Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 22 Jul 2012 17:31:37 +0000 (UTC)
From:      Robert Watson <rwatson@FreeBSD.org>
To:        src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org
Subject:   svn commit: r238699 - head/sys/netinet
Message-ID:  <201207221731.q6MHVbxN009024@svn.freebsd.org>

next in thread | raw e-mail | index | archive | help
Author: rwatson
Date: Sun Jul 22 17:31:36 2012
New Revision: 238699
URL: http://svn.freebsd.org/changeset/base/238699

Log:
  Update some stale comments regarding tcbinfo locking in the TCP input
  path: read locks on tcbinfo are no longer used, so won't happen.  No
  functional change.
  
  MFC after:	3 days

Modified:
  head/sys/netinet/tcp_input.c

Modified: head/sys/netinet/tcp_input.c
==============================================================================
--- head/sys/netinet/tcp_input.c	Sun Jul 22 16:56:59 2012	(r238698)
+++ head/sys/netinet/tcp_input.c	Sun Jul 22 17:31:36 2012	(r238699)
@@ -909,7 +909,7 @@ findpcb:
 	/*
 	 * A previous connection in TIMEWAIT state is supposed to catch stray
 	 * or duplicate segments arriving late.  If this segment was a
-	 * legitimate new connection attempt the old INPCB gets removed and
+	 * legitimate new connection attempt, the old INPCB gets removed and
 	 * we can try again to find a listening socket.
 	 *
 	 * At this point, due to earlier optimism, we may hold only an inpcb
@@ -1438,15 +1438,8 @@ tcp_do_segment(struct mbuf *m, struct tc
 	/*
 	 * If this is either a state-changing packet or current state isn't
 	 * established, we require a write lock on tcbinfo.  Otherwise, we
-	 * allow either a read lock or a write lock, as we may have acquired
-	 * a write lock due to a race.
-	 *
-	 * Require a global write lock for SYN/FIN/RST segments or
-	 * non-established connections; otherwise accept either a read or
-	 * write lock, as we may have conservatively acquired a write lock in
-	 * certain cases in tcp_input() (is this still true?).  Currently we
-	 * will never enter with no lock, so we try to drop it quickly in the
-	 * common pure ack/pure data cases.
+	 * allow the tcbinfo to be in either alocked or unlocked, as the
+	 * caller may have unnecessarily acquired a write lock due to a race.
 	 */
 	if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
 	    tp->t_state != TCPS_ESTABLISHED) {



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201207221731.q6MHVbxN009024>