datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
   3  * Distributed under the terms of the GNU GPL version 2.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 /* Functions for managing the dp interface/device. */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/init.h>
  14 #include <linux/module.h>
  15 #include <linux/if_arp.h>
  16 #include <linux/if_vlan.h>
  17 #include <linux/in.h>
  18 #include <linux/ip.h>
  19 #include <linux/jhash.h>
  20 #include <linux/delay.h>
  21 #include <linux/time.h>
  22 #include <linux/etherdevice.h>
  23 #include <linux/genetlink.h>
  24 #include <linux/kernel.h>
  25 #include <linux/kthread.h>
  26 #include <linux/mutex.h>
  27 #include <linux/percpu.h>
  28 #include <linux/rcupdate.h>
  29 #include <linux/tcp.h>
  30 #include <linux/udp.h>
  31 #include <linux/version.h>
  32 #include <linux/ethtool.h>
  33 #include <linux/wait.h>
  34 #include <asm/system.h>
  35 #include <asm/div64.h>
  36 #include <linux/highmem.h>
  37 #include <linux/netfilter_bridge.h>
  38 #include <linux/netfilter_ipv4.h>
  39 #include <linux/inetdevice.h>
  40 #include <linux/list.h>
  41 #include <linux/openvswitch.h>
  42 #include <linux/rculist.h>
  43 #include <linux/dmi.h>
  44 #include <net/genetlink.h>
  45
  46 #include "checksum.h"
  47 #include "datapath.h"
  48 #include "flow.h"
  49 #include "vlan.h"
  50 #include "tunnel.h"
  51 #include "vport-internal_dev.h"
  52
  53 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
  54     LINUX_VERSION_CODE > KERNEL_VERSION(3,2,0)
  55 #error Kernels before 2.6.18 or after 3.2 are not supported by this version of Open vSwitch.
  56 #endif
  57
  58 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
  59 EXPORT_SYMBOL(dp_ioctl_hook);
  60
  61 /**
  62  * DOC: Locking:
  63  *
  64  * Writes to device state (add/remove datapath, port, set operations on vports,
  65  * etc.) are protected by RTNL.
  66  *
  67  * Writes to other state (flow table modifications, set miscellaneous datapath
  68  * parameters, etc.) are protected by genl_mutex.  The RTNL lock nests inside
  69  * genl_mutex.
  70  *
  71  * Reads are protected by RCU.
  72  *
  73  * There are a few special cases (mostly stats) that have their own
  74  * synchronization but they nest under all of above and don't interact with
  75  * each other.
  76  */
  77
  78 /* Global list of datapaths to enable dumping them all out.
  79  * Protected by genl_mutex.
  80  */
  81 static LIST_HEAD(dps);
  82
  83 static struct vport *new_vport(const struct vport_parms *);
  84 static int queue_gso_packets(int dp_ifindex, struct sk_buff *,
  85                              const struct dp_upcall_info *);
  86 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *,
  87                                   const struct dp_upcall_info *);
  88
  89 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  90 struct datapath *get_dp(int dp_ifindex)
  91 {
  92         struct datapath *dp = NULL;
  93         struct net_device *dev;
  94
  95         rcu_read_lock();
  96         dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
  97         if (dev) {
  98                 struct vport *vport = internal_dev_get_vport(dev);
  99                 if (vport)
 100                         dp = vport->dp;
 101         }
 102         rcu_read_unlock();
 103
 104         return dp;
 105 }
 106 EXPORT_SYMBOL_GPL(get_dp);
 107
 108 /* Must be called with genl_mutex. */
 109 static struct flow_table *get_table_protected(struct datapath *dp)
 110 {
 111         return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
 112 }
 113
 114 /* Must be called with rcu_read_lock or RTNL lock. */
 115 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
 116 {
 117         return rcu_dereference_rtnl(dp->ports[port_no]);
 118 }
 119
 120 /* Must be called with rcu_read_lock or RTNL lock. */
 121 const char *dp_name(const struct datapath *dp)
 122 {
 123         struct vport *vport = rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]);
 124         return vport->ops->get_name(vport);
 125 }
 126
 127 static int get_dpifindex(struct datapath *dp)
 128 {
 129         struct vport *local;
 130         int ifindex;
 131
 132         rcu_read_lock();
 133
 134         local = get_vport_protected(dp, OVSP_LOCAL);
 135         if (local)
 136                 ifindex = local->ops->get_ifindex(local);
 137         else
 138                 ifindex = 0;
 139
 140         rcu_read_unlock();
 141
 142         return ifindex;
 143 }
 144
 145 static size_t br_nlmsg_size(void)
 146 {
 147         return NLMSG_ALIGN(sizeof(struct ifinfomsg))
 148                + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
 149                + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
 150                + nla_total_size(4) /* IFLA_MASTER */
 151                + nla_total_size(4) /* IFLA_MTU */
 152                + nla_total_size(1); /* IFLA_OPERSTATE */
 153 }
 154
 155 /* Caller must hold RTNL lock. */
 156 static int dp_fill_ifinfo(struct sk_buff *skb,
 157                           const struct vport *port,
 158                           int event, unsigned int flags)
 159 {
 160         struct datapath *dp = port->dp;
 161         struct ifinfomsg *hdr;
 162         struct nlmsghdr *nlh;
 163
 164         if (!port->ops->get_ifindex)
 165                 return -ENODEV;
 166
 167         nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
 168         if (nlh == NULL)
 169                 return -EMSGSIZE;
 170
 171         hdr = nlmsg_data(nlh);
 172         hdr->ifi_family = AF_BRIDGE;
 173         hdr->__ifi_pad = 0;
 174         hdr->ifi_type = ARPHRD_ETHER;
 175         hdr->ifi_index = port->ops->get_ifindex(port);
 176         hdr->ifi_flags = port->ops->get_dev_flags(port);
 177         hdr->ifi_change = 0;
 178
 179         NLA_PUT_STRING(skb, IFLA_IFNAME, port->ops->get_name(port));
 180         NLA_PUT_U32(skb, IFLA_MASTER, get_dpifindex(dp));
 181         NLA_PUT_U32(skb, IFLA_MTU, port->ops->get_mtu(port));
 182 #ifdef IFLA_OPERSTATE
 183         NLA_PUT_U8(skb, IFLA_OPERSTATE,
 184                    port->ops->is_running(port)
 185                         ? port->ops->get_operstate(port)
 186                         : IF_OPER_DOWN);
 187 #endif
 188
 189         NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, port->ops->get_addr(port));
 190
 191         return nlmsg_end(skb, nlh);
 192
 193 nla_put_failure:
 194         nlmsg_cancel(skb, nlh);
 195         return -EMSGSIZE;
 196 }
 197
 198 /* Caller must hold RTNL lock. */
 199 static void dp_ifinfo_notify(int event, struct vport *port)
 200 {
 201         struct sk_buff *skb;
 202         int err;
 203
 204         skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
 205         if (!skb) {
 206                 err = -ENOBUFS;
 207                 goto err;
 208         }
 209
 210         err = dp_fill_ifinfo(skb, port, event, 0);
 211         if (err < 0) {
 212                 if (err == -ENODEV) {
 213                         goto out;
 214                 } else {
 215                         /* -EMSGSIZE implies BUG in br_nlmsg_size() */
 216                         WARN_ON(err == -EMSGSIZE);
 217                         goto err;
 218                 }
 219         }
 220
 221         rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 222
 223         return;
 224 err:
 225         rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
 226 out:
 227         kfree_skb(skb);
 228 }
 229
 230 static void release_dp(struct kobject *kobj)
 231 {
 232         struct datapath *dp = container_of(kobj, struct datapath, ifobj);
 233         kfree(dp);
 234 }
 235
 236 static struct kobj_type dp_ktype = {
 237         .release = release_dp
 238 };
 239
 240 static void destroy_dp_rcu(struct rcu_head *rcu)
 241 {
 242         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 243
 244         flow_tbl_destroy(dp->table);
 245         free_percpu(dp->stats_percpu);
 246         kobject_put(&dp->ifobj);
 247 }
 248
 249 /* Called with RTNL lock and genl_lock. */
 250 static struct vport *new_vport(const struct vport_parms *parms)
 251 {
 252         struct vport *vport;
 253
 254         vport = vport_add(parms);
 255         if (!IS_ERR(vport)) {
 256                 struct datapath *dp = parms->dp;
 257
 258                 rcu_assign_pointer(dp->ports[parms->port_no], vport);
 259                 list_add(&vport->node, &dp->port_list);
 260
 261                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 262         }
 263
 264         return vport;
 265 }
 266
 267 /* Called with RTNL lock. */
 268 void dp_detach_port(struct vport *p)
 269 {
 270         ASSERT_RTNL();
 271
 272         if (p->port_no != OVSP_LOCAL)
 273                 dp_sysfs_del_if(p);
 274         dp_ifinfo_notify(RTM_DELLINK, p);
 275
 276         /* First drop references to device. */
 277         list_del(&p->node);
 278         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 279
 280         /* Then destroy it. */
 281         vport_del(p);
 282 }
 283
 284 /* Must be called with rcu_read_lock. */
 285 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 286 {
 287         struct datapath *dp = p->dp;
 288         struct sw_flow *flow;
 289         struct dp_stats_percpu *stats;
 290         u64 *stats_counter;
 291         int error;
 292
 293         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 294
 295         if (!OVS_CB(skb)->flow) {
 296                 struct sw_flow_key key;
 297                 int key_len;
 298
 299                 /* Extract flow from 'skb' into 'key'. */
 300                 error = flow_extract(skb, p->port_no, &key, &key_len);
 301                 if (unlikely(error)) {
 302                         kfree_skb(skb);
 303                         return;
 304                 }
 305
 306                 /* Look up flow. */
 307                 flow = flow_tbl_lookup(rcu_dereference(dp->table),
 308                                        &key, key_len);
 309                 if (unlikely(!flow)) {
 310                         struct dp_upcall_info upcall;
 311
 312                         upcall.cmd = OVS_PACKET_CMD_MISS;
 313                         upcall.key = &key;
 314                         upcall.userdata = NULL;
 315                         upcall.pid = p->upcall_pid;
 316                         dp_upcall(dp, skb, &upcall);
 317                         consume_skb(skb);
 318                         stats_counter = &stats->n_missed;
 319                         goto out;
 320                 }
 321
 322                 OVS_CB(skb)->flow = flow;
 323         }
 324
 325         stats_counter = &stats->n_hit;
 326         flow_used(OVS_CB(skb)->flow, skb);
 327         execute_actions(dp, skb);
 328
 329 out:
 330         /* Update datapath statistics. */
 331
 332         write_seqcount_begin(&stats->seqlock);
 333         (*stats_counter)++;
 334         write_seqcount_end(&stats->seqlock);
 335 }
 336
 337 static struct genl_family dp_packet_genl_family = {
 338         .id = GENL_ID_GENERATE,
 339         .hdrsize = sizeof(struct ovs_header),
 340         .name = OVS_PACKET_FAMILY,
 341         .version = OVS_PACKET_VERSION,
 342         .maxattr = OVS_PACKET_ATTR_MAX
 343 };
 344
 345 int dp_upcall(struct datapath *dp, struct sk_buff *skb,
 346               const struct dp_upcall_info *upcall_info)
 347 {
 348         struct dp_stats_percpu *stats;
 349         int dp_ifindex;
 350         int err;
 351
 352         if (upcall_info->pid == 0) {
 353                 err = -ENOTCONN;
 354                 goto err;
 355         }
 356
 357         dp_ifindex = get_dpifindex(dp);
 358         if (!dp_ifindex) {
 359                 err = -ENODEV;
 360                 goto err;
 361         }
 362
 363         forward_ip_summed(skb, true);
 364
 365         if (!skb_is_gso(skb))
 366                 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
 367         else
 368                 err = queue_gso_packets(dp_ifindex, skb, upcall_info);
 369         if (err)
 370                 goto err;
 371
 372         return 0;
 373
 374 err:
 375         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 376
 377         write_seqcount_begin(&stats->seqlock);
 378         stats->n_lost++;
 379         write_seqcount_end(&stats->seqlock);
 380
 381         return err;
 382 }
 383
 384 static int queue_gso_packets(int dp_ifindex, struct sk_buff *skb,
 385                              const struct dp_upcall_info *upcall_info)
 386 {
 387         struct dp_upcall_info later_info;
 388         struct sw_flow_key later_key;
 389         struct sk_buff *segs, *nskb;
 390         int err;
 391
 392         segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 393         if (IS_ERR(skb))
 394                 return PTR_ERR(skb);
 395
 396         /* Queue all of the segments. */
 397         skb = segs;
 398         do {
 399                 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
 400                 if (err)
 401                         break;
 402
 403                 if (skb == segs && skb_shinfo(skb)->gso_type & SKB_GSO_UDP) {
 404                         /* The initial flow key extracted by flow_extract() in
 405                          * this case is for a first fragment, so we need to
 406                          * properly mark later fragments.
 407                          */
 408                         later_key = *upcall_info->key;
 409                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 410
 411                         later_info = *upcall_info;
 412                         later_info.key = &later_key;
 413                         upcall_info = &later_info;
 414                 }
 415         } while ((skb = skb->next));
 416
 417         /* Free all of the segments. */
 418         skb = segs;
 419         do {
 420                 nskb = skb->next;
 421                 if (err)
 422                         kfree_skb(skb);
 423                 else
 424                         consume_skb(skb);
 425         } while ((skb = nskb));
 426         return err;
 427 }
 428
 429 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *skb,
 430                                   const struct dp_upcall_info *upcall_info)
 431 {
 432         struct ovs_header *upcall;
 433         struct sk_buff *user_skb; /* to be queued to userspace */
 434         struct nlattr *nla;
 435         unsigned int len;
 436         int err;
 437
 438         err = vlan_deaccel_tag(skb);
 439         if (unlikely(err))
 440                 return err;
 441
 442         if (nla_attr_size(skb->len) > USHRT_MAX)
 443                 return -EFBIG;
 444
 445         len = sizeof(struct ovs_header);
 446         len += nla_total_size(skb->len);
 447         len += nla_total_size(FLOW_BUFSIZE);
 448         if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
 449                 len += nla_total_size(8);
 450
 451         user_skb = genlmsg_new(len, GFP_ATOMIC);
 452         if (!user_skb)
 453                 return -ENOMEM;
 454
 455         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 456                              0, upcall_info->cmd);
 457         upcall->dp_ifindex = dp_ifindex;
 458
 459         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 460         flow_to_nlattrs(upcall_info->key, user_skb);
 461         nla_nest_end(user_skb, nla);
 462
 463         if (upcall_info->userdata)
 464                 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
 465                             nla_get_u64(upcall_info->userdata));
 466
 467         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 468
 469         skb_copy_and_csum_dev(skb, nla_data(nla));
 470
 471         return genlmsg_unicast(&init_net, user_skb, upcall_info->pid);
 472 }
 473
 474 /* Called with genl_mutex. */
 475 static int flush_flows(int dp_ifindex)
 476 {
 477         struct flow_table *old_table;
 478         struct flow_table *new_table;
 479         struct datapath *dp;
 480
 481         dp = get_dp(dp_ifindex);
 482         if (!dp)
 483                 return -ENODEV;
 484
 485         old_table = get_table_protected(dp);
 486         new_table = flow_tbl_alloc(TBL_MIN_BUCKETS);
 487         if (!new_table)
 488                 return -ENOMEM;
 489
 490         rcu_assign_pointer(dp->table, new_table);
 491
 492         flow_tbl_deferred_destroy(old_table);
 493         return 0;
 494 }
 495
 496 static int validate_actions(const struct nlattr *attr,
 497                                 const struct sw_flow_key *key, int depth);
 498
 499 static int validate_sample(const struct nlattr *attr,
 500                                 const struct sw_flow_key *key, int depth)
 501 {
 502         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 503         const struct nlattr *probability, *actions;
 504         const struct nlattr *a;
 505         int rem;
 506
 507         memset(attrs, 0, sizeof(attrs));
 508         nla_for_each_nested(a, attr, rem) {
 509                 int type = nla_type(a);
 510                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 511                         return -EINVAL;
 512                 attrs[type] = a;
 513         }
 514         if (rem)
 515                 return -EINVAL;
 516
 517         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 518         if (!probability || nla_len(probability) != sizeof(u32))
 519                 return -EINVAL;
 520
 521         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 522         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 523                 return -EINVAL;
 524         return validate_actions(actions, key, depth + 1);
 525 }
 526
 527 static int validate_set(const struct nlattr *a,
 528                         const struct sw_flow_key *flow_key)
 529 {
 530         const struct nlattr *ovs_key = nla_data(a);
 531         int key_type = nla_type(ovs_key);
 532
 533         /* There can be only one key in a action */
 534         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 535                 return -EINVAL;
 536
 537         if (key_type > OVS_KEY_ATTR_MAX ||
 538             nla_len(ovs_key) != ovs_key_lens[key_type])
 539                 return -EINVAL;
 540
 541         switch (key_type) {
 542         const struct ovs_key_ipv4 *ipv4_key;
 543
 544         case OVS_KEY_ATTR_PRIORITY:
 545         case OVS_KEY_ATTR_TUN_ID:
 546         case OVS_KEY_ATTR_ETHERNET:
 547                 break;
 548
 549         case OVS_KEY_ATTR_IPV4:
 550                 if (flow_key->eth.type != htons(ETH_P_IP))
 551                         return -EINVAL;
 552
 553                 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
 554                         return -EINVAL;
 555
 556                 ipv4_key = nla_data(ovs_key);
 557                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 558                         return -EINVAL;
 559
 560                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 561                         return -EINVAL;
 562
 563                 break;
 564
 565         case OVS_KEY_ATTR_TCP:
 566                 if (flow_key->ip.proto != IPPROTO_TCP)
 567                         return -EINVAL;
 568
 569                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 570                         return -EINVAL;
 571
 572                 break;
 573
 574         case OVS_KEY_ATTR_UDP:
 575                 if (flow_key->ip.proto != IPPROTO_UDP)
 576                         return -EINVAL;
 577
 578                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 579                         return -EINVAL;
 580                 break;
 581
 582         default:
 583                 return -EINVAL;
 584         }
 585
 586         return 0;
 587 }
 588
 589 static int validate_userspace(const struct nlattr *attr)
 590 {
 591         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 592                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 593                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
 594         };
 595         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 596         int error;
 597
 598         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 599                                  attr, userspace_policy);
 600         if (error)
 601                 return error;
 602
 603         if (!a[OVS_USERSPACE_ATTR_PID] ||
 604             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 605                 return -EINVAL;
 606
 607         return 0;
 608 }
 609
 610 static int validate_actions(const struct nlattr *attr,
 611                                 const struct sw_flow_key *key,  int depth)
 612 {
 613         const struct nlattr *a;
 614         int rem, err;
 615
 616         if (depth >= SAMPLE_ACTION_DEPTH)
 617                 return -EOVERFLOW;
 618
 619         nla_for_each_nested(a, attr, rem) {
 620                 /* Expected argument lengths, (u32)-1 for variable length. */
 621                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 622                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 623                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 624                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 625                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 626                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 627                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 628                 };
 629                 const struct ovs_action_push_vlan *vlan;
 630                 int type = nla_type(a);
 631
 632                 if (type > OVS_ACTION_ATTR_MAX ||
 633                     (action_lens[type] != nla_len(a) &&
 634                      action_lens[type] != (u32)-1))
 635                         return -EINVAL;
 636
 637                 switch (type) {
 638                 case OVS_ACTION_ATTR_UNSPEC:
 639                         return -EINVAL;
 640
 641                 case OVS_ACTION_ATTR_USERSPACE:
 642                         err = validate_userspace(a);
 643                         if (err)
 644                                 return err;
 645                         break;
 646
 647                 case OVS_ACTION_ATTR_OUTPUT:
 648                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 649                                 return -EINVAL;
 650                         break;
 651
 652
 653                 case OVS_ACTION_ATTR_POP_VLAN:
 654                         break;
 655
 656                 case OVS_ACTION_ATTR_PUSH_VLAN:
 657                         vlan = nla_data(a);
 658                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 659                                 return -EINVAL;
 660                         if (vlan->vlan_tci & htons(VLAN_TAG_PRESENT))
 661                                 return -EINVAL;
 662                         break;
 663
 664                 case OVS_ACTION_ATTR_SET:
 665                         err = validate_set(a, key);
 666                         if (err)
 667                                 return err;
 668                         break;
 669
 670                 case OVS_ACTION_ATTR_SAMPLE:
 671                         err = validate_sample(a, key, depth);
 672                         if (err)
 673                                 return err;
 674                         break;
 675
 676                 default:
 677                         return -EINVAL;
 678                 }
 679         }
 680
 681         if (rem > 0)
 682                 return -EINVAL;
 683
 684         return 0;
 685 }
 686
 687 static void clear_stats(struct sw_flow *flow)
 688 {
 689         flow->used = 0;
 690         flow->tcp_flags = 0;
 691         flow->packet_count = 0;
 692         flow->byte_count = 0;
 693 }
 694
 695 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 696 {
 697         struct ovs_header *ovs_header = info->userhdr;
 698         struct nlattr **a = info->attrs;
 699         struct sw_flow_actions *acts;
 700         struct sk_buff *packet;
 701         struct sw_flow *flow;
 702         struct datapath *dp;
 703         struct ethhdr *eth;
 704         int len;
 705         int err;
 706         int key_len;
 707
 708         err = -EINVAL;
 709         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 710             !a[OVS_PACKET_ATTR_ACTIONS] ||
 711             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 712                 goto err;
 713
 714         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 715         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 716         err = -ENOMEM;
 717         if (!packet)
 718                 goto err;
 719         skb_reserve(packet, NET_IP_ALIGN);
 720
 721         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 722
 723         skb_reset_mac_header(packet);
 724         eth = eth_hdr(packet);
 725
 726         /* Normally, setting the skb 'protocol' field would be handled by a
 727          * call to eth_type_trans(), but it assumes there's a sending
 728          * device, which we may not have. */
 729         if (ntohs(eth->h_proto) >= 1536)
 730                 packet->protocol = eth->h_proto;
 731         else
 732                 packet->protocol = htons(ETH_P_802_2);
 733
 734         /* Build an sw_flow for sending this packet. */
 735         flow = flow_alloc();
 736         err = PTR_ERR(flow);
 737         if (IS_ERR(flow))
 738                 goto err_kfree_skb;
 739
 740         err = flow_extract(packet, -1, &flow->key, &key_len);
 741         if (err)
 742                 goto err_flow_put;
 743
 744         err = flow_metadata_from_nlattrs(&flow->key.phy.priority,
 745                                          &flow->key.phy.in_port,
 746                                          &flow->key.phy.tun_id,
 747                                          a[OVS_PACKET_ATTR_KEY]);
 748         if (err)
 749                 goto err_flow_put;
 750
 751         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
 752         if (err)
 753                 goto err_flow_put;
 754
 755         flow->hash = flow_hash(&flow->key, key_len);
 756
 757         acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 758         err = PTR_ERR(acts);
 759         if (IS_ERR(acts))
 760                 goto err_flow_put;
 761         rcu_assign_pointer(flow->sf_acts, acts);
 762
 763         OVS_CB(packet)->flow = flow;
 764         packet->priority = flow->key.phy.priority;
 765
 766         rcu_read_lock();
 767         dp = get_dp(ovs_header->dp_ifindex);
 768         err = -ENODEV;
 769         if (!dp)
 770                 goto err_unlock;
 771
 772         local_bh_disable();
 773         err = execute_actions(dp, packet);
 774         local_bh_enable();
 775         rcu_read_unlock();
 776
 777         flow_put(flow);
 778         return err;
 779
 780 err_unlock:
 781         rcu_read_unlock();
 782 err_flow_put:
 783         flow_put(flow);
 784 err_kfree_skb:
 785         kfree_skb(packet);
 786 err:
 787         return err;
 788 }
 789
 790 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 791         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 792         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 793         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 794 };
 795
 796 static struct genl_ops dp_packet_genl_ops[] = {
 797         { .cmd = OVS_PACKET_CMD_EXECUTE,
 798           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 799           .policy = packet_policy,
 800           .doit = ovs_packet_cmd_execute
 801         }
 802 };
 803
 804 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 805 {
 806         int i;
 807         struct flow_table *table = get_table_protected(dp);
 808
 809         stats->n_flows = flow_tbl_count(table);
 810
 811         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 812         for_each_possible_cpu(i) {
 813                 const struct dp_stats_percpu *percpu_stats;
 814                 struct dp_stats_percpu local_stats;
 815                 unsigned seqcount;
 816
 817                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 818
 819                 do {
 820                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 821                         local_stats = *percpu_stats;
 822                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 823
 824                 stats->n_hit += local_stats.n_hit;
 825                 stats->n_missed += local_stats.n_missed;
 826                 stats->n_lost += local_stats.n_lost;
 827         }
 828 }
 829
 830 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 831         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 832         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 833         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 834 };
 835
 836 static struct genl_family dp_flow_genl_family = {
 837         .id = GENL_ID_GENERATE,
 838         .hdrsize = sizeof(struct ovs_header),
 839         .name = OVS_FLOW_FAMILY,
 840         .version = OVS_FLOW_VERSION,
 841         .maxattr = OVS_FLOW_ATTR_MAX
 842 };
 843
 844 static struct genl_multicast_group dp_flow_multicast_group = {
 845         .name = OVS_FLOW_MCGROUP
 846 };
 847
 848 /* Called with genl_lock. */
 849 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 850                                   struct sk_buff *skb, u32 pid,
 851                                   u32 seq, u32 flags, u8 cmd)
 852 {
 853         const int skb_orig_len = skb->len;
 854         const struct sw_flow_actions *sf_acts;
 855         struct ovs_flow_stats stats;
 856         struct ovs_header *ovs_header;
 857         struct nlattr *nla;
 858         unsigned long used;
 859         u8 tcp_flags;
 860         int err;
 861
 862         sf_acts = rcu_dereference_protected(flow->sf_acts,
 863                                             lockdep_genl_is_held());
 864
 865         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 866         if (!ovs_header)
 867                 return -EMSGSIZE;
 868
 869         ovs_header->dp_ifindex = get_dpifindex(dp);
 870
 871         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 872         if (!nla)
 873                 goto nla_put_failure;
 874         err = flow_to_nlattrs(&flow->key, skb);
 875         if (err)
 876                 goto error;
 877         nla_nest_end(skb, nla);
 878
 879         spin_lock_bh(&flow->lock);
 880         used = flow->used;
 881         stats.n_packets = flow->packet_count;
 882         stats.n_bytes = flow->byte_count;
 883         tcp_flags = flow->tcp_flags;
 884         spin_unlock_bh(&flow->lock);
 885
 886         if (used)
 887                 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
 888
 889         if (stats.n_packets)
 890                 NLA_PUT(skb, OVS_FLOW_ATTR_STATS,
 891                         sizeof(struct ovs_flow_stats), &stats);
 892
 893         if (tcp_flags)
 894                 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 895
 896         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 897          * this is the first flow to be dumped into 'skb'.  This is unusual for
 898          * Netlink but individual action lists can be longer than
 899          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 900          * The userspace caller can always fetch the actions separately if it
 901          * really wants them.  (Most userspace callers in fact don't care.)
 902          *
 903          * This can only fail for dump operations because the skb is always
 904          * properly sized for single flows.
 905          */
 906         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 907                       sf_acts->actions);
 908         if (err < 0 && skb_orig_len)
 909                 goto error;
 910
 911         return genlmsg_end(skb, ovs_header);
 912
 913 nla_put_failure:
 914         err = -EMSGSIZE;
 915 error:
 916         genlmsg_cancel(skb, ovs_header);
 917         return err;
 918 }
 919
 920 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 921 {
 922         const struct sw_flow_actions *sf_acts;
 923         int len;
 924
 925         sf_acts = rcu_dereference_protected(flow->sf_acts,
 926                                             lockdep_genl_is_held());
 927
 928         /* OVS_FLOW_ATTR_KEY */
 929         len = nla_total_size(FLOW_BUFSIZE);
 930         /* OVS_FLOW_ATTR_ACTIONS */
 931         len += nla_total_size(sf_acts->actions_len);
 932         /* OVS_FLOW_ATTR_STATS */
 933         len += nla_total_size(sizeof(struct ovs_flow_stats));
 934         /* OVS_FLOW_ATTR_TCP_FLAGS */
 935         len += nla_total_size(1);
 936         /* OVS_FLOW_ATTR_USED */
 937         len += nla_total_size(8);
 938
 939         len += NLMSG_ALIGN(sizeof(struct ovs_header));
 940
 941         return genlmsg_new(len, GFP_KERNEL);
 942 }
 943
 944 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 945                                                struct datapath *dp,
 946                                                u32 pid, u32 seq, u8 cmd)
 947 {
 948         struct sk_buff *skb;
 949         int retval;
 950
 951         skb = ovs_flow_cmd_alloc_info(flow);
 952         if (!skb)
 953                 return ERR_PTR(-ENOMEM);
 954
 955         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 956         BUG_ON(retval < 0);
 957         return skb;
 958 }
 959
 960 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 961 {
 962         struct nlattr **a = info->attrs;
 963         struct ovs_header *ovs_header = info->userhdr;
 964         struct sw_flow_key key;
 965         struct sw_flow *flow;
 966         struct sk_buff *reply;
 967         struct datapath *dp;
 968         struct flow_table *table;
 969         int error;
 970         int key_len;
 971
 972         /* Extract key. */
 973         error = -EINVAL;
 974         if (!a[OVS_FLOW_ATTR_KEY])
 975                 goto error;
 976         error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
 977         if (error)
 978                 goto error;
 979
 980         /* Validate actions. */
 981         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 982                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
 983                 if (error)
 984                         goto error;
 985         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 986                 error = -EINVAL;
 987                 goto error;
 988         }
 989
 990         dp = get_dp(ovs_header->dp_ifindex);
 991         error = -ENODEV;
 992         if (!dp)
 993                 goto error;
 994
 995         table = get_table_protected(dp);
 996         flow = flow_tbl_lookup(table, &key, key_len);
 997         if (!flow) {
 998                 struct sw_flow_actions *acts;
 999
1000                 /* Bail out if we're not allowed to create a new flow. */
1001                 error = -ENOENT;
1002                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1003                         goto error;
1004
1005                 /* Expand table, if necessary, to make room. */
1006                 if (flow_tbl_need_to_expand(table)) {
1007                         struct flow_table *new_table;
1008
1009                         new_table = flow_tbl_expand(table);
1010                         if (!IS_ERR(new_table)) {
1011                                 rcu_assign_pointer(dp->table, new_table);
1012                                 flow_tbl_deferred_destroy(table);
1013                                 table = get_table_protected(dp);
1014                         }
1015                 }
1016
1017                 /* Allocate flow. */
1018                 flow = flow_alloc();
1019                 if (IS_ERR(flow)) {
1020                         error = PTR_ERR(flow);
1021                         goto error;
1022                 }
1023                 flow->key = key;
1024                 clear_stats(flow);
1025
1026                 /* Obtain actions. */
1027                 acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1028                 error = PTR_ERR(acts);
1029                 if (IS_ERR(acts))
1030                         goto error_free_flow;
1031                 rcu_assign_pointer(flow->sf_acts, acts);
1032
1033                 /* Put flow in bucket. */
1034                 flow->hash = flow_hash(&key, key_len);
1035                 flow_tbl_insert(table, flow);
1036
1037                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1038                                                 info->snd_seq,
1039                                                 OVS_FLOW_CMD_NEW);
1040         } else {
1041                 /* We found a matching flow. */
1042                 struct sw_flow_actions *old_acts;
1043                 struct nlattr *acts_attrs;
1044
1045                 /* Bail out if we're not allowed to modify an existing flow.
1046                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1047                  * because Generic Netlink treats the latter as a dump
1048                  * request.  We also accept NLM_F_EXCL in case that bug ever
1049                  * gets fixed.
1050                  */
1051                 error = -EEXIST;
1052                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1053                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1054                         goto error;
1055
1056                 /* Update actions. */
1057                 old_acts = rcu_dereference_protected(flow->sf_acts,
1058                                                      lockdep_genl_is_held());
1059                 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1060                 if (acts_attrs &&
1061                    (old_acts->actions_len != nla_len(acts_attrs) ||
1062                    memcmp(old_acts->actions, nla_data(acts_attrs),
1063                           old_acts->actions_len))) {
1064                         struct sw_flow_actions *new_acts;
1065
1066                         new_acts = flow_actions_alloc(acts_attrs);
1067                         error = PTR_ERR(new_acts);
1068                         if (IS_ERR(new_acts))
1069                                 goto error;
1070
1071                         rcu_assign_pointer(flow->sf_acts, new_acts);
1072                         flow_deferred_free_acts(old_acts);
1073                 }
1074
1075                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1076                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1077
1078                 /* Clear stats. */
1079                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1080                         spin_lock_bh(&flow->lock);
1081                         clear_stats(flow);
1082                         spin_unlock_bh(&flow->lock);
1083                 }
1084         }
1085
1086         if (!IS_ERR(reply))
1087                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1088                            dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1089         else
1090                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1091                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1092         return 0;
1093
1094 error_free_flow:
1095         flow_put(flow);
1096 error:
1097         return error;
1098 }
1099
1100 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1101 {
1102         struct nlattr **a = info->attrs;
1103         struct ovs_header *ovs_header = info->userhdr;
1104         struct sw_flow_key key;
1105         struct sk_buff *reply;
1106         struct sw_flow *flow;
1107         struct datapath *dp;
1108         struct flow_table *table;
1109         int err;
1110         int key_len;
1111
1112         if (!a[OVS_FLOW_ATTR_KEY])
1113                 return -EINVAL;
1114         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1115         if (err)
1116                 return err;
1117
1118         dp = get_dp(ovs_header->dp_ifindex);
1119         if (!dp)
1120                 return -ENODEV;
1121
1122         table = get_table_protected(dp);
1123         flow = flow_tbl_lookup(table, &key, key_len);
1124         if (!flow)
1125                 return -ENOENT;
1126
1127         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1128                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1129         if (IS_ERR(reply))
1130                 return PTR_ERR(reply);
1131
1132         return genlmsg_reply(reply, info);
1133 }
1134
1135 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1136 {
1137         struct nlattr **a = info->attrs;
1138         struct ovs_header *ovs_header = info->userhdr;
1139         struct sw_flow_key key;
1140         struct sk_buff *reply;
1141         struct sw_flow *flow;
1142         struct datapath *dp;
1143         struct flow_table *table;
1144         int err;
1145         int key_len;
1146
1147         if (!a[OVS_FLOW_ATTR_KEY])
1148                 return flush_flows(ovs_header->dp_ifindex);
1149         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1150         if (err)
1151                 return err;
1152
1153         dp = get_dp(ovs_header->dp_ifindex);
1154         if (!dp)
1155                 return -ENODEV;
1156
1157         table = get_table_protected(dp);
1158         flow = flow_tbl_lookup(table, &key, key_len);
1159         if (!flow)
1160                 return -ENOENT;
1161
1162         reply = ovs_flow_cmd_alloc_info(flow);
1163         if (!reply)
1164                 return -ENOMEM;
1165
1166         flow_tbl_remove(table, flow);
1167
1168         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1169                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1170         BUG_ON(err < 0);
1171
1172         flow_deferred_free(flow);
1173
1174         genl_notify(reply, genl_info_net(info), info->snd_pid,
1175                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1176         return 0;
1177 }
1178
1179 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1180 {
1181         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1182         struct datapath *dp;
1183
1184         dp = get_dp(ovs_header->dp_ifindex);
1185         if (!dp)
1186                 return -ENODEV;
1187
1188         for (;;) {
1189                 struct sw_flow *flow;
1190                 u32 bucket, obj;
1191
1192                 bucket = cb->args[0];
1193                 obj = cb->args[1];
1194                 flow = flow_tbl_next(get_table_protected(dp), &bucket, &obj);
1195                 if (!flow)
1196                         break;
1197
1198                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1199                                            NETLINK_CB(cb->skb).pid,
1200                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1201                                            OVS_FLOW_CMD_NEW) < 0)
1202                         break;
1203
1204                 cb->args[0] = bucket;
1205                 cb->args[1] = obj;
1206         }
1207         return skb->len;
1208 }
1209
1210 static struct genl_ops dp_flow_genl_ops[] = {
1211         { .cmd = OVS_FLOW_CMD_NEW,
1212           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1213           .policy = flow_policy,
1214           .doit = ovs_flow_cmd_new_or_set
1215         },
1216         { .cmd = OVS_FLOW_CMD_DEL,
1217           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1218           .policy = flow_policy,
1219           .doit = ovs_flow_cmd_del
1220         },
1221         { .cmd = OVS_FLOW_CMD_GET,
1222           .flags = 0,               /* OK for unprivileged users. */
1223           .policy = flow_policy,
1224           .doit = ovs_flow_cmd_get,
1225           .dumpit = ovs_flow_cmd_dump
1226         },
1227         { .cmd = OVS_FLOW_CMD_SET,
1228           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1229           .policy = flow_policy,
1230           .doit = ovs_flow_cmd_new_or_set,
1231         },
1232 };
1233
1234 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1235 #ifdef HAVE_NLA_NUL_STRING
1236         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1237 #endif
1238         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1239 };
1240
1241 static struct genl_family dp_datapath_genl_family = {
1242         .id = GENL_ID_GENERATE,
1243         .hdrsize = sizeof(struct ovs_header),
1244         .name = OVS_DATAPATH_FAMILY,
1245         .version = OVS_DATAPATH_VERSION,
1246         .maxattr = OVS_DP_ATTR_MAX
1247 };
1248
1249 static struct genl_multicast_group dp_datapath_multicast_group = {
1250         .name = OVS_DATAPATH_MCGROUP
1251 };
1252
1253 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1254                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1255 {
1256         struct ovs_header *ovs_header;
1257         struct nlattr *nla;
1258         int err;
1259
1260         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1261                                    flags, cmd);
1262         if (!ovs_header)
1263                 goto error;
1264
1265         ovs_header->dp_ifindex = get_dpifindex(dp);
1266
1267         rcu_read_lock();
1268         err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
1269         rcu_read_unlock();
1270         if (err)
1271                 goto nla_put_failure;
1272
1273         nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats));
1274         if (!nla)
1275                 goto nla_put_failure;
1276         get_dp_stats(dp, nla_data(nla));
1277
1278         return genlmsg_end(skb, ovs_header);
1279
1280 nla_put_failure:
1281         genlmsg_cancel(skb, ovs_header);
1282 error:
1283         return -EMSGSIZE;
1284 }
1285
1286 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1287                                              u32 seq, u8 cmd)
1288 {
1289         struct sk_buff *skb;
1290         int retval;
1291
1292         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1293         if (!skb)
1294                 return ERR_PTR(-ENOMEM);
1295
1296         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1297         if (retval < 0) {
1298                 kfree_skb(skb);
1299                 return ERR_PTR(retval);
1300         }
1301         return skb;
1302 }
1303
1304 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1305 {
1306         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1307 }
1308
1309 /* Called with genl_mutex and optionally with RTNL lock also. */
1310 static struct datapath *lookup_datapath(struct ovs_header *ovs_header,
1311                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1312 {
1313         struct datapath *dp;
1314
1315         if (!a[OVS_DP_ATTR_NAME])
1316                 dp = get_dp(ovs_header->dp_ifindex);
1317         else {
1318                 struct vport *vport;
1319
1320                 rcu_read_lock();
1321                 vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1322                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1323                 rcu_read_unlock();
1324         }
1325         return dp ? dp : ERR_PTR(-ENODEV);
1326 }
1327
1328 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1329 {
1330         struct nlattr **a = info->attrs;
1331         struct vport_parms parms;
1332         struct sk_buff *reply;
1333         struct datapath *dp;
1334         struct vport *vport;
1335         int err;
1336
1337         err = -EINVAL;
1338         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1339                 goto err;
1340
1341         err = ovs_dp_cmd_validate(a);
1342         if (err)
1343                 goto err;
1344
1345         rtnl_lock();
1346         err = -ENODEV;
1347         if (!try_module_get(THIS_MODULE))
1348                 goto err_unlock_rtnl;
1349
1350         err = -ENOMEM;
1351         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1352         if (dp == NULL)
1353                 goto err_put_module;
1354         INIT_LIST_HEAD(&dp->port_list);
1355
1356         /* Initialize kobject for bridge.  This will be added as
1357          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1358         dp->ifobj.kset = NULL;
1359         kobject_init(&dp->ifobj, &dp_ktype);
1360
1361         /* Allocate table. */
1362         err = -ENOMEM;
1363         rcu_assign_pointer(dp->table, flow_tbl_alloc(TBL_MIN_BUCKETS));
1364         if (!dp->table)
1365                 goto err_free_dp;
1366
1367         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1368         if (!dp->stats_percpu) {
1369                 err = -ENOMEM;
1370                 goto err_destroy_table;
1371         }
1372
1373         /* Set up our datapath device. */
1374         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1375         parms.type = OVS_VPORT_TYPE_INTERNAL;
1376         parms.options = NULL;
1377         parms.dp = dp;
1378         parms.port_no = OVSP_LOCAL;
1379         parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1380
1381         vport = new_vport(&parms);
1382         if (IS_ERR(vport)) {
1383                 err = PTR_ERR(vport);
1384                 if (err == -EBUSY)
1385                         err = -EEXIST;
1386
1387                 goto err_destroy_percpu;
1388         }
1389
1390         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1391                                       info->snd_seq, OVS_DP_CMD_NEW);
1392         err = PTR_ERR(reply);
1393         if (IS_ERR(reply))
1394                 goto err_destroy_local_port;
1395
1396         list_add_tail(&dp->list_node, &dps);
1397         dp_sysfs_add_dp(dp);
1398
1399         rtnl_unlock();
1400
1401         genl_notify(reply, genl_info_net(info), info->snd_pid,
1402                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1403         return 0;
1404
1405 err_destroy_local_port:
1406         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1407 err_destroy_percpu:
1408         free_percpu(dp->stats_percpu);
1409 err_destroy_table:
1410         flow_tbl_destroy(get_table_protected(dp));
1411 err_free_dp:
1412         kfree(dp);
1413 err_put_module:
1414         module_put(THIS_MODULE);
1415 err_unlock_rtnl:
1416         rtnl_unlock();
1417 err:
1418         return err;
1419 }
1420
1421 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1422 {
1423         struct vport *vport, *next_vport;
1424         struct sk_buff *reply;
1425         struct datapath *dp;
1426         int err;
1427
1428         err = ovs_dp_cmd_validate(info->attrs);
1429         if (err)
1430                 goto exit;
1431
1432         rtnl_lock();
1433         dp = lookup_datapath(info->userhdr, info->attrs);
1434         err = PTR_ERR(dp);
1435         if (IS_ERR(dp))
1436                 goto exit_unlock;
1437
1438         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1439                                       info->snd_seq, OVS_DP_CMD_DEL);
1440         err = PTR_ERR(reply);
1441         if (IS_ERR(reply))
1442                 goto exit_unlock;
1443
1444         list_for_each_entry_safe(vport, next_vport, &dp->port_list, node)
1445                 if (vport->port_no != OVSP_LOCAL)
1446                         dp_detach_port(vport);
1447
1448         dp_sysfs_del_dp(dp);
1449         list_del(&dp->list_node);
1450         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1451
1452         /* rtnl_unlock() will wait until all the references to devices that
1453          * are pending unregistration have been dropped.  We do it here to
1454          * ensure that any internal devices (which contain DP pointers) are
1455          * fully destroyed before freeing the datapath.
1456          */
1457         rtnl_unlock();
1458
1459         call_rcu(&dp->rcu, destroy_dp_rcu);
1460         module_put(THIS_MODULE);
1461
1462         genl_notify(reply, genl_info_net(info), info->snd_pid,
1463                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1464
1465         return 0;
1466
1467 exit_unlock:
1468         rtnl_unlock();
1469 exit:
1470         return err;
1471 }
1472
1473 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1474 {
1475         struct sk_buff *reply;
1476         struct datapath *dp;
1477         int err;
1478
1479         err = ovs_dp_cmd_validate(info->attrs);
1480         if (err)
1481                 return err;
1482
1483         dp = lookup_datapath(info->userhdr, info->attrs);
1484         if (IS_ERR(dp))
1485                 return PTR_ERR(dp);
1486
1487         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1488                                       info->snd_seq, OVS_DP_CMD_NEW);
1489         if (IS_ERR(reply)) {
1490                 err = PTR_ERR(reply);
1491                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1492                                 dp_datapath_multicast_group.id, err);
1493                 return 0;
1494         }
1495
1496         genl_notify(reply, genl_info_net(info), info->snd_pid,
1497                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1498         return 0;
1499 }
1500
1501 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1502 {
1503         struct sk_buff *reply;
1504         struct datapath *dp;
1505         int err;
1506
1507         err = ovs_dp_cmd_validate(info->attrs);
1508         if (err)
1509                 return err;
1510
1511         dp = lookup_datapath(info->userhdr, info->attrs);
1512         if (IS_ERR(dp))
1513                 return PTR_ERR(dp);
1514
1515         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1516                                       info->snd_seq, OVS_DP_CMD_NEW);
1517         if (IS_ERR(reply))
1518                 return PTR_ERR(reply);
1519
1520         return genlmsg_reply(reply, info);
1521 }
1522
1523 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1524 {
1525         struct datapath *dp;
1526         int skip = cb->args[0];
1527         int i = 0;
1528
1529         list_for_each_entry(dp, &dps, list_node) {
1530                 if (i < skip)
1531                         continue;
1532                 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1533                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1534                                          OVS_DP_CMD_NEW) < 0)
1535                         break;
1536                 i++;
1537         }
1538
1539         cb->args[0] = i;
1540
1541         return skb->len;
1542 }
1543
1544 static struct genl_ops dp_datapath_genl_ops[] = {
1545         { .cmd = OVS_DP_CMD_NEW,
1546           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1547           .policy = datapath_policy,
1548           .doit = ovs_dp_cmd_new
1549         },
1550         { .cmd = OVS_DP_CMD_DEL,
1551           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1552           .policy = datapath_policy,
1553           .doit = ovs_dp_cmd_del
1554         },
1555         { .cmd = OVS_DP_CMD_GET,
1556           .flags = 0,               /* OK for unprivileged users. */
1557           .policy = datapath_policy,
1558           .doit = ovs_dp_cmd_get,
1559           .dumpit = ovs_dp_cmd_dump
1560         },
1561         { .cmd = OVS_DP_CMD_SET,
1562           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1563           .policy = datapath_policy,
1564           .doit = ovs_dp_cmd_set,
1565         },
1566 };
1567
1568 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1569 #ifdef HAVE_NLA_NUL_STRING
1570         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1571         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1572         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1573 #else
1574         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1575         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1576 #endif
1577         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1578         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1579         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1580         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1581 };
1582
1583 static struct genl_family dp_vport_genl_family = {
1584         .id = GENL_ID_GENERATE,
1585         .hdrsize = sizeof(struct ovs_header),
1586         .name = OVS_VPORT_FAMILY,
1587         .version = OVS_VPORT_VERSION,
1588         .maxattr = OVS_VPORT_ATTR_MAX
1589 };
1590
1591 struct genl_multicast_group dp_vport_multicast_group = {
1592         .name = OVS_VPORT_MCGROUP
1593 };
1594
1595 /* Called with RTNL lock or RCU read lock. */
1596 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1597                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1598 {
1599         struct ovs_header *ovs_header;
1600         struct nlattr *nla;
1601         int err;
1602
1603         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1604                                  flags, cmd);
1605         if (!ovs_header)
1606                 return -EMSGSIZE;
1607
1608         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1609
1610         NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1611         NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type);
1612         NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport));
1613         NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1614
1615         nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS,
1616                           sizeof(struct ovs_vport_stats));
1617         if (!nla)
1618                 goto nla_put_failure;
1619
1620         vport_get_stats(vport, nla_data(nla));
1621
1622         NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN,
1623                 vport->ops->get_addr(vport));
1624
1625         err = vport_get_options(vport, skb);
1626         if (err == -EMSGSIZE)
1627                 goto error;
1628
1629         return genlmsg_end(skb, ovs_header);
1630
1631 nla_put_failure:
1632         err = -EMSGSIZE;
1633 error:
1634         genlmsg_cancel(skb, ovs_header);
1635         return err;
1636 }
1637
1638 /* Called with RTNL lock or RCU read lock. */
1639 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1640                                          u32 seq, u8 cmd)
1641 {
1642         struct sk_buff *skb;
1643         int retval;
1644
1645         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1646         if (!skb)
1647                 return ERR_PTR(-ENOMEM);
1648
1649         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1650         if (retval < 0) {
1651                 kfree_skb(skb);
1652                 return ERR_PTR(retval);
1653         }
1654         return skb;
1655 }
1656
1657 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1658 {
1659         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1660 }
1661
1662 /* Called with RTNL lock or RCU read lock. */
1663 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1664                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1665 {
1666         struct datapath *dp;
1667         struct vport *vport;
1668
1669         if (a[OVS_VPORT_ATTR_NAME]) {
1670                 vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1671                 if (!vport)
1672                         return ERR_PTR(-ENODEV);
1673                 return vport;
1674         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1675                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1676
1677                 if (port_no >= DP_MAX_PORTS)
1678                         return ERR_PTR(-EFBIG);
1679
1680                 dp = get_dp(ovs_header->dp_ifindex);
1681                 if (!dp)
1682                         return ERR_PTR(-ENODEV);
1683
1684                 vport = get_vport_protected(dp, port_no);
1685                 if (!vport)
1686                         return ERR_PTR(-ENOENT);
1687                 return vport;
1688         } else
1689                 return ERR_PTR(-EINVAL);
1690 }
1691
1692 /* Called with RTNL lock. */
1693 static int change_vport(struct vport *vport,
1694                         struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1695 {
1696         int err = 0;
1697
1698         if (a[OVS_VPORT_ATTR_STATS])
1699                 vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1700
1701         if (a[OVS_VPORT_ATTR_ADDRESS])
1702                 err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1703
1704         return err;
1705 }
1706
1707 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1708 {
1709         struct nlattr **a = info->attrs;
1710         struct ovs_header *ovs_header = info->userhdr;
1711         struct vport_parms parms;
1712         struct sk_buff *reply;
1713         struct vport *vport;
1714         struct datapath *dp;
1715         u32 port_no;
1716         int err;
1717
1718         err = -EINVAL;
1719         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1720             !a[OVS_VPORT_ATTR_UPCALL_PID])
1721                 goto exit;
1722
1723         err = ovs_vport_cmd_validate(a);
1724         if (err)
1725                 goto exit;
1726
1727         rtnl_lock();
1728         dp = get_dp(ovs_header->dp_ifindex);
1729         err = -ENODEV;
1730         if (!dp)
1731                 goto exit_unlock;
1732
1733         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1734                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1735
1736                 err = -EFBIG;
1737                 if (port_no >= DP_MAX_PORTS)
1738                         goto exit_unlock;
1739
1740                 vport = get_vport_protected(dp, port_no);
1741                 err = -EBUSY;
1742                 if (vport)
1743                         goto exit_unlock;
1744         } else {
1745                 for (port_no = 1; ; port_no++) {
1746                         if (port_no >= DP_MAX_PORTS) {
1747                                 err = -EFBIG;
1748                                 goto exit_unlock;
1749                         }
1750                         vport = get_vport_protected(dp, port_no);
1751                         if (!vport)
1752                                 break;
1753                 }
1754         }
1755
1756         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1757         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1758         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1759         parms.dp = dp;
1760         parms.port_no = port_no;
1761         parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1762
1763         vport = new_vport(&parms);
1764         err = PTR_ERR(vport);
1765         if (IS_ERR(vport))
1766                 goto exit_unlock;
1767
1768         dp_sysfs_add_if(vport);
1769
1770         err = change_vport(vport, a);
1771         if (!err) {
1772                 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1773                                                  info->snd_seq,
1774                                                  OVS_VPORT_CMD_NEW);
1775                 if (IS_ERR(reply))
1776                         err = PTR_ERR(reply);
1777         }
1778         if (err) {
1779                 dp_detach_port(vport);
1780                 goto exit_unlock;
1781         }
1782         genl_notify(reply, genl_info_net(info), info->snd_pid,
1783                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1784
1785
1786 exit_unlock:
1787         rtnl_unlock();
1788 exit:
1789         return err;
1790 }
1791
1792 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1793 {
1794         struct nlattr **a = info->attrs;
1795         struct sk_buff *reply;
1796         struct vport *vport;
1797         int err;
1798
1799         err = ovs_vport_cmd_validate(a);
1800         if (err)
1801                 goto exit;
1802
1803         rtnl_lock();
1804         vport = lookup_vport(info->userhdr, a);
1805         err = PTR_ERR(vport);
1806         if (IS_ERR(vport))
1807                 goto exit_unlock;
1808
1809         err = 0;
1810         if (a[OVS_VPORT_ATTR_TYPE] &&
1811             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1812                 err = -EINVAL;
1813
1814         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1815                 err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1816         if (!err)
1817                 err = change_vport(vport, a);
1818         if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1819                 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1820
1821         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1822                                          OVS_VPORT_CMD_NEW);
1823         if (IS_ERR(reply)) {
1824                 err = PTR_ERR(reply);
1825                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1826                                 dp_vport_multicast_group.id, err);
1827                 return 0;
1828         }
1829
1830         genl_notify(reply, genl_info_net(info), info->snd_pid,
1831                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1832
1833 exit_unlock:
1834         rtnl_unlock();
1835 exit:
1836         return err;
1837 }
1838
1839 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1840 {
1841         struct nlattr **a = info->attrs;
1842         struct sk_buff *reply;
1843         struct vport *vport;
1844         int err;
1845
1846         err = ovs_vport_cmd_validate(a);
1847         if (err)
1848                 goto exit;
1849
1850         rtnl_lock();
1851         vport = lookup_vport(info->userhdr, a);
1852         err = PTR_ERR(vport);
1853         if (IS_ERR(vport))
1854                 goto exit_unlock;
1855
1856         if (vport->port_no == OVSP_LOCAL) {
1857                 err = -EINVAL;
1858                 goto exit_unlock;
1859         }
1860
1861         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1862                                          OVS_VPORT_CMD_DEL);
1863         err = PTR_ERR(reply);
1864         if (IS_ERR(reply))
1865                 goto exit_unlock;
1866
1867         dp_detach_port(vport);
1868
1869         genl_notify(reply, genl_info_net(info), info->snd_pid,
1870                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1871
1872 exit_unlock:
1873         rtnl_unlock();
1874 exit:
1875         return err;
1876 }
1877
1878 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1879 {
1880         struct nlattr **a = info->attrs;
1881         struct ovs_header *ovs_header = info->userhdr;
1882         struct sk_buff *reply;
1883         struct vport *vport;
1884         int err;
1885
1886         err = ovs_vport_cmd_validate(a);
1887         if (err)
1888                 goto exit;
1889
1890         rcu_read_lock();
1891         vport = lookup_vport(ovs_header, a);
1892         err = PTR_ERR(vport);
1893         if (IS_ERR(vport))
1894                 goto exit_unlock;
1895
1896         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1897                                          OVS_VPORT_CMD_NEW);
1898         err = PTR_ERR(reply);
1899         if (IS_ERR(reply))
1900                 goto exit_unlock;
1901
1902         rcu_read_unlock();
1903
1904         return genlmsg_reply(reply, info);
1905
1906 exit_unlock:
1907         rcu_read_unlock();
1908 exit:
1909         return err;
1910 }
1911
1912 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1913 {
1914         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1915         struct datapath *dp;
1916         u32 port_no;
1917         int retval;
1918
1919         dp = get_dp(ovs_header->dp_ifindex);
1920         if (!dp)
1921                 return -ENODEV;
1922
1923         rcu_read_lock();
1924         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1925                 struct vport *vport;
1926
1927                 vport = get_vport_protected(dp, port_no);
1928                 if (!vport)
1929                         continue;
1930
1931                 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1932                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1933                                             OVS_VPORT_CMD_NEW) < 0)
1934                         break;
1935         }
1936         rcu_read_unlock();
1937
1938         cb->args[0] = port_no;
1939         retval = skb->len;
1940
1941         return retval;
1942 }
1943
1944 static struct genl_ops dp_vport_genl_ops[] = {
1945         { .cmd = OVS_VPORT_CMD_NEW,
1946           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1947           .policy = vport_policy,
1948           .doit = ovs_vport_cmd_new
1949         },
1950         { .cmd = OVS_VPORT_CMD_DEL,
1951           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1952           .policy = vport_policy,
1953           .doit = ovs_vport_cmd_del
1954         },
1955         { .cmd = OVS_VPORT_CMD_GET,
1956           .flags = 0,               /* OK for unprivileged users. */
1957           .policy = vport_policy,
1958           .doit = ovs_vport_cmd_get,
1959           .dumpit = ovs_vport_cmd_dump
1960         },
1961         { .cmd = OVS_VPORT_CMD_SET,
1962           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1963           .policy = vport_policy,
1964           .doit = ovs_vport_cmd_set,
1965         },
1966 };
1967
1968 struct genl_family_and_ops {
1969         struct genl_family *family;
1970         struct genl_ops *ops;
1971         int n_ops;
1972         struct genl_multicast_group *group;
1973 };
1974
1975 static const struct genl_family_and_ops dp_genl_families[] = {
1976         { &dp_datapath_genl_family,
1977           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1978           &dp_datapath_multicast_group },
1979         { &dp_vport_genl_family,
1980           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1981           &dp_vport_multicast_group },
1982         { &dp_flow_genl_family,
1983           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1984           &dp_flow_multicast_group },
1985         { &dp_packet_genl_family,
1986           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1987           NULL },
1988 };
1989
1990 static void dp_unregister_genl(int n_families)
1991 {
1992         int i;
1993
1994         for (i = 0; i < n_families; i++)
1995                 genl_unregister_family(dp_genl_families[i].family);
1996 }
1997
1998 static int dp_register_genl(void)
1999 {
2000         int n_registered;
2001         int err;
2002         int i;
2003
2004         n_registered = 0;
2005         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2006                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2007
2008                 err = genl_register_family_with_ops(f->family, f->ops,
2009                                                     f->n_ops);
2010                 if (err)
2011                         goto error;
2012                 n_registered++;
2013
2014                 if (f->group) {
2015                         err = genl_register_mc_group(f->family, f->group);
2016                         if (err)
2017                                 goto error;
2018                 }
2019         }
2020
2021         return 0;
2022
2023 error:
2024         dp_unregister_genl(n_registered);
2025         return err;
2026 }
2027
2028 static int __init dp_init(void)
2029 {
2030         struct sk_buff *dummy_skb;
2031         int err;
2032
2033         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2034
2035         pr_info("Open vSwitch %s, built "__DATE__" "__TIME__"\n",
2036                 VERSION BUILDNR);
2037
2038         err = tnl_init();
2039         if (err)
2040                 goto error;
2041
2042         err = flow_init();
2043         if (err)
2044                 goto error_tnl_exit;
2045
2046         err = vport_init();
2047         if (err)
2048                 goto error_flow_exit;
2049
2050         err = register_netdevice_notifier(&dp_device_notifier);
2051         if (err)
2052                 goto error_vport_exit;
2053
2054         err = dp_register_genl();
2055         if (err < 0)
2056                 goto error_unreg_notifier;
2057
2058         return 0;
2059
2060 error_unreg_notifier:
2061         unregister_netdevice_notifier(&dp_device_notifier);
2062 error_vport_exit:
2063         vport_exit();
2064 error_flow_exit:
2065         flow_exit();
2066 error_tnl_exit:
2067         tnl_exit();
2068 error:
2069         return err;
2070 }
2071
2072 static void dp_cleanup(void)
2073 {
2074         rcu_barrier();
2075         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2076         unregister_netdevice_notifier(&dp_device_notifier);
2077         vport_exit();
2078         flow_exit();
2079         tnl_exit();
2080 }
2081
2082 module_init(dp_init);
2083 module_exit(dp_cleanup);
2084
2085 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2086 MODULE_LICENSE("GPL");