[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20080910042247.GA2544@gondor.apana.org.au>
Date: Wed, 10 Sep 2008 14:22:47 +1000
From: Herbert Xu <herbert@...dor.apana.org.au>
To: David Miller <davem@...emloft.net>
Cc: timo.teras@....fi, netdev@...r.kernel.org
Subject: Re: xfrm_state locking regression...
On Tue, Sep 09, 2008 at 09:06:54PM -0700, David Miller wrote:
>
> Only AF_KEY gave an error, and this ended the dump. This was
> one of Timo's goals, to make AF_KEY continue where it left
> off in subsequent dump calls done by the user, when we hit the
> socket limit.
I see.
> > While it is possible to restore the old order and save a little
> > bit of memory, I certainly don't think this is very urgent.
>
> Too late, I already implemented it :-)
Awesome :) Your patch is much simpler than I thought.
> diff --git a/include/net/xfrm.h b/include/net/xfrm.h
> index 4bb9499..ba0e47c 100644
> --- a/include/net/xfrm.h
> +++ b/include/net/xfrm.h
> @@ -120,7 +120,6 @@ extern struct mutex xfrm_cfg_mutex;
> /* Full description of state of transformer. */
> struct xfrm_state
> {
> - struct list_head all;
> union {
> struct list_head gclist;
> struct hlist_node bydst;
This union now needs to move across to bysrc or byspi since bydst
needs to be preserved for walking even after a node is added to
the GC.
> @@ -566,7 +564,6 @@ int __xfrm_state_delete(struct xfrm_state *x)
> x->km.state = XFRM_STATE_DEAD;
> spin_lock(&xfrm_state_lock);
> x->lastused = xfrm_state_walk_ongoing;
> - list_del_rcu(&x->all);
> hlist_del(&x->bydst);
We need hlist_del_rcu on bydst for the same reason.
> @@ -1571,24 +1564,29 @@ int xfrm_state_walk(struct xfrm_state_walk *walk,
> old = x = walk->state;
> walk->state = NULL;
> spin_lock_bh(&xfrm_state_lock);
> - if (x == NULL)
> - x = list_first_entry(&xfrm_state_all, struct xfrm_state, all);
> - list_for_each_entry_from(x, &xfrm_state_all, all) {
> - if (x->km.state == XFRM_STATE_DEAD)
> - continue;
> - if (!xfrm_id_proto_match(x->id.proto, walk->proto))
> - continue;
> - if (last) {
> - err = func(last, walk->count, data);
> - if (err) {
> - xfrm_state_hold(last);
> - walk->state = last;
> - xfrm_state_walk_ongoing++;
> - goto out;
> + for (i = walk->chain; i <= xfrm_state_hmask; i++) {
> + struct hlist_head *h = xfrm_state_bydst + i;
> + struct hlist_node *entry;
> +
> + if (!x)
> + x = hlist_entry(h->first, struct xfrm_state, bydst);
> + hlist_for_each_entry_from(x, entry, bydst) {
> + if (!xfrm_id_proto_match(x->id.proto, walk->proto))
> + continue;
> + if (last) {
> + err = func(last, walk->count, data);
> + if (err) {
This is a bug in the original patch. When we get an error on
the very last node we'll lose that node instead of dumping it
later. I wonder if we could just get rid of last altogether...
Thanks,
--
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <herbert@...dor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists