net: netfilter: Add kfuncs to allocate and insert CT
authorLorenzo Bianconi <lorenzo@kernel.org>
Thu, 21 Jul 2022 13:42:39 +0000 (15:42 +0200)
committerAlexei Starovoitov <ast@kernel.org>
Fri, 22 Jul 2022 04:03:16 +0000 (21:03 -0700)
Introduce bpf_xdp_ct_alloc, bpf_skb_ct_alloc and bpf_ct_insert_entry
kfuncs in order to insert a new entry from XDP and TC programs.
Introduce bpf_nf_ct_tuple_parse utility routine to consolidate common
code.

We extract out a helper __nf_ct_set_timeout, used by the ctnetlink and
nf_conntrack_bpf code, extract it out to nf_conntrack_core, so that
nf_conntrack_bpf doesn't need a dependency on CONFIG_NF_CT_NETLINK.
Later this helper will be reused as a helper to set timeout of allocated
but not yet inserted CT entry.

The allocation functions return struct nf_conn___init instead of
nf_conn, to distinguish allocated CT from an already inserted or looked
up CT. This is later used to enforce restrictions on what kfuncs
allocated CT can be used with.

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Co-developed-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20220721134245.2450-8-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
include/net/netfilter/nf_conntrack_core.h
net/netfilter/nf_conntrack_bpf.c
net/netfilter/nf_conntrack_netlink.c

index 37866c8386e234b261e5f665db29110bea262e06..83a60c684e6cc5dfb8c110227c0a247b474e175c 100644 (file)
@@ -84,4 +84,19 @@ void nf_conntrack_lock(spinlock_t *lock);
 
 extern spinlock_t nf_conntrack_expect_lock;
 
+/* ctnetlink code shared by both ctnetlink and nf_conntrack_bpf */
+
+#if (IS_BUILTIN(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \
+    (IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES) || \
+    IS_ENABLED(CONFIG_NF_CT_NETLINK))
+
+static inline void __nf_ct_set_timeout(struct nf_conn *ct, u64 timeout)
+{
+       if (timeout > INT_MAX)
+               timeout = INT_MAX;
+       WRITE_ONCE(ct->timeout, nfct_time_stamp + (u32)timeout);
+}
+
+#endif
+
 #endif /* _NF_CONNTRACK_CORE_H */
index 16304869264f1b00571c2fb352637de6d4d284e6..cac4a9558968ebfb706e0a26e477805bfaa146f7 100644 (file)
@@ -55,6 +55,94 @@ enum {
        NF_BPF_CT_OPTS_SZ = 12,
 };
 
+static int bpf_nf_ct_tuple_parse(struct bpf_sock_tuple *bpf_tuple,
+                                u32 tuple_len, u8 protonum, u8 dir,
+                                struct nf_conntrack_tuple *tuple)
+{
+       union nf_inet_addr *src = dir ? &tuple->dst.u3 : &tuple->src.u3;
+       union nf_inet_addr *dst = dir ? &tuple->src.u3 : &tuple->dst.u3;
+       union nf_conntrack_man_proto *sport = dir ? (void *)&tuple->dst.u
+                                                 : &tuple->src.u;
+       union nf_conntrack_man_proto *dport = dir ? &tuple->src.u
+                                                 : (void *)&tuple->dst.u;
+
+       if (unlikely(protonum != IPPROTO_TCP && protonum != IPPROTO_UDP))
+               return -EPROTO;
+
+       memset(tuple, 0, sizeof(*tuple));
+
+       switch (tuple_len) {
+       case sizeof(bpf_tuple->ipv4):
+               tuple->src.l3num = AF_INET;
+               src->ip = bpf_tuple->ipv4.saddr;
+               sport->tcp.port = bpf_tuple->ipv4.sport;
+               dst->ip = bpf_tuple->ipv4.daddr;
+               dport->tcp.port = bpf_tuple->ipv4.dport;
+               break;
+       case sizeof(bpf_tuple->ipv6):
+               tuple->src.l3num = AF_INET6;
+               memcpy(src->ip6, bpf_tuple->ipv6.saddr, sizeof(bpf_tuple->ipv6.saddr));
+               sport->tcp.port = bpf_tuple->ipv6.sport;
+               memcpy(dst->ip6, bpf_tuple->ipv6.daddr, sizeof(bpf_tuple->ipv6.daddr));
+               dport->tcp.port = bpf_tuple->ipv6.dport;
+               break;
+       default:
+               return -EAFNOSUPPORT;
+       }
+       tuple->dst.protonum = protonum;
+       tuple->dst.dir = dir;
+
+       return 0;
+}
+
+static struct nf_conn *
+__bpf_nf_ct_alloc_entry(struct net *net, struct bpf_sock_tuple *bpf_tuple,
+                       u32 tuple_len, struct bpf_ct_opts *opts, u32 opts_len,
+                       u32 timeout)
+{
+       struct nf_conntrack_tuple otuple, rtuple;
+       struct nf_conn *ct;
+       int err;
+
+       if (!opts || !bpf_tuple || opts->reserved[0] || opts->reserved[1] ||
+           opts_len != NF_BPF_CT_OPTS_SZ)
+               return ERR_PTR(-EINVAL);
+
+       if (unlikely(opts->netns_id < BPF_F_CURRENT_NETNS))
+               return ERR_PTR(-EINVAL);
+
+       err = bpf_nf_ct_tuple_parse(bpf_tuple, tuple_len, opts->l4proto,
+                                   IP_CT_DIR_ORIGINAL, &otuple);
+       if (err < 0)
+               return ERR_PTR(err);
+
+       err = bpf_nf_ct_tuple_parse(bpf_tuple, tuple_len, opts->l4proto,
+                                   IP_CT_DIR_REPLY, &rtuple);
+       if (err < 0)
+               return ERR_PTR(err);
+
+       if (opts->netns_id >= 0) {
+               net = get_net_ns_by_id(net, opts->netns_id);
+               if (unlikely(!net))
+                       return ERR_PTR(-ENONET);
+       }
+
+       ct = nf_conntrack_alloc(net, &nf_ct_zone_dflt, &otuple, &rtuple,
+                               GFP_ATOMIC);
+       if (IS_ERR(ct))
+               goto out;
+
+       memset(&ct->proto, 0, sizeof(ct->proto));
+       __nf_ct_set_timeout(ct, timeout * HZ);
+       ct->status |= IPS_CONFIRMED;
+
+out:
+       if (opts->netns_id >= 0)
+               put_net(net);
+
+       return ct;
+}
+
 static struct nf_conn *__bpf_nf_ct_lookup(struct net *net,
                                          struct bpf_sock_tuple *bpf_tuple,
                                          u32 tuple_len, struct bpf_ct_opts *opts,
@@ -63,6 +151,7 @@ static struct nf_conn *__bpf_nf_ct_lookup(struct net *net,
        struct nf_conntrack_tuple_hash *hash;
        struct nf_conntrack_tuple tuple;
        struct nf_conn *ct;
+       int err;
 
        if (!opts || !bpf_tuple || opts->reserved[0] || opts->reserved[1] ||
            opts_len != NF_BPF_CT_OPTS_SZ)
@@ -72,27 +161,10 @@ static struct nf_conn *__bpf_nf_ct_lookup(struct net *net,
        if (unlikely(opts->netns_id < BPF_F_CURRENT_NETNS))
                return ERR_PTR(-EINVAL);
 
-       memset(&tuple, 0, sizeof(tuple));
-       switch (tuple_len) {
-       case sizeof(bpf_tuple->ipv4):
-               tuple.src.l3num = AF_INET;
-               tuple.src.u3.ip = bpf_tuple->ipv4.saddr;
-               tuple.src.u.tcp.port = bpf_tuple->ipv4.sport;
-               tuple.dst.u3.ip = bpf_tuple->ipv4.daddr;
-               tuple.dst.u.tcp.port = bpf_tuple->ipv4.dport;
-               break;
-       case sizeof(bpf_tuple->ipv6):
-               tuple.src.l3num = AF_INET6;
-               memcpy(tuple.src.u3.ip6, bpf_tuple->ipv6.saddr, sizeof(bpf_tuple->ipv6.saddr));
-               tuple.src.u.tcp.port = bpf_tuple->ipv6.sport;
-               memcpy(tuple.dst.u3.ip6, bpf_tuple->ipv6.daddr, sizeof(bpf_tuple->ipv6.daddr));
-               tuple.dst.u.tcp.port = bpf_tuple->ipv6.dport;
-               break;
-       default:
-               return ERR_PTR(-EAFNOSUPPORT);
-       }
-
-       tuple.dst.protonum = opts->l4proto;
+       err = bpf_nf_ct_tuple_parse(bpf_tuple, tuple_len, opts->l4proto,
+                                   IP_CT_DIR_ORIGINAL, &tuple);
+       if (err < 0)
+               return ERR_PTR(err);
 
        if (opts->netns_id >= 0) {
                net = get_net_ns_by_id(net, opts->netns_id);
@@ -116,6 +188,43 @@ __diag_push();
 __diag_ignore_all("-Wmissing-prototypes",
                  "Global functions as their definitions will be in nf_conntrack BTF");
 
+struct nf_conn___init {
+       struct nf_conn ct;
+};
+
+/* bpf_xdp_ct_alloc - Allocate a new CT entry
+ *
+ * Parameters:
+ * @xdp_ctx    - Pointer to ctx (xdp_md) in XDP program
+ *                 Cannot be NULL
+ * @bpf_tuple  - Pointer to memory representing the tuple to look up
+ *                 Cannot be NULL
+ * @tuple__sz  - Length of the tuple structure
+ *                 Must be one of sizeof(bpf_tuple->ipv4) or
+ *                 sizeof(bpf_tuple->ipv6)
+ * @opts       - Additional options for allocation (documented above)
+ *                 Cannot be NULL
+ * @opts__sz   - Length of the bpf_ct_opts structure
+ *                 Must be NF_BPF_CT_OPTS_SZ (12)
+ */
+struct nf_conn___init *
+bpf_xdp_ct_alloc(struct xdp_md *xdp_ctx, struct bpf_sock_tuple *bpf_tuple,
+                u32 tuple__sz, struct bpf_ct_opts *opts, u32 opts__sz)
+{
+       struct xdp_buff *ctx = (struct xdp_buff *)xdp_ctx;
+       struct nf_conn *nfct;
+
+       nfct = __bpf_nf_ct_alloc_entry(dev_net(ctx->rxq->dev), bpf_tuple, tuple__sz,
+                                      opts, opts__sz, 10);
+       if (IS_ERR(nfct)) {
+               if (opts)
+                       opts->error = PTR_ERR(nfct);
+               return NULL;
+       }
+
+       return (struct nf_conn___init *)nfct;
+}
+
 /* bpf_xdp_ct_lookup - Lookup CT entry for the given tuple, and acquire a
  *                    reference to it
  *
@@ -150,6 +259,40 @@ bpf_xdp_ct_lookup(struct xdp_md *xdp_ctx, struct bpf_sock_tuple *bpf_tuple,
        return nfct;
 }
 
+/* bpf_skb_ct_alloc - Allocate a new CT entry
+ *
+ * Parameters:
+ * @skb_ctx    - Pointer to ctx (__sk_buff) in TC program
+ *                 Cannot be NULL
+ * @bpf_tuple  - Pointer to memory representing the tuple to look up
+ *                 Cannot be NULL
+ * @tuple__sz  - Length of the tuple structure
+ *                 Must be one of sizeof(bpf_tuple->ipv4) or
+ *                 sizeof(bpf_tuple->ipv6)
+ * @opts       - Additional options for allocation (documented above)
+ *                 Cannot be NULL
+ * @opts__sz   - Length of the bpf_ct_opts structure
+ *                 Must be NF_BPF_CT_OPTS_SZ (12)
+ */
+struct nf_conn___init *
+bpf_skb_ct_alloc(struct __sk_buff *skb_ctx, struct bpf_sock_tuple *bpf_tuple,
+                u32 tuple__sz, struct bpf_ct_opts *opts, u32 opts__sz)
+{
+       struct sk_buff *skb = (struct sk_buff *)skb_ctx;
+       struct nf_conn *nfct;
+       struct net *net;
+
+       net = skb->dev ? dev_net(skb->dev) : sock_net(skb->sk);
+       nfct = __bpf_nf_ct_alloc_entry(net, bpf_tuple, tuple__sz, opts, opts__sz, 10);
+       if (IS_ERR(nfct)) {
+               if (opts)
+                       opts->error = PTR_ERR(nfct);
+               return NULL;
+       }
+
+       return (struct nf_conn___init *)nfct;
+}
+
 /* bpf_skb_ct_lookup - Lookup CT entry for the given tuple, and acquire a
  *                    reference to it
  *
@@ -184,6 +327,26 @@ bpf_skb_ct_lookup(struct __sk_buff *skb_ctx, struct bpf_sock_tuple *bpf_tuple,
        return nfct;
 }
 
+/* bpf_ct_insert_entry - Add the provided entry into a CT map
+ *
+ * This must be invoked for referenced PTR_TO_BTF_ID.
+ *
+ * @nfct__ref   - Pointer to referenced nf_conn___init object, obtained
+ *                using bpf_xdp_ct_alloc or bpf_skb_ct_alloc.
+ */
+struct nf_conn *bpf_ct_insert_entry(struct nf_conn___init *nfct__ref)
+{
+       struct nf_conn *nfct = (struct nf_conn *)nfct__ref;
+       int err;
+
+       err = nf_conntrack_hash_check_insert(nfct);
+       if (err < 0) {
+               nf_conntrack_free(nfct);
+               return NULL;
+       }
+       return nfct;
+}
+
 /* bpf_ct_release - Release acquired nf_conn object
  *
  * This must be invoked for referenced PTR_TO_BTF_ID, and the verifier rejects
@@ -204,8 +367,11 @@ void bpf_ct_release(struct nf_conn *nfct)
 __diag_pop()
 
 BTF_SET8_START(nf_ct_kfunc_set)
+BTF_ID_FLAGS(func, bpf_xdp_ct_alloc, KF_ACQUIRE | KF_RET_NULL)
 BTF_ID_FLAGS(func, bpf_xdp_ct_lookup, KF_ACQUIRE | KF_RET_NULL)
+BTF_ID_FLAGS(func, bpf_skb_ct_alloc, KF_ACQUIRE | KF_RET_NULL)
 BTF_ID_FLAGS(func, bpf_skb_ct_lookup, KF_ACQUIRE | KF_RET_NULL)
+BTF_ID_FLAGS(func, bpf_ct_insert_entry, KF_ACQUIRE | KF_RET_NULL | KF_RELEASE)
 BTF_ID_FLAGS(func, bpf_ct_release, KF_RELEASE)
 BTF_SET8_END(nf_ct_kfunc_set)
 
index 722af5e309ba280a17553e0d4164dcd25d0754d9..0729b2f0d44fdccc93c7117553e4780126c1c33c 100644 (file)
@@ -2025,9 +2025,7 @@ static int ctnetlink_change_timeout(struct nf_conn *ct,
 {
        u64 timeout = (u64)ntohl(nla_get_be32(cda[CTA_TIMEOUT])) * HZ;
 
-       if (timeout > INT_MAX)
-               timeout = INT_MAX;
-       WRITE_ONCE(ct->timeout, nfct_time_stamp + (u32)timeout);
+       __nf_ct_set_timeout(ct, timeout);
 
        if (test_bit(IPS_DYING_BIT, &ct->status))
                return -ETIME;
@@ -2292,9 +2290,7 @@ ctnetlink_create_conntrack(struct net *net,
                goto err1;
 
        timeout = (u64)ntohl(nla_get_be32(cda[CTA_TIMEOUT])) * HZ;
-       if (timeout > INT_MAX)
-               timeout = INT_MAX;
-       ct->timeout = (u32)timeout + nfct_time_stamp;
+       __nf_ct_set_timeout(ct, timeout);
 
        rcu_read_lock();
        if (cda[CTA_HELP]) {