datapath: handle recirculation loop detection
[sliver-openvswitch.git] / datapath / actions.c
index 33633df..f71f1f6 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2007-2013 Nicira, Inc.
+ * Copyright (c) 2007-2014 Nicira, Inc.
  *
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of version 2 of the GNU General Public
@@ -34,7 +34,6 @@
 #include <net/dsfield.h>
 #include <net/sctp/checksum.h>
 
-#include "checksum.h"
 #include "datapath.h"
 #include "vlan.h"
 #include "vport.h"
@@ -60,7 +59,7 @@ static int __pop_vlan_tci(struct sk_buff *skb, __be16 *current_tci)
        if (unlikely(err))
                return err;
 
-       if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
+       if (skb->ip_summed == CHECKSUM_COMPLETE)
                skb->csum = csum_sub(skb->csum, csum_partial(skb->data
                                        + (2 * ETH_ALEN), VLAN_HLEN, 0));
 
@@ -117,7 +116,7 @@ static int push_vlan(struct sk_buff *skb, const struct ovs_action_push_vlan *vla
                if (!__vlan_put_tag(skb, skb->vlan_proto, current_tag))
                        return -ENOMEM;
 
-               if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
+               if (skb->ip_summed == CHECKSUM_COMPLETE)
                        skb->csum = csum_add(skb->csum, csum_partial(skb->data
                                        + (2 * ETH_ALEN), VLAN_HLEN, 0));
 
@@ -134,16 +133,12 @@ static int set_eth_addr(struct sk_buff *skb,
        if (unlikely(err))
                return err;
 
-       if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
-               skb->csum = csum_sub(skb->csum, csum_partial(eth_hdr(skb),
-                                                            ETH_ALEN * 2, 0));
+       skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 
-       memcpy(eth_hdr(skb)->h_source, eth_key->eth_src, ETH_ALEN);
-       memcpy(eth_hdr(skb)->h_dest, eth_key->eth_dst, ETH_ALEN);
+       ether_addr_copy(eth_hdr(skb)->h_source, eth_key->eth_src);
+       ether_addr_copy(eth_hdr(skb)->h_dest, eth_key->eth_dst);
 
-       if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
-               skb->csum = csum_add(skb->csum, csum_partial(eth_hdr(skb),
-                                                            ETH_ALEN * 2, 0));
+       ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 
        return 0;
 }
@@ -161,8 +156,7 @@ static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
                if (likely(transport_len >= sizeof(struct udphdr))) {
                        struct udphdr *uh = udp_hdr(skb);
 
-                       if (uh->check ||
-                           get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
+                       if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
                                inet_proto_csum_replace4(&uh->check, skb,
                                                         *addr, new_addr, 1);
                                if (!uh->check)
@@ -189,8 +183,7 @@ static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
                if (likely(transport_len >= sizeof(struct udphdr))) {
                        struct udphdr *uh = udp_hdr(skb);
 
-                       if (uh->check ||
-                           get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
+                       if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
                                inet_proto_csum_replace16(&uh->check, skb,
                                                          addr, new_addr, 1);
                                if (!uh->check)
@@ -311,7 +304,7 @@ static void set_udp_port(struct sk_buff *skb, __be16 *port, __be16 new_port)
 {
        struct udphdr *uh = udp_hdr(skb);
 
-       if (uh->check && get_ip_summed(skb) != OVS_CSUM_PARTIAL) {
+       if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
                set_tp_port(skb, port, new_port, &uh->check);
 
                if (!uh->check)
@@ -467,6 +460,21 @@ static int sample(struct datapath *dp, struct sk_buff *skb,
                                  nla_len(acts_list), true);
 }
 
+static void execute_hash(struct sk_buff *skb, const struct nlattr *attr)
+{
+       struct sw_flow_key *key = OVS_CB(skb)->pkt_key;
+       struct ovs_action_hash *hash_act = nla_data(attr);
+       u32 hash = 0;
+
+       /* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
+       hash = skb_get_rxhash(skb);
+       hash = jhash_1word(hash, hash_act->hash_basis);
+       if (!hash)
+               hash = 0x1;
+
+       key->ovs_flow_hash = hash;
+}
+
 static int execute_set_action(struct sk_buff *skb,
                                 const struct nlattr *nested_attr)
 {
@@ -513,6 +521,26 @@ static int execute_set_action(struct sk_buff *skb,
        return err;
 }
 
+static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
+                                const struct nlattr *a)
+{
+       struct sw_flow_key recirc_key;
+       const struct vport *p = OVS_CB(skb)->input_vport;
+       uint32_t hash = OVS_CB(skb)->pkt_key->ovs_flow_hash;
+       int err;
+
+       err = ovs_flow_extract(skb, p->port_no, &recirc_key);
+       if (err)
+               return err;
+
+       recirc_key.ovs_flow_hash = hash;
+       recirc_key.recirc_id = nla_get_u32(a);
+
+       ovs_dp_process_packet_with_key(skb, &recirc_key, true);
+
+       return 0;
+}
+
 /* Execute a list of actions against 'skb'. */
 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
                        const struct nlattr *attr, int len, bool keep_skb)
@@ -543,6 +571,10 @@ static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
                        output_userspace(dp, skb, a);
                        break;
 
+               case OVS_ACTION_ATTR_HASH:
+                       execute_hash(skb, a);
+                       break;
+
                case OVS_ACTION_ATTR_PUSH_VLAN:
                        err = push_vlan(skb, nla_data(a));
                        if (unlikely(err)) /* skb already freed. */
@@ -553,12 +585,31 @@ static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
                        err = pop_vlan(skb);
                        break;
 
+               case OVS_ACTION_ATTR_RECIRC: {
+                       struct sk_buff *recirc_skb;
+                       const bool last_action = (a->nla_len == rem);
+
+                       if (!last_action || keep_skb)
+                               recirc_skb = skb_clone(skb, GFP_ATOMIC);
+                       else
+                               recirc_skb = skb;
+
+                       err = execute_recirc(dp, recirc_skb, a);
+
+                       if (last_action || err)
+                               return err;
+
+                       break;
+               }
+
                case OVS_ACTION_ATTR_SET:
                        err = execute_set_action(skb, nla_data(a));
                        break;
 
                case OVS_ACTION_ATTR_SAMPLE:
                        err = sample(dp, skb, a);
+                       if (unlikely(err)) /* skb already freed. */
+                               return err;
                        break;
                }
 
@@ -580,11 +631,18 @@ static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 }
 
 /* We limit the number of times that we pass into execute_actions()
- * to avoid blowing out the stack in the event that we have a loop. */
-#define MAX_LOOPS 4
+ * to avoid blowing out the stack in the event that we have a loop.
+ *
+ * Each loop adds some (estimated) cost to the kernel stack.
+ * The loop terminates when the max cost is exceeded.
+ * */
+#define RECIRC_STACK_COST 1
+#define DEFAULT_STACK_COST 4
+/* Allow up to 4 regular services, and up to 3 recirculations */
+#define MAX_STACK_COST (DEFAULT_STACK_COST * 4 + RECIRC_STACK_COST * 3)
 
 struct loop_counter {
-       u8 count;               /* Count. */
+       u8 stack_cost;          /* loop stack cost. */
        bool looping;           /* Loop detected? */
 };
 
@@ -593,22 +651,24 @@ static DEFINE_PER_CPU(struct loop_counter, loop_counters);
 static int loop_suppress(struct datapath *dp, struct sw_flow_actions *actions)
 {
        if (net_ratelimit())
-               pr_warn("%s: flow looped %d times, dropping\n",
-                               ovs_dp_name(dp), MAX_LOOPS);
+               pr_warn("%s: flow loop detected, dropping\n",
+                               ovs_dp_name(dp));
        actions->actions_len = 0;
        return -ELOOP;
 }
 
 /* Execute a list of actions against 'skb'. */
-int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb)
+int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb, bool recirc)
 {
        struct sw_flow_actions *acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
+       const u8 stack_cost = recirc ? RECIRC_STACK_COST : DEFAULT_STACK_COST;
        struct loop_counter *loop;
        int error;
 
        /* Check whether we've looped too much. */
        loop = &__get_cpu_var(loop_counters);
-       if (unlikely(++loop->count > MAX_LOOPS))
+       loop->stack_cost += stack_cost;
+       if (unlikely(loop->stack_cost > MAX_STACK_COST))
                loop->looping = true;
        if (unlikely(loop->looping)) {
                error = loop_suppress(dp, acts);
@@ -625,8 +685,9 @@ int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb)
                error = loop_suppress(dp, acts);
 
 out_loop:
-       /* Decrement loop counter. */
-       if (!--loop->count)
+       /* Decrement loop stack cost. */
+       loop->stack_cost -= stack_cost;
+       if (!loop->stack_cost)
                loop->looping = false;
 
        return error;