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 static 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
 115 /* Must be called with genl_mutex. */
 116 static struct flow_table *get_table_protected(struct datapath *dp)
 117 {
 118         return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
 119 }
 120
 121 /* Must be called with rcu_read_lock or RTNL lock. */
 122 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
 123 {
 124         return rcu_dereference_rtnl(dp->ports[port_no]);
 125 }
 126
 127 /* Must be called with rcu_read_lock or RTNL lock. */
 128 const char *dp_name(const struct datapath *dp)
 129 {
 130         struct vport *vport = rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]);
 131         return vport->ops->get_name(vport);
 132 }
 133
 134 static int get_dpifindex(struct datapath *dp)
 135 {
 136         struct vport *local;
 137         int ifindex;
 138
 139         rcu_read_lock();
 140
 141         local = get_vport_protected(dp, OVSP_LOCAL);
 142         if (local)
 143                 ifindex = local->ops->get_ifindex(local);
 144         else
 145                 ifindex = 0;
 146
 147         rcu_read_unlock();
 148
 149         return ifindex;
 150 }
 151
 152 static size_t br_nlmsg_size(void)
 153 {
 154         return NLMSG_ALIGN(sizeof(struct ifinfomsg))
 155                + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
 156                + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
 157                + nla_total_size(4) /* IFLA_MASTER */
 158                + nla_total_size(4) /* IFLA_MTU */
 159                + nla_total_size(1); /* IFLA_OPERSTATE */
 160 }
 161
 162 /* Caller must hold RTNL lock. */
 163 static int dp_fill_ifinfo(struct sk_buff *skb,
 164                           const struct vport *port,
 165                           int event, unsigned int flags)
 166 {
 167         struct datapath *dp = port->dp;
 168         struct ifinfomsg *hdr;
 169         struct nlmsghdr *nlh;
 170
 171         if (!port->ops->get_ifindex)
 172                 return -ENODEV;
 173
 174         nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
 175         if (nlh == NULL)
 176                 return -EMSGSIZE;
 177
 178         hdr = nlmsg_data(nlh);
 179         hdr->ifi_family = AF_BRIDGE;
 180         hdr->__ifi_pad = 0;
 181         hdr->ifi_type = ARPHRD_ETHER;
 182         hdr->ifi_index = port->ops->get_ifindex(port);
 183         hdr->ifi_flags = port->ops->get_dev_flags(port);
 184         hdr->ifi_change = 0;
 185
 186         NLA_PUT_STRING(skb, IFLA_IFNAME, port->ops->get_name(port));
 187         NLA_PUT_U32(skb, IFLA_MASTER, get_dpifindex(dp));
 188         NLA_PUT_U32(skb, IFLA_MTU, port->ops->get_mtu(port));
 189 #ifdef IFLA_OPERSTATE
 190         NLA_PUT_U8(skb, IFLA_OPERSTATE,
 191                    port->ops->is_running(port)
 192                         ? port->ops->get_operstate(port)
 193                         : IF_OPER_DOWN);
 194 #endif
 195
 196         NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, port->ops->get_addr(port));
 197
 198         return nlmsg_end(skb, nlh);
 199
 200 nla_put_failure:
 201         nlmsg_cancel(skb, nlh);
 202         return -EMSGSIZE;
 203 }
 204
 205 /* Caller must hold RTNL lock. */
 206 static void dp_ifinfo_notify(int event, struct vport *port)
 207 {
 208         struct sk_buff *skb;
 209         int err;
 210
 211         skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
 212         if (!skb) {
 213                 err = -ENOBUFS;
 214                 goto err;
 215         }
 216
 217         err = dp_fill_ifinfo(skb, port, event, 0);
 218         if (err < 0) {
 219                 if (err == -ENODEV) {
 220                         goto out;
 221                 } else {
 222                         /* -EMSGSIZE implies BUG in br_nlmsg_size() */
 223                         WARN_ON(err == -EMSGSIZE);
 224                         goto err;
 225                 }
 226         }
 227
 228         rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 229
 230         return;
 231 err:
 232         rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
 233 out:
 234         kfree_skb(skb);
 235 }
 236
 237 static void release_dp(struct kobject *kobj)
 238 {
 239         struct datapath *dp = container_of(kobj, struct datapath, ifobj);
 240         kfree(dp);
 241 }
 242
 243 static struct kobj_type dp_ktype = {
 244         .release = release_dp
 245 };
 246
 247 static void destroy_dp_rcu(struct rcu_head *rcu)
 248 {
 249         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 250
 251         flow_tbl_destroy((__force struct flow_table *)dp->table);
 252         free_percpu(dp->stats_percpu);
 253         kobject_put(&dp->ifobj);
 254 }
 255
 256 /* Called with RTNL lock and genl_lock. */
 257 static struct vport *new_vport(const struct vport_parms *parms)
 258 {
 259         struct vport *vport;
 260
 261         vport = vport_add(parms);
 262         if (!IS_ERR(vport)) {
 263                 struct datapath *dp = parms->dp;
 264
 265                 rcu_assign_pointer(dp->ports[parms->port_no], vport);
 266                 list_add(&vport->node, &dp->port_list);
 267
 268                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 269         }
 270
 271         return vport;
 272 }
 273
 274 /* Called with RTNL lock. */
 275 void dp_detach_port(struct vport *p)
 276 {
 277         ASSERT_RTNL();
 278
 279         if (p->port_no != OVSP_LOCAL)
 280                 dp_sysfs_del_if(p);
 281         dp_ifinfo_notify(RTM_DELLINK, p);
 282
 283         /* First drop references to device. */
 284         list_del(&p->node);
 285         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 286
 287         /* Then destroy it. */
 288         vport_del(p);
 289 }
 290
 291 /* Must be called with rcu_read_lock. */
 292 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 293 {
 294         struct datapath *dp = p->dp;
 295         struct sw_flow *flow;
 296         struct dp_stats_percpu *stats;
 297         u64 *stats_counter;
 298         int error;
 299
 300         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 301
 302         if (!OVS_CB(skb)->flow) {
 303                 struct sw_flow_key key;
 304                 int key_len;
 305
 306                 /* Extract flow from 'skb' into 'key'. */
 307                 error = flow_extract(skb, p->port_no, &key, &key_len);
 308                 if (unlikely(error)) {
 309                         kfree_skb(skb);
 310                         return;
 311                 }
 312
 313                 /* Look up flow. */
 314                 flow = flow_tbl_lookup(rcu_dereference(dp->table),
 315                                        &key, key_len);
 316                 if (unlikely(!flow)) {
 317                         struct dp_upcall_info upcall;
 318
 319                         upcall.cmd = OVS_PACKET_CMD_MISS;
 320                         upcall.key = &key;
 321                         upcall.userdata = NULL;
 322                         upcall.pid = p->upcall_pid;
 323                         dp_upcall(dp, skb, &upcall);
 324                         consume_skb(skb);
 325                         stats_counter = &stats->n_missed;
 326                         goto out;
 327                 }
 328
 329                 OVS_CB(skb)->flow = flow;
 330         }
 331
 332         stats_counter = &stats->n_hit;
 333         flow_used(OVS_CB(skb)->flow, skb);
 334         execute_actions(dp, skb);
 335
 336 out:
 337         /* Update datapath statistics. */
 338         u64_stats_update_begin(&stats->sync);
 339         (*stats_counter)++;
 340         u64_stats_update_end(&stats->sync);
 341 }
 342
 343 static struct genl_family dp_packet_genl_family = {
 344         .id = GENL_ID_GENERATE,
 345         .hdrsize = sizeof(struct ovs_header),
 346         .name = OVS_PACKET_FAMILY,
 347         .version = OVS_PACKET_VERSION,
 348         .maxattr = OVS_PACKET_ATTR_MAX
 349 };
 350
 351 int dp_upcall(struct datapath *dp, struct sk_buff *skb,
 352               const struct dp_upcall_info *upcall_info)
 353 {
 354         struct dp_stats_percpu *stats;
 355         int dp_ifindex;
 356         int err;
 357
 358         if (upcall_info->pid == 0) {
 359                 err = -ENOTCONN;
 360                 goto err;
 361         }
 362
 363         dp_ifindex = get_dpifindex(dp);
 364         if (!dp_ifindex) {
 365                 err = -ENODEV;
 366                 goto err;
 367         }
 368
 369         forward_ip_summed(skb, true);
 370
 371         if (!skb_is_gso(skb))
 372                 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
 373         else
 374                 err = queue_gso_packets(dp_ifindex, skb, upcall_info);
 375         if (err)
 376                 goto err;
 377
 378         return 0;
 379
 380 err:
 381         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 382
 383         u64_stats_update_begin(&stats->sync);
 384         stats->n_lost++;
 385         u64_stats_update_end(&stats->sync);
 386
 387         return err;
 388 }
 389
 390 static int queue_gso_packets(int dp_ifindex, struct sk_buff *skb,
 391                              const struct dp_upcall_info *upcall_info)
 392 {
 393         struct dp_upcall_info later_info;
 394         struct sw_flow_key later_key;
 395         struct sk_buff *segs, *nskb;
 396         int err;
 397
 398         segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 399         if (IS_ERR(skb))
 400                 return PTR_ERR(skb);
 401
 402         /* Queue all of the segments. */
 403         skb = segs;
 404         do {
 405                 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
 406                 if (err)
 407                         break;
 408
 409                 if (skb == segs && skb_shinfo(skb)->gso_type & SKB_GSO_UDP) {
 410                         /* The initial flow key extracted by flow_extract() in
 411                          * this case is for a first fragment, so we need to
 412                          * properly mark later fragments.
 413                          */
 414                         later_key = *upcall_info->key;
 415                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 416
 417                         later_info = *upcall_info;
 418                         later_info.key = &later_key;
 419                         upcall_info = &later_info;
 420                 }
 421         } while ((skb = skb->next));
 422
 423         /* Free all of the segments. */
 424         skb = segs;
 425         do {
 426                 nskb = skb->next;
 427                 if (err)
 428                         kfree_skb(skb);
 429                 else
 430                         consume_skb(skb);
 431         } while ((skb = nskb));
 432         return err;
 433 }
 434
 435 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *skb,
 436                                   const struct dp_upcall_info *upcall_info)
 437 {
 438         struct ovs_header *upcall;
 439         struct sk_buff *nskb = NULL;
 440         struct sk_buff *user_skb; /* to be queued to userspace */
 441         struct nlattr *nla;
 442         unsigned int len;
 443         int err;
 444
 445         if (vlan_tx_tag_present(skb)) {
 446                 nskb = skb_clone(skb, GFP_ATOMIC);
 447                 if (!nskb)
 448                         return -ENOMEM;
 449
 450                 err = vlan_deaccel_tag(nskb);
 451                 if (err)
 452                         return err;
 453
 454                 skb = nskb;
 455         }
 456
 457         if (nla_attr_size(skb->len) > USHRT_MAX) {
 458                 err = -EFBIG;
 459                 goto out;
 460         }
 461
 462         len = sizeof(struct ovs_header);
 463         len += nla_total_size(skb->len);
 464         len += nla_total_size(FLOW_BUFSIZE);
 465         if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
 466                 len += nla_total_size(8);
 467
 468         user_skb = genlmsg_new(len, GFP_ATOMIC);
 469         if (!user_skb) {
 470                 err = -ENOMEM;
 471                 goto out;
 472         }
 473
 474         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 475                              0, upcall_info->cmd);
 476         upcall->dp_ifindex = dp_ifindex;
 477
 478         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 479         flow_to_nlattrs(upcall_info->key, user_skb);
 480         nla_nest_end(user_skb, nla);
 481
 482         if (upcall_info->userdata)
 483                 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
 484                             nla_get_u64(upcall_info->userdata));
 485
 486         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 487
 488         skb_copy_and_csum_dev(skb, nla_data(nla));
 489
 490         err = genlmsg_unicast(&init_net, user_skb, upcall_info->pid);
 491
 492 out:
 493         kfree_skb(nskb);
 494         return err;
 495 }
 496
 497 /* Called with genl_mutex. */
 498 static int flush_flows(int dp_ifindex)
 499 {
 500         struct flow_table *old_table;
 501         struct flow_table *new_table;
 502         struct datapath *dp;
 503
 504         dp = get_dp(dp_ifindex);
 505         if (!dp)
 506                 return -ENODEV;
 507
 508         old_table = get_table_protected(dp);
 509         new_table = flow_tbl_alloc(TBL_MIN_BUCKETS);
 510         if (!new_table)
 511                 return -ENOMEM;
 512
 513         rcu_assign_pointer(dp->table, new_table);
 514
 515         flow_tbl_deferred_destroy(old_table);
 516         return 0;
 517 }
 518
 519 static int validate_actions(const struct nlattr *attr,
 520                                 const struct sw_flow_key *key, int depth);
 521
 522 static int validate_sample(const struct nlattr *attr,
 523                                 const struct sw_flow_key *key, int depth)
 524 {
 525         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 526         const struct nlattr *probability, *actions;
 527         const struct nlattr *a;
 528         int rem;
 529
 530         memset(attrs, 0, sizeof(attrs));
 531         nla_for_each_nested(a, attr, rem) {
 532                 int type = nla_type(a);
 533                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 534                         return -EINVAL;
 535                 attrs[type] = a;
 536         }
 537         if (rem)
 538                 return -EINVAL;
 539
 540         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 541         if (!probability || nla_len(probability) != sizeof(u32))
 542                 return -EINVAL;
 543
 544         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 545         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 546                 return -EINVAL;
 547         return validate_actions(actions, key, depth + 1);
 548 }
 549
 550 static int validate_set(const struct nlattr *a,
 551                         const struct sw_flow_key *flow_key)
 552 {
 553         const struct nlattr *ovs_key = nla_data(a);
 554         int key_type = nla_type(ovs_key);
 555
 556         /* There can be only one key in a action */
 557         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 558                 return -EINVAL;
 559
 560         if (key_type > OVS_KEY_ATTR_MAX ||
 561             nla_len(ovs_key) != ovs_key_lens[key_type])
 562                 return -EINVAL;
 563
 564         switch (key_type) {
 565         const struct ovs_key_ipv4 *ipv4_key;
 566
 567         case OVS_KEY_ATTR_PRIORITY:
 568         case OVS_KEY_ATTR_TUN_ID:
 569         case OVS_KEY_ATTR_ETHERNET:
 570                 break;
 571
 572         case OVS_KEY_ATTR_IPV4:
 573                 if (flow_key->eth.type != htons(ETH_P_IP))
 574                         return -EINVAL;
 575
 576                 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
 577                         return -EINVAL;
 578
 579                 ipv4_key = nla_data(ovs_key);
 580                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 581                         return -EINVAL;
 582
 583                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 584                         return -EINVAL;
 585
 586                 break;
 587
 588         case OVS_KEY_ATTR_TCP:
 589                 if (flow_key->ip.proto != IPPROTO_TCP)
 590                         return -EINVAL;
 591
 592                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 593                         return -EINVAL;
 594
 595                 break;
 596
 597         case OVS_KEY_ATTR_UDP:
 598                 if (flow_key->ip.proto != IPPROTO_UDP)
 599                         return -EINVAL;
 600
 601                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 602                         return -EINVAL;
 603                 break;
 604
 605         default:
 606                 return -EINVAL;
 607         }
 608
 609         return 0;
 610 }
 611
 612 static int validate_userspace(const struct nlattr *attr)
 613 {
 614         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 615                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 616                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
 617         };
 618         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 619         int error;
 620
 621         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 622                                  attr, userspace_policy);
 623         if (error)
 624                 return error;
 625
 626         if (!a[OVS_USERSPACE_ATTR_PID] ||
 627             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 628                 return -EINVAL;
 629
 630         return 0;
 631 }
 632
 633 static int validate_actions(const struct nlattr *attr,
 634                                 const struct sw_flow_key *key,  int depth)
 635 {
 636         const struct nlattr *a;
 637         int rem, err;
 638
 639         if (depth >= SAMPLE_ACTION_DEPTH)
 640                 return -EOVERFLOW;
 641
 642         nla_for_each_nested(a, attr, rem) {
 643                 /* Expected argument lengths, (u32)-1 for variable length. */
 644                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 645                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 646                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 647                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 648                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 649                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 650                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 651                 };
 652                 const struct ovs_action_push_vlan *vlan;
 653                 int type = nla_type(a);
 654
 655                 if (type > OVS_ACTION_ATTR_MAX ||
 656                     (action_lens[type] != nla_len(a) &&
 657                      action_lens[type] != (u32)-1))
 658                         return -EINVAL;
 659
 660                 switch (type) {
 661                 case OVS_ACTION_ATTR_UNSPEC:
 662                         return -EINVAL;
 663
 664                 case OVS_ACTION_ATTR_USERSPACE:
 665                         err = validate_userspace(a);
 666                         if (err)
 667                                 return err;
 668                         break;
 669
 670                 case OVS_ACTION_ATTR_OUTPUT:
 671                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 672                                 return -EINVAL;
 673                         break;
 674
 675
 676                 case OVS_ACTION_ATTR_POP_VLAN:
 677                         break;
 678
 679                 case OVS_ACTION_ATTR_PUSH_VLAN:
 680                         vlan = nla_data(a);
 681                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 682                                 return -EINVAL;
 683                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 684                                 return -EINVAL;
 685                         break;
 686
 687                 case OVS_ACTION_ATTR_SET:
 688                         err = validate_set(a, key);
 689                         if (err)
 690                                 return err;
 691                         break;
 692
 693                 case OVS_ACTION_ATTR_SAMPLE:
 694                         err = validate_sample(a, key, depth);
 695                         if (err)
 696                                 return err;
 697                         break;
 698
 699                 default:
 700                         return -EINVAL;
 701                 }
 702         }
 703
 704         if (rem > 0)
 705                 return -EINVAL;
 706
 707         return 0;
 708 }
 709
 710 static void clear_stats(struct sw_flow *flow)
 711 {
 712         flow->used = 0;
 713         flow->tcp_flags = 0;
 714         flow->packet_count = 0;
 715         flow->byte_count = 0;
 716 }
 717
 718 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 719 {
 720         struct ovs_header *ovs_header = info->userhdr;
 721         struct nlattr **a = info->attrs;
 722         struct sw_flow_actions *acts;
 723         struct sk_buff *packet;
 724         struct sw_flow *flow;
 725         struct datapath *dp;
 726         struct ethhdr *eth;
 727         int len;
 728         int err;
 729         int key_len;
 730
 731         err = -EINVAL;
 732         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 733             !a[OVS_PACKET_ATTR_ACTIONS] ||
 734             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 735                 goto err;
 736
 737         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 738         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 739         err = -ENOMEM;
 740         if (!packet)
 741                 goto err;
 742         skb_reserve(packet, NET_IP_ALIGN);
 743
 744         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 745
 746         skb_reset_mac_header(packet);
 747         eth = eth_hdr(packet);
 748
 749         /* Normally, setting the skb 'protocol' field would be handled by a
 750          * call to eth_type_trans(), but it assumes there's a sending
 751          * device, which we may not have. */
 752         if (ntohs(eth->h_proto) >= 1536)
 753                 packet->protocol = eth->h_proto;
 754         else
 755                 packet->protocol = htons(ETH_P_802_2);
 756
 757         /* Build an sw_flow for sending this packet. */
 758         flow = flow_alloc();
 759         err = PTR_ERR(flow);
 760         if (IS_ERR(flow))
 761                 goto err_kfree_skb;
 762
 763         err = flow_extract(packet, -1, &flow->key, &key_len);
 764         if (err)
 765                 goto err_flow_put;
 766
 767         err = flow_metadata_from_nlattrs(&flow->key.phy.priority,
 768                                          &flow->key.phy.in_port,
 769                                          &flow->key.phy.tun_id,
 770                                          a[OVS_PACKET_ATTR_KEY]);
 771         if (err)
 772                 goto err_flow_put;
 773
 774         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
 775         if (err)
 776                 goto err_flow_put;
 777
 778         flow->hash = flow_hash(&flow->key, key_len);
 779
 780         acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 781         err = PTR_ERR(acts);
 782         if (IS_ERR(acts))
 783                 goto err_flow_put;
 784         rcu_assign_pointer(flow->sf_acts, acts);
 785
 786         OVS_CB(packet)->flow = flow;
 787         packet->priority = flow->key.phy.priority;
 788
 789         rcu_read_lock();
 790         dp = get_dp(ovs_header->dp_ifindex);
 791         err = -ENODEV;
 792         if (!dp)
 793                 goto err_unlock;
 794
 795         local_bh_disable();
 796         err = execute_actions(dp, packet);
 797         local_bh_enable();
 798         rcu_read_unlock();
 799
 800         flow_put(flow);
 801         return err;
 802
 803 err_unlock:
 804         rcu_read_unlock();
 805 err_flow_put:
 806         flow_put(flow);
 807 err_kfree_skb:
 808         kfree_skb(packet);
 809 err:
 810         return err;
 811 }
 812
 813 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 814         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 815         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 816         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 817 };
 818
 819 static struct genl_ops dp_packet_genl_ops[] = {
 820         { .cmd = OVS_PACKET_CMD_EXECUTE,
 821           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 822           .policy = packet_policy,
 823           .doit = ovs_packet_cmd_execute
 824         }
 825 };
 826
 827 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 828 {
 829         int i;
 830         struct flow_table *table = get_table_protected(dp);
 831
 832         stats->n_flows = flow_tbl_count(table);
 833
 834         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 835         for_each_possible_cpu(i) {
 836                 const struct dp_stats_percpu *percpu_stats;
 837                 struct dp_stats_percpu local_stats;
 838                 unsigned int start;
 839
 840                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 841
 842                 do {
 843                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 844                         local_stats = *percpu_stats;
 845                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 846
 847                 stats->n_hit += local_stats.n_hit;
 848                 stats->n_missed += local_stats.n_missed;
 849                 stats->n_lost += local_stats.n_lost;
 850         }
 851 }
 852
 853 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 854         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 855         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 856         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 857 };
 858
 859 static struct genl_family dp_flow_genl_family = {
 860         .id = GENL_ID_GENERATE,
 861         .hdrsize = sizeof(struct ovs_header),
 862         .name = OVS_FLOW_FAMILY,
 863         .version = OVS_FLOW_VERSION,
 864         .maxattr = OVS_FLOW_ATTR_MAX
 865 };
 866
 867 static struct genl_multicast_group dp_flow_multicast_group = {
 868         .name = OVS_FLOW_MCGROUP
 869 };
 870
 871 /* Called with genl_lock. */
 872 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 873                                   struct sk_buff *skb, u32 pid,
 874                                   u32 seq, u32 flags, u8 cmd)
 875 {
 876         const int skb_orig_len = skb->len;
 877         const struct sw_flow_actions *sf_acts;
 878         struct ovs_flow_stats stats;
 879         struct ovs_header *ovs_header;
 880         struct nlattr *nla;
 881         unsigned long used;
 882         u8 tcp_flags;
 883         int err;
 884
 885         sf_acts = rcu_dereference_protected(flow->sf_acts,
 886                                             lockdep_genl_is_held());
 887
 888         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 889         if (!ovs_header)
 890                 return -EMSGSIZE;
 891
 892         ovs_header->dp_ifindex = get_dpifindex(dp);
 893
 894         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 895         if (!nla)
 896                 goto nla_put_failure;
 897         err = flow_to_nlattrs(&flow->key, skb);
 898         if (err)
 899                 goto error;
 900         nla_nest_end(skb, nla);
 901
 902         spin_lock_bh(&flow->lock);
 903         used = flow->used;
 904         stats.n_packets = flow->packet_count;
 905         stats.n_bytes = flow->byte_count;
 906         tcp_flags = flow->tcp_flags;
 907         spin_unlock_bh(&flow->lock);
 908
 909         if (used)
 910                 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
 911
 912         if (stats.n_packets)
 913                 NLA_PUT(skb, OVS_FLOW_ATTR_STATS,
 914                         sizeof(struct ovs_flow_stats), &stats);
 915
 916         if (tcp_flags)
 917                 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 918
 919         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 920          * this is the first flow to be dumped into 'skb'.  This is unusual for
 921          * Netlink but individual action lists can be longer than
 922          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 923          * The userspace caller can always fetch the actions separately if it
 924          * really wants them.  (Most userspace callers in fact don't care.)
 925          *
 926          * This can only fail for dump operations because the skb is always
 927          * properly sized for single flows.
 928          */
 929         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 930                       sf_acts->actions);
 931         if (err < 0 && skb_orig_len)
 932                 goto error;
 933
 934         return genlmsg_end(skb, ovs_header);
 935
 936 nla_put_failure:
 937         err = -EMSGSIZE;
 938 error:
 939         genlmsg_cancel(skb, ovs_header);
 940         return err;
 941 }
 942
 943 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 944 {
 945         const struct sw_flow_actions *sf_acts;
 946         int len;
 947
 948         sf_acts = rcu_dereference_protected(flow->sf_acts,
 949                                             lockdep_genl_is_held());
 950
 951         /* OVS_FLOW_ATTR_KEY */
 952         len = nla_total_size(FLOW_BUFSIZE);
 953         /* OVS_FLOW_ATTR_ACTIONS */
 954         len += nla_total_size(sf_acts->actions_len);
 955         /* OVS_FLOW_ATTR_STATS */
 956         len += nla_total_size(sizeof(struct ovs_flow_stats));
 957         /* OVS_FLOW_ATTR_TCP_FLAGS */
 958         len += nla_total_size(1);
 959         /* OVS_FLOW_ATTR_USED */
 960         len += nla_total_size(8);
 961
 962         len += NLMSG_ALIGN(sizeof(struct ovs_header));
 963
 964         return genlmsg_new(len, GFP_KERNEL);
 965 }
 966
 967 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 968                                                struct datapath *dp,
 969                                                u32 pid, u32 seq, u8 cmd)
 970 {
 971         struct sk_buff *skb;
 972         int retval;
 973
 974         skb = ovs_flow_cmd_alloc_info(flow);
 975         if (!skb)
 976                 return ERR_PTR(-ENOMEM);
 977
 978         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 979         BUG_ON(retval < 0);
 980         return skb;
 981 }
 982
 983 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 984 {
 985         struct nlattr **a = info->attrs;
 986         struct ovs_header *ovs_header = info->userhdr;
 987         struct sw_flow_key key;
 988         struct sw_flow *flow;
 989         struct sk_buff *reply;
 990         struct datapath *dp;
 991         struct flow_table *table;
 992         int error;
 993         int key_len;
 994
 995         /* Extract key. */
 996         error = -EINVAL;
 997         if (!a[OVS_FLOW_ATTR_KEY])
 998                 goto error;
 999         error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1000         if (error)
1001                 goto error;
1002
1003         /* Validate actions. */
1004         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1005                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
1006                 if (error)
1007                         goto error;
1008         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1009                 error = -EINVAL;
1010                 goto error;
1011         }
1012
1013         dp = get_dp(ovs_header->dp_ifindex);
1014         error = -ENODEV;
1015         if (!dp)
1016                 goto error;
1017
1018         table = get_table_protected(dp);
1019         flow = flow_tbl_lookup(table, &key, key_len);
1020         if (!flow) {
1021                 struct sw_flow_actions *acts;
1022
1023                 /* Bail out if we're not allowed to create a new flow. */
1024                 error = -ENOENT;
1025                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1026                         goto error;
1027
1028                 /* Expand table, if necessary, to make room. */
1029                 if (flow_tbl_need_to_expand(table)) {
1030                         struct flow_table *new_table;
1031
1032                         new_table = flow_tbl_expand(table);
1033                         if (!IS_ERR(new_table)) {
1034                                 rcu_assign_pointer(dp->table, new_table);
1035                                 flow_tbl_deferred_destroy(table);
1036                                 table = get_table_protected(dp);
1037                         }
1038                 }
1039
1040                 /* Allocate flow. */
1041                 flow = flow_alloc();
1042                 if (IS_ERR(flow)) {
1043                         error = PTR_ERR(flow);
1044                         goto error;
1045                 }
1046                 flow->key = key;
1047                 clear_stats(flow);
1048
1049                 /* Obtain actions. */
1050                 acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1051                 error = PTR_ERR(acts);
1052                 if (IS_ERR(acts))
1053                         goto error_free_flow;
1054                 rcu_assign_pointer(flow->sf_acts, acts);
1055
1056                 /* Put flow in bucket. */
1057                 flow->hash = flow_hash(&key, key_len);
1058                 flow_tbl_insert(table, flow);
1059
1060                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1061                                                 info->snd_seq,
1062                                                 OVS_FLOW_CMD_NEW);
1063         } else {
1064                 /* We found a matching flow. */
1065                 struct sw_flow_actions *old_acts;
1066                 struct nlattr *acts_attrs;
1067
1068                 /* Bail out if we're not allowed to modify an existing flow.
1069                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1070                  * because Generic Netlink treats the latter as a dump
1071                  * request.  We also accept NLM_F_EXCL in case that bug ever
1072                  * gets fixed.
1073                  */
1074                 error = -EEXIST;
1075                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1076                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1077                         goto error;
1078
1079                 /* Update actions. */
1080                 old_acts = rcu_dereference_protected(flow->sf_acts,
1081                                                      lockdep_genl_is_held());
1082                 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1083                 if (acts_attrs &&
1084                    (old_acts->actions_len != nla_len(acts_attrs) ||
1085                    memcmp(old_acts->actions, nla_data(acts_attrs),
1086                           old_acts->actions_len))) {
1087                         struct sw_flow_actions *new_acts;
1088
1089                         new_acts = flow_actions_alloc(acts_attrs);
1090                         error = PTR_ERR(new_acts);
1091                         if (IS_ERR(new_acts))
1092                                 goto error;
1093
1094                         rcu_assign_pointer(flow->sf_acts, new_acts);
1095                         flow_deferred_free_acts(old_acts);
1096                 }
1097
1098                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1099                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1100
1101                 /* Clear stats. */
1102                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1103                         spin_lock_bh(&flow->lock);
1104                         clear_stats(flow);
1105                         spin_unlock_bh(&flow->lock);
1106                 }
1107         }
1108
1109         if (!IS_ERR(reply))
1110                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1111                            dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1112         else
1113                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1114                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1115         return 0;
1116
1117 error_free_flow:
1118         flow_put(flow);
1119 error:
1120         return error;
1121 }
1122
1123 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1124 {
1125         struct nlattr **a = info->attrs;
1126         struct ovs_header *ovs_header = info->userhdr;
1127         struct sw_flow_key key;
1128         struct sk_buff *reply;
1129         struct sw_flow *flow;
1130         struct datapath *dp;
1131         struct flow_table *table;
1132         int err;
1133         int key_len;
1134
1135         if (!a[OVS_FLOW_ATTR_KEY])
1136                 return -EINVAL;
1137         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1138         if (err)
1139                 return err;
1140
1141         dp = get_dp(ovs_header->dp_ifindex);
1142         if (!dp)
1143                 return -ENODEV;
1144
1145         table = get_table_protected(dp);
1146         flow = flow_tbl_lookup(table, &key, key_len);
1147         if (!flow)
1148                 return -ENOENT;
1149
1150         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1151                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1152         if (IS_ERR(reply))
1153                 return PTR_ERR(reply);
1154
1155         return genlmsg_reply(reply, info);
1156 }
1157
1158 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1159 {
1160         struct nlattr **a = info->attrs;
1161         struct ovs_header *ovs_header = info->userhdr;
1162         struct sw_flow_key key;
1163         struct sk_buff *reply;
1164         struct sw_flow *flow;
1165         struct datapath *dp;
1166         struct flow_table *table;
1167         int err;
1168         int key_len;
1169
1170         if (!a[OVS_FLOW_ATTR_KEY])
1171                 return flush_flows(ovs_header->dp_ifindex);
1172         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1173         if (err)
1174                 return err;
1175
1176         dp = get_dp(ovs_header->dp_ifindex);
1177         if (!dp)
1178                 return -ENODEV;
1179
1180         table = get_table_protected(dp);
1181         flow = flow_tbl_lookup(table, &key, key_len);
1182         if (!flow)
1183                 return -ENOENT;
1184
1185         reply = ovs_flow_cmd_alloc_info(flow);
1186         if (!reply)
1187                 return -ENOMEM;
1188
1189         flow_tbl_remove(table, flow);
1190
1191         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1192                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1193         BUG_ON(err < 0);
1194
1195         flow_deferred_free(flow);
1196
1197         genl_notify(reply, genl_info_net(info), info->snd_pid,
1198                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1199         return 0;
1200 }
1201
1202 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1203 {
1204         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1205         struct datapath *dp;
1206
1207         dp = get_dp(ovs_header->dp_ifindex);
1208         if (!dp)
1209                 return -ENODEV;
1210
1211         for (;;) {
1212                 struct sw_flow *flow;
1213                 u32 bucket, obj;
1214
1215                 bucket = cb->args[0];
1216                 obj = cb->args[1];
1217                 flow = flow_tbl_next(get_table_protected(dp), &bucket, &obj);
1218                 if (!flow)
1219                         break;
1220
1221                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1222                                            NETLINK_CB(cb->skb).pid,
1223                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1224                                            OVS_FLOW_CMD_NEW) < 0)
1225                         break;
1226
1227                 cb->args[0] = bucket;
1228                 cb->args[1] = obj;
1229         }
1230         return skb->len;
1231 }
1232
1233 static struct genl_ops dp_flow_genl_ops[] = {
1234         { .cmd = OVS_FLOW_CMD_NEW,
1235           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1236           .policy = flow_policy,
1237           .doit = ovs_flow_cmd_new_or_set
1238         },
1239         { .cmd = OVS_FLOW_CMD_DEL,
1240           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1241           .policy = flow_policy,
1242           .doit = ovs_flow_cmd_del
1243         },
1244         { .cmd = OVS_FLOW_CMD_GET,
1245           .flags = 0,               /* OK for unprivileged users. */
1246           .policy = flow_policy,
1247           .doit = ovs_flow_cmd_get,
1248           .dumpit = ovs_flow_cmd_dump
1249         },
1250         { .cmd = OVS_FLOW_CMD_SET,
1251           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1252           .policy = flow_policy,
1253           .doit = ovs_flow_cmd_new_or_set,
1254         },
1255 };
1256
1257 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1258 #ifdef HAVE_NLA_NUL_STRING
1259         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1260 #endif
1261         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1262 };
1263
1264 static struct genl_family dp_datapath_genl_family = {
1265         .id = GENL_ID_GENERATE,
1266         .hdrsize = sizeof(struct ovs_header),
1267         .name = OVS_DATAPATH_FAMILY,
1268         .version = OVS_DATAPATH_VERSION,
1269         .maxattr = OVS_DP_ATTR_MAX
1270 };
1271
1272 static struct genl_multicast_group dp_datapath_multicast_group = {
1273         .name = OVS_DATAPATH_MCGROUP
1274 };
1275
1276 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1277                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1278 {
1279         struct ovs_header *ovs_header;
1280         struct ovs_dp_stats dp_stats;
1281         int err;
1282
1283         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1284                                    flags, cmd);
1285         if (!ovs_header)
1286                 goto error;
1287
1288         ovs_header->dp_ifindex = get_dpifindex(dp);
1289
1290         rcu_read_lock();
1291         err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
1292         rcu_read_unlock();
1293         if (err)
1294                 goto nla_put_failure;
1295
1296         get_dp_stats(dp, &dp_stats);
1297         NLA_PUT(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats);
1298
1299         return genlmsg_end(skb, ovs_header);
1300
1301 nla_put_failure:
1302         genlmsg_cancel(skb, ovs_header);
1303 error:
1304         return -EMSGSIZE;
1305 }
1306
1307 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1308                                              u32 seq, u8 cmd)
1309 {
1310         struct sk_buff *skb;
1311         int retval;
1312
1313         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1314         if (!skb)
1315                 return ERR_PTR(-ENOMEM);
1316
1317         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1318         if (retval < 0) {
1319                 kfree_skb(skb);
1320                 return ERR_PTR(retval);
1321         }
1322         return skb;
1323 }
1324
1325 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1326 {
1327         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1328 }
1329
1330 /* Called with genl_mutex and optionally with RTNL lock also. */
1331 static struct datapath *lookup_datapath(struct ovs_header *ovs_header,
1332                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1333 {
1334         struct datapath *dp;
1335
1336         if (!a[OVS_DP_ATTR_NAME])
1337                 dp = get_dp(ovs_header->dp_ifindex);
1338         else {
1339                 struct vport *vport;
1340
1341                 rcu_read_lock();
1342                 vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1343                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1344                 rcu_read_unlock();
1345         }
1346         return dp ? dp : ERR_PTR(-ENODEV);
1347 }
1348
1349 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1350 {
1351         struct nlattr **a = info->attrs;
1352         struct vport_parms parms;
1353         struct sk_buff *reply;
1354         struct datapath *dp;
1355         struct vport *vport;
1356         int err;
1357
1358         err = -EINVAL;
1359         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1360                 goto err;
1361
1362         err = ovs_dp_cmd_validate(a);
1363         if (err)
1364                 goto err;
1365
1366         rtnl_lock();
1367         err = -ENODEV;
1368         if (!try_module_get(THIS_MODULE))
1369                 goto err_unlock_rtnl;
1370
1371         err = -ENOMEM;
1372         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1373         if (dp == NULL)
1374                 goto err_put_module;
1375         INIT_LIST_HEAD(&dp->port_list);
1376
1377         /* Initialize kobject for bridge.  This will be added as
1378          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1379         dp->ifobj.kset = NULL;
1380         kobject_init(&dp->ifobj, &dp_ktype);
1381
1382         /* Allocate table. */
1383         err = -ENOMEM;
1384         rcu_assign_pointer(dp->table, 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         /* Set up our datapath device. */
1395         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1396         parms.type = OVS_VPORT_TYPE_INTERNAL;
1397         parms.options = NULL;
1398         parms.dp = dp;
1399         parms.port_no = OVSP_LOCAL;
1400         parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1401
1402         vport = new_vport(&parms);
1403         if (IS_ERR(vport)) {
1404                 err = PTR_ERR(vport);
1405                 if (err == -EBUSY)
1406                         err = -EEXIST;
1407
1408                 goto err_destroy_percpu;
1409         }
1410
1411         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1412                                       info->snd_seq, OVS_DP_CMD_NEW);
1413         err = PTR_ERR(reply);
1414         if (IS_ERR(reply))
1415                 goto err_destroy_local_port;
1416
1417         list_add_tail(&dp->list_node, &dps);
1418         dp_sysfs_add_dp(dp);
1419
1420         rtnl_unlock();
1421
1422         genl_notify(reply, genl_info_net(info), info->snd_pid,
1423                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1424         return 0;
1425
1426 err_destroy_local_port:
1427         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1428 err_destroy_percpu:
1429         free_percpu(dp->stats_percpu);
1430 err_destroy_table:
1431         flow_tbl_destroy(get_table_protected(dp));
1432 err_free_dp:
1433         kfree(dp);
1434 err_put_module:
1435         module_put(THIS_MODULE);
1436 err_unlock_rtnl:
1437         rtnl_unlock();
1438 err:
1439         return err;
1440 }
1441
1442 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1443 {
1444         struct vport *vport, *next_vport;
1445         struct sk_buff *reply;
1446         struct datapath *dp;
1447         int err;
1448
1449         err = ovs_dp_cmd_validate(info->attrs);
1450         if (err)
1451                 goto exit;
1452
1453         rtnl_lock();
1454         dp = lookup_datapath(info->userhdr, info->attrs);
1455         err = PTR_ERR(dp);
1456         if (IS_ERR(dp))
1457                 goto exit_unlock;
1458
1459         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1460                                       info->snd_seq, OVS_DP_CMD_DEL);
1461         err = PTR_ERR(reply);
1462         if (IS_ERR(reply))
1463                 goto exit_unlock;
1464
1465         list_for_each_entry_safe(vport, next_vport, &dp->port_list, node)
1466                 if (vport->port_no != OVSP_LOCAL)
1467                         dp_detach_port(vport);
1468
1469         dp_sysfs_del_dp(dp);
1470         list_del(&dp->list_node);
1471         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1472
1473         /* rtnl_unlock() will wait until all the references to devices that
1474          * are pending unregistration have been dropped.  We do it here to
1475          * ensure that any internal devices (which contain DP pointers) are
1476          * fully destroyed before freeing the datapath.
1477          */
1478         rtnl_unlock();
1479
1480         call_rcu(&dp->rcu, destroy_dp_rcu);
1481         module_put(THIS_MODULE);
1482
1483         genl_notify(reply, genl_info_net(info), info->snd_pid,
1484                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1485
1486         return 0;
1487
1488 exit_unlock:
1489         rtnl_unlock();
1490 exit:
1491         return err;
1492 }
1493
1494 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1495 {
1496         struct sk_buff *reply;
1497         struct datapath *dp;
1498         int err;
1499
1500         err = ovs_dp_cmd_validate(info->attrs);
1501         if (err)
1502                 return err;
1503
1504         dp = lookup_datapath(info->userhdr, info->attrs);
1505         if (IS_ERR(dp))
1506                 return PTR_ERR(dp);
1507
1508         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1509                                       info->snd_seq, OVS_DP_CMD_NEW);
1510         if (IS_ERR(reply)) {
1511                 err = PTR_ERR(reply);
1512                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1513                                 dp_datapath_multicast_group.id, err);
1514                 return 0;
1515         }
1516
1517         genl_notify(reply, genl_info_net(info), info->snd_pid,
1518                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1519         return 0;
1520 }
1521
1522 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1523 {
1524         struct sk_buff *reply;
1525         struct datapath *dp;
1526         int err;
1527
1528         err = ovs_dp_cmd_validate(info->attrs);
1529         if (err)
1530                 return err;
1531
1532         dp = lookup_datapath(info->userhdr, info->attrs);
1533         if (IS_ERR(dp))
1534                 return PTR_ERR(dp);
1535
1536         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1537                                       info->snd_seq, OVS_DP_CMD_NEW);
1538         if (IS_ERR(reply))
1539                 return PTR_ERR(reply);
1540
1541         return genlmsg_reply(reply, info);
1542 }
1543
1544 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1545 {
1546         struct datapath *dp;
1547         int skip = cb->args[0];
1548         int i = 0;
1549
1550         list_for_each_entry(dp, &dps, list_node) {
1551                 if (i < skip)
1552                         continue;
1553                 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1554                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1555                                          OVS_DP_CMD_NEW) < 0)
1556                         break;
1557                 i++;
1558         }
1559
1560         cb->args[0] = i;
1561
1562         return skb->len;
1563 }
1564
1565 static struct genl_ops dp_datapath_genl_ops[] = {
1566         { .cmd = OVS_DP_CMD_NEW,
1567           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1568           .policy = datapath_policy,
1569           .doit = ovs_dp_cmd_new
1570         },
1571         { .cmd = OVS_DP_CMD_DEL,
1572           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1573           .policy = datapath_policy,
1574           .doit = ovs_dp_cmd_del
1575         },
1576         { .cmd = OVS_DP_CMD_GET,
1577           .flags = 0,               /* OK for unprivileged users. */
1578           .policy = datapath_policy,
1579           .doit = ovs_dp_cmd_get,
1580           .dumpit = ovs_dp_cmd_dump
1581         },
1582         { .cmd = OVS_DP_CMD_SET,
1583           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1584           .policy = datapath_policy,
1585           .doit = ovs_dp_cmd_set,
1586         },
1587 };
1588
1589 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1590 #ifdef HAVE_NLA_NUL_STRING
1591         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1592         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1593         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1594 #else
1595         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1596         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1597 #endif
1598         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1599         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1600         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1601         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1602 };
1603
1604 static struct genl_family dp_vport_genl_family = {
1605         .id = GENL_ID_GENERATE,
1606         .hdrsize = sizeof(struct ovs_header),
1607         .name = OVS_VPORT_FAMILY,
1608         .version = OVS_VPORT_VERSION,
1609         .maxattr = OVS_VPORT_ATTR_MAX
1610 };
1611
1612 struct genl_multicast_group dp_vport_multicast_group = {
1613         .name = OVS_VPORT_MCGROUP
1614 };
1615
1616 /* Called with RTNL lock or RCU read lock. */
1617 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1618                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1619 {
1620         struct ovs_header *ovs_header;
1621         struct ovs_vport_stats vport_stats;
1622         int err;
1623
1624         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1625                                  flags, cmd);
1626         if (!ovs_header)
1627                 return -EMSGSIZE;
1628
1629         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1630
1631         NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1632         NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type);
1633         NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport));
1634         NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1635
1636         vport_get_stats(vport, &vport_stats);
1637         NLA_PUT(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1638                 &vport_stats);
1639
1640         NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN,
1641                 vport->ops->get_addr(vport));
1642
1643         err = vport_get_options(vport, skb);
1644         if (err == -EMSGSIZE)
1645                 goto error;
1646
1647         return genlmsg_end(skb, ovs_header);
1648
1649 nla_put_failure:
1650         err = -EMSGSIZE;
1651 error:
1652         genlmsg_cancel(skb, ovs_header);
1653         return err;
1654 }
1655
1656 /* Called with RTNL lock or RCU read lock. */
1657 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1658                                          u32 seq, u8 cmd)
1659 {
1660         struct sk_buff *skb;
1661         int retval;
1662
1663         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1664         if (!skb)
1665                 return ERR_PTR(-ENOMEM);
1666
1667         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1668         if (retval < 0) {
1669                 kfree_skb(skb);
1670                 return ERR_PTR(retval);
1671         }
1672         return skb;
1673 }
1674
1675 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1676 {
1677         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1678 }
1679
1680 /* Called with RTNL lock or RCU read lock. */
1681 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1682                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1683 {
1684         struct datapath *dp;
1685         struct vport *vport;
1686
1687         if (a[OVS_VPORT_ATTR_NAME]) {
1688                 vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1689                 if (!vport)
1690                         return ERR_PTR(-ENODEV);
1691                 return vport;
1692         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1693                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1694
1695                 if (port_no >= DP_MAX_PORTS)
1696                         return ERR_PTR(-EFBIG);
1697
1698                 dp = get_dp(ovs_header->dp_ifindex);
1699                 if (!dp)
1700                         return ERR_PTR(-ENODEV);
1701
1702                 vport = get_vport_protected(dp, port_no);
1703                 if (!vport)
1704                         return ERR_PTR(-ENOENT);
1705                 return vport;
1706         } else
1707                 return ERR_PTR(-EINVAL);
1708 }
1709
1710 /* Called with RTNL lock. */
1711 static int change_vport(struct vport *vport,
1712                         struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1713 {
1714         int err = 0;
1715
1716         if (a[OVS_VPORT_ATTR_STATS])
1717                 vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1718
1719         if (a[OVS_VPORT_ATTR_ADDRESS])
1720                 err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1721
1722         return err;
1723 }
1724
1725 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1726 {
1727         struct nlattr **a = info->attrs;
1728         struct ovs_header *ovs_header = info->userhdr;
1729         struct vport_parms parms;
1730         struct sk_buff *reply;
1731         struct vport *vport;
1732         struct datapath *dp;
1733         u32 port_no;
1734         int err;
1735
1736         err = -EINVAL;
1737         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1738             !a[OVS_VPORT_ATTR_UPCALL_PID])
1739                 goto exit;
1740
1741         err = ovs_vport_cmd_validate(a);
1742         if (err)
1743                 goto exit;
1744
1745         rtnl_lock();
1746         dp = get_dp(ovs_header->dp_ifindex);
1747         err = -ENODEV;
1748         if (!dp)
1749                 goto exit_unlock;
1750
1751         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1752                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1753
1754                 err = -EFBIG;
1755                 if (port_no >= DP_MAX_PORTS)
1756                         goto exit_unlock;
1757
1758                 vport = get_vport_protected(dp, port_no);
1759                 err = -EBUSY;
1760                 if (vport)
1761                         goto exit_unlock;
1762         } else {
1763                 for (port_no = 1; ; port_no++) {
1764                         if (port_no >= DP_MAX_PORTS) {
1765                                 err = -EFBIG;
1766                                 goto exit_unlock;
1767                         }
1768                         vport = get_vport_protected(dp, port_no);
1769                         if (!vport)
1770                                 break;
1771                 }
1772         }
1773
1774         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1775         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1776         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1777         parms.dp = dp;
1778         parms.port_no = port_no;
1779         parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1780
1781         vport = new_vport(&parms);
1782         err = PTR_ERR(vport);
1783         if (IS_ERR(vport))
1784                 goto exit_unlock;
1785
1786         dp_sysfs_add_if(vport);
1787
1788         err = change_vport(vport, a);
1789         if (!err) {
1790                 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1791                                                  info->snd_seq,
1792                                                  OVS_VPORT_CMD_NEW);
1793                 if (IS_ERR(reply))
1794                         err = PTR_ERR(reply);
1795         }
1796         if (err) {
1797                 dp_detach_port(vport);
1798                 goto exit_unlock;
1799         }
1800         genl_notify(reply, genl_info_net(info), info->snd_pid,
1801                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1802
1803
1804 exit_unlock:
1805         rtnl_unlock();
1806 exit:
1807         return err;
1808 }
1809
1810 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1811 {
1812         struct nlattr **a = info->attrs;
1813         struct sk_buff *reply;
1814         struct vport *vport;
1815         int err;
1816
1817         err = ovs_vport_cmd_validate(a);
1818         if (err)
1819                 goto exit;
1820
1821         rtnl_lock();
1822         vport = lookup_vport(info->userhdr, a);
1823         err = PTR_ERR(vport);
1824         if (IS_ERR(vport))
1825                 goto exit_unlock;
1826
1827         err = 0;
1828         if (a[OVS_VPORT_ATTR_TYPE] &&
1829             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1830                 err = -EINVAL;
1831
1832         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1833                 err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1834         if (!err)
1835                 err = change_vport(vport, a);
1836         if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1837                 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1838
1839         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1840                                          OVS_VPORT_CMD_NEW);
1841         if (IS_ERR(reply)) {
1842                 err = PTR_ERR(reply);
1843                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1844                                 dp_vport_multicast_group.id, err);
1845                 return 0;
1846         }
1847
1848         genl_notify(reply, genl_info_net(info), info->snd_pid,
1849                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1850
1851 exit_unlock:
1852         rtnl_unlock();
1853 exit:
1854         return err;
1855 }
1856
1857 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1858 {
1859         struct nlattr **a = info->attrs;
1860         struct sk_buff *reply;
1861         struct vport *vport;
1862         int err;
1863
1864         err = ovs_vport_cmd_validate(a);
1865         if (err)
1866                 goto exit;
1867
1868         rtnl_lock();
1869         vport = lookup_vport(info->userhdr, a);
1870         err = PTR_ERR(vport);
1871         if (IS_ERR(vport))
1872                 goto exit_unlock;
1873
1874         if (vport->port_no == OVSP_LOCAL) {
1875                 err = -EINVAL;
1876                 goto exit_unlock;
1877         }
1878
1879         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1880                                          OVS_VPORT_CMD_DEL);
1881         err = PTR_ERR(reply);
1882         if (IS_ERR(reply))
1883                 goto exit_unlock;
1884
1885         dp_detach_port(vport);
1886
1887         genl_notify(reply, genl_info_net(info), info->snd_pid,
1888                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1889
1890 exit_unlock:
1891         rtnl_unlock();
1892 exit:
1893         return err;
1894 }
1895
1896 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1897 {
1898         struct nlattr **a = info->attrs;
1899         struct ovs_header *ovs_header = info->userhdr;
1900         struct sk_buff *reply;
1901         struct vport *vport;
1902         int err;
1903
1904         err = ovs_vport_cmd_validate(a);
1905         if (err)
1906                 goto exit;
1907
1908         rcu_read_lock();
1909         vport = lookup_vport(ovs_header, a);
1910         err = PTR_ERR(vport);
1911         if (IS_ERR(vport))
1912                 goto exit_unlock;
1913
1914         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1915                                          OVS_VPORT_CMD_NEW);
1916         err = PTR_ERR(reply);
1917         if (IS_ERR(reply))
1918                 goto exit_unlock;
1919
1920         rcu_read_unlock();
1921
1922         return genlmsg_reply(reply, info);
1923
1924 exit_unlock:
1925         rcu_read_unlock();
1926 exit:
1927         return err;
1928 }
1929
1930 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1931 {
1932         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1933         struct datapath *dp;
1934         u32 port_no;
1935         int retval;
1936
1937         dp = get_dp(ovs_header->dp_ifindex);
1938         if (!dp)
1939                 return -ENODEV;
1940
1941         rcu_read_lock();
1942         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1943                 struct vport *vport;
1944
1945                 vport = get_vport_protected(dp, port_no);
1946                 if (!vport)
1947                         continue;
1948
1949                 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1950                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1951                                             OVS_VPORT_CMD_NEW) < 0)
1952                         break;
1953         }
1954         rcu_read_unlock();
1955
1956         cb->args[0] = port_no;
1957         retval = skb->len;
1958
1959         return retval;
1960 }
1961
1962 static struct genl_ops dp_vport_genl_ops[] = {
1963         { .cmd = OVS_VPORT_CMD_NEW,
1964           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1965           .policy = vport_policy,
1966           .doit = ovs_vport_cmd_new
1967         },
1968         { .cmd = OVS_VPORT_CMD_DEL,
1969           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1970           .policy = vport_policy,
1971           .doit = ovs_vport_cmd_del
1972         },
1973         { .cmd = OVS_VPORT_CMD_GET,
1974           .flags = 0,               /* OK for unprivileged users. */
1975           .policy = vport_policy,
1976           .doit = ovs_vport_cmd_get,
1977           .dumpit = ovs_vport_cmd_dump
1978         },
1979         { .cmd = OVS_VPORT_CMD_SET,
1980           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1981           .policy = vport_policy,
1982           .doit = ovs_vport_cmd_set,
1983         },
1984 };
1985
1986 struct genl_family_and_ops {
1987         struct genl_family *family;
1988         struct genl_ops *ops;
1989         int n_ops;
1990         struct genl_multicast_group *group;
1991 };
1992
1993 static const struct genl_family_and_ops dp_genl_families[] = {
1994         { &dp_datapath_genl_family,
1995           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1996           &dp_datapath_multicast_group },
1997         { &dp_vport_genl_family,
1998           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1999           &dp_vport_multicast_group },
2000         { &dp_flow_genl_family,
2001           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2002           &dp_flow_multicast_group },
2003         { &dp_packet_genl_family,
2004           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2005           NULL },
2006 };
2007
2008 static void dp_unregister_genl(int n_families)
2009 {
2010         int i;
2011
2012         for (i = 0; i < n_families; i++)
2013                 genl_unregister_family(dp_genl_families[i].family);
2014 }
2015
2016 static int dp_register_genl(void)
2017 {
2018         int n_registered;
2019         int err;
2020         int i;
2021
2022         n_registered = 0;
2023         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2024                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2025
2026                 err = genl_register_family_with_ops(f->family, f->ops,
2027                                                     f->n_ops);
2028                 if (err)
2029                         goto error;
2030                 n_registered++;
2031
2032                 if (f->group) {
2033                         err = genl_register_mc_group(f->family, f->group);
2034                         if (err)
2035                                 goto error;
2036                 }
2037         }
2038
2039         return 0;
2040
2041 error:
2042         dp_unregister_genl(n_registered);
2043         return err;
2044 }
2045
2046 static int __init dp_init(void)
2047 {
2048         struct sk_buff *dummy_skb;
2049         int err;
2050
2051         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2052
2053         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2054                 VERSION BUILDNR);
2055
2056         err = tnl_init();
2057         if (err)
2058                 goto error;
2059
2060         err = flow_init();
2061         if (err)
2062                 goto error_tnl_exit;
2063
2064         err = vport_init();
2065         if (err)
2066                 goto error_flow_exit;
2067
2068         err = register_netdevice_notifier(&dp_device_notifier);
2069         if (err)
2070                 goto error_vport_exit;
2071
2072         err = dp_register_genl();
2073         if (err < 0)
2074                 goto error_unreg_notifier;
2075
2076         return 0;
2077
2078 error_unreg_notifier:
2079         unregister_netdevice_notifier(&dp_device_notifier);
2080 error_vport_exit:
2081         vport_exit();
2082 error_flow_exit:
2083         flow_exit();
2084 error_tnl_exit:
2085         tnl_exit();
2086 error:
2087         return err;
2088 }
2089
2090 static void dp_cleanup(void)
2091 {
2092         rcu_barrier();
2093         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2094         unregister_netdevice_notifier(&dp_device_notifier);
2095         vport_exit();
2096         flow_exit();
2097         tnl_exit();
2098 }
2099
2100 module_init(dp_init);
2101 module_exit(dp_cleanup);
2102
2103 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2104 MODULE_LICENSE("GPL");