commit a6bae93bf835039b8ce31cf393f49e72029db76f
author David Benjamin <davidben@google.com> Sat Jul 01 01:08:14 2017 -0400
committer David Benjamin <davidben@google.com> Wed Jul 05 20:32:47 2017 +0000
tree 78b286809ec052dcb9e18fab6b27f4a23224d088
parent c94998ae9511add69918821f5700dab8da9207c0

Never set not_resumable on an immutable session.

Once passed to the outside world, an SSL_SESSION is immutable. It is not
thread-safe to set not_resumable. In most cases, the session is already
expired anyway. In other cases, making all this remove session be unlink rather than
destroy is sound and consistent with how we treat sessions elsewhere.

In particular, SSL_CTX_free calls SSL_CTX_flush_sessions(0), and
bulk-invalidating everything like this is interfering with some
follow-up changes to improve the fuzzer.

Change-Id: I2a19b8ce32d9effc1efaa72e934e015ebbbfbf9a
Reviewed-on: https://boringssl-review.googlesource.com/17530
Reviewed-by: David Benjamin <davidben@google.com>

ssl/ssl_session.c

diff --git a/ssl/ssl_session.c b/ssl/ssl_session.c
index 1670eb7..3e2c9f4 100644
--- a/ssl/ssl_session.c
+++ b/ssl/ssl_session.c
@@ -959,7 +959,6 @@
     }
 
     if (ret) {
-      found_session->not_resumable = 1;
       if (ctx->remove_session_cb != NULL) {
         ctx->remove_session_cb(ctx, found_session);
       }
@@ -1038,7 +1037,6 @@
      * save on locking overhead */
     (void) lh_SSL_SESSION_delete(param->cache, session);
     SSL_SESSION_list_remove(param->ctx, session);
-    session->not_resumable = 1;
     if (param->ctx->remove_session_cb != NULL) {
       param->ctx->remove_session_cb(param->ctx, session);
     }