datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007-2013 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/version.h>
  40 #include <linux/ethtool.h>
  41 #include <linux/wait.h>
  42 #include <asm/div64.h>
  43 #include <linux/highmem.h>
  44 #include <linux/netfilter_bridge.h>
  45 #include <linux/netfilter_ipv4.h>
  46 #include <linux/inetdevice.h>
  47 #include <linux/list.h>
  48 #include <linux/openvswitch.h>
  49 #include <linux/rculist.h>
  50 #include <linux/dmi.h>
  51 #include <linux/genetlink.h>
  52 #include <net/genetlink.h>
  53 #include <net/genetlink.h>
  54 #include <net/net_namespace.h>
  55 #include <net/netns/generic.h>
  56
  57 #include "datapath.h"
  58 #include "flow.h"
  59 #include "flow_table.h"
  60 #include "flow_netlink.h"
  61 #include "vlan.h"
  62 #include "vport-internal_dev.h"
  63 #include "vport-netdev.h"
  64
  65 int ovs_net_id __read_mostly;
  66
  67 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
  68                        struct genl_multicast_group *grp)
  69 {
  70         genl_notify(skb, genl_info_net(info), info->snd_portid,
  71                     grp->id, info->nlhdr, GFP_KERNEL);
  72 }
  73
  74 /**
  75  * DOC: Locking:
  76  *
  77  * All writes e.g. Writes to device state (add/remove datapath, port, set
  78  * operations on vports, etc.), Writes to other state (flow table
  79  * modifications, set miscellaneous datapath parameters, etc.) are protected
  80  * by ovs_lock.
  81  *
  82  * Reads are protected by RCU.
  83  *
  84  * There are a few special cases (mostly stats) that have their own
  85  * synchronization but they nest under all of above and don't interact with
  86  * each other.
  87  *
  88  * The RTNL lock nests inside ovs_mutex.
  89  */
  90
  91 static DEFINE_MUTEX(ovs_mutex);
  92
  93 void ovs_lock(void)
  94 {
  95         mutex_lock(&ovs_mutex);
  96 }
  97
  98 void ovs_unlock(void)
  99 {
 100         mutex_unlock(&ovs_mutex);
 101 }
 102
 103 #ifdef CONFIG_LOCKDEP
 104 int lockdep_ovsl_is_held(void)
 105 {
 106         if (debug_locks)
 107                 return lockdep_is_held(&ovs_mutex);
 108         else
 109                 return 1;
 110 }
 111 #endif
 112
 113 static struct vport *new_vport(const struct vport_parms *);
 114 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 115                              const struct dp_upcall_info *);
 116 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 117                                   const struct dp_upcall_info *);
 118
 119 /* Must be called with rcu_read_lock or ovs_mutex. */
 120 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 121 {
 122         struct datapath *dp = NULL;
 123         struct net_device *dev;
 124
 125         rcu_read_lock();
 126         dev = dev_get_by_index_rcu(net, dp_ifindex);
 127         if (dev) {
 128                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 129                 if (vport)
 130                         dp = vport->dp;
 131         }
 132         rcu_read_unlock();
 133
 134         return dp;
 135 }
 136
 137 /* Must be called with rcu_read_lock or ovs_mutex. */
 138 const char *ovs_dp_name(const struct datapath *dp)
 139 {
 140         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 141         return vport->ops->get_name(vport);
 142 }
 143
 144 static int get_dpifindex(struct datapath *dp)
 145 {
 146         struct vport *local;
 147         int ifindex;
 148
 149         rcu_read_lock();
 150
 151         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 152         if (local)
 153                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 154         else
 155                 ifindex = 0;
 156
 157         rcu_read_unlock();
 158
 159         return ifindex;
 160 }
 161
 162 static void destroy_dp_rcu(struct rcu_head *rcu)
 163 {
 164         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 165
 166         free_percpu(dp->stats_percpu);
 167         release_net(ovs_dp_get_net(dp));
 168         kfree(dp->ports);
 169         kfree(dp);
 170 }
 171
 172 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 173                                             u16 port_no)
 174 {
 175         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 176 }
 177
 178 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 179 {
 180         struct vport *vport;
 181         struct hlist_head *head;
 182
 183         head = vport_hash_bucket(dp, port_no);
 184         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 185                 if (vport->port_no == port_no)
 186                         return vport;
 187         }
 188         return NULL;
 189 }
 190
 191 /* Called with ovs_mutex. */
 192 static struct vport *new_vport(const struct vport_parms *parms)
 193 {
 194         struct vport *vport;
 195
 196         vport = ovs_vport_add(parms);
 197         if (!IS_ERR(vport)) {
 198                 struct datapath *dp = parms->dp;
 199                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 200
 201                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 202         }
 203         return vport;
 204 }
 205
 206 void ovs_dp_detach_port(struct vport *p)
 207 {
 208         ASSERT_OVSL();
 209
 210         /* First drop references to device. */
 211         hlist_del_rcu(&p->dp_hash_node);
 212
 213         /* Then destroy it. */
 214         ovs_vport_del(p);
 215 }
 216
 217 /* Must be called with rcu_read_lock. */
 218 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 219 {
 220         struct datapath *dp = p->dp;
 221         struct sw_flow *flow;
 222         struct dp_stats_percpu *stats;
 223         struct sw_flow_key key;
 224         u64 *stats_counter;
 225         u32 n_mask_hit;
 226         int error;
 227
 228         stats = this_cpu_ptr(dp->stats_percpu);
 229
 230         /* Extract flow from 'skb' into 'key'. */
 231         error = ovs_flow_extract(skb, p->port_no, &key);
 232         if (unlikely(error)) {
 233                 kfree_skb(skb);
 234                 return;
 235         }
 236
 237         /* Look up flow. */
 238         flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
 239         if (unlikely(!flow)) {
 240                 struct dp_upcall_info upcall;
 241
 242                 upcall.cmd = OVS_PACKET_CMD_MISS;
 243                 upcall.key = &key;
 244                 upcall.userdata = NULL;
 245                 upcall.portid = p->upcall_portid;
 246                 ovs_dp_upcall(dp, skb, &upcall);
 247                 consume_skb(skb);
 248                 stats_counter = &stats->n_missed;
 249                 goto out;
 250         }
 251
 252         OVS_CB(skb)->flow = flow;
 253         OVS_CB(skb)->pkt_key = &key;
 254
 255         ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
 256         ovs_execute_actions(dp, skb);
 257         stats_counter = &stats->n_hit;
 258
 259 out:
 260         /* Update datapath statistics. */
 261         u64_stats_update_begin(&stats->sync);
 262         (*stats_counter)++;
 263         stats->n_mask_hit += n_mask_hit;
 264         u64_stats_update_end(&stats->sync);
 265 }
 266
 267 static struct genl_family dp_packet_genl_family = {
 268         .id = GENL_ID_GENERATE,
 269         .hdrsize = sizeof(struct ovs_header),
 270         .name = OVS_PACKET_FAMILY,
 271         .version = OVS_PACKET_VERSION,
 272         .maxattr = OVS_PACKET_ATTR_MAX,
 273         .netnsok = true,
 274          SET_PARALLEL_OPS
 275 };
 276
 277 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 278                   const struct dp_upcall_info *upcall_info)
 279 {
 280         struct dp_stats_percpu *stats;
 281         int err;
 282
 283         if (upcall_info->portid == 0) {
 284                 err = -ENOTCONN;
 285                 goto err;
 286         }
 287
 288         if (!skb_is_gso(skb))
 289                 err = queue_userspace_packet(dp, skb, upcall_info);
 290         else
 291                 err = queue_gso_packets(dp, skb, upcall_info);
 292         if (err)
 293                 goto err;
 294
 295         return 0;
 296
 297 err:
 298         stats = this_cpu_ptr(dp->stats_percpu);
 299
 300         u64_stats_update_begin(&stats->sync);
 301         stats->n_lost++;
 302         u64_stats_update_end(&stats->sync);
 303
 304         return err;
 305 }
 306
 307 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 308                              const struct dp_upcall_info *upcall_info)
 309 {
 310         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 311         struct dp_upcall_info later_info;
 312         struct sw_flow_key later_key;
 313         struct sk_buff *segs, *nskb;
 314         int err;
 315
 316         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 317         if (IS_ERR(segs))
 318                 return PTR_ERR(segs);
 319
 320         /* Queue all of the segments. */
 321         skb = segs;
 322         do {
 323                 err = queue_userspace_packet(dp, skb, upcall_info);
 324                 if (err)
 325                         break;
 326
 327                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 328                         /* The initial flow key extracted by ovs_flow_extract()
 329                          * in this case is for a first fragment, so we need to
 330                          * properly mark later fragments.
 331                          */
 332                         later_key = *upcall_info->key;
 333                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 334
 335                         later_info = *upcall_info;
 336                         later_info.key = &later_key;
 337                         upcall_info = &later_info;
 338                 }
 339         } while ((skb = skb->next));
 340
 341         /* Free all of the segments. */
 342         skb = segs;
 343         do {
 344                 nskb = skb->next;
 345                 if (err)
 346                         kfree_skb(skb);
 347                 else
 348                         consume_skb(skb);
 349         } while ((skb = nskb));
 350         return err;
 351 }
 352
 353 static size_t key_attr_size(void)
 354 {
 355         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 356                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 357                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
 358                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 359                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 360                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
 361                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
 362                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 363                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
 364                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 365                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 366                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 367                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 368                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 369                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 370                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 371                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 372                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 373                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 374 }
 375
 376 static size_t upcall_msg_size(const struct nlattr *userdata,
 377                               unsigned int hdrlen)
 378 {
 379         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 380                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 381                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 382
 383         /* OVS_PACKET_ATTR_USERDATA */
 384         if (userdata)
 385                 size += NLA_ALIGN(userdata->nla_len);
 386
 387         return size;
 388 }
 389
 390 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 391                                   const struct dp_upcall_info *upcall_info)
 392 {
 393         struct ovs_header *upcall;
 394         struct sk_buff *nskb = NULL;
 395         struct sk_buff *user_skb; /* to be queued to userspace */
 396         struct nlattr *nla;
 397         struct genl_info info = {
 398 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,14,0)
 399                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 400 #endif
 401                 .snd_portid = upcall_info->portid,
 402         };
 403         size_t len;
 404         unsigned int hlen;
 405         int err, dp_ifindex;
 406
 407         dp_ifindex = get_dpifindex(dp);
 408         if (!dp_ifindex)
 409                 return -ENODEV;
 410
 411         if (vlan_tx_tag_present(skb)) {
 412                 nskb = skb_clone(skb, GFP_ATOMIC);
 413                 if (!nskb)
 414                         return -ENOMEM;
 415
 416                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 417                 if (!nskb)
 418                         return -ENOMEM;
 419
 420                 vlan_set_tci(nskb, 0);
 421
 422                 skb = nskb;
 423         }
 424
 425         if (nla_attr_size(skb->len) > USHRT_MAX) {
 426                 err = -EFBIG;
 427                 goto out;
 428         }
 429
 430         /* Complete checksum if needed */
 431         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 432             (err = skb_checksum_help(skb)))
 433                 goto out;
 434
 435         /* Older versions of OVS user space enforce alignment of the last
 436          * Netlink attribute to NLA_ALIGNTO which would require extensive
 437          * padding logic. Only perform zerocopy if padding is not required.
 438          */
 439         if (dp->user_features & OVS_DP_F_UNALIGNED)
 440                 hlen = skb_zerocopy_headlen(skb);
 441         else
 442                 hlen = skb->len;
 443
 444         len = upcall_msg_size(upcall_info->userdata, hlen);
 445         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 446         if (!user_skb) {
 447                 err = -ENOMEM;
 448                 goto out;
 449         }
 450
 451         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 452                              0, upcall_info->cmd);
 453         upcall->dp_ifindex = dp_ifindex;
 454
 455         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 456         ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
 457         nla_nest_end(user_skb, nla);
 458
 459         if (upcall_info->userdata)
 460                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 461                           nla_len(upcall_info->userdata),
 462                           nla_data(upcall_info->userdata));
 463
 464         /* Only reserve room for attribute header, packet data is added
 465          * in skb_zerocopy() */
 466         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 467                 err = -ENOBUFS;
 468                 goto out;
 469         }
 470         nla->nla_len = nla_attr_size(skb->len);
 471
 472         skb_zerocopy(user_skb, skb, skb->len, hlen);
 473
 474         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 475         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 476                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 477
 478                 if (plen > 0)
 479                         memset(skb_put(user_skb, plen), 0, plen);
 480         }
 481
 482         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 483
 484         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 485 out:
 486         kfree_skb(nskb);
 487         return err;
 488 }
 489
 490 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 491 {
 492         struct ovs_header *ovs_header = info->userhdr;
 493         struct nlattr **a = info->attrs;
 494         struct sw_flow_actions *acts;
 495         struct sk_buff *packet;
 496         struct sw_flow *flow;
 497         struct datapath *dp;
 498         struct ethhdr *eth;
 499         int len;
 500         int err;
 501
 502         err = -EINVAL;
 503         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 504             !a[OVS_PACKET_ATTR_ACTIONS])
 505                 goto err;
 506
 507         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 508         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 509         err = -ENOMEM;
 510         if (!packet)
 511                 goto err;
 512         skb_reserve(packet, NET_IP_ALIGN);
 513
 514         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 515
 516         skb_reset_mac_header(packet);
 517         eth = eth_hdr(packet);
 518
 519         /* Normally, setting the skb 'protocol' field would be handled by a
 520          * call to eth_type_trans(), but it assumes there's a sending
 521          * device, which we may not have. */
 522         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 523                 packet->protocol = eth->h_proto;
 524         else
 525                 packet->protocol = htons(ETH_P_802_2);
 526
 527         /* Build an sw_flow for sending this packet. */
 528         flow = ovs_flow_alloc(false);
 529         err = PTR_ERR(flow);
 530         if (IS_ERR(flow))
 531                 goto err_kfree_skb;
 532
 533         err = ovs_flow_extract(packet, -1, &flow->key);
 534         if (err)
 535                 goto err_flow_free;
 536
 537         err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
 538         if (err)
 539                 goto err_flow_free;
 540         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 541         err = PTR_ERR(acts);
 542         if (IS_ERR(acts))
 543                 goto err_flow_free;
 544
 545         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 546                                    &flow->key, 0, &acts);
 547         rcu_assign_pointer(flow->sf_acts, acts);
 548         if (err)
 549                 goto err_flow_free;
 550
 551         OVS_CB(packet)->flow = flow;
 552         OVS_CB(packet)->pkt_key = &flow->key;
 553         packet->priority = flow->key.phy.priority;
 554         packet->mark = flow->key.phy.skb_mark;
 555
 556         rcu_read_lock();
 557         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 558         err = -ENODEV;
 559         if (!dp)
 560                 goto err_unlock;
 561
 562         local_bh_disable();
 563         err = ovs_execute_actions(dp, packet);
 564         local_bh_enable();
 565         rcu_read_unlock();
 566
 567         ovs_flow_free(flow, false);
 568         return err;
 569
 570 err_unlock:
 571         rcu_read_unlock();
 572 err_flow_free:
 573         ovs_flow_free(flow, false);
 574 err_kfree_skb:
 575         kfree_skb(packet);
 576 err:
 577         return err;
 578 }
 579
 580 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 581         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 582         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 583         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 584 };
 585
 586 static struct genl_ops dp_packet_genl_ops[] = {
 587         { .cmd = OVS_PACKET_CMD_EXECUTE,
 588           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 589           .policy = packet_policy,
 590           .doit = ovs_packet_cmd_execute
 591         }
 592 };
 593
 594 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
 595                          struct ovs_dp_megaflow_stats *mega_stats)
 596 {
 597         int i;
 598
 599         memset(mega_stats, 0, sizeof(*mega_stats));
 600
 601         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 602         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 603
 604         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 605
 606         for_each_possible_cpu(i) {
 607                 const struct dp_stats_percpu *percpu_stats;
 608                 struct dp_stats_percpu local_stats;
 609                 unsigned int start;
 610
 611                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 612
 613                 do {
 614                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 615                         local_stats = *percpu_stats;
 616                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 617
 618                 stats->n_hit += local_stats.n_hit;
 619                 stats->n_missed += local_stats.n_missed;
 620                 stats->n_lost += local_stats.n_lost;
 621                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 622         }
 623 }
 624
 625 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 626         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 627         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 628         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 629 };
 630
 631 static struct genl_family dp_flow_genl_family = {
 632         .id = GENL_ID_GENERATE,
 633         .hdrsize = sizeof(struct ovs_header),
 634         .name = OVS_FLOW_FAMILY,
 635         .version = OVS_FLOW_VERSION,
 636         .maxattr = OVS_FLOW_ATTR_MAX,
 637         .netnsok = true,
 638          SET_PARALLEL_OPS
 639 };
 640
 641 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 642         .name = OVS_FLOW_MCGROUP
 643 };
 644
 645 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 646 {
 647         return NLMSG_ALIGN(sizeof(struct ovs_header))
 648                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 649                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 650                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 651                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 652                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 653                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 654 }
 655
 656 /* Called with ovs_mutex. */
 657 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 658                                   struct sk_buff *skb, u32 portid,
 659                                   u32 seq, u32 flags, u8 cmd)
 660 {
 661         const int skb_orig_len = skb->len;
 662         struct nlattr *start;
 663         struct ovs_flow_stats stats;
 664         __be16 tcp_flags;
 665         unsigned long used;
 666         struct ovs_header *ovs_header;
 667         struct nlattr *nla;
 668         int err;
 669
 670         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 671         if (!ovs_header)
 672                 return -EMSGSIZE;
 673
 674         ovs_header->dp_ifindex = get_dpifindex(dp);
 675
 676         /* Fill flow key. */
 677         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 678         if (!nla)
 679                 goto nla_put_failure;
 680
 681         err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
 682         if (err)
 683                 goto error;
 684         nla_nest_end(skb, nla);
 685
 686         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 687         if (!nla)
 688                 goto nla_put_failure;
 689
 690         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 691         if (err)
 692                 goto error;
 693
 694         nla_nest_end(skb, nla);
 695
 696         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 697         if (used &&
 698             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 699                 goto nla_put_failure;
 700
 701         if (stats.n_packets &&
 702             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 703                 goto nla_put_failure;
 704
 705         if ((u8)ntohs(tcp_flags) &&
 706              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 707                 goto nla_put_failure;
 708
 709         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 710          * this is the first flow to be dumped into 'skb'.  This is unusual for
 711          * Netlink but individual action lists can be longer than
 712          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 713          * The userspace caller can always fetch the actions separately if it
 714          * really wants them.  (Most userspace callers in fact don't care.)
 715          *
 716          * This can only fail for dump operations because the skb is always
 717          * properly sized for single flows.
 718          */
 719         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 720         if (start) {
 721                 const struct sw_flow_actions *sf_acts;
 722
 723                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 724
 725                 err = ovs_nla_put_actions(sf_acts->actions,
 726                                           sf_acts->actions_len, skb);
 727                 if (!err)
 728                         nla_nest_end(skb, start);
 729                 else {
 730                         if (skb_orig_len)
 731                                 goto error;
 732
 733                         nla_nest_cancel(skb, start);
 734                 }
 735         } else if (skb_orig_len)
 736                 goto nla_put_failure;
 737
 738         return genlmsg_end(skb, ovs_header);
 739
 740 nla_put_failure:
 741         err = -EMSGSIZE;
 742 error:
 743         genlmsg_cancel(skb, ovs_header);
 744         return err;
 745 }
 746
 747 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow,
 748                                                struct genl_info *info)
 749 {
 750         size_t len;
 751
 752         len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts));
 753
 754         return genlmsg_new_unicast(len, info, GFP_KERNEL);
 755 }
 756
 757 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 758                                                struct datapath *dp,
 759                                                struct genl_info *info,
 760                                                u8 cmd)
 761 {
 762         struct sk_buff *skb;
 763         int retval;
 764
 765         skb = ovs_flow_cmd_alloc_info(flow, info);
 766         if (!skb)
 767                 return ERR_PTR(-ENOMEM);
 768
 769         retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid,
 770                                         info->snd_seq, 0, cmd);
 771         BUG_ON(retval < 0);
 772         return skb;
 773 }
 774
 775 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 776 {
 777         struct nlattr **a = info->attrs;
 778         struct ovs_header *ovs_header = info->userhdr;
 779         struct sw_flow_key key, masked_key;
 780         struct sw_flow *flow = NULL;
 781         struct sw_flow_mask mask;
 782         struct sk_buff *reply;
 783         struct datapath *dp;
 784         struct sw_flow_actions *acts = NULL;
 785         struct sw_flow_match match;
 786         bool exact_5tuple;
 787         int error;
 788
 789         /* Extract key. */
 790         error = -EINVAL;
 791         if (!a[OVS_FLOW_ATTR_KEY])
 792                 goto error;
 793
 794         ovs_match_init(&match, &key, &mask);
 795         error = ovs_nla_get_match(&match, &exact_5tuple,
 796                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 797         if (error)
 798                 goto error;
 799
 800         /* Validate actions. */
 801         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 802                 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 803                 error = PTR_ERR(acts);
 804                 if (IS_ERR(acts))
 805                         goto error;
 806
 807                 ovs_flow_mask_key(&masked_key, &key, &mask);
 808                 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
 809                                              &masked_key, 0, &acts);
 810                 if (error) {
 811                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 812                         goto err_kfree;
 813                 }
 814         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 815                 error = -EINVAL;
 816                 goto error;
 817         }
 818
 819         ovs_lock();
 820         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 821         error = -ENODEV;
 822         if (!dp)
 823                 goto err_unlock_ovs;
 824
 825         /* Check if this is a duplicate flow */
 826         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 827         if (!flow) {
 828                 /* Bail out if we're not allowed to create a new flow. */
 829                 error = -ENOENT;
 830                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
 831                         goto err_unlock_ovs;
 832
 833                 /* Allocate flow. */
 834                 flow = ovs_flow_alloc(!exact_5tuple);
 835                 if (IS_ERR(flow)) {
 836                         error = PTR_ERR(flow);
 837                         goto err_unlock_ovs;
 838                 }
 839
 840                 flow->key = masked_key;
 841                 flow->unmasked_key = key;
 842                 rcu_assign_pointer(flow->sf_acts, acts);
 843
 844                 /* Put flow in bucket. */
 845                 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
 846                 if (error) {
 847                         acts = NULL;
 848                         goto err_flow_free;
 849                 }
 850
 851                 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
 852         } else {
 853                 /* We found a matching flow. */
 854                 struct sw_flow_actions *old_acts;
 855
 856                 /* Bail out if we're not allowed to modify an existing flow.
 857                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 858                  * because Generic Netlink treats the latter as a dump
 859                  * request.  We also accept NLM_F_EXCL in case that bug ever
 860                  * gets fixed.
 861                  */
 862                 error = -EEXIST;
 863                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
 864                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
 865                         goto err_unlock_ovs;
 866
 867                 /* The unmasked key has to be the same for flow updates. */
 868                 if (!ovs_flow_cmp_unmasked_key(flow, &match))
 869                         goto err_unlock_ovs;
 870
 871                 /* Update actions. */
 872                 old_acts = ovsl_dereference(flow->sf_acts);
 873                 rcu_assign_pointer(flow->sf_acts, acts);
 874                 ovs_nla_free_flow_actions(old_acts);
 875
 876                 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
 877
 878                 /* Clear stats. */
 879                 if (a[OVS_FLOW_ATTR_CLEAR])
 880                         ovs_flow_stats_clear(flow);
 881         }
 882         ovs_unlock();
 883
 884         if (!IS_ERR(reply))
 885                 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
 886         else
 887                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
 888                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
 889         return 0;
 890
 891 err_flow_free:
 892         ovs_flow_free(flow, false);
 893 err_unlock_ovs:
 894         ovs_unlock();
 895 err_kfree:
 896         kfree(acts);
 897 error:
 898         return error;
 899 }
 900
 901 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
 902 {
 903         struct nlattr **a = info->attrs;
 904         struct ovs_header *ovs_header = info->userhdr;
 905         struct sw_flow_key key;
 906         struct sk_buff *reply;
 907         struct sw_flow *flow;
 908         struct datapath *dp;
 909         struct sw_flow_match match;
 910         int err;
 911
 912         if (!a[OVS_FLOW_ATTR_KEY]) {
 913                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
 914                 return -EINVAL;
 915         }
 916
 917         ovs_match_init(&match, &key, NULL);
 918         err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
 919         if (err)
 920                 return err;
 921
 922         ovs_lock();
 923         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 924         if (!dp) {
 925                 err = -ENODEV;
 926                 goto unlock;
 927         }
 928
 929         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 930         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
 931                 err = -ENOENT;
 932                 goto unlock;
 933         }
 934
 935         reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
 936         if (IS_ERR(reply)) {
 937                 err = PTR_ERR(reply);
 938                 goto unlock;
 939         }
 940
 941         ovs_unlock();
 942         return genlmsg_reply(reply, info);
 943 unlock:
 944         ovs_unlock();
 945         return err;
 946 }
 947
 948 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
 949 {
 950         struct nlattr **a = info->attrs;
 951         struct ovs_header *ovs_header = info->userhdr;
 952         struct sw_flow_key key;
 953         struct sk_buff *reply;
 954         struct sw_flow *flow;
 955         struct datapath *dp;
 956         struct sw_flow_match match;
 957         int err;
 958
 959         ovs_lock();
 960         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 961         if (!dp) {
 962                 err = -ENODEV;
 963                 goto unlock;
 964         }
 965
 966         if (!a[OVS_FLOW_ATTR_KEY]) {
 967                 err = ovs_flow_tbl_flush(&dp->table);
 968                 goto unlock;
 969         }
 970
 971         ovs_match_init(&match, &key, NULL);
 972         err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
 973         if (err)
 974                 goto unlock;
 975
 976         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 977         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
 978                 err = -ENOENT;
 979                 goto unlock;
 980         }
 981
 982         reply = ovs_flow_cmd_alloc_info(flow, info);
 983         if (!reply) {
 984                 err = -ENOMEM;
 985                 goto unlock;
 986         }
 987
 988         ovs_flow_tbl_remove(&dp->table, flow);
 989
 990         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
 991                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
 992         BUG_ON(err < 0);
 993
 994         ovs_flow_free(flow, true);
 995         ovs_unlock();
 996
 997         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
 998         return 0;
 999 unlock:
1000         ovs_unlock();
1001         return err;
1002 }
1003
1004 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1005 {
1006         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1007         struct table_instance *ti;
1008         struct datapath *dp;
1009
1010         rcu_read_lock();
1011         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1012         if (!dp) {
1013                 rcu_read_unlock();
1014                 return -ENODEV;
1015         }
1016
1017         ti = rcu_dereference(dp->table.ti);
1018         for (;;) {
1019                 struct sw_flow *flow;
1020                 u32 bucket, obj;
1021
1022                 bucket = cb->args[0];
1023                 obj = cb->args[1];
1024                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1025                 if (!flow)
1026                         break;
1027
1028                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1029                                            NETLINK_CB(cb->skb).portid,
1030                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1031                                            OVS_FLOW_CMD_NEW) < 0)
1032                         break;
1033
1034                 cb->args[0] = bucket;
1035                 cb->args[1] = obj;
1036         }
1037         rcu_read_unlock();
1038         return skb->len;
1039 }
1040
1041 static struct genl_ops dp_flow_genl_ops[] = {
1042         { .cmd = OVS_FLOW_CMD_NEW,
1043           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1044           .policy = flow_policy,
1045           .doit = ovs_flow_cmd_new_or_set
1046         },
1047         { .cmd = OVS_FLOW_CMD_DEL,
1048           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1049           .policy = flow_policy,
1050           .doit = ovs_flow_cmd_del
1051         },
1052         { .cmd = OVS_FLOW_CMD_GET,
1053           .flags = 0,               /* OK for unprivileged users. */
1054           .policy = flow_policy,
1055           .doit = ovs_flow_cmd_get,
1056           .dumpit = ovs_flow_cmd_dump
1057         },
1058         { .cmd = OVS_FLOW_CMD_SET,
1059           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1060           .policy = flow_policy,
1061           .doit = ovs_flow_cmd_new_or_set,
1062         },
1063 };
1064
1065 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1066         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1067         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1068         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1069 };
1070
1071 static struct genl_family dp_datapath_genl_family = {
1072         .id = GENL_ID_GENERATE,
1073         .hdrsize = sizeof(struct ovs_header),
1074         .name = OVS_DATAPATH_FAMILY,
1075         .version = OVS_DATAPATH_VERSION,
1076         .maxattr = OVS_DP_ATTR_MAX,
1077         .netnsok = true,
1078          SET_PARALLEL_OPS
1079 };
1080
1081 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1082         .name = OVS_DATAPATH_MCGROUP
1083 };
1084
1085 static size_t ovs_dp_cmd_msg_size(void)
1086 {
1087         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1088
1089         msgsize += nla_total_size(IFNAMSIZ);
1090         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1091         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1092         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1093
1094         return msgsize;
1095 }
1096
1097 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1098                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1099 {
1100         struct ovs_header *ovs_header;
1101         struct ovs_dp_stats dp_stats;
1102         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1103         int err;
1104
1105         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1106                                    flags, cmd);
1107         if (!ovs_header)
1108                 goto error;
1109
1110         ovs_header->dp_ifindex = get_dpifindex(dp);
1111
1112         rcu_read_lock();
1113         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1114         rcu_read_unlock();
1115         if (err)
1116                 goto nla_put_failure;
1117
1118         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1119         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1120                         &dp_stats))
1121                 goto nla_put_failure;
1122
1123         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1124                         sizeof(struct ovs_dp_megaflow_stats),
1125                         &dp_megaflow_stats))
1126                 goto nla_put_failure;
1127
1128         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1129                 goto nla_put_failure;
1130
1131         return genlmsg_end(skb, ovs_header);
1132
1133 nla_put_failure:
1134         genlmsg_cancel(skb, ovs_header);
1135 error:
1136         return -EMSGSIZE;
1137 }
1138
1139 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp,
1140                                              struct genl_info *info, u8 cmd)
1141 {
1142         struct sk_buff *skb;
1143         int retval;
1144
1145         skb = genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1146         if (!skb)
1147                 return ERR_PTR(-ENOMEM);
1148
1149         retval = ovs_dp_cmd_fill_info(dp, skb, info->snd_portid, info->snd_seq, 0, cmd);
1150         if (retval < 0) {
1151                 kfree_skb(skb);
1152                 return ERR_PTR(retval);
1153         }
1154         return skb;
1155 }
1156
1157 /* Called with ovs_mutex. */
1158 static struct datapath *lookup_datapath(struct net *net,
1159                                         struct ovs_header *ovs_header,
1160                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1161 {
1162         struct datapath *dp;
1163
1164         if (!a[OVS_DP_ATTR_NAME])
1165                 dp = get_dp(net, ovs_header->dp_ifindex);
1166         else {
1167                 struct vport *vport;
1168
1169                 rcu_read_lock();
1170                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1171                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1172                 rcu_read_unlock();
1173         }
1174         return dp ? dp : ERR_PTR(-ENODEV);
1175 }
1176
1177 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1178 {
1179         struct datapath *dp;
1180
1181         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1182         if (!dp)
1183                 return;
1184
1185         WARN(dp->user_features, "Dropping previously announced user features\n");
1186         dp->user_features = 0;
1187 }
1188
1189 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1190 {
1191         if (a[OVS_DP_ATTR_USER_FEATURES])
1192                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1193 }
1194
1195 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1196 {
1197         struct nlattr **a = info->attrs;
1198         struct vport_parms parms;
1199         struct sk_buff *reply;
1200         struct datapath *dp;
1201         struct vport *vport;
1202         struct ovs_net *ovs_net;
1203         int err, i;
1204
1205         err = -EINVAL;
1206         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1207                 goto err;
1208
1209         ovs_lock();
1210
1211         err = -ENOMEM;
1212         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1213         if (dp == NULL)
1214                 goto err_unlock_ovs;
1215
1216         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1217
1218         /* Allocate table. */
1219         err = ovs_flow_tbl_init(&dp->table);
1220         if (err)
1221                 goto err_free_dp;
1222
1223         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1224         if (!dp->stats_percpu) {
1225                 err = -ENOMEM;
1226                 goto err_destroy_table;
1227         }
1228
1229         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1230                             GFP_KERNEL);
1231         if (!dp->ports) {
1232                 err = -ENOMEM;
1233                 goto err_destroy_percpu;
1234         }
1235
1236         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1237                 INIT_HLIST_HEAD(&dp->ports[i]);
1238
1239         /* Set up our datapath device. */
1240         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1241         parms.type = OVS_VPORT_TYPE_INTERNAL;
1242         parms.options = NULL;
1243         parms.dp = dp;
1244         parms.port_no = OVSP_LOCAL;
1245         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1246
1247         ovs_dp_change(dp, a);
1248
1249         vport = new_vport(&parms);
1250         if (IS_ERR(vport)) {
1251                 err = PTR_ERR(vport);
1252                 if (err == -EBUSY)
1253                         err = -EEXIST;
1254
1255                 if (err == -EEXIST) {
1256                         /* An outdated user space instance that does not understand
1257                          * the concept of user_features has attempted to create a new
1258                          * datapath and is likely to reuse it. Drop all user features.
1259                          */
1260                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1261                                 ovs_dp_reset_user_features(skb, info);
1262                 }
1263
1264                 goto err_destroy_ports_array;
1265         }
1266
1267         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1268         err = PTR_ERR(reply);
1269         if (IS_ERR(reply))
1270                 goto err_destroy_local_port;
1271
1272         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1273         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1274
1275         ovs_unlock();
1276
1277         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1278         return 0;
1279
1280 err_destroy_local_port:
1281         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1282 err_destroy_ports_array:
1283         kfree(dp->ports);
1284 err_destroy_percpu:
1285         free_percpu(dp->stats_percpu);
1286 err_destroy_table:
1287         ovs_flow_tbl_destroy(&dp->table, false);
1288 err_free_dp:
1289         release_net(ovs_dp_get_net(dp));
1290         kfree(dp);
1291 err_unlock_ovs:
1292         ovs_unlock();
1293 err:
1294         return err;
1295 }
1296
1297 /* Called with ovs_mutex. */
1298 static void __dp_destroy(struct datapath *dp)
1299 {
1300         int i;
1301
1302         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1303                 struct vport *vport;
1304                 struct hlist_node *n;
1305
1306                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1307                         if (vport->port_no != OVSP_LOCAL)
1308                                 ovs_dp_detach_port(vport);
1309         }
1310
1311         list_del_rcu(&dp->list_node);
1312
1313         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1314          * all ports in datapath are destroyed first before freeing datapath.
1315          */
1316         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1317
1318         /* RCU destroy the flow table */
1319         ovs_flow_tbl_destroy(&dp->table, true);
1320
1321         call_rcu(&dp->rcu, destroy_dp_rcu);
1322 }
1323
1324 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1325 {
1326         struct sk_buff *reply;
1327         struct datapath *dp;
1328         int err;
1329
1330         ovs_lock();
1331         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1332         err = PTR_ERR(dp);
1333         if (IS_ERR(dp))
1334                 goto unlock;
1335
1336         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_DEL);
1337         err = PTR_ERR(reply);
1338         if (IS_ERR(reply))
1339                 goto unlock;
1340
1341         __dp_destroy(dp);
1342         ovs_unlock();
1343
1344         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1345
1346         return 0;
1347 unlock:
1348         ovs_unlock();
1349         return err;
1350 }
1351
1352 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1353 {
1354         struct sk_buff *reply;
1355         struct datapath *dp;
1356         int err;
1357
1358         ovs_lock();
1359         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1360         err = PTR_ERR(dp);
1361         if (IS_ERR(dp))
1362                 goto unlock;
1363
1364         ovs_dp_change(dp, info->attrs);
1365
1366         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1367         if (IS_ERR(reply)) {
1368                 err = PTR_ERR(reply);
1369                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1370                                 ovs_dp_datapath_multicast_group.id, err);
1371                 err = 0;
1372                 goto unlock;
1373         }
1374
1375         ovs_unlock();
1376         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1377
1378         return 0;
1379 unlock:
1380         ovs_unlock();
1381         return err;
1382 }
1383
1384 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1385 {
1386         struct sk_buff *reply;
1387         struct datapath *dp;
1388         int err;
1389
1390         ovs_lock();
1391         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1392         if (IS_ERR(dp)) {
1393                 err = PTR_ERR(dp);
1394                 goto unlock;
1395         }
1396
1397         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1398         if (IS_ERR(reply)) {
1399                 err = PTR_ERR(reply);
1400                 goto unlock;
1401         }
1402
1403         ovs_unlock();
1404         return genlmsg_reply(reply, info);
1405
1406 unlock:
1407         ovs_unlock();
1408         return err;
1409 }
1410
1411 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1412 {
1413         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1414         struct datapath *dp;
1415         int skip = cb->args[0];
1416         int i = 0;
1417
1418         rcu_read_lock();
1419         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1420                 if (i >= skip &&
1421                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1422                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1423                                          OVS_DP_CMD_NEW) < 0)
1424                         break;
1425                 i++;
1426         }
1427         rcu_read_unlock();
1428
1429         cb->args[0] = i;
1430
1431         return skb->len;
1432 }
1433
1434 static struct genl_ops dp_datapath_genl_ops[] = {
1435         { .cmd = OVS_DP_CMD_NEW,
1436           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1437           .policy = datapath_policy,
1438           .doit = ovs_dp_cmd_new
1439         },
1440         { .cmd = OVS_DP_CMD_DEL,
1441           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1442           .policy = datapath_policy,
1443           .doit = ovs_dp_cmd_del
1444         },
1445         { .cmd = OVS_DP_CMD_GET,
1446           .flags = 0,               /* OK for unprivileged users. */
1447           .policy = datapath_policy,
1448           .doit = ovs_dp_cmd_get,
1449           .dumpit = ovs_dp_cmd_dump
1450         },
1451         { .cmd = OVS_DP_CMD_SET,
1452           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1453           .policy = datapath_policy,
1454           .doit = ovs_dp_cmd_set,
1455         },
1456 };
1457
1458 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1459         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1460         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1461         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1462         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1463         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1464         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1465 };
1466
1467 static struct genl_family dp_vport_genl_family = {
1468         .id = GENL_ID_GENERATE,
1469         .hdrsize = sizeof(struct ovs_header),
1470         .name = OVS_VPORT_FAMILY,
1471         .version = OVS_VPORT_VERSION,
1472         .maxattr = OVS_VPORT_ATTR_MAX,
1473         .netnsok = true,
1474          SET_PARALLEL_OPS
1475 };
1476
1477 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1478         .name = OVS_VPORT_MCGROUP
1479 };
1480
1481 /* Called with ovs_mutex or RCU read lock. */
1482 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1483                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1484 {
1485         struct ovs_header *ovs_header;
1486         struct ovs_vport_stats vport_stats;
1487         int err;
1488
1489         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1490                                  flags, cmd);
1491         if (!ovs_header)
1492                 return -EMSGSIZE;
1493
1494         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1495
1496         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1497             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1498             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1499             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1500                 goto nla_put_failure;
1501
1502         ovs_vport_get_stats(vport, &vport_stats);
1503         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1504                     &vport_stats))
1505                 goto nla_put_failure;
1506
1507         err = ovs_vport_get_options(vport, skb);
1508         if (err == -EMSGSIZE)
1509                 goto error;
1510
1511         return genlmsg_end(skb, ovs_header);
1512
1513 nla_put_failure:
1514         err = -EMSGSIZE;
1515 error:
1516         genlmsg_cancel(skb, ovs_header);
1517         return err;
1518 }
1519
1520 /* Called with ovs_mutex or RCU read lock. */
1521 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1522                                          u32 seq, u8 cmd)
1523 {
1524         struct sk_buff *skb;
1525         int retval;
1526
1527         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1528         if (!skb)
1529                 return ERR_PTR(-ENOMEM);
1530
1531         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1532         BUG_ON(retval < 0);
1533
1534         return skb;
1535 }
1536
1537 /* Called with ovs_mutex or RCU read lock. */
1538 static struct vport *lookup_vport(struct net *net,
1539                                   struct ovs_header *ovs_header,
1540                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1541 {
1542         struct datapath *dp;
1543         struct vport *vport;
1544
1545         if (a[OVS_VPORT_ATTR_NAME]) {
1546                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1547                 if (!vport)
1548                         return ERR_PTR(-ENODEV);
1549                 if (ovs_header->dp_ifindex &&
1550                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1551                         return ERR_PTR(-ENODEV);
1552                 return vport;
1553         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1554                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1555
1556                 if (port_no >= DP_MAX_PORTS)
1557                         return ERR_PTR(-EFBIG);
1558
1559                 dp = get_dp(net, ovs_header->dp_ifindex);
1560                 if (!dp)
1561                         return ERR_PTR(-ENODEV);
1562
1563                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1564                 if (!vport)
1565                         return ERR_PTR(-ENODEV);
1566                 return vport;
1567         } else
1568                 return ERR_PTR(-EINVAL);
1569 }
1570
1571 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1572 {
1573         struct nlattr **a = info->attrs;
1574         struct ovs_header *ovs_header = info->userhdr;
1575         struct vport_parms parms;
1576         struct sk_buff *reply;
1577         struct vport *vport;
1578         struct datapath *dp;
1579         u32 port_no;
1580         int err;
1581
1582         err = -EINVAL;
1583         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1584             !a[OVS_VPORT_ATTR_UPCALL_PID])
1585                 goto exit;
1586
1587         ovs_lock();
1588         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1589         err = -ENODEV;
1590         if (!dp)
1591                 goto exit_unlock;
1592
1593         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1594                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1595
1596                 err = -EFBIG;
1597                 if (port_no >= DP_MAX_PORTS)
1598                         goto exit_unlock;
1599
1600                 vport = ovs_vport_ovsl(dp, port_no);
1601                 err = -EBUSY;
1602                 if (vport)
1603                         goto exit_unlock;
1604         } else {
1605                 for (port_no = 1; ; port_no++) {
1606                         if (port_no >= DP_MAX_PORTS) {
1607                                 err = -EFBIG;
1608                                 goto exit_unlock;
1609                         }
1610                         vport = ovs_vport_ovsl(dp, port_no);
1611                         if (!vport)
1612                                 break;
1613                 }
1614         }
1615
1616         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1617         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1618         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1619         parms.dp = dp;
1620         parms.port_no = port_no;
1621         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1622
1623         vport = new_vport(&parms);
1624         err = PTR_ERR(vport);
1625         if (IS_ERR(vport))
1626                 goto exit_unlock;
1627
1628         err = 0;
1629         if (a[OVS_VPORT_ATTR_STATS])
1630                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1631
1632         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1633                                          OVS_VPORT_CMD_NEW);
1634         if (IS_ERR(reply)) {
1635                 err = PTR_ERR(reply);
1636                 ovs_dp_detach_port(vport);
1637                 goto exit_unlock;
1638         }
1639
1640         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1641
1642 exit_unlock:
1643         ovs_unlock();
1644 exit:
1645         return err;
1646 }
1647
1648 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1649 {
1650         struct nlattr **a = info->attrs;
1651         struct sk_buff *reply;
1652         struct vport *vport;
1653         int err;
1654
1655         ovs_lock();
1656         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1657         err = PTR_ERR(vport);
1658         if (IS_ERR(vport))
1659                 goto exit_unlock;
1660
1661         if (a[OVS_VPORT_ATTR_TYPE] &&
1662             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1663                 err = -EINVAL;
1664                 goto exit_unlock;
1665         }
1666
1667         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1668         if (!reply) {
1669                 err = -ENOMEM;
1670                 goto exit_unlock;
1671         }
1672
1673         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1674                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1675                 if (err)
1676                         goto exit_free;
1677         }
1678
1679         if (a[OVS_VPORT_ATTR_STATS])
1680                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1681
1682         if (a[OVS_VPORT_ATTR_UPCALL_PID])
1683                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1684
1685         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1686                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1687         BUG_ON(err < 0);
1688
1689         ovs_unlock();
1690         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1691         return 0;
1692
1693 exit_free:
1694         kfree_skb(reply);
1695 exit_unlock:
1696         ovs_unlock();
1697         return err;
1698 }
1699
1700 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1701 {
1702         struct nlattr **a = info->attrs;
1703         struct sk_buff *reply;
1704         struct vport *vport;
1705         int err;
1706
1707         ovs_lock();
1708         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1709         err = PTR_ERR(vport);
1710         if (IS_ERR(vport))
1711                 goto exit_unlock;
1712
1713         if (vport->port_no == OVSP_LOCAL) {
1714                 err = -EINVAL;
1715                 goto exit_unlock;
1716         }
1717
1718         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1719                                          info->snd_seq, OVS_VPORT_CMD_DEL);
1720         err = PTR_ERR(reply);
1721         if (IS_ERR(reply))
1722                 goto exit_unlock;
1723
1724         err = 0;
1725         ovs_dp_detach_port(vport);
1726
1727         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1728
1729 exit_unlock:
1730         ovs_unlock();
1731         return err;
1732 }
1733
1734 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1735 {
1736         struct nlattr **a = info->attrs;
1737         struct ovs_header *ovs_header = info->userhdr;
1738         struct sk_buff *reply;
1739         struct vport *vport;
1740         int err;
1741
1742         rcu_read_lock();
1743         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1744         err = PTR_ERR(vport);
1745         if (IS_ERR(vport))
1746                 goto exit_unlock;
1747
1748         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1749                                          info->snd_seq, OVS_VPORT_CMD_NEW);
1750         err = PTR_ERR(reply);
1751         if (IS_ERR(reply))
1752                 goto exit_unlock;
1753
1754         rcu_read_unlock();
1755
1756         return genlmsg_reply(reply, info);
1757
1758 exit_unlock:
1759         rcu_read_unlock();
1760         return err;
1761 }
1762
1763 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1764 {
1765         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1766         struct datapath *dp;
1767         int bucket = cb->args[0], skip = cb->args[1];
1768         int i, j = 0;
1769
1770         rcu_read_lock();
1771         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1772         if (!dp) {
1773                 rcu_read_unlock();
1774                 return -ENODEV;
1775         }
1776         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1777                 struct vport *vport;
1778
1779                 j = 0;
1780                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1781                         if (j >= skip &&
1782                             ovs_vport_cmd_fill_info(vport, skb,
1783                                                     NETLINK_CB(cb->skb).portid,
1784                                                     cb->nlh->nlmsg_seq,
1785                                                     NLM_F_MULTI,
1786                                                     OVS_VPORT_CMD_NEW) < 0)
1787                                 goto out;
1788
1789                         j++;
1790                 }
1791                 skip = 0;
1792         }
1793 out:
1794         rcu_read_unlock();
1795
1796         cb->args[0] = i;
1797         cb->args[1] = j;
1798
1799         return skb->len;
1800 }
1801
1802 static struct genl_ops dp_vport_genl_ops[] = {
1803         { .cmd = OVS_VPORT_CMD_NEW,
1804           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1805           .policy = vport_policy,
1806           .doit = ovs_vport_cmd_new
1807         },
1808         { .cmd = OVS_VPORT_CMD_DEL,
1809           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1810           .policy = vport_policy,
1811           .doit = ovs_vport_cmd_del
1812         },
1813         { .cmd = OVS_VPORT_CMD_GET,
1814           .flags = 0,               /* OK for unprivileged users. */
1815           .policy = vport_policy,
1816           .doit = ovs_vport_cmd_get,
1817           .dumpit = ovs_vport_cmd_dump
1818         },
1819         { .cmd = OVS_VPORT_CMD_SET,
1820           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1821           .policy = vport_policy,
1822           .doit = ovs_vport_cmd_set,
1823         },
1824 };
1825
1826 struct genl_family_and_ops {
1827         struct genl_family *family;
1828         struct genl_ops *ops;
1829         int n_ops;
1830         struct genl_multicast_group *group;
1831 };
1832
1833 static const struct genl_family_and_ops dp_genl_families[] = {
1834         { &dp_datapath_genl_family,
1835           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1836           &ovs_dp_datapath_multicast_group },
1837         { &dp_vport_genl_family,
1838           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1839           &ovs_dp_vport_multicast_group },
1840         { &dp_flow_genl_family,
1841           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1842           &ovs_dp_flow_multicast_group },
1843         { &dp_packet_genl_family,
1844           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1845           NULL },
1846 };
1847
1848 static void dp_unregister_genl(int n_families)
1849 {
1850         int i;
1851
1852         for (i = 0; i < n_families; i++)
1853                 genl_unregister_family(dp_genl_families[i].family);
1854 }
1855
1856 static int dp_register_genl(void)
1857 {
1858         int n_registered;
1859         int err;
1860         int i;
1861
1862         n_registered = 0;
1863         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1864                 const struct genl_family_and_ops *f = &dp_genl_families[i];
1865
1866                 err = genl_register_family_with_ops(f->family, f->ops,
1867                                                     f->n_ops);
1868                 if (err)
1869                         goto error;
1870                 n_registered++;
1871
1872                 if (f->group) {
1873                         err = genl_register_mc_group(f->family, f->group);
1874                         if (err)
1875                                 goto error;
1876                 }
1877         }
1878
1879         return 0;
1880
1881 error:
1882         dp_unregister_genl(n_registered);
1883         return err;
1884 }
1885
1886 static int __net_init ovs_init_net(struct net *net)
1887 {
1888         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1889
1890         INIT_LIST_HEAD(&ovs_net->dps);
1891         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1892         return 0;
1893 }
1894
1895 static void __net_exit ovs_exit_net(struct net *net)
1896 {
1897         struct datapath *dp, *dp_next;
1898         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1899
1900         ovs_lock();
1901         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1902                 __dp_destroy(dp);
1903         ovs_unlock();
1904
1905         cancel_work_sync(&ovs_net->dp_notify_work);
1906 }
1907
1908 static struct pernet_operations ovs_net_ops = {
1909         .init = ovs_init_net,
1910         .exit = ovs_exit_net,
1911         .id   = &ovs_net_id,
1912         .size = sizeof(struct ovs_net),
1913 };
1914
1915 DEFINE_COMPAT_PNET_REG_FUNC(device);
1916
1917 static int __init dp_init(void)
1918 {
1919         int err;
1920
1921         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1922
1923         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
1924                 VERSION);
1925
1926         err = ovs_flow_init();
1927         if (err)
1928                 goto error;
1929
1930         err = ovs_vport_init();
1931         if (err)
1932                 goto error_flow_exit;
1933
1934         err = register_pernet_device(&ovs_net_ops);
1935         if (err)
1936                 goto error_vport_exit;
1937
1938         err = register_netdevice_notifier(&ovs_dp_device_notifier);
1939         if (err)
1940                 goto error_netns_exit;
1941
1942         err = dp_register_genl();
1943         if (err < 0)
1944                 goto error_unreg_notifier;
1945
1946         return 0;
1947
1948 error_unreg_notifier:
1949         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1950 error_netns_exit:
1951         unregister_pernet_device(&ovs_net_ops);
1952 error_vport_exit:
1953         ovs_vport_exit();
1954 error_flow_exit:
1955         ovs_flow_exit();
1956 error:
1957         return err;
1958 }
1959
1960 static void dp_cleanup(void)
1961 {
1962         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1963         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1964         unregister_pernet_device(&ovs_net_ops);
1965         rcu_barrier();
1966         ovs_vport_exit();
1967         ovs_flow_exit();
1968 }
1969
1970 module_init(dp_init);
1971 module_exit(dp_cleanup);
1972
1973 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1974 MODULE_LICENSE("GPL");
1975 MODULE_VERSION(VERSION);