From: Ben Pfaff Date: Wed, 6 Oct 2010 18:37:03 +0000 (-0700) Subject: Merge "master" into "wdp". X-Git-Url: http://git.onelab.eu/?a=commitdiff_plain;h=d15a5ee594f46a855788652e687aa5d736d530fb;p=sliver-openvswitch.git Merge "master" into "wdp". --- d15a5ee594f46a855788652e687aa5d736d530fb diff --cc datapath/datapath.c index 131dcafcf,e9f30f8a8..abf751677 --- a/datapath/datapath.c +++ b/datapath/datapath.c @@@ -553,31 -548,39 +548,39 @@@ void dp_process_received_packet(struct OVS_CB(skb)->dp_port = p; - /* Extract flow from 'skb' into 'key'. */ - error = flow_extract(skb, p ? p->port_no : XFLOWP_NONE, &key); - if (unlikely(error)) { - kfree_skb(skb); - return; - } + if (!OVS_CB(skb)->flow) { - struct odp_flow_key key; ++ struct xflow_key key; + struct tbl_node *flow_node; + bool is_frag; - if (OVS_CB(skb)->is_frag && 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, &is_frag); ++ error = flow_extract(skb, p ? p->port_no : XFLOWP_NONE, &key, &is_frag); + if (unlikely(error)) { + kfree_skb(skb); + return; + } - /* Look up flow. */ - flow_node = tbl_lookup(rcu_dereference(dp->table), &key, flow_hash(&key), flow_cmp); - if (unlikely(!flow_node)) { - dp_output_control(dp, skb, _XFLOWL_MISS_NR, OVS_CB(skb)->tun_id); - stats_counter_off = offsetof(struct dp_stats_percpu, n_missed); - goto out; + if (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 (unlikely(!flow_node)) { - dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id); ++ dp_output_control(dp, skb, _XFLOWL_MISS_NR, OVS_CB(skb)->tun_id); + stats_counter_off = offsetof(struct dp_stats_percpu, n_missed); + goto out; + } + + OVS_CB(skb)->flow = flow_cast(flow_node); } - flow = flow_cast(flow_node); - flow_used(flow, skb); + flow_used(OVS_CB(skb)->flow, skb); - acts = rcu_dereference(flow->sf_acts); + acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts); /* Check whether we've looped too much. */ loop = &get_cpu_var(dp_loop_counters).counters[!!in_interrupt()]; diff --cc datapath/flow.c index 48bebca52,1aa6e291b..de2d3f3d6 --- a/datapath/flow.c +++ b/datapath/flow.c @@@ -108,10 -108,13 +108,13 @@@ struct sw_flow_actions *flow_actions_al { struct sw_flow_actions *sfa; - if (n_actions > (PAGE_SIZE - sizeof *sfa) / sizeof(union xflow_action)) + /* At least DP_MAX_PORTS actions are required to be able to flood a + * packet to every port. Factor of 2 allows for setting VLAN tags, + * etc. */ + if (n_actions > 2 * DP_MAX_PORTS) return ERR_PTR(-EINVAL); - sfa = kmalloc(sizeof *sfa + n_actions * sizeof(union odp_action), + sfa = kmalloc(sizeof *sfa + n_actions * sizeof(union xflow_action), GFP_KERNEL); if (!sfa) return ERR_PTR(-ENOMEM); @@@ -238,14 -267,16 +266,15 @@@ static __be16 parse_ethertype(struct sk * Sets OVS_CB(skb)->is_frag to %true if @skb is an IPv4 fragment, otherwise to * %false. */ - int flow_extract(struct sk_buff *skb, u16 in_port, struct xflow_key *key) -int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key, ++int flow_extract(struct sk_buff *skb, u16 in_port, struct xflow_key *key, + bool *is_frag) { struct ethhdr *eth; memset(key, 0, sizeof *key); key->tun_id = OVS_CB(skb)->tun_id; key->in_port = in_port; - OVS_CB(skb)->is_frag = false; - key->dl_vlan = htons(ODP_VLAN_NONE); + *is_frag = false; /* * We would really like to pull as many bytes as we could possibly diff --cc datapath/flow.h index 528e15aa3,25b720449..fab57c758 --- a/datapath/flow.h +++ b/datapath/flow.h @@@ -33,9 -33,12 +33,12 @@@ struct sw_flow struct rcu_head rcu; struct tbl_node tbl_node; - struct odp_flow_key key; + struct xflow_key key; struct sw_flow_actions *sf_acts; + atomic_t refcnt; + bool dead; + spinlock_t lock; /* Lock for values below. */ unsigned long used; /* Last used time (in jiffies). */ u64 packet_count; /* Number of packets matched. */ @@@ -58,20 -61,24 +61,24 @@@ struct arp_eth_heade unsigned char ar_tip[4]; /* target IP address */ } __attribute__((packed)); - extern struct kmem_cache *flow_cache; + int flow_init(void); + void flow_exit(void); - struct sw_flow_actions *flow_actions_alloc(size_t n_actions); + struct sw_flow *flow_alloc(void); void flow_deferred_free(struct sw_flow *); + void flow_free_tbl(struct tbl_node *); + + struct sw_flow_actions *flow_actions_alloc(size_t n_actions); void flow_deferred_free_acts(struct sw_flow_actions *); - int flow_extract(struct sk_buff *, u16 in_port, struct xflow_key *); + + void flow_hold(struct sw_flow *); + void flow_put(struct sw_flow *); + -int flow_extract(struct sk_buff *, u16 in_port, struct odp_flow_key *, bool *is_frag); ++int flow_extract(struct sk_buff *, u16 in_port, struct xflow_key *, bool *is_frag); void flow_used(struct sw_flow *, struct sk_buff *); -u32 flow_hash(const struct odp_flow_key *key); +u32 flow_hash(const struct xflow_key *key); int flow_cmp(const struct tbl_node *, void *target); - void flow_free_tbl(struct tbl_node *); - - int flow_init(void); - void flow_exit(void); static inline struct sw_flow *flow_cast(const struct tbl_node *node) { diff --cc datapath/tunnel.c index 6fa369be0,ad4522808..c752fe8af --- a/datapath/tunnel.c +++ b/datapath/tunnel.c @@@ -514,98 -688,283 +688,283 @@@ bool tnl_frag_needed(struct vport *vpor return true; } - static struct sk_buff *check_headroom(struct sk_buff *skb, int headroom) + static bool check_mtu(struct sk_buff *skb, + struct vport *vport, + const struct tnl_mutable_config *mutable, + const struct rtable *rt, __be16 *frag_offp) { - if (skb_headroom(skb) < headroom || skb_header_cloned(skb)) { - struct sk_buff *nskb = skb_realloc_headroom(skb, headroom + 16); - if (unlikely(!nskb)) { - kfree_skb(skb); - return ERR_PTR(-ENOMEM); + int mtu; + __be16 frag_off; + + frag_off = (mutable->port_config.flags & TNL_F_PMTUD) ? htons(IP_DF) : 0; + if (frag_off) + mtu = dst_mtu(&rt_dst(rt)) + - ETH_HLEN + - mutable->tunnel_hlen + - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0); + else + mtu = mutable->mtu; + + if (skb->protocol == htons(ETH_P_IP)) { + struct iphdr *old_iph = ip_hdr(skb); + + frag_off |= old_iph->frag_off & htons(IP_DF); + mtu = max(mtu, IP_MIN_MTU); + + if ((old_iph->frag_off & htons(IP_DF)) && + mtu < ntohs(old_iph->tot_len)) { + if (tnl_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id)) + goto drop; } + } + #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) + else if (skb->protocol == htons(ETH_P_IPV6)) { + unsigned int packet_length = skb->len - ETH_HLEN + - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0); - set_skb_csum_bits(skb, nskb); + mtu = max(mtu, IPV6_MIN_MTU); - if (skb->sk) - skb_set_owner_w(nskb, skb->sk); + /* IPv6 requires PMTUD if the packet is above the minimum MTU. */ + if (packet_length > IPV6_MIN_MTU) + frag_off = htons(IP_DF); - dev_kfree_skb(skb); - return nskb; + if (mtu < packet_length) { + if (tnl_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id)) + goto drop; + } } + #endif - return skb; + *frag_offp = frag_off; + return true; + + drop: + *frag_offp = 0; + return false; + } + + static void create_tunnel_header(const struct vport *vport, + const struct tnl_mutable_config *mutable, + const struct rtable *rt, void *header) + { + struct tnl_vport *tnl_vport = tnl_vport_priv(vport); + struct iphdr *iph = header; + + iph->version = 4; + iph->ihl = sizeof(struct iphdr) >> 2; + iph->frag_off = htons(IP_DF); + iph->protocol = tnl_vport->tnl_ops->ipproto; + iph->tos = mutable->port_config.tos; + iph->daddr = rt->rt_dst; + iph->saddr = rt->rt_src; + iph->ttl = mutable->port_config.ttl; + if (!iph->ttl) + iph->ttl = dst_metric(&rt_dst(rt), RTAX_HOPLIMIT); + + tnl_vport->tnl_ops->build_header(vport, mutable, iph + 1); } - static inline u8 ecn_encapsulate(u8 tos, struct sk_buff *skb) + static inline void *get_cached_header(const struct tnl_cache *cache) { - u8 inner; + return (void *)cache + ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN); + } - if (skb->protocol == htons(ETH_P_IP)) - inner = ((struct iphdr *)skb_network_header(skb))->tos; - #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) - else if (skb->protocol == htons(ETH_P_IPV6)) - inner = ipv6_get_dsfield((struct ipv6hdr *)skb_network_header(skb)); + static inline bool check_cache_valid(const struct tnl_cache *cache, + const struct tnl_mutable_config *mutable) + { + return cache && + #ifdef NEED_CACHE_TIMEOUT + time_before(jiffies, cache->expiration) && #endif - else - inner = 0; - - return INET_ECN_encapsulate(tos, inner); + #ifdef HAVE_RT_GENID + atomic_read(&init_net.ipv4.rt_genid) == cache->rt->rt_genid && + #endif + #ifdef HAVE_HH_SEQ + rt_dst(cache->rt).hh->hh_lock.sequence == cache->hh_seq && + #endif + mutable->seq == cache->mutable_seq && + (!is_internal_dev(rt_dst(cache->rt).dev) || + (cache->flow && !cache->flow->dead)); } - static inline void ecn_decapsulate(struct sk_buff *skb) + static int cache_cleaner_cb(struct tbl_node *tbl_node, void *aux) { - u8 tos = ip_hdr(skb)->tos; + struct tnl_vport *tnl_vport = tnl_vport_table_cast(tbl_node); + const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable); + const struct tnl_cache *cache = rcu_dereference(tnl_vport->cache); - if (INET_ECN_is_ce(tos)) { - __be16 protocol = skb->protocol; - unsigned int nw_header = skb_network_header(skb) - skb->data; + if (cache && !check_cache_valid(cache, mutable) && + spin_trylock_bh(&tnl_vport->cache_lock)) { + assign_cache_rcu(tnl_vport_to_vport(tnl_vport), NULL); + spin_unlock_bh(&tnl_vport->cache_lock); + } - if (skb->protocol == htons(ETH_P_8021Q)) { - if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN))) - return; + return 0; + } - protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto; - nw_header += VLAN_HLEN; - } + static void cache_cleaner(struct work_struct *work) + { + schedule_cache_cleaner(); - if (protocol == htons(ETH_P_IP)) { - if (unlikely(!pskb_may_pull(skb, nw_header - + sizeof(struct iphdr)))) - return; + rcu_read_lock(); + tbl_foreach(port_table, cache_cleaner_cb, NULL); + rcu_read_unlock(); + } - IP_ECN_set_ce((struct iphdr *)(nw_header + skb->data)); - } - #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) - else if (protocol == htons(ETH_P_IPV6)) { - if (unlikely(!pskb_may_pull(skb, nw_header - + sizeof(struct ipv6hdr)))) - return; + static inline void create_eth_hdr(struct tnl_cache *cache, + const struct rtable *rt) + { + void *cache_data = get_cached_header(cache); + int hh_len = rt_dst(rt).hh->hh_len; + int hh_off = HH_DATA_ALIGN(rt_dst(rt).hh->hh_len) - hh_len; - IP6_ECN_set_ce((struct ipv6hdr *)(nw_header - + skb->data)); - } + #ifdef HAVE_HH_SEQ + unsigned hh_seq; + + do { + hh_seq = read_seqbegin(&rt_dst(rt).hh->hh_lock); + memcpy(cache_data, (void *)rt_dst(rt).hh->hh_data + hh_off, hh_len); + } while (read_seqretry(&rt_dst(rt).hh->hh_lock, hh_seq)); + + cache->hh_seq = hh_seq; + #else + read_lock_bh(&rt_dst(rt).hh->hh_lock); + memcpy(cache_data, (void *)rt_dst(rt).hh->hh_data + hh_off, hh_len); + read_unlock_bh(&rt_dst(rt).hh->hh_lock); #endif - } } - static struct sk_buff *handle_gso(struct sk_buff *skb) + static struct tnl_cache *build_cache(struct vport *vport, + const struct tnl_mutable_config *mutable, + struct rtable *rt) { - if (skb_is_gso(skb)) { - struct sk_buff *nskb = skb_gso_segment(skb, 0); + struct tnl_vport *tnl_vport = tnl_vport_priv(vport); + struct tnl_cache *cache; + void *cache_data; + int cache_len; - dev_kfree_skb(skb); - return nskb; + if (!(mutable->port_config.flags & TNL_F_HDR_CACHE)) + return NULL; + + /* + * If there is no entry in the ARP cache or if this device does not + * support hard header caching just fall back to the IP stack. + */ + if (!rt_dst(rt).hh) + return NULL; + + /* + * If lock is contended fall back to directly building the header. + * We're not going to help performance by sitting here spinning. + */ + if (!spin_trylock_bh(&tnl_vport->cache_lock)) + return NULL; + + cache = tnl_vport->cache; + if (check_cache_valid(cache, mutable)) + goto unlock; + else + cache = NULL; + + cache_len = rt_dst(rt).hh->hh_len + mutable->tunnel_hlen; + + cache = kzalloc(ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN) + + cache_len, GFP_ATOMIC); + if (!cache) + goto unlock; + + cache->len = cache_len; + + create_eth_hdr(cache, rt); + cache_data = get_cached_header(cache) + rt_dst(rt).hh->hh_len; + + create_tunnel_header(vport, mutable, rt, cache_data); + + cache->mutable_seq = mutable->seq; + cache->rt = rt; + #ifdef NEED_CACHE_TIMEOUT + cache->expiration = jiffies + tnl_vport->cache_exp_interval; + #endif + + if (is_internal_dev(rt_dst(rt).dev)) { + int err; + struct vport *vport; + struct dp_port *dp_port; + struct sk_buff *skb; + bool is_frag; - struct odp_flow_key flow_key; ++ struct xflow_key flow_key; + struct tbl_node *flow_node; + + vport = internal_dev_get_vport(rt_dst(rt).dev); + if (!vport) + goto done; + + dp_port = vport_get_dp_port(vport); + if (!dp_port) + goto done; + + skb = alloc_skb(cache->len, GFP_ATOMIC); + if (!skb) + goto done; + + __skb_put(skb, cache->len); + memcpy(skb->data, get_cached_header(cache), cache->len); + + err = flow_extract(skb, dp_port->port_no, &flow_key, &is_frag); + + kfree_skb(skb); + if (err || is_frag) + goto done; + + flow_node = tbl_lookup(rcu_dereference(dp_port->dp->table), + &flow_key, flow_hash(&flow_key), + flow_cmp); + if (flow_node) { + struct sw_flow *flow = flow_cast(flow_node); + + cache->flow = flow; + flow_hold(flow); + } } - return skb; + done: + assign_cache_rcu(vport, cache); + + unlock: + spin_unlock_bh(&tnl_vport->cache_lock); + + return cache; } - static int handle_csum_offload(struct sk_buff *skb) + static struct rtable *find_route(struct vport *vport, + const struct tnl_mutable_config *mutable, + u8 tos, struct tnl_cache **cache) { - if (skb->ip_summed == CHECKSUM_PARTIAL) - return skb_checksum_help(skb); - else { - skb->ip_summed = CHECKSUM_NONE; - return 0; + struct tnl_vport *tnl_vport = tnl_vport_priv(vport); + struct tnl_cache *cur_cache = rcu_dereference(tnl_vport->cache); + + *cache = NULL; + tos = RT_TOS(tos); + + if (likely(tos == mutable->port_config.tos && + check_cache_valid(cur_cache, mutable))) { + *cache = cur_cache; + return cur_cache->rt; + } else { + struct rtable *rt; + struct flowi fl = { .nl_u = { .ip4_u = + { .daddr = mutable->port_config.daddr, + .saddr = mutable->port_config.saddr, + .tos = tos } }, + .proto = tnl_vport->tnl_ops->ipproto }; + + if (unlikely(ip_route_output_key(&init_net, &rt, &fl))) + return NULL; + + if (likely(tos == mutable->port_config.tos)) + *cache = build_cache(vport, mutable, rt); + + return rt; } } diff --cc lib/classifier.c index f844f5232,e0c57ebe4..f6f0b5a53 --- a/lib/classifier.c +++ b/lib/classifier.c @@@ -333,24 -326,24 +332,24 @@@ classifier_find_rule_exactly(const stru int table_idx; uint32_t hash; - if (!wildcards) { + if (!target->wildcards) { /* Ignores 'priority'. */ - return search_exact_table(cls, flow_hash(target, 0), target); + return search_exact_table(cls, flow_hash_headers(target, 0), target); } - assert(wildcards == (wildcards & OVSFW_ALL)); - table_idx = table_idx_from_wildcards(wildcards); + assert(target->wildcards == (target->wildcards & OVSFW_ALL)); + table_idx = table_idx_from_wildcards(target->wildcards); hash = hash_fields(target, table_idx); - HMAP_FOR_EACH_WITH_HASH (bucket, struct cls_bucket, hmap_node, hash, + HMAP_FOR_EACH_WITH_HASH (bucket, hmap_node, hash, &cls->tables[table_idx]) { if (equal_fields(&bucket->fixed, target, table_idx)) { struct cls_rule *pos; - LIST_FOR_EACH (pos, struct cls_rule, node.list, &bucket->rules) { + LIST_FOR_EACH (pos, node.list, &bucket->rules) { - if (pos->priority < priority) { + if (pos->flow.priority < target->priority) { return NULL; - } else if (pos->priority == priority && - pos->wc.wildcards == wildcards && - flow_equal(target, &pos->flow)) { + } else if (pos->flow.priority == target->priority && + pos->flow.wildcards == target->wildcards && + flow_equal_headers(target, &pos->flow)) { return pos; } } @@@ -378,13 -373,12 +377,12 @@@ classifier_rule_overlaps(const struct c for (tbl = &cls->tables[0]; tbl < &cls->tables[CLS_N_FIELDS]; tbl++) { struct cls_bucket *bucket; - HMAP_FOR_EACH (bucket, struct cls_bucket, hmap_node, tbl) { + HMAP_FOR_EACH (bucket, hmap_node, tbl) { struct cls_rule *rule; - LIST_FOR_EACH (rule, struct cls_rule, node.list, - &bucket->rules) { + LIST_FOR_EACH (rule, node.list, &bucket->rules) { - if (rule->priority == priority + if (rule->flow.priority == target->priority - && rules_match_2wild(rule, &target_rule, 0)) { + && rules_match_2wild(rule, &target_rule, 0)) { return true; } } @@@ -433,14 -419,10 +430,13 @@@ classifier_for_each_match(const struct * bucket itself will be destroyed. The bucket contains the * list head so that's a use-after-free error. */ prev_rule = NULL; - LIST_FOR_EACH (rule, struct cls_rule, node.list, - &bucket->rules) { + LIST_FOR_EACH (rule, node.list, &bucket->rules) { - if (rules_match_1wild(rule, target, 0)) { + if (rules_match_1wild(rule, &target, 0)) { if (prev_rule) { - callback(prev_rule, aux); + int retval = callback(prev_rule, aux); + if (retval) { + return retval; + } } prev_rule = rule; } @@@ -456,16 -435,13 +452,16 @@@ } if (include & CLS_INC_EXACT) { - if (target->wc.wildcards) { + if (target.flow.wildcards) { struct cls_rule *rule, *next_rule; - HMAP_FOR_EACH_SAFE (rule, next_rule, struct cls_rule, node.hmap, + HMAP_FOR_EACH_SAFE (rule, next_rule, node.hmap, &cls->exact_table) { - if (rules_match_1wild(rule, target, 0)) { - callback(rule, aux); + if (rules_match_1wild(rule, &target, 0)) { + int retval = callback(rule, aux); + if (retval) { + return retval; + } } } } else { @@@ -515,13 -481,9 +510,12 @@@ classifier_for_each(const struct classi * bucket itself will be destroyed. The bucket contains the * list head so that's a use-after-free error. */ prev_rule = NULL; - LIST_FOR_EACH (rule, struct cls_rule, node.list, - &bucket->rules) { + LIST_FOR_EACH (rule, node.list, &bucket->rules) { if (prev_rule) { - callback(prev_rule, aux); + int retval = callback(prev_rule, aux); + if (retval) { + return retval; + } } prev_rule = rule; } @@@ -538,16 -497,10 +532,15 @@@ if (include & CLS_INC_EXACT) { struct cls_rule *rule, *next_rule; - HMAP_FOR_EACH_SAFE (rule, next_rule, - struct cls_rule, node.hmap, &cls->exact_table) { + HMAP_FOR_EACH_SAFE (rule, next_rule, node.hmap, &cls->exact_table) { - callback(rule, aux); + int retval = callback(rule, aux); + if (retval) { + return retval; + } } } + + return 0; } static struct cls_bucket *create_bucket(struct hmap *, size_t hash, @@@ -681,9 -634,9 +674,9 @@@ static struct cls_rule bucket_insert(struct cls_bucket *bucket, struct cls_rule *rule) { struct cls_rule *pos; - LIST_FOR_EACH (pos, struct cls_rule, node.list, &bucket->rules) { + LIST_FOR_EACH (pos, node.list, &bucket->rules) { - if (pos->priority == rule->priority) { - if (pos->wc.wildcards == rule->wc.wildcards + if (pos->flow.priority == rule->flow.priority) { + if (pos->flow.wildcards == rule->flow.wildcards && rules_match_1wild(pos, rule, rule->table_idx)) { list_replace(&rule->node.list, &pos->node.list); @@@ -934,9 -886,8 +926,8 @@@ search_exact_table(const struct classif { struct cls_rule *rule; - HMAP_FOR_EACH_WITH_HASH (rule, struct cls_rule, node.hmap, - hash, &cls->exact_table) { + HMAP_FOR_EACH_WITH_HASH (rule, node.hmap, hash, &cls->exact_table) { - if (flow_equal(&rule->flow, target)) { + if (flow_equal_headers(&rule->flow, target)) { return rule; } } diff --cc lib/list.h index 59e3e86cb,0481477ae..997298270 --- a/lib/list.h +++ b/lib/list.h @@@ -57,22 -53,18 +57,22 @@@ struct list *list_back(struct list *) size_t list_size(const struct list *); bool list_is_empty(const struct list *); - #define LIST_FOR_EACH(ITER, STRUCT, MEMBER, LIST) \ - for (ITER = CONTAINER_OF((LIST)->next, STRUCT, MEMBER); \ - &(ITER)->MEMBER != (LIST); \ - ITER = CONTAINER_OF((ITER)->MEMBER.next, STRUCT, MEMBER)) - #define LIST_FOR_EACH_REVERSE(ITER, STRUCT, MEMBER, LIST) \ - for (ITER = CONTAINER_OF((LIST)->prev, STRUCT, MEMBER); \ - &(ITER)->MEMBER != (LIST); \ - ITER = CONTAINER_OF((ITER)->MEMBER.prev, STRUCT, MEMBER)) - #define LIST_FOR_EACH_SAFE(ITER, NEXT, STRUCT, MEMBER, LIST) \ - for (ITER = CONTAINER_OF((LIST)->next, STRUCT, MEMBER); \ - (NEXT = CONTAINER_OF((ITER)->MEMBER.next, STRUCT, MEMBER), \ - &(ITER)->MEMBER != (LIST)); \ + #define LIST_FOR_EACH(ITER, MEMBER, LIST) \ + for (ITER = OBJECT_CONTAINING((LIST)->next, ITER, MEMBER); \ + &(ITER)->MEMBER != (LIST); \ + ITER = OBJECT_CONTAINING((ITER)->MEMBER.next, ITER, MEMBER)) + #define LIST_FOR_EACH_REVERSE(ITER, MEMBER, LIST) \ + for (ITER = OBJECT_CONTAINING((LIST)->prev, ITER, MEMBER); \ + &(ITER)->MEMBER != (LIST); \ + ITER = OBJECT_CONTAINING((ITER)->MEMBER.prev, ITER, MEMBER)) + #define LIST_FOR_EACH_SAFE(ITER, NEXT, MEMBER, LIST) \ + for (ITER = OBJECT_CONTAINING((LIST)->next, ITER, MEMBER); \ + (NEXT = OBJECT_CONTAINING((ITER)->MEMBER.next, ITER, MEMBER), \ + &(ITER)->MEMBER != (LIST)); \ ITER = NEXT) + +#ifdef __cplusplus +} +#endif #endif /* list.h */ diff --cc lib/ofp-parse.c index a17d75c04,7a888801e..01dae0ca8 --- a/lib/ofp-parse.c +++ b/lib/ofp-parse.c @@@ -501,3 -514,65 +518,68 @@@ parse_ofp_str(char *string, struct ofp_ free(new); } } + + /* Parses 'string' as a OFPT_FLOW_MOD with subtype OFPFC_ADD and returns an - * ofpbuf that contains it. */ ++ * ofpbuf that contains it. Sets '*table_idx' to the index of the table to ++ * which the flow should be added, or to 0xff if none was specified. */ + struct ofpbuf * -parse_ofp_add_flow_str(char *string) ++parse_ofp_add_flow_str(char *string, uint8_t *table_idx) + { + struct ofpbuf *buffer; + struct ofp_flow_mod *ofm; + uint16_t priority, idle_timeout, hard_timeout; + uint64_t cookie; + struct ofp_match match; + + /* parse_ofp_str() will expand and reallocate the data in 'buffer', so we + * can't keep pointers to across the parse_ofp_str() call. */ + make_openflow(sizeof *ofm, OFPT_FLOW_MOD, &buffer); - parse_ofp_str(string, &match, buffer, - NULL, NULL, &priority, &idle_timeout, &hard_timeout, - &cookie); ++ parse_ofp_str(string, &match, buffer, table_idx, NULL, &priority, ++ &idle_timeout, &hard_timeout, &cookie); + ofm = buffer->data; + ofm->match = match; + ofm->command = htons(OFPFC_ADD); ++ if (*table_idx != 0xff) { ++ ofm->command |= htons(*table_idx << 8); ++ } + ofm->cookie = htonll(cookie); + ofm->idle_timeout = htons(idle_timeout); + ofm->hard_timeout = htons(hard_timeout); + ofm->buffer_id = htonl(UINT32_MAX); + ofm->priority = htons(priority); + update_openflow_length(buffer); + + return buffer; + } + + /* Parses an OFPT_FLOW_MOD with subtype OFPFC_ADD from 'stream' and returns an + * ofpbuf that contains it. Returns a null pointer if end-of-file is reached + * before reading a flow. */ + struct ofpbuf * -parse_ofp_add_flow_file(FILE *stream) ++parse_ofp_add_flow_file(FILE *stream, uint8_t *table_idx) + { + struct ofpbuf *b = NULL; + struct ds s = DS_EMPTY_INITIALIZER; + + while (!ds_get_line(&s, stream)) { + char *line = ds_cstr(&s); + char *comment; + + /* Delete comments. */ + comment = strchr(line, '#'); + if (comment) { + *comment = '\0'; + } + + /* Drop empty lines. */ + if (line[strspn(line, " \t\n")] == '\0') { + continue; + } + - b = parse_ofp_add_flow_str(line); ++ b = parse_ofp_add_flow_str(line, table_idx); + break; + } + ds_destroy(&s); + + return b; + } diff --cc lib/ofp-parse.h index aa0489c0c,ac8e6d2f1..7a44f0b9a --- a/lib/ofp-parse.h +++ b/lib/ofp-parse.h @@@ -29,5 -30,7 +30,7 @@@ void parse_ofp_str(char *string, struc uint16_t *out_port, uint16_t *priority, uint16_t *idle_timeout, uint16_t *hard_timeout, uint64_t *cookie); -struct ofpbuf *parse_ofp_add_flow_str(char *string); -struct ofpbuf *parse_ofp_add_flow_file(FILE *); ++struct ofpbuf *parse_ofp_add_flow_str(char *string, uint8_t *table_idx); ++struct ofpbuf *parse_ofp_add_flow_file(FILE *, uint8_t *table_idx); #endif /* ofp-parse.h */ diff --cc lib/ofp-print.c index e44cf36a1,1eaaa27d8..8de48a381 --- a/lib/ofp-print.c +++ b/lib/ofp-print.c @@@ -746,36 -762,51 +762,55 @@@ ofp_print_flow_mod(struct ds *string, c int verbosity) { const struct ofp_flow_mod *ofm = oh; + unsigned int command = ntohs(ofm->command); + ds_put_char(string, ' '); ofp_print_match(string, &ofm->match, verbosity); + if (ds_last(string) != ' ') { + ds_put_char(string, ' '); + } + - switch (ntohs(ofm->command)) { + switch (command & 0xff) { case OFPFC_ADD: - ds_put_cstr(string, " ADD: "); + ds_put_cstr(string, "ADD:"); break; case OFPFC_MODIFY: - ds_put_cstr(string, " MOD: "); + ds_put_cstr(string, "MOD:"); break; case OFPFC_MODIFY_STRICT: - ds_put_cstr(string, " MOD_STRICT: "); + ds_put_cstr(string, "MOD_STRICT:"); break; case OFPFC_DELETE: - ds_put_cstr(string, " DEL: "); + ds_put_cstr(string, "DEL:"); break; case OFPFC_DELETE_STRICT: - ds_put_cstr(string, " DEL_STRICT: "); + ds_put_cstr(string, "DEL_STRICT:"); break; default: - ds_put_format(string, " cmd:%u ", command); - ds_put_format(string, "cmd:%d", ntohs(ofm->command)); ++ ds_put_format(string, "cmd:%u", command & 0xff); + } + if (command & 0xff00) { - ds_put_format(string, "table_id:%u ", command >> 8); ++ ds_put_format(string, " table_id:%u", command >> 8); + } + if (ofm->cookie != htonll(0)) { + ds_put_format(string, " cookie:0x%"PRIx64, ntohll(ofm->cookie)); + } + if (ofm->idle_timeout != htons(OFP_FLOW_PERMANENT)) { + ds_put_format(string, " idle:%d", ntohs(ofm->idle_timeout)); } - ds_put_format(string, "cookie:0x%"PRIx64" idle:%d hard:%d pri:%d " - "buf:%#x flags:%"PRIx16" ", ntohll(ofm->cookie), - ntohs(ofm->idle_timeout), ntohs(ofm->hard_timeout), - ofm->match.wildcards ? ntohs(ofm->priority) : (uint16_t)-1, - ntohl(ofm->buffer_id), ntohs(ofm->flags)); + if (ofm->hard_timeout != htons(OFP_FLOW_PERMANENT)) { + ds_put_format(string, " hard:%d", ntohs(ofm->hard_timeout)); + } + if (ofm->priority != htons(32768)) { + ds_put_format(string, " pri:%"PRIu16, ntohs(ofm->priority)); + } + if (ofm->buffer_id != htonl(UINT32_MAX)) { + ds_put_format(string, " buf:%#"PRIx32, ntohl(ofm->buffer_id)); + } + if (ofm->flags != htons(0)) { + ds_put_format(string, " flags:%"PRIx16, ntohs(ofm->flags)); + } + ds_put_cstr(string, " "); ofp_print_actions(string, ofm->actions, len - offsetof(struct ofp_flow_mod, actions)); ds_put_char(string, '\n'); diff --cc lib/ofp-util.c index d527d430e,7a2e17cb2..6e0e41ff9 --- a/lib/ofp-util.c +++ b/lib/ofp-util.c @@@ -438,6 -438,6 +438,21 @@@ check_ofp_packet_out(const struct ofp_h return 0; } ++struct ofpbuf * ++make_nxt_flow_mod_table_id(bool enable) ++{ ++ struct nxt_flow_mod_table_id *flow_mod_table_id; ++ struct ofpbuf *buffer; ++ ++ flow_mod_table_id = make_openflow(sizeof *flow_mod_table_id, OFPT_VENDOR, ++ &buffer); ++ ++ flow_mod_table_id->vendor = htonl(NX_VENDOR_ID); ++ flow_mod_table_id->subtype = htonl(NXT_FLOW_MOD_TABLE_ID); ++ flow_mod_table_id->set = enable; ++ return buffer; ++} ++ const struct ofp_flow_stats * flow_stats_first(struct flow_stats_iterator *iter, const struct ofp_stats_reply *osr) diff --cc lib/ofp-util.h index 0e7281518,b4af179d6..64de2a079 --- a/lib/ofp-util.h +++ b/lib/ofp-util.h @@@ -62,6 -62,6 +62,8 @@@ int check_ofp_message_array(const struc int check_ofp_packet_out(const struct ofp_header *, struct ofpbuf *data, int *n_actions, int max_ports); ++struct ofpbuf *make_nxt_flow_mod_table_id(bool enable); ++ struct flow_stats_iterator { const uint8_t *pos, *end; }; diff --cc lib/vlog-modules.def index 4487662fa,7d614b412..c82275733 --- a/lib/vlog-modules.def +++ b/lib/vlog-modules.def @@@ -76,18 -76,16 +76,19 @@@ VLOG_MODULE(stream_unix VLOG_MODULE(svec) VLOG_MODULE(timeval) VLOG_MODULE(socket_util) + VLOG_MODULE(system_stats) VLOG_MODULE(unixctl) VLOG_MODULE(util) -VLOG_MODULE(vconn_stream) VLOG_MODULE(vconn) -VLOG_MODULE(vsctl) +VLOG_MODULE(vconn_stream) VLOG_MODULE(vlog) +VLOG_MODULE(vsctl) VLOG_MODULE(vswitchd) +VLOG_MODULE(wdp) +VLOG_MODULE(wdp_xflow) VLOG_MODULE(xenserver) +VLOG_MODULE(xfif) +VLOG_MODULE(xfif_linux) +VLOG_MODULE(xfif_netdev) #undef VLOG_MODULE diff --cc lib/xfif-linux.c index 7bfa840c1,635fe9411..82fa1d7b8 --- a/lib/xfif-linux.c +++ b/lib/xfif-linux.c @@@ -328,18 -328,17 +328,17 @@@ xfif_linux_port_list(const struct xfif } static int -dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep) +xfif_linux_port_poll(const struct xfif *xfif_, char **devnamep) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct xfif_linux *xfif = xfif_linux_cast(xfif_); - if (dpif->change_error) { - dpif->change_error = false; - shash_clear(&dpif->changed_ports); + if (xfif->change_error) { + xfif->change_error = false; + shash_clear(&xfif->changed_ports); return ENOBUFS; - } else if (!shash_is_empty(&dpif->changed_ports)) { - struct shash_node *node = shash_first(&dpif->changed_ports); - *devnamep = shash_steal(&dpif->changed_ports, node); + } else if (!shash_is_empty(&xfif->changed_ports)) { + struct shash_node *node = shash_first(&xfif->changed_ports); - *devnamep = xstrdup(node->name); - shash_delete(&xfif->changed_ports, node); ++ *devnamep = shash_steal(&xfif->changed_ports, node); return 0; } else { return EAGAIN; @@@ -478,9 -477,8 +477,8 @@@ xfif_linux_recv(struct xfif *xfif_, str int retval; int error; - buf = ofpbuf_new(65536 + XFIF_RECV_MSG_PADDING); - ofpbuf_reserve(buf, XFIF_RECV_MSG_PADDING); - buf = ofpbuf_new_with_headroom(65536, DPIF_RECV_MSG_PADDING); - retval = read(dpif->fd, ofpbuf_tail(buf), ofpbuf_tailroom(buf)); ++ buf = ofpbuf_new_with_headroom(65536, XFIF_RECV_MSG_PADDING); + retval = read(xfif->fd, ofpbuf_tail(buf), ofpbuf_tailroom(buf)); if (retval < 0) { error = errno; if (error != EAGAIN) { diff --cc lib/xfif-netdev.c index 1afb08c81,000000000..25493fbd4 mode 100644,000000..100644 --- a/lib/xfif-netdev.c +++ b/lib/xfif-netdev.c @@@ -1,1402 -1,0 +1,1400 @@@ +/* + * Copyright (c) 2009, 2010 Nicira Networks. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include +#include "xfif.h" + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "csum.h" +#include "flow.h" +#include "hmap.h" +#include "list.h" +#include "netdev.h" +#include "xflow-util.h" +#include "ofp-print.h" +#include "ofpbuf.h" +#include "packets.h" +#include "poll-loop.h" +#include "queue.h" +#include "timeval.h" +#include "util.h" +#include "vlog.h" +#include "xfif-provider.h" + +VLOG_DEFINE_THIS_MODULE(xfif_netdev) + +/* Configuration parameters. */ +enum { N_QUEUES = 2 }; /* Number of queues for xfif_recv(). */ +enum { MAX_QUEUE_LEN = 100 }; /* Maximum number of packets per queue. */ +enum { N_GROUPS = 16 }; /* Number of port groups. */ +enum { MAX_PORTS = 256 }; /* Maximum number of ports. */ +enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */ + +/* Enough headroom to add a vlan tag, plus an extra 2 bytes to allow IP + * headers to be aligned on a 4-byte boundary. */ +enum { XF_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN }; + +/* Datapath based on the network device interface from netdev.h. */ +struct xf_netdev { + struct list node; + int xf_idx; + int open_cnt; + bool destroyed; + + bool drop_frags; /* Drop all IP fragments, if true. */ + struct ovs_queue queues[N_QUEUES]; /* Messages queued for xfif_recv(). */ + struct hmap flow_table; /* Flow table. */ + struct xflow_port_group groups[N_GROUPS]; + + /* Statistics. */ + long long int n_frags; /* Number of dropped IP fragments. */ + long long int n_hit; /* Number of flow table matches. */ + long long int n_missed; /* Number of flow table misses. */ + long long int n_lost; /* Number of misses not passed to client. */ + + /* Ports. */ + int n_ports; + struct xf_netdev_port *ports[MAX_PORTS]; + struct list port_list; + unsigned int serial; +}; + +/* A port in a netdev-based datapath. */ +struct xf_netdev_port { + int port_no; /* Index into xf_netdev's 'ports'. */ + struct list node; /* Element in xf_netdev's 'port_list'. */ + struct netdev *netdev; + bool internal; /* Internal port (as XFLOW_PORT_INTERNAL)? */ +}; + +/* A flow in xf_netdev's 'flow_table'. */ +struct xf_netdev_flow { + struct hmap_node node; /* Element in xf_netdev's 'flow_table'. */ + struct xflow_key key; + + /* Statistics. */ + struct timespec used; /* Last used time. */ + long long int packet_count; /* Number of packets matched. */ + long long int byte_count; /* Number of bytes matched. */ + uint16_t tcp_ctl; /* Bitwise-OR of seen tcp_ctl values. */ + + /* Actions. */ + union xflow_action *actions; + unsigned int n_actions; +}; + +/* Interface to netdev-based datapath. */ +struct xfif_netdev { + struct xfif xfif; + struct xf_netdev *xf; + int listen_mask; + unsigned int xf_serial; +}; + +/* All netdev-based datapaths. */ +static struct xf_netdev *xf_netdevs[256]; +struct list xf_netdev_list = LIST_INITIALIZER(&xf_netdev_list); +enum { N_XF_NETDEVS = ARRAY_SIZE(xf_netdevs) }; + +/* Maximum port MTU seen so far. */ +static int max_mtu = ETH_PAYLOAD_MAX; + +static int get_port_by_number(struct xf_netdev *, uint16_t port_no, + struct xf_netdev_port **portp); +static int get_port_by_name(struct xf_netdev *, const char *devname, + struct xf_netdev_port **portp); +static void xf_netdev_free(struct xf_netdev *); +static void xf_netdev_flow_flush(struct xf_netdev *); +static int do_add_port(struct xf_netdev *, const char *devname, uint16_t flags, + uint16_t port_no); +static int do_del_port(struct xf_netdev *, uint16_t port_no); +static int xf_netdev_output_control(struct xf_netdev *, const struct ofpbuf *, + int queue_no, int port_no, uint32_t arg); +static int xf_netdev_execute_actions(struct xf_netdev *, + struct ofpbuf *, struct xflow_key *, + const union xflow_action *, int n); + +static struct xfif_netdev * +xfif_netdev_cast(const struct xfif *xfif) +{ + xfif_assert_class(xfif, &xfif_netdev_class); + return CONTAINER_OF(xfif, struct xfif_netdev, xfif); +} + +static struct xf_netdev * +get_xf_netdev(const struct xfif *xfif) +{ + return xfif_netdev_cast(xfif)->xf; +} + +static int +name_to_xf_idx(const char *name) +{ + if (!strncmp(name, "xf", 2) && isdigit((unsigned char)name[2])) { + int xf_idx = atoi(name + 2); + if (xf_idx >= 0 && xf_idx < N_XF_NETDEVS) { + return xf_idx; + } + } + return -1; +} + +static struct xf_netdev * +find_xf_netdev(const char *name) +{ + int xf_idx; + size_t i; + + xf_idx = name_to_xf_idx(name); + if (xf_idx >= 0) { + return xf_netdevs[xf_idx]; + } + + for (i = 0; i < N_XF_NETDEVS; i++) { + struct xf_netdev *xf = xf_netdevs[i]; + if (xf) { + struct xf_netdev_port *port; + if (!get_port_by_name(xf, name, &port)) { + return xf; + } + } + } + return NULL; +} + +static struct xfif * +create_xfif_netdev(struct xf_netdev *xf) +{ + struct xfif_netdev *xfif; + char *xfname; + + xf->open_cnt++; + + xfname = xasprintf("xf%d", xf->xf_idx); + xfif = xmalloc(sizeof *xfif); + xfif_init(&xfif->xfif, &xfif_netdev_class, xfname, xf->xf_idx, xf->xf_idx); + xfif->xf = xf; + xfif->listen_mask = 0; + xfif->xf_serial = xf->serial; + free(xfname); + + return &xfif->xfif; +} + +static int +create_xf_netdev(const char *name, int xf_idx, struct xfif **xfifp) +{ + struct xf_netdev *xf; + int error; + int i; + + if (xf_netdevs[xf_idx]) { + return EBUSY; + } + + /* Create datapath. */ + xf_netdevs[xf_idx] = xf = xzalloc(sizeof *xf); + list_push_back(&xf_netdev_list, &xf->node); + xf->xf_idx = xf_idx; + xf->open_cnt = 0; + xf->drop_frags = false; + for (i = 0; i < N_QUEUES; i++) { + queue_init(&xf->queues[i]); + } + hmap_init(&xf->flow_table); + for (i = 0; i < N_GROUPS; i++) { + xf->groups[i].ports = NULL; + xf->groups[i].n_ports = 0; + xf->groups[i].group = i; + } + list_init(&xf->port_list); + error = do_add_port(xf, name, XFLOW_PORT_INTERNAL, XFLOWP_LOCAL); + if (error) { + xf_netdev_free(xf); + return ENODEV; + } + + *xfifp = create_xfif_netdev(xf); + return 0; +} + +static int +xfif_netdev_open(const char *name, const char *type OVS_UNUSED, bool create, + struct xfif **xfifp) +{ + if (create) { + if (find_xf_netdev(name)) { + return EEXIST; + } else { + int xf_idx = name_to_xf_idx(name); + if (xf_idx >= 0) { + return create_xf_netdev(name, xf_idx, xfifp); + } else { + /* Scan for unused xf_idx number. */ + for (xf_idx = 0; xf_idx < N_XF_NETDEVS; xf_idx++) { + int error = create_xf_netdev(name, xf_idx, xfifp); + if (error != EBUSY) { + return error; + } + } + + /* All datapath numbers in use. */ + return ENOBUFS; + } + } + } else { + struct xf_netdev *xf = find_xf_netdev(name); + if (xf) { + *xfifp = create_xfif_netdev(xf); + return 0; + } else { + return ENODEV; + } + } +} + +static void +xf_netdev_free(struct xf_netdev *xf) +{ + int i; + + xf_netdev_flow_flush(xf); + while (xf->n_ports > 0) { + struct xf_netdev_port *port = CONTAINER_OF( + xf->port_list.next, struct xf_netdev_port, node); + do_del_port(xf, port->port_no); + } + for (i = 0; i < N_QUEUES; i++) { + queue_destroy(&xf->queues[i]); + } + hmap_destroy(&xf->flow_table); + for (i = 0; i < N_GROUPS; i++) { + free(xf->groups[i].ports); + } + xf_netdevs[xf->xf_idx] = NULL; + list_remove(&xf->node); + free(xf); +} + +static void +xfif_netdev_close(struct xfif *xfif) +{ + struct xf_netdev *xf = get_xf_netdev(xfif); + assert(xf->open_cnt > 0); + if (--xf->open_cnt == 0 && xf->destroyed) { + xf_netdev_free(xf); + } + free(xfif); +} + +static int +xfif_netdev_destroy(struct xfif *xfif) +{ + struct xf_netdev *xf = get_xf_netdev(xfif); + xf->destroyed = true; + return 0; +} + +static int +xfif_netdev_get_stats(const struct xfif *xfif, struct xflow_stats *stats) +{ + struct xf_netdev *xf = get_xf_netdev(xfif); + memset(stats, 0, sizeof *stats); + stats->n_flows = hmap_count(&xf->flow_table); + stats->cur_capacity = hmap_capacity(&xf->flow_table); + stats->max_capacity = MAX_FLOWS; + stats->n_ports = xf->n_ports; + stats->max_ports = MAX_PORTS; + stats->max_groups = N_GROUPS; + stats->n_frags = xf->n_frags; + stats->n_hit = xf->n_hit; + stats->n_missed = xf->n_missed; + stats->n_lost = xf->n_lost; + stats->max_miss_queue = MAX_QUEUE_LEN; + stats->max_action_queue = MAX_QUEUE_LEN; + return 0; +} + +static int +xfif_netdev_get_drop_frags(const struct xfif *xfif, bool *drop_fragsp) +{ + struct xf_netdev *xf = get_xf_netdev(xfif); + *drop_fragsp = xf->drop_frags; + return 0; +} + +static int +xfif_netdev_set_drop_frags(struct xfif *xfif, bool drop_frags) +{ + struct xf_netdev *xf = get_xf_netdev(xfif); + xf->drop_frags = drop_frags; + return 0; +} + +static int +do_add_port(struct xf_netdev *xf, const char *devname, uint16_t flags, + uint16_t port_no) +{ + bool internal = (flags & XFLOW_PORT_INTERNAL) != 0; + struct xf_netdev_port *port; + struct netdev_options netdev_options; + struct netdev *netdev; + int mtu; + int error; + + /* XXX reject devices already in some xf_netdev. */ + + /* Open and validate network device. */ + memset(&netdev_options, 0, sizeof netdev_options); + netdev_options.name = devname; + netdev_options.ethertype = NETDEV_ETH_TYPE_ANY; + if (internal) { + netdev_options.type = "tap"; + } + + error = netdev_open(&netdev_options, &netdev); + if (error) { + return error; + } + /* XXX reject loopback devices */ + /* XXX reject non-Ethernet devices */ + + error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, false); + if (error) { + netdev_close(netdev); + return error; + } + + port = xmalloc(sizeof *port); + port->port_no = port_no; + port->netdev = netdev; + port->internal = internal; + + netdev_get_mtu(netdev, &mtu); + if (mtu > max_mtu) { + max_mtu = mtu; + } + + list_push_back(&xf->port_list, &port->node); + xf->ports[port_no] = port; + xf->n_ports++; + xf->serial++; + + return 0; +} + +static int +xfif_netdev_port_add(struct xfif *xfif, const char *devname, uint16_t flags, + uint16_t *port_nop) +{ + struct xf_netdev *xf = get_xf_netdev(xfif); + int port_no; + + for (port_no = 0; port_no < MAX_PORTS; port_no++) { + if (!xf->ports[port_no]) { + *port_nop = port_no; + return do_add_port(xf, devname, flags, port_no); + } + } + return EFBIG; +} + +static int +xfif_netdev_port_del(struct xfif *xfif, uint16_t port_no) +{ + struct xf_netdev *xf = get_xf_netdev(xfif); + return port_no == XFLOWP_LOCAL ? EINVAL : do_del_port(xf, port_no); +} + +static bool +is_valid_port_number(uint16_t port_no) +{ + return port_no < MAX_PORTS; +} + +static int +get_port_by_number(struct xf_netdev *xf, + uint16_t port_no, struct xf_netdev_port **portp) +{ + if (!is_valid_port_number(port_no)) { + *portp = NULL; + return EINVAL; + } else { + *portp = xf->ports[port_no]; + return *portp ? 0 : ENOENT; + } +} + +static int +get_port_by_name(struct xf_netdev *xf, + const char *devname, struct xf_netdev_port **portp) +{ + struct xf_netdev_port *port; + - LIST_FOR_EACH (port, struct xf_netdev_port, node, &xf->port_list) { ++ LIST_FOR_EACH (port, node, &xf->port_list) { + if (!strcmp(netdev_get_name(port->netdev), devname)) { + *portp = port; + return 0; + } + } + return ENOENT; +} + +static int +do_del_port(struct xf_netdev *xf, uint16_t port_no) +{ + struct xf_netdev_port *port; + char *name; + int error; + + error = get_port_by_number(xf, port_no, &port); + if (error) { + return error; + } + + list_remove(&port->node); + xf->ports[port->port_no] = NULL; + xf->n_ports--; + xf->serial++; + + name = xstrdup(netdev_get_name(port->netdev)); + netdev_close(port->netdev); + + free(name); + free(port); + + return 0; +} + +static void +answer_port_query(const struct xf_netdev_port *port, struct xflow_port *xflow_port) +{ + memset(xflow_port, 0, sizeof *xflow_port); + ovs_strlcpy(xflow_port->devname, netdev_get_name(port->netdev), + sizeof xflow_port->devname); + xflow_port->port = port->port_no; + xflow_port->flags = port->internal ? XFLOW_PORT_INTERNAL : 0; +} + +static int +xfif_netdev_port_query_by_number(const struct xfif *xfif, uint16_t port_no, + struct xflow_port *xflow_port) +{ + struct xf_netdev *xf = get_xf_netdev(xfif); + struct xf_netdev_port *port; + int error; + + error = get_port_by_number(xf, port_no, &port); + if (!error) { + answer_port_query(port, xflow_port); + } + return error; +} + +static int +xfif_netdev_port_query_by_name(const struct xfif *xfif, const char *devname, + struct xflow_port *xflow_port) +{ + struct xf_netdev *xf = get_xf_netdev(xfif); + struct xf_netdev_port *port; + int error; + + error = get_port_by_name(xf, devname, &port); + if (!error) { + answer_port_query(port, xflow_port); + } + return error; +} + +static void +xf_netdev_free_flow(struct xf_netdev *xf, struct xf_netdev_flow *flow) +{ + hmap_remove(&xf->flow_table, &flow->node); + free(flow->actions); + free(flow); +} + +static void +xf_netdev_flow_flush(struct xf_netdev *xf) +{ + struct xf_netdev_flow *flow, *next; + - HMAP_FOR_EACH_SAFE (flow, next, struct xf_netdev_flow, node, - &xf->flow_table) { ++ HMAP_FOR_EACH_SAFE (flow, next, node, &xf->flow_table) { + xf_netdev_free_flow(xf, flow); + } +} + +static int +xfif_netdev_flow_flush(struct xfif *xfif) +{ + struct xf_netdev *xf = get_xf_netdev(xfif); + xf_netdev_flow_flush(xf); + return 0; +} + +static int +xfif_netdev_port_list(const struct xfif *xfif, struct xflow_port *ports, int n) +{ + struct xf_netdev *xf = get_xf_netdev(xfif); + struct xf_netdev_port *port; + int i; + + i = 0; - LIST_FOR_EACH (port, struct xf_netdev_port, node, &xf->port_list) { ++ LIST_FOR_EACH (port, node, &xf->port_list) { + struct xflow_port *xflow_port = &ports[i]; + if (i >= n) { + break; + } + answer_port_query(port, xflow_port); + i++; + } + return xf->n_ports; +} + +static int +xfif_netdev_port_poll(const struct xfif *xfif_, char **devnamep OVS_UNUSED) +{ + struct xfif_netdev *xfif = xfif_netdev_cast(xfif_); + if (xfif->xf_serial != xfif->xf->serial) { + xfif->xf_serial = xfif->xf->serial; + return ENOBUFS; + } else { + return EAGAIN; + } +} + +static void +xfif_netdev_port_poll_wait(const struct xfif *xfif_) +{ + struct xfif_netdev *xfif = xfif_netdev_cast(xfif_); + if (xfif->xf_serial != xfif->xf->serial) { + poll_immediate_wake(); + } +} + +static int +get_port_group(const struct xfif *xfif, int group_no, + struct xflow_port_group **groupp) +{ + struct xf_netdev *xf = get_xf_netdev(xfif); + + if (group_no >= 0 && group_no < N_GROUPS) { + *groupp = &xf->groups[group_no]; + return 0; + } else { + *groupp = NULL; + return EINVAL; + } +} + +static int +xfif_netdev_port_group_get(const struct xfif *xfif, int group_no, + uint16_t ports[], int n) +{ + struct xflow_port_group *group; + int error; + + if (n < 0) { + return -EINVAL; + } + + error = get_port_group(xfif, group_no, &group); + if (!error) { + memcpy(ports, group->ports, MIN(n, group->n_ports) * sizeof *ports); + return group->n_ports; + } else { + return -error; + } +} + +static int +xfif_netdev_port_group_set(struct xfif *xfif, int group_no, + const uint16_t ports[], int n) +{ + struct xflow_port_group *group; + int error; + + if (n < 0 || n > MAX_PORTS) { + return EINVAL; + } + + error = get_port_group(xfif, group_no, &group); + if (!error) { + free(group->ports); + group->ports = xmemdup(ports, n * sizeof *group->ports); + group->n_ports = n; + group->group = group_no; + } + return error; +} + +static struct xf_netdev_flow * +xf_netdev_lookup_flow(const struct xf_netdev *xf, + const struct xflow_key *key) +{ + struct xf_netdev_flow *flow; + - HMAP_FOR_EACH_WITH_HASH (flow, struct xf_netdev_flow, node, ++ HMAP_FOR_EACH_WITH_HASH (flow, node, + xflow_key_hash(key, 0), &xf->flow_table) { + if (xflow_key_equal(&flow->key, key)) { + return flow; + } + } + return NULL; +} + +static void +answer_flow_query(struct xf_netdev_flow *flow, uint32_t query_flags, + struct xflow_flow *xflow_flow) +{ + if (flow) { + xflow_flow->key = flow->key; + xflow_flow->stats.n_packets = flow->packet_count; + xflow_flow->stats.n_bytes = flow->byte_count; + xflow_flow->stats.used_sec = flow->used.tv_sec; + xflow_flow->stats.used_nsec = flow->used.tv_nsec; + xflow_flow->stats.tcp_flags = TCP_FLAGS(flow->tcp_ctl); + xflow_flow->stats.reserved = 0; + xflow_flow->stats.error = 0; + if (xflow_flow->n_actions > 0) { + unsigned int n = MIN(xflow_flow->n_actions, flow->n_actions); + memcpy(xflow_flow->actions, flow->actions, + n * sizeof *xflow_flow->actions); + xflow_flow->n_actions = flow->n_actions; + } + + if (query_flags & XFLOWFF_ZERO_TCP_FLAGS) { + flow->tcp_ctl = 0; + } + + } else { + xflow_flow->stats.error = ENOENT; + } +} + +static int +xfif_netdev_flow_get(const struct xfif *xfif, struct xflow_flow flows[], int n) +{ + struct xf_netdev *xf = get_xf_netdev(xfif); + int i; + + for (i = 0; i < n; i++) { + struct xflow_flow *xflow_flow = &flows[i]; + answer_flow_query(xf_netdev_lookup_flow(xf, &xflow_flow->key), + xflow_flow->flags, xflow_flow); + } + return 0; +} + +static int +xfif_netdev_validate_actions(const union xflow_action *actions, int n_actions, + bool *mutates) +{ + unsigned int i; + + *mutates = false; + for (i = 0; i < n_actions; i++) { + const union xflow_action *a = &actions[i]; + switch (a->type) { + case XFLOWAT_OUTPUT: + if (a->output.port >= MAX_PORTS) { + return EINVAL; + } + break; + + case XFLOWAT_OUTPUT_GROUP: + *mutates = true; + if (a->output_group.group >= N_GROUPS) { + return EINVAL; + } + break; + + case XFLOWAT_CONTROLLER: + break; + + case XFLOWAT_SET_DL_TCI: + *mutates = true; + if (a->dl_tci.mask != htons(VLAN_VID_MASK) + && a->dl_tci.mask != htons(VLAN_PCP_MASK) + && a->dl_tci.mask != htons(VLAN_VID_MASK | VLAN_PCP_MASK)) { + return EINVAL; + } + if (a->dl_tci.tci & ~a->dl_tci.mask){ + return EINVAL; + } + break; + + case XFLOWAT_SET_NW_TOS: + *mutates = true; + if (a->nw_tos.nw_tos & IP_ECN_MASK) { + return EINVAL; + } + break; + + case XFLOWAT_STRIP_VLAN: + case XFLOWAT_SET_DL_SRC: + case XFLOWAT_SET_DL_DST: + case XFLOWAT_SET_NW_SRC: + case XFLOWAT_SET_NW_DST: + case XFLOWAT_SET_TP_SRC: + case XFLOWAT_SET_TP_DST: + *mutates = true; + break; + + default: + return EOPNOTSUPP; + } + } + return 0; +} + +static int +set_flow_actions(struct xf_netdev_flow *flow, struct xflow_flow *xflow_flow) +{ + size_t n_bytes; + bool mutates; + int error; + + if (xflow_flow->n_actions >= 4096 / sizeof *xflow_flow->actions) { + return EINVAL; + } + error = xfif_netdev_validate_actions(xflow_flow->actions, + xflow_flow->n_actions, &mutates); + if (error) { + return error; + } + + n_bytes = xflow_flow->n_actions * sizeof *flow->actions; + flow->actions = xrealloc(flow->actions, n_bytes); + flow->n_actions = xflow_flow->n_actions; + memcpy(flow->actions, xflow_flow->actions, n_bytes); + return 0; +} + +static int +add_flow(struct xfif *xfif, struct xflow_flow *xflow_flow) +{ + struct xf_netdev *xf = get_xf_netdev(xfif); + struct xf_netdev_flow *flow; + int error; + + flow = xzalloc(sizeof *flow); + flow->key = xflow_flow->key; + + error = set_flow_actions(flow, xflow_flow); + if (error) { + free(flow); + return error; + } + + hmap_insert(&xf->flow_table, &flow->node, + xflow_key_hash(&flow->key, 0)); + return 0; +} + +static void +clear_stats(struct xf_netdev_flow *flow) +{ + flow->used.tv_sec = 0; + flow->used.tv_nsec = 0; + flow->packet_count = 0; + flow->byte_count = 0; + flow->tcp_ctl = 0; +} + +static int +xfif_netdev_flow_put(struct xfif *xfif, struct xflow_flow_put *put) +{ + struct xf_netdev *xf = get_xf_netdev(xfif); + struct xf_netdev_flow *flow; + + flow = xf_netdev_lookup_flow(xf, &put->flow.key); + if (!flow) { + if (put->flags & XFLOWPF_CREATE) { + if (hmap_count(&xf->flow_table) < MAX_FLOWS) { + return add_flow(xfif, &put->flow); + } else { + return EFBIG; + } + } else { + return ENOENT; + } + } else { + if (put->flags & XFLOWPF_MODIFY) { + int error = set_flow_actions(flow, &put->flow); + if (!error && put->flags & XFLOWPF_ZERO_STATS) { + clear_stats(flow); + } + return error; + } else { + return EEXIST; + } + } +} + + +static int +xfif_netdev_flow_del(struct xfif *xfif, struct xflow_flow *xflow_flow) +{ + struct xf_netdev *xf = get_xf_netdev(xfif); + struct xf_netdev_flow *flow; + + flow = xf_netdev_lookup_flow(xf, &xflow_flow->key); + if (flow) { + answer_flow_query(flow, 0, xflow_flow); + xf_netdev_free_flow(xf, flow); + return 0; + } else { + return ENOENT; + } +} + +static int +xfif_netdev_flow_list(const struct xfif *xfif, struct xflow_flow flows[], int n) +{ + struct xf_netdev *xf = get_xf_netdev(xfif); + struct xf_netdev_flow *flow; + int i; + + i = 0; - HMAP_FOR_EACH (flow, struct xf_netdev_flow, node, &xf->flow_table) { ++ HMAP_FOR_EACH (flow, node, &xf->flow_table) { + if (i >= n) { + break; + } + answer_flow_query(flow, 0, &flows[i++]); + } + return hmap_count(&xf->flow_table); +} + +static int +xfif_netdev_execute(struct xfif *xfif, uint16_t in_port, + const union xflow_action actions[], int n_actions, + const struct ofpbuf *packet) +{ + struct xf_netdev *xf = get_xf_netdev(xfif); + struct ofpbuf copy; + bool mutates; + struct xflow_key key; + flow_t flow; + int error; + + if (packet->size < ETH_HEADER_LEN || packet->size > UINT16_MAX) { + return EINVAL; + } + + error = xfif_netdev_validate_actions(actions, n_actions, &mutates); + if (error) { + return error; + } + + if (mutates) { + /* We need a deep copy of 'packet' since we're going to modify its + * data. */ + ofpbuf_init(©, XF_NETDEV_HEADROOM + packet->size); + copy.data = (char*)copy.base + XF_NETDEV_HEADROOM; + ofpbuf_put(©, packet->data, packet->size); + } else { + /* We still need a shallow copy of 'packet', even though we won't + * modify its data, because flow_extract() modifies packet->l2, etc. + * We could probably get away with modifying those but it's more polite + * if we don't. */ + copy = *packet; + } + flow_extract(©, 0, in_port, &flow); + xflow_key_from_flow(&key, &flow); + error = xf_netdev_execute_actions(xf, ©, &key, actions, n_actions); + if (mutates) { + ofpbuf_uninit(©); + } + return error; +} + +static int +xfif_netdev_recv_get_mask(const struct xfif *xfif, int *listen_mask) +{ + struct xfif_netdev *xfif_netdev = xfif_netdev_cast(xfif); + *listen_mask = xfif_netdev->listen_mask; + return 0; +} + +static int +xfif_netdev_recv_set_mask(struct xfif *xfif, int listen_mask) +{ + struct xfif_netdev *xfif_netdev = xfif_netdev_cast(xfif); + if (!(listen_mask & ~XFLOWL_ALL)) { + xfif_netdev->listen_mask = listen_mask; + return 0; + } else { + return EINVAL; + } +} + +static struct ovs_queue * +find_nonempty_queue(struct xfif *xfif) +{ + struct xfif_netdev *xfif_netdev = xfif_netdev_cast(xfif); + struct xf_netdev *xf = get_xf_netdev(xfif); + int mask = xfif_netdev->listen_mask; + int i; + + for (i = 0; i < N_QUEUES; i++) { + struct ovs_queue *q = &xf->queues[i]; + if (q->n && mask & (1u << i)) { + return q; + } + } + return NULL; +} + +static int +xfif_netdev_recv(struct xfif *xfif, struct ofpbuf **bufp) +{ + struct ovs_queue *q = find_nonempty_queue(xfif); + if (q) { + *bufp = queue_pop_head(q); + return 0; + } else { + return EAGAIN; + } +} + +static void +xfif_netdev_recv_wait(struct xfif *xfif) +{ + struct ovs_queue *q = find_nonempty_queue(xfif); + if (q) { + poll_immediate_wake(); + } else { + /* No messages ready to be received, and xf_wait() will ensure that we + * wake up to queue new messages, so there is nothing to do. */ + } +} + +static void +xf_netdev_flow_used(struct xf_netdev_flow *flow, + const struct xflow_key *key, + const struct ofpbuf *packet) +{ + time_timespec(&flow->used); + flow->packet_count++; + flow->byte_count += packet->size; + if (key->dl_type == htons(ETH_TYPE_IP) && key->nw_proto == IPPROTO_TCP) { + struct tcp_header *th = packet->l4; + flow->tcp_ctl |= th->tcp_ctl; + } +} + +static void +xf_netdev_port_input(struct xf_netdev *xf, struct xf_netdev_port *port, + struct ofpbuf *packet) +{ + struct xf_netdev_flow *flow; + struct xflow_key key; + flow_t f; + + if (packet->size < ETH_HEADER_LEN) { + return; + } + if (flow_extract(packet, 0, port->port_no, &f) && xf->drop_frags) { + xf->n_frags++; + return; + } + xflow_key_from_flow(&key, &f); + + flow = xf_netdev_lookup_flow(xf, &key); + if (flow) { + xf_netdev_flow_used(flow, &key, packet); + xf_netdev_execute_actions(xf, packet, &key, + flow->actions, flow->n_actions); + xf->n_hit++; + } else { + xf->n_missed++; + xf_netdev_output_control(xf, packet, _XFLOWL_MISS_NR, port->port_no, 0); + } +} + +static void +xf_netdev_run(void) +{ + struct ofpbuf packet; + struct xf_netdev *xf; + + ofpbuf_init(&packet, XF_NETDEV_HEADROOM + max_mtu); - LIST_FOR_EACH (xf, struct xf_netdev, node, &xf_netdev_list) { ++ LIST_FOR_EACH (xf, node, &xf_netdev_list) { + struct xf_netdev_port *port; + - LIST_FOR_EACH (port, struct xf_netdev_port, node, &xf->port_list) { ++ LIST_FOR_EACH (port, node, &xf->port_list) { + int error; + + /* Reset packet contents. */ + packet.data = (char*)packet.base + XF_NETDEV_HEADROOM; + packet.size = 0; + + error = netdev_recv(port->netdev, &packet); + if (!error) { + xf_netdev_port_input(xf, port, &packet); + } else if (error != EAGAIN) { + struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_ERR_RL(&rl, "error receiving data from %s: %s", + netdev_get_name(port->netdev), strerror(error)); + } + } + } + ofpbuf_uninit(&packet); +} + +static void +xf_netdev_wait(void) +{ + struct xf_netdev *xf; + - LIST_FOR_EACH (xf, struct xf_netdev, node, &xf_netdev_list) { ++ LIST_FOR_EACH (xf, node, &xf_netdev_list) { + struct xf_netdev_port *port; - LIST_FOR_EACH (port, struct xf_netdev_port, node, &xf->port_list) { ++ LIST_FOR_EACH (port, node, &xf->port_list) { + netdev_recv_wait(port->netdev); + } + } +} + + +/* Modify or add a 802.1Q header in 'packet' according to 'a'. */ +static void +xf_netdev_set_dl_tci(struct ofpbuf *packet, + const struct xflow_action_dl_tci *a) +{ + struct vlan_eth_header *veh; + struct eth_header *eh; + + eh = packet->l2; + if (packet->size >= sizeof(struct vlan_eth_header) + && eh->eth_type == htons(ETH_TYPE_VLAN)) { + veh = packet->l2; + veh->veth_tci = (veh->veth_tci & ~a->mask) | a->tci; + } else { + /* Insert new 802.1Q header. */ - struct eth_header *eh = packet->l2; + struct vlan_eth_header tmp; + memcpy(tmp.veth_dst, eh->eth_dst, ETH_ADDR_LEN); + memcpy(tmp.veth_src, eh->eth_src, ETH_ADDR_LEN); + tmp.veth_type = htons(ETH_TYPE_VLAN); + tmp.veth_tci = htons(a->tci); + tmp.veth_next_type = eh->eth_type; + + veh = ofpbuf_push_uninit(packet, VLAN_HEADER_LEN); + memcpy(veh, &tmp, sizeof tmp); + packet->l2 = (char*)packet->l2 - VLAN_HEADER_LEN; + } +} + +static void +xf_netdev_strip_vlan(struct ofpbuf *packet, struct xflow_key *key) +{ + struct vlan_eth_header *veh = packet->l2; + if (packet->size >= sizeof *veh + && veh->veth_type == htons(ETH_TYPE_VLAN)) { + struct eth_header tmp; + + memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN); + memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN); + tmp.eth_type = veh->veth_next_type; + + packet->size -= VLAN_HEADER_LEN; + packet->data = (char*)packet->data + VLAN_HEADER_LEN; + packet->l2 = (char*)packet->l2 + VLAN_HEADER_LEN; + memcpy(packet->data, &tmp, sizeof tmp); + + key->dl_tci = htons(0); + } +} + +static void +xf_netdev_set_dl_src(struct ofpbuf *packet, + const uint8_t dl_addr[ETH_ADDR_LEN]) +{ + struct eth_header *eh = packet->l2; + memcpy(eh->eth_src, dl_addr, sizeof eh->eth_src); +} + +static void +xf_netdev_set_dl_dst(struct ofpbuf *packet, + const uint8_t dl_addr[ETH_ADDR_LEN]) +{ + struct eth_header *eh = packet->l2; + memcpy(eh->eth_dst, dl_addr, sizeof eh->eth_dst); +} + +static bool +is_ip(const struct ofpbuf *packet, const struct xflow_key *key) +{ + return key->dl_type == htons(ETH_TYPE_IP) && packet->l4; +} + +static void +xf_netdev_set_nw_addr(struct ofpbuf *packet, const struct xflow_key *key, + const struct xflow_action_nw_addr *a) +{ + if (is_ip(packet, key)) { + struct ip_header *nh = packet->l3; + uint32_t *field; + + field = a->type == XFLOWAT_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst; + if (key->nw_proto == IP_TYPE_TCP && packet->l7) { + struct tcp_header *th = packet->l4; + th->tcp_csum = recalc_csum32(th->tcp_csum, *field, a->nw_addr); + } else if (key->nw_proto == IP_TYPE_UDP && packet->l7) { + struct udp_header *uh = packet->l4; + if (uh->udp_csum) { + uh->udp_csum = recalc_csum32(uh->udp_csum, *field, a->nw_addr); + if (!uh->udp_csum) { + uh->udp_csum = 0xffff; + } + } + } + nh->ip_csum = recalc_csum32(nh->ip_csum, *field, a->nw_addr); + *field = a->nw_addr; + } +} + +static void +xf_netdev_set_nw_tos(struct ofpbuf *packet, const struct xflow_key *key, + const struct xflow_action_nw_tos *a) +{ + if (is_ip(packet, key)) { + struct ip_header *nh = packet->l3; + uint8_t *field = &nh->ip_tos; + + /* Set the DSCP bits and preserve the ECN bits. */ + uint8_t new = a->nw_tos | (nh->ip_tos & IP_ECN_MASK); + + nh->ip_csum = recalc_csum16(nh->ip_csum, htons((uint16_t)*field), + htons((uint16_t)a->nw_tos)); + *field = new; + } +} + +static void +xf_netdev_set_tp_port(struct ofpbuf *packet, const struct xflow_key *key, + const struct xflow_action_tp_port *a) +{ + if (is_ip(packet, key)) { + uint16_t *field; + if (key->nw_proto == IPPROTO_TCP && packet->l7) { + struct tcp_header *th = packet->l4; + field = a->type == XFLOWAT_SET_TP_SRC ? &th->tcp_src : &th->tcp_dst; + th->tcp_csum = recalc_csum16(th->tcp_csum, *field, a->tp_port); + *field = a->tp_port; + } else if (key->nw_proto == IPPROTO_UDP && packet->l7) { + struct udp_header *uh = packet->l4; + field = a->type == XFLOWAT_SET_TP_SRC ? &uh->udp_src : &uh->udp_dst; + uh->udp_csum = recalc_csum16(uh->udp_csum, *field, a->tp_port); + *field = a->tp_port; + } else { + return; + } + } +} + +static void +xf_netdev_output_port(struct xf_netdev *xf, struct ofpbuf *packet, + uint16_t out_port) +{ + struct xf_netdev_port *p = xf->ports[out_port]; + if (p) { + netdev_send(p->netdev, packet); + } +} + +static void +xf_netdev_output_group(struct xf_netdev *xf, uint16_t group, uint16_t in_port, + struct ofpbuf *packet) +{ + struct xflow_port_group *g = &xf->groups[group]; + int i; + + for (i = 0; i < g->n_ports; i++) { + uint16_t out_port = g->ports[i]; + if (out_port != in_port) { + xf_netdev_output_port(xf, packet, out_port); + } + } +} + +static int +xf_netdev_output_control(struct xf_netdev *xf, const struct ofpbuf *packet, + int queue_no, int port_no, uint32_t arg) +{ + struct ovs_queue *q = &xf->queues[queue_no]; + struct xflow_msg *header; + struct ofpbuf *msg; + size_t msg_size; + + if (q->n >= MAX_QUEUE_LEN) { + xf->n_lost++; + return ENOBUFS; + } + + msg_size = sizeof *header + packet->size; + msg = ofpbuf_new(msg_size + XFIF_RECV_MSG_PADDING); + header = ofpbuf_put_uninit(msg, sizeof *header); + ofpbuf_reserve(msg, XFIF_RECV_MSG_PADDING); + header->type = queue_no; + header->length = msg_size; + header->port = port_no; + header->arg = arg; + ofpbuf_put(msg, packet->data, packet->size); + queue_push_tail(q, msg); + + return 0; +} + +/* Returns true if 'packet' is an invalid Ethernet+IPv4 ARP packet: one with + * screwy or truncated header fields or one whose inner and outer Ethernet + * address differ. */ +static bool +xf_netdev_is_spoofed_arp(struct ofpbuf *packet, const struct xflow_key *key) +{ + struct arp_eth_header *arp; + struct eth_header *eth; + ptrdiff_t l3_size; + + if (key->dl_type != htons(ETH_TYPE_ARP)) { + return false; + } + + l3_size = (char *) ofpbuf_end(packet) - (char *) packet->l3; + if (l3_size < sizeof(struct arp_eth_header)) { + return true; + } + + eth = packet->l2; + arp = packet->l3; + return (arp->ar_hrd != htons(ARP_HRD_ETHERNET) + || arp->ar_pro != htons(ARP_PRO_IP) + || arp->ar_hln != ETH_HEADER_LEN + || arp->ar_pln != 4 + || !eth_addr_equals(arp->ar_sha, eth->eth_src)); +} + +static int +xf_netdev_execute_actions(struct xf_netdev *xf, + struct ofpbuf *packet, struct xflow_key *key, + const union xflow_action *actions, int n_actions) +{ + int i; + for (i = 0; i < n_actions; i++) { + const union xflow_action *a = &actions[i]; + + switch (a->type) { + case XFLOWAT_OUTPUT: + xf_netdev_output_port(xf, packet, a->output.port); + break; + + case XFLOWAT_OUTPUT_GROUP: + xf_netdev_output_group(xf, a->output_group.group, key->in_port, + packet); + break; + + case XFLOWAT_CONTROLLER: + xf_netdev_output_control(xf, packet, _XFLOWL_ACTION_NR, + key->in_port, a->controller.arg); + break; + + case XFLOWAT_SET_DL_TCI: + xf_netdev_set_dl_tci(packet, &a->dl_tci); + break; + + case XFLOWAT_STRIP_VLAN: + xf_netdev_strip_vlan(packet, key); + break; + + case XFLOWAT_SET_DL_SRC: + xf_netdev_set_dl_src(packet, a->dl_addr.dl_addr); + break; + + case XFLOWAT_SET_DL_DST: + xf_netdev_set_dl_dst(packet, a->dl_addr.dl_addr); + break; + + case XFLOWAT_SET_NW_SRC: + case XFLOWAT_SET_NW_DST: + xf_netdev_set_nw_addr(packet, key, &a->nw_addr); + break; + + case XFLOWAT_SET_NW_TOS: + xf_netdev_set_nw_tos(packet, key, &a->nw_tos); + break; + + case XFLOWAT_SET_TP_SRC: + case XFLOWAT_SET_TP_DST: + xf_netdev_set_tp_port(packet, key, &a->tp_port); + break; + + case XFLOWAT_DROP_SPOOFED_ARP: + if (xf_netdev_is_spoofed_arp(packet, key)) { + return 0; + } + } + } + return 0; +} + +const struct xfif_class xfif_netdev_class = { + "netdev", + xf_netdev_run, + xf_netdev_wait, + NULL, /* enumerate */ + xfif_netdev_open, + xfif_netdev_close, + NULL, /* get_all_names */ + xfif_netdev_destroy, + xfif_netdev_get_stats, + xfif_netdev_get_drop_frags, + xfif_netdev_set_drop_frags, + xfif_netdev_port_add, + xfif_netdev_port_del, + xfif_netdev_port_query_by_number, + xfif_netdev_port_query_by_name, + xfif_netdev_port_list, + xfif_netdev_port_poll, + xfif_netdev_port_poll_wait, + xfif_netdev_port_group_get, + xfif_netdev_port_group_set, + xfif_netdev_flow_get, + xfif_netdev_flow_put, + xfif_netdev_flow_del, + xfif_netdev_flow_flush, + xfif_netdev_flow_list, + xfif_netdev_execute, + xfif_netdev_recv_get_mask, + xfif_netdev_recv_set_mask, + NULL, /* get_sflow_probability */ + NULL, /* set_sflow_probability */ + NULL, /* queue_to_priority */ + xfif_netdev_recv, + xfif_netdev_recv_wait, +}; diff --cc lib/xflow-util.c index e11da07fc,000000000..622f9efdc mode 100644,000000..100644 --- a/lib/xflow-util.c +++ b/lib/xflow-util.c @@@ -1,207 -1,0 +1,210 @@@ +/* + * Copyright (c) 2009, 2010 Nicira Networks. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include +#include "xflow-util.h" +#include +#include +#include +#include "coverage.h" +#include "dynamic-string.h" +#include "flow.h" +#include "packets.h" +#include "timeval.h" +#include "util.h" + +union xflow_action * +xflow_actions_add(struct xflow_actions *actions, uint16_t type) +{ + union xflow_action *a; + size_t idx; + + idx = actions->n_actions++ & (MAX_XFLOW_ACTIONS - 1); + a = &actions->actions[idx]; + memset(a, 0, sizeof *a); + a->type = type; + return a; +} + +void +format_xflow_key(struct ds *ds, const struct xflow_key *key) +{ + ds_put_format(ds, "tunnel%"PRIx32":in_port%04x", + key->tun_id, key->in_port); + if (key->dl_tci) { + ds_put_format(ds, ":vlan%"PRIu16":pcp%d", + vlan_tci_to_vid(key->dl_tci), + vlan_tci_to_pcp(key->dl_tci)); + } + ds_put_format(ds, " mac"ETH_ADDR_FMT"->"ETH_ADDR_FMT" type%04x " + "proto%"PRId8" tos%"PRIu8" ip"IP_FMT"->"IP_FMT" port%d->%d", + ETH_ADDR_ARGS(key->dl_src), ETH_ADDR_ARGS(key->dl_dst), + ntohs(key->dl_type), key->nw_proto, key->nw_tos, + IP_ARGS(&key->nw_src), IP_ARGS(&key->nw_dst), + ntohs(key->tp_src), ntohs(key->tp_dst)); +} + +void +format_xflow_action(struct ds *ds, const union xflow_action *a) +{ + switch (a->type) { + case XFLOWAT_OUTPUT: + ds_put_format(ds, "%"PRIu16, a->output.port); + break; + case XFLOWAT_OUTPUT_GROUP: + ds_put_format(ds, "g%"PRIu16, a->output_group.group); + break; + case XFLOWAT_CONTROLLER: + ds_put_format(ds, "ctl(%"PRIu32")", a->controller.arg); + break; + case XFLOWAT_SET_TUNNEL: + ds_put_format(ds, "set_tunnel(0x%08"PRIx32")", ntohl(a->tunnel.tun_id)); + break; + case XFLOWAT_SET_DL_TCI: + ds_put_format(ds, "set_tci(%04"PRIx16",mask=%04"PRIx16")", + ntohs(a->dl_tci.tci), ntohs(a->dl_tci.mask)); + break; + case XFLOWAT_STRIP_VLAN: + ds_put_format(ds, "strip_vlan"); + break; + case XFLOWAT_SET_DL_SRC: + ds_put_format(ds, "set_dl_src("ETH_ADDR_FMT")", + ETH_ADDR_ARGS(a->dl_addr.dl_addr)); + break; + case XFLOWAT_SET_DL_DST: + ds_put_format(ds, "set_dl_dst("ETH_ADDR_FMT")", + ETH_ADDR_ARGS(a->dl_addr.dl_addr)); + break; + case XFLOWAT_SET_NW_SRC: + ds_put_format(ds, "set_nw_src("IP_FMT")", + IP_ARGS(&a->nw_addr.nw_addr)); + break; + case XFLOWAT_SET_NW_DST: + ds_put_format(ds, "set_nw_dst("IP_FMT")", + IP_ARGS(&a->nw_addr.nw_addr)); + break; + case XFLOWAT_SET_NW_TOS: + ds_put_format(ds, "set_nw_tos(%"PRIu8")", a->nw_tos.nw_tos); + break; + case XFLOWAT_SET_TP_SRC: + ds_put_format(ds, "set_tp_src(%"PRIu16")", ntohs(a->tp_port.tp_port)); + break; + case XFLOWAT_SET_TP_DST: + ds_put_format(ds, "set_tp_dst(%"PRIu16")", ntohs(a->tp_port.tp_port)); + break; + case XFLOWAT_SET_PRIORITY: + ds_put_format(ds, "set_priority(0x%"PRIx32")", a->priority.priority); + break; + case XFLOWAT_POP_PRIORITY: + ds_put_cstr(ds, "pop_priority"); + break; ++ case XFLOWAT_DROP_SPOOFED_ARP: ++ ds_put_cstr(ds, "drop_spoofed_arp"); ++ break; + default: + ds_put_format(ds, "***bad action 0x%"PRIx16"***", a->type); + break; + } +} + +void +format_xflow_actions(struct ds *ds, const union xflow_action *actions, + size_t n_actions) +{ + size_t i; + for (i = 0; i < n_actions; i++) { + if (i) { + ds_put_char(ds, ','); + } + format_xflow_action(ds, &actions[i]); + } + if (!n_actions) { + ds_put_cstr(ds, "drop"); + } +} + +void +format_xflow_flow_stats(struct ds *ds, const struct xflow_flow_stats *s) +{ + ds_put_format(ds, "packets:%llu, bytes:%llu, used:", + (unsigned long long int) s->n_packets, + (unsigned long long int) s->n_bytes); + if (s->used_sec) { + long long int used = s->used_sec * 1000 + s->used_nsec / 1000000; + ds_put_format(ds, "%.3fs", (time_msec() - used) / 1000.0); + } else { + ds_put_format(ds, "never"); + } +} + +void +format_xflow_flow(struct ds *ds, const struct xflow_flow *f) +{ + format_xflow_key(ds, &f->key); + ds_put_cstr(ds, ", "); + format_xflow_flow_stats(ds, &f->stats); + ds_put_cstr(ds, ", actions:"); + format_xflow_actions(ds, f->actions, f->n_actions); +} + +void +xflow_key_from_flow(struct xflow_key *key, const struct flow *flow) +{ + key->tun_id = flow->tun_id; + key->nw_src = flow->nw_src; + key->nw_dst = flow->nw_dst; + key->in_port = ofp_port_to_xflow_port(flow->in_port); + if (flow->dl_vlan == htons(OFP_VLAN_NONE)) { + key->dl_tci = htons(0); + } else { + uint16_t vid = flow->dl_vlan & htons(VLAN_VID_MASK); + uint16_t pcp = htons((flow->dl_vlan_pcp << VLAN_PCP_SHIFT) + & VLAN_PCP_MASK); + key->dl_tci = vid | pcp | htons(XFLOW_TCI_PRESENT); + } + key->dl_type = flow->dl_type; + key->tp_src = flow->tp_src; + key->tp_dst = flow->tp_dst; + memcpy(key->dl_src, flow->dl_src, ETH_ADDR_LEN); + memcpy(key->dl_dst, flow->dl_dst, ETH_ADDR_LEN); + key->nw_proto = flow->nw_proto; + key->nw_tos = flow->nw_tos; +} + +void +xflow_key_to_flow(const struct xflow_key *key, struct flow *flow) +{ + flow->wildcards = 0; + flow->priority = 0xffff; + flow->tun_id = key->tun_id; + flow->nw_src = key->nw_src; + flow->nw_dst = key->nw_dst; + flow->in_port = xflow_port_to_ofp_port(key->in_port); + if (key->dl_tci) { + flow->dl_vlan = htons(vlan_tci_to_vid(key->dl_tci)); + flow->dl_vlan_pcp = vlan_tci_to_pcp(key->dl_tci); + } else { + flow->dl_vlan = htons(OFP_VLAN_NONE); + flow->dl_vlan_pcp = 0; + } + flow->dl_type = key->dl_type; + flow->tp_src = key->tp_src; + flow->tp_dst = key->tp_dst; + memcpy(flow->dl_src, key->dl_src, ETH_ADDR_LEN); + memcpy(flow->dl_dst, key->dl_dst, ETH_ADDR_LEN); + flow->nw_proto = key->nw_proto; + flow->nw_tos = key->nw_tos; +} diff --cc ofproto/ofproto-sflow.c index 7ed61af10,f129d38be..372491159 --- a/ofproto/ofproto-sflow.c +++ b/ofproto/ofproto-sflow.c @@@ -20,13 -20,14 +20,14 @@@ #include #include #include "collectors.h" -#include "dpif.h" #include "compiler.h" + #include "hash.h" + #include "hmap.h" #include "netdev.h" #include "ofpbuf.h" #include "ofproto.h" +#include "packets.h" #include "poll-loop.h" - #include "port-array.h" #include "sflow_api.h" #include "socket-util.h" #include "timeval.h" @@@ -37,8 -36,10 +38,10 @@@ VLOG_DEFINE_THIS_MODULE(sflow) struct ofproto_sflow_port { + struct hmap_node hmap_node; /* In struct ofproto_sflow's "ports" hmap. */ struct netdev *netdev; /* Underlying network device, for stats. */ SFLDataSource_instance dsi; /* sFlow library's notion of port number. */ - uint16_t odp_port; /* ODP port number. */ ++ uint16_t xflow_port; /* xflow port number. */ }; struct ofproto_sflow { @@@ -46,12 -47,15 +49,15 @@@ struct collectors *collectors; SFLAgent *sflow_agent; struct ofproto_sflow_options *options; - struct dpif *dpif; + struct wdp *wdp; time_t next_tick; size_t n_flood, n_all; - struct port_array ports; /* Indexed by XFLOW port number. */ + struct hmap ports; /* Contains "struct ofproto_sflow_port"s. */ }; + static void ofproto_sflow_del_port__(struct ofproto_sflow *, + struct ofproto_sflow_port *); + #define RECEIVER_INDEX 1 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); @@@ -131,6 -135,20 +137,20 @@@ sflow_agent_send_packet_cb(void *os_, S collectors_send(os->collectors, pkt, pktLen); } + static struct ofproto_sflow_port * -ofproto_sflow_find_port(const struct ofproto_sflow *os, uint16_t odp_port) ++ofproto_sflow_find_port(const struct ofproto_sflow *os, uint16_t xflow_port) + { + struct ofproto_sflow_port *osp; + + HMAP_FOR_EACH_IN_BUCKET (osp, hmap_node, - hash_int(odp_port, 0), &os->ports) { - if (osp->odp_port == odp_port) { ++ hash_int(xflow_port, 0), &os->ports) { ++ if (osp->xflow_port == xflow_port) { + return osp; + } + } + return NULL; + } + static void sflow_agent_get_counters(void *os_, SFLPoller *poller, SFL_COUNTERS_SAMPLE_TYPE *cs) @@@ -266,9 -284,9 +286,9 @@@ ofproto_sflow_create(struct wdp *wdp struct ofproto_sflow *os; os = xcalloc(1, sizeof *os); - os->dpif = dpif; + os->wdp = wdp; os->next_tick = time_now() + 1; - port_array_init(&os->ports); + hmap_init(&os->ports); return os; } @@@ -333,10 -350,11 +352,11 @@@ ofproto_sflow_add_port(struct ofproto_s osp->netdev = netdev; ifindex = netdev_get_ifindex(netdev); if (ifindex <= 0) { - ifindex = (os->sflow_agent->subId << 16) + odp_port; + ifindex = (os->sflow_agent->subId << 16) + xflow_port; } SFL_DS_SET(osp->dsi, 0, ifindex, 0); - port_array_set(&os->ports, xflow_port, osp); - osp->odp_port = odp_port; - hmap_insert(&os->ports, &osp->hmap_node, hash_int(odp_port, 0)); ++ osp->xflow_port = xflow_port; ++ hmap_insert(&os->ports, &osp->hmap_node, hash_int(xflow_port, 0)); /* Add poller and sampler. */ if (os->sflow_agent) { @@@ -345,18 -363,25 +365,25 @@@ } } + static void + ofproto_sflow_del_port__(struct ofproto_sflow *os, + struct ofproto_sflow_port *osp) + { + if (os->sflow_agent) { + sfl_agent_removePoller(os->sflow_agent, &osp->dsi); + sfl_agent_removeSampler(os->sflow_agent, &osp->dsi); + } + netdev_close(osp->netdev); + hmap_remove(&os->ports, &osp->hmap_node); + free(osp); + } + void -ofproto_sflow_del_port(struct ofproto_sflow *os, uint16_t odp_port) +ofproto_sflow_del_port(struct ofproto_sflow *os, uint16_t xflow_port) { - struct ofproto_sflow_port *osp = port_array_get(&os->ports, xflow_port); - struct ofproto_sflow_port *osp = ofproto_sflow_find_port(os, odp_port); ++ struct ofproto_sflow_port *osp = ofproto_sflow_find_port(os, xflow_port); if (osp) { - if (os->sflow_agent) { - sfl_agent_removePoller(os->sflow_agent, &osp->dsi); - sfl_agent_removeSampler(os->sflow_agent, &osp->dsi); - } - netdev_close(osp->netdev); - free(osp); - port_array_delete(&os->ports, xflow_port); + ofproto_sflow_del_port__(os, osp); } } @@@ -434,21 -458,21 +460,21 @@@ ofproto_sflow_set_options(struct ofprot sfl_receiver_set_sFlowRcvrTimeout(receiver, 0xffffffff); /* Set the sampling_rate down in the datapath. */ - dpif_set_sflow_probability(os->dpif, - MAX(1, UINT32_MAX / options->sampling_rate)); + wdp_set_sflow_probability(os->wdp, + MAX(1, UINT32_MAX / options->sampling_rate)); /* Add samplers and pollers for the currently known ports. */ - PORT_ARRAY_FOR_EACH (osp, &os->ports, xflow_port) { - ofproto_sflow_add_poller(os, osp, xflow_port); + HMAP_FOR_EACH (osp, hmap_node, &os->ports) { - ofproto_sflow_add_poller(os, osp, osp->odp_port); ++ ofproto_sflow_add_poller(os, osp, osp->xflow_port); ofproto_sflow_add_sampler(os, osp); } } static int -ofproto_sflow_odp_port_to_ifindex(const struct ofproto_sflow *os, - uint16_t odp_port) +ofproto_sflow_xflow_port_to_ifindex(const struct ofproto_sflow *os, - uint16_t xflow_port) ++ uint16_t xflow_port) { - struct ofproto_sflow_port *osp = port_array_get(&os->ports, xflow_port); - struct ofproto_sflow_port *osp = ofproto_sflow_find_port(os, odp_port); ++ struct ofproto_sflow_port *osp = ofproto_sflow_find_port(os, xflow_port); return osp ? SFL_DS_INDEX(osp->dsi) : 0; } diff --cc ofproto/ofproto.c index 3c165c71b,3d2989a62..cb67deb2b --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@@ -27,8 -27,11 +27,10 @@@ #include "classifier.h" #include "coverage.h" #include "discovery.h" -#include "dpif.h" #include "dynamic-string.h" #include "fail-open.h" + #include "hash.h" + #include "hmap.h" #include "in-band.h" #include "mac-learning.h" #include "netdev.h" @@@ -821,9 -949,9 +820,8 @@@ ofproto_destroy(struct ofproto *p free(p->extra_in_band_remotes); ofproto_flush_flows(p); - classifier_destroy(&p->cls); - LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node, - &p->all_conns) { + LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) { ofconn_destroy(ofconn); } hmap_destroy(&p->controllers); @@@ -907,30 -1055,6 +904,30 @@@ add_snooper(struct ofproto *ofproto, st } } +static void +ofproto_port_poll_cb(const struct ofp_phy_port *opp, uint8_t reason, + void *ofproto_) +{ + /* XXX Should limit the number of queued port status change messages. */ + struct ofproto *ofproto = ofproto_; + struct ofconn *ofconn; + - LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) { ++ LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { + struct ofp_port_status *ops; + struct ofpbuf *b; + + if (!ofconn_receives_async_msgs(ofconn)) { + continue; + } + + ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b); + ops->reason = reason; + ops->desc = *opp; + hton_ofp_phy_port(&ops->desc); + queue_tx(b, ofconn, NULL); + } +} + int ofproto_run1(struct ofproto *p) { @@@ -1037,9 -1206,10 +1032,9 @@@ ofproto_wait(struct ofproto *p struct ofconn *ofconn; size_t i; - dpif_recv_wait(p->dpif); - dpif_port_poll_wait(p->dpif); - netdev_monitor_poll_wait(p->netdev_monitor); + wdp_recv_wait(p->wdp); + wdp_port_poll_wait(p->wdp); - LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { + LIST_FOR_EACH (ofconn, node, &p->all_conns) { ofconn_wait(ofconn); } if (p->in_band) { @@@ -1052,7 -1222,17 +1047,7 @@@ if (p->sflow) { ofproto_sflow_wait(p->sflow); } - HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) { - if (!tag_set_is_empty(&p->revalidate_set)) { - poll_immediate_wake(); - } - if (p->need_revalidate) { - /* Shouldn't happen, but if it does just go around again. */ - VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); - poll_immediate_wake(); - } else if (p->next_expiration != LLONG_MAX) { - poll_timer_wait_until(p->next_expiration); - } + HMAP_FOR_EACH (ofservice, node, &p->services) { pvconn_wait(ofservice->pvconn); } for (i = 0; i < p->n_snoops; i++) { @@@ -1945,8 -3423,8 +1940,8 @@@ handle_aggregate_stats_request(struct o struct queue_stats_cbdata { struct ofconn *ofconn; - struct ofport *ofport; ++ struct wdp_port *wdp_port; struct ofpbuf *msg; - uint16_t port_no; }; static void @@@ -1956,7 -3434,7 +1951,7 @@@ put_queue_stats(struct queue_stats_cbda struct ofp_queue_stats *reply; reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg); - reply->port_no = htons(cbdata->port_no); - reply->port_no = htons(cbdata->ofport->opp.port_no); ++ reply->port_no = htons(cbdata->wdp_port->opp.port_no); memset(reply->pad, 0, sizeof reply->pad); reply->queue_id = htonl(queue_id); reply->tx_bytes = htonll(stats->tx_bytes); @@@ -1975,10 -3453,10 +1970,10 @@@ handle_queue_stats_dump_cb(uint32_t que } static void -handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id, +handle_queue_stats_for_port(struct wdp_port *port, uint32_t queue_id, struct queue_stats_cbdata *cbdata) { - cbdata->port_no = port->opp.port_no; - cbdata->ofport = port; ++ cbdata->wdp_port = port; if (queue_id == OFPQ_ALL) { netdev_dump_queue_stats(port->netdev, handle_queue_stats_dump_cb, cbdata); @@@ -2764,50 -4316,137 +2759,50 @@@ delete_flow(struct ofproto *p, struct w * being added (and expiring). (It also prevents processing OpenFlow * requests that would not add new flows, so it is imperfect.) */ - prev = NULL; - LIST_FOR_EACH (ofconn, node, &p->all_conns) { - if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn) - && ofconn_receives_async_msgs(ofconn)) { - if (prev) { - queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter); - } else { - buf = compose_flow_removed(p, rule, now, reason); - } - prev = ofconn; - } - } - if (prev) { - queue_tx(buf, prev, prev->reply_counter); - } -} - - -static void -expire_rule(struct cls_rule *cls_rule, void *p_) -{ - struct ofproto *p = p_; - struct rule *rule = rule_from_cls_rule(cls_rule); - long long int hard_expire, idle_expire, expire, now; - - hard_expire = (rule->hard_timeout - ? rule->created + rule->hard_timeout * 1000 - : LLONG_MAX); - idle_expire = (rule->idle_timeout - && (rule->super || list_is_empty(&rule->list)) - ? rule->used + rule->idle_timeout * 1000 - : LLONG_MAX); - expire = MIN(hard_expire, idle_expire); - - now = time_msec(); - if (now < expire) { - if (rule->installed && now >= rule->used + 5000) { - uninstall_idle_flow(p, rule); - } else if (!rule->cr.wc.wildcards) { - active_timeout(p, rule); - } - - return; - } - - COVERAGE_INC(ofproto_expired); + struct ofproto_rule *ofproto_rule = ofproto_rule_cast(rule); + struct wdp_flow_stats stats; + struct ofpbuf *buf; + int error; - /* Update stats. This code will be a no-op if the rule expired - * due to an idle timeout. */ - if (rule->cr.wc.wildcards) { - struct rule *subrule, *next; - LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) { - rule_remove(p, subrule); - } + if (ofproto_rule->send_flow_removed) { + /* Compose most of the ofp_flow_removed before 'rule' is destroyed. */ + buf = compose_flow_removed(p, rule, reason); } else { - rule_uninstall(p, rule); - } - - if (!rule_is_hidden(rule)) { - send_flow_removed(p, rule, now, - (now >= hard_expire - ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT)); - } - rule_remove(p, rule); -} - -static void -active_timeout(struct ofproto *ofproto, struct rule *rule) -{ - if (ofproto->netflow && !is_controller_rule(rule) && - netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) { - struct ofexpired expired; - struct odp_flow odp_flow; - - /* Get updated flow stats. */ - memset(&odp_flow, 0, sizeof odp_flow); - if (rule->installed) { - odp_flow.key = rule->cr.flow; - odp_flow.flags = ODPFF_ZERO_TCP_FLAGS; - dpif_flow_get(ofproto->dpif, &odp_flow); - - if (odp_flow.stats.n_packets) { - update_time(ofproto, rule, &odp_flow.stats); - netflow_flow_update_flags(&rule->nf_flow, - odp_flow.stats.tcp_flags); - } - } - - expired.flow = rule->cr.flow; - expired.packet_count = rule->packet_count + - odp_flow.stats.n_packets; - expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes; - expired.used = rule->used; - - netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); - - /* Schedule us to send the accumulated records once we have - * collected all of them. */ - poll_immediate_wake(); + buf = NULL; } -} - -static void -update_used(struct ofproto *p) -{ - struct odp_flow *flows; - size_t n_flows; - size_t i; - int error; - error = dpif_flow_list_all(p->dpif, &flows, &n_flows); + error = wdp_flow_delete(p->wdp, rule, &stats); if (error) { - return; + return error; } - for (i = 0; i < n_flows; i++) { - struct odp_flow *f = &flows[i]; - struct rule *rule; + if (buf) { + struct ofp_flow_removed *ofr; + struct ofconn *prev = NULL; + struct ofconn *ofconn; - rule = rule_from_cls_rule( - classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX)); - if (!rule || !rule->installed) { - COVERAGE_INC(ofproto_unexpected_rule); - dpif_flow_del(p->dpif, f); - continue; - } + /* Compose the parts of the ofp_flow_removed that require stats. */ + ofr = buf->data; + ofr->packet_count = htonll(stats.n_packets); + ofr->byte_count = htonll(stats.n_bytes); - LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { - update_time(p, rule, &f->stats); - rule_account(p, rule, f->stats.n_bytes); ++ LIST_FOR_EACH (ofconn, node, &p->all_conns) { + if (rconn_is_connected(ofconn->rconn)) { + if (prev) { + queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter); + } + prev = ofconn; + } + } + if (prev) { + queue_tx(buf, prev, prev->reply_counter); + } else { + ofpbuf_delete(buf); + } } - free(flows); + free(ofproto_rule); + + return 0; } /* pinsched callback for sending 'packet' on 'ofconn'. */ diff --cc ofproto/wdp-xflow.c index ce7ff2f1c,000000000..26830e666 mode 100644,000000..100644 --- a/ofproto/wdp-xflow.c +++ b/ofproto/wdp-xflow.c @@@ -1,2659 -1,0 +1,2724 @@@ +/* + * Copyright (c) 2010 Nicira Networks. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include + +#include "wdp-xflow.h" + +#include +#include + +#include "coverage.h" +#include "dhcp.h" +#include "mac-learning.h" +#include "netdev.h" +#include "netflow.h" +#include "ofp-util.h" +#include "ofpbuf.h" +#include "ofproto.h" +#include "openflow/nicira-ext.h" +#include "openflow/openflow.h" +#include "packets.h" +#include "poll-loop.h" +#include "port-array.h" +#include "queue.h" +#include "shash.h" +#include "svec.h" +#include "timeval.h" +#include "util.h" +#include "vconn.h" +#include "wdp-provider.h" +#include "xfif.h" +#include "xflow-util.h" +#include "vlog.h" +#include "xtoxll.h" + +VLOG_DEFINE_THIS_MODULE(wdp_xflow) + +enum { + TABLEID_HASH = 0, + TABLEID_CLASSIFIER = 1 +}; + +static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + +/* Maximum numbers of rules. */ +#define WX_MAX_WILD 65536 /* Wildcarded rules. */ +#define WX_MAX_EXACT 1048576 /* Exact-match rules. */ + ++struct wx_port { ++ struct hmap_node hmap_node; ++ struct wdp_port wdp_port; ++ uint16_t xflow_port; ++}; ++ +struct wx { + struct list list_node; + struct wdp wdp; + struct xfif *xfif; + struct classifier cls; + struct netdev_monitor *netdev_monitor; - struct port_array ports; /* Index is xflow port nr; - * wdp_port->opp.port_no is OFP port nr. */ ++ struct hmap ports; /* Contains "struct wx_port"s. */ + struct shash port_by_name; + long long int next_expiration; + int wdp_listen_mask; + + /* Rules that might need to be revalidated. */ + bool need_revalidate; /* Revalidate all subrules? */ + bool revalidate_all; /* Revalidate all subrules and other rules? */ + struct tag_set revalidate_set; /* Tag set of (sub)rules to revalidate. */ + + /* Hooks for ovs-vswitchd. */ + const struct ofhooks *ofhooks; + void *aux; + + /* Used by default ofhooks. */ + struct mac_learning *ml; + + /* List of "struct wdp_packets" queued for the controller by + * execute_xflow_actions(). */ +#define MAX_CTL_PACKETS 50 + struct list ctl_packets; + int n_ctl_packets; +}; + +static const struct ofhooks default_ofhooks; + +static struct list all_wx = LIST_INITIALIZER(&all_wx); + +static int wx_port_init(struct wx *); ++static struct wx_port *wx_port_get(const struct wx *, uint16_t xflow_port); +static void wx_port_process_change(struct wx *wx, int error, char *devname, + wdp_port_poll_cb_func *cb, void *aux); +static void wx_port_refresh_groups(struct wx *); + +static void wx_purge_ctl_packets__(struct wx *); + +enum { + WX_GROUP_FLOOD = 0, + WX_GROUP_ALL = 1 +}; + +static struct wx * +wx_cast(const struct wdp *wdp) +{ + return CONTAINER_OF(wdp, struct wx, wdp); +} + +static int +wx_xlate_actions(struct wx *, const union ofp_action *, size_t n, + const flow_t *flow, const struct ofpbuf *packet, + tag_type *tags, struct xflow_actions *out, + bool *may_set_up_flow); + +struct wx_rule { + struct wdp_rule wr; + + uint64_t packet_count; /* Number of packets received. */ + uint64_t byte_count; /* Number of bytes received. */ + uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */ + long long int used; /* Last-used time (0 if never used). */ + tag_type tags; /* Tags (set only by hooks). */ + + /* If 'super' is non-NULL, this rule is a subrule, that is, it is an + * exact-match rule (having cr.wc.wildcards of 0) generated from the + * wildcard rule 'super'. In this case, 'list' is an element of the + * super-rule's list. + * + * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of + * a list of subrules. A super-rule with no wildcards (where + * cr.wc.wildcards is 0) will never have any subrules. */ + struct wx_rule *super; + struct list list; + + /* Datapath actions. + * + * A super-rule with wildcard fields never has xflow actions (since the + * datapath only supports exact-match flows). */ + bool installed; /* Installed in datapath? */ + bool may_install; /* True ordinarily; false if actions must + * be reassessed for every packet. */ + int n_xflow_actions; + union xflow_action *xflow_actions; +}; + +static void wx_rule_destroy(struct wx *, struct wx_rule *); +static void wx_rule_update_actions(struct wx *, struct wx_rule *); +static void wx_rule_execute(struct wx *, struct wx_rule *, + struct ofpbuf *packet, const flow_t *); +static bool wx_rule_make_actions(struct wx *, struct wx_rule *, + const struct ofpbuf *packet); +static void wx_rule_install(struct wx *, struct wx_rule *, + struct wx_rule *displaced_rule); + +static struct wx_rule * +wx_rule_cast(const struct cls_rule *cls_rule) +{ + return cls_rule ? CONTAINER_OF(cls_rule, struct wx_rule, wr.cr) : NULL; +} + +/* Returns true if 'rule' is merely an implementation detail that should be + * hidden from the client. */ +static inline bool +wx_rule_is_hidden(const struct wx_rule *rule) +{ + return rule->super != NULL; +} + +static void +wx_rule_free(struct wx_rule *rule) +{ + wdp_rule_uninit(&rule->wr); + free(rule->xflow_actions); + free(rule); +} + +static void +wx_rule_account(struct wx *wx OVS_UNUSED, struct wx_rule *rule OVS_UNUSED, + uint64_t extra_bytes OVS_UNUSED) +{ + /* XXX call account_cb hook */ +} + +static void +wx_rule_post_uninstall(struct wx *wx, struct wx_rule *rule) +{ + struct wx_rule *super = rule->super; + + wx_rule_account(wx, rule, 0); + + /* XXX netflow expiration */ + + if (super) { + super->packet_count += rule->packet_count; + super->byte_count += rule->byte_count; + + /* Reset counters to prevent double counting if the rule ever gets + * reinstalled. */ + rule->packet_count = 0; + rule->byte_count = 0; + rule->accounted_bytes = 0; + + //XXX netflow_flow_clear(&rule->nf_flow); + } +} + +static long long int +xflow_flow_stats_to_msec(const struct xflow_flow_stats *stats) +{ + return (stats->used_sec + ? stats->used_sec * 1000 + stats->used_nsec / 1000000 + : 0); +} + +static void +wx_rule_update_time(struct wx *wx OVS_UNUSED, struct wx_rule *rule, + const struct xflow_flow_stats *stats) +{ + long long int used = xflow_flow_stats_to_msec(stats); + if (used > rule->used) { + rule->used = used; + if (rule->super && used > rule->super->used) { + rule->super->used = used; + } + //XXX netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used); + } +} + +static void +wx_rule_update_stats(struct wx *wx, struct wx_rule *rule, + const struct xflow_flow_stats *stats) +{ + if (stats->n_packets) { + wx_rule_update_time(wx, rule, stats); + rule->packet_count += stats->n_packets; + rule->byte_count += stats->n_bytes; + /* XXX netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags); */ + } +} + +static void +wx_rule_uninstall(struct wx *wx, struct wx_rule *rule) +{ + assert(!rule->wr.cr.flow.wildcards); + if (rule->installed) { + struct xflow_flow xflow_flow; + + xflow_key_from_flow(&xflow_flow.key, &rule->wr.cr.flow); + xflow_flow.actions = NULL; + xflow_flow.n_actions = 0; + xflow_flow.flags = 0; + if (!xfif_flow_del(wx->xfif, &xflow_flow)) { + wx_rule_update_stats(wx, rule, &xflow_flow.stats); + } + rule->installed = false; + + wx_rule_post_uninstall(wx, rule); + } +} + +#if 0 +static bool +is_controller_rule(struct wx_rule *rule) +{ + /* If the only action is send to the controller then don't report + * NetFlow expiration messages since it is just part of the control + * logic for the network and not real traffic. */ + + return (rule + && rule->super + && rule->super->n_actions == 1 + && action_outputs_to_port(&rule->super->actions[0], + htons(OFPP_CONTROLLER))); +} +#endif + +static void +wx_rule_remove(struct wx *wx, struct wx_rule *rule) +{ + if (rule->wr.cr.flow.wildcards) { + COVERAGE_INC(wx_del_wc_flow); + wx->need_revalidate = true; + } else { + wx_rule_uninstall(wx, rule); + } + classifier_remove(&wx->cls, &rule->wr.cr); + wx_rule_destroy(wx, rule); +} + +static bool +wx_rule_revalidate(struct wx *wx, struct wx_rule *rule) +{ + const flow_t *flow = &rule->wr.cr.flow; + + COVERAGE_INC(wx_rule_revalidate); + if (rule->super) { + struct wx_rule *super; + super = wx_rule_cast(classifier_lookup_wild(&wx->cls, flow)); + if (!super) { + wx_rule_remove(wx, rule); + return false; + } else if (super != rule->super) { + COVERAGE_INC(wx_revalidate_moved); + list_remove(&rule->list); + list_push_back(&super->list, &rule->list); + rule->super = super; + rule->wr.hard_timeout = super->wr.hard_timeout; + rule->wr.idle_timeout = super->wr.idle_timeout; + rule->wr.created = super->wr.created; + rule->used = 0; + } + } + + wx_rule_update_actions(wx, rule); + return true; +} + +/* Destroys 'rule'. If 'rule' is a subrule, also removes it from its + * super-rule's list of subrules. If 'rule' is a super-rule, also iterates + * through all of its subrules and revalidates them, destroying any that no + * longer has a super-rule (which is probably all of them). + * + * Before calling this function, the caller must make have removed 'rule' from + * the classifier. If 'rule' is an exact-match rule, the caller is also + * responsible for ensuring that it has been uninstalled from the datapath. */ +static void +wx_rule_destroy(struct wx *wx, struct wx_rule *rule) +{ + if (!rule->super) { + struct wx_rule *subrule, *next; - LIST_FOR_EACH_SAFE (subrule, next, struct wx_rule, list, &rule->list) { ++ LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) { + wx_rule_revalidate(wx, subrule); + } + } else { + list_remove(&rule->list); + } + wx_rule_free(rule); +} + +#if 0 +static bool +wx_rule_has_out_port(const struct wx_rule *rule, uint16_t out_port) +{ + const union ofp_action *oa; + struct actions_iterator i; + + if (out_port == htons(OFPP_NONE)) { + return true; + } + for (oa = actions_first(&i, rule->wr.actions, + rule->wr.n_actions); + oa; + oa = actions_next(&i)) { + if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) { + return true; + } + } + return false; +} +#endif + +/* Caller is responsible for initializing the 'cr' and ofp_table_id members of + * the returned rule. */ +static struct wx_rule * +wx_rule_create(struct wx_rule *super, + const union ofp_action *actions, size_t n_actions, + uint16_t idle_timeout, uint16_t hard_timeout) +{ + struct wx_rule *rule = xzalloc(sizeof *rule); + wdp_rule_init(&rule->wr, actions, n_actions); + rule->wr.idle_timeout = idle_timeout; + rule->wr.hard_timeout = hard_timeout; + rule->used = rule->wr.created; + rule->super = super; + if (super) { + list_push_back(&super->list, &rule->list); + } else { + list_init(&rule->list); + } +#if 0 + netflow_flow_clear(&rule->nf_flow); + netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created); +#endif + + return rule; +} + +/* Executes, within 'wx', the 'n_actions' actions in 'actions' on 'packet', + * which arrived on 'in_port'. + * + * Takes ownership of 'packet'. */ +static bool +execute_xflow_actions(struct wx *wx, uint16_t in_port, + const union xflow_action *actions, size_t n_actions, + struct ofpbuf *packet) +{ + if (n_actions == 1 && actions[0].type == XFLOWAT_CONTROLLER + && wx->n_ctl_packets < MAX_CTL_PACKETS) { + /* As an optimization, avoid a round-trip from userspace to kernel to + * userspace. This also avoids possibly filling up kernel packet + * buffers along the way. */ + struct wdp_packet *wdp_packet; + + if (!(wx->wdp_listen_mask & WDP_CHAN_ACTION)) { + return true; + } + + wdp_packet = xmalloc(sizeof *wdp_packet); + wdp_packet->channel = WDP_CHAN_ACTION; + wdp_packet->tun_id = 0; + wdp_packet->in_port = in_port; + wdp_packet->send_len = actions[0].controller.arg; + wdp_packet->payload = packet; + + list_push_back(&wx->ctl_packets, &wdp_packet->list); + + return true; + } else { + int error; + + error = xfif_execute(wx->xfif, in_port, actions, n_actions, packet); + ofpbuf_delete(packet); + return !error; + } +} + +/* Executes the actions indicated by 'rule' on 'packet', which is in flow + * 'flow' and is considered to have arrived on xflow port 'in_port'. 'packet' + * must have at least sizeof(struct ofp_packet_in) bytes of headroom. + * + * The flow that 'packet' actually contains does not need to actually match + * 'rule'; the actions in 'rule' will be applied to it either way. Likewise, + * the packet and byte counters for 'rule' will be credited for the packet sent + * out whether or not the packet actually matches 'rule'. + * + * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow, + * the caller must already have accurately composed xflow actions for it given + * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if + * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this + * function will compose a set of xflow actions based on 'rule''s OpenFlow + * actions and apply them to 'packet'. + * + * Takes ownership of 'packet'. */ +static void +wx_rule_execute(struct wx *wx, struct wx_rule *rule, + struct ofpbuf *packet, const flow_t *flow) +{ + const union xflow_action *actions; + struct xflow_flow_stats stats; + size_t n_actions; + struct xflow_actions a; + + assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); + + /* Grab or compose the xflow actions. + * + * The special case for an exact-match 'rule' where 'flow' is not the + * rule's flow is important to avoid, e.g., sending a packet out its input + * port simply because the xflow actions were composed for the wrong + * scenario. */ + if (rule->wr.cr.flow.wildcards + || !flow_equal_headers(flow, &rule->wr.cr.flow)) + { + struct wx_rule *super = rule->super ? rule->super : rule; + if (wx_xlate_actions(wx, super->wr.actions, super->wr.n_actions, flow, + packet, NULL, &a, NULL)) { + ofpbuf_delete(packet); + return; + } + actions = a.actions; + n_actions = a.n_actions; + } else { + actions = rule->xflow_actions; + n_actions = rule->n_xflow_actions; + } + + /* Execute the xflow actions. */ + flow_extract_stats(flow, packet, &stats); + if (!execute_xflow_actions(wx, flow->in_port, + actions, n_actions, packet)) { + wx_rule_update_stats(wx, rule, &stats); + rule->used = time_msec(); + //XXX netflow_flow_update_time(wx->netflow, &rule->nf_flow, rule->used); + } +} + +/* Inserts 'rule' into 'p''s flow table. + * + * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s + * actions on it and credits the statistics for sending the packet to 'rule'. + * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of + * headroom. */ +static void +wx_rule_insert(struct wx *wx, struct wx_rule *rule, struct ofpbuf *packet, + uint16_t in_port) +{ + struct wx_rule *displaced_rule; + + /* Insert the rule in the classifier. */ + displaced_rule = wx_rule_cast(classifier_insert(&wx->cls, &rule->wr.cr)); + if (!rule->wr.cr.flow.wildcards) { + wx_rule_make_actions(wx, rule, packet); + } + + /* Send the packet and credit it to the rule. */ + if (packet) { + flow_t flow; + flow_extract(packet, 0, in_port, &flow); + wx_rule_execute(wx, rule, packet, &flow); + } + + /* Install the rule in the datapath only after sending the packet, to + * avoid packet reordering. */ + if (rule->wr.cr.flow.wildcards) { + COVERAGE_INC(wx_add_wc_flow); + wx->need_revalidate = true; + } else { + wx_rule_install(wx, rule, displaced_rule); + } + + /* Free the rule that was displaced, if any. */ + if (displaced_rule) { + rule->wr.client_data = displaced_rule->wr.client_data; + wx_rule_destroy(wx, displaced_rule); + } +} + +static struct wx_rule * +wx_rule_create_subrule(struct wx *wx, struct wx_rule *rule, const flow_t *flow) +{ + struct wx_rule *subrule; + + subrule = wx_rule_create(rule, NULL, 0, + rule->wr.idle_timeout, + rule->wr.hard_timeout); + /* Subrules aren't really in any OpenFlow table, so don't bother with + * subrule->wr.ofp_table_id. */ + COVERAGE_INC(wx_subrule_create); + cls_rule_from_flow(flow, &subrule->wr.cr); + classifier_insert_exact(&wx->cls, &subrule->wr.cr); + + return subrule; +} + +/* Returns true if the actions changed, false otherwise. */ +static bool +wx_rule_make_actions(struct wx *wx, struct wx_rule *rule, + const struct ofpbuf *packet) +{ + const struct wx_rule *super; + struct xflow_actions a; + size_t actions_len; + + assert(!rule->wr.cr.flow.wildcards); + + super = rule->super ? rule->super : rule; + wx_xlate_actions(wx, super->wr.actions, super->wr.n_actions, + &rule->wr.cr.flow, packet, + &rule->tags, &a, &rule->may_install); + + actions_len = a.n_actions * sizeof *a.actions; + if (rule->n_xflow_actions != a.n_actions + || memcmp(rule->xflow_actions, a.actions, actions_len)) { + COVERAGE_INC(wx_xflow_unchanged); + free(rule->xflow_actions); + rule->n_xflow_actions = a.n_actions; + rule->xflow_actions = xmemdup(a.actions, actions_len); + return true; + } else { + return false; + } +} + +static int +do_put_flow(struct wx *wx, struct wx_rule *rule, int flags, + struct xflow_flow_put *put) +{ + memset(&put->flow.stats, 0, sizeof put->flow.stats); + xflow_key_from_flow(&put->flow.key, &rule->wr.cr.flow); + put->flow.actions = rule->xflow_actions; + put->flow.n_actions = rule->n_xflow_actions; + put->flow.flags = 0; + put->flags = flags; + return xfif_flow_put(wx->xfif, put); +} + +static void +wx_rule_install(struct wx *wx, struct wx_rule *rule, struct wx_rule *displaced_rule) +{ + assert(!rule->wr.cr.flow.wildcards); + + if (rule->may_install) { + struct xflow_flow_put put; + if (!do_put_flow(wx, rule, + XFLOWPF_CREATE | XFLOWPF_MODIFY | XFLOWPF_ZERO_STATS, + &put)) { + rule->installed = true; + if (displaced_rule) { + wx_rule_update_stats(wx, displaced_rule, &put.flow.stats); + wx_rule_post_uninstall(wx, displaced_rule); + } + } + } else if (displaced_rule) { + wx_rule_uninstall(wx, displaced_rule); + } +} + +static void +wx_rule_reinstall(struct wx *wx, struct wx_rule *rule) +{ + if (rule->installed) { + struct xflow_flow_put put; + COVERAGE_INC(wx_dp_missed); + do_put_flow(wx, rule, XFLOWPF_CREATE | XFLOWPF_MODIFY, &put); + } else { + wx_rule_install(wx, rule, NULL); + } +} + +static void +wx_rule_update_actions(struct wx *wx, struct wx_rule *rule) +{ + bool actions_changed; +#if 0 + uint16_t new_out_iface, old_out_iface; + + old_out_iface = rule->nf_flow.output_iface; +#endif + actions_changed = wx_rule_make_actions(wx, rule, NULL); + + if (rule->may_install) { + if (rule->installed) { + if (actions_changed) { + struct xflow_flow_put put; + do_put_flow(wx, rule, XFLOWPF_CREATE | XFLOWPF_MODIFY + | XFLOWPF_ZERO_STATS, &put); + wx_rule_update_stats(wx, rule, &put.flow.stats); +#if 0 + /* Temporarily set the old output iface so that NetFlow + * messages have the correct output interface for the old + * stats. */ + new_out_iface = rule->nf_flow.output_iface; + rule->nf_flow.output_iface = old_out_iface; +#endif + wx_rule_post_uninstall(wx, rule); + //rule->nf_flow.output_iface = new_out_iface; + } + } else { + wx_rule_install(wx, rule, NULL); + } + } else { + wx_rule_uninstall(wx, rule); + } +} + +static void +add_output_group_action(struct xflow_actions *actions, uint16_t group, + uint16_t *nf_output_iface) +{ + xflow_actions_add(actions, XFLOWAT_OUTPUT_GROUP)->output_group.group = group; + + if (group == WX_GROUP_ALL || group == WX_GROUP_FLOOD) { + *nf_output_iface = NF_OUT_FLOOD; + } +} + +static void +add_controller_action(struct xflow_actions *actions, uint16_t max_len) +{ + union xflow_action *a = xflow_actions_add(actions, XFLOWAT_CONTROLLER); + a->controller.arg = max_len; +} + +struct wx_xlate_ctx { + /* Input. */ + flow_t flow; /* Flow to which these actions correspond. */ + int recurse; /* Recursion level, via xlate_table_action. */ + struct wx *wx; + const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a + * null pointer if we are revalidating + * without a packet to refer to. */ + + /* Output. */ + struct xflow_actions *out; /* Datapath actions. */ + tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */ + bool may_set_up_flow; /* True ordinarily; false if the actions must + * be reassessed for every packet. */ + uint16_t nf_output_iface; /* Output interface index for NetFlow. */ +}; + +static void do_xlate_actions(const union ofp_action *in, size_t n_in, + struct wx_xlate_ctx *ctx); + +static void +add_output_action(struct wx_xlate_ctx *ctx, uint16_t port) +{ - const struct wdp_port *wdp_port = port_array_get(&ctx->wx->ports, port); ++ const struct wx_port *wx_port = wx_port_get(ctx->wx, port); + - if (wdp_port) { - if (wdp_port->opp.config & OFPPC_NO_FWD) { ++ if (wx_port) { ++ if (wx_port->wdp_port.opp.config & OFPPC_NO_FWD) { + /* Forwarding disabled on port. */ + return; + } + } else { + /* + * We don't have an ofport record for this port, but it doesn't hurt to + * allow forwarding to it anyhow. Maybe such a port will appear later + * and we're pre-populating the flow table. + */ + } + + xflow_actions_add(ctx->out, XFLOWAT_OUTPUT)->output.port = port; + //ctx->nf_output_iface = port; +} + +static struct wx_rule * +wx_rule_lookup_valid(struct wx *wx, const flow_t *flow) +{ + struct wx_rule *rule = wx_rule_cast(classifier_lookup(&wx->cls, flow)); + + /* The rule we found might not be valid, since we could be in need of + * revalidation. If it is not valid, don't return it. */ + if (rule + && rule->super + && wx->need_revalidate + && !wx_rule_revalidate(wx, rule)) { + COVERAGE_INC(wx_invalidated); + return NULL; + } + + return rule; +} + +static void +xlate_table_action(struct wx_xlate_ctx *ctx, uint16_t in_port) +{ + if (!ctx->recurse) { + uint16_t old_in_port; + struct wx_rule *rule; + + /* Look up a flow with 'in_port' as the input port. Then restore the + * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will + * have surprising behavior). */ + old_in_port = ctx->flow.in_port; + ctx->flow.in_port = in_port; + rule = wx_rule_lookup_valid(ctx->wx, &ctx->flow); + ctx->flow.in_port = old_in_port; + + if (rule) { + if (rule->super) { + rule = rule->super; + } + + ctx->recurse++; + do_xlate_actions(rule->wr.actions, rule->wr.n_actions, ctx); + ctx->recurse--; + } + } +} + +static void +xlate_output_action__(struct wx_xlate_ctx *ctx, + uint16_t port, uint16_t max_len) +{ + uint16_t xflow_port; + uint16_t prev_nf_output_iface = ctx->nf_output_iface; + + ctx->nf_output_iface = NF_OUT_DROP; + + switch (port) { + case OFPP_IN_PORT: + add_output_action(ctx, ctx->flow.in_port); + break; + case OFPP_TABLE: + xlate_table_action(ctx, ctx->flow.in_port); + break; + case OFPP_NORMAL: + if (!ctx->wx->ofhooks->normal_cb(&ctx->flow, ctx->packet, + ctx->out, ctx->tags, + &ctx->nf_output_iface, + ctx->wx->aux)) { + COVERAGE_INC(wx_uninstallable); + ctx->may_set_up_flow = false; + } + break; + + case OFPP_FLOOD: + add_output_group_action(ctx->out, WX_GROUP_FLOOD, + &ctx->nf_output_iface); + break; + case OFPP_ALL: + add_output_group_action(ctx->out, WX_GROUP_ALL, &ctx->nf_output_iface); + break; + case OFPP_CONTROLLER: + add_controller_action(ctx->out, max_len); + break; + case OFPP_LOCAL: + add_output_action(ctx, XFLOWP_LOCAL); + break; + default: + xflow_port = ofp_port_to_xflow_port(port); + if (xflow_port != ctx->flow.in_port) { + add_output_action(ctx, xflow_port); + } + break; + } + + if (prev_nf_output_iface == NF_OUT_FLOOD) { + ctx->nf_output_iface = NF_OUT_FLOOD; + } else if (ctx->nf_output_iface == NF_OUT_DROP) { + ctx->nf_output_iface = prev_nf_output_iface; + } else if (prev_nf_output_iface != NF_OUT_DROP && + ctx->nf_output_iface != NF_OUT_FLOOD) { + ctx->nf_output_iface = NF_OUT_MULTI; + } +} + +static void +xlate_output_action(struct wx_xlate_ctx *ctx, + const struct ofp_action_output *oao) +{ + xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); +} + +/* If the final xflow action in 'ctx' is "pop priority", drop it, as an + * optimization, because we're going to add another action that sets the + * priority immediately after, or because there are no actions following the + * pop. */ +static void +remove_pop_action(struct wx_xlate_ctx *ctx) +{ + size_t n = ctx->out->n_actions; + if (n > 0 && ctx->out->actions[n - 1].type == XFLOWAT_POP_PRIORITY) { + ctx->out->n_actions--; + } +} + +static void +xlate_enqueue_action(struct wx_xlate_ctx *ctx, + const struct ofp_action_enqueue *oae) +{ + uint16_t ofp_port, xflow_port; + uint32_t priority; + int error; + + error = xfif_queue_to_priority(ctx->wx->xfif, ntohl(oae->queue_id), + &priority); + if (error) { + /* Fall back to ordinary output action. */ + xlate_output_action__(ctx, ntohs(oae->port), 0); + return; + } + + /* Figure out xflow output port. */ + ofp_port = ntohs(oae->port); + if (ofp_port != OFPP_IN_PORT) { + xflow_port = ofp_port_to_xflow_port(ofp_port); + } else { + xflow_port = ctx->flow.in_port; + } + + /* Add xflow actions. */ + remove_pop_action(ctx); + xflow_actions_add(ctx->out, XFLOWAT_SET_PRIORITY)->priority.priority + = priority; + add_output_action(ctx, xflow_port); + xflow_actions_add(ctx->out, XFLOWAT_POP_PRIORITY); + + /* Update NetFlow output port. */ + if (ctx->nf_output_iface == NF_OUT_DROP) { + ctx->nf_output_iface = xflow_port; + } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { + ctx->nf_output_iface = NF_OUT_MULTI; + } +} + ++static void ++xlate_set_queue_action(struct wx_xlate_ctx *ctx, ++ const struct nx_action_set_queue *nasq) ++{ ++ uint32_t priority; ++ int error; ++ ++ error = xfif_queue_to_priority(ctx->wx->xfif, ntohl(nasq->queue_id), ++ &priority); ++ if (error) { ++ /* Couldn't translate queue to a priority, so ignore. A warning ++ * has already been logged. */ ++ return; ++ } ++ ++ remove_pop_action(ctx); ++ xflow_actions_add(ctx->out, XFLOWAT_SET_PRIORITY)->priority.priority ++ = priority; ++} ++ +static void +xlate_nicira_action(struct wx_xlate_ctx *ctx, + const struct nx_action_header *nah) +{ + const struct nx_action_resubmit *nar; + const struct nx_action_set_tunnel *nast; ++ const struct nx_action_set_queue *nasq; + union xflow_action *oa; + int subtype = ntohs(nah->subtype); + + assert(nah->vendor == htonl(NX_VENDOR_ID)); + switch (subtype) { + case NXAST_RESUBMIT: + nar = (const struct nx_action_resubmit *) nah; + xlate_table_action(ctx, ofp_port_to_xflow_port(ntohs(nar->in_port))); + break; + + case NXAST_SET_TUNNEL: + nast = (const struct nx_action_set_tunnel *) nah; + oa = xflow_actions_add(ctx->out, XFLOWAT_SET_TUNNEL); + ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id; + break; + + case NXAST_DROP_SPOOFED_ARP: + if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) { + xflow_actions_add(ctx->out, XFLOWAT_DROP_SPOOFED_ARP); + } + break; + ++ case NXAST_SET_QUEUE: ++ nasq = (const struct nx_action_set_queue *) nah; ++ xlate_set_queue_action(ctx, nasq); ++ break; ++ ++ case NXAST_POP_QUEUE: ++ xflow_actions_add(ctx->out, XFLOWAT_POP_PRIORITY); ++ break; ++ + /* If you add a new action here that modifies flow data, don't forget to + * update the flow key in ctx->flow at the same time. */ + + default: + VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype); + break; + } +} + +static void +do_xlate_actions(const union ofp_action *in, size_t n_in, + struct wx_xlate_ctx *ctx) +{ + struct actions_iterator iter; + const union ofp_action *ia; - const struct wdp_port *port; ++ const struct wx_port *port; + - port = port_array_get(&ctx->wx->ports, ctx->flow.in_port); - if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) && - port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) - ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) { - /* Drop this flow. */ - return; ++ port = wx_port_get(ctx->wx, ctx->flow.in_port); ++ if (port) { ++ const struct ofp_phy_port *opp = &port->wdp_port.opp; ++ if (opp->config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) && ++ opp->config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) ++ ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) { ++ /* Drop this flow. */ ++ return; ++ } + } + + for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) { + uint16_t type = ntohs(ia->type); + union xflow_action *oa; + + switch (type) { + case OFPAT_OUTPUT: + xlate_output_action(ctx, &ia->output); + break; + + case OFPAT_SET_VLAN_VID: + oa = xflow_actions_add(ctx->out, XFLOWAT_SET_DL_TCI); + oa->dl_tci.tci = ia->vlan_vid.vlan_vid & htons(VLAN_VID_MASK); + oa->dl_tci.mask = htons(VLAN_VID_MASK); + ctx->flow.dl_vlan = ia->vlan_vid.vlan_vid; + break; + + case OFPAT_SET_VLAN_PCP: + oa = xflow_actions_add(ctx->out, XFLOWAT_SET_DL_TCI); + oa->dl_tci.tci = htons((ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) + & VLAN_PCP_MASK); + oa->dl_tci.mask = htons(VLAN_PCP_MASK); + + if (ctx->flow.dl_vlan == htons(OFP_VLAN_NONE)) { + ctx->flow.dl_vlan = htons(0); + } + ctx->flow.dl_vlan_pcp = ia->vlan_pcp.vlan_pcp; + break; + + case OFPAT_STRIP_VLAN: + xflow_actions_add(ctx->out, XFLOWAT_STRIP_VLAN); + ctx->flow.dl_vlan = htons(OFP_VLAN_NONE); + ctx->flow.dl_vlan_pcp = 0; + break; + + case OFPAT_SET_DL_SRC: + oa = xflow_actions_add(ctx->out, XFLOWAT_SET_DL_SRC); + memcpy(oa->dl_addr.dl_addr, + ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); + memcpy(ctx->flow.dl_src, + ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); + break; + + case OFPAT_SET_DL_DST: + oa = xflow_actions_add(ctx->out, XFLOWAT_SET_DL_DST); + memcpy(oa->dl_addr.dl_addr, + ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); + memcpy(ctx->flow.dl_dst, + ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); + break; + + case OFPAT_SET_NW_SRC: + oa = xflow_actions_add(ctx->out, XFLOWAT_SET_NW_SRC); + ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; + break; + + case OFPAT_SET_NW_DST: + oa = xflow_actions_add(ctx->out, XFLOWAT_SET_NW_DST); + ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; + break; + + case OFPAT_SET_NW_TOS: + oa = xflow_actions_add(ctx->out, XFLOWAT_SET_NW_TOS); + ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos; + break; + + case OFPAT_SET_TP_SRC: + oa = xflow_actions_add(ctx->out, XFLOWAT_SET_TP_SRC); + ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port; + break; + + case OFPAT_SET_TP_DST: + oa = xflow_actions_add(ctx->out, XFLOWAT_SET_TP_DST); + ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port; + break; + + case OFPAT_ENQUEUE: + xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); + break; + + case OFPAT_VENDOR: + xlate_nicira_action(ctx, (const struct nx_action_header *) ia); + break; + + default: + VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type); + break; + } + } +} + +/* Returns true if 'flow' and 'actions' may be set up as a flow in the kernel. + * This is true most of the time, but we don't allow flows that would prevent + * DHCP replies from being seen by the local port to be set up in the + * kernel. + * + * We only need this, strictly speaking, when in-band control is turned on. */ +static bool +wx_may_set_up(const flow_t *flow, const struct xflow_actions *actions) +{ + if (flow->dl_type == htons(ETH_TYPE_IP) + && flow->nw_proto == IP_TYPE_UDP + && flow->tp_src == htons(DHCP_SERVER_PORT) + && flow->tp_dst == htons(DHCP_CLIENT_PORT)) { + int i; + + for (i = 0; i < actions->n_actions; i++) { + const struct xflow_action_output *oao = &actions->actions[i].output; + if (oao->type == XFLOWAT_OUTPUT && oao->port == XFLOWP_LOCAL) { + return true; + } + } + return false; + } + + return true; +} + +static int +wx_xlate_actions(struct wx *wx, const union ofp_action *in, size_t n_in, + const flow_t *flow, const struct ofpbuf *packet, + tag_type *tags, struct xflow_actions *out, + bool *may_set_up_flow) +{ + tag_type no_tags = 0; + struct wx_xlate_ctx ctx; + COVERAGE_INC(wx_ofp2xflow); + xflow_actions_init(out); + ctx.flow = *flow; + ctx.recurse = 0; + ctx.wx = wx; + ctx.packet = packet; + ctx.out = out; + ctx.tags = tags ? tags : &no_tags; + ctx.may_set_up_flow = true; + ctx.nf_output_iface = NF_OUT_DROP; + do_xlate_actions(in, n_in, &ctx); + remove_pop_action(&ctx); + + if (may_set_up_flow) { + *may_set_up_flow = ctx.may_set_up_flow && wx_may_set_up(flow, out); + } +#if 0 + if (nf_output_iface) { + *nf_output_iface = ctx.nf_output_iface; + } +#endif + if (xflow_actions_overflow(out)) { + COVERAGE_INC(xflow_overflow); + xflow_actions_init(out); + return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY); + } + return 0; +} + +static void +update_used(struct wx *wx) +{ + struct xflow_flow *flows; + size_t n_flows; + size_t i; + int error; + + error = xfif_flow_list_all(wx->xfif, &flows, &n_flows); + if (error) { + return; + } + + for (i = 0; i < n_flows; i++) { + struct xflow_flow *f = &flows[i]; + struct wx_rule *rule; + flow_t flow; + + xflow_key_to_flow(&f->key, &flow); + rule = wx_rule_cast(classifier_find_rule_exactly(&wx->cls, &flow)); + if (!rule || !rule->installed) { + COVERAGE_INC(wx_unexpected_rule); + xfif_flow_del(wx->xfif, f); + continue; + } + + wx_rule_update_time(wx, rule, &f->stats); + wx_rule_account(wx, rule, f->stats.n_bytes); + } + free(flows); +} + +static void +uninstall_idle_flow(struct wx *wx, struct wx_rule *rule) +{ + assert(rule->installed); + assert(!rule->wr.cr.flow.wildcards); + + if (rule->super) { + wx_rule_remove(wx, rule); + } else { + wx_rule_uninstall(wx, rule); + } +} + +static int +expire_rule(struct cls_rule *cls_rule, void *wx_) +{ + struct wx *wx = wx_; + struct wx_rule *rule = wx_rule_cast(cls_rule); + long long int hard_expire, idle_expire, expire, now; + + hard_expire = (rule->wr.hard_timeout + ? rule->wr.created + rule->wr.hard_timeout * 1000 + : LLONG_MAX); + idle_expire = (rule->wr.idle_timeout + && (rule->super || list_is_empty(&rule->list)) + ? rule->used + rule->wr.idle_timeout * 1000 + : LLONG_MAX); + expire = MIN(hard_expire, idle_expire); + + now = time_msec(); + if (now < expire) { + if (rule->installed && now >= rule->used + 5000) { + uninstall_idle_flow(wx, rule); + } else if (!rule->wr.cr.flow.wildcards) { + //XXX active_timeout(wx, rule); + } + + return 0; + } + + COVERAGE_INC(wx_expired); + + /* Update stats. This code will be a no-op if the rule expired + * due to an idle timeout. */ + if (rule->wr.cr.flow.wildcards) { + struct wx_rule *subrule, *next; - LIST_FOR_EACH_SAFE (subrule, next, struct wx_rule, list, &rule->list) { ++ LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) { + wx_rule_remove(wx, subrule); + } + } else { + wx_rule_uninstall(wx, rule); + } + +#if 0 /* XXX */ + if (!wx_rule_is_hidden(rule)) { + send_flow_removed(wx, rule, now, + (now >= hard_expire + ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT)); + } +#endif + wx_rule_remove(wx, rule); + + return 0; +} + +struct revalidate_cbdata { + struct wx *wx; + bool revalidate_all; /* Revalidate all exact-match rules? */ + bool revalidate_subrules; /* Revalidate all exact-match subrules? */ + struct tag_set revalidate_set; /* Set of tags to revalidate. */ +}; + +static bool +revalidate_rule(struct wx *wx, struct wx_rule *rule) +{ + const flow_t *flow = &rule->wr.cr.flow; + + COVERAGE_INC(wx_revalidate_rule); + if (rule->super) { + struct wx_rule *super; + super = wx_rule_cast(classifier_lookup_wild(&wx->cls, flow)); + if (!super) { + wx_rule_remove(wx, rule); + return false; + } else if (super != rule->super) { + COVERAGE_INC(wx_revalidate_moved); + list_remove(&rule->list); + list_push_back(&super->list, &rule->list); + rule->super = super; + rule->wr.hard_timeout = super->wr.hard_timeout; + rule->wr.idle_timeout = super->wr.idle_timeout; + rule->wr.created = super->wr.created; + rule->used = 0; + } + } + + wx_rule_update_actions(wx, rule); + return true; +} + +static int +revalidate_cb(struct cls_rule *sub_, void *cbdata_) +{ + struct wx_rule *sub = wx_rule_cast(sub_); + struct revalidate_cbdata *cbdata = cbdata_; + + if (cbdata->revalidate_all + || (cbdata->revalidate_subrules && sub->super) + || tag_set_intersects(&cbdata->revalidate_set, sub->tags)) { + revalidate_rule(cbdata->wx, sub); + } + return 0; +} + +static void +wx_run_one(struct wx *wx) +{ + if (time_msec() >= wx->next_expiration) { + COVERAGE_INC(wx_expiration); + wx->next_expiration = time_msec() + 1000; + update_used(wx); + + classifier_for_each(&wx->cls, CLS_INC_ALL, expire_rule, wx); + + /* XXX account_checkpoint_cb */ + } + + if (wx->need_revalidate || !tag_set_is_empty(&wx->revalidate_set)) { + struct revalidate_cbdata cbdata; + cbdata.wx = wx; + cbdata.revalidate_all = wx->revalidate_all; + cbdata.revalidate_subrules = wx->need_revalidate; + cbdata.revalidate_set = wx->revalidate_set; + tag_set_init(&wx->revalidate_set); + COVERAGE_INC(wx_revalidate); + classifier_for_each(&wx->cls, CLS_INC_EXACT, revalidate_cb, &cbdata); + wx->need_revalidate = false; + } +} + +static void +wx_run(void) +{ + struct wx *wx; + - LIST_FOR_EACH (wx, struct wx, list_node, &all_wx) { ++ LIST_FOR_EACH (wx, list_node, &all_wx) { + wx_run_one(wx); + } + xf_run(); +} + +static void +wx_wait_one(struct wx *wx) +{ + if (wx->need_revalidate || !tag_set_is_empty(&wx->revalidate_set)) { + poll_immediate_wake(); + } else if (wx->next_expiration != LLONG_MAX) { + poll_timer_wait_until(wx->next_expiration); + } +} + +static void +wx_wait(void) +{ + struct wx *wx; + - LIST_FOR_EACH (wx, struct wx, list_node, &all_wx) { ++ LIST_FOR_EACH (wx, list_node, &all_wx) { + wx_wait_one(wx); + } + xf_wait(); +} + +static int wx_flow_flush(struct wdp *); + +static int +wx_enumerate(const struct wdp_class *wdp_class, struct svec *all_wdps) +{ + struct svec names = SVEC_EMPTY_INITIALIZER; + int error = xf_enumerate_names(wdp_class->type, &names); + svec_move(all_wdps, &names); + return error; +} + +static int +wx_open(const struct wdp_class *wdp_class, const char *name, bool create, + struct wdp **wdpp) +{ + struct xfif *xfif; + int error; + + error = (create + ? xfif_create_and_open(name, wdp_class->type, &xfif) + : xfif_open(name, wdp_class->type, &xfif)); + if (!error) { + struct wx *wx; + + wx = xzalloc(sizeof *wx); + list_push_back(&all_wx, &wx->list_node); + wdp_init(&wx->wdp, wdp_class, name, 0, 0); + wx->xfif = xfif; + classifier_init(&wx->cls); + wx->netdev_monitor = netdev_monitor_create(); - port_array_init(&wx->ports); ++ hmap_init(&wx->ports); + shash_init(&wx->port_by_name); + wx->next_expiration = time_msec() + 1000; + tag_set_init(&wx->revalidate_set); + + wx_port_init(wx); + + wx->ofhooks = &default_ofhooks; + wx->aux = wx; + wx->ml = mac_learning_create(); + + list_init(&wx->ctl_packets); + + *wdpp = &wx->wdp; + } + + return error; +} + +static void +wx_close(struct wdp *wdp) +{ + struct wx *wx = wx_cast(wdp); + + wx_flow_flush(wdp); + xfif_close(wx->xfif); + classifier_destroy(&wx->cls); + netdev_monitor_destroy(wx->netdev_monitor); + list_remove(&wx->list_node); + mac_learning_destroy(wx->ml); ++ hmap_destroy(&wx->ports); ++ shash_destroy(&wx->port_by_name); + free(wx); +} + +static int +wx_get_all_names(const struct wdp *wdp, struct svec *all_names) +{ + struct wx *wx = wx_cast(wdp); + + return xfif_get_all_names(wx->xfif, all_names); +} + +static int +wx_destroy(struct wdp *wdp) +{ + struct wx *wx = wx_cast(wdp); + + return xfif_delete(wx->xfif); +} + +static int +wx_get_features(const struct wdp *wdp, struct ofpbuf **featuresp) +{ + struct wx *wx = wx_cast(wdp); + struct ofp_switch_features *osf; + struct ofpbuf *buf; - unsigned int port_no; - struct wdp_port *port; ++ struct wx_port *port; + + buf = ofpbuf_new(sizeof *osf); + osf = ofpbuf_put_zeros(buf, sizeof *osf); + osf->n_tables = 2; + osf->capabilities = htonl(OFPC_ARP_MATCH_IP); + osf->actions = htonl((1u << OFPAT_OUTPUT) | + (1u << OFPAT_SET_VLAN_VID) | + (1u << OFPAT_SET_VLAN_PCP) | + (1u << OFPAT_STRIP_VLAN) | + (1u << OFPAT_SET_DL_SRC) | + (1u << OFPAT_SET_DL_DST) | + (1u << OFPAT_SET_NW_SRC) | + (1u << OFPAT_SET_NW_DST) | + (1u << OFPAT_SET_NW_TOS) | + (1u << OFPAT_SET_TP_SRC) | + (1u << OFPAT_SET_TP_DST) | + (1u << OFPAT_ENQUEUE)); + - PORT_ARRAY_FOR_EACH (port, &wx->ports, port_no) { - hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp)); ++ HMAP_FOR_EACH (port, hmap_node, &wx->ports) { ++ const struct ofp_phy_port *opp = &port->wdp_port.opp; ++ hton_ofp_phy_port(ofpbuf_put(buf, opp, sizeof *opp)); + } + + *featuresp = buf; + return 0; +} + +static int +count_subrules(struct cls_rule *cls_rule, void *n_subrules_) +{ + struct wx_rule *rule = wx_rule_cast(cls_rule); + int *n_subrules = n_subrules_; + + if (rule->super) { + (*n_subrules)++; + } + return 0; +} + +static int +wx_get_stats(const struct wdp *wdp, struct wdp_stats *stats) +{ + struct wx *wx = wx_cast(wdp); + struct xflow_stats xflow_stats; + int error; + + error = xfif_get_xf_stats(wx->xfif, &xflow_stats); + stats->max_ports = xflow_stats.max_ports; + return error; +} + +static int +wx_get_table_stats(const struct wdp *wdp, struct ofpbuf *stats) +{ + struct wx *wx = wx_cast(wdp); + struct xflow_stats xflow_stats; + struct ofp_table_stats *exact, *wild; + int n_subrules; + + xfif_get_xf_stats(wx->xfif, &xflow_stats); + /* XXX should pass up errors, but there are no appropriate OpenFlow error + * codes. */ + + n_subrules = 0; + classifier_for_each(&wx->cls, CLS_INC_EXACT, count_subrules, &n_subrules); + + exact = ofpbuf_put_zeros(stats, sizeof *exact); + exact->table_id = TABLEID_HASH; + strcpy(exact->name, "exact"); + exact->wildcards = htonl(0); + exact->max_entries = htonl(MIN(WX_MAX_EXACT, xflow_stats.max_capacity)); + exact->active_count = htonl(classifier_count_exact(&wx->cls) - n_subrules); + exact->lookup_count = htonll(xflow_stats.n_hit + xflow_stats.n_missed); + exact->matched_count = htonll(xflow_stats.n_hit); + + wild = ofpbuf_put_zeros(stats, sizeof *exact); + wild->table_id = TABLEID_CLASSIFIER; + strcpy(wild->name, "classifier"); + wild->wildcards = htonl(OVSFW_ALL); + wild->max_entries = htonl(WX_MAX_WILD); + wild->active_count = htonl(classifier_count_wild(&wx->cls)); + wild->lookup_count = htonll(0); /* XXX */ + wild->matched_count = htonll(0); /* XXX */ + + return 0; +} + +static int +wx_get_drop_frags(const struct wdp *wdp, bool *drop_frags) +{ + struct wx *wx = wx_cast(wdp); + + return xfif_get_drop_frags(wx->xfif, drop_frags); +} + +static int +wx_set_drop_frags(struct wdp *wdp, bool drop_frags) +{ + struct wx *wx = wx_cast(wdp); + + return xfif_set_drop_frags(wx->xfif, drop_frags); +} + +static int +wx_port_add(struct wdp *wdp, const char *devname, + bool internal, uint16_t *port_no) +{ + struct wx *wx = wx_cast(wdp); + uint16_t xflow_flags = internal ? XFLOW_PORT_INTERNAL : 0; + return xfif_port_add(wx->xfif, devname, xflow_flags, port_no); +} + +static int +wx_port_del(struct wdp *wdp, uint16_t port_no) +{ + struct wx *wx = wx_cast(wdp); + + return xfif_port_del(wx->xfif, port_no); +} + +static int - wx_answer_port_query(const struct wdp_port *port, struct wdp_port *portp) ++wx_answer_port_query(const struct wx_port *port, struct wdp_port *portp) +{ + if (port) { - wdp_port_copy(portp, port); ++ wdp_port_copy(portp, &port->wdp_port); + return 0; + } else { + return ENOENT; + } +} + +static int +wx_port_query_by_number(const struct wdp *wdp, uint16_t port_no, + struct wdp_port *portp) +{ + struct wx *wx = wx_cast(wdp); - const struct wdp_port *port; ++ struct wx_port *wx_port = wx_port_get(wx, ofp_port_to_xflow_port(port_no)); + - port = port_array_get(&wx->ports, ofp_port_to_xflow_port(port_no)); - return wx_answer_port_query(port, portp); ++ return wx_answer_port_query(wx_port, portp); +} + +static int +wx_port_query_by_name(const struct wdp *wdp, const char *devname, + struct wdp_port *portp) +{ + struct wx *wx = wx_cast(wdp); + + return wx_answer_port_query(shash_find_data(&wx->port_by_name, devname), + portp); +} + +static int +wx_port_set_config(struct wdp *wdp, uint16_t port_no, uint32_t config) +{ + struct wx *wx = wx_cast(wdp); - struct wdp_port *port; ++ struct wx_port *port; ++ struct ofp_phy_port *opp; + uint32_t changes; + - port = port_array_get(&wx->ports, ofp_port_to_xflow_port(port_no)); ++ port = wx_port_get(wx, ofp_port_to_xflow_port(port_no)); + if (!port) { + return ENOENT; + } - changes = config ^ port->opp.config; ++ opp = &port->wdp_port.opp; ++ changes = config ^ opp->config; + + if (changes & OFPPC_PORT_DOWN) { ++ struct netdev *netdev = port->wdp_port.netdev; + int error; ++ + if (config & OFPPC_PORT_DOWN) { - error = netdev_turn_flags_off(port->netdev, NETDEV_UP, true); ++ error = netdev_turn_flags_off(netdev, NETDEV_UP, true); + } else { - error = netdev_turn_flags_on(port->netdev, NETDEV_UP, true); ++ error = netdev_turn_flags_on(netdev, NETDEV_UP, true); + } + if (!error) { - port->opp.config ^= OFPPC_PORT_DOWN; ++ opp->config ^= OFPPC_PORT_DOWN; + } + } + +#define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD) + if (changes & REVALIDATE_BITS) { + COVERAGE_INC(wx_costly_flags); - port->opp.config ^= changes & REVALIDATE_BITS; ++ opp->config ^= changes & REVALIDATE_BITS; + wx->need_revalidate = true; + } +#undef REVALIDATE_BITS + + if (changes & OFPPC_NO_FLOOD) { - port->opp.config ^= OFPPC_NO_FLOOD; ++ opp->config ^= OFPPC_NO_FLOOD; + wx_port_refresh_groups(wx); + } + + if (changes & OFPPC_NO_PACKET_IN) { - port->opp.config ^= OFPPC_NO_PACKET_IN; ++ opp->config ^= OFPPC_NO_PACKET_IN; + } + + return 0; +} + +static int +wx_port_list(const struct wdp *wdp, struct wdp_port **portsp, size_t *n_portsp) +{ + struct wx *wx = wx_cast(wdp); - struct wdp_port *ports, *port; - unsigned int port_no; ++ struct wdp_port *ports; ++ struct wx_port *port; + size_t n_ports, i; + - *n_portsp = n_ports = port_array_count(&wx->ports); ++ *n_portsp = n_ports = hmap_count(&wx->ports); + *portsp = ports = xmalloc(n_ports * sizeof *ports); + i = 0; - PORT_ARRAY_FOR_EACH (port, &wx->ports, port_no) { - wdp_port_copy(&ports[i++], port); ++ HMAP_FOR_EACH (port, hmap_node, &wx->ports) { ++ wdp_port_copy(&ports[i++], &port->wdp_port); + } + assert(i == n_ports); + + return 0; +} + +static int +wx_port_poll(struct wdp *wdp, wdp_port_poll_cb_func *cb, void *aux) +{ + struct wx *wx = wx_cast(wdp); + char *devname; + int retval; + int error; + + retval = 0; + while ((error = xfif_port_poll(wx->xfif, &devname)) != EAGAIN) { + wx_port_process_change(wx, error, devname, cb, aux); + if (error && error != ENOBUFS) { + retval = error; + } + } + while ((error = netdev_monitor_poll(wx->netdev_monitor, + &devname)) != EAGAIN) { + wx_port_process_change(wx, error, devname, cb, aux); + if (error && error != ENOBUFS) { + retval = error; + } + } + return retval; +} + +static int +wx_port_poll_wait(const struct wdp *wdp) +{ + struct wx *wx = wx_cast(wdp); + + xfif_port_poll_wait(wx->xfif); + netdev_monitor_poll_wait(wx->netdev_monitor); + return 0; +} + +static struct wdp_rule * +wx_flow_get(const struct wdp *wdp, const flow_t *flow, unsigned int include) +{ + struct wx *wx = wx_cast(wdp); + struct wx_rule *rule; + int table_id; + + table_id = flow->wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH; + if (!(include & (1u << table_id))) { + return NULL; + } + + rule = wx_rule_cast(classifier_find_rule_exactly(&wx->cls, flow)); + return rule && !wx_rule_is_hidden(rule) ? &rule->wr : NULL; +} + +static struct wdp_rule * +wx_flow_match(const struct wdp *wdp, const flow_t *flow) +{ + struct wx *wx = wx_cast(wdp); + struct wx_rule *rule; + + rule = wx_rule_cast(classifier_lookup(&wx->cls, flow)); + if (rule) { + if (wx_rule_is_hidden(rule)) { + rule = rule->super; + } + return &rule->wr; + } else { + return NULL; + } +} + +struct wx_for_each_thunk_aux { + wdp_flow_cb_func *client_callback; + void *client_aux; +}; + +static int +wx_for_each_thunk(struct cls_rule *cls_rule, void *aux_) +{ + struct wx_for_each_thunk_aux *aux = aux_; + struct wx_rule *rule = wx_rule_cast(cls_rule); + + if (!wx_rule_is_hidden(rule)) { + return aux->client_callback(&rule->wr, aux->client_aux); + } + return 0; +} + +static int +wx_flow_for_each_match(const struct wdp *wdp, const flow_t *target, + unsigned int include, + wdp_flow_cb_func *client_callback, void *client_aux) +{ + struct wx *wx = wx_cast(wdp); + struct wx_for_each_thunk_aux aux; + int cls_include; + + cls_include = 0; + if (include & (1u << TABLEID_HASH)) { + cls_include |= CLS_INC_EXACT; + } + if (include & (1u << TABLEID_CLASSIFIER)) { + cls_include |= CLS_INC_WILD; + } + + aux.client_callback = client_callback; + aux.client_aux = client_aux; + return classifier_for_each_match(&wx->cls, target, cls_include, + wx_for_each_thunk, &aux); +} + +/* Obtains statistic counters for 'rule' within 'wx' and stores them into + * '*stats'. If 'rule' is a wildcarded rule, the returned statistic include + * statistics for all of 'rule''s subrules. */ +static void +query_stats(struct wx *wx, struct wx_rule *rule, struct wdp_flow_stats *stats) +{ + struct wx_rule *subrule; + struct xflow_flow *xflow_flows; + size_t n_xflow_flows; + + /* Start from historical data for 'rule' itself that are no longer tracked + * by the datapath. This counts, for example, subrules that have + * expired. */ + stats->n_packets = rule->packet_count; + stats->n_bytes = rule->byte_count; + stats->inserted = rule->wr.created; + stats->used = LLONG_MIN; + stats->tcp_flags = 0; + stats->ip_tos = 0; + + /* Prepare to ask the datapath for statistics on 'rule', or if it is + * wildcarded then on all of its subrules. + * + * Also, add any statistics that are not tracked by the datapath for each + * subrule. This includes, for example, statistics for packets that were + * executed "by hand" by ofproto via xfif_execute() but must be accounted + * to a flow. */ + n_xflow_flows = rule->wr.cr.flow.wildcards ? list_size(&rule->list) : 1; + xflow_flows = xzalloc(n_xflow_flows * sizeof *xflow_flows); + if (rule->wr.cr.flow.wildcards) { + size_t i = 0; - LIST_FOR_EACH (subrule, struct wx_rule, list, &rule->list) { ++ LIST_FOR_EACH (subrule, list, &rule->list) { + xflow_key_from_flow(&xflow_flows[i++].key, &subrule->wr.cr.flow); + stats->n_packets += subrule->packet_count; + stats->n_bytes += subrule->byte_count; + } + } else { + xflow_key_from_flow(&xflow_flows[0].key, &rule->wr.cr.flow); + } + + /* Fetch up-to-date statistics from the datapath and add them in. */ + if (!xfif_flow_get_multiple(wx->xfif, xflow_flows, n_xflow_flows)) { + size_t i; + for (i = 0; i < n_xflow_flows; i++) { + struct xflow_flow *xflow_flow = &xflow_flows[i]; + long long int used; + + stats->n_packets += xflow_flow->stats.n_packets; + stats->n_bytes += xflow_flow->stats.n_bytes; + used = xflow_flow_stats_to_msec(&xflow_flow->stats); + if (used > stats->used) { + stats->used = used; + } + stats->tcp_flags |= xflow_flow->stats.tcp_flags; + } + } + free(xflow_flows); +} + +static int +wx_flow_get_stats(const struct wdp *wdp, + const struct wdp_rule *wdp_rule, + struct wdp_flow_stats *stats) +{ + struct wx *wx = wx_cast(wdp); + struct wx_rule *rule = wx_rule_cast(&wdp_rule->cr); + + query_stats(wx, rule, stats); + return 0; +} + +static bool +wx_flow_overlaps(const struct wdp *wdp, const flow_t *flow) +{ + struct wx *wx = wx_cast(wdp); + + /* XXX overlap with a subrule? */ + return classifier_rule_overlaps(&wx->cls, flow); +} + +static int +wx_flow_put(struct wdp *wdp, const struct wdp_flow_put *put, + struct wdp_flow_stats *old_stats, struct wdp_rule **rulep) +{ + struct wx *wx = wx_cast(wdp); + struct wx_rule *rule; + uint8_t ofp_table_id; + + ofp_table_id = put->flow->wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH; + if (put->ofp_table_id != 0xff && put->ofp_table_id != ofp_table_id) { + return ofp_mkerr_nicira(OFPET_FLOW_MOD_FAILED, NXFMFC_BAD_TABLE_ID); + } + + rule = wx_rule_cast(classifier_find_rule_exactly(&wx->cls, put->flow)); + if (rule && wx_rule_is_hidden(rule)) { + rule = NULL; + } + + if (rule) { + if (!(put->flags & WDP_PUT_MODIFY)) { + return EEXIST; + } + } else { + if (!(put->flags & WDP_PUT_CREATE)) { + return EINVAL; + } + if ((put->flow->wildcards + ? classifier_count_wild(&wx->cls) >= WX_MAX_WILD + : classifier_count_exact(&wx->cls) >= WX_MAX_EXACT)) { + /* XXX subrules should not count against exact-match limit */ + return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL); + } + } + + rule = wx_rule_create(NULL, put->actions, put->n_actions, + put->idle_timeout, put->hard_timeout); + cls_rule_from_flow(put->flow, &rule->wr.cr); + rule->wr.ofp_table_id = ofp_table_id; + wx_rule_insert(wx, rule, NULL, 0); + + if (old_stats) { + /* XXX */ + memset(old_stats, 0, sizeof *old_stats); + } + if (rulep) { + *rulep = &rule->wr; + } + + return 0; +} + +static int +wx_flow_delete(struct wdp *wdp, struct wdp_rule *wdp_rule, + struct wdp_flow_stats *final_stats) +{ + struct wx *wx = wx_cast(wdp); + struct wx_rule *rule = wx_rule_cast(&wdp_rule->cr); + + wx_rule_remove(wx, rule); + if (final_stats) { + memset(final_stats, 0, sizeof *final_stats); /* XXX */ + } + return 0; +} + +static int +wx_flush_rule(struct cls_rule *cls_rule, void *wx_) +{ + struct wx_rule *rule = wx_rule_cast(cls_rule); + struct wx *wx = wx_; + + /* Mark the flow as not installed, even though it might really be + * installed, so that wx_rule_remove() doesn't bother trying to uninstall + * it. There is no point in uninstalling it individually since we are + * about to blow away all the flows with xfif_flow_flush(). */ + rule->installed = false; + + wx_rule_remove(wx, rule); + + return 0; +} + +static int +wx_flow_flush(struct wdp *wdp) +{ + struct wx *wx = wx_cast(wdp); + + COVERAGE_INC(wx_flow_flush); + classifier_for_each(&wx->cls, CLS_INC_ALL, wx_flush_rule, wx); + xfif_flow_flush(wx->xfif); + return 0; +} + +static int +wx_execute(struct wdp *wdp, uint16_t in_port, + const union ofp_action actions[], int n_actions, + const struct ofpbuf *packet) +{ + struct wx *wx = wx_cast(wdp); + struct xflow_actions xflow_actions; + flow_t flow; + int error; + + flow_extract((struct ofpbuf *) packet, 0, in_port, &flow); + error = wx_xlate_actions(wx, actions, n_actions, &flow, packet, + NULL, &xflow_actions, NULL); + if (error) { + return error; + } + return xfif_execute(wx->xfif, ofp_port_to_xflow_port(in_port), + xflow_actions.actions, xflow_actions.n_actions, + packet); +} + +static int +wx_flow_inject(struct wdp *wdp, struct wdp_rule *wdp_rule, + uint16_t in_port, const struct ofpbuf *packet) +{ + struct wx_rule *rule = wx_rule_cast(&wdp_rule->cr); + int error; + + error = wx_execute(wdp, in_port, rule->wr.actions, rule->wr.n_actions, + packet); + if (!error) { + rule->packet_count++; + rule->byte_count += packet->size; + rule->used = time_msec(); + } + return error; +} + +static int +wx_recv_get_mask(const struct wdp *wdp, int *listen_mask) +{ + struct wx *wx = wx_cast(wdp); + int xflow_listen_mask; + int error; + + error = xfif_recv_get_mask(wx->xfif, &xflow_listen_mask); + if (!error) { + *listen_mask = 0; + if (xflow_listen_mask & XFLOWL_MISS) { + *listen_mask |= 1 << WDP_CHAN_MISS; + } + if (xflow_listen_mask & XFLOWL_ACTION) { + *listen_mask |= 1 << WDP_CHAN_ACTION; + } + if (xflow_listen_mask & XFLOWL_SFLOW) { + *listen_mask |= 1 << WDP_CHAN_SFLOW; + } + } + return error; +} + +static int +wx_recv_set_mask(struct wdp *wdp, int listen_mask) +{ + struct wx *wx = wx_cast(wdp); + int xflow_listen_mask; + + wx->wdp_listen_mask = listen_mask; + + xflow_listen_mask = 0; + if (listen_mask & (1 << WDP_CHAN_MISS)) { + xflow_listen_mask |= XFLOWL_MISS; + } + if (listen_mask & (1 << WDP_CHAN_ACTION)) { + xflow_listen_mask |= XFLOWL_ACTION; + } else { + wx_purge_ctl_packets__(wx); + } + if (listen_mask & (1 << WDP_CHAN_SFLOW)) { + xflow_listen_mask |= XFLOWL_SFLOW; + } + + return xfif_recv_set_mask(wx->xfif, xflow_listen_mask); +} + +static int +wx_get_sflow_probability(const struct wdp *wdp, uint32_t *probability) +{ + struct wx *wx = wx_cast(wdp); + + return xfif_get_sflow_probability(wx->xfif, probability); +} + +static int +wx_set_sflow_probability(struct wdp *wdp, uint32_t probability) +{ + struct wx *wx = wx_cast(wdp); + + return xfif_set_sflow_probability(wx->xfif, probability); +} + +static int +wx_translate_xflow_msg(struct xflow_msg *msg, struct ofpbuf *payload, + struct wdp_packet *packet) +{ + packet->in_port = xflow_port_to_ofp_port(msg->port); + packet->send_len = 0; + packet->tun_id = 0; + + switch (msg->type) { + case _XFLOWL_MISS_NR: + packet->channel = WDP_CHAN_MISS; + packet->payload = payload; + packet->tun_id = msg->arg; + return 0; + + case _XFLOWL_ACTION_NR: + packet->channel = WDP_CHAN_ACTION; + packet->payload = payload; + packet->send_len = msg->arg; + return 0; + + case _XFLOWL_SFLOW_NR: + /* XXX */ + ofpbuf_delete(payload); + return ENOSYS; + + default: + VLOG_WARN_RL(&rl, "received XFLOW message of unexpected type %"PRIu32, + msg->type); + ofpbuf_delete(payload); + return ENOSYS; + } +} + +static const uint8_t * +get_local_mac(const struct wx *wx) +{ - const struct wdp_port *port = port_array_get(&wx->ports, XFLOWP_LOCAL); - return port ? port->opp.hw_addr : NULL; ++ const struct wx_port *port = wx_port_get(wx, XFLOWP_LOCAL); ++ return port ? port->wdp_port.opp.hw_addr : NULL; +} + +/* Returns true if 'packet' is a DHCP reply to the local port. Such a reply + * should be sent to the local port regardless of the flow table. + * + * We only need this, strictly speaking, when in-band control is turned on. */ +static bool +wx_is_local_dhcp_reply(const struct wx *wx, + const flow_t *flow, const struct ofpbuf *packet) +{ + if (flow->dl_type == htons(ETH_TYPE_IP) + && flow->nw_proto == IP_TYPE_UDP + && flow->tp_src == htons(DHCP_SERVER_PORT) + && flow->tp_dst == htons(DHCP_CLIENT_PORT) + && packet->l7) + { + const uint8_t *local_mac = get_local_mac(wx); + struct dhcp_header *dhcp = ofpbuf_at( + packet, (char *)packet->l7 - (char *)packet->data, sizeof *dhcp); + return dhcp && local_mac && eth_addr_equals(dhcp->chaddr, local_mac); + } + + return false; +} + +/* Determines whether 'payload' that arrived on 'in_port' is included in any of + * the flows in 'wx''s OpenFlow flow table. If so, then it adds a + * corresponding flow to the xfif's exact-match flow table, taking ownership of + * 'payload', and returns true. If not, it returns false and the caller + * retains ownership of 'payload'. */ +static bool +wx_explode_rule(struct wx *wx, uint16_t in_port, struct ofpbuf *payload) +{ + struct wx_rule *rule; + flow_t flow; + + flow_extract(payload, 0, xflow_port_to_ofp_port(in_port), &flow); + + if (wx_is_local_dhcp_reply(wx, &flow, payload)) { + union xflow_action action; + + memset(&action, 0, sizeof(action)); + action.output.type = XFLOWAT_OUTPUT; + action.output.port = XFLOWP_LOCAL; + xfif_execute(wx->xfif, in_port, &action, 1, payload); + } + + rule = wx_rule_lookup_valid(wx, &flow); + if (!rule) { + return false; + } + + if (rule->wr.cr.flow.wildcards) { + rule = wx_rule_create_subrule(wx, rule, &flow); + wx_rule_make_actions(wx, rule, payload); + } else { + if (!rule->may_install) { + /* The rule is not installable, that is, we need to process every + * packet, so process the current packet and set its actions into + * 'subrule'. */ + wx_rule_make_actions(wx, rule, payload); + } else { + /* XXX revalidate rule if it needs it */ + } + } + + wx_rule_execute(wx, rule, payload, &flow); + wx_rule_reinstall(wx, rule); + + return true; +} + +static int +wx_recv(struct wdp *wdp, struct wdp_packet *packet) +{ + struct wx *wx = wx_cast(wdp); + int i; + + if (wx->n_ctl_packets) { + struct wdp_packet *wdp_packet; + + wdp_packet = CONTAINER_OF(list_pop_front(&wx->ctl_packets), + struct wdp_packet, list); + wx->n_ctl_packets--; + + *packet = *wdp_packet; + free(wdp_packet); + + return 0; + } + + /* XXX need to avoid 50*50 potential cost for caller. */ + for (i = 0; i < 50; i++) { + struct xflow_msg *msg; + struct ofpbuf *buf; + int error; + + error = xfif_recv(wx->xfif, &buf); + if (error) { + return error; + } + + msg = ofpbuf_pull(buf, sizeof *msg); + if (msg->type != _XFLOWL_MISS_NR + || !wx_explode_rule(wx, msg->port, buf)) { + return wx_translate_xflow_msg(msg, buf, packet); + } + } + return EAGAIN; +} + +static void +wx_recv_purge_queue__(struct wx *wx, int max, int xflow_listen_mask, + int *errorp) +{ + int error; + + error = xfif_recv_set_mask(wx->xfif, xflow_listen_mask); + if (!error) { + struct ofpbuf *buf; + + while (max > 0 && (error = xfif_recv(wx->xfif, &buf)) == 0) { + ofpbuf_delete(buf); + max--; + } + } + if (error && error != EAGAIN) { + *errorp = error; + } +} + +static void +wx_purge_ctl_packets__(struct wx *wx) +{ + struct wdp_packet *this, *next; + - LIST_FOR_EACH_SAFE (this, next, struct wdp_packet, list, - &wx->ctl_packets) { ++ LIST_FOR_EACH_SAFE (this, next, list, &wx->ctl_packets) { + list_remove(&this->list); + ofpbuf_delete(this->payload); + free(this); + } + wx->n_ctl_packets = 0; +} + +static int +wx_recv_purge(struct wdp *wdp) +{ + struct wx *wx = wx_cast(wdp); + struct xflow_stats xflow_stats; + int xflow_listen_mask; + int retval, error; + + xfif_get_xf_stats(wx->xfif, &xflow_stats); + + error = xfif_recv_get_mask(wx->xfif, &xflow_listen_mask); + if (error || !(xflow_listen_mask & XFLOWL_ALL)) { + return error; + } + + if (xflow_listen_mask & XFLOWL_MISS) { + wx_recv_purge_queue__(wx, xflow_stats.max_miss_queue, XFLOWL_MISS, + &error); + } + if (xflow_listen_mask & XFLOWL_ACTION) { + wx_recv_purge_queue__(wx, xflow_stats.max_action_queue, XFLOWL_ACTION, + &error); + wx_purge_ctl_packets__(wx); + } + if (xflow_listen_mask & XFLOWL_SFLOW) { + wx_recv_purge_queue__(wx, xflow_stats.max_sflow_queue, XFLOWL_SFLOW, + &error); + } + + retval = xfif_recv_set_mask(wx->xfif, xflow_listen_mask); + return retval ? retval : error; +} + + +static void +wx_recv_wait(struct wdp *wdp) +{ + struct wx *wx = wx_cast(wdp); + + if (wx->n_ctl_packets) { + poll_immediate_wake(); + } else { + xfif_recv_wait(wx->xfif); + } +} + +static int +wx_set_ofhooks(struct wdp *wdp, const struct ofhooks *ofhooks, void *aux) +{ + struct wx *wx = wx_cast(wdp); + + if (wx->ofhooks == &default_ofhooks) { + mac_learning_destroy(wx->ml); + wx->ml = NULL; + } + + wx->ofhooks = ofhooks; + wx->aux = aux; + return 0; +} + +static void +wx_revalidate(struct wdp *wdp, tag_type tag) +{ + struct wx *wx = wx_cast(wdp); + + tag_set_add(&wx->revalidate_set, tag); +} + +static void +wx_revalidate_all(struct wdp *wdp) +{ + struct wx *wx = wx_cast(wdp); + + wx->revalidate_all = true; +} + +static void wx_port_update(struct wx *, const char *devname, + wdp_port_poll_cb_func *cb, void *aux); +static void wx_port_reinit(struct wx *, wdp_port_poll_cb_func *cb, void *aux); + +static void +wx_port_process_change(struct wx *wx, int error, char *devname, + wdp_port_poll_cb_func *cb, void *aux) +{ + if (error == ENOBUFS) { + wx_port_reinit(wx, cb, aux); + } else if (!error) { + wx_port_update(wx, devname, cb, aux); + free(devname); + } +} + +static size_t +wx_port_refresh_group(struct wx *wx, unsigned int group) +{ + uint16_t *ports; + size_t n_ports; - struct wdp_port *port; - unsigned int port_no; ++ struct wx_port *port; + + assert(group == WX_GROUP_ALL || group == WX_GROUP_FLOOD); + - ports = xmalloc(port_array_count(&wx->ports) * sizeof *ports); ++ ports = xmalloc(hmap_count(&wx->ports) * sizeof *ports); + n_ports = 0; - PORT_ARRAY_FOR_EACH (port, &wx->ports, port_no) { - if (group == WX_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) { - ports[n_ports++] = port_no; ++ HMAP_FOR_EACH (port, hmap_node, &wx->ports) { ++ const struct ofp_phy_port *opp = &port->wdp_port.opp; ++ if (group == WX_GROUP_ALL || !(opp->config & OFPPC_NO_FLOOD)) { ++ ports[n_ports++] = port->xflow_port; + } + } + xfif_port_group_set(wx->xfif, group, ports, n_ports); + free(ports); + + return n_ports; +} + +static void +wx_port_refresh_groups(struct wx *wx) +{ + wx_port_refresh_group(wx, WX_GROUP_FLOOD); + wx_port_refresh_group(wx, WX_GROUP_ALL); +} + +static void +wx_port_reinit(struct wx *wx, wdp_port_poll_cb_func *cb, void *aux) +{ + struct svec devnames; - struct wdp_port *wdp_port; - unsigned int port_no; ++ struct wx_port *wx_port; + struct xflow_port *xflow_ports; + size_t n_xflow_ports; + size_t i; + + svec_init(&devnames); - PORT_ARRAY_FOR_EACH (wdp_port, &wx->ports, port_no) { - svec_add (&devnames, (char *) wdp_port->opp.name); ++ HMAP_FOR_EACH (wx_port, hmap_node, &wx->ports) { ++ svec_add (&devnames, (char *) wx_port->wdp_port.opp.name); + } + xfif_port_list(wx->xfif, &xflow_ports, &n_xflow_ports); + for (i = 0; i < n_xflow_ports; i++) { + svec_add(&devnames, xflow_ports[i].devname); + } + free(xflow_ports); + + svec_sort_unique(&devnames); + for (i = 0; i < devnames.n; i++) { + wx_port_update(wx, devnames.names[i], cb, aux); + } + svec_destroy(&devnames); + + wx_port_refresh_groups(wx); +} + - static struct wdp_port * - make_wdp_port(const struct xflow_port *xflow_port) ++static struct wx_port * ++make_wx_port(const struct xflow_port *xflow_port) +{ + struct netdev_options netdev_options; + enum netdev_flags flags; ++ struct wx_port *wx_port; + struct wdp_port *wdp_port; + struct netdev *netdev; + bool carrier; + int error; + + memset(&netdev_options, 0, sizeof netdev_options); + netdev_options.name = xflow_port->devname; + netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; + + error = netdev_open(&netdev_options, &netdev); + if (error) { + VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s " + "cannot be opened (%s)", + xflow_port->devname, xflow_port->port, + xflow_port->devname, strerror(error)); + return NULL; + } + - wdp_port = xmalloc(sizeof *wdp_port); ++ wx_port = xmalloc(sizeof *wx_port); ++ wx_port->xflow_port = xflow_port->port; ++ wdp_port = &wx_port->wdp_port; + wdp_port->netdev = netdev; + wdp_port->opp.port_no = xflow_port_to_ofp_port(xflow_port->port); + netdev_get_etheraddr(netdev, wdp_port->opp.hw_addr); + strncpy((char *) wdp_port->opp.name, xflow_port->devname, + sizeof wdp_port->opp.name); + wdp_port->opp.name[sizeof wdp_port->opp.name - 1] = '\0'; + + netdev_get_flags(netdev, &flags); + wdp_port->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN; + + netdev_get_carrier(netdev, &carrier); + wdp_port->opp.state = carrier ? 0 : OFPPS_LINK_DOWN; + + netdev_get_features(netdev, + &wdp_port->opp.curr, &wdp_port->opp.advertised, + &wdp_port->opp.supported, &wdp_port->opp.peer); + + wdp_port->devname = xstrdup(xflow_port->devname); + wdp_port->internal = (xflow_port->flags & XFLOW_PORT_INTERNAL) != 0; - return wdp_port; ++ return wx_port; +} + +static bool +wx_port_conflicts(const struct wx *wx, const struct xflow_port *xflow_port) +{ - if (port_array_get(&wx->ports, xflow_port->port)) { ++ if (wx_port_get(wx, xflow_port->port)) { + VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath", + xflow_port->port); + return true; + } else if (shash_find(&wx->port_by_name, xflow_port->devname)) { + VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath", + xflow_port->devname); + return true; + } else { + return false; + } +} + +static int - wdp_port_equal(const struct wdp_port *a_, const struct wdp_port *b_) ++wx_port_equal(const struct wx_port *a_, const struct wx_port *b_) +{ - const struct ofp_phy_port *a = &a_->opp; - const struct ofp_phy_port *b = &b_->opp; ++ const struct ofp_phy_port *a = &a_->wdp_port.opp; ++ const struct ofp_phy_port *b = &b_->wdp_port.opp; + + BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */ + return (a->port_no == b->port_no + && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) + && !strcmp((char *) a->name, (char *) b->name) + && a->state == b->state + && a->config == b->config + && a->curr == b->curr + && a->advertised == b->advertised + && a->supported == b->supported + && a->peer == b->peer); +} + +static void - wx_port_install(struct wx *wx, struct wdp_port *wdp_port) ++wx_port_install(struct wx *wx, struct wx_port *wx_port) +{ - uint16_t xflow_port = ofp_port_to_xflow_port(wdp_port->opp.port_no); - const char *netdev_name = (const char *) wdp_port->opp.name; ++ const struct ofp_phy_port *opp = &wx_port->wdp_port.opp; ++ uint16_t xflow_port = ofp_port_to_xflow_port(opp->port_no); ++ const char *name = (const char *) opp->name; + - netdev_monitor_add(wx->netdev_monitor, wdp_port->netdev); - port_array_set(&wx->ports, xflow_port, wdp_port); - shash_add(&wx->port_by_name, netdev_name, wdp_port); ++ netdev_monitor_add(wx->netdev_monitor, wx_port->wdp_port.netdev); ++ hmap_insert(&wx->ports, &wx_port->hmap_node, hash_int(xflow_port, 0)); ++ shash_add(&wx->port_by_name, name, wx_port); +} + +static void - wx_port_remove(struct wx *wx, struct wdp_port *wdp_port) ++wx_port_remove(struct wx *wx, struct wx_port *wx_port) +{ - uint16_t xflow_port = ofp_port_to_xflow_port(wdp_port->opp.port_no); ++ const struct ofp_phy_port *opp = &wx_port->wdp_port.opp; ++ const char *name = (const char *) opp->name; + - netdev_monitor_remove(wx->netdev_monitor, wdp_port->netdev); - port_array_delete(&wx->ports, xflow_port); - shash_delete(&wx->port_by_name, - shash_find(&wx->port_by_name, (char *) wdp_port->opp.name)); ++ netdev_monitor_remove(wx->netdev_monitor, wx_port->wdp_port.netdev); ++ hmap_remove(&wx->ports, &wx_port->hmap_node); ++ shash_delete(&wx->port_by_name, shash_find(&wx->port_by_name, name)); +} + +static void - wx_port_free(struct wdp_port *wdp_port) ++wx_port_free(struct wx_port *wx_port) +{ - wdp_port_free(wdp_port); - free(wdp_port); ++ if (wx_port) { ++ wdp_port_free(&wx_port->wdp_port); ++ free(wx_port); ++ } +} + +static void +wx_port_update(struct wx *wx, const char *devname, + wdp_port_poll_cb_func *cb, void *aux) +{ + struct xflow_port xflow_port; - struct wdp_port *old_wdp_port; - struct wdp_port *new_wdp_port; ++ struct wx_port *old_wx_port; ++ struct wx_port *new_wx_port; + int error; + + COVERAGE_INC(wx_update_port); + + /* Query the datapath for port information. */ + error = xfif_port_query_by_name(wx->xfif, devname, &xflow_port); + - /* Find the old wdp_port. */ - old_wdp_port = shash_find_data(&wx->port_by_name, devname); ++ /* Find the old wx_port. */ ++ old_wx_port = shash_find_data(&wx->port_by_name, devname); + if (!error) { - if (!old_wdp_port) { ++ if (!old_wx_port) { + /* There's no port named 'devname' but there might be a port with + * the same port number. This could happen if a port is deleted + * and then a new one added in its place very quickly, or if a port + * is renamed. In the former case we want to send an OFPPR_DELETE + * and an OFPPR_ADD, and in the latter case we want to send a + * single OFPPR_MODIFY. We can distinguish the cases by comparing + * the old port's ifindex against the new port, or perhaps less + * reliably but more portably by comparing the old port's MAC + * against the new port's MAC. However, this code isn't that smart + * and always sends an OFPPR_MODIFY (XXX). */ - old_wdp_port = port_array_get(&wx->ports, xflow_port.port); ++ old_wx_port = wx_port_get(wx, xflow_port.port); + } + } else if (error != ENOENT && error != ENODEV) { + VLOG_WARN_RL(&rl, "xfif_port_query_by_name returned unexpected error " + "%s", strerror(error)); + return; + } + - /* Create a new wdp_port. */ - new_wdp_port = !error ? make_wdp_port(&xflow_port) : NULL; ++ /* Create a new wx_port. */ ++ new_wx_port = !error ? make_wx_port(&xflow_port) : NULL; + + /* Eliminate a few pathological cases. */ - if (!old_wdp_port && !new_wdp_port) { ++ if (!old_wx_port && !new_wx_port) { + return; - } else if (old_wdp_port && new_wdp_port) { ++ } else if (old_wx_port && new_wx_port) { + /* Most of the 'config' bits are OpenFlow soft state, but + * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the - * OpenFlow bits from old_wdp_port. (make_wdp_port() only sets ++ * OpenFlow bits from old_wx_port. (make_wx_port() only sets + * OFPPC_PORT_DOWN and leaves the other bits 0.) */ - new_wdp_port->opp.config |= old_wdp_port->opp.config & ~OFPPC_PORT_DOWN; ++ struct ofp_phy_port *new_opp = &new_wx_port->wdp_port.opp; ++ struct ofp_phy_port *old_opp = &old_wx_port->wdp_port.opp; ++ new_opp->config |= old_opp->config & ~OFPPC_PORT_DOWN; + - if (wdp_port_equal(old_wdp_port, new_wdp_port)) { ++ if (wx_port_equal(old_wx_port, new_wx_port)) { + /* False alarm--no change. */ - wx_port_free(new_wdp_port); ++ wx_port_free(new_wx_port); + return; + } + } + + /* Now deal with the normal cases. */ - if (old_wdp_port) { - wx_port_remove(wx, old_wdp_port); ++ if (old_wx_port) { ++ wx_port_remove(wx, old_wx_port); + } - if (new_wdp_port) { - wx_port_install(wx, new_wdp_port); ++ if (new_wx_port) { ++ wx_port_install(wx, new_wx_port); + } + + /* Call back. */ - if (!old_wdp_port) { - (*cb)(&new_wdp_port->opp, OFPPR_ADD, aux); - } else if (!new_wdp_port) { - (*cb)(&old_wdp_port->opp, OFPPR_DELETE, aux); ++ if (!old_wx_port) { ++ (*cb)(&new_wx_port->wdp_port.opp, OFPPR_ADD, aux); ++ } else if (!new_wx_port) { ++ (*cb)(&old_wx_port->wdp_port.opp, OFPPR_DELETE, aux); + } else { - (*cb)(&new_wdp_port->opp, OFPPR_MODIFY, aux); ++ (*cb)(&new_wx_port->wdp_port.opp, OFPPR_MODIFY, aux); + } + + /* Update port groups. */ + wx_port_refresh_groups(wx); + + /* Clean up. */ - wx_port_free(old_wdp_port); ++ wx_port_free(old_wx_port); +} + +static int +wx_port_init(struct wx *wx) +{ + struct xflow_port *ports; + size_t n_ports; + size_t i; + int error; + + error = xfif_port_list(wx->xfif, &ports, &n_ports); + if (error) { + return error; + } + + for (i = 0; i < n_ports; i++) { + const struct xflow_port *xflow_port = &ports[i]; + if (!wx_port_conflicts(wx, xflow_port)) { - struct wdp_port *wdp_port = make_wdp_port(xflow_port); - if (wdp_port) { - wx_port_install(wx, wdp_port); ++ struct wx_port *wx_port = make_wx_port(xflow_port); ++ if (wx_port) { ++ wx_port_install(wx, wx_port); + } + } + } + free(ports); + wx_port_refresh_groups(wx); + return 0; +} ++ ++/* Returns the port in 'wx' with xflow port number 'xflow_port'. */ ++static struct wx_port * ++wx_port_get(const struct wx *wx, uint16_t xflow_port) ++{ ++ struct wx_port *port; ++ ++ HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_int(xflow_port, 0), ++ &wx->ports) { ++ if (port->xflow_port == xflow_port) { ++ return port; ++ } ++ } ++ return NULL; ++} + +void +wdp_xflow_register(void) +{ + static const struct wdp_class wdp_xflow_class = { + NULL, /* name */ + wx_run, + wx_wait, + wx_enumerate, + wx_open, + wx_close, + wx_get_all_names, + wx_destroy, + wx_get_features, + wx_get_stats, + wx_get_table_stats, + wx_get_drop_frags, + wx_set_drop_frags, + wx_port_add, + wx_port_del, + wx_port_query_by_number, + wx_port_query_by_name, + wx_port_list, + wx_port_set_config, + wx_port_poll, + wx_port_poll_wait, + wx_flow_get, + wx_flow_match, + wx_flow_for_each_match, + wx_flow_get_stats, + wx_flow_overlaps, + wx_flow_put, + wx_flow_delete, + wx_flow_flush, + wx_flow_inject, + wx_execute, + wx_recv_get_mask, + wx_recv_set_mask, + wx_get_sflow_probability, + wx_set_sflow_probability, + wx_recv, + wx_recv_purge, + wx_recv_wait, + wx_set_ofhooks, + wx_revalidate, + wx_revalidate_all, + }; + + static bool inited = false; + + struct svec types; + const char *type; + bool registered; + int i; + + if (inited) { + return; + } + inited = true; + + svec_init(&types); + xf_enumerate_types(&types); + + registered = false; + SVEC_FOR_EACH (i, type, &types) { + struct wdp_class *class; + + class = xmalloc(sizeof *class); + *class = wdp_xflow_class; + class->type = xstrdup(type); + if (registered) { + class->run = NULL; + class->wait = NULL; + } + if (!wdp_register_provider(class)) { + registered = true; + } + } + + svec_destroy(&types); +} + +static bool +default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, + struct xflow_actions *actions, tag_type *tags, + uint16_t *nf_output_iface, void *wx_) +{ + struct wx *wx = wx_; + int out_port; + + /* Drop frames for reserved multicast addresses. */ + if (eth_addr_is_reserved(flow->dl_dst)) { + return true; + } + + /* Learn source MAC (but don't try to learn from revalidation). */ + if (packet != NULL) { + tag_type rev_tag = mac_learning_learn(wx->ml, flow->dl_src, + 0, flow->in_port, + GRAT_ARP_LOCK_NONE); + if (rev_tag) { + /* The log messages here could actually be useful in debugging, + * so keep the rate limit relatively high. */ + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); + VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16, + ETH_ADDR_ARGS(flow->dl_src), flow->in_port); + tag_set_add(&wx->revalidate_set, rev_tag); + } + } + + /* Determine output port. */ + out_port = mac_learning_lookup_tag(wx->ml, flow->dl_dst, 0, tags, + NULL); + if (out_port < 0) { + add_output_group_action(actions, WX_GROUP_FLOOD, nf_output_iface); + } else if (out_port != flow->in_port) { + xflow_actions_add(actions, XFLOWAT_OUTPUT)->output.port = out_port; + *nf_output_iface = out_port; + } else { + /* Drop. */ + } + + return true; +} + +static const struct ofhooks default_ofhooks = { + NULL, + default_normal_ofhook_cb, + NULL, + NULL +}; diff --cc utilities/ovs-controller.c index b18959ad6,26a1fc3f5..e72b74da3 --- a/utilities/ovs-controller.c +++ b/utilities/ovs-controller.c @@@ -28,6 -28,7 +28,8 @@@ #include "compiler.h" #include "daemon.h" #include "learning-switch.h" + #include "ofp-parse.h" ++#include "ofp-util.h" #include "ofpbuf.h" #include "openflow/openflow.h" #include "poll-loop.h" @@@ -254,6 -257,45 +258,52 @@@ do_switching(struct switch_ *sw : EAGAIN); } + static void + read_flow_file(const char *name) + { ++ bool table_id_enabled = false; ++ uint8_t table_idx; + struct ofpbuf *b; + FILE *stream; + + stream = fopen(optarg, "r"); + if (!stream) { + ovs_fatal(errno, "%s: open", name); + } + - while ((b = parse_ofp_add_flow_file(stream)) != NULL) { ++ while ((b = parse_ofp_add_flow_file(stream, &table_idx)) != NULL) { ++ if ((table_idx != 0xff) != table_id_enabled) { ++ table_id_enabled = table_idx != 0xff; ++ queue_push_tail(&default_flows, ++ make_nxt_flow_mod_table_id(table_id_enabled)); ++ } + queue_push_tail(&default_flows, b); + } + + fclose(stream); + } + + static void + add_port_queue(char *s) + { + char *save_ptr = NULL; + char *port_name; + char *queue_id; + + port_name = strtok_r(s, ":", &save_ptr); + queue_id = strtok_r(NULL, "", &save_ptr); + if (!queue_id) { + ovs_fatal(0, "argument to -Q or --port-queue should take the form " + "\":\""); + } + + if (!shash_add_once(&port_queues, port_name, + (void *) (uintptr_t) atoi(queue_id))) { + ovs_fatal(0, " arguments for -Q or --port-queue must " + "be unique"); + } + } + static void parse_options(int argc, char *argv[]) { diff --cc utilities/ovs-ofctl.c index 60c63c79a,13f583e7e..aba8f8487 --- a/utilities/ovs-ofctl.c +++ b/utilities/ovs-ofctl.c @@@ -466,19 -465,28 +467,34 @@@ do_dump_aggregate(int argc, char *argv[ dump_stats_transaction(argv[1], request); } +static void - enable_flow_mod_table_id_ext(struct vconn *vconn, uint8_t enable) ++enable_flow_mod_table_id_ext(struct vconn *vconn, bool enable) +{ - struct nxt_flow_mod_table_id *flow_mod_table_id; - struct ofpbuf *buffer; ++ send_openflow_buffer(vconn, make_nxt_flow_mod_table_id(enable)); ++} + - flow_mod_table_id = make_openflow(sizeof *flow_mod_table_id, OFPT_VENDOR, &buffer); + static void + do_queue_stats(int argc, char *argv[]) + { + struct ofp_queue_stats_request *req; + struct ofpbuf *request; - flow_mod_table_id->vendor = htonl(NX_VENDOR_ID); - flow_mod_table_id->subtype = htonl(NXT_FLOW_MOD_TABLE_ID); - flow_mod_table_id->set = enable; + req = alloc_stats_request(sizeof *req, OFPST_QUEUE, &request); - send_openflow_buffer(vconn, buffer); + if (argc > 2 && argv[2][0] && strcasecmp(argv[2], "all")) { + req->port_no = htons(str_to_port_no(argv[1], argv[2])); + } else { + req->port_no = htons(OFPP_ALL); + } + if (argc > 3 && argv[3][0] && strcasecmp(argv[3], "all")) { + req->queue_id = htonl(atoi(argv[3])); + } else { + req->queue_id = htonl(OFPQ_ALL); + } + + memset(req->pad, 0, sizeof req->pad); + + dump_stats_transaction(argv[1], request); } static void @@@ -521,10 -524,8 +537,10 @@@ static voi do_add_flows(int argc OVS_UNUSED, char *argv[]) { struct vconn *vconn; + struct ofpbuf *b; FILE *file; - char line[1024]; ++ bool table_id_enabled = false; + uint8_t table_idx; - int table_id_enabled = 0; file = fopen(argv[2], "r"); if (file == NULL) { @@@ -532,54 -533,8 +548,12 @@@ } open_vconn(argv[1], &vconn); - while (fgets(line, sizeof line, file)) { - struct ofpbuf *buffer; - struct ofp_flow_mod *ofm; - uint16_t priority, idle_timeout, hard_timeout; - uint64_t cookie; - struct ofp_match match; - - char *comment; - - /* Delete comments. */ - comment = strchr(line, '#'); - if (comment) { - *comment = '\0'; - } - - /* Drop empty lines. */ - if (line[strspn(line, " \t\n")] == '\0') { - continue; - } - - /* Parse and send. parse_ofp_str() will expand and reallocate - * the data in 'buffer', so we can't keep pointers to across the - * parse_ofp_str() call. */ - make_openflow(sizeof *ofm, OFPT_FLOW_MOD, &buffer); - parse_ofp_str(line, &match, buffer, &table_idx, NULL, &priority, - &idle_timeout, &hard_timeout, &cookie); - ofm = buffer->data; - ofm->match = match; - ofm->command = htons(OFPFC_ADD); - ofm->cookie = htonll(cookie); - ofm->idle_timeout = htons(idle_timeout); - ofm->hard_timeout = htons(hard_timeout); - ofm->buffer_id = htonl(UINT32_MAX); - ofm->priority = htons(priority); - - if (table_idx != 0xff) { - if (!table_id_enabled) { - enable_flow_mod_table_id_ext(vconn, 1); - table_id_enabled = 1; - } - ofm->command = htons(ntohs(ofm->command) | (table_idx << 8)); - } else { - if (table_id_enabled) { - enable_flow_mod_table_id_ext(vconn, 0); - table_id_enabled = 0; - } - while ((b = parse_ofp_add_flow_file(file)) != NULL) { ++ while ((b = parse_ofp_add_flow_file(file, &table_idx)) != NULL) { ++ if ((table_idx != 0xff) != table_id_enabled) { ++ table_id_enabled = table_idx != 0xff; ++ enable_flow_mod_table_id_ext(vconn, table_id_enabled); + } - send_openflow_buffer(vconn, buffer); + send_openflow_buffer(vconn, b); } vconn_close(vconn); fclose(file); @@@ -914,6 -859,26 +888,27 @@@ do_benchmark(int argc OVS_UNUSED, char count * message_size / (duration / 1000.0)); } + /* This command is really only useful for testing the flow parser (ofp_parse), + * so it is undocumented. */ + static void + do_parse_flows(int argc OVS_UNUSED, char *argv[]) + { ++ uint8_t table_idx; + struct ofpbuf *b; + FILE *file; + + file = fopen(argv[1], "r"); + if (file == NULL) { + ovs_fatal(errno, "%s: open", argv[2]); + } + - while ((b = parse_ofp_add_flow_file(file)) != NULL) { ++ while ((b = parse_ofp_add_flow_file(file, &table_idx)) != NULL) { + ofp_print(stdout, b->data, b->size, 0); + ofpbuf_delete(b); + } + fclose(file); + } + static void do_help(int argc OVS_UNUSED, char *argv[] OVS_UNUSED) { diff --cc vswitchd/bridge.c index 45b32bb64,60a0d6e1c..45ce7277c --- a/vswitchd/bridge.c +++ b/vswitchd/bridge.c @@@ -84,7 -85,8 +85,8 @@@ struct iface /* These members are valid only after bridge_reconfigure() causes them to * be initialized. */ - struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */ - int dp_ifidx; /* Index within kernel datapath. */ ++ struct hmap_node xf_ifidx_node; /* In struct bridge's "ifaces" hmap. */ + int xf_ifidx; /* Index within kernel datapath. */ struct netdev *netdev; /* Network device. */ bool enabled; /* May be chosen for flows? */ const struct ovsrec_interface *cfg; @@@ -163,8 -165,8 +165,8 @@@ struct bridge struct ofproto *ofproto; /* OpenFlow switch. */ /* Kernel datapath information. */ - struct dpif *dpif; /* Datapath. */ + struct xfif *xfif; /* Datapath. */ - struct port_array ifaces; /* Indexed by kernel datapath port number. */ + struct hmap ifaces; /* Contains "struct iface"s. */ /* Bridge ports. */ struct port **ports; @@@ -605,42 -622,41 +621,41 @@@ bridge_reconfigure(const struct ovsrec_ * The kernel will reject any attempt to add a given port to a datapath if * that port already belongs to a different datapath, so we must do all * port deletions before any port additions. */ - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { + LIST_FOR_EACH (br, node, &all_bridges) { - struct odp_port *dpif_ports; - size_t n_dpif_ports; + struct xflow_port *xfif_ports; + size_t n_xfif_ports; struct shash want_ifaces; - dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); + xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports); bridge_get_all_ifaces(br, &want_ifaces); - for (i = 0; i < n_dpif_ports; i++) { - const struct odp_port *p = &dpif_ports[i]; + for (i = 0; i < n_xfif_ports; i++) { + const struct xflow_port *p = &xfif_ports[i]; if (!shash_find(&want_ifaces, p->devname) && strcmp(p->devname, br->name)) { - int retval = dpif_port_del(br->dpif, p->port); + int retval = xfif_port_del(br->xfif, p->port); if (retval) { VLOG_ERR("failed to remove %s interface from %s: %s", - p->devname, dpif_name(br->dpif), + p->devname, xfif_name(br->xfif), strerror(retval)); } } } shash_destroy(&want_ifaces); - free(dpif_ports); + free(xfif_ports); } - LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { + LIST_FOR_EACH (br, node, &all_bridges) { - struct odp_port *dpif_ports; - size_t n_dpif_ports; + struct xflow_port *xfif_ports; + size_t n_xfif_ports; struct shash cur_ifaces, want_ifaces; - struct shash_node *node; /* Get the set of interfaces currently in this datapath. */ - dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); + xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports); shash_init(&cur_ifaces); - for (i = 0; i < n_dpif_ports; i++) { - const char *name = dpif_ports[i].devname; + for (i = 0; i < n_xfif_ports; i++) { + const char *name = xfif_ports[i].devname; shash_add_once(&cur_ifaces, name, NULL); } - free(dpif_ports); + free(xfif_ports); /* Get the set of interfaces we want on this datapath. */ bridge_get_all_ifaces(br, &want_ifaces); @@@ -1309,15 -1343,15 +1340,15 @@@ bridge_destroy(struct bridge *br port_destroy(br->ports[br->n_ports - 1]); } list_remove(&br->node); - error = dpif_delete(br->dpif); + error = xfif_delete(br->xfif); if (error && error != ENOENT) { VLOG_ERR("failed to delete %s: %s", - dpif_name(br->dpif), strerror(error)); + xfif_name(br->xfif), strerror(error)); } - dpif_close(br->dpif); + xfif_close(br->xfif); ofproto_destroy(br->ofproto); mac_learning_destroy(br->ml); - port_array_destroy(&br->ifaces); + hmap_destroy(&br->ifaces); shash_destroy(&br->port_by_name); shash_destroy(&br->iface_by_name); free(br->ports); @@@ -1709,25 -1753,26 +1740,26 @@@ bridge_fetch_dp_ifaces(struct bridge *b struct port *port = br->ports[i]; for (j = 0; j < port->n_ifaces; j++) { struct iface *iface = port->ifaces[j]; - iface->dp_ifidx = -1; + iface->xf_ifidx = -1; } } - port_array_clear(&br->ifaces); + hmap_clear(&br->ifaces); - dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); - for (i = 0; i < n_dpif_ports; i++) { - struct odp_port *p = &dpif_ports[i]; + xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports); + for (i = 0; i < n_xfif_ports; i++) { + struct xflow_port *p = &xfif_ports[i]; struct iface *iface = iface_lookup(br, p->devname); if (iface) { - if (iface->dp_ifidx >= 0) { + if (iface->xf_ifidx >= 0) { VLOG_WARN("%s reported interface %s twice", - dpif_name(br->dpif), p->devname); - } else if (iface_from_dp_ifidx(br, p->port)) { + xfif_name(br->xfif), p->devname); + } else if (iface_from_xf_ifidx(br, p->port)) { VLOG_WARN("%s reported interface %"PRIu16" twice", - dpif_name(br->dpif), p->port); + xfif_name(br->xfif), p->port); } else { - port_array_set(&br->ifaces, p->port, iface); - iface->dp_ifidx = p->port; - hmap_insert(&br->ifaces, &iface->dp_ifidx_node, - hash_int(iface->dp_ifidx, 0)); + iface->xf_ifidx = p->port; ++ hmap_insert(&br->ifaces, &iface->xf_ifidx_node, ++ hash_int(iface->xf_ifidx, 0)); } if (iface->cfg) { @@@ -2891,9 -2934,9 +2922,9 @@@ bond_send_learning_packets(struct port ofpbuf_init(&packet, 128); error = n_packets = n_errors = 0; - LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) { + LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { union ofp_action actions[2], *a; - uint16_t dp_ifidx; + uint16_t xf_ifidx; tag_type tags = 0; flow_t flow; int retval; @@@ -3042,9 -3085,8 +3073,8 @@@ bond_unixctl_show(struct unixctl_conn * hash, be->tx_bytes / 1024); /* MACs. */ - LIST_FOR_EACH (me, struct mac_entry, lru_node, - &port->bridge->ml->lrus) { + LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) { - uint16_t dp_ifidx; + uint16_t xf_ifidx; tag_type tags = 0; if (bond_hash(me->mac) == hash && me->port != port->port_idx @@@ -3693,8 -3734,8 +3722,8 @@@ iface_destroy(struct iface *iface shash_find_and_delete_assert(&br->iface_by_name, iface->name); - if (iface->dp_ifidx >= 0) { - hmap_remove(&br->ifaces, &iface->dp_ifidx_node); + if (iface->xf_ifidx >= 0) { - port_array_set(&br->ifaces, iface->xf_ifidx, NULL); ++ hmap_remove(&br->ifaces, &iface->xf_ifidx_node); } del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces]; @@@ -3722,9 -3763,17 +3751,17 @@@ iface_lookup(const struct bridge *br, c } static struct iface * -iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) +iface_from_xf_ifidx(const struct bridge *br, uint16_t xf_ifidx) { - return port_array_get(&br->ifaces, xf_ifidx); + struct iface *iface; + - HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node, - hash_int(dp_ifidx, 0), &br->ifaces) { - if (iface->dp_ifidx == dp_ifidx) { ++ HMAP_FOR_EACH_IN_BUCKET (iface, xf_ifidx_node, ++ hash_int(xf_ifidx, 0), &br->ifaces) { ++ if (iface->xf_ifidx == xf_ifidx) { + return iface; + } + } + return NULL; } /* Returns true if 'iface' is the name of an "internal" interface on bridge