Re: Adler32 addition to Nbase in r12676 causing rare segfault

[Daniel Roethlisberger <daniel () roe ch>]

Brandon Enright <bmenrigh () ucsd edu> 2009-04-15:
> I just ran into a segfault in the new Adler32 code in Nbase.  I'm
> seriously pressed for time at the moment so I won't be able to help
> troubleshoot for a day or two.
>
> Here is what I have:
>
> Program terminated with signal 11, Segmentation fault.
> [New process 15092]
> #0  adler32 (buf=<value optimized out>, len=123137424) at nbase_misc.c:460
> 460         s1 = (s1 + buf[n]) % ADLER32_BASE;
> (gdb) bt
> #0  adler32 (buf=<value optimized out>, len=123137424) at nbase_misc.c:460
> #1  0x00007f2309cb0ee8 in inflate () from /lib/libz.so.1

Are you sure this isn't actually a segmentation fault in zlib's
adler32(), and gdb is messing up the namespace, using the debug
information and/or source code for adler32() in nbase?

Nbase adler32() is not called from anywhere at the moment in
trunk, and I'd be rather surprised if your rtld would somehow
cause deflate() in libz.so.1 to call nbase adler32() instead of
zlib's internal adler32().

> #2  0x00007f230aebb4d8 in ?? () from /usr/lib/libcrypto.so.0.9.8
> #3  0x00007f230aeba8f1 in COMP_expand_block () from /usr/lib/libcrypto.so.0.9.8
> #4  0x00007f230b157f3d in ssl3_do_uncompress () from /usr/lib/libssl.so.0.9.8
> #5  0x00007f230b158470 in ssl3_read_bytes () from /usr/lib/libssl.so.0.9.8
> #6  0x00007f230b15921d in ssl3_get_message () from /usr/lib/libssl.so.0.9.8
> #7  0x00007f230b1597f2 in ssl3_get_finished () from /usr/lib/libssl.so.0.9.8
> #8  0x00007f230b154cdf in ssl3_connect () from /usr/lib/libssl.so.0.9.8
> #9  0x00000000004755f0 in handle_connect_result (ms=0x750bc60, nse=0x750f170, 
>     status=16) at nsock_core.c:348
> #10 0x0000000000476b5d in nsock_loop (nsp=0x750bc60, msec_timeout=-1)
>     at nsock_core.c:828
> #11 0x000000000045d788 in service_scan (Targets=@0x7fff139e6b10)
>     at service_scan.cc:2435
> #12 0x000000000041db2d in nmap_main (argc=35, argv=0x7fff139e9db8)
>     at nmap.cc:1787
> #13 0x0000000000418f37 in main (argc=35, argv=0x7fff139e9db8) at main.cc:215
> Current language:  auto; currently c
>
>
> I've gotten quite a handful of these crashes (~20), here is an
> interesting print from another core dump:
>
> Program terminated with signal 11, Segmentation fault.
> [New process 24252]
> #0  0x00000000004767e8 in adler32 (buf=0x1 <Address 0x1 out of bounds>, 
>     len=110187248) at nbase_misc.c:460
> 460         s1 = (s1 + buf[n]) % ADLER32_BASE;
>
>
> If I had to guess, I'd say that the length field is insanely big,
> causing a overflow somewhere and overwriting the buf pointer.  Since I
> haven't looked at any code yet this is 110% guessing.

The len field looks like the high dword of a 64 bit pointer,
probably pointing to the heap.  This could equally be either gdb
misinterpreting things based on the wrong source code, or code
compiled against the signature of zlib adler32() actually calling
nbase adler32().

> If the problem isn't obvious, I'll suggest adding a few
> assert()s here and there.  I'll try to reproduce the problem in
> Valgrind in a few days if it isn't fixed by then.

Unless someone can reproduce these segfaults, I don't think I can
fix this without some debugging / verification help from you.

Can you add an assert(0) on the first line of nbase adler32() to
make sure that our adler32() is not called when your segfaults
happen?

-- 
Daniel Roethlisberger
http://daniel.roe.ch/

_______________________________________________
Sent through the nmap-dev mailing list
http://cgi.insecure.org/mailman/listinfo/nmap-dev
Archived at http://SecLists.Org