[PATCH 12/19] NET: set PF_FSTRANS while holding rtnl_lock

[NeilBrown <neilb@xxxxxxx>]

As rtnl_mutex can be taken while holding sk_lock, and sk_lock can be
taken while performing memory reclaim (at least when loop-back NFS is
active), any memory allocation under rtnl_mutex must avoid __GFP_FS,
which is most easily done by setting PF_MEMALLOC.


        CPU0                    CPU1
        ----                    ----
   lock(rtnl_mutex);
                                lock(sk_lock-AF_INET);
                                lock(rtnl_mutex);
   <Memory allocation/reclaim>
     lock(sk_lock-AF_INET);

  *** DEADLOCK ***

1/ rtnl_mutex is taken while holding sk_lock:

    [<ffffffff81abb442>] rtnl_lock+0x12/0x20
    [<ffffffff81b28c3a>] ip_mc_leave_group+0x2a/0x160
    [<ffffffff81aec70b>] do_ip_setsockopt.isra.18+0x96b/0xed0
    [<ffffffff81aecc97>] ip_setsockopt+0x27/0x90
    [<ffffffff81b151c6>] udp_setsockopt+0x16/0x30
    [<ffffffff81a9144f>] sock_common_setsockopt+0xf/0x20
    [<ffffffff81a907de>] SyS_setsockopt+0x5e/0xc0

2/ memory is allocated under rtnl_mutex:
    [<ffffffff8166eb41>] kobject_set_name_vargs+0x21/0x70
    [<ffffffff81840d92>] dev_set_name+0x42/0x50
    [<ffffffff81ac5e97>] netdev_register_kobject+0x57/0x130
    [<ffffffff81aaf574>] register_netdevice+0x354/0x550
    [<ffffffff81aaf785>] register_netdev+0x15/0x30


Signed-off-by: NeilBrown <neilb@xxxxxxx>
---
 net/core/rtnetlink.c |    9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c
index 120eecc0f5a4..6870211e93a6 100644
--- a/net/core/rtnetlink.c
+++ b/net/core/rtnetlink.c
@@ -61,15 +61,18 @@ struct rtnl_link {
 };
 
 static DEFINE_MUTEX(rtnl_mutex);
+static int rtnl_pflags;
 
 void rtnl_lock(void)
 {
 	mutex_lock(&rtnl_mutex);
+	current_set_flags_nested(&rtnl_pflags, PF_FSTRANS);
 }
 EXPORT_SYMBOL(rtnl_lock);
 
 void __rtnl_unlock(void)
 {
+	current_restore_flags_nested(&rtnl_pflags, PF_FSTRANS);
 	mutex_unlock(&rtnl_mutex);
 }
 
@@ -82,7 +85,11 @@ EXPORT_SYMBOL(rtnl_unlock);
 
 int rtnl_trylock(void)
 {
-	return mutex_trylock(&rtnl_mutex);
+	if (mutex_trylock(&rtnl_mutex)) {
+		current_set_flags_nested(&rtnl_pflags, PF_FSTRANS);
+		return 1;
+	}
+	return 0;
 }
 EXPORT_SYMBOL(rtnl_trylock);
 


--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html