X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=lib%2Fbond.c;h=bb5d49968b70186e389ff4b4cacff4c7aa0d9a21;hb=2c8fcc9cd6a7bbb948f6c79879e89c7ed791c9b1;hp=9a688be6988bd1331bb44b84354e76a53675ffd7;hpb=296f651909b7ea4de7a9450c0787fc3c4051efc1;p=openvswitch diff --git a/lib/bond.c b/lib/bond.c index 9a688be6..bb5d4996 100644 --- a/lib/bond.c +++ b/lib/bond.c @@ -59,6 +59,7 @@ 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. */ @@ -103,9 +104,6 @@ struct bond { /* BM_STABLE specific bonding info. */ tag_type stb_tag; /* Tag associated with this bond. */ - /* Monitoring. */ - struct netdev_monitor *monitor; /* detect == BLSM_CARRIER only. */ - /* Legacy compatibility. */ long long int next_fake_iface_update; /* LLONG_MAX if disabled. */ @@ -189,7 +187,6 @@ bond_create(const struct bond_settings *s) bond->no_slaves_tag = tag_create_random(); bond->stb_tag = tag_create_random(); bond->next_fake_iface_update = LLONG_MAX; - bond->monitor = netdev_monitor_create(); bond_reconfigure(bond, s); @@ -219,9 +216,6 @@ bond_destroy(struct bond *bond) hmap_destroy(&bond->slaves); free(bond->hash); - - netdev_monitor_destroy(bond->monitor); - free(bond->name); free(bond); } @@ -284,15 +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 (slave->netdev) { - netdev_monitor_remove(bond->monitor, slave->netdev); - } - netdev_monitor_add(bond->monitor, netdev); slave->netdev = netdev; + slave->change_seq = 0; } } @@ -334,7 +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)); @@ -350,7 +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); } } @@ -368,7 +358,6 @@ bond_slave_unregister(struct bond *bond, const void *slave_) return; } - netdev_monitor_remove(bond->monitor, slave->netdev); bond_enable_slave(slave, false, NULL); del_active = bond->active_slave == slave; @@ -419,13 +408,10 @@ bond_run(struct bond *bond, struct tag_set *tags, bool lacp_negotiated) bond->lacp_negotiated = lacp_negotiated; - 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); @@ -468,12 +454,14 @@ bond_wait(struct bond *bond) { struct bond_slave *slave; - netdev_monitor_poll_wait(bond->monitor); - 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) { @@ -496,14 +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. @@ -519,37 +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 @@ -571,10 +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. */ @@ -585,12 +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 @@ -632,7 +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; } @@ -984,11 +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)); @@ -1172,7 +1185,7 @@ bond_unixctl_hash(struct unixctl_conn *conn, const char *args_, return; } } else { - vlan = OFP_VLAN_NONE; + vlan = 0; } if (basis_s) { @@ -1199,16 +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 @@ -1305,8 +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 @@ -1500,7 +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); }