datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007-2011 Nicira Networks.
   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/version.h>
  40 #include <linux/ethtool.h>
  41 #include <linux/wait.h>
  42 #include <asm/system.h>
  43 #include <asm/div64.h>
  44 #include <linux/highmem.h>
  45 #include <linux/netfilter_bridge.h>
  46 #include <linux/netfilter_ipv4.h>
  47 #include <linux/inetdevice.h>
  48 #include <linux/list.h>
  49 #include <linux/openvswitch.h>
  50 #include <linux/rculist.h>
  51 #include <linux/dmi.h>
  52 #include <net/genetlink.h>
  53
  54 #include "checksum.h"
  55 #include "datapath.h"
  56 #include "flow.h"
  57 #include "vlan.h"
  58 #include "tunnel.h"
  59 #include "vport-internal_dev.h"
  60
  61 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
  62     LINUX_VERSION_CODE > KERNEL_VERSION(3,2,0)
  63 #error Kernels before 2.6.18 or after 3.2 are not supported by this version of Open vSwitch.
  64 #endif
  65
  66 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
  67 EXPORT_SYMBOL(dp_ioctl_hook);
  68
  69 /**
  70  * DOC: Locking:
  71  *
  72  * Writes to device state (add/remove datapath, port, set operations on vports,
  73  * etc.) are protected by RTNL.
  74  *
  75  * Writes to other state (flow table modifications, set miscellaneous datapath
  76  * parameters, etc.) are protected by genl_mutex.  The RTNL lock nests inside
  77  * genl_mutex.
  78  *
  79  * Reads are protected by RCU.
  80  *
  81  * There are a few special cases (mostly stats) that have their own
  82  * synchronization but they nest under all of above and don't interact with
  83  * each other.
  84  */
  85
  86 /* Global list of datapaths to enable dumping them all out.
  87  * Protected by genl_mutex.
  88  */
  89 static LIST_HEAD(dps);
  90
  91 static struct vport *new_vport(const struct vport_parms *);
  92 static int queue_gso_packets(int dp_ifindex, struct sk_buff *,
  93                              const struct dp_upcall_info *);
  94 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *,
  95                                   const struct dp_upcall_info *);
  96
  97 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  98 struct datapath *get_dp(int dp_ifindex)
  99 {
 100         struct datapath *dp = NULL;
 101         struct net_device *dev;
 102
 103         rcu_read_lock();
 104         dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
 105         if (dev) {
 106                 struct vport *vport = internal_dev_get_vport(dev);
 107                 if (vport)
 108                         dp = vport->dp;
 109         }
 110         rcu_read_unlock();
 111
 112         return dp;
 113 }
 114 EXPORT_SYMBOL_GPL(get_dp);
 115
 116 /* Must be called with genl_mutex. */
 117 static struct flow_table *get_table_protected(struct datapath *dp)
 118 {
 119         return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
 120 }
 121
 122 /* Must be called with rcu_read_lock or RTNL lock. */
 123 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
 124 {
 125         return rcu_dereference_rtnl(dp->ports[port_no]);
 126 }
 127
 128 /* Must be called with rcu_read_lock or RTNL lock. */
 129 const char *dp_name(const struct datapath *dp)
 130 {
 131         struct vport *vport = rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]);
 132         return vport->ops->get_name(vport);
 133 }
 134
 135 static int get_dpifindex(struct datapath *dp)
 136 {
 137         struct vport *local;
 138         int ifindex;
 139
 140         rcu_read_lock();
 141
 142         local = get_vport_protected(dp, OVSP_LOCAL);
 143         if (local)
 144                 ifindex = local->ops->get_ifindex(local);
 145         else
 146                 ifindex = 0;
 147
 148         rcu_read_unlock();
 149
 150         return ifindex;
 151 }
 152
 153 static size_t br_nlmsg_size(void)
 154 {
 155         return NLMSG_ALIGN(sizeof(struct ifinfomsg))
 156                + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
 157                + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
 158                + nla_total_size(4) /* IFLA_MASTER */
 159                + nla_total_size(4) /* IFLA_MTU */
 160                + nla_total_size(1); /* IFLA_OPERSTATE */
 161 }
 162
 163 /* Caller must hold RTNL lock. */
 164 static int dp_fill_ifinfo(struct sk_buff *skb,
 165                           const struct vport *port,
 166                           int event, unsigned int flags)
 167 {
 168         struct datapath *dp = port->dp;
 169         struct ifinfomsg *hdr;
 170         struct nlmsghdr *nlh;
 171
 172         if (!port->ops->get_ifindex)
 173                 return -ENODEV;
 174
 175         nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
 176         if (nlh == NULL)
 177                 return -EMSGSIZE;
 178
 179         hdr = nlmsg_data(nlh);
 180         hdr->ifi_family = AF_BRIDGE;
 181         hdr->__ifi_pad = 0;
 182         hdr->ifi_type = ARPHRD_ETHER;
 183         hdr->ifi_index = port->ops->get_ifindex(port);
 184         hdr->ifi_flags = port->ops->get_dev_flags(port);
 185         hdr->ifi_change = 0;
 186
 187         NLA_PUT_STRING(skb, IFLA_IFNAME, port->ops->get_name(port));
 188         NLA_PUT_U32(skb, IFLA_MASTER, get_dpifindex(dp));
 189         NLA_PUT_U32(skb, IFLA_MTU, port->ops->get_mtu(port));
 190 #ifdef IFLA_OPERSTATE
 191         NLA_PUT_U8(skb, IFLA_OPERSTATE,
 192                    port->ops->is_running(port)
 193                         ? port->ops->get_operstate(port)
 194                         : IF_OPER_DOWN);
 195 #endif
 196
 197         NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, port->ops->get_addr(port));
 198
 199         return nlmsg_end(skb, nlh);
 200
 201 nla_put_failure:
 202         nlmsg_cancel(skb, nlh);
 203         return -EMSGSIZE;
 204 }
 205
 206 /* Caller must hold RTNL lock. */
 207 static void dp_ifinfo_notify(int event, struct vport *port)
 208 {
 209         struct sk_buff *skb;
 210         int err;
 211
 212         skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
 213         if (!skb) {
 214                 err = -ENOBUFS;
 215                 goto err;
 216         }
 217
 218         err = dp_fill_ifinfo(skb, port, event, 0);
 219         if (err < 0) {
 220                 if (err == -ENODEV) {
 221                         goto out;
 222                 } else {
 223                         /* -EMSGSIZE implies BUG in br_nlmsg_size() */
 224                         WARN_ON(err == -EMSGSIZE);
 225                         goto err;
 226                 }
 227         }
 228
 229         rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 230
 231         return;
 232 err:
 233         rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
 234 out:
 235         kfree_skb(skb);
 236 }
 237
 238 static void release_dp(struct kobject *kobj)
 239 {
 240         struct datapath *dp = container_of(kobj, struct datapath, ifobj);
 241         kfree(dp);
 242 }
 243
 244 static struct kobj_type dp_ktype = {
 245         .release = release_dp
 246 };
 247
 248 static void destroy_dp_rcu(struct rcu_head *rcu)
 249 {
 250         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 251
 252         flow_tbl_destroy(dp->table);
 253         free_percpu(dp->stats_percpu);
 254         kobject_put(&dp->ifobj);
 255 }
 256
 257 /* Called with RTNL lock and genl_lock. */
 258 static struct vport *new_vport(const struct vport_parms *parms)
 259 {
 260         struct vport *vport;
 261
 262         vport = vport_add(parms);
 263         if (!IS_ERR(vport)) {
 264                 struct datapath *dp = parms->dp;
 265
 266                 rcu_assign_pointer(dp->ports[parms->port_no], vport);
 267                 list_add(&vport->node, &dp->port_list);
 268
 269                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 270         }
 271
 272         return vport;
 273 }
 274
 275 /* Called with RTNL lock. */
 276 void dp_detach_port(struct vport *p)
 277 {
 278         ASSERT_RTNL();
 279
 280         if (p->port_no != OVSP_LOCAL)
 281                 dp_sysfs_del_if(p);
 282         dp_ifinfo_notify(RTM_DELLINK, p);
 283
 284         /* First drop references to device. */
 285         list_del(&p->node);
 286         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 287
 288         /* Then destroy it. */
 289         vport_del(p);
 290 }
 291
 292 /* Must be called with rcu_read_lock. */
 293 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 294 {
 295         struct datapath *dp = p->dp;
 296         struct sw_flow *flow;
 297         struct dp_stats_percpu *stats;
 298         u64 *stats_counter;
 299         int error;
 300
 301         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 302
 303         if (!OVS_CB(skb)->flow) {
 304                 struct sw_flow_key key;
 305                 int key_len;
 306
 307                 /* Extract flow from 'skb' into 'key'. */
 308                 error = flow_extract(skb, p->port_no, &key, &key_len);
 309                 if (unlikely(error)) {
 310                         kfree_skb(skb);
 311                         return;
 312                 }
 313
 314                 /* Look up flow. */
 315                 flow = flow_tbl_lookup(rcu_dereference(dp->table),
 316                                        &key, key_len);
 317                 if (unlikely(!flow)) {
 318                         struct dp_upcall_info upcall;
 319
 320                         upcall.cmd = OVS_PACKET_CMD_MISS;
 321                         upcall.key = &key;
 322                         upcall.userdata = NULL;
 323                         upcall.pid = p->upcall_pid;
 324                         dp_upcall(dp, skb, &upcall);
 325                         consume_skb(skb);
 326                         stats_counter = &stats->n_missed;
 327                         goto out;
 328                 }
 329
 330                 OVS_CB(skb)->flow = flow;
 331         }
 332
 333         stats_counter = &stats->n_hit;
 334         flow_used(OVS_CB(skb)->flow, skb);
 335         execute_actions(dp, skb);
 336
 337 out:
 338         /* Update datapath statistics. */
 339
 340         write_seqcount_begin(&stats->seqlock);
 341         (*stats_counter)++;
 342         write_seqcount_end(&stats->seqlock);
 343 }
 344
 345 static struct genl_family dp_packet_genl_family = {
 346         .id = GENL_ID_GENERATE,
 347         .hdrsize = sizeof(struct ovs_header),
 348         .name = OVS_PACKET_FAMILY,
 349         .version = OVS_PACKET_VERSION,
 350         .maxattr = OVS_PACKET_ATTR_MAX
 351 };
 352
 353 int dp_upcall(struct datapath *dp, struct sk_buff *skb,
 354               const struct dp_upcall_info *upcall_info)
 355 {
 356         struct dp_stats_percpu *stats;
 357         int dp_ifindex;
 358         int err;
 359
 360         if (upcall_info->pid == 0) {
 361                 err = -ENOTCONN;
 362                 goto err;
 363         }
 364
 365         dp_ifindex = get_dpifindex(dp);
 366         if (!dp_ifindex) {
 367                 err = -ENODEV;
 368                 goto err;
 369         }
 370
 371         forward_ip_summed(skb, true);
 372
 373         if (!skb_is_gso(skb))
 374                 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
 375         else
 376                 err = queue_gso_packets(dp_ifindex, skb, upcall_info);
 377         if (err)
 378                 goto err;
 379
 380         return 0;
 381
 382 err:
 383         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 384
 385         write_seqcount_begin(&stats->seqlock);
 386         stats->n_lost++;
 387         write_seqcount_end(&stats->seqlock);
 388
 389         return err;
 390 }
 391
 392 static int queue_gso_packets(int dp_ifindex, struct sk_buff *skb,
 393                              const struct dp_upcall_info *upcall_info)
 394 {
 395         struct dp_upcall_info later_info;
 396         struct sw_flow_key later_key;
 397         struct sk_buff *segs, *nskb;
 398         int err;
 399
 400         segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 401         if (IS_ERR(skb))
 402                 return PTR_ERR(skb);
 403
 404         /* Queue all of the segments. */
 405         skb = segs;
 406         do {
 407                 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
 408                 if (err)
 409                         break;
 410
 411                 if (skb == segs && skb_shinfo(skb)->gso_type & SKB_GSO_UDP) {
 412                         /* The initial flow key extracted by flow_extract() in
 413                          * this case is for a first fragment, so we need to
 414                          * properly mark later fragments.
 415                          */
 416                         later_key = *upcall_info->key;
 417                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 418
 419                         later_info = *upcall_info;
 420                         later_info.key = &later_key;
 421                         upcall_info = &later_info;
 422                 }
 423         } while ((skb = skb->next));
 424
 425         /* Free all of the segments. */
 426         skb = segs;
 427         do {
 428                 nskb = skb->next;
 429                 if (err)
 430                         kfree_skb(skb);
 431                 else
 432                         consume_skb(skb);
 433         } while ((skb = nskb));
 434         return err;
 435 }
 436
 437 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *skb,
 438                                   const struct dp_upcall_info *upcall_info)
 439 {
 440         struct ovs_header *upcall;
 441         struct sk_buff *user_skb; /* to be queued to userspace */
 442         struct nlattr *nla;
 443         unsigned int len;
 444         int err;
 445
 446         err = vlan_deaccel_tag(skb);
 447         if (unlikely(err))
 448                 return err;
 449
 450         if (nla_attr_size(skb->len) > USHRT_MAX)
 451                 return -EFBIG;
 452
 453         len = sizeof(struct ovs_header);
 454         len += nla_total_size(skb->len);
 455         len += nla_total_size(FLOW_BUFSIZE);
 456         if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
 457                 len += nla_total_size(8);
 458
 459         user_skb = genlmsg_new(len, GFP_ATOMIC);
 460         if (!user_skb)
 461                 return -ENOMEM;
 462
 463         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 464                              0, upcall_info->cmd);
 465         upcall->dp_ifindex = dp_ifindex;
 466
 467         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 468         flow_to_nlattrs(upcall_info->key, user_skb);
 469         nla_nest_end(user_skb, nla);
 470
 471         if (upcall_info->userdata)
 472                 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
 473                             nla_get_u64(upcall_info->userdata));
 474
 475         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 476
 477         skb_copy_and_csum_dev(skb, nla_data(nla));
 478
 479         return genlmsg_unicast(&init_net, user_skb, upcall_info->pid);
 480 }
 481
 482 /* Called with genl_mutex. */
 483 static int flush_flows(int dp_ifindex)
 484 {
 485         struct flow_table *old_table;
 486         struct flow_table *new_table;
 487         struct datapath *dp;
 488
 489         dp = get_dp(dp_ifindex);
 490         if (!dp)
 491                 return -ENODEV;
 492
 493         old_table = get_table_protected(dp);
 494         new_table = flow_tbl_alloc(TBL_MIN_BUCKETS);
 495         if (!new_table)
 496                 return -ENOMEM;
 497
 498         rcu_assign_pointer(dp->table, new_table);
 499
 500         flow_tbl_deferred_destroy(old_table);
 501         return 0;
 502 }
 503
 504 static int validate_actions(const struct nlattr *attr,
 505                                 const struct sw_flow_key *key, int depth);
 506
 507 static int validate_sample(const struct nlattr *attr,
 508                                 const struct sw_flow_key *key, int depth)
 509 {
 510         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 511         const struct nlattr *probability, *actions;
 512         const struct nlattr *a;
 513         int rem;
 514
 515         memset(attrs, 0, sizeof(attrs));
 516         nla_for_each_nested(a, attr, rem) {
 517                 int type = nla_type(a);
 518                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 519                         return -EINVAL;
 520                 attrs[type] = a;
 521         }
 522         if (rem)
 523                 return -EINVAL;
 524
 525         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 526         if (!probability || nla_len(probability) != sizeof(u32))
 527                 return -EINVAL;
 528
 529         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 530         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 531                 return -EINVAL;
 532         return validate_actions(actions, key, depth + 1);
 533 }
 534
 535 static int validate_set(const struct nlattr *a,
 536                         const struct sw_flow_key *flow_key)
 537 {
 538         const struct nlattr *ovs_key = nla_data(a);
 539         int key_type = nla_type(ovs_key);
 540
 541         /* There can be only one key in a action */
 542         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 543                 return -EINVAL;
 544
 545         if (key_type > OVS_KEY_ATTR_MAX ||
 546             nla_len(ovs_key) != ovs_key_lens[key_type])
 547                 return -EINVAL;
 548
 549         switch (key_type) {
 550         const struct ovs_key_ipv4 *ipv4_key;
 551
 552         case OVS_KEY_ATTR_PRIORITY:
 553         case OVS_KEY_ATTR_TUN_ID:
 554         case OVS_KEY_ATTR_ETHERNET:
 555                 break;
 556
 557         case OVS_KEY_ATTR_IPV4:
 558                 if (flow_key->eth.type != htons(ETH_P_IP))
 559                         return -EINVAL;
 560
 561                 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
 562                         return -EINVAL;
 563
 564                 ipv4_key = nla_data(ovs_key);
 565                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 566                         return -EINVAL;
 567
 568                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 569                         return -EINVAL;
 570
 571                 break;
 572
 573         case OVS_KEY_ATTR_TCP:
 574                 if (flow_key->ip.proto != IPPROTO_TCP)
 575                         return -EINVAL;
 576
 577                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 578                         return -EINVAL;
 579
 580                 break;
 581
 582         case OVS_KEY_ATTR_UDP:
 583                 if (flow_key->ip.proto != IPPROTO_UDP)
 584                         return -EINVAL;
 585
 586                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 587                         return -EINVAL;
 588                 break;
 589
 590         default:
 591                 return -EINVAL;
 592         }
 593
 594         return 0;
 595 }
 596
 597 static int validate_userspace(const struct nlattr *attr)
 598 {
 599         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 600                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 601                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
 602         };
 603         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 604         int error;
 605
 606         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 607                                  attr, userspace_policy);
 608         if (error)
 609                 return error;
 610
 611         if (!a[OVS_USERSPACE_ATTR_PID] ||
 612             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 613                 return -EINVAL;
 614
 615         return 0;
 616 }
 617
 618 static int validate_actions(const struct nlattr *attr,
 619                                 const struct sw_flow_key *key,  int depth)
 620 {
 621         const struct nlattr *a;
 622         int rem, err;
 623
 624         if (depth >= SAMPLE_ACTION_DEPTH)
 625                 return -EOVERFLOW;
 626
 627         nla_for_each_nested(a, attr, rem) {
 628                 /* Expected argument lengths, (u32)-1 for variable length. */
 629                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 630                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 631                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 632                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 633                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 634                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 635                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 636                 };
 637                 const struct ovs_action_push_vlan *vlan;
 638                 int type = nla_type(a);
 639
 640                 if (type > OVS_ACTION_ATTR_MAX ||
 641                     (action_lens[type] != nla_len(a) &&
 642                      action_lens[type] != (u32)-1))
 643                         return -EINVAL;
 644
 645                 switch (type) {
 646                 case OVS_ACTION_ATTR_UNSPEC:
 647                         return -EINVAL;
 648
 649                 case OVS_ACTION_ATTR_USERSPACE:
 650                         err = validate_userspace(a);
 651                         if (err)
 652                                 return err;
 653                         break;
 654
 655                 case OVS_ACTION_ATTR_OUTPUT:
 656                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 657                                 return -EINVAL;
 658                         break;
 659
 660
 661                 case OVS_ACTION_ATTR_POP_VLAN:
 662                         break;
 663
 664                 case OVS_ACTION_ATTR_PUSH_VLAN:
 665                         vlan = nla_data(a);
 666                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 667                                 return -EINVAL;
 668                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 669                                 return -EINVAL;
 670                         break;
 671
 672                 case OVS_ACTION_ATTR_SET:
 673                         err = validate_set(a, key);
 674                         if (err)
 675                                 return err;
 676                         break;
 677
 678                 case OVS_ACTION_ATTR_SAMPLE:
 679                         err = validate_sample(a, key, depth);
 680                         if (err)
 681                                 return err;
 682                         break;
 683
 684                 default:
 685                         return -EINVAL;
 686                 }
 687         }
 688
 689         if (rem > 0)
 690                 return -EINVAL;
 691
 692         return 0;
 693 }
 694
 695 static void clear_stats(struct sw_flow *flow)
 696 {
 697         flow->used = 0;
 698         flow->tcp_flags = 0;
 699         flow->packet_count = 0;
 700         flow->byte_count = 0;
 701 }
 702
 703 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 704 {
 705         struct ovs_header *ovs_header = info->userhdr;
 706         struct nlattr **a = info->attrs;
 707         struct sw_flow_actions *acts;
 708         struct sk_buff *packet;
 709         struct sw_flow *flow;
 710         struct datapath *dp;
 711         struct ethhdr *eth;
 712         int len;
 713         int err;
 714         int key_len;
 715
 716         err = -EINVAL;
 717         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 718             !a[OVS_PACKET_ATTR_ACTIONS] ||
 719             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 720                 goto err;
 721
 722         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 723         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 724         err = -ENOMEM;
 725         if (!packet)
 726                 goto err;
 727         skb_reserve(packet, NET_IP_ALIGN);
 728
 729         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 730
 731         skb_reset_mac_header(packet);
 732         eth = eth_hdr(packet);
 733
 734         /* Normally, setting the skb 'protocol' field would be handled by a
 735          * call to eth_type_trans(), but it assumes there's a sending
 736          * device, which we may not have. */
 737         if (ntohs(eth->h_proto) >= 1536)
 738                 packet->protocol = eth->h_proto;
 739         else
 740                 packet->protocol = htons(ETH_P_802_2);
 741
 742         /* Build an sw_flow for sending this packet. */
 743         flow = flow_alloc();
 744         err = PTR_ERR(flow);
 745         if (IS_ERR(flow))
 746                 goto err_kfree_skb;
 747
 748         err = flow_extract(packet, -1, &flow->key, &key_len);
 749         if (err)
 750                 goto err_flow_put;
 751
 752         err = flow_metadata_from_nlattrs(&flow->key.phy.priority,
 753                                          &flow->key.phy.in_port,
 754                                          &flow->key.phy.tun_id,
 755                                          a[OVS_PACKET_ATTR_KEY]);
 756         if (err)
 757                 goto err_flow_put;
 758
 759         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
 760         if (err)
 761                 goto err_flow_put;
 762
 763         flow->hash = flow_hash(&flow->key, key_len);
 764
 765         acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 766         err = PTR_ERR(acts);
 767         if (IS_ERR(acts))
 768                 goto err_flow_put;
 769         rcu_assign_pointer(flow->sf_acts, acts);
 770
 771         OVS_CB(packet)->flow = flow;
 772         packet->priority = flow->key.phy.priority;
 773
 774         rcu_read_lock();
 775         dp = get_dp(ovs_header->dp_ifindex);
 776         err = -ENODEV;
 777         if (!dp)
 778                 goto err_unlock;
 779
 780         local_bh_disable();
 781         err = execute_actions(dp, packet);
 782         local_bh_enable();
 783         rcu_read_unlock();
 784
 785         flow_put(flow);
 786         return err;
 787
 788 err_unlock:
 789         rcu_read_unlock();
 790 err_flow_put:
 791         flow_put(flow);
 792 err_kfree_skb:
 793         kfree_skb(packet);
 794 err:
 795         return err;
 796 }
 797
 798 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 799         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 800         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 801         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 802 };
 803
 804 static struct genl_ops dp_packet_genl_ops[] = {
 805         { .cmd = OVS_PACKET_CMD_EXECUTE,
 806           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 807           .policy = packet_policy,
 808           .doit = ovs_packet_cmd_execute
 809         }
 810 };
 811
 812 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 813 {
 814         int i;
 815         struct flow_table *table = get_table_protected(dp);
 816
 817         stats->n_flows = flow_tbl_count(table);
 818
 819         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 820         for_each_possible_cpu(i) {
 821                 const struct dp_stats_percpu *percpu_stats;
 822                 struct dp_stats_percpu local_stats;
 823                 unsigned seqcount;
 824
 825                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 826
 827                 do {
 828                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 829                         local_stats = *percpu_stats;
 830                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 831
 832                 stats->n_hit += local_stats.n_hit;
 833                 stats->n_missed += local_stats.n_missed;
 834                 stats->n_lost += local_stats.n_lost;
 835         }
 836 }
 837
 838 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 839         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 840         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 841         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 842 };
 843
 844 static struct genl_family dp_flow_genl_family = {
 845         .id = GENL_ID_GENERATE,
 846         .hdrsize = sizeof(struct ovs_header),
 847         .name = OVS_FLOW_FAMILY,
 848         .version = OVS_FLOW_VERSION,
 849         .maxattr = OVS_FLOW_ATTR_MAX
 850 };
 851
 852 static struct genl_multicast_group dp_flow_multicast_group = {
 853         .name = OVS_FLOW_MCGROUP
 854 };
 855
 856 /* Called with genl_lock. */
 857 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 858                                   struct sk_buff *skb, u32 pid,
 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 = rcu_dereference_protected(flow->sf_acts,
 871                                             lockdep_genl_is_held());
 872
 873         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 874         if (!ovs_header)
 875                 return -EMSGSIZE;
 876
 877         ovs_header->dp_ifindex = get_dpifindex(dp);
 878
 879         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 880         if (!nla)
 881                 goto nla_put_failure;
 882         err = flow_to_nlattrs(&flow->key, skb);
 883         if (err)
 884                 goto error;
 885         nla_nest_end(skb, nla);
 886
 887         spin_lock_bh(&flow->lock);
 888         used = flow->used;
 889         stats.n_packets = flow->packet_count;
 890         stats.n_bytes = flow->byte_count;
 891         tcp_flags = flow->tcp_flags;
 892         spin_unlock_bh(&flow->lock);
 893
 894         if (used)
 895                 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
 896
 897         if (stats.n_packets)
 898                 NLA_PUT(skb, OVS_FLOW_ATTR_STATS,
 899                         sizeof(struct ovs_flow_stats), &stats);
 900
 901         if (tcp_flags)
 902                 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 903
 904         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 905          * this is the first flow to be dumped into 'skb'.  This is unusual for
 906          * Netlink but individual action lists can be longer than
 907          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 908          * The userspace caller can always fetch the actions separately if it
 909          * really wants them.  (Most userspace callers in fact don't care.)
 910          *
 911          * This can only fail for dump operations because the skb is always
 912          * properly sized for single flows.
 913          */
 914         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 915                       sf_acts->actions);
 916         if (err < 0 && skb_orig_len)
 917                 goto error;
 918
 919         return genlmsg_end(skb, ovs_header);
 920
 921 nla_put_failure:
 922         err = -EMSGSIZE;
 923 error:
 924         genlmsg_cancel(skb, ovs_header);
 925         return err;
 926 }
 927
 928 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 929 {
 930         const struct sw_flow_actions *sf_acts;
 931         int len;
 932
 933         sf_acts = rcu_dereference_protected(flow->sf_acts,
 934                                             lockdep_genl_is_held());
 935
 936         /* OVS_FLOW_ATTR_KEY */
 937         len = nla_total_size(FLOW_BUFSIZE);
 938         /* OVS_FLOW_ATTR_ACTIONS */
 939         len += nla_total_size(sf_acts->actions_len);
 940         /* OVS_FLOW_ATTR_STATS */
 941         len += nla_total_size(sizeof(struct ovs_flow_stats));
 942         /* OVS_FLOW_ATTR_TCP_FLAGS */
 943         len += nla_total_size(1);
 944         /* OVS_FLOW_ATTR_USED */
 945         len += nla_total_size(8);
 946
 947         len += NLMSG_ALIGN(sizeof(struct ovs_header));
 948
 949         return genlmsg_new(len, GFP_KERNEL);
 950 }
 951
 952 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 953                                                struct datapath *dp,
 954                                                u32 pid, u32 seq, u8 cmd)
 955 {
 956         struct sk_buff *skb;
 957         int retval;
 958
 959         skb = ovs_flow_cmd_alloc_info(flow);
 960         if (!skb)
 961                 return ERR_PTR(-ENOMEM);
 962
 963         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 964         BUG_ON(retval < 0);
 965         return skb;
 966 }
 967
 968 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 969 {
 970         struct nlattr **a = info->attrs;
 971         struct ovs_header *ovs_header = info->userhdr;
 972         struct sw_flow_key key;
 973         struct sw_flow *flow;
 974         struct sk_buff *reply;
 975         struct datapath *dp;
 976         struct flow_table *table;
 977         int error;
 978         int key_len;
 979
 980         /* Extract key. */
 981         error = -EINVAL;
 982         if (!a[OVS_FLOW_ATTR_KEY])
 983                 goto error;
 984         error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
 985         if (error)
 986                 goto error;
 987
 988         /* Validate actions. */
 989         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 990                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
 991                 if (error)
 992                         goto error;
 993         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 994                 error = -EINVAL;
 995                 goto error;
 996         }
 997
 998         dp = get_dp(ovs_header->dp_ifindex);
 999         error = -ENODEV;
1000         if (!dp)
1001                 goto error;
1002
1003         table = get_table_protected(dp);
1004         flow = flow_tbl_lookup(table, &key, key_len);
1005         if (!flow) {
1006                 struct sw_flow_actions *acts;
1007
1008                 /* Bail out if we're not allowed to create a new flow. */
1009                 error = -ENOENT;
1010                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1011                         goto error;
1012
1013                 /* Expand table, if necessary, to make room. */
1014                 if (flow_tbl_need_to_expand(table)) {
1015                         struct flow_table *new_table;
1016
1017                         new_table = flow_tbl_expand(table);
1018                         if (!IS_ERR(new_table)) {
1019                                 rcu_assign_pointer(dp->table, new_table);
1020                                 flow_tbl_deferred_destroy(table);
1021                                 table = get_table_protected(dp);
1022                         }
1023                 }
1024
1025                 /* Allocate flow. */
1026                 flow = flow_alloc();
1027                 if (IS_ERR(flow)) {
1028                         error = PTR_ERR(flow);
1029                         goto error;
1030                 }
1031                 flow->key = key;
1032                 clear_stats(flow);
1033
1034                 /* Obtain actions. */
1035                 acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1036                 error = PTR_ERR(acts);
1037                 if (IS_ERR(acts))
1038                         goto error_free_flow;
1039                 rcu_assign_pointer(flow->sf_acts, acts);
1040
1041                 /* Put flow in bucket. */
1042                 flow->hash = flow_hash(&key, key_len);
1043                 flow_tbl_insert(table, flow);
1044
1045                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1046                                                 info->snd_seq,
1047                                                 OVS_FLOW_CMD_NEW);
1048         } else {
1049                 /* We found a matching flow. */
1050                 struct sw_flow_actions *old_acts;
1051                 struct nlattr *acts_attrs;
1052
1053                 /* Bail out if we're not allowed to modify an existing flow.
1054                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1055                  * because Generic Netlink treats the latter as a dump
1056                  * request.  We also accept NLM_F_EXCL in case that bug ever
1057                  * gets fixed.
1058                  */
1059                 error = -EEXIST;
1060                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1061                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1062                         goto error;
1063
1064                 /* Update actions. */
1065                 old_acts = rcu_dereference_protected(flow->sf_acts,
1066                                                      lockdep_genl_is_held());
1067                 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1068                 if (acts_attrs &&
1069                    (old_acts->actions_len != nla_len(acts_attrs) ||
1070                    memcmp(old_acts->actions, nla_data(acts_attrs),
1071                           old_acts->actions_len))) {
1072                         struct sw_flow_actions *new_acts;
1073
1074                         new_acts = flow_actions_alloc(acts_attrs);
1075                         error = PTR_ERR(new_acts);
1076                         if (IS_ERR(new_acts))
1077                                 goto error;
1078
1079                         rcu_assign_pointer(flow->sf_acts, new_acts);
1080                         flow_deferred_free_acts(old_acts);
1081                 }
1082
1083                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1084                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1085
1086                 /* Clear stats. */
1087                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1088                         spin_lock_bh(&flow->lock);
1089                         clear_stats(flow);
1090                         spin_unlock_bh(&flow->lock);
1091                 }
1092         }
1093
1094         if (!IS_ERR(reply))
1095                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1096                            dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1097         else
1098                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1099                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1100         return 0;
1101
1102 error_free_flow:
1103         flow_put(flow);
1104 error:
1105         return error;
1106 }
1107
1108 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1109 {
1110         struct nlattr **a = info->attrs;
1111         struct ovs_header *ovs_header = info->userhdr;
1112         struct sw_flow_key key;
1113         struct sk_buff *reply;
1114         struct sw_flow *flow;
1115         struct datapath *dp;
1116         struct flow_table *table;
1117         int err;
1118         int key_len;
1119
1120         if (!a[OVS_FLOW_ATTR_KEY])
1121                 return -EINVAL;
1122         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1123         if (err)
1124                 return err;
1125
1126         dp = get_dp(ovs_header->dp_ifindex);
1127         if (!dp)
1128                 return -ENODEV;
1129
1130         table = get_table_protected(dp);
1131         flow = flow_tbl_lookup(table, &key, key_len);
1132         if (!flow)
1133                 return -ENOENT;
1134
1135         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1136                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1137         if (IS_ERR(reply))
1138                 return PTR_ERR(reply);
1139
1140         return genlmsg_reply(reply, info);
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         if (!a[OVS_FLOW_ATTR_KEY])
1156                 return flush_flows(ovs_header->dp_ifindex);
1157         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1158         if (err)
1159                 return err;
1160
1161         dp = get_dp(ovs_header->dp_ifindex);
1162         if (!dp)
1163                 return -ENODEV;
1164
1165         table = get_table_protected(dp);
1166         flow = flow_tbl_lookup(table, &key, key_len);
1167         if (!flow)
1168                 return -ENOENT;
1169
1170         reply = ovs_flow_cmd_alloc_info(flow);
1171         if (!reply)
1172                 return -ENOMEM;
1173
1174         flow_tbl_remove(table, flow);
1175
1176         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1177                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1178         BUG_ON(err < 0);
1179
1180         flow_deferred_free(flow);
1181
1182         genl_notify(reply, genl_info_net(info), info->snd_pid,
1183                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1184         return 0;
1185 }
1186
1187 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1188 {
1189         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1190         struct datapath *dp;
1191
1192         dp = get_dp(ovs_header->dp_ifindex);
1193         if (!dp)
1194                 return -ENODEV;
1195
1196         for (;;) {
1197                 struct sw_flow *flow;
1198                 u32 bucket, obj;
1199
1200                 bucket = cb->args[0];
1201                 obj = cb->args[1];
1202                 flow = flow_tbl_next(get_table_protected(dp), &bucket, &obj);
1203                 if (!flow)
1204                         break;
1205
1206                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1207                                            NETLINK_CB(cb->skb).pid,
1208                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1209                                            OVS_FLOW_CMD_NEW) < 0)
1210                         break;
1211
1212                 cb->args[0] = bucket;
1213                 cb->args[1] = obj;
1214         }
1215         return skb->len;
1216 }
1217
1218 static struct genl_ops dp_flow_genl_ops[] = {
1219         { .cmd = OVS_FLOW_CMD_NEW,
1220           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1221           .policy = flow_policy,
1222           .doit = ovs_flow_cmd_new_or_set
1223         },
1224         { .cmd = OVS_FLOW_CMD_DEL,
1225           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1226           .policy = flow_policy,
1227           .doit = ovs_flow_cmd_del
1228         },
1229         { .cmd = OVS_FLOW_CMD_GET,
1230           .flags = 0,               /* OK for unprivileged users. */
1231           .policy = flow_policy,
1232           .doit = ovs_flow_cmd_get,
1233           .dumpit = ovs_flow_cmd_dump
1234         },
1235         { .cmd = OVS_FLOW_CMD_SET,
1236           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1237           .policy = flow_policy,
1238           .doit = ovs_flow_cmd_new_or_set,
1239         },
1240 };
1241
1242 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1243 #ifdef HAVE_NLA_NUL_STRING
1244         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1245 #endif
1246         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1247 };
1248
1249 static struct genl_family dp_datapath_genl_family = {
1250         .id = GENL_ID_GENERATE,
1251         .hdrsize = sizeof(struct ovs_header),
1252         .name = OVS_DATAPATH_FAMILY,
1253         .version = OVS_DATAPATH_VERSION,
1254         .maxattr = OVS_DP_ATTR_MAX
1255 };
1256
1257 static struct genl_multicast_group dp_datapath_multicast_group = {
1258         .name = OVS_DATAPATH_MCGROUP
1259 };
1260
1261 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1262                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1263 {
1264         struct ovs_header *ovs_header;
1265         struct nlattr *nla;
1266         int err;
1267
1268         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1269                                    flags, cmd);
1270         if (!ovs_header)
1271                 goto error;
1272
1273         ovs_header->dp_ifindex = get_dpifindex(dp);
1274
1275         rcu_read_lock();
1276         err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
1277         rcu_read_unlock();
1278         if (err)
1279                 goto nla_put_failure;
1280
1281         nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats));
1282         if (!nla)
1283                 goto nla_put_failure;
1284         get_dp_stats(dp, nla_data(nla));
1285
1286         return genlmsg_end(skb, ovs_header);
1287
1288 nla_put_failure:
1289         genlmsg_cancel(skb, ovs_header);
1290 error:
1291         return -EMSGSIZE;
1292 }
1293
1294 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1295                                              u32 seq, u8 cmd)
1296 {
1297         struct sk_buff *skb;
1298         int retval;
1299
1300         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1301         if (!skb)
1302                 return ERR_PTR(-ENOMEM);
1303
1304         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1305         if (retval < 0) {
1306                 kfree_skb(skb);
1307                 return ERR_PTR(retval);
1308         }
1309         return skb;
1310 }
1311
1312 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1313 {
1314         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1315 }
1316
1317 /* Called with genl_mutex and optionally with RTNL lock also. */
1318 static struct datapath *lookup_datapath(struct ovs_header *ovs_header,
1319                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1320 {
1321         struct datapath *dp;
1322
1323         if (!a[OVS_DP_ATTR_NAME])
1324                 dp = get_dp(ovs_header->dp_ifindex);
1325         else {
1326                 struct vport *vport;
1327
1328                 rcu_read_lock();
1329                 vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1330                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1331                 rcu_read_unlock();
1332         }
1333         return dp ? dp : ERR_PTR(-ENODEV);
1334 }
1335
1336 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1337 {
1338         struct nlattr **a = info->attrs;
1339         struct vport_parms parms;
1340         struct sk_buff *reply;
1341         struct datapath *dp;
1342         struct vport *vport;
1343         int err;
1344
1345         err = -EINVAL;
1346         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1347                 goto err;
1348
1349         err = ovs_dp_cmd_validate(a);
1350         if (err)
1351                 goto err;
1352
1353         rtnl_lock();
1354         err = -ENODEV;
1355         if (!try_module_get(THIS_MODULE))
1356                 goto err_unlock_rtnl;
1357
1358         err = -ENOMEM;
1359         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1360         if (dp == NULL)
1361                 goto err_put_module;
1362         INIT_LIST_HEAD(&dp->port_list);
1363
1364         /* Initialize kobject for bridge.  This will be added as
1365          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1366         dp->ifobj.kset = NULL;
1367         kobject_init(&dp->ifobj, &dp_ktype);
1368
1369         /* Allocate table. */
1370         err = -ENOMEM;
1371         rcu_assign_pointer(dp->table, flow_tbl_alloc(TBL_MIN_BUCKETS));
1372         if (!dp->table)
1373                 goto err_free_dp;
1374
1375         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1376         if (!dp->stats_percpu) {
1377                 err = -ENOMEM;
1378                 goto err_destroy_table;
1379         }
1380
1381         /* Set up our datapath device. */
1382         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1383         parms.type = OVS_VPORT_TYPE_INTERNAL;
1384         parms.options = NULL;
1385         parms.dp = dp;
1386         parms.port_no = OVSP_LOCAL;
1387         parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1388
1389         vport = new_vport(&parms);
1390         if (IS_ERR(vport)) {
1391                 err = PTR_ERR(vport);
1392                 if (err == -EBUSY)
1393                         err = -EEXIST;
1394
1395                 goto err_destroy_percpu;
1396         }
1397
1398         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1399                                       info->snd_seq, OVS_DP_CMD_NEW);
1400         err = PTR_ERR(reply);
1401         if (IS_ERR(reply))
1402                 goto err_destroy_local_port;
1403
1404         list_add_tail(&dp->list_node, &dps);
1405         dp_sysfs_add_dp(dp);
1406
1407         rtnl_unlock();
1408
1409         genl_notify(reply, genl_info_net(info), info->snd_pid,
1410                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1411         return 0;
1412
1413 err_destroy_local_port:
1414         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1415 err_destroy_percpu:
1416         free_percpu(dp->stats_percpu);
1417 err_destroy_table:
1418         flow_tbl_destroy(get_table_protected(dp));
1419 err_free_dp:
1420         kfree(dp);
1421 err_put_module:
1422         module_put(THIS_MODULE);
1423 err_unlock_rtnl:
1424         rtnl_unlock();
1425 err:
1426         return err;
1427 }
1428
1429 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1430 {
1431         struct vport *vport, *next_vport;
1432         struct sk_buff *reply;
1433         struct datapath *dp;
1434         int err;
1435
1436         err = ovs_dp_cmd_validate(info->attrs);
1437         if (err)
1438                 goto exit;
1439
1440         rtnl_lock();
1441         dp = lookup_datapath(info->userhdr, info->attrs);
1442         err = PTR_ERR(dp);
1443         if (IS_ERR(dp))
1444                 goto exit_unlock;
1445
1446         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1447                                       info->snd_seq, OVS_DP_CMD_DEL);
1448         err = PTR_ERR(reply);
1449         if (IS_ERR(reply))
1450                 goto exit_unlock;
1451
1452         list_for_each_entry_safe(vport, next_vport, &dp->port_list, node)
1453                 if (vport->port_no != OVSP_LOCAL)
1454                         dp_detach_port(vport);
1455
1456         dp_sysfs_del_dp(dp);
1457         list_del(&dp->list_node);
1458         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1459
1460         /* rtnl_unlock() will wait until all the references to devices that
1461          * are pending unregistration have been dropped.  We do it here to
1462          * ensure that any internal devices (which contain DP pointers) are
1463          * fully destroyed before freeing the datapath.
1464          */
1465         rtnl_unlock();
1466
1467         call_rcu(&dp->rcu, destroy_dp_rcu);
1468         module_put(THIS_MODULE);
1469
1470         genl_notify(reply, genl_info_net(info), info->snd_pid,
1471                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1472
1473         return 0;
1474
1475 exit_unlock:
1476         rtnl_unlock();
1477 exit:
1478         return err;
1479 }
1480
1481 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1482 {
1483         struct sk_buff *reply;
1484         struct datapath *dp;
1485         int err;
1486
1487         err = ovs_dp_cmd_validate(info->attrs);
1488         if (err)
1489                 return err;
1490
1491         dp = lookup_datapath(info->userhdr, info->attrs);
1492         if (IS_ERR(dp))
1493                 return PTR_ERR(dp);
1494
1495         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1496                                       info->snd_seq, OVS_DP_CMD_NEW);
1497         if (IS_ERR(reply)) {
1498                 err = PTR_ERR(reply);
1499                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1500                                 dp_datapath_multicast_group.id, err);
1501                 return 0;
1502         }
1503
1504         genl_notify(reply, genl_info_net(info), info->snd_pid,
1505                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1506         return 0;
1507 }
1508
1509 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1510 {
1511         struct sk_buff *reply;
1512         struct datapath *dp;
1513         int err;
1514
1515         err = ovs_dp_cmd_validate(info->attrs);
1516         if (err)
1517                 return err;
1518
1519         dp = lookup_datapath(info->userhdr, info->attrs);
1520         if (IS_ERR(dp))
1521                 return PTR_ERR(dp);
1522
1523         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1524                                       info->snd_seq, OVS_DP_CMD_NEW);
1525         if (IS_ERR(reply))
1526                 return PTR_ERR(reply);
1527
1528         return genlmsg_reply(reply, info);
1529 }
1530
1531 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1532 {
1533         struct datapath *dp;
1534         int skip = cb->args[0];
1535         int i = 0;
1536
1537         list_for_each_entry(dp, &dps, list_node) {
1538                 if (i < skip)
1539                         continue;
1540                 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1541                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1542                                          OVS_DP_CMD_NEW) < 0)
1543                         break;
1544                 i++;
1545         }
1546
1547         cb->args[0] = i;
1548
1549         return skb->len;
1550 }
1551
1552 static struct genl_ops dp_datapath_genl_ops[] = {
1553         { .cmd = OVS_DP_CMD_NEW,
1554           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1555           .policy = datapath_policy,
1556           .doit = ovs_dp_cmd_new
1557         },
1558         { .cmd = OVS_DP_CMD_DEL,
1559           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1560           .policy = datapath_policy,
1561           .doit = ovs_dp_cmd_del
1562         },
1563         { .cmd = OVS_DP_CMD_GET,
1564           .flags = 0,               /* OK for unprivileged users. */
1565           .policy = datapath_policy,
1566           .doit = ovs_dp_cmd_get,
1567           .dumpit = ovs_dp_cmd_dump
1568         },
1569         { .cmd = OVS_DP_CMD_SET,
1570           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1571           .policy = datapath_policy,
1572           .doit = ovs_dp_cmd_set,
1573         },
1574 };
1575
1576 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1577 #ifdef HAVE_NLA_NUL_STRING
1578         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1579         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1580         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1581 #else
1582         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1583         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1584 #endif
1585         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1586         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1587         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1588         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1589 };
1590
1591 static struct genl_family dp_vport_genl_family = {
1592         .id = GENL_ID_GENERATE,
1593         .hdrsize = sizeof(struct ovs_header),
1594         .name = OVS_VPORT_FAMILY,
1595         .version = OVS_VPORT_VERSION,
1596         .maxattr = OVS_VPORT_ATTR_MAX
1597 };
1598
1599 struct genl_multicast_group dp_vport_multicast_group = {
1600         .name = OVS_VPORT_MCGROUP
1601 };
1602
1603 /* Called with RTNL lock or RCU read lock. */
1604 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1605                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1606 {
1607         struct ovs_header *ovs_header;
1608         struct nlattr *nla;
1609         int err;
1610
1611         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1612                                  flags, cmd);
1613         if (!ovs_header)
1614                 return -EMSGSIZE;
1615
1616         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1617
1618         NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1619         NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type);
1620         NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport));
1621         NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1622
1623         nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS,
1624                           sizeof(struct ovs_vport_stats));
1625         if (!nla)
1626                 goto nla_put_failure;
1627
1628         vport_get_stats(vport, nla_data(nla));
1629
1630         NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN,
1631                 vport->ops->get_addr(vport));
1632
1633         err = vport_get_options(vport, skb);
1634         if (err == -EMSGSIZE)
1635                 goto error;
1636
1637         return genlmsg_end(skb, ovs_header);
1638
1639 nla_put_failure:
1640         err = -EMSGSIZE;
1641 error:
1642         genlmsg_cancel(skb, ovs_header);
1643         return err;
1644 }
1645
1646 /* Called with RTNL lock or RCU read lock. */
1647 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1648                                          u32 seq, u8 cmd)
1649 {
1650         struct sk_buff *skb;
1651         int retval;
1652
1653         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1654         if (!skb)
1655                 return ERR_PTR(-ENOMEM);
1656
1657         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1658         if (retval < 0) {
1659                 kfree_skb(skb);
1660                 return ERR_PTR(retval);
1661         }
1662         return skb;
1663 }
1664
1665 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1666 {
1667         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1668 }
1669
1670 /* Called with RTNL lock or RCU read lock. */
1671 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1672                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1673 {
1674         struct datapath *dp;
1675         struct vport *vport;
1676
1677         if (a[OVS_VPORT_ATTR_NAME]) {
1678                 vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1679                 if (!vport)
1680                         return ERR_PTR(-ENODEV);
1681                 return vport;
1682         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1683                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1684
1685                 if (port_no >= DP_MAX_PORTS)
1686                         return ERR_PTR(-EFBIG);
1687
1688                 dp = get_dp(ovs_header->dp_ifindex);
1689                 if (!dp)
1690                         return ERR_PTR(-ENODEV);
1691
1692                 vport = get_vport_protected(dp, port_no);
1693                 if (!vport)
1694                         return ERR_PTR(-ENOENT);
1695                 return vport;
1696         } else
1697                 return ERR_PTR(-EINVAL);
1698 }
1699
1700 /* Called with RTNL lock. */
1701 static int change_vport(struct vport *vport,
1702                         struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1703 {
1704         int err = 0;
1705
1706         if (a[OVS_VPORT_ATTR_STATS])
1707                 vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1708
1709         if (a[OVS_VPORT_ATTR_ADDRESS])
1710                 err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1711
1712         return err;
1713 }
1714
1715 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1716 {
1717         struct nlattr **a = info->attrs;
1718         struct ovs_header *ovs_header = info->userhdr;
1719         struct vport_parms parms;
1720         struct sk_buff *reply;
1721         struct vport *vport;
1722         struct datapath *dp;
1723         u32 port_no;
1724         int err;
1725
1726         err = -EINVAL;
1727         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1728             !a[OVS_VPORT_ATTR_UPCALL_PID])
1729                 goto exit;
1730
1731         err = ovs_vport_cmd_validate(a);
1732         if (err)
1733                 goto exit;
1734
1735         rtnl_lock();
1736         dp = get_dp(ovs_header->dp_ifindex);
1737         err = -ENODEV;
1738         if (!dp)
1739                 goto exit_unlock;
1740
1741         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1742                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1743
1744                 err = -EFBIG;
1745                 if (port_no >= DP_MAX_PORTS)
1746                         goto exit_unlock;
1747
1748                 vport = get_vport_protected(dp, port_no);
1749                 err = -EBUSY;
1750                 if (vport)
1751                         goto exit_unlock;
1752         } else {
1753                 for (port_no = 1; ; port_no++) {
1754                         if (port_no >= DP_MAX_PORTS) {
1755                                 err = -EFBIG;
1756                                 goto exit_unlock;
1757                         }
1758                         vport = get_vport_protected(dp, port_no);
1759                         if (!vport)
1760                                 break;
1761                 }
1762         }
1763
1764         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1765         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1766         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1767         parms.dp = dp;
1768         parms.port_no = port_no;
1769         parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1770
1771         vport = new_vport(&parms);
1772         err = PTR_ERR(vport);
1773         if (IS_ERR(vport))
1774                 goto exit_unlock;
1775
1776         dp_sysfs_add_if(vport);
1777
1778         err = change_vport(vport, a);
1779         if (!err) {
1780                 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1781                                                  info->snd_seq,
1782                                                  OVS_VPORT_CMD_NEW);
1783                 if (IS_ERR(reply))
1784                         err = PTR_ERR(reply);
1785         }
1786         if (err) {
1787                 dp_detach_port(vport);
1788                 goto exit_unlock;
1789         }
1790         genl_notify(reply, genl_info_net(info), info->snd_pid,
1791                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1792
1793
1794 exit_unlock:
1795         rtnl_unlock();
1796 exit:
1797         return err;
1798 }
1799
1800 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1801 {
1802         struct nlattr **a = info->attrs;
1803         struct sk_buff *reply;
1804         struct vport *vport;
1805         int err;
1806
1807         err = ovs_vport_cmd_validate(a);
1808         if (err)
1809                 goto exit;
1810
1811         rtnl_lock();
1812         vport = lookup_vport(info->userhdr, a);
1813         err = PTR_ERR(vport);
1814         if (IS_ERR(vport))
1815                 goto exit_unlock;
1816
1817         err = 0;
1818         if (a[OVS_VPORT_ATTR_TYPE] &&
1819             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1820                 err = -EINVAL;
1821
1822         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1823                 err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1824         if (!err)
1825                 err = change_vport(vport, a);
1826         if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1827                 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1828
1829         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1830                                          OVS_VPORT_CMD_NEW);
1831         if (IS_ERR(reply)) {
1832                 err = PTR_ERR(reply);
1833                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1834                                 dp_vport_multicast_group.id, err);
1835                 return 0;
1836         }
1837
1838         genl_notify(reply, genl_info_net(info), info->snd_pid,
1839                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1840
1841 exit_unlock:
1842         rtnl_unlock();
1843 exit:
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         err = ovs_vport_cmd_validate(a);
1855         if (err)
1856                 goto exit;
1857
1858         rtnl_lock();
1859         vport = lookup_vport(info->userhdr, a);
1860         err = PTR_ERR(vport);
1861         if (IS_ERR(vport))
1862                 goto exit_unlock;
1863
1864         if (vport->port_no == OVSP_LOCAL) {
1865                 err = -EINVAL;
1866                 goto exit_unlock;
1867         }
1868
1869         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1870                                          OVS_VPORT_CMD_DEL);
1871         err = PTR_ERR(reply);
1872         if (IS_ERR(reply))
1873                 goto exit_unlock;
1874
1875         dp_detach_port(vport);
1876
1877         genl_notify(reply, genl_info_net(info), info->snd_pid,
1878                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1879
1880 exit_unlock:
1881         rtnl_unlock();
1882 exit:
1883         return err;
1884 }
1885
1886 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1887 {
1888         struct nlattr **a = info->attrs;
1889         struct ovs_header *ovs_header = info->userhdr;
1890         struct sk_buff *reply;
1891         struct vport *vport;
1892         int err;
1893
1894         err = ovs_vport_cmd_validate(a);
1895         if (err)
1896                 goto exit;
1897
1898         rcu_read_lock();
1899         vport = lookup_vport(ovs_header, a);
1900         err = PTR_ERR(vport);
1901         if (IS_ERR(vport))
1902                 goto exit_unlock;
1903
1904         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1905                                          OVS_VPORT_CMD_NEW);
1906         err = PTR_ERR(reply);
1907         if (IS_ERR(reply))
1908                 goto exit_unlock;
1909
1910         rcu_read_unlock();
1911
1912         return genlmsg_reply(reply, info);
1913
1914 exit_unlock:
1915         rcu_read_unlock();
1916 exit:
1917         return err;
1918 }
1919
1920 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1921 {
1922         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1923         struct datapath *dp;
1924         u32 port_no;
1925         int retval;
1926
1927         dp = get_dp(ovs_header->dp_ifindex);
1928         if (!dp)
1929                 return -ENODEV;
1930
1931         rcu_read_lock();
1932         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1933                 struct vport *vport;
1934
1935                 vport = get_vport_protected(dp, port_no);
1936                 if (!vport)
1937                         continue;
1938
1939                 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1940                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1941                                             OVS_VPORT_CMD_NEW) < 0)
1942                         break;
1943         }
1944         rcu_read_unlock();
1945
1946         cb->args[0] = port_no;
1947         retval = skb->len;
1948
1949         return retval;
1950 }
1951
1952 static struct genl_ops dp_vport_genl_ops[] = {
1953         { .cmd = OVS_VPORT_CMD_NEW,
1954           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1955           .policy = vport_policy,
1956           .doit = ovs_vport_cmd_new
1957         },
1958         { .cmd = OVS_VPORT_CMD_DEL,
1959           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1960           .policy = vport_policy,
1961           .doit = ovs_vport_cmd_del
1962         },
1963         { .cmd = OVS_VPORT_CMD_GET,
1964           .flags = 0,               /* OK for unprivileged users. */
1965           .policy = vport_policy,
1966           .doit = ovs_vport_cmd_get,
1967           .dumpit = ovs_vport_cmd_dump
1968         },
1969         { .cmd = OVS_VPORT_CMD_SET,
1970           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1971           .policy = vport_policy,
1972           .doit = ovs_vport_cmd_set,
1973         },
1974 };
1975
1976 struct genl_family_and_ops {
1977         struct genl_family *family;
1978         struct genl_ops *ops;
1979         int n_ops;
1980         struct genl_multicast_group *group;
1981 };
1982
1983 static const struct genl_family_and_ops dp_genl_families[] = {
1984         { &dp_datapath_genl_family,
1985           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1986           &dp_datapath_multicast_group },
1987         { &dp_vport_genl_family,
1988           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1989           &dp_vport_multicast_group },
1990         { &dp_flow_genl_family,
1991           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1992           &dp_flow_multicast_group },
1993         { &dp_packet_genl_family,
1994           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1995           NULL },
1996 };
1997
1998 static void dp_unregister_genl(int n_families)
1999 {
2000         int i;
2001
2002         for (i = 0; i < n_families; i++)
2003                 genl_unregister_family(dp_genl_families[i].family);
2004 }
2005
2006 static int dp_register_genl(void)
2007 {
2008         int n_registered;
2009         int err;
2010         int i;
2011
2012         n_registered = 0;
2013         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2014                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2015
2016                 err = genl_register_family_with_ops(f->family, f->ops,
2017                                                     f->n_ops);
2018                 if (err)
2019                         goto error;
2020                 n_registered++;
2021
2022                 if (f->group) {
2023                         err = genl_register_mc_group(f->family, f->group);
2024                         if (err)
2025                                 goto error;
2026                 }
2027         }
2028
2029         return 0;
2030
2031 error:
2032         dp_unregister_genl(n_registered);
2033         return err;
2034 }
2035
2036 static int __init dp_init(void)
2037 {
2038         struct sk_buff *dummy_skb;
2039         int err;
2040
2041         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2042
2043         pr_info("Open vSwitch %s, built "__DATE__" "__TIME__"\n",
2044                 VERSION BUILDNR);
2045
2046         err = tnl_init();
2047         if (err)
2048                 goto error;
2049
2050         err = flow_init();
2051         if (err)
2052                 goto error_tnl_exit;
2053
2054         err = vport_init();
2055         if (err)
2056                 goto error_flow_exit;
2057
2058         err = register_netdevice_notifier(&dp_device_notifier);
2059         if (err)
2060                 goto error_vport_exit;
2061
2062         err = dp_register_genl();
2063         if (err < 0)
2064                 goto error_unreg_notifier;
2065
2066         return 0;
2067
2068 error_unreg_notifier:
2069         unregister_netdevice_notifier(&dp_device_notifier);
2070 error_vport_exit:
2071         vport_exit();
2072 error_flow_exit:
2073         flow_exit();
2074 error_tnl_exit:
2075         tnl_exit();
2076 error:
2077         return err;
2078 }
2079
2080 static void dp_cleanup(void)
2081 {
2082         rcu_barrier();
2083         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2084         unregister_netdevice_notifier(&dp_device_notifier);
2085         vport_exit();
2086         flow_exit();
2087         tnl_exit();
2088 }
2089
2090 module_init(dp_init);
2091 module_exit(dp_cleanup);
2092
2093 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2094 MODULE_LICENSE("GPL");