Bug 1405635

+++ This bug was initially created as a clone of Bug #1405205 +++

Description of problem:
crm_report collects /var/log/lastlog file (or at least is looking for patterns in there for some reason).

Usually that is harmless, because the file is small. If it gets large, it creates a problem for grep used by crm_report, as it scans through that binary file and buffers everything up to a newline char.

You can quite easily create a file that is very large and grep will try to buffer it. Then it runs out of RAM and gets killed.

Easy way to create a large /var/log/lastlog is:
useradd -b /mnt/brawl -m -U -c "quota-sanity user" -u 10000000 quota-user-kPRAaCKm

Even though these circumstances are not exactly the common ones, we do crash on a file that most likely does not contain useful info. Can you please add a check there or just remove the grepping through it completely?

Ideally no part of crm_report should be able to eat up all memory :).

Version-Release number of selected component (if applicable):
pacemaker-cli-1.1.15-3.el6.x86_64

How reproducible:
always

Steps to Reproduce:
1. on any cluster node: useradd -b /mnt/brawl -m -U -c "quota-sanity user" -u 10000000 quota-user-kPRAaCKm
2. run crm_report that collects info from all nodes
3.

Actual results:
grep crashes (signal 6) on the affected node, maybe some files will be missing in the report.

Expected results:
* ideally skip /var/log/lastlog collection
* crm_report being more cautious on binary files with grep (doing a size check would be neat)

Additional info:

--- Additional comment from Jaroslav Kortus on 2016-12-15 16:06:36 EST ---

QA note: testable using brawl_quick + crm_report after (in ~3hrs) or by just running the quota subtest of it (ping refried), or just manually via reproducer.

--- Additional comment from Ken Gaillot on 2016-12-16 16:58:07 EST ---

The problem is that crm_report dynamically detects what system logs are used for the cluster by grepping for a particular pattern in (up to) all files in /var/log.

It's already on the long-term plan to convert crm_report from a shell script to python, to make the file handling much more efficient.

But for 6.9 timeframe, I can make sure "file" returns "text" or "compressed" before doing the grep. That will at least skip lastlog, wtmp, etc.