Break out tcp_output() spinning problem from socket buffer corruption in

tcp_output(): the latter is fixed and merged, but the former remains a
problem.  Note that it is actively being worked on.
This commit is contained in:
Robert Watson 2004-10-26 09:50:03 +00:00
parent 0107d20850
commit 547ea0ede1
Notes: svn2git 2020-12-08 03:00:23 +00:00
svn path=/www/; revision=22685

View file

@ -1,7 +1,7 @@
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" [
<!ENTITY base CDATA "../..">
<!ENTITY email 'freebsd-qa'>
<!ENTITY date "$FreeBSD: www/en/releases/5.3R/todo.sgml,v 1.103 2004/10/22 10:14:43 rwatson Exp $">
<!ENTITY date "$FreeBSD: www/en/releases/5.3R/todo.sgml,v 1.104 2004/10/26 09:45:36 rwatson Exp $">
<!ENTITY title "FreeBSD 5.3 Open Issues">
<!ENTITY % includes SYSTEM "../../includes.sgml"> %includes;
<!ENTITY % developers SYSTEM "../../developers.sgml"> %developers;
@ -62,6 +62,22 @@
The patch will be merged to 6.x and 5.x shortly.</td>
</tr>
<tr>
<td>Reports of TCP-related instability under extremely high load;
possibly related to SACK</td>
<td>&status.wip;</td>
<td>&a.gnn;, &a.rwatson;, &a.scottl</td>
<td>There have been reports that, under extremely high load, the
tcp_output() routine may appear to run for extended periods, resulting
in the appearance of a hang for an extended period (up to 30 minutes),
followed by recovery. This may be a result of a bug in the TCP
selective acknowledgement implementation introduced following 5.2;
the release engineering team is currently working with the submitters
to diagnose the problem. Depending on the nature of the problem, it
may be appropriate to release with SACK disabled, or to correct the
bug prior to 5.3.</td>
</tr>
</table>
<h3>Required features for 5.3-RELEASE</h3>
@ -539,15 +555,13 @@
</tr>
<tr>
<td>Reports of tcp_output() spinning; socket buffer corruption</td>
<td>&status.untested;</td>
<td>Reports of socket buffer corruption in tcp_output()</td>
<td>&status.done;</td>
<td>&a.rwatson;</td>
<td>There have been reports of occasional corruption of socket buffers
and/or spinning in tcp_output(). This may have been the result of
missing socket buffer locking in tcp_output(), which has now been
corrected in 6.x and 5.x; the release engineering team is awaiting
word as to whether this fixes the problems experienced by the
submitter.</td>
<td>There have been reports of occasional corruption of socket buffers.
This may have been the result of missing socket buffer locking in
tcp_output(), which has now been corrected in 6.x and 5.x; this
problem is believed to be fixed.</td>
</tr>
<tr>