#include "vlog.h"
#include "sflow_api.h"
#include "vlan-bitmap.h"
+#include "packets.h"
VLOG_DEFINE_THIS_MODULE(bridge);
char *name; /* Host network device name. */
struct netdev *netdev; /* Network device. */
ofp_port_t ofp_port; /* OpenFlow port number. */
+ uint64_t change_seq;
/* These members are valid only within bridge_reconfigure(). */
const char *type; /* Usually same as cfg->type. */
/* Track changes to port connectivity. */
static uint64_t connectivity_seqno = LLONG_MIN;
+/* Status update to database.
+ *
+ * Some information in the database must be kept as up-to-date as possible to
+ * allow controllers to respond rapidly to network outages. Those status are
+ * updated via the 'status_txn'.
+ *
+ * We use the global connectivity sequence number to detect the status change.
+ * Also, to prevent the status update from sending too much to the database,
+ * we check the return status of each update transaction and do not start new
+ * update if the previous transaction status is 'TXN_INCOMPLETE'.
+ *
+ * 'status_txn' is NULL if there is no ongoing status update.
+ *
+ * If the previous database transaction was incomplete or failed (is not
+ * 'TXN_SUCCESS' or 'TXN_UNCHANGED'), 'force_status_commit' is set to true.
+ * This means that 'status_txn' must be committed next iteration of bridge_run()
+ * even if the connectivity or netdev sequence numbers do not change.
+ */
+static struct ovsdb_idl_txn *status_txn;
+static bool force_status_commit = true;
+
+/* When the status update transaction returns 'TXN_INCOMPLETE', should register a
+ * timeout in 'STATUS_CHECK_AGAIN_MSEC' to check again. */
+#define STATUS_CHECK_AGAIN_MSEC 100
+
/* Each time this timer expires, the bridge fetches interface and mirror
* statistics and pushes them into the database. */
-#define IFACE_STATS_INTERVAL (5 * 1000) /* In milliseconds. */
-static long long int iface_stats_timer = LLONG_MIN;
+static int stats_timer_interval;
+static long long int stats_timer = LLONG_MIN;
-/* Set to true to allow experimental use of OpenFlow 1.4.
- * This is false initially because OpenFlow 1.4 is not yet safe to use: it can
- * abort due to unimplemented features. */
-static bool allow_of14;
+/* Current stats database transaction, NULL if there is no ongoing
+ * transaction. */
+static struct ovsdb_idl_txn *stats_txn;
/* In some datapaths, creating and destroying OpenFlow ports can be extremely
* expensive. This can cause bridge_reconfigure() to take a long time during
static const char *iface_get_type(const struct ovsrec_interface *,
const struct ovsrec_bridge *);
static void iface_destroy(struct iface *);
+static void iface_destroy__(struct iface *);
static struct iface *iface_lookup(const struct bridge *, const char *name);
static struct iface *iface_find(const char *name);
static struct iface *iface_from_ofp_port(const struct bridge *,
static void iface_configure_cfm(struct iface *);
static void iface_refresh_cfm_stats(struct iface *);
static void iface_refresh_stats(struct iface *);
-static void iface_refresh_status(struct iface *);
+static void iface_refresh_netdev_status(struct iface *);
+static void iface_refresh_ofproto_status(struct iface *);
static bool iface_is_synthetic(const struct iface *);
static ofp_port_t iface_get_requested_ofp_port(
const struct ovsrec_interface *);
ovsdb_idl_omit_alert(idl, &ovsrec_port_col_status);
ovsdb_idl_omit_alert(idl, &ovsrec_port_col_statistics);
+ ovsdb_idl_omit_alert(idl, &ovsrec_port_col_bond_active_slave);
ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_admin_state);
ovsdb_idl_destroy(idl);
}
-/* Enables use of OpenFlow 1.4. This is off by default because OpenFlow 1.4 is
- * not yet safe to use: it can abort due to unimplemented features. */
-void
-bridge_enable_of14(void)
-{
- allow_of14 = true;
-}
-
/* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
* addresses and ports into '*managersp' and '*n_managersp'. The caller is
* responsible for freeing '*managersp' (with free()).
bridge_configure_stp(br);
bridge_configure_tables(br);
bridge_configure_dp_desc(br);
-
- if (smap_get(&br->cfg->other_config, "flow-eviction-threshold")) {
- /* XXX: Remove this warning message eventually. */
- VLOG_WARN_ONCE("As of June 2013, flow-eviction-threshold has been"
- " moved to the Open_vSwitch table. Ignoring its"
- " setting in the bridge table.");
- }
}
free(managers);
struct ofproto_port ofproto_port;
struct ofproto_port_dump dump;
+ struct sset ofproto_ports;
+ struct port *port, *port_next;
+
/* List of "ofp_port"s to delete. We make a list instead of deleting them
* right away because ofproto implementations aren't necessarily able to
* iterate through a changing list of ports in an entirely robust way. */
del = NULL;
n = allocated = 0;
+ sset_init(&ofproto_ports);
+ /* Main task: Iterate over the ports in 'br->ofproto' and remove the ports
+ * that are not configured in the database. (This commonly happens when
+ * ports have been deleted, e.g. with "ovs-vsctl del-port".)
+ *
+ * Side tasks: Reconfigure the ports that are still in 'br'. Delete ports
+ * that have the wrong OpenFlow port number (and arrange to add them back
+ * with the correct OpenFlow port number). */
OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, br->ofproto) {
ofp_port_t requested_ofp_port;
struct iface *iface;
+ sset_add(&ofproto_ports, ofproto_port.name);
+
iface = iface_lookup(br, ofproto_port.name);
if (!iface) {
/* No such iface is configured, so we should delete this
iface_destroy(iface);
del = add_ofp_port(ofproto_port.ofp_port, del, &n, &allocated);
}
-
for (i = 0; i < n; i++) {
ofproto_port_del(br->ofproto, del[i]);
}
free(del);
+
+ /* Iterate over this module's idea of interfaces in 'br'. Remove any ports
+ * that we didn't see when we iterated through the datapath, i.e. ports
+ * that disappeared underneath use. This is an unusual situation, but it
+ * can happen in some cases:
+ *
+ * - An admin runs a command like "ovs-dpctl del-port" (which is a bad
+ * idea but could happen).
+ *
+ * - The port represented a device that disappeared, e.g. a tuntap
+ * device destroyed via "tunctl -d", a physical Ethernet device
+ * whose module was just unloaded via "rmmod", or a virtual NIC for a
+ * VM whose VM was just terminated. */
+ HMAP_FOR_EACH_SAFE (port, port_next, hmap_node, &br->ports) {
+ struct iface *iface, *iface_next;
+
+ LIST_FOR_EACH_SAFE (iface, iface_next, port_elem, &port->ifaces) {
+ if (!sset_contains(&ofproto_ports, iface->name)) {
+ iface_destroy__(iface);
+ }
+ }
+
+ if (list_is_empty(&port->ifaces)) {
+ port_destroy(port);
+ }
+ }
+ sset_destroy(&ofproto_ports);
}
static void
s.vlan_mode = PORT_VLAN_NATIVE_UNTAGGED;
} else {
/* This "can't happen" because ovsdb-server should prevent it. */
- VLOG_ERR("unknown VLAN mode %s", cfg->vlan_mode);
+ VLOG_WARN("port %s: unknown VLAN mode %s, falling "
+ "back to trunk mode", port->name, cfg->vlan_mode);
s.vlan_mode = PORT_VLAN_TRUNK;
}
} else {
if (s.vlan >= 0) {
s.vlan_mode = PORT_VLAN_ACCESS;
if (cfg->n_trunks) {
- VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
- port->name);
+ VLOG_WARN("port %s: ignoring trunks in favor of implicit vlan",
+ port->name);
}
} else {
s.vlan_mode = PORT_VLAN_TRUNK;
static uint32_t
bridge_get_allowed_versions(struct bridge *br)
{
- uint32_t allowed_versions;
-
if (!br->cfg->n_protocols)
return 0;
- allowed_versions = ofputil_versions_from_strings(br->cfg->protocols,
- br->cfg->n_protocols);
- if (!allow_of14) {
- allowed_versions &= ~(1u << OFP14_VERSION);
- }
- return allowed_versions;
+ return ofputil_versions_from_strings(br->cfg->protocols,
+ br->cfg->n_protocols);
}
/* Set NetFlow configuration on 'br'. */
VLOG_INFO("bridge %s: added interface %s on port %d",
br->name, iface_cfg->name, *ofp_portp);
- if ((port_cfg->vlan_mode && !strcmp(port_cfg->vlan_mode, "splinter"))
- || iface_is_internal(iface_cfg, br->cfg)) {
+ if (port_cfg->vlan_mode && !strcmp(port_cfg->vlan_mode, "splinter")) {
netdev_turn_flags_on(netdev, NETDEV_UP, NULL);
}
/* Populate initial status in database. */
iface_refresh_stats(iface);
- iface_refresh_status(iface);
/* Add bond fake iface if necessary. */
if (port_is_bond_fake_iface(port)) {
}
static void
-iface_refresh_status(struct iface *iface)
+iface_refresh_netdev_status(struct iface *iface)
{
struct smap smap;
enum netdev_features current;
- int64_t bps;
- int mtu;
- int64_t mtu_64;
+ enum netdev_flags flags;
+ const char *link_state;
uint8_t mac[ETH_ADDR_LEN];
- int64_t ifindex64;
- int error;
+ int64_t bps, mtu_64, ifindex64, link_resets;
+ int mtu, error;
if (iface_is_synthetic(iface)) {
return;
}
+ if (iface->change_seq == netdev_get_change_seq(iface->netdev)
+ && !force_status_commit) {
+ return;
+ }
+
+ iface->change_seq = netdev_get_change_seq(iface->netdev);
+
smap_init(&smap);
if (!netdev_get_status(iface->netdev, &smap)) {
smap_destroy(&smap);
+ error = netdev_get_flags(iface->netdev, &flags);
+ if (!error) {
+ const char *state = flags & NETDEV_UP ? "up" : "down";
+
+ ovsrec_interface_set_admin_state(iface->cfg, state);
+ } else {
+ ovsrec_interface_set_admin_state(iface->cfg, NULL);
+ }
+
+ link_state = netdev_get_carrier(iface->netdev) ? "up" : "down";
+ ovsrec_interface_set_link_state(iface->cfg, link_state);
+
+ link_resets = netdev_get_carrier_resets(iface->netdev);
+ ovsrec_interface_set_link_resets(iface->cfg, &link_resets, 1);
+
error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
bps = !error ? netdev_features_to_bps(current, 0) : 0;
if (bps) {
ovsrec_interface_set_ifindex(iface->cfg, &ifindex64, 1);
}
+static void
+iface_refresh_ofproto_status(struct iface *iface)
+{
+ struct smap smap;
+ int current, error;
+
+ if (iface_is_synthetic(iface)) {
+ return;
+ }
+
+ current = ofproto_port_is_lacp_current(iface->port->bridge->ofproto,
+ iface->ofp_port);
+ if (current >= 0) {
+ bool bl = current;
+ ovsrec_interface_set_lacp_current(iface->cfg, &bl, 1);
+ } else {
+ ovsrec_interface_set_lacp_current(iface->cfg, NULL, 0);
+ }
+
+ iface_refresh_cfm_stats(iface);
+
+ smap_init(&smap);
+ error = ofproto_port_get_bfd_status(iface->port->bridge->ofproto,
+ iface->ofp_port, force_status_commit,
+ &smap);
+ if (error >= 0) {
+ ovsrec_interface_set_bfd_status(iface->cfg, &smap);
+ }
+ smap_destroy(&smap);
+}
+
/* Writes 'iface''s CFM statistics to the database. 'iface' must not be
* synthetic. */
static void
{
const struct ovsrec_interface *cfg = iface->cfg;
struct ofproto_cfm_status status;
+ int error;
- if (!ofproto_port_get_cfm_status(iface->port->bridge->ofproto,
- iface->ofp_port, &status)) {
+ error = ofproto_port_get_cfm_status(iface->port->bridge->ofproto,
+ iface->ofp_port, force_status_commit,
+ &status);
+ if (error < 0) {
+ /* Do nothing if there is no status change since last update. */
+ } else if (error > 0) {
ovsrec_interface_set_cfm_fault(cfg, NULL, 0);
ovsrec_interface_set_cfm_fault_status(cfg, NULL, 0);
ovsrec_interface_set_cfm_remote_opstate(cfg, NULL);
ARRAY_SIZE(int_values));
}
+static void
+port_refresh_bond_status(struct port *port, bool force_update)
+{
+ uint8_t mac[6];
+
+ /* Return if port is not a bond */
+ if (list_is_singleton(&port->ifaces)) {
+ return;
+ }
+
+ if (bond_get_changed_active_slave(port->name, mac, force_update)) {
+ struct ds mac_s;
+
+ ds_init(&mac_s);
+ ds_put_format(&mac_s, ETH_ADDR_FMT, ETH_ADDR_ARGS(mac));
+ ovsrec_port_set_bond_active_slave(port->cfg, ds_cstr(&mac_s));
+ ds_destroy(&mac_s);
+ }
+}
+
static bool
enable_system_stats(const struct ovsrec_open_vswitch *cfg)
{
ofproto_free_ofproto_controller_info(&info);
}
-\f
-/* "Instant" stats.
- *
- * Some information in the database must be kept as up-to-date as possible to
- * allow controllers to respond rapidly to network outages. We call these
- * statistics "instant" stats.
- *
- * We wish to update these statistics every INSTANT_INTERVAL_MSEC milliseconds,
- * assuming that they've changed. The only means we have to determine whether
- * they have changed are:
- *
- * - Try to commit changes to the database. If nothing changed, then
- * ovsdb_idl_txn_commit() returns TXN_UNCHANGED, otherwise some other
- * value.
- *
- * - instant_stats_run() is called late in the run loop, after anything that
- * might change any of the instant stats.
- *
- * We use these two facts together to avoid waking the process up every
- * INSTANT_INTERVAL_MSEC whether there is any change or not.
- */
-
-/* Minimum interval between writing updates to the instant stats to the
- * database. */
-#define INSTANT_INTERVAL_MSEC 100
-
-/* Current instant stats database transaction, NULL if there is no ongoing
- * transaction. */
-static struct ovsdb_idl_txn *instant_txn;
-
-/* Next time (in msec on monotonic clock) at which we will update the instant
- * stats. */
-static long long int instant_next_txn = LLONG_MIN;
-
-/* True if the run loop has run since we last saw that the instant stats were
- * unchanged, that is, this is true if we need to wake up at 'instant_next_txn'
- * to refresh the instant stats. */
-static bool instant_stats_could_have_changed;
-static void
-instant_stats_run(void)
-{
- enum ovsdb_idl_txn_status status;
-
- instant_stats_could_have_changed = true;
-
- if (!instant_txn) {
- struct bridge *br;
- uint64_t seq;
-
- if (time_msec() < instant_next_txn) {
- return;
- }
- instant_next_txn = time_msec() + INSTANT_INTERVAL_MSEC;
-
- seq = seq_read(connectivity_seq_get());
- if (seq == connectivity_seqno) {
- return;
- }
- connectivity_seqno = seq;
-
- instant_txn = ovsdb_idl_txn_create(idl);
- HMAP_FOR_EACH (br, node, &all_bridges) {
- struct iface *iface;
- struct port *port;
-
- br_refresh_stp_status(br);
-
- HMAP_FOR_EACH (port, hmap_node, &br->ports) {
- port_refresh_stp_status(port);
- }
-
- HMAP_FOR_EACH (iface, name_node, &br->iface_by_name) {
- enum netdev_flags flags;
- struct smap smap;
- const char *link_state;
- int64_t link_resets;
- int current, error;
-
- if (iface_is_synthetic(iface)) {
- continue;
- }
-
- current = ofproto_port_is_lacp_current(br->ofproto,
- iface->ofp_port);
- if (current >= 0) {
- bool bl = current;
- ovsrec_interface_set_lacp_current(iface->cfg, &bl, 1);
- } else {
- ovsrec_interface_set_lacp_current(iface->cfg, NULL, 0);
- }
-
- error = netdev_get_flags(iface->netdev, &flags);
- if (!error) {
- const char *state = flags & NETDEV_UP ? "up" : "down";
- ovsrec_interface_set_admin_state(iface->cfg, state);
- } else {
- ovsrec_interface_set_admin_state(iface->cfg, NULL);
- }
-
- link_state = netdev_get_carrier(iface->netdev) ? "up" : "down";
- ovsrec_interface_set_link_state(iface->cfg, link_state);
-
- link_resets = netdev_get_carrier_resets(iface->netdev);
- ovsrec_interface_set_link_resets(iface->cfg, &link_resets, 1);
-
- iface_refresh_cfm_stats(iface);
-
- smap_init(&smap);
- ofproto_port_get_bfd_status(br->ofproto, iface->ofp_port,
- &smap);
- ovsrec_interface_set_bfd_status(iface->cfg, &smap);
- smap_destroy(&smap);
- }
- }
- }
-
- status = ovsdb_idl_txn_commit(instant_txn);
- if (status != TXN_INCOMPLETE) {
- ovsdb_idl_txn_destroy(instant_txn);
- instant_txn = NULL;
- }
- if (status == TXN_UNCHANGED) {
- instant_stats_could_have_changed = false;
- }
-}
-
-static void
-instant_stats_wait(void)
-{
- if (instant_txn) {
- ovsdb_idl_txn_wait(instant_txn);
- } else if (instant_stats_could_have_changed) {
- poll_timer_wait_until(instant_next_txn);
- }
-}
-\f
static void
bridge_run__(void)
{
bool vlan_splinters_changed;
struct bridge *br;
+ int stats_interval;
ovsrec_open_vswitch_init(&null_cfg);
* with the current situation of multiple ovs-vswitchd daemons,
* disable system stats collection. */
system_stats_enable(false);
+ /* This prevents the process from constantly waking up on
+ * connectivity seq. */
+ connectivity_seqno = seq_read(connectivity_seq_get());
return;
} else if (!ovsdb_idl_has_lock(idl)) {
return;
}
}
+ /* Statistics update interval should always be greater than or equal to
+ * 5000 ms. */
+ if (cfg) {
+ stats_interval = MAX(smap_get_int(&cfg->other_config,
+ "stats-update-interval",
+ 5000), 5000);
+ } else {
+ stats_interval = 5000;
+ }
+ if (stats_timer_interval != stats_interval) {
+ stats_timer_interval = stats_interval;
+ stats_timer = LLONG_MIN;
+ }
+
/* Refresh interface and mirror stats if necessary. */
- if (time_msec() >= iface_stats_timer) {
- if (cfg) {
- struct ovsdb_idl_txn *txn;
+ if (time_msec() >= stats_timer && cfg) {
+ enum ovsdb_idl_txn_status status;
- txn = ovsdb_idl_txn_create(idl);
+ /* Rate limit the update. Do not start a new update if the
+ * previous one is not done. */
+ if (!stats_txn) {
+ stats_txn = ovsdb_idl_txn_create(idl);
HMAP_FOR_EACH (br, node, &all_bridges) {
struct port *port;
struct mirror *m;
LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
iface_refresh_stats(iface);
- iface_refresh_status(iface);
}
-
port_refresh_stp_stats(port);
}
-
HMAP_FOR_EACH (m, hmap_node, &br->mirrors) {
mirror_refresh_stats(m);
}
-
}
refresh_controller_status();
- ovsdb_idl_txn_commit(txn);
- ovsdb_idl_txn_destroy(txn); /* XXX */
}
- iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
+ status = ovsdb_idl_txn_commit(stats_txn);
+ if (status != TXN_INCOMPLETE) {
+ stats_timer = time_msec() + stats_timer_interval;
+ ovsdb_idl_txn_destroy(stats_txn);
+ stats_txn = NULL;
+ }
+ }
+
+ if (!status_txn) {
+ uint64_t seq;
+
+ /* Check the need to update status. */
+ seq = seq_read(connectivity_seq_get());
+ if (seq != connectivity_seqno || force_status_commit) {
+ connectivity_seqno = seq;
+ status_txn = ovsdb_idl_txn_create(idl);
+ HMAP_FOR_EACH (br, node, &all_bridges) {
+ struct port *port;
+
+ br_refresh_stp_status(br);
+ HMAP_FOR_EACH (port, hmap_node, &br->ports) {
+ struct iface *iface;
+
+ port_refresh_stp_status(port);
+ port_refresh_bond_status(port, force_status_commit);
+ LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
+ iface_refresh_netdev_status(iface);
+ iface_refresh_ofproto_status(iface);
+ }
+ }
+ }
+ }
+ }
+
+ if (status_txn) {
+ enum ovsdb_idl_txn_status status;
+
+ status = ovsdb_idl_txn_commit(status_txn);
+
+ /* If the transaction is incomplete or fails, 'status_txn'
+ * needs to be committed next iteration of bridge_run() even if
+ * connectivity or netdev sequence numbers do not change. */
+ if (status == TXN_SUCCESS || status == TXN_UNCHANGED)
+ {
+ force_status_commit = false;
+ } else {
+ force_status_commit = true;
+ }
+
+ /* Do not destroy "status_txn" if the transaction is
+ * "TXN_INCOMPLETE". */
+ if (status != TXN_INCOMPLETE) {
+ ovsdb_idl_txn_destroy(status_txn);
+ status_txn = NULL;
+ }
}
run_system_stats();
- instant_stats_run();
}
void
HMAP_FOR_EACH (br, node, &all_bridges) {
ofproto_wait(br->ofproto);
}
- poll_timer_wait_until(iface_stats_timer);
+
+ poll_timer_wait_until(stats_timer);
+ }
+
+ /* If the status database transaction is 'TXN_INCOMPLETE' or is
+ * unsuccessful, register a timeout in 'STATUS_CHECK_AGAIN_MSEC'. Else,
+ * wait on the global connectivity sequence number. Note, this also helps
+ * batch multiple status changes into one transaction. */
+ if (force_status_commit) {
+ poll_timer_wait_until(time_msec() + STATUS_CHECK_AGAIN_MSEC);
+ } else {
+ seq_wait(connectivity_seq_get(), connectivity_seqno);
}
system_stats_wait();
- instant_stats_wait();
}
/* Adds some memory usage statistics for bridges into 'usage', for use with
oc->rate_limit = 0;
oc->burst_limit = 0;
oc->enable_async_msgs = true;
+ oc->dscp = 0;
}
/* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
\f
/* Port functions. */
-static void iface_destroy__(struct iface *);
-
static struct port *
port_create(struct bridge *br, const struct ovsrec_port *cfg)
{
{
const char *detect_s;
struct iface *iface;
+ const char *mac_s;
int miimon_interval;
s->name = port->name;
LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
netdev_set_miimon_interval(iface->netdev, miimon_interval);
}
+
+ mac_s = port->cfg->bond_active_slave;
+ if (!mac_s || !ovs_scan(mac_s, ETH_ADDR_SCAN_FMT,
+ ETH_ADDR_SCAN_ARGS(s->active_slave_mac))) {
+ /* OVSDB did not store the last active interface */
+ memset(s->active_slave_mac, 0, sizeof(s->active_slave_mac));
+ }
}
/* Returns true if 'port' is synthetic, that is, if we constructed it locally
}
if (iface->ofp_port != OFPP_NONE) {
- const struct ofproto_port_queue *port_queues = queues_buf.data;
- size_t n_queues = queues_buf.size / sizeof *port_queues;
+ const struct ofproto_port_queue *port_queues = ofpbuf_data(&queues_buf);
+ size_t n_queues = ofpbuf_size(&queues_buf) / sizeof *port_queues;
ofproto_port_set_queues(iface->port->bridge->ofproto, iface->ofp_port,
port_queues, n_queues);