net/openvswitch/datapath.c

   1 /*
   2  * Copyright (c) 2007-2013 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/ethtool.h>
  40 #include <linux/wait.h>
  41 #include <asm/div64.h>
  42 #include <linux/highmem.h>
  43 #include <linux/netfilter_bridge.h>
  44 #include <linux/netfilter_ipv4.h>
  45 #include <linux/inetdevice.h>
  46 #include <linux/list.h>
  47 #include <linux/lockdep.h>
  48 #include <linux/openvswitch.h>
  49 #include <linux/rculist.h>
  50 #include <linux/dmi.h>
  51 #include <linux/workqueue.h>
  52 #include <net/genetlink.h>
  53 #include <net/net_namespace.h>
  54 #include <net/netns/generic.h>
  55
  56 #include "datapath.h"
  57 #include "flow.h"
  58 #include "flow_table.h"
  59 #include "flow_netlink.h"
  60 #include "vport-internal_dev.h"
  61 #include "vport-netdev.h"
  62
  63 int ovs_net_id __read_mostly;
  64
  65 static void ovs_notify(struct genl_family *family,
  66                        struct sk_buff *skb, struct genl_info *info)
  67 {
  68         genl_notify(family, skb, genl_info_net(info), info->snd_portid,
  69                     0, info->nlhdr, GFP_KERNEL);
  70 }
  71
  72 /**
  73  * DOC: Locking:
  74  *
  75  * All writes e.g. Writes to device state (add/remove datapath, port, set
  76  * operations on vports, etc.), Writes to other state (flow table
  77  * modifications, set miscellaneous datapath parameters, etc.) are protected
  78  * by ovs_lock.
  79  *
  80  * Reads are protected by RCU.
  81  *
  82  * There are a few special cases (mostly stats) that have their own
  83  * synchronization but they nest under all of above and don't interact with
  84  * each other.
  85  *
  86  * The RTNL lock nests inside ovs_mutex.
  87  */
  88
  89 static DEFINE_MUTEX(ovs_mutex);
  90
  91 void ovs_lock(void)
  92 {
  93         mutex_lock(&ovs_mutex);
  94 }
  95
  96 void ovs_unlock(void)
  97 {
  98         mutex_unlock(&ovs_mutex);
  99 }
 100
 101 #ifdef CONFIG_LOCKDEP
 102 int lockdep_ovsl_is_held(void)
 103 {
 104         if (debug_locks)
 105                 return lockdep_is_held(&ovs_mutex);
 106         else
 107                 return 1;
 108 }
 109 #endif
 110
 111 static struct vport *new_vport(const struct vport_parms *);
 112 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 113                              const struct dp_upcall_info *);
 114 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 115                                   const struct dp_upcall_info *);
 116
 117 /* Must be called with rcu_read_lock or ovs_mutex. */
 118 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 119 {
 120         struct datapath *dp = NULL;
 121         struct net_device *dev;
 122
 123         rcu_read_lock();
 124         dev = dev_get_by_index_rcu(net, dp_ifindex);
 125         if (dev) {
 126                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 127                 if (vport)
 128                         dp = vport->dp;
 129         }
 130         rcu_read_unlock();
 131
 132         return dp;
 133 }
 134
 135 /* Must be called with rcu_read_lock or ovs_mutex. */
 136 static const char *ovs_dp_name(const struct datapath *dp)
 137 {
 138         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 139         return vport->ops->get_name(vport);
 140 }
 141
 142 static int get_dpifindex(struct datapath *dp)
 143 {
 144         struct vport *local;
 145         int ifindex;
 146
 147         rcu_read_lock();
 148
 149         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 150         if (local)
 151                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 152         else
 153                 ifindex = 0;
 154
 155         rcu_read_unlock();
 156
 157         return ifindex;
 158 }
 159
 160 static void destroy_dp_rcu(struct rcu_head *rcu)
 161 {
 162         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 163
 164         free_percpu(dp->stats_percpu);
 165         release_net(ovs_dp_get_net(dp));
 166         kfree(dp->ports);
 167         kfree(dp);
 168 }
 169
 170 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 171                                             u16 port_no)
 172 {
 173         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 174 }
 175
 176 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 177 {
 178         struct vport *vport;
 179         struct hlist_head *head;
 180
 181         head = vport_hash_bucket(dp, port_no);
 182         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 183                 if (vport->port_no == port_no)
 184                         return vport;
 185         }
 186         return NULL;
 187 }
 188
 189 /* Called with ovs_mutex. */
 190 static struct vport *new_vport(const struct vport_parms *parms)
 191 {
 192         struct vport *vport;
 193
 194         vport = ovs_vport_add(parms);
 195         if (!IS_ERR(vport)) {
 196                 struct datapath *dp = parms->dp;
 197                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 198
 199                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 200         }
 201         return vport;
 202 }
 203
 204 void ovs_dp_detach_port(struct vport *p)
 205 {
 206         ASSERT_OVSL();
 207
 208         /* First drop references to device. */
 209         hlist_del_rcu(&p->dp_hash_node);
 210
 211         /* Then destroy it. */
 212         ovs_vport_del(p);
 213 }
 214
 215 /* Must be called with rcu_read_lock. */
 216 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 217 {
 218         struct datapath *dp = p->dp;
 219         struct sw_flow *flow;
 220         struct dp_stats_percpu *stats;
 221         struct sw_flow_key key;
 222         u64 *stats_counter;
 223         u32 n_mask_hit;
 224         int error;
 225
 226         stats = this_cpu_ptr(dp->stats_percpu);
 227
 228         /* Extract flow from 'skb' into 'key'. */
 229         error = ovs_flow_extract(skb, p->port_no, &key);
 230         if (unlikely(error)) {
 231                 kfree_skb(skb);
 232                 return;
 233         }
 234
 235         /* Look up flow. */
 236         flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
 237         if (unlikely(!flow)) {
 238                 struct dp_upcall_info upcall;
 239
 240                 upcall.cmd = OVS_PACKET_CMD_MISS;
 241                 upcall.key = &key;
 242                 upcall.userdata = NULL;
 243                 upcall.portid = p->upcall_portid;
 244                 ovs_dp_upcall(dp, skb, &upcall);
 245                 consume_skb(skb);
 246                 stats_counter = &stats->n_missed;
 247                 goto out;
 248         }
 249
 250         OVS_CB(skb)->flow = flow;
 251         OVS_CB(skb)->pkt_key = &key;
 252
 253         ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
 254         ovs_execute_actions(dp, skb);
 255         stats_counter = &stats->n_hit;
 256
 257 out:
 258         /* Update datapath statistics. */
 259         u64_stats_update_begin(&stats->syncp);
 260         (*stats_counter)++;
 261         stats->n_mask_hit += n_mask_hit;
 262         u64_stats_update_end(&stats->syncp);
 263 }
 264
 265 static struct genl_family dp_packet_genl_family = {
 266         .id = GENL_ID_GENERATE,
 267         .hdrsize = sizeof(struct ovs_header),
 268         .name = OVS_PACKET_FAMILY,
 269         .version = OVS_PACKET_VERSION,
 270         .maxattr = OVS_PACKET_ATTR_MAX,
 271         .netnsok = true,
 272         .parallel_ops = true,
 273 };
 274
 275 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 276                   const struct dp_upcall_info *upcall_info)
 277 {
 278         struct dp_stats_percpu *stats;
 279         int err;
 280
 281         if (upcall_info->portid == 0) {
 282                 err = -ENOTCONN;
 283                 goto err;
 284         }
 285
 286         if (!skb_is_gso(skb))
 287                 err = queue_userspace_packet(dp, skb, upcall_info);
 288         else
 289                 err = queue_gso_packets(dp, skb, upcall_info);
 290         if (err)
 291                 goto err;
 292
 293         return 0;
 294
 295 err:
 296         stats = this_cpu_ptr(dp->stats_percpu);
 297
 298         u64_stats_update_begin(&stats->syncp);
 299         stats->n_lost++;
 300         u64_stats_update_end(&stats->syncp);
 301
 302         return err;
 303 }
 304
 305 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 306                              const struct dp_upcall_info *upcall_info)
 307 {
 308         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 309         struct dp_upcall_info later_info;
 310         struct sw_flow_key later_key;
 311         struct sk_buff *segs, *nskb;
 312         int err;
 313
 314         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 315         if (IS_ERR(segs))
 316                 return PTR_ERR(segs);
 317
 318         /* Queue all of the segments. */
 319         skb = segs;
 320         do {
 321                 err = queue_userspace_packet(dp, skb, upcall_info);
 322                 if (err)
 323                         break;
 324
 325                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 326                         /* The initial flow key extracted by ovs_flow_extract()
 327                          * in this case is for a first fragment, so we need to
 328                          * properly mark later fragments.
 329                          */
 330                         later_key = *upcall_info->key;
 331                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 332
 333                         later_info = *upcall_info;
 334                         later_info.key = &later_key;
 335                         upcall_info = &later_info;
 336                 }
 337         } while ((skb = skb->next));
 338
 339         /* Free all of the segments. */
 340         skb = segs;
 341         do {
 342                 nskb = skb->next;
 343                 if (err)
 344                         kfree_skb(skb);
 345                 else
 346                         consume_skb(skb);
 347         } while ((skb = nskb));
 348         return err;
 349 }
 350
 351 static size_t key_attr_size(void)
 352 {
 353         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 354                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 355                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
 356                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 357                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 358                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
 359                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
 360                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 361                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
 362                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 363                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 364                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 365                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 366                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 367                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 368                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 369                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 370                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 371                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 372 }
 373
 374 static size_t upcall_msg_size(const struct nlattr *userdata,
 375                               unsigned int hdrlen)
 376 {
 377         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 378                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 379                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 380
 381         /* OVS_PACKET_ATTR_USERDATA */
 382         if (userdata)
 383                 size += NLA_ALIGN(userdata->nla_len);
 384
 385         return size;
 386 }
 387
 388 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 389                                   const struct dp_upcall_info *upcall_info)
 390 {
 391         struct ovs_header *upcall;
 392         struct sk_buff *nskb = NULL;
 393         struct sk_buff *user_skb; /* to be queued to userspace */
 394         struct nlattr *nla;
 395         struct genl_info info = {
 396                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 397                 .snd_portid = upcall_info->portid,
 398         };
 399         size_t len;
 400         unsigned int hlen;
 401         int err, dp_ifindex;
 402
 403         dp_ifindex = get_dpifindex(dp);
 404         if (!dp_ifindex)
 405                 return -ENODEV;
 406
 407         if (vlan_tx_tag_present(skb)) {
 408                 nskb = skb_clone(skb, GFP_ATOMIC);
 409                 if (!nskb)
 410                         return -ENOMEM;
 411
 412                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 413                 if (!nskb)
 414                         return -ENOMEM;
 415
 416                 nskb->vlan_tci = 0;
 417                 skb = nskb;
 418         }
 419
 420         if (nla_attr_size(skb->len) > USHRT_MAX) {
 421                 err = -EFBIG;
 422                 goto out;
 423         }
 424
 425         /* Complete checksum if needed */
 426         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 427             (err = skb_checksum_help(skb)))
 428                 goto out;
 429
 430         /* Older versions of OVS user space enforce alignment of the last
 431          * Netlink attribute to NLA_ALIGNTO which would require extensive
 432          * padding logic. Only perform zerocopy if padding is not required.
 433          */
 434         if (dp->user_features & OVS_DP_F_UNALIGNED)
 435                 hlen = skb_zerocopy_headlen(skb);
 436         else
 437                 hlen = skb->len;
 438
 439         len = upcall_msg_size(upcall_info->userdata, hlen);
 440         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 441         if (!user_skb) {
 442                 err = -ENOMEM;
 443                 goto out;
 444         }
 445
 446         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 447                              0, upcall_info->cmd);
 448         upcall->dp_ifindex = dp_ifindex;
 449
 450         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 451         ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
 452         nla_nest_end(user_skb, nla);
 453
 454         if (upcall_info->userdata)
 455                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 456                           nla_len(upcall_info->userdata),
 457                           nla_data(upcall_info->userdata));
 458
 459         /* Only reserve room for attribute header, packet data is added
 460          * in skb_zerocopy() */
 461         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 462                 err = -ENOBUFS;
 463                 goto out;
 464         }
 465         nla->nla_len = nla_attr_size(skb->len);
 466
 467         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 468         if (err)
 469                 goto out;
 470
 471         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 472         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 473                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 474
 475                 if (plen > 0)
 476                         memset(skb_put(user_skb, plen), 0, plen);
 477         }
 478
 479         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 480
 481         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 482 out:
 483         if (err)
 484                 skb_tx_error(skb);
 485         kfree_skb(nskb);
 486         return err;
 487 }
 488
 489 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 490 {
 491         struct ovs_header *ovs_header = info->userhdr;
 492         struct nlattr **a = info->attrs;
 493         struct sw_flow_actions *acts;
 494         struct sk_buff *packet;
 495         struct sw_flow *flow;
 496         struct datapath *dp;
 497         struct ethhdr *eth;
 498         int len;
 499         int err;
 500
 501         err = -EINVAL;
 502         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 503             !a[OVS_PACKET_ATTR_ACTIONS])
 504                 goto err;
 505
 506         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 507         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 508         err = -ENOMEM;
 509         if (!packet)
 510                 goto err;
 511         skb_reserve(packet, NET_IP_ALIGN);
 512
 513         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 514
 515         skb_reset_mac_header(packet);
 516         eth = eth_hdr(packet);
 517
 518         /* Normally, setting the skb 'protocol' field would be handled by a
 519          * call to eth_type_trans(), but it assumes there's a sending
 520          * device, which we may not have. */
 521         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 522                 packet->protocol = eth->h_proto;
 523         else
 524                 packet->protocol = htons(ETH_P_802_2);
 525
 526         /* Build an sw_flow for sending this packet. */
 527         flow = ovs_flow_alloc();
 528         err = PTR_ERR(flow);
 529         if (IS_ERR(flow))
 530                 goto err_kfree_skb;
 531
 532         err = ovs_flow_extract(packet, -1, &flow->key);
 533         if (err)
 534                 goto err_flow_free;
 535
 536         err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
 537         if (err)
 538                 goto err_flow_free;
 539         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 540         err = PTR_ERR(acts);
 541         if (IS_ERR(acts))
 542                 goto err_flow_free;
 543
 544         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 545                                    &flow->key, 0, &acts);
 546         rcu_assign_pointer(flow->sf_acts, acts);
 547         if (err)
 548                 goto err_flow_free;
 549
 550         OVS_CB(packet)->flow = flow;
 551         OVS_CB(packet)->pkt_key = &flow->key;
 552         packet->priority = flow->key.phy.priority;
 553         packet->mark = flow->key.phy.skb_mark;
 554
 555         rcu_read_lock();
 556         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 557         err = -ENODEV;
 558         if (!dp)
 559                 goto err_unlock;
 560
 561         local_bh_disable();
 562         err = ovs_execute_actions(dp, packet);
 563         local_bh_enable();
 564         rcu_read_unlock();
 565
 566         ovs_flow_free(flow, false);
 567         return err;
 568
 569 err_unlock:
 570         rcu_read_unlock();
 571 err_flow_free:
 572         ovs_flow_free(flow, false);
 573 err_kfree_skb:
 574         kfree_skb(packet);
 575 err:
 576         return err;
 577 }
 578
 579 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 580         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 581         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 582         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 583 };
 584
 585 static const struct genl_ops dp_packet_genl_ops[] = {
 586         { .cmd = OVS_PACKET_CMD_EXECUTE,
 587           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 588           .policy = packet_policy,
 589           .doit = ovs_packet_cmd_execute
 590         }
 591 };
 592
 593 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
 594                          struct ovs_dp_megaflow_stats *mega_stats)
 595 {
 596         int i;
 597
 598         memset(mega_stats, 0, sizeof(*mega_stats));
 599
 600         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 601         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 602
 603         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 604
 605         for_each_possible_cpu(i) {
 606                 const struct dp_stats_percpu *percpu_stats;
 607                 struct dp_stats_percpu local_stats;
 608                 unsigned int start;
 609
 610                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 611
 612                 do {
 613                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 614                         local_stats = *percpu_stats;
 615                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 616
 617                 stats->n_hit += local_stats.n_hit;
 618                 stats->n_missed += local_stats.n_missed;
 619                 stats->n_lost += local_stats.n_lost;
 620                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 621         }
 622 }
 623
 624 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 625         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 626         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 627         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 628 };
 629
 630 static struct genl_family dp_flow_genl_family = {
 631         .id = GENL_ID_GENERATE,
 632         .hdrsize = sizeof(struct ovs_header),
 633         .name = OVS_FLOW_FAMILY,
 634         .version = OVS_FLOW_VERSION,
 635         .maxattr = OVS_FLOW_ATTR_MAX,
 636         .netnsok = true,
 637         .parallel_ops = true,
 638 };
 639
 640 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 641         .name = OVS_FLOW_MCGROUP
 642 };
 643
 644 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 645 {
 646         return NLMSG_ALIGN(sizeof(struct ovs_header))
 647                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 648                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 649                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 650                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 651                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 652                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 653 }
 654
 655 /* Called with ovs_mutex. */
 656 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 657                                   struct sk_buff *skb, u32 portid,
 658                                   u32 seq, u32 flags, u8 cmd)
 659 {
 660         const int skb_orig_len = skb->len;
 661         struct nlattr *start;
 662         struct ovs_flow_stats stats;
 663         __be16 tcp_flags;
 664         unsigned long used;
 665         struct ovs_header *ovs_header;
 666         struct nlattr *nla;
 667         int err;
 668
 669         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 670         if (!ovs_header)
 671                 return -EMSGSIZE;
 672
 673         ovs_header->dp_ifindex = get_dpifindex(dp);
 674
 675         /* Fill flow key. */
 676         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 677         if (!nla)
 678                 goto nla_put_failure;
 679
 680         err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
 681         if (err)
 682                 goto error;
 683         nla_nest_end(skb, nla);
 684
 685         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 686         if (!nla)
 687                 goto nla_put_failure;
 688
 689         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 690         if (err)
 691                 goto error;
 692
 693         nla_nest_end(skb, nla);
 694
 695         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 696         if (used &&
 697             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 698                 goto nla_put_failure;
 699
 700         if (stats.n_packets &&
 701             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 702                 goto nla_put_failure;
 703
 704         if ((u8)ntohs(tcp_flags) &&
 705              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 706                 goto nla_put_failure;
 707
 708         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 709          * this is the first flow to be dumped into 'skb'.  This is unusual for
 710          * Netlink but individual action lists can be longer than
 711          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 712          * The userspace caller can always fetch the actions separately if it
 713          * really wants them.  (Most userspace callers in fact don't care.)
 714          *
 715          * This can only fail for dump operations because the skb is always
 716          * properly sized for single flows.
 717          */
 718         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 719         if (start) {
 720                 const struct sw_flow_actions *sf_acts;
 721
 722                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 723
 724                 err = ovs_nla_put_actions(sf_acts->actions,
 725                                           sf_acts->actions_len, skb);
 726                 if (!err)
 727                         nla_nest_end(skb, start);
 728                 else {
 729                         if (skb_orig_len)
 730                                 goto error;
 731
 732                         nla_nest_cancel(skb, start);
 733                 }
 734         } else if (skb_orig_len)
 735                 goto nla_put_failure;
 736
 737         return genlmsg_end(skb, ovs_header);
 738
 739 nla_put_failure:
 740         err = -EMSGSIZE;
 741 error:
 742         genlmsg_cancel(skb, ovs_header);
 743         return err;
 744 }
 745
 746 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow,
 747                                                struct genl_info *info)
 748 {
 749         size_t len;
 750
 751         len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts));
 752
 753         return genlmsg_new_unicast(len, info, GFP_KERNEL);
 754 }
 755
 756 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 757                                                struct datapath *dp,
 758                                                struct genl_info *info,
 759                                                u8 cmd)
 760 {
 761         struct sk_buff *skb;
 762         int retval;
 763
 764         skb = ovs_flow_cmd_alloc_info(flow, info);
 765         if (!skb)
 766                 return ERR_PTR(-ENOMEM);
 767
 768         retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid,
 769                                         info->snd_seq, 0, cmd);
 770         BUG_ON(retval < 0);
 771         return skb;
 772 }
 773
 774 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 775 {
 776         struct nlattr **a = info->attrs;
 777         struct ovs_header *ovs_header = info->userhdr;
 778         struct sw_flow_key key, masked_key;
 779         struct sw_flow *flow = NULL;
 780         struct sw_flow_mask mask;
 781         struct sk_buff *reply;
 782         struct datapath *dp;
 783         struct sw_flow_actions *acts = NULL;
 784         struct sw_flow_match match;
 785         int error;
 786
 787         /* Extract key. */
 788         error = -EINVAL;
 789         if (!a[OVS_FLOW_ATTR_KEY])
 790                 goto error;
 791
 792         ovs_match_init(&match, &key, &mask);
 793         error = ovs_nla_get_match(&match,
 794                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 795         if (error)
 796                 goto error;
 797
 798         /* Validate actions. */
 799         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 800                 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 801                 error = PTR_ERR(acts);
 802                 if (IS_ERR(acts))
 803                         goto error;
 804
 805                 ovs_flow_mask_key(&masked_key, &key, &mask);
 806                 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
 807                                              &masked_key, 0, &acts);
 808                 if (error) {
 809                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 810                         goto err_kfree;
 811                 }
 812         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 813                 error = -EINVAL;
 814                 goto error;
 815         }
 816
 817         ovs_lock();
 818         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 819         error = -ENODEV;
 820         if (!dp)
 821                 goto err_unlock_ovs;
 822
 823         /* Check if this is a duplicate flow */
 824         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 825         if (!flow) {
 826                 /* Bail out if we're not allowed to create a new flow. */
 827                 error = -ENOENT;
 828                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
 829                         goto err_unlock_ovs;
 830
 831                 /* Allocate flow. */
 832                 flow = ovs_flow_alloc();
 833                 if (IS_ERR(flow)) {
 834                         error = PTR_ERR(flow);
 835                         goto err_unlock_ovs;
 836                 }
 837
 838                 flow->key = masked_key;
 839                 flow->unmasked_key = key;
 840                 rcu_assign_pointer(flow->sf_acts, acts);
 841
 842                 /* Put flow in bucket. */
 843                 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
 844                 if (error) {
 845                         acts = NULL;
 846                         goto err_flow_free;
 847                 }
 848
 849                 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
 850         } else {
 851                 /* We found a matching flow. */
 852                 struct sw_flow_actions *old_acts;
 853
 854                 /* Bail out if we're not allowed to modify an existing flow.
 855                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 856                  * because Generic Netlink treats the latter as a dump
 857                  * request.  We also accept NLM_F_EXCL in case that bug ever
 858                  * gets fixed.
 859                  */
 860                 error = -EEXIST;
 861                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
 862                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
 863                         goto err_unlock_ovs;
 864
 865                 /* The unmasked key has to be the same for flow updates. */
 866                 if (!ovs_flow_cmp_unmasked_key(flow, &match))
 867                         goto err_unlock_ovs;
 868
 869                 /* Update actions. */
 870                 old_acts = ovsl_dereference(flow->sf_acts);
 871                 rcu_assign_pointer(flow->sf_acts, acts);
 872                 ovs_nla_free_flow_actions(old_acts);
 873
 874                 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
 875
 876                 /* Clear stats. */
 877                 if (a[OVS_FLOW_ATTR_CLEAR])
 878                         ovs_flow_stats_clear(flow);
 879         }
 880         ovs_unlock();
 881
 882         if (!IS_ERR(reply))
 883                 ovs_notify(&dp_flow_genl_family, reply, info);
 884         else
 885                 genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
 886                              0, PTR_ERR(reply));
 887         return 0;
 888
 889 err_flow_free:
 890         ovs_flow_free(flow, false);
 891 err_unlock_ovs:
 892         ovs_unlock();
 893 err_kfree:
 894         kfree(acts);
 895 error:
 896         return error;
 897 }
 898
 899 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
 900 {
 901         struct nlattr **a = info->attrs;
 902         struct ovs_header *ovs_header = info->userhdr;
 903         struct sw_flow_key key;
 904         struct sk_buff *reply;
 905         struct sw_flow *flow;
 906         struct datapath *dp;
 907         struct sw_flow_match match;
 908         int err;
 909
 910         if (!a[OVS_FLOW_ATTR_KEY]) {
 911                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
 912                 return -EINVAL;
 913         }
 914
 915         ovs_match_init(&match, &key, NULL);
 916         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
 917         if (err)
 918                 return err;
 919
 920         ovs_lock();
 921         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 922         if (!dp) {
 923                 err = -ENODEV;
 924                 goto unlock;
 925         }
 926
 927         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 928         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
 929                 err = -ENOENT;
 930                 goto unlock;
 931         }
 932
 933         reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
 934         if (IS_ERR(reply)) {
 935                 err = PTR_ERR(reply);
 936                 goto unlock;
 937         }
 938
 939         ovs_unlock();
 940         return genlmsg_reply(reply, info);
 941 unlock:
 942         ovs_unlock();
 943         return err;
 944 }
 945
 946 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
 947 {
 948         struct nlattr **a = info->attrs;
 949         struct ovs_header *ovs_header = info->userhdr;
 950         struct sw_flow_key key;
 951         struct sk_buff *reply;
 952         struct sw_flow *flow;
 953         struct datapath *dp;
 954         struct sw_flow_match match;
 955         int err;
 956
 957         ovs_lock();
 958         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 959         if (!dp) {
 960                 err = -ENODEV;
 961                 goto unlock;
 962         }
 963
 964         if (!a[OVS_FLOW_ATTR_KEY]) {
 965                 err = ovs_flow_tbl_flush(&dp->table);
 966                 goto unlock;
 967         }
 968
 969         ovs_match_init(&match, &key, NULL);
 970         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
 971         if (err)
 972                 goto unlock;
 973
 974         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 975         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
 976                 err = -ENOENT;
 977                 goto unlock;
 978         }
 979
 980         reply = ovs_flow_cmd_alloc_info(flow, info);
 981         if (!reply) {
 982                 err = -ENOMEM;
 983                 goto unlock;
 984         }
 985
 986         ovs_flow_tbl_remove(&dp->table, flow);
 987
 988         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
 989                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
 990         BUG_ON(err < 0);
 991
 992         ovs_flow_free(flow, true);
 993         ovs_unlock();
 994
 995         ovs_notify(&dp_flow_genl_family, reply, info);
 996         return 0;
 997 unlock:
 998         ovs_unlock();
 999         return err;
1000 }
1001
1002 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1003 {
1004         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1005         struct table_instance *ti;
1006         struct datapath *dp;
1007
1008         rcu_read_lock();
1009         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1010         if (!dp) {
1011                 rcu_read_unlock();
1012                 return -ENODEV;
1013         }
1014
1015         ti = rcu_dereference(dp->table.ti);
1016         for (;;) {
1017                 struct sw_flow *flow;
1018                 u32 bucket, obj;
1019
1020                 bucket = cb->args[0];
1021                 obj = cb->args[1];
1022                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1023                 if (!flow)
1024                         break;
1025
1026                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1027                                            NETLINK_CB(cb->skb).portid,
1028                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1029                                            OVS_FLOW_CMD_NEW) < 0)
1030                         break;
1031
1032                 cb->args[0] = bucket;
1033                 cb->args[1] = obj;
1034         }
1035         rcu_read_unlock();
1036         return skb->len;
1037 }
1038
1039 static const struct genl_ops dp_flow_genl_ops[] = {
1040         { .cmd = OVS_FLOW_CMD_NEW,
1041           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1042           .policy = flow_policy,
1043           .doit = ovs_flow_cmd_new_or_set
1044         },
1045         { .cmd = OVS_FLOW_CMD_DEL,
1046           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1047           .policy = flow_policy,
1048           .doit = ovs_flow_cmd_del
1049         },
1050         { .cmd = OVS_FLOW_CMD_GET,
1051           .flags = 0,               /* OK for unprivileged users. */
1052           .policy = flow_policy,
1053           .doit = ovs_flow_cmd_get,
1054           .dumpit = ovs_flow_cmd_dump
1055         },
1056         { .cmd = OVS_FLOW_CMD_SET,
1057           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1058           .policy = flow_policy,
1059           .doit = ovs_flow_cmd_new_or_set,
1060         },
1061 };
1062
1063 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1064         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1065         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1066         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1067 };
1068
1069 static struct genl_family dp_datapath_genl_family = {
1070         .id = GENL_ID_GENERATE,
1071         .hdrsize = sizeof(struct ovs_header),
1072         .name = OVS_DATAPATH_FAMILY,
1073         .version = OVS_DATAPATH_VERSION,
1074         .maxattr = OVS_DP_ATTR_MAX,
1075         .netnsok = true,
1076         .parallel_ops = true,
1077 };
1078
1079 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1080         .name = OVS_DATAPATH_MCGROUP
1081 };
1082
1083 static size_t ovs_dp_cmd_msg_size(void)
1084 {
1085         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1086
1087         msgsize += nla_total_size(IFNAMSIZ);
1088         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1089         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1090         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1091
1092         return msgsize;
1093 }
1094
1095 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1096                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1097 {
1098         struct ovs_header *ovs_header;
1099         struct ovs_dp_stats dp_stats;
1100         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1101         int err;
1102
1103         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1104                                    flags, cmd);
1105         if (!ovs_header)
1106                 goto error;
1107
1108         ovs_header->dp_ifindex = get_dpifindex(dp);
1109
1110         rcu_read_lock();
1111         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1112         rcu_read_unlock();
1113         if (err)
1114                 goto nla_put_failure;
1115
1116         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1117         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1118                         &dp_stats))
1119                 goto nla_put_failure;
1120
1121         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1122                         sizeof(struct ovs_dp_megaflow_stats),
1123                         &dp_megaflow_stats))
1124                 goto nla_put_failure;
1125
1126         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1127                 goto nla_put_failure;
1128
1129         return genlmsg_end(skb, ovs_header);
1130
1131 nla_put_failure:
1132         genlmsg_cancel(skb, ovs_header);
1133 error:
1134         return -EMSGSIZE;
1135 }
1136
1137 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp,
1138                                              struct genl_info *info, u8 cmd)
1139 {
1140         struct sk_buff *skb;
1141         int retval;
1142
1143         skb = genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1144         if (!skb)
1145                 return ERR_PTR(-ENOMEM);
1146
1147         retval = ovs_dp_cmd_fill_info(dp, skb, info->snd_portid, info->snd_seq, 0, cmd);
1148         if (retval < 0) {
1149                 kfree_skb(skb);
1150                 return ERR_PTR(retval);
1151         }
1152         return skb;
1153 }
1154
1155 /* Called with ovs_mutex. */
1156 static struct datapath *lookup_datapath(struct net *net,
1157                                         struct ovs_header *ovs_header,
1158                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1159 {
1160         struct datapath *dp;
1161
1162         if (!a[OVS_DP_ATTR_NAME])
1163                 dp = get_dp(net, ovs_header->dp_ifindex);
1164         else {
1165                 struct vport *vport;
1166
1167                 rcu_read_lock();
1168                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1169                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1170                 rcu_read_unlock();
1171         }
1172         return dp ? dp : ERR_PTR(-ENODEV);
1173 }
1174
1175 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1176 {
1177         struct datapath *dp;
1178
1179         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1180         if (IS_ERR(dp))
1181                 return;
1182
1183         WARN(dp->user_features, "Dropping previously announced user features\n");
1184         dp->user_features = 0;
1185 }
1186
1187 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1188 {
1189         if (a[OVS_DP_ATTR_USER_FEATURES])
1190                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1191 }
1192
1193 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1194 {
1195         struct nlattr **a = info->attrs;
1196         struct vport_parms parms;
1197         struct sk_buff *reply;
1198         struct datapath *dp;
1199         struct vport *vport;
1200         struct ovs_net *ovs_net;
1201         int err, i;
1202
1203         err = -EINVAL;
1204         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1205                 goto err;
1206
1207         ovs_lock();
1208
1209         err = -ENOMEM;
1210         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1211         if (dp == NULL)
1212                 goto err_unlock_ovs;
1213
1214         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1215
1216         /* Allocate table. */
1217         err = ovs_flow_tbl_init(&dp->table);
1218         if (err)
1219                 goto err_free_dp;
1220
1221         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1222         if (!dp->stats_percpu) {
1223                 err = -ENOMEM;
1224                 goto err_destroy_table;
1225         }
1226
1227         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1228                             GFP_KERNEL);
1229         if (!dp->ports) {
1230                 err = -ENOMEM;
1231                 goto err_destroy_percpu;
1232         }
1233
1234         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1235                 INIT_HLIST_HEAD(&dp->ports[i]);
1236
1237         /* Set up our datapath device. */
1238         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1239         parms.type = OVS_VPORT_TYPE_INTERNAL;
1240         parms.options = NULL;
1241         parms.dp = dp;
1242         parms.port_no = OVSP_LOCAL;
1243         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1244
1245         ovs_dp_change(dp, a);
1246
1247         vport = new_vport(&parms);
1248         if (IS_ERR(vport)) {
1249                 err = PTR_ERR(vport);
1250                 if (err == -EBUSY)
1251                         err = -EEXIST;
1252
1253                 if (err == -EEXIST) {
1254                         /* An outdated user space instance that does not understand
1255                          * the concept of user_features has attempted to create a new
1256                          * datapath and is likely to reuse it. Drop all user features.
1257                          */
1258                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1259                                 ovs_dp_reset_user_features(skb, info);
1260                 }
1261
1262                 goto err_destroy_ports_array;
1263         }
1264
1265         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1266         err = PTR_ERR(reply);
1267         if (IS_ERR(reply))
1268                 goto err_destroy_local_port;
1269
1270         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1271         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1272
1273         ovs_unlock();
1274
1275         ovs_notify(&dp_datapath_genl_family, reply, info);
1276         return 0;
1277
1278 err_destroy_local_port:
1279         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1280 err_destroy_ports_array:
1281         kfree(dp->ports);
1282 err_destroy_percpu:
1283         free_percpu(dp->stats_percpu);
1284 err_destroy_table:
1285         ovs_flow_tbl_destroy(&dp->table, false);
1286 err_free_dp:
1287         release_net(ovs_dp_get_net(dp));
1288         kfree(dp);
1289 err_unlock_ovs:
1290         ovs_unlock();
1291 err:
1292         return err;
1293 }
1294
1295 /* Called with ovs_mutex. */
1296 static void __dp_destroy(struct datapath *dp)
1297 {
1298         int i;
1299
1300         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1301                 struct vport *vport;
1302                 struct hlist_node *n;
1303
1304                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1305                         if (vport->port_no != OVSP_LOCAL)
1306                                 ovs_dp_detach_port(vport);
1307         }
1308
1309         list_del_rcu(&dp->list_node);
1310
1311         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1312          * all ports in datapath are destroyed first before freeing datapath.
1313          */
1314         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1315
1316         /* RCU destroy the flow table */
1317         ovs_flow_tbl_destroy(&dp->table, true);
1318
1319         call_rcu(&dp->rcu, destroy_dp_rcu);
1320 }
1321
1322 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1323 {
1324         struct sk_buff *reply;
1325         struct datapath *dp;
1326         int err;
1327
1328         ovs_lock();
1329         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1330         err = PTR_ERR(dp);
1331         if (IS_ERR(dp))
1332                 goto unlock;
1333
1334         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_DEL);
1335         err = PTR_ERR(reply);
1336         if (IS_ERR(reply))
1337                 goto unlock;
1338
1339         __dp_destroy(dp);
1340         ovs_unlock();
1341
1342         ovs_notify(&dp_datapath_genl_family, reply, info);
1343
1344         return 0;
1345 unlock:
1346         ovs_unlock();
1347         return err;
1348 }
1349
1350 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1351 {
1352         struct sk_buff *reply;
1353         struct datapath *dp;
1354         int err;
1355
1356         ovs_lock();
1357         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1358         err = PTR_ERR(dp);
1359         if (IS_ERR(dp))
1360                 goto unlock;
1361
1362         ovs_dp_change(dp, info->attrs);
1363
1364         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1365         if (IS_ERR(reply)) {
1366                 err = PTR_ERR(reply);
1367                 genl_set_err(&dp_datapath_genl_family, sock_net(skb->sk), 0,
1368                              0, err);
1369                 err = 0;
1370                 goto unlock;
1371         }
1372
1373         ovs_unlock();
1374         ovs_notify(&dp_datapath_genl_family, reply, info);
1375
1376         return 0;
1377 unlock:
1378         ovs_unlock();
1379         return err;
1380 }
1381
1382 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1383 {
1384         struct sk_buff *reply;
1385         struct datapath *dp;
1386         int err;
1387
1388         ovs_lock();
1389         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1390         if (IS_ERR(dp)) {
1391                 err = PTR_ERR(dp);
1392                 goto unlock;
1393         }
1394
1395         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1396         if (IS_ERR(reply)) {
1397                 err = PTR_ERR(reply);
1398                 goto unlock;
1399         }
1400
1401         ovs_unlock();
1402         return genlmsg_reply(reply, info);
1403
1404 unlock:
1405         ovs_unlock();
1406         return err;
1407 }
1408
1409 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1410 {
1411         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1412         struct datapath *dp;
1413         int skip = cb->args[0];
1414         int i = 0;
1415
1416         rcu_read_lock();
1417         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1418                 if (i >= skip &&
1419                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1420                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1421                                          OVS_DP_CMD_NEW) < 0)
1422                         break;
1423                 i++;
1424         }
1425         rcu_read_unlock();
1426
1427         cb->args[0] = i;
1428
1429         return skb->len;
1430 }
1431
1432 static const struct genl_ops dp_datapath_genl_ops[] = {
1433         { .cmd = OVS_DP_CMD_NEW,
1434           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1435           .policy = datapath_policy,
1436           .doit = ovs_dp_cmd_new
1437         },
1438         { .cmd = OVS_DP_CMD_DEL,
1439           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1440           .policy = datapath_policy,
1441           .doit = ovs_dp_cmd_del
1442         },
1443         { .cmd = OVS_DP_CMD_GET,
1444           .flags = 0,               /* OK for unprivileged users. */
1445           .policy = datapath_policy,
1446           .doit = ovs_dp_cmd_get,
1447           .dumpit = ovs_dp_cmd_dump
1448         },
1449         { .cmd = OVS_DP_CMD_SET,
1450           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1451           .policy = datapath_policy,
1452           .doit = ovs_dp_cmd_set,
1453         },
1454 };
1455
1456 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1457         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1458         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1459         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1460         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1461         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1462         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1463 };
1464
1465 struct genl_family dp_vport_genl_family = {
1466         .id = GENL_ID_GENERATE,
1467         .hdrsize = sizeof(struct ovs_header),
1468         .name = OVS_VPORT_FAMILY,
1469         .version = OVS_VPORT_VERSION,
1470         .maxattr = OVS_VPORT_ATTR_MAX,
1471         .netnsok = true,
1472         .parallel_ops = true,
1473 };
1474
1475 static struct genl_multicast_group ovs_dp_vport_multicast_group = {
1476         .name = OVS_VPORT_MCGROUP
1477 };
1478
1479 /* Called with ovs_mutex or RCU read lock. */
1480 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1481                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1482 {
1483         struct ovs_header *ovs_header;
1484         struct ovs_vport_stats vport_stats;
1485         int err;
1486
1487         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1488                                  flags, cmd);
1489         if (!ovs_header)
1490                 return -EMSGSIZE;
1491
1492         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1493
1494         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1495             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1496             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1497             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1498                 goto nla_put_failure;
1499
1500         ovs_vport_get_stats(vport, &vport_stats);
1501         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1502                     &vport_stats))
1503                 goto nla_put_failure;
1504
1505         err = ovs_vport_get_options(vport, skb);
1506         if (err == -EMSGSIZE)
1507                 goto error;
1508
1509         return genlmsg_end(skb, ovs_header);
1510
1511 nla_put_failure:
1512         err = -EMSGSIZE;
1513 error:
1514         genlmsg_cancel(skb, ovs_header);
1515         return err;
1516 }
1517
1518 /* Called with ovs_mutex or RCU read lock. */
1519 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1520                                          u32 seq, u8 cmd)
1521 {
1522         struct sk_buff *skb;
1523         int retval;
1524
1525         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1526         if (!skb)
1527                 return ERR_PTR(-ENOMEM);
1528
1529         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1530         BUG_ON(retval < 0);
1531
1532         return skb;
1533 }
1534
1535 /* Called with ovs_mutex or RCU read lock. */
1536 static struct vport *lookup_vport(struct net *net,
1537                                   struct ovs_header *ovs_header,
1538                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1539 {
1540         struct datapath *dp;
1541         struct vport *vport;
1542
1543         if (a[OVS_VPORT_ATTR_NAME]) {
1544                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1545                 if (!vport)
1546                         return ERR_PTR(-ENODEV);
1547                 if (ovs_header->dp_ifindex &&
1548                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1549                         return ERR_PTR(-ENODEV);
1550                 return vport;
1551         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1552                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1553
1554                 if (port_no >= DP_MAX_PORTS)
1555                         return ERR_PTR(-EFBIG);
1556
1557                 dp = get_dp(net, ovs_header->dp_ifindex);
1558                 if (!dp)
1559                         return ERR_PTR(-ENODEV);
1560
1561                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1562                 if (!vport)
1563                         return ERR_PTR(-ENODEV);
1564                 return vport;
1565         } else
1566                 return ERR_PTR(-EINVAL);
1567 }
1568
1569 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1570 {
1571         struct nlattr **a = info->attrs;
1572         struct ovs_header *ovs_header = info->userhdr;
1573         struct vport_parms parms;
1574         struct sk_buff *reply;
1575         struct vport *vport;
1576         struct datapath *dp;
1577         u32 port_no;
1578         int err;
1579
1580         err = -EINVAL;
1581         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1582             !a[OVS_VPORT_ATTR_UPCALL_PID])
1583                 goto exit;
1584
1585         ovs_lock();
1586         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1587         err = -ENODEV;
1588         if (!dp)
1589                 goto exit_unlock;
1590
1591         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1592                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1593
1594                 err = -EFBIG;
1595                 if (port_no >= DP_MAX_PORTS)
1596                         goto exit_unlock;
1597
1598                 vport = ovs_vport_ovsl(dp, port_no);
1599                 err = -EBUSY;
1600                 if (vport)
1601                         goto exit_unlock;
1602         } else {
1603                 for (port_no = 1; ; port_no++) {
1604                         if (port_no >= DP_MAX_PORTS) {
1605                                 err = -EFBIG;
1606                                 goto exit_unlock;
1607                         }
1608                         vport = ovs_vport_ovsl(dp, port_no);
1609                         if (!vport)
1610                                 break;
1611                 }
1612         }
1613
1614         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1615         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1616         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1617         parms.dp = dp;
1618         parms.port_no = port_no;
1619         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1620
1621         vport = new_vport(&parms);
1622         err = PTR_ERR(vport);
1623         if (IS_ERR(vport))
1624                 goto exit_unlock;
1625
1626         err = 0;
1627         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1628                                          OVS_VPORT_CMD_NEW);
1629         if (IS_ERR(reply)) {
1630                 err = PTR_ERR(reply);
1631                 ovs_dp_detach_port(vport);
1632                 goto exit_unlock;
1633         }
1634
1635         ovs_notify(&dp_vport_genl_family, reply, info);
1636
1637 exit_unlock:
1638         ovs_unlock();
1639 exit:
1640         return err;
1641 }
1642
1643 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1644 {
1645         struct nlattr **a = info->attrs;
1646         struct sk_buff *reply;
1647         struct vport *vport;
1648         int err;
1649
1650         ovs_lock();
1651         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1652         err = PTR_ERR(vport);
1653         if (IS_ERR(vport))
1654                 goto exit_unlock;
1655
1656         if (a[OVS_VPORT_ATTR_TYPE] &&
1657             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1658                 err = -EINVAL;
1659                 goto exit_unlock;
1660         }
1661
1662         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1663         if (!reply) {
1664                 err = -ENOMEM;
1665                 goto exit_unlock;
1666         }
1667
1668         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1669                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1670                 if (err)
1671                         goto exit_free;
1672         }
1673
1674         if (a[OVS_VPORT_ATTR_UPCALL_PID])
1675                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1676
1677         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1678                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1679         BUG_ON(err < 0);
1680
1681         ovs_unlock();
1682         ovs_notify(&dp_vport_genl_family, reply, info);
1683         return 0;
1684
1685 exit_free:
1686         kfree_skb(reply);
1687 exit_unlock:
1688         ovs_unlock();
1689         return err;
1690 }
1691
1692 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1693 {
1694         struct nlattr **a = info->attrs;
1695         struct sk_buff *reply;
1696         struct vport *vport;
1697         int err;
1698
1699         ovs_lock();
1700         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1701         err = PTR_ERR(vport);
1702         if (IS_ERR(vport))
1703                 goto exit_unlock;
1704
1705         if (vport->port_no == OVSP_LOCAL) {
1706                 err = -EINVAL;
1707                 goto exit_unlock;
1708         }
1709
1710         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1711                                          info->snd_seq, OVS_VPORT_CMD_DEL);
1712         err = PTR_ERR(reply);
1713         if (IS_ERR(reply))
1714                 goto exit_unlock;
1715
1716         err = 0;
1717         ovs_dp_detach_port(vport);
1718
1719         ovs_notify(&dp_vport_genl_family, reply, info);
1720
1721 exit_unlock:
1722         ovs_unlock();
1723         return err;
1724 }
1725
1726 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1727 {
1728         struct nlattr **a = info->attrs;
1729         struct ovs_header *ovs_header = info->userhdr;
1730         struct sk_buff *reply;
1731         struct vport *vport;
1732         int err;
1733
1734         rcu_read_lock();
1735         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1736         err = PTR_ERR(vport);
1737         if (IS_ERR(vport))
1738                 goto exit_unlock;
1739
1740         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1741                                          info->snd_seq, OVS_VPORT_CMD_NEW);
1742         err = PTR_ERR(reply);
1743         if (IS_ERR(reply))
1744                 goto exit_unlock;
1745
1746         rcu_read_unlock();
1747
1748         return genlmsg_reply(reply, info);
1749
1750 exit_unlock:
1751         rcu_read_unlock();
1752         return err;
1753 }
1754
1755 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1756 {
1757         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1758         struct datapath *dp;
1759         int bucket = cb->args[0], skip = cb->args[1];
1760         int i, j = 0;
1761
1762         rcu_read_lock();
1763         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1764         if (!dp) {
1765                 rcu_read_unlock();
1766                 return -ENODEV;
1767         }
1768         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1769                 struct vport *vport;
1770
1771                 j = 0;
1772                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1773                         if (j >= skip &&
1774                             ovs_vport_cmd_fill_info(vport, skb,
1775                                                     NETLINK_CB(cb->skb).portid,
1776                                                     cb->nlh->nlmsg_seq,
1777                                                     NLM_F_MULTI,
1778                                                     OVS_VPORT_CMD_NEW) < 0)
1779                                 goto out;
1780
1781                         j++;
1782                 }
1783                 skip = 0;
1784         }
1785 out:
1786         rcu_read_unlock();
1787
1788         cb->args[0] = i;
1789         cb->args[1] = j;
1790
1791         return skb->len;
1792 }
1793
1794 static const struct genl_ops dp_vport_genl_ops[] = {
1795         { .cmd = OVS_VPORT_CMD_NEW,
1796           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1797           .policy = vport_policy,
1798           .doit = ovs_vport_cmd_new
1799         },
1800         { .cmd = OVS_VPORT_CMD_DEL,
1801           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1802           .policy = vport_policy,
1803           .doit = ovs_vport_cmd_del
1804         },
1805         { .cmd = OVS_VPORT_CMD_GET,
1806           .flags = 0,               /* OK for unprivileged users. */
1807           .policy = vport_policy,
1808           .doit = ovs_vport_cmd_get,
1809           .dumpit = ovs_vport_cmd_dump
1810         },
1811         { .cmd = OVS_VPORT_CMD_SET,
1812           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1813           .policy = vport_policy,
1814           .doit = ovs_vport_cmd_set,
1815         },
1816 };
1817
1818 struct genl_family_and_ops {
1819         struct genl_family *family;
1820         const struct genl_ops *ops;
1821         int n_ops;
1822         const struct genl_multicast_group *group;
1823 };
1824
1825 static const struct genl_family_and_ops dp_genl_families[] = {
1826         { &dp_datapath_genl_family,
1827           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1828           &ovs_dp_datapath_multicast_group },
1829         { &dp_vport_genl_family,
1830           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1831           &ovs_dp_vport_multicast_group },
1832         { &dp_flow_genl_family,
1833           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1834           &ovs_dp_flow_multicast_group },
1835         { &dp_packet_genl_family,
1836           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1837           NULL },
1838 };
1839
1840 static void dp_unregister_genl(int n_families)
1841 {
1842         int i;
1843
1844         for (i = 0; i < n_families; i++)
1845                 genl_unregister_family(dp_genl_families[i].family);
1846 }
1847
1848 static int dp_register_genl(void)
1849 {
1850         int n_registered;
1851         int err;
1852         int i;
1853
1854         n_registered = 0;
1855         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1856                 const struct genl_family_and_ops *f = &dp_genl_families[i];
1857
1858                 f->family->ops = f->ops;
1859                 f->family->n_ops = f->n_ops;
1860                 f->family->mcgrps = f->group;
1861                 f->family->n_mcgrps = f->group ? 1 : 0;
1862                 err = genl_register_family(f->family);
1863                 if (err)
1864                         goto error;
1865                 n_registered++;
1866         }
1867
1868         return 0;
1869
1870 error:
1871         dp_unregister_genl(n_registered);
1872         return err;
1873 }
1874
1875 static int __net_init ovs_init_net(struct net *net)
1876 {
1877         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1878
1879         INIT_LIST_HEAD(&ovs_net->dps);
1880         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1881         return 0;
1882 }
1883
1884 static void __net_exit ovs_exit_net(struct net *net)
1885 {
1886         struct datapath *dp, *dp_next;
1887         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1888
1889         ovs_lock();
1890         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1891                 __dp_destroy(dp);
1892         ovs_unlock();
1893
1894         cancel_work_sync(&ovs_net->dp_notify_work);
1895 }
1896
1897 static struct pernet_operations ovs_net_ops = {
1898         .init = ovs_init_net,
1899         .exit = ovs_exit_net,
1900         .id   = &ovs_net_id,
1901         .size = sizeof(struct ovs_net),
1902 };
1903
1904 static int __init dp_init(void)
1905 {
1906         int err;
1907
1908         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1909
1910         pr_info("Open vSwitch switching datapath\n");
1911
1912         err = ovs_flow_init();
1913         if (err)
1914                 goto error;
1915
1916         err = ovs_vport_init();
1917         if (err)
1918                 goto error_flow_exit;
1919
1920         err = register_pernet_device(&ovs_net_ops);
1921         if (err)
1922                 goto error_vport_exit;
1923
1924         err = register_netdevice_notifier(&ovs_dp_device_notifier);
1925         if (err)
1926                 goto error_netns_exit;
1927
1928         err = dp_register_genl();
1929         if (err < 0)
1930                 goto error_unreg_notifier;
1931
1932         return 0;
1933
1934 error_unreg_notifier:
1935         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1936 error_netns_exit:
1937         unregister_pernet_device(&ovs_net_ops);
1938 error_vport_exit:
1939         ovs_vport_exit();
1940 error_flow_exit:
1941         ovs_flow_exit();
1942 error:
1943         return err;
1944 }
1945
1946 static void dp_cleanup(void)
1947 {
1948         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1949         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1950         unregister_pernet_device(&ovs_net_ops);
1951         rcu_barrier();
1952         ovs_vport_exit();
1953         ovs_flow_exit();
1954 }
1955
1956 module_init(dp_init);
1957 module_exit(dp_cleanup);
1958
1959 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1960 MODULE_LICENSE("GPL");