Created attachment 157101 [details]
patch to correct the bug

Hi,

This bug report is about the four rsyslog ports: sysutils/rsyslog8, sysutils/rsyslog7, sysutils/rsyslog8-devel, sysutils/rsyslog7-devel.

When rsyslog is started at boot time (for instance by setting rsyslogd_enable="YES" in rc.conf(5), by adding "service rsyslogd onestart" in rc.local(8), or by simply adding the daemon path and args in rc.local(8)), and when rsyslog is simultaneously configured to output some streams to /dev/console, the daemon will start correctly but, at the near end of the boot sequence of the operating system, it will fail permanently, in an endless loop, using 100% CPU.

Attached to this bug report, please find a patch to correct this behaviour. To add this patch to the port, just add the file to the "files" directory. Note that the patch is named *.c2 since another patch for the same file is currently discussed in bug report 200270.

Here is a complete explanation of the steps that make the bug happen:

1- when init(8) launches rc(8) scripts, rsyslogd starts

2- rsyslogd reads rsyslog.conf(5) and if some stuff must be logged to the console, because of a line like "*.err;kern.warning;auth.notice;mail.crit /dev/console" in the configuration file, the daemon calls open(2) to get a file descriptor to write on "/dev/console". It starts writing corresponding logs to this descriptor.

3- Later during the boot sequence, init(8) configures the console, and for this to be done, it starts by calling the revoke(2) syscall: revoke("/dev/console") is the first line of code in /usr/src/sbin/init/init.c:open_console().

4- Once /dev/console is revoked, further writes to any file descriptor previously opened on this file return -1 with ENXIO as errno.

5- thus, rsyslogd gets this error in runtime/Stream.c:doWriteCall(), and calls runtime/stream.c:tryTTYRecover() since the error occured on a tty.

6- but runtime/stream.c:tryTTYRecover() tries to reopen the tty only if the error is EIO on Linux or EBADF on any other operating system. Since the error is ENXIO, that is distinct from EBADF, runtime/stream.c:tryTTYRecover() returns RS_RET_OK and runtime/stream.c:doWriteCall() loops, endlessly.

Sincerely,