Merge branch 'net-timestamp-next'

Willem de Bruijn says:

====================
net-timestamp: new tx tstamps and tcp

Extend socket tx timestamping:
- allow multiple types of software timestamps aside from send (1)
- add software timestamp on enter packet scheduling (4)
- add software timestamp for TCP (5)
- add software timestamp for TCP on ACK (6)

The sk_flags option space is nearly exhausted. Also move the
many timestamp options to a new sk->sk_tstamps (2).

To disambiguate data when tstamps may arrive out of order,
optionally return a sequential ID assigned at send (3).

Extend Linux tx timestamping to monitoring of latency
incurred within the kernel stack and to protocols embedded in TCP.
Complex kernel setups may have multiple layers of queueing, including
multiple instances of packet scheduling, and many classes per layer.
Many applications embed discrete payloads into TCP bytestreams for
reliability, flow control, etcetera. Detecting application tail
latency in such scenarios relies on identifying the exact queue
responsible if on the host, or the network latency if otherwise.

Changelog:
v4->v5
  - define SCM_TSTAMP_SND == 0, for legacy behavior
  - add TCP tstamps without changing the generated byte stream
    - modify GSO and ACK to find offset: slightly more complex
      than previous invariant that it is the last byte
  - consistent naming of packet scheduling
    - rename SCM_TSTAMP_ENQ to SCM_TSTAMP_SCHED
  - add unique key in ee_data
  - add id field in ee_info to disambiguate tstamps
    - optional, only on new flag SOF_TIMESTAMPING_OPT_ID
    - for bytestream, in bytes

v3->v4
  - (v3 review comment) removed skb->mark packet identification (*A)
  - (v3 review comment) fixed indentation
  - tcp: fixed poll() to return POLLERR on non-zero queue
  - rebased to work without syststamp
  - comments: removed all traces of MSG_TSTAMP_.. (*B)

v2->v3
  - extend the SO_TIMESTAMPING API, instead of defining a new one.
  - add protocol independent support to correlate tstamps with data,
    based on returning skb->mark.
  - removed no-payload optimization and documentation (for now):

    I have a follow-on patch that reintroduces MSG_TSTAMP along with a
    new socket option SOF_TIMESTAMPING_OPT_ONFLAG. This is equivalent
    to sequence setsockopt(<enable>); send(..); setsockopt(<disable>),
    but avoids the need to define a MSG_TSTAMP_<TYPE> for each type.

    I will leave these three patches as follow-on, as this patchset is
    large enough as is.

v1->v2
  - expand timestamping (existing and new) to SOCK_RAW and ping sockets
  - rename sock_errqueue_timestamping to scm_timestamping
  - change timestamp data format: do not add fields to scm_timestamping.
      Doing so could break legacy applications. Instead, communicate
      through an existing, but unused, field in the error message.
  - rename SOF_.._OPT_TX_NO_PAYLOAD to shorter SOF_.._OPT_TSONLY
  - move msg_tstamp test app out of patchset and to github
      git://github.com/wdebruij/kerneltools.git
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

9 files changed, 111 insertions, 42 deletions

diff --git a/net/core/dev.c b/net/core/dev.c
index b370230fe1d3..1c15b189c52b 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -132,6 +132,7 @@
 #include <linux/hashtable.h>
 #include <linux/vmalloc.h>
 #include <linux/if_macvlan.h>
+#include <linux/errqueue.h>
 
 #include "net-sysfs.h"
 
@@ -2876,6 +2877,9 @@ static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
 
 	skb_reset_mac_header(skb);
 
+	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
+		__skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
+
 	/* Disable soft irqs for various locks below. Also
 	 * stops preemption for RCU.
 	 */
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index c1a33033cbe2..3dec0293a7c5 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -3490,10 +3490,10 @@ int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
 }
 EXPORT_SYMBOL(sock_queue_err_skb);
 
-void skb_tstamp_tx(struct sk_buff *orig_skb,
-		struct skb_shared_hwtstamps *hwtstamps)
+void __skb_tstamp_tx(struct sk_buff *orig_skb,
+		     struct skb_shared_hwtstamps *hwtstamps,
+		     struct sock *sk, int tstype)
 {
-	struct sock *sk = orig_skb->sk;
 	struct sock_exterr_skb *serr;
 	struct sk_buff *skb;
 	int err;
@@ -3521,12 +3521,26 @@ void skb_tstamp_tx(struct sk_buff *orig_skb,
 	memset(serr, 0, sizeof(*serr));
 	serr->ee.ee_errno = ENOMSG;
 	serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
+	serr->ee.ee_info = tstype;
+	if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
+		serr->ee.ee_data = skb_shinfo(skb)->tskey;
+		if (sk->sk_protocol == IPPROTO_TCP)
+			serr->ee.ee_data -= sk->sk_tskey;
+	}
 
 	err = sock_queue_err_skb(sk, skb);
 
 	if (err)
 		kfree_skb(skb);
 }
+EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
+
+void skb_tstamp_tx(struct sk_buff *orig_skb,
+		   struct skb_shared_hwtstamps *hwtstamps)
+{
+	return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
+			       SCM_TSTAMP_SND);
+}
 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
 
 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
diff --git a/net/core/sock.c b/net/core/sock.c
index a741163568fa..2714811afbd8 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -848,22 +848,25 @@ set_rcvbuf:
 			ret = -EINVAL;
 			break;
 		}
-		sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
-				  val & SOF_TIMESTAMPING_TX_HARDWARE);
-		sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
-				  val & SOF_TIMESTAMPING_TX_SOFTWARE);
-		sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
-				  val & SOF_TIMESTAMPING_RX_HARDWARE);
+		if (val & SOF_TIMESTAMPING_OPT_ID &&
+		    !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
+			if (sk->sk_protocol == IPPROTO_TCP) {
+				if (sk->sk_state != TCP_ESTABLISHED) {
+					ret = -EINVAL;
+					break;
+				}
+				sk->sk_tskey = tcp_sk(sk)->snd_una;
+			} else {
+				sk->sk_tskey = 0;
+			}
+		}
+		sk->sk_tsflags = val;
 		if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
 			sock_enable_timestamp(sk,
 					      SOCK_TIMESTAMPING_RX_SOFTWARE);
 		else
 			sock_disable_timestamp(sk,
 					       (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
-		sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
-				  val & SOF_TIMESTAMPING_SOFTWARE);
-		sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
-				  val & SOF_TIMESTAMPING_RAW_HARDWARE);
 		break;
 
 	case SO_RCVLOWAT:
@@ -1089,19 +1092,7 @@ int sock_getsockopt(struct socket *sock, int level, int optname,
 		break;
 
 	case SO_TIMESTAMPING:
-		v.val = 0;
-		if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
-			v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
-		if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
-			v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
-		if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
-			v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
-		if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
-			v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
-		if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
-			v.val |= SOF_TIMESTAMPING_SOFTWARE;
-		if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
-			v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
+		v.val = sk->sk_tsflags;
 		break;
 
 	case SO_RCVTIMEO:
diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
index b16556836d66..215af2b155cb 100644
--- a/net/ipv4/ip_output.c
+++ b/net/ipv4/ip_output.c
@@ -855,11 +855,15 @@ static int __ip_append_data(struct sock *sk,
 	unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
 	int csummode = CHECKSUM_NONE;
 	struct rtable *rt = (struct rtable *)cork->dst;
+	u32 tskey = 0;
 
 	skb = skb_peek_tail(queue);
 
 	exthdrlen = !skb ? rt->dst.header_len : 0;
 	mtu = cork->fragsize;
+	if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
+	    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
+		tskey = sk->sk_tskey++;
 
 	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
 
@@ -976,6 +980,8 @@ alloc_new_skb:
 			/* only the initial fragment is time stamped */
 			skb_shinfo(skb)->tx_flags = cork->tx_flags;
 			cork->tx_flags = 0;
+			skb_shinfo(skb)->tskey = tskey;
+			tskey = 0;
 
 			/*
 			 *	Find where to start putting bytes.
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 9d2118e5fbc7..744af67a5989 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -426,6 +426,15 @@ void tcp_init_sock(struct sock *sk)
 }
 EXPORT_SYMBOL(tcp_init_sock);
 
+void tcp_tx_timestamp(struct sock *sk, struct sk_buff *skb)
+{
+	struct skb_shared_info *shinfo = skb_shinfo(skb);
+
+	sock_tx_timestamp(sk, &shinfo->tx_flags);
+	if (shinfo->tx_flags & SKBTX_ANY_SW_TSTAMP)
+		shinfo->tskey = TCP_SKB_CB(skb)->seq + skb->len - 1;
+}
+
 /*
  *	Wait for a TCP event.
  *
@@ -523,7 +532,7 @@ unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait)
 	}
 	/* This barrier is coupled with smp_wmb() in tcp_reset() */
 	smp_rmb();
-	if (sk->sk_err)
+	if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
 		mask |= POLLERR;
 
 	return mask;
@@ -959,8 +968,10 @@ new_segment:
 
 		copied += copy;
 		offset += copy;
-		if (!(size -= copy))
+		if (!(size -= copy)) {
+			tcp_tx_timestamp(sk, skb);
 			goto out;
+		}
 
 		if (skb->len < size_goal || (flags & MSG_OOB))
 			continue;
@@ -1252,8 +1263,10 @@ new_segment:
 
 			from += copy;
 			copied += copy;
-			if ((seglen -= copy) == 0 && iovlen == 0)
+			if ((seglen -= copy) == 0 && iovlen == 0) {
+				tcp_tx_timestamp(sk, skb);
 				goto out;
+			}
 
 			if (skb->len < max || (flags & MSG_OOB) || unlikely(tp->repair))
 				continue;
@@ -1617,6 +1630,9 @@ int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
 	struct sk_buff *skb;
 	u32 urg_hole = 0;
 
+	if (unlikely(flags & MSG_ERRQUEUE))
+		return ip_recv_error(sk, msg, len, addr_len);
+
 	if (sk_can_busy_loop(sk) && skb_queue_empty(&sk->sk_receive_queue) &&
 	    (sk->sk_state == TCP_ESTABLISHED))
 		sk_busy_loop(sk, nonblock);
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 6a2984507755..a3d47af01906 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -74,6 +74,7 @@
 #include <linux/ipsec.h>
 #include <asm/unaligned.h>
 #include <net/netdma.h>
+#include <linux/errqueue.h>
 
 int sysctl_tcp_timestamps __read_mostly = 1;
 int sysctl_tcp_window_scaling __read_mostly = 1;
@@ -3106,6 +3107,11 @@ static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets,
 			tp->retrans_stamp = 0;
 		}
 
+		if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_ACK_TSTAMP) &&
+		    between(skb_shinfo(skb)->tskey, prior_snd_una,
+			    tp->snd_una + 1))
+			__skb_tstamp_tx(skb, NULL, sk, SCM_TSTAMP_ACK);
+
 		if (!fully_acked)
 			break;
 
diff --git a/net/ipv4/tcp_offload.c b/net/ipv4/tcp_offload.c
index 55046ecd083e..f597119fc4e7 100644
--- a/net/ipv4/tcp_offload.c
+++ b/net/ipv4/tcp_offload.c
@@ -14,6 +14,21 @@
 #include <net/tcp.h>
 #include <net/protocol.h>
 
+void tcp_gso_tstamp(struct sk_buff *skb, unsigned int ts_seq, unsigned int seq,
+		    unsigned int mss)
+{
+	while (skb) {
+		if (ts_seq < (__u64) seq + mss) {
+			skb_shinfo(skb)->tx_flags = SKBTX_SW_TSTAMP;
+			skb_shinfo(skb)->tskey = ts_seq;
+			return;
+		}
+
+		skb = skb->next;
+		seq += mss;
+	}
+}
+
 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
 				netdev_features_t features)
 {
@@ -91,6 +106,9 @@ struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
 	th = tcp_hdr(skb);
 	seq = ntohl(th->seq);
 
+	if (unlikely(skb_shinfo(gso_skb)->tx_flags & SKBTX_SW_TSTAMP))
+		tcp_gso_tstamp(segs, skb_shinfo(gso_skb)->tskey, seq, mss);
+
 	newcheck = ~csum_fold((__force __wsum)((__force u32)th->check +
 					       (__force u32)delta));
 
diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
index f5dafe609f8b..315a55d66079 100644
--- a/net/ipv6/ip6_output.c
+++ b/net/ipv6/ip6_output.c
@@ -1157,6 +1157,7 @@ int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
 	int err;
 	int offset = 0;
 	__u8 tx_flags = 0;
+	u32 tskey = 0;
 
 	if (flags&MSG_PROBE)
 		return 0;
@@ -1272,8 +1273,12 @@ emsgsize:
 		}
 	}
 
-	if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW)
+	if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
 		sock_tx_timestamp(sk, &tx_flags);
+		if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
+		    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
+			tskey = sk->sk_tskey++;
+	}
 
 	/*
 	 * Let's try using as much space as possible.
@@ -1397,6 +1402,8 @@ alloc_new_skb:
 			/* Only the initial fragment is time stamped */
 			skb_shinfo(skb)->tx_flags = tx_flags;
 			tx_flags = 0;
+			skb_shinfo(skb)->tskey = tskey;
+			tskey = 0;
 
 			/*
 			 *	Find where to start putting bytes
diff --git a/net/socket.c b/net/socket.c
index d8222c025061..ae89569a2db5 100644
--- a/net/socket.c
+++ b/net/socket.c
@@ -106,6 +106,7 @@
 #include <linux/sockios.h>
 #include <linux/atalk.h>
 #include <net/busy_poll.h>
+#include <linux/errqueue.h>
 
 #ifdef CONFIG_NET_RX_BUSY_POLL
 unsigned int sysctl_net_busy_read __read_mostly;
@@ -612,10 +613,15 @@ EXPORT_SYMBOL(sock_release);
 void sock_tx_timestamp(struct sock *sk, __u8 *tx_flags)
 {
 	*tx_flags = 0;
-	if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
+	if (sk->sk_tsflags & SOF_TIMESTAMPING_TX_HARDWARE)
 		*tx_flags |= SKBTX_HW_TSTAMP;
-	if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
+	if (sk->sk_tsflags & SOF_TIMESTAMPING_TX_SOFTWARE)
 		*tx_flags |= SKBTX_SW_TSTAMP;
+	if (sk->sk_tsflags & SOF_TIMESTAMPING_TX_SCHED)
+		*tx_flags |= SKBTX_SCHED_TSTAMP;
+	if (sk->sk_tsflags & SOF_TIMESTAMPING_TX_ACK)
+		*tx_flags |= SKBTX_ACK_TSTAMP;
+
 	if (sock_flag(sk, SOCK_WIFI_STATUS))
 		*tx_flags |= SKBTX_WIFI_STATUS;
 }
@@ -697,7 +703,7 @@ void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
 	struct sk_buff *skb)
 {
 	int need_software_tstamp = sock_flag(sk, SOCK_RCVTSTAMP);
-	struct timespec ts[3];
+	struct scm_timestamping tss;
 	int empty = 1;
 	struct skb_shared_hwtstamps *shhwtstamps =
 		skb_hwtstamps(skb);
@@ -714,24 +720,25 @@ void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
 			put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP,
 				 sizeof(tv), &tv);
 		} else {
-			skb_get_timestampns(skb, &ts[0]);
+			struct timespec ts;
+			skb_get_timestampns(skb, &ts);
 			put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMPNS,
-				 sizeof(ts[0]), &ts[0]);
+				 sizeof(ts), &ts);
 		}
 	}
 
-
-	memset(ts, 0, sizeof(ts));
-	if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE) &&
-	    ktime_to_timespec_cond(skb->tstamp, ts + 0))
+	memset(&tss, 0, sizeof(tss));
+	if ((sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE ||
+	     skb_shinfo(skb)->tx_flags & SKBTX_ANY_SW_TSTAMP) &&
+	    ktime_to_timespec_cond(skb->tstamp, tss.ts + 0))
 		empty = 0;
 	if (shhwtstamps &&
-	    sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE) &&
-	    ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts + 2))
+	    (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
+	    ktime_to_timespec_cond(shhwtstamps->hwtstamp, tss.ts + 2))
 		empty = 0;
 	if (!empty)
 		put_cmsg(msg, SOL_SOCKET,
-			 SCM_TIMESTAMPING, sizeof(ts), &ts);
+			 SCM_TIMESTAMPING, sizeof(tss), &tss);
 }
 EXPORT_SYMBOL_GPL(__sock_recv_timestamp);