Please backport agent/mibgroup/host/hr_swrun.c to 5.4.1

[Sergey Svishchev]

There are two problems in snmpd 5.4.1~dfsg-12ubuntu7 (version currently in Karmic):

1) crash due to bad pointer dereference in hr_swrun.c:748. Kernel message looks like:

segfault at 1 ip 00007f984e0ee3e0 sp 00007fffe41e1270 error 4 in libnetsnmpmibs.so.15.1.0[7f984e0bd000+10d000]

2) hrSWRunPath may not be populated for all processes (/proc/.../status parsing is faulty)

Both problems are fixed in current code, but there's no 5.5 package yet.

[Chuck Short (zulcss)]

Is there any upstream bugs that are used to track these?

[Sergey Svishchev (svs)]

1) http://sourceforge.net/tracker/index.php?func=detail&aid=1774612&group_id=12694&atid=312694

2) http://net-snmp.svn.sourceforge.net/viewvc/net-snmp/trunk/net-snmp/agent/mibgroup/host/hr_swrun.c?r1=17265&r2=17460

"Proper check for empty process status file.
(Avoids truncated walk of hrSWRunPath on SuSE systems)
Reported by Kamil Kisiel."

[Chuck Short (zulcss)]

Im going to close this as this is fixed in lucid. However, this is probably a good candidate for an SRU.

regards
chuck

[Zach Wily (zach-zwily)]

I don't think this is fixed in Lucid. Lucid has version 5.4.2.1, but I think the change was committed upstream after that. From the source code for the package included in Lucid:

            if ((fp = fopen(string, "r")) == NULL)
                return NULL;
            fgets(buf, sizeof(buf), fp); /* Name: process name */
            if ( cp == NULL ) {
                fclose(fp);
                return NULL; /* the process probably died */
            }

If you look at the change linked to by Sergey (#2), you can see that it's not applied. Also, the hrSWRunPath is still not available.

[Ben Ritcey (i-launchpad-lists-ritcey-com)]

Agreed - hrSWRunPath is still unpopulated in Lucid w/ 5.4.2.1~dfsg0ubuntu1-0ubuntu2

[Sergey Svishchev (svs)]

I'd like to see this bug fixed in Lucid, it's affecting monitoring infrastructure that I use (false alarms like "no cron process present").

[Sergey Svishchev (svs)]

test case:

$ snmpbulkget -v 2c -c ... -Cr10 ... hrSWRunPath
HOST-RESOURCES-MIB::hrSWRunPath.1 = STRING: "/sbin/init"
HOST-RESOURCES-MIB::hrSWRunParameters.1 = ""
HOST-RESOURCES-MIB::hrSWRunParameters.2 = ""
HOST-RESOURCES-MIB::hrSWRunParameters.3 = ""
HOST-RESOURCES-MIB::hrSWRunParameters.4 = ""
HOST-RESOURCES-MIB::hrSWRunParameters.5 = ""
HOST-RESOURCES-MIB::hrSWRunParameters.6 = ""
HOST-RESOURCES-MIB::hrSWRunParameters.7 = ""
HOST-RESOURCES-MIB::hrSWRunParameters.8 = ""
HOST-RESOURCES-MIB::hrSWRunParameters.9 = ""

As you can see, hrSWRunPath table is truncated.

[Brian O'Neill (boneill-inforonics)]

I have just come across this myself, and similarly it has broken our process monitoring of servers that are running Lucid.

Will we see the fix rolled out?

[Sergey Svishchev (svs)]

Maybe not...

[pdf (pdffs)]

A proposed solution is to simply backport hr_swrun.c from 5.4.4, tested and problem resolved here using this patch.