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 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
  75                        struct genl_multicast_group *grp)
  76 {
  77         genl_notify(skb, genl_info_net(info), info->snd_portid,
  78                     grp->id, info->nlhdr, GFP_KERNEL);
  79 }
  80
  81 /**
  82  * DOC: Locking:
  83  *
  84  * Writes to device state (add/remove datapath, port, set operations on vports,
  85  * etc.) are protected by RTNL.
  86  *
  87  * Writes to other state (flow table modifications, set miscellaneous datapath
  88  * parameters, etc.) are protected by genl_mutex.  The RTNL lock nests inside
  89  * genl_mutex.
  90  *
  91  * Reads are protected by RCU.
  92  *
  93  * There are a few special cases (mostly stats) that have their own
  94  * synchronization but they nest under all of above and don't interact with
  95  * each other.
  96  */
  97
  98 static struct vport *new_vport(const struct vport_parms *);
  99 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
 100                              const struct dp_upcall_info *);
 101 static int queue_userspace_packet(struct net *, int dp_ifindex,
 102                                   struct sk_buff *,
 103                                   const struct dp_upcall_info *);
 104
 105 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
 106 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 107 {
 108         struct datapath *dp = NULL;
 109         struct net_device *dev;
 110
 111         rcu_read_lock();
 112         dev = dev_get_by_index_rcu(net, dp_ifindex);
 113         if (dev) {
 114                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 115                 if (vport)
 116                         dp = vport->dp;
 117         }
 118         rcu_read_unlock();
 119
 120         return dp;
 121 }
 122
 123 /* Must be called with rcu_read_lock or RTNL lock. */
 124 const char *ovs_dp_name(const struct datapath *dp)
 125 {
 126         struct vport *vport = ovs_vport_rtnl_rcu(dp, OVSP_LOCAL);
 127         return vport->ops->get_name(vport);
 128 }
 129
 130 static int get_dpifindex(struct datapath *dp)
 131 {
 132         struct vport *local;
 133         int ifindex;
 134
 135         rcu_read_lock();
 136
 137         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 138         if (local)
 139                 ifindex = local->ops->get_ifindex(local);
 140         else
 141                 ifindex = 0;
 142
 143         rcu_read_unlock();
 144
 145         return ifindex;
 146 }
 147
 148 static void destroy_dp_rcu(struct rcu_head *rcu)
 149 {
 150         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 151
 152         ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
 153         free_percpu(dp->stats_percpu);
 154         release_net(ovs_dp_get_net(dp));
 155         kfree(dp->ports);
 156         kfree(dp);
 157 }
 158
 159 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 160                                             u16 port_no)
 161 {
 162         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 163 }
 164
 165 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 166 {
 167         struct vport *vport;
 168         struct hlist_head *head;
 169
 170         head = vport_hash_bucket(dp, port_no);
 171         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 172                 if (vport->port_no == port_no)
 173                         return vport;
 174         }
 175         return NULL;
 176 }
 177
 178 /* Called with RTNL lock and genl_lock. */
 179 static struct vport *new_vport(const struct vport_parms *parms)
 180 {
 181         struct vport *vport;
 182
 183         vport = ovs_vport_add(parms);
 184         if (!IS_ERR(vport)) {
 185                 struct datapath *dp = parms->dp;
 186                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 187
 188                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 189         }
 190
 191         return vport;
 192 }
 193
 194 /* Called with RTNL lock. */
 195 void ovs_dp_detach_port(struct vport *p)
 196 {
 197         ASSERT_RTNL();
 198
 199         /* First drop references to device. */
 200         hlist_del_rcu(&p->dp_hash_node);
 201
 202         /* Then destroy it. */
 203         ovs_vport_del(p);
 204 }
 205
 206 /* Must be called with rcu_read_lock. */
 207 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 208 {
 209         struct datapath *dp = p->dp;
 210         struct sw_flow *flow;
 211         struct dp_stats_percpu *stats;
 212         struct sw_flow_key key;
 213         u64 *stats_counter;
 214         int error;
 215         int key_len;
 216
 217         stats = this_cpu_ptr(dp->stats_percpu);
 218
 219         /* Extract flow from 'skb' into 'key'. */
 220         error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
 221         if (unlikely(error)) {
 222                 kfree_skb(skb);
 223                 return;
 224         }
 225
 226         /* Look up flow. */
 227         flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
 228         if (unlikely(!flow)) {
 229                 struct dp_upcall_info upcall;
 230
 231                 upcall.cmd = OVS_PACKET_CMD_MISS;
 232                 upcall.key = &key;
 233                 upcall.userdata = NULL;
 234                 upcall.portid = p->upcall_portid;
 235                 ovs_dp_upcall(dp, skb, &upcall);
 236                 consume_skb(skb);
 237                 stats_counter = &stats->n_missed;
 238                 goto out;
 239         }
 240
 241         OVS_CB(skb)->flow = flow;
 242
 243         stats_counter = &stats->n_hit;
 244         ovs_flow_used(OVS_CB(skb)->flow, skb);
 245         ovs_execute_actions(dp, skb);
 246
 247 out:
 248         /* Update datapath statistics. */
 249         u64_stats_update_begin(&stats->sync);
 250         (*stats_counter)++;
 251         u64_stats_update_end(&stats->sync);
 252 }
 253
 254 static struct genl_family dp_packet_genl_family = {
 255         .id = GENL_ID_GENERATE,
 256         .hdrsize = sizeof(struct ovs_header),
 257         .name = OVS_PACKET_FAMILY,
 258         .version = OVS_PACKET_VERSION,
 259         .maxattr = OVS_PACKET_ATTR_MAX,
 260         .netnsok = true
 261 };
 262
 263 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 264                   const struct dp_upcall_info *upcall_info)
 265 {
 266         struct dp_stats_percpu *stats;
 267         int dp_ifindex;
 268         int err;
 269
 270         if (upcall_info->portid == 0) {
 271                 err = -ENOTCONN;
 272                 goto err;
 273         }
 274
 275         dp_ifindex = get_dpifindex(dp);
 276         if (!dp_ifindex) {
 277                 err = -ENODEV;
 278                 goto err;
 279         }
 280
 281         if (!skb_is_gso(skb))
 282                 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 283         else
 284                 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 285         if (err)
 286                 goto err;
 287
 288         return 0;
 289
 290 err:
 291         stats = this_cpu_ptr(dp->stats_percpu);
 292
 293         u64_stats_update_begin(&stats->sync);
 294         stats->n_lost++;
 295         u64_stats_update_end(&stats->sync);
 296
 297         return err;
 298 }
 299
 300 static int queue_gso_packets(struct net *net, int dp_ifindex,
 301                              struct sk_buff *skb,
 302                              const struct dp_upcall_info *upcall_info)
 303 {
 304         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 305         struct dp_upcall_info later_info;
 306         struct sw_flow_key later_key;
 307         struct sk_buff *segs, *nskb;
 308         int err;
 309
 310         segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
 311         if (IS_ERR(segs))
 312                 return PTR_ERR(segs);
 313
 314         /* Queue all of the segments. */
 315         skb = segs;
 316         do {
 317                 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
 318                 if (err)
 319                         break;
 320
 321                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 322                         /* The initial flow key extracted by ovs_flow_extract()
 323                          * in this case is for a first fragment, so we need to
 324                          * properly mark later fragments.
 325                          */
 326                         later_key = *upcall_info->key;
 327                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 328
 329                         later_info = *upcall_info;
 330                         later_info.key = &later_key;
 331                         upcall_info = &later_info;
 332                 }
 333         } while ((skb = skb->next));
 334
 335         /* Free all of the segments. */
 336         skb = segs;
 337         do {
 338                 nskb = skb->next;
 339                 if (err)
 340                         kfree_skb(skb);
 341                 else
 342                         consume_skb(skb);
 343         } while ((skb = nskb));
 344         return err;
 345 }
 346
 347 static size_t key_attr_size(void)
 348 {
 349         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 350                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 351                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 352                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 353                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 354                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 355                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 356                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 357                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 358                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 359                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 360 }
 361
 362 static size_t upcall_msg_size(const struct sk_buff *skb,
 363                               const struct nlattr *userdata)
 364 {
 365         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 366                 + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
 367                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 368
 369         /* OVS_PACKET_ATTR_USERDATA */
 370         if (userdata)
 371                 size += NLA_ALIGN(userdata->nla_len);
 372
 373         return size;
 374 }
 375
 376 static int queue_userspace_packet(struct net *net, int dp_ifindex,
 377                                   struct sk_buff *skb,
 378                                   const struct dp_upcall_info *upcall_info)
 379 {
 380         struct ovs_header *upcall;
 381         struct sk_buff *nskb = NULL;
 382         struct sk_buff *user_skb; /* to be queued to userspace */
 383         struct nlattr *nla;
 384         int err;
 385
 386         if (vlan_tx_tag_present(skb)) {
 387                 nskb = skb_clone(skb, GFP_ATOMIC);
 388                 if (!nskb)
 389                         return -ENOMEM;
 390
 391                 nskb = __vlan_put_tag(nskb, vlan_tx_tag_get(nskb));
 392                 if (!nskb)
 393                         return -ENOMEM;
 394
 395                 nskb->vlan_tci = 0;
 396                 skb = nskb;
 397         }
 398
 399         if (nla_attr_size(skb->len) > USHRT_MAX) {
 400                 err = -EFBIG;
 401                 goto out;
 402         }
 403
 404         user_skb = genlmsg_new(upcall_msg_size(skb, upcall_info->userdata), GFP_ATOMIC);
 405         if (!user_skb) {
 406                 err = -ENOMEM;
 407                 goto out;
 408         }
 409
 410         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 411                              0, upcall_info->cmd);
 412         upcall->dp_ifindex = dp_ifindex;
 413
 414         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 415         ovs_flow_to_nlattrs(upcall_info->key, user_skb);
 416         nla_nest_end(user_skb, nla);
 417
 418         if (upcall_info->userdata)
 419                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 420                           nla_len(upcall_info->userdata),
 421                           nla_data(upcall_info->userdata));
 422
 423         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 424
 425         skb_copy_and_csum_dev(skb, nla_data(nla));
 426
 427         genlmsg_end(user_skb, upcall);
 428         err = genlmsg_unicast(net, user_skb, upcall_info->portid);
 429
 430 out:
 431         kfree_skb(nskb);
 432         return err;
 433 }
 434
 435 /* Called with genl_mutex. */
 436 static int flush_flows(struct datapath *dp)
 437 {
 438         struct flow_table *old_table;
 439         struct flow_table *new_table;
 440
 441         old_table = genl_dereference(dp->table);
 442         new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
 443         if (!new_table)
 444                 return -ENOMEM;
 445
 446         rcu_assign_pointer(dp->table, new_table);
 447
 448         ovs_flow_tbl_deferred_destroy(old_table);
 449         return 0;
 450 }
 451
 452 static int validate_actions(const struct nlattr *attr,
 453                                 const struct sw_flow_key *key, int depth);
 454
 455 static int validate_sample(const struct nlattr *attr,
 456                                 const struct sw_flow_key *key, int depth)
 457 {
 458         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 459         const struct nlattr *probability, *actions;
 460         const struct nlattr *a;
 461         int rem;
 462
 463         memset(attrs, 0, sizeof(attrs));
 464         nla_for_each_nested(a, attr, rem) {
 465                 int type = nla_type(a);
 466                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 467                         return -EINVAL;
 468                 attrs[type] = a;
 469         }
 470         if (rem)
 471                 return -EINVAL;
 472
 473         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 474         if (!probability || nla_len(probability) != sizeof(u32))
 475                 return -EINVAL;
 476
 477         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 478         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 479                 return -EINVAL;
 480         return validate_actions(actions, key, depth + 1);
 481 }
 482
 483 static int validate_tp_port(const struct sw_flow_key *flow_key)
 484 {
 485         if (flow_key->eth.type == htons(ETH_P_IP)) {
 486                 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
 487                         return 0;
 488         } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
 489                 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
 490                         return 0;
 491         }
 492
 493         return -EINVAL;
 494 }
 495
 496 static int validate_set(const struct nlattr *a,
 497                         const struct sw_flow_key *flow_key)
 498 {
 499         const struct nlattr *ovs_key = nla_data(a);
 500         int key_type = nla_type(ovs_key);
 501
 502         /* There can be only one key in a action */
 503         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 504                 return -EINVAL;
 505
 506         if (key_type > OVS_KEY_ATTR_MAX ||
 507             nla_len(ovs_key) != ovs_key_lens[key_type])
 508                 return -EINVAL;
 509
 510         switch (key_type) {
 511         const struct ovs_key_ipv4 *ipv4_key;
 512         const struct ovs_key_ipv6 *ipv6_key;
 513
 514         case OVS_KEY_ATTR_PRIORITY:
 515         case OVS_KEY_ATTR_SKB_MARK:
 516         case OVS_KEY_ATTR_ETHERNET:
 517                 break;
 518
 519         case OVS_KEY_ATTR_IPV4:
 520                 if (flow_key->eth.type != htons(ETH_P_IP))
 521                         return -EINVAL;
 522
 523                 if (!flow_key->ip.proto)
 524                         return -EINVAL;
 525
 526                 ipv4_key = nla_data(ovs_key);
 527                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 528                         return -EINVAL;
 529
 530                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 531                         return -EINVAL;
 532
 533                 break;
 534
 535         case OVS_KEY_ATTR_IPV6:
 536                 if (flow_key->eth.type != htons(ETH_P_IPV6))
 537                         return -EINVAL;
 538
 539                 if (!flow_key->ip.proto)
 540                         return -EINVAL;
 541
 542                 ipv6_key = nla_data(ovs_key);
 543                 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
 544                         return -EINVAL;
 545
 546                 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
 547                         return -EINVAL;
 548
 549                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
 550                         return -EINVAL;
 551
 552                 break;
 553
 554         case OVS_KEY_ATTR_TCP:
 555                 if (flow_key->ip.proto != IPPROTO_TCP)
 556                         return -EINVAL;
 557
 558                 return validate_tp_port(flow_key);
 559
 560         case OVS_KEY_ATTR_UDP:
 561                 if (flow_key->ip.proto != IPPROTO_UDP)
 562                         return -EINVAL;
 563
 564                 return validate_tp_port(flow_key);
 565
 566         default:
 567                 return -EINVAL;
 568         }
 569
 570         return 0;
 571 }
 572
 573 static int validate_userspace(const struct nlattr *attr)
 574 {
 575         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 576                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 577                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
 578         };
 579         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 580         int error;
 581
 582         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 583                                  attr, userspace_policy);
 584         if (error)
 585                 return error;
 586
 587         if (!a[OVS_USERSPACE_ATTR_PID] ||
 588             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 589                 return -EINVAL;
 590
 591         return 0;
 592 }
 593
 594 static int validate_actions(const struct nlattr *attr,
 595                                 const struct sw_flow_key *key,  int depth)
 596 {
 597         const struct nlattr *a;
 598         int rem, err;
 599
 600         if (depth >= SAMPLE_ACTION_DEPTH)
 601                 return -EOVERFLOW;
 602
 603         nla_for_each_nested(a, attr, rem) {
 604                 /* Expected argument lengths, (u32)-1 for variable length. */
 605                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 606                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 607                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 608                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 609                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 610                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 611                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 612                 };
 613                 const struct ovs_action_push_vlan *vlan;
 614                 int type = nla_type(a);
 615
 616                 if (type > OVS_ACTION_ATTR_MAX ||
 617                     (action_lens[type] != nla_len(a) &&
 618                      action_lens[type] != (u32)-1))
 619                         return -EINVAL;
 620
 621                 switch (type) {
 622                 case OVS_ACTION_ATTR_UNSPEC:
 623                         return -EINVAL;
 624
 625                 case OVS_ACTION_ATTR_USERSPACE:
 626                         err = validate_userspace(a);
 627                         if (err)
 628                                 return err;
 629                         break;
 630
 631                 case OVS_ACTION_ATTR_OUTPUT:
 632                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 633                                 return -EINVAL;
 634                         break;
 635
 636
 637                 case OVS_ACTION_ATTR_POP_VLAN:
 638                         break;
 639
 640                 case OVS_ACTION_ATTR_PUSH_VLAN:
 641                         vlan = nla_data(a);
 642                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 643                                 return -EINVAL;
 644                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 645                                 return -EINVAL;
 646                         break;
 647
 648                 case OVS_ACTION_ATTR_SET:
 649                         err = validate_set(a, key);
 650                         if (err)
 651                                 return err;
 652                         break;
 653
 654                 case OVS_ACTION_ATTR_SAMPLE:
 655                         err = validate_sample(a, key, depth);
 656                         if (err)
 657                                 return err;
 658                         break;
 659
 660                 default:
 661                         return -EINVAL;
 662                 }
 663         }
 664
 665         if (rem > 0)
 666                 return -EINVAL;
 667
 668         return 0;
 669 }
 670
 671 static void clear_stats(struct sw_flow *flow)
 672 {
 673         flow->used = 0;
 674         flow->tcp_flags = 0;
 675         flow->packet_count = 0;
 676         flow->byte_count = 0;
 677 }
 678
 679 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 680 {
 681         struct ovs_header *ovs_header = info->userhdr;
 682         struct nlattr **a = info->attrs;
 683         struct sw_flow_actions *acts;
 684         struct sk_buff *packet;
 685         struct sw_flow *flow;
 686         struct datapath *dp;
 687         struct ethhdr *eth;
 688         int len;
 689         int err;
 690         int key_len;
 691
 692         err = -EINVAL;
 693         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 694             !a[OVS_PACKET_ATTR_ACTIONS])
 695                 goto err;
 696
 697         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 698         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 699         err = -ENOMEM;
 700         if (!packet)
 701                 goto err;
 702         skb_reserve(packet, NET_IP_ALIGN);
 703
 704         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 705
 706         skb_reset_mac_header(packet);
 707         eth = eth_hdr(packet);
 708
 709         /* Normally, setting the skb 'protocol' field would be handled by a
 710          * call to eth_type_trans(), but it assumes there's a sending
 711          * device, which we may not have. */
 712         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 713                 packet->protocol = eth->h_proto;
 714         else
 715                 packet->protocol = htons(ETH_P_802_2);
 716
 717         /* Build an sw_flow for sending this packet. */
 718         flow = ovs_flow_alloc();
 719         err = PTR_ERR(flow);
 720         if (IS_ERR(flow))
 721                 goto err_kfree_skb;
 722
 723         err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
 724         if (err)
 725                 goto err_flow_free;
 726
 727         err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
 728                                              &flow->key.phy.skb_mark,
 729                                              &flow->key.phy.in_port,
 730                                              a[OVS_PACKET_ATTR_KEY]);
 731         if (err)
 732                 goto err_flow_free;
 733
 734         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
 735         if (err)
 736                 goto err_flow_free;
 737
 738         flow->hash = ovs_flow_hash(&flow->key, key_len);
 739
 740         acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 741         err = PTR_ERR(acts);
 742         if (IS_ERR(acts))
 743                 goto err_flow_free;
 744         rcu_assign_pointer(flow->sf_acts, acts);
 745
 746         OVS_CB(packet)->flow = flow;
 747         packet->priority = flow->key.phy.priority;
 748         packet->mark = flow->key.phy.skb_mark;
 749
 750         rcu_read_lock();
 751         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 752         err = -ENODEV;
 753         if (!dp)
 754                 goto err_unlock;
 755
 756         local_bh_disable();
 757         err = ovs_execute_actions(dp, packet);
 758         local_bh_enable();
 759         rcu_read_unlock();
 760
 761         ovs_flow_free(flow);
 762         return err;
 763
 764 err_unlock:
 765         rcu_read_unlock();
 766 err_flow_free:
 767         ovs_flow_free(flow);
 768 err_kfree_skb:
 769         kfree_skb(packet);
 770 err:
 771         return err;
 772 }
 773
 774 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 775         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 776         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 777         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 778 };
 779
 780 static struct genl_ops dp_packet_genl_ops[] = {
 781         { .cmd = OVS_PACKET_CMD_EXECUTE,
 782           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 783           .policy = packet_policy,
 784           .doit = ovs_packet_cmd_execute
 785         }
 786 };
 787
 788 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 789 {
 790         int i;
 791         struct flow_table *table = genl_dereference(dp->table);
 792
 793         stats->n_flows = ovs_flow_tbl_count(table);
 794
 795         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 796         for_each_possible_cpu(i) {
 797                 const struct dp_stats_percpu *percpu_stats;
 798                 struct dp_stats_percpu local_stats;
 799                 unsigned int start;
 800
 801                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 802
 803                 do {
 804                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 805                         local_stats = *percpu_stats;
 806                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 807
 808                 stats->n_hit += local_stats.n_hit;
 809                 stats->n_missed += local_stats.n_missed;
 810                 stats->n_lost += local_stats.n_lost;
 811         }
 812 }
 813
 814 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 815         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 816         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 817         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 818 };
 819
 820 static struct genl_family dp_flow_genl_family = {
 821         .id = GENL_ID_GENERATE,
 822         .hdrsize = sizeof(struct ovs_header),
 823         .name = OVS_FLOW_FAMILY,
 824         .version = OVS_FLOW_VERSION,
 825         .maxattr = OVS_FLOW_ATTR_MAX,
 826         .netnsok = true
 827 };
 828
 829 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 830         .name = OVS_FLOW_MCGROUP
 831 };
 832
 833 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 834 {
 835         return NLMSG_ALIGN(sizeof(struct ovs_header))
 836                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 837                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 838                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 839                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 840                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 841 }
 842
 843 /* Called with genl_lock. */
 844 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 845                                   struct sk_buff *skb, u32 portid,
 846                                   u32 seq, u32 flags, u8 cmd)
 847 {
 848         const int skb_orig_len = skb->len;
 849         const struct sw_flow_actions *sf_acts;
 850         struct ovs_flow_stats stats;
 851         struct ovs_header *ovs_header;
 852         struct nlattr *nla;
 853         unsigned long used;
 854         u8 tcp_flags;
 855         int err;
 856
 857         sf_acts = rcu_dereference_protected(flow->sf_acts,
 858                                             lockdep_genl_is_held());
 859
 860         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 861         if (!ovs_header)
 862                 return -EMSGSIZE;
 863
 864         ovs_header->dp_ifindex = get_dpifindex(dp);
 865
 866         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 867         if (!nla)
 868                 goto nla_put_failure;
 869         err = ovs_flow_to_nlattrs(&flow->key, skb);
 870         if (err)
 871                 goto error;
 872         nla_nest_end(skb, nla);
 873
 874         spin_lock_bh(&flow->lock);
 875         used = flow->used;
 876         stats.n_packets = flow->packet_count;
 877         stats.n_bytes = flow->byte_count;
 878         tcp_flags = flow->tcp_flags;
 879         spin_unlock_bh(&flow->lock);
 880
 881         if (used &&
 882             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 883                 goto nla_put_failure;
 884
 885         if (stats.n_packets &&
 886             nla_put(skb, OVS_FLOW_ATTR_STATS,
 887                     sizeof(struct ovs_flow_stats), &stats))
 888                 goto nla_put_failure;
 889
 890         if (tcp_flags &&
 891             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
 892                 goto nla_put_failure;
 893
 894         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 895          * this is the first flow to be dumped into 'skb'.  This is unusual for
 896          * Netlink but individual action lists can be longer than
 897          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 898          * The userspace caller can always fetch the actions separately if it
 899          * really wants them.  (Most userspace callers in fact don't care.)
 900          *
 901          * This can only fail for dump operations because the skb is always
 902          * properly sized for single flows.
 903          */
 904         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 905                       sf_acts->actions);
 906         if (err < 0 && skb_orig_len)
 907                 goto error;
 908
 909         return genlmsg_end(skb, ovs_header);
 910
 911 nla_put_failure:
 912         err = -EMSGSIZE;
 913 error:
 914         genlmsg_cancel(skb, ovs_header);
 915         return err;
 916 }
 917
 918 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 919 {
 920         const struct sw_flow_actions *sf_acts;
 921
 922         sf_acts = rcu_dereference_protected(flow->sf_acts,
 923                                             lockdep_genl_is_held());
 924
 925         return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
 926 }
 927
 928 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 929                                                struct datapath *dp,
 930                                                u32 portid, u32 seq, u8 cmd)
 931 {
 932         struct sk_buff *skb;
 933         int retval;
 934
 935         skb = ovs_flow_cmd_alloc_info(flow);
 936         if (!skb)
 937                 return ERR_PTR(-ENOMEM);
 938
 939         retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
 940         BUG_ON(retval < 0);
 941         return skb;
 942 }
 943
 944 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 945 {
 946         struct nlattr **a = info->attrs;
 947         struct ovs_header *ovs_header = info->userhdr;
 948         struct sw_flow_key key;
 949         struct sw_flow *flow;
 950         struct sk_buff *reply;
 951         struct datapath *dp;
 952         struct flow_table *table;
 953         int error;
 954         int key_len;
 955
 956         /* Extract key. */
 957         error = -EINVAL;
 958         if (!a[OVS_FLOW_ATTR_KEY])
 959                 goto error;
 960         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
 961         if (error)
 962                 goto error;
 963
 964         /* Validate actions. */
 965         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 966                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
 967                 if (error)
 968                         goto error;
 969         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 970                 error = -EINVAL;
 971                 goto error;
 972         }
 973
 974         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 975         error = -ENODEV;
 976         if (!dp)
 977                 goto error;
 978
 979         table = genl_dereference(dp->table);
 980         flow = ovs_flow_tbl_lookup(table, &key, key_len);
 981         if (!flow) {
 982                 struct sw_flow_actions *acts;
 983
 984                 /* Bail out if we're not allowed to create a new flow. */
 985                 error = -ENOENT;
 986                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
 987                         goto error;
 988
 989                 /* Expand table, if necessary, to make room. */
 990                 if (ovs_flow_tbl_need_to_expand(table)) {
 991                         struct flow_table *new_table;
 992
 993                         new_table = ovs_flow_tbl_expand(table);
 994                         if (!IS_ERR(new_table)) {
 995                                 rcu_assign_pointer(dp->table, new_table);
 996                                 ovs_flow_tbl_deferred_destroy(table);
 997                                 table = genl_dereference(dp->table);
 998                         }
 999                 }
1000
1001                 /* Allocate flow. */
1002                 flow = ovs_flow_alloc();
1003                 if (IS_ERR(flow)) {
1004                         error = PTR_ERR(flow);
1005                         goto error;
1006                 }
1007                 flow->key = key;
1008                 clear_stats(flow);
1009
1010                 /* Obtain actions. */
1011                 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1012                 error = PTR_ERR(acts);
1013                 if (IS_ERR(acts))
1014                         goto error_free_flow;
1015                 rcu_assign_pointer(flow->sf_acts, acts);
1016
1017                 /* Put flow in bucket. */
1018                 flow->hash = ovs_flow_hash(&key, key_len);
1019                 ovs_flow_tbl_insert(table, flow);
1020
1021                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1022                                                 info->snd_seq,
1023                                                 OVS_FLOW_CMD_NEW);
1024         } else {
1025                 /* We found a matching flow. */
1026                 struct sw_flow_actions *old_acts;
1027                 struct nlattr *acts_attrs;
1028
1029                 /* Bail out if we're not allowed to modify an existing flow.
1030                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1031                  * because Generic Netlink treats the latter as a dump
1032                  * request.  We also accept NLM_F_EXCL in case that bug ever
1033                  * gets fixed.
1034                  */
1035                 error = -EEXIST;
1036                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1037                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1038                         goto error;
1039
1040                 /* Update actions. */
1041                 old_acts = rcu_dereference_protected(flow->sf_acts,
1042                                                      lockdep_genl_is_held());
1043                 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1044                 if (acts_attrs &&
1045                    (old_acts->actions_len != nla_len(acts_attrs) ||
1046                    memcmp(old_acts->actions, nla_data(acts_attrs),
1047                           old_acts->actions_len))) {
1048                         struct sw_flow_actions *new_acts;
1049
1050                         new_acts = ovs_flow_actions_alloc(acts_attrs);
1051                         error = PTR_ERR(new_acts);
1052                         if (IS_ERR(new_acts))
1053                                 goto error;
1054
1055                         rcu_assign_pointer(flow->sf_acts, new_acts);
1056                         ovs_flow_deferred_free_acts(old_acts);
1057                 }
1058
1059                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1060                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1061
1062                 /* Clear stats. */
1063                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1064                         spin_lock_bh(&flow->lock);
1065                         clear_stats(flow);
1066                         spin_unlock_bh(&flow->lock);
1067                 }
1068         }
1069
1070         if (!IS_ERR(reply))
1071                 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1072         else
1073                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1074                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1075         return 0;
1076
1077 error_free_flow:
1078         ovs_flow_free(flow);
1079 error:
1080         return error;
1081 }
1082
1083 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1084 {
1085         struct nlattr **a = info->attrs;
1086         struct ovs_header *ovs_header = info->userhdr;
1087         struct sw_flow_key key;
1088         struct sk_buff *reply;
1089         struct sw_flow *flow;
1090         struct datapath *dp;
1091         struct flow_table *table;
1092         int err;
1093         int key_len;
1094
1095         if (!a[OVS_FLOW_ATTR_KEY])
1096                 return -EINVAL;
1097         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1098         if (err)
1099                 return err;
1100
1101         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1102         if (!dp)
1103                 return -ENODEV;
1104
1105         table = genl_dereference(dp->table);
1106         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1107         if (!flow)
1108                 return -ENOENT;
1109
1110         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1111                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1112         if (IS_ERR(reply))
1113                 return PTR_ERR(reply);
1114
1115         return genlmsg_reply(reply, info);
1116 }
1117
1118 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1119 {
1120         struct nlattr **a = info->attrs;
1121         struct ovs_header *ovs_header = info->userhdr;
1122         struct sw_flow_key key;
1123         struct sk_buff *reply;
1124         struct sw_flow *flow;
1125         struct datapath *dp;
1126         struct flow_table *table;
1127         int err;
1128         int key_len;
1129
1130         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1131         if (!dp)
1132                 return -ENODEV;
1133
1134         if (!a[OVS_FLOW_ATTR_KEY])
1135                 return flush_flows(dp);
1136
1137         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1138         if (err)
1139                 return err;
1140
1141         table = genl_dereference(dp->table);
1142         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1143         if (!flow)
1144                 return -ENOENT;
1145
1146         reply = ovs_flow_cmd_alloc_info(flow);
1147         if (!reply)
1148                 return -ENOMEM;
1149
1150         ovs_flow_tbl_remove(table, flow);
1151
1152         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1153                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1154         BUG_ON(err < 0);
1155
1156         ovs_flow_deferred_free(flow);
1157
1158         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1159         return 0;
1160 }
1161
1162 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1163 {
1164         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1165         struct datapath *dp;
1166         struct flow_table *table;
1167
1168         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1169         if (!dp)
1170                 return -ENODEV;
1171
1172         table = genl_dereference(dp->table);
1173
1174         for (;;) {
1175                 struct sw_flow *flow;
1176                 u32 bucket, obj;
1177
1178                 bucket = cb->args[0];
1179                 obj = cb->args[1];
1180                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1181                 if (!flow)
1182                         break;
1183
1184                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1185                                            NETLINK_CB(cb->skb).portid,
1186                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1187                                            OVS_FLOW_CMD_NEW) < 0)
1188                         break;
1189
1190                 cb->args[0] = bucket;
1191                 cb->args[1] = obj;
1192         }
1193         return skb->len;
1194 }
1195
1196 static struct genl_ops dp_flow_genl_ops[] = {
1197         { .cmd = OVS_FLOW_CMD_NEW,
1198           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1199           .policy = flow_policy,
1200           .doit = ovs_flow_cmd_new_or_set
1201         },
1202         { .cmd = OVS_FLOW_CMD_DEL,
1203           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1204           .policy = flow_policy,
1205           .doit = ovs_flow_cmd_del
1206         },
1207         { .cmd = OVS_FLOW_CMD_GET,
1208           .flags = 0,               /* OK for unprivileged users. */
1209           .policy = flow_policy,
1210           .doit = ovs_flow_cmd_get,
1211           .dumpit = ovs_flow_cmd_dump
1212         },
1213         { .cmd = OVS_FLOW_CMD_SET,
1214           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1215           .policy = flow_policy,
1216           .doit = ovs_flow_cmd_new_or_set,
1217         },
1218 };
1219
1220 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1221         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1222         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1223 };
1224
1225 static struct genl_family dp_datapath_genl_family = {
1226         .id = GENL_ID_GENERATE,
1227         .hdrsize = sizeof(struct ovs_header),
1228         .name = OVS_DATAPATH_FAMILY,
1229         .version = OVS_DATAPATH_VERSION,
1230         .maxattr = OVS_DP_ATTR_MAX,
1231         .netnsok = true
1232 };
1233
1234 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1235         .name = OVS_DATAPATH_MCGROUP
1236 };
1237
1238 static size_t ovs_dp_cmd_msg_size(void)
1239 {
1240         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1241
1242         msgsize += nla_total_size(IFNAMSIZ);
1243         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1244
1245         return msgsize;
1246 }
1247
1248 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1249                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1250 {
1251         struct ovs_header *ovs_header;
1252         struct ovs_dp_stats dp_stats;
1253         int err;
1254
1255         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1256                                    flags, cmd);
1257         if (!ovs_header)
1258                 goto error;
1259
1260         ovs_header->dp_ifindex = get_dpifindex(dp);
1261
1262         rcu_read_lock();
1263         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1264         rcu_read_unlock();
1265         if (err)
1266                 goto nla_put_failure;
1267
1268         get_dp_stats(dp, &dp_stats);
1269         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1270                 goto nla_put_failure;
1271
1272         return genlmsg_end(skb, ovs_header);
1273
1274 nla_put_failure:
1275         genlmsg_cancel(skb, ovs_header);
1276 error:
1277         return -EMSGSIZE;
1278 }
1279
1280 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1281                                              u32 seq, u8 cmd)
1282 {
1283         struct sk_buff *skb;
1284         int retval;
1285
1286         skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1287         if (!skb)
1288                 return ERR_PTR(-ENOMEM);
1289
1290         retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1291         if (retval < 0) {
1292                 kfree_skb(skb);
1293                 return ERR_PTR(retval);
1294         }
1295         return skb;
1296 }
1297
1298 /* Called with genl_mutex and optionally with RTNL lock also. */
1299 static struct datapath *lookup_datapath(struct net *net,
1300                                         struct ovs_header *ovs_header,
1301                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1302 {
1303         struct datapath *dp;
1304
1305         if (!a[OVS_DP_ATTR_NAME])
1306                 dp = get_dp(net, ovs_header->dp_ifindex);
1307         else {
1308                 struct vport *vport;
1309
1310                 rcu_read_lock();
1311                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1312                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1313                 rcu_read_unlock();
1314         }
1315         return dp ? dp : ERR_PTR(-ENODEV);
1316 }
1317
1318 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1319 {
1320         struct nlattr **a = info->attrs;
1321         struct vport_parms parms;
1322         struct sk_buff *reply;
1323         struct datapath *dp;
1324         struct vport *vport;
1325         struct ovs_net *ovs_net;
1326         int err, i;
1327
1328         err = -EINVAL;
1329         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1330                 goto err;
1331
1332         rtnl_lock();
1333
1334         err = -ENOMEM;
1335         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1336         if (dp == NULL)
1337                 goto err_unlock_rtnl;
1338
1339         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1340
1341         /* Allocate table. */
1342         err = -ENOMEM;
1343         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1344         if (!dp->table)
1345                 goto err_free_dp;
1346
1347         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1348         if (!dp->stats_percpu) {
1349                 err = -ENOMEM;
1350                 goto err_destroy_table;
1351         }
1352
1353         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1354                         GFP_KERNEL);
1355         if (!dp->ports) {
1356                 err = -ENOMEM;
1357                 goto err_destroy_percpu;
1358         }
1359
1360         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1361                 INIT_HLIST_HEAD(&dp->ports[i]);
1362
1363         /* Set up our datapath device. */
1364         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1365         parms.type = OVS_VPORT_TYPE_INTERNAL;
1366         parms.options = NULL;
1367         parms.dp = dp;
1368         parms.port_no = OVSP_LOCAL;
1369         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1370
1371         vport = new_vport(&parms);
1372         if (IS_ERR(vport)) {
1373                 err = PTR_ERR(vport);
1374                 if (err == -EBUSY)
1375                         err = -EEXIST;
1376
1377                 goto err_destroy_ports_array;
1378         }
1379
1380         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1381                                       info->snd_seq, OVS_DP_CMD_NEW);
1382         err = PTR_ERR(reply);
1383         if (IS_ERR(reply))
1384                 goto err_destroy_local_port;
1385
1386         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1387         list_add_tail(&dp->list_node, &ovs_net->dps);
1388         rtnl_unlock();
1389
1390         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1391         return 0;
1392
1393 err_destroy_local_port:
1394         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1395 err_destroy_ports_array:
1396         kfree(dp->ports);
1397 err_destroy_percpu:
1398         free_percpu(dp->stats_percpu);
1399 err_destroy_table:
1400         ovs_flow_tbl_destroy(genl_dereference(dp->table));
1401 err_free_dp:
1402         release_net(ovs_dp_get_net(dp));
1403         kfree(dp);
1404 err_unlock_rtnl:
1405         rtnl_unlock();
1406 err:
1407         return err;
1408 }
1409
1410 /* Called with genl_mutex. */
1411 static void __dp_destroy(struct datapath *dp)
1412 {
1413         int i;
1414
1415         rtnl_lock();
1416
1417         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1418                 struct vport *vport;
1419                 struct hlist_node *n;
1420
1421                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1422                         if (vport->port_no != OVSP_LOCAL)
1423                                 ovs_dp_detach_port(vport);
1424         }
1425
1426         list_del(&dp->list_node);
1427         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1428
1429         /* rtnl_unlock() will wait until all the references to devices that
1430          * are pending unregistration have been dropped.  We do it here to
1431          * ensure that any internal devices (which contain DP pointers) are
1432          * fully destroyed before freeing the datapath.
1433          */
1434         rtnl_unlock();
1435
1436         call_rcu(&dp->rcu, destroy_dp_rcu);
1437 }
1438
1439 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1440 {
1441         struct sk_buff *reply;
1442         struct datapath *dp;
1443         int err;
1444
1445         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1446         err = PTR_ERR(dp);
1447         if (IS_ERR(dp))
1448                 return err;
1449
1450         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1451                                       info->snd_seq, OVS_DP_CMD_DEL);
1452         err = PTR_ERR(reply);
1453         if (IS_ERR(reply))
1454                 return err;
1455
1456         __dp_destroy(dp);
1457
1458         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1459
1460         return 0;
1461 }
1462
1463 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1464 {
1465         struct sk_buff *reply;
1466         struct datapath *dp;
1467         int err;
1468
1469         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1470         if (IS_ERR(dp))
1471                 return PTR_ERR(dp);
1472
1473         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1474                                       info->snd_seq, OVS_DP_CMD_NEW);
1475         if (IS_ERR(reply)) {
1476                 err = PTR_ERR(reply);
1477                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1478                                 ovs_dp_datapath_multicast_group.id, err);
1479                 return 0;
1480         }
1481
1482         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1483
1484         return 0;
1485 }
1486
1487 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1488 {
1489         struct sk_buff *reply;
1490         struct datapath *dp;
1491
1492         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1493         if (IS_ERR(dp))
1494                 return PTR_ERR(dp);
1495
1496         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1497                                       info->snd_seq, OVS_DP_CMD_NEW);
1498         if (IS_ERR(reply))
1499                 return PTR_ERR(reply);
1500
1501         return genlmsg_reply(reply, info);
1502 }
1503
1504 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1505 {
1506         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1507         struct datapath *dp;
1508         int skip = cb->args[0];
1509         int i = 0;
1510
1511         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1512                 if (i >= skip &&
1513                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1514                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1515                                          OVS_DP_CMD_NEW) < 0)
1516                         break;
1517                 i++;
1518         }
1519
1520         cb->args[0] = i;
1521
1522         return skb->len;
1523 }
1524
1525 static struct genl_ops dp_datapath_genl_ops[] = {
1526         { .cmd = OVS_DP_CMD_NEW,
1527           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1528           .policy = datapath_policy,
1529           .doit = ovs_dp_cmd_new
1530         },
1531         { .cmd = OVS_DP_CMD_DEL,
1532           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1533           .policy = datapath_policy,
1534           .doit = ovs_dp_cmd_del
1535         },
1536         { .cmd = OVS_DP_CMD_GET,
1537           .flags = 0,               /* OK for unprivileged users. */
1538           .policy = datapath_policy,
1539           .doit = ovs_dp_cmd_get,
1540           .dumpit = ovs_dp_cmd_dump
1541         },
1542         { .cmd = OVS_DP_CMD_SET,
1543           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1544           .policy = datapath_policy,
1545           .doit = ovs_dp_cmd_set,
1546         },
1547 };
1548
1549 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1550         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1551         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1552         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1553         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1554         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1555         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1556 };
1557
1558 static struct genl_family dp_vport_genl_family = {
1559         .id = GENL_ID_GENERATE,
1560         .hdrsize = sizeof(struct ovs_header),
1561         .name = OVS_VPORT_FAMILY,
1562         .version = OVS_VPORT_VERSION,
1563         .maxattr = OVS_VPORT_ATTR_MAX,
1564         .netnsok = true
1565 };
1566
1567 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1568         .name = OVS_VPORT_MCGROUP
1569 };
1570
1571 /* Called with RTNL lock or RCU read lock. */
1572 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1573                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1574 {
1575         struct ovs_header *ovs_header;
1576         struct ovs_vport_stats vport_stats;
1577         int err;
1578
1579         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1580                                  flags, cmd);
1581         if (!ovs_header)
1582                 return -EMSGSIZE;
1583
1584         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1585
1586         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1587             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1588             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1589             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1590                 goto nla_put_failure;
1591
1592         ovs_vport_get_stats(vport, &vport_stats);
1593         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1594                     &vport_stats))
1595                 goto nla_put_failure;
1596
1597         err = ovs_vport_get_options(vport, skb);
1598         if (err == -EMSGSIZE)
1599                 goto error;
1600
1601         return genlmsg_end(skb, ovs_header);
1602
1603 nla_put_failure:
1604         err = -EMSGSIZE;
1605 error:
1606         genlmsg_cancel(skb, ovs_header);
1607         return err;
1608 }
1609
1610 /* Called with RTNL lock or RCU read lock. */
1611 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1612                                          u32 seq, u8 cmd)
1613 {
1614         struct sk_buff *skb;
1615         int retval;
1616
1617         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1618         if (!skb)
1619                 return ERR_PTR(-ENOMEM);
1620
1621         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1622         if (retval < 0) {
1623                 kfree_skb(skb);
1624                 return ERR_PTR(retval);
1625         }
1626         return skb;
1627 }
1628
1629 /* Called with RTNL lock or RCU read lock. */
1630 static struct vport *lookup_vport(struct net *net,
1631                                   struct ovs_header *ovs_header,
1632                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1633 {
1634         struct datapath *dp;
1635         struct vport *vport;
1636
1637         if (a[OVS_VPORT_ATTR_NAME]) {
1638                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1639                 if (!vport)
1640                         return ERR_PTR(-ENODEV);
1641                 if (ovs_header->dp_ifindex &&
1642                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1643                         return ERR_PTR(-ENODEV);
1644                 return vport;
1645         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1646                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1647
1648                 if (port_no >= DP_MAX_PORTS)
1649                         return ERR_PTR(-EFBIG);
1650
1651                 dp = get_dp(net, ovs_header->dp_ifindex);
1652                 if (!dp)
1653                         return ERR_PTR(-ENODEV);
1654
1655                 vport = ovs_vport_rtnl_rcu(dp, port_no);
1656                 if (!vport)
1657                         return ERR_PTR(-ENODEV);
1658                 return vport;
1659         } else
1660                 return ERR_PTR(-EINVAL);
1661 }
1662
1663 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1664 {
1665         struct nlattr **a = info->attrs;
1666         struct ovs_header *ovs_header = info->userhdr;
1667         struct vport_parms parms;
1668         struct sk_buff *reply;
1669         struct vport *vport;
1670         struct datapath *dp;
1671         u32 port_no;
1672         int err;
1673
1674         err = -EINVAL;
1675         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1676             !a[OVS_VPORT_ATTR_UPCALL_PID])
1677                 goto exit;
1678
1679         rtnl_lock();
1680         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1681         err = -ENODEV;
1682         if (!dp)
1683                 goto exit_unlock;
1684
1685         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1686                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1687
1688                 err = -EFBIG;
1689                 if (port_no >= DP_MAX_PORTS)
1690                         goto exit_unlock;
1691
1692                 vport = ovs_vport_rtnl_rcu(dp, port_no);
1693                 err = -EBUSY;
1694                 if (vport)
1695                         goto exit_unlock;
1696         } else {
1697                 for (port_no = 1; ; port_no++) {
1698                         if (port_no >= DP_MAX_PORTS) {
1699                                 err = -EFBIG;
1700                                 goto exit_unlock;
1701                         }
1702                         vport = ovs_vport_rtnl(dp, port_no);
1703                         if (!vport)
1704                                 break;
1705                 }
1706         }
1707
1708         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1709         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1710         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1711         parms.dp = dp;
1712         parms.port_no = port_no;
1713         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1714
1715         vport = new_vport(&parms);
1716         err = PTR_ERR(vport);
1717         if (IS_ERR(vport))
1718                 goto exit_unlock;
1719
1720         err = 0;
1721         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1722                                          OVS_VPORT_CMD_NEW);
1723         if (IS_ERR(reply)) {
1724                 err = PTR_ERR(reply);
1725                 ovs_dp_detach_port(vport);
1726                 goto exit_unlock;
1727         }
1728
1729         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1730
1731 exit_unlock:
1732         rtnl_unlock();
1733 exit:
1734         return err;
1735 }
1736
1737 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1738 {
1739         struct nlattr **a = info->attrs;
1740         struct sk_buff *reply;
1741         struct vport *vport;
1742         int err;
1743
1744         rtnl_lock();
1745         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1746         err = PTR_ERR(vport);
1747         if (IS_ERR(vport))
1748                 goto exit_unlock;
1749
1750         err = 0;
1751         if (a[OVS_VPORT_ATTR_TYPE] &&
1752             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1753                 err = -EINVAL;
1754
1755         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1756                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1757         if (err)
1758                 goto exit_unlock;
1759         if (a[OVS_VPORT_ATTR_UPCALL_PID])
1760                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1761
1762         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1763                                          OVS_VPORT_CMD_NEW);
1764         if (IS_ERR(reply)) {
1765                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1766                                 ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
1767                 goto exit_unlock;
1768         }
1769
1770         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1771
1772 exit_unlock:
1773         rtnl_unlock();
1774         return err;
1775 }
1776
1777 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1778 {
1779         struct nlattr **a = info->attrs;
1780         struct sk_buff *reply;
1781         struct vport *vport;
1782         int err;
1783
1784         rtnl_lock();
1785         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1786         err = PTR_ERR(vport);
1787         if (IS_ERR(vport))
1788                 goto exit_unlock;
1789
1790         if (vport->port_no == OVSP_LOCAL) {
1791                 err = -EINVAL;
1792                 goto exit_unlock;
1793         }
1794
1795         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1796                                          OVS_VPORT_CMD_DEL);
1797         err = PTR_ERR(reply);
1798         if (IS_ERR(reply))
1799                 goto exit_unlock;
1800
1801         err = 0;
1802         ovs_dp_detach_port(vport);
1803
1804         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1805
1806 exit_unlock:
1807         rtnl_unlock();
1808         return err;
1809 }
1810
1811 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1812 {
1813         struct nlattr **a = info->attrs;
1814         struct ovs_header *ovs_header = info->userhdr;
1815         struct sk_buff *reply;
1816         struct vport *vport;
1817         int err;
1818
1819         rcu_read_lock();
1820         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1821         err = PTR_ERR(vport);
1822         if (IS_ERR(vport))
1823                 goto exit_unlock;
1824
1825         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1826                                          OVS_VPORT_CMD_NEW);
1827         err = PTR_ERR(reply);
1828         if (IS_ERR(reply))
1829                 goto exit_unlock;
1830
1831         rcu_read_unlock();
1832
1833         return genlmsg_reply(reply, info);
1834
1835 exit_unlock:
1836         rcu_read_unlock();
1837         return err;
1838 }
1839
1840 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1841 {
1842         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1843         struct datapath *dp;
1844         int bucket = cb->args[0], skip = cb->args[1];
1845         int i, j = 0;
1846
1847         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1848         if (!dp)
1849                 return -ENODEV;
1850
1851         rcu_read_lock();
1852         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1853                 struct vport *vport;
1854
1855                 j = 0;
1856                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1857                         if (j >= skip &&
1858                             ovs_vport_cmd_fill_info(vport, skb,
1859                                                     NETLINK_CB(cb->skb).portid,
1860                                                     cb->nlh->nlmsg_seq,
1861                                                     NLM_F_MULTI,
1862                                                     OVS_VPORT_CMD_NEW) < 0)
1863                                 goto out;
1864
1865                         j++;
1866                 }
1867                 skip = 0;
1868         }
1869 out:
1870         rcu_read_unlock();
1871
1872         cb->args[0] = i;
1873         cb->args[1] = j;
1874
1875         return skb->len;
1876 }
1877
1878 static struct genl_ops dp_vport_genl_ops[] = {
1879         { .cmd = OVS_VPORT_CMD_NEW,
1880           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1881           .policy = vport_policy,
1882           .doit = ovs_vport_cmd_new
1883         },
1884         { .cmd = OVS_VPORT_CMD_DEL,
1885           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1886           .policy = vport_policy,
1887           .doit = ovs_vport_cmd_del
1888         },
1889         { .cmd = OVS_VPORT_CMD_GET,
1890           .flags = 0,               /* OK for unprivileged users. */
1891           .policy = vport_policy,
1892           .doit = ovs_vport_cmd_get,
1893           .dumpit = ovs_vport_cmd_dump
1894         },
1895         { .cmd = OVS_VPORT_CMD_SET,
1896           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1897           .policy = vport_policy,
1898           .doit = ovs_vport_cmd_set,
1899         },
1900 };
1901
1902 struct genl_family_and_ops {
1903         struct genl_family *family;
1904         struct genl_ops *ops;
1905         int n_ops;
1906         struct genl_multicast_group *group;
1907 };
1908
1909 static const struct genl_family_and_ops dp_genl_families[] = {
1910         { &dp_datapath_genl_family,
1911           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1912           &ovs_dp_datapath_multicast_group },
1913         { &dp_vport_genl_family,
1914           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1915           &ovs_dp_vport_multicast_group },
1916         { &dp_flow_genl_family,
1917           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1918           &ovs_dp_flow_multicast_group },
1919         { &dp_packet_genl_family,
1920           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1921           NULL },
1922 };
1923
1924 static void dp_unregister_genl(int n_families)
1925 {
1926         int i;
1927
1928         for (i = 0; i < n_families; i++)
1929                 genl_unregister_family(dp_genl_families[i].family);
1930 }
1931
1932 static int dp_register_genl(void)
1933 {
1934         int n_registered;
1935         int err;
1936         int i;
1937
1938         n_registered = 0;
1939         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1940                 const struct genl_family_and_ops *f = &dp_genl_families[i];
1941
1942                 err = genl_register_family_with_ops(f->family, f->ops,
1943                                                     f->n_ops);
1944                 if (err)
1945                         goto error;
1946                 n_registered++;
1947
1948                 if (f->group) {
1949                         err = genl_register_mc_group(f->family, f->group);
1950                         if (err)
1951                                 goto error;
1952                 }
1953         }
1954
1955         return 0;
1956
1957 error:
1958         dp_unregister_genl(n_registered);
1959         return err;
1960 }
1961
1962 static void rehash_flow_table(struct work_struct *work)
1963 {
1964         struct datapath *dp;
1965         struct net *net;
1966
1967         genl_lock();
1968         rtnl_lock();
1969         for_each_net(net) {
1970                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1971
1972                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1973                         struct flow_table *old_table = genl_dereference(dp->table);
1974                         struct flow_table *new_table;
1975
1976                         new_table = ovs_flow_tbl_rehash(old_table);
1977                         if (!IS_ERR(new_table)) {
1978                                 rcu_assign_pointer(dp->table, new_table);
1979                                 ovs_flow_tbl_deferred_destroy(old_table);
1980                         }
1981                 }
1982         }
1983         rtnl_unlock();
1984         genl_unlock();
1985
1986         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
1987 }
1988
1989 static int __net_init ovs_init_net(struct net *net)
1990 {
1991         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1992
1993         INIT_LIST_HEAD(&ovs_net->dps);
1994         return 0;
1995 }
1996
1997 static void __net_exit ovs_exit_net(struct net *net)
1998 {
1999         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2000         struct datapath *dp, *dp_next;
2001
2002         genl_lock();
2003         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2004                 __dp_destroy(dp);
2005         genl_unlock();
2006 }
2007
2008 static struct pernet_operations ovs_net_ops = {
2009         .init = ovs_init_net,
2010         .exit = ovs_exit_net,
2011         .id   = &ovs_net_id,
2012         .size = sizeof(struct ovs_net),
2013 };
2014
2015 static int __init dp_init(void)
2016 {
2017         int err;
2018
2019         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2020
2021         pr_info("Open vSwitch switching datapath\n");
2022
2023         err = ovs_flow_init();
2024         if (err)
2025                 goto error;
2026
2027         err = ovs_vport_init();
2028         if (err)
2029                 goto error_flow_exit;
2030
2031         err = register_pernet_device(&ovs_net_ops);
2032         if (err)
2033                 goto error_vport_exit;
2034
2035         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2036         if (err)
2037                 goto error_netns_exit;
2038
2039         err = dp_register_genl();
2040         if (err < 0)
2041                 goto error_unreg_notifier;
2042
2043         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2044
2045         return 0;
2046
2047 error_unreg_notifier:
2048         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2049 error_netns_exit:
2050         unregister_pernet_device(&ovs_net_ops);
2051 error_vport_exit:
2052         ovs_vport_exit();
2053 error_flow_exit:
2054         ovs_flow_exit();
2055 error:
2056         return err;
2057 }
2058
2059 static void dp_cleanup(void)
2060 {
2061         cancel_delayed_work_sync(&rehash_flow_wq);
2062         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2063         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2064         unregister_pernet_device(&ovs_net_ops);
2065         rcu_barrier();
2066         ovs_vport_exit();
2067         ovs_flow_exit();
2068 }
2069
2070 module_init(dp_init);
2071 module_exit(dp_cleanup);
2072
2073 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2074 MODULE_LICENSE("GPL");