Merge branch 'master' into next

Conflicts:
	COPYING
	datapath/datapath.h
	lib/automake.mk
	lib/dpif-provider.h
	lib/dpif.c
	lib/hmap.h
	lib/netdev-provider.h
	lib/netdev.c
	lib/stream-ssl.h
	ofproto/executer.c
	ofproto/ofproto.c
	ofproto/ofproto.h
	tests/automake.mk
	utilities/ovs-ofctl.c
	utilities/ovs-vsctl.in
	vswitchd/ovs-vswitchd.conf.5.in
	xenserver/etc_init.d_vswitch
	xenserver/etc_xensource_scripts_vif
	xenserver/opt_xensource_libexec_interface-reconfigure

7 files changed, 771 insertions, 23 deletions

diff --git a/ofproto/automake.mk b/ofproto/automake.mk
index 0f05e5358..0c99b4907 100644
--- a/ofproto/automake.mk
+++ b/ofproto/automake.mk
@@ -19,6 +19,8 @@ ofproto_libofproto_a_SOURCES = \
 	ofproto/netflow.h \
 	ofproto/ofproto.c \
 	ofproto/ofproto.h \
+	ofproto/ofproto-sflow.c \
+	ofproto/ofproto-sflow.h \
 	ofproto/pktbuf.c \
 	ofproto/pktbuf.h \
 	ofproto/pinsched.c \
diff --git a/ofproto/collectors.c b/ofproto/collectors.c
index f7cb1dbe4..4589f3298 100644
--- a/ofproto/collectors.c
+++ b/ofproto/collectors.c
@@ -121,3 +121,9 @@ collectors_send(const struct collectors *c, const void *payload, size_t n)
         }
     }
 }
+
+int
+collectors_count(const struct collectors *c)
+{
+    return c->n_fds;
+}
diff --git a/ofproto/collectors.h b/ofproto/collectors.h
index a4abb6312..ac70f3759 100644
--- a/ofproto/collectors.h
+++ b/ofproto/collectors.h
@@ -28,4 +28,6 @@ void collectors_destroy(struct collectors *);
 
 void collectors_send(const struct collectors *, const void *, size_t);
 
+int collectors_count(const struct collectors *);
+
 #endif /* collectors.h */
diff --git a/ofproto/ofproto-sflow.c b/ofproto/ofproto-sflow.c
new file mode 100644
index 000000000..1b659d1ab
--- /dev/null
+++ b/ofproto/ofproto-sflow.c
@@ -0,0 +1,607 @@
+/*
+ * Copyright (c) 2009, 2010 InMon Corp.
+ * Copyright (c) 2009 Nicira Networks.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <config.h>
+#include "ofproto-sflow.h"
+#include <inttypes.h>
+#include <stdlib.h>
+#include "collectors.h"
+#include "dpif.h"
+#include "compiler.h"
+#include "netdev.h"
+#include "ofpbuf.h"
+#include "ofproto.h"
+#include "poll-loop.h"
+#include "port-array.h"
+#include "sflow_api.h"
+#include "socket-util.h"
+#include "timeval.h"
+
+#define THIS_MODULE VLM_sflow
+#include "vlog.h"
+
+struct ofproto_sflow_port {
+    struct netdev *netdev;      /* Underlying network device, for stats. */
+    SFLDataSource_instance dsi; /* sFlow library's notion of port number. */
+};
+
+struct ofproto_sflow {
+    struct ofproto *ofproto;
+    struct collectors *collectors;
+    SFLAgent *sflow_agent;
+    struct ofproto_sflow_options *options;
+    struct dpif *dpif;
+    time_t next_tick;
+    size_t n_flood, n_all;
+    struct port_array ports;    /* Indexed by ODP port number. */
+};
+
+#define RECEIVER_INDEX 1
+
+static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+
+static bool
+nullable_string_is_equal(const char *a, const char *b)
+{
+    return a ? b && !strcmp(a, b) : !b;
+}
+
+static bool
+ofproto_sflow_options_equal(const struct ofproto_sflow_options *a,
+                            const struct ofproto_sflow_options *b)
+{
+    return (svec_equal(&a->targets, &b->targets)
+            && a->sampling_rate == b->sampling_rate
+            && a->polling_interval == b->polling_interval
+            && a->header_len == b->header_len
+            && a->sub_id == b->sub_id
+            && nullable_string_is_equal(a->agent_device, b->agent_device)
+            && nullable_string_is_equal(a->control_ip, b->control_ip));
+}
+
+static struct ofproto_sflow_options *
+ofproto_sflow_options_clone(const struct ofproto_sflow_options *old)
+{
+    struct ofproto_sflow_options *new = xmemdup(old, sizeof *old);
+    svec_clone(&new->targets, &old->targets);
+    new->agent_device = old->agent_device ? xstrdup(old->agent_device) : NULL;
+    new->control_ip = old->control_ip ? xstrdup(old->control_ip) : NULL;
+    return new;
+}
+
+static void
+ofproto_sflow_options_destroy(struct ofproto_sflow_options *options)
+{
+    if (options) {
+        svec_destroy(&options->targets);
+        free(options->agent_device);
+        free(options->control_ip);
+        free(options);
+    }
+}
+
+/* sFlow library callback to allocate memory. */
+static void *
+sflow_agent_alloc_cb(void *magic UNUSED, SFLAgent *agent UNUSED, size_t bytes)
+{
+    return calloc(1, bytes);
+}
+
+/* sFlow library callback to free memory. */
+static int
+sflow_agent_free_cb(void *magic UNUSED, SFLAgent *agent UNUSED, void *obj)
+{
+    free(obj);
+    return 0;
+}
+
+/* sFlow library callback to report error. */
+static void
+sflow_agent_error_cb(void *magic UNUSED, SFLAgent *agent UNUSED, char *msg)
+{
+    VLOG_WARN("sFlow agent error: %s", msg);
+}
+
+/* sFlow library callback to send datagram. */
+static void
+sflow_agent_send_packet_cb(void *os_, SFLAgent *agent UNUSED,
+                           SFLReceiver *receiver UNUSED, u_char *pkt,
+                           uint32_t pktLen)
+{
+    struct ofproto_sflow *os = os_;
+    collectors_send(os->collectors, pkt, pktLen);
+}
+
+static void
+sflow_agent_get_counters(void *os_, SFLPoller *poller,
+                         SFL_COUNTERS_SAMPLE_TYPE *cs)
+{
+    struct ofproto_sflow *os = os_;
+    SFLCounters_sample_element elem;
+    struct ofproto_sflow_port *osp;
+    SFLIf_counters *counters;
+    struct netdev_stats stats;
+    enum netdev_flags flags;
+    uint32_t current;
+
+    osp = port_array_get(&os->ports, poller->bridgePort);
+    if (!osp) {
+        return;
+    }
+
+    elem.tag = SFLCOUNTERS_GENERIC;
+    counters = &elem.counterBlock.generic;
+    counters->ifIndex = SFL_DS_INDEX(poller->dsi);
+    counters->ifType = 6;
+    if (!netdev_get_features(osp->netdev, &current, NULL, NULL, NULL)) {
+      /* The values of ifDirection come from MAU MIB (RFC 2668): 0 = unknown,
+         1 = full-duplex, 2 = half-duplex, 3 = in, 4=out */
+        counters->ifSpeed = netdev_features_to_bps(current);
+        counters->ifDirection = (netdev_features_is_full_duplex(current)
+                                 ? 1 : 2);
+    } else {
+        counters->ifSpeed = 100000000;
+        counters->ifDirection = 0;
+    }
+    if (!netdev_get_flags(osp->netdev, &flags) && flags & NETDEV_UP) {
+        bool carrier;
+
+        counters->ifStatus = 1; /* ifAdminStatus up. */
+        if (!netdev_get_carrier(osp->netdev, &carrier) && carrier) {
+            counters->ifStatus |= 2; /* ifOperStatus us. */
+        }
+    } else {
+        counters->ifStatus = 0;  /* Down. */
+    }
+
+    /* XXX
+       1. Is the multicast counter filled in?
+       2. Does the multicast counter include broadcasts?
+       3. Does the rx_packets counter include multicasts/broadcasts?
+    */
+    netdev_get_stats(osp->netdev, &stats);
+    counters->ifInOctets = stats.rx_bytes;
+    counters->ifInUcastPkts = stats.rx_packets;
+    counters->ifInMulticastPkts = stats.multicast;
+    counters->ifInBroadcastPkts = -1;
+    counters->ifInDiscards = stats.rx_dropped;
+    counters->ifInErrors = stats.rx_errors;
+    counters->ifInUnknownProtos = -1;
+    counters->ifOutOctets = stats.tx_bytes;
+    counters->ifOutUcastPkts = stats.tx_packets;
+    counters->ifOutMulticastPkts = -1;
+    counters->ifOutBroadcastPkts = -1;
+    counters->ifOutDiscards = stats.tx_dropped;
+    counters->ifOutErrors = stats.tx_errors;
+    counters->ifPromiscuousMode = 0;
+
+    SFLADD_ELEMENT(cs, &elem);
+    sfl_poller_writeCountersSample(poller, cs);
+}
+
+/* Obtains an address to use for the local sFlow agent and stores it into
+ * '*agent_addr'.  Returns true if successful, false on failure.
+ *
+ * The sFlow agent address should be a local IP address that is persistent and
+ * reachable over the network, if possible.  The IP address associated with
+ * 'agent_device' is used if it has one, and otherwise 'control_ip', the IP
+ * address used to talk to the controller. */
+static bool
+sflow_choose_agent_address(const char *agent_device, const char *control_ip,
+                           SFLAddress *agent_addr)
+{
+    struct in_addr in4;
+
+    memset(agent_addr, 0, sizeof *agent_addr);
+    agent_addr->type = SFLADDRESSTYPE_IP_V4;
+
+    if (agent_device) {
+        struct netdev *netdev;
+
+        if (!netdev_open_default(agent_device, &netdev)) {
+            int error = netdev_get_in4(netdev, &in4, NULL);
+            netdev_close(netdev);
+            if (!error) {
+                goto success;
+            }
+        }
+    }
+
+    if (control_ip && !lookup_ip(control_ip, &in4)) {
+        goto success;
+    }
+
+    VLOG_ERR("could not determine IP address for sFlow agent");
+    return false;
+
+success:
+    agent_addr->address.ip_v4.addr = in4.s_addr;
+    return true;
+}
+
+void
+ofproto_sflow_clear(struct ofproto_sflow *os)
+{
+    struct ofproto_sflow_port *osp;
+    unsigned int odp_port;
+
+    if (os->sflow_agent) {
+        sfl_agent_release(os->sflow_agent);
+        os->sflow_agent = NULL;
+    }
+    collectors_destroy(os->collectors);
+    os->collectors = NULL;
+    ofproto_sflow_options_destroy(os->options);
+    os->options = NULL;
+
+    PORT_ARRAY_FOR_EACH (osp, &os->ports, odp_port) {
+        ofproto_sflow_del_port(os, odp_port);
+    }
+    port_array_clear(&os->ports);
+
+    /* Turn off sampling to save CPU cycles. */
+    dpif_set_sflow_probability(os->dpif, 0);
+}
+
+bool
+ofproto_sflow_is_enabled(const struct ofproto_sflow *os)
+{
+    return os->collectors != NULL;
+}
+
+struct ofproto_sflow *
+ofproto_sflow_create(struct dpif *dpif)
+{
+    struct ofproto_sflow *os;
+
+    os = xcalloc(1, sizeof *os);
+    os->dpif = dpif;
+    os->next_tick = time_now() + 1;
+    port_array_init(&os->ports);
+    return os;
+}
+
+void
+ofproto_sflow_destroy(struct ofproto_sflow *os)
+{
+    if (os) {
+        ofproto_sflow_clear(os);
+        port_array_destroy(&os->ports);
+        free(os);
+    }
+}
+
+static void
+ofproto_sflow_add_poller(struct ofproto_sflow *os,
+                         struct ofproto_sflow_port *osp, uint16_t odp_port)
+{
+    SFLPoller *poller = sfl_agent_addPoller(os->sflow_agent, &osp->dsi, os,
+                                            sflow_agent_get_counters);
+    sfl_poller_set_sFlowCpInterval(poller, os->options->polling_interval);
+    sfl_poller_set_sFlowCpReceiver(poller, RECEIVER_INDEX);
+    sfl_poller_set_bridgePort(poller, odp_port);
+}
+
+static void
+ofproto_sflow_add_sampler(struct ofproto_sflow *os,
+			  struct ofproto_sflow_port *osp,
+			  u_int32_t sampling_rate, u_int32_t header_len)
+{
+    SFLSampler *sampler = sfl_agent_addSampler(os->sflow_agent, &osp->dsi);
+    sfl_sampler_set_sFlowFsPacketSamplingRate(sampler, sampling_rate);
+    sfl_sampler_set_sFlowFsMaximumHeaderSize(sampler, header_len);
+    sfl_sampler_set_sFlowFsReceiver(sampler, RECEIVER_INDEX);
+}
+
+void
+ofproto_sflow_add_port(struct ofproto_sflow *os, uint16_t odp_port,
+                       const char *netdev_name)
+{
+    struct ofproto_sflow_port *osp;
+    struct netdev *netdev;
+    uint32_t ifindex;
+    int error;
+
+    ofproto_sflow_del_port(os, odp_port);
+
+    /* Open network device. */
+    error = netdev_open_default(netdev_name, &netdev);
+    if (error) {
+        VLOG_WARN_RL(&rl, "failed to open network device \"%s\": %s",
+                     netdev_name, strerror(error));
+        return;
+    }
+
+    /* Add to table of ports. */
+    osp = xmalloc(sizeof *osp);
+    osp->netdev = netdev;
+    ifindex = netdev_get_ifindex(netdev);
+    if (ifindex <= 0) {
+        ifindex = (os->sflow_agent->subId << 16) + odp_port;
+    }
+    SFL_DS_SET(osp->dsi, 0, ifindex, 0);
+    port_array_set(&os->ports, odp_port, osp);
+
+    /* Add poller. */
+    if (os->sflow_agent) {
+        ofproto_sflow_add_poller(os, osp, odp_port);
+    }
+}
+
+void
+ofproto_sflow_del_port(struct ofproto_sflow *os, uint16_t odp_port)
+{
+    struct ofproto_sflow_port *osp = port_array_get(&os->ports, odp_port);
+    if (osp) {
+        if (os->sflow_agent) {
+            sfl_agent_removePoller(os->sflow_agent, &osp->dsi);
+            sfl_agent_removeSampler(os->sflow_agent, &osp->dsi);
+        }
+        netdev_close(osp->netdev);
+        free(osp);
+        port_array_set(&os->ports, odp_port, NULL);
+    }
+}
+
+void
+ofproto_sflow_set_options(struct ofproto_sflow *os,
+                          const struct ofproto_sflow_options *options)
+{
+    struct ofproto_sflow_port *osp;
+    bool options_changed;
+    SFLReceiver *receiver;
+    unsigned int odp_port;
+    SFLAddress agentIP;
+    time_t now;
+    int error;
+
+    if (!options->targets.n || !options->sampling_rate) {
+        /* No point in doing any work if there are no targets or nothing to
+         * sample. */
+        ofproto_sflow_clear(os);
+        return;
+    }
+
+    options_changed = (!os->options
+                       || !ofproto_sflow_options_equal(options, os->options));
+
+    /* Configure collectors if options have changed or if we're shortchanged in
+     * collectors (which indicates that opening one or more of the configured
+     * collectors failed, so that we should retry). */
+    if (options_changed
+        || collectors_count(os->collectors) < options->targets.n) {
+        collectors_destroy(os->collectors);
+        error = collectors_create(&options->targets,
+                                  SFL_DEFAULT_COLLECTOR_PORT, &os->collectors);
+        if (os->collectors == NULL) {
+            VLOG_WARN_RL(&rl, "no collectors could be initialized, "
+                         "sFlow disabled");
+            ofproto_sflow_clear(os);
+            return;
+        }
+    }
+
+    /* Avoid reconfiguring if options didn't change. */
+    if (!options_changed) {
+        return;
+    }
+    ofproto_sflow_options_destroy(os->options);
+    os->options = ofproto_sflow_options_clone(options);
+
+    /* Choose agent IP address. */
+    if (!sflow_choose_agent_address(options->agent_device,
+                                    options->control_ip, &agentIP)) {
+        ofproto_sflow_clear(os);
+        return;
+    }
+
+    /* Create agent. */
+    VLOG_INFO("creating sFlow agent %d", options->sub_id);
+    if (os->sflow_agent) {
+        sfl_agent_release(os->sflow_agent);
+    }
+    os->sflow_agent = xcalloc(1, sizeof *os->sflow_agent);
+    now = time_now();
+    sfl_agent_init(os->sflow_agent,
+                   &agentIP,
+                   options->sub_id,
+                   now,         /* Boot time. */
+                   now,         /* Current time. */
+                   os,          /* Pointer supplied to callbacks. */
+                   sflow_agent_alloc_cb,
+                   sflow_agent_free_cb,
+                   sflow_agent_error_cb,
+                   sflow_agent_send_packet_cb);
+
+    receiver = sfl_agent_addReceiver(os->sflow_agent);
+    sfl_receiver_set_sFlowRcvrOwner(receiver, "Open vSwitch sFlow");
+    sfl_receiver_set_sFlowRcvrTimeout(receiver, 0xffffffff);
+
+    /* Set the sampling_rate down in the datapath. */
+    dpif_set_sflow_probability(os->dpif,
+                               MAX(1, UINT32_MAX / options->sampling_rate));
+
+    /* Add samplers and pollers for the currently known ports. */
+    PORT_ARRAY_FOR_EACH (osp, &os->ports, odp_port) {
+        ofproto_sflow_add_sampler(os, osp,
+                                  options->sampling_rate, options->header_len);
+    }
+}
+
+static int
+ofproto_sflow_odp_port_to_ifindex(const struct ofproto_sflow *os,
+                                  uint16_t odp_port)
+{
+    struct ofproto_sflow_port *osp = port_array_get(&os->ports, odp_port);
+    return osp ? SFL_DS_INDEX(osp->dsi) : 0;
+}
+
+void
+ofproto_sflow_received(struct ofproto_sflow *os, struct odp_msg *msg)
+{
+    SFL_FLOW_SAMPLE_TYPE fs;
+    SFLFlow_sample_element hdrElem;
+    SFLSampled_header *header;
+    SFLFlow_sample_element switchElem;
+    SFLSampler *sampler;
+    const struct odp_sflow_sample_header *hdr;
+    const union odp_action *actions;
+    struct ofpbuf payload;
+    size_t n_actions, n_outputs;
+    size_t min_size;
+    flow_t flow;
+    size_t i;
+
+    /* Get odp_sflow_sample_header. */
+    min_size = sizeof *msg + sizeof *hdr;
+    if (min_size > msg->length) {
+        VLOG_WARN_RL(&rl, "sFlow packet too small (%"PRIu32" < %zu)",
+                     msg->length, min_size);
+        return;
+    }
+    hdr = (const struct odp_sflow_sample_header *) (msg + 1);
+
+    /* Get actions. */
+    n_actions = hdr->n_actions;
+    if (n_actions > 65536 / sizeof *actions) {
+        VLOG_WARN_RL(&rl, "too many actions in sFlow packet (%zu > %zu)",
+                     65536 / sizeof *actions, n_actions);
+        return;
+    }
+    min_size += n_actions * sizeof *actions;
+    if (min_size > msg->length) {
+        VLOG_WARN_RL(&rl, "sFlow packet with %zu actions too small "
+                     "(%"PRIu32" < %zu)",
+                     n_actions, msg->length, min_size);
+        return;
+    }
+    actions = (const union odp_action *) (hdr + 1);
+
+    /* Get packet payload and extract flow. */
+    payload.data = (union odp_action *) (actions + n_actions);
+    payload.size = msg->length - min_size;
+    flow_extract(&payload, msg->port, &flow);
+
+    /* Build a flow sample */
+    memset(&fs, 0, sizeof fs);
+    fs.input = ofproto_sflow_odp_port_to_ifindex(os, msg->port);
+    fs.output = 0;              /* Filled in correctly below. */
+    fs.sample_pool = hdr->sample_pool;
+
+    /* We are going to give it to the sampler that represents this input port.
+     * By implementing "ingress-only" sampling like this we ensure that we
+     * never have to offer the same sample to more than one sampler. */
+    sampler = sfl_agent_getSamplerByIfIndex(os->sflow_agent, fs.input);
+    if (!sampler) {
+        VLOG_WARN_RL(&rl, "no sampler for input ifIndex (%"PRIu32")",
+                     fs.input);
+        return;
+    }
+
+    /* Sampled header. */
+    memset(&hdrElem, 0, sizeof hdrElem);
+    hdrElem.tag = SFLFLOW_HEADER;
+    header = &hdrElem.flowType.header;
+    header->header_protocol = SFLHEADER_ETHERNET_ISO8023;
+    header->frame_length = payload.size;
+    header->stripped = 4; /* Ethernet FCS stripped off. */
+    header->header_length = MIN(payload.size,
+                                sampler->sFlowFsMaximumHeaderSize);
+    header->header_bytes = payload.data;
+
+    /* Add extended switch element. */
+    memset(&switchElem, 0, sizeof(switchElem));
+    switchElem.tag = SFLFLOW_EX_SWITCH;
+    switchElem.flowType.sw.src_vlan = ntohs(flow.dl_vlan);
+    switchElem.flowType.sw.src_priority = -1; /* XXX */
+    switchElem.flowType.sw.dst_vlan = -1;     /* Filled in correctly below. */
+    switchElem.flowType.sw.dst_priority = switchElem.flowType.sw.src_priority;
+
+    /* Figure out the output ports. */
+    n_outputs = 0;
+    for (i = 0; i < n_actions; i++) {
+        const union odp_action *a = &actions[i];
+
+        switch (a->type) {
+        case ODPAT_OUTPUT:
+            fs.output = ofproto_sflow_odp_port_to_ifindex(os, a->output.port);
+            n_outputs++;
+            break;
+
+        case ODPAT_OUTPUT_GROUP:
+            n_outputs += (a->output_group.group == DP_GROUP_FLOOD ? os->n_flood
+                          : a->output_group.group == DP_GROUP_ALL ? os->n_all
+                          : 0);
+            break;
+
+        case ODPAT_SET_VLAN_VID:
+            switchElem.flowType.sw.dst_vlan = ntohs(a->vlan_vid.vlan_vid);
+            break;
+
+        case ODPAT_SET_VLAN_PCP:
+            switchElem.flowType.sw.dst_priority = a->vlan_pcp.vlan_pcp;
+            break;
+
+        default:
+            break;
+        }
+    }
+
+    /* Set output port, as defined by http://www.sflow.org/sflow_version_5.txt
+       (search for "Input/output port information"). */
+    if (!n_outputs) {
+        /* This value indicates that the packet was dropped for an unknown
+         * reason. */
+        fs.output = 0x40000000 | 256;
+    } else if (n_outputs > 1 || !fs.output) {
+        /* Setting the high bit means "multiple output ports". */
+        fs.output = 0x80000000 | n_outputs;
+    }
+
+    /* Submit the flow sample to be encoded into the next datagram. */
+    SFLADD_ELEMENT(&fs, &hdrElem);
+    SFLADD_ELEMENT(&fs, &switchElem);
+    sfl_sampler_writeFlowSample(sampler, &fs);
+}
+
+void
+ofproto_sflow_set_group_sizes(struct ofproto_sflow *os,
+                              size_t n_flood, size_t n_all)
+{
+    os->n_flood = n_flood;
+    os->n_all = n_all;
+}
+
+void
+ofproto_sflow_run(struct ofproto_sflow *os)
+{
+    if (ofproto_sflow_is_enabled(os)) {
+        time_t now = time_now();
+        if (now >= os->next_tick) {
+            sfl_agent_tick(os->sflow_agent, now);
+            os->next_tick = now + 1;
+        }
+    }
+}
+
+void
+ofproto_sflow_wait(struct ofproto_sflow *os)
+{
+    if (ofproto_sflow_is_enabled(os)) {
+        poll_timer_wait(os->next_tick * 1000 - time_msec());
+    }
+}
diff --git a/ofproto/ofproto-sflow.h b/ofproto/ofproto-sflow.h
new file mode 100644
index 000000000..ec86d1159
--- /dev/null
+++ b/ofproto/ofproto-sflow.h
@@ -0,0 +1,46 @@
+/*
+ * Copyright (c) 2009 InMon Corp.
+ * Copyright (c) 2009 Nicira Networks.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef OFPROTO_SFLOW_H
+#define OFPROTO_SFLOW_H 1
+
+#include <stdint.h>
+#include "svec.h"
+
+struct dpif;
+struct odp_msg;
+struct ofproto_sflow_options;
+
+struct ofproto_sflow *ofproto_sflow_create(struct dpif *);
+void ofproto_sflow_destroy(struct ofproto_sflow *);
+void ofproto_sflow_set_options(struct ofproto_sflow *,
+                               const struct ofproto_sflow_options *);
+void ofproto_sflow_clear(struct ofproto_sflow *);
+bool ofproto_sflow_is_enabled(const struct ofproto_sflow *);
+
+void ofproto_sflow_add_port(struct ofproto_sflow *, uint16_t odp_port,
+                            const char *netdev_name);
+void ofproto_sflow_del_port(struct ofproto_sflow *, uint16_t odp_port);
+void ofproto_sflow_set_group_sizes(struct ofproto_sflow *,
+                                   size_t n_flood, size_t n_all);
+
+void ofproto_sflow_run(struct ofproto_sflow *);
+void ofproto_sflow_wait(struct ofproto_sflow *);
+
+void ofproto_sflow_received(struct ofproto_sflow *, struct odp_msg *);
+
+#endif /* ofproto/ofproto-sflow.h */
diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c
index 10b4796f1..c44762c46 100644
--- a/ofproto/ofproto.c
+++ b/ofproto/ofproto.c
@@ -34,6 +34,7 @@
 #include "netflow.h"
 #include "odp-util.h"
 #include "ofp-print.h"
+#include "ofproto-sflow.h"
 #include "ofpbuf.h"
 #include "openflow/nicira-ext.h"
 #include "openflow/openflow.h"
@@ -58,10 +59,7 @@
 #define THIS_MODULE VLM_ofproto
 #include "vlog.h"
 
-enum {
-    DP_GROUP_FLOOD = 0,
-    DP_GROUP_ALL = 1
-};
+#include "sflow_api.h"
 
 enum {
     TABLEID_HASH = 0,
@@ -205,6 +203,7 @@ struct ofproto {
     struct fail_open *fail_open;
     struct pinsched *miss_sched, *action_sched;
     struct netflow *netflow;
+    struct ofproto_sflow *sflow;
 
     /* Flow table. */
     struct classifier cls;
@@ -249,7 +248,8 @@ static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
 static void handle_openflow(struct ofconn *, struct ofproto *,
                             struct ofpbuf *);
 
-static void refresh_port_group(struct ofproto *, unsigned int group);
+static void refresh_port_groups(struct ofproto *);
+
 static void update_port(struct ofproto *, const char *devname);
 static int init_ports(struct ofproto *);
 static void reinit_ports(struct ofproto *);
@@ -279,7 +279,7 @@ ofproto_create(const char *datapath, const char *datapath_type,
         dpif_close(dpif);
         return error;
     }
-    error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION);
+    error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
     if (error) {
         VLOG_ERR("failed to listen on datapath %s: %s",
                  datapath, strerror(error));
@@ -312,6 +312,7 @@ ofproto_create(const char *datapath, const char *datapath_type,
     p->fail_open = NULL;
     p->miss_sched = p->action_sched = NULL;
     p->netflow = NULL;
+    p->sflow = NULL;
 
     /* Initialize flow table. */
     classifier_init(&p->cls);
@@ -532,6 +533,30 @@ ofproto_set_netflow(struct ofproto *ofproto,
 }
 
 void
+ofproto_set_sflow(struct ofproto *ofproto,
+                  const struct ofproto_sflow_options *oso)
+{
+    struct ofproto_sflow *os = ofproto->sflow;
+    if (oso) {
+        if (!os) {
+            struct ofport *ofport;
+            unsigned int odp_port;
+
+            os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
+            refresh_port_groups(ofproto);
+            PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
+                ofproto_sflow_add_port(os, odp_port,
+                                       netdev_get_name(ofport->netdev));
+            }
+        }
+        ofproto_sflow_set_options(os, oso);
+    } else {
+        ofproto_sflow_destroy(os);
+        ofproto->sflow = NULL;
+    }
+}
+
+void
 ofproto_set_failure(struct ofproto *ofproto, bool fail_open)
 {
     if (fail_open) {
@@ -676,6 +701,7 @@ ofproto_destroy(struct ofproto *p)
     pinsched_destroy(p->miss_sched);
     pinsched_destroy(p->action_sched);
     netflow_destroy(p->netflow);
+    ofproto_sflow_destroy(p->sflow);
 
     switch_status_unregister(p->ss_cat);
 
@@ -829,6 +855,9 @@ ofproto_run1(struct ofproto *p)
     if (p->netflow) {
         netflow_run(p->netflow);
     }
+    if (p->sflow) {
+        ofproto_sflow_run(p->sflow);
+    }
 
     return 0;
 }
@@ -882,6 +911,9 @@ ofproto_wait(struct ofproto *p)
     }
     pinsched_wait(p->miss_sched);
     pinsched_wait(p->action_sched);
+    if (p->sflow) {
+        ofproto_sflow_wait(p->sflow);
+    }
     if (!tag_set_is_empty(&p->revalidate_set)) {
         poll_immediate_wake();
     }
@@ -1022,7 +1054,7 @@ reinit_ports(struct ofproto *p)
     svec_destroy(&devnames);
 }
 
-static void
+static size_t
 refresh_port_group(struct ofproto *p, unsigned int group)
 {
     uint16_t *ports;
@@ -1041,13 +1073,18 @@ refresh_port_group(struct ofproto *p, unsigned int group)
     }
     dpif_port_group_set(p->dpif, group, ports, n_ports);
     free(ports);
+
+    return n_ports;
 }
 
 static void
 refresh_port_groups(struct ofproto *p)
 {
-    refresh_port_group(p, DP_GROUP_FLOOD);
-    refresh_port_group(p, DP_GROUP_ALL);
+    size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
+    size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
+    if (p->sflow) {
+        ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
+    }
 }
 
 static struct ofport *
@@ -1152,19 +1189,29 @@ send_port_status(struct ofproto *p, const struct ofport *ofport,
 static void
 ofport_install(struct ofproto *p, struct ofport *ofport)
 {
+    uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
+    const char *netdev_name = (const char *) ofport->opp.name;
+
     netdev_monitor_add(p->netdev_monitor, ofport->netdev);
-    port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no),
-                   ofport);
-    shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport);
+    port_array_set(&p->ports, odp_port, ofport);
+    shash_add(&p->port_by_name, netdev_name, ofport);
+    if (p->sflow) {
+        ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
+    }
 }
 
 static void
 ofport_remove(struct ofproto *p, struct ofport *ofport)
 {
+    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, ofp_port_to_odp_port(ofport->opp.port_no), NULL);
+    port_array_set(&p->ports, odp_port, NULL);
     shash_delete(&p->port_by_name,
                  shash_find(&p->port_by_name, (char *) ofport->opp.name));
+    if (p->sflow) {
+        ofproto_sflow_del_port(p->sflow, odp_port);
+    }
 }
 
 static void
@@ -2249,7 +2296,7 @@ update_port_config(struct ofproto *p, struct ofport *port,
 #undef REVALIDATE_BITS
     if (mask & OFPPC_NO_FLOOD) {
         port->opp.config ^= OFPPC_NO_FLOOD;
-        refresh_port_group(p, DP_GROUP_FLOOD);
+        refresh_port_groups(p);
     }
     if (mask & OFPPC_NO_PACKET_IN) {
         port->opp.config ^= OFPPC_NO_PACKET_IN;
@@ -2999,7 +3046,7 @@ handle_openflow(struct ofconn *ofconn, struct ofproto *p,
 }
 
 static void
-handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
+handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
 {
     struct odp_msg *msg = packet->data;
     uint16_t in_port = odp_port_to_ofp_port(msg->port);
@@ -3007,14 +3054,6 @@ handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
     struct ofpbuf payload;
     flow_t flow;
 
-    /* Handle controller actions. */
-    if (msg->type == _ODPL_ACTION_NR) {
-        COVERAGE_INC(ofproto_ctlr_action);
-        pinsched_send(p->action_sched, in_port, packet,
-                      send_packet_in_action, p);
-        return;
-    }
-
     payload.data = msg + 1;
     payload.size = msg->length - sizeof *msg;
     flow_extract(&payload, msg->port, &flow);
@@ -3084,6 +3123,36 @@ handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
         ofpbuf_delete(packet);
     }
 }
+
+static void
+handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
+{
+    struct odp_msg *msg = packet->data;
+
+    switch (msg->type) {
+    case _ODPL_ACTION_NR:
+        COVERAGE_INC(ofproto_ctlr_action);
+        pinsched_send(p->action_sched, odp_port_to_ofp_port(msg->port), packet,
+                      send_packet_in_action, p);
+        break;
+
+    case _ODPL_SFLOW_NR:
+        if (p->sflow) {
+            ofproto_sflow_received(p->sflow, msg);
+        }
+        ofpbuf_delete(packet);
+        break;
+
+    case _ODPL_MISS_NR:
+        handle_odp_miss_msg(p, packet);
+        break;
+
+    default:
+        VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
+                     msg->type);
+        break;
+    }
+}
 
 static void
 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
diff --git a/ofproto/ofproto.h b/ofproto/ofproto.h
index ddc34483d..a94c8b59b 100644
--- a/ofproto/ofproto.h
+++ b/ofproto/ofproto.h
@@ -29,6 +29,11 @@ struct ofhooks;
 struct ofproto;
 struct svec;
 
+enum {
+    DP_GROUP_FLOOD = 0,
+    DP_GROUP_ALL = 1
+};
+
 struct ofexpired {
     flow_t flow;
     uint64_t packet_count;      /* Packets from subrules. */
@@ -36,6 +41,16 @@ struct ofexpired {
     long long int used;         /* Last-used time (0 if never used). */
 };
 
+struct ofproto_sflow_options {
+    struct svec targets;
+    uint32_t sampling_rate;
+    uint32_t polling_interval;
+    uint32_t header_len;
+    uint32_t sub_id;
+    char *agent_device;
+    char *control_ip;
+};
+
 int ofproto_create(const char *datapath, const char *datapath_type,
                    const struct ofhooks *, void *aux,
                    struct ofproto **ofprotop);
@@ -62,6 +77,7 @@ int ofproto_set_listeners(struct ofproto *, const struct svec *listeners);
 int ofproto_set_snoops(struct ofproto *, const struct svec *snoops);
 int ofproto_set_netflow(struct ofproto *,
                         const struct netflow_options *nf_options);
+void ofproto_set_sflow(struct ofproto *, const struct ofproto_sflow_options *);
 void ofproto_set_failure(struct ofproto *, bool fail_open);
 void ofproto_set_rate_limit(struct ofproto *, int rate_limit, int burst_limit);
 int ofproto_set_stp(struct ofproto *, bool enable_stp);