text
stringlengths
0
1.99k
in '/kernel/sys/syscall.c'.
long sys_open(const char * file, long flags, long mode) {
PTR_VALIDATE(file);
if (!file) return -EFAULT;
fs_node_t * node = kopen((char *)file, flags);
int access_bits = 0;
if (node && (flags & O_CREAT) && (flags & O_EXCL)) {
close_fs(node);
return -EEXIST;
}
...
The first thing the kernel does is to check that 'file' is a valid user
space pointer.
The 'ptr_validate()' function checks that the address is in user space
and is mapped with appropriate flags. This will be important later.
It then tries to open that file with 'kopen' and then performs access
checks to determine if the file already exists. Afterwards, it continues
to perform access checks.
This is how the OS enforces file system access permissions. If you want
to open a file it will check all of the permissions before the file is
ever visible in user mode.
...
int fd = process_append_fd(this_core->current_process, node);
...
return fd;
}
If all the checks have passed 'process_append_fd()' is called and the
file descriptor is now visible in the user mode process.
'fd' is then returned from the system call and the libc then returns it
from 'open()'.
Since the checks here look sane, we need to change either the files
permissions or elevate our privileges. Let's take a look at elevating
privileges.
---[ 4.2 - Becoming root normally
You may have wondered how 'sudo' can make you 'root' on a Linux system.
It is definitely one of those "obvious" things I mentioned earlier, so
you may never have given it a second thought. But if you do, it seems a
little odd.
'sudo' is a program that runs in 'user mode' in ring 3 like any other.
It can't issue a magic CPU instruction that changes the user and it
can't write in kernel memory. If it could then so could any other user
mode process.
Clearly it uses the 'setuid()' libc function, but using it to switch to
another user requires privileges.
But we can run 'sudo' as a low-privileged user to become root, so what
makes 'sudo' special?
You probably already know that the way it works is that the file system
doesn't just store permissions for read/write/execute access, but can
also store flags and capabilities.
Particularly the SUID flag denotes that a program should be executed
not as the user that starts it, but as the user that owns the file.
On ToaruOS it works exactly the same way as it does on Linux:
local@livecd ~$ ls -al /bin/sudo
-r-sr-xr-x 1 root root 10384 Mar 16 17:26 /bin/sudo
Note that instead of 'x' it shows 's' for the execute permission,
showing the SUID bit is set.
---[ 4.3 - SUID on the kernel side
The implementation of the SUID bit is very straight-forward in ToaruOS
and can be found in 'elf_exec()' in '/kernel/misc/elf64.c'.
if ((file->mask & S_ISUID) &&
!(this_core->current_process->flags &
(PROC_FLAG_TRACE_SYSCALLS | PROC_FLAG_TRACE_SIGNALS)))
{
/* setuid */
this_core->current_process->user = file->uid;
}
This is already the full implementation. If the 'S_ISUID' flag of the