We received a bug report from one of our customers about their application crashing when they enable Datadog profiler.

The crash stack trace is as follows:
---
C [libc.so.6+0xf53cd] readdir+0x2d
V [libjvm.so+0x6d5a20] JfrOSInterface::initialize()+0xe0
V [libjvm.so+0x6e4a32] JfrRecorder::create_components()+0x1b2
V [libjvm.so+0x6e4b9d] JfrRecorder::create(bool)+0x1d
V [libjvm.so+0x6ce717] jfr_create_jfr+0x187
j jdk.jfr.internal.JVM.createJFR(Z)Z+0
j jdk.jfr.internal.JVM.createNativeJFR()V+3
j jdk.jfr.internal.PlatformRecorder.<init>()V+48
j jdk.jfr.FlightRecorder.getFlightRecorder()Ljdk/jfr/FlightRecorder;+25
j jdk.jfr.Recording.<init>(Ljava/util/Map;)V+4
j jdk.jfr.Recording.<init>()V+8
---

The crash is not specific to our profiler - any attempt to start JFR will trigger it.
The customer is using SELinux in a very restrictive setup. Eg. they deny access to `/proc` directory.


The issue is that in os_perf_linux.cpp (https://github.com/openjdk/jdk/blob/bab70193ddaae66a1f9039b0f1af1a0dc0e39cf9/src/hotspot/os/linux/os_perf_linux.cpp#L848) an attempt to open `/proc` dir is made but no error is expected.
Therefore, it that call fails, the `_dir` variable will be NULL and the subsequent call to `readdir(_dir)` (https://github.com/openjdk/jdk/blob/bab70193ddaae66a1f9039b0f1af1a0dc0e39cf9/src/hotspot/os/linux/os_perf_linux.cpp#L829) will crash.