Fri, 21 Jul 2000 22:58:18 EDT
[log in to unmask] quoted and then wrote:
>+> I don't know how many of you follow the security bulletins
>> in the Unix-world, but there has recently been a lot of
>> trouble with the more widely used FTP daemons (one of them
>> is WU-FTPD).
>I'm not familiar with this particular problem, but many
>security problems in internet clients and servers are the
>result of buffer overflow bugs. An example is the recent
>flap over Outlook and executable code in MIME headers.
>Please do not be so confident as to assume this doesn't
>happen if you were to implement the code in Ada. You
>certainly can prevent it in pure Ada code, but if you are
>interfacing to the operating system, you are still subject
>to potential problems.
I am familiar with a relatively large development organization
that makes little use of Ada but has only recently started to
use C in volume. They seem to do much better than average at
avoiding buffer overflows because of a longstanding discipline
of carefully checking sizes and lengths. So for the project
proposed, I think the fact that the implementation language
would be Ada might color the degree of discipline of those
doing the programming.
>I recently ported some Ada code which contained an interface
>to the TM time structure (time.h) from Solaris to Linux. I
>did not carefully examine the Linux include file, just compiled
>the code. Turns out that the Linux (Slackware) had an extra
>long and char* in the structure that weren't in the Solaris
>definition. Took me quite awhile to find the bug. The code
>worked when I had a Text_IO debug statement, but not when I
>removed it. That's the only clue I had!
But the lesson I take from that is that the proposed FTP project
would end up not being so "portable" as desired even if it were
restricted to the set of Unix-like operating systems (a much
more narrow scope than I would like to see in a premier Ada
My own security experience says that even if you avoid buffer
overflows, there are _many_ additional defects just waiting
to happen in this problem domain, especially if one attempts
a high-performance implementation.