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