firefox23 doesn't start up correctly when certain syscalls are wrapped

[rr-debugger]

firefox23 doesn't start up correctly when certain syscalls are wrapped

[joneschrisg]

So hm. #32 shows that all the tests are passing when run with the filter lib. So this appears to be some new issue specific to FF.

@rocallahan do you recall the last time you successfully recorded FF with the filter lib? What was the FF version?

[joneschrisg]

Disabling the seccomp-bpf and bisecting narrows the problem down to some subset of the socket interface

accept4
accept
getpeername
getsockname
getsockopt
recv
recvfrom
socket
bind
connect
listen
sendmsg
send
sendto
setsockopt
shutdown

[joneschrisg]

Ended up being pretty complex. Here's what's busted

clock_gettime
gettimeofday
accept4 - shutdown (socketcall API)

I'm running an x86 build in an x86-64 VM. I know that the ABI for clock_gettime/gettimeofday is significantly different across x86/x86-64, so this is pretty suspicious. Will either disable or switch (back, sigh) to an x86 VM.

[joneschrisg]

Broken the same way in an x86 VM, too ...

[joneschrisg]

"Fixed" (worked around temporarily) by #35.

[rocallahan]

It worked for me a couple of months ago in Fedora 17.