Hi Hugo,

I've been spining around my email box today and I found this. Please,
next time cc: netfilter-devel.

Hugo Mildenberger wrote:
> Pablo,
> 
> thanks for checking in timer.h.  Attached you'll find a 
> patch which fixes the logfile-reopen problem after 
> sending a SIGHUP to ulogd. The reason was that 
> config_parse_file() stuffed a stack-allocated string 
> named 'args' (stemming from wordbuf) to logfile_open  
> which in turn got assigned to a global variable 
> named 'ulogd_logfile'. 

I have applied the following patch based on yours, the credits are
stored in git changelog so IMO there's no need to keep it in the C file.

> There is reason to  assume that there are more issues 
> like this during module configuration. Currently I have 
> not yet understood which modules might be affected.
> 
> 
> Some other issues:
> 
> 1.) Perhaps using inotify instead of (or in addition to) 
> SIGUSR1?

I don't think that we need such thing, I prefer that the sysadmin have
to explicitly order a reload of the configuration file.

> 2.) grepping through kernel modules, I found in 
> nfnetlink_log.c:470 that beneath socket's fuid also 
> fgid is transmitted, while ulogd_inppkt_NFLOG.c
> and printpkt.c only want to know about uid.

I think that Eric has fixed this a one of his patchsets.

> 3.) For some reasons, I would be happy to see also the 
> pid of the socket owning process appearing at least in 
> a mysql table. I know that the somehow related "owner 
> match"  is marked "broken on SMP", but the code for 
> non-smp machines is still in place. Could you just give 
> me a hint, what the problem on smp regarding pid/owner 
> match really is, and if there are reason not to 
> transmit this information via nfnetlink_log.c.

See patch: "Remove tasklist_lock abuse in ipt{,6}owner" from Patrick
McHardy. The owner match in linux kernel <= 2.6.14 gets semaphore in a
BH which can produce a deadlock in SMP.

-- 
"Los honestos son inadaptados sociales" -- Les Luthiers