Bug 1683388

Description of problem:

When starting the nginx service via systemd using the provided unit file, the log files created in /var/log/nginx are owned by root:root, even though the log directory is owned by nginx:nginx (mode: 700). However, the default logrotate config (/etc/logrotate.d/nginx) that comes with the nginx package creates log files as nginx:nginx and then sends a signal to nginx's process to reopen the logs as the nginx user like so:

postrotate
        /bin/kill -USR1 `cat /run/nginx.pid 2>/dev/null` 2>/dev/null || true
endscript


Therefore, the ownership of the /var/log/nginx directory cannot be changed to a user other than nginx or else logrotate will fail to reopen the logs as the nginx user, and show this in the error.log:

... [emerg] 31850#0: open() "/var/log/nginx/error.log" failed (13: Permission denied)
... [emerg] 31850#0: open() "/var/log/nginx/access.log" failed (13: Permission denied)

The service will still work fine if you just started it with systemctl since the logs are then opened as root.

At best, both systemctl and logrotate should create and open the log files with the same ownership and same user.

Version-Release number of selected component (if applicable):

Tested with nginx-1.12.2-2.el7.x86_64, but did not validate with previous versions.


How reproducible:

Very easily.

Steps to Reproduce:
1. Install the epel/x86_64 yum repository
2. Install nginx package
3. Check ownership of /var/log/nginx (should be mode 700/nginx:nginx)
4. Execute: systemctl start nginx
5. Check ownership of log files in /var/log/nginx (should be mode 644/root:root)
6. Execute: curl -I http://localhost at least once
7. Confirm it works. Should get a response and the access.log file should contain lines confirming your access.
8. Change the ownership of the /var/log/nginx directory to a user other than nginx (could be any)
9. Execute: systemctl restart nginx
10. Run steps 6-7 again, and confirm it still works
11. Force a log rotation: logrorate -f /etc/logrotate.d/nginx -v
12. Check the log files in /var/log/nginx. You should see new log files owned by nginx:nginx, and the error.log file should contain the lines mentioned in the description confirming the nginx user is unable to open the files.
13. Execute the curl commands again and confirm it still works, but no logs are created anymore.
14. Change the ownership of /var/log/nginx back to the nginx user
15. Force a log rotation again and confirm that it was able to (new error.log file should be empty and if you execute curl again you should see entries in acess.log)


Actual results:

(See above)

Expected results:

(See above)

Additional info:

N/A