podman create: Unable to close conn: \"close unixpacket @->/proc/self/fd/11/attach: use of closed network connection

[edsantiago]

New flake exposed by #11776:

# # podman create docker-archive:/tmp/podman_bats.lhLlzc/archive.tar
...
# dea245e90d8c9b4a645a71d9c20b8e80a3b913e3c41b5327016fe6d9a421533d
# # podman start --attach dea245e90d8c9b4a645a71d9c20b8e80a3b913e3c41b5327016fe6d9a421533d
# h6XQEUDR3lEH7VjgGUcd
# time="2021-10-01T07:26:23-05:00" level=error msg="Unable to close conn: \"close unixpacket @->/proc/self/fd/11/attach: use of closed network connection\""
# #/vvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvv
# #|     FAIL: 'podman run' of 'podman-create docker-archive'
# #| expected: 'h6XQEUDR3lEH7VjgGUcd'
# #|   actual: 'h6XQEUDR3lEH7VjgGUcd'
# #|         > 'time="2021-10-01T07:26:23-05:00" level=error msg="Unable to close conn: \"close unixpacket @->/proc/self/fd/11/attach: use of closed network connection\""'
# #\^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

sys: podman run docker-archive

• fedora-33 : sys podman fedora-33 rootless host
  • PR Pod Events Logging Fix #11816
    • 10-01 16:04
• fedora-34 : sys podman fedora-34 root host
  • PR pkg/specgen: cache image in generator #11808
    • 10-01 08:45

Looking through cirrus-flake-grep, I see this happening as far back as June (when I started collecting logs). This is an old bug, not a new one.

[edsantiago]

another one

[Luap99]

The code will call close twice.

podman/libpod/oci_attach_linux.go

Lines 220 to 224 in c963a50

	if err == nil {
	if connErr := conn.CloseWrite(); connErr != nil {
	logrus.Errorf("Unable to close conn: %q", connErr)
	}
	}

podman/libpod/oci_attach_linux.go

Lines 163 to 167 in c963a50

	defer func() {
	if err := conn.Close(); err != nil {
	logrus.Errorf("Unable to close socket: %q", err)
	}
	}()

If CloseWrite is called before Close, everything is fine but if Close was called first, CloseWrite will fail.
I think we can just remove the CloseWrite call.