Bug 1812032

Description of problem:
The logger (/usr/bin/logger from util-linux-2.23.2-59.el7.x86_64) has potential security issue on a way it handles the TCP connections to the syslog server. 
If the backend server is restarted logger will for some time be trying to send the messages to a socket, which is already closed from the other side.
I have discovered this issue when trying to integrate Apache logs using pipe to logger to log to central syslog server.


Version-Release number of selected component (if applicable):
util-linux-2.23.2-59.el7.x86_64


How reproducible:
100%


Steps to Reproduce:
1. run the fake TCP syslog server
Session1_Server: 
nc -l -p 9514

2. open the connection to the fake syslog server from logger:
Session2_Client_logger:
cat | TZ=GMT /usr/bin/logger --tag myserver --tcp --port 9514 --priority local0.warn --server localhost

3. Log AAA in Session2 to test connection is working
Session2_Client_logger:
AAA

Session1_Server:
<132>Mar 10 11:53:07 mytestag: AAA

4. Now when the session is established restart the server:
Session1_Server:
Ctrl-C
nc -l -p 9514

5. Try to send some more on the logger side:
Session2_Client_logger:
BBB
CCC
DDD


Actual results:
The situation is actually worse than if I would use for the same scenario UDP. Using TCP for this scenario I have lost 3 log messages. 

After third line/message the logger quits with the error code 141.
It actually doesn't write any message to the error output, so when used for example in Apache pipe to log it is really hard to detect that this happened as it is not writing anything to the Apache error_log.


Expected results:
When using TCP I would expect that I won't be loosing any log messages.
I would expect that the logger would try to re-establish the broken TCP or quits as soon as the TCP session is closed (between sending AAA and BBB to the logger). 


Additional info:
In Fedora 31 (util-linux-2.34-4.fc31.x86_64) the logger quits after sending another message - losing 1 log message in the process of syslog server restart.