Skip to contentSkip to conversation

preliminary support for 'ss'

RFC PATCH In review Davide Caratti 4 participants 8 messages May 14, 2019 – May 17, 2019

Subsystems (3)
lore
Share & tools
lore

Davide Caratti

author
#0 lore raw

cover preliminary support for 'ss'

dumping subflow-related information from userspace tools can be useful also in the early stages of developement, for writing functional tests and for debugging purposes.

- patch 1 adds support for ULP-specific information in the output of INET_DIAG_INFO request.
- patch2 is a POC that dumps several members of struct subflow_context.

Davide Caratti (2):
tcp: ulp: add functions to dump ulp-specific information
mptcp: allow dumping subflow context to userspace

 include/net/tcp.h              |  3 ++
 include/uapi/linux/inet_diag.h |  7 ++++
 include/uapi/linux/mptcp.h     | 21 ++++++++++
 net/ipv4/tcp_diag.c            | 33 ++++++++++++++-
 net/mptcp/Makefile             |  2 +-
 net/mptcp/diag.c               | 73 ++++++++++++++++++++++++++++++++++
 net/mptcp/protocol.h           |  3 ++
 net/mptcp/subflow.c            |  2 +
 8 files changed, 141 insertions(+), 3 deletions(-)
 create mode 100644 include/uapi/linux/mptcp.h
 create mode 100644 net/mptcp/diag.c
Signature · 1 line
2.20.1
author
#1 lore raw

RFC PATCH 1/2 tcp: ulp: add functions to dump ulp-specific information

currently, only getsockopt() can be invoked to know if a ULP is on top of a TCP socket. idiag_get_aux() and idiag_get_aux_size() are currently used only by TCP MD5 extension: extend them to dump ULP-related information to userspace tools like 'ss' when they query the kernel with INET_DIAG_INFO.

Signed-off-by Davide Caratti <dcaratti(a)redhat.com>
---
 include/net/tcp.h              |  3 +++
 include/uapi/linux/inet_diag.h |  2 ++
 net/ipv4/tcp_diag.c            | 33 +++++++++++++++++++++++++++++++--
 3 files changed, 36 insertions(+), 2 deletions(-)
include/net/tcp.h +3 -0
index 02df4f946c06..16161dbbdea0 100644
@@ -2100,6 +2100,9 @@ struct tcp_ulp_ops {
 	/* clone ulp */
 	void (*clone)(const struct request_sock *req, struct sock *newsk,
 		      const gfp_t priority);
+	/* diagnostic */
+	int (*get_info)(struct sock *sk, struct sk_buff *skb);
+	size_t (*get_info_size)(struct sock *sk);
 
 	char		name[TCP_ULP_NAME_MAX];
 	struct module	*owner;
include/uapi/linux/inet_diag.h +2 -0
index e8baca85bac6..195e10550dc8 100644
@@ -153,6 +153,8 @@ enum {
 	INET_DIAG_BBRINFO,	/* request as INET_DIAG_VEGASINFO */
 	INET_DIAG_CLASS_ID,	/* request as INET_DIAG_TCLASS */
 	INET_DIAG_MD5SIG,
+	INET_DIAG_ULPNAME,
+	INET_DIAG_ULPINFO,
 	__INET_DIAG_MAX,
 };
 
net/ipv4/tcp_diag.c +31 -2
index 81148f7a2323..c37a0eb7a9b8 100644
@@ -88,10 +88,12 @@ static int tcp_diag_put_md5sig(struct sk_buff *skb,
 static int tcp_diag_get_aux(struct sock *sk, bool net_admin,
 			    struct sk_buff *skb)
 {
+	struct inet_connection_sock *icsk = inet_csk(sk);
+	int err = 0;
+
 #ifdef CONFIG_TCP_MD5SIG
 	if (net_admin) {
 		struct tcp_md5sig_info *md5sig;
-		int err = 0;
 
 		rcu_read_lock();
 		md5sig = rcu_dereference(tcp_sk(sk)->md5sig_info);
@@ -102,12 +104,33 @@ static int tcp_diag_get_aux(struct sock *sk, bool net_admin,
 			return err;
 	}
 #endif
+	if (net_admin && icsk->icsk_ulp_ops) {
+		struct nlattr *nest;
 
-	return 0;
+		err = nla_put_string(skb, INET_DIAG_ULPNAME,
+				     icsk->icsk_ulp_ops->name);
+		if (err < 0 || !icsk->icsk_ulp_ops->get_info)
+			return err;
+
+		nest = nla_nest_start_noflag(skb, INET_DIAG_ULPINFO);
+		if (!nest) {
+			err = -EMSGSIZE;
+			goto nla_failure;
+		}
+		err = icsk->icsk_ulp_ops->get_info(sk, skb);
+		if (err < 0) {
+nla_failure:
+			nla_nest_cancel(skb, nest);
+			return err;
+		}
+		nla_nest_end(skb, nest);
+	}
+	return err;
 }
 
 static size_t tcp_diag_get_aux_size(struct sock *sk, bool net_admin)
 {
+	struct inet_connection_sock *icsk = inet_csk(sk);
 	size_t size = 0;
 
 #ifdef CONFIG_TCP_MD5SIG
@@ -127,6 +150,12 @@ static size_t tcp_diag_get_aux_size(struct sock *sk, bool net_admin)
 				       sizeof(struct tcp_diag_md5sig));
 	}
 #endif
+	if (net_admin && icsk->icsk_ulp_ops) {
+		size += nla_total_size(TCP_ULP_NAME_MAX);
+		if (icsk->icsk_ulp_ops->get_info_size)
+			size += nla_total_size(0) + /* INET_DIAG_ULPINFO */
+				icsk->icsk_ulp_ops->get_info_size(sk);
+	}
 
 	return size;
 }
Signature · 1 line
2.20.1
author
#2 lore raw

RFC PATCH 2/2 mptcp: allow dumping subflow context to userspace

In reply to Davide Caratti

add ulp-specific diagnostic functions, so that subflow information can be dumped to userspace programs like 'ss'.

Signed-off-by Davide Caratti <dcaratti(a)redhat.com>

---

Notes:
- this is a POC patch where I only dump few members of the subflow, and put them in a nested attribute so that it's possible to read them in a program regardless the value of 'subflow_ulp_ops.name'. Please let me know if we should dump additional information, or if some of those I included don't make sense.

 include/uapi/linux/inet_diag.h |  5 +++
 include/uapi/linux/mptcp.h     | 21 ++++++++++
 net/mptcp/Makefile             |  2 +-
 net/mptcp/diag.c               | 73 ++++++++++++++++++++++++++++++++++
 net/mptcp/protocol.h           |  3 ++
 net/mptcp/subflow.c            |  2 +
 6 files changed, 105 insertions(+), 1 deletion(-)
 create mode 100644 include/uapi/linux/mptcp.h
 create mode 100644 net/mptcp/diag.c
include/uapi/linux/inet_diag.h +5 -0
index 195e10550dc8..ae341c9bc4e6 100644
@@ -204,4 +204,9 @@ union tcp_cc_info {
 	struct tcp_dctcp_info	dctcp;
 	struct tcp_bbr_info	bbr;
 };
+
+enum {
+	ULPINFO_MPTCP_SUBFLOW,
+	__ULPINFO_MAX
+};
 #endif /* _UAPI_INET_DIAG_H_ */
include/uapi/linux/mptcp.h new file +21 -0
new file mode 100644
index 000000000000..bba74a97e6c2
@@ -0,0 +1,21 @@
+/* SPDX-License-Identifier: GPL-2.0+ WITH Linux-syscall-note */
+#ifndef _UAPI_MPTCP_H
+#define _UAPI_MPTCP_H
+
+#include <linux/types.h>
+
+#define SUBFLOW_FLAGS_MP_REMOTE BIT(0)
+#define SUBFLOW_FLAGS_MP_LOCAL  BIT(1)
+#define SUBFLOW_FLAGS_DSS_CSUM  BIT(2)
+
+enum {
+	MPTCP_SUBFLOW_LOCAL_KEY,
+	MPTCP_SUBFLOW_REMOTE_KEY,
+	MPTCP_SUBFLOW_FLAGS,
+	MPTCP_SUBFLOW_PAD,
+	__MPTCP_SUBFLOW_MAX
+};
+
+#define MPTCP_SUBFLOW_MAX (__MPTCP_SUBFLOW_MAX - 1)
+
+#endif /* _UAPI_MPTCP_H */
net/mptcp/Makefile +1 -1
index 178ae81d8b66..f93ac5bc1b90 100644
@@ -1,4 +1,4 @@
 # SPDX-License-Identifier: GPL-2.0
 obj-$(CONFIG_MPTCP) += mptcp.o
 
-mptcp-y := protocol.o subflow.o options.o token.o crypto.o
+mptcp-y := protocol.o subflow.o options.o token.o crypto.o diag.o
net/mptcp/diag.c new file +73 -0

73 lines not on this page · raw message

net/mptcp/protocol.h +3 -0
index 81260c737046..721b28e26bbf 100644
@@ -144,4 +144,7 @@ static inline struct mptcp_ext *mptcp_get_ext(struct sk_buff *skb)
 	return (struct mptcp_ext *)skb_ext_find(skb, SKB_EXT_MPTCP);
 }
 
+size_t subflow_get_info_size(struct sock *sk);
+int subflow_get_info(struct sock *sk, struct sk_buff *skb);
+
 #endif /* __MPTCP_PROTOCOL_H */
net/mptcp/subflow.c +2 -0
index f88bc48ca997..2d4f810bc2d2 100644
@@ -262,6 +262,8 @@ static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
 	.init		= subflow_ulp_init,
 	.release	= subflow_ulp_release,
 	.clone		= subflow_ulp_clone,
+	.get_info	= subflow_get_info,
+	.get_info_size	= subflow_get_info_size,
 };
 
 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
Signature · 1 line
2.20.1
#3 lore raw

Re: [MPTCP] [PATCH RFC mptcp 2/2] mptcp: allow dumping subflow context to userspace

Hi,

Quoted 5 lines · Davide Caratti

- this is a POC patch where I only dump few members of the subflow, and put them in a nested attribute so that it's possible to read them in a program regardless the value of 'subflow_ulp_ops.name'. Please let me know if we should dump additional information, or if some of those I included don't make sense.

I personally think that SSN and mapping related info would be useful and worthy to add.

Paolo

#5 lore raw

Re: [MPTCP] [PATCH RFC mptcp 2/2] mptcp: allow dumping subflow context to userspace

In reply to Davide Caratti
Quoted 50 lines · Davide Caratti

add ulp-specific diagnostic functions, so that subflow information can be dumped to userspace programs like 'ss'.

Signed-off-by Davide Caratti <dcaratti(a)redhat.com>

---

Notes:
- this is a POC patch where I only dump few members of the subflow, and put them in a nested attribute so that it's possible to read them in a program regardless the value of 'subflow_ulp_ops.name'. Please let me know if we should dump additional information, or if some of those I included don't make sense.

include/uapi/linux/inet_diag.h |  5 +++
include/uapi/linux/mptcp.h     | 21 ++++++++++
net/mptcp/Makefile             |  2 +-
net/mptcp/diag.c               | 73 ++++++++++++++++++++++++++++++++++
net/mptcp/protocol.h           |  3 ++
net/mptcp/subflow.c            |  2 +
6 files changed, 105 insertions(+), 1 deletion(-)

create mode 100644 include/uapi/linux/mptcp.h
create mode 100644 net/mptcp/diag.c

include/uapi/linux/inet_diag.h +5 -0
index 195e10550dc8..ae341c9bc4e6 100644
@@ -204,4 +204,9 @@ union tcp_cc_info {
 	struct tcp_dctcp_info	dctcp;
 	struct tcp_bbr_info	bbr;
 };
+
+enum {
+	ULPINFO_MPTCP_SUBFLOW,
+	__ULPINFO_MAX
+};
 #endif /* _UAPI_INET_DIAG_H_ */
include/uapi/linux/mptcp.h new file +7 -0
new file mode 100644
index 000000000000..bba74a97e6c2
@@ -0,0 +1,21 @@
+/* SPDX-License-Identifier: GPL-2.0+ WITH Linux-syscall-note */
+#ifndef _UAPI_MPTCP_H
+#define _UAPI_MPTCP_H
+
+#include <linux/types.h>
+
+#define SUBFLOW_FLAGS_MP_REMOTE BIT(0)

include/uapi/linux/mptcp.h line 17

Quoted 10 lines
include/uapi/linux/mptcp.h new file +10 -0
+#define SUBFLOW_FLAGS_MP_LOCAL  BIT(1)
+#define SUBFLOW_FLAGS_DSS_CSUM  BIT(2)
+
+enum {
+	MPTCP_SUBFLOW_LOCAL_KEY,
+	MPTCP_SUBFLOW_REMOTE_KEY,
+	MPTCP_SUBFLOW_FLAGS,
+	MPTCP_SUBFLOW_PAD,
+	__MPTCP_SUBFLOW_MAX
+};

The 32-bit subflow->token is also useful, since that's what would be relevant for future MP_JOIN attempts.

Quoted 43 lines
include/uapi/linux/mptcp.h new file +4 -0
+
+#define MPTCP_SUBFLOW_MAX (__MPTCP_SUBFLOW_MAX - 1)
+
+#endif /* _UAPI_MPTCP_H */
net/mptcp/Makefile +0 -0
index 178ae81d8b66..f93ac5bc1b90 100644
@@ -1,4 +1,4 @@
 # SPDX-License-Identifier: GPL-2.0

obj-$(CONFIG_MPTCP) += mptcp.o

net/mptcp/Makefile +1 -1
-mptcp-y := protocol.o subflow.o options.o token.o crypto.o
+mptcp-y := protocol.o subflow.o options.o token.o crypto.o diag.o
net/mptcp/diag.c new file +23 -0
new file mode 100644
index 000000000000..27266192da88
@@ -0,0 +1,73 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include <linux/module.h>
+#include <linux/net.h>
+#include <linux/sock_diag.h>
+#include <linux/inet_diag.h>
+#include <linux/tcp.h>
+
+#include <net/netlink.h>
+#include <net/tcp.h>
+#include <net/mptcp.h>
+
+#include <uapi/linux/mptcp.h>
+#include "protocol.h"
+
+int subflow_get_info(struct sock *sk, struct sk_buff *skb)
+{
+	struct mptcp_subflow_info *info;
+	struct subflow_context *subflow = subflow_ctx(sk);
+	struct nlattr *start = 0;
+	u32 flags = 0;
+	int ret = 0;
+

net/mptcp/diag.c line 33

Quoted 10 lines
net/mptcp/diag.c new file +10 -0
+	if (!subflow)
+		goto end;
+
+	start = nla_nest_start_noflag(skb, ULPINFO_MPTCP_SUBFLOW);
+	if (!start) {
+		ret = -EMSGSIZE;
+		goto nla_failure;
+	}
+	if (subflow->mp_capable)
+		flags |= SUBFLOW_FLAGS_MP_REMOTE;

If mp_capable is true, the socket is using MPTCP. If it's false, it has done a fallback to TCP. For userspace, might be better to name it SUBFLOW_FLAGS_IN_TCP_FALLBACK (or similar) if !mp_capable.

net/mptcp/diag.c line 35

Quoted 2 lines
net/mptcp/diag.c new file +2 -0
+	if (subflow->request_mptcp)
+		flags |= SUBFLOW_FLAGS_MP_LOCAL;

This is only used at connect time and is only false if ULP setup failed, so I don't think it's useful.

net/mptcp/diag.c line 37

Quoted 2 lines
net/mptcp/diag.c new file +2 -0
+	if (subflow->request_cksum)
+		flags |= SUBFLOW_FLAGS_DSS_CSUM;

subflow->use_checksum is the field to check for this flag.

Thanks for the patch, it will be helpful to have more visibility at run time.

Mat

Quoted 69 lines
net/mptcp/diag.c new file +36 -0
+
+	if (nla_put_be64(skb, MPTCP_SUBFLOW_LOCAL_KEY,
+			 cpu_to_be64(subflow->local_key),
+			 MPTCP_SUBFLOW_PAD) ||
+	    nla_put_be64(skb, MPTCP_SUBFLOW_REMOTE_KEY,
+			 cpu_to_be64(subflow->remote_key),
+			 MPTCP_SUBFLOW_PAD) ||
+	    nla_put_u32(skb, MPTCP_SUBFLOW_FLAGS, flags)) {
+		ret = -EMSGSIZE;
+		goto nla_failure;
+	}
+	nla_nest_end(skb, start);
+end:
+	return ret;
+
+nla_failure:
+	nla_nest_cancel(skb, start);
+	goto end;
+}
+
+size_t subflow_get_info_size(struct sock *sk)
+{
+	struct subflow_context *subflow = subflow_ctx(sk);
+	size_t size = 0;
+
+	if (!subflow)
+		goto end;
+
+	size += nla_total_size(0) +  /* ULPINFO_MPTCP_SUBFLOW */
+		nla_total_size_64bit(8) + /* MPTCP_SUBFLOW_LOCAL_KEY */
+		nla_total_size_64bit(8) + /* MPTCP_SUBFLOW_REMOTE_KEY */
+		nla_total_size(4) + /* MPTCP_SUBFLOW_FLAGS */
+		0;
+end:
+	return size;
+}
net/mptcp/protocol.h +3 -0
index 81260c737046..721b28e26bbf 100644
@@ -144,4 +144,7 @@ static inline struct mptcp_ext *mptcp_get_ext(struct sk_buff *skb)
 	return (struct mptcp_ext *)skb_ext_find(skb, SKB_EXT_MPTCP);
 }
 
+size_t subflow_get_info_size(struct sock *sk);
+int subflow_get_info(struct sock *sk, struct sk_buff *skb);
+
 #endif /* __MPTCP_PROTOCOL_H */
net/mptcp/subflow.c +2 -0
index f88bc48ca997..2d4f810bc2d2 100644
@@ -262,6 +262,8 @@ static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
 	.init		= subflow_ulp_init,
 	.release	= subflow_ulp_release,
 	.clone		= subflow_ulp_clone,
+	.get_info	= subflow_get_info,
+	.get_info_size	= subflow_get_info_size,
 };
 
 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
Signature · 6 lines
2.20.1

_______________________________________________
mptcp mailing list
mptcp(a)lists.01.org
https://lists.01.org/mailman/listinfo/mptcp

--
Mat Martineau
Intel

#4 lore raw

Re: [MPTCP] [PATCH RFC mptcp 2/2] mptcp: allow dumping subflow context to userspace

In reply to Paolo Abeni
Quoted 12 lines · Paolo Abeni

Hi,

Quoted 5 lines · Davide Caratti

- this is a POC patch where I only dump few members of the subflow, and put them in a nested attribute so that it's possible to read them in a program regardless the value of 'subflow_ulp_ops.name'. Please let me know if we should dump additional information, or if some of those I included don't make sense.

I personally think that SSN and mapping related info would be useful and worthy to add.

As we implement multiple subflows and deal with MPTCP-level out-of-order reassembly, the mapping-related info will likely move in to the MPTCP-level socket. If we expose information about mappings, I think we would register a different inet_diag_handler for the MPTCP socket and report it there.

--
Mat Martineau
Intel

#6 lore raw

Re: [MPTCP] [PATCH RFC mptcp 2/2] mptcp: allow dumping subflow context to userspace

In reply to Mat Martineau

Hi Mat, Davide,

Thank you for the review and the patches!

Quoted 21 lines · Mat Martineau
Quoted 18 lines · Davide Caratti
include/uapi/linux/mptcp.h new file +12 -0
new file mode 100644
index 000000000000..bba74a97e6c2
@@ -0,0 +1,21 @@
+/* SPDX-License-Identifier: GPL-2.0+ WITH Linux-syscall-note */
+#ifndef _UAPI_MPTCP_H
+#define _UAPI_MPTCP_H
+
+#include <linux/types.h>
+
+#define SUBFLOW_FLAGS_MP_REMOTE BIT(0)
+#define SUBFLOW_FLAGS_MP_LOCAL  BIT(1)
+#define SUBFLOW_FLAGS_DSS_CSUM  BIT(2)
+
+enum {
+    MPTCP_SUBFLOW_LOCAL_KEY,
Quoted 8 lines

include/uapi/linux/mptcp.h line 17

Quoted 5 lines
include/uapi/linux/mptcp.h new file +5 -0
+    MPTCP_SUBFLOW_REMOTE_KEY,
+    MPTCP_SUBFLOW_FLAGS,
+    MPTCP_SUBFLOW_PAD,
+    __MPTCP_SUBFLOW_MAX
+};

The 32-bit subflow->token is also useful, since that's what would be relevant for future MP_JOIN attempts.

Just a small note to say that it is not recommended to expose this token to regular users for security reasons. Please see this (famous -- Christoph would remember it :) ) commit from mptcp.org:

https://github.com/multipath-tcp/mptcp/commit/e993b775d8abda232ceecbbed135b40c546f5088

Cheers,
Matt

Signature · 5 lines
Matthieu Baerts | R&D Engineer
matthieu.baerts(a)tessares.net
Tessares SA | Hybrid Access Solutions
www.tessares.net
1 Avenue Jean Monnet, 1348 Louvain-la-Neuve, Belgium
#7 lore raw

Re: [MPTCP] [PATCH RFC mptcp 2/2] mptcp: allow dumping subflow context to userspace

Quoted 35 lines · Matthieu Baerts

Hi Mat, Davide,

Thank you for the review and the patches!

Quoted 21 lines · Mat Martineau
Quoted 18 lines · Davide Caratti
include/uapi/linux/mptcp.h new file +12 -0
new file mode 100644
index 000000000000..bba74a97e6c2
@@ -0,0 +1,21 @@
+/* SPDX-License-Identifier: GPL-2.0+ WITH Linux-syscall-note */
+#ifndef _UAPI_MPTCP_H
+#define _UAPI_MPTCP_H
+
+#include <linux/types.h>
+
+#define SUBFLOW_FLAGS_MP_REMOTE BIT(0)
+#define SUBFLOW_FLAGS_MP_LOCAL  BIT(1)
+#define SUBFLOW_FLAGS_DSS_CSUM  BIT(2)
+
+enum {
+    MPTCP_SUBFLOW_LOCAL_KEY,
Quoted 8 lines

include/uapi/linux/mptcp.h line 17

Quoted 5 lines
include/uapi/linux/mptcp.h new file +5 -0
+    MPTCP_SUBFLOW_REMOTE_KEY,
+    MPTCP_SUBFLOW_FLAGS,
+    MPTCP_SUBFLOW_PAD,
+    __MPTCP_SUBFLOW_MAX
+};

The 32-bit subflow->token is also useful, since that's what would be relevant for future MP_JOIN attempts.

Quoted 7 lines

Just a small note to say that it is not recommended to expose this token to regular users for security reasons. Please see this (famous -- Christoph would remember it :) ) commit from mptcp.org:

https://github.com/multipath-tcp/mptcp/commit/e993b775d8abda232ceecbbed135b40c546f5088

Yes! I had seen that Davide's first patch checks for net_admin privileges, but it's best to make sure everyone knows that access to the tokens should have this restriction.

--
Mat Martineau
Intel