/*
- * Copyright (c) 2008, 2009, 2010, 2011 Nicira, Inc.
+ * 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 "packets.h"
#include "poll-loop.h"
#include "shash.h"
-#include "tag.h"
#include "timeval.h"
#include "unixctl.h"
#include "vlog.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. */
struct list list_node; /* In bond_slave's 'entries' list. */
};
long long delay_expires; /* Time after which 'enabled' may change. */
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
/* Bonding info. */
enum bond_mode balance; /* Balancing mode, one of BM_*. */
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. */
long long int next_rebalance; /* Next rebalancing time. */
bool send_learning_packets;
- /* 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. */
+ bool lacp_fallback_ab; /* Fallback to active-backup on LACP failure. */
- /* Tag set saved for next bond_run(). This tag set is a kluge for cases
- * where we can't otherwise provide revalidation feedback to the client.
- * That's only unixctl commands now; I hope no other cases will arise. */
- struct tag_set unixctl_tags;
+ atomic_int ref_cnt;
};
-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 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 struct ovs_rwlock rwlock = OVS_RWLOCK_INITIALIZER;
+static struct hmap all_bonds__ = HMAP_INITIALIZER(&all_bonds__);
+static struct hmap *const all_bonds OVS_GUARDED_BY(rwlock) = &all_bonds__;
+
+static void bond_entry_reset(struct bond *) OVS_REQ_WRLOCK(rwlock);
+static struct bond_slave *bond_slave_lookup(struct bond *, const void *slave_)
+ OVS_REQ_RDLOCK(rwlock);
+static void bond_enable_slave(struct bond_slave *, bool enable)
+ OVS_REQ_WRLOCK(rwlock);
+static void bond_link_status_update(struct bond_slave *)
+ OVS_REQ_WRLOCK(rwlock);
+static void bond_choose_active_slave(struct bond *)
+ OVS_REQ_WRLOCK(rwlock);;
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,
uint32_t basis);
static struct bond_entry *lookup_bond_entry(const struct bond *,
const struct flow *,
- uint16_t vlan);
-static tag_type bond_get_active_slave_tag(const struct bond *);
+ uint16_t vlan)
+ OVS_REQ_RDLOCK(rwlock);
static struct bond_slave *choose_output_slave(const struct bond *,
const struct flow *,
- uint16_t vlan);
-static void bond_update_fake_slave_stats(struct bond *);
+ struct flow_wildcards *,
+ uint16_t vlan)
+ OVS_REQ_RDLOCK(rwlock);
+static void bond_update_fake_slave_stats(struct bond *)
+ OVS_REQ_RDLOCK(rwlock);
/* Attempts to parse 's' as the name of a bond balancing mode. If successful,
* stores the mode in '*balance' and returns true. Otherwise returns false
*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 {
return "balance-tcp";
case BM_SLB:
return "balance-slb";
- case BM_STABLE:
- return "stable";
case BM_AB:
return "active-backup";
}
bond = xzalloc(sizeof *bond);
hmap_init(&bond->slaves);
- bond->no_slaves_tag = tag_create_random();
- bond->stb_tag = tag_create_random();
bond->next_fake_iface_update = LLONG_MAX;
+ atomic_init(&bond->ref_cnt, 1);
bond_reconfigure(bond, s);
+ return bond;
+}
- tag_set_init(&bond->unixctl_tags);
+struct bond *
+bond_ref(const struct bond *bond_)
+{
+ struct bond *bond = CONST_CAST(struct bond *, bond_);
+ if (bond) {
+ int orig;
+ atomic_add(&bond->ref_cnt, 1, &orig);
+ ovs_assert(orig > 0);
+ }
return bond;
}
/* Frees 'bond'. */
void
-bond_destroy(struct bond *bond)
+bond_unref(struct bond *bond)
{
struct bond_slave *slave, *next_slave;
+ int orig;
if (!bond) {
return;
}
- hmap_remove(&all_bonds, &bond->hmap_node);
+ atomic_sub(&bond->ref_cnt, 1, &orig);
+ ovs_assert(orig > 0);
+ if (orig != 1) {
+ return;
+ }
+
+ ovs_rwlock_wrlock(&rwlock);
+ hmap_remove(all_bonds, &bond->hmap_node);
+ ovs_rwlock_unlock(&rwlock);
HMAP_FOR_EACH_SAFE (slave, next_slave, hmap_node, &bond->slaves) {
hmap_remove(&bond->slaves, &slave->hmap_node);
{
bool revalidate = false;
+ ovs_rwlock_wrlock(&rwlock);
if (!bond->name || strcmp(bond->name, s->name)) {
if (bond->name) {
- hmap_remove(&all_bonds, &bond->hmap_node);
+ hmap_remove(all_bonds, &bond->hmap_node);
free(bond->name);
}
bond->name = xstrdup(s->name);
- hmap_insert(&all_bonds, &bond->hmap_node, hash_string(bond->name, 0));
+ hmap_insert(all_bonds, &bond->hmap_node, hash_string(bond->name, 0));
}
bond->updelay = s->up_delay;
bond->downdelay = s->down_delay;
+ if (bond->lacp_fallback_ab != s->lacp_fallback_ab_cfg) {
+ bond->lacp_fallback_ab = s->lacp_fallback_ab_cfg;
+ revalidate = true;
+ }
+
if (bond->rebalance_interval != s->rebalance_interval) {
bond->rebalance_interval = s->rebalance_interval;
revalidate = true;
bond_entry_reset(bond);
}
+ ovs_rwlock_unlock(&rwlock);
return revalidate;
}
static void
bond_slave_set_netdev__(struct bond_slave *slave, struct netdev *netdev)
+ OVS_REQ_WRLOCK(rwlock)
{
if (slave->netdev != netdev) {
slave->netdev = netdev;
* 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'.
*/
void
-bond_slave_register(struct bond *bond, void *slave_, uint32_t stb_id,
- struct netdev *netdev)
+bond_slave_register(struct bond *bond, void *slave_, struct netdev *netdev)
{
- struct bond_slave *slave = bond_slave_lookup(bond, slave_);
+ struct bond_slave *slave;
+ ovs_rwlock_wrlock(&rwlock);
+ slave = bond_slave_lookup(bond, slave_);
if (!slave) {
slave = xzalloc(sizeof *slave);
bond->bond_revalidate = true;
slave->enabled = false;
- bond_enable_slave(slave, netdev_get_carrier(netdev), NULL);
- }
-
- if (slave->stb_id != stb_id) {
- slave->stb_id = stb_id;
- bond->bond_revalidate = true;
+ bond_enable_slave(slave, netdev_get_carrier(netdev));
}
bond_slave_set_netdev__(slave, netdev);
free(slave->name);
slave->name = xstrdup(netdev_get_name(netdev));
+ ovs_rwlock_unlock(&rwlock);
}
/* Updates the network device to be used with 'slave_' to 'netdev'.
void
bond_slave_set_netdev(struct bond *bond, void *slave_, struct netdev *netdev)
{
- struct bond_slave *slave = bond_slave_lookup(bond, slave_);
+ struct bond_slave *slave;
+
+ ovs_rwlock_wrlock(&rwlock);
+ slave = bond_slave_lookup(bond, slave_);
if (slave) {
bond_slave_set_netdev__(slave, netdev);
}
+ ovs_rwlock_unlock(&rwlock);
}
/* Unregisters 'slave_' from 'bond'. If 'bond' does not contain such a slave
void
bond_slave_unregister(struct bond *bond, const void *slave_)
{
- struct bond_slave *slave = bond_slave_lookup(bond, slave_);
+ struct bond_slave *slave;
bool del_active;
+ ovs_rwlock_wrlock(&rwlock);
+ slave = bond_slave_lookup(bond, slave_);
if (!slave) {
- return;
+ goto out;
}
- bond_enable_slave(slave, false, NULL);
+ bond->bond_revalidate = true;
+ bond_enable_slave(slave, false);
del_active = bond->active_slave == slave;
if (bond->hash) {
free(slave);
if (del_active) {
- struct tag_set tags;
-
- tag_set_init(&tags);
- bond_choose_active_slave(bond, &tags);
+ bond_choose_active_slave(bond);
bond->send_learning_packets = true;
}
+out:
+ ovs_rwlock_unlock(&rwlock);
}
/* Should be called on each slave in 'bond' before bond_run() to indicate
void
bond_slave_set_may_enable(struct bond *bond, void *slave_, bool may_enable)
{
+ ovs_rwlock_wrlock(&rwlock);
bond_slave_lookup(bond, slave_)->may_enable = may_enable;
+ ovs_rwlock_unlock(&rwlock);
}
-/* Performs periodic maintenance on 'bond'. The caller must provide 'tags' to
- * allow tagged flows to be invalidated.
+/* Performs periodic maintenance on 'bond'.
+ *
+ * Returns true if the caller should revalidate its flows.
*
* The caller should check bond_should_send_learning_packets() afterward. */
-void
-bond_run(struct bond *bond, struct tag_set *tags, enum lacp_status lacp_status)
+bool
+bond_run(struct bond *bond, enum lacp_status lacp_status)
{
struct bond_slave *slave;
+ bool revalidate;
+ ovs_rwlock_wrlock(&rwlock);
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);
+ bond_link_status_update(slave);
slave->change_seq = netdev_change_seq(slave->netdev);
}
if (!bond->active_slave || !bond->active_slave->enabled) {
- bond_choose_active_slave(bond, tags);
+ bond_choose_active_slave(bond);
}
/* Update fake bond interface stats. */
bond->next_fake_iface_update = time_msec() + 1000;
}
- if (bond->bond_revalidate) {
- bond->bond_revalidate = false;
-
- bond_entry_reset(bond);
- 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);
- }
+ revalidate = bond->bond_revalidate;
+ bond->bond_revalidate = false;
+ ovs_rwlock_unlock(&rwlock);
- /* Invalidate any tags required by */
- tag_set_union(tags, &bond->unixctl_tags);
- tag_set_init(&bond->unixctl_tags);
+ return revalidate;
}
/* Causes poll_block() to wake up when 'bond' needs something to be done. */
{
struct bond_slave *slave;
+ ovs_rwlock_rdlock(&rwlock);
HMAP_FOR_EACH (slave, hmap_node, &bond->slaves) {
if (slave->delay_expires != LLONG_MAX) {
poll_timer_wait_until(slave->delay_expires);
poll_timer_wait_until(bond->next_fake_iface_update);
}
- /* Ensure that any saved tags get revalidated right away. */
- if (!tag_set_is_empty(&bond->unixctl_tags)) {
+ if (bond->bond_revalidate) {
poll_immediate_wake();
}
+ ovs_rwlock_unlock(&rwlock);
/* We don't wait for bond->next_rebalance because rebalancing can only run
* at a flow account checkpoint. ofproto does checkpointing on its own
static bool
may_send_learning_packets(const struct bond *bond)
{
- return bond->lacp_status == LACP_DISABLED
- && bond->balance != BM_AB
+ return ((bond->lacp_status == LACP_DISABLED
+ && (bond->balance == BM_SLB || bond->balance == BM_AB))
+ || (bond->lacp_fallback_ab && bond->lacp_status == LACP_CONFIGURED))
&& bond->active_slave;
}
* 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
bond_should_send_learning_packets(struct bond *bond)
{
- bool send = bond->send_learning_packets && may_send_learning_packets(bond);
+ bool send;
+
+ ovs_rwlock_wrlock(&rwlock);
+ send = bond->send_learning_packets && may_send_learning_packets(bond);
bond->send_learning_packets = false;
+ ovs_rwlock_unlock(&rwlock);
return send;
}
struct ofpbuf *packet;
struct flow flow;
- assert(may_send_learning_packets(bond));
-
+ ovs_rwlock_rdlock(&rwlock);
+ 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, NULL, vlan);
packet = ofpbuf_new(0);
compose_rarp(packet, eth_src);
}
*port_aux = slave->aux;
+ ovs_rwlock_unlock(&rwlock);
return packet;
}
\f
*/
enum bond_verdict
bond_check_admissibility(struct bond *bond, const void *slave_,
- const uint8_t eth_dst[ETH_ADDR_LEN], tag_type *tags)
+ const uint8_t eth_dst[ETH_ADDR_LEN])
{
- struct bond_slave *slave = bond_slave_lookup(bond, slave_);
+ enum bond_verdict verdict = BV_DROP;
+ struct bond_slave *slave;
+
+ ovs_rwlock_rdlock(&rwlock);
+ slave = bond_slave_lookup(bond, slave_);
+ if (!slave) {
+ goto out;
+ }
/* LACP bonds have very loose admissibility restrictions because we can
* assume the remote switch is aware of the bond and will "do the right
* packets to them.
*
* If LACP is configured, but LACP negotiations have been unsuccessful, we
- * drop all incoming traffic. */
+ * drop all incoming traffic except if lacp_fallback_ab is enabled. */
switch (bond->lacp_status) {
- case LACP_NEGOTIATED: return slave->enabled ? BV_ACCEPT : BV_DROP;
- case LACP_CONFIGURED: return BV_DROP;
- case LACP_DISABLED: break;
+ case LACP_NEGOTIATED:
+ verdict = slave->enabled ? BV_ACCEPT : BV_DROP;
+ goto out;
+ case LACP_CONFIGURED:
+ if (!bond->lacp_fallback_ab) {
+ goto out;
+ }
+ case LACP_DISABLED:
+ break;
}
/* Drop all multicast packets on inactive slaves. */
if (eth_addr_is_multicast(eth_dst)) {
- *tags |= bond_get_active_slave_tag(bond);
- if (bond->active_slave != bond_slave_lookup(bond, slave_)) {
- return BV_DROP;
+ if (bond->active_slave != slave) {
+ goto out;
}
}
switch (bond->balance) {
+ case BM_TCP:
+ /* TCP balanced bonds require successful LACP negotiations. Based on the
+ * above check, LACP is off or lacp_fallback_ab is true on this bond.
+ * If lacp_fallback_ab is true fall through to BM_AB case else, we
+ * drop all incoming traffic. */
+ if (!bond->lacp_fallback_ab) {
+ goto out;
+ }
+
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;
+ goto out;
}
- 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;
+ verdict = BV_ACCEPT;
+ goto out;
case BM_SLB:
/* Drop all packets for which we have learned a different input port,
* 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;
+ verdict = BV_DROP_IF_MOVED;
+ goto out;
}
NOT_REACHED();
+out:
+ ovs_rwlock_unlock(&rwlock);
+ return verdict;
+
}
/* Returns the slave (registered on 'bond' by bond_slave_register()) to which
* nonzero only for trunk ports), whereas 'vlan' is the logical VLAN that the
* packet belongs to (so for an access port it will be the access port's VLAN).
*
- * Adds a tag to '*tags' that associates the flow with the returned slave.
+ * If 'wc' is non-NULL, bitwise-OR's 'wc' with the set of bits that were
+ * significant in the selection. At some point earlier, 'wc' should
+ * have been initialized (e.g., by flow_wildcards_init_catchall()).
*/
void *
bond_choose_output_slave(struct bond *bond, const struct flow *flow,
- uint16_t vlan, tag_type *tags)
+ struct flow_wildcards *wc, uint16_t vlan)
{
- struct bond_slave *slave = choose_output_slave(bond, flow, vlan);
- if (slave) {
- *tags |= bond->balance == BM_STABLE ? bond->stb_tag : slave->tag;
- return slave->aux;
- } else {
- *tags |= bond->no_slaves_tag;
- return NULL;
- }
+ struct bond_slave *slave;
+ void *aux;
+
+ ovs_rwlock_rdlock(&rwlock);
+ slave = choose_output_slave(bond, flow, wc, vlan);
+ aux = slave ? slave->aux : NULL;
+ ovs_rwlock_unlock(&rwlock);
+
+ return aux;
}
\f
/* Rebalancing. */
static bool
-bond_is_balanced(const struct bond *bond)
+bond_is_balanced(const struct bond *bond) OVS_REQ_RDLOCK(rwlock)
{
return bond->rebalance_interval
&& (bond->balance == BM_SLB || bond->balance == BM_TCP);
bond_account(struct bond *bond, const struct flow *flow, uint16_t vlan,
uint64_t n_bytes)
{
+ ovs_rwlock_wrlock(&rwlock);
if (bond_is_balanced(bond)) {
lookup_bond_entry(bond, flow, vlan)->tx_bytes += n_bytes;
}
+ ovs_rwlock_unlock(&rwlock);
}
static struct bond_slave *
-bond_slave_from_bal_node(struct list *bal)
+bond_slave_from_bal_node(struct list *bal) OVS_REQ_RDLOCK(rwlock)
{
return CONTAINER_OF(bal, struct bond_slave, bal_node);
}
/* Shifts 'hash' from its current slave to 'to'. */
static void
-bond_shift_load(struct bond_entry *hash, struct bond_slave *to,
- struct tag_set *set)
+bond_shift_load(struct bond_entry *hash, struct bond_slave *to)
{
struct bond_slave *from = hash->slave;
struct bond *bond = from->bond;
to->tx_bytes += delta;
/* Arrange for flows to be revalidated. */
- tag_set_add(set, hash->tag);
hash->slave = to;
- hash->tag = tag_create_random();
+ bond->bond_revalidate = true;
}
-/* 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;
}
}
* The caller should have called bond_account() for each active flow, to ensure
* that flow data is consistently accounted at this point. */
void
-bond_rebalance(struct bond *bond, struct tag_set *tags)
+bond_rebalance(struct bond *bond)
{
struct bond_slave *slave;
struct bond_entry *e;
struct list bals;
+ ovs_rwlock_wrlock(&rwlock);
if (!bond_is_balanced(bond) || time_msec() < bond->next_rebalance) {
+ ovs_rwlock_unlock(&rwlock);
return;
}
bond->next_rebalance = time_msec() + bond->rebalance_interval;
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);
+ bond_shift_load(e, to);
/* Delete element from from->entries.
*
e->slave = NULL;
}
}
+ ovs_rwlock_unlock(&rwlock);
}
\f
/* Bonding unixctl user interface functions. */
static struct bond *
-bond_find(const char *name)
+bond_find(const char *name) OVS_REQ_RDLOCK(rwlock)
{
struct bond *bond;
HMAP_FOR_EACH_WITH_HASH (bond, hmap_node, hash_string(name, 0),
- &all_bonds) {
+ all_bonds) {
if (!strcmp(bond->name, name)) {
return bond;
}
ds_put_cstr(&ds, "bond\ttype\tslaves\n");
- HMAP_FOR_EACH (bond, hmap_node, &all_bonds) {
+ ovs_rwlock_rdlock(&rwlock);
+ HMAP_FOR_EACH (bond, hmap_node, all_bonds) {
const struct bond_slave *slave;
size_t i;
}
ds_put_char(&ds, '\n');
}
+ ovs_rwlock_unlock(&rwlock);
unixctl_command_reply(conn, ds_cstr(&ds));
ds_destroy(&ds);
}
static void
bond_print_details(struct ds *ds, const struct bond *bond)
+ OVS_REQ_RDLOCK(rwlock)
{
struct shash slave_shash = SHASH_INITIALIZER(&slave_shash);
const struct shash_node **sorted_slaves = NULL;
{
struct ds ds = DS_EMPTY_INITIALIZER;
+ ovs_rwlock_rdlock(&rwlock);
if (argc > 1) {
const struct bond *bond = bond_find(argv[1]);
if (!bond) {
unixctl_command_reply_error(conn, "no such bond");
- return;
+ goto out;
}
bond_print_details(&ds, bond);
} else {
const struct bond *bond;
- HMAP_FOR_EACH (bond, hmap_node, &all_bonds) {
+ HMAP_FOR_EACH (bond, hmap_node, all_bonds) {
bond_print_details(&ds, bond);
}
}
unixctl_command_reply(conn, ds_cstr(&ds));
ds_destroy(&ds);
+
+out:
+ ovs_rwlock_unlock(&rwlock);
}
static void
struct bond_entry *entry;
int hash;
+ ovs_rwlock_wrlock(&rwlock);
bond = bond_find(bond_s);
if (!bond) {
unixctl_command_reply_error(conn, "no such bond");
- return;
+ goto out;
}
if (bond->balance != BM_SLB) {
unixctl_command_reply_error(conn, "not an SLB bond");
- return;
+ goto out;
}
if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
hash = atoi(hash_s) & BOND_MASK;
} else {
unixctl_command_reply_error(conn, "bad hash");
- return;
+ goto out;
}
slave = bond_lookup_slave(bond, slave_s);
if (!slave) {
unixctl_command_reply_error(conn, "no such slave");
- return;
+ goto out;
}
if (!slave->enabled) {
unixctl_command_reply_error(conn, "cannot migrate to disabled slave");
- return;
+ goto out;
}
entry = &bond->hash[hash];
- tag_set_add(&bond->unixctl_tags, entry->tag);
+ bond->bond_revalidate = true;
entry->slave = slave;
- entry->tag = tag_create_random();
unixctl_command_reply(conn, "migrated");
+
+out:
+ ovs_rwlock_unlock(&rwlock);
}
static void
struct bond *bond;
struct bond_slave *slave;
+ ovs_rwlock_wrlock(&rwlock);
bond = bond_find(bond_s);
if (!bond) {
unixctl_command_reply_error(conn, "no such bond");
- return;
+ goto out;
}
slave = bond_lookup_slave(bond, slave_s);
if (!slave) {
unixctl_command_reply_error(conn, "no such slave");
- return;
+ goto out;
}
if (!slave->enabled) {
unixctl_command_reply_error(conn, "cannot make disabled slave active");
- return;
+ goto out;
}
if (bond->active_slave != slave) {
- tag_set_add(&bond->unixctl_tags, bond_get_active_slave_tag(bond));
+ bond->bond_revalidate = true;
bond->active_slave = slave;
- bond->active_slave->tag = tag_create_random();
VLOG_INFO("bond %s: active interface is now %s",
bond->name, slave->name);
bond->send_learning_packets = true;
} else {
unixctl_command_reply(conn, "no change");
}
+out:
+ ovs_rwlock_unlock(&rwlock);
}
static void
struct bond *bond;
struct bond_slave *slave;
+ ovs_rwlock_wrlock(&rwlock);
bond = bond_find(bond_s);
if (!bond) {
unixctl_command_reply_error(conn, "no such bond");
- return;
+ goto out;
}
slave = bond_lookup_slave(bond, slave_s);
if (!slave) {
unixctl_command_reply_error(conn, "no such slave");
- return;
+ goto out;
}
- bond_enable_slave(slave, enable, &bond->unixctl_tags);
+ bond_enable_slave(slave, enable);
unixctl_command_reply(conn, enable ? "enabled" : "disabled");
+
+out:
+ ovs_rwlock_unlock(&rwlock);
}
static void
}
static void
-bond_enable_slave(struct bond_slave *slave, bool enable, struct tag_set *tags)
+bond_enable_slave(struct bond_slave *slave, bool enable)
{
- struct bond *bond = slave->bond;
slave->delay_expires = LLONG_MAX;
if (enable != slave->enabled) {
+ slave->bond->bond_revalidate = true;
slave->enabled = enable;
- if (!slave->enabled) {
- VLOG_WARN("interface %s: disabled", slave->name);
- if (tags) {
- tag_set_add(tags, slave->tag);
- }
- } else {
- VLOG_WARN("interface %s: enabled", slave->name);
- slave->tag = tag_create_random();
- }
-
- if (bond->balance == BM_STABLE) {
- bond->bond_revalidate = true;
- }
+ VLOG_INFO("interface %s: %s", slave->name,
+ slave->enabled ? "enabled" : "disabled");
}
}
static void
-bond_link_status_update(struct bond_slave *slave, struct tag_set *tags)
+bond_link_status_update(struct bond_slave *slave)
{
struct bond *bond = slave->bond;
bool up;
}
if (time_msec() >= slave->delay_expires) {
- bond_enable_slave(slave, up, tags);
+ bond_enable_slave(slave, up);
}
}
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);
+ ovs_assert(bond->balance == BM_TCP || bond->balance == BM_SLB);
return (bond->balance == BM_TCP
? bond_hash_tcp(flow, vlan, bond->basis)
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 *
choose_output_slave(const struct bond *bond, const struct flow *flow,
- uint16_t vlan)
+ struct flow_wildcards *wc, uint16_t vlan)
{
struct bond_entry *e;
+ int balance;
+ balance = bond->balance;
if (bond->lacp_status == LACP_CONFIGURED) {
/* LACP has been configured on this bond but negotiations were
- * unsuccussful. Drop all traffic. */
- return NULL;
+ * unsuccussful. If lacp_fallback_ab is enabled use active-
+ * backup mode else drop all traffic. */
+ if (!bond->lacp_fallback_ab) {
+ return NULL;
+ }
+ balance = BM_AB;
}
- switch (bond->balance) {
+ switch (balance) {
case BM_AB:
return bond->active_slave;
- 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;
}
+ if (wc) {
+ flow_mask_hash_fields(flow, wc, NX_HASH_FIELDS_SYMMETRIC_L4);
+ }
/* Fall Through. */
case BM_SLB:
- if (!bond_is_balanced(bond)) {
- return choose_stb_slave(bond, bond_hash(bond, flow, vlan));
+ if (wc) {
+ flow_mask_hash_fields(flow, wc, NX_HASH_FIELDS_ETH_SRC);
}
e = lookup_bond_entry(bond, flow, vlan);
if (!e->slave || !e->slave->enabled) {
if (!e->slave->enabled) {
e->slave = bond->active_slave;
}
- e->tag = tag_create_random();
}
return e->slave;
}
static void
-bond_choose_active_slave(struct bond *bond, struct tag_set *tags)
+bond_choose_active_slave(struct bond *bond)
{
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
struct bond_slave *old_active_slave = bond->active_slave;
"remaining %lld ms updelay (since no interface was "
"enabled)", bond->name, bond->active_slave->name,
bond->active_slave->delay_expires - time_msec());
- bond_enable_slave(bond->active_slave, true, tags);
- }
-
- if (!old_active_slave) {
- tag_set_add(tags, bond->no_slaves_tag);
+ bond_enable_slave(bond->active_slave, true);
}
bond->send_learning_packets = true;
} else if (old_active_slave) {
- VLOG_WARN_RL(&rl, "bond %s: all interfaces disabled", bond->name);
+ VLOG_INFO_RL(&rl, "bond %s: all interfaces disabled", bond->name);
}
}
-/* Returns the tag for 'bond''s active slave, or 'bond''s no_slaves_tag if
- * there is no active slave. */
-static tag_type
-bond_get_active_slave_tag(const struct bond *bond)
-{
- return (bond->active_slave
- ? bond->active_slave->tag
- : bond->no_slaves_tag);
-}
-
/* Attempts to make the sum of the bond slaves' statistics appear on the fake
* bond interface. */
static void
git://git.onelab.eu / sliver-openvswitch.git / blobdiff