X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=datapath%2Fflow.c;h=0987fd2afc00ebaa69e6ed54af667368865ecedd;hb=d656937779f5b987ec021c21189a0deab29fd64d;hp=b80e0e295082b148085815d75e07d9c19b141491;hpb=ca78c6b69c01508713b8a8e50065843fcaf53936;p=sliver-openvswitch.git diff --git a/datapath/flow.c b/datapath/flow.c index b80e0e295..0987fd2af 100644 --- a/datapath/flow.c +++ b/datapath/flow.c @@ -1,6 +1,6 @@ /* * Distributed under the terms of the GNU GPL version 2. - * Copyright (c) 2007, 2008, 2009, 2010 Nicira Networks. + * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks. * * Significant portions of this file may be copied from parts of the Linux * kernel, by Linus Torvalds and others. @@ -8,6 +8,7 @@ #include "flow.h" #include "datapath.h" +#include #include #include #include @@ -29,10 +30,8 @@ #include #include -#include "compat.h" - -struct kmem_cache *flow_cache; -static unsigned int hash_seed; +static struct kmem_cache *flow_cache; +static unsigned int hash_seed __read_mostly; static inline bool arphdr_ok(struct sk_buff *skb) { @@ -104,43 +103,57 @@ void flow_used(struct sw_flow *flow, struct sk_buff *skb) spin_unlock_bh(&flow->lock); } -struct sw_flow_actions *flow_actions_alloc(size_t n_actions) +struct sw_flow_actions *flow_actions_alloc(u32 actions_len) { struct sw_flow_actions *sfa; - if (n_actions > (PAGE_SIZE - sizeof *sfa) / sizeof(union odp_action)) + if (actions_len % NLA_ALIGNTO) + return ERR_PTR(-EINVAL); + + /* 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 (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4)) return ERR_PTR(-EINVAL); - sfa = kmalloc(sizeof *sfa + n_actions * sizeof(union odp_action), - GFP_KERNEL); + sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL); if (!sfa) return ERR_PTR(-ENOMEM); - sfa->n_actions = n_actions; + sfa->actions_len = actions_len; return sfa; } - -/* Frees 'flow' immediately. */ -static void flow_free(struct sw_flow *flow) +struct sw_flow *flow_alloc(void) { - if (unlikely(!flow)) - return; - kfree(flow->sf_acts); - kmem_cache_free(flow_cache, flow); + struct sw_flow *flow; + + flow = kmem_cache_alloc(flow_cache, GFP_KERNEL); + if (!flow) + return ERR_PTR(-ENOMEM); + + spin_lock_init(&flow->lock); + atomic_set(&flow->refcnt, 1); + flow->dead = false; + + return flow; } void flow_free_tbl(struct tbl_node *node) { struct sw_flow *flow = flow_cast(node); - flow_free(flow); + + flow->dead = true; + flow_put(flow); } /* RCU callback used by flow_deferred_free. */ static void rcu_free_flow_callback(struct rcu_head *rcu) { struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu); - flow_free(flow); + + flow->dead = true; + flow_put(flow); } /* Schedules 'flow' to be freed after the next RCU grace period. @@ -150,10 +163,26 @@ void flow_deferred_free(struct sw_flow *flow) call_rcu(&flow->rcu, rcu_free_flow_callback); } +void flow_hold(struct sw_flow *flow) +{ + atomic_inc(&flow->refcnt); +} + +void flow_put(struct sw_flow *flow) +{ + if (unlikely(!flow)) + return; + + if (atomic_dec_and_test(&flow->refcnt)) { + kfree((struct sf_flow_acts __force *)flow->sf_acts); + kmem_cache_free(flow_cache, flow); + } +} + /* RCU callback used by flow_deferred_free_acts. */ static void rcu_free_acts_callback(struct rcu_head *rcu) { - struct sw_flow_actions *sf_acts = container_of(rcu, + struct sw_flow_actions *sf_acts = container_of(rcu, struct sw_flow_actions, rcu); kfree(sf_acts); } @@ -165,7 +194,7 @@ void flow_deferred_free_acts(struct sw_flow_actions *sf_acts) call_rcu(&sf_acts->rcu, rcu_free_acts_callback); } -static void parse_vlan(struct sk_buff *skb, struct odp_flow_key *key) +static void parse_vlan(struct sk_buff *skb, struct sw_flow_key *key) { struct qtag_prefix { __be16 eth_type; /* ETH_P_8021Q */ @@ -177,8 +206,7 @@ static void parse_vlan(struct sk_buff *skb, struct odp_flow_key *key) return; qp = (struct qtag_prefix *) skb->data; - key->dl_vlan = qp->tci & htons(VLAN_VID_MASK); - key->dl_vlan_pcp = (ntohs(qp->tci) & VLAN_PCP_MASK) >> VLAN_PCP_SHIFT; + key->dl_tci = qp->tci | htons(VLAN_TAG_PRESENT); __skb_pull(skb, sizeof(struct qtag_prefix)); } @@ -189,7 +217,7 @@ static __be16 parse_ethertype(struct sk_buff *skb) u8 ssap; /* Always 0xAA */ u8 ctrl; u8 oui[3]; - u16 ethertype; + __be16 ethertype; }; struct llc_snap_hdr *llc; __be16 proto; @@ -197,17 +225,17 @@ static __be16 parse_ethertype(struct sk_buff *skb) proto = *(__be16 *) skb->data; __skb_pull(skb, sizeof(__be16)); - if (ntohs(proto) >= ODP_DL_TYPE_ETH2_CUTOFF) + if (ntohs(proto) >= 1536) return proto; if (unlikely(skb->len < sizeof(struct llc_snap_hdr))) - return htons(ODP_DL_TYPE_NOT_ETH_TYPE); + return htons(ETH_P_802_2); llc = (struct llc_snap_hdr *) skb->data; if (llc->dsap != LLC_SAP_SNAP || llc->ssap != LLC_SAP_SNAP || (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0) - return htons(ODP_DL_TYPE_NOT_ETH_TYPE); + return htons(ETH_P_802_2); __skb_pull(skb, sizeof(struct llc_snap_hdr)); return llc->ethertype; @@ -219,6 +247,8 @@ static __be16 parse_ethertype(struct sk_buff *skb) * Ethernet header * @in_port: port number on which @skb was received. * @key: output flow key + * @is_frag: set to 1 if @skb contains an IPv4 fragment, or to 0 if @skb does + * not contain an IPv4 packet or if it is not a fragment. * * The caller must ensure that skb->len >= ETH_HLEN. * @@ -235,19 +265,16 @@ static __be16 parse_ethertype(struct sk_buff *skb) * past the IPv4 header, if one is present and of a correct length, * otherwise the same as skb->network_header. For other key->dl_type * values it is left untouched. - * - * 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 odp_flow_key *key) +int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key, + bool *is_frag) { struct ethhdr *eth; - memset(key, 0, sizeof *key); + memset(key, 0, sizeof(*key)); key->tun_id = OVS_CB(skb)->tun_id; key->in_port = in_port; - key->dl_vlan = htons(ODP_VLAN_NONE); - OVS_CB(skb)->is_frag = false; + *is_frag = false; /* * We would really like to pull as many bytes as we could possibly @@ -305,7 +332,8 @@ int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key) key->nw_proto = nh->protocol; /* Transport layer. */ - if (!(nh->frag_off & htons(IP_MF | IP_OFFSET))) { + if (!(nh->frag_off & htons(IP_MF | IP_OFFSET)) && + !(skb_shinfo(skb)->gso_type & SKB_GSO_UDP)) { if (key->nw_proto == IPPROTO_TCP) { if (tcphdr_ok(skb)) { struct tcphdr *tcp = tcp_hdr(skb); @@ -328,9 +356,9 @@ int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key) key->tp_dst = htons(icmp->code); } } - } else { - OVS_CB(skb)->is_frag = true; - } + } else + *is_frag = true; + } else if (key->dl_type == htons(ETH_P_ARP) && arphdr_ok(skb)) { struct arp_eth_header *arp; @@ -342,11 +370,10 @@ int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key) && arp->ar_pln == 4) { /* We only match on the lower 8 bits of the opcode. */ - if (ntohs(arp->ar_op) <= 0xff) { + if (ntohs(arp->ar_op) <= 0xff) key->nw_proto = ntohs(arp->ar_op); - } - if (key->nw_proto == ARPOP_REQUEST + if (key->nw_proto == ARPOP_REQUEST || key->nw_proto == ARPOP_REPLY) { memcpy(&key->nw_src, arp->ar_sip, sizeof(key->nw_src)); memcpy(&key->nw_dst, arp->ar_tip, sizeof(key->nw_dst)); @@ -356,17 +383,281 @@ int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key) return 0; } -u32 flow_hash(const struct odp_flow_key *key) +u32 flow_hash(const struct sw_flow_key *key) { - return jhash2((u32*)key, sizeof *key / sizeof(u32), hash_seed); + return jhash2((u32*)key, sizeof(*key) / sizeof(u32), hash_seed); } int flow_cmp(const struct tbl_node *node, void *key2_) { - const struct odp_flow_key *key1 = &flow_cast(node)->key; - const struct odp_flow_key *key2 = key2_; + const struct sw_flow_key *key1 = &flow_cast(node)->key; + const struct sw_flow_key *key2 = key2_; + + return !memcmp(key1, key2, sizeof(struct sw_flow_key)); +} + +/** + * flow_from_nlattrs - parses Netlink attributes into a flow key. + * @swkey: receives the extracted flow key. + * @key: Netlink attribute holding nested %ODP_KEY_ATTR_* Netlink attribute + * sequence. + * + * This state machine accepts the following forms, with [] for optional + * elements and | for alternatives: + * + * [tun_id] in_port ethernet [8021q] [ethertype [IP [TCP|UDP|ICMP] | ARP] + */ +int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *attr) +{ + const struct nlattr *nla; + u16 prev_type; + int rem; + + memset(swkey, 0, sizeof(*swkey)); + swkey->dl_type = htons(ETH_P_802_2); + + prev_type = ODP_KEY_ATTR_UNSPEC; + nla_for_each_nested(nla, attr, rem) { + static const u32 key_lens[ODP_KEY_ATTR_MAX + 1] = { + [ODP_KEY_ATTR_TUN_ID] = 8, + [ODP_KEY_ATTR_IN_PORT] = 4, + [ODP_KEY_ATTR_ETHERNET] = sizeof(struct odp_key_ethernet), + [ODP_KEY_ATTR_8021Q] = sizeof(struct odp_key_8021q), + [ODP_KEY_ATTR_ETHERTYPE] = 2, + [ODP_KEY_ATTR_IPV4] = sizeof(struct odp_key_ipv4), + [ODP_KEY_ATTR_TCP] = sizeof(struct odp_key_tcp), + [ODP_KEY_ATTR_UDP] = sizeof(struct odp_key_udp), + [ODP_KEY_ATTR_ICMP] = sizeof(struct odp_key_icmp), + [ODP_KEY_ATTR_ARP] = sizeof(struct odp_key_arp), + }; + + const struct odp_key_ethernet *eth_key; + const struct odp_key_8021q *q_key; + const struct odp_key_ipv4 *ipv4_key; + const struct odp_key_tcp *tcp_key; + const struct odp_key_udp *udp_key; + const struct odp_key_icmp *icmp_key; + const struct odp_key_arp *arp_key; + + int type = nla_type(nla); + + if (type > ODP_KEY_ATTR_MAX || nla_len(nla) != key_lens[type]) + return -EINVAL; + +#define TRANSITION(PREV_TYPE, TYPE) (((PREV_TYPE) << 16) | (TYPE)) + switch (TRANSITION(prev_type, type)) { + case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_TUN_ID): + swkey->tun_id = nla_get_be64(nla); + break; + + case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_IN_PORT): + case TRANSITION(ODP_KEY_ATTR_TUN_ID, ODP_KEY_ATTR_IN_PORT): + if (nla_get_u32(nla) >= DP_MAX_PORTS) + return -EINVAL; + swkey->in_port = nla_get_u32(nla); + break; + + case TRANSITION(ODP_KEY_ATTR_IN_PORT, ODP_KEY_ATTR_ETHERNET): + eth_key = nla_data(nla); + memcpy(swkey->dl_src, eth_key->eth_src, ETH_ALEN); + memcpy(swkey->dl_dst, eth_key->eth_dst, ETH_ALEN); + break; + + case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_8021Q): + q_key = nla_data(nla); + /* Only standard 0x8100 VLANs currently supported. */ + if (q_key->q_tpid != htons(ETH_P_8021Q)) + return -EINVAL; + if (q_key->q_tci & htons(VLAN_TAG_PRESENT)) + return -EINVAL; + swkey->dl_tci = q_key->q_tci | htons(VLAN_TAG_PRESENT); + break; + + case TRANSITION(ODP_KEY_ATTR_8021Q, ODP_KEY_ATTR_ETHERTYPE): + case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_ETHERTYPE): + swkey->dl_type = nla_get_be16(nla); + if (ntohs(swkey->dl_type) < 1536) + return -EINVAL; + break; + + case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_IPV4): + if (swkey->dl_type != htons(ETH_P_IP)) + return -EINVAL; + ipv4_key = nla_data(nla); + swkey->nw_src = ipv4_key->ipv4_src; + swkey->nw_dst = ipv4_key->ipv4_dst; + swkey->nw_proto = ipv4_key->ipv4_proto; + swkey->nw_tos = ipv4_key->ipv4_tos; + if (swkey->nw_tos & INET_ECN_MASK) + return -EINVAL; + break; + + case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_TCP): + if (swkey->nw_proto != IPPROTO_TCP) + return -EINVAL; + tcp_key = nla_data(nla); + swkey->tp_src = tcp_key->tcp_src; + swkey->tp_dst = tcp_key->tcp_dst; + break; + + case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_UDP): + if (swkey->nw_proto != IPPROTO_UDP) + return -EINVAL; + udp_key = nla_data(nla); + swkey->tp_src = udp_key->udp_src; + swkey->tp_dst = udp_key->udp_dst; + break; + + case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_ICMP): + if (swkey->nw_proto != IPPROTO_ICMP) + return -EINVAL; + icmp_key = nla_data(nla); + swkey->tp_src = htons(icmp_key->icmp_type); + swkey->tp_dst = htons(icmp_key->icmp_code); + break; + + case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_ARP): + if (swkey->dl_type != htons(ETH_P_ARP)) + return -EINVAL; + arp_key = nla_data(nla); + swkey->nw_src = arp_key->arp_sip; + swkey->nw_dst = arp_key->arp_tip; + if (arp_key->arp_op & htons(0xff00)) + return -EINVAL; + swkey->nw_proto = ntohs(arp_key->arp_op); + break; + + default: + return -EINVAL; + } + + prev_type = type; + } + if (rem) + return -EINVAL; + + switch (prev_type) { + case ODP_KEY_ATTR_UNSPEC: + return -EINVAL; + + case ODP_KEY_ATTR_TUN_ID: + case ODP_KEY_ATTR_IN_PORT: + return -EINVAL; + + case ODP_KEY_ATTR_ETHERNET: + case ODP_KEY_ATTR_8021Q: + return 0; + + case ODP_KEY_ATTR_ETHERTYPE: + if (swkey->dl_type == htons(ETH_P_IP) || + swkey->dl_type == htons(ETH_P_ARP)) + return -EINVAL; + return 0; + + case ODP_KEY_ATTR_IPV4: + if (swkey->nw_proto == IPPROTO_TCP || + swkey->nw_proto == IPPROTO_UDP || + swkey->nw_proto == IPPROTO_ICMP) + return -EINVAL; + return 0; + + case ODP_KEY_ATTR_TCP: + case ODP_KEY_ATTR_UDP: + case ODP_KEY_ATTR_ICMP: + case ODP_KEY_ATTR_ARP: + return 0; + } + + WARN_ON_ONCE(1); + return -EINVAL; +} + +int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb) +{ + struct odp_key_ethernet *eth_key; + struct nlattr *nla; + + if (swkey->tun_id != cpu_to_be64(0)) + NLA_PUT_BE64(skb, ODP_KEY_ATTR_TUN_ID, swkey->tun_id); + + NLA_PUT_U32(skb, ODP_KEY_ATTR_IN_PORT, swkey->in_port); + + nla = nla_reserve(skb, ODP_KEY_ATTR_ETHERNET, sizeof(*eth_key)); + if (!nla) + goto nla_put_failure; + eth_key = nla_data(nla); + memcpy(eth_key->eth_src, swkey->dl_src, ETH_ALEN); + memcpy(eth_key->eth_dst, swkey->dl_dst, ETH_ALEN); + + if (swkey->dl_tci != htons(0)) { + struct odp_key_8021q q_key; + + q_key.q_tpid = htons(ETH_P_8021Q); + q_key.q_tci = swkey->dl_tci & ~htons(VLAN_TAG_PRESENT); + NLA_PUT(skb, ODP_KEY_ATTR_8021Q, sizeof(q_key), &q_key); + } + + if (swkey->dl_type == htons(ETH_P_802_2)) + return 0; + + NLA_PUT_BE16(skb, ODP_KEY_ATTR_ETHERTYPE, swkey->dl_type); + + if (swkey->dl_type == htons(ETH_P_IP)) { + struct odp_key_ipv4 *ipv4_key; + + nla = nla_reserve(skb, ODP_KEY_ATTR_IPV4, sizeof(*ipv4_key)); + if (!nla) + goto nla_put_failure; + ipv4_key = nla_data(nla); + ipv4_key->ipv4_src = swkey->nw_src; + ipv4_key->ipv4_dst = swkey->nw_dst; + ipv4_key->ipv4_proto = swkey->nw_proto; + ipv4_key->ipv4_tos = swkey->nw_tos; + + if (swkey->nw_proto == IPPROTO_TCP) { + struct odp_key_tcp *tcp_key; + + nla = nla_reserve(skb, ODP_KEY_ATTR_TCP, sizeof(*tcp_key)); + if (!nla) + goto nla_put_failure; + tcp_key = nla_data(nla); + tcp_key->tcp_src = swkey->tp_src; + tcp_key->tcp_dst = swkey->tp_dst; + } else if (swkey->nw_proto == IPPROTO_UDP) { + struct odp_key_udp *udp_key; + + nla = nla_reserve(skb, ODP_KEY_ATTR_UDP, sizeof(*udp_key)); + if (!nla) + goto nla_put_failure; + udp_key = nla_data(nla); + udp_key->udp_src = swkey->tp_src; + udp_key->udp_dst = swkey->tp_dst; + } else if (swkey->nw_proto == IPPROTO_ICMP) { + struct odp_key_icmp *icmp_key; + + nla = nla_reserve(skb, ODP_KEY_ATTR_ICMP, sizeof(*icmp_key)); + if (!nla) + goto nla_put_failure; + icmp_key = nla_data(nla); + icmp_key->icmp_type = ntohs(swkey->tp_src); + icmp_key->icmp_code = ntohs(swkey->tp_dst); + } + } else if (swkey->dl_type == htons(ETH_P_ARP)) { + struct odp_key_arp *arp_key; + + nla = nla_reserve(skb, ODP_KEY_ATTR_ARP, sizeof(*arp_key)); + if (!nla) + goto nla_put_failure; + arp_key = nla_data(nla); + arp_key->arp_sip = swkey->nw_src; + arp_key->arp_tip = swkey->nw_dst; + arp_key->arp_op = htons(swkey->nw_proto); + } + + return 0; - return !memcmp(key1, key2, sizeof(struct odp_flow_key)); +nla_put_failure: + return -EMSGSIZE; } /* Initializes the flow module. @@ -378,7 +669,7 @@ int flow_init(void) if (flow_cache == NULL) return -ENOMEM; - get_random_bytes(&hash_seed, sizeof hash_seed); + get_random_bytes(&hash_seed, sizeof(hash_seed)); return 0; }