ktrace -d broken on current/stable-9

[Jilles Tjoelker]

On Mon, Jan 14, 2013 at 06:48:13PM -0800, Garrett Cooper wrote:
> I tried using ktrace on a kernel compiled a week ago, and it appears
> to not be following forks like it should on amd64:

> # ktrace -d ./regress -l
> [snip]

> Not sure how it broke, but it was working a couple months ago (in
> particular I remember it working either around October or November),
> and the bug seems to have worked its way back to 9-STABLE (I'm running
> into the same problem if I do ktrace -d, enter a shell, then exec
> another shell from that shell). Haven't spent the time to bisect the
> commits looking for the culprit (yet), but if need be I'll trace down
> the culprit sometime this week.

> truss works, so it doesn't seem like ptrace(2) is broken.

ktrace -d is not really useful in the synopsis with a command. It only
means that the child processes of ktrace (at a time just before it
executes the utility) should be traced as well. This is almost always an
empty set, unless you do things like
  cmd1 & ktrace -d cmd2
which will trace cmd2 and part of cmd1.

You probably want ktrace -i.

-- 
Jilles Tjoelker