Re: 2.4+ptrace exploit fix breaks root's ability to strace

Re: 2.4+ptrace exploit fix breaks root's ability to strace

[Yusuf Wilajati Purna]

Hi,

On 2003-03-22 17:28:54, Arjan van de Ven wrote:
>On Sat, Mar 22, 2003 at 05:13:12PM +0000, Russell King wrote:
>> 
>> int ptrace_check_attach(struct task_struct *child, int kill)
>> {
>> 	...
>> +       if (!is_dumpable(child))
>> +               return -EPERM;
>> }
>> 
>> So, we went from being able to ptrace daemons as root, to being able to
>> attach daemons and then being unable to do anything with them, even if
>> you're root (or have the CAP_SYS_PTRACE capability).  I think this
>> behaviour is getting on for being described as "insane" 8) and is
>> clearly wrong.
>
>ok it seems this check is too strong. It *has* to check
>child->task_dumpable and return -EPERM, but child->mm->dumpable is not
>needed.

So, do you mean that the following is enough:

int ptrace_check_attach(struct task_struct *child, int kill)
{
      ...
+       if (!child->task_dumpable)
+               return -EPERM;
}

Regards,

Purna
         		

Re: 2.4+ptrace exploit fix breaks root's ability to strace

[Bernhard Kaindl]

On Thu, 17 Apr 2003, Yusuf Wilajati Purna wrote:
> On 2003-03-22 17:28:54, Arjan van de Ven wrote:
> >On Sat, Mar 22, 2003 at 05:13:12PM +0000, Russell King wrote:
> >>
> >> int ptrace_check_attach(struct task_struct *child, int kill)
> >> {
> >> 	...
> >> +       if (!is_dumpable(child))
> >> +               return -EPERM;
> >> }
> >>
> >> So, we went from being able to ptrace daemons as root, to being able to
> >> attach daemons and then being unable to do anything with them, even if
> >> you're root (or have the CAP_SYS_PTRACE capability).  I think this
> >> behaviour is getting on for being described as "insane" 8) and is
> >> clearly wrong.
> >
> >ok it seems this check is too strong. It *has* to check
> >child->task_dumpable and return -EPERM, but child->mm->dumpable is not
> >needed.
>
> So, do you mean that the following is enough:
>
> int ptrace_check_attach(struct task_struct *child, int kill)
> {
>       ...
> +       if (!child->task_dumpable)
> +               return -EPERM;
> }

It's enough to still be safe against the ptrace/kmod exploits.

I could not find a security problem in it yet because
compute_cred() ignores the suid bit on exec when the
process is being traced, so the strace does not get
access to privileges from somebody else and ptrace_attach
uses is_dumpable() which also checks task->mm->dumpable
so a tracer can't attach to a suid program.

It will also help the case Russell King describes above
where root failed to trace a daemon which changed uids
or a suid program, AFAICS.

It is not the complete fix for it because the ptrace functions
also use access_process_vm() where the patch had this hunk:

@@ -123,6 +127,8 @@ int access_process_vm(struct task_struct
        /* Worry about races with exit() */
        task_lock(tsk);
        mm = tsk->mm;
+       if (!is_dumpable(tsk) || (&init_mm == mm))
+               mm = NULL;
        if (mm)
                atomic_inc(&mm->mm_users);
        task_unlock(tsk);

You need to backout the tsk->mm->dumpable check done within is_dumpable
here by just checking task_dumpable and then ptracing from root works
prperly again.

As the kmod ptrace fix relies on task_dumpable for it's protection against
kernel thread trace, and you just remove the tsk->mm->dumpable check by
replacing !is_dumpable(tsk) with !tsk->task_dumpable here also, you don't
affect the kmod ptrace exploit protection in any way while fixing the
ability of root to trace any task.

This also fixes the problem /proc/<pid>/cmdline being empty (also for root)
if <pid> is not dumpable, which is the other bug introduced by this hunk
and broke process managment tools as it was also read on l-k.

Of course now people may say that this opens a security hole because a
normal user ist now able to read /proc/*/cmdline again (also for not
dumpable processes) but I'm answering to this that it has been well
known that it is insecure to put secrets onto the command line space
and e.g. if programs don't clear it properly, they are buggy and should
be audited to fix the bugs and not add a workaound to the kernel for
such programs.

Bernd

Re: 2.4+ptrace exploit fix breaks root's ability to strace

[Yusuf Wilajati Purna]

Hi,

Thanks for the clarification. :-)

Bernhard Kaindl wrote:

>On Thu, 17 Apr 2003, Yusuf Wilajati Purna wrote:
>
>>On 2003-03-22 17:28:54, Arjan van de Ven wrote:
>>
>>>On Sat, Mar 22, 2003 at 05:13:12PM +0000, Russell King wrote:
>>>
>>>>int ptrace_check_attach(struct task_struct *child, int kill)
>>>>{
>>>>	...
>>>>+       if (!is_dumpable(child))
>>>>+               return -EPERM;
>>>>}
>>>>
>>>>So, we went from being able to ptrace daemons as root, to being able to
>>>>attach daemons and then being unable to do anything with them, even if
>>>>you're root (or have the CAP_SYS_PTRACE capability).  I think this
>>>>behaviour is getting on for being described as "insane" 8) and is
>>>>clearly wrong.
>>>>
>>>ok it seems this check is too strong. It *has* to check
>>>child->task_dumpable and return -EPERM, but child->mm->dumpable is not
>>>needed.
>>>
>>So, do you mean that the following is enough:
>>
>>int ptrace_check_attach(struct task_struct *child, int kill)
>>{
>>      ...
>>+       if (!child->task_dumpable)
>>+               return -EPERM;
>>}
>>
>
>It's enough to still be safe against the ptrace/kmod exploits.
>
>I could not find a security problem in it yet because
>compute_cred() ignores the suid bit on exec when the
>process is being traced, so the strace does not get
>access to privileges from somebody else and ptrace_attach
>uses is_dumpable() which also checks task->mm->dumpable
>so a tracer can't attach to a suid program.
>
>It will also help the case Russell King describes above
>where root failed to trace a daemon which changed uids
>or a suid program, AFAICS.
>
>It is not the complete fix for it because the ptrace functions
>also use access_process_vm() where the patch had this hunk:
>
>@@ -123,6 +127,8 @@ int access_process_vm(struct task_struct
>        /* Worry about races with exit() */
>        task_lock(tsk);
>        mm = tsk->mm;
>+       if (!is_dumpable(tsk) || (&init_mm == mm))
>+               mm = NULL;
>        if (mm)
>                atomic_inc(&mm->mm_users);
>        task_unlock(tsk);
>
>You need to backout the tsk->mm->dumpable check done within is_dumpable
>here by just checking task_dumpable and then ptracing from root works
>prperly again.
>
>As the kmod ptrace fix relies on task_dumpable for it's protection against
>kernel thread trace, and you just remove the tsk->mm->dumpable check by
>replacing !is_dumpable(tsk) with !tsk->task_dumpable here also, you don't
>affect the kmod ptrace exploit protection in any way while fixing the
>ability of root to trace any task.
>
>This also fixes the problem /proc/<pid>/cmdline being empty (also for root)
>if <pid> is not dumpable, which is the other bug introduced by this hunk
>and broke process managment tools as it was also read on l-k.
>
Just to recapitulate,
The following changes to the original patch (Alan's patch):

   int ptrace_check_attach(struct task_struct *child, int kill)
   {
           ...
   +      if (!child->task_dumpable)
   +      return -EPERM;
   }

   int access_process_vm(struct task_struct *tsk, unsigned long addr, 
void *buf, int len, int write)
   {
           ...
           /* Worry about races with exit() */
           task_lock(tsk);
           mm = tsk->mm;
   +      if (!tsk->task_dumpable || (&init_mm == mm))
   +                mm = NULL;
           ...
   }

can solve the following side-effects introduced by the original patch:

- /proc/PID/cmdline and /proc/PID/environ are empty for non-dumpable 
process es
  even for root. (ps displays those processes in [] brackets.)
  http://marc.theaimsgroup.com/?l=linux-kernel&m=104807368719299&w=2

- strace started by root cannot ptrace user threads or such non-dumpable 
processes.
  http://marc.theaimsgroup.com/?l=linux-kernel&m=104835339619706&w=2

At least, I have confirmed this on an i386/IA-32 platform. And I have 
checked also
that ptrace/kmod exploits such as isec-ptrace-kmod-exploit.c, ptrace.c, 
km3.c cannot
get root privilege with the changes.

Any comments?

Thanks,
Purna

[PATCH][2.4+ptrace] fix side effects of the kmod/ptrace secfix

[Bernhard Kaindl]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 6604 bytes --]

Hello Marcelo, Hello Yusuf!

  I've attached a patch which fixes the remaining side effects
which the ptrace fix posted by Alan introduced(which affect production
systems) and I'm sending this because I think 2.4.20-rc1 should
not be released as 2.4.21 without these problems fixed.

On Tue, 22 Apr 2003, Yusuf Wilajati Purna wrote:
>
> Thanks for the clarification. :-)

Sorry if my descriptions in my previos mail did not have any word too
much(really short) but I tried to make the point straight for people
which know the code. I'm adding a little bit more verbosity now :-)

The check added by Alan's patch to ptrace_check_attach was:

+       if (!is_dumpable(child))
+               return -EPERM;

New, replacement check in ptrace_check_attach:

+       if (!child->task_dumpable)
+               return -EPERM;

I want to explain now, why the above use of is_dumpable() broke ptrace
of setuid programs by root:

is_dumpable() checks if both, task_dumpable and mm->dumpable are set, and
evaluates to false, if one of them is false.

The new kernel_thread() function added by Alan's patch sets task_dumpable
(which is 1 by default) for the new kernel thread to 0, and this is the
only place where his new variable is set to 0, so "non_kernel_thread"
would accurately describe what it is saying.

By adding is_dumpable(child) to ptrace_check_attach(), the patch posted
by Alan, not only a check if the task is a kernel thread, but also a
check if the task changed it's uid's was added(what mm->dumpable says)
so even root was blocked out by this check.

So, removing the wrong check to child->mm->dumpable and only checking
child->task_dumpable (wnich really means "non_kernel_thread") is the
first part of the fix.

The other place which needed to be touched to fix Alan's patch was
access_process_vm(), where Alan's patch did this change:

> >@@ -123,6 +127,8 @@ int access_process_vm(struct task_struct
> >        /* Worry about races with exit() */
> >        task_lock(tsk);
> >        mm = tsk->mm;
> >+       if (!is_dumpable(tsk) || (&init_mm == mm))
> >+               mm = NULL;
> >        if (mm)
> >                atomic_inc(&mm->mm_users);
> >        task_unlock(tsk);

access_process_vm() is in the same code patch as ptrace_check_attach.

If you read the sys_ptrace implementation in arch/i386/kernel/ptrace.c,
you'll find a call to ptrace_check_attach() and then, shortly afterwards,
depending on what ptrace action was requested, a call to access_process_vm()

So the !is_dumpable(tsk) check above it just a repetition if the previous
check which you can also replace with !tsk->task_dumpable which you correctly
understood and you show below in your change:

> Just to recapitulate,
> The following changes to the original patch (Alan's patch):
>
>    int ptrace_check_attach(struct task_struct *child, int kill)
>    {
>            ...
>    +      if (!child->task_dumpable)
>    +      return -EPERM;
>    }
>
>    int access_process_vm(struct task_struct *tsk, unsigned long addr,
> void *buf, int len, int write)
>    {
>            ...
>            /* Worry about races with exit() */
>            task_lock(tsk);
>            mm = tsk->mm;
>    +      if (!tsk->task_dumpable || (&init_mm == mm))
>    +                mm = NULL;

Note, in addtion to breaking root's ability to trace setuid programs,
having the tsk->mm->dumpable checked by !is_dumpable(tsk) at this place
also broke /proc/PID/cmdline and /proc/PID/environ because access_process_vm()
is also used by these proc functions.

If somebody says this opens a securtiy leak, I'd have to say:

 If a suid task leaks such information thru it's cmdline buffer, it's
 the problem of the suid process not acting secure and should be reviewed.

 You would need to restrict cmdline access to all root processes(not only
 suid) and maybe even to all processes with different capabilites and uid/gid
 to work around problems in such processes. But you would break even more
 system monitoring stuff this way(I've even heard shutdown is affected)

>            ...
>    }
>
> can solve the following side-effects introduced by the original patch:
>
> - /proc/PID/cmdline and /proc/PID/environ are empty for non-dumpable
> process es
>   even for root. (ps displays those processes in [] brackets.)
>   http://marc.theaimsgroup.com/?l=linux-kernel&m=104807368719299&w=2
>
> - strace started by root cannot ptrace user threads or such non-dumpable
> processes.
>   http://marc.theaimsgroup.com/?l=linux-kernel&m=104835339619706&w=2

Yes exacly, they do fix these side-effects, as your test correctly gives:

> At least, I have confirmed this on an i386/IA-32 platform. And I have
> checked also
> that ptrace/kmod exploits such as isec-ptrace-kmod-exploit.c, ptrace.c,
> km3.c cannot
> get root privilege with the changes.

Exactly, the change (effectve removal of the task->mm->dumpable flag check,
which is not part of the kernel_thread/ptrace issue, fix the the two side
effects you describe above while maintaining same security against the
kmod/ptrace exploits because it only removes code that has nothing to
do with the kernel_thread/ptrace issue, which was added by Alan's patch
and introduced these side effects.

It's really that easy, you just have to look and the code and see it ;-)

Ok, you need to understand how the ptrace code works and how Alan's
patch effectively blocks all possible trace attempts and backdoors,
but once you understood how it works, it's easy to identify the parts
of Alan's patch which cause these side effects.

It's just cleanup, nothing more, nothing very creative, but a nice
opportunity to learn a little bit about the kernel, no deep knowledge
about VM or something really complex is needed.

<tiny font>
If you look sharper, you can even start cleaning up more of code added
by the patch Alan sent (the above checks are completely unneccesary ;-)
but you need the big picture for this and I have to give you this big
picture in a separate mail to make this point.
</tiny font>

> Any comments?

I'm sorry that I did not send a patch the first time to make the
change 100% clear to anybody, but I'm doing this now.

Incremental patch which applies on top of the patch posted by
Alan and also on top of 2.4.21-rc1 is attached now.

With only this patch applied I'd think 2.4.21 could be released,
but not without this minimum fix.

2.4.21-rc1, if it would be released as is, has the potential of
breaking lots of systems which rely on not seeing the side effects
Yusuf Wilajati Purna describes above and are fixed by this
incremental fix.

Best Regards,
Bernhard Kaindl
SuSE Linux

[-- Attachment #2: Type: TEXT/PLAIN, Size: 546 bytes --]

--- kernel/ptrace.c	2003/04/22 21:14:20	1.1
+++ kernel/ptrace.c	2003/04/22 21:15:40
@@ -22,7 +22,7 @@
 int ptrace_check_attach(struct task_struct *child, int kill)
 {
 	mb();
-	if (!is_dumpable(child))
+	if (!child->task_dumpable)
 		return -EPERM;
 
 	if (!(child->ptrace & PT_PTRACED))
@@ -127,7 +127,7 @@
 	/* Worry about races with exit() */
 	task_lock(tsk);
 	mm = tsk->mm;
-	if (!is_dumpable(tsk) || (&init_mm == mm))
+	if (!tsk->task_dumpable || (&init_mm == mm))
 		mm = NULL;
 	if (mm)
 		atomic_inc(&mm->mm_users);

Re: [PATCH][2.4+ptrace] fix side effects of the kmod/ptrace secfix

[Nuno Silva]

Good morning! :)

I'd like to ear an "official" word on this subject, please. :)
Is this patch still secure?

I don't like having a known root hole in my servers but i'd like to have 
the same functionallity as before. Right now it seams that I can't have 
both :(

Alan? Marcelo? Linus? HELPPPPPPPPP! :)

Regards,
Nuno Silva

Bernhard Kaindl wrote:
> Hello Marcelo, Hello Yusuf!
> 
>   I've attached a patch which fixes the remaining side effects
> which the ptrace fix posted by Alan introduced(which affect production
> systems) and I'm sending this because I think 2.4.20-rc1 should
> not be released as 2.4.21 without these problems fixed.
> 
> On Tue, 22 Apr 2003, Yusuf Wilajati Purna wrote:
> 
>>Thanks for the clarification. :-)
> 
> 
> Sorry if my descriptions in my previos mail did not have any word too
> much(really short) but I tried to make the point straight for people
> which know the code. I'm adding a little bit more verbosity now :-)
> 
> The check added by Alan's patch to ptrace_check_attach was:
> 
> +       if (!is_dumpable(child))
> +               return -EPERM;
> 
> New, replacement check in ptrace_check_attach:
> 
> +       if (!child->task_dumpable)
> +               return -EPERM;
> 
> I want to explain now, why the above use of is_dumpable() broke ptrace
> of setuid programs by root:
> 
> is_dumpable() checks if both, task_dumpable and mm->dumpable are set, and
> evaluates to false, if one of them is false.
> 
> The new kernel_thread() function added by Alan's patch sets task_dumpable
> (which is 1 by default) for the new kernel thread to 0, and this is the
> only place where his new variable is set to 0, so "non_kernel_thread"
> would accurately describe what it is saying.
> 
> By adding is_dumpable(child) to ptrace_check_attach(), the patch posted
> by Alan, not only a check if the task is a kernel thread, but also a
> check if the task changed it's uid's was added(what mm->dumpable says)
> so even root was blocked out by this check.
> 
> So, removing the wrong check to child->mm->dumpable and only checking
> child->task_dumpable (wnich really means "non_kernel_thread") is the
> first part of the fix.
> 
> The other place which needed to be touched to fix Alan's patch was
> access_process_vm(), where Alan's patch did this change:
> 
> 
>>>@@ -123,6 +127,8 @@ int access_process_vm(struct task_struct
>>>       /* Worry about races with exit() */
>>>       task_lock(tsk);
>>>       mm = tsk->mm;
>>>+       if (!is_dumpable(tsk) || (&init_mm == mm))
>>>+               mm = NULL;
>>>       if (mm)
>>>               atomic_inc(&mm->mm_users);
>>>       task_unlock(tsk);
> 
> 
> access_process_vm() is in the same code patch as ptrace_check_attach.
> 
> If you read the sys_ptrace implementation in arch/i386/kernel/ptrace.c,
> you'll find a call to ptrace_check_attach() and then, shortly afterwards,
> depending on what ptrace action was requested, a call to access_process_vm()
> 
> So the !is_dumpable(tsk) check above it just a repetition if the previous
> check which you can also replace with !tsk->task_dumpable which you correctly
> understood and you show below in your change:
> 
> 
>>Just to recapitulate,
>>The following changes to the original patch (Alan's patch):
>>
>>   int ptrace_check_attach(struct task_struct *child, int kill)
>>   {
>>           ...
>>   +      if (!child->task_dumpable)
>>   +      return -EPERM;
>>   }
>>
>>   int access_process_vm(struct task_struct *tsk, unsigned long addr,
>>void *buf, int len, int write)
>>   {
>>           ...
>>           /* Worry about races with exit() */
>>           task_lock(tsk);
>>           mm = tsk->mm;
>>   +      if (!tsk->task_dumpable || (&init_mm == mm))
>>   +                mm = NULL;
> 
> 
> Note, in addtion to breaking root's ability to trace setuid programs,
> having the tsk->mm->dumpable checked by !is_dumpable(tsk) at this place
> also broke /proc/PID/cmdline and /proc/PID/environ because access_process_vm()
> is also used by these proc functions.
> 
> If somebody says this opens a securtiy leak, I'd have to say:
> 
>  If a suid task leaks such information thru it's cmdline buffer, it's
>  the problem of the suid process not acting secure and should be reviewed.
> 
>  You would need to restrict cmdline access to all root processes(not only
>  suid) and maybe even to all processes with different capabilites and uid/gid
>  to work around problems in such processes. But you would break even more
>  system monitoring stuff this way(I've even heard shutdown is affected)
> 
> 
>>           ...
>>   }
>>
>>can solve the following side-effects introduced by the original patch:
>>
>>- /proc/PID/cmdline and /proc/PID/environ are empty for non-dumpable
>>process es
>>  even for root. (ps displays those processes in [] brackets.)
>>  http://marc.theaimsgroup.com/?l=linux-kernel&m=104807368719299&w=2
>>
>>- strace started by root cannot ptrace user threads or such non-dumpable
>>processes.
>>  http://marc.theaimsgroup.com/?l=linux-kernel&m=104835339619706&w=2
> 
> 
> Yes exacly, they do fix these side-effects, as your test correctly gives:
> 
> 
>>At least, I have confirmed this on an i386/IA-32 platform. And I have
>>checked also
>>that ptrace/kmod exploits such as isec-ptrace-kmod-exploit.c, ptrace.c,
>>km3.c cannot
>>get root privilege with the changes.
> 
> 
> Exactly, the change (effectve removal of the task->mm->dumpable flag check,
> which is not part of the kernel_thread/ptrace issue, fix the the two side
> effects you describe above while maintaining same security against the
> kmod/ptrace exploits because it only removes code that has nothing to
> do with the kernel_thread/ptrace issue, which was added by Alan's patch
> and introduced these side effects.
> 
> It's really that easy, you just have to look and the code and see it ;-)
> 
> Ok, you need to understand how the ptrace code works and how Alan's
> patch effectively blocks all possible trace attempts and backdoors,
> but once you understood how it works, it's easy to identify the parts
> of Alan's patch which cause these side effects.
> 
> It's just cleanup, nothing more, nothing very creative, but a nice
> opportunity to learn a little bit about the kernel, no deep knowledge
> about VM or something really complex is needed.
> 
> <tiny font>
> If you look sharper, you can even start cleaning up more of code added
> by the patch Alan sent (the above checks are completely unneccesary ;-)
> but you need the big picture for this and I have to give you this big
> picture in a separate mail to make this point.
> </tiny font>
> 
>>Any comments?
> 
> 
> I'm sorry that I did not send a patch the first time to make the
> change 100% clear to anybody, but I'm doing this now.
> 
> Incremental patch which applies on top of the patch posted by
> Alan and also on top of 2.4.21-rc1 is attached now.
> 
> With only this patch applied I'd think 2.4.21 could be released,
> but not without this minimum fix.
> 
> 2.4.21-rc1, if it would be released as is, has the potential of
> breaking lots of systems which rely on not seeing the side effects
> Yusuf Wilajati Purna describes above and are fixed by this
> incremental fix.
> 
> Best Regards,
> Bernhard Kaindl
> SuSE Linux
> 
> 
> ------------------------------------------------------------------------
> 
> --- kernel/ptrace.c	2003/04/22 21:14:20	1.1
> +++ kernel/ptrace.c	2003/04/22 21:15:40
> @@ -22,7 +22,7 @@
>  int ptrace_check_attach(struct task_struct *child, int kill)
>  {
>  	mb();
> -	if (!is_dumpable(child))
> +	if (!child->task_dumpable)
>  		return -EPERM;
>  
>  	if (!(child->ptrace & PT_PTRACED))
> @@ -127,7 +127,7 @@
>  	/* Worry about races with exit() */
>  	task_lock(tsk);
>  	mm = tsk->mm;
> -	if (!is_dumpable(tsk) || (&init_mm == mm))
> +	if (!tsk->task_dumpable || (&init_mm == mm))
>  		mm = NULL;
>  	if (mm)
>  		atomic_inc(&mm->mm_users);

Re: [PATCH][2.4+ptrace] fix side effects of the kmod/ptrace secfix

[Arjan van de Ven]

On Thu, Apr 24, 2003 at 06:40:55AM +0100, Nuno Silva wrote:
> Good morning! :)
> 
> I'd like to ear an "official" word on this subject, please. :)
> Is this patch still secure?

The check is loosend too much.

Re: [PATCH][2.4+ptrace] fix side effects of the kmod/ptrace secfix

[Bernhard Kaindl]

Hi Arjan!

On Thu, 24 Apr 2003, Arjan van de Ven wrote:
> On Thu, Apr 24, 2003 at 06:40:55AM +0100, Nuno Silva wrote:
> > Good morning! :)
> >
> > I'd like to ear an "official" word on this subject, please. :)
> > Is this patch still secure?
>
> The check is loosend too much.

Last month, you sounded different:

On 2003-03-22 17:28:54, Arjan van de Ven wrote:
>On Sat, Mar 22, 2003 at 05:13:12PM +0000, Russell King wrote:
>>
>> int ptrace_check_attach(struct task_struct *child, int kill)
>> {
>>    ...
>> +       if (!is_dumpable(child))
>> +               return -EPERM;
>> }
>>
>> So, we went from being able to ptrace daemons as root, to being able to
>> attach daemons and then being unable to do anything with them, even if
>> you're root (or have the CAP_SYS_PTRACE capability).  I think this
>> behaviour is getting on for being described as "insane" 8) and is
>> clearly wrong.
>
>ok it seems this check is too strong. It *has* to check
>child->task_dumpable and return -EPERM, but child->mm->dumpable is not
>needed.

Can you give me an explanation that changed with regard to the
kmod/ptrace fix?

Im private discussuion this possible scenario has been brought up:

> execed setuid
> opens RAW_SOCKET
> setuid back
>
> 	ptrace_attach()
> ...

AFAICS, ptrace_attach() will abort, because the "setuid back" does
not set mm->dumpable back to 1, it is left at 0 and ptrace_attach()
aborts then.

Of course you could do an exec() to get a new mm and then, you may
get an mm->dumpable is set != 0.

But especially in the case of an exec() I think the setuid program
should not go back to the original uid because the user may send
any signal it wants to the suid program then, read it's environment
and do all other sorts of bad stuff e.g. if the program is stupid
enough to not use a safe directory for temp files and such.

If we really need to protect such (IMHO insecure) progam to be
somewhat more secure, we could inherit task_dumpable over exec's,
but I'm not sure what kind of side effects whis would have when
using programs like su and login...

I think by adding such workarounds which are unrelated to kmod,
we would introduce more unwanted side effects insted of fixing
what we alread have because of too strong checks.

Best Regards,
Bernhard Kaindl

Re: [PATCH][2.4+ptrace] fix side effects of the kmod/ptrace secfix

[Andreas Gietl]

>  You would need to restrict cmdline access to all root processes(not only
>  suid) and maybe even to all processes with different capabilites and > 
uid/gid
>  to work around problems in such processes. But you would break even more
> system monitoring stuff this way(I've even heard shutdown is affected)

i can confirm that shutdown (halt|reboot) does not work on my 2.4.21-rc1-ac1 
boxes. (gentoo + redhat).

But your patch does not seem to fix it.

-- 
e-admin internet gmbh
Andreas Gietl                                            tel +49 941 3810884
Ludwig-Thoma-Strasse 35                      fax +49 89 244329104
93051 Regensburg                                  mobil +49 171 6070008

PGP/GPG-Key unter http://www.e-admin.de/gpg.html