Re: [LART] inode mismanagement in hugetlb code

[William Lee Irwin III <wli@holomorphy.com>]

Alexander Viro wrote:
>>         a) inodes MUST have an address of valid struct super_block in their
>> ->i_sb.  Had been discussed quite a few times already.

On Thu, Oct 10, 2002 at 12:22:28PM -0700, Andrew Morton wrote:
> afaict, that code only wants an inode because it is borrowing
> the pagecache functions for page lookup.  It's using i_ino as
> a search key too.  It has no superblock.
> Solutions might be: 1) allocate a private <int key, radix tree>
> structure or 2) require that these inodes come from hugetlbfs,
> although the "key" makes that a bit tricky.

shm clobbers hugetlbfs-assigned inode numbers with the shp->id
somewhere around newseg(), (which doesn't matter because it's the
only user of the kern_mounted fs) so ->i_ino will need to avoid
clashing there. But there are several ways to wean hugetlbpage.c off
->i_ino so it's not a big deal.

Bill