Skip to contentSkip to conversation

net/smc: Introduce smc_ops

PATCH Superseded D. Wythe 3 participants 13 messages Jan 7, 2025 – Jan 14, 2025

Subsystems (11)

v5 of 7 Superseded Latest is v7 →

lore
Share & tools
lore

D. Wythe

author maintainer
#0 lore raw

v5 cover net/smc: Introduce smc_ops

This patch aims to introduce BPF injection capabilities for SMC and includes a self-test to ensure code stability.

Since the SMC protocol isn't ideal for every situation, especially short-lived ones, most applications can't guarantee the absence of such scenarios. Consequently, applications may need specific strategies to decide whether to use SMC. For example, an application might limit SMC usage to certain IP addresses or ports.

To maintain the principle of transparent replacement, we want applications to remain unaffected even if they need specific SMC strategies. In other words, they should not require recompilation of their code.

Additionally, we need to ensure the scalability of strategy implementation. While using socket options or sysctl might be straightforward, it could complicate future expansions.

Fortunately, BPF addresses these concerns effectively. Users can write their own strategies in eBPF to determine whether to use SMC, and they can easily modify those strategies in the future.

v2:
1. Rename smc_bpf_ops to smc_ops.
2. Change the scope of smc_ops from global to per netns.
3. Directly pass parameters to ops instead of smc_ops_ctx.
4. Remove struct smc_ops_ctx.
5. Remove exports that are no longer needed.

v3:
1. Remove find_ksym_btf_id_by_prefix_kind.
2. Enhance selftest, introduce a complete ops for filtering smc connections based on ip pairs and a realistic topology test to verify it.

v4:
1. Remove unless func: smc_bpf_ops_check_member()
2. Remove unless inline func: smc_ops_find_by_name()
3. Change CONFIG_SMC=y to complete CI testing
4. Change smc_sock to smc_sock___local in test to avoid compiling failed with CONFIG_SMC=y
5. Improve test cases, remove unnecessary timeouts and multi-thread test, using network_helpers to start testing between server and client.
6. Fix issues when the return value of the ops function is neither 0

     nor 1.

v5:
1. Fix incorrect CI config from CONFIG_SMC=Y to CONFIG_SMC=y.

D. Wythe (5):
bpf: export necessary sympols for modules with struct_ops net/smc: Introduce generic hook smc_ops
net/smc: bpf: register smc_ops info struct_ops
libbpf: fix error when st-prefix_ops and ops from differ btf bpf/selftests: add selftest for bpf_smc_ops

 include/net/netns/smc.h                       |   3 +
 include/net/smc.h                             |  51 +++
 kernel/bpf/bpf_struct_ops.c                   |   2 +
 kernel/bpf/syscall.c                          |   1 +
 net/ipv4/tcp_output.c                         |  15 +-
 net/smc/Kconfig                               |  12 +
 net/smc/Makefile                              |   1 +
 net/smc/af_smc.c                              |  10 +
 net/smc/smc_ops.c                             | 130 ++++++
 net/smc/smc_ops.h                             |  30 ++
 net/smc/smc_sysctl.c                          |  95 +++++
 tools/lib/bpf/libbpf.c                        |  25 +-
 tools/testing/selftests/bpf/config            |   4 +
 .../selftests/bpf/prog_tests/test_bpf_smc.c   | 390 ++++++++++++++++++
 tools/testing/selftests/bpf/progs/bpf_smc.c   | 116 ++++++
 15 files changed, 873 insertions(+), 12 deletions(-)
 create mode 100644 net/smc/smc_ops.c
 create mode 100644 net/smc/smc_ops.h
 create mode 100644 tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c
 create mode 100644 tools/testing/selftests/bpf/progs/bpf_smc.c
Signature · 1 line
2.45.0
author maintainer
#1 lore raw

PATCH v5 1/5 bpf: export necessary sympols for modules with struct_ops

Exports three necessary symbols for implementing struct_ops with tristate subsystem.

To hold or release refcnt of struct_ops refcnt by inline funcs bpf_try_module_get and bpf_module_put which use bpf_struct_ops_get(put) conditionally.

And to copy obj name from one to the other with effective checks by bpf_obj_name_cpy.

---
 kernel/bpf/bpf_struct_ops.c | 2 ++
 kernel/bpf/syscall.c        | 1 +
 2 files changed, 3 insertions(+)
kernel/bpf/bpf_struct_ops.c +2 -0
index 606efe32485a..00c212e0ad39 100644
@@ -1119,6 +1119,7 @@ bool bpf_struct_ops_get(const void *kdata)
 	map = __bpf_map_inc_not_zero(&st_map->map, false);
 	return !IS_ERR(map);
 }
+EXPORT_SYMBOL_GPL(bpf_struct_ops_get);
 
 void bpf_struct_ops_put(const void *kdata)
 {
@@ -1130,6 +1131,7 @@ void bpf_struct_ops_put(const void *kdata)
 
 	bpf_map_put(&st_map->map);
 }
+EXPORT_SYMBOL_GPL(bpf_struct_ops_put);
 
 int bpf_struct_ops_supported(const struct bpf_struct_ops *st_ops, u32 moff)
 {
kernel/bpf/syscall.c +1 -0
index 5684e8ce132d..62238ec989dc 100644
@@ -1167,6 +1167,7 @@ int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
 
 	return src - orig_src;
 }
+EXPORT_SYMBOL_GPL(bpf_obj_name_cpy);
 
 int map_check_no_btf(const struct bpf_map *map,
 		     const struct btf *btf,
Signature · 1 line
2.45.0
author maintainer
#2 lore raw

PATCH v5 2/5 net/smc: Introduce generic hook smc_ops

In reply to D. Wythe

The introduction of IPPROTO_SMC enables eBPF programs to determine whether to use SMC based on the context of socket creation, such as network namespaces, PID and comm name, etc.

As a subsequent enhancement, to introduce a new generic hook that allows decisions on whether to use SMC or not at runtime, including but not limited to local/remote IP address or ports.

Moreover, in the future, we can achieve more complex extensions to the protocol stack by extending this ops.

---
 include/net/netns/smc.h |  3 ++
 include/net/smc.h       | 51 ++++++++++++++++++++++
 net/ipv4/tcp_output.c   | 15 +++++--
 net/smc/Kconfig         | 12 ++++++
 net/smc/Makefile        |  1 +
 net/smc/smc_ops.c       | 51 ++++++++++++++++++++++
 net/smc/smc_ops.h       | 25 +++++++++++
 net/smc/smc_sysctl.c    | 95 +++++++++++++++++++++++++++++++++++++++++
 8 files changed, 249 insertions(+), 4 deletions(-)
 create mode 100644 net/smc/smc_ops.c
 create mode 100644 net/smc/smc_ops.h
include/net/netns/smc.h +3 -0
index fc752a50f91b..59d069f56b2d 100644
@@ -17,6 +17,9 @@ struct netns_smc {
 #ifdef CONFIG_SYSCTL
 	struct ctl_table_header		*smc_hdr;
 #endif
+#if IS_ENABLED(CONFIG_SMC_OPS)
+	struct smc_ops __rcu *ops;
+#endif /* CONFIG_SMC_OPS */
 	unsigned int			sysctl_autocorking_size;
 	unsigned int			sysctl_smcr_buf_type;
 	int				sysctl_smcr_testlink_time;
include/net/smc.h +51 -0

61 lines not on this page · raw message

net/ipv4/tcp_output.c +11 -4
index 5485a70b5fe5..7b402167fb4d 100644
@@ -48,6 +48,7 @@
 #include <linux/skbuff_ref.h>
 
 #include <trace/events/tcp.h>
+#include <net/smc.h>
 
 /* Refresh clocks of a TCP socket,
  * ensuring monotically increasing values.
@@ -759,14 +760,17 @@ static void tcp_options_write(struct tcphdr *th, struct tcp_sock *tp,
 	mptcp_options_write(th, ptr, tp, opts);
 }
 
-static void smc_set_option(const struct tcp_sock *tp,
+static void smc_set_option(struct tcp_sock *tp,
 			   struct tcp_out_options *opts,
 			   unsigned int *remaining)
 {
 #if IS_ENABLED(CONFIG_SMC)
 	if (static_branch_unlikely(&tcp_have_smc)) {
 		if (tp->syn_smc) {
-			if (*remaining >= TCPOLEN_EXP_SMC_BASE_ALIGNED) {
+			tp->syn_smc = smc_call_retops(1, &tp->inet_conn.icsk_inet.sk,
+						      set_option, tp);
+			/* re-check syn_smc */
+			if (tp->syn_smc && *remaining >= TCPOLEN_EXP_SMC_BASE_ALIGNED) {
 				opts->options |= OPTION_SMC;
 				*remaining -= TCPOLEN_EXP_SMC_BASE_ALIGNED;
 			}
@@ -776,14 +780,17 @@ static void smc_set_option(const struct tcp_sock *tp,
 }
 
 static void smc_set_option_cond(const struct tcp_sock *tp,
-				const struct inet_request_sock *ireq,
+				struct inet_request_sock *ireq,
 				struct tcp_out_options *opts,
 				unsigned int *remaining)
 {
 #if IS_ENABLED(CONFIG_SMC)
 	if (static_branch_unlikely(&tcp_have_smc)) {
 		if (tp->syn_smc && ireq->smc_ok) {
-			if (*remaining >= TCPOLEN_EXP_SMC_BASE_ALIGNED) {
+			ireq->smc_ok = smc_call_retops(1, &tp->inet_conn.icsk_inet.sk,
+						       set_option_cond, tp, ireq);
+			/* re-check smc_ok */
+			if (ireq->smc_ok && *remaining >= TCPOLEN_EXP_SMC_BASE_ALIGNED) {
 				opts->options |= OPTION_SMC;
 				*remaining -= TCPOLEN_EXP_SMC_BASE_ALIGNED;
 			}
net/smc/Kconfig +12 -0
index ba5e6a2dd2fd..0ee16ec8dceb 100644
@@ -33,3 +33,15 @@ config SMC_LO
 	  of architecture or hardware.
 
 	  if unsure, say N.
+
+config SMC_OPS
+	bool "Generic hook for SMC subsystem"
+	depends on SMC && BPF_SYSCALL
+	default n
+	help
+	  SMC_OPS enables support to register genericfor hook via eBPF programs
+	  for SMC subsystem. eBPF programs offer much greater flexibility
+	  in modifying the behavior of the SMC protocol stack compared
+	  to a complete kernel-based approach.
+
+	  if unsure, say N.
net/smc/Makefile +1 -0
index 60f1c87d5212..5dd706b2927a 100644
@@ -7,3 +7,4 @@ smc-y += smc_cdc.o smc_tx.o smc_rx.o smc_close.o smc_ism.o smc_netlink.o smc_sta
 smc-y += smc_tracepoint.o smc_inet.o
 smc-$(CONFIG_SYSCTL) += smc_sysctl.o
 smc-$(CONFIG_SMC_LO) += smc_loopback.o
+smc-$(CONFIG_SMC_OPS) += smc_ops.o
\ No newline at end of file
net/smc/smc_ops.c new file +51 -0

51 lines not on this page · raw message

net/smc/smc_ops.h new file +25 -0
new file mode 100644
index 000000000000..4b33dc2d65b9
@@ -0,0 +1,25 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ *  Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  Generic hook for SMC subsystem.
+ *
+ *  Copyright IBM Corp. 2016
+ *  Copyright (c) 2024, Alibaba Inc.
+ *
+ *  Author: D. Wythe <[email protected]>
+ */
+
+#ifndef __SMC_OPS
+#define __SMC_OPS
+
+#include <net/smc.h>
+
+/* Find ops by the target name, which required to be a c-string.
+ * Return NULL if no such ops was found,otherwise, return a valid ops.
+ *
+ * Note: Caller MUST ensure it's was invoked under rcu_read_lock.
+ */
+struct smc_ops *smc_ops_find_by_name(const char *name);
+
+#endif /* __SMC_OPS */
net/smc/smc_sysctl.c +95 -0

131 lines not on this page · raw message

Signature · 1 line
2.45.0
maintainer
#3 lore raw

Re: [PATCH bpf-next v5 2/5] net/smc: Introduce generic hook smc_ops

Quoted 30 lines · D. Wythe

The introduction of IPPROTO_SMC enables eBPF programs to determine whether to use SMC based on the context of socket creation, such as network namespaces, PID and comm name, etc.

As a subsequent enhancement, to introduce a new generic hook that allows decisions on whether to use SMC or not at runtime, including but not limited to local/remote IP address or ports.

Moreover, in the future, we can achieve more complex extensions to the protocol stack by extending this ops.

---

include/net/netns/smc.h |  3 ++
include/net/smc.h       | 51 ++++++++++++++++++++++
net/ipv4/tcp_output.c   | 15 +++++--
net/smc/Kconfig         | 12 ++++++
net/smc/Makefile        |  1 +
net/smc/smc_ops.c       | 51 ++++++++++++++++++++++
net/smc/smc_ops.h       | 25 +++++++++++
net/smc/smc_sysctl.c    | 95 +++++++++++++++++++++++++++++++++++++++++
8 files changed, 249 insertions(+), 4 deletions(-)

create mode 100644 net/smc/smc_ops.c
create mode 100644 net/smc/smc_ops.h

include/net/netns/smc.h +0 -0
index fc752a50f91b..59d069f56b2d 100644

include/net/netns/smc.h struct netns_smc { · line 21

Quoted 5 lines
include/net/netns/smc.h +2 -0
@@ -17,6 +17,9 @@ struct netns_smc {
 #ifdef CONFIG_SYSCTL
 	struct ctl_table_header		*smc_hdr;
 #endif
+#if IS_ENABLED(CONFIG_SMC_OPS)
+	struct smc_ops __rcu *ops;
                              ^^^

Align with other fields

Quoted 76 lines
include/net/netns/smc.h +1 -0
+#endif /* CONFIG_SMC_OPS */
 	unsigned int			sysctl_autocorking_size;
 	unsigned int			sysctl_smcr_buf_type;
 	int				sysctl_smcr_testlink_time;
include/net/smc.h +51 -0
index db84e4e35080..326a217001d4 100644
@@ -18,6 +18,8 @@
 #include "linux/ism.h"
 
 struct sock;
+struct tcp_sock;
+struct inet_request_sock;
 
 #define SMC_MAX_PNETID_LEN	16	/* Max. length of PNET id */
 
@@ -97,4 +99,53 @@ struct smcd_dev {
 	u8 going_away : 1;
 };
 
+#define  SMC_OPS_NAME_MAX 16
+
+enum {
+	/* ops can be inherit from init_net */
+	SMC_OPS_FLAG_INHERITABLE = 0x1,
+
+	SMC_OPS_ALL_FLAGS = SMC_OPS_FLAG_INHERITABLE,
+};
+
+struct smc_ops {
+	/* priavte */
+
+	struct list_head list;
+	struct module *owner;
+
+	/* public */
+
+	/* unique name */
+	char name[SMC_OPS_NAME_MAX];
+	int flags;
+
+	/* Invoked before computing SMC option for SYN packets.
+	 * We can control whether to set SMC options by returning varios value.
+	 * Return 0 to disable SMC, or return any other value to enable it.
+	 */
+	int (*set_option)(struct tcp_sock *tp);
+
+	/* Invoked before Set up SMC options for SYN-ACK packets
+	 * We can control whether to respond SMC options by returning varios value.
+	 * Return 0 to disable SMC, or return any other value to enable it.
+	 */
+	int (*set_option_cond)(const struct tcp_sock *tp, struct inet_request_sock *ireq);
+};
+
+#if IS_ENABLED(CONFIG_SMC_OPS)
+#define smc_call_retops(init_val, sk, func, ...) ({	\
+	typeof(init_val) __ret = (init_val);		\
+	struct smc_ops *ops;				\
+	rcu_read_lock();				\
+	ops = READ_ONCE(sock_net(sk)->smc.ops);		\
+	if (ops && ops->func)				\
+		__ret = ops->func(__VA_ARGS__);		\
+	rcu_read_unlock();				\
+	!!__ret;					\
+})
+#else
+#define smc_call_retops(init_val, ...) (init_val)
+#endif /* CONFIG_SMC_OPS */
+
 #endif	/* _SMC_H */
net/ipv4/tcp_output.c +0 -0
index 5485a70b5fe5..7b402167fb4d 100644

net/ipv4/tcp_output.c line 51

Quoted 4 lines
net/ipv4/tcp_output.c +1 -0
@@ -48,6 +48,7 @@
 #include <linux/skbuff_ref.h>
 
 #include <trace/events/tcp.h>
+#include <net/smc.h>

Put this include right after #include <net/proto_memory.h> ? I notice all the includes here are well classified.

Quoted 48 lines
net/ipv4/tcp_output.c +10 -4
 /* Refresh clocks of a TCP socket,
 * ensuring monotically increasing values.
@@ -759,14 +760,17 @@ static void tcp_options_write(struct tcphdr *th, struct tcp_sock *tp,
 	mptcp_options_write(th, ptr, tp, opts);
 }
 
-static void smc_set_option(const struct tcp_sock *tp,
+static void smc_set_option(struct tcp_sock *tp,
 			   struct tcp_out_options *opts,
 			   unsigned int *remaining)
 {
 #if IS_ENABLED(CONFIG_SMC)
 	if (static_branch_unlikely(&tcp_have_smc)) {
 		if (tp->syn_smc) {
-			if (*remaining >= TCPOLEN_EXP_SMC_BASE_ALIGNED) {
+			tp->syn_smc = smc_call_retops(1, &tp->inet_conn.icsk_inet.sk,
+						      set_option, tp);
+			/* re-check syn_smc */
+			if (tp->syn_smc && *remaining >= TCPOLEN_EXP_SMC_BASE_ALIGNED) {
 				opts->options |= OPTION_SMC;
 				*remaining -= TCPOLEN_EXP_SMC_BASE_ALIGNED;
 			}
@@ -776,14 +780,17 @@ static void smc_set_option(const struct tcp_sock *tp,
 }
 
 static void smc_set_option_cond(const struct tcp_sock *tp,
-				const struct inet_request_sock *ireq,
+				struct inet_request_sock *ireq,
 				struct tcp_out_options *opts,
 				unsigned int *remaining)
 {
 #if IS_ENABLED(CONFIG_SMC)
 	if (static_branch_unlikely(&tcp_have_smc)) {
 		if (tp->syn_smc && ireq->smc_ok) {
-			if (*remaining >= TCPOLEN_EXP_SMC_BASE_ALIGNED) {
+			ireq->smc_ok = smc_call_retops(1, &tp->inet_conn.icsk_inet.sk,
+						       set_option_cond, tp, ireq);
+			/* re-check smc_ok */
+			if (ireq->smc_ok && *remaining >= TCPOLEN_EXP_SMC_BASE_ALIGNED) {
 				opts->options |= OPTION_SMC;
 				*remaining -= TCPOLEN_EXP_SMC_BASE_ALIGNED;
 			}
net/smc/Kconfig +0 -0
index ba5e6a2dd2fd..0ee16ec8dceb 100644

net/smc/Kconfig config SMC_LO · line 42

Quoted 10 lines
net/smc/Kconfig +7 -0
@@ -33,3 +33,15 @@ config SMC_LO
 	  of architecture or hardware.
 
 	  if unsure, say N.
+
+config SMC_OPS
+	bool "Generic hook for SMC subsystem"
+	depends on SMC && BPF_SYSCALL
+	default n
+	help
+	  SMC_OPS enables support to register genericfor hook via eBPF programs

genericfor ?

Quoted 147 lines
net/smc/Kconfig +5 -0
+	  for SMC subsystem. eBPF programs offer much greater flexibility
+	  in modifying the behavior of the SMC protocol stack compared
+	  to a complete kernel-based approach.
+
+	  if unsure, say N.
net/smc/Makefile +0 -0
index 60f1c87d5212..5dd706b2927a 100644
@@ -7,3 +7,4 @@ smc-y += smc_cdc.o smc_tx.o smc_rx.o smc_close.o smc_ism.o smc_netlink.o smc_sta

smc-y += smc_tracepoint.o smc_inet.o
smc-$(CONFIG_SYSCTL) += smc_sysctl.o
smc-$(CONFIG_SMC_LO) += smc_loopback.o

net/smc/Makefile +1 -0
+smc-$(CONFIG_SMC_OPS) += smc_ops.o
\ No newline at end of file
net/smc/smc_ops.c new file +51 -0
new file mode 100644
index 000000000000..0fc19cadd760
@@ -0,0 +1,51 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ *  Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  Generic hook for SMC subsystem.
+ *
+ *  Copyright IBM Corp. 2016
+ *  Copyright (c) 2024, Alibaba Inc.
+ *
+ *  Author: D. Wythe <[email protected]>
+ */
+
+#include "smc_ops.h"
+
+static DEFINE_SPINLOCK(smc_ops_list_lock);
+static LIST_HEAD(smc_ops_list);
+
+static int smc_ops_reg(struct smc_ops *ops)
+{
+	int ret = 0;
+
+	spin_lock(&smc_ops_list_lock);
+	/* already exist or duplicate name */
+	if (smc_ops_find_by_name(ops->name))
+		ret = -EEXIST;
+	else
+		list_add_tail_rcu(&ops->list, &smc_ops_list);
+	spin_unlock(&smc_ops_list_lock);
+	return ret;
+}
+
+static void smc_ops_unreg(struct smc_ops *ops)
+{
+	spin_lock(&smc_ops_list_lock);
+	list_del_rcu(&ops->list);
+	spin_unlock(&smc_ops_list_lock);
+
+	/* Ensure that all readers to complete */
+	synchronize_rcu();
+}
+
+struct smc_ops *smc_ops_find_by_name(const char *name)
+{
+	struct smc_ops *ops;
+
+	list_for_each_entry_rcu(ops, &smc_ops_list, list) {
+		if (strcmp(ops->name, name) == 0)
+			return ops;
+	}
+	return NULL;
+}
net/smc/smc_ops.h new file +25 -0
new file mode 100644
index 000000000000..4b33dc2d65b9
@@ -0,0 +1,25 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ *  Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  Generic hook for SMC subsystem.
+ *
+ *  Copyright IBM Corp. 2016
+ *  Copyright (c) 2024, Alibaba Inc.
+ *
+ *  Author: D. Wythe <[email protected]>
+ */
+
+#ifndef __SMC_OPS
+#define __SMC_OPS
+
+#include <net/smc.h>
+
+/* Find ops by the target name, which required to be a c-string.
+ * Return NULL if no such ops was found,otherwise, return a valid ops.
+ *
+ * Note: Caller MUST ensure it's was invoked under rcu_read_lock.
+ */
+struct smc_ops *smc_ops_find_by_name(const char *name);
+
+#endif /* __SMC_OPS */
net/smc/smc_sysctl.c +29 -0
index 2fab6456f765..2aa3f19025f4 100644
@@ -18,6 +18,7 @@
 #include "smc_core.h"
 #include "smc_llc.h"
 #include "smc_sysctl.h"
+#include "smc_ops.h"
 
 static int min_sndbuf = SMC_BUF_MIN_SIZE;
 static int min_rcvbuf = SMC_BUF_MIN_SIZE;
@@ -30,6 +31,70 @@ static int links_per_lgr_max = SMC_LINKS_ADD_LNK_MAX;
 static int conns_per_lgr_min = SMC_CONN_PER_LGR_MIN;
 static int conns_per_lgr_max = SMC_CONN_PER_LGR_MAX;
 
+#if IS_ENABLED(CONFIG_SMC_OPS)
+static int smc_net_replace_smc_ops(struct net *net, const char *name)
+{
+	struct smc_ops *ops = NULL;
+
+	rcu_read_lock();
+	/* null or empty name ask to clear current ops */
+	if (name && name[0]) {
+		ops = smc_ops_find_by_name(name);
+		if (!ops) {
+			rcu_read_unlock();
+			return -EINVAL;
+		}
+		/* no change, just return */
+		if (ops == rcu_dereference(net->smc.ops)) {
+			rcu_read_unlock();
+			return 0;
+		}
+	}
+	if (!ops || bpf_try_module_get(ops, ops->owner)) {
+		/* xhcg */
+		ops = rcu_replace_pointer(net->smc.ops, ops, true);
+		/* release old ops */
+		if (ops)
+			bpf_module_put(ops, ops->owner);
+	} else if (ops) {
+		rcu_read_unlock();
+		return -EBUSY;

net/smc/smc_sysctl.c static int links_per_lgr_max = SMC_LINKS_ADD_LNK_MAX; · line 71

Quoted 10 lines
net/smc/smc_sysctl.c +10 -0
+	}
+	rcu_read_unlock();
+	return 0;
+}
+
+static int proc_smc_ops(const struct ctl_table *ctl, int write,
+			void *buffer, size_t *lenp, loff_t *ppos)
+{
+	struct net *net = container_of(ctl->data, struct net,
+				       smc.ops);

This line won't exceed 80 characters, no need to split into 2 lines.

Best regards,
Dust

author maintainer
#4 lore raw

Re: [PATCH bpf-next v5 2/5] net/smc: Introduce generic hook smc_ops

Quoted 37 lines · Dust Li
Quoted 30 lines · D. Wythe

The introduction of IPPROTO_SMC enables eBPF programs to determine whether to use SMC based on the context of socket creation, such as network namespaces, PID and comm name, etc.

As a subsequent enhancement, to introduce a new generic hook that allows decisions on whether to use SMC or not at runtime, including but not limited to local/remote IP address or ports.

Moreover, in the future, we can achieve more complex extensions to the protocol stack by extending this ops.

---

include/net/netns/smc.h |  3 ++
include/net/smc.h       | 51 ++++++++++++++++++++++
net/ipv4/tcp_output.c   | 15 +++++--
net/smc/Kconfig         | 12 ++++++
net/smc/Makefile        |  1 +
net/smc/smc_ops.c       | 51 ++++++++++++++++++++++
net/smc/smc_ops.h       | 25 +++++++++++
net/smc/smc_sysctl.c    | 95 +++++++++++++++++++++++++++++++++++++++++
8 files changed, 249 insertions(+), 4 deletions(-)

create mode 100644 net/smc/smc_ops.c
create mode 100644 net/smc/smc_ops.h

include/net/netns/smc.h +0 -0
index fc752a50f91b..59d069f56b2d 100644

include/net/netns/smc.h struct netns_smc { · line 21

Quoted 5 lines
include/net/netns/smc.h +2 -0
@@ -17,6 +17,9 @@ struct netns_smc {
 #ifdef CONFIG_SYSCTL
 	struct ctl_table_header		*smc_hdr;
 #endif
+#if IS_ENABLED(CONFIG_SMC_OPS)
+	struct smc_ops __rcu *ops;
                              ^^^
Quoted 2 lines

Align with other fields

Oops, nice catch.

Quoted 4 lines
Quoted 4 lines
include/net/netns/smc.h +1 -0
+#endif /* CONFIG_SMC_OPS */
 	unsigned int			sysctl_autocorking_size;
 	unsigned int			sysctl_smcr_buf_type;
 	int				sysctl_smcr_testlink_time;
Quoted 10 lines

include/net/smc.h

Quoted 5 lines
include/net/smc.h +1 -0
index db84e4e35080..326a217001d4 100644
+#include <net/smc.h>

Put this include right after #include <net/proto_memory.h> ? I notice all the includes here are well classified.

Done, I didn't notice this detail before.

Quoted 7 lines
Quoted 7 lines

/* Refresh clocks of a TCP socket,
* ensuring monotically increasing values.

include/net/smc.h +0 -0
@@ -759,14 +760,17 @@ static void tcp_options_write(struct tcphdr *th, struct tcp_sock *tp,
 	mptcp_options_write(th, ptr, tp, opts);
 }
 
Quoted 9 lines

include/net/smc.h static void tcp_options_write(struct tcphdr *th, struct tcp_sock *tp, · line 767

Quoted 6 lines
include/net/smc.h +2 -1
-static void smc_set_option(const struct tcp_sock *tp,
+static void smc_set_option(struct tcp_sock *tp,
 			   struct tcp_out_options *opts,
 			   unsigned int *remaining)
 {
+	  SMC_OPS enables support to register genericfor hook via eBPF programs

genericfor ?

generic.. I will fix it in next version.

Quoted 5 lines
Quoted 5 lines
include/net/smc.h +5 -0
+	  for SMC subsystem. eBPF programs offer much greater flexibility
+	  in modifying the behavior of the SMC protocol stack compared
+	  to a complete kernel-based approach.
+
+	  if unsure, say N.
Quoted 9 lines

net/smc/Makefile

Quoted 6 lines
net/smc/Makefile +2 -0
index 60f1c87d5212..5dd706b2927a 100644
+	struct net *net = container_of(ctl->data, struct net,
+				       smc.ops);

This line won't exceed 80 characters, no need to split into 2 lines.

Take it. Thanks for all the advices.

Best wishes,
D. Wythe

Quoted 3 lines

Best regards,
Dust

maintainer
#5 lore raw

Re: [PATCH bpf-next v5 2/5] net/smc: Introduce generic hook smc_ops

In reply to D. Wythe
Quoted 56 lines · D. Wythe

The introduction of IPPROTO_SMC enables eBPF programs to determine whether to use SMC based on the context of socket creation, such as network namespaces, PID and comm name, etc.

As a subsequent enhancement, to introduce a new generic hook that allows decisions on whether to use SMC or not at runtime, including but not limited to local/remote IP address or ports.

Moreover, in the future, we can achieve more complex extensions to the protocol stack by extending this ops.

---

include/net/netns/smc.h |  3 ++
include/net/smc.h       | 51 ++++++++++++++++++++++
net/ipv4/tcp_output.c   | 15 +++++--
net/smc/Kconfig         | 12 ++++++
net/smc/Makefile        |  1 +
net/smc/smc_ops.c       | 51 ++++++++++++++++++++++
net/smc/smc_ops.h       | 25 +++++++++++
net/smc/smc_sysctl.c    | 95 +++++++++++++++++++++++++++++++++++++++++
8 files changed, 249 insertions(+), 4 deletions(-)

create mode 100644 net/smc/smc_ops.c
create mode 100644 net/smc/smc_ops.h

include/net/netns/smc.h +3 -0
index fc752a50f91b..59d069f56b2d 100644
@@ -17,6 +17,9 @@ struct netns_smc {
 #ifdef CONFIG_SYSCTL
 	struct ctl_table_header		*smc_hdr;
 #endif
+#if IS_ENABLED(CONFIG_SMC_OPS)
+	struct smc_ops __rcu *ops;
+#endif /* CONFIG_SMC_OPS */
 	unsigned int			sysctl_autocorking_size;
 	unsigned int			sysctl_smcr_buf_type;
 	int				sysctl_smcr_testlink_time;
include/net/smc.h +2 -0
index db84e4e35080..326a217001d4 100644
@@ -18,6 +18,8 @@
 #include "linux/ism.h"
 
 struct sock;
+struct tcp_sock;
+struct inet_request_sock;
 
 #define SMC_MAX_PNETID_LEN	16	/* Max. length of PNET id */
 
@@ -97,4 +99,53 @@ struct smcd_dev {
 	u8 going_away : 1;
 };
 

include/net/smc.h struct smcd_dev { · line 111

Quoted 10 lines
include/net/smc.h +10 -0
+#define  SMC_OPS_NAME_MAX 16
+
+enum {
+	/* ops can be inherit from init_net */
+	SMC_OPS_FLAG_INHERITABLE = 0x1,
+
+	SMC_OPS_ALL_FLAGS = SMC_OPS_FLAG_INHERITABLE,
+};
+
+struct smc_ops {

One more thing.
Can we call it smc_bpf_ops ? I think smc_ops is a bit ambiguous. Same for smc_ops.h/c source file.

Best regards,
Dust

author maintainer
#6 lore raw

Re: [PATCH bpf-next v5 2/5] net/smc: Introduce generic hook smc_ops

Quoted 26 lines · Dust Li
Quoted 25 lines · D. Wythe

The introduction of IPPROTO_SMC enables eBPF programs to determine whether to use SMC based on the context of socket creation, such as network namespaces, PID and comm name, etc.

As a subsequent enhancement, to introduce a new generic hook that allows decisions on whether to use SMC or not at runtime, including but not limited to local/remote IP address or ports.

Moreover, in the future, we can achieve more complex extensions to the protocol stack by extending this ops.

---

include/net/netns/smc.h |  3 ++
include/net/smc.h       | 51 ++++++++++++++++++++++
net/ipv4/tcp_output.c   | 15 +++++--
net/smc/Kconfig         | 12 ++++++
net/smc/Makefile        |  1 +
net/smc/smc_ops.c       | 51 ++++++++++++++++++++++
net/smc/smc_ops.h       | 25 +++++++++++
net/smc/smc_sysctl.c    | 95 +++++++++++++++++++++++++++++++++++++++++
8 files changed, 249 insertions(+), 4 deletions(-)

create mode 100644 net/smc/smc_ops.c
create mode 100644 net/smc/smc_ops.h

Quoted 6 lines

Review

Quoted 2 lines
(quoted diff) +2 -0
+
+struct smc_ops {

One more thing.
Can we call it smc_bpf_ops ? I think smc_ops is a bit ambiguous. Same for smc_ops.h/c source file.

I don't think smc_bpf_ops is a good idea. BPF is just a way to implement smc_ops. Similarly, we can also implement this ops within the kernel module, just like tcp_congestion_ops dose. If you think this is ambiguous, perhaps we can call it as smc_handshake_ops ? This should eliminate the ambiguity.

Best wishes,
D. Wythe

Quoted 3 lines

Best regards,
Dust

author maintainer
#7 lore raw

PATCH v5 3/5 net/smc: bpf: register smc_ops info struct_ops

In reply to D. Wythe

To implement injection capability for smc via struct_ops, so that user can make their own smc_ops to modify the behavior of smc stack.

Currently, user can write their own implememtion to choose whether to use SMC or not before TCP 3rd handshake to be comleted. In the future, users can implement more complex functions on smc by expanding it.

---
 net/smc/af_smc.c  | 10 ++++++
 net/smc/smc_ops.c | 79 +++++++++++++++++++++++++++++++++++++++++++++++
 net/smc/smc_ops.h |  5 +++
 3 files changed, 94 insertions(+)
net/smc/af_smc.c +10 -0
index 9d76e902fd77..6adedae2986d 100644
@@ -55,6 +55,7 @@
 #include "smc_sysctl.h"
 #include "smc_loopback.h"
 #include "smc_inet.h"
+#include "smc_ops.h"
 
 static DEFINE_MUTEX(smc_server_lgr_pending);	/* serialize link group
 						 * creation on server
@@ -3576,8 +3577,17 @@ static int __init smc_init(void)
 		pr_err("%s: smc_inet_init fails with %d\n", __func__, rc);
 		goto out_ulp;
 	}
+
+	rc = smc_bpf_struct_ops_init();
+	if (rc) {
+		pr_err("%s: smc_bpf_struct_ops_init fails with %d\n", __func__, rc);
+		goto out_inet;
+	}
+
 	static_branch_enable(&tcp_have_smc);
 	return 0;
+out_inet:
+	smc_inet_exit();
 out_ulp:
 	tcp_unregister_ulp(&smc_ulp_ops);
 out_lo:
net/smc/smc_ops.c +79 -0

88 lines not on this page · raw message

net/smc/smc_ops.h +5 -0
index 4b33dc2d65b9..06dc30363a83 100644
@@ -21,5 +21,10 @@
  * Note: Caller MUST ensure it's was invoked under rcu_read_lock.
  */
 struct smc_ops *smc_ops_find_by_name(const char *name);
+#if IS_ENABLED(CONFIG_SMC_OPS)
+int smc_bpf_struct_ops_init(void);
+#else
+static inline int smc_bpf_struct_ops_init(void) { return 0; }
+#endif
 
 #endif /* __SMC_OPS */
Signature · 1 line
2.45.0
author maintainer
#8 lore raw

PATCH v5 4/5 libbpf: fix error when st-prefix_ops and ops from differ btf

In reply to D. Wythe

When a struct_ops named xxx_ops was registered by a module, and it will be used in both built-in modules and the module itself, so that the btf_type of xxx_ops will be present in btf_vmlinux instead of in btf_mod, which means that the btf_type of bpf_struct_ops_xxx_ops and xxx_ops will not be in the same btf.

Here are four possible case:

+--------+-------------+-------------+---------------------------------+
|        | st_opx_xxx  | xxx         |                                 |
+--------+-------------+-------------+---------------------------------+
| case 0 | btf_vmlinux | bft_vmlinux | be used and reg only in vmlinux |
+--------+-------------+-------------+---------------------------------+
| case 1 | btf_vmlinux | bpf_mod     | INVALID                         |
+--------+-------------+-------------+---------------------------------+
| case 2 | btf_mod     | btf_vmlinux | reg in mod but be used both in  |
|        |             |             | vmlinux and mod.                |
+--------+-------------+-------------+---------------------------------+
| case 3 | btf_mod     | btf_mod     | be used and reg only in mod     |
+--------+-------------+-------------+---------------------------------+

At present, cases 0, 1, and 3 can be correctly identified, because st_ops_xxx is searched from the same btf with xxx. In order to handle case 2 correctly without affecting other cases, we cannot simply change the search method for st_ops_xxx from find_btf_by_prefix_kind() to find_ksym_btf_id(), because in this way, case 1 will not be recognized anymore.

To address this issue, if st_ops_xxx cannot be found in the btf with xxx and mod_btf does not exist, do find_ksym_btf_id() again to avoid such issue.

Fixes 590a00888250 ("bpf: libbpf: Add STRUCT_OPS support")
---
 tools/lib/bpf/libbpf.c | 25 +++++++++++++++++--------
 1 file changed, 17 insertions(+), 8 deletions(-)
tools/lib/bpf/libbpf.c +17 -8
index 66173ddb5a2d..56bf74894110 100644
@@ -1005,7 +1005,8 @@ find_struct_ops_kern_types(struct bpf_object *obj, const char *tname_raw,
 	const struct btf_member *kern_data_member;
 	struct btf *btf = NULL;
 	__s32 kern_vtype_id, kern_type_id;
-	char tname[256];
+	char tname[256], stname[256];
+	int ret;
 	__u32 i;
 
 	snprintf(tname, sizeof(tname), "%.*s",
@@ -1020,17 +1021,25 @@ find_struct_ops_kern_types(struct bpf_object *obj, const char *tname_raw,
 	}
 	kern_type = btf__type_by_id(btf, kern_type_id);
 
+	ret = snprintf(stname, sizeof(stname), "%s%s", STRUCT_OPS_VALUE_PREFIX, tname);
+	if (ret < 0 || ret >= sizeof(stname))
+		return -ENAMETOOLONG;
+
 	/* Find the corresponding "map_value" type that will be used
 	 * in map_update(BPF_MAP_TYPE_STRUCT_OPS).  For example,
 	 * find "struct bpf_struct_ops_tcp_congestion_ops" from the
 	 * btf_vmlinux.
 	 */
-	kern_vtype_id = find_btf_by_prefix_kind(btf, STRUCT_OPS_VALUE_PREFIX,
-						tname, BTF_KIND_STRUCT);
+	kern_vtype_id = btf__find_by_name_kind(btf, stname, BTF_KIND_STRUCT);
 	if (kern_vtype_id < 0) {
-		pr_warn("struct_ops init_kern: struct %s%s is not found in kernel BTF\n",
-			STRUCT_OPS_VALUE_PREFIX, tname);
-		return kern_vtype_id;
+		if (kern_vtype_id == -ENOENT && !*mod_btf)
+			kern_vtype_id = find_ksym_btf_id(obj, stname, BTF_KIND_STRUCT,
+							 &btf, mod_btf);
+		if (kern_vtype_id < 0) {
+			pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n",
+				stname);
+			return kern_vtype_id;
+		}
 	}
 	kern_vtype = btf__type_by_id(btf, kern_vtype_id);
 
@@ -1046,8 +1055,8 @@ find_struct_ops_kern_types(struct bpf_object *obj, const char *tname_raw,
 			break;
 	}
 	if (i == btf_vlen(kern_vtype)) {
-		pr_warn("struct_ops init_kern: struct %s data is not found in struct %s%s\n",
-			tname, STRUCT_OPS_VALUE_PREFIX, tname);
+		pr_warn("struct_ops init_kern: struct %s data is not found in struct %s\n",
+			tname, stname);
 		return -EINVAL;
 	}
 
Signature · 1 line
2.45.0
author maintainer
#9 lore raw

PATCH v5 5/5 bpf/selftests: add selftest for bpf_smc_ops

In reply to D. Wythe

This tests introduces a tiny smc_ops for filtering SMC connections based on IP pairs, and also adds a realistic topology model to verify this ops.

Also, we can only use SMC loopback under CI test, so an additional configuration needs to be enabled.

Follow the steps below to run this test.

make -C tools/testing/selftests/bpf
cd tools/testing/selftests/bpf
sudo ./test_progs -t smc

Results shows:
Summary: 1/1 PASSED, 0 SKIPPED, 0 FAILED

---
 tools/testing/selftests/bpf/config            |   4 +
 .../selftests/bpf/prog_tests/test_bpf_smc.c   | 390 ++++++++++++++++++
 tools/testing/selftests/bpf/progs/bpf_smc.c   | 116 ++++++
 3 files changed, 510 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c
 create mode 100644 tools/testing/selftests/bpf/progs/bpf_smc.c
tools/testing/selftests/bpf/config +4 -0
index c378d5d07e02..fac2f2a9d02f 100644
@@ -113,3 +113,7 @@ CONFIG_XDP_SOCKETS=y
 CONFIG_XFRM_INTERFACE=y
 CONFIG_TCP_CONG_DCTCP=y
 CONFIG_TCP_CONG_BBR=y
+CONFIG_INFINIBAND=y
+CONFIG_SMC=y
+CONFIG_SMC_OPS=y
+CONFIG_SMC_LO=y
\ No newline at end of file
tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c new file +390 -0

390 lines not on this page · raw message

tools/testing/selftests/bpf/progs/bpf_smc.c new file +116 -0

116 lines not on this page · raw message

Signature · 1 line
2.45.0
#10 lore raw

Re: [PATCH bpf-next v5 5/5] bpf/selftests: add selftest for bpf_smc_ops

Quoted 27 lines · D. Wythe
tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c new file +27 -0
+static int send_cmd(int fd, __u16 nlmsg_type, __u32 nlmsg_pid, __u16 nlmsg_flags,
+			__u8 genl_cmd, __u16 nla_type,
+			void *nla_data, int nla_len)
+{
+	struct nlattr *na;
+	struct sockaddr_nl nladdr;
+	int r, buflen;
+	char *buf;
+
+	struct msgtemplate msg = {0};
+
+	msg.n.nlmsg_len = NLMSG_LENGTH(GENL_HDRLEN);
+	msg.n.nlmsg_type = nlmsg_type;
+	msg.n.nlmsg_flags = nlmsg_flags;
+	msg.n.nlmsg_seq = 0;
+	msg.n.nlmsg_pid = nlmsg_pid;
+	msg.g.cmd = genl_cmd;
+	msg.g.version = 1;
+	na = (struct nlattr *) GENLMSG_DATA(&msg);
+	na->nla_type = nla_type;
+	na->nla_len = nla_len + 1 + NLA_HDRLEN;
+	memcpy(NLA_DATA(na), nla_data, nla_len);
+	msg.n.nlmsg_len += NLMSG_ALIGN(na->nla_len);
+
+	buf = (char *) &msg;
+	buflen = msg.n.nlmsg_len;
+	memset(&nladdr, 0, sizeof(nladdr));

tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c line 95

Quoted 10 lines
tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c new file +10 -0
+	nladdr.nl_family = AF_NETLINK;
+
+	while ((r = sendto(fd, buf, buflen, 0, (struct sockaddr *) &nladdr,
+			   sizeof(nladdr))) < buflen) {
+		if (r > 0) {
+			buf += r;
+			buflen -= r;
+		} else if (errno != EAGAIN)
+			return -1;
+		}

The "}" indentation is off.

I was wondering if it missed a "}" for the while loop. Turns out the "else if" does not have braces while the "if" has. I would add braces to the "else if" also to avoid confusion like this.

Quoted 68 lines
tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c new file +68 -0
+	return 0;
+}
+
+static bool get_smc_nl_family_id(void)
+{
+	struct sockaddr_nl nl_src;
+	struct msgtemplate msg;
+	struct nlattr *nl;
+	int fd, ret;
+	pid_t pid;
+
+	fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_GENERIC);
+	if (!ASSERT_GT(fd, 0, "nl_family socket"))
+		return false;
+
+	pid = getpid();
+
+	memset(&nl_src, 0, sizeof(nl_src));
+	nl_src.nl_family = AF_NETLINK;
+	nl_src.nl_pid = pid;
+
+	ret = bind(fd, (struct sockaddr *) &nl_src, sizeof(nl_src));
+	if (!ASSERT_GE(ret, 0, "nl_family bind"))
+		goto fail;
+
+	ret = send_cmd(fd, GENL_ID_CTRL, pid,
+		       NLM_F_REQUEST, CTRL_CMD_GETFAMILY,
+		       CTRL_ATTR_FAMILY_NAME, (void *)SMC_GENL_FAMILY_NAME,
+		       strlen(SMC_GENL_FAMILY_NAME));
+	if (!ASSERT_EQ(ret, 0, "nl_family query"))
+		goto fail;
+
+	ret = recv(fd, &msg, sizeof(msg), 0);
+	if (!ASSERT_FALSE(msg.n.nlmsg_type == NLMSG_ERROR || (ret < 0) ||
+			  !NLMSG_OK((&msg.n), ret), "nl_family response"))
+		goto fail;
+
+	nl = (struct nlattr *) GENLMSG_DATA(&msg);
+	nl = (struct nlattr *) ((char *) nl + NLA_ALIGN(nl->nla_len));
+	if (!ASSERT_EQ(nl->nla_type, CTRL_ATTR_FAMILY_ID, "nl_family nla type"))
+		goto fail;
+
+	smc_nl_family_id = *(uint16_t *) NLA_DATA(nl);
+	close(fd);
+	return true;
+fail:
+	close(fd);
+	return false;
+}
+
+static bool smc_ueid(int op)
+{
+	struct sockaddr_nl nl_src;
+	struct msgtemplate msg;
+	struct nlmsgerr *err;
+	char test_ueid[32];
+	int fd, ret;
+	pid_t pid;
+
+	/* UEID required */
+	memset(test_ueid, '\x20', sizeof(test_ueid));
+	memcpy(test_ueid, SMC_BPFTEST_UEID, strlen(SMC_BPFTEST_UEID));
+	fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_GENERIC);
+	if (!ASSERT_GT(fd, 0, "ueid socket"))
+		return false;
+
+	pid = getpid();
+	memset(&nl_src, 0, sizeof(nl_src));

tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c line 173

Quoted 10 lines
tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c new file +10 -0
+	nl_src.nl_family = AF_NETLINK;
+	nl_src.nl_pid = pid;
+
+	ret = bind(fd, (struct sockaddr *) &nl_src, sizeof(nl_src));
+	if (!ASSERT_GE(ret, 0, "ueid bind"))
+		goto fail;
+
+	ret = send_cmd(fd, smc_nl_family_id, pid,
+		       NLM_F_REQUEST | NLM_F_ACK, op, SMC_NLA_EID_TABLE_ENTRY,
+	(void *)test_ueid, sizeof(test_ueid));

Same. Indentation is off.

Quoted 25 lines
tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c new file +25 -0
+	if (!ASSERT_EQ(ret, 0, "ueid cmd"))
+		goto fail;
+
+	ret = recv(fd, &msg, sizeof(msg), 0);
+	if (!ASSERT_FALSE((ret < 0) || !NLMSG_OK((&msg.n), ret), "ueid response"))
+		goto fail;
+
+	if (msg.n.nlmsg_type == NLMSG_ERROR) {
+		err = NLMSG_DATA(&msg);
+		switch (op) {
+		case SMC_NETLINK_REMOVE_UEID:
+			if (!ASSERT_FALSE((err->error && err->error != -ENOENT), "ueid remove"))
+				goto fail;
+			break;
+		case SMC_NETLINK_ADD_UEID:
+			if (!ASSERT_EQ(err->error, 0, "ueid add"))
+				goto fail;
+			break;
+		default:
+			break;
+		}
+	}
+	close(fd);
+	return true;
+fail:

tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c line 208

Quoted 10 lines
tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c new file +10 -0
+	close(fd);
+	return false;
+}
+
+static bool setup_netns(void)
+{
+	if (!ASSERT_OK(make_netns(TEST_NS), "create net namespace"))
+		return false;
+
+	nstoken = open_netns(TEST_NS);

Instead of make_netns and then immediately open_netns, try netns_new(TEST_NS, true) from the test_progs.c.

Quoted 133 lines
tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c new file +133 -0
+	if (!ASSERT_OK_PTR(nstoken, "open net namespace"))
+		goto fail_open_netns;
+
+	if (!ASSERT_OK(system("ip addr add 127.0.1.0/8 dev lo"), "add server node"))
+		goto fail_ip;
+
+	if (!ASSERT_OK(system("ip addr add 127.0.2.0/8 dev lo"), "server via risk path"))
+		goto fail_ip;
+
+	return true;
+fail_open_netns:
+	remove_netns(TEST_NS);
+fail_ip:
+	close_netns(nstoken);
+	return false;
+}
+
+static void cleanup_netns(void)
+{
+	close_netns(nstoken);
+	remove_netns(TEST_NS);
+}
+
+static bool setup_ueid(void)
+{
+	/* get smc nl id */
+	if (!get_smc_nl_family_id())
+		return false;
+	/* clear old ueid for bpftest */
+	smc_ueid(SMC_NETLINK_REMOVE_UEID);
+	/* smc-loopback required ueid */
+	return smc_ueid(SMC_NETLINK_ADD_UEID);
+}
+
+static void cleanup_ueid(void)
+{
+	smc_ueid(SMC_NETLINK_REMOVE_UEID);
+}
+
+static bool setup_smc(void)
+{
+	if (!setup_ueid())
+		return false;
+
+	if (!setup_netns())
+		goto fail_netns;
+
+	return true;
+fail_netns:
+	cleanup_ueid();
+	return false;
+}
+
+static int set_client_addr_cb(int fd, void *opts)
+{
+	const char *src = (const char *)opts;
+	struct sockaddr_in localaddr;
+
+	localaddr.sin_family = AF_INET;
+	localaddr.sin_port = htons(0);
+	localaddr.sin_addr.s_addr = inet_addr(src);
+	return !ASSERT_EQ(bind(fd, &localaddr, sizeof(localaddr)), 0, "client bind");
+}
+
+static void run_link(const char *src, const char *dst, int port)
+{
+	struct network_helper_opts opts = {0};
+	int server, client;
+
+	server = start_server_str(AF_INET, SOCK_STREAM, dst, port, NULL);
+	if (!ASSERT_OK_FD(server, "start service_1"))
+		return;
+
+	opts.proto = IPPROTO_TCP;
+	opts.post_socket_cb = set_client_addr_cb;
+	opts.cb_opts = (void *)src;
+
+	client = connect_to_fd_opts(server, &opts);
+	if (!ASSERT_OK_FD(client, "start connect"))
+		goto fail_client;
+
+	close(client);
+fail_client:
+	close(server);
+}
+
+static void block_link(int map_fd, const char *src, const char *dst)
+{
+	struct smc_strat_ip_value val = { .mode = /* block */ 0 };
+	struct smc_strat_ip_key key = {
+		.sip = inet_addr(src),
+		.dip = inet_addr(dst),
+	};
+
+	bpf_map_update_elem(map_fd, &key, &val, BPF_ANY);
+}
+
+/*
+ * This test describes a real-life service topology as follows:
+ *
+ *                             +-------------> service_1
+ *            link1            |                     |
+ *   +--------------------> server                   |  link 2
+ *   |                         |                     V
+ *   |                         +-------------> service_2
+ *   |        link 3
+ *  client -------------------> server_via_unsafe_path -> service_3
+ *
+ * Among them,
+ * 1. link-1 is very suitable for using SMC.
+ * 2. link-2 is not suitable for using SMC, because the mode of this link is kind of
+ *     short-link services.
+ * 3. link-3 is also not suitable for using SMC, because the RDMA link is unavailable and
+ *     needs to go through a long timeout before it can fallback to TCP.
+ *
+ * To achieve this goal, we use a customized SMC ip strategy via smc_ops.
+ */
+static void test_topo(void)
+{
+	struct bpf_smc *skel;
+	int rc, map_fd;
+
+	skel = bpf_smc__open_and_load();
+	if (!ASSERT_OK_PTR(skel, "bpf_smc__open_and_load"))
+		return;
+
+	rc = bpf_smc__attach(skel);
+	if (!ASSERT_EQ(rc, 0, "bpf_smc__attach"))
+		goto fail;
+
+	map_fd = bpf_map__fd(skel->maps.smc_strats_ip);
+	if (!ASSERT_GT(map_fd, 0, "bpf_map__fd"))
+		goto fail;

tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c line 351

Quoted 10 lines
tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c new file +10 -0
+
+	/* Mock the process of transparent replacement, since we will modify protocol
+	 * to ipproto_smc accropding to it via fmod_ret/update_socket_protocol.
+	 */
+	system("sysctl -w net.smc.ops=linkcheck");
+
+	/* Configure ip strat */
+	block_link(map_fd, CLIENT_IP, SERVER_IP_VIA_RISK_PATH);
+	block_link(map_fd, SERVER_IP, SERVER_IP);
+	close(map_fd);

No need to close(map-fd) here. bpf_smc__destroy(skel) will do it.

It seems the new selftest fails also. not always though which is concerning.

pw-bot: cr

Quoted 24 lines
tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c new file +24 -0
+
+	/* should go with smc */
+	run_link(CLIENT_IP, SERVER_IP, SERVICE_1);
+	/* should go with smc fallback */
+	run_link(SERVER_IP, SERVER_IP, SERVICE_2);
+
+	ASSERT_EQ(skel->bss->smc_cnt, 2, "smc count");
+	ASSERT_EQ(skel->bss->fallback_cnt, 1, "fallback count");
+
+	/* should go with smc */
+	run_link(CLIENT_IP, SERVER_IP, SERVICE_2);
+
+	ASSERT_EQ(skel->bss->smc_cnt, 3, "smc count");
+	ASSERT_EQ(skel->bss->fallback_cnt, 1, "fallback count");
+
+	/* should go with smc fallback */
+	run_link(CLIENT_IP, SERVER_IP_VIA_RISK_PATH, SERVICE_3);
+
+	ASSERT_EQ(skel->bss->smc_cnt, 4, "smc count");
+	ASSERT_EQ(skel->bss->fallback_cnt, 2, "fallback count");
+
+fail:
+	bpf_smc__destroy(skel);
+}
author maintainer
#11 lore raw

Re: [PATCH bpf-next v5 5/5] bpf/selftests: add selftest for bpf_smc_ops

Quoted 14 lines · Martin KaFai Lau

tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c line 95

Quoted 10 lines · D. Wythe
tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c new file +10 -0
+static int send_cmd(int fd, __u16 nlmsg_type, __u32 nlmsg_pid, __u16 nlmsg_flags,
+			__u8 genl_cmd, __u16 nla_type,
+	while ((r = sendto(fd, buf, buflen, 0, (struct sockaddr *) &nladdr,
+			   sizeof(nladdr))) < buflen) {
+		if (r > 0) {
+			buf += r;
+			buflen -= r;
+		} else if (errno != EAGAIN)
+			return -1;
+		}

The "}" indentation is off.

Quoted 4 lines

I was wondering if it missed a "}" for the while loop. Turns out the "else if" does not have braces while the "if" has. I would add braces to the "else if" also to avoid confusion like this.

Take it. I fix change it in next version.

Quoted 4 lines
Quoted 4 lines
tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c new file +4 -0
+	return 0;
+}
+
+static bool get_smc_nl_family_id(void)
Quoted 8 lines

tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c line 173

Quoted 6 lines
tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c new file +6 -0
+{
+	struct sockaddr_nl nl_src;
+	struct msgtemplate msg;
+	ret = send_cmd(fd, smc_nl_family_id, pid,
+		       NLM_F_REQUEST | NLM_F_ACK, op, SMC_NLA_EID_TABLE_ENTRY,
+	(void *)test_ueid, sizeof(test_ueid));

Same. Indentation is off.

Take it. Thanks for pointing it out.

Quoted 8 lines

tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c line 208

Quoted 5 lines
tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c new file +4 -0
+	if (!ASSERT_EQ(ret, 0, "ueid cmd"))
+		goto fail;
+
+	nstoken = open_netns(TEST_NS);

Instead of make_netns and then immediately open_netns, try netns_new(TEST_NS, true) from the test_progs.c.

Got it, I'll try it in next version.

Quoted 10 lines
Quoted 10 lines
tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c new file +9 -0
+	if (!ASSERT_OK_PTR(nstoken, "open net namespace"))
+		goto fail_open_netns;
+
+	if (!ASSERT_OK(system("ip addr add 127.0.1.0/8 dev lo"), "add server node"))
+		goto fail_ip;
+
+	if (!ASSERT_OK(system("ip addr add 127.0.2.0/8 dev lo"), "server via risk path"))
+	close_netns(nstoken);
+	return false;
Quoted 8 lines

tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c line 351

Quoted 6 lines
tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c new file +6 -0
+}
+
+	/* Configure ip strat */
+	block_link(map_fd, CLIENT_IP, SERVER_IP_VIA_RISK_PATH);
+	block_link(map_fd, SERVER_IP, SERVER_IP);
+	close(map_fd);

No need to close(map-fd) here. bpf_smc__destroy(skel) will do it.

Got it. Many thanks.

Quoted 3 lines

It seems the new selftest fails also. not always though which is concerning.

This might not be a random failure, but rather related to s390x, which carries a seid by default, which may affect my action of deleting ueid. I am requesting IBM folks to help me analyze this issue since i have no s390x machine.

Anyway, I will solve it in the next version.

Best wishes,
D. Wythe

Quoted 27 lines

pw-bot: cr

Quoted 24 lines
tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c new file +24 -0
+
+	/* should go with smc */
+	run_link(CLIENT_IP, SERVER_IP, SERVICE_1);
+	/* should go with smc fallback */
+	run_link(SERVER_IP, SERVER_IP, SERVICE_2);
+
+	ASSERT_EQ(skel->bss->smc_cnt, 2, "smc count");
+	ASSERT_EQ(skel->bss->fallback_cnt, 1, "fallback count");
+
+	/* should go with smc */
+	run_link(CLIENT_IP, SERVER_IP, SERVICE_2);
+
+	ASSERT_EQ(skel->bss->smc_cnt, 3, "smc count");
+	ASSERT_EQ(skel->bss->fallback_cnt, 1, "fallback count");
+
+	/* should go with smc fallback */
+	run_link(CLIENT_IP, SERVER_IP_VIA_RISK_PATH, SERVICE_3);
+
+	ASSERT_EQ(skel->bss->smc_cnt, 4, "smc count");
+	ASSERT_EQ(skel->bss->fallback_cnt, 2, "fallback count");
+
+fail:
+	bpf_smc__destroy(skel);
+}
author maintainer
#12 lore raw
In reply to D. Wythe

On Tue, Jan 07, 2025 at 12:17:10PM +0800, D. Wythe wrote:

Hi smc folks,

Given that there are almost no more serious issues blocking the progress of this series in the BPF community regarding the BPF part, your input on the SMC part of this series are very valuable.

Additionally, I encountered some errors during CI testing on the S390x architecture. These errors seem to be related to adding ueid via netlink, and I suspect they might be connected to seid. However, since I have no s390x machine, I am unable to further pinpoint the exact cause of the problem. If IBM could help confirm this, it would be greatly appreciated. If there is no time to address this issue, I plan to temporarily skip the BPF CI testing of my case in S390x.

Best wishes,
D. Wythe

Quoted 80 lines

This patch aims to introduce BPF injection capabilities for SMC and includes a self-test to ensure code stability.

Since the SMC protocol isn't ideal for every situation, especially short-lived ones, most applications can't guarantee the absence of such scenarios. Consequently, applications may need specific strategies to decide whether to use SMC. For example, an application might limit SMC usage to certain IP addresses or ports.

To maintain the principle of transparent replacement, we want applications to remain unaffected even if they need specific SMC strategies. In other words, they should not require recompilation of their code.

Additionally, we need to ensure the scalability of strategy implementation. While using socket options or sysctl might be straightforward, it could complicate future expansions.

Fortunately, BPF addresses these concerns effectively. Users can write their own strategies in eBPF to determine whether to use SMC, and they can easily modify those strategies in the future.

v2:
1. Rename smc_bpf_ops to smc_ops.
2. Change the scope of smc_ops from global to per netns.
3. Directly pass parameters to ops instead of smc_ops_ctx.
4. Remove struct smc_ops_ctx.
5. Remove exports that are no longer needed.

v3:
1. Remove find_ksym_btf_id_by_prefix_kind.
2. Enhance selftest, introduce a complete ops for filtering smc connections based on ip pairs and a realistic topology test to verify it.

v4:
1. Remove unless func: smc_bpf_ops_check_member()
2. Remove unless inline func: smc_ops_find_by_name()
3. Change CONFIG_SMC=y to complete CI testing
4. Change smc_sock to smc_sock___local in test to avoid compiling failed with CONFIG_SMC=y
5. Improve test cases, remove unnecessary timeouts and multi-thread test, using network_helpers to start testing between server and client.
6. Fix issues when the return value of the ops function is neither 0

     nor 1.

v5:
1. Fix incorrect CI config from CONFIG_SMC=Y to CONFIG_SMC=y.

D. Wythe (5):
bpf: export necessary sympols for modules with struct_ops net/smc: Introduce generic hook smc_ops
net/smc: bpf: register smc_ops info struct_ops
libbpf: fix error when st-prefix_ops and ops from differ btf bpf/selftests: add selftest for bpf_smc_ops

 include/net/netns/smc.h                       |   3 +
 include/net/smc.h                             |  51 +++
 kernel/bpf/bpf_struct_ops.c                   |   2 +
 kernel/bpf/syscall.c                          |   1 +
 net/ipv4/tcp_output.c                         |  15 +-
 net/smc/Kconfig                               |  12 +
 net/smc/Makefile                              |   1 +
 net/smc/af_smc.c                              |  10 +
 net/smc/smc_ops.c                             | 130 ++++++
 net/smc/smc_ops.h                             |  30 ++
 net/smc/smc_sysctl.c                          |  95 +++++
 tools/lib/bpf/libbpf.c                        |  25 +-
 tools/testing/selftests/bpf/config            |   4 +
 .../selftests/bpf/prog_tests/test_bpf_smc.c   | 390 ++++++++++++++++++
 tools/testing/selftests/bpf/progs/bpf_smc.c   | 116 ++++++
 15 files changed, 873 insertions(+), 12 deletions(-)
 create mode 100644 net/smc/smc_ops.c
 create mode 100644 net/smc/smc_ops.h
 create mode 100644 tools/testing/selftests/bpf/prog_tests/test_bpf_smc.c
 create mode 100644 tools/testing/selftests/bpf/progs/bpf_smc.c
Signature · 1 line
2.45.0