sctp: Fix list corruption resulting from freeing an association on a list
authorNeil Horman <nhorman@tuxdriver.com>
Mon, 16 Jul 2012 09:13:51 +0000 (09:13 +0000)
committerDavid S. Miller <davem@davemloft.net>
Tue, 17 Jul 2012 05:32:26 +0000 (22:32 -0700)
A few days ago Dave Jones reported this oops:

[22766.294255] general protection fault: 0000 [#1] PREEMPT SMP
[22766.295376] CPU 0
[22766.295384] Modules linked in:
[22766.387137]  ffffffffa169f292 6b6b6b6b6b6b6b6b ffff880147c03a90
ffff880147c03a74
[22766.387135] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 00000000000
[22766.387136] Process trinity-watchdo (pid: 10896, threadinfo ffff88013e7d2000,
[22766.387137] Stack:
[22766.387140]  ffff880147c03a10
[22766.387140]  ffffffffa169f2b6
[22766.387140]  ffff88013ed95728
[22766.387143]  0000000000000002
[22766.387143]  0000000000000000
[22766.387143]  ffff880003fad062
[22766.387144]  ffff88013c120000
[22766.387144]
[22766.387145] Call Trace:
[22766.387145]  <IRQ>
[22766.387150]  [<ffffffffa169f292>] ? __sctp_lookup_association+0x62/0xd0
[sctp]
[22766.387154]  [<ffffffffa169f2b6>] __sctp_lookup_association+0x86/0xd0 [sctp]
[22766.387157]  [<ffffffffa169f597>] sctp_rcv+0x207/0xbb0 [sctp]
[22766.387161]  [<ffffffff810d4da8>] ? trace_hardirqs_off_caller+0x28/0xd0
[22766.387163]  [<ffffffff815827e3>] ? nf_hook_slow+0x133/0x210
[22766.387166]  [<ffffffff815902fc>] ? ip_local_deliver_finish+0x4c/0x4c0
[22766.387168]  [<ffffffff8159043d>] ip_local_deliver_finish+0x18d/0x4c0
[22766.387169]  [<ffffffff815902fc>] ? ip_local_deliver_finish+0x4c/0x4c0
[22766.387171]  [<ffffffff81590a07>] ip_local_deliver+0x47/0x80
[22766.387172]  [<ffffffff8158fd80>] ip_rcv_finish+0x150/0x680
[22766.387174]  [<ffffffff81590c54>] ip_rcv+0x214/0x320
[22766.387176]  [<ffffffff81558c07>] __netif_receive_skb+0x7b7/0x910
[22766.387178]  [<ffffffff8155856c>] ? __netif_receive_skb+0x11c/0x910
[22766.387180]  [<ffffffff810d423e>] ? put_lock_stats.isra.25+0xe/0x40
[22766.387182]  [<ffffffff81558f83>] netif_receive_skb+0x23/0x1f0
[22766.387183]  [<ffffffff815596a9>] ? dev_gro_receive+0x139/0x440
[22766.387185]  [<ffffffff81559280>] napi_skb_finish+0x70/0xa0
[22766.387187]  [<ffffffff81559cb5>] napi_gro_receive+0xf5/0x130
[22766.387218]  [<ffffffffa01c4679>] e1000_receive_skb+0x59/0x70 [e1000e]
[22766.387242]  [<ffffffffa01c5aab>] e1000_clean_rx_irq+0x28b/0x460 [e1000e]
[22766.387266]  [<ffffffffa01c9c18>] e1000e_poll+0x78/0x430 [e1000e]
[22766.387268]  [<ffffffff81559fea>] net_rx_action+0x1aa/0x3d0
[22766.387270]  [<ffffffff810a495f>] ? account_system_vtime+0x10f/0x130
[22766.387273]  [<ffffffff810734d0>] __do_softirq+0xe0/0x420
[22766.387275]  [<ffffffff8169826c>] call_softirq+0x1c/0x30
[22766.387278]  [<ffffffff8101db15>] do_softirq+0xd5/0x110
[22766.387279]  [<ffffffff81073bc5>] irq_exit+0xd5/0xe0
[22766.387281]  [<ffffffff81698b03>] do_IRQ+0x63/0xd0
[22766.387283]  [<ffffffff8168ee2f>] common_interrupt+0x6f/0x6f
[22766.387283]  <EOI>
[22766.387284]
[22766.387285]  [<ffffffff8168eed9>] ? retint_swapgs+0x13/0x1b
[22766.387285] Code: c0 90 5d c3 66 0f 1f 44 00 00 4c 89 c8 5d c3 0f 1f 00 55 48
89 e5 48 83
ec 20 48 89 5d e8 4c 89 65 f0 4c 89 6d f8 66 66 66 66 90 <0f> b7 87 98 00 00 00
48 89 fb
49 89 f5 66 c1 c0 08 66 39 46 02
[22766.387307]
[22766.387307] RIP
[22766.387311]  [<ffffffffa168a2c9>] sctp_assoc_is_match+0x19/0x90 [sctp]
[22766.387311]  RSP <ffff880147c039b0>
[22766.387142]  ffffffffa16ab120
[22766.599537] ---[ end trace 3f6dae82e37b17f5 ]---
[22766.601221] Kernel panic - not syncing: Fatal exception in interrupt

It appears from his analysis and some staring at the code that this is likely
occuring because an association is getting freed while still on the
sctp_assoc_hashtable.  As a result, we get a gpf when traversing the hashtable
while a freed node corrupts part of the list.

Nominally I would think that an mibalanced refcount was responsible for this,
but I can't seem to find any obvious imbalance.  What I did note however was
that the two places where we create an association using
sctp_primitive_ASSOCIATE (__sctp_connect and sctp_sendmsg), have failure paths
which free a newly created association after calling sctp_primitive_ASSOCIATE.
sctp_primitive_ASSOCIATE brings us into the sctp_sf_do_prm_asoc path, which
issues a SCTP_CMD_NEW_ASOC side effect, which in turn adds a new association to
the aforementioned hash table.  the sctp command interpreter that process side
effects has not way to unwind previously processed commands, so freeing the
association from the __sctp_connect or sctp_sendmsg error path would lead to a
freed association remaining on this hash table.

I've fixed this but modifying sctp_[un]hash_established to use hlist_del_init,
which allows us to proerly use hlist_unhashed to check if the node is on a
hashlist safely during a delete.  That in turn alows us to safely call
sctp_unhash_established in the __sctp_connect and sctp_sendmsg error paths
before freeing them, regardles of what the associations state is on the hash
list.

I noted, while I was doing this, that the __sctp_unhash_endpoint was using
hlist_unhsashed in a simmilar fashion, but never nullified any removed nodes
pointers to make that function work properly, so I fixed that up in a simmilar
fashion.

I attempted to test this using a virtual guest running the SCTP_RR test from
netperf in a loop while running the trinity fuzzer, both in a loop.  I wasn't
able to recreate the problem prior to this fix, nor was I able to trigger the
failure after (neither of which I suppose is suprising).  Given the trace above
however, I think its likely that this is what we hit.

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Reported-by: davej@redhat.com
CC: davej@redhat.com
CC: "David S. Miller" <davem@davemloft.net>
CC: Vlad Yasevich <vyasevich@gmail.com>
CC: Sridhar Samudrala <sri@us.ibm.com>
CC: linux-sctp@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
net/sctp/input.c
net/sctp/socket.c

index 80564fe03024634dc479280a8fe8d60ea643e1cc..8b9b6790a3dfc60b562a3a6ea30dccf248612cc4 100644 (file)
@@ -736,15 +736,12 @@ static void __sctp_unhash_endpoint(struct sctp_endpoint *ep)
 
        epb = &ep->base;
 
-       if (hlist_unhashed(&epb->node))
-               return;
-
        epb->hashent = sctp_ep_hashfn(epb->bind_addr.port);
 
        head = &sctp_ep_hashtable[epb->hashent];
 
        sctp_write_lock(&head->lock);
-       __hlist_del(&epb->node);
+       hlist_del_init(&epb->node);
        sctp_write_unlock(&head->lock);
 }
 
@@ -825,7 +822,7 @@ static void __sctp_unhash_established(struct sctp_association *asoc)
        head = &sctp_assoc_hashtable[epb->hashent];
 
        sctp_write_lock(&head->lock);
-       __hlist_del(&epb->node);
+       hlist_del_init(&epb->node);
        sctp_write_unlock(&head->lock);
 }
 
index b3b8a8d813eb663f18a6fd488d3f0472f2fe1263..31c7bfcd9b5872aa136e2b513123bb6d582bdff6 100644 (file)
@@ -1231,8 +1231,14 @@ out_free:
        SCTP_DEBUG_PRINTK("About to exit __sctp_connect() free asoc: %p"
                          " kaddrs: %p err: %d\n",
                          asoc, kaddrs, err);
-       if (asoc)
+       if (asoc) {
+               /* sctp_primitive_ASSOCIATE may have added this association
+                * To the hash table, try to unhash it, just in case, its a noop
+                * if it wasn't hashed so we're safe
+                */
+               sctp_unhash_established(asoc);
                sctp_association_free(asoc);
+       }
        return err;
 }
 
@@ -1942,8 +1948,10 @@ SCTP_STATIC int sctp_sendmsg(struct kiocb *iocb, struct sock *sk,
        goto out_unlock;
 
 out_free:
-       if (new_asoc)
+       if (new_asoc) {
+               sctp_unhash_established(asoc);
                sctp_association_free(asoc);
+       }
 out_unlock:
        sctp_release_sock(sk);