Pileus Git - ~andy/linux/blob - net/openvswitch/datapath.c

   1 /*
   2  * Copyright (c) 2007-2012 Nicira Networks.
   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/openvswitch.h>
  48 #include <linux/rculist.h>
  49 #include <linux/dmi.h>
  50 #include <linux/workqueue.h>
  51 #include <net/genetlink.h>
  52
  53 #include "datapath.h"
  54 #include "flow.h"
  55 #include "vport-internal_dev.h"
  56
  57 /**
  58  * DOC: Locking:
  59  *
  60  * Writes to device state (add/remove datapath, port, set operations on vports,
  61  * etc.) are protected by RTNL.
  62  *
  63  * Writes to other state (flow table modifications, set miscellaneous datapath
  64  * parameters, etc.) are protected by genl_mutex.  The RTNL lock nests inside
  65  * genl_mutex.
  66  *
  67  * Reads are protected by RCU.
  68  *
  69  * There are a few special cases (mostly stats) that have their own
  70  * synchronization but they nest under all of above and don't interact with
  71  * each other.
  72  */
  73
  74 /* Global list of datapaths to enable dumping them all out.
  75  * Protected by genl_mutex.
  76  */
  77 static LIST_HEAD(dps);
  78
  79 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
  80 static void rehash_flow_table(struct work_struct *work);
  81 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
  82
  83 static struct vport *new_vport(const struct vport_parms *);
  84 static int queue_gso_packets(int dp_ifindex, struct sk_buff *,
  85                              const struct dp_upcall_info *);
  86 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *,
  87                                   const struct dp_upcall_info *);
  88
  89 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  90 static struct datapath *get_dp(int dp_ifindex)
  91 {
  92         struct datapath *dp = NULL;
  93         struct net_device *dev;
  94
  95         rcu_read_lock();
  96         dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
  97         if (dev) {
  98                 struct vport *vport = ovs_internal_dev_get_vport(dev);
  99                 if (vport)
 100                         dp = vport->dp;
 101         }
 102         rcu_read_unlock();
 103
 104         return dp;
 105 }
 106
 107 /* Must be called with rcu_read_lock or RTNL lock. */
 108 const char *ovs_dp_name(const struct datapath *dp)
 109 {
 110         struct vport *vport = rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]);
 111         return vport->ops->get_name(vport);
 112 }
 113
 114 static int get_dpifindex(struct datapath *dp)
 115 {
 116         struct vport *local;
 117         int ifindex;
 118
 119         rcu_read_lock();
 120
 121         local = rcu_dereference(dp->ports[OVSP_LOCAL]);
 122         if (local)
 123                 ifindex = local->ops->get_ifindex(local);
 124         else
 125                 ifindex = 0;
 126
 127         rcu_read_unlock();
 128
 129         return ifindex;
 130 }
 131
 132 static void destroy_dp_rcu(struct rcu_head *rcu)
 133 {
 134         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 135
 136         ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
 137         free_percpu(dp->stats_percpu);
 138         kfree(dp);
 139 }
 140
 141 /* Called with RTNL lock and genl_lock. */
 142 static struct vport *new_vport(const struct vport_parms *parms)
 143 {
 144         struct vport *vport;
 145
 146         vport = ovs_vport_add(parms);
 147         if (!IS_ERR(vport)) {
 148                 struct datapath *dp = parms->dp;
 149
 150                 rcu_assign_pointer(dp->ports[parms->port_no], vport);
 151                 list_add(&vport->node, &dp->port_list);
 152         }
 153
 154         return vport;
 155 }
 156
 157 /* Called with RTNL lock. */
 158 void ovs_dp_detach_port(struct vport *p)
 159 {
 160         ASSERT_RTNL();
 161
 162         /* First drop references to device. */
 163         list_del(&p->node);
 164         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 165
 166         /* Then destroy it. */
 167         ovs_vport_del(p);
 168 }
 169
 170 /* Must be called with rcu_read_lock. */
 171 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 172 {
 173         struct datapath *dp = p->dp;
 174         struct sw_flow *flow;
 175         struct dp_stats_percpu *stats;
 176         struct sw_flow_key key;
 177         u64 *stats_counter;
 178         int error;
 179         int key_len;
 180
 181         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 182
 183         /* Extract flow from 'skb' into 'key'. */
 184         error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
 185         if (unlikely(error)) {
 186                 kfree_skb(skb);
 187                 return;
 188         }
 189
 190         /* Look up flow. */
 191         flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
 192         if (unlikely(!flow)) {
 193                 struct dp_upcall_info upcall;
 194
 195                 upcall.cmd = OVS_PACKET_CMD_MISS;
 196                 upcall.key = &key;
 197                 upcall.userdata = NULL;
 198                 upcall.pid = p->upcall_pid;
 199                 ovs_dp_upcall(dp, skb, &upcall);
 200                 consume_skb(skb);
 201                 stats_counter = &stats->n_missed;
 202                 goto out;
 203         }
 204
 205         OVS_CB(skb)->flow = flow;
 206
 207         stats_counter = &stats->n_hit;
 208         ovs_flow_used(OVS_CB(skb)->flow, skb);
 209         ovs_execute_actions(dp, skb);
 210
 211 out:
 212         /* Update datapath statistics. */
 213         u64_stats_update_begin(&stats->sync);
 214         (*stats_counter)++;
 215         u64_stats_update_end(&stats->sync);
 216 }
 217
 218 static struct genl_family dp_packet_genl_family = {
 219         .id = GENL_ID_GENERATE,
 220         .hdrsize = sizeof(struct ovs_header),
 221         .name = OVS_PACKET_FAMILY,
 222         .version = OVS_PACKET_VERSION,
 223         .maxattr = OVS_PACKET_ATTR_MAX
 224 };
 225
 226 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 227               const struct dp_upcall_info *upcall_info)
 228 {
 229         struct dp_stats_percpu *stats;
 230         int dp_ifindex;
 231         int err;
 232
 233         if (upcall_info->pid == 0) {
 234                 err = -ENOTCONN;
 235                 goto err;
 236         }
 237
 238         dp_ifindex = get_dpifindex(dp);
 239         if (!dp_ifindex) {
 240                 err = -ENODEV;
 241                 goto err;
 242         }
 243
 244         if (!skb_is_gso(skb))
 245                 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
 246         else
 247                 err = queue_gso_packets(dp_ifindex, skb, upcall_info);
 248         if (err)
 249                 goto err;
 250
 251         return 0;
 252
 253 err:
 254         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 255
 256         u64_stats_update_begin(&stats->sync);
 257         stats->n_lost++;
 258         u64_stats_update_end(&stats->sync);
 259
 260         return err;
 261 }
 262
 263 static int queue_gso_packets(int dp_ifindex, struct sk_buff *skb,
 264                              const struct dp_upcall_info *upcall_info)
 265 {
 266         struct dp_upcall_info later_info;
 267         struct sw_flow_key later_key;
 268         struct sk_buff *segs, *nskb;
 269         int err;
 270
 271         segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 272         if (IS_ERR(skb))
 273                 return PTR_ERR(skb);
 274
 275         /* Queue all of the segments. */
 276         skb = segs;
 277         do {
 278                 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
 279                 if (err)
 280                         break;
 281
 282                 if (skb == segs && skb_shinfo(skb)->gso_type & SKB_GSO_UDP) {
 283                         /* The initial flow key extracted by ovs_flow_extract()
 284                          * in this case is for a first fragment, so we need to
 285                          * properly mark later fragments.
 286                          */
 287                         later_key = *upcall_info->key;
 288                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 289
 290                         later_info = *upcall_info;
 291                         later_info.key = &later_key;
 292                         upcall_info = &later_info;
 293                 }
 294         } while ((skb = skb->next));
 295
 296         /* Free all of the segments. */
 297         skb = segs;
 298         do {
 299                 nskb = skb->next;
 300                 if (err)
 301                         kfree_skb(skb);
 302                 else
 303                         consume_skb(skb);
 304         } while ((skb = nskb));
 305         return err;
 306 }
 307
 308 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *skb,
 309                                   const struct dp_upcall_info *upcall_info)
 310 {
 311         struct ovs_header *upcall;
 312         struct sk_buff *nskb = NULL;
 313         struct sk_buff *user_skb; /* to be queued to userspace */
 314         struct nlattr *nla;
 315         unsigned int len;
 316         int err;
 317
 318         if (vlan_tx_tag_present(skb)) {
 319                 nskb = skb_clone(skb, GFP_ATOMIC);
 320                 if (!nskb)
 321                         return -ENOMEM;
 322
 323                 nskb = __vlan_put_tag(nskb, vlan_tx_tag_get(nskb));
 324                 if (!skb)
 325                         return -ENOMEM;
 326
 327                 nskb->vlan_tci = 0;
 328                 skb = nskb;
 329         }
 330
 331         if (nla_attr_size(skb->len) > USHRT_MAX) {
 332                 err = -EFBIG;
 333                 goto out;
 334         }
 335
 336         len = sizeof(struct ovs_header);
 337         len += nla_total_size(skb->len);
 338         len += nla_total_size(FLOW_BUFSIZE);
 339         if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
 340                 len += nla_total_size(8);
 341
 342         user_skb = genlmsg_new(len, GFP_ATOMIC);
 343         if (!user_skb) {
 344                 err = -ENOMEM;
 345                 goto out;
 346         }
 347
 348         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 349                              0, upcall_info->cmd);
 350         upcall->dp_ifindex = dp_ifindex;
 351
 352         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 353         ovs_flow_to_nlattrs(upcall_info->key, user_skb);
 354         nla_nest_end(user_skb, nla);
 355
 356         if (upcall_info->userdata)
 357                 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
 358                             nla_get_u64(upcall_info->userdata));
 359
 360         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 361
 362         skb_copy_and_csum_dev(skb, nla_data(nla));
 363
 364         err = genlmsg_unicast(&init_net, user_skb, upcall_info->pid);
 365
 366 out:
 367         kfree_skb(nskb);
 368         return err;
 369 }
 370
 371 /* Called with genl_mutex. */
 372 static int flush_flows(int dp_ifindex)
 373 {
 374         struct flow_table *old_table;
 375         struct flow_table *new_table;
 376         struct datapath *dp;
 377
 378         dp = get_dp(dp_ifindex);
 379         if (!dp)
 380                 return -ENODEV;
 381
 382         old_table = genl_dereference(dp->table);
 383         new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
 384         if (!new_table)
 385                 return -ENOMEM;
 386
 387         rcu_assign_pointer(dp->table, new_table);
 388
 389         ovs_flow_tbl_deferred_destroy(old_table);
 390         return 0;
 391 }
 392
 393 static int validate_actions(const struct nlattr *attr,
 394                                 const struct sw_flow_key *key, int depth);
 395
 396 static int validate_sample(const struct nlattr *attr,
 397                                 const struct sw_flow_key *key, int depth)
 398 {
 399         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 400         const struct nlattr *probability, *actions;
 401         const struct nlattr *a;
 402         int rem;
 403
 404         memset(attrs, 0, sizeof(attrs));
 405         nla_for_each_nested(a, attr, rem) {
 406                 int type = nla_type(a);
 407                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 408                         return -EINVAL;
 409                 attrs[type] = a;
 410         }
 411         if (rem)
 412                 return -EINVAL;
 413
 414         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 415         if (!probability || nla_len(probability) != sizeof(u32))
 416                 return -EINVAL;
 417
 418         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 419         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 420                 return -EINVAL;
 421         return validate_actions(actions, key, depth + 1);
 422 }
 423
 424 static int validate_set(const struct nlattr *a,
 425                         const struct sw_flow_key *flow_key)
 426 {
 427         const struct nlattr *ovs_key = nla_data(a);
 428         int key_type = nla_type(ovs_key);
 429
 430         /* There can be only one key in a action */
 431         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 432                 return -EINVAL;
 433
 434         if (key_type > OVS_KEY_ATTR_MAX ||
 435             nla_len(ovs_key) != ovs_key_lens[key_type])
 436                 return -EINVAL;
 437
 438         switch (key_type) {
 439         const struct ovs_key_ipv4 *ipv4_key;
 440
 441         case OVS_KEY_ATTR_PRIORITY:
 442         case OVS_KEY_ATTR_ETHERNET:
 443                 break;
 444
 445         case OVS_KEY_ATTR_IPV4:
 446                 if (flow_key->eth.type != htons(ETH_P_IP))
 447                         return -EINVAL;
 448
 449                 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
 450                         return -EINVAL;
 451
 452                 ipv4_key = nla_data(ovs_key);
 453                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 454                         return -EINVAL;
 455
 456                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 457                         return -EINVAL;
 458
 459                 break;
 460
 461         case OVS_KEY_ATTR_TCP:
 462                 if (flow_key->ip.proto != IPPROTO_TCP)
 463                         return -EINVAL;
 464
 465                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 466                         return -EINVAL;
 467
 468                 break;
 469
 470         case OVS_KEY_ATTR_UDP:
 471                 if (flow_key->ip.proto != IPPROTO_UDP)
 472                         return -EINVAL;
 473
 474                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 475                         return -EINVAL;
 476                 break;
 477
 478         default:
 479                 return -EINVAL;
 480         }
 481
 482         return 0;
 483 }
 484
 485 static int validate_userspace(const struct nlattr *attr)
 486 {
 487         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 488                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 489                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
 490         };
 491         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 492         int error;
 493
 494         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 495                                  attr, userspace_policy);
 496         if (error)
 497                 return error;
 498
 499         if (!a[OVS_USERSPACE_ATTR_PID] ||
 500             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 501                 return -EINVAL;
 502
 503         return 0;
 504 }
 505
 506 static int validate_actions(const struct nlattr *attr,
 507                                 const struct sw_flow_key *key,  int depth)
 508 {
 509         const struct nlattr *a;
 510         int rem, err;
 511
 512         if (depth >= SAMPLE_ACTION_DEPTH)
 513                 return -EOVERFLOW;
 514
 515         nla_for_each_nested(a, attr, rem) {
 516                 /* Expected argument lengths, (u32)-1 for variable length. */
 517                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 518                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 519                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 520                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 521                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 522                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 523                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 524                 };
 525                 const struct ovs_action_push_vlan *vlan;
 526                 int type = nla_type(a);
 527
 528                 if (type > OVS_ACTION_ATTR_MAX ||
 529                     (action_lens[type] != nla_len(a) &&
 530                      action_lens[type] != (u32)-1))
 531                         return -EINVAL;
 532
 533                 switch (type) {
 534                 case OVS_ACTION_ATTR_UNSPEC:
 535                         return -EINVAL;
 536
 537                 case OVS_ACTION_ATTR_USERSPACE:
 538                         err = validate_userspace(a);
 539                         if (err)
 540                                 return err;
 541                         break;
 542
 543                 case OVS_ACTION_ATTR_OUTPUT:
 544                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 545                                 return -EINVAL;
 546                         break;
 547
 548
 549                 case OVS_ACTION_ATTR_POP_VLAN:
 550                         break;
 551
 552                 case OVS_ACTION_ATTR_PUSH_VLAN:
 553                         vlan = nla_data(a);
 554                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 555                                 return -EINVAL;
 556                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 557                                 return -EINVAL;
 558                         break;
 559
 560                 case OVS_ACTION_ATTR_SET:
 561                         err = validate_set(a, key);
 562                         if (err)
 563                                 return err;
 564                         break;
 565
 566                 case OVS_ACTION_ATTR_SAMPLE:
 567                         err = validate_sample(a, key, depth);
 568                         if (err)
 569                                 return err;
 570                         break;
 571
 572                 default:
 573                         return -EINVAL;
 574                 }
 575         }
 576
 577         if (rem > 0)
 578                 return -EINVAL;
 579
 580         return 0;
 581 }
 582
 583 static void clear_stats(struct sw_flow *flow)
 584 {
 585         flow->used = 0;
 586         flow->tcp_flags = 0;
 587         flow->packet_count = 0;
 588         flow->byte_count = 0;
 589 }
 590
 591 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 592 {
 593         struct ovs_header *ovs_header = info->userhdr;
 594         struct nlattr **a = info->attrs;
 595         struct sw_flow_actions *acts;
 596         struct sk_buff *packet;
 597         struct sw_flow *flow;
 598         struct datapath *dp;
 599         struct ethhdr *eth;
 600         int len;
 601         int err;
 602         int key_len;
 603
 604         err = -EINVAL;
 605         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 606             !a[OVS_PACKET_ATTR_ACTIONS] ||
 607             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 608                 goto err;
 609
 610         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 611         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 612         err = -ENOMEM;
 613         if (!packet)
 614                 goto err;
 615         skb_reserve(packet, NET_IP_ALIGN);
 616
 617         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 618
 619         skb_reset_mac_header(packet);
 620         eth = eth_hdr(packet);
 621
 622         /* Normally, setting the skb 'protocol' field would be handled by a
 623          * call to eth_type_trans(), but it assumes there's a sending
 624          * device, which we may not have. */
 625         if (ntohs(eth->h_proto) >= 1536)
 626                 packet->protocol = eth->h_proto;
 627         else
 628                 packet->protocol = htons(ETH_P_802_2);
 629
 630         /* Build an sw_flow for sending this packet. */
 631         flow = ovs_flow_alloc();
 632         err = PTR_ERR(flow);
 633         if (IS_ERR(flow))
 634                 goto err_kfree_skb;
 635
 636         err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
 637         if (err)
 638                 goto err_flow_free;
 639
 640         err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
 641                                              &flow->key.phy.in_port,
 642                                              a[OVS_PACKET_ATTR_KEY]);
 643         if (err)
 644                 goto err_flow_free;
 645
 646         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
 647         if (err)
 648                 goto err_flow_free;
 649
 650         flow->hash = ovs_flow_hash(&flow->key, key_len);
 651
 652         acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 653         err = PTR_ERR(acts);
 654         if (IS_ERR(acts))
 655                 goto err_flow_free;
 656         rcu_assign_pointer(flow->sf_acts, acts);
 657
 658         OVS_CB(packet)->flow = flow;
 659         packet->priority = flow->key.phy.priority;
 660
 661         rcu_read_lock();
 662         dp = get_dp(ovs_header->dp_ifindex);
 663         err = -ENODEV;
 664         if (!dp)
 665                 goto err_unlock;
 666
 667         local_bh_disable();
 668         err = ovs_execute_actions(dp, packet);
 669         local_bh_enable();
 670         rcu_read_unlock();
 671
 672         ovs_flow_free(flow);
 673         return err;
 674
 675 err_unlock:
 676         rcu_read_unlock();
 677 err_flow_free:
 678         ovs_flow_free(flow);
 679 err_kfree_skb:
 680         kfree_skb(packet);
 681 err:
 682         return err;
 683 }
 684
 685 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 686         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 687         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 688         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 689 };
 690
 691 static struct genl_ops dp_packet_genl_ops[] = {
 692         { .cmd = OVS_PACKET_CMD_EXECUTE,
 693           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 694           .policy = packet_policy,
 695           .doit = ovs_packet_cmd_execute
 696         }
 697 };
 698
 699 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 700 {
 701         int i;
 702         struct flow_table *table = genl_dereference(dp->table);
 703
 704         stats->n_flows = ovs_flow_tbl_count(table);
 705
 706         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 707         for_each_possible_cpu(i) {
 708                 const struct dp_stats_percpu *percpu_stats;
 709                 struct dp_stats_percpu local_stats;
 710                 unsigned int start;
 711
 712                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 713
 714                 do {
 715                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 716                         local_stats = *percpu_stats;
 717                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 718
 719                 stats->n_hit += local_stats.n_hit;
 720                 stats->n_missed += local_stats.n_missed;
 721                 stats->n_lost += local_stats.n_lost;
 722         }
 723 }
 724
 725 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 726         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 727         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 728         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 729 };
 730
 731 static struct genl_family dp_flow_genl_family = {
 732         .id = GENL_ID_GENERATE,
 733         .hdrsize = sizeof(struct ovs_header),
 734         .name = OVS_FLOW_FAMILY,
 735         .version = OVS_FLOW_VERSION,
 736         .maxattr = OVS_FLOW_ATTR_MAX
 737 };
 738
 739 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 740         .name = OVS_FLOW_MCGROUP
 741 };
 742
 743 /* Called with genl_lock. */
 744 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 745                                   struct sk_buff *skb, u32 pid,
 746                                   u32 seq, u32 flags, u8 cmd)
 747 {
 748         const int skb_orig_len = skb->len;
 749         const struct sw_flow_actions *sf_acts;
 750         struct ovs_flow_stats stats;
 751         struct ovs_header *ovs_header;
 752         struct nlattr *nla;
 753         unsigned long used;
 754         u8 tcp_flags;
 755         int err;
 756
 757         sf_acts = rcu_dereference_protected(flow->sf_acts,
 758                                             lockdep_genl_is_held());
 759
 760         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 761         if (!ovs_header)
 762                 return -EMSGSIZE;
 763
 764         ovs_header->dp_ifindex = get_dpifindex(dp);
 765
 766         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 767         if (!nla)
 768                 goto nla_put_failure;
 769         err = ovs_flow_to_nlattrs(&flow->key, skb);
 770         if (err)
 771                 goto error;
 772         nla_nest_end(skb, nla);
 773
 774         spin_lock_bh(&flow->lock);
 775         used = flow->used;
 776         stats.n_packets = flow->packet_count;
 777         stats.n_bytes = flow->byte_count;
 778         tcp_flags = flow->tcp_flags;
 779         spin_unlock_bh(&flow->lock);
 780
 781         if (used &&
 782             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 783                 goto nla_put_failure;
 784
 785         if (stats.n_packets &&
 786             nla_put(skb, OVS_FLOW_ATTR_STATS,
 787                     sizeof(struct ovs_flow_stats), &stats))
 788                 goto nla_put_failure;
 789
 790         if (tcp_flags &&
 791             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
 792                 goto nla_put_failure;
 793
 794         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 795          * this is the first flow to be dumped into 'skb'.  This is unusual for
 796          * Netlink but individual action lists can be longer than
 797          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 798          * The userspace caller can always fetch the actions separately if it
 799          * really wants them.  (Most userspace callers in fact don't care.)
 800          *
 801          * This can only fail for dump operations because the skb is always
 802          * properly sized for single flows.
 803          */
 804         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 805                       sf_acts->actions);
 806         if (err < 0 && skb_orig_len)
 807                 goto error;
 808
 809         return genlmsg_end(skb, ovs_header);
 810
 811 nla_put_failure:
 812         err = -EMSGSIZE;
 813 error:
 814         genlmsg_cancel(skb, ovs_header);
 815         return err;
 816 }
 817
 818 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 819 {
 820         const struct sw_flow_actions *sf_acts;
 821         int len;
 822
 823         sf_acts = rcu_dereference_protected(flow->sf_acts,
 824                                             lockdep_genl_is_held());
 825
 826         /* OVS_FLOW_ATTR_KEY */
 827         len = nla_total_size(FLOW_BUFSIZE);
 828         /* OVS_FLOW_ATTR_ACTIONS */
 829         len += nla_total_size(sf_acts->actions_len);
 830         /* OVS_FLOW_ATTR_STATS */
 831         len += nla_total_size(sizeof(struct ovs_flow_stats));
 832         /* OVS_FLOW_ATTR_TCP_FLAGS */
 833         len += nla_total_size(1);
 834         /* OVS_FLOW_ATTR_USED */
 835         len += nla_total_size(8);
 836
 837         len += NLMSG_ALIGN(sizeof(struct ovs_header));
 838
 839         return genlmsg_new(len, GFP_KERNEL);
 840 }
 841
 842 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 843                                                struct datapath *dp,
 844                                                u32 pid, u32 seq, u8 cmd)
 845 {
 846         struct sk_buff *skb;
 847         int retval;
 848
 849         skb = ovs_flow_cmd_alloc_info(flow);
 850         if (!skb)
 851                 return ERR_PTR(-ENOMEM);
 852
 853         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 854         BUG_ON(retval < 0);
 855         return skb;
 856 }
 857
 858 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 859 {
 860         struct nlattr **a = info->attrs;
 861         struct ovs_header *ovs_header = info->userhdr;
 862         struct sw_flow_key key;
 863         struct sw_flow *flow;
 864         struct sk_buff *reply;
 865         struct datapath *dp;
 866         struct flow_table *table;
 867         int error;
 868         int key_len;
 869
 870         /* Extract key. */
 871         error = -EINVAL;
 872         if (!a[OVS_FLOW_ATTR_KEY])
 873                 goto error;
 874         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
 875         if (error)
 876                 goto error;
 877
 878         /* Validate actions. */
 879         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 880                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
 881                 if (error)
 882                         goto error;
 883         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 884                 error = -EINVAL;
 885                 goto error;
 886         }
 887
 888         dp = get_dp(ovs_header->dp_ifindex);
 889         error = -ENODEV;
 890         if (!dp)
 891                 goto error;
 892
 893         table = genl_dereference(dp->table);
 894         flow = ovs_flow_tbl_lookup(table, &key, key_len);
 895         if (!flow) {
 896                 struct sw_flow_actions *acts;
 897
 898                 /* Bail out if we're not allowed to create a new flow. */
 899                 error = -ENOENT;
 900                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
 901                         goto error;
 902
 903                 /* Expand table, if necessary, to make room. */
 904                 if (ovs_flow_tbl_need_to_expand(table)) {
 905                         struct flow_table *new_table;
 906
 907                         new_table = ovs_flow_tbl_expand(table);
 908                         if (!IS_ERR(new_table)) {
 909                                 rcu_assign_pointer(dp->table, new_table);
 910                                 ovs_flow_tbl_deferred_destroy(table);
 911                                 table = genl_dereference(dp->table);
 912                         }
 913                 }
 914
 915                 /* Allocate flow. */
 916                 flow = ovs_flow_alloc();
 917                 if (IS_ERR(flow)) {
 918                         error = PTR_ERR(flow);
 919                         goto error;
 920                 }
 921                 flow->key = key;
 922                 clear_stats(flow);
 923
 924                 /* Obtain actions. */
 925                 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
 926                 error = PTR_ERR(acts);
 927                 if (IS_ERR(acts))
 928                         goto error_free_flow;
 929                 rcu_assign_pointer(flow->sf_acts, acts);
 930
 931                 /* Put flow in bucket. */
 932                 flow->hash = ovs_flow_hash(&key, key_len);
 933                 ovs_flow_tbl_insert(table, flow);
 934
 935                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
 936                                                 info->snd_seq,
 937                                                 OVS_FLOW_CMD_NEW);
 938         } else {
 939                 /* We found a matching flow. */
 940                 struct sw_flow_actions *old_acts;
 941                 struct nlattr *acts_attrs;
 942
 943                 /* Bail out if we're not allowed to modify an existing flow.
 944                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 945                  * because Generic Netlink treats the latter as a dump
 946                  * request.  We also accept NLM_F_EXCL in case that bug ever
 947                  * gets fixed.
 948                  */
 949                 error = -EEXIST;
 950                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
 951                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
 952                         goto error;
 953
 954                 /* Update actions. */
 955                 old_acts = rcu_dereference_protected(flow->sf_acts,
 956                                                      lockdep_genl_is_held());
 957                 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
 958                 if (acts_attrs &&
 959                    (old_acts->actions_len != nla_len(acts_attrs) ||
 960                    memcmp(old_acts->actions, nla_data(acts_attrs),
 961                           old_acts->actions_len))) {
 962                         struct sw_flow_actions *new_acts;
 963
 964                         new_acts = ovs_flow_actions_alloc(acts_attrs);
 965                         error = PTR_ERR(new_acts);
 966                         if (IS_ERR(new_acts))
 967                                 goto error;
 968
 969                         rcu_assign_pointer(flow->sf_acts, new_acts);
 970                         ovs_flow_deferred_free_acts(old_acts);
 971                 }
 972
 973                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
 974                                                info->snd_seq, OVS_FLOW_CMD_NEW);
 975
 976                 /* Clear stats. */
 977                 if (a[OVS_FLOW_ATTR_CLEAR]) {
 978                         spin_lock_bh(&flow->lock);
 979                         clear_stats(flow);
 980                         spin_unlock_bh(&flow->lock);
 981                 }
 982         }
 983
 984         if (!IS_ERR(reply))
 985                 genl_notify(reply, genl_info_net(info), info->snd_pid,
 986                            ovs_dp_flow_multicast_group.id, info->nlhdr,
 987                            GFP_KERNEL);
 988         else
 989                 netlink_set_err(init_net.genl_sock, 0,
 990                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
 991         return 0;
 992
 993 error_free_flow:
 994         ovs_flow_free(flow);
 995 error:
 996         return error;
 997 }
 998
 999 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1000 {
1001         struct nlattr **a = info->attrs;
1002         struct ovs_header *ovs_header = info->userhdr;
1003         struct sw_flow_key key;
1004         struct sk_buff *reply;
1005         struct sw_flow *flow;
1006         struct datapath *dp;
1007         struct flow_table *table;
1008         int err;
1009         int key_len;
1010
1011         if (!a[OVS_FLOW_ATTR_KEY])
1012                 return -EINVAL;
1013         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1014         if (err)
1015                 return err;
1016
1017         dp = get_dp(ovs_header->dp_ifindex);
1018         if (!dp)
1019                 return -ENODEV;
1020
1021         table = genl_dereference(dp->table);
1022         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1023         if (!flow)
1024                 return -ENOENT;
1025
1026         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1027                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1028         if (IS_ERR(reply))
1029                 return PTR_ERR(reply);
1030
1031         return genlmsg_reply(reply, info);
1032 }
1033
1034 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1035 {
1036         struct nlattr **a = info->attrs;
1037         struct ovs_header *ovs_header = info->userhdr;
1038         struct sw_flow_key key;
1039         struct sk_buff *reply;
1040         struct sw_flow *flow;
1041         struct datapath *dp;
1042         struct flow_table *table;
1043         int err;
1044         int key_len;
1045
1046         if (!a[OVS_FLOW_ATTR_KEY])
1047                 return flush_flows(ovs_header->dp_ifindex);
1048         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1049         if (err)
1050                 return err;
1051
1052         dp = get_dp(ovs_header->dp_ifindex);
1053         if (!dp)
1054                 return -ENODEV;
1055
1056         table = genl_dereference(dp->table);
1057         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1058         if (!flow)
1059                 return -ENOENT;
1060
1061         reply = ovs_flow_cmd_alloc_info(flow);
1062         if (!reply)
1063                 return -ENOMEM;
1064
1065         ovs_flow_tbl_remove(table, flow);
1066
1067         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1068                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1069         BUG_ON(err < 0);
1070
1071         ovs_flow_deferred_free(flow);
1072
1073         genl_notify(reply, genl_info_net(info), info->snd_pid,
1074                     ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1075         return 0;
1076 }
1077
1078 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1079 {
1080         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1081         struct datapath *dp;
1082         struct flow_table *table;
1083
1084         dp = get_dp(ovs_header->dp_ifindex);
1085         if (!dp)
1086                 return -ENODEV;
1087
1088         table = genl_dereference(dp->table);
1089
1090         for (;;) {
1091                 struct sw_flow *flow;
1092                 u32 bucket, obj;
1093
1094                 bucket = cb->args[0];
1095                 obj = cb->args[1];
1096                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1097                 if (!flow)
1098                         break;
1099
1100                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1101                                            NETLINK_CB(cb->skb).pid,
1102                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1103                                            OVS_FLOW_CMD_NEW) < 0)
1104                         break;
1105
1106                 cb->args[0] = bucket;
1107                 cb->args[1] = obj;
1108         }
1109         return skb->len;
1110 }
1111
1112 static struct genl_ops dp_flow_genl_ops[] = {
1113         { .cmd = OVS_FLOW_CMD_NEW,
1114           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1115           .policy = flow_policy,
1116           .doit = ovs_flow_cmd_new_or_set
1117         },
1118         { .cmd = OVS_FLOW_CMD_DEL,
1119           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1120           .policy = flow_policy,
1121           .doit = ovs_flow_cmd_del
1122         },
1123         { .cmd = OVS_FLOW_CMD_GET,
1124           .flags = 0,               /* OK for unprivileged users. */
1125           .policy = flow_policy,
1126           .doit = ovs_flow_cmd_get,
1127           .dumpit = ovs_flow_cmd_dump
1128         },
1129         { .cmd = OVS_FLOW_CMD_SET,
1130           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1131           .policy = flow_policy,
1132           .doit = ovs_flow_cmd_new_or_set,
1133         },
1134 };
1135
1136 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1137         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1138         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1139 };
1140
1141 static struct genl_family dp_datapath_genl_family = {
1142         .id = GENL_ID_GENERATE,
1143         .hdrsize = sizeof(struct ovs_header),
1144         .name = OVS_DATAPATH_FAMILY,
1145         .version = OVS_DATAPATH_VERSION,
1146         .maxattr = OVS_DP_ATTR_MAX
1147 };
1148
1149 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1150         .name = OVS_DATAPATH_MCGROUP
1151 };
1152
1153 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1154                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1155 {
1156         struct ovs_header *ovs_header;
1157         struct ovs_dp_stats dp_stats;
1158         int err;
1159
1160         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1161                                    flags, cmd);
1162         if (!ovs_header)
1163                 goto error;
1164
1165         ovs_header->dp_ifindex = get_dpifindex(dp);
1166
1167         rcu_read_lock();
1168         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1169         rcu_read_unlock();
1170         if (err)
1171                 goto nla_put_failure;
1172
1173         get_dp_stats(dp, &dp_stats);
1174         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1175                 goto nla_put_failure;
1176
1177         return genlmsg_end(skb, ovs_header);
1178
1179 nla_put_failure:
1180         genlmsg_cancel(skb, ovs_header);
1181 error:
1182         return -EMSGSIZE;
1183 }
1184
1185 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1186                                              u32 seq, u8 cmd)
1187 {
1188         struct sk_buff *skb;
1189         int retval;
1190
1191         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1192         if (!skb)
1193                 return ERR_PTR(-ENOMEM);
1194
1195         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1196         if (retval < 0) {
1197                 kfree_skb(skb);
1198                 return ERR_PTR(retval);
1199         }
1200         return skb;
1201 }
1202
1203 /* Called with genl_mutex and optionally with RTNL lock also. */
1204 static struct datapath *lookup_datapath(struct ovs_header *ovs_header,
1205                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1206 {
1207         struct datapath *dp;
1208
1209         if (!a[OVS_DP_ATTR_NAME])
1210                 dp = get_dp(ovs_header->dp_ifindex);
1211         else {
1212                 struct vport *vport;
1213
1214                 rcu_read_lock();
1215                 vport = ovs_vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1216                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1217                 rcu_read_unlock();
1218         }
1219         return dp ? dp : ERR_PTR(-ENODEV);
1220 }
1221
1222 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1223 {
1224         struct nlattr **a = info->attrs;
1225         struct vport_parms parms;
1226         struct sk_buff *reply;
1227         struct datapath *dp;
1228         struct vport *vport;
1229         int err;
1230
1231         err = -EINVAL;
1232         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1233                 goto err;
1234
1235         rtnl_lock();
1236         err = -ENODEV;
1237         if (!try_module_get(THIS_MODULE))
1238                 goto err_unlock_rtnl;
1239
1240         err = -ENOMEM;
1241         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1242         if (dp == NULL)
1243                 goto err_put_module;
1244         INIT_LIST_HEAD(&dp->port_list);
1245
1246         /* Allocate table. */
1247         err = -ENOMEM;
1248         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1249         if (!dp->table)
1250                 goto err_free_dp;
1251
1252         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1253         if (!dp->stats_percpu) {
1254                 err = -ENOMEM;
1255                 goto err_destroy_table;
1256         }
1257
1258         /* Set up our datapath device. */
1259         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1260         parms.type = OVS_VPORT_TYPE_INTERNAL;
1261         parms.options = NULL;
1262         parms.dp = dp;
1263         parms.port_no = OVSP_LOCAL;
1264         parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1265
1266         vport = new_vport(&parms);
1267         if (IS_ERR(vport)) {
1268                 err = PTR_ERR(vport);
1269                 if (err == -EBUSY)
1270                         err = -EEXIST;
1271
1272                 goto err_destroy_percpu;
1273         }
1274
1275         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1276                                       info->snd_seq, OVS_DP_CMD_NEW);
1277         err = PTR_ERR(reply);
1278         if (IS_ERR(reply))
1279                 goto err_destroy_local_port;
1280
1281         list_add_tail(&dp->list_node, &dps);
1282         rtnl_unlock();
1283
1284         genl_notify(reply, genl_info_net(info), info->snd_pid,
1285                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1286                     GFP_KERNEL);
1287         return 0;
1288
1289 err_destroy_local_port:
1290         ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL]));
1291 err_destroy_percpu:
1292         free_percpu(dp->stats_percpu);
1293 err_destroy_table:
1294         ovs_flow_tbl_destroy(genl_dereference(dp->table));
1295 err_free_dp:
1296         kfree(dp);
1297 err_put_module:
1298         module_put(THIS_MODULE);
1299 err_unlock_rtnl:
1300         rtnl_unlock();
1301 err:
1302         return err;
1303 }
1304
1305 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1306 {
1307         struct vport *vport, *next_vport;
1308         struct sk_buff *reply;
1309         struct datapath *dp;
1310         int err;
1311
1312         rtnl_lock();
1313         dp = lookup_datapath(info->userhdr, info->attrs);
1314         err = PTR_ERR(dp);
1315         if (IS_ERR(dp))
1316                 goto exit_unlock;
1317
1318         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1319                                       info->snd_seq, OVS_DP_CMD_DEL);
1320         err = PTR_ERR(reply);
1321         if (IS_ERR(reply))
1322                 goto exit_unlock;
1323
1324         list_for_each_entry_safe(vport, next_vport, &dp->port_list, node)
1325                 if (vport->port_no != OVSP_LOCAL)
1326                         ovs_dp_detach_port(vport);
1327
1328         list_del(&dp->list_node);
1329         ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL]));
1330
1331         /* rtnl_unlock() will wait until all the references to devices that
1332          * are pending unregistration have been dropped.  We do it here to
1333          * ensure that any internal devices (which contain DP pointers) are
1334          * fully destroyed before freeing the datapath.
1335          */
1336         rtnl_unlock();
1337
1338         call_rcu(&dp->rcu, destroy_dp_rcu);
1339         module_put(THIS_MODULE);
1340
1341         genl_notify(reply, genl_info_net(info), info->snd_pid,
1342                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1343                     GFP_KERNEL);
1344
1345         return 0;
1346
1347 exit_unlock:
1348         rtnl_unlock();
1349         return err;
1350 }
1351
1352 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1353 {
1354         struct sk_buff *reply;
1355         struct datapath *dp;
1356         int err;
1357
1358         dp = lookup_datapath(info->userhdr, info->attrs);
1359         if (IS_ERR(dp))
1360                 return PTR_ERR(dp);
1361
1362         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1363                                       info->snd_seq, OVS_DP_CMD_NEW);
1364         if (IS_ERR(reply)) {
1365                 err = PTR_ERR(reply);
1366                 netlink_set_err(init_net.genl_sock, 0,
1367                                 ovs_dp_datapath_multicast_group.id, err);
1368                 return 0;
1369         }
1370
1371         genl_notify(reply, genl_info_net(info), info->snd_pid,
1372                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1373                     GFP_KERNEL);
1374
1375         return 0;
1376 }
1377
1378 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1379 {
1380         struct sk_buff *reply;
1381         struct datapath *dp;
1382
1383         dp = lookup_datapath(info->userhdr, info->attrs);
1384         if (IS_ERR(dp))
1385                 return PTR_ERR(dp);
1386
1387         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1388                                       info->snd_seq, OVS_DP_CMD_NEW);
1389         if (IS_ERR(reply))
1390                 return PTR_ERR(reply);
1391
1392         return genlmsg_reply(reply, info);
1393 }
1394
1395 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1396 {
1397         struct datapath *dp;
1398         int skip = cb->args[0];
1399         int i = 0;
1400
1401         list_for_each_entry(dp, &dps, list_node) {
1402                 if (i >= skip &&
1403                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1404                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1405                                          OVS_DP_CMD_NEW) < 0)
1406                         break;
1407                 i++;
1408         }
1409
1410         cb->args[0] = i;
1411
1412         return skb->len;
1413 }
1414
1415 static struct genl_ops dp_datapath_genl_ops[] = {
1416         { .cmd = OVS_DP_CMD_NEW,
1417           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1418           .policy = datapath_policy,
1419           .doit = ovs_dp_cmd_new
1420         },
1421         { .cmd = OVS_DP_CMD_DEL,
1422           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1423           .policy = datapath_policy,
1424           .doit = ovs_dp_cmd_del
1425         },
1426         { .cmd = OVS_DP_CMD_GET,
1427           .flags = 0,               /* OK for unprivileged users. */
1428           .policy = datapath_policy,
1429           .doit = ovs_dp_cmd_get,
1430           .dumpit = ovs_dp_cmd_dump
1431         },
1432         { .cmd = OVS_DP_CMD_SET,
1433           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1434           .policy = datapath_policy,
1435           .doit = ovs_dp_cmd_set,
1436         },
1437 };
1438
1439 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1440         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1441         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1442         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1443         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1444         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1445         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1446 };
1447
1448 static struct genl_family dp_vport_genl_family = {
1449         .id = GENL_ID_GENERATE,
1450         .hdrsize = sizeof(struct ovs_header),
1451         .name = OVS_VPORT_FAMILY,
1452         .version = OVS_VPORT_VERSION,
1453         .maxattr = OVS_VPORT_ATTR_MAX
1454 };
1455
1456 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1457         .name = OVS_VPORT_MCGROUP
1458 };
1459
1460 /* Called with RTNL lock or RCU read lock. */
1461 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1462                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1463 {
1464         struct ovs_header *ovs_header;
1465         struct ovs_vport_stats vport_stats;
1466         int err;
1467
1468         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1469                                  flags, cmd);
1470         if (!ovs_header)
1471                 return -EMSGSIZE;
1472
1473         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1474
1475         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1476             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1477             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1478             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid))
1479                 goto nla_put_failure;
1480
1481         ovs_vport_get_stats(vport, &vport_stats);
1482         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1483                     &vport_stats))
1484                 goto nla_put_failure;
1485
1486         err = ovs_vport_get_options(vport, skb);
1487         if (err == -EMSGSIZE)
1488                 goto error;
1489
1490         return genlmsg_end(skb, ovs_header);
1491
1492 nla_put_failure:
1493         err = -EMSGSIZE;
1494 error:
1495         genlmsg_cancel(skb, ovs_header);
1496         return err;
1497 }
1498
1499 /* Called with RTNL lock or RCU read lock. */
1500 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1501                                          u32 seq, u8 cmd)
1502 {
1503         struct sk_buff *skb;
1504         int retval;
1505
1506         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1507         if (!skb)
1508                 return ERR_PTR(-ENOMEM);
1509
1510         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1511         if (retval < 0) {
1512                 kfree_skb(skb);
1513                 return ERR_PTR(retval);
1514         }
1515         return skb;
1516 }
1517
1518 /* Called with RTNL lock or RCU read lock. */
1519 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1520                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1521 {
1522         struct datapath *dp;
1523         struct vport *vport;
1524
1525         if (a[OVS_VPORT_ATTR_NAME]) {
1526                 vport = ovs_vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1527                 if (!vport)
1528                         return ERR_PTR(-ENODEV);
1529                 if (ovs_header->dp_ifindex &&
1530                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1531                         return ERR_PTR(-ENODEV);
1532                 return vport;
1533         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1534                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1535
1536                 if (port_no >= DP_MAX_PORTS)
1537                         return ERR_PTR(-EFBIG);
1538
1539                 dp = get_dp(ovs_header->dp_ifindex);
1540                 if (!dp)
1541                         return ERR_PTR(-ENODEV);
1542
1543                 vport = rcu_dereference_rtnl(dp->ports[port_no]);
1544                 if (!vport)
1545                         return ERR_PTR(-ENOENT);
1546                 return vport;
1547         } else
1548                 return ERR_PTR(-EINVAL);
1549 }
1550
1551 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1552 {
1553         struct nlattr **a = info->attrs;
1554         struct ovs_header *ovs_header = info->userhdr;
1555         struct vport_parms parms;
1556         struct sk_buff *reply;
1557         struct vport *vport;
1558         struct datapath *dp;
1559         u32 port_no;
1560         int err;
1561
1562         err = -EINVAL;
1563         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1564             !a[OVS_VPORT_ATTR_UPCALL_PID])
1565                 goto exit;
1566
1567         rtnl_lock();
1568         dp = get_dp(ovs_header->dp_ifindex);
1569         err = -ENODEV;
1570         if (!dp)
1571                 goto exit_unlock;
1572
1573         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1574                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1575
1576                 err = -EFBIG;
1577                 if (port_no >= DP_MAX_PORTS)
1578                         goto exit_unlock;
1579
1580                 vport = rtnl_dereference(dp->ports[port_no]);
1581                 err = -EBUSY;
1582                 if (vport)
1583                         goto exit_unlock;
1584         } else {
1585                 for (port_no = 1; ; port_no++) {
1586                         if (port_no >= DP_MAX_PORTS) {
1587                                 err = -EFBIG;
1588                                 goto exit_unlock;
1589                         }
1590                         vport = rtnl_dereference(dp->ports[port_no]);
1591                         if (!vport)
1592                                 break;
1593                 }
1594         }
1595
1596         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1597         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1598         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1599         parms.dp = dp;
1600         parms.port_no = port_no;
1601         parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1602
1603         vport = new_vport(&parms);
1604         err = PTR_ERR(vport);
1605         if (IS_ERR(vport))
1606                 goto exit_unlock;
1607
1608         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1609                                          OVS_VPORT_CMD_NEW);
1610         if (IS_ERR(reply)) {
1611                 err = PTR_ERR(reply);
1612                 ovs_dp_detach_port(vport);
1613                 goto exit_unlock;
1614         }
1615         genl_notify(reply, genl_info_net(info), info->snd_pid,
1616                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1617
1618 exit_unlock:
1619         rtnl_unlock();
1620 exit:
1621         return err;
1622 }
1623
1624 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1625 {
1626         struct nlattr **a = info->attrs;
1627         struct sk_buff *reply;
1628         struct vport *vport;
1629         int err;
1630
1631         rtnl_lock();
1632         vport = lookup_vport(info->userhdr, a);
1633         err = PTR_ERR(vport);
1634         if (IS_ERR(vport))
1635                 goto exit_unlock;
1636
1637         err = 0;
1638         if (a[OVS_VPORT_ATTR_TYPE] &&
1639             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1640                 err = -EINVAL;
1641
1642         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1643                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1644         if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1645                 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1646
1647         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1648                                          OVS_VPORT_CMD_NEW);
1649         if (IS_ERR(reply)) {
1650                 err = PTR_ERR(reply);
1651                 netlink_set_err(init_net.genl_sock, 0,
1652                                 ovs_dp_vport_multicast_group.id, err);
1653                 return 0;
1654         }
1655
1656         genl_notify(reply, genl_info_net(info), info->snd_pid,
1657                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1658
1659 exit_unlock:
1660         rtnl_unlock();
1661         return err;
1662 }
1663
1664 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1665 {
1666         struct nlattr **a = info->attrs;
1667         struct sk_buff *reply;
1668         struct vport *vport;
1669         int err;
1670
1671         rtnl_lock();
1672         vport = lookup_vport(info->userhdr, a);
1673         err = PTR_ERR(vport);
1674         if (IS_ERR(vport))
1675                 goto exit_unlock;
1676
1677         if (vport->port_no == OVSP_LOCAL) {
1678                 err = -EINVAL;
1679                 goto exit_unlock;
1680         }
1681
1682         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1683                                          OVS_VPORT_CMD_DEL);
1684         err = PTR_ERR(reply);
1685         if (IS_ERR(reply))
1686                 goto exit_unlock;
1687
1688         ovs_dp_detach_port(vport);
1689
1690         genl_notify(reply, genl_info_net(info), info->snd_pid,
1691                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1692
1693 exit_unlock:
1694         rtnl_unlock();
1695         return err;
1696 }
1697
1698 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1699 {
1700         struct nlattr **a = info->attrs;
1701         struct ovs_header *ovs_header = info->userhdr;
1702         struct sk_buff *reply;
1703         struct vport *vport;
1704         int err;
1705
1706         rcu_read_lock();
1707         vport = lookup_vport(ovs_header, a);
1708         err = PTR_ERR(vport);
1709         if (IS_ERR(vport))
1710                 goto exit_unlock;
1711
1712         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1713                                          OVS_VPORT_CMD_NEW);
1714         err = PTR_ERR(reply);
1715         if (IS_ERR(reply))
1716                 goto exit_unlock;
1717
1718         rcu_read_unlock();
1719
1720         return genlmsg_reply(reply, info);
1721
1722 exit_unlock:
1723         rcu_read_unlock();
1724         return err;
1725 }
1726
1727 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1728 {
1729         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1730         struct datapath *dp;
1731         u32 port_no;
1732         int retval;
1733
1734         dp = get_dp(ovs_header->dp_ifindex);
1735         if (!dp)
1736                 return -ENODEV;
1737
1738         rcu_read_lock();
1739         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1740                 struct vport *vport;
1741
1742                 vport = rcu_dereference(dp->ports[port_no]);
1743                 if (!vport)
1744                         continue;
1745
1746                 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1747                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1748                                             OVS_VPORT_CMD_NEW) < 0)
1749                         break;
1750         }
1751         rcu_read_unlock();
1752
1753         cb->args[0] = port_no;
1754         retval = skb->len;
1755
1756         return retval;
1757 }
1758
1759 static void rehash_flow_table(struct work_struct *work)
1760 {
1761         struct datapath *dp;
1762
1763         genl_lock();
1764
1765         list_for_each_entry(dp, &dps, list_node) {
1766                 struct flow_table *old_table = genl_dereference(dp->table);
1767                 struct flow_table *new_table;
1768
1769                 new_table = ovs_flow_tbl_rehash(old_table);
1770                 if (!IS_ERR(new_table)) {
1771                         rcu_assign_pointer(dp->table, new_table);
1772                         ovs_flow_tbl_deferred_destroy(old_table);
1773                 }
1774         }
1775
1776         genl_unlock();
1777
1778         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
1779 }
1780
1781 static struct genl_ops dp_vport_genl_ops[] = {
1782         { .cmd = OVS_VPORT_CMD_NEW,
1783           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1784           .policy = vport_policy,
1785           .doit = ovs_vport_cmd_new
1786         },
1787         { .cmd = OVS_VPORT_CMD_DEL,
1788           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1789           .policy = vport_policy,
1790           .doit = ovs_vport_cmd_del
1791         },
1792         { .cmd = OVS_VPORT_CMD_GET,
1793           .flags = 0,               /* OK for unprivileged users. */
1794           .policy = vport_policy,
1795           .doit = ovs_vport_cmd_get,
1796           .dumpit = ovs_vport_cmd_dump
1797         },
1798         { .cmd = OVS_VPORT_CMD_SET,
1799           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1800           .policy = vport_policy,
1801           .doit = ovs_vport_cmd_set,
1802         },
1803 };
1804
1805 struct genl_family_and_ops {
1806         struct genl_family *family;
1807         struct genl_ops *ops;
1808         int n_ops;
1809         struct genl_multicast_group *group;
1810 };
1811
1812 static const struct genl_family_and_ops dp_genl_families[] = {
1813         { &dp_datapath_genl_family,
1814           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1815           &ovs_dp_datapath_multicast_group },
1816         { &dp_vport_genl_family,
1817           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1818           &ovs_dp_vport_multicast_group },
1819         { &dp_flow_genl_family,
1820           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1821           &ovs_dp_flow_multicast_group },
1822         { &dp_packet_genl_family,
1823           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1824           NULL },
1825 };
1826
1827 static void dp_unregister_genl(int n_families)
1828 {
1829         int i;
1830
1831         for (i = 0; i < n_families; i++)
1832                 genl_unregister_family(dp_genl_families[i].family);
1833 }
1834
1835 static int dp_register_genl(void)
1836 {
1837         int n_registered;
1838         int err;
1839         int i;
1840
1841         n_registered = 0;
1842         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1843                 const struct genl_family_and_ops *f = &dp_genl_families[i];
1844
1845                 err = genl_register_family_with_ops(f->family, f->ops,
1846                                                     f->n_ops);
1847                 if (err)
1848                         goto error;
1849                 n_registered++;
1850
1851                 if (f->group) {
1852                         err = genl_register_mc_group(f->family, f->group);
1853                         if (err)
1854                                 goto error;
1855                 }
1856         }
1857
1858         return 0;
1859
1860 error:
1861         dp_unregister_genl(n_registered);
1862         return err;
1863 }
1864
1865 static int __init dp_init(void)
1866 {
1867         struct sk_buff *dummy_skb;
1868         int err;
1869
1870         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
1871
1872         pr_info("Open vSwitch switching datapath\n");
1873
1874         err = ovs_flow_init();
1875         if (err)
1876                 goto error;
1877
1878         err = ovs_vport_init();
1879         if (err)
1880                 goto error_flow_exit;
1881
1882         err = register_netdevice_notifier(&ovs_dp_device_notifier);
1883         if (err)
1884                 goto error_vport_exit;
1885
1886         err = dp_register_genl();
1887         if (err < 0)
1888                 goto error_unreg_notifier;
1889
1890         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
1891
1892         return 0;
1893
1894 error_unreg_notifier:
1895         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1896 error_vport_exit:
1897         ovs_vport_exit();
1898 error_flow_exit:
1899         ovs_flow_exit();
1900 error:
1901         return err;
1902 }
1903
1904 static void dp_cleanup(void)
1905 {
1906         cancel_delayed_work_sync(&rehash_flow_wq);
1907         rcu_barrier();
1908         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1909         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1910         ovs_vport_exit();
1911         ovs_flow_exit();
1912 }
1913
1914 module_init(dp_init);
1915 module_exit(dp_cleanup);
1916
1917 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1918 MODULE_LICENSE("GPL");