Created attachment 1228669 [details]
Patch - take read lock

Description of problem:

Running any workload results in a core dump with this message:
Illegal instruction (core dumped)

Version-Release number of selected component (if applicable):
filebench-1.4.9.1-5.fc24.x86_64

How reproducible:
Always

Steps to Reproduce:
1. Use this sample workload file (any valid workload will do):

set $dir=/tmp
set $nfiles=100
set $meandirwidth=20
set $meanfilesize=16k
set $iosize=1m
set $nthreads=1

set mode quit firstdone

define fileset name=bigfileset,path=$dir,size=$meanfilesize,entries=$nfiles,dirwidth=$meandirwidth,prealloc=100,paralloc
define fileset name=destfiles,path=$dir,size=$meanfilesize,entries=$nfiles,dirwidth=$meandirwidth

define process name=filereader,instances=1
{
  thread name=filereaderthread,memsize=10m,instances=$nthreads
  {
    flowop openfile name=openfile1,filesetname=bigfileset,fd=1
    flowop readwholefile name=readfile1,fd=1,iosize=$iosize
    flowop createfile name=createfile2,filesetname=destfiles,fd=2
    flowop writewholefile name=writefile2,fd=2,srcfd=1,iosize=$iosize
    flowop closefile name=closefile1,fd=1
    flowop closefile name=closefile2,fd=2
  }
}

echo  "Copyfiles Version 3.0 personality successfully loaded"
run 10

2. filebench -f <workload_file>

Actual results:
Core dumped just before reaching the "running" stage:
Illegal instruction (core dumped)

Expected results:


Additional info:

From the core file produced when this happens in a VM, it looks like it crashes in pthread_rwlock_unlock().  Looking at the source, the main thread tries to signal the worker threads with:

/* Release the read lock, allowing threads to start */
(void) pthread_rwlock_unlock(&filebench_shm->shm_run_lock);

... but fails to actually take the lock at any point before doing that.

See attached small patch for a proposed fix.  Note that more recent upstream (currently at 1.5 alpha 3) does not have this particular issue.