From fa066f015f716c74315fed1a08912d003dfe3f8c Mon Sep 17 00:00:00 2001 From: Ben Pfaff Date: Wed, 11 May 2011 12:26:06 -0700 Subject: [PATCH] bridge: Move packet processing functionality into ofproto. Until now, packet processing in ovs-vswitchd has been split between two components: ofproto, for basic OpenFlow functionality, and bridge, for OFPP_NORMAL processing. This architecture will not work as Open vSwitch starts to support a wider variety of underlying hardware, because it imposes a model in which the bridge needs to be able to look at every exact-match flow within a OpenFlow flow, which most hardware doesn't support. Therefore, this commit moves all of the packet processing code in bridge into ofproto, as preparation for generalizing further. --- PORTING | 16 + ofproto/ofproto.c | 1563 ++++++++++++++++++++++++++++----- ofproto/ofproto.h | 83 +- utilities/ovs-openflowd.c | 2 +- vswitchd/bridge.c | 1742 ++++++++----------------------------- 5 files changed, 1763 insertions(+), 1643 deletions(-) diff --git a/PORTING b/PORTING index 911761e6..6af9f51d 100644 --- a/PORTING +++ b/PORTING @@ -7,6 +7,22 @@ are most likely to be necessary in porting OVS to Unix-like platforms. (Porting OVS to other kinds of platforms is likely to be more difficult.) +Vocabulary +---------- + +For historical reasons, different words are used for essentially the +same concept in different areas of the Open vSwitch source tree. Here +is a concordance, indexed by the area of the source tree: + + datapath/ vport --- + vswitchd/ iface port + ofproto/ port bundle + lib/bond.c slave bond + lib/lacp.c slave lacp + lib/netdev.c netdev --- + database Interface Port + + Open vSwitch Architectural Overview ----------------------------------- diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index f8a67fcf..002470ec 100644 --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@ -25,6 +25,8 @@ #include #include #include "autopath.h" +#include "bitmap.h" +#include "bond.h" #include "byte-order.h" #include "cfm.h" #include "classifier.h" @@ -35,7 +37,9 @@ #include "fail-open.h" #include "hash.h" #include "hmap.h" +#include "hmapx.h" #include "in-band.h" +#include "lacp.h" #include "mac-learning.h" #include "multipath.h" #include "netdev.h" @@ -65,6 +69,7 @@ #include "unaligned.h" #include "unixctl.h" #include "vconn.h" +#include "vlan-bitmap.h" #include "vlog.h" VLOG_DEFINE_THIS_MODULE(ofproto); @@ -99,13 +104,67 @@ COVERAGE_DEFINE(ofproto_update_port); struct rule; +#define MAX_MIRRORS 32 +typedef uint32_t mirror_mask_t; +#define MIRROR_MASK_C(X) UINT32_C(X) +BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); +struct ofmirror { + struct ofproto *ofproto; /* Owning ofproto. */ + size_t idx; /* In ofproto's "mirrors" array. */ + void *aux; /* Key supplied by ofproto's client. */ + char *name; /* Identifier for log messages. */ + + /* Selection criteria. */ + struct hmapx srcs; /* Contains "struct ofbundle *"s. */ + struct hmapx dsts; /* Contains "struct ofbundle *"s. */ + unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */ + + /* Output (mutually exclusive). */ + struct ofbundle *out; /* Output port or NULL. */ + int out_vlan; /* Output VLAN or -1. */ +}; + +static void ofproto_mirror_destroy(struct ofmirror *); + +/* A group of one or more OpenFlow ports. */ +#define OFBUNDLE_FLOOD ((struct ofbundle *) 1) +struct ofbundle { + struct ofproto *ofproto; /* Owning ofproto. */ + struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ + void *aux; /* Key supplied by ofproto's client. */ + char *name; /* Identifier for log messages. */ + + /* Configuration. */ + struct list ports; /* Contains "struct ofport"s. */ + int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ + unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. + * NULL if all VLANs are trunked. */ + struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ + struct bond *bond; /* Bonding setup if more than one port, + * otherwise NULL. */ + + /* Status. */ + bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */ + + /* Port mirroring info. */ + mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ + mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ + mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */ +}; + +/* An OpenFlow port. */ struct ofport { struct ofproto *ofproto; /* Owning ofproto. */ struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */ struct netdev *netdev; struct ofp_phy_port opp; uint16_t odp_port; + + /* Bridging. */ + struct ofbundle *bundle; /* Bundle that contains this port, if any. */ + struct list bundle_node; /* In struct ofbundle's "ports" list. */ struct cfm *cfm; /* Connectivity Fault Management, if any. */ + tag_type tag; /* Tag associated with this port. */ }; static void ofport_free(struct ofport *); @@ -251,7 +310,7 @@ struct facet { * be reassessed for every packet. */ size_t actions_len; /* Number of bytes in actions[]. */ struct nlattr *actions; /* Datapath actions. */ - tag_type tags; /* Tags (set only by hooks). */ + tag_type tags; /* Tags. */ struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ }; @@ -297,9 +356,13 @@ struct ofproto { struct shash port_by_name; uint32_t max_ports; - /* Configuration. */ + /* Bridging. */ struct netflow *netflow; struct ofproto_sflow *sflow; + struct hmap bundles; /* Contains "struct ofbundle"s. */ + struct mac_learning *ml; + struct ofmirror *mirrors[MAX_MIRRORS]; + bool has_bonded_bundles; /* Flow table. */ struct classifier cls; @@ -312,13 +375,6 @@ struct ofproto { /* OpenFlow connections. */ struct connmgr *connmgr; - - /* Hooks for ovs-vswitchd. */ - const struct ofhooks *ofhooks; - void *aux; - - /* Used by default ofhooks. */ - struct mac_learning *ml; }; /* Map from dpif name to struct ofproto, for use by unixctl commands. */ @@ -326,8 +382,6 @@ static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos); static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); -static const struct ofhooks default_ofhooks; - static uint64_t pick_datapath_id(const struct ofproto *); static uint64_t pick_fallback_dpid(void); @@ -346,17 +400,24 @@ static void update_port(struct ofproto *, const char *devname); static int init_ports(struct ofproto *); static void reinit_ports(struct ofproto *); +static void update_learning_table(struct ofproto *, + const struct flow *, int vlan, + struct ofbundle *); +static bool is_admissible(struct ofproto *, const struct flow *, + bool have_packet, tag_type *, int *vlanp, + struct ofbundle **in_bundlep); + static void ofproto_unixctl_init(void); int ofproto_create(const char *datapath, const char *datapath_type, - const struct ofhooks *ofhooks, void *aux, struct ofproto **ofprotop) { char local_name[IF_NAMESIZE]; struct ofproto *p; struct dpif *dpif; int error; + int i; *ofprotop = NULL; @@ -408,9 +469,15 @@ ofproto_create(const char *datapath, const char *datapath_type, shash_init(&p->port_by_name); p->max_ports = dpif_get_max_ports(dpif); - /* Initialize submodules. */ + /* Initialize bridging. */ p->netflow = NULL; p->sflow = NULL; + hmap_init(&p->bundles); + p->ml = mac_learning_create(); + for (i = 0; i < MAX_MIRRORS; i++) { + p->mirrors[i] = NULL; + } + p->has_bonded_bundles = false; /* Initialize flow table. */ classifier_init(&p->cls); @@ -421,17 +488,6 @@ ofproto_create(const char *datapath, const char *datapath_type, p->need_revalidate = false; tag_set_init(&p->revalidate_set); - /* Initialize hooks. */ - if (ofhooks) { - p->ofhooks = ofhooks; - p->aux = aux; - p->ml = NULL; - } else { - p->ofhooks = &default_ofhooks; - p->aux = p; - p->ml = mac_learning_create(); - } - /* Pick final datapath ID. */ p->datapath_id = pick_datapath_id(p); VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id); @@ -657,19 +713,568 @@ ofproto_port_get_cfm(struct ofproto *ofproto, uint32_t port_no) struct ofport *ofport = get_port(ofproto, port_no); return ofport ? ofport->cfm : NULL; } + +/* Checks the status of LACP negotiation for 'ofp_port' within ofproto. + * Returns 1 if LACP partner information for 'ofp_port' is up-to-date, + * 0 if LACP partner information is not current (generally indicating a + * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */ +int +ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port) +{ + struct ofport *ofport = get_port(ofproto, ofp_port_to_odp_port(ofp_port)); + return (ofport && ofport->bundle && ofport->bundle->lacp + ? lacp_slave_is_current(ofport->bundle->lacp, ofport) + : -1); +} -uint64_t -ofproto_get_datapath_id(const struct ofproto *ofproto) +/* Bundles. */ + +/* Expires all MAC learning entries associated with 'port' and forces ofproto + * to revalidate every flow. */ +static void +ofproto_bundle_flush_macs(struct ofbundle *bundle) +{ + struct ofproto *ofproto = bundle->ofproto; + struct mac_learning *ml = ofproto->ml; + struct mac_entry *mac, *next_mac; + + ofproto->need_revalidate = true; + LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { + if (mac->port.p == bundle) { + mac_learning_expire(ml, mac); + } + } +} + +static struct ofbundle * +ofproto_bundle_lookup(const struct ofproto *ofproto, void *aux) +{ + struct ofbundle *bundle; + + HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), + &ofproto->bundles) { + if (bundle->aux == aux) { + return bundle; + } + } + return NULL; +} + +/* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the + * ones that are found to 'bundles'. */ +static void +ofproto_bundle_lookup_multiple(struct ofproto *ofproto, + void **auxes, size_t n_auxes, + struct hmapx *bundles) +{ + size_t i; + + hmapx_init(bundles); + for (i = 0; i < n_auxes; i++) { + struct ofbundle *bundle = ofproto_bundle_lookup(ofproto, auxes[i]); + if (bundle) { + hmapx_add(bundles, bundle); + } + } +} + +static void +ofproto_bundle_del_port(struct ofport *port) +{ + struct ofbundle *bundle = port->bundle; + + list_remove(&port->bundle_node); + port->bundle = NULL; + + if (bundle->lacp) { + lacp_slave_unregister(bundle->lacp, port); + } + if (bundle->bond) { + bond_slave_unregister(bundle->bond, port); + } + + bundle->floodable = true; + LIST_FOR_EACH (port, bundle_node, &bundle->ports) { + if (port->opp.config & htonl(OFPPC_NO_FLOOD)) { + bundle->floodable = false; + } + } +} + +static bool +ofproto_bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, + struct lacp_slave_settings *lacp) +{ + struct ofport *port; + + port = get_port(bundle->ofproto, ofp_port_to_odp_port(ofp_port)); + if (!port) { + return false; + } + + if (port->bundle != bundle) { + if (port->bundle) { + ofproto_bundle_del_port(port); + } + + port->bundle = bundle; + list_push_back(&bundle->ports, &port->bundle_node); + if (port->opp.config & htonl(OFPPC_NO_FLOOD)) { + bundle->floodable = false; + } + } + + if (lacp) { + lacp_slave_register(bundle->lacp, port, lacp); + } + + return true; +} + +void +ofproto_bundle_register(struct ofproto *ofproto, void *aux, + const struct ofproto_bundle_settings *s) +{ + bool need_flush = false; + const unsigned long *trunks; + struct ofbundle *bundle; + struct ofport *port; + size_t i; + bool ok; + + assert(s->n_slaves == 1 || s->bond != NULL); + assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); + + bundle = ofproto_bundle_lookup(ofproto, aux); + if (!bundle) { + bundle = xmalloc(sizeof *bundle); + + bundle->ofproto = ofproto; + hmap_insert(&ofproto->bundles, &bundle->hmap_node, + hash_pointer(aux, 0)); + bundle->aux = aux; + bundle->name = NULL; + + list_init(&bundle->ports); + bundle->vlan = -1; + bundle->trunks = NULL; + bundle->bond = NULL; + bundle->lacp = NULL; + + bundle->floodable = true; + + bundle->src_mirrors = 0; + bundle->dst_mirrors = 0; + bundle->mirror_out = 0; + } + + if (!bundle->name || strcmp(s->name, bundle->name)) { + free(bundle->name); + bundle->name = xstrdup(s->name); + } + + /* LACP. */ + if (s->lacp) { + if (!bundle->lacp) { + bundle->lacp = lacp_create(); + } + lacp_configure(bundle->lacp, s->lacp); + } else { + lacp_destroy(bundle->lacp); + bundle->lacp = NULL; + } + + /* Update set of ports. */ + ok = true; + for (i = 0; i < s->n_slaves; i++) { + if (!ofproto_bundle_add_port(bundle, s->slaves[i], + s->lacp ? &s->lacp_slaves[i] : NULL)) { + ok = false; + } + } + if (!ok || list_size(&bundle->ports) != s->n_slaves) { + struct ofport *next_port; + + LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { + for (i = 0; i < s->n_slaves; i++) { + if (s->slaves[i] == odp_port_to_ofp_port(port->odp_port)) { + goto found; + } + } + + ofproto_bundle_del_port(port); + found: ; + } + } + assert(list_size(&bundle->ports) <= s->n_slaves); + + if (list_is_empty(&bundle->ports)) { + ofproto_bundle_unregister(ofproto, aux); + return; + } + + /* Set VLAN tag. */ + if (s->vlan != bundle->vlan) { + bundle->vlan = s->vlan; + need_flush = true; + } + + /* Get trunked VLANs. */ + trunks = s->vlan == -1 ? NULL : s->trunks; + if (!vlan_bitmap_equal(trunks, bundle->trunks)) { + free(bundle->trunks); + bundle->trunks = vlan_bitmap_clone(trunks); + need_flush = true; + } + + /* Bonding. */ + if (!list_is_short(&bundle->ports)) { + bundle->ofproto->has_bonded_bundles = true; + if (bundle->bond) { + if (bond_reconfigure(bundle->bond, s->bond)) { + ofproto->need_revalidate = true; + } + } else { + bundle->bond = bond_create(s->bond); + } + + LIST_FOR_EACH (port, bundle_node, &bundle->ports) { + uint16_t stable_id = (bundle->lacp + ? lacp_slave_get_port_id(bundle->lacp, port) + : port->odp_port); + bond_slave_register(bundle->bond, port, stable_id, port->netdev); + } + } else { + bond_destroy(bundle->bond); + bundle->bond = NULL; + } + + /* If we changed something that would affect MAC learning, un-learn + * everything on this port and force flow revalidation. */ + if (need_flush) { + ofproto_bundle_flush_macs(bundle); + } +} + +static void +ofproto_bundle_destroy(struct ofbundle *bundle) { - return ofproto->datapath_id; + struct ofproto *ofproto; + struct ofport *port, *next_port; + int i; + + if (!bundle) { + return; + } + + ofproto = bundle->ofproto; + for (i = 0; i < MAX_MIRRORS; i++) { + struct ofmirror *m = ofproto->mirrors[i]; + if (m) { + if (m->out == bundle) { + ofproto_mirror_destroy(m); + } else if (hmapx_find_and_delete(&m->srcs, bundle) + || hmapx_find_and_delete(&m->dsts, bundle)) { + ofproto->need_revalidate = true; + } + } + } + + LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { + ofproto_bundle_del_port(port); + } + + ofproto_bundle_flush_macs(bundle); + hmap_remove(&ofproto->bundles, &bundle->hmap_node); + free(bundle->name); + free(bundle->trunks); + bond_destroy(bundle->bond); + lacp_destroy(bundle->lacp); + free(bundle); } -enum ofproto_fail_mode -ofproto_get_fail_mode(const struct ofproto *p) +void +ofproto_bundle_unregister(struct ofproto *ofproto, void *aux) { - return connmgr_get_fail_mode(p->connmgr); + ofproto_bundle_destroy(ofproto_bundle_lookup(ofproto, aux)); +} + +static void +send_pdu_cb(void *port_, const struct lacp_pdu *pdu) +{ + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); + struct ofport *port = port_; + uint8_t ea[ETH_ADDR_LEN]; + int error; + + error = netdev_get_etheraddr(port->netdev, ea); + if (!error) { + struct lacp_pdu *packet_pdu; + struct ofpbuf packet; + + ofpbuf_init(&packet, 0); + packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, + sizeof *packet_pdu); + *packet_pdu = *pdu; + error = netdev_send(port->netdev, &packet); + if (error) { + VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed " + "(%s)", port->bundle->name, + netdev_get_name(port->netdev), strerror(error)); + } + ofpbuf_uninit(&packet); + } else { + VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " + "%s (%s)", port->bundle->name, + netdev_get_name(port->netdev), strerror(error)); + } +} + +static void +ofproto_bundle_send_learning_packets(struct ofbundle *bundle) +{ + struct ofproto *ofproto = bundle->ofproto; + int error, n_packets, n_errors; + struct mac_entry *e; + + error = n_packets = n_errors = 0; + LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { + if (e->port.p != bundle) { + int ret = bond_send_learning_packet(bundle->bond, e->mac, e->vlan); + if (ret) { + error = ret; + n_errors++; + } + n_packets++; + } + } + + if (n_errors) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " + "packets, last error was: %s", + bundle->name, n_errors, n_packets, strerror(error)); + } else { + VLOG_DBG("bond %s: sent %d gratuitous learning packets", + bundle->name, n_packets); + } +} + +static void +ofproto_bundle_run(struct ofbundle *bundle) +{ + if (bundle->lacp) { + lacp_run(bundle->lacp, send_pdu_cb); + } + if (bundle->bond) { + struct ofport *port; + + LIST_FOR_EACH (port, bundle_node, &bundle->ports) { + bool may_enable = lacp_slave_may_enable(bundle->lacp, port); + bond_slave_set_lacp_may_enable(bundle->bond, port, may_enable); + } + + bond_run(bundle->bond, &bundle->ofproto->revalidate_set, + lacp_negotiated(bundle->lacp)); + if (bond_should_send_learning_packets(bundle->bond)) { + ofproto_bundle_send_learning_packets(bundle); + } + } +} + +static void +ofproto_bundle_wait(struct ofbundle *bundle) +{ + if (bundle->lacp) { + lacp_wait(bundle->lacp); + } + if (bundle->bond) { + bond_wait(bundle->bond); + } +} + +static int +ofproto_mirror_scan(struct ofproto *ofproto) +{ + int idx; + + for (idx = 0; idx < MAX_MIRRORS; idx++) { + if (!ofproto->mirrors[idx]) { + return idx; + } + } + return -1; +} + +static struct ofmirror * +ofproto_mirror_lookup(struct ofproto *ofproto, void *aux) +{ + int i; + + for (i = 0; i < MAX_MIRRORS; i++) { + struct ofmirror *mirror = ofproto->mirrors[i]; + if (mirror && mirror->aux == aux) { + return mirror; + } + } + + return NULL; +} + +void +ofproto_mirror_register(struct ofproto *ofproto, void *aux, + const struct ofproto_mirror_settings *s) +{ + mirror_mask_t mirror_bit; + struct ofbundle *bundle; + struct ofmirror *mirror; + struct ofbundle *out; + struct hmapx srcs; /* Contains "struct ofbundle *"s. */ + struct hmapx dsts; /* Contains "struct ofbundle *"s. */ + int out_vlan; + + mirror = ofproto_mirror_lookup(ofproto, aux); + if (!mirror) { + int idx; + + idx = ofproto_mirror_scan(ofproto); + if (idx < 0) { + VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " + "cannot create %s", + ofproto->name, MAX_MIRRORS, s->name); + return; + } + + mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); + mirror->ofproto = ofproto; + mirror->idx = idx; + mirror->out_vlan = -1; + mirror->name = NULL; + } + + if (!mirror->name || strcmp(s->name, mirror->name)) { + free(mirror->name); + mirror->name = xstrdup(s->name); + } + + /* Get the new configuration. */ + if (s->out_bundle) { + out = ofproto_bundle_lookup(ofproto, s->out_bundle); + if (!out) { + ofproto_mirror_destroy(mirror); + return; + } + out_vlan = -1; + } else { + out = NULL; + out_vlan = s->out_vlan; + } + ofproto_bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); + ofproto_bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); + + /* If the configuration has not changed, do nothing. */ + if (hmapx_equals(&srcs, &mirror->srcs) + && hmapx_equals(&dsts, &mirror->dsts) + && vlan_bitmap_equal(mirror->vlans, s->src_vlans) + && mirror->out == out + && mirror->out_vlan == out_vlan) + { + hmapx_destroy(&srcs); + hmapx_destroy(&dsts); + return; + } + + hmapx_swap(&srcs, &mirror->srcs); + hmapx_destroy(&srcs); + + hmapx_swap(&dsts, &mirror->dsts); + hmapx_destroy(&dsts); + + free(mirror->vlans); + mirror->vlans = vlan_bitmap_clone(s->src_vlans); + + mirror->out = out; + mirror->out_vlan = out_vlan; + + /* Update bundles. */ + mirror_bit = MIRROR_MASK_C(1) << mirror->idx; + HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { + if (hmapx_contains(&mirror->srcs, bundle)) { + bundle->src_mirrors |= mirror_bit; + } else { + bundle->src_mirrors &= ~mirror_bit; + } + + if (hmapx_contains(&mirror->dsts, bundle)) { + bundle->dst_mirrors |= mirror_bit; + } else { + bundle->dst_mirrors &= ~mirror_bit; + } + + if (mirror->out == bundle) { + bundle->mirror_out |= mirror_bit; + } else { + bundle->mirror_out &= ~mirror_bit; + } + } + + ofproto->need_revalidate = true; + mac_learning_flush(ofproto->ml); +} + +static void +ofproto_mirror_destroy(struct ofmirror *mirror) +{ + mirror_mask_t mirror_bit; + struct ofbundle *bundle; + struct ofproto *ofproto; + + if (!mirror) { + return; + } + + ofproto = mirror->ofproto; + ofproto->need_revalidate = true; + mac_learning_flush(ofproto->ml); + + mirror_bit = MIRROR_MASK_C(1) << mirror->idx; + HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { + bundle->src_mirrors &= ~mirror_bit; + bundle->dst_mirrors &= ~mirror_bit; + bundle->mirror_out &= ~mirror_bit; + } + + hmapx_destroy(&mirror->srcs); + hmapx_destroy(&mirror->dsts); + free(mirror->vlans); + + ofproto->mirrors[mirror->idx] = NULL; + free(mirror->name); + free(mirror); +} + +void +ofproto_mirror_unregister(struct ofproto *ofproto, void *aux) +{ + ofproto_mirror_destroy(ofproto_mirror_lookup(ofproto, aux)); +} + +void +ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans) +{ + if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { + ofproto->need_revalidate = true; + mac_learning_flush(ofproto->ml); + } } +bool +ofproto_is_mirror_output_bundle(struct ofproto *ofproto, void *aux) +{ + struct ofbundle *bundle = ofproto_bundle_lookup(ofproto, aux); + return bundle && bundle->mirror_out != 0; +} + bool ofproto_has_snoops(const struct ofproto *ofproto) { @@ -682,10 +1287,11 @@ ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops) connmgr_get_snoops(ofproto->connmgr, snoops); } -static void -ofproto_destroy__(struct ofproto *p, bool delete) +void +ofproto_destroy(struct ofproto *p) { struct ofport *ofport, *next_ofport; + int i; if (!p) { return; @@ -693,18 +1299,14 @@ ofproto_destroy__(struct ofproto *p, bool delete) hmap_remove(&all_ofprotos, &p->hmap_node); + for (i = 0; i < MAX_MIRRORS; i++) { + ofproto_mirror_destroy(p->mirrors[i]); + } ofproto_flush_flows__(p); connmgr_destroy(p->connmgr); classifier_destroy(&p->cls); hmap_destroy(&p->facets); - if (delete) { - int error = dpif_delete(p->dpif); - if (error && error != ENOENT) { - VLOG_ERR("bridge %s: failed to destroy (%s)", - p->name, strerror(error)); - } - } dpif_close(p->dpif); netdev_monitor_destroy(p->netdev_monitor); @@ -717,8 +1319,6 @@ ofproto_destroy__(struct ofproto *p, bool delete) netflow_destroy(p->netflow); ofproto_sflow_destroy(p->sflow); - mac_learning_destroy(p->ml); - free(p->mfr_desc); free(p->hw_desc); free(p->sw_desc); @@ -731,28 +1331,6 @@ ofproto_destroy__(struct ofproto *p, bool delete) free(p); } -void -ofproto_destroy(struct ofproto *p) -{ - ofproto_destroy__(p, false); -} - -void -ofproto_destroy_and_delete(struct ofproto *p) -{ - ofproto_destroy__(p, true); -} - -int -ofproto_run(struct ofproto *p) -{ - int error = ofproto_run1(p); - if (!error) { - error = ofproto_run2(p, false); - } - return error; -} - static void process_port_change(struct ofproto *ofproto, int error, char *devname) { @@ -765,8 +1343,9 @@ process_port_change(struct ofproto *ofproto, int error, char *devname) } int -ofproto_run1(struct ofproto *p) +ofproto_run(struct ofproto *p) { + struct ofbundle *bundle; struct ofport *ofport; char *devname; int error; @@ -804,6 +1383,10 @@ ofproto_run1(struct ofproto *p) ofport_run(ofport); } + HMAP_FOR_EACH (bundle, hmap_node, &p->bundles) { + ofproto_bundle_run(bundle); + } + connmgr_run(p->connmgr, handle_openflow); if (timer_expired(&p->next_expiration)) { @@ -819,26 +1402,16 @@ ofproto_run1(struct ofproto *p) ofproto_sflow_run(p->sflow); } - return 0; -} - -int -ofproto_run2(struct ofproto *p, bool revalidate_all) -{ - /* Figure out what we need to revalidate now, if anything. */ - struct tag_set revalidate_set = p->revalidate_set; - if (p->need_revalidate) { - revalidate_all = true; - } - - /* Clear the revalidation flags. */ - tag_set_init(&p->revalidate_set); - p->need_revalidate = false; - /* Now revalidate if there's anything to do. */ - if (revalidate_all || !tag_set_is_empty(&revalidate_set)) { + if (p->need_revalidate || !tag_set_is_empty(&p->revalidate_set)) { + struct tag_set revalidate_set = p->revalidate_set; + bool revalidate_all = p->need_revalidate; struct facet *facet, *next; + /* Clear the revalidation flags. */ + tag_set_init(&p->revalidate_set); + p->need_revalidate = false; + HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) { if (revalidate_all || tag_set_intersects(&revalidate_set, facet->tags)) { @@ -853,11 +1426,15 @@ ofproto_run2(struct ofproto *p, bool revalidate_all) void ofproto_wait(struct ofproto *p) { + struct ofbundle *bundle; struct ofport *ofport; HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { ofport_wait(ofport); } + HMAP_FOR_EACH (bundle, hmap_node, &p->bundles) { + ofproto_bundle_wait(bundle); + } dpif_recv_wait(p->dpif); dpif_port_poll_wait(p->dpif); netdev_monitor_poll_wait(p->netdev_monitor); @@ -877,18 +1454,6 @@ ofproto_wait(struct ofproto *p) connmgr_wait(p->connmgr); } -void -ofproto_revalidate(struct ofproto *ofproto, tag_type tag) -{ - tag_set_add(&ofproto->revalidate_set, tag); -} - -struct tag_set * -ofproto_get_revalidate_set(struct ofproto *ofproto) -{ - return &ofproto->revalidate_set; -} - bool ofproto_is_alive(const struct ofproto *p) { @@ -1081,15 +1646,6 @@ ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port) return error; } -/* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns - * true if 'odp_port' exists and should be included, false otherwise. */ -bool -ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port) -{ - struct ofport *ofport = get_port(ofproto, odp_port); - return ofport && !(ofport->opp.config & htonl(OFPPC_NO_FLOOD)); -} - /* Sends 'packet' out of port 'port_no' within 'p'. If 'vlan_tci' is zero the * packet will not have any 802.1Q hader; if it is nonzero, then the packet * will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'. @@ -1301,7 +1857,9 @@ ofport_install(struct ofproto *p, ofport->netdev = netdev; ofport->opp = *opp; ofport->odp_port = ofp_port_to_odp_port(ntohs(opp->port_no)); + ofport->bundle = NULL; ofport->cfm = NULL; + ofport->tag = tag_create_random(); /* Add port to 'p'. */ netdev_monitor_add(p->netdev_monitor, ofport->netdev); @@ -1316,19 +1874,8 @@ ofport_install(struct ofproto *p, static void ofport_remove(struct ofport *ofport) { - struct ofproto *p = ofport->ofproto; - - connmgr_send_port_status(p->connmgr, &ofport->opp, OFPPR_DELETE); - - netdev_monitor_remove(p->netdev_monitor, ofport->netdev); - hmap_remove(&p->ports, &ofport->hmap_node); - shash_delete(&p->port_by_name, - shash_find(&p->port_by_name, - netdev_get_name(ofport->netdev))); - if (p->sflow) { - ofproto_sflow_del_port(p->sflow, ofport->odp_port); - } - + connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->opp, + OFPPR_DELETE); ofport_free(ofport); } @@ -1353,6 +1900,10 @@ ofport_modified(struct ofport *port, { struct ofproto *ofproto = port->ofproto; + if (port->bundle && port->bundle->bond) { + bond_slave_set_netdev(port->bundle->bond, port, netdev); + } + memcpy(port->opp.hw_addr, opp->hw_addr, ETH_ADDR_LEN); port->opp.config = ((port->opp.config & ~htonl(OFPPC_PORT_DOWN)) | (opp->config & htonl(OFPPC_PORT_DOWN))); @@ -1400,12 +1951,52 @@ ofport_wait(struct ofport *ofport) } static void -ofport_free(struct ofport *ofport) +ofport_unregister(struct ofport *port) { - if (ofport) { - cfm_destroy(ofport->cfm); - netdev_close(ofport->netdev); - free(ofport); + struct ofbundle *bundle = port->bundle; + + if (bundle) { + ofproto_bundle_del_port(port); + if (list_is_empty(&bundle->ports)) { + ofproto_bundle_destroy(bundle); + } else if (list_is_short(&bundle->ports)) { + bond_destroy(bundle->bond); + bundle->bond = NULL; + } + } + + cfm_destroy(port->cfm); + port->cfm = NULL; +} + +void +ofproto_port_unregister(struct ofproto *ofproto, uint32_t ofp_port) +{ + struct ofport *port = get_port(ofproto, ofp_port_to_odp_port(ofp_port)); + if (port) { + ofport_unregister(port); + } +} + +static void +ofport_free(struct ofport *port) +{ + if (port) { + struct ofproto *ofproto = port->ofproto; + const char *name = netdev_get_name(port->netdev); + + ofport_unregister(port); + + netdev_monitor_remove(ofproto->netdev_monitor, port->netdev); + hmap_remove(&ofproto->ports, &port->hmap_node); + shash_delete(&ofproto->port_by_name, + shash_find(&ofproto->port_by_name, name)); + if (ofproto->sflow) { + ofproto_sflow_del_port(ofproto->sflow, port->odp_port); + } + + netdev_close(port->netdev); + free(port); } } @@ -1829,21 +2420,49 @@ facet_install(struct ofproto *p, struct facet *facet, bool zero_stats) } } -/* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up - * to the accounting hook function in the ofhooks structure. */ static void facet_account(struct ofproto *ofproto, struct facet *facet, uint64_t extra_bytes) { - uint64_t total_bytes = facet->byte_count + extra_bytes; + uint64_t total_bytes, n_bytes; + struct ofbundle *in_bundle; + const struct nlattr *a; + tag_type dummy = 0; + unsigned int left; + int vlan; - if (ofproto->ofhooks->account_flow_cb - && total_bytes > facet->accounted_bytes) - { - ofproto->ofhooks->account_flow_cb( - &facet->flow, facet->tags, facet->actions, facet->actions_len, - total_bytes - facet->accounted_bytes, ofproto->aux); - facet->accounted_bytes = total_bytes; + total_bytes = facet->byte_count + extra_bytes; + if (total_bytes <= facet->accounted_bytes) { + return; + } + n_bytes = total_bytes - facet->accounted_bytes; + facet->accounted_bytes = total_bytes; + + /* Test that 'tags' is nonzero to ensure that only flows that include an + * OFPP_NORMAL action are used for learning and bond slave rebalancing. + * This works because OFPP_NORMAL always sets a nonzero tag value. + * + * 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. */ + if (!facet->tags + || !is_admissible(ofproto, &facet->flow, false, &dummy, + &vlan, &in_bundle)) { + return; + } + + update_learning_table(ofproto, &facet->flow, vlan, in_bundle); + + if (!ofproto->has_bonded_bundles) { + return; + } + NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) { + if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) { + struct ofport *port = get_port(ofproto, nl_attr_get_u32(a)); + if (port && port->bundle && port->bundle->bond) { + bond_account(port->bundle->bond, &facet->flow, vlan, n_bytes); + } + } } } @@ -2041,6 +2660,507 @@ facet_revalidate(struct ofproto *ofproto, struct facet *facet) return true; } +/* Bridge packet processing functions. */ + +struct dst { + struct ofport *port; + uint16_t vlan; +}; + +struct dst_set { + struct dst builtin[32]; + struct dst *dsts; + size_t n, allocated; +}; + +static void dst_set_init(struct dst_set *); +static void dst_set_add(struct dst_set *, const struct dst *); +static void dst_set_free(struct dst_set *); + +static struct ofport *ofbundle_get_a_port(const struct ofbundle *); + +static bool +set_dst(struct action_xlate_ctx *ctx, struct dst *dst, + const struct ofbundle *in_bundle, const struct ofbundle *out_bundle) +{ + dst->vlan = (out_bundle->vlan >= 0 ? OFP_VLAN_NONE + : in_bundle->vlan >= 0 ? in_bundle->vlan + : ctx->flow.vlan_tci == 0 ? OFP_VLAN_NONE + : vlan_tci_to_vid(ctx->flow.vlan_tci)); + + dst->port = (!out_bundle->bond + ? ofbundle_get_a_port(out_bundle) + : bond_choose_output_slave(out_bundle->bond, &ctx->flow, + dst->vlan, &ctx->tags)); + + return dst->port != NULL; +} + +static int +mirror_mask_ffs(mirror_mask_t mask) +{ + BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); + return ffs(mask); +} + +static void +dst_set_init(struct dst_set *set) +{ + set->dsts = set->builtin; + set->n = 0; + set->allocated = ARRAY_SIZE(set->builtin); +} + +static void +dst_set_add(struct dst_set *set, const struct dst *dst) +{ + if (set->n >= set->allocated) { + size_t new_allocated; + struct dst *new_dsts; + + new_allocated = set->allocated * 2; + new_dsts = xmalloc(new_allocated * sizeof *new_dsts); + memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts); + + dst_set_free(set); + + set->dsts = new_dsts; + set->allocated = new_allocated; + } + set->dsts[set->n++] = *dst; +} + +static void +dst_set_free(struct dst_set *set) +{ + if (set->dsts != set->builtin) { + free(set->dsts); + } +} + +static bool +dst_is_duplicate(const struct dst_set *set, const struct dst *test) +{ + size_t i; + for (i = 0; i < set->n; i++) { + if (set->dsts[i].vlan == test->vlan + && set->dsts[i].port == test->port) { + return true; + } + } + return false; +} + +static bool +ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) +{ + return bundle->vlan < 0 && vlan_bitmap_contains(bundle->trunks, vlan); +} + +static bool +ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) +{ + return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); +} + +/* Returns an arbitrary interface within 'bundle'. */ +static struct ofport * +ofbundle_get_a_port(const struct ofbundle *bundle) +{ + return CONTAINER_OF(list_front(&bundle->ports), + struct ofport, bundle_node); +} + +static void +compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan, + const struct ofbundle *in_bundle, + const struct ofbundle *out_bundle, struct dst_set *set) +{ + struct dst dst; + + if (out_bundle == OFBUNDLE_FLOOD) { + struct ofbundle *bundle; + + HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { + if (bundle != in_bundle + && ofbundle_includes_vlan(bundle, vlan) + && bundle->floodable + && !bundle->mirror_out + && set_dst(ctx, &dst, in_bundle, bundle)) { + dst_set_add(set, &dst); + } + } + ctx->nf_output_iface = NF_OUT_FLOOD; + } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) { + dst_set_add(set, &dst); + ctx->nf_output_iface = dst.port->odp_port; + } +} + +static bool +vlan_is_mirrored(const struct ofmirror *m, int vlan) +{ + return vlan_bitmap_contains(m->vlans, vlan); +} + +static void +compose_mirror_dsts(struct action_xlate_ctx *ctx, + uint16_t vlan, const struct ofbundle *in_bundle, + struct dst_set *set) +{ + struct ofproto *ofproto = ctx->ofproto; + mirror_mask_t mirrors; + int flow_vlan; + size_t i; + + mirrors = in_bundle->src_mirrors; + for (i = 0; i < set->n; i++) { + mirrors |= set->dsts[i].port->bundle->dst_mirrors; + } + + if (!mirrors) { + return; + } + + flow_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci); + if (flow_vlan == 0) { + flow_vlan = OFP_VLAN_NONE; + } + + while (mirrors) { + struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; + if (vlan_is_mirrored(m, vlan)) { + struct dst dst; + + if (m->out) { + if (set_dst(ctx, &dst, in_bundle, m->out) + && !dst_is_duplicate(set, &dst)) { + dst_set_add(set, &dst); + } + } else { + struct ofbundle *bundle; + + HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { + if (ofbundle_includes_vlan(bundle, m->out_vlan) + && set_dst(ctx, &dst, in_bundle, bundle)) + { + if (bundle->vlan < 0) { + dst.vlan = m->out_vlan; + } + if (dst_is_duplicate(set, &dst)) { + continue; + } + + /* Use the vlan tag on the original flow instead of + * the one passed in the vlan parameter. This ensures + * that we compare the vlan from before any implicit + * tagging tags place. This is necessary because + * dst->vlan is the final vlan, after removing implicit + * tags. */ + if (bundle == in_bundle && dst.vlan == flow_vlan) { + /* Don't send out input port on same VLAN. */ + continue; + } + dst_set_add(set, &dst); + } + } + } + } + mirrors &= mirrors - 1; + } +} + +static void +compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan, + const struct ofbundle *in_bundle, + const struct ofbundle *out_bundle) +{ + uint16_t initial_vlan, cur_vlan; + const struct dst *dst; + struct dst_set set; + + dst_set_init(&set); + compose_dsts(ctx, vlan, in_bundle, out_bundle, &set); + compose_mirror_dsts(ctx, vlan, in_bundle, &set); + + /* Output all the packets we can without having to change the VLAN. */ + initial_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci); + if (initial_vlan == 0) { + initial_vlan = OFP_VLAN_NONE; + } + for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) { + if (dst->vlan != initial_vlan) { + continue; + } + nl_msg_put_u32(ctx->odp_actions, + ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port); + } + + /* Then output the rest. */ + cur_vlan = initial_vlan; + for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) { + if (dst->vlan == initial_vlan) { + continue; + } + if (dst->vlan != cur_vlan) { + if (dst->vlan == OFP_VLAN_NONE) { + nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN); + } else { + ovs_be16 tci; + tci = htons(dst->vlan & VLAN_VID_MASK); + tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); + nl_msg_put_be16(ctx->odp_actions, + ODP_ACTION_ATTR_SET_DL_TCI, tci); + } + cur_vlan = dst->vlan; + } + nl_msg_put_u32(ctx->odp_actions, + ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port); + } + + dst_set_free(&set); +} + +/* Returns the effective vlan of a packet, taking into account both the + * 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 ofproto *ofproto, const struct flow *flow, + struct ofbundle *in_bundle, bool have_packet) +{ + int vlan = vlan_tci_to_vid(flow->vlan_tci); + if (in_bundle->vlan >= 0) { + if (vlan) { + if (have_packet) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " + "packet received on port %s configured with " + "implicit VLAN %"PRIu16, + ofproto->name, vlan, + in_bundle->name, in_bundle->vlan); + } + return -1; + } + vlan = in_bundle->vlan; + } else { + if (!ofbundle_includes_vlan(in_bundle, vlan)) { + if (have_packet) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " + "packet received on port %s not configured for " + "trunking VLAN %d", + ofproto->name, vlan, in_bundle->name, vlan); + } + return -1; + } + } + + 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 struct flow *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 ofproto *ofproto, + const struct flow *flow, int vlan, + struct ofbundle *in_bundle) +{ + struct mac_entry *mac; + + if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { + return; + } + + mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); + if (is_gratuitous_arp(flow)) { + /* We don't want to learn from gratuitous ARP packets that are + * reflected back over bond slaves so we lock the learning table. */ + if (!in_bundle->bond) { + mac_entry_set_grat_arp_lock(mac); + } else if (mac_entry_is_grat_arp_locked(mac)) { + return; + } + } + + if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { + /* The log messages here could actually be useful in debugging, + * so keep the rate limit relatively high. */ + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); + VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " + "on port %s in VLAN %d", + ofproto->name, ETH_ADDR_ARGS(flow->dl_src), + in_bundle->name, vlan); + + mac->port.p = in_bundle; + tag_set_add(&ofproto->revalidate_set, + mac_learning_changed(ofproto->ml, mac)); + } +} + +/* 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_admissible(struct ofproto *ofproto, const struct flow *flow, + bool have_packet, + tag_type *tags, int *vlanp, struct ofbundle **in_bundlep) +{ + struct ofport *in_port; + struct ofbundle *in_bundle; + int vlan; + + /* Find the port and bundle for the received packet. */ + in_port = get_port(ofproto, flow->in_port); + *in_bundlep = in_bundle = in_port->bundle; + if (!in_port || !in_bundle) { + /* No interface? Something fishy... */ + if (have_packet) { + /* Odd. A few possible reasons here: + * + * - We deleted a port but there are still a few packets queued up + * from it. + * + * - Someone externally added a port (e.g. "ovs-dpctl add-if") that + * we don't know about. + * + * - Packet arrived on the local port but the local port is not + * part of a bundle. + */ + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + + VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " + "port %"PRIu16, + ofproto->name, flow->in_port); + } + return false; + } + *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet); + if (vlan < 0) { + return false; + } + + /* Drop frames for reserved multicast addresses. */ + if (eth_addr_is_reserved(flow->dl_dst)) { + return false; + } + + /* Drop frames on bundles reserved for mirroring. */ + if (in_bundle->mirror_out) { + 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", + ofproto->name, in_bundle->name); + } + return false; + } + + if (in_bundle->bond) { + struct mac_entry *mac; + + switch (bond_check_admissibility(in_bundle->bond, in_port, + flow->dl_dst, tags)) { + case BV_ACCEPT: + break; + + case BV_DROP: + return false; + + case BV_DROP_IF_MOVED: + mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); + if (mac && mac->port.p != in_bundle && + (!is_gratuitous_arp(flow) + || mac_entry_is_grat_arp_locked(mac))) { + return false; + } + break; + } + } + + 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 +xlate_normal(struct action_xlate_ctx *ctx) +{ + struct ofbundle *in_bundle; + struct ofbundle *out_bundle; + struct mac_entry *mac; + int vlan; + + /* Check whether we should drop packets in this flow. */ + if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL, + &ctx->tags, &vlan, &in_bundle)) { + out_bundle = NULL; + goto done; + } + + /* Learn source MAC (but don't try to learn from revalidation). */ + if (ctx->packet) { + update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); + } + + /* Determine output bundle. */ + mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, + &ctx->tags); + if (mac) { + out_bundle = mac->port.p; + } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) { + /* If we are revalidating but don't have a learning entry then eject + * the flow. Installing a flow that floods packets opens up a window + * of time where we could learn from a packet reflected on a bond and + * blackhole packets before the learning table is updated to reflect + * the correct port. */ + return false; + } else { + out_bundle = OFBUNDLE_FLOOD; + } + + /* Don't send packets out their input bundles. */ + if (in_bundle == out_bundle) { + out_bundle = NULL; + } + +done: + if (in_bundle) { + compose_actions(ctx, vlan, in_bundle, out_bundle); + } + + return true; +} + static void send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh, int error) @@ -2238,13 +3358,7 @@ xlate_output_action__(struct action_xlate_ctx *ctx, xlate_table_action(ctx, ctx->flow.in_port); break; case OFPP_NORMAL: - if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet, - ctx->odp_actions, &ctx->tags, - &ctx->nf_output_iface, - ctx->ofproto->aux)) { - COVERAGE_INC(ofproto_uninstallable); - ctx->may_set_up_flow = false; - } + xlate_normal(ctx); break; case OFPP_FLOOD: flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(OFPPC_NO_FLOOD), @@ -2402,6 +3516,27 @@ update_reg_state(struct action_xlate_ctx *ctx, } } +static void +xlate_autopath(struct action_xlate_ctx *ctx, + const struct nx_action_autopath *naa) +{ + uint16_t ofp_port = ntohl(naa->id); + struct ofport *port; + + port = get_port(ctx->ofproto, ofp_port_to_odp_port(ofp_port)); + if (!port || !port->bundle) { + ofp_port = OFPP_NONE; + } else if (port->bundle->bond) { + /* Autopath does not support VLAN hashing. */ + struct ofport *slave = bond_choose_output_slave( + port->bundle->bond, &ctx->flow, OFP_VLAN_NONE, &ctx->tags); + if (slave) { + ofp_port = odp_port_to_ofp_port(slave->odp_port); + } + } + autopath_execute(naa, &ctx->flow, ofp_port); +} + static void xlate_nicira_action(struct action_xlate_ctx *ctx, const struct nx_action_header *nah) @@ -2412,9 +3547,7 @@ xlate_nicira_action(struct action_xlate_ctx *ctx, const struct nx_action_multipath *nam; const struct nx_action_autopath *naa; enum nx_action_subtype subtype = ntohs(nah->subtype); - const struct ofhooks *ofhooks = ctx->ofproto->ofhooks; struct xlate_reg_state state; - uint16_t autopath_port; ovs_be64 tun_id; assert(nah->vendor == htonl(NX_VENDOR_ID)); @@ -2478,11 +3611,7 @@ xlate_nicira_action(struct action_xlate_ctx *ctx, case NXAST_AUTOPATH: naa = (const struct nx_action_autopath *) nah; - autopath_port = (ofhooks->autopath_cb - ? ofhooks->autopath_cb(&ctx->flow, ntohl(naa->id), - &ctx->tags, ctx->ofproto->aux) - : OFPP_NONE); - autopath_execute(naa, &ctx->flow, autopath_port); + xlate_autopath(ctx, naa); break; /* If you add a new action here that modifies flow data, don't forget to @@ -2610,16 +3739,27 @@ action_xlate_ctx_init(struct action_xlate_ctx *ctx, ctx->check_special = true; } -static void -ofproto_process_cfm(struct ofproto *ofproto, const struct flow *flow, - const struct ofpbuf *packet) +static bool +ofproto_process_special(struct ofproto *ofproto, const struct flow *flow, + const struct ofpbuf *packet) { - struct ofport *ofport; - - ofport = get_port(ofproto, flow->in_port); - if (ofport && ofport->cfm) { - cfm_process_heartbeat(ofport->cfm, packet); + if (cfm_should_process_flow(flow)) { + struct ofport *ofport = get_port(ofproto, flow->in_port); + if (ofport && ofport->cfm) { + cfm_process_heartbeat(ofport->cfm, packet); + } + return true; + } else if (flow->dl_type == htons(ETH_TYPE_LACP)) { + struct ofport *port = get_port(ofproto, flow->in_port); + if (port && port->bundle && port->bundle->lacp) { + const struct lacp_pdu *pdu = parse_lacp_packet(packet); + if (pdu) { + lacp_process_pdu(port->bundle->lacp, port, pdu); + } + return true; + } } + return false; } static struct ofpbuf * @@ -2635,15 +3775,8 @@ xlate_actions(struct action_xlate_ctx *ctx, ctx->recurse = 0; ctx->last_pop_priority = -1; - if (ctx->check_special && cfm_should_process_flow(&ctx->flow)) { - if (ctx->packet) { - ofproto_process_cfm(ctx->ofproto, &ctx->flow, ctx->packet); - } - ctx->may_set_up_flow = false; - } else if (ctx->check_special - && ctx->ofproto->ofhooks->special_cb - && !ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet, - ctx->ofproto->aux)) { + if (ctx->check_special + && ofproto_process_special(ctx->ofproto, &ctx->flow, ctx->packet)) { ctx->may_set_up_flow = false; } else { do_xlate_actions(in, n_in, ctx); @@ -3940,12 +5073,8 @@ handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall) /* Set header pointers in 'flow'. */ flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow); - if (cfm_should_process_flow(&flow)) { - ofproto_process_cfm(p, &flow, upcall->packet); - ofpbuf_delete(upcall->packet); - return; - } else if (p->ofhooks->special_cb - && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) { + /* Handle 802.1ag and LACP. */ + if (ofproto_process_special(p, &flow, upcall->packet)) { ofpbuf_delete(upcall->packet); return; } @@ -4068,12 +5197,16 @@ ofproto_expire(struct ofproto *ofproto) rule_expire(ofproto, rule); } - /* Let the hook know that we're at a stable point: all outstanding data - * in existing flows has been accounted to the account_cb. Thus, the - * hook can now reasonably do operations that depend on having accurate - * flow volume accounting (currently, that's just bond rebalancing). */ - if (ofproto->ofhooks->account_checkpoint_cb) { - ofproto->ofhooks->account_checkpoint_cb(ofproto->aux); + /* All outstanding data in existing flows has been accounted, so it's a + * good time to do bond rebalancing. */ + if (ofproto->has_bonded_bundles) { + struct ofbundle *bundle; + + HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { + if (bundle->bond) { + bond_rebalance(bundle->bond, &ofproto->revalidate_set); + } + } } return MIN(dp_max_idle, 1000); @@ -4571,6 +5704,31 @@ exit: free(args); } +static void +ofproto_unixctl_fdb_show(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + const struct ofproto *ofproto; + const struct mac_entry *e; + + ofproto = ofproto_lookup(args); + if (!ofproto) { + unixctl_command_reply(conn, 501, "no such bridge"); + return; + } + + ds_put_cstr(&ds, " port VLAN MAC Age\n"); + LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { + struct ofbundle *bundle = e->port.p; + ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", + ofbundle_get_a_port(bundle)->odp_port, + e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); + } + unixctl_command_reply(conn, 200, ds_cstr(&ds)); + ds_destroy(&ds); +} + static void ofproto_unixctl_init(void) { @@ -4582,62 +5740,5 @@ ofproto_unixctl_init(void) unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL); unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL); + unixctl_command_register("fdb/show", ofproto_unixctl_fdb_show, NULL); } - -static bool -default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, - struct ofpbuf *odp_actions, tag_type *tags, - uint16_t *nf_output_iface, void *ofproto_) -{ - struct ofproto *ofproto = ofproto_; - struct mac_entry *dst_mac; - - /* Drop frames for reserved multicast addresses. */ - if (eth_addr_is_reserved(flow->dl_dst)) { - return true; - } - - /* Learn source MAC (but don't try to learn from revalidation). */ - if (packet != NULL - && mac_learning_may_learn(ofproto->ml, flow->dl_src, 0)) { - struct mac_entry *src_mac; - - src_mac = mac_learning_insert(ofproto->ml, flow->dl_src, 0); - if (mac_entry_is_new(src_mac) || src_mac->port.i != flow->in_port) { - /* The log messages here could actually be useful in debugging, - * so keep the rate limit relatively high. */ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); - VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16, - ETH_ADDR_ARGS(flow->dl_src), flow->in_port); - - ofproto_revalidate(ofproto, - mac_learning_changed(ofproto->ml, src_mac)); - src_mac->port.i = flow->in_port; - } - } - - /* Determine output port. */ - dst_mac = mac_learning_lookup(ofproto->ml, flow->dl_dst, 0, tags); - if (!dst_mac) { - flood_packets(ofproto, flow->in_port, htonl(OFPPC_NO_FLOOD), - nf_output_iface, odp_actions); - } else { - int out_port = dst_mac->port.i; - if (out_port != flow->in_port) { - nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port); - *nf_output_iface = out_port; - } else { - /* Drop. */ - } - } - - return true; -} - -static const struct ofhooks default_ofhooks = { - default_normal_ofhook_cb, - NULL, - NULL, - NULL, - NULL -}; diff --git a/ofproto/ofproto.h b/ofproto/ofproto.h index cd3671c4..2b8011da 100644 --- a/ofproto/ofproto.h +++ b/ofproto/ofproto.h @@ -35,8 +35,6 @@ struct cfm; struct cls_rule; struct dpif_port; struct netdev; -struct nlattr; -struct ofhooks; struct ofproto; struct shash; @@ -96,13 +94,9 @@ struct ofproto_controller { #define DEFAULT_DP_DESC "None" int ofproto_create(const char *datapath, const char *datapath_type, - const struct ofhooks *, void *aux, struct ofproto **ofprotop); void ofproto_destroy(struct ofproto *); -void ofproto_destroy_and_delete(struct ofproto *); int ofproto_run(struct ofproto *); -int ofproto_run1(struct ofproto *); -int ofproto_run2(struct ofproto *, bool revalidate_all); void ofproto_wait(struct ofproto *); bool ofproto_is_alive(const struct ofproto *); @@ -142,7 +136,6 @@ int ofproto_port_dump_done(struct ofproto_port_dump *); int ofproto_port_add(struct ofproto *, struct netdev *, uint16_t *ofp_portp); int ofproto_port_del(struct ofproto *, uint16_t ofp_port); -bool ofproto_port_is_floodable(struct ofproto *, uint16_t odp_port); int ofproto_port_query_by_name(const struct ofproto *, const char *devname, struct ofproto_port *); @@ -165,42 +158,72 @@ int ofproto_set_netflow(struct ofproto *, const struct netflow_options *nf_options); void ofproto_set_sflow(struct ofproto *, const struct ofproto_sflow_options *); -/* Configuration of individual interfaces. */ +/* Configuration of ports. */ + +void ofproto_port_unregister(struct ofproto *, uint32_t port_no); + void ofproto_port_clear_cfm(struct ofproto *, uint32_t port_no); void ofproto_port_set_cfm(struct ofproto *, uint32_t port_no, const struct cfm *, const uint16_t *remote_mps, size_t n_remote_mps); const struct cfm *ofproto_port_get_cfm(struct ofproto *, uint32_t port_no); +int ofproto_port_is_lacp_current(struct ofproto *, uint16_t ofp_port); + +/* Configuration of bundles. */ +struct ofproto_bundle_settings { + char *name; /* For use in log messages. */ + + uint32_t *slaves; /* OpenFlow port numbers for slaves. */ + size_t n_slaves; + + int vlan; /* VLAN if access port, -1 if trunk port. */ + unsigned long *trunks; /* vlan_bitmap, NULL to trunk all VLANs. */ + + struct bond_settings *bond; /* Must be nonnull iff if n_slaves > 1. */ + + struct lacp_settings *lacp; /* Nonnull to enable LACP. */ + struct lacp_slave_settings *lacp_slaves; /* Array of n_slaves elements. */ +}; + +void ofproto_bundle_register(struct ofproto *, void *aux, + const struct ofproto_bundle_settings *); +void ofproto_bundle_unregister(struct ofproto *, void *aux); + +/* Configuration of mirrors. */ +struct ofproto_mirror_settings { + /* Name for log messages. */ + char *name; + + /* Bundles that select packets for mirroring upon ingress. */ + void **srcs; /* A set of registered ofbundle handles. */ + size_t n_srcs; + + /* Bundles that select packets for mirroring upon egress. */ + void **dsts; /* A set of registered ofbundle handles. */ + size_t n_dsts; + + /* VLANs of packets to select for mirroring. */ + unsigned long *src_vlans; /* vlan_bitmap, NULL selects all VLANs. */ + + /* Output (mutually exclusive). */ + void *out_bundle; /* A registered ofbundle handle or NULL. */ + uint16_t out_vlan; /* Output VLAN, only if out_bundle is NULL. */ +}; + +void ofproto_mirror_register(struct ofproto *, void *aux, + const struct ofproto_mirror_settings *); +void ofproto_mirror_unregister(struct ofproto *, void *aux); + +void ofproto_set_flood_vlans(struct ofproto *, unsigned long *flood_vlans); +bool ofproto_is_mirror_output_bundle(struct ofproto *, void *aux); /* Configuration querying. */ -uint64_t ofproto_get_datapath_id(const struct ofproto *); -enum ofproto_fail_mode ofproto_get_fail_mode(const struct ofproto *); -void ofproto_get_listeners(const struct ofproto *, struct sset *); bool ofproto_has_snoops(const struct ofproto *); void ofproto_get_snoops(const struct ofproto *, struct sset *); void ofproto_get_all_flows(struct ofproto *p, struct ds *); void ofproto_get_netflow_ids(const struct ofproto *, uint8_t *engine_type, uint8_t *engine_id); -/* Hooks for ovs-vswitchd. */ -struct ofhooks { - bool (*normal_cb)(const struct flow *, const struct ofpbuf *packet, - struct ofpbuf *odp_actions, tag_type *, - uint16_t *nf_output_iface, void *aux); - bool (*special_cb)(const struct flow *flow, const struct ofpbuf *packet, - void *aux); - void (*account_flow_cb)(const struct flow *, tag_type tags, - const struct nlattr *odp_actions, - size_t actions_len, - uint64_t n_bytes, void *aux); - void (*account_checkpoint_cb)(void *aux); - - uint16_t (*autopath_cb)(const struct flow *, uint32_t id, - tag_type *, void *aux); -}; -void ofproto_revalidate(struct ofproto *, tag_type); -struct tag_set *ofproto_get_revalidate_set(struct ofproto *); - void ofproto_get_ofproto_controller_info(const struct ofproto *, struct shash *); void ofproto_free_ofproto_controller_info(struct shash *); diff --git a/utilities/ovs-openflowd.c b/utilities/ovs-openflowd.c index f54eb355..c4d41d1e 100644 --- a/utilities/ovs-openflowd.c +++ b/utilities/ovs-openflowd.c @@ -115,7 +115,7 @@ main(int argc, char *argv[]) VLOG_INFO("Open vSwitch version %s", VERSION BUILDNR); VLOG_INFO("OpenFlow protocol version 0x%02x", OFP_VERSION); - error = ofproto_create(s.dp_name, s.dp_type, NULL, NULL, &ofproto); + error = ofproto_create(s.dp_name, s.dp_type, &ofproto); if (error) { VLOG_FATAL("could not initialize OpenFlow switch (%s)", strerror(error)); diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c index 876ed54c..0aa395e0 100644 --- a/vswitchd/bridge.c +++ b/vswitchd/bridge.c @@ -77,25 +77,8 @@ VLOG_DEFINE_THIS_MODULE(bridge); -COVERAGE_DEFINE(bridge_flush); -COVERAGE_DEFINE(bridge_process_flow); COVERAGE_DEFINE(bridge_reconfigure); -struct dst { - struct iface *iface; - uint16_t vlan; -}; - -struct dst_set { - struct dst builtin[32]; - struct dst *dsts; - size_t n, allocated; -}; - -static void dst_set_init(struct dst_set *); -static void dst_set_add(struct dst_set *, const struct dst *); -static void dst_set_free(struct dst_set *); - struct iface { /* These members are always valid. */ struct list port_elem; /* Element in struct port's "ifaces" list. */ @@ -113,58 +96,29 @@ struct iface { const struct ovsrec_interface *cfg; }; -#define MAX_MIRRORS 32 -typedef uint32_t mirror_mask_t; -#define MIRROR_MASK_C(X) UINT32_C(X) -BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); struct mirror { + struct uuid uuid; /* UUID of this "mirror" record in database. */ + struct hmap_node hmap_node; /* In struct bridge's "mirrors" hmap. */ struct bridge *bridge; - size_t idx; char *name; - struct uuid uuid; /* UUID of this "mirror" record in database. */ - - /* Selection criteria. */ - struct sset src_ports; /* Source port names. */ - struct sset dst_ports; /* Destination port names. */ - int *vlans; - size_t n_vlans; - - /* Output. */ - struct port *out_port; - int out_vlan; }; -#define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */ struct port { struct bridge *bridge; struct hmap_node hmap_node; /* Element in struct bridge's "ports" hmap. */ char *name; - int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ - unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. - * NULL if all VLANs are trunked. */ const struct ovsrec_port *cfg; /* An ordinary bridge port has 1 interface. * A bridge port for bonding has at least 2 interfaces. */ struct list ifaces; /* List of "struct iface"s. */ - - struct lacp *lacp; /* NULL if LACP is not enabled. */ - - /* Bonding info. */ - struct bond *bond; - - /* 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? */ }; struct bridge { struct hmap_node node; /* In 'all_bridges'. */ char *name; /* User-specified arbitrary name. */ char *type; /* Datapath type. */ - struct mac_learning *ml; /* MAC learning table. */ uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */ uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */ const struct ovsrec_bridge *cfg; @@ -177,14 +131,8 @@ struct bridge { struct hmap ifaces; /* "struct iface"s indexed by dp_ifidx. */ struct hmap iface_by_name; /* "struct iface"s indexed by name. */ - /* Bonding. */ - bool has_bonded_ports; - - /* Flow tracking. */ - bool flush; - /* Port mirroring. */ - struct mirror *mirrors[MAX_MIRRORS]; + struct hmap mirrors; /* "struct mirror" indexed by UUID. */ /* Synthetic local port if necessary. */ struct ovsrec_port synth_local_port; @@ -217,7 +165,6 @@ 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 size_t bridge_get_controllers(const struct bridge *br, struct ovsrec_controller ***controllersp); static void bridge_add_del_ports(struct bridge *); @@ -227,10 +174,9 @@ static void bridge_refresh_dp_ifidx(struct bridge *); static void bridge_configure_datapath_id(struct bridge *); static void bridge_configure_netflow(struct bridge *); static void bridge_configure_sflow(struct bridge *, int *sflow_bridge_number); -static void bridge_reconfigure_remotes(struct bridge *, - const struct sockaddr_in *managers, - size_t n_managers); -static void bridge_flush(struct bridge *); +static void bridge_configure_remotes(struct bridge *, + const struct sockaddr_in *managers, + size_t n_managers); static void bridge_pick_local_hw_addr(struct bridge *, uint8_t ea[ETH_ADDR_LEN], struct iface **hw_addr_iface); @@ -242,31 +188,26 @@ static bool bridge_has_bond_fake_iface(const struct bridge *, const char *name); static bool port_is_bond_fake_iface(const struct port *); -static unixctl_cb_func bridge_unixctl_fdb_show; static unixctl_cb_func cfm_unixctl_show; static unixctl_cb_func qos_unixctl_show; -static void port_run(struct port *); -static void port_wait(struct port *); -static void port_reconfigure(struct port *); static struct port *port_create(struct bridge *, const struct ovsrec_port *); static void port_add_ifaces(struct port *); static void port_del_ifaces(struct port *); static void port_destroy(struct port *); static struct port *port_lookup(const struct bridge *, const char *name); -static struct iface *port_get_an_iface(const struct port *); -static struct port *port_from_dp_ifidx(const struct bridge *, - uint16_t dp_ifidx); -static void port_reconfigure_lacp(struct port *); -static void port_reconfigure_bond(struct port *); -static void port_send_learning_packets(struct port *); - -static void mirror_create(struct bridge *, struct ovsrec_mirror *); +static void port_configure(struct port *); +static struct lacp_settings *port_configure_lacp(struct port *, + struct lacp_settings *); +static void port_configure_bond(struct port *, struct bond_settings *); + +static void bridge_configure_mirrors(struct bridge *); +static struct mirror *mirror_create(struct bridge *, + const 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 *); -static bool vlan_is_mirrored(const struct mirror *, int vlan); +static bool mirror_configure(struct mirror *, const struct ovsrec_mirror *); +static void iface_configure_lacp(struct iface *, struct lacp_slave_settings *); static struct iface *iface_create(struct port *port, const struct ovsrec_interface *if_cfg); static void iface_destroy(struct iface *); @@ -286,9 +227,6 @@ static void shash_from_ovs_idl_map(char **keys, char **values, size_t n, struct shash *); static void shash_to_ovs_idl_map(struct shash *, char ***keys, char ***values, size_t *n); - -/* Hooks into ofproto processing. */ -static struct ofhooks bridge_ofhooks; /* Public functions. */ @@ -353,7 +291,6 @@ bridge_init(const char *remote) ovsdb_idl_omit(idl, &ovsrec_ssl_col_external_ids); /* Register unixctl commands. */ - unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL); unixctl_command_register("cfm/show", cfm_unixctl_show, NULL); unixctl_command_register("qos/show", qos_unixctl_show, NULL); unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows, @@ -487,13 +424,10 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) HMAP_FOR_EACH (br, node, &all_bridges) { struct port *port; - br->has_bonded_ports = false; HMAP_FOR_EACH (port, hmap_node, &br->ports) { struct iface *iface; - port_reconfigure(port); - port_reconfigure_lacp(port); - port_reconfigure_bond(port); + port_configure(port); HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) { iface_configure_cfm(iface); @@ -501,10 +435,9 @@ bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) iface_set_mac(iface); } } - mirror_reconfigure(br); - + bridge_configure_mirrors(br); bridge_configure_datapath_id(br); - bridge_reconfigure_remotes(br, managers, n_managers); + bridge_configure_remotes(br, managers, n_managers); bridge_configure_netflow(br); bridge_configure_sflow(br, &sflow_bridge_number); } @@ -550,18 +483,87 @@ bridge_del_dps(void) static bool bridge_add_dp(struct bridge *br) { - int error; - - error = ofproto_create(br->name, br->type, &bridge_ofhooks, br, - &br->ofproto); + int error = ofproto_create(br->name, br->type, &br->ofproto); if (error) { - VLOG_ERR("failed to create bridge %s: %s", - br->name, strerror(error)); + VLOG_ERR("failed to create bridge %s: %s", br->name, strerror(error)); return false; } return true; } +static void +port_configure(struct port *port) +{ + const struct ovsrec_port *cfg = port->cfg; + struct bond_settings bond_settings; + struct lacp_settings lacp_settings; + struct ofproto_bundle_settings s; + struct iface *iface; + + /* Get name. */ + s.name = port->name; + + /* Get slaves. */ + s.n_slaves = 0; + s.slaves = xmalloc(list_size(&port->ifaces) * sizeof *s.slaves); + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + s.slaves[s.n_slaves++] = iface->dp_ifidx; + } + + /* Get VLAN tag. */ + s.vlan = -1; + if (cfg->tag) { + if (list_is_short(&port->ifaces)) { + if (*cfg->tag >= 0 && *cfg->tag <= 4095) { + s.vlan = *cfg->tag; + VLOG_DBG("port %s: assigning VLAN tag %d", port->name, s.vlan); + } + } else { + /* It's possible that bonded, VLAN-tagged ports make sense. Maybe + * they even work as-is. But they have not been tested. */ + VLOG_WARN("port %s: VLAN tags not supported on bonded ports", + port->name); + } + } + + /* Get VLAN trunks. */ + s.trunks = NULL; + if (s.vlan < 0 && cfg->n_trunks) { + s.trunks = vlan_bitmap_from_array(cfg->trunks, cfg->n_trunks); + } else if (s.vlan >= 0 && cfg->n_trunks) { + VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan", + port->name); + } + + /* Get LACP settings. */ + s.lacp = port_configure_lacp(port, &lacp_settings); + if (s.lacp) { + size_t i = 0; + + s.lacp_slaves = xmalloc(s.n_slaves * sizeof *s.lacp_slaves); + LIST_FOR_EACH (iface, port_elem, &port->ifaces) { + iface_configure_lacp(iface, &s.lacp_slaves[i++]); + } + } else { + s.lacp_slaves = NULL; + } + + /* Get bond settings. */ + if (s.n_slaves > 1) { + port_configure_bond(port, &bond_settings); + s.bond = &bond_settings; + } else { + s.bond = NULL; + } + + /* Register. */ + ofproto_bundle_register(port->bridge->ofproto, port, &s); + + /* Clean up. */ + free(s.trunks); + free(s.lacp_slaves); +} + /* Pick local port hardware address and datapath ID for 'br'. */ static void bridge_configure_datapath_id(struct bridge *br) @@ -800,13 +802,7 @@ bridge_del_ofproto_ports(struct bridge *br) br->name, name, strerror(error)); } if (iface) { - if (iface->port->bond) { - /* The bond has a pointer to the netdev, so remove it from - * the bond before closing the netdev. The slave will get - * added back to the bond later, after a new netdev is - * available. */ - bond_slave_unregister(iface->port->bond, iface); - } + ofproto_port_unregister(br->ofproto, ofproto_port.ofp_port); netdev_close(iface->netdev); iface->netdev = NULL; } @@ -1016,7 +1012,7 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], struct iface *iface; /* Mirror output ports don't participate. */ - if (port->is_mirror_output_port) { + if (ofproto_is_mirror_output_bundle(br->ofproto, port)) { continue; } @@ -1283,22 +1279,16 @@ iface_refresh_cfm_stats(struct iface *iface) static bool iface_refresh_lacp_stats(struct iface *iface) { - bool *db_current = iface->cfg->lacp_current; - bool changed = false; - - if (iface->port->lacp) { - bool current = lacp_slave_is_current(iface->port->lacp, iface); + struct ofproto *ofproto = iface->port->bridge->ofproto; + uint16_t ofp_port = odp_port_to_ofp_port(iface->dp_ifidx); + int old = iface->cfg->lacp_current ? *iface->cfg->lacp_current : -1; + int new = ofproto_port_is_lacp_current(ofproto, ofp_port); - if (!db_current || *db_current != current) { - changed = true; - ovsrec_interface_set_lacp_current(iface->cfg, ¤t, 1); - } - } else if (db_current) { - changed = true; - ovsrec_interface_set_lacp_current(iface->cfg, NULL, 0); + if (old != new) { + bool current = new; + ovsrec_interface_set_lacp_current(iface->cfg, ¤t, new >= 0); } - - return changed; + return old != new; } static void @@ -1421,7 +1411,7 @@ bridge_run(void) /* Let each bridge do the work that it needs to do. */ datapath_destroyed = false; HMAP_FOR_EACH (br, node, &all_bridges) { - int error = bridge_run_one(br); + int error = ofproto_run(br->ofproto); if (error) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, " @@ -1525,13 +1515,7 @@ bridge_wait(void) struct bridge *br; HMAP_FOR_EACH (br, node, &all_bridges) { - struct port *port; - ofproto_wait(br->ofproto); - mac_learning_wait(br->ml); - HMAP_FOR_EACH (port, hmap_node, &br->ports) { - port_wait(port); - } } ovsdb_idl_wait(idl); poll_timer_wait_until(stats_timer); @@ -1540,41 +1524,6 @@ bridge_wait(void) poll_timer_wait_until(db_limiter); } } - -/* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s - * configuration changes. */ -static void -bridge_flush(struct bridge *br) -{ - COVERAGE_INC(bridge_flush); - br->flush = true; -} - -/* Bridge unixctl user interface functions. */ -static void -bridge_unixctl_fdb_show(struct unixctl_conn *conn, - const char *args, void *aux OVS_UNUSED) -{ - struct ds ds = DS_EMPTY_INITIALIZER; - const struct bridge *br; - const struct mac_entry *e; - - br = bridge_lookup(args); - if (!br) { - unixctl_command_reply(conn, 501, "no such bridge"); - return; - } - - ds_put_cstr(&ds, " port VLAN MAC Age\n"); - LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { - struct port *port = e->port.p; - ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", - port_get_an_iface(port)->dp_ifidx, - e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); - } - unixctl_command_reply(conn, 200, ds_cstr(&ds)); - ds_destroy(&ds); -} /* CFM unixctl user interface functions. */ static void @@ -1708,14 +1657,12 @@ bridge_create(const struct ovsrec_bridge *br_cfg) br->name = xstrdup(br_cfg->name); br->type = xstrdup(dpif_normalize_type(br_cfg->datapath_type)); br->cfg = br_cfg; - br->ml = mac_learning_create(); eth_addr_nicira_random(br->default_ea); hmap_init(&br->ports); hmap_init(&br->ifaces); hmap_init(&br->iface_by_name); - - br->flush = false; + hmap_init(&br->mirrors); hmap_insert(&all_bridges, &br->node, hash_string(br->name, 0)); } @@ -1724,21 +1671,21 @@ static void bridge_destroy(struct bridge *br) { if (br) { - struct port *port, *next; - int i; + struct mirror *mirror, *next_mirror; + struct port *port, *next_port; - HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) { + HMAP_FOR_EACH_SAFE (port, next_port, hmap_node, &br->ports) { port_destroy(port); } - for (i = 0; i < MAX_MIRRORS; i++) { - mirror_destroy(br->mirrors[i]); + HMAP_FOR_EACH_SAFE (mirror, next_mirror, hmap_node, &br->mirrors) { + mirror_destroy(mirror); } hmap_remove(&all_bridges, &br->node); ofproto_destroy(br->ofproto); - mac_learning_destroy(br->ml); hmap_destroy(&br->ifaces); hmap_destroy(&br->ports); hmap_destroy(&br->iface_by_name); + hmap_destroy(&br->mirrors); free(br->name); free(br->type); free(br); @@ -1803,29 +1750,6 @@ bridge_unixctl_reconnect(struct unixctl_conn *conn, unixctl_command_reply(conn, 200, NULL); } -static int -bridge_run_one(struct bridge *br) -{ - struct port *port; - int error; - - error = ofproto_run1(br->ofproto); - if (error) { - return error; - } - - mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto)); - - HMAP_FOR_EACH (port, hmap_node, &br->ports) { - port_run(port); - } - - error = ofproto_run2(br->ofproto, br->flush); - br->flush = false; - - return error; -} - static size_t bridge_get_controllers(const struct bridge *br, struct ovsrec_controller ***controllersp) @@ -1989,9 +1913,8 @@ bridge_configure_local_iface_netdev(struct bridge *br, } static void -bridge_reconfigure_remotes(struct bridge *br, - const struct sockaddr_in *managers, - size_t n_managers) +bridge_configure_remotes(struct bridge *br, + const struct sockaddr_in *managers, size_t n_managers) { const char *disable_ib_str, *queue_id_str; bool disable_in_band = false; @@ -2075,911 +1998,150 @@ bridge_reconfigure_remotes(struct bridge *br, } } -/* Bridge packet processing functions. */ +/* Port functions. */ -static bool -set_dst(struct dst *dst, const struct flow *flow, - const struct port *in_port, const struct port *out_port, - tag_type *tags) +static struct port * +port_create(struct bridge *br, const struct ovsrec_port *cfg) { - dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE - : in_port->vlan >= 0 ? in_port->vlan - : flow->vlan_tci == 0 ? OFP_VLAN_NONE - : vlan_tci_to_vid(flow->vlan_tci)); + struct port *port; - dst->iface = (!out_port->bond - ? port_get_an_iface(out_port) - : bond_choose_output_slave(out_port->bond, flow, - dst->vlan, tags)); + port = xzalloc(sizeof *port); + port->bridge = br; + port->name = xstrdup(cfg->name); + port->cfg = cfg; + list_init(&port->ifaces); - return dst->iface != NULL; -} + hmap_insert(&br->ports, &port->hmap_node, hash_string(port->name, 0)); -static int -mirror_mask_ffs(mirror_mask_t mask) -{ - BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); - return ffs(mask); -} + VLOG_INFO("created port %s on bridge %s", port->name, br->name); -static void -dst_set_init(struct dst_set *set) -{ - set->dsts = set->builtin; - set->n = 0; - set->allocated = ARRAY_SIZE(set->builtin); + return port; } -static void -dst_set_add(struct dst_set *set, const struct dst *dst) +static const char * +get_port_other_config(const struct ovsrec_port *port, const char *key, + const char *default_value) { - if (set->n >= set->allocated) { - size_t new_allocated; - struct dst *new_dsts; + const char *value; - new_allocated = set->allocated * 2; - new_dsts = xmalloc(new_allocated * sizeof *new_dsts); - memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts); + value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config, + key); + return value ? value : default_value; +} - dst_set_free(set); +static const char * +get_interface_other_config(const struct ovsrec_interface *iface, + const char *key, const char *default_value) +{ + const char *value; - set->dsts = new_dsts; - set->allocated = new_allocated; - } - set->dsts[set->n++] = *dst; + value = get_ovsrec_key_value(&iface->header_, + &ovsrec_interface_col_other_config, key); + return value ? value : default_value; } +/* Deletes interfaces from 'port' that are no longer configured for it. */ static void -dst_set_free(struct dst_set *set) +port_del_ifaces(struct port *port) { - if (set->dsts != set->builtin) { - free(set->dsts); + struct iface *iface, *next; + struct sset new_ifaces; + size_t i; + + /* Collect list of new interfaces. */ + sset_init(&new_ifaces); + for (i = 0; i < port->cfg->n_interfaces; i++) { + const char *name = port->cfg->interfaces[i]->name; + sset_add(&new_ifaces, name); } -} -static bool -dst_is_duplicate(const struct dst_set *set, const struct dst *test) -{ - size_t i; - for (i = 0; i < set->n; i++) { - if (set->dsts[i].vlan == test->vlan - && set->dsts[i].iface == test->iface) { - return true; + /* Get rid of deleted interfaces. */ + LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) { + if (!sset_contains(&new_ifaces, iface->name)) { + iface_destroy(iface); } } - return false; -} - -static bool -port_trunks_vlan(const struct port *port, uint16_t vlan) -{ - return (port->vlan < 0 || vlan_bitmap_contains(port->trunks, vlan)); -} -static bool -port_includes_vlan(const struct port *port, uint16_t vlan) -{ - return vlan == port->vlan || port_trunks_vlan(port, vlan); + sset_destroy(&new_ifaces); } -static bool -port_is_floodable(const struct port *port) +/* Adds new interfaces to 'port' and updates 'type' and 'cfg' members of + * existing ones. */ +static void +port_add_ifaces(struct port *port) { - struct iface *iface; + struct shash new_ifaces; + struct shash_node *node; + size_t i; - LIST_FOR_EACH (iface, port_elem, &port->ifaces) { - if (!ofproto_port_is_floodable(port->bridge->ofproto, - iface->dp_ifidx)) { - return false; + /* Collect new ifaces. */ + shash_init(&new_ifaces); + for (i = 0; i < port->cfg->n_interfaces; i++) { + const struct ovsrec_interface *cfg = port->cfg->interfaces[i]; + if (!shash_add_once(&new_ifaces, cfg->name, cfg)) { + VLOG_WARN("port %s: %s specified twice as port interface", + port->name, cfg->name); + iface_set_ofport(cfg, -1); } } - return true; -} - -/* Returns an arbitrary interface within 'port'. */ -static struct iface * -port_get_an_iface(const struct port *port) -{ - return CONTAINER_OF(list_front(&port->ifaces), struct iface, port_elem); -} - -static void -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_set *set, tag_type *tags, uint16_t *nf_output_iface) -{ - struct dst dst; - if (out_port == FLOOD_PORT) { - struct port *port; + /* Create new interfaces. + * Update interface types and 'cfg' members. */ + SHASH_FOR_EACH (node, &new_ifaces) { + const struct ovsrec_interface *cfg = node->data; + const char *iface_name = node->name; + struct iface *iface; - HMAP_FOR_EACH (port, hmap_node, &br->ports) { - 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)) { - dst_set_add(set, &dst); - } + iface = iface_lookup(port->bridge, iface_name); + if (!iface) { + iface = iface_create(port, cfg); + } else { + iface->cfg = cfg; } - *nf_output_iface = NF_OUT_FLOOD; - } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) { - dst_set_add(set, &dst); - *nf_output_iface = dst.iface->dp_ifidx; + + /* 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(iface_name, port->bridge->name) ? "internal" + : cfg->type[0] ? cfg->type + : "system"); } + shash_destroy(&new_ifaces); } static void -compose_mirror_dsts(const struct bridge *br, const struct flow *flow, - uint16_t vlan, const struct port *in_port, - struct dst_set *set, tag_type *tags) +port_destroy(struct port *port) { - mirror_mask_t mirrors; - int flow_vlan; - size_t i; - - mirrors = in_port->src_mirrors; - for (i = 0; i < set->n; i++) { - mirrors |= set->dsts[i].iface->port->dst_mirrors; - } + if (port) { + struct bridge *br = port->bridge; + struct iface *iface, *next; - if (!mirrors) { - return; - } + if (br->ofproto) { + ofproto_bundle_unregister(br->ofproto, port); + } - flow_vlan = vlan_tci_to_vid(flow->vlan_tci); - if (flow_vlan == 0) { - flow_vlan = OFP_VLAN_NONE; - } + LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) { + iface_destroy(iface); + } - while (mirrors) { - struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1]; - if (!m->n_vlans || vlan_is_mirrored(m, vlan)) { - struct dst dst; + hmap_remove(&br->ports, &port->hmap_node); - if (m->out_port) { - if (set_dst(&dst, flow, in_port, m->out_port, tags) - && !dst_is_duplicate(set, &dst)) { - dst_set_add(set, &dst); - } - } else { - struct port *port; + VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name); - HMAP_FOR_EACH (port, hmap_node, &br->ports) { - if (port_includes_vlan(port, m->out_vlan) - && set_dst(&dst, flow, in_port, port, tags)) - { - if (port->vlan < 0) { - dst.vlan = m->out_vlan; - } - if (dst_is_duplicate(set, &dst)) { - continue; - } - - /* Use the vlan tag on the original flow instead of - * the one passed in the vlan parameter. This ensures - * that we compare the vlan from before any implicit - * tagging tags place. This is necessary because - * dst->vlan is the final vlan, after removing implicit - * tags. */ - if (port == in_port && dst.vlan == flow_vlan) { - /* Don't send out input port on same VLAN. */ - continue; - } - dst_set_add(set, &dst); - } - } - } - } - mirrors &= mirrors - 1; + free(port->name); + free(port); } } -static void -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 ofpbuf *actions, - uint16_t *nf_output_iface) -{ - uint16_t initial_vlan, cur_vlan; - const struct dst *dst; - struct dst_set set; - - dst_set_init(&set); - compose_dsts(br, flow, vlan, in_port, out_port, &set, tags, - nf_output_iface); - compose_mirror_dsts(br, flow, vlan, in_port, &set, tags); - - /* Output all the packets we can without having to change the VLAN. */ - initial_vlan = vlan_tci_to_vid(flow->vlan_tci); - if (initial_vlan == 0) { - initial_vlan = OFP_VLAN_NONE; - } - for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) { - if (dst->vlan != initial_vlan) { - continue; - } - nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->iface->dp_ifidx); - } - - /* Then output the rest. */ - cur_vlan = initial_vlan; - for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) { - if (dst->vlan == initial_vlan) { - continue; - } - if (dst->vlan != cur_vlan) { - if (dst->vlan == OFP_VLAN_NONE) { - nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN); - } else { - ovs_be16 tci; - tci = htons(dst->vlan & VLAN_VID_MASK); - tci |= flow->vlan_tci & htons(VLAN_PCP_MASK); - nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci); - } - cur_vlan = dst->vlan; - } - nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->iface->dp_ifidx); - } - - dst_set_free(&set); -} - -/* Returns the effective vlan of a packet, taking into account both the - * 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 struct flow *flow, - struct port *in_port, bool have_packet) -{ - int vlan = vlan_tci_to_vid(flow->vlan_tci); - if (in_port->vlan >= 0) { - if (vlan) { - if (have_packet) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " - "packet received on port %s configured with " - "implicit VLAN %"PRIu16, - br->name, vlan, in_port->name, in_port->vlan); - } - return -1; - } - vlan = in_port->vlan; - } else { - if (!port_includes_vlan(in_port, vlan)) { - if (have_packet) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " - "packet received on port %s not configured for " - "trunking VLAN %d", - br->name, vlan, in_port->name, vlan); - } - return -1; - } - } - - 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 struct flow *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 struct flow *flow, int vlan, - struct port *in_port) -{ - struct mac_entry *mac; - - if (!mac_learning_may_learn(br->ml, flow->dl_src, vlan)) { - return; - } - - mac = mac_learning_insert(br->ml, flow->dl_src, vlan); - if (is_gratuitous_arp(flow)) { - /* We don't want to learn from gratuitous ARP packets that are - * reflected back over bond slaves so we lock the learning table. */ - if (!in_port->bond) { - mac_entry_set_grat_arp_lock(mac); - } else if (mac_entry_is_grat_arp_locked(mac)) { - return; - } - } - - if (mac_entry_is_new(mac) || mac->port.p != in_port) { - /* The log messages here could actually be useful in debugging, - * so keep the rate limit relatively high. */ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); - VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " - "on port %s in VLAN %d", - br->name, ETH_ADDR_ARGS(flow->dl_src), - in_port->name, vlan); - - mac->port.p = in_port; - ofproto_revalidate(br->ofproto, mac_learning_changed(br->ml, mac)); - } -} - -/* 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_admissible(struct bridge *br, const struct flow *flow, bool have_packet, - tag_type *tags, int *vlanp, struct port **in_portp) -{ - struct iface *in_iface; - struct port *in_port; - int vlan; - - /* 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 (have_packet) { - /* Odd. A few possible reasons here: - * - * - We deleted an interface but there are still a few packets - * queued up from it. - * - * - Someone externally added an interface (e.g. with "ovs-dpctl - * add-if") that we don't know about. - * - * - Packet arrived on the local port but the local port is not - * one of our bridge ports. - */ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - - VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " - "interface %"PRIu16, br->name, flow->in_port); - } - - *in_portp = NULL; - return false; - } - *in_portp = in_port = in_iface->port; - *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet); - if (vlan < 0) { - return false; - } - - /* Drop frames for reserved multicast addresses. */ - if (eth_addr_is_reserved(flow->dl_dst)) { - return false; - } - - /* Drop frames on ports reserved for mirroring. */ - if (in_port->is_mirror_output_port) { - 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; - } - - if (in_port->bond) { - struct mac_entry *mac; - - switch (bond_check_admissibility(in_port->bond, in_iface, - flow->dl_dst, tags)) { - case BV_ACCEPT: - break; - - case BV_DROP: - return false; - - case BV_DROP_IF_MOVED: - mac = mac_learning_lookup(br->ml, flow->dl_src, vlan, NULL); - if (mac && mac->port.p != in_port && - (!is_gratuitous_arp(flow) - || mac_entry_is_grat_arp_locked(mac))) { - return false; - } - break; - } - } - - 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 struct flow *flow, - const struct ofpbuf *packet, struct ofpbuf *actions, - tag_type *tags, uint16_t *nf_output_iface) -{ - struct port *in_port; - struct port *out_port; - struct mac_entry *mac; - int vlan; - - /* 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. */ - mac = mac_learning_lookup(br->ml, flow->dl_dst, vlan, tags); - if (mac) { - out_port = mac->port.p; - } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) { - /* If we are revalidating but don't have a learning entry then - * eject the flow. Installing a flow that floods packets opens - * up a window of time where we could learn from a packet reflected - * 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. */ - if (in_port == out_port) { - out_port = NULL; - } - -done: - if (in_port) { - compose_actions(br, flow, vlan, in_port, out_port, tags, actions, - nf_output_iface); - } - - return true; -} - -static bool -bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, - struct ofpbuf *actions, tag_type *tags, - uint16_t *nf_output_iface, void *br_) -{ - struct bridge *br = br_; - - COVERAGE_INC(bridge_process_flow); - return process_flow(br, flow, packet, actions, tags, nf_output_iface); -} - -static bool -bridge_special_ofhook_cb(const struct flow *flow, - const struct ofpbuf *packet, void *br_) -{ - struct iface *iface; - struct bridge *br = br_; - - iface = iface_from_dp_ifidx(br, flow->in_port); - - if (flow->dl_type == htons(ETH_TYPE_LACP)) { - if (iface && iface->port->lacp && packet) { - const struct lacp_pdu *pdu = parse_lacp_packet(packet); - if (pdu) { - lacp_process_pdu(iface->port->lacp, iface, pdu); - } - } - return false; - } - - return true; -} - -static void -bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags, - const struct nlattr *actions, - size_t actions_len, - uint64_t n_bytes, void *br_) -{ - struct bridge *br = br_; - const struct nlattr *a; - struct port *in_port; - tag_type dummy = 0; - unsigned int left; - 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. - * - * 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; - } - NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) { - if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) { - struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a)); - if (out_port && out_port->bond) { - uint16_t vlan = (flow->vlan_tci - ? vlan_tci_to_vid(flow->vlan_tci) - : OFP_VLAN_NONE); - bond_account(out_port->bond, flow, vlan, n_bytes); - } - } - } -} - -static void -bridge_account_checkpoint_ofhook_cb(void *br_) -{ - struct bridge *br = br_; - struct port *port; - - HMAP_FOR_EACH (port, hmap_node, &br->ports) { - if (port->bond) { - bond_rebalance(port->bond, - ofproto_get_revalidate_set(br->ofproto)); - } - } -} - -static uint16_t -bridge_autopath_ofhook_cb(const struct flow *flow, uint32_t ofp_port, - tag_type *tags, void *br_) -{ - struct bridge *br = br_; - uint16_t odp_port = ofp_port_to_odp_port(ofp_port); - struct port *port = port_from_dp_ifidx(br, odp_port); - uint16_t ret; - - if (!port) { - ret = ODPP_NONE; - } else if (list_is_short(&port->ifaces)) { - ret = odp_port; - } else { - struct iface *iface; - - /* Autopath does not support VLAN hashing. */ - iface = bond_choose_output_slave(port->bond, flow, - OFP_VLAN_NONE, tags); - ret = iface ? iface->dp_ifidx : ODPP_NONE; - } - - return odp_port_to_ofp_port(ret); -} - -static struct ofhooks bridge_ofhooks = { - bridge_normal_ofhook_cb, - bridge_special_ofhook_cb, - bridge_account_flow_ofhook_cb, - bridge_account_checkpoint_ofhook_cb, - bridge_autopath_ofhook_cb, -}; - -/* Port functions. */ - -static void -lacp_send_pdu_cb(void *iface_, const struct lacp_pdu *pdu) -{ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); - struct iface *iface = iface_; - uint8_t ea[ETH_ADDR_LEN]; - int error; - - error = netdev_get_etheraddr(iface->netdev, ea); - if (!error) { - struct lacp_pdu *packet_pdu; - struct ofpbuf packet; - - ofpbuf_init(&packet, 0); - packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, - sizeof *packet_pdu); - *packet_pdu = *pdu; - error = netdev_send(iface->netdev, &packet); - if (error) { - VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed " - "(%s)", iface->port->name, iface->name, - strerror(error)); - } - ofpbuf_uninit(&packet); - } else { - VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " - "%s (%s)", iface->port->name, iface->name, - strerror(error)); - } -} - -static void -port_run(struct port *port) -{ - if (port->lacp) { - lacp_run(port->lacp, lacp_send_pdu_cb); - } - - if (port->bond) { - struct iface *iface; - - LIST_FOR_EACH (iface, port_elem, &port->ifaces) { - bool may_enable = lacp_slave_may_enable(port->lacp, iface); - bond_slave_set_lacp_may_enable(port->bond, iface, may_enable); - } - - bond_run(port->bond, - ofproto_get_revalidate_set(port->bridge->ofproto), - lacp_negotiated(port->lacp)); - if (bond_should_send_learning_packets(port->bond)) { - port_send_learning_packets(port); - } - } -} - -static void -port_wait(struct port *port) -{ - if (port->lacp) { - lacp_wait(port->lacp); - } - - if (port->bond) { - bond_wait(port->bond); - } -} - -static struct port * -port_create(struct bridge *br, const struct ovsrec_port *cfg) -{ - struct port *port; - - port = xzalloc(sizeof *port); - port->bridge = br; - port->name = xstrdup(cfg->name); - port->cfg = cfg; - list_init(&port->ifaces); - - hmap_insert(&br->ports, &port->hmap_node, hash_string(port->name, 0)); - - VLOG_INFO("created port %s on bridge %s", port->name, br->name); - - 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 const char * -get_interface_other_config(const struct ovsrec_interface *iface, - const char *key, const char *default_value) -{ - const char *value; - - value = get_ovsrec_key_value(&iface->header_, - &ovsrec_interface_col_other_config, key); - return value ? value : default_value; -} - -/* Deletes interfaces from 'port' that are no longer configured for it. */ -static void -port_del_ifaces(struct port *port) -{ - struct iface *iface, *next; - struct sset new_ifaces; - size_t i; - - /* Collect list of new interfaces. */ - sset_init(&new_ifaces); - for (i = 0; i < port->cfg->n_interfaces; i++) { - const char *name = port->cfg->interfaces[i]->name; - sset_add(&new_ifaces, name); - } - - /* Get rid of deleted interfaces. */ - LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) { - if (!sset_contains(&new_ifaces, iface->name)) { - iface_destroy(iface); - } - } - - sset_destroy(&new_ifaces); -} - -/* Adds new interfaces to 'port' and updates 'type' and 'cfg' members of - * existing ones. */ -static void -port_add_ifaces(struct port *port) -{ - struct shash new_ifaces; - struct shash_node *node; - size_t i; - - /* Collect new ifaces. */ - shash_init(&new_ifaces); - for (i = 0; i < port->cfg->n_interfaces; i++) { - const struct ovsrec_interface *cfg = port->cfg->interfaces[i]; - if (!shash_add_once(&new_ifaces, cfg->name, cfg)) { - VLOG_WARN("port %s: %s specified twice as port interface", - port->name, cfg->name); - iface_set_ofport(cfg, -1); - } - } - - /* Create new interfaces. - * Update interface types and 'cfg' members. */ - SHASH_FOR_EACH (node, &new_ifaces) { - const struct ovsrec_interface *cfg = node->data; - const char *iface_name = node->name; - struct iface *iface; - - iface = iface_lookup(port->bridge, iface_name); - if (!iface) { - iface = iface_create(port, cfg); - } else { - iface->cfg = 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(iface_name, port->bridge->name) ? "internal" - : cfg->type[0] ? cfg->type - : "system"); - } - shash_destroy(&new_ifaces); -} - -/* Expires all MAC learning entries associated with 'port' and forces ofproto - * to revalidate every flow. */ -static void -port_flush_macs(struct port *port) -{ - struct bridge *br = port->bridge; - struct mac_learning *ml = br->ml; - struct mac_entry *mac, *next_mac; - - bridge_flush(br); - LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { - if (mac->port.p == port) { - mac_learning_expire(ml, mac); - } - } -} - -static void -port_reconfigure(struct port *port) -{ - const struct ovsrec_port *cfg = port->cfg; - bool need_flush = false; - unsigned long *trunks; - int vlan; - - /* Get VLAN tag. */ - vlan = -1; - if (cfg->tag) { - if (list_is_short(&port->ifaces)) { - vlan = *cfg->tag; - if (vlan >= 0 && vlan <= 4095) { - VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan); - } else { - vlan = -1; - } - } else { - /* It's possible that bonded, VLAN-tagged ports make sense. Maybe - * they even work as-is. But they have not been tested. */ - VLOG_WARN("port %s: VLAN tags not supported on bonded ports", - port->name); - } - } - if (port->vlan != vlan) { - port->vlan = vlan; - need_flush = true; - } - - /* Get trunked VLANs. */ - trunks = NULL; - if (vlan < 0 && cfg->n_trunks) { - trunks = vlan_bitmap_from_array(cfg->trunks, cfg->n_trunks); - if (!trunks) { - VLOG_ERR("port %s: no valid trunks, trunking all VLANs", - port->name); - } - } else if (vlan >= 0 && cfg->n_trunks) { - VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan", - port->name); - } - if (!vlan_bitmap_equal(trunks, port->trunks)) { - need_flush = true; - } - bitmap_free(port->trunks); - port->trunks = trunks; - - if (need_flush) { - port_flush_macs(port); - } -} - -static void -port_destroy(struct port *port) -{ - if (port) { - struct bridge *br = port->bridge; - struct iface *iface, *next; - int i; - - for (i = 0; i < MAX_MIRRORS; i++) { - struct mirror *m = br->mirrors[i]; - if (m && m->out_port == port) { - mirror_destroy(m); - } - } - - LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) { - iface_destroy(iface); - } - - hmap_remove(&br->ports, &port->hmap_node); - - VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name); - - bond_destroy(port->bond); - lacp_destroy(port->lacp); - port_flush_macs(port); - - bitmap_free(port->trunks); - free(port->name); - free(port); - } -} - -static struct port * -port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) -{ - struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx); - return iface ? iface->port : NULL; -} - -static struct port * -port_lookup(const struct bridge *br, const char *name) -{ - struct port *port; - - HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_string(name, 0), - &br->ports) { - if (!strcmp(port->name, name)) { - return port; +static struct port * +port_lookup(const struct bridge *br, const char *name) +{ + struct port *port; + + HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_string(name, 0), + &br->ports) { + if (!strcmp(port->name, name)) { + return port; } } return NULL; @@ -3007,185 +2169,107 @@ enable_lacp(struct port *port, bool *activep) } } -static void -iface_reconfigure_lacp(struct iface *iface) -{ - struct lacp_slave_settings s; - int priority, portid; - - portid = atoi(get_interface_other_config(iface->cfg, "lacp-port-id", "0")); - priority = atoi(get_interface_other_config(iface->cfg, - "lacp-port-priority", "0")); - - if (portid <= 0 || portid > UINT16_MAX) { - portid = iface->dp_ifidx; - } - - if (priority <= 0 || priority > UINT16_MAX) { - priority = UINT16_MAX; - } - - s.name = iface->name; - s.id = portid; - s.priority = priority; - lacp_slave_register(iface->port->lacp, iface, &s); -} - -static void -port_reconfigure_lacp(struct port *port) +static struct lacp_settings * +port_configure_lacp(struct port *port, struct lacp_settings *s) { - static struct lacp_settings s; - struct iface *iface; - uint8_t sysid[ETH_ADDR_LEN]; - const char *sysid_str; const char *lacp_time; long long int custom_time; int priority; - if (!enable_lacp(port, &s.active)) { - lacp_destroy(port->lacp); - port->lacp = NULL; - return; - } - - sysid_str = get_port_other_config(port->cfg, "lacp-system-id", NULL); - if (sysid_str && eth_addr_from_string(sysid_str, sysid)) { - memcpy(s.id, sysid, ETH_ADDR_LEN); - } else { - memcpy(s.id, port->bridge->ea, ETH_ADDR_LEN); + if (!enable_lacp(port, &s->active)) { + return NULL; } - s.name = port->name; + s->name = port->name; + memcpy(s->id, port->bridge->ea, ETH_ADDR_LEN); /* Prefer bondable links if unspecified. */ priority = atoi(get_port_other_config(port->cfg, "lacp-system-priority", "0")); - s.priority = (priority > 0 && priority <= UINT16_MAX - ? priority - : UINT16_MAX - !list_is_short(&port->ifaces)); + s->priority = (priority > 0 && priority <= UINT16_MAX + ? priority + : UINT16_MAX - !list_is_short(&port->ifaces)); - s.strict = !strcmp(get_port_other_config(port->cfg, "lacp-strict", - "false"), - "true"); + s->strict = !strcmp(get_port_other_config(port->cfg, "lacp-strict", + "false"), + "true"); lacp_time = get_port_other_config(port->cfg, "lacp-time", "slow"); custom_time = atoi(lacp_time); if (!strcmp(lacp_time, "fast")) { - s.lacp_time = LACP_TIME_FAST; + s->lacp_time = LACP_TIME_FAST; } else if (!strcmp(lacp_time, "slow")) { - s.lacp_time = LACP_TIME_SLOW; + s->lacp_time = LACP_TIME_SLOW; } else if (custom_time > 0) { - s.lacp_time = LACP_TIME_CUSTOM; - s.custom_time = custom_time; + s->lacp_time = LACP_TIME_CUSTOM; + s->custom_time = custom_time; } else { - s.lacp_time = LACP_TIME_SLOW; + s->lacp_time = LACP_TIME_SLOW; } - if (!port->lacp) { - port->lacp = lacp_create(); - } + return s; +} - lacp_configure(port->lacp, &s); +static void +iface_configure_lacp(struct iface *iface, struct lacp_slave_settings *s) +{ + int priority, portid; - LIST_FOR_EACH (iface, port_elem, &port->ifaces) { - iface_reconfigure_lacp(iface); + portid = atoi(get_interface_other_config(iface->cfg, "lacp-port-id", "0")); + priority = atoi(get_interface_other_config(iface->cfg, + "lacp-port-priority", "0")); + + if (portid <= 0 || portid > UINT16_MAX) { + portid = iface->dp_ifidx; } + + if (priority <= 0 || priority > UINT16_MAX) { + priority = UINT16_MAX; + } + + s->name = iface->name; + s->id = iface->dp_ifidx; + s->id = portid; + s->priority = priority; } static void -port_reconfigure_bond(struct port *port) +port_configure_bond(struct port *port, struct bond_settings *s) { - struct bond_settings s; const char *detect_s; - struct iface *iface; - if (list_is_short(&port->ifaces)) { - /* Not a bonded port. */ - bond_destroy(port->bond); - port->bond = NULL; - return; - } - - port->bridge->has_bonded_ports = true; - - s.name = port->name; - s.balance = BM_SLB; + s->name = port->name; + s->balance = BM_SLB; if (port->cfg->bond_mode - && !bond_mode_from_string(&s.balance, port->cfg->bond_mode)) { + && !bond_mode_from_string(&s->balance, port->cfg->bond_mode)) { VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s", port->name, port->cfg->bond_mode, - bond_mode_to_string(s.balance)); + bond_mode_to_string(s->balance)); } - s.detect = BLSM_CARRIER; + s->detect = BLSM_CARRIER; detect_s = get_port_other_config(port->cfg, "bond-detect-mode", NULL); - if (detect_s && !bond_detect_mode_from_string(&s.detect, detect_s)) { + if (detect_s && !bond_detect_mode_from_string(&s->detect, detect_s)) { VLOG_WARN("port %s: unsupported bond-detect-mode %s, " "defaulting to %s", - port->name, detect_s, bond_detect_mode_to_string(s.detect)); + port->name, detect_s, bond_detect_mode_to_string(s->detect)); } - s.miimon_interval = atoi( + s->miimon_interval = atoi( get_port_other_config(port->cfg, "bond-miimon-interval", "200")); - if (s.miimon_interval < 100) { - s.miimon_interval = 100; + if (s->miimon_interval < 100) { + s->miimon_interval = 100; } - s.up_delay = MAX(0, port->cfg->bond_updelay); - s.down_delay = MAX(0, port->cfg->bond_downdelay); - s.rebalance_interval = atoi( + s->up_delay = MAX(0, port->cfg->bond_updelay); + s->down_delay = MAX(0, port->cfg->bond_downdelay); + s->rebalance_interval = atoi( get_port_other_config(port->cfg, "bond-rebalance-interval", "10000")); - if (s.rebalance_interval < 1000) { - s.rebalance_interval = 1000; - } - - s.fake_iface = port->cfg->bond_fake_iface; - - if (!port->bond) { - port->bond = bond_create(&s); - } else { - if (bond_reconfigure(port->bond, &s)) { - bridge_flush(port->bridge); - } + if (s->rebalance_interval < 1000) { + s->rebalance_interval = 1000; } - LIST_FOR_EACH (iface, port_elem, &port->ifaces) { - uint16_t stable_id = (port->lacp - ? lacp_slave_get_port_id(port->lacp, iface) - : iface->dp_ifidx); - bond_slave_register(iface->port->bond, iface, stable_id, - iface->netdev); - } -} - -static void -port_send_learning_packets(struct port *port) -{ - struct bridge *br = port->bridge; - int error, n_packets, n_errors; - struct mac_entry *e; - - error = n_packets = n_errors = 0; - LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { - if (e->port.p != port) { - int ret = bond_send_learning_packet(port->bond, e->mac, e->vlan); - if (ret) { - error = ret; - n_errors++; - } - n_packets++; - } - } - - if (n_errors) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " - "packets, last error was: %s", - port->name, n_errors, n_packets, strerror(error)); - } else { - VLOG_DBG("bond %s: sent %d gratuitous learning packets", - port->name, n_packets); - } + s->fake_iface = port->cfg->bond_fake_iface; } /* Interface functions. */ @@ -3211,8 +2295,6 @@ iface_create(struct port *port, const struct ovsrec_interface *if_cfg) VLOG_DBG("attached network device %s to port %s", iface->name, port->name); - bridge_flush(br); - return iface; } @@ -3223,12 +2305,8 @@ iface_destroy(struct iface *iface) struct port *port = iface->port; struct bridge *br = port->bridge; - if (port->bond) { - bond_slave_unregister(port->bond, iface); - } - - if (port->lacp) { - lacp_slave_unregister(port->lacp, iface); + if (br->ofproto && iface->dp_ifidx >= 0) { + ofproto_port_unregister(br->ofproto, iface->dp_ifidx); } if (iface->dp_ifidx >= 0) { @@ -3242,8 +2320,6 @@ iface_destroy(struct iface *iface) free(iface->name); free(iface); - - bridge_flush(port->bridge); } } @@ -3433,6 +2509,10 @@ iface_configure_qos(struct iface *iface, const struct ovsrec_qos *qos) shash_destroy(&details); } } + + netdev_set_policing(iface->netdev, + iface->cfg->ingress_policing_rate, + iface->cfg->ingress_policing_burst); } static void @@ -3495,11 +2575,10 @@ iface_is_synthetic(const struct iface *iface) static struct mirror * mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid) { - int i; + struct mirror *m; - for (i = 0; i < MAX_MIRRORS; i++) { - struct mirror *m = br->mirrors[i]; - if (m && uuid_equals(uuid, &m->uuid)) { + HMAP_FOR_EACH_IN_BUCKET (m, hmap_node, uuid_hash(uuid), &br->mirrors) { + if (uuid_equals(uuid, &m->uuid)) { return m; } } @@ -3507,96 +2586,55 @@ mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid) } static void -mirror_reconfigure(struct bridge *br) +bridge_configure_mirrors(struct bridge *br) { - unsigned long *rspan_vlans; - struct port *port; - int i; + const struct ovsdb_datum *mc; + unsigned long *flood_vlans; + struct mirror *m, *next; + size_t i; /* Get rid of deleted mirrors. */ - for (i = 0; i < MAX_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); - } + mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID); + HMAP_FOR_EACH_SAFE (m, next, hmap_node, &br->mirrors) { + union ovsdb_atom atom; + + atom.uuid = m->uuid; + if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) { + mirror_destroy(m); } } /* Add new mirrors and reconfigure existing ones. */ for (i = 0; i < br->cfg->n_mirrors; i++) { - struct ovsrec_mirror *cfg = br->cfg->mirrors[i]; + const struct ovsrec_mirror *cfg = br->cfg->mirrors[i]; struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid); - if (m) { - mirror_reconfigure_one(m, cfg); - } else { - mirror_create(br, cfg); + if (!m) { + m = mirror_create(br, cfg); } - } - - /* Update port reserved status. */ - HMAP_FOR_EACH (port, hmap_node, &br->ports) { - port->is_mirror_output_port = false; - } - for (i = 0; i < MAX_MIRRORS; i++) { - struct mirror *m = br->mirrors[i]; - if (m && m->out_port) { - m->out_port->is_mirror_output_port = true; + if (!mirror_configure(m, cfg)) { + mirror_destroy(m); } } /* Update flooded vlans (for RSPAN). */ - rspan_vlans = NULL; - if (br->cfg->n_flood_vlans) { - rspan_vlans = vlan_bitmap_from_array(br->cfg->flood_vlans, - br->cfg->n_flood_vlans); - } - if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) { - bridge_flush(br); - mac_learning_flush(br->ml); - } - free(rspan_vlans); + flood_vlans = vlan_bitmap_from_array(br->cfg->flood_vlans, + br->cfg->n_flood_vlans); + ofproto_set_flood_vlans(br->ofproto, flood_vlans); + bitmap_free(flood_vlans); } -static void -mirror_create(struct bridge *br, struct ovsrec_mirror *cfg) +static struct mirror * +mirror_create(struct bridge *br, const 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, cfg->name); - return; - } - if (!br->mirrors[i]) { - break; - } - } - - VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name); - bridge_flush(br); - mac_learning_flush(br->ml); - - br->mirrors[i] = m = xzalloc(sizeof *m); + m = xzalloc(sizeof *m); m->uuid = cfg->header_.uuid; + hmap_insert(&br->mirrors, &m->hmap_node, uuid_hash(&m->uuid)); m->bridge = br; - m->idx = i; m->name = xstrdup(cfg->name); - sset_init(&m->src_ports); - sset_init(&m->dst_ports); - m->vlans = NULL; - m->n_vlans = 0; - m->out_vlan = -1; - m->out_port = NULL; - mirror_reconfigure_one(m, cfg); + return m; } static void @@ -3604,104 +2642,63 @@ mirror_destroy(struct mirror *m) { if (m) { struct bridge *br = m->bridge; - struct port *port; - HMAP_FOR_EACH (port, hmap_node, &br->ports) { - port->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); - port->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); + if (br->ofproto) { + ofproto_mirror_unregister(br->ofproto, m); } - sset_destroy(&m->src_ports); - sset_destroy(&m->dst_ports); - free(m->vlans); - - m->bridge->mirrors[m->idx] = NULL; + hmap_remove(&br->mirrors, &m->hmap_node); free(m->name); free(m); - - bridge_flush(br); - mac_learning_flush(br->ml); } } static void -mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports, - struct sset *names) +mirror_collect_ports(struct mirror *m, + struct ovsrec_port **in_ports, int n_in_ports, + void ***out_portsp, size_t *n_out_portsp) { + void **out_ports = xmalloc(n_in_ports * sizeof *out_ports); + size_t n_out_ports = 0; size_t i; - for (i = 0; i < n_ports; i++) { - const char *name = ports[i]->name; - if (port_lookup(m->bridge, name)) { - sset_add(names, name); + for (i = 0; i < n_in_ports; i++) { + const char *name = in_ports[i]->name; + struct port *port = port_lookup(m->bridge, name); + if (port) { + out_ports[n_out_ports++] = port; } else { VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent " "port %s", m->bridge->name, m->name, name); } } -} - -static size_t -mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg, - int **vlans) -{ - size_t n_vlans; - size_t i; - - *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]; - if (vlan < 0 || vlan > 4095) { - VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64, - m->bridge->name, m->name, vlan); - } else { - (*vlans)[n_vlans++] = vlan; - } - } - return n_vlans; + *out_portsp = out_ports; + *n_out_portsp = n_out_ports; } static bool -vlan_is_mirrored(const struct mirror *m, int vlan) -{ - size_t i; - - for (i = 0; i < m->n_vlans; i++) { - if (m->vlans[i] == vlan) { - return true; - } - } - return false; -} - -static void -mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) +mirror_configure(struct mirror *m, const struct ovsrec_mirror *cfg) { - struct sset src_ports, dst_ports; - mirror_mask_t mirror_bit; + struct ofproto_mirror_settings s; struct port *out_port; struct port *port; - int out_vlan; - size_t n_vlans; - int *vlans; /* Set name. */ if (strcmp(cfg->name, m->name)) { free(m->name); m->name = xstrdup(cfg->name); } + s.name = m->name; - /* Get output port. */ + /* Get output port or VLAN. */ if (cfg->output_port) { - out_port = port_lookup(m->bridge, cfg->output_port->name); + s.out_bundle = port_lookup(m->bridge, cfg->output_port->name); if (!out_port) { VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge", m->bridge->name, m->name); - mirror_destroy(m); - return; + return false; } - out_vlan = -1; + s.out_vlan = UINT16_MAX; if (cfg->output_vlan) { VLOG_ERR("bridge %s: mirror %s specifies both output port and " @@ -3709,70 +2706,53 @@ mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) m->bridge->name, m->name); } } else if (cfg->output_vlan) { - out_port = NULL; - out_vlan = *cfg->output_vlan; + /* The database should prevent invalid VLAN values. */ + s.out_bundle = NULL; + s.out_vlan = *cfg->output_vlan; } else { VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring", m->bridge->name, m->name); - mirror_destroy(m); - return; + return false; } - sset_init(&src_ports); - sset_init(&dst_ports); + /* Get port selection. */ if (cfg->select_all) { + size_t n_ports = hmap_count(&m->bridge->ports); + void **ports = xmalloc(n_ports * sizeof *ports); + size_t i; + + i = 0; HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) { - sset_add(&src_ports, port->name); - sset_add(&dst_ports, port->name); + ports[i++] = port; } - vlans = NULL; - n_vlans = 0; + + s.srcs = ports; + s.n_srcs = n_ports; + + s.dsts = ports; + s.n_dsts = n_ports; } else { - /* Get ports, and drop duplicates and ports that don't exist. */ + /* Get ports, dropping ports that don't exist. + * The IDL ensures that there are no duplicates. */ mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port, - &src_ports); + &s.srcs, &s.n_srcs); 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); - } - - /* Update mirror data. */ - if (!sset_equals(&m->src_ports, &src_ports) - || !sset_equals(&m->dst_ports, &dst_ports) - || m->n_vlans != n_vlans - || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans) - || m->out_port != out_port - || m->out_vlan != out_vlan) { - bridge_flush(m->bridge); - mac_learning_flush(m->bridge->ml); - } - sset_swap(&m->src_ports, &src_ports); - sset_swap(&m->dst_ports, &dst_ports); - free(m->vlans); - m->vlans = vlans; - m->n_vlans = n_vlans; - m->out_port = out_port; - m->out_vlan = out_vlan; - - /* Update ports. */ - mirror_bit = MIRROR_MASK_C(1) << m->idx; - HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) { - if (sset_contains(&m->src_ports, port->name)) { - port->src_mirrors |= mirror_bit; - } else { - port->src_mirrors &= ~mirror_bit; - } + &s.dsts, &s.n_dsts); - if (sset_contains(&m->dst_ports, port->name)) { - port->dst_mirrors |= mirror_bit; - } else { - port->dst_mirrors &= ~mirror_bit; - } } + /* Get VLAN selection. */ + s.src_vlans = vlan_bitmap_from_array(cfg->select_vlan, cfg->n_select_vlan); + + /* Configure. */ + ofproto_mirror_register(m->bridge->ofproto, m, &s); + /* Clean up. */ - sset_destroy(&src_ports); - sset_destroy(&dst_ports); + if (s.srcs != s.dsts) { + free(s.dsts); + } + free(s.srcs); + free(s.src_vlans); + + return true; } -- 2.30.2