1 |
On Thu, 2004-03-25 at 20:11, Chris Bainbridge wrote: |
2 |
|
3 |
> Do you believe that this is the case for all developers? Even if it were, |
4 |
> there are many programs other than daemons that connect to the internet that |
5 |
> may contain exploitable code. Web browsers, email, irc, ftp, cvs, ssh, |
6 |
> usenet, p2p; developers use these programs often, and many exploits have been |
7 |
> found in client implementations in the past. |
8 |
So shouldn't we focus on creating reliable software and only add package |
9 |
managment etc. when the software is "good enough" ? |
10 |
We have to accept a certain level of "bugginess", otherwise allmost no |
11 |
software would exist. |
12 |
|
13 |
> Know that a developer checks out |
14 |
> cvs from some other project? Hack the server, and when his ip address |
15 |
> connects, send trojan code. Then there are physical attacks; is a developer |
16 |
> at your uni? Pick his door lock and install a hardware key logger. |
17 |
So ... what stops me from r00ting an rsync server? |
18 |
What stops me from taking some devs as hostages to get the keys? |
19 |
|
20 |
I think that we _should_ focus on high-profile risks. That includes the |
21 |
compromise of a small number of computers. Any large-scale attack will |
22 |
succeed. |
23 |
|
24 |
> All of these things might individually be less likely than a direct attack, |
25 |
> but together the possibility that one small security breach, for a single |
26 |
> developer, might occur is more than comparable to the possibility that the |
27 |
> rsync code, which has been extensively audited, might contain an external |
28 |
> exploit. |
29 |
So ... what if hypothetically there was more than one service running on an |
30 |
rsync server and the server was compromised due to something unrelated to rsync? |
31 |
|
32 |
Hmmm, that reminds me of a gentoo rsync server that was compromised... |
33 |
|
34 |
</sarcasm> |
35 |
|
36 |
It is good that you, as well as many other Gentooists, are very aware of |
37 |
security risks, but I think that this discussion is slowly slipping into |
38 |
a paranoid worst-case analysis that says that we can't trust nobody. |
39 |
|
40 |
The security system has to be robust (because people make mistakes) and |
41 |
simple (because we are lazy). So any complex n+1 signatures with |
42 |
certificate from m+1 servers will not work because some devs will not |
43 |
keep the high standard of security that would be needed for such a |
44 |
system. (e.g. ignore warnings because there are too many) |
45 |
|
46 |
|
47 |
hth, |
48 |
Patrick |
49 |
|
50 |
|
51 |
-- |
52 |
gentoo-dev@g.o mailing list |