Linux NULL pointer dereference due to incorrect proto_ops initializations ------------------------------------------------------------------------- In the Linux kernel, each socket has an associated struct of operations called proto_ops which contain pointers to functions implementing various features, such as accept, bind, shutdown, and so on. If an operation on a particular socket is unimplemented, they are expected to point the associated function pointer to predefined stubs, for example if the "accept" operation is undefined it would point to sock_no_accept(). However, we have found that this is not always the case and some of these pointers are left uninitialized. This is not always a security issue, as the kernel validates the pointers at the call site, such as this example from sock_splice_read: static ssize_t sock_splice_read(struct file *file, loff_t *ppos, struct pipe_inode_info *pipe, size_t len, unsigned int flags) { struct socket *sock = file->private_data; if (unlikely(!sock->ops->splice_read)) return -EINVAL; return sock->ops->splice_read(sock, ppos, pipe, len, flags); } But we have found an example where this is not the case; the sock_sendpage() routine does not validate the function pointer is valid before dereferencing it, and therefore relies on the correct initialization of the proto_ops structure. We have identified several examples where the initialization is incomplete: - The SOCKOPS_WRAP macro defined in include/linux/net.h, which appears correct at first glance, was actually affected. This includes PF_APPLETALK, PF_IPX, PF_IRDA, PF_X25 and PF_AX25 families. - Initializations were missing in other protocols, including PF_BLUETOOTH, PF_IUCV, PF_INET6 (with IPPROTO_SCTP), PF_PPPOX and PF_ISDN. -------------------- Affected Software ------------------------ All Linux 2.4/2.6 versions since May 2001 are believed to be affected: - Linux 2.4, from 2.4.4 up to and including 2.4.37.4 - Linux 2.6, from 2.6.0 up to and including 2.6.30.4 -------------------- Consequences ----------------------- This issue is easily exploitable for local privilege escalation. In order to exploit this, an attacker would create a mapping at address zero containing code to be executed with privileges of the kernel, and then trigger a vulnerable operation using a sequence like this: /* ... */ int fdin = mkstemp(template); int fdout = socket(PF_PPPOX, SOCK_DGRAM, 0); unlink(template); ftruncate(fdin, PAGE_SIZE); sendfile(fdout, fdin, NULL, PAGE_SIZE); /* ... */ Please note, sendfile() is just one of many ways to cause a sendpage operation on a socket. Successful exploitation will lead to complete attacker control of the system. ------------------- Mitigation ----------------------- Recent kernels with mmap_min_addr support may prevent exploitation if the sysctl vm.mmap_min_addr is set above zero. However, administrators should be aware that LSM based mandatory access control systems, such as SELinux, may alter this functionality. It should also be noted that all kernels up to 2.6.30.2 are vulnerable to published attacks against mmap_min_addr. ------------------- Solution ----------------------- Linus committed a patch correcting this issue on 13th August 2009. http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=e694958388c50148389b0e9b9e9e8945cf0f1b98 ------------------- Credit ----------------------- This bug was discovered by Tavis Ormandy and Julien Tinnes of the Google Security Team.