userland core files.
By timatworkhomeandinbetween on Apr 30, 2005
During the week I collected several escalations that required analysis of userland core files. In both cases I was supplied the core file and a showrev -p and asked for a quick analysis. So my choices were to either..
Point our jumpstart configuration scripts at the showrev-p output and install a machine so that I have all the right libraries in the right place. Machines are an expensive resource and it seems a waste to book and occupy a whole machine just for this.
- Use the shared library checksums in the core file to extract the
correct libraries from our patch database and get dbx to use
those. I haven't written the scripts to do this yet so it would take a
Ask the customer for some more files from his machine. The binary that generated the core and then all the shared libraries listed in the output of pldd core. With those I can build a dbxrc that maps the loads of the libraries to use the customer's ones. Once I have these I can debug the core on any machine where dbx is available.
In solaris 10 coreadm gives you much better control over what parts of the faulting processes address space end up in the core file. The default will include the shared library segments so core files will be bigger but debugging them wont need all the other files.