Bug#879603: systemd: network service enters failed state when client goes away too fast

[Julien Cristau]

On 10/23/2017 01:31 PM, Michael Biebl wrote:
> Am 23.10.2017 um 12:21 schrieb Julien Cristau:
>> Package: systemd
>> Version: 232-25+deb9u1
>> User: debian-admin at lists.debian.org
>> Usertag: needed-by-DSA-Team
>>
>> It looks like systemd has a race condition where instead of giving up
>> gracefully when a client goes away, it enters a bogus failure mode.
> 
> Please share the complete service file you are using and describe in
> more detail what's happening.
> Can you provide a minimal test case which would allow to reproduce the
> error and explain what kind of behaviour you expect.
> 
1) service file

# /etc/systemd/system/vsftpd-security at .service
[Unit]
Description=vsftpd security

[Service]
ExecStart=-/usr/sbin/vsftpd /etc/vsftpd-security.conf
StandardInput=socket
StandardError=journal
CapabilityBoundingSet=CAP_SYS_CHROOT CAP_SETUID CAP_SETGID
PrivateDevices=true
ProtectHome=true
ProtectSystem=full

# /etc/systemd/system/vsftpd-security.socket
[Unit]
Description=vsftpd security (socket)

[Socket]
ListenStream=149.20.4.14:21
ListenStream=[2001:4f8:1:c::14]:21
Accept=true
FreeBind=true
MaxConnections=200

[Install]
WantedBy=sockets.target

2) what I expect is for the service not to end up in a failed state for
something which, as far as I can tell, is not an error condition.  I
expect that if a remote client goes away before systemd can exec vsftpd
(which is what the error messages hint at; if that's not what's going
on, then there's a separate bug that the errors are misleading), it
should treat that as business as usual, and basically ignore it.

Cheers,
Julien