#include <arpa/inet.h>
#include <ctype.h>
#include <inttypes.h>
+#include <sys/socket.h>
#include <net/if.h>
#include <openflow/openflow.h>
#include <signal.h>
#include "dynamic-string.h"
#include "flow.h"
#include "hash.h"
+#include "jsonrpc.h"
#include "list.h"
#include "mac-learning.h"
#include "netdev.h"
#include "ofpbuf.h"
#include "ofproto/netflow.h"
#include "ofproto/ofproto.h"
+#include "ovsdb-data.h"
#include "packets.h"
#include "poll-loop.h"
#include "port-array.h"
#include "vconn.h"
#include "vswitchd/vswitch-idl.h"
#include "xenserver.h"
+#include "vlog.h"
#include "xtoxll.h"
+#include "sflow_api.h"
-#define THIS_MODULE VLM_bridge
-#include "vlog.h"
+VLOG_DEFINE_THIS_MODULE(bridge)
struct dst {
uint16_t vlan;
long long delay_expires; /* Time after which 'enabled' may change. */
/* These members are valid only after bridge_reconfigure() causes them to
- * be initialized.*/
+ * be initialized. */
int dp_ifidx; /* Index within kernel datapath. */
struct netdev *netdev; /* Network device. */
bool enabled; /* May be chosen for flows? */
-
- /* This member is only valid *during* bridge_reconfigure(). */
const struct ovsrec_interface *cfg;
};
struct bridge *bridge;
size_t idx;
char *name;
+ struct uuid uuid; /* UUID of this "mirror" record in database. */
/* Selection criteria. */
struct shash src_ports; /* Name is port name; data is always NULL. */
struct bridge *bridge;
size_t port_idx;
int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
- unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
+ unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
+ * NULL if all VLANs are trunked. */
+ const struct ovsrec_port *cfg;
char *name;
/* An ordinary bridge port has 1 interface.
int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */
bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
+ long long int bond_next_fake_iface_update; /* Time of next update. */
+ int bond_rebalance_interval; /* Interval between rebalances, in ms. */
+ long long int bond_next_rebalance; /* Next rebalancing time. */
/* Port mirroring info. */
mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
bool is_mirror_output_port; /* Does port mirroring send frames here? */
-
- /* This member is only valid *during* bridge_reconfigure(). */
- const struct ovsrec_port *cfg;
};
#define DP_MAX_PORTS 255
struct list node; /* Node in global list of bridges. */
char *name; /* User-specified arbitrary name. */
struct mac_learning *ml; /* MAC learning table. */
- bool sent_config_request; /* Successfully sent config request? */
uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
-
- /* Support for remote controllers. */
- char *controller; /* NULL if there is no remote controller;
- * "discover" to do controller discovery;
- * otherwise a vconn name. */
+ const struct ovsrec_bridge *cfg;
/* OpenFlow switch processing. */
struct ofproto *ofproto; /* OpenFlow switch. */
/* Bridge ports. */
struct port **ports;
size_t n_ports, allocated_ports;
+ struct shash iface_by_name; /* "struct iface"s indexed by name. */
+ struct shash port_by_name; /* "struct port"s indexed by name. */
/* Bonding. */
bool has_bonded_ports;
- long long int bond_next_rebalance;
/* Flow tracking. */
bool flush;
- /* Flow statistics gathering. */
- time_t next_stats_request;
-
/* Port mirroring. */
struct mirror *mirrors[MAX_MIRRORS];
-
- /* This member is only valid *during* bridge_reconfigure(). */
- const struct ovsrec_bridge *cfg;
};
/* List of all bridges. */
static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
-/* Maximum number of datapaths. */
-enum { DP_MAX = 256 };
+/* OVSDB IDL used to obtain configuration. */
+static struct ovsdb_idl *idl;
+
+/* Each time this timer expires, the bridge fetches statistics for every
+ * interface and pushes them into the database. */
+#define IFACE_STATS_INTERVAL (5 * 1000) /* In milliseconds. */
+static long long int iface_stats_timer = LLONG_MIN;
static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
static void bridge_destroy(struct bridge *);
static struct bridge *bridge_lookup(const char *name);
static unixctl_cb_func bridge_unixctl_dump_flows;
+static unixctl_cb_func bridge_unixctl_reconnect;
static int bridge_run_one(struct bridge *);
-static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *,
- struct bridge *);
-static void bridge_reconfigure_controller(const struct ovsrec_open_vswitch *,
- struct bridge *);
+static size_t bridge_get_controllers(const struct bridge *br,
+ struct ovsrec_controller ***controllersp);
+static void bridge_reconfigure_one(struct bridge *);
+static void bridge_reconfigure_remotes(struct bridge *,
+ const struct sockaddr_in *managers,
+ size_t n_managers);
static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
static void bridge_fetch_dp_ifaces(struct bridge *);
static void bridge_flush(struct bridge *);
static struct port *port_create(struct bridge *, const char *name);
static void port_reconfigure(struct port *, const struct ovsrec_port *);
+static void port_del_ifaces(struct port *, const struct ovsrec_port *);
static void port_destroy(struct port *);
static struct port *port_lookup(const struct bridge *, const char *name);
static struct iface *port_lookup_iface(const struct port *, const char *name);
static void port_update_vlan_compat(struct port *);
static void port_update_bonding(struct port *);
-static struct mirror *mirror_create(struct bridge *, const char *name);
+static void mirror_create(struct bridge *, struct ovsrec_mirror *);
static void mirror_destroy(struct mirror *);
static void mirror_reconfigure(struct bridge *);
static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
uint16_t dp_ifidx);
static bool iface_is_internal(const struct bridge *, const char *name);
static void iface_set_mac(struct iface *);
+static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
/* Hooks into ofproto processing. */
static struct ofhooks bridge_ofhooks;
\f
/* Public functions. */
-/* Adds the name of each interface used by a bridge, including local and
- * internal ports, to 'svec'. */
+/* Initializes the bridge module, configuring it to obtain its configuration
+ * from an OVSDB server accessed over 'remote', which should be a string in a
+ * form acceptable to ovsdb_idl_create(). */
void
-bridge_get_ifaces(struct svec *svec)
+bridge_init(const char *remote)
{
- struct bridge *br, *next;
- size_t i, j;
+ /* Create connection to database. */
+ idl = ovsdb_idl_create(remote, &ovsrec_idl_class);
- LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
- for (i = 0; i < br->n_ports; i++) {
- struct port *port = br->ports[i];
+ ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_cur_cfg);
+ ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_statistics);
+ ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
- for (j = 0; j < port->n_ifaces; j++) {
- struct iface *iface = port->ifaces[j];
- if (iface->dp_ifidx < 0) {
- VLOG_ERR("%s interface not in datapath %s, ignoring",
- iface->name, dpif_name(br->dpif));
- } else {
- if (iface->dp_ifidx != ODPP_LOCAL) {
- svec_add(svec, iface->name);
- }
- }
- }
- }
- }
+ ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
+
+ ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
+ ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
+
+ ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_ofport);
+ ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_statistics);
+ ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
+
+ /* Register unixctl commands. */
+ unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
+ unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
+ NULL);
+ unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
+ NULL);
+ bond_init();
}
-void
-bridge_init(const struct ovsrec_open_vswitch *cfg)
+/* Performs configuration that is only necessary once at ovs-vswitchd startup,
+ * but for which the ovs-vswitchd configuration 'cfg' is required. */
+static void
+bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
{
+ static bool already_configured_once;
struct svec bridge_names;
struct svec dpif_names, dpif_types;
size_t i;
- unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
+ /* Only do this once per ovs-vswitchd run. */
+ if (already_configured_once) {
+ return;
+ }
+ already_configured_once = true;
+
+ iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
+ /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
svec_init(&bridge_names);
for (i = 0; i < cfg->n_bridges; i++) {
svec_add(&bridge_names, cfg->bridges[i]->name);
}
svec_sort(&bridge_names);
+ /* Iterate over all system dpifs and delete any of them that do not appear
+ * in 'cfg'. */
svec_init(&dpif_names);
svec_init(&dpif_types);
dp_enumerate_types(&dpif_types);
dp_enumerate_names(dpif_types.names[i], &dpif_names);
+ /* For each dpif... */
for (j = 0; j < dpif_names.n; j++) {
retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
if (!retval) {
struct svec all_names;
size_t k;
+ /* ...check whether any of its names is in 'bridge_names'. */
svec_init(&all_names);
dpif_get_all_names(dpif, &all_names);
for (k = 0; k < all_names.n; k++) {
goto found;
}
}
+
+ /* No. Delete the dpif. */
dpif_delete(dpif);
+
found:
svec_destroy(&all_names);
dpif_close(dpif);
}
}
}
+ svec_destroy(&bridge_names);
svec_destroy(&dpif_names);
svec_destroy(&dpif_types);
-
- unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
- NULL);
-
- bond_init();
- bridge_reconfigure(cfg);
-}
-
-#ifdef HAVE_OPENSSL
-static bool
-config_string_change(const char *value, char **valuep)
-{
- if (value && (!*valuep || strcmp(value, *valuep))) {
- free(*valuep);
- *valuep = xstrdup(value);
- return true;
- } else {
- return false;
- }
}
-static void
-bridge_configure_ssl(const struct ovsrec_ssl *ssl)
-{
- /* XXX SSL should be configurable on a per-bridge basis.
- * XXX should be possible to de-configure SSL. */
- static char *private_key_file;
- static char *certificate_file;
- static char *cacert_file;
- struct stat s;
-
- if (!ssl) {
- /* XXX We can't un-set SSL settings. */
- return;
- }
-
- if (config_string_change(ssl->private_key, &private_key_file)) {
- stream_ssl_set_private_key_file(private_key_file);
- }
-
- if (config_string_change(ssl->certificate, &certificate_file)) {
- stream_ssl_set_certificate_file(certificate_file);
- }
-
- /* We assume that even if the filename hasn't changed, if the CA cert
- * file has been removed, that we want to move back into
- * boot-strapping mode. This opens a small security hole, because
- * the old certificate will still be trusted until vSwitch is
- * restarted. We may want to address this in vconn's SSL library. */
- if (config_string_change(ssl->ca_cert, &cacert_file)
- || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) {
- stream_ssl_set_ca_cert_file(cacert_file, ssl->bootstrap_ca_cert);
- }
-}
-#endif
-
/* Attempt to create the network device 'iface_name' through the netdev
* library. */
static int
set_up_iface(const struct ovsrec_interface *iface_cfg, struct iface *iface,
bool create)
{
- struct shash_node *node;
struct shash options;
int error = 0;
size_t i;
memset(&netdev_options, 0, sizeof netdev_options);
netdev_options.name = iface_cfg->name;
- netdev_options.type = iface_cfg->type;
+ if (!strcmp(iface_cfg->type, "internal")) {
+ /* An "internal" config type maps to a netdev "system" type. */
+ netdev_options.type = "system";
+ } else {
+ netdev_options.type = iface_cfg->type;
+ }
netdev_options.args = &options;
netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
- netdev_options.may_create = true;
- if (iface_is_internal(iface->port->bridge, iface_cfg->name)) {
- netdev_options.may_open = true;
- }
error = netdev_open(&netdev_options, &iface->netdev);
const char *iface_type = iface_cfg->type && strlen(iface_cfg->type)
? iface_cfg->type : NULL;
+ /* An "internal" config type maps to a netdev "system" type. */
+ if (iface_type && !strcmp(iface_type, "internal")) {
+ iface_type = "system";
+ }
+
if (!iface_type || !strcmp(netdev_type, iface_type)) {
error = netdev_reconfigure(iface->netdev, &options);
} else {
error = EINVAL;
}
}
-
- SHASH_FOR_EACH (node, &options) {
- free(node->data);
- }
- shash_destroy(&options);
+ shash_destroy_free_data(&options);
return error;
}
}
static bool
-check_iface_netdev(struct bridge *br UNUSED, struct iface *iface,
- void *aux UNUSED)
+check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface,
+ void *aux OVS_UNUSED)
{
if (!iface->netdev) {
int error = set_up_iface(iface->cfg, iface, true);
}
static bool
-check_iface_dp_ifidx(struct bridge *br, struct iface *iface, void *aux UNUSED)
+check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
+ void *aux OVS_UNUSED)
{
if (iface->dp_ifidx >= 0) {
VLOG_DBG("%s has interface %s on port %d",
}
static bool
-set_iface_properties(struct bridge *br UNUSED, struct iface *iface,
- void *aux UNUSED)
+set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
+ void *aux OVS_UNUSED)
{
/* Set policing attributes. */
netdev_set_policing(iface->netdev,
}
}
-void
+/* 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()).
+ *
+ * You may be asking yourself "why does ovs-vswitchd care?", because
+ * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
+ * should not be and in fact is not directly involved in that. But
+ * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
+ * it has to tell in-band control where the managers are to enable that.
+ */
+static void
+collect_managers(const struct ovsrec_open_vswitch *ovs_cfg,
+ struct sockaddr_in **managersp, size_t *n_managersp)
+{
+ struct sockaddr_in *managers = NULL;
+ size_t n_managers = 0;
+
+ if (ovs_cfg->n_managers > 0) {
+ size_t i;
+
+ managers = xmalloc(ovs_cfg->n_managers * sizeof *managers);
+ for (i = 0; i < ovs_cfg->n_managers; i++) {
+ const char *name = ovs_cfg->managers[i];
+ struct sockaddr_in *sin = &managers[i];
+
+ if ((!strncmp(name, "tcp:", 4)
+ && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) ||
+ (!strncmp(name, "ssl:", 4)
+ && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) {
+ n_managers++;
+ }
+ }
+ }
+
+ *managersp = managers;
+ *n_managersp = n_managers;
+}
+
+static void
bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
{
- struct ovsdb_idl_txn *txn;
struct shash old_br, new_br;
struct shash_node *node;
struct bridge *br, *next;
+ struct sockaddr_in *managers;
+ size_t n_managers;
size_t i;
+ int sflow_bridge_number;
COVERAGE_INC(bridge_reconfigure);
- txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl);
+ collect_managers(ovs_cfg, &managers, &n_managers);
/* Collect old and new bridges. */
shash_init(&old_br);
shash_destroy(&old_br);
shash_destroy(&new_br);
-#ifdef HAVE_OPENSSL
- /* Configure SSL. */
- bridge_configure_ssl(ovs_cfg->ssl);
-#endif
-
/* Reconfigure all bridges. */
LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
- bridge_reconfigure_one(ovs_cfg, br);
+ bridge_reconfigure_one(br);
}
/* Add and delete ports on all datapaths.
shash_init(&cur_ifaces);
for (i = 0; i < n_dpif_ports; i++) {
const char *name = dpif_ports[i].devname;
- if (!shash_find(&cur_ifaces, name)) {
- shash_add(&cur_ifaces, name, NULL);
- }
+ shash_add_once(&cur_ifaces, name, NULL);
}
free(dpif_ports);
shash_destroy(&cur_ifaces);
shash_destroy(&want_ifaces);
}
+ sflow_bridge_number = 0;
LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
uint8_t ea[8];
uint64_t dpid;
dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
ofproto_set_datapath_id(br->ofproto, dpid);
- dpid_string = xasprintf("%012"PRIx64, dpid);
+ dpid_string = xasprintf("%016"PRIx64, dpid);
ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
free(dpid_string);
ofproto_set_netflow(br->ofproto, NULL);
}
+ /* Set sFlow configuration on this bridge. */
+ if (br->cfg->sflow) {
+ const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
+ struct ovsrec_controller **controllers;
+ struct ofproto_sflow_options oso;
+ size_t n_controllers;
+ size_t i;
+
+ memset(&oso, 0, sizeof oso);
+
+ oso.targets.n = sflow_cfg->n_targets;
+ oso.targets.names = sflow_cfg->targets;
+
+ oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
+ if (sflow_cfg->sampling) {
+ oso.sampling_rate = *sflow_cfg->sampling;
+ }
+
+ oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
+ if (sflow_cfg->polling) {
+ oso.polling_interval = *sflow_cfg->polling;
+ }
+
+ oso.header_len = SFL_DEFAULT_HEADER_SIZE;
+ if (sflow_cfg->header) {
+ oso.header_len = *sflow_cfg->header;
+ }
+
+ oso.sub_id = sflow_bridge_number++;
+ oso.agent_device = sflow_cfg->agent;
+
+ oso.control_ip = NULL;
+ n_controllers = bridge_get_controllers(br, &controllers);
+ for (i = 0; i < n_controllers; i++) {
+ if (controllers[i]->local_ip) {
+ oso.control_ip = controllers[i]->local_ip;
+ break;
+ }
+ }
+ ofproto_set_sflow(br->ofproto, &oso);
+
+ /* Do not destroy oso.targets because it is owned by sflow_cfg. */
+ } else {
+ ofproto_set_sflow(br->ofproto, NULL);
+ }
+
/* Update the controller and related settings. It would be more
* straightforward to call this from bridge_reconfigure_one(), but we
* can't do it there for two reasons. First, and most importantly, at
* yet; when a controller is configured, resetting the datapath ID will
* immediately disconnect from the controller, so it's better to set
* the datapath ID before the controller. */
- bridge_reconfigure_controller(ovs_cfg, br);
+ bridge_reconfigure_remotes(br, managers, n_managers);
}
LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
for (i = 0; i < br->n_ports; i++) {
struct port *port = br->ports[i];
+ int j;
port_update_vlan_compat(port);
port_update_bonding(port);
+
+ for (j = 0; j < port->n_ifaces; j++) {
+ iface_update_qos(port->ifaces[j], port->cfg->qos);
+ }
}
}
LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
iterate_and_prune_ifaces(br, set_iface_properties, NULL);
}
- ovsrec_open_vswitch_set_cur_cfg(ovs_cfg, ovs_cfg->next_cfg);
+ free(managers);
+}
- ovsdb_idl_txn_commit(txn);
- ovsdb_idl_txn_destroy(txn); /* XXX */
+static const char *
+get_ovsrec_key_value(const struct ovsdb_idl_row *row,
+ const struct ovsdb_idl_column *column,
+ const char *key)
+{
+ const struct ovsdb_datum *datum;
+ union ovsdb_atom atom;
+ unsigned int idx;
+
+ datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
+ atom.string = (char *) key;
+ idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
+ return idx == UINT_MAX ? NULL : datum->values[idx].string;
}
static const char *
bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
{
- size_t i;
-
- for (i = 0; i < br_cfg->n_other_config; i++) {
- if (!strcmp(br_cfg->key_other_config[i], key)) {
- return br_cfg->value_other_config[i];
- }
- }
- return NULL;
+ return get_ovsrec_key_value(&br_cfg->header_,
+ &ovsrec_bridge_col_other_config, key);
}
static void
return eth_addr_to_uint64(hash);
}
-int
+static void
+iface_refresh_stats(struct iface *iface)
+{
+ struct iface_stat {
+ char *name;
+ int offset;
+ };
+ static const struct iface_stat iface_stats[] = {
+ { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
+ { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
+ { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
+ { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
+ { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
+ { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
+ { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
+ { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
+ { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
+ { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
+ { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
+ { "collisions", offsetof(struct netdev_stats, collisions) },
+ };
+ enum { N_STATS = ARRAY_SIZE(iface_stats) };
+ const struct iface_stat *s;
+
+ char *keys[N_STATS];
+ int64_t values[N_STATS];
+ int n;
+
+ struct netdev_stats stats;
+
+ /* Intentionally ignore return value, since errors will set 'stats' to
+ * all-1s, and we will deal with that correctly below. */
+ netdev_get_stats(iface->netdev, &stats);
+
+ n = 0;
+ for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
+ uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
+ if (value != UINT64_MAX) {
+ keys[n] = s->name;
+ values[n] = value;
+ n++;
+ }
+ }
+
+ ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
+}
+
+void
bridge_run(void)
{
- struct bridge *br, *next;
- int retval;
+ const struct ovsrec_open_vswitch *cfg;
- retval = 0;
- LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
+ bool datapath_destroyed;
+ bool database_changed;
+ struct bridge *br;
+
+ /* Let each bridge do the work that it needs to do. */
+ datapath_destroyed = false;
+ LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
int error = bridge_run_one(br);
if (error) {
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
"forcing reconfiguration", br->name);
- if (!retval) {
- retval = error;
+ datapath_destroyed = true;
+ }
+ }
+
+ /* (Re)configure if necessary. */
+ database_changed = ovsdb_idl_run(idl);
+ cfg = ovsrec_open_vswitch_first(idl);
+ if (database_changed || datapath_destroyed) {
+ if (cfg) {
+ struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
+
+ bridge_configure_once(cfg);
+ bridge_reconfigure(cfg);
+
+ ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
+ ovsdb_idl_txn_commit(txn);
+ ovsdb_idl_txn_destroy(txn); /* XXX */
+ } else {
+ /* We still need to reconfigure to avoid dangling pointers to
+ * now-destroyed ovsrec structures inside bridge data. */
+ static const struct ovsrec_open_vswitch null_cfg;
+
+ bridge_reconfigure(&null_cfg);
+ }
+ }
+
+#ifdef HAVE_OPENSSL
+ /* Re-configure SSL. We do this on every trip through the main loop,
+ * instead of just when the database changes, because the contents of the
+ * key and certificate files can change without the database changing. */
+ if (cfg && cfg->ssl) {
+ const struct ovsrec_ssl *ssl = cfg->ssl;
+
+ stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
+ stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
+ }
+#endif
+
+ /* Refresh interface stats if necessary. */
+ if (time_msec() >= iface_stats_timer) {
+ struct ovsdb_idl_txn *txn;
+
+ txn = ovsdb_idl_txn_create(idl);
+ LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
+ size_t i;
+
+ for (i = 0; i < br->n_ports; i++) {
+ struct port *port = br->ports[i];
+ size_t j;
+
+ for (j = 0; j < port->n_ifaces; j++) {
+ struct iface *iface = port->ifaces[j];
+ iface_refresh_stats(iface);
+ }
}
}
+ ovsdb_idl_txn_commit(txn);
+ ovsdb_idl_txn_destroy(txn); /* XXX */
+
+ iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
}
- return retval;
}
void
LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
ofproto_wait(br->ofproto);
- if (br->controller) {
+ if (ofproto_has_primary_controller(br->ofproto)) {
continue;
}
mac_learning_wait(br->ml);
bond_wait(br);
}
+ ovsdb_idl_wait(idl);
+ poll_timer_wait_until(iface_stats_timer);
}
/* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
/* Bridge unixctl user interface functions. */
static void
bridge_unixctl_fdb_show(struct unixctl_conn *conn,
- const char *args, void *aux UNUSED)
+ const char *args, void *aux OVS_UNUSED)
{
struct ds ds = DS_EMPTY_INITIALIZER;
const struct bridge *br;
br->name = xstrdup(br_cfg->name);
br->cfg = br_cfg;
br->ml = mac_learning_create();
- br->sent_config_request = false;
- eth_addr_random(br->default_ea);
+ eth_addr_nicira_random(br->default_ea);
port_array_init(&br->ifaces);
+ shash_init(&br->port_by_name);
+ shash_init(&br->iface_by_name);
+
br->flush = false;
- br->bond_next_rebalance = time_msec() + 10000;
list_push_back(&all_bridges, &br->node);
}
dpif_close(br->dpif);
ofproto_destroy(br->ofproto);
- free(br->controller);
mac_learning_destroy(br->ml);
port_array_destroy(&br->ifaces);
+ shash_destroy(&br->port_by_name);
+ shash_destroy(&br->iface_by_name);
free(br->ports);
free(br->name);
free(br);
return NULL;
}
-bool
-bridge_exists(const char *name)
-{
- return bridge_lookup(name) ? true : false;
-}
-
-uint64_t
-bridge_get_datapathid(const char *name)
-{
- struct bridge *br = bridge_lookup(name);
- return br ? ofproto_get_datapath_id(br->ofproto) : 0;
-}
-
/* Handle requests for a listing of all flows known by the OpenFlow
* stack, including those normally hidden. */
static void
bridge_unixctl_dump_flows(struct unixctl_conn *conn,
- const char *args, void *aux UNUSED)
+ const char *args, void *aux OVS_UNUSED)
{
struct bridge *br;
struct ds results;
ds_destroy(&results);
}
+/* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
+ * connections and reconnect. If BRIDGE is not specified, then all bridges
+ * drop their controller connections and reconnect. */
+static void
+bridge_unixctl_reconnect(struct unixctl_conn *conn,
+ const char *args, void *aux OVS_UNUSED)
+{
+ struct bridge *br;
+ if (args[0] != '\0') {
+ br = bridge_lookup(args);
+ if (!br) {
+ unixctl_command_reply(conn, 501, "Unknown bridge");
+ return;
+ }
+ ofproto_reconnect_controllers(br->ofproto);
+ } else {
+ LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
+ ofproto_reconnect_controllers(br->ofproto);
+ }
+ }
+ unixctl_command_reply(conn, 200, NULL);
+}
+
static int
bridge_run_one(struct bridge *br)
{
return error;
}
-static const struct ovsrec_controller *
-bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg,
- const struct bridge *br)
+static size_t
+bridge_get_controllers(const struct bridge *br,
+ struct ovsrec_controller ***controllersp)
{
- const struct ovsrec_controller *controller;
+ struct ovsrec_controller **controllers;
+ size_t n_controllers;
- controller = (br->cfg->controller ? br->cfg->controller
- : ovs_cfg->controller ? ovs_cfg->controller
- : NULL);
+ controllers = br->cfg->controller;
+ n_controllers = br->cfg->n_controller;
- if (controller && !strcmp(controller->target, "none")) {
- return NULL;
+ if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
+ controllers = NULL;
+ n_controllers = 0;
}
- return controller;
-}
-
-static bool
-check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_)
-{
- struct svec *ifaces = ifaces_;
- if (!svec_contains(ifaces, iface->name)) {
- svec_add(ifaces, iface->name);
- svec_sort(ifaces);
- return true;
- } else {
- VLOG_ERR("bridge %s: %s interface is on multiple ports, "
- "removing from %s",
- br->name, iface->name, iface->port->name);
- return false;
+ if (controllersp) {
+ *controllersp = controllers;
}
+ return n_controllers;
}
static void
-bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg,
- struct bridge *br)
+bridge_reconfigure_one(struct bridge *br)
{
struct shash old_ports, new_ports;
- struct svec ifaces;
- struct svec listeners, old_listeners;
struct svec snoops, old_snoops;
struct shash_node *node;
+ enum ofproto_fail_mode fail_mode;
size_t i;
/* Collect old ports. */
* user didn't specify one.
*
* XXX perhaps we should synthesize a port ourselves in this case. */
- if (bridge_get_controller(ovs_cfg, br)) {
+ if (bridge_get_controllers(br, NULL)) {
char local_name[IF_NAMESIZE];
int error;
}
}
- /* Get rid of deleted ports and add new ports. */
+ /* Get rid of deleted ports.
+ * Get rid of deleted interfaces on ports that still exist. */
SHASH_FOR_EACH (node, &old_ports) {
- if (!shash_find(&new_ports, node->name)) {
- port_destroy(node->data);
+ struct port *port = node->data;
+ const struct ovsrec_port *port_cfg;
+
+ port_cfg = shash_find_data(&new_ports, node->name);
+ if (!port_cfg) {
+ port_destroy(port);
+ } else {
+ port_del_ifaces(port, port_cfg);
}
}
+
+ /* Create new ports.
+ * Add new interfaces to existing ports.
+ * Reconfigure existing ports. */
SHASH_FOR_EACH (node, &new_ports) {
struct port *port = shash_find_data(&old_ports, node->name);
if (!port) {
port = port_create(br, node->name);
}
+
port_reconfigure(port, node->data);
+ if (!port->n_ifaces) {
+ VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
+ br->name, port->name);
+ port_destroy(port);
+ }
}
shash_destroy(&old_ports);
shash_destroy(&new_ports);
- /* Check and delete duplicate interfaces. */
- svec_init(&ifaces);
- iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces);
- svec_destroy(&ifaces);
+ /* Set the fail-mode */
+ fail_mode = !br->cfg->fail_mode
+ || !strcmp(br->cfg->fail_mode, "standalone")
+ ? OFPROTO_FAIL_STANDALONE
+ : OFPROTO_FAIL_SECURE;
+ if (ofproto_get_fail_mode(br->ofproto) != fail_mode
+ && !ofproto_has_primary_controller(br->ofproto)) {
+ ofproto_flush_flows(br->ofproto);
+ }
+ ofproto_set_fail_mode(br->ofproto, fail_mode);
/* Delete all flows if we're switching from connected to standalone or vice
* versa. (XXX Should we delete all flows if we are switching from one
* controller to another?) */
-#if 0
- /* Configure OpenFlow management listeners. */
- svec_init(&listeners);
- cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
- if (!listeners.n) {
- svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
- ovs_rundir, br->name));
- } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
- svec_clear(&listeners);
- }
- svec_sort_unique(&listeners);
-
- svec_init(&old_listeners);
- ofproto_get_listeners(br->ofproto, &old_listeners);
- svec_sort_unique(&old_listeners);
-
- if (!svec_equal(&listeners, &old_listeners)) {
- ofproto_set_listeners(br->ofproto, &listeners);
- }
- svec_destroy(&listeners);
- svec_destroy(&old_listeners);
-
/* Configure OpenFlow controller connection snooping. */
svec_init(&snoops);
- cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
- if (!snoops.n) {
- svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
- ovs_rundir, br->name));
- } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
- svec_clear(&snoops);
- }
- svec_sort_unique(&snoops);
-
- svec_init(&old_snoops);
- ofproto_get_snoops(br->ofproto, &old_snoops);
- svec_sort_unique(&old_snoops);
-
- if (!svec_equal(&snoops, &old_snoops)) {
- ofproto_set_snoops(br->ofproto, &snoops);
- }
- svec_destroy(&snoops);
- svec_destroy(&old_snoops);
-#else
- /* Default listener. */
- svec_init(&listeners);
- svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
- ovs_rundir, br->name));
- svec_init(&old_listeners);
- ofproto_get_listeners(br->ofproto, &old_listeners);
- if (!svec_equal(&listeners, &old_listeners)) {
- ofproto_set_listeners(br->ofproto, &listeners);
- }
- svec_destroy(&listeners);
- svec_destroy(&old_listeners);
-
- /* Default snoop. */
- svec_init(&snoops);
svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
ovs_rundir, br->name));
svec_init(&old_snoops);
}
svec_destroy(&snoops);
svec_destroy(&old_snoops);
-#endif
mirror_reconfigure(br);
}
+/* Initializes 'oc' appropriately as a management service controller for
+ * 'br'.
+ *
+ * The caller must free oc->target when it is no longer needed. */
static void
-bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg,
- struct bridge *br)
+bridge_ofproto_controller_for_mgmt(const struct bridge *br,
+ struct ofproto_controller *oc)
{
- char *pfx = xasprintf("bridge.%s.controller", br->name);
- const struct ovsrec_controller *c;
+ oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir, br->name);
+ oc->max_backoff = 0;
+ oc->probe_interval = 60;
+ oc->band = OFPROTO_OUT_OF_BAND;
+ oc->accept_re = NULL;
+ oc->update_resolv_conf = false;
+ oc->rate_limit = 0;
+ oc->burst_limit = 0;
+}
- c = bridge_get_controller(ovs_cfg, br);
- if ((br->controller != NULL) != (c != NULL)) {
- ofproto_flush_flows(br->ofproto);
+/* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
+static void
+bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
+ struct ofproto_controller *oc)
+{
+ oc->target = c->target;
+ oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
+ oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
+ oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
+ ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
+ oc->accept_re = c->discover_accept_regex;
+ oc->update_resolv_conf = c->discover_update_resolv_conf;
+ oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
+ oc->burst_limit = (c->controller_burst_limit
+ ? *c->controller_burst_limit : 0);
+}
+
+/* Configures the IP stack for 'br''s local interface properly according to the
+ * configuration in 'c'. */
+static void
+bridge_configure_local_iface_netdev(struct bridge *br,
+ struct ovsrec_controller *c)
+{
+ struct netdev *netdev;
+ struct in_addr mask, gateway;
+
+ struct iface *local_iface;
+ struct in_addr ip;
+
+ /* Controller discovery does its own TCP/IP configuration later. */
+ if (strcmp(c->target, "discover")) {
+ return;
}
- free(br->controller);
- br->controller = c ? xstrdup(c->target) : NULL;
- if (c) {
- int max_backoff, probe;
- int rate_limit, burst_limit;
+ /* If there's no local interface or no IP address, give up. */
+ local_iface = bridge_get_local_iface(br);
+ if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
+ return;
+ }
- if (!strcmp(c->target, "discover")) {
- ofproto_set_discovery(br->ofproto, true,
- c->discover_accept_regex,
- c->discover_update_resolv_conf);
- } else {
- struct iface *local_iface;
- struct in_addr ip;
- bool in_band;
-
- in_band = (!c->connection_mode
- || !strcmp(c->connection_mode, "out-of-band"));
- ofproto_set_discovery(br->ofproto, false, NULL, NULL);
- ofproto_set_in_band(br->ofproto, in_band);
-
- local_iface = bridge_get_local_iface(br);
- if (local_iface && c->local_ip && inet_aton(c->local_ip, &ip)) {
- struct netdev *netdev = local_iface->netdev;
- struct in_addr ip, mask, gateway;
-
- if (!c->local_netmask || !inet_aton(c->local_netmask, &mask)) {
- mask.s_addr = 0;
- }
- if (!c->local_gateway
- || !inet_aton(c->local_gateway, &gateway)) {
- gateway.s_addr = 0;
- }
+ /* Bring up the local interface. */
+ netdev = local_iface->netdev;
+ netdev_turn_flags_on(netdev, NETDEV_UP, true);
- netdev_turn_flags_on(netdev, NETDEV_UP, true);
- if (!mask.s_addr) {
- mask.s_addr = guess_netmask(ip.s_addr);
- }
- if (!netdev_set_in4(netdev, ip, mask)) {
- VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
- "netmask "IP_FMT,
- br->name, IP_ARGS(&ip.s_addr),
- IP_ARGS(&mask.s_addr));
- }
+ /* Configure the IP address and netmask. */
+ if (!c->local_netmask
+ || !inet_aton(c->local_netmask, &mask)
+ || !mask.s_addr) {
+ mask.s_addr = guess_netmask(ip.s_addr);
+ }
+ if (!netdev_set_in4(netdev, ip, mask)) {
+ VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
+ br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
+ }
- if (gateway.s_addr) {
- if (!netdev_add_router(netdev, gateway)) {
- VLOG_INFO("bridge %s: configured gateway "IP_FMT,
- br->name, IP_ARGS(&gateway.s_addr));
- }
- }
- }
+ /* Configure the default gateway. */
+ if (c->local_gateway
+ && inet_aton(c->local_gateway, &gateway)
+ && gateway.s_addr) {
+ if (!netdev_add_router(netdev, gateway)) {
+ VLOG_INFO("bridge %s: configured gateway "IP_FMT,
+ br->name, IP_ARGS(&gateway.s_addr));
+ }
+ }
+}
+
+static void
+bridge_reconfigure_remotes(struct bridge *br,
+ const struct sockaddr_in *managers,
+ size_t n_managers)
+{
+ struct ovsrec_controller **controllers;
+ size_t n_controllers;
+ bool had_primary;
+
+ struct ofproto_controller *ocs;
+ size_t n_ocs;
+ size_t i;
+
+ ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
+ had_primary = ofproto_has_primary_controller(br->ofproto);
+
+ n_controllers = bridge_get_controllers(br, &controllers);
+
+ ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
+ n_ocs = 0;
+
+ bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
+ for (i = 0; i < n_controllers; i++) {
+ struct ovsrec_controller *c = controllers[i];
+
+ if (!strncmp(c->target, "punix:", 6)
+ || !strncmp(c->target, "unix:", 5)) {
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+
+ /* Prevent remote ovsdb-server users from accessing arbitrary Unix
+ * domain sockets and overwriting arbitrary local files. */
+ VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
+ "\"%s\" due to possibility for remote exploit",
+ dpif_name(br->dpif), c->target);
+ continue;
}
- ofproto_set_failure(br->ofproto,
- (!c->fail_mode
- || !strcmp(c->fail_mode, "standalone")
- || !strcmp(c->fail_mode, "open")));
+ bridge_configure_local_iface_netdev(br, c);
+ bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs++]);
+ }
- probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
- ofproto_set_probe_interval(br->ofproto, probe);
+ ofproto_set_controllers(br->ofproto, ocs, n_ocs);
+ free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
+ free(ocs);
- max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
- ofproto_set_max_backoff(br->ofproto, max_backoff);
+ if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
+ ofproto_flush_flows(br->ofproto);
+ }
- rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
- burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0;
- ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
- } else {
+ /* If there are no controllers and the bridge is in standalone
+ * mode, set up a flow that matches every packet and directs
+ * them to OFPP_NORMAL (which goes to us). Otherwise, the
+ * switch is in secure mode and we won't pass any traffic until
+ * a controller has been defined and it tells us to do so. */
+ if (!n_controllers
+ && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
union ofp_action action;
flow_t flow;
- /* Set up a flow that matches every packet and directs them to
- * OFPP_NORMAL (which goes to us). */
memset(&action, 0, sizeof action);
action.type = htons(OFPAT_OUTPUT);
action.output.len = htons(sizeof action);
action.output.port = htons(OFPP_NORMAL);
memset(&flow, 0, sizeof flow);
- ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
- &action, 1, 0);
-
- ofproto_set_in_band(br->ofproto, false);
- ofproto_set_max_backoff(br->ofproto, 1);
- ofproto_set_probe_interval(br->ofproto, 5);
- ofproto_set_failure(br->ofproto, false);
+ ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
}
- free(pfx);
-
- ofproto_set_controller(br->ofproto, br->controller);
}
static void
port->bond_compat_is_stale = true;
}
+/* Attempts to make the sum of the bond slaves' statistics appear on the fake
+ * bond interface. */
+static void
+bond_update_fake_iface_stats(struct port *port)
+{
+ struct netdev_stats bond_stats;
+ struct netdev *bond_dev;
+ size_t i;
+
+ memset(&bond_stats, 0, sizeof bond_stats);
+
+ for (i = 0; i < port->n_ifaces; i++) {
+ struct netdev_stats slave_stats;
+
+ if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) {
+ /* XXX: We swap the stats here because they are swapped back when
+ * reported by the internal device. The reason for this is
+ * internal devices normally represent packets going into the system
+ * but when used as fake bond device they represent packets leaving
+ * the system. We really should do this in the internal device
+ * itself because changing it here reverses the counts from the
+ * perspective of the switch. However, the internal device doesn't
+ * know what type of device it represents so we have to do it here
+ * for now. */
+ bond_stats.tx_packets += slave_stats.rx_packets;
+ bond_stats.tx_bytes += slave_stats.rx_bytes;
+ bond_stats.rx_packets += slave_stats.tx_packets;
+ bond_stats.rx_bytes += slave_stats.tx_bytes;
+ }
+ }
+
+ if (!netdev_open_default(port->name, &bond_dev)) {
+ netdev_set_stats(bond_dev, &bond_stats);
+ netdev_close(bond_dev);
+ }
+}
+
static void
bond_run(struct bridge *br)
{
bond_enable_slave(iface, !iface->enabled);
}
}
+
+ if (port->bond_fake_iface
+ && time_msec() >= port->bond_next_fake_iface_update) {
+ bond_update_fake_iface_stats(port);
+ port->bond_next_fake_iface_update = time_msec() + 1000;
+ }
}
if (port->bond_compat_is_stale) {
for (j = 0; j < port->n_ifaces; j++) {
struct iface *iface = port->ifaces[j];
if (iface->delay_expires != LLONG_MAX) {
- poll_timer_wait(iface->delay_expires - time_msec());
+ poll_timer_wait_until(iface->delay_expires);
}
}
+ if (port->bond_fake_iface) {
+ poll_timer_wait_until(port->bond_next_fake_iface_update);
+ }
}
}
static bool
port_trunks_vlan(const struct port *port, uint16_t vlan)
{
- return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
+ return (port->vlan < 0
+ && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
}
static bool
return dst - dsts;
}
-static void UNUSED
+static void OVS_UNUSED
print_dsts(const struct dst *dsts, size_t n)
{
for (; n--; dsts++) {
return vlan;
}
+/* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
+ * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
+ * indicate this; newer upstream kernels use gratuitous ARP requests. */
+static bool
+is_gratuitous_arp(const flow_t *flow)
+{
+ return (flow->dl_type == htons(ETH_TYPE_ARP)
+ && eth_addr_is_broadcast(flow->dl_dst)
+ && (flow->nw_proto == ARP_OP_REPLY
+ || (flow->nw_proto == ARP_OP_REQUEST
+ && flow->nw_src == flow->nw_dst)));
+}
+
static void
update_learning_table(struct bridge *br, const flow_t *flow, int vlan,
struct port *in_port)
{
- tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
- vlan, in_port->port_idx);
+ enum grat_arp_lock_type lock_type;
+ tag_type rev_tag;
+
+ /* We don't want to learn from gratuitous ARP packets that are reflected
+ * back over bond slaves so we lock the learning table. */
+ lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE :
+ (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET :
+ GRAT_ARP_LOCK_CHECK;
+
+ rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx,
+ lock_type);
if (rev_tag) {
/* The log messages here could actually be useful in debugging,
* so keep the rate limit relatively high. */
}
}
+/* Determines whether packets in 'flow' within 'br' should be forwarded or
+ * dropped. Returns true if they may be forwarded, false if they should be
+ * dropped.
+ *
+ * If 'have_packet' is true, it indicates that the caller is processing a
+ * received packet. If 'have_packet' is false, then the caller is just
+ * revalidating an existing flow because configuration has changed. Either
+ * way, 'have_packet' only affects logging (there is no point in logging errors
+ * during revalidation).
+ *
+ * Sets '*in_portp' to the input port. This will be a null pointer if
+ * flow->in_port does not designate a known input port (in which case
+ * is_admissible() returns false).
+ *
+ * When returning true, sets '*vlanp' to the effective VLAN of the input
+ * packet, as returned by flow_get_vlan().
+ *
+ * May also add tags to '*tags', although the current implementation only does
+ * so in one special case.
+ */
static bool
-is_bcast_arp_reply(const flow_t *flow)
-{
- return (flow->dl_type == htons(ETH_TYPE_ARP)
- && flow->nw_proto == ARP_OP_REPLY
- && eth_addr_is_broadcast(flow->dl_dst));
-}
-
-/* If the composed actions may be applied to any packet in the given 'flow',
- * returns true. Otherwise, the actions should only be applied to 'packet', or
- * not at all, if 'packet' was NULL. */
-static bool
-process_flow(struct bridge *br, const flow_t *flow,
- const struct ofpbuf *packet, struct odp_actions *actions,
- tag_type *tags, uint16_t *nf_output_iface)
+is_admissible(struct bridge *br, const flow_t *flow, bool have_packet,
+ tag_type *tags, int *vlanp, struct port **in_portp)
{
struct iface *in_iface;
struct port *in_port;
- struct port *out_port = NULL; /* By default, drop the packet/flow. */
int vlan;
- int out_port_idx;
/* Find the interface and port structure for the received packet. */
in_iface = iface_from_dp_ifidx(br, flow->in_port);
if (!in_iface) {
/* No interface? Something fishy... */
- if (packet != NULL) {
+ if (have_packet) {
/* Odd. A few possible reasons here:
*
* - We deleted an interface but there are still a few packets
"interface %"PRIu16, br->name, flow->in_port);
}
- /* Return without adding any actions, to drop packets on this flow. */
- return true;
+ *in_portp = NULL;
+ return false;
}
- in_port = in_iface->port;
- vlan = flow_get_vlan(br, flow, in_port, !!packet);
+ *in_portp = in_port = in_iface->port;
+ *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
if (vlan < 0) {
- goto done;
+ return false;
}
/* Drop frames for reserved multicast addresses. */
if (eth_addr_is_reserved(flow->dl_dst)) {
- goto done;
+ return false;
}
/* Drop frames on ports reserved for mirroring. */
if (in_port->is_mirror_output_port) {
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
- VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
- "which is reserved exclusively for mirroring",
- br->name, in_port->name);
- goto done;
+ if (have_packet) {
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+ VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
+ "%s, which is reserved exclusively for mirroring",
+ br->name, in_port->name);
+ }
+ return false;
}
/* Packets received on bonds need special attention to avoid duplicates. */
if (in_port->n_ifaces > 1) {
int src_idx;
+ bool is_grat_arp_locked;
if (eth_addr_is_multicast(flow->dl_dst)) {
*tags |= in_port->active_iface_tag;
if (in_port->active_iface != in_iface->port_ifidx) {
/* Drop all multicast packets on inactive slaves. */
- goto done;
+ return false;
}
}
/* 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. Broadcast ARP replies are an exception
- * to this rule: the host has moved to another switch. */
- src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
+ * 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. If this is the case, just drop the
+ * packet now. */
+ src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan,
+ &is_grat_arp_locked);
if (src_idx != -1 && src_idx != in_port->port_idx &&
- !is_bcast_arp_reply(flow)) {
- goto done;
+ (!is_gratuitous_arp(flow) || is_grat_arp_locked)) {
+ return false;
}
}
- /* MAC learning. */
- out_port = FLOOD_PORT;
+ return true;
+}
+
+/* If the composed actions may be applied to any packet in the given 'flow',
+ * returns true. Otherwise, the actions should only be applied to 'packet', or
+ * not at all, if 'packet' was NULL. */
+static bool
+process_flow(struct bridge *br, const flow_t *flow,
+ const struct ofpbuf *packet, struct odp_actions *actions,
+ tag_type *tags, uint16_t *nf_output_iface)
+{
+ struct port *in_port;
+ struct port *out_port;
+ int vlan;
+ int out_port_idx;
+
+ /* Check whether we should drop packets in this flow. */
+ if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
+ out_port = NULL;
+ goto done;
+ }
+
/* Learn source MAC (but don't try to learn from revalidation). */
if (packet) {
update_learning_table(br, flow, vlan, in_port);
}
/* Determine output port. */
- out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
- tags);
+ out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags,
+ NULL);
if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
out_port = br->ports[out_port_idx];
} else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
* on a bond and blackhole packets before the learning table is
* updated to reflect the correct port. */
return false;
+ } else {
+ out_port = FLOOD_PORT;
}
/* Don't send packets out their input ports. */
}
done:
- compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
- nf_output_iface);
+ if (in_port) {
+ compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
+ nf_output_iface);
+ }
return true;
}
struct bridge *br = br_;
COVERAGE_INC(bridge_process_flow);
+
return process_flow(br, flow, packet, actions, tags, nf_output_iface);
}
static void
-bridge_account_flow_ofhook_cb(const flow_t *flow,
+bridge_account_flow_ofhook_cb(const flow_t *flow, tag_type tags,
const union odp_action *actions,
size_t n_actions, unsigned long long int n_bytes,
void *br_)
{
struct bridge *br = br_;
- struct port *in_port;
const union odp_action *a;
+ struct port *in_port;
+ tag_type dummy = 0;
+ int vlan;
- /* Feed information from the active flows back into the learning table
- * to ensure that table is always in sync with what is actually flowing
- * through the datapath. */
- in_port = port_from_dp_ifidx(br, flow->in_port);
- if (in_port) {
- int vlan = flow_get_vlan(br, flow, in_port, false);
- if (vlan >= 0) {
- update_learning_table(br, flow, vlan, in_port);
- }
+ /* Feed information from the active flows back into the learning table to
+ * ensure that table is always in sync with what is actually flowing
+ * through the datapath.
+ *
+ * We test that 'tags' is nonzero to ensure that only flows that include an
+ * OFPP_NORMAL action are used for learning. This works because
+ * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
+ if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
+ update_learning_table(br, flow, vlan, in_port);
}
+ /* Account for bond slave utilization. */
if (!br->has_bonded_ports) {
return;
}
-
for (a = actions; a < &actions[n_actions]; a++) {
if (a->type == ODPAT_OUTPUT) {
struct port *out_port = port_from_dp_ifidx(br, a->output.port);
bridge_account_checkpoint_ofhook_cb(void *br_)
{
struct bridge *br = br_;
+ long long int now;
size_t i;
if (!br->has_bonded_ports) {
return;
}
- /* The current ofproto implementation calls this callback at least once a
- * second, so this timer implementation is sufficient. */
- if (time_msec() < br->bond_next_rebalance) {
- return;
- }
- br->bond_next_rebalance = time_msec() + 10000;
-
+ now = time_msec();
for (i = 0; i < br->n_ports; i++) {
struct port *port = br->ports[i];
- if (port->n_ifaces > 1) {
+ if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) {
+ port->bond_next_rebalance = now + port->bond_rebalance_interval;
bond_rebalance_port(port);
}
}
n_packets++;
compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
e->mac);
- flow_extract(&packet, ODPP_NONE, &flow);
+ flow_extract(&packet, 0, ODPP_NONE, &flow);
retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
&packet);
if (retval) {
static void
bond_unixctl_list(struct unixctl_conn *conn,
- const char *args UNUSED, void *aux UNUSED)
+ const char *args OVS_UNUSED, void *aux OVS_UNUSED)
{
struct ds ds = DS_EMPTY_INITIALIZER;
const struct bridge *br;
static void
bond_unixctl_show(struct unixctl_conn *conn,
- const char *args, void *aux UNUSED)
+ const char *args, void *aux OVS_UNUSED)
{
struct ds ds = DS_EMPTY_INITIALIZER;
const struct port *port;
ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
ds_put_format(&ds, "next rebalance: %lld ms\n",
- port->bridge->bond_next_rebalance - time_msec());
+ port->bond_next_rebalance - time_msec());
for (j = 0; j < port->n_ifaces; j++) {
const struct iface *iface = port->ifaces[j];
struct bond_entry *be;
static void
bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
- void *aux UNUSED)
+ void *aux OVS_UNUSED)
{
char *args = (char *) args_;
char *save_ptr = NULL;
static void
bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
- void *aux UNUSED)
+ void *aux OVS_UNUSED)
{
char *args = (char *) args_;
char *save_ptr = NULL;
static void
bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
- void *aux UNUSED)
+ void *aux OVS_UNUSED)
{
enable_slave(conn, args, true);
}
static void
bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
- void *aux UNUSED)
+ void *aux OVS_UNUSED)
{
enable_slave(conn, args, false);
}
static void
bond_unixctl_hash(struct unixctl_conn *conn, const char *args,
- void *aux UNUSED)
+ void *aux OVS_UNUSED)
{
uint8_t mac[ETH_ADDR_LEN];
uint8_t hash;
sizeof *br->ports);
}
br->ports[br->n_ports++] = port;
+ shash_add_assert(&br->port_by_name, port->name, port);
VLOG_INFO("created port %s on bridge %s", port->name, br->name);
bridge_flush(br);
return port;
}
+static const char *
+get_port_other_config(const struct ovsrec_port *port, const char *key,
+ const char *default_value)
+{
+ const char *value;
+
+ value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
+ key);
+ return value ? value : default_value;
+}
+
static void
-port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
+port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
{
- struct shash old_ifaces, new_ifaces;
- struct shash_node *node;
- unsigned long *trunks;
- int vlan;
+ struct shash new_ifaces;
size_t i;
- port->cfg = cfg;
-
- /* Collect old and new interfaces. */
- shash_init(&old_ifaces);
+ /* Collect list of new interfaces. */
shash_init(&new_ifaces);
- for (i = 0; i < port->n_ifaces; i++) {
- shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]);
- }
for (i = 0; i < cfg->n_interfaces; i++) {
const char *name = cfg->interfaces[i]->name;
- if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) {
- VLOG_WARN("port %s: %s specified twice as port interface",
- port->name, name);
+ shash_add_once(&new_ifaces, name, NULL);
+ }
+
+ /* Get rid of deleted interfaces. */
+ for (i = 0; i < port->n_ifaces; ) {
+ if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) {
+ iface_destroy(port->ifaces[i]);
+ } else {
+ i++;
}
}
+
+ shash_destroy(&new_ifaces);
+}
+
+static void
+port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
+{
+ struct shash new_ifaces;
+ long long int next_rebalance;
+ unsigned long *trunks;
+ int vlan;
+ size_t i;
+
+ port->cfg = cfg;
+
+ /* Update settings. */
port->updelay = cfg->bond_updelay;
if (port->updelay < 0) {
port->updelay = 0;
}
- port->updelay = cfg->bond_downdelay;
+ port->downdelay = cfg->bond_downdelay;
if (port->downdelay < 0) {
port->downdelay = 0;
}
-
- /* Get rid of deleted interfaces and add new interfaces. */
- SHASH_FOR_EACH (node, &old_ifaces) {
- if (!shash_find(&new_ifaces, node->name)) {
- iface_destroy(node->data);
- }
+ port->bond_rebalance_interval = atoi(
+ get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
+ if (port->bond_rebalance_interval < 1000) {
+ port->bond_rebalance_interval = 1000;
+ }
+ next_rebalance = time_msec() + port->bond_rebalance_interval;
+ if (port->bond_next_rebalance > next_rebalance) {
+ port->bond_next_rebalance = next_rebalance;
}
- SHASH_FOR_EACH (node, &new_ifaces) {
- const struct ovsrec_interface *if_cfg = node->data;
+
+ /* Add new interfaces and update 'cfg' member of existing ones. */
+ shash_init(&new_ifaces);
+ for (i = 0; i < cfg->n_interfaces; i++) {
+ const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
struct iface *iface;
- iface = shash_find_data(&old_ifaces, if_cfg->name);
- if (!iface) {
- iface = iface_create(port, if_cfg);
- } else {
+ if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) {
+ VLOG_WARN("port %s: %s specified twice as port interface",
+ port->name, if_cfg->name);
+ continue;
+ }
+
+ iface = iface_lookup(port->bridge, if_cfg->name);
+ if (iface) {
+ if (iface->port != port) {
+ VLOG_ERR("bridge %s: %s interface is on multiple ports, "
+ "removing from %s",
+ port->bridge->name, if_cfg->name, iface->port->name);
+ continue;
+ }
iface->cfg = if_cfg;
+ } else {
+ iface_create(port, if_cfg);
}
}
+ shash_destroy(&new_ifaces);
/* Get VLAN tag. */
vlan = -1;
/* Get trunked VLANs. */
trunks = NULL;
- if (vlan < 0) {
+ if (vlan < 0 && cfg->n_trunks) {
size_t n_errors;
size_t i;
port->name, cfg->n_trunks);
}
if (n_errors == cfg->n_trunks) {
- if (n_errors) {
- VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
- port->name);
- }
- bitmap_set_multiple(trunks, 0, 4096, 1);
- }
- } else {
- if (cfg->n_trunks) {
- VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
+ VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
port->name);
+ bitmap_free(trunks);
+ trunks = NULL;
}
+ } else if (vlan >= 0 && cfg->n_trunks) {
+ VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
+ port->name);
}
if (trunks == NULL
? port->trunks != NULL
}
bitmap_free(port->trunks);
port->trunks = trunks;
-
- shash_destroy(&old_ifaces);
- shash_destroy(&new_ifaces);
}
static void
iface_destroy(port->ifaces[port->n_ifaces - 1]);
}
+ shash_find_and_delete_assert(&br->port_by_name, port->name);
+
del = br->ports[port->port_idx] = br->ports[--br->n_ports];
del->port_idx = port->port_idx;
static struct port *
port_lookup(const struct bridge *br, const char *name)
{
- size_t i;
-
- for (i = 0; i < br->n_ports; i++) {
- struct port *port = br->ports[i];
- if (!strcmp(port->name, name)) {
- return port;
- }
- }
- return NULL;
+ return shash_find_data(&br->port_by_name, name);
}
static struct iface *
port_lookup_iface(const struct port *port, const char *name)
{
- size_t j;
-
- for (j = 0; j < port->n_ifaces; j++) {
- struct iface *iface = port->ifaces[j];
- if (!strcmp(iface->name, name)) {
- return iface;
- }
- }
- return NULL;
+ struct iface *iface = iface_lookup(port->bridge, name);
+ return iface && iface->port == port ? iface : NULL;
}
static void
}
port->no_ifaces_tag = tag_create_random();
bond_choose_active_iface(port);
+ port->bond_next_rebalance
+ = time_msec() + port->bond_rebalance_interval;
+
+ if (port->cfg->bond_fake_iface) {
+ port->bond_next_fake_iface_update = time_msec();
+ }
}
port->bond_compat_is_stale = true;
port->bond_fake_iface = port->cfg->bond_fake_iface;
static struct iface *
iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
{
+ struct bridge *br = port->bridge;
struct iface *iface;
char *name = if_cfg->name;
int error;
iface->netdev = NULL;
iface->cfg = if_cfg;
+ shash_add_assert(&br->iface_by_name, iface->name, iface);
+
+ /* Attempt to create the network interface in case it doesn't exist yet. */
+ if (!iface_is_internal(br, iface->name)) {
+ error = set_up_iface(if_cfg, iface, true);
+ if (error) {
+ VLOG_WARN("could not create iface %s: %s", iface->name,
+ strerror(error));
+
+ shash_find_and_delete_assert(&br->iface_by_name, iface->name);
+ free(iface->name);
+ free(iface);
+ return NULL;
+ }
+ }
+
if (port->n_ifaces >= port->allocated_ifaces) {
port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
sizeof *port->ifaces);
}
port->ifaces[port->n_ifaces++] = iface;
if (port->n_ifaces > 1) {
- port->bridge->has_bonded_ports = true;
- }
-
- /* Attempt to create the network interface in case it
- * doesn't exist yet. */
- if (!iface_is_internal(port->bridge, iface->name)) {
- error = set_up_iface(if_cfg, iface, true);
- if (error) {
- VLOG_WARN("could not create iface %s: %s", iface->name,
- strerror(error));
- }
+ br->has_bonded_ports = true;
}
VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
- bridge_flush(port->bridge);
+ bridge_flush(br);
return iface;
}
bool del_active = port->active_iface == iface->port_ifidx;
struct iface *del;
+ shash_find_and_delete_assert(&br->iface_by_name, iface->name);
+
if (iface->dp_ifidx >= 0) {
port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
}
static struct iface *
iface_lookup(const struct bridge *br, const char *name)
{
- size_t i, j;
-
- for (i = 0; i < br->n_ports; i++) {
- struct port *port = br->ports[i];
- for (j = 0; j < port->n_ifaces; j++) {
- struct iface *iface = port->ifaces[j];
- if (!strcmp(iface->name, name)) {
- return iface;
- }
- }
- }
- return NULL;
+ return shash_find_data(&br->iface_by_name, name);
}
static struct iface *
static bool
iface_is_internal(const struct bridge *br, const char *if_name)
{
- /* XXX wastes time */
struct iface *iface;
struct port *port;
}
}
}
+
+static void
+shash_from_ovs_idl_map(char **keys, char **values, size_t n,
+ struct shash *shash)
+{
+ size_t i;
+
+ shash_init(shash);
+ for (i = 0; i < n; i++) {
+ shash_add(shash, keys[i], values[i]);
+ }
+}
+
+struct iface_delete_queues_cbdata {
+ struct netdev *netdev;
+ const struct ovsdb_datum *queues;
+};
+
+static bool
+queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
+{
+ union ovsdb_atom atom;
+
+ atom.integer = target;
+ return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
+}
+
+static void
+iface_delete_queues(unsigned int queue_id,
+ const struct shash *details OVS_UNUSED, void *cbdata_)
+{
+ struct iface_delete_queues_cbdata *cbdata = cbdata_;
+
+ if (!queue_ids_include(cbdata->queues, queue_id)) {
+ netdev_delete_queue(cbdata->netdev, queue_id);
+ }
+}
+
+static void
+iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
+{
+ if (!qos || qos->type[0] == '\0') {
+ netdev_set_qos(iface->netdev, NULL, NULL);
+ } else {
+ struct iface_delete_queues_cbdata cbdata;
+ struct shash details;
+ size_t i;
+
+ /* Configure top-level Qos for 'iface'. */
+ shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
+ qos->n_other_config, &details);
+ netdev_set_qos(iface->netdev, qos->type, &details);
+ shash_destroy(&details);
+
+ /* Deconfigure queues that were deleted. */
+ cbdata.netdev = iface->netdev;
+ cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
+ OVSDB_TYPE_UUID);
+ netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
+
+ /* Configure queues for 'iface'. */
+ for (i = 0; i < qos->n_queues; i++) {
+ const struct ovsrec_queue *queue = qos->value_queues[i];
+ unsigned int queue_id = qos->key_queues[i];
+
+ shash_from_ovs_idl_map(queue->key_other_config,
+ queue->value_other_config,
+ queue->n_other_config, &details);
+ netdev_set_queue(iface->netdev, queue_id, &details);
+ shash_destroy(&details);
+ }
+ }
+}
\f
/* Port mirroring. */
+static struct mirror *
+mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
+{
+ int i;
+
+ for (i = 0; i < MAX_MIRRORS; i++) {
+ struct mirror *m = br->mirrors[i];
+ if (m && uuid_equals(uuid, &m->uuid)) {
+ return m;
+ }
+ }
+ return NULL;
+}
+
static void
mirror_reconfigure(struct bridge *br)
{
- struct shash old_mirrors, new_mirrors;
- struct shash_node *node;
unsigned long *rspan_vlans;
int i;
- /* Collect old mirrors. */
- shash_init(&old_mirrors);
+ /* Get rid of deleted mirrors. */
for (i = 0; i < MAX_MIRRORS; i++) {
- if (br->mirrors[i]) {
- shash_add(&old_mirrors, br->mirrors[i]->name, br->mirrors[i]);
+ struct mirror *m = br->mirrors[i];
+ if (m) {
+ const struct ovsdb_datum *mc;
+ union ovsdb_atom atom;
+
+ mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
+ atom.uuid = br->mirrors[i]->uuid;
+ if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
+ mirror_destroy(m);
+ }
}
}
- /* Collect new mirrors. */
- shash_init(&new_mirrors);
+ /* Add new mirrors and reconfigure existing ones. */
for (i = 0; i < br->cfg->n_mirrors; i++) {
struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
- if (!shash_add_once(&new_mirrors, cfg->name, cfg)) {
- VLOG_WARN("bridge %s: %s specified twice as mirror",
- br->name, cfg->name);
- }
- }
-
- /* Get rid of deleted mirrors and add new mirrors. */
- SHASH_FOR_EACH (node, &old_mirrors) {
- if (!shash_find(&new_mirrors, node->name)) {
- mirror_destroy(node->data);
- }
- }
- SHASH_FOR_EACH (node, &new_mirrors) {
- struct mirror *mirror = shash_find_data(&old_mirrors, node->name);
- if (!mirror) {
- mirror = mirror_create(br, node->name);
- if (!mirror) {
- break;
- }
+ struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
+ if (m) {
+ mirror_reconfigure_one(m, cfg);
+ } else {
+ mirror_create(br, cfg);
}
- mirror_reconfigure_one(mirror, node->data);
}
- shash_destroy(&old_mirrors);
- shash_destroy(&new_mirrors);
/* Update port reserved status. */
for (i = 0; i < br->n_ports; i++) {
}
}
-static struct mirror *
-mirror_create(struct bridge *br, const char *name)
+static void
+mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
{
struct mirror *m;
size_t i;
for (i = 0; ; i++) {
if (i >= MAX_MIRRORS) {
VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
- "cannot create %s", br->name, MAX_MIRRORS, name);
- return NULL;
+ "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
+ return;
}
if (!br->mirrors[i]) {
break;
}
}
- VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
+ VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
bridge_flush(br);
br->mirrors[i] = m = xzalloc(sizeof *m);
m->bridge = br;
m->idx = i;
- m->name = xstrdup(name);
+ m->name = xstrdup(cfg->name);
shash_init(&m->src_ports);
shash_init(&m->dst_ports);
m->vlans = NULL;
m->out_vlan = -1;
m->out_port = NULL;
- return m;
+ mirror_reconfigure_one(m, cfg);
}
static void
free(m->vlans);
m->bridge->mirrors[m->idx] = NULL;
+ free(m->name);
free(m);
bridge_flush(br);
size_t n_vlans;
size_t i;
- *vlans = xmalloc(sizeof *vlans * cfg->n_select_vlan);
+ *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
n_vlans = 0;
for (i = 0; i < cfg->n_select_vlan; i++) {
int64_t vlan = cfg->select_vlan[i];
size_t n_vlans;
int *vlans;
size_t i;
- bool mirror_all_ports;
- bool any_ports_specified;
- bool any_vlans_specified;
+
+ /* Set name. */
+ if (strcmp(cfg->name, m->name)) {
+ free(m->name);
+ m->name = xstrdup(cfg->name);
+ }
/* Get output port. */
if (cfg->output_port) {
return;
}
- /* Get all the ports, and drop duplicates and ports that don't exist. */
shash_init(&src_ports);
shash_init(&dst_ports);
- mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
- &src_ports);
- mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
- &dst_ports);
- any_ports_specified = cfg->n_select_dst_port || cfg->n_select_dst_port;
- if (any_ports_specified
- && shash_is_empty(&src_ports) && shash_is_empty(&dst_ports)) {
- VLOG_ERR("bridge %s: disabling mirror %s since none of the specified "
- "selection ports exists", m->bridge->name, m->name);
- mirror_destroy(m);
- goto exit;
- }
+ if (cfg->select_all) {
+ for (i = 0; i < m->bridge->n_ports; i++) {
+ const char *name = m->bridge->ports[i]->name;
+ shash_add_once(&src_ports, name, NULL);
+ shash_add_once(&dst_ports, name, NULL);
+ }
+ vlans = NULL;
+ n_vlans = 0;
+ } else {
+ /* Get ports, and drop duplicates and ports that don't exist. */
+ mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
+ &src_ports);
+ mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
+ &dst_ports);
- /* Get all the vlans, and drop duplicate and invalid vlans. */
- n_vlans = mirror_collect_vlans(m, cfg, &vlans);
- any_vlans_specified = cfg->n_select_vlan > 0;
- if (any_vlans_specified && !n_vlans) {
- VLOG_ERR("bridge %s: disabling mirror %s since none of the specified "
- "VLANs exists", m->bridge->name, m->name);
- mirror_destroy(m);
- goto exit;
+ /* Get all the vlans, and drop duplicate and invalid vlans. */
+ n_vlans = mirror_collect_vlans(m, cfg, &vlans);
}
/* Update mirror data. */
m->out_port = out_port;
m->out_vlan = out_vlan;
- /* If no selection criteria have been given, mirror for all ports. */
- mirror_all_ports = !any_ports_specified && !any_vlans_specified;
-
/* Update ports. */
mirror_bit = MIRROR_MASK_C(1) << m->idx;
for (i = 0; i < m->bridge->n_ports; i++) {
struct port *port = m->bridge->ports[i];
- if (mirror_all_ports
- || shash_find(&m->src_ports, port->name)
+ if (shash_find(&m->src_ports, port->name)
|| (m->n_vlans
&& (!port->vlan
? port_trunks_any_mirrored_vlan(m, port)
port->src_mirrors &= ~mirror_bit;
}
- if (mirror_all_ports || shash_find(&m->dst_ports, port->name)) {
+ if (shash_find(&m->dst_ports, port->name)) {
port->dst_mirrors |= mirror_bit;
} else {
port->dst_mirrors &= ~mirror_bit;
}
/* Clean up. */
-exit:
shash_destroy(&src_ports);
shash_destroy(&dst_ports);
}
git://git.onelab.eu / sliver-openvswitch.git / blobdiff