Re: [Xen-devel] xl errors if /var/log/xen doesn't exist

[Ian Campbell <Ian.Campbell@xxxxxxxxxx>]

create ^
thanks

Thanks for the report, I've created a bug for tracking purposes.

On Mon, 2014-01-13 at 14:32 +0000, Andrew Cooper wrote:
> Hello,
> 
> As part of running XenRT on a XenServer experimental build on top of
> Xen-4.4.0-rc1, I encountered the following error which is far from helpful:
> 
> $ xl create /root/minios.cfg
> xl: fatal error: xl_cmdimpl.c:457: No such file or directory: logfile =
> open(fullname, O_WRONLY|O_CREAT|O_APPEND, 0644)
> Parsing config from /root/minios.cfg
> libxl: error: libxl_exec.c:118:libxl_report_child_exitstatus:
> daemonizing child [32476] exited with error status 3
> Parsing config from /root/minios.cfg
> 
> The root cause of the issue is that the directly /var/log/xen didn't
> exist

Right, I do think we ought to cope with this a bit more gracefully.

>  (because I messed up the packaging).
> 
> Because of the CHK_SYSCALL() macro which prints a string-ized version of
> its argument, the error message omits the key detail (i.e. which file
> doesn't exist).

Oops!

> Furthermore, the string 'fullname' is a complete path
> which has been returned from "libxl_create_logfile()" without error.
> 
> If the failure to create a logfile is a fatal error (which itself is
> probably not correct - what about a RO dom0 which strictly logs to
> rsyslog?),

Even a R/O dom0 would need a writeable /var I think -- otherwise xl's
behaviour is the least of your worries. A tmpfs for example.

In any case presumably that use case would warrant making further
changes in order to have the option to direct these logs to syslog too.

>  then either xl or libxl should assume the responsibility of
> ensuring that its logging directory actually exists, and
> "libxl_create_logfile()" seems like a prime candidate to assume this
> responsibility.

It sounds like it, yes.

Ian.


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel