Re: [syslog-ng] TCP recv bug in syslog-ng v2.09?
Yeah, now the only problem is all the ancient boxes with no new 12 series code or outdated SMARTNet agreements. But those are a problem regardless of this issue! On Thu, Aug 19, 2010 at 04:05:11PM -0400, Clayton Dukes wrote:
Aha! Thanks for pointing me to my own company's documentation! hehe. So, it does appear as though they have provided a fix in later versions of the code. That is probably about the best we can ask for, and in my case, it's not a terrible option since we're running this on IP SLA shadow routers which can easily be upgraded.
A HUGE thank you!
______________________________________________________________
Clayton Dukes ______________________________________________________________
On Thu, Aug 19, 2010 at 3:50 PM, Matthew Hall <mhall@mhcomputing.net> wrote:
It's a pretty famous problem among the logging daemon maintainers.
Many logging daemons are struggling with this issue, and they have come up with various hacks to try to split out the broken messages using timing since they can't always get a '\n'.
Fortunately there is a fix available in recent firmware it appears.
To append a line feed character at the end of each syslog message over TCP as delimiter, use the logging delimiter tcp command. To turn off the delimiter for Syslog over TCP, use the no form of this command.
logging delimiter tcp no logging delimiter tcp
By default, the logging delimiter tcp function is enabled.
This command was introduced in 12.4(24)T.
Usage Guidelines
You can use the logging delimiter tcp to append a line feed character at the end of each syslog message over TCP as the delimiter.
Since TCP is a stream protocol, the delimiter helps the external TCP syslog listeners to accurately detect the end of each syslog message from the incoming TCP stream. In case the external TCP Syslog listeners do not work well with the delimiter, use the no form of this command to turn it off.
On Thu, Aug 19, 2010 at 03:44:12PM -0400, Clayton Dukes wrote:
lol, thanks :-) I'm actually a bit surprised that nobody has said anything before (like our customers). Maybe they have, but I did a bug search and didn't find anything. I'll have to dig deeper.
I know the concept of tcp syslog is still relatively new, so maybe it just hasn't come up yet. I'll still have to see if what they have was done on purpose/by design or if it is, in fact, a bug (or design flaw?). I dunno, hopefully someone smarter than me can answer that, which shouldn't be too hard, heh.
______________________________________________________________
Clayton Dukes ______________________________________________________________
On Thu, Aug 19, 2010 at 3:26 PM, Matthew Hall <mhall@mhcomputing.net> wrote:
Thanks Clayton,
I know how to explain things for networking people because I used to work at HP ProCurve. ;-)
I really hope you can get this fixed because I think it will cause me problems at my new job.
Matthew.
On Thu, Aug 19, 2010 at 03:13:24PM -0400, Clayton Dukes wrote:
Matt, Thanks for that very good explanation, it's good ammo for me to approach the developers @ Cisco with :-)
______________________________________________________________
Clayton Dukes ______________________________________________________________
On Thu, Aug 19, 2010 at 1:21 PM, Matthew Hall <mhall@mhcomputing.net
wrote:
This understanding is Not quite right. What we are talking about
here is
the difference between SOCK_STREAM and SOCK_DGRAM socket semantics.
In a SOCK_DGRAM each PDU or segment is considered to be a separate atomic message, and each is delivered to the application separately. For a SOCK_DGRAM, a connection just means you have bound the socket to a specific remote address. It does not open and close for each message because it is not connection oriented.
In a SOCK_STREAM each PDU or segment is considered to be additional bytes in a continuous flow of data. The application which receives data from a SOCK_STREAM has no way of seeing each PDU or segment individually, because the PDUs or segments are reassembled inside the kernel.
There is no standard way of simply receiving each SOCK_STREAM packet separately because there is no relationship between the size of each PDU or segment and the size of the message. The size of the message is defined by the '\n'.
This difference is also seen in UDP DNS versus TCP DNS. TCP DNS sends extra fields in the packets to keep track of the message sizes and UDP DNS does not.
Matthew.
On Thu, Aug 19, 2010 at 11:34:39AM -0400, Clayton Dukes wrote: > bing! (light bulb just went off). > I wasn't thinking about the whole established vs. non established connection > thing. > I couldn't figure out why, if UDP wasn't sending a newline, it wasn't > causing the same problem. I wasn't thinking about the fact that the > connection was closing, thus ending the stream. > > Thanks for the help! > > ______________________________________________________________ > > Clayton Dukes > ______________________________________________________________ > > > On Thu, Aug 19, 2010 at 11:29 AM, <syslogng@feystorm.net> wrote: > > > I explained it already :-) > > When the message comes in over TCP and doesnt end with a newline, syslog-ng > > assumes the message is going to be continued in another packet. When the > > cumulative total of all the messages exceeds the max message size it flushes > > the buffer out and you get all the messages mashed together at once. > > You can try filing a bug report on bugzilla.balabit.com and request a new > > flag or something that treats each packet on a tcp source as a separate > > message, but I'd say the problem is more cisco than syslog-ng since > > syslog-ng works fine with all other sources except cisco devices :-/ > > Look at it this way, every thing that sends logs out to tcp expects the > > receiving syslog daemon to treat a packet without a newline as a message to > > be continued in a later packet. If syslog-ng changed that default behavior, > > all these other things that expect the behavior would break. > > > > -Patrick > > > > > > Sent: Thursday, August 19, 2010 9:12:36 AM > > > > From: Clayton Dukes <cdukes@gmail.com> <cdukes@gmail.com> > > To: Syslog-ng users' and developers' mailing list > > <syslog-ng@lists.balabit.hu> <syslog-ng@lists.balabit.hu> > > Subject: Re: [syslog-ng] TCP recv bug in syslog-ng v2.09? > > > > Following up on this: > > > > Interestingly, however, is that when I use tcpdump, I DO see each message > > coming in via TCP...so now I'm leaning back towards syslog-ng being the > > problem. > > > > > > > > Here's what I can see: > > > > > > > > UDP: > > > > #tcpdump -vvv host 14.3.23.50 -i eth0 > > > > tcpdump: listening on eth0, link-type EN10MB (Ethernet), capture size 96 > > bytes > > > > 09:13:31.465239 IP (tos 0xb8, ttl 251, id 1323, offset 0, flags [none], > > proto UDP (17), length 134) > > > > 14.3.23.50.51526 > server.x.com.syslog: SYSLOG, length: 106 > > > > Facility local7 (23), Severity notice (5) > > > > Msg: 2975: *Aug 19 12:34:19.465: %SYS-5-CONFIG_I[|syslog] > > > > > > > > > > > > TCP: > > > > #tcpdump -vvv host 14.3.23.50 -i eth0 > > > > 09:46:29.902063 IP (tos 0x0, ttl 251, id 12253, offset 0, flags [none], > > proto TCP (6), length 146) > > > > 14.3.23.50.31746 > server.x.com.601: Flags [.], seq 233:339, ack 1, win > > 4128, length 106 > > > > 09:46:29.902077 IP (tos 0x0, ttl 64, id 27779, offset 0, flags [DF], proto > > TCP (6), length 40) > > > > server.x.com.601 > 14.3.23.50.31746: Flags [.], cksum 0xa3bb (correct), > > seq 1, ack 339, win 5840, length 0 > > > > > > > > > > > > So, it looks like the syslog message does end, but why is syslog-ng > > buffering and showing multiple TCP-based messages as a single message? > > > > Do I have something misconfigured? > > > > > > > > > > ______________________________________________________________ > > > > Clayton Dukes > > ______________________________________________________________ > > > > > > On Tue, Aug 17, 2010 at 3:14 PM, Clayton Dukes < cdukes@gmail.com> wrote: > > > >> yikes! > >> seriously? > >> Guess I'll have to file a bug internally :-) > >> Can someone else positively verify this? > >> Or any suggestions on how I can so that we can recreate it in a lab? > >> > >> ______________________________________________________________ > >> > >> Clayton Dukes > >> ______________________________________________________________ > >> > >> > >> > >> On Tue, Aug 17, 2010 at 2:52 PM, <syslogng@feystorm.net> wrote: > >> > >>> If I recall correctly its because cisco equipment doesnt terminate its > >>> log entries with newlines, so when sending via TCP, syslog-ng thinks the > >>> message is going to be continued in another packet (UDP is assumed to be 1 > >>> packet per log entry). > >>> The only way to fix this is an ugly hack to set the timeout so that when > >>> it doesnt get a reply within a certain time, it assumes the log entry ended. > >>> but if several log entries are sent within the timeout, then they'll all be > >>> mashed together into 1 syslog-ng entry. > >>> > >>> > >>> Sent: Tuesday, August 17, 2010 12:28:28 PM > >>> From: Clayton Dukes <cdukes@gmail.com> <cdukes@gmail.com> > >>> To: Syslog-ng users' and developers' mailing list > >>> <syslog-ng@lists.balabit.hu> <syslog-ng@lists.balabit.hu> > >>> Subject: [syslog-ng] TCP recv bug in syslog-ng v2.09? > >>> > >>> Hey guys, > >>> Are there any known bugs for syslog-ng v2.09 that won't allow a cisco > >>> router to send logs over tcp? > >>> I can see a connection established in syslog-ng. > >>> I also see the message come in via tcpdump, but nothing in syslog-ng's > >>> output. > >>> If I change the router from tcp to udp, messages come in as expected. > >>> > >>> *Router config:* > >>> > >>> logging source-interface Loopback0 > >>> logging 172.18.224.150 > >>> logging host 172.18.224.190 transport tcp > >>> > >>> > >>> *syslog-ng config:* > >>> > >>> source s_all { > >>> udp(); > >>> tcp(ip(11.31.130.99) port(8002) max-connections(300)); > >>> tcp(ip(172.18.224.190) port(601) max-connections(300)); > >>> }; > >>> > >>> > >>> *debug output:* > >>> I commented out the line above for the other interface (11.31.130.99), > >>> restarted and this is all I see: > >>> Syslog connection accepted; from='AF_INET(14.3.23.50:63845 )', > >>> to='AF_INET(172.18.224.190:601)' > >>> > >>> > >>> *tcpdump:* > >>> > >>> 14:13:46.914566 IP (tos 0x0, ttl 251, id 4303, offset 0, flags [none], > >>> proto TCP (6), length 134) > >>> 14.3.23.50.63845 > xxx.com.601: Flags [.], seq 230:324, ack 1, win > >>> 4128, length 94 > >>> > >>> > >>> *Router debug:* > >>> > >>> *Aug 17 17:34:19.772: %SYS-5-CONFIG_I: Configured from console by pnoc > >>> on vty0 (172.18.224.151) > >>> *Aug 17 17:34:20.776: Released port 15205 in Transport Port Agent for > >>> TCP IP type 1 delay 240000 > >>> *Aug 17 17:34:20.776: TCB 0x850F9754 destroyed > >>> *Aug 17 17:34:25.775: TCB83648E60 created > >>> *Aug 17 17:34:25.775: TCB83648E60 setting property TCP_PID (8) 845083E4 > >>> *Aug 17 17:34:25.775: TCB83648E60 setting property TCP_NO_DELAY (1) > >>> 845083E8 > >>> *Aug 17 17:34:25.775: TCB83648E60 setting property TCP keepalive timeout > >>> (17) 845084A0 > >>> *Aug 17 17:34:25.775: TCP: Random local port generated 63845, network 1 > >>> *Aug 17 17:34:25.775: TCB83648E60 bound to 14.3.23.50.63845 > >>> *Aug 17 17:34:25.775: Reserved port 63845 in Transport Port Agent for > >>> TCP IP type 1 > >>> *Aug 17 17:34:25.775: TCP: sending SYN, seq 3300233565, ack 0 > >>> *Aug 17 17:34:25.775: TCP0: Connection to 172.18.224.190:601 , > >>> advertising MSS 536 > >>> *Aug 17 17:34:25.775: TCP0: state was CLOSED -> SYNSENT [63845 -> > >>> 172.18.224.190(601)] > >>> *Aug 17 17:34:25.779: TCP0: state was SYNSENT -> ESTAB [63845 -> > >>> 172.18.224.190(601)] > >>> *Aug 17 17:34:25.779: TCP: tcb 83648E60 connection to 172.18.224.190:601, > >>> peer MSS 1460, MSS is 536 > >>> *Aug 17 17:34:25.779: TCB83648E60 connected to 172.18.224.190.601 > >>> *Aug 17 17:34:25.779: %SYS-6-LOGGINGHOST_STARTSTOP: Logging to host > >>> 172.18.224.190 port 601 started - reconnection > >>> > >>>
> >>> > >>> Clayton Dukes > >>>
> >>> > >>> ------------------------------ > >>> > >>>
> >>> Member info: https://lists.balabit.hu/mailman/listinfo/syslog-ng > >>> Documentation: http://www.balabit.com/support/documentation/?product=syslog-ng > >>> FAQ: http://www.campin.net/syslog-ng/faq.html > >>> > >>> > >>> > >>>
> >>> Member info: https://lists.balabit.hu/mailman/listinfo/syslog-ng > >>> Documentation: > >>> http://www.balabit.com/support/documentation/?product=syslog-ng > >>> FAQ: http://www.campin.net/syslog-ng/faq.html > >>> > >>> > >>> > >> > > ------------------------------ > > > >
> > Member info: https://lists.balabit.hu/mailman/listinfo/syslog-ng > > Documentation: http://www.balabit.com/support/documentation/?product=syslog-ng > > FAQ: http://www.campin.net/syslog-ng/faq.html > > > > > > > >
> > Member info: https://lists.balabit.hu/mailman/listinfo/syslog-ng > > Documentation: > > http://www.balabit.com/support/documentation/?product=syslog-ng > > FAQ: http://www.campin.net/syslog-ng/faq.html > > > > > >
>
> Member info: https://lists.balabit.hu/mailman/listinfo/syslog-ng > Documentation: http://www.balabit.com/support/documentation/?product=syslog-ng > FAQ: http://www.campin.net/syslog-ng/faq.html >
participants (1)
-
Matthew Hall