +#include <cstdio> +#include <cstring> +#include <map> +#include <memory> +#include <unordered_map> +#include <unordered_set> +#include <utility> +#include <vector> + +#include "gmock/gmock.h" +#include "gtest/gtest.h" +#include "gtest/gtest.h" +#include "test/syscalls/linux/ip_socket_test_util.h" +#include "test/syscalls/linux/socket_bind_to_device_util.h" +#include "test/syscalls/linux/socket_test_util.h" +#include "test/util/capability_util.h" +#include "test/util/test_util.h" +#include "test/util/thread_util.h" + +namespace gvisor { +namespace testing { + +using std::string; +using std::vector; + +// Test fixture for SO_BINDTODEVICE tests the results of sequences of socket +// binding. +class BindToDeviceSequenceTest : public ::testing::TestWithParam<SocketKind> { + protected: + void SetUp() override { + printf("Testing case: %s\n", GetParam().description.c_str()); + ASSERT_TRUE(ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) + << "CAP_NET_RAW is required to use SO_BINDTODEVICE"; + socket_factory_ = GetParam(); + + interface_names_ = GetInterfaceNames(); + } + + PosixErrorOr<std::unique_ptr<FileDescriptor>> NewSocket() const { + return socket_factory_.Create(); + } + + // Gets a device by device_id. If the device_id has been seen before, returns + // the previously returned device. If not, finds or creates a new device. + // Returns an empty string on failure. + void GetDevice(int device_id, string *device_name) { + auto device = devices_.find(device_id); + if (device != devices_.end()) { + *device_name = device->second; + return; + } + + // Need to pick a new device. Try ethernet first. + *device_name = absl::StrCat("eth", next_unused_eth_); + if (interface_names_.find(*device_name) != interface_names_.end()) { + devices_[device_id] = *device_name; + next_unused_eth_++; + return; + } + + // Need to make a new tunnel device. gVisor tests should have enough + // ethernet devices to never reach here. + ASSERT_FALSE(IsRunningOnGvisor()); + // Need a tunnel. + tunnels_.push_back(ASSERT_NO_ERRNO_AND_VALUE(Tunnel::New())); + devices_[device_id] = tunnels_.back()->GetName(); + *device_name = devices_[device_id]; + } + + // Release the socket + void ReleaseSocket(int socket_id) { + // Close the socket that was made in a previous action. The socket_id + // indicates which socket to close based on index into the list of actions. + sockets_to_close_.erase(socket_id); + } + + // Bind a socket with the reuse option and bind_to_device options. Checks + // that all steps succeed and that the bind command's error matches want. + // Sets the socket_id to uniquely identify the socket bound if it is not + // nullptr. + void BindSocket(bool reuse, int device_id = 0, int want = 0, + int *socket_id = nullptr) { + next_socket_id_++; + sockets_to_close_[next_socket_id_] = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + auto socket_fd = sockets_to_close_[next_socket_id_]->get(); + if (socket_id != nullptr) { + *socket_id = next_socket_id_; + } + + // If reuse is indicated, do that. + if (reuse) { + EXPECT_THAT(setsockopt(socket_fd, SOL_SOCKET, SO_REUSEPORT, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceedsWithValue(0)); + } + + // If the device is non-zero, bind to that device. + if (device_id != 0) { + string device_name; + ASSERT_NO_FATAL_FAILURE(GetDevice(device_id, &device_name)); + EXPECT_THAT(setsockopt(socket_fd, SOL_SOCKET, SO_BINDTODEVICE, + device_name.c_str(), device_name.size() + 1), + SyscallSucceedsWithValue(0)); + char get_device[100]; + socklen_t get_device_size = 100; + EXPECT_THAT(getsockopt(socket_fd, SOL_SOCKET, SO_BINDTODEVICE, get_device, + &get_device_size), + SyscallSucceedsWithValue(0)); + } + + struct sockaddr_in addr = {}; + addr.sin_family = AF_INET; + addr.sin_addr.s_addr = htonl(INADDR_LOOPBACK); + addr.sin_port = port_; + if (want == 0) { + ASSERT_THAT( + bind(socket_fd, reinterpret_cast<const struct sockaddr *>(&addr), + sizeof(addr)), + SyscallSucceeds()); + } else { + ASSERT_THAT( + bind(socket_fd, reinterpret_cast<const struct sockaddr *>(&addr), + sizeof(addr)), + SyscallFailsWithErrno(want)); + } + + if (port_ == 0) { + // We don't yet know what port we'll be using so we need to fetch it and + // remember it for future commands. + socklen_t addr_size = sizeof(addr); + ASSERT_THAT( + getsockname(socket_fd, reinterpret_cast<struct sockaddr *>(&addr), + &addr_size), + SyscallSucceeds()); + port_ = addr.sin_port; + } + } + + private: + SocketKind socket_factory_; + // devices maps from the device id in the test case to the name of the device. + std::unordered_map<int, string> devices_; + // These are the tunnels that were created for the test and will be destroyed + // by the destructor. + vector<std::unique_ptr<Tunnel>> tunnels_; + // A list of all interface names before the test started. + std::unordered_set<string> interface_names_; + // The next ethernet device to use when requested a device. + int next_unused_eth_ = 1; + // The port for all tests. Originally 0 (any) and later set to the port that + // all further commands will use. + in_port_t port_ = 0; + // sockets_to_close_ is a map from action index to the socket that was + // created. + std::unordered_map<int, + std::unique_ptr<gvisor::testing::FileDescriptor>> + sockets_to_close_; + int next_socket_id_ = 0; +}; + +TEST_P(BindToDeviceSequenceTest, BindTwiceWithDeviceFails) { + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 3)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 3, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, BindToDevice) { + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 1)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 2)); +} + +TEST_P(BindToDeviceSequenceTest, BindToDeviceAndThenWithoutDevice) { + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 0, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, BindWithoutDevice) { + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse */ false)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 0, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, BindWithDevice) { + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 123, 0)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 456, 0)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 789, 0)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 0, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, BindWithReuse) { + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse */ true)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse */ true, /* bind_to_device */ 0)); +} + +TEST_P(BindToDeviceSequenceTest, BindingWithReuseAndDevice) { + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 123, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 0, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 456)); + ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse */ true)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 789)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 999, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, MixingReuseAndNotReuseByBindingToDevice) { + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 123, 0)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 456, 0)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 789, 0)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 999, 0)); +} + +TEST_P(BindToDeviceSequenceTest, CannotBindTo0AfterMixingReuseAndNotReuse) { + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 456)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 0, EADDRINUSE)); +} + +TEST_P(BindToDeviceSequenceTest, BindAndRelease) { + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 123)); + int to_release; + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 0, 0, &to_release)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 345, EADDRINUSE)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 789)); + // Release the bind to device 0 and try again. + ASSERT_NO_FATAL_FAILURE(ReleaseSocket(to_release)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 345)); +} + +TEST_P(BindToDeviceSequenceTest, BindTwiceWithReuseOnce) { + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ false, /* bind_to_device */ 123)); + ASSERT_NO_FATAL_FAILURE( + BindSocket(/* reuse */ true, /* bind_to_device */ 0, EADDRINUSE)); +} + +INSTANTIATE_TEST_SUITE_P(BindToDeviceTest, BindToDeviceSequenceTest, + ::testing::Values(IPv4UDPUnboundSocket(0), + IPv4TCPUnboundSocket(0))); + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/socket_bind_to_device_util.cc b/test/syscalls/linux/socket_bind_to_device_util.cc new file mode 100644 index 000000000..f4ee775bd --- /dev/null +++ b/test/syscalls/linux/socket_bind_to_device_util.cc @@ -0,0 +1,75 @@ +// Copyright 2019 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 "test/syscalls/linux/socket_bind_to_device_util.h" + +#include <arpa/inet.h> +#include <fcntl.h> +#include <linux/if_tun.h> +#include <net/if.h> +#include <netinet/in.h> +#include <sys/ioctl.h> +#include <sys/socket.h> +#include <sys/types.h> +#include <sys/un.h> +#include <unistd.h> + +#include <cstdio> +#include <cstring> +#include <map> +#include <memory> +#include <unordered_map> +#include <unordered_set> +#include <utility> +#include <vector> + +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +using std::string; + +PosixErrorOr<std::unique_ptr<Tunnel>> Tunnel::New(string tunnel_name) { + int fd; + RETURN_ERROR_IF_SYSCALL_FAIL(fd = open("/dev/net/tun", O_RDWR)); + + // Using `new` to access a non-public constructor. + auto new_tunnel = absl::WrapUnique(new Tunnel(fd)); + + ifreq ifr = {}; + ifr.ifr_flags = IFF_TUN; + strncpy(ifr.ifr_name, tunnel_name.c_str(), sizeof(ifr.ifr_name)); + + RETURN_ERROR_IF_SYSCALL_FAIL(ioctl(fd, TUNSETIFF, &ifr)); + new_tunnel->name_ = ifr.ifr_name; + return new_tunnel; +} + +std::unordered_set<string> GetInterfaceNames() { + struct if_nameindex* interfaces = if_nameindex(); + std::unordered_set<string> names; + if (interfaces == nullptr) { + return names; + } + for (auto interface = interfaces; + interface->if_index != 0 || interface->if_name != nullptr; interface++) { + names.insert(interface->if_name); + } + if_freenameindex(interfaces); + return names; +} + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/socket_bind_to_device_util.h b/test/syscalls/linux/socket_bind_to_device_util.h new file mode 100644 index 000000000..f941ccc86 --- /dev/null +++ b/test/syscalls/linux/socket_bind_to_device_util.h @@ -0,0 +1,67 @@ +// Copyright 2019 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. + +#ifndef GVISOR_TEST_SYSCALLS_SOCKET_BIND_TO_DEVICE_UTILS_H_ +#define GVISOR_TEST_SYSCALLS_SOCKET_BIND_TO_DEVICE_UTILS_H_ + +#include <arpa/inet.h> +#include <linux/if_tun.h> +#include <net/if.h> +#include <netinet/in.h> +#include <sys/ioctl.h> +#include <sys/socket.h> +#include <sys/types.h> +#include <sys/un.h> +#include <unistd.h> + +#include <cstdio> +#include <cstring> +#include <map> +#include <memory> +#include <string> +#include <unordered_map> +#include <unordered_set> +#include <utility> +#include <vector> + +#include "absl/memory/memory.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +class Tunnel { + public: + static PosixErrorOr<std::unique_ptr<Tunnel>> New( + std::string tunnel_name = ""); + const std::string& GetName() const { return name_; } + + ~Tunnel() { + if (fd_ != -1) { + close(fd_); + } + } + + private: + Tunnel(int fd) : fd_(fd) {} + int fd_ = -1; + std::string name_; +}; + +std::unordered_set<std::string> GetInterfaceNames(); + +} // namespace testing +} // namespace gvisor + +#endif // GVISOR_TEST_SYSCALLS_SOCKET_BIND_TO_DEVICE_UTILS_H_ diff --git a/test/syscalls/linux/uidgid.cc b/test/syscalls/linux/uidgid.cc index d48453a93..6218fbce1 100644 --- a/test/syscalls/linux/uidgid.cc +++ b/test/syscalls/linux/uidgid.cc @@ -25,6 +25,7 @@ #include "test/util/posix_error.h" #include "test/util/test_util.h" #include "test/util/thread_util.h" +#include "test/util/uid_util.h" ABSL_FLAG(int32_t, scratch_uid1, 65534, "first scratch UID"); ABSL_FLAG(int32_t, scratch_uid2, 65533, "second scratch UID"); @@ -68,30 +69,6 @@ TEST(UidGidTest, Getgroups) { // here; see the setgroups test below. } -// If the caller's real/effective/saved user/group IDs are all 0, IsRoot returns -// true. Otherwise IsRoot logs an explanatory message and returns false. -PosixErrorOr<bool> IsRoot() { - uid_t ruid, euid, suid; - int rc = getresuid(&ruid, &euid, &suid); - MaybeSave(); - if (rc < 0) { - return PosixError(errno, "getresuid"); - } - if (ruid != 0 || euid != 0 || suid != 0) { - return false; - } - gid_t rgid, egid, sgid; - rc = getresgid(&rgid, &egid, &sgid); - MaybeSave(); - if (rc < 0) { - return PosixError(errno, "getresgid"); - } - if (rgid != 0 || egid != 0 || sgid != 0) { - return false; - } - return true; -} - // Checks that the calling process' real/effective/saved user IDs are // ruid/euid/suid respectively. PosixError CheckUIDs(uid_t ruid, uid_t euid, uid_t suid) { diff --git a/test/syscalls/linux/uname.cc b/test/syscalls/linux/uname.cc index 0a5d91017..d8824b171 100644 --- a/test/syscalls/linux/uname.cc +++ b/test/syscalls/linux/uname.cc @@ -41,6 +41,19 @@ TEST(UnameTest, Sanity) { TEST(UnameTest, SetNames) { SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN))); + char hostname[65]; + ASSERT_THAT(sethostname("0123456789", 3), SyscallSucceeds()); + EXPECT_THAT(gethostname(hostname, sizeof(hostname)), SyscallSucceeds()); + EXPECT_EQ(absl::string_view(hostname), "012"); + + ASSERT_THAT(sethostname("0123456789\0xxx", 11), SyscallSucceeds()); + EXPECT_THAT(gethostname(hostname, sizeof(hostname)), SyscallSucceeds()); + EXPECT_EQ(absl::string_view(hostname), "0123456789"); + + ASSERT_THAT(sethostname("0123456789\0xxx", 12), SyscallSucceeds()); + EXPECT_THAT(gethostname(hostname, sizeof(hostname)), SyscallSucceeds()); + EXPECT_EQ(absl::string_view(hostname), "0123456789"); + constexpr char kHostname[] = "wubbalubba"; ASSERT_THAT(sethostname(kHostname, sizeof(kHostname)), SyscallSucceeds()); @@ -54,7 +67,6 @@ TEST(UnameTest, SetNames) { EXPECT_EQ(absl::string_view(buf.domainname), kDomainname); // These should just be glibc wrappers that also call uname(2). - char hostname[65]; EXPECT_THAT(gethostname(hostname, sizeof(hostname)), SyscallSucceeds()); EXPECT_EQ(absl::string_view(hostname), kHostname); diff --git a/test/util/BUILD b/test/util/BUILD index 25ed9c944..5d2a9cc2c 100644 --- a/test/util/BUILD +++ b/test/util/BUILD @@ -324,3 +324,14 @@ cc_library( ":test_util", ], ) + +cc_library( + name = "uid_util", + testonly = 1, + srcs = ["uid_util.cc"], + hdrs = ["uid_util.h"], + deps = [ + ":posix_error", + ":save_util", + ], +) diff --git a/test/util/proc_util.cc b/test/util/proc_util.cc index 75b24da37..34d636ba9 100644 --- a/test/util/proc_util.cc +++ b/test/util/proc_util.cc @@ -88,7 +88,7 @@ PosixErrorOr<std::vector<ProcMapsEntry>> ParseProcMaps( std::vector<ProcMapsEntry> entries; auto lines = absl::StrSplit(contents, '\n', absl::SkipEmpty()); for (const auto& l : lines) { - std::cout << "line: " << l; + std::cout << "line: " << l << std::endl; ASSIGN_OR_RETURN_ERRNO(auto entry, ParseProcMapsLine(l)); entries.push_back(entry); } diff --git a/test/util/uid_util.cc b/test/util/uid_util.cc new file mode 100644 index 000000000..b131b4b99 --- /dev/null +++ b/test/util/uid_util.cc @@ -0,0 +1,44 @@ +// 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 "test/util/posix_error.h" +#include "test/util/save_util.h" + +namespace gvisor { +namespace testing { + +PosixErrorOr<bool> IsRoot() { + uid_t ruid, euid, suid; + int rc = getresuid(&ruid, &euid, &suid); + MaybeSave(); + if (rc < 0) { + return PosixError(errno, "getresuid"); + } + if (ruid != 0 || euid != 0 || suid != 0) { + return false; + } + gid_t rgid, egid, sgid; + rc = getresgid(&rgid, &egid, &sgid); + MaybeSave(); + if (rc < 0) { + return PosixError(errno, "getresgid"); + } + if (rgid != 0 || egid != 0 || sgid != 0) { + return false; + } + return true; +} + +} // namespace testing +} // namespace gvisor diff --git a/test/util/uid_util.h b/test/util/uid_util.h new file mode 100644 index 000000000..2cd387fb0 --- /dev/null +++ b/test/util/uid_util.h @@ -0,0 +1,29 @@ +// 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. + +#ifndef GVISOR_TEST_SYSCALLS_UID_UTIL_H_ +#define GVISOR_TEST_SYSCALLS_UID_UTIL_H_ + +#include "test/util/posix_error.h" + +namespace gvisor { +namespace testing { + +// Returns true if the caller's real/effective/saved user/group IDs are all 0. +PosixErrorOr<bool> IsRoot(); + +} // namespace testing +} // namespace gvisor + +#endif // GVISOR_TEST_SYSCALLS_UID_UTIL_H_ |