r15764@catbus: nickm | 2007-10-14 04:43:10 -0400

Backport r11829: Downgrade warning that caused bug 463; comment; resolve.


svn:r11929

ChangeLog

@@ -40,6 +40,9 @@ Changes in version 0.1.2.18 - 2007-??-??
       key from getting rotated.
     - On some platforms, accept() can return a broken address.  Detect
       this more quietly, and deal accordingly.  (Fixes bug 483.)
+    - It's not actually an error to find a non-pending entry in the DNS
+      cache when canceling a pending resolve.  Don't log unless stuff
+      is fishy.  Resolves bug 463.
 
 
 Changes in version 0.1.2.17 - 2007-08-30

doc/TODO.012

@@ -14,6 +14,6 @@ P - r10579: new addsysuser implementation for osx (??)
     (this will break some existing test-network configurations, yes?)
   o r11499, r11500, r11501: hidserv hexdigests rather than nicknames
 P - r11548, the osx /tmp fix
-N - r11829: Don't warn when cancel_pending_resolve() finds a cached failure.
+  o r11829: Don't warn when cancel_pending_resolve() finds a cached failure.
 R - r11915: just because you hup, don't publish a near-duplicate descriptor
 

src/or/dns.c

@@ -839,9 +839,19 @@ dns_cancel_pending_resolve(const char *address)
   strlcpy(search.address, address, sizeof(search.address));
 
   resolve = HT_FIND(cache_map, &cache_root, &search);
-  if (!resolve || resolve->state != CACHE_STATE_PENDING) {
-    log_notice(LD_BUG,"Address %s is not pending. Dropping.",
+  if (!resolve)
+    return;
+
+  if (resolve->state != CACHE_STATE_PENDING) {
+    /* We can get into this state if we never actually created the pending
+     * resolve, due to finding an earlier cached error or something.  Just
+     * ignore it. */
+    if (resolve->pending_connections) {
+      log_warn(LD_BUG,
+               "Address %s is not pending but has pending connections!",
                escaped_safe_str(address));
+      tor_fragile_assert();
+    }
     return;
   }