summaryrefslogtreecommitdiffhomepage
diff options
context:
space:
mode:
authorIan Lewis <ianlewis@google.com>2020-01-08 16:35:43 -0800
committergVisor bot <gvisor-bot@google.com>2020-01-08 17:24:05 -0800
commitfbb2c008e26a7e9d860f6cbf796ea7c375858502 (patch)
tree698e17f595ea9a689579cf009011171c4bbdcf9a
parent565b64148314018e1234196182b55c4f01772e77 (diff)
Return correct length with MSG_TRUNC for unix sockets.
This change calls a new Truncate method on the EndpointReader in RecvMsg for both netlink and unix sockets. This allows readers such as sockets to peek at the length of data without actually reading it to a buffer. Fixes #993 #1240 PiperOrigin-RevId: 288800167
-rw-r--r--pkg/sentry/socket/netlink/BUILD1
-rw-r--r--pkg/sentry/socket/netlink/socket.go29
-rw-r--r--pkg/sentry/socket/unix/io.go13
-rw-r--r--pkg/sentry/socket/unix/unix.go23
-rw-r--r--test/syscalls/linux/BUILD1
-rw-r--r--test/syscalls/linux/socket_non_stream.cc113
-rw-r--r--test/syscalls/linux/socket_non_stream_blocking.cc37
-rw-r--r--test/syscalls/linux/socket_stream.cc55
8 files changed, 250 insertions, 22 deletions
diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD
index 79589e3c8..136821963 100644
--- a/pkg/sentry/socket/netlink/BUILD
+++ b/pkg/sentry/socket/netlink/BUILD
@@ -22,7 +22,6 @@ go_library(
"//pkg/sentry/kernel",
"//pkg/sentry/kernel/auth",
"//pkg/sentry/kernel/time",
- "//pkg/sentry/safemem",
"//pkg/sentry/socket",
"//pkg/sentry/socket/netlink/port",
"//pkg/sentry/socket/unix",
diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go
index 4a1b87a9a..d2e3644a6 100644
--- a/pkg/sentry/socket/netlink/socket.go
+++ b/pkg/sentry/socket/netlink/socket.go
@@ -29,7 +29,6 @@ import (
"gvisor.dev/gvisor/pkg/sentry/kernel"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time"
- "gvisor.dev/gvisor/pkg/sentry/safemem"
"gvisor.dev/gvisor/pkg/sentry/socket"
"gvisor.dev/gvisor/pkg/sentry/socket/netlink/port"
"gvisor.dev/gvisor/pkg/sentry/socket/unix"
@@ -500,29 +499,29 @@ func (s *Socket) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, have
trunc := flags&linux.MSG_TRUNC != 0
r := unix.EndpointReader{
+ Ctx: t,
Endpoint: s.ep,
Peek: flags&linux.MSG_PEEK != 0,
}
+ doRead := func() (int64, error) {
+ return dst.CopyOutFrom(t, &r)
+ }
+
// If MSG_TRUNC is set with a zero byte destination then we still need
// to read the message and discard it, or in the case where MSG_PEEK is
// set, leave it be. In both cases the full message length must be
- // returned. However, the memory manager for the destination will not read
- // the endpoint if the destination is zero length.
- //
- // In order for the endpoint to be read when the destination size is zero,
- // we must cause a read of the endpoint by using a separate fake zero
- // length block sequence and calling the EndpointReader directly.
+ // returned.
if trunc && dst.Addrs.NumBytes() == 0 {
- // Perform a read to a zero byte block sequence. We can ignore the
- // original destination since it was zero bytes. The length returned by
- // ReadToBlocks is ignored and we return the full message length to comply
- // with MSG_TRUNC.
- _, err := r.ReadToBlocks(safemem.BlockSeqOf(safemem.BlockFromSafeSlice(make([]byte, 0))))
- return int(r.MsgSize), linux.MSG_TRUNC, from, fromLen, socket.ControlMessages{}, syserr.FromError(err)
+ doRead = func() (int64, error) {
+ err := r.Truncate()
+ // Always return zero for bytes read since the destination size is
+ // zero.
+ return 0, err
+ }
}
- if n, err := dst.CopyOutFrom(t, &r); err != syserror.ErrWouldBlock || flags&linux.MSG_DONTWAIT != 0 {
+ if n, err := doRead(); err != syserror.ErrWouldBlock || flags&linux.MSG_DONTWAIT != 0 {
var mflags int
if n < int64(r.MsgSize) {
mflags |= linux.MSG_TRUNC
@@ -540,7 +539,7 @@ func (s *Socket) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, have
defer s.EventUnregister(&e)
for {
- if n, err := dst.CopyOutFrom(t, &r); err != syserror.ErrWouldBlock {
+ if n, err := doRead(); err != syserror.ErrWouldBlock {
var mflags int
if n < int64(r.MsgSize) {
mflags |= linux.MSG_TRUNC
diff --git a/pkg/sentry/socket/unix/io.go b/pkg/sentry/socket/unix/io.go
index 2ec1a662d..2447f24ef 100644
--- a/pkg/sentry/socket/unix/io.go
+++ b/pkg/sentry/socket/unix/io.go
@@ -83,6 +83,19 @@ type EndpointReader struct {
ControlTrunc bool
}
+// Truncate calls RecvMsg on the endpoint without writing to a destination.
+func (r *EndpointReader) Truncate() error {
+ // Ignore bytes read since it will always be zero.
+ _, ms, c, ct, err := r.Endpoint.RecvMsg(r.Ctx, [][]byte{}, r.Creds, r.NumRights, r.Peek, r.From)
+ r.Control = c
+ r.ControlTrunc = ct
+ r.MsgSize = ms
+ if err != nil {
+ return err.ToError()
+ }
+ return nil
+}
+
// ReadToBlocks implements safemem.Reader.ReadToBlocks.
func (r *EndpointReader) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) {
return safemem.FromVecReaderFunc{func(bufs [][]byte) (int64, error) {
diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go
index 885758054..91effe89a 100644
--- a/pkg/sentry/socket/unix/unix.go
+++ b/pkg/sentry/socket/unix/unix.go
@@ -544,8 +544,27 @@ func (s *SocketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags
if senderRequested {
r.From = &tcpip.FullAddress{}
}
+
+ doRead := func() (int64, error) {
+ return dst.CopyOutFrom(t, &r)
+ }
+
+ // If MSG_TRUNC is set with a zero byte destination then we still need
+ // to read the message and discard it, or in the case where MSG_PEEK is
+ // set, leave it be. In both cases the full message length must be
+ // returned.
+ if trunc && dst.Addrs.NumBytes() == 0 {
+ doRead = func() (int64, error) {
+ err := r.Truncate()
+ // Always return zero for bytes read since the destination size is
+ // zero.
+ return 0, err
+ }
+
+ }
+
var total int64
- if n, err := dst.CopyOutFrom(t, &r); err != syserror.ErrWouldBlock || dontWait {
+ if n, err := doRead(); err != syserror.ErrWouldBlock || dontWait {
var from linux.SockAddr
var fromLen uint32
if r.From != nil && len([]byte(r.From.Addr)) != 0 {
@@ -580,7 +599,7 @@ func (s *SocketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags
defer s.EventUnregister(&e)
for {
- if n, err := dst.CopyOutFrom(t, &r); err != syserror.ErrWouldBlock {
+ if n, err := doRead(); err != syserror.ErrWouldBlock {
var from linux.SockAddr
var fromLen uint32
if r.From != nil {
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/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) {