Debug a futex lock

Question

I have a process waiting on a futex:

# strace -p 5538
Process 5538 attached - interrupt to quit
futex(0x7f86c9ed6a0c, FUTEX_WAIT, 20, NULL

How can I best debug such a situation? Can I identify who holds the futex? Are there any tools similar to ipcs and ipcrm but for futexes?

Larimor

Answer 1

Try using gdb -p *PID* and then run where or bt to see a backtrace.

It won't be spectacularly useful with binaries and libraries that have had their debugging symbols stripped, but you may be able to deduce a fair bit from the context. It might be able to indicate to you which part of a complex process is hanging, and then you could examine the right part of the sources to search for the lock.

Schuh

Answer 2

I have the same problem with a piece of c++ code. Running ubuntu 12.10 64bit. It looks to like a similar problem in 2007, where the libc was buggy (and maybe still is?).

I start a pthread which runs a traceroute in a system call. Printf before and after the system indicate, that the operating system hangs on the system call, WITHOUT executing the traceroute.

I am not sure if my linux is broken once again because of the ubuntu update, or if it's a libc related bug. Since a lot of applications seem to have "similar" problems, I assume it's stuck somewhere in the userspace.

My c++ code runs perfectly on 32bit systems and even 64bit osx, so i assume that ubuntu 12.10 + 64bit libc combination is broken.

Condensed