connector: fix regression introduced by sid connector
Christian Borntraeger [Mon, 26 Oct 2009 23:49:34 +0000 (16:49 -0700)]
Since commit 02b51df1b07b4e9ca823c89284e704cadb323cd1 (proc connector: add
event for process becoming session leader) we have the following warning:

Badness at kernel/softirq.c:143
Krnl PSW : 0404c00180000000 00000000001481d4 (local_bh_enable+0xb0/0xe0)
Call Trace:
([<000000013fe04100>] 0x13fe04100)
 [<000000000048a946>] sk_filter+0x9a/0xd0
 [<000000000049d938>] netlink_broadcast+0x2c0/0x53c
 [<00000000003ba9ae>] cn_netlink_send+0x272/0x2b0
 [<00000000003baef0>] proc_sid_connector+0xc4/0xd4
 [<0000000000142604>] __set_special_pids+0x58/0x90
 [<0000000000159938>] sys_setsid+0xb4/0xd8
 [<00000000001187fe>] sysc_noemu+0x10/0x16
 [<00000041616cb266>] 0x41616cb266

The warning is
--->    WARN_ON_ONCE(in_irq() || irqs_disabled());

The network code must not be called with disabled interrupts but
sys_setsid holds the tasklist_lock with spinlock_irq while calling the

After a discussion we agreed that we can move proc_sid_connector from
__set_special_pids to sys_setsid.

We also agreed that it is sufficient to change the check from
task_session(curr) != pid into err > 0, since if we don't change the
session, this means we were already the leader and return -EPERM.

One last thing:
There is also daemonize(), and some people might want to get a
notification in that case. Since daemonize() is only needed if a user
space does kernel_thread this does not look important (and there seems
to be no consensus if this connector should be called in daemonize). If
we really want this, we can add proc_sid_connector to daemonize() in an
additional patch (Scott?)

Signed-off-by: Christian Borntraeger <>
Cc: Scott James Remnant <>
Cc: Matt Helsley <>
Cc: David S. Miller <>
Acked-by: Oleg Nesterov <>
Acked-by: Evgeniy Polyakov <>
Acked-by: David Rientjes <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>


index e61891f..f7864ac 100644 (file)
@@ -359,10 +359,8 @@ void __set_special_pids(struct pid *pid)
        struct task_struct *curr = current->group_leader;
-       if (task_session(curr) != pid) {
+       if (task_session(curr) != pid)
                change_pid(curr, PIDTYPE_SID, pid);
-               proc_sid_connector(curr);
-       }
        if (task_pgrp(curr) != pid)
                change_pid(curr, PIDTYPE_PGID, pid);
index 255475d..1828f8d 100644 (file)
@@ -1110,6 +1110,8 @@ SYSCALL_DEFINE0(setsid)
        err = session;
+       if (err > 0)
+               proc_sid_connector(group_leader);
        return err;