Merge branch 'Split bpf_sock dst_port field'

Jakub Sitnicki says:

====================

This is a follow-up to discussion around the idea of making dst_port in struct
bpf_sock a 16-bit field that happened in [1].

v2:
- use an anonymous field for zero padding (Alexei)

v1:
- keep dst_field offset unchanged to prevent existing BPF program breakage
  (Martin)
- allow 8-bit loads from dst_port[0] and [1]
- add test coverage for the verifier and the context access converter

[1] https://lore.kernel.org/bpf/87sftbobys.fsf@cloudflare.com/
====================
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

include/uapi/linux/bpf.h

@@ -5574,7 +5574,8 @@ struct bpf_sock {
 	__u32 src_ip4;
 	__u32 src_ip6[4];
 	__u32 src_port;		/* host byte order */
-	__u32 dst_port;		/* network byte order */
+	__be16 dst_port;	/* network byte order */
+	__u16 :16;		/* zero padding */
 	__u32 dst_ip4;
 	__u32 dst_ip6[4];
 	__u32 state;

net/core/filter.c

@@ -8265,6 +8265,7 @@ bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
 			      struct bpf_insn_access_aux *info)
 {
 	const int size_default = sizeof(__u32);
+	int field_size;
 
 	if (off < 0 || off >= sizeof(struct bpf_sock))
 		return false;
@@ -8276,7 +8277,6 @@ bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
 	case offsetof(struct bpf_sock, family):
 	case offsetof(struct bpf_sock, type):
 	case offsetof(struct bpf_sock, protocol):
-	case offsetof(struct bpf_sock, dst_port):
 	case offsetof(struct bpf_sock, src_port):
 	case offsetof(struct bpf_sock, rx_queue_mapping):
 	case bpf_ctx_range(struct bpf_sock, src_ip4):
@@ -8285,6 +8285,14 @@ bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
 	case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
 		bpf_ctx_record_field_size(info, size_default);
 		return bpf_ctx_narrow_access_ok(off, size, size_default);
+	case bpf_ctx_range(struct bpf_sock, dst_port):
+		field_size = size == size_default ?
+			size_default : sizeof_field(struct bpf_sock, dst_port);
+		bpf_ctx_record_field_size(info, field_size);
+		return bpf_ctx_narrow_access_ok(off, size, field_size);
+	case offsetofend(struct bpf_sock, dst_port) ...
+	     offsetof(struct bpf_sock, dst_ip4) - 1:
+		return false;
 	}
 
 	return size == size_default;

tools/include/uapi/linux/bpf.h

@@ -5574,7 +5574,8 @@ struct bpf_sock {
 	__u32 src_ip4;
 	__u32 src_ip6[4];
 	__u32 src_port;		/* host byte order */
-	__u32 dst_port;		/* network byte order */
+	__be16 dst_port;	/* network byte order */
+	__u16 :16;		/* zero padding */
 	__u32 dst_ip4;
 	__u32 dst_ip6[4];
 	__u32 state;

tools/testing/selftests/bpf/prog_tests/sock_fields.c

 // SPDX-License-Identifier: GPL-2.0
 /* Copyright (c) 2019 Facebook */
 
+#define _GNU_SOURCE
 #include <netinet/in.h>
 #include <arpa/inet.h>
 #include <unistd.h>
+#include <sched.h>
 #include <stdlib.h>
 #include <string.h>
 #include <errno.h>
@@ -20,6 +22,7 @@
 enum bpf_linum_array_idx {
 	EGRESS_LINUM_IDX,
 	INGRESS_LINUM_IDX,
+	READ_SK_DST_PORT_LINUM_IDX,
 	__NR_BPF_LINUM_ARRAY_IDX,
 };
 
@@ -42,8 +45,16 @@ static __u64 child_cg_id;
 static int linum_map_fd;
 static __u32 duration;
 
-static __u32 egress_linum_idx = EGRESS_LINUM_IDX;
-static __u32 ingress_linum_idx = INGRESS_LINUM_IDX;
+static bool create_netns(void)
+{
+	if (!ASSERT_OK(unshare(CLONE_NEWNET), "create netns"))
+		return false;
+
+	if (!ASSERT_OK(system("ip link set dev lo up"), "bring up lo"))
+		return false;
+
+	return true;
+}
 
 static void print_sk(const struct bpf_sock *sk, const char *prefix)
 {
@@ -91,19 +102,24 @@ static void check_result(void)
 {
 	struct bpf_tcp_sock srv_tp, cli_tp, listen_tp;
 	struct bpf_sock srv_sk, cli_sk, listen_sk;
-	__u32 ingress_linum, egress_linum;
+	__u32 idx, ingress_linum, egress_linum, linum;
 	int err;
 
-	err = bpf_map_lookup_elem(linum_map_fd, &egress_linum_idx,
-				  &egress_linum);
+	idx = EGRESS_LINUM_IDX;
+	err = bpf_map_lookup_elem(linum_map_fd, &idx, &egress_linum);
 	CHECK(err < 0, "bpf_map_lookup_elem(linum_map_fd)",
 	      "err:%d errno:%d\n", err, errno);
 
-	err = bpf_map_lookup_elem(linum_map_fd, &ingress_linum_idx,
-				  &ingress_linum);
+	idx = INGRESS_LINUM_IDX;
+	err = bpf_map_lookup_elem(linum_map_fd, &idx, &ingress_linum);
 	CHECK(err < 0, "bpf_map_lookup_elem(linum_map_fd)",
 	      "err:%d errno:%d\n", err, errno);
 
+	idx = READ_SK_DST_PORT_LINUM_IDX;
+	err = bpf_map_lookup_elem(linum_map_fd, &idx, &linum);
+	ASSERT_OK(err, "bpf_map_lookup_elem(linum_map_fd, READ_SK_DST_PORT_IDX)");
+	ASSERT_EQ(linum, 0, "failure in read_sk_dst_port on line");
+
 	memcpy(&srv_sk, &skel->bss->srv_sk, sizeof(srv_sk));
 	memcpy(&srv_tp, &skel->bss->srv_tp, sizeof(srv_tp));
 	memcpy(&cli_sk, &skel->bss->cli_sk, sizeof(cli_sk));
@@ -262,7 +278,7 @@ static void test(void)
 	char buf[DATA_LEN];
 
 	/* Prepare listen_fd */
-	listen_fd = start_server(AF_INET6, SOCK_STREAM, "::1", 0, 0);
+	listen_fd = start_server(AF_INET6, SOCK_STREAM, "::1", 0xcafe, 0);
 	/* start_server() has logged the error details */
 	if (CHECK_FAIL(listen_fd == -1))
 		goto done;
@@ -330,8 +346,12 @@ static void test(void)
 
 void serial_test_sock_fields(void)
 {
-	struct bpf_link *egress_link = NULL, *ingress_link = NULL;
 	int parent_cg_fd = -1, child_cg_fd = -1;
+	struct bpf_link *link;
+
+	/* Use a dedicated netns to have a fixed listen port */
+	if (!create_netns())
+		return;
 
 	/* Create a cgroup, get fd, and join it */
 	parent_cg_fd = test__join_cgroup(PARENT_CGROUP);
@@ -352,15 +372,20 @@ void serial_test_sock_fields(void)
 	if (CHECK(!skel, "test_sock_fields__open_and_load", "failed\n"))
 		goto done;
 
-	egress_link = bpf_program__attach_cgroup(skel->progs.egress_read_sock_fields,
-						 child_cg_fd);
-	if (!ASSERT_OK_PTR(egress_link, "attach_cgroup(egress)"))
+	link = bpf_program__attach_cgroup(skel->progs.egress_read_sock_fields, child_cg_fd);
+	if (!ASSERT_OK_PTR(link, "attach_cgroup(egress_read_sock_fields)"))
+		goto done;
+	skel->links.egress_read_sock_fields = link;
+
+	link = bpf_program__attach_cgroup(skel->progs.ingress_read_sock_fields, child_cg_fd);
+	if (!ASSERT_OK_PTR(link, "attach_cgroup(ingress_read_sock_fields)"))
 		goto done;
+	skel->links.ingress_read_sock_fields = link;
 
-	ingress_link = bpf_program__attach_cgroup(skel->progs.ingress_read_sock_fields,
-						  child_cg_fd);
-	if (!ASSERT_OK_PTR(ingress_link, "attach_cgroup(ingress)"))
+	link = bpf_program__attach_cgroup(skel->progs.read_sk_dst_port, child_cg_fd);
+	if (!ASSERT_OK_PTR(link, "attach_cgroup(read_sk_dst_port"))
 		goto done;
+	skel->links.read_sk_dst_port = link;
 
 	linum_map_fd = bpf_map__fd(skel->maps.linum_map);
 	sk_pkt_out_cnt_fd = bpf_map__fd(skel->maps.sk_pkt_out_cnt);
@@ -369,8 +394,7 @@ void serial_test_sock_fields(void)
 	test();
 
 done:
-	bpf_link__destroy(egress_link);
-	bpf_link__destroy(ingress_link);
+	test_sock_fields__detach(skel);
 	test_sock_fields__destroy(skel);
 	if (child_cg_fd >= 0)
 		close(child_cg_fd);

tools/testing/selftests/bpf/progs/test_sock_fields.c

@@ -12,6 +12,7 @@
 enum bpf_linum_array_idx {
 	EGRESS_LINUM_IDX,
 	INGRESS_LINUM_IDX,
+	READ_SK_DST_PORT_LINUM_IDX,
 	__NR_BPF_LINUM_ARRAY_IDX,
 };
 
@@ -250,4 +251,44 @@ int ingress_read_sock_fields(struct __sk_buff *skb)
 	return CG_OK;
 }
 
+static __noinline bool sk_dst_port__load_word(struct bpf_sock *sk)
+{
+	__u32 *word = (__u32 *)&sk->dst_port;
+	return word[0] == bpf_htonl(0xcafe0000);
+}
+
+static __noinline bool sk_dst_port__load_half(struct bpf_sock *sk)
+{
+	__u16 *half = (__u16 *)&sk->dst_port;
+	return half[0] == bpf_htons(0xcafe);
+}
+
+static __noinline bool sk_dst_port__load_byte(struct bpf_sock *sk)
+{
+	__u8 *byte = (__u8 *)&sk->dst_port;
+	return byte[0] == 0xca && byte[1] == 0xfe;
+}
+
+SEC("cgroup_skb/egress")
+int read_sk_dst_port(struct __sk_buff *skb)
+{
+	__u32 linum, linum_idx;
+	struct bpf_sock *sk;
+
+	linum_idx = READ_SK_DST_PORT_LINUM_IDX;
+
+	sk = skb->sk;
+	if (!sk)
+		RET_LOG();
+
+	if (!sk_dst_port__load_word(sk))
+		RET_LOG();
+	if (!sk_dst_port__load_half(sk))
+		RET_LOG();
+	if (!sk_dst_port__load_byte(sk))
+		RET_LOG();
+
+	return CG_OK;
+}
+
 char _license[] SEC("license") = "GPL";

tools/testing/selftests/bpf/verifier/sock.c

@@ -121,7 +121,25 @@
 	.result = ACCEPT,
 },
 {
-	"sk_fullsock(skb->sk): sk->dst_port [narrow load]",
+	"sk_fullsock(skb->sk): sk->dst_port [word load] (backward compatibility)",
+	.insns = {
+	BPF_LDX_MEM(BPF_DW, BPF_REG_1, BPF_REG_1, offsetof(struct __sk_buff, sk)),
+	BPF_JMP_IMM(BPF_JNE, BPF_REG_1, 0, 2),
+	BPF_MOV64_IMM(BPF_REG_0, 0),
+	BPF_EXIT_INSN(),
+	BPF_EMIT_CALL(BPF_FUNC_sk_fullsock),
+	BPF_JMP_IMM(BPF_JNE, BPF_REG_0, 0, 2),
+	BPF_MOV64_IMM(BPF_REG_0, 0),
+	BPF_EXIT_INSN(),
+	BPF_LDX_MEM(BPF_W, BPF_REG_0, BPF_REG_0, offsetof(struct bpf_sock, dst_port)),
+	BPF_MOV64_IMM(BPF_REG_0, 0),
+	BPF_EXIT_INSN(),
+	},
+	.prog_type = BPF_PROG_TYPE_CGROUP_SKB,
+	.result = ACCEPT,
+},
+{
+	"sk_fullsock(skb->sk): sk->dst_port [half load]",
 	.insns = {
 	BPF_LDX_MEM(BPF_DW, BPF_REG_1, BPF_REG_1, offsetof(struct __sk_buff, sk)),
 	BPF_JMP_IMM(BPF_JNE, BPF_REG_1, 0, 2),
@@ -139,7 +157,64 @@
 	.result = ACCEPT,
 },
 {
-	"sk_fullsock(skb->sk): sk->dst_port [load 2nd byte]",
+	"sk_fullsock(skb->sk): sk->dst_port [half load] (invalid)",
+	.insns = {
+	BPF_LDX_MEM(BPF_DW, BPF_REG_1, BPF_REG_1, offsetof(struct __sk_buff, sk)),
+	BPF_JMP_IMM(BPF_JNE, BPF_REG_1, 0, 2),
+	BPF_MOV64_IMM(BPF_REG_0, 0),
+	BPF_EXIT_INSN(),
+	BPF_EMIT_CALL(BPF_FUNC_sk_fullsock),
+	BPF_JMP_IMM(BPF_JNE, BPF_REG_0, 0, 2),
+	BPF_MOV64_IMM(BPF_REG_0, 0),
+	BPF_EXIT_INSN(),
+	BPF_LDX_MEM(BPF_H, BPF_REG_0, BPF_REG_0, offsetof(struct bpf_sock, dst_port) + 2),
+	BPF_MOV64_IMM(BPF_REG_0, 0),
+	BPF_EXIT_INSN(),
+	},
+	.prog_type = BPF_PROG_TYPE_CGROUP_SKB,
+	.result = REJECT,
+	.errstr = "invalid sock access",
+},
+{
+	"sk_fullsock(skb->sk): sk->dst_port [byte load]",
+	.insns = {
+	BPF_LDX_MEM(BPF_DW, BPF_REG_1, BPF_REG_1, offsetof(struct __sk_buff, sk)),
+	BPF_JMP_IMM(BPF_JNE, BPF_REG_1, 0, 2),
+	BPF_MOV64_IMM(BPF_REG_0, 0),
+	BPF_EXIT_INSN(),
+	BPF_EMIT_CALL(BPF_FUNC_sk_fullsock),
+	BPF_JMP_IMM(BPF_JNE, BPF_REG_0, 0, 2),
+	BPF_MOV64_IMM(BPF_REG_0, 0),
+	BPF_EXIT_INSN(),
+	BPF_LDX_MEM(BPF_B, BPF_REG_2, BPF_REG_0, offsetof(struct bpf_sock, dst_port)),
+	BPF_LDX_MEM(BPF_B, BPF_REG_2, BPF_REG_0, offsetof(struct bpf_sock, dst_port) + 1),
+	BPF_MOV64_IMM(BPF_REG_0, 0),
+	BPF_EXIT_INSN(),
+	},
+	.prog_type = BPF_PROG_TYPE_CGROUP_SKB,
+	.result = ACCEPT,
+},
+{
+	"sk_fullsock(skb->sk): sk->dst_port [byte load] (invalid)",
+	.insns = {
+	BPF_LDX_MEM(BPF_DW, BPF_REG_1, BPF_REG_1, offsetof(struct __sk_buff, sk)),
+	BPF_JMP_IMM(BPF_JNE, BPF_REG_1, 0, 2),
+	BPF_MOV64_IMM(BPF_REG_0, 0),
+	BPF_EXIT_INSN(),
+	BPF_EMIT_CALL(BPF_FUNC_sk_fullsock),
+	BPF_JMP_IMM(BPF_JNE, BPF_REG_0, 0, 2),
+	BPF_MOV64_IMM(BPF_REG_0, 0),
+	BPF_EXIT_INSN(),
+	BPF_LDX_MEM(BPF_B, BPF_REG_0, BPF_REG_0, offsetof(struct bpf_sock, dst_port) + 2),
+	BPF_MOV64_IMM(BPF_REG_0, 0),
+	BPF_EXIT_INSN(),
+	},
+	.prog_type = BPF_PROG_TYPE_CGROUP_SKB,
+	.result = REJECT,
+	.errstr = "invalid sock access",
+},
+{
+	"sk_fullsock(skb->sk): past sk->dst_port [half load] (invalid)",
 	.insns = {
 	BPF_LDX_MEM(BPF_DW, BPF_REG_1, BPF_REG_1, offsetof(struct __sk_buff, sk)),
 	BPF_JMP_IMM(BPF_JNE, BPF_REG_1, 0, 2),
@@ -149,7 +224,7 @@
 	BPF_JMP_IMM(BPF_JNE, BPF_REG_0, 0, 2),
 	BPF_MOV64_IMM(BPF_REG_0, 0),
 	BPF_EXIT_INSN(),
-	BPF_LDX_MEM(BPF_B, BPF_REG_0, BPF_REG_0, offsetof(struct bpf_sock, dst_port) + 1),
+	BPF_LDX_MEM(BPF_H, BPF_REG_0, BPF_REG_0, offsetofend(struct bpf_sock, dst_port)),
 	BPF_MOV64_IMM(BPF_REG_0, 0),
 	BPF_EXIT_INSN(),
 	},