NODE 6f37966ecfs & remailers
allan@elvis.tamu.edu (Allan Bailey)Wed, 17 Aug 94 08:25:55 PDT
Has anyone considered using a CFS directory (or directories) for
a remailer's files, spool, etc?
Any thoughts about such security measures?
--
Allan Bailey, allan@elvis.tamu.edu | "Freedom is not free."
Infinite Diversity in Infinite Combinations | allan.bailey@tamu.edu
Esperanto: MondLingvo, lingvo internacia.
NODE 7eecc302Re: cfs & remailers
Bill O'Hanlon <wmo@digibd.com>Wed, 17 Aug 94 09:21:13 PDT
On Wed, 17 Aug 94 10:22:19 -0500 Allan Bailey wrote:
--------
>
> Has anyone considered using a CFS directory (or directories) for
> a remailer's files, spool, etc?
>
> Any thoughts about such security measures?
>
I considered it, for the remailer@rebma.mn.org. I'm already running CFS
for personal entertainment & education, so it's a possibility.
Here's my assumptions about how I'd operate it:
1) CFS file systems are mounted sometime after boot, manually, by me. The
passphrase is entered at mount time. (Obviously, supplying the passphrase
via an /etc/rc script defeats any security that CFS might add.)
2) The file systems remain mounted throughout the uptime of the system, since
mail can come in at any arbitrary time, primarily while I sleep.
3) If someone comes knocking loudly at my door to do the raid thing, I'll
have bigger things to worry about than unmounting the CFS file systems.
My wife and daughter will be formost on my mind.
I thought of two problems with it.
1) I'd not only have to put the home directory of the remailer user under CFS,
but also the uucp and sendmail spool directories. (Rebma has a UUCP
connection for getting mail.) Otherwise, security would be pointless, since
the messages would be coming in the clear to the spool directories. Maybe
this wouldn't be so bad, but it seems like I'd have to do a lot of
tinkering before I'd trust that sendmail wasn't gonna drop my other mail
on the floor. (I get some consulting-type mail on this machine.
Potentially, I can miss out on financial opportunity if my mail is not
dependable. Chalk my caution up to pure greed.)
2) I'd have to come up with some kludge to spool the incoming mail files in
a directory if the CFS file systems weren't mounted. (For example, if
power failed on the machine, or it crashed and otherwise rebooted, and I
didn't notice and wasn't around to type the passphrase in to remount the
CFS system.)
Those two thoughts make me question what security I'm buying for my trouble.
Seems to me what I'm getting is protection from a law enforcement type or other
computer thief who unplugs my machine and takes it away. (If they want to
make a backup before turning the machine off, with the CFS file systems
mounted, they have to spend some time at it.)
The people whose security would be helped are those who do a single hop or
send unencrypted mail through the remailer. People who use the remailer
properly already have encrypted their mail.
I guess that I thought it was too much effort to do, given that the only people
who would derive added security are those who were too stupid to use the
remailer properly in the first place.
Anyone see a flaw in my reasoning? I actually was considering doing it anyway,
just for the fun of it, when I had free time. If there is some valid security
reason, it might move up on my to-do list.
NODE 9428e720Re: cfs & remailers
Matt Blaze <mab@crypto.com>Wed, 17 Aug 94 10:22:44 PDT
"Bill O'Hanlon" <wmo@digibd.com> writes:
>On Wed, 17 Aug 94 10:22:19 -0500 Allan Bailey wrote:
>>
>> Has anyone considered using a CFS directory (or directories) for
>> a remailer's files, spool, etc?
>>
...
>I thought of two problems with it.
>
>1) I'd not only have to put the home directory of the remailer user under CFS,
> but also the uucp and sendmail spool directories. (Rebma has a UUCP
> connection for getting mail.) Otherwise, security would be pointless, sinc
e
> the messages would be coming in the clear to the spool directories. Maybe
> this wouldn't be so bad, but it seems like I'd have to do a lot of
> tinkering before I'd trust that sendmail wasn't gonna drop my other mail
> on the floor. (I get some consulting-type mail on this machine.
> Potentially, I can miss out on financial opportunity if my mail is not
> dependable. Chalk my caution up to pure greed.)
>
>
>2) I'd have to come up with some kludge to spool the incoming mail files in
> a directory if the CFS file systems weren't mounted. (For example, if
> power failed on the machine, or it crashed and otherwise rebooted, and I
> didn't notice and wasn't around to type the passphrase in to remount the
> CFS system.)
...
I'm working (with very low priority, unfortunately) on a sendmail hack
that spools mail (instead of bouncing) if the mailbox write fails.
This will be intended for a secure mail system that I'm working on
that uses CFS for its storage. Stay tuned...
Another potential problem with sendmail->cfs interaction is that
CFS doesn't implement NFS file locking. This isn't much of an issue with
a single host and a single instance of CFS, but could be a problem if the
mailboxes are read and written by other machines or are remotely mounted
by the machine running sendmail.
By the way, another mode of operation you might consider is to use a
"permanent" key (that you supply at boot time) for the spool directories
and a temporary key (assigned randomly by the machine at boot time)
for temp files that have only local significance but that may have
sensitive data. /usr/tmp points to /crypt/tmp on my machine for this
service (do a cmkdir and cattach at boot time. You also have to hack
cfs to make /crypt/tmp be mode 777).
-matt
NODE f1da29abRe: cfs & remailers
Jim Hart <hart@chaos.bsu.edu>Wed, 17 Aug 94 14:44:46 PDT
> The people whose security would be helped are those who do a single hop or
> send unencrypted mail through the remailer. People who use the remailer
> properly already have encrypted their mail.
But they'd still be in your logs, unless you immediately delete
or encrypt your logs. If you keep logs to help debug your
system snoop-proofing those logs is a good idea.
Also CFSing mail spools just for regular e-mail is a good idea,
to help enforce the ECPA. I hope this becomes standard policy
on the Internet.
(Of course, don't forget SecureDrive available for DOS).
Jim Hart
hart@chaos.bsu.edu
NODE 02ea2e9fRe: cfs & remailers
Bill O'Hanlon <wmo@digibd.com>Wed, 17 Aug 94 16:16:18 PDT
On Wed, 17 Aug 1994 16:42:33 -0500 (EST) Jim Hart wrote:
--------
>
> > The people whose security would be helped are those who do a single hop or
> > send unencrypted mail through the remailer. People who use the remailer
> > properly already have encrypted their mail.
>
> But they'd still be in your logs, unless you immediately delete
> or encrypt your logs. If you keep logs to help debug your
> system snoop-proofing those logs is a good idea.
I skipped a step in giving my assumptions. By "people who use the remailer
properly" I mean people who encrypt AND chain through multiple remailers.
In that case, even if I were to keep logs, all that anyone would know from
a message is that a particular user used a remailer, or that a particular
cleartext message had a certain remailer as its jumpoff point. Not both.
(Unless, of course, I'm in collusion with other remailer operators. But that's
also a non-code issue.)
I'm not interested/concerned with preserving the security of the people who
don't chain and encrypt.
>
> Also CFSing mail spools just for regular e-mail is a good idea,
> to help enforce the ECPA. I hope this becomes standard policy
> on the Internet.
That's an interesting and valid point. I can see some sense in an
encrypted file system for mail spools, just to highlight a philosophical
point or to help create a new net-wide philosophy for the handling of email.
I'm not sure that security is improved, however.
I half-expect Eric or Tim to jump in here to point out that this is one
of those situations where you have to define who your enemy is, and to make
sure that your efforts apply to the situation.
My personal situation is, I run a remailer on a home Unix machine via a
phone line UUCP feed. I am the only user of this machine, so I do not
have to defend against users with local access. My efforts are intended to
block the following foes: my service provider and any node upstream of it,
thieves/misguided law enforcement types, and phone taps. Encrypting something
that I receive in the clear over an insecure line isn't useful.
Of course, this is very specific to my situation. I expect that there exists
sites where running CFS for the spools makes sense. The fact that Matt Blaze
has said he has put some effort into making that possible just reinforces that.
This conversation is making me think that I should follow some other remailers
and make the remailer at rebma only allow encrypted traffic, since I have such
a low-opinion of unencrypted traffic.
Now, when we're all running our mail traffic over something like swIPe, such
that all connections are encrypted... And if I got an encrypted UUCP
connection... That might change things.
Then again, if you want security, encrypt it and chain remailers, regardless.
Sorry. I'm rambling. I won't dignify it by calling it "brainstorming."....
-Bill
NODE ba82ae99Re: cfs & remailers
Jim Hart <hart@chaos.bsu.edu>Fri, 19 Aug 94 00:58:16 PDT
Bill O Hanlon:
> In that case, even if I were to keep logs, all that anyone would know from
> a message is that a particular user used a remailer, or that a particular
> cleartext message had a certain remailer as its jumpoff point. Not both.
They'd learn both if they had snooped the entire remail chain (which
is the equivalent of collusion). Going back and retrieving logs
for all the the links, after the snoopers have discovered an
important message they want to trace, is both an easier and a more
likely attack than wiretapping all the links in real time in
anticipation of an important message -- unless the remailer
operators snoop-proof their logs.
Also keep in mind that, given the lack of a good user interface,
there is currently too little properly encrypted and nested remailer
traffic to create anything approaching a true digital mix.
Jim Hart
hart@chaos.bsu.edu