Re-check with p_candebug() that credentials of the target are still fine.
Block execve() around the check and fget_remote() call with execve_block(), which prevents suid exec under us.

Same bug assuming the number of queues from iflib will be a power of two, was the only other network driver I found with the same bug

How's this relate to D57240 ?

In D57242#1311904, @bz wrote:

Not saying that I'll take them but where did you get the two Realtek driver files from?
The ones I am aware off are different, that's why I am asking.

exp-run requested

Not saying that I'll take them but where did you get the two Realtek driver files from?
The ones I am aware off are different, that's why I am asking.

In D57124#1311901, @arrowd wrote:

This looks promising! Here's my plan on how to exploit this:

• The "client" program gets a procdesc fd for itself with pdopenpid(getpid()).
• It then sends the procdesc fd to the "server" program via UNIX socket.
• The "server" program checks that client's PID matches the passed procdesc using pdgetpid().
• If the descriptor is OK the "server" then polls (via poll, select or kevent) on it to have a notification when the process exits. I don't even need the exit code.

If I understand it correctly, my plan should be possible once this diff lands.

This looks promising! Here's my plan on how to exploit this:

Initialize uio_td.

respond to olce@'s comments

Add p_comm

In D57231#1311555, @glebius wrote:

To me this looks like a more generic problem rather than just the watchdogd's problem. Should every daemon have this feature? Or should we say that /var/run shall be mdmfs(8) on hosts with read-only root?

I added authors of the pidfile(3) to add people who definitely thought a lot on this problem.

In D57247#1311835, @glebius wrote:

P.S. Oleksandr, if you got review from any other domain expert than Warner, consider this approved by mentor (me).

P.S. Oleksandr, if you got review from any other domain expert than Warner, consider this approved by mentor (me).

Not a domain expert, so leaving this 100% to Warner. Code & style wise everything LGTM.

This update takes care of PERSONALITY_PATH in addition to PKG_DEFAULT_PATH.

Fill uio_seg and uio_rw.

Thanks! Applied the feedback now.

In D51589#1311788, @emaste wrote:

Is it reasonable to commit this (with the understanding that there will still be issues post-resume)?

Is it reasonable to commit this (with the understanding that there will still be issues post-resume)?

In D57244#1311781, @markj wrote:

Shall we update the kill test in tests/sys/capsicum/capmode.cc to verify sigqueue too?

Shall we update the kill test in tests/sys/capsicum/capmode.cc to verify sigqueue too?

The default pattern is to repeat the queues across the RX queues so if there are 4 RSS queues but 8 RX queues you end up with 1 2 3 4 1 2 3 4 but this is actually configurable at runtime for traffic steering and whatnot