On Aug 21, 2018, at 4:34 PM, Nataraj <incoming-centos at rjl.com> wrote: > > On 08/21/2018 02:20 PM, Warren Young wrote: >> On Aug 21, 2018, at 1:27 PM, Nataraj <incoming-centos at rjl.com> wrote: >>> I have a web application which uses sudo to invoke python scripts as the >>> user under which the application runs (NO root access). >> Why is the web app not running with that user’s permissions in the first place? >> > The php code runs as user apache under the webserver. Okay, that’s useful to know, and it’s something we’re just now learning. You’ll get better advice if you include such details when using for help. > If the php ran as the app users it would have full access to all of the data in the app. …and that’s a problem why, exactly? What could happen if that were allowed? I understand that you’re creating a privilege separation scheme here, but if you want good advice, we need to know what you want to achieve with the scheme and why that is necessary. What resources does this non-php user own that user php must not be allowed to have access to? Once we know that, we can advise on how to protect those resources. > Using sudo the app can only invoke one specific python script (which is the command name in the sudoers file) to do what it needs to do, without having access to the rest of the apps data and other python scripts used by other functions in the app). Another way to go about it would be to have the background service running as the non-php user, then provide access to it over the many IPC mechanisms available in a Linux system: named pipes, SysV message queues, shared memory… Add to that all of the higher-level services available like message-oriented middleware: https://en.wikipedia.org/wiki/Category:Message-oriented_middleware Such services let one process tell another, “Hey, I need you to do something for me,” then wait for the answer, received as a single coherent message. Many of these schemes let you split that worker process off into a separate machine, or even a cluster of machines. That could help you to get off CentOS 5: move the worker process onto a C7 box, then when that’s well-validated, move the PHP bits over. Or, move the PHP bits to a *second* C7 box, and now you’ve got much stronger privilege separation. You may remember all of the ways that Shellshock — a local-only exploit — was able to be exploited over HTTP, because local web app code was using the shell, thus converting it into a remote-exploitable security hole. By separating the worker processes to a separate machine, that now can’t happen on the second box. If the front-end box has no sensitive material on it, that’s enough security: just wipe it and re-image it if it’s ever compromised. However you do this, these mechanisms give you hard privilege separation without SELinux bugging you. > I could be convinced otherwise if I could see where running the php as the app users, would make more sense. That depends on whether the boundary between user php and this unknown “appuser” is bidirectional or not. If there are things owned by user “php” that “appuser” should not see, then continuing to run the web app as two separate users makes sense. If “appuser” can be said to own everything in the web app, and the only reason you’re converting user php privileges to “appuser” privileges is so you don’t have to give user php access to everything in the web app, then I’d say my prior suggestion holds. Now that I know you’re using PHP, I can recommend something like PHP-FPM: https://php-fpm.org/ That’s the old external project. It’s now part of the PHP core: https://secure.php.net/manual/en/install.fpm.php You’ll have to use the old version with C5, though, as that happened after C5 was released. FPM isn’t the only way to go, just one idea, which happens to be well-supported within the PHP community. Regardless of the exact method, this lets you run your PHP code as a non-php user, letting Apache proxy to it using mod_fcgi. Now you’ve got strong separation between things Apache is allowed to read and things it must talk down through PHP to get access to. > It could be that giving sudo sys_ptrace access could increase the risk to the security of the system Once you give a process ptrace ability, it’s pretty much game over when it comes to security. The scope of what one process can do to another via ptrace(2) is HUUUUGE. I’d very much resist placating SELinux in this way. SELinux might in fact be warning you about a real attack here, which would explain why it’s intermittent.