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 (flow_stats.tcp_flags &&
 680             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, flow_stats.tcp_flags))
 681                 goto nla_put_failure;
 682
 683         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 684          * this is the first flow to be dumped into 'skb'.  This is unusual for
 685          * Netlink but individual action lists can be longer than
 686          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 687          * The userspace caller can always fetch the actions separately if it
 688          * really wants them.  (Most userspace callers in fact don't care.)
 689          *
 690          * This can only fail for dump operations because the skb is always
 691          * properly sized for single flows.
 692          */
 693         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 694         if (start) {
 695                 const struct sw_flow_actions *sf_acts;
 696
 697                 sf_acts = rcu_dereference_check(flow->sf_acts,
 698                                                 lockdep_ovsl_is_held());
 699
 700                 err = ovs_nla_put_actions(sf_acts->actions,
 701                                           sf_acts->actions_len, skb);
 702                 if (!err)
 703                         nla_nest_end(skb, start);
 704                 else {
 705                         if (skb_orig_len)
 706                                 goto error;
 707
 708                         nla_nest_cancel(skb, start);
 709                 }
 710         } else if (skb_orig_len)
 711                 goto nla_put_failure;
 712
 713         return genlmsg_end(skb, ovs_header);
 714
 715 nla_put_failure:
 716         err = -EMSGSIZE;
 717 error:
 718         genlmsg_cancel(skb, ovs_header);
 719         return err;
 720 }
 721
 722 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 723 {
 724         const struct sw_flow_actions *sf_acts;
 725
 726         sf_acts = ovsl_dereference(flow->sf_acts);
 727
 728         return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
 729 }
 730
 731 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 732                                                struct datapath *dp,
 733                                                u32 portid, u32 seq, u8 cmd)
 734 {
 735         struct sk_buff *skb;
 736         int retval;
 737
 738         skb = ovs_flow_cmd_alloc_info(flow);
 739         if (!skb)
 740                 return ERR_PTR(-ENOMEM);
 741
 742         retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
 743         BUG_ON(retval < 0);
 744         return skb;
 745 }
 746
 747 static struct sw_flow *__ovs_flow_tbl_lookup(struct flow_table *tbl,
 748                                               const struct sw_flow_key *key)
 749 {
 750         u32 __always_unused n_mask_hit;
 751
 752         return ovs_flow_tbl_lookup(tbl, key, &n_mask_hit);
 753 }
 754
 755 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 756 {
 757         struct nlattr **a = info->attrs;
 758         struct ovs_header *ovs_header = info->userhdr;
 759         struct sw_flow_key key, masked_key;
 760         struct sw_flow *flow = NULL;
 761         struct sw_flow_mask mask;
 762         struct sk_buff *reply;
 763         struct datapath *dp;
 764         struct sw_flow_actions *acts = NULL;
 765         struct sw_flow_match match;
 766         int error;
 767
 768         /* Extract key. */
 769         error = -EINVAL;
 770         if (!a[OVS_FLOW_ATTR_KEY])
 771                 goto error;
 772
 773         ovs_match_init(&match, &key, &mask);
 774         error = ovs_nla_get_match(&match,
 775                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 776         if (error)
 777                 goto error;
 778
 779         /* Validate actions. */
 780         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 781                 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 782                 error = PTR_ERR(acts);
 783                 if (IS_ERR(acts))
 784                         goto error;
 785
 786                 ovs_flow_mask_key(&masked_key, &key, &mask);
 787                 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
 788                                              &masked_key, 0, &acts);
 789                 if (error) {
 790                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 791                         goto err_kfree;
 792                 }
 793         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 794                 error = -EINVAL;
 795                 goto error;
 796         }
 797
 798         ovs_lock();
 799         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 800         error = -ENODEV;
 801         if (!dp)
 802                 goto err_unlock_ovs;
 803
 804         /* Check if this is a duplicate flow */
 805         flow = __ovs_flow_tbl_lookup(&dp->table, &key);
 806         if (!flow) {
 807                 /* Bail out if we're not allowed to create a new flow. */
 808                 error = -ENOENT;
 809                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
 810                         goto err_unlock_ovs;
 811
 812                 /* Allocate flow. */
 813                 flow = ovs_flow_alloc();
 814                 if (IS_ERR(flow)) {
 815                         error = PTR_ERR(flow);
 816                         goto err_unlock_ovs;
 817                 }
 818
 819                 flow->key = masked_key;
 820                 flow->unmasked_key = key;
 821                 rcu_assign_pointer(flow->sf_acts, acts);
 822
 823                 /* Put flow in bucket. */
 824                 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
 825                 if (error) {
 826                         acts = NULL;
 827                         goto err_flow_free;
 828                 }
 829
 830                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
 831                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
 832         } else {
 833                 /* We found a matching flow. */
 834                 struct sw_flow_actions *old_acts;
 835
 836                 /* Bail out if we're not allowed to modify an existing flow.
 837                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 838                  * because Generic Netlink treats the latter as a dump
 839                  * request.  We also accept NLM_F_EXCL in case that bug ever
 840                  * gets fixed.
 841                  */
 842                 error = -EEXIST;
 843                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
 844                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
 845                         goto err_unlock_ovs;
 846
 847                 /* The unmasked key has to be the same for flow updates. */
 848                 error = -EINVAL;
 849                 if (!ovs_flow_cmp_unmasked_key(flow, &match)) {
 850                         OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
 851                         goto err_unlock_ovs;
 852                 }
 853
 854                 /* Update actions. */
 855                 old_acts = ovsl_dereference(flow->sf_acts);
 856                 rcu_assign_pointer(flow->sf_acts, acts);
 857                 ovs_nla_free_flow_actions(old_acts);
 858
 859                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
 860                                                info->snd_seq, OVS_FLOW_CMD_NEW);
 861
 862                 /* Clear stats. */
 863                 if (a[OVS_FLOW_ATTR_CLEAR])
 864                         ovs_flow_stats_clear(flow);
 865         }
 866         ovs_unlock();
 867
 868         if (!IS_ERR(reply))
 869                 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
 870         else
 871                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
 872                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
 873         return 0;
 874
 875 err_flow_free:
 876         ovs_flow_free(flow, false);
 877 err_unlock_ovs:
 878         ovs_unlock();
 879 err_kfree:
 880         kfree(acts);
 881 error:
 882         return error;
 883 }
 884
 885 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
 886 {
 887         struct nlattr **a = info->attrs;
 888         struct ovs_header *ovs_header = info->userhdr;
 889         struct sw_flow_key key;
 890         struct sk_buff *reply;
 891         struct sw_flow *flow;
 892         struct datapath *dp;
 893         struct sw_flow_match match;
 894         int err;
 895
 896         if (!a[OVS_FLOW_ATTR_KEY]) {
 897                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
 898                 return -EINVAL;
 899         }
 900
 901         ovs_match_init(&match, &key, NULL);
 902         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
 903         if (err)
 904                 return err;
 905
 906         ovs_lock();
 907         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 908         if (!dp) {
 909                 err = -ENODEV;
 910                 goto unlock;
 911         }
 912
 913         flow = __ovs_flow_tbl_lookup(&dp->table, &key);
 914         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
 915                 err = -ENOENT;
 916                 goto unlock;
 917         }
 918
 919         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
 920                                         info->snd_seq, 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);
 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, u32 portid,
1120                                              u32 seq, u8 cmd)
1121 {
1122         struct sk_buff *skb;
1123         int retval;
1124
1125         skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1126         if (!skb)
1127                 return ERR_PTR(-ENOMEM);
1128
1129         retval = ovs_dp_cmd_fill_info(dp, skb, portid, 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->snd_portid,
1219                                       info->snd_seq, OVS_DP_CMD_NEW);
1220         err = PTR_ERR(reply);
1221         if (IS_ERR(reply))
1222                 goto err_destroy_local_port;
1223
1224         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1225         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1226
1227         ovs_unlock();
1228
1229         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1230         return 0;
1231
1232 err_destroy_local_port:
1233         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1234 err_destroy_ports_array:
1235         kfree(dp->ports);
1236 err_destroy_percpu:
1237         free_percpu(dp->stats_percpu);
1238 err_destroy_table:
1239         ovs_flow_tbl_destroy(&dp->table);
1240 err_free_dp:
1241         release_net(ovs_dp_get_net(dp));
1242         kfree(dp);
1243 err_unlock_ovs:
1244         ovs_unlock();
1245 err:
1246         return err;
1247 }
1248
1249 /* Called with ovs_mutex. */
1250 static void __dp_destroy(struct datapath *dp)
1251 {
1252         int i;
1253
1254         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1255                 struct vport *vport;
1256                 struct hlist_node *n;
1257
1258                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1259                         if (vport->port_no != OVSP_LOCAL)
1260                                 ovs_dp_detach_port(vport);
1261         }
1262
1263         list_del_rcu(&dp->list_node);
1264
1265         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1266          * all port in datapath are destroyed first before freeing datapath.
1267          */
1268         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1269
1270         call_rcu(&dp->rcu, destroy_dp_rcu);
1271 }
1272
1273 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1274 {
1275         struct sk_buff *reply;
1276         struct datapath *dp;
1277         int err;
1278
1279         ovs_lock();
1280         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1281         err = PTR_ERR(dp);
1282         if (IS_ERR(dp))
1283                 goto unlock;
1284
1285         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1286                                       info->snd_seq, OVS_DP_CMD_DEL);
1287         err = PTR_ERR(reply);
1288         if (IS_ERR(reply))
1289                 goto unlock;
1290
1291         __dp_destroy(dp);
1292         ovs_unlock();
1293
1294         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1295
1296         return 0;
1297 unlock:
1298         ovs_unlock();
1299         return err;
1300 }
1301
1302 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1303 {
1304         struct sk_buff *reply;
1305         struct datapath *dp;
1306         int err;
1307
1308         ovs_lock();
1309         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1310         err = PTR_ERR(dp);
1311         if (IS_ERR(dp))
1312                 goto unlock;
1313
1314         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1315                                       info->snd_seq, OVS_DP_CMD_NEW);
1316         if (IS_ERR(reply)) {
1317                 err = PTR_ERR(reply);
1318                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1319                                 ovs_dp_datapath_multicast_group.id, err);
1320                 err = 0;
1321                 goto unlock;
1322         }
1323
1324         ovs_unlock();
1325         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1326
1327         return 0;
1328 unlock:
1329         ovs_unlock();
1330         return err;
1331 }
1332
1333 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1334 {
1335         struct sk_buff *reply;
1336         struct datapath *dp;
1337         int err;
1338
1339         ovs_lock();
1340         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1341         if (IS_ERR(dp)) {
1342                 err = PTR_ERR(dp);
1343                 goto unlock;
1344         }
1345
1346         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1347                                       info->snd_seq, OVS_DP_CMD_NEW);
1348         if (IS_ERR(reply)) {
1349                 err = PTR_ERR(reply);
1350                 goto unlock;
1351         }
1352
1353         ovs_unlock();
1354         return genlmsg_reply(reply, info);
1355
1356 unlock:
1357         ovs_unlock();
1358         return err;
1359 }
1360
1361 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1362 {
1363         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1364         struct datapath *dp;
1365         int skip = cb->args[0];
1366         int i = 0;
1367
1368         rcu_read_lock();
1369         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1370                 if (i >= skip &&
1371                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1372                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1373                                          OVS_DP_CMD_NEW) < 0)
1374                         break;
1375                 i++;
1376         }
1377         rcu_read_unlock();
1378
1379         cb->args[0] = i;
1380
1381         return skb->len;
1382 }
1383
1384 static struct genl_ops dp_datapath_genl_ops[] = {
1385         { .cmd = OVS_DP_CMD_NEW,
1386           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1387           .policy = datapath_policy,
1388           .doit = ovs_dp_cmd_new
1389         },
1390         { .cmd = OVS_DP_CMD_DEL,
1391           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1392           .policy = datapath_policy,
1393           .doit = ovs_dp_cmd_del
1394         },
1395         { .cmd = OVS_DP_CMD_GET,
1396           .flags = 0,               /* OK for unprivileged users. */
1397           .policy = datapath_policy,
1398           .doit = ovs_dp_cmd_get,
1399           .dumpit = ovs_dp_cmd_dump
1400         },
1401         { .cmd = OVS_DP_CMD_SET,
1402           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1403           .policy = datapath_policy,
1404           .doit = ovs_dp_cmd_set,
1405         },
1406 };
1407
1408 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1409         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1410         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1411         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1412         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1413         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1414         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1415 };
1416
1417 static struct genl_family dp_vport_genl_family = {
1418         .id = GENL_ID_GENERATE,
1419         .hdrsize = sizeof(struct ovs_header),
1420         .name = OVS_VPORT_FAMILY,
1421         .version = OVS_VPORT_VERSION,
1422         .maxattr = OVS_VPORT_ATTR_MAX,
1423         .netnsok = true,
1424          SET_PARALLEL_OPS
1425 };
1426
1427 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1428         .name = OVS_VPORT_MCGROUP
1429 };
1430
1431 /* Called with ovs_mutex or RCU read lock. */
1432 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1433                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1434 {
1435         struct ovs_header *ovs_header;
1436         struct ovs_vport_stats vport_stats;
1437         int err;
1438
1439         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1440                                  flags, cmd);
1441         if (!ovs_header)
1442                 return -EMSGSIZE;
1443
1444         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1445
1446         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1447             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1448             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1449             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1450                 goto nla_put_failure;
1451
1452         ovs_vport_get_stats(vport, &vport_stats);
1453         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1454                     &vport_stats))
1455                 goto nla_put_failure;
1456
1457         err = ovs_vport_get_options(vport, skb);
1458         if (err == -EMSGSIZE)
1459                 goto error;
1460
1461         return genlmsg_end(skb, ovs_header);
1462
1463 nla_put_failure:
1464         err = -EMSGSIZE;
1465 error:
1466         genlmsg_cancel(skb, ovs_header);
1467         return err;
1468 }
1469
1470 /* Called with ovs_mutex or RCU read lock. */
1471 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1472                                          u32 seq, u8 cmd)
1473 {
1474         struct sk_buff *skb;
1475         int retval;
1476
1477         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1478         if (!skb)
1479                 return ERR_PTR(-ENOMEM);
1480
1481         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1482         BUG_ON(retval < 0);
1483
1484         return skb;
1485 }
1486
1487 /* Called with ovs_mutex or RCU read lock. */
1488 static struct vport *lookup_vport(struct net *net,
1489                                   struct ovs_header *ovs_header,
1490                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1491 {
1492         struct datapath *dp;
1493         struct vport *vport;
1494
1495         if (a[OVS_VPORT_ATTR_NAME]) {
1496                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1497                 if (!vport)
1498                         return ERR_PTR(-ENODEV);
1499                 if (ovs_header->dp_ifindex &&
1500                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1501                         return ERR_PTR(-ENODEV);
1502                 return vport;
1503         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1504                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1505
1506                 if (port_no >= DP_MAX_PORTS)
1507                         return ERR_PTR(-EFBIG);
1508
1509                 dp = get_dp(net, ovs_header->dp_ifindex);
1510                 if (!dp)
1511                         return ERR_PTR(-ENODEV);
1512
1513                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1514                 if (!vport)
1515                         return ERR_PTR(-ENODEV);
1516                 return vport;
1517         } else
1518                 return ERR_PTR(-EINVAL);
1519 }
1520
1521 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1522 {
1523         struct nlattr **a = info->attrs;
1524         struct ovs_header *ovs_header = info->userhdr;
1525         struct vport_parms parms;
1526         struct sk_buff *reply;
1527         struct vport *vport;
1528         struct datapath *dp;
1529         u32 port_no;
1530         int err;
1531
1532         err = -EINVAL;
1533         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1534             !a[OVS_VPORT_ATTR_UPCALL_PID])
1535                 goto exit;
1536
1537         ovs_lock();
1538         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1539         err = -ENODEV;
1540         if (!dp)
1541                 goto exit_unlock;
1542
1543         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1544                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1545
1546                 err = -EFBIG;
1547                 if (port_no >= DP_MAX_PORTS)
1548                         goto exit_unlock;
1549
1550                 vport = ovs_vport_ovsl(dp, port_no);
1551                 err = -EBUSY;
1552                 if (vport)
1553                         goto exit_unlock;
1554         } else {
1555                 for (port_no = 1; ; port_no++) {
1556                         if (port_no >= DP_MAX_PORTS) {
1557                                 err = -EFBIG;
1558                                 goto exit_unlock;
1559                         }
1560                         vport = ovs_vport_ovsl(dp, port_no);
1561                         if (!vport)
1562                                 break;
1563                 }
1564         }
1565
1566         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1567         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1568         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1569         parms.dp = dp;
1570         parms.port_no = port_no;
1571         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1572
1573         vport = new_vport(&parms);
1574         err = PTR_ERR(vport);
1575         if (IS_ERR(vport))
1576                 goto exit_unlock;
1577
1578         err = 0;
1579         if (a[OVS_VPORT_ATTR_STATS])
1580                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1581
1582         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1583                                          OVS_VPORT_CMD_NEW);
1584         if (IS_ERR(reply)) {
1585                 err = PTR_ERR(reply);
1586                 ovs_dp_detach_port(vport);
1587                 goto exit_unlock;
1588         }
1589
1590         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1591
1592 exit_unlock:
1593         ovs_unlock();
1594 exit:
1595         return err;
1596 }
1597
1598 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1599 {
1600         struct nlattr **a = info->attrs;
1601         struct sk_buff *reply;
1602         struct vport *vport;
1603         int err;
1604
1605         ovs_lock();
1606         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1607         err = PTR_ERR(vport);
1608         if (IS_ERR(vport))
1609                 goto exit_unlock;
1610
1611         if (a[OVS_VPORT_ATTR_TYPE] &&
1612             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1613                 err = -EINVAL;
1614                 goto exit_unlock;
1615         }
1616
1617         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1618         if (!reply) {
1619                 err = -ENOMEM;
1620                 goto exit_unlock;
1621         }
1622
1623         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1624                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1625                 if (err)
1626                         goto exit_free;
1627         }
1628
1629         if (a[OVS_VPORT_ATTR_STATS])
1630                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1631
1632         if (a[OVS_VPORT_ATTR_UPCALL_PID])
1633                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1634
1635         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1636                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1637         BUG_ON(err < 0);
1638
1639         ovs_unlock();
1640         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1641         return 0;
1642
1643 exit_free:
1644         kfree_skb(reply);
1645 exit_unlock:
1646         ovs_unlock();
1647         return err;
1648 }
1649
1650 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1651 {
1652         struct nlattr **a = info->attrs;
1653         struct sk_buff *reply;
1654         struct vport *vport;
1655         int err;
1656
1657         ovs_lock();
1658         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1659         err = PTR_ERR(vport);
1660         if (IS_ERR(vport))
1661                 goto exit_unlock;
1662
1663         if (vport->port_no == OVSP_LOCAL) {
1664                 err = -EINVAL;
1665                 goto exit_unlock;
1666         }
1667
1668         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1669                                          info->snd_seq, OVS_VPORT_CMD_DEL);
1670         err = PTR_ERR(reply);
1671         if (IS_ERR(reply))
1672                 goto exit_unlock;
1673
1674         err = 0;
1675         ovs_dp_detach_port(vport);
1676
1677         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1678
1679 exit_unlock:
1680         ovs_unlock();
1681         return err;
1682 }
1683
1684 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1685 {
1686         struct nlattr **a = info->attrs;
1687         struct ovs_header *ovs_header = info->userhdr;
1688         struct sk_buff *reply;
1689         struct vport *vport;
1690         int err;
1691
1692         rcu_read_lock();
1693         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1694         err = PTR_ERR(vport);
1695         if (IS_ERR(vport))
1696                 goto exit_unlock;
1697
1698         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1699                                          info->snd_seq, OVS_VPORT_CMD_NEW);
1700         err = PTR_ERR(reply);
1701         if (IS_ERR(reply))
1702                 goto exit_unlock;
1703
1704         rcu_read_unlock();
1705
1706         return genlmsg_reply(reply, info);
1707
1708 exit_unlock:
1709         rcu_read_unlock();
1710         return err;
1711 }
1712
1713 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1714 {
1715         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1716         struct datapath *dp;
1717         int bucket = cb->args[0], skip = cb->args[1];
1718         int i, j = 0;
1719
1720         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1721         if (!dp)
1722                 return -ENODEV;
1723
1724         rcu_read_lock();
1725         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1726                 struct vport *vport;
1727
1728                 j = 0;
1729                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1730                         if (j >= skip &&
1731                             ovs_vport_cmd_fill_info(vport, skb,
1732                                                     NETLINK_CB(cb->skb).portid,
1733                                                     cb->nlh->nlmsg_seq,
1734                                                     NLM_F_MULTI,
1735                                                     OVS_VPORT_CMD_NEW) < 0)
1736                                 goto out;
1737
1738                         j++;
1739                 }
1740                 skip = 0;
1741         }
1742 out:
1743         rcu_read_unlock();
1744
1745         cb->args[0] = i;
1746         cb->args[1] = j;
1747
1748         return skb->len;
1749 }
1750
1751 static struct genl_ops dp_vport_genl_ops[] = {
1752         { .cmd = OVS_VPORT_CMD_NEW,
1753           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1754           .policy = vport_policy,
1755           .doit = ovs_vport_cmd_new
1756         },
1757         { .cmd = OVS_VPORT_CMD_DEL,
1758           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1759           .policy = vport_policy,
1760           .doit = ovs_vport_cmd_del
1761         },
1762         { .cmd = OVS_VPORT_CMD_GET,
1763           .flags = 0,               /* OK for unprivileged users. */
1764           .policy = vport_policy,
1765           .doit = ovs_vport_cmd_get,
1766           .dumpit = ovs_vport_cmd_dump
1767         },
1768         { .cmd = OVS_VPORT_CMD_SET,
1769           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1770           .policy = vport_policy,
1771           .doit = ovs_vport_cmd_set,
1772         },
1773 };
1774
1775 struct genl_family_and_ops {
1776         struct genl_family *family;
1777         struct genl_ops *ops;
1778         int n_ops;
1779         struct genl_multicast_group *group;
1780 };
1781
1782 static const struct genl_family_and_ops dp_genl_families[] = {
1783         { &dp_datapath_genl_family,
1784           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1785           &ovs_dp_datapath_multicast_group },
1786         { &dp_vport_genl_family,
1787           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1788           &ovs_dp_vport_multicast_group },
1789         { &dp_flow_genl_family,
1790           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1791           &ovs_dp_flow_multicast_group },
1792         { &dp_packet_genl_family,
1793           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1794           NULL },
1795 };
1796
1797 static void dp_unregister_genl(int n_families)
1798 {
1799         int i;
1800
1801         for (i = 0; i < n_families; i++)
1802                 genl_unregister_family(dp_genl_families[i].family);
1803 }
1804
1805 static int dp_register_genl(void)
1806 {
1807         int n_registered;
1808         int err;
1809         int i;
1810
1811         n_registered = 0;
1812         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1813                 const struct genl_family_and_ops *f = &dp_genl_families[i];
1814
1815                 err = genl_register_family_with_ops(f->family, f->ops,
1816                                                     f->n_ops);
1817                 if (err)
1818                         goto error;
1819                 n_registered++;
1820
1821                 if (f->group) {
1822                         err = genl_register_mc_group(f->family, f->group);
1823                         if (err)
1824                                 goto error;
1825                 }
1826         }
1827
1828         return 0;
1829
1830 error:
1831         dp_unregister_genl(n_registered);
1832         return err;
1833 }
1834
1835 static int __net_init ovs_init_net(struct net *net)
1836 {
1837         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1838
1839         INIT_LIST_HEAD(&ovs_net->dps);
1840         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1841         return 0;
1842 }
1843
1844 static void __net_exit ovs_exit_net(struct net *net)
1845 {
1846         struct datapath *dp, *dp_next;
1847         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1848
1849         ovs_lock();
1850         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1851                 __dp_destroy(dp);
1852         ovs_unlock();
1853
1854         cancel_work_sync(&ovs_net->dp_notify_work);
1855 }
1856
1857 static struct pernet_operations ovs_net_ops = {
1858         .init = ovs_init_net,
1859         .exit = ovs_exit_net,
1860         .id   = &ovs_net_id,
1861         .size = sizeof(struct ovs_net),
1862 };
1863
1864 DEFINE_COMPAT_PNET_REG_FUNC(device);
1865
1866 static int __init dp_init(void)
1867 {
1868         int err;
1869
1870         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1871
1872         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
1873                 VERSION);
1874
1875         err = ovs_flow_init();
1876         if (err)
1877                 goto error;
1878
1879         err = ovs_vport_init();
1880         if (err)
1881                 goto error_flow_exit;
1882
1883         err = register_pernet_device(&ovs_net_ops);
1884         if (err)
1885                 goto error_vport_exit;
1886
1887         err = register_netdevice_notifier(&ovs_dp_device_notifier);
1888         if (err)
1889                 goto error_netns_exit;
1890
1891         err = dp_register_genl();
1892         if (err < 0)
1893                 goto error_unreg_notifier;
1894
1895         return 0;
1896
1897 error_unreg_notifier:
1898         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1899 error_netns_exit:
1900         unregister_pernet_device(&ovs_net_ops);
1901 error_vport_exit:
1902         ovs_vport_exit();
1903 error_flow_exit:
1904         ovs_flow_exit();
1905 error:
1906         return err;
1907 }
1908
1909 static void dp_cleanup(void)
1910 {
1911         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1912         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1913         unregister_pernet_device(&ovs_net_ops);
1914         rcu_barrier();
1915         ovs_vport_exit();
1916         ovs_flow_exit();
1917 }
1918
1919 module_init(dp_init);
1920 module_exit(dp_cleanup);
1921
1922 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1923 MODULE_LICENSE("GPL");
1924 MODULE_VERSION(VERSION);