Re: patch to fix 2.1.49 busy inodes

Linus Torvalds (torvalds@transmeta.com)
Thu, 14 Aug 1997 09:58:00 -0700 (PDT)


On Thu, 14 Aug 1997, Alan Cox wrote:
> > AC - Should abstract sockets even have inodes? This problems rings a
> > bell from 15-18 months ago.
>
> Its the easiest way to make normal file operations do the right thing,
> stuff like read/write/fstat..

Indeed. Think of inodes as the "IO" entity - they have very little
connection other than the name to their on-disk counterparts. So you need
an inode for anything that is supposed to be able to handle IO.

Bill's patch applied, although it should no longer make any difference in
my tree (I no longer use "i_dev" at all in inode.c - the inodes are
indexed with "i_ino" and "i_sb", and nothing else matters for my inode.c)

Linus