net/openvswitch/datapath.c

   1 /*
   2  * Copyright (c) 2007-2012 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/ethtool.h>
  40 #include <linux/wait.h>
  41 #include <asm/div64.h>
  42 #include <linux/highmem.h>
  43 #include <linux/netfilter_bridge.h>
  44 #include <linux/netfilter_ipv4.h>
  45 #include <linux/inetdevice.h>
  46 #include <linux/list.h>
  47 #include <linux/openvswitch.h>
  48 #include <linux/rculist.h>
  49 #include <linux/dmi.h>
  50 #include <linux/workqueue.h>
  51 #include <net/genetlink.h>
  52 #include <net/net_namespace.h>
  53 #include <net/netns/generic.h>
  54
  55 #include "datapath.h"
  56 #include "flow.h"
  57 #include "vport-internal_dev.h"
  58
  59 /**
  60  * struct ovs_net - Per net-namespace data for ovs.
  61  * @dps: List of datapaths to enable dumping them all out.
  62  * Protected by genl_mutex.
  63  */
  64 struct ovs_net {
  65         struct list_head dps;
  66 };
  67
  68 static int ovs_net_id __read_mostly;
  69
  70 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
  71 static void rehash_flow_table(struct work_struct *work);
  72 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
  73
  74 /**
  75  * DOC: Locking:
  76  *
  77  * Writes to device state (add/remove datapath, port, set operations on vports,
  78  * etc.) are protected by RTNL.
  79  *
  80  * Writes to other state (flow table modifications, set miscellaneous datapath
  81  * parameters, etc.) are protected by genl_mutex.  The RTNL lock nests inside
  82  * genl_mutex.
  83  *
  84  * Reads are protected by RCU.
  85  *
  86  * There are a few special cases (mostly stats) that have their own
  87  * synchronization but they nest under all of above and don't interact with
  88  * each other.
  89  */
  90
  91 static struct vport *new_vport(const struct vport_parms *);
  92 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
  93                              const struct dp_upcall_info *);
  94 static int queue_userspace_packet(struct net *, int dp_ifindex,
  95                                   struct sk_buff *,
  96                                   const struct dp_upcall_info *);
  97
  98 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  99 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 100 {
 101         struct datapath *dp = NULL;
 102         struct net_device *dev;
 103
 104         rcu_read_lock();
 105         dev = dev_get_by_index_rcu(net, dp_ifindex);
 106         if (dev) {
 107                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 108                 if (vport)
 109                         dp = vport->dp;
 110         }
 111         rcu_read_unlock();
 112
 113         return dp;
 114 }
 115
 116 /* Must be called with rcu_read_lock or RTNL lock. */
 117 const char *ovs_dp_name(const struct datapath *dp)
 118 {
 119         struct vport *vport = ovs_vport_rtnl_rcu(dp, OVSP_LOCAL);
 120         return vport->ops->get_name(vport);
 121 }
 122
 123 static int get_dpifindex(struct datapath *dp)
 124 {
 125         struct vport *local;
 126         int ifindex;
 127
 128         rcu_read_lock();
 129
 130         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 131         if (local)
 132                 ifindex = local->ops->get_ifindex(local);
 133         else
 134                 ifindex = 0;
 135
 136         rcu_read_unlock();
 137
 138         return ifindex;
 139 }
 140
 141 static void destroy_dp_rcu(struct rcu_head *rcu)
 142 {
 143         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 144
 145         ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
 146         free_percpu(dp->stats_percpu);
 147         release_net(ovs_dp_get_net(dp));
 148         kfree(dp->ports);
 149         kfree(dp);
 150 }
 151
 152 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 153                                             u16 port_no)
 154 {
 155         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 156 }
 157
 158 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 159 {
 160         struct vport *vport;
 161         struct hlist_node *n;
 162         struct hlist_head *head;
 163
 164         head = vport_hash_bucket(dp, port_no);
 165         hlist_for_each_entry_rcu(vport, n, head, dp_hash_node) {
 166                 if (vport->port_no == port_no)
 167                         return vport;
 168         }
 169         return NULL;
 170 }
 171
 172 /* Called with RTNL lock and genl_lock. */
 173 static struct vport *new_vport(const struct vport_parms *parms)
 174 {
 175         struct vport *vport;
 176
 177         vport = ovs_vport_add(parms);
 178         if (!IS_ERR(vport)) {
 179                 struct datapath *dp = parms->dp;
 180                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 181
 182                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 183         }
 184
 185         return vport;
 186 }
 187
 188 /* Called with RTNL lock. */
 189 void ovs_dp_detach_port(struct vport *p)
 190 {
 191         ASSERT_RTNL();
 192
 193         /* First drop references to device. */
 194         hlist_del_rcu(&p->dp_hash_node);
 195
 196         /* Then destroy it. */
 197         ovs_vport_del(p);
 198 }
 199
 200 /* Must be called with rcu_read_lock. */
 201 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 202 {
 203         struct datapath *dp = p->dp;
 204         struct sw_flow *flow;
 205         struct dp_stats_percpu *stats;
 206         struct sw_flow_key key;
 207         u64 *stats_counter;
 208         int error;
 209         int key_len;
 210
 211         stats = this_cpu_ptr(dp->stats_percpu);
 212
 213         /* Extract flow from 'skb' into 'key'. */
 214         error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
 215         if (unlikely(error)) {
 216                 kfree_skb(skb);
 217                 return;
 218         }
 219
 220         /* Look up flow. */
 221         flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
 222         if (unlikely(!flow)) {
 223                 struct dp_upcall_info upcall;
 224
 225                 upcall.cmd = OVS_PACKET_CMD_MISS;
 226                 upcall.key = &key;
 227                 upcall.userdata = NULL;
 228                 upcall.portid = p->upcall_portid;
 229                 ovs_dp_upcall(dp, skb, &upcall);
 230                 consume_skb(skb);
 231                 stats_counter = &stats->n_missed;
 232                 goto out;
 233         }
 234
 235         OVS_CB(skb)->flow = flow;
 236
 237         stats_counter = &stats->n_hit;
 238         ovs_flow_used(OVS_CB(skb)->flow, skb);
 239         ovs_execute_actions(dp, skb);
 240
 241 out:
 242         /* Update datapath statistics. */
 243         u64_stats_update_begin(&stats->sync);
 244         (*stats_counter)++;
 245         u64_stats_update_end(&stats->sync);
 246 }
 247
 248 static struct genl_family dp_packet_genl_family = {
 249         .id = GENL_ID_GENERATE,
 250         .hdrsize = sizeof(struct ovs_header),
 251         .name = OVS_PACKET_FAMILY,
 252         .version = OVS_PACKET_VERSION,
 253         .maxattr = OVS_PACKET_ATTR_MAX,
 254         .netnsok = true
 255 };
 256
 257 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 258                   const struct dp_upcall_info *upcall_info)
 259 {
 260         struct dp_stats_percpu *stats;
 261         int dp_ifindex;
 262         int err;
 263
 264         if (upcall_info->portid == 0) {
 265                 err = -ENOTCONN;
 266                 goto err;
 267         }
 268
 269         dp_ifindex = get_dpifindex(dp);
 270         if (!dp_ifindex) {
 271                 err = -ENODEV;
 272                 goto err;
 273         }
 274
 275         if (!skb_is_gso(skb))
 276                 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 277         else
 278                 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 279         if (err)
 280                 goto err;
 281
 282         return 0;
 283
 284 err:
 285         stats = this_cpu_ptr(dp->stats_percpu);
 286
 287         u64_stats_update_begin(&stats->sync);
 288         stats->n_lost++;
 289         u64_stats_update_end(&stats->sync);
 290
 291         return err;
 292 }
 293
 294 static int queue_gso_packets(struct net *net, int dp_ifindex,
 295                              struct sk_buff *skb,
 296                              const struct dp_upcall_info *upcall_info)
 297 {
 298         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 299         struct dp_upcall_info later_info;
 300         struct sw_flow_key later_key;
 301         struct sk_buff *segs, *nskb;
 302         int err;
 303
 304         segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 305         if (IS_ERR(segs))
 306                 return PTR_ERR(segs);
 307
 308         /* Queue all of the segments. */
 309         skb = segs;
 310         do {
 311                 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
 312                 if (err)
 313                         break;
 314
 315                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 316                         /* The initial flow key extracted by ovs_flow_extract()
 317                          * in this case is for a first fragment, so we need to
 318                          * properly mark later fragments.
 319                          */
 320                         later_key = *upcall_info->key;
 321                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 322
 323                         later_info = *upcall_info;
 324                         later_info.key = &later_key;
 325                         upcall_info = &later_info;
 326                 }
 327         } while ((skb = skb->next));
 328
 329         /* Free all of the segments. */
 330         skb = segs;
 331         do {
 332                 nskb = skb->next;
 333                 if (err)
 334                         kfree_skb(skb);
 335                 else
 336                         consume_skb(skb);
 337         } while ((skb = nskb));
 338         return err;
 339 }
 340
 341 static int queue_userspace_packet(struct net *net, int dp_ifindex,
 342                                   struct sk_buff *skb,
 343                                   const struct dp_upcall_info *upcall_info)
 344 {
 345         struct ovs_header *upcall;
 346         struct sk_buff *nskb = NULL;
 347         struct sk_buff *user_skb; /* to be queued to userspace */
 348         struct nlattr *nla;
 349         unsigned int len;
 350         int err;
 351
 352         if (vlan_tx_tag_present(skb)) {
 353                 nskb = skb_clone(skb, GFP_ATOMIC);
 354                 if (!nskb)
 355                         return -ENOMEM;
 356
 357                 nskb = __vlan_put_tag(nskb, vlan_tx_tag_get(nskb));
 358                 if (!nskb)
 359                         return -ENOMEM;
 360
 361                 nskb->vlan_tci = 0;
 362                 skb = nskb;
 363         }
 364
 365         if (nla_attr_size(skb->len) > USHRT_MAX) {
 366                 err = -EFBIG;
 367                 goto out;
 368         }
 369
 370         len = sizeof(struct ovs_header);
 371         len += nla_total_size(skb->len);
 372         len += nla_total_size(FLOW_BUFSIZE);
 373         if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
 374                 len += nla_total_size(8);
 375
 376         user_skb = genlmsg_new(len, GFP_ATOMIC);
 377         if (!user_skb) {
 378                 err = -ENOMEM;
 379                 goto out;
 380         }
 381
 382         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 383                              0, upcall_info->cmd);
 384         upcall->dp_ifindex = dp_ifindex;
 385
 386         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 387         ovs_flow_to_nlattrs(upcall_info->key, user_skb);
 388         nla_nest_end(user_skb, nla);
 389
 390         if (upcall_info->userdata)
 391                 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
 392                             nla_get_u64(upcall_info->userdata));
 393
 394         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 395
 396         skb_copy_and_csum_dev(skb, nla_data(nla));
 397
 398         err = genlmsg_unicast(net, user_skb, upcall_info->portid);
 399
 400 out:
 401         kfree_skb(nskb);
 402         return err;
 403 }
 404
 405 /* Called with genl_mutex. */
 406 static int flush_flows(struct datapath *dp)
 407 {
 408         struct flow_table *old_table;
 409         struct flow_table *new_table;
 410
 411         old_table = genl_dereference(dp->table);
 412         new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
 413         if (!new_table)
 414                 return -ENOMEM;
 415
 416         rcu_assign_pointer(dp->table, new_table);
 417
 418         ovs_flow_tbl_deferred_destroy(old_table);
 419         return 0;
 420 }
 421
 422 static int validate_actions(const struct nlattr *attr,
 423                                 const struct sw_flow_key *key, int depth);
 424
 425 static int validate_sample(const struct nlattr *attr,
 426                                 const struct sw_flow_key *key, int depth)
 427 {
 428         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 429         const struct nlattr *probability, *actions;
 430         const struct nlattr *a;
 431         int rem;
 432
 433         memset(attrs, 0, sizeof(attrs));
 434         nla_for_each_nested(a, attr, rem) {
 435                 int type = nla_type(a);
 436                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 437                         return -EINVAL;
 438                 attrs[type] = a;
 439         }
 440         if (rem)
 441                 return -EINVAL;
 442
 443         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 444         if (!probability || nla_len(probability) != sizeof(u32))
 445                 return -EINVAL;
 446
 447         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 448         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 449                 return -EINVAL;
 450         return validate_actions(actions, key, depth + 1);
 451 }
 452
 453 static int validate_tp_port(const struct sw_flow_key *flow_key)
 454 {
 455         if (flow_key->eth.type == htons(ETH_P_IP)) {
 456                 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
 457                         return 0;
 458         } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
 459                 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
 460                         return 0;
 461         }
 462
 463         return -EINVAL;
 464 }
 465
 466 static int validate_set(const struct nlattr *a,
 467                         const struct sw_flow_key *flow_key)
 468 {
 469         const struct nlattr *ovs_key = nla_data(a);
 470         int key_type = nla_type(ovs_key);
 471
 472         /* There can be only one key in a action */
 473         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 474                 return -EINVAL;
 475
 476         if (key_type > OVS_KEY_ATTR_MAX ||
 477             nla_len(ovs_key) != ovs_key_lens[key_type])
 478                 return -EINVAL;
 479
 480         switch (key_type) {
 481         const struct ovs_key_ipv4 *ipv4_key;
 482         const struct ovs_key_ipv6 *ipv6_key;
 483
 484         case OVS_KEY_ATTR_PRIORITY:
 485         case OVS_KEY_ATTR_ETHERNET:
 486                 break;
 487
 488         case OVS_KEY_ATTR_IPV4:
 489                 if (flow_key->eth.type != htons(ETH_P_IP))
 490                         return -EINVAL;
 491
 492                 if (!flow_key->ip.proto)
 493                         return -EINVAL;
 494
 495                 ipv4_key = nla_data(ovs_key);
 496                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 497                         return -EINVAL;
 498
 499                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 500                         return -EINVAL;
 501
 502                 break;
 503
 504         case OVS_KEY_ATTR_IPV6:
 505                 if (flow_key->eth.type != htons(ETH_P_IPV6))
 506                         return -EINVAL;
 507
 508                 if (!flow_key->ip.proto)
 509                         return -EINVAL;
 510
 511                 ipv6_key = nla_data(ovs_key);
 512                 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
 513                         return -EINVAL;
 514
 515                 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
 516                         return -EINVAL;
 517
 518                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
 519                         return -EINVAL;
 520
 521                 break;
 522
 523         case OVS_KEY_ATTR_TCP:
 524                 if (flow_key->ip.proto != IPPROTO_TCP)
 525                         return -EINVAL;
 526
 527                 return validate_tp_port(flow_key);
 528
 529         case OVS_KEY_ATTR_UDP:
 530                 if (flow_key->ip.proto != IPPROTO_UDP)
 531                         return -EINVAL;
 532
 533                 return validate_tp_port(flow_key);
 534
 535         default:
 536                 return -EINVAL;
 537         }
 538
 539         return 0;
 540 }
 541
 542 static int validate_userspace(const struct nlattr *attr)
 543 {
 544         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 545                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 546                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
 547         };
 548         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 549         int error;
 550
 551         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 552                                  attr, userspace_policy);
 553         if (error)
 554                 return error;
 555
 556         if (!a[OVS_USERSPACE_ATTR_PID] ||
 557             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 558                 return -EINVAL;
 559
 560         return 0;
 561 }
 562
 563 static int validate_actions(const struct nlattr *attr,
 564                                 const struct sw_flow_key *key,  int depth)
 565 {
 566         const struct nlattr *a;
 567         int rem, err;
 568
 569         if (depth >= SAMPLE_ACTION_DEPTH)
 570                 return -EOVERFLOW;
 571
 572         nla_for_each_nested(a, attr, rem) {
 573                 /* Expected argument lengths, (u32)-1 for variable length. */
 574                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 575                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 576                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 577                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 578                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 579                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 580                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 581                 };
 582                 const struct ovs_action_push_vlan *vlan;
 583                 int type = nla_type(a);
 584
 585                 if (type > OVS_ACTION_ATTR_MAX ||
 586                     (action_lens[type] != nla_len(a) &&
 587                      action_lens[type] != (u32)-1))
 588                         return -EINVAL;
 589
 590                 switch (type) {
 591                 case OVS_ACTION_ATTR_UNSPEC:
 592                         return -EINVAL;
 593
 594                 case OVS_ACTION_ATTR_USERSPACE:
 595                         err = validate_userspace(a);
 596                         if (err)
 597                                 return err;
 598                         break;
 599
 600                 case OVS_ACTION_ATTR_OUTPUT:
 601                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 602                                 return -EINVAL;
 603                         break;
 604
 605
 606                 case OVS_ACTION_ATTR_POP_VLAN:
 607                         break;
 608
 609                 case OVS_ACTION_ATTR_PUSH_VLAN:
 610                         vlan = nla_data(a);
 611                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 612                                 return -EINVAL;
 613                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 614                                 return -EINVAL;
 615                         break;
 616
 617                 case OVS_ACTION_ATTR_SET:
 618                         err = validate_set(a, key);
 619                         if (err)
 620                                 return err;
 621                         break;
 622
 623                 case OVS_ACTION_ATTR_SAMPLE:
 624                         err = validate_sample(a, key, depth);
 625                         if (err)
 626                                 return err;
 627                         break;
 628
 629                 default:
 630                         return -EINVAL;
 631                 }
 632         }
 633
 634         if (rem > 0)
 635                 return -EINVAL;
 636
 637         return 0;
 638 }
 639
 640 static void clear_stats(struct sw_flow *flow)
 641 {
 642         flow->used = 0;
 643         flow->tcp_flags = 0;
 644         flow->packet_count = 0;
 645         flow->byte_count = 0;
 646 }
 647
 648 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 649 {
 650         struct ovs_header *ovs_header = info->userhdr;
 651         struct nlattr **a = info->attrs;
 652         struct sw_flow_actions *acts;
 653         struct sk_buff *packet;
 654         struct sw_flow *flow;
 655         struct datapath *dp;
 656         struct ethhdr *eth;
 657         int len;
 658         int err;
 659         int key_len;
 660
 661         err = -EINVAL;
 662         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 663             !a[OVS_PACKET_ATTR_ACTIONS] ||
 664             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 665                 goto err;
 666
 667         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 668         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 669         err = -ENOMEM;
 670         if (!packet)
 671                 goto err;
 672         skb_reserve(packet, NET_IP_ALIGN);
 673
 674         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 675
 676         skb_reset_mac_header(packet);
 677         eth = eth_hdr(packet);
 678
 679         /* Normally, setting the skb 'protocol' field would be handled by a
 680          * call to eth_type_trans(), but it assumes there's a sending
 681          * device, which we may not have. */
 682         if (ntohs(eth->h_proto) >= 1536)
 683                 packet->protocol = eth->h_proto;
 684         else
 685                 packet->protocol = htons(ETH_P_802_2);
 686
 687         /* Build an sw_flow for sending this packet. */
 688         flow = ovs_flow_alloc();
 689         err = PTR_ERR(flow);
 690         if (IS_ERR(flow))
 691                 goto err_kfree_skb;
 692
 693         err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
 694         if (err)
 695                 goto err_flow_free;
 696
 697         err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
 698                                              &flow->key.phy.in_port,
 699                                              a[OVS_PACKET_ATTR_KEY]);
 700         if (err)
 701                 goto err_flow_free;
 702
 703         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
 704         if (err)
 705                 goto err_flow_free;
 706
 707         flow->hash = ovs_flow_hash(&flow->key, key_len);
 708
 709         acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 710         err = PTR_ERR(acts);
 711         if (IS_ERR(acts))
 712                 goto err_flow_free;
 713         rcu_assign_pointer(flow->sf_acts, acts);
 714
 715         OVS_CB(packet)->flow = flow;
 716         packet->priority = flow->key.phy.priority;
 717
 718         rcu_read_lock();
 719         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 720         err = -ENODEV;
 721         if (!dp)
 722                 goto err_unlock;
 723
 724         local_bh_disable();
 725         err = ovs_execute_actions(dp, packet);
 726         local_bh_enable();
 727         rcu_read_unlock();
 728
 729         ovs_flow_free(flow);
 730         return err;
 731
 732 err_unlock:
 733         rcu_read_unlock();
 734 err_flow_free:
 735         ovs_flow_free(flow);
 736 err_kfree_skb:
 737         kfree_skb(packet);
 738 err:
 739         return err;
 740 }
 741
 742 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 743         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 744         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 745         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 746 };
 747
 748 static struct genl_ops dp_packet_genl_ops[] = {
 749         { .cmd = OVS_PACKET_CMD_EXECUTE,
 750           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 751           .policy = packet_policy,
 752           .doit = ovs_packet_cmd_execute
 753         }
 754 };
 755
 756 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 757 {
 758         int i;
 759         struct flow_table *table = genl_dereference(dp->table);
 760
 761         stats->n_flows = ovs_flow_tbl_count(table);
 762
 763         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 764         for_each_possible_cpu(i) {
 765                 const struct dp_stats_percpu *percpu_stats;
 766                 struct dp_stats_percpu local_stats;
 767                 unsigned int start;
 768
 769                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 770
 771                 do {
 772                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 773                         local_stats = *percpu_stats;
 774                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 775
 776                 stats->n_hit += local_stats.n_hit;
 777                 stats->n_missed += local_stats.n_missed;
 778                 stats->n_lost += local_stats.n_lost;
 779         }
 780 }
 781
 782 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 783         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 784         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 785         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 786 };
 787
 788 static struct genl_family dp_flow_genl_family = {
 789         .id = GENL_ID_GENERATE,
 790         .hdrsize = sizeof(struct ovs_header),
 791         .name = OVS_FLOW_FAMILY,
 792         .version = OVS_FLOW_VERSION,
 793         .maxattr = OVS_FLOW_ATTR_MAX,
 794         .netnsok = true
 795 };
 796
 797 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 798         .name = OVS_FLOW_MCGROUP
 799 };
 800
 801 /* Called with genl_lock. */
 802 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 803                                   struct sk_buff *skb, u32 portid,
 804                                   u32 seq, u32 flags, u8 cmd)
 805 {
 806         const int skb_orig_len = skb->len;
 807         const struct sw_flow_actions *sf_acts;
 808         struct ovs_flow_stats stats;
 809         struct ovs_header *ovs_header;
 810         struct nlattr *nla;
 811         unsigned long used;
 812         u8 tcp_flags;
 813         int err;
 814
 815         sf_acts = rcu_dereference_protected(flow->sf_acts,
 816                                             lockdep_genl_is_held());
 817
 818         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 819         if (!ovs_header)
 820                 return -EMSGSIZE;
 821
 822         ovs_header->dp_ifindex = get_dpifindex(dp);
 823
 824         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 825         if (!nla)
 826                 goto nla_put_failure;
 827         err = ovs_flow_to_nlattrs(&flow->key, skb);
 828         if (err)
 829                 goto error;
 830         nla_nest_end(skb, nla);
 831
 832         spin_lock_bh(&flow->lock);
 833         used = flow->used;
 834         stats.n_packets = flow->packet_count;
 835         stats.n_bytes = flow->byte_count;
 836         tcp_flags = flow->tcp_flags;
 837         spin_unlock_bh(&flow->lock);
 838
 839         if (used &&
 840             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 841                 goto nla_put_failure;
 842
 843         if (stats.n_packets &&
 844             nla_put(skb, OVS_FLOW_ATTR_STATS,
 845                     sizeof(struct ovs_flow_stats), &stats))
 846                 goto nla_put_failure;
 847
 848         if (tcp_flags &&
 849             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
 850                 goto nla_put_failure;
 851
 852         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 853          * this is the first flow to be dumped into 'skb'.  This is unusual for
 854          * Netlink but individual action lists can be longer than
 855          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 856          * The userspace caller can always fetch the actions separately if it
 857          * really wants them.  (Most userspace callers in fact don't care.)
 858          *
 859          * This can only fail for dump operations because the skb is always
 860          * properly sized for single flows.
 861          */
 862         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 863                       sf_acts->actions);
 864         if (err < 0 && skb_orig_len)
 865                 goto error;
 866
 867         return genlmsg_end(skb, ovs_header);
 868
 869 nla_put_failure:
 870         err = -EMSGSIZE;
 871 error:
 872         genlmsg_cancel(skb, ovs_header);
 873         return err;
 874 }
 875
 876 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 877 {
 878         const struct sw_flow_actions *sf_acts;
 879         int len;
 880
 881         sf_acts = rcu_dereference_protected(flow->sf_acts,
 882                                             lockdep_genl_is_held());
 883
 884         /* OVS_FLOW_ATTR_KEY */
 885         len = nla_total_size(FLOW_BUFSIZE);
 886         /* OVS_FLOW_ATTR_ACTIONS */
 887         len += nla_total_size(sf_acts->actions_len);
 888         /* OVS_FLOW_ATTR_STATS */
 889         len += nla_total_size(sizeof(struct ovs_flow_stats));
 890         /* OVS_FLOW_ATTR_TCP_FLAGS */
 891         len += nla_total_size(1);
 892         /* OVS_FLOW_ATTR_USED */
 893         len += nla_total_size(8);
 894
 895         len += NLMSG_ALIGN(sizeof(struct ovs_header));
 896
 897         return genlmsg_new(len, GFP_KERNEL);
 898 }
 899
 900 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 901                                                struct datapath *dp,
 902                                                u32 portid, u32 seq, u8 cmd)
 903 {
 904         struct sk_buff *skb;
 905         int retval;
 906
 907         skb = ovs_flow_cmd_alloc_info(flow);
 908         if (!skb)
 909                 return ERR_PTR(-ENOMEM);
 910
 911         retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
 912         BUG_ON(retval < 0);
 913         return skb;
 914 }
 915
 916 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 917 {
 918         struct nlattr **a = info->attrs;
 919         struct ovs_header *ovs_header = info->userhdr;
 920         struct sw_flow_key key;
 921         struct sw_flow *flow;
 922         struct sk_buff *reply;
 923         struct datapath *dp;
 924         struct flow_table *table;
 925         int error;
 926         int key_len;
 927
 928         /* Extract key. */
 929         error = -EINVAL;
 930         if (!a[OVS_FLOW_ATTR_KEY])
 931                 goto error;
 932         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
 933         if (error)
 934                 goto error;
 935
 936         /* Validate actions. */
 937         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 938                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
 939                 if (error)
 940                         goto error;
 941         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 942                 error = -EINVAL;
 943                 goto error;
 944         }
 945
 946         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 947         error = -ENODEV;
 948         if (!dp)
 949                 goto error;
 950
 951         table = genl_dereference(dp->table);
 952         flow = ovs_flow_tbl_lookup(table, &key, key_len);
 953         if (!flow) {
 954                 struct sw_flow_actions *acts;
 955
 956                 /* Bail out if we're not allowed to create a new flow. */
 957                 error = -ENOENT;
 958                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
 959                         goto error;
 960
 961                 /* Expand table, if necessary, to make room. */
 962                 if (ovs_flow_tbl_need_to_expand(table)) {
 963                         struct flow_table *new_table;
 964
 965                         new_table = ovs_flow_tbl_expand(table);
 966                         if (!IS_ERR(new_table)) {
 967                                 rcu_assign_pointer(dp->table, new_table);
 968                                 ovs_flow_tbl_deferred_destroy(table);
 969                                 table = genl_dereference(dp->table);
 970                         }
 971                 }
 972
 973                 /* Allocate flow. */
 974                 flow = ovs_flow_alloc();
 975                 if (IS_ERR(flow)) {
 976                         error = PTR_ERR(flow);
 977                         goto error;
 978                 }
 979                 flow->key = key;
 980                 clear_stats(flow);
 981
 982                 /* Obtain actions. */
 983                 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
 984                 error = PTR_ERR(acts);
 985                 if (IS_ERR(acts))
 986                         goto error_free_flow;
 987                 rcu_assign_pointer(flow->sf_acts, acts);
 988
 989                 /* Put flow in bucket. */
 990                 flow->hash = ovs_flow_hash(&key, key_len);
 991                 ovs_flow_tbl_insert(table, flow);
 992
 993                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
 994                                                 info->snd_seq,
 995                                                 OVS_FLOW_CMD_NEW);
 996         } else {
 997                 /* We found a matching flow. */
 998                 struct sw_flow_actions *old_acts;
 999                 struct nlattr *acts_attrs;
1000
1001                 /* Bail out if we're not allowed to modify an existing flow.
1002                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1003                  * because Generic Netlink treats the latter as a dump
1004                  * request.  We also accept NLM_F_EXCL in case that bug ever
1005                  * gets fixed.
1006                  */
1007                 error = -EEXIST;
1008                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1009                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1010                         goto error;
1011
1012                 /* Update actions. */
1013                 old_acts = rcu_dereference_protected(flow->sf_acts,
1014                                                      lockdep_genl_is_held());
1015                 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1016                 if (acts_attrs &&
1017                    (old_acts->actions_len != nla_len(acts_attrs) ||
1018                    memcmp(old_acts->actions, nla_data(acts_attrs),
1019                           old_acts->actions_len))) {
1020                         struct sw_flow_actions *new_acts;
1021
1022                         new_acts = ovs_flow_actions_alloc(acts_attrs);
1023                         error = PTR_ERR(new_acts);
1024                         if (IS_ERR(new_acts))
1025                                 goto error;
1026
1027                         rcu_assign_pointer(flow->sf_acts, new_acts);
1028                         ovs_flow_deferred_free_acts(old_acts);
1029                 }
1030
1031                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1032                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1033
1034                 /* Clear stats. */
1035                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1036                         spin_lock_bh(&flow->lock);
1037                         clear_stats(flow);
1038                         spin_unlock_bh(&flow->lock);
1039                 }
1040         }
1041
1042         if (!IS_ERR(reply))
1043                 genl_notify(reply, genl_info_net(info), info->snd_portid,
1044                            ovs_dp_flow_multicast_group.id, info->nlhdr,
1045                            GFP_KERNEL);
1046         else
1047                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1048                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1049         return 0;
1050
1051 error_free_flow:
1052         ovs_flow_free(flow);
1053 error:
1054         return error;
1055 }
1056
1057 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1058 {
1059         struct nlattr **a = info->attrs;
1060         struct ovs_header *ovs_header = info->userhdr;
1061         struct sw_flow_key key;
1062         struct sk_buff *reply;
1063         struct sw_flow *flow;
1064         struct datapath *dp;
1065         struct flow_table *table;
1066         int err;
1067         int key_len;
1068
1069         if (!a[OVS_FLOW_ATTR_KEY])
1070                 return -EINVAL;
1071         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1072         if (err)
1073                 return err;
1074
1075         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1076         if (!dp)
1077                 return -ENODEV;
1078
1079         table = genl_dereference(dp->table);
1080         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1081         if (!flow)
1082                 return -ENOENT;
1083
1084         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1085                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1086         if (IS_ERR(reply))
1087                 return PTR_ERR(reply);
1088
1089         return genlmsg_reply(reply, info);
1090 }
1091
1092 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1093 {
1094         struct nlattr **a = info->attrs;
1095         struct ovs_header *ovs_header = info->userhdr;
1096         struct sw_flow_key key;
1097         struct sk_buff *reply;
1098         struct sw_flow *flow;
1099         struct datapath *dp;
1100         struct flow_table *table;
1101         int err;
1102         int key_len;
1103
1104         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1105         if (!dp)
1106                 return -ENODEV;
1107
1108         if (!a[OVS_FLOW_ATTR_KEY])
1109                 return flush_flows(dp);
1110
1111         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1112         if (err)
1113                 return err;
1114
1115         table = genl_dereference(dp->table);
1116         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1117         if (!flow)
1118                 return -ENOENT;
1119
1120         reply = ovs_flow_cmd_alloc_info(flow);
1121         if (!reply)
1122                 return -ENOMEM;
1123
1124         ovs_flow_tbl_remove(table, flow);
1125
1126         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1127                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1128         BUG_ON(err < 0);
1129
1130         ovs_flow_deferred_free(flow);
1131
1132         genl_notify(reply, genl_info_net(info), info->snd_portid,
1133                     ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1134         return 0;
1135 }
1136
1137 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1138 {
1139         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1140         struct datapath *dp;
1141         struct flow_table *table;
1142
1143         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1144         if (!dp)
1145                 return -ENODEV;
1146
1147         table = genl_dereference(dp->table);
1148
1149         for (;;) {
1150                 struct sw_flow *flow;
1151                 u32 bucket, obj;
1152
1153                 bucket = cb->args[0];
1154                 obj = cb->args[1];
1155                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1156                 if (!flow)
1157                         break;
1158
1159                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1160                                            NETLINK_CB(cb->skb).portid,
1161                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1162                                            OVS_FLOW_CMD_NEW) < 0)
1163                         break;
1164
1165                 cb->args[0] = bucket;
1166                 cb->args[1] = obj;
1167         }
1168         return skb->len;
1169 }
1170
1171 static struct genl_ops dp_flow_genl_ops[] = {
1172         { .cmd = OVS_FLOW_CMD_NEW,
1173           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1174           .policy = flow_policy,
1175           .doit = ovs_flow_cmd_new_or_set
1176         },
1177         { .cmd = OVS_FLOW_CMD_DEL,
1178           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1179           .policy = flow_policy,
1180           .doit = ovs_flow_cmd_del
1181         },
1182         { .cmd = OVS_FLOW_CMD_GET,
1183           .flags = 0,               /* OK for unprivileged users. */
1184           .policy = flow_policy,
1185           .doit = ovs_flow_cmd_get,
1186           .dumpit = ovs_flow_cmd_dump
1187         },
1188         { .cmd = OVS_FLOW_CMD_SET,
1189           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1190           .policy = flow_policy,
1191           .doit = ovs_flow_cmd_new_or_set,
1192         },
1193 };
1194
1195 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1196         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1197         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1198 };
1199
1200 static struct genl_family dp_datapath_genl_family = {
1201         .id = GENL_ID_GENERATE,
1202         .hdrsize = sizeof(struct ovs_header),
1203         .name = OVS_DATAPATH_FAMILY,
1204         .version = OVS_DATAPATH_VERSION,
1205         .maxattr = OVS_DP_ATTR_MAX,
1206         .netnsok = true
1207 };
1208
1209 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1210         .name = OVS_DATAPATH_MCGROUP
1211 };
1212
1213 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1214                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1215 {
1216         struct ovs_header *ovs_header;
1217         struct ovs_dp_stats dp_stats;
1218         int err;
1219
1220         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1221                                    flags, cmd);
1222         if (!ovs_header)
1223                 goto error;
1224
1225         ovs_header->dp_ifindex = get_dpifindex(dp);
1226
1227         rcu_read_lock();
1228         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1229         rcu_read_unlock();
1230         if (err)
1231                 goto nla_put_failure;
1232
1233         get_dp_stats(dp, &dp_stats);
1234         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1235                 goto nla_put_failure;
1236
1237         return genlmsg_end(skb, ovs_header);
1238
1239 nla_put_failure:
1240         genlmsg_cancel(skb, ovs_header);
1241 error:
1242         return -EMSGSIZE;
1243 }
1244
1245 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1246                                              u32 seq, u8 cmd)
1247 {
1248         struct sk_buff *skb;
1249         int retval;
1250
1251         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1252         if (!skb)
1253                 return ERR_PTR(-ENOMEM);
1254
1255         retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1256         if (retval < 0) {
1257                 kfree_skb(skb);
1258                 return ERR_PTR(retval);
1259         }
1260         return skb;
1261 }
1262
1263 /* Called with genl_mutex and optionally with RTNL lock also. */
1264 static struct datapath *lookup_datapath(struct net *net,
1265                                         struct ovs_header *ovs_header,
1266                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1267 {
1268         struct datapath *dp;
1269
1270         if (!a[OVS_DP_ATTR_NAME])
1271                 dp = get_dp(net, ovs_header->dp_ifindex);
1272         else {
1273                 struct vport *vport;
1274
1275                 rcu_read_lock();
1276                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1277                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1278                 rcu_read_unlock();
1279         }
1280         return dp ? dp : ERR_PTR(-ENODEV);
1281 }
1282
1283 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1284 {
1285         struct nlattr **a = info->attrs;
1286         struct vport_parms parms;
1287         struct sk_buff *reply;
1288         struct datapath *dp;
1289         struct vport *vport;
1290         struct ovs_net *ovs_net;
1291         int err, i;
1292
1293         err = -EINVAL;
1294         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1295                 goto err;
1296
1297         rtnl_lock();
1298
1299         err = -ENOMEM;
1300         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1301         if (dp == NULL)
1302                 goto err_unlock_rtnl;
1303
1304         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1305
1306         /* Allocate table. */
1307         err = -ENOMEM;
1308         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1309         if (!dp->table)
1310                 goto err_free_dp;
1311
1312         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1313         if (!dp->stats_percpu) {
1314                 err = -ENOMEM;
1315                 goto err_destroy_table;
1316         }
1317
1318         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1319                         GFP_KERNEL);
1320         if (!dp->ports) {
1321                 err = -ENOMEM;
1322                 goto err_destroy_percpu;
1323         }
1324
1325         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1326                 INIT_HLIST_HEAD(&dp->ports[i]);
1327
1328         /* Set up our datapath device. */
1329         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1330         parms.type = OVS_VPORT_TYPE_INTERNAL;
1331         parms.options = NULL;
1332         parms.dp = dp;
1333         parms.port_no = OVSP_LOCAL;
1334         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1335
1336         vport = new_vport(&parms);
1337         if (IS_ERR(vport)) {
1338                 err = PTR_ERR(vport);
1339                 if (err == -EBUSY)
1340                         err = -EEXIST;
1341
1342                 goto err_destroy_ports_array;
1343         }
1344
1345         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1346                                       info->snd_seq, OVS_DP_CMD_NEW);
1347         err = PTR_ERR(reply);
1348         if (IS_ERR(reply))
1349                 goto err_destroy_local_port;
1350
1351         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1352         list_add_tail(&dp->list_node, &ovs_net->dps);
1353         rtnl_unlock();
1354
1355         genl_notify(reply, genl_info_net(info), info->snd_portid,
1356                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1357                     GFP_KERNEL);
1358         return 0;
1359
1360 err_destroy_local_port:
1361         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1362 err_destroy_ports_array:
1363         kfree(dp->ports);
1364 err_destroy_percpu:
1365         free_percpu(dp->stats_percpu);
1366 err_destroy_table:
1367         ovs_flow_tbl_destroy(genl_dereference(dp->table));
1368 err_free_dp:
1369         release_net(ovs_dp_get_net(dp));
1370         kfree(dp);
1371 err_unlock_rtnl:
1372         rtnl_unlock();
1373 err:
1374         return err;
1375 }
1376
1377 /* Called with genl_mutex. */
1378 static void __dp_destroy(struct datapath *dp)
1379 {
1380         int i;
1381
1382         rtnl_lock();
1383
1384         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1385                 struct vport *vport;
1386                 struct hlist_node *node, *n;
1387
1388                 hlist_for_each_entry_safe(vport, node, n, &dp->ports[i], dp_hash_node)
1389                         if (vport->port_no != OVSP_LOCAL)
1390                                 ovs_dp_detach_port(vport);
1391         }
1392
1393         list_del(&dp->list_node);
1394         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1395
1396         /* rtnl_unlock() will wait until all the references to devices that
1397          * are pending unregistration have been dropped.  We do it here to
1398          * ensure that any internal devices (which contain DP pointers) are
1399          * fully destroyed before freeing the datapath.
1400          */
1401         rtnl_unlock();
1402
1403         call_rcu(&dp->rcu, destroy_dp_rcu);
1404 }
1405
1406 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1407 {
1408         struct sk_buff *reply;
1409         struct datapath *dp;
1410         int err;
1411
1412         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1413         err = PTR_ERR(dp);
1414         if (IS_ERR(dp))
1415                 return err;
1416
1417         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1418                                       info->snd_seq, OVS_DP_CMD_DEL);
1419         err = PTR_ERR(reply);
1420         if (IS_ERR(reply))
1421                 return err;
1422
1423         __dp_destroy(dp);
1424
1425         genl_notify(reply, genl_info_net(info), info->snd_portid,
1426                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1427                     GFP_KERNEL);
1428
1429         return 0;
1430 }
1431
1432 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1433 {
1434         struct sk_buff *reply;
1435         struct datapath *dp;
1436         int err;
1437
1438         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1439         if (IS_ERR(dp))
1440                 return PTR_ERR(dp);
1441
1442         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1443                                       info->snd_seq, OVS_DP_CMD_NEW);
1444         if (IS_ERR(reply)) {
1445                 err = PTR_ERR(reply);
1446                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1447                                 ovs_dp_datapath_multicast_group.id, err);
1448                 return 0;
1449         }
1450
1451         genl_notify(reply, genl_info_net(info), info->snd_portid,
1452                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1453                     GFP_KERNEL);
1454
1455         return 0;
1456 }
1457
1458 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1459 {
1460         struct sk_buff *reply;
1461         struct datapath *dp;
1462
1463         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1464         if (IS_ERR(dp))
1465                 return PTR_ERR(dp);
1466
1467         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1468                                       info->snd_seq, OVS_DP_CMD_NEW);
1469         if (IS_ERR(reply))
1470                 return PTR_ERR(reply);
1471
1472         return genlmsg_reply(reply, info);
1473 }
1474
1475 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1476 {
1477         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1478         struct datapath *dp;
1479         int skip = cb->args[0];
1480         int i = 0;
1481
1482         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1483                 if (i >= skip &&
1484                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1485                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1486                                          OVS_DP_CMD_NEW) < 0)
1487                         break;
1488                 i++;
1489         }
1490
1491         cb->args[0] = i;
1492
1493         return skb->len;
1494 }
1495
1496 static struct genl_ops dp_datapath_genl_ops[] = {
1497         { .cmd = OVS_DP_CMD_NEW,
1498           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1499           .policy = datapath_policy,
1500           .doit = ovs_dp_cmd_new
1501         },
1502         { .cmd = OVS_DP_CMD_DEL,
1503           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1504           .policy = datapath_policy,
1505           .doit = ovs_dp_cmd_del
1506         },
1507         { .cmd = OVS_DP_CMD_GET,
1508           .flags = 0,               /* OK for unprivileged users. */
1509           .policy = datapath_policy,
1510           .doit = ovs_dp_cmd_get,
1511           .dumpit = ovs_dp_cmd_dump
1512         },
1513         { .cmd = OVS_DP_CMD_SET,
1514           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1515           .policy = datapath_policy,
1516           .doit = ovs_dp_cmd_set,
1517         },
1518 };
1519
1520 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1521         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1522         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1523         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1524         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1525         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1526         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1527 };
1528
1529 static struct genl_family dp_vport_genl_family = {
1530         .id = GENL_ID_GENERATE,
1531         .hdrsize = sizeof(struct ovs_header),
1532         .name = OVS_VPORT_FAMILY,
1533         .version = OVS_VPORT_VERSION,
1534         .maxattr = OVS_VPORT_ATTR_MAX,
1535         .netnsok = true
1536 };
1537
1538 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1539         .name = OVS_VPORT_MCGROUP
1540 };
1541
1542 /* Called with RTNL lock or RCU read lock. */
1543 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1544                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1545 {
1546         struct ovs_header *ovs_header;
1547         struct ovs_vport_stats vport_stats;
1548         int err;
1549
1550         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1551                                  flags, cmd);
1552         if (!ovs_header)
1553                 return -EMSGSIZE;
1554
1555         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1556
1557         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1558             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1559             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1560             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1561                 goto nla_put_failure;
1562
1563         ovs_vport_get_stats(vport, &vport_stats);
1564         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1565                     &vport_stats))
1566                 goto nla_put_failure;
1567
1568         err = ovs_vport_get_options(vport, skb);
1569         if (err == -EMSGSIZE)
1570                 goto error;
1571
1572         return genlmsg_end(skb, ovs_header);
1573
1574 nla_put_failure:
1575         err = -EMSGSIZE;
1576 error:
1577         genlmsg_cancel(skb, ovs_header);
1578         return err;
1579 }
1580
1581 /* Called with RTNL lock or RCU read lock. */
1582 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1583                                          u32 seq, u8 cmd)
1584 {
1585         struct sk_buff *skb;
1586         int retval;
1587
1588         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1589         if (!skb)
1590                 return ERR_PTR(-ENOMEM);
1591
1592         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1593         if (retval < 0) {
1594                 kfree_skb(skb);
1595                 return ERR_PTR(retval);
1596         }
1597         return skb;
1598 }
1599
1600 /* Called with RTNL lock or RCU read lock. */
1601 static struct vport *lookup_vport(struct net *net,
1602                                   struct ovs_header *ovs_header,
1603                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1604 {
1605         struct datapath *dp;
1606         struct vport *vport;
1607
1608         if (a[OVS_VPORT_ATTR_NAME]) {
1609                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1610                 if (!vport)
1611                         return ERR_PTR(-ENODEV);
1612                 if (ovs_header->dp_ifindex &&
1613                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1614                         return ERR_PTR(-ENODEV);
1615                 return vport;
1616         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1617                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1618
1619                 if (port_no >= DP_MAX_PORTS)
1620                         return ERR_PTR(-EFBIG);
1621
1622                 dp = get_dp(net, ovs_header->dp_ifindex);
1623                 if (!dp)
1624                         return ERR_PTR(-ENODEV);
1625
1626                 vport = ovs_vport_rtnl_rcu(dp, port_no);
1627                 if (!vport)
1628                         return ERR_PTR(-ENOENT);
1629                 return vport;
1630         } else
1631                 return ERR_PTR(-EINVAL);
1632 }
1633
1634 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1635 {
1636         struct nlattr **a = info->attrs;
1637         struct ovs_header *ovs_header = info->userhdr;
1638         struct vport_parms parms;
1639         struct sk_buff *reply;
1640         struct vport *vport;
1641         struct datapath *dp;
1642         u32 port_no;
1643         int err;
1644
1645         err = -EINVAL;
1646         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1647             !a[OVS_VPORT_ATTR_UPCALL_PID])
1648                 goto exit;
1649
1650         rtnl_lock();
1651         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1652         err = -ENODEV;
1653         if (!dp)
1654                 goto exit_unlock;
1655
1656         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1657                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1658
1659                 err = -EFBIG;
1660                 if (port_no >= DP_MAX_PORTS)
1661                         goto exit_unlock;
1662
1663                 vport = ovs_vport_rtnl_rcu(dp, port_no);
1664                 err = -EBUSY;
1665                 if (vport)
1666                         goto exit_unlock;
1667         } else {
1668                 for (port_no = 1; ; port_no++) {
1669                         if (port_no >= DP_MAX_PORTS) {
1670                                 err = -EFBIG;
1671                                 goto exit_unlock;
1672                         }
1673                         vport = ovs_vport_rtnl(dp, port_no);
1674                         if (!vport)
1675                                 break;
1676                 }
1677         }
1678
1679         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1680         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1681         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1682         parms.dp = dp;
1683         parms.port_no = port_no;
1684         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1685
1686         vport = new_vport(&parms);
1687         err = PTR_ERR(vport);
1688         if (IS_ERR(vport))
1689                 goto exit_unlock;
1690
1691         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1692                                          OVS_VPORT_CMD_NEW);
1693         if (IS_ERR(reply)) {
1694                 err = PTR_ERR(reply);
1695                 ovs_dp_detach_port(vport);
1696                 goto exit_unlock;
1697         }
1698         genl_notify(reply, genl_info_net(info), info->snd_portid,
1699                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1700
1701 exit_unlock:
1702         rtnl_unlock();
1703 exit:
1704         return err;
1705 }
1706
1707 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1708 {
1709         struct nlattr **a = info->attrs;
1710         struct sk_buff *reply;
1711         struct vport *vport;
1712         int err;
1713
1714         rtnl_lock();
1715         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1716         err = PTR_ERR(vport);
1717         if (IS_ERR(vport))
1718                 goto exit_unlock;
1719
1720         err = 0;
1721         if (a[OVS_VPORT_ATTR_TYPE] &&
1722             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1723                 err = -EINVAL;
1724
1725         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1726                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1727         if (err)
1728                 goto exit_unlock;
1729         if (a[OVS_VPORT_ATTR_UPCALL_PID])
1730                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1731
1732         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1733                                          OVS_VPORT_CMD_NEW);
1734         if (IS_ERR(reply)) {
1735                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1736                                 ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
1737                 goto exit_unlock;
1738         }
1739
1740         genl_notify(reply, genl_info_net(info), info->snd_portid,
1741                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1742
1743 exit_unlock:
1744         rtnl_unlock();
1745         return err;
1746 }
1747
1748 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1749 {
1750         struct nlattr **a = info->attrs;
1751         struct sk_buff *reply;
1752         struct vport *vport;
1753         int err;
1754
1755         rtnl_lock();
1756         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1757         err = PTR_ERR(vport);
1758         if (IS_ERR(vport))
1759                 goto exit_unlock;
1760
1761         if (vport->port_no == OVSP_LOCAL) {
1762                 err = -EINVAL;
1763                 goto exit_unlock;
1764         }
1765
1766         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1767                                          OVS_VPORT_CMD_DEL);
1768         err = PTR_ERR(reply);
1769         if (IS_ERR(reply))
1770                 goto exit_unlock;
1771
1772         ovs_dp_detach_port(vport);
1773
1774         genl_notify(reply, genl_info_net(info), info->snd_portid,
1775                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1776
1777 exit_unlock:
1778         rtnl_unlock();
1779         return err;
1780 }
1781
1782 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1783 {
1784         struct nlattr **a = info->attrs;
1785         struct ovs_header *ovs_header = info->userhdr;
1786         struct sk_buff *reply;
1787         struct vport *vport;
1788         int err;
1789
1790         rcu_read_lock();
1791         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1792         err = PTR_ERR(vport);
1793         if (IS_ERR(vport))
1794                 goto exit_unlock;
1795
1796         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1797                                          OVS_VPORT_CMD_NEW);
1798         err = PTR_ERR(reply);
1799         if (IS_ERR(reply))
1800                 goto exit_unlock;
1801
1802         rcu_read_unlock();
1803
1804         return genlmsg_reply(reply, info);
1805
1806 exit_unlock:
1807         rcu_read_unlock();
1808         return err;
1809 }
1810
1811 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1812 {
1813         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1814         struct datapath *dp;
1815         int bucket = cb->args[0], skip = cb->args[1];
1816         int i, j = 0;
1817
1818         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1819         if (!dp)
1820                 return -ENODEV;
1821
1822         rcu_read_lock();
1823         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1824                 struct vport *vport;
1825                 struct hlist_node *n;
1826
1827                 j = 0;
1828                 hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node) {
1829                         if (j >= skip &&
1830                             ovs_vport_cmd_fill_info(vport, skb,
1831                                                     NETLINK_CB(cb->skb).portid,
1832                                                     cb->nlh->nlmsg_seq,
1833                                                     NLM_F_MULTI,
1834                                                     OVS_VPORT_CMD_NEW) < 0)
1835                                 goto out;
1836
1837                         j++;
1838                 }
1839                 skip = 0;
1840         }
1841 out:
1842         rcu_read_unlock();
1843
1844         cb->args[0] = i;
1845         cb->args[1] = j;
1846
1847         return skb->len;
1848 }
1849
1850 static struct genl_ops dp_vport_genl_ops[] = {
1851         { .cmd = OVS_VPORT_CMD_NEW,
1852           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1853           .policy = vport_policy,
1854           .doit = ovs_vport_cmd_new
1855         },
1856         { .cmd = OVS_VPORT_CMD_DEL,
1857           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1858           .policy = vport_policy,
1859           .doit = ovs_vport_cmd_del
1860         },
1861         { .cmd = OVS_VPORT_CMD_GET,
1862           .flags = 0,               /* OK for unprivileged users. */
1863           .policy = vport_policy,
1864           .doit = ovs_vport_cmd_get,
1865           .dumpit = ovs_vport_cmd_dump
1866         },
1867         { .cmd = OVS_VPORT_CMD_SET,
1868           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1869           .policy = vport_policy,
1870           .doit = ovs_vport_cmd_set,
1871         },
1872 };
1873
1874 struct genl_family_and_ops {
1875         struct genl_family *family;
1876         struct genl_ops *ops;
1877         int n_ops;
1878         struct genl_multicast_group *group;
1879 };
1880
1881 static const struct genl_family_and_ops dp_genl_families[] = {
1882         { &dp_datapath_genl_family,
1883           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1884           &ovs_dp_datapath_multicast_group },
1885         { &dp_vport_genl_family,
1886           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1887           &ovs_dp_vport_multicast_group },
1888         { &dp_flow_genl_family,
1889           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1890           &ovs_dp_flow_multicast_group },
1891         { &dp_packet_genl_family,
1892           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1893           NULL },
1894 };
1895
1896 static void dp_unregister_genl(int n_families)
1897 {
1898         int i;
1899
1900         for (i = 0; i < n_families; i++)
1901                 genl_unregister_family(dp_genl_families[i].family);
1902 }
1903
1904 static int dp_register_genl(void)
1905 {
1906         int n_registered;
1907         int err;
1908         int i;
1909
1910         n_registered = 0;
1911         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1912                 const struct genl_family_and_ops *f = &dp_genl_families[i];
1913
1914                 err = genl_register_family_with_ops(f->family, f->ops,
1915                                                     f->n_ops);
1916                 if (err)
1917                         goto error;
1918                 n_registered++;
1919
1920                 if (f->group) {
1921                         err = genl_register_mc_group(f->family, f->group);
1922                         if (err)
1923                                 goto error;
1924                 }
1925         }
1926
1927         return 0;
1928
1929 error:
1930         dp_unregister_genl(n_registered);
1931         return err;
1932 }
1933
1934 static void rehash_flow_table(struct work_struct *work)
1935 {
1936         struct datapath *dp;
1937         struct net *net;
1938
1939         genl_lock();
1940         rtnl_lock();
1941         for_each_net(net) {
1942                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1943
1944                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1945                         struct flow_table *old_table = genl_dereference(dp->table);
1946                         struct flow_table *new_table;
1947
1948                         new_table = ovs_flow_tbl_rehash(old_table);
1949                         if (!IS_ERR(new_table)) {
1950                                 rcu_assign_pointer(dp->table, new_table);
1951                                 ovs_flow_tbl_deferred_destroy(old_table);
1952                         }
1953                 }
1954         }
1955         rtnl_unlock();
1956         genl_unlock();
1957
1958         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
1959 }
1960
1961 static int __net_init ovs_init_net(struct net *net)
1962 {
1963         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1964
1965         INIT_LIST_HEAD(&ovs_net->dps);
1966         return 0;
1967 }
1968
1969 static void __net_exit ovs_exit_net(struct net *net)
1970 {
1971         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1972         struct datapath *dp, *dp_next;
1973
1974         genl_lock();
1975         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1976                 __dp_destroy(dp);
1977         genl_unlock();
1978 }
1979
1980 static struct pernet_operations ovs_net_ops = {
1981         .init = ovs_init_net,
1982         .exit = ovs_exit_net,
1983         .id   = &ovs_net_id,
1984         .size = sizeof(struct ovs_net),
1985 };
1986
1987 static int __init dp_init(void)
1988 {
1989         struct sk_buff *dummy_skb;
1990         int err;
1991
1992         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
1993
1994         pr_info("Open vSwitch switching datapath\n");
1995
1996         err = ovs_flow_init();
1997         if (err)
1998                 goto error;
1999
2000         err = ovs_vport_init();
2001         if (err)
2002                 goto error_flow_exit;
2003
2004         err = register_pernet_device(&ovs_net_ops);
2005         if (err)
2006                 goto error_vport_exit;
2007
2008         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2009         if (err)
2010                 goto error_netns_exit;
2011
2012         err = dp_register_genl();
2013         if (err < 0)
2014                 goto error_unreg_notifier;
2015
2016         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2017
2018         return 0;
2019
2020 error_unreg_notifier:
2021         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2022 error_netns_exit:
2023         unregister_pernet_device(&ovs_net_ops);
2024 error_vport_exit:
2025         ovs_vport_exit();
2026 error_flow_exit:
2027         ovs_flow_exit();
2028 error:
2029         return err;
2030 }
2031
2032 static void dp_cleanup(void)
2033 {
2034         cancel_delayed_work_sync(&rehash_flow_wq);
2035         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2036         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2037         unregister_pernet_device(&ovs_net_ops);
2038         rcu_barrier();
2039         ovs_vport_exit();
2040         ovs_flow_exit();
2041 }
2042
2043 module_init(dp_init);
2044 module_exit(dp_cleanup);
2045
2046 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2047 MODULE_LICENSE("GPL");