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 struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa, int attr_len)
 468 {
 469
 470         struct sw_flow_actions *acts;
 471         int new_acts_size;
 472         int req_size = NLA_ALIGN(attr_len);
 473         int next_offset = offsetof(struct sw_flow_actions, actions) +
 474                                         (*sfa)->actions_len;
 475
 476         if (req_size <= (ksize(*sfa) - next_offset))
 477                 goto out;
 478
 479         new_acts_size = ksize(*sfa) * 2;
 480
 481         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
 482                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
 483                         return ERR_PTR(-EMSGSIZE);
 484                 new_acts_size = MAX_ACTIONS_BUFSIZE;
 485         }
 486
 487         acts = ovs_flow_actions_alloc(new_acts_size);
 488         if (IS_ERR(acts))
 489                 return (void *)acts;
 490
 491         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
 492         acts->actions_len = (*sfa)->actions_len;
 493         kfree(*sfa);
 494         *sfa = acts;
 495
 496 out:
 497         (*sfa)->actions_len += req_size;
 498         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
 499 }
 500
 501 static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, int len)
 502 {
 503         struct nlattr *a;
 504
 505         a = reserve_sfa_size(sfa, nla_attr_size(len));
 506         if (IS_ERR(a))
 507                 return PTR_ERR(a);
 508
 509         a->nla_type = attrtype;
 510         a->nla_len = nla_attr_size(len);
 511
 512         if (data)
 513                 memcpy(nla_data(a), data, len);
 514         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
 515
 516         return 0;
 517 }
 518
 519 static inline int add_nested_action_start(struct sw_flow_actions **sfa, int attrtype)
 520 {
 521         int used = (*sfa)->actions_len;
 522         int err;
 523
 524         err = add_action(sfa, attrtype, NULL, 0);
 525         if (err)
 526                 return err;
 527
 528         return used;
 529 }
 530
 531 static inline void add_nested_action_end(struct sw_flow_actions *sfa, int st_offset)
 532 {
 533         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions + st_offset);
 534
 535         a->nla_len = sfa->actions_len - st_offset;
 536 }
 537
 538 static int validate_and_copy_actions(const struct nlattr *attr,
 539                                      const struct sw_flow_key *key, int depth,
 540                                      struct sw_flow_actions **sfa);
 541
 542 static int validate_and_copy_sample(const struct nlattr *attr,
 543                                     const struct sw_flow_key *key, int depth,
 544                                     struct sw_flow_actions **sfa)
 545 {
 546         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 547         const struct nlattr *probability, *actions;
 548         const struct nlattr *a;
 549         int rem, start, err, st_acts;
 550
 551         memset(attrs, 0, sizeof(attrs));
 552         nla_for_each_nested(a, attr, rem) {
 553                 int type = nla_type(a);
 554                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 555                         return -EINVAL;
 556                 attrs[type] = a;
 557         }
 558         if (rem)
 559                 return -EINVAL;
 560
 561         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 562         if (!probability || nla_len(probability) != sizeof(u32))
 563                 return -EINVAL;
 564
 565         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 566         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 567                 return -EINVAL;
 568
 569         /* validation done, copy sample action. */
 570         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
 571         if (start < 0)
 572                 return start;
 573         err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY, nla_data(probability), sizeof(u32));
 574         if (err)
 575                 return err;
 576         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
 577         if (st_acts < 0)
 578                 return st_acts;
 579
 580         err = validate_and_copy_actions(actions, key, depth + 1, sfa);
 581         if (err)
 582                 return err;
 583
 584         add_nested_action_end(*sfa, st_acts);
 585         add_nested_action_end(*sfa, start);
 586
 587         return 0;
 588 }
 589
 590 static int validate_tp_port(const struct sw_flow_key *flow_key)
 591 {
 592         if (flow_key->eth.type == htons(ETH_P_IP)) {
 593                 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
 594                         return 0;
 595         } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
 596                 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
 597                         return 0;
 598         }
 599
 600         return -EINVAL;
 601 }
 602
 603 static int validate_set(const struct nlattr *a,
 604                         const struct sw_flow_key *flow_key)
 605 {
 606         const struct nlattr *ovs_key = nla_data(a);
 607         int key_type = nla_type(ovs_key);
 608
 609         /* There can be only one key in a action */
 610         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 611                 return -EINVAL;
 612
 613         if (key_type > OVS_KEY_ATTR_MAX ||
 614             nla_len(ovs_key) != ovs_key_lens[key_type])
 615                 return -EINVAL;
 616
 617         switch (key_type) {
 618         const struct ovs_key_ipv4 *ipv4_key;
 619         const struct ovs_key_ipv6 *ipv6_key;
 620
 621         case OVS_KEY_ATTR_PRIORITY:
 622         case OVS_KEY_ATTR_SKB_MARK:
 623         case OVS_KEY_ATTR_ETHERNET:
 624                 break;
 625
 626         case OVS_KEY_ATTR_IPV4:
 627                 if (flow_key->eth.type != htons(ETH_P_IP))
 628                         return -EINVAL;
 629
 630                 if (!flow_key->ip.proto)
 631                         return -EINVAL;
 632
 633                 ipv4_key = nla_data(ovs_key);
 634                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 635                         return -EINVAL;
 636
 637                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 638                         return -EINVAL;
 639
 640                 break;
 641
 642         case OVS_KEY_ATTR_IPV6:
 643                 if (flow_key->eth.type != htons(ETH_P_IPV6))
 644                         return -EINVAL;
 645
 646                 if (!flow_key->ip.proto)
 647                         return -EINVAL;
 648
 649                 ipv6_key = nla_data(ovs_key);
 650                 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
 651                         return -EINVAL;
 652
 653                 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
 654                         return -EINVAL;
 655
 656                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
 657                         return -EINVAL;
 658
 659                 break;
 660
 661         case OVS_KEY_ATTR_TCP:
 662                 if (flow_key->ip.proto != IPPROTO_TCP)
 663                         return -EINVAL;
 664
 665                 return validate_tp_port(flow_key);
 666
 667         case OVS_KEY_ATTR_UDP:
 668                 if (flow_key->ip.proto != IPPROTO_UDP)
 669                         return -EINVAL;
 670
 671                 return validate_tp_port(flow_key);
 672
 673         default:
 674                 return -EINVAL;
 675         }
 676
 677         return 0;
 678 }
 679
 680 static int validate_userspace(const struct nlattr *attr)
 681 {
 682         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 683                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 684                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
 685         };
 686         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 687         int error;
 688
 689         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 690                                  attr, userspace_policy);
 691         if (error)
 692                 return error;
 693
 694         if (!a[OVS_USERSPACE_ATTR_PID] ||
 695             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 696                 return -EINVAL;
 697
 698         return 0;
 699 }
 700
 701 static int copy_action(const struct nlattr *from,
 702                        struct sw_flow_actions **sfa)
 703 {
 704         int totlen = NLA_ALIGN(from->nla_len);
 705         struct nlattr *to;
 706
 707         to = reserve_sfa_size(sfa, from->nla_len);
 708         if (IS_ERR(to))
 709                 return PTR_ERR(to);
 710
 711         memcpy(to, from, totlen);
 712         return 0;
 713 }
 714
 715 static int validate_and_copy_actions(const struct nlattr *attr,
 716                                      const struct sw_flow_key *key,
 717                                      int depth,
 718                                      struct sw_flow_actions **sfa)
 719 {
 720         const struct nlattr *a;
 721         int rem, err;
 722
 723         if (depth >= SAMPLE_ACTION_DEPTH)
 724                 return -EOVERFLOW;
 725
 726         nla_for_each_nested(a, attr, rem) {
 727                 /* Expected argument lengths, (u32)-1 for variable length. */
 728                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 729                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 730                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 731                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 732                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 733                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 734                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 735                 };
 736                 const struct ovs_action_push_vlan *vlan;
 737                 int type = nla_type(a);
 738                 bool skip_copy;
 739
 740                 if (type > OVS_ACTION_ATTR_MAX ||
 741                     (action_lens[type] != nla_len(a) &&
 742                      action_lens[type] != (u32)-1))
 743                         return -EINVAL;
 744
 745                 skip_copy = false;
 746                 switch (type) {
 747                 case OVS_ACTION_ATTR_UNSPEC:
 748                         return -EINVAL;
 749
 750                 case OVS_ACTION_ATTR_USERSPACE:
 751                         err = validate_userspace(a);
 752                         if (err)
 753                                 return err;
 754                         break;
 755
 756                 case OVS_ACTION_ATTR_OUTPUT:
 757                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 758                                 return -EINVAL;
 759                         break;
 760
 761
 762                 case OVS_ACTION_ATTR_POP_VLAN:
 763                         break;
 764
 765                 case OVS_ACTION_ATTR_PUSH_VLAN:
 766                         vlan = nla_data(a);
 767                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 768                                 return -EINVAL;
 769                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 770                                 return -EINVAL;
 771                         break;
 772
 773                 case OVS_ACTION_ATTR_SET:
 774                         err = validate_set(a, key);
 775                         if (err)
 776                                 return err;
 777                         break;
 778
 779                 case OVS_ACTION_ATTR_SAMPLE:
 780                         err = validate_and_copy_sample(a, key, depth, sfa);
 781                         if (err)
 782                                 return err;
 783                         skip_copy = true;
 784                         break;
 785
 786                 default:
 787                         return -EINVAL;
 788                 }
 789                 if (!skip_copy) {
 790                         err = copy_action(a, sfa);
 791                         if (err)
 792                                 return err;
 793                 }
 794         }
 795
 796         if (rem > 0)
 797                 return -EINVAL;
 798
 799         return 0;
 800 }
 801
 802 static void clear_stats(struct sw_flow *flow)
 803 {
 804         flow->used = 0;
 805         flow->tcp_flags = 0;
 806         flow->packet_count = 0;
 807         flow->byte_count = 0;
 808 }
 809
 810 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 811 {
 812         struct ovs_header *ovs_header = info->userhdr;
 813         struct nlattr **a = info->attrs;
 814         struct sw_flow_actions *acts;
 815         struct sk_buff *packet;
 816         struct sw_flow *flow;
 817         struct datapath *dp;
 818         struct ethhdr *eth;
 819         int len;
 820         int err;
 821         int key_len;
 822
 823         err = -EINVAL;
 824         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 825             !a[OVS_PACKET_ATTR_ACTIONS])
 826                 goto err;
 827
 828         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 829         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 830         err = -ENOMEM;
 831         if (!packet)
 832                 goto err;
 833         skb_reserve(packet, NET_IP_ALIGN);
 834
 835         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 836
 837         skb_reset_mac_header(packet);
 838         eth = eth_hdr(packet);
 839
 840         /* Normally, setting the skb 'protocol' field would be handled by a
 841          * call to eth_type_trans(), but it assumes there's a sending
 842          * device, which we may not have. */
 843         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 844                 packet->protocol = eth->h_proto;
 845         else
 846                 packet->protocol = htons(ETH_P_802_2);
 847
 848         /* Build an sw_flow for sending this packet. */
 849         flow = ovs_flow_alloc();
 850         err = PTR_ERR(flow);
 851         if (IS_ERR(flow))
 852                 goto err_kfree_skb;
 853
 854         err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
 855         if (err)
 856                 goto err_flow_free;
 857
 858         err = ovs_flow_metadata_from_nlattrs(flow, a[OVS_PACKET_ATTR_KEY]);
 859         if (err)
 860                 goto err_flow_free;
 861         flow->hash = ovs_flow_hash(&flow->key, key_len);
 862         acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 863         err = PTR_ERR(acts);
 864         if (IS_ERR(acts))
 865                 goto err_flow_free;
 866
 867         err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
 868         rcu_assign_pointer(flow->sf_acts, acts);
 869         if (err)
 870                 goto err_flow_free;
 871
 872         OVS_CB(packet)->flow = flow;
 873         packet->priority = flow->key.phy.priority;
 874         packet->mark = flow->key.phy.skb_mark;
 875
 876         rcu_read_lock();
 877         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 878         err = -ENODEV;
 879         if (!dp)
 880                 goto err_unlock;
 881
 882         local_bh_disable();
 883         err = ovs_execute_actions(dp, packet);
 884         local_bh_enable();
 885         rcu_read_unlock();
 886
 887         ovs_flow_free(flow);
 888         return err;
 889
 890 err_unlock:
 891         rcu_read_unlock();
 892 err_flow_free:
 893         ovs_flow_free(flow);
 894 err_kfree_skb:
 895         kfree_skb(packet);
 896 err:
 897         return err;
 898 }
 899
 900 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 901         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 902         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 903         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 904 };
 905
 906 static struct genl_ops dp_packet_genl_ops[] = {
 907         { .cmd = OVS_PACKET_CMD_EXECUTE,
 908           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 909           .policy = packet_policy,
 910           .doit = ovs_packet_cmd_execute
 911         }
 912 };
 913
 914 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 915 {
 916         int i;
 917         struct flow_table *table = ovsl_dereference(dp->table);
 918
 919         stats->n_flows = ovs_flow_tbl_count(table);
 920
 921         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 922         for_each_possible_cpu(i) {
 923                 const struct dp_stats_percpu *percpu_stats;
 924                 struct dp_stats_percpu local_stats;
 925                 unsigned int start;
 926
 927                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 928
 929                 do {
 930                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 931                         local_stats = *percpu_stats;
 932                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 933
 934                 stats->n_hit += local_stats.n_hit;
 935                 stats->n_missed += local_stats.n_missed;
 936                 stats->n_lost += local_stats.n_lost;
 937         }
 938 }
 939
 940 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 941         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 942         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 943         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 944 };
 945
 946 static struct genl_family dp_flow_genl_family = {
 947         .id = GENL_ID_GENERATE,
 948         .hdrsize = sizeof(struct ovs_header),
 949         .name = OVS_FLOW_FAMILY,
 950         .version = OVS_FLOW_VERSION,
 951         .maxattr = OVS_FLOW_ATTR_MAX,
 952         .netnsok = true,
 953         .parallel_ops = true,
 954 };
 955
 956 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 957         .name = OVS_FLOW_MCGROUP
 958 };
 959
 960 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
 961 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
 962 {
 963         const struct nlattr *a;
 964         struct nlattr *start;
 965         int err = 0, rem;
 966
 967         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
 968         if (!start)
 969                 return -EMSGSIZE;
 970
 971         nla_for_each_nested(a, attr, rem) {
 972                 int type = nla_type(a);
 973                 struct nlattr *st_sample;
 974
 975                 switch (type) {
 976                 case OVS_SAMPLE_ATTR_PROBABILITY:
 977                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
 978                                 return -EMSGSIZE;
 979                         break;
 980                 case OVS_SAMPLE_ATTR_ACTIONS:
 981                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
 982                         if (!st_sample)
 983                                 return -EMSGSIZE;
 984                         err = actions_to_attr(nla_data(a), nla_len(a), skb);
 985                         if (err)
 986                                 return err;
 987                         nla_nest_end(skb, st_sample);
 988                         break;
 989                 }
 990         }
 991
 992         nla_nest_end(skb, start);
 993         return err;
 994 }
 995
 996 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
 997 {
 998         const struct nlattr *a;
 999         int rem, err;
1000
1001         nla_for_each_attr(a, attr, len, rem) {
1002                 int type = nla_type(a);
1003
1004                 switch (type) {
1005                 case OVS_ACTION_ATTR_SAMPLE:
1006                         err = sample_action_to_attr(a, skb);
1007                         if (err)
1008                                 return err;
1009                         break;
1010                 default:
1011                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
1012                                 return -EMSGSIZE;
1013                         break;
1014                 }
1015         }
1016
1017         return 0;
1018 }
1019
1020 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
1021 {
1022         return NLMSG_ALIGN(sizeof(struct ovs_header))
1023                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
1024                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
1025                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
1026                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
1027                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
1028 }
1029
1030 /* Called with ovs_mutex. */
1031 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1032                                   struct sk_buff *skb, u32 portid,
1033                                   u32 seq, u32 flags, u8 cmd)
1034 {
1035         const int skb_orig_len = skb->len;
1036         const struct sw_flow_actions *sf_acts;
1037         struct nlattr *start;
1038         struct ovs_flow_stats stats;
1039         struct ovs_header *ovs_header;
1040         struct nlattr *nla;
1041         unsigned long used;
1042         u8 tcp_flags;
1043         int err;
1044
1045         sf_acts = ovsl_dereference(flow->sf_acts);
1046
1047         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1048         if (!ovs_header)
1049                 return -EMSGSIZE;
1050
1051         ovs_header->dp_ifindex = get_dpifindex(dp);
1052
1053         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1054         if (!nla)
1055                 goto nla_put_failure;
1056         err = ovs_flow_to_nlattrs(&flow->key, skb);
1057         if (err)
1058                 goto error;
1059         nla_nest_end(skb, nla);
1060
1061         spin_lock_bh(&flow->lock);
1062         used = flow->used;
1063         stats.n_packets = flow->packet_count;
1064         stats.n_bytes = flow->byte_count;
1065         tcp_flags = flow->tcp_flags;
1066         spin_unlock_bh(&flow->lock);
1067
1068         if (used &&
1069             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1070                 goto nla_put_failure;
1071
1072         if (stats.n_packets &&
1073             nla_put(skb, OVS_FLOW_ATTR_STATS,
1074                     sizeof(struct ovs_flow_stats), &stats))
1075                 goto nla_put_failure;
1076
1077         if (tcp_flags &&
1078             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1079                 goto nla_put_failure;
1080
1081         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1082          * this is the first flow to be dumped into 'skb'.  This is unusual for
1083          * Netlink but individual action lists can be longer than
1084          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1085          * The userspace caller can always fetch the actions separately if it
1086          * really wants them.  (Most userspace callers in fact don't care.)
1087          *
1088          * This can only fail for dump operations because the skb is always
1089          * properly sized for single flows.
1090          */
1091         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1092         if (start) {
1093                 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1094                 if (!err)
1095                         nla_nest_end(skb, start);
1096                 else {
1097                         if (skb_orig_len)
1098                                 goto error;
1099
1100                         nla_nest_cancel(skb, start);
1101                 }
1102         } else if (skb_orig_len)
1103                 goto nla_put_failure;
1104
1105         return genlmsg_end(skb, ovs_header);
1106
1107 nla_put_failure:
1108         err = -EMSGSIZE;
1109 error:
1110         genlmsg_cancel(skb, ovs_header);
1111         return err;
1112 }
1113
1114 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1115 {
1116         const struct sw_flow_actions *sf_acts;
1117
1118         sf_acts = ovsl_dereference(flow->sf_acts);
1119
1120         return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
1121 }
1122
1123 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1124                                                struct datapath *dp,
1125                                                u32 portid, u32 seq, u8 cmd)
1126 {
1127         struct sk_buff *skb;
1128         int retval;
1129
1130         skb = ovs_flow_cmd_alloc_info(flow);
1131         if (!skb)
1132                 return ERR_PTR(-ENOMEM);
1133
1134         retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1135         BUG_ON(retval < 0);
1136         return skb;
1137 }
1138
1139 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1140 {
1141         struct nlattr **a = info->attrs;
1142         struct ovs_header *ovs_header = info->userhdr;
1143         struct sw_flow_key key;
1144         struct sw_flow *flow;
1145         struct sk_buff *reply;
1146         struct datapath *dp;
1147         struct flow_table *table;
1148         struct sw_flow_actions *acts = NULL;
1149         int error;
1150         int key_len;
1151
1152         /* Extract key. */
1153         error = -EINVAL;
1154         if (!a[OVS_FLOW_ATTR_KEY])
1155                 goto error;
1156         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1157         if (error)
1158                 goto error;
1159
1160         /* Validate actions. */
1161         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1162                 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1163                 error = PTR_ERR(acts);
1164                 if (IS_ERR(acts))
1165                         goto error;
1166
1167                 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0, &acts);
1168                 if (error)
1169                         goto err_kfree;
1170         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1171                 error = -EINVAL;
1172                 goto error;
1173         }
1174
1175         ovs_lock();
1176         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1177         error = -ENODEV;
1178         if (!dp)
1179                 goto err_unlock_ovs;
1180
1181         table = ovsl_dereference(dp->table);
1182         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1183         if (!flow) {
1184                 /* Bail out if we're not allowed to create a new flow. */
1185                 error = -ENOENT;
1186                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1187                         goto err_unlock_ovs;
1188
1189                 /* Expand table, if necessary, to make room. */
1190                 if (ovs_flow_tbl_need_to_expand(table)) {
1191                         struct flow_table *new_table;
1192
1193                         new_table = ovs_flow_tbl_expand(table);
1194                         if (!IS_ERR(new_table)) {
1195                                 rcu_assign_pointer(dp->table, new_table);
1196                                 ovs_flow_tbl_deferred_destroy(table);
1197                                 table = ovsl_dereference(dp->table);
1198                         }
1199                 }
1200
1201                 /* Allocate flow. */
1202                 flow = ovs_flow_alloc();
1203                 if (IS_ERR(flow)) {
1204                         error = PTR_ERR(flow);
1205                         goto err_unlock_ovs;
1206                 }
1207                 flow->key = key;
1208                 clear_stats(flow);
1209
1210                 rcu_assign_pointer(flow->sf_acts, acts);
1211
1212                 /* Put flow in bucket. */
1213                 flow->hash = ovs_flow_hash(&key, key_len);
1214                 ovs_flow_tbl_insert(table, flow);
1215
1216                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1217                                                 info->snd_seq,
1218                                                 OVS_FLOW_CMD_NEW);
1219         } else {
1220                 /* We found a matching flow. */
1221                 struct sw_flow_actions *old_acts;
1222
1223                 /* Bail out if we're not allowed to modify an existing flow.
1224                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1225                  * because Generic Netlink treats the latter as a dump
1226                  * request.  We also accept NLM_F_EXCL in case that bug ever
1227                  * gets fixed.
1228                  */
1229                 error = -EEXIST;
1230                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1231                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1232                         goto err_unlock_ovs;
1233
1234                 /* Update actions. */
1235                 old_acts = ovsl_dereference(flow->sf_acts);
1236                 rcu_assign_pointer(flow->sf_acts, acts);
1237                 ovs_flow_deferred_free_acts(old_acts);
1238
1239                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1240                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1241
1242                 /* Clear stats. */
1243                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1244                         spin_lock_bh(&flow->lock);
1245                         clear_stats(flow);
1246                         spin_unlock_bh(&flow->lock);
1247                 }
1248         }
1249         ovs_unlock();
1250
1251         if (!IS_ERR(reply))
1252                 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1253         else
1254                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1255                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1256         return 0;
1257
1258 err_unlock_ovs:
1259         ovs_unlock();
1260 err_kfree:
1261         kfree(acts);
1262 error:
1263         return error;
1264 }
1265
1266 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1267 {
1268         struct nlattr **a = info->attrs;
1269         struct ovs_header *ovs_header = info->userhdr;
1270         struct sw_flow_key key;
1271         struct sk_buff *reply;
1272         struct sw_flow *flow;
1273         struct datapath *dp;
1274         struct flow_table *table;
1275         int err;
1276         int key_len;
1277
1278         if (!a[OVS_FLOW_ATTR_KEY])
1279                 return -EINVAL;
1280         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1281         if (err)
1282                 return err;
1283
1284         ovs_lock();
1285         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1286         if (!dp) {
1287                 err = -ENODEV;
1288                 goto unlock;
1289         }
1290
1291         table = ovsl_dereference(dp->table);
1292         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1293         if (!flow) {
1294                 err = -ENOENT;
1295                 goto unlock;
1296         }
1297
1298         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1299                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1300         if (IS_ERR(reply)) {
1301                 err = PTR_ERR(reply);
1302                 goto unlock;
1303         }
1304
1305         ovs_unlock();
1306         return genlmsg_reply(reply, info);
1307 unlock:
1308         ovs_unlock();
1309         return err;
1310 }
1311
1312 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1313 {
1314         struct nlattr **a = info->attrs;
1315         struct ovs_header *ovs_header = info->userhdr;
1316         struct sw_flow_key key;
1317         struct sk_buff *reply;
1318         struct sw_flow *flow;
1319         struct datapath *dp;
1320         struct flow_table *table;
1321         int err;
1322         int key_len;
1323
1324         ovs_lock();
1325         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1326         if (!dp) {
1327                 err = -ENODEV;
1328                 goto unlock;
1329         }
1330
1331         if (!a[OVS_FLOW_ATTR_KEY]) {
1332                 err = flush_flows(dp);
1333                 goto unlock;
1334         }
1335         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1336         if (err)
1337                 goto unlock;
1338
1339         table = ovsl_dereference(dp->table);
1340         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1341         if (!flow) {
1342                 err = -ENOENT;
1343                 goto unlock;
1344         }
1345
1346         reply = ovs_flow_cmd_alloc_info(flow);
1347         if (!reply) {
1348                 err = -ENOMEM;
1349                 goto unlock;
1350         }
1351
1352         ovs_flow_tbl_remove(table, flow);
1353
1354         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1355                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1356         BUG_ON(err < 0);
1357
1358         ovs_flow_deferred_free(flow);
1359         ovs_unlock();
1360
1361         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1362         return 0;
1363 unlock:
1364         ovs_unlock();
1365         return err;
1366 }
1367
1368 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1369 {
1370         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1371         struct datapath *dp;
1372         struct flow_table *table;
1373
1374         ovs_lock();
1375         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1376         if (!dp) {
1377                 ovs_unlock();
1378                 return -ENODEV;
1379         }
1380
1381         table = ovsl_dereference(dp->table);
1382
1383         for (;;) {
1384                 struct sw_flow *flow;
1385                 u32 bucket, obj;
1386
1387                 bucket = cb->args[0];
1388                 obj = cb->args[1];
1389                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1390                 if (!flow)
1391                         break;
1392
1393                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1394                                            NETLINK_CB(cb->skb).portid,
1395                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1396                                            OVS_FLOW_CMD_NEW) < 0)
1397                         break;
1398
1399                 cb->args[0] = bucket;
1400                 cb->args[1] = obj;
1401         }
1402         ovs_unlock();
1403         return skb->len;
1404 }
1405
1406 static struct genl_ops dp_flow_genl_ops[] = {
1407         { .cmd = OVS_FLOW_CMD_NEW,
1408           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1409           .policy = flow_policy,
1410           .doit = ovs_flow_cmd_new_or_set
1411         },
1412         { .cmd = OVS_FLOW_CMD_DEL,
1413           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1414           .policy = flow_policy,
1415           .doit = ovs_flow_cmd_del
1416         },
1417         { .cmd = OVS_FLOW_CMD_GET,
1418           .flags = 0,               /* OK for unprivileged users. */
1419           .policy = flow_policy,
1420           .doit = ovs_flow_cmd_get,
1421           .dumpit = ovs_flow_cmd_dump
1422         },
1423         { .cmd = OVS_FLOW_CMD_SET,
1424           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1425           .policy = flow_policy,
1426           .doit = ovs_flow_cmd_new_or_set,
1427         },
1428 };
1429
1430 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1431         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1432         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1433 };
1434
1435 static struct genl_family dp_datapath_genl_family = {
1436         .id = GENL_ID_GENERATE,
1437         .hdrsize = sizeof(struct ovs_header),
1438         .name = OVS_DATAPATH_FAMILY,
1439         .version = OVS_DATAPATH_VERSION,
1440         .maxattr = OVS_DP_ATTR_MAX,
1441         .netnsok = true,
1442         .parallel_ops = true,
1443 };
1444
1445 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1446         .name = OVS_DATAPATH_MCGROUP
1447 };
1448
1449 static size_t ovs_dp_cmd_msg_size(void)
1450 {
1451         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1452
1453         msgsize += nla_total_size(IFNAMSIZ);
1454         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1455
1456         return msgsize;
1457 }
1458
1459 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1460                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1461 {
1462         struct ovs_header *ovs_header;
1463         struct ovs_dp_stats dp_stats;
1464         int err;
1465
1466         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1467                                    flags, cmd);
1468         if (!ovs_header)
1469                 goto error;
1470
1471         ovs_header->dp_ifindex = get_dpifindex(dp);
1472
1473         rcu_read_lock();
1474         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1475         rcu_read_unlock();
1476         if (err)
1477                 goto nla_put_failure;
1478
1479         get_dp_stats(dp, &dp_stats);
1480         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1481                 goto nla_put_failure;
1482
1483         return genlmsg_end(skb, ovs_header);
1484
1485 nla_put_failure:
1486         genlmsg_cancel(skb, ovs_header);
1487 error:
1488         return -EMSGSIZE;
1489 }
1490
1491 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1492                                              u32 seq, u8 cmd)
1493 {
1494         struct sk_buff *skb;
1495         int retval;
1496
1497         skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1498         if (!skb)
1499                 return ERR_PTR(-ENOMEM);
1500
1501         retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1502         if (retval < 0) {
1503                 kfree_skb(skb);
1504                 return ERR_PTR(retval);
1505         }
1506         return skb;
1507 }
1508
1509 /* Called with ovs_mutex. */
1510 static struct datapath *lookup_datapath(struct net *net,
1511                                         struct ovs_header *ovs_header,
1512                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1513 {
1514         struct datapath *dp;
1515
1516         if (!a[OVS_DP_ATTR_NAME])
1517                 dp = get_dp(net, ovs_header->dp_ifindex);
1518         else {
1519                 struct vport *vport;
1520
1521                 rcu_read_lock();
1522                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1523                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1524                 rcu_read_unlock();
1525         }
1526         return dp ? dp : ERR_PTR(-ENODEV);
1527 }
1528
1529 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1530 {
1531         struct nlattr **a = info->attrs;
1532         struct vport_parms parms;
1533         struct sk_buff *reply;
1534         struct datapath *dp;
1535         struct vport *vport;
1536         struct ovs_net *ovs_net;
1537         int err, i;
1538
1539         err = -EINVAL;
1540         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1541                 goto err;
1542
1543         ovs_lock();
1544
1545         err = -ENOMEM;
1546         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1547         if (dp == NULL)
1548                 goto err_unlock_ovs;
1549
1550         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1551
1552         /* Allocate table. */
1553         err = -ENOMEM;
1554         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1555         if (!dp->table)
1556                 goto err_free_dp;
1557
1558         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1559         if (!dp->stats_percpu) {
1560                 err = -ENOMEM;
1561                 goto err_destroy_table;
1562         }
1563
1564         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1565                         GFP_KERNEL);
1566         if (!dp->ports) {
1567                 err = -ENOMEM;
1568                 goto err_destroy_percpu;
1569         }
1570
1571         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1572                 INIT_HLIST_HEAD(&dp->ports[i]);
1573
1574         /* Set up our datapath device. */
1575         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1576         parms.type = OVS_VPORT_TYPE_INTERNAL;
1577         parms.options = NULL;
1578         parms.dp = dp;
1579         parms.port_no = OVSP_LOCAL;
1580         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1581
1582         vport = new_vport(&parms);
1583         if (IS_ERR(vport)) {
1584                 err = PTR_ERR(vport);
1585                 if (err == -EBUSY)
1586                         err = -EEXIST;
1587
1588                 goto err_destroy_ports_array;
1589         }
1590
1591         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1592                                       info->snd_seq, OVS_DP_CMD_NEW);
1593         err = PTR_ERR(reply);
1594         if (IS_ERR(reply))
1595                 goto err_destroy_local_port;
1596
1597         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1598         list_add_tail(&dp->list_node, &ovs_net->dps);
1599
1600         ovs_unlock();
1601
1602         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1603         return 0;
1604
1605 err_destroy_local_port:
1606         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1607 err_destroy_ports_array:
1608         kfree(dp->ports);
1609 err_destroy_percpu:
1610         free_percpu(dp->stats_percpu);
1611 err_destroy_table:
1612         ovs_flow_tbl_destroy(ovsl_dereference(dp->table));
1613 err_free_dp:
1614         release_net(ovs_dp_get_net(dp));
1615         kfree(dp);
1616 err_unlock_ovs:
1617         ovs_unlock();
1618 err:
1619         return err;
1620 }
1621
1622 /* Called with ovs_mutex. */
1623 static void __dp_destroy(struct datapath *dp)
1624 {
1625         int i;
1626
1627         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1628                 struct vport *vport;
1629                 struct hlist_node *n;
1630
1631                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1632                         if (vport->port_no != OVSP_LOCAL)
1633                                 ovs_dp_detach_port(vport);
1634         }
1635
1636         list_del(&dp->list_node);
1637
1638         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1639          * all port in datapath are destroyed first before freeing datapath.
1640          */
1641         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1642
1643         call_rcu(&dp->rcu, destroy_dp_rcu);
1644 }
1645
1646 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1647 {
1648         struct sk_buff *reply;
1649         struct datapath *dp;
1650         int err;
1651
1652         ovs_lock();
1653         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1654         err = PTR_ERR(dp);
1655         if (IS_ERR(dp))
1656                 goto unlock;
1657
1658         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1659                                       info->snd_seq, OVS_DP_CMD_DEL);
1660         err = PTR_ERR(reply);
1661         if (IS_ERR(reply))
1662                 goto unlock;
1663
1664         __dp_destroy(dp);
1665         ovs_unlock();
1666
1667         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1668
1669         return 0;
1670 unlock:
1671         ovs_unlock();
1672         return err;
1673 }
1674
1675 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1676 {
1677         struct sk_buff *reply;
1678         struct datapath *dp;
1679         int err;
1680
1681         ovs_lock();
1682         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1683         err = PTR_ERR(dp);
1684         if (IS_ERR(dp))
1685                 goto unlock;
1686
1687         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1688                                       info->snd_seq, OVS_DP_CMD_NEW);
1689         if (IS_ERR(reply)) {
1690                 err = PTR_ERR(reply);
1691                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1692                                 ovs_dp_datapath_multicast_group.id, err);
1693                 err = 0;
1694                 goto unlock;
1695         }
1696
1697         ovs_unlock();
1698         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1699
1700         return 0;
1701 unlock:
1702         ovs_unlock();
1703         return err;
1704 }
1705
1706 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1707 {
1708         struct sk_buff *reply;
1709         struct datapath *dp;
1710         int err;
1711
1712         ovs_lock();
1713         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1714         if (IS_ERR(dp)) {
1715                 err = PTR_ERR(dp);
1716                 goto unlock;
1717         }
1718
1719         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1720                                       info->snd_seq, OVS_DP_CMD_NEW);
1721         if (IS_ERR(reply)) {
1722                 err = PTR_ERR(reply);
1723                 goto unlock;
1724         }
1725
1726         ovs_unlock();
1727         return genlmsg_reply(reply, info);
1728
1729 unlock:
1730         ovs_unlock();
1731         return err;
1732 }
1733
1734 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1735 {
1736         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1737         struct datapath *dp;
1738         int skip = cb->args[0];
1739         int i = 0;
1740
1741         ovs_lock();
1742         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1743                 if (i >= skip &&
1744                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1745                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1746                                          OVS_DP_CMD_NEW) < 0)
1747                         break;
1748                 i++;
1749         }
1750         ovs_unlock();
1751
1752         cb->args[0] = i;
1753
1754         return skb->len;
1755 }
1756
1757 static struct genl_ops dp_datapath_genl_ops[] = {
1758         { .cmd = OVS_DP_CMD_NEW,
1759           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1760           .policy = datapath_policy,
1761           .doit = ovs_dp_cmd_new
1762         },
1763         { .cmd = OVS_DP_CMD_DEL,
1764           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1765           .policy = datapath_policy,
1766           .doit = ovs_dp_cmd_del
1767         },
1768         { .cmd = OVS_DP_CMD_GET,
1769           .flags = 0,               /* OK for unprivileged users. */
1770           .policy = datapath_policy,
1771           .doit = ovs_dp_cmd_get,
1772           .dumpit = ovs_dp_cmd_dump
1773         },
1774         { .cmd = OVS_DP_CMD_SET,
1775           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1776           .policy = datapath_policy,
1777           .doit = ovs_dp_cmd_set,
1778         },
1779 };
1780
1781 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1782         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1783         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1784         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1785         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1786         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1787         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1788 };
1789
1790 static struct genl_family dp_vport_genl_family = {
1791         .id = GENL_ID_GENERATE,
1792         .hdrsize = sizeof(struct ovs_header),
1793         .name = OVS_VPORT_FAMILY,
1794         .version = OVS_VPORT_VERSION,
1795         .maxattr = OVS_VPORT_ATTR_MAX,
1796         .netnsok = true,
1797         .parallel_ops = true,
1798 };
1799
1800 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1801         .name = OVS_VPORT_MCGROUP
1802 };
1803
1804 /* Called with ovs_mutex or RCU read lock. */
1805 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1806                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1807 {
1808         struct ovs_header *ovs_header;
1809         struct ovs_vport_stats vport_stats;
1810         int err;
1811
1812         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1813                                  flags, cmd);
1814         if (!ovs_header)
1815                 return -EMSGSIZE;
1816
1817         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1818
1819         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1820             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1821             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1822             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1823                 goto nla_put_failure;
1824
1825         ovs_vport_get_stats(vport, &vport_stats);
1826         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1827                     &vport_stats))
1828                 goto nla_put_failure;
1829
1830         err = ovs_vport_get_options(vport, skb);
1831         if (err == -EMSGSIZE)
1832                 goto error;
1833
1834         return genlmsg_end(skb, ovs_header);
1835
1836 nla_put_failure:
1837         err = -EMSGSIZE;
1838 error:
1839         genlmsg_cancel(skb, ovs_header);
1840         return err;
1841 }
1842
1843 /* Called with ovs_mutex or RCU read lock. */
1844 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1845                                          u32 seq, u8 cmd)
1846 {
1847         struct sk_buff *skb;
1848         int retval;
1849
1850         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1851         if (!skb)
1852                 return ERR_PTR(-ENOMEM);
1853
1854         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1855         BUG_ON(retval < 0);
1856
1857         return skb;
1858 }
1859
1860 /* Called with ovs_mutex or RCU read lock. */
1861 static struct vport *lookup_vport(struct net *net,
1862                                   struct ovs_header *ovs_header,
1863                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1864 {
1865         struct datapath *dp;
1866         struct vport *vport;
1867
1868         if (a[OVS_VPORT_ATTR_NAME]) {
1869                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1870                 if (!vport)
1871                         return ERR_PTR(-ENODEV);
1872                 if (ovs_header->dp_ifindex &&
1873                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1874                         return ERR_PTR(-ENODEV);
1875                 return vport;
1876         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1877                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1878
1879                 if (port_no >= DP_MAX_PORTS)
1880                         return ERR_PTR(-EFBIG);
1881
1882                 dp = get_dp(net, ovs_header->dp_ifindex);
1883                 if (!dp)
1884                         return ERR_PTR(-ENODEV);
1885
1886                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1887                 if (!vport)
1888                         return ERR_PTR(-ENODEV);
1889                 return vport;
1890         } else
1891                 return ERR_PTR(-EINVAL);
1892 }
1893
1894 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1895 {
1896         struct nlattr **a = info->attrs;
1897         struct ovs_header *ovs_header = info->userhdr;
1898         struct vport_parms parms;
1899         struct sk_buff *reply;
1900         struct vport *vport;
1901         struct datapath *dp;
1902         u32 port_no;
1903         int err;
1904
1905         err = -EINVAL;
1906         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1907             !a[OVS_VPORT_ATTR_UPCALL_PID])
1908                 goto exit;
1909
1910         ovs_lock();
1911         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1912         err = -ENODEV;
1913         if (!dp)
1914                 goto exit_unlock;
1915
1916         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1917                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1918
1919                 err = -EFBIG;
1920                 if (port_no >= DP_MAX_PORTS)
1921                         goto exit_unlock;
1922
1923                 vport = ovs_vport_ovsl(dp, port_no);
1924                 err = -EBUSY;
1925                 if (vport)
1926                         goto exit_unlock;
1927         } else {
1928                 for (port_no = 1; ; port_no++) {
1929                         if (port_no >= DP_MAX_PORTS) {
1930                                 err = -EFBIG;
1931                                 goto exit_unlock;
1932                         }
1933                         vport = ovs_vport_ovsl(dp, port_no);
1934                         if (!vport)
1935                                 break;
1936                 }
1937         }
1938
1939         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1940         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1941         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1942         parms.dp = dp;
1943         parms.port_no = port_no;
1944         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1945
1946         vport = new_vport(&parms);
1947         err = PTR_ERR(vport);
1948         if (IS_ERR(vport))
1949                 goto exit_unlock;
1950
1951         err = 0;
1952         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1953                                          OVS_VPORT_CMD_NEW);
1954         if (IS_ERR(reply)) {
1955                 err = PTR_ERR(reply);
1956                 ovs_dp_detach_port(vport);
1957                 goto exit_unlock;
1958         }
1959
1960         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1961
1962 exit_unlock:
1963         ovs_unlock();
1964 exit:
1965         return err;
1966 }
1967
1968 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1969 {
1970         struct nlattr **a = info->attrs;
1971         struct sk_buff *reply;
1972         struct vport *vport;
1973         int err;
1974
1975         ovs_lock();
1976         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1977         err = PTR_ERR(vport);
1978         if (IS_ERR(vport))
1979                 goto exit_unlock;
1980
1981         if (a[OVS_VPORT_ATTR_TYPE] &&
1982             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1983                 err = -EINVAL;
1984                 goto exit_unlock;
1985         }
1986
1987         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1988         if (!reply) {
1989                 err = -ENOMEM;
1990                 goto exit_unlock;
1991         }
1992
1993         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1994                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1995                 if (err)
1996                         goto exit_free;
1997         }
1998
1999         if (a[OVS_VPORT_ATTR_UPCALL_PID])
2000                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2001
2002         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2003                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2004         BUG_ON(err < 0);
2005
2006         ovs_unlock();
2007         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2008         return 0;
2009
2010         rtnl_unlock();
2011         return 0;
2012
2013 exit_free:
2014         kfree_skb(reply);
2015 exit_unlock:
2016         ovs_unlock();
2017         return err;
2018 }
2019
2020 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2021 {
2022         struct nlattr **a = info->attrs;
2023         struct sk_buff *reply;
2024         struct vport *vport;
2025         int err;
2026
2027         ovs_lock();
2028         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2029         err = PTR_ERR(vport);
2030         if (IS_ERR(vport))
2031                 goto exit_unlock;
2032
2033         if (vport->port_no == OVSP_LOCAL) {
2034                 err = -EINVAL;
2035                 goto exit_unlock;
2036         }
2037
2038         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2039                                          info->snd_seq, OVS_VPORT_CMD_DEL);
2040         err = PTR_ERR(reply);
2041         if (IS_ERR(reply))
2042                 goto exit_unlock;
2043
2044         err = 0;
2045         ovs_dp_detach_port(vport);
2046
2047         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2048
2049 exit_unlock:
2050         ovs_unlock();
2051         return err;
2052 }
2053
2054 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2055 {
2056         struct nlattr **a = info->attrs;
2057         struct ovs_header *ovs_header = info->userhdr;
2058         struct sk_buff *reply;
2059         struct vport *vport;
2060         int err;
2061
2062         rcu_read_lock();
2063         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2064         err = PTR_ERR(vport);
2065         if (IS_ERR(vport))
2066                 goto exit_unlock;
2067
2068         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2069                                          info->snd_seq, OVS_VPORT_CMD_NEW);
2070         err = PTR_ERR(reply);
2071         if (IS_ERR(reply))
2072                 goto exit_unlock;
2073
2074         rcu_read_unlock();
2075
2076         return genlmsg_reply(reply, info);
2077
2078 exit_unlock:
2079         rcu_read_unlock();
2080         return err;
2081 }
2082
2083 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2084 {
2085         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2086         struct datapath *dp;
2087         int bucket = cb->args[0], skip = cb->args[1];
2088         int i, j = 0;
2089
2090         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2091         if (!dp)
2092                 return -ENODEV;
2093
2094         rcu_read_lock();
2095         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2096                 struct vport *vport;
2097
2098                 j = 0;
2099                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2100                         if (j >= skip &&
2101                             ovs_vport_cmd_fill_info(vport, skb,
2102                                                     NETLINK_CB(cb->skb).portid,
2103                                                     cb->nlh->nlmsg_seq,
2104                                                     NLM_F_MULTI,
2105                                                     OVS_VPORT_CMD_NEW) < 0)
2106                                 goto out;
2107
2108                         j++;
2109                 }
2110                 skip = 0;
2111         }
2112 out:
2113         rcu_read_unlock();
2114
2115         cb->args[0] = i;
2116         cb->args[1] = j;
2117
2118         return skb->len;
2119 }
2120
2121 static struct genl_ops dp_vport_genl_ops[] = {
2122         { .cmd = OVS_VPORT_CMD_NEW,
2123           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2124           .policy = vport_policy,
2125           .doit = ovs_vport_cmd_new
2126         },
2127         { .cmd = OVS_VPORT_CMD_DEL,
2128           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2129           .policy = vport_policy,
2130           .doit = ovs_vport_cmd_del
2131         },
2132         { .cmd = OVS_VPORT_CMD_GET,
2133           .flags = 0,               /* OK for unprivileged users. */
2134           .policy = vport_policy,
2135           .doit = ovs_vport_cmd_get,
2136           .dumpit = ovs_vport_cmd_dump
2137         },
2138         { .cmd = OVS_VPORT_CMD_SET,
2139           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2140           .policy = vport_policy,
2141           .doit = ovs_vport_cmd_set,
2142         },
2143 };
2144
2145 struct genl_family_and_ops {
2146         struct genl_family *family;
2147         struct genl_ops *ops;
2148         int n_ops;
2149         struct genl_multicast_group *group;
2150 };
2151
2152 static const struct genl_family_and_ops dp_genl_families[] = {
2153         { &dp_datapath_genl_family,
2154           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2155           &ovs_dp_datapath_multicast_group },
2156         { &dp_vport_genl_family,
2157           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2158           &ovs_dp_vport_multicast_group },
2159         { &dp_flow_genl_family,
2160           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2161           &ovs_dp_flow_multicast_group },
2162         { &dp_packet_genl_family,
2163           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2164           NULL },
2165 };
2166
2167 static void dp_unregister_genl(int n_families)
2168 {
2169         int i;
2170
2171         for (i = 0; i < n_families; i++)
2172                 genl_unregister_family(dp_genl_families[i].family);
2173 }
2174
2175 static int dp_register_genl(void)
2176 {
2177         int n_registered;
2178         int err;
2179         int i;
2180
2181         n_registered = 0;
2182         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2183                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2184
2185                 err = genl_register_family_with_ops(f->family, f->ops,
2186                                                     f->n_ops);
2187                 if (err)
2188                         goto error;
2189                 n_registered++;
2190
2191                 if (f->group) {
2192                         err = genl_register_mc_group(f->family, f->group);
2193                         if (err)
2194                                 goto error;
2195                 }
2196         }
2197
2198         return 0;
2199
2200 error:
2201         dp_unregister_genl(n_registered);
2202         return err;
2203 }
2204
2205 static void rehash_flow_table(struct work_struct *work)
2206 {
2207         struct datapath *dp;
2208         struct net *net;
2209
2210         ovs_lock();
2211         rtnl_lock();
2212         for_each_net(net) {
2213                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2214
2215                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2216                         struct flow_table *old_table = ovsl_dereference(dp->table);
2217                         struct flow_table *new_table;
2218
2219                         new_table = ovs_flow_tbl_rehash(old_table);
2220                         if (!IS_ERR(new_table)) {
2221                                 rcu_assign_pointer(dp->table, new_table);
2222                                 ovs_flow_tbl_deferred_destroy(old_table);
2223                         }
2224                 }
2225         }
2226         rtnl_unlock();
2227         ovs_unlock();
2228         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2229 }
2230
2231 static int __net_init ovs_init_net(struct net *net)
2232 {
2233         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2234
2235         INIT_LIST_HEAD(&ovs_net->dps);
2236         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2237         return 0;
2238 }
2239
2240 static void __net_exit ovs_exit_net(struct net *net)
2241 {
2242         struct datapath *dp, *dp_next;
2243         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2244
2245         ovs_lock();
2246         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2247                 __dp_destroy(dp);
2248         ovs_unlock();
2249
2250         cancel_work_sync(&ovs_net->dp_notify_work);
2251 }
2252
2253 static struct pernet_operations ovs_net_ops = {
2254         .init = ovs_init_net,
2255         .exit = ovs_exit_net,
2256         .id   = &ovs_net_id,
2257         .size = sizeof(struct ovs_net),
2258 };
2259
2260 static int __init dp_init(void)
2261 {
2262         int err;
2263
2264         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2265
2266         pr_info("Open vSwitch switching datapath\n");
2267
2268         err = ovs_flow_init();
2269         if (err)
2270                 goto error;
2271
2272         err = ovs_vport_init();
2273         if (err)
2274                 goto error_flow_exit;
2275
2276         err = register_pernet_device(&ovs_net_ops);
2277         if (err)
2278                 goto error_vport_exit;
2279
2280         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2281         if (err)
2282                 goto error_netns_exit;
2283
2284         err = dp_register_genl();
2285         if (err < 0)
2286                 goto error_unreg_notifier;
2287
2288         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2289
2290         return 0;
2291
2292 error_unreg_notifier:
2293         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2294 error_netns_exit:
2295         unregister_pernet_device(&ovs_net_ops);
2296 error_vport_exit:
2297         ovs_vport_exit();
2298 error_flow_exit:
2299         ovs_flow_exit();
2300 error:
2301         return err;
2302 }
2303
2304 static void dp_cleanup(void)
2305 {
2306         cancel_delayed_work_sync(&rehash_flow_wq);
2307         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2308         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2309         unregister_pernet_device(&ovs_net_ops);
2310         rcu_barrier();
2311         ovs_vport_exit();
2312         ovs_flow_exit();
2313 }
2314
2315 module_init(dp_init);
2316 module_exit(dp_cleanup);
2317
2318 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2319 MODULE_LICENSE("GPL");