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