CVE-2026-64563

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> rhashtable: clear stale iter-&gt;p on table restart<br /> <br /> rhashtable_walk_start_check() has two restart paths when resuming a walk.<br /> When iter-&gt;walker.tbl is valid, it re-validates iter-&gt;p against the table<br /> and sets iter-&gt;p = NULL if the object is gone. When iter-&gt;walker.tbl is<br /> NULL (table was freed during resize), it resets slot and skip but forgets<br /> to clear iter-&gt;p.<br /> <br /> rhashtable_walk_next() then dereferences the stale iter-&gt;p, reading<br /> freed memory. This is a use-after-free.<br /> <br /> Any caller that does multi-fragment rhashtable walks across<br /> walk_stop/walk_start boundaries is affected. Concrete cases include<br /> netlink_diag (__netlink_diag_dump in net/netlink/diag.c) and TIPC<br /> (tipc_nl_sk_walk in net/tipc/socket.c).<br /> <br /> Crash stack (netlink_diag):<br /> BUG: KASAN: slab-use-after-free in rhashtable_walk_next+0x365/0x3c0<br /> Read of size 8 at addr ffff88801a9d2438 (freed kmalloc-2k, offset 1080)<br /> Call Trace:<br /> rhashtable_walk_next+0x365/0x3c0 (lib/rhashtable.c:1016)<br /> __netlink_diag_dump+0x160/0x760 (net/netlink/diag.c:122)<br /> netlink_diag_dump+0xc2/0x240<br /> netlink_dump+0x5bc/0x1270<br /> netlink_recvmsg+0x7a3/0x980<br /> sock_recvmsg+0x1bc/0x200<br /> __sys_recvfrom+0x1d4/0x2c0