datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007-2011 Nicira Networks.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/version.h>
  40 #include <linux/ethtool.h>
  41 #include <linux/wait.h>
  42 #include <asm/system.h>
  43 #include <asm/div64.h>
  44 #include <linux/highmem.h>
  45 #include <linux/netfilter_bridge.h>
  46 #include <linux/netfilter_ipv4.h>
  47 #include <linux/inetdevice.h>
  48 #include <linux/list.h>
  49 #include <linux/openvswitch.h>
  50 #include <linux/rculist.h>
  51 #include <linux/dmi.h>
  52 #include <net/genetlink.h>
  53
  54 #include "checksum.h"
  55 #include "datapath.h"
  56 #include "flow.h"
  57 #include "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(3,2,0)
  64 #error Kernels before 2.6.18 or after 3.2 are not supported by this version of Open vSwitch.
  65 #endif
  66
  67 int (*ovs_dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
  68 EXPORT_SYMBOL(ovs_dp_ioctl_hook);
  69
  70 /**
  71  * DOC: Locking:
  72  *
  73  * Writes to device state (add/remove datapath, port, set operations on vports,
  74  * etc.) are protected by RTNL.
  75  *
  76  * Writes to other state (flow table modifications, set miscellaneous datapath
  77  * parameters, etc.) are protected by genl_mutex.  The RTNL lock nests inside
  78  * genl_mutex.
  79  *
  80  * Reads are protected by RCU.
  81  *
  82  * There are a few special cases (mostly stats) that have their own
  83  * synchronization but they nest under all of above and don't interact with
  84  * each other.
  85  */
  86
  87 /* Global list of datapaths to enable dumping them all out.
  88  * Protected by genl_mutex.
  89  */
  90 static LIST_HEAD(dps);
  91
  92 static struct vport *new_vport(const struct vport_parms *);
  93 static int queue_gso_packets(int dp_ifindex, struct sk_buff *,
  94                              const struct dp_upcall_info *);
  95 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *,
  96                                   const struct dp_upcall_info *);
  97
  98 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  99 static struct datapath *get_dp(int dp_ifindex)
 100 {
 101         struct datapath *dp = NULL;
 102         struct net_device *dev;
 103
 104         rcu_read_lock();
 105         dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
 106         if (dev) {
 107                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 108                 if (vport)
 109                         dp = vport->dp;
 110         }
 111         rcu_read_unlock();
 112
 113         return dp;
 114 }
 115
 116 /* Must be called with rcu_read_lock or RTNL lock. */
 117 const char *ovs_dp_name(const struct datapath *dp)
 118 {
 119         struct vport *vport = rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]);
 120         return vport->ops->get_name(vport);
 121 }
 122
 123 static int get_dpifindex(struct datapath *dp)
 124 {
 125         struct vport *local;
 126         int ifindex;
 127
 128         rcu_read_lock();
 129
 130         local = rcu_dereference(dp->ports[OVSP_LOCAL]);
 131         if (local)
 132                 ifindex = local->ops->get_ifindex(local);
 133         else
 134                 ifindex = 0;
 135
 136         rcu_read_unlock();
 137
 138         return ifindex;
 139 }
 140
 141 static size_t br_nlmsg_size(void)
 142 {
 143         return NLMSG_ALIGN(sizeof(struct ifinfomsg))
 144                + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
 145                + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
 146                + nla_total_size(4) /* IFLA_MASTER */
 147                + nla_total_size(4) /* IFLA_MTU */
 148                + nla_total_size(1); /* IFLA_OPERSTATE */
 149 }
 150
 151 /* Caller must hold RTNL lock. */
 152 static int dp_fill_ifinfo(struct sk_buff *skb,
 153                           const struct vport *port,
 154                           int event, unsigned int flags)
 155 {
 156         struct datapath *dp = port->dp;
 157         struct ifinfomsg *hdr;
 158         struct nlmsghdr *nlh;
 159
 160         if (!port->ops->get_ifindex)
 161                 return -ENODEV;
 162
 163         nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
 164         if (nlh == NULL)
 165                 return -EMSGSIZE;
 166
 167         hdr = nlmsg_data(nlh);
 168         hdr->ifi_family = AF_BRIDGE;
 169         hdr->__ifi_pad = 0;
 170         hdr->ifi_type = ARPHRD_ETHER;
 171         hdr->ifi_index = port->ops->get_ifindex(port);
 172         hdr->ifi_flags = port->ops->get_dev_flags(port);
 173         hdr->ifi_change = 0;
 174
 175         NLA_PUT_STRING(skb, IFLA_IFNAME, port->ops->get_name(port));
 176         NLA_PUT_U32(skb, IFLA_MASTER, get_dpifindex(dp));
 177         NLA_PUT_U32(skb, IFLA_MTU, port->ops->get_mtu(port));
 178 #ifdef IFLA_OPERSTATE
 179         NLA_PUT_U8(skb, IFLA_OPERSTATE,
 180                    port->ops->is_running(port)
 181                         ? port->ops->get_operstate(port)
 182                         : IF_OPER_DOWN);
 183 #endif
 184
 185         NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, port->ops->get_addr(port));
 186
 187         return nlmsg_end(skb, nlh);
 188
 189 nla_put_failure:
 190         nlmsg_cancel(skb, nlh);
 191         return -EMSGSIZE;
 192 }
 193
 194 /* Caller must hold RTNL lock. */
 195 static void dp_ifinfo_notify(int event, struct vport *port)
 196 {
 197         struct sk_buff *skb;
 198         int err;
 199
 200         skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
 201         if (!skb) {
 202                 err = -ENOBUFS;
 203                 goto err;
 204         }
 205
 206         err = dp_fill_ifinfo(skb, port, event, 0);
 207         if (err < 0) {
 208                 if (err == -ENODEV) {
 209                         goto out;
 210                 } else {
 211                         /* -EMSGSIZE implies BUG in br_nlmsg_size() */
 212                         WARN_ON(err == -EMSGSIZE);
 213                         goto err;
 214                 }
 215         }
 216
 217         rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 218
 219         return;
 220 err:
 221         rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
 222 out:
 223         kfree_skb(skb);
 224 }
 225
 226 static void release_dp(struct kobject *kobj)
 227 {
 228         struct datapath *dp = container_of(kobj, struct datapath, ifobj);
 229         kfree(dp);
 230 }
 231
 232 static struct kobj_type dp_ktype = {
 233         .release = release_dp
 234 };
 235
 236 static void destroy_dp_rcu(struct rcu_head *rcu)
 237 {
 238         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 239
 240         ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
 241         free_percpu(dp->stats_percpu);
 242         kobject_put(&dp->ifobj);
 243 }
 244
 245 /* Called with RTNL lock and genl_lock. */
 246 static struct vport *new_vport(const struct vport_parms *parms)
 247 {
 248         struct vport *vport;
 249
 250         vport = ovs_vport_add(parms);
 251         if (!IS_ERR(vport)) {
 252                 struct datapath *dp = parms->dp;
 253
 254                 rcu_assign_pointer(dp->ports[parms->port_no], vport);
 255                 list_add(&vport->node, &dp->port_list);
 256
 257                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 258         }
 259
 260         return vport;
 261 }
 262
 263 /* Called with RTNL lock. */
 264 void ovs_dp_detach_port(struct vport *p)
 265 {
 266         ASSERT_RTNL();
 267
 268         if (p->port_no != OVSP_LOCAL)
 269                 ovs_dp_sysfs_del_if(p);
 270         dp_ifinfo_notify(RTM_DELLINK, p);
 271
 272         /* First drop references to device. */
 273         list_del(&p->node);
 274         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 275
 276         /* Then destroy it. */
 277         ovs_vport_del(p);
 278 }
 279
 280 /* Must be called with rcu_read_lock. */
 281 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 282 {
 283         struct datapath *dp = p->dp;
 284         struct sw_flow *flow;
 285         struct dp_stats_percpu *stats;
 286         u64 *stats_counter;
 287         int error;
 288
 289         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 290
 291         if (!OVS_CB(skb)->flow) {
 292                 struct sw_flow_key key;
 293                 int key_len;
 294
 295                 /* Extract flow from 'skb' into 'key'. */
 296                 error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
 297                 if (unlikely(error)) {
 298                         kfree_skb(skb);
 299                         return;
 300                 }
 301
 302                 /* Look up flow. */
 303                 flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table),
 304                                            &key, key_len);
 305                 if (unlikely(!flow)) {
 306                         struct dp_upcall_info upcall;
 307
 308                         upcall.cmd = OVS_PACKET_CMD_MISS;
 309                         upcall.key = &key;
 310                         upcall.userdata = NULL;
 311                         upcall.pid = p->upcall_pid;
 312                         ovs_dp_upcall(dp, skb, &upcall);
 313                         consume_skb(skb);
 314                         stats_counter = &stats->n_missed;
 315                         goto out;
 316                 }
 317
 318                 OVS_CB(skb)->flow = flow;
 319         }
 320
 321         stats_counter = &stats->n_hit;
 322         ovs_flow_used(OVS_CB(skb)->flow, skb);
 323         ovs_execute_actions(dp, skb);
 324
 325 out:
 326         /* Update datapath statistics. */
 327         u64_stats_update_begin(&stats->sync);
 328         (*stats_counter)++;
 329         u64_stats_update_end(&stats->sync);
 330 }
 331
 332 static struct genl_family dp_packet_genl_family = {
 333         .id = GENL_ID_GENERATE,
 334         .hdrsize = sizeof(struct ovs_header),
 335         .name = OVS_PACKET_FAMILY,
 336         .version = OVS_PACKET_VERSION,
 337         .maxattr = OVS_PACKET_ATTR_MAX
 338 };
 339
 340 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 341                   const struct dp_upcall_info *upcall_info)
 342 {
 343         struct dp_stats_percpu *stats;
 344         int dp_ifindex;
 345         int err;
 346
 347         if (upcall_info->pid == 0) {
 348                 err = -ENOTCONN;
 349                 goto err;
 350         }
 351
 352         dp_ifindex = get_dpifindex(dp);
 353         if (!dp_ifindex) {
 354                 err = -ENODEV;
 355                 goto err;
 356         }
 357
 358         forward_ip_summed(skb, true);
 359
 360         if (!skb_is_gso(skb))
 361                 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
 362         else
 363                 err = queue_gso_packets(dp_ifindex, skb, upcall_info);
 364         if (err)
 365                 goto err;
 366
 367         return 0;
 368
 369 err:
 370         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 371
 372         u64_stats_update_begin(&stats->sync);
 373         stats->n_lost++;
 374         u64_stats_update_end(&stats->sync);
 375
 376         return err;
 377 }
 378
 379 static int queue_gso_packets(int dp_ifindex, struct sk_buff *skb,
 380                              const struct dp_upcall_info *upcall_info)
 381 {
 382         struct dp_upcall_info later_info;
 383         struct sw_flow_key later_key;
 384         struct sk_buff *segs, *nskb;
 385         int err;
 386
 387         segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 388         if (IS_ERR(skb))
 389                 return PTR_ERR(skb);
 390
 391         /* Queue all of the segments. */
 392         skb = segs;
 393         do {
 394                 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
 395                 if (err)
 396                         break;
 397
 398                 if (skb == segs && skb_shinfo(skb)->gso_type & SKB_GSO_UDP) {
 399                         /* The initial flow key extracted by ovs_flow_extract()
 400                          * in this case is for a first fragment, so we need to
 401                          * properly mark later fragments.
 402                          */
 403                         later_key = *upcall_info->key;
 404                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 405
 406                         later_info = *upcall_info;
 407                         later_info.key = &later_key;
 408                         upcall_info = &later_info;
 409                 }
 410         } while ((skb = skb->next));
 411
 412         /* Free all of the segments. */
 413         skb = segs;
 414         do {
 415                 nskb = skb->next;
 416                 if (err)
 417                         kfree_skb(skb);
 418                 else
 419                         consume_skb(skb);
 420         } while ((skb = nskb));
 421         return err;
 422 }
 423
 424 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *skb,
 425                                   const struct dp_upcall_info *upcall_info)
 426 {
 427         struct ovs_header *upcall;
 428         struct sk_buff *nskb = NULL;
 429         struct sk_buff *user_skb; /* to be queued to userspace */
 430         struct nlattr *nla;
 431         unsigned int len;
 432         int err;
 433
 434         if (vlan_tx_tag_present(skb)) {
 435                 nskb = skb_clone(skb, GFP_ATOMIC);
 436                 if (!nskb)
 437                         return -ENOMEM;
 438
 439                 err = vlan_deaccel_tag(nskb);
 440                 if (err)
 441                         return err;
 442
 443                 skb = nskb;
 444         }
 445
 446         if (nla_attr_size(skb->len) > USHRT_MAX) {
 447                 err = -EFBIG;
 448                 goto out;
 449         }
 450
 451         len = sizeof(struct ovs_header);
 452         len += nla_total_size(skb->len);
 453         len += nla_total_size(FLOW_BUFSIZE);
 454         if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
 455                 len += nla_total_size(8);
 456
 457         user_skb = genlmsg_new(len, GFP_ATOMIC);
 458         if (!user_skb) {
 459                 err = -ENOMEM;
 460                 goto out;
 461         }
 462
 463         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 464                              0, upcall_info->cmd);
 465         upcall->dp_ifindex = dp_ifindex;
 466
 467         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 468         ovs_flow_to_nlattrs(upcall_info->key, user_skb);
 469         nla_nest_end(user_skb, nla);
 470
 471         if (upcall_info->userdata)
 472                 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
 473                             nla_get_u64(upcall_info->userdata));
 474
 475         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 476
 477         skb_copy_and_csum_dev(skb, nla_data(nla));
 478
 479         err = genlmsg_unicast(&init_net, user_skb, upcall_info->pid);
 480
 481 out:
 482         kfree_skb(nskb);
 483         return err;
 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 = genl_dereference(dp->table);
 498         new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
 499         if (!new_table)
 500                 return -ENOMEM;
 501
 502         rcu_assign_pointer(dp->table, new_table);
 503
 504         ovs_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_set(const struct nlattr *a,
 540                         const struct sw_flow_key *flow_key)
 541 {
 542         const struct nlattr *ovs_key = nla_data(a);
 543         int key_type = nla_type(ovs_key);
 544
 545         /* There can be only one key in a action */
 546         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 547                 return -EINVAL;
 548
 549         if (key_type > OVS_KEY_ATTR_MAX ||
 550             nla_len(ovs_key) != ovs_key_lens[key_type])
 551                 return -EINVAL;
 552
 553         switch (key_type) {
 554         const struct ovs_key_ipv4 *ipv4_key;
 555
 556         case OVS_KEY_ATTR_PRIORITY:
 557         case OVS_KEY_ATTR_TUN_ID:
 558         case OVS_KEY_ATTR_ETHERNET:
 559                 break;
 560
 561         case OVS_KEY_ATTR_IPV4:
 562                 if (flow_key->eth.type != htons(ETH_P_IP))
 563                         return -EINVAL;
 564
 565                 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
 566                         return -EINVAL;
 567
 568                 ipv4_key = nla_data(ovs_key);
 569                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 570                         return -EINVAL;
 571
 572                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 573                         return -EINVAL;
 574
 575                 break;
 576
 577         case OVS_KEY_ATTR_TCP:
 578                 if (flow_key->ip.proto != IPPROTO_TCP)
 579                         return -EINVAL;
 580
 581                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 582                         return -EINVAL;
 583
 584                 break;
 585
 586         case OVS_KEY_ATTR_UDP:
 587                 if (flow_key->ip.proto != IPPROTO_UDP)
 588                         return -EINVAL;
 589
 590                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 591                         return -EINVAL;
 592                 break;
 593
 594         default:
 595                 return -EINVAL;
 596         }
 597
 598         return 0;
 599 }
 600
 601 static int validate_userspace(const struct nlattr *attr)
 602 {
 603         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 604                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 605                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
 606         };
 607         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 608         int error;
 609
 610         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 611                                  attr, userspace_policy);
 612         if (error)
 613                 return error;
 614
 615         if (!a[OVS_USERSPACE_ATTR_PID] ||
 616             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 617                 return -EINVAL;
 618
 619         return 0;
 620 }
 621
 622 static int validate_actions(const struct nlattr *attr,
 623                                 const struct sw_flow_key *key,  int depth)
 624 {
 625         const struct nlattr *a;
 626         int rem, err;
 627
 628         if (depth >= SAMPLE_ACTION_DEPTH)
 629                 return -EOVERFLOW;
 630
 631         nla_for_each_nested(a, attr, rem) {
 632                 /* Expected argument lengths, (u32)-1 for variable length. */
 633                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 634                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 635                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 636                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 637                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 638                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 639                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 640                 };
 641                 const struct ovs_action_push_vlan *vlan;
 642                 int type = nla_type(a);
 643
 644                 if (type > OVS_ACTION_ATTR_MAX ||
 645                     (action_lens[type] != nla_len(a) &&
 646                      action_lens[type] != (u32)-1))
 647                         return -EINVAL;
 648
 649                 switch (type) {
 650                 case OVS_ACTION_ATTR_UNSPEC:
 651                         return -EINVAL;
 652
 653                 case OVS_ACTION_ATTR_USERSPACE:
 654                         err = validate_userspace(a);
 655                         if (err)
 656                                 return err;
 657                         break;
 658
 659                 case OVS_ACTION_ATTR_OUTPUT:
 660                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 661                                 return -EINVAL;
 662                         break;
 663
 664
 665                 case OVS_ACTION_ATTR_POP_VLAN:
 666                         break;
 667
 668                 case OVS_ACTION_ATTR_PUSH_VLAN:
 669                         vlan = nla_data(a);
 670                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 671                                 return -EINVAL;
 672                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 673                                 return -EINVAL;
 674                         break;
 675
 676                 case OVS_ACTION_ATTR_SET:
 677                         err = validate_set(a, key);
 678                         if (err)
 679                                 return err;
 680                         break;
 681
 682                 case OVS_ACTION_ATTR_SAMPLE:
 683                         err = validate_sample(a, key, depth);
 684                         if (err)
 685                                 return err;
 686                         break;
 687
 688                 default:
 689                         return -EINVAL;
 690                 }
 691         }
 692
 693         if (rem > 0)
 694                 return -EINVAL;
 695
 696         return 0;
 697 }
 698
 699 static void clear_stats(struct sw_flow *flow)
 700 {
 701         flow->used = 0;
 702         flow->tcp_flags = 0;
 703         flow->packet_count = 0;
 704         flow->byte_count = 0;
 705 }
 706
 707 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 708 {
 709         struct ovs_header *ovs_header = info->userhdr;
 710         struct nlattr **a = info->attrs;
 711         struct sw_flow_actions *acts;
 712         struct sk_buff *packet;
 713         struct sw_flow *flow;
 714         struct datapath *dp;
 715         struct ethhdr *eth;
 716         int len;
 717         int err;
 718         int key_len;
 719
 720         err = -EINVAL;
 721         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 722             !a[OVS_PACKET_ATTR_ACTIONS] ||
 723             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 724                 goto err;
 725
 726         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 727         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 728         err = -ENOMEM;
 729         if (!packet)
 730                 goto err;
 731         skb_reserve(packet, NET_IP_ALIGN);
 732
 733         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 734
 735         skb_reset_mac_header(packet);
 736         eth = eth_hdr(packet);
 737
 738         /* Normally, setting the skb 'protocol' field would be handled by a
 739          * call to eth_type_trans(), but it assumes there's a sending
 740          * device, which we may not have. */
 741         if (ntohs(eth->h_proto) >= 1536)
 742                 packet->protocol = eth->h_proto;
 743         else
 744                 packet->protocol = htons(ETH_P_802_2);
 745
 746         /* Build an sw_flow for sending this packet. */
 747         flow = ovs_flow_alloc();
 748         err = PTR_ERR(flow);
 749         if (IS_ERR(flow))
 750                 goto err_kfree_skb;
 751
 752         err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
 753         if (err)
 754                 goto err_flow_put;
 755
 756         err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
 757                                              &flow->key.phy.in_port,
 758                                              &flow->key.phy.tun_id,
 759                                              a[OVS_PACKET_ATTR_KEY]);
 760         if (err)
 761                 goto err_flow_put;
 762
 763         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
 764         if (err)
 765                 goto err_flow_put;
 766
 767         flow->hash = ovs_flow_hash(&flow->key, key_len);
 768
 769         acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 770         err = PTR_ERR(acts);
 771         if (IS_ERR(acts))
 772                 goto err_flow_put;
 773         rcu_assign_pointer(flow->sf_acts, acts);
 774
 775         OVS_CB(packet)->flow = flow;
 776         packet->priority = flow->key.phy.priority;
 777
 778         rcu_read_lock();
 779         dp = get_dp(ovs_header->dp_ifindex);
 780         err = -ENODEV;
 781         if (!dp)
 782                 goto err_unlock;
 783
 784         local_bh_disable();
 785         err = ovs_execute_actions(dp, packet);
 786         local_bh_enable();
 787         rcu_read_unlock();
 788
 789         ovs_flow_put(flow);
 790         return err;
 791
 792 err_unlock:
 793         rcu_read_unlock();
 794 err_flow_put:
 795         ovs_flow_put(flow);
 796 err_kfree_skb:
 797         kfree_skb(packet);
 798 err:
 799         return err;
 800 }
 801
 802 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 803         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 804         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 805         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 806 };
 807
 808 static struct genl_ops dp_packet_genl_ops[] = {
 809         { .cmd = OVS_PACKET_CMD_EXECUTE,
 810           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 811           .policy = packet_policy,
 812           .doit = ovs_packet_cmd_execute
 813         }
 814 };
 815
 816 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 817 {
 818         int i;
 819         struct flow_table *table = genl_dereference(dp->table);
 820
 821         stats->n_flows = ovs_flow_tbl_count(table);
 822
 823         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 824         for_each_possible_cpu(i) {
 825                 const struct dp_stats_percpu *percpu_stats;
 826                 struct dp_stats_percpu local_stats;
 827                 unsigned int start;
 828
 829                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 830
 831                 do {
 832                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 833                         local_stats = *percpu_stats;
 834                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 835
 836                 stats->n_hit += local_stats.n_hit;
 837                 stats->n_missed += local_stats.n_missed;
 838                 stats->n_lost += local_stats.n_lost;
 839         }
 840 }
 841
 842 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 843         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 844         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 845         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 846 };
 847
 848 static struct genl_family dp_flow_genl_family = {
 849         .id = GENL_ID_GENERATE,
 850         .hdrsize = sizeof(struct ovs_header),
 851         .name = OVS_FLOW_FAMILY,
 852         .version = OVS_FLOW_VERSION,
 853         .maxattr = OVS_FLOW_ATTR_MAX
 854 };
 855
 856 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 857         .name = OVS_FLOW_MCGROUP
 858 };
 859
 860 /* Called with genl_lock. */
 861 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 862                                   struct sk_buff *skb, u32 pid,
 863                                   u32 seq, u32 flags, u8 cmd)
 864 {
 865         const int skb_orig_len = skb->len;
 866         const struct sw_flow_actions *sf_acts;
 867         struct ovs_flow_stats stats;
 868         struct ovs_header *ovs_header;
 869         struct nlattr *nla;
 870         unsigned long used;
 871         u8 tcp_flags;
 872         int err;
 873
 874         sf_acts = rcu_dereference_protected(flow->sf_acts,
 875                                             lockdep_genl_is_held());
 876
 877         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 878         if (!ovs_header)
 879                 return -EMSGSIZE;
 880
 881         ovs_header->dp_ifindex = get_dpifindex(dp);
 882
 883         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 884         if (!nla)
 885                 goto nla_put_failure;
 886         err = ovs_flow_to_nlattrs(&flow->key, skb);
 887         if (err)
 888                 goto error;
 889         nla_nest_end(skb, nla);
 890
 891         spin_lock_bh(&flow->lock);
 892         used = flow->used;
 893         stats.n_packets = flow->packet_count;
 894         stats.n_bytes = flow->byte_count;
 895         tcp_flags = flow->tcp_flags;
 896         spin_unlock_bh(&flow->lock);
 897
 898         if (used)
 899                 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used));
 900
 901         if (stats.n_packets)
 902                 NLA_PUT(skb, OVS_FLOW_ATTR_STATS,
 903                         sizeof(struct ovs_flow_stats), &stats);
 904
 905         if (tcp_flags)
 906                 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 907
 908         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 909          * this is the first flow to be dumped into 'skb'.  This is unusual for
 910          * Netlink but individual action lists can be longer than
 911          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 912          * The userspace caller can always fetch the actions separately if it
 913          * really wants them.  (Most userspace callers in fact don't care.)
 914          *
 915          * This can only fail for dump operations because the skb is always
 916          * properly sized for single flows.
 917          */
 918         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 919                       sf_acts->actions);
 920         if (err < 0 && skb_orig_len)
 921                 goto error;
 922
 923         return genlmsg_end(skb, ovs_header);
 924
 925 nla_put_failure:
 926         err = -EMSGSIZE;
 927 error:
 928         genlmsg_cancel(skb, ovs_header);
 929         return err;
 930 }
 931
 932 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 933 {
 934         const struct sw_flow_actions *sf_acts;
 935         int len;
 936
 937         sf_acts = rcu_dereference_protected(flow->sf_acts,
 938                                             lockdep_genl_is_held());
 939
 940         /* OVS_FLOW_ATTR_KEY */
 941         len = nla_total_size(FLOW_BUFSIZE);
 942         /* OVS_FLOW_ATTR_ACTIONS */
 943         len += nla_total_size(sf_acts->actions_len);
 944         /* OVS_FLOW_ATTR_STATS */
 945         len += nla_total_size(sizeof(struct ovs_flow_stats));
 946         /* OVS_FLOW_ATTR_TCP_FLAGS */
 947         len += nla_total_size(1);
 948         /* OVS_FLOW_ATTR_USED */
 949         len += nla_total_size(8);
 950
 951         len += NLMSG_ALIGN(sizeof(struct ovs_header));
 952
 953         return genlmsg_new(len, GFP_KERNEL);
 954 }
 955
 956 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 957                                                struct datapath *dp,
 958                                                u32 pid, u32 seq, u8 cmd)
 959 {
 960         struct sk_buff *skb;
 961         int retval;
 962
 963         skb = ovs_flow_cmd_alloc_info(flow);
 964         if (!skb)
 965                 return ERR_PTR(-ENOMEM);
 966
 967         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 968         BUG_ON(retval < 0);
 969         return skb;
 970 }
 971
 972 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 973 {
 974         struct nlattr **a = info->attrs;
 975         struct ovs_header *ovs_header = info->userhdr;
 976         struct sw_flow_key key;
 977         struct sw_flow *flow;
 978         struct sk_buff *reply;
 979         struct datapath *dp;
 980         struct flow_table *table;
 981         int error;
 982         int key_len;
 983
 984         /* Extract key. */
 985         error = -EINVAL;
 986         if (!a[OVS_FLOW_ATTR_KEY])
 987                 goto error;
 988         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
 989         if (error)
 990                 goto error;
 991
 992         /* Validate actions. */
 993         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 994                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
 995                 if (error)
 996                         goto error;
 997         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 998                 error = -EINVAL;
 999                 goto error;
1000         }
1001
1002         dp = get_dp(ovs_header->dp_ifindex);
1003         error = -ENODEV;
1004         if (!dp)
1005                 goto error;
1006
1007         table = genl_dereference(dp->table);
1008         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1009         if (!flow) {
1010                 struct sw_flow_actions *acts;
1011
1012                 /* Bail out if we're not allowed to create a new flow. */
1013                 error = -ENOENT;
1014                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1015                         goto error;
1016
1017                 /* Expand table, if necessary, to make room. */
1018                 if (ovs_flow_tbl_need_to_expand(table)) {
1019                         struct flow_table *new_table;
1020
1021                         new_table = ovs_flow_tbl_expand(table);
1022                         if (!IS_ERR(new_table)) {
1023                                 rcu_assign_pointer(dp->table, new_table);
1024                                 ovs_flow_tbl_deferred_destroy(table);
1025                                 table = genl_dereference(dp->table);
1026                         }
1027                 }
1028
1029                 /* Allocate flow. */
1030                 flow = ovs_flow_alloc();
1031                 if (IS_ERR(flow)) {
1032                         error = PTR_ERR(flow);
1033                         goto error;
1034                 }
1035                 flow->key = key;
1036                 clear_stats(flow);
1037
1038                 /* Obtain actions. */
1039                 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1040                 error = PTR_ERR(acts);
1041                 if (IS_ERR(acts))
1042                         goto error_free_flow;
1043                 rcu_assign_pointer(flow->sf_acts, acts);
1044
1045                 /* Put flow in bucket. */
1046                 flow->hash = ovs_flow_hash(&key, key_len);
1047                 ovs_flow_tbl_insert(table, flow);
1048
1049                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1050                                                 info->snd_seq,
1051                                                 OVS_FLOW_CMD_NEW);
1052         } else {
1053                 /* We found a matching flow. */
1054                 struct sw_flow_actions *old_acts;
1055                 struct nlattr *acts_attrs;
1056
1057                 /* Bail out if we're not allowed to modify an existing flow.
1058                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1059                  * because Generic Netlink treats the latter as a dump
1060                  * request.  We also accept NLM_F_EXCL in case that bug ever
1061                  * gets fixed.
1062                  */
1063                 error = -EEXIST;
1064                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1065                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1066                         goto error;
1067
1068                 /* Update actions. */
1069                 old_acts = rcu_dereference_protected(flow->sf_acts,
1070                                                      lockdep_genl_is_held());
1071                 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1072                 if (acts_attrs &&
1073                    (old_acts->actions_len != nla_len(acts_attrs) ||
1074                    memcmp(old_acts->actions, nla_data(acts_attrs),
1075                           old_acts->actions_len))) {
1076                         struct sw_flow_actions *new_acts;
1077
1078                         new_acts = ovs_flow_actions_alloc(acts_attrs);
1079                         error = PTR_ERR(new_acts);
1080                         if (IS_ERR(new_acts))
1081                                 goto error;
1082
1083                         rcu_assign_pointer(flow->sf_acts, new_acts);
1084                         ovs_flow_deferred_free_acts(old_acts);
1085                 }
1086
1087                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1088                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1089
1090                 /* Clear stats. */
1091                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1092                         spin_lock_bh(&flow->lock);
1093                         clear_stats(flow);
1094                         spin_unlock_bh(&flow->lock);
1095                 }
1096         }
1097
1098         if (!IS_ERR(reply))
1099                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1100                            ovs_dp_flow_multicast_group.id, info->nlhdr,
1101                            GFP_KERNEL);
1102         else
1103                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1104                                 ovs_dp_flow_multicast_group.id,
1105                                 PTR_ERR(reply));
1106         return 0;
1107
1108 error_free_flow:
1109         ovs_flow_put(flow);
1110 error:
1111         return error;
1112 }
1113
1114 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1115 {
1116         struct nlattr **a = info->attrs;
1117         struct ovs_header *ovs_header = info->userhdr;
1118         struct sw_flow_key key;
1119         struct sk_buff *reply;
1120         struct sw_flow *flow;
1121         struct datapath *dp;
1122         struct flow_table *table;
1123         int err;
1124         int key_len;
1125
1126         if (!a[OVS_FLOW_ATTR_KEY])
1127                 return -EINVAL;
1128         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1129         if (err)
1130                 return err;
1131
1132         dp = get_dp(ovs_header->dp_ifindex);
1133         if (!dp)
1134                 return -ENODEV;
1135
1136         table = genl_dereference(dp->table);
1137         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1138         if (!flow)
1139                 return -ENOENT;
1140
1141         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1142                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1143         if (IS_ERR(reply))
1144                 return PTR_ERR(reply);
1145
1146         return genlmsg_reply(reply, info);
1147 }
1148
1149 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1150 {
1151         struct nlattr **a = info->attrs;
1152         struct ovs_header *ovs_header = info->userhdr;
1153         struct sw_flow_key key;
1154         struct sk_buff *reply;
1155         struct sw_flow *flow;
1156         struct datapath *dp;
1157         struct flow_table *table;
1158         int err;
1159         int key_len;
1160
1161         if (!a[OVS_FLOW_ATTR_KEY])
1162                 return flush_flows(ovs_header->dp_ifindex);
1163         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1164         if (err)
1165                 return err;
1166
1167         dp = get_dp(ovs_header->dp_ifindex);
1168         if (!dp)
1169                 return -ENODEV;
1170
1171         table = genl_dereference(dp->table);
1172         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1173         if (!flow)
1174                 return -ENOENT;
1175
1176         reply = ovs_flow_cmd_alloc_info(flow);
1177         if (!reply)
1178                 return -ENOMEM;
1179
1180         ovs_flow_tbl_remove(table, flow);
1181
1182         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1183                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1184         BUG_ON(err < 0);
1185
1186         ovs_flow_deferred_free(flow);
1187
1188         genl_notify(reply, genl_info_net(info), info->snd_pid,
1189                     ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1190         return 0;
1191 }
1192
1193 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1194 {
1195         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1196         struct datapath *dp;
1197         struct flow_table *table;
1198
1199         dp = get_dp(ovs_header->dp_ifindex);
1200         if (!dp)
1201                 return -ENODEV;
1202
1203         table = genl_dereference(dp->table);
1204
1205         for (;;) {
1206                 struct sw_flow *flow;
1207                 u32 bucket, obj;
1208
1209                 bucket = cb->args[0];
1210                 obj = cb->args[1];
1211                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1212                 if (!flow)
1213                         break;
1214
1215                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1216                                            NETLINK_CB(cb->skb).pid,
1217                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1218                                            OVS_FLOW_CMD_NEW) < 0)
1219                         break;
1220
1221                 cb->args[0] = bucket;
1222                 cb->args[1] = obj;
1223         }
1224         return skb->len;
1225 }
1226
1227 static struct genl_ops dp_flow_genl_ops[] = {
1228         { .cmd = OVS_FLOW_CMD_NEW,
1229           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1230           .policy = flow_policy,
1231           .doit = ovs_flow_cmd_new_or_set
1232         },
1233         { .cmd = OVS_FLOW_CMD_DEL,
1234           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1235           .policy = flow_policy,
1236           .doit = ovs_flow_cmd_del
1237         },
1238         { .cmd = OVS_FLOW_CMD_GET,
1239           .flags = 0,               /* OK for unprivileged users. */
1240           .policy = flow_policy,
1241           .doit = ovs_flow_cmd_get,
1242           .dumpit = ovs_flow_cmd_dump
1243         },
1244         { .cmd = OVS_FLOW_CMD_SET,
1245           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1246           .policy = flow_policy,
1247           .doit = ovs_flow_cmd_new_or_set,
1248         },
1249 };
1250
1251 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1252 #ifdef HAVE_NLA_NUL_STRING
1253         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1254 #endif
1255         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1256 };
1257
1258 static struct genl_family dp_datapath_genl_family = {
1259         .id = GENL_ID_GENERATE,
1260         .hdrsize = sizeof(struct ovs_header),
1261         .name = OVS_DATAPATH_FAMILY,
1262         .version = OVS_DATAPATH_VERSION,
1263         .maxattr = OVS_DP_ATTR_MAX
1264 };
1265
1266 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1267         .name = OVS_DATAPATH_MCGROUP
1268 };
1269
1270 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1271                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1272 {
1273         struct ovs_header *ovs_header;
1274         struct ovs_dp_stats dp_stats;
1275         int err;
1276
1277         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1278                                    flags, cmd);
1279         if (!ovs_header)
1280                 goto error;
1281
1282         ovs_header->dp_ifindex = get_dpifindex(dp);
1283
1284         rcu_read_lock();
1285         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1286         rcu_read_unlock();
1287         if (err)
1288                 goto nla_put_failure;
1289
1290         get_dp_stats(dp, &dp_stats);
1291         NLA_PUT(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats);
1292
1293         return genlmsg_end(skb, ovs_header);
1294
1295 nla_put_failure:
1296         genlmsg_cancel(skb, ovs_header);
1297 error:
1298         return -EMSGSIZE;
1299 }
1300
1301 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1302                                              u32 seq, u8 cmd)
1303 {
1304         struct sk_buff *skb;
1305         int retval;
1306
1307         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1308         if (!skb)
1309                 return ERR_PTR(-ENOMEM);
1310
1311         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1312         if (retval < 0) {
1313                 kfree_skb(skb);
1314                 return ERR_PTR(retval);
1315         }
1316         return skb;
1317 }
1318
1319 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1320 {
1321         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1322 }
1323
1324 /* Called with genl_mutex and optionally with RTNL lock also. */
1325 static struct datapath *lookup_datapath(struct ovs_header *ovs_header,
1326                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1327 {
1328         struct datapath *dp;
1329
1330         if (!a[OVS_DP_ATTR_NAME])
1331                 dp = get_dp(ovs_header->dp_ifindex);
1332         else {
1333                 struct vport *vport;
1334
1335                 rcu_read_lock();
1336                 vport = ovs_vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1337                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1338                 rcu_read_unlock();
1339         }
1340         return dp ? dp : ERR_PTR(-ENODEV);
1341 }
1342
1343 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1344 {
1345         struct nlattr **a = info->attrs;
1346         struct vport_parms parms;
1347         struct sk_buff *reply;
1348         struct datapath *dp;
1349         struct vport *vport;
1350         int err;
1351
1352         err = -EINVAL;
1353         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1354                 goto err;
1355
1356         err = ovs_dp_cmd_validate(a);
1357         if (err)
1358                 goto err;
1359
1360         rtnl_lock();
1361         err = -ENODEV;
1362         if (!try_module_get(THIS_MODULE))
1363                 goto err_unlock_rtnl;
1364
1365         err = -ENOMEM;
1366         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1367         if (dp == NULL)
1368                 goto err_put_module;
1369         INIT_LIST_HEAD(&dp->port_list);
1370
1371         /* Initialize kobject for bridge.  This will be added as
1372          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1373         dp->ifobj.kset = NULL;
1374         kobject_init(&dp->ifobj, &dp_ktype);
1375
1376         /* Allocate table. */
1377         err = -ENOMEM;
1378         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1379         if (!dp->table)
1380                 goto err_free_dp;
1381
1382         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1383         if (!dp->stats_percpu) {
1384                 err = -ENOMEM;
1385                 goto err_destroy_table;
1386         }
1387
1388         /* Set up our datapath device. */
1389         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1390         parms.type = OVS_VPORT_TYPE_INTERNAL;
1391         parms.options = NULL;
1392         parms.dp = dp;
1393         parms.port_no = OVSP_LOCAL;
1394         parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1395
1396         vport = new_vport(&parms);
1397         if (IS_ERR(vport)) {
1398                 err = PTR_ERR(vport);
1399                 if (err == -EBUSY)
1400                         err = -EEXIST;
1401
1402                 goto err_destroy_percpu;
1403         }
1404
1405         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1406                                       info->snd_seq, OVS_DP_CMD_NEW);
1407         err = PTR_ERR(reply);
1408         if (IS_ERR(reply))
1409                 goto err_destroy_local_port;
1410
1411         list_add_tail(&dp->list_node, &dps);
1412         ovs_dp_sysfs_add_dp(dp);
1413
1414         rtnl_unlock();
1415
1416         genl_notify(reply, genl_info_net(info), info->snd_pid,
1417                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1418                     GFP_KERNEL);
1419         return 0;
1420
1421 err_destroy_local_port:
1422         ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL]));
1423 err_destroy_percpu:
1424         free_percpu(dp->stats_percpu);
1425 err_destroy_table:
1426         ovs_flow_tbl_destroy(genl_dereference(dp->table));
1427 err_free_dp:
1428         kfree(dp);
1429 err_put_module:
1430         module_put(THIS_MODULE);
1431 err_unlock_rtnl:
1432         rtnl_unlock();
1433 err:
1434         return err;
1435 }
1436
1437 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1438 {
1439         struct vport *vport, *next_vport;
1440         struct sk_buff *reply;
1441         struct datapath *dp;
1442         int err;
1443
1444         err = ovs_dp_cmd_validate(info->attrs);
1445         if (err)
1446                 goto exit;
1447
1448         rtnl_lock();
1449         dp = lookup_datapath(info->userhdr, info->attrs);
1450         err = PTR_ERR(dp);
1451         if (IS_ERR(dp))
1452                 goto exit_unlock;
1453
1454         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1455                                       info->snd_seq, OVS_DP_CMD_DEL);
1456         err = PTR_ERR(reply);
1457         if (IS_ERR(reply))
1458                 goto exit_unlock;
1459
1460         list_for_each_entry_safe(vport, next_vport, &dp->port_list, node)
1461                 if (vport->port_no != OVSP_LOCAL)
1462                         ovs_dp_detach_port(vport);
1463
1464         ovs_dp_sysfs_del_dp(dp);
1465         list_del(&dp->list_node);
1466         ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL]));
1467
1468         /* rtnl_unlock() will wait until all the references to devices that
1469          * are pending unregistration have been dropped.  We do it here to
1470          * ensure that any internal devices (which contain DP pointers) are
1471          * fully destroyed before freeing the datapath.
1472          */
1473         rtnl_unlock();
1474
1475         call_rcu(&dp->rcu, destroy_dp_rcu);
1476         module_put(THIS_MODULE);
1477
1478         genl_notify(reply, genl_info_net(info), info->snd_pid,
1479                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1480                     GFP_KERNEL);
1481
1482         return 0;
1483
1484 exit_unlock:
1485         rtnl_unlock();
1486 exit:
1487         return err;
1488 }
1489
1490 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1491 {
1492         struct sk_buff *reply;
1493         struct datapath *dp;
1494         int err;
1495
1496         err = ovs_dp_cmd_validate(info->attrs);
1497         if (err)
1498                 return err;
1499
1500         dp = lookup_datapath(info->userhdr, info->attrs);
1501         if (IS_ERR(dp))
1502                 return PTR_ERR(dp);
1503
1504         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1505                                       info->snd_seq, OVS_DP_CMD_NEW);
1506         if (IS_ERR(reply)) {
1507                 err = PTR_ERR(reply);
1508                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1509                                 ovs_dp_datapath_multicast_group.id, err);
1510                 return 0;
1511         }
1512
1513         genl_notify(reply, genl_info_net(info), info->snd_pid,
1514                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1515                     GFP_KERNEL);
1516
1517         return 0;
1518 }
1519
1520 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1521 {
1522         struct sk_buff *reply;
1523         struct datapath *dp;
1524         int err;
1525
1526         err = ovs_dp_cmd_validate(info->attrs);
1527         if (err)
1528                 return err;
1529
1530         dp = lookup_datapath(info->userhdr, info->attrs);
1531         if (IS_ERR(dp))
1532                 return PTR_ERR(dp);
1533
1534         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1535                                       info->snd_seq, OVS_DP_CMD_NEW);
1536         if (IS_ERR(reply))
1537                 return PTR_ERR(reply);
1538
1539         return genlmsg_reply(reply, info);
1540 }
1541
1542 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1543 {
1544         struct datapath *dp;
1545         int skip = cb->args[0];
1546         int i = 0;
1547
1548         list_for_each_entry(dp, &dps, list_node) {
1549                 if (i < skip)
1550                         continue;
1551                 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1552                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1553                                          OVS_DP_CMD_NEW) < 0)
1554                         break;
1555                 i++;
1556         }
1557
1558         cb->args[0] = i;
1559
1560         return skb->len;
1561 }
1562
1563 static struct genl_ops dp_datapath_genl_ops[] = {
1564         { .cmd = OVS_DP_CMD_NEW,
1565           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1566           .policy = datapath_policy,
1567           .doit = ovs_dp_cmd_new
1568         },
1569         { .cmd = OVS_DP_CMD_DEL,
1570           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1571           .policy = datapath_policy,
1572           .doit = ovs_dp_cmd_del
1573         },
1574         { .cmd = OVS_DP_CMD_GET,
1575           .flags = 0,               /* OK for unprivileged users. */
1576           .policy = datapath_policy,
1577           .doit = ovs_dp_cmd_get,
1578           .dumpit = ovs_dp_cmd_dump
1579         },
1580         { .cmd = OVS_DP_CMD_SET,
1581           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1582           .policy = datapath_policy,
1583           .doit = ovs_dp_cmd_set,
1584         },
1585 };
1586
1587 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1588 #ifdef HAVE_NLA_NUL_STRING
1589         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1590         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1591         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1592 #else
1593         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1594         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1595 #endif
1596         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1597         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1598         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1599         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1600 };
1601
1602 static struct genl_family dp_vport_genl_family = {
1603         .id = GENL_ID_GENERATE,
1604         .hdrsize = sizeof(struct ovs_header),
1605         .name = OVS_VPORT_FAMILY,
1606         .version = OVS_VPORT_VERSION,
1607         .maxattr = OVS_VPORT_ATTR_MAX
1608 };
1609
1610 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1611         .name = OVS_VPORT_MCGROUP
1612 };
1613
1614 /* Called with RTNL lock or RCU read lock. */
1615 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1616                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1617 {
1618         struct ovs_header *ovs_header;
1619         struct ovs_vport_stats vport_stats;
1620         int err;
1621
1622         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1623                                  flags, cmd);
1624         if (!ovs_header)
1625                 return -EMSGSIZE;
1626
1627         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1628
1629         NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1630         NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type);
1631         NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport));
1632         NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1633
1634         ovs_vport_get_stats(vport, &vport_stats);
1635         NLA_PUT(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1636                 &vport_stats);
1637
1638         NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN,
1639                 vport->ops->get_addr(vport));
1640
1641         err = ovs_vport_get_options(vport, skb);
1642         if (err == -EMSGSIZE)
1643                 goto error;
1644
1645         return genlmsg_end(skb, ovs_header);
1646
1647 nla_put_failure:
1648         err = -EMSGSIZE;
1649 error:
1650         genlmsg_cancel(skb, ovs_header);
1651         return err;
1652 }
1653
1654 /* Called with RTNL lock or RCU read lock. */
1655 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1656                                          u32 seq, u8 cmd)
1657 {
1658         struct sk_buff *skb;
1659         int retval;
1660
1661         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1662         if (!skb)
1663                 return ERR_PTR(-ENOMEM);
1664
1665         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1666         if (retval < 0) {
1667                 kfree_skb(skb);
1668                 return ERR_PTR(retval);
1669         }
1670         return skb;
1671 }
1672
1673 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1674 {
1675         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1676 }
1677
1678 /* Called with RTNL lock or RCU read lock. */
1679 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1680                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1681 {
1682         struct datapath *dp;
1683         struct vport *vport;
1684
1685         if (a[OVS_VPORT_ATTR_NAME]) {
1686                 vport = ovs_vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1687                 if (!vport)
1688                         return ERR_PTR(-ENODEV);
1689                 return vport;
1690         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1691                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1692
1693                 if (port_no >= DP_MAX_PORTS)
1694                         return ERR_PTR(-EFBIG);
1695
1696                 dp = get_dp(ovs_header->dp_ifindex);
1697                 if (!dp)
1698                         return ERR_PTR(-ENODEV);
1699
1700                 vport = rcu_dereference_rtnl(dp->ports[port_no]);
1701                 if (!vport)
1702                         return ERR_PTR(-ENOENT);
1703                 return vport;
1704         } else
1705                 return ERR_PTR(-EINVAL);
1706 }
1707
1708 /* Called with RTNL lock. */
1709 static int change_vport(struct vport *vport,
1710                         struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1711 {
1712         int err = 0;
1713
1714         if (a[OVS_VPORT_ATTR_STATS])
1715                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1716
1717         if (a[OVS_VPORT_ATTR_ADDRESS])
1718                 err = ovs_vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1719
1720         return err;
1721 }
1722
1723 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1724 {
1725         struct nlattr **a = info->attrs;
1726         struct ovs_header *ovs_header = info->userhdr;
1727         struct vport_parms parms;
1728         struct sk_buff *reply;
1729         struct vport *vport;
1730         struct datapath *dp;
1731         u32 port_no;
1732         int err;
1733
1734         err = -EINVAL;
1735         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1736             !a[OVS_VPORT_ATTR_UPCALL_PID])
1737                 goto exit;
1738
1739         err = ovs_vport_cmd_validate(a);
1740         if (err)
1741                 goto exit;
1742
1743         rtnl_lock();
1744         dp = get_dp(ovs_header->dp_ifindex);
1745         err = -ENODEV;
1746         if (!dp)
1747                 goto exit_unlock;
1748
1749         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1750                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1751
1752                 err = -EFBIG;
1753                 if (port_no >= DP_MAX_PORTS)
1754                         goto exit_unlock;
1755
1756                 vport = rtnl_dereference(dp->ports[port_no]);
1757                 err = -EBUSY;
1758                 if (vport)
1759                         goto exit_unlock;
1760         } else {
1761                 for (port_no = 1; ; port_no++) {
1762                         if (port_no >= DP_MAX_PORTS) {
1763                                 err = -EFBIG;
1764                                 goto exit_unlock;
1765                         }
1766                         vport = rtnl_dereference(dp->ports[port_no]);
1767                         if (!vport)
1768                                 break;
1769                 }
1770         }
1771
1772         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1773         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1774         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1775         parms.dp = dp;
1776         parms.port_no = port_no;
1777         parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1778
1779         vport = new_vport(&parms);
1780         err = PTR_ERR(vport);
1781         if (IS_ERR(vport))
1782                 goto exit_unlock;
1783
1784         ovs_dp_sysfs_add_if(vport);
1785
1786         err = change_vport(vport, a);
1787         if (!err) {
1788                 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1789                                                  info->snd_seq,
1790                                                  OVS_VPORT_CMD_NEW);
1791                 if (IS_ERR(reply))
1792                         err = PTR_ERR(reply);
1793         }
1794         if (err) {
1795                 ovs_dp_detach_port(vport);
1796                 goto exit_unlock;
1797         }
1798         genl_notify(reply, genl_info_net(info), info->snd_pid,
1799                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1800
1801 exit_unlock:
1802         rtnl_unlock();
1803 exit:
1804         return err;
1805 }
1806
1807 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1808 {
1809         struct nlattr **a = info->attrs;
1810         struct sk_buff *reply;
1811         struct vport *vport;
1812         int err;
1813
1814         err = ovs_vport_cmd_validate(a);
1815         if (err)
1816                 goto exit;
1817
1818         rtnl_lock();
1819         vport = lookup_vport(info->userhdr, a);
1820         err = PTR_ERR(vport);
1821         if (IS_ERR(vport))
1822                 goto exit_unlock;
1823
1824         err = 0;
1825         if (a[OVS_VPORT_ATTR_TYPE] &&
1826             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1827                 err = -EINVAL;
1828
1829         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1830                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1831         if (!err)
1832                 err = change_vport(vport, a);
1833         if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1834                 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1835
1836         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1837                                          OVS_VPORT_CMD_NEW);
1838         if (IS_ERR(reply)) {
1839                 err = PTR_ERR(reply);
1840                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1841                                 ovs_dp_vport_multicast_group.id, err);
1842                 return 0;
1843         }
1844
1845         genl_notify(reply, genl_info_net(info), info->snd_pid,
1846                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1847
1848 exit_unlock:
1849         rtnl_unlock();
1850 exit:
1851         return err;
1852 }
1853
1854 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1855 {
1856         struct nlattr **a = info->attrs;
1857         struct sk_buff *reply;
1858         struct vport *vport;
1859         int err;
1860
1861         err = ovs_vport_cmd_validate(a);
1862         if (err)
1863                 goto exit;
1864
1865         rtnl_lock();
1866         vport = lookup_vport(info->userhdr, a);
1867         err = PTR_ERR(vport);
1868         if (IS_ERR(vport))
1869                 goto exit_unlock;
1870
1871         if (vport->port_no == OVSP_LOCAL) {
1872                 err = -EINVAL;
1873                 goto exit_unlock;
1874         }
1875
1876         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1877                                          OVS_VPORT_CMD_DEL);
1878         err = PTR_ERR(reply);
1879         if (IS_ERR(reply))
1880                 goto exit_unlock;
1881
1882         ovs_dp_detach_port(vport);
1883
1884         genl_notify(reply, genl_info_net(info), info->snd_pid,
1885                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1886
1887 exit_unlock:
1888         rtnl_unlock();
1889 exit:
1890         return err;
1891 }
1892
1893 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1894 {
1895         struct nlattr **a = info->attrs;
1896         struct ovs_header *ovs_header = info->userhdr;
1897         struct sk_buff *reply;
1898         struct vport *vport;
1899         int err;
1900
1901         err = ovs_vport_cmd_validate(a);
1902         if (err)
1903                 goto exit;
1904
1905         rcu_read_lock();
1906         vport = lookup_vport(ovs_header, a);
1907         err = PTR_ERR(vport);
1908         if (IS_ERR(vport))
1909                 goto exit_unlock;
1910
1911         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1912                                          OVS_VPORT_CMD_NEW);
1913         err = PTR_ERR(reply);
1914         if (IS_ERR(reply))
1915                 goto exit_unlock;
1916
1917         rcu_read_unlock();
1918
1919         return genlmsg_reply(reply, info);
1920
1921 exit_unlock:
1922         rcu_read_unlock();
1923 exit:
1924         return err;
1925 }
1926
1927 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1928 {
1929         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1930         struct datapath *dp;
1931         u32 port_no;
1932         int retval;
1933
1934         dp = get_dp(ovs_header->dp_ifindex);
1935         if (!dp)
1936                 return -ENODEV;
1937
1938         rcu_read_lock();
1939         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1940                 struct vport *vport;
1941
1942                 vport = rcu_dereference(dp->ports[port_no]);
1943                 if (!vport)
1944                         continue;
1945
1946                 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1947                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1948                                             OVS_VPORT_CMD_NEW) < 0)
1949                         break;
1950         }
1951         rcu_read_unlock();
1952
1953         cb->args[0] = port_no;
1954         retval = skb->len;
1955
1956         return retval;
1957 }
1958
1959 static struct genl_ops dp_vport_genl_ops[] = {
1960         { .cmd = OVS_VPORT_CMD_NEW,
1961           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1962           .policy = vport_policy,
1963           .doit = ovs_vport_cmd_new
1964         },
1965         { .cmd = OVS_VPORT_CMD_DEL,
1966           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1967           .policy = vport_policy,
1968           .doit = ovs_vport_cmd_del
1969         },
1970         { .cmd = OVS_VPORT_CMD_GET,
1971           .flags = 0,               /* OK for unprivileged users. */
1972           .policy = vport_policy,
1973           .doit = ovs_vport_cmd_get,
1974           .dumpit = ovs_vport_cmd_dump
1975         },
1976         { .cmd = OVS_VPORT_CMD_SET,
1977           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1978           .policy = vport_policy,
1979           .doit = ovs_vport_cmd_set,
1980         },
1981 };
1982
1983 struct genl_family_and_ops {
1984         struct genl_family *family;
1985         struct genl_ops *ops;
1986         int n_ops;
1987         struct genl_multicast_group *group;
1988 };
1989
1990 static const struct genl_family_and_ops dp_genl_families[] = {
1991         { &dp_datapath_genl_family,
1992           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1993           &ovs_dp_datapath_multicast_group },
1994         { &dp_vport_genl_family,
1995           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1996           &ovs_dp_vport_multicast_group },
1997         { &dp_flow_genl_family,
1998           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1999           &ovs_dp_flow_multicast_group },
2000         { &dp_packet_genl_family,
2001           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2002           NULL },
2003 };
2004
2005 static void dp_unregister_genl(int n_families)
2006 {
2007         int i;
2008
2009         for (i = 0; i < n_families; i++)
2010                 genl_unregister_family(dp_genl_families[i].family);
2011 }
2012
2013 static int dp_register_genl(void)
2014 {
2015         int n_registered;
2016         int err;
2017         int i;
2018
2019         n_registered = 0;
2020         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2021                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2022
2023                 err = genl_register_family_with_ops(f->family, f->ops,
2024                                                     f->n_ops);
2025                 if (err)
2026                         goto error;
2027                 n_registered++;
2028
2029                 if (f->group) {
2030                         err = genl_register_mc_group(f->family, f->group);
2031                         if (err)
2032                                 goto error;
2033                 }
2034         }
2035
2036         return 0;
2037
2038 error:
2039         dp_unregister_genl(n_registered);
2040         return err;
2041 }
2042
2043 static int __init dp_init(void)
2044 {
2045         struct sk_buff *dummy_skb;
2046         int err;
2047
2048         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2049
2050         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2051                 VERSION BUILDNR);
2052
2053         err = genl_exec_init();
2054         if (err)
2055                 goto error;
2056
2057         err = ovs_workqueues_init();
2058         if (err)
2059                 goto error_genl_exec;
2060
2061         err = ovs_tnl_init();
2062         if (err)
2063                 goto error_wq;
2064
2065         err = ovs_flow_init();
2066         if (err)
2067                 goto error_tnl_exit;
2068
2069         err = ovs_vport_init();
2070         if (err)
2071                 goto error_flow_exit;
2072
2073         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2074         if (err)
2075                 goto error_vport_exit;
2076
2077         err = dp_register_genl();
2078         if (err < 0)
2079                 goto error_unreg_notifier;
2080
2081         return 0;
2082
2083 error_unreg_notifier:
2084         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2085 error_vport_exit:
2086         ovs_vport_exit();
2087 error_flow_exit:
2088         ovs_flow_exit();
2089 error_tnl_exit:
2090         ovs_tnl_exit();
2091 error_wq:
2092         ovs_workqueues_exit();
2093 error_genl_exec:
2094         genl_exec_exit();
2095 error:
2096         return err;
2097 }
2098
2099 static void dp_cleanup(void)
2100 {
2101         rcu_barrier();
2102         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2103         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2104         ovs_vport_exit();
2105         ovs_flow_exit();
2106         ovs_tnl_exit();
2107         ovs_workqueues_exit();
2108         genl_exec_exit();
2109 }
2110
2111 module_init(dp_init);
2112 module_exit(dp_cleanup);
2113
2114 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2115 MODULE_LICENSE("GPL");