X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=datapath%2Ftunnel.c;h=67fbd2ba3fe821cc2b9bbdedf12edefee868bb60;hb=03a8a29eb87ef21d0e56c200ebe43878ef773d63;hp=2bf61597d7a80eabe48d0cf452a004a60223d187;hpb=bf16ba4a9b1b4a38c867349e70e39d551b406b32;p=openvswitch diff --git a/datapath/tunnel.c b/datapath/tunnel.c index 2bf61597..67fbd2ba 100644 --- a/datapath/tunnel.c +++ b/datapath/tunnel.c @@ -13,9 +13,11 @@ #include #include #include +#include #include #include #include +#include #include #include @@ -31,7 +33,6 @@ #include "actions.h" #include "checksum.h" #include "datapath.h" -#include "table.h" #include "tunnel.h" #include "vlan.h" #include "vport.h" @@ -68,8 +69,10 @@ #define CACHE_CLEANER_INTERVAL (5 * HZ) #define CACHE_DATA_ALIGN 16 +#define PORT_TABLE_SIZE 1024 -static struct tbl __rcu *port_table __read_mostly; +static struct hlist_head *port_table __read_mostly; +static int port_table_count; static void cache_cleaner(struct work_struct *work); static DECLARE_DELAYED_WORK(cache_cleaner_wq, cache_cleaner); @@ -95,11 +98,6 @@ static inline struct vport *tnl_vport_to_vport(const struct tnl_vport *tnl_vport return vport_from_priv(tnl_vport); } -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. */ @@ -186,11 +184,9 @@ struct port_lookup_key { * Modifies 'target' to store the rcu_dereferenced pointer that was used to do * the comparision. */ -static int port_cmp(const struct tbl_node *node, void *target, int unused) +static int port_cmp(const struct tnl_vport *tnl_vport, + struct port_lookup_key *lookup) { - const struct tnl_vport *tnl_vport = tnl_vport_table_cast(node); - struct port_lookup_key *lookup = target; - lookup->mutable = rcu_dereference_rtnl(tnl_vport->mutable); return (lookup->mutable->tunnel_type == lookup->tunnel_type && @@ -218,104 +214,69 @@ static u32 mutable_hash(const struct tnl_mutable_config *mutable) return port_hash(&lookup); } -static void check_table_empty(void) -{ - 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); - } +static inline struct hlist_head *find_bucket(u32 hash) +{ + return &port_table[(hash & (PORT_TABLE_SIZE - 1))]; } -static int add_port(struct vport *vport) +static void port_table_add_port(struct vport *vport) { - struct tbl *cur_table = rtnl_dereference(port_table); struct tnl_vport *tnl_vport = tnl_vport_priv(vport); - int err; - - if (!port_table) { - struct tbl *new_table; + u32 hash = mutable_hash(rtnl_dereference(tnl_vport->mutable)); - new_table = tbl_create(TBL_MIN_BUCKETS); - if (!new_table) - return -ENOMEM; - - rcu_assign_pointer(port_table, new_table); + if (port_table_count == 0) schedule_cache_cleaner(); - } else if (tbl_count(cur_table) > tbl_n_buckets(cur_table)) { - struct tbl *new_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(cur_table, NULL); - } - - 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; - } + hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash)); + port_table_count++; (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++; - - return 0; } -static int move_port(struct vport *vport, struct tnl_mutable_config *new_mutable) +static void port_table_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; hash = mutable_hash(new_mutable); - if (hash == tnl_vport->tbl_node.hash) - goto table_updated; - - /* - * Ideally we should make this move atomic to avoid having gaps in - * finding tunnels or the possibility of failure. However, if we do - * find a tunnel it will always be consistent. - */ - err = tbl_remove(cur_table, &tnl_vport->tbl_node); - if (err) - return err; + hlist_del_init_rcu(&tnl_vport->hash_node); + hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(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; } -static int del_port(struct vport *vport) +static void port_table_remove_port(struct vport *vport) { struct tnl_vport *tnl_vport = tnl_vport_priv(vport); - int err; - err = tbl_remove(rtnl_dereference(port_table), &tnl_vport->tbl_node); - if (err) - return err; + hlist_del_init_rcu(&tnl_vport->hash_node); + + port_table_count--; + if (port_table_count == 0) + cancel_delayed_work_sync(&cache_cleaner_wq); - check_table_empty(); (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--; +} - return 0; +static struct tnl_vport *port_table_lookup(struct port_lookup_key *lookup) +{ + struct hlist_node *n; + struct hlist_head *bucket; + u32 hash = port_hash(lookup); + struct tnl_vport * tnl_vport; + + bucket = find_bucket(hash); + + hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node) { + if (port_cmp(tnl_vport, lookup)) + return tnl_vport; + } + + return NULL; } struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be64 key, @@ -323,11 +284,7 @@ struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be64 key, const struct tnl_mutable_config **mutable) { struct port_lookup_key lookup; - struct tbl *table = rcu_dereference_rtnl(port_table); - struct tbl_node *tbl_node; - - if (unlikely(!table)) - return NULL; + struct tnl_vport * tnl_vport; lookup.saddr = saddr; lookup.daddr = daddr; @@ -337,18 +294,15 @@ struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be64 key, lookup.tunnel_type = tunnel_type & ~TNL_T_KEY_MATCH; if (key_local_remote_ports) { - tbl_node = tbl_lookup(table, &lookup, sizeof(lookup), - port_hash(&lookup), port_cmp); - if (tbl_node) + tnl_vport = port_table_lookup(&lookup); + if (tnl_vport) goto found; } if (key_remote_ports) { lookup.saddr = 0; - - tbl_node = tbl_lookup(table, &lookup, sizeof(lookup), - port_hash(&lookup), port_cmp); - if (tbl_node) + tnl_vport = port_table_lookup(&lookup); + if (tnl_vport) goto found; lookup.saddr = saddr; @@ -360,18 +314,15 @@ struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be64 key, lookup.tunnel_type = tunnel_type & ~TNL_T_KEY_EXACT; if (local_remote_ports) { - tbl_node = tbl_lookup(table, &lookup, sizeof(lookup), - port_hash(&lookup), port_cmp); - if (tbl_node) + tnl_vport = port_table_lookup(&lookup); + if (tnl_vport) goto found; } if (remote_ports) { lookup.saddr = 0; - - tbl_node = tbl_lookup(table, &lookup, sizeof(lookup), - port_hash(&lookup), port_cmp); - if (tbl_node) + tnl_vport = port_table_lookup(&lookup); + if (tnl_vport) goto found; } } @@ -380,7 +331,7 @@ struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be64 key, found: *mutable = lookup.mutable; - return tnl_vport_to_vport(tnl_vport_table_cast(tbl_node)); + return tnl_vport_to_vport(tnl_vport); } static void ecn_decapsulate(struct sk_buff *skb, u8 tos) @@ -450,9 +401,13 @@ void tnl_rcv(struct vport *vport, struct sk_buff *skb, u8 tos) secpath_reset(skb); ecn_decapsulate(skb, tos); - compute_ip_summed(skb, false); vlan_set_tci(skb, 0); + if (unlikely(compute_ip_summed(skb, false))) { + kfree_skb(skb); + return; + } + vport_receive(vport, skb); } @@ -718,7 +673,11 @@ bool tnl_frag_needed(struct vport *vport, const struct tnl_mutable_config *mutab (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION)) OVS_CB(nskb)->tun_id = flow_key; - compute_ip_summed(nskb, false); + if (unlikely(compute_ip_summed(nskb, false))) { + kfree_skb(nskb); + return false; + } + vport_receive(vport, nskb); return true; @@ -838,10 +797,10 @@ static inline bool check_cache_valid(const struct tnl_cache *cache, (cache->flow && !cache->flow->dead)); } -static int cache_cleaner_cb(struct tbl_node *tbl_node, void *aux) +static void __cache_cleaner(struct tnl_vport *tnl_vport) { - 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_mutable_config *mutable = + rcu_dereference(tnl_vport->mutable); const struct tnl_cache *cache = rcu_dereference(tnl_vport->cache); if (cache && !check_cache_valid(cache, mutable) && @@ -849,16 +808,24 @@ static int cache_cleaner_cb(struct tbl_node *tbl_node, void *aux) assign_cache_rcu(tnl_vport_to_vport(tnl_vport), NULL); spin_unlock_bh(&tnl_vport->cache_lock); } - - return 0; } static void cache_cleaner(struct work_struct *work) { + int i; + schedule_cache_cleaner(); rcu_read_lock(); - tbl_foreach(rcu_dereference(port_table), cache_cleaner_cb, NULL); + for (i = 0; i < PORT_TABLE_SIZE; i++) { + struct hlist_node *n; + struct hlist_head *bucket; + struct tnl_vport *tnl_vport; + + bucket = &port_table[i]; + hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node) + __cache_cleaner(tnl_vport); + } rcu_read_unlock(); } @@ -939,12 +906,12 @@ static struct tnl_cache *build_cache(struct vport *vport, if (is_internal_dev(rt_dst(rt).dev)) { struct sw_flow_key flow_key; - struct tbl_node *flow_node; struct vport *dst_vport; struct sk_buff *skb; bool is_frag; int err; int flow_key_len; + struct sw_flow *flow; dst_vport = internal_dev_get_vport(rt_dst(rt).dev); if (!dst_vport) @@ -960,17 +927,13 @@ static struct tnl_cache *build_cache(struct vport *vport, err = flow_extract(skb, dst_vport->port_no, &flow_key, &flow_key_len, &is_frag); - kfree_skb(skb); + consume_skb(skb); if (err || is_frag) goto done; - flow_node = tbl_lookup(rcu_dereference(dst_vport->dp->table), - &flow_key, flow_key_len, - flow_hash(&flow_key, flow_key_len), - flow_cmp); - if (flow_node) { - struct sw_flow *flow = flow_cast(flow_node); - + flow = flow_tbl_lookup(rcu_dereference(dst_vport->dp->table), + &flow_key, flow_key_len); + if (flow) { cache->flow = flow; flow_hold(flow); } @@ -1053,12 +1016,6 @@ static struct sk_buff *handle_offloads(struct sk_buff *skb, int min_headroom; int err; - forward_ip_summed(skb); - - err = vswitch_skb_checksum_setup(skb); - if (unlikely(err)) - goto error_free; - min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len + mutable->tunnel_hlen + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0); @@ -1073,18 +1030,21 @@ static struct sk_buff *handle_offloads(struct sk_buff *skb, goto error_free; } + forward_ip_summed(skb, true); + if (skb_is_gso(skb)) { struct sk_buff *nskb; nskb = skb_gso_segment(skb, 0); - kfree_skb(skb); if (IS_ERR(nskb)) { + kfree_skb(skb); err = PTR_ERR(nskb); goto error; } + consume_skb(skb); skb = nskb; - } else if (skb->ip_summed == CHECKSUM_PARTIAL) { + } else if (get_ip_summed(skb) == OVS_CSUM_PARTIAL) { /* Pages aren't locked and could change at any time. * If this happens after we compute the checksum, the * checksum will be wrong. We linearize now to avoid @@ -1099,8 +1059,9 @@ static struct sk_buff *handle_offloads(struct sk_buff *skb, err = skb_checksum_help(skb); if (unlikely(err)) goto error_free; - } else if (skb->ip_summed == CHECKSUM_COMPLETE) - skb->ip_summed = CHECKSUM_NONE; + } + + set_ip_summed(skb, OVS_CSUM_NONE); return skb; @@ -1295,8 +1256,12 @@ int tnl_send(struct vport *vport, struct sk_buff *skb) ip_send_check(iph); if (cache_vport) { + if (unlikely(compute_ip_summed(skb, true))) { + kfree_skb(skb); + goto next; + } + OVS_CB(skb)->flow = cache->flow; - compute_ip_summed(skb, true); vport_receive(cache_vport, skb); sent_len += orig_len; } else { @@ -1329,64 +1294,64 @@ out: return sent_len; } -static const struct nla_policy tnl_policy[ODP_TUNNEL_ATTR_MAX + 1] = { - [ODP_TUNNEL_ATTR_FLAGS] = { .type = NLA_U32 }, - [ODP_TUNNEL_ATTR_DST_IPV4] = { .type = NLA_U32 }, - [ODP_TUNNEL_ATTR_SRC_IPV4] = { .type = NLA_U32 }, - [ODP_TUNNEL_ATTR_OUT_KEY] = { .type = NLA_U64 }, - [ODP_TUNNEL_ATTR_IN_KEY] = { .type = NLA_U64 }, - [ODP_TUNNEL_ATTR_TOS] = { .type = NLA_U8 }, - [ODP_TUNNEL_ATTR_TTL] = { .type = NLA_U8 }, +static const struct nla_policy tnl_policy[OVS_TUNNEL_ATTR_MAX + 1] = { + [OVS_TUNNEL_ATTR_FLAGS] = { .type = NLA_U32 }, + [OVS_TUNNEL_ATTR_DST_IPV4] = { .type = NLA_U32 }, + [OVS_TUNNEL_ATTR_SRC_IPV4] = { .type = NLA_U32 }, + [OVS_TUNNEL_ATTR_OUT_KEY] = { .type = NLA_U64 }, + [OVS_TUNNEL_ATTR_IN_KEY] = { .type = NLA_U64 }, + [OVS_TUNNEL_ATTR_TOS] = { .type = NLA_U8 }, + [OVS_TUNNEL_ATTR_TTL] = { .type = NLA_U8 }, }; -/* Sets ODP_TUNNEL_ATTR_* fields in 'mutable', which must initially be zeroed. */ +/* Sets OVS_TUNNEL_ATTR_* fields in 'mutable', which must initially be zeroed. */ static int tnl_set_config(struct nlattr *options, 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; - struct nlattr *a[ODP_TUNNEL_ATTR_MAX + 1]; + struct nlattr *a[OVS_TUNNEL_ATTR_MAX + 1]; int err; if (!options) return -EINVAL; - err = nla_parse_nested(a, ODP_TUNNEL_ATTR_MAX, options, tnl_policy); + err = nla_parse_nested(a, OVS_TUNNEL_ATTR_MAX, options, tnl_policy); if (err) return err; - if (!a[ODP_TUNNEL_ATTR_FLAGS] || !a[ODP_TUNNEL_ATTR_DST_IPV4]) + if (!a[OVS_TUNNEL_ATTR_FLAGS] || !a[OVS_TUNNEL_ATTR_DST_IPV4]) return -EINVAL; - mutable->flags = nla_get_u32(a[ODP_TUNNEL_ATTR_FLAGS]) & TNL_F_PUBLIC; + mutable->flags = nla_get_u32(a[OVS_TUNNEL_ATTR_FLAGS]) & TNL_F_PUBLIC; - if (a[ODP_TUNNEL_ATTR_SRC_IPV4]) - mutable->saddr = nla_get_be32(a[ODP_TUNNEL_ATTR_SRC_IPV4]); - mutable->daddr = nla_get_be32(a[ODP_TUNNEL_ATTR_DST_IPV4]); + if (a[OVS_TUNNEL_ATTR_SRC_IPV4]) + mutable->saddr = nla_get_be32(a[OVS_TUNNEL_ATTR_SRC_IPV4]); + mutable->daddr = nla_get_be32(a[OVS_TUNNEL_ATTR_DST_IPV4]); - if (a[ODP_TUNNEL_ATTR_TOS]) { - mutable->tos = nla_get_u8(a[ODP_TUNNEL_ATTR_TOS]); + if (a[OVS_TUNNEL_ATTR_TOS]) { + mutable->tos = nla_get_u8(a[OVS_TUNNEL_ATTR_TOS]); if (mutable->tos != RT_TOS(mutable->tos)) return -EINVAL; } - if (a[ODP_TUNNEL_ATTR_TTL]) - mutable->ttl = nla_get_u8(a[ODP_TUNNEL_ATTR_TTL]); + if (a[OVS_TUNNEL_ATTR_TTL]) + mutable->ttl = nla_get_u8(a[OVS_TUNNEL_ATTR_TTL]); mutable->tunnel_type = tnl_ops->tunnel_type; - if (!a[ODP_TUNNEL_ATTR_IN_KEY]) { + if (!a[OVS_TUNNEL_ATTR_IN_KEY]) { mutable->tunnel_type |= TNL_T_KEY_MATCH; mutable->flags |= TNL_F_IN_KEY_MATCH; } else { mutable->tunnel_type |= TNL_T_KEY_EXACT; - mutable->in_key = nla_get_be64(a[ODP_TUNNEL_ATTR_IN_KEY]); + mutable->in_key = nla_get_be64(a[OVS_TUNNEL_ATTR_IN_KEY]); } - if (!a[ODP_TUNNEL_ATTR_OUT_KEY]) + if (!a[OVS_TUNNEL_ATTR_OUT_KEY]) mutable->flags |= TNL_F_OUT_KEY_ACTION; else - mutable->out_key = nla_get_be64(a[ODP_TUNNEL_ATTR_OUT_KEY]); + mutable->out_key = nla_get_be64(a[OVS_TUNNEL_ATTR_OUT_KEY]); mutable->tunnel_hlen = tnl_ops->hdr_len(mutable); if (mutable->tunnel_hlen < 0) @@ -1449,10 +1414,7 @@ struct vport *tnl_create(const struct vport_parms *parms, rcu_assign_pointer(tnl_vport->mutable, mutable); - err = add_port(vport); - if (err) - goto error_free_mutable; - + port_table_add_port(vport); return vport; error_free_mutable: @@ -1486,9 +1448,8 @@ int tnl_set_options(struct vport *vport, struct nlattr *options) if (err) goto error_free; - err = move_port(vport, mutable); - if (err) - goto error_free; + if (mutable_hash(mutable) != mutable_hash(old_mutable)) + port_table_move_port(vport, mutable); return 0; @@ -1503,19 +1464,19 @@ int tnl_get_options(const struct vport *vport, struct sk_buff *skb) const struct tnl_vport *tnl_vport = tnl_vport_priv(vport); const struct tnl_mutable_config *mutable = rcu_dereference_rtnl(tnl_vport->mutable); - NLA_PUT_U32(skb, ODP_TUNNEL_ATTR_FLAGS, mutable->flags & TNL_F_PUBLIC); - NLA_PUT_BE32(skb, ODP_TUNNEL_ATTR_DST_IPV4, mutable->daddr); + NLA_PUT_U32(skb, OVS_TUNNEL_ATTR_FLAGS, mutable->flags & TNL_F_PUBLIC); + NLA_PUT_BE32(skb, OVS_TUNNEL_ATTR_DST_IPV4, mutable->daddr); if (!(mutable->flags & TNL_F_IN_KEY_MATCH)) - NLA_PUT_BE64(skb, ODP_TUNNEL_ATTR_IN_KEY, mutable->in_key); + NLA_PUT_BE64(skb, OVS_TUNNEL_ATTR_IN_KEY, mutable->in_key); if (!(mutable->flags & TNL_F_OUT_KEY_ACTION)) - NLA_PUT_BE64(skb, ODP_TUNNEL_ATTR_OUT_KEY, mutable->out_key); + NLA_PUT_BE64(skb, OVS_TUNNEL_ATTR_OUT_KEY, mutable->out_key); if (mutable->saddr) - NLA_PUT_BE32(skb, ODP_TUNNEL_ATTR_SRC_IPV4, mutable->saddr); + NLA_PUT_BE32(skb, OVS_TUNNEL_ATTR_SRC_IPV4, mutable->saddr); if (mutable->tos) - NLA_PUT_U8(skb, ODP_TUNNEL_ATTR_TOS, mutable->tos); + NLA_PUT_U8(skb, OVS_TUNNEL_ATTR_TOS, mutable->tos); if (mutable->ttl) - NLA_PUT_U8(skb, ODP_TUNNEL_ATTR_TTL, mutable->ttl); + NLA_PUT_U8(skb, OVS_TUNNEL_ATTR_TTL, mutable->ttl); return 0; @@ -1533,21 +1494,14 @@ static void free_port_rcu(struct rcu_head *rcu) vport_free(tnl_vport_to_vport(tnl_vport)); } -int tnl_destroy(struct vport *vport) +void tnl_destroy(struct vport *vport) { struct tnl_vport *tnl_vport = tnl_vport_priv(vport); - const struct tnl_mutable_config *mutable, *old_mutable; + const struct tnl_mutable_config *mutable; mutable = rtnl_dereference(tnl_vport->mutable); - - if (vport == tnl_find_port(mutable->saddr, mutable->daddr, - mutable->in_key, mutable->tunnel_type, - &old_mutable)) - del_port(vport); - + port_table_remove_port(vport); call_rcu(&tnl_vport->rcu, free_port_rcu); - - return 0; } int tnl_set_addr(struct vport *vport, const unsigned char *addr) @@ -1586,3 +1540,37 @@ void tnl_free_linked_skbs(struct sk_buff *skb) skb = next; } } + +int tnl_init(void) +{ + int i; + + port_table = kmalloc(PORT_TABLE_SIZE * sizeof(struct hlist_head *), + GFP_KERNEL); + if (!port_table) + return -ENOMEM; + + for (i = 0; i < PORT_TABLE_SIZE; i++) + INIT_HLIST_HEAD(&port_table[i]); + + return 0; +} + +void tnl_exit(void) +{ + int i; + + for (i = 0; i < PORT_TABLE_SIZE; i++) { + struct tnl_vport * tnl_vport; + struct hlist_head *hash_head; + struct hlist_node *n; + + hash_head = &port_table[i]; + hlist_for_each_entry(tnl_vport, n, hash_head, hash_node) { + BUG(); + goto out; + } + } +out: + kfree(port_table); +}