You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@trafficserver.apache.org by zw...@apache.org on 2018/03/06 18:16:20 UTC

[trafficserver] 02/04: Prevent potential dispatch of HttpTunnel events to a dead consumer.

This is an automated email from the ASF dual-hosted git repository.

zwoop pushed a commit to branch 7.1.x
in repository https://gitbox.apache.org/repos/asf/trafficserver.git

commit dd51074bc5a9c0a15712e2030935473b3d4b4885
Author: Alan M. Carroll <so...@yahoo-inc.com>
AuthorDate: Fri Jun 2 13:07:36 2017 -0500

    Prevent potential dispatch of HttpTunnel events to a dead consumer.
    
    (cherry picked from commit 6f0bc840b2cbfcbead88c4a411dc719c4f9d6f25)
---
 proxy/http/HttpTunnel.h | 30 +++++++++++++++++++++++++-----
 1 file changed, 25 insertions(+), 5 deletions(-)

diff --git a/proxy/http/HttpTunnel.h b/proxy/http/HttpTunnel.h
index 42c6fd4..54c83f0 100644
--- a/proxy/http/HttpTunnel.h
+++ b/proxy/http/HttpTunnel.h
@@ -446,12 +446,32 @@ HttpTunnel::get_producer(HttpTunnelType_t type)
 inline HttpTunnelConsumer *
 HttpTunnel::get_consumer(VConnection *vc)
 {
-  for (int i = 0; i < MAX_CONSUMERS; i++) {
-    if (consumers[i].vc == vc) {
-      return consumers + i;
+  /** Rare but persistent problem in which a @c INKVConnInternal is used by a consumer, released,
+      and then re-allocated for a different consumer. This causes two consumers to have the same VC
+      pointer resulting in this method returning the wrong consumer. Note this is a not a bad use of
+      the tunnel, but an unfortunate interaction with the FIFO free lists.
+
+      It's not correct to check for the consumer being alive - at a minimum `HTTP_TUNNEL_EVENT_DONE`
+      is dispatched against a consumer after the consumer is not alive. Instead if a non-alive
+      consumer matches it is stored as a candidate and returned if no other match is found. If a
+      live matching consumer is found, it is immediately returned. It is never valid to have the
+      same VC in more than one active consumer. This should avoid a performance impact because in
+      the usual case the consumer will be alive.
+
+      In the case of a deliberate dispatch of an event to a dead consumer that has a duplicate vc
+      address, this will select the last consumer which will be correct as the consumers are added
+      in order therefore the latter consumer will be the most recent / appropriate target.
+  */
+  HttpTunnelConsumer *zret = nullptr;
+  for (HttpTunnelConsumer &c : consumers) {
+    if (c.vc == vc) {
+      zret = &c;
+      if (c.alive) { // a match that's alive is always the best.
+        break;
+      }
     }
   }
-  return NULL;
+  return zret;
 }
 
 inline HttpTunnelProducer *
@@ -470,7 +490,7 @@ HttpTunnel::get_consumer(VIO *vio)
 {
   if (vio) {
     for (int i = 0; i < MAX_CONSUMERS; i++) {
-      if (consumers[i].write_vio == vio || consumers[i].vc == vio->vc_server) {
+      if (consumers[i].alive && (consumers[i].write_vio == vio || consumers[i].vc == vio->vc_server)) {
         return consumers + i;
       }
     }

-- 
To stop receiving notification emails like this one, please contact
zwoop@apache.org.