In the /prod/PID subset of procfs, why are the exe, cwd, and root symlinks
considered priviledged information?
Exe is the big one for me, as this one can be usually infered from reading
/prod/PID/maps. Root I guess can't be inferred in any unpriviledged way,
and neither can cwd. At any rate.. I am not sure behind the philosophy to
make these symlinks' destinations priviledged... can someone clarify
this?
Thanks,
-Calin
On Thu, Nov 07, 2002 at 10:57:06AM -0500, Calin A. Culianu wrote:
>
> In the /prod/PID subset of procfs, why are the exe, cwd, and root symlinks
> considered priviledged information?
>
> Exe is the big one for me, as this one can be usually infered from reading
> /prod/PID/maps. Root I guess can't be inferred in any unpriviledged way,
> and neither can cwd. At any rate.. I am not sure behind the philosophy to
> make these symlinks' destinations priviledged... can someone clarify
> this?
This came up a little while ago. The answer is that maps should be
priviledged also.
For instance:
You can protect a directory by giving its parent directory no read
permissions. The name of the directory is now secret. You don't want
to reveal it in cwd.
--
Daniel Jacobowitz
MontaVista Software Debian GNU/Linux Developer
On Thu, Nov 07, 2002 at 11:05:21AM -0500, Daniel Jacobowitz wrote:
> On Thu, Nov 07, 2002 at 10:57:06AM -0500, Calin A. Culianu wrote:
> >
> > In the /prod/PID subset of procfs, why are the exe, cwd, and root symlinks
> > considered priviledged information?
> >
> > Exe is the big one for me, as this one can be usually infered from reading
> > /prod/PID/maps. Root I guess can't be inferred in any unpriviledged way,
> > and neither can cwd. At any rate.. I am not sure behind the philosophy to
> > make these symlinks' destinations priviledged... can someone clarify
> > this?
>
> This came up a little while ago. The answer is that maps should be
> priviledged also.
>
> For instance:
> You can protect a directory by giving its parent directory no read
> permissions. The name of the directory is now secret. You don't want
> to reveal it in cwd.
>
Daniel is correct in that the issue came up recently. He
gives _his_ answer above. If you believe in security
through obscurity you will agree with him. I don't.
I will agree that there should be no real reason to need
access to this information.
With ACLs you will be able to explicitly grant access and
you won't have to depend on keeping shared info secret.
Then this will be less of an issue.
--
________________________________________________________________
J.W. Schultz Pegasystems Technologies
email address: [email protected]
Remember Cernan and Schmitt
On Thu, Nov 07, 2002 at 02:16:15PM -0800, jw schultz wrote:
> On Thu, Nov 07, 2002 at 11:05:21AM -0500, Daniel Jacobowitz wrote:
> > On Thu, Nov 07, 2002 at 10:57:06AM -0500, Calin A. Culianu wrote:
> > >
> > > In the /prod/PID subset of procfs, why are the exe, cwd, and root symlinks
> > > considered priviledged information?
> > >
> > > Exe is the big one for me, as this one can be usually infered from reading
> > > /prod/PID/maps. Root I guess can't be inferred in any unpriviledged way,
> > > and neither can cwd. At any rate.. I am not sure behind the philosophy to
> > > make these symlinks' destinations priviledged... can someone clarify
> > > this?
> >
> > This came up a little while ago. The answer is that maps should be
> > priviledged also.
> >
> > For instance:
> > You can protect a directory by giving its parent directory no read
> > permissions. The name of the directory is now secret. You don't want
> > to reveal it in cwd.
> >
>
> Daniel is correct in that the issue came up recently. He
> gives _his_ answer above. If you believe in security
> through obscurity you will agree with him. I don't.
> I will agree that there should be no real reason to need
> access to this information.
>
> With ACLs you will be able to explicitly grant access and
> you won't have to depend on keeping shared info secret.
> Then this will be less of an issue.
I recommend you go think about what security through obscurity actually
_means_. If you think that an unreadable directory and a
randomly-generated subdirectory is security through obscurity, then in
what way is it actually different from a _password_? That's what it
is.
Yes, this is poor-man's-ACLs. It works in a lot of places when ACLs
won't. For instance, an anonymous FTP server...
--
Daniel Jacobowitz
MontaVista Software Debian GNU/Linux Developer
On Thursday 07 November 2002 04:28 pm, Daniel Jacobowitz wrote:
> On Thu, Nov 07, 2002 at 02:16:15PM -0800, jw schultz wrote:
> > On Thu, Nov 07, 2002 at 11:05:21AM -0500, Daniel Jacobowitz wrote:
> > > On Thu, Nov 07, 2002 at 10:57:06AM -0500, Calin A. Culianu wrote:
> > > > In the /prod/PID subset of procfs, why are the exe, cwd, and root
> > > > symlinks considered priviledged information?
> > > >
> > > > Exe is the big one for me, as this one can be usually infered from
> > > > reading /prod/PID/maps. Root I guess can't be inferred in any
> > > > unpriviledged way, and neither can cwd. At any rate.. I am not sure
> > > > behind the philosophy to make these symlinks' destinations
> > > > priviledged... can someone clarify this?
> > >
> > > This came up a little while ago. The answer is that maps should be
> > > priviledged also.
> > >
> > > For instance:
> > > You can protect a directory by giving its parent directory no read
> > > permissions. The name of the directory is now secret. You don't want
> > > to reveal it in cwd.
> >
> > Daniel is correct in that the issue came up recently. He
> > gives _his_ answer above. If you believe in security
> > through obscurity you will agree with him. I don't.
> > I will agree that there should be no real reason to need
> > access to this information.
> >
> > With ACLs you will be able to explicitly grant access and
> > you won't have to depend on keeping shared info secret.
> > Then this will be less of an issue.
>
> I recommend you go think about what security through obscurity actually
> _means_. If you think that an unreadable directory and a
> randomly-generated subdirectory is security through obscurity, then in
> what way is it actually different from a _password_? That's what it
> is.
>
> Yes, this is poor-man's-ACLs. It works in a lot of places when ACLs
> won't. For instance, an anonymous FTP server...
It also isn't necessarily just for obscurity.
If the directory contains customer contact lists, where each customer
is represented by a directory entry, and all communications to/from that
customer is stored in files. You don't want the cwd to be exposed by
a process that may set its' workspace to a specific customer.
That is NOT security by obscurity. Exposure is called a "leak".
And it is entirely possible for sensitive information to be embeded
in pathnames, process names, and parameters (most often).
--
-------------------------------------------------------------------------
Jesse I Pollard, II
Email: [email protected]
Any opinions expressed are solely my own.
jw schultz wrote:
>On Thu, Nov 07, 2002 at 11:05:21AM -0500, Daniel Jacobowitz wrote:
>> For instance:
>> You can protect a directory by giving its parent directory no read
>> permissions. The name of the directory is now secret. You don't want
>> to reveal it in cwd.
>
>Daniel is correct in that the issue came up recently. He
>gives _his_ answer above. If you believe in security
>through obscurity you will agree with him.
Huh? This has nothing to do with security through obscurity.
This has to do with the Unix permissions model. /proc/pid probably
shouldn't leak information that would otherwise be kept confidential in the
standard Unix permissions model (principle of least surprise, etc.).
>I will agree that there should be no real reason to need
>access to this information.
They why is it allowed? Allowing something that noone needs and
that breaks most people's intuitions of how security will work seems
like a pretty questionable idea.
On Thu, Nov 07, 2002 at 04:41:01PM -0600, Jesse Pollard wrote:
> On Thursday 07 November 2002 04:28 pm, Daniel Jacobowitz wrote:
> > On Thu, Nov 07, 2002 at 02:16:15PM -0800, jw schultz wrote:
> > > On Thu, Nov 07, 2002 at 11:05:21AM -0500, Daniel Jacobowitz wrote:
> > > > On Thu, Nov 07, 2002 at 10:57:06AM -0500, Calin A. Culianu wrote:
> > > > > In the /prod/PID subset of procfs, why are the exe, cwd, and root
> > > > > symlinks considered priviledged information?
> > > > >
> > > > > Exe is the big one for me, as this one can be usually infered from
> > > > > reading /prod/PID/maps. Root I guess can't be inferred in any
> > > > > unpriviledged way, and neither can cwd. At any rate.. I am not sure
> > > > > behind the philosophy to make these symlinks' destinations
> > > > > priviledged... can someone clarify this?
> > > >
> > > > This came up a little while ago. The answer is that maps should be
> > > > priviledged also.
> > > >
> > > > For instance:
> > > > You can protect a directory by giving its parent directory no read
> > > > permissions. The name of the directory is now secret. You don't want
> > > > to reveal it in cwd.
> > >
> > > Daniel is correct in that the issue came up recently. He
> > > gives _his_ answer above. If you believe in security
> > > through obscurity you will agree with him. I don't.
> > > I will agree that there should be no real reason to need
> > > access to this information.
> > >
> > > With ACLs you will be able to explicitly grant access and
> > > you won't have to depend on keeping shared info secret.
> > > Then this will be less of an issue.
> >
> > I recommend you go think about what security through obscurity actually
> > _means_. If you think that an unreadable directory and a
> > randomly-generated subdirectory is security through obscurity, then in
> > what way is it actually different from a _password_? That's what it
> > is.
> >
> > Yes, this is poor-man's-ACLs. It works in a lot of places when ACLs
> > won't. For instance, an anonymous FTP server...
>
> It also isn't necessarily just for obscurity.
>
> If the directory contains customer contact lists, where each customer
> is represented by a directory entry, and all communications to/from that
> customer is stored in files. You don't want the cwd to be exposed by
> a process that may set its' workspace to a specific customer.
>
> That is NOT security by obscurity. Exposure is called a "leak".
> And it is entirely possible for sensitive information to be embeded
> in pathnames, process names, and parameters (most often).
I painted with too broad a brush. I apologize.
I'm afraid i'll never consider shared secrets to be secure.
They may provide privacy but not security.
Jesse's example where confidential information is part of
paths not permitted is an extremely good example of why
access to /proc/pid/maps should be restricted and by no
means qualifies for the pejorative i flung so thoughtlessly.
--
________________________________________________________________
J.W. Schultz Pegasystems Technologies
email address: [email protected]
Remember Cernan and Schmitt
On Friday 08 November 2002 01:55, jw schultz wrote:
> I painted with too broad a brush. I apologize.
>
> I'm afraid i'll never consider shared secrets to be secure.
Don't use SSH then, or any SSL derivative. They only use public key
cryptography at the start of the session, primarily to exchange a pair of
symmetric keys, which are then used to encrypt the bulk data transmitted
during the rest of the session.
Public key cryptography is computationally expensive. Symmetric key
cryptography is quick and easy, and provably breakable only by brute force if
you know they haven't got the key.
And of course, the data you're transmitting is itself a shared secret, isn't
it? At the end of the session, anyway. That's the whole point.
> They may provide privacy but not security.
You can have security without privacy?
How?
Privacy is "necessary but not sufficient" to security. All cryptography boils
down to there being something you don't want somebody else to know.
Rob
--
http://penguicon.sf.net - Terry Pratchett, Eric Raymond, Pete Abrams, Illiad,
CmdrTaco, liquid nitrogen ice cream, and caffienated jello. Well why not?