netbird/gvisor
0
0
Fork 0
mirror of https://github.com/netbirdio/gvisor.git synced 2026-05-22 17:12:49 -07:00
Code Issues Packages Projects Releases Wiki Activity

Try to avoid accept, incoming ACK race

Browse Source 
This test checks if an incoming ACK is dropped by the listener when the
accept queue is full. The ACK receive handling could race with the test
invoking accept on the DUT, causing the test to be flaky. Add a wait
time before invoking accept on the DUT to give cycles for the incoming
ACK to be handled/dropped by the listener.

PiperOrigin-RevId: 378770225
This commit is contained in:
Mithun Iyer
2021-06-10 17:23:34 -07:00
committed by gVisor bot
parent 3fcbad5093
commit 3c91fa42ae
1 changed files with 11 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/packetimpact/tests/tcp_listen_backlog_test.go
+11 -3
View File
@@ -55,15 +55,23 @@ func TestTCPListenBacklog(t *testing.T) {
// Send the ACK to complete handshake.
establishedConn.Send(t, testbench.TCP{Flags: testbench.TCPFlags(header.TCPFlagAck)})
// Poll for the established connection ready for accept.
dut.PollOne(t, listenFd, unix.POLLIN, time.Second)
// Send the ACK to complete handshake, expect this to be ignored by the
// listener.
// Send the ACK to complete handshake, expect this to be dropped by the
// listener as the accept queue would be full because of the previous
// handshake.
incompleteConn.Send(t, testbench.TCP{Flags: testbench.TCPFlags(header.TCPFlagAck)})
// Let the test wait for sometime so that the ACK is indeed dropped by
// the listener. Without such a wait, the DUT accept can race with
// ACK handling (dropping) causing the test to be flaky.
time.Sleep(100 * time.Millisecond)
// Drain the accept queue to enable poll for subsequent connections on the
// listener.
dut.Accept(t, listenFd)
fd, _ := dut.Accept(t, listenFd)
dut.Close(t, fd)
// The ACK for the incomplete connection should be ignored by the
// listening endpoint and the poll on listener should now time out.