Skip to contentSkip to conversation

request: capabilities that allow users to drop privileges further

DISCUSSION Felix von Leitner 6 participants 10 messages Dec 15, 2003 – Dec 17, 2003

lore
Share & tools
lore
#0 lore raw

I would like to be able to drop capabilities that every normal user has, so that network servers can limit the impact of possible future security problems further. For example, I want my non-cgi web server to be able to drop the capabilities to

* fork
* execve
* ptrace
* load kernel modules
* mknod
* write to the file system

and I would like to modify my smtpd to not be able to

* fork
* execve
* ptrace
* load kernel modules
* mknod

I can kludge around some of these, for example I can disable fork with resource limits, and I can limit writing to the file system with chroot and proper permissions in the file systems, but I'm not aware of a way to disable ptrace for example, or pthread_create.

I know that there are patches to provide an extended "jail" chroot support, but being able to drop capabilities like this would be a more general solution.

Felix

#1 lore raw
Quoted 25 lines · Felix von Leitner

I would like to be able to drop capabilities that every normal user has, so that network servers can limit the impact of possible future security problems further. For example, I want my non-cgi web server to be able to drop the capabilities to

* fork
* execve
* ptrace
* load kernel modules
* mknod
* write to the file system

and I would like to modify my smtpd to not be able to

* fork
* execve
* ptrace
* load kernel modules
* mknod

I can kludge around some of these, for example I can disable fork with resource limits, and I can limit writing to the file system with chroot and proper permissions in the file systems, but I'm not aware of a way to disable ptrace for example, or pthread_create.

Quoted 5 lines

I know that there are patches to provide an extended "jail" chroot support, but being able to drop capabilities like this would be a more general solution.

Felix

So you expect kernel support? Normally, real people write or modify applications to provide for specific exceptions to the standards. They don't expect an operating system to modify itself to unique situations. That's not what operating systems have generally done in the past.

The 'C' runtime library interfaces to the kernel. You can use the ld.so.preload capabilities to substitute private functions for fork(), etc. This has the additional benefit of allowing crappy, poorly-written, executables that may have buffer overflows to be executed with increased confidence. Of course, some root-shell programs bypass the 'C' runtime libraries.

Cheers,
Dick Johnson
Penguin : Linux version 2.4.22 on an i686 machine (797.90 BogoMips).
Note 96.31% of all statistics are fiction.

#2 lore raw
Quoted 2 lines · Richard B. Johnson
Quoted 1 line · Felix von Leitner

I would like to be able to drop capabilities that every normal user

[...]

Quoted 1 line
Quoted 1 line

security problems further. For example, I want my non-cgi web server

[...]

Quoted 3 lines
Quoted 3 lines

* fork
* execve
* ptrace

[...]

Quoted 5 lines

So you expect kernel support? Normally, real people write or modify applications to provide for specific exceptions to the standards. They don't expect an operating system to modify itself to unique situations. That's not what operating systems have generally done in the past.

[...]

I dont agree. Policy is userspace but enforcing the policy very often needs kernel support.

Having ACL in 2.6 is an example where operating system already adopted to special needs. Furthermore, the kernel is already able to drop special capabilites, like module loading. Having a generalised capabilites model is a good idea and there are already some more or less usable security modules.

cheers

Christian

#3 lore raw
In reply to Richard B. Johnson

hi :)

Quoted 1 line · Richard B. Johnson

Of course, some root-shell programs bypass the 'C' runtime libraries.

of course, they have to as shellcode won't include a dynamic linker. ;)

so your approach does not help from a security point of view, and felix only was concerned about security.

Signature · 5 lines
CU,		  / Friedrich-Alexander University Erlangen, Germany
Martin Waitz	//  Department of Computer Science 3       _________
______________/// - - - - - - - - - - - - - - - - - - - - ///
dies ist eine manuell generierte mail, sie beinhaltet    //
tippfehler und ist auch ohne grossbuchstaben gueltig.   /
#4 lore raw
In reply to Felix von Leitner
Quoted 11 lines · Felix von Leitner

I would like to be able to drop capabilities that every normal user has, so that network servers can limit the impact of possible future security problems further. For example, I want my non-cgi web server to be able to drop the capabilities to

* fork
* execve
* ptrace
* load kernel modules
* mknod
* write to the file system

You can have a look at
http://lsm.immunix.org/ and
http://lsm.immunix.org/lsm_modules.html
if there is something that fits your need.
If not, feel free to write a security module, that is able to do just what you want. ;-)

cheers

Christian

#5 lore raw
In reply to Felix von Leitner
Quoted 4 lines · * Felix von Leitner ([email protected]) wrote:

I would like to be able to drop capabilities that every normal user has, so that network servers can limit the impact of possible future security problems further. For example, I want my non-cgi web server to be able to drop the capabilities to

Using existing capabilities system you can limit many of these. Just dropping privs from uid = 0 to anything else is a good start.

Quoted 1 line

* fork

rlimit

Quoted 1 line

* execve

mount fs noexec

Quoted 1 line

* ptrace

drop CAP_SYS_PTRACE

Quoted 1 line

* load kernel modules

drop CAP_SYS_MODULE

Quoted 1 line

* mknod

drop CAP_MKNOD

Quoted 1 line

* write to the file system

mount fs r/o.

In general, most of what you ask for is already there. Otherwise use some MAC policy that gives you the control you want (check out SELinux for example).

thanks,
-chris

Signature · 1 line
Linux Security Modules     http://lsm.immunix.org     http://lsm.bkbits.net
#6 lore raw

hi :)

Quoted 3 lines · Chris Wright
Quoted 1 line

* ptrace

drop CAP_SYS_PTRACE

that will only help agains ptracing foreign processes. you can still debug your own ones.

so this does not help agains buffer overflows&co in ptrace

Signature · 5 lines
CU,		  / Friedrich-Alexander University Erlangen, Germany
Martin Waitz	//  Department of Computer Science 3       _________
______________/// - - - - - - - - - - - - - - - - - - - - ///
dies ist eine manuell generierte mail, sie beinhaltet    //
tippfehler und ist auch ohne grossbuchstaben gueltig.   /
#7 lore raw
In reply to Chris Wright

Thus spake Chris Wright ([email protected]):

Quoted 5 lines
Quoted 4 lines

I would like to be able to drop capabilities that every normal user has, so that network servers can limit the impact of possible future security problems further. For example, I want my non-cgi web server to be able to drop the capabilities to

Using existing capabilities system you can limit many of these.

No. The administrator can limit many of these. I want the software to be able to limit itself.

Quoted 1 line

Just dropping privs from uid = 0 to anything else is a good start.

If you give administrators tools to limit privileges, some well-educated admins may do it, but the bulk will not.

If you give software authors the tools to drop privileges they don't need, and you help everyone.

Quoted 2 lines
Quoted 1 line

* execve

mount fs noexec

Not an option for a web server or smtpd.

Quoted 2 lines
Quoted 1 line

* ptrace

drop CAP_SYS_PTRACE

That will still allow the process to ptrace other processes of the same uid.

Quoted 2 lines
Quoted 1 line

* write to the file system

mount fs r/o.

Again, not an option.

I hope I made myself more clear this time.

I would be happy with a well documented and fine grained capabilities system, and maybe a new syscall:

  abstain(__NR_execve);

I would also like a way to say that I want to never access the network except through sockets that are already open (and presumably passed to me at process creation). Dan Bernstein has a good rationale on what I think is needed: http://cr.yp.to/unix/disablenetwork.html, however I think if we do this, we might as well opt for more flexibility.

Imagine being able to call "gzip -dc" in a pipe and denying it write access to the file system, network I/O and other harmful operations. If programs can restrict themselves, we could write email client software that uses external untrusted plugins without fear of buffer overflows or catching yourself a root kit.

Felix

#8 lore raw
Quoted 5 lines · * Felix von Leitner ([email protected]) wrote:

Imagine being able to call "gzip -dc" in a pipe and denying it write access to the file system, network I/O and other harmful operations. If programs can restrict themselves, we could write email client software that uses external untrusted plugins without fear of buffer overflows or catching yourself a root kit.

Write some SELinux policies for the email and web server that do what you'd like. The LSM infrastructure allows you to control all these things, and SELinux gives a configuration language to do this with. Or you can write a simple module to do just what you'd like.

thanks,
-chris

Signature · 1 line
Linux Security Modules     http://lsm.immunix.org     http://lsm.bkbits.net
#9 lore raw
In reply to Felix von Leitner
Quoted 12 lines · Felix von Leitner

I would like to be able to drop capabilities that every normal user has, so that network servers can limit the impact of possible future security problems further. For example, I want my non-cgi web server to be able to drop the capabilities to

* fork
* execve
* ptrace
* load kernel modules
* mknod
* write to the file system

Quoted 7 lines

and I would like to modify my smtpd to not be able to

* fork
* execve
* ptrace
* load kernel modules
* mknod

You can specify policy under SELinux to acheive this (without modifying any applications).

- James

Signature · 2 lines
James Morris
<[email protected]>