X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;ds=sidebyside;f=vswitchd%2Fbridge.c;h=41fcba52287ce861a118502dc3df04f44cb67033;hb=85da620e9788b473797b95212b916327974e6942;hp=5210da161ac6d136f419f6a4779d33f7a5285907;hpb=43776b8fa5df28adfe08e7bdc8bffdac252b1e01;p=sliver-openvswitch.git diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index 5210da161..41fcba522 100644 --- a/vswitchd/bridge.c +++ b/vswitchd/bridge.c @@ -89,6 +89,7 @@ struct iface { int dp_ifidx; /* Index within kernel datapath. */ struct netdev *netdev; /* Network device. */ bool enabled; /* May be chosen for flows? */ + const char *type; /* Usually same as cfg->type. */ const struct ovsrec_interface *cfg; }; @@ -146,6 +147,7 @@ struct port { 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. */ + struct netdev_monitor *monitor; /* Tracks carrier up/down status. */ /* Port mirroring info. */ mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ @@ -252,8 +254,8 @@ static void iface_destroy(struct iface *); static struct iface *iface_lookup(const struct bridge *, const char *name); static struct iface *iface_from_dp_ifidx(const struct bridge *, 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_set_ofport(const struct ovsrec_interface *, int64_t ofport); static void iface_update_qos(struct iface *, const struct ovsrec_qos *); static void shash_from_ovs_idl_map(char **keys, char **values, size_t n, @@ -389,6 +391,15 @@ iface_get_options(const struct ovsrec_interface *if_cfg, struct shash *options) } } +/* Returns the type of network device that 'iface' should have. (This is + * ordinarily the same type as the interface, but the network devices for + * "internal" ports have type "system".) */ +static const char * +iface_get_netdev_type(const struct iface *iface) +{ + return !strcmp(iface->type, "internal") ? "system" : iface->type; +} + /* Attempt to create the network device for 'iface' through the netdev * library. */ static int @@ -400,12 +411,7 @@ create_iface_netdev(struct iface *iface) memset(&netdev_options, 0, sizeof netdev_options); netdev_options.name = iface->cfg->name; - 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.type = iface_get_netdev_type(iface); netdev_options.args = &options; netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; @@ -414,7 +420,7 @@ create_iface_netdev(struct iface *iface) error = netdev_open(&netdev_options, &iface->netdev); if (iface->netdev) { - netdev_get_carrier(iface->netdev, &iface->enabled); + iface->enabled = netdev_get_carrier(iface->netdev); } shash_destroy(&options); @@ -431,9 +437,7 @@ reconfigure_iface_netdev(struct iface *iface) /* Skip reconfiguration if the device has the wrong type. This shouldn't * happen, but... */ - iface_type = (!iface->cfg->type[0] ? NULL - : !strcmp(iface->cfg->type, "internal") ? "system" - : iface->cfg->type); + iface_type = iface_get_netdev_type(iface); netdev_type = netdev_get_type(iface->netdev); if (iface_type && strcmp(netdev_type, iface_type)) { VLOG_WARN("%s: attempting change device type from %s to %s", @@ -449,38 +453,27 @@ reconfigure_iface_netdev(struct iface *iface) return error; } +/* Callback for iterate_and_prune_ifaces(). */ static bool -check_iface_netdev(struct bridge *br OVS_UNUSED, struct iface *iface, - void *aux OVS_UNUSED) +check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED) { if (!iface->netdev) { - int error = create_iface_netdev(iface); - if (error) { - VLOG_WARN("could not open netdev on %s, dropping: %s", iface->name, - strerror(error)); - return false; - } + /* We already reported a related error, don't bother duplicating it. */ + return false; } - return true; -} - -static bool -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", - dpif_name(br->dpif), - iface->name, iface->dp_ifidx); - return true; - } else { + if (iface->dp_ifidx < 0) { VLOG_ERR("%s interface not in %s, dropping", iface->name, dpif_name(br->dpif)); return false; } + + VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif), + iface->name, iface->dp_ifidx); + return true; } +/* Callback for iterate_and_prune_ifaces(). */ static bool set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface, void *aux OVS_UNUSED) @@ -492,8 +485,7 @@ set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface, /* Set MAC address of internal interfaces other than the local * interface. */ - if (iface->dp_ifidx != ODPP_LOCAL - && iface_is_internal(br, iface->name)) { + if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) { iface_set_mac(iface); } @@ -518,6 +510,7 @@ iterate_and_prune_ifaces(struct bridge *br, if (cb(br, iface, aux)) { j++; } else { + iface_set_ofport(iface->cfg, -1); iface_destroy(iface); } } @@ -666,30 +659,58 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) shash_init(&cur_ifaces); for (i = 0; i < n_dpif_ports; i++) { const char *name = dpif_ports[i].devname; - shash_add_once(&cur_ifaces, name, NULL); + shash_add_once(&cur_ifaces, name, &dpif_ports[i]); } - free(dpif_ports); /* Get the set of interfaces we want on this datapath. */ bridge_get_all_ifaces(br, &want_ifaces); + hmap_clear(&br->ifaces); SHASH_FOR_EACH (node, &want_ifaces) { const char *if_name = node->name; struct iface *iface = node->data; + bool internal = !iface || !strcmp(iface->type, "internal"); + struct odp_port *dpif_port = shash_find_data(&cur_ifaces, if_name); + int error; + + /* If we have a port or a netdev already, and it's not the type we + * want, then delete the port (if any) and close the netdev (if + * any). */ + if (internal + ? dpif_port && !(dpif_port->flags & ODP_PORT_INTERNAL) + : (iface->netdev + && strcmp(iface->type, netdev_get_type(iface->netdev)))) + { + if (dpif_port) { + error = ofproto_port_del(br->ofproto, dpif_port->port); + if (error) { + continue; + } + dpif_port = NULL; + } + if (iface) { + netdev_close(iface->netdev); + iface->netdev = NULL; + } + } - if (shash_find(&cur_ifaces, if_name)) { - /* Already exists on the datapath. If we have it open, - * reconfigure it; otherwise we'll open it later. */ - if (iface && iface->netdev) { + /* If it's not an internal port, open (possibly create) the + * netdev. */ + if (!internal) { + if (!iface->netdev) { + error = create_iface_netdev(iface); + if (error) { + VLOG_WARN("could not create iface %s: %s", iface->name, + strerror(error)); + continue; + } + } else { reconfigure_iface_netdev(iface); } - } else { - /* Need to add to datapath. */ - bool internal; - int error; + } - /* Add to datapath. */ - internal = iface_is_internal(br, if_name); + /* If it's not part of the datapath, add it. */ + if (!dpif_port) { error = dpif_port_add(br->dpif, if_name, internal ? ODP_PORT_INTERNAL : 0, NULL); if (error == EFBIG) { @@ -699,9 +720,25 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) } else if (error) { VLOG_ERR("failed to add %s interface to %s: %s", if_name, dpif_name(br->dpif), strerror(error)); + continue; } } + + /* If it's an internal port, open the netdev. */ + if (internal) { + if (iface && !iface->netdev) { + error = create_iface_netdev(iface); + if (error) { + VLOG_WARN("could not create iface %s: %s", iface->name, + strerror(error)); + continue; + } + } + } else { + assert(iface->netdev != NULL); + } } + free(dpif_ports); shash_destroy(&cur_ifaces); shash_destroy(&want_ifaces); } @@ -715,8 +752,7 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) bridge_fetch_dp_ifaces(br); - iterate_and_prune_ifaces(br, check_iface_netdev, NULL); - iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL); + iterate_and_prune_ifaces(br, check_iface, NULL); /* Pick local port hardware address, datapath ID. */ bridge_pick_local_hw_addr(br, ea, &hw_addr_iface); @@ -1668,12 +1704,25 @@ bridge_reconfigure_remotes(struct bridge *br, struct ovsrec_controller **controllers; size_t n_controllers; bool had_primary; + const char *disable_ib_str; + bool disable_in_band = false; struct ofproto_controller *ocs; size_t n_ocs; size_t i; - ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers); + + /* Check if we should disable in-band control on this bridge. */ + disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band"); + if (disable_ib_str && !strcmp(disable_ib_str, "true")) { + disable_in_band = true; + } + + if (disable_in_band) { + ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0); + } else { + 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); @@ -1698,7 +1747,11 @@ bridge_reconfigure_remotes(struct bridge *br, } bridge_configure_local_iface_netdev(br, c); - bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs++]); + bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]); + if (disable_in_band) { + ocs[n_ocs].band = OFPROTO_OUT_OF_BAND; + } + n_ocs++; } ofproto_set_controllers(br->ofproto, ocs, n_ocs); @@ -1717,7 +1770,7 @@ bridge_reconfigure_remotes(struct bridge *br, if (!n_controllers && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) { union ofp_action action; - flow_t flow; + struct flow flow; memset(&action, 0, sizeof action); action.type = htons(OFPAT_OUTPUT); @@ -1788,12 +1841,10 @@ bridge_fetch_dp_ifaces(struct bridge *br) hash_int(iface->dp_ifidx, 0)); } - if (iface->cfg) { - int64_t ofport = (iface->dp_ifidx >= 0 - ? odp_port_to_ofp_port(iface->dp_ifidx) - : -1); - ovsrec_interface_set_ofport(iface->cfg, &ofport, 1); - } + iface_set_ofport(iface->cfg, + (iface->dp_ifidx >= 0 + ? odp_port_to_ofp_port(iface->dp_ifidx) + : -1)); } } free(dpif_ports); @@ -2021,6 +2072,22 @@ bond_run(struct bridge *br) struct port *port = br->ports[i]; if (port->n_ifaces >= 2) { + char *devname; + + /* Track carrier going up and down on interfaces. */ + while (!netdev_monitor_poll(port->monitor, &devname)) { + struct iface *iface; + + iface = port_lookup_iface(port, devname); + if (iface) { + bool carrier = netdev_get_carrier(iface->netdev); + + bond_link_status_update(iface, carrier); + port_update_bond_compat(port); + } + free(devname); + } + for (j = 0; j < port->n_ifaces; j++) { struct iface *iface = port->ifaces[j]; if (time_msec() >= iface->delay_expires) { @@ -2052,6 +2119,7 @@ bond_wait(struct bridge *br) if (port->n_ifaces < 2) { continue; } + netdev_monitor_poll_wait(port->monitor); for (j = 0; j < port->n_ifaces; j++) { struct iface *iface = port->ifaces[j]; if (iface->delay_expires != LLONG_MAX) { @@ -2065,7 +2133,7 @@ bond_wait(struct bridge *br) } static bool -set_dst(struct dst *p, const flow_t *flow, +set_dst(struct dst *p, const struct flow *flow, const struct port *in_port, const struct port *out_port, tag_type *tags) { @@ -2153,8 +2221,22 @@ port_includes_vlan(const struct port *port, uint16_t vlan) return vlan == port->vlan || port_trunks_vlan(port, vlan); } +static bool +port_is_floodable(const struct port *port) +{ + int i; + + for (i = 0; i < port->n_ifaces; i++) { + if (!ofproto_port_is_floodable(port->bridge->ofproto, + port->ifaces[i]->dp_ifidx)) { + return false; + } + } + return true; +} + static size_t -compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan, +compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan, const struct port *in_port, const struct port *out_port, struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface) { @@ -2167,7 +2249,9 @@ compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan, /* XXX even better, define each VLAN as a datapath port group */ for (i = 0; i < br->n_ports; i++) { struct port *port = br->ports[i]; - if (port != in_port && port_includes_vlan(port, vlan) + if (port != in_port + && port_is_floodable(port) + && port_includes_vlan(port, vlan) && !port->is_mirror_output_port && set_dst(dst, flow, in_port, port, tags)) { mirrors |= port->dst_mirrors; @@ -2242,7 +2326,7 @@ print_dsts(const struct dst *dsts, size_t n) } static void -compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan, +compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan, const struct port *in_port, const struct port *out_port, tag_type *tags, struct odp_actions *actions, uint16_t *nf_output_iface) @@ -2262,8 +2346,9 @@ compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan, if (p->vlan == OFP_VLAN_NONE) { odp_actions_add(actions, ODPAT_STRIP_VLAN); } else { - a = odp_actions_add(actions, ODPAT_SET_VLAN_VID); - a->vlan_vid.vlan_vid = htons(p->vlan); + a = odp_actions_add(actions, ODPAT_SET_DL_TCI); + a->dl_tci.tci = htons(p->vlan & VLAN_VID_MASK); + a->dl_tci.tci |= htons(flow->dl_vlan_pcp << VLAN_PCP_SHIFT); } cur_vlan = p->vlan; } @@ -2276,7 +2361,7 @@ compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan, * 802.1Q header and implicitly tagged ports. A value of 0 indicates that * the packet is untagged and -1 indicates it has an invalid header and * should be dropped. */ -static int flow_get_vlan(struct bridge *br, const flow_t *flow, +static int flow_get_vlan(struct bridge *br, const struct flow *flow, struct port *in_port, bool have_packet) { /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet @@ -2322,7 +2407,7 @@ static int flow_get_vlan(struct bridge *br, const flow_t *flow, * 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) +is_gratuitous_arp(const struct flow *flow) { return (flow->dl_type == htons(ETH_TYPE_ARP) && eth_addr_is_broadcast(flow->dl_dst) @@ -2332,7 +2417,7 @@ is_gratuitous_arp(const flow_t *flow) } static void -update_learning_table(struct bridge *br, const flow_t *flow, int vlan, +update_learning_table(struct bridge *br, const struct flow *flow, int vlan, struct port *in_port) { enum grat_arp_lock_type lock_type; @@ -2380,7 +2465,7 @@ update_learning_table(struct bridge *br, const flow_t *flow, int vlan, * so in one special case. */ static bool -is_admissible(struct bridge *br, const flow_t *flow, bool have_packet, +is_admissible(struct bridge *br, const struct flow *flow, bool have_packet, tag_type *tags, int *vlanp, struct port **in_portp) { struct iface *in_iface; @@ -2469,7 +2554,7 @@ is_admissible(struct bridge *br, const flow_t *flow, bool have_packet, * 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, +process_flow(struct bridge *br, const struct flow *flow, const struct ofpbuf *packet, struct odp_actions *actions, tag_type *tags, uint16_t *nf_output_iface) { @@ -2519,45 +2604,8 @@ done: return true; } -/* Careful: 'opp' is in host byte order and opp->port_no is an OFP port - * number. */ -static void -bridge_port_changed_ofhook_cb(enum ofp_port_reason reason, - const struct ofp_phy_port *opp, - void *br_) -{ - struct bridge *br = br_; - struct iface *iface; - struct port *port; - - iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no)); - if (!iface) { - return; - } - port = iface->port; - - if (reason == OFPPR_DELETE) { - VLOG_WARN("bridge %s: interface %s deleted unexpectedly", - br->name, iface->name); - iface_destroy(iface); - if (!port->n_ifaces) { - VLOG_WARN("bridge %s: port %s has no interfaces, dropping", - br->name, port->name); - port_destroy(port); - } - - bridge_flush(br); - } else { - if (port->n_ifaces > 1) { - bool up = !(opp->state & OFPPS_LINK_DOWN); - bond_link_status_update(iface, up); - port_update_bond_compat(port); - } - } -} - static bool -bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, +bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, struct odp_actions *actions, tag_type *tags, uint16_t *nf_output_iface, void *br_) { @@ -2569,7 +2617,7 @@ bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, } static void -bridge_account_flow_ofhook_cb(const flow_t *flow, tag_type tags, +bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags, const union odp_action *actions, size_t n_actions, unsigned long long int n_bytes, void *br_) @@ -2629,7 +2677,6 @@ bridge_account_checkpoint_ofhook_cb(void *br_) } static struct ofhooks bridge_ofhooks = { - bridge_port_changed_ofhook_cb, bridge_normal_ofhook_cb, bridge_account_flow_ofhook_cb, bridge_account_checkpoint_ofhook_cb, @@ -2951,7 +2998,7 @@ bond_send_learning_packets(struct port *port) union ofp_action actions[2], *a; uint16_t dp_ifidx; tag_type tags = 0; - flow_t flow; + struct flow flow; int retval; if (e->port == port->port_idx @@ -3406,6 +3453,7 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) 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); + iface_set_ofport(if_cfg, -1); continue; } @@ -3419,8 +3467,15 @@ port_reconfigure(struct port *port, const struct ovsrec_port *cfg) } iface->cfg = if_cfg; } else { - iface_create(port, if_cfg); + iface = iface_create(port, if_cfg); } + + /* Determine interface type. The local port always has type + * "internal". Other ports take their type from the database and + * default to "system" if none is specified. */ + iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal" + : if_cfg->type[0] ? if_cfg->type + : "system"); } shash_destroy(&new_ifaces); @@ -3511,6 +3566,9 @@ port_destroy(struct port *port) del = br->ports[port->port_idx] = br->ports[--br->n_ports]; del->port_idx = port->port_idx; + VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name); + + netdev_monitor_destroy(port->monitor); free(port->ifaces); bitmap_free(port->trunks); free(port->name); @@ -3542,6 +3600,10 @@ port_lookup_iface(const struct port *port, const char *name) static void port_update_bonding(struct port *port) { + if (port->monitor) { + netdev_monitor_destroy(port->monitor); + port->monitor = NULL; + } if (port->n_ifaces < 2) { /* Not a bonded port. */ if (port->bond_hash) { @@ -3551,9 +3613,9 @@ port_update_bonding(struct port *port) port->bond_fake_iface = false; } } else { - if (!port->bond_hash) { - size_t i; + size_t i; + if (!port->bond_hash) { port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash); for (i = 0; i <= BOND_MASK; i++) { struct bond_entry *e = &port->bond_hash[i]; @@ -3571,6 +3633,11 @@ port_update_bonding(struct port *port) } port->bond_compat_is_stale = true; port->bond_fake_iface = port->cfg->bond_fake_iface; + + port->monitor = netdev_monitor_create(); + for (i = 0; i < port->n_ifaces; i++) { + netdev_monitor_add(port->monitor, port->ifaces[i]->netdev); + } } } @@ -3692,7 +3759,6 @@ 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 = xzalloc(sizeof *iface); iface->port = port; @@ -3706,20 +3772,6 @@ iface_create(struct port *port, const struct ovsrec_interface *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 = create_iface_netdev(iface); - 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); @@ -3789,38 +3841,6 @@ iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) return NULL; } -/* Returns true if 'iface' is the name of an "internal" interface on bridge - * 'br', that is, an interface that is entirely simulated within the datapath. - * The local port (ODPP_LOCAL) is always an internal interface. Other local - * interfaces are created by setting "iface..internal = true". - * - * In addition, we have a kluge-y feature that creates an internal port with - * the name of a bonded port if "bonding..fake-iface = true" is set. - * This feature needs to go away in the long term. Until then, this is one - * reason why this function takes a name instead of a struct iface: the fake - * interfaces created this way do not have a struct iface. */ -static bool -iface_is_internal(const struct bridge *br, const char *if_name) -{ - struct iface *iface; - struct port *port; - - if (!strcmp(if_name, br->name)) { - return true; - } - - iface = iface_lookup(br, if_name); - if (iface && !strcmp(iface->cfg->type, "internal")) { - return true; - } - - port = port_lookup(br, if_name); - if (port && port->n_ifaces > 1 && port->cfg->bond_fake_iface) { - return true; - } - return false; -} - /* Set Ethernet address of 'iface', if one is specified in the configuration * file. */ static void @@ -3845,6 +3865,15 @@ iface_set_mac(struct iface *iface) } } +/* Sets the ofport column of 'if_cfg' to 'ofport'. */ +static void +iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport) +{ + if (if_cfg) { + ovsrec_interface_set_ofport(if_cfg, &ofport, 1); + } +} + /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'. * * The value strings in '*shash' are taken directly from values[], not copied,