From: [email protected]

Hi Mark,

snmpd segfaults on a simple query on CPU usage query (OID .1.3.6.1.4.1.2021.11.52.0)

open("/proc/stat", O_RDONLY) = 12

Kernel: Linux 2.6.7

Please retest this with 5.1.2-1. This seems to be the same as #244721,
however snmpd still shouldn't crash in such cases.

Thanks,
Jochen


--
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]

--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)

From: [email protected]

Jochen Friedrich wrote:

Please retest this with 5.1.2-1. This seems to be the same as #244721, however snmpd still shouldn't crash in such cases.

5.1.2-1 didn't fix it.

It isn't the same as #244721 btw, as 5.1-5 is fine (i.e. doesn't crash,
and responds to the queries normally), but does output these (annoying) messages.

--
Mark

[email protected]


--
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]

--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)

From: [email protected]

Hi Mark,

It isn't the same as #244721 btw, as 5.1-5 is fine (i.e. doesn't crash,
and responds to the queries normally), but does output these (annoying) messages.

Hmm, all i can think of in this case is lm-sensors (again!). From the
strace, it looks like some global memory has been overwritten. The main
change between 5.1-5 and later versions of net-snmp is the re-enabled
support for lm-sensors (except for several bugfixes from upstream).

Do you have a chance to recompile net-snmp without lm-sensors support and
check the result?

Just download the sources (apt-get source snmpd),
comment out the following 4 lines in debian/rules:

ifeq "$(ARCH)" "i386"
MIB_MODULES += ucd-snmp/lmSensors
LIBSENSORS = -lsensors
endif

And then run "dpkg-buildpackage -us -uc" in the source tree.

Thanks,
Jochen


--
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]

--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)

From: [email protected]

Jochen Friedrich wrote:

Hmm, all i can think of in this case is lm-sensors (again!). From the
strace, it looks like some global memory has been overwritten. The main change between 5.1-5 and later versions of net-snmp is the re-enabled
support for lm-sensors (except for several bugfixes from upstream).

Do you have a chance to recompile net-snmp without lm-sensors support and check the result?

Yes, that seems to have done the trick. 5.1.2-1 without lm-sensors is
running without any problems now.

--
Mark

[email protected]


--
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]

--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)

From: [email protected]

tag 264442 -sid
tag 264442 +experimental upstream
thanks

Hi Mark,

Yes, that seems to have done the trick. 5.1.2-1 without lm-sensors is
running without any problems now.

I just uploaded 5.1.2-2 without lm-sensors support to unstable and
5.1.2-2exp1 with lm-sensors and diskio to experimental to further chasing
the bug without affecting sarge. Apparently, lm-sensors is still to buggy
to go into a stable release.

Thanks,
Jochen


--
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]

--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)