Re: iptables userspace API broken due to added value in nf_inet_hooks

[Pablo Neira Ayuso <pablo@netfilter.org>]

On Wed, Oct 14, 2020 at 02:59:47PM +0200, Jason A. Donenfeld wrote:
> Hey Pablo,
> 
> In 60a3815da702fd9e4759945f26cce5c47d3967ad, you added another enum
> value to nf_inet_hooks:
> 
> --- a/include/uapi/linux/netfilter.h
> +++ b/include/uapi/linux/netfilter.h
> @@ -45,6 +45,7 @@ enum nf_inet_hooks {
>        NF_INET_FORWARD,
>        NF_INET_LOCAL_OUT,
>        NF_INET_POST_ROUTING,
> +       NF_INET_INGRESS,
>        NF_INET_NUMHOOKS
> };
> 
> That seems fine, but actually it changes the value of
> NF_INET_NUMHOOKS, which is used in struct ipt_getinfo:
> 
> /* The argument to IPT_SO_GET_INFO */
> struct ipt_getinfo {
>        /* Which table: caller fills this in. */
>        char name[XT_TABLE_MAXNAMELEN];
> 
>        /* Kernel fills these in. */
>        /* Which hook entry points are valid: bitmask */
>        unsigned int valid_hooks;
> 
>        /* Hook entry points: one per netfilter hook. */
>        unsigned int hook_entry[NF_INET_NUMHOOKS];
> 
>        /* Underflow points. */
>        unsigned int underflow[NF_INET_NUMHOOKS];
> 
>        /* Number of entries */
>        unsigned int num_entries;
> 
>        /* Size of entries. */
>        unsigned int size;
> };
> 
> This in turn makes that struct bigger, which means this check in
> net/ipv4/netfilter/ip_tables.c fails:
> 
> static int get_info(struct net *net, void __user *user, const int *len)
> {
>        char name[XT_TABLE_MAXNAMELEN];
>        struct xt_table *t;
>        int ret;
> 
>        if (*len != sizeof(struct ipt_getinfo))
>                return -EINVAL;
> 
> This is affecting my CI, which attempts to use an older iptables with
> net-next and fails with:
> 
> iptables v1.8.4 (legacy): can't initialize iptables table `filter':
> Module is wrong version
> Perhaps iptables or your kernel needs to be upgraded.
> 
> Is this kind of breakage okay? If there's an exception carved out for
> breaking the iptables API, just let me know, and I'll look into making
> adjustments to work around it in my CI. On the other hand, if this
> breakage was unintentional, now you know.

Oh right, I'll need a new IPT_INET_NUMHOOKS for this.

I'll submit a patch, that's for the heads up.