kcm: switch order of device registration to fix a crash
authorJiri Slaby <jslaby@suse.cz>
Fri, 29 Mar 2019 11:19:46 +0000 (12:19 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 17 Apr 2019 06:36:44 +0000 (08:36 +0200)
[ Upstream commit 3c446e6f96997f2a95bf0037ef463802162d2323 ]

When kcm is loaded while many processes try to create a KCM socket, a
crash occurs:
 BUG: unable to handle kernel NULL pointer dereference at 000000000000000e
 IP: mutex_lock+0x27/0x40 kernel/locking/mutex.c:240
 PGD 8000000016ef2067 P4D 8000000016ef2067 PUD 3d6e9067 PMD 0
 Oops: 0002 [#1] SMP KASAN PTI
 CPU: 0 PID: 7005 Comm: syz-executor.5 Not tainted 4.12.14-396-default #1 SLE15-SP1 (unreleased)
 RIP: 0010:mutex_lock+0x27/0x40 kernel/locking/mutex.c:240
 RSP: 0018:ffff88000d487a00 EFLAGS: 00010246
 RAX: 0000000000000000 RBX: 000000000000000e RCX: 1ffff100082b0719
 ...
 CR2: 000000000000000e CR3: 000000004b1bc003 CR4: 0000000000060ef0
 Call Trace:
  kcm_create+0x600/0xbf0 [kcm]
  __sock_create+0x324/0x750 net/socket.c:1272
 ...

This is due to race between sock_create and unfinished
register_pernet_device. kcm_create tries to do "net_generic(net,
kcm_net_id)". but kcm_net_id is not initialized yet.

So switch the order of the two to close the race.

This can be reproduced with mutiple processes doing socket(PF_KCM, ...)
and one process doing module removal.

Fixes: ab7ac4eb9832 ("kcm: Kernel Connection Multiplexor module")
Reviewed-by: Michal Kubecek <mkubecek@suse.cz>
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
net/kcm/kcmsock.c

index 553d0ad..2f3cd09 100644 (file)
@@ -2058,14 +2058,14 @@ static int __init kcm_init(void)
        if (err)
                goto fail;
 
-       err = sock_register(&kcm_family_ops);
-       if (err)
-               goto sock_register_fail;
-
        err = register_pernet_device(&kcm_net_ops);
        if (err)
                goto net_ops_fail;
 
+       err = sock_register(&kcm_family_ops);
+       if (err)
+               goto sock_register_fail;
+
        err = kcm_proc_init();
        if (err)
                goto proc_init_fail;
@@ -2073,12 +2073,12 @@ static int __init kcm_init(void)
        return 0;
 
 proc_init_fail:
-       unregister_pernet_device(&kcm_net_ops);
-
-net_ops_fail:
        sock_unregister(PF_KCM);
 
 sock_register_fail:
+       unregister_pernet_device(&kcm_net_ops);
+
+net_ops_fail:
        proto_unregister(&kcm_proto);
 
 fail:
@@ -2094,8 +2094,8 @@ fail:
 static void __exit kcm_exit(void)
 {
        kcm_proc_exit();
-       unregister_pernet_device(&kcm_net_ops);
        sock_unregister(PF_KCM);
+       unregister_pernet_device(&kcm_net_ops);
        proto_unregister(&kcm_proto);
        destroy_workqueue(kcm_wq);