Merge branch 'master' into iptables-write-input-drop

14 files changed, 297 insertions, 23 deletions

diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go
index 5927eb017..0c9dee8fe 100644
--- a/test/iptables/iptables_test.go
+++ b/test/iptables/iptables_test.go
@@ -15,7 +15,6 @@
 package iptables
 
 import (
-	"flag"
 	"fmt"
 	"net"
 	"os"
@@ -23,6 +22,7 @@ import (
 	"testing"
 	"time"
 
+	"flag"
 	"gvisor.dev/gvisor/pkg/log"
 	"gvisor.dev/gvisor/runsc/dockerutil"
 	"gvisor.dev/gvisor/runsc/testutil"
diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD
index 064ce8429..ce8abe217 100644
--- a/test/syscalls/linux/BUILD
+++ b/test/syscalls/linux/BUILD
@@ -2888,7 +2888,6 @@ cc_library(
         ":unix_domain_socket_test_util",
         "//test/util:test_util",
         "//test/util:thread_util",
-        "//test/util:timer_util",
         "@com_google_absl//absl/time",
         "@com_google_googletest//:gtest",
     ],
diff --git a/test/syscalls/linux/inotify.cc b/test/syscalls/linux/inotify.cc
index 59ec9940a..fdef646eb 100644
--- a/test/syscalls/linux/inotify.cc
+++ b/test/syscalls/linux/inotify.cc
@@ -977,7 +977,7 @@ TEST(Inotify, WatchOnRelativePath) {
       ASSERT_NO_ERRNO_AND_VALUE(Open(file1.path(), O_RDONLY));
 
   // Change working directory to root.
-  const char* old_working_dir = get_current_dir_name();
+  const FileDescriptor cwd = ASSERT_NO_ERRNO_AND_VALUE(Open(".", O_PATH));
   EXPECT_THAT(chdir(root.path().c_str()), SyscallSucceeds());
 
   // Add a watch on file1 with a relative path.
@@ -997,7 +997,7 @@ TEST(Inotify, WatchOnRelativePath) {
   // continue to hold a reference, random save/restore tests can fail if a save
   // is triggered after "root" is unlinked; we can't save deleted fs objects
   // with active references.
-  EXPECT_THAT(chdir(old_working_dir), SyscallSucceeds());
+  EXPECT_THAT(fchdir(cwd.get()), SyscallSucceeds());
 }
 
 TEST(Inotify, ZeroLengthReadWriteDoesNotGenerateEvent) {
diff --git a/test/syscalls/linux/ip_socket_test_util.cc b/test/syscalls/linux/ip_socket_test_util.cc
index 8398fc95f..6b472eb2f 100644
--- a/test/syscalls/linux/ip_socket_test_util.cc
+++ b/test/syscalls/linux/ip_socket_test_util.cc
@@ -187,24 +187,24 @@ PosixErrorOr<int> IfAddrHelper::GetIndex(std::string name) {
   return InterfaceIndex(name);
 }
 
-std::string GetAddr4Str(in_addr* a) {
+std::string GetAddr4Str(const in_addr* a) {
   char str[INET_ADDRSTRLEN];
   inet_ntop(AF_INET, a, str, sizeof(str));
   return std::string(str);
 }
 
-std::string GetAddr6Str(in6_addr* a) {
+std::string GetAddr6Str(const in6_addr* a) {
   char str[INET6_ADDRSTRLEN];
   inet_ntop(AF_INET6, a, str, sizeof(str));
   return std::string(str);
 }
 
-std::string GetAddrStr(sockaddr* a) {
+std::string GetAddrStr(const sockaddr* a) {
   if (a->sa_family == AF_INET) {
-    auto src = &(reinterpret_cast<sockaddr_in*>(a)->sin_addr);
+    auto src = &(reinterpret_cast<const sockaddr_in*>(a)->sin_addr);
     return GetAddr4Str(src);
   } else if (a->sa_family == AF_INET6) {
-    auto src = &(reinterpret_cast<sockaddr_in6*>(a)->sin6_addr);
+    auto src = &(reinterpret_cast<const sockaddr_in6*>(a)->sin6_addr);
     return GetAddr6Str(src);
   }
   return std::string("<invalid>");
diff --git a/test/syscalls/linux/ip_socket_test_util.h b/test/syscalls/linux/ip_socket_test_util.h
index 9cb4566db..0f58e0f77 100644
--- a/test/syscalls/linux/ip_socket_test_util.h
+++ b/test/syscalls/linux/ip_socket_test_util.h
@@ -105,14 +105,14 @@ class IfAddrHelper {
 };
 
 // GetAddr4Str returns the given IPv4 network address structure as a string.
-std::string GetAddr4Str(in_addr* a);
+std::string GetAddr4Str(const in_addr* a);
 
 // GetAddr6Str returns the given IPv6 network address structure as a string.
-std::string GetAddr6Str(in6_addr* a);
+std::string GetAddr6Str(const in6_addr* a);
 
 // GetAddrStr returns the given IPv4 or IPv6 network address structure as a
 // string.
-std::string GetAddrStr(sockaddr* a);
+std::string GetAddrStr(const sockaddr* a);
 
 }  // namespace testing
 }  // namespace gvisor
diff --git a/test/syscalls/linux/poll.cc b/test/syscalls/linux/poll.cc
index 9e5aa7fd0..c42472474 100644
--- a/test/syscalls/linux/poll.cc
+++ b/test/syscalls/linux/poll.cc
@@ -275,7 +275,8 @@ TEST_F(PollTest, Nfds) {
   // Each entry in the 'fds' array refers to the eventfd and polls for
   // "writable" events (events=POLLOUT). This essentially guarantees that the
   // poll() is a no-op and allows negative testing of the 'nfds' parameter.
-  std::vector<struct pollfd> fds(max_fds, {.fd = efd.get(), .events = POLLOUT});
+  std::vector<struct pollfd> fds(max_fds + 1,
+                                 {.fd = efd.get(), .events = POLLOUT});
 
   // Verify that 'nfds' up to RLIMIT_NOFILE are allowed.
   EXPECT_THAT(RetryEINTR(poll)(fds.data(), 1, 1), SyscallSucceedsWithValue(1));
diff --git a/test/syscalls/linux/preadv2.cc b/test/syscalls/linux/preadv2.cc
index c9246367d..cd936ea90 100644
--- a/test/syscalls/linux/preadv2.cc
+++ b/test/syscalls/linux/preadv2.cc
@@ -202,7 +202,7 @@ TEST(Preadv2Test, TestInvalidOffset) {
   iov[0].iov_len = 0;
 
   EXPECT_THAT(preadv2(fd.get(), iov.get(), /*iovcnt=*/1, /*offset=*/-8,
-                      /*flags=*/RWF_HIPRI),
+                      /*flags=*/0),
               SyscallFailsWithErrno(EINVAL));
 }
 
diff --git a/test/syscalls/linux/readv_common.cc b/test/syscalls/linux/readv_common.cc
index 491d5f40f..2694dc64f 100644
--- a/test/syscalls/linux/readv_common.cc
+++ b/test/syscalls/linux/readv_common.cc
@@ -154,7 +154,7 @@ void ReadBuffersOverlapping(int fd) {
   char* expected_ptr = expected.data();
   memcpy(expected_ptr, &kReadvTestData[overlap_bytes], overlap_bytes);
   memcpy(&expected_ptr[overlap_bytes], &kReadvTestData[overlap_bytes],
-         kReadvTestDataSize);
+         kReadvTestDataSize - overlap_bytes);
 
   struct iovec iovs[2];
   iovs[0].iov_base = buffer.data();
diff --git a/test/syscalls/linux/socket_bind_to_device_distribution.cc b/test/syscalls/linux/socket_bind_to_device_distribution.cc
index 5767181a1..5ed57625c 100644
--- a/test/syscalls/linux/socket_bind_to_device_distribution.cc
+++ b/test/syscalls/linux/socket_bind_to_device_distribution.cc
@@ -183,7 +183,14 @@ TEST_P(BindToDeviceDistributionTest, Tcp) {
             }
             // Receive some data from a socket to be sure that the connect()
             // system call has been completed on another side.
-            int data;
+            // Do a short read and then close the socket to trigger a RST. This
+            // ensures that both ends of the connection are cleaned up and no
+            // goroutines hang around in TIME-WAIT. We do this so that this test
+            // does not timeout under gotsan runs where lots of goroutines can
+            // cause the test to use absurd amounts of memory.
+            //
+            // See: https://tools.ietf.org/html/rfc2525#page-50 section 2.17
+            uint16_t data;
             EXPECT_THAT(
                 RetryEINTR(recv)(fd.ValueOrDie().get(), &data, sizeof(data), 0),
                 SyscallSucceedsWithValue(sizeof(data)));
@@ -198,15 +205,29 @@ TEST_P(BindToDeviceDistributionTest, Tcp) {
   }
 
   for (int i = 0; i < kConnectAttempts; i++) {
-    FileDescriptor const fd = ASSERT_NO_ERRNO_AND_VALUE(
+    const FileDescriptor 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());
 
+    // Do two separate sends to ensure two segments are received. This is
+    // required for netstack where read is incorrectly assuming a whole
+    // segment is read when endpoint.Read() is called which is technically
+    // incorrect as the syscall that invoked endpoint.Read() may only
+    // consume it partially. This results in a case where a close() of
+    // such a socket does not trigger a RST in netstack due to the
+    // endpoint assuming that the endpoint has no unread data.
     EXPECT_THAT(RetryEINTR(send)(fd.get(), &i, sizeof(i), 0),
                 SyscallSucceedsWithValue(sizeof(i)));
+
+    // TODO(gvisor.dev/issue/1449): Remove this block once netstack correctly
+    //   generates a RST.
+    if (IsRunningOnGvisor()) {
+      EXPECT_THAT(RetryEINTR(send)(fd.get(), &i, sizeof(i), 0),
+                  SyscallSucceedsWithValue(sizeof(i)));
+    }
   }
 
   // Join threads to be sure that all connections have been counted.
diff --git a/test/syscalls/linux/socket_inet_loopback.cc b/test/syscalls/linux/socket_inet_loopback.cc
index 619d41901..138024d9e 100644
--- a/test/syscalls/linux/socket_inet_loopback.cc
+++ b/test/syscalls/linux/socket_inet_loopback.cc
@@ -714,7 +714,7 @@ TEST_P(SocketInetReusePortTest, TcpPortReuseMultiThread_NoRandomSave) {
   sockaddr_storage listen_addr = listener.addr;
   sockaddr_storage conn_addr = connector.addr;
   constexpr int kThreadCount = 3;
-  constexpr int kConnectAttempts = 4096;
+  constexpr int kConnectAttempts = 10000;
 
   // Create the listening socket.
   FileDescriptor listener_fds[kThreadCount];
@@ -729,7 +729,7 @@ TEST_P(SocketInetReusePortTest, TcpPortReuseMultiThread_NoRandomSave) {
     ASSERT_THAT(
         bind(fd, reinterpret_cast<sockaddr*>(&listen_addr), listener.addr_len),
         SyscallSucceeds());
-    ASSERT_THAT(listen(fd, kConnectAttempts / 3), SyscallSucceeds());
+    ASSERT_THAT(listen(fd, 40), SyscallSucceeds());
 
     // On the first bind we need to determine which port was bound.
     if (i != 0) {
@@ -772,7 +772,14 @@ TEST_P(SocketInetReusePortTest, TcpPortReuseMultiThread_NoRandomSave) {
             }
             // Receive some data from a socket to be sure that the connect()
             // system call has been completed on another side.
-            int data;
+            // Do a short read and then close the socket to trigger a RST. This
+            // ensures that both ends of the connection are cleaned up and no
+            // goroutines hang around in TIME-WAIT. We do this so that this test
+            // does not timeout under gotsan runs where lots of goroutines can
+            // cause the test to use absurd amounts of memory.
+            //
+            // See: https://tools.ietf.org/html/rfc2525#page-50 section 2.17
+            uint16_t data;
             EXPECT_THAT(
                 RetryEINTR(recv)(fd.ValueOrDie().get(), &data, sizeof(data), 0),
                 SyscallSucceedsWithValue(sizeof(data)));
@@ -795,8 +802,22 @@ TEST_P(SocketInetReusePortTest, TcpPortReuseMultiThread_NoRandomSave) {
                               connector.addr_len),
           SyscallSucceeds());
 
+      // Do two separate sends to ensure two segments are received. This is
+      // required for netstack where read is incorrectly assuming a whole
+      // segment is read when endpoint.Read() is called which is technically
+      // incorrect as the syscall that invoked endpoint.Read() may only
+      // consume it partially. This results in a case where a close() of
+      // such a socket does not trigger a RST in netstack due to the
+      // endpoint assuming that the endpoint has no unread data.
       EXPECT_THAT(RetryEINTR(send)(fd.get(), &i, sizeof(i), 0),
                   SyscallSucceedsWithValue(sizeof(i)));
+
+      // TODO(gvisor.dev/issue/1449): Remove this block once netstack correctly
+      //   generates a RST.
+      if (IsRunningOnGvisor()) {
+        EXPECT_THAT(RetryEINTR(send)(fd.get(), &i, sizeof(i), 0),
+                    SyscallSucceedsWithValue(sizeof(i)));
+      }
     }
   });
 
diff --git a/test/syscalls/linux/socket_ip_unbound.cc b/test/syscalls/linux/socket_ip_unbound.cc
index b6754111f..ca597e267 100644
--- a/test/syscalls/linux/socket_ip_unbound.cc
+++ b/test/syscalls/linux/socket_ip_unbound.cc
@@ -129,6 +129,7 @@ TEST_P(IPUnboundSocketTest, InvalidNegativeTtl) {
 struct TOSOption {
   int level;
   int option;
+  int cmsg_level;
 };
 
 constexpr int INET_ECN_MASK = 3;
@@ -139,10 +140,12 @@ static TOSOption GetTOSOption(int domain) {
     case AF_INET:
       opt.level = IPPROTO_IP;
       opt.option = IP_TOS;
+      opt.cmsg_level = SOL_IP;
       break;
     case AF_INET6:
       opt.level = IPPROTO_IPV6;
       opt.option = IPV6_TCLASS;
+      opt.cmsg_level = SOL_IPV6;
       break;
   }
   return opt;
@@ -386,6 +389,36 @@ TEST_P(IPUnboundSocketTest, NullTOS) {
               SyscallFailsWithErrno(EFAULT));
 }
 
+TEST_P(IPUnboundSocketTest, InsufficientBufferTOS) {
+  SKIP_IF(GetParam().protocol == IPPROTO_TCP);
+
+  auto socket = ASSERT_NO_ERRNO_AND_VALUE(NewSocket());
+  TOSOption t = GetTOSOption(GetParam().domain);
+
+  in_addr addr4;
+  in6_addr addr6;
+  ASSERT_THAT(inet_pton(AF_INET, "127.0.0.1", &addr4), ::testing::Eq(1));
+  ASSERT_THAT(inet_pton(AF_INET6, "fe80::", &addr6), ::testing::Eq(1));
+
+  cmsghdr cmsg = {};
+  cmsg.cmsg_len = sizeof(cmsg);
+  cmsg.cmsg_level = t.cmsg_level;
+  cmsg.cmsg_type = t.option;
+
+  msghdr msg = {};
+  msg.msg_control = &cmsg;
+  msg.msg_controllen = sizeof(cmsg);
+  if (GetParam().domain == AF_INET) {
+    msg.msg_name = &addr4;
+    msg.msg_namelen = sizeof(addr4);
+  } else {
+    msg.msg_name = &addr6;
+    msg.msg_namelen = sizeof(addr6);
+  }
+
+  EXPECT_THAT(sendmsg(socket->get(), &msg, 0), SyscallFailsWithErrno(EINVAL));
+}
+
 INSTANTIATE_TEST_SUITE_P(
     IPUnboundSockets, IPUnboundSocketTest,
     ::testing::ValuesIn(VecCat<SocketKind>(VecCat<SocketKind>(
diff --git a/test/syscalls/linux/socket_non_stream.cc b/test/syscalls/linux/socket_non_stream.cc
index d91c5ed39..c61817f14 100644
--- a/test/syscalls/linux/socket_non_stream.cc
+++ b/test/syscalls/linux/socket_non_stream.cc
@@ -113,7 +113,7 @@ TEST_P(NonStreamSocketPairTest, RecvmsgMsghdrFlagMsgTrunc) {
   EXPECT_EQ(0, memcmp(received_data, sent_data, sizeof(received_data)));
 
   // Check that msghdr flags were updated.
-  EXPECT_EQ(msg.msg_flags, MSG_TRUNC);
+  EXPECT_EQ(msg.msg_flags & MSG_TRUNC, MSG_TRUNC);
 }
 
 // Stream sockets allow data sent with multiple sends to be peeked at in a
@@ -193,7 +193,7 @@ TEST_P(NonStreamSocketPairTest, MsgTruncTruncationRecvmsgMsghdrFlagMsgTrunc) {
   EXPECT_EQ(0, memcmp(received_data, sent_data, sizeof(received_data)));
 
   // Check that msghdr flags were updated.
-  EXPECT_EQ(msg.msg_flags, MSG_TRUNC);
+  EXPECT_EQ(msg.msg_flags & MSG_TRUNC, MSG_TRUNC);
 }
 
 TEST_P(NonStreamSocketPairTest, MsgTruncSameSize) {
@@ -224,5 +224,114 @@ TEST_P(NonStreamSocketPairTest, MsgTruncNotFull) {
   EXPECT_EQ(0, memcmp(sent_data, received_data, sizeof(sent_data)));
 }
 
+// This test tests reading from a socket with MSG_TRUNC and a zero length
+// receive buffer. The user should be able to get the message length.
+TEST_P(NonStreamSocketPairTest, RecvmsgMsgTruncZeroLen) {
+  auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair());
+
+  char sent_data[10];
+  RandomizeBuffer(sent_data, sizeof(sent_data));
+  ASSERT_THAT(
+      RetryEINTR(send)(sockets->first_fd(), sent_data, sizeof(sent_data), 0),
+      SyscallSucceedsWithValue(sizeof(sent_data)));
+
+  // The receive buffer is of zero length.
+  char received_data[0] = {};
+
+  struct iovec iov;
+  iov.iov_base = received_data;
+  iov.iov_len = sizeof(received_data);
+  struct msghdr msg = {};
+  msg.msg_flags = -1;
+  msg.msg_iov = &iov;
+  msg.msg_iovlen = 1;
+
+  // The syscall succeeds returning the full size of the message on the socket.
+  ASSERT_THAT(RetryEINTR(recvmsg)(sockets->second_fd(), &msg, MSG_TRUNC),
+              SyscallSucceedsWithValue(sizeof(sent_data)));
+
+  // Check that MSG_TRUNC is set on msghdr flags.
+  EXPECT_EQ(msg.msg_flags & MSG_TRUNC, MSG_TRUNC);
+}
+
+// This test tests reading from a socket with MSG_TRUNC | MSG_PEEK and a zero
+// length receive buffer. The user should be able to get the message length
+// without reading data off the socket.
+TEST_P(NonStreamSocketPairTest, RecvmsgMsgTruncMsgPeekZeroLen) {
+  auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair());
+
+  char sent_data[10];
+  RandomizeBuffer(sent_data, sizeof(sent_data));
+  ASSERT_THAT(
+      RetryEINTR(send)(sockets->first_fd(), sent_data, sizeof(sent_data), 0),
+      SyscallSucceedsWithValue(sizeof(sent_data)));
+
+  // The receive buffer is of zero length.
+  char peek_data[0] = {};
+
+  struct iovec peek_iov;
+  peek_iov.iov_base = peek_data;
+  peek_iov.iov_len = sizeof(peek_data);
+  struct msghdr peek_msg = {};
+  peek_msg.msg_flags = -1;
+  peek_msg.msg_iov = &peek_iov;
+  peek_msg.msg_iovlen = 1;
+
+  // The syscall succeeds returning the full size of the message on the socket.
+  ASSERT_THAT(RetryEINTR(recvmsg)(sockets->second_fd(), &peek_msg,
+                                  MSG_TRUNC | MSG_PEEK),
+              SyscallSucceedsWithValue(sizeof(sent_data)));
+
+  // Check that MSG_TRUNC is set on msghdr flags because the receive buffer is
+  // smaller than the message size.
+  EXPECT_EQ(peek_msg.msg_flags & MSG_TRUNC, MSG_TRUNC);
+
+  char received_data[sizeof(sent_data)] = {};
+
+  struct iovec received_iov;
+  received_iov.iov_base = received_data;
+  received_iov.iov_len = sizeof(received_data);
+  struct msghdr received_msg = {};
+  received_msg.msg_flags = -1;
+  received_msg.msg_iov = &received_iov;
+  received_msg.msg_iovlen = 1;
+
+  // Next we can read the actual data.
+  ASSERT_THAT(
+      RetryEINTR(recvmsg)(sockets->second_fd(), &received_msg, MSG_TRUNC),
+      SyscallSucceedsWithValue(sizeof(sent_data)));
+
+  EXPECT_EQ(0, memcmp(sent_data, received_data, sizeof(sent_data)));
+
+  // Check that MSG_TRUNC is not set on msghdr flags because we read the whole
+  // message.
+  EXPECT_EQ(received_msg.msg_flags & MSG_TRUNC, 0);
+}
+
+// This test tests reading from a socket with MSG_TRUNC | MSG_PEEK and a zero
+// length receive buffer and MSG_DONTWAIT. The user should be able to get an
+// EAGAIN or EWOULDBLOCK error response.
+TEST_P(NonStreamSocketPairTest, RecvmsgTruncPeekDontwaitZeroLen) {
+  auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair());
+
+  // NOTE: We don't send any data on the socket.
+
+  // The receive buffer is of zero length.
+  char peek_data[0] = {};
+
+  struct iovec peek_iov;
+  peek_iov.iov_base = peek_data;
+  peek_iov.iov_len = sizeof(peek_data);
+  struct msghdr peek_msg = {};
+  peek_msg.msg_flags = -1;
+  peek_msg.msg_iov = &peek_iov;
+  peek_msg.msg_iovlen = 1;
+
+  // recvmsg fails with EAGAIN because no data is available on the socket.
+  ASSERT_THAT(RetryEINTR(recvmsg)(sockets->second_fd(), &peek_msg,
+                                  MSG_TRUNC | MSG_PEEK | MSG_DONTWAIT),
+              SyscallFailsWithErrno(EAGAIN));
+}
+
 }  // namespace testing
 }  // namespace gvisor
diff --git a/test/syscalls/linux/socket_non_stream_blocking.cc b/test/syscalls/linux/socket_non_stream_blocking.cc
index 62d87c1af..b052f6e61 100644
--- a/test/syscalls/linux/socket_non_stream_blocking.cc
+++ b/test/syscalls/linux/socket_non_stream_blocking.cc
@@ -25,6 +25,7 @@
 #include "test/syscalls/linux/socket_test_util.h"
 #include "test/syscalls/linux/unix_domain_socket_test_util.h"
 #include "test/util/test_util.h"
+#include "test/util/thread_util.h"
 
 namespace gvisor {
 namespace testing {
@@ -44,5 +45,41 @@ TEST_P(BlockingNonStreamSocketPairTest, RecvLessThanBufferWaitAll) {
               SyscallSucceedsWithValue(sizeof(sent_data)));
 }
 
+// This test tests reading from a socket with MSG_TRUNC | MSG_PEEK and a zero
+// length receive buffer and MSG_DONTWAIT. The recvmsg call should block on
+// reading the data.
+TEST_P(BlockingNonStreamSocketPairTest,
+       RecvmsgTruncPeekDontwaitZeroLenBlocking) {
+  auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair());
+
+  // NOTE: We don't initially send any data on the socket.
+  const int data_size = 10;
+  char sent_data[data_size];
+  RandomizeBuffer(sent_data, data_size);
+
+  // The receive buffer is of zero length.
+  char peek_data[0] = {};
+
+  struct iovec peek_iov;
+  peek_iov.iov_base = peek_data;
+  peek_iov.iov_len = sizeof(peek_data);
+  struct msghdr peek_msg = {};
+  peek_msg.msg_flags = -1;
+  peek_msg.msg_iov = &peek_iov;
+  peek_msg.msg_iovlen = 1;
+
+  ScopedThread t([&]() {
+    // The syscall succeeds returning the full size of the message on the
+    // socket. This should block until there is data on the socket.
+    ASSERT_THAT(RetryEINTR(recvmsg)(sockets->second_fd(), &peek_msg,
+                                    MSG_TRUNC | MSG_PEEK),
+                SyscallSucceedsWithValue(data_size));
+  });
+
+  absl::SleepFor(absl::Seconds(1));
+  ASSERT_THAT(RetryEINTR(send)(sockets->first_fd(), sent_data, data_size, 0),
+              SyscallSucceedsWithValue(data_size));
+}
+
 }  // namespace testing
 }  // namespace gvisor
diff --git a/test/syscalls/linux/socket_stream.cc b/test/syscalls/linux/socket_stream.cc
index 346443f96..6522b2e01 100644
--- a/test/syscalls/linux/socket_stream.cc
+++ b/test/syscalls/linux/socket_stream.cc
@@ -104,7 +104,60 @@ TEST_P(StreamSocketPairTest, RecvmsgMsghdrFlagsNoMsgTrunc) {
   EXPECT_EQ(0, memcmp(received_data, sent_data, sizeof(received_data)));
 
   // Check that msghdr flags were cleared (MSG_TRUNC was not set).
-  EXPECT_EQ(msg.msg_flags, 0);
+  ASSERT_EQ(msg.msg_flags & MSG_TRUNC, 0);
+}
+
+TEST_P(StreamSocketPairTest, RecvmsgTruncZeroLen) {
+  auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair());
+
+  char sent_data[10];
+  RandomizeBuffer(sent_data, sizeof(sent_data));
+  ASSERT_THAT(
+      RetryEINTR(send)(sockets->first_fd(), sent_data, sizeof(sent_data), 0),
+      SyscallSucceedsWithValue(sizeof(sent_data)));
+
+  char received_data[0] = {};
+
+  struct iovec iov;
+  iov.iov_base = received_data;
+  iov.iov_len = sizeof(received_data);
+  struct msghdr msg = {};
+  msg.msg_flags = -1;
+  msg.msg_iov = &iov;
+  msg.msg_iovlen = 1;
+
+  ASSERT_THAT(RetryEINTR(recvmsg)(sockets->second_fd(), &msg, MSG_TRUNC),
+              SyscallSucceedsWithValue(0));
+
+  // Check that msghdr flags were cleared (MSG_TRUNC was not set).
+  ASSERT_EQ(msg.msg_flags & MSG_TRUNC, 0);
+}
+
+TEST_P(StreamSocketPairTest, RecvmsgTruncPeekZeroLen) {
+  auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair());
+
+  char sent_data[10];
+  RandomizeBuffer(sent_data, sizeof(sent_data));
+  ASSERT_THAT(
+      RetryEINTR(send)(sockets->first_fd(), sent_data, sizeof(sent_data), 0),
+      SyscallSucceedsWithValue(sizeof(sent_data)));
+
+  char received_data[0] = {};
+
+  struct iovec iov;
+  iov.iov_base = received_data;
+  iov.iov_len = sizeof(received_data);
+  struct msghdr msg = {};
+  msg.msg_flags = -1;
+  msg.msg_iov = &iov;
+  msg.msg_iovlen = 1;
+
+  ASSERT_THAT(
+      RetryEINTR(recvmsg)(sockets->second_fd(), &msg, MSG_TRUNC | MSG_PEEK),
+      SyscallSucceedsWithValue(0));
+
+  // Check that msghdr flags were cleared (MSG_TRUNC was not set).
+  ASSERT_EQ(msg.msg_flags & MSG_TRUNC, 0);
 }
 
 TEST_P(StreamSocketPairTest, MsgTrunc) {