Re^2: how would you detect a math expression

[Molt]

Sounds more like a dangerous idea.. try Googling for system('rm -rf /'), see if it takes their entire system down.

I know that's a very extreme example, and would require a very badly set up system, but bear in mind you've just allowed your random passer-by to do anything the user the webapp's running as could do.

Running eval() on user input is a very dangerous hobby, much better to use Parse::RecDescent to parse it into a sane, and safe, mini-languase.

[shmem]

Sounds more like a dangerous idea..

Uh-oh. Of course you're right; I forgot to mention that before that, a double fork and subsequent chroot /dev/null is required :-)

Good point, Molt++

[Molt]

Even with the double fork and chroot there're still a lot of potential security holes which are open, for example consider the following two fairly common attacks for a couple off the top of my head:
• Denial of service attacks: Make the machine talk to itself a lot.. use LWP::Simple; while (1) { get("http://localhost/something_intensive.cgi")} (bonus points for calling the the calculator again with the same effect), fork bombing with while (1) { fork }, or getting a list of the current user's active processes and kill -9'ing them.
• Mailing back details from your machine to the aggressor to get information for other attacks: Grab the script, grab the config files listed in the script, connect to the database with credentials from config file, delete all data or subtly corrupt it as you see fit. If this application has no config file there're probably other web apps on the machine who's config file you can get.
Basically this is the Whack-A-Mole kind of thing where you think of two avenues of attack and block them, the attacker thinks of a third and causes you pain. The kind of security provided by the Safe module could be a step in the right direction but that'd need careful config, whereas the route means that the operations the user can execute are strictly restricted to a known-safe subset, and also means that the user can enter terms such as '2x2.23kg in lbs' which'd never run in Perl even with constants carefully defined.

[shmem]

Even with the double fork and chroot there're still a lot of potential security holes which are open, for example consider the following two fairly common attacks for a couple off the top of my head:

of course taint mode is required, as well as sanitizing the input, limiting it to a safe subset of expressions. And it seems that I am required to advise the usage of strict and warnings also... but that's not the point of the thread, is it?