You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Karl Fogel <kf...@newton.ch.collab.net> on 2003/01/14 21:26:45 UTC
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
svn-agent is not an equivalent substitute for on-disk caching. Y'all
hang out with too technical a crowd or something, I think :-). Many
people will not remember to start up their agents. People will not
understand how it works. They will be confused, then they will read
the docs and be more confused. (I know plenty of people who can't get
ssh-agent working right. They're not stupid, they just have other
things to pay attention to.)
"Oh, Subversion itself will start the agent!" But what if one's
already running? "It'll cache a PID file in ~/.subversion/, and
Subversion will check there first!" And what if the machine crashes
before that PID file has a chance to be removed? "Well, if Subversion
tries to contact the process and can't, it'll erase the PID file and
start a new one." So superuser could crash the machine and restart
with a trojan horse at the old agent PID, and harvest passwords? Or
heck, why not just replace the binary with a trojan horse in the first
place, if you're root?
In other words, this is not really any more secure than just storing
the auth data in a ~/.subversion/ area readable only by the user (I
hope & assume Windows per-user registries are equally secure.)
But it *is* a lot more complex, in terms of portability, code
complexity, and number of things that can go wrong at run time.
What's our win here? I'm not seeing it.
-Karl
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by Colin Watson <cj...@flatline.org.uk>.
On Thu, Jan 16, 2003 at 08:42:18AM -0800, rbb@rkbloom.net wrote:
> On Thu, 16 Jan 2003, Colin Watson wrote:
> > On Thu, Jan 16, 2003 at 08:18:55AM -0800, rbb@rkbloom.net wrote:
> > > Yes, ra_svn over SSH is exactly what CVS does. ra_dav + SSL
> > > without auth-caching is also perfectly secure. The only remaining
> > > problem (once passwords are moved out of the wc, is that the
> > > default is insecure, and the docs glance over the issue. The
> > > reason that svn-agent came up at all is because people want both
> > > security and auth-caching, which requires something like
> > > svn-agent.
> >
> > If ra_svn is tunnelled over ssh, why can it not consider ssh to have
> > already performed authentication (if indeed it doesn't use this
> > logic already)? If so, ssh-agent is sufficient and indeed
> > preferable; you'll usually need it anyway to stop ssh prompting for
> > authentication.
>
> If you use ra_svn over ssh, ssh does do the authentication. And,
> ssh-agent is required to get passphrase caching.
Right, so then this should be documented in big flashing lights as the
way to get security and auth-caching at the same time.
--
Colin Watson [cjwatson@flatline.org.uk]
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by rb...@rkbloom.net.
On Thu, 16 Jan 2003, Colin Watson wrote:
> On Thu, Jan 16, 2003 at 08:18:55AM -0800, rbb@rkbloom.net wrote:
> > On 16 Jan 2003 cmpilato@collab.net wrote:
> > > IANA-SecurityGuy, but. Can't ra_svn be SSH-tunneled? If so, then it
> > > would seem that that is a good mapping to CVS using :ext:SSH. And by
> > > using mod_dav + SSL and disabling auth caching altogether, isn't that
> > > an exact match of CVS's most secure model?
> >
> > Yes, ra_svn over SSH is exactly what CVS does. ra_dav + SSL without
> > auth-caching is also perfectly secure. The only remaining problem (once
> > passwords are moved out of the wc, is that the default is insecure, and
> > the docs glance over the issue. The reason that svn-agent came up at all
> > is because people want both security and auth-caching, which requires
> > something like svn-agent.
>
> If ra_svn is tunnelled over ssh, why can it not consider ssh to have
> already performed authentication (if indeed it doesn't use this logic
> already)? If so, ssh-agent is sufficient and indeed preferable; you'll
> usually need it anyway to stop ssh prompting for authentication.
If you use ra_svn over ssh, ssh does do the authentication. And,
ssh-agent is required to get passphrase caching.
Ryan
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by Colin Watson <cj...@flatline.org.uk>.
On Thu, Jan 16, 2003 at 08:18:55AM -0800, rbb@rkbloom.net wrote:
> On 16 Jan 2003 cmpilato@collab.net wrote:
> > IANA-SecurityGuy, but. Can't ra_svn be SSH-tunneled? If so, then it
> > would seem that that is a good mapping to CVS using :ext:SSH. And by
> > using mod_dav + SSL and disabling auth caching altogether, isn't that
> > an exact match of CVS's most secure model?
>
> Yes, ra_svn over SSH is exactly what CVS does. ra_dav + SSL without
> auth-caching is also perfectly secure. The only remaining problem (once
> passwords are moved out of the wc, is that the default is insecure, and
> the docs glance over the issue. The reason that svn-agent came up at all
> is because people want both security and auth-caching, which requires
> something like svn-agent.
If ra_svn is tunnelled over ssh, why can it not consider ssh to have
already performed authentication (if indeed it doesn't use this logic
already)? If so, ssh-agent is sufficient and indeed preferable; you'll
usually need it anyway to stop ssh prompting for authentication.
--
Colin Watson [cjwatson@flatline.org.uk]
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by cm...@collab.net.
mark benedetto king <mb...@earthquake.boredom.org> writes:
> Moving the passwords into ~/.subversion should probably block on
> repos UUIDs, which I'm working on right now.
+1.
> I have the FS side implemented.
Coolness. Can't wait to see the patches.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by cm...@collab.net.
<rb...@rkbloom.net> writes:
> Just so I understand it, what is the relationship between the UUID's and
> the passwords? Is it making ~/.subversion understand different repos, or
> is there some other relationship that I am missing?
Yah. Given Subversion's public URL scheme (which it uses for its
resource identifiers), there's no way to tell if two things are in the
same repository.
Given 'http://server/foo/bar/A' and 'http://server/foo/baz/B' might
both be in the repository whose root is at 'http://server/foo', or
they could be in two different ones (at 'http://server/foo/bar' and
http://server/foo/baz', respectively).
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by rb...@rkbloom.net.
On Thu, 16 Jan 2003, mark benedetto king wrote:
> On Thu, Jan 16, 2003 at 08:27:49AM -0800, rbb@rkbloom.net wrote:
> >
> > Which was the original plan, yes. The only hard part about this is moving
> > the passwords into ~/.subversion, because that directory currently doesn't
> > understand multiple repositories. I haven't had any time to work on this
> > yet. If somebody else wants to go ahead and implement it, I won't be
> > upset, otherwise it is likely to take me a few weeks to finish. I get at
> > most an hour a night to code.
> >
>
> Moving the passwords into ~/.subversion should probably block on
> repos UUIDs, which I'm working on right now. I have the FS side implemented.
Just so I understand it, what is the relationship between the UUID's and
the passwords? Is it making ~/.subversion understand different repos, or
is there some other relationship that I am missing?
Ryan
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by mark benedetto king <mb...@earthquake.boredom.org>.
On Thu, Jan 16, 2003 at 08:27:49AM -0800, rbb@rkbloom.net wrote:
>
> Which was the original plan, yes. The only hard part about this is moving
> the passwords into ~/.subversion, because that directory currently doesn't
> understand multiple repositories. I haven't had any time to work on this
> yet. If somebody else wants to go ahead and implement it, I won't be
> upset, otherwise it is likely to take me a few weeks to finish. I get at
> most an hour a night to code.
>
Moving the passwords into ~/.subversion should probably block on
repos UUIDs, which I'm working on right now. I have the FS side implemented.
I've decided to publish the UUIDs to the client as entry metadata.
Basically, every entry has a URL (or an implicit one inherited from
its parent dir), and every entry should have a repo UUID as well.
This, IMO, accurately models reality (always nice), and will be reasonably
easy to implement. As an added bonus, it makes the "svn switch --rewrite-urls"
validation easy, too.
--ben
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
RE: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by rb...@rkbloom.net.
On Thu, 16 Jan 2003, Sander Striker wrote:
> > From: rbb@rkbloom.net [mailto:rbb@rkbloom.net]
> > Sent: Thursday, January 16, 2003 5:19 PM
>
> > On 16 Jan 2003 cmpilato@collab.net wrote:
> >
> >> <rb...@rkbloom.net> writes:
> >>
> >>> I still disagree that implementing #2 brings us to the same point as CVS.
> >>> CVS only caches passwords to your disk if you are using :pserver:, which
> >>> most sites just don't do unless they are offering anonymous CVS. (Yes,
> >>> there are some that do, but it is rare).
> >>>
> >>> What that means, is that by implementing #2, you have brought subversion
> >>> up to the very least that CVS does. This makes subversion useful for
> >>> public access, but leaves it unsuitable for use with private passwords.
> >>> Emulating a feature of CVS that most people consider to be a security
> >>> problem does not sound like the correct way to replace CVS.
> >>
> >> IANA-SecurityGuy, but. Can't ra_svn be SSH-tunneled? If so, then it
> >> would seem that that is a good mapping to CVS using :ext:SSH. And by
> >> using mod_dav + SSL and disabling auth caching altogether, isn't that
> >> an exact match of CVS's most secure model?
> >
> > Yes, ra_svn over SSH is exactly what CVS does. ra_dav + SSL without
> > auth-caching is also perfectly secure. The only remaining problem (once
> > passwords are moved out of the wc, is that the default is insecure, and
> > the docs glance over the issue. The reason that svn-agent came up at all
> > is because people want both security and auth-caching, which requires
> > something like svn-agent.
>
> Right. So to improve fast, move to option #2 (and change the default to
> no-caching), and when svn-agent is done go for option #3.
Which was the original plan, yes. The only hard part about this is moving
the passwords into ~/.subversion, because that directory currently doesn't
understand multiple repositories. I haven't had any time to work on this
yet. If somebody else wants to go ahead and implement it, I won't be
upset, otherwise it is likely to take me a few weeks to finish. I get at
most an hour a night to code.
Ryan
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
RE: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by Sander Striker <st...@apache.org>.
> From: rbb@rkbloom.net [mailto:rbb@rkbloom.net]
> Sent: Thursday, January 16, 2003 5:19 PM
> On 16 Jan 2003 cmpilato@collab.net wrote:
>
>> <rb...@rkbloom.net> writes:
>>
>>> I still disagree that implementing #2 brings us to the same point as CVS.
>>> CVS only caches passwords to your disk if you are using :pserver:, which
>>> most sites just don't do unless they are offering anonymous CVS. (Yes,
>>> there are some that do, but it is rare).
>>>
>>> What that means, is that by implementing #2, you have brought subversion
>>> up to the very least that CVS does. This makes subversion useful for
>>> public access, but leaves it unsuitable for use with private passwords.
>>> Emulating a feature of CVS that most people consider to be a security
>>> problem does not sound like the correct way to replace CVS.
>>
>> IANA-SecurityGuy, but. Can't ra_svn be SSH-tunneled? If so, then it
>> would seem that that is a good mapping to CVS using :ext:SSH. And by
>> using mod_dav + SSL and disabling auth caching altogether, isn't that
>> an exact match of CVS's most secure model?
>
> Yes, ra_svn over SSH is exactly what CVS does. ra_dav + SSL without
> auth-caching is also perfectly secure. The only remaining problem (once
> passwords are moved out of the wc, is that the default is insecure, and
> the docs glance over the issue. The reason that svn-agent came up at all
> is because people want both security and auth-caching, which requires
> something like svn-agent.
Right. So to improve fast, move to option #2 (and change the default to
no-caching), and when svn-agent is done go for option #3.
Sander
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by rb...@rkbloom.net.
On 16 Jan 2003 cmpilato@collab.net wrote:
> <rb...@rkbloom.net> writes:
>
> > I still disagree that implementing #2 brings us to the same point as CVS.
> > CVS only caches passwords to your disk if you are using :pserver:, which
> > most sites just don't do unless they are offering anonymous CVS. (Yes,
> > there are some that do, but it is rare).
> >
> > What that means, is that by implementing #2, you have brought subversion
> > up to the very least that CVS does. This makes subversion useful for
> > public access, but leaves it unsuitable for use with private passwords.
> > Emulating a feature of CVS that most people consider to be a security
> > problem does not sound like the correct way to replace CVS.
>
> IANA-SecurityGuy, but. Can't ra_svn be SSH-tunneled? If so, then it
> would seem that that is a good mapping to CVS using :ext:SSH. And by
> using mod_dav + SSL and disabling auth caching altogether, isn't that
> an exact match of CVS's most secure model?
Yes, ra_svn over SSH is exactly what CVS does. ra_dav + SSL without
auth-caching is also perfectly secure. The only remaining problem (once
passwords are moved out of the wc, is that the default is insecure, and
the docs glance over the issue. The reason that svn-agent came up at all
is because people want both security and auth-caching, which requires
something like svn-agent.
Ryan
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by cm...@collab.net.
<rb...@rkbloom.net> writes:
> I still disagree that implementing #2 brings us to the same point as CVS.
> CVS only caches passwords to your disk if you are using :pserver:, which
> most sites just don't do unless they are offering anonymous CVS. (Yes,
> there are some that do, but it is rare).
>
> What that means, is that by implementing #2, you have brought subversion
> up to the very least that CVS does. This makes subversion useful for
> public access, but leaves it unsuitable for use with private passwords.
> Emulating a feature of CVS that most people consider to be a security
> problem does not sound like the correct way to replace CVS.
IANA-SecurityGuy, but. Can't ra_svn be SSH-tunneled? If so, then it
would seem that that is a good mapping to CVS using :ext:SSH. And by
using mod_dav + SSL and disabling auth caching altogether, isn't that
an exact match of CVS's most secure model?
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by Greg Stein <gs...@lyra.org>.
Just configure your server to only do Digest auth, and Neon will do that
rather than Basic. To the svn user, the two are exactly the same (user and
password).
Hmm. In fact, we should probably change our documentation examples to use
Digest configuration instead of Basic. Since we're based on Apache 2.0, then
it is known to be available (well, "most likely" available). If we get
feedback that most prepackaged Apache's aren't enabling mod_auth_digest,
then we can add an appendix or something to show how to configure Basic.
Cheers,
-g
On Thu, Jan 16, 2003 at 10:34:31AM -0700, David Waite wrote:
> Yes.
>
> -David Waite
>
> rbb@rkbloom.net wrote:
>
> >Oh, this also begs another question, does svn (or Neon) support Digest
> >authentication? Much better from a security POV.
> >
> >Ryan
--
Greg Stein, http://www.lyra.org/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by David Waite <ma...@akuma.org>.
Yes.
-David Waite
rbb@rkbloom.net wrote:
>Oh, this also begs another question, does svn (or Neon) support Digest
>authentication? Much better from a security POV.
>
>Ryan
>
>
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by rb...@rkbloom.net.
Oh, this also begs another question, does svn (or Neon) support Digest
authentication? Much better from a security POV.
Ryan
On Thu, 16 Jan 2003, Jani Averbach wrote:
> On 16 Jan 2003, Karl Fogel wrote:
>
> > It would be useful to have a catalog of our auth methods and how we
> > store their data, at this point. We've got:
> >
> > http basic auth -- Stored in cleartext, and goes over the wire in
> > cleartext too. Basically like CVS pserver.
> > At least, it *will* be, once we stop storing
> > the auth data in the working copy! :-)
> >
> > ssl -- Stored in cleartext, but goes over the wire
> > encrypted. Like CVS pserver over an ssh
> > tunnel. Secure as far as the network goes,
> > not so secure from attacks based on the client
> > host.
> >
> > ra_svn ssh tunnel -- No cleartext password stored on client side;
> > basically, this is like CVS's ":ext:" protocol
> > when run over ssh. It's only ra_svn, though,
> > right?
> >
> > Anything else? (Not a rhetorical question; quite possible I've missed
> > stuff here, or misunderstood our existing options, in which case
> > please educate me.)
> >
>
> http basic auth
> over ssh tunnel -- Much more secure at the moment than ssl. We
> don't have at the moment any kind certificate
> handling, right?
> And you will need only one ssh accout (if you
> like), thanks to the http auth
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by Jani Averbach <ja...@cc.jyu.fi>.
On 16 Jan 2003, Karl Fogel wrote:
> It would be useful to have a catalog of our auth methods and how we
> store their data, at this point. We've got:
>
> http basic auth -- Stored in cleartext, and goes over the wire in
> cleartext too. Basically like CVS pserver.
> At least, it *will* be, once we stop storing
> the auth data in the working copy! :-)
>
> ssl -- Stored in cleartext, but goes over the wire
> encrypted. Like CVS pserver over an ssh
> tunnel. Secure as far as the network goes,
> not so secure from attacks based on the client
> host.
>
> ra_svn ssh tunnel -- No cleartext password stored on client side;
> basically, this is like CVS's ":ext:" protocol
> when run over ssh. It's only ra_svn, though,
> right?
>
> Anything else? (Not a rhetorical question; quite possible I've missed
> stuff here, or misunderstood our existing options, in which case
> please educate me.)
>
http basic auth
over ssh tunnel -- Much more secure at the moment than ssl. We
don't have at the moment any kind certificate
handling, right?
And you will need only one ssh accout (if you
like), thanks to the http auth
BR, Jani
--
Jani Averbach
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by Karl Fogel <kf...@newton.ch.collab.net>.
<rb...@rkbloom.net> writes:
> There is nothing unreasonable in that statement. The only remaining
> question is what the default should be, which is where a big portion of
> the discussion has been. If we get svn-agent working, then the default
> for ra_dav can be to store the password on disk, because we have an easy
> way to make auth caching secure. If we don't get svn-agent working, then
> the default should be not storing the password on disk, because there is
> no easy way to make authentication caching secure. All my opinion of
> course.
Cool -- and I completely agree.
(We can keep convenience by prompting to ask whether or not to store
the password, at the same time we prompt for the password itself, as
described in an earlier mail.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by rb...@rkbloom.net.
<Last response from me on this.>
I agree with everything in this message. I will be working on patches to
implement it, but they will come slowly.
On 16 Jan 2003, Karl Fogel wrote:
> .....
> I feel like what I'm saying is not controversial; yet somehow people
> keep coming back with slightly orthogonal responses. Is there
> anything unreasonable in the statement
>
> Subversion should offer at least the same range of
> security/convenience tradeoffs as CVS does.
There is nothing unreasonable in that statement. The only remaining
question is what the default should be, which is where a big portion of
the discussion has been. If we get svn-agent working, then the default
for ra_dav can be to store the password on disk, because we have an easy
way to make auth caching secure. If we don't get svn-agent working, then
the default should be not storing the password on disk, because there is
no easy way to make authentication caching secure. All my opinion of
course.
(Honestly, this conversation got out of hand.)
Ryan
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by Karl Fogel <kf...@newton.ch.collab.net>.
<rb...@rkbloom.net> writes:
> I still disagree that implementing #2 brings us to the same point as CVS.
> CVS only caches passwords to your disk if you are using :pserver:, which
> most sites just don't do unless they are offering anonymous CVS. (Yes,
> there are some that do, but it is rare).
Pserver is one particular point in the security/convenience continuum.
Many sites use it (and no, not only public sites, by the way; I've
seen plenty that use it for private data). No, it's not the most
secure, but it's secure enough for a lot of uses, and SVN has to to
offer an equivalent.
> What that means, is that by implementing #2, you have brought subversion
> up to the very least that CVS does. This makes subversion useful for
> public access, but leaves it unsuitable for use with private passwords.
> Emulating a feature of CVS that most people consider to be a security
> problem does not sound like the correct way to replace CVS.
It would be useful to have a catalog of our auth methods and how we
store their data, at this point. We've got:
http basic auth -- Stored in cleartext, and goes over the wire in
cleartext too. Basically like CVS pserver.
At least, it *will* be, once we stop storing
the auth data in the working copy! :-)
ssl -- Stored in cleartext, but goes over the wire
encrypted. Like CVS pserver over an ssh
tunnel. Secure as far as the network goes,
not so secure from attacks based on the client
host.
ra_svn ssh tunnel -- No cleartext password stored on client side;
basically, this is like CVS's ":ext:" protocol
when run over ssh. It's only ra_svn, though,
right?
Anything else? (Not a rhetorical question; quite possible I've missed
stuff here, or misunderstood our existing options, in which case
please educate me.)
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by rb...@rkbloom.net.
On 16 Jan 2003, Karl Fogel wrote:
> Benjamin Pflugmann <be...@pflugmann.de> writes:
> > No, I meant the way ssh-agent works: encrypted key on disk, plain text
> > key (or passphrase) in memory. Else, you would have to type your
> > passphrase each time it is accessed and nothing would be won (in terms
> > of usability).
>
> Sorry; of course you're right. But the point remains that now we've
> introduced another passphrase into the user experience.
>
> Hmmm. We're getting a bit off-track here, as far as immediate,
> pre-1.0 concerns for Subversion. All I'm trying to say is:
>
> 1. We currently store passwords in the working copy. Ryan has
> correctly pointed out that this is bad, because it makes sharing
> working copies a security compromise.
>
> 2. If we store the auth data into ~/.subversion/ (or equivalent)
> instead, we move up to the same security as CVS, which, after
> all, is the software we're trying to replace.
>
> 3. With svn-agent, we can get even more security, but at the cost
> of additional inconvenience for users and additional development
> and maintenance complexity for ourselves. Therefore, although
> svn-agent is certainly a good idea, it cannot be our only 1.0
> security option. We should first support option (2), because
> that's the security/convenience tradeoff people are accustomed
> to from CVS (and it's a sensible one, for this application).
>
> That's the big picture, IMHO.
I still disagree that implementing #2 brings us to the same point as CVS.
CVS only caches passwords to your disk if you are using :pserver:, which
most sites just don't do unless they are offering anonymous CVS. (Yes,
there are some that do, but it is rare).
What that means, is that by implementing #2, you have brought subversion
up to the very least that CVS does. This makes subversion useful for
public access, but leaves it unsuitable for use with private passwords.
Emulating a feature of CVS that most people consider to be a security
problem does not sound like the correct way to replace CVS.
Ryan
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by Karl Fogel <kf...@newton.ch.collab.net>.
Benjamin Pflugmann <be...@pflugmann.de> writes:
> No, I meant the way ssh-agent works: encrypted key on disk, plain text
> key (or passphrase) in memory. Else, you would have to type your
> passphrase each time it is accessed and nothing would be won (in terms
> of usability).
Sorry; of course you're right. But the point remains that now we've
introduced another passphrase into the user experience.
Hmmm. We're getting a bit off-track here, as far as immediate,
pre-1.0 concerns for Subversion. All I'm trying to say is:
1. We currently store passwords in the working copy. Ryan has
correctly pointed out that this is bad, because it makes sharing
working copies a security compromise.
2. If we store the auth data into ~/.subversion/ (or equivalent)
instead, we move up to the same security as CVS, which, after
all, is the software we're trying to replace.
3. With svn-agent, we can get even more security, but at the cost
of additional inconvenience for users and additional development
and maintenance complexity for ourselves. Therefore, although
svn-agent is certainly a good idea, it cannot be our only 1.0
security option. We should first support option (2), because
that's the security/convenience tradeoff people are accustomed
to from CVS (and it's a sensible one, for this application).
That's the big picture, IMHO.
-K
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by Benjamin Pflugmann <be...@pflugmann.de>.
On Wed 2003-01-15 at 22:58:06 -0600, Karl Fogel wrote:
> Benjamin Pflugmann <be...@pflugmann.de> writes:
> > Although I am no security expert, I am a bit interested in that topic.
> > The claim you made (that a plain text password in a read-protected
> > file is equally secure as an encypted one which is only held in plain
> > in memory) sounds absurd to me. For a start, the exposure time is not
> > the same: The one is always available, even after you deleted it
> > (backups), while the other is only available while you are working
> > with it (the encrypted form is always avaiable, but let's hope that
> > the encryption is good and your passphrase, too).
>
> If the password is stored in memory *encrypted*, then yes, it is more
> secure.
No, I meant the way ssh-agent works: encrypted key on disk, plain text
key (or passphrase) in memory. Else, you would have to type your
passphrase each time it is accessed and nothing would be won (in terms
of usability).
If you don't want to use a key, but only password authentication, it
would store nothing on disk and simply cache your plain text password
in memory.
> But it is also less convenient, because now we've introduced one
> more passphrase into the user's experience -- so we're no longer
> comparing the same things.
It sounds as if you meant the simple password caching case
originally. It doesn't matter, my point was about the storing
mechanism: short time in-memory vs. long time on disk.
> If we don't introduce that extra passphrase, then the security is
> comparable between the two schemes, though the agent password is
> harder to get at, in terms of inconvenience for the attacker.
IMNSHO, it isn't. It is a difference by at least an order of
magnitude.
For a start, "harder to get at" is all what security is about. There
is no way to prevent 100% that someone gets at your data. You can only
make it as hard as possible (and reasonable). So you effectively just
said yourself that it is more secure.
Also, the window of exposure that I mentioned in my last mail, should
be enough to make it obvious. If it is cached in memory after I typed
it, an attacker has to target my session and his attack has to work
the first time or he may not get a second try (if I notice the attack
due to a crash or something). If it is saved on disk, he can attack
any time he likes, even if I am on vacation for a month.
Benjamin.
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by Karl Fogel <kf...@newton.ch.collab.net>.
Benjamin Pflugmann <be...@pflugmann.de> writes:
> Although I am no security expert, I am a bit interested in that topic.
> The claim you made (that a plain text password in a read-protected
> file is equally secure as an encypted one which is only held in plain
> in memory) sounds absurd to me. For a start, the exposure time is not
> the same: The one is always available, even after you deleted it
> (backups), while the other is only available while you are working
> with it (the encrypted form is always avaiable, but let's hope that
> the encryption is good and your passphrase, too).
If the password is stored in memory *encrypted*, then yes, it is more
secure. But it is also less convenient, because now we've introduced
one more passphrase into the user's experience -- so we're no longer
comparing the same things.
If we don't introduce that extra passphrase, then the security is
comparable between the two schemes, though the agent password is
harder to get at, in terms of inconvenience for the attacker.
> As a last word: I do not claim that subversion has to handle
> authentication data as secure as ssh does (although I can see the case
> for this) or that the complexity would be worth it. But I strongly
> resent the idea that storing a plain text password would be equally
> secure.
Sorry, I never meant to claim that storing a plaintext password is the
same as storing an encrypted one. I only meant to say that storing a
plaintext password is the same as storing a plaintext one.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by Benjamin Pflugmann <be...@pflugmann.de>.
On Tue 2003-01-14 at 15:26:45 -0600, Karl Fogel wrote:
> svn-agent is not an equivalent substitute for on-disk caching. Y'all
> hang out with too technical a crowd or something, I think :-). Many
> people will not remember to start up their agents. People will not
> understand how it works. They will be confused, then they will read
> the docs and be more confused. (I know plenty of people who can't get
> ssh-agent working right. They're not stupid, they just have other
> things to pay attention to.)
>
> "Oh, Subversion itself will start the agent!" But what if one's
> already running? "It'll cache a PID file in ~/.subversion/, and
> Subversion will check there first!" And what if the machine crashes
> before that PID file has a chance to be removed? "Well, if Subversion
> tries to contact the process and can't, it'll erase the PID file and
> start a new one." So superuser could crash the machine and restart
> with a trojan horse at the old agent PID, and harvest passwords? Or
> heck, why not just replace the binary with a trojan horse in the first
> place, if you're root?
>
> In other words, this is not really any more secure than just storing
> the auth data in a ~/.subversion/ area readable only by the user (I
> hope & assume Windows per-user registries are equally secure.)
I earnestly hope you are not serious about this. If you are, I am not
sure where to start, because your basic assumptions are miles away
from the consensus of any security-related discussion I have been
involved with. No offense meant.
Although I am no security expert, I am a bit interested in that topic.
The claim you made (that a plain text password in a read-protected
file is equally secure as an encypted one which is only held in plain
in memory) sounds absurd to me. For a start, the exposure time is not
the same: The one is always available, even after you deleted it
(backups), while the other is only available while you are working
with it (the encrypted form is always avaiable, but let's hope that
the encryption is good and your passphrase, too).
You are right, if your root is after it, there is no way to stop him.
You can only try to make it harder. So using a malicious root as
argument is void, IMHO.
For example, I trust my root not to sniff memory for the data
ssh-agent stores for me. But do I trust everybody who has access to
the backups that are made of home dirs? Even after when I am gone from
that company or another sysadmin has been hired? Of course not.
If someone manages to break into your account. With plain password you
have already lost, with the encrypted one intruder has still to set up
some way to get it. And so on.
I am not done yet. There are a lot of other points I am not going into
yet. I hope you see where I am getting at, anyhow.
As a last word: I do not claim that subversion has to handle
authentication data as secure as ssh does (although I can see the case
for this) or that the complexity would be worth it. But I strongly
resent the idea that storing a plain text password would be equally
secure.
Benjamin.
Re: Writing svn-agent
Posted by Zack Weinberg <za...@codesourcery.com>.
Karl Fogel <kf...@newton.ch.collab.net> writes:
> (Among other things, we shouldn't be paying the same price to
> protect svn passwords as to protect login passwords, because they're
> not as valuable as login passwords.)
That depends on the situation. If someone steals my login password
they can read my email, which is mostly publicly archived mailing
lists so big deal, and they can delete my files, which would only be
an inconvenience; I have backups. If someone steals my SSH passphrase
they could trash sources.redhat.com, which is also backed up but would
inconvenience hundreds of people all over the world instead of just
me; or they could inject malicious code into one of the repositories
there, which would be a major catastrophe.
No prize for guessing which of these is more closely held.
zw
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by Karl Fogel <kf...@newton.ch.collab.net>.
<rb...@rkbloom.net> writes:
> Well, how are your friends who can't get ssh-agent working using ssh? Do
> they have passphrases in their keys? If so, then they are typing their
> passwords all the time.
No, they just type their regular passwords, which are much shorter.
But that's comparing apples and oranges, I don't think it's relevant
to this discussion. (Among other things, we shouldn't be paying the
same price to protect svn passwords as to protect login passwords,
because they're not as valuable as login passwords.)
> If not, then svn-agent will work the same way as ssh-agent.
>
> The only place that svn-agent doesn't replace auth caching, is if you
> aren't using client-side certs, and you don't have svn-agent running.
> If we can make svn-agent easy to use though, this case will go away.
Sure, I understand the advantages of an agent. But still, my question
is: "What's our win here?" I don't see much of a security gain, and I
do see a convenience loss (both for users and for code developers).
There is no way we can make it as easy to use as straight disk
caching. We can do our best, but it will *always* be more error
prone, because it's got more surfaces.
Obviously, you can work on what you want, and I don't think the agent
is a bad idea. It's just not a solution to our current auth problems,
IMHO.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Writing svn-agent (Was Re: [PATCH] default to --no-auth-cache)
Posted by rb...@rkbloom.net.
Well, how are your friends who can't get ssh-agent working using ssh? Do
they have passphrases in their keys? If so, then they are typing their
passwords all the time. If not, then svn-agent will work the same way as
ssh-agent.
The only place that svn-agent doesn't replace auth caching, is if you
aren't using client-side certs, and you don't have svn-agent running.
If we can make svn-agent easy to use though, this case will go away.
Ryan
On 14 Jan 2003, Karl Fogel wrote:
> svn-agent is not an equivalent substitute for on-disk caching. Y'all
> hang out with too technical a crowd or something, I think :-). Many
> people will not remember to start up their agents. People will not
> understand how it works. They will be confused, then they will read
> the docs and be more confused. (I know plenty of people who can't get
> ssh-agent working right. They're not stupid, they just have other
> things to pay attention to.)
>
> "Oh, Subversion itself will start the agent!" But what if one's
> already running? "It'll cache a PID file in ~/.subversion/, and
> Subversion will check there first!" And what if the machine crashes
> before that PID file has a chance to be removed? "Well, if Subversion
> tries to contact the process and can't, it'll erase the PID file and
> start a new one." So superuser could crash the machine and restart
> with a trojan horse at the old agent PID, and harvest passwords? Or
> heck, why not just replace the binary with a trojan horse in the first
> place, if you're root?
>
> In other words, this is not really any more secure than just storing
> the auth data in a ~/.subversion/ area readable only by the user (I
> hope & assume Windows per-user registries are equally secure.)
>
> But it *is* a lot more complex, in terms of portability, code
> complexity, and number of things that can go wrong at run time.
>
> What's our win here? I'm not seeing it.
>
> -Karl
>
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org