CVE-2024-26923

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> af_unix: Fix garbage collector racing against connect()<br /> <br /> Garbage collector does not take into account the risk of embryo getting<br /> enqueued during the garbage collection. If such embryo has a peer that<br /> carries SCM_RIGHTS, two consecutive passes of scan_children() may see a<br /> different set of children. Leading to an incorrectly elevated inflight<br /> count, and then a dangling pointer within the gc_inflight_list.<br /> <br /> sockets are AF_UNIX/SOCK_STREAM<br /> S is an unconnected socket<br /> L is a listening in-flight socket bound to addr, not in fdtable<br /> V&amp;#39;s fd will be passed via sendmsg(), gets inflight count bumped<br /> <br /> connect(S, addr) sendmsg(S, [V]); close(V) __unix_gc()<br /> ---------------- ------------------------- -----------<br /> <br /> NS = unix_create1()<br /> skb1 = sock_wmalloc(NS)<br /> L = unix_find_other(addr)<br /> unix_state_lock(L)<br /> unix_peer(S) = NS<br /> // V count=1 inflight=0<br /> <br /> NS = unix_peer(S)<br /> skb2 = sock_alloc()<br /> skb_queue_tail(NS, skb2[V])<br /> <br /> // V became in-flight<br /> // V count=2 inflight=1<br /> <br /> close(V)<br /> <br /> // V count=1 inflight=1<br /> // GC candidate condition met<br /> <br /> for u in gc_inflight_list:<br /> if (total_refs == inflight_refs)<br /> add u to gc_candidates<br /> <br /> // gc_candidates={L, V}<br /> <br /> for u in gc_candidates:<br /> scan_children(u, dec_inflight)<br /> <br /> // embryo (skb1) was not<br /> // reachable from L yet, so V&amp;#39;s<br /> // inflight remains unchanged<br /> __skb_queue_tail(L, skb1)<br /> unix_state_unlock(L)<br /> for u in gc_candidates:<br /> if (u.inflight)<br /> scan_children(u, inc_inflight_move_tail)<br /> <br /> // V count=1 inflight=2 (!)<br /> <br /> If there is a GC-candidate listening socket, lock/unlock its state. This<br /> makes GC wait until the end of any ongoing connect() to that socket. After<br /> flipping the lock, a possibly SCM-laden embryo is already enqueued. And if<br /> there is another embryo coming, it can not possibly carry SCM_RIGHTS. At<br /> this point, unix_inflight() can not happen because unix_gc_lock is already<br /> taken. Inflight graph remains unaffected.