X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=datapath%2Fdatapath.c;h=a34049f931383fd52478b182531fa815476261a8;hb=7d0ab001dbc7bd4285aaf1dbcb881312ec32608c;hp=eb260e33ab2c8d6b02b57d10adf30551a91c2a4e;hpb=fceb2a5bb2063023777fc75c68a2670b5169fa13;p=openvswitch

diff --git a/datapath/datapath.c b/datapath/datapath.c
index eb260e33..a34049f9 100644
--- a/datapath/datapath.c
+++ b/datapath/datapath.c
@@ -69,8 +69,22 @@ EXPORT_SYMBOL(dp_ioctl_hook);
 static struct datapath *dps[ODP_MAX];
 static DEFINE_MUTEX(dp_mutex);
 
-/* Number of milliseconds between runs of the maintenance thread. */
-#define MAINT_SLEEP_MSECS 1000
+/* We limit the number of times that we pass into dp_process_received_packet()
+ * to avoid blowing out the stack in the event that we have a loop. */
+struct loop_counter {
+	int count;		/* Count. */
+	bool looping;		/* Loop detected? */
+};
+
+#define DP_MAX_LOOPS 5
+
+/* We use a separate counter for each CPU for both interrupt and non-interrupt
+ * context in order to keep the limit deterministic for a given packet. */
+struct percpu_loop_counters {
+	struct loop_counter counters[2];
+};
+
+static DEFINE_PER_CPU(struct percpu_loop_counters, dp_loop_counters);
 
 static int new_dp_port(struct datapath *, struct odp_port *, int port_no);
 
@@ -380,6 +394,7 @@ static int new_dp_port(struct datapath *dp, struct odp_port *odp_port, int port_
 
 	p->port_no = port_no;
 	p->dp = dp;
+	p->vport = vport;
 	atomic_set(&p->sflow_pool, 0);
 
 	err = vport_attach(vport, p);
@@ -513,6 +528,14 @@ out:
 	return err;
 }
 
+static void suppress_loop(struct datapath *dp, struct sw_flow_actions *actions)
+{
+	if (net_ratelimit())
+		printk(KERN_WARNING "%s: flow looped %d times, dropping\n",
+		       dp_name(dp), DP_MAX_LOOPS);
+	actions->n_actions = 0;
+}
+
 /* Must be called with rcu_read_lock. */
 void dp_process_received_packet(struct dp_port *p, struct sk_buff *skb)
 {
@@ -521,37 +544,71 @@ void dp_process_received_packet(struct dp_port *p, struct sk_buff *skb)
 	int stats_counter_off;
 	struct odp_flow_key key;
 	struct tbl_node *flow_node;
-
-	WARN_ON_ONCE(skb_shared(skb));
-	skb_warn_if_lro(skb);
+	struct sw_flow *flow;
+	struct sw_flow_actions *acts;
+	struct loop_counter *loop;
+	int error;
 
 	OVS_CB(skb)->dp_port = p;
 
-	if (flow_extract(skb, p ? p->port_no : ODPP_NONE, &key)) {
-		if (dp->drop_frags) {
-			kfree_skb(skb);
-			stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
-			goto out;
-		}
+	/* Extract flow from 'skb' into 'key'. */
+	error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key);
+	if (unlikely(error)) {
+		kfree_skb(skb);
+		return;
+	}
+
+	if (OVS_CB(skb)->is_frag && dp->drop_frags) {
+		kfree_skb(skb);
+		stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
+		goto out;
 	}
 
+	/* Look up flow. */
 	flow_node = tbl_lookup(rcu_dereference(dp->table), &key, flow_hash(&key), flow_cmp);
-	if (flow_node) {
-		struct sw_flow *flow = flow_cast(flow_node);
-		struct sw_flow_actions *acts = rcu_dereference(flow->sf_acts);
-		flow_used(flow, skb);
-		execute_actions(dp, skb, &key, acts->actions, acts->n_actions,
-				GFP_ATOMIC);
-		stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
-	} else {
-		stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
+	if (unlikely(!flow_node)) {
 		dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id);
+		stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
+		goto out;
+	}
+
+	flow = flow_cast(flow_node);
+	flow_used(flow, skb);
+
+	acts = rcu_dereference(flow->sf_acts);
+
+	/* Check whether we've looped too much. */
+	loop = &get_cpu_var(dp_loop_counters).counters[!!in_interrupt()];
+	if (unlikely(++loop->count > DP_MAX_LOOPS))
+		loop->looping = true;
+	if (unlikely(loop->looping)) {
+		suppress_loop(dp, acts);
+		goto out_loop;
 	}
 
+	/* Execute actions. */
+	execute_actions(dp, skb, &key, acts->actions, acts->n_actions, GFP_ATOMIC);
+	stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
+
+	/* Check whether sub-actions looped too much. */
+	if (unlikely(loop->looping))
+		suppress_loop(dp, acts);
+
+out_loop:
+	/* Decrement loop counter. */
+	if (!--loop->count)
+		loop->looping = false;
+	put_cpu_var(dp_loop_counters);
+
 out:
+	/* Update datapath statistics. */
 	local_bh_disable();
 	stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
+
+	write_seqcount_begin(&stats->seqlock);
 	(*(u64 *)((u8 *)stats + stats_counter_off))++;
+	write_seqcount_end(&stats->seqlock);
+
 	local_bh_enable();
 }
 
@@ -812,7 +869,11 @@ err_kfree_skb:
 err:
 	local_bh_disable();
 	stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
+
+	write_seqcount_begin(&stats->seqlock);
 	stats->n_lost++;
+	write_seqcount_end(&stats->seqlock);
+
 	local_bh_enable();
 
 	return err;
@@ -903,27 +964,43 @@ error:
 	return ERR_PTR(error);
 }
 
-static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats)
+static struct timespec get_time_offset(void)
+{
+	struct timespec now_mono, now_jiffies;
+
+	ktime_get_ts(&now_mono);
+	jiffies_to_timespec(jiffies, &now_jiffies);
+	return timespec_sub(now_mono, now_jiffies);
+}
+
+static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats,
+		      struct timespec time_offset)
 {
-	if (flow->used.tv_sec) {
-		stats->used_sec = flow->used.tv_sec;
-		stats->used_nsec = flow->used.tv_nsec;
+	if (flow->used) {
+		struct timespec flow_ts, used;
+
+		jiffies_to_timespec(flow->used, &flow_ts);
+		set_normalized_timespec(&used, flow_ts.tv_sec + time_offset.tv_sec,
+					flow_ts.tv_nsec + time_offset.tv_nsec);
+
+		stats->used_sec = used.tv_sec;
+		stats->used_nsec = used.tv_nsec;
 	} else {
 		stats->used_sec = 0;
 		stats->used_nsec = 0;
 	}
+
 	stats->n_packets = flow->packet_count;
 	stats->n_bytes = flow->byte_count;
-	stats->ip_tos = flow->ip_tos;
+	stats->reserved = 0;
 	stats->tcp_flags = flow->tcp_flags;
 	stats->error = 0;
 }
 
 static void clear_stats(struct sw_flow *flow)
 {
-	flow->used.tv_sec = flow->used.tv_nsec = 0;
+	flow->used = 0;
 	flow->tcp_flags = 0;
-	flow->ip_tos = 0;
 	flow->packet_count = 0;
 	flow->byte_count = 0;
 }
@@ -1021,7 +1098,7 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
 
 		/* Fetch stats, then clear them if necessary. */
 		spin_lock_bh(&flow->lock);
-		get_stats(flow, stats);
+		get_stats(flow, stats, get_time_offset());
 		if (uf->flags & ODPPF_ZERO_STATS)
 			clear_stats(flow);
 		spin_unlock_bh(&flow->lock);
@@ -1058,6 +1135,7 @@ static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
 }
 
 static int do_answer_query(struct sw_flow *flow, u32 query_flags,
+			   struct timespec time_offset,
 			   struct odp_flow_stats __user *ustats,
 			   union odp_action __user *actions,
 			   u32 __user *n_actionsp)
@@ -1067,7 +1145,7 @@ static int do_answer_query(struct sw_flow *flow, u32 query_flags,
 	u32 n_actions;
 
 	spin_lock_bh(&flow->lock);
-	get_stats(flow, &stats);
+	get_stats(flow, &stats, time_offset);
 	if (query_flags & ODPFF_ZERO_TCP_FLAGS)
 		flow->tcp_flags = 0;
 
@@ -1091,6 +1169,7 @@ static int do_answer_query(struct sw_flow *flow, u32 query_flags,
 }
 
 static int answer_query(struct sw_flow *flow, u32 query_flags,
+			struct timespec time_offset,
 			struct odp_flow __user *ufp)
 {
 	union odp_action *actions;
@@ -1098,7 +1177,7 @@ static int answer_query(struct sw_flow *flow, u32 query_flags,
 	if (get_user(actions, &ufp->actions))
 		return -EFAULT;
 
-	return do_answer_query(flow, query_flags,
+	return do_answer_query(flow, query_flags, time_offset,
 			       &ufp->stats, actions, &ufp->n_actions);
 }
 
@@ -1137,7 +1216,7 @@ static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
 	if (IS_ERR(flow))
 		return PTR_ERR(flow);
 
-	error = answer_query(flow, 0, ufp);
+	error = answer_query(flow, 0, get_time_offset(), ufp);
 	flow_deferred_free(flow);
 	return error;
 }
@@ -1145,8 +1224,11 @@ static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
 static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
 {
 	struct tbl *table = rcu_dereference(dp->table);
+	struct timespec time_offset;
 	u32 i;
 
+	time_offset = get_time_offset();
+
 	for (i = 0; i < flowvec->n_flows; i++) {
 		struct odp_flow __user *ufp = &flowvec->flows[i];
 		struct odp_flow uf;
@@ -1161,7 +1243,7 @@ static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec
 		if (!flow_node)
 			error = put_user(ENOENT, &ufp->stats.error);
 		else
-			error = answer_query(flow_cast(flow_node), uf.flags, ufp);
+			error = answer_query(flow_cast(flow_node), uf.flags, time_offset, ufp);
 		if (error)
 			return -EFAULT;
 	}
@@ -1172,6 +1254,7 @@ struct list_flows_cbdata {
 	struct odp_flow __user *uflows;
 	u32 n_flows;
 	u32 listed_flows;
+	struct timespec time_offset;
 };
 
 static int list_flow(struct tbl_node *node, void *cbdata_)
@@ -1183,7 +1266,7 @@ static int list_flow(struct tbl_node *node, void *cbdata_)
 
 	if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
 		return -EFAULT;
-	error = answer_query(flow, 0, ufp);
+	error = answer_query(flow, 0, cbdata->time_offset, ufp);
 	if (error)
 		return error;
 
@@ -1203,6 +1286,8 @@ static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
 	cbdata.uflows = flowvec->flows;
 	cbdata.n_flows = flowvec->n_flows;
 	cbdata.listed_flows = 0;
+	cbdata.time_offset = get_time_offset();
+
 	error = tbl_foreach(rcu_dereference(dp->table), list_flow, &cbdata);
 	return error ? error : cbdata.listed_flows;
 }
@@ -1280,7 +1365,9 @@ static int do_execute(struct datapath *dp, const struct odp_execute *execute)
 	else
 		skb->protocol = htons(ETH_P_802_2);
 
-	flow_extract(skb, execute->in_port, &key);
+	err = flow_extract(skb, execute->in_port, &key);
+	if (err)
+		goto error_free_skb;
 
 	rcu_read_lock();
 	err = execute_actions(dp, skb, &key, actions->actions,
@@ -1322,12 +1409,21 @@ static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
 	stats.max_groups = DP_MAX_GROUPS;
 	stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0;
 	for_each_possible_cpu(i) {
-		const struct dp_stats_percpu *s;
-		s = per_cpu_ptr(dp->stats_percpu, i);
-		stats.n_frags += s->n_frags;
-		stats.n_hit += s->n_hit;
-		stats.n_missed += s->n_missed;
-		stats.n_lost += s->n_lost;
+		const struct dp_stats_percpu *percpu_stats;
+		struct dp_stats_percpu local_stats;
+		unsigned seqcount;
+
+		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
+
+		do {
+			seqcount = read_seqcount_begin(&percpu_stats->seqlock);
+			local_stats = *percpu_stats;
+		} while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
+
+		stats.n_frags += local_stats.n_frags;
+		stats.n_hit += local_stats.n_hit;
+		stats.n_missed += local_stats.n_missed;
+		stats.n_lost += local_stats.n_lost;
 	}
 	stats.max_miss_queue = DP_MAX_QUEUE_LEN;
 	stats.max_action_queue = DP_MAX_QUEUE_LEN;
@@ -1811,6 +1907,7 @@ static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __use
 }
 
 static int compat_answer_query(struct sw_flow *flow, u32 query_flags,
+			       struct timespec time_offset,
 			       struct compat_odp_flow __user *ufp)
 {
 	compat_uptr_t actions;
@@ -1818,7 +1915,7 @@ static int compat_answer_query(struct sw_flow *flow, u32 query_flags,
 	if (get_user(actions, &ufp->actions))
 		return -EFAULT;
 
-	return do_answer_query(flow, query_flags, &ufp->stats,
+	return do_answer_query(flow, query_flags, time_offset, &ufp->stats,
 			       compat_ptr(actions), &ufp->n_actions);
 }
 
@@ -1835,7 +1932,7 @@ static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *u
 	if (IS_ERR(flow))
 		return PTR_ERR(flow);
 
-	error = compat_answer_query(flow, 0, ufp);
+	error = compat_answer_query(flow, 0, get_time_offset(), ufp);
 	flow_deferred_free(flow);
 	return error;
 }
@@ -1843,8 +1940,11 @@ static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *u
 static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
 {
 	struct tbl *table = rcu_dereference(dp->table);
+	struct timespec time_offset;
 	u32 i;
 
+	time_offset = get_time_offset();
+
 	for (i = 0; i < n_flows; i++) {
 		struct compat_odp_flow __user *ufp = &flows[i];
 		struct odp_flow uf;
@@ -1859,7 +1959,7 @@ static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows
 		if (!flow_node)
 			error = put_user(ENOENT, &ufp->stats.error);
 		else
-			error = compat_answer_query(flow_cast(flow_node), uf.flags, ufp);
+			error = compat_answer_query(flow_cast(flow_node), uf.flags, time_offset, ufp);
 		if (error)
 			return -EFAULT;
 	}
@@ -1870,6 +1970,7 @@ struct compat_list_flows_cbdata {
 	struct compat_odp_flow __user *uflows;
 	u32 n_flows;
 	u32 listed_flows;
+	struct timespec time_offset;
 };
 
 static int compat_list_flow(struct tbl_node *node, void *cbdata_)
@@ -1881,7 +1982,7 @@ static int compat_list_flow(struct tbl_node *node, void *cbdata_)
 
 	if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
 		return -EFAULT;
-	error = compat_answer_query(flow, 0, ufp);
+	error = compat_answer_query(flow, 0, cbdata->time_offset, ufp);
 	if (error)
 		return error;
 
@@ -1901,6 +2002,8 @@ static int compat_list_flows(struct datapath *dp, struct compat_odp_flow *flows,
 	cbdata.uflows = flows;
 	cbdata.n_flows = n_flows;
 	cbdata.listed_flows = 0;
+	cbdata.time_offset = get_time_offset();
+
 	error = tbl_foreach(rcu_dereference(dp->table), compat_list_flow, &cbdata);
 	return error ? error : cbdata.listed_flows;
 }