Luna/libos/include/os/Security.h

49 lines
2.2 KiB
C++

/**
* @file Security.h
* @author apio (cloudapio.eu)
* @brief Functions to restrict process operations.
*
* @copyright Copyright (c) 2023, the Luna authors.
*
*/
#pragma once
#include <luna/Result.h>
namespace os
{
namespace Security
{
/**
* @brief Restrict system operations.
*
* The pledge() call, borrowed from OpenBSD, is a simple function to sandbox a process effectively.
*
* Syscalls are divided into a number of categories ("promises"), the following are the ones implemented on
* Luna: stdio rpath wpath cpath fattr chown unix tty proc exec prot_exec id mount signal host error
*
* The way pledge() works is: the process "tells" the kernel which subset of functions it will use, and if it
* suddenly uses something it has not promised (probably because the process was hacked, using ROP or something
* else) the kernel kills the process immediately with an uncatchable SIGABRT. Alternatively, if the process has
* pledged the "error" promise, the call will fail with ENOSYS.
*
* Pledges are not inherited across exec, although one may specify another set of promises to apply on the next
* execve() call. Thus, pledge() is not a way to restrict untrusted programs (unless the "exec" pledge is
* removed), but more of a way to protect trusted local programs from vulnerabilities.
*
* One may call pledge() several times, but only to remove promises, not to add them.
*
* A typical call to pledge would look like this:
*
* TRY(os::Security::pledge("stdio rpath wpath unix proc", nullptr));
*
* @param promises The promises to apply immediately, separated by spaces. If empty, the process may only call
* _exit(2). If NULL, the promises are not changed.
* @param execpromises The promises to apply on the next call to execve(2), separated by spaces. If empty, the
* process may only call _exit(2). If NULL, the execpromises are not changed.
* @return Result<void> Whether the operation succeeded.
*/
Result<void> pledge(const char* promises, const char* execpromises);
}
}