RFR(xs): 8216559: [JFR] Native libraries not correctly parsed from /proc/self/maps

[Markus Gronlund]

Hi Severin,

Looks good, thanks for fixing.

Markus

-----Original Message-----
From: Severin Gehwolf <sgehwolf at redhat.com> 
Sent: den 11 januari 2019 14:35
To: hotspot-dev <hotspot-dev at openjdk.java.net>; hotspot-jfr-dev at openjdk.java.net
Subject: RFR(xs): 8216559: [JFR] Native libraries not correctly parsed from /proc/self/maps

Hi,

Could I please get a review of this tiny fix for native libraries parsing from /proc/<pid>/maps? It's a Linux-only change and seems to surface only on certain systems (with 3-digit device ids in /proc). The issue has been noticed since TestNativeLibrariesEvent.java was failing for me.

What's the problem? The Linux impl of os::get_loaded_modules_info reads /proc/self/maps and scans each line into variables via sscanf(). The format for the device pointer allowed up to 5 characters. In my case it was 6 characters long. As a result, the last digit of the device id got interpreted as inode value and the inode value as name. Thus, the failing test was showing strange library names like "25847051".

According to the spec, major:minor device numbers may be up to 3-digits in length. Hence, I propose to change the format from '%5s' to '%7s': 3 digits for major/minor each, plus one for ':'.

Bug: https://bugs.openjdk.java.net/browse/JDK-8216559
webrev: http://cr.openjdk.java.net/~sgehwolf/webrevs/JDK-8216559/webrev.01/

Testing: jdk_jfr tests on Linux x86_64 including TestNativeLibrariesEvent.java
         which now passes.

Thoughts?

Thanks,
Severin