kprobes: Skip clearing aggrprobe's post_handler in kprobe-on-ftrace case
authorLi Huafei <lihuafei1@huawei.com>
Fri, 18 Nov 2022 01:15:34 +0000 (10:15 +0900)
committerMasami Hiramatsu (Google) <mhiramat@kernel.org>
Fri, 18 Nov 2022 01:15:34 +0000 (10:15 +0900)
In __unregister_kprobe_top(), if the currently unregistered probe has
post_handler but other child probes of the aggrprobe do not have
post_handler, the post_handler of the aggrprobe is cleared. If this is
a ftrace-based probe, there is a problem. In later calls to
disarm_kprobe(), we will use kprobe_ftrace_ops because post_handler is
NULL. But we're armed with kprobe_ipmodify_ops. This triggers a WARN in
__disarm_kprobe_ftrace() and may even cause use-after-free:

  Failed to disarm kprobe-ftrace at kernel_clone+0x0/0x3c0 (error -2)
  WARNING: CPU: 5 PID: 137 at kernel/kprobes.c:1135 __disarm_kprobe_ftrace.isra.21+0xcf/0xe0
  Modules linked in: testKprobe_007(-)
  CPU: 5 PID: 137 Comm: rmmod Not tainted 6.1.0-rc4-dirty #18
  [...]
  Call Trace:
   <TASK>
   __disable_kprobe+0xcd/0xe0
   __unregister_kprobe_top+0x12/0x150
   ? mutex_lock+0xe/0x30
   unregister_kprobes.part.23+0x31/0xa0
   unregister_kprobe+0x32/0x40
   __x64_sys_delete_module+0x15e/0x260
   ? do_user_addr_fault+0x2cd/0x6b0
   do_syscall_64+0x3a/0x90
   entry_SYSCALL_64_after_hwframe+0x63/0xcd
   [...]

For the kprobe-on-ftrace case, we keep the post_handler setting to
identify this aggrprobe armed with kprobe_ipmodify_ops. This way we
can disarm it correctly.

Link: https://lore.kernel.org/all/20221112070000.35299-1-lihuafei1@huawei.com/
Fixes: 0bc11ed5ab60 ("kprobes: Allow kprobes coexist with livepatch")
Reported-by: Zhao Gongyi <zhaogongyi@huawei.com>
Suggested-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Signed-off-by: Li Huafei <lihuafei1@huawei.com>
Acked-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Signed-off-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
kernel/kprobes.c

index cd9f5a6..3050631 100644 (file)
@@ -1766,7 +1766,13 @@ static int __unregister_kprobe_top(struct kprobe *p)
                                if ((list_p != p) && (list_p->post_handler))
                                        goto noclean;
                        }
-                       ap->post_handler = NULL;
+                       /*
+                        * For the kprobe-on-ftrace case, we keep the
+                        * post_handler setting to identify this aggrprobe
+                        * armed with kprobe_ipmodify_ops.
+                        */
+                       if (!kprobe_ftrace(ap))
+                               ap->post_handler = NULL;
                }
 noclean:
                /*