infinite loops with scallion and tor-0.2.4.x

[robgjansen]

As a result of ticket 9731, Tor changed the functions connection_consider_empty_write_buckets and connection_consider_empty_read_buckets, which caused inifinite loops in shadow. I believe the infinite loops were fixed in the following commits:

https://gitweb.torproject.org/tor.git/commitdiff/96f92f2062118b61a43134170b92001d3be1c128?hp=702c0502cf3121e2e698963cf53adb00ae2d136d
https://gitweb.torproject.org/tor.git/commitdiff/8f793c38fbcd885d40adf8312ef1767e172e1be4?hp=96f92f2062118b61a43134170b92001d3be1c128

The fix on Tor's end has not been backported into 0.2.4.x yet. So the workaround is making the changes above manually or using tor-0.2.5.x. I opened Tor ticket 11638 requesting a fix on 0.2.4.

[robgjansen]

This seems to work correctly using the following

shadow

• master branch at commit 3acc9c2

shadow-plugin-tor

• master branch at commit shadow/shadow-plugin-tor@01e9701

Tor

• Merge branch bug11638_024 of Nick's repo (https://git.torproject.org/nickm/tor.git) with branch maint-0.2.4 from the main Tor repo (https://git.torproject.org/tor.git) (currently at commit b45f0f8fb9b11effcd48159c078f09a3fbc04b11).

[robgjansen]

I ran 10 simulations with the above code and the 400 relay config here as well as a 3600 relay experiment, and they all completed successfully.

[robgjansen]

Tor will not be merging Nick's fix into 0.2.4.x. Here is the diff of the fix against the latest stable tor-0.2.4.23. If you run into this infinite loop problem, please either merge Nick's branch as described above, or merge this patch.

diff --git a/changes/bug9731_full_backport b/changes/bug9731_full_backport
new file mode 100644
index 0000000..16806eb
--- /dev/null
+++ b/changes/bug9731_full_backport
@@ -0,0 +1,7 @@
+  o Minor bugfixes (backport from 0.2.5.1-alpha):
+    - No longer apply rate-limiting to any non-rate-limited connection.  Our
+      earlier fix in 0.2.4.18-rc applied later in the calculation, applied
+      only to CPUWorker connections, and led to infinite loops with the
+      Shadow network simulation tool. Fixes bug 11638, and backports an
+      improved fix for bug 9731. Bugfix on 0.2.4.18-rc.
+
diff --git a/src/or/connection.c b/src/or/connection.c
index 4f74a1d..275d562 100644
--- a/src/or/connection.c
+++ b/src/or/connection.c
@@ -2508,6 +2508,9 @@ connection_consider_empty_read_buckets(connection_t *conn)
 {
   const char *reason;

+  if (!connection_is_rate_limited(conn))
+    return; /* Always okay. */
+
   if (global_read_bucket <= 0) {
     reason = "global read bucket exhausted. Pausing.";
   } else if (connection_counts_as_relayed_traffic(conn, approx_time()) &&
@@ -2520,9 +2523,6 @@ connection_consider_empty_read_buckets(connection_t *conn)
   } else
     return; /* all good, no need to stop it */

-  if (conn->type == CONN_TYPE_CPUWORKER)
-    return; /* Always okay. */
-
   LOG_FN_CONN(conn, (LOG_DEBUG, LD_NET, "%s", reason));
   conn->read_blocked_on_bw = 1;
   connection_stop_reading(conn);
@@ -2535,6 +2535,9 @@ connection_consider_empty_write_buckets(connection_t *conn)
 {
   const char *reason;

+  if (!connection_is_rate_limited(conn))
+    return; /* Always okay. */
+
   if (global_write_bucket <= 0) {
     reason = "global write bucket exhausted. Pausing.";
   } else if (connection_counts_as_relayed_traffic(conn, approx_time()) &&
@@ -2547,9 +2550,6 @@ connection_consider_empty_write_buckets(connection_t *conn)
   } else
     return; /* all good, no need to stop it */

-  if (conn->type == CONN_TYPE_CPUWORKER)
-    return; /* Always okay. */
-
   LOG_FN_CONN(conn, (LOG_DEBUG, LD_NET, "%s", reason));
   conn->write_blocked_on_bw = 1;
   connection_stop_writing(conn);