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, plen;
 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             (plen = (ALIGN(user_skb->len, NLA_ALIGNTO) - user_skb->len)) > 0)
 477                 memset(skb_put(user_skb, plen), 0, plen);
 478
 479         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 480
 481         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 482 out:
 483         kfree_skb(nskb);
 484         return err;
 485 }
 486
 487 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 488 {
 489         struct ovs_header *ovs_header = info->userhdr;
 490         struct nlattr **a = info->attrs;
 491         struct sw_flow_actions *acts;
 492         struct sk_buff *packet;
 493         struct sw_flow *flow;
 494         struct datapath *dp;
 495         struct ethhdr *eth;
 496         int len;
 497         int err;
 498
 499         err = -EINVAL;
 500         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 501             !a[OVS_PACKET_ATTR_ACTIONS])
 502                 goto err;
 503
 504         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 505         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 506         err = -ENOMEM;
 507         if (!packet)
 508                 goto err;
 509         skb_reserve(packet, NET_IP_ALIGN);
 510
 511         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 512
 513         skb_reset_mac_header(packet);
 514         eth = eth_hdr(packet);
 515
 516         /* Normally, setting the skb 'protocol' field would be handled by a
 517          * call to eth_type_trans(), but it assumes there's a sending
 518          * device, which we may not have. */
 519         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 520                 packet->protocol = eth->h_proto;
 521         else
 522                 packet->protocol = htons(ETH_P_802_2);
 523
 524         /* Build an sw_flow for sending this packet. */
 525         flow = ovs_flow_alloc(false);
 526         err = PTR_ERR(flow);
 527         if (IS_ERR(flow))
 528                 goto err_kfree_skb;
 529
 530         err = ovs_flow_extract(packet, -1, &flow->key);
 531         if (err)
 532                 goto err_flow_free;
 533
 534         err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
 535         if (err)
 536                 goto err_flow_free;
 537         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 538         err = PTR_ERR(acts);
 539         if (IS_ERR(acts))
 540                 goto err_flow_free;
 541
 542         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 543                                    &flow->key, 0, &acts);
 544         rcu_assign_pointer(flow->sf_acts, acts);
 545         if (err)
 546                 goto err_flow_free;
 547
 548         OVS_CB(packet)->flow = flow;
 549         OVS_CB(packet)->pkt_key = &flow->key;
 550         packet->priority = flow->key.phy.priority;
 551         packet->mark = flow->key.phy.skb_mark;
 552
 553         rcu_read_lock();
 554         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 555         err = -ENODEV;
 556         if (!dp)
 557                 goto err_unlock;
 558
 559         local_bh_disable();
 560         err = ovs_execute_actions(dp, packet);
 561         local_bh_enable();
 562         rcu_read_unlock();
 563
 564         ovs_flow_free(flow, false);
 565         return err;
 566
 567 err_unlock:
 568         rcu_read_unlock();
 569 err_flow_free:
 570         ovs_flow_free(flow, false);
 571 err_kfree_skb:
 572         kfree_skb(packet);
 573 err:
 574         return err;
 575 }
 576
 577 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 578         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 579         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 580         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 581 };
 582
 583 static struct genl_ops dp_packet_genl_ops[] = {
 584         { .cmd = OVS_PACKET_CMD_EXECUTE,
 585           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 586           .policy = packet_policy,
 587           .doit = ovs_packet_cmd_execute
 588         }
 589 };
 590
 591 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
 592                          struct ovs_dp_megaflow_stats *mega_stats)
 593 {
 594         int i;
 595
 596         memset(mega_stats, 0, sizeof(*mega_stats));
 597
 598         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 599         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 600
 601         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 602
 603         for_each_possible_cpu(i) {
 604                 const struct dp_stats_percpu *percpu_stats;
 605                 struct dp_stats_percpu local_stats;
 606                 unsigned int start;
 607
 608                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 609
 610                 do {
 611                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 612                         local_stats = *percpu_stats;
 613                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 614
 615                 stats->n_hit += local_stats.n_hit;
 616                 stats->n_missed += local_stats.n_missed;
 617                 stats->n_lost += local_stats.n_lost;
 618                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 619         }
 620 }
 621
 622 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 623         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 624         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 625         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 626 };
 627
 628 static struct genl_family dp_flow_genl_family = {
 629         .id = GENL_ID_GENERATE,
 630         .hdrsize = sizeof(struct ovs_header),
 631         .name = OVS_FLOW_FAMILY,
 632         .version = OVS_FLOW_VERSION,
 633         .maxattr = OVS_FLOW_ATTR_MAX,
 634         .netnsok = true,
 635          SET_PARALLEL_OPS
 636 };
 637
 638 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 639         .name = OVS_FLOW_MCGROUP
 640 };
 641
 642 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 643 {
 644         return NLMSG_ALIGN(sizeof(struct ovs_header))
 645                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 646                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 647                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 648                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 649                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 650                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 651 }
 652
 653 /* Called with ovs_mutex. */
 654 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 655                                   struct sk_buff *skb, u32 portid,
 656                                   u32 seq, u32 flags, u8 cmd)
 657 {
 658         const int skb_orig_len = skb->len;
 659         struct nlattr *start;
 660         struct ovs_flow_stats stats;
 661         __be16 tcp_flags;
 662         unsigned long used;
 663         struct ovs_header *ovs_header;
 664         struct nlattr *nla;
 665         int err;
 666
 667         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 668         if (!ovs_header)
 669                 return -EMSGSIZE;
 670
 671         ovs_header->dp_ifindex = get_dpifindex(dp);
 672
 673         /* Fill flow key. */
 674         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 675         if (!nla)
 676                 goto nla_put_failure;
 677
 678         err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
 679         if (err)
 680                 goto error;
 681         nla_nest_end(skb, nla);
 682
 683         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 684         if (!nla)
 685                 goto nla_put_failure;
 686
 687         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 688         if (err)
 689                 goto error;
 690
 691         nla_nest_end(skb, nla);
 692
 693         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 694         if (used &&
 695             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 696                 goto nla_put_failure;
 697
 698         if (stats.n_packets &&
 699             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 700                 goto nla_put_failure;
 701
 702         if ((u8)ntohs(tcp_flags) &&
 703              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 704                 goto nla_put_failure;
 705
 706         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 707          * this is the first flow to be dumped into 'skb'.  This is unusual for
 708          * Netlink but individual action lists can be longer than
 709          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 710          * The userspace caller can always fetch the actions separately if it
 711          * really wants them.  (Most userspace callers in fact don't care.)
 712          *
 713          * This can only fail for dump operations because the skb is always
 714          * properly sized for single flows.
 715          */
 716         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 717         if (start) {
 718                 const struct sw_flow_actions *sf_acts;
 719
 720                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 721
 722                 err = ovs_nla_put_actions(sf_acts->actions,
 723                                           sf_acts->actions_len, skb);
 724                 if (!err)
 725                         nla_nest_end(skb, start);
 726                 else {
 727                         if (skb_orig_len)
 728                                 goto error;
 729
 730                         nla_nest_cancel(skb, start);
 731                 }
 732         } else if (skb_orig_len)
 733                 goto nla_put_failure;
 734
 735         return genlmsg_end(skb, ovs_header);
 736
 737 nla_put_failure:
 738         err = -EMSGSIZE;
 739 error:
 740         genlmsg_cancel(skb, ovs_header);
 741         return err;
 742 }
 743
 744 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow,
 745                                                struct genl_info *info)
 746 {
 747         size_t len;
 748
 749         len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts));
 750
 751         return genlmsg_new_unicast(len, info, GFP_KERNEL);
 752 }
 753
 754 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 755                                                struct datapath *dp,
 756                                                struct genl_info *info,
 757                                                u8 cmd)
 758 {
 759         struct sk_buff *skb;
 760         int retval;
 761
 762         skb = ovs_flow_cmd_alloc_info(flow, info);
 763         if (!skb)
 764                 return ERR_PTR(-ENOMEM);
 765
 766         retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid,
 767                                         info->snd_seq, 0, cmd);
 768         BUG_ON(retval < 0);
 769         return skb;
 770 }
 771
 772 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 773 {
 774         struct nlattr **a = info->attrs;
 775         struct ovs_header *ovs_header = info->userhdr;
 776         struct sw_flow_key key, masked_key;
 777         struct sw_flow *flow = NULL;
 778         struct sw_flow_mask mask;
 779         struct sk_buff *reply;
 780         struct datapath *dp;
 781         struct sw_flow_actions *acts = NULL;
 782         struct sw_flow_match match;
 783         bool exact_5tuple;
 784         int error;
 785
 786         /* Extract key. */
 787         error = -EINVAL;
 788         if (!a[OVS_FLOW_ATTR_KEY])
 789                 goto error;
 790
 791         ovs_match_init(&match, &key, &mask);
 792         error = ovs_nla_get_match(&match, &exact_5tuple,
 793                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 794         if (error)
 795                 goto error;
 796
 797         /* Validate actions. */
 798         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 799                 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 800                 error = PTR_ERR(acts);
 801                 if (IS_ERR(acts))
 802                         goto error;
 803
 804                 ovs_flow_mask_key(&masked_key, &key, &mask);
 805                 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
 806                                              &masked_key, 0, &acts);
 807                 if (error) {
 808                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 809                         goto err_kfree;
 810                 }
 811         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 812                 error = -EINVAL;
 813                 goto error;
 814         }
 815
 816         ovs_lock();
 817         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 818         error = -ENODEV;
 819         if (!dp)
 820                 goto err_unlock_ovs;
 821
 822         /* Check if this is a duplicate flow */
 823         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 824         if (!flow) {
 825                 /* Bail out if we're not allowed to create a new flow. */
 826                 error = -ENOENT;
 827                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
 828                         goto err_unlock_ovs;
 829
 830                 /* Allocate flow. */
 831                 flow = ovs_flow_alloc(!exact_5tuple);
 832                 if (IS_ERR(flow)) {
 833                         error = PTR_ERR(flow);
 834                         goto err_unlock_ovs;
 835                 }
 836
 837                 flow->key = masked_key;
 838                 flow->unmasked_key = key;
 839                 rcu_assign_pointer(flow->sf_acts, acts);
 840
 841                 /* Put flow in bucket. */
 842                 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
 843                 if (error) {
 844                         acts = NULL;
 845                         goto err_flow_free;
 846                 }
 847
 848                 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
 849         } else {
 850                 /* We found a matching flow. */
 851                 struct sw_flow_actions *old_acts;
 852
 853                 /* Bail out if we're not allowed to modify an existing flow.
 854                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 855                  * because Generic Netlink treats the latter as a dump
 856                  * request.  We also accept NLM_F_EXCL in case that bug ever
 857                  * gets fixed.
 858                  */
 859                 error = -EEXIST;
 860                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
 861                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
 862                         goto err_unlock_ovs;
 863
 864                 /* The unmasked key has to be the same for flow updates. */
 865                 error = -EINVAL;
 866                 if (!ovs_flow_cmp_unmasked_key(flow, &match)) {
 867                         OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
 868                         goto err_unlock_ovs;
 869                 }
 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
1093         return msgsize;
1094 }
1095
1096 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1097                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1098 {
1099         struct ovs_header *ovs_header;
1100         struct ovs_dp_stats dp_stats;
1101         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1102         int err;
1103
1104         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1105                                    flags, cmd);
1106         if (!ovs_header)
1107                 goto error;
1108
1109         ovs_header->dp_ifindex = get_dpifindex(dp);
1110
1111         rcu_read_lock();
1112         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1113         rcu_read_unlock();
1114         if (err)
1115                 goto nla_put_failure;
1116
1117         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1118         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1119                         &dp_stats))
1120                 goto nla_put_failure;
1121
1122         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1123                         sizeof(struct ovs_dp_megaflow_stats),
1124                         &dp_megaflow_stats))
1125                 goto nla_put_failure;
1126
1127         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1128                 goto nla_put_failure;
1129
1130         return genlmsg_end(skb, ovs_header);
1131
1132 nla_put_failure:
1133         genlmsg_cancel(skb, ovs_header);
1134 error:
1135         return -EMSGSIZE;
1136 }
1137
1138 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp,
1139                                              struct genl_info *info, u8 cmd)
1140 {
1141         struct sk_buff *skb;
1142         int retval;
1143
1144         skb = genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1145         if (!skb)
1146                 return ERR_PTR(-ENOMEM);
1147
1148         retval = ovs_dp_cmd_fill_info(dp, skb, info->snd_portid, info->snd_seq, 0, cmd);
1149         if (retval < 0) {
1150                 kfree_skb(skb);
1151                 return ERR_PTR(retval);
1152         }
1153         return skb;
1154 }
1155
1156 /* Called with ovs_mutex. */
1157 static struct datapath *lookup_datapath(struct net *net,
1158                                         struct ovs_header *ovs_header,
1159                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1160 {
1161         struct datapath *dp;
1162
1163         if (!a[OVS_DP_ATTR_NAME])
1164                 dp = get_dp(net, ovs_header->dp_ifindex);
1165         else {
1166                 struct vport *vport;
1167
1168                 rcu_read_lock();
1169                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1170                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1171                 rcu_read_unlock();
1172         }
1173         return dp ? dp : ERR_PTR(-ENODEV);
1174 }
1175
1176 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1177 {
1178         struct datapath *dp;
1179
1180         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1181         if (!dp)
1182                 return;
1183
1184         WARN(dp->user_features, "Dropping previously announced user features\n");
1185         dp->user_features = 0;
1186 }
1187
1188 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1189 {
1190         if (a[OVS_DP_ATTR_USER_FEATURES])
1191                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1192 }
1193
1194 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1195 {
1196         struct nlattr **a = info->attrs;
1197         struct vport_parms parms;
1198         struct sk_buff *reply;
1199         struct datapath *dp;
1200         struct vport *vport;
1201         struct ovs_net *ovs_net;
1202         int err, i;
1203
1204         err = -EINVAL;
1205         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1206                 goto err;
1207
1208         ovs_lock();
1209
1210         err = -ENOMEM;
1211         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1212         if (dp == NULL)
1213                 goto err_unlock_ovs;
1214
1215         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1216
1217         /* Allocate table. */
1218         err = ovs_flow_tbl_init(&dp->table);
1219         if (err)
1220                 goto err_free_dp;
1221
1222         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1223         if (!dp->stats_percpu) {
1224                 err = -ENOMEM;
1225                 goto err_destroy_table;
1226         }
1227
1228         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1229                             GFP_KERNEL);
1230         if (!dp->ports) {
1231                 err = -ENOMEM;
1232                 goto err_destroy_percpu;
1233         }
1234
1235         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1236                 INIT_HLIST_HEAD(&dp->ports[i]);
1237
1238         /* Set up our datapath device. */
1239         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1240         parms.type = OVS_VPORT_TYPE_INTERNAL;
1241         parms.options = NULL;
1242         parms.dp = dp;
1243         parms.port_no = OVSP_LOCAL;
1244         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1245
1246         ovs_dp_change(dp, a);
1247
1248         vport = new_vport(&parms);
1249         if (IS_ERR(vport)) {
1250                 err = PTR_ERR(vport);
1251                 if (err == -EBUSY)
1252                         err = -EEXIST;
1253
1254                 if (err == -EEXIST) {
1255                         /* An outdated user space instance that does not understand
1256                          * the concept of user_features has attempted to create a new
1257                          * datapath and is likely to reuse it. Drop all user features.
1258                          */
1259                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1260                                 ovs_dp_reset_user_features(skb, info);
1261                 }
1262
1263                 goto err_destroy_ports_array;
1264         }
1265
1266         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1267         err = PTR_ERR(reply);
1268         if (IS_ERR(reply))
1269                 goto err_destroy_local_port;
1270
1271         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1272         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1273
1274         ovs_unlock();
1275
1276         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1277         return 0;
1278
1279 err_destroy_local_port:
1280         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1281 err_destroy_ports_array:
1282         kfree(dp->ports);
1283 err_destroy_percpu:
1284         free_percpu(dp->stats_percpu);
1285 err_destroy_table:
1286         ovs_flow_tbl_destroy(&dp->table, false);
1287 err_free_dp:
1288         release_net(ovs_dp_get_net(dp));
1289         kfree(dp);
1290 err_unlock_ovs:
1291         ovs_unlock();
1292 err:
1293         return err;
1294 }
1295
1296 /* Called with ovs_mutex. */
1297 static void __dp_destroy(struct datapath *dp)
1298 {
1299         int i;
1300
1301         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1302                 struct vport *vport;
1303                 struct hlist_node *n;
1304
1305                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1306                         if (vport->port_no != OVSP_LOCAL)
1307                                 ovs_dp_detach_port(vport);
1308         }
1309
1310         list_del_rcu(&dp->list_node);
1311
1312         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1313          * all ports in datapath are destroyed first before freeing datapath.
1314          */
1315         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1316
1317         /* RCU destroy the flow table */
1318         ovs_flow_tbl_destroy(&dp->table, true);
1319
1320         call_rcu(&dp->rcu, destroy_dp_rcu);
1321 }
1322
1323 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1324 {
1325         struct sk_buff *reply;
1326         struct datapath *dp;
1327         int err;
1328
1329         ovs_lock();
1330         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1331         err = PTR_ERR(dp);
1332         if (IS_ERR(dp))
1333                 goto unlock;
1334
1335         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_DEL);
1336         err = PTR_ERR(reply);
1337         if (IS_ERR(reply))
1338                 goto unlock;
1339
1340         __dp_destroy(dp);
1341         ovs_unlock();
1342
1343         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1344
1345         return 0;
1346 unlock:
1347         ovs_unlock();
1348         return err;
1349 }
1350
1351 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1352 {
1353         struct sk_buff *reply;
1354         struct datapath *dp;
1355         int err;
1356
1357         ovs_lock();
1358         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1359         err = PTR_ERR(dp);
1360         if (IS_ERR(dp))
1361                 goto unlock;
1362
1363         ovs_dp_change(dp, info->attrs);
1364
1365         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1366         if (IS_ERR(reply)) {
1367                 err = PTR_ERR(reply);
1368                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1369                                 ovs_dp_datapath_multicast_group.id, err);
1370                 err = 0;
1371                 goto unlock;
1372         }
1373
1374         ovs_unlock();
1375         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1376
1377         return 0;
1378 unlock:
1379         ovs_unlock();
1380         return err;
1381 }
1382
1383 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1384 {
1385         struct sk_buff *reply;
1386         struct datapath *dp;
1387         int err;
1388
1389         ovs_lock();
1390         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1391         if (IS_ERR(dp)) {
1392                 err = PTR_ERR(dp);
1393                 goto unlock;
1394         }
1395
1396         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1397         if (IS_ERR(reply)) {
1398                 err = PTR_ERR(reply);
1399                 goto unlock;
1400         }
1401
1402         ovs_unlock();
1403         return genlmsg_reply(reply, info);
1404
1405 unlock:
1406         ovs_unlock();
1407         return err;
1408 }
1409
1410 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1411 {
1412         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1413         struct datapath *dp;
1414         int skip = cb->args[0];
1415         int i = 0;
1416
1417         rcu_read_lock();
1418         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1419                 if (i >= skip &&
1420                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1421                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1422                                          OVS_DP_CMD_NEW) < 0)
1423                         break;
1424                 i++;
1425         }
1426         rcu_read_unlock();
1427
1428         cb->args[0] = i;
1429
1430         return skb->len;
1431 }
1432
1433 static struct genl_ops dp_datapath_genl_ops[] = {
1434         { .cmd = OVS_DP_CMD_NEW,
1435           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1436           .policy = datapath_policy,
1437           .doit = ovs_dp_cmd_new
1438         },
1439         { .cmd = OVS_DP_CMD_DEL,
1440           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1441           .policy = datapath_policy,
1442           .doit = ovs_dp_cmd_del
1443         },
1444         { .cmd = OVS_DP_CMD_GET,
1445           .flags = 0,               /* OK for unprivileged users. */
1446           .policy = datapath_policy,
1447           .doit = ovs_dp_cmd_get,
1448           .dumpit = ovs_dp_cmd_dump
1449         },
1450         { .cmd = OVS_DP_CMD_SET,
1451           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1452           .policy = datapath_policy,
1453           .doit = ovs_dp_cmd_set,
1454         },
1455 };
1456
1457 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1458         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1459         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1460         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1461         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1462         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1463         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1464 };
1465
1466 static struct genl_family dp_vport_genl_family = {
1467         .id = GENL_ID_GENERATE,
1468         .hdrsize = sizeof(struct ovs_header),
1469         .name = OVS_VPORT_FAMILY,
1470         .version = OVS_VPORT_VERSION,
1471         .maxattr = OVS_VPORT_ATTR_MAX,
1472         .netnsok = true,
1473          SET_PARALLEL_OPS
1474 };
1475
1476 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1477         .name = OVS_VPORT_MCGROUP
1478 };
1479
1480 /* Called with ovs_mutex or RCU read lock. */
1481 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1482                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1483 {
1484         struct ovs_header *ovs_header;
1485         struct ovs_vport_stats vport_stats;
1486         int err;
1487
1488         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1489                                  flags, cmd);
1490         if (!ovs_header)
1491                 return -EMSGSIZE;
1492
1493         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1494
1495         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1496             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1497             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1498             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1499                 goto nla_put_failure;
1500
1501         ovs_vport_get_stats(vport, &vport_stats);
1502         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1503                     &vport_stats))
1504                 goto nla_put_failure;
1505
1506         err = ovs_vport_get_options(vport, skb);
1507         if (err == -EMSGSIZE)
1508                 goto error;
1509
1510         return genlmsg_end(skb, ovs_header);
1511
1512 nla_put_failure:
1513         err = -EMSGSIZE;
1514 error:
1515         genlmsg_cancel(skb, ovs_header);
1516         return err;
1517 }
1518
1519 /* Called with ovs_mutex or RCU read lock. */
1520 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1521                                          u32 seq, u8 cmd)
1522 {
1523         struct sk_buff *skb;
1524         int retval;
1525
1526         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1527         if (!skb)
1528                 return ERR_PTR(-ENOMEM);
1529
1530         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1531         BUG_ON(retval < 0);
1532
1533         return skb;
1534 }
1535
1536 /* Called with ovs_mutex or RCU read lock. */
1537 static struct vport *lookup_vport(struct net *net,
1538                                   struct ovs_header *ovs_header,
1539                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1540 {
1541         struct datapath *dp;
1542         struct vport *vport;
1543
1544         if (a[OVS_VPORT_ATTR_NAME]) {
1545                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1546                 if (!vport)
1547                         return ERR_PTR(-ENODEV);
1548                 if (ovs_header->dp_ifindex &&
1549                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1550                         return ERR_PTR(-ENODEV);
1551                 return vport;
1552         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1553                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1554
1555                 if (port_no >= DP_MAX_PORTS)
1556                         return ERR_PTR(-EFBIG);
1557
1558                 dp = get_dp(net, ovs_header->dp_ifindex);
1559                 if (!dp)
1560                         return ERR_PTR(-ENODEV);
1561
1562                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1563                 if (!vport)
1564                         return ERR_PTR(-ENODEV);
1565                 return vport;
1566         } else
1567                 return ERR_PTR(-EINVAL);
1568 }
1569
1570 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1571 {
1572         struct nlattr **a = info->attrs;
1573         struct ovs_header *ovs_header = info->userhdr;
1574         struct vport_parms parms;
1575         struct sk_buff *reply;
1576         struct vport *vport;
1577         struct datapath *dp;
1578         u32 port_no;
1579         int err;
1580
1581         err = -EINVAL;
1582         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1583             !a[OVS_VPORT_ATTR_UPCALL_PID])
1584                 goto exit;
1585
1586         ovs_lock();
1587         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1588         err = -ENODEV;
1589         if (!dp)
1590                 goto exit_unlock;
1591
1592         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1593                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1594
1595                 err = -EFBIG;
1596                 if (port_no >= DP_MAX_PORTS)
1597                         goto exit_unlock;
1598
1599                 vport = ovs_vport_ovsl(dp, port_no);
1600                 err = -EBUSY;
1601                 if (vport)
1602                         goto exit_unlock;
1603         } else {
1604                 for (port_no = 1; ; port_no++) {
1605                         if (port_no >= DP_MAX_PORTS) {
1606                                 err = -EFBIG;
1607                                 goto exit_unlock;
1608                         }
1609                         vport = ovs_vport_ovsl(dp, port_no);
1610                         if (!vport)
1611                                 break;
1612                 }
1613         }
1614
1615         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1616         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1617         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1618         parms.dp = dp;
1619         parms.port_no = port_no;
1620         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1621
1622         vport = new_vport(&parms);
1623         err = PTR_ERR(vport);
1624         if (IS_ERR(vport))
1625                 goto exit_unlock;
1626
1627         err = 0;
1628         if (a[OVS_VPORT_ATTR_STATS])
1629                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1630
1631         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1632                                          OVS_VPORT_CMD_NEW);
1633         if (IS_ERR(reply)) {
1634                 err = PTR_ERR(reply);
1635                 ovs_dp_detach_port(vport);
1636                 goto exit_unlock;
1637         }
1638
1639         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1640
1641 exit_unlock:
1642         ovs_unlock();
1643 exit:
1644         return err;
1645 }
1646
1647 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1648 {
1649         struct nlattr **a = info->attrs;
1650         struct sk_buff *reply;
1651         struct vport *vport;
1652         int err;
1653
1654         ovs_lock();
1655         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1656         err = PTR_ERR(vport);
1657         if (IS_ERR(vport))
1658                 goto exit_unlock;
1659
1660         if (a[OVS_VPORT_ATTR_TYPE] &&
1661             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1662                 err = -EINVAL;
1663                 goto exit_unlock;
1664         }
1665
1666         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1667         if (!reply) {
1668                 err = -ENOMEM;
1669                 goto exit_unlock;
1670         }
1671
1672         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1673                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1674                 if (err)
1675                         goto exit_free;
1676         }
1677
1678         if (a[OVS_VPORT_ATTR_STATS])
1679                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1680
1681         if (a[OVS_VPORT_ATTR_UPCALL_PID])
1682                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1683
1684         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1685                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1686         BUG_ON(err < 0);
1687
1688         ovs_unlock();
1689         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1690         return 0;
1691
1692 exit_free:
1693         kfree_skb(reply);
1694 exit_unlock:
1695         ovs_unlock();
1696         return err;
1697 }
1698
1699 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1700 {
1701         struct nlattr **a = info->attrs;
1702         struct sk_buff *reply;
1703         struct vport *vport;
1704         int err;
1705
1706         ovs_lock();
1707         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1708         err = PTR_ERR(vport);
1709         if (IS_ERR(vport))
1710                 goto exit_unlock;
1711
1712         if (vport->port_no == OVSP_LOCAL) {
1713                 err = -EINVAL;
1714                 goto exit_unlock;
1715         }
1716
1717         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1718                                          info->snd_seq, OVS_VPORT_CMD_DEL);
1719         err = PTR_ERR(reply);
1720         if (IS_ERR(reply))
1721                 goto exit_unlock;
1722
1723         err = 0;
1724         ovs_dp_detach_port(vport);
1725
1726         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1727
1728 exit_unlock:
1729         ovs_unlock();
1730         return err;
1731 }
1732
1733 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1734 {
1735         struct nlattr **a = info->attrs;
1736         struct ovs_header *ovs_header = info->userhdr;
1737         struct sk_buff *reply;
1738         struct vport *vport;
1739         int err;
1740
1741         rcu_read_lock();
1742         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1743         err = PTR_ERR(vport);
1744         if (IS_ERR(vport))
1745                 goto exit_unlock;
1746
1747         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1748                                          info->snd_seq, OVS_VPORT_CMD_NEW);
1749         err = PTR_ERR(reply);
1750         if (IS_ERR(reply))
1751                 goto exit_unlock;
1752
1753         rcu_read_unlock();
1754
1755         return genlmsg_reply(reply, info);
1756
1757 exit_unlock:
1758         rcu_read_unlock();
1759         return err;
1760 }
1761
1762 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1763 {
1764         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1765         struct datapath *dp;
1766         int bucket = cb->args[0], skip = cb->args[1];
1767         int i, j = 0;
1768
1769         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1770         if (!dp)
1771                 return -ENODEV;
1772
1773         rcu_read_lock();
1774         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1775                 struct vport *vport;
1776
1777                 j = 0;
1778                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1779                         if (j >= skip &&
1780                             ovs_vport_cmd_fill_info(vport, skb,
1781                                                     NETLINK_CB(cb->skb).portid,
1782                                                     cb->nlh->nlmsg_seq,
1783                                                     NLM_F_MULTI,
1784                                                     OVS_VPORT_CMD_NEW) < 0)
1785                                 goto out;
1786
1787                         j++;
1788                 }
1789                 skip = 0;
1790         }
1791 out:
1792         rcu_read_unlock();
1793
1794         cb->args[0] = i;
1795         cb->args[1] = j;
1796
1797         return skb->len;
1798 }
1799
1800 static struct genl_ops dp_vport_genl_ops[] = {
1801         { .cmd = OVS_VPORT_CMD_NEW,
1802           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1803           .policy = vport_policy,
1804           .doit = ovs_vport_cmd_new
1805         },
1806         { .cmd = OVS_VPORT_CMD_DEL,
1807           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1808           .policy = vport_policy,
1809           .doit = ovs_vport_cmd_del
1810         },
1811         { .cmd = OVS_VPORT_CMD_GET,
1812           .flags = 0,               /* OK for unprivileged users. */
1813           .policy = vport_policy,
1814           .doit = ovs_vport_cmd_get,
1815           .dumpit = ovs_vport_cmd_dump
1816         },
1817         { .cmd = OVS_VPORT_CMD_SET,
1818           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1819           .policy = vport_policy,
1820           .doit = ovs_vport_cmd_set,
1821         },
1822 };
1823
1824 struct genl_family_and_ops {
1825         struct genl_family *family;
1826         struct genl_ops *ops;
1827         int n_ops;
1828         struct genl_multicast_group *group;
1829 };
1830
1831 static const struct genl_family_and_ops dp_genl_families[] = {
1832         { &dp_datapath_genl_family,
1833           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1834           &ovs_dp_datapath_multicast_group },
1835         { &dp_vport_genl_family,
1836           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1837           &ovs_dp_vport_multicast_group },
1838         { &dp_flow_genl_family,
1839           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1840           &ovs_dp_flow_multicast_group },
1841         { &dp_packet_genl_family,
1842           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1843           NULL },
1844 };
1845
1846 static void dp_unregister_genl(int n_families)
1847 {
1848         int i;
1849
1850         for (i = 0; i < n_families; i++)
1851                 genl_unregister_family(dp_genl_families[i].family);
1852 }
1853
1854 static int dp_register_genl(void)
1855 {
1856         int n_registered;
1857         int err;
1858         int i;
1859
1860         n_registered = 0;
1861         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1862                 const struct genl_family_and_ops *f = &dp_genl_families[i];
1863
1864                 err = genl_register_family_with_ops(f->family, f->ops,
1865                                                     f->n_ops);
1866                 if (err)
1867                         goto error;
1868                 n_registered++;
1869
1870                 if (f->group) {
1871                         err = genl_register_mc_group(f->family, f->group);
1872                         if (err)
1873                                 goto error;
1874                 }
1875         }
1876
1877         return 0;
1878
1879 error:
1880         dp_unregister_genl(n_registered);
1881         return err;
1882 }
1883
1884 static int __net_init ovs_init_net(struct net *net)
1885 {
1886         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1887
1888         INIT_LIST_HEAD(&ovs_net->dps);
1889         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1890         return 0;
1891 }
1892
1893 static void __net_exit ovs_exit_net(struct net *net)
1894 {
1895         struct datapath *dp, *dp_next;
1896         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1897
1898         ovs_lock();
1899         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1900                 __dp_destroy(dp);
1901         ovs_unlock();
1902
1903         cancel_work_sync(&ovs_net->dp_notify_work);
1904 }
1905
1906 static struct pernet_operations ovs_net_ops = {
1907         .init = ovs_init_net,
1908         .exit = ovs_exit_net,
1909         .id   = &ovs_net_id,
1910         .size = sizeof(struct ovs_net),
1911 };
1912
1913 DEFINE_COMPAT_PNET_REG_FUNC(device);
1914
1915 static int __init dp_init(void)
1916 {
1917         int err;
1918
1919         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1920
1921         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
1922                 VERSION);
1923
1924         err = ovs_flow_init();
1925         if (err)
1926                 goto error;
1927
1928         err = ovs_vport_init();
1929         if (err)
1930                 goto error_flow_exit;
1931
1932         err = register_pernet_device(&ovs_net_ops);
1933         if (err)
1934                 goto error_vport_exit;
1935
1936         err = register_netdevice_notifier(&ovs_dp_device_notifier);
1937         if (err)
1938                 goto error_netns_exit;
1939
1940         err = dp_register_genl();
1941         if (err < 0)
1942                 goto error_unreg_notifier;
1943
1944         return 0;
1945
1946 error_unreg_notifier:
1947         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1948 error_netns_exit:
1949         unregister_pernet_device(&ovs_net_ops);
1950 error_vport_exit:
1951         ovs_vport_exit();
1952 error_flow_exit:
1953         ovs_flow_exit();
1954 error:
1955         return err;
1956 }
1957
1958 static void dp_cleanup(void)
1959 {
1960         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1961         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1962         unregister_pernet_device(&ovs_net_ops);
1963         rcu_barrier();
1964         ovs_vport_exit();
1965         ovs_flow_exit();
1966 }
1967
1968 module_init(dp_init);
1969 module_exit(dp_cleanup);
1970
1971 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1972 MODULE_LICENSE("GPL");
1973 MODULE_VERSION(VERSION);