net: ipv6: fib: don't sleep inside atomic lock
authorBenjamin Block <bebl@mageta.org>
Thu, 21 Aug 2014 17:37:48 +0000 (19:37 +0200)
committerDavid S. Miller <davem@davemloft.net>
Fri, 22 Aug 2014 17:54:49 +0000 (10:54 -0700)
The function fib6_commit_metrics() allocates a piece of memory in mode
GFP_KERNEL while holding an atomic lock from higher up in the stack, in
the function __ip6_ins_rt(). This produces the following BUG:

> BUG: sleeping function called from invalid context at mm/slub.c:1250
> in_atomic(): 1, irqs_disabled(): 0, pid: 2909, name: dhcpcd
> 2 locks held by dhcpcd/2909:
>  #0:  (rtnl_mutex){+.+.+.}, at: [<ffffffff81978e67>] rtnl_lock+0x17/0x20
>  #1:  (&tb->tb6_lock){++--+.}, at: [<ffffffff81a6951a>] ip6_route_add+0x65a/0x800
> CPU: 1 PID: 2909 Comm: dhcpcd Not tainted 3.17.0-rc1 #1
> Hardware name: ASUS All Series/Q87T, BIOS 0216 10/16/2013
>  0000000000000008 ffff8800c8f13858 ffffffff81af135a 0000000000000000
>  ffff880212202430 ffff8800c8f13878 ffffffff810f8d3a ffff880212202c98
>  0000000000000010 ffff8800c8f138c8 ffffffff8121ad0e 0000000000000001
> Call Trace:
>  [<ffffffff81af135a>] dump_stack+0x4e/0x68
>  [<ffffffff810f8d3a>] __might_sleep+0x10a/0x120
>  [<ffffffff8121ad0e>] kmem_cache_alloc_trace+0x4e/0x190
>  [<ffffffff81a6bcd6>] ? fib6_commit_metrics+0x66/0x110
>  [<ffffffff81a6bcd6>] fib6_commit_metrics+0x66/0x110
>  [<ffffffff81a6cbf3>] fib6_add+0x883/0xa80
>  [<ffffffff81a6951a>] ? ip6_route_add+0x65a/0x800
>  [<ffffffff81a69535>] ip6_route_add+0x675/0x800
>  [<ffffffff81a68f2a>] ? ip6_route_add+0x6a/0x800
>  [<ffffffff81a6990c>] inet6_rtm_newroute+0x5c/0x80
>  [<ffffffff8197cf01>] rtnetlink_rcv_msg+0x211/0x260
>  [<ffffffff81978e67>] ? rtnl_lock+0x17/0x20
>  [<ffffffff81119708>] ? lock_release_holdtime+0x28/0x180
>  [<ffffffff81978e67>] ? rtnl_lock+0x17/0x20
>  [<ffffffff8197ccf0>] ? __rtnl_unlock+0x20/0x20
>  [<ffffffff819a989e>] netlink_rcv_skb+0x6e/0xd0
>  [<ffffffff81978ee5>] rtnetlink_rcv+0x25/0x40
>  [<ffffffff819a8e59>] netlink_unicast+0xd9/0x180
>  [<ffffffff819a9600>] netlink_sendmsg+0x700/0x770
>  [<ffffffff81103735>] ? local_clock+0x25/0x30
>  [<ffffffff8194e83c>] sock_sendmsg+0x6c/0x90
>  [<ffffffff811f98e3>] ? might_fault+0xa3/0xb0
>  [<ffffffff8195ca6d>] ? verify_iovec+0x7d/0xf0
>  [<ffffffff8194ec3e>] ___sys_sendmsg+0x37e/0x3b0
>  [<ffffffff8111ef15>] ? trace_hardirqs_on_caller+0x185/0x220
>  [<ffffffff81af979e>] ? mutex_unlock+0xe/0x10
>  [<ffffffff819a55ec>] ? netlink_insert+0xbc/0xe0
>  [<ffffffff819a65e5>] ? netlink_autobind.isra.30+0x125/0x150
>  [<ffffffff819a6520>] ? netlink_autobind.isra.30+0x60/0x150
>  [<ffffffff819a84f9>] ? netlink_bind+0x159/0x230
>  [<ffffffff811f989a>] ? might_fault+0x5a/0xb0
>  [<ffffffff8194f25e>] ? SYSC_bind+0x7e/0xd0
>  [<ffffffff8194f8cd>] __sys_sendmsg+0x4d/0x80
>  [<ffffffff8194f912>] SyS_sendmsg+0x12/0x20
>  [<ffffffff81afc692>] system_call_fastpath+0x16/0x1b

Fixing this by replacing the mode GFP_KERNEL with GFP_ATOMIC.

Signed-off-by: Benjamin Block <bebl@mageta.org>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv6/ip6_fib.c

index cb4459b..76b7f5e 100644 (file)
@@ -643,7 +643,7 @@ static int fib6_commit_metrics(struct dst_entry *dst,
        if (dst->flags & DST_HOST) {
                mp = dst_metrics_write_ptr(dst);
        } else {
-               mp = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
+               mp = kzalloc(sizeof(u32) * RTAX_MAX, GFP_ATOMIC);
                if (!mp)
                        return -ENOMEM;
                dst_init_metrics(dst, mp, 0);