Murphy v1.0

Written by Zach Miller and Todd Tannenbaum, with help from Ben Liblit. Please contacts us with questions or comments!

See also the Murphy Project Page for more material, including research papers about how Murphy was created and some of the scary bugs it has found.

Subdirectories and Licenses

• classads-1.0.10: This is the ClassAd Library source, used as the policy language in Murphy. Distributed via the Apache 2.0 License from http://research.cs.wisc.edu/condor/classad/.

• murphy-1.0: This is the Murphy source code, implemented as a derivative work of the Parrot tool included as part of the Cooperative Computing Tools (cctools). cctools is released under the GPL v2 License from http://nd.edu/~ccl/software/.

To Build

On a 64-bit Linux system, in this subdirectory enter:

./build_murphy

If all goes well with the build, the binary “Murphy” will end up in this subdirectory.

Note: Murphy will not run properly on a 32-bit system.

To Run

To use Murphy, quick and dirty, in this directory enter “source setup.sh” or “source setup.csh”, then “Murphy ./my_program”.

More Detailed Info

Gremlin Configuration File

The environment variable MURPHY_CONFIG should point to a gremlin configuration file. There is an example in murphy_config.example that has all gremlins listed but disabled. You can modify this according to what you want.

The format is one entry per line, with four colon-delimited fields on each line:

gremlin:percent:seed:constraint

1. gremlin can be: readone, readless, readone_s, writeone, writeone_s, writezero, eintr, eagain, enospc, closefail, or selectfdset

2. percent is an integer between 0 and 100 inclusive, if you want random invocation.

3. seed is an integer used to seed the pseudo-random number generator

4. constraint is a classad constraint which must evaluate to true in order for the gremlin to be invoked. See the classad language and reference manual.

In the configuration file, blank lines and lines starting with “#” are ignored.

Running Murphy

Add the directory containing Murphy to your path. This isn’t strictly necessary, but is probably a good idea and will save you time.

To support programs that use vfork(), you must set the environment variable PARROT_HELPER to point to the file libparrot_helper.so.

To run Murphy, use “Murphy [options] <command> ...”, where options and environment variables are:

• -c <file>: Path to Murphy config file, defaulting to $MURPHY_CONFIG.
• -g <file>: Send gremlin activation log to this file.
• -r <file>: Replay a gremlin activation log from this file.
• -a: Attach with gdb when replay of a gremlin activation log completes.
• -d <name>: Enable debugging for the named sub-system.
• -H: Disable use of helper library.
• -h: Show brief help information.
• -l <path>: Path to ld.so to use.
• -o <file>: Send debugging messages to this file, defaulting to stderr.
• -O <bytes>: Rotate debug files of this size.

Debuggable subsystems for use with the -d flag are: syscall, notice, process, pstree, alloc, cache, poll, debug, murphy, user, all, time, and pid. The most useful ones for Murphy are “-d murphy” and “-d syscall”.

Run-Time Steering

The Murphy run-time control function is mkdir(). Four commands are currently implemented:

• mkdir /Murphy/set-metadata/<MeTaDaTa>: Set the per-process metadata to the string “<MeTaDaTa>”. The string is limited to 1000 characters, but any non-null character is legal.

• mkdir /Murphy/update-config/<config-entry>: Dynamically set a new config entry, using the same format that is used in the murphy_config file.

• mkdir /Murphy/suspend: Suspend your program and detach Murphy. You are now free to attach with your own debugger, send it SIGCONT to continue, kill it, etc. Don’t forget to clean it up or it will lurk forever.

• mkdir /Murphy/suspend-and-debug: Detach Murphy from the program under test, then reattach to it with gdb. This is useful for most people and allows you to inspect data, get a stack trace, etc.