SSD Advisory – Linux Kernel XFRM Privilege Escalation Full report: https://blogs.securiteam.com/index.php/archives/3535 Twitter: @SecuriTeam_SSD Weibo: SecuriTeam_SSD Vulnerability Summary The following advisory describes a Use-after-free vulnerability found in Linux kernel that can lead to privilege escalation. The vulnerability found in Netlink socket subsystem – XFRM. Netlink is used to transfer information between the kernel and user-space processes. It consists of a standard sockets-based interface for user space processes and an internal kernel API for kernel modules. Credit An independent security researcher, Mohamed Ghannam, has reported this vulnerability to Beyond Security’s SecuriTeam Secure Disclosure program Vendor response The vulnerability has been addressed as part of 1137b5e (“ipsec: Fix aborted xfrm policy dump crash”) Vulnerability details An unprivileged user can change Netlink socket subsystem – XFRM value sk->sk_rcvbuf (sk == struct sock object). The value can be changed into specific range via setsockopt(SO_RCVBUF). sk_rcvbuf is the total number of bytes of a buffer receiving data via recvmsg/recv/read. The sk_rcvbuf value is how many bytes the kernel should allocate for the skb (struct sk_buff objects). skb->trusize is a variable which keep track of how many bytes of memory are consumed, in order to not wasting and manage memory, the kernel can handle the skb size at run time. For example, if we allocate a large socket buffer (skb) and we only received 1-byte packet size, the kernel will adjust this by calling skb_set_owner_r. By calling skb_set_owner_r the sk->sk_rmem_alloc (refers to an atomic variable sk->sk_backlog.rmem_alloc) is modified. When we create a XFRM netlink socket, xfrm_dump_policy is called, when we close the socket xfrm_dump_policy_done is called. xfrm_dump_policy_done is called whenever cb_running for netlink_sock object is true. The xfrm_dump_policy_done tries to clean-up a xfrm walk entry which is managed by netlink_callback object. When netlink_skb_set_owner_r is called (like skb_set_owner_r) it updates the sk_rmem_alloc. netlink_dump(): In above snippet we can see that netlink_dump() check fails when sk->sk_rcvbuf is smaller than sk_rmem_alloc (notice that we can control sk->sk_rcvbuf via stockpot). When this condition fails, it jumps to the end of a function and quit with failure and the value of cb_running doesn’t changed to false. nlk->cb_running is true, thus xfrm_dump_policy_done() is being called. nlk->cb.done points to xfrm_dump_policy_done, it worth noting that this function handles a doubly linked list, so if we can tweak this vulnerability to reference a controlled buffer, we could have a read/write what/where primitive. -- Thanks Maor Shwartz Beyond Security GPG Key ID: 6D273779F52A9FC2