Submitter : lord *Status : closed
Assigned Engineer : lord *Fixed By : ananth
*Fixed By Domain : engr *Closed Date : 08/31/00
Priority : 1 *Modified Date : 08/31/00
*Modified User : ananth *Modified User Domain : engr
*Fix Description :
From: steve lord <lord@xxxxxxx> (PARTIAL)
Date: Aug 31 2000 11:16:49AM
[pvnews version: 1.71]
----------------------------
At least one more to go - which may be restricted to the kiocluster
mount option.
Date: Wed Aug 30 14:54:03 PDT 2000
Workarea: jen.americas.sgi.com:/src/lord/xfs-linux.2.4.0-test5
.....
==========================
ADDITIONAL INFORMATION (TAKE)
From: ananth ananthanarayanan <ananth@waco>
Date: Aug 31 2000 05:35:03PM
[pvnews version: 1.71]
==========================
This fixes known corruption problems with
respect to large cp's and rpm builds.
Also tested with doio 1- and 2-threads.
Date: Thu Aug 31 17:28:57 PDT 2000
Workarea: waco.engr.sgi.com:/build1/ananth/xfs
The following file(s) were checked into:
bonnie.engr.sgi.com:/isms/slinx/2.4.0-test1-xfs
Modid: 2.4.0-test1-xfs:slinx:73466a
linux/fs/pagebuf/page_buf.c - 1.26
- Fix a problem where kiobuf I/O assumptions were being violated;
that is, a READ with some valid pages will erase recent
data contained only in the in-memory version of the valid pages.
Also support clustered pages which are locked on lookup.
linux/fs/pagebuf/page_buf_io.c - 1.27
- Fix a bug where if the clustering was stopped because of
reaching the tuneable count, the last selected pages was
left out of the I/O. Also, support lock pages to be clustered
on lookup, closing holes between lookup and start of I/O.
Description :
The glibc RPM build is once again failing in an XFS filesystem,
not much more information than this is available at the moment,
The location of the failure appears to be semi-random.
==========================
ADDITIONAL INFORMATION (UPDATE)
From: lord@xxxxxxx (BugWorks)
Date: Aug 30 2000 01:24:42PM
==========================
.....
==========================
ADDITIONAL INFORMATION (REASSIGN)
From: lord@xxxxxxx (BugWorks)
Date: Aug 30 2000 02:46:29PM
==========================
Reassign, managers don't do PV's.
Also, we appear to be making some progress, first there is
more than one bug here. The write full page function is broken,
so dirty data is not always swapped out correctly. A fix for
this will be along shortly.
Secondly, a copy of a file several times bigger than system
memory can generate corruption within the file which appears
consistent with some data being discarded before being written
to disk. In this case kiocluster appears to be necessary to
cause corruption.
|