[Pkg-openssl-devel] Bug#459619: Bug#459619: openssl speed segfaults for rsa

[Kurt Roeckx]

severity 459619 serious
thanks

On Mon, Jan 07, 2008 at 05:53:52PM +0100, Elmar Hoffmann wrote:
> Package: openssl
> Version: 0.9.8g-3
> Severity: normal
> 
> Hi,
> 
> running openssl speed leads to a segfault, at least on powerpc.
> The last line printed is "Doing cast cbc for 3s on 8192 size blocks:
> [...]", ie. the last test before rsa.

I can reproduce your problem:
kroeckx at bruckner:~$ openssl speed
Doing md2 for 3s on 16 size blocks: 53618 md2's in 3.00s
Doing md2 for 3s on 64 size blocks: 28433 md2's in 3.00s
[...]
Doing blowfish cbc for 3s on 8192 size blocks: 2465 blowfish cbc's in 1.84s
Doing cast cbc for 3s on 16 size blocks: 1234613 cast cbc's in 2.55s
Doing cast cbc for 3s on 64 size blocks: 404089 cast cbc's in 3.00s
Doing cast cbc for 3s on 256 size blocks: 103739 cast cbc's in 3.01s
Doing cast cbc for 3s on 1024 size blocks: 26119 cast cbc's in 3.00s
Doing cast cbc for 3s on 8192 size blocks: 3272 cast cbc's in 3.00s
Segmentation fault

The weird thing is that when running cast or rsa on it's own it doesn't
do that.  I can also perfectly run this test on an amd64 and it works
perfectly.

I'll try to dig deeper into this.



> The following backtrace has the unprintable variable contents of m and
> sigret removed:
> 
> #0  RSA_sign (type=114, m=0x1007eed0, m_len=36, sigret=0x10080ed8,
>     siglen=0x7fe0e748, rsa=0x7257469f) at rsa_sign.c:79
> #1  0x10030fec in ?? ()
> #2  0x1000e254 in ?? ()
> #3  0x1000e8bc in ?? ()
> #4  0x0fc22b10 in ?? () from /lib/libc.so.6
> #5  0x0fc22cd0 in __libc_start_main () from /lib/libc.so.6
> #6  0x00000000 in ?? ()

Did you try installing the libssl0.9.8-dbg?  It might give a better
backtrace.  But I guess you have and wouldn't even get the variable
names and source lines without that.


Kurt