[PATCH bpf-next 1/2] net: Parse IPv6 ext headers from TCP sock_ops

[PATCH bpf-next 1/2] net: Parse IPv6 ext headers from TCP sock_ops

[Mathieu Jadin]

Add a flag that, if set, triggers the call of eBPF program for each
packet holding an IPv6 extension header. Also add a sock_ops operator
that identifies such call.

This change uses skb_data and skb_data_end introduced for TCP options'
parsing but these pointer cover the IPv6 header and its extension
headers.

For instance, this change allows to read an eBPF sock_ops program to
read complex Segment Routing Headers carrying complex messages in TLV or
observing its intermediate segments as soon as they are received.

Signed-off-by: Mathieu Jadin <mathjadin@gmail.com>
---
 include/uapi/linux/bpf.h       | 26 +++++++++++++++++++++++++-
 net/ipv6/tcp_ipv6.c            | 26 ++++++++++++++++++++++++--
 tools/include/uapi/linux/bpf.h | 26 +++++++++++++++++++++++++-
 3 files changed, 74 insertions(+), 4 deletions(-)

diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
index 6297eafdc40f..79968e57b0b0 100644
--- a/include/uapi/linux/bpf.h
+++ b/include/uapi/linux/bpf.h
@@ -5822,6 +5822,10 @@ struct bpf_sock_ops {
 	 *					the 3WHS.
 	 *
 	 * bpf_load_hdr_opt() can also be used to read a particular option.
+	 *
+	 * Under sock_ops->op ==  BPF_SOCK_OPS_PARSE_IP6_HDR_CB,
+	 * [skb_data, skb_data_end] covers the whole IPv6 header
+	 * with its extension headers.
 	 */
 	__bpf_md_ptr(void *, skb_data);
 	__bpf_md_ptr(void *, skb_data_end);
@@ -5890,8 +5894,15 @@ enum {
 	 * options first before the BPF program does.
 	 */
 	BPF_SOCK_OPS_WRITE_HDR_OPT_CB_FLAG = (1<<6),
+	/* Call bpf for all received IPv6 extension headers.  The bpf prog will
+	 * be called under sock_ops->op == BPF_SOCK_OPS_PARSE_IPV6_HDR_CB and
+	 * will be able to parse the IPv6 header and its extension headers.
+	 *
+	 * The bpf prog will usually turn this off in the common cases.
+	 */
+	BPF_SOCK_OPS_PARSE_IPV6_HDR_CB_FLAG = (1<<7),
 /* Mask of all currently supported cb flags */
-	BPF_SOCK_OPS_ALL_CB_FLAGS       = 0x7F,
+	BPF_SOCK_OPS_ALL_CB_FLAGS       = 0xFF,
 };
 
 /* List of known BPF sock_ops operators.
@@ -6004,6 +6015,19 @@ enum {
 					 * by the kernel or the
 					 * earlier bpf-progs.
 					 */
+	BPF_SOCK_OPS_PARSE_IPV6_HDR_CB,	/* Parse the IPv6 extension
+					 * header option.
+					 * It will be called to handle
+					 * the packets received at
+					 * an already established
+					 * connection with an extension
+					 * header.
+					 *
+					 * sock_ops->skb_data:
+					 * Referring to the received skb.
+					 * It covers the IPv6 header and
+					 * its extension headers only.
+					 */
 };
 
 /* List of TCP states. There is a build check in net/ipv4/tcp.c to detect
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index 3b7d6ede1364..20c83c089ebf 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -1471,7 +1471,7 @@ int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
 {
 	struct ipv6_pinfo *np = tcp_inet6_sk(sk);
 	struct sk_buff *opt_skb = NULL;
-	struct tcp_sock *tp;
+	struct tcp_sock *tp = tcp_sk(sk);
 
 	/* Imagine: socket is IPv6. IPv4 packet arrives,
 	   goes to IPv4 receive handler and backlogged.
@@ -1519,6 +1519,29 @@ int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
 			}
 		}
 
+		/* Call ebpf on packets with extension headers */
+		if (BPF_SOCK_OPS_TEST_FLAG(tp, BPF_SOCK_OPS_PARSE_IPV6_HDR_CB_FLAG) &&
+		    ipv6_hdr(skb)->nexthdr != IPPROTO_TCP) {
+			struct bpf_sock_ops_kern sock_ops;
+			void *old_data_ptr;
+
+			memset(&sock_ops, 0,
+			       offsetof(struct bpf_sock_ops_kern, temp));
+			if (sk_fullsock(sk)) {
+				sock_ops.is_fullsock = 1;
+				sock_owned_by_me(sk);
+			}
+			sock_ops.op = BPF_SOCK_OPS_PARSE_IPV6_HDR_CB;
+			sock_ops.sk = sk;
+			sock_ops.skb = skb;
+			/* Temporary use the network header as skb data */
+			sock_ops.skb_data_end = skb_transport_header(skb);
+			old_data_ptr = skb->data;
+			skb->data = skb_network_header(skb);
+			BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
+			skb->data = old_data_ptr;
+		}
+
 		tcp_rcv_established(sk, skb);
 		if (opt_skb)
 			goto ipv6_pktoptions;
@@ -1572,7 +1595,6 @@ int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
 	   3. socket is not in passive state.
 	   4. Finally, it really contains options, which user wants to receive.
 	 */
-	tp = tcp_sk(sk);
 	if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
 	    !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
 		if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h
index 6297eafdc40f..79968e57b0b0 100644
--- a/tools/include/uapi/linux/bpf.h
+++ b/tools/include/uapi/linux/bpf.h
@@ -5822,6 +5822,10 @@ struct bpf_sock_ops {
 	 *					the 3WHS.
 	 *
 	 * bpf_load_hdr_opt() can also be used to read a particular option.
+	 *
+	 * Under sock_ops->op ==  BPF_SOCK_OPS_PARSE_IP6_HDR_CB,
+	 * [skb_data, skb_data_end] covers the whole IPv6 header
+	 * with its extension headers.
 	 */
 	__bpf_md_ptr(void *, skb_data);
 	__bpf_md_ptr(void *, skb_data_end);
@@ -5890,8 +5894,15 @@ enum {
 	 * options first before the BPF program does.
 	 */
 	BPF_SOCK_OPS_WRITE_HDR_OPT_CB_FLAG = (1<<6),
+	/* Call bpf for all received IPv6 extension headers.  The bpf prog will
+	 * be called under sock_ops->op == BPF_SOCK_OPS_PARSE_IPV6_HDR_CB and
+	 * will be able to parse the IPv6 header and its extension headers.
+	 *
+	 * The bpf prog will usually turn this off in the common cases.
+	 */
+	BPF_SOCK_OPS_PARSE_IPV6_HDR_CB_FLAG = (1<<7),
 /* Mask of all currently supported cb flags */
-	BPF_SOCK_OPS_ALL_CB_FLAGS       = 0x7F,
+	BPF_SOCK_OPS_ALL_CB_FLAGS       = 0xFF,
 };
 
 /* List of known BPF sock_ops operators.
@@ -6004,6 +6015,19 @@ enum {
 					 * by the kernel or the
 					 * earlier bpf-progs.
 					 */
+	BPF_SOCK_OPS_PARSE_IPV6_HDR_CB,	/* Parse the IPv6 extension
+					 * header option.
+					 * It will be called to handle
+					 * the packets received at
+					 * an already established
+					 * connection with an extension
+					 * header.
+					 *
+					 * sock_ops->skb_data:
+					 * Referring to the received skb.
+					 * It covers the IPv6 header and
+					 * its extension headers only.
+					 */
 };
 
 /* List of TCP states. There is a build check in net/ipv4/tcp.c to detect
-- 
2.32.0

[PATCH bpf-next 2/2] selftests/bpf: Test for IPv6 ext header parsing

[Mathieu Jadin]

This test creates a client and a server exchanging a single byte
with a Segment Routing Header and the eBPF program saves
the inner segment in a sk_storage. The test program checks that
the segment is correct.

Signed-off-by: Mathieu Jadin <mathjadin@gmail.com>
---
 .../bpf/prog_tests/tcp_ipv6_exthdr_srh.c      | 171 ++++++++++++++++++
 .../selftests/bpf/progs/tcp_ipv6_exthdr_srh.c |  78 ++++++++
 2 files changed, 249 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/tcp_ipv6_exthdr_srh.c
 create mode 100644 tools/testing/selftests/bpf/progs/tcp_ipv6_exthdr_srh.c

diff --git a/tools/testing/selftests/bpf/prog_tests/tcp_ipv6_exthdr_srh.c b/tools/testing/selftests/bpf/prog_tests/tcp_ipv6_exthdr_srh.c
new file mode 100644
index 000000000000..70f7ee230975
--- /dev/null
+++ b/tools/testing/selftests/bpf/prog_tests/tcp_ipv6_exthdr_srh.c
@@ -0,0 +1,171 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <test_progs.h>
+#include <linux/seg6.h>
+#include "cgroup_helpers.h"
+#include "network_helpers.h"
+
+struct tcp_srh_storage {
+	struct in6_addr inner_segment;
+};
+
+static void send_byte(int fd)
+{
+	char b = 0x55;
+
+	if (CHECK_FAIL(send(fd, &b, sizeof(b), 0) != 1))
+		perror("Failed to send single byte");
+}
+
+static int verify_srh(int map_fd, int server_fd, struct ipv6_sr_hdr *client_srh)
+{
+	int err = 0;
+	struct tcp_srh_storage val;
+
+	if (CHECK_FAIL(bpf_map_lookup_elem(map_fd, &server_fd, &val) < 0)) {
+		perror("Failed to read socket storage");
+		return -1;
+	}
+
+	if (memcmp(&val.inner_segment, &client_srh->segments[1],
+		   sizeof(struct in6_addr))) {
+		log_err("The inner segment of the received SRH differs from the sent one");
+		err++;
+	}
+
+	return err;
+}
+
+static int run_test(int cgroup_fd, int listen_fd)
+{
+	struct bpf_prog_load_attr attr = {
+		.prog_type = BPF_PROG_TYPE_SOCK_OPS,
+		.file = "./tcp_ipv6_exthdr_srh.o",
+		.expected_attach_type = BPF_CGROUP_SOCK_OPS,
+	};
+	size_t srh_size = sizeof(struct ipv6_sr_hdr) +
+		2 * sizeof(struct in6_addr);
+	struct ipv6_sr_hdr *client_srh;
+	struct bpf_object *obj;
+	struct bpf_map *map;
+	struct timeval tv;
+	int client_fd;
+	int server_fd;
+	int prog_fd;
+	int map_fd;
+	char byte;
+	int err;
+
+	err = bpf_prog_load_xattr(&attr, &obj, &prog_fd);
+	if (err) {
+		log_err("Failed to load BPF object");
+		return -1;
+	}
+
+	map = bpf_object__next_map(obj, NULL);
+	map_fd = bpf_map__fd(map);
+
+	err = bpf_prog_attach(prog_fd, cgroup_fd, BPF_CGROUP_SOCK_OPS, 0);
+	if (err) {
+		log_err("Failed to attach BPF program");
+		goto close_bpf_object;
+	}
+
+	client_fd = connect_to_fd(listen_fd, 0);
+	if (client_fd < 0) {
+		err = -1;
+		goto close_bpf_object;
+	}
+
+	server_fd = accept(listen_fd, NULL, 0);
+	if (server_fd < 0) {
+		err = -1;
+		goto close_client_fd;
+	}
+
+	/* Set an SRH with ::1 as an intermediate segment on the client */
+
+	client_srh = calloc(1, srh_size);
+	if (!client_srh) {
+		log_err("Failed to create the SRH to send");
+		goto close_server_fd;
+	}
+	client_srh->type = IPV6_SRCRT_TYPE_4;
+	// We do not count the first 8 bytes (RFC 8200 Section 4.4)
+	client_srh->hdrlen = (2 * sizeof(struct in6_addr)) >> 3;
+	client_srh->segments_left = 1;
+	client_srh->first_segment = 1;
+	// client_srh->segments[0] is set by the kernel
+	memcpy(&client_srh->segments[1], &in6addr_loopback,
+	       sizeof(struct in6_addr));
+
+	if (setsockopt(client_fd, SOL_IPV6, IPV6_RTHDR, client_srh,
+		       srh_size)) {
+		log_err("Failed to set the SRH on the client");
+		goto free_srh;
+	}
+
+	/* Send traffic with this SRH
+	 * and check its parsing on the server side
+	 */
+
+	tv.tv_sec = 1;
+	tv.tv_usec = 0;
+	if (setsockopt(server_fd, SOL_SOCKET, SO_RCVTIMEO, (const char *)&tv,
+		       sizeof(tv))) {
+		log_err("Failed to set the receive timeout on the server");
+		err = -1;
+		goto free_srh;
+	}
+
+	send_byte(client_fd);
+	if (recv(server_fd, &byte, 1, 0) != 1) {
+		log_err("Failed to get the byte under one second on the server 2");
+		err = -1;
+		goto free_srh;
+	}
+
+	err += verify_srh(map_fd, server_fd, client_srh);
+
+free_srh:
+	free(client_srh);
+close_server_fd:
+	close(server_fd);
+close_client_fd:
+	close(client_fd);
+close_bpf_object:
+	bpf_object__close(obj);
+	return err;
+}
+
+void test_tcp_ipv6_exthdr_srh(void)
+{
+	int server_fd, cgroup_fd;
+
+	cgroup_fd = test__join_cgroup("/tcp_ipv6_exthdr_srh");
+	if (CHECK_FAIL(cgroup_fd < 0))
+		return;
+
+	server_fd = start_server(AF_INET6, SOCK_STREAM, "::1", 0, 0);
+	if (CHECK_FAIL(server_fd < 0))
+		goto close_cgroup_fd;
+
+	if (CHECK_FAIL(system("sysctl net.ipv6.conf.all.seg6_enabled=1")))
+		goto close_server;
+
+	if (CHECK_FAIL(system("sysctl net.ipv6.conf.lo.seg6_enabled=1")))
+		goto reset_sysctl;
+
+	CHECK_FAIL(run_test(cgroup_fd, server_fd));
+
+	if (CHECK_FAIL(system("sysctl net.ipv6.conf.lo.seg6_enabled=0")))
+		log_err("Cannot reset sysctl net.ipv6.conf.lo.seg6_enabled to 0");
+
+reset_sysctl:
+	if (CHECK_FAIL(system("sysctl net.ipv6.conf.all.seg6_enabled=0")))
+		log_err("Cannot reset sysctl net.ipv6.conf.all.seg6_enabled to 0");
+
+close_server:
+	close(server_fd);
+close_cgroup_fd:
+	close(cgroup_fd);
+}
diff --git a/tools/testing/selftests/bpf/progs/tcp_ipv6_exthdr_srh.c b/tools/testing/selftests/bpf/progs/tcp_ipv6_exthdr_srh.c
new file mode 100644
index 000000000000..276bda8bbecb
--- /dev/null
+++ b/tools/testing/selftests/bpf/progs/tcp_ipv6_exthdr_srh.c
@@ -0,0 +1,78 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/types.h>
+#include <bpf/bpf_helpers.h>
+#include <linux/in6.h>
+#include <linux/ipv6.h>
+#include <linux/seg6.h>
+#include <linux/bpf.h>
+
+char _license[] SEC("license") = "GPL";
+
+#define NEXTHDR_ROUTING	43
+
+struct tcp_srh_storage {
+	struct in6_addr inner_segment;
+};
+
+struct {
+	__uint(type, BPF_MAP_TYPE_SK_STORAGE);
+	__uint(map_flags, BPF_F_NO_PREALLOC);
+	__type(key, int);
+	__type(value, struct tcp_srh_storage);
+} socket_storage_map SEC(".maps");
+
+/* Check the header received from the active side */
+static int read_incoming_srh(struct bpf_sock_ops *skops,
+			     struct tcp_srh_storage *storage)
+{
+	__u32 seg_size = 2 * sizeof(struct in6_addr);
+	struct ipv6_sr_hdr *srh;
+	struct ipv6hdr *ip6;
+	void *seg_list;
+	int ret = 1;
+
+	ip6 = (struct ipv6hdr *)skops->skb_data;
+	if (ip6 + 1 <= skops->skb_data_end && ip6->nexthdr == NEXTHDR_ROUTING) {
+		srh = (struct ipv6_sr_hdr *)(ip6 + 1);
+		if (srh + 1 <= skops->skb_data_end) {
+			if (srh->type != IPV6_SRCRT_TYPE_4)
+				return ret;
+
+			seg_list = (void *)(srh + 1);
+			if (seg_list + seg_size <= skops->skb_data_end) {
+				// This is an SRH with at least 2 segments
+				storage->inner_segment = srh->segments[1];
+				ret = 0;
+			}
+		}
+	}
+
+	return ret;
+}
+
+SEC("sockops")
+int srh_read(struct bpf_sock_ops *skops)
+{
+	struct tcp_srh_storage *storage;
+	int true_val = 1;
+
+	if (!skops->sk)
+		return 1;
+
+	storage = bpf_sk_storage_get(&socket_storage_map, skops->sk, 0,
+				     BPF_SK_STORAGE_GET_F_CREATE);
+	if (!storage)
+		return 1;
+
+	switch (skops->op) {
+	case BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB:
+	case BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB:
+		bpf_sock_ops_cb_flags_set(skops, skops->bpf_sock_ops_cb_flags |
+				  BPF_SOCK_OPS_PARSE_IPV6_HDR_CB_FLAG);
+		break;
+	case BPF_SOCK_OPS_PARSE_IPV6_HDR_CB:
+		return read_incoming_srh(skops, storage);
+	}
+
+	return 0;
+}
-- 
2.32.0