Re: Socket locking

David S. Miller (davem@jenolan.rutgers.edu)
Sat, 4 Oct 1997 14:24:11 -0400


From: Andi Kleen <ak@muc.de>
Date: 04 Oct 1997 16:03:22 +0200

BTW with more stuff moving into the dentries - how about getting
rid of the socket inode and putting a socket pointer directly into
the dentry?

I think this is a good idea. Really dentry->d_inode is "backing store
object" so just changing that member of struct dentry to be
"void *d_object" would suffice.

But here is the other issue, don't we use the permission stuff in the
inode for socket processing in certain cases? And furthermore someone
else noted that we do use the locks in the inode so some sort of
replacement would need to be placed in the socket struct. But this is
OK because in the end we'll be saving space just by taking the inode
out of the picture. Come to think of it, this will make socket
creation and destruction much faster overall.

Later,
David "Sparc" Miller
davem@caip.rutgers.edu