xfs
[Top] [All Lists]

Re: [PATCH] Fix reference counting race on log buffers

To: Christoph Hellwig <hch@xxxxxxxxxxxxx>
Subject: Re: [PATCH] Fix reference counting race on log buffers
From: Lachlan McIlroy <lachlan@xxxxxxx>
Date: Fri, 15 Aug 2008 13:21:02 +1000
Cc: Dave Chinner <david@xxxxxxxxxxxxx>, xfs@xxxxxxxxxxx
In-reply-to: <20080815012807.GA547@xxxxxxxxxxxxx>
References: <1215752481-6862-1-git-send-email-david@xxxxxxxxxxxxx> <20080814180603.GA3087@xxxxxxxxxxxxx> <48A4DB09.7020702@xxxxxxx> <20080815012807.GA547@xxxxxxxxxxxxx>
Reply-to: lachlan@xxxxxxx
Sender: xfs-bounce@xxxxxxxxxxx
User-agent: Thunderbird 2.0.0.16 (X11/20080707)
Christoph Hellwig wrote:
On Fri, Aug 15, 2008 at 11:25:29AM +1000, Lachlan McIlroy wrote:
Christoph Hellwig wrote:
Btw, while this patch made it to mainline it never went into the oss
cvs tree (and thus probably the ptools tree).

There also are a lot of other difference between the trees, I'll see
how to get them back into sync..
Yeah, we're aware of these differences and we're trying to sync them up.
If you identify any specific commits from mainline (other than this log
race fix) that aren't in the cvs tree please let us know.

I think I have most of them covered:

 - log reace
 - the endianess helpers patch I sent a note out about
 - the mnt_want_write patches I've sent today
 - just yesterday the creds preparation in mainline replaced some
   XFS current_* macros with generic ones, I'll send a patch for these
   once the changes in mainline have settled a little

The long term differences are:

 - a different xfs_version.h - IMHO we should just kill this versioning
   scheme completely..
 - xfsidbg.  Maybe now with kgdb in the kgdb folks get something
   submitted and then this is taken care off..
That stuff is there for kdb.  Are you suggesting we switch to kgdb and
drop this file?

 - modular quota.  IMHO we should just stop doing this in CVS, it
   doesn't really buy us anything.  If people are okay with that
   I'll send a patch
I don't see any problem with this.  Seems like a good idea.

 - dmapi - well, this won't get merged and currently I don't see anyone
   doing HSM for mainline properly.  But I have some ideas how to make
   it less intrusive at least
All that would be very helpful.


Other than that there are some difference dueto a different base kernel
version, which are temporary until the XFS tree is merged up to kernel
including those.

Last time I checked the fs/xfs differences between mainline and our
internal tree resulted in a patch around 470KB.  Most of that is changes
that we don't push to mainline such as fs/xfs/dmapi and fs/xfs/xfsidbg.c.
The rest I'm working though.


<Prev in Thread] Current Thread [Next in Thread>