/*
- * 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.
#include <limits.h>
#include <stdint.h>
#include <stdlib.h>
+#include <math.h>
#include "coverage.h"
#include "dynamic-string.h"
#include "flow.h"
#include "hmap.h"
+#include "lacp.h"
#include "list.h"
#include "netdev.h"
#include "odp-util.h"
#include "ofpbuf.h"
#include "packets.h"
#include "poll-loop.h"
+#include "shash.h"
#include "tag.h"
#include "timeval.h"
#include "unixctl.h"
struct bond_entry {
struct bond_slave *slave; /* Assigned slave, NULL if unassigned. */
uint64_t tx_bytes; /* Count of bytes recently transmitted. */
- tag_type tag; /* Tag for entry<->slave association. */
+ tag_type tag; /* Tag for entry<->facet association. */
struct list list_node; /* In bond_slave's 'entries' list. */
};
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. */
- bool lacp_negotiated; /* LACP negotiations were successful. */
+ 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. */
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, uint32_t basis);
static unsigned int bond_hash_tcp(const struct flow *, uint16_t vlan,
static tag_type bond_get_active_slave_tag(const struct bond *);
static struct bond_slave *choose_output_slave(const struct bond *,
const struct flow *,
- uint16_t vlan);
+ uint16_t vlan, tag_type *tags);
static void bond_update_fake_slave_stats(struct bond *);
/* Attempts to parse 's' as the name of a bond balancing mode. If successful,
bond->updelay = s->up_delay;
bond->downdelay = s->down_delay;
- bond->rebalance_interval = s->rebalance_interval;
+
+ if (bond->rebalance_interval != s->rebalance_interval) {
+ bond->rebalance_interval = s->rebalance_interval;
+ revalidate = true;
+ }
if (bond->balance != s->balance) {
bond->balance = s->balance;
revalidate = true;
+
+ if (bond->balance == BM_STABLE) {
+ VLOG_WARN_ONCE("Stable bond mode is deprecated and may be removed"
+ " in February 2013. Please email"
+ " dev@openvswitch.org with concerns.");
+ }
}
if (bond->basis != s->basis) {
*
* The caller should check bond_should_send_learning_packets() afterward. */
void
-bond_run(struct bond *bond, struct tag_set *tags, bool lacp_negotiated)
+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);
- bond->lacp_negotiated = lacp_negotiated;
+ 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->next_fake_iface_update = time_msec() + 1000;
}
- if (is_tcp_hash != bond_is_tcp_hash(bond)) {
- bond->bond_revalidate = true;
- }
-
if (bond->bond_revalidate) {
bond->bond_revalidate = false;
static bool
may_send_learning_packets(const struct bond *bond)
{
- return !bond->lacp_negotiated && bond->balance != BM_AB &&
- bond->active_slave;
+ return bond->lacp_status == LACP_DISABLED
+ && (bond->balance == BM_SLB || bond->balance == BM_AB)
+ && bond->active_slave;
}
/* Returns true if 'bond' needs the client to send out packets to assist with
* is located. For each MAC that has been learned on a port other than 'bond',
* 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.
+ * This function will only return true if 'bond' is in SLB or active-backup
+ * mode and LACP is not negotiated. Otherwise sending learning packets isn't
+ * necessary.
*
* Calling this function resets the state that it checks. */
bool
{
struct bond_slave *slave;
struct ofpbuf *packet;
+ tag_type tags = 0;
struct flow flow;
- assert(may_send_learning_packets(bond));
+ ovs_assert(may_send_learning_packets(bond));
memset(&flow, 0, sizeof flow);
memcpy(flow.dl_src, eth_src, ETH_ADDR_LEN);
- slave = choose_output_slave(bond, &flow, vlan);
+ slave = choose_output_slave(bond, &flow, vlan, &tags);
packet = ofpbuf_new(0);
- compose_benign_packet(packet, "Open vSwitch Bond Failover", 0xf177,
- eth_src);
+ compose_rarp(packet, eth_src);
if (vlan) {
eth_push_vlan(packet, htons(vlan));
}
* 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 slave->enabled ? BV_ACCEPT : BV_DROP;
+ * 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. */
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. */
+ /* 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
bond_choose_output_slave(struct bond *bond, const struct flow *flow,
uint16_t vlan, tag_type *tags)
{
- struct bond_slave *slave = choose_output_slave(bond, flow, vlan);
+ struct bond_slave *slave = choose_output_slave(bond, flow, vlan, tags);
if (slave) {
*tags |= bond->balance == BM_STABLE ? bond->stb_tag : slave->tag;
return slave->aux;
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'. */
hash->tag = tag_create_random();
}
-/* Pick and returns a bond_entry to migrate to 'to' (the least-loaded slave),
+/* Picks and returns a bond_entry to migrate from 'from' (the most heavily
+ * loaded bond slave) to a bond slave that has 'to_tx_bytes' bytes of load,
* given that doing so must decrease the ratio of the load on the two slaves by
* at least 0.1. Returns NULL if there is no appropriate entry.
*
delta = e->tx_bytes;
old_ratio = (double)from->tx_bytes / to_tx_bytes;
new_ratio = (double)(from->tx_bytes - delta) / (to_tx_bytes + delta);
- if (old_ratio - new_ratio > 0.1) {
- /* Would decrease the ratio, move it. */
+ if (old_ratio - new_ratio > 0.1
+ && fabs(new_ratio - 1.0) < fabs(old_ratio - 1.0)) {
+ /* We're aiming for an ideal ratio of 1, meaning both the 'from'
+ and 'to' slave have the same load. Therefore, we only move an
+ entry if it decreases the load on 'from', and brings us closer
+ to equal traffic load. */
return e;
}
}
break;
}
- /* 'from' is carrying significantly more load than 'to', and that load
- * is split across at least two different hashes. */
+ /* 'from' is carrying significantly more load than 'to'. Pick a hash
+ * to move from 'from' to 'to'. */
e = choose_entry_to_migrate(from, to->tx_bytes);
if (e) {
bond_shift_load(e, to, tags);
}
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,
- int argc OVS_UNUSED, const char *argv[],
- 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(argv[1]);
- 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->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-hash-basis: %"PRIu32"\n", bond->basis);
- 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_format(&ds, "lacp_negotiated: %s\n",
- bond->lacp_negotiated ? "true" : "false");
+ 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, "<unknown>\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;
+ 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",
+ ds_put_format(ds, "\tmay_enable: %s\n",
slave->may_enable ? "true" : "false");
if (!bond_is_balanced(bond)) {
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);
/* XXX How can we list the MACs assigned to hashes of SLB bonds? */
}
}
- unixctl_command_reply(conn, 200, ds_cstr(&ds));
+ shash_destroy(&slave_shash);
+ free(sorted_slaves);
+ ds_put_cstr(ds, "\n");
+}
+
+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, ds_cstr(&ds));
ds_destroy(&ds);
}
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;
}
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 = 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;
}
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");
}
}
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
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 {
if (basis_s) {
if (sscanf(basis_s, "%"PRIu32, &basis) != 1) {
- unixctl_command_reply(conn, 501, "invalid basis");
+ unixctl_command_reply_error(conn, "invalid basis");
return;
}
} else {
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");
}
}
bond_init(void)
{
unixctl_command_register("bond/list", "", 0, 0, bond_unixctl_list, NULL);
- unixctl_command_register("bond/show", "port", 1, 1,
- bond_unixctl_show, NULL);
+ unixctl_command_register("bond/show", "[port]", 0, 1, bond_unixctl_show,
+ 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,
VLOG_INFO_RL(&rl, "interface %s: will not be %s",
slave->name, up ? "disabled" : "enabled");
} else {
- int delay = (bond->lacp_negotiated ? 0
+ int delay = (bond->lacp_status != LACP_DISABLED ? 0
: up ? bond->updelay : bond->downdelay);
slave->delay_expires = time_msec() + delay;
if (delay) {
}
}
-static bool
-bond_is_tcp_hash(const struct bond *bond)
-{
- return (bond->balance == BM_TCP && bond->lacp_negotiated)
- || bond->balance == BM_STABLE;
-}
-
static unsigned int
bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan, uint32_t basis)
{
static unsigned int
bond_hash(const struct bond *bond, const struct flow *flow, uint16_t vlan)
{
- assert(bond->balance != BM_AB);
+ ovs_assert(bond->balance == BM_TCP || bond->balance == BM_SLB);
- return (bond_is_tcp_hash(bond)
+ return (bond->balance == BM_TCP
? bond_hash_tcp(flow, vlan, bond->basis)
: bond_hash_src(flow->dl_src, vlan, bond->basis));
}
* 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, const struct flow *flow,
- uint16_t vlan)
+choose_stb_slave(const struct bond *bond, uint32_t flow_hash)
{
struct bond_slave *best, *slave;
- uint32_t best_hash, flow_hash;
+ uint32_t best_hash;
best = NULL;
best_hash = 0;
- flow_hash = bond_hash(bond, flow, vlan);
HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) {
if (slave->enabled) {
uint32_t hash;
static struct bond_slave *
choose_output_slave(const struct bond *bond, const struct flow *flow,
- uint16_t vlan)
+ uint16_t vlan, tag_type *tags)
{
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_STABLE:
- return choose_stb_slave(bond, flow, vlan);
- case BM_SLB:
+ 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),
}
e->tag = tag_create_random();
}
+ *tags |= e->tag;
return e->slave;
default: