2004-01-01 22:01:22

by Theodore Ts'o

[permalink] [raw]
Subject: [PATCH] [2.4.24-pre3] 0/5 EXT2/3 Updates

Hi Marcelo,

I would appreciate it if you would consider applying these patches for
2.4.24:

Patch 1: Fix VM overcommit accounting of truncated pages pinned by ext3
Patch 2: Fix Compatibility problem with 2.6 SELinux users who try to
boot 2.4 kernels (from sct)
Patch 3: Add forward compatibility for on-line resizing
Patch 4: Add forward compatibility for expanded inodes
Patch 5: Add HTREE indexed directory support

Most of these patches have been tested for quite a while as independent
patches against 2.4, as well as in Linux 2.6.0. Patches 2, 3, and 4 add
better forward compatibility for the EXT2 and EXT3 filesystems. Patch
#5 is the largests, and adds the HTREE support. This is a roll-up patch
that includes all of the various bugfixes that are in Linux 2.6.0.

(Note: There are some additional 2.4 ext2/3 patches including those for
the Orlov allocator, that are not included in this patch set; the ones
included here are the ones which I believe are most critical for
inclusion in the 2.4 mainline.)


- Ted


2004-01-04 00:07:49

by Frank van Maarseveen

[permalink] [raw]
Subject: Re: [PATCH] [2.4.24-pre3] 0/5 EXT2/3 Updates

On Thu, Jan 01, 2004 at 04:50:20PM -0500, Theodore Ts'o wrote:
> Patch 5: Add HTREE indexed directory support

Will this work on an NFS server? I remember an issue regarding a cookie
representing the directory offset: The problem was that the conversion
to HTREE of a directory while at the same time reading that directory
over NFS would basically invalidate the directory offset.

--
Frank

2004-01-04 00:44:59

by Theodore Ts'o

[permalink] [raw]
Subject: Re: [PATCH] [2.4.24-pre3] 0/5 EXT2/3 Updates

On Sun, Jan 04, 2004 at 01:07:47AM +0100, Frank van Maarseveen wrote:
> On Thu, Jan 01, 2004 at 04:50:20PM -0500, Theodore Ts'o wrote:
> > Patch 5: Add HTREE indexed directory support
>
> Will this work on an NFS server? I remember an issue regarding a cookie
> representing the directory offset: The problem was that the conversion
> to HTREE of a directory while at the same time reading that directory
> over NFS would basically invalidate the directory offset.

Yes, it will work over an NFS server. We solve this problem by having
readdir() always return directory entries of small directories (which
are eligible to be converted into HTREE format) in hash order.

- Ted