[asterisk-commits] russell: trunk r152877 - /trunk/channels/chan_sip.c

SVN commits to the Asterisk project asterisk-commits at lists.digium.com
Thu Oct 30 14:21:53 CDT 2008


Author: russell
Date: Thu Oct 30 14:21:53 2008
New Revision: 152877

URL: http://svn.digium.com/view/asterisk?view=rev&rev=152877
Log:
Modify the documentation of the sip_registry struct
 - Remove a comment that says that the monitor thread is the only one that
   ever touches these objects.  This is no longer the case with TCP.  Also,
   I would eventually like to get the scheduler in its own thread, so this
   is just a poor assumption to make.
 - Note that reference counting of these objects with respect to scheduler
   entries is not complete.  There are some leaked references when deleting
   scheduler entries.

Modified:
    trunk/channels/chan_sip.c

Modified: trunk/channels/chan_sip.c
URL: http://svn.digium.com/view/asterisk/trunk/channels/chan_sip.c?view=diff&rev=152877&r1=152876&r2=152877
==============================================================================
--- trunk/channels/chan_sip.c (original)
+++ trunk/channels/chan_sip.c Thu Oct 30 14:21:53 2008
@@ -1657,7 +1657,9 @@
 };
 
 
-/*! \brief Registrations with other SIP proxies
+/*! 
+ * \brief Registrations with other SIP proxies
+ *
  * Created by sip_register(), the entry is linked in the 'regl' list,
  * and never deleted (other than at 'sip reload' or module unload times).
  * The entry always has a pending timeout, either waiting for an ACK to
@@ -1666,11 +1668,12 @@
  * or once the previously completed registration one expires).
  * The registration can be in one of many states, though at the moment
  * the handling is a bit mixed.
- * Note that the entire evolution of sip_registry (transmissions,
- * incoming packets and timeouts) is driven by one single thread,
- * do_monitor(), so there is almost no synchronization issue.
- * The only exception  is the sip_pvt creation/lookup,
- * as the dialoglist is also manipulated by other threads.
+ *
+ * XXX \todo Reference count handling for this object has some problems with
+ * respect to scheduler entries.  The ref count is handled in some places,
+ * but not all of them.  There are some places where references get leaked
+ * when this scheduler entry gets cancelled.  At worst, this would cause
+ * memory leaks on reloads if registrations get removed from configuration.
  */
 struct sip_registry {
 	ASTOBJ_COMPONENTS_FULL(struct sip_registry,1,1);
@@ -4756,7 +4759,6 @@
 	ast_atomic_fetchadd_int(&regobjs, -1);
 	ast_dnsmgr_release(reg->dnsmgr);
 	ast_free(reg);
-	
 }
 
 /*! \brief Destroy MWI subscription object */




More information about the asterisk-commits mailing list