net: infrastructure for hardware time stamping

The additional per-packet information (16 bytes for time stamps, 1
byte for flags) is stored for all packets in the skb_shared_info
struct. This implementation detail is hidden from users of that
information via skb_* accessor functions. A separate struct resp.
union is used for the additional information so that it can be
stored/copied easily outside of skb_shared_info.

Compared to previous implementations (reusing the tstamp field
depending on the context, optional additional structures) this
is the simplest solution. It does not extend sk_buff itself.

TX time stamping is implemented in software if the device driver
doesn't support hardware time stamping.

The new semantic for hardware/software time stamping around
ndo_start_xmit() is based on two assumptions about existing
network device drivers which don't support hardware time
stamping and know nothing about it:
 - they leave the new skb_shared_tx unmodified
 - the keep the connection to the originating socket in skb->sk
   alive, i.e., don't call skb_orphan()

Given that skb_shared_tx is new, the first assumption is safe.
The second is only true for some drivers. As a result, software
TX time stamping currently works with the bnx2 driver, but not
with the unmodified igb driver (the two drivers this patch series
was tested with).
Signed-off-by: Patrick Ohly <patrick.ohly@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

include/linux/skbuff.h

@@ -132,6 +132,57 @@ struct skb_frag_struct {
 	__u32 size;
 };
 
+#define HAVE_HW_TIME_STAMP
+
+/**
+ * skb_shared_hwtstamps - hardware time stamps
+ *
+ * @hwtstamp:	hardware time stamp transformed into duration
+ *		since arbitrary point in time
+ * @syststamp:	hwtstamp transformed to system time base
+ *
+ * Software time stamps generated by ktime_get_real() are stored in
+ * skb->tstamp. The relation between the different kinds of time
+ * stamps is as follows:
+ *
+ * syststamp and tstamp can be compared against each other in
+ * arbitrary combinations.  The accuracy of a
+ * syststamp/tstamp/"syststamp from other device" comparison is
+ * limited by the accuracy of the transformation into system time
+ * base. This depends on the device driver and its underlying
+ * hardware.
+ *
+ * hwtstamps can only be compared against other hwtstamps from
+ * the same device.
+ *
+ * This structure is attached to packets as part of the
+ * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
+ */
+struct skb_shared_hwtstamps {
+	ktime_t	hwtstamp;
+	ktime_t	syststamp;
+};
+
+/**
+ * skb_shared_tx - instructions for time stamping of outgoing packets
+ *
+ * @hardware:		generate hardware time stamp
+ * @software:		generate software time stamp
+ * @in_progress:	device driver is going to provide
+ *			hardware time stamp
+ *
+ * These flags are attached to packets as part of the
+ * &skb_shared_info. Use skb_tx() to get a pointer.
+ */
+union skb_shared_tx {
+	struct {
+		__u8	hardware:1,
+			software:1,
+			in_progress:1;
+	};
+	__u8 flags;
+};
+
 /* This data is invariant across clones and lives at
  * the end of the header data, ie. at skb->end.
  */
@@ -143,10 +194,12 @@ struct skb_shared_info {
 	unsigned short	gso_segs;
 	unsigned short  gso_type;
 	__be32          ip6_frag_id;
+	union skb_shared_tx tx_flags;
 #ifdef CONFIG_HAS_DMA
 	unsigned int	num_dma_maps;
 #endif
 	struct sk_buff	*frag_list;
+	struct skb_shared_hwtstamps hwtstamps;
 	skb_frag_t	frags[MAX_SKB_FRAGS];
 #ifdef CONFIG_HAS_DMA
 	dma_addr_t	dma_maps[MAX_SKB_FRAGS + 1];
@@ -465,6 +518,16 @@ static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
 /* Internal */
 #define skb_shinfo(SKB)	((struct skb_shared_info *)(skb_end_pointer(SKB)))
 
+static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
+{
+	return &skb_shinfo(skb)->hwtstamps;
+}
+
+static inline union skb_shared_tx *skb_tx(struct sk_buff *skb)
+{
+	return &skb_shinfo(skb)->tx_flags;
+}
+
 /**
  *	skb_queue_empty - check if a queue is empty
  *	@list: queue head
@@ -1730,6 +1793,11 @@ static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
 
 extern void skb_init(void);
 
+static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
+{
+	return skb->tstamp;
+}
+
 /**
  *	skb_get_timestamp - get timestamp from a skb
  *	@skb: skb to get stamp from
@@ -1739,11 +1807,18 @@ extern void skb_init(void);
  *	This function converts the offset back to a struct timeval and stores
  *	it in stamp.
  */
-static inline void skb_get_timestamp(const struct sk_buff *skb, struct timeval *stamp)
+static inline void skb_get_timestamp(const struct sk_buff *skb,
+				     struct timeval *stamp)
 {
 	*stamp = ktime_to_timeval(skb->tstamp);
 }
 
+static inline void skb_get_timestampns(const struct sk_buff *skb,
+				       struct timespec *stamp)
+{
+	*stamp = ktime_to_timespec(skb->tstamp);
+}
+
 static inline void __net_timestamp(struct sk_buff *skb)
 {
 	skb->tstamp = ktime_get_real();
@@ -1759,6 +1834,20 @@ static inline ktime_t net_invalid_timestamp(void)
 	return ktime_set(0, 0);
 }
 
+/**
+ * skb_tstamp_tx - queue clone of skb with send time stamps
+ * @orig_skb:	the original outgoing packet
+ * @hwtstamps:	hardware time stamps, may be NULL if not available
+ *
+ * If the skb has a socket associated, then this function clones the
+ * skb (thus sharing the actual data and optional structures), stores
+ * the optional hardware time stamping information (if non NULL) or
+ * generates a software time stamp (otherwise), then queues the clone
+ * to the error queue of the socket.  Errors are silently ignored.
+ */
+extern void skb_tstamp_tx(struct sk_buff *orig_skb,
+			struct skb_shared_hwtstamps *hwtstamps);
+
 extern __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
 extern __sum16 __skb_checksum_complete(struct sk_buff *skb);
 

net/core/dev.c

@@ -1672,10 +1672,21 @@ static int dev_gso_segment(struct sk_buff *skb)
 	return 0;
 }
 
+static void tstamp_tx(struct sk_buff *skb)
+{
+	union skb_shared_tx *shtx =
+		skb_tx(skb);
+	if (unlikely(shtx->software &&
+			!shtx->in_progress)) {
+		skb_tstamp_tx(skb, NULL);
+	}
+}
+
 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
 			struct netdev_queue *txq)
 {
 	const struct net_device_ops *ops = dev->netdev_ops;
+	int rc;
 
 	prefetch(&dev->netdev_ops->ndo_start_xmit);
 	if (likely(!skb->next)) {
@@ -1689,13 +1700,29 @@ int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
 				goto gso;
 		}
 
-		return ops->ndo_start_xmit(skb, dev);
+		rc = ops->ndo_start_xmit(skb, dev);
+		/*
+		 * TODO: if skb_orphan() was called by
+		 * dev->hard_start_xmit() (for example, the unmodified
+		 * igb driver does that; bnx2 doesn't), then
+		 * skb_tx_software_timestamp() will be unable to send
+		 * back the time stamp.
+		 *
+		 * How can this be prevented? Always create another
+		 * reference to the socket before calling
+		 * dev->hard_start_xmit()? Prevent that skb_orphan()
+		 * does anything in dev->hard_start_xmit() by clearing
+		 * the skb destructor before the call and restoring it
+		 * afterwards, then doing the skb_orphan() ourselves?
+		 */
+		if (likely(!rc))
+			tstamp_tx(skb);
+		return rc;
 	}
 
 gso:
 	do {
 		struct sk_buff *nskb = skb->next;
-		int rc;
 
 		skb->next = nskb->next;
 		nskb->next = NULL;
@@ -1705,6 +1732,7 @@ int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
 			skb->next = nskb;
 			return rc;
 		}
+		tstamp_tx(skb);
 		if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
 			return NETDEV_TX_BUSY;
 	} while (skb->next);

net/core/skbuff.c

@@ -55,6 +55,7 @@
 #include <linux/rtnetlink.h>
 #include <linux/init.h>
 #include <linux/scatterlist.h>
+#include <linux/errqueue.h>
 
 #include <net/protocol.h>
 #include <net/dst.h>
@@ -215,7 +216,9 @@ struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
 	shinfo->gso_segs = 0;
 	shinfo->gso_type = 0;
 	shinfo->ip6_frag_id = 0;
+	shinfo->tx_flags.flags = 0;
 	shinfo->frag_list = NULL;
+	memset(&shinfo->hwtstamps, 0, sizeof(shinfo->hwtstamps));
 
 	if (fclone) {
 		struct sk_buff *child = skb + 1;
@@ -2945,6 +2948,44 @@ int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
 }
 EXPORT_SYMBOL_GPL(skb_cow_data);
 
+void skb_tstamp_tx(struct sk_buff *orig_skb,
+		struct skb_shared_hwtstamps *hwtstamps)
+{
+	struct sock *sk = orig_skb->sk;
+	struct sock_exterr_skb *serr;
+	struct sk_buff *skb;
+	int err;
+
+	if (!sk)
+		return;
+
+	skb = skb_clone(orig_skb, GFP_ATOMIC);
+	if (!skb)
+		return;
+
+	if (hwtstamps) {
+		*skb_hwtstamps(skb) =
+			*hwtstamps;
+	} else {
+		/*
+		 * no hardware time stamps available,
+		 * so keep the skb_shared_tx and only
+		 * store software time stamp
+		 */
+		skb->tstamp = ktime_get_real();
+	}
+
+	serr = SKB_EXT_ERR(skb);
+	memset(serr, 0, sizeof(*serr));
+	serr->ee.ee_errno = ENOMSG;
+	serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
+	err = sock_queue_err_skb(sk, skb);
+	if (err)
+		kfree_skb(skb);
+}
+EXPORT_SYMBOL_GPL(skb_tstamp_tx);
+
+
 /**
  * skb_partial_csum_set - set up and verify partial csum values for packet
  * @skb: the skb to set