On 11/20/17, 7:19 AM, "[email protected] on behalf of Aaron
Conole" <[email protected] on behalf of [email protected]> wrote:
Hi Darrell,
Darrell Ball <[email protected]> writes:
> Presently, alg processing is enabled by default to exercise testing.
> This is similar to kernels before 4.7. The recommended default
> behavior in the kernel is to only process algs if a helper is
> supplied in a conntrack rule. The behavior is changed to match the
> later kernels.
>
> Signed-off-by: Darrell Ball <[email protected]>
> ---
I like having the alg processing off by default. However, at this point
some folks may be relying (or not care about) on the default behavior of
the ct() action (which is to assume
ALG=whatever-is-appropriate-to-the-port). I think we probably will need
to have a knob to enable / disable this.
This part has me worried.
The kernel has taken the position that the user should operate such that a
helper is
needed to be specified. This gives the user total granular control as well.
i.e. the kernel has taken the position that net.netfilter.nf_conntrack_helper
should be off by default
in all cases.
Allowing the user to turn this on is a double-edged sword. Yes, it allows the
user to not have to specify
alg=blah, but is needing to specify alg= all that onerous?
On the flip side, allowing the user to skip specifying “alg=blah” opens a
security hole that will never be
noticed until a problem occurs – that is really nasty. Given the kernel
position, it may be ok to do the
right thing here and not allow the user to shoot himself in the foot accidently.
More below.
> lib/conntrack.c | 35 +++++++++++++++++++++++++++++------
> 1 file changed, 29 insertions(+), 6 deletions(-)
>
> diff --git a/lib/conntrack.c b/lib/conntrack.c
> index 7fbcfba..dea2fed 100644
> --- a/lib/conntrack.c
> +++ b/lib/conntrack.c
> @@ -789,13 +789,34 @@ conn_clean(struct conntrack *ct, struct conn *conn,
> }
> }
>
> +static bool
> +ct_verify_helper(const char *helper, enum ct_alg_ctl_type ct_alg_ctl)
> +{
Since we know the algorithm at this point, doesn't it make sense to
just associate the alg with the connection? I say that because if I
read this correctly (and I haven't had coffee yet - maybe I missed
something) non-standard ports won't get their associations properly (so
if I want to run ftp on port 2122 for example, the helper will not get
associated).
If we take the approach I mentioned in 2/4, we can just associate the
function pointer. After all, the user asked for it - we should do what
the user wants.
It is orthogonal.
This of CT_ALG_CTL_FTP and CT_ALG_CTL_TFTP as array indexes (I’ll explain this
more in my response to patch 2).
i.e. CT_ALG_CTL_FTP and CT_ALG_CTL_TFTP are not linked to specific L4 port
numbers.
> + if (ct_alg_ctl == CT_ALG_CTL_NONE) {
> + return true;
> + } else if (helper) {
> + if ((ct_alg_ctl == CT_ALG_CTL_FTP) &&
> + !strncmp(helper, "ftp", strlen("ftp"))) {
> + return true;
> + } else if ((ct_alg_ctl == CT_ALG_CTL_TFTP) &&
> + !strncmp(helper, "tftp", strlen("tftp"))) {
> + return true;
> + } else {
> + return false;
> + }
> + } else {
> + return false;
> + }
> +}
> +
> /* This function is called with the bucket lock held. */
> static struct conn *
> conn_not_found(struct conntrack *ct, struct dp_packet *pkt,
> struct conn_lookup_ctx *ctx, bool commit, long long now,
> const struct nat_action_info_t *nat_action_info,
> struct conn *conn_for_un_nat_copy,
> - const char *helper, const struct alg_exp_node *alg_exp)
> + const char *helper, const struct alg_exp_node *alg_exp,
> + enum ct_alg_ctl_type ct_alg_ctl)
> {
> struct conn *nc = NULL;
>
> @@ -819,15 +840,16 @@ conn_not_found(struct conntrack *ct, struct
dp_packet *pkt,
> return nc;
> }
>
> + if (!ct_verify_helper(helper, ct_alg_ctl)) {
> + return nc;
> + }
> +
> unsigned bucket = hash_to_bucket(ctx->hash);
> nc = new_conn(&ct->buckets[bucket], pkt, &ctx->key, now);
> ctx->conn = nc;
> nc->rev_key = nc->key;
> conn_key_reverse(&nc->rev_key);
> -
> - if (helper) {
> - nc->alg = xstrdup(helper);
> - }
> + nc->alg = nullable_xstrdup(helper);
>
> if (alg_exp) {
> nc->alg_related = true;
> @@ -1182,7 +1204,8 @@ process_one(struct conntrack *ct, struct dp_packet
*pkt,
> }
> ct_rwlock_unlock(&ct->resources_lock);
> conn = conn_not_found(ct, pkt, ctx, commit, now, nat_action_info,
> - &conn_for_un_nat_copy, helper, alg_exp);
> + &conn_for_un_nat_copy, helper, alg_exp,
> + ct_alg_ctl);
> }
> write_ct_md(pkt, zone, conn, &ctx->key, alg_exp);
_______________________________________________
dev mailing list
[email protected]
https://urldefense.proofpoint.com/v2/url?u=https-3A__mail.openvswitch.org_mailman_listinfo_ovs-2Ddev&d=DwICAg&c=uilaK90D4TOVoH58JNXRgQ&r=BVhFA09CGX7JQ5Ih-uZnsw&m=phyp9Vtggl--DceKrVCrgijrJyeAjHO8WeqZHo1QALQ&s=sqNKc1aLhgXDSZ_0NvVLqf264Uz_PSeJHqBUr2Ufsho&e=
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
