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, a[OVS_PACKET_ATTR_KEY]);
 743         if (err)
 744                 goto err_flow_free;
 745
 746         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
 747         if (err)
 748                 goto err_flow_free;
 749
 750         flow->hash = ovs_flow_hash(&flow->key, key_len);
 751
 752         acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 753         err = PTR_ERR(acts);
 754         if (IS_ERR(acts))
 755                 goto err_flow_free;
 756         rcu_assign_pointer(flow->sf_acts, acts);
 757
 758         OVS_CB(packet)->flow = flow;
 759         packet->priority = flow->key.phy.priority;
 760         packet->mark = flow->key.phy.skb_mark;
 761
 762         rcu_read_lock();
 763         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 764         err = -ENODEV;
 765         if (!dp)
 766                 goto err_unlock;
 767
 768         local_bh_disable();
 769         err = ovs_execute_actions(dp, packet);
 770         local_bh_enable();
 771         rcu_read_unlock();
 772
 773         ovs_flow_free(flow);
 774         return err;
 775
 776 err_unlock:
 777         rcu_read_unlock();
 778 err_flow_free:
 779         ovs_flow_free(flow);
 780 err_kfree_skb:
 781         kfree_skb(packet);
 782 err:
 783         return err;
 784 }
 785
 786 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 787         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 788         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 789         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 790 };
 791
 792 static struct genl_ops dp_packet_genl_ops[] = {
 793         { .cmd = OVS_PACKET_CMD_EXECUTE,
 794           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 795           .policy = packet_policy,
 796           .doit = ovs_packet_cmd_execute
 797         }
 798 };
 799
 800 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 801 {
 802         int i;
 803         struct flow_table *table = ovsl_dereference(dp->table);
 804
 805         stats->n_flows = ovs_flow_tbl_count(table);
 806
 807         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 808         for_each_possible_cpu(i) {
 809                 const struct dp_stats_percpu *percpu_stats;
 810                 struct dp_stats_percpu local_stats;
 811                 unsigned int start;
 812
 813                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 814
 815                 do {
 816                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 817                         local_stats = *percpu_stats;
 818                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 819
 820                 stats->n_hit += local_stats.n_hit;
 821                 stats->n_missed += local_stats.n_missed;
 822                 stats->n_lost += local_stats.n_lost;
 823         }
 824 }
 825
 826 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 827         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 828         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 829         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 830 };
 831
 832 static struct genl_family dp_flow_genl_family = {
 833         .id = GENL_ID_GENERATE,
 834         .hdrsize = sizeof(struct ovs_header),
 835         .name = OVS_FLOW_FAMILY,
 836         .version = OVS_FLOW_VERSION,
 837         .maxattr = OVS_FLOW_ATTR_MAX,
 838         .netnsok = true,
 839         .parallel_ops = true,
 840 };
 841
 842 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 843         .name = OVS_FLOW_MCGROUP
 844 };
 845
 846 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 847 {
 848         return NLMSG_ALIGN(sizeof(struct ovs_header))
 849                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 850                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 851                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 852                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 853                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 854 }
 855
 856 /* Called with ovs_mutex. */
 857 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 858                                   struct sk_buff *skb, u32 portid,
 859                                   u32 seq, u32 flags, u8 cmd)
 860 {
 861         const int skb_orig_len = skb->len;
 862         const struct sw_flow_actions *sf_acts;
 863         struct ovs_flow_stats stats;
 864         struct ovs_header *ovs_header;
 865         struct nlattr *nla;
 866         unsigned long used;
 867         u8 tcp_flags;
 868         int err;
 869
 870         sf_acts = ovsl_dereference(flow->sf_acts);
 871
 872         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 873         if (!ovs_header)
 874                 return -EMSGSIZE;
 875
 876         ovs_header->dp_ifindex = get_dpifindex(dp);
 877
 878         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 879         if (!nla)
 880                 goto nla_put_failure;
 881         err = ovs_flow_to_nlattrs(&flow->key, skb);
 882         if (err)
 883                 goto error;
 884         nla_nest_end(skb, nla);
 885
 886         spin_lock_bh(&flow->lock);
 887         used = flow->used;
 888         stats.n_packets = flow->packet_count;
 889         stats.n_bytes = flow->byte_count;
 890         tcp_flags = flow->tcp_flags;
 891         spin_unlock_bh(&flow->lock);
 892
 893         if (used &&
 894             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 895                 goto nla_put_failure;
 896
 897         if (stats.n_packets &&
 898             nla_put(skb, OVS_FLOW_ATTR_STATS,
 899                     sizeof(struct ovs_flow_stats), &stats))
 900                 goto nla_put_failure;
 901
 902         if (tcp_flags &&
 903             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
 904                 goto nla_put_failure;
 905
 906         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 907          * this is the first flow to be dumped into 'skb'.  This is unusual for
 908          * Netlink but individual action lists can be longer than
 909          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 910          * The userspace caller can always fetch the actions separately if it
 911          * really wants them.  (Most userspace callers in fact don't care.)
 912          *
 913          * This can only fail for dump operations because the skb is always
 914          * properly sized for single flows.
 915          */
 916         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 917                       sf_acts->actions);
 918         if (err < 0 && skb_orig_len)
 919                 goto error;
 920
 921         return genlmsg_end(skb, ovs_header);
 922
 923 nla_put_failure:
 924         err = -EMSGSIZE;
 925 error:
 926         genlmsg_cancel(skb, ovs_header);
 927         return err;
 928 }
 929
 930 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 931 {
 932         const struct sw_flow_actions *sf_acts;
 933
 934         sf_acts = ovsl_dereference(flow->sf_acts);
 935
 936         return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
 937 }
 938
 939 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 940                                                struct datapath *dp,
 941                                                u32 portid, u32 seq, u8 cmd)
 942 {
 943         struct sk_buff *skb;
 944         int retval;
 945
 946         skb = ovs_flow_cmd_alloc_info(flow);
 947         if (!skb)
 948                 return ERR_PTR(-ENOMEM);
 949
 950         retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
 951         BUG_ON(retval < 0);
 952         return skb;
 953 }
 954
 955 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 956 {
 957         struct nlattr **a = info->attrs;
 958         struct ovs_header *ovs_header = info->userhdr;
 959         struct sw_flow_key key;
 960         struct sw_flow *flow;
 961         struct sk_buff *reply;
 962         struct datapath *dp;
 963         struct flow_table *table;
 964         int error;
 965         int key_len;
 966
 967         /* Extract key. */
 968         error = -EINVAL;
 969         if (!a[OVS_FLOW_ATTR_KEY])
 970                 goto error;
 971         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
 972         if (error)
 973                 goto error;
 974
 975         /* Validate actions. */
 976         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 977                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
 978                 if (error)
 979                         goto error;
 980         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 981                 error = -EINVAL;
 982                 goto error;
 983         }
 984
 985         ovs_lock();
 986         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 987         error = -ENODEV;
 988         if (!dp)
 989                 goto err_unlock_ovs;
 990
 991         table = ovsl_dereference(dp->table);
 992         flow = ovs_flow_tbl_lookup(table, &key, key_len);
 993         if (!flow) {
 994                 struct sw_flow_actions *acts;
 995
 996                 /* Bail out if we're not allowed to create a new flow. */
 997                 error = -ENOENT;
 998                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
 999                         goto err_unlock_ovs;
1000
1001                 /* Expand table, if necessary, to make room. */
1002                 if (ovs_flow_tbl_need_to_expand(table)) {
1003                         struct flow_table *new_table;
1004
1005                         new_table = ovs_flow_tbl_expand(table);
1006                         if (!IS_ERR(new_table)) {
1007                                 rcu_assign_pointer(dp->table, new_table);
1008                                 ovs_flow_tbl_deferred_destroy(table);
1009                                 table = ovsl_dereference(dp->table);
1010                         }
1011                 }
1012
1013                 /* Allocate flow. */
1014                 flow = ovs_flow_alloc();
1015                 if (IS_ERR(flow)) {
1016                         error = PTR_ERR(flow);
1017                         goto err_unlock_ovs;
1018                 }
1019                 flow->key = key;
1020                 clear_stats(flow);
1021
1022                 /* Obtain actions. */
1023                 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1024                 error = PTR_ERR(acts);
1025                 if (IS_ERR(acts))
1026                         goto error_free_flow;
1027                 rcu_assign_pointer(flow->sf_acts, acts);
1028
1029                 /* Put flow in bucket. */
1030                 flow->hash = ovs_flow_hash(&key, key_len);
1031                 ovs_flow_tbl_insert(table, flow);
1032
1033                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1034                                                 info->snd_seq,
1035                                                 OVS_FLOW_CMD_NEW);
1036         } else {
1037                 /* We found a matching flow. */
1038                 struct sw_flow_actions *old_acts;
1039                 struct nlattr *acts_attrs;
1040
1041                 /* Bail out if we're not allowed to modify an existing flow.
1042                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1043                  * because Generic Netlink treats the latter as a dump
1044                  * request.  We also accept NLM_F_EXCL in case that bug ever
1045                  * gets fixed.
1046                  */
1047                 error = -EEXIST;
1048                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1049                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1050                         goto err_unlock_ovs;
1051
1052                 /* Update actions. */
1053                 old_acts = ovsl_dereference(flow->sf_acts);
1054                 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1055                 if (acts_attrs &&
1056                    (old_acts->actions_len != nla_len(acts_attrs) ||
1057                    memcmp(old_acts->actions, nla_data(acts_attrs),
1058                           old_acts->actions_len))) {
1059                         struct sw_flow_actions *new_acts;
1060
1061                         new_acts = ovs_flow_actions_alloc(acts_attrs);
1062                         error = PTR_ERR(new_acts);
1063                         if (IS_ERR(new_acts))
1064                                 goto err_unlock_ovs;
1065
1066                         rcu_assign_pointer(flow->sf_acts, new_acts);
1067                         ovs_flow_deferred_free_acts(old_acts);
1068                 }
1069
1070                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1071                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1072
1073                 /* Clear stats. */
1074                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1075                         spin_lock_bh(&flow->lock);
1076                         clear_stats(flow);
1077                         spin_unlock_bh(&flow->lock);
1078                 }
1079         }
1080         ovs_unlock();
1081
1082         if (!IS_ERR(reply))
1083                 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1084         else
1085                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1086                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1087         return 0;
1088
1089 error_free_flow:
1090         ovs_flow_free(flow);
1091 err_unlock_ovs:
1092         ovs_unlock();
1093 error:
1094         return error;
1095 }
1096
1097 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1098 {
1099         struct nlattr **a = info->attrs;
1100         struct ovs_header *ovs_header = info->userhdr;
1101         struct sw_flow_key key;
1102         struct sk_buff *reply;
1103         struct sw_flow *flow;
1104         struct datapath *dp;
1105         struct flow_table *table;
1106         int err;
1107         int key_len;
1108
1109         if (!a[OVS_FLOW_ATTR_KEY])
1110                 return -EINVAL;
1111         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1112         if (err)
1113                 return err;
1114
1115         ovs_lock();
1116         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1117         if (!dp) {
1118                 err = -ENODEV;
1119                 goto unlock;
1120         }
1121
1122         table = ovsl_dereference(dp->table);
1123         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1124         if (!flow) {
1125                 err = -ENOENT;
1126                 goto unlock;
1127         }
1128
1129         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1130                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1131         if (IS_ERR(reply)) {
1132                 err = PTR_ERR(reply);
1133                 goto unlock;
1134         }
1135
1136         ovs_unlock();
1137         return genlmsg_reply(reply, info);
1138 unlock:
1139         ovs_unlock();
1140         return err;
1141 }
1142
1143 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1144 {
1145         struct nlattr **a = info->attrs;
1146         struct ovs_header *ovs_header = info->userhdr;
1147         struct sw_flow_key key;
1148         struct sk_buff *reply;
1149         struct sw_flow *flow;
1150         struct datapath *dp;
1151         struct flow_table *table;
1152         int err;
1153         int key_len;
1154
1155         ovs_lock();
1156         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1157         if (!dp) {
1158                 err = -ENODEV;
1159                 goto unlock;
1160         }
1161
1162         if (!a[OVS_FLOW_ATTR_KEY]) {
1163                 err = flush_flows(dp);
1164                 goto unlock;
1165         }
1166         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1167         if (err)
1168                 goto unlock;
1169
1170         table = ovsl_dereference(dp->table);
1171         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1172         if (!flow) {
1173                 err = -ENOENT;
1174                 goto unlock;
1175         }
1176
1177         reply = ovs_flow_cmd_alloc_info(flow);
1178         if (!reply) {
1179                 err = -ENOMEM;
1180                 goto unlock;
1181         }
1182
1183         ovs_flow_tbl_remove(table, flow);
1184
1185         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1186                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1187         BUG_ON(err < 0);
1188
1189         ovs_flow_deferred_free(flow);
1190         ovs_unlock();
1191
1192         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1193         return 0;
1194 unlock:
1195         ovs_unlock();
1196         return err;
1197 }
1198
1199 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1200 {
1201         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1202         struct datapath *dp;
1203         struct flow_table *table;
1204
1205         ovs_lock();
1206         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1207         if (!dp) {
1208                 ovs_unlock();
1209                 return -ENODEV;
1210         }
1211
1212         table = ovsl_dereference(dp->table);
1213
1214         for (;;) {
1215                 struct sw_flow *flow;
1216                 u32 bucket, obj;
1217
1218                 bucket = cb->args[0];
1219                 obj = cb->args[1];
1220                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1221                 if (!flow)
1222                         break;
1223
1224                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1225                                            NETLINK_CB(cb->skb).portid,
1226                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1227                                            OVS_FLOW_CMD_NEW) < 0)
1228                         break;
1229
1230                 cb->args[0] = bucket;
1231                 cb->args[1] = obj;
1232         }
1233         ovs_unlock();
1234         return skb->len;
1235 }
1236
1237 static struct genl_ops dp_flow_genl_ops[] = {
1238         { .cmd = OVS_FLOW_CMD_NEW,
1239           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1240           .policy = flow_policy,
1241           .doit = ovs_flow_cmd_new_or_set
1242         },
1243         { .cmd = OVS_FLOW_CMD_DEL,
1244           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1245           .policy = flow_policy,
1246           .doit = ovs_flow_cmd_del
1247         },
1248         { .cmd = OVS_FLOW_CMD_GET,
1249           .flags = 0,               /* OK for unprivileged users. */
1250           .policy = flow_policy,
1251           .doit = ovs_flow_cmd_get,
1252           .dumpit = ovs_flow_cmd_dump
1253         },
1254         { .cmd = OVS_FLOW_CMD_SET,
1255           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1256           .policy = flow_policy,
1257           .doit = ovs_flow_cmd_new_or_set,
1258         },
1259 };
1260
1261 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1262         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1263         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1264 };
1265
1266 static struct genl_family dp_datapath_genl_family = {
1267         .id = GENL_ID_GENERATE,
1268         .hdrsize = sizeof(struct ovs_header),
1269         .name = OVS_DATAPATH_FAMILY,
1270         .version = OVS_DATAPATH_VERSION,
1271         .maxattr = OVS_DP_ATTR_MAX,
1272         .netnsok = true,
1273         .parallel_ops = true,
1274 };
1275
1276 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1277         .name = OVS_DATAPATH_MCGROUP
1278 };
1279
1280 static size_t ovs_dp_cmd_msg_size(void)
1281 {
1282         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1283
1284         msgsize += nla_total_size(IFNAMSIZ);
1285         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1286
1287         return msgsize;
1288 }
1289
1290 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1291                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1292 {
1293         struct ovs_header *ovs_header;
1294         struct ovs_dp_stats dp_stats;
1295         int err;
1296
1297         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1298                                    flags, cmd);
1299         if (!ovs_header)
1300                 goto error;
1301
1302         ovs_header->dp_ifindex = get_dpifindex(dp);
1303
1304         rcu_read_lock();
1305         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1306         rcu_read_unlock();
1307         if (err)
1308                 goto nla_put_failure;
1309
1310         get_dp_stats(dp, &dp_stats);
1311         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1312                 goto nla_put_failure;
1313
1314         return genlmsg_end(skb, ovs_header);
1315
1316 nla_put_failure:
1317         genlmsg_cancel(skb, ovs_header);
1318 error:
1319         return -EMSGSIZE;
1320 }
1321
1322 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1323                                              u32 seq, u8 cmd)
1324 {
1325         struct sk_buff *skb;
1326         int retval;
1327
1328         skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1329         if (!skb)
1330                 return ERR_PTR(-ENOMEM);
1331
1332         retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1333         if (retval < 0) {
1334                 kfree_skb(skb);
1335                 return ERR_PTR(retval);
1336         }
1337         return skb;
1338 }
1339
1340 /* Called with ovs_mutex. */
1341 static struct datapath *lookup_datapath(struct net *net,
1342                                         struct ovs_header *ovs_header,
1343                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1344 {
1345         struct datapath *dp;
1346
1347         if (!a[OVS_DP_ATTR_NAME])
1348                 dp = get_dp(net, ovs_header->dp_ifindex);
1349         else {
1350                 struct vport *vport;
1351
1352                 rcu_read_lock();
1353                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1354                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1355                 rcu_read_unlock();
1356         }
1357         return dp ? dp : ERR_PTR(-ENODEV);
1358 }
1359
1360 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1361 {
1362         struct nlattr **a = info->attrs;
1363         struct vport_parms parms;
1364         struct sk_buff *reply;
1365         struct datapath *dp;
1366         struct vport *vport;
1367         struct ovs_net *ovs_net;
1368         int err, i;
1369
1370         err = -EINVAL;
1371         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1372                 goto err;
1373
1374         ovs_lock();
1375
1376         err = -ENOMEM;
1377         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1378         if (dp == NULL)
1379                 goto err_unlock_ovs;
1380
1381         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1382
1383         /* Allocate table. */
1384         err = -ENOMEM;
1385         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1386         if (!dp->table)
1387                 goto err_free_dp;
1388
1389         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1390         if (!dp->stats_percpu) {
1391                 err = -ENOMEM;
1392                 goto err_destroy_table;
1393         }
1394
1395         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1396                         GFP_KERNEL);
1397         if (!dp->ports) {
1398                 err = -ENOMEM;
1399                 goto err_destroy_percpu;
1400         }
1401
1402         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1403                 INIT_HLIST_HEAD(&dp->ports[i]);
1404
1405         /* Set up our datapath device. */
1406         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1407         parms.type = OVS_VPORT_TYPE_INTERNAL;
1408         parms.options = NULL;
1409         parms.dp = dp;
1410         parms.port_no = OVSP_LOCAL;
1411         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1412
1413         vport = new_vport(&parms);
1414         if (IS_ERR(vport)) {
1415                 err = PTR_ERR(vport);
1416                 if (err == -EBUSY)
1417                         err = -EEXIST;
1418
1419                 goto err_destroy_ports_array;
1420         }
1421
1422         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1423                                       info->snd_seq, OVS_DP_CMD_NEW);
1424         err = PTR_ERR(reply);
1425         if (IS_ERR(reply))
1426                 goto err_destroy_local_port;
1427
1428         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1429         list_add_tail(&dp->list_node, &ovs_net->dps);
1430
1431         ovs_unlock();
1432
1433         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1434         return 0;
1435
1436 err_destroy_local_port:
1437         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1438 err_destroy_ports_array:
1439         kfree(dp->ports);
1440 err_destroy_percpu:
1441         free_percpu(dp->stats_percpu);
1442 err_destroy_table:
1443         ovs_flow_tbl_destroy(ovsl_dereference(dp->table));
1444 err_free_dp:
1445         release_net(ovs_dp_get_net(dp));
1446         kfree(dp);
1447 err_unlock_ovs:
1448         ovs_unlock();
1449 err:
1450         return err;
1451 }
1452
1453 /* Called with ovs_mutex. */
1454 static void __dp_destroy(struct datapath *dp)
1455 {
1456         int i;
1457
1458         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1459                 struct vport *vport;
1460                 struct hlist_node *n;
1461
1462                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1463                         if (vport->port_no != OVSP_LOCAL)
1464                                 ovs_dp_detach_port(vport);
1465         }
1466
1467         list_del(&dp->list_node);
1468
1469         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1470          * all port in datapath are destroyed first before freeing datapath.
1471          */
1472         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1473
1474         call_rcu(&dp->rcu, destroy_dp_rcu);
1475 }
1476
1477 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1478 {
1479         struct sk_buff *reply;
1480         struct datapath *dp;
1481         int err;
1482
1483         ovs_lock();
1484         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1485         err = PTR_ERR(dp);
1486         if (IS_ERR(dp))
1487                 goto unlock;
1488
1489         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1490                                       info->snd_seq, OVS_DP_CMD_DEL);
1491         err = PTR_ERR(reply);
1492         if (IS_ERR(reply))
1493                 goto unlock;
1494
1495         __dp_destroy(dp);
1496         ovs_unlock();
1497
1498         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1499
1500         return 0;
1501 unlock:
1502         ovs_unlock();
1503         return err;
1504 }
1505
1506 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1507 {
1508         struct sk_buff *reply;
1509         struct datapath *dp;
1510         int err;
1511
1512         ovs_lock();
1513         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1514         err = PTR_ERR(dp);
1515         if (IS_ERR(dp))
1516                 goto unlock;
1517
1518         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1519                                       info->snd_seq, OVS_DP_CMD_NEW);
1520         if (IS_ERR(reply)) {
1521                 err = PTR_ERR(reply);
1522                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1523                                 ovs_dp_datapath_multicast_group.id, err);
1524                 err = 0;
1525                 goto unlock;
1526         }
1527
1528         ovs_unlock();
1529         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1530
1531         return 0;
1532 unlock:
1533         ovs_unlock();
1534         return err;
1535 }
1536
1537 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1538 {
1539         struct sk_buff *reply;
1540         struct datapath *dp;
1541         int err;
1542
1543         ovs_lock();
1544         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1545         if (IS_ERR(dp)) {
1546                 err = PTR_ERR(dp);
1547                 goto unlock;
1548         }
1549
1550         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1551                                       info->snd_seq, OVS_DP_CMD_NEW);
1552         if (IS_ERR(reply)) {
1553                 err = PTR_ERR(reply);
1554                 goto unlock;
1555         }
1556
1557         ovs_unlock();
1558         return genlmsg_reply(reply, info);
1559
1560 unlock:
1561         ovs_unlock();
1562         return err;
1563 }
1564
1565 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1566 {
1567         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1568         struct datapath *dp;
1569         int skip = cb->args[0];
1570         int i = 0;
1571
1572         ovs_lock();
1573         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1574                 if (i >= skip &&
1575                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1576                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1577                                          OVS_DP_CMD_NEW) < 0)
1578                         break;
1579                 i++;
1580         }
1581         ovs_unlock();
1582
1583         cb->args[0] = i;
1584
1585         return skb->len;
1586 }
1587
1588 static struct genl_ops dp_datapath_genl_ops[] = {
1589         { .cmd = OVS_DP_CMD_NEW,
1590           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1591           .policy = datapath_policy,
1592           .doit = ovs_dp_cmd_new
1593         },
1594         { .cmd = OVS_DP_CMD_DEL,
1595           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1596           .policy = datapath_policy,
1597           .doit = ovs_dp_cmd_del
1598         },
1599         { .cmd = OVS_DP_CMD_GET,
1600           .flags = 0,               /* OK for unprivileged users. */
1601           .policy = datapath_policy,
1602           .doit = ovs_dp_cmd_get,
1603           .dumpit = ovs_dp_cmd_dump
1604         },
1605         { .cmd = OVS_DP_CMD_SET,
1606           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1607           .policy = datapath_policy,
1608           .doit = ovs_dp_cmd_set,
1609         },
1610 };
1611
1612 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1613         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1614         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1615         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1616         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1617         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1618         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1619 };
1620
1621 static struct genl_family dp_vport_genl_family = {
1622         .id = GENL_ID_GENERATE,
1623         .hdrsize = sizeof(struct ovs_header),
1624         .name = OVS_VPORT_FAMILY,
1625         .version = OVS_VPORT_VERSION,
1626         .maxattr = OVS_VPORT_ATTR_MAX,
1627         .netnsok = true,
1628         .parallel_ops = true,
1629 };
1630
1631 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1632         .name = OVS_VPORT_MCGROUP
1633 };
1634
1635 /* Called with ovs_mutex or RCU read lock. */
1636 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1637                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1638 {
1639         struct ovs_header *ovs_header;
1640         struct ovs_vport_stats vport_stats;
1641         int err;
1642
1643         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1644                                  flags, cmd);
1645         if (!ovs_header)
1646                 return -EMSGSIZE;
1647
1648         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1649
1650         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1651             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1652             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1653             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1654                 goto nla_put_failure;
1655
1656         ovs_vport_get_stats(vport, &vport_stats);
1657         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1658                     &vport_stats))
1659                 goto nla_put_failure;
1660
1661         err = ovs_vport_get_options(vport, skb);
1662         if (err == -EMSGSIZE)
1663                 goto error;
1664
1665         return genlmsg_end(skb, ovs_header);
1666
1667 nla_put_failure:
1668         err = -EMSGSIZE;
1669 error:
1670         genlmsg_cancel(skb, ovs_header);
1671         return err;
1672 }
1673
1674 /* Called with ovs_mutex or RCU read lock. */
1675 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1676                                          u32 seq, u8 cmd)
1677 {
1678         struct sk_buff *skb;
1679         int retval;
1680
1681         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1682         if (!skb)
1683                 return ERR_PTR(-ENOMEM);
1684
1685         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1686         BUG_ON(retval < 0);
1687
1688         return skb;
1689 }
1690
1691 /* Called with ovs_mutex or RCU read lock. */
1692 static struct vport *lookup_vport(struct net *net,
1693                                   struct ovs_header *ovs_header,
1694                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1695 {
1696         struct datapath *dp;
1697         struct vport *vport;
1698
1699         if (a[OVS_VPORT_ATTR_NAME]) {
1700                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1701                 if (!vport)
1702                         return ERR_PTR(-ENODEV);
1703                 if (ovs_header->dp_ifindex &&
1704                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1705                         return ERR_PTR(-ENODEV);
1706                 return vport;
1707         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1708                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1709
1710                 if (port_no >= DP_MAX_PORTS)
1711                         return ERR_PTR(-EFBIG);
1712
1713                 dp = get_dp(net, ovs_header->dp_ifindex);
1714                 if (!dp)
1715                         return ERR_PTR(-ENODEV);
1716
1717                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1718                 if (!vport)
1719                         return ERR_PTR(-ENODEV);
1720                 return vport;
1721         } else
1722                 return ERR_PTR(-EINVAL);
1723 }
1724
1725 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1726 {
1727         struct nlattr **a = info->attrs;
1728         struct ovs_header *ovs_header = info->userhdr;
1729         struct vport_parms parms;
1730         struct sk_buff *reply;
1731         struct vport *vport;
1732         struct datapath *dp;
1733         u32 port_no;
1734         int err;
1735
1736         err = -EINVAL;
1737         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1738             !a[OVS_VPORT_ATTR_UPCALL_PID])
1739                 goto exit;
1740
1741         ovs_lock();
1742         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1743         err = -ENODEV;
1744         if (!dp)
1745                 goto exit_unlock;
1746
1747         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1748                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1749
1750                 err = -EFBIG;
1751                 if (port_no >= DP_MAX_PORTS)
1752                         goto exit_unlock;
1753
1754                 vport = ovs_vport_ovsl(dp, port_no);
1755                 err = -EBUSY;
1756                 if (vport)
1757                         goto exit_unlock;
1758         } else {
1759                 for (port_no = 1; ; port_no++) {
1760                         if (port_no >= DP_MAX_PORTS) {
1761                                 err = -EFBIG;
1762                                 goto exit_unlock;
1763                         }
1764                         vport = ovs_vport_ovsl(dp, port_no);
1765                         if (!vport)
1766                                 break;
1767                 }
1768         }
1769
1770         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1771         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1772         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1773         parms.dp = dp;
1774         parms.port_no = port_no;
1775         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1776
1777         vport = new_vport(&parms);
1778         err = PTR_ERR(vport);
1779         if (IS_ERR(vport))
1780                 goto exit_unlock;
1781
1782         err = 0;
1783         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1784                                          OVS_VPORT_CMD_NEW);
1785         if (IS_ERR(reply)) {
1786                 err = PTR_ERR(reply);
1787                 ovs_dp_detach_port(vport);
1788                 goto exit_unlock;
1789         }
1790
1791         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1792
1793 exit_unlock:
1794         ovs_unlock();
1795 exit:
1796         return err;
1797 }
1798
1799 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1800 {
1801         struct nlattr **a = info->attrs;
1802         struct sk_buff *reply;
1803         struct vport *vport;
1804         int err;
1805
1806         ovs_lock();
1807         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1808         err = PTR_ERR(vport);
1809         if (IS_ERR(vport))
1810                 goto exit_unlock;
1811
1812         if (a[OVS_VPORT_ATTR_TYPE] &&
1813             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1814                 err = -EINVAL;
1815                 goto exit_unlock;
1816         }
1817
1818         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1819         if (!reply) {
1820                 err = -ENOMEM;
1821                 goto exit_unlock;
1822         }
1823
1824         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1825                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1826                 if (err)
1827                         goto exit_free;
1828         }
1829
1830         if (a[OVS_VPORT_ATTR_UPCALL_PID])
1831                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1832
1833         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1834                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1835         BUG_ON(err < 0);
1836
1837         ovs_unlock();
1838         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1839         return 0;
1840
1841         rtnl_unlock();
1842         return 0;
1843
1844 exit_free:
1845         kfree_skb(reply);
1846 exit_unlock:
1847         ovs_unlock();
1848         return err;
1849 }
1850
1851 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1852 {
1853         struct nlattr **a = info->attrs;
1854         struct sk_buff *reply;
1855         struct vport *vport;
1856         int err;
1857
1858         ovs_lock();
1859         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1860         err = PTR_ERR(vport);
1861         if (IS_ERR(vport))
1862                 goto exit_unlock;
1863
1864         if (vport->port_no == OVSP_LOCAL) {
1865                 err = -EINVAL;
1866                 goto exit_unlock;
1867         }
1868
1869         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1870                                          OVS_VPORT_CMD_DEL);
1871         err = PTR_ERR(reply);
1872         if (IS_ERR(reply))
1873                 goto exit_unlock;
1874
1875         err = 0;
1876         ovs_dp_detach_port(vport);
1877
1878         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1879
1880 exit_unlock:
1881         ovs_unlock();
1882         return err;
1883 }
1884
1885 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1886 {
1887         struct nlattr **a = info->attrs;
1888         struct ovs_header *ovs_header = info->userhdr;
1889         struct sk_buff *reply;
1890         struct vport *vport;
1891         int err;
1892
1893         rcu_read_lock();
1894         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1895         err = PTR_ERR(vport);
1896         if (IS_ERR(vport))
1897                 goto exit_unlock;
1898
1899         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1900                                          OVS_VPORT_CMD_NEW);
1901         err = PTR_ERR(reply);
1902         if (IS_ERR(reply))
1903                 goto exit_unlock;
1904
1905         rcu_read_unlock();
1906
1907         return genlmsg_reply(reply, info);
1908
1909 exit_unlock:
1910         rcu_read_unlock();
1911         return err;
1912 }
1913
1914 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1915 {
1916         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1917         struct datapath *dp;
1918         int bucket = cb->args[0], skip = cb->args[1];
1919         int i, j = 0;
1920
1921         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1922         if (!dp)
1923                 return -ENODEV;
1924
1925         rcu_read_lock();
1926         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1927                 struct vport *vport;
1928
1929                 j = 0;
1930                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1931                         if (j >= skip &&
1932                             ovs_vport_cmd_fill_info(vport, skb,
1933                                                     NETLINK_CB(cb->skb).portid,
1934                                                     cb->nlh->nlmsg_seq,
1935                                                     NLM_F_MULTI,
1936                                                     OVS_VPORT_CMD_NEW) < 0)
1937                                 goto out;
1938
1939                         j++;
1940                 }
1941                 skip = 0;
1942         }
1943 out:
1944         rcu_read_unlock();
1945
1946         cb->args[0] = i;
1947         cb->args[1] = j;
1948
1949         return skb->len;
1950 }
1951
1952 static struct genl_ops dp_vport_genl_ops[] = {
1953         { .cmd = OVS_VPORT_CMD_NEW,
1954           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1955           .policy = vport_policy,
1956           .doit = ovs_vport_cmd_new
1957         },
1958         { .cmd = OVS_VPORT_CMD_DEL,
1959           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1960           .policy = vport_policy,
1961           .doit = ovs_vport_cmd_del
1962         },
1963         { .cmd = OVS_VPORT_CMD_GET,
1964           .flags = 0,               /* OK for unprivileged users. */
1965           .policy = vport_policy,
1966           .doit = ovs_vport_cmd_get,
1967           .dumpit = ovs_vport_cmd_dump
1968         },
1969         { .cmd = OVS_VPORT_CMD_SET,
1970           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1971           .policy = vport_policy,
1972           .doit = ovs_vport_cmd_set,
1973         },
1974 };
1975
1976 struct genl_family_and_ops {
1977         struct genl_family *family;
1978         struct genl_ops *ops;
1979         int n_ops;
1980         struct genl_multicast_group *group;
1981 };
1982
1983 static const struct genl_family_and_ops dp_genl_families[] = {
1984         { &dp_datapath_genl_family,
1985           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1986           &ovs_dp_datapath_multicast_group },
1987         { &dp_vport_genl_family,
1988           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1989           &ovs_dp_vport_multicast_group },
1990         { &dp_flow_genl_family,
1991           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1992           &ovs_dp_flow_multicast_group },
1993         { &dp_packet_genl_family,
1994           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1995           NULL },
1996 };
1997
1998 static void dp_unregister_genl(int n_families)
1999 {
2000         int i;
2001
2002         for (i = 0; i < n_families; i++)
2003                 genl_unregister_family(dp_genl_families[i].family);
2004 }
2005
2006 static int dp_register_genl(void)
2007 {
2008         int n_registered;
2009         int err;
2010         int i;
2011
2012         n_registered = 0;
2013         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2014                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2015
2016                 err = genl_register_family_with_ops(f->family, f->ops,
2017                                                     f->n_ops);
2018                 if (err)
2019                         goto error;
2020                 n_registered++;
2021
2022                 if (f->group) {
2023                         err = genl_register_mc_group(f->family, f->group);
2024                         if (err)
2025                                 goto error;
2026                 }
2027         }
2028
2029         return 0;
2030
2031 error:
2032         dp_unregister_genl(n_registered);
2033         return err;
2034 }
2035
2036 static void rehash_flow_table(struct work_struct *work)
2037 {
2038         struct datapath *dp;
2039         struct net *net;
2040
2041         ovs_lock();
2042         rtnl_lock();
2043         for_each_net(net) {
2044                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2045
2046                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2047                         struct flow_table *old_table = ovsl_dereference(dp->table);
2048                         struct flow_table *new_table;
2049
2050                         new_table = ovs_flow_tbl_rehash(old_table);
2051                         if (!IS_ERR(new_table)) {
2052                                 rcu_assign_pointer(dp->table, new_table);
2053                                 ovs_flow_tbl_deferred_destroy(old_table);
2054                         }
2055                 }
2056         }
2057         rtnl_unlock();
2058         ovs_unlock();
2059         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2060 }
2061
2062 static int __net_init ovs_init_net(struct net *net)
2063 {
2064         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2065
2066         INIT_LIST_HEAD(&ovs_net->dps);
2067         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2068         return 0;
2069 }
2070
2071 static void __net_exit ovs_exit_net(struct net *net)
2072 {
2073         struct datapath *dp, *dp_next;
2074         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2075
2076         ovs_lock();
2077         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2078                 __dp_destroy(dp);
2079         ovs_unlock();
2080
2081         cancel_work_sync(&ovs_net->dp_notify_work);
2082 }
2083
2084 static struct pernet_operations ovs_net_ops = {
2085         .init = ovs_init_net,
2086         .exit = ovs_exit_net,
2087         .id   = &ovs_net_id,
2088         .size = sizeof(struct ovs_net),
2089 };
2090
2091 static int __init dp_init(void)
2092 {
2093         int err;
2094
2095         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2096
2097         pr_info("Open vSwitch switching datapath\n");
2098
2099         err = ovs_flow_init();
2100         if (err)
2101                 goto error;
2102
2103         err = ovs_vport_init();
2104         if (err)
2105                 goto error_flow_exit;
2106
2107         err = register_pernet_device(&ovs_net_ops);
2108         if (err)
2109                 goto error_vport_exit;
2110
2111         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2112         if (err)
2113                 goto error_netns_exit;
2114
2115         err = dp_register_genl();
2116         if (err < 0)
2117                 goto error_unreg_notifier;
2118
2119         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2120
2121         return 0;
2122
2123 error_unreg_notifier:
2124         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2125 error_netns_exit:
2126         unregister_pernet_device(&ovs_net_ops);
2127 error_vport_exit:
2128         ovs_vport_exit();
2129 error_flow_exit:
2130         ovs_flow_exit();
2131 error:
2132         return err;
2133 }
2134
2135 static void dp_cleanup(void)
2136 {
2137         cancel_delayed_work_sync(&rehash_flow_wq);
2138         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2139         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2140         unregister_pernet_device(&ovs_net_ops);
2141         rcu_barrier();
2142         ovs_vport_exit();
2143         ovs_flow_exit();
2144 }
2145
2146 module_init(dp_init);
2147 module_exit(dp_cleanup);
2148
2149 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2150 MODULE_LICENSE("GPL");