datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
   3  * Distributed under the terms of the GNU GPL version 2.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 /* Functions for managing the dp interface/device. */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/init.h>
  14 #include <linux/module.h>
  15 #include <linux/if_arp.h>
  16 #include <linux/if_vlan.h>
  17 #include <linux/in.h>
  18 #include <linux/ip.h>
  19 #include <linux/jhash.h>
  20 #include <linux/delay.h>
  21 #include <linux/time.h>
  22 #include <linux/etherdevice.h>
  23 #include <linux/genetlink.h>
  24 #include <linux/kernel.h>
  25 #include <linux/kthread.h>
  26 #include <linux/mutex.h>
  27 #include <linux/percpu.h>
  28 #include <linux/rcupdate.h>
  29 #include <linux/tcp.h>
  30 #include <linux/udp.h>
  31 #include <linux/version.h>
  32 #include <linux/ethtool.h>
  33 #include <linux/wait.h>
  34 #include <asm/system.h>
  35 #include <asm/div64.h>
  36 #include <linux/highmem.h>
  37 #include <linux/netfilter_bridge.h>
  38 #include <linux/netfilter_ipv4.h>
  39 #include <linux/inetdevice.h>
  40 #include <linux/list.h>
  41 #include <linux/openvswitch.h>
  42 #include <linux/rculist.h>
  43 #include <linux/dmi.h>
  44 #include <net/inet_ecn.h>
  45 #include <net/genetlink.h>
  46
  47 #include "checksum.h"
  48 #include "datapath.h"
  49 #include "actions.h"
  50 #include "flow.h"
  51 #include "vlan.h"
  52 #include "tunnel.h"
  53 #include "vport-internal_dev.h"
  54
  55 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
  56     LINUX_VERSION_CODE >= KERNEL_VERSION(3,2,0)
  57 #error Kernels before 2.6.18 or after 3.1 are not supported by this version of Open vSwitch.
  58 #endif
  59
  60 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
  61 EXPORT_SYMBOL(dp_ioctl_hook);
  62
  63 /**
  64  * DOC: Locking:
  65  *
  66  * Writes to device state (add/remove datapath, port, set operations on vports,
  67  * etc.) are protected by RTNL.
  68  *
  69  * Writes to other state (flow table modifications, set miscellaneous datapath
  70  * parameters, etc.) are protected by genl_mutex.  The RTNL lock nests inside
  71  * genl_mutex.
  72  *
  73  * Reads are protected by RCU.
  74  *
  75  * There are a few special cases (mostly stats) that have their own
  76  * synchronization but they nest under all of above and don't interact with
  77  * each other.
  78  */
  79
  80 /* Global list of datapaths to enable dumping them all out.
  81  * Protected by genl_mutex.
  82  */
  83 static LIST_HEAD(dps);
  84
  85 static struct vport *new_vport(const struct vport_parms *);
  86 static int queue_gso_packets(int dp_ifindex, struct sk_buff *,
  87                              const struct dp_upcall_info *);
  88 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *,
  89                                   const struct dp_upcall_info *);
  90
  91 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  92 struct datapath *get_dp(int dp_ifindex)
  93 {
  94         struct datapath *dp = NULL;
  95         struct net_device *dev;
  96
  97         rcu_read_lock();
  98         dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
  99         if (dev) {
 100                 struct vport *vport = internal_dev_get_vport(dev);
 101                 if (vport)
 102                         dp = vport->dp;
 103         }
 104         rcu_read_unlock();
 105
 106         return dp;
 107 }
 108 EXPORT_SYMBOL_GPL(get_dp);
 109
 110 /* Must be called with genl_mutex. */
 111 static struct flow_table *get_table_protected(struct datapath *dp)
 112 {
 113         return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
 114 }
 115
 116 /* Must be called with rcu_read_lock or RTNL lock. */
 117 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
 118 {
 119         return rcu_dereference_rtnl(dp->ports[port_no]);
 120 }
 121
 122 /* Must be called with rcu_read_lock or RTNL lock. */
 123 const char *dp_name(const struct datapath *dp)
 124 {
 125         return vport_get_name(rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]));
 126 }
 127
 128 static int get_dpifindex(struct datapath *dp)
 129 {
 130         struct vport *local;
 131         int ifindex;
 132
 133         rcu_read_lock();
 134
 135         local = get_vport_protected(dp, OVSP_LOCAL);
 136         if (local)
 137                 ifindex = vport_get_ifindex(local);
 138         else
 139                 ifindex = 0;
 140
 141         rcu_read_unlock();
 142
 143         return ifindex;
 144 }
 145
 146 static size_t br_nlmsg_size(void)
 147 {
 148         return NLMSG_ALIGN(sizeof(struct ifinfomsg))
 149                + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
 150                + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
 151                + nla_total_size(4) /* IFLA_MASTER */
 152                + nla_total_size(4) /* IFLA_MTU */
 153                + nla_total_size(1); /* IFLA_OPERSTATE */
 154 }
 155
 156 /* Caller must hold RTNL lock. */
 157 static int dp_fill_ifinfo(struct sk_buff *skb,
 158                           const struct vport *port,
 159                           int event, unsigned int flags)
 160 {
 161         struct datapath *dp = port->dp;
 162         int ifindex = vport_get_ifindex(port);
 163         struct ifinfomsg *hdr;
 164         struct nlmsghdr *nlh;
 165
 166         if (ifindex < 0)
 167                 return ifindex;
 168
 169         nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
 170         if (nlh == NULL)
 171                 return -EMSGSIZE;
 172
 173         hdr = nlmsg_data(nlh);
 174         hdr->ifi_family = AF_BRIDGE;
 175         hdr->__ifi_pad = 0;
 176         hdr->ifi_type = ARPHRD_ETHER;
 177         hdr->ifi_index = ifindex;
 178         hdr->ifi_flags = vport_get_flags(port);
 179         hdr->ifi_change = 0;
 180
 181         NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
 182         NLA_PUT_U32(skb, IFLA_MASTER, get_dpifindex(dp));
 183         NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
 184 #ifdef IFLA_OPERSTATE
 185         NLA_PUT_U8(skb, IFLA_OPERSTATE,
 186                    vport_is_running(port)
 187                         ? vport_get_operstate(port)
 188                         : IF_OPER_DOWN);
 189 #endif
 190
 191         NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
 192
 193         return nlmsg_end(skb, nlh);
 194
 195 nla_put_failure:
 196         nlmsg_cancel(skb, nlh);
 197         return -EMSGSIZE;
 198 }
 199
 200 /* Caller must hold RTNL lock. */
 201 static void dp_ifinfo_notify(int event, struct vport *port)
 202 {
 203         struct sk_buff *skb;
 204         int err = -ENOBUFS;
 205
 206         skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
 207         if (skb == NULL)
 208                 goto errout;
 209
 210         err = dp_fill_ifinfo(skb, port, event, 0);
 211         if (err < 0) {
 212                 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
 213                 WARN_ON(err == -EMSGSIZE);
 214                 kfree_skb(skb);
 215                 goto errout;
 216         }
 217         rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 218         return;
 219 errout:
 220         if (err < 0)
 221                 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
 222 }
 223
 224 static void release_dp(struct kobject *kobj)
 225 {
 226         struct datapath *dp = container_of(kobj, struct datapath, ifobj);
 227         kfree(dp);
 228 }
 229
 230 static struct kobj_type dp_ktype = {
 231         .release = release_dp
 232 };
 233
 234 static void destroy_dp_rcu(struct rcu_head *rcu)
 235 {
 236         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 237
 238         flow_tbl_destroy(dp->table);
 239         free_percpu(dp->stats_percpu);
 240         kobject_put(&dp->ifobj);
 241 }
 242
 243 /* Called with RTNL lock and genl_lock. */
 244 static struct vport *new_vport(const struct vport_parms *parms)
 245 {
 246         struct vport *vport;
 247
 248         vport = vport_add(parms);
 249         if (!IS_ERR(vport)) {
 250                 struct datapath *dp = parms->dp;
 251
 252                 rcu_assign_pointer(dp->ports[parms->port_no], vport);
 253                 list_add(&vport->node, &dp->port_list);
 254
 255                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 256         }
 257
 258         return vport;
 259 }
 260
 261 /* Called with RTNL lock. */
 262 void dp_detach_port(struct vport *p)
 263 {
 264         ASSERT_RTNL();
 265
 266         if (p->port_no != OVSP_LOCAL)
 267                 dp_sysfs_del_if(p);
 268         dp_ifinfo_notify(RTM_DELLINK, p);
 269
 270         /* First drop references to device. */
 271         list_del(&p->node);
 272         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 273
 274         /* Then destroy it. */
 275         vport_del(p);
 276 }
 277
 278 /* Must be called with rcu_read_lock. */
 279 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 280 {
 281         struct datapath *dp = p->dp;
 282         struct sw_flow *flow;
 283         struct dp_stats_percpu *stats;
 284         u64 *stats_counter;
 285         int error;
 286
 287         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 288
 289         if (!OVS_CB(skb)->flow) {
 290                 struct sw_flow_key key;
 291                 int key_len;
 292
 293                 /* Extract flow from 'skb' into 'key'. */
 294                 error = flow_extract(skb, p->port_no, &key, &key_len);
 295                 if (unlikely(error)) {
 296                         kfree_skb(skb);
 297                         return;
 298                 }
 299
 300                 /* Look up flow. */
 301                 flow = flow_tbl_lookup(rcu_dereference(dp->table),
 302                                        &key, key_len);
 303                 if (unlikely(!flow)) {
 304                         struct dp_upcall_info upcall;
 305
 306                         upcall.cmd = OVS_PACKET_CMD_MISS;
 307                         upcall.key = &key;
 308                         upcall.userdata = NULL;
 309                         upcall.pid = p->upcall_pid;
 310                         dp_upcall(dp, skb, &upcall);
 311                         consume_skb(skb);
 312                         stats_counter = &stats->n_missed;
 313                         goto out;
 314                 }
 315
 316                 OVS_CB(skb)->flow = flow;
 317         }
 318
 319         stats_counter = &stats->n_hit;
 320         flow_used(OVS_CB(skb)->flow, skb);
 321         execute_actions(dp, skb);
 322
 323 out:
 324         /* Update datapath statistics. */
 325
 326         write_seqcount_begin(&stats->seqlock);
 327         (*stats_counter)++;
 328         write_seqcount_end(&stats->seqlock);
 329 }
 330
 331 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
 332 {
 333         u16 csum_start, csum_offset;
 334         __wsum csum;
 335
 336         get_skb_csum_pointers(skb, &csum_start, &csum_offset);
 337         csum_start -= skb_headroom(skb);
 338
 339         skb_copy_bits(skb, 0, to, csum_start);
 340
 341         csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
 342                                       skb->len - csum_start, 0);
 343         *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
 344 }
 345
 346 static struct genl_family dp_packet_genl_family = {
 347         .id = GENL_ID_GENERATE,
 348         .hdrsize = sizeof(struct ovs_header),
 349         .name = OVS_PACKET_FAMILY,
 350         .version = OVS_PACKET_VERSION,
 351         .maxattr = OVS_PACKET_ATTR_MAX
 352 };
 353
 354 int dp_upcall(struct datapath *dp, struct sk_buff *skb,
 355               const struct dp_upcall_info *upcall_info)
 356 {
 357         struct dp_stats_percpu *stats;
 358         int dp_ifindex;
 359         int err;
 360
 361         if (upcall_info->pid == 0) {
 362                 err = -ENOTCONN;
 363                 goto err;
 364         }
 365
 366         dp_ifindex = get_dpifindex(dp);
 367         if (!dp_ifindex) {
 368                 err = -ENODEV;
 369                 goto err;
 370         }
 371
 372         forward_ip_summed(skb, true);
 373
 374         if (!skb_is_gso(skb))
 375                 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
 376         else
 377                 err = queue_gso_packets(dp_ifindex, skb, upcall_info);
 378         if (err)
 379                 goto err;
 380
 381         return 0;
 382
 383 err:
 384         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 385
 386         write_seqcount_begin(&stats->seqlock);
 387         stats->n_lost++;
 388         write_seqcount_end(&stats->seqlock);
 389
 390         return err;
 391 }
 392
 393 static int queue_gso_packets(int dp_ifindex, struct sk_buff *skb,
 394                              const struct dp_upcall_info *upcall_info)
 395 {
 396         struct dp_upcall_info later_info;
 397         struct sw_flow_key later_key;
 398         struct sk_buff *segs, *nskb;
 399         int err;
 400
 401         segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 402         if (IS_ERR(skb))
 403                 return PTR_ERR(skb);
 404
 405         /* Queue all of the segments. */
 406         skb = segs;
 407         do {
 408                 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
 409                 if (err)
 410                         break;
 411
 412                 if (skb == segs && skb_shinfo(skb)->gso_type & SKB_GSO_UDP) {
 413                         /* The initial flow key extracted by flow_extract() in
 414                          * this case is for a first fragment, so we need to
 415                          * properly mark later fragments.
 416                          */
 417                         later_key = *upcall_info->key;
 418                         later_key.ip.tos_frag &= ~OVS_FRAG_TYPE_MASK;
 419                         later_key.ip.tos_frag |= OVS_FRAG_TYPE_LATER;
 420
 421                         later_info = *upcall_info;
 422                         later_info.key = &later_key;
 423                         upcall_info = &later_info;
 424                 }
 425         } while ((skb = skb->next));
 426
 427         /* Free all of the segments. */
 428         skb = segs;
 429         do {
 430                 nskb = skb->next;
 431                 if (err)
 432                         kfree_skb(skb);
 433                 else
 434                         consume_skb(skb);
 435         } while ((skb = nskb));
 436         return err;
 437 }
 438
 439 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *skb,
 440                                   const struct dp_upcall_info *upcall_info)
 441 {
 442         struct ovs_header *upcall;
 443         struct sk_buff *user_skb; /* to be queued to userspace */
 444         struct nlattr *nla;
 445         unsigned int len;
 446         int err;
 447
 448         err = vlan_deaccel_tag(skb);
 449         if (unlikely(err))
 450                 return err;
 451
 452         if (nla_attr_size(skb->len) > USHRT_MAX)
 453                 return -EFBIG;
 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                 return -ENOMEM;
 464
 465         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 466                              0, upcall_info->cmd);
 467         upcall->dp_ifindex = dp_ifindex;
 468
 469         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 470         flow_to_nlattrs(upcall_info->key, user_skb);
 471         nla_nest_end(user_skb, nla);
 472
 473         if (upcall_info->userdata)
 474                 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
 475                             nla_get_u64(upcall_info->userdata));
 476
 477         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 478         if (skb->ip_summed == CHECKSUM_PARTIAL)
 479                 copy_and_csum_skb(skb, nla_data(nla));
 480         else
 481                 skb_copy_bits(skb, 0, nla_data(nla), skb->len);
 482
 483         return genlmsg_unicast(&init_net, user_skb, upcall_info->pid);
 484 }
 485
 486 /* Called with genl_mutex. */
 487 static int flush_flows(int dp_ifindex)
 488 {
 489         struct flow_table *old_table;
 490         struct flow_table *new_table;
 491         struct datapath *dp;
 492
 493         dp = get_dp(dp_ifindex);
 494         if (!dp)
 495                 return -ENODEV;
 496
 497         old_table = get_table_protected(dp);
 498         new_table = flow_tbl_alloc(TBL_MIN_BUCKETS);
 499         if (!new_table)
 500                 return -ENOMEM;
 501
 502         rcu_assign_pointer(dp->table, new_table);
 503
 504         flow_tbl_deferred_destroy(old_table);
 505         return 0;
 506 }
 507
 508 static int validate_actions(const struct nlattr *attr,
 509                                 const struct sw_flow_key *key, int depth);
 510
 511 static int validate_sample(const struct nlattr *attr,
 512                                 const struct sw_flow_key *key, int depth)
 513 {
 514         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 515         const struct nlattr *probability, *actions;
 516         const struct nlattr *a;
 517         int rem;
 518
 519         memset(attrs, 0, sizeof(attrs));
 520         nla_for_each_nested(a, attr, rem) {
 521                 int type = nla_type(a);
 522                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 523                         return -EINVAL;
 524                 attrs[type] = a;
 525         }
 526         if (rem)
 527                 return -EINVAL;
 528
 529         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 530         if (!probability || nla_len(probability) != sizeof(u32))
 531                 return -EINVAL;
 532
 533         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 534         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 535                 return -EINVAL;
 536         return validate_actions(actions, key, depth + 1);
 537 }
 538
 539 static int validate_action_key(const struct nlattr *a,
 540                                 const struct sw_flow_key *flow_key)
 541 {
 542         int act_type = nla_type(a);
 543         const struct nlattr *ovs_key = nla_data(a);
 544         int key_type = nla_type(ovs_key);
 545
 546         /* There can be only one key in a action */
 547         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 548                 return -EINVAL;
 549
 550         if (key_type > OVS_KEY_ATTR_MAX ||
 551             nla_len(ovs_key) != ovs_key_lens[key_type])
 552                 return -EINVAL;
 553
 554 #define ACTION(act, key)        (((act) << 8) | (key))
 555
 556         switch (ACTION(act_type, key_type)) {
 557         const struct ovs_key_ipv4 *ipv4_key;
 558         const struct ovs_key_8021q *q_key;
 559
 560         case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_PRIORITY):
 561         case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_TUN_ID):
 562         case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_ETHERNET):
 563                 break;
 564
 565         case ACTION(OVS_ACTION_ATTR_PUSH, OVS_KEY_ATTR_8021Q):
 566                 q_key = nla_data(ovs_key);
 567                 if (q_key->q_tpid != htons(ETH_P_8021Q))
 568                         return -EINVAL;
 569
 570                 if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
 571                         return -EINVAL;
 572                 break;
 573
 574         case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_IPV4):
 575                 if (flow_key->eth.type != htons(ETH_P_IP))
 576                         return -EINVAL;
 577
 578                 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
 579                         return -EINVAL;
 580
 581                 ipv4_key = nla_data(ovs_key);
 582                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 583                         return -EINVAL;
 584
 585                 if (ipv4_key->ipv4_tos & INET_ECN_MASK)
 586                         return -EINVAL;
 587
 588                 if (ipv4_key->ipv4_frag !=
 589                     (flow_key->ip.tos_frag & OVS_FRAG_TYPE_MASK))
 590                         return -EINVAL;
 591
 592                 break;
 593
 594         case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_TCP):
 595                 if (flow_key->ip.proto != IPPROTO_TCP)
 596                         return -EINVAL;
 597
 598                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 599                         return -EINVAL;
 600
 601                 break;
 602
 603         case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_UDP):
 604                 if (flow_key->ip.proto != IPPROTO_UDP)
 605                         return -EINVAL;
 606
 607                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 608                         return -EINVAL;
 609                 break;
 610
 611         default:
 612                 return -EINVAL;
 613         }
 614 #undef ACTION
 615         return 0;
 616 }
 617
 618 static int validate_userspace(const struct nlattr *attr)
 619 {
 620         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 621                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 622                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
 623         };
 624         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 625         int error;
 626
 627         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 628                                  attr, userspace_policy);
 629         if (error)
 630                 return error;
 631
 632         if (!a[OVS_USERSPACE_ATTR_PID] ||
 633             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 634                 return -EINVAL;
 635
 636         return 0;
 637 }
 638
 639 static int validate_actions(const struct nlattr *attr,
 640                                 const struct sw_flow_key *key,  int depth)
 641 {
 642         const struct nlattr *a;
 643         int rem, err;
 644
 645         if (depth >= SAMPLE_ACTION_DEPTH)
 646                 return -EOVERFLOW;
 647
 648         nla_for_each_nested(a, attr, rem) {
 649                 /* Expected argument lengths, (u32)-1 for variable length. */
 650                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 651                         [OVS_ACTION_ATTR_OUTPUT] = 4,
 652                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 653                         [OVS_ACTION_ATTR_PUSH] = (u32)-1,
 654                         [OVS_ACTION_ATTR_POP] = 2,
 655                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 656                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 657                 };
 658                 int type = nla_type(a);
 659
 660                 if (type > OVS_ACTION_ATTR_MAX ||
 661                     (action_lens[type] != nla_len(a) &&
 662                      action_lens[type] != (u32)-1))
 663                         return -EINVAL;
 664
 665                 switch (type) {
 666                 case OVS_ACTION_ATTR_UNSPEC:
 667                         return -EINVAL;
 668
 669                 case OVS_ACTION_ATTR_USERSPACE:
 670                         err = validate_userspace(a);
 671                         if (err)
 672                                 return err;
 673                         break;
 674
 675                 case OVS_ACTION_ATTR_OUTPUT:
 676                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 677                                 return -EINVAL;
 678                         break;
 679
 680
 681                 case OVS_ACTION_ATTR_POP:
 682                         if (nla_get_u16(a) != OVS_KEY_ATTR_8021Q)
 683                                 return -EINVAL;
 684                         break;
 685
 686                 case OVS_ACTION_ATTR_SET:
 687                 case OVS_ACTION_ATTR_PUSH:
 688                         err = validate_action_key(a, key);
 689                         if (err)
 690                                 return err;
 691                         break;
 692
 693                 case OVS_ACTION_ATTR_SAMPLE:
 694                         err = validate_sample(a, key, depth);
 695                         if (err)
 696                                 return err;
 697                         break;
 698
 699                 default:
 700                         return -EINVAL;
 701                 }
 702         }
 703
 704         if (rem > 0)
 705                 return -EINVAL;
 706
 707         return 0;
 708 }
 709
 710 static void clear_stats(struct sw_flow *flow)
 711 {
 712         flow->used = 0;
 713         flow->tcp_flags = 0;
 714         flow->packet_count = 0;
 715         flow->byte_count = 0;
 716 }
 717
 718 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 719 {
 720         struct ovs_header *ovs_header = info->userhdr;
 721         struct nlattr **a = info->attrs;
 722         struct sw_flow_actions *acts;
 723         struct sk_buff *packet;
 724         struct sw_flow *flow;
 725         struct datapath *dp;
 726         struct ethhdr *eth;
 727         int len;
 728         int err;
 729         int key_len;
 730
 731         err = -EINVAL;
 732         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 733             !a[OVS_PACKET_ATTR_ACTIONS] ||
 734             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 735                 goto err;
 736
 737         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 738         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 739         err = -ENOMEM;
 740         if (!packet)
 741                 goto err;
 742         skb_reserve(packet, NET_IP_ALIGN);
 743
 744         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 745
 746         skb_reset_mac_header(packet);
 747         eth = eth_hdr(packet);
 748
 749         /* Normally, setting the skb 'protocol' field would be handled by a
 750          * call to eth_type_trans(), but it assumes there's a sending
 751          * device, which we may not have. */
 752         if (ntohs(eth->h_proto) >= 1536)
 753                 packet->protocol = eth->h_proto;
 754         else
 755                 packet->protocol = htons(ETH_P_802_2);
 756
 757         /* Build an sw_flow for sending this packet. */
 758         flow = flow_alloc();
 759         err = PTR_ERR(flow);
 760         if (IS_ERR(flow))
 761                 goto err_kfree_skb;
 762
 763         err = flow_extract(packet, -1, &flow->key, &key_len);
 764         if (err)
 765                 goto err_flow_put;
 766
 767         err = flow_metadata_from_nlattrs(&flow->key.phy.priority,
 768                                          &flow->key.phy.in_port,
 769                                          &flow->key.phy.tun_id,
 770                                          a[OVS_PACKET_ATTR_KEY]);
 771         if (err)
 772                 goto err_flow_put;
 773
 774         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
 775         if (err)
 776                 goto err_flow_put;
 777
 778         flow->hash = flow_hash(&flow->key, key_len);
 779
 780         acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 781         err = PTR_ERR(acts);
 782         if (IS_ERR(acts))
 783                 goto err_flow_put;
 784         rcu_assign_pointer(flow->sf_acts, acts);
 785
 786         OVS_CB(packet)->flow = flow;
 787         packet->priority = flow->key.phy.priority;
 788
 789         rcu_read_lock();
 790         dp = get_dp(ovs_header->dp_ifindex);
 791         err = -ENODEV;
 792         if (!dp)
 793                 goto err_unlock;
 794
 795         local_bh_disable();
 796         err = execute_actions(dp, packet);
 797         local_bh_enable();
 798         rcu_read_unlock();
 799
 800         flow_put(flow);
 801         return err;
 802
 803 err_unlock:
 804         rcu_read_unlock();
 805 err_flow_put:
 806         flow_put(flow);
 807 err_kfree_skb:
 808         kfree_skb(packet);
 809 err:
 810         return err;
 811 }
 812
 813 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 814         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 815         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 816         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 817 };
 818
 819 static struct genl_ops dp_packet_genl_ops[] = {
 820         { .cmd = OVS_PACKET_CMD_EXECUTE,
 821           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 822           .policy = packet_policy,
 823           .doit = ovs_packet_cmd_execute
 824         }
 825 };
 826
 827 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 828 {
 829         int i;
 830         struct flow_table *table = get_table_protected(dp);
 831
 832         stats->n_flows = flow_tbl_count(table);
 833
 834         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 835         for_each_possible_cpu(i) {
 836                 const struct dp_stats_percpu *percpu_stats;
 837                 struct dp_stats_percpu local_stats;
 838                 unsigned seqcount;
 839
 840                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 841
 842                 do {
 843                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 844                         local_stats = *percpu_stats;
 845                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 846
 847                 stats->n_hit += local_stats.n_hit;
 848                 stats->n_missed += local_stats.n_missed;
 849                 stats->n_lost += local_stats.n_lost;
 850         }
 851 }
 852
 853 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 854         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 855         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 856         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 857 };
 858
 859 static struct genl_family dp_flow_genl_family = {
 860         .id = GENL_ID_GENERATE,
 861         .hdrsize = sizeof(struct ovs_header),
 862         .name = OVS_FLOW_FAMILY,
 863         .version = OVS_FLOW_VERSION,
 864         .maxattr = OVS_FLOW_ATTR_MAX
 865 };
 866
 867 static struct genl_multicast_group dp_flow_multicast_group = {
 868         .name = OVS_FLOW_MCGROUP
 869 };
 870
 871 /* Called with genl_lock. */
 872 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 873                                   struct sk_buff *skb, u32 pid,
 874                                   u32 seq, u32 flags, u8 cmd)
 875 {
 876         const int skb_orig_len = skb->len;
 877         const struct sw_flow_actions *sf_acts;
 878         struct ovs_flow_stats stats;
 879         struct ovs_header *ovs_header;
 880         struct nlattr *nla;
 881         unsigned long used;
 882         u8 tcp_flags;
 883         int err;
 884
 885         sf_acts = rcu_dereference_protected(flow->sf_acts,
 886                                             lockdep_genl_is_held());
 887
 888         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 889         if (!ovs_header)
 890                 return -EMSGSIZE;
 891
 892         ovs_header->dp_ifindex = get_dpifindex(dp);
 893
 894         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 895         if (!nla)
 896                 goto nla_put_failure;
 897         err = flow_to_nlattrs(&flow->key, skb);
 898         if (err)
 899                 goto error;
 900         nla_nest_end(skb, nla);
 901
 902         spin_lock_bh(&flow->lock);
 903         used = flow->used;
 904         stats.n_packets = flow->packet_count;
 905         stats.n_bytes = flow->byte_count;
 906         tcp_flags = flow->tcp_flags;
 907         spin_unlock_bh(&flow->lock);
 908
 909         if (used)
 910                 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
 911
 912         if (stats.n_packets)
 913                 NLA_PUT(skb, OVS_FLOW_ATTR_STATS,
 914                         sizeof(struct ovs_flow_stats), &stats);
 915
 916         if (tcp_flags)
 917                 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 918
 919         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 920          * this is the first flow to be dumped into 'skb'.  This is unusual for
 921          * Netlink but individual action lists can be longer than
 922          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 923          * The userspace caller can always fetch the actions separately if it
 924          * really wants them.  (Most userspace callers in fact don't care.)
 925          *
 926          * This can only fail for dump operations because the skb is always
 927          * properly sized for single flows.
 928          */
 929         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 930                       sf_acts->actions);
 931         if (err < 0 && skb_orig_len)
 932                 goto error;
 933
 934         return genlmsg_end(skb, ovs_header);
 935
 936 nla_put_failure:
 937         err = -EMSGSIZE;
 938 error:
 939         genlmsg_cancel(skb, ovs_header);
 940         return err;
 941 }
 942
 943 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 944 {
 945         const struct sw_flow_actions *sf_acts;
 946         int len;
 947
 948         sf_acts = rcu_dereference_protected(flow->sf_acts,
 949                                             lockdep_genl_is_held());
 950
 951         /* OVS_FLOW_ATTR_KEY */
 952         len = nla_total_size(FLOW_BUFSIZE);
 953         /* OVS_FLOW_ATTR_ACTIONS */
 954         len += nla_total_size(sf_acts->actions_len);
 955         /* OVS_FLOW_ATTR_STATS */
 956         len += nla_total_size(sizeof(struct ovs_flow_stats));
 957         /* OVS_FLOW_ATTR_TCP_FLAGS */
 958         len += nla_total_size(1);
 959         /* OVS_FLOW_ATTR_USED */
 960         len += nla_total_size(8);
 961
 962         len += NLMSG_ALIGN(sizeof(struct ovs_header));
 963
 964         return genlmsg_new(len, GFP_KERNEL);
 965 }
 966
 967 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 968                                                struct datapath *dp,
 969                                                u32 pid, u32 seq, u8 cmd)
 970 {
 971         struct sk_buff *skb;
 972         int retval;
 973
 974         skb = ovs_flow_cmd_alloc_info(flow);
 975         if (!skb)
 976                 return ERR_PTR(-ENOMEM);
 977
 978         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 979         BUG_ON(retval < 0);
 980         return skb;
 981 }
 982
 983 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 984 {
 985         struct nlattr **a = info->attrs;
 986         struct ovs_header *ovs_header = info->userhdr;
 987         struct sw_flow_key key;
 988         struct sw_flow *flow;
 989         struct sk_buff *reply;
 990         struct datapath *dp;
 991         struct flow_table *table;
 992         int error;
 993         int key_len;
 994
 995         /* Extract key. */
 996         error = -EINVAL;
 997         if (!a[OVS_FLOW_ATTR_KEY])
 998                 goto error;
 999         error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1000         if (error)
1001                 goto error;
1002
1003         /* Validate actions. */
1004         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1005                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
1006                 if (error)
1007                         goto error;
1008         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1009                 error = -EINVAL;
1010                 goto error;
1011         }
1012
1013         dp = get_dp(ovs_header->dp_ifindex);
1014         error = -ENODEV;
1015         if (!dp)
1016                 goto error;
1017
1018         table = get_table_protected(dp);
1019         flow = flow_tbl_lookup(table, &key, key_len);
1020         if (!flow) {
1021                 struct sw_flow_actions *acts;
1022
1023                 /* Bail out if we're not allowed to create a new flow. */
1024                 error = -ENOENT;
1025                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1026                         goto error;
1027
1028                 /* Expand table, if necessary, to make room. */
1029                 if (flow_tbl_need_to_expand(table)) {
1030                         struct flow_table *new_table;
1031
1032                         new_table = flow_tbl_expand(table);
1033                         if (!IS_ERR(new_table)) {
1034                                 rcu_assign_pointer(dp->table, new_table);
1035                                 flow_tbl_deferred_destroy(table);
1036                                 table = get_table_protected(dp);
1037                         }
1038                 }
1039
1040                 /* Allocate flow. */
1041                 flow = flow_alloc();
1042                 if (IS_ERR(flow)) {
1043                         error = PTR_ERR(flow);
1044                         goto error;
1045                 }
1046                 flow->key = key;
1047                 clear_stats(flow);
1048
1049                 /* Obtain actions. */
1050                 acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1051                 error = PTR_ERR(acts);
1052                 if (IS_ERR(acts))
1053                         goto error_free_flow;
1054                 rcu_assign_pointer(flow->sf_acts, acts);
1055
1056                 /* Put flow in bucket. */
1057                 flow->hash = flow_hash(&key, key_len);
1058                 flow_tbl_insert(table, flow);
1059
1060                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1061                                                 info->snd_seq,
1062                                                 OVS_FLOW_CMD_NEW);
1063         } else {
1064                 /* We found a matching flow. */
1065                 struct sw_flow_actions *old_acts;
1066                 struct nlattr *acts_attrs;
1067
1068                 /* Bail out if we're not allowed to modify an existing flow.
1069                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1070                  * because Generic Netlink treats the latter as a dump
1071                  * request.  We also accept NLM_F_EXCL in case that bug ever
1072                  * gets fixed.
1073                  */
1074                 error = -EEXIST;
1075                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1076                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1077                         goto error;
1078
1079                 /* Update actions. */
1080                 old_acts = rcu_dereference_protected(flow->sf_acts,
1081                                                      lockdep_genl_is_held());
1082                 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1083                 if (acts_attrs &&
1084                    (old_acts->actions_len != nla_len(acts_attrs) ||
1085                    memcmp(old_acts->actions, nla_data(acts_attrs),
1086                           old_acts->actions_len))) {
1087                         struct sw_flow_actions *new_acts;
1088
1089                         new_acts = flow_actions_alloc(acts_attrs);
1090                         error = PTR_ERR(new_acts);
1091                         if (IS_ERR(new_acts))
1092                                 goto error;
1093
1094                         rcu_assign_pointer(flow->sf_acts, new_acts);
1095                         flow_deferred_free_acts(old_acts);
1096                 }
1097
1098                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1099                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1100
1101                 /* Clear stats. */
1102                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1103                         spin_lock_bh(&flow->lock);
1104                         clear_stats(flow);
1105                         spin_unlock_bh(&flow->lock);
1106                 }
1107         }
1108
1109         if (!IS_ERR(reply))
1110                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1111                            dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1112         else
1113                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1114                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1115         return 0;
1116
1117 error_free_flow:
1118         flow_put(flow);
1119 error:
1120         return error;
1121 }
1122
1123 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1124 {
1125         struct nlattr **a = info->attrs;
1126         struct ovs_header *ovs_header = info->userhdr;
1127         struct sw_flow_key key;
1128         struct sk_buff *reply;
1129         struct sw_flow *flow;
1130         struct datapath *dp;
1131         struct flow_table *table;
1132         int err;
1133         int key_len;
1134
1135         if (!a[OVS_FLOW_ATTR_KEY])
1136                 return -EINVAL;
1137         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1138         if (err)
1139                 return err;
1140
1141         dp = get_dp(ovs_header->dp_ifindex);
1142         if (!dp)
1143                 return -ENODEV;
1144
1145         table = get_table_protected(dp);
1146         flow = flow_tbl_lookup(table, &key, key_len);
1147         if (!flow)
1148                 return -ENOENT;
1149
1150         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1151                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1152         if (IS_ERR(reply))
1153                 return PTR_ERR(reply);
1154
1155         return genlmsg_reply(reply, info);
1156 }
1157
1158 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1159 {
1160         struct nlattr **a = info->attrs;
1161         struct ovs_header *ovs_header = info->userhdr;
1162         struct sw_flow_key key;
1163         struct sk_buff *reply;
1164         struct sw_flow *flow;
1165         struct datapath *dp;
1166         struct flow_table *table;
1167         int err;
1168         int key_len;
1169
1170         if (!a[OVS_FLOW_ATTR_KEY])
1171                 return flush_flows(ovs_header->dp_ifindex);
1172         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1173         if (err)
1174                 return err;
1175
1176         dp = get_dp(ovs_header->dp_ifindex);
1177         if (!dp)
1178                 return -ENODEV;
1179
1180         table = get_table_protected(dp);
1181         flow = flow_tbl_lookup(table, &key, key_len);
1182         if (!flow)
1183                 return -ENOENT;
1184
1185         reply = ovs_flow_cmd_alloc_info(flow);
1186         if (!reply)
1187                 return -ENOMEM;
1188
1189         flow_tbl_remove(table, flow);
1190
1191         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1192                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1193         BUG_ON(err < 0);
1194
1195         flow_deferred_free(flow);
1196
1197         genl_notify(reply, genl_info_net(info), info->snd_pid,
1198                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1199         return 0;
1200 }
1201
1202 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1203 {
1204         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1205         struct datapath *dp;
1206
1207         dp = get_dp(ovs_header->dp_ifindex);
1208         if (!dp)
1209                 return -ENODEV;
1210
1211         for (;;) {
1212                 struct sw_flow *flow;
1213                 u32 bucket, obj;
1214
1215                 bucket = cb->args[0];
1216                 obj = cb->args[1];
1217                 flow = flow_tbl_next(get_table_protected(dp), &bucket, &obj);
1218                 if (!flow)
1219                         break;
1220
1221                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1222                                            NETLINK_CB(cb->skb).pid,
1223                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1224                                            OVS_FLOW_CMD_NEW) < 0)
1225                         break;
1226
1227                 cb->args[0] = bucket;
1228                 cb->args[1] = obj;
1229         }
1230         return skb->len;
1231 }
1232
1233 static struct genl_ops dp_flow_genl_ops[] = {
1234         { .cmd = OVS_FLOW_CMD_NEW,
1235           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1236           .policy = flow_policy,
1237           .doit = ovs_flow_cmd_new_or_set
1238         },
1239         { .cmd = OVS_FLOW_CMD_DEL,
1240           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1241           .policy = flow_policy,
1242           .doit = ovs_flow_cmd_del
1243         },
1244         { .cmd = OVS_FLOW_CMD_GET,
1245           .flags = 0,               /* OK for unprivileged users. */
1246           .policy = flow_policy,
1247           .doit = ovs_flow_cmd_get,
1248           .dumpit = ovs_flow_cmd_dump
1249         },
1250         { .cmd = OVS_FLOW_CMD_SET,
1251           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1252           .policy = flow_policy,
1253           .doit = ovs_flow_cmd_new_or_set,
1254         },
1255 };
1256
1257 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1258 #ifdef HAVE_NLA_NUL_STRING
1259         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1260 #endif
1261         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1262 };
1263
1264 static struct genl_family dp_datapath_genl_family = {
1265         .id = GENL_ID_GENERATE,
1266         .hdrsize = sizeof(struct ovs_header),
1267         .name = OVS_DATAPATH_FAMILY,
1268         .version = OVS_DATAPATH_VERSION,
1269         .maxattr = OVS_DP_ATTR_MAX
1270 };
1271
1272 static struct genl_multicast_group dp_datapath_multicast_group = {
1273         .name = OVS_DATAPATH_MCGROUP
1274 };
1275
1276 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1277                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1278 {
1279         struct ovs_header *ovs_header;
1280         struct nlattr *nla;
1281         int err;
1282
1283         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1284                                    flags, cmd);
1285         if (!ovs_header)
1286                 goto error;
1287
1288         ovs_header->dp_ifindex = get_dpifindex(dp);
1289
1290         rcu_read_lock();
1291         err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
1292         rcu_read_unlock();
1293         if (err)
1294                 goto nla_put_failure;
1295
1296         nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats));
1297         if (!nla)
1298                 goto nla_put_failure;
1299         get_dp_stats(dp, nla_data(nla));
1300
1301         return genlmsg_end(skb, ovs_header);
1302
1303 nla_put_failure:
1304         genlmsg_cancel(skb, ovs_header);
1305 error:
1306         return -EMSGSIZE;
1307 }
1308
1309 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1310                                              u32 seq, u8 cmd)
1311 {
1312         struct sk_buff *skb;
1313         int retval;
1314
1315         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1316         if (!skb)
1317                 return ERR_PTR(-ENOMEM);
1318
1319         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1320         if (retval < 0) {
1321                 kfree_skb(skb);
1322                 return ERR_PTR(retval);
1323         }
1324         return skb;
1325 }
1326
1327 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1328 {
1329         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1330 }
1331
1332 /* Called with genl_mutex and optionally with RTNL lock also. */
1333 static struct datapath *lookup_datapath(struct ovs_header *ovs_header,
1334                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1335 {
1336         struct datapath *dp;
1337
1338         if (!a[OVS_DP_ATTR_NAME])
1339                 dp = get_dp(ovs_header->dp_ifindex);
1340         else {
1341                 struct vport *vport;
1342
1343                 rcu_read_lock();
1344                 vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1345                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1346                 rcu_read_unlock();
1347         }
1348         return dp ? dp : ERR_PTR(-ENODEV);
1349 }
1350
1351 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1352 {
1353         struct nlattr **a = info->attrs;
1354         struct vport_parms parms;
1355         struct sk_buff *reply;
1356         struct datapath *dp;
1357         struct vport *vport;
1358         int err;
1359
1360         err = -EINVAL;
1361         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1362                 goto err;
1363
1364         err = ovs_dp_cmd_validate(a);
1365         if (err)
1366                 goto err;
1367
1368         rtnl_lock();
1369         err = -ENODEV;
1370         if (!try_module_get(THIS_MODULE))
1371                 goto err_unlock_rtnl;
1372
1373         err = -ENOMEM;
1374         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1375         if (dp == NULL)
1376                 goto err_put_module;
1377         INIT_LIST_HEAD(&dp->port_list);
1378
1379         /* Initialize kobject for bridge.  This will be added as
1380          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1381         dp->ifobj.kset = NULL;
1382         kobject_init(&dp->ifobj, &dp_ktype);
1383
1384         /* Allocate table. */
1385         err = -ENOMEM;
1386         rcu_assign_pointer(dp->table, flow_tbl_alloc(TBL_MIN_BUCKETS));
1387         if (!dp->table)
1388                 goto err_free_dp;
1389
1390         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1391         if (!dp->stats_percpu) {
1392                 err = -ENOMEM;
1393                 goto err_destroy_table;
1394         }
1395
1396         /* Set up our datapath device. */
1397         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1398         parms.type = OVS_VPORT_TYPE_INTERNAL;
1399         parms.options = NULL;
1400         parms.dp = dp;
1401         parms.port_no = OVSP_LOCAL;
1402         parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1403
1404         vport = new_vport(&parms);
1405         if (IS_ERR(vport)) {
1406                 err = PTR_ERR(vport);
1407                 if (err == -EBUSY)
1408                         err = -EEXIST;
1409
1410                 goto err_destroy_percpu;
1411         }
1412
1413         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1414                                       info->snd_seq, OVS_DP_CMD_NEW);
1415         err = PTR_ERR(reply);
1416         if (IS_ERR(reply))
1417                 goto err_destroy_local_port;
1418
1419         list_add_tail(&dp->list_node, &dps);
1420         dp_sysfs_add_dp(dp);
1421
1422         rtnl_unlock();
1423
1424         genl_notify(reply, genl_info_net(info), info->snd_pid,
1425                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1426         return 0;
1427
1428 err_destroy_local_port:
1429         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1430 err_destroy_percpu:
1431         free_percpu(dp->stats_percpu);
1432 err_destroy_table:
1433         flow_tbl_destroy(get_table_protected(dp));
1434 err_free_dp:
1435         kfree(dp);
1436 err_put_module:
1437         module_put(THIS_MODULE);
1438 err_unlock_rtnl:
1439         rtnl_unlock();
1440 err:
1441         return err;
1442 }
1443
1444 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1445 {
1446         struct vport *vport, *next_vport;
1447         struct sk_buff *reply;
1448         struct datapath *dp;
1449         int err;
1450
1451         err = ovs_dp_cmd_validate(info->attrs);
1452         if (err)
1453                 goto exit;
1454
1455         rtnl_lock();
1456         dp = lookup_datapath(info->userhdr, info->attrs);
1457         err = PTR_ERR(dp);
1458         if (IS_ERR(dp))
1459                 goto exit_unlock;
1460
1461         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1462                                       info->snd_seq, OVS_DP_CMD_DEL);
1463         err = PTR_ERR(reply);
1464         if (IS_ERR(reply))
1465                 goto exit_unlock;
1466
1467         list_for_each_entry_safe(vport, next_vport, &dp->port_list, node)
1468                 if (vport->port_no != OVSP_LOCAL)
1469                         dp_detach_port(vport);
1470
1471         dp_sysfs_del_dp(dp);
1472         list_del(&dp->list_node);
1473         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1474
1475         /* rtnl_unlock() will wait until all the references to devices that
1476          * are pending unregistration have been dropped.  We do it here to
1477          * ensure that any internal devices (which contain DP pointers) are
1478          * fully destroyed before freeing the datapath.
1479          */
1480         rtnl_unlock();
1481
1482         call_rcu(&dp->rcu, destroy_dp_rcu);
1483         module_put(THIS_MODULE);
1484
1485         genl_notify(reply, genl_info_net(info), info->snd_pid,
1486                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1487
1488         return 0;
1489
1490 exit_unlock:
1491         rtnl_unlock();
1492 exit:
1493         return err;
1494 }
1495
1496 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1497 {
1498         struct sk_buff *reply;
1499         struct datapath *dp;
1500         int err;
1501
1502         err = ovs_dp_cmd_validate(info->attrs);
1503         if (err)
1504                 return err;
1505
1506         dp = lookup_datapath(info->userhdr, info->attrs);
1507         if (IS_ERR(dp))
1508                 return PTR_ERR(dp);
1509
1510         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1511                                       info->snd_seq, OVS_DP_CMD_NEW);
1512         if (IS_ERR(reply)) {
1513                 err = PTR_ERR(reply);
1514                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1515                                 dp_datapath_multicast_group.id, err);
1516                 return 0;
1517         }
1518
1519         genl_notify(reply, genl_info_net(info), info->snd_pid,
1520                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
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                         continue;
1555                 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1556                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1557                                          OVS_DP_CMD_NEW) < 0)
1558                         break;
1559                 i++;
1560         }
1561
1562         cb->args[0] = i;
1563
1564         return skb->len;
1565 }
1566
1567 static struct genl_ops dp_datapath_genl_ops[] = {
1568         { .cmd = OVS_DP_CMD_NEW,
1569           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1570           .policy = datapath_policy,
1571           .doit = ovs_dp_cmd_new
1572         },
1573         { .cmd = OVS_DP_CMD_DEL,
1574           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1575           .policy = datapath_policy,
1576           .doit = ovs_dp_cmd_del
1577         },
1578         { .cmd = OVS_DP_CMD_GET,
1579           .flags = 0,               /* OK for unprivileged users. */
1580           .policy = datapath_policy,
1581           .doit = ovs_dp_cmd_get,
1582           .dumpit = ovs_dp_cmd_dump
1583         },
1584         { .cmd = OVS_DP_CMD_SET,
1585           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1586           .policy = datapath_policy,
1587           .doit = ovs_dp_cmd_set,
1588         },
1589 };
1590
1591 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1592 #ifdef HAVE_NLA_NUL_STRING
1593         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1594         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1595         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1596 #else
1597         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1598         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1599 #endif
1600         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1601         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1602         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1603         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1604 };
1605
1606 static struct genl_family dp_vport_genl_family = {
1607         .id = GENL_ID_GENERATE,
1608         .hdrsize = sizeof(struct ovs_header),
1609         .name = OVS_VPORT_FAMILY,
1610         .version = OVS_VPORT_VERSION,
1611         .maxattr = OVS_VPORT_ATTR_MAX
1612 };
1613
1614 struct genl_multicast_group dp_vport_multicast_group = {
1615         .name = OVS_VPORT_MCGROUP
1616 };
1617
1618 /* Called with RTNL lock or RCU read lock. */
1619 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1620                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1621 {
1622         struct ovs_header *ovs_header;
1623         struct nlattr *nla;
1624         int err;
1625
1626         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1627                                  flags, cmd);
1628         if (!ovs_header)
1629                 return -EMSGSIZE;
1630
1631         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1632
1633         NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1634         NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport_get_type(vport));
1635         NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport_get_name(vport));
1636         NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1637
1638         nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS,
1639                           sizeof(struct ovs_vport_stats));
1640         if (!nla)
1641                 goto nla_put_failure;
1642
1643         vport_get_stats(vport, nla_data(nla));
1644
1645         NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1646
1647         err = vport_get_options(vport, skb);
1648         if (err == -EMSGSIZE)
1649                 goto error;
1650
1651         return genlmsg_end(skb, ovs_header);
1652
1653 nla_put_failure:
1654         err = -EMSGSIZE;
1655 error:
1656         genlmsg_cancel(skb, ovs_header);
1657         return err;
1658 }
1659
1660 /* Called with RTNL lock or RCU read lock. */
1661 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1662                                          u32 seq, u8 cmd)
1663 {
1664         struct sk_buff *skb;
1665         int retval;
1666
1667         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1668         if (!skb)
1669                 return ERR_PTR(-ENOMEM);
1670
1671         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1672         if (retval < 0) {
1673                 kfree_skb(skb);
1674                 return ERR_PTR(retval);
1675         }
1676         return skb;
1677 }
1678
1679 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1680 {
1681         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1682 }
1683
1684 /* Called with RTNL lock or RCU read lock. */
1685 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1686                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1687 {
1688         struct datapath *dp;
1689         struct vport *vport;
1690
1691         if (a[OVS_VPORT_ATTR_NAME]) {
1692                 vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1693                 if (!vport)
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 = get_vport_protected(dp, 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                 vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1722
1723         if (a[OVS_VPORT_ATTR_ADDRESS])
1724                 err = 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 = get_vport_protected(dp, 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 = get_vport_protected(dp, 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         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                 dp_detach_port(vport);
1802                 goto exit_unlock;
1803         }
1804         genl_notify(reply, genl_info_net(info), info->snd_pid,
1805                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1806
1807
1808 exit_unlock:
1809         rtnl_unlock();
1810 exit:
1811         return err;
1812 }
1813
1814 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1815 {
1816         struct nlattr **a = info->attrs;
1817         struct sk_buff *reply;
1818         struct vport *vport;
1819         int err;
1820
1821         err = ovs_vport_cmd_validate(a);
1822         if (err)
1823                 goto exit;
1824
1825         rtnl_lock();
1826         vport = lookup_vport(info->userhdr, a);
1827         err = PTR_ERR(vport);
1828         if (IS_ERR(vport))
1829                 goto exit_unlock;
1830
1831         err = 0;
1832         if (a[OVS_VPORT_ATTR_TYPE] &&
1833             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport_get_type(vport))
1834                 err = -EINVAL;
1835
1836         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1837                 err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1838         if (!err)
1839                 err = change_vport(vport, a);
1840         if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1841                 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1842
1843         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1844                                          OVS_VPORT_CMD_NEW);
1845         if (IS_ERR(reply)) {
1846                 err = PTR_ERR(reply);
1847                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1848                                 dp_vport_multicast_group.id, err);
1849                 return 0;
1850         }
1851
1852         genl_notify(reply, genl_info_net(info), info->snd_pid,
1853                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1854
1855 exit_unlock:
1856         rtnl_unlock();
1857 exit:
1858         return err;
1859 }
1860
1861 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1862 {
1863         struct nlattr **a = info->attrs;
1864         struct sk_buff *reply;
1865         struct vport *vport;
1866         int err;
1867
1868         err = ovs_vport_cmd_validate(a);
1869         if (err)
1870                 goto exit;
1871
1872         rtnl_lock();
1873         vport = lookup_vport(info->userhdr, a);
1874         err = PTR_ERR(vport);
1875         if (IS_ERR(vport))
1876                 goto exit_unlock;
1877
1878         if (vport->port_no == OVSP_LOCAL) {
1879                 err = -EINVAL;
1880                 goto exit_unlock;
1881         }
1882
1883         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1884                                          OVS_VPORT_CMD_DEL);
1885         err = PTR_ERR(reply);
1886         if (IS_ERR(reply))
1887                 goto exit_unlock;
1888
1889         dp_detach_port(vport);
1890
1891         genl_notify(reply, genl_info_net(info), info->snd_pid,
1892                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1893
1894 exit_unlock:
1895         rtnl_unlock();
1896 exit:
1897         return err;
1898 }
1899
1900 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1901 {
1902         struct nlattr **a = info->attrs;
1903         struct ovs_header *ovs_header = info->userhdr;
1904         struct sk_buff *reply;
1905         struct vport *vport;
1906         int err;
1907
1908         err = ovs_vport_cmd_validate(a);
1909         if (err)
1910                 goto exit;
1911
1912         rcu_read_lock();
1913         vport = lookup_vport(ovs_header, a);
1914         err = PTR_ERR(vport);
1915         if (IS_ERR(vport))
1916                 goto exit_unlock;
1917
1918         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1919                                          OVS_VPORT_CMD_NEW);
1920         err = PTR_ERR(reply);
1921         if (IS_ERR(reply))
1922                 goto exit_unlock;
1923
1924         rcu_read_unlock();
1925
1926         return genlmsg_reply(reply, info);
1927
1928 exit_unlock:
1929         rcu_read_unlock();
1930 exit:
1931         return err;
1932 }
1933
1934 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1935 {
1936         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1937         struct datapath *dp;
1938         u32 port_no;
1939         int retval;
1940
1941         dp = get_dp(ovs_header->dp_ifindex);
1942         if (!dp)
1943                 return -ENODEV;
1944
1945         rcu_read_lock();
1946         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1947                 struct vport *vport;
1948
1949                 vport = get_vport_protected(dp, port_no);
1950                 if (!vport)
1951                         continue;
1952
1953                 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1954                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1955                                             OVS_VPORT_CMD_NEW) < 0)
1956                         break;
1957         }
1958         rcu_read_unlock();
1959
1960         cb->args[0] = port_no;
1961         retval = skb->len;
1962
1963         return retval;
1964 }
1965
1966 static struct genl_ops dp_vport_genl_ops[] = {
1967         { .cmd = OVS_VPORT_CMD_NEW,
1968           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1969           .policy = vport_policy,
1970           .doit = ovs_vport_cmd_new
1971         },
1972         { .cmd = OVS_VPORT_CMD_DEL,
1973           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1974           .policy = vport_policy,
1975           .doit = ovs_vport_cmd_del
1976         },
1977         { .cmd = OVS_VPORT_CMD_GET,
1978           .flags = 0,               /* OK for unprivileged users. */
1979           .policy = vport_policy,
1980           .doit = ovs_vport_cmd_get,
1981           .dumpit = ovs_vport_cmd_dump
1982         },
1983         { .cmd = OVS_VPORT_CMD_SET,
1984           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1985           .policy = vport_policy,
1986           .doit = ovs_vport_cmd_set,
1987         },
1988 };
1989
1990 struct genl_family_and_ops {
1991         struct genl_family *family;
1992         struct genl_ops *ops;
1993         int n_ops;
1994         struct genl_multicast_group *group;
1995 };
1996
1997 static const struct genl_family_and_ops dp_genl_families[] = {
1998         { &dp_datapath_genl_family,
1999           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2000           &dp_datapath_multicast_group },
2001         { &dp_vport_genl_family,
2002           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2003           &dp_vport_multicast_group },
2004         { &dp_flow_genl_family,
2005           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2006           &dp_flow_multicast_group },
2007         { &dp_packet_genl_family,
2008           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2009           NULL },
2010 };
2011
2012 static void dp_unregister_genl(int n_families)
2013 {
2014         int i;
2015
2016         for (i = 0; i < n_families; i++)
2017                 genl_unregister_family(dp_genl_families[i].family);
2018 }
2019
2020 static int dp_register_genl(void)
2021 {
2022         int n_registered;
2023         int err;
2024         int i;
2025
2026         n_registered = 0;
2027         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2028                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2029
2030                 err = genl_register_family_with_ops(f->family, f->ops,
2031                                                     f->n_ops);
2032                 if (err)
2033                         goto error;
2034                 n_registered++;
2035
2036                 if (f->group) {
2037                         err = genl_register_mc_group(f->family, f->group);
2038                         if (err)
2039                                 goto error;
2040                 }
2041         }
2042
2043         return 0;
2044
2045 error:
2046         dp_unregister_genl(n_registered);
2047         return err;
2048 }
2049
2050 static int __init dp_init(void)
2051 {
2052         struct sk_buff *dummy_skb;
2053         int err;
2054
2055         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2056
2057         pr_info("Open vSwitch %s, built "__DATE__" "__TIME__"\n",
2058                 VERSION BUILDNR);
2059
2060         err = tnl_init();
2061         if (err)
2062                 goto error;
2063
2064         err = flow_init();
2065         if (err)
2066                 goto error_tnl_exit;
2067
2068         err = vport_init();
2069         if (err)
2070                 goto error_flow_exit;
2071
2072         err = register_netdevice_notifier(&dp_device_notifier);
2073         if (err)
2074                 goto error_vport_exit;
2075
2076         err = dp_register_genl();
2077         if (err < 0)
2078                 goto error_unreg_notifier;
2079
2080         return 0;
2081
2082 error_unreg_notifier:
2083         unregister_netdevice_notifier(&dp_device_notifier);
2084 error_vport_exit:
2085         vport_exit();
2086 error_flow_exit:
2087         flow_exit();
2088 error_tnl_exit:
2089         tnl_exit();
2090 error:
2091         return err;
2092 }
2093
2094 static void dp_cleanup(void)
2095 {
2096         rcu_barrier();
2097         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2098         unregister_netdevice_notifier(&dp_device_notifier);
2099         vport_exit();
2100         flow_exit();
2101         tnl_exit();
2102 }
2103
2104 module_init(dp_init);
2105 module_exit(dp_cleanup);
2106
2107 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2108 MODULE_LICENSE("GPL");