bpf: Add support for writing to nf_conn:mark

Support direct writes to nf_conn:mark from TC and XDP prog types. This
is useful when applications want to store per-connection metadata. This
is also particularly useful for applications that run both bpf and
iptables/nftables because the latter can trivially access this metadata.

One example use case would be if a bpf prog is responsible for advanced
packet classification and iptables/nftables is later used for routing
due to pre-existing/legacy code.

Signed-off-by: Daniel Xu <[e-mail]>
Link: https://lore.kernel.org/r/[e-mail]
Signed-off-by: Alexei Starovoitov <[e-mail]>

include/net/netfilter/nf_conntrack_bpf.h

···
 #ifndef _NF_CONNTRACK_BPF_H
 #define _NF_CONNTRACK_BPF_H
 
+#include <linux/bpf.h>
 #include <linux/btf.h>
 #include <linux/kconfig.h>
+#include <linux/mutex.h>
 
 #if (IS_BUILTIN(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \
     (IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES))
 
 extern int register_nf_conntrack_bpf(void);
+extern void cleanup_nf_conntrack_bpf(void);
+
+extern struct mutex nf_conn_btf_access_lock;
+extern int (*nfct_bsa)(struct bpf_verifier_log *log, const struct btf *btf,
+		       const struct btf_type *t, int off, int size,
+		       enum bpf_access_type atype, u32 *next_btf_id,
+		       enum bpf_type_flag *flag);
 
 #else
 
 static inline int register_nf_conntrack_bpf(void)
 {
 	return 0;
+}
+
+static inline void cleanup_nf_conntrack_bpf(void)
+{
+}
+
+static inline int nf_conntrack_btf_struct_access(struct bpf_verifier_log *log,
+						 const struct btf *btf,
+						 const struct btf_type *t, int off,
+						 int size, enum bpf_access_type atype,
+						 u32 *next_btf_id,
+						 enum bpf_type_flag *flag)
+{
+	return -EACCES;
 }
 
 #endif

net/core/filter.c

···
  */
 
 #include <linux/atomic.h>
+#include <linux/bpf_verifier.h>
 #include <linux/module.h>
 #include <linux/types.h>
 #include <linux/mm.h>
···
 	return bpf_skb_is_valid_access(off, size, type, prog, info);
 }
 
+DEFINE_MUTEX(nf_conn_btf_access_lock);
+EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock);
+
+int (*nfct_bsa)(struct bpf_verifier_log *log, const struct btf *btf,
+		const struct btf_type *t, int off, int size,
+		enum bpf_access_type atype, u32 *next_btf_id,
+		enum bpf_type_flag *flag);
+EXPORT_SYMBOL_GPL(nfct_bsa);
+
+static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log,
+					const struct btf *btf,
+					const struct btf_type *t, int off,
+					int size, enum bpf_access_type atype,
+					u32 *next_btf_id,
+					enum bpf_type_flag *flag)
+{
+	int ret = -EACCES;
+
+	if (atype == BPF_READ)
+		return btf_struct_access(log, btf, t, off, size, atype, next_btf_id,
+					 flag);
+
+	mutex_lock(&nf_conn_btf_access_lock);
+	if (nfct_bsa)
+		ret = nfct_bsa(log, btf, t, off, size, atype, next_btf_id, flag);
+	mutex_unlock(&nf_conn_btf_access_lock);
+
+	return ret;
+}
+
 static bool __is_valid_xdp_access(int off, int size)
 {
 	if (off < 0 || off >= sizeof(struct xdp_md))
···
 		     act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
 }
 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
+
+static int xdp_btf_struct_access(struct bpf_verifier_log *log,
+				 const struct btf *btf,
+				 const struct btf_type *t, int off,
+				 int size, enum bpf_access_type atype,
+				 u32 *next_btf_id,
+				 enum bpf_type_flag *flag)
+{
+	int ret = -EACCES;
+
+	if (atype == BPF_READ)
+		return btf_struct_access(log, btf, t, off, size, atype, next_btf_id,
+					 flag);
+
+	mutex_lock(&nf_conn_btf_access_lock);
+	if (nfct_bsa)
+		ret = nfct_bsa(log, btf, t, off, size, atype, next_btf_id, flag);
+	mutex_unlock(&nf_conn_btf_access_lock);
+
+	return ret;
+}
 
 static bool sock_addr_is_valid_access(int off, int size,
 				      enum bpf_access_type type,
···
 	.convert_ctx_access	= tc_cls_act_convert_ctx_access,
 	.gen_prologue		= tc_cls_act_prologue,
 	.gen_ld_abs		= bpf_gen_ld_abs,
+	.btf_struct_access	= tc_cls_act_btf_struct_access,
 };
 
 const struct bpf_prog_ops tc_cls_act_prog_ops = {
···
 	.is_valid_access	= xdp_is_valid_access,
 	.convert_ctx_access	= xdp_convert_ctx_access,
 	.gen_prologue		= bpf_noop_prologue,
+	.btf_struct_access	= xdp_btf_struct_access,
 };
 
 const struct bpf_prog_ops xdp_prog_ops = {

net/netfilter/nf_conntrack_bpf.c

···
  * are exposed through to BPF programs is explicitly unstable.
  */
 
+#include <linux/bpf_verifier.h>
 #include <linux/bpf.h>
 #include <linux/btf.h>
+#include <linux/mutex.h>
 #include <linux/types.h>
 #include <linux/btf_ids.h>
 #include <linux/net_namespace.h>
···
 	opts->dir = NF_CT_DIRECTION(hash);
 
 	return ct;
+}
+
+BTF_ID_LIST(btf_nf_conn_ids)
+BTF_ID(struct, nf_conn)
+BTF_ID(struct, nf_conn___init)
+
+/* Check writes into `struct nf_conn` */
+static int _nf_conntrack_btf_struct_access(struct bpf_verifier_log *log,
+					   const struct btf *btf,
+					   const struct btf_type *t, int off,
+					   int size, enum bpf_access_type atype,
+					   u32 *next_btf_id,
+					   enum bpf_type_flag *flag)
+{
+	const struct btf_type *ncit;
+	const struct btf_type *nct;
+	size_t end;
+
+	ncit = btf_type_by_id(btf, btf_nf_conn_ids[1]);
+	nct = btf_type_by_id(btf, btf_nf_conn_ids[0]);
+
+	if (t != nct && t != ncit) {
+		bpf_log(log, "only read is supported\n");
+		return -EACCES;
+	}
+
+	/* `struct nf_conn` and `struct nf_conn___init` have the same layout
+	 * so we are safe to simply merge offset checks here
+	 */
+	switch (off) {
+#if defined(CONFIG_NF_CONNTRACK_MARK)
+	case offsetof(struct nf_conn, mark):
+		end = offsetofend(struct nf_conn, mark);
+		break;
+#endif
+	default:
+		bpf_log(log, "no write support to nf_conn at off %d\n", off);
+		return -EACCES;
+	}
+
+	if (off + size > end) {
+		bpf_log(log,
+			"write access at off %d with size %d beyond the member of nf_conn ended at %zu\n",
+			off, size, end);
+		return -EACCES;
+	}
+
+	return 0;
 }
 
 __diag_push();
···
 	int ret;
 
 	ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &nf_conntrack_kfunc_set);
-	return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &nf_conntrack_kfunc_set);
+	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &nf_conntrack_kfunc_set);
+	if (!ret) {
+		mutex_lock(&nf_conn_btf_access_lock);
+		nfct_bsa = _nf_conntrack_btf_struct_access;
+		mutex_unlock(&nf_conn_btf_access_lock);
+	}
+
+	return ret;
+}
+
+void cleanup_nf_conntrack_bpf(void)
+{
+	mutex_lock(&nf_conn_btf_access_lock);
+	nfct_bsa = NULL;
+	mutex_unlock(&nf_conn_btf_access_lock);
 }

net/netfilter/nf_conntrack_core.c

···
 
 void nf_conntrack_cleanup_start(void)
 {
+	cleanup_nf_conntrack_bpf();
 	conntrack_gc_work.exiting = true;
 }