[PATCH net-next 0/3] Namespace-ify some sysctl in net/core

[PATCH net-next 0/3] Namespace-ify some sysctl in net/core

[menglong8.dong]

From: Menglong Dong <dong.menglong@zte.com.cn>

For now, most sysctl in 'net/core' are globally unified, such as
sysctl_wmem_default, sysctl_rmem_default, sysctl_wmem_default,
sysctl_rmem_default, etc.

It's not convenient in some case. For example, when we use docker
and try to control the default udp socket receive buffer for
each container by sysctl_rmem_default.

For that reason, I namespace-ify some sysctl in 'net/core', which
are sysctl_wmem_default, sysctl_rmem_default, sysctl_wmem_default
and sysctl_rmem_default.

In the first patch, I made some adjustments to the initialization
of netns_core_table.

The second patch make sysctl_wmem_default and sysctl_rmem_default
per-namespace, and the third patch make sysctl_wmem_max and
sysctl_rmem_max per-namespace.

After these patch, sysctl above are pre-namespace, for example:

$ cat /proc/sys/net/core/rmem_default
1024000
$ ip netns exec test cat /proc/sys/net/core/rmem_default
212992
$ ip netns exec test2 cat /proc/sys/net/core/rmem_default
2048000

Thanks for Christian's patient explaining to make these patches a
single series~

Menglong Dong (3):
  net: core: init every ctl_table in netns_core_table
  net: core: Namespace-ify sysctl_wmem_default and sysctl_rmem_default
  net: core: Namespace-ify sysctl_rmem_max and sysctl_wmem_max

 include/net/netns/core.h        |  4 ++
 include/net/sock.h              |  6 ---
 net/core/filter.c               |  4 +-
 net/core/net_namespace.c        |  4 ++
 net/core/sock.c                 | 18 +++-----
 net/core/sysctl_net_core.c      | 76 +++++++++++++++++----------------
 net/ipv4/ip_output.c            |  2 +-
 net/ipv4/tcp_output.c           |  2 +-
 net/netfilter/ipvs/ip_vs_sync.c |  4 +-
 9 files changed, 60 insertions(+), 60 deletions(-)


base-commit: 5ee88057889bbca5f5bb96031b62b3756b33e164
-- 
2.30.0

[PATCH net-next 1/3] net: core: init every ctl_table in netns_core_table

[menglong8.dong]

From: Menglong Dong <dong.menglong@zte.com.cn>

For now, there is only one element in netns_core_table, and it is inited
directly in sysctl_core_net_init. To make it more flexible, we can init
every element at once, just like what ipv4_sysctl_init_net() did.

Signed-off-by: Menglong Dong <dong.menglong@zte.com.cn>
---
 net/core/sysctl_net_core.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/net/core/sysctl_net_core.c b/net/core/sysctl_net_core.c
index d86d8d11cfe4..966d976dee84 100644
--- a/net/core/sysctl_net_core.c
+++ b/net/core/sysctl_net_core.c
@@ -606,15 +606,19 @@ static __net_init int sysctl_core_net_init(struct net *net)
 
 	tbl = netns_core_table;
 	if (!net_eq(net, &init_net)) {
+		int i;
+
 		tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
 		if (tbl == NULL)
 			goto err_dup;
 
-		tbl[0].data = &net->core.sysctl_somaxconn;
+		/* Update the variables to point into the current struct net */
+		for (i = 0; i < ARRAY_SIZE(netns_core_table) - 1; i++) {
+			tbl[i].data += (void *)net - (void *)&init_net;
 
-		/* Don't export any sysctls to unprivileged users */
-		if (net->user_ns != &init_user_ns) {
-			tbl[0].procname = NULL;
+			/* Don't export any sysctls to unprivileged users */
+			if (net->user_ns != &init_user_ns)
+				tbl[i].procname = NULL;
 		}
 	}
 
-- 
2.30.0

[PATCH net-next 2/3] net: core: Namespace-ify sysctl_wmem_default and sysctl_rmem_default

[menglong8.dong]

From: Menglong Dong <dong.menglong@zte.com.cn>

For now, sysctl_wmem_default and sysctl_rmem_default are globally
unified. It's not convenient in some case. For example, when we
use docker and try to control the default udp socket receive buffer
for each container.

For that reason, make sysctl_wmem_default and sysctl_rmem_default
per-namespace.

Signed-off-by: Menglong Dong <dong.menglong@zte.com.cn>
---
 include/net/netns/core.h   |  2 ++
 include/net/sock.h         |  3 ---
 net/core/net_namespace.c   |  2 ++
 net/core/sock.c            |  6 ++----
 net/core/sysctl_net_core.c | 32 ++++++++++++++++----------------
 net/ipv4/ip_output.c       |  2 +-
 6 files changed, 23 insertions(+), 24 deletions(-)

diff --git a/include/net/netns/core.h b/include/net/netns/core.h
index 36c2d998a43c..317b47df6d08 100644
--- a/include/net/netns/core.h
+++ b/include/net/netns/core.h
@@ -9,6 +9,8 @@ struct netns_core {
 	/* core sysctls */
 	struct ctl_table_header	*sysctl_hdr;
 
+	int sysctl_wmem_default;
+	int sysctl_rmem_default;
 	int	sysctl_somaxconn;
 
 #ifdef CONFIG_PROC_FS
diff --git a/include/net/sock.h b/include/net/sock.h
index bdc4323ce53c..b846a6d24459 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -2653,9 +2653,6 @@ extern __u32 sysctl_rmem_max;
 extern int sysctl_tstamp_allow_data;
 extern int sysctl_optmem_max;
 
-extern __u32 sysctl_wmem_default;
-extern __u32 sysctl_rmem_default;
-
 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
 
 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
diff --git a/net/core/net_namespace.c b/net/core/net_namespace.c
index 2ef3b4557f40..eb4ea99131d6 100644
--- a/net/core/net_namespace.c
+++ b/net/core/net_namespace.c
@@ -374,6 +374,8 @@ static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
 
 static int __net_init net_defaults_init_net(struct net *net)
 {
+	net->core.sysctl_rmem_default = SK_RMEM_MAX;
+	net->core.sysctl_wmem_default = SK_WMEM_MAX;
 	net->core.sysctl_somaxconn = SOMAXCONN;
 	return 0;
 }
diff --git a/net/core/sock.c b/net/core/sock.c
index bbcd4b97eddd..2421e4ea1915 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -270,8 +270,6 @@ __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
 EXPORT_SYMBOL(sysctl_wmem_max);
 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
 EXPORT_SYMBOL(sysctl_rmem_max);
-__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
-__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
 
 /* Maximal space eaten by iovec or ancillary data plus some space */
 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
@@ -2970,8 +2968,8 @@ void sock_init_data(struct socket *sock, struct sock *sk)
 	timer_setup(&sk->sk_timer, NULL, 0);
 
 	sk->sk_allocation	=	GFP_KERNEL;
-	sk->sk_rcvbuf		=	sysctl_rmem_default;
-	sk->sk_sndbuf		=	sysctl_wmem_default;
+	sk->sk_rcvbuf		=	sock_net(sk)->core.sysctl_rmem_default;
+	sk->sk_sndbuf		=	sock_net(sk)->core.sysctl_wmem_default;
 	sk->sk_state		=	TCP_CLOSE;
 	sk_set_socket(sk, sock);
 
diff --git a/net/core/sysctl_net_core.c b/net/core/sysctl_net_core.c
index 966d976dee84..5c1c75e42a09 100644
--- a/net/core/sysctl_net_core.c
+++ b/net/core/sysctl_net_core.c
@@ -326,22 +326,6 @@ static struct ctl_table net_core_table[] = {
 		.proc_handler	= proc_dointvec_minmax,
 		.extra1		= &min_rcvbuf,
 	},
-	{
-		.procname	= "wmem_default",
-		.data		= &sysctl_wmem_default,
-		.maxlen		= sizeof(int),
-		.mode		= 0644,
-		.proc_handler	= proc_dointvec_minmax,
-		.extra1		= &min_sndbuf,
-	},
-	{
-		.procname	= "rmem_default",
-		.data		= &sysctl_rmem_default,
-		.maxlen		= sizeof(int),
-		.mode		= 0644,
-		.proc_handler	= proc_dointvec_minmax,
-		.extra1		= &min_rcvbuf,
-	},
 	{
 		.procname	= "dev_weight",
 		.data		= &weight_p,
@@ -584,6 +568,22 @@ static struct ctl_table netns_core_table[] = {
 		.extra1		= SYSCTL_ZERO,
 		.proc_handler	= proc_dointvec_minmax
 	},
+	{
+		.procname	= "wmem_default",
+		.data		= &init_net.core.sysctl_wmem_default,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &min_sndbuf,
+	},
+	{
+		.procname	= "rmem_default",
+		.data		= &init_net.core.sysctl_rmem_default,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &min_rcvbuf,
+	},
 	{ }
 };
 
diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
index 2ed0b01f72f0..0fbdcda6f314 100644
--- a/net/ipv4/ip_output.c
+++ b/net/ipv4/ip_output.c
@@ -1709,7 +1709,7 @@ void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
 
 	sk->sk_protocol = ip_hdr(skb)->protocol;
 	sk->sk_bound_dev_if = arg->bound_dev_if;
-	sk->sk_sndbuf = sysctl_wmem_default;
+	sk->sk_sndbuf = sock_net(sk)->core.sysctl_wmem_default;
 	ipc.sockc.mark = fl4.flowi4_mark;
 	err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
 			     len, 0, &ipc, &rt, MSG_DONTWAIT);
-- 
2.30.0

[PATCH net-next 3/3] net: core: Namespace-ify sysctl_rmem_max and sysctl_wmem_max

[menglong8.dong]

From: Menglong Dong <dong.menglong@zte.com.cn>

For now, sysctl_wmem_max and sysctl_rmem_max are globally unified.
It's not convenient in some case. For example, when we use docker
and try to control the default udp socket receive buffer for each
container.

For that reason, make sysctl_wmem_max and sysctl_rmem_max
per-namespace.

Signed-off-by: Menglong Dong <dong.menglong@zte.com.cn>
---
 include/net/netns/core.h        |  2 ++
 include/net/sock.h              |  3 ---
 net/core/filter.c               |  4 ++--
 net/core/net_namespace.c        |  2 ++
 net/core/sock.c                 | 12 ++++--------
 net/core/sysctl_net_core.c      | 32 ++++++++++++++++----------------
 net/ipv4/tcp_output.c           |  2 +-
 net/netfilter/ipvs/ip_vs_sync.c |  4 ++--
 8 files changed, 29 insertions(+), 32 deletions(-)

diff --git a/include/net/netns/core.h b/include/net/netns/core.h
index 317b47df6d08..b4aecac6e8ce 100644
--- a/include/net/netns/core.h
+++ b/include/net/netns/core.h
@@ -11,6 +11,8 @@ struct netns_core {
 
 	int sysctl_wmem_default;
 	int sysctl_rmem_default;
+	int sysctl_wmem_max;
+	int sysctl_rmem_max;
 	int	sysctl_somaxconn;
 
 #ifdef CONFIG_PROC_FS
diff --git a/include/net/sock.h b/include/net/sock.h
index b846a6d24459..f6b0f2c482ad 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -2647,9 +2647,6 @@ void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
 #define SK_WMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 #define SK_RMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 
-extern __u32 sysctl_wmem_max;
-extern __u32 sysctl_rmem_max;
-
 extern int sysctl_tstamp_allow_data;
 extern int sysctl_optmem_max;
 
diff --git a/net/core/filter.c b/net/core/filter.c
index 255aeee72402..3dca58f6c40c 100644
--- a/net/core/filter.c
+++ b/net/core/filter.c
@@ -4717,13 +4717,13 @@ static int _bpf_setsockopt(struct sock *sk, int level, int optname,
 		/* Only some socketops are supported */
 		switch (optname) {
 		case SO_RCVBUF:
-			val = min_t(u32, val, sysctl_rmem_max);
+			val = min_t(u32, val, sock_net(sk)->core.sysctl_rmem_max);
 			sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 			WRITE_ONCE(sk->sk_rcvbuf,
 				   max_t(int, val * 2, SOCK_MIN_RCVBUF));
 			break;
 		case SO_SNDBUF:
-			val = min_t(u32, val, sysctl_wmem_max);
+			val = min_t(u32, val, sock_net(sk)->core.sysctl_wmem_max);
 			sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 			WRITE_ONCE(sk->sk_sndbuf,
 				   max_t(int, val * 2, SOCK_MIN_SNDBUF));
diff --git a/net/core/net_namespace.c b/net/core/net_namespace.c
index eb4ea99131d6..552e3c5b2a41 100644
--- a/net/core/net_namespace.c
+++ b/net/core/net_namespace.c
@@ -376,6 +376,8 @@ static int __net_init net_defaults_init_net(struct net *net)
 {
 	net->core.sysctl_rmem_default = SK_RMEM_MAX;
 	net->core.sysctl_wmem_default = SK_WMEM_MAX;
+	net->core.sysctl_rmem_max = SK_RMEM_MAX;
+	net->core.sysctl_wmem_max = SK_WMEM_MAX;
 	net->core.sysctl_somaxconn = SOMAXCONN;
 	return 0;
 }
diff --git a/net/core/sock.c b/net/core/sock.c
index 2421e4ea1915..eb7eaaa840ce 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -265,12 +265,6 @@ static struct lock_class_key af_wlock_keys[AF_MAX];
 static struct lock_class_key af_elock_keys[AF_MAX];
 static struct lock_class_key af_kern_callback_keys[AF_MAX];
 
-/* Run time adjustable parameters. */
-__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
-EXPORT_SYMBOL(sysctl_wmem_max);
-__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
-EXPORT_SYMBOL(sysctl_rmem_max);
-
 /* Maximal space eaten by iovec or ancillary data plus some space */
 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
 EXPORT_SYMBOL(sysctl_optmem_max);
@@ -877,7 +871,7 @@ int sock_setsockopt(struct socket *sock, int level, int optname,
 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
 		 * are treated in BSD as hints
 		 */
-		val = min_t(u32, val, sysctl_wmem_max);
+		val = min_t(u32, val, sock_net(sk)->core.sysctl_wmem_max);
 set_sndbuf:
 		/* Ensure val * 2 fits into an int, to prevent max_t()
 		 * from treating it as a negative value.
@@ -909,7 +903,9 @@ int sock_setsockopt(struct socket *sock, int level, int optname,
 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
 		 * are treated in BSD as hints
 		 */
-		__sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max));
+		__sock_set_rcvbuf(sk,
+				  min_t(u32, val,
+					sock_net(sk)->core.sysctl_rmem_max));
 		break;
 
 	case SO_RCVBUFFORCE:
diff --git a/net/core/sysctl_net_core.c b/net/core/sysctl_net_core.c
index 5c1c75e42a09..30a8e3a324ec 100644
--- a/net/core/sysctl_net_core.c
+++ b/net/core/sysctl_net_core.c
@@ -310,22 +310,6 @@ proc_dolongvec_minmax_bpf_restricted(struct ctl_table *table, int write,
 
 static struct ctl_table net_core_table[] = {
 #ifdef CONFIG_NET
-	{
-		.procname	= "wmem_max",
-		.data		= &sysctl_wmem_max,
-		.maxlen		= sizeof(int),
-		.mode		= 0644,
-		.proc_handler	= proc_dointvec_minmax,
-		.extra1		= &min_sndbuf,
-	},
-	{
-		.procname	= "rmem_max",
-		.data		= &sysctl_rmem_max,
-		.maxlen		= sizeof(int),
-		.mode		= 0644,
-		.proc_handler	= proc_dointvec_minmax,
-		.extra1		= &min_rcvbuf,
-	},
 	{
 		.procname	= "dev_weight",
 		.data		= &weight_p,
@@ -584,6 +568,22 @@ static struct ctl_table netns_core_table[] = {
 		.proc_handler	= proc_dointvec_minmax,
 		.extra1		= &min_rcvbuf,
 	},
+	{
+		.procname	= "wmem_max",
+		.data		= &init_net.core.sysctl_wmem_max,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &min_sndbuf,
+	},
+	{
+		.procname	= "rmem_max",
+		.data		= &init_net.core.sysctl_rmem_max,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &min_rcvbuf,
+	},
 	{ }
 };
 
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index f322e798a351..8c1b2b0e6211 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -241,7 +241,7 @@ void tcp_select_initial_window(const struct sock *sk, int __space, __u32 mss,
 	if (wscale_ok) {
 		/* Set window scaling on max possible window */
 		space = max_t(u32, space, sock_net(sk)->ipv4.sysctl_tcp_rmem[2]);
-		space = max_t(u32, space, sysctl_rmem_max);
+		space = max_t(u32, space, sock_net(sk)->core.sysctl_rmem_max);
 		space = min_t(u32, space, *window_clamp);
 		*rcv_wscale = clamp_t(int, ilog2(space) - 15,
 				      0, TCP_MAX_WSCALE);
diff --git a/net/netfilter/ipvs/ip_vs_sync.c b/net/netfilter/ipvs/ip_vs_sync.c
index 9d43277b8b4f..2e7e10b76c36 100644
--- a/net/netfilter/ipvs/ip_vs_sync.c
+++ b/net/netfilter/ipvs/ip_vs_sync.c
@@ -1280,12 +1280,12 @@ static void set_sock_size(struct sock *sk, int mode, int val)
 	lock_sock(sk);
 	if (mode) {
 		val = clamp_t(int, val, (SOCK_MIN_SNDBUF + 1) / 2,
-			      sysctl_wmem_max);
+			      sock_net(sk)->core.sysctl_wmem_max);
 		sk->sk_sndbuf = val * 2;
 		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 	} else {
 		val = clamp_t(int, val, (SOCK_MIN_RCVBUF + 1) / 2,
-			      sysctl_rmem_max);
+			      sock_net(sk)->core.sysctl_rmem_max);
 		sk->sk_rcvbuf = val * 2;
 		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 	}
-- 
2.30.0

Re: [PATCH net-next 3/3] net: core: Namespace-ify sysctl_rmem_max and sysctl_wmem_max

[Florian Westphal]

menglong8.dong@gmail.com <menglong8.dong@gmail.com> wrote:
> From: Menglong Dong <dong.menglong@zte.com.cn>
> 
> For now, sysctl_wmem_max and sysctl_rmem_max are globally unified.
> It's not convenient in some case. For example, when we use docker
> and try to control the default udp socket receive buffer for each
> container.
> 
> For that reason, make sysctl_wmem_max and sysctl_rmem_max
> per-namespace.

I think having those values be restricted by init netns is a desirable
property.

Re: [PATCH net-next 3/3] net: core: Namespace-ify sysctl_rmem_max and sysctl_wmem_max

[Menglong Dong]

Hello~

On Wed, Jan 20, 2021 at 6:46 PM Florian Westphal <fw@strlen.de> wrote:
>
> >
> > For that reason, make sysctl_wmem_max and sysctl_rmem_max
> > per-namespace.
>
> I think having those values be restricted by init netns is a desirable
> property.

I just thought that having these values per-namespace can be more flexible,
and users can have more choices. Is there any bad influence that I didn't
realize?

Thanks~
Menglong Dong

Re: [PATCH net-next 3/3] net: core: Namespace-ify sysctl_rmem_max and sysctl_wmem_max

[Nicolas Dichtel]

Le 20/01/2021 à 14:28, Menglong Dong a écrit :
[snip]
>>> For that reason, make sysctl_wmem_max and sysctl_rmem_max
>>> per-namespace.
>>
>> I think having those values be restricted by init netns is a desirable
>> property.
> 
> I just thought that having these values per-namespace can be more flexible,
> and users can have more choices. Is there any bad influence that I didn't
> realize?
You can have a look here:
https://lore.kernel.org/netdev/1501495652.1876.17.camel@edumazet-glaptop3.roam.corp.google.com/
https://patchwork.ozlabs.org/project/netdev/patch/20170726170333.24580-1-mcroce@redhat.com/


Regards,
Nicolas