regression after FreeBSD-EN-15:17.libc

[Peter Blok]

Dear hackers,

I have a strange problem with signal handling after the above errata was implemented. It exposes itself after a slogin to a FreeBSD-10.2-RELEASE-p7 and pressing Ctrl-C. It terminates the csh and disconnects. This happens on two systems, one physical Octacore Atom and on a virtual system at Hetzner. Both systems are in production and amd64.

Sources are in sync with svn. .cshrc is standard.

I have tried to setup different test systems - one quad core Xeon and one VMWare Fusion, running the exact same code, but they don’t exhibit the problem.

Besides the termination of the csh, I have seen corruptions in a db5 database, after reboots. For example the sshguard database was garbled after reboot. My suspicion is that it is signal related, caused by reboot. This is why I am investigating further.

I had a feeling it happened after the above errata change was implemented. If I backout the changes everything works ok. If I put them back in, it fails again.

Checked the changes, but I can’t see anything wrong with them in relation to the problem.

Some other data points.

- it doesn’t happen with ksh93
- it doesn’t happen after "exec csh -F”   which doesn’t use vfork

At one time I used ktrace and noticed the signal was delivered twice.

I’ll make an exact clone of the Hetzner image and try to reproduce it, but any other advice is welcome.

Peter