diff mbox series

[v3] net: fully namespace net.core.{r,w}mem_{default,max} sysctls

Message ID 20250321043504.9729-1-danny@orbstack.dev (mailing list archive)
State New
Delegated to: Netdev Maintainers
Headers show
Series [v3] net: fully namespace net.core.{r,w}mem_{default,max} sysctls | expand

Checks

Context Check Description
netdev/series_format warning Single patches do not need cover letters; Target tree name not specified in the subject
netdev/tree_selection success Guessed tree name to be net-next, async
netdev/ynl success Generated files up to date; no warnings/errors; no diff in generated;
netdev/fixes_present success Fixes tag not required for -next series
netdev/header_inline success No static functions without inline keyword in header files
netdev/build_32bit success Errors and warnings before: 38 this patch: 38
netdev/build_tools success Errors and warnings before: 26 (+0) this patch: 26 (+0)
netdev/cc_maintainers warning 12 maintainers not CCed: kuniyu@amazon.com lvs-devel@vger.kernel.org ncardwell@google.com linux-kselftest@vger.kernel.org coreteam@netfilter.org netfilter-devel@vger.kernel.org willemb@google.com pablo@netfilter.org kadlec@netfilter.org ja@ssi.bg kernelxing@tencent.com shuah@kernel.org
netdev/build_clang success Errors and warnings before: 73 this patch: 73
netdev/verify_signedoff success Signed-off-by tag matches author and committer
netdev/deprecated_api success None detected
netdev/check_selftest success net selftest script(s) already in Makefile
netdev/verify_fixes success No Fixes tag
netdev/build_allmodconfig_warn success Errors and warnings before: 4240 this patch: 4240
netdev/checkpatch warning WARNING: Missing a blank line after declarations WARNING: line length of 102 exceeds 80 columns WARNING: line length of 85 exceeds 80 columns WARNING: line length of 86 exceeds 80 columns WARNING: line length of 88 exceeds 80 columns WARNING: line length of 89 exceeds 80 columns WARNING: line length of 90 exceeds 80 columns
netdev/build_clang_rust success No Rust files in patch. Skipping build
netdev/kdoc success Errors and warnings before: 17 this patch: 17
netdev/source_inline success Was 0 now: 0
netdev/contest warning net-next-2025-03-21--18-00 (tests: 896)

Commit Message

Danny Lin March 21, 2025, 4:34 a.m. UTC
This builds on commit 19249c0724f2 ("net: make net.core.{r,w}mem_{default,max} namespaced")
by adding support for writing the sysctls from within net namespaces,
rather than only reading the values that were set in init_net. These are
relatively commonly-used sysctls, so programs may try to set them without
knowing that they're in a container. It can be surprising for such attempts
to fail with EACCES.

Unlike other net sysctls that were converted to namespaced ones, many
systems have a sysctl.conf (or other configs) that globally write to
net.core.rmem_default on boot and expect the value to propagate to
containers, and programs running in containers may depend on the increased
buffer sizes in order to work properly. This means that namespacing the
sysctls and using the kernel default values in each new netns would break
existing workloads.

As a compromise, inherit the initial net.core.*mem_* values from the
current process' netns when creating a new netns. This is not standard
behavior for most netns sysctls, but it avoids breaking existing workloads.

Signed-off-by: Danny Lin <danny@orbstack.dev>
---
 include/net/netns/core.h                    |  5 +++++
 include/net/sock.h                          |  6 -----
 net/core/net_namespace.c                    | 21 +++++++++++++++++
 net/core/sock.c                             | 16 ++++---------
 net/core/sysctl_net_core.c                  | 25 ++++-----------------
 net/ipv4/ip_output.c                        |  2 +-
 net/ipv4/tcp_output.c                       |  2 +-
 net/netfilter/ipvs/ip_vs_sync.c             |  4 ++--
 tools/testing/selftests/net/netns-sysctl.sh | 24 ++++++++++++++------
 9 files changed, 55 insertions(+), 50 deletions(-)
diff mbox series

Patch

diff --git a/include/net/netns/core.h b/include/net/netns/core.h
index 9b36f0ff0c20..0459523602cb 100644
--- a/include/net/netns/core.h
+++ b/include/net/netns/core.h
@@ -17,6 +17,11 @@  struct netns_core {
 	u8	sysctl_txrehash;
 	u8	sysctl_tstamp_allow_data;
 
+	u32 sysctl_wmem_max;
+	u32 sysctl_rmem_max;
+	u32 sysctl_wmem_default;
+	u32 sysctl_rmem_default;
+
 #ifdef CONFIG_PROC_FS
 	struct prot_inuse __percpu *prot_inuse;
 #endif
diff --git a/include/net/sock.h b/include/net/sock.h
index 8daf1b3b12c6..7aeddba44919 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -2849,12 +2849,6 @@  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 __u32 sysctl_wmem_default;
-extern __u32 sysctl_rmem_default;
-
 #define SKB_FRAG_PAGE_ORDER	get_order(32768)
 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
 
diff --git a/net/core/net_namespace.c b/net/core/net_namespace.c
index 4303f2a49262..61a417f499e9 100644
--- a/net/core/net_namespace.c
+++ b/net/core/net_namespace.c
@@ -317,6 +317,27 @@  static __net_init void preinit_net_sysctl(struct net *net)
 	net->core.sysctl_optmem_max = 128 * 1024;
 	net->core.sysctl_txrehash = SOCK_TXREHASH_ENABLED;
 	net->core.sysctl_tstamp_allow_data = 1;
+
+	/*
+	 * net.core.{r,w}mem_{default,max} used to be non-namespaced.
+	 * For backward compatibility, inherit values from the current netns
+	 * when creating a new one, so that setting them in init_net
+	 * affects new namespaces like it used to. This avoids causing
+	 * surprising performance regressions for namespaced applications
+	 * relying on tuned rmem/wmem.
+	 */
+	if (net == &init_net) {
+		net->core.sysctl_wmem_max = SK_WMEM_MAX;
+		net->core.sysctl_rmem_max = SK_RMEM_MAX;
+		net->core.sysctl_wmem_default = SK_WMEM_MAX;
+		net->core.sysctl_rmem_default = SK_RMEM_MAX;
+	} else {
+		struct net *current_net = current->nsproxy->net_ns;
+		net->core.sysctl_wmem_max = current_net->core.sysctl_wmem_max;
+		net->core.sysctl_rmem_max = current_net->core.sysctl_rmem_max;
+		net->core.sysctl_wmem_default = current_net->core.sysctl_wmem_default;
+		net->core.sysctl_rmem_default = current_net->core.sysctl_rmem_default;
+	}
 }
 
 /* init code that must occur even if setup_net() is not called. */
diff --git a/net/core/sock.c b/net/core/sock.c
index 323892066def..88694cd59abd 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -278,14 +278,6 @@  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);
-__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
-__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
-
 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
 EXPORT_SYMBOL_GPL(memalloc_socks_key);
 
@@ -1333,7 +1325,7 @@  int sk_setsockopt(struct sock *sk, int level, int optname,
 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
 		 * are treated in BSD as hints
 		 */
-		val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
+		val = min_t(u32, val, READ_ONCE(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.
@@ -1365,7 +1357,7 @@  int sk_setsockopt(struct sock *sk, 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, READ_ONCE(sysctl_rmem_max)));
+		__sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sock_net(sk)->core.sysctl_rmem_max)));
 		break;
 
 	case SO_RCVBUFFORCE:
@@ -3618,8 +3610,8 @@  void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
 	timer_setup(&sk->sk_timer, NULL, 0);
 
 	sk->sk_allocation	=	GFP_KERNEL;
-	sk->sk_rcvbuf		=	READ_ONCE(sysctl_rmem_default);
-	sk->sk_sndbuf		=	READ_ONCE(sysctl_wmem_default);
+	sk->sk_rcvbuf		=	READ_ONCE(sock_net(sk)->core.sysctl_rmem_default);
+	sk->sk_sndbuf		=	READ_ONCE(sock_net(sk)->core.sysctl_wmem_default);
 	sk->sk_state		=	TCP_CLOSE;
 	sk->sk_use_task_frag	=	true;
 	sk_set_socket(sk, sock);
diff --git a/net/core/sysctl_net_core.c b/net/core/sysctl_net_core.c
index c7769ee0d9c5..aedc249bf0e2 100644
--- a/net/core/sysctl_net_core.c
+++ b/net/core/sysctl_net_core.c
@@ -676,21 +676,9 @@  static struct ctl_table netns_core_table[] = {
 		.extra2		= SYSCTL_ONE,
 		.proc_handler	= proc_dou8vec_minmax,
 	},
-	{
-		.procname	= "tstamp_allow_data",
-		.data		= &init_net.core.sysctl_tstamp_allow_data,
-		.maxlen		= sizeof(u8),
-		.mode		= 0644,
-		.proc_handler	= proc_dou8vec_minmax,
-		.extra1		= SYSCTL_ZERO,
-		.extra2		= SYSCTL_ONE
-	},
-	/* sysctl_core_net_init() will set the values after this
-	 * to readonly in network namespaces
-	 */
 	{
 		.procname	= "wmem_max",
-		.data		= &sysctl_wmem_max,
+		.data		= &init_net.core.sysctl_wmem_max,
 		.maxlen		= sizeof(int),
 		.mode		= 0644,
 		.proc_handler	= proc_dointvec_minmax,
@@ -698,7 +686,7 @@  static struct ctl_table netns_core_table[] = {
 	},
 	{
 		.procname	= "rmem_max",
-		.data		= &sysctl_rmem_max,
+		.data		= &init_net.core.sysctl_rmem_max,
 		.maxlen		= sizeof(int),
 		.mode		= 0644,
 		.proc_handler	= proc_dointvec_minmax,
@@ -706,7 +694,7 @@  static struct ctl_table netns_core_table[] = {
 	},
 	{
 		.procname	= "wmem_default",
-		.data		= &sysctl_wmem_default,
+		.data		= &init_net.core.sysctl_wmem_default,
 		.maxlen		= sizeof(int),
 		.mode		= 0644,
 		.proc_handler	= proc_dointvec_minmax,
@@ -714,7 +702,7 @@  static struct ctl_table netns_core_table[] = {
 	},
 	{
 		.procname	= "rmem_default",
-		.data		= &sysctl_rmem_default,
+		.data		= &init_net.core.sysctl_rmem_default,
 		.maxlen		= sizeof(int),
 		.mode		= 0644,
 		.proc_handler	= proc_dointvec_minmax,
@@ -748,13 +736,8 @@  static __net_init int sysctl_core_net_init(struct net *net)
 			goto err_dup;
 
 		for (i = 0; i < table_size; ++i) {
-			if (tbl[i].data == &sysctl_wmem_max)
-				break;
-
 			tbl[i].data += (char *)net - (char *)&init_net;
 		}
-		for (; i < table_size; ++i)
-			tbl[i].mode &= ~0222;
 	}
 
 	net->core.sysctl_hdr = register_net_sysctl_sz(net, "net/core", tbl, table_size);
diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
index 6e18d7ec5062..7b7c5e4ea5d8 100644
--- a/net/ipv4/ip_output.c
+++ b/net/ipv4/ip_output.c
@@ -1643,7 +1643,7 @@  void ip_send_unicast_reply(struct sock *sk, const struct sock *orig_sk,
 
 	sk->sk_protocol = ip_hdr(skb)->protocol;
 	sk->sk_bound_dev_if = arg->bound_dev_if;
-	sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
+	sk->sk_sndbuf = READ_ONCE(net->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);
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index e0a4e5432399..8cb319315e76 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, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2]));
-		space = max_t(u32, space, READ_ONCE(sysctl_rmem_max));
+		space = max_t(u32, space, READ_ONCE(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 3402675bf521..62f30d5c25c7 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,
-			      READ_ONCE(sysctl_wmem_max));
+			      READ_ONCE(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,
-			      READ_ONCE(sysctl_rmem_max));
+			      READ_ONCE(sock_net(sk)->core.sysctl_rmem_max));
 		sk->sk_rcvbuf = val * 2;
 		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 	}
diff --git a/tools/testing/selftests/net/netns-sysctl.sh b/tools/testing/selftests/net/netns-sysctl.sh
index 45c34a3b9aae..bb6f173a28e2 100755
--- a/tools/testing/selftests/net/netns-sysctl.sh
+++ b/tools/testing/selftests/net/netns-sysctl.sh
@@ -20,21 +20,31 @@  fail() {
 setup_ns test_ns
 
 for sc in {r,w}mem_{default,max}; do
-	# check that this is writable in a netns
+	initial_value="$(sysctl -n "net.core.$sc")"
+
+	# check that this is writable in the init netns
 	[ -w "/proc/sys/net/core/$sc" ] ||
 		fail "$sc isn't writable in the init netns!"
 
-	# change the value in the host netns
+	# change the value in the init netns
 	sysctl -qw "net.core.$sc=300000" ||
 		fail "Can't write $sc in init netns!"
 
-	# check that the value is read from the init netns
-	[ "$(ip netns exec $test_ns sysctl -n "net.core.$sc")" -eq 300000 ] ||
+	# check that the value did not change in the test netns
+	[ "$(ip netns exec $test_ns sysctl -n "net.core.$sc")" -eq "$initial_value" ] ||
 		fail "Value for $sc mismatch!"
 
-	# check that this isn't writable in a netns
-	ip netns exec $test_ns [ -w "/proc/sys/net/core/$sc" ] &&
-		fail "$sc is writable in a netns!"
+	# check that this is also writable in the test netns
+	ip netns exec $test_ns [ -w "/proc/sys/net/core/$sc" ] ||
+		fail "$sc isn't writable in the test netns!"
+
+	# change the value in the test netns
+	ip netns exec $test_ns sysctl -qw "net.core.$sc=200000" ||
+		fail "Can't write $sc in test netns!"
+
+	# check that the value is read from the test netns
+	[ "$(ip netns exec $test_ns sysctl -n "net.core.$sc")" -eq 200000 ] ||
+		fail "Value for $sc mismatch!"
 done
 
 echo 'Test passed OK'