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(&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         int err;
 408
 409         if (vlan_tx_tag_present(skb)) {
 410                 nskb = skb_clone(skb, GFP_ATOMIC);
 411                 if (!nskb)
 412                         return -ENOMEM;
 413
 414                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 415                 if (!nskb)
 416                         return -ENOMEM;
 417
 418                 vlan_set_tci(nskb, 0);
 419
 420                 skb = nskb;
 421         }
 422
 423         if (nla_attr_size(skb->len) > USHRT_MAX) {
 424                 err = -EFBIG;
 425                 goto out;
 426         }
 427
 428         user_skb = genlmsg_new(upcall_msg_size(skb, upcall_info->userdata), GFP_ATOMIC);
 429         if (!user_skb) {
 430                 err = -ENOMEM;
 431                 goto out;
 432         }
 433
 434         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 435                              0, upcall_info->cmd);
 436         upcall->dp_ifindex = dp_ifindex;
 437
 438         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 439         ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
 440         nla_nest_end(user_skb, nla);
 441
 442         if (upcall_info->userdata)
 443                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 444                           nla_len(upcall_info->userdata),
 445                           nla_data(upcall_info->userdata));
 446
 447         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 448
 449         skb_copy_and_csum_dev(skb, nla_data(nla));
 450
 451         genlmsg_end(user_skb, upcall);
 452         err = genlmsg_unicast(net, user_skb, upcall_info->portid);
 453
 454 out:
 455         kfree_skb(nskb);
 456         return err;
 457 }
 458
 459 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 460 {
 461         struct ovs_header *ovs_header = info->userhdr;
 462         struct nlattr **a = info->attrs;
 463         struct sw_flow_actions *acts;
 464         struct sk_buff *packet;
 465         struct sw_flow *flow;
 466         struct datapath *dp;
 467         struct ethhdr *eth;
 468         int len;
 469         int err;
 470
 471         err = -EINVAL;
 472         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 473             !a[OVS_PACKET_ATTR_ACTIONS])
 474                 goto err;
 475
 476         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 477         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 478         err = -ENOMEM;
 479         if (!packet)
 480                 goto err;
 481         skb_reserve(packet, NET_IP_ALIGN);
 482
 483         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 484
 485         skb_reset_mac_header(packet);
 486         eth = eth_hdr(packet);
 487
 488         /* Normally, setting the skb 'protocol' field would be handled by a
 489          * call to eth_type_trans(), but it assumes there's a sending
 490          * device, which we may not have. */
 491         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 492                 packet->protocol = eth->h_proto;
 493         else
 494                 packet->protocol = htons(ETH_P_802_2);
 495
 496         /* Build an sw_flow for sending this packet. */
 497         flow = ovs_flow_alloc();
 498         err = PTR_ERR(flow);
 499         if (IS_ERR(flow))
 500                 goto err_kfree_skb;
 501
 502         err = ovs_flow_extract(packet, -1, &flow->key);
 503         if (err)
 504                 goto err_flow_free;
 505
 506         err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
 507         if (err)
 508                 goto err_flow_free;
 509         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 510         err = PTR_ERR(acts);
 511         if (IS_ERR(acts))
 512                 goto err_flow_free;
 513
 514         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 515                                    &flow->key, 0, &acts);
 516         rcu_assign_pointer(flow->sf_acts, acts);
 517         if (err)
 518                 goto err_flow_free;
 519
 520         OVS_CB(packet)->flow = flow;
 521         OVS_CB(packet)->pkt_key = &flow->key;
 522         packet->priority = flow->key.phy.priority;
 523         packet->mark = flow->key.phy.skb_mark;
 524
 525         rcu_read_lock();
 526         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 527         err = -ENODEV;
 528         if (!dp)
 529                 goto err_unlock;
 530
 531         local_bh_disable();
 532         err = ovs_execute_actions(dp, packet);
 533         local_bh_enable();
 534         rcu_read_unlock();
 535
 536         ovs_flow_free(flow, false);
 537         return err;
 538
 539 err_unlock:
 540         rcu_read_unlock();
 541 err_flow_free:
 542         ovs_flow_free(flow, false);
 543 err_kfree_skb:
 544         kfree_skb(packet);
 545 err:
 546         return err;
 547 }
 548
 549 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 550         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 551         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 552         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 553 };
 554
 555 static struct genl_ops dp_packet_genl_ops[] = {
 556         { .cmd = OVS_PACKET_CMD_EXECUTE,
 557           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 558           .policy = packet_policy,
 559           .doit = ovs_packet_cmd_execute
 560         }
 561 };
 562
 563 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
 564                          struct ovs_dp_megaflow_stats *mega_stats)
 565 {
 566         int i;
 567
 568         memset(mega_stats, 0, sizeof(*mega_stats));
 569
 570         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 571         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 572
 573         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 574
 575         for_each_possible_cpu(i) {
 576                 const struct dp_stats_percpu *percpu_stats;
 577                 struct dp_stats_percpu local_stats;
 578                 unsigned int start;
 579
 580                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 581
 582                 do {
 583                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 584                         local_stats = *percpu_stats;
 585                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 586
 587                 stats->n_hit += local_stats.n_hit;
 588                 stats->n_missed += local_stats.n_missed;
 589                 stats->n_lost += local_stats.n_lost;
 590                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 591         }
 592 }
 593
 594 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 595         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 596         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 597         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 598 };
 599
 600 static struct genl_family dp_flow_genl_family = {
 601         .id = GENL_ID_GENERATE,
 602         .hdrsize = sizeof(struct ovs_header),
 603         .name = OVS_FLOW_FAMILY,
 604         .version = OVS_FLOW_VERSION,
 605         .maxattr = OVS_FLOW_ATTR_MAX,
 606         .netnsok = true,
 607          SET_PARALLEL_OPS
 608 };
 609
 610 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 611         .name = OVS_FLOW_MCGROUP
 612 };
 613
 614 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 615 {
 616         return NLMSG_ALIGN(sizeof(struct ovs_header))
 617                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 618                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 619                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 620                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 621                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 622                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 623 }
 624
 625 /* Called with ovs_mutex. */
 626 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 627                                   struct sk_buff *skb, u32 portid,
 628                                   u32 seq, u32 flags, u8 cmd)
 629 {
 630         const int skb_orig_len = skb->len;
 631         struct nlattr *start;
 632         struct sw_flow_stats flow_stats;
 633         struct ovs_header *ovs_header;
 634         struct nlattr *nla;
 635         int err;
 636
 637         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 638         if (!ovs_header)
 639                 return -EMSGSIZE;
 640
 641         ovs_header->dp_ifindex = get_dpifindex(dp);
 642
 643         /* Fill flow key. */
 644         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 645         if (!nla)
 646                 goto nla_put_failure;
 647
 648         err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
 649         if (err)
 650                 goto error;
 651         nla_nest_end(skb, nla);
 652
 653         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 654         if (!nla)
 655                 goto nla_put_failure;
 656
 657         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 658         if (err)
 659                 goto error;
 660
 661         nla_nest_end(skb, nla);
 662
 663         ovs_flow_stats_get(flow, &flow_stats);
 664         if (flow_stats.used &&
 665             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(flow_stats.used)))
 666                 goto nla_put_failure;
 667
 668         if (flow_stats.packet_count) {
 669                 struct ovs_flow_stats stats = {
 670                         .n_packets = flow_stats.packet_count,
 671                         .n_bytes = flow_stats.byte_count,
 672                 };
 673
 674                 if (nla_put(skb, OVS_FLOW_ATTR_STATS,
 675                             sizeof(struct ovs_flow_stats), &stats))
 676                         goto nla_put_failure;
 677         }
 678
 679         if ((u8)ntohs(flow_stats.tcp_flags) &&
 680             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS,
 681                        (u8)ntohs(flow_stats.tcp_flags)))
 682                 goto nla_put_failure;
 683
 684         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 685          * this is the first flow to be dumped into 'skb'.  This is unusual for
 686          * Netlink but individual action lists can be longer than
 687          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 688          * The userspace caller can always fetch the actions separately if it
 689          * really wants them.  (Most userspace callers in fact don't care.)
 690          *
 691          * This can only fail for dump operations because the skb is always
 692          * properly sized for single flows.
 693          */
 694         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 695         if (start) {
 696                 const struct sw_flow_actions *sf_acts;
 697
 698                 sf_acts = rcu_dereference_check(flow->sf_acts,
 699                                                 lockdep_ovsl_is_held());
 700
 701                 err = ovs_nla_put_actions(sf_acts->actions,
 702                                           sf_acts->actions_len, skb);
 703                 if (!err)
 704                         nla_nest_end(skb, start);
 705                 else {
 706                         if (skb_orig_len)
 707                                 goto error;
 708
 709                         nla_nest_cancel(skb, start);
 710                 }
 711         } else if (skb_orig_len)
 712                 goto nla_put_failure;
 713
 714         return genlmsg_end(skb, ovs_header);
 715
 716 nla_put_failure:
 717         err = -EMSGSIZE;
 718 error:
 719         genlmsg_cancel(skb, ovs_header);
 720         return err;
 721 }
 722
 723 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 724 {
 725         const struct sw_flow_actions *sf_acts;
 726
 727         sf_acts = ovsl_dereference(flow->sf_acts);
 728
 729         return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
 730 }
 731
 732 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 733                                                struct datapath *dp,
 734                                                u32 portid, u32 seq, u8 cmd)
 735 {
 736         struct sk_buff *skb;
 737         int retval;
 738
 739         skb = ovs_flow_cmd_alloc_info(flow);
 740         if (!skb)
 741                 return ERR_PTR(-ENOMEM);
 742
 743         retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
 744         BUG_ON(retval < 0);
 745         return skb;
 746 }
 747
 748 static struct sw_flow *__ovs_flow_tbl_lookup(struct flow_table *tbl,
 749                                               const struct sw_flow_key *key)
 750 {
 751         u32 __always_unused n_mask_hit;
 752
 753         return ovs_flow_tbl_lookup(tbl, key, &n_mask_hit);
 754 }
 755
 756 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 757 {
 758         struct nlattr **a = info->attrs;
 759         struct ovs_header *ovs_header = info->userhdr;
 760         struct sw_flow_key key, masked_key;
 761         struct sw_flow *flow = NULL;
 762         struct sw_flow_mask mask;
 763         struct sk_buff *reply;
 764         struct datapath *dp;
 765         struct sw_flow_actions *acts = NULL;
 766         struct sw_flow_match match;
 767         int error;
 768
 769         /* Extract key. */
 770         error = -EINVAL;
 771         if (!a[OVS_FLOW_ATTR_KEY])
 772                 goto error;
 773
 774         ovs_match_init(&match, &key, &mask);
 775         error = ovs_nla_get_match(&match,
 776                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 777         if (error)
 778                 goto error;
 779
 780         /* Validate actions. */
 781         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 782                 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 783                 error = PTR_ERR(acts);
 784                 if (IS_ERR(acts))
 785                         goto error;
 786
 787                 ovs_flow_mask_key(&masked_key, &key, &mask);
 788                 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
 789                                              &masked_key, 0, &acts);
 790                 if (error) {
 791                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 792                         goto err_kfree;
 793                 }
 794         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 795                 error = -EINVAL;
 796                 goto error;
 797         }
 798
 799         ovs_lock();
 800         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 801         error = -ENODEV;
 802         if (!dp)
 803                 goto err_unlock_ovs;
 804
 805         /* Check if this is a duplicate flow */
 806         flow = __ovs_flow_tbl_lookup(&dp->table, &key);
 807         if (!flow) {
 808                 /* Bail out if we're not allowed to create a new flow. */
 809                 error = -ENOENT;
 810                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
 811                         goto err_unlock_ovs;
 812
 813                 /* Allocate flow. */
 814                 flow = ovs_flow_alloc();
 815                 if (IS_ERR(flow)) {
 816                         error = PTR_ERR(flow);
 817                         goto err_unlock_ovs;
 818                 }
 819
 820                 flow->key = masked_key;
 821                 flow->unmasked_key = key;
 822                 rcu_assign_pointer(flow->sf_acts, acts);
 823
 824                 /* Put flow in bucket. */
 825                 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
 826                 if (error) {
 827                         acts = NULL;
 828                         goto err_flow_free;
 829                 }
 830
 831                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
 832                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
 833         } else {
 834                 /* We found a matching flow. */
 835                 struct sw_flow_actions *old_acts;
 836
 837                 /* Bail out if we're not allowed to modify an existing flow.
 838                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 839                  * because Generic Netlink treats the latter as a dump
 840                  * request.  We also accept NLM_F_EXCL in case that bug ever
 841                  * gets fixed.
 842                  */
 843                 error = -EEXIST;
 844                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
 845                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
 846                         goto err_unlock_ovs;
 847
 848                 /* The unmasked key has to be the same for flow updates. */
 849                 error = -EINVAL;
 850                 if (!ovs_flow_cmp_unmasked_key(flow, &match)) {
 851                         OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
 852                         goto err_unlock_ovs;
 853                 }
 854
 855                 /* Update actions. */
 856                 old_acts = ovsl_dereference(flow->sf_acts);
 857                 rcu_assign_pointer(flow->sf_acts, acts);
 858                 ovs_nla_free_flow_actions(old_acts);
 859
 860                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
 861                                                info->snd_seq, OVS_FLOW_CMD_NEW);
 862
 863                 /* Clear stats. */
 864                 if (a[OVS_FLOW_ATTR_CLEAR])
 865                         ovs_flow_stats_clear(flow);
 866         }
 867         ovs_unlock();
 868
 869         if (!IS_ERR(reply))
 870                 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
 871         else
 872                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
 873                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
 874         return 0;
 875
 876 err_flow_free:
 877         ovs_flow_free(flow, false);
 878 err_unlock_ovs:
 879         ovs_unlock();
 880 err_kfree:
 881         kfree(acts);
 882 error:
 883         return error;
 884 }
 885
 886 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
 887 {
 888         struct nlattr **a = info->attrs;
 889         struct ovs_header *ovs_header = info->userhdr;
 890         struct sw_flow_key key;
 891         struct sk_buff *reply;
 892         struct sw_flow *flow;
 893         struct datapath *dp;
 894         struct sw_flow_match match;
 895         int err;
 896
 897         if (!a[OVS_FLOW_ATTR_KEY]) {
 898                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
 899                 return -EINVAL;
 900         }
 901
 902         ovs_match_init(&match, &key, NULL);
 903         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
 904         if (err)
 905                 return err;
 906
 907         ovs_lock();
 908         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 909         if (!dp) {
 910                 err = -ENODEV;
 911                 goto unlock;
 912         }
 913
 914         flow = __ovs_flow_tbl_lookup(&dp->table, &key);
 915         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
 916                 err = -ENOENT;
 917                 goto unlock;
 918         }
 919
 920         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
 921                                         info->snd_seq, OVS_FLOW_CMD_NEW);
 922         if (IS_ERR(reply)) {
 923                 err = PTR_ERR(reply);
 924                 goto unlock;
 925         }
 926
 927         ovs_unlock();
 928         return genlmsg_reply(reply, info);
 929 unlock:
 930         ovs_unlock();
 931         return err;
 932 }
 933
 934 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
 935 {
 936         struct nlattr **a = info->attrs;
 937         struct ovs_header *ovs_header = info->userhdr;
 938         struct sw_flow_key key;
 939         struct sk_buff *reply;
 940         struct sw_flow *flow;
 941         struct datapath *dp;
 942         struct sw_flow_match match;
 943         int err;
 944
 945         ovs_lock();
 946         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 947         if (!dp) {
 948                 err = -ENODEV;
 949                 goto unlock;
 950         }
 951
 952         if (!a[OVS_FLOW_ATTR_KEY]) {
 953                 err = ovs_flow_tbl_flush(&dp->table);
 954                 goto unlock;
 955         }
 956
 957         ovs_match_init(&match, &key, NULL);
 958         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
 959         if (err)
 960                 goto unlock;
 961
 962         flow = __ovs_flow_tbl_lookup(&dp->table, &key);
 963         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
 964                 err = -ENOENT;
 965                 goto unlock;
 966         }
 967
 968         reply = ovs_flow_cmd_alloc_info(flow);
 969         if (!reply) {
 970                 err = -ENOMEM;
 971                 goto unlock;
 972         }
 973
 974         ovs_flow_tbl_remove(&dp->table, flow);
 975
 976         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
 977                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
 978         BUG_ON(err < 0);
 979
 980         ovs_flow_free(flow, true);
 981         ovs_unlock();
 982
 983         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
 984         return 0;
 985 unlock:
 986         ovs_unlock();
 987         return err;
 988 }
 989
 990 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
 991 {
 992         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
 993         struct table_instance *ti;
 994         struct datapath *dp;
 995
 996         rcu_read_lock();
 997         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 998         if (!dp) {
 999                 rcu_read_unlock();
1000                 return -ENODEV;
1001         }
1002
1003         ti = rcu_dereference(dp->table.ti);
1004         for (;;) {
1005                 struct sw_flow *flow;
1006                 u32 bucket, obj;
1007
1008                 bucket = cb->args[0];
1009                 obj = cb->args[1];
1010                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1011                 if (!flow)
1012                         break;
1013
1014                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1015                                            NETLINK_CB(cb->skb).portid,
1016                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1017                                            OVS_FLOW_CMD_NEW) < 0)
1018                         break;
1019
1020                 cb->args[0] = bucket;
1021                 cb->args[1] = obj;
1022         }
1023         rcu_read_unlock();
1024         return skb->len;
1025 }
1026
1027 static struct genl_ops dp_flow_genl_ops[] = {
1028         { .cmd = OVS_FLOW_CMD_NEW,
1029           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1030           .policy = flow_policy,
1031           .doit = ovs_flow_cmd_new_or_set
1032         },
1033         { .cmd = OVS_FLOW_CMD_DEL,
1034           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1035           .policy = flow_policy,
1036           .doit = ovs_flow_cmd_del
1037         },
1038         { .cmd = OVS_FLOW_CMD_GET,
1039           .flags = 0,               /* OK for unprivileged users. */
1040           .policy = flow_policy,
1041           .doit = ovs_flow_cmd_get,
1042           .dumpit = ovs_flow_cmd_dump
1043         },
1044         { .cmd = OVS_FLOW_CMD_SET,
1045           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1046           .policy = flow_policy,
1047           .doit = ovs_flow_cmd_new_or_set,
1048         },
1049 };
1050
1051 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1052         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1053         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1054 };
1055
1056 static struct genl_family dp_datapath_genl_family = {
1057         .id = GENL_ID_GENERATE,
1058         .hdrsize = sizeof(struct ovs_header),
1059         .name = OVS_DATAPATH_FAMILY,
1060         .version = OVS_DATAPATH_VERSION,
1061         .maxattr = OVS_DP_ATTR_MAX,
1062         .netnsok = true,
1063          SET_PARALLEL_OPS
1064 };
1065
1066 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1067         .name = OVS_DATAPATH_MCGROUP
1068 };
1069
1070 static size_t ovs_dp_cmd_msg_size(void)
1071 {
1072         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1073
1074         msgsize += nla_total_size(IFNAMSIZ);
1075         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1076         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1077
1078         return msgsize;
1079 }
1080
1081 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1082                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1083 {
1084         struct ovs_header *ovs_header;
1085         struct ovs_dp_stats dp_stats;
1086         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1087         int err;
1088
1089         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1090                                    flags, cmd);
1091         if (!ovs_header)
1092                 goto error;
1093
1094         ovs_header->dp_ifindex = get_dpifindex(dp);
1095
1096         rcu_read_lock();
1097         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1098         rcu_read_unlock();
1099         if (err)
1100                 goto nla_put_failure;
1101
1102         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1103         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1104                         &dp_stats))
1105                 goto nla_put_failure;
1106
1107         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1108                         sizeof(struct ovs_dp_megaflow_stats),
1109                         &dp_megaflow_stats))
1110                 goto nla_put_failure;
1111
1112         return genlmsg_end(skb, ovs_header);
1113
1114 nla_put_failure:
1115         genlmsg_cancel(skb, ovs_header);
1116 error:
1117         return -EMSGSIZE;
1118 }
1119
1120 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1121                                              u32 seq, u8 cmd)
1122 {
1123         struct sk_buff *skb;
1124         int retval;
1125
1126         skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1127         if (!skb)
1128                 return ERR_PTR(-ENOMEM);
1129
1130         retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1131         if (retval < 0) {
1132                 kfree_skb(skb);
1133                 return ERR_PTR(retval);
1134         }
1135         return skb;
1136 }
1137
1138 /* Called with ovs_mutex. */
1139 static struct datapath *lookup_datapath(struct net *net,
1140                                         struct ovs_header *ovs_header,
1141                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1142 {
1143         struct datapath *dp;
1144
1145         if (!a[OVS_DP_ATTR_NAME])
1146                 dp = get_dp(net, ovs_header->dp_ifindex);
1147         else {
1148                 struct vport *vport;
1149
1150                 rcu_read_lock();
1151                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1152                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1153                 rcu_read_unlock();
1154         }
1155         return dp ? dp : ERR_PTR(-ENODEV);
1156 }
1157
1158 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1159 {
1160         struct nlattr **a = info->attrs;
1161         struct vport_parms parms;
1162         struct sk_buff *reply;
1163         struct datapath *dp;
1164         struct vport *vport;
1165         struct ovs_net *ovs_net;
1166         int err, i;
1167
1168         err = -EINVAL;
1169         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1170                 goto err;
1171
1172         ovs_lock();
1173
1174         err = -ENOMEM;
1175         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1176         if (dp == NULL)
1177                 goto err_unlock_ovs;
1178
1179         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1180
1181         /* Allocate table. */
1182         err = ovs_flow_tbl_init(&dp->table);
1183         if (err)
1184                 goto err_free_dp;
1185
1186         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1187         if (!dp->stats_percpu) {
1188                 err = -ENOMEM;
1189                 goto err_destroy_table;
1190         }
1191
1192         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1193                             GFP_KERNEL);
1194         if (!dp->ports) {
1195                 err = -ENOMEM;
1196                 goto err_destroy_percpu;
1197         }
1198
1199         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1200                 INIT_HLIST_HEAD(&dp->ports[i]);
1201
1202         /* Set up our datapath device. */
1203         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1204         parms.type = OVS_VPORT_TYPE_INTERNAL;
1205         parms.options = NULL;
1206         parms.dp = dp;
1207         parms.port_no = OVSP_LOCAL;
1208         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1209
1210         vport = new_vport(&parms);
1211         if (IS_ERR(vport)) {
1212                 err = PTR_ERR(vport);
1213                 if (err == -EBUSY)
1214                         err = -EEXIST;
1215
1216                 goto err_destroy_ports_array;
1217         }
1218
1219         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1220                                       info->snd_seq, OVS_DP_CMD_NEW);
1221         err = PTR_ERR(reply);
1222         if (IS_ERR(reply))
1223                 goto err_destroy_local_port;
1224
1225         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1226         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1227
1228         ovs_unlock();
1229
1230         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1231         return 0;
1232
1233 err_destroy_local_port:
1234         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1235 err_destroy_ports_array:
1236         kfree(dp->ports);
1237 err_destroy_percpu:
1238         free_percpu(dp->stats_percpu);
1239 err_destroy_table:
1240         ovs_flow_tbl_destroy(&dp->table);
1241 err_free_dp:
1242         release_net(ovs_dp_get_net(dp));
1243         kfree(dp);
1244 err_unlock_ovs:
1245         ovs_unlock();
1246 err:
1247         return err;
1248 }
1249
1250 /* Called with ovs_mutex. */
1251 static void __dp_destroy(struct datapath *dp)
1252 {
1253         int i;
1254
1255         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1256                 struct vport *vport;
1257                 struct hlist_node *n;
1258
1259                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1260                         if (vport->port_no != OVSP_LOCAL)
1261                                 ovs_dp_detach_port(vport);
1262         }
1263
1264         list_del_rcu(&dp->list_node);
1265
1266         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1267          * all port in datapath are destroyed first before freeing datapath.
1268          */
1269         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1270
1271         call_rcu(&dp->rcu, destroy_dp_rcu);
1272 }
1273
1274 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1275 {
1276         struct sk_buff *reply;
1277         struct datapath *dp;
1278         int err;
1279
1280         ovs_lock();
1281         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1282         err = PTR_ERR(dp);
1283         if (IS_ERR(dp))
1284                 goto unlock;
1285
1286         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1287                                       info->snd_seq, OVS_DP_CMD_DEL);
1288         err = PTR_ERR(reply);
1289         if (IS_ERR(reply))
1290                 goto unlock;
1291
1292         __dp_destroy(dp);
1293         ovs_unlock();
1294
1295         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1296
1297         return 0;
1298 unlock:
1299         ovs_unlock();
1300         return err;
1301 }
1302
1303 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1304 {
1305         struct sk_buff *reply;
1306         struct datapath *dp;
1307         int err;
1308
1309         ovs_lock();
1310         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1311         err = PTR_ERR(dp);
1312         if (IS_ERR(dp))
1313                 goto unlock;
1314
1315         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1316                                       info->snd_seq, OVS_DP_CMD_NEW);
1317         if (IS_ERR(reply)) {
1318                 err = PTR_ERR(reply);
1319                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1320                                 ovs_dp_datapath_multicast_group.id, err);
1321                 err = 0;
1322                 goto unlock;
1323         }
1324
1325         ovs_unlock();
1326         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1327
1328         return 0;
1329 unlock:
1330         ovs_unlock();
1331         return err;
1332 }
1333
1334 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1335 {
1336         struct sk_buff *reply;
1337         struct datapath *dp;
1338         int err;
1339
1340         ovs_lock();
1341         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1342         if (IS_ERR(dp)) {
1343                 err = PTR_ERR(dp);
1344                 goto unlock;
1345         }
1346
1347         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1348                                       info->snd_seq, OVS_DP_CMD_NEW);
1349         if (IS_ERR(reply)) {
1350                 err = PTR_ERR(reply);
1351                 goto unlock;
1352         }
1353
1354         ovs_unlock();
1355         return genlmsg_reply(reply, info);
1356
1357 unlock:
1358         ovs_unlock();
1359         return err;
1360 }
1361
1362 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1363 {
1364         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1365         struct datapath *dp;
1366         int skip = cb->args[0];
1367         int i = 0;
1368
1369         rcu_read_lock();
1370         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1371                 if (i >= skip &&
1372                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1373                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1374                                          OVS_DP_CMD_NEW) < 0)
1375                         break;
1376                 i++;
1377         }
1378         rcu_read_unlock();
1379
1380         cb->args[0] = i;
1381
1382         return skb->len;
1383 }
1384
1385 static struct genl_ops dp_datapath_genl_ops[] = {
1386         { .cmd = OVS_DP_CMD_NEW,
1387           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1388           .policy = datapath_policy,
1389           .doit = ovs_dp_cmd_new
1390         },
1391         { .cmd = OVS_DP_CMD_DEL,
1392           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1393           .policy = datapath_policy,
1394           .doit = ovs_dp_cmd_del
1395         },
1396         { .cmd = OVS_DP_CMD_GET,
1397           .flags = 0,               /* OK for unprivileged users. */
1398           .policy = datapath_policy,
1399           .doit = ovs_dp_cmd_get,
1400           .dumpit = ovs_dp_cmd_dump
1401         },
1402         { .cmd = OVS_DP_CMD_SET,
1403           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1404           .policy = datapath_policy,
1405           .doit = ovs_dp_cmd_set,
1406         },
1407 };
1408
1409 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1410         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1411         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1412         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1413         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1414         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1415         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1416 };
1417
1418 static struct genl_family dp_vport_genl_family = {
1419         .id = GENL_ID_GENERATE,
1420         .hdrsize = sizeof(struct ovs_header),
1421         .name = OVS_VPORT_FAMILY,
1422         .version = OVS_VPORT_VERSION,
1423         .maxattr = OVS_VPORT_ATTR_MAX,
1424         .netnsok = true,
1425          SET_PARALLEL_OPS
1426 };
1427
1428 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1429         .name = OVS_VPORT_MCGROUP
1430 };
1431
1432 /* Called with ovs_mutex or RCU read lock. */
1433 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1434                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1435 {
1436         struct ovs_header *ovs_header;
1437         struct ovs_vport_stats vport_stats;
1438         int err;
1439
1440         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1441                                  flags, cmd);
1442         if (!ovs_header)
1443                 return -EMSGSIZE;
1444
1445         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1446
1447         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1448             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1449             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1450             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1451                 goto nla_put_failure;
1452
1453         ovs_vport_get_stats(vport, &vport_stats);
1454         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1455                     &vport_stats))
1456                 goto nla_put_failure;
1457
1458         err = ovs_vport_get_options(vport, skb);
1459         if (err == -EMSGSIZE)
1460                 goto error;
1461
1462         return genlmsg_end(skb, ovs_header);
1463
1464 nla_put_failure:
1465         err = -EMSGSIZE;
1466 error:
1467         genlmsg_cancel(skb, ovs_header);
1468         return err;
1469 }
1470
1471 /* Called with ovs_mutex or RCU read lock. */
1472 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1473                                          u32 seq, u8 cmd)
1474 {
1475         struct sk_buff *skb;
1476         int retval;
1477
1478         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1479         if (!skb)
1480                 return ERR_PTR(-ENOMEM);
1481
1482         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1483         BUG_ON(retval < 0);
1484
1485         return skb;
1486 }
1487
1488 /* Called with ovs_mutex or RCU read lock. */
1489 static struct vport *lookup_vport(struct net *net,
1490                                   struct ovs_header *ovs_header,
1491                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1492 {
1493         struct datapath *dp;
1494         struct vport *vport;
1495
1496         if (a[OVS_VPORT_ATTR_NAME]) {
1497                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1498                 if (!vport)
1499                         return ERR_PTR(-ENODEV);
1500                 if (ovs_header->dp_ifindex &&
1501                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1502                         return ERR_PTR(-ENODEV);
1503                 return vport;
1504         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1505                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1506
1507                 if (port_no >= DP_MAX_PORTS)
1508                         return ERR_PTR(-EFBIG);
1509
1510                 dp = get_dp(net, ovs_header->dp_ifindex);
1511                 if (!dp)
1512                         return ERR_PTR(-ENODEV);
1513
1514                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1515                 if (!vport)
1516                         return ERR_PTR(-ENODEV);
1517                 return vport;
1518         } else
1519                 return ERR_PTR(-EINVAL);
1520 }
1521
1522 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1523 {
1524         struct nlattr **a = info->attrs;
1525         struct ovs_header *ovs_header = info->userhdr;
1526         struct vport_parms parms;
1527         struct sk_buff *reply;
1528         struct vport *vport;
1529         struct datapath *dp;
1530         u32 port_no;
1531         int err;
1532
1533         err = -EINVAL;
1534         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1535             !a[OVS_VPORT_ATTR_UPCALL_PID])
1536                 goto exit;
1537
1538         ovs_lock();
1539         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1540         err = -ENODEV;
1541         if (!dp)
1542                 goto exit_unlock;
1543
1544         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1545                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1546
1547                 err = -EFBIG;
1548                 if (port_no >= DP_MAX_PORTS)
1549                         goto exit_unlock;
1550
1551                 vport = ovs_vport_ovsl(dp, port_no);
1552                 err = -EBUSY;
1553                 if (vport)
1554                         goto exit_unlock;
1555         } else {
1556                 for (port_no = 1; ; port_no++) {
1557                         if (port_no >= DP_MAX_PORTS) {
1558                                 err = -EFBIG;
1559                                 goto exit_unlock;
1560                         }
1561                         vport = ovs_vport_ovsl(dp, port_no);
1562                         if (!vport)
1563                                 break;
1564                 }
1565         }
1566
1567         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1568         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1569         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1570         parms.dp = dp;
1571         parms.port_no = port_no;
1572         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1573
1574         vport = new_vport(&parms);
1575         err = PTR_ERR(vport);
1576         if (IS_ERR(vport))
1577                 goto exit_unlock;
1578
1579         err = 0;
1580         if (a[OVS_VPORT_ATTR_STATS])
1581                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1582
1583         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1584                                          OVS_VPORT_CMD_NEW);
1585         if (IS_ERR(reply)) {
1586                 err = PTR_ERR(reply);
1587                 ovs_dp_detach_port(vport);
1588                 goto exit_unlock;
1589         }
1590
1591         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1592
1593 exit_unlock:
1594         ovs_unlock();
1595 exit:
1596         return err;
1597 }
1598
1599 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1600 {
1601         struct nlattr **a = info->attrs;
1602         struct sk_buff *reply;
1603         struct vport *vport;
1604         int err;
1605
1606         ovs_lock();
1607         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1608         err = PTR_ERR(vport);
1609         if (IS_ERR(vport))
1610                 goto exit_unlock;
1611
1612         if (a[OVS_VPORT_ATTR_TYPE] &&
1613             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1614                 err = -EINVAL;
1615                 goto exit_unlock;
1616         }
1617
1618         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1619         if (!reply) {
1620                 err = -ENOMEM;
1621                 goto exit_unlock;
1622         }
1623
1624         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1625                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1626                 if (err)
1627                         goto exit_free;
1628         }
1629
1630         if (a[OVS_VPORT_ATTR_STATS])
1631                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1632
1633         if (a[OVS_VPORT_ATTR_UPCALL_PID])
1634                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1635
1636         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1637                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1638         BUG_ON(err < 0);
1639
1640         ovs_unlock();
1641         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1642         return 0;
1643
1644 exit_free:
1645         kfree_skb(reply);
1646 exit_unlock:
1647         ovs_unlock();
1648         return err;
1649 }
1650
1651 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1652 {
1653         struct nlattr **a = info->attrs;
1654         struct sk_buff *reply;
1655         struct vport *vport;
1656         int err;
1657
1658         ovs_lock();
1659         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1660         err = PTR_ERR(vport);
1661         if (IS_ERR(vport))
1662                 goto exit_unlock;
1663
1664         if (vport->port_no == OVSP_LOCAL) {
1665                 err = -EINVAL;
1666                 goto exit_unlock;
1667         }
1668
1669         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1670                                          info->snd_seq, OVS_VPORT_CMD_DEL);
1671         err = PTR_ERR(reply);
1672         if (IS_ERR(reply))
1673                 goto exit_unlock;
1674
1675         err = 0;
1676         ovs_dp_detach_port(vport);
1677
1678         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1679
1680 exit_unlock:
1681         ovs_unlock();
1682         return err;
1683 }
1684
1685 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1686 {
1687         struct nlattr **a = info->attrs;
1688         struct ovs_header *ovs_header = info->userhdr;
1689         struct sk_buff *reply;
1690         struct vport *vport;
1691         int err;
1692
1693         rcu_read_lock();
1694         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1695         err = PTR_ERR(vport);
1696         if (IS_ERR(vport))
1697                 goto exit_unlock;
1698
1699         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1700                                          info->snd_seq, OVS_VPORT_CMD_NEW);
1701         err = PTR_ERR(reply);
1702         if (IS_ERR(reply))
1703                 goto exit_unlock;
1704
1705         rcu_read_unlock();
1706
1707         return genlmsg_reply(reply, info);
1708
1709 exit_unlock:
1710         rcu_read_unlock();
1711         return err;
1712 }
1713
1714 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1715 {
1716         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1717         struct datapath *dp;
1718         int bucket = cb->args[0], skip = cb->args[1];
1719         int i, j = 0;
1720
1721         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1722         if (!dp)
1723                 return -ENODEV;
1724
1725         rcu_read_lock();
1726         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1727                 struct vport *vport;
1728
1729                 j = 0;
1730                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1731                         if (j >= skip &&
1732                             ovs_vport_cmd_fill_info(vport, skb,
1733                                                     NETLINK_CB(cb->skb).portid,
1734                                                     cb->nlh->nlmsg_seq,
1735                                                     NLM_F_MULTI,
1736                                                     OVS_VPORT_CMD_NEW) < 0)
1737                                 goto out;
1738
1739                         j++;
1740                 }
1741                 skip = 0;
1742         }
1743 out:
1744         rcu_read_unlock();
1745
1746         cb->args[0] = i;
1747         cb->args[1] = j;
1748
1749         return skb->len;
1750 }
1751
1752 static struct genl_ops dp_vport_genl_ops[] = {
1753         { .cmd = OVS_VPORT_CMD_NEW,
1754           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1755           .policy = vport_policy,
1756           .doit = ovs_vport_cmd_new
1757         },
1758         { .cmd = OVS_VPORT_CMD_DEL,
1759           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1760           .policy = vport_policy,
1761           .doit = ovs_vport_cmd_del
1762         },
1763         { .cmd = OVS_VPORT_CMD_GET,
1764           .flags = 0,               /* OK for unprivileged users. */
1765           .policy = vport_policy,
1766           .doit = ovs_vport_cmd_get,
1767           .dumpit = ovs_vport_cmd_dump
1768         },
1769         { .cmd = OVS_VPORT_CMD_SET,
1770           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1771           .policy = vport_policy,
1772           .doit = ovs_vport_cmd_set,
1773         },
1774 };
1775
1776 struct genl_family_and_ops {
1777         struct genl_family *family;
1778         struct genl_ops *ops;
1779         int n_ops;
1780         struct genl_multicast_group *group;
1781 };
1782
1783 static const struct genl_family_and_ops dp_genl_families[] = {
1784         { &dp_datapath_genl_family,
1785           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1786           &ovs_dp_datapath_multicast_group },
1787         { &dp_vport_genl_family,
1788           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1789           &ovs_dp_vport_multicast_group },
1790         { &dp_flow_genl_family,
1791           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1792           &ovs_dp_flow_multicast_group },
1793         { &dp_packet_genl_family,
1794           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1795           NULL },
1796 };
1797
1798 static void dp_unregister_genl(int n_families)
1799 {
1800         int i;
1801
1802         for (i = 0; i < n_families; i++)
1803                 genl_unregister_family(dp_genl_families[i].family);
1804 }
1805
1806 static int dp_register_genl(void)
1807 {
1808         int n_registered;
1809         int err;
1810         int i;
1811
1812         n_registered = 0;
1813         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1814                 const struct genl_family_and_ops *f = &dp_genl_families[i];
1815
1816                 err = genl_register_family_with_ops(f->family, f->ops,
1817                                                     f->n_ops);
1818                 if (err)
1819                         goto error;
1820                 n_registered++;
1821
1822                 if (f->group) {
1823                         err = genl_register_mc_group(f->family, f->group);
1824                         if (err)
1825                                 goto error;
1826                 }
1827         }
1828
1829         return 0;
1830
1831 error:
1832         dp_unregister_genl(n_registered);
1833         return err;
1834 }
1835
1836 static int __net_init ovs_init_net(struct net *net)
1837 {
1838         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1839
1840         INIT_LIST_HEAD(&ovs_net->dps);
1841         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1842         return 0;
1843 }
1844
1845 static void __net_exit ovs_exit_net(struct net *net)
1846 {
1847         struct datapath *dp, *dp_next;
1848         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1849
1850         ovs_lock();
1851         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1852                 __dp_destroy(dp);
1853         ovs_unlock();
1854
1855         cancel_work_sync(&ovs_net->dp_notify_work);
1856 }
1857
1858 static struct pernet_operations ovs_net_ops = {
1859         .init = ovs_init_net,
1860         .exit = ovs_exit_net,
1861         .id   = &ovs_net_id,
1862         .size = sizeof(struct ovs_net),
1863 };
1864
1865 DEFINE_COMPAT_PNET_REG_FUNC(device);
1866
1867 static int __init dp_init(void)
1868 {
1869         int err;
1870
1871         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1872
1873         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
1874                 VERSION);
1875
1876         err = ovs_flow_init();
1877         if (err)
1878                 goto error;
1879
1880         err = ovs_vport_init();
1881         if (err)
1882                 goto error_flow_exit;
1883
1884         err = register_pernet_device(&ovs_net_ops);
1885         if (err)
1886                 goto error_vport_exit;
1887
1888         err = register_netdevice_notifier(&ovs_dp_device_notifier);
1889         if (err)
1890                 goto error_netns_exit;
1891
1892         err = dp_register_genl();
1893         if (err < 0)
1894                 goto error_unreg_notifier;
1895
1896         return 0;
1897
1898 error_unreg_notifier:
1899         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1900 error_netns_exit:
1901         unregister_pernet_device(&ovs_net_ops);
1902 error_vport_exit:
1903         ovs_vport_exit();
1904 error_flow_exit:
1905         ovs_flow_exit();
1906 error:
1907         return err;
1908 }
1909
1910 static void dp_cleanup(void)
1911 {
1912         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1913         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1914         unregister_pernet_device(&ovs_net_ops);
1915         rcu_barrier();
1916         ovs_vport_exit();
1917         ovs_flow_exit();
1918 }
1919
1920 module_init(dp_init);
1921 module_exit(dp_cleanup);
1922
1923 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1924 MODULE_LICENSE("GPL");
1925 MODULE_VERSION(VERSION);