commit 4ab6ad136b62add35598f3c354231701b7c14c65
author Klement Sekera <ksekera@cisco.com> Fri Dec 17 12:47:51 2021 +0000
committer Ole Tr�an <otroan@employees.org> Mon Dec 20 10:05:20 2021 +0000
tree 5cff25e84550e4b7855774e7213efd61bfe7a5cc
parent e3241f04410fa45099f958f77636b80293904fc3

ip: SVR fix race condition

There could be a race condition where two fragments of one chain end up
at the same time on different workers, one overwriting others hash
entry. Add a check for that and restart processing on the unlucky worker
who ends up being second from hash table POV. This will then result in a
proper handover to worker now owning this reassembly.

Type: fix
Fixes: de34c35fc73226943538149fae9dbc5cfbdc6e75
Signed-off-by: Klement Sekera <ksekera@cisco.com>
Change-Id: I9eb29c5cb1ffe3b5eb1d5a638e17ab7ba2628d28

src/vnet/ip/reass/ip4_sv_reass.c

diff --git a/src/vnet/ip/reass/ip4_sv_reass.c b/src/vnet/ip/reass/ip4_sv_reass.c
index f5fa575..9971daf 100644
--- a/src/vnet/ip/reass/ip4_sv_reass.c
+++ b/src/vnet/ip/reass/ip4_sv_reass.c
@@ -321,6 +321,8 @@
   ip4_sv_reass_t *reass = NULL;
   f64 now = vlib_time_now (vm);
 
+again:
+
   if (!clib_bihash_search_16_8 (&rm->hash, &kv->kv, &kv->kv))
     {
       if (vm->thread_index != kv->v.thread_index)
@@ -375,10 +377,14 @@
   kv->v.thread_index = vm->thread_index;
   reass->last_heard = now;
 
-  if (clib_bihash_add_del_16_8 (&rm->hash, &kv->kv, 1))
+  int rv = clib_bihash_add_del_16_8 (&rm->hash, &kv->kv, 2);
+  if (rv)
     {
       ip4_sv_reass_free (vm, rm, rt, reass);
       reass = NULL;
+      // if other worker created a context already work with the other copy
+      if (-2 == rv)
+	goto again;
     }
 
   return reass;

src/vnet/ip/reass/ip6_sv_reass.c

diff --git a/src/vnet/ip/reass/ip6_sv_reass.c b/src/vnet/ip/reass/ip6_sv_reass.c
index 58c7d8d..23ae678 100644
--- a/src/vnet/ip/reass/ip6_sv_reass.c
+++ b/src/vnet/ip/reass/ip6_sv_reass.c
@@ -311,6 +311,8 @@
   ip6_sv_reass_t *reass = NULL;
   f64 now = vlib_time_now (vm);
 
+again:
+
   if (!clib_bihash_search_48_8 (&rm->hash, &kv->kv, &kv->kv))
     {
       if (vm->thread_index != kv->v.thread_index)
@@ -370,10 +372,14 @@
   kv->v.thread_index = vm->thread_index;
   reass->last_heard = now;
 
-  if (clib_bihash_add_del_48_8 (&rm->hash, &kv->kv, 1))
+  int rv = clib_bihash_add_del_48_8 (&rm->hash, &kv->kv, 2);
+  if (rv)
     {
       ip6_sv_reass_free (vm, rm, rt, reass);
       reass = NULL;
+      // if other worker created a context already work with the other copy
+      if (-2 == rv)
+	goto again;
     }
 
   return reass;