Re: [PATCH] netfilter: ipset: add ip_set lock to ip_set_test

From: Jozsef Kadlecsik
Date: Mon Oct 02 2023 - 15:14:10 EST


Hi,

On Mon, 2 Oct 2023, Simon Horman wrote:

> On Wed, Sep 27, 2023 at 09:03:09PM +0800, xiaolinkui wrote:
> > From: Linkui Xiao <xiaolinkui@xxxxxxxxxx>
> >
> > If the ip_set is not locked during ip_set_test, the following situations
> > may occur:
> >
> > CPU0 CPU1
> > ip_rcv->
> > ip_rcv_finish->
> > ip_local_deliver->
> > nf_hook_slow->
> > iptable_filter_hook->
> > ipt_do_table->
> > set_match_v4->
> > ip_set_test-> list_set_destroy->
> > hash_net4_kadt-> set->data = NULL
>
> I'm having a bit of trouble analysing this.
> In particular, I'm concerned that in such a scenario set
> itself will be also freed, which seems likely to lead to problems.
>
> Can you provide a more complete call stack for CPU1 ?

ip_set_test() runs intentionally without holding a spinlock, it uses RCU.

But I don't understand the scenario at all:

CPU0: CPU1:
hash_net4_kadt list_set_destroy

so it's a hash:net type which works on a list
of set type of sets only

The list type of set can freely be destroyed (when not referenced), the
destroy operation has no effect whatsoever on its possible hash:net type
of member set.

Moreover, kernel side add/del/test can only be performed when the set in
question is referenced. Referenced sets cannot be deleted.

So what is the scenario really in this case?

Best regards,
Jozsef

> > h = set->data
> > .cidr = INIT_CIDR(h->nets[0].cidr[0], HOST_MASK)
> >
> > The set->data is empty, continuing to access set->data will result in a
> > kernel NULL pointer. The call trace is as follows:
> >
> > [2350616.024418] Call trace:
> > [2350616.024670] hash_net4_kadt+0x38/0x148 [ip_set_hash_net]
> > [2350616.025147] ip_set_test+0xbc/0x230 [ip_set]
> > [2350616.025549] set_match_v4+0xac/0xd0 [xt_set]
> > [2350616.025951] ipt_do_table+0x32c/0x678 [ip_tables]
> > [2350616.026391] iptable_filter_hook+0x30/0x40 [iptable_filter]
> > [2350616.026905] nf_hook_slow+0x50/0x100
> > [2350616.027256] ip_local_deliver+0xd4/0xe8
> > [2350616.027616] ip_rcv_finish+0x90/0xb0
> > [2350616.027961] ip_rcv+0x50/0xb0
> > [2350616.028261] __netif_receive_skb_one_core+0x58/0x68
> > [2350616.028716] __netif_receive_skb+0x28/0x80
> > [2350616.029098] netif_receive_skb_internal+0x3c/0xa8
> > [2350616.029533] napi_gro_receive+0xf8/0x170
> > [2350616.029898] receive_buf+0xec/0xa08 [virtio_net]
> > [2350616.030323] virtnet_poll+0x144/0x310 [virtio_net]
> > [2350616.030761] net_rx_action+0x158/0x3a0
> > [2350616.031124] __do_softirq+0x11c/0x33c
> > [2350616.031470] irq_exit+0x11c/0x128
> > [2350616.031793] __handle_domain_irq+0x6c/0xc0
> > [2350616.032172] gic_handle_irq+0x6c/0x170
> > [2350616.032528] el1_irq+0xb8/0x140
> > [2350616.032835] arch_cpu_idle+0x38/0x1c0
> > [2350616.033183] default_idle_call+0x24/0x58
> > [2350616.033549] do_idle+0x1a4/0x268
> > [2350616.033859] cpu_startup_entry+0x28/0x78
> > [2350616.034234] secondary_start_kernel+0x17c/0x1c8
> >
> > Signed-off-by: Linkui Xiao <xiaolinkui@xxxxxxxxxx>
> > ---
> > net/netfilter/ipset/ip_set_core.c | 2 ++
> > 1 file changed, 2 insertions(+)
> >
> > diff --git a/net/netfilter/ipset/ip_set_core.c b/net/netfilter/ipset/ip_set_core.c
> > index 35d2f9c9ada0..46f4f47e29e4 100644
> > --- a/net/netfilter/ipset/ip_set_core.c
> > +++ b/net/netfilter/ipset/ip_set_core.c
> > @@ -747,7 +747,9 @@ ip_set_test(ip_set_id_t index, const struct sk_buff *skb,
> > !(opt->family == set->family || set->family == NFPROTO_UNSPEC))
> > return 0;
> >
> > + ip_set_lock(set);
> > ret = set->variant->kadt(set, skb, par, IPSET_TEST, opt);
> > + ip_set_unlock(set);
> >
> > if (ret == -EAGAIN) {
> > /* Type requests element to be completed */
> > --
> > 2.17.1
> >
> >
>

-
E-mail : kadlec@xxxxxxxxxxxxxxxxx, kadlecsik.jozsef@xxxxxxxxx
PGP key : https://wigner.hu/~kadlec/pgp_public_key.txt
Address : Wigner Research Centre for Physics
H-1525 Budapest 114, POB. 49, Hungary