openwall
Some formats fail '-internal mode internal mask' test
It is hard to reproduce. So far I've seen it only using AMD CPU driver (but it is NOT a driver bug). To reproduce:
rm -f ../run/opencl/*.bin *.pot; make -sj8 &&
../run/john -lws=2 -gws=2 tst-.in --skip --mask=0?d -form=xsha512-opencl -dev=7
A patch follows ("my formats" only), but I'm not closing the bug. I also saw a failure (SIGSEGV) in: [edited]
-
mysql-sha1-opencl. -
raw-sha1-opencl -
raw-md4-opencl(return code: 135) -
raw-md5-opencl
Tested on "well" now, with ./jtrts.pl -pass=-dev=7 -internal mask
- cryptosafe-opencl segfaults
- mscash-opencl segfaults
- nt-opencl segfaults
- ntlmv2-opencl segfaults
- raw-sha1-opencl segfaults
The three other formats you mentioned passed just fine. Intermittent?
When instead running john -test=0 -form=mask -dev=7, all 17 format passes with no problem.
After more TS runs, it becomes obvious most of these segfaults are intermittent
Intermittent?
Sure.
Well, I can't remember about a fix for raw-md * formats. Anyway, a fix is also a possibility.
Using Intel CPU driver instead (dev 1), some formats do show up too:
form=oldoffice-opencl guesses: 12 N/A N/A [PASSED]
form=raw-md4-opencl guesses: 0 -show= 0 0:00:00:00 DONE : Expected count(s) (13)(-show13) [!!!FAILED1!!! ret_val=0]
form=raw-md5-opencl guesses: 0 -show= 0 0:00:00:00 DONE : Expected count(s) (6)(-show6) [!!!FAILED1!!! ret_val=0]
form=raw-sha1-opencl guesses: 0 -show= 0 0:00:00:00 DONE : Expected count(s) (23)(-show23) [!!!FAILED1!!! ret_val=0]
form=salted-sha1-opencl guesses: 0 -show= 0 0:00:00:00 DONE : Expected count(s) (21)(-show21) [!!!FAILED1!!! ret_val=0]
form=sl3-opencl guesses: 0 -show= 0 0:00:00:00 DONE : Expected count(s) (2)(-show2) [!!!FAILED1!!! ret_val=0]
It wasn't in my case (#4633). But, I haven't tested/debugged all formats.
IMO, JtR is failing to properly handle OpenCL initialization in some formats and errors like this are a side effect of this. I do believe some formats are not doing the right thing. So, bugs exist in our code.