X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=ofproto%2Fin-band.c;h=42b3efbed43c367efa561599842dd8f22fc797f8;hb=HEAD;hp=e04ae60926561dbeea843c475d12bc2bf62c1981;hpb=b2fda3effc787f265b5ad5dfa967ac00627bd075;p=sliver-openvswitch.git diff --git a/ofproto/in-band.c b/ofproto/in-band.c index e04ae6092..42b3efbed 100644 --- a/ofproto/in-band.c +++ b/ofproto/in-band.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks. + * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 Nicira, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -25,17 +25,17 @@ #include #include "classifier.h" #include "dhcp.h" -#include "dpif.h" #include "flow.h" #include "netdev.h" #include "netlink.h" #include "odp-util.h" +#include "ofp-actions.h" #include "ofproto.h" #include "ofpbuf.h" +#include "ofproto-provider.h" #include "openflow/openflow.h" #include "packets.h" #include "poll-loop.h" -#include "private.h" #include "timeval.h" #include "vlog.h" @@ -73,9 +73,23 @@ struct in_band_remote { struct netdev *remote_netdev; /* Device to send to next-hop MAC. */ }; +/* What to do to an in_band_rule. */ +enum in_band_op { + ADD, /* Add the rule to ofproto's flow table. */ + DEL /* Delete the rule from ofproto's flow table. */ +}; + +/* A rule to add to or delete from ofproto's flow table. */ +struct in_band_rule { + struct hmap_node hmap_node; /* In struct in_band's "rules" hmap. */ + struct match match; + unsigned int priority; + enum in_band_op op; +}; + struct in_band { struct ofproto *ofproto; - int queue_id, prev_queue_id; + int queue_id; /* Remote information. */ time_t next_remote_refresh; /* Refresh timer. */ @@ -87,12 +101,8 @@ struct in_band { uint8_t local_mac[ETH_ADDR_LEN]; /* Current MAC. */ struct netdev *local_netdev; /* Local port's network device. */ - /* Local and remote addresses that are installed as flows. */ - uint8_t installed_local_mac[ETH_ADDR_LEN]; - struct sockaddr_in *remote_addrs; - size_t n_remote_addrs; - uint8_t *remote_macs; - size_t n_remote_macs; + /* Flow tracking. */ + struct hmap rules; /* Contains "struct in_band_rule"s. */ }; static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 60); @@ -110,7 +120,8 @@ refresh_remote(struct in_band *ib, struct in_band_remote *r) &next_hop_inaddr, &next_hop_dev); if (retval) { VLOG_WARN("cannot find route for controller ("IP_FMT"): %s", - IP_ARGS(&r->remote_addr.sin_addr), strerror(retval)); + IP_ARGS(r->remote_addr.sin_addr.s_addr), + ovs_strerror(retval)); return 1; } if (!next_hop_inaddr.s_addr) { @@ -123,12 +134,12 @@ refresh_remote(struct in_band *ib, struct in_band_remote *r) { netdev_close(r->remote_netdev); - retval = netdev_open_default(next_hop_dev, &r->remote_netdev); + retval = netdev_open(next_hop_dev, "system", &r->remote_netdev); if (retval) { VLOG_WARN_RL(&rl, "cannot open netdev %s (next hop " "to controller "IP_FMT"): %s", - next_hop_dev, IP_ARGS(&r->remote_addr.sin_addr), - strerror(retval)); + next_hop_dev, IP_ARGS(r->remote_addr.sin_addr.s_addr), + ovs_strerror(retval)); free(next_hop_dev); return 1; } @@ -140,7 +151,7 @@ refresh_remote(struct in_band *ib, struct in_band_remote *r) r->remote_mac); if (retval) { VLOG_DBG_RL(&rl, "cannot look up remote MAC address ("IP_FMT"): %s", - IP_ARGS(&next_hop_inaddr.s_addr), strerror(retval)); + IP_ARGS(next_hop_inaddr.s_addr), ovs_strerror(retval)); } /* If we don't have a MAC address, then refresh quickly, since we probably @@ -212,292 +223,183 @@ refresh_local(struct in_band *ib) return true; } -/* Returns true if 'packet' should be sent to the local port regardless - * of the flow table. */ +/* Returns true if packets in 'flow' should be directed to the local port. + * (This keeps the flow table from preventing DHCP replies from being seen by + * the local port.) */ bool -in_band_msg_in_hook(struct in_band *in_band, const struct flow *flow, - const struct ofpbuf *packet) +in_band_must_output_to_local_port(const struct flow *flow) { - /* Regardless of how the flow table is configured, we want to be - * able to see replies to our DHCP requests. */ - if (flow->dl_type == htons(ETH_TYPE_IP) + return (flow->dl_type == htons(ETH_TYPE_IP) && flow->nw_proto == IPPROTO_UDP && flow->tp_src == htons(DHCP_SERVER_PORT) - && flow->tp_dst == htons(DHCP_CLIENT_PORT) - && packet->l7) { - struct dhcp_header *dhcp; - - dhcp = ofpbuf_at(packet, (char *)packet->l7 - (char *)packet->data, - sizeof *dhcp); - if (!dhcp) { - return false; - } - - refresh_local(in_band); - if (!eth_addr_is_zero(in_band->local_mac) - && eth_addr_equals(dhcp->chaddr, in_band->local_mac)) { - return true; - } - } - - return false; + && flow->tp_dst == htons(DHCP_CLIENT_PORT)); } -/* Returns true if the rule that would match 'flow' with 'actions' is - * allowed to be set up in the datapath. */ -bool -in_band_rule_check(const struct flow *flow, - const struct nlattr *actions, size_t actions_len) +static void +add_rule(struct in_band *ib, const struct match *match, unsigned int priority) { - /* Don't allow flows that would prevent DHCP replies from being seen - * by the local port. */ - if (flow->dl_type == htons(ETH_TYPE_IP) - && flow->nw_proto == IPPROTO_UDP - && flow->tp_src == htons(DHCP_SERVER_PORT) - && flow->tp_dst == htons(DHCP_CLIENT_PORT)) { - const struct nlattr *a; - unsigned int left; - - NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) { - if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT - && nl_attr_get_u32(a) == ODPP_LOCAL) { - return true; - } + uint32_t hash = match_hash(match, 0); + struct in_band_rule *rule; + + HMAP_FOR_EACH_WITH_HASH (rule, hmap_node, hash, &ib->rules) { + if (match_equal(&rule->match, match)) { + rule->op = ADD; + return; } - return false; } - return true; + rule = xmalloc(sizeof *rule); + rule->match = *match; + rule->priority = priority; + rule->op = ADD; + hmap_insert(&ib->rules, &rule->hmap_node, hash); } static void -make_rules(struct in_band *ib, - void (*cb)(struct in_band *, const struct cls_rule *)) +update_rules(struct in_band *ib) { - struct cls_rule rule; - size_t i; + struct in_band_rule *ib_rule; + struct in_band_remote *r; + struct match match; + + /* Mark all the existing rules for deletion. (Afterward we will re-add any + * rules that are still valid.) */ + HMAP_FOR_EACH (ib_rule, hmap_node, &ib->rules) { + ib_rule->op = DEL; + } - if (!eth_addr_is_zero(ib->installed_local_mac)) { + if (ib->n_remotes && !eth_addr_is_zero(ib->local_mac)) { /* (a) Allow DHCP requests sent from the local port. */ - cls_rule_init_catchall(&rule, IBR_FROM_LOCAL_DHCP); - cls_rule_set_in_port(&rule, ODPP_LOCAL); - cls_rule_set_dl_type(&rule, htons(ETH_TYPE_IP)); - cls_rule_set_dl_src(&rule, ib->installed_local_mac); - cls_rule_set_nw_proto(&rule, IPPROTO_UDP); - cls_rule_set_tp_src(&rule, htons(DHCP_CLIENT_PORT)); - cls_rule_set_tp_dst(&rule, htons(DHCP_SERVER_PORT)); - cb(ib, &rule); + match_init_catchall(&match); + match_set_in_port(&match, OFPP_LOCAL); + match_set_dl_type(&match, htons(ETH_TYPE_IP)); + match_set_dl_src(&match, ib->local_mac); + match_set_nw_proto(&match, IPPROTO_UDP); + match_set_tp_src(&match, htons(DHCP_CLIENT_PORT)); + match_set_tp_dst(&match, htons(DHCP_SERVER_PORT)); + add_rule(ib, &match, IBR_FROM_LOCAL_DHCP); /* (b) Allow ARP replies to the local port's MAC address. */ - cls_rule_init_catchall(&rule, IBR_TO_LOCAL_ARP); - cls_rule_set_dl_type(&rule, htons(ETH_TYPE_ARP)); - cls_rule_set_dl_dst(&rule, ib->installed_local_mac); - cls_rule_set_nw_proto(&rule, ARP_OP_REPLY); - cb(ib, &rule); + match_init_catchall(&match); + match_set_dl_type(&match, htons(ETH_TYPE_ARP)); + match_set_dl_dst(&match, ib->local_mac); + match_set_nw_proto(&match, ARP_OP_REPLY); + add_rule(ib, &match, IBR_TO_LOCAL_ARP); /* (c) Allow ARP requests from the local port's MAC address. */ - cls_rule_init_catchall(&rule, IBR_FROM_LOCAL_ARP); - cls_rule_set_dl_type(&rule, htons(ETH_TYPE_ARP)); - cls_rule_set_dl_src(&rule, ib->installed_local_mac); - cls_rule_set_nw_proto(&rule, ARP_OP_REQUEST); - cb(ib, &rule); + match_init_catchall(&match); + match_set_dl_type(&match, htons(ETH_TYPE_ARP)); + match_set_dl_src(&match, ib->local_mac); + match_set_nw_proto(&match, ARP_OP_REQUEST); + add_rule(ib, &match, IBR_FROM_LOCAL_ARP); } - for (i = 0; i < ib->n_remote_macs; i++) { - const uint8_t *remote_mac = &ib->remote_macs[i * ETH_ADDR_LEN]; + for (r = ib->remotes; r < &ib->remotes[ib->n_remotes]; r++) { + const uint8_t *remote_mac = r->remote_mac; - if (i > 0) { - const uint8_t *prev_mac = &ib->remote_macs[(i - 1) * ETH_ADDR_LEN]; - if (eth_addr_equals(remote_mac, prev_mac)) { - /* Skip duplicates. */ - continue; - } + if (eth_addr_is_zero(remote_mac)) { + continue; } /* (d) Allow ARP replies to the next hop's MAC address. */ - cls_rule_init_catchall(&rule, IBR_TO_NEXT_HOP_ARP); - cls_rule_set_dl_type(&rule, htons(ETH_TYPE_ARP)); - cls_rule_set_dl_dst(&rule, remote_mac); - cls_rule_set_nw_proto(&rule, ARP_OP_REPLY); - cb(ib, &rule); + match_init_catchall(&match); + match_set_dl_type(&match, htons(ETH_TYPE_ARP)); + match_set_dl_dst(&match, remote_mac); + match_set_nw_proto(&match, ARP_OP_REPLY); + add_rule(ib, &match, IBR_TO_NEXT_HOP_ARP); /* (e) Allow ARP requests from the next hop's MAC address. */ - cls_rule_init_catchall(&rule, IBR_FROM_NEXT_HOP_ARP); - cls_rule_set_dl_type(&rule, htons(ETH_TYPE_ARP)); - cls_rule_set_dl_src(&rule, remote_mac); - cls_rule_set_nw_proto(&rule, ARP_OP_REQUEST); - cb(ib, &rule); - } - - for (i = 0; i < ib->n_remote_addrs; i++) { - const struct sockaddr_in *a = &ib->remote_addrs[i]; - - if (!i || a->sin_addr.s_addr != a[-1].sin_addr.s_addr) { - /* (f) Allow ARP replies containing the remote's IP address as a - * target. */ - cls_rule_init_catchall(&rule, IBR_TO_REMOTE_ARP); - cls_rule_set_dl_type(&rule, htons(ETH_TYPE_ARP)); - cls_rule_set_nw_proto(&rule, ARP_OP_REPLY); - cls_rule_set_nw_dst(&rule, a->sin_addr.s_addr); - cb(ib, &rule); - - /* (g) Allow ARP requests containing the remote's IP address as a - * source. */ - cls_rule_init_catchall(&rule, IBR_FROM_REMOTE_ARP); - cls_rule_set_dl_type(&rule, htons(ETH_TYPE_ARP)); - cls_rule_set_nw_proto(&rule, ARP_OP_REQUEST); - cls_rule_set_nw_src(&rule, a->sin_addr.s_addr); - cb(ib, &rule); - } - - if (!i - || a->sin_addr.s_addr != a[-1].sin_addr.s_addr - || a->sin_port != a[-1].sin_port) { - /* (h) Allow TCP traffic to the remote's IP and port. */ - cls_rule_init_catchall(&rule, IBR_TO_REMOTE_TCP); - cls_rule_set_dl_type(&rule, htons(ETH_TYPE_IP)); - cls_rule_set_nw_proto(&rule, IPPROTO_TCP); - cls_rule_set_nw_dst(&rule, a->sin_addr.s_addr); - cls_rule_set_tp_dst(&rule, a->sin_port); - cb(ib, &rule); - - /* (i) Allow TCP traffic from the remote's IP and port. */ - cls_rule_init_catchall(&rule, IBR_FROM_REMOTE_TCP); - cls_rule_set_dl_type(&rule, htons(ETH_TYPE_IP)); - cls_rule_set_nw_proto(&rule, IPPROTO_TCP); - cls_rule_set_nw_src(&rule, a->sin_addr.s_addr); - cls_rule_set_tp_src(&rule, a->sin_port); - cb(ib, &rule); - } - } -} - -static void -drop_rule(struct in_band *ib, const struct cls_rule *rule) -{ - ofproto_delete_flow(ib->ofproto, rule); -} - -/* Drops from the flow table all of the flows set up by 'ib', then clears out - * the information about the installed flows so that they can be filled in - * again if necessary. */ -static void -drop_rules(struct in_band *ib) -{ - /* Drop rules. */ - make_rules(ib, drop_rule); - - /* Clear out state. */ - memset(ib->installed_local_mac, 0, sizeof ib->installed_local_mac); - - free(ib->remote_addrs); - ib->remote_addrs = NULL; - ib->n_remote_addrs = 0; - - free(ib->remote_macs); - ib->remote_macs = NULL; - ib->n_remote_macs = 0; -} - -static void -add_rule(struct in_band *ib, const struct cls_rule *rule) -{ - struct { - struct nx_action_set_queue nxsq; - struct ofp_action_output oao; - } actions; - - memset(&actions, 0, sizeof actions); - - actions.oao.type = htons(OFPAT_OUTPUT); - actions.oao.len = htons(sizeof actions.oao); - actions.oao.port = htons(OFPP_NORMAL); - actions.oao.max_len = htons(0); - - if (ib->queue_id < 0) { - ofproto_add_flow(ib->ofproto, rule, - (union ofp_action *) &actions.oao, 1); - } else { - actions.nxsq.type = htons(OFPAT_VENDOR); - actions.nxsq.len = htons(sizeof actions.nxsq); - actions.nxsq.vendor = htonl(NX_VENDOR_ID); - actions.nxsq.subtype = htons(NXAST_SET_QUEUE); - actions.nxsq.queue_id = htonl(ib->queue_id); - - ofproto_add_flow(ib->ofproto, rule, (union ofp_action *) &actions, - sizeof actions / sizeof(union ofp_action)); + match_init_catchall(&match); + match_set_dl_type(&match, htons(ETH_TYPE_ARP)); + match_set_dl_src(&match, remote_mac); + match_set_nw_proto(&match, ARP_OP_REQUEST); + add_rule(ib, &match, IBR_FROM_NEXT_HOP_ARP); } -} -/* Inserts flows into the flow table for the current state of 'ib'. */ -static void -add_rules(struct in_band *ib) -{ - make_rules(ib, add_rule); -} - -static int -compare_addrs(const void *a_, const void *b_) -{ - const struct sockaddr_in *a = a_; - const struct sockaddr_in *b = b_; - int cmp; - - cmp = memcmp(&a->sin_addr.s_addr, - &b->sin_addr.s_addr, - sizeof a->sin_addr.s_addr); - if (cmp) { - return cmp; + for (r = ib->remotes; r < &ib->remotes[ib->n_remotes]; r++) { + const struct sockaddr_in *a = &r->remote_addr; + + /* (f) Allow ARP replies containing the remote's IP address as a + * target. */ + match_init_catchall(&match); + match_set_dl_type(&match, htons(ETH_TYPE_ARP)); + match_set_nw_proto(&match, ARP_OP_REPLY); + match_set_nw_dst(&match, a->sin_addr.s_addr); + add_rule(ib, &match, IBR_TO_REMOTE_ARP); + + /* (g) Allow ARP requests containing the remote's IP address as a + * source. */ + match_init_catchall(&match); + match_set_dl_type(&match, htons(ETH_TYPE_ARP)); + match_set_nw_proto(&match, ARP_OP_REQUEST); + match_set_nw_src(&match, a->sin_addr.s_addr); + add_rule(ib, &match, IBR_FROM_REMOTE_ARP); + + /* (h) Allow TCP traffic to the remote's IP and port. */ + match_init_catchall(&match); + match_set_dl_type(&match, htons(ETH_TYPE_IP)); + match_set_nw_proto(&match, IPPROTO_TCP); + match_set_nw_dst(&match, a->sin_addr.s_addr); + match_set_tp_dst(&match, a->sin_port); + add_rule(ib, &match, IBR_TO_REMOTE_TCP); + + /* (i) Allow TCP traffic from the remote's IP and port. */ + match_init_catchall(&match); + match_set_dl_type(&match, htons(ETH_TYPE_IP)); + match_set_nw_proto(&match, IPPROTO_TCP); + match_set_nw_src(&match, a->sin_addr.s_addr); + match_set_tp_src(&match, a->sin_port); + add_rule(ib, &match, IBR_FROM_REMOTE_TCP); } - return memcmp(&a->sin_port, &b->sin_port, sizeof a->sin_port); } -static int -compare_macs(const void *a, const void *b) -{ - return eth_addr_compare_3way(a, b); -} - -void +/* Updates the OpenFlow flow table for the current state of in-band control. + * Returns true ordinarily. Returns false if no remotes are configured on 'ib' + * and 'ib' doesn't have any rules left to remove from the OpenFlow flow + * table. Thus, a false return value means that the caller can destroy 'ib' + * without leaving extra flows hanging around in the flow table. */ +bool in_band_run(struct in_band *ib) { - bool local_change, remote_change, queue_id_change; - struct in_band_remote *r; + uint64_t ofpacts_stub[128 / 8]; + struct ofpbuf ofpacts; - local_change = refresh_local(ib); - remote_change = refresh_remotes(ib); - queue_id_change = ib->queue_id != ib->prev_queue_id; - if (!local_change && !remote_change && !queue_id_change) { - /* Nothing changed, nothing to do. */ - return; - } - ib->prev_queue_id = ib->queue_id; + struct in_band_rule *rule, *next; - /* Drop old rules. */ - drop_rules(ib); + ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); - /* Figure out new rules. */ - memcpy(ib->installed_local_mac, ib->local_mac, ETH_ADDR_LEN); - ib->remote_addrs = xmalloc(ib->n_remotes * sizeof *ib->remote_addrs); - ib->n_remote_addrs = 0; - ib->remote_macs = xmalloc(ib->n_remotes * ETH_ADDR_LEN); - ib->n_remote_macs = 0; - for (r = ib->remotes; r < &ib->remotes[ib->n_remotes]; r++) { - ib->remote_addrs[ib->n_remote_addrs++] = r->remote_addr; - if (!eth_addr_is_zero(r->remote_mac)) { - memcpy(&ib->remote_macs[ib->n_remote_macs * ETH_ADDR_LEN], - r->remote_mac, ETH_ADDR_LEN); - ib->n_remote_macs++; + if (ib->queue_id >= 0) { + ofpact_put_SET_QUEUE(&ofpacts)->queue_id = ib->queue_id; + } + ofpact_put_OUTPUT(&ofpacts)->port = OFPP_NORMAL; + + refresh_local(ib); + refresh_remotes(ib); + + update_rules(ib); + + HMAP_FOR_EACH_SAFE (rule, next, hmap_node, &ib->rules) { + switch (rule->op) { + case ADD: + ofproto_add_flow(ib->ofproto, &rule->match, rule->priority, + ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts)); + break; + + case DEL: + if (ofproto_delete_flow(ib->ofproto, + &rule->match, rule->priority)) { + /* ofproto doesn't have the rule anymore so there's no reason + * for us to track it any longer. */ + hmap_remove(&ib->rules, &rule->hmap_node); + free(rule); + } + break; } } - /* Sort, to allow make_rules() to easily skip duplicates. */ - qsort(ib->remote_addrs, ib->n_remote_addrs, sizeof *ib->remote_addrs, - compare_addrs); - qsort(ib->remote_macs, ib->n_remote_macs, ETH_ADDR_LEN, compare_macs); + ofpbuf_uninit(&ofpacts); - /* Add new rules. */ - add_rules(ib); + return ib->n_remotes || !hmap_is_empty(&ib->rules); } void @@ -508,14 +410,6 @@ in_band_wait(struct in_band *in_band) poll_timer_wait_until(wakeup * 1000); } -/* ofproto has flushed all flows from the flow table and it is calling us back - * to allow us to reinstall the ones that are important to us. */ -void -in_band_flushed(struct in_band *in_band) -{ - add_rules(in_band); -} - int in_band_create(struct ofproto *ofproto, const char *local_name, struct in_band **in_bandp) @@ -525,19 +419,21 @@ in_band_create(struct ofproto *ofproto, const char *local_name, int error; *in_bandp = NULL; - error = netdev_open_default(local_name, &local_netdev); + error = netdev_open(local_name, "internal", &local_netdev); if (error) { VLOG_ERR("failed to initialize in-band control: cannot open " - "datapath local port %s (%s)", local_name, strerror(error)); + "datapath local port %s (%s)", + local_name, ovs_strerror(error)); return error; } in_band = xzalloc(sizeof *in_band); in_band->ofproto = ofproto; - in_band->queue_id = in_band->prev_queue_id = -1; + in_band->queue_id = -1; in_band->next_remote_refresh = TIME_MIN; in_band->next_local_refresh = TIME_MIN; in_band->local_netdev = local_netdev; + hmap_init(&in_band->rules); *in_bandp = in_band; @@ -548,7 +444,13 @@ void in_band_destroy(struct in_band *ib) { if (ib) { - drop_rules(ib); + struct in_band_rule *rule, *next; + + HMAP_FOR_EACH_SAFE (rule, next, hmap_node, &ib->rules) { + hmap_remove(&ib->rules, &rule->hmap_node); + free(rule); + } + hmap_destroy(&ib->rules); in_band_set_remotes(ib, NULL, 0); netdev_close(ib->local_netdev); free(ib);