- Add support for OpenFlow error message type.
[sliver-openvswitch.git] / datapath / datapath.c
index ab3f866..cfebab8 100644 (file)
@@ -6,6 +6,7 @@
 
 /* Functions for managing the dp interface/device. */
 
+#include <linux/init.h>
 #include <linux/module.h>
 #include <linux/if_arp.h>
 #include <linux/if_bridge.h>
@@ -45,6 +46,7 @@
 #define BRIDGE_PORT_NO_FLOOD   0x00000001 
 
 #define UINT32_MAX                       4294967295U
+#define UINT16_MAX                       65535
 #define MAX(X, Y) ((X) > (Y) ? (X) : (Y))
 
 struct net_bridge_port {
@@ -77,77 +79,123 @@ static int dp_maint_func(void *data);
 static int send_port_status(struct net_bridge_port *p, uint8_t status);
 
 
-/* nla_unreserve - reduce amount of space reserved by nla_reserve  
+/* nla_shrink - reduce amount of space reserved by nla_reserve
  * @skb: socket buffer from which to recover room
  * @nla: netlink attribute to adjust
- * @len: amount by which to reduce attribute payload
+ * @len: new length of attribute payload
  *
  * Reduces amount of space reserved by a call to nla_reserve.
  *
  * No other attributes may be added between calling nla_reserve and this
  * function, since it will create a hole in the message.
  */
-void nla_unreserve(struct sk_buff *skb, struct nlattr *nla, int len)
+void nla_shrink(struct sk_buff *skb, struct nlattr *nla, int len)
 {
-       skb->tail -= len;
-       skb->len  -= len;
-
-       nla->nla_len -= len;
+       int delta = nla_total_size(len) - nla_total_size(nla_len(nla));
+       BUG_ON(delta > 0);
+       skb->tail += delta;
+       skb->len  += delta;
+       nla->nla_len = nla_attr_size(len);
 }
 
+/* Puts a set of openflow headers for a message of the given 'type' into 'skb'.
+ * If 'sender' is nonnull, then it is used as the message's destination.  'dp'
+ * must specify the datapath to use.
+ *
+ * '*max_openflow_len' receives the maximum number of bytes that are available
+ * for the embedded OpenFlow message.  The caller must call
+ * resize_openflow_skb() to set the actual size of the message to this number
+ * of bytes or less.
+ *
+ * Returns the openflow header if successful, otherwise (if 'skb' is too small)
+ * an error code. */
 static void *
-alloc_openflow_skb(struct datapath *dp, size_t openflow_len, uint8_t type,
-                  const struct sender *sender, struct sk_buff **pskb) 
+put_openflow_headers(struct datapath *dp, struct sk_buff *skb, uint8_t type,
+                    const struct sender *sender, int *max_openflow_len)
 {
-       size_t genl_len;
-       struct sk_buff *skb;
-       struct nlattr *attr;
        struct ofp_header *oh;
-
-       genl_len = nla_total_size(sizeof(uint32_t)); /* DP_GENL_A_DP_IDX */
-       genl_len += nla_total_size(openflow_len);    /* DP_GENL_A_OPENFLOW */
-       skb = *pskb = genlmsg_new(genl_len, GFP_ATOMIC);
-       if (!skb) {
-               if (net_ratelimit())
-                       printk("alloc_openflow_skb: genlmsg_new failed\n");
-               return NULL;
-       }
+       struct nlattr *attr;
+       int openflow_len;
 
        /* Assemble the Generic Netlink wrapper. */
        if (!genlmsg_put(skb,
                         sender ? sender->pid : 0,
                         sender ? sender->seq : 0,
                         &dp_genl_family, 0, DP_GENL_C_OPENFLOW))
-               BUG();
+               return ERR_PTR(-ENOBUFS);
        if (nla_put_u32(skb, DP_GENL_A_DP_IDX, dp->dp_idx) < 0)
-               BUG();
+               return ERR_PTR(-ENOBUFS);
+       openflow_len = (skb_tailroom(skb) - NLA_HDRLEN) & ~(NLA_ALIGNTO - 1);
+       if (openflow_len < sizeof *oh)
+               return ERR_PTR(-ENOBUFS);
+       *max_openflow_len = openflow_len;
        attr = nla_reserve(skb, DP_GENL_A_OPENFLOW, openflow_len);
        BUG_ON(!attr);
-       nlmsg_end(skb, (struct nlmsghdr *) skb->data);
 
-       /* Fill in the header. */
+       /* Fill in the header.  The caller is responsible for the length. */
        oh = nla_data(attr);
        oh->version = OFP_VERSION;
        oh->type = type;
-       oh->length = htons(openflow_len);
        oh->xid = sender ? sender->xid : 0;
 
        return oh;
 }
 
+/* Resizes OpenFlow header 'oh', which must be at the tail end of 'skb', to new
+ * length 'new_length' (in bytes), adjusting pointers and size values as
+ * necessary. */
 static void
 resize_openflow_skb(struct sk_buff *skb,
                    struct ofp_header *oh, size_t new_length)
 {
-       struct nlattr *attr;
-
-       BUG_ON(new_length > ntohs(oh->length));
-       attr = ((void *) oh) - NLA_HDRLEN;
-       nla_unreserve(skb, attr, ntohs(oh->length) - new_length);
+       struct nlattr *attr = ((void *) oh) - NLA_HDRLEN;
+       nla_shrink(skb, attr, new_length);
        oh->length = htons(new_length);
        nlmsg_end(skb, (struct nlmsghdr *) skb->data);
 }
 
+/* Allocates a new skb to contain an OpenFlow message 'openflow_len' bytes in
+ * length.  Returns a null pointer if memory is unavailable, otherwise returns
+ * the OpenFlow header and stores a pointer to the skb in '*pskb'. 
+ *
+ * 'type' is the OpenFlow message type.  If 'sender' is nonnull, then it is
+ * used as the message's destination.  'dp' must specify the datapath to
+ * use.  */
+static void *
+alloc_openflow_skb(struct datapath *dp, size_t openflow_len, uint8_t type,
+                  const struct sender *sender, struct sk_buff **pskb) 
+{
+       struct ofp_header *oh;
+       size_t genl_len;
+       struct sk_buff *skb;
+       int max_openflow_len;
+
+       if ((openflow_len + sizeof(struct ofp_header)) > UINT16_MAX) {
+               if (net_ratelimit())
+                       printk("alloc_openflow_skb: openflow message too large: %d\n", 
+                                       openflow_len);
+               return NULL;
+       }
+
+       genl_len = nlmsg_total_size(GENL_HDRLEN + dp_genl_family.hdrsize);
+       genl_len += nla_total_size(sizeof(uint32_t)); /* DP_GENL_A_DP_IDX */
+       genl_len += nla_total_size(openflow_len);    /* DP_GENL_A_OPENFLOW */
+       skb = *pskb = genlmsg_new(genl_len, GFP_ATOMIC);
+       if (!skb) {
+               if (net_ratelimit())
+                       printk("alloc_openflow_skb: genlmsg_new failed\n");
+               return NULL;
+       }
+
+       oh = put_openflow_headers(dp, skb, type, sender, &max_openflow_len);
+       BUG_ON(!oh || IS_ERR(oh));
+       resize_openflow_skb(skb, oh, openflow_len);
+
+       return oh;
+}
+
+/* Sends 'skb' to 'sender' if it is nonnull, otherwise multicasts 'skb' to all
+ * listeners. */
 static int
 send_openflow_skb(struct sk_buff *skb, const struct sender *sender) 
 {
@@ -239,8 +287,8 @@ static int new_dp(int dp_idx)
                printk("datapath: problem setting up 'of' device\n");
 #endif
 
-       dp->config.flags = 0;
-       dp->config.miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
+       dp->flags = 0;
+       dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
 
        dp->dp_task = kthread_run(dp_maint_func, dp, "dp%d", dp_idx);
        if (IS_ERR(dp->dp_task))
@@ -531,7 +579,19 @@ int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port)
        else if (out_port == OFPP_CONTROLLER)
                return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
                                                  OFPR_ACTION);
-       else if (out_port >= OFPP_MAX)
+       else if (out_port == OFPP_TABLE) {
+               struct sw_flow_key key;
+               struct sw_flow *flow;
+
+               flow_extract(skb, skb->dev->br_port->port_no, &key);
+               flow = chain_lookup(dp->chain, &key);
+               if (likely(flow != NULL)) {
+                       flow_used(flow, skb);
+                       execute_actions(dp, skb, &key, flow->actions, flow->n_actions);
+                       return 0;
+               }
+               return -ESRCH;
+       } else if (out_port >= OFPP_MAX)
                goto bad_port;
 
        p = dp->ports[out_port];
@@ -580,6 +640,10 @@ dp_output_control(struct datapath *dp, struct sk_buff *skb,
 
        opi_len = offsetof(struct ofp_packet_in, data) + fwd_len;
        opi = alloc_openflow_skb(dp, opi_len, OFPT_PACKET_IN, NULL, &f_skb);
+       if (!opi) {
+               err = -ENOMEM;
+               goto out;
+       }
        opi->buffer_id      = htonl(buffer_id);
        opi->total_len      = htons(skb->len);
        opi->in_port        = htons(skb->dev->br_port->port_no);
@@ -588,8 +652,8 @@ dp_output_control(struct datapath *dp, struct sk_buff *skb,
        memcpy(opi->data, skb_mac_header(skb), fwd_len);
        err = send_openflow_skb(f_skb, NULL);
 
+out:
        kfree_skb(skb);
-
        return err;
 }
 
@@ -640,7 +704,6 @@ fill_features_reply(struct datapath *dp, struct ofp_switch_features *ofr)
        ofr->datapath_id    = cpu_to_be64(dp->id); 
 
        ofr->n_exact        = htonl(2 * TABLE_HASH_MAX_FLOWS);
-       ofr->n_mac_only     = htonl(TABLE_MAC_MAX_FLOWS);
        ofr->n_compression  = 0;                                           /* Not supported */
        ofr->n_general      = htonl(TABLE_LINEAR_MAX_FLOWS);
        ofr->buffer_mb      = htonl(UINT32_MAX);
@@ -686,13 +749,14 @@ dp_send_config_reply(struct datapath *dp, const struct sender *sender)
        struct sk_buff *skb;
        struct ofp_switch_config *osc;
 
-       osc = alloc_openflow_skb(dp, sizeof *osc, OFPT_PORT_STATUS, sender,
+       osc = alloc_openflow_skb(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY, sender,
                                 &skb);
        if (!osc)
                return -ENOMEM;
-       memcpy(((char *)osc) + sizeof osc->header,
-              ((char *)&dp->config) + sizeof dp->config.header,
-              sizeof dp->config - sizeof dp->config.header);
+
+       osc->flags = htons(dp->flags);
+       osc->miss_send_len = htons(dp->miss_send_len);
+
        return send_openflow_skb(skb, sender);
 }
 
@@ -748,6 +812,121 @@ dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow)
        return send_openflow_skb(skb, NULL);
 }
 
+int
+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 sk_buff *skb;
+       struct ofp_error_msg *oem;
+
+
+       oem = alloc_openflow_skb(dp, sizeof(*oem)+len, OFPT_ERROR_MSG, 
+                       sender, &skb);
+       if (!oem)
+               return -ENOMEM;
+
+       oem->type = htons(type);
+       oem->code = htons(code);
+       memcpy(oem->data, data, len);
+
+       return send_openflow_skb(skb, sender);
+}
+
+static void
+fill_flow_stats(struct ofp_flow_stats *ofs, struct sw_flow *flow,
+               int table_idx)
+{
+       ofs->match.wildcards = htons(flow->key.wildcards);
+       ofs->match.in_port   = flow->key.in_port;
+       memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
+       memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
+       ofs->match.dl_vlan   = flow->key.dl_vlan;
+       ofs->match.dl_type   = flow->key.dl_type;
+       ofs->match.nw_src    = flow->key.nw_src;
+       ofs->match.nw_dst    = flow->key.nw_dst;
+       ofs->match.nw_proto  = flow->key.nw_proto;
+       memset(ofs->match.pad, 0, sizeof ofs->match.pad);
+       ofs->match.tp_src    = flow->key.tp_src;
+       ofs->match.tp_dst    = flow->key.tp_dst;
+       ofs->duration        = htonl((jiffies - flow->init_time) / HZ);
+       ofs->priority        = htons(flow->priority);
+       ofs->table_id        = table_idx;
+       ofs->packet_count    = cpu_to_be64(flow->packet_count);
+       ofs->byte_count      = cpu_to_be64(flow->byte_count);
+}
+
+static int 
+fill_port_stats_reply(struct datapath *dp, struct ofp_port_stats_reply *psr)
+{
+       struct net_bridge_port *p;
+       int port_count = 0;
+
+       list_for_each_entry_rcu (p, &dp->port_list, node) {
+               struct ofp_port_stats *ps = &psr->ports[port_count++];
+               struct net_device_stats *stats = p->dev->get_stats(p->dev);
+               ps->port_no = htons(p->port_no);
+               memset(ps->pad, 0, sizeof ps->pad);
+               ps->rx_count = cpu_to_be64(stats->rx_packets);
+               ps->tx_count = cpu_to_be64(stats->tx_packets);
+               ps->drop_count = cpu_to_be64(stats->rx_dropped
+                                            + stats->tx_dropped);
+       }
+
+       return port_count;
+}
+
+int
+dp_send_port_stats(struct datapath *dp, const struct sender *sender)
+{
+       struct sk_buff *skb;
+       struct ofp_port_stats_reply *psr;
+       size_t psr_len, port_max_len;
+       int port_count;
+
+       /* Overallocate. */
+       port_max_len = sizeof(struct ofp_port_stats) * OFPP_MAX;
+       psr = alloc_openflow_skb(dp, sizeof *psr + port_max_len,
+                                OFPT_PORT_STATS_REPLY, sender, &skb);
+       if (!psr)
+               return -ENOMEM;
+
+       /* Fill. */
+       port_count = fill_port_stats_reply(dp, psr);
+
+       /* Shrink to fit. */
+       psr_len = sizeof *psr + sizeof(struct ofp_port_stats) * port_count;
+       resize_openflow_skb(skb, &psr->header, psr_len);
+       return send_openflow_skb(skb, sender);
+}
+
+int
+dp_send_table_stats(struct datapath *dp, const struct sender *sender)
+{
+       struct sk_buff *skb;
+       struct ofp_table_stats_reply *tsr;
+       int i, n_tables;
+
+       n_tables = dp->chain->n_tables;
+       tsr = alloc_openflow_skb(dp, (offsetof(struct ofp_table_stats_reply,
+                                              tables)
+                                     + sizeof tsr->tables[0] * n_tables),
+                                OFPT_TABLE_STATS_REPLY, sender, &skb);
+       if (!tsr)
+               return -ENOMEM;
+       for (i = 0; i < n_tables; i++) {
+               struct ofp_table_stats *ots = &tsr->tables[i];
+               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;
+               ots->pad[0] = ots->pad[1] = 0;
+               ots->max_entries = htonl(stats.max_flows);
+               ots->active_count = htonl(stats.n_flows);
+               ots->matched_count = cpu_to_be64(0); /* FIXME */
+       }
+       return send_openflow_skb(skb, sender);
+}
+
 /* Generic Netlink interface.
  *
  * See netlink(7) for an introduction to netlink.  See
@@ -848,7 +1027,7 @@ static int dp_genl_query(struct sk_buff *skb, struct genl_info *info)
                err = -ENOENT;
        else {
                void *data;
-               ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
+               ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
                if (!ans_skb) {
                        err = -ENOMEM;
                        goto err;
@@ -875,360 +1054,6 @@ nla_put_failure:
        return err;
 }
 
-/*
- * Fill flow entry for nl flow query.  Called with rcu_lock  
- *
- */
-static
-int
-dp_fill_flow(struct ofp_flow_mod* ofm, struct swt_iterator* iter)
-{
-       ofm->header.version  = OFP_VERSION;
-       ofm->header.type     = OFPT_FLOW_MOD;
-       ofm->header.length   = htons(sizeof(struct ofp_flow_mod) 
-                               + sizeof(ofm->actions[0]));
-       ofm->header.xid      = htonl(0);
-
-       ofm->match.wildcards = htons(iter->flow->key.wildcards);
-       ofm->match.in_port   = iter->flow->key.in_port;
-       ofm->match.dl_vlan   = iter->flow->key.dl_vlan;
-       memcpy(ofm->match.dl_src, iter->flow->key.dl_src, ETH_ALEN);
-       memcpy(ofm->match.dl_dst, iter->flow->key.dl_dst, ETH_ALEN);
-       ofm->match.dl_type   = iter->flow->key.dl_type;
-       ofm->match.nw_src    = iter->flow->key.nw_src;
-       ofm->match.nw_dst    = iter->flow->key.nw_dst;
-       ofm->match.nw_proto  = iter->flow->key.nw_proto;
-       ofm->match.tp_src    = iter->flow->key.tp_src;
-       ofm->match.tp_dst    = iter->flow->key.tp_dst;
-       ofm->group_id        = iter->flow->group_id;
-       ofm->max_idle        = iter->flow->max_idle;
-       /* TODO support multiple actions  */
-       ofm->actions[0]      = iter->flow->actions[0];
-
-       return 0;
-}
-
-/* Convenience function */
-static
-void* 
-dp_init_nl_flow_msg(uint32_t dp_idx, uint16_t table_idx, 
-               struct genl_info *info, struct sk_buff* skb)
-{
-       void* data;
-
-       data = genlmsg_put_reply(skb, info, &dp_genl_family, 0, 
-                               DP_GENL_C_QUERY_FLOW);
-       if (data == NULL)
-               return NULL;
-       NLA_PUT_U32(skb, DP_GENL_A_DP_IDX,   dp_idx);
-       NLA_PUT_U16(skb, DP_GENL_A_TABLEIDX, table_idx);
-
-       return data;
-
-nla_put_failure:
-       return NULL;
-}
-
-/*  Iterate through the specified table and send all flow entries over
- *  netlink to userspace.  Each flow message has the following format:
- *
- *  32bit dpix
- *  16bit tabletype
- *  32bit number of flows
- *  openflow-flow-entries
- *
- *  The full table may require multiple messages.  A message with 0 flows
- *  signifies end-of message.
- */
-
-static 
-int 
-dp_dump_table(struct datapath *dp, uint16_t table_idx, struct genl_info *info, struct ofp_flow_mod* matchme) 
-{ 
-       struct sk_buff  *skb = 0; 
-       struct sw_table *table = 0;
-       struct swt_iterator iter;
-       struct sw_flow_key in_flow; 
-       struct nlattr   *attr;
-       int count = 0, sum_count = 0;
-       void *data; 
-       uint8_t* ofm_ptr = 0;
-       struct nlattr   *num_attr; 
-       int err = -ENOMEM;
-
-       table = dp->chain->tables[table_idx]; 
-       if ( table == NULL ) {
-               dprintk("dp::dp_dump_table error, non-existant table at position %d\n", table_idx);
-               return -EINVAL;
-       }
-
-       if (!table->iterator(table, &iter)) {
-               dprintk("dp::dp_dump_table couldn't initialize empty table iterator\n");
-               return -ENOMEM;
-       }
-
-       while (iter.flow) {
-
-               /* verify that we can fit all NL_FLOWS_PER_MESSAGE in a single
-                * sk_buf */
-               if( (sizeof(dp_genl_family) + sizeof(uint32_t) + sizeof(uint16_t) + sizeof(uint32_t) + 
-                                       (NL_FLOWS_PER_MESSAGE * sizeof(struct ofp_flow_mod))) > (8192 - 64)){
-                       dprintk("dp::dp_dump_table NL_FLOWS_PER_MESSAGE may cause overrun in skbuf\n");
-                       return -ENOMEM;
-               }
-
-               skb = nlmsg_new(NLMSG_GOODSIZE, GFP_ATOMIC);
-               if (skb == NULL) {
-                       return -ENOMEM;
-               }
-
-               data = dp_init_nl_flow_msg(dp->dp_idx, table_idx, info, skb);
-               if (data == NULL){
-                       err= -ENOMEM;   
-                       goto error_free_skb;
-               } 
-
-               /* reserve space to put the number of flows for this message, to
-                * be filled after the loop*/
-               num_attr = nla_reserve(skb, DP_GENL_A_NUMFLOWS, sizeof(uint32_t));
-               if(!num_attr){
-                       err = -ENOMEM;
-                       goto error_free_skb;
-               }
-
-               /* Only load NL_FLOWS_PER_MESSAGE flows at a time */
-               attr = nla_reserve(skb, DP_GENL_A_FLOW, 
-                               (sizeof(struct ofp_flow_mod) + sizeof(struct ofp_action)) * NL_FLOWS_PER_MESSAGE);
-               if (!attr){
-                       err = -ENOMEM;
-                       goto error_free_skb;
-               }
-
-               /* internal loop to fill NL_FLOWS_PER_MESSAGE flows */
-               ofm_ptr = nla_data(attr);
-               flow_extract_match(&in_flow, &matchme->match);
-               while (iter.flow && count < NL_FLOWS_PER_MESSAGE) {
-                       if(flow_matches(&in_flow, &iter.flow->key)){
-                               if((err = dp_fill_flow((struct ofp_flow_mod*)ofm_ptr, &iter))) 
-                                       goto error_free_skb;
-                               count++; 
-                               /* TODO support multiple actions  */
-                               ofm_ptr += sizeof(struct ofp_flow_mod) + sizeof(struct ofp_action);
-                       }
-                       table->iterator_next(&iter);
-               }
-
-               *((uint32_t*)nla_data(num_attr)) = count;
-               genlmsg_end(skb, data); 
-
-               sum_count += count; 
-               count = 0;
-
-               err = genlmsg_unicast(skb, info->snd_pid); 
-               skb = 0;
-       }
-
-       /* send a sentinal message saying we're done */
-       skb = nlmsg_new(NLMSG_GOODSIZE, GFP_ATOMIC);
-       if (skb == NULL) {
-               return -ENOMEM;
-       }
-       data = dp_init_nl_flow_msg(dp->dp_idx, table_idx, info, skb);
-       if (data == NULL){
-               err= -ENOMEM;   
-               goto error_free_skb;
-       } 
-
-       NLA_PUT_U32(skb, DP_GENL_A_NUMFLOWS,   0);
-       /* dummy flow so nl doesn't complain */
-       attr = nla_reserve(skb, DP_GENL_A_FLOW, sizeof(struct ofp_flow_mod));
-       if (!attr){
-               err = -ENOMEM;
-               goto error_free_skb;
-       }
-       genlmsg_end(skb, data); 
-       err = genlmsg_reply(skb, info); skb = 0;
-
-nla_put_failure:
-error_free_skb:
-       if(skb)
-               kfree_skb(skb);
-       return err;
-}
-
-/* Helper function to query_table which creates and sends a message packed with
- * table stats.  Message form is:
- *
- * u32 DP_IDX
- * u32 NUM_TABLES
- * OFP_TABLE (list of OFP_TABLES)
- *
- */
-
-static 
-int 
-dp_dump_table_stats(struct datapath *dp, int dp_idx, struct genl_info *info) 
-{ 
-       struct sk_buff   *skb = 0; 
-       struct ofp_table *ot = 0;
-       struct nlattr   *attr;
-       struct sw_table_stats stats; 
-       size_t len;
-       void *data; 
-       int err = -ENOMEM;
-       int i = 0;
-       int nt = dp->chain->n_tables;
-
-       len = 4 + 4 + (sizeof(struct ofp_table) * nt);
-
-       /* u32 IDX, u32 NUMTABLES, list-of-tables */
-       skb = nlmsg_new(MAX(len, NLMSG_GOODSIZE), GFP_ATOMIC);
-       if (skb == NULL) {
-               return -ENOMEM;
-       }
-       
-       data = genlmsg_put_reply(skb, info, &dp_genl_family, 0, 
-                               DP_GENL_C_QUERY_TABLE);
-       if (data == NULL){
-               return -ENOMEM;
-       } 
-
-       NLA_PUT_U32(skb, DP_GENL_A_DP_IDX,      dp_idx);
-       NLA_PUT_U32(skb, DP_GENL_A_NUMTABLES, nt);
-
-       /* ... we assume that all tables can fit in a single message.
-        * Probably a reasonable assumption seeing that we only have
-        * 3 atm */
-       attr = nla_reserve(skb, DP_GENL_A_TABLE, (sizeof(struct ofp_table) * nt));
-       if (!attr){
-               err = -ENOMEM;
-               goto error_free_skb;
-       }
-
-       ot = nla_data(attr);
-
-       for (i = 0; i < nt; ++i) {
-               dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
-               ot->header.version = OFP_VERSION;
-               ot->header.type    = OFPT_TABLE;
-               ot->header.length  = htons(sizeof(struct ofp_table));
-               ot->header.xid     = htonl(0);
-
-               strncpy(ot->name, stats.name, OFP_MAX_TABLE_NAME_LEN); 
-               ot->table_id  = htons(i);
-               ot->n_flows   = htonl(stats.n_flows);
-               ot->max_flows = htonl(stats.max_flows);
-               ot++;
-       }
-
-       genlmsg_end(skb, data); 
-       err = genlmsg_reply(skb, info); skb = 0;
-
-nla_put_failure:
-error_free_skb:
-       if(skb)
-               kfree_skb(skb);
-       return err;
-}
-
-/* 
- * Queries a datapath for flow-table statistics 
- */
-
-
-static int dp_genl_table_query(struct sk_buff *skb, struct genl_info *info)
-{
-       struct   datapath* dp;
-       int       err = 0;
-
-       if (!info->attrs[DP_GENL_A_DP_IDX]) {
-               dprintk("dp::dp_genl_table_query received message with missing attributes\n");
-               return -EINVAL;
-       }
-
-       rcu_read_lock();
-       dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
-       if (!dp) {
-               err = -ENOENT;
-               goto err_out;
-       }
-
-       err = dp_dump_table_stats(dp, nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]), info); 
-
-err_out:
-       rcu_read_unlock();
-       return err;
-}
-
-/* 
- * Queries a datapath for flow-table entries.
- */
-
-static int dp_genl_flow_query(struct sk_buff *skb, struct genl_info *info)
-{
-       struct datapath* dp;
-       struct ofp_flow_mod*  ofm;
-       u16     table_idx;
-       int     err = 0;
-
-       if (!info->attrs[DP_GENL_A_DP_IDX]
-                               || !info->attrs[DP_GENL_A_TABLEIDX]
-                               || !info->attrs[DP_GENL_A_FLOW]) {
-               dprintk("dp::dp_genl_flow_query received message with missing attributes\n");
-               return -EINVAL;
-       }
-
-       rcu_read_lock();
-       dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
-       if (!dp) {
-               err = -ENOENT;
-               goto err_out;
-       }
-
-       table_idx = nla_get_u16(info->attrs[DP_GENL_A_TABLEIDX]);
-
-       if (dp->chain->n_tables <= table_idx){
-               printk("table index %d invalid (dp has %d tables)\n",
-                               table_idx, dp->chain->n_tables);
-       err = -EINVAL;
-               goto err_out;
-       }
-
-       ofm = nla_data(info->attrs[DP_GENL_A_FLOW]);
-       err = dp_dump_table(dp, table_idx, info, ofm); 
-
-err_out:
-       rcu_read_unlock();
-       return err;
-}
-
-static struct nla_policy dp_genl_flow_policy[DP_GENL_A_MAX + 1] = {
-       [DP_GENL_A_DP_IDX]      = { .type = NLA_U32 },
-       [DP_GENL_A_TABLEIDX] = { .type = NLA_U16 },
-       [DP_GENL_A_NUMFLOWS]  = { .type = NLA_U32 },
-};
-
-static struct genl_ops dp_genl_ops_query_flow = {
-       .cmd    = DP_GENL_C_QUERY_FLOW,
-       .flags  = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
-       .policy = dp_genl_flow_policy,
-       .doit   = dp_genl_flow_query,
-       .dumpit = NULL,
-};
-
-static struct nla_policy dp_genl_table_policy[DP_GENL_A_MAX + 1] = {
-       [DP_GENL_A_DP_IDX]      = { .type = NLA_U32 },
-};
-
-static struct genl_ops dp_genl_ops_query_table = {
-       .cmd    = DP_GENL_C_QUERY_TABLE,
-       .flags  = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
-       .policy = dp_genl_table_policy,
-       .doit   = dp_genl_table_query,
-       .dumpit = NULL,
-};
-
-
 static struct genl_ops dp_genl_ops_query_dp = {
        .cmd = DP_GENL_C_QUERY_DP,
        .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
@@ -1314,7 +1139,6 @@ static int dp_genl_openflow(struct sk_buff *skb, struct genl_info *info)
                goto out;
        }
 
-       va = info->attrs[DP_GENL_A_OPENFLOW];
        if (nla_len(va) < sizeof(struct ofp_header)) {
                err = -EINVAL;
                goto out;
@@ -1335,12 +1159,156 @@ static struct nla_policy dp_genl_openflow_policy[DP_GENL_A_MAX + 1] = {
        [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
 };
 
+struct flow_stats_cb_state {
+       int dp_idx;
+       int table_idx;
+       struct sw_table_position position;
+       struct ofp_flow_stats_request *rq;
+       int sent_terminator;
+
+       struct ofp_flow_stats *flows;
+       int n_flows, max_flows;
+};
+
+static int muster_callback(struct sw_flow *flow, void *private)
+{
+       struct flow_stats_cb_state *s = private;
+
+       fill_flow_stats(&s->flows[s->n_flows], flow, s->table_idx);
+       return ++s->n_flows >= s->max_flows;
+}
+
+int
+muster_flow_stats(struct datapath *dp, struct flow_stats_cb_state *s,
+                 const struct sender *sender, struct sk_buff *skb)
+{
+       struct ofp_flow_stats_reply *fsr;
+       size_t header_size, flow_size;
+       struct sw_flow_key match_key;
+       int max_openflow_len;
+       size_t size;
+
+       fsr = put_openflow_headers(dp, skb, OFPT_FLOW_STATS_REPLY, sender,
+                                  &max_openflow_len);
+       if (IS_ERR(fsr))
+               return PTR_ERR(fsr);
+       resize_openflow_skb(skb, &fsr->header, max_openflow_len);
+
+       header_size = offsetof(struct ofp_flow_stats_reply, flows);
+       flow_size = sizeof fsr->flows[0];
+       s->max_flows = (max_openflow_len - header_size) / flow_size;
+       if (s->max_flows <= 0)
+               return -ENOMEM;
+       s->flows = fsr->flows;
+
+       flow_extract_match(&match_key, &s->rq->match);
+       s->n_flows = 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,
+                                  muster_callback, s))
+                       break;
+
+               s->table_idx++;
+               memset(&s->position, 0, sizeof s->position);
+       }
+       if (!s->n_flows) {
+               /* Signal dump completion. */
+               if (s->sent_terminator) {
+                       return 0;
+               }
+               s->sent_terminator = 1;
+       }
+       size = header_size + flow_size * s->n_flows;
+       resize_openflow_skb(skb, &fsr->header, size);
+       return skb->len;
+}
+
+static int
+dp_genl_openflow_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
+{
+       struct datapath *dp;
+       struct sender sender;
+       struct flow_stats_cb_state *state;
+       int err;
+
+       if (!cb->args[0]) {
+               struct nlattr *attrs[DP_GENL_A_MAX + 1];
+               struct ofp_flow_stats_request *rq;
+               struct nlattr *va;
+
+               err = nlmsg_parse(cb->nlh, GENL_HDRLEN, attrs, DP_GENL_A_MAX,
+                                 dp_genl_openflow_policy);
+               if (err < 0)
+                       return err;
+
+               if (!attrs[DP_GENL_A_DP_IDX])
+                       return -EINVAL;
+
+               va = attrs[DP_GENL_A_OPENFLOW];
+               if (!va || nla_len(va) != sizeof *state->rq)
+                       return -EINVAL;
+
+               rq = nla_data(va);
+               if (rq->header.version != OFP_VERSION
+                   || rq->header.type != OFPT_FLOW_STATS_REQUEST
+                   || ntohs(rq->header.length) != sizeof *rq)
+                       return -EINVAL;
+
+               state = kmalloc(sizeof *state, GFP_KERNEL);
+               if (!state)
+                       return -ENOMEM;
+               state->dp_idx = nla_get_u32(attrs[DP_GENL_A_DP_IDX]);
+               state->table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
+               memset(&state->position, 0, sizeof state->position);
+               state->rq = rq;
+               state->sent_terminator = 0;
+
+               cb->args[0] = (long) state;
+       } else {
+               state = (struct flow_stats_cb_state *) cb->args[0];
+       }
+
+       if (state->rq->type != OFPFS_INDIV) {
+               return -ENOTSUPP;
+       }
+
+       rcu_read_lock();
+       dp = dp_get(state->dp_idx);
+       if (!dp) {
+               err = -ENOENT;
+               goto out;
+       }
+
+       sender.xid = state->rq->header.xid;
+       sender.pid = NETLINK_CB(cb->skb).pid;
+       sender.seq = cb->nlh->nlmsg_seq;
+       err = muster_flow_stats(dp, state, &sender, skb);
+
+out:
+       rcu_read_unlock();
+       return err;
+}
+
+static int
+dp_genl_openflow_done(struct netlink_callback *cb)
+{
+       struct flow_stats_cb_state *state;
+       state = (struct flow_stats_cb_state *) cb->args[0];
+       kfree(state);
+       return 0;
+}
+
 static struct genl_ops dp_genl_ops_openflow = {
        .cmd = DP_GENL_C_OPENFLOW,
        .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
        .policy = dp_genl_openflow_policy,
        .doit = dp_genl_openflow,
-       .dumpit = NULL,
+       .dumpit = dp_genl_openflow_dumpit,
+       .done = dp_genl_openflow_done,
 };
 
 static struct nla_policy dp_genl_benchmark_policy[DP_GENL_A_MAX + 1] = {
@@ -1363,8 +1331,6 @@ static struct genl_ops *dp_genl_all_ops[] = {
         * front. */
        &dp_genl_ops_openflow,
 
-       &dp_genl_ops_query_flow,
-       &dp_genl_ops_query_table,
        &dp_genl_ops_add_dp,
        &dp_genl_ops_del_dp,
        &dp_genl_ops_query_dp,