Merge branch 'hash-rethink'

Akhmat Karakotov says:

====================
Make hash rethink configurable

As it was shown in the report by Alexander Azimov, hash rethink at the
client-side may lead to connection timeout toward stateful anycast
services. Tom Herbert created a patchset to address this issue by applying
hash rethink only after a negative routing event (3RTOs) [1]. This change
also affects server-side behavior, which we found undesirable. This
patchset changes defaults in a way to make them safe: hash rethink at the
client-side is disabled and enabled at the server-side upon each RTO
event or in case of duplicate acknowledgments.

This patchset provides two options to change default behaviour. The hash
rethink may be disabled at the server-side by the new sysctl option.
Changes in the sysctl option don't affect default behavior at the
client-side.

Hash rethink can also be enabled/disabled with socket option or bpf
syscalls which ovewrite both default and sysctl settings. This socket
option is available on both client and server-side. This should provide
mechanics to enable hash rethink inside administrative domain, such as DC,
where hash rethink at the client-side can be desirable.

[1] https://lore.kernel.org/netdev/20210809185314.38187-1-tom@herbertland.com/

v2:
	- Changed sysctl default to ENABLED in all patches. Reduced sysctl
	  and socket option size to u8. Fixed netns bug reported by kernel
	  test robot.

v3:
	- Fixed bug with bad u8 comparison. Moved sk_txrehash to use less
	  bytes in struct. Added WRITE_ONCE() in setsockopt in and
	  READ_ONCE() in tcp_rtx_synack.

v4:
	- Rebase and add documentation for sysctl option.

v5:
	- Move sk_txrehash out of busy poll ifdef.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Documentation/admin-guide/sysctl/net.rst

@@ -365,6 +365,15 @@ new netns has been created.
 
 Default : 0  (for compatibility reasons)
 
+txrehash
+--------
+
+Controls default hash rethink behaviour on listening socket when SO_TXREHASH
+option is set to SOCK_TXREHASH_DEFAULT (i. e. not overridden by setsockopt).
+
+If set to 1 (default), hash rethink is performed on listening socket.
+If set to 0, hash rethink is not performed.
+
 2. /proc/sys/net/unix - Parameters for Unix domain sockets
 ----------------------------------------------------------
 

arch/alpha/include/uapi/asm/socket.h

@@ -133,6 +133,8 @@
 
 #define SO_RESERVE_MEM		73
 
+#define SO_TXREHASH		74
+
 #if !defined(__KERNEL__)
 
 #if __BITS_PER_LONG == 64

arch/mips/include/uapi/asm/socket.h

@@ -144,6 +144,8 @@
 
 #define SO_RESERVE_MEM		73
 
+#define SO_TXREHASH		74
+
 #if !defined(__KERNEL__)
 
 #if __BITS_PER_LONG == 64

arch/parisc/include/uapi/asm/socket.h

@@ -125,6 +125,8 @@
 
 #define SO_RESERVE_MEM		0x4047
 
+#define SO_TXREHASH		0x4048
+
 #if !defined(__KERNEL__)
 
 #if __BITS_PER_LONG == 64

arch/sparc/include/uapi/asm/socket.h

@@ -126,6 +126,8 @@
 
 #define SO_RESERVE_MEM           0x0052
 
+#define SO_TXREHASH              0x0053
+
 
 #if !defined(__KERNEL__)
 

include/net/netns/core.h

@@ -10,6 +10,7 @@ struct netns_core {
 	struct ctl_table_header	*sysctl_hdr;
 
 	int	sysctl_somaxconn;
+	u8	sysctl_txrehash;
 
 #ifdef CONFIG_PROC_FS
 	struct prot_inuse __percpu *prot_inuse;

include/net/sock.h

@@ -316,6 +316,7 @@ struct sk_filter;
   *	@sk_rcvtimeo: %SO_RCVTIMEO setting
   *	@sk_sndtimeo: %SO_SNDTIMEO setting
   *	@sk_txhash: computed flow hash for use on transmit
+  *	@sk_txrehash: enable TX hash rethink
   *	@sk_filter: socket filtering instructions
   *	@sk_timer: sock cleanup timer
   *	@sk_stamp: time stamp of last packet received
@@ -491,6 +492,7 @@ struct sock {
 	u32			sk_ack_backlog;
 	u32			sk_max_ack_backlog;
 	kuid_t			sk_uid;
+	u8			sk_txrehash;
 #ifdef CONFIG_NET_RX_BUSY_POLL
 	u8			sk_prefer_busy_poll;
 	u16			sk_busy_poll_budget;
@@ -587,6 +589,18 @@ static inline bool sk_user_data_is_nocopy(const struct sock *sk)
 			   __tmp | SK_USER_DATA_NOCOPY);		\
 })
 
+static inline
+struct net *sock_net(const struct sock *sk)
+{
+	return read_pnet(&sk->sk_net);
+}
+
+static inline
+void sock_net_set(struct sock *sk, struct net *net)
+{
+	write_pnet(&sk->sk_net, net);
+}
+
 /*
  * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
  * or not whether his port will be reused by someone else. SK_FORCE_REUSE
@@ -2054,7 +2068,7 @@ static inline void sk_set_txhash(struct sock *sk)
 
 static inline bool sk_rethink_txhash(struct sock *sk)
 {
-	if (sk->sk_txhash) {
+	if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) {
 		sk_set_txhash(sk);
 		return true;
 	}
@@ -2704,18 +2718,6 @@ static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
 	__kfree_skb(skb);
 }
 
-static inline
-struct net *sock_net(const struct sock *sk)
-{
-	return read_pnet(&sk->sk_net);
-}
-
-static inline
-void sock_net_set(struct sock *sk, struct net *net)
-{
-	write_pnet(&sk->sk_net, net);
-}
-
 static inline bool
 skb_sk_is_prefetched(struct sk_buff *skb)
 {

include/uapi/asm-generic/socket.h

@@ -128,6 +128,8 @@
 
 #define SO_RESERVE_MEM		73
 
+#define SO_TXREHASH		74
+
 #if !defined(__KERNEL__)
 
 #if __BITS_PER_LONG == 64 || (defined(__x86_64__) && defined(__ILP32__))

include/uapi/linux/socket.h

@@ -31,4 +31,8 @@ struct __kernel_sockaddr_storage {
 
 #define SOCK_BUF_LOCK_MASK (SOCK_SNDBUF_LOCK | SOCK_RCVBUF_LOCK)
 
+#define SOCK_TXREHASH_DEFAULT	((u8)-1)
+#define SOCK_TXREHASH_DISABLED	0
+#define SOCK_TXREHASH_ENABLED	1
+
 #endif /* _UAPI_LINUX_SOCKET_H */

net/core/filter.c

@@ -5091,6 +5091,13 @@ static int _bpf_setsockopt(struct sock *sk, int level, int optname,
 		case SO_REUSEPORT:
 			sk->sk_reuseport = valbool;
 			break;
+		case SO_TXREHASH:
+			if (val < -1 || val > 1) {
+				ret = -EINVAL;
+				break;
+			}
+			sk->sk_txrehash = (u8)val;
+			break;
 		default:
 			ret = -EINVAL;
 		}
@@ -5269,6 +5276,9 @@ static int _bpf_getsockopt(struct sock *sk, int level, int optname,
 		case SO_REUSEPORT:
 			*((int *)optval) = sk->sk_reuseport;
 			break;
+		case SO_TXREHASH:
+			*((int *)optval) = sk->sk_txrehash;
+			break;
 		default:
 			goto err_clear;
 		}

net/core/net_namespace.c

@@ -364,6 +364,8 @@ 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_somaxconn = SOMAXCONN;
+	net->core.sysctl_txrehash = SOCK_TXREHASH_ENABLED;
+
 	return 0;
 }
 

net/core/sock.c

@@ -1447,6 +1447,15 @@ int sock_setsockopt(struct socket *sock, int level, int optname,
 		break;
 	}
 
+	case SO_TXREHASH:
+		if (val < -1 || val > 1) {
+			ret = -EINVAL;
+			break;
+		}
+		/* Paired with READ_ONCE() in tcp_rtx_synack() */
+		WRITE_ONCE(sk->sk_txrehash, (u8)val);
+		break;
+
 	default:
 		ret = -ENOPROTOOPT;
 		break;
@@ -1834,6 +1843,10 @@ int sock_getsockopt(struct socket *sock, int level, int optname,
 		v.val = sk->sk_reserved_mem;
 		break;
 
+	case SO_TXREHASH:
+		v.val = sk->sk_txrehash;
+		break;
+
 	default:
 		/* We implement the SO_SNDLOWAT etc to not be settable
 		 * (1003.1g 7).
@@ -3279,6 +3292,7 @@ void sock_init_data(struct socket *sock, struct sock *sk)
 	sk->sk_pacing_rate = ~0UL;
 	WRITE_ONCE(sk->sk_pacing_shift, 10);
 	sk->sk_incoming_cpu = -1;
+	sk->sk_txrehash = SOCK_TXREHASH_DEFAULT;
 
 	sk_rx_queue_clear(sk);
 	/*

net/core/sysctl_net_core.c

@@ -593,6 +593,15 @@ static struct ctl_table netns_core_table[] = {
 		.extra1		= SYSCTL_ZERO,
 		.proc_handler	= proc_dointvec_minmax
 	},
+	{
+		.procname	= "txrehash",
+		.data		= &init_net.core.sysctl_txrehash,
+		.maxlen		= sizeof(u8),
+		.mode		= 0644,
+		.extra1		= SYSCTL_ZERO,
+		.extra2		= SYSCTL_ONE,
+		.proc_handler	= proc_dou8vec_minmax,
+	},
 	{ }
 };
 
@@ -611,7 +620,7 @@ __setup("fb_tunnels=", fb_tunnels_only_for_init_net_sysctl_setup);
 
 static __net_init int sysctl_core_net_init(struct net *net)
 {
-	struct ctl_table *tbl;
+	struct ctl_table *tbl, *tmp;
 
 	tbl = netns_core_table;
 	if (!net_eq(net, &init_net)) {
@@ -619,7 +628,8 @@ static __net_init int sysctl_core_net_init(struct net *net)
 		if (tbl == NULL)
 			goto err_dup;
 
-		tbl[0].data = &net->core.sysctl_somaxconn;
+		for (tmp = tbl; tmp->procname; tmp++)
+			tmp->data += (char *)net - (char *)&init_net;
 
 		/* Don't export any sysctls to unprivileged users */
 		if (net->user_ns != &init_user_ns) {

net/ipv4/inet_connection_sock.c

@@ -1046,6 +1046,9 @@ int inet_csk_listen_start(struct sock *sk)
 	sk->sk_ack_backlog = 0;
 	inet_csk_delack_init(sk);
 
+	if (sk->sk_txrehash == SOCK_TXREHASH_DEFAULT)
+		sk->sk_txrehash = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
+
 	/* There is race window here: we announce ourselves listening,
 	 * but this transition is still not validated by get_port().
 	 * It is OK, because this socket enters to hash table only

net/ipv4/tcp_output.c

@@ -4092,6 +4092,8 @@ int tcp_rtx_synack(const struct sock *sk, struct request_sock *req)
 	struct flowi fl;
 	int res;
 
+	/* Paired with WRITE_ONCE() in sock_setsockopt() */
+	if (READ_ONCE(sk->sk_txrehash) == SOCK_TXREHASH_ENABLED)
 		tcp_rsk(req)->txhash = net_tx_rndhash();
 	res = af_ops->send_synack(sk, NULL, &fl, req, NULL, TCP_SYNACK_NORMAL,
 				  NULL);