X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fbond.c;h=54f2d0e0e3987b7266cb3f642d3e2c52397d594b;hb=refs%2Fheads%2Fbsd-port;hp=dad34ada9164f7a8c2939db5e4bf77e938bc5da4;hpb=1b137691320fbf0d1e545a146495d8988cd62200;p=sliver-openvswitch.git diff --git a/lib/bond.c b/lib/bond.c index dad34ada9..54f2d0e0e 100644 --- a/lib/bond.c +++ b/lib/bond.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks. + * Copyright (c) 2008, 2009, 2010, 2011 Nicira, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -33,6 +33,7 @@ #include "ofpbuf.h" #include "packets.h" #include "poll-loop.h" +#include "shash.h" #include "tag.h" #include "timeval.h" #include "unixctl.h" @@ -40,8 +41,6 @@ VLOG_DEFINE_THIS_MODULE(bond); -COVERAGE_DEFINE(bond_process_lacp); - /* Bit-mask for hashing a flow down to a bucket. * There are (BOND_MASK + 1) buckets. */ #define BOND_MASK 0xff @@ -62,18 +61,22 @@ 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 may_enable; /* Client considers this slave bondable. */ tag_type tag; /* Tag associated with this slave. */ /* Rebalancing info. Used only by bond_rebalance(). */ struct list bal_node; /* In bond_rebalance()'s 'bals' list. */ struct list entries; /* 'struct bond_entry's assigned here. */ uint64_t tx_bytes; /* Sum across 'tx_bytes' of entries. */ + + /* BM_STABLE specific bonding info. */ + uint32_t stb_id; /* ID used for 'stb_slaves' ordering. */ }; /* A bond, that is, a set of network devices grouped to improve performance or @@ -90,6 +93,9 @@ struct bond { struct bond_slave *active_slave; tag_type no_slaves_tag; /* Tag for flows when all slaves disabled. */ int updelay, downdelay; /* Delay before slave goes up/down, in ms. */ + enum lacp_status lacp_status; /* Status of LACP negotiations. */ + bool bond_revalidate; /* True if flows need revalidation. */ + uint32_t basis; /* Basis for flow hash function. */ /* SLB specific bonding info. */ struct bond_entry *hash; /* An array of (BOND_MASK + 1) elements. */ @@ -97,14 +103,8 @@ struct bond { long long int next_rebalance; /* Next rebalancing time. */ bool send_learning_packets; - /* LACP. */ - struct lacp *lacp; /* LACP object. NULL if LACP is disabled. */ - - /* 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. */ + /* BM_STABLE specific bonding info. */ + tag_type stb_tag; /* Tag associated with this bond. */ /* Legacy compatibility. */ long long int next_fake_iface_update; /* LLONG_MAX if disabled. */ @@ -119,15 +119,14 @@ 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 *); static void bond_choose_active_slave(struct bond *, struct tag_set *); -static bool bond_is_tcp_hash(const struct bond *); static unsigned int bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], - uint16_t vlan); -static unsigned int bond_hash_tcp(const struct flow *, uint16_t vlan); + uint16_t vlan, uint32_t basis); +static unsigned int bond_hash_tcp(const struct flow *, uint16_t vlan, + uint32_t basis); static struct bond_entry *lookup_bond_entry(const struct bond *, const struct flow *, uint16_t vlan); @@ -147,6 +146,8 @@ bond_mode_from_string(enum bond_mode *balance, const char *s) *balance = BM_TCP; } else if (!strcmp(s, bond_mode_to_string(BM_SLB))) { *balance = BM_SLB; + } else if (!strcmp(s, bond_mode_to_string(BM_STABLE))) { + *balance = BM_STABLE; } else if (!strcmp(s, bond_mode_to_string(BM_AB))) { *balance = BM_AB; } else { @@ -163,40 +164,14 @@ bond_mode_to_string(enum bond_mode balance) { return "balance-tcp"; case BM_SLB: return "balance-slb"; + case BM_STABLE: + return "stable"; case BM_AB: return "active-backup"; } 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'. @@ -211,7 +186,7 @@ bond_create(const struct bond_settings *s) bond = xzalloc(sizeof *bond); hmap_init(&bond->slaves); bond->no_slaves_tag = tag_create_random(); - bond->miimon_next_update = LLONG_MAX; + bond->stb_tag = tag_create_random(); bond->next_fake_iface_update = LLONG_MAX; bond_reconfigure(bond, s); @@ -242,11 +217,6 @@ bond_destroy(struct bond *bond) hmap_destroy(&bond->slaves); free(bond->hash); - - lacp_destroy(bond->lacp); - - netdev_monitor_destroy(bond->monitor); - free(bond->name); free(bond); } @@ -255,9 +225,7 @@ bond_destroy(struct bond *bond) * * The caller should register each slave on 'bond' by calling * bond_slave_register(). This is optional if none of the slaves' - * configuration has changed, except that it is mandatory if 's' enables LACP - * and 'bond' previously didn't have LACP enabled. In any case it can't - * hurt. + * configuration has changed. In any case it can't hurt. * * Returns true if the configuration has changed in such a way that requires * flow revalidation. @@ -276,44 +244,22 @@ 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; - if (bond->balance != s->balance) { - bond->balance = s->balance; + if (bond->rebalance_interval != s->rebalance_interval) { + bond->rebalance_interval = s->rebalance_interval; 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 (bond->balance != s->balance) { + bond->balance = s->balance; + revalidate = true; } - if (s->lacp) { - if (!bond->lacp) { - bond->lacp = lacp_create(); - } - lacp_configure(bond->lacp, s->lacp); - } else { - lacp_destroy(bond->lacp); - bond->lacp = NULL; + if (bond->basis != s->basis) { + bond->basis = s->basis; + revalidate = true; } if (s->fake_iface) { @@ -324,6 +270,11 @@ bond_reconfigure(struct bond *bond, const struct bond_settings *s) bond->next_fake_iface_update = LLONG_MAX; } + if (bond->bond_revalidate) { + revalidate = true; + bond->bond_revalidate = false; + } + if (bond->balance == BM_AB || !bond->hash || revalidate) { bond_entry_reset(bond); } @@ -331,21 +282,31 @@ bond_reconfigure(struct bond *bond, const struct bond_settings *s) return revalidate; } +static void +bond_slave_set_netdev__(struct bond_slave *slave, struct netdev *netdev) +{ + if (slave->netdev != netdev) { + slave->netdev = netdev; + slave->change_seq = 0; + } +} + /* Registers 'slave_' as a slave of 'bond'. The 'slave_' pointer is an * arbitrary client-provided pointer that uniquely identifies a slave within a * bond. If 'slave_' already exists within 'bond' then this function * reconfigures the existing slave. * + * 'stb_id' is used in BM_STABLE bonds to guarantee consistent slave choices + * across restarts and distributed vswitch instances. It should be unique per + * slave, and preferably consistent across restarts and reconfigurations. + * * 'netdev' must be the network device that 'slave_' represents. It is owned * by the client, so the client must not close it before either unregistering * 'slave_' or destroying 'bond'. - * - * If 'bond' has a LACP configuration then 'lacp_settings' must point to LACP - * settings for 'slave_'; otherwise 'lacp_settings' is ignored. */ void -bond_slave_register(struct bond *bond, void *slave_, struct netdev *netdev, - const struct lacp_slave_settings *lacp_settings) +bond_slave_register(struct bond *bond, void *slave_, uint32_t stb_id, + struct netdev *netdev) { struct bond_slave *slave = bond_slave_lookup(bond, slave_); @@ -356,18 +317,35 @@ bond_slave_register(struct bond *bond, void *slave_, struct netdev *netdev, 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) { + slave->stb_id = stb_id; + bond->bond_revalidate = true; } - slave->netdev = netdev; + bond_slave_set_netdev__(slave, netdev); + free(slave->name); slave->name = xstrdup(netdev_get_name(netdev)); +} - if (bond->lacp) { - assert(lacp_settings != NULL); - lacp_slave_register(bond->lacp, slave, lacp_settings); +/* Updates the network device to be used with 'slave_' to 'netdev'. + * + * This is useful if the caller closes and re-opens the network device + * registered with bond_slave_register() but doesn't need to change anything + * else. */ +void +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__(slave, netdev); } } @@ -412,36 +390,15 @@ bond_slave_unregister(struct bond *bond, const void *slave_) } } -/* Callback for lacp_run(). */ -static void -bond_send_pdu_cb(void *slave_, const struct lacp_pdu *pdu) +/* 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_may_enable(struct bond *bond, void *slave_, bool may_enable) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); - struct bond_slave *slave = slave_; - uint8_t ea[ETH_ADDR_LEN]; - int error; - - error = netdev_get_etheraddr(slave->netdev, ea); - if (!error) { - struct lacp_pdu *packet_pdu; - struct ofpbuf packet; - - ofpbuf_init(&packet, 0); - packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, - sizeof *packet_pdu); - *packet_pdu = *pdu; - error = netdev_send(slave->netdev, &packet); - if (error) { - VLOG_WARN_RL(&rl, "bond %s: sending LACP PDU on slave %s failed " - "(%s)", - slave->bond->name, slave->name, strerror(error)); - } - ofpbuf_uninit(&packet); - } else { - VLOG_ERR_RL(&rl, "bond %s: cannot obtain Ethernet address of slave " - "%s (%s)", - slave->bond->name, slave->name, strerror(error)); - } + bond_slave_lookup(bond, slave_)->may_enable = may_enable; } /* Performs periodic maintenance on 'bond'. The caller must provide 'tags' to @@ -449,33 +406,19 @@ bond_send_pdu_cb(void *slave_, const struct lacp_pdu *pdu) * * The caller should check bond_should_send_learning_packets() afterward. */ void -bond_run(struct bond *bond, struct tag_set *tags) +bond_run(struct bond *bond, struct tag_set *tags, enum lacp_status lacp_status) { struct bond_slave *slave; - bool is_tcp_hash = bond_is_tcp_hash(bond); - - /* 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; - } - /* Update LACP. */ - if (bond->lacp) { - HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { - lacp_slave_enable(bond->lacp, slave, slave->enabled); - } - - lacp_run(bond->lacp, bond_send_pdu_cb); + if (bond->lacp_status != lacp_status) { + bond->lacp_status = lacp_status; + bond->bond_revalidate = true; } /* 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); @@ -487,13 +430,20 @@ bond_run(struct bond *bond, struct tag_set *tags) bond->next_fake_iface_update = time_msec() + 1000; } - if (is_tcp_hash != bond_is_tcp_hash(bond)) { - struct bond_slave *slave; + if (bond->bond_revalidate) { + bond->bond_revalidate = false; bond_entry_reset(bond); - HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { - tag_set_add(tags, slave->tag); + if (bond->balance != BM_STABLE) { + struct bond_slave *slave; + + HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { + tag_set_add(tags, slave->tag); + } + } else { + tag_set_add(tags, bond->stb_tag); } + tag_set_add(tags, bond->no_slaves_tag); } /* Invalidate any tags required by */ @@ -507,16 +457,14 @@ 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) { @@ -539,14 +487,16 @@ bond_wait(struct bond *bond) static bool may_send_learning_packets(const struct bond *bond) { - return !lacp_negotiated(bond->lacp) && bond->balance != BM_AB; + return bond->lacp_status == LACP_DISABLED + && 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. @@ -562,37 +512,32 @@ 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, - eth_src); + packet = ofpbuf_new(0); + compose_rarp(packet, 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 @@ -614,10 +559,20 @@ 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". */ - if (lacp_negotiated(bond->lacp)) { - return BV_ACCEPT; + 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 LACP is configured, but LACP negotiations have been unsuccessful, we + * drop all incoming traffic. */ + switch (bond->lacp_status) { + case LACP_NEGOTIATED: return slave->enabled ? BV_ACCEPT : BV_DROP; + case LACP_CONFIGURED: return BV_DROP; + case LACP_DISABLED: break; } /* Drop all multicast packets on inactive slaves. */ @@ -628,12 +583,41 @@ 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 balanced bonds require successful LACP negotiated. Based on the + * above check, LACP is off on this bond. Therfore, we drop all + * incoming traffic. */ + return BV_DROP; + + 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 @@ -654,38 +638,21 @@ bond_choose_output_slave(struct bond *bond, const struct flow *flow, { struct bond_slave *slave = choose_output_slave(bond, flow, vlan); if (slave) { - *tags |= slave->tag; + *tags |= bond->balance == BM_STABLE ? bond->stb_tag : slave->tag; return slave->aux; } else { *tags |= bond->no_slaves_tag; return NULL; } } - -/* Processes LACP packet 'packet', which was received on 'slave_' within - * 'bond'. - * - * The client should use this function to pass along LACP messages received on - * any of 'bond''s slaves. */ -void -bond_process_lacp(struct bond *bond, void *slave_, const struct ofpbuf *packet) -{ - if (bond->lacp) { - struct bond_slave *slave = bond_slave_lookup(bond, slave_); - const struct lacp_pdu *pdu = parse_lacp_packet(packet); - if (slave && pdu) { - COVERAGE_INC(bond_process_lacp); - lacp_process_pdu(bond->lacp, slave, pdu); - } - } -} /* Rebalancing. */ static bool bond_is_balanced(const struct bond *bond) { - return bond->balance == BM_SLB || bond->balance == BM_TCP; + return bond->rebalance_interval + && (bond->balance == BM_SLB || bond->balance == BM_TCP); } /* Notifies 'bond' that 'n_bytes' bytes were sent in 'flow' within 'vlan'. */ @@ -693,7 +660,6 @@ 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; } @@ -950,7 +916,8 @@ bond_lookup_slave(struct bond *bond, const char *slave_name) static void bond_unixctl_list(struct unixctl_conn *conn, - const char *args OVS_UNUSED, void *aux OVS_UNUSED) + int argc OVS_UNUSED, const char *argv[] OVS_UNUSED, + void *aux OVS_UNUSED) { struct ds ds = DS_EMPTY_INITIALIZER; const struct bond *bond; @@ -973,77 +940,78 @@ bond_unixctl_list(struct unixctl_conn *conn, } ds_put_char(&ds, '\n'); } - unixctl_command_reply(conn, 200, ds_cstr(&ds)); + unixctl_command_reply(conn, ds_cstr(&ds)); ds_destroy(&ds); } static void -bond_unixctl_show(struct unixctl_conn *conn, - const char *args, void *aux OVS_UNUSED) +bond_print_details(struct ds *ds, const struct bond *bond) { - struct ds ds = DS_EMPTY_INITIALIZER; + struct shash slave_shash = SHASH_INITIALIZER(&slave_shash); + const struct shash_node **sorted_slaves = NULL; const struct bond_slave *slave; - const struct bond *bond; + int i; - bond = bond_find(args); - if (!bond) { - unixctl_command_reply(conn, 501, "no such bond"); - return; - } - - ds_put_format(&ds, "bond_mode: %s\n", + ds_put_format(ds, "---- %s ----\n", bond->name); + ds_put_format(ds, "bond_mode: %s\n", bond_mode_to_string(bond->balance)); - if (bond->lacp) { - ds_put_format(&ds, "lacp: %s\n", - lacp_is_active(bond->lacp) ? "active" : "passive"); - } else { - ds_put_cstr(&ds, "lacp: off\n"); - } - - if (bond->balance != BM_AB) { - ds_put_format(&ds, "bond-hash-algorithm: %s\n", - bond_is_tcp_hash(bond) ? "balance-tcp" : "balance-slb"); - } + 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); + ds_put_format(ds, "updelay: %d ms\n", bond->updelay); + ds_put_format(ds, "downdelay: %d ms\n", bond->downdelay); if (bond_is_balanced(bond)) { - ds_put_format(&ds, "next rebalance: %lld ms\n", + ds_put_format(ds, "next rebalance: %lld ms\n", bond->next_rebalance - time_msec()); } + ds_put_cstr(ds, "lacp_status: "); + switch (bond->lacp_status) { + case LACP_NEGOTIATED: + ds_put_cstr(ds, "negotiated\n"); + break; + case LACP_CONFIGURED: + ds_put_cstr(ds, "configured\n"); + break; + case LACP_DISABLED: + ds_put_cstr(ds, "off\n"); + break; + default: + ds_put_cstr(ds, "\n"); + break; + } + HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { + shash_add(&slave_shash, slave->name, slave); + } + sorted_slaves = shash_sort(&slave_shash); + + for (i = 0; i < shash_count(&slave_shash); i++) { struct bond_entry *be; - struct flow flow; + + slave = sorted_slaves[i]->data; /* Basic info. */ - ds_put_format(&ds, "\nslave %s: %s\n", + ds_put_format(ds, "\nslave %s: %s\n", slave->name, slave->enabled ? "enabled" : "disabled"); if (slave == bond->active_slave) { - ds_put_cstr(&ds, "\tactive slave\n"); + ds_put_cstr(ds, "\tactive slave\n"); } if (slave->delay_expires != LLONG_MAX) { - ds_put_format(&ds, "\t%s expires in %lld ms\n", + ds_put_format(ds, "\t%s expires in %lld ms\n", slave->enabled ? "downdelay" : "updelay", slave->delay_expires - time_msec()); } + 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; @@ -1051,67 +1019,83 @@ bond_unixctl_show(struct unixctl_conn *conn, continue; } - ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n", + 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 of SLB bonds? */ + } + } + shash_destroy(&slave_shash); + free(sorted_slaves); + ds_put_cstr(ds, "\n"); +} - /* XXX How can we list the MACs assigned to hashes? */ +static void +bond_unixctl_show(struct unixctl_conn *conn, + int argc, const char *argv[], + void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + + if (argc > 1) { + const struct bond *bond = bond_find(argv[1]); + + if (!bond) { + unixctl_command_reply_error(conn, "no such bond"); + return; + } + bond_print_details(&ds, bond); + } else { + const struct bond *bond; + + HMAP_FOR_EACH (bond, hmap_node, &all_bonds) { + bond_print_details(&ds, bond); } } - unixctl_command_reply(conn, 200, ds_cstr(&ds)); + + unixctl_command_reply(conn, ds_cstr(&ds)); ds_destroy(&ds); } static void -bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, +bond_unixctl_migrate(struct unixctl_conn *conn, + int argc OVS_UNUSED, const char *argv[], void *aux OVS_UNUSED) { - char *args = (char *) args_; - char *save_ptr = NULL; - char *bond_s, *hash_s, *slave_s; + const char *bond_s = argv[1]; + const char *hash_s = argv[2]; + const char *slave_s = argv[3]; struct bond *bond; struct bond_slave *slave; struct bond_entry *entry; int hash; - bond_s = strtok_r(args, " ", &save_ptr); - hash_s = strtok_r(NULL, " ", &save_ptr); - slave_s = strtok_r(NULL, " ", &save_ptr); - if (!slave_s) { - unixctl_command_reply(conn, 501, - "usage: bond/migrate BOND HASH SLAVE"); - return; - } - bond = bond_find(bond_s); if (!bond) { - unixctl_command_reply(conn, 501, "no such bond"); + unixctl_command_reply_error(conn, "no such bond"); return; } if (bond->balance != BM_SLB) { - unixctl_command_reply(conn, 501, "not an SLB bond"); + unixctl_command_reply_error(conn, "not an SLB bond"); return; } if (strspn(hash_s, "0123456789") == strlen(hash_s)) { hash = atoi(hash_s) & BOND_MASK; } else { - unixctl_command_reply(conn, 501, "bad hash"); + unixctl_command_reply_error(conn, "bad hash"); return; } slave = bond_lookup_slave(bond, slave_s); if (!slave) { - unixctl_command_reply(conn, 501, "no such slave"); + unixctl_command_reply_error(conn, "no such slave"); return; } if (!slave->enabled) { - unixctl_command_reply(conn, 501, "cannot migrate to disabled slave"); + unixctl_command_reply_error(conn, "cannot migrate to disabled slave"); return; } @@ -1119,41 +1103,33 @@ bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, tag_set_add(&bond->unixctl_tags, entry->tag); entry->slave = slave; entry->tag = tag_create_random(); - unixctl_command_reply(conn, 200, "migrated"); + unixctl_command_reply(conn, "migrated"); } static void -bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_, +bond_unixctl_set_active_slave(struct unixctl_conn *conn, + int argc OVS_UNUSED, const char *argv[], void *aux OVS_UNUSED) { - char *args = (char *) args_; - char *save_ptr = NULL; - char *bond_s, *slave_s; + const char *bond_s = argv[1]; + const char *slave_s = argv[2]; struct bond *bond; struct bond_slave *slave; - bond_s = strtok_r(args, " ", &save_ptr); - slave_s = strtok_r(NULL, " ", &save_ptr); - if (!slave_s) { - unixctl_command_reply(conn, 501, - "usage: bond/set-active-slave BOND SLAVE"); - return; - } - bond = bond_find(bond_s); if (!bond) { - unixctl_command_reply(conn, 501, "no such bond"); + unixctl_command_reply_error(conn, "no such bond"); return; } slave = bond_lookup_slave(bond, slave_s); if (!slave) { - unixctl_command_reply(conn, 501, "no such slave"); + unixctl_command_reply_error(conn, "no such slave"); return; } if (!slave->enabled) { - unixctl_command_reply(conn, 501, "cannot make disabled slave active"); + unixctl_command_reply_error(conn, "cannot make disabled slave active"); return; } @@ -1164,112 +1140,111 @@ bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_, VLOG_INFO("bond %s: active interface is now %s", bond->name, slave->name); bond->send_learning_packets = true; - unixctl_command_reply(conn, 200, "done"); + unixctl_command_reply(conn, "done"); } else { - unixctl_command_reply(conn, 200, "no change"); + unixctl_command_reply(conn, "no change"); } } static void -enable_slave(struct unixctl_conn *conn, const char *args_, bool enable) +enable_slave(struct unixctl_conn *conn, const char *argv[], bool enable) { - char *args = (char *) args_; - char *save_ptr = NULL; - char *bond_s, *slave_s; + const char *bond_s = argv[1]; + const char *slave_s = argv[2]; struct bond *bond; struct bond_slave *slave; - bond_s = strtok_r(args, " ", &save_ptr); - slave_s = strtok_r(NULL, " ", &save_ptr); - if (!slave_s) { - char *usage = xasprintf("usage: bond/%s-slave BOND SLAVE", - enable ? "enable" : "disable"); - unixctl_command_reply(conn, 501, usage); - free(usage); - return; - } - bond = bond_find(bond_s); if (!bond) { - unixctl_command_reply(conn, 501, "no such bond"); + unixctl_command_reply_error(conn, "no such bond"); return; } slave = bond_lookup_slave(bond, slave_s); if (!slave) { - unixctl_command_reply(conn, 501, "no such slave"); + unixctl_command_reply_error(conn, "no such slave"); return; } bond_enable_slave(slave, enable, &bond->unixctl_tags); - unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled"); + unixctl_command_reply(conn, enable ? "enabled" : "disabled"); } static void -bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args, +bond_unixctl_enable_slave(struct unixctl_conn *conn, + int argc OVS_UNUSED, const char *argv[], void *aux OVS_UNUSED) { - enable_slave(conn, args, true); + enable_slave(conn, argv, true); } static void -bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args, +bond_unixctl_disable_slave(struct unixctl_conn *conn, + int argc OVS_UNUSED, const char *argv[], void *aux OVS_UNUSED) { - enable_slave(conn, args, false); + enable_slave(conn, argv, false); } static void -bond_unixctl_hash(struct unixctl_conn *conn, const char *args_, +bond_unixctl_hash(struct unixctl_conn *conn, int argc, const char *argv[], void *aux OVS_UNUSED) { - char *args = (char *) args_; + const char *mac_s = argv[1]; + const char *vlan_s = argc > 2 ? argv[2] : NULL; + const char *basis_s = argc > 3 ? argv[3] : NULL; uint8_t mac[ETH_ADDR_LEN]; uint8_t hash; char *hash_cstr; unsigned int vlan; - char *mac_s, *vlan_s; - char *save_ptr = NULL; - - mac_s = strtok_r(args, " ", &save_ptr); - vlan_s = strtok_r(NULL, " ", &save_ptr); + uint32_t basis; if (vlan_s) { if (sscanf(vlan_s, "%u", &vlan) != 1) { - unixctl_command_reply(conn, 501, "invalid vlan"); + unixctl_command_reply_error(conn, "invalid vlan"); + return; + } + } else { + vlan = 0; + } + + if (basis_s) { + if (sscanf(basis_s, "%"PRIu32, &basis) != 1) { + unixctl_command_reply_error(conn, "invalid basis"); return; } } else { - vlan = OFP_VLAN_NONE; + basis = 0; } if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac)) == ETH_ADDR_SCAN_COUNT) { - hash = bond_hash_src(mac, vlan) & BOND_MASK; + hash = bond_hash_src(mac, vlan, basis) & BOND_MASK; hash_cstr = xasprintf("%u", hash); - unixctl_command_reply(conn, 200, hash_cstr); + unixctl_command_reply(conn, hash_cstr); free(hash_cstr); } else { - unixctl_command_reply(conn, 501, "invalid mac"); + unixctl_command_reply_error(conn, "invalid mac"); } } void bond_init(void) { - lacp_init(); - - 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", - 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, + unixctl_command_register("bond/list", "", 0, 0, bond_unixctl_list, NULL); + unixctl_command_register("bond/show", "[port]", 0, 1, bond_unixctl_show, NULL); - unixctl_command_register("bond/hash", bond_unixctl_hash, NULL); + unixctl_command_register("bond/migrate", "port hash slave", 3, 3, + bond_unixctl_migrate, NULL); + unixctl_command_register("bond/set-active-slave", "port slave", 2, 2, + bond_unixctl_set_active_slave, NULL); + unixctl_command_register("bond/enable-slave", "port slave", 2, 2, + bond_unixctl_enable_slave, NULL); + unixctl_command_register("bond/disable-slave", "port slave", 2, 2, + bond_unixctl_disable_slave, NULL); + unixctl_command_register("bond/hash", "mac [vlan] [basis]", 1, 3, + bond_unixctl_hash, NULL); } static void @@ -1305,17 +1280,10 @@ 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) { + struct bond *bond = slave->bond; slave->delay_expires = LLONG_MAX; if (enable != slave->enabled) { slave->enabled = enable; @@ -1328,6 +1296,10 @@ bond_enable_slave(struct bond_slave *slave, bool enable, struct tag_set *tags) VLOG_WARN("interface %s: enabled", slave->name); slave->tag = tag_create_random(); } + + if (bond->balance == BM_STABLE) { + bond->bond_revalidate = true; + } } } @@ -1337,7 +1309,7 @@ bond_link_status_update(struct bond_slave *slave, struct tag_set *tags) struct bond *bond = slave->bond; bool up; - up = slave->up && lacp_slave_may_enable(bond->lacp, slave); + 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", @@ -1347,7 +1319,7 @@ bond_link_status_update(struct bond_slave *slave, struct tag_set *tags) VLOG_INFO_RL(&rl, "interface %s: will not be %s", slave->name, up ? "disabled" : "enabled"); } else { - int delay = (lacp_negotiated(bond->lacp) ? 0 + int delay = (bond->lacp_status != LACP_DISABLED ? 0 : up ? bond->updelay : bond->downdelay); slave->delay_expires = time_msec() + delay; if (delay) { @@ -1366,38 +1338,68 @@ 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 && lacp_negotiated(bond->lacp); -} - static unsigned int -bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan) +bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan, uint32_t basis) { - return hash_bytes(mac, ETH_ADDR_LEN, vlan); + return hash_3words(hash_bytes(mac, ETH_ADDR_LEN, 0), vlan, basis); } static unsigned int -bond_hash_tcp(const struct flow *flow, uint16_t vlan) +bond_hash_tcp(const struct flow *flow, uint16_t vlan, uint32_t basis) { struct flow hash_flow = *flow; - hash_flow.vlan_tci = vlan; + hash_flow.vlan_tci = htons(vlan); /* The symmetric quality of this hash function is not required, but * flow_hash_symmetric_l4 already exists, and is sufficient for our * purposes, so we use it out of convenience. */ - return flow_hash_symmetric_l4(&hash_flow, 0); + return flow_hash_symmetric_l4(&hash_flow, basis); +} + +static unsigned int +bond_hash(const struct bond *bond, const struct flow *flow, uint16_t vlan) +{ + assert(bond->balance == BM_TCP || bond->balance == BM_SLB); + + return (bond->balance == BM_TCP + ? bond_hash_tcp(flow, vlan, bond->basis) + : bond_hash_src(flow->dl_src, vlan, bond->basis)); } static struct bond_entry * lookup_bond_entry(const struct bond *bond, const struct flow *flow, uint16_t vlan) { - assert(bond->balance != BM_AB); - return &bond->hash[(bond_is_tcp_hash(bond) - ? bond_hash_tcp(flow, vlan) - : bond_hash_src(flow->dl_src, vlan)) & BOND_MASK]; + return &bond->hash[bond_hash(bond, flow, vlan) & BOND_MASK]; +} + +/* This function uses Highest Random Weight hashing to choose an output slave. + * This approach only reassigns a minimal number of flows when slaves are + * enabled or disabled. Unfortunately, it has O(n) performance against the + * number of slaves. There exist algorithms which are O(1), but have slightly + * more complex implementations and require the use of memory. This may need + * to be reimplemented if it becomes a performance bottleneck. */ +static struct bond_slave * +choose_stb_slave(const struct bond *bond, uint32_t flow_hash) +{ + struct bond_slave *best, *slave; + uint32_t best_hash; + + best = NULL; + best_hash = 0; + HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { + if (slave->enabled) { + uint32_t hash; + + hash = hash_2words(flow_hash, slave->stb_id); + if (!best || hash > best_hash) { + best = slave; + best_hash = hash; + } + } + } + + return best; } static struct bond_slave * @@ -1406,12 +1408,29 @@ choose_output_slave(const struct bond *bond, const struct flow *flow, { struct bond_entry *e; + if (bond->lacp_status == LACP_CONFIGURED) { + /* LACP has been configured on this bond but negotiations were + * unsuccussful. Drop all traffic. */ + return NULL; + } + switch (bond->balance) { case BM_AB: return bond->active_slave; - case BM_SLB: + case BM_STABLE: + return choose_stb_slave(bond, bond_hash_tcp(flow, vlan, bond->basis)); + case BM_TCP: + if (bond->lacp_status != LACP_NEGOTIATED) { + /* Must have LACP negotiations for TCP balanced bonds. */ + return NULL; + } + /* Fall Through. */ + case BM_SLB: + if (!bond_is_balanced(bond)) { + return choose_stb_slave(bond, bond_hash(bond, flow, vlan)); + } e = lookup_bond_entry(bond, flow, vlan); if (!e->slave || !e->slave->enabled) { e->slave = CONTAINER_OF(hmap_random_node(&bond->slaves), @@ -1445,7 +1464,7 @@ bond_choose_slave(const struct bond *bond) best = NULL; HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) { if (slave->delay_expires != LLONG_MAX - && lacp_slave_may_enable(bond->lacp, slave) + && slave->may_enable && (!best || slave->delay_expires < best->delay_expires)) { best = slave; } @@ -1523,7 +1542,7 @@ 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); }