Herbert Xu wrote:

On Wed, Sep 03, 2008 at 08:07:33AM +0300, Timo Teräs wrote:

quoted

But then caller of xfrm_state_walk() can be still holding a
reference to the entry and dumping of the whole chain fails,
because the entry is no longer part of the chain when dumping
continues later.

The walking coding skips entries which are dead so the
entry is only kept temporarily to make a full traversal of
the list.

Ah I see your point now.

Actually another thing we should do is to postpone the unlinking
into the GC process.  This doesn't really need to be immediate.

Yes, might be a good idea.

And here is a patch to fix a couple of changes in user-behaviour.

ipsec: Restore larval states and socket policies in dump

The commit commit 4c563f7669c10a12354b72b518c2287ffc6ebfb3 ("[XFRM]:
Speed up xfrm_policy and xfrm_state walking") inadvertently removed
larval states and socket policies from netlink dumps.  This patch
restores them.

Oh, I seemed to have missed that there was multiple places where
the list insertion happens. Would it make sense to make function
for inserting the entries in the lists?

- Timo