Merge branch 'net-smc-introduce-smc_hs_ctrl'

D. Wythe says:

====================
net/smc: Introduce smc_hs_ctrl

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.

This is a rework of the series from [1]. Changes since [1] are limited to
the SMC parts:

1. Rename smc_ops to smc_hs_ctrl and change interface name.
2. Squash SMC patches, removing standalone non-BPF hook capability.
3. Fix typos

[1]: https://lore.kernel.org/bpf/[email protected]/#t

v2 -> v1:
  - Removed the fixes patch, which have already been merged on current branch.
  - Fixed compilation warning of smc_call_hsbpf() when CONFIG_SMC_HS_CTRL_BPF
    is not enabled.
  - Changed the default value of CONFIG_SMC_HS_CTRL_BPF to Y.
  - Fix typo and renamed some variables

v3 -> v2:
  - Removed the libbpf patch, which have already been merged on current branch.
  - Fixed sparse warning of smc_call_hsbpf() and xchg().

v4 -> v3:
   - Rebased on latest bpf-next, updated SMC loopback config from SMC_LO to DIBS_LO
     per upstream changes.

v5 -> v4:
    - Removed the redundant sk parameter from smc_call_hsbpf
    - Reject registration when bpf_link is set, link support will be added in the
      future.
    - Updated selftests with new test heplers.
====================

Link: https://patch.msgid.link/[email protected]
Signed-off-by: Martin KaFai Lau <[email protected]>

Author: Martin KaFai Lau

include/net/netns/smc.h

@@ -17,6 +17,9 @@ struct netns_smc {
 #ifdef CONFIG_SYSCTL
 	struct ctl_table_header		*smc_hdr;
 #endif
+#if IS_ENABLED(CONFIG_SMC_HS_CTRL_BPF)
+	struct smc_hs_ctrl __rcu	*hs_ctrl;
+#endif /* CONFIG_SMC_HS_CTRL_BPF */
 	unsigned int			sysctl_autocorking_size;
 	unsigned int			sysctl_smcr_buf_type;
 	int				sysctl_smcr_testlink_time;

include/net/smc.h

@@ -17,6 +17,8 @@
 #include <linux/wait.h>
 #include <linux/dibs.h>
 
+struct tcp_sock;
+struct inet_request_sock;
 struct sock;
 
 #define SMC_MAX_PNETID_LEN	16	/* Max. length of PNET id */
@@ -50,4 +52,55 @@ struct smcd_dev {
 	u8 going_away : 1;
 };
 
+#define SMC_HS_CTRL_NAME_MAX 16
+
+enum {
+	/* ops can be inherit from init_net */
+	SMC_HS_CTRL_FLAG_INHERITABLE = 0x1,
+
+	SMC_HS_CTRL_ALL_FLAGS = SMC_HS_CTRL_FLAG_INHERITABLE,
+};
+
+struct smc_hs_ctrl {
+	/* private */
+
+	struct list_head list;
+	struct module *owner;
+
+	/* public */
+
+	/* unique name */
+	char name[SMC_HS_CTRL_NAME_MAX];
+	int flags;
+
+	/* Invoked before computing SMC option for SYN packets.
+	 * We can control whether to set SMC options by returning various value.
+	 * Return 0 to disable SMC, or return any other value to enable it.
+	 */
+	int (*syn_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 various
+	 * value. Return 0 to disable SMC, or return any other value to enable
+	 * it.
+	 */
+	int (*synack_option)(const struct tcp_sock *tp,
+			     struct inet_request_sock *ireq);
+};
+
+#if IS_ENABLED(CONFIG_SMC_HS_CTRL_BPF)
+#define smc_call_hsbpf(init_val, tp, func, ...) ({				\
+	typeof(init_val) __ret = (init_val);					\
+	struct smc_hs_ctrl *ctrl;						\
+	rcu_read_lock();							\
+	ctrl = rcu_dereference(sock_net((struct sock *)(tp))->smc.hs_ctrl);	\
+	if (ctrl && ctrl->func)							\
+		__ret = ctrl->func(tp, ##__VA_ARGS__);				\
+	rcu_read_unlock();							\
+	__ret;									\
+})
+#else
+#define smc_call_hsbpf(init_val, tp, ...)  ({ (void)(tp); (init_val); })
+#endif /* CONFIG_SMC_HS_CTRL_BPF */
+
 #endif	/* _SMC_H */

kernel/bpf/bpf_struct_ops.c

@@ -1162,6 +1162,7 @@ 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)
 {
@@ -1173,6 +1174,7 @@ void bpf_struct_ops_put(const void *kdata)
 
 	bpf_map_put(&st_map->map);
 }
+EXPORT_SYMBOL_GPL(bpf_struct_ops_put);
 
 u32 bpf_struct_ops_id(const void *kdata)
 {

kernel/bpf/syscall.c

@@ -1234,6 +1234,7 @@ 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,

net/ipv4/tcp_output.c

@@ -40,6 +40,7 @@
 #include <net/tcp.h>
 #include <net/tcp_ecn.h>
 #include <net/mptcp.h>
+#include <net/smc.h>
 #include <net/proto_memory.h>
 #include <net/psp.h>
 
@@ -802,34 +803,36 @@ 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) {
-				opts->options |= OPTION_SMC;
-				*remaining -= TCPOLEN_EXP_SMC_BASE_ALIGNED;
-			}
+	if (static_branch_unlikely(&tcp_have_smc) && tp->syn_smc) {
+		tp->syn_smc = !!smc_call_hsbpf(1, tp, syn_option);
+		/* re-check syn_smc */
+		if (tp->syn_smc &&
+		    *remaining >= TCPOLEN_EXP_SMC_BASE_ALIGNED) {
+			opts->options |= OPTION_SMC;
+			*remaining -= TCPOLEN_EXP_SMC_BASE_ALIGNED;
 		}
 	}
 #endif
 }
 
 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) {
-				opts->options |= OPTION_SMC;
-				*remaining -= TCPOLEN_EXP_SMC_BASE_ALIGNED;
-			}
+	if (static_branch_unlikely(&tcp_have_smc) && tp->syn_smc && ireq->smc_ok) {
+		ireq->smc_ok = !!smc_call_hsbpf(1, tp, synack_option, 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;
 		}
 	}
 #endif

net/smc/Kconfig

@@ -19,3 +19,13 @@ config SMC_DIAG
 	  smcss.
 
 	  if unsure, say Y.
+
+config SMC_HS_CTRL_BPF
+	bool "Generic eBPF hook for SMC handshake flow"
+	depends on SMC && BPF_SYSCALL
+	default y
+	help
+	  SMC_HS_CTRL_BPF enables support to register generic eBPF hook for SMC
+	  handshake flow, which offer much greater flexibility in modifying the behavior
+	  of the SMC protocol stack compared to a complete kernel-based approach. Select
+	  this option if you want filtring the handshake process via eBPF programs.

net/smc/Makefile

@@ -6,3 +6,4 @@ smc-y := af_smc.o smc_pnet.o smc_ib.o smc_clc.o smc_core.o smc_wr.o smc_llc.o
 smc-y += smc_cdc.o smc_tx.o smc_rx.o smc_close.o smc_ism.o smc_netlink.o smc_stats.o
 smc-y += smc_tracepoint.o smc_inet.o
 smc-$(CONFIG_SYSCTL) += smc_sysctl.o
+smc-$(CONFIG_SMC_HS_CTRL_BPF) += smc_hs_bpf.o

net/smc/af_smc.c

@@ -58,6 +58,7 @@
 #include "smc_tracepoint.h"
 #include "smc_sysctl.h"
 #include "smc_inet.h"
+#include "smc_hs_bpf.h"
 
 static DEFINE_MUTEX(smc_server_lgr_pending);	/* serialize link group
 						 * creation on server
@@ -3600,8 +3601,16 @@ static int __init smc_init(void)
 		pr_err("%s: smc_inet_init fails with %d\n", __func__, rc);
 		goto out_ulp;
 	}
+	rc = bpf_smc_hs_ctrl_init();
+	if (rc) {
+		pr_err("%s: bpf_smc_hs_ctrl_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_ib:

net/smc/smc_hs_bpf.c

@@ -0,0 +1,140 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ *  Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  Generic hook for SMC handshake flow.
+ *
+ *  Copyright IBM Corp. 2016
+ *  Copyright (c) 2025, Alibaba Inc.
+ *
+ *  Author: D. Wythe <[email protected]>
+ */
+
+#include <linux/bpf_verifier.h>
+#include <linux/bpf.h>
+#include <linux/btf.h>
+#include <linux/rculist.h>
+
+#include "smc_hs_bpf.h"
+
+static DEFINE_SPINLOCK(smc_hs_ctrl_list_lock);
+static LIST_HEAD(smc_hs_ctrl_list);
+
+static int smc_hs_ctrl_reg(struct smc_hs_ctrl *ctrl)
+{
+	int ret = 0;
+
+	spin_lock(&smc_hs_ctrl_list_lock);
+	/* already exist or duplicate name */
+	if (smc_hs_ctrl_find_by_name(ctrl->name))
+		ret = -EEXIST;
+	else
+		list_add_tail_rcu(&ctrl->list, &smc_hs_ctrl_list);
+	spin_unlock(&smc_hs_ctrl_list_lock);
+	return ret;
+}
+
+static void smc_hs_ctrl_unreg(struct smc_hs_ctrl *ctrl)
+{
+	spin_lock(&smc_hs_ctrl_list_lock);
+	list_del_rcu(&ctrl->list);
+	spin_unlock(&smc_hs_ctrl_list_lock);
+
+	/* Ensure that all readers to complete */
+	synchronize_rcu();
+}
+
+struct smc_hs_ctrl *smc_hs_ctrl_find_by_name(const char *name)
+{
+	struct smc_hs_ctrl *ctrl;
+
+	list_for_each_entry_rcu(ctrl, &smc_hs_ctrl_list, list) {
+		if (strcmp(ctrl->name, name) == 0)
+			return ctrl;
+	}
+	return NULL;
+}
+
+static int __smc_bpf_stub_set_tcp_option(struct tcp_sock *tp) { return 1; }
+static int __smc_bpf_stub_set_tcp_option_cond(const struct tcp_sock *tp,
+					      struct inet_request_sock *ireq)
+{
+	return 1;
+}
+
+static struct smc_hs_ctrl __smc_bpf_hs_ctrl = {
+	.syn_option	= __smc_bpf_stub_set_tcp_option,
+	.synack_option	= __smc_bpf_stub_set_tcp_option_cond,
+};
+
+static int smc_bpf_hs_ctrl_init(struct btf *btf) { return 0; }
+
+static int smc_bpf_hs_ctrl_reg(void *kdata, struct bpf_link *link)
+{
+	if (link)
+		return -EOPNOTSUPP;
+
+	return smc_hs_ctrl_reg(kdata);
+}
+
+static void smc_bpf_hs_ctrl_unreg(void *kdata, struct bpf_link *link)
+{
+	smc_hs_ctrl_unreg(kdata);
+}
+
+static int smc_bpf_hs_ctrl_init_member(const struct btf_type *t,
+				       const struct btf_member *member,
+				       void *kdata, const void *udata)
+{
+	const struct smc_hs_ctrl *u_ctrl;
+	struct smc_hs_ctrl *k_ctrl;
+	u32 moff;
+
+	u_ctrl = (const struct smc_hs_ctrl *)udata;
+	k_ctrl = (struct smc_hs_ctrl *)kdata;
+
+	moff = __btf_member_bit_offset(t, member) / 8;
+	switch (moff) {
+	case offsetof(struct smc_hs_ctrl, name):
+		if (bpf_obj_name_cpy(k_ctrl->name, u_ctrl->name,
+				     sizeof(u_ctrl->name)) <= 0)
+			return -EINVAL;
+		return 1;
+	case offsetof(struct smc_hs_ctrl, flags):
+		if (u_ctrl->flags & ~SMC_HS_CTRL_ALL_FLAGS)
+			return -EINVAL;
+		k_ctrl->flags = u_ctrl->flags;
+		return 1;
+	default:
+		break;
+	}
+
+	return 0;
+}
+
+static const struct bpf_func_proto *
+bpf_smc_hs_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
+{
+	return bpf_base_func_proto(func_id, prog);
+}
+
+static const struct bpf_verifier_ops smc_bpf_verifier_ops = {
+	.get_func_proto		= bpf_smc_hs_func_proto,
+	.is_valid_access	= bpf_tracing_btf_ctx_access,
+};
+
+static struct bpf_struct_ops bpf_smc_hs_ctrl_ops = {
+	.name		= "smc_hs_ctrl",
+	.init		= smc_bpf_hs_ctrl_init,
+	.reg		= smc_bpf_hs_ctrl_reg,
+	.unreg		= smc_bpf_hs_ctrl_unreg,
+	.cfi_stubs	= &__smc_bpf_hs_ctrl,
+	.verifier_ops	= &smc_bpf_verifier_ops,
+	.init_member	= smc_bpf_hs_ctrl_init_member,
+	.owner		= THIS_MODULE,
+};
+
+int bpf_smc_hs_ctrl_init(void)
+{
+	return register_bpf_struct_ops(&bpf_smc_hs_ctrl_ops, smc_hs_ctrl);
+}

net/smc/smc_hs_bpf.h

@@ -0,0 +1,31 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ *  Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  Generic hook for SMC handshake flow.
+ *
+ *  Copyright IBM Corp. 2016
+ *  Copyright (c) 2025, Alibaba Inc.
+ *
+ *  Author: D. Wythe <[email protected]>
+ */
+
+#ifndef __SMC_HS_CTRL
+#define __SMC_HS_CTRL
+
+#include <net/smc.h>
+
+/* Find hs_ctrl by the target name, which required to be a c-string.
+ * Return NULL if no such ctrl was found,otherwise, return a valid ctrl.
+ *
+ * Note: Caller MUST ensure it's was invoked under rcu_read_lock.
+ */
+struct smc_hs_ctrl *smc_hs_ctrl_find_by_name(const char *name);
+
+#if IS_ENABLED(CONFIG_SMC_HS_CTRL_BPF)
+int bpf_smc_hs_ctrl_init(void);
+#else
+static inline int bpf_smc_hs_ctrl_init(void) { return 0; }
+#endif /* CONFIG_SMC_HS_CTRL_BPF */
+
+#endif /* __SMC_HS_CTRL */