net: Implement read-only protection and COW'ing of metrics.

Routing metrics are now copy-on-write.

Initially a route entry points it's metrics at a read-only location.
If a routing table entry exists, it will point there. Else it will
point at the all zero metric place-holder called 'dst_default_metrics'.

The writeability state of the metrics is stored in the low bits of the
metrics pointer, we have two bits left to spare if we want to store
more states.

For the initial implementation, COW is implemented simply via kmalloc.
However future enhancements will change this to place the writable
metrics somewhere else, in order to increase sharing. Very likely
this "somewhere else" will be the inetpeer cache.

Note also that this means that metrics updates may transiently fail
if we cannot COW the metrics successfully.

But even by itself, this patch should decrease memory usage and
increase cache locality especially for routing workloads. In those
cases the read-only metric copies stay in place and never get written
to.

TCP workloads where metrics get updated, and those rare cases where
PMTU triggers occur, will take a very slight performance hit. But
that hit will be alleviated when the long-term writable metrics
move to a more sharable location.

Since the metrics storage went from a u32 array of RTAX_MAX entries to
what is essentially a pointer, some retooling of the dst_entry layout
was necessary.

Most importantly, we need to preserve the alignment of the reference
count so that it doesn't share cache lines with the read-mostly state,
as per Eric Dumazet's alignment assertion checks.

The only non-trivial bit here is the move of the 'flags' member into
the writeable cacheline. This is OK since we are always accessing the
flags around the same moment when we made a modification to the
reference count.

Signed-off-by: David S. Miller <[e-mail]>

include/net/dst.h

···
 	struct rcu_head		rcu_head;
 	struct dst_entry	*child;
 	struct net_device       *dev;
-	short			error;
-	short			obsolete;
-	int			flags;
-#define DST_HOST		0x0001
-#define DST_NOXFRM		0x0002
-#define DST_NOPOLICY		0x0004
-#define DST_NOHASH		0x0008
-#define DST_NOCACHE		0x0010
+	struct  dst_ops	        *ops;
+	unsigned long		_metrics;
 	unsigned long		expires;
-
-	unsigned short		header_len;	/* more space at head required */
-	unsigned short		trailer_len;	/* space to reserve at tail */
-
-	unsigned int		rate_tokens;
-	unsigned long		rate_last;	/* rate limiting for ICMP */
-
 	struct dst_entry	*path;
-
 	struct neighbour	*neighbour;
 	struct hh_cache		*hh;
 #ifdef CONFIG_XFRM
···
 	int			(*input)(struct sk_buff*);
 	int			(*output)(struct sk_buff*);
 
-	struct  dst_ops	        *ops;
-
-	u32			_metrics[RTAX_MAX];
-
+	short			error;
+	short			obsolete;
+	unsigned short		header_len;	/* more space at head required */
+	unsigned short		trailer_len;	/* space to reserve at tail */
 #ifdef CONFIG_IP_ROUTE_CLASSID
 	__u32			tclassid;
 #else
 	__u32			__pad2;
 #endif
-
 
 	/*
 	 * Align __refcnt to a 64 bytes alignment
···
 	atomic_t		__refcnt;	/* client references	*/
 	int			__use;
 	unsigned long		lastuse;
+	unsigned long		rate_last;	/* rate limiting for ICMP */
+	unsigned int		rate_tokens;
+	int			flags;
+#define DST_HOST		0x0001
+#define DST_NOXFRM		0x0002
+#define DST_NOPOLICY		0x0004
+#define DST_NOHASH		0x0008
+#define DST_NOCACHE		0x0010
 	union {
 		struct dst_entry	*next;
 		struct rtable __rcu	*rt_next;
···
 
 #ifdef __KERNEL__
 
+extern u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old);
+
+#define DST_METRICS_READ_ONLY	0x1UL
+#define __DST_METRICS_PTR(Y)	\
+	((u32 *)((Y) & ~DST_METRICS_READ_ONLY))
+#define DST_METRICS_PTR(X)	__DST_METRICS_PTR((X)->_metrics)
+
+static inline bool dst_metrics_read_only(const struct dst_entry *dst)
+{
+	return dst->_metrics & DST_METRICS_READ_ONLY;
+}
+
+extern void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old);
+
+static inline void dst_destroy_metrics_generic(struct dst_entry *dst)
+{
+	unsigned long val = dst->_metrics;
+	if (!(val & DST_METRICS_READ_ONLY))
+		__dst_destroy_metrics_generic(dst, val);
+}
+
+static inline u32 *dst_metrics_write_ptr(struct dst_entry *dst)
+{
+	unsigned long p = dst->_metrics;
+
+	if (p & DST_METRICS_READ_ONLY)
+		return dst->ops->cow_metrics(dst, p);
+	return __DST_METRICS_PTR(p);
+}
+
+/* This may only be invoked before the entry has reached global
+ * visibility.
+ */
+static inline void dst_init_metrics(struct dst_entry *dst,
+				    const u32 *src_metrics,
+				    bool read_only)
+{
+	dst->_metrics = ((unsigned long) src_metrics) |
+		(read_only ? DST_METRICS_READ_ONLY : 0);
+}
+
+static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src)
+{
+	u32 *dst_metrics = dst_metrics_write_ptr(dest);
+
+	if (dst_metrics) {
+		u32 *src_metrics = DST_METRICS_PTR(src);
+
+		memcpy(dst_metrics, src_metrics, RTAX_MAX * sizeof(u32));
+	}
+}
+
+static inline u32 *dst_metrics_ptr(struct dst_entry *dst)
+{
+	return DST_METRICS_PTR(dst);
+}
+
 static inline u32
 dst_metric_raw(const struct dst_entry *dst, const int metric)
 {
-	return dst->_metrics[metric-1];
+	u32 *p = DST_METRICS_PTR(dst);
+
+	return p[metric-1];
 }
 
 static inline u32
···
 
 static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val)
 {
-	dst->_metrics[metric-1] = val;
-}
+	u32 *p = dst_metrics_write_ptr(dst);
 
-static inline void dst_import_metrics(struct dst_entry *dst, const u32 *src_metrics)
-{
-	memcpy(dst->_metrics, src_metrics, RTAX_MAX * sizeof(u32));
-}
-
-static inline void dst_copy_metrics(struct dst_entry *dest, const struct dst_entry *src)
-{
-	dst_import_metrics(dest, src->_metrics);
-}
-
-static inline u32 *dst_metrics_ptr(struct dst_entry *dst)
-{
-	return dst->_metrics;
+	if (p)
+		p[metric-1] = val;
 }
 
 static inline u32

include/net/dst_ops.h

···
 	struct dst_entry *	(*check)(struct dst_entry *, __u32 cookie);
 	unsigned int		(*default_advmss)(const struct dst_entry *);
 	unsigned int		(*default_mtu)(const struct dst_entry *);
+	u32 *			(*cow_metrics)(struct dst_entry *, unsigned long);
 	void			(*destroy)(struct dst_entry *);
 	void			(*ifdown)(struct dst_entry *,
 					  struct net_device *dev, int how);

include/net/route.h

···
 
 struct fib_nh;
 struct inet_peer;
+struct fib_info;
 struct rtable {
 	struct dst_entry	dst;
 
···
 	/* Miscellaneous cached information */
 	__be32			rt_spec_dst; /* RFC1122 specific destination */
 	struct inet_peer	*peer; /* long-living peer info */
+	struct fib_info		*fi; /* for client ref to shared metrics */
 };
 
 static inline bool rt_is_input_route(struct rtable *rt)

net/core/dst.c

···
 }
 EXPORT_SYMBOL(dst_discard);
 
+static const u32 dst_default_metrics[RTAX_MAX];
+
 void *dst_alloc(struct dst_ops *ops)
 {
 	struct dst_entry *dst;
···
 	dst->lastuse = jiffies;
 	dst->path = dst;
 	dst->input = dst->output = dst_discard;
+	dst_init_metrics(dst, dst_default_metrics, true);
 #if RT_CACHE_DEBUG >= 2
 	atomic_inc(&dst_total);
 #endif
···
 	}
 }
 EXPORT_SYMBOL(dst_release);
+
+u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old)
+{
+	u32 *p = kmalloc(sizeof(u32) * RTAX_MAX, GFP_ATOMIC);
+
+	if (p) {
+		u32 *old_p = __DST_METRICS_PTR(old);
+		unsigned long prev, new;
+
+		memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
+
+		new = (unsigned long) p;
+		prev = cmpxchg(&dst->_metrics, old, new);
+
+		if (prev != old) {
+			kfree(p);
+			p = __DST_METRICS_PTR(prev);
+			if (prev & DST_METRICS_READ_ONLY)
+				p = NULL;
+		}
+	}
+	return p;
+}
+EXPORT_SYMBOL(dst_cow_metrics_generic);
+
+/* Caller asserts that dst_metrics_read_only(dst) is false.  */
+void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old)
+{
+	unsigned long prev, new;
+
+	new = (unsigned long) dst_default_metrics;
+	prev = cmpxchg(&dst->_metrics, old, new);
+	if (prev == old)
+		kfree(__DST_METRICS_PTR(old));
+}
+EXPORT_SYMBOL(__dst_destroy_metrics_generic);
 
 /**
  * skb_dst_set_noref - sets skb dst, without a reference

net/decnet/dn_route.c

···
 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst);
 static unsigned int dn_dst_default_mtu(const struct dst_entry *dst);
+static void dn_dst_destroy(struct dst_entry *);
 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
 static void dn_dst_link_failure(struct sk_buff *);
 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
···
 	.check =		dn_dst_check,
 	.default_advmss =	dn_dst_default_advmss,
 	.default_mtu =		dn_dst_default_mtu,
+	.cow_metrics =		dst_cow_metrics_generic,
+	.destroy =		dn_dst_destroy,
 	.negative_advice =	dn_dst_negative_advice,
 	.link_failure =		dn_dst_link_failure,
 	.update_pmtu =		dn_dst_update_pmtu,
 };
+
+static void dn_dst_destroy(struct dst_entry *dst)
+{
+	dst_destroy_metrics_generic(dst);
+}
 
 static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
 {
···
 {
 	struct dn_fib_info *fi = res->fi;
 	struct net_device *dev = rt->dst.dev;
+	unsigned int mss_metric;
 	struct neighbour *n;
-	unsigned int metric;
 
 	if (fi) {
 		if (DN_FIB_RES_GW(*res) &&
 		    DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
 			rt->rt_gateway = DN_FIB_RES_GW(*res);
-		dst_import_metrics(&rt->dst, fi->fib_metrics);
+		dst_init_metrics(&rt->dst, fi->fib_metrics, true);
 	}
 	rt->rt_type = res->type;
 
···
 
 	if (dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
 		dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
-	metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
-	if (metric) {
+	mss_metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
+	if (mss_metric) {
 		unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
-		if (metric > mss)
+		if (mss_metric > mss)
 			dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
 	}
 	return 0;

net/ipv4/route.c

···
 {
 }
 
+static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
+{
+	u32 *p = kmalloc(sizeof(u32) * RTAX_MAX, GFP_ATOMIC);
+
+	if (p) {
+		u32 *old_p = __DST_METRICS_PTR(old);
+		unsigned long prev, new;
+
+		memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
+
+		new = (unsigned long) p;
+		prev = cmpxchg(&dst->_metrics, old, new);
+
+		if (prev != old) {
+			kfree(p);
+			p = __DST_METRICS_PTR(prev);
+			if (prev & DST_METRICS_READ_ONLY)
+				p = NULL;
+		} else {
+			struct rtable *rt = (struct rtable *) dst;
+
+			if (rt->fi) {
+				fib_info_put(rt->fi);
+				rt->fi = NULL;
+			}
+		}
+	}
+	return p;
+}
+
 static struct dst_ops ipv4_dst_ops = {
 	.family =		AF_INET,
 	.protocol =		cpu_to_be16(ETH_P_IP),
···
 	.check =		ipv4_dst_check,
 	.default_advmss =	ipv4_default_advmss,
 	.default_mtu =		ipv4_default_mtu,
+	.cow_metrics =		ipv4_cow_metrics,
 	.destroy =		ipv4_dst_destroy,
 	.ifdown =		ipv4_dst_ifdown,
 	.negative_advice =	ipv4_negative_advice,
···
 
 				if (rt->peer)
 					atomic_inc(&rt->peer->refcnt);
+				if (rt->fi)
+					atomic_inc(&rt->fi->fib_clntref);
 
 				if (arp_bind_neighbour(&rt->dst) ||
 				    !(rt->dst.neighbour->nud_state &
···
 	struct rtable *rt = (struct rtable *) dst;
 	struct inet_peer *peer = rt->peer;
 
+	dst_destroy_metrics_generic(dst);
+	if (rt->fi) {
+		fib_info_put(rt->fi);
+		rt->fi = NULL;
+	}
 	if (peer) {
 		rt->peer = NULL;
 		inet_putpeer(peer);
···
 		if (FIB_RES_GW(*res) &&
 		    FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
 			rt->rt_gateway = FIB_RES_GW(*res);
-		dst_import_metrics(dst, fi->fib_metrics);
+		rt->fi = fi;
+		atomic_inc(&fi->fib_clntref);
+		dst_init_metrics(dst, fi->fib_metrics, true);
 #ifdef CONFIG_IP_ROUTE_CLASSID
 		dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
 #endif
···
 		rt->peer = ort->peer;
 		if (rt->peer)
 			atomic_inc(&rt->peer->refcnt);
+		rt->fi = ort->fi;
+		if (rt->fi)
+			atomic_inc(&rt->fi->fib_clntref);
 
 		dst_free(new);
 	}

net/ipv4/xfrm4_policy.c

···
 {
 	struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
 
+	dst_destroy_metrics_generic(dst);
+
 	if (likely(xdst->u.rt.peer))
 		inet_putpeer(xdst->u.rt.peer);
+
 	xfrm_dst_destroy(xdst);
 }
 
···
 	.protocol =		cpu_to_be16(ETH_P_IP),
 	.gc =			xfrm4_garbage_collect,
 	.update_pmtu =		xfrm4_update_pmtu,
+	.cow_metrics =		dst_cow_metrics_generic,
 	.destroy =		xfrm4_dst_destroy,
 	.ifdown =		xfrm4_dst_ifdown,
 	.local_out =		__ip_local_out,

net/ipv6/route.c

···
 	.check			=	ip6_dst_check,
 	.default_advmss		=	ip6_default_advmss,
 	.default_mtu		=	ip6_default_mtu,
+	.cow_metrics		=	dst_cow_metrics_generic,
 	.destroy		=	ip6_dst_destroy,
 	.ifdown			=	ip6_dst_ifdown,
 	.negative_advice	=	ip6_negative_advice,
···
 	.destroy		=	ip6_dst_destroy,
 	.check			=	ip6_dst_check,
 	.update_pmtu		=	ip6_rt_blackhole_update_pmtu,
+};
+
+static const u32 ip6_template_metrics[RTAX_MAX] = {
+	[RTAX_HOPLIMIT - 1] = 255,
 };
 
 static struct rt6_info ip6_null_entry_template = {
···
 		rt->rt6i_idev = NULL;
 		in6_dev_put(idev);
 	}
+	dst_destroy_metrics_generic(dst);
 	if (peer) {
 		BUG_ON(!(rt->rt6i_flags & RTF_CACHE));
 		rt->rt6i_peer = NULL;
···
 	net->ipv6.ip6_null_entry->dst.path =
 		(struct dst_entry *)net->ipv6.ip6_null_entry;
 	net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
-	dst_metric_set(&net->ipv6.ip6_null_entry->dst, RTAX_HOPLIMIT, 255);
+	dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
+			 ip6_template_metrics, true);
 
 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 	net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
···
 	net->ipv6.ip6_prohibit_entry->dst.path =
 		(struct dst_entry *)net->ipv6.ip6_prohibit_entry;
 	net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
-	dst_metric_set(&net->ipv6.ip6_prohibit_entry->dst, RTAX_HOPLIMIT, 255);
+	dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
+			 ip6_template_metrics, true);
 
 	net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
 					       sizeof(*net->ipv6.ip6_blk_hole_entry),
···
 	net->ipv6.ip6_blk_hole_entry->dst.path =
 		(struct dst_entry *)net->ipv6.ip6_blk_hole_entry;
 	net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
-	dst_metric_set(&net->ipv6.ip6_blk_hole_entry->dst, RTAX_HOPLIMIT, 255);
+	dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
+			 ip6_template_metrics, true);
 #endif
 
 	net->ipv6.sysctl.flush_delay = 0;

net/ipv6/xfrm6_policy.c

···
 
 	if (likely(xdst->u.rt6.rt6i_idev))
 		in6_dev_put(xdst->u.rt6.rt6i_idev);
+	dst_destroy_metrics_generic(dst);
 	if (likely(xdst->u.rt6.rt6i_peer))
 		inet_putpeer(xdst->u.rt6.rt6i_peer);
 	xfrm_dst_destroy(xdst);
···
 	.protocol =		cpu_to_be16(ETH_P_IPV6),
 	.gc =			xfrm6_garbage_collect,
 	.update_pmtu =		xfrm6_update_pmtu,
+	.cow_metrics =		dst_cow_metrics_generic,
 	.destroy =		xfrm6_dst_destroy,
 	.ifdown =		xfrm6_dst_ifdown,
 	.local_out =		__ip6_local_out,