[openvswitch] / switch / datapath.c

/* Copyright (c) 2008 The Board of Trustees of The Leland Stanford
 * Junior University
 * 
 * We are making the OpenFlow specification and associated documentation
 * (Software) available for public use and benefit with the expectation
 * that others will use, modify and enhance the Software and contribute
 * those enhancements back to the community. However, since we would
 * like to make the Software available for broadest use, with as few
 * restrictions as possible permission is hereby granted, free of
 * charge, to any person obtaining a copy of this Software to deal in
 * the Software under the copyrights without restriction, including
 * without limitation the rights to use, copy, modify, merge, publish,
 * distribute, sublicense, and/or sell copies of the Software, and to
 * permit persons to whom the Software is furnished to do so, subject to
 * the following conditions:
 * 
 * The above copyright notice and this permission notice shall be
 * included in all copies or substantial portions of the Software.
 * 
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT.  IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 * 
 * The name and trademarks of copyright holder(s) may NOT be used in
 * advertising or publicity pertaining to the Software or any
 * derivatives without specific, written prior permission.
 */

#include "datapath.h"
#include <arpa/inet.h>
#include <assert.h>
#include <errno.h>
#include <inttypes.h>
#include <stdlib.h>
#include <string.h>
#include "buffer.h"
#include "chain.h"
#include "csum.h"
#include "flow.h"
#include "netdev.h"
#include "packets.h"
#include "poll-loop.h"
#include "rconn.h"
#include "vconn.h"
#include "table.h"
#include "xtoxll.h"

#define THIS_MODULE VLM_datapath
#include "vlog.h"

#define BRIDGE_PORT_NO_FLOOD    0x00000001

/* Capabilities supported by this implementation. */
#define OFP_SUPPORTED_CAPABILITIES ( OFPC_FLOW_STATS \
        | OFPC_TABLE_STATS \
        | OFPC_PORT_STATS \
        | OFPC_MULTI_PHY_TX )

/* Actions supported by this implementation. */
#define OFP_SUPPORTED_ACTIONS ( (1 << OFPAT_OUTPUT)         \
                                | (1 << OFPAT_SET_DL_VLAN)  \
                                | (1 << OFPAT_SET_DL_SRC)   \
                                | (1 << OFPAT_SET_DL_DST)   \
                                | (1 << OFPAT_SET_NW_SRC)   \
                                | (1 << OFPAT_SET_NW_DST)   \
                                | (1 << OFPAT_SET_TP_SRC)   \
                                | (1 << OFPAT_SET_TP_DST) )

struct sw_port {
    uint32_t flags;
    struct datapath *dp;
    struct netdev *netdev;
    struct list node; /* Element in datapath.ports. */
    unsigned long long int rx_count, tx_count, drop_count;
};

/* The origin of a received OpenFlow message, to enable sending a reply. */
struct sender {
    struct remote *remote;      /* The device that sent the message. */
    uint32_t xid;               /* The OpenFlow transaction ID. */
};

/* A connection to a controller or a management device. */
struct remote {
    struct list node;
    struct rconn *rconn;

    /* Support for reliable, multi-message replies to requests.
     *
     * If an incoming request needs to have a reliable reply that might
     * require multiple messages, it can use remote_start_dump() to set up
     * a callback that will be called as buffer space for replies. */
    int (*cb_dump)(struct datapath *, void *aux);
    void (*cb_done)(void *aux);
    void *cb_aux;
};

struct datapath {
    /* Remote connections. */
    struct remote *controller;  /* Connection to controller. */
    struct list remotes;        /* All connections (including controller). */
    struct vconn *listen_vconn;

    time_t last_timeout;

    /* Unique identifier for this datapath */
    uint64_t  id;

    struct sw_chain *chain;  /* Forwarding rules. */

    struct ofp_switch_config config;

    /* Switch ports. */
    struct sw_port ports[OFPP_MAX];
    struct list port_list; /* List of ports, for flooding. */
};

static struct remote *remote_create(struct datapath *, struct rconn *);
static void remote_run(struct datapath *, struct remote *);
static void remote_wait(struct remote *);
static void remote_destroy(struct remote *);

void dp_output_port(struct datapath *, struct buffer *,
                    int in_port, int out_port);
void dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp);
void dp_output_control(struct datapath *, struct buffer *, int in_port,
                       size_t max_len, int reason);
static void send_flow_expired(struct datapath *, struct sw_flow *);
static void send_port_status(struct sw_port *p, uint8_t status);
static void del_switch_port(struct sw_port *p);
static void execute_actions(struct datapath *, struct buffer *,
                            int in_port, const struct sw_flow_key *,
                            const struct ofp_action *, int n_actions);
static void modify_vlan(struct buffer *buffer, const struct sw_flow_key *key,
                        const struct ofp_action *a);
static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
                      uint8_t nw_proto, const struct ofp_action *a);
static void modify_th(struct buffer *buffer, uint16_t eth_proto,
                          uint8_t nw_proto, const struct ofp_action *a);

/* Buffers are identified to userspace by a 31-bit opaque ID.  We divide the ID
 * into a buffer number (low bits) and a cookie (high bits).  The buffer number
 * is an index into an array of buffers.  The cookie distinguishes between
 * different packets that have occupied a single buffer.  Thus, the more
 * buffers we have, the lower-quality the cookie... */
#define PKT_BUFFER_BITS 8
#define N_PKT_BUFFERS (1 << PKT_BUFFER_BITS)
#define PKT_BUFFER_MASK (N_PKT_BUFFERS - 1)

#define PKT_COOKIE_BITS (32 - PKT_BUFFER_BITS)

void fwd_port_input(struct datapath *, struct buffer *, int in_port);
int fwd_control_input(struct datapath *, const struct sender *,
                      const void *, size_t);

uint32_t save_buffer(struct buffer *);
static struct buffer *retrieve_buffer(uint32_t id);
static void discard_buffer(uint32_t id);

static int port_no(struct datapath *dp, struct sw_port *p) 
{
    assert(p >= dp->ports && p < &dp->ports[ARRAY_SIZE(dp->ports)]);
    return p - dp->ports;
}

/* Generates and returns a random datapath id. */
static uint64_t
gen_datapath_id(void)
{
    uint8_t ea[ETH_ADDR_LEN];
    eth_addr_random(ea);
    return eth_addr_to_uint64(ea);
}

int
dp_new(struct datapath **dp_, uint64_t dpid, struct rconn *rconn)
{
    struct datapath *dp;

    dp = calloc(1, sizeof *dp);
    if (!dp) {
        return ENOMEM;
    }

    dp->last_timeout = time(0);
    list_init(&dp->remotes);
    dp->controller = remote_create(dp, rconn);
    dp->listen_vconn = NULL;
    dp->id = dpid <= UINT64_C(0xffffffffffff) ? dpid : gen_datapath_id();
    dp->chain = chain_create();
    if (!dp->chain) {
        VLOG_ERR("could not create chain");
        free(dp);
        return ENOMEM;
    }

    list_init(&dp->port_list);
    dp->config.flags = 0;
    dp->config.miss_send_len = htons(OFP_DEFAULT_MISS_SEND_LEN);
    *dp_ = dp;
    return 0;
}

int
dp_add_port(struct datapath *dp, const char *name)
{
    struct netdev *netdev;
    struct in6_addr in6;
    struct in_addr in4;
    struct sw_port *p;
    int error;

    error = netdev_open(name, &netdev);
    if (error) {
        return error;
    }
    error = netdev_set_flags(netdev, NETDEV_UP | NETDEV_PROMISC);
    if (error) {
        VLOG_ERR("Couldn't set promiscuous mode on %s device", name);
        netdev_close(netdev);
        return error;
    }
    if (netdev_get_in4(netdev, &in4)) {
        VLOG_ERR("%s device has assigned IP address %s", name, inet_ntoa(in4));
    }
    if (netdev_get_in6(netdev, &in6)) {
        char in6_name[INET6_ADDRSTRLEN + 1];
        inet_ntop(AF_INET6, &in6, in6_name, sizeof in6_name);
        VLOG_ERR("%s device has assigned IPv6 address %s", name, in6_name);
    }

    for (p = dp->ports; ; p++) {
        if (p >= &dp->ports[ARRAY_SIZE(dp->ports)]) {
            return EXFULL;
        } else if (!p->netdev) {
            break;
        }
    }

    p->dp = dp;
    p->netdev = netdev;
    p->tx_count = 0;
    p->rx_count = 0;
    p->drop_count = 0;
    list_push_back(&dp->port_list, &p->node);

    /* Notify the ctlpath that this port has been added */
    send_port_status(p, OFPPR_ADD);

    return 0;
}

void
dp_add_listen_vconn(struct datapath *dp, struct vconn *listen_vconn)
{
    assert(!dp->listen_vconn);
    dp->listen_vconn = listen_vconn;
}

void
dp_run(struct datapath *dp)
{
    time_t now = time(0);
    struct sw_port *p, *pn;
    struct remote *r, *rn;
    struct buffer *buffer = NULL;

    if (now != dp->last_timeout) {
        struct list deleted = LIST_INITIALIZER(&deleted);
        struct sw_flow *f, *n;

        chain_timeout(dp->chain, &deleted);
        LIST_FOR_EACH_SAFE (f, n, struct sw_flow, node, &deleted) {
            send_flow_expired(dp, f);
            list_remove(&f->node);
            flow_free(f);
        }
        dp->last_timeout = now;
    }
    poll_timer_wait(1000);
    
    LIST_FOR_EACH_SAFE (p, pn, struct sw_port, node, &dp->port_list) {
        int error;

        if (!buffer) {
            /* Allocate buffer with some headroom to add headers in forwarding
             * to the controller or adding a vlan tag, plus an extra 2 bytes to
             * allow IP headers to be aligned on a 4-byte boundary.  */
            const int headroom = 128 + 2;
            const int hard_header = VLAN_ETH_HEADER_LEN;
            const int mtu = netdev_get_mtu(p->netdev);
            buffer = buffer_new(headroom + hard_header + mtu);
            buffer->data += headroom;
        }
        error = netdev_recv(p->netdev, buffer);
        if (!error) {
            p->rx_count++;
            fwd_port_input(dp, buffer, port_no(dp, p));
            buffer = NULL;
        } else if (error != EAGAIN) {
            VLOG_ERR("Error receiving data from %s: %s",
                     netdev_get_name(p->netdev), strerror(error));
            del_switch_port(p);
        }
    }
    buffer_delete(buffer);

    /* Talk to remotes. */
    LIST_FOR_EACH_SAFE (r, rn, struct remote, node, &dp->remotes) {
        remote_run(dp, r);
    }
    if (dp->listen_vconn) {
        for (;;) {
            struct vconn *new_vconn;
            int retval;

            retval = vconn_accept(dp->listen_vconn, &new_vconn);
            if (retval) {
                if (retval != EAGAIN) {
                    VLOG_WARN("accept failed (%s)", strerror(retval));
                }
                break;
            }
            remote_create(dp, rconn_new_from_vconn("passive", 128, new_vconn));
        }
    }
}

static void
remote_run(struct datapath *dp, struct remote *r)
{
    int i;

    rconn_run(r->rconn);

    /* Do some remote processing, but cap it at a reasonable amount so that
     * other processing doesn't starve. */
    for (i = 0; i < 50; i++) {
        if (!r->cb_dump) {
            struct buffer *buffer;
            struct ofp_header *oh;

            buffer = rconn_recv(r->rconn);
            if (!buffer) {
                break;
            }

            if (buffer->size >= sizeof *oh) {
                struct sender sender;

                oh = buffer->data;
                sender.remote = r;
                sender.xid = oh->xid;
                fwd_control_input(dp, &sender, buffer->data, buffer->size);
            } else {
                VLOG_WARN("received too-short OpenFlow message");
            }
            buffer_delete(buffer); 
        } else {
            if (!rconn_is_full(r->rconn)) {
                int error = r->cb_dump(dp, r->cb_aux);
                if (error <= 0) {
                    if (error) {
                        VLOG_WARN("dump callback error: %s", strerror(-error));
                    }
                    r->cb_done(r->cb_aux);
                    r->cb_dump = NULL;
                }
            } else {
                break;
            }
        }
    }

    if (!rconn_is_alive(r->rconn)) {
        remote_destroy(r);
    }
}

static void
remote_wait(struct remote *r) 
{
    rconn_run_wait(r->rconn);
    rconn_recv_wait(r->rconn);
}

static void
remote_destroy(struct remote *r)
{
    if (r) {
        if (r->cb_dump && r->cb_done) {
            r->cb_done(r->cb_aux);
        }
        list_remove(&r->node);
        rconn_destroy(r->rconn);
        free(r);
    }
}

static struct remote *
remote_create(struct datapath *dp, struct rconn *rconn) 
{
    struct remote *remote = xmalloc(sizeof *remote);
    list_push_back(&dp->remotes, &remote->node);
    remote->rconn = rconn;
    remote->cb_dump = NULL;
    return remote;
}

/* Starts a callback-based, reliable, possibly multi-message reply to a
 * request made by 'remote'.
 *
 * 'dump' designates a function that will be called when the 'remote' send
 * queue has an empty slot.  It should compose a message and send it on
 * 'remote'.  On success, it should return 1 if it should be called again when
 * another send queue slot opens up, 0 if its transmissions are complete, or a
 * negative errno value on failure.
 *
 * 'done' designates a function to clean up any resources allocated for the
 * dump.  It must handle being called before the dump is complete (which will
 * happen if 'remote' is closed unexpectedly).
 *
 * 'aux' is passed to 'dump' and 'done'. */
static void
remote_start_dump(struct remote *remote,
                  int (*dump)(struct datapath *, void *),
                  void (*done)(void *),
                  void *aux) 
{
    assert(!remote->cb_dump);
    remote->cb_dump = dump;
    remote->cb_done = done;
    remote->cb_aux = aux;
}

void
dp_wait(struct datapath *dp) 
{
    struct sw_port *p;
    struct remote *r;

    LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
        netdev_recv_wait(p->netdev);
    }
    LIST_FOR_EACH (r, struct remote, node, &dp->remotes) {
        remote_wait(r);
    }
    if (dp->listen_vconn) {
        vconn_accept_wait(dp->listen_vconn);
    }
}

/* Delete 'p' from switch. */
static void
del_switch_port(struct sw_port *p)
{
    send_port_status(p, OFPPR_DELETE);
    netdev_close(p->netdev);
    p->netdev = NULL;
    list_remove(&p->node);
}

void
dp_destroy(struct datapath *dp)
{
    struct sw_port *p, *n;

    if (!dp) {
        return;
    }

    LIST_FOR_EACH_SAFE (p, n, struct sw_port, node, &dp->port_list) {
        del_switch_port(p); 
    }
    chain_destroy(dp->chain);
    free(dp);
}

/* Send packets out all the ports except the originating one.  If the
 * "flood" argument is set, don't send out ports with flooding disabled.
 */
static int
output_all(struct datapath *dp, struct buffer *buffer, int in_port, int flood)
{
    struct sw_port *p;
    int prev_port;

    prev_port = -1;
    LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
        if (port_no(dp, p) == in_port) {
            continue;
        }
        if (flood && p->flags & BRIDGE_PORT_NO_FLOOD) {
            continue;
        }
        if (prev_port != -1) {
            dp_output_port(dp, buffer_clone(buffer), in_port, prev_port);
        }
        prev_port = port_no(dp, p);
    }
    if (prev_port != -1)
        dp_output_port(dp, buffer, in_port, prev_port);
    else
        buffer_delete(buffer);

    return 0;
}

void
output_packet(struct datapath *dp, struct buffer *buffer, int out_port) 
{
    if (out_port >= 0 && out_port < OFPP_MAX) { 
        struct sw_port *p = &dp->ports[out_port];
        if (p->netdev != NULL) {
            if (!netdev_send(p->netdev, buffer)) {
                p->tx_count++;
            } else {
                p->drop_count++;
            }
            return;
        }
    }

    buffer_delete(buffer);
    /* FIXME: ratelimit */
    VLOG_DBG("can't forward to bad port %d\n", out_port);
}

/* Takes ownership of 'buffer' and transmits it to 'out_port' on 'dp'.
 */
void
dp_output_port(struct datapath *dp, struct buffer *buffer,
               int in_port, int out_port)
{

    assert(buffer);
    if (out_port == OFPP_FLOOD) {
        output_all(dp, buffer, in_port, 1); 
    } else if (out_port == OFPP_ALL) {
        output_all(dp, buffer, in_port, 0); 
    } else if (out_port == OFPP_CONTROLLER) {
        dp_output_control(dp, buffer, in_port, 0, OFPR_ACTION); 
    } else if (out_port == OFPP_TABLE) {
        struct sw_flow_key key;
        struct sw_flow *flow;

        key.wildcards = 0;
        flow_extract(buffer, in_port, &key.flow);
        flow = chain_lookup(dp->chain, &key);
        if (flow != NULL) {
            flow_used(flow, buffer);
            execute_actions(dp, buffer, in_port, &key, 
                            flow->actions, flow->n_actions);
        }
    } else {
        output_packet(dp, buffer, out_port);
    }
}

static void *
alloc_openflow_buffer(struct datapath *dp, size_t openflow_len, uint8_t type,
                      const struct sender *sender, struct buffer **bufferp)
{
    struct buffer *buffer;
    struct ofp_header *oh;

    buffer = *bufferp = buffer_new(openflow_len);
    oh = buffer_put_uninit(buffer, openflow_len);
    oh->version = OFP_VERSION;
    oh->type = type;
    oh->length = 0;             /* Filled in by send_openflow_buffer(). */
    oh->xid = sender ? sender->xid : 0;
    return oh;
}

static int
send_openflow_buffer(struct datapath *dp, struct buffer *buffer,
                     const struct sender *sender)
{
    struct remote *remote = sender ? sender->remote : dp->controller;
    struct rconn *rconn = remote->rconn;
    struct ofp_header *oh;
    int retval;

    oh = buffer_at_assert(buffer, 0, sizeof *oh);
    oh->length = htons(buffer->size);

    retval = rconn_send(rconn, buffer);
    if (retval) {
        VLOG_WARN("send to %s failed: %s",
                  rconn_get_name(rconn), strerror(retval));
        buffer_delete(buffer);
    }
    return retval;
}

/* Takes ownership of 'buffer' and transmits it to 'dp''s controller.  If the
 * packet can be saved in a buffer, then only the first max_len bytes of
 * 'buffer' are sent; otherwise, all of 'buffer' is sent.  'reason' indicates
 * why 'buffer' is being sent. 'max_len' sets the maximum number of bytes that
 * the caller wants to be sent; a value of 0 indicates the entire packet should
 * be sent. */
void
dp_output_control(struct datapath *dp, struct buffer *buffer, int in_port,
                  size_t max_len, int reason)
{
    struct ofp_packet_in *opi;
    size_t total_len;
    uint32_t buffer_id;

    buffer_id = save_buffer(buffer);
    total_len = buffer->size;
    if (buffer_id != UINT32_MAX && buffer->size > max_len) {
        buffer->size = max_len;
    }

    opi = buffer_push_uninit(buffer, offsetof(struct ofp_packet_in, data));
    opi->header.version = OFP_VERSION;
    opi->header.type    = OFPT_PACKET_IN;
    opi->header.length  = htons(buffer->size);
    opi->header.xid     = htonl(0);
    opi->buffer_id      = htonl(buffer_id);
    opi->total_len      = htons(total_len);
    opi->in_port        = htons(in_port);
    opi->reason         = reason;
    opi->pad            = 0;
    send_openflow_buffer(dp, buffer, NULL);
}

static void fill_port_desc(struct datapath *dp, struct sw_port *p,
                           struct ofp_phy_port *desc)
{
    desc->port_no = htons(port_no(dp, p));
    strncpy((char *) desc->name, netdev_get_name(p->netdev),
            sizeof desc->name);
    desc->name[sizeof desc->name - 1] = '\0';
    memcpy(desc->hw_addr, netdev_get_etheraddr(p->netdev), ETH_ADDR_LEN);
    desc->flags = htonl(p->flags);
    desc->features = htonl(netdev_get_features(p->netdev));
    desc->speed = htonl(netdev_get_speed(p->netdev));
}

static void
dp_send_features_reply(struct datapath *dp, const struct sender *sender)
{
    struct buffer *buffer;
    struct ofp_switch_features *ofr;
    struct sw_port *p;

    ofr = alloc_openflow_buffer(dp, sizeof *ofr, OFPT_FEATURES_REPLY,
                                sender, &buffer);
    ofr->datapath_id    = htonll(dp->id); 
    ofr->n_exact        = htonl(2 * TABLE_HASH_MAX_FLOWS);
    ofr->n_compression  = 0;         /* Not supported */
    ofr->n_general      = htonl(TABLE_LINEAR_MAX_FLOWS);
    ofr->buffer_mb      = htonl(UINT32_MAX);
    ofr->n_buffers      = htonl(N_PKT_BUFFERS);
    ofr->capabilities   = htonl(OFP_SUPPORTED_CAPABILITIES);
    ofr->actions        = htonl(OFP_SUPPORTED_ACTIONS);
    LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
        struct ofp_phy_port *opp = buffer_put_uninit(buffer, sizeof *opp);
        memset(opp, 0, sizeof *opp);
        fill_port_desc(dp, p, opp);
    }
    send_openflow_buffer(dp, buffer, sender);
}

void
dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp)
{
    int port_no = ntohs(opp->port_no);
    if (port_no < OFPP_MAX) {
        struct sw_port *p = &dp->ports[port_no];

        /* Make sure the port id hasn't changed since this was sent */
        if (!p || memcmp(opp->hw_addr, netdev_get_etheraddr(p->netdev),
                         ETH_ADDR_LEN) != 0) {
            return;
        }
        p->flags = htonl(opp->flags); 
    }
}

static void
send_port_status(struct sw_port *p, uint8_t status) 
{
    struct buffer *buffer;
    struct ofp_port_status *ops;
    ops = alloc_openflow_buffer(p->dp, sizeof *ops, OFPT_PORT_STATUS, NULL,
                                &buffer);
    ops->reason = status;
    memset(ops->pad, 0, sizeof ops->pad);
    fill_port_desc(p->dp, p, &ops->desc);

    send_openflow_buffer(p->dp, buffer, NULL);
}

void
send_flow_expired(struct datapath *dp, struct sw_flow *flow)
{
    struct buffer *buffer;
    struct ofp_flow_expired *ofe;
    ofe = alloc_openflow_buffer(dp, sizeof *ofe, OFPT_FLOW_EXPIRED, NULL,
                                &buffer);
    flow_fill_match(&ofe->match, &flow->key);

    memset(ofe->pad, 0, sizeof ofe->pad);
    ofe->priority = htons(flow->priority);

    ofe->duration     = htonl(flow->timeout - flow->max_idle - flow->created);
    ofe->packet_count = htonll(flow->packet_count);
    ofe->byte_count   = htonll(flow->byte_count);
    send_openflow_buffer(dp, buffer, NULL);
}

void
dp_send_error_msg(struct datapath *dp, const struct sender *sender,
        uint16_t type, uint16_t code, const uint8_t *data, size_t len)
{
    struct buffer *buffer;
    struct ofp_error_msg *oem;
    oem = alloc_openflow_buffer(dp, sizeof(*oem)+len, OFPT_ERROR_MSG, 
                                sender, &buffer);
    oem->type = htons(type);
    oem->code = htons(code);
    memcpy(oem->data, data, len);
    send_openflow_buffer(dp, buffer, sender);
}

static void
fill_flow_stats(struct buffer *buffer, struct sw_flow *flow,
                int table_idx, time_t now)
{
    struct ofp_flow_stats *ofs;
    int length = sizeof *ofs + sizeof *ofs->actions * flow->n_actions;
    ofs = buffer_put_uninit(buffer, length);
    ofs->length          = htons(length);
    ofs->table_id        = table_idx;
    ofs->pad             = 0;
    ofs->match.wildcards = htons(flow->key.wildcards);
    ofs->match.in_port   = flow->key.flow.in_port;
    memcpy(ofs->match.dl_src, flow->key.flow.dl_src, ETH_ADDR_LEN);
    memcpy(ofs->match.dl_dst, flow->key.flow.dl_dst, ETH_ADDR_LEN);
    ofs->match.dl_vlan   = flow->key.flow.dl_vlan;
    ofs->match.dl_type   = flow->key.flow.dl_type;
    ofs->match.nw_src    = flow->key.flow.nw_src;
    ofs->match.nw_dst    = flow->key.flow.nw_dst;
    ofs->match.nw_proto  = flow->key.flow.nw_proto;
    memset(ofs->match.pad, 0, sizeof ofs->match.pad);
    ofs->match.tp_src    = flow->key.flow.tp_src;
    ofs->match.tp_dst    = flow->key.flow.tp_dst;
    ofs->duration        = htonl(now - flow->created);
    ofs->packet_count    = htonll(flow->packet_count);
    ofs->byte_count      = htonll(flow->byte_count);
    ofs->priority        = htons(flow->priority);
    ofs->max_idle        = htons(flow->max_idle);
    memcpy(ofs->actions, flow->actions,
           sizeof *ofs->actions * flow->n_actions);
}

\f
/* 'buffer' was received on 'in_port', a physical switch port between 0 and
 * OFPP_MAX.  Process it according to 'chain'. */
void fwd_port_input(struct datapath *dp, struct buffer *buffer, int in_port)
{
    struct sw_flow_key key;
    struct sw_flow *flow;

    key.wildcards = 0;
    flow_extract(buffer, in_port, &key.flow);
    flow = chain_lookup(dp->chain, &key);
    if (flow != NULL) {
        flow_used(flow, buffer);
        execute_actions(dp, buffer, in_port, &key,
                        flow->actions, flow->n_actions);
    } else {
        dp_output_control(dp, buffer, in_port, ntohs(dp->config.miss_send_len),
                          OFPR_NO_MATCH);
    }
}

static void
do_output(struct datapath *dp, struct buffer *buffer, int in_port,
          size_t max_len, int out_port)
{
    if (out_port != OFPP_CONTROLLER) {
        dp_output_port(dp, buffer, in_port, out_port);
    } else {
        dp_output_control(dp, buffer, in_port, max_len, OFPR_ACTION);
    }
}

static void
execute_actions(struct datapath *dp, struct buffer *buffer,
                int in_port, const struct sw_flow_key *key,
                const struct ofp_action *actions, int n_actions)
{
    /* Every output action needs a separate clone of 'buffer', but the common
     * case is just a single output action, so that doing a clone and then
     * freeing the original buffer is wasteful.  So the following code is
     * slightly obscure just to avoid that. */
    int prev_port;
    size_t max_len=0;        /* Initialze to make compiler happy */
    uint16_t eth_proto;
    int i;

    prev_port = -1;
    eth_proto = ntohs(key->flow.dl_type);

    for (i = 0; i < n_actions; i++) {
        const struct ofp_action *a = &actions[i];
        struct eth_header *eh = buffer->l2;

        if (prev_port != -1) {
            do_output(dp, buffer_clone(buffer), in_port, max_len, prev_port);
            prev_port = -1;
        }

        switch (ntohs(a->type)) {
        case OFPAT_OUTPUT:
            prev_port = ntohs(a->arg.output.port);
            max_len = ntohs(a->arg.output.max_len);
            break;

        case OFPAT_SET_DL_VLAN:
            modify_vlan(buffer, key, a);
            break;

        case OFPAT_SET_DL_SRC:
            memcpy(eh->eth_src, a->arg.dl_addr, sizeof eh->eth_src);
            break;

        case OFPAT_SET_DL_DST:
            memcpy(eh->eth_dst, a->arg.dl_addr, sizeof eh->eth_dst);
            break;

        case OFPAT_SET_NW_SRC:
        case OFPAT_SET_NW_DST:
            modify_nh(buffer, eth_proto, key->flow.nw_proto, a);
            break;

        case OFPAT_SET_TP_SRC:
        case OFPAT_SET_TP_DST:
            modify_th(buffer, eth_proto, key->flow.nw_proto, a);
            break;

        default:
            NOT_REACHED();
        }
    }
    if (prev_port != -1)
        do_output(dp, buffer, in_port, max_len, prev_port);
    else
        buffer_delete(buffer);
}

static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
                      uint8_t nw_proto, const struct ofp_action *a)
{
    if (eth_proto == ETH_TYPE_IP) {
        struct ip_header *nh = buffer->l3;
        uint32_t new, *field;

        new = a->arg.nw_addr;
        field = a->type == OFPAT_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst;
        if (nw_proto == IP_TYPE_TCP) {
            struct tcp_header *th = buffer->l4;
            th->tcp_csum = recalc_csum32(th->tcp_csum, *field, new);
        } else if (nw_proto == IP_TYPE_UDP) {
            struct udp_header *th = buffer->l4;
            if (th->udp_csum) {
                th->udp_csum = recalc_csum32(th->udp_csum, *field, new);
                if (!th->udp_csum) {
                    th->udp_csum = 0xffff;
                }
            }
        }
        nh->ip_csum = recalc_csum32(nh->ip_csum, *field, new);
        *field = new;
    }
}

static void modify_th(struct buffer *buffer, uint16_t eth_proto,
                      uint8_t nw_proto, const struct ofp_action *a)
{
    if (eth_proto == ETH_TYPE_IP) {
        uint16_t new, *field;

        new = a->arg.tp;

        if (nw_proto == IP_TYPE_TCP) {
            struct tcp_header *th = buffer->l4;
            field = a->type == OFPAT_SET_TP_SRC ? &th->tcp_src : &th->tcp_dst;
            th->tcp_csum = recalc_csum16(th->tcp_csum, *field, new);
            *field = new;
        } else if (nw_proto == IP_TYPE_UDP) {
            struct udp_header *th = buffer->l4;
            field = a->type == OFPAT_SET_TP_SRC ? &th->udp_src : &th->udp_dst;
            th->udp_csum = recalc_csum16(th->udp_csum, *field, new);
            *field = new;
        }
    }
}

static void
modify_vlan(struct buffer *buffer,
            const struct sw_flow_key *key, const struct ofp_action *a)
{
    uint16_t new_id = a->arg.vlan_id;
    struct vlan_eth_header *veh;

    if (new_id != htons(OFP_VLAN_NONE)) {
        if (key->flow.dl_vlan != htons(OFP_VLAN_NONE)) {
            /* Modify vlan id, but maintain other TCI values */
            veh = buffer->l2;
            veh->veth_tci &= ~htons(VLAN_VID);
            veh->veth_tci |= new_id;
        } else {
            /* Insert new vlan id. */
            struct eth_header *eh = buffer->l2;
            struct vlan_eth_header tmp;
            memcpy(tmp.veth_dst, eh->eth_dst, ETH_ADDR_LEN);
            memcpy(tmp.veth_src, eh->eth_src, ETH_ADDR_LEN);
            tmp.veth_type = htons(ETH_TYPE_VLAN);
            tmp.veth_tci = new_id;
            tmp.veth_next_type = eh->eth_type;
            
            veh = buffer_push_uninit(buffer, VLAN_HEADER_LEN);
            memcpy(veh, &tmp, sizeof tmp);
            buffer->l2 -= VLAN_HEADER_LEN;
        }
    } else  {
        /* Remove an existing vlan header if it exists */
        veh = buffer->l2;
        if (veh->veth_type == htons(ETH_TYPE_VLAN)) {
            struct eth_header tmp;
            
            memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN);
            memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN);
            tmp.eth_type = veh->veth_next_type;
            
            buffer->size -= VLAN_HEADER_LEN;
            buffer->data += VLAN_HEADER_LEN;
            buffer->l2 += VLAN_HEADER_LEN;
            memcpy(buffer->data, &tmp, sizeof tmp);
        }
    }
}

static int
recv_features_request(struct datapath *dp, const struct sender *sender,
                      const void *msg) 
{
    dp_send_features_reply(dp, sender);
    return 0;
}

static int
recv_get_config_request(struct datapath *dp, const struct sender *sender,
                        const void *msg) 
{
    struct buffer *buffer;
    struct ofp_switch_config *osc;

    osc = alloc_openflow_buffer(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY,
                                sender, &buffer);

    assert(sizeof *osc == sizeof dp->config);
    memcpy(((char *)osc) + sizeof osc->header,
           ((char *)&dp->config) + sizeof dp->config.header,
           sizeof dp->config - sizeof dp->config.header);

    return send_openflow_buffer(dp, buffer, sender);
}

static int
recv_set_config(struct datapath *dp, const struct sender *sender UNUSED,
                const void *msg)
{
    const struct ofp_switch_config *osc = msg;
    dp->config = *osc;
    return 0;
}

static int
recv_packet_out(struct datapath *dp, const struct sender *sender UNUSED,
                const void *msg)
{
    const struct ofp_packet_out *opo = msg;

    if (ntohl(opo->buffer_id) == (uint32_t) -1) {
        /* FIXME: can we avoid copying data here? */
        int data_len = ntohs(opo->header.length) - sizeof *opo;
        struct buffer *buffer = buffer_new(data_len);
        buffer_put(buffer, opo->u.data, data_len);
        dp_output_port(dp, buffer,
                       ntohs(opo->in_port), ntohs(opo->out_port));
    } else {
        struct sw_flow_key key;
        struct buffer *buffer;
        int n_acts;

        buffer = retrieve_buffer(ntohl(opo->buffer_id));
        if (!buffer) {
            return -ESRCH; 
        }

        n_acts = (ntohs(opo->header.length) - sizeof *opo) 
            / sizeof *opo->u.actions;
        flow_extract(buffer, ntohs(opo->in_port), &key.flow);
        execute_actions(dp, buffer, ntohs(opo->in_port),
                        &key, opo->u.actions, n_acts);
    }
    return 0;
}

static int
recv_port_mod(struct datapath *dp, const struct sender *sender UNUSED,
              const void *msg)
{
    const struct ofp_port_mod *opm = msg;

    dp_update_port_flags(dp, &opm->desc);

    return 0;
}

static int
add_flow(struct datapath *dp, const struct ofp_flow_mod *ofm)
{
    int error = -ENOMEM;
    int n_acts;
    int i;
    struct sw_flow *flow;


    /* To prevent loops, make sure there's no action to send to the
     * OFP_TABLE virtual port.
     */
    n_acts = (ntohs(ofm->header.length) - sizeof *ofm) / sizeof *ofm->actions;
    for (i=0; i<n_acts; i++) {
        const struct ofp_action *a = &ofm->actions[i];

        if (a->type == htons(OFPAT_OUTPUT)
                    && (a->arg.output.port == htons(OFPP_TABLE)
                        || a->arg.output.port == htons(OFPP_NONE))) {
            /* xxx Send fancy new error message? */
            goto error;
        }
    }

    /* Allocate memory. */
    flow = flow_alloc(n_acts);
    if (flow == NULL)
        goto error;

    /* Fill out flow. */
    flow_extract_match(&flow->key, &ofm->match);
    flow->max_idle = ntohs(ofm->max_idle);
    flow->priority = flow->key.wildcards ? ntohs(ofm->priority) : -1;
    flow->timeout = time(0) + flow->max_idle; /* FIXME */
    flow->n_actions = n_acts;
    flow->created = time(0);    /* FIXME */
    flow->byte_count = 0;
    flow->packet_count = 0;
    memcpy(flow->actions, ofm->actions, n_acts * sizeof *flow->actions);

    /* Act. */
    error = chain_insert(dp->chain, flow);
    if (error) {
        goto error_free_flow; 
    }
    error = 0;
    if (ntohl(ofm->buffer_id) != UINT32_MAX) {
        struct buffer *buffer = retrieve_buffer(ntohl(ofm->buffer_id));
        if (buffer) {
            struct sw_flow_key key;
            uint16_t in_port = ntohs(ofm->match.in_port);
            flow_used(flow, buffer);
            flow_extract(buffer, in_port, &key.flow);
            execute_actions(dp, buffer, in_port, &key, ofm->actions, n_acts);
        } else {
            error = -ESRCH; 
        }
    }
    return error;

error_free_flow:
    flow_free(flow);
error:
    if (ntohl(ofm->buffer_id) != (uint32_t) -1)
        discard_buffer(ntohl(ofm->buffer_id));
    return error;
}

static int
recv_flow(struct datapath *dp, const struct sender *sender UNUSED,
          const void *msg)
{
    const struct ofp_flow_mod *ofm = msg;
    uint16_t command = ntohs(ofm->command);

    if (command == OFPFC_ADD) {
        return add_flow(dp, ofm);
    }  else if (command == OFPFC_DELETE) {
        struct sw_flow_key key;
        flow_extract_match(&key, &ofm->match);
        return chain_delete(dp->chain, &key, 0, 0) ? 0 : -ESRCH;
    } else if (command == OFPFC_DELETE_STRICT) {
        struct sw_flow_key key;
        uint16_t priority;
        flow_extract_match(&key, &ofm->match);
        priority = key.wildcards ? ntohs(ofm->priority) : -1;
        return chain_delete(dp->chain, &key, priority, 1) ? 0 : -ESRCH;
    } else {
        return -ENODEV;
    }
}

struct flow_stats_state {
    int table_idx;
    struct sw_table_position position;
    struct ofp_flow_stats_request rq;
    time_t now;

    struct buffer *buffer;
};

#define MAX_FLOW_STATS_BYTES 4096

static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
                           void **state)
{
    const struct ofp_flow_stats_request *fsr = body;
    struct flow_stats_state *s = xmalloc(sizeof *s);
    s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
    memset(&s->position, 0, sizeof s->position);
    s->rq = *fsr;
    *state = s;
    return 0;
}

static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
{
    struct flow_stats_state *s = private;
    fill_flow_stats(s->buffer, flow, s->table_idx, s->now);
    return s->buffer->size >= MAX_FLOW_STATS_BYTES;
}

static int flow_stats_dump(struct datapath *dp, void *state,
                           struct buffer *buffer)
{
    struct flow_stats_state *s = state;
    struct sw_flow_key match_key;

    flow_extract_match(&match_key, &s->rq.match);
    s->buffer = buffer;
    s->now = time(0);
    while (s->table_idx < dp->chain->n_tables
           && (s->rq.table_id == 0xff || s->rq.table_id == s->table_idx))
    {
        struct sw_table *table = dp->chain->tables[s->table_idx];

        if (table->iterate(table, &match_key, &s->position,
                           flow_stats_dump_callback, s))
            break;

        s->table_idx++;
        memset(&s->position, 0, sizeof s->position);
    }
    return s->buffer->size >= MAX_FLOW_STATS_BYTES;
}

static void flow_stats_done(void *state)
{
    free(state);
}

struct aggregate_stats_state {
    struct ofp_aggregate_stats_request rq;
};

static int aggregate_stats_init(struct datapath *dp,
                                const void *body, int body_len,
                                void **state)
{
    const struct ofp_aggregate_stats_request *rq = body;
    struct aggregate_stats_state *s = xmalloc(sizeof *s);
    s->rq = *rq;
    *state = s;
    return 0;
}

static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
{
    struct ofp_aggregate_stats_reply *rpy = private;
    rpy->packet_count += flow->packet_count;
    rpy->byte_count += flow->byte_count;
    rpy->flow_count++;
    return 0;
}

static int aggregate_stats_dump(struct datapath *dp, void *state,
                                struct buffer *buffer)
{
    struct aggregate_stats_state *s = state;
    struct ofp_aggregate_stats_request *rq = &s->rq;
    struct ofp_aggregate_stats_reply *rpy;
    struct sw_table_position position;
    struct sw_flow_key match_key;
    int table_idx;

    rpy = buffer_put_uninit(buffer, sizeof *rpy);
    memset(rpy, 0, sizeof *rpy);

    flow_extract_match(&match_key, &rq->match);
    table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
    memset(&position, 0, sizeof position);
    while (table_idx < dp->chain->n_tables
           && (rq->table_id == 0xff || rq->table_id == table_idx))
    {
        struct sw_table *table = dp->chain->tables[table_idx];
        int error;

        error = table->iterate(table, &match_key, &position,
                               aggregate_stats_dump_callback, rpy);
        if (error)
            return error;

        table_idx++;
        memset(&position, 0, sizeof position);
    }

    rpy->packet_count = htonll(rpy->packet_count);
    rpy->byte_count = htonll(rpy->byte_count);
    rpy->flow_count = htonl(rpy->flow_count);
    return 0;
}

static void aggregate_stats_done(void *state) 
{
    free(state);
}

static int table_stats_dump(struct datapath *dp, void *state,
                            struct buffer *buffer)
{
    int i;
    for (i = 0; i < dp->chain->n_tables; i++) {
        struct ofp_table_stats *ots = buffer_put_uninit(buffer, sizeof *ots);
        struct sw_table_stats stats;
        dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
        strncpy(ots->name, stats.name, sizeof ots->name);
        ots->table_id = i;
        memset(ots->pad, 0, sizeof ots->pad);
        ots->max_entries = htonl(stats.max_flows);
        ots->active_count = htonl(stats.n_flows);
        ots->matched_count = htonll(0); /* FIXME */
    }
    return 0;
}

struct port_stats_state {
    int port;
};

static int port_stats_init(struct datapath *dp, const void *body, int body_len,
               void **state)
{
    struct port_stats_state *s = xmalloc(sizeof *s);
    s->port = 0;
    *state = s;
    return 0;
}

static int port_stats_dump(struct datapath *dp, void *state,
                           struct buffer *buffer)
{
    struct port_stats_state *s = state;
    int i;

    for (i = s->port; i < OFPP_MAX; i++) {
        struct sw_port *p = &dp->ports[i];
        struct ofp_port_stats *ops;
        if (!p->netdev) {
            continue;
        }
        ops = buffer_put_uninit(buffer, sizeof *ops);
        ops->port_no = htons(port_no(dp, p));
        memset(ops->pad, 0, sizeof ops->pad);
        ops->rx_count = htonll(p->rx_count);
        ops->tx_count = htonll(p->tx_count);
        ops->drop_count = htonll(p->drop_count);
        ops++;
    }
    s->port = i;
    return 0;
}

static void port_stats_done(void *state)
{
    free(state);
}

struct stats_type {
    /* Minimum and maximum acceptable number of bytes in body member of
     * struct ofp_stats_request. */
    size_t min_body, max_body;

    /* Prepares to dump some kind of statistics on 'dp'.  'body' and
     * 'body_len' are the 'body' member of the struct ofp_stats_request.
     * Returns zero if successful, otherwise a negative error code.
     * May initialize '*state' to state information.  May be null if no
     * initialization is required.*/
    int (*init)(struct datapath *dp, const void *body, int body_len,
            void **state);

    /* Appends statistics for 'dp' to 'buffer', which initially contains a
     * struct ofp_stats_reply.  On success, it should return 1 if it should be
     * called again later with another buffer, 0 if it is done, or a negative
     * errno value on failure. */
    int (*dump)(struct datapath *dp, void *state, struct buffer *buffer);

    /* Cleans any state created by the init or dump functions.  May be null
     * if no cleanup is required. */
    void (*done)(void *state);
};

static const struct stats_type stats[] = {
    [OFPST_FLOW] = {
        sizeof(struct ofp_flow_stats_request),
        sizeof(struct ofp_flow_stats_request),
        flow_stats_init,
        flow_stats_dump,
        flow_stats_done
    },
    [OFPST_AGGREGATE] = {
        sizeof(struct ofp_aggregate_stats_request),
        sizeof(struct ofp_aggregate_stats_request),
        aggregate_stats_init,
        aggregate_stats_dump,
        aggregate_stats_done
    },
    [OFPST_TABLE] = {
        0,
        0,
        NULL,
        table_stats_dump,
        NULL
    },
    [OFPST_PORT] = {
        0,
        0,
        port_stats_init,
        port_stats_dump,
        port_stats_done
    },
};

struct stats_dump_cb {
    bool done;
    struct ofp_stats_request *rq;
    struct sender sender;
    const struct stats_type *s;
    void *state;
};

static int
stats_dump(struct datapath *dp, void *cb_)
{
    struct stats_dump_cb *cb = cb_;
    struct ofp_stats_reply *osr;
    struct buffer *buffer;
    int err;

    if (cb->done) {
        return 0;
    }

    osr = alloc_openflow_buffer(dp, sizeof *osr, OFPT_STATS_REPLY, &cb->sender,
                                &buffer);
    osr->type = htons(cb->s - stats);
    osr->flags = 0;

    err = cb->s->dump(dp, cb->state, buffer);
    if (err >= 0) {
        int err2;
        if (!err) {
            cb->done = true;
        } else {
            /* Buffer might have been reallocated, so find our data again. */
            osr = buffer_at_assert(buffer, 0, sizeof *osr);
            osr->flags = ntohs(OFPSF_REPLY_MORE);
        }
        err2 = send_openflow_buffer(dp, buffer, &cb->sender);
        if (err2) {
            err = err2;
        }
    }

    return err;
}

static void
stats_done(void *cb_)
{
    struct stats_dump_cb *cb = cb_;
    if (cb) {
        if (cb->s->done) {
            cb->s->done(cb->state);
        }
        free(cb);
    }
}

static int
recv_stats_request(struct datapath *dp, const struct sender *sender,
                   const void *oh)
{
    const struct ofp_stats_request *rq = oh;
    size_t rq_len = ntohs(rq->header.length);
    struct stats_dump_cb *cb;
    int type, body_len;
    int err;

    type = ntohs(rq->type);
    if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
        VLOG_WARN("received stats request of unknown type %d", type);
        return -EINVAL;
    }

    cb = xmalloc(sizeof *cb);
    cb->done = false;
    cb->rq = xmemdup(rq, rq_len);
    cb->sender = *sender;
    cb->s = &stats[type];
    cb->state = NULL;
    
    body_len = rq_len - offsetof(struct ofp_stats_request, body);
    if (body_len < cb->s->min_body || body_len > cb->s->max_body) {
        VLOG_WARN("stats request type %d with bad body length %d",
                  type, body_len);
        err = -EINVAL;
        goto error;
    }

    if (cb->s->init) {
        err = cb->s->init(dp, rq->body, body_len, &cb->state);
        if (err) {
            VLOG_WARN("failed initialization of stats request type %d: %s",
                      type, strerror(-err));
            goto error;
        }
    }

    remote_start_dump(sender->remote, stats_dump, stats_done, cb);
    return 0;

error:
    free(cb->rq);
    free(cb);
    return err;
}

static int
recv_echo_request(struct datapath *dp, const struct sender *sender,
                  const void *oh)
{
    return send_openflow_buffer(dp, make_echo_reply(oh), sender);
}

static int
recv_echo_reply(struct datapath *dp UNUSED, const struct sender *sender UNUSED,
                  const void *oh UNUSED)
{
    return 0;
}

/* 'msg', which is 'length' bytes long, was received from the control path.
 * Apply it to 'chain'. */
int
fwd_control_input(struct datapath *dp, const struct sender *sender,
                  const void *msg, size_t length)
{
    struct openflow_packet {
        size_t min_size;
        int (*handler)(struct datapath *, const struct sender *, const void *);
    };

    static const struct openflow_packet packets[] = {
        [OFPT_FEATURES_REQUEST] = {
            sizeof (struct ofp_header),
            recv_features_request,
        },
        [OFPT_GET_CONFIG_REQUEST] = {
            sizeof (struct ofp_header),
            recv_get_config_request,
        },
        [OFPT_SET_CONFIG] = {
            sizeof (struct ofp_switch_config),
            recv_set_config,
        },
        [OFPT_PACKET_OUT] = {
            sizeof (struct ofp_packet_out),
            recv_packet_out,
        },
        [OFPT_FLOW_MOD] = {
            sizeof (struct ofp_flow_mod),
            recv_flow,
        },
        [OFPT_PORT_MOD] = {
            sizeof (struct ofp_port_mod),
            recv_port_mod,
        },
        [OFPT_STATS_REQUEST] = {
            sizeof (struct ofp_stats_request),
            recv_stats_request,
        },
        [OFPT_ECHO_REQUEST] = {
            sizeof (struct ofp_header),
            recv_echo_request,
        },
        [OFPT_ECHO_REPLY] = {
            sizeof (struct ofp_header),
            recv_echo_reply,
        },
    };

    const struct openflow_packet *pkt;
    struct ofp_header *oh;

    oh = (struct ofp_header *) msg;
    if (oh->version != OFP_VERSION || oh->type >= ARRAY_SIZE(packets)
        || ntohs(oh->length) > length)
        return -EINVAL;

    pkt = &packets[oh->type];
    if (!pkt->handler)
        return -ENOSYS;
    if (length < pkt->min_size)
        return -EFAULT;

    return pkt->handler(dp, sender, msg);
}
\f
/* Packet buffering. */

#define OVERWRITE_SECS  1

struct packet_buffer {
    struct buffer *buffer;
    uint32_t cookie;
    time_t timeout;
};

static struct packet_buffer buffers[N_PKT_BUFFERS];
static unsigned int buffer_idx;

uint32_t save_buffer(struct buffer *buffer)
{
    struct packet_buffer *p;
    uint32_t id;

    buffer_idx = (buffer_idx + 1) & PKT_BUFFER_MASK;
    p = &buffers[buffer_idx];
    if (p->buffer) {
        /* Don't buffer packet if existing entry is less than
         * OVERWRITE_SECS old. */
        if (time(0) < p->timeout) { /* FIXME */
            return -1;
        } else {
            buffer_delete(p->buffer); 
        }
    }
    /* Don't use maximum cookie value since the all-bits-1 id is
     * special. */
    if (++p->cookie >= (1u << PKT_COOKIE_BITS) - 1)
        p->cookie = 0;
    p->buffer = buffer_clone(buffer);      /* FIXME */
    p->timeout = time(0) + OVERWRITE_SECS; /* FIXME */
    id = buffer_idx | (p->cookie << PKT_BUFFER_BITS);

    return id;
}

static struct buffer *retrieve_buffer(uint32_t id)
{
    struct buffer *buffer = NULL;
    struct packet_buffer *p;

    p = &buffers[id & PKT_BUFFER_MASK];
    if (p->cookie == id >> PKT_BUFFER_BITS) {
        buffer = p->buffer;
        p->buffer = NULL;
    } else {
        printf("cookie mismatch: %x != %x\n",
               id >> PKT_BUFFER_BITS, p->cookie);
    }

    return buffer;
}

static void discard_buffer(uint32_t id)
{
    struct packet_buffer *p;

    p = &buffers[id & PKT_BUFFER_MASK];
    if (p->cookie == id >> PKT_BUFFER_BITS) {
        buffer_delete(p->buffer);
        p->buffer = NULL;
    }
}