Corey Minyard a =E9crit :
quoted text
> I believe there is a race in this patch:
>=20
> +    sk_for_each_rcu(sk, node, &hslot->head) {
> +        /*
> +         * lockless reader, and SLAB_DESTROY_BY_RCU items:
> +         * We must check this item was not moved to another chain
> +         */
> +        if (udp_hashfn(net, sk->sk_hash) !=3D hash)
> +            goto begin;
>         score =3D compute_score(sk, net, hnum, saddr, sport, daddr, dpo=
rt,=20
quoted text
> dif);
>         if (score > badness) {
>             result =3D sk;
>             badness =3D score;
>         }
>     }
>=20
> If the socket is moved from one list to another list in-between the tim=
e=20
quoted text
> the hash is calculated and the next field is accessed, and the socket=20
> has moved to the end of the new list, the traversal will not complete=20
> properly on the list it should have, since the socket will be on the en=
d=20
quoted text
> of the new list and there's not a way to tell it's on a new list and=20
> restart the list traversal.  I think that this can be solved by=20
> pre-fetching the "next" field (with proper barriers) before checking th=
e=20
quoted text
> hash.
>=20

You are absolutely right. As we *need* next pointer anyway for the prefet=
ch(),
we can store it, so that its value is kept.

quoted text
> I also might be nice to have a way to avoid recomputing the score the=20
> second time, perhaps using a sequence number of some type.

Well, computing score is really cheap, everything is in cache, granted
the chain was not really huge and high score item at the beginning.

Adding yet another field in sock structure should be avoided if possible.=


Thanks

[PATCH] udp: introduce sk_for_each_rcu_safenext()

Corey Minyard found a race added in commit 271b72c7fa82c2c7a795bc16896149=
933110672d
(udp: RCU handling for Unicast packets.)

 "If the socket is moved from one list to another list in-between the tim=
e=20
  the hash is calculated and the next field is accessed, and the socket=20
  has moved to the end of the new list, the traversal will not complete=20
  properly on the list it should have, since the socket will be on the en=
d=20
  of the new list and there's not a way to tell it's on a new list and=20
  restart the list traversal.  I think that this can be solved by=20
  pre-fetching the "next" field (with proper barriers) before checking th=
e=20
  hash."

This patch corrects this problem, introducing a new sk_for_each_rcu_safen=
ext()
macro.

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
---
 include/linux/rculist.h |   17 +++++++++++++++++
 include/net/sock.h      |    4 ++--
 net/ipv4/udp.c          |    4 ++--
 net/ipv6/udp.c          |    4 ++--
 4 files changed, 23 insertions(+), 6 deletions(-)