Re: [patch 7/8] fdmap v2 - implement sys_socket2

From: Kyle Moffett
Date: Sat Jun 09 2007 - 15:28:40 EST

On Jun 09, 2007, at 13:24:29, Al Viro wrote:
On Sat, Jun 09, 2007 at 10:08:59AM -0700, Ulrich Drepper wrote:
- - there are two interface to use: open + fcntl. This is racy. And don't tell me this doesn't matter.
Racy with respect to what? Return-to-libc exploits from another thread?

How about racy with respect to normal open or fork+exec from another thread? Specifically there are cases where libc or other libraries want to create a backend thread dealing with file descriptors in response to the program's straightforward calls into that library (Examples include using syslets or event-based polling threads).


Program Thread: Library Thread:
fd = socket(AF_*, SOCK_*, 0);
int x = FD_CLOEXEC;
fcntl(fd, F_SETFD, &x);

New Process:

Whoops!!! Suddenly the user process executed by the (theoretically) single-threaded program got a handle to a netlink socket affecting some system resource!!!


Program Thread: Async libc getpwent()-cache syslet
fd = open("/etc/shadow");

Here we were trying to safely call into code which reads from stdin and shouldn't be given privileged data, but the syslet makes the common paradigm 'close(0); open("/dev/null");' horribly insecure.

If you extend all the FD syscalls to all take a "flags" parameter and add the appropriate flags, then you can pass O_CLOEXEC|O_RANDFD to whatever syscall you are using and both problems vanish.

Kyle Moffett

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at