[PATCH 2.6.8.1] bonding: bonding might sleep with lock held

[Jay Vosburgh <fubar@xxxxxxxxxx>]

        I posted a patch for this a while back, Andi Kleen suggested an
alternate patch, which never made it into the mainline; here it is.

        To summarize the problem: In 802.3ad mode, the bond_close()
function will eventually call dev_remove_pack() with a lock held.  In
turn, dev_remove_pack() calls synchronize_net() which eventually might
sleep in wait_for_completion().

        Patch generated against 2.6.7, but applied ok to 2.6.8.1.

        -J

---
        -Jay Vosburgh, IBM Linux Technology Center, fubar@xxxxxxxxxx

Signed-off-by: Jay Vosburgh <fubar@xxxxxxxxxx>

diff -urN linux-2.6.7-virgin/drivers/net/bonding/bond_main.c 
linux-2.6.7/drivers/net/bonding/bond_main.c
--- linux-2.6.7-virgin/drivers/net/bonding/bond_main.c  2004-08-17 
14:56:04.411984128 -0700
+++ linux-2.6.7/drivers/net/bonding/bond_main.c 2004-08-17 14:57:08.441968256 
-0700
@@ -3566,15 +3566,15 @@
 {
        struct bonding *bond = bond_dev->priv;
 
-       write_lock_bh(&bond->lock);
-
-       bond_mc_list_destroy(bond);
-
        if (bond->params.mode == BOND_MODE_8023AD) {
                /* Unregister the receive of LACPDUs */
                bond_unregister_lacpdu(bond);
        }
 
+       write_lock_bh(&bond->lock);
+
+       bond_mc_list_destroy(bond);
+
        /* signal timers not to re-arm */
        bond->kill_timers = 1;