X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=datapath%2Ftunnel.c;h=bf66f4f0932abc14df598aad53fcf6659ae2bb3e;hb=8fe2b9682c74b4735a5a7503568a586e3353ca85;hp=ad45228087e65b6cab5d84b549399431b1e721cb;hpb=4bee421f3ab28492aebc32b8b13e41ca5d12a936;p=sliver-openvswitch.git diff --git a/datapath/tunnel.c b/datapath/tunnel.c index ad4522808..bf66f4f09 100644 --- a/datapath/tunnel.c +++ b/datapath/tunnel.c @@ -29,6 +29,7 @@ #include #include "actions.h" +#include "checksum.h" #include "datapath.h" #include "table.h" #include "tunnel.h" @@ -67,21 +68,20 @@ #define CACHE_DATA_ALIGN 16 -/* Protected by RCU. */ -static struct tbl *port_table; +static struct tbl __rcu *port_table __read_mostly; static void cache_cleaner(struct work_struct *work); -DECLARE_DELAYED_WORK(cache_cleaner_wq, cache_cleaner); +static DECLARE_DELAYED_WORK(cache_cleaner_wq, cache_cleaner); /* * These are just used as an optimization: they don't require any kind of * synchronization because we could have just as easily read the value before * the port change happened. */ -static unsigned int key_local_remote_ports; -static unsigned int key_remote_ports; -static unsigned int local_remote_ports; -static unsigned int remote_ports; +static unsigned int key_local_remote_ports __read_mostly; +static unsigned int key_remote_ports __read_mostly; +static unsigned int local_remote_ports __read_mostly; +static unsigned int remote_ports __read_mostly; #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) #define rt_dst(rt) (rt->dst) @@ -99,6 +99,15 @@ static inline struct tnl_vport *tnl_vport_table_cast(const struct tbl_node *node return container_of(node, struct tnl_vport, tbl_node); } +/* This is analogous to rtnl_dereference for the tunnel cache. It checks that + * cache_lock is held, so it is only for update side code. + */ +static inline struct tnl_cache *cache_dereference(struct tnl_vport *tnl_vport) +{ + return rcu_dereference_protected(tnl_vport->cache, + lockdep_is_held(&tnl_vport->cache_lock)); +} + static inline void schedule_cache_cleaner(void) { schedule_delayed_work(&cache_cleaner_wq, CACHE_CLEANER_INTERVAL); @@ -132,7 +141,7 @@ static void assign_config_rcu(struct vport *vport, struct tnl_vport *tnl_vport = tnl_vport_priv(vport); struct tnl_mutable_config *old_config; - old_config = tnl_vport->mutable; + old_config = rtnl_dereference(tnl_vport->mutable); rcu_assign_pointer(tnl_vport->mutable, new_config); call_rcu(&old_config->rcu, free_config_rcu); } @@ -142,7 +151,7 @@ static void assign_cache_rcu(struct vport *vport, struct tnl_cache *new_cache) struct tnl_vport *tnl_vport = tnl_vport_priv(vport); struct tnl_cache *old_cache; - old_cache = tnl_vport->cache; + old_cache = cache_dereference(tnl_vport); rcu_assign_pointer(tnl_vport->cache, new_cache); if (old_cache) @@ -164,16 +173,12 @@ static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable) } } -enum lookup_key { - LOOKUP_TUNNEL_TYPE = 0, - LOOKUP_SADDR = 1, - LOOKUP_DADDR = 2, - LOOKUP_KEY = 3, -}; - struct port_lookup_key { - u32 vals[4]; /* Contains enum lookup_key keys. */ const struct tnl_mutable_config *mutable; + __be64 key; + u32 tunnel_type; + __be32 saddr; + __be32 daddr; }; /* @@ -185,36 +190,38 @@ static int port_cmp(const struct tbl_node *node, void *target) const struct tnl_vport *tnl_vport = tnl_vport_table_cast(node); struct port_lookup_key *lookup = target; - lookup->mutable = rcu_dereference(tnl_vport->mutable); + lookup->mutable = rcu_dereference_rtnl(tnl_vport->mutable); - return (lookup->mutable->tunnel_type == lookup->vals[LOOKUP_TUNNEL_TYPE]) && - lookup->mutable->port_config.daddr == lookup->vals[LOOKUP_DADDR] && - lookup->mutable->port_config.in_key == lookup->vals[LOOKUP_KEY] && - lookup->mutable->port_config.saddr == lookup->vals[LOOKUP_SADDR]; + return (lookup->mutable->tunnel_type == lookup->tunnel_type && + lookup->mutable->port_config.daddr == lookup->daddr && + lookup->mutable->port_config.in_key == lookup->key && + lookup->mutable->port_config.saddr == lookup->saddr); } -static u32 port_hash(struct port_lookup_key *lookup) +static u32 port_hash(struct port_lookup_key *k) { - return jhash2(lookup->vals, ARRAY_SIZE(lookup->vals), 0); + u32 x = jhash_3words((__force u32)k->saddr, (__force u32)k->daddr, + k->tunnel_type, 0); + return jhash_2words((__force u64)k->key >> 32, (__force u32)k->key, x); } static u32 mutable_hash(const struct tnl_mutable_config *mutable) { struct port_lookup_key lookup; - lookup.vals[LOOKUP_SADDR] = mutable->port_config.saddr; - lookup.vals[LOOKUP_DADDR] = mutable->port_config.daddr; - lookup.vals[LOOKUP_KEY] = mutable->port_config.in_key; - lookup.vals[LOOKUP_TUNNEL_TYPE] = mutable->tunnel_type; + lookup.saddr = mutable->port_config.saddr; + lookup.daddr = mutable->port_config.daddr; + lookup.key = mutable->port_config.in_key; + lookup.tunnel_type = mutable->tunnel_type; return port_hash(&lookup); } static void check_table_empty(void) { - if (tbl_count(port_table) == 0) { - struct tbl *old_table = port_table; + struct tbl *old_table = rtnl_dereference(port_table); + if (tbl_count(old_table) == 0) { cancel_delayed_work_sync(&cache_cleaner_wq); rcu_assign_pointer(port_table, NULL); tbl_deferred_destroy(old_table, NULL); @@ -223,6 +230,7 @@ static void check_table_empty(void) static int add_port(struct vport *vport) { + struct tbl *cur_table = rtnl_dereference(port_table); struct tnl_vport *tnl_vport = tnl_vport_priv(vport); int err; @@ -236,25 +244,25 @@ static int add_port(struct vport *vport) rcu_assign_pointer(port_table, new_table); schedule_cache_cleaner(); - } else if (tbl_count(port_table) > tbl_n_buckets(port_table)) { - struct tbl *old_table = port_table; + } else if (tbl_count(cur_table) > tbl_n_buckets(cur_table)) { struct tbl *new_table; - new_table = tbl_expand(old_table); + new_table = tbl_expand(cur_table); if (IS_ERR(new_table)) return PTR_ERR(new_table); rcu_assign_pointer(port_table, new_table); - tbl_deferred_destroy(old_table, NULL); + tbl_deferred_destroy(cur_table, NULL); } - err = tbl_insert(port_table, &tnl_vport->tbl_node, mutable_hash(tnl_vport->mutable)); + err = tbl_insert(rtnl_dereference(port_table), &tnl_vport->tbl_node, + mutable_hash(rtnl_dereference(tnl_vport->mutable))); if (err) { check_table_empty(); return err; } - (*find_port_pool(tnl_vport->mutable))++; + (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++; return 0; } @@ -262,6 +270,7 @@ static int add_port(struct vport *vport) static int move_port(struct vport *vport, struct tnl_mutable_config *new_mutable) { int err; + struct tbl *cur_table = rtnl_dereference(port_table); struct tnl_vport *tnl_vport = tnl_vport_priv(vport); u32 hash; @@ -274,18 +283,21 @@ static int move_port(struct vport *vport, struct tnl_mutable_config *new_mutable * finding tunnels or the possibility of failure. However, if we do * find a tunnel it will always be consistent. */ - err = tbl_remove(port_table, &tnl_vport->tbl_node); + err = tbl_remove(cur_table, &tnl_vport->tbl_node); if (err) return err; - err = tbl_insert(port_table, &tnl_vport->tbl_node, hash); + err = tbl_insert(cur_table, &tnl_vport->tbl_node, hash); if (err) { + (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--; check_table_empty(); return err; } table_updated: + (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--; assign_config_rcu(vport, new_mutable); + (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++; return 0; } @@ -295,33 +307,33 @@ static int del_port(struct vport *vport) struct tnl_vport *tnl_vport = tnl_vport_priv(vport); int err; - err = tbl_remove(port_table, &tnl_vport->tbl_node); + err = tbl_remove(rtnl_dereference(port_table), &tnl_vport->tbl_node); if (err) return err; check_table_empty(); - (*find_port_pool(tnl_vport->mutable))--; + (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--; return 0; } -struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be32 key, +struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be64 key, int tunnel_type, const struct tnl_mutable_config **mutable) { struct port_lookup_key lookup; - struct tbl *table = rcu_dereference(port_table); + struct tbl *table = rcu_dereference_rtnl(port_table); struct tbl_node *tbl_node; if (unlikely(!table)) return NULL; - lookup.vals[LOOKUP_SADDR] = saddr; - lookup.vals[LOOKUP_DADDR] = daddr; + lookup.saddr = saddr; + lookup.daddr = daddr; if (tunnel_type & TNL_T_KEY_EXACT) { - lookup.vals[LOOKUP_KEY] = key; - lookup.vals[LOOKUP_TUNNEL_TYPE] = tunnel_type & ~TNL_T_KEY_MATCH; + lookup.key = key; + lookup.tunnel_type = tunnel_type & ~TNL_T_KEY_MATCH; if (key_local_remote_ports) { tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp); @@ -330,19 +342,19 @@ struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be32 key, } if (key_remote_ports) { - lookup.vals[LOOKUP_SADDR] = 0; + lookup.saddr = 0; tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp); if (tbl_node) goto found; - lookup.vals[LOOKUP_SADDR] = saddr; + lookup.saddr = saddr; } } if (tunnel_type & TNL_T_KEY_MATCH) { - lookup.vals[LOOKUP_KEY] = 0; - lookup.vals[LOOKUP_TUNNEL_TYPE] = tunnel_type & ~TNL_T_KEY_EXACT; + lookup.key = 0; + lookup.tunnel_type = tunnel_type & ~TNL_T_KEY_EXACT; if (local_remote_ports) { tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp); @@ -351,7 +363,7 @@ struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be32 key, } if (remote_ports) { - lookup.vals[LOOKUP_SADDR] = 0; + lookup.saddr = 0; tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp); if (tbl_node) @@ -368,34 +380,37 @@ found: static inline void ecn_decapsulate(struct sk_buff *skb) { - u8 tos = ip_hdr(skb)->tos; - - if (INET_ECN_is_ce(tos)) { + /* This is accessing the outer IP header of the tunnel, which we've + * already validated to be OK. skb->data is currently set to the start + * of the inner Ethernet header, and we've validated ETH_HLEN. + */ + if (unlikely(INET_ECN_is_ce(ip_hdr(skb)->tos))) { __be16 protocol = skb->protocol; - unsigned int nw_header = skb_network_offset(skb); + + skb_set_network_header(skb, ETH_HLEN); if (skb->protocol == htons(ETH_P_8021Q)) { if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN))) return; protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto; - nw_header += VLAN_HLEN; + skb_set_network_header(skb, VLAN_ETH_HLEN); } if (protocol == htons(ETH_P_IP)) { - if (unlikely(!pskb_may_pull(skb, nw_header + if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb) + sizeof(struct iphdr)))) return; - IP_ECN_set_ce((struct iphdr *)(skb->data + nw_header)); + IP_ECN_set_ce(ip_hdr(skb)); } #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) else if (protocol == htons(ETH_P_IPV6)) { - if (unlikely(!pskb_may_pull(skb, nw_header + if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb) + sizeof(struct ipv6hdr)))) return; - IP6_ECN_set_ce((struct ipv6hdr *)(skb->data + nw_header)); + IP6_ECN_set_ce(ipv6_hdr(skb)); } #endif } @@ -404,17 +419,25 @@ static inline void ecn_decapsulate(struct sk_buff *skb) /* Called with rcu_read_lock. */ void tnl_rcv(struct vport *vport, struct sk_buff *skb) { - skb->pkt_type = PACKET_HOST; - skb->protocol = eth_type_trans(skb, skb->dev); + /* Packets received by this function are in the following state: + * - skb->data points to the inner Ethernet header. + * - The inner Ethernet header is in the linear data area. + * - skb->csum does not include the inner Ethernet header. + * - The layer pointers point at the outer headers. + */ + + struct ethhdr *eh = (struct ethhdr *)skb->data; + + if (likely(ntohs(eh->h_proto) >= 1536)) + skb->protocol = eh->h_proto; + else + skb->protocol = htons(ETH_P_802_2); skb_dst_drop(skb); nf_reset(skb); secpath_reset(skb); - skb_reset_network_header(skb); ecn_decapsulate(skb); - - skb_push(skb, ETH_HLEN); compute_ip_summed(skb, false); vport_receive(vport, skb); @@ -587,7 +610,7 @@ static void ipv6_build_icmp(struct sk_buff *skb, struct sk_buff *nskb, #endif /* IPv6 */ bool tnl_frag_needed(struct vport *vport, const struct tnl_mutable_config *mutable, - struct sk_buff *skb, unsigned int mtu, __be32 flow_key) + struct sk_buff *skb, unsigned int mtu, __be64 flow_key) { unsigned int eth_hdr_len = ETH_HLEN; unsigned int total_length = 0, header_length = 0, payload_length; @@ -807,7 +830,7 @@ static void cache_cleaner(struct work_struct *work) schedule_cache_cleaner(); rcu_read_lock(); - tbl_foreach(port_table, cache_cleaner_cb, NULL); + tbl_foreach(rcu_dereference(port_table), cache_cleaner_cb, NULL); rcu_read_unlock(); } @@ -860,7 +883,7 @@ static struct tnl_cache *build_cache(struct vport *vport, if (!spin_trylock_bh(&tnl_vport->cache_lock)) return NULL; - cache = tnl_vport->cache; + cache = cache_dereference(tnl_vport); if (check_cache_valid(cache, mutable)) goto unlock; else @@ -887,22 +910,17 @@ static struct tnl_cache *build_cache(struct vport *vport, #endif if (is_internal_dev(rt_dst(rt).dev)) { - int err; + struct odp_flow_key flow_key; + struct tbl_node *flow_node; struct vport *vport; - struct dp_port *dp_port; struct sk_buff *skb; bool is_frag; - struct odp_flow_key flow_key; - struct tbl_node *flow_node; + int err; 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; @@ -910,13 +928,13 @@ static struct tnl_cache *build_cache(struct vport *vport, __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); + err = flow_extract(skb, vport->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_node = tbl_lookup(rcu_dereference(vport->dp->table), &flow_key, flow_hash(&flow_key), flow_cmp); if (flow_node) { @@ -1038,7 +1056,7 @@ static struct sk_buff *handle_offloads(struct sk_buff *skb, */ if (skb_headroom(skb) < min_headroom) { skb = check_headroom(skb, min_headroom); - if (unlikely(IS_ERR(skb))) { + if (IS_ERR(skb)) { err = PTR_ERR(skb); goto error; } @@ -1046,7 +1064,7 @@ static struct sk_buff *handle_offloads(struct sk_buff *skb, nskb = skb_gso_segment(skb, 0); kfree_skb(skb); - if (unlikely(IS_ERR(nskb))) { + if (IS_ERR(nskb)) { err = PTR_ERR(nskb); goto error; } @@ -1054,7 +1072,7 @@ static struct sk_buff *handle_offloads(struct sk_buff *skb, skb = nskb; } else { skb = check_headroom(skb, min_headroom); - if (unlikely(IS_ERR(skb))) { + if (IS_ERR(skb)) { err = PTR_ERR(skb); goto error; } @@ -1099,6 +1117,7 @@ static int send_frags(struct sk_buff *skb, int frag_len = skb->len - mutable->tunnel_hlen; skb->next = NULL; + memset(IPCB(skb), 0, sizeof(*IPCB(skb))); err = ip_local_out(skb); if (likely(net_xmit_eval(err) == 0)) @@ -1191,7 +1210,7 @@ int tnl_send(struct vport *vport, struct sk_buff *skb) /* Offloading */ skb = handle_offloads(skb, mutable, rt); - if (unlikely(IS_ERR(skb))) + if (IS_ERR(skb)) goto error; /* MTU */ @@ -1247,10 +1266,6 @@ int tnl_send(struct vport *vport, struct sk_buff *skb) skb_dst_set(skb, unattached_dst); unattached_dst = NULL; } - - - memset(&IPCB(skb)->opt, 0, sizeof(IPCB(skb)->opt)); - IPCB(skb)->flags = 0; } skb_set_transport_header(skb, skb_network_offset(skb) + sizeof(struct iphdr)); @@ -1266,15 +1281,16 @@ int tnl_send(struct vport *vport, struct sk_buff *skb) if (likely(cache)) { int orig_len = skb->len - cache->len; + struct vport *cache_vport = internal_dev_get_vport(rt_dst(rt).dev); skb->protocol = htons(ETH_P_IP); iph->tot_len = htons(skb->len - skb_network_offset(skb)); ip_send_check(iph); - if (is_internal_dev(rt_dst(rt).dev)) { + if (cache_vport) { OVS_CB(skb)->flow = cache->flow; compute_ip_summed(skb, true); - vport_receive(internal_dev_get_vport(rt_dst(rt).dev), skb); + vport_receive(cache_vport, skb); sent_len += orig_len; } else { int err; @@ -1306,15 +1322,14 @@ out: return sent_len; } -static int set_config(const void __user *uconfig, const struct tnl_ops *tnl_ops, +static int set_config(const void *config, const struct tnl_ops *tnl_ops, const struct vport *cur_vport, struct tnl_mutable_config *mutable) { const struct vport *old_vport; const struct tnl_mutable_config *old_mutable; - if (copy_from_user(&mutable->port_config, uconfig, sizeof(struct tnl_port_config))) - return -EFAULT; + mutable->port_config = *(struct tnl_port_config *)config; if (mutable->port_config.daddr == 0) return -EINVAL; @@ -1350,7 +1365,7 @@ static int set_config(const void __user *uconfig, const struct tnl_ops *tnl_ops, return 0; } -struct vport *tnl_create(const char *name, const void __user *config, +struct vport *tnl_create(const struct vport_parms *parms, const struct vport_ops *vport_ops, const struct tnl_ops *tnl_ops) { @@ -1359,7 +1374,7 @@ struct vport *tnl_create(const char *name, const void __user *config, int initial_frag_id; int err; - vport = vport_alloc(sizeof(struct tnl_vport), vport_ops); + vport = vport_alloc(sizeof(struct tnl_vport), vport_ops, parms); if (IS_ERR(vport)) { err = PTR_ERR(vport); goto error; @@ -1367,7 +1382,7 @@ struct vport *tnl_create(const char *name, const void __user *config, tnl_vport = tnl_vport_priv(vport); - strcpy(tnl_vport->name, name); + strcpy(tnl_vport->name, parms->name); tnl_vport->tnl_ops = tnl_ops; tnl_vport->mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL); @@ -1382,7 +1397,7 @@ struct vport *tnl_create(const char *name, const void __user *config, get_random_bytes(&initial_frag_id, sizeof(int)); atomic_set(&tnl_vport->frag_id, initial_frag_id); - err = set_config(config, tnl_ops, NULL, tnl_vport->mutable); + err = set_config(parms->config, tnl_ops, NULL, tnl_vport->mutable); if (err) goto error_free_mutable; @@ -1407,7 +1422,7 @@ error: return ERR_PTR(err); } -int tnl_modify(struct vport *vport, const void __user *config) +int tnl_modify(struct vport *vport, struct odp_port *port) { struct tnl_vport *tnl_vport = tnl_vport_priv(vport); struct tnl_mutable_config *mutable; @@ -1419,7 +1434,7 @@ int tnl_modify(struct vport *vport, const void __user *config) goto error; } - err = set_config(config, tnl_vport->tnl_ops, vport, mutable); + err = set_config(port->config, tnl_vport->tnl_ops, vport, mutable); if (err) goto error_free; @@ -1505,13 +1520,13 @@ const char *tnl_get_name(const struct vport *vport) const unsigned char *tnl_get_addr(const struct vport *vport) { const struct tnl_vport *tnl_vport = tnl_vport_priv(vport); - return rcu_dereference(tnl_vport->mutable)->eth_addr; + return rcu_dereference_rtnl(tnl_vport->mutable)->eth_addr; } int tnl_get_mtu(const struct vport *vport) { const struct tnl_vport *tnl_vport = tnl_vport_priv(vport); - return rcu_dereference(tnl_vport->mutable)->mtu; + return rcu_dereference_rtnl(tnl_vport->mutable)->mtu; } void tnl_free_linked_skbs(struct sk_buff *skb)