Skip to contentSkip to conversation

How to make stack executable on demand?

DISCUSSION H. J. Lu 6 participants 9 messages Apr 16, 2004 – Apr 22, 2004

lore
Share & tools
lore

H. J. Lu

author
#0 lore raw

With the non-executable stack kernel, how can I make stack executable on demand? If I set kernel with non-executable stack, only those binaries marked with executable PT_GNU_STACK will have executable stack. But a binary with non-executable PT_GNU_STACK may dlopen a DSO with executable PT_GNU_STACK. The dynamic linker will try to make stack executable with mprotect. But it will either fail if kernel is set with non-executable stack, or be a no-op if kernel is set with executable stack. Is there a third option that a process starts with non-executable stack and can change the stack permission later?

H.J.

#1 lore raw
Quoted 4 lines · H. J. Lu

is set with executable stack. Is there a third option that a process starts with non-executable stack and can change the stack permission later?

Well, in my opinion your request is equivalent to "keep all these cute buffer overflows forever". Take any protected app, LD_PRELOAD or drop in a bad/malicious library and your're done for good. Not really a good idea.

Signature · 1 line
Andreas Steinmetz                       SPAMmers use [email protected]
author
#2 lore raw
Quoted 9 lines · Andreas Steinmetz
Quoted 4 lines · H. J. Lu

is set with executable stack. Is there a third option that a process starts with non-executable stack and can change the stack permission later?

Well, in my opinion your request is equivalent to "keep all these cute buffer overflows forever". Take any protected app, LD_PRELOAD or drop in a bad/malicious library and your're done for good. Not really a good idea.

The current scheme doesn't work too well. Linker doesn't combine PT_GNU_STACK from DSO:

http://sources.redhat.com/ml/binutils/2004-04/msg00341.html

for a reason. It expects the dynamic linker to do that at the run-time, which kernel won't allow. I am looking for a reasonable solution.

H.J.

#4 lore raw
In reply to H. J. Lu
Quoted 2 lines

But it will either fail if
kernel is set with non-executable stack,

eh no. mprotect with prot_exec is still supposed to work. The stacks still have MAY_EXEC attribute, just not the actual EXEC attribute

author
#5 lore raw
Quoted 5 lines · Arjan van de Ven
Quoted 2 lines

But it will either fail if
kernel is set with non-executable stack,

eh no. mprotect with prot_exec is still supposed to work. The stacks still have MAY_EXEC attribute, just not the actual EXEC attribute

Ok. It looks like a bug in Red Hat EL 3 kernel. In fs/exec.c, there are

                if (executable_stack)
                        mpnt->vm_flags = VM_STACK_FLAGS | VM_MAYEXEC | VM_EXEC;
		else
                        mpnt->vm_flags = VM_STACK_FLAGS & ~(VM_MAYEXEC|VM_EXEC); 

That is if an executabl is not marked with executable stack, the VM_MAYEXEC bit is turned off. But 2.6.5-mm6 has

                if (unlikely(executable_stack == EXSTACK_ENABLE_X))
                        mpnt->vm_flags = VM_STACK_FLAGS |  VM_EXEC;
                else if (executable_stack == EXSTACK_DISABLE_X)
                        mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
                else
                        mpnt->vm_flags = VM_STACK_FLAGS;

The VM_MAYEXEC bit is untouched. Now the question is if it is a good idea for user to change stack permission.

H.J.

#6 lore raw
Quoted 8 lines · H. J. Lu
Quoted 5 lines · Arjan van de Ven
Quoted 2 lines

But it will either fail if
kernel is set with non-executable stack,

eh no. mprotect with prot_exec is still supposed to work. The stacks still have MAY_EXEC attribute, just not the actual EXEC attribute

Ok. It looks like a bug in Red Hat EL 3 kernel. In fs/exec.c, there

That version of exec-shield is quite dated. For the latest version, look on http://people.redhat.com/mingo, or the Fedora kernels.

I'm pretty sure that Exec-shield isn't even enabled in the EL kernels at this time, so it's quite out of date in respect to the others.

Dave

#7 lore raw
In reply to H. J. Lu
Quoted 7 lines · H. J. Lu
Quoted 5 lines · Arjan van de Ven
Quoted 2 lines

But it will either fail if
kernel is set with non-executable stack,

eh no. mprotect with prot_exec is still supposed to work. The stacks still have MAY_EXEC attribute, just not the actual EXEC attribute

Quoted 7 lines

Ok. It looks like a bug in Red Hat EL 3 kernel. In fs/exec.c, there are

                if (executable_stack)
                        mpnt->vm_flags = VM_STACK_FLAGS | VM_MAYEXEC | VM_EXEC;
		else
                        mpnt->vm_flags = VM_STACK_FLAGS & ~(VM_MAYEXEC|VM_EXEC); 

yep that's a bug

Quoted 2 lines

The VM_MAYEXEC bit is untouched. Now the question is if it is a good idea for user to change stack permission.

it's required for correct operation and "security wise" it doesn't matter, if someone can execute an mprotect syscall the game is over anyway

#8 lore raw
In reply to H. J. Lu
Quoted 10 lines · H. J. Lu
Quoted 5 lines · Arjan van de Ven
Quoted 2 lines

But it will either fail if
kernel is set with non-executable stack,

eh no. mprotect with prot_exec is still supposed to work. The stacks still have MAY_EXEC attribute, just not the actual EXEC attribute

[...]
The VM_MAYEXEC bit is untouched. Now the question is if it is a good idea for user to change stack permission.

You can create a new executable data area with mmap(), copy the stack to it, unmap the stack, and mremap() to move the copy to where the stack was. The run time linker can do this if you're on a kernel where mprotect() fails.

In other words, even those kernels which disable VM_MAYEXEC don't protect against this alternative way of simulating mprotect(). There is no point in them prohibiting it.

-- Jamie

#3 lore raw
In reply to Andreas Steinmetz

Hi!

Quoted 9 lines
Quoted 4 lines

is set with executable stack. Is there a third option that a process starts with non-executable stack and can change the stack permission later?

Well, in my opinion your request is equivalent to "keep all these cute buffer overflows forever". Take any protected app, LD_PRELOAD or drop in a bad/malicious library and your're done for good. Not really a good idea.

With malicious libraries, you have *way* bigger problems than buffer overruns.
Pavel

Signature · 1 line
64 bytes from 195.113.31.123: icmp_seq=28 ttl=51 time=448769.1 ms