Java process segfault in LCMS library

Description

Dunno if that's a java issue, a geotools issue, or a geoserver issue, but some innocuous getmap request on a layergroup manages to crash java/tomcat/geoserver with a segfault. This happened twice in the past, it crashed the two tomcat instances i'm using in a loadbalanced environment, and this time the request came from a monitoring client doing random bbox requests on the same layer all the time.

http://ids.craig.fr/wxs/public/ows?EXCEPTIONS=XML&BBOX=699667.8967252899%2C6499926.519292068%2C734929.2462655391%2C6535187.868832317&CRS=EPSG%3A2154&VERSION=1.3.0&FORMAT=image%2Fpng&SERVICE=WMS&SERVICE=WMS&HEIGHT=256&_SPATINEO_ID=12260&LAYERS=procedures&REQUEST=GetMap&STYLES=&WIDTH=256

Replaying the same request doesnt crash tomcat, so it doesnt seem related to the request parameters.

Environment

tomcat 8.0.14-1+deb8u1, JRE 7u95-2.6.4-1~deb8u1, liblcms 2.6.3+b3.

Activity

Show:
Andrea Aime
March 24, 2016, 8:30 AM

"Can reproduce" being the key. The developer that is going to fix this one needs to be able to reproduce too

Landry Breuil
March 24, 2016, 9:41 AM

Well, i'd be very happy to provide detailed STR, but last i looked at the previous segfaults (with java7), replaying the exact same queries that triggered the initial crash wasnt re-crashing tomcat, so im not sure it's related to a particular dataset/style/request.

What's also strange is that the two load-balanced/independant tomcat instances crashes together in series..

I'll try to have a deeper look into the logs.

Ben Caradoc-Davies
March 24, 2016, 10:08 AM

Landry,

by definition, any JRE crash is a JRE bug. No JRE should segfault regardless of bytcode input. Our ability to help is likely to be limited to reproducing the segfault, identifying the circumstances that lead to it, documenting any mitigation measures, and reporting upstream; this bug can stay open until these tasks are done. The scope of this project does not extend to fixing bugs in OpenJDK. Nonetheless, your report may help others who encounter a similar segfault. I look forward to your STR.

Kind regards,
Ben.

Landry Breuil
January 16, 2020, 9:23 AM

It seems the last crashes we had were in october 2018, when we upgraded geoserver from 2.8.1 to 2.12.1 it seems they completely disappeared.

 

Right now we’re on 2.12.5 with liblcms 2.8-4+deb9u1, jdk 8u232-b09-1~deb9u1 and tomcat 8.5.50-0+deb9u1 on debian jessie, and i didnt experience any jdk/lcms crashes.

 

Assignee

Unassigned

Reporter

Landry Breuil

Triage

None

Fix versions

None

Affects versions

Components

Priority

High
Configure