Pileus Git - ~andy/linux/blob - 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/lockdep.h>
  48 #include <linux/openvswitch.h>
  49 #include <linux/rculist.h>
  50 #include <linux/dmi.h>
  51 #include <linux/workqueue.h>
  52 #include <net/genetlink.h>
  53 #include <net/net_namespace.h>
  54 #include <net/netns/generic.h>
  55
  56 #include "datapath.h"
  57 #include "flow.h"
  58 #include "vport-internal_dev.h"
  59 #include "vport-netdev.h"
  60
  61
  62 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
  63 static void rehash_flow_table(struct work_struct *work);
  64 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
  65
  66 int ovs_net_id __read_mostly;
  67
  68 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
  69                        struct genl_multicast_group *grp)
  70 {
  71         genl_notify(skb, genl_info_net(info), info->snd_portid,
  72                     grp->id, info->nlhdr, GFP_KERNEL);
  73 }
  74
  75 /**
  76  * DOC: Locking:
  77  *
  78  * All writes e.g. Writes to device state (add/remove datapath, port, set
  79  * operations on vports, etc.), Writes to other state (flow table
  80  * modifications, set miscellaneous datapath parameters, etc.) are protected
  81  * by ovs_lock.
  82  *
  83  * Reads are protected by RCU.
  84  *
  85  * There are a few special cases (mostly stats) that have their own
  86  * synchronization but they nest under all of above and don't interact with
  87  * each other.
  88  *
  89  * The RTNL lock nests inside ovs_mutex.
  90  */
  91
  92 static DEFINE_MUTEX(ovs_mutex);
  93
  94 void ovs_lock(void)
  95 {
  96         mutex_lock(&ovs_mutex);
  97 }
  98
  99 void ovs_unlock(void)
 100 {
 101         mutex_unlock(&ovs_mutex);
 102 }
 103
 104 #ifdef CONFIG_LOCKDEP
 105 int lockdep_ovsl_is_held(void)
 106 {
 107         if (debug_locks)
 108                 return lockdep_is_held(&ovs_mutex);
 109         else
 110                 return 1;
 111 }
 112 #endif
 113
 114 static struct vport *new_vport(const struct vport_parms *);
 115 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
 116                              const struct dp_upcall_info *);
 117 static int queue_userspace_packet(struct net *, int dp_ifindex,
 118                                   struct sk_buff *,
 119                                   const struct dp_upcall_info *);
 120
 121 /* Must be called with rcu_read_lock or ovs_mutex. */
 122 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 123 {
 124         struct datapath *dp = NULL;
 125         struct net_device *dev;
 126
 127         rcu_read_lock();
 128         dev = dev_get_by_index_rcu(net, dp_ifindex);
 129         if (dev) {
 130                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 131                 if (vport)
 132                         dp = vport->dp;
 133         }
 134         rcu_read_unlock();
 135
 136         return dp;
 137 }
 138
 139 /* Must be called with rcu_read_lock or ovs_mutex. */
 140 const char *ovs_dp_name(const struct datapath *dp)
 141 {
 142         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 143         return vport->ops->get_name(vport);
 144 }
 145
 146 static int get_dpifindex(struct datapath *dp)
 147 {
 148         struct vport *local;
 149         int ifindex;
 150
 151         rcu_read_lock();
 152
 153         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 154         if (local)
 155                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 156         else
 157                 ifindex = 0;
 158
 159         rcu_read_unlock();
 160
 161         return ifindex;
 162 }
 163
 164 static void destroy_dp_rcu(struct rcu_head *rcu)
 165 {
 166         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 167
 168         ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
 169         free_percpu(dp->stats_percpu);
 170         release_net(ovs_dp_get_net(dp));
 171         kfree(dp->ports);
 172         kfree(dp);
 173 }
 174
 175 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 176                                             u16 port_no)
 177 {
 178         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 179 }
 180
 181 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 182 {
 183         struct vport *vport;
 184         struct hlist_head *head;
 185
 186         head = vport_hash_bucket(dp, port_no);
 187         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 188                 if (vport->port_no == port_no)
 189                         return vport;
 190         }
 191         return NULL;
 192 }
 193
 194 /* Called with ovs_mutex. */
 195 static struct vport *new_vport(const struct vport_parms *parms)
 196 {
 197         struct vport *vport;
 198
 199         vport = ovs_vport_add(parms);
 200         if (!IS_ERR(vport)) {
 201                 struct datapath *dp = parms->dp;
 202                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 203
 204                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 205         }
 206         return vport;
 207 }
 208
 209 void ovs_dp_detach_port(struct vport *p)
 210 {
 211         ASSERT_OVSL();
 212
 213         /* First drop references to device. */
 214         hlist_del_rcu(&p->dp_hash_node);
 215
 216         /* Then destroy it. */
 217         ovs_vport_del(p);
 218 }
 219
 220 /* Must be called with rcu_read_lock. */
 221 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 222 {
 223         struct datapath *dp = p->dp;
 224         struct sw_flow *flow;
 225         struct dp_stats_percpu *stats;
 226         struct sw_flow_key key;
 227         u64 *stats_counter;
 228         int error;
 229         int key_len;
 230
 231         stats = this_cpu_ptr(dp->stats_percpu);
 232
 233         /* Extract flow from 'skb' into 'key'. */
 234         error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
 235         if (unlikely(error)) {
 236                 kfree_skb(skb);
 237                 return;
 238         }
 239
 240         /* Look up flow. */
 241         flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
 242         if (unlikely(!flow)) {
 243                 struct dp_upcall_info upcall;
 244
 245                 upcall.cmd = OVS_PACKET_CMD_MISS;
 246                 upcall.key = &key;
 247                 upcall.userdata = NULL;
 248                 upcall.portid = p->upcall_portid;
 249                 ovs_dp_upcall(dp, skb, &upcall);
 250                 consume_skb(skb);
 251                 stats_counter = &stats->n_missed;
 252                 goto out;
 253         }
 254
 255         OVS_CB(skb)->flow = flow;
 256
 257         stats_counter = &stats->n_hit;
 258         ovs_flow_used(OVS_CB(skb)->flow, skb);
 259         ovs_execute_actions(dp, skb);
 260
 261 out:
 262         /* Update datapath statistics. */
 263         u64_stats_update_begin(&stats->sync);
 264         (*stats_counter)++;
 265         u64_stats_update_end(&stats->sync);
 266 }
 267
 268 static struct genl_family dp_packet_genl_family = {
 269         .id = GENL_ID_GENERATE,
 270         .hdrsize = sizeof(struct ovs_header),
 271         .name = OVS_PACKET_FAMILY,
 272         .version = OVS_PACKET_VERSION,
 273         .maxattr = OVS_PACKET_ATTR_MAX,
 274         .netnsok = true,
 275         .parallel_ops = true,
 276 };
 277
 278 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 279                   const struct dp_upcall_info *upcall_info)
 280 {
 281         struct dp_stats_percpu *stats;
 282         int dp_ifindex;
 283         int err;
 284
 285         if (upcall_info->portid == 0) {
 286                 err = -ENOTCONN;
 287                 goto err;
 288         }
 289
 290         dp_ifindex = get_dpifindex(dp);
 291         if (!dp_ifindex) {
 292                 err = -ENODEV;
 293                 goto err;
 294         }
 295
 296         if (!skb_is_gso(skb))
 297                 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 298         else
 299                 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 300         if (err)
 301                 goto err;
 302
 303         return 0;
 304
 305 err:
 306         stats = this_cpu_ptr(dp->stats_percpu);
 307
 308         u64_stats_update_begin(&stats->sync);
 309         stats->n_lost++;
 310         u64_stats_update_end(&stats->sync);
 311
 312         return err;
 313 }
 314
 315 static int queue_gso_packets(struct net *net, int dp_ifindex,
 316                              struct sk_buff *skb,
 317                              const struct dp_upcall_info *upcall_info)
 318 {
 319         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 320         struct dp_upcall_info later_info;
 321         struct sw_flow_key later_key;
 322         struct sk_buff *segs, *nskb;
 323         int err;
 324
 325         segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
 326         if (IS_ERR(segs))
 327                 return PTR_ERR(segs);
 328
 329         /* Queue all of the segments. */
 330         skb = segs;
 331         do {
 332                 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
 333                 if (err)
 334                         break;
 335
 336                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 337                         /* The initial flow key extracted by ovs_flow_extract()
 338                          * in this case is for a first fragment, so we need to
 339                          * properly mark later fragments.
 340                          */
 341                         later_key = *upcall_info->key;
 342                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 343
 344                         later_info = *upcall_info;
 345                         later_info.key = &later_key;
 346                         upcall_info = &later_info;
 347                 }
 348         } while ((skb = skb->next));
 349
 350         /* Free all of the segments. */
 351         skb = segs;
 352         do {
 353                 nskb = skb->next;
 354                 if (err)
 355                         kfree_skb(skb);
 356                 else
 357                         consume_skb(skb);
 358         } while ((skb = nskb));
 359         return err;
 360 }
 361
 362 static size_t key_attr_size(void)
 363 {
 364         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 365                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 366                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 367                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 368                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 369                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 370                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 371                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 372                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 373                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 374                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 375 }
 376
 377 static size_t upcall_msg_size(const struct sk_buff *skb,
 378                               const struct nlattr *userdata)
 379 {
 380         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 381                 + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
 382                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 383
 384         /* OVS_PACKET_ATTR_USERDATA */
 385         if (userdata)
 386                 size += NLA_ALIGN(userdata->nla_len);
 387
 388         return size;
 389 }
 390
 391 static int queue_userspace_packet(struct net *net, int dp_ifindex,
 392                                   struct sk_buff *skb,
 393                                   const struct dp_upcall_info *upcall_info)
 394 {
 395         struct ovs_header *upcall;
 396         struct sk_buff *nskb = NULL;
 397         struct sk_buff *user_skb; /* to be queued to userspace */
 398         struct nlattr *nla;
 399         int err;
 400
 401         if (vlan_tx_tag_present(skb)) {
 402                 nskb = skb_clone(skb, GFP_ATOMIC);
 403                 if (!nskb)
 404                         return -ENOMEM;
 405
 406                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 407                 if (!nskb)
 408                         return -ENOMEM;
 409
 410                 nskb->vlan_tci = 0;
 411                 skb = nskb;
 412         }
 413
 414         if (nla_attr_size(skb->len) > USHRT_MAX) {
 415                 err = -EFBIG;
 416                 goto out;
 417         }
 418
 419         user_skb = genlmsg_new(upcall_msg_size(skb, upcall_info->userdata), GFP_ATOMIC);
 420         if (!user_skb) {
 421                 err = -ENOMEM;
 422                 goto out;
 423         }
 424
 425         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 426                              0, upcall_info->cmd);
 427         upcall->dp_ifindex = dp_ifindex;
 428
 429         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 430         ovs_flow_to_nlattrs(upcall_info->key, user_skb);
 431         nla_nest_end(user_skb, nla);
 432
 433         if (upcall_info->userdata)
 434                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 435                           nla_len(upcall_info->userdata),
 436                           nla_data(upcall_info->userdata));
 437
 438         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 439
 440         skb_copy_and_csum_dev(skb, nla_data(nla));
 441
 442         genlmsg_end(user_skb, upcall);
 443         err = genlmsg_unicast(net, user_skb, upcall_info->portid);
 444
 445 out:
 446         kfree_skb(nskb);
 447         return err;
 448 }
 449
 450 /* Called with ovs_mutex. */
 451 static int flush_flows(struct datapath *dp)
 452 {
 453         struct flow_table *old_table;
 454         struct flow_table *new_table;
 455
 456         old_table = ovsl_dereference(dp->table);
 457         new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
 458         if (!new_table)
 459                 return -ENOMEM;
 460
 461         rcu_assign_pointer(dp->table, new_table);
 462
 463         ovs_flow_tbl_deferred_destroy(old_table);
 464         return 0;
 465 }
 466
 467 static int validate_actions(const struct nlattr *attr,
 468                                 const struct sw_flow_key *key, int depth);
 469
 470 static int validate_sample(const struct nlattr *attr,
 471                                 const struct sw_flow_key *key, int depth)
 472 {
 473         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 474         const struct nlattr *probability, *actions;
 475         const struct nlattr *a;
 476         int rem;
 477
 478         memset(attrs, 0, sizeof(attrs));
 479         nla_for_each_nested(a, attr, rem) {
 480                 int type = nla_type(a);
 481                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 482                         return -EINVAL;
 483                 attrs[type] = a;
 484         }
 485         if (rem)
 486                 return -EINVAL;
 487
 488         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 489         if (!probability || nla_len(probability) != sizeof(u32))
 490                 return -EINVAL;
 491
 492         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 493         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 494                 return -EINVAL;
 495         return validate_actions(actions, key, depth + 1);
 496 }
 497
 498 static int validate_tp_port(const struct sw_flow_key *flow_key)
 499 {
 500         if (flow_key->eth.type == htons(ETH_P_IP)) {
 501                 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
 502                         return 0;
 503         } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
 504                 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
 505                         return 0;
 506         }
 507
 508         return -EINVAL;
 509 }
 510
 511 static int validate_set(const struct nlattr *a,
 512                         const struct sw_flow_key *flow_key)
 513 {
 514         const struct nlattr *ovs_key = nla_data(a);
 515         int key_type = nla_type(ovs_key);
 516
 517         /* There can be only one key in a action */
 518         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 519                 return -EINVAL;
 520
 521         if (key_type > OVS_KEY_ATTR_MAX ||
 522             nla_len(ovs_key) != ovs_key_lens[key_type])
 523                 return -EINVAL;
 524
 525         switch (key_type) {
 526         const struct ovs_key_ipv4 *ipv4_key;
 527         const struct ovs_key_ipv6 *ipv6_key;
 528
 529         case OVS_KEY_ATTR_PRIORITY:
 530         case OVS_KEY_ATTR_SKB_MARK:
 531         case OVS_KEY_ATTR_ETHERNET:
 532                 break;
 533
 534         case OVS_KEY_ATTR_IPV4:
 535                 if (flow_key->eth.type != htons(ETH_P_IP))
 536                         return -EINVAL;
 537
 538                 if (!flow_key->ip.proto)
 539                         return -EINVAL;
 540
 541                 ipv4_key = nla_data(ovs_key);
 542                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 543                         return -EINVAL;
 544
 545                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 546                         return -EINVAL;
 547
 548                 break;
 549
 550         case OVS_KEY_ATTR_IPV6:
 551                 if (flow_key->eth.type != htons(ETH_P_IPV6))
 552                         return -EINVAL;
 553
 554                 if (!flow_key->ip.proto)
 555                         return -EINVAL;
 556
 557                 ipv6_key = nla_data(ovs_key);
 558                 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
 559                         return -EINVAL;
 560
 561                 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
 562                         return -EINVAL;
 563
 564                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
 565                         return -EINVAL;
 566
 567                 break;
 568
 569         case OVS_KEY_ATTR_TCP:
 570                 if (flow_key->ip.proto != IPPROTO_TCP)
 571                         return -EINVAL;
 572
 573                 return validate_tp_port(flow_key);
 574
 575         case OVS_KEY_ATTR_UDP:
 576                 if (flow_key->ip.proto != IPPROTO_UDP)
 577                         return -EINVAL;
 578
 579                 return validate_tp_port(flow_key);
 580
 581         default:
 582                 return -EINVAL;
 583         }
 584
 585         return 0;
 586 }
 587
 588 static int validate_userspace(const struct nlattr *attr)
 589 {
 590         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 591                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 592                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
 593         };
 594         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 595         int error;
 596
 597         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 598                                  attr, userspace_policy);
 599         if (error)
 600                 return error;
 601
 602         if (!a[OVS_USERSPACE_ATTR_PID] ||
 603             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 604                 return -EINVAL;
 605
 606         return 0;
 607 }
 608
 609 static int validate_actions(const struct nlattr *attr,
 610                                 const struct sw_flow_key *key,  int depth)
 611 {
 612         const struct nlattr *a;
 613         int rem, err;
 614
 615         if (depth >= SAMPLE_ACTION_DEPTH)
 616                 return -EOVERFLOW;
 617
 618         nla_for_each_nested(a, attr, rem) {
 619                 /* Expected argument lengths, (u32)-1 for variable length. */
 620                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 621                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 622                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 623                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 624                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 625                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 626                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 627                 };
 628                 const struct ovs_action_push_vlan *vlan;
 629                 int type = nla_type(a);
 630
 631                 if (type > OVS_ACTION_ATTR_MAX ||
 632                     (action_lens[type] != nla_len(a) &&
 633                      action_lens[type] != (u32)-1))
 634                         return -EINVAL;
 635
 636                 switch (type) {
 637                 case OVS_ACTION_ATTR_UNSPEC:
 638                         return -EINVAL;
 639
 640                 case OVS_ACTION_ATTR_USERSPACE:
 641                         err = validate_userspace(a);
 642                         if (err)
 643                                 return err;
 644                         break;
 645
 646                 case OVS_ACTION_ATTR_OUTPUT:
 647                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 648                                 return -EINVAL;
 649                         break;
 650
 651
 652                 case OVS_ACTION_ATTR_POP_VLAN:
 653                         break;
 654
 655                 case OVS_ACTION_ATTR_PUSH_VLAN:
 656                         vlan = nla_data(a);
 657                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 658                                 return -EINVAL;
 659                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 660                                 return -EINVAL;
 661                         break;
 662
 663                 case OVS_ACTION_ATTR_SET:
 664                         err = validate_set(a, key);
 665                         if (err)
 666                                 return err;
 667                         break;
 668
 669                 case OVS_ACTION_ATTR_SAMPLE:
 670                         err = validate_sample(a, key, depth);
 671                         if (err)
 672                                 return err;
 673                         break;
 674
 675                 default:
 676                         return -EINVAL;
 677                 }
 678         }
 679
 680         if (rem > 0)
 681                 return -EINVAL;
 682
 683         return 0;
 684 }
 685
 686 static void clear_stats(struct sw_flow *flow)
 687 {
 688         flow->used = 0;
 689         flow->tcp_flags = 0;
 690         flow->packet_count = 0;
 691         flow->byte_count = 0;
 692 }
 693
 694 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 695 {
 696         struct ovs_header *ovs_header = info->userhdr;
 697         struct nlattr **a = info->attrs;
 698         struct sw_flow_actions *acts;
 699         struct sk_buff *packet;
 700         struct sw_flow *flow;
 701         struct datapath *dp;
 702         struct ethhdr *eth;
 703         int len;
 704         int err;
 705         int key_len;
 706
 707         err = -EINVAL;
 708         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 709             !a[OVS_PACKET_ATTR_ACTIONS])
 710                 goto err;
 711
 712         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 713         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 714         err = -ENOMEM;
 715         if (!packet)
 716                 goto err;
 717         skb_reserve(packet, NET_IP_ALIGN);
 718
 719         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 720
 721         skb_reset_mac_header(packet);
 722         eth = eth_hdr(packet);
 723
 724         /* Normally, setting the skb 'protocol' field would be handled by a
 725          * call to eth_type_trans(), but it assumes there's a sending
 726          * device, which we may not have. */
 727         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 728                 packet->protocol = eth->h_proto;
 729         else
 730                 packet->protocol = htons(ETH_P_802_2);
 731
 732         /* Build an sw_flow for sending this packet. */
 733         flow = ovs_flow_alloc();
 734         err = PTR_ERR(flow);
 735         if (IS_ERR(flow))
 736                 goto err_kfree_skb;
 737
 738         err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
 739         if (err)
 740                 goto err_flow_free;
 741
 742         err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
 743                                              &flow->key.phy.skb_mark,
 744                                              &flow->key.phy.in_port,
 745                                              a[OVS_PACKET_ATTR_KEY]);
 746         if (err)
 747                 goto err_flow_free;
 748
 749         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
 750         if (err)
 751                 goto err_flow_free;
 752
 753         flow->hash = ovs_flow_hash(&flow->key, key_len);
 754
 755         acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 756         err = PTR_ERR(acts);
 757         if (IS_ERR(acts))
 758                 goto err_flow_free;
 759         rcu_assign_pointer(flow->sf_acts, acts);
 760
 761         OVS_CB(packet)->flow = flow;
 762         packet->priority = flow->key.phy.priority;
 763         packet->mark = flow->key.phy.skb_mark;
 764
 765         rcu_read_lock();
 766         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 767         err = -ENODEV;
 768         if (!dp)
 769                 goto err_unlock;
 770
 771         local_bh_disable();
 772         err = ovs_execute_actions(dp, packet);
 773         local_bh_enable();
 774         rcu_read_unlock();
 775
 776         ovs_flow_free(flow);
 777         return err;
 778
 779 err_unlock:
 780         rcu_read_unlock();
 781 err_flow_free:
 782         ovs_flow_free(flow);
 783 err_kfree_skb:
 784         kfree_skb(packet);
 785 err:
 786         return err;
 787 }
 788
 789 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 790         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 791         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 792         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 793 };
 794
 795 static struct genl_ops dp_packet_genl_ops[] = {
 796         { .cmd = OVS_PACKET_CMD_EXECUTE,
 797           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 798           .policy = packet_policy,
 799           .doit = ovs_packet_cmd_execute
 800         }
 801 };
 802
 803 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 804 {
 805         int i;
 806         struct flow_table *table = ovsl_dereference(dp->table);
 807
 808         stats->n_flows = ovs_flow_tbl_count(table);
 809
 810         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 811         for_each_possible_cpu(i) {
 812                 const struct dp_stats_percpu *percpu_stats;
 813                 struct dp_stats_percpu local_stats;
 814                 unsigned int start;
 815
 816                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 817
 818                 do {
 819                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 820                         local_stats = *percpu_stats;
 821                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 822
 823                 stats->n_hit += local_stats.n_hit;
 824                 stats->n_missed += local_stats.n_missed;
 825                 stats->n_lost += local_stats.n_lost;
 826         }
 827 }
 828
 829 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 830         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 831         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 832         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 833 };
 834
 835 static struct genl_family dp_flow_genl_family = {
 836         .id = GENL_ID_GENERATE,
 837         .hdrsize = sizeof(struct ovs_header),
 838         .name = OVS_FLOW_FAMILY,
 839         .version = OVS_FLOW_VERSION,
 840         .maxattr = OVS_FLOW_ATTR_MAX,
 841         .netnsok = true,
 842         .parallel_ops = true,
 843 };
 844
 845 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 846         .name = OVS_FLOW_MCGROUP
 847 };
 848
 849 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 850 {
 851         return NLMSG_ALIGN(sizeof(struct ovs_header))
 852                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 853                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 854                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 855                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 856                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 857 }
 858
 859 /* Called with ovs_mutex. */
 860 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 861                                   struct sk_buff *skb, u32 portid,
 862                                   u32 seq, u32 flags, u8 cmd)
 863 {
 864         const int skb_orig_len = skb->len;
 865         const struct sw_flow_actions *sf_acts;
 866         struct ovs_flow_stats stats;
 867         struct ovs_header *ovs_header;
 868         struct nlattr *nla;
 869         unsigned long used;
 870         u8 tcp_flags;
 871         int err;
 872
 873         sf_acts = ovsl_dereference(flow->sf_acts);
 874
 875         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 876         if (!ovs_header)
 877                 return -EMSGSIZE;
 878
 879         ovs_header->dp_ifindex = get_dpifindex(dp);
 880
 881         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 882         if (!nla)
 883                 goto nla_put_failure;
 884         err = ovs_flow_to_nlattrs(&flow->key, skb);
 885         if (err)
 886                 goto error;
 887         nla_nest_end(skb, nla);
 888
 889         spin_lock_bh(&flow->lock);
 890         used = flow->used;
 891         stats.n_packets = flow->packet_count;
 892         stats.n_bytes = flow->byte_count;
 893         tcp_flags = flow->tcp_flags;
 894         spin_unlock_bh(&flow->lock);
 895
 896         if (used &&
 897             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 898                 goto nla_put_failure;
 899
 900         if (stats.n_packets &&
 901             nla_put(skb, OVS_FLOW_ATTR_STATS,
 902                     sizeof(struct ovs_flow_stats), &stats))
 903                 goto nla_put_failure;
 904
 905         if (tcp_flags &&
 906             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
 907                 goto nla_put_failure;
 908
 909         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 910          * this is the first flow to be dumped into 'skb'.  This is unusual for
 911          * Netlink but individual action lists can be longer than
 912          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 913          * The userspace caller can always fetch the actions separately if it
 914          * really wants them.  (Most userspace callers in fact don't care.)
 915          *
 916          * This can only fail for dump operations because the skb is always
 917          * properly sized for single flows.
 918          */
 919         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 920                       sf_acts->actions);
 921         if (err < 0 && skb_orig_len)
 922                 goto error;
 923
 924         return genlmsg_end(skb, ovs_header);
 925
 926 nla_put_failure:
 927         err = -EMSGSIZE;
 928 error:
 929         genlmsg_cancel(skb, ovs_header);
 930         return err;
 931 }
 932
 933 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 934 {
 935         const struct sw_flow_actions *sf_acts;
 936
 937         sf_acts = ovsl_dereference(flow->sf_acts);
 938
 939         return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
 940 }
 941
 942 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 943                                                struct datapath *dp,
 944                                                u32 portid, u32 seq, u8 cmd)
 945 {
 946         struct sk_buff *skb;
 947         int retval;
 948
 949         skb = ovs_flow_cmd_alloc_info(flow);
 950         if (!skb)
 951                 return ERR_PTR(-ENOMEM);
 952
 953         retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
 954         BUG_ON(retval < 0);
 955         return skb;
 956 }
 957
 958 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 959 {
 960         struct nlattr **a = info->attrs;
 961         struct ovs_header *ovs_header = info->userhdr;
 962         struct sw_flow_key key;
 963         struct sw_flow *flow;
 964         struct sk_buff *reply;
 965         struct datapath *dp;
 966         struct flow_table *table;
 967         int error;
 968         int key_len;
 969
 970         /* Extract key. */
 971         error = -EINVAL;
 972         if (!a[OVS_FLOW_ATTR_KEY])
 973                 goto error;
 974         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
 975         if (error)
 976                 goto error;
 977
 978         /* Validate actions. */
 979         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 980                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
 981                 if (error)
 982                         goto error;
 983         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 984                 error = -EINVAL;
 985                 goto error;
 986         }
 987
 988         ovs_lock();
 989         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 990         error = -ENODEV;
 991         if (!dp)
 992                 goto err_unlock_ovs;
 993
 994         table = ovsl_dereference(dp->table);
 995         flow = ovs_flow_tbl_lookup(table, &key, key_len);
 996         if (!flow) {
 997                 struct sw_flow_actions *acts;
 998
 999                 /* Bail out if we're not allowed to create a new flow. */
1000                 error = -ENOENT;
1001                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1002                         goto err_unlock_ovs;
1003
1004                 /* Expand table, if necessary, to make room. */
1005                 if (ovs_flow_tbl_need_to_expand(table)) {
1006                         struct flow_table *new_table;
1007
1008                         new_table = ovs_flow_tbl_expand(table);
1009                         if (!IS_ERR(new_table)) {
1010                                 rcu_assign_pointer(dp->table, new_table);
1011                                 ovs_flow_tbl_deferred_destroy(table);
1012                                 table = ovsl_dereference(dp->table);
1013                         }
1014                 }
1015
1016                 /* Allocate flow. */
1017                 flow = ovs_flow_alloc();
1018                 if (IS_ERR(flow)) {
1019                         error = PTR_ERR(flow);
1020                         goto err_unlock_ovs;
1021                 }
1022                 flow->key = key;
1023                 clear_stats(flow);
1024
1025                 /* Obtain actions. */
1026                 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1027                 error = PTR_ERR(acts);
1028                 if (IS_ERR(acts))
1029                         goto error_free_flow;
1030                 rcu_assign_pointer(flow->sf_acts, acts);
1031
1032                 /* Put flow in bucket. */
1033                 flow->hash = ovs_flow_hash(&key, key_len);
1034                 ovs_flow_tbl_insert(table, flow);
1035
1036                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1037                                                 info->snd_seq,
1038                                                 OVS_FLOW_CMD_NEW);
1039         } else {
1040                 /* We found a matching flow. */
1041                 struct sw_flow_actions *old_acts;
1042                 struct nlattr *acts_attrs;
1043
1044                 /* Bail out if we're not allowed to modify an existing flow.
1045                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1046                  * because Generic Netlink treats the latter as a dump
1047                  * request.  We also accept NLM_F_EXCL in case that bug ever
1048                  * gets fixed.
1049                  */
1050                 error = -EEXIST;
1051                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1052                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1053                         goto err_unlock_ovs;
1054
1055                 /* Update actions. */
1056                 old_acts = ovsl_dereference(flow->sf_acts);
1057                 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1058                 if (acts_attrs &&
1059                    (old_acts->actions_len != nla_len(acts_attrs) ||
1060                    memcmp(old_acts->actions, nla_data(acts_attrs),
1061                           old_acts->actions_len))) {
1062                         struct sw_flow_actions *new_acts;
1063
1064                         new_acts = ovs_flow_actions_alloc(acts_attrs);
1065                         error = PTR_ERR(new_acts);
1066                         if (IS_ERR(new_acts))
1067                                 goto err_unlock_ovs;
1068
1069                         rcu_assign_pointer(flow->sf_acts, new_acts);
1070                         ovs_flow_deferred_free_acts(old_acts);
1071                 }
1072
1073                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1074                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1075
1076                 /* Clear stats. */
1077                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1078                         spin_lock_bh(&flow->lock);
1079                         clear_stats(flow);
1080                         spin_unlock_bh(&flow->lock);
1081                 }
1082         }
1083         ovs_unlock();
1084
1085         if (!IS_ERR(reply))
1086                 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1087         else
1088                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1089                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1090         return 0;
1091
1092 error_free_flow:
1093         ovs_flow_free(flow);
1094 err_unlock_ovs:
1095         ovs_unlock();
1096 error:
1097         return error;
1098 }
1099
1100 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1101 {
1102         struct nlattr **a = info->attrs;
1103         struct ovs_header *ovs_header = info->userhdr;
1104         struct sw_flow_key key;
1105         struct sk_buff *reply;
1106         struct sw_flow *flow;
1107         struct datapath *dp;
1108         struct flow_table *table;
1109         int err;
1110         int key_len;
1111
1112         if (!a[OVS_FLOW_ATTR_KEY])
1113                 return -EINVAL;
1114         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1115         if (err)
1116                 return err;
1117
1118         ovs_lock();
1119         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1120         if (!dp) {
1121                 err = -ENODEV;
1122                 goto unlock;
1123         }
1124
1125         table = ovsl_dereference(dp->table);
1126         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1127         if (!flow) {
1128                 err = -ENOENT;
1129                 goto unlock;
1130         }
1131
1132         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1133                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1134         if (IS_ERR(reply)) {
1135                 err = PTR_ERR(reply);
1136                 goto unlock;
1137         }
1138
1139         ovs_unlock();
1140         return genlmsg_reply(reply, info);
1141 unlock:
1142         ovs_unlock();
1143         return err;
1144 }
1145
1146 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1147 {
1148         struct nlattr **a = info->attrs;
1149         struct ovs_header *ovs_header = info->userhdr;
1150         struct sw_flow_key key;
1151         struct sk_buff *reply;
1152         struct sw_flow *flow;
1153         struct datapath *dp;
1154         struct flow_table *table;
1155         int err;
1156         int key_len;
1157
1158         ovs_lock();
1159         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1160         if (!dp) {
1161                 err = -ENODEV;
1162                 goto unlock;
1163         }
1164
1165         if (!a[OVS_FLOW_ATTR_KEY]) {
1166                 err = flush_flows(dp);
1167                 goto unlock;
1168         }
1169         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1170         if (err)
1171                 goto unlock;
1172
1173         table = ovsl_dereference(dp->table);
1174         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1175         if (!flow) {
1176                 err = -ENOENT;
1177                 goto unlock;
1178         }
1179
1180         reply = ovs_flow_cmd_alloc_info(flow);
1181         if (!reply) {
1182                 err = -ENOMEM;
1183                 goto unlock;
1184         }
1185
1186         ovs_flow_tbl_remove(table, flow);
1187
1188         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1189                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1190         BUG_ON(err < 0);
1191
1192         ovs_flow_deferred_free(flow);
1193         ovs_unlock();
1194
1195         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1196         return 0;
1197 unlock:
1198         ovs_unlock();
1199         return err;
1200 }
1201
1202 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1203 {
1204         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1205         struct datapath *dp;
1206         struct flow_table *table;
1207
1208         ovs_lock();
1209         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1210         if (!dp) {
1211                 ovs_unlock();
1212                 return -ENODEV;
1213         }
1214
1215         table = ovsl_dereference(dp->table);
1216
1217         for (;;) {
1218                 struct sw_flow *flow;
1219                 u32 bucket, obj;
1220
1221                 bucket = cb->args[0];
1222                 obj = cb->args[1];
1223                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1224                 if (!flow)
1225                         break;
1226
1227                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1228                                            NETLINK_CB(cb->skb).portid,
1229                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1230                                            OVS_FLOW_CMD_NEW) < 0)
1231                         break;
1232
1233                 cb->args[0] = bucket;
1234                 cb->args[1] = obj;
1235         }
1236         ovs_unlock();
1237         return skb->len;
1238 }
1239
1240 static struct genl_ops dp_flow_genl_ops[] = {
1241         { .cmd = OVS_FLOW_CMD_NEW,
1242           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1243           .policy = flow_policy,
1244           .doit = ovs_flow_cmd_new_or_set
1245         },
1246         { .cmd = OVS_FLOW_CMD_DEL,
1247           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1248           .policy = flow_policy,
1249           .doit = ovs_flow_cmd_del
1250         },
1251         { .cmd = OVS_FLOW_CMD_GET,
1252           .flags = 0,               /* OK for unprivileged users. */
1253           .policy = flow_policy,
1254           .doit = ovs_flow_cmd_get,
1255           .dumpit = ovs_flow_cmd_dump
1256         },
1257         { .cmd = OVS_FLOW_CMD_SET,
1258           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1259           .policy = flow_policy,
1260           .doit = ovs_flow_cmd_new_or_set,
1261         },
1262 };
1263
1264 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1265         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1266         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1267 };
1268
1269 static struct genl_family dp_datapath_genl_family = {
1270         .id = GENL_ID_GENERATE,
1271         .hdrsize = sizeof(struct ovs_header),
1272         .name = OVS_DATAPATH_FAMILY,
1273         .version = OVS_DATAPATH_VERSION,
1274         .maxattr = OVS_DP_ATTR_MAX,
1275         .netnsok = true,
1276         .parallel_ops = true,
1277 };
1278
1279 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1280         .name = OVS_DATAPATH_MCGROUP
1281 };
1282
1283 static size_t ovs_dp_cmd_msg_size(void)
1284 {
1285         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1286
1287         msgsize += nla_total_size(IFNAMSIZ);
1288         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1289
1290         return msgsize;
1291 }
1292
1293 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1294                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1295 {
1296         struct ovs_header *ovs_header;
1297         struct ovs_dp_stats dp_stats;
1298         int err;
1299
1300         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1301                                    flags, cmd);
1302         if (!ovs_header)
1303                 goto error;
1304
1305         ovs_header->dp_ifindex = get_dpifindex(dp);
1306
1307         rcu_read_lock();
1308         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1309         rcu_read_unlock();
1310         if (err)
1311                 goto nla_put_failure;
1312
1313         get_dp_stats(dp, &dp_stats);
1314         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1315                 goto nla_put_failure;
1316
1317         return genlmsg_end(skb, ovs_header);
1318
1319 nla_put_failure:
1320         genlmsg_cancel(skb, ovs_header);
1321 error:
1322         return -EMSGSIZE;
1323 }
1324
1325 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1326                                              u32 seq, u8 cmd)
1327 {
1328         struct sk_buff *skb;
1329         int retval;
1330
1331         skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1332         if (!skb)
1333                 return ERR_PTR(-ENOMEM);
1334
1335         retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1336         if (retval < 0) {
1337                 kfree_skb(skb);
1338                 return ERR_PTR(retval);
1339         }
1340         return skb;
1341 }
1342
1343 /* Called with ovs_mutex. */
1344 static struct datapath *lookup_datapath(struct net *net,
1345                                         struct ovs_header *ovs_header,
1346                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1347 {
1348         struct datapath *dp;
1349
1350         if (!a[OVS_DP_ATTR_NAME])
1351                 dp = get_dp(net, ovs_header->dp_ifindex);
1352         else {
1353                 struct vport *vport;
1354
1355                 rcu_read_lock();
1356                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1357                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1358                 rcu_read_unlock();
1359         }
1360         return dp ? dp : ERR_PTR(-ENODEV);
1361 }
1362
1363 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1364 {
1365         struct nlattr **a = info->attrs;
1366         struct vport_parms parms;
1367         struct sk_buff *reply;
1368         struct datapath *dp;
1369         struct vport *vport;
1370         struct ovs_net *ovs_net;
1371         int err, i;
1372
1373         err = -EINVAL;
1374         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1375                 goto err;
1376
1377         ovs_lock();
1378
1379         err = -ENOMEM;
1380         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1381         if (dp == NULL)
1382                 goto err_unlock_ovs;
1383
1384         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1385
1386         /* Allocate table. */
1387         err = -ENOMEM;
1388         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1389         if (!dp->table)
1390                 goto err_free_dp;
1391
1392         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1393         if (!dp->stats_percpu) {
1394                 err = -ENOMEM;
1395                 goto err_destroy_table;
1396         }
1397
1398         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1399                         GFP_KERNEL);
1400         if (!dp->ports) {
1401                 err = -ENOMEM;
1402                 goto err_destroy_percpu;
1403         }
1404
1405         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1406                 INIT_HLIST_HEAD(&dp->ports[i]);
1407
1408         /* Set up our datapath device. */
1409         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1410         parms.type = OVS_VPORT_TYPE_INTERNAL;
1411         parms.options = NULL;
1412         parms.dp = dp;
1413         parms.port_no = OVSP_LOCAL;
1414         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1415
1416         vport = new_vport(&parms);
1417         if (IS_ERR(vport)) {
1418                 err = PTR_ERR(vport);
1419                 if (err == -EBUSY)
1420                         err = -EEXIST;
1421
1422                 goto err_destroy_ports_array;
1423         }
1424
1425         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1426                                       info->snd_seq, OVS_DP_CMD_NEW);
1427         err = PTR_ERR(reply);
1428         if (IS_ERR(reply))
1429                 goto err_destroy_local_port;
1430
1431         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1432         list_add_tail(&dp->list_node, &ovs_net->dps);
1433
1434         ovs_unlock();
1435
1436         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1437         return 0;
1438
1439 err_destroy_local_port:
1440         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1441 err_destroy_ports_array:
1442         kfree(dp->ports);
1443 err_destroy_percpu:
1444         free_percpu(dp->stats_percpu);
1445 err_destroy_table:
1446         ovs_flow_tbl_destroy(ovsl_dereference(dp->table));
1447 err_free_dp:
1448         release_net(ovs_dp_get_net(dp));
1449         kfree(dp);
1450 err_unlock_ovs:
1451         ovs_unlock();
1452 err:
1453         return err;
1454 }
1455
1456 /* Called with ovs_mutex. */
1457 static void __dp_destroy(struct datapath *dp)
1458 {
1459         int i;
1460
1461         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1462                 struct vport *vport;
1463                 struct hlist_node *n;
1464
1465                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1466                         if (vport->port_no != OVSP_LOCAL)
1467                                 ovs_dp_detach_port(vport);
1468         }
1469
1470         list_del(&dp->list_node);
1471
1472         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1473          * all port in datapath are destroyed first before freeing datapath.
1474          */
1475         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1476
1477         call_rcu(&dp->rcu, destroy_dp_rcu);
1478 }
1479
1480 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1481 {
1482         struct sk_buff *reply;
1483         struct datapath *dp;
1484         int err;
1485
1486         ovs_lock();
1487         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1488         err = PTR_ERR(dp);
1489         if (IS_ERR(dp))
1490                 goto unlock;
1491
1492         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1493                                       info->snd_seq, OVS_DP_CMD_DEL);
1494         err = PTR_ERR(reply);
1495         if (IS_ERR(reply))
1496                 goto unlock;
1497
1498         __dp_destroy(dp);
1499         ovs_unlock();
1500
1501         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1502
1503         return 0;
1504 unlock:
1505         ovs_unlock();
1506         return err;
1507 }
1508
1509 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1510 {
1511         struct sk_buff *reply;
1512         struct datapath *dp;
1513         int err;
1514
1515         ovs_lock();
1516         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1517         err = PTR_ERR(dp);
1518         if (IS_ERR(dp))
1519                 goto unlock;
1520
1521         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1522                                       info->snd_seq, OVS_DP_CMD_NEW);
1523         if (IS_ERR(reply)) {
1524                 err = PTR_ERR(reply);
1525                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1526                                 ovs_dp_datapath_multicast_group.id, err);
1527                 err = 0;
1528                 goto unlock;
1529         }
1530
1531         ovs_unlock();
1532         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1533
1534         return 0;
1535 unlock:
1536         ovs_unlock();
1537         return err;
1538 }
1539
1540 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1541 {
1542         struct sk_buff *reply;
1543         struct datapath *dp;
1544         int err;
1545
1546         ovs_lock();
1547         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1548         if (IS_ERR(dp)) {
1549                 err = PTR_ERR(dp);
1550                 goto unlock;
1551         }
1552
1553         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1554                                       info->snd_seq, OVS_DP_CMD_NEW);
1555         if (IS_ERR(reply)) {
1556                 err = PTR_ERR(reply);
1557                 goto unlock;
1558         }
1559
1560         ovs_unlock();
1561         return genlmsg_reply(reply, info);
1562
1563 unlock:
1564         ovs_unlock();
1565         return err;
1566 }
1567
1568 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1569 {
1570         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1571         struct datapath *dp;
1572         int skip = cb->args[0];
1573         int i = 0;
1574
1575         ovs_lock();
1576         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1577                 if (i >= skip &&
1578                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1579                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1580                                          OVS_DP_CMD_NEW) < 0)
1581                         break;
1582                 i++;
1583         }
1584         ovs_unlock();
1585
1586         cb->args[0] = i;
1587
1588         return skb->len;
1589 }
1590
1591 static struct genl_ops dp_datapath_genl_ops[] = {
1592         { .cmd = OVS_DP_CMD_NEW,
1593           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1594           .policy = datapath_policy,
1595           .doit = ovs_dp_cmd_new
1596         },
1597         { .cmd = OVS_DP_CMD_DEL,
1598           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1599           .policy = datapath_policy,
1600           .doit = ovs_dp_cmd_del
1601         },
1602         { .cmd = OVS_DP_CMD_GET,
1603           .flags = 0,               /* OK for unprivileged users. */
1604           .policy = datapath_policy,
1605           .doit = ovs_dp_cmd_get,
1606           .dumpit = ovs_dp_cmd_dump
1607         },
1608         { .cmd = OVS_DP_CMD_SET,
1609           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1610           .policy = datapath_policy,
1611           .doit = ovs_dp_cmd_set,
1612         },
1613 };
1614
1615 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1616         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1617         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1618         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1619         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1620         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1621         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1622 };
1623
1624 static struct genl_family dp_vport_genl_family = {
1625         .id = GENL_ID_GENERATE,
1626         .hdrsize = sizeof(struct ovs_header),
1627         .name = OVS_VPORT_FAMILY,
1628         .version = OVS_VPORT_VERSION,
1629         .maxattr = OVS_VPORT_ATTR_MAX,
1630         .netnsok = true,
1631         .parallel_ops = true,
1632 };
1633
1634 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1635         .name = OVS_VPORT_MCGROUP
1636 };
1637
1638 /* Called with ovs_mutex or RCU read lock. */
1639 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1640                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1641 {
1642         struct ovs_header *ovs_header;
1643         struct ovs_vport_stats vport_stats;
1644         int err;
1645
1646         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1647                                  flags, cmd);
1648         if (!ovs_header)
1649                 return -EMSGSIZE;
1650
1651         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1652
1653         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1654             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1655             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1656             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1657                 goto nla_put_failure;
1658
1659         ovs_vport_get_stats(vport, &vport_stats);
1660         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1661                     &vport_stats))
1662                 goto nla_put_failure;
1663
1664         err = ovs_vport_get_options(vport, skb);
1665         if (err == -EMSGSIZE)
1666                 goto error;
1667
1668         return genlmsg_end(skb, ovs_header);
1669
1670 nla_put_failure:
1671         err = -EMSGSIZE;
1672 error:
1673         genlmsg_cancel(skb, ovs_header);
1674         return err;
1675 }
1676
1677 /* Called with ovs_mutex or RCU read lock. */
1678 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1679                                          u32 seq, u8 cmd)
1680 {
1681         struct sk_buff *skb;
1682         int retval;
1683
1684         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1685         if (!skb)
1686                 return ERR_PTR(-ENOMEM);
1687
1688         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1689         BUG_ON(retval < 0);
1690
1691         return skb;
1692 }
1693
1694 /* Called with ovs_mutex or RCU read lock. */
1695 static struct vport *lookup_vport(struct net *net,
1696                                   struct ovs_header *ovs_header,
1697                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1698 {
1699         struct datapath *dp;
1700         struct vport *vport;
1701
1702         if (a[OVS_VPORT_ATTR_NAME]) {
1703                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1704                 if (!vport)
1705                         return ERR_PTR(-ENODEV);
1706                 if (ovs_header->dp_ifindex &&
1707                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1708                         return ERR_PTR(-ENODEV);
1709                 return vport;
1710         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1711                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1712
1713                 if (port_no >= DP_MAX_PORTS)
1714                         return ERR_PTR(-EFBIG);
1715
1716                 dp = get_dp(net, ovs_header->dp_ifindex);
1717                 if (!dp)
1718                         return ERR_PTR(-ENODEV);
1719
1720                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1721                 if (!vport)
1722                         return ERR_PTR(-ENODEV);
1723                 return vport;
1724         } else
1725                 return ERR_PTR(-EINVAL);
1726 }
1727
1728 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1729 {
1730         struct nlattr **a = info->attrs;
1731         struct ovs_header *ovs_header = info->userhdr;
1732         struct vport_parms parms;
1733         struct sk_buff *reply;
1734         struct vport *vport;
1735         struct datapath *dp;
1736         u32 port_no;
1737         int err;
1738
1739         err = -EINVAL;
1740         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1741             !a[OVS_VPORT_ATTR_UPCALL_PID])
1742                 goto exit;
1743
1744         ovs_lock();
1745         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1746         err = -ENODEV;
1747         if (!dp)
1748                 goto exit_unlock;
1749
1750         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1751                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1752
1753                 err = -EFBIG;
1754                 if (port_no >= DP_MAX_PORTS)
1755                         goto exit_unlock;
1756
1757                 vport = ovs_vport_ovsl(dp, port_no);
1758                 err = -EBUSY;
1759                 if (vport)
1760                         goto exit_unlock;
1761         } else {
1762                 for (port_no = 1; ; port_no++) {
1763                         if (port_no >= DP_MAX_PORTS) {
1764                                 err = -EFBIG;
1765                                 goto exit_unlock;
1766                         }
1767                         vport = ovs_vport_ovsl(dp, port_no);
1768                         if (!vport)
1769                                 break;
1770                 }
1771         }
1772
1773         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1774         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1775         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1776         parms.dp = dp;
1777         parms.port_no = port_no;
1778         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1779
1780         vport = new_vport(&parms);
1781         err = PTR_ERR(vport);
1782         if (IS_ERR(vport))
1783                 goto exit_unlock;
1784
1785         err = 0;
1786         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1787                                          OVS_VPORT_CMD_NEW);
1788         if (IS_ERR(reply)) {
1789                 err = PTR_ERR(reply);
1790                 ovs_dp_detach_port(vport);
1791                 goto exit_unlock;
1792         }
1793
1794         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1795
1796 exit_unlock:
1797         ovs_unlock();
1798 exit:
1799         return err;
1800 }
1801
1802 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1803 {
1804         struct nlattr **a = info->attrs;
1805         struct sk_buff *reply;
1806         struct vport *vport;
1807         int err;
1808
1809         ovs_lock();
1810         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1811         err = PTR_ERR(vport);
1812         if (IS_ERR(vport))
1813                 goto exit_unlock;
1814
1815         err = 0;
1816         if (a[OVS_VPORT_ATTR_TYPE] &&
1817             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1818                 err = -EINVAL;
1819
1820         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1821         if (!reply) {
1822                 err = -ENOMEM;
1823                 goto exit_unlock;
1824         }
1825
1826         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1827                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1828         if (err)
1829                 goto exit_free;
1830
1831         if (a[OVS_VPORT_ATTR_UPCALL_PID])
1832                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1833
1834         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1835                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1836         BUG_ON(err < 0);
1837
1838         ovs_unlock();
1839         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1840         return 0;
1841
1842         rtnl_unlock();
1843         return 0;
1844
1845 exit_free:
1846         kfree_skb(reply);
1847 exit_unlock:
1848         ovs_unlock();
1849         return err;
1850 }
1851
1852 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1853 {
1854         struct nlattr **a = info->attrs;
1855         struct sk_buff *reply;
1856         struct vport *vport;
1857         int err;
1858
1859         ovs_lock();
1860         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1861         err = PTR_ERR(vport);
1862         if (IS_ERR(vport))
1863                 goto exit_unlock;
1864
1865         if (vport->port_no == OVSP_LOCAL) {
1866                 err = -EINVAL;
1867                 goto exit_unlock;
1868         }
1869
1870         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1871                                          OVS_VPORT_CMD_DEL);
1872         err = PTR_ERR(reply);
1873         if (IS_ERR(reply))
1874                 goto exit_unlock;
1875
1876         err = 0;
1877         ovs_dp_detach_port(vport);
1878
1879         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1880
1881 exit_unlock:
1882         ovs_unlock();
1883         return err;
1884 }
1885
1886 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1887 {
1888         struct nlattr **a = info->attrs;
1889         struct ovs_header *ovs_header = info->userhdr;
1890         struct sk_buff *reply;
1891         struct vport *vport;
1892         int err;
1893
1894         rcu_read_lock();
1895         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1896         err = PTR_ERR(vport);
1897         if (IS_ERR(vport))
1898                 goto exit_unlock;
1899
1900         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1901                                          OVS_VPORT_CMD_NEW);
1902         err = PTR_ERR(reply);
1903         if (IS_ERR(reply))
1904                 goto exit_unlock;
1905
1906         rcu_read_unlock();
1907
1908         return genlmsg_reply(reply, info);
1909
1910 exit_unlock:
1911         rcu_read_unlock();
1912         return err;
1913 }
1914
1915 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1916 {
1917         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1918         struct datapath *dp;
1919         int bucket = cb->args[0], skip = cb->args[1];
1920         int i, j = 0;
1921
1922         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1923         if (!dp)
1924                 return -ENODEV;
1925
1926         rcu_read_lock();
1927         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1928                 struct vport *vport;
1929
1930                 j = 0;
1931                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1932                         if (j >= skip &&
1933                             ovs_vport_cmd_fill_info(vport, skb,
1934                                                     NETLINK_CB(cb->skb).portid,
1935                                                     cb->nlh->nlmsg_seq,
1936                                                     NLM_F_MULTI,
1937                                                     OVS_VPORT_CMD_NEW) < 0)
1938                                 goto out;
1939
1940                         j++;
1941                 }
1942                 skip = 0;
1943         }
1944 out:
1945         rcu_read_unlock();
1946
1947         cb->args[0] = i;
1948         cb->args[1] = j;
1949
1950         return skb->len;
1951 }
1952
1953 static struct genl_ops dp_vport_genl_ops[] = {
1954         { .cmd = OVS_VPORT_CMD_NEW,
1955           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1956           .policy = vport_policy,
1957           .doit = ovs_vport_cmd_new
1958         },
1959         { .cmd = OVS_VPORT_CMD_DEL,
1960           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1961           .policy = vport_policy,
1962           .doit = ovs_vport_cmd_del
1963         },
1964         { .cmd = OVS_VPORT_CMD_GET,
1965           .flags = 0,               /* OK for unprivileged users. */
1966           .policy = vport_policy,
1967           .doit = ovs_vport_cmd_get,
1968           .dumpit = ovs_vport_cmd_dump
1969         },
1970         { .cmd = OVS_VPORT_CMD_SET,
1971           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1972           .policy = vport_policy,
1973           .doit = ovs_vport_cmd_set,
1974         },
1975 };
1976
1977 struct genl_family_and_ops {
1978         struct genl_family *family;
1979         struct genl_ops *ops;
1980         int n_ops;
1981         struct genl_multicast_group *group;
1982 };
1983
1984 static const struct genl_family_and_ops dp_genl_families[] = {
1985         { &dp_datapath_genl_family,
1986           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1987           &ovs_dp_datapath_multicast_group },
1988         { &dp_vport_genl_family,
1989           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1990           &ovs_dp_vport_multicast_group },
1991         { &dp_flow_genl_family,
1992           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1993           &ovs_dp_flow_multicast_group },
1994         { &dp_packet_genl_family,
1995           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1996           NULL },
1997 };
1998
1999 static void dp_unregister_genl(int n_families)
2000 {
2001         int i;
2002
2003         for (i = 0; i < n_families; i++)
2004                 genl_unregister_family(dp_genl_families[i].family);
2005 }
2006
2007 static int dp_register_genl(void)
2008 {
2009         int n_registered;
2010         int err;
2011         int i;
2012
2013         n_registered = 0;
2014         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2015                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2016
2017                 err = genl_register_family_with_ops(f->family, f->ops,
2018                                                     f->n_ops);
2019                 if (err)
2020                         goto error;
2021                 n_registered++;
2022
2023                 if (f->group) {
2024                         err = genl_register_mc_group(f->family, f->group);
2025                         if (err)
2026                                 goto error;
2027                 }
2028         }
2029
2030         return 0;
2031
2032 error:
2033         dp_unregister_genl(n_registered);
2034         return err;
2035 }
2036
2037 static void rehash_flow_table(struct work_struct *work)
2038 {
2039         struct datapath *dp;
2040         struct net *net;
2041
2042         ovs_lock();
2043         rtnl_lock();
2044         for_each_net(net) {
2045                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2046
2047                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2048                         struct flow_table *old_table = ovsl_dereference(dp->table);
2049                         struct flow_table *new_table;
2050
2051                         new_table = ovs_flow_tbl_rehash(old_table);
2052                         if (!IS_ERR(new_table)) {
2053                                 rcu_assign_pointer(dp->table, new_table);
2054                                 ovs_flow_tbl_deferred_destroy(old_table);
2055                         }
2056                 }
2057         }
2058         rtnl_unlock();
2059         ovs_unlock();
2060         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2061 }
2062
2063 static int __net_init ovs_init_net(struct net *net)
2064 {
2065         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2066
2067         INIT_LIST_HEAD(&ovs_net->dps);
2068         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2069         return 0;
2070 }
2071
2072 static void __net_exit ovs_exit_net(struct net *net)
2073 {
2074         struct datapath *dp, *dp_next;
2075         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2076
2077         ovs_lock();
2078         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2079                 __dp_destroy(dp);
2080         ovs_unlock();
2081
2082         cancel_work_sync(&ovs_net->dp_notify_work);
2083 }
2084
2085 static struct pernet_operations ovs_net_ops = {
2086         .init = ovs_init_net,
2087         .exit = ovs_exit_net,
2088         .id   = &ovs_net_id,
2089         .size = sizeof(struct ovs_net),
2090 };
2091
2092 static int __init dp_init(void)
2093 {
2094         int err;
2095
2096         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2097
2098         pr_info("Open vSwitch switching datapath\n");
2099
2100         err = ovs_flow_init();
2101         if (err)
2102                 goto error;
2103
2104         err = ovs_vport_init();
2105         if (err)
2106                 goto error_flow_exit;
2107
2108         err = register_pernet_device(&ovs_net_ops);
2109         if (err)
2110                 goto error_vport_exit;
2111
2112         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2113         if (err)
2114                 goto error_netns_exit;
2115
2116         err = dp_register_genl();
2117         if (err < 0)
2118                 goto error_unreg_notifier;
2119
2120         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2121
2122         return 0;
2123
2124 error_unreg_notifier:
2125         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2126 error_netns_exit:
2127         unregister_pernet_device(&ovs_net_ops);
2128 error_vport_exit:
2129         ovs_vport_exit();
2130 error_flow_exit:
2131         ovs_flow_exit();
2132 error:
2133         return err;
2134 }
2135
2136 static void dp_cleanup(void)
2137 {
2138         cancel_delayed_work_sync(&rehash_flow_wq);
2139         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2140         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2141         unregister_pernet_device(&ovs_net_ops);
2142         rcu_barrier();
2143         ovs_vport_exit();
2144         ovs_flow_exit();
2145 }
2146
2147 module_init(dp_init);
2148 module_exit(dp_cleanup);
2149
2150 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2151 MODULE_LICENSE("GPL");