Re: kjournald() with DIO

[Mingming Cao <cmm@us.ibm.com>]

On Fri, 2005-09-16 at 14:42 +0100, Stephen C. Tweedie wrote: 
> Hi,
> 
> On Tue, 2005-09-13 at 01:29, Andrew Morton wrote:
> 
> > Have you spoken with Mingming about this?  She's chasing a similar race. 
> > She's currently working on getting the ext3-debug patch working so we can
> > find out how those buffers (which apparently aren't even attached to the
> > journal) came to have an elevated refcount.
> 
> Here's the latest version of ext3 buffer tracing which I've been using. 
> It's basically just the same as one from Andrew from about 2.6.5
> vintage, with a few changes in places to deal with new bits of code that
> need temporary buffer_heads initialised properly for tracing.
> 
> It has survived a couple of hours of fsx and fsstress in parallel, but I
> haven't run any O_DIRECT tests with it on the current git kernel, so
> there may stil be problems in that area.

Your patch works smoothly running iozone DIO tests overnight.  It also
successfully shows the last 32 activities history of the busy buffer
which caused the EIO failure here, and the results matches what the
previous analysis.

A minor improvement, to translate BJ_Locked state into a proper string.

Signed-off-by: Mingming Cao cmm@us.ibm.com
---

 linux-2.6.5-7.191-cmm/fs/jbd-kernel.c |    1 +
 1 files changed, 1 insertion(+)

diff -puN fs/jbd-kernel.c~buffer-trace-fix fs/jbd-kernel.c
--- linux-2.6.5-7.191/fs/jbd-kernel.c~buffer-trace-fix	2005-09-19 15:48:03.000000000 -0700
+++ linux-2.6.5-7.191-cmm/fs/jbd-kernel.c	2005-09-19 15:50:16.000000000 -0700
@@ -163,6 +163,7 @@ static const char *b_jlist_to_string(uns
 	case BJ_Shadow:		return "BJ_Shadow";
 	case BJ_LogCtl:		return "BJ_LogCtl";
 	case BJ_Reserved:	return "BJ_Reserved";
+	case BJ_Locked:		return "BJ_Locked";
 #endif
 	default:		return "Bad b_jlist";
 	}

_