datapath/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/version.h>
  40 #include <linux/ethtool.h>
  41 #include <linux/wait.h>
  42 #include <asm/div64.h>
  43 #include <linux/highmem.h>
  44 #include <linux/netfilter_bridge.h>
  45 #include <linux/netfilter_ipv4.h>
  46 #include <linux/inetdevice.h>
  47 #include <linux/list.h>
  48 #include <linux/openvswitch.h>
  49 #include <linux/rculist.h>
  50 #include <linux/dmi.h>
  51 #include <linux/genetlink.h>
  52 #include <net/genetlink.h>
  53 #include <net/genetlink.h>
  54 #include <net/net_namespace.h>
  55 #include <net/netns/generic.h>
  56
  57 #include "datapath.h"
  58 #include "flow.h"
  59 #include "flow_netlink.h"
  60 #include "vlan.h"
  61 #include "vport-internal_dev.h"
  62 #include "vport-netdev.h"
  63
  64 int ovs_net_id __read_mostly;
  65
  66 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
  67                        struct genl_multicast_group *grp)
  68 {
  69         genl_notify(skb, genl_info_net(info), info->snd_portid,
  70                     grp->id, info->nlhdr, GFP_KERNEL);
  71 }
  72
  73 /**
  74  * DOC: Locking:
  75  *
  76  * All writes e.g. Writes to device state (add/remove datapath, port, set
  77  * operations on vports, etc.), Writes to other state (flow table
  78  * modifications, set miscellaneous datapath parameters, etc.) are protected
  79  * by ovs_lock.
  80  *
  81  * Reads are protected by RCU.
  82  *
  83  * There are a few special cases (mostly stats) that have their own
  84  * synchronization but they nest under all of above and don't interact with
  85  * each other.
  86  *
  87  * The RTNL lock nests inside ovs_mutex.
  88  */
  89
  90 static DEFINE_MUTEX(ovs_mutex);
  91
  92 void ovs_lock(void)
  93 {
  94         mutex_lock(&ovs_mutex);
  95 }
  96
  97 void ovs_unlock(void)
  98 {
  99         mutex_unlock(&ovs_mutex);
 100 }
 101
 102 #ifdef CONFIG_LOCKDEP
 103 int lockdep_ovsl_is_held(void)
 104 {
 105         if (debug_locks)
 106                 return lockdep_is_held(&ovs_mutex);
 107         else
 108                 return 1;
 109 }
 110 #endif
 111
 112 static struct vport *new_vport(const struct vport_parms *);
 113 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
 114                              const struct dp_upcall_info *);
 115 static int queue_userspace_packet(struct net *, int dp_ifindex,
 116                                   struct sk_buff *,
 117                                   const struct dp_upcall_info *);
 118
 119 /* Must be called with rcu_read_lock or ovs_mutex. */
 120 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 121 {
 122         struct datapath *dp = NULL;
 123         struct net_device *dev;
 124
 125         rcu_read_lock();
 126         dev = dev_get_by_index_rcu(net, dp_ifindex);
 127         if (dev) {
 128                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 129                 if (vport)
 130                         dp = vport->dp;
 131         }
 132         rcu_read_unlock();
 133
 134         return dp;
 135 }
 136
 137 /* Must be called with rcu_read_lock or ovs_mutex. */
 138 const char *ovs_dp_name(const struct datapath *dp)
 139 {
 140         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 141         return vport->ops->get_name(vport);
 142 }
 143
 144 static int get_dpifindex(struct datapath *dp)
 145 {
 146         struct vport *local;
 147         int ifindex;
 148
 149         rcu_read_lock();
 150
 151         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 152         if (local)
 153                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 154         else
 155                 ifindex = 0;
 156
 157         rcu_read_unlock();
 158
 159         return ifindex;
 160 }
 161
 162 static void destroy_dp_rcu(struct rcu_head *rcu)
 163 {
 164         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 165
 166         ovs_flow_tbl_destroy(&dp->table);
 167         free_percpu(dp->stats_percpu);
 168         release_net(ovs_dp_get_net(dp));
 169         kfree(dp->ports);
 170         kfree(dp);
 171 }
 172
 173 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 174                                             u16 port_no)
 175 {
 176         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 177 }
 178
 179 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 180 {
 181         struct vport *vport;
 182         struct hlist_head *head;
 183
 184         head = vport_hash_bucket(dp, port_no);
 185         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 186                 if (vport->port_no == port_no)
 187                         return vport;
 188         }
 189         return NULL;
 190 }
 191
 192 /* Called with ovs_mutex. */
 193 static struct vport *new_vport(const struct vport_parms *parms)
 194 {
 195         struct vport *vport;
 196
 197         vport = ovs_vport_add(parms);
 198         if (!IS_ERR(vport)) {
 199                 struct datapath *dp = parms->dp;
 200                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 201
 202                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 203         }
 204         return vport;
 205 }
 206
 207 void ovs_dp_detach_port(struct vport *p)
 208 {
 209         ASSERT_OVSL();
 210
 211         /* First drop references to device. */
 212         hlist_del_rcu(&p->dp_hash_node);
 213
 214         /* Then destroy it. */
 215         ovs_vport_del(p);
 216 }
 217
 218 /* Must be called with rcu_read_lock. */
 219 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 220 {
 221         struct datapath *dp = p->dp;
 222         struct sw_flow *flow;
 223         struct dp_stats_percpu *stats;
 224         struct sw_flow_key key;
 225         u64 *stats_counter;
 226         u32 n_mask_hit;
 227         int error;
 228
 229         stats = this_cpu_ptr(dp->stats_percpu);
 230
 231         /* Extract flow from 'skb' into 'key'. */
 232         error = ovs_flow_extract(skb, p->port_no, &key);
 233         if (unlikely(error)) {
 234                 kfree_skb(skb);
 235                 return;
 236         }
 237
 238         /* Look up flow. */
 239         flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
 240         if (unlikely(!flow)) {
 241                 struct dp_upcall_info upcall;
 242
 243                 upcall.cmd = OVS_PACKET_CMD_MISS;
 244                 upcall.key = &key;
 245                 upcall.userdata = NULL;
 246                 upcall.portid = p->upcall_portid;
 247                 ovs_dp_upcall(dp, skb, &upcall);
 248                 consume_skb(skb);
 249                 stats_counter = &stats->n_missed;
 250                 goto out;
 251         }
 252
 253         OVS_CB(skb)->flow = flow;
 254         OVS_CB(skb)->pkt_key = &key;
 255
 256         ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
 257         ovs_execute_actions(dp, skb);
 258         stats_counter = &stats->n_hit;
 259
 260 out:
 261         /* Update datapath statistics. */
 262         u64_stats_update_begin(&stats->sync);
 263         (*stats_counter)++;
 264         stats->n_mask_hit += n_mask_hit;
 265         u64_stats_update_end(&stats->sync);
 266 }
 267
 268 static struct genl_family dp_packet_genl_family = {
 269         .id = GENL_ID_GENERATE,
 270         .hdrsize = sizeof(struct ovs_header),
 271         .name = OVS_PACKET_FAMILY,
 272         .version = OVS_PACKET_VERSION,
 273         .maxattr = OVS_PACKET_ATTR_MAX,
 274         .netnsok = true,
 275          SET_PARALLEL_OPS
 276 };
 277
 278 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 279                   const struct dp_upcall_info *upcall_info)
 280 {
 281         struct dp_stats_percpu *stats;
 282         int dp_ifindex;
 283         int err;
 284
 285         if (upcall_info->portid == 0) {
 286                 err = -ENOTCONN;
 287                 goto err;
 288         }
 289
 290         dp_ifindex = get_dpifindex(dp);
 291         if (!dp_ifindex) {
 292                 err = -ENODEV;
 293                 goto err;
 294         }
 295
 296         if (!skb_is_gso(skb))
 297                 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 298         else
 299                 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 300         if (err)
 301                 goto err;
 302
 303         return 0;
 304
 305 err:
 306         stats = this_cpu_ptr(dp->stats_percpu);
 307
 308         u64_stats_update_begin(&stats->sync);
 309         stats->n_lost++;
 310         u64_stats_update_end(&stats->sync);
 311
 312         return err;
 313 }
 314
 315 static int queue_gso_packets(struct net *net, int dp_ifindex,
 316                              struct sk_buff *skb,
 317                              const struct dp_upcall_info *upcall_info)
 318 {
 319         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 320         struct dp_upcall_info later_info;
 321         struct sw_flow_key later_key;
 322         struct sk_buff *segs, *nskb;
 323         int err;
 324
 325         segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
 326         if (IS_ERR(segs))
 327                 return PTR_ERR(segs);
 328
 329         /* Queue all of the segments. */
 330         skb = segs;
 331         do {
 332                 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
 333                 if (err)
 334                         break;
 335
 336                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 337                         /* The initial flow key extracted by ovs_flow_extract()
 338                          * in this case is for a first fragment, so we need to
 339                          * properly mark later fragments.
 340                          */
 341                         later_key = *upcall_info->key;
 342                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 343
 344                         later_info = *upcall_info;
 345                         later_info.key = &later_key;
 346                         upcall_info = &later_info;
 347                 }
 348         } while ((skb = skb->next));
 349
 350         /* Free all of the segments. */
 351         skb = segs;
 352         do {
 353                 nskb = skb->next;
 354                 if (err)
 355                         kfree_skb(skb);
 356                 else
 357                         consume_skb(skb);
 358         } while ((skb = nskb));
 359         return err;
 360 }
 361
 362 static size_t key_attr_size(void)
 363 {
 364         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 365                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 366                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
 367                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 368                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 369                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
 370                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
 371                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 372                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
 373                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 374                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 375                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 376                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 377                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 378                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 379                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 380                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 381                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 382                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 383 }
 384
 385 static size_t upcall_msg_size(const struct sk_buff *skb,
 386                               const struct nlattr *userdata)
 387 {
 388         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 389                 + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
 390                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 391
 392         /* OVS_PACKET_ATTR_USERDATA */
 393         if (userdata)
 394                 size += NLA_ALIGN(userdata->nla_len);
 395
 396         return size;
 397 }
 398
 399 static int queue_userspace_packet(struct net *net, int dp_ifindex,
 400                                   struct sk_buff *skb,
 401                                   const struct dp_upcall_info *upcall_info)
 402 {
 403         struct ovs_header *upcall;
 404         struct sk_buff *nskb = NULL;
 405         struct sk_buff *user_skb; /* to be queued to userspace */
 406         struct nlattr *nla;
 407         struct genl_info info = {
 408 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,14,0)
 409                 .dst_sk = net->genl_sock,
 410 #endif
 411                 .snd_portid = upcall_info->portid,
 412         };
 413         size_t len;
 414         int err;
 415
 416         if (vlan_tx_tag_present(skb)) {
 417                 nskb = skb_clone(skb, GFP_ATOMIC);
 418                 if (!nskb)
 419                         return -ENOMEM;
 420
 421                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 422                 if (!nskb)
 423                         return -ENOMEM;
 424
 425                 vlan_set_tci(nskb, 0);
 426
 427                 skb = nskb;
 428         }
 429
 430         if (nla_attr_size(skb->len) > USHRT_MAX) {
 431                 err = -EFBIG;
 432                 goto out;
 433         }
 434
 435         len = upcall_msg_size(skb, upcall_info->userdata);
 436         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 437         if (!user_skb) {
 438                 err = -ENOMEM;
 439                 goto out;
 440         }
 441
 442         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 443                              0, upcall_info->cmd);
 444         upcall->dp_ifindex = dp_ifindex;
 445
 446         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 447         ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
 448         nla_nest_end(user_skb, nla);
 449
 450         if (upcall_info->userdata)
 451                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 452                           nla_len(upcall_info->userdata),
 453                           nla_data(upcall_info->userdata));
 454
 455         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 456
 457         skb_copy_and_csum_dev(skb, nla_data(nla));
 458
 459         genlmsg_end(user_skb, upcall);
 460         err = genlmsg_unicast(net, user_skb, upcall_info->portid);
 461
 462 out:
 463         kfree_skb(nskb);
 464         return err;
 465 }
 466
 467 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 468 {
 469         struct ovs_header *ovs_header = info->userhdr;
 470         struct nlattr **a = info->attrs;
 471         struct sw_flow_actions *acts;
 472         struct sk_buff *packet;
 473         struct sw_flow *flow;
 474         struct datapath *dp;
 475         struct ethhdr *eth;
 476         int len;
 477         int err;
 478
 479         err = -EINVAL;
 480         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 481             !a[OVS_PACKET_ATTR_ACTIONS])
 482                 goto err;
 483
 484         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 485         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 486         err = -ENOMEM;
 487         if (!packet)
 488                 goto err;
 489         skb_reserve(packet, NET_IP_ALIGN);
 490
 491         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 492
 493         skb_reset_mac_header(packet);
 494         eth = eth_hdr(packet);
 495
 496         /* Normally, setting the skb 'protocol' field would be handled by a
 497          * call to eth_type_trans(), but it assumes there's a sending
 498          * device, which we may not have. */
 499         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 500                 packet->protocol = eth->h_proto;
 501         else
 502                 packet->protocol = htons(ETH_P_802_2);
 503
 504         /* Build an sw_flow for sending this packet. */
 505         flow = ovs_flow_alloc(false);
 506         err = PTR_ERR(flow);
 507         if (IS_ERR(flow))
 508                 goto err_kfree_skb;
 509
 510         err = ovs_flow_extract(packet, -1, &flow->key);
 511         if (err)
 512                 goto err_flow_free;
 513
 514         err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
 515         if (err)
 516                 goto err_flow_free;
 517         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 518         err = PTR_ERR(acts);
 519         if (IS_ERR(acts))
 520                 goto err_flow_free;
 521
 522         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 523                                    &flow->key, 0, &acts);
 524         rcu_assign_pointer(flow->sf_acts, acts);
 525         if (err)
 526                 goto err_flow_free;
 527
 528         OVS_CB(packet)->flow = flow;
 529         OVS_CB(packet)->pkt_key = &flow->key;
 530         packet->priority = flow->key.phy.priority;
 531         packet->mark = flow->key.phy.skb_mark;
 532
 533         rcu_read_lock();
 534         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 535         err = -ENODEV;
 536         if (!dp)
 537                 goto err_unlock;
 538
 539         local_bh_disable();
 540         err = ovs_execute_actions(dp, packet);
 541         local_bh_enable();
 542         rcu_read_unlock();
 543
 544         ovs_flow_free(flow, false);
 545         return err;
 546
 547 err_unlock:
 548         rcu_read_unlock();
 549 err_flow_free:
 550         ovs_flow_free(flow, false);
 551 err_kfree_skb:
 552         kfree_skb(packet);
 553 err:
 554         return err;
 555 }
 556
 557 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 558         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 559         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 560         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 561 };
 562
 563 static struct genl_ops dp_packet_genl_ops[] = {
 564         { .cmd = OVS_PACKET_CMD_EXECUTE,
 565           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 566           .policy = packet_policy,
 567           .doit = ovs_packet_cmd_execute
 568         }
 569 };
 570
 571 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
 572                          struct ovs_dp_megaflow_stats *mega_stats)
 573 {
 574         int i;
 575
 576         memset(mega_stats, 0, sizeof(*mega_stats));
 577
 578         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 579         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 580
 581         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 582
 583         for_each_possible_cpu(i) {
 584                 const struct dp_stats_percpu *percpu_stats;
 585                 struct dp_stats_percpu local_stats;
 586                 unsigned int start;
 587
 588                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 589
 590                 do {
 591                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 592                         local_stats = *percpu_stats;
 593                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 594
 595                 stats->n_hit += local_stats.n_hit;
 596                 stats->n_missed += local_stats.n_missed;
 597                 stats->n_lost += local_stats.n_lost;
 598                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 599         }
 600 }
 601
 602 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 603         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 604         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 605         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 606 };
 607
 608 static struct genl_family dp_flow_genl_family = {
 609         .id = GENL_ID_GENERATE,
 610         .hdrsize = sizeof(struct ovs_header),
 611         .name = OVS_FLOW_FAMILY,
 612         .version = OVS_FLOW_VERSION,
 613         .maxattr = OVS_FLOW_ATTR_MAX,
 614         .netnsok = true,
 615          SET_PARALLEL_OPS
 616 };
 617
 618 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 619         .name = OVS_FLOW_MCGROUP
 620 };
 621
 622 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 623 {
 624         return NLMSG_ALIGN(sizeof(struct ovs_header))
 625                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 626                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 627                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 628                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 629                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 630                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 631 }
 632
 633 /* Called with ovs_mutex. */
 634 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 635                                   struct sk_buff *skb, u32 portid,
 636                                   u32 seq, u32 flags, u8 cmd)
 637 {
 638         const int skb_orig_len = skb->len;
 639         struct nlattr *start;
 640         struct ovs_flow_stats stats;
 641         __be16 tcp_flags;
 642         unsigned long used;
 643         struct ovs_header *ovs_header;
 644         struct nlattr *nla;
 645         int err;
 646
 647         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 648         if (!ovs_header)
 649                 return -EMSGSIZE;
 650
 651         ovs_header->dp_ifindex = get_dpifindex(dp);
 652
 653         /* Fill flow key. */
 654         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 655         if (!nla)
 656                 goto nla_put_failure;
 657
 658         err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
 659         if (err)
 660                 goto error;
 661         nla_nest_end(skb, nla);
 662
 663         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 664         if (!nla)
 665                 goto nla_put_failure;
 666
 667         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 668         if (err)
 669                 goto error;
 670
 671         nla_nest_end(skb, nla);
 672
 673         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 674         if (used &&
 675             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 676                 goto nla_put_failure;
 677
 678         if (stats.n_packets &&
 679             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 680                 goto nla_put_failure;
 681
 682         if ((u8)ntohs(tcp_flags) &&
 683              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 684                 goto nla_put_failure;
 685
 686         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 687          * this is the first flow to be dumped into 'skb'.  This is unusual for
 688          * Netlink but individual action lists can be longer than
 689          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 690          * The userspace caller can always fetch the actions separately if it
 691          * really wants them.  (Most userspace callers in fact don't care.)
 692          *
 693          * This can only fail for dump operations because the skb is always
 694          * properly sized for single flows.
 695          */
 696         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 697         if (start) {
 698                 const struct sw_flow_actions *sf_acts;
 699
 700                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 701
 702                 err = ovs_nla_put_actions(sf_acts->actions,
 703                                           sf_acts->actions_len, skb);
 704                 if (!err)
 705                         nla_nest_end(skb, start);
 706                 else {
 707                         if (skb_orig_len)
 708                                 goto error;
 709
 710                         nla_nest_cancel(skb, start);
 711                 }
 712         } else if (skb_orig_len)
 713                 goto nla_put_failure;
 714
 715         return genlmsg_end(skb, ovs_header);
 716
 717 nla_put_failure:
 718         err = -EMSGSIZE;
 719 error:
 720         genlmsg_cancel(skb, ovs_header);
 721         return err;
 722 }
 723
 724 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow,
 725                                                struct genl_info *info)
 726 {
 727         size_t len;
 728
 729         len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts));
 730
 731         return genlmsg_new_unicast(len, info, GFP_KERNEL);
 732 }
 733
 734 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 735                                                struct datapath *dp,
 736                                                struct genl_info *info,
 737                                                u8 cmd)
 738 {
 739         struct sk_buff *skb;
 740         int retval;
 741
 742         skb = ovs_flow_cmd_alloc_info(flow, info);
 743         if (!skb)
 744                 return ERR_PTR(-ENOMEM);
 745
 746         retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid,
 747                                         info->snd_seq, 0, cmd);
 748         BUG_ON(retval < 0);
 749         return skb;
 750 }
 751
 752 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 753 {
 754         struct nlattr **a = info->attrs;
 755         struct ovs_header *ovs_header = info->userhdr;
 756         struct sw_flow_key key, masked_key;
 757         struct sw_flow *flow = NULL;
 758         struct sw_flow_mask mask;
 759         struct sk_buff *reply;
 760         struct datapath *dp;
 761         struct sw_flow_actions *acts = NULL;
 762         struct sw_flow_match match;
 763         bool exact_5tuple;
 764         int error;
 765
 766         /* Extract key. */
 767         error = -EINVAL;
 768         if (!a[OVS_FLOW_ATTR_KEY])
 769                 goto error;
 770
 771         ovs_match_init(&match, &key, &mask);
 772         error = ovs_nla_get_match(&match, &exact_5tuple,
 773                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 774         if (error)
 775                 goto error;
 776
 777         /* Validate actions. */
 778         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 779                 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 780                 error = PTR_ERR(acts);
 781                 if (IS_ERR(acts))
 782                         goto error;
 783
 784                 ovs_flow_mask_key(&masked_key, &key, &mask);
 785                 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
 786                                              &masked_key, 0, &acts);
 787                 if (error) {
 788                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 789                         goto err_kfree;
 790                 }
 791         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 792                 error = -EINVAL;
 793                 goto error;
 794         }
 795
 796         ovs_lock();
 797         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 798         error = -ENODEV;
 799         if (!dp)
 800                 goto err_unlock_ovs;
 801
 802         /* Check if this is a duplicate flow */
 803         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 804         if (!flow) {
 805                 /* Bail out if we're not allowed to create a new flow. */
 806                 error = -ENOENT;
 807                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
 808                         goto err_unlock_ovs;
 809
 810                 /* Allocate flow. */
 811                 flow = ovs_flow_alloc(!exact_5tuple);
 812                 if (IS_ERR(flow)) {
 813                         error = PTR_ERR(flow);
 814                         goto err_unlock_ovs;
 815                 }
 816
 817                 flow->key = masked_key;
 818                 flow->unmasked_key = key;
 819                 rcu_assign_pointer(flow->sf_acts, acts);
 820
 821                 /* Put flow in bucket. */
 822                 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
 823                 if (error) {
 824                         acts = NULL;
 825                         goto err_flow_free;
 826                 }
 827
 828                 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
 829         } else {
 830                 /* We found a matching flow. */
 831                 struct sw_flow_actions *old_acts;
 832
 833                 /* Bail out if we're not allowed to modify an existing flow.
 834                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 835                  * because Generic Netlink treats the latter as a dump
 836                  * request.  We also accept NLM_F_EXCL in case that bug ever
 837                  * gets fixed.
 838                  */
 839                 error = -EEXIST;
 840                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
 841                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
 842                         goto err_unlock_ovs;
 843
 844                 /* The unmasked key has to be the same for flow updates. */
 845                 error = -EINVAL;
 846                 if (!ovs_flow_cmp_unmasked_key(flow, &match)) {
 847                         OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
 848                         goto err_unlock_ovs;
 849                 }
 850
 851                 /* Update actions. */
 852                 old_acts = ovsl_dereference(flow->sf_acts);
 853                 rcu_assign_pointer(flow->sf_acts, acts);
 854                 ovs_nla_free_flow_actions(old_acts);
 855
 856                 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
 857
 858                 /* Clear stats. */
 859                 if (a[OVS_FLOW_ATTR_CLEAR])
 860                         ovs_flow_stats_clear(flow);
 861         }
 862         ovs_unlock();
 863
 864         if (!IS_ERR(reply))
 865                 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
 866         else
 867                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
 868                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
 869         return 0;
 870
 871 err_flow_free:
 872         ovs_flow_free(flow, false);
 873 err_unlock_ovs:
 874         ovs_unlock();
 875 err_kfree:
 876         kfree(acts);
 877 error:
 878         return error;
 879 }
 880
 881 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
 882 {
 883         struct nlattr **a = info->attrs;
 884         struct ovs_header *ovs_header = info->userhdr;
 885         struct sw_flow_key key;
 886         struct sk_buff *reply;
 887         struct sw_flow *flow;
 888         struct datapath *dp;
 889         struct sw_flow_match match;
 890         int err;
 891
 892         if (!a[OVS_FLOW_ATTR_KEY]) {
 893                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
 894                 return -EINVAL;
 895         }
 896
 897         ovs_match_init(&match, &key, NULL);
 898         err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
 899         if (err)
 900                 return err;
 901
 902         ovs_lock();
 903         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 904         if (!dp) {
 905                 err = -ENODEV;
 906                 goto unlock;
 907         }
 908
 909         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 910         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
 911                 err = -ENOENT;
 912                 goto unlock;
 913         }
 914
 915         reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
 916         if (IS_ERR(reply)) {
 917                 err = PTR_ERR(reply);
 918                 goto unlock;
 919         }
 920
 921         ovs_unlock();
 922         return genlmsg_reply(reply, info);
 923 unlock:
 924         ovs_unlock();
 925         return err;
 926 }
 927
 928 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
 929 {
 930         struct nlattr **a = info->attrs;
 931         struct ovs_header *ovs_header = info->userhdr;
 932         struct sw_flow_key key;
 933         struct sk_buff *reply;
 934         struct sw_flow *flow;
 935         struct datapath *dp;
 936         struct sw_flow_match match;
 937         int err;
 938
 939         ovs_lock();
 940         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 941         if (!dp) {
 942                 err = -ENODEV;
 943                 goto unlock;
 944         }
 945
 946         if (!a[OVS_FLOW_ATTR_KEY]) {
 947                 err = ovs_flow_tbl_flush(&dp->table);
 948                 goto unlock;
 949         }
 950
 951         ovs_match_init(&match, &key, NULL);
 952         err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
 953         if (err)
 954                 goto unlock;
 955
 956         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 957         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
 958                 err = -ENOENT;
 959                 goto unlock;
 960         }
 961
 962         reply = ovs_flow_cmd_alloc_info(flow, info);
 963         if (!reply) {
 964                 err = -ENOMEM;
 965                 goto unlock;
 966         }
 967
 968         ovs_flow_tbl_remove(&dp->table, flow);
 969
 970         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
 971                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
 972         BUG_ON(err < 0);
 973
 974         ovs_flow_free(flow, true);
 975         ovs_unlock();
 976
 977         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
 978         return 0;
 979 unlock:
 980         ovs_unlock();
 981         return err;
 982 }
 983
 984 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
 985 {
 986         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
 987         struct table_instance *ti;
 988         struct datapath *dp;
 989
 990         rcu_read_lock();
 991         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 992         if (!dp) {
 993                 rcu_read_unlock();
 994                 return -ENODEV;
 995         }
 996
 997         ti = rcu_dereference(dp->table.ti);
 998         for (;;) {
 999                 struct sw_flow *flow;
1000                 u32 bucket, obj;
1001
1002                 bucket = cb->args[0];
1003                 obj = cb->args[1];
1004                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1005                 if (!flow)
1006                         break;
1007
1008                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1009                                            NETLINK_CB(cb->skb).portid,
1010                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1011                                            OVS_FLOW_CMD_NEW) < 0)
1012                         break;
1013
1014                 cb->args[0] = bucket;
1015                 cb->args[1] = obj;
1016         }
1017         rcu_read_unlock();
1018         return skb->len;
1019 }
1020
1021 static struct genl_ops dp_flow_genl_ops[] = {
1022         { .cmd = OVS_FLOW_CMD_NEW,
1023           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1024           .policy = flow_policy,
1025           .doit = ovs_flow_cmd_new_or_set
1026         },
1027         { .cmd = OVS_FLOW_CMD_DEL,
1028           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1029           .policy = flow_policy,
1030           .doit = ovs_flow_cmd_del
1031         },
1032         { .cmd = OVS_FLOW_CMD_GET,
1033           .flags = 0,               /* OK for unprivileged users. */
1034           .policy = flow_policy,
1035           .doit = ovs_flow_cmd_get,
1036           .dumpit = ovs_flow_cmd_dump
1037         },
1038         { .cmd = OVS_FLOW_CMD_SET,
1039           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1040           .policy = flow_policy,
1041           .doit = ovs_flow_cmd_new_or_set,
1042         },
1043 };
1044
1045 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1046         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1047         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1048         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1049 };
1050
1051 static struct genl_family dp_datapath_genl_family = {
1052         .id = GENL_ID_GENERATE,
1053         .hdrsize = sizeof(struct ovs_header),
1054         .name = OVS_DATAPATH_FAMILY,
1055         .version = OVS_DATAPATH_VERSION,
1056         .maxattr = OVS_DP_ATTR_MAX,
1057         .netnsok = true,
1058          SET_PARALLEL_OPS
1059 };
1060
1061 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1062         .name = OVS_DATAPATH_MCGROUP
1063 };
1064
1065 static size_t ovs_dp_cmd_msg_size(void)
1066 {
1067         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1068
1069         msgsize += nla_total_size(IFNAMSIZ);
1070         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1071         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1072
1073         return msgsize;
1074 }
1075
1076 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1077                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1078 {
1079         struct ovs_header *ovs_header;
1080         struct ovs_dp_stats dp_stats;
1081         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1082         int err;
1083
1084         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1085                                    flags, cmd);
1086         if (!ovs_header)
1087                 goto error;
1088
1089         ovs_header->dp_ifindex = get_dpifindex(dp);
1090
1091         rcu_read_lock();
1092         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1093         rcu_read_unlock();
1094         if (err)
1095                 goto nla_put_failure;
1096
1097         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1098         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1099                         &dp_stats))
1100                 goto nla_put_failure;
1101
1102         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1103                         sizeof(struct ovs_dp_megaflow_stats),
1104                         &dp_megaflow_stats))
1105                 goto nla_put_failure;
1106
1107         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1108                 goto nla_put_failure;
1109
1110         return genlmsg_end(skb, ovs_header);
1111
1112 nla_put_failure:
1113         genlmsg_cancel(skb, ovs_header);
1114 error:
1115         return -EMSGSIZE;
1116 }
1117
1118 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp,
1119                                              struct genl_info *info, u8 cmd)
1120 {
1121         struct sk_buff *skb;
1122         int retval;
1123
1124         skb = genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1125         if (!skb)
1126                 return ERR_PTR(-ENOMEM);
1127
1128         retval = ovs_dp_cmd_fill_info(dp, skb, info->snd_portid, info->snd_seq, 0, cmd);
1129         if (retval < 0) {
1130                 kfree_skb(skb);
1131                 return ERR_PTR(retval);
1132         }
1133         return skb;
1134 }
1135
1136 /* Called with ovs_mutex. */
1137 static struct datapath *lookup_datapath(struct net *net,
1138                                         struct ovs_header *ovs_header,
1139                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1140 {
1141         struct datapath *dp;
1142
1143         if (!a[OVS_DP_ATTR_NAME])
1144                 dp = get_dp(net, ovs_header->dp_ifindex);
1145         else {
1146                 struct vport *vport;
1147
1148                 rcu_read_lock();
1149                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1150                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1151                 rcu_read_unlock();
1152         }
1153         return dp ? dp : ERR_PTR(-ENODEV);
1154 }
1155
1156 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1157 {
1158         struct datapath *dp;
1159
1160         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1161         if (!dp)
1162                 return;
1163
1164         WARN(dp->user_features, "Dropping previously announced user features\n");
1165         dp->user_features = 0;
1166 }
1167
1168 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1169 {
1170         if (a[OVS_DP_ATTR_USER_FEATURES])
1171                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1172 }
1173
1174 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1175 {
1176         struct nlattr **a = info->attrs;
1177         struct vport_parms parms;
1178         struct sk_buff *reply;
1179         struct datapath *dp;
1180         struct vport *vport;
1181         struct ovs_net *ovs_net;
1182         int err, i;
1183
1184         err = -EINVAL;
1185         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1186                 goto err;
1187
1188         ovs_lock();
1189
1190         err = -ENOMEM;
1191         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1192         if (dp == NULL)
1193                 goto err_unlock_ovs;
1194
1195         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1196
1197         /* Allocate table. */
1198         err = ovs_flow_tbl_init(&dp->table);
1199         if (err)
1200                 goto err_free_dp;
1201
1202         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1203         if (!dp->stats_percpu) {
1204                 err = -ENOMEM;
1205                 goto err_destroy_table;
1206         }
1207
1208         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1209                             GFP_KERNEL);
1210         if (!dp->ports) {
1211                 err = -ENOMEM;
1212                 goto err_destroy_percpu;
1213         }
1214
1215         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1216                 INIT_HLIST_HEAD(&dp->ports[i]);
1217
1218         /* Set up our datapath device. */
1219         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1220         parms.type = OVS_VPORT_TYPE_INTERNAL;
1221         parms.options = NULL;
1222         parms.dp = dp;
1223         parms.port_no = OVSP_LOCAL;
1224         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1225
1226         ovs_dp_change(dp, a);
1227
1228         vport = new_vport(&parms);
1229         if (IS_ERR(vport)) {
1230                 err = PTR_ERR(vport);
1231                 if (err == -EBUSY)
1232                         err = -EEXIST;
1233
1234                 if (err == -EEXIST) {
1235                         /* An outdated user space instance that does not understand
1236                          * the concept of user_features has attempted to create a new
1237                          * datapath and is likely to reuse it. Drop all user features.
1238                          */
1239                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1240                                 ovs_dp_reset_user_features(skb, info);
1241                 }
1242
1243                 goto err_destroy_ports_array;
1244         }
1245
1246         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1247         err = PTR_ERR(reply);
1248         if (IS_ERR(reply))
1249                 goto err_destroy_local_port;
1250
1251         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1252         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1253
1254         ovs_unlock();
1255
1256         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1257         return 0;
1258
1259 err_destroy_local_port:
1260         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1261 err_destroy_ports_array:
1262         kfree(dp->ports);
1263 err_destroy_percpu:
1264         free_percpu(dp->stats_percpu);
1265 err_destroy_table:
1266         ovs_flow_tbl_destroy(&dp->table);
1267 err_free_dp:
1268         release_net(ovs_dp_get_net(dp));
1269         kfree(dp);
1270 err_unlock_ovs:
1271         ovs_unlock();
1272 err:
1273         return err;
1274 }
1275
1276 /* Called with ovs_mutex. */
1277 static void __dp_destroy(struct datapath *dp)
1278 {
1279         int i;
1280
1281         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1282                 struct vport *vport;
1283                 struct hlist_node *n;
1284
1285                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1286                         if (vport->port_no != OVSP_LOCAL)
1287                                 ovs_dp_detach_port(vport);
1288         }
1289
1290         list_del_rcu(&dp->list_node);
1291
1292         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1293          * all port in datapath are destroyed first before freeing datapath.
1294          */
1295         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1296
1297         call_rcu(&dp->rcu, destroy_dp_rcu);
1298 }
1299
1300 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1301 {
1302         struct sk_buff *reply;
1303         struct datapath *dp;
1304         int err;
1305
1306         ovs_lock();
1307         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1308         err = PTR_ERR(dp);
1309         if (IS_ERR(dp))
1310                 goto unlock;
1311
1312         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_DEL);
1313         err = PTR_ERR(reply);
1314         if (IS_ERR(reply))
1315                 goto unlock;
1316
1317         __dp_destroy(dp);
1318         ovs_unlock();
1319
1320         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1321
1322         return 0;
1323 unlock:
1324         ovs_unlock();
1325         return err;
1326 }
1327
1328 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1329 {
1330         struct sk_buff *reply;
1331         struct datapath *dp;
1332         int err;
1333
1334         ovs_lock();
1335         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1336         err = PTR_ERR(dp);
1337         if (IS_ERR(dp))
1338                 goto unlock;
1339
1340         ovs_dp_change(dp, info->attrs);
1341
1342         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1343         if (IS_ERR(reply)) {
1344                 err = PTR_ERR(reply);
1345                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1346                                 ovs_dp_datapath_multicast_group.id, err);
1347                 err = 0;
1348                 goto unlock;
1349         }
1350
1351         ovs_unlock();
1352         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1353
1354         return 0;
1355 unlock:
1356         ovs_unlock();
1357         return err;
1358 }
1359
1360 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1361 {
1362         struct sk_buff *reply;
1363         struct datapath *dp;
1364         int err;
1365
1366         ovs_lock();
1367         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1368         if (IS_ERR(dp)) {
1369                 err = PTR_ERR(dp);
1370                 goto unlock;
1371         }
1372
1373         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1374         if (IS_ERR(reply)) {
1375                 err = PTR_ERR(reply);
1376                 goto unlock;
1377         }
1378
1379         ovs_unlock();
1380         return genlmsg_reply(reply, info);
1381
1382 unlock:
1383         ovs_unlock();
1384         return err;
1385 }
1386
1387 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1388 {
1389         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1390         struct datapath *dp;
1391         int skip = cb->args[0];
1392         int i = 0;
1393
1394         rcu_read_lock();
1395         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1396                 if (i >= skip &&
1397                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1398                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1399                                          OVS_DP_CMD_NEW) < 0)
1400                         break;
1401                 i++;
1402         }
1403         rcu_read_unlock();
1404
1405         cb->args[0] = i;
1406
1407         return skb->len;
1408 }
1409
1410 static struct genl_ops dp_datapath_genl_ops[] = {
1411         { .cmd = OVS_DP_CMD_NEW,
1412           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1413           .policy = datapath_policy,
1414           .doit = ovs_dp_cmd_new
1415         },
1416         { .cmd = OVS_DP_CMD_DEL,
1417           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1418           .policy = datapath_policy,
1419           .doit = ovs_dp_cmd_del
1420         },
1421         { .cmd = OVS_DP_CMD_GET,
1422           .flags = 0,               /* OK for unprivileged users. */
1423           .policy = datapath_policy,
1424           .doit = ovs_dp_cmd_get,
1425           .dumpit = ovs_dp_cmd_dump
1426         },
1427         { .cmd = OVS_DP_CMD_SET,
1428           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1429           .policy = datapath_policy,
1430           .doit = ovs_dp_cmd_set,
1431         },
1432 };
1433
1434 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1435         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1436         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1437         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1438         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1439         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1440         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1441 };
1442
1443 static struct genl_family dp_vport_genl_family = {
1444         .id = GENL_ID_GENERATE,
1445         .hdrsize = sizeof(struct ovs_header),
1446         .name = OVS_VPORT_FAMILY,
1447         .version = OVS_VPORT_VERSION,
1448         .maxattr = OVS_VPORT_ATTR_MAX,
1449         .netnsok = true,
1450          SET_PARALLEL_OPS
1451 };
1452
1453 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1454         .name = OVS_VPORT_MCGROUP
1455 };
1456
1457 /* Called with ovs_mutex or RCU read lock. */
1458 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1459                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1460 {
1461         struct ovs_header *ovs_header;
1462         struct ovs_vport_stats vport_stats;
1463         int err;
1464
1465         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1466                                  flags, cmd);
1467         if (!ovs_header)
1468                 return -EMSGSIZE;
1469
1470         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1471
1472         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1473             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1474             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1475             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1476                 goto nla_put_failure;
1477
1478         ovs_vport_get_stats(vport, &vport_stats);
1479         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1480                     &vport_stats))
1481                 goto nla_put_failure;
1482
1483         err = ovs_vport_get_options(vport, skb);
1484         if (err == -EMSGSIZE)
1485                 goto error;
1486
1487         return genlmsg_end(skb, ovs_header);
1488
1489 nla_put_failure:
1490         err = -EMSGSIZE;
1491 error:
1492         genlmsg_cancel(skb, ovs_header);
1493         return err;
1494 }
1495
1496 /* Called with ovs_mutex or RCU read lock. */
1497 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1498                                          u32 seq, u8 cmd)
1499 {
1500         struct sk_buff *skb;
1501         int retval;
1502
1503         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1504         if (!skb)
1505                 return ERR_PTR(-ENOMEM);
1506
1507         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1508         BUG_ON(retval < 0);
1509
1510         return skb;
1511 }
1512
1513 /* Called with ovs_mutex or RCU read lock. */
1514 static struct vport *lookup_vport(struct net *net,
1515                                   struct ovs_header *ovs_header,
1516                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1517 {
1518         struct datapath *dp;
1519         struct vport *vport;
1520
1521         if (a[OVS_VPORT_ATTR_NAME]) {
1522                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1523                 if (!vport)
1524                         return ERR_PTR(-ENODEV);
1525                 if (ovs_header->dp_ifindex &&
1526                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1527                         return ERR_PTR(-ENODEV);
1528                 return vport;
1529         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1530                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1531
1532                 if (port_no >= DP_MAX_PORTS)
1533                         return ERR_PTR(-EFBIG);
1534
1535                 dp = get_dp(net, ovs_header->dp_ifindex);
1536                 if (!dp)
1537                         return ERR_PTR(-ENODEV);
1538
1539                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1540                 if (!vport)
1541                         return ERR_PTR(-ENODEV);
1542                 return vport;
1543         } else
1544                 return ERR_PTR(-EINVAL);
1545 }
1546
1547 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1548 {
1549         struct nlattr **a = info->attrs;
1550         struct ovs_header *ovs_header = info->userhdr;
1551         struct vport_parms parms;
1552         struct sk_buff *reply;
1553         struct vport *vport;
1554         struct datapath *dp;
1555         u32 port_no;
1556         int err;
1557
1558         err = -EINVAL;
1559         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1560             !a[OVS_VPORT_ATTR_UPCALL_PID])
1561                 goto exit;
1562
1563         ovs_lock();
1564         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1565         err = -ENODEV;
1566         if (!dp)
1567                 goto exit_unlock;
1568
1569         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1570                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1571
1572                 err = -EFBIG;
1573                 if (port_no >= DP_MAX_PORTS)
1574                         goto exit_unlock;
1575
1576                 vport = ovs_vport_ovsl(dp, port_no);
1577                 err = -EBUSY;
1578                 if (vport)
1579                         goto exit_unlock;
1580         } else {
1581                 for (port_no = 1; ; port_no++) {
1582                         if (port_no >= DP_MAX_PORTS) {
1583                                 err = -EFBIG;
1584                                 goto exit_unlock;
1585                         }
1586                         vport = ovs_vport_ovsl(dp, port_no);
1587                         if (!vport)
1588                                 break;
1589                 }
1590         }
1591
1592         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1593         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1594         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1595         parms.dp = dp;
1596         parms.port_no = port_no;
1597         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1598
1599         vport = new_vport(&parms);
1600         err = PTR_ERR(vport);
1601         if (IS_ERR(vport))
1602                 goto exit_unlock;
1603
1604         err = 0;
1605         if (a[OVS_VPORT_ATTR_STATS])
1606                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1607
1608         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, 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
1616         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1617
1618 exit_unlock:
1619         ovs_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         ovs_lock();
1632         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1633         err = PTR_ERR(vport);
1634         if (IS_ERR(vport))
1635                 goto exit_unlock;
1636
1637         if (a[OVS_VPORT_ATTR_TYPE] &&
1638             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1639                 err = -EINVAL;
1640                 goto exit_unlock;
1641         }
1642
1643         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1644         if (!reply) {
1645                 err = -ENOMEM;
1646                 goto exit_unlock;
1647         }
1648
1649         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1650                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1651                 if (err)
1652                         goto exit_free;
1653         }
1654
1655         if (a[OVS_VPORT_ATTR_STATS])
1656                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1657
1658         if (a[OVS_VPORT_ATTR_UPCALL_PID])
1659                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1660
1661         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1662                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1663         BUG_ON(err < 0);
1664
1665         ovs_unlock();
1666         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1667         return 0;
1668
1669 exit_free:
1670         kfree_skb(reply);
1671 exit_unlock:
1672         ovs_unlock();
1673         return err;
1674 }
1675
1676 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1677 {
1678         struct nlattr **a = info->attrs;
1679         struct sk_buff *reply;
1680         struct vport *vport;
1681         int err;
1682
1683         ovs_lock();
1684         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1685         err = PTR_ERR(vport);
1686         if (IS_ERR(vport))
1687                 goto exit_unlock;
1688
1689         if (vport->port_no == OVSP_LOCAL) {
1690                 err = -EINVAL;
1691                 goto exit_unlock;
1692         }
1693
1694         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1695                                          info->snd_seq, OVS_VPORT_CMD_DEL);
1696         err = PTR_ERR(reply);
1697         if (IS_ERR(reply))
1698                 goto exit_unlock;
1699
1700         err = 0;
1701         ovs_dp_detach_port(vport);
1702
1703         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1704
1705 exit_unlock:
1706         ovs_unlock();
1707         return err;
1708 }
1709
1710 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1711 {
1712         struct nlattr **a = info->attrs;
1713         struct ovs_header *ovs_header = info->userhdr;
1714         struct sk_buff *reply;
1715         struct vport *vport;
1716         int err;
1717
1718         rcu_read_lock();
1719         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1720         err = PTR_ERR(vport);
1721         if (IS_ERR(vport))
1722                 goto exit_unlock;
1723
1724         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1725                                          info->snd_seq, OVS_VPORT_CMD_NEW);
1726         err = PTR_ERR(reply);
1727         if (IS_ERR(reply))
1728                 goto exit_unlock;
1729
1730         rcu_read_unlock();
1731
1732         return genlmsg_reply(reply, info);
1733
1734 exit_unlock:
1735         rcu_read_unlock();
1736         return err;
1737 }
1738
1739 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1740 {
1741         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1742         struct datapath *dp;
1743         int bucket = cb->args[0], skip = cb->args[1];
1744         int i, j = 0;
1745
1746         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1747         if (!dp)
1748                 return -ENODEV;
1749
1750         rcu_read_lock();
1751         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1752                 struct vport *vport;
1753
1754                 j = 0;
1755                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1756                         if (j >= skip &&
1757                             ovs_vport_cmd_fill_info(vport, skb,
1758                                                     NETLINK_CB(cb->skb).portid,
1759                                                     cb->nlh->nlmsg_seq,
1760                                                     NLM_F_MULTI,
1761                                                     OVS_VPORT_CMD_NEW) < 0)
1762                                 goto out;
1763
1764                         j++;
1765                 }
1766                 skip = 0;
1767         }
1768 out:
1769         rcu_read_unlock();
1770
1771         cb->args[0] = i;
1772         cb->args[1] = j;
1773
1774         return skb->len;
1775 }
1776
1777 static struct genl_ops dp_vport_genl_ops[] = {
1778         { .cmd = OVS_VPORT_CMD_NEW,
1779           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1780           .policy = vport_policy,
1781           .doit = ovs_vport_cmd_new
1782         },
1783         { .cmd = OVS_VPORT_CMD_DEL,
1784           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1785           .policy = vport_policy,
1786           .doit = ovs_vport_cmd_del
1787         },
1788         { .cmd = OVS_VPORT_CMD_GET,
1789           .flags = 0,               /* OK for unprivileged users. */
1790           .policy = vport_policy,
1791           .doit = ovs_vport_cmd_get,
1792           .dumpit = ovs_vport_cmd_dump
1793         },
1794         { .cmd = OVS_VPORT_CMD_SET,
1795           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1796           .policy = vport_policy,
1797           .doit = ovs_vport_cmd_set,
1798         },
1799 };
1800
1801 struct genl_family_and_ops {
1802         struct genl_family *family;
1803         struct genl_ops *ops;
1804         int n_ops;
1805         struct genl_multicast_group *group;
1806 };
1807
1808 static const struct genl_family_and_ops dp_genl_families[] = {
1809         { &dp_datapath_genl_family,
1810           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1811           &ovs_dp_datapath_multicast_group },
1812         { &dp_vport_genl_family,
1813           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1814           &ovs_dp_vport_multicast_group },
1815         { &dp_flow_genl_family,
1816           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1817           &ovs_dp_flow_multicast_group },
1818         { &dp_packet_genl_family,
1819           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1820           NULL },
1821 };
1822
1823 static void dp_unregister_genl(int n_families)
1824 {
1825         int i;
1826
1827         for (i = 0; i < n_families; i++)
1828                 genl_unregister_family(dp_genl_families[i].family);
1829 }
1830
1831 static int dp_register_genl(void)
1832 {
1833         int n_registered;
1834         int err;
1835         int i;
1836
1837         n_registered = 0;
1838         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1839                 const struct genl_family_and_ops *f = &dp_genl_families[i];
1840
1841                 err = genl_register_family_with_ops(f->family, f->ops,
1842                                                     f->n_ops);
1843                 if (err)
1844                         goto error;
1845                 n_registered++;
1846
1847                 if (f->group) {
1848                         err = genl_register_mc_group(f->family, f->group);
1849                         if (err)
1850                                 goto error;
1851                 }
1852         }
1853
1854         return 0;
1855
1856 error:
1857         dp_unregister_genl(n_registered);
1858         return err;
1859 }
1860
1861 static int __net_init ovs_init_net(struct net *net)
1862 {
1863         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1864
1865         INIT_LIST_HEAD(&ovs_net->dps);
1866         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1867         return 0;
1868 }
1869
1870 static void __net_exit ovs_exit_net(struct net *net)
1871 {
1872         struct datapath *dp, *dp_next;
1873         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1874
1875         ovs_lock();
1876         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1877                 __dp_destroy(dp);
1878         ovs_unlock();
1879
1880         cancel_work_sync(&ovs_net->dp_notify_work);
1881 }
1882
1883 static struct pernet_operations ovs_net_ops = {
1884         .init = ovs_init_net,
1885         .exit = ovs_exit_net,
1886         .id   = &ovs_net_id,
1887         .size = sizeof(struct ovs_net),
1888 };
1889
1890 DEFINE_COMPAT_PNET_REG_FUNC(device);
1891
1892 static int __init dp_init(void)
1893 {
1894         int err;
1895
1896         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1897
1898         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
1899                 VERSION);
1900
1901         err = ovs_flow_init();
1902         if (err)
1903                 goto error;
1904
1905         err = ovs_vport_init();
1906         if (err)
1907                 goto error_flow_exit;
1908
1909         err = register_pernet_device(&ovs_net_ops);
1910         if (err)
1911                 goto error_vport_exit;
1912
1913         err = register_netdevice_notifier(&ovs_dp_device_notifier);
1914         if (err)
1915                 goto error_netns_exit;
1916
1917         err = dp_register_genl();
1918         if (err < 0)
1919                 goto error_unreg_notifier;
1920
1921         return 0;
1922
1923 error_unreg_notifier:
1924         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1925 error_netns_exit:
1926         unregister_pernet_device(&ovs_net_ops);
1927 error_vport_exit:
1928         ovs_vport_exit();
1929 error_flow_exit:
1930         ovs_flow_exit();
1931 error:
1932         return err;
1933 }
1934
1935 static void dp_cleanup(void)
1936 {
1937         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1938         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1939         unregister_pernet_device(&ovs_net_ops);
1940         rcu_barrier();
1941         ovs_vport_exit();
1942         ovs_flow_exit();
1943 }
1944
1945 module_init(dp_init);
1946 module_exit(dp_cleanup);
1947
1948 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1949 MODULE_LICENSE("GPL");
1950 MODULE_VERSION(VERSION);