p = (char *)(path + strlen(path) - 4);
This is a very bad idea as path is user-supplied and has to be treated as malicious. An attacker can omit the string-terminator...
This is a very bad idea as path is user-supplied and has to be treated as malicious. An attacker can omit the string-terminator...
So your assumption on omitting the null terminator is wrong, BUT there's another flaw in this. The length of the string can be less than four bytes. By allocating the string in a nice page boundary this can be used to cause a memory protection fault.
But the kernel does not call strlen() on the user-supplied string...
The PATH_MAX guarantees that the string copy will terminate even if no null character is present.
Now if any routines that call strncpy_from_user assume that it leaves them with a null terminated string there is still a potential for trouble but that does not stem from the use of strncpy_from_user.
Check the implementation for do_getname:
http://lxr.free-electrons.com/source/fs/namei.c
And you'll see that it returns -ENAMETOOLONG when the name is >= the buffer allocated, only < len is allowed. (line 133 in the link above).
Anyone who's using open() can kill the kernel after loading this module.
And if you have that ability, worrying about the security of a module such as this seems, to me, totally and utterly pointless, because you could be inserting a rootkit, or a module that randomly deletes a file every time you press the letter A etc.
So could you explain what you mean please? Why be concerned about security when security is already compromised? How does it being multi-user make a difference?
Suppose your admin is a funny guy and loads this kernel module at April 1st. I (Jonny non-root user) can no longer listen to my favorite mp3, because the kernel module changes the path all the time. The admin loughs and has fun.
But then later I call open() with some fancy arguments and take over the whole machine (or at least crash it). This is the part where I lough.
is a absolute no-go. It's a nice and funny kernel module but it dooms the kernel's security.
What this means is that the kernel can no longer see if a user is writing to a read-only location, which is a major part of memory protection. It has major security implications. I don't know whether other parts of the kernel enable or disable this at context switches or at other times.
What this also means is that this kernel module will only work on x86 CPU's. So no rickrolling you ARM boxes or smartphones.
No, disabling WP in CR0 is only relevant for privileged code. Writes to write protected pages in user mode will still trap. It might mess up some in-kernel COW stuff though.