datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007-2012 Nicira Networks.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/version.h>
  40 #include <linux/ethtool.h>
  41 #include <linux/wait.h>
  42 #include <asm/system.h>
  43 #include <asm/div64.h>
  44 #include <linux/highmem.h>
  45 #include <linux/netfilter_bridge.h>
  46 #include <linux/netfilter_ipv4.h>
  47 #include <linux/inetdevice.h>
  48 #include <linux/list.h>
  49 #include <linux/openvswitch.h>
  50 #include <linux/rculist.h>
  51 #include <linux/dmi.h>
  52 #include <net/genetlink.h>
  53
  54 #include "checksum.h"
  55 #include "datapath.h"
  56 #include "flow.h"
  57 #include "genl_exec.h"
  58 #include "vlan.h"
  59 #include "tunnel.h"
  60 #include "vport-internal_dev.h"
  61
  62 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
  63     LINUX_VERSION_CODE >= KERNEL_VERSION(3,3,0)
  64 #error Kernels before 2.6.18 or after 3.2 are not supported by this version of Open vSwitch.
  65 #endif
  66
  67 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
  68 static void rehash_flow_table(struct work_struct *work);
  69 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
  70
  71 int (*ovs_dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
  72 EXPORT_SYMBOL(ovs_dp_ioctl_hook);
  73
  74 /**
  75  * DOC: Locking:
  76  *
  77  * Writes to device state (add/remove datapath, port, set operations on vports,
  78  * etc.) are protected by RTNL.
  79  *
  80  * Writes to other state (flow table modifications, set miscellaneous datapath
  81  * parameters, etc.) are protected by genl_mutex.  The RTNL lock nests inside
  82  * genl_mutex.
  83  *
  84  * Reads are protected by RCU.
  85  *
  86  * There are a few special cases (mostly stats) that have their own
  87  * synchronization but they nest under all of above and don't interact with
  88  * each other.
  89  */
  90
  91 /* Global list of datapaths to enable dumping them all out.
  92  * Protected by genl_mutex.
  93  */
  94 static LIST_HEAD(dps);
  95
  96 static struct vport *new_vport(const struct vport_parms *);
  97 static int queue_gso_packets(int dp_ifindex, struct sk_buff *,
  98                              const struct dp_upcall_info *);
  99 static int queue_userspace_packet(int dp_ifindex, 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(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(&init_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 = 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 = rcu_dereference(dp->ports[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         ovs_flow_tbl_destroy((__force struct flow_table *)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 = ovs_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 ovs_dp_detach_port(struct vport *p)
 269 {
 270         ASSERT_RTNL();
 271
 272         if (p->port_no != OVSP_LOCAL)
 273                 ovs_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         ovs_vport_del(p);
 282 }
 283
 284 /* Must be called with rcu_read_lock. */
 285 void ovs_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 = ovs_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 = ovs_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                         ovs_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         ovs_flow_used(OVS_CB(skb)->flow, skb);
 327         ovs_execute_actions(dp, skb);
 328
 329 out:
 330         /* Update datapath statistics. */
 331         u64_stats_update_begin(&stats->sync);
 332         (*stats_counter)++;
 333         u64_stats_update_end(&stats->sync);
 334 }
 335
 336 static struct genl_family dp_packet_genl_family = {
 337         .id = GENL_ID_GENERATE,
 338         .hdrsize = sizeof(struct ovs_header),
 339         .name = OVS_PACKET_FAMILY,
 340         .version = OVS_PACKET_VERSION,
 341         .maxattr = OVS_PACKET_ATTR_MAX
 342 };
 343
 344 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 345                   const struct dp_upcall_info *upcall_info)
 346 {
 347         struct dp_stats_percpu *stats;
 348         int dp_ifindex;
 349         int err;
 350
 351         if (upcall_info->pid == 0) {
 352                 err = -ENOTCONN;
 353                 goto err;
 354         }
 355
 356         dp_ifindex = get_dpifindex(dp);
 357         if (!dp_ifindex) {
 358                 err = -ENODEV;
 359                 goto err;
 360         }
 361
 362         forward_ip_summed(skb, true);
 363
 364         if (!skb_is_gso(skb))
 365                 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
 366         else
 367                 err = queue_gso_packets(dp_ifindex, skb, upcall_info);
 368         if (err)
 369                 goto err;
 370
 371         return 0;
 372
 373 err:
 374         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 375
 376         u64_stats_update_begin(&stats->sync);
 377         stats->n_lost++;
 378         u64_stats_update_end(&stats->sync);
 379
 380         return err;
 381 }
 382
 383 static int queue_gso_packets(int dp_ifindex, struct sk_buff *skb,
 384                              const struct dp_upcall_info *upcall_info)
 385 {
 386         struct dp_upcall_info later_info;
 387         struct sw_flow_key later_key;
 388         struct sk_buff *segs, *nskb;
 389         int err;
 390
 391         segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 392         if (IS_ERR(skb))
 393                 return PTR_ERR(skb);
 394
 395         /* Queue all of the segments. */
 396         skb = segs;
 397         do {
 398                 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
 399                 if (err)
 400                         break;
 401
 402                 if (skb == segs && skb_shinfo(skb)->gso_type & SKB_GSO_UDP) {
 403                         /* The initial flow key extracted by ovs_flow_extract()
 404                          * in this case is for a first fragment, so we need to
 405                          * properly mark later fragments.
 406                          */
 407                         later_key = *upcall_info->key;
 408                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 409
 410                         later_info = *upcall_info;
 411                         later_info.key = &later_key;
 412                         upcall_info = &later_info;
 413                 }
 414         } while ((skb = skb->next));
 415
 416         /* Free all of the segments. */
 417         skb = segs;
 418         do {
 419                 nskb = skb->next;
 420                 if (err)
 421                         kfree_skb(skb);
 422                 else
 423                         consume_skb(skb);
 424         } while ((skb = nskb));
 425         return err;
 426 }
 427
 428 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *skb,
 429                                   const struct dp_upcall_info *upcall_info)
 430 {
 431         struct ovs_header *upcall;
 432         struct sk_buff *nskb = NULL;
 433         struct sk_buff *user_skb; /* to be queued to userspace */
 434         struct nlattr *nla;
 435         unsigned int len;
 436         int err;
 437
 438         if (vlan_tx_tag_present(skb)) {
 439                 nskb = skb_clone(skb, GFP_ATOMIC);
 440                 if (!nskb)
 441                         return -ENOMEM;
 442
 443                 err = vlan_deaccel_tag(nskb);
 444                 if (err)
 445                         return err;
 446
 447                 skb = nskb;
 448         }
 449
 450         if (nla_attr_size(skb->len) > USHRT_MAX) {
 451                 err = -EFBIG;
 452                 goto out;
 453         }
 454
 455         len = sizeof(struct ovs_header);
 456         len += nla_total_size(skb->len);
 457         len += nla_total_size(FLOW_BUFSIZE);
 458         if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
 459                 len += nla_total_size(8);
 460
 461         user_skb = genlmsg_new(len, GFP_ATOMIC);
 462         if (!user_skb) {
 463                 err = -ENOMEM;
 464                 goto out;
 465         }
 466
 467         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 468                              0, upcall_info->cmd);
 469         upcall->dp_ifindex = dp_ifindex;
 470
 471         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 472         ovs_flow_to_nlattrs(upcall_info->key, user_skb);
 473         nla_nest_end(user_skb, nla);
 474
 475         if (upcall_info->userdata)
 476                 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
 477                             nla_get_u64(upcall_info->userdata));
 478
 479         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 480
 481         skb_copy_and_csum_dev(skb, nla_data(nla));
 482
 483         err = genlmsg_unicast(&init_net, user_skb, upcall_info->pid);
 484
 485 out:
 486         kfree_skb(nskb);
 487         return err;
 488 }
 489
 490 /* Called with genl_mutex. */
 491 static int flush_flows(int dp_ifindex)
 492 {
 493         struct flow_table *old_table;
 494         struct flow_table *new_table;
 495         struct datapath *dp;
 496
 497         dp = get_dp(dp_ifindex);
 498         if (!dp)
 499                 return -ENODEV;
 500
 501         old_table = genl_dereference(dp->table);
 502         new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
 503         if (!new_table)
 504                 return -ENOMEM;
 505
 506         rcu_assign_pointer(dp->table, new_table);
 507
 508         ovs_flow_tbl_deferred_destroy(old_table);
 509         return 0;
 510 }
 511
 512 static int validate_actions(const struct nlattr *attr,
 513                                 const struct sw_flow_key *key, int depth);
 514
 515 static int validate_sample(const struct nlattr *attr,
 516                                 const struct sw_flow_key *key, int depth)
 517 {
 518         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 519         const struct nlattr *probability, *actions;
 520         const struct nlattr *a;
 521         int rem;
 522
 523         memset(attrs, 0, sizeof(attrs));
 524         nla_for_each_nested(a, attr, rem) {
 525                 int type = nla_type(a);
 526                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 527                         return -EINVAL;
 528                 attrs[type] = a;
 529         }
 530         if (rem)
 531                 return -EINVAL;
 532
 533         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 534         if (!probability || nla_len(probability) != sizeof(u32))
 535                 return -EINVAL;
 536
 537         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 538         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 539                 return -EINVAL;
 540         return validate_actions(actions, key, depth + 1);
 541 }
 542
 543 static int validate_set(const struct nlattr *a,
 544                         const struct sw_flow_key *flow_key)
 545 {
 546         const struct nlattr *ovs_key = nla_data(a);
 547         int key_type = nla_type(ovs_key);
 548
 549         /* There can be only one key in a action */
 550         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 551                 return -EINVAL;
 552
 553         if (key_type > OVS_KEY_ATTR_MAX ||
 554             nla_len(ovs_key) != ovs_key_lens[key_type])
 555                 return -EINVAL;
 556
 557         switch (key_type) {
 558         const struct ovs_key_ipv4 *ipv4_key;
 559
 560         case OVS_KEY_ATTR_PRIORITY:
 561         case OVS_KEY_ATTR_TUN_ID:
 562         case OVS_KEY_ATTR_ETHERNET:
 563                 break;
 564
 565         case OVS_KEY_ATTR_IPV4:
 566                 if (flow_key->eth.type != htons(ETH_P_IP))
 567                         return -EINVAL;
 568
 569                 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
 570                         return -EINVAL;
 571
 572                 ipv4_key = nla_data(ovs_key);
 573                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 574                         return -EINVAL;
 575
 576                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 577                         return -EINVAL;
 578
 579                 break;
 580
 581         case OVS_KEY_ATTR_TCP:
 582                 if (flow_key->ip.proto != IPPROTO_TCP)
 583                         return -EINVAL;
 584
 585                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 586                         return -EINVAL;
 587
 588                 break;
 589
 590         case OVS_KEY_ATTR_UDP:
 591                 if (flow_key->ip.proto != IPPROTO_UDP)
 592                         return -EINVAL;
 593
 594                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 595                         return -EINVAL;
 596                 break;
 597
 598         default:
 599                 return -EINVAL;
 600         }
 601
 602         return 0;
 603 }
 604
 605 static int validate_userspace(const struct nlattr *attr)
 606 {
 607         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 608                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 609                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
 610         };
 611         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 612         int error;
 613
 614         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 615                                  attr, userspace_policy);
 616         if (error)
 617                 return error;
 618
 619         if (!a[OVS_USERSPACE_ATTR_PID] ||
 620             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 621                 return -EINVAL;
 622
 623         return 0;
 624 }
 625
 626 static int validate_actions(const struct nlattr *attr,
 627                                 const struct sw_flow_key *key,  int depth)
 628 {
 629         const struct nlattr *a;
 630         int rem, err;
 631
 632         if (depth >= SAMPLE_ACTION_DEPTH)
 633                 return -EOVERFLOW;
 634
 635         nla_for_each_nested(a, attr, rem) {
 636                 /* Expected argument lengths, (u32)-1 for variable length. */
 637                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 638                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 639                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 640                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 641                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 642                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 643                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 644                 };
 645                 const struct ovs_action_push_vlan *vlan;
 646                 int type = nla_type(a);
 647
 648                 if (type > OVS_ACTION_ATTR_MAX ||
 649                     (action_lens[type] != nla_len(a) &&
 650                      action_lens[type] != (u32)-1))
 651                         return -EINVAL;
 652
 653                 switch (type) {
 654                 case OVS_ACTION_ATTR_UNSPEC:
 655                         return -EINVAL;
 656
 657                 case OVS_ACTION_ATTR_USERSPACE:
 658                         err = validate_userspace(a);
 659                         if (err)
 660                                 return err;
 661                         break;
 662
 663                 case OVS_ACTION_ATTR_OUTPUT:
 664                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 665                                 return -EINVAL;
 666                         break;
 667
 668
 669                 case OVS_ACTION_ATTR_POP_VLAN:
 670                         break;
 671
 672                 case OVS_ACTION_ATTR_PUSH_VLAN:
 673                         vlan = nla_data(a);
 674                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 675                                 return -EINVAL;
 676                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 677                                 return -EINVAL;
 678                         break;
 679
 680                 case OVS_ACTION_ATTR_SET:
 681                         err = validate_set(a, key);
 682                         if (err)
 683                                 return err;
 684                         break;
 685
 686                 case OVS_ACTION_ATTR_SAMPLE:
 687                         err = validate_sample(a, key, depth);
 688                         if (err)
 689                                 return err;
 690                         break;
 691
 692                 default:
 693                         return -EINVAL;
 694                 }
 695         }
 696
 697         if (rem > 0)
 698                 return -EINVAL;
 699
 700         return 0;
 701 }
 702
 703 static void clear_stats(struct sw_flow *flow)
 704 {
 705         flow->used = 0;
 706         flow->tcp_flags = 0;
 707         flow->packet_count = 0;
 708         flow->byte_count = 0;
 709 }
 710
 711 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 712 {
 713         struct ovs_header *ovs_header = info->userhdr;
 714         struct nlattr **a = info->attrs;
 715         struct sw_flow_actions *acts;
 716         struct sk_buff *packet;
 717         struct sw_flow *flow;
 718         struct datapath *dp;
 719         struct ethhdr *eth;
 720         int len;
 721         int err;
 722         int key_len;
 723
 724         err = -EINVAL;
 725         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 726             !a[OVS_PACKET_ATTR_ACTIONS] ||
 727             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 728                 goto err;
 729
 730         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 731         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 732         err = -ENOMEM;
 733         if (!packet)
 734                 goto err;
 735         skb_reserve(packet, NET_IP_ALIGN);
 736
 737         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 738
 739         skb_reset_mac_header(packet);
 740         eth = eth_hdr(packet);
 741
 742         /* Normally, setting the skb 'protocol' field would be handled by a
 743          * call to eth_type_trans(), but it assumes there's a sending
 744          * device, which we may not have. */
 745         if (ntohs(eth->h_proto) >= 1536)
 746                 packet->protocol = eth->h_proto;
 747         else
 748                 packet->protocol = htons(ETH_P_802_2);
 749
 750         /* Build an sw_flow for sending this packet. */
 751         flow = ovs_flow_alloc();
 752         err = PTR_ERR(flow);
 753         if (IS_ERR(flow))
 754                 goto err_kfree_skb;
 755
 756         err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
 757         if (err)
 758                 goto err_flow_put;
 759
 760         err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
 761                                              &flow->key.phy.in_port,
 762                                              &flow->key.phy.tun_id,
 763                                              a[OVS_PACKET_ATTR_KEY]);
 764         if (err)
 765                 goto err_flow_put;
 766
 767         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
 768         if (err)
 769                 goto err_flow_put;
 770
 771         flow->hash = ovs_flow_hash(&flow->key, key_len);
 772
 773         acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 774         err = PTR_ERR(acts);
 775         if (IS_ERR(acts))
 776                 goto err_flow_put;
 777         rcu_assign_pointer(flow->sf_acts, acts);
 778
 779         OVS_CB(packet)->flow = flow;
 780         packet->priority = flow->key.phy.priority;
 781
 782         rcu_read_lock();
 783         dp = get_dp(ovs_header->dp_ifindex);
 784         err = -ENODEV;
 785         if (!dp)
 786                 goto err_unlock;
 787
 788         local_bh_disable();
 789         err = ovs_execute_actions(dp, packet);
 790         local_bh_enable();
 791         rcu_read_unlock();
 792
 793         ovs_flow_put(flow);
 794         return err;
 795
 796 err_unlock:
 797         rcu_read_unlock();
 798 err_flow_put:
 799         ovs_flow_put(flow);
 800 err_kfree_skb:
 801         kfree_skb(packet);
 802 err:
 803         return err;
 804 }
 805
 806 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 807         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 808         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 809         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 810 };
 811
 812 static struct genl_ops dp_packet_genl_ops[] = {
 813         { .cmd = OVS_PACKET_CMD_EXECUTE,
 814           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 815           .policy = packet_policy,
 816           .doit = ovs_packet_cmd_execute
 817         }
 818 };
 819
 820 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 821 {
 822         int i;
 823         struct flow_table *table = genl_dereference(dp->table);
 824
 825         stats->n_flows = ovs_flow_tbl_count(table);
 826
 827         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 828         for_each_possible_cpu(i) {
 829                 const struct dp_stats_percpu *percpu_stats;
 830                 struct dp_stats_percpu local_stats;
 831                 unsigned int start;
 832
 833                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 834
 835                 do {
 836                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 837                         local_stats = *percpu_stats;
 838                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 839
 840                 stats->n_hit += local_stats.n_hit;
 841                 stats->n_missed += local_stats.n_missed;
 842                 stats->n_lost += local_stats.n_lost;
 843         }
 844 }
 845
 846 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 847         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 848         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 849         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 850 };
 851
 852 static struct genl_family dp_flow_genl_family = {
 853         .id = GENL_ID_GENERATE,
 854         .hdrsize = sizeof(struct ovs_header),
 855         .name = OVS_FLOW_FAMILY,
 856         .version = OVS_FLOW_VERSION,
 857         .maxattr = OVS_FLOW_ATTR_MAX
 858 };
 859
 860 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 861         .name = OVS_FLOW_MCGROUP
 862 };
 863
 864 /* Called with genl_lock. */
 865 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 866                                   struct sk_buff *skb, u32 pid,
 867                                   u32 seq, u32 flags, u8 cmd)
 868 {
 869         const int skb_orig_len = skb->len;
 870         const struct sw_flow_actions *sf_acts;
 871         struct ovs_flow_stats stats;
 872         struct ovs_header *ovs_header;
 873         struct nlattr *nla;
 874         unsigned long used;
 875         u8 tcp_flags;
 876         int err;
 877
 878         sf_acts = rcu_dereference_protected(flow->sf_acts,
 879                                             lockdep_genl_is_held());
 880
 881         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 882         if (!ovs_header)
 883                 return -EMSGSIZE;
 884
 885         ovs_header->dp_ifindex = get_dpifindex(dp);
 886
 887         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 888         if (!nla)
 889                 goto nla_put_failure;
 890         err = ovs_flow_to_nlattrs(&flow->key, skb);
 891         if (err)
 892                 goto error;
 893         nla_nest_end(skb, nla);
 894
 895         spin_lock_bh(&flow->lock);
 896         used = flow->used;
 897         stats.n_packets = flow->packet_count;
 898         stats.n_bytes = flow->byte_count;
 899         tcp_flags = flow->tcp_flags;
 900         spin_unlock_bh(&flow->lock);
 901
 902         if (used)
 903                 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used));
 904
 905         if (stats.n_packets)
 906                 NLA_PUT(skb, OVS_FLOW_ATTR_STATS,
 907                         sizeof(struct ovs_flow_stats), &stats);
 908
 909         if (tcp_flags)
 910                 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 911
 912         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 913          * this is the first flow to be dumped into 'skb'.  This is unusual for
 914          * Netlink but individual action lists can be longer than
 915          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 916          * The userspace caller can always fetch the actions separately if it
 917          * really wants them.  (Most userspace callers in fact don't care.)
 918          *
 919          * This can only fail for dump operations because the skb is always
 920          * properly sized for single flows.
 921          */
 922         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 923                       sf_acts->actions);
 924         if (err < 0 && skb_orig_len)
 925                 goto error;
 926
 927         return genlmsg_end(skb, ovs_header);
 928
 929 nla_put_failure:
 930         err = -EMSGSIZE;
 931 error:
 932         genlmsg_cancel(skb, ovs_header);
 933         return err;
 934 }
 935
 936 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 937 {
 938         const struct sw_flow_actions *sf_acts;
 939         int len;
 940
 941         sf_acts = rcu_dereference_protected(flow->sf_acts,
 942                                             lockdep_genl_is_held());
 943
 944         /* OVS_FLOW_ATTR_KEY */
 945         len = nla_total_size(FLOW_BUFSIZE);
 946         /* OVS_FLOW_ATTR_ACTIONS */
 947         len += nla_total_size(sf_acts->actions_len);
 948         /* OVS_FLOW_ATTR_STATS */
 949         len += nla_total_size(sizeof(struct ovs_flow_stats));
 950         /* OVS_FLOW_ATTR_TCP_FLAGS */
 951         len += nla_total_size(1);
 952         /* OVS_FLOW_ATTR_USED */
 953         len += nla_total_size(8);
 954
 955         len += NLMSG_ALIGN(sizeof(struct ovs_header));
 956
 957         return genlmsg_new(len, GFP_KERNEL);
 958 }
 959
 960 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 961                                                struct datapath *dp,
 962                                                u32 pid, u32 seq, u8 cmd)
 963 {
 964         struct sk_buff *skb;
 965         int retval;
 966
 967         skb = ovs_flow_cmd_alloc_info(flow);
 968         if (!skb)
 969                 return ERR_PTR(-ENOMEM);
 970
 971         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 972         BUG_ON(retval < 0);
 973         return skb;
 974 }
 975
 976 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 977 {
 978         struct nlattr **a = info->attrs;
 979         struct ovs_header *ovs_header = info->userhdr;
 980         struct sw_flow_key key;
 981         struct sw_flow *flow;
 982         struct sk_buff *reply;
 983         struct datapath *dp;
 984         struct flow_table *table;
 985         int error;
 986         int key_len;
 987
 988         /* Extract key. */
 989         error = -EINVAL;
 990         if (!a[OVS_FLOW_ATTR_KEY])
 991                 goto error;
 992         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
 993         if (error)
 994                 goto error;
 995
 996         /* Validate actions. */
 997         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 998                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
 999                 if (error)
1000                         goto error;
1001         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1002                 error = -EINVAL;
1003                 goto error;
1004         }
1005
1006         dp = get_dp(ovs_header->dp_ifindex);
1007         error = -ENODEV;
1008         if (!dp)
1009                 goto error;
1010
1011         table = genl_dereference(dp->table);
1012         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1013         if (!flow) {
1014                 struct sw_flow_actions *acts;
1015
1016                 /* Bail out if we're not allowed to create a new flow. */
1017                 error = -ENOENT;
1018                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1019                         goto error;
1020
1021                 /* Expand table, if necessary, to make room. */
1022                 if (ovs_flow_tbl_need_to_expand(table)) {
1023                         struct flow_table *new_table;
1024
1025                         new_table = ovs_flow_tbl_expand(table);
1026                         if (!IS_ERR(new_table)) {
1027                                 rcu_assign_pointer(dp->table, new_table);
1028                                 ovs_flow_tbl_deferred_destroy(table);
1029                                 table = genl_dereference(dp->table);
1030                         }
1031                 }
1032
1033                 /* Allocate flow. */
1034                 flow = ovs_flow_alloc();
1035                 if (IS_ERR(flow)) {
1036                         error = PTR_ERR(flow);
1037                         goto error;
1038                 }
1039                 flow->key = key;
1040                 clear_stats(flow);
1041
1042                 /* Obtain actions. */
1043                 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1044                 error = PTR_ERR(acts);
1045                 if (IS_ERR(acts))
1046                         goto error_free_flow;
1047                 rcu_assign_pointer(flow->sf_acts, acts);
1048
1049                 /* Put flow in bucket. */
1050                 flow->hash = ovs_flow_hash(&key, key_len);
1051                 ovs_flow_tbl_insert(table, flow);
1052
1053                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1054                                                 info->snd_seq,
1055                                                 OVS_FLOW_CMD_NEW);
1056         } else {
1057                 /* We found a matching flow. */
1058                 struct sw_flow_actions *old_acts;
1059                 struct nlattr *acts_attrs;
1060
1061                 /* Bail out if we're not allowed to modify an existing flow.
1062                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1063                  * because Generic Netlink treats the latter as a dump
1064                  * request.  We also accept NLM_F_EXCL in case that bug ever
1065                  * gets fixed.
1066                  */
1067                 error = -EEXIST;
1068                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1069                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1070                         goto error;
1071
1072                 /* Update actions. */
1073                 old_acts = rcu_dereference_protected(flow->sf_acts,
1074                                                      lockdep_genl_is_held());
1075                 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1076                 if (acts_attrs &&
1077                    (old_acts->actions_len != nla_len(acts_attrs) ||
1078                    memcmp(old_acts->actions, nla_data(acts_attrs),
1079                           old_acts->actions_len))) {
1080                         struct sw_flow_actions *new_acts;
1081
1082                         new_acts = ovs_flow_actions_alloc(acts_attrs);
1083                         error = PTR_ERR(new_acts);
1084                         if (IS_ERR(new_acts))
1085                                 goto error;
1086
1087                         rcu_assign_pointer(flow->sf_acts, new_acts);
1088                         ovs_flow_deferred_free_acts(old_acts);
1089                 }
1090
1091                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1092                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1093
1094                 /* Clear stats. */
1095                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1096                         spin_lock_bh(&flow->lock);
1097                         clear_stats(flow);
1098                         spin_unlock_bh(&flow->lock);
1099                 }
1100         }
1101
1102         if (!IS_ERR(reply))
1103                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1104                            ovs_dp_flow_multicast_group.id, info->nlhdr,
1105                            GFP_KERNEL);
1106         else
1107                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1108                                 ovs_dp_flow_multicast_group.id,
1109                                 PTR_ERR(reply));
1110         return 0;
1111
1112 error_free_flow:
1113         ovs_flow_put(flow);
1114 error:
1115         return error;
1116 }
1117
1118 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1119 {
1120         struct nlattr **a = info->attrs;
1121         struct ovs_header *ovs_header = info->userhdr;
1122         struct sw_flow_key key;
1123         struct sk_buff *reply;
1124         struct sw_flow *flow;
1125         struct datapath *dp;
1126         struct flow_table *table;
1127         int err;
1128         int key_len;
1129
1130         if (!a[OVS_FLOW_ATTR_KEY])
1131                 return -EINVAL;
1132         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1133         if (err)
1134                 return err;
1135
1136         dp = get_dp(ovs_header->dp_ifindex);
1137         if (!dp)
1138                 return -ENODEV;
1139
1140         table = genl_dereference(dp->table);
1141         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1142         if (!flow)
1143                 return -ENOENT;
1144
1145         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1146                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1147         if (IS_ERR(reply))
1148                 return PTR_ERR(reply);
1149
1150         return genlmsg_reply(reply, info);
1151 }
1152
1153 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1154 {
1155         struct nlattr **a = info->attrs;
1156         struct ovs_header *ovs_header = info->userhdr;
1157         struct sw_flow_key key;
1158         struct sk_buff *reply;
1159         struct sw_flow *flow;
1160         struct datapath *dp;
1161         struct flow_table *table;
1162         int err;
1163         int key_len;
1164
1165         if (!a[OVS_FLOW_ATTR_KEY])
1166                 return flush_flows(ovs_header->dp_ifindex);
1167         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1168         if (err)
1169                 return err;
1170
1171         dp = get_dp(ovs_header->dp_ifindex);
1172         if (!dp)
1173                 return -ENODEV;
1174
1175         table = genl_dereference(dp->table);
1176         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1177         if (!flow)
1178                 return -ENOENT;
1179
1180         reply = ovs_flow_cmd_alloc_info(flow);
1181         if (!reply)
1182                 return -ENOMEM;
1183
1184         ovs_flow_tbl_remove(table, flow);
1185
1186         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1187                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1188         BUG_ON(err < 0);
1189
1190         ovs_flow_deferred_free(flow);
1191
1192         genl_notify(reply, genl_info_net(info), info->snd_pid,
1193                     ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1194         return 0;
1195 }
1196
1197 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1198 {
1199         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1200         struct datapath *dp;
1201         struct flow_table *table;
1202
1203         dp = get_dp(ovs_header->dp_ifindex);
1204         if (!dp)
1205                 return -ENODEV;
1206
1207         table = genl_dereference(dp->table);
1208
1209         for (;;) {
1210                 struct sw_flow *flow;
1211                 u32 bucket, obj;
1212
1213                 bucket = cb->args[0];
1214                 obj = cb->args[1];
1215                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1216                 if (!flow)
1217                         break;
1218
1219                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1220                                            NETLINK_CB(cb->skb).pid,
1221                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1222                                            OVS_FLOW_CMD_NEW) < 0)
1223                         break;
1224
1225                 cb->args[0] = bucket;
1226                 cb->args[1] = obj;
1227         }
1228         return skb->len;
1229 }
1230
1231 static struct genl_ops dp_flow_genl_ops[] = {
1232         { .cmd = OVS_FLOW_CMD_NEW,
1233           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1234           .policy = flow_policy,
1235           .doit = ovs_flow_cmd_new_or_set
1236         },
1237         { .cmd = OVS_FLOW_CMD_DEL,
1238           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1239           .policy = flow_policy,
1240           .doit = ovs_flow_cmd_del
1241         },
1242         { .cmd = OVS_FLOW_CMD_GET,
1243           .flags = 0,               /* OK for unprivileged users. */
1244           .policy = flow_policy,
1245           .doit = ovs_flow_cmd_get,
1246           .dumpit = ovs_flow_cmd_dump
1247         },
1248         { .cmd = OVS_FLOW_CMD_SET,
1249           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1250           .policy = flow_policy,
1251           .doit = ovs_flow_cmd_new_or_set,
1252         },
1253 };
1254
1255 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1256 #ifdef HAVE_NLA_NUL_STRING
1257         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1258 #endif
1259         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1260 };
1261
1262 static struct genl_family dp_datapath_genl_family = {
1263         .id = GENL_ID_GENERATE,
1264         .hdrsize = sizeof(struct ovs_header),
1265         .name = OVS_DATAPATH_FAMILY,
1266         .version = OVS_DATAPATH_VERSION,
1267         .maxattr = OVS_DP_ATTR_MAX
1268 };
1269
1270 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1271         .name = OVS_DATAPATH_MCGROUP
1272 };
1273
1274 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1275                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1276 {
1277         struct ovs_header *ovs_header;
1278         struct ovs_dp_stats dp_stats;
1279         int err;
1280
1281         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1282                                    flags, cmd);
1283         if (!ovs_header)
1284                 goto error;
1285
1286         ovs_header->dp_ifindex = get_dpifindex(dp);
1287
1288         rcu_read_lock();
1289         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1290         rcu_read_unlock();
1291         if (err)
1292                 goto nla_put_failure;
1293
1294         get_dp_stats(dp, &dp_stats);
1295         NLA_PUT(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats);
1296
1297         return genlmsg_end(skb, ovs_header);
1298
1299 nla_put_failure:
1300         genlmsg_cancel(skb, ovs_header);
1301 error:
1302         return -EMSGSIZE;
1303 }
1304
1305 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1306                                              u32 seq, u8 cmd)
1307 {
1308         struct sk_buff *skb;
1309         int retval;
1310
1311         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1312         if (!skb)
1313                 return ERR_PTR(-ENOMEM);
1314
1315         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1316         if (retval < 0) {
1317                 kfree_skb(skb);
1318                 return ERR_PTR(retval);
1319         }
1320         return skb;
1321 }
1322
1323 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1324 {
1325         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1326 }
1327
1328 /* Called with genl_mutex and optionally with RTNL lock also. */
1329 static struct datapath *lookup_datapath(struct ovs_header *ovs_header,
1330                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1331 {
1332         struct datapath *dp;
1333
1334         if (!a[OVS_DP_ATTR_NAME])
1335                 dp = get_dp(ovs_header->dp_ifindex);
1336         else {
1337                 struct vport *vport;
1338
1339                 rcu_read_lock();
1340                 vport = ovs_vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1341                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1342                 rcu_read_unlock();
1343         }
1344         return dp ? dp : ERR_PTR(-ENODEV);
1345 }
1346
1347 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1348 {
1349         struct nlattr **a = info->attrs;
1350         struct vport_parms parms;
1351         struct sk_buff *reply;
1352         struct datapath *dp;
1353         struct vport *vport;
1354         int err;
1355
1356         err = -EINVAL;
1357         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1358                 goto err;
1359
1360         err = ovs_dp_cmd_validate(a);
1361         if (err)
1362                 goto err;
1363
1364         rtnl_lock();
1365         err = -ENODEV;
1366         if (!try_module_get(THIS_MODULE))
1367                 goto err_unlock_rtnl;
1368
1369         err = -ENOMEM;
1370         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1371         if (dp == NULL)
1372                 goto err_put_module;
1373         INIT_LIST_HEAD(&dp->port_list);
1374
1375         /* Initialize kobject for bridge.  This will be added as
1376          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1377         dp->ifobj.kset = NULL;
1378         kobject_init(&dp->ifobj, &dp_ktype);
1379
1380         /* Allocate table. */
1381         err = -ENOMEM;
1382         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1383         if (!dp->table)
1384                 goto err_free_dp;
1385
1386         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1387         if (!dp->stats_percpu) {
1388                 err = -ENOMEM;
1389                 goto err_destroy_table;
1390         }
1391
1392         /* Set up our datapath device. */
1393         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1394         parms.type = OVS_VPORT_TYPE_INTERNAL;
1395         parms.options = NULL;
1396         parms.dp = dp;
1397         parms.port_no = OVSP_LOCAL;
1398         parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1399
1400         vport = new_vport(&parms);
1401         if (IS_ERR(vport)) {
1402                 err = PTR_ERR(vport);
1403                 if (err == -EBUSY)
1404                         err = -EEXIST;
1405
1406                 goto err_destroy_percpu;
1407         }
1408
1409         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1410                                       info->snd_seq, OVS_DP_CMD_NEW);
1411         err = PTR_ERR(reply);
1412         if (IS_ERR(reply))
1413                 goto err_destroy_local_port;
1414
1415         list_add_tail(&dp->list_node, &dps);
1416         ovs_dp_sysfs_add_dp(dp);
1417
1418         rtnl_unlock();
1419
1420         genl_notify(reply, genl_info_net(info), info->snd_pid,
1421                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1422                     GFP_KERNEL);
1423         return 0;
1424
1425 err_destroy_local_port:
1426         ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL]));
1427 err_destroy_percpu:
1428         free_percpu(dp->stats_percpu);
1429 err_destroy_table:
1430         ovs_flow_tbl_destroy(genl_dereference(dp->table));
1431 err_free_dp:
1432         kfree(dp);
1433 err_put_module:
1434         module_put(THIS_MODULE);
1435 err_unlock_rtnl:
1436         rtnl_unlock();
1437 err:
1438         return err;
1439 }
1440
1441 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1442 {
1443         struct vport *vport, *next_vport;
1444         struct sk_buff *reply;
1445         struct datapath *dp;
1446         int err;
1447
1448         err = ovs_dp_cmd_validate(info->attrs);
1449         if (err)
1450                 goto exit;
1451
1452         rtnl_lock();
1453         dp = lookup_datapath(info->userhdr, info->attrs);
1454         err = PTR_ERR(dp);
1455         if (IS_ERR(dp))
1456                 goto exit_unlock;
1457
1458         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1459                                       info->snd_seq, OVS_DP_CMD_DEL);
1460         err = PTR_ERR(reply);
1461         if (IS_ERR(reply))
1462                 goto exit_unlock;
1463
1464         list_for_each_entry_safe(vport, next_vport, &dp->port_list, node)
1465                 if (vport->port_no != OVSP_LOCAL)
1466                         ovs_dp_detach_port(vport);
1467
1468         ovs_dp_sysfs_del_dp(dp);
1469         list_del(&dp->list_node);
1470         ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL]));
1471
1472         /* rtnl_unlock() will wait until all the references to devices that
1473          * are pending unregistration have been dropped.  We do it here to
1474          * ensure that any internal devices (which contain DP pointers) are
1475          * fully destroyed before freeing the datapath.
1476          */
1477         rtnl_unlock();
1478
1479         call_rcu(&dp->rcu, destroy_dp_rcu);
1480         module_put(THIS_MODULE);
1481
1482         genl_notify(reply, genl_info_net(info), info->snd_pid,
1483                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1484                     GFP_KERNEL);
1485
1486         return 0;
1487
1488 exit_unlock:
1489         rtnl_unlock();
1490 exit:
1491         return err;
1492 }
1493
1494 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1495 {
1496         struct sk_buff *reply;
1497         struct datapath *dp;
1498         int err;
1499
1500         err = ovs_dp_cmd_validate(info->attrs);
1501         if (err)
1502                 return err;
1503
1504         dp = lookup_datapath(info->userhdr, info->attrs);
1505         if (IS_ERR(dp))
1506                 return PTR_ERR(dp);
1507
1508         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1509                                       info->snd_seq, OVS_DP_CMD_NEW);
1510         if (IS_ERR(reply)) {
1511                 err = PTR_ERR(reply);
1512                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1513                                 ovs_dp_datapath_multicast_group.id, err);
1514                 return 0;
1515         }
1516
1517         genl_notify(reply, genl_info_net(info), info->snd_pid,
1518                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1519                     GFP_KERNEL);
1520
1521         return 0;
1522 }
1523
1524 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1525 {
1526         struct sk_buff *reply;
1527         struct datapath *dp;
1528         int err;
1529
1530         err = ovs_dp_cmd_validate(info->attrs);
1531         if (err)
1532                 return err;
1533
1534         dp = lookup_datapath(info->userhdr, info->attrs);
1535         if (IS_ERR(dp))
1536                 return PTR_ERR(dp);
1537
1538         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1539                                       info->snd_seq, OVS_DP_CMD_NEW);
1540         if (IS_ERR(reply))
1541                 return PTR_ERR(reply);
1542
1543         return genlmsg_reply(reply, info);
1544 }
1545
1546 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1547 {
1548         struct datapath *dp;
1549         int skip = cb->args[0];
1550         int i = 0;
1551
1552         list_for_each_entry(dp, &dps, list_node) {
1553                 if (i >= skip &&
1554                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1555                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1556                                          OVS_DP_CMD_NEW) < 0)
1557                         break;
1558                 i++;
1559         }
1560
1561         cb->args[0] = i;
1562
1563         return skb->len;
1564 }
1565
1566 static struct genl_ops dp_datapath_genl_ops[] = {
1567         { .cmd = OVS_DP_CMD_NEW,
1568           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1569           .policy = datapath_policy,
1570           .doit = ovs_dp_cmd_new
1571         },
1572         { .cmd = OVS_DP_CMD_DEL,
1573           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1574           .policy = datapath_policy,
1575           .doit = ovs_dp_cmd_del
1576         },
1577         { .cmd = OVS_DP_CMD_GET,
1578           .flags = 0,               /* OK for unprivileged users. */
1579           .policy = datapath_policy,
1580           .doit = ovs_dp_cmd_get,
1581           .dumpit = ovs_dp_cmd_dump
1582         },
1583         { .cmd = OVS_DP_CMD_SET,
1584           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1585           .policy = datapath_policy,
1586           .doit = ovs_dp_cmd_set,
1587         },
1588 };
1589
1590 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1591 #ifdef HAVE_NLA_NUL_STRING
1592         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1593         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1594         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1595 #else
1596         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1597         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1598 #endif
1599         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1600         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1601         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1602         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1603 };
1604
1605 static struct genl_family dp_vport_genl_family = {
1606         .id = GENL_ID_GENERATE,
1607         .hdrsize = sizeof(struct ovs_header),
1608         .name = OVS_VPORT_FAMILY,
1609         .version = OVS_VPORT_VERSION,
1610         .maxattr = OVS_VPORT_ATTR_MAX
1611 };
1612
1613 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1614         .name = OVS_VPORT_MCGROUP
1615 };
1616
1617 /* Called with RTNL lock or RCU read lock. */
1618 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1619                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1620 {
1621         struct ovs_header *ovs_header;
1622         struct ovs_vport_stats vport_stats;
1623         int err;
1624
1625         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1626                                  flags, cmd);
1627         if (!ovs_header)
1628                 return -EMSGSIZE;
1629
1630         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1631
1632         NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1633         NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type);
1634         NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport));
1635         NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1636
1637         ovs_vport_get_stats(vport, &vport_stats);
1638         NLA_PUT(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1639                 &vport_stats);
1640
1641         NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN,
1642                 vport->ops->get_addr(vport));
1643
1644         err = ovs_vport_get_options(vport, skb);
1645         if (err == -EMSGSIZE)
1646                 goto error;
1647
1648         return genlmsg_end(skb, ovs_header);
1649
1650 nla_put_failure:
1651         err = -EMSGSIZE;
1652 error:
1653         genlmsg_cancel(skb, ovs_header);
1654         return err;
1655 }
1656
1657 /* Called with RTNL lock or RCU read lock. */
1658 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1659                                          u32 seq, u8 cmd)
1660 {
1661         struct sk_buff *skb;
1662         int retval;
1663
1664         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1665         if (!skb)
1666                 return ERR_PTR(-ENOMEM);
1667
1668         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1669         if (retval < 0) {
1670                 kfree_skb(skb);
1671                 return ERR_PTR(retval);
1672         }
1673         return skb;
1674 }
1675
1676 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1677 {
1678         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1679 }
1680
1681 /* Called with RTNL lock or RCU read lock. */
1682 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1683                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1684 {
1685         struct datapath *dp;
1686         struct vport *vport;
1687
1688         if (a[OVS_VPORT_ATTR_NAME]) {
1689                 vport = ovs_vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1690                 if (!vport)
1691                         return ERR_PTR(-ENODEV);
1692                 if (ovs_header->dp_ifindex &&
1693                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1694                         return ERR_PTR(-ENODEV);
1695                 return vport;
1696         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1697                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1698
1699                 if (port_no >= DP_MAX_PORTS)
1700                         return ERR_PTR(-EFBIG);
1701
1702                 dp = get_dp(ovs_header->dp_ifindex);
1703                 if (!dp)
1704                         return ERR_PTR(-ENODEV);
1705
1706                 vport = rcu_dereference_rtnl(dp->ports[port_no]);
1707                 if (!vport)
1708                         return ERR_PTR(-ENOENT);
1709                 return vport;
1710         } else
1711                 return ERR_PTR(-EINVAL);
1712 }
1713
1714 /* Called with RTNL lock. */
1715 static int change_vport(struct vport *vport,
1716                         struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1717 {
1718         int err = 0;
1719
1720         if (a[OVS_VPORT_ATTR_STATS])
1721                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1722
1723         if (a[OVS_VPORT_ATTR_ADDRESS])
1724                 err = ovs_vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1725
1726         return err;
1727 }
1728
1729 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1730 {
1731         struct nlattr **a = info->attrs;
1732         struct ovs_header *ovs_header = info->userhdr;
1733         struct vport_parms parms;
1734         struct sk_buff *reply;
1735         struct vport *vport;
1736         struct datapath *dp;
1737         u32 port_no;
1738         int err;
1739
1740         err = -EINVAL;
1741         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1742             !a[OVS_VPORT_ATTR_UPCALL_PID])
1743                 goto exit;
1744
1745         err = ovs_vport_cmd_validate(a);
1746         if (err)
1747                 goto exit;
1748
1749         rtnl_lock();
1750         dp = get_dp(ovs_header->dp_ifindex);
1751         err = -ENODEV;
1752         if (!dp)
1753                 goto exit_unlock;
1754
1755         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1756                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1757
1758                 err = -EFBIG;
1759                 if (port_no >= DP_MAX_PORTS)
1760                         goto exit_unlock;
1761
1762                 vport = rtnl_dereference(dp->ports[port_no]);
1763                 err = -EBUSY;
1764                 if (vport)
1765                         goto exit_unlock;
1766         } else {
1767                 for (port_no = 1; ; port_no++) {
1768                         if (port_no >= DP_MAX_PORTS) {
1769                                 err = -EFBIG;
1770                                 goto exit_unlock;
1771                         }
1772                         vport = rtnl_dereference(dp->ports[port_no]);
1773                         if (!vport)
1774                                 break;
1775                 }
1776         }
1777
1778         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1779         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1780         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1781         parms.dp = dp;
1782         parms.port_no = port_no;
1783         parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1784
1785         vport = new_vport(&parms);
1786         err = PTR_ERR(vport);
1787         if (IS_ERR(vport))
1788                 goto exit_unlock;
1789
1790         ovs_dp_sysfs_add_if(vport);
1791
1792         err = change_vport(vport, a);
1793         if (!err) {
1794                 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1795                                                  info->snd_seq,
1796                                                  OVS_VPORT_CMD_NEW);
1797                 if (IS_ERR(reply))
1798                         err = PTR_ERR(reply);
1799         }
1800         if (err) {
1801                 ovs_dp_detach_port(vport);
1802                 goto exit_unlock;
1803         }
1804         genl_notify(reply, genl_info_net(info), info->snd_pid,
1805                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1806
1807 exit_unlock:
1808         rtnl_unlock();
1809 exit:
1810         return err;
1811 }
1812
1813 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1814 {
1815         struct nlattr **a = info->attrs;
1816         struct sk_buff *reply;
1817         struct vport *vport;
1818         int err;
1819
1820         err = ovs_vport_cmd_validate(a);
1821         if (err)
1822                 goto exit;
1823
1824         rtnl_lock();
1825         vport = lookup_vport(info->userhdr, a);
1826         err = PTR_ERR(vport);
1827         if (IS_ERR(vport))
1828                 goto exit_unlock;
1829
1830         err = 0;
1831         if (a[OVS_VPORT_ATTR_TYPE] &&
1832             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1833                 err = -EINVAL;
1834
1835         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1836                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1837         if (!err)
1838                 err = change_vport(vport, a);
1839         else
1840                 goto exit_unlock;
1841         if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1842                 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1843
1844         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1845                                          OVS_VPORT_CMD_NEW);
1846         if (IS_ERR(reply)) {
1847                 err = PTR_ERR(reply);
1848                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1849                                 ovs_dp_vport_multicast_group.id, err);
1850                 return 0;
1851         }
1852
1853         genl_notify(reply, genl_info_net(info), info->snd_pid,
1854                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1855
1856 exit_unlock:
1857         rtnl_unlock();
1858 exit:
1859         return err;
1860 }
1861
1862 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1863 {
1864         struct nlattr **a = info->attrs;
1865         struct sk_buff *reply;
1866         struct vport *vport;
1867         int err;
1868
1869         err = ovs_vport_cmd_validate(a);
1870         if (err)
1871                 goto exit;
1872
1873         rtnl_lock();
1874         vport = lookup_vport(info->userhdr, a);
1875         err = PTR_ERR(vport);
1876         if (IS_ERR(vport))
1877                 goto exit_unlock;
1878
1879         if (vport->port_no == OVSP_LOCAL) {
1880                 err = -EINVAL;
1881                 goto exit_unlock;
1882         }
1883
1884         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1885                                          OVS_VPORT_CMD_DEL);
1886         err = PTR_ERR(reply);
1887         if (IS_ERR(reply))
1888                 goto exit_unlock;
1889
1890         ovs_dp_detach_port(vport);
1891
1892         genl_notify(reply, genl_info_net(info), info->snd_pid,
1893                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1894
1895 exit_unlock:
1896         rtnl_unlock();
1897 exit:
1898         return err;
1899 }
1900
1901 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1902 {
1903         struct nlattr **a = info->attrs;
1904         struct ovs_header *ovs_header = info->userhdr;
1905         struct sk_buff *reply;
1906         struct vport *vport;
1907         int err;
1908
1909         err = ovs_vport_cmd_validate(a);
1910         if (err)
1911                 goto exit;
1912
1913         rcu_read_lock();
1914         vport = lookup_vport(ovs_header, a);
1915         err = PTR_ERR(vport);
1916         if (IS_ERR(vport))
1917                 goto exit_unlock;
1918
1919         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1920                                          OVS_VPORT_CMD_NEW);
1921         err = PTR_ERR(reply);
1922         if (IS_ERR(reply))
1923                 goto exit_unlock;
1924
1925         rcu_read_unlock();
1926
1927         return genlmsg_reply(reply, info);
1928
1929 exit_unlock:
1930         rcu_read_unlock();
1931 exit:
1932         return err;
1933 }
1934
1935 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1936 {
1937         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1938         struct datapath *dp;
1939         u32 port_no;
1940         int retval;
1941
1942         dp = get_dp(ovs_header->dp_ifindex);
1943         if (!dp)
1944                 return -ENODEV;
1945
1946         rcu_read_lock();
1947         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1948                 struct vport *vport;
1949
1950                 vport = rcu_dereference(dp->ports[port_no]);
1951                 if (!vport)
1952                         continue;
1953
1954                 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1955                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1956                                             OVS_VPORT_CMD_NEW) < 0)
1957                         break;
1958         }
1959         rcu_read_unlock();
1960
1961         cb->args[0] = port_no;
1962         retval = skb->len;
1963
1964         return retval;
1965 }
1966
1967 static struct genl_ops dp_vport_genl_ops[] = {
1968         { .cmd = OVS_VPORT_CMD_NEW,
1969           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1970           .policy = vport_policy,
1971           .doit = ovs_vport_cmd_new
1972         },
1973         { .cmd = OVS_VPORT_CMD_DEL,
1974           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1975           .policy = vport_policy,
1976           .doit = ovs_vport_cmd_del
1977         },
1978         { .cmd = OVS_VPORT_CMD_GET,
1979           .flags = 0,               /* OK for unprivileged users. */
1980           .policy = vport_policy,
1981           .doit = ovs_vport_cmd_get,
1982           .dumpit = ovs_vport_cmd_dump
1983         },
1984         { .cmd = OVS_VPORT_CMD_SET,
1985           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1986           .policy = vport_policy,
1987           .doit = ovs_vport_cmd_set,
1988         },
1989 };
1990
1991 struct genl_family_and_ops {
1992         struct genl_family *family;
1993         struct genl_ops *ops;
1994         int n_ops;
1995         struct genl_multicast_group *group;
1996 };
1997
1998 static const struct genl_family_and_ops dp_genl_families[] = {
1999         { &dp_datapath_genl_family,
2000           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2001           &ovs_dp_datapath_multicast_group },
2002         { &dp_vport_genl_family,
2003           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2004           &ovs_dp_vport_multicast_group },
2005         { &dp_flow_genl_family,
2006           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2007           &ovs_dp_flow_multicast_group },
2008         { &dp_packet_genl_family,
2009           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2010           NULL },
2011 };
2012
2013 static void dp_unregister_genl(int n_families)
2014 {
2015         int i;
2016
2017         for (i = 0; i < n_families; i++)
2018                 genl_unregister_family(dp_genl_families[i].family);
2019 }
2020
2021 static int dp_register_genl(void)
2022 {
2023         int n_registered;
2024         int err;
2025         int i;
2026
2027         n_registered = 0;
2028         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2029                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2030
2031                 err = genl_register_family_with_ops(f->family, f->ops,
2032                                                     f->n_ops);
2033                 if (err)
2034                         goto error;
2035                 n_registered++;
2036
2037                 if (f->group) {
2038                         err = genl_register_mc_group(f->family, f->group);
2039                         if (err)
2040                                 goto error;
2041                 }
2042         }
2043
2044         return 0;
2045
2046 error:
2047         dp_unregister_genl(n_registered);
2048         return err;
2049 }
2050
2051 static int __rehash_flow_table(void *dummy)
2052 {
2053         struct datapath *dp;
2054
2055         list_for_each_entry(dp, &dps, list_node) {
2056                 struct flow_table *old_table = genl_dereference(dp->table);
2057                 struct flow_table *new_table;
2058
2059                 new_table = ovs_flow_tbl_rehash(old_table);
2060                 if (!IS_ERR(new_table)) {
2061                         rcu_assign_pointer(dp->table, new_table);
2062                         ovs_flow_tbl_deferred_destroy(old_table);
2063                 }
2064         }
2065         return 0;
2066 }
2067
2068 static void rehash_flow_table(struct work_struct *work)
2069 {
2070         genl_exec(__rehash_flow_table, NULL);
2071         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2072 }
2073
2074 static int __init dp_init(void)
2075 {
2076         struct sk_buff *dummy_skb;
2077         int err;
2078
2079         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2080
2081         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2082                 VERSION BUILDNR);
2083
2084         err = genl_exec_init();
2085         if (err)
2086                 goto error;
2087
2088         err = ovs_workqueues_init();
2089         if (err)
2090                 goto error_genl_exec;
2091
2092         err = ovs_tnl_init();
2093         if (err)
2094                 goto error_wq;
2095
2096         err = ovs_flow_init();
2097         if (err)
2098                 goto error_tnl_exit;
2099
2100         err = ovs_vport_init();
2101         if (err)
2102                 goto error_flow_exit;
2103
2104         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2105         if (err)
2106                 goto error_vport_exit;
2107
2108         err = dp_register_genl();
2109         if (err < 0)
2110                 goto error_unreg_notifier;
2111
2112         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2113
2114         return 0;
2115
2116 error_unreg_notifier:
2117         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2118 error_vport_exit:
2119         ovs_vport_exit();
2120 error_flow_exit:
2121         ovs_flow_exit();
2122 error_tnl_exit:
2123         ovs_tnl_exit();
2124 error_wq:
2125         ovs_workqueues_exit();
2126 error_genl_exec:
2127         genl_exec_exit();
2128 error:
2129         return err;
2130 }
2131
2132 static void dp_cleanup(void)
2133 {
2134         cancel_delayed_work_sync(&rehash_flow_wq);
2135         rcu_barrier();
2136         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2137         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2138         ovs_vport_exit();
2139         ovs_flow_exit();
2140         ovs_tnl_exit();
2141         ovs_workqueues_exit();
2142         genl_exec_exit();
2143 }
2144
2145 module_init(dp_init);
2146 module_exit(dp_cleanup);
2147
2148 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2149 MODULE_LICENSE("GPL");