CVE-2022-50409

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> net: If sock is dead don&amp;#39;t access sock&amp;#39;s sk_wq in sk_stream_wait_memory<br /> <br /> Fixes the below NULL pointer dereference:<br /> <br /> [...]<br /> [ 14.471200] Call Trace:<br /> [ 14.471562] <br /> [ 14.471882] lock_acquire+0x245/0x2e0<br /> [ 14.472416] ? remove_wait_queue+0x12/0x50<br /> [ 14.473014] ? _raw_spin_lock_irqsave+0x17/0x50<br /> [ 14.473681] _raw_spin_lock_irqsave+0x3d/0x50<br /> [ 14.474318] ? remove_wait_queue+0x12/0x50<br /> [ 14.474907] remove_wait_queue+0x12/0x50<br /> [ 14.475480] sk_stream_wait_memory+0x20d/0x340<br /> [ 14.476127] ? do_wait_intr_irq+0x80/0x80<br /> [ 14.476704] do_tcp_sendpages+0x287/0x600<br /> [ 14.477283] tcp_bpf_push+0xab/0x260<br /> [ 14.477817] tcp_bpf_sendmsg_redir+0x297/0x500<br /> [ 14.478461] ? __local_bh_enable_ip+0x77/0xe0<br /> [ 14.479096] tcp_bpf_send_verdict+0x105/0x470<br /> [ 14.479729] tcp_bpf_sendmsg+0x318/0x4f0<br /> [ 14.480311] sock_sendmsg+0x2d/0x40<br /> [ 14.480822] ____sys_sendmsg+0x1b4/0x1c0<br /> [ 14.481390] ? copy_msghdr_from_user+0x62/0x80<br /> [ 14.482048] ___sys_sendmsg+0x78/0xb0<br /> [ 14.482580] ? vmf_insert_pfn_prot+0x91/0x150<br /> [ 14.483215] ? __do_fault+0x2a/0x1a0<br /> [ 14.483738] ? do_fault+0x15e/0x5d0<br /> [ 14.484246] ? __handle_mm_fault+0x56b/0x1040<br /> [ 14.484874] ? lock_is_held_type+0xdf/0x130<br /> [ 14.485474] ? find_held_lock+0x2d/0x90<br /> [ 14.486046] ? __sys_sendmsg+0x41/0x70<br /> [ 14.486587] __sys_sendmsg+0x41/0x70<br /> [ 14.487105] ? intel_pmu_drain_pebs_core+0x350/0x350<br /> [ 14.487822] do_syscall_64+0x34/0x80<br /> [ 14.488345] entry_SYSCALL_64_after_hwframe+0x63/0xcd<br /> [...]<br /> <br /> The test scenario has the following flow:<br /> <br /> thread1 thread2<br /> ----------- ---------------<br /> tcp_bpf_sendmsg<br /> tcp_bpf_send_verdict<br /> tcp_bpf_sendmsg_redir sock_close<br /> tcp_bpf_push_locked __sock_release<br /> tcp_bpf_push //inet_release<br /> do_tcp_sendpages sock-&gt;ops-&gt;release<br /> sk_stream_wait_memory // tcp_close<br /> sk_wait_event sk-&gt;sk_prot-&gt;close<br /> release_sock(__sk);<br /> ***<br /> lock_sock(sk);<br /> __tcp_close<br /> sock_orphan(sk)<br /> sk-&gt;sk_wq = NULL<br /> release_sock<br /> ****<br /> lock_sock(__sk);<br /> remove_wait_queue(sk_sleep(sk), &amp;wait);<br /> sk_sleep(sk)<br /> //NULL pointer dereference<br /> &amp;rcu_dereference_raw(sk-&gt;sk_wq)-&gt;wait<br /> <br /> While waiting for memory in thread1, the socket is released with its wait<br /> queue because thread2 has closed it. This caused by tcp_bpf_send_verdict<br /> didn&amp;#39;t increase the f_count of psock-&gt;sk_redir-&gt;sk_socket-&gt;file in thread1.<br /> <br /> We should check if SOCK_DEAD flag is set on wakeup in sk_stream_wait_memory<br /> before accessing the wait queue.