]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commitdiff
ipv4: fix the rcu race between free_fib_info and ip_route_output_slow
authorYanmin Zhang <yanmin_zhang@linux.intel.com>
Wed, 23 May 2012 15:39:45 +0000 (15:39 +0000)
committerDavid S. Miller <davem@davemloft.net>
Thu, 24 May 2012 04:28:21 +0000 (00:28 -0400)
We hit a kernel OOPS.

<3>[23898.789643] BUG: sleeping function called from invalid context at
/data/buildbot/workdir/ics/hardware/intel/linux-2.6/arch/x86/mm/fault.c:1103
<3>[23898.862215] in_atomic(): 0, irqs_disabled(): 0, pid: 10526, name:
Thread-6683
<4>[23898.967805] HSU serial 0000:00:05.1: 0000:00:05.2:HSU serial prevented me
to suspend...
<4>[23899.258526] Pid: 10526, comm: Thread-6683 Tainted: G        W
3.0.8-137685-ge7742f9 #1
<4>[23899.357404] HSU serial 0000:00:05.1: 0000:00:05.2:HSU serial prevented me
to suspend...
<4>[23899.904225] Call Trace:
<4>[23899.989209]  [<c1227f50>] ? pgtable_bad+0x130/0x130
<4>[23900.000416]  [<c1238c2a>] __might_sleep+0x10a/0x110
<4>[23900.007357]  [<c1228021>] do_page_fault+0xd1/0x3c0
<4>[23900.013764]  [<c18e9ba9>] ? restore_all+0xf/0xf
<4>[23900.024024]  [<c17c007b>] ? napi_complete+0x8b/0x690
<4>[23900.029297]  [<c1227f50>] ? pgtable_bad+0x130/0x130
<4>[23900.123739]  [<c1227f50>] ? pgtable_bad+0x130/0x130
<4>[23900.128955]  [<c18ea0c3>] error_code+0x5f/0x64
<4>[23900.133466]  [<c1227f50>] ? pgtable_bad+0x130/0x130
<4>[23900.138450]  [<c17f6298>] ? __ip_route_output_key+0x698/0x7c0
<4>[23900.144312]  [<c17f5f8d>] ? __ip_route_output_key+0x38d/0x7c0
<4>[23900.150730]  [<c17f63df>] ip_route_output_flow+0x1f/0x60
<4>[23900.156261]  [<c181de58>] ip4_datagram_connect+0x188/0x2b0
<4>[23900.161960]  [<c18e981f>] ? _raw_spin_unlock_bh+0x1f/0x30
<4>[23900.167834]  [<c18298d6>] inet_dgram_connect+0x36/0x80
<4>[23900.173224]  [<c14f9e88>] ? _copy_from_user+0x48/0x140
<4>[23900.178817]  [<c17ab9da>] sys_connect+0x9a/0xd0
<4>[23900.183538]  [<c132e93c>] ? alloc_file+0xdc/0x240
<4>[23900.189111]  [<c123925d>] ? sub_preempt_count+0x3d/0x50

Function free_fib_info resets nexthop_nh->nh_dev to NULL before releasing
fi. Other cpu might be accessing fi. Fixing it by delaying the releasing.

With the patch, we ran MTBF testing on Android mobile for 12 hours
and didn't trigger the issue.

Thank Eric for very detailed review/checking the issue.

Signed-off-by: Yanmin Zhang <yanmin_zhang@linux.intel.com>
Signed-off-by: Kun Jiang <kunx.jiang@intel.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/fib_semantics.c

index a8bdf7405433f55adcd0ef540dea1f0cf858c5f9..e5b7182fa099e683ee0484b3772c9aa58eba0d2f 100644 (file)
@@ -145,6 +145,12 @@ static void free_fib_info_rcu(struct rcu_head *head)
 {
        struct fib_info *fi = container_of(head, struct fib_info, rcu);
 
+       change_nexthops(fi) {
+               if (nexthop_nh->nh_dev)
+                       dev_put(nexthop_nh->nh_dev);
+       } endfor_nexthops(fi);
+
+       release_net(fi->fib_net);
        if (fi->fib_metrics != (u32 *) dst_default_metrics)
                kfree(fi->fib_metrics);
        kfree(fi);
@@ -156,13 +162,7 @@ void free_fib_info(struct fib_info *fi)
                pr_warn("Freeing alive fib_info %p\n", fi);
                return;
        }
-       change_nexthops(fi) {
-               if (nexthop_nh->nh_dev)
-                       dev_put(nexthop_nh->nh_dev);
-               nexthop_nh->nh_dev = NULL;
-       } endfor_nexthops(fi);
        fib_info_cnt--;
-       release_net(fi->fib_net);
        call_rcu(&fi->rcu, free_fib_info_rcu);
 }