X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fbond.c;h=bb5d49968b70186e389ff4b4cacff4c7aa0d9a21;hb=579a77e024b93ba5dfb840468c2fcd804e576d7b;hp=9caa3f2efb745c6bd33cf87722c4d07f252c8349;hpb=b2fda3effc787f265b5ad5dfa967ac00627bd075;p=sliver-openvswitch.git diff --git a/lib/bond.c b/lib/bond.c index 9caa3f2ef..bb5d49968 100644 --- a/lib/bond.c +++ b/lib/bond.c @@ -59,13 +59,13 @@ struct bond_slave { void *aux; /* Client-provided handle for this slave. */ struct netdev *netdev; /* Network device, owned by the client. */ + unsigned int change_seq; /* Tracks changes in 'netdev'. */ char *name; /* Name (a copy of netdev_get_name(netdev)). */ /* Link status. */ long long delay_expires; /* Time after which 'enabled' may change. */ - bool up; /* Last link status read from netdev. */ bool enabled; /* May be chosen for flows? */ - bool lacp_may_enable; /* LACP considers this interface bondable. */ + bool may_enable; /* Client considers this slave bondable. */ tag_type tag; /* Tag associated with this slave. */ /* Rebalancing info. Used only by bond_rebalance(). */ @@ -104,12 +104,6 @@ struct bond { /* BM_STABLE specific bonding info. */ tag_type stb_tag; /* Tag associated with this bond. */ - /* Monitoring. */ - enum bond_detect_mode detect; /* Link status mode, one of BLSM_*. */ - struct netdev_monitor *monitor; /* detect == BLSM_CARRIER only. */ - long long int miimon_interval; /* Miimon status refresh interval. */ - long long int miimon_next_update; /* Time of next miimon update. */ - /* Legacy compatibility. */ long long int next_fake_iface_update; /* LLONG_MAX if disabled. */ @@ -123,7 +117,6 @@ static struct hmap all_bonds = HMAP_INITIALIZER(&all_bonds); static void bond_entry_reset(struct bond *); static struct bond_slave *bond_slave_lookup(struct bond *, const void *slave_); -static bool bond_is_link_up(struct bond *, struct netdev *); static void bond_enable_slave(struct bond_slave *, bool enable, struct tag_set *); static void bond_link_status_update(struct bond_slave *, struct tag_set *); @@ -178,34 +171,6 @@ bond_mode_to_string(enum bond_mode balance) { NOT_REACHED(); } -/* Attempts to parse 's' as the name of a bond link status detection mode. If - * successful, stores the mode in '*detect' and returns true. Otherwise - * returns false without modifying '*detect'. */ -bool -bond_detect_mode_from_string(enum bond_detect_mode *detect, const char *s) -{ - if (!strcmp(s, bond_detect_mode_to_string(BLSM_CARRIER))) { - *detect = BLSM_CARRIER; - } else if (!strcmp(s, bond_detect_mode_to_string(BLSM_MIIMON))) { - *detect = BLSM_MIIMON; - } else { - return false; - } - return true; -} - -/* Returns a string representing 'detect'. */ -const char * -bond_detect_mode_to_string(enum bond_detect_mode detect) -{ - switch (detect) { - case BLSM_CARRIER: - return "carrier"; - case BLSM_MIIMON: - return "miimon"; - } - NOT_REACHED(); -} /* Creates and returns a new bond whose configuration is initially taken from * 's'. @@ -221,7 +186,6 @@ bond_create(const struct bond_settings *s) hmap_init(&bond->slaves); bond->no_slaves_tag = tag_create_random(); bond->stb_tag = tag_create_random(); - bond->miimon_next_update = LLONG_MAX; bond->next_fake_iface_update = LLONG_MAX; bond_reconfigure(bond, s); @@ -252,9 +216,6 @@ bond_destroy(struct bond *bond) hmap_destroy(&bond->slaves); free(bond->hash); - - netdev_monitor_destroy(bond->monitor); - free(bond->name); free(bond); } @@ -282,8 +243,6 @@ bond_reconfigure(struct bond *bond, const struct bond_settings *s) hmap_insert(&all_bonds, &bond->hmap_node, hash_string(bond->name, 0)); } - bond->detect = s->detect; - bond->miimon_interval = s->miimon_interval; bond->updelay = s->up_delay; bond->downdelay = s->down_delay; bond->rebalance_interval = s->rebalance_interval; @@ -298,25 +257,6 @@ bond_reconfigure(struct bond *bond, const struct bond_settings *s) revalidate = true; } - if (bond->detect == BLSM_CARRIER) { - struct bond_slave *slave; - - if (!bond->monitor) { - bond->monitor = netdev_monitor_create(); - } - - HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { - netdev_monitor_add(bond->monitor, slave->netdev); - } - } else { - netdev_monitor_destroy(bond->monitor); - bond->monitor = NULL; - - if (bond->miimon_next_update == LLONG_MAX) { - bond->miimon_next_update = time_msec() + bond->miimon_interval; - } - } - if (s->fake_iface) { if (bond->next_fake_iface_update == LLONG_MAX) { bond->next_fake_iface_update = time_msec(); @@ -338,17 +278,11 @@ bond_reconfigure(struct bond *bond, const struct bond_settings *s) } static void -bond_slave_set_netdev__(struct bond *bond, struct bond_slave *slave, - struct netdev *netdev) +bond_slave_set_netdev__(struct bond_slave *slave, struct netdev *netdev) { if (slave->netdev != netdev) { - if (bond->monitor) { - if (slave->netdev) { - netdev_monitor_remove(bond->monitor, slave->netdev); - } - netdev_monitor_add(bond->monitor, netdev); - } slave->netdev = netdev; + slave->change_seq = 0; } } @@ -378,12 +312,11 @@ bond_slave_register(struct bond *bond, void *slave_, uint32_t stb_id, slave->bond = bond; slave->aux = slave_; slave->delay_expires = LLONG_MAX; - slave->up = bond_is_link_up(bond, netdev); slave->name = xstrdup(netdev_get_name(netdev)); bond->bond_revalidate = true; slave->enabled = false; - bond_enable_slave(slave, slave->up, NULL); + bond_enable_slave(slave, netdev_get_carrier(netdev), NULL); } if (slave->stb_id != stb_id) { @@ -391,7 +324,7 @@ bond_slave_register(struct bond *bond, void *slave_, uint32_t stb_id, bond->bond_revalidate = true; } - bond_slave_set_netdev__(bond, slave, netdev); + bond_slave_set_netdev__(slave, netdev); free(slave->name); slave->name = xstrdup(netdev_get_name(netdev)); @@ -407,7 +340,7 @@ bond_slave_set_netdev(struct bond *bond, void *slave_, struct netdev *netdev) { struct bond_slave *slave = bond_slave_lookup(bond, slave_); if (slave) { - bond_slave_set_netdev__(bond, slave, netdev); + bond_slave_set_netdev__(slave, netdev); } } @@ -425,10 +358,6 @@ bond_slave_unregister(struct bond *bond, const void *slave_) return; } - if (bond->monitor) { - netdev_monitor_remove(bond->monitor, slave->netdev); - } - bond_enable_slave(slave, false, NULL); del_active = bond->active_slave == slave; @@ -456,13 +385,15 @@ bond_slave_unregister(struct bond *bond, const void *slave_) } } -/* Should be called on each slave in 'bond' before bond_run() to indicate the - * results of lacp_slave_may_enable() on 'slave_'. */ +/* Should be called on each slave in 'bond' before bond_run() to indicate + * whether or not 'slave_' may be enabled. This function is intended to allow + * other protocols to have some impact on bonding decisions. For example LACP + * or high level link monitoring protocols may decide that a given slave should + * not be able to send traffic. */ void -bond_slave_set_lacp_may_enable(struct bond *bond, void *slave_, - bool may_enable) +bond_slave_set_may_enable(struct bond *bond, void *slave_, bool may_enable) { - bond_slave_lookup(bond, slave_)->lacp_may_enable = may_enable; + bond_slave_lookup(bond, slave_)->may_enable = may_enable; } /* Performs periodic maintenance on 'bond'. The caller must provide 'tags' to @@ -477,23 +408,10 @@ bond_run(struct bond *bond, struct tag_set *tags, bool lacp_negotiated) bond->lacp_negotiated = lacp_negotiated; - /* Update link status. */ - if (bond->detect == BLSM_CARRIER - || time_msec() >= bond->miimon_next_update) - { - HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { - slave->up = bond_is_link_up(bond, slave->netdev); - } - bond->miimon_next_update = time_msec() + bond->miimon_interval; - } - - if (bond->monitor) { - netdev_monitor_flush(bond->monitor); - } - /* Enable slaves based on link status and LACP feedback. */ HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { bond_link_status_update(slave, tags); + slave->change_seq = netdev_change_seq(slave->netdev); } if (!bond->active_slave || !bond->active_slave->enabled) { bond_choose_active_slave(bond, tags); @@ -536,16 +454,14 @@ bond_wait(struct bond *bond) { struct bond_slave *slave; - if (bond->detect == BLSM_CARRIER) { - netdev_monitor_poll_wait(bond->monitor); - } else { - poll_timer_wait_until(bond->miimon_next_update); - } - HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { if (slave->delay_expires != LLONG_MAX) { poll_timer_wait_until(slave->delay_expires); } + + if (slave->change_seq != netdev_change_seq(slave->netdev)) { + poll_immediate_wake(); + } } if (bond->next_fake_iface_update != LLONG_MAX) { @@ -568,14 +484,15 @@ bond_wait(struct bond *bond) static bool may_send_learning_packets(const struct bond *bond) { - return !bond->lacp_negotiated && bond->balance != BM_AB; + return !bond->lacp_negotiated && bond->balance != BM_AB && + bond->active_slave; } /* Returns true if 'bond' needs the client to send out packets to assist with * MAC learning on 'bond'. If this function returns true, then the client * should iterate through its MAC learning table for the bridge on which 'bond' * is located. For each MAC that has been learned on a port other than 'bond', - * it should call bond_send_learning_packet(). + * it should call bond_compose_learning_packet(). * * This function will only return true if 'bond' is in SLB mode and LACP is not * negotiated. Otherwise sending learning packets isn't necessary. @@ -591,37 +508,33 @@ bond_should_send_learning_packets(struct bond *bond) /* Sends a gratuitous learning packet on 'bond' from 'eth_src' on 'vlan'. * - * See bond_should_send_learning_packets() for description of usage. */ -int -bond_send_learning_packet(struct bond *bond, - const uint8_t eth_src[ETH_ADDR_LEN], - uint16_t vlan) + * See bond_should_send_learning_packets() for description of usage. The + * caller should send the composed packet on the port associated with + * port_aux and takes ownership of the returned ofpbuf. */ +struct ofpbuf * +bond_compose_learning_packet(struct bond *bond, + const uint8_t eth_src[ETH_ADDR_LEN], + uint16_t vlan, void **port_aux) { struct bond_slave *slave; - struct ofpbuf packet; + struct ofpbuf *packet; struct flow flow; - int error; assert(may_send_learning_packets(bond)); - if (!bond->active_slave) { - /* Nowhere to send the learning packet. */ - return 0; - } memset(&flow, 0, sizeof flow); memcpy(flow.dl_src, eth_src, ETH_ADDR_LEN); slave = choose_output_slave(bond, &flow, vlan); - ofpbuf_init(&packet, 0); - compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177, + packet = ofpbuf_new(0); + compose_benign_packet(packet, "Open vSwitch Bond Failover", 0xf177, eth_src); if (vlan) { - eth_set_vlan_tci(&packet, htons(vlan)); + eth_push_vlan(packet, htons(vlan)); } - error = netdev_send(slave->netdev, &packet); - ofpbuf_uninit(&packet); - return error; + *port_aux = slave->aux; + return packet; } /* Checks whether a packet that arrived on 'slave_' within 'bond', with an @@ -643,10 +556,15 @@ enum bond_verdict bond_check_admissibility(struct bond *bond, const void *slave_, const uint8_t eth_dst[ETH_ADDR_LEN], tag_type *tags) { - /* Admit all packets if LACP has been negotiated, because that means that - * the remote switch is aware of the bond and will "do the right thing". */ + struct bond_slave *slave = bond_slave_lookup(bond, slave_); + + /* LACP bonds have very loose admissibility restrictions because we can + * assume the remote switch is aware of the bond and will "do the right + * thing". However, as a precaution we drop packets on disabled slaves + * because no correctly implemented partner switch should be sending + * packets to them. */ if (bond->lacp_negotiated) { - return BV_ACCEPT; + return slave->enabled ? BV_ACCEPT : BV_DROP; } /* Drop all multicast packets on inactive slaves. */ @@ -657,12 +575,40 @@ bond_check_admissibility(struct bond *bond, const void *slave_, } } - /* Drop all packets for which we have learned a different input port, - * because we probably sent the packet on one slave and got it back on the - * other. Gratuitous ARP packets are an exception to this rule: the host - * has moved to another switch. The exception to the exception is if we - * locked the learning table to avoid reflections on bond slaves. */ - return BV_DROP_IF_MOVED; + switch (bond->balance) { + case BM_AB: + /* Drop all packets which arrive on backup slaves. This is similar to + * how Linux bonding handles active-backup bonds. */ + *tags |= bond_get_active_slave_tag(bond); + if (bond->active_slave != slave) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + + VLOG_DBG_RL(&rl, "active-backup bond received packet on backup" + " slave (%s) destined for " ETH_ADDR_FMT, + slave->name, ETH_ADDR_ARGS(eth_dst)); + return BV_DROP; + } + return BV_ACCEPT; + + case BM_TCP: + /* TCP balancing has degraded to SLB (otherwise the + * bond->lacp_negotiated check above would have processed this). + * + * Fall through. */ + case BM_SLB: + /* Drop all packets for which we have learned a different input port, + * because we probably sent the packet on one slave and got it back on + * the other. Gratuitous ARP packets are an exception to this rule: + * the host has moved to another switch. The exception to the + * exception is if we locked the learning table to avoid reflections on + * bond slaves. */ + return BV_DROP_IF_MOVED; + + case BM_STABLE: + return BV_ACCEPT; + } + + NOT_REACHED(); } /* Returns the slave (registered on 'bond' by bond_slave_register()) to which @@ -704,7 +650,6 @@ void bond_account(struct bond *bond, const struct flow *flow, uint16_t vlan, uint64_t n_bytes) { - if (bond_is_balanced(bond)) { lookup_bond_entry(bond, flow, vlan)->tx_bytes += n_bytes; } @@ -1012,14 +957,6 @@ bond_unixctl_show(struct unixctl_conn *conn, ds_put_format(&ds, "bond-hash-basis: %"PRIu32"\n", bond->basis); - ds_put_format(&ds, "bond-detect-mode: %s\n", - bond->monitor ? "carrier" : "miimon"); - - if (!bond->monitor) { - ds_put_format(&ds, "bond-miimon-interval: %lld\n", - bond->miimon_interval); - } - ds_put_format(&ds, "updelay: %d ms\n", bond->updelay); ds_put_format(&ds, "downdelay: %d ms\n", bond->downdelay); @@ -1033,7 +970,6 @@ bond_unixctl_show(struct unixctl_conn *conn, HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { struct bond_entry *be; - struct flow flow; /* Basic info. */ ds_put_format(&ds, "\nslave %s: %s\n", @@ -1047,15 +983,14 @@ bond_unixctl_show(struct unixctl_conn *conn, slave->delay_expires - time_msec()); } - ds_put_format(&ds, "\tlacp_may_enable: %s\n", - slave->lacp_may_enable ? "true" : "false"); + ds_put_format(&ds, "\tmay_enable: %s\n", + slave->may_enable ? "true" : "false"); if (!bond_is_balanced(bond)) { continue; } /* Hashes. */ - memset(&flow, 0, sizeof flow); for (be = bond->hash; be <= &bond->hash[BOND_MASK]; be++) { int hash = be - bond->hash; @@ -1066,11 +1001,7 @@ bond_unixctl_show(struct unixctl_conn *conn, ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n", hash, be->tx_bytes / 1024); - if (bond->balance != BM_SLB) { - continue; - } - - /* XXX How can we list the MACs assigned to hashes? */ + /* XXX How can we list the MACs assigned to hashes of SLB bonds? */ } } unixctl_command_reply(conn, 200, ds_cstr(&ds)); @@ -1254,7 +1185,7 @@ bond_unixctl_hash(struct unixctl_conn *conn, const char *args_, return; } } else { - vlan = OFP_VLAN_NONE; + vlan = 0; } if (basis_s) { @@ -1281,16 +1212,18 @@ bond_unixctl_hash(struct unixctl_conn *conn, const char *args_, void bond_init(void) { - unixctl_command_register("bond/list", bond_unixctl_list, NULL); - unixctl_command_register("bond/show", bond_unixctl_show, NULL); - unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL); - unixctl_command_register("bond/set-active-slave", + unixctl_command_register("bond/list", "", bond_unixctl_list, NULL); + unixctl_command_register("bond/show", "port", bond_unixctl_show, NULL); + unixctl_command_register("bond/migrate", "port hash slave", + bond_unixctl_migrate, NULL); + unixctl_command_register("bond/set-active-slave", "port slave", bond_unixctl_set_active_slave, NULL); - unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave, - NULL); - unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave, - NULL); - unixctl_command_register("bond/hash", bond_unixctl_hash, NULL); + unixctl_command_register("bond/enable-slave", "port slave", + bond_unixctl_enable_slave, NULL); + unixctl_command_register("bond/disable-slave", "port slave", + bond_unixctl_disable_slave, NULL); + unixctl_command_register("bond/hash", "mac [vlan] [basis]", + bond_unixctl_hash, NULL); } static void @@ -1326,14 +1259,6 @@ bond_slave_lookup(struct bond *bond, const void *slave_) return NULL; } -static bool -bond_is_link_up(struct bond *bond, struct netdev *netdev) -{ - return (bond->detect == BLSM_CARRIER - ? netdev_get_carrier(netdev) - : netdev_get_miimon(netdev)); -} - static void bond_enable_slave(struct bond_slave *slave, bool enable, struct tag_set *tags) { @@ -1363,7 +1288,7 @@ bond_link_status_update(struct bond_slave *slave, struct tag_set *tags) struct bond *bond = slave->bond; bool up; - up = slave->up && slave->lacp_may_enable; + up = netdev_get_carrier(slave->netdev) && slave->may_enable; if ((up == slave->enabled) != (slave->delay_expires == LLONG_MAX)) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); VLOG_INFO_RL(&rl, "interface %s: link state %s", @@ -1395,8 +1320,8 @@ bond_link_status_update(struct bond_slave *slave, struct tag_set *tags) static bool bond_is_tcp_hash(const struct bond *bond) { - return (bond->balance == BM_TCP || bond->balance == BM_STABLE) - && bond->lacp_negotiated; + return (bond->balance == BM_TCP && bond->lacp_negotiated) + || bond->balance == BM_STABLE; } static unsigned int @@ -1512,7 +1437,7 @@ bond_choose_slave(const struct bond *bond) best = NULL; HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { if (slave->delay_expires != LLONG_MAX - && slave->lacp_may_enable + && slave->may_enable && (!best || slave->delay_expires < best->delay_expires)) { best = slave; } @@ -1590,7 +1515,7 @@ bond_update_fake_slave_stats(struct bond *bond) } } - if (!netdev_open_default(bond->name, &bond_dev)) { + if (!netdev_open(bond->name, "system", &bond_dev)) { netdev_set_stats(bond_dev, &bond_stats); netdev_close(bond_dev); }