Dtrace bugs

Hm. Dtrace has a lot of bugs in it...as I try to ensure dtrace
cannot crash the linux kernel, I am stumbling on to some "thinko"
errors in dtrace.

Try the following on a Mac:

$ dtrace -l >/dev/null & dtrace -l >/dev/null & dtrace -l >/dev/null & dtrace -l >/dev/null &

We get interrupted system calls and issues on the major/minor numbers
(see 'dmesg' after it fails).

Next up is DTRACE_ENABLEIOC. This calls dtrace_copyin_dof() which validates
the main dtrace mutex is not asserted. Normally it isnt, unless someone
else is running a heavy handed probe trace, in which case a nasty race
condition exists. Unless a VFS lock is applied, this could spell out
danger or kernel panic (on Solaris, FreeBSD and MacOS).

dtrace_xcall is another area of potential for kernel deadlock if
multiple dtraces are firing on multiple cpus.

I have nearly finished my "safety" checks in dtrace, having written
my own dtrace_xcall, and am checking that kernels dont deadlock on you.

But its uncovering some nasty race conditions in dtrace as a whole.


Post created by CRiSP v10.0.5a-b5971