Michal Trojnara wrote:
Dear OpenSSL users,
When performing stress-testing of stunnel with session cache
disabled I receive core dumps on concurrent SSL_connect() calls.
Here is an example stack backtrace:
#0 0xa7e60d41 in BN_ucmp (a=0x80a28fc, b=0x80a1f08) at bn_lib.c:662
662 bn_check_top(b);
(gdb) bt
#0 0xa7e60d41 in BN_ucmp (a=0x80a28fc, b=0x80a1f08) at bn_lib.c:662
#1 0xa7e6c803 in BN_from_montgomery (ret=0x80a28fc, a=0x80a2924,
mont=0x80a1ef0, ctx=0x80a2578) at bn_mont.c:281
#2 0xa7e6c3ca in BN_mod_mul_montgomery (r=0x80a28fc, a=0x80a28fc,
b=0x80a28fc, mont=0x80a1ef0, ctx=0x80a2578) at bn_mont.c:145
#3 0xa7e5e555 in BN_mod_exp_mont (rr=0x80a28d4, a=0x80a28c0,
p=0x80a1d80, m=0x80a22a8, ctx=0x80a2578, in_mont=0x80a1ef0)
at bn_exp.c:464
#4 0xa7e81a8c in RSA_eay_public_encrypt (flen=48,
from=0xa7fa0960 "\003", to=0x808083c "", rsa=0x80e88c0, padding=1)
at rsa_eay.c:238
#5 0xa7e83e99 in RSA_public_encrypt (flen=48,
from=0xa7fa0960 "\003", to=0x808083c "", rsa=0x80e88c0, padding=1)
at rsa_lib.c:282
#6 0xa7f55fb1 in ssl3_send_client_key_exchange (s=0x80e7f10)
at s3_clnt.c:1680
#7 0xa7f531a9 in ssl3_connect (s=0x80e7f10) at s3_clnt.c:327
#8 0xa7f6a6fd in SSL_connect (s=0x80e7f10) at ssl_lib.c:850
#9 0x0804c96b in init_ssl (c=0x2) at client.c:322
#10 0x0804c581 in do_client (c=0x806eb68) at client.c:208
#11 0x0804c52e in run_client (c=0x806eb68) at client.c:151
#12 0x0804c29b in client (arg=0x806eb68) at client.c:124
#13 0x4dd17ca3 in start_thread ()
from /lib/tls/i686/cmov/libpthread.so.0
#14 0x4db98f5a in clone () from /lib/tls/i686/cmov/libc.so.6
(gdb) l bn_lib.c:662
657 {
658 int i;
659 BN_ULONG t1,t2,*ap,*bp;
660
661 bn_check_top(a);
662 bn_check_top(b);
663
664 i=a->top-b->top;
665 if (i != 0) return(i);
666 ap=a->d;
Serializing requests solves the problem, but is definitely not a solution for
real life applications. 8-)
Calling SSL_connect() and SSL_accept() from a critical section helps a little.
Instead of core dumps I get the following errors:
SSL_connect: 1408C095:
error:1408C095:SSL routines:SSL3_GET_FINISHED:digest check failed
The library is OpenSSL 0.9.8d configured with:
./Configure threads shared zlib debug-linux-elf
Can you help me find a solution or a better workaround?
does it help is you build openssl without BN_DEBUG ? When BN_DEBUG
is defined bn_check_top() is a macro that should check if the BIGNUM::top
value is correclty set (note: this isn't really a solution as
bn_check_top() (without BN_DEBUG_RAND) should be mt-safe).
Cheers,
Nils
______________________________________________________________________
OpenSSL Project http://www.openssl.org
User Support Mailing List openssl-users@openssl.org
Automated List Manager [EMAIL PROTECTED]
