/*
* Copyright (c) 2009, 2010 Nicira Networks.
+ * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#include "ofproto.h"
#include <errno.h>
#include <inttypes.h>
+#include <sys/socket.h>
#include <net/if.h>
#include <netinet/in.h>
#include <stdbool.h>
#include "netflow.h"
#include "odp-util.h"
#include "ofp-print.h"
+#include "ofp-util.h"
#include "ofproto-sflow.h"
#include "ofpbuf.h"
#include "openflow/nicira-ext.h"
#include "rconn.h"
#include "shash.h"
#include "status.h"
-#include "stp.h"
#include "stream-ssl.h"
#include "svec.h"
#include "tag.h"
#include "timeval.h"
#include "unixctl.h"
#include "vconn.h"
+#include "vlog.h"
#include "xtoxll.h"
-#define THIS_MODULE VLM_ofproto
-#include "vlog.h"
+VLOG_DEFINE_THIS_MODULE(ofproto)
#include "sflow_api.h"
struct rule {
struct cls_rule cr;
- uint64_t flow_cookie; /* Controller-issued identifier.
+ uint64_t flow_cookie; /* Controller-issued identifier.
(Kept in network-byte order.) */
uint16_t idle_timeout; /* In seconds from time of last use. */
uint16_t hard_timeout; /* In seconds from time of creation. */
/* ofproto supports two kinds of OpenFlow connections:
*
- * - "Controller connections": Connections to ordinary OpenFlow controllers.
- * ofproto maintains persistent connections to these controllers and by
- * default sends them asynchronous messages such as packet-ins.
+ * - "Primary" connections to ordinary OpenFlow controllers. ofproto
+ * maintains persistent connections to these controllers and by default
+ * sends them asynchronous messages such as packet-ins.
*
- * - "Transient connections", e.g. from ovs-ofctl. When these connections
+ * - "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_CONTROLLER, /* An OpenFlow controller. */
- OFCONN_TRANSIENT /* A transient connection. */
+ 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. */
#define OFCONN_REPLY_MAX 100
struct rconn_packet_counter *reply_counter;
- /* type == OFCONN_CONTROLLER only. */
+ /* 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. */
static void ofconn_destroy(struct ofconn *);
static void ofconn_run(struct ofconn *, struct ofproto *);
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);
/* 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;
+
/* Flow table. */
struct classifier cls;
bool need_revalidate;
/* OpenFlow connections. */
struct hmap controllers; /* Controller "struct ofconn"s. */
struct list all_conns; /* Contains "struct ofconn"s. */
- struct pvconn **listeners;
- size_t n_listeners;
+ enum ofproto_fail_mode fail_mode;
+
+ /* OpenFlow listeners. */
+ struct hmap services; /* Contains "struct ofservice"s. */
struct pvconn **snoops;
size_t n_snoops;
/* Initialize OpenFlow connections. */
list_init(&p->all_conns);
hmap_init(&p->controllers);
- p->listeners = NULL;
- p->n_listeners = 0;
+ hmap_init(&p->services);
p->snoops = NULL;
p->n_snoops = 0;
uint64_t old_dpid = p->datapath_id;
p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
if (p->datapath_id != old_dpid) {
- struct ofconn *ofconn;
-
VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
/* Force all active connections to reconnect, since there is no way to
* notify a controller that the datapath ID has changed. */
- LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
- rconn_reconnect(ofconn->rconn);
- }
+ ofproto_reconnect_controllers(p);
}
}
discovery = NULL;
}
- ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_CONTROLLER);
+ 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 {
- rconn_connect(ofconn->rconn, c->target);
+ 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));
static void
update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
{
- struct ofproto *ofproto = ofconn->ofproto;
int probe_interval;
- int i;
ofconn->band = (is_in_band_controller(c)
? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
}
- for (i = 0; i < N_SCHEDULERS; i++) {
- struct pinsched **s = &ofconn->schedulers[i];
-
- if (c->rate_limit > 0) {
- if (!*s) {
- *s = pinsched_create(c->rate_limit, c->burst_limit,
- ofproto->switch_status);
- } else {
- pinsched_set_limits(*s, c->rate_limit, c->burst_limit);
- }
- } else {
- pinsched_destroy(*s);
- *s = NULL;
- }
- }
+ 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_name(ofconn->rconn);
+ return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
}
static struct ofconn *
{
const struct ofconn *ofconn;
struct sockaddr_in *addrs;
+ size_t max_addrs, n_addrs;
bool discovery;
- size_t n_addrs;
-
+ size_t i;
- addrs = xmalloc(hmap_count(&ofproto->controllers) * sizeof *addrs);
+ /* 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. */
HMAP_FOR_EACH (ofconn, struct 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);
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.
*
in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
&ofproto->in_band);
}
- in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
+ if (ofproto->in_band) {
+ in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
+ }
ofproto->next_in_band_update = time_msec() + 1000;
} else {
in_band_destroy(ofproto->in_band);
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, struct 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;
- enum ofproto_fail_mode fail_mode;
- struct ofconn *ofconn, *next;
+ 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];
- shash_add_once(&new_controllers, c->target, &controllers[i]);
- if (!find_controller_by_target(p, c->target)) {
- add_controller(p, c);
+ 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]);
}
- fail_mode = OFPROTO_FAIL_STANDALONE;
+ /* Delete controllers that are no longer configured.
+ * Update configuration of all now-existing controllers. */
ss_exists = false;
- HMAP_FOR_EACH_SAFE (ofconn, next, struct ofconn, hmap_node,
+ HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, hmap_node,
&p->controllers) {
struct ofproto_controller *c;
if (ofconn->ss) {
ss_exists = true;
}
- if (c->fail == OFPROTO_FAIL_SECURE) {
- fail_mode = OFPROTO_FAIL_SECURE;
- }
}
}
- shash_destroy(&new_controllers);
-
- update_in_band_remotes(p);
- if (!hmap_is_empty(&p->controllers)
- && fail_mode == OFPROTO_FAIL_STANDALONE) {
- struct rconn **rconns;
- size_t n;
+ /* Delete services that are no longer configured.
+ * Update configuration of all now-existing services. */
+ HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, struct ofservice, node,
+ &p->services) {
+ struct ofproto_controller *c;
- if (!p->fail_open) {
- p->fail_open = fail_open_create(p, p->switch_status);
+ c = shash_find_data(&new_controllers,
+ pvconn_get_name(ofservice->pvconn));
+ if (!c) {
+ ofservice_destroy(p, ofservice);
+ } else {
+ ofservice_reconfigure(ofservice, c);
}
+ }
- n = 0;
- rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
- HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) {
- rconns[n++] = ofconn->rconn;
- }
+ shash_destroy(&new_controllers);
- 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;
- }
+ update_in_band_remotes(p);
+ update_fail_open(p);
if (!hmap_is_empty(&p->controllers) && !ss_exists) {
ofconn = CONTAINER_OF(hmap_first(&p->controllers),
}
}
+void
+ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
+{
+ p->fail_mode = fail_mode;
+ update_fail_open(p);
+}
+
+/* Drops the connections between 'ofproto' and all of its controllers, forcing
+ * them to reconnect. */
+void
+ofproto_reconnect_controllers(struct ofproto *ofproto)
+{
+ struct ofconn *ofconn;
+
+ LIST_FOR_EACH (ofconn, struct 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;
+}
+
+/* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
+ * in-band control should guarantee access, in the same way that in-band
+ * control guarantees access to OpenFlow controllers. */
+void
+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);
+}
+
void
ofproto_set_desc(struct ofproto *p,
const char *mfr_desc, const char *hw_desc,
return retval;
}
-int
-ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
-{
- return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
-}
-
int
ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
{
}
}
-int
-ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
-{
- /* XXX */
- if (enable_stp) {
- VLOG_WARN("STP is not yet implemented");
- return EINVAL;
- } else {
- return 0;
- }
-}
-
uint64_t
ofproto_get_datapath_id(const struct ofproto *ofproto)
{
}
bool
-ofproto_has_controller(const struct ofproto *ofproto)
+ofproto_has_primary_controller(const struct ofproto *ofproto)
{
return !hmap_is_empty(&ofproto->controllers);
}
-void
-ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
+enum ofproto_fail_mode
+ofproto_get_fail_mode(const struct ofproto *p)
{
- size_t i;
-
- for (i = 0; i < ofproto->n_listeners; i++) {
- svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
- }
+ return p->fail_mode;
}
void
void
ofproto_destroy(struct ofproto *p)
{
+ struct ofservice *ofservice, *next_ofservice;
struct ofconn *ofconn, *next_ofconn;
struct ofport *ofport;
unsigned int port_no;
in_band_destroy(p->in_band);
p->in_band = NULL;
+ free(p->extra_in_band_remotes);
ofproto_flush_flows(p);
classifier_destroy(&p->cls);
netflow_destroy(p->netflow);
ofproto_sflow_destroy(p->sflow);
- for (i = 0; i < p->n_listeners; i++) {
- pvconn_close(p->listeners[i]);
+ HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, struct ofservice, node,
+ &p->services) {
+ ofservice_destroy(p, ofservice);
}
- free(p->listeners);
+ hmap_destroy(&p->services);
for (i = 0; i < p->n_snoops; i++) {
pvconn_close(p->snoops[i]);
}
}
+/* 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 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;
+ }
+}
+
/* 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)
{
- struct ofconn *ofconn;
+ struct ofconn *ofconn, *best;
- /* Arbitrarily pick the first controller in the list for monitoring. We
- * could do something smarter or more flexible later, if it ever proves
- * useful. */
+ /* Pick a controller for monitoring. */
+ best = NULL;
LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
- if (ofconn->type == OFCONN_CONTROLLER) {
- rconn_add_monitor(ofconn->rconn, vconn);
- return;
+ if (ofconn->type == OFCONN_PRIMARY
+ && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
+ best = ofconn;
}
+ }
+ if (best) {
+ rconn_add_monitor(best->rconn, vconn);
+ } else {
+ VLOG_INFO_RL(&rl, "no controller connection to snoop");
+ vconn_close(vconn);
}
- VLOG_INFO_RL(&rl, "no controller connection to monitor");
- vconn_close(vconn);
}
int
ofproto_run1(struct ofproto *p)
{
struct ofconn *ofconn, *next_ofconn;
+ struct ofservice *ofservice;
char *devname;
int error;
int i;
fail_open_run(p->fail_open);
}
- for (i = 0; i < p->n_listeners; i++) {
+ HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) {
struct vconn *vconn;
int retval;
- retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
+ retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
if (!retval) {
- ofconn_create(p, rconn_new_from_vconn("passive", vconn),
- OFCONN_TRANSIENT);
+ struct ofconn *ofconn;
+ 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);
+
+ 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));
}
void
ofproto_wait(struct ofproto *p)
{
+ struct ofservice *ofservice;
struct ofconn *ofconn;
size_t i;
ofconn_wait(ofconn);
}
if (p->in_band) {
- poll_timer_wait(p->next_in_band_update - time_msec());
+ poll_timer_wait_until(p->next_in_band_update);
in_band_wait(p->in_band);
}
if (p->fail_open) {
VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
poll_immediate_wake();
} else if (p->next_expiration != LLONG_MAX) {
- poll_timer_wait(p->next_expiration - time_msec());
+ poll_timer_wait_until(p->next_expiration);
}
- for (i = 0; i < p->n_listeners; i++) {
- pvconn_wait(p->listeners[i]);
+ HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) {
+ pvconn_wait(ofservice->pvconn);
}
for (i = 0; i < p->n_snoops; i++) {
pvconn_wait(p->snoops[i]);
{
struct rule *rule;
rule = rule_create(p, NULL, actions, n_actions,
- idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
+ idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
0, 0, false);
cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
rule_insert(p, rule, NULL, 0);
memset(&netdev_options, 0, sizeof netdev_options);
netdev_options.name = odp_port->devname;
netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
- netdev_options.may_open = true;
error = netdev_open(&netdev_options, &netdev);
if (error) {
struct ofp_port_status *ops;
struct ofpbuf *b;
- if (ofconn->role == NX_ROLE_SLAVE) {
+ if (!ofconn_receives_async_msgs(ofconn)) {
continue;
}
uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
- port_array_set(&p->ports, odp_port, NULL);
+ port_array_delete(&p->ports, odp_port);
shash_delete(&p->port_by_name,
shash_find(&p->port_by_name, (char *) ofport->opp.name));
if (p->sflow) {
static void
ofconn_destroy(struct ofconn *ofconn)
{
- if (ofconn->type == OFCONN_CONTROLLER) {
+ if (ofconn->type == OFCONN_PRIMARY) {
hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
}
discovery_destroy(ofconn->discovery);
}
if (discovery_run(ofconn->discovery, &controller_name)) {
if (controller_name) {
- rconn_connect(ofconn->rconn, 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);
}
COVERAGE_INC(ofproto_ofconn_stuck);
}
}
+
+/* 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;
+ }
+}
+
+/* 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)
+{
+ return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
+}
+
+static void
+ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
+{
+ int i;
+
+ for (i = 0; i < N_SCHEDULERS; i++) {
+ struct pinsched **s = &ofconn->schedulers[i];
+
+ if (rate > 0) {
+ if (!*s) {
+ *s = pinsched_create(rate, burst,
+ ofconn->ofproto->switch_status);
+ } else {
+ pinsched_set_limits(*s, rate, burst);
+ }
+ } else {
+ pinsched_destroy(*s);
+ *s = NULL;
+ }
+ }
+}
+\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;
+}
+
+/* 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)
+{
+ struct ofservice *ofservice;
+
+ HMAP_FOR_EACH_WITH_HASH (ofservice, struct ofservice, node,
+ hash_string(target, 0), &ofproto->services) {
+ if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
+ return ofservice;
+ }
+ }
+ return NULL;
+}
\f
/* Caller is responsible for initializing the 'cr' member of the returned
* rule. */
}
for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
oa = actions_next(&i)) {
- if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
+ if (action_outputs_to_port(oa, out_port)) {
return true;
}
}
return false;
}
+/* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
+ * 'packet', which arrived on 'in_port'.
+ *
+ * Takes ownership of 'packet'. */
+static bool
+execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
+ const union odp_action *actions, size_t n_actions,
+ struct ofpbuf *packet)
+{
+ if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) {
+ /* As an optimization, avoid a round-trip from userspace to kernel to
+ * userspace. This also avoids possibly filling up kernel packet
+ * buffers along the way. */
+ struct odp_msg *msg;
+
+ msg = ofpbuf_push_uninit(packet, sizeof *msg);
+ msg->type = _ODPL_ACTION_NR;
+ msg->length = sizeof(struct odp_msg) + packet->size;
+ msg->port = in_port;
+ msg->reserved = 0;
+ msg->arg = actions[0].controller.arg;
+
+ send_packet_in(ofproto, packet);
+
+ return true;
+ } else {
+ int error;
+
+ error = dpif_execute(ofproto->dpif, in_port,
+ actions, n_actions, packet);
+ ofpbuf_delete(packet);
+ return !error;
+ }
+}
+
/* Executes the actions indicated by 'rule' on 'packet', which is in flow
- * 'flow' and is considered to have arrived on ODP port 'in_port'.
+ * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet'
+ * must have at least sizeof(struct ofp_packet_in) bytes of headroom.
*
* The flow that 'packet' actually contains does not need to actually match
* 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
* 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
* 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
* function will compose a set of ODP actions based on 'rule''s OpenFlow
- * actions and apply them to 'packet'. */
+ * actions and apply them to 'packet'.
+ *
+ * Takes ownership of 'packet'. */
static void
rule_execute(struct ofproto *ofproto, struct rule *rule,
struct ofpbuf *packet, const flow_t *flow)
{
const union odp_action *actions;
+ struct odp_flow_stats stats;
size_t n_actions;
struct odp_actions a;
+ assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
+
/* Grab or compose the ODP actions.
*
* The special case for an exact-match 'rule' where 'flow' is not the
struct rule *super = rule->super ? rule->super : rule;
if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
packet, &a, NULL, 0, NULL)) {
+ ofpbuf_delete(packet);
return;
}
actions = a.actions;
}
/* Execute the ODP actions. */
- if (!dpif_execute(ofproto->dpif, flow->in_port,
- actions, n_actions, packet)) {
- struct odp_flow_stats stats;
- flow_extract_stats(flow, packet, &stats);
+ flow_extract_stats(flow, packet, &stats);
+ if (execute_odp_actions(ofproto, flow->in_port,
+ actions, n_actions, packet)) {
update_stats(ofproto, rule, &stats);
rule->used = time_msec();
netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
}
}
+/* Inserts 'rule' into 'p''s flow table.
+ *
+ * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s
+ * actions on it and credits the statistics for sending the packet to 'rule'.
+ * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of
+ * headroom. */
static void
rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
uint16_t in_port)
&& total_bytes > rule->accounted_bytes)
{
ofproto->ofhooks->account_flow_cb(
- &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
+ &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions,
total_bytes - rule->accounted_bytes, ofproto->aux);
rule->accounted_bytes = total_bytes;
}
* NetFlow expiration messages since it is just part of the control
* logic for the network and not real traffic. */
- if (rule && rule->super) {
- struct rule *super = rule->super;
-
- return super->n_actions == 1 &&
- super->actions[0].type == htons(OFPAT_OUTPUT) &&
- super->actions[0].output.port == htons(OFPP_CONTROLLER);
- }
-
- return false;
+ return (rule
+ && rule->super
+ && rule->super->n_actions == 1
+ && action_outputs_to_port(&rule->super->actions[0],
+ htons(OFPP_CONTROLLER)));
}
static void
(1u << OFPAT_SET_NW_DST) |
(1u << OFPAT_SET_NW_TOS) |
(1u << OFPAT_SET_TP_SRC) |
- (1u << OFPAT_SET_TP_DST));
+ (1u << OFPAT_SET_TP_DST) |
+ (1u << OFPAT_ENQUEUE));
PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
}
flags = ntohs(osc->flags);
- if (ofconn->type == OFCONN_CONTROLLER && ofconn->role != NX_ROLE_SLAVE) {
+ if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
switch (flags & OFPC_FRAG_MASK) {
case OFPC_FRAG_NORMAL:
dpif_set_drop_frags(p->dpif, false);
}
static void
-add_controller_action(struct odp_actions *actions,
- const struct ofp_action_output *oao)
+add_controller_action(struct odp_actions *actions, uint16_t max_len)
{
union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
- a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
+ a->controller.arg = max_len;
}
struct action_xlate_ctx {
uint16_t nf_output_iface; /* Output interface index for NetFlow. */
};
+/* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
+ * flow translation. */
+#define MAX_RESUBMIT_RECURSION 8
+
static void do_xlate_actions(const union ofp_action *in, size_t n_in,
struct action_xlate_ctx *ctx);
static void
xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
{
- if (!ctx->recurse) {
+ if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
uint16_t old_in_port;
struct rule *rule;
do_xlate_actions(rule->actions, rule->n_actions, ctx);
ctx->recurse--;
}
+ } else {
+ struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
+
+ VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
+ MAX_RESUBMIT_RECURSION);
}
}
static void
-xlate_output_action(struct action_xlate_ctx *ctx,
- const struct ofp_action_output *oao)
+xlate_output_action__(struct action_xlate_ctx *ctx,
+ uint16_t port, uint16_t max_len)
{
uint16_t odp_port;
uint16_t prev_nf_output_iface = ctx->nf_output_iface;
ctx->nf_output_iface = NF_OUT_DROP;
- switch (ntohs(oao->port)) {
+ switch (port) {
case OFPP_IN_PORT:
add_output_action(ctx, ctx->flow.in_port);
break;
add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
break;
case OFPP_CONTROLLER:
- add_controller_action(ctx->out, oao);
+ add_controller_action(ctx->out, max_len);
break;
case OFPP_LOCAL:
add_output_action(ctx, ODPP_LOCAL);
break;
default:
- odp_port = ofp_port_to_odp_port(ntohs(oao->port));
+ odp_port = ofp_port_to_odp_port(port);
if (odp_port != ctx->flow.in_port) {
add_output_action(ctx, odp_port);
}
}
}
+static void
+xlate_output_action(struct action_xlate_ctx *ctx,
+ const struct ofp_action_output *oao)
+{
+ xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
+}
+
+/* If the final ODP action in 'ctx' is "pop priority", drop it, as an
+ * optimization, because we're going to add another action that sets the
+ * priority immediately after, or because there are no actions following the
+ * pop. */
+static void
+remove_pop_action(struct action_xlate_ctx *ctx)
+{
+ size_t n = ctx->out->n_actions;
+ if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
+ ctx->out->n_actions--;
+ }
+}
+
+static void
+xlate_enqueue_action(struct action_xlate_ctx *ctx,
+ const struct ofp_action_enqueue *oae)
+{
+ uint16_t ofp_port, odp_port;
+ uint32_t priority;
+ int error;
+
+ error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
+ &priority);
+ if (error) {
+ /* Fall back to ordinary output action. */
+ xlate_output_action__(ctx, ntohs(oae->port), 0);
+ return;
+ }
+
+ /* Figure out ODP output port. */
+ ofp_port = ntohs(oae->port);
+ if (ofp_port != OFPP_IN_PORT) {
+ odp_port = ofp_port_to_odp_port(ofp_port);
+ } else {
+ odp_port = ctx->flow.in_port;
+ }
+
+ /* Add ODP actions. */
+ remove_pop_action(ctx);
+ odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
+ = priority;
+ add_output_action(ctx, odp_port);
+ odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
+
+ /* Update NetFlow output port. */
+ if (ctx->nf_output_iface == NF_OUT_DROP) {
+ ctx->nf_output_iface = odp_port;
+ } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
+ ctx->nf_output_iface = NF_OUT_MULTI;
+ }
+}
+
static void
xlate_nicira_action(struct action_xlate_ctx *ctx,
const struct nx_action_header *nah)
ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
break;
+ case NXAST_DROP_SPOOFED_ARP:
+ if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
+ odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP);
+ }
+ break;
+
/* If you add a new action here that modifies flow data, don't forget to
- * update the flow key in ctx->flow in the same key. */
+ * update the flow key in ctx->flow at the same time. */
default:
VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
- port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr)
+ port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
/* Drop this flow. */
return;
case OFPAT_STRIP_VLAN:
odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
- ctx->flow.dl_vlan = OFP_VLAN_NONE;
+ ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
ctx->flow.dl_vlan_pcp = 0;
break;
xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
break;
+ case OFPAT_ENQUEUE:
+ xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
+ break;
+
default:
VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
break;
ctx.may_set_up_flow = true;
ctx.nf_output_iface = NF_OUT_DROP;
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. */
*nf_output_iface = ctx.nf_output_iface;
}
if (odp_actions_overflow(out)) {
+ COVERAGE_INC(odp_overflow);
odp_actions_init(out);
return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
}
static int
reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
{
- if (ofconn->type == OFCONN_CONTROLLER && ofconn->role == NX_ROLE_SLAVE) {
+ if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
char *type_name;
}
static void
-append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
- struct ofpbuf *msg)
+append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
+ struct ofpbuf **msgp)
{
struct netdev_stats stats;
struct ofp_port_stats *ops;
- /* Intentionally ignore return value, since errors will set
- * 'stats' to all-1s, which is correct for OpenFlow, and
+ /* Intentionally ignore return value, since errors will set
+ * 'stats' to all-1s, which is correct for OpenFlow, and
* netdev_get_stats() will log errors. */
netdev_get_stats(port->netdev, &stats);
- ops = append_stats_reply(sizeof *ops, ofconn, &msg);
+ ops = append_stats_reply(sizeof *ops, ofconn, msgp);
ops->port_no = htons(odp_port_to_ofp_port(port_no));
memset(ops->pad, 0, sizeof ops->pad);
ops->rx_packets = htonll(stats.rx_packets);
msg = start_stats_reply(osr, sizeof *ops * 16);
if (psr->port_no != htons(OFPP_NONE)) {
- port = port_array_get(&p->ports,
+ port = port_array_get(&p->ports,
ofp_port_to_odp_port(ntohs(psr->port_no)));
if (port) {
- append_port_stat(port, ntohs(psr->port_no), ofconn, msg);
+ append_port_stat(port, ntohs(psr->port_no), ofconn, &msg);
}
} else {
PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
- append_port_stat(port, port_no, ofconn, msg);
+ append_port_stat(port, port_no, ofconn, &msg);
}
}
ds_put_cstr(results, "\n");
}
-/* Adds a pretty-printed description of all flows to 'results', including
+/* Adds a pretty-printed description of all flows to 'results', including
* those marked hidden by secchan (e.g., by in-band control). */
void
ofproto_get_all_flows(struct ofproto *p, struct ds *results)
return 0;
}
+struct queue_stats_cbdata {
+ struct ofconn *ofconn;
+ struct ofpbuf *msg;
+ uint16_t port_no;
+};
+
+static void
+put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
+ const struct netdev_queue_stats *stats)
+{
+ struct ofp_queue_stats *reply;
+
+ reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
+ reply->port_no = htons(cbdata->port_no);
+ memset(reply->pad, 0, sizeof reply->pad);
+ reply->queue_id = htonl(queue_id);
+ reply->tx_bytes = htonll(stats->tx_bytes);
+ reply->tx_packets = htonll(stats->tx_packets);
+ reply->tx_errors = htonll(stats->tx_errors);
+}
+
+static void
+handle_queue_stats_dump_cb(uint32_t queue_id,
+ struct netdev_queue_stats *stats,
+ void *cbdata_)
+{
+ struct queue_stats_cbdata *cbdata = cbdata_;
+
+ put_queue_stats(cbdata, queue_id, stats);
+}
+
+static void
+handle_queue_stats_for_port(struct ofport *port, uint16_t port_no,
+ uint32_t queue_id,
+ struct queue_stats_cbdata *cbdata)
+{
+ cbdata->port_no = port_no;
+ if (queue_id == OFPQ_ALL) {
+ netdev_dump_queue_stats(port->netdev,
+ handle_queue_stats_dump_cb, cbdata);
+ } else {
+ struct netdev_queue_stats stats;
+
+ netdev_get_queue_stats(port->netdev, queue_id, &stats);
+ put_queue_stats(cbdata, queue_id, &stats);
+ }
+}
+
+static int
+handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn,
+ const struct ofp_stats_request *osr,
+ size_t arg_size)
+{
+ struct ofp_queue_stats_request *qsr;
+ struct queue_stats_cbdata cbdata;
+ struct ofport *port;
+ unsigned int port_no;
+ uint32_t queue_id;
+
+ if (arg_size != sizeof *qsr) {
+ return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
+ }
+ qsr = (struct ofp_queue_stats_request *) osr->body;
+
+ COVERAGE_INC(ofproto_queue_req);
+
+ cbdata.ofconn = ofconn;
+ cbdata.msg = start_stats_reply(osr, 128);
+
+ port_no = ntohs(qsr->port_no);
+ queue_id = ntohl(qsr->queue_id);
+ if (port_no == OFPP_ALL) {
+ PORT_ARRAY_FOR_EACH (port, &ofproto->ports, port_no) {
+ handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
+ }
+ } else if (port_no < ofproto->max_ports) {
+ port = port_array_get(&ofproto->ports, port_no);
+ if (port) {
+ handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
+ }
+ } else {
+ ofpbuf_delete(cbdata.msg);
+ return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
+ }
+ queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
+
+ return 0;
+}
+
static int
handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
struct ofp_header *oh)
case OFPST_PORT:
return handle_port_stats_request(p, ofconn, osr, arg_size);
+ case OFPST_QUEUE:
+ return handle_queue_stats_request(p, ofconn, osr, arg_size);
+
case OFPST_VENDOR:
return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
update_time(ofproto, rule, stats);
rule->packet_count += stats->n_packets;
rule->byte_count += stats->n_bytes;
- netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
- stats->tcp_flags);
+ netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags);
}
}
}
rule_insert(p, rule, packet, in_port);
- ofpbuf_delete(packet);
return error;
}
flow_extract(packet, 0, in_port, &flow);
rule_execute(ofproto, rule, packet, &flow);
- ofpbuf_delete(packet);
return 0;
}
handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
struct ofp_flow_mod *ofm)
{
+ struct ofp_match orig_match;
size_t n_actions;
int error;
return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
}
+ /* Normalize ofp->match. If normalization actually changes anything, then
+ * log the differences. */
+ ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
+ orig_match = ofm->match;
normalize_match(&ofm->match);
+ if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
+ static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
+ if (!VLOG_DROP_INFO(&normal_rl)) {
+ char *old = ofp_match_to_literal_string(&orig_match);
+ char *new = ofp_match_to_literal_string(&ofm->match);
+ VLOG_INFO("%s: normalization changed ofp_match, details:",
+ rconn_get_name(ofconn->rconn));
+ VLOG_INFO(" pre: %s", old);
+ VLOG_INFO("post: %s", new);
+ free(old);
+ free(new);
+ }
+ }
+
if (!ofm->match.wildcards) {
ofm->priority = htons(UINT16_MAX);
}
uint32_t role;
if (ntohs(msg->header.length) != sizeof *nrr) {
- VLOG_WARN_RL(&rl, "received role request of length %zu (expected %zu)",
+ VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
ntohs(msg->header.length), sizeof *nrr);
return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
}
nrr = (struct nx_role_request *) msg;
- if (ofconn->type != OFCONN_CONTROLLER) {
+ if (ofconn->type != OFCONN_PRIMARY) {
VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
"connection");
return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
struct nicira_header *nh;
if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
- VLOG_WARN_RL(&rl, "received vendor message of length %zu "
+ VLOG_WARN_RL(&rl, "received vendor message of length %u "
"(expected at least %zu)",
ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
}
if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
- VLOG_WARN_RL(&rl, "received Nicira vendor message of length %zu "
+ VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
"(expected at least %zu)",
ntohs(ovh->header.length), sizeof(struct nicira_header));
return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
}
}
- rule_execute(p, rule, &payload, &flow);
- rule_reinstall(p, rule);
-
if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
/*
* Extra-special case for fail-open mode.
*
* See the top-level comment in fail-open.c for more information.
*/
- send_packet_in(p, packet);
- } else {
- ofpbuf_delete(packet);
+ send_packet_in(p, ofpbuf_clone(packet));
}
+
+ ofpbuf_pull(packet, sizeof *msg);
+ rule_execute(p, rule, packet, &flow);
+ rule_reinstall(p, rule);
}
static void
prev = NULL;
LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
- && ofconn->role != NX_ROLE_SLAVE) {
+ && ofconn_receives_async_msgs(ofconn)) {
if (prev) {
queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
} else {
if (odp_flow.stats.n_packets) {
update_time(ofproto, rule, &odp_flow.stats);
- netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
+ netflow_flow_update_flags(&rule->nf_flow,
odp_flow.stats.tcp_flags);
}
}
free(flows);
}
+/* pinsched callback for sending 'packet' on 'ofconn'. */
static void
do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
{
struct ofconn *ofconn = ofconn_;
+
+ rconn_send_with_limit(ofconn->rconn, packet,
+ ofconn->packet_in_counter, 100);
+}
+
+/* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
+ * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
+ * packet scheduler for sending.
+ *
+ * 'max_len' specifies the maximum number of bytes of the packet to send on
+ * 'ofconn' (INT_MAX specifies no limit).
+ *
+ * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
+ * ownership is transferred to this function. */
+static void
+schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
+ bool clone)
+{
struct ofproto *ofproto = ofconn->ofproto;
- struct odp_msg *msg = packet->data;
- struct ofpbuf payload;
- struct ofpbuf *opi;
+ struct ofp_packet_in *opi = packet->data;
+ uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
+ int send_len, trim_size;
uint32_t buffer_id;
- int send_len;
- /* Extract packet payload from 'msg'. */
- payload.data = msg + 1;
- payload.size = msg->length - sizeof *msg;
+ /* Get buffer. */
+ if (opi->reason == OFPR_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 {
+ struct ofpbuf payload;
+ payload.data = opi->data;
+ payload.size = packet->size - offsetof(struct ofp_packet_in, data);
+ buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
+ }
- /* Construct packet-in message. */
- send_len = INT_MAX;
+ /* Figure out how much of the packet to send. */
+ send_len = ntohs(opi->total_len);
+ if (buffer_id != UINT32_MAX) {
+ send_len = MIN(send_len, ofconn->miss_send_len);
+ }
+ send_len = MIN(send_len, max_len);
+
+ /* Adjust packet length and clone if necessary. */
+ trim_size = offsetof(struct ofp_packet_in, data) + send_len;
+ if (clone) {
+ packet = ofpbuf_clone_data(packet->data, trim_size);
+ opi = packet->data;
+ } else {
+ packet->size = trim_size;
+ }
+
+ /* Update packet headers. */
+ 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()). */
+ pinsched_send(ofconn->schedulers[opi->reason], in_port,
+ packet, do_send_packet_in, ofconn);
+}
+
+/* Replace struct odp_msg header in 'packet' by equivalent struct
+ * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
+ * returned by dpif_recv()).
+ *
+ * The conversion is not complete: the caller still needs to trim any unneeded
+ * payload off the end of the buffer, set the length in the OpenFlow header,
+ * and set buffer_id. Those require us to know the controller settings and so
+ * must be done on a per-controller basis.
+ *
+ * Returns the maximum number of bytes of the packet that should be sent to
+ * the controller (INT_MAX if no limit). */
+static int
+do_convert_to_packet_in(struct ofpbuf *packet)
+{
+ struct odp_msg *msg = packet->data;
+ struct ofp_packet_in *opi;
+ uint8_t reason;
+ uint16_t total_len;
+ uint16_t in_port;
+ int max_len;
+
+ /* Extract relevant header fields */
if (msg->type == _ODPL_ACTION_NR) {
- buffer_id = UINT32_MAX;
+ reason = OFPR_ACTION;
+ max_len = msg->arg;
} else {
- if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
- buffer_id = pktbuf_get_null();
- } else {
- buffer_id = pktbuf_save(ofconn->pktbuf, &payload, msg->port);
- }
- if (buffer_id != UINT32_MAX) {
- send_len = ofconn->miss_send_len;
- }
+ reason = OFPR_NO_MATCH;
+ max_len = INT_MAX;
}
- opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port),
- msg->type, &payload, send_len);
+ total_len = msg->length - sizeof *msg;
+ in_port = odp_port_to_ofp_port(msg->port);
- /* Send. */
- rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
+ /* Repurpose packet buffer by overwriting header. */
+ ofpbuf_pull(packet, sizeof(struct odp_msg));
+ opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
+ opi->header.version = OFP_VERSION;
+ opi->header.type = OFPT_PACKET_IN;
+ opi->total_len = htons(total_len);
+ opi->in_port = htons(in_port);
+ opi->reason = reason;
- ofpbuf_delete(packet);
+ return max_len;
}
+/* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
+ * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
+ * as necessary according to their individual configurations.
+ *
+ * 'packet' must have sufficient headroom to convert it into a struct
+ * ofp_packet_in (e.g. as returned by dpif_recv()).
+ *
+ * Takes ownership of 'packet'. */
static void
send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
{
- struct odp_msg *msg = packet->data;
struct ofconn *ofconn, *prev;
+ int max_len;
- assert(msg->type == _ODPL_MISS_NR || msg->type == _ODPL_ACTION_NR);
+ max_len = do_convert_to_packet_in(packet);
prev = NULL;
LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
- if (ofconn->role != NX_ROLE_SLAVE) {
+ if (ofconn_receives_async_msgs(ofconn)) {
if (prev) {
- pinsched_send(prev->schedulers[msg->type], msg->port,
- ofpbuf_clone(packet), do_send_packet_in, prev);
+ schedule_packet_in(prev, packet, max_len, true);
}
prev = ofconn;
}
}
if (prev) {
- pinsched_send(prev->schedulers[msg->type], msg->port,
- packet, do_send_packet_in, prev);
+ schedule_packet_in(prev, packet, max_len, false);
} else {
ofpbuf_delete(packet);
}
/* 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);
+ 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. */
}
/* Determine output port. */
- out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
+ out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
+ NULL);
if (out_port < 0) {
add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
} else if (out_port != flow->in_port) {
git://git.onelab.eu / sliver-openvswitch.git / blobdiff