Skip to contentSkip to conversation

add skc_to_mptcp_sock

PATCH Superseded Geliang Tang 3 participants 8 messages Mar 25, 2022

Subsystems (7)

v13 of 14 Superseded Latest is v14 →

lore
Share & tools
lore
#0 lore raw

v13 cover add skc_to_mptcp_sock

v13:
patch 1:
As Mat mentioned in the "BPF packet scheduler" series, BPF packet
scheduler depends on CONFIG_BPF_SYSCALL. In fact, the series depends on CONFIG_BPF_SYSCALL too.
- change "CONFIG_BPF" to "CONFIG_BPF_SYSCALL" in Makefile
- change "ifdef CONFIG_MPTCP" in include/linux/bpf.h to:

    "#if defined(CONFIG_MPTCP) && defined(CONFIG_BPF_SYSCALL)"

- drop "#ifdef CONFIG_MPTCP" in net/core/filter.c
patch 2:
Address to Matt's comments in v12:
- use bpf_object__find_program_by_name, bpf_object__find_map_by_name
- use -EIO
- drop close_client_fd label
- add more CHECK_FAIL

v12:
patch 1:
- merged the squash-to patch for v11.
- rebased to export/20220324T054815.
patch 2:
- fix the libbpf 0.8 build error.
patch 4:
- use fgets instead of fread to get null terminated string.

v11:
- merge "check more values from mptcp_sock" series into this series.
- patch 2:
- rebased.
- patch 3:
- rebased.
- add 'Copyright' entry as Matt suggested in v9.
- patch 4:
- add test__start_subtest.
- add some 'Copyright' entries.
- patch 5:
- change monitor_log_path to static
- drop bpf_trace_printk()
- patch 6:
- use TCP_CA_NAME_MAX instead of strlen(val.ca_name) in strncmp().
- use 'sysctl -b net.ipv4.tcp_congestion_control' to get ca_name.

v10:
- merge the squash-to patch.
- update patch 3 as Matt suggested.
- add sync() in get_msk_token() before read().
- add a comment for get_msk_token().
- update subjects and commit logs.

v9:
- update progs/mptcp.c in patch 4 and 5

v8:
- update as Matt suggested.

v7:
- parse msk token from the output of 'ip mptcp monitor'.
- add Nicolas and Matt's SoB tags.

v6:
- add skc_to_mptcp_sock helper and test

RESEND:
- fix the CI Build Failure.
v5:
- fix incorrect token value
- verify the token in selftest

v4:
- define bpf_mptcp_sock_proto as a static function, no longer export it in linux/bpf.h

v3:
- use RET_PTR_TO_BTF_ID_OR_NULL instead of RET_PTR_TO_MPTCP_SOCK_OR_NULL
- add a new bpf_id BTF_SOCK_TYPE_MPTCP

v2:
- keep RET_PTR_TO_MPTCP_SOCK_OR_NULL. If we use RET_PTR_TO_BTF_ID_OR_NULL
instead of RET_PTR_TO_MPTCP_SOCK_OR_NULL as Alexei suggested, the "userspace" tests developed by Nicolas will break.

Geliang Tang (5):
bpf: add bpf_skc_to_mptcp_sock_proto
Squash to "selftests: bpf: add MPTCP test base"
selftests: bpf: test bpf_skc_to_mptcp_sock
selftests: bpf: verify ca_name of struct mptcp_sock
selftests: bpf: verify first subflow of mptcp_sock

 include/linux/bpf.h                           |   9 +
 include/linux/btf_ids.h                       |   3 +-
 include/uapi/linux/bpf.h                      |   7 +
 net/core/filter.c                             |  16 ++
 net/mptcp/Makefile                            |   2 +
 net/mptcp/bpf.c                               |  22 ++
 scripts/bpf_doc.py                            |   2 +
 tools/include/uapi/linux/bpf.h                |   7 +
 tools/testing/selftests/bpf/bpf_tcp_helpers.h |  12 +
 .../testing/selftests/bpf/prog_tests/mptcp.c  | 209 +++++++++++++++---
 tools/testing/selftests/bpf/progs/mptcp.c     |  44 +++-
 11 files changed, 289 insertions(+), 44 deletions(-)
 create mode 100644 net/mptcp/bpf.c
Signature · 1 line
2.34.1
#1 lore raw

PATCH v13 1/5 bpf: add bpf_skc_to_mptcp_sock_proto

This patch implemented a new bpf_func_proto bpf_skc_to_mptcp_sock_proto. Defined a new bpf_id BTF_SOCK_TYPE_MPTCP, implemented the helper bpf_skc_to_mptcp_sock, and added a new helper bpf_mptcp_sock_from_subflow to get struct mptcp_sock from a given subflow socket.

---
 include/linux/bpf.h            |  9 +++++++++
 include/linux/btf_ids.h        |  3 ++-
 include/uapi/linux/bpf.h       |  7 +++++++
 net/core/filter.c              | 16 ++++++++++++++++
 net/mptcp/Makefile             |  2 ++
 net/mptcp/bpf.c                | 22 ++++++++++++++++++++++
 scripts/bpf_doc.py             |  2 ++
 tools/include/uapi/linux/bpf.h |  7 +++++++
 8 files changed, 67 insertions(+), 1 deletion(-)
 create mode 100644 net/mptcp/bpf.c
include/linux/bpf.h +9 -0
index bdb5298735ce..117ae3f9451e 100644
@@ -2372,6 +2372,15 @@ static inline u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
 }
 #endif /* CONFIG_INET */
 
+#if defined(CONFIG_MPTCP) && defined(CONFIG_BPF_SYSCALL)
+struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk);
+#else
+static inline struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk)
+{
+	return NULL;
+}
+#endif /* defined(CONFIG_MPTCP) && defined(CONFIG_BPF_SYSCALL) */
+
 enum bpf_text_poke_type {
 	BPF_MOD_CALL,
 	BPF_MOD_JUMP,
include/linux/btf_ids.h +2 -1
index bc5d9cc34e4c..335a19092368 100644
@@ -178,7 +178,8 @@ extern struct btf_id_set name;
 	BTF_SOCK_TYPE(BTF_SOCK_TYPE_TCP6, tcp6_sock)			\
 	BTF_SOCK_TYPE(BTF_SOCK_TYPE_UDP, udp_sock)			\
 	BTF_SOCK_TYPE(BTF_SOCK_TYPE_UDP6, udp6_sock)			\
-	BTF_SOCK_TYPE(BTF_SOCK_TYPE_UNIX, unix_sock)
+	BTF_SOCK_TYPE(BTF_SOCK_TYPE_UNIX, unix_sock)			\
+	BTF_SOCK_TYPE(BTF_SOCK_TYPE_MPTCP, mptcp_sock)
 
 enum {
 #define BTF_SOCK_TYPE(name, str) name,
include/uapi/linux/bpf.h +7 -0
index 9ef1f3e1c22f..785f2cb15495 100644
@@ -5143,6 +5143,12 @@ union bpf_attr {
  *		The **hash_algo** is returned on success,
  *		**-EOPNOTSUP** if the hash calculation failed or **-EINVAL** if
  *		invalid arguments are passed.
+ *
+ * struct mptcp_sock *bpf_skc_to_mptcp_sock(void *sk)
+ *	Description
+ *		Dynamically cast a *sk* pointer to a *mptcp_sock* pointer.
+ *	Return
+ *		*sk* if casting is valid, or **NULL** otherwise.
  */
 #define __BPF_FUNC_MAPPER(FN)		\
 	FN(unspec),			\
@@ -5339,6 +5345,7 @@ union bpf_attr {
 	FN(copy_from_user_task),	\
 	FN(skb_set_tstamp),		\
 	FN(ima_file_hash),		\
+	FN(skc_to_mptcp_sock),		\
 	/* */
 
 /* integer value in 'imm' field of BPF_CALL instruction selects which helper
net/core/filter.c +16 -0
index 258821e58ec6..53f353b81a1f 100644
@@ -11270,6 +11270,19 @@ const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
 	.ret_btf_id		= &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
 };
 
+BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
+{
+	return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
+}
+
+static const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
+	.func		= bpf_skc_to_mptcp_sock,
+	.gpl_only	= false,
+	.ret_type	= RET_PTR_TO_BTF_ID_OR_NULL,
+	.arg1_type	= ARG_PTR_TO_SOCK_COMMON,
+	.ret_btf_id	= &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
+};
+
 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
 {
 	return (unsigned long)sock_from_file(file);
@@ -11312,6 +11325,9 @@ bpf_sk_base_func_proto(enum bpf_func_id func_id)
 	case BPF_FUNC_skc_to_unix_sock:
 		func = &bpf_skc_to_unix_sock_proto;
 		break;
+	case BPF_FUNC_skc_to_mptcp_sock:
+		func = &bpf_skc_to_mptcp_sock_proto;
+		break;
 	case BPF_FUNC_ktime_get_coarse_ns:
 		return &bpf_ktime_get_coarse_ns_proto;
 	default:
net/mptcp/Makefile +2 -0
index e54daceac58b..99dddf08ca73 100644
@@ -10,3 +10,5 @@ obj-$(CONFIG_INET_MPTCP_DIAG) += mptcp_diag.o
 mptcp_crypto_test-objs := crypto_test.o
 mptcp_token_test-objs := token_test.o
 obj-$(CONFIG_MPTCP_KUNIT_TEST) += mptcp_crypto_test.o mptcp_token_test.o
+
+obj-$(CONFIG_BPF_SYSCALL) += bpf.o
net/mptcp/bpf.c new file +22 -0
new file mode 100644
index 000000000000..535602ba2582
@@ -0,0 +1,22 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Multipath TCP
+ *
+ * Copyright (c) 2020, Tessares SA.
+ * Copyright (c) 2022, SUSE.
+ *
+ * Author: Nicolas Rybowski <[email protected]>
+ */
+
+#define pr_fmt(fmt) "MPTCP: " fmt
+
+#include <linux/bpf.h>
+#include "protocol.h"
+
+struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk)
+{
+	if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP && sk_is_mptcp(sk))
+		return mptcp_sk(mptcp_subflow_ctx(sk)->conn);
+
+	return NULL;
+}
+EXPORT_SYMBOL(bpf_mptcp_sock_from_subflow);
scripts/bpf_doc.py +2 -0
index 096625242475..0e5a9e69ae59 100755
@@ -623,6 +623,7 @@ class PrinterHelpers(Printer):
             'struct udp6_sock',
             'struct unix_sock',
             'struct task_struct',
+            'struct mptcp_sock',
 
             'struct __sk_buff',
             'struct sk_msg_md',
@@ -682,6 +683,7 @@ class PrinterHelpers(Printer):
             'struct socket',
             'struct file',
             'struct bpf_timer',
+            'struct mptcp_sock',
     }
     mapped_types = {
             'u8': '__u8',
tools/include/uapi/linux/bpf.h +7 -0
index 9ef1f3e1c22f..785f2cb15495 100644
@@ -5143,6 +5143,12 @@ union bpf_attr {
  *		The **hash_algo** is returned on success,
  *		**-EOPNOTSUP** if the hash calculation failed or **-EINVAL** if
  *		invalid arguments are passed.
+ *
+ * struct mptcp_sock *bpf_skc_to_mptcp_sock(void *sk)
+ *	Description
+ *		Dynamically cast a *sk* pointer to a *mptcp_sock* pointer.
+ *	Return
+ *		*sk* if casting is valid, or **NULL** otherwise.
  */
 #define __BPF_FUNC_MAPPER(FN)		\
 	FN(unspec),			\
@@ -5339,6 +5345,7 @@ union bpf_attr {
 	FN(copy_from_user_task),	\
 	FN(skb_set_tstamp),		\
 	FN(ima_file_hash),		\
+	FN(skc_to_mptcp_sock),		\
 	/* */
 
 /* integer value in 'imm' field of BPF_CALL instruction selects which helper
Signature · 1 line
2.34.1
#3 lore raw

PATCH v13 2/5 Squash to "selftests: bpf: add MPTCP test base"

In reply to Geliang Tang

fix libbpf 0.8 build break:

prog_tests/mptcp.c:176:2: error: 'bpf_prog_load_xattr' is deprecated: libbpf v0.8+: use bpf_object__open() and bpf_object__load() instead [-Werror=deprecated-declarations]

Some cleanups.

---
 .../testing/selftests/bpf/prog_tests/mptcp.c  | 77 +++++++++++--------
 tools/testing/selftests/bpf/progs/mptcp.c     | 16 ++--
 2 files changed, 56 insertions(+), 37 deletions(-)
tools/testing/selftests/bpf/prog_tests/mptcp.c +47 -30
index 04aef0f147dc..bfd521e17711 100644
@@ -1,4 +1,6 @@
 // SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2020, Tessares SA. */
+
 #include <test_progs.h>
 #include "cgroup_helpers.h"
 #include "network_helpers.h"
@@ -13,12 +15,6 @@ static int verify_sk(int map_fd, int client_fd, const char *msg, __u32 is_mptcp)
 	int err = 0, cfd = client_fd;
 	struct mptcp_storage val;
 
-	/* Currently there is no easy way to get back the subflow sk from the MPTCP
-	 * sk, thus we cannot access here the sk_storage associated to the subflow
-	 * sk. Also, there is no sk_storage associated with the MPTCP sk since it
-	 * does not trigger sockops events.
-	 * We silently pass this situation at the moment.
-	 */
 	if (is_mptcp == 1)
 		return 0;
 
@@ -28,14 +24,14 @@ static int verify_sk(int map_fd, int client_fd, const char *msg, __u32 is_mptcp)
 	}
 
 	if (val.invoked != 1) {
-		log_err("%s: unexpected invoked count %d != %d",
-			msg, val.invoked, 1);
+		log_err("%s: unexpected invoked count %d != 1",
+			msg, val.invoked);
 		err++;
 	}
 
-	if (val.is_mptcp != is_mptcp) {
-		log_err("%s: unexpected bpf_tcp_sock.is_mptcp %d != %d",
-			msg, val.is_mptcp, is_mptcp);
+	if (val.is_mptcp != 0) {
+		log_err("%s: unexpected bpf_tcp_sock.is_mptcp %d != 0",
+			msg, val.is_mptcp);
 		err++;
 	}
 
@@ -45,49 +41,64 @@ static int verify_sk(int map_fd, int client_fd, const char *msg, __u32 is_mptcp)
 static int run_test(int cgroup_fd, int server_fd, bool is_mptcp)
 {
 	int client_fd, prog_fd, map_fd, err;
+	struct bpf_program *prog;
 	struct bpf_object *obj;
 	struct bpf_map *map;
 
-	struct bpf_prog_load_attr attr = {
-		.prog_type = BPF_PROG_TYPE_SOCK_OPS,
-		.file = "./mptcp.o",
-		.expected_attach_type = BPF_CGROUP_SOCK_OPS,
-	};
+	obj = bpf_object__open("./mptcp.o");
+	if (libbpf_get_error(obj))
+		return -EIO;
 
-	err = bpf_prog_load_xattr(&attr, &obj, &prog_fd);
-	if (err) {
-		log_err("Failed to load BPF object");
-		return -1;
+	err = bpf_object__load(obj);
+	if (CHECK_FAIL(err))
+		goto out;
+
+	prog = bpf_object__find_program_by_name(obj, "_sockops");
+	if (CHECK_FAIL(!prog)) {
+		err = -EIO;
+		goto out;
+	}
+
+	prog_fd = bpf_program__fd(prog);
+	if (CHECK_FAIL(prog_fd < 0)) {
+		err = -EIO;
+		goto out;
+	}
+
+	map = bpf_object__find_map_by_name(obj, "socket_storage_map");
+	if (CHECK_FAIL(!map)) {
+		err = -EIO;
+		goto out;
 	}
 
-	map = bpf_object__next_map(obj, NULL);
 	map_fd = bpf_map__fd(map);
+	if (CHECK_FAIL(map_fd < 0)) {
+		err = -EIO;
+		goto out;
+	}
 
 	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;
-	}
+	if (CHECK_FAIL(err))
+		goto out;
 
 	client_fd = is_mptcp ? connect_to_mptcp_fd(server_fd, 0) :
 			       connect_to_fd(server_fd, 0);
 	if (client_fd < 0) {
-		err = -1;
-		goto close_client_fd;
+		err = -EIO;
+		goto out;
 	}
 
 	err += is_mptcp ? verify_sk(map_fd, client_fd, "MPTCP subflow socket", 1) :
 			  verify_sk(map_fd, client_fd, "plain TCP socket", 0);
 
-close_client_fd:
 	close(client_fd);
 
-close_bpf_object:
+out:
 	bpf_object__close(obj);
 	return err;
 }
 
-void test_mptcp(void)
+void test_base(void)
 {
 	int server_fd, cgroup_fd;
 
@@ -117,3 +128,9 @@ void test_mptcp(void)
 close_cgroup_fd:
 	close(cgroup_fd);
 }
+
+void test_mptcp(void)
+{
+	if (test__start_subtest("base"))
+		test_base();
+}
tools/testing/selftests/bpf/progs/mptcp.c +9 -7
index be5ee8dac2b3..0d65fb889d03 100644
@@ -1,4 +1,6 @@
 // SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2020, Tessares SA. */
+
 #include <linux/bpf.h>
 #include <bpf/bpf_helpers.h>
 
@@ -25,22 +27,22 @@ int _sockops(struct bpf_sock_ops *ctx)
 	int op = (int)ctx->op;
 	struct bpf_sock *sk;
 
+	if (op != BPF_SOCK_OPS_TCP_CONNECT_CB)
+		return 1;
+
 	sk = ctx->sk;
 	if (!sk)
 		return 1;
 
+	tcp_sk = bpf_tcp_sock(sk);
+	if (!tcp_sk)
+		return 1;
+
 	storage = bpf_sk_storage_get(&socket_storage_map, sk, 0,
 				     BPF_SK_STORAGE_GET_F_CREATE);
 	if (!storage)
 		return 1;
 
-	if (op != BPF_SOCK_OPS_TCP_CONNECT_CB)
-		return 1;
-
-	tcp_sk = bpf_tcp_sock(sk);
-	if (!tcp_sk)
-		return 1;
-
 	storage->invoked++;
 	storage->is_mptcp = tcp_sk->is_mptcp;
 
Signature · 1 line
2.34.1
#5 lore raw

PATCH v13 3/5 selftests: bpf: test bpf_skc_to_mptcp_sock

In reply to Geliang Tang

This patch extended the MPTCP test base, to test the new helper bpf_skc_to_mptcp_sock().

Added a new function verify_msk() to verify the msk token, and a new function get_msk_token() to parse the msk token from the output of the command 'ip mptcp monitor'.

---
 tools/testing/selftests/bpf/bpf_tcp_helpers.h |   6 ++
 .../testing/selftests/bpf/prog_tests/mptcp.c  | 102 ++++++++++++++++--
 tools/testing/selftests/bpf/progs/mptcp.c     |  25 ++++-
 3 files changed, 123 insertions(+), 10 deletions(-)
tools/testing/selftests/bpf/bpf_tcp_helpers.h +6 -0
index b1ede6f0b821..9ea6687c8f4d 100644
@@ -83,6 +83,12 @@ struct tcp_sock {
 	__u64	tcp_mstamp;	/* most recent packet received/sent */
 } __attribute__((preserve_access_index));
 
+struct mptcp_sock {
+	struct inet_connection_sock	sk;
+
+	__u32		token;
+} __attribute__((preserve_access_index));
+
 static __always_inline struct inet_connection_sock *inet_csk(const struct sock *sk)
 {
 	return (struct inet_connection_sock *)sk;
tools/testing/selftests/bpf/prog_tests/mptcp.c +96 -6

144 lines not on this page · raw message

tools/testing/selftests/bpf/progs/mptcp.c +21 -4
index 0d65fb889d03..afacea5ad9ca 100644
@@ -3,6 +3,7 @@
 
 #include <linux/bpf.h>
 #include <bpf/bpf_helpers.h>
+#include "bpf_tcp_helpers.h"
 
 char _license[] SEC("license") = "GPL";
 __u32 _version SEC("version") = 1;
@@ -10,6 +11,7 @@ __u32 _version SEC("version") = 1;
 struct mptcp_storage {
 	__u32 invoked;
 	__u32 is_mptcp;
+	__u32 token;
 };
 
 struct {
@@ -24,6 +26,7 @@ int _sockops(struct bpf_sock_ops *ctx)
 {
 	struct mptcp_storage *storage;
 	struct bpf_tcp_sock *tcp_sk;
+	struct mptcp_sock *msk;
 	int op = (int)ctx->op;
 	struct bpf_sock *sk;
 
@@ -38,11 +41,25 @@ int _sockops(struct bpf_sock_ops *ctx)
 	if (!tcp_sk)
 		return 1;
 
-	storage = bpf_sk_storage_get(&socket_storage_map, sk, 0,
-				     BPF_SK_STORAGE_GET_F_CREATE);
-	if (!storage)
-		return 1;
+	if (!tcp_sk->is_mptcp) {
+		storage = bpf_sk_storage_get(&socket_storage_map, sk, 0,
+					     BPF_SK_STORAGE_GET_F_CREATE);
+		if (!storage)
+			return 1;
+
+		storage->token = 0;
+	} else {
+		msk = bpf_skc_to_mptcp_sock(sk);
+		if (!msk)
+			return 1;
+
+		storage = bpf_sk_storage_get(&socket_storage_map, msk, 0,
+					     BPF_SK_STORAGE_GET_F_CREATE);
+		if (!storage)
+			return 1;
 
+		storage->token = msk->token;
+	}
 	storage->invoked++;
 	storage->is_mptcp = tcp_sk->is_mptcp;
 
Signature · 1 line
2.34.1
#6 lore raw

PATCH v13 4/5 selftests: bpf: verify ca_name of struct mptcp_sock

In reply to Geliang Tang

This patch verified another member of struct mptcp_sock, ca_name. Added a new function get_msk_ca_name() to read the sysctl tcp_congestion_control and verified it in verify_msk().

---
 tools/testing/selftests/bpf/bpf_tcp_helpers.h |  5 ++++
 .../testing/selftests/bpf/prog_tests/mptcp.c  | 24 +++++++++++++++++++
 tools/testing/selftests/bpf/progs/mptcp.c     |  4 ++++
 3 files changed, 33 insertions(+)
tools/testing/selftests/bpf/bpf_tcp_helpers.h +5 -0
index 9ea6687c8f4d..bebe382bcc7a 100644
@@ -16,6 +16,10 @@ BPF_PROG(name, args)
 #define SOL_TCP 6
 #endif
 
+#ifndef TCP_CA_NAME_MAX
+#define TCP_CA_NAME_MAX	16
+#endif
+
 #define tcp_jiffies32 ((__u32)bpf_jiffies64())
 
 struct sock_common {
@@ -87,6 +91,7 @@ struct mptcp_sock {
 	struct inet_connection_sock	sk;
 
 	__u32		token;
+	char		ca_name[TCP_CA_NAME_MAX];
 } __attribute__((preserve_access_index));
 
 static __always_inline struct inet_connection_sock *inet_csk(const struct sock *sk)
tools/testing/selftests/bpf/prog_tests/mptcp.c +24 -0
index 3232ace122c9..ea8dfcdb230e 100644
@@ -5,10 +5,15 @@
 #include "cgroup_helpers.h"
 #include "network_helpers.h"
 
+#ifndef TCP_CA_NAME_MAX
+#define TCP_CA_NAME_MAX	16
+#endif
+
 struct mptcp_storage {
 	__u32 invoked;
 	__u32 is_mptcp;
 	__u32 token;
+	char ca_name[TCP_CA_NAME_MAX];
 };
 
 static char monitor_log_path[64];
@@ -79,11 +84,22 @@ static __u32 get_msk_token(void)
 	return token;
 }
 
+void get_msk_ca_name(char ca_name[])
+{
+	FILE *stream = popen("sysctl -b net.ipv4.tcp_congestion_control", "r");
+
+	if (fgets(ca_name, TCP_CA_NAME_MAX, stream) == NULL)
+		log_err("Failed to read ca_name");
+
+	pclose(stream);
+}
+
 static int verify_msk(int map_fd, int client_fd)
 {
 	char *msg = "MPTCP subflow socket";
 	int err = 0, cfd = client_fd;
 	struct mptcp_storage val;
+	char ca_name[TCP_CA_NAME_MAX];
 	__u32 token;
 
 	token = get_msk_token();
@@ -92,6 +108,8 @@ static int verify_msk(int map_fd, int client_fd)
 		return -1;
 	}
 
+	get_msk_ca_name(ca_name);
+
 	if (CHECK_FAIL(bpf_map_lookup_elem(map_fd, &cfd, &val) < 0)) {
 		perror("Failed to read socket storage");
 		return -1;
@@ -115,6 +133,12 @@ static int verify_msk(int map_fd, int client_fd)
 		err++;
 	}
 
+	if (strncmp(val.ca_name, ca_name, TCP_CA_NAME_MAX)) {
+		log_err("Unexpected mptcp_sock.ca_name %s != %s",
+			val.ca_name, ca_name);
+		err++;
+	}
+
 	return err;
 }
 
tools/testing/selftests/bpf/progs/mptcp.c +4 -0
index afacea5ad9ca..6e7d2abf6ce3 100644
@@ -1,6 +1,7 @@
 // SPDX-License-Identifier: GPL-2.0
 /* Copyright (c) 2020, Tessares SA. */
 
+#include <string.h>
 #include <linux/bpf.h>
 #include <bpf/bpf_helpers.h>
 #include "bpf_tcp_helpers.h"
@@ -12,6 +13,7 @@ struct mptcp_storage {
 	__u32 invoked;
 	__u32 is_mptcp;
 	__u32 token;
+	char ca_name[TCP_CA_NAME_MAX];
 };
 
 struct {
@@ -48,6 +50,7 @@ int _sockops(struct bpf_sock_ops *ctx)
 			return 1;
 
 		storage->token = 0;
+		bzero(storage->ca_name, TCP_CA_NAME_MAX);
 	} else {
 		msk = bpf_skc_to_mptcp_sock(sk);
 		if (!msk)
@@ -59,6 +62,7 @@ int _sockops(struct bpf_sock_ops *ctx)
 			return 1;
 
 		storage->token = msk->token;
+		memcpy(storage->ca_name, msk->ca_name, TCP_CA_NAME_MAX);
 	}
 	storage->invoked++;
 	storage->is_mptcp = tcp_sk->is_mptcp;
Signature · 1 line
2.34.1
#7 lore raw

PATCH v13 5/5 selftests: bpf: verify first subflow of mptcp_sock

In reply to Geliang Tang

This patch verified the 'first' struct member of struct mptcp_sock, which pointed to the first subflow of msk. Saved 'sk' in mptcp_storage, and verified it with 'first' in verify_msk().

---
 tools/testing/selftests/bpf/bpf_tcp_helpers.h  | 1 +
 tools/testing/selftests/bpf/prog_tests/mptcp.c | 8 ++++++++
 tools/testing/selftests/bpf/progs/mptcp.c      | 5 +++++
 3 files changed, 14 insertions(+)
tools/testing/selftests/bpf/bpf_tcp_helpers.h +1 -0
index bebe382bcc7a..f92357597e63 100644
@@ -91,6 +91,7 @@ struct mptcp_sock {
 	struct inet_connection_sock	sk;
 
 	__u32		token;
+	struct sock	*first;
 	char		ca_name[TCP_CA_NAME_MAX];
 } __attribute__((preserve_access_index));
 
tools/testing/selftests/bpf/prog_tests/mptcp.c +8 -0
index ea8dfcdb230e..888d9e9b8870 100644
@@ -12,7 +12,9 @@
 struct mptcp_storage {
 	__u32 invoked;
 	__u32 is_mptcp;
+	struct sock *sk;
 	__u32 token;
+	struct sock *first;
 	char ca_name[TCP_CA_NAME_MAX];
 };
 
@@ -133,6 +135,12 @@ static int verify_msk(int map_fd, int client_fd)
 		err++;
 	}
 
+	if (val.first != val.sk) {
+		log_err("Unexpected mptcp_sock.first %p != %p",
+			val.first, val.sk);
+		err++;
+	}
+
 	if (strncmp(val.ca_name, ca_name, TCP_CA_NAME_MAX)) {
 		log_err("Unexpected mptcp_sock.ca_name %s != %s",
 			val.ca_name, ca_name);
tools/testing/selftests/bpf/progs/mptcp.c +5 -0
index 6e7d2abf6ce3..16055f553f29 100644
@@ -12,7 +12,9 @@ __u32 _version SEC("version") = 1;
 struct mptcp_storage {
 	__u32 invoked;
 	__u32 is_mptcp;
+	struct sock *sk;
 	__u32 token;
+	struct sock *first;
 	char ca_name[TCP_CA_NAME_MAX];
 };
 
@@ -51,6 +53,7 @@ int _sockops(struct bpf_sock_ops *ctx)
 
 		storage->token = 0;
 		bzero(storage->ca_name, TCP_CA_NAME_MAX);
+		storage->first = NULL;
 	} else {
 		msk = bpf_skc_to_mptcp_sock(sk);
 		if (!msk)
@@ -63,9 +66,11 @@ int _sockops(struct bpf_sock_ops *ctx)
 
 		storage->token = msk->token;
 		memcpy(storage->ca_name, msk->ca_name, TCP_CA_NAME_MAX);
+		storage->first = msk->first;
 	}
 	storage->invoked++;
 	storage->is_mptcp = tcp_sk->is_mptcp;
+	storage->sk = (struct sock *)sk;
 
 	return 1;
 }
Signature · 1 line
2.34.1
Acked-by on v13 2/5
#4 lore raw

Re: [PATCH mptcp-next v13 2/5] Squash to "selftests: bpf: add MPTCP test base"

In reply to Geliang Tang

Hi Geliang,

Quoted 5 lines · Geliang Tang

fix libbpf 0.8 build break:

prog_tests/mptcp.c:176:2: error: 'bpf_prog_load_xattr' is deprecated: libbpf v0.8+: use bpf_object__open() and bpf_object__load() instead [-Werror=deprecated-declarations]

Some cleanups.

Thank you for the update!

Quoted 13 lines
tools/testing/selftests/bpf/prog_tests/mptcp.c +1 -5
@@ -45,49 +41,64 @@ static int verify_sk(int map_fd, int client_fd, const char *msg, __u32 is_mptcp)
 static int run_test(int cgroup_fd, int server_fd, bool is_mptcp)
 {
 	int client_fd, prog_fd, map_fd, err;
+	struct bpf_program *prog;
 	struct bpf_object *obj;
 	struct bpf_map *map;
 
-	struct bpf_prog_load_attr attr = {
-		.prog_type = BPF_PROG_TYPE_SOCK_OPS,
-		.file = "./mptcp.o",
-		.expected_attach_type = BPF_CGROUP_SOCK_OPS,
-	};

tools/testing/selftests/bpf/prog_tests/mptcp.c static int verify_sk(int map_fd, int client_fd, const char *msg, __u32 is_mptcp) · line 53

Quoted 10 lines
tools/testing/selftests/bpf/prog_tests/mptcp.c +5 -4
+	obj = bpf_object__open("./mptcp.o");
+	if (libbpf_get_error(obj))
+		return -EIO;
 
-	err = bpf_prog_load_xattr(&attr, &obj, &prog_fd);
-	if (err) {
-		log_err("Failed to load BPF object");
-		return -1;
+	err = bpf_object__load(obj);
+	if (CHECK_FAIL(err))

I think it would be nice to add specific messages for each errors but if it is done like in other BPF selftests and it is fine for BTF maintainers, fine for me because at the end, these errors you would get here and below will not be linked to MPTCP code.

Quoted 51 lines
tools/testing/selftests/bpf/prog_tests/mptcp.c +26 -9
+		goto out;
+
+	prog = bpf_object__find_program_by_name(obj, "_sockops");
+	if (CHECK_FAIL(!prog)) {
+		err = -EIO;
+		goto out;
+	}
+
+	prog_fd = bpf_program__fd(prog);
+	if (CHECK_FAIL(prog_fd < 0)) {
+		err = -EIO;
+		goto out;
+	}
+
+	map = bpf_object__find_map_by_name(obj, "socket_storage_map");
+	if (CHECK_FAIL(!map)) {
+		err = -EIO;
+		goto out;
 	}
 
-	map = bpf_object__next_map(obj, NULL);
 	map_fd = bpf_map__fd(map);
+	if (CHECK_FAIL(map_fd < 0)) {
+		err = -EIO;
+		goto out;
+	}
 
 	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;
-	}
+	if (CHECK_FAIL(err))
+		goto out;
 
 	client_fd = is_mptcp ? connect_to_mptcp_fd(server_fd, 0) :
 			       connect_to_fd(server_fd, 0);
 	if (client_fd < 0) {
-		err = -1;
-		goto close_client_fd;
+		err = -EIO;
+		goto out;
 	}
 
 	err += is_mptcp ? verify_sk(map_fd, client_fd, "MPTCP subflow socket", 1) :
 			  verify_sk(map_fd, client_fd, "plain TCP socket", 0);
 
-close_client_fd:
 	close(client_fd);
 
-close_bpf_object:

tools/testing/selftests/bpf/prog_tests/mptcp.c static int verify_sk(int map_fd, int client_fd, const char *msg, __u32 is_mptcp) · line 104

Quoted 10 lines
tools/testing/selftests/bpf/prog_tests/mptcp.c +2 -1
+out:
 	bpf_object__close(obj);
 	return err;
 }
 
-void test_mptcp(void)
+void test_base(void)
 {
 	int server_fd, cgroup_fd;
 

So:

@Mat: OK for you? :)

Cheers,
Matt

Signature · 2 lines
Tessares | Belgium | Hybrid Access Solutions
www.tessares.net
#2 lore raw

Re: [PATCH mptcp-next v13 1/5] bpf: add bpf_skc_to_mptcp_sock_proto

In reply to Geliang Tang
Quoted 22 lines · Geliang Tang

This patch implemented a new bpf_func_proto bpf_skc_to_mptcp_sock_proto. Defined a new bpf_id BTF_SOCK_TYPE_MPTCP, implemented the helper bpf_skc_to_mptcp_sock, and added a new helper bpf_mptcp_sock_from_subflow to get struct mptcp_sock from a given subflow socket.

---

include/linux/bpf.h            |  9 +++++++++
include/linux/btf_ids.h        |  3 ++-
include/uapi/linux/bpf.h       |  7 +++++++
net/core/filter.c              | 16 ++++++++++++++++
net/mptcp/Makefile             |  2 ++
net/mptcp/bpf.c                | 22 ++++++++++++++++++++++
scripts/bpf_doc.py             |  2 ++
tools/include/uapi/linux/bpf.h |  7 +++++++
8 files changed, 67 insertions(+), 1 deletion(-)

create mode 100644 net/mptcp/bpf.c

...

net/mptcp/Makefile obj-$(CONFIG_INET_MPTCP_DIAG) += mptcp_diag.o · line 14

Quoted 10 lines
net/mptcp/Makefile +0 -0
index e54daceac58b..99dddf08ca73 100644
@@ -10,3 +10,5 @@ obj-$(CONFIG_INET_MPTCP_DIAG) += mptcp_diag.o

mptcp_crypto_test-objs := crypto_test.o mptcp_token_test-objs := token_test.o obj-$(CONFIG_MPTCP_KUNIT_TEST) += mptcp_crypto_test.o mptcp_token_test.o

net/mptcp/Makefile +2 -0
+
+obj-$(CONFIG_BPF_SYSCALL) += bpf.o

Ok, I see that net/ipv4/bpf_tcp_ca.c is also included based on CONFIG_BPF_SYSCALL, so it would probably be overkill to add a new CONFIG_MPTCP_BPF_SCHEDULER option. Using CONFIG_BPF_SYSCALL is fine.

I noticed that the TCP CA conditional also depends on CONFIG_BPF_JIT:

ifeq ($(CONFIG_BPF_JIT),y)
obj-$(CONFIG_BPF_SYSCALL) += bpf_tcp_ca.o
endif

Presumably this is to reduce the overhead of BPF code that's called frequently. Should we do the same for the MPTCP scheduler?

--
Mat Martineau
Intel