ovs/ofproto/ofproto-dpif-upcall.c

/* Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
 *
 * 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-dpif-upcall.h"

#include <errno.h>
#include <stdbool.h>
#include <inttypes.h>

#include "connmgr.h"
#include "coverage.h"
#include "dpif.h"
#include "dynamic-string.h"
#include "fail-open.h"
#include "guarded-list.h"
#include "latch.h"
#include "list.h"
#include "netlink.h"
#include "ofpbuf.h"
#include "ofproto-dpif-ipfix.h"
#include "ofproto-dpif-sflow.h"
#include "ofproto-dpif-xlate.h"
#include "ovs-rcu.h"
#include "packets.h"
#include "poll-loop.h"
#include "seq.h"
#include "unixctl.h"
#include "vlog.h"

#define MAX_QUEUE_LENGTH 512
#define FLOW_MISS_MAX_BATCH 50
#define REVALIDATE_MAX_BATCH 50

VLOG_DEFINE_THIS_MODULE(ofproto_dpif_upcall);

COVERAGE_DEFINE(upcall_duplicate_flow);

/* A thread that reads upcalls from dpif, forwards each upcall's packet,
 * and possibly sets up a kernel flow as a cache. */
struct handler {
    struct udpif *udpif;               /* Parent udpif. */
    pthread_t thread;                  /* Thread ID. */
    uint32_t handler_id;               /* Handler id. */
};

/* A thread that processes datapath flows, updates OpenFlow statistics, and
 * updates or removes them if necessary. */
struct revalidator {
    struct udpif *udpif;               /* Parent udpif. */
    pthread_t thread;                  /* Thread ID. */
    unsigned int id;                   /* ovsthread_id_self(). */
    struct hmap *ukeys;                /* Points into udpif->ukeys for this
                                          revalidator. Used for GC phase. */
};

/* An upcall handler for ofproto_dpif.
 *
 * udpif keeps records of two kind of logically separate units:
 *
 * upcall handling
 * ---------------
 *
 *    - An array of 'struct handler's for upcall handling and flow
 *      installation.
 *
 * flow revalidation
 * -----------------
 *
 *    - Revalidation threads which read the datapath flow table and maintains
 *      them.
 */
struct udpif {
    struct list list_node;             /* In all_udpifs list. */

    struct dpif *dpif;                 /* Datapath handle. */
    struct dpif_backer *backer;        /* Opaque dpif_backer pointer. */

    uint32_t secret;                   /* Random seed for upcall hash. */

    struct handler *handlers;          /* Upcall handlers. */
    size_t n_handlers;

    struct revalidator *revalidators;  /* Flow revalidators. */
    size_t n_revalidators;

    struct latch exit_latch;           /* Tells child threads to exit. */

    /* Revalidation. */
    struct seq *reval_seq;             /* Incremented to force revalidation. */
    bool need_revalidate;              /* As indicated by 'reval_seq'. */
    bool reval_exit;                   /* Set by leader on 'exit_latch. */
    pthread_barrier_t reval_barrier;   /* Barrier used by revalidators. */
    struct dpif_flow_dump dump;        /* DPIF flow dump state. */
    long long int dump_duration;       /* Duration of the last flow dump. */
    struct seq *dump_seq;              /* Increments each dump iteration. */

    /* There are 'n_revalidators' ukey hmaps. Each revalidator retains a
     * reference to one of these for garbage collection.
     *
     * During the flow dump phase, revalidators insert into these with a random
     * distribution. During the garbage collection phase, each revalidator
     * takes care of garbage collecting one of these hmaps. */
    struct {
        struct ovs_mutex mutex;        /* Guards the following. */
        struct hmap hmap OVS_GUARDED;  /* Datapath flow keys. */
    } *ukeys;

    /* Datapath flow statistics. */
    unsigned int max_n_flows;
    unsigned int avg_n_flows;

    /* Following fields are accessed and modified by different threads. */
    atomic_uint flow_limit;            /* Datapath flow hard limit. */

    /* n_flows_mutex prevents multiple threads updating these concurrently. */
    atomic_ulong n_flows;           /* Number of flows in the datapath. */
    atomic_llong n_flows_timestamp;    /* Last time n_flows was updated. */
    struct ovs_mutex n_flows_mutex;
};

enum upcall_type {
    BAD_UPCALL,                 /* Some kind of bug somewhere. */
    MISS_UPCALL,                /* A flow miss.  */
    SFLOW_UPCALL,               /* sFlow sample. */
    FLOW_SAMPLE_UPCALL,         /* Per-flow sampling. */
    IPFIX_UPCALL                /* Per-bridge sampling. */
};

struct upcall {
    struct flow_miss *flow_miss;    /* This upcall's flow_miss. */

    /* Raw upcall plus data for keeping track of the memory backing it. */
    struct dpif_upcall dpif_upcall; /* As returned by dpif_recv() */
    struct ofpbuf upcall_buf;       /* Owns some data in 'dpif_upcall'. */
    uint64_t upcall_stub[512 / 8];  /* Buffer to reduce need for malloc(). */
};

/* 'udpif_key's are responsible for tracking the little bit of state udpif
 * needs to do flow expiration which can't be pulled directly from the
 * datapath.  They may be created or maintained by any revalidator during
 * the dump phase, but are owned by a single revalidator, and are destroyed
 * by that revalidator during the garbage-collection phase.
 *
 * While some elements of a udpif_key are protected by a mutex, the ukey itself
 * is not.  Therefore it is not safe to destroy a udpif_key except when all
 * revalidators are in garbage collection phase, or they aren't running. */
struct udpif_key {
    struct hmap_node hmap_node;     /* In parent revalidator 'ukeys' map. */

    /* These elements are read only once created, and therefore aren't
     * protected by a mutex. */
    const struct nlattr *key;      /* Datapath flow key. */
    size_t key_len;                /* Length of 'key'. */

    struct ovs_mutex mutex;                   /* Guards the following. */
    struct dpif_flow_stats stats OVS_GUARDED; /* Last known stats.*/
    long long int created OVS_GUARDED;        /* Estimate of creation time. */
    bool mark OVS_GUARDED;                    /* For mark and sweep garbage
                                                 collection. */
    bool flow_exists OVS_GUARDED;             /* Ensures flows are only deleted
                                                 once. */

    struct xlate_cache *xcache OVS_GUARDED;   /* Cache for xlate entries that
                                               * are affected by this ukey.
                                               * Used for stats and learning.*/
    struct odputil_keybuf key_buf;            /* Memory for 'key'. */
};

/* Flow miss batching.
 *
 * Some dpifs implement operations faster when you hand them off in a batch.
 * To allow batching, "struct flow_miss" queues the dpif-related work needed
 * for a given flow.  Each "struct flow_miss" corresponds to sending one or
 * more packets, plus possibly installing the flow in the dpif. */
struct flow_miss {
    struct hmap_node hmap_node;
    struct ofproto_dpif *ofproto;

    struct flow flow;
    const struct nlattr *key;
    size_t key_len;
    enum dpif_upcall_type upcall_type;
    struct dpif_flow_stats stats;
    odp_port_t odp_in_port;

    uint64_t slow_path_buf[128 / 8];
    struct odputil_keybuf mask_buf;

    struct xlate_out xout;

    bool put;
};

static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
static struct list all_udpifs = LIST_INITIALIZER(&all_udpifs);

static size_t read_upcalls(struct handler *,
                           struct upcall upcalls[FLOW_MISS_MAX_BATCH],
                           struct flow_miss miss_buf[FLOW_MISS_MAX_BATCH],
                           struct hmap *);
static void handle_upcalls(struct handler *, struct hmap *, struct upcall *,
                           size_t n_upcalls);
static void udpif_stop_threads(struct udpif *);
static void udpif_start_threads(struct udpif *, size_t n_handlers,
                                size_t n_revalidators);
static void *udpif_upcall_handler(void *);
static void *udpif_revalidator(void *);
static unsigned long udpif_get_n_flows(struct udpif *);
static void revalidate(struct revalidator *);
static void revalidator_sweep(struct revalidator *);
static void revalidator_purge(struct revalidator *);
static void upcall_unixctl_show(struct unixctl_conn *conn, int argc,
                                const char *argv[], void *aux);
static void upcall_unixctl_disable_megaflows(struct unixctl_conn *, int argc,
                                             const char *argv[], void *aux);
static void upcall_unixctl_enable_megaflows(struct unixctl_conn *, int argc,
                                            const char *argv[], void *aux);
static void upcall_unixctl_set_flow_limit(struct unixctl_conn *conn, int argc,
                                            const char *argv[], void *aux);

static struct udpif_key *ukey_create(const struct nlattr *key, size_t key_len,
                                     long long int used);
static void ukey_delete(struct revalidator *, struct udpif_key *);

static atomic_bool enable_megaflows = ATOMIC_VAR_INIT(true);

struct udpif *
udpif_create(struct dpif_backer *backer, struct dpif *dpif)
{
    static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
    struct udpif *udpif = xzalloc(sizeof *udpif);

    if (ovsthread_once_start(&once)) {
        unixctl_command_register("upcall/show", "", 0, 0, upcall_unixctl_show,
                                 NULL);
        unixctl_command_register("upcall/disable-megaflows", "", 0, 0,
                                 upcall_unixctl_disable_megaflows, NULL);
        unixctl_command_register("upcall/enable-megaflows", "", 0, 0,
                                 upcall_unixctl_enable_megaflows, NULL);
        unixctl_command_register("upcall/set-flow-limit", "", 1, 1,
                                 upcall_unixctl_set_flow_limit, NULL);
        ovsthread_once_done(&once);
    }

    udpif->dpif = dpif;
    udpif->backer = backer;
    atomic_init(&udpif->flow_limit, MIN(ofproto_flow_limit, 10000));
    udpif->secret = random_uint32();
    udpif->reval_seq = seq_create();
    udpif->dump_seq = seq_create();
    latch_init(&udpif->exit_latch);
    list_push_back(&all_udpifs, &udpif->list_node);
    atomic_init(&udpif->n_flows, 0);
    atomic_init(&udpif->n_flows_timestamp, LLONG_MIN);
    ovs_mutex_init(&udpif->n_flows_mutex);

    return udpif;
}

void
udpif_destroy(struct udpif *udpif)
{
    udpif_stop_threads(udpif);

    list_remove(&udpif->list_node);
    latch_destroy(&udpif->exit_latch);
    seq_destroy(udpif->reval_seq);
    seq_destroy(udpif->dump_seq);
    ovs_mutex_destroy(&udpif->n_flows_mutex);
    free(udpif);
}

/* Stops the handler and revalidator threads, must be enclosed in
 * ovsrcu quiescent state unless when destroying udpif. */
static void
udpif_stop_threads(struct udpif *udpif)
{
    if (udpif && (udpif->n_handlers != 0 || udpif->n_revalidators != 0)) {
        size_t i;

        latch_set(&udpif->exit_latch);

        for (i = 0; i < udpif->n_handlers; i++) {
            struct handler *handler = &udpif->handlers[i];

            xpthread_join(handler->thread, NULL);
        }

        for (i = 0; i < udpif->n_revalidators; i++) {
            xpthread_join(udpif->revalidators[i].thread, NULL);
        }

        for (i = 0; i < udpif->n_revalidators; i++) {
            struct revalidator *revalidator = &udpif->revalidators[i];

            /* Delete ukeys, and delete all flows from the datapath to prevent
             * double-counting stats. */
            revalidator_purge(revalidator);

            hmap_destroy(&udpif->ukeys[i].hmap);
            ovs_mutex_destroy(&udpif->ukeys[i].mutex);
        }

        latch_poll(&udpif->exit_latch);

        xpthread_barrier_destroy(&udpif->reval_barrier);

        free(udpif->revalidators);
        udpif->revalidators = NULL;
        udpif->n_revalidators = 0;

        free(udpif->handlers);
        udpif->handlers = NULL;
        udpif->n_handlers = 0;

        free(udpif->ukeys);
        udpif->ukeys = NULL;
    }
}

/* Starts the handler and revalidator threads, must be enclosed in
 * ovsrcu quiescent state. */
static void
udpif_start_threads(struct udpif *udpif, size_t n_handlers,
                    size_t n_revalidators)
{
    if (udpif && n_handlers && n_revalidators) {
        size_t i;

        udpif->n_handlers = n_handlers;
        udpif->n_revalidators = n_revalidators;

        udpif->handlers = xzalloc(udpif->n_handlers * sizeof *udpif->handlers);
        for (i = 0; i < udpif->n_handlers; i++) {
            struct handler *handler = &udpif->handlers[i];

            handler->udpif = udpif;
            handler->handler_id = i;
            handler->thread = ovs_thread_create(
                "handler", udpif_upcall_handler, handler);
        }

        xpthread_barrier_init(&udpif->reval_barrier, NULL,
                              udpif->n_revalidators);
        udpif->reval_exit = false;
        udpif->revalidators = xzalloc(udpif->n_revalidators
                                      * sizeof *udpif->revalidators);
        udpif->ukeys = xmalloc(sizeof *udpif->ukeys * n_revalidators);
        for (i = 0; i < udpif->n_revalidators; i++) {
            struct revalidator *revalidator = &udpif->revalidators[i];

            revalidator->udpif = udpif;
            hmap_init(&udpif->ukeys[i].hmap);
            ovs_mutex_init(&udpif->ukeys[i].mutex);
            revalidator->ukeys = &udpif->ukeys[i].hmap;
            revalidator->thread = ovs_thread_create(
                "revalidator", udpif_revalidator, revalidator);
        }
    }
}

/* Tells 'udpif' how many threads it should use to handle upcalls.
 * 'n_handlers' and 'n_revalidators' can never be zero.  'udpif''s
 * datapath handle must have packet reception enabled before starting
 * threads. */
void
udpif_set_threads(struct udpif *udpif, size_t n_handlers,
                  size_t n_revalidators)
{
    ovs_assert(udpif);
    ovs_assert(n_handlers && n_revalidators);

    ovsrcu_quiesce_start();
    if (udpif->n_handlers != n_handlers
        || udpif->n_revalidators != n_revalidators) {
        udpif_stop_threads(udpif);
    }

    if (!udpif->handlers && !udpif->revalidators) {
        int error;

        error = dpif_handlers_set(udpif->dpif, n_handlers);
        if (error) {
            VLOG_ERR("failed to configure handlers in dpif %s: %s",
                     dpif_name(udpif->dpif), ovs_strerror(error));
            return;
        }

        udpif_start_threads(udpif, n_handlers, n_revalidators);
    }
    ovsrcu_quiesce_end();
}

/* Waits for all ongoing upcall translations to complete.  This ensures that
 * there are no transient references to any removed ofprotos (or other
 * objects).  In particular, this should be called after an ofproto is removed
 * (e.g. via xlate_remove_ofproto()) but before it is destroyed. */
void
udpif_synchronize(struct udpif *udpif)
{
    /* This is stronger than necessary.  It would be sufficient to ensure
     * (somehow) that each handler and revalidator thread had passed through
     * its main loop once. */
    size_t n_handlers = udpif->n_handlers;
    size_t n_revalidators = udpif->n_revalidators;

    ovsrcu_quiesce_start();
    udpif_stop_threads(udpif);
    udpif_start_threads(udpif, n_handlers, n_revalidators);
    ovsrcu_quiesce_end();
}

/* Notifies 'udpif' that something changed which may render previous
 * xlate_actions() results invalid. */
void
udpif_revalidate(struct udpif *udpif)
{
    seq_change(udpif->reval_seq);
}

/* Returns a seq which increments every time 'udpif' pulls stats from the
 * datapath.  Callers can use this to get a sense of when might be a good time
 * to do periodic work which relies on relatively up to date statistics. */
struct seq *
udpif_dump_seq(struct udpif *udpif)
{
    return udpif->dump_seq;
}

void
udpif_get_memory_usage(struct udpif *udpif, struct simap *usage)
{
    size_t i;

    simap_increase(usage, "handlers", udpif->n_handlers);

    simap_increase(usage, "revalidators", udpif->n_revalidators);
    for (i = 0; i < udpif->n_revalidators; i++) {
        ovs_mutex_lock(&udpif->ukeys[i].mutex);
        simap_increase(usage, "udpif keys", hmap_count(&udpif->ukeys[i].hmap));
        ovs_mutex_unlock(&udpif->ukeys[i].mutex);
    }
}

/* Remove flows from a single datapath. */
void
udpif_flush(struct udpif *udpif)
{
    size_t n_handlers, n_revalidators;

    n_handlers = udpif->n_handlers;
    n_revalidators = udpif->n_revalidators;

    ovsrcu_quiesce_start();

    udpif_stop_threads(udpif);
    dpif_flow_flush(udpif->dpif);
    udpif_start_threads(udpif, n_handlers, n_revalidators);

    ovsrcu_quiesce_end();
}

/* Removes all flows from all datapaths. */
static void
udpif_flush_all_datapaths(void)
{
    struct udpif *udpif;

    LIST_FOR_EACH (udpif, list_node, &all_udpifs) {
        udpif_flush(udpif);
    }
}


static unsigned long
udpif_get_n_flows(struct udpif *udpif)
{
    long long int time, now;
    unsigned long flow_count;

    now = time_msec();
    atomic_read(&udpif->n_flows_timestamp, &time);
    if (time < now - 100 && !ovs_mutex_trylock(&udpif->n_flows_mutex)) {
        struct dpif_dp_stats stats;

        atomic_store(&udpif->n_flows_timestamp, now);
        dpif_get_dp_stats(udpif->dpif, &stats);
        flow_count = stats.n_flows;
        atomic_store(&udpif->n_flows, flow_count);
        ovs_mutex_unlock(&udpif->n_flows_mutex);
    } else {
        atomic_read(&udpif->n_flows, &flow_count);
    }
    return flow_count;
}

/* The upcall handler thread tries to read a batch of FLOW_MISS_MAX_BATCH
 * upcalls from dpif, processes the batch and installs corresponding flows
 * in dpif. */
static void *
udpif_upcall_handler(void *arg)
{
    struct handler *handler = arg;
    struct udpif *udpif = handler->udpif;
    struct hmap misses = HMAP_INITIALIZER(&misses);

    while (!latch_is_set(&handler->udpif->exit_latch)) {
        struct upcall upcalls[FLOW_MISS_MAX_BATCH];
        struct flow_miss miss_buf[FLOW_MISS_MAX_BATCH];
        struct flow_miss *miss;
        size_t n_upcalls, i;

        n_upcalls = read_upcalls(handler, upcalls, miss_buf, &misses);
        if (!n_upcalls) {
            dpif_recv_wait(udpif->dpif, handler->handler_id);
            latch_wait(&udpif->exit_latch);
            poll_block();
        } else {
            handle_upcalls(handler, &misses, upcalls, n_upcalls);

            HMAP_FOR_EACH (miss, hmap_node, &misses) {
                xlate_out_uninit(&miss->xout);
            }
            hmap_clear(&misses);
            for (i = 0; i < n_upcalls; i++) {
                ofpbuf_uninit(&upcalls[i].dpif_upcall.packet);
                ofpbuf_uninit(&upcalls[i].upcall_buf);
            }
        }
        coverage_clear();
    }
    hmap_destroy(&misses);

    return NULL;
}

static void *
udpif_revalidator(void *arg)
{
    /* Used by all revalidators. */
    struct revalidator *revalidator = arg;
    struct udpif *udpif = revalidator->udpif;
    bool leader = revalidator == &udpif->revalidators[0];

    /* Used only by the leader. */
    long long int start_time = 0;
    uint64_t last_reval_seq = 0;
    unsigned int flow_limit = 0;
    size_t n_flows = 0;

    revalidator->id = ovsthread_id_self();
    for (;;) {
        if (leader) {
            uint64_t reval_seq;

            reval_seq = seq_read(udpif->reval_seq);
            udpif->need_revalidate = last_reval_seq != reval_seq;
            last_reval_seq = reval_seq;

            n_flows = udpif_get_n_flows(udpif);
            udpif->max_n_flows = MAX(n_flows, udpif->max_n_flows);
            udpif->avg_n_flows = (udpif->avg_n_flows + n_flows) / 2;

            /* Only the leader checks the exit latch to prevent a race where
             * some threads think it's true and exit and others think it's
             * false and block indefinitely on the reval_barrier */
            udpif->reval_exit = latch_is_set(&udpif->exit_latch);

            start_time = time_msec();
            if (!udpif->reval_exit) {
                dpif_flow_dump_start(&udpif->dump, udpif->dpif);
            }
        }

        /* Wait for the leader to start the flow dump. */
        xpthread_barrier_wait(&udpif->reval_barrier);
        if (udpif->reval_exit) {
            break;
        }
        revalidate(revalidator);

        /* Wait for all flows to have been dumped before we garbage collect. */
        xpthread_barrier_wait(&udpif->reval_barrier);
        revalidator_sweep(revalidator);

        /* Wait for all revalidators to finish garbage collection. */
        xpthread_barrier_wait(&udpif->reval_barrier);

        if (leader) {
            long long int duration;

            dpif_flow_dump_done(&udpif->dump);
            seq_change(udpif->dump_seq);

            duration = MAX(time_msec() - start_time, 1);
            atomic_read(&udpif->flow_limit, &flow_limit);
            udpif->dump_duration = duration;
            if (duration > 2000) {
                flow_limit /= duration / 1000;
            } else if (duration > 1300) {
                flow_limit = flow_limit * 3 / 4;
            } else if (duration < 1000 && n_flows > 2000
                       && flow_limit < n_flows * 1000 / duration) {
                flow_limit += 1000;
            }
            flow_limit = MIN(ofproto_flow_limit, MAX(flow_limit, 1000));
            atomic_store(&udpif->flow_limit, flow_limit);

            if (duration > 2000) {
                VLOG_INFO("Spent an unreasonably long %lldms dumping flows",
                          duration);
            }

            poll_timer_wait_until(start_time + MIN(ofproto_max_idle, 500));
            seq_wait(udpif->reval_seq, last_reval_seq);
            latch_wait(&udpif->exit_latch);
            poll_block();
        }
    }

    return NULL;
}

static enum upcall_type
classify_upcall(const struct upcall *upcall)
{
    const struct dpif_upcall *dpif_upcall = &upcall->dpif_upcall;
    union user_action_cookie cookie;
    size_t userdata_len;

    /* First look at the upcall type. */
    switch (dpif_upcall->type) {
    case DPIF_UC_ACTION:
        break;

    case DPIF_UC_MISS:
        return MISS_UPCALL;

    case DPIF_N_UC_TYPES:
    default:
        VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
                     dpif_upcall->type);
        return BAD_UPCALL;
    }

    /* "action" upcalls need a closer look. */
    if (!dpif_upcall->userdata) {
        VLOG_WARN_RL(&rl, "action upcall missing cookie");
        return BAD_UPCALL;
    }
    userdata_len = nl_attr_get_size(dpif_upcall->userdata);
    if (userdata_len < sizeof cookie.type
        || userdata_len > sizeof cookie) {
        VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %"PRIuSIZE,
                     userdata_len);
        return BAD_UPCALL;
    }
    memset(&cookie, 0, sizeof cookie);
    memcpy(&cookie, nl_attr_get(dpif_upcall->userdata), userdata_len);
    if (userdata_len == MAX(8, sizeof cookie.sflow)
        && cookie.type == USER_ACTION_COOKIE_SFLOW) {
        return SFLOW_UPCALL;
    } else if (userdata_len == MAX(8, sizeof cookie.slow_path)
               && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
        return MISS_UPCALL;
    } else if (userdata_len == MAX(8, sizeof cookie.flow_sample)
               && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
        return FLOW_SAMPLE_UPCALL;
    } else if (userdata_len == MAX(8, sizeof cookie.ipfix)
               && cookie.type == USER_ACTION_COOKIE_IPFIX) {
        return IPFIX_UPCALL;
    } else {
        VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
                     " and size %"PRIuSIZE, cookie.type, userdata_len);
        return BAD_UPCALL;
    }
}

/* Calculates slow path actions for 'xout'.  'buf' must statically be
 * initialized with at least 128 bytes of space. */
static void
compose_slow_path(struct udpif *udpif, struct xlate_out *xout,
                  struct flow *flow, odp_port_t odp_in_port,
                  struct ofpbuf *buf)
{
    union user_action_cookie cookie;
    odp_port_t port;
    uint32_t pid;

    cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
    cookie.slow_path.unused = 0;
    cookie.slow_path.reason = xout->slow;

    port = xout->slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)
        ? ODPP_NONE
        : odp_in_port;
    pid = dpif_port_get_pid(udpif->dpif, port, flow_hash_5tuple(flow, 0));
    odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, buf);
}

static struct flow_miss *
flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
               const struct flow *flow, uint32_t hash)
{
    struct flow_miss *miss;

    HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
        if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
            return miss;
        }
    }

    return NULL;
}

/* Reads and classifies upcalls.  Returns the number of upcalls successfully
 * read. */
static size_t
read_upcalls(struct handler *handler,
             struct upcall upcalls[FLOW_MISS_MAX_BATCH],
             struct flow_miss miss_buf[FLOW_MISS_MAX_BATCH],
             struct hmap *misses)
{
    struct udpif *udpif = handler->udpif;
    size_t i;
    size_t n_misses = 0;
    size_t n_upcalls = 0;

    /*
     * Try reading FLOW_MISS_MAX_BATCH upcalls from dpif.
     *
     * Extract the flow from each upcall.  Construct in 'misses' a hash table
     * that maps each unique flow to a 'struct flow_miss'.
     *
     * Most commonly there is a single packet per flow_miss, but there are
     * several reasons why there might be more than one, e.g.:
     *
     *   - The dpif packet interface does not support TSO (or UFO, etc.), so a
     *     large packet sent to userspace is split into a sequence of smaller
     *     ones.
     *
     *   - A stream of quickly arriving packets in an established "slow-pathed"
     *     flow.
     *
     *   - Rarely, a stream of quickly arriving packets in a flow not yet
     *     established.  (This is rare because most protocols do not send
     *     multiple back-to-back packets before receiving a reply from the
     *     other end of the connection, which gives OVS a chance to set up a
     *     datapath flow.)
     */
    for (i = 0; i < FLOW_MISS_MAX_BATCH; i++) {
        struct upcall *upcall = &upcalls[n_upcalls];
        struct flow_miss *miss = &miss_buf[n_misses];
        struct dpif_upcall *dupcall;
        struct ofpbuf *packet;
        struct flow_miss *existing_miss;
        struct ofproto_dpif *ofproto;
        struct dpif_sflow *sflow;
        struct dpif_ipfix *ipfix;
        struct flow flow;
        enum upcall_type type;
        odp_port_t odp_in_port;
        int error;

        ofpbuf_use_stub(&upcall->upcall_buf, upcall->upcall_stub,
                        sizeof upcall->upcall_stub);
        error = dpif_recv(udpif->dpif, handler->handler_id,
                          &upcall->dpif_upcall, &upcall->upcall_buf);
        if (error) {
            ofpbuf_uninit(&upcall->upcall_buf);
            break;
        }

        dupcall = &upcall->dpif_upcall;
        packet = &dupcall->packet;
        error = xlate_receive(udpif->backer, packet, dupcall->key,
                              dupcall->key_len, &flow,
                              &ofproto, &ipfix, &sflow, NULL, &odp_in_port);
        if (error) {
            if (error == ENODEV) {
                /* Received packet on datapath port for which we couldn't
                 * associate an ofproto.  This can happen if a port is removed
                 * while traffic is being received.  Print a rate-limited
                 * message in case it happens frequently.  Install a drop flow
                 * so that future packets of the flow are inexpensively dropped
                 * in the kernel. */
                VLOG_INFO_RL(&rl, "received packet on unassociated datapath "
                             "port %"PRIu32, odp_in_port);
                dpif_flow_put(udpif->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
                              dupcall->key, dupcall->key_len, NULL, 0, NULL, 0,
                              NULL);
            }
            goto destroy_upcall;
        }

        type = classify_upcall(upcall);
        if (type == MISS_UPCALL) {
            uint32_t hash;
            struct pkt_metadata md = pkt_metadata_from_flow(&flow);

            flow_extract(packet, &md, &miss->flow);
            hash = flow_hash(&miss->flow, 0);
            existing_miss = flow_miss_find(misses, ofproto, &miss->flow,
                                           hash);
            if (!existing_miss) {
                hmap_insert(misses, &miss->hmap_node, hash);
                miss->ofproto = ofproto;
                miss->key = dupcall->key;
                miss->key_len = dupcall->key_len;
                miss->upcall_type = dupcall->type;
                miss->stats.n_packets = 0;
                miss->stats.n_bytes = 0;
                miss->stats.used = time_msec();
                miss->stats.tcp_flags = 0;
                miss->odp_in_port = odp_in_port;
                miss->put = false;
                n_misses++;
            } else {
                miss = existing_miss;
            }
            miss->stats.tcp_flags |= ntohs(miss->flow.tcp_flags);
            miss->stats.n_bytes += ofpbuf_size(packet);
            miss->stats.n_packets++;

            upcall->flow_miss = miss;
            n_upcalls++;
            continue;
        }

        switch (type) {
        case SFLOW_UPCALL:
            if (sflow) {
                union user_action_cookie cookie;

                memset(&cookie, 0, sizeof cookie);
                memcpy(&cookie, nl_attr_get(dupcall->userdata),
                       sizeof cookie.sflow);
                dpif_sflow_received(sflow, packet, &flow, odp_in_port,
                                    &cookie);
            }
            break;
        case IPFIX_UPCALL:
            if (ipfix) {
                dpif_ipfix_bridge_sample(ipfix, packet, &flow);
            }
            break;
        case FLOW_SAMPLE_UPCALL:
            if (ipfix) {
                union user_action_cookie cookie;

                memset(&cookie, 0, sizeof cookie);
                memcpy(&cookie, nl_attr_get(dupcall->userdata),
                       sizeof cookie.flow_sample);

                /* The flow reflects exactly the contents of the packet.
                 * Sample the packet using it. */
                dpif_ipfix_flow_sample(ipfix, packet, &flow,
                                       cookie.flow_sample.collector_set_id,
                                       cookie.flow_sample.probability,
                                       cookie.flow_sample.obs_domain_id,
                                       cookie.flow_sample.obs_point_id);
            }
            break;
        case BAD_UPCALL:
            break;
        case MISS_UPCALL:
            OVS_NOT_REACHED();
        }

        dpif_ipfix_unref(ipfix);
        dpif_sflow_unref(sflow);

destroy_upcall:
        ofpbuf_uninit(&upcall->dpif_upcall.packet);
        ofpbuf_uninit(&upcall->upcall_buf);
    }

    return n_upcalls;
}

static void
handle_upcalls(struct handler *handler, struct hmap *misses,
               struct upcall *upcalls, size_t n_upcalls)
{
    struct udpif *udpif = handler->udpif;
    struct dpif_op *opsp[FLOW_MISS_MAX_BATCH * 2];
    struct dpif_op ops[FLOW_MISS_MAX_BATCH * 2];
    struct flow_miss *miss;
    size_t n_ops, i;
    unsigned int flow_limit;
    bool fail_open, may_put;

    atomic_read(&udpif->flow_limit, &flow_limit);
    may_put = udpif_get_n_flows(udpif) < flow_limit;

    /* Initialize each 'struct flow_miss's ->xout.
     *
     * We do this per-flow_miss rather than per-packet because, most commonly,
     * all the packets in a flow can use the same translation.
     *
     * We can't do this in the previous loop because we need the TCP flags for
     * all the packets in each miss. */
    fail_open = false;
    HMAP_FOR_EACH (miss, hmap_node, misses) {
        struct xlate_in xin;

        xlate_in_init(&xin, miss->ofproto, &miss->flow, NULL,
                      miss->stats.tcp_flags, NULL);
        xin.may_learn = true;

        if (miss->upcall_type == DPIF_UC_MISS) {
            xin.resubmit_stats = &miss->stats;
        } else {
            /* For non-miss upcalls, there's a flow in the datapath which this
             * packet was accounted to.  Presumably the revalidators will deal
             * with pushing its stats eventually. */
        }

        xlate_actions(&xin, &miss->xout);
        fail_open = fail_open || miss->xout.fail_open;
    }

    /* Now handle the packets individually in order of arrival.  In the common
     * case each packet of a miss can share the same actions, but slow-pathed
     * packets need to be translated individually:
     *
     *   - For SLOW_CFM, SLOW_LACP, SLOW_STP, and SLOW_BFD, translation is what
     *     processes received packets for these protocols.
     *
     *   - For SLOW_CONTROLLER, translation sends the packet to the OpenFlow
     *     controller.
     *
     * The loop fills 'ops' with an array of operations to execute in the
     * datapath. */
    n_ops = 0;
    for (i = 0; i < n_upcalls; i++) {
        struct upcall *upcall = &upcalls[i];
        struct flow_miss *miss = upcall->flow_miss;
        struct ofpbuf *packet = &upcall->dpif_upcall.packet;
        struct dpif_op *op;
        ovs_be16 flow_vlan_tci;

        /* Save a copy of flow.vlan_tci in case it is changed to
         * generate proper mega flow masks for VLAN splinter flows. */
        flow_vlan_tci = miss->flow.vlan_tci;

        if (miss->xout.slow) {
            struct xlate_in xin;

            xlate_in_init(&xin, miss->ofproto, &miss->flow, NULL, 0, packet);
            xlate_actions_for_side_effects(&xin);
        }

        if (miss->flow.in_port.ofp_port
            != vsp_realdev_to_vlandev(miss->ofproto,
                                      miss->flow.in_port.ofp_port,
                                      miss->flow.vlan_tci)) {
            /* This packet was received on a VLAN splinter port.  We
             * added a VLAN to the packet to make the packet resemble
             * the flow, but the actions were composed assuming that
             * the packet contained no VLAN.  So, we must remove the
             * VLAN header from the packet before trying to execute the
             * actions. */
            if (ofpbuf_size(&miss->xout.odp_actions)) {
                eth_pop_vlan(packet);
            }

            /* Remove the flow vlan tags inserted by vlan splinter logic
             * to ensure megaflow masks generated match the data path flow. */
            miss->flow.vlan_tci = 0;
        }

        /* Do not install a flow into the datapath if:
         *
         *    - The datapath already has too many flows.
         *
         *    - An earlier iteration of this loop already put the same flow.
         *
         *    - We received this packet via some flow installed in the kernel
         *      already. */
        if (may_put
            && !miss->put
            && upcall->dpif_upcall.type == DPIF_UC_MISS) {
            struct ofpbuf mask;
            bool megaflow;

            miss->put = true;

            atomic_read(&enable_megaflows, &megaflow);
            ofpbuf_use_stack(&mask, &miss->mask_buf, sizeof miss->mask_buf);
            if (megaflow) {
                size_t max_mpls;
                bool recirc;

                recirc = ofproto_dpif_get_enable_recirc(miss->ofproto);
                max_mpls = ofproto_dpif_get_max_mpls_depth(miss->ofproto);
                odp_flow_key_from_mask(&mask, &miss->xout.wc.masks,
                                       &miss->flow, UINT32_MAX, max_mpls,
                                       recirc);
            }

            op = &ops[n_ops++];
            op->type = DPIF_OP_FLOW_PUT;
            op->u.flow_put.flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
            op->u.flow_put.key = miss->key;
            op->u.flow_put.key_len = miss->key_len;
            op->u.flow_put.mask = ofpbuf_data(&mask);
            op->u.flow_put.mask_len = ofpbuf_size(&mask);
            op->u.flow_put.stats = NULL;

            if (!miss->xout.slow) {
                op->u.flow_put.actions = ofpbuf_data(&miss->xout.odp_actions);
                op->u.flow_put.actions_len = ofpbuf_size(&miss->xout.odp_actions);
            } else {
                struct ofpbuf buf;

                ofpbuf_use_stack(&buf, miss->slow_path_buf,
                                 sizeof miss->slow_path_buf);
                compose_slow_path(udpif, &miss->xout, &miss->flow,
                                  miss->odp_in_port, &buf);
                op->u.flow_put.actions = ofpbuf_data(&buf);
                op->u.flow_put.actions_len = ofpbuf_size(&buf);
            }
        }

        /*
         * The 'miss' may be shared by multiple upcalls. Restore
         * the saved flow vlan_tci field before processing the next
         * upcall. */
        miss->flow.vlan_tci = flow_vlan_tci;

        if (ofpbuf_size(&miss->xout.odp_actions)) {

            op = &ops[n_ops++];
            op->type = DPIF_OP_EXECUTE;
            op->u.execute.packet = packet;
            odp_key_to_pkt_metadata(miss->key, miss->key_len,
                                    &op->u.execute.md);
            op->u.execute.actions = ofpbuf_data(&miss->xout.odp_actions);
            op->u.execute.actions_len = ofpbuf_size(&miss->xout.odp_actions);
            op->u.execute.needs_help = (miss->xout.slow & SLOW_ACTION) != 0;
        }
    }

    /* Special case for fail-open mode.
     *
     * If we are in fail-open mode, but we are connected to a controller too,
     * then we should send the packet up to the controller in the hope that it
     * will try to set up a flow and thereby allow us to exit fail-open.
     *
     * See the top-level comment in fail-open.c for more information.
     *
     * Copy packets before they are modified by execution. */
    if (fail_open) {
        for (i = 0; i < n_upcalls; i++) {
            struct upcall *upcall = &upcalls[i];
            struct flow_miss *miss = upcall->flow_miss;
            struct ofpbuf *packet = &upcall->dpif_upcall.packet;
            struct ofproto_packet_in *pin;

            pin = xmalloc(sizeof *pin);
            pin->up.packet = xmemdup(ofpbuf_data(packet), ofpbuf_size(packet));
            pin->up.packet_len = ofpbuf_size(packet);
            pin->up.reason = OFPR_NO_MATCH;
            pin->up.table_id = 0;
            pin->up.cookie = OVS_BE64_MAX;
            flow_get_metadata(&miss->flow, &pin->up.fmd);
            pin->send_len = 0; /* Not used for flow table misses. */
            pin->miss_type = OFPROTO_PACKET_IN_NO_MISS;
            ofproto_dpif_send_packet_in(miss->ofproto, pin);
        }
    }

    /* Execute batch. */
    for (i = 0; i < n_ops; i++) {
        opsp[i] = &ops[i];
    }
    dpif_operate(udpif->dpif, opsp, n_ops);
}

/* Must be called with udpif->ukeys[hash % udpif->n_revalidators].mutex. */
static struct udpif_key *
ukey_lookup__(struct udpif *udpif, const struct nlattr *key, size_t key_len,
              uint32_t hash)
{
    struct udpif_key *ukey;
    struct hmap *hmap = &udpif->ukeys[hash % udpif->n_revalidators].hmap;

    HMAP_FOR_EACH_WITH_HASH (ukey, hmap_node, hash, hmap) {
        if (ukey->key_len == key_len && !memcmp(ukey->key, key, key_len)) {
            return ukey;
        }
    }
    return NULL;
}

static struct udpif_key *
ukey_lookup(struct udpif *udpif, const struct nlattr *key, size_t key_len,
            uint32_t hash)
{
    struct udpif_key *ukey;
    uint32_t idx = hash % udpif->n_revalidators;

    ovs_mutex_lock(&udpif->ukeys[idx].mutex);
    ukey = ukey_lookup__(udpif, key, key_len, hash);
    ovs_mutex_unlock(&udpif->ukeys[idx].mutex);

    return ukey;
}

static struct udpif_key *
ukey_create(const struct nlattr *key, size_t key_len, long long int used)
{
    struct udpif_key *ukey = xmalloc(sizeof *ukey);
    ovs_mutex_init(&ukey->mutex);

    ukey->key = (struct nlattr *) &ukey->key_buf;
    memcpy(&ukey->key_buf, key, key_len);
    ukey->key_len = key_len;

    ovs_mutex_lock(&ukey->mutex);
    ukey->mark = false;
    ukey->flow_exists = true;
    ukey->created = used ? used : time_msec();
    memset(&ukey->stats, 0, sizeof ukey->stats);
    ukey->xcache = NULL;
    ovs_mutex_unlock(&ukey->mutex);

    return ukey;
}

/* Checks for a ukey in 'udpif->ukeys' with the same 'ukey->key' and 'hash',
 * and inserts 'ukey' if it does not exist.
 *
 * Returns true if 'ukey' was inserted into 'udpif->ukeys', false otherwise. */
static bool
udpif_insert_ukey(struct udpif *udpif, struct udpif_key *ukey, uint32_t hash)
{
    struct udpif_key *duplicate;
    uint32_t idx = hash % udpif->n_revalidators;
    bool ok;

    ovs_mutex_lock(&udpif->ukeys[idx].mutex);
    duplicate = ukey_lookup__(udpif, ukey->key, ukey->key_len, hash);
    if (duplicate) {
        ok = false;
    } else {
        hmap_insert(&udpif->ukeys[idx].hmap, &ukey->hmap_node, hash);
        ok = true;
    }
    ovs_mutex_unlock(&udpif->ukeys[idx].mutex);

    return ok;
}

static void
ukey_delete(struct revalidator *revalidator, struct udpif_key *ukey)
    OVS_NO_THREAD_SAFETY_ANALYSIS
{
    if (revalidator) {
        hmap_remove(revalidator->ukeys, &ukey->hmap_node);
    }
    xlate_cache_delete(ukey->xcache);
    ovs_mutex_destroy(&ukey->mutex);
    free(ukey);
}

static bool
should_revalidate(uint64_t packets, long long int used)
{
    long long int metric, now, duration;

    /* Calculate the mean time between seeing these packets. If this
     * exceeds the threshold, then delete the flow rather than performing
     * costly revalidation for flows that aren't being hit frequently.
     *
     * This is targeted at situations where the dump_duration is high (~1s),
     * and revalidation is triggered by a call to udpif_revalidate(). In
     * these situations, revalidation of all flows causes fluctuations in the
     * flow_limit due to the interaction with the dump_duration and max_idle.
     * This tends to result in deletion of low-throughput flows anyway, so
     * skip the revalidation and just delete those flows. */
    packets = MAX(packets, 1);
    now = MAX(used, time_msec());
    duration = now - used;
    metric = duration / packets;

    if (metric > 200) {
        return false;
    }
    return true;
}

static bool
revalidate_ukey(struct udpif *udpif, struct udpif_key *ukey,
                const struct nlattr *mask, size_t mask_len,
                const struct nlattr *actions, size_t actions_len,
                const struct dpif_flow_stats *stats)
{
    uint64_t slow_path_buf[128 / 8];
    struct xlate_out xout, *xoutp;
    struct netflow *netflow;
    struct ofproto_dpif *ofproto;
    struct dpif_flow_stats push;
    struct ofpbuf xout_actions;
    struct flow flow, dp_mask;
    uint32_t *dp32, *xout32;
    odp_port_t odp_in_port;
    struct xlate_in xin;
    long long int last_used;
    int error;
    size_t i;
    bool may_learn, ok;

    ok = false;
    xoutp = NULL;
    netflow = NULL;

    ovs_mutex_lock(&ukey->mutex);
    last_used = ukey->stats.used;
    push.used = stats->used;
    push.tcp_flags = stats->tcp_flags;
    push.n_packets = stats->n_packets > ukey->stats.n_packets
        ? stats->n_packets - ukey->stats.n_packets
        : 0;
    push.n_bytes = stats->n_bytes > ukey->stats.n_bytes
        ? stats->n_bytes - ukey->stats.n_bytes
        : 0;

    if (!ukey->flow_exists) {
        /* Don't bother revalidating if the flow was already deleted. */
        goto exit;
    }

    if (udpif->need_revalidate && last_used
        && !should_revalidate(push.n_packets, last_used)) {
        ok = false;
        goto exit;
    }

    /* We will push the stats, so update the ukey stats cache. */
    ukey->stats = *stats;
    if (!push.n_packets && !udpif->need_revalidate) {
        ok = true;
        goto exit;
    }

    may_learn = push.n_packets > 0;
    if (ukey->xcache && !udpif->need_revalidate) {
        xlate_push_stats(ukey->xcache, may_learn, &push);
        ok = true;
        goto exit;
    }

    error = xlate_receive(udpif->backer, NULL, ukey->key, ukey->key_len, &flow,
                          &ofproto, NULL, NULL, &netflow, &odp_in_port);
    if (error) {
        goto exit;
    }

    if (udpif->need_revalidate) {
        xlate_cache_clear(ukey->xcache);
    }
    if (!ukey->xcache) {
        ukey->xcache = xlate_cache_new();
    }

    xlate_in_init(&xin, ofproto, &flow, NULL, push.tcp_flags, NULL);
    xin.resubmit_stats = push.n_packets ? &push : NULL;
    xin.xcache = ukey->xcache;
    xin.may_learn = may_learn;
    xin.skip_wildcards = !udpif->need_revalidate;
    xlate_actions(&xin, &xout);
    xoutp = &xout;

    if (!udpif->need_revalidate) {
        ok = true;
        goto exit;
    }

    if (!xout.slow) {
        ofpbuf_use_const(&xout_actions, ofpbuf_data(&xout.odp_actions),
                         ofpbuf_size(&xout.odp_actions));
    } else {
        ofpbuf_use_stack(&xout_actions, slow_path_buf, sizeof slow_path_buf);
        compose_slow_path(udpif, &xout, &flow, odp_in_port, &xout_actions);
    }

    if (actions_len != ofpbuf_size(&xout_actions)
        || memcmp(ofpbuf_data(&xout_actions), actions, actions_len)) {
        goto exit;
    }

    if (odp_flow_key_to_mask(mask, mask_len, &dp_mask, &flow)
        == ODP_FIT_ERROR) {
        goto exit;
    }

    /* Since the kernel is free to ignore wildcarded bits in the mask, we can't
     * directly check that the masks are the same.  Instead we check that the
     * mask in the kernel is more specific i.e. less wildcarded, than what
     * we've calculated here.  This guarantees we don't catch any packets we
     * shouldn't with the megaflow. */
    dp32 = (uint32_t *) &dp_mask;
    xout32 = (uint32_t *) &xout.wc.masks;
    for (i = 0; i < FLOW_U32S; i++) {
        if ((dp32[i] | xout32[i]) != dp32[i]) {
            goto exit;
        }
    }
    ok = true;

exit:
    ovs_mutex_unlock(&ukey->mutex);
    if (netflow) {
        if (!ok) {
            netflow_expire(netflow, &flow);
            netflow_flow_clear(netflow, &flow);
        }
        netflow_unref(netflow);
    }
    xlate_out_uninit(xoutp);
    return ok;
}

struct dump_op {
    struct udpif_key *ukey;
    struct dpif_flow_stats stats; /* Stats for 'op'. */
    struct dpif_op op;            /* Flow del operation. */
};

static void
dump_op_init(struct dump_op *op, const struct nlattr *key, size_t key_len,
             struct udpif_key *ukey)
{
    op->ukey = ukey;
    op->op.type = DPIF_OP_FLOW_DEL;
    op->op.u.flow_del.key = key;
    op->op.u.flow_del.key_len = key_len;
    op->op.u.flow_del.stats = &op->stats;
}

static void
push_dump_ops__(struct udpif *udpif, struct dump_op *ops, size_t n_ops)
{
    struct dpif_op *opsp[REVALIDATE_MAX_BATCH];
    size_t i;

    ovs_assert(n_ops <= REVALIDATE_MAX_BATCH);
    for (i = 0; i < n_ops; i++) {
        opsp[i] = &ops[i].op;
    }
    dpif_operate(udpif->dpif, opsp, n_ops);

    for (i = 0; i < n_ops; i++) {
        struct dump_op *op = &ops[i];
        struct dpif_flow_stats *push, *stats, push_buf;

        stats = op->op.u.flow_del.stats;
        if (op->ukey) {
            push = &push_buf;
            ovs_mutex_lock(&op->ukey->mutex);
            push->used = MAX(stats->used, op->ukey->stats.used);
            push->tcp_flags = stats->tcp_flags | op->ukey->stats.tcp_flags;
            push->n_packets = stats->n_packets - op->ukey->stats.n_packets;
            push->n_bytes = stats->n_bytes - op->ukey->stats.n_bytes;
            ovs_mutex_unlock(&op->ukey->mutex);
        } else {
            push = stats;
        }

        if (push->n_packets || netflow_exists()) {
            struct ofproto_dpif *ofproto;
            struct netflow *netflow;
            struct flow flow;
            bool may_learn;

            may_learn = push->n_packets > 0;
            if (op->ukey) {
                ovs_mutex_lock(&op->ukey->mutex);
                if (op->ukey->xcache) {
                    xlate_push_stats(op->ukey->xcache, may_learn, push);
                    ovs_mutex_unlock(&op->ukey->mutex);
                    continue;
                }
                ovs_mutex_unlock(&op->ukey->mutex);
            }

            if (!xlate_receive(udpif->backer, NULL, op->op.u.flow_del.key,
                               op->op.u.flow_del.key_len, &flow, &ofproto,
                               NULL, NULL, &netflow, NULL)) {
                struct xlate_in xin;

                xlate_in_init(&xin, ofproto, &flow, NULL, push->tcp_flags,
                              NULL);
                xin.resubmit_stats = push->n_packets ? push : NULL;
                xin.may_learn = may_learn;
                xin.skip_wildcards = true;
                xlate_actions_for_side_effects(&xin);

                if (netflow) {
                    netflow_expire(netflow, &flow);
                    netflow_flow_clear(netflow, &flow);
                    netflow_unref(netflow);
                }
            }
        }
    }
}

static void
push_dump_ops(struct revalidator *revalidator,
              struct dump_op *ops, size_t n_ops)
{
    int i;

    push_dump_ops__(revalidator->udpif, ops, n_ops);
    for (i = 0; i < n_ops; i++) {
        ukey_delete(revalidator, ops[i].ukey);
    }
}

static void
revalidate(struct revalidator *revalidator)
{
    struct udpif *udpif = revalidator->udpif;

    struct dump_op ops[REVALIDATE_MAX_BATCH];
    const struct nlattr *key, *mask, *actions;
    size_t key_len, mask_len, actions_len;
    const struct dpif_flow_stats *stats;
    long long int now;
    unsigned int flow_limit;
    size_t n_ops;
    void *state;

    n_ops = 0;
    now = time_msec();
    atomic_read(&udpif->flow_limit, &flow_limit);

    dpif_flow_dump_state_init(udpif->dpif, &state);
    while (dpif_flow_dump_next(&udpif->dump, state, &key, &key_len, &mask,
                               &mask_len, &actions, &actions_len, &stats)) {
        struct udpif_key *ukey;
        bool mark, may_destroy;
        long long int used, max_idle;
        uint32_t hash;
        size_t n_flows;

        hash = hash_bytes(key, key_len, udpif->secret);
        ukey = ukey_lookup(udpif, key, key_len, hash);

        used = stats->used;
        if (!used && ukey) {
            ovs_mutex_lock(&ukey->mutex);

            if (ukey->mark || !ukey->flow_exists) {
                /* The flow has already been dumped. This can occasionally
                 * occur if the datapath is changed in the middle of a flow
                 * dump. Rather than perform the same work twice, skip the
                 * flow this time. */
                ovs_mutex_unlock(&ukey->mutex);
                COVERAGE_INC(upcall_duplicate_flow);
                goto next;
            }

            used = ukey->created;
            ovs_mutex_unlock(&ukey->mutex);
        }

        n_flows = udpif_get_n_flows(udpif);
        max_idle = ofproto_max_idle;
        if (n_flows > flow_limit) {
            max_idle = 100;
        }

        if ((used && used < now - max_idle) || n_flows > flow_limit * 2) {
            mark = false;
        } else {
            if (!ukey) {
                ukey = ukey_create(key, key_len, used);
                if (!udpif_insert_ukey(udpif, ukey, hash)) {
                    /* The same ukey has already been created. This means that
                     * another revalidator is processing this flow
                     * concurrently, so don't bother processing it. */
                    ukey_delete(NULL, ukey);
                    goto next;
                }
            }

            mark = revalidate_ukey(udpif, ukey, mask, mask_len, actions,
                                   actions_len, stats);
        }

        if (ukey) {
            ovs_mutex_lock(&ukey->mutex);
            ukey->mark = ukey->flow_exists = mark;
            ovs_mutex_unlock(&ukey->mutex);
        }

        if (!mark) {
            dump_op_init(&ops[n_ops++], key, key_len, ukey);
        }

    next:
        may_destroy = dpif_flow_dump_next_may_destroy_keys(&udpif->dump,
                                                           state);

        /* Only update 'now' immediately before 'buffer' will be updated.
         * This gives us the current time relative to the time the datapath
         * will write into 'stats'. */
        if (may_destroy) {
            now = time_msec();
        }

        /* Only do a dpif_operate when we've hit our maximum batch, or when our
         * memory is about to be clobbered by the next call to
         * dpif_flow_dump_next(). */
        if (n_ops == REVALIDATE_MAX_BATCH || (n_ops && may_destroy)) {
            push_dump_ops__(udpif, ops, n_ops);
            n_ops = 0;
        }
    }

    if (n_ops) {
        push_dump_ops__(udpif, ops, n_ops);
    }

    dpif_flow_dump_state_uninit(udpif->dpif, state);
}

static void
revalidator_sweep__(struct revalidator *revalidator, bool purge)
    OVS_NO_THREAD_SAFETY_ANALYSIS
{
    struct dump_op ops[REVALIDATE_MAX_BATCH];
    struct udpif_key *ukey, *next;
    size_t n_ops;

    n_ops = 0;

    /* During garbage collection, this revalidator completely owns its ukeys
     * map, and therefore doesn't need to do any locking. */
    HMAP_FOR_EACH_SAFE (ukey, next, hmap_node, revalidator->ukeys) {
        if (!purge && ukey->mark) {
            ukey->mark = false;
        } else if (!ukey->flow_exists) {
            ukey_delete(revalidator, ukey);
        } else {
            struct dump_op *op = &ops[n_ops++];

            /* If we have previously seen a flow in the datapath, but didn't
             * see it during the most recent dump, delete it. This allows us
             * to clean up the ukey and keep the statistics consistent. */
            dump_op_init(op, ukey->key, ukey->key_len, ukey);
            if (n_ops == REVALIDATE_MAX_BATCH) {
                push_dump_ops(revalidator, ops, n_ops);
                n_ops = 0;
            }
        }
    }

    if (n_ops) {
        push_dump_ops(revalidator, ops, n_ops);
    }
}

static void
revalidator_sweep(struct revalidator *revalidator)
{
    revalidator_sweep__(revalidator, false);
}

static void
revalidator_purge(struct revalidator *revalidator)
{
    revalidator_sweep__(revalidator, true);
}

static void
upcall_unixctl_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
                    const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
{
    struct ds ds = DS_EMPTY_INITIALIZER;
    struct udpif *udpif;

    LIST_FOR_EACH (udpif, list_node, &all_udpifs) {
        unsigned int flow_limit;
        size_t i;

        atomic_read(&udpif->flow_limit, &flow_limit);

        ds_put_format(&ds, "%s:\n", dpif_name(udpif->dpif));
        ds_put_format(&ds, "\tflows         : (current %lu)"
            " (avg %u) (max %u) (limit %u)\n", udpif_get_n_flows(udpif),
            udpif->avg_n_flows, udpif->max_n_flows, flow_limit);
        ds_put_format(&ds, "\tdump duration : %lldms\n", udpif->dump_duration);

        ds_put_char(&ds, '\n');
        for (i = 0; i < n_revalidators; i++) {
            struct revalidator *revalidator = &udpif->revalidators[i];

            ovs_mutex_lock(&udpif->ukeys[i].mutex);
            ds_put_format(&ds, "\t%u: (keys %"PRIuSIZE")\n",
                          revalidator->id, hmap_count(&udpif->ukeys[i].hmap));
            ovs_mutex_unlock(&udpif->ukeys[i].mutex);
        }
    }

    unixctl_command_reply(conn, ds_cstr(&ds));
    ds_destroy(&ds);
}

/* Disable using the megaflows.
 *
 * This command is only needed for advanced debugging, so it's not
 * documented in the man page. */
static void
upcall_unixctl_disable_megaflows(struct unixctl_conn *conn,
                                 int argc OVS_UNUSED,
                                 const char *argv[] OVS_UNUSED,
                                 void *aux OVS_UNUSED)
{
    atomic_store(&enable_megaflows, false);
    udpif_flush_all_datapaths();
    unixctl_command_reply(conn, "megaflows disabled");
}

/* Re-enable using megaflows.
 *
 * This command is only needed for advanced debugging, so it's not
 * documented in the man page. */
static void
upcall_unixctl_enable_megaflows(struct unixctl_conn *conn,
                                int argc OVS_UNUSED,
                                const char *argv[] OVS_UNUSED,
                                void *aux OVS_UNUSED)
{
    atomic_store(&enable_megaflows, true);
    udpif_flush_all_datapaths();
    unixctl_command_reply(conn, "megaflows enabled");
}

/* Set the flow limit.
 *
 * This command is only needed for advanced debugging, so it's not
 * documented in the man page. */
static void
upcall_unixctl_set_flow_limit(struct unixctl_conn *conn,
                              int argc OVS_UNUSED,
                              const char *argv[] OVS_UNUSED,
                              void *aux OVS_UNUSED)
{
    struct ds ds = DS_EMPTY_INITIALIZER;
    struct udpif *udpif;
    unsigned int flow_limit = atoi(argv[1]);

    LIST_FOR_EACH (udpif, list_node, &all_udpifs) {
        atomic_store(&udpif->flow_limit, flow_limit);
    }
    ds_put_format(&ds, "set flow_limit to %u\n", flow_limit);
    unixctl_command_reply(conn, ds_cstr(&ds));
    ds_destroy(&ds);
}