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