datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007-2012 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/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 #include <net/net_namespace.h>
  54 #include <net/netns/generic.h>
  55
  56 #include "checksum.h"
  57 #include "datapath.h"
  58 #include "flow.h"
  59 #include "genl_exec.h"
  60 #include "vlan.h"
  61 #include "tunnel.h"
  62 #include "vport-internal_dev.h"
  63
  64 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
  65     LINUX_VERSION_CODE >= KERNEL_VERSION(3,4,0)
  66 #error Kernels before 2.6.18 or after 3.3 are not supported by this version of Open vSwitch.
  67 #endif
  68
  69 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
  70 static void rehash_flow_table(struct work_struct *work);
  71 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
  72
  73 int ovs_net_id __read_mostly;
  74
  75 int (*ovs_dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
  76 EXPORT_SYMBOL(ovs_dp_ioctl_hook);
  77
  78 /**
  79  * DOC: Locking:
  80  *
  81  * Writes to device state (add/remove datapath, port, set operations on vports,
  82  * etc.) are protected by RTNL.
  83  *
  84  * Writes to other state (flow table modifications, set miscellaneous datapath
  85  * parameters, etc.) are protected by genl_mutex.  The RTNL lock nests inside
  86  * genl_mutex.
  87  *
  88  * Reads are protected by RCU.
  89  *
  90  * There are a few special cases (mostly stats) that have their own
  91  * synchronization but they nest under all of above and don't interact with
  92  * each other.
  93  */
  94
  95 static struct vport *new_vport(const struct vport_parms *);
  96 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
  97                              const struct dp_upcall_info *);
  98 static int queue_userspace_packet(struct net *, int dp_ifindex,
  99                                   struct sk_buff *,
 100                                   const struct dp_upcall_info *);
 101
 102 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
 103 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 104 {
 105         struct datapath *dp = NULL;
 106         struct net_device *dev;
 107
 108         rcu_read_lock();
 109         dev = dev_get_by_index_rcu(net, dp_ifindex);
 110         if (dev) {
 111                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 112                 if (vport)
 113                         dp = vport->dp;
 114         }
 115         rcu_read_unlock();
 116
 117         return dp;
 118 }
 119
 120 /* Must be called with rcu_read_lock or RTNL lock. */
 121 const char *ovs_dp_name(const struct datapath *dp)
 122 {
 123         struct vport *vport = ovs_vport_rtnl_rcu(dp, 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 = ovs_vport_rcu(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         if (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             nla_put(skb, IFLA_ADDRESS, ETH_ALEN, port->ops->get_addr(port)))
 189                 goto nla_put_failure;
 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, ovs_dp_get_net(port->dp), 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 222
 223         return;
 224 err:
 225         rtnl_set_sk_err(ovs_dp_get_net(port->dp), 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         ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
 245         free_percpu(dp->stats_percpu);
 246         release_net(ovs_dp_get_net(dp));
 247         kfree(dp->ports);
 248         kobject_put(&dp->ifobj);
 249 }
 250
 251 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 252                                             u16 port_no)
 253 {
 254         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 255 }
 256
 257 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 258 {
 259         struct vport *vport;
 260         struct hlist_node *n;
 261         struct hlist_head *head;
 262
 263         head = vport_hash_bucket(dp, port_no);
 264         hlist_for_each_entry_rcu(vport, n, head, dp_hash_node) {
 265                 if (vport->port_no == port_no)
 266                         return vport;
 267         }
 268         return NULL;
 269 }
 270
 271 /* Called with RTNL lock and genl_lock. */
 272 static struct vport *new_vport(const struct vport_parms *parms)
 273 {
 274         struct vport *vport;
 275
 276         vport = ovs_vport_add(parms);
 277         if (!IS_ERR(vport)) {
 278                 struct datapath *dp = parms->dp;
 279                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 280
 281                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 282                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 283         }
 284         return vport;
 285 }
 286
 287 /* Called with RTNL lock. */
 288 void ovs_dp_detach_port(struct vport *p)
 289 {
 290         ASSERT_RTNL();
 291
 292         if (p->port_no != OVSP_LOCAL)
 293                 ovs_dp_sysfs_del_if(p);
 294
 295         dp_ifinfo_notify(RTM_DELLINK, p);
 296
 297         /* First drop references to device. */
 298         hlist_del_rcu(&p->dp_hash_node);
 299
 300         /* Then destroy it. */
 301         ovs_vport_del(p);
 302 }
 303
 304 /* Must be called with rcu_read_lock. */
 305 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 306 {
 307         struct datapath *dp = p->dp;
 308         struct sw_flow *flow;
 309         struct dp_stats_percpu *stats;
 310         u64 *stats_counter;
 311         int error;
 312
 313         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 314
 315         if (!OVS_CB(skb)->flow) {
 316                 struct sw_flow_key key;
 317                 int key_len;
 318
 319                 /* Extract flow from 'skb' into 'key'. */
 320                 error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
 321                 if (unlikely(error)) {
 322                         kfree_skb(skb);
 323                         return;
 324                 }
 325
 326                 /* Look up flow. */
 327                 flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table),
 328                                            &key, key_len);
 329                 if (unlikely(!flow)) {
 330                         struct dp_upcall_info upcall;
 331
 332                         upcall.cmd = OVS_PACKET_CMD_MISS;
 333                         upcall.key = &key;
 334                         upcall.userdata = NULL;
 335                         upcall.pid = p->upcall_pid;
 336                         ovs_dp_upcall(dp, skb, &upcall);
 337                         consume_skb(skb);
 338                         stats_counter = &stats->n_missed;
 339                         goto out;
 340                 }
 341
 342                 OVS_CB(skb)->flow = flow;
 343         }
 344
 345         stats_counter = &stats->n_hit;
 346         ovs_flow_used(OVS_CB(skb)->flow, skb);
 347         ovs_execute_actions(dp, skb);
 348
 349 out:
 350         /* Update datapath statistics. */
 351         u64_stats_update_begin(&stats->sync);
 352         (*stats_counter)++;
 353         u64_stats_update_end(&stats->sync);
 354 }
 355
 356 static struct genl_family dp_packet_genl_family = {
 357         .id = GENL_ID_GENERATE,
 358         .hdrsize = sizeof(struct ovs_header),
 359         .name = OVS_PACKET_FAMILY,
 360         .version = OVS_PACKET_VERSION,
 361         .maxattr = OVS_PACKET_ATTR_MAX,
 362          SET_NETNSOK
 363 };
 364
 365 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 366                   const struct dp_upcall_info *upcall_info)
 367 {
 368         struct dp_stats_percpu *stats;
 369         int dp_ifindex;
 370         int err;
 371
 372         if (upcall_info->pid == 0) {
 373                 err = -ENOTCONN;
 374                 goto err;
 375         }
 376
 377         dp_ifindex = get_dpifindex(dp);
 378         if (!dp_ifindex) {
 379                 err = -ENODEV;
 380                 goto err;
 381         }
 382
 383         forward_ip_summed(skb, true);
 384
 385         if (!skb_is_gso(skb))
 386                 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 387         else
 388                 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 389         if (err)
 390                 goto err;
 391
 392         return 0;
 393
 394 err:
 395         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 396
 397         u64_stats_update_begin(&stats->sync);
 398         stats->n_lost++;
 399         u64_stats_update_end(&stats->sync);
 400
 401         return err;
 402 }
 403
 404 static int queue_gso_packets(struct net *net, int dp_ifindex,
 405                              struct sk_buff *skb,
 406                              const struct dp_upcall_info *upcall_info)
 407 {
 408         struct dp_upcall_info later_info;
 409         struct sw_flow_key later_key;
 410         struct sk_buff *segs, *nskb;
 411         int err;
 412
 413         segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 414         if (IS_ERR(skb))
 415                 return PTR_ERR(skb);
 416
 417         /* Queue all of the segments. */
 418         skb = segs;
 419         do {
 420                 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
 421                 if (err)
 422                         break;
 423
 424                 if (skb == segs && skb_shinfo(skb)->gso_type & SKB_GSO_UDP) {
 425                         /* The initial flow key extracted by ovs_flow_extract()
 426                          * in this case is for a first fragment, so we need to
 427                          * properly mark later fragments.
 428                          */
 429                         later_key = *upcall_info->key;
 430                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 431
 432                         later_info = *upcall_info;
 433                         later_info.key = &later_key;
 434                         upcall_info = &later_info;
 435                 }
 436         } while ((skb = skb->next));
 437
 438         /* Free all of the segments. */
 439         skb = segs;
 440         do {
 441                 nskb = skb->next;
 442                 if (err)
 443                         kfree_skb(skb);
 444                 else
 445                         consume_skb(skb);
 446         } while ((skb = nskb));
 447         return err;
 448 }
 449
 450 static int queue_userspace_packet(struct net *net, int dp_ifindex,
 451                                   struct sk_buff *skb,
 452                                   const struct dp_upcall_info *upcall_info)
 453 {
 454         struct ovs_header *upcall;
 455         struct sk_buff *nskb = NULL;
 456         struct sk_buff *user_skb; /* to be queued to userspace */
 457         struct nlattr *nla;
 458         unsigned int len;
 459         int err;
 460
 461         if (vlan_tx_tag_present(skb)) {
 462                 nskb = skb_clone(skb, GFP_ATOMIC);
 463                 if (!nskb)
 464                         return -ENOMEM;
 465
 466                 err = vlan_deaccel_tag(nskb);
 467                 if (err)
 468                         return err;
 469
 470                 skb = nskb;
 471         }
 472
 473         if (nla_attr_size(skb->len) > USHRT_MAX) {
 474                 err = -EFBIG;
 475                 goto out;
 476         }
 477
 478         len = sizeof(struct ovs_header);
 479         len += nla_total_size(skb->len);
 480         len += nla_total_size(FLOW_BUFSIZE);
 481         if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
 482                 len += nla_total_size(8);
 483
 484         user_skb = genlmsg_new(len, GFP_ATOMIC);
 485         if (!user_skb) {
 486                 err = -ENOMEM;
 487                 goto out;
 488         }
 489
 490         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 491                              0, upcall_info->cmd);
 492         upcall->dp_ifindex = dp_ifindex;
 493
 494         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 495         ovs_flow_to_nlattrs(upcall_info->key, user_skb);
 496         nla_nest_end(user_skb, nla);
 497
 498         if (upcall_info->userdata)
 499                 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
 500                             nla_get_u64(upcall_info->userdata));
 501
 502         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 503
 504         skb_copy_and_csum_dev(skb, nla_data(nla));
 505
 506         err = genlmsg_unicast(net, user_skb, upcall_info->pid);
 507
 508 out:
 509         kfree_skb(nskb);
 510         return err;
 511 }
 512
 513 /* Called with genl_mutex. */
 514 static int flush_flows(struct datapath *dp)
 515 {
 516         struct flow_table *old_table;
 517         struct flow_table *new_table;
 518
 519         old_table = genl_dereference(dp->table);
 520         new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
 521         if (!new_table)
 522                 return -ENOMEM;
 523
 524         rcu_assign_pointer(dp->table, new_table);
 525
 526         ovs_flow_tbl_deferred_destroy(old_table);
 527         return 0;
 528 }
 529
 530 static int validate_actions(const struct nlattr *attr,
 531                                 const struct sw_flow_key *key, int depth);
 532
 533 static int validate_sample(const struct nlattr *attr,
 534                                 const struct sw_flow_key *key, int depth)
 535 {
 536         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 537         const struct nlattr *probability, *actions;
 538         const struct nlattr *a;
 539         int rem;
 540
 541         memset(attrs, 0, sizeof(attrs));
 542         nla_for_each_nested(a, attr, rem) {
 543                 int type = nla_type(a);
 544                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 545                         return -EINVAL;
 546                 attrs[type] = a;
 547         }
 548         if (rem)
 549                 return -EINVAL;
 550
 551         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 552         if (!probability || nla_len(probability) != sizeof(u32))
 553                 return -EINVAL;
 554
 555         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 556         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 557                 return -EINVAL;
 558         return validate_actions(actions, key, depth + 1);
 559 }
 560
 561 static int validate_set(const struct nlattr *a,
 562                         const struct sw_flow_key *flow_key)
 563 {
 564         const struct nlattr *ovs_key = nla_data(a);
 565         int key_type = nla_type(ovs_key);
 566
 567         /* There can be only one key in a action */
 568         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 569                 return -EINVAL;
 570
 571         if (key_type > OVS_KEY_ATTR_MAX ||
 572             nla_len(ovs_key) != ovs_key_lens[key_type])
 573                 return -EINVAL;
 574
 575         switch (key_type) {
 576         const struct ovs_key_ipv4 *ipv4_key;
 577
 578         case OVS_KEY_ATTR_PRIORITY:
 579         case OVS_KEY_ATTR_TUN_ID:
 580         case OVS_KEY_ATTR_ETHERNET:
 581                 break;
 582
 583         case OVS_KEY_ATTR_IPV4:
 584                 if (flow_key->eth.type != htons(ETH_P_IP))
 585                         return -EINVAL;
 586
 587                 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
 588                         return -EINVAL;
 589
 590                 ipv4_key = nla_data(ovs_key);
 591                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 592                         return -EINVAL;
 593
 594                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 595                         return -EINVAL;
 596
 597                 break;
 598
 599         case OVS_KEY_ATTR_TCP:
 600                 if (flow_key->ip.proto != IPPROTO_TCP)
 601                         return -EINVAL;
 602
 603                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 604                         return -EINVAL;
 605
 606                 break;
 607
 608         case OVS_KEY_ATTR_UDP:
 609                 if (flow_key->ip.proto != IPPROTO_UDP)
 610                         return -EINVAL;
 611
 612                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 613                         return -EINVAL;
 614                 break;
 615
 616         default:
 617                 return -EINVAL;
 618         }
 619
 620         return 0;
 621 }
 622
 623 static int validate_userspace(const struct nlattr *attr)
 624 {
 625         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 626                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 627                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
 628         };
 629         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 630         int error;
 631
 632         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 633                                  attr, userspace_policy);
 634         if (error)
 635                 return error;
 636
 637         if (!a[OVS_USERSPACE_ATTR_PID] ||
 638             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 639                 return -EINVAL;
 640
 641         return 0;
 642 }
 643
 644 static int validate_actions(const struct nlattr *attr,
 645                                 const struct sw_flow_key *key,  int depth)
 646 {
 647         const struct nlattr *a;
 648         int rem, err;
 649
 650         if (depth >= SAMPLE_ACTION_DEPTH)
 651                 return -EOVERFLOW;
 652
 653         nla_for_each_nested(a, attr, rem) {
 654                 /* Expected argument lengths, (u32)-1 for variable length. */
 655                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 656                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 657                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 658                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 659                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 660                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 661                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 662                 };
 663                 const struct ovs_action_push_vlan *vlan;
 664                 int type = nla_type(a);
 665
 666                 if (type > OVS_ACTION_ATTR_MAX ||
 667                     (action_lens[type] != nla_len(a) &&
 668                      action_lens[type] != (u32)-1))
 669                         return -EINVAL;
 670
 671                 switch (type) {
 672                 case OVS_ACTION_ATTR_UNSPEC:
 673                         return -EINVAL;
 674
 675                 case OVS_ACTION_ATTR_USERSPACE:
 676                         err = validate_userspace(a);
 677                         if (err)
 678                                 return err;
 679                         break;
 680
 681                 case OVS_ACTION_ATTR_OUTPUT:
 682                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 683                                 return -EINVAL;
 684                         break;
 685
 686
 687                 case OVS_ACTION_ATTR_POP_VLAN:
 688                         break;
 689
 690                 case OVS_ACTION_ATTR_PUSH_VLAN:
 691                         vlan = nla_data(a);
 692                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 693                                 return -EINVAL;
 694                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 695                                 return -EINVAL;
 696                         break;
 697
 698                 case OVS_ACTION_ATTR_SET:
 699                         err = validate_set(a, key);
 700                         if (err)
 701                                 return err;
 702                         break;
 703
 704                 case OVS_ACTION_ATTR_SAMPLE:
 705                         err = validate_sample(a, key, depth);
 706                         if (err)
 707                                 return err;
 708                         break;
 709
 710                 default:
 711                         return -EINVAL;
 712                 }
 713         }
 714
 715         if (rem > 0)
 716                 return -EINVAL;
 717
 718         return 0;
 719 }
 720
 721 static void clear_stats(struct sw_flow *flow)
 722 {
 723         flow->used = 0;
 724         flow->tcp_flags = 0;
 725         flow->packet_count = 0;
 726         flow->byte_count = 0;
 727 }
 728
 729 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 730 {
 731         struct ovs_header *ovs_header = info->userhdr;
 732         struct nlattr **a = info->attrs;
 733         struct sw_flow_actions *acts;
 734         struct sk_buff *packet;
 735         struct sw_flow *flow;
 736         struct datapath *dp;
 737         struct ethhdr *eth;
 738         int len;
 739         int err;
 740         int key_len;
 741
 742         err = -EINVAL;
 743         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 744             !a[OVS_PACKET_ATTR_ACTIONS] ||
 745             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 746                 goto err;
 747
 748         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 749         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 750         err = -ENOMEM;
 751         if (!packet)
 752                 goto err;
 753         skb_reserve(packet, NET_IP_ALIGN);
 754
 755         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 756
 757         skb_reset_mac_header(packet);
 758         eth = eth_hdr(packet);
 759
 760         /* Normally, setting the skb 'protocol' field would be handled by a
 761          * call to eth_type_trans(), but it assumes there's a sending
 762          * device, which we may not have. */
 763         if (ntohs(eth->h_proto) >= 1536)
 764                 packet->protocol = eth->h_proto;
 765         else
 766                 packet->protocol = htons(ETH_P_802_2);
 767
 768         /* Build an sw_flow for sending this packet. */
 769         flow = ovs_flow_alloc();
 770         err = PTR_ERR(flow);
 771         if (IS_ERR(flow))
 772                 goto err_kfree_skb;
 773
 774         err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
 775         if (err)
 776                 goto err_flow_put;
 777
 778         err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
 779                                              &flow->key.phy.in_port,
 780                                              &flow->key.phy.tun_id,
 781                                              a[OVS_PACKET_ATTR_KEY]);
 782         if (err)
 783                 goto err_flow_put;
 784
 785         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
 786         if (err)
 787                 goto err_flow_put;
 788
 789         flow->hash = ovs_flow_hash(&flow->key, key_len);
 790
 791         acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 792         err = PTR_ERR(acts);
 793         if (IS_ERR(acts))
 794                 goto err_flow_put;
 795         rcu_assign_pointer(flow->sf_acts, acts);
 796
 797         OVS_CB(packet)->flow = flow;
 798         packet->priority = flow->key.phy.priority;
 799
 800         rcu_read_lock();
 801         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 802         err = -ENODEV;
 803         if (!dp)
 804                 goto err_unlock;
 805
 806         local_bh_disable();
 807         err = ovs_execute_actions(dp, packet);
 808         local_bh_enable();
 809         rcu_read_unlock();
 810
 811         ovs_flow_put(flow);
 812         return err;
 813
 814 err_unlock:
 815         rcu_read_unlock();
 816 err_flow_put:
 817         ovs_flow_put(flow);
 818 err_kfree_skb:
 819         kfree_skb(packet);
 820 err:
 821         return err;
 822 }
 823
 824 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 825         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 826         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 827         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 828 };
 829
 830 static struct genl_ops dp_packet_genl_ops[] = {
 831         { .cmd = OVS_PACKET_CMD_EXECUTE,
 832           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 833           .policy = packet_policy,
 834           .doit = ovs_packet_cmd_execute
 835         }
 836 };
 837
 838 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 839 {
 840         int i;
 841         struct flow_table *table = genl_dereference(dp->table);
 842
 843         stats->n_flows = ovs_flow_tbl_count(table);
 844
 845         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 846         for_each_possible_cpu(i) {
 847                 const struct dp_stats_percpu *percpu_stats;
 848                 struct dp_stats_percpu local_stats;
 849                 unsigned int start;
 850
 851                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 852
 853                 do {
 854                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 855                         local_stats = *percpu_stats;
 856                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 857
 858                 stats->n_hit += local_stats.n_hit;
 859                 stats->n_missed += local_stats.n_missed;
 860                 stats->n_lost += local_stats.n_lost;
 861         }
 862 }
 863
 864 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 865         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 866         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 867         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 868 };
 869
 870 static struct genl_family dp_flow_genl_family = {
 871         .id = GENL_ID_GENERATE,
 872         .hdrsize = sizeof(struct ovs_header),
 873         .name = OVS_FLOW_FAMILY,
 874         .version = OVS_FLOW_VERSION,
 875         .maxattr = OVS_FLOW_ATTR_MAX,
 876          SET_NETNSOK
 877 };
 878
 879 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 880         .name = OVS_FLOW_MCGROUP
 881 };
 882
 883 /* Called with genl_lock. */
 884 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 885                                   struct sk_buff *skb, u32 pid,
 886                                   u32 seq, u32 flags, u8 cmd)
 887 {
 888         const int skb_orig_len = skb->len;
 889         const struct sw_flow_actions *sf_acts;
 890         struct ovs_flow_stats stats;
 891         struct ovs_header *ovs_header;
 892         struct nlattr *nla;
 893         unsigned long used;
 894         u8 tcp_flags;
 895         int err;
 896
 897         sf_acts = rcu_dereference_protected(flow->sf_acts,
 898                                             lockdep_genl_is_held());
 899
 900         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 901         if (!ovs_header)
 902                 return -EMSGSIZE;
 903
 904         ovs_header->dp_ifindex = get_dpifindex(dp);
 905
 906         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 907         if (!nla)
 908                 goto nla_put_failure;
 909         err = ovs_flow_to_nlattrs(&flow->key, skb);
 910         if (err)
 911                 goto error;
 912         nla_nest_end(skb, nla);
 913
 914         spin_lock_bh(&flow->lock);
 915         used = flow->used;
 916         stats.n_packets = flow->packet_count;
 917         stats.n_bytes = flow->byte_count;
 918         tcp_flags = flow->tcp_flags;
 919         spin_unlock_bh(&flow->lock);
 920
 921         if (used &&
 922             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 923                 goto nla_put_failure;
 924
 925         if (stats.n_packets &&
 926             nla_put(skb, OVS_FLOW_ATTR_STATS,
 927                     sizeof(struct ovs_flow_stats), &stats))
 928                 goto nla_put_failure;
 929
 930         if (tcp_flags &&
 931             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
 932                 goto nla_put_failure;
 933
 934         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 935          * this is the first flow to be dumped into 'skb'.  This is unusual for
 936          * Netlink but individual action lists can be longer than
 937          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 938          * The userspace caller can always fetch the actions separately if it
 939          * really wants them.  (Most userspace callers in fact don't care.)
 940          *
 941          * This can only fail for dump operations because the skb is always
 942          * properly sized for single flows.
 943          */
 944         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 945                       sf_acts->actions);
 946         if (err < 0 && skb_orig_len)
 947                 goto error;
 948
 949         return genlmsg_end(skb, ovs_header);
 950
 951 nla_put_failure:
 952         err = -EMSGSIZE;
 953 error:
 954         genlmsg_cancel(skb, ovs_header);
 955         return err;
 956 }
 957
 958 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 959 {
 960         const struct sw_flow_actions *sf_acts;
 961         int len;
 962
 963         sf_acts = rcu_dereference_protected(flow->sf_acts,
 964                                             lockdep_genl_is_held());
 965
 966         /* OVS_FLOW_ATTR_KEY */
 967         len = nla_total_size(FLOW_BUFSIZE);
 968         /* OVS_FLOW_ATTR_ACTIONS */
 969         len += nla_total_size(sf_acts->actions_len);
 970         /* OVS_FLOW_ATTR_STATS */
 971         len += nla_total_size(sizeof(struct ovs_flow_stats));
 972         /* OVS_FLOW_ATTR_TCP_FLAGS */
 973         len += nla_total_size(1);
 974         /* OVS_FLOW_ATTR_USED */
 975         len += nla_total_size(8);
 976
 977         len += NLMSG_ALIGN(sizeof(struct ovs_header));
 978
 979         return genlmsg_new(len, GFP_KERNEL);
 980 }
 981
 982 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 983                                                struct datapath *dp,
 984                                                u32 pid, u32 seq, u8 cmd)
 985 {
 986         struct sk_buff *skb;
 987         int retval;
 988
 989         skb = ovs_flow_cmd_alloc_info(flow);
 990         if (!skb)
 991                 return ERR_PTR(-ENOMEM);
 992
 993         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 994         BUG_ON(retval < 0);
 995         return skb;
 996 }
 997
 998 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 999 {
1000         struct nlattr **a = info->attrs;
1001         struct ovs_header *ovs_header = info->userhdr;
1002         struct sw_flow_key key;
1003         struct sw_flow *flow;
1004         struct sk_buff *reply;
1005         struct datapath *dp;
1006         struct flow_table *table;
1007         int error;
1008         int key_len;
1009
1010         /* Extract key. */
1011         error = -EINVAL;
1012         if (!a[OVS_FLOW_ATTR_KEY])
1013                 goto error;
1014         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1015         if (error)
1016                 goto error;
1017
1018         /* Validate actions. */
1019         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1020                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
1021                 if (error)
1022                         goto error;
1023         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1024                 error = -EINVAL;
1025                 goto error;
1026         }
1027
1028         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1029         error = -ENODEV;
1030         if (!dp)
1031                 goto error;
1032
1033         table = genl_dereference(dp->table);
1034         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1035         if (!flow) {
1036                 struct sw_flow_actions *acts;
1037
1038                 /* Bail out if we're not allowed to create a new flow. */
1039                 error = -ENOENT;
1040                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1041                         goto error;
1042
1043                 /* Expand table, if necessary, to make room. */
1044                 if (ovs_flow_tbl_need_to_expand(table)) {
1045                         struct flow_table *new_table;
1046
1047                         new_table = ovs_flow_tbl_expand(table);
1048                         if (!IS_ERR(new_table)) {
1049                                 rcu_assign_pointer(dp->table, new_table);
1050                                 ovs_flow_tbl_deferred_destroy(table);
1051                                 table = genl_dereference(dp->table);
1052                         }
1053                 }
1054
1055                 /* Allocate flow. */
1056                 flow = ovs_flow_alloc();
1057                 if (IS_ERR(flow)) {
1058                         error = PTR_ERR(flow);
1059                         goto error;
1060                 }
1061                 flow->key = key;
1062                 clear_stats(flow);
1063
1064                 /* Obtain actions. */
1065                 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1066                 error = PTR_ERR(acts);
1067                 if (IS_ERR(acts))
1068                         goto error_free_flow;
1069                 rcu_assign_pointer(flow->sf_acts, acts);
1070
1071                 /* Put flow in bucket. */
1072                 flow->hash = ovs_flow_hash(&key, key_len);
1073                 ovs_flow_tbl_insert(table, flow);
1074
1075                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1076                                                 info->snd_seq,
1077                                                 OVS_FLOW_CMD_NEW);
1078         } else {
1079                 /* We found a matching flow. */
1080                 struct sw_flow_actions *old_acts;
1081                 struct nlattr *acts_attrs;
1082
1083                 /* Bail out if we're not allowed to modify an existing flow.
1084                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1085                  * because Generic Netlink treats the latter as a dump
1086                  * request.  We also accept NLM_F_EXCL in case that bug ever
1087                  * gets fixed.
1088                  */
1089                 error = -EEXIST;
1090                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1091                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1092                         goto error;
1093
1094                 /* Update actions. */
1095                 old_acts = rcu_dereference_protected(flow->sf_acts,
1096                                                      lockdep_genl_is_held());
1097                 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1098                 if (acts_attrs &&
1099                    (old_acts->actions_len != nla_len(acts_attrs) ||
1100                    memcmp(old_acts->actions, nla_data(acts_attrs),
1101                           old_acts->actions_len))) {
1102                         struct sw_flow_actions *new_acts;
1103
1104                         new_acts = ovs_flow_actions_alloc(acts_attrs);
1105                         error = PTR_ERR(new_acts);
1106                         if (IS_ERR(new_acts))
1107                                 goto error;
1108
1109                         rcu_assign_pointer(flow->sf_acts, new_acts);
1110                         ovs_flow_deferred_free_acts(old_acts);
1111                 }
1112
1113                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1114                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1115
1116                 /* Clear stats. */
1117                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1118                         spin_lock_bh(&flow->lock);
1119                         clear_stats(flow);
1120                         spin_unlock_bh(&flow->lock);
1121                 }
1122         }
1123
1124         if (!IS_ERR(reply))
1125                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1126                            ovs_dp_flow_multicast_group.id, info->nlhdr,
1127                            GFP_KERNEL);
1128         else
1129                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1130                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1131         return 0;
1132
1133 error_free_flow:
1134         ovs_flow_put(flow);
1135 error:
1136         return error;
1137 }
1138
1139 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1140 {
1141         struct nlattr **a = info->attrs;
1142         struct ovs_header *ovs_header = info->userhdr;
1143         struct sw_flow_key key;
1144         struct sk_buff *reply;
1145         struct sw_flow *flow;
1146         struct datapath *dp;
1147         struct flow_table *table;
1148         int err;
1149         int key_len;
1150
1151         if (!a[OVS_FLOW_ATTR_KEY])
1152                 return -EINVAL;
1153         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1154         if (err)
1155                 return err;
1156
1157         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1158         if (!dp)
1159                 return -ENODEV;
1160
1161         table = genl_dereference(dp->table);
1162         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1163         if (!flow)
1164                 return -ENOENT;
1165
1166         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1167                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1168         if (IS_ERR(reply))
1169                 return PTR_ERR(reply);
1170
1171         return genlmsg_reply(reply, info);
1172 }
1173
1174 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1175 {
1176         struct nlattr **a = info->attrs;
1177         struct ovs_header *ovs_header = info->userhdr;
1178         struct sw_flow_key key;
1179         struct sk_buff *reply;
1180         struct sw_flow *flow;
1181         struct datapath *dp;
1182         struct flow_table *table;
1183         int err;
1184         int key_len;
1185
1186         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1187         if (!dp)
1188                 return -ENODEV;
1189
1190         if (!a[OVS_FLOW_ATTR_KEY])
1191                 return flush_flows(dp);
1192
1193         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1194         if (err)
1195                 return err;
1196
1197         table = genl_dereference(dp->table);
1198         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1199         if (!flow)
1200                 return -ENOENT;
1201
1202         reply = ovs_flow_cmd_alloc_info(flow);
1203         if (!reply)
1204                 return -ENOMEM;
1205
1206         ovs_flow_tbl_remove(table, flow);
1207
1208         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1209                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1210         BUG_ON(err < 0);
1211
1212         ovs_flow_deferred_free(flow);
1213
1214         genl_notify(reply, genl_info_net(info), info->snd_pid,
1215                     ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1216         return 0;
1217 }
1218
1219 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1220 {
1221         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1222         struct datapath *dp;
1223         struct flow_table *table;
1224
1225         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1226         if (!dp)
1227                 return -ENODEV;
1228
1229         table = genl_dereference(dp->table);
1230
1231         for (;;) {
1232                 struct sw_flow *flow;
1233                 u32 bucket, obj;
1234
1235                 bucket = cb->args[0];
1236                 obj = cb->args[1];
1237                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1238                 if (!flow)
1239                         break;
1240
1241                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1242                                            NETLINK_CB(cb->skb).pid,
1243                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1244                                            OVS_FLOW_CMD_NEW) < 0)
1245                         break;
1246
1247                 cb->args[0] = bucket;
1248                 cb->args[1] = obj;
1249         }
1250         return skb->len;
1251 }
1252
1253 static struct genl_ops dp_flow_genl_ops[] = {
1254         { .cmd = OVS_FLOW_CMD_NEW,
1255           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1256           .policy = flow_policy,
1257           .doit = ovs_flow_cmd_new_or_set
1258         },
1259         { .cmd = OVS_FLOW_CMD_DEL,
1260           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1261           .policy = flow_policy,
1262           .doit = ovs_flow_cmd_del
1263         },
1264         { .cmd = OVS_FLOW_CMD_GET,
1265           .flags = 0,               /* OK for unprivileged users. */
1266           .policy = flow_policy,
1267           .doit = ovs_flow_cmd_get,
1268           .dumpit = ovs_flow_cmd_dump
1269         },
1270         { .cmd = OVS_FLOW_CMD_SET,
1271           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1272           .policy = flow_policy,
1273           .doit = ovs_flow_cmd_new_or_set,
1274         },
1275 };
1276
1277 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1278 #ifdef HAVE_NLA_NUL_STRING
1279         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1280 #endif
1281         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1282 };
1283
1284 static struct genl_family dp_datapath_genl_family = {
1285         .id = GENL_ID_GENERATE,
1286         .hdrsize = sizeof(struct ovs_header),
1287         .name = OVS_DATAPATH_FAMILY,
1288         .version = OVS_DATAPATH_VERSION,
1289         .maxattr = OVS_DP_ATTR_MAX,
1290          SET_NETNSOK
1291 };
1292
1293 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1294         .name = OVS_DATAPATH_MCGROUP
1295 };
1296
1297 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1298                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1299 {
1300         struct ovs_header *ovs_header;
1301         struct ovs_dp_stats dp_stats;
1302         int err;
1303
1304         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1305                                    flags, cmd);
1306         if (!ovs_header)
1307                 goto error;
1308
1309         ovs_header->dp_ifindex = get_dpifindex(dp);
1310
1311         rcu_read_lock();
1312         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1313         rcu_read_unlock();
1314         if (err)
1315                 goto nla_put_failure;
1316
1317         get_dp_stats(dp, &dp_stats);
1318         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1319                 goto nla_put_failure;
1320
1321         return genlmsg_end(skb, ovs_header);
1322
1323 nla_put_failure:
1324         genlmsg_cancel(skb, ovs_header);
1325 error:
1326         return -EMSGSIZE;
1327 }
1328
1329 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1330                                              u32 seq, u8 cmd)
1331 {
1332         struct sk_buff *skb;
1333         int retval;
1334
1335         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1336         if (!skb)
1337                 return ERR_PTR(-ENOMEM);
1338
1339         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1340         if (retval < 0) {
1341                 kfree_skb(skb);
1342                 return ERR_PTR(retval);
1343         }
1344         return skb;
1345 }
1346
1347 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1348 {
1349         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1350 }
1351
1352 /* Called with genl_mutex and optionally with RTNL lock also. */
1353 static struct datapath *lookup_datapath(struct net *net,
1354                                         struct ovs_header *ovs_header,
1355                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1356 {
1357         struct datapath *dp;
1358
1359         if (!a[OVS_DP_ATTR_NAME])
1360                 dp = get_dp(net, ovs_header->dp_ifindex);
1361         else {
1362                 struct vport *vport;
1363
1364                 rcu_read_lock();
1365                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1366                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1367                 rcu_read_unlock();
1368         }
1369         return dp ? dp : ERR_PTR(-ENODEV);
1370 }
1371
1372 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1373 {
1374         struct nlattr **a = info->attrs;
1375         struct vport_parms parms;
1376         struct sk_buff *reply;
1377         struct datapath *dp;
1378         struct vport *vport;
1379         struct ovs_net *ovs_net;
1380         int err, i;
1381
1382         err = -EINVAL;
1383         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1384                 goto err;
1385
1386         err = ovs_dp_cmd_validate(a);
1387         if (err)
1388                 goto err;
1389
1390         rtnl_lock();
1391
1392         err = -ENOMEM;
1393         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1394         if (dp == NULL)
1395                 goto err_unlock_rtnl;
1396
1397         /* Initialize kobject for bridge.  This will be added as
1398          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1399         dp->ifobj.kset = NULL;
1400         kobject_init(&dp->ifobj, &dp_ktype);
1401
1402         /* Allocate table. */
1403         err = -ENOMEM;
1404         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1405         if (!dp->table)
1406                 goto err_free_dp;
1407
1408         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1409         if (!dp->stats_percpu) {
1410                 err = -ENOMEM;
1411                 goto err_destroy_table;
1412         }
1413         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1414
1415         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1416                             GFP_KERNEL);
1417         if (!dp->ports) {
1418                 err = -ENOMEM;
1419                 goto err_destroy_percpu;
1420         }
1421
1422         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1423                 INIT_HLIST_HEAD(&dp->ports[i]);
1424
1425         /* Set up our datapath device. */
1426         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1427         parms.type = OVS_VPORT_TYPE_INTERNAL;
1428         parms.options = NULL;
1429         parms.dp = dp;
1430         parms.port_no = OVSP_LOCAL;
1431         parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1432
1433         vport = new_vport(&parms);
1434         if (IS_ERR(vport)) {
1435                 err = PTR_ERR(vport);
1436                 if (err == -EBUSY)
1437                         err = -EEXIST;
1438
1439                 goto err_destroy_ports_array;
1440         }
1441
1442         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1443                                       info->snd_seq, OVS_DP_CMD_NEW);
1444         err = PTR_ERR(reply);
1445         if (IS_ERR(reply))
1446                 goto err_destroy_local_port;
1447
1448         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1449         list_add_tail(&dp->list_node, &ovs_net->dps);
1450         ovs_dp_sysfs_add_dp(dp);
1451
1452         rtnl_unlock();
1453
1454         genl_notify(reply, genl_info_net(info), info->snd_pid,
1455                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1456                     GFP_KERNEL);
1457         return 0;
1458
1459 err_destroy_local_port:
1460         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1461 err_destroy_ports_array:
1462         kfree(dp->ports);
1463 err_destroy_percpu:
1464         free_percpu(dp->stats_percpu);
1465 err_destroy_table:
1466         ovs_flow_tbl_destroy(genl_dereference(dp->table));
1467 err_free_dp:
1468         kfree(dp);
1469 err_unlock_rtnl:
1470         rtnl_unlock();
1471 err:
1472         return err;
1473 }
1474
1475 /* Called with genl_mutex. */
1476 static void __dp_destroy(struct datapath *dp)
1477 {
1478         int i;
1479
1480         rtnl_lock();
1481
1482         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1483                 struct vport *vport;
1484                 struct hlist_node *node, *n;
1485
1486                 hlist_for_each_entry_safe(vport, node, n, &dp->ports[i], dp_hash_node)
1487                         if (vport->port_no != OVSP_LOCAL)
1488                                 ovs_dp_detach_port(vport);
1489         }
1490
1491         ovs_dp_sysfs_del_dp(dp);
1492         list_del(&dp->list_node);
1493         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1494
1495         /* rtnl_unlock() will wait until all the references to devices that
1496          * are pending unregistration have been dropped.  We do it here to
1497          * ensure that any internal devices (which contain DP pointers) are
1498          * fully destroyed before freeing the datapath.
1499          */
1500         rtnl_unlock();
1501
1502         call_rcu(&dp->rcu, destroy_dp_rcu);
1503 }
1504
1505 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1506 {
1507         struct sk_buff *reply;
1508         struct datapath *dp;
1509         int err;
1510
1511         err = ovs_dp_cmd_validate(info->attrs);
1512         if (err)
1513                 return err;
1514
1515         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1516         err = PTR_ERR(dp);
1517         if (IS_ERR(dp))
1518                 return err;
1519
1520         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1521                                       info->snd_seq, OVS_DP_CMD_DEL);
1522         err = PTR_ERR(reply);
1523         if (IS_ERR(reply))
1524                 return err;
1525
1526         __dp_destroy(dp);
1527
1528         genl_notify(reply, genl_info_net(info), info->snd_pid,
1529                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1530                     GFP_KERNEL);
1531
1532         return 0;
1533 }
1534
1535 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1536 {
1537         struct sk_buff *reply;
1538         struct datapath *dp;
1539         int err;
1540
1541         err = ovs_dp_cmd_validate(info->attrs);
1542         if (err)
1543                 return err;
1544
1545         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1546         if (IS_ERR(dp))
1547                 return PTR_ERR(dp);
1548
1549         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1550                                       info->snd_seq, OVS_DP_CMD_NEW);
1551         if (IS_ERR(reply)) {
1552                 err = PTR_ERR(reply);
1553                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1554                                 ovs_dp_datapath_multicast_group.id, err);
1555                 return 0;
1556         }
1557
1558         genl_notify(reply, genl_info_net(info), info->snd_pid,
1559                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1560                     GFP_KERNEL);
1561
1562         return 0;
1563 }
1564
1565 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1566 {
1567         struct sk_buff *reply;
1568         struct datapath *dp;
1569         int err;
1570
1571         err = ovs_dp_cmd_validate(info->attrs);
1572         if (err)
1573                 return err;
1574
1575         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1576         if (IS_ERR(dp))
1577                 return PTR_ERR(dp);
1578
1579         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1580                                       info->snd_seq, OVS_DP_CMD_NEW);
1581         if (IS_ERR(reply))
1582                 return PTR_ERR(reply);
1583
1584         return genlmsg_reply(reply, info);
1585 }
1586
1587 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1588 {
1589         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1590         struct datapath *dp;
1591         int skip = cb->args[0];
1592         int i = 0;
1593
1594         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1595                 if (i >= skip &&
1596                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1597                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1598                                          OVS_DP_CMD_NEW) < 0)
1599                         break;
1600                 i++;
1601         }
1602
1603         cb->args[0] = i;
1604
1605         return skb->len;
1606 }
1607
1608 static struct genl_ops dp_datapath_genl_ops[] = {
1609         { .cmd = OVS_DP_CMD_NEW,
1610           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1611           .policy = datapath_policy,
1612           .doit = ovs_dp_cmd_new
1613         },
1614         { .cmd = OVS_DP_CMD_DEL,
1615           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1616           .policy = datapath_policy,
1617           .doit = ovs_dp_cmd_del
1618         },
1619         { .cmd = OVS_DP_CMD_GET,
1620           .flags = 0,               /* OK for unprivileged users. */
1621           .policy = datapath_policy,
1622           .doit = ovs_dp_cmd_get,
1623           .dumpit = ovs_dp_cmd_dump
1624         },
1625         { .cmd = OVS_DP_CMD_SET,
1626           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1627           .policy = datapath_policy,
1628           .doit = ovs_dp_cmd_set,
1629         },
1630 };
1631
1632 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1633 #ifdef HAVE_NLA_NUL_STRING
1634         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1635         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1636         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1637 #else
1638         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1639         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1640 #endif
1641         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1642         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1643         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1644         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1645 };
1646
1647 static struct genl_family dp_vport_genl_family = {
1648         .id = GENL_ID_GENERATE,
1649         .hdrsize = sizeof(struct ovs_header),
1650         .name = OVS_VPORT_FAMILY,
1651         .version = OVS_VPORT_VERSION,
1652         .maxattr = OVS_VPORT_ATTR_MAX,
1653          SET_NETNSOK
1654 };
1655
1656 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1657         .name = OVS_VPORT_MCGROUP
1658 };
1659
1660 /* Called with RTNL lock or RCU read lock. */
1661 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1662                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1663 {
1664         struct ovs_header *ovs_header;
1665         struct ovs_vport_stats vport_stats;
1666         int err;
1667
1668         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1669                                  flags, cmd);
1670         if (!ovs_header)
1671                 return -EMSGSIZE;
1672
1673         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1674
1675         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1676             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1677             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1678             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid))
1679                 goto nla_put_failure;
1680
1681         ovs_vport_get_stats(vport, &vport_stats);
1682         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1683                     &vport_stats))
1684                 goto nla_put_failure;
1685
1686         if (nla_put(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN,
1687                     vport->ops->get_addr(vport)))
1688                 goto nla_put_failure;
1689
1690         err = ovs_vport_get_options(vport, skb);
1691         if (err == -EMSGSIZE)
1692                 goto error;
1693
1694         return genlmsg_end(skb, ovs_header);
1695
1696 nla_put_failure:
1697         err = -EMSGSIZE;
1698 error:
1699         genlmsg_cancel(skb, ovs_header);
1700         return err;
1701 }
1702
1703 /* Called with RTNL lock or RCU read lock. */
1704 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1705                                          u32 seq, u8 cmd)
1706 {
1707         struct sk_buff *skb;
1708         int retval;
1709
1710         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1711         if (!skb)
1712                 return ERR_PTR(-ENOMEM);
1713
1714         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1715         if (retval < 0) {
1716                 kfree_skb(skb);
1717                 return ERR_PTR(retval);
1718         }
1719         return skb;
1720 }
1721
1722 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1723 {
1724         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1725 }
1726
1727 /* Called with RTNL lock or RCU read lock. */
1728 static struct vport *lookup_vport(struct net *net,
1729                                   struct ovs_header *ovs_header,
1730                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1731 {
1732         struct datapath *dp;
1733         struct vport *vport;
1734
1735         if (a[OVS_VPORT_ATTR_NAME]) {
1736                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1737                 if (!vport)
1738                         return ERR_PTR(-ENODEV);
1739                 if (ovs_header->dp_ifindex &&
1740                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1741                         return ERR_PTR(-ENODEV);
1742                 return vport;
1743         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1744                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1745
1746                 if (port_no >= DP_MAX_PORTS)
1747                         return ERR_PTR(-EFBIG);
1748
1749                 dp = get_dp(net, ovs_header->dp_ifindex);
1750                 if (!dp)
1751                         return ERR_PTR(-ENODEV);
1752
1753                 vport = ovs_vport_rtnl_rcu(dp, port_no);
1754                 if (!vport)
1755                         return ERR_PTR(-ENOENT);
1756                 return vport;
1757         } else
1758                 return ERR_PTR(-EINVAL);
1759 }
1760
1761 /* Called with RTNL lock. */
1762 static int change_vport(struct vport *vport,
1763                         struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1764 {
1765         int err = 0;
1766
1767         if (a[OVS_VPORT_ATTR_STATS])
1768                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1769
1770         if (a[OVS_VPORT_ATTR_ADDRESS])
1771                 err = ovs_vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1772
1773         return err;
1774 }
1775
1776 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1777 {
1778         struct nlattr **a = info->attrs;
1779         struct ovs_header *ovs_header = info->userhdr;
1780         struct vport_parms parms;
1781         struct sk_buff *reply;
1782         struct vport *vport;
1783         struct datapath *dp;
1784         u32 port_no;
1785         int err;
1786
1787         err = -EINVAL;
1788         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1789             !a[OVS_VPORT_ATTR_UPCALL_PID])
1790                 goto exit;
1791
1792         err = ovs_vport_cmd_validate(a);
1793         if (err)
1794                 goto exit;
1795
1796         rtnl_lock();
1797         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1798         err = -ENODEV;
1799         if (!dp)
1800                 goto exit_unlock;
1801
1802         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1803                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1804
1805                 err = -EFBIG;
1806                 if (port_no >= DP_MAX_PORTS)
1807                         goto exit_unlock;
1808
1809                 vport = ovs_vport_rtnl(dp, port_no);
1810                 err = -EBUSY;
1811                 if (vport)
1812                         goto exit_unlock;
1813         } else {
1814                 for (port_no = 1; ; port_no++) {
1815                         if (port_no >= DP_MAX_PORTS) {
1816                                 err = -EFBIG;
1817                                 goto exit_unlock;
1818                         }
1819                         vport = ovs_vport_rtnl(dp, port_no);
1820                         if (!vport)
1821                                 break;
1822                 }
1823         }
1824
1825         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1826         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1827         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1828         parms.dp = dp;
1829         parms.port_no = port_no;
1830         parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1831
1832         vport = new_vport(&parms);
1833         err = PTR_ERR(vport);
1834         if (IS_ERR(vport))
1835                 goto exit_unlock;
1836
1837         ovs_dp_sysfs_add_if(vport);
1838
1839         err = change_vport(vport, a);
1840         if (!err) {
1841                 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1842                                                  info->snd_seq,
1843                                                  OVS_VPORT_CMD_NEW);
1844                 if (IS_ERR(reply))
1845                         err = PTR_ERR(reply);
1846         }
1847         if (err) {
1848                 ovs_dp_detach_port(vport);
1849                 goto exit_unlock;
1850         }
1851         genl_notify(reply, genl_info_net(info), info->snd_pid,
1852                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1853
1854 exit_unlock:
1855         rtnl_unlock();
1856 exit:
1857         return err;
1858 }
1859
1860 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1861 {
1862         struct nlattr **a = info->attrs;
1863         struct sk_buff *reply;
1864         struct vport *vport;
1865         int err;
1866
1867         err = ovs_vport_cmd_validate(a);
1868         if (err)
1869                 goto exit;
1870
1871         rtnl_lock();
1872         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1873         err = PTR_ERR(vport);
1874         if (IS_ERR(vport))
1875                 goto exit_unlock;
1876
1877         err = 0;
1878         if (a[OVS_VPORT_ATTR_TYPE] &&
1879             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1880                 err = -EINVAL;
1881
1882         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1883                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1884         if (!err)
1885                 err = change_vport(vport, a);
1886         else
1887                 goto exit_unlock;
1888         if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1889                 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1890
1891         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1892                                          OVS_VPORT_CMD_NEW);
1893         if (IS_ERR(reply)) {
1894                 err = PTR_ERR(reply);
1895                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1896                                 ovs_dp_vport_multicast_group.id, err);
1897                 return 0;
1898         }
1899
1900         genl_notify(reply, genl_info_net(info), info->snd_pid,
1901                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1902
1903 exit_unlock:
1904         rtnl_unlock();
1905 exit:
1906         return err;
1907 }
1908
1909 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1910 {
1911         struct nlattr **a = info->attrs;
1912         struct sk_buff *reply;
1913         struct vport *vport;
1914         int err;
1915
1916         err = ovs_vport_cmd_validate(a);
1917         if (err)
1918                 goto exit;
1919
1920         rtnl_lock();
1921         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1922         err = PTR_ERR(vport);
1923         if (IS_ERR(vport))
1924                 goto exit_unlock;
1925
1926         if (vport->port_no == OVSP_LOCAL) {
1927                 err = -EINVAL;
1928                 goto exit_unlock;
1929         }
1930
1931         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1932                                          OVS_VPORT_CMD_DEL);
1933         err = PTR_ERR(reply);
1934         if (IS_ERR(reply))
1935                 goto exit_unlock;
1936
1937         ovs_dp_detach_port(vport);
1938
1939         genl_notify(reply, genl_info_net(info), info->snd_pid,
1940                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1941
1942 exit_unlock:
1943         rtnl_unlock();
1944 exit:
1945         return err;
1946 }
1947
1948 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1949 {
1950         struct nlattr **a = info->attrs;
1951         struct ovs_header *ovs_header = info->userhdr;
1952         struct sk_buff *reply;
1953         struct vport *vport;
1954         int err;
1955
1956         err = ovs_vport_cmd_validate(a);
1957         if (err)
1958                 goto exit;
1959
1960         rcu_read_lock();
1961         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1962         err = PTR_ERR(vport);
1963         if (IS_ERR(vport))
1964                 goto exit_unlock;
1965
1966         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1967                                          OVS_VPORT_CMD_NEW);
1968         err = PTR_ERR(reply);
1969         if (IS_ERR(reply))
1970                 goto exit_unlock;
1971
1972         rcu_read_unlock();
1973
1974         return genlmsg_reply(reply, info);
1975
1976 exit_unlock:
1977         rcu_read_unlock();
1978 exit:
1979         return err;
1980 }
1981
1982 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1983 {
1984         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1985         struct datapath *dp;
1986         int bucket = cb->args[0], skip = cb->args[1];
1987         int i, j = 0;
1988
1989         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1990         if (!dp)
1991                 return -ENODEV;
1992
1993         rcu_read_lock();
1994         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1995                 struct vport *vport;
1996                 struct hlist_node *n;
1997
1998                 j = 0;
1999                 hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node) {
2000                         if (j >= skip &&
2001                             ovs_vport_cmd_fill_info(vport, skb,
2002                                                     NETLINK_CB(cb->skb).pid,
2003                                                     cb->nlh->nlmsg_seq,
2004                                                     NLM_F_MULTI,
2005                                                     OVS_VPORT_CMD_NEW) < 0)
2006                                 goto out;
2007
2008                         j++;
2009                 }
2010                 skip = 0;
2011         }
2012 out:
2013         rcu_read_unlock();
2014
2015         cb->args[0] = i;
2016         cb->args[1] = j;
2017
2018         return skb->len;
2019 }
2020
2021 static struct genl_ops dp_vport_genl_ops[] = {
2022         { .cmd = OVS_VPORT_CMD_NEW,
2023           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2024           .policy = vport_policy,
2025           .doit = ovs_vport_cmd_new
2026         },
2027         { .cmd = OVS_VPORT_CMD_DEL,
2028           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2029           .policy = vport_policy,
2030           .doit = ovs_vport_cmd_del
2031         },
2032         { .cmd = OVS_VPORT_CMD_GET,
2033           .flags = 0,               /* OK for unprivileged users. */
2034           .policy = vport_policy,
2035           .doit = ovs_vport_cmd_get,
2036           .dumpit = ovs_vport_cmd_dump
2037         },
2038         { .cmd = OVS_VPORT_CMD_SET,
2039           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2040           .policy = vport_policy,
2041           .doit = ovs_vport_cmd_set,
2042         },
2043 };
2044
2045 struct genl_family_and_ops {
2046         struct genl_family *family;
2047         struct genl_ops *ops;
2048         int n_ops;
2049         struct genl_multicast_group *group;
2050 };
2051
2052 static const struct genl_family_and_ops dp_genl_families[] = {
2053         { &dp_datapath_genl_family,
2054           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2055           &ovs_dp_datapath_multicast_group },
2056         { &dp_vport_genl_family,
2057           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2058           &ovs_dp_vport_multicast_group },
2059         { &dp_flow_genl_family,
2060           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2061           &ovs_dp_flow_multicast_group },
2062         { &dp_packet_genl_family,
2063           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2064           NULL },
2065 };
2066
2067 static void dp_unregister_genl(int n_families)
2068 {
2069         int i;
2070
2071         for (i = 0; i < n_families; i++)
2072                 genl_unregister_family(dp_genl_families[i].family);
2073 }
2074
2075 static int dp_register_genl(void)
2076 {
2077         int n_registered;
2078         int err;
2079         int i;
2080
2081         n_registered = 0;
2082         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2083                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2084
2085                 err = genl_register_family_with_ops(f->family, f->ops,
2086                                                     f->n_ops);
2087                 if (err)
2088                         goto error;
2089                 n_registered++;
2090
2091                 if (f->group) {
2092                         err = genl_register_mc_group(f->family, f->group);
2093                         if (err)
2094                                 goto error;
2095                 }
2096         }
2097
2098         return 0;
2099
2100 error:
2101         dp_unregister_genl(n_registered);
2102         return err;
2103 }
2104
2105 static int __rehash_flow_table(void *dummy)
2106 {
2107         struct datapath *dp;
2108         struct net *net;
2109
2110         rtnl_lock();
2111         for_each_net(net) {
2112                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2113
2114                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2115                         struct flow_table *old_table = genl_dereference(dp->table);
2116                         struct flow_table *new_table;
2117
2118                         new_table = ovs_flow_tbl_rehash(old_table);
2119                         if (!IS_ERR(new_table)) {
2120                                 rcu_assign_pointer(dp->table, new_table);
2121                                 ovs_flow_tbl_deferred_destroy(old_table);
2122                         }
2123                 }
2124         }
2125         rtnl_unlock();
2126         return 0;
2127 }
2128
2129 static void rehash_flow_table(struct work_struct *work)
2130 {
2131         genl_exec(__rehash_flow_table, NULL);
2132         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2133 }
2134
2135 static int dp_destroy_all(void *data)
2136 {
2137         struct datapath *dp, *dp_next;
2138         struct ovs_net *ovs_net = data;
2139
2140         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2141                 __dp_destroy(dp);
2142
2143         return 0;
2144 }
2145
2146 static int __net_init ovs_init_net(struct net *net)
2147 {
2148         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2149
2150         INIT_LIST_HEAD(&ovs_net->dps);
2151         return 0;
2152 }
2153
2154 static void __net_exit ovs_exit_net(struct net *net)
2155 {
2156         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2157
2158         genl_exec(dp_destroy_all, ovs_net);
2159 }
2160
2161 static struct pernet_operations ovs_net_ops = {
2162         .init = ovs_init_net,
2163         .exit = ovs_exit_net,
2164         .id   = &ovs_net_id,
2165         .size = sizeof(struct ovs_net),
2166 };
2167
2168 static int __init dp_init(void)
2169 {
2170         struct sk_buff *dummy_skb;
2171         int err;
2172
2173         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2174
2175         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2176                 VERSION);
2177
2178         err = genl_exec_init();
2179         if (err)
2180                 goto error;
2181
2182         err = ovs_workqueues_init();
2183         if (err)
2184                 goto error_genl_exec;
2185
2186         err = ovs_tnl_init();
2187         if (err)
2188                 goto error_wq;
2189
2190         err = ovs_flow_init();
2191         if (err)
2192                 goto error_tnl_exit;
2193
2194         err = ovs_vport_init();
2195         if (err)
2196                 goto error_flow_exit;
2197
2198         err = register_pernet_device(&ovs_net_ops);
2199         if (err)
2200                 goto error_vport_exit;
2201
2202         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2203         if (err)
2204                 goto error_netns_exit;
2205
2206         err = dp_register_genl();
2207         if (err < 0)
2208                 goto error_unreg_notifier;
2209
2210         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2211
2212         return 0;
2213
2214 error_unreg_notifier:
2215         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2216 error_netns_exit:
2217         unregister_pernet_device(&ovs_net_ops);
2218 error_vport_exit:
2219         ovs_vport_exit();
2220 error_flow_exit:
2221         ovs_flow_exit();
2222 error_tnl_exit:
2223         ovs_tnl_exit();
2224 error_wq:
2225         ovs_workqueues_exit();
2226 error_genl_exec:
2227         genl_exec_exit();
2228 error:
2229         return err;
2230 }
2231
2232 static void dp_cleanup(void)
2233 {
2234         cancel_delayed_work_sync(&rehash_flow_wq);
2235         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2236         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2237         unregister_pernet_device(&ovs_net_ops);
2238         rcu_barrier();
2239         ovs_vport_exit();
2240         ovs_flow_exit();
2241         ovs_tnl_exit();
2242         ovs_workqueues_exit();
2243         genl_exec_exit();
2244 }
2245
2246 module_init(dp_init);
2247 module_exit(dp_cleanup);
2248
2249 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2250 MODULE_LICENSE("GPL");