#include <netinet/in.h>
#include <stdbool.h>
#include <stdlib.h>
+#include "autopath.h"
+#include "bitmap.h"
+#include "bond.h"
#include "byte-order.h"
+#include "cfm.h"
#include "classifier.h"
+#include "connmgr.h"
#include "coverage.h"
-#include "discovery.h"
#include "dpif.h"
#include "dynamic-string.h"
#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"
#include "pinsched.h"
#include "pktbuf.h"
#include "poll-loop.h"
+#include "private.h"
#include "rconn.h"
#include "shash.h"
-#include "status.h"
+#include "sset.h"
#include "stream-ssl.h"
-#include "svec.h"
#include "tag.h"
+#include "timer.h"
#include "timeval.h"
#include "unaligned.h"
#include "unixctl.h"
#include "vconn.h"
+#include "vlan-bitmap.h"
#include "vlog.h"
VLOG_DEFINE_THIS_MODULE(ofproto);
COVERAGE_DEFINE(ofproto_flush);
COVERAGE_DEFINE(ofproto_invalidated);
COVERAGE_DEFINE(ofproto_no_packet_in);
-COVERAGE_DEFINE(ofproto_ofconn_stuck);
COVERAGE_DEFINE(ofproto_ofp2odp);
COVERAGE_DEFINE(ofproto_packet_in);
COVERAGE_DEFINE(ofproto_packet_out);
COVERAGE_DEFINE(ofproto_uninstallable);
COVERAGE_DEFINE(ofproto_update_port);
-#include "sflow_api.h"
-
/* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
* flow translation. */
#define MAX_RESUBMIT_RECURSION 16
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; /* In host byte order. */
+ 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 *);
-static void hton_ofp_phy_port(struct ofp_phy_port *);
+static void ofport_run(struct ofport *);
+static void ofport_wait(struct ofport *);
struct action_xlate_ctx {
/* action_xlate_ctx_init() initializes these members. */
* 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. */
};
const struct dpif_flow_stats *);
static void facet_push_stats(struct ofproto *, struct facet *);
-/* ofproto supports two kinds of OpenFlow connections:
- *
- * - "Primary" connections to ordinary OpenFlow controllers. ofproto
- * maintains persistent connections to these controllers and by default
- * sends them asynchronous messages such as packet-ins.
- *
- * - "Service" connections, e.g. from ovs-ofctl. When these connections
- * drop, it is the other side's responsibility to reconnect them if
- * necessary. ofproto does not send them asynchronous messages by default.
- *
- * Currently, active (tcp, ssl, unix) connections are always "primary"
- * connections and passive (ptcp, pssl, punix) connections are always "service"
- * connections. There is no inherent reason for this, but it reflects the
- * common case.
- */
-enum ofconn_type {
- OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
- OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
-};
-
-/* A listener for incoming OpenFlow "service" connections. */
-struct ofservice {
- struct hmap_node node; /* In struct ofproto's "services" hmap. */
- struct pvconn *pvconn; /* OpenFlow connection listener. */
-
- /* These are not used by ofservice directly. They are settings for
- * accepted "struct ofconn"s from the pvconn. */
- int probe_interval; /* Max idle time before probing, in seconds. */
- int rate_limit; /* Max packet-in rate in packets per second. */
- int burst_limit; /* Limit on accumulating packet credits. */
-};
-
-static struct ofservice *ofservice_lookup(struct ofproto *,
- const char *target);
-static int ofservice_create(struct ofproto *,
- const struct ofproto_controller *);
-static void ofservice_reconfigure(struct ofservice *,
- const struct ofproto_controller *);
-static void ofservice_destroy(struct ofproto *, struct ofservice *);
-
-/* An OpenFlow connection. */
-struct ofconn {
- struct ofproto *ofproto; /* The ofproto that owns this connection. */
- struct list node; /* In struct ofproto's "all_conns" list. */
- struct rconn *rconn; /* OpenFlow connection. */
- enum ofconn_type type; /* Type. */
- enum nx_flow_format flow_format; /* Currently selected flow format. */
-
- /* OFPT_PACKET_IN related data. */
- struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
-#define N_SCHEDULERS 2
- struct pinsched *schedulers[N_SCHEDULERS];
- struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
- int miss_send_len; /* Bytes to send of buffered packets. */
-
- /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
- * requests, and the maximum number before we stop reading OpenFlow
- * requests. */
-#define OFCONN_REPLY_MAX 100
- struct rconn_packet_counter *reply_counter;
-
- /* type == OFCONN_PRIMARY only. */
- enum nx_role role; /* Role. */
- struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
- struct discovery *discovery; /* Controller discovery object, if enabled. */
- struct status_category *ss; /* Switch status category. */
- enum ofproto_band band; /* In-band or out-of-band? */
-};
-
-
-static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
- enum ofconn_type);
-static void ofconn_destroy(struct ofconn *);
-static void ofconn_run(struct ofconn *);
-static void ofconn_wait(struct ofconn *);
-static bool ofconn_receives_async_msgs(const struct ofconn *);
-static char *ofconn_make_name(const struct ofproto *, const char *target);
-static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
-
-static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
- struct rconn_packet_counter *counter);
-
static void send_packet_in(struct ofproto *, struct dpif_upcall *,
const struct flow *, bool clone);
-static void do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn);
struct ofproto {
+ char *name; /* Datapath name. */
+ struct hmap_node hmap_node; /* In global 'all_ofprotos' hmap. */
+
/* Settings. */
uint64_t datapath_id; /* Datapath ID. */
uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
struct shash port_by_name;
uint32_t max_ports;
- /* Configuration. */
- struct switch_status *switch_status;
- struct fail_open *fail_open;
+ /* Bridging. */
struct netflow *netflow;
struct ofproto_sflow *sflow;
-
- /* In-band control. */
- struct in_band *in_band;
- long long int next_in_band_update;
- struct sockaddr_in *extra_in_band_remotes;
- size_t n_extra_remotes;
- int in_band_queue;
+ 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;
- long long int next_expiration;
+ struct timer next_expiration;
/* Facets. */
struct hmap facets;
struct tag_set revalidate_set;
/* OpenFlow connections. */
- struct hmap controllers; /* Controller "struct ofconn"s. */
- struct list all_conns; /* Contains "struct ofconn"s. */
- enum ofproto_fail_mode fail_mode;
-
- /* OpenFlow listeners. */
- struct hmap services; /* Contains "struct ofservice"s. */
- struct pvconn **snoops;
- size_t n_snoops;
-
- /* Hooks for ovs-vswitchd. */
- const struct ofhooks *ofhooks;
- void *aux;
-
- /* Used by default ofhooks. */
- struct mac_learning *ml;
+ struct connmgr *connmgr;
};
/* Map from dpif name to struct ofproto, for use by unixctl commands. */
-static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
+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);
+static void ofproto_flush_flows__(struct ofproto *);
static int ofproto_expire(struct ofproto *);
static void flow_push_stats(struct ofproto *, const struct rule *,
struct flow *, uint64_t packets, uint64_t bytes,
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;
ofproto_unixctl_init();
/* Connect to datapath and start listening for messages. */
- error = dpif_open(datapath, datapath_type, &dpif);
+ error = dpif_create_and_open(datapath, datapath_type, &dpif);
if (error) {
VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
return error;
dpif_flow_flush(dpif);
dpif_recv_purge(dpif);
+ error = dpif_port_get_name(dpif, ODPP_LOCAL,
+ local_name, sizeof local_name);
+ if (error) {
+ VLOG_ERR("%s: cannot get name of datapath local port (%s)",
+ datapath, strerror(error));
+ return error;
+ }
+
/* Initialize settings. */
p = xzalloc(sizeof *p);
+ p->name = xstrdup(dpif_name(dpif));
+ hmap_insert(&all_ofprotos, &p->hmap_node, hash_string(p->name, 0));
p->fallback_dpid = pick_fallback_dpid();
p->datapath_id = p->fallback_dpid;
p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
shash_init(&p->port_by_name);
p->max_ports = dpif_get_max_ports(dpif);
- /* Initialize submodules. */
- p->switch_status = switch_status_create(p);
- p->fail_open = NULL;
+ /* Initialize bridging. */
p->netflow = NULL;
p->sflow = NULL;
-
- /* Initialize in-band control. */
- p->in_band = NULL;
- p->in_band_queue = -1;
+ 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);
- p->next_expiration = time_msec() + 1000;
+ timer_set_duration(&p->next_expiration, 1000);
/* Initialize facet table. */
hmap_init(&p->facets);
p->need_revalidate = false;
tag_set_init(&p->revalidate_set);
- /* Initialize OpenFlow connections. */
- list_init(&p->all_conns);
- hmap_init(&p->controllers);
- hmap_init(&p->services);
- p->snoops = NULL;
- p->n_snoops = 0;
-
- /* 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);
- shash_add_once(&all_ofprotos, dpif_name(p->dpif), p);
+ /* Initialize OpenFlow connections. */
+ p->connmgr = connmgr_create(p, datapath, local_name);
+
+ init_ports(p);
*ofprotop = p;
return 0;
}
}
-static bool
-is_discovery_controller(const struct ofproto_controller *c)
-{
- return !strcmp(c->target, "discover");
-}
-
-static bool
-is_in_band_controller(const struct ofproto_controller *c)
-{
- return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
-}
-
-/* Creates a new controller in 'ofproto'. Some of the settings are initially
- * drawn from 'c', but update_controller() needs to be called later to finish
- * the new ofconn's configuration. */
-static void
-add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
-{
- struct discovery *discovery;
- struct ofconn *ofconn;
-
- if (is_discovery_controller(c)) {
- int error = discovery_create(c->accept_re, c->update_resolv_conf,
- ofproto->dpif, ofproto->switch_status,
- &discovery);
- if (error) {
- return;
- }
- } else {
- discovery = NULL;
- }
-
- ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
- ofconn->pktbuf = pktbuf_create();
- ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
- if (discovery) {
- ofconn->discovery = discovery;
- } else {
- char *name = ofconn_make_name(ofproto, c->target);
- rconn_connect(ofconn->rconn, c->target, name);
- free(name);
- }
- hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
- hash_string(c->target, 0));
-}
-
-/* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
- * target or turn discovery on or off (these are done by creating new ofconns
- * and deleting old ones), but it can update the rest of an ofconn's
- * settings. */
-static void
-update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
-{
- int probe_interval;
-
- ofconn->band = (is_in_band_controller(c)
- ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
-
- rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
-
- probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
- rconn_set_probe_interval(ofconn->rconn, probe_interval);
-
- if (ofconn->discovery) {
- discovery_set_update_resolv_conf(ofconn->discovery,
- c->update_resolv_conf);
- discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
- }
-
- ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
-}
-
-static const char *
-ofconn_get_target(const struct ofconn *ofconn)
-{
- return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
-}
-
-static struct ofconn *
-find_controller_by_target(struct ofproto *ofproto, const char *target)
-{
- struct ofconn *ofconn;
-
- HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
- hash_string(target, 0), &ofproto->controllers) {
- if (!strcmp(ofconn_get_target(ofconn), target)) {
- return ofconn;
- }
- }
- return NULL;
-}
-
-static void
-update_in_band_remotes(struct ofproto *ofproto)
-{
- const struct ofconn *ofconn;
- struct sockaddr_in *addrs;
- size_t max_addrs, n_addrs;
- bool discovery;
- size_t i;
-
- /* Allocate enough memory for as many remotes as we could possibly have. */
- max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
- addrs = xmalloc(max_addrs * sizeof *addrs);
- n_addrs = 0;
-
- /* Add all the remotes. */
- discovery = false;
- HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
- struct sockaddr_in *sin = &addrs[n_addrs];
-
- if (ofconn->band == OFPROTO_OUT_OF_BAND) {
- continue;
- }
-
- sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
- if (sin->sin_addr.s_addr) {
- sin->sin_port = rconn_get_remote_port(ofconn->rconn);
- n_addrs++;
- }
- if (ofconn->discovery) {
- discovery = true;
- }
- }
- for (i = 0; i < ofproto->n_extra_remotes; i++) {
- addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
- }
-
- /* Create or update or destroy in-band.
- *
- * Ordinarily we only enable in-band if there's at least one remote
- * address, but discovery needs the in-band rules for DHCP to be installed
- * even before we know any remote addresses. */
- if (n_addrs || discovery) {
- if (!ofproto->in_band) {
- in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
- &ofproto->in_band);
- }
- if (ofproto->in_band) {
- in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
- }
- in_band_set_queue(ofproto->in_band, ofproto->in_band_queue);
- ofproto->next_in_band_update = time_msec() + 1000;
- } else {
- in_band_destroy(ofproto->in_band);
- ofproto->in_band = NULL;
- }
-
- /* Clean up. */
- free(addrs);
-}
-
-static void
-update_fail_open(struct ofproto *p)
-{
- struct ofconn *ofconn;
-
- if (!hmap_is_empty(&p->controllers)
- && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
- struct rconn **rconns;
- size_t n;
-
- if (!p->fail_open) {
- p->fail_open = fail_open_create(p, p->switch_status);
- }
-
- n = 0;
- rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
- HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
- rconns[n++] = ofconn->rconn;
- }
-
- fail_open_set_controllers(p->fail_open, rconns, n);
- /* p->fail_open takes ownership of 'rconns'. */
- } else {
- fail_open_destroy(p->fail_open);
- p->fail_open = NULL;
- }
-}
-
void
ofproto_set_controllers(struct ofproto *p,
const struct ofproto_controller *controllers,
size_t n_controllers)
{
- struct shash new_controllers;
- struct ofconn *ofconn, *next_ofconn;
- struct ofservice *ofservice, *next_ofservice;
- bool ss_exists;
- size_t i;
-
- /* Create newly configured controllers and services.
- * Create a name to ofproto_controller mapping in 'new_controllers'. */
- shash_init(&new_controllers);
- for (i = 0; i < n_controllers; i++) {
- const struct ofproto_controller *c = &controllers[i];
-
- if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
- if (!find_controller_by_target(p, c->target)) {
- add_controller(p, c);
- }
- } else if (!pvconn_verify_name(c->target)) {
- if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
- continue;
- }
- } else {
- VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
- dpif_name(p->dpif), c->target);
- continue;
- }
-
- shash_add_once(&new_controllers, c->target, &controllers[i]);
- }
-
- /* Delete controllers that are no longer configured.
- * Update configuration of all now-existing controllers. */
- ss_exists = false;
- HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
- struct ofproto_controller *c;
-
- c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
- if (!c) {
- ofconn_destroy(ofconn);
- } else {
- update_controller(ofconn, c);
- if (ofconn->ss) {
- ss_exists = true;
- }
- }
- }
-
- /* Delete services that are no longer configured.
- * Update configuration of all now-existing services. */
- HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
- struct ofproto_controller *c;
-
- c = shash_find_data(&new_controllers,
- pvconn_get_name(ofservice->pvconn));
- if (!c) {
- ofservice_destroy(p, ofservice);
- } else {
- ofservice_reconfigure(ofservice, c);
- }
- }
-
- shash_destroy(&new_controllers);
-
- update_in_band_remotes(p);
- update_fail_open(p);
-
- if (!hmap_is_empty(&p->controllers) && !ss_exists) {
- ofconn = CONTAINER_OF(hmap_first(&p->controllers),
- struct ofconn, hmap_node);
- ofconn->ss = switch_status_register(p->switch_status, "remote",
- rconn_status_cb, ofconn->rconn);
- }
+ connmgr_set_controllers(p->connmgr, controllers, n_controllers);
}
void
ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
{
- p->fail_mode = fail_mode;
- update_fail_open(p);
+ connmgr_set_fail_mode(p->connmgr, fail_mode);
}
/* Drops the connections between 'ofproto' and all of its controllers, forcing
void
ofproto_reconnect_controllers(struct ofproto *ofproto)
{
- struct ofconn *ofconn;
-
- LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
- rconn_reconnect(ofconn->rconn);
- }
-}
-
-static bool
-any_extras_changed(const struct ofproto *ofproto,
- const struct sockaddr_in *extras, size_t n)
-{
- size_t i;
-
- if (n != ofproto->n_extra_remotes) {
- return true;
- }
-
- for (i = 0; i < n; i++) {
- const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
- const struct sockaddr_in *new = &extras[i];
-
- if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
- old->sin_port != new->sin_port) {
- return true;
- }
- }
-
- return false;
+ connmgr_reconnect(ofproto->connmgr);
}
/* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
const struct sockaddr_in *extras, size_t n)
{
- if (!any_extras_changed(ofproto, extras, n)) {
- return;
- }
-
- free(ofproto->extra_in_band_remotes);
- ofproto->n_extra_remotes = n;
- ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
-
- update_in_band_remotes(ofproto);
+ connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
}
/* Sets the OpenFlow queue used by flows set up by in-band control on
void
ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
{
- if (queue_id != ofproto->in_band_queue) {
- ofproto->in_band_queue = queue_id;
- update_in_band_remotes(ofproto);
- }
+ connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
}
void
}
}
-static int
-set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
- const struct svec *svec)
-{
- struct pvconn **pvconns = *pvconnsp;
- size_t n_pvconns = *n_pvconnsp;
- int retval = 0;
- size_t i;
-
- for (i = 0; i < n_pvconns; i++) {
- pvconn_close(pvconns[i]);
- }
- free(pvconns);
-
- pvconns = xmalloc(svec->n * sizeof *pvconns);
- n_pvconns = 0;
- for (i = 0; i < svec->n; i++) {
- const char *name = svec->names[i];
- struct pvconn *pvconn;
- int error;
-
- error = pvconn_open(name, &pvconn);
- if (!error) {
- pvconns[n_pvconns++] = pvconn;
- } else {
- VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
- if (!retval) {
- retval = error;
- }
- }
- }
-
- *pvconnsp = pvconns;
- *n_pvconnsp = n_pvconns;
-
- return retval;
-}
-
int
-ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
+ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
{
- return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
+ return connmgr_set_snoops(ofproto->connmgr, snoops);
}
int
ofproto_set_netflow(struct ofproto *ofproto,
const struct netflow_options *nf_options)
{
- if (nf_options && nf_options->collectors.n) {
+ if (nf_options && !sset_is_empty(&nf_options->collectors)) {
if (!ofproto->netflow) {
ofproto->netflow = netflow_create();
}
ofproto->sflow = NULL;
}
}
+\f
+/* Connectivity Fault Management configuration. */
+
+/* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
+void
+ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
+{
+ struct ofport *ofport = get_port(ofproto, ofp_port_to_odp_port(ofp_port));
+ if (ofport && ofport->cfm){
+ cfm_destroy(ofport->cfm);
+ ofport->cfm = NULL;
+ }
+}
-uint64_t
-ofproto_get_datapath_id(const struct ofproto *ofproto)
+/* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
+ * basic configuration from the configuration members in 'cfm', and the set of
+ * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'.
+ * Ignores the statistics members of 'cfm'.
+ *
+ * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
+void
+ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
+ const struct cfm *cfm,
+ const uint16_t *remote_mps, size_t n_remote_mps)
{
- return ofproto->datapath_id;
+ struct ofport *ofport;
+
+ ofport = get_port(ofproto, ofp_port_to_odp_port(ofp_port));
+ if (!ofport) {
+ VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
+ ofproto->name, ofp_port);
+ return;
+ }
+
+ if (!ofport->cfm) {
+ ofport->cfm = cfm_create();
+ }
+
+ ofport->cfm->mpid = cfm->mpid;
+ ofport->cfm->interval = cfm->interval;
+ memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN);
+
+ cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps);
+
+ if (!cfm_configure(ofport->cfm)) {
+ VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed",
+ ofproto->name, ofp_port,
+ netdev_get_name(ofport->netdev));
+ cfm_destroy(ofport->cfm);
+ ofport->cfm = NULL;
+ }
}
-bool
-ofproto_has_primary_controller(const struct ofproto *ofproto)
+/* Returns the connectivity fault management object associated with 'ofp_port'
+ * within 'ofproto', or a null pointer if 'ofproto' does not have a port
+ * 'ofp_port' or if that port does not have CFM configured. The caller must
+ * not modify or destroy the returned object. */
+const struct cfm *
+ofproto_port_get_cfm(struct ofproto *ofproto, uint16_t ofp_port)
{
- return !hmap_is_empty(&ofproto->controllers);
+ struct ofport *ofport = get_port(ofproto, ofp_port_to_odp_port(ofp_port));
+ return ofport ? ofport->cfm : NULL;
}
-enum ofproto_fail_mode
-ofproto_get_fail_mode(const struct ofproto *p)
+/* 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)
{
- return p->fail_mode;
+ 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);
}
+\f
+/* Bundles. */
-void
-ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
+/* Expires all MAC learning entries associated with 'port' and forces ofproto
+ * to revalidate every flow. */
+static void
+ofproto_bundle_flush_macs(struct ofbundle *bundle)
{
- size_t i;
+ struct ofproto *ofproto = bundle->ofproto;
+ struct mac_learning *ml = ofproto->ml;
+ struct mac_entry *mac, *next_mac;
- for (i = 0; i < ofproto->n_snoops; i++) {
- svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
+ 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);
+ }
}
}
-void
-ofproto_destroy(struct ofproto *p)
+static struct ofbundle *
+ofproto_bundle_lookup(const struct ofproto *ofproto, void *aux)
{
- struct ofservice *ofservice, *next_ofservice;
- struct ofconn *ofconn, *next_ofconn;
- struct ofport *ofport, *next_ofport;
- size_t i;
+ struct ofbundle *bundle;
- if (!p) {
- return;
+ HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
+ &ofproto->bundles) {
+ if (bundle->aux == aux) {
+ return bundle;
+ }
}
+ return NULL;
+}
- shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif));
+/* 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;
- /* Destroy fail-open and in-band early, since they touch the classifier. */
- fail_open_destroy(p->fail_open);
- p->fail_open = NULL;
+ 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);
+ }
+ }
+}
- in_band_destroy(p->in_band);
- p->in_band = NULL;
- free(p->extra_in_band_remotes);
+static void
+ofproto_bundle_del_port(struct ofport *port)
+{
+ struct ofbundle *bundle = port->bundle;
- ofproto_flush_flows(p);
- classifier_destroy(&p->cls);
- hmap_destroy(&p->facets);
+ list_remove(&port->bundle_node);
+ port->bundle = NULL;
- LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
- ofconn_destroy(ofconn);
+ if (bundle->lacp) {
+ lacp_slave_unregister(bundle->lacp, port);
+ }
+ if (bundle->bond) {
+ bond_slave_unregister(bundle->bond, port);
}
- hmap_destroy(&p->controllers);
- dpif_close(p->dpif);
- netdev_monitor_destroy(p->netdev_monitor);
- HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
- hmap_remove(&p->ports, &ofport->hmap_node);
- ofport_free(ofport);
+ bundle->floodable = true;
+ LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
+ if (port->opp.config & htonl(OFPPC_NO_FLOOD)) {
+ bundle->floodable = false;
+ }
}
- shash_destroy(&p->port_by_name);
+}
- switch_status_destroy(p->switch_status);
- netflow_destroy(p->netflow);
- ofproto_sflow_destroy(p->sflow);
+static bool
+ofproto_bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
+ struct lacp_slave_settings *lacp)
+{
+ struct ofport *port;
- HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
- ofservice_destroy(p, ofservice);
+ port = get_port(bundle->ofproto, ofp_port_to_odp_port(ofp_port));
+ if (!port) {
+ return false;
}
- hmap_destroy(&p->services);
- for (i = 0; i < p->n_snoops; i++) {
- pvconn_close(p->snoops[i]);
+ 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;
+ }
}
- free(p->snoops);
- mac_learning_destroy(p->ml);
+ if (lacp) {
+ lacp_slave_register(bundle->lacp, port, lacp);
+ }
- free(p->mfr_desc);
- free(p->hw_desc);
- free(p->sw_desc);
- free(p->serial_desc);
- free(p->dp_desc);
+ return true;
+}
- hmap_destroy(&p->ports);
+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;
- free(p);
+ 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);
+ }
}
-int
-ofproto_run(struct ofproto *p)
+static void
+ofproto_bundle_destroy(struct ofbundle *bundle)
{
- int error = ofproto_run1(p);
+ 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);
+}
+
+void
+ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
+{
+ 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) {
- error = ofproto_run2(p, false);
+ 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));
}
- return error;
}
static void
-process_port_change(struct ofproto *ofproto, int error, char *devname)
+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 (error == ENOBUFS) {
- reinit_ports(ofproto);
- } else if (!error) {
- update_port(ofproto, devname);
- free(devname);
+ 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);
+ }
}
}
-/* Returns a "preference level" for snooping 'ofconn'. A higher return value
- * means that 'ofconn' is more interesting for monitoring than a lower return
- * value. */
+static void
+ofproto_bundle_wait(struct ofbundle *bundle)
+{
+ if (bundle->lacp) {
+ lacp_wait(bundle->lacp);
+ }
+ if (bundle->bond) {
+ bond_wait(bundle->bond);
+ }
+}
+\f
static int
-snoop_preference(const struct ofconn *ofconn)
-{
- switch (ofconn->role) {
- case NX_ROLE_MASTER:
- return 3;
- case NX_ROLE_OTHER:
- return 2;
- case NX_ROLE_SLAVE:
- return 1;
- default:
- /* Shouldn't happen. */
- return 0;
+ofproto_mirror_scan(struct ofproto *ofproto)
+{
+ int idx;
+
+ for (idx = 0; idx < MAX_MIRRORS; idx++) {
+ if (!ofproto->mirrors[idx]) {
+ return idx;
+ }
}
+ return -1;
}
-/* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
- * Connects this vconn to a controller. */
-static void
-add_snooper(struct ofproto *ofproto, struct vconn *vconn)
+static struct ofmirror *
+ofproto_mirror_lookup(struct ofproto *ofproto, void *aux)
{
- struct ofconn *ofconn, *best;
+ int i;
- /* Pick a controller for monitoring. */
- best = NULL;
- LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
- if (ofconn->type == OFCONN_PRIMARY
- && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
- best = ofconn;
+ for (i = 0; i < MAX_MIRRORS; i++) {
+ struct ofmirror *mirror = ofproto->mirrors[i];
+ if (mirror && mirror->aux == aux) {
+ return mirror;
}
}
- if (best) {
- rconn_add_monitor(best->rconn, vconn);
+ 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 {
- VLOG_INFO_RL(&rl, "no controller connection to snoop");
- vconn_close(vconn);
+ 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;
+}
+\f
+bool
+ofproto_has_snoops(const struct ofproto *ofproto)
+{
+ return connmgr_has_snoops(ofproto->connmgr);
+}
+
+void
+ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
+{
+ connmgr_get_snoops(ofproto->connmgr, snoops);
+}
+
+void
+ofproto_destroy(struct ofproto *p)
+{
+ struct ofport *ofport, *next_ofport;
+ int i;
+
+ if (!p) {
+ return;
+ }
+
+ 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);
+
+ dpif_close(p->dpif);
+
+ netdev_monitor_destroy(p->netdev_monitor);
+ HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
+ hmap_remove(&p->ports, &ofport->hmap_node);
+ ofport_free(ofport);
+ }
+ shash_destroy(&p->port_by_name);
+
+ netflow_destroy(p->netflow);
+ ofproto_sflow_destroy(p->sflow);
+
+ free(p->mfr_desc);
+ free(p->hw_desc);
+ free(p->sw_desc);
+ free(p->serial_desc);
+ free(p->dp_desc);
+
+ hmap_destroy(&p->ports);
+
+ free(p->name);
+ free(p);
+}
+
+static void
+process_port_change(struct ofproto *ofproto, int error, char *devname)
+{
+ if (error == ENOBUFS) {
+ reinit_ports(ofproto);
+ } else if (!error) {
+ update_port(ofproto, devname);
+ free(devname);
}
}
int
-ofproto_run1(struct ofproto *p)
+ofproto_run(struct ofproto *p)
{
- struct ofconn *ofconn, *next_ofconn;
- struct ofservice *ofservice;
+ struct ofbundle *bundle;
+ struct ofport *ofport;
char *devname;
int error;
int i;
- if (shash_is_empty(&p->port_by_name)) {
- init_ports(p);
- }
+ dpif_run(p->dpif);
for (i = 0; i < 50; i++) {
struct dpif_upcall packet;
* spin from here on out. */
static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
- dpif_name(p->dpif));
+ p->name);
return ENODEV;
}
break;
process_port_change(p, error, devname);
}
- if (p->in_band) {
- if (time_msec() >= p->next_in_band_update) {
- update_in_band_remotes(p);
- }
- in_band_run(p->in_band);
- }
-
- LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
- ofconn_run(ofconn);
+ HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
+ ofport_run(ofport);
}
- /* Fail-open maintenance. Do this after processing the ofconns since
- * fail-open checks the status of the controller rconn. */
- if (p->fail_open) {
- fail_open_run(p->fail_open);
+ HMAP_FOR_EACH (bundle, hmap_node, &p->bundles) {
+ ofproto_bundle_run(bundle);
}
- HMAP_FOR_EACH (ofservice, node, &p->services) {
- struct vconn *vconn;
- int retval;
-
- retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
- if (!retval) {
- struct rconn *rconn;
- char *name;
-
- rconn = rconn_create(ofservice->probe_interval, 0);
- name = ofconn_make_name(p, vconn_get_name(vconn));
- rconn_connect_unreliably(rconn, vconn, name);
- free(name);
+ connmgr_run(p->connmgr, handle_openflow);
- ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
- ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
- ofservice->burst_limit);
- } else if (retval != EAGAIN) {
- VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
- }
- }
-
- for (i = 0; i < p->n_snoops; i++) {
- struct vconn *vconn;
- int retval;
-
- retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
- if (!retval) {
- add_snooper(p, vconn);
- } else if (retval != EAGAIN) {
- VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
- }
- }
-
- if (time_msec() >= p->next_expiration) {
+ if (timer_expired(&p->next_expiration)) {
int delay = ofproto_expire(p);
- p->next_expiration = time_msec() + delay;
+ timer_set_duration(&p->next_expiration, delay);
COVERAGE_INC(ofproto_expiration);
}
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)) {
void
ofproto_wait(struct ofproto *p)
{
- struct ofservice *ofservice;
- struct ofconn *ofconn;
- size_t i;
+ struct ofbundle *bundle;
+ struct ofport *ofport;
+ dpif_wait(p->dpif);
+ 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);
- LIST_FOR_EACH (ofconn, node, &p->all_conns) {
- ofconn_wait(ofconn);
- }
- if (p->in_band) {
- poll_timer_wait_until(p->next_in_band_update);
- in_band_wait(p->in_band);
- }
- if (p->fail_open) {
- fail_open_wait(p->fail_open);
- }
if (p->sflow) {
ofproto_sflow_wait(p->sflow);
}
/* Shouldn't happen, but if it does just go around again. */
VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
poll_immediate_wake();
- } else if (p->next_expiration != LLONG_MAX) {
- poll_timer_wait_until(p->next_expiration);
- }
- HMAP_FOR_EACH (ofservice, node, &p->services) {
- pvconn_wait(ofservice->pvconn);
- }
- for (i = 0; i < p->n_snoops; i++) {
- pvconn_wait(p->snoops[i]);
+ } else {
+ timer_wait(&p->next_expiration);
}
+ connmgr_wait(p->connmgr);
+}
+
+bool
+ofproto_is_alive(const struct ofproto *p)
+{
+ return connmgr_has_controllers(p->connmgr);
}
void
-ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
+ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
+ struct shash *info)
{
- tag_set_add(&ofproto->revalidate_set, tag);
+ connmgr_get_controller_info(ofproto->connmgr, info);
}
-struct tag_set *
-ofproto_get_revalidate_set(struct ofproto *ofproto)
+void
+ofproto_free_ofproto_controller_info(struct shash *info)
{
- return &ofproto->revalidate_set;
+ struct shash_node *node;
+
+ SHASH_FOR_EACH (node, info) {
+ struct ofproto_controller_info *cinfo = node->data;
+ while (cinfo->pairs.n) {
+ free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
+ }
+ free(cinfo);
+ }
+ shash_destroy(info);
}
-bool
-ofproto_is_alive(const struct ofproto *p)
+/* Makes a deep copy of 'old' into 'port'. */
+void
+ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
{
- return !hmap_is_empty(&p->controllers);
+ port->name = xstrdup(old->name);
+ port->type = xstrdup(old->type);
+ port->ofp_port = old->ofp_port;
}
+/* Frees memory allocated to members of 'ofproto_port'.
+ *
+ * Do not call this function on an ofproto_port obtained from
+ * ofproto_port_dump_next(): that function retains ownership of the data in the
+ * ofproto_port. */
void
-ofproto_get_ofproto_controller_info(const struct ofproto * ofproto,
- struct shash *info)
+ofproto_port_destroy(struct ofproto_port *ofproto_port)
{
- const struct ofconn *ofconn;
+ free(ofproto_port->name);
+ free(ofproto_port->type);
+}
- shash_init(info);
+/* Converts a dpif_port into an ofproto_port.
+ *
+ * This only makes a shallow copy, so make sure that the dpif_port doesn't get
+ * freed while the ofproto_port is still in use. You can choose to free the
+ * ofproto_port instead of the dpif_port. */
+static void
+ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
+ struct dpif_port *dpif_port)
+{
+ ofproto_port->name = dpif_port->name;
+ ofproto_port->type = dpif_port->type;
+ ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
+}
- HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
- const struct rconn *rconn = ofconn->rconn;
- const int last_error = rconn_get_last_error(rconn);
- struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo);
+/* Initializes 'dump' to begin dumping the ports in an ofproto.
+ *
+ * This function provides no status indication. An error status for the entire
+ * dump operation is provided when it is completed by calling
+ * ofproto_port_dump_done().
+ */
+void
+ofproto_port_dump_start(struct ofproto_port_dump *dump,
+ const struct ofproto *ofproto)
+{
+ struct dpif_port_dump *dpif_dump;
- shash_add(info, rconn_get_target(rconn), cinfo);
+ dump->state = dpif_dump = xmalloc(sizeof *dpif_dump);
+ dpif_port_dump_start(dpif_dump, ofproto->dpif);
+}
- cinfo->is_connected = rconn_is_connected(rconn);
- cinfo->role = ofconn->role;
+/* Attempts to retrieve another port from 'dump', which must have been created
+ * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
+ * 'port' and returns true. On failure, returns false.
+ *
+ * Failure might indicate an actual error or merely that the last port has been
+ * dumped. An error status for the entire dump operation is provided when it
+ * is completed by calling ofproto_port_dump_done().
+ *
+ * The ofproto owns the data stored in 'port'. It will remain valid until at
+ * least the next time 'dump' is passed to ofproto_port_dump_next() or
+ * ofproto_port_dump_done(). */
+bool
+ofproto_port_dump_next(struct ofproto_port_dump *dump,
+ struct ofproto_port *port)
+{
+ struct dpif_port_dump *dpif_dump = dump->state;
+ struct dpif_port dpif_port;
+ bool ok;
- cinfo->pairs.n = 0;
+ ok = dpif_port_dump_next(dpif_dump, &dpif_port);
+ if (ok) {
+ ofproto_port_from_dpif_port(port, &dpif_port);
+ }
+ return ok;
+}
- if (last_error) {
- cinfo->pairs.keys[cinfo->pairs.n] = "last_error";
- cinfo->pairs.values[cinfo->pairs.n++] =
- xstrdup(ovs_retval_to_string(last_error));
- }
+/* Completes port table dump operation 'dump', which must have been created
+ * with ofproto_port_dump_start(). Returns 0 if the dump operation was
+ * error-free, otherwise a positive errno value describing the problem. */
+int
+ofproto_port_dump_done(struct ofproto_port_dump *dump)
+{
+ struct dpif_port_dump *dpif_dump = dump->state;
+ int error = dpif_port_dump_done(dpif_dump);
+ free(dpif_dump);
+ return error;
+}
- cinfo->pairs.keys[cinfo->pairs.n] = "state";
- cinfo->pairs.values[cinfo->pairs.n++] =
- xstrdup(rconn_get_state(rconn));
+/* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
+ * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
+ * is non-null). On failure, returns a positive errno value and sets
+ * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
+int
+ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
+ uint16_t *ofp_portp)
+{
+ uint16_t odp_port;
+ int error;
- cinfo->pairs.keys[cinfo->pairs.n] = "time_in_state";
- cinfo->pairs.values[cinfo->pairs.n++] =
- xasprintf("%u", rconn_get_state_elapsed(rconn));
+ error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
+ if (!error) {
+ update_port(ofproto, netdev_get_name(netdev));
+ }
+ if (ofp_portp) {
+ *ofp_portp = error ? OFPP_NONE : odp_port_to_ofp_port(odp_port);
}
+ return error;
}
-void
-ofproto_free_ofproto_controller_info(struct shash *info)
+/* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
+ * initializes '*port' appropriately; on failure, returns a positive errno
+ * value.
+ *
+ * The caller owns the data in 'port' and must free it with
+ * ofproto_port_destroy() when it is no longer needed. */
+int
+ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
+ struct ofproto_port *port)
{
- struct shash_node *node;
+ struct dpif_port dpif_port;
+ int error;
- SHASH_FOR_EACH (node, info) {
- struct ofproto_controller_info *cinfo = node->data;
- while (cinfo->pairs.n) {
- free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
- }
- free(cinfo);
+ error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
+ if (!error) {
+ ofproto_port_from_dpif_port(port, &dpif_port);
}
- shash_destroy(info);
+ return error;
}
-/* Deletes port number 'odp_port' from the datapath for 'ofproto'.
- *
- * This is almost the same as calling dpif_port_del() directly on the
- * datapath, but it also makes 'ofproto' close its open netdev for the port
- * (if any). This makes it possible to create a new netdev of a different
- * type under the same name, which otherwise the netdev library would refuse
- * to do because of the conflict. (The netdev would eventually get closed on
- * the next trip through ofproto_run(), but this interface is more direct.)
- *
+/* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
* Returns 0 if successful, otherwise a positive errno. */
int
-ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
+ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
{
+ uint32_t odp_port = ofp_port_to_odp_port(ofp_port);
struct ofport *ofport = get_port(ofproto, odp_port);
- const char *name = ofport ? ofport->opp.name : "<unknown>";
+ const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
int error;
error = dpif_port_del(ofproto->dpif, odp_port);
- if (error) {
- VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
- dpif_name(ofproto->dpif), odp_port, name, strerror(error));
- } else if (ofport) {
- /* 'name' is ofport->opp.name and update_port() is going to destroy
- * 'ofport'. Just in case update_port() refers to 'name' after it
+ if (!error && ofport) {
+ /* 'name' is the netdev's name and update_port() is going to close the
+ * netdev. Just in case update_port() refers to 'name' after it
* destroys 'ofport', make a copy of it around the update_port()
* call. */
char *devname = xstrdup(name);
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 & OFPPC_NO_FLOOD);
-}
-
-int
-ofproto_send_packet(struct ofproto *p, const struct flow *flow,
- const union ofp_action *actions, size_t n_actions,
+/* 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'.
+ *
+ * Returns 0 if successful, otherwise a positive errno value. */
+static int
+ofproto_send_packet(struct ofproto *ofproto,
+ uint32_t port_no, uint16_t vlan_tci,
const struct ofpbuf *packet)
{
- struct action_xlate_ctx ctx;
- struct ofpbuf *odp_actions;
-
- action_xlate_ctx_init(&ctx, p, flow, packet);
- /* Always xlate packets originated in this function. */
- ctx.check_special = false;
- odp_actions = xlate_actions(&ctx, actions, n_actions);
-
- /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
- * error code? */
- dpif_execute(p->dpif, odp_actions->data, odp_actions->size, packet);
+ struct ofpbuf odp_actions;
+ int error;
- ofpbuf_delete(odp_actions);
+ ofpbuf_init(&odp_actions, 32);
+ if (vlan_tci != 0) {
+ nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
+ ntohs(vlan_tci & ~VLAN_CFI));
+ }
+ nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, port_no);
+ error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size,
+ packet);
+ ofpbuf_uninit(&odp_actions);
- return 0;
+ if (error) {
+ VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
+ ofproto->name, port_no, strerror(error));
+ }
+ return error;
}
/* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
}
}
-void
-ofproto_flush_flows(struct ofproto *ofproto)
+static void
+ofproto_flush_flows__(struct ofproto *ofproto)
{
struct facet *facet, *next_facet;
struct rule *rule, *next_rule;
* individually since we are about to blow away all the facets with
* dpif_flow_flush(). */
facet->installed = false;
+ facet->dp_packet_count = 0;
+ facet->dp_byte_count = 0;
facet_remove(ofproto, facet);
}
}
dpif_flow_flush(ofproto->dpif);
- if (ofproto->in_band) {
- in_band_flushed(ofproto->in_band);
- }
- if (ofproto->fail_open) {
- fail_open_flushed(ofproto->fail_open);
- }
+}
+
+void
+ofproto_flush_flows(struct ofproto *ofproto)
+{
+ ofproto_flush_flows__(ofproto);
+ connmgr_flushed(ofproto->connmgr);
}
\f
static void
reinit_ports(struct ofproto *p)
{
struct dpif_port_dump dump;
- struct shash_node *node;
- struct shash devnames;
+ struct sset devnames;
struct ofport *ofport;
struct dpif_port dpif_port;
+ const char *devname;
COVERAGE_INC(ofproto_reinit_ports);
- shash_init(&devnames);
+ sset_init(&devnames);
HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
- shash_add_once (&devnames, ofport->opp.name, NULL);
+ sset_add(&devnames, netdev_get_name(ofport->netdev));
}
DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
- shash_add_once (&devnames, dpif_port.name, NULL);
+ sset_add(&devnames, dpif_port.name);
}
- SHASH_FOR_EACH (node, &devnames) {
- update_port(p, node->name);
+ SSET_FOR_EACH (devname, &devnames) {
+ update_port(p, devname);
}
- shash_destroy(&devnames);
+ sset_destroy(&devnames);
}
-static struct ofport *
-make_ofport(const struct dpif_port *dpif_port)
+/* Opens and returns a netdev for 'dpif_port', or a null pointer if the netdev
+ * cannot be opened. On success, also fills in 'opp'. */
+static struct netdev *
+ofport_open(const struct dpif_port *dpif_port, struct ofp_phy_port *opp)
{
+ uint32_t curr, advertised, supported, peer;
struct netdev_options netdev_options;
enum netdev_flags flags;
- struct ofport *ofport;
struct netdev *netdev;
int error;
return NULL;
}
- ofport = xzalloc(sizeof *ofport);
- ofport->netdev = netdev;
- ofport->odp_port = dpif_port->port_no;
- ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no);
- netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
- ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name);
-
netdev_get_flags(netdev, &flags);
- ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
+ netdev_get_features(netdev, &curr, &advertised, &supported, &peer);
- ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
+ opp->port_no = htons(odp_port_to_ofp_port(dpif_port->port_no));
+ netdev_get_etheraddr(netdev, opp->hw_addr);
+ ovs_strzcpy(opp->name, dpif_port->name, sizeof opp->name);
+ opp->config = flags & NETDEV_UP ? 0 : htonl(OFPPC_PORT_DOWN);
+ opp->state = netdev_get_carrier(netdev) ? 0 : htonl(OFPPS_LINK_DOWN);
+ opp->curr = htonl(curr);
+ opp->advertised = htonl(advertised);
+ opp->supported = htonl(supported);
+ opp->peer = htonl(peer);
- netdev_get_features(netdev,
- &ofport->opp.curr, &ofport->opp.advertised,
- &ofport->opp.supported, &ofport->opp.peer);
- return ofport;
+ return netdev;
}
static bool
}
}
-static int
-ofport_equal(const struct ofport *a_, const struct ofport *b_)
+/* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
+ * port number, and 'config' bits other than OFPPC_PORT_DOWN are
+ * disregarded. */
+static bool
+ofport_equal(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
{
- const struct ofp_phy_port *a = &a_->opp;
- const struct ofp_phy_port *b = &b_->opp;
-
BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
- return (a->port_no == b->port_no
- && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
- && !strcmp(a->name, b->name)
+ return (!memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
&& a->state == b->state
- && a->config == b->config
+ && !((a->config ^ b->config) & htonl(OFPPC_PORT_DOWN))
&& a->curr == b->curr
&& a->advertised == b->advertised
&& a->supported == b->supported
&& a->peer == b->peer);
}
+/* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
+ * The caller must ensure that 'p' does not have a conflicting ofport (that is,
+ * one with the same name or port number). */
static void
-send_port_status(struct ofproto *p, const struct ofport *ofport,
- uint8_t reason)
+ofport_install(struct ofproto *p,
+ struct netdev *netdev, const struct ofp_phy_port *opp)
{
- /* XXX Should limit the number of queued port status change messages. */
- struct ofconn *ofconn;
- LIST_FOR_EACH (ofconn, node, &p->all_conns) {
- struct ofp_port_status *ops;
- struct ofpbuf *b;
-
- /* Primary controllers, even slaves, should always get port status
- updates. Otherwise obey ofconn_receives_async_msgs(). */
- if (ofconn->type != OFCONN_PRIMARY
- && !ofconn_receives_async_msgs(ofconn)) {
- continue;
- }
+ const char *netdev_name = netdev_get_name(netdev);
+ struct ofport *ofport;
- ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
- ops->reason = reason;
- ops->desc = ofport->opp;
- hton_ofp_phy_port(&ops->desc);
- queue_tx(b, ofconn, NULL);
- }
-}
+ connmgr_send_port_status(p->connmgr, opp, OFPPR_ADD);
-static void
-ofport_install(struct ofproto *p, struct ofport *ofport)
-{
- const char *netdev_name = ofport->opp.name;
+ /* Create ofport. */
+ ofport = xmalloc(sizeof *ofport);
+ ofport->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);
hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
shash_add(&p->port_by_name, netdev_name, ofport);
}
}
+/* Removes 'ofport' from 'p' and destroys it. */
static void
-ofport_remove(struct ofproto *p, struct ofport *ofport)
+ofport_remove(struct ofport *ofport)
{
- 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, ofport->opp.name));
- 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);
}
+/* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
+ * destroys it. */
static void
-ofport_free(struct ofport *ofport)
-{
- if (ofport) {
- netdev_close(ofport->netdev);
- free(ofport);
- }
-}
-
-static struct ofport *
-get_port(const struct ofproto *ofproto, uint16_t odp_port)
+ofport_remove_with_name(struct ofproto *ofproto, const char *name)
{
- struct ofport *port;
-
- HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
- hash_int(odp_port, 0), &ofproto->ports) {
- if (port->odp_port == odp_port) {
- return port;
- }
+ struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
+ if (port) {
+ ofport_remove(port);
}
- return NULL;
}
+/* Updates 'port' within 'ofproto' with the new 'netdev' and 'opp'.
+ *
+ * Does not handle a name or port number change. The caller must implement
+ * such a change as a delete followed by an add. */
static void
-update_port(struct ofproto *p, const char *devname)
+ofport_modified(struct ofport *port,
+ struct netdev *netdev, struct ofp_phy_port *opp)
{
- struct dpif_port dpif_port;
- struct ofport *old_ofport;
- struct ofport *new_ofport;
- int error;
-
- COVERAGE_INC(ofproto_update_port);
+ struct ofproto *ofproto = port->ofproto;
- /* Query the datapath for port information. */
- error = dpif_port_query_by_name(p->dpif, devname, &dpif_port);
-
- /* Find the old ofport. */
- old_ofport = shash_find_data(&p->port_by_name, devname);
- if (!error) {
- if (!old_ofport) {
- /* There's no port named 'devname' but there might be a port with
- * the same port number. This could happen if a port is deleted
- * and then a new one added in its place very quickly, or if a port
- * is renamed. In the former case we want to send an OFPPR_DELETE
- * and an OFPPR_ADD, and in the latter case we want to send a
- * single OFPPR_MODIFY. We can distinguish the cases by comparing
- * the old port's ifindex against the new port, or perhaps less
- * reliably but more portably by comparing the old port's MAC
- * against the new port's MAC. However, this code isn't that smart
- * and always sends an OFPPR_MODIFY (XXX). */
- old_ofport = get_port(p, dpif_port.port_no);
- }
- } else if (error != ENOENT && error != ENODEV) {
- VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
- "%s", strerror(error));
- goto exit;
+ if (port->bundle && port->bundle->bond) {
+ bond_slave_set_netdev(port->bundle->bond, port, netdev);
}
- /* Create a new ofport. */
- new_ofport = !error ? make_ofport(&dpif_port) : NULL;
+ 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)));
+ port->opp.state = opp->state;
+ port->opp.curr = opp->curr;
+ port->opp.advertised = opp->advertised;
+ port->opp.supported = opp->supported;
+ port->opp.peer = opp->peer;
- /* Eliminate a few pathological cases. */
- if (!old_ofport && !new_ofport) {
- goto exit;
- } else if (old_ofport && new_ofport) {
- /* Most of the 'config' bits are OpenFlow soft state, but
- * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
- * OpenFlow bits from old_ofport. (make_ofport() only sets
- * OFPPC_PORT_DOWN and leaves the other bits 0.) */
- new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
-
- if (ofport_equal(old_ofport, new_ofport)) {
- /* False alarm--no change. */
- ofport_free(new_ofport);
- goto exit;
- }
- }
+ netdev_monitor_remove(ofproto->netdev_monitor, port->netdev);
+ netdev_monitor_add(ofproto->netdev_monitor, netdev);
- /* Now deal with the normal cases. */
- if (old_ofport) {
- ofport_remove(p, old_ofport);
- }
- if (new_ofport) {
- ofport_install(p, new_ofport);
- }
- send_port_status(p, new_ofport ? new_ofport : old_ofport,
- (!old_ofport ? OFPPR_ADD
- : !new_ofport ? OFPPR_DELETE
- : OFPPR_MODIFY));
- ofport_free(old_ofport);
+ netdev_close(port->netdev);
+ port->netdev = netdev;
-exit:
- dpif_port_destroy(&dpif_port);
+ connmgr_send_port_status(ofproto->connmgr, &port->opp, OFPPR_MODIFY);
}
-static int
-init_ports(struct ofproto *p)
+static void
+ofport_run(struct ofport *ofport)
{
- struct dpif_port_dump dump;
- struct dpif_port dpif_port;
+ if (ofport->cfm) {
+ cfm_run(ofport->cfm);
- DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
- if (!ofport_conflicts(p, &dpif_port)) {
- struct ofport *ofport = make_ofport(&dpif_port);
- if (ofport) {
- ofport_install(p, ofport);
- }
+ if (cfm_should_send_ccm(ofport->cfm)) {
+ struct ofpbuf packet;
+ struct ccm *ccm;
+
+ ofpbuf_init(&packet, 0);
+ ccm = eth_compose(&packet, eth_addr_ccm, ofport->opp.hw_addr,
+ ETH_TYPE_CFM, sizeof *ccm);
+ cfm_compose_ccm(ofport->cfm, ccm);
+ ofproto_send_packet(ofport->ofproto, ofport->odp_port, 0, &packet);
+ ofpbuf_uninit(&packet);
}
}
-
- return 0;
-}
-\f
-static struct ofconn *
-ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
-{
- struct ofconn *ofconn = xzalloc(sizeof *ofconn);
- ofconn->ofproto = p;
- list_push_back(&p->all_conns, &ofconn->node);
- ofconn->rconn = rconn;
- ofconn->type = type;
- ofconn->flow_format = NXFF_OPENFLOW10;
- ofconn->role = NX_ROLE_OTHER;
- ofconn->packet_in_counter = rconn_packet_counter_create ();
- ofconn->pktbuf = NULL;
- ofconn->miss_send_len = 0;
- ofconn->reply_counter = rconn_packet_counter_create ();
- return ofconn;
}
static void
-ofconn_destroy(struct ofconn *ofconn)
+ofport_wait(struct ofport *ofport)
{
- if (ofconn->type == OFCONN_PRIMARY) {
- hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
+ if (ofport->cfm) {
+ cfm_wait(ofport->cfm);
}
- discovery_destroy(ofconn->discovery);
-
- list_remove(&ofconn->node);
- switch_status_unregister(ofconn->ss);
- rconn_destroy(ofconn->rconn);
- rconn_packet_counter_destroy(ofconn->packet_in_counter);
- rconn_packet_counter_destroy(ofconn->reply_counter);
- pktbuf_destroy(ofconn->pktbuf);
- free(ofconn);
}
static void
-ofconn_run(struct ofconn *ofconn)
+ofport_unregister(struct ofport *port)
{
- struct ofproto *p = ofconn->ofproto;
- int iteration;
- size_t i;
+ struct ofbundle *bundle = port->bundle;
- if (ofconn->discovery) {
- char *controller_name;
- if (rconn_is_connectivity_questionable(ofconn->rconn)) {
- discovery_question_connectivity(ofconn->discovery);
+ 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;
}
- if (discovery_run(ofconn->discovery, &controller_name)) {
- if (controller_name) {
- char *ofconn_name = ofconn_make_name(p, controller_name);
- rconn_connect(ofconn->rconn, controller_name, ofconn_name);
- free(ofconn_name);
- } else {
- rconn_disconnect(ofconn->rconn);
- }
- }
- }
-
- for (i = 0; i < N_SCHEDULERS; i++) {
- pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
}
- rconn_run(ofconn->rconn);
-
- if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
- /* Limit the number of iterations to prevent other tasks from
- * starving. */
- for (iteration = 0; iteration < 50; iteration++) {
- struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
- if (!of_msg) {
- break;
- }
- if (p->fail_open) {
- fail_open_maybe_recover(p->fail_open);
- }
- handle_openflow(ofconn, of_msg);
- ofpbuf_delete(of_msg);
- }
- }
+ cfm_destroy(port->cfm);
+ port->cfm = NULL;
+}
- if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
- ofconn_destroy(ofconn);
+void
+ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
+{
+ struct ofport *port = get_port(ofproto, ofp_port_to_odp_port(ofp_port));
+ if (port) {
+ ofport_unregister(port);
}
}
static void
-ofconn_wait(struct ofconn *ofconn)
+ofport_free(struct ofport *port)
{
- int i;
+ if (port) {
+ struct ofproto *ofproto = port->ofproto;
+ const char *name = netdev_get_name(port->netdev);
- if (ofconn->discovery) {
- discovery_wait(ofconn->discovery);
- }
- for (i = 0; i < N_SCHEDULERS; i++) {
- pinsched_wait(ofconn->schedulers[i]);
- }
- rconn_run_wait(ofconn->rconn);
- if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
- rconn_recv_wait(ofconn->rconn);
- } else {
- COVERAGE_INC(ofproto_ofconn_stuck);
- }
-}
+ ofport_unregister(port);
-/* Returns true if 'ofconn' should receive asynchronous messages. */
-static bool
-ofconn_receives_async_msgs(const struct ofconn *ofconn)
-{
- if (ofconn->type == OFCONN_PRIMARY) {
- /* Primary controllers always get asynchronous messages unless they
- * have configured themselves as "slaves". */
- return ofconn->role != NX_ROLE_SLAVE;
- } else {
- /* Service connections don't get asynchronous messages unless they have
- * explicitly asked for them by setting a nonzero miss send length. */
- return ofconn->miss_send_len > 0;
+ 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);
}
}
-/* Returns a human-readable name for an OpenFlow connection between 'ofproto'
- * and 'target', suitable for use in log messages for identifying the
- * connection.
- *
- * The name is dynamically allocated. The caller should free it (with free())
- * when it is no longer needed. */
-static char *
-ofconn_make_name(const struct ofproto *ofproto, const char *target)
+static struct ofport *
+get_port(const struct ofproto *ofproto, uint16_t odp_port)
{
- return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
+ struct ofport *port;
+
+ HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
+ hash_int(odp_port, 0), &ofproto->ports) {
+ if (port->odp_port == odp_port) {
+ return port;
+ }
+ }
+ return NULL;
}
static void
-ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
+update_port(struct ofproto *ofproto, const char *name)
{
- int i;
+ struct dpif_port dpif_port;
+ struct ofp_phy_port opp;
+ struct netdev *netdev;
+ struct ofport *port;
- for (i = 0; i < N_SCHEDULERS; i++) {
- struct pinsched **s = &ofconn->schedulers[i];
+ COVERAGE_INC(ofproto_update_port);
- if (rate > 0) {
- if (!*s) {
- *s = pinsched_create(rate, burst,
- ofconn->ofproto->switch_status);
+ /* Fetch 'name''s location and properties from the datapath. */
+ netdev = (!dpif_port_query_by_name(ofproto->dpif, name, &dpif_port)
+ ? ofport_open(&dpif_port, &opp)
+ : NULL);
+ if (netdev) {
+ port = get_port(ofproto, dpif_port.port_no);
+ if (port && !strcmp(netdev_get_name(port->netdev), name)) {
+ /* 'name' hasn't changed location. Any properties changed? */
+ if (!ofport_equal(&port->opp, &opp)) {
+ ofport_modified(port, netdev, &opp);
} else {
- pinsched_set_limits(*s, rate, burst);
+ netdev_close(netdev);
}
} else {
- pinsched_destroy(*s);
- *s = NULL;
+ /* If 'port' is nonnull then its name differs from 'name' and thus
+ * we should delete it. If we think there's a port named 'name'
+ * then its port number must be wrong now so delete it too. */
+ if (port) {
+ ofport_remove(port);
+ }
+ ofport_remove_with_name(ofproto, name);
+ ofport_install(ofproto, netdev, &opp);
}
+ } else {
+ /* Any port named 'name' is gone now. */
+ ofport_remove_with_name(ofproto, name);
}
-}
-\f
-static void
-ofservice_reconfigure(struct ofservice *ofservice,
- const struct ofproto_controller *c)
-{
- ofservice->probe_interval = c->probe_interval;
- ofservice->rate_limit = c->rate_limit;
- ofservice->burst_limit = c->burst_limit;
+ dpif_port_destroy(&dpif_port);
}
-/* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
- * positive errno value. */
static int
-ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
-{
- struct ofservice *ofservice;
- struct pvconn *pvconn;
- int error;
-
- error = pvconn_open(c->target, &pvconn);
- if (error) {
- return error;
- }
-
- ofservice = xzalloc(sizeof *ofservice);
- hmap_insert(&ofproto->services, &ofservice->node,
- hash_string(c->target, 0));
- ofservice->pvconn = pvconn;
-
- ofservice_reconfigure(ofservice, c);
-
- return 0;
-}
-
-static void
-ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
-{
- hmap_remove(&ofproto->services, &ofservice->node);
- pvconn_close(ofservice->pvconn);
- free(ofservice);
-}
-
-/* Finds and returns the ofservice within 'ofproto' that has the given
- * 'target', or a null pointer if none exists. */
-static struct ofservice *
-ofservice_lookup(struct ofproto *ofproto, const char *target)
+init_ports(struct ofproto *p)
{
- struct ofservice *ofservice;
+ struct dpif_port_dump dump;
+ struct dpif_port dpif_port;
+
+ DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
+ if (!ofport_conflicts(p, &dpif_port)) {
+ struct ofp_phy_port opp;
+ struct netdev *netdev;
- HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
- &ofproto->services) {
- if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
- return ofservice;
+ netdev = ofport_open(&dpif_port, &opp);
+ if (netdev) {
+ ofport_install(p, netdev, &opp);
+ }
}
}
- return NULL;
+
+ return 0;
}
\f
/* Returns true if 'rule' should be hidden from the controller.
assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
flow_extract_stats(&facet->flow, packet, &stats);
+ stats.used = time_msec();
if (execute_odp_actions(ofproto, &facet->flow,
facet->actions, facet->actions_len, packet)) {
- facet->used = time_msec();
facet_update_stats(ofproto, facet, &stats);
- netflow_flow_update_time(ofproto->netflow,
- &facet->nf_flow, facet->used);
}
}
const struct nlattr *actions, size_t actions_len,
struct dpif_flow_stats *stats)
{
- uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
+ struct odputil_keybuf keybuf;
enum dpif_flow_put_flags flags;
struct ofpbuf key;
facet->dp_byte_count = 0;
}
- ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
+ ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
odp_flow_key_from_flow(&key, &facet->flow);
- assert(key.base == keybuf);
return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
actions, actions_len, 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);
+ }
+ }
}
}
facet_uninstall(struct ofproto *p, struct facet *facet)
{
if (facet->installed) {
- uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
+ struct odputil_keybuf keybuf;
struct dpif_flow_stats stats;
struct ofpbuf key;
- ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
+ ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
odp_flow_key_from_flow(&key, &facet->flow);
- assert(key.base == keybuf);
if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
facet_update_stats(p, facet, &stats);
facet->installed = false;
facet->dp_packet_count = 0;
facet->dp_byte_count = 0;
+ } else {
+ assert(facet->dp_packet_count == 0);
+ assert(facet->dp_byte_count == 0);
}
}
}
}
- return NULL;
-}
-
-/* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
- * Returns it if found, otherwise a null pointer.
- *
- * The returned facet is guaranteed to be valid. */
-static struct facet *
-facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
-{
- struct facet *facet = facet_find(ofproto, flow);
+ return NULL;
+}
+
+/* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
+ * Returns it if found, otherwise a null pointer.
+ *
+ * The returned facet is guaranteed to be valid. */
+static struct facet *
+facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
+{
+ struct facet *facet = facet_find(ofproto, flow);
+
+ /* The facet we found might not be valid, since we could be in need of
+ * revalidation. If it is not valid, don't return it. */
+ if (facet
+ && ofproto->need_revalidate
+ && !facet_revalidate(ofproto, facet)) {
+ COVERAGE_INC(ofproto_invalidated);
+ return NULL;
+ }
+
+ return facet;
+}
+
+/* Re-searches 'ofproto''s classifier for a rule matching 'facet':
+ *
+ * - If the rule found is different from 'facet''s current rule, moves
+ * 'facet' to the new rule and recompiles its actions.
+ *
+ * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
+ * where it is and recompiles its actions anyway.
+ *
+ * - If there is none, destroys 'facet'.
+ *
+ * Returns true if 'facet' still exists, false if it has been destroyed. */
+static bool
+facet_revalidate(struct ofproto *ofproto, struct facet *facet)
+{
+ struct action_xlate_ctx ctx;
+ struct ofpbuf *odp_actions;
+ struct rule *new_rule;
+ bool actions_changed;
+
+ COVERAGE_INC(facet_revalidate);
+
+ /* Determine the new rule. */
+ new_rule = rule_lookup(ofproto, &facet->flow);
+ if (!new_rule) {
+ /* No new rule, so delete the facet. */
+ facet_remove(ofproto, facet);
+ return false;
+ }
+
+ /* Calculate new ODP actions.
+ *
+ * We do not modify any 'facet' state yet, because we might need to, e.g.,
+ * emit a NetFlow expiration and, if so, we need to have the old state
+ * around to properly compose it. */
+ action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
+ odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
+ actions_changed = (facet->actions_len != odp_actions->size
+ || memcmp(facet->actions, odp_actions->data,
+ facet->actions_len));
+
+ /* If the ODP actions changed or the installability changed, then we need
+ * to talk to the datapath. */
+ if (actions_changed || ctx.may_set_up_flow != facet->installed) {
+ if (ctx.may_set_up_flow) {
+ struct dpif_flow_stats stats;
+
+ facet_put__(ofproto, facet,
+ odp_actions->data, odp_actions->size, &stats);
+ facet_update_stats(ofproto, facet, &stats);
+ } else {
+ facet_uninstall(ofproto, facet);
+ }
+
+ /* The datapath flow is gone or has zeroed stats, so push stats out of
+ * 'facet' into 'rule'. */
+ facet_flush_stats(ofproto, facet);
+ }
+
+ /* Update 'facet' now that we've taken care of all the old state. */
+ facet->tags = ctx.tags;
+ facet->nf_flow.output_iface = ctx.nf_output_iface;
+ facet->may_install = ctx.may_set_up_flow;
+ if (actions_changed) {
+ free(facet->actions);
+ facet->actions_len = odp_actions->size;
+ facet->actions = xmemdup(odp_actions->data, odp_actions->size);
+ }
+ if (facet->rule != new_rule) {
+ COVERAGE_INC(facet_changed_rule);
+ list_remove(&facet->list_node);
+ list_push_back(&new_rule->facets, &facet->list_node);
+ facet->rule = new_rule;
+ facet->used = new_rule->created;
+ facet->rs_used = facet->used;
+ }
+
+ ofpbuf_delete(odp_actions);
+
+ return true;
+}
+\f
+/* 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);
- /* The facet we found might not be valid, since we could be in need of
- * revalidation. If it is not valid, don't return it. */
- if (facet
- && ofproto->need_revalidate
- && !facet_revalidate(ofproto, facet)) {
- COVERAGE_INC(ofproto_invalidated);
- return NULL;
+ mac->port.p = in_bundle;
+ tag_set_add(&ofproto->revalidate_set,
+ mac_learning_changed(ofproto->ml, mac));
}
-
- return facet;
}
-/* Re-searches 'ofproto''s classifier for a rule matching 'facet':
+/* 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 the rule found is different from 'facet''s current rule, moves
- * 'facet' to the new rule and recompiles its actions.
+ * 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).
*
- * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
- * where it is and recompiles its actions anyway.
+ * 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).
*
- * - If there is none, destroys 'facet'.
+ * When returning true, sets '*vlanp' to the effective VLAN of the input
+ * packet, as returned by flow_get_vlan().
*
- * Returns true if 'facet' still exists, false if it has been destroyed. */
+ * May also add tags to '*tags', although the current implementation only does
+ * so in one special case.
+ */
static bool
-facet_revalidate(struct ofproto *ofproto, struct facet *facet)
-{
- struct action_xlate_ctx ctx;
- struct ofpbuf *odp_actions;
- struct rule *new_rule;
- bool actions_changed;
+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;
+ }
- COVERAGE_INC(facet_revalidate);
+ /* Drop frames for reserved multicast addresses. */
+ if (eth_addr_is_reserved(flow->dl_dst)) {
+ return false;
+ }
- /* Determine the new rule. */
- new_rule = rule_lookup(ofproto, &facet->flow);
- if (!new_rule) {
- /* No new rule, so delete the facet. */
- facet_remove(ofproto, facet);
+ /* 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;
}
- /* Calculate new ODP actions.
- *
- * We do not modify any 'facet' state yet, because we might need to, e.g.,
- * emit a NetFlow expiration and, if so, we need to have the old state
- * around to properly compose it. */
- action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
- odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
- actions_changed = (facet->actions_len != odp_actions->size
- || memcmp(facet->actions, odp_actions->data,
- facet->actions_len));
+ if (in_bundle->bond) {
+ struct mac_entry *mac;
- /* If the ODP actions changed or the installability changed, then we need
- * to talk to the datapath. */
- if (actions_changed || ctx.may_set_up_flow != facet->installed) {
- if (ctx.may_set_up_flow) {
- struct dpif_flow_stats stats;
+ switch (bond_check_admissibility(in_bundle->bond, in_port,
+ flow->dl_dst, tags)) {
+ case BV_ACCEPT:
+ break;
- facet_put__(ofproto, facet,
- odp_actions->data, odp_actions->size, &stats);
- facet_update_stats(ofproto, facet, &stats);
- } else {
- facet_uninstall(ofproto, facet);
+ 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;
}
+ }
- /* The datapath flow is gone or has zeroed stats, so push stats out of
- * 'facet' into 'rule'. */
- facet_flush_stats(ofproto, facet);
+ 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;
}
- /* Update 'facet' now that we've taken care of all the old state. */
- facet->tags = ctx.tags;
- facet->nf_flow.output_iface = ctx.nf_output_iface;
- facet->may_install = ctx.may_set_up_flow;
- if (actions_changed) {
- free(facet->actions);
- facet->actions_len = odp_actions->size;
- facet->actions = xmemdup(odp_actions->data, odp_actions->size);
+ /* 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;
}
- if (facet->rule != new_rule) {
- COVERAGE_INC(facet_changed_rule);
- list_remove(&facet->list_node);
- list_push_back(&new_rule->facets, &facet->list_node);
- facet->rule = new_rule;
- facet->used = new_rule->created;
+
+ /* Don't send packets out their input bundles. */
+ if (in_bundle == out_bundle) {
+ out_bundle = NULL;
}
- ofpbuf_delete(odp_actions);
+done:
+ if (in_bundle) {
+ compose_actions(ctx, vlan, in_bundle, out_bundle);
+ }
return true;
}
\f
-static void
-queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
- struct rconn_packet_counter *counter)
-{
- update_openflow_length(msg);
- if (rconn_send(ofconn->rconn, msg, counter)) {
- ofpbuf_delete(msg);
- }
-}
-
static void
send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
int error)
struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
if (buf) {
COVERAGE_INC(ofproto_error);
- queue_tx(buf, ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, buf);
}
}
-static void
-hton_ofp_phy_port(struct ofp_phy_port *opp)
-{
- opp->port_no = htons(opp->port_no);
- opp->config = htonl(opp->config);
- opp->state = htonl(opp->state);
- opp->curr = htonl(opp->curr);
- opp->advertised = htonl(opp->advertised);
- opp->supported = htonl(opp->supported);
- opp->peer = htonl(opp->peer);
-}
-
static int
handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
{
- queue_tx(make_echo_reply(oh), ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, make_echo_reply(oh));
return 0;
}
static int
handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
struct ofp_switch_features *osf;
struct ofpbuf *buf;
struct ofport *port;
osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
- osf->datapath_id = htonll(ofconn->ofproto->datapath_id);
+ osf->datapath_id = htonll(ofproto->datapath_id);
osf->n_buffers = htonl(pktbuf_capacity());
osf->n_tables = 2;
osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
(1u << OFPAT_SET_TP_DST) |
(1u << OFPAT_ENQUEUE));
- HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) {
- hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
+ HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
+ ofpbuf_put(buf, &port->opp, sizeof port->opp);
}
- queue_tx(buf, ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, buf);
return 0;
}
static int
handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
struct ofpbuf *buf;
struct ofp_switch_config *osc;
uint16_t flags;
bool drop_frags;
/* Figure out flags. */
- dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags);
+ dpif_get_drop_frags(ofproto->dpif, &drop_frags);
flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
/* Send reply. */
osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
osc->flags = htons(flags);
- osc->miss_send_len = htons(ofconn->miss_send_len);
- queue_tx(buf, ofconn, ofconn->reply_counter);
+ osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
+ ofconn_send_reply(ofconn, buf);
return 0;
}
static int
handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
uint16_t flags = ntohs(osc->flags);
- if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
+ if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
+ && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
switch (flags & OFPC_FRAG_MASK) {
case OFPC_FRAG_NORMAL:
- dpif_set_drop_frags(ofconn->ofproto->dpif, false);
+ dpif_set_drop_frags(ofproto->dpif, false);
break;
case OFPC_FRAG_DROP:
- dpif_set_drop_frags(ofconn->ofproto->dpif, true);
+ dpif_set_drop_frags(ofproto->dpif, true);
break;
default:
VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
}
}
- ofconn->miss_send_len = ntohs(osc->miss_send_len);
+ ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
return 0;
}
const struct ofport *ofport = get_port(ctx->ofproto, port);
if (ofport) {
- if (ofport->opp.config & OFPPC_NO_FWD) {
+ if (ofport->opp.config & htonl(OFPPC_NO_FWD)) {
/* Forwarding disabled on port. */
return;
}
}
static void
-flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
+flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, ovs_be32 mask,
uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
{
struct ofport *ofport;
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, OFPPC_NO_FLOOD,
+ flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(OFPPC_NO_FLOOD),
&ctx->nf_output_iface, ctx->odp_actions);
break;
case OFPP_ALL:
- flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
+ flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(0),
&ctx->nf_output_iface, ctx->odp_actions);
break;
case OFPP_CONTROLLER:
}
}
+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)
const struct nx_action_set_tunnel *nast;
const struct nx_action_set_queue *nasq;
const struct nx_action_multipath *nam;
+ const struct nx_action_autopath *naa;
enum nx_action_subtype subtype = ntohs(nah->subtype);
struct xlate_reg_state state;
ovs_be64 tun_id;
multipath_execute(nam, &ctx->flow);
break;
+ case NXAST_AUTOPATH:
+ naa = (const struct nx_action_autopath *) nah;
+ xlate_autopath(ctx, naa);
+ break;
+
/* If you add a new action here that modifies flow data, don't forget to
* update the flow key in ctx->flow at the same time. */
const struct ofport *port;
port = get_port(ctx->ofproto, ctx->flow.in_port);
- if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
+ if (port && port->opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
- ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
+ ? htonl(OFPPC_NO_RECV_STP)
+ : htonl(OFPPC_NO_RECV))) {
/* Drop this flow. */
return;
}
ctx->check_special = true;
}
+static bool
+ofproto_process_special(struct ofproto *ofproto, const struct flow *flow,
+ const struct ofpbuf *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 *
xlate_actions(struct action_xlate_ctx *ctx,
const union ofp_action *in, size_t n_in)
ctx->recurse = 0;
ctx->last_pop_priority = -1;
- if (!ctx->check_special
- || !ctx->ofproto->ofhooks->special_cb
- || ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet,
- ctx->ofproto->aux)) {
- do_xlate_actions(in, n_in, ctx);
- } else {
+ 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);
}
remove_pop_action(ctx);
/* Check with in-band control to see if we're allowed to set up this
* flow. */
- if (!in_band_rule_check(ctx->ofproto->in_band, &ctx->flow,
- ctx->odp_actions->data, ctx->odp_actions->size)) {
+ if (!connmgr_may_set_up_flow(ctx->ofproto->connmgr, &ctx->flow,
+ ctx->odp_actions->data,
+ ctx->odp_actions->size)) {
ctx->may_set_up_flow = false;
}
static int
reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
{
- if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
+ if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
+ && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
msg_type);
static int
handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
{
- struct ofproto *p = ofconn->ofproto;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
struct ofp_packet_out *opo;
struct ofpbuf payload, *buffer;
union ofp_action *ofp_actions;
/* Get payload. */
if (opo->buffer_id != htonl(UINT32_MAX)) {
- error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
- &buffer, &in_port);
+ error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
+ &buffer, &in_port);
if (error || !buffer) {
return error;
}
static void
update_port_config(struct ofproto *p, struct ofport *port,
- uint32_t config, uint32_t mask)
+ ovs_be32 config, ovs_be32 mask)
{
mask &= config ^ port->opp.config;
- if (mask & OFPPC_PORT_DOWN) {
- if (config & OFPPC_PORT_DOWN) {
+ if (mask & htonl(OFPPC_PORT_DOWN)) {
+ if (config & htonl(OFPPC_PORT_DOWN)) {
netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
} else {
netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
}
#define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
OFPPC_NO_FWD | OFPPC_NO_FLOOD)
- if (mask & REVALIDATE_BITS) {
+ if (mask & htonl(REVALIDATE_BITS)) {
COVERAGE_INC(ofproto_costly_flags);
- port->opp.config ^= mask & REVALIDATE_BITS;
+ port->opp.config ^= mask & htonl(REVALIDATE_BITS);
p->need_revalidate = true;
}
#undef REVALIDATE_BITS
- if (mask & OFPPC_NO_PACKET_IN) {
- port->opp.config ^= OFPPC_NO_PACKET_IN;
+ if (mask & htonl(OFPPC_NO_PACKET_IN)) {
+ port->opp.config ^= htonl(OFPPC_NO_PACKET_IN);
}
}
static int
handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
{
- struct ofproto *p = ofconn->ofproto;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
struct ofport *port;
int error;
} else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
} else {
- update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
+ update_port_config(p, port, opm->config, opm->mask);
if (opm->advertise) {
netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
}
struct ofp_stats_reply *reply = msg->data;
reply->flags = htons(OFPSF_REPLY_MORE);
*msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
- queue_tx(msg, ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, msg);
}
return ofpbuf_put_uninit(*msgp, nbytes);
}
struct nicira_stats_msg *reply = msg->data;
reply->flags = htons(OFPSF_REPLY_MORE);
*msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
- queue_tx(msg, ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, msg);
}
ofpbuf_prealloc_tailroom(*msgp, nbytes);
}
handle_desc_stats_request(struct ofconn *ofconn,
const struct ofp_header *request)
{
- struct ofproto *p = ofconn->ofproto;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
struct ofp_desc_stats *ods;
struct ofpbuf *msg;
ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
- queue_tx(msg, ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, msg);
return 0;
}
handle_table_stats_request(struct ofconn *ofconn,
const struct ofp_header *request)
{
- struct ofproto *p = ofconn->ofproto;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
struct ofp_table_stats *ots;
struct ofpbuf *msg;
ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
memset(ots, 0, sizeof *ots);
strcpy(ots->name, "classifier");
- ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10
+ ots->wildcards = (ofconn_get_flow_format(ofconn) == NXFF_OPENFLOW10
? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
ots->active_count = htonl(classifier_count(&p->cls));
put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
- queue_tx(msg, ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, msg);
return 0;
}
netdev_get_stats(port->netdev, &stats);
ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
- ops->port_no = htons(port->opp.port_no);
+ ops->port_no = port->opp.port_no;
memset(ops->pad, 0, sizeof ops->pad);
put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
static int
handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
{
- struct ofproto *p = ofconn->ofproto;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
struct ofp_port_stats *ops;
struct ofpbuf *msg;
}
}
- queue_tx(msg, ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, msg);
return 0;
}
static void
-calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec)
+calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
{
long long int msecs = time_msec() - start;
- *sec = htonl(msecs / 1000);
- *nsec = htonl((msecs % 1000) * (1000 * 1000));
+ *sec = msecs / 1000;
+ *nsec = (msecs % 1000) * (1000 * 1000);
+}
+
+static void
+calc_flow_duration(long long int start, ovs_be32 *sec_be, ovs_be32 *nsec_be)
+{
+ uint32_t sec, nsec;
+
+ calc_flow_duration__(start, &sec, &nsec);
+ *sec_be = htonl(sec);
+ *nsec_be = htonl(nsec);
}
static void
ofs->length = htons(len);
ofs->table_id = 0;
ofs->pad = 0;
- ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match,
- rule->flow_cookie, &cookie);
+ ofputil_cls_rule_to_match(&rule->cr, ofconn_get_flow_format(ofconn),
+ &ofs->match, rule->flow_cookie, &cookie);
put_32aligned_be64(&ofs->cookie, cookie);
calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
ofs->priority = htons(rule->cr.priority);
static bool
is_valid_table(uint8_t table_id)
{
- return table_id == 0 || table_id == 0xff;
+ if (table_id == 0 || table_id == 0xff) {
+ return true;
+ } else {
+ /* It would probably be better to reply with an error but there doesn't
+ * seem to be any appropriate value, so that might just be
+ * confusing. */
+ VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
+ table_id);
+ return false;
+ }
}
static int
handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
{
const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
struct ofpbuf *reply;
COVERAGE_INC(ofproto_flows_req);
ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
&target);
- cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
+ cls_cursor_init(&cursor, &ofproto->cls, &target);
CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
}
}
- queue_tx(reply, ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, reply);
return 0;
}
static int
handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
struct nx_flow_stats_request *nfsr;
struct cls_rule target;
struct ofpbuf *reply;
struct cls_cursor cursor;
struct rule *rule;
- cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
+ cls_cursor_init(&cursor, &ofproto->cls, &target);
CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
}
}
- queue_tx(reply, ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, reply);
return 0;
}
}
/* Adds a pretty-printed description of all flows to 'results', including
- * those marked hidden by secchan (e.g., by in-band control). */
+ * hidden flows (e.g., set up by in-band control). */
void
ofproto_get_all_flows(struct ofproto *p, struct ds *results)
{
}
}
+/* Obtains the NetFlow engine type and engine ID for 'ofproto' into
+ * '*engine_type' and '*engine_id', respectively. */
+void
+ofproto_get_netflow_ids(const struct ofproto *ofproto,
+ uint8_t *engine_type, uint8_t *engine_id)
+{
+ dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
+}
+
static void
query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
ovs_be16 out_port, uint8_t table_id,
const struct ofp_header *oh)
{
const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
struct ofp_aggregate_stats_reply *reply;
struct cls_rule target;
struct ofpbuf *msg;
msg = start_ofp_stats_reply(oh, sizeof *reply);
reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
- query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
+ query_aggregate_stats(ofproto, &target, request->out_port,
request->table_id, reply);
- queue_tx(msg, ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, msg);
return 0;
}
static int
handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
struct nx_aggregate_stats_request *request;
struct ofp_aggregate_stats_reply *reply;
struct cls_rule target;
COVERAGE_INC(ofproto_flows_req);
buf = start_nxstats_reply(&request->nsm, sizeof *reply);
reply = ofpbuf_put_uninit(buf, sizeof *reply);
- query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
+ query_aggregate_stats(ofproto, &target, request->out_port,
request->table_id, reply);
- queue_tx(buf, ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, buf);
return 0;
}
struct ofp_queue_stats *reply;
reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
- reply->port_no = htons(cbdata->ofport->opp.port_no);
+ reply->port_no = cbdata->ofport->opp.port_no;
memset(reply->pad, 0, sizeof reply->pad);
reply->queue_id = htonl(queue_id);
put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
static int
handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
{
- struct ofproto *ofproto = ofconn->ofproto;
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
const struct ofp_queue_stats_request *qsr;
struct queue_stats_cbdata cbdata;
struct ofport *port;
ofpbuf_delete(cbdata.msg);
return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
}
- queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, cbdata.msg);
return 0;
}
facet_update_stats(struct ofproto *ofproto, struct facet *facet,
const struct dpif_flow_stats *stats)
{
- if (stats->n_packets) {
+ if (stats->n_packets || stats->used > facet->used) {
facet_update_time(ofproto, facet, stats->used);
facet->packet_count += stats->n_packets;
facet->byte_count += stats->n_bytes;
* in which no matching flow already exists in the flow table.
*
* Adds the flow specified by 'ofm', which is followed by 'n_actions'
- * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
+ * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
* OpenFlow error code as encoded by ofp_mkerr() on failure.
*
* 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
static int
add_flow(struct ofconn *ofconn, struct flow_mod *fm)
{
- struct ofproto *p = ofconn->ofproto;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
struct ofpbuf *packet;
struct rule *rule;
uint16_t in_port;
error = 0;
if (fm->buffer_id != UINT32_MAX) {
- error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
- &packet, &in_port);
+ error = ofconn_pktbuf_retrieve(ofconn, fm->buffer_id,
+ &packet, &in_port);
} else {
packet = NULL;
in_port = UINT16_MAX;
send_buffered_packet(struct ofconn *ofconn,
struct rule *rule, uint32_t buffer_id)
{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
struct ofpbuf *packet;
uint16_t in_port;
int error;
return 0;
}
- error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
+ error = ofconn_pktbuf_retrieve(ofconn, buffer_id, &packet, &in_port);
if (error) {
return error;
}
- rule_execute(ofconn->ofproto, rule, in_port, packet);
+ rule_execute(ofproto, rule, in_port, packet);
return 0;
}
static int
modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
{
- struct ofproto *p = ofconn->ofproto;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
struct rule *match = NULL;
struct cls_cursor cursor;
struct rule *rule;
static int
modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
{
- struct ofproto *p = ofconn->ofproto;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
struct rule *rule = find_flow_strict(p, fm);
if (rule && !rule_is_hidden(rule)) {
modify_flow(p, fm, rule);
static int
handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
{
- struct ofproto *p = ofconn->ofproto;
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
struct flow_mod fm;
int error;
return error;
}
- error = ofputil_decode_flow_mod(&fm, oh, ofconn->flow_format);
+ error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_flow_format(ofconn));
if (error) {
return error;
}
{
const struct nxt_tun_id_cookie *msg
= (const struct nxt_tun_id_cookie *) oh;
+ enum nx_flow_format flow_format;
+
+ flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
+ ofconn_set_flow_format(ofconn, flow_format);
- ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
return 0;
}
struct ofpbuf *buf;
uint32_t role;
- if (ofconn->type != OFCONN_PRIMARY) {
- VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
- "connection");
+ if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
+ VLOG_WARN_RL(&rl, "ignoring role request on service connection");
return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
}
return ofp_mkerr(OFPET_BAD_REQUEST, -1);
}
- if (role == NX_ROLE_MASTER) {
- struct ofconn *other;
-
- HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) {
- if (other->role == NX_ROLE_MASTER) {
- other->role = NX_ROLE_SLAVE;
- }
- }
- }
- ofconn->role = role;
+ ofconn_set_role(ofconn, role);
reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
reply->role = htonl(role);
- queue_tx(buf, ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, buf);
return 0;
}
if (format == NXFF_OPENFLOW10
|| format == NXFF_TUN_ID_FROM_COOKIE
|| format == NXFF_NXM) {
- ofconn->flow_format = format;
+ ofconn_set_flow_format(ofconn, format);
return 0;
} else {
return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
/* Currently, everything executes synchronously, so we can just
* immediately send the barrier reply. */
ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
- queue_tx(buf, ofconn, ofconn->reply_counter);
+ ofconn_send_reply(ofconn, buf);
return 0;
}
return 0;
/* Nicira extension requests. */
- case OFPUTIL_NXT_STATUS_REQUEST:
- return switch_status_handle_request(
- ofconn->ofproto->switch_status, ofconn->rconn, oh);
-
case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
return handle_tun_id_from_cookie(ofconn, oh);
case OFPUTIL_OFPST_PORT_REPLY:
case OFPUTIL_OFPST_TABLE_REPLY:
case OFPUTIL_OFPST_AGGREGATE_REPLY:
- case OFPUTIL_NXT_STATUS_REPLY:
case OFPUTIL_NXT_ROLE_REPLY:
case OFPUTIL_NXT_FLOW_REMOVED:
case OFPUTIL_NXST_FLOW_REPLY:
/* Set header pointers in 'flow'. */
flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
- 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;
}
/* Check with in-band control to see if this packet should be sent
* to the local port regardless of the flow table. */
- if (in_band_msg_in_hook(p->in_band, &flow, upcall->packet)) {
- struct ofpbuf odp_actions;
-
- ofpbuf_init(&odp_actions, 32);
- nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, ODPP_LOCAL);
- dpif_execute(p->dpif, odp_actions.data, odp_actions.size,
- upcall->packet);
- ofpbuf_uninit(&odp_actions);
+ if (connmgr_msg_in_hook(p->connmgr, &flow, upcall->packet)) {
+ ofproto_send_packet(p, ODPP_LOCAL, 0, upcall->packet);
}
facet = facet_lookup_valid(p, &flow);
/* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
struct ofport *port = get_port(p, flow.in_port);
if (port) {
- if (port->opp.config & OFPPC_NO_PACKET_IN) {
+ if (port->opp.config & htonl(OFPPC_NO_PACKET_IN)) {
COVERAGE_INC(ofproto_no_packet_in);
/* XXX install 'drop' flow entry */
ofpbuf_delete(upcall->packet);
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);
ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
}
}
- VLOG_INFO("%s: %s (msec:count)",
- dpif_name(ofproto->dpif), ds_cstr(&s));
+ VLOG_INFO("%s: %s (msec:count)", ofproto->name, ds_cstr(&s));
ds_destroy(&s);
}
rule_remove(ofproto, rule);
}
\f
-static struct ofpbuf *
-compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule,
- uint8_t reason)
-{
- struct ofp_flow_removed *ofr;
- struct ofpbuf *buf;
-
- ofr = make_openflow_xid(sizeof *ofr, OFPT_FLOW_REMOVED, htonl(0), &buf);
- ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match,
- rule->flow_cookie, &ofr->cookie);
- ofr->priority = htons(rule->cr.priority);
- ofr->reason = reason;
- calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec);
- ofr->idle_timeout = htons(rule->idle_timeout);
- ofr->packet_count = htonll(rule->packet_count);
- ofr->byte_count = htonll(rule->byte_count);
-
- return buf;
-}
-
-static struct ofpbuf *
-compose_nx_flow_removed(const struct rule *rule, uint8_t reason)
-{
- struct nx_flow_removed *nfr;
- struct ofpbuf *buf;
- int match_len;
-
- make_nxmsg_xid(sizeof *nfr, NXT_FLOW_REMOVED, htonl(0), &buf);
- match_len = nx_put_match(buf, &rule->cr);
-
- nfr = buf->data;
- nfr->cookie = rule->flow_cookie;
- nfr->priority = htons(rule->cr.priority);
- nfr->reason = reason;
- calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec);
- nfr->idle_timeout = htons(rule->idle_timeout);
- nfr->match_len = htons(match_len);
- nfr->packet_count = htonll(rule->packet_count);
- nfr->byte_count = htonll(rule->byte_count);
-
- return buf;
-}
-
static void
rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
{
- struct ofconn *ofconn;
+ struct ofputil_flow_removed fr;
if (!rule->send_flow_removed) {
return;
}
- LIST_FOR_EACH (ofconn, node, &p->all_conns) {
- struct ofpbuf *msg;
-
- if (!rconn_is_connected(ofconn->rconn)
- || !ofconn_receives_async_msgs(ofconn)) {
- continue;
- }
+ fr.rule = rule->cr;
+ fr.cookie = rule->flow_cookie;
+ fr.reason = reason;
+ calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
+ fr.idle_timeout = rule->idle_timeout;
+ fr.packet_count = rule->packet_count;
+ fr.byte_count = rule->byte_count;
- msg = (ofconn->flow_format == NXFF_NXM
- ? compose_nx_flow_removed(rule, reason)
- : compose_ofp_flow_removed(ofconn, rule, reason));
-
- /* Account flow expirations under ofconn->reply_counter, the counter
- * for replies to OpenFlow requests. That works because preventing
- * OpenFlow requests from being processed also prevents new flows from
- * being added (and expiring). (It also prevents processing OpenFlow
- * requests that would not add new flows, so it is imperfect.) */
- queue_tx(msg, ofconn, ofconn->reply_counter);
- }
+ connmgr_send_flow_removed(p->connmgr, &fr);
}
/* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
*bytes = b;
}
-/* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */
-static void
-do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_)
-{
- struct ofconn *ofconn = ofconn_;
-
- rconn_send_with_limit(ofconn->rconn, ofp_packet_in,
- ofconn->packet_in_counter, 100);
-}
-
-/* Takes 'upcall', whose packet has the flow specified by 'flow', composes an
- * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet
- * scheduler for sending.
- *
- * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
- * Otherwise, ownership is transferred to this function. */
-static void
-schedule_packet_in(struct ofconn *ofconn, struct dpif_upcall *upcall,
- const struct flow *flow, bool clone)
-{
- enum { OPI_SIZE = offsetof(struct ofp_packet_in, data) };
- struct ofproto *ofproto = ofconn->ofproto;
- struct ofp_packet_in *opi;
- int total_len, send_len;
- struct ofpbuf *packet;
- uint32_t buffer_id;
- int idx;
-
- /* Get OpenFlow buffer_id. */
- if (upcall->type == DPIF_UC_ACTION) {
- buffer_id = UINT32_MAX;
- } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
- buffer_id = pktbuf_get_null();
- } else if (!ofconn->pktbuf) {
- buffer_id = UINT32_MAX;
- } else {
- buffer_id = pktbuf_save(ofconn->pktbuf, upcall->packet, flow->in_port);
- }
-
- /* Figure out how much of the packet to send. */
- total_len = send_len = upcall->packet->size;
- if (buffer_id != UINT32_MAX) {
- send_len = MIN(send_len, ofconn->miss_send_len);
- }
- if (upcall->type == DPIF_UC_ACTION) {
- send_len = MIN(send_len, upcall->userdata);
- }
-
- /* Copy or steal buffer for OFPT_PACKET_IN. */
- if (clone) {
- packet = ofpbuf_clone_data_with_headroom(upcall->packet->data,
- send_len, OPI_SIZE);
- } else {
- packet = upcall->packet;
- packet->size = send_len;
- }
-
- /* Add OFPT_PACKET_IN. */
- opi = ofpbuf_push_zeros(packet, OPI_SIZE);
- opi->header.version = OFP_VERSION;
- opi->header.type = OFPT_PACKET_IN;
- opi->total_len = htons(total_len);
- opi->in_port = htons(odp_port_to_ofp_port(flow->in_port));
- opi->reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
- opi->buffer_id = htonl(buffer_id);
- update_openflow_length(packet);
-
- /* Hand over to packet scheduler. It might immediately call into
- * do_send_packet_in() or it might buffer it for a while (until a later
- * call to pinsched_run()). */
- idx = upcall->type == DPIF_UC_MISS ? 0 : 1;
- pinsched_send(ofconn->schedulers[idx], flow->in_port,
- packet, do_send_packet_in, ofconn);
-}
-
/* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
* OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
* their individual configurations.
*
- * Takes ownership of 'packet'. */
+ * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
+ * Otherwise, ownership is transferred to this function. */
static void
send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
const struct flow *flow, bool clone)
{
- struct ofconn *ofconn, *prev;
+ struct ofputil_packet_in pin;
- prev = NULL;
- LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
- if (ofconn_receives_async_msgs(ofconn)) {
- if (prev) {
- schedule_packet_in(prev, upcall, flow, true);
- }
- prev = ofconn;
- }
- }
- if (prev) {
- schedule_packet_in(prev, upcall, flow, clone);
- } else if (!clone) {
- ofpbuf_delete(upcall->packet);
- }
+ pin.packet = upcall->packet;
+ pin.in_port = odp_port_to_ofp_port(flow->in_port);
+ pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
+ pin.buffer_id = 0; /* not yet known */
+ pin.send_len = upcall->userdata;
+ connmgr_send_packet_in(ofproto->connmgr, upcall, flow,
+ clone ? NULL : upcall->packet);
}
static uint64_t
return eth_addr_to_uint64(ea);
}
\f
+static struct ofproto *
+ofproto_lookup(const char *name)
+{
+ struct ofproto *ofproto;
+
+ HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
+ &all_ofprotos) {
+ if (!strcmp(ofproto->name, name)) {
+ return ofproto;
+ }
+ }
+ return NULL;
+}
+
static void
ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
void *aux OVS_UNUSED)
{
- const struct shash_node *node;
+ struct ofproto *ofproto;
struct ds results;
ds_init(&results);
- SHASH_FOR_EACH (node, &all_ofprotos) {
- ds_put_format(&results, "%s\n", node->name);
+ HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
+ ds_put_format(&results, "%s\n", ofproto->name);
}
unixctl_command_reply(conn, 200, ds_cstr(&results));
ds_destroy(&results);
goto exit;
}
- ofproto = shash_find_data(&all_ofprotos, dpname);
+ ofproto = ofproto_lookup(dpname);
if (!ofproto) {
unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
"for help)");
goto exit;
}
- tun_id = htonll(strtoull(tun_id_s, NULL, 10));
+ tun_id = htonll(strtoull(tun_id_s, NULL, 0));
in_port = ofp_port_to_odp_port(atoi(in_port_s));
packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
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)
{
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);
}
-\f
-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_;
- int out_port;
-
- /* 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) {
- tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
- 0, flow->in_port,
- GRAT_ARP_LOCK_NONE);
- if (rev_tag) {
- /* 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, rev_tag);
- }
- }
-
- /* Determine output port. */
- out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
- NULL);
- if (out_port < 0) {
- flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
- nf_output_iface, odp_actions);
- } else 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
-};