net/openvswitch/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/ethtool.h>
  40 #include <linux/wait.h>
  41 #include <asm/div64.h>
  42 #include <linux/highmem.h>
  43 #include <linux/netfilter_bridge.h>
  44 #include <linux/netfilter_ipv4.h>
  45 #include <linux/inetdevice.h>
  46 #include <linux/list.h>
  47 #include <linux/lockdep.h>
  48 #include <linux/openvswitch.h>
  49 #include <linux/rculist.h>
  50 #include <linux/dmi.h>
  51 #include <linux/workqueue.h>
  52 #include <net/genetlink.h>
  53 #include <net/net_namespace.h>
  54 #include <net/netns/generic.h>
  55
  56 #include "datapath.h"
  57 #include "flow.h"
  58 #include "flow_table.h"
  59 #include "flow_netlink.h"
  60 #include "vport-internal_dev.h"
  61 #include "vport-netdev.h"
  62
  63 int ovs_net_id __read_mostly;
  64
  65 /* Check if need to build a reply message.
  66  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  67 static bool ovs_must_notify(struct genl_info *info,
  68                             const struct genl_multicast_group *grp)
  69 {
  70         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  71                 netlink_has_listeners(genl_info_net(info)->genl_sock, 0);
  72 }
  73
  74 static void ovs_notify(struct genl_family *family,
  75                        struct sk_buff *skb, struct genl_info *info)
  76 {
  77         genl_notify(family, skb, genl_info_net(info), info->snd_portid,
  78                     0, info->nlhdr, GFP_KERNEL);
  79 }
  80
  81 /**
  82  * DOC: Locking:
  83  *
  84  * All writes e.g. Writes to device state (add/remove datapath, port, set
  85  * operations on vports, etc.), Writes to other state (flow table
  86  * modifications, set miscellaneous datapath parameters, etc.) are protected
  87  * by ovs_lock.
  88  *
  89  * Reads are protected by RCU.
  90  *
  91  * There are a few special cases (mostly stats) that have their own
  92  * synchronization but they nest under all of above and don't interact with
  93  * each other.
  94  *
  95  * The RTNL lock nests inside ovs_mutex.
  96  */
  97
  98 static DEFINE_MUTEX(ovs_mutex);
  99
 100 void ovs_lock(void)
 101 {
 102         mutex_lock(&ovs_mutex);
 103 }
 104
 105 void ovs_unlock(void)
 106 {
 107         mutex_unlock(&ovs_mutex);
 108 }
 109
 110 #ifdef CONFIG_LOCKDEP
 111 int lockdep_ovsl_is_held(void)
 112 {
 113         if (debug_locks)
 114                 return lockdep_is_held(&ovs_mutex);
 115         else
 116                 return 1;
 117 }
 118 #endif
 119
 120 static struct vport *new_vport(const struct vport_parms *);
 121 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 122                              const struct dp_upcall_info *);
 123 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 124                                   const struct dp_upcall_info *);
 125
 126 /* Must be called with rcu_read_lock or ovs_mutex. */
 127 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 128 {
 129         struct datapath *dp = NULL;
 130         struct net_device *dev;
 131
 132         rcu_read_lock();
 133         dev = dev_get_by_index_rcu(net, dp_ifindex);
 134         if (dev) {
 135                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 136                 if (vport)
 137                         dp = vport->dp;
 138         }
 139         rcu_read_unlock();
 140
 141         return dp;
 142 }
 143
 144 /* Must be called with rcu_read_lock or ovs_mutex. */
 145 static const char *ovs_dp_name(const struct datapath *dp)
 146 {
 147         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 148         return vport->ops->get_name(vport);
 149 }
 150
 151 static int get_dpifindex(struct datapath *dp)
 152 {
 153         struct vport *local;
 154         int ifindex;
 155
 156         rcu_read_lock();
 157
 158         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 159         if (local)
 160                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 161         else
 162                 ifindex = 0;
 163
 164         rcu_read_unlock();
 165
 166         return ifindex;
 167 }
 168
 169 static void destroy_dp_rcu(struct rcu_head *rcu)
 170 {
 171         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 172
 173         free_percpu(dp->stats_percpu);
 174         release_net(ovs_dp_get_net(dp));
 175         kfree(dp->ports);
 176         kfree(dp);
 177 }
 178
 179 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 180                                             u16 port_no)
 181 {
 182         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 183 }
 184
 185 /* Called with ovs_mutex or RCU read lock. */
 186 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 187 {
 188         struct vport *vport;
 189         struct hlist_head *head;
 190
 191         head = vport_hash_bucket(dp, port_no);
 192         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 193                 if (vport->port_no == port_no)
 194                         return vport;
 195         }
 196         return NULL;
 197 }
 198
 199 /* Called with ovs_mutex. */
 200 static struct vport *new_vport(const struct vport_parms *parms)
 201 {
 202         struct vport *vport;
 203
 204         vport = ovs_vport_add(parms);
 205         if (!IS_ERR(vport)) {
 206                 struct datapath *dp = parms->dp;
 207                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 208
 209                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 210         }
 211         return vport;
 212 }
 213
 214 void ovs_dp_detach_port(struct vport *p)
 215 {
 216         ASSERT_OVSL();
 217
 218         /* First drop references to device. */
 219         hlist_del_rcu(&p->dp_hash_node);
 220
 221         /* Then destroy it. */
 222         ovs_vport_del(p);
 223 }
 224
 225 /* Must be called with rcu_read_lock. */
 226 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 227 {
 228         struct datapath *dp = p->dp;
 229         struct sw_flow *flow;
 230         struct dp_stats_percpu *stats;
 231         struct sw_flow_key key;
 232         u64 *stats_counter;
 233         u32 n_mask_hit;
 234         int error;
 235
 236         stats = this_cpu_ptr(dp->stats_percpu);
 237
 238         /* Extract flow from 'skb' into 'key'. */
 239         error = ovs_flow_extract(skb, p->port_no, &key);
 240         if (unlikely(error)) {
 241                 kfree_skb(skb);
 242                 return;
 243         }
 244
 245         /* Look up flow. */
 246         flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
 247         if (unlikely(!flow)) {
 248                 struct dp_upcall_info upcall;
 249
 250                 upcall.cmd = OVS_PACKET_CMD_MISS;
 251                 upcall.key = &key;
 252                 upcall.userdata = NULL;
 253                 upcall.portid = p->upcall_portid;
 254                 ovs_dp_upcall(dp, skb, &upcall);
 255                 consume_skb(skb);
 256                 stats_counter = &stats->n_missed;
 257                 goto out;
 258         }
 259
 260         OVS_CB(skb)->flow = flow;
 261         OVS_CB(skb)->pkt_key = &key;
 262
 263         ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
 264         ovs_execute_actions(dp, skb);
 265         stats_counter = &stats->n_hit;
 266
 267 out:
 268         /* Update datapath statistics. */
 269         u64_stats_update_begin(&stats->syncp);
 270         (*stats_counter)++;
 271         stats->n_mask_hit += n_mask_hit;
 272         u64_stats_update_end(&stats->syncp);
 273 }
 274
 275 static struct genl_family dp_packet_genl_family = {
 276         .id = GENL_ID_GENERATE,
 277         .hdrsize = sizeof(struct ovs_header),
 278         .name = OVS_PACKET_FAMILY,
 279         .version = OVS_PACKET_VERSION,
 280         .maxattr = OVS_PACKET_ATTR_MAX,
 281         .netnsok = true,
 282         .parallel_ops = true,
 283 };
 284
 285 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 286                   const struct dp_upcall_info *upcall_info)
 287 {
 288         struct dp_stats_percpu *stats;
 289         int err;
 290
 291         if (upcall_info->portid == 0) {
 292                 err = -ENOTCONN;
 293                 goto err;
 294         }
 295
 296         if (!skb_is_gso(skb))
 297                 err = queue_userspace_packet(dp, skb, upcall_info);
 298         else
 299                 err = queue_gso_packets(dp, skb, upcall_info);
 300         if (err)
 301                 goto err;
 302
 303         return 0;
 304
 305 err:
 306         stats = this_cpu_ptr(dp->stats_percpu);
 307
 308         u64_stats_update_begin(&stats->syncp);
 309         stats->n_lost++;
 310         u64_stats_update_end(&stats->syncp);
 311
 312         return err;
 313 }
 314
 315 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 316                              const struct dp_upcall_info *upcall_info)
 317 {
 318         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 319         struct dp_upcall_info later_info;
 320         struct sw_flow_key later_key;
 321         struct sk_buff *segs, *nskb;
 322         int err;
 323
 324         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 325         if (IS_ERR(segs))
 326                 return PTR_ERR(segs);
 327
 328         /* Queue all of the segments. */
 329         skb = segs;
 330         do {
 331                 err = queue_userspace_packet(dp, skb, upcall_info);
 332                 if (err)
 333                         break;
 334
 335                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 336                         /* The initial flow key extracted by ovs_flow_extract()
 337                          * in this case is for a first fragment, so we need to
 338                          * properly mark later fragments.
 339                          */
 340                         later_key = *upcall_info->key;
 341                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 342
 343                         later_info = *upcall_info;
 344                         later_info.key = &later_key;
 345                         upcall_info = &later_info;
 346                 }
 347         } while ((skb = skb->next));
 348
 349         /* Free all of the segments. */
 350         skb = segs;
 351         do {
 352                 nskb = skb->next;
 353                 if (err)
 354                         kfree_skb(skb);
 355                 else
 356                         consume_skb(skb);
 357         } while ((skb = nskb));
 358         return err;
 359 }
 360
 361 static size_t key_attr_size(void)
 362 {
 363         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 364                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 365                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
 366                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 367                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 368                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
 369                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
 370                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 371                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
 372                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 373                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 374                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 375                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 376                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 377                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 378                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 379                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 380                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 381                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 382 }
 383
 384 static size_t upcall_msg_size(const struct nlattr *userdata,
 385                               unsigned int hdrlen)
 386 {
 387         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 388                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 389                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 390
 391         /* OVS_PACKET_ATTR_USERDATA */
 392         if (userdata)
 393                 size += NLA_ALIGN(userdata->nla_len);
 394
 395         return size;
 396 }
 397
 398 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 399                                   const struct dp_upcall_info *upcall_info)
 400 {
 401         struct ovs_header *upcall;
 402         struct sk_buff *nskb = NULL;
 403         struct sk_buff *user_skb; /* to be queued to userspace */
 404         struct nlattr *nla;
 405         struct genl_info info = {
 406                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 407                 .snd_portid = upcall_info->portid,
 408         };
 409         size_t len;
 410         unsigned int hlen;
 411         int err, dp_ifindex;
 412
 413         dp_ifindex = get_dpifindex(dp);
 414         if (!dp_ifindex)
 415                 return -ENODEV;
 416
 417         if (vlan_tx_tag_present(skb)) {
 418                 nskb = skb_clone(skb, GFP_ATOMIC);
 419                 if (!nskb)
 420                         return -ENOMEM;
 421
 422                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 423                 if (!nskb)
 424                         return -ENOMEM;
 425
 426                 nskb->vlan_tci = 0;
 427                 skb = nskb;
 428         }
 429
 430         if (nla_attr_size(skb->len) > USHRT_MAX) {
 431                 err = -EFBIG;
 432                 goto out;
 433         }
 434
 435         /* Complete checksum if needed */
 436         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 437             (err = skb_checksum_help(skb)))
 438                 goto out;
 439
 440         /* Older versions of OVS user space enforce alignment of the last
 441          * Netlink attribute to NLA_ALIGNTO which would require extensive
 442          * padding logic. Only perform zerocopy if padding is not required.
 443          */
 444         if (dp->user_features & OVS_DP_F_UNALIGNED)
 445                 hlen = skb_zerocopy_headlen(skb);
 446         else
 447                 hlen = skb->len;
 448
 449         len = upcall_msg_size(upcall_info->userdata, hlen);
 450         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 451         if (!user_skb) {
 452                 err = -ENOMEM;
 453                 goto out;
 454         }
 455
 456         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 457                              0, upcall_info->cmd);
 458         upcall->dp_ifindex = dp_ifindex;
 459
 460         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 461         ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
 462         nla_nest_end(user_skb, nla);
 463
 464         if (upcall_info->userdata)
 465                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 466                           nla_len(upcall_info->userdata),
 467                           nla_data(upcall_info->userdata));
 468
 469         /* Only reserve room for attribute header, packet data is added
 470          * in skb_zerocopy() */
 471         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 472                 err = -ENOBUFS;
 473                 goto out;
 474         }
 475         nla->nla_len = nla_attr_size(skb->len);
 476
 477         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 478         if (err)
 479                 goto out;
 480
 481         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 482         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 483                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 484
 485                 if (plen > 0)
 486                         memset(skb_put(user_skb, plen), 0, plen);
 487         }
 488
 489         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 490
 491         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 492 out:
 493         if (err)
 494                 skb_tx_error(skb);
 495         kfree_skb(nskb);
 496         return err;
 497 }
 498
 499 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 500 {
 501         struct ovs_header *ovs_header = info->userhdr;
 502         struct nlattr **a = info->attrs;
 503         struct sw_flow_actions *acts;
 504         struct sk_buff *packet;
 505         struct sw_flow *flow;
 506         struct datapath *dp;
 507         struct ethhdr *eth;
 508         int len;
 509         int err;
 510
 511         err = -EINVAL;
 512         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 513             !a[OVS_PACKET_ATTR_ACTIONS])
 514                 goto err;
 515
 516         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 517         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 518         err = -ENOMEM;
 519         if (!packet)
 520                 goto err;
 521         skb_reserve(packet, NET_IP_ALIGN);
 522
 523         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 524
 525         skb_reset_mac_header(packet);
 526         eth = eth_hdr(packet);
 527
 528         /* Normally, setting the skb 'protocol' field would be handled by a
 529          * call to eth_type_trans(), but it assumes there's a sending
 530          * device, which we may not have. */
 531         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 532                 packet->protocol = eth->h_proto;
 533         else
 534                 packet->protocol = htons(ETH_P_802_2);
 535
 536         /* Build an sw_flow for sending this packet. */
 537         flow = ovs_flow_alloc();
 538         err = PTR_ERR(flow);
 539         if (IS_ERR(flow))
 540                 goto err_kfree_skb;
 541
 542         err = ovs_flow_extract(packet, -1, &flow->key);
 543         if (err)
 544                 goto err_flow_free;
 545
 546         err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
 547         if (err)
 548                 goto err_flow_free;
 549         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 550         err = PTR_ERR(acts);
 551         if (IS_ERR(acts))
 552                 goto err_flow_free;
 553
 554         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 555                                    &flow->key, 0, &acts);
 556         rcu_assign_pointer(flow->sf_acts, acts);
 557         if (err)
 558                 goto err_flow_free;
 559
 560         OVS_CB(packet)->flow = flow;
 561         OVS_CB(packet)->pkt_key = &flow->key;
 562         packet->priority = flow->key.phy.priority;
 563         packet->mark = flow->key.phy.skb_mark;
 564
 565         rcu_read_lock();
 566         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 567         err = -ENODEV;
 568         if (!dp)
 569                 goto err_unlock;
 570
 571         local_bh_disable();
 572         err = ovs_execute_actions(dp, packet);
 573         local_bh_enable();
 574         rcu_read_unlock();
 575
 576         ovs_flow_free(flow, false);
 577         return err;
 578
 579 err_unlock:
 580         rcu_read_unlock();
 581 err_flow_free:
 582         ovs_flow_free(flow, false);
 583 err_kfree_skb:
 584         kfree_skb(packet);
 585 err:
 586         return err;
 587 }
 588
 589 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 590         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 591         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 592         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 593 };
 594
 595 static const struct genl_ops dp_packet_genl_ops[] = {
 596         { .cmd = OVS_PACKET_CMD_EXECUTE,
 597           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 598           .policy = packet_policy,
 599           .doit = ovs_packet_cmd_execute
 600         }
 601 };
 602
 603 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
 604                          struct ovs_dp_megaflow_stats *mega_stats)
 605 {
 606         int i;
 607
 608         memset(mega_stats, 0, sizeof(*mega_stats));
 609
 610         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 611         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 612
 613         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 614
 615         for_each_possible_cpu(i) {
 616                 const struct dp_stats_percpu *percpu_stats;
 617                 struct dp_stats_percpu local_stats;
 618                 unsigned int start;
 619
 620                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 621
 622                 do {
 623                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 624                         local_stats = *percpu_stats;
 625                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 626
 627                 stats->n_hit += local_stats.n_hit;
 628                 stats->n_missed += local_stats.n_missed;
 629                 stats->n_lost += local_stats.n_lost;
 630                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 631         }
 632 }
 633
 634 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 635         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 636         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 637         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 638 };
 639
 640 static struct genl_family dp_flow_genl_family = {
 641         .id = GENL_ID_GENERATE,
 642         .hdrsize = sizeof(struct ovs_header),
 643         .name = OVS_FLOW_FAMILY,
 644         .version = OVS_FLOW_VERSION,
 645         .maxattr = OVS_FLOW_ATTR_MAX,
 646         .netnsok = true,
 647         .parallel_ops = true,
 648 };
 649
 650 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 651         .name = OVS_FLOW_MCGROUP
 652 };
 653
 654 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 655 {
 656         return NLMSG_ALIGN(sizeof(struct ovs_header))
 657                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 658                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 659                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 660                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 661                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 662                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 663 }
 664
 665 /* Called with ovs_mutex or RCU read lock. */
 666 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 667                                   struct sk_buff *skb, u32 portid,
 668                                   u32 seq, u32 flags, u8 cmd)
 669 {
 670         const int skb_orig_len = skb->len;
 671         struct nlattr *start;
 672         struct ovs_flow_stats stats;
 673         __be16 tcp_flags;
 674         unsigned long used;
 675         struct ovs_header *ovs_header;
 676         struct nlattr *nla;
 677         int err;
 678
 679         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 680         if (!ovs_header)
 681                 return -EMSGSIZE;
 682
 683         ovs_header->dp_ifindex = get_dpifindex(dp);
 684
 685         /* Fill flow key. */
 686         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 687         if (!nla)
 688                 goto nla_put_failure;
 689
 690         err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
 691         if (err)
 692                 goto error;
 693         nla_nest_end(skb, nla);
 694
 695         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 696         if (!nla)
 697                 goto nla_put_failure;
 698
 699         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 700         if (err)
 701                 goto error;
 702
 703         nla_nest_end(skb, nla);
 704
 705         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 706         if (used &&
 707             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 708                 goto nla_put_failure;
 709
 710         if (stats.n_packets &&
 711             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 712                 goto nla_put_failure;
 713
 714         if ((u8)ntohs(tcp_flags) &&
 715              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 716                 goto nla_put_failure;
 717
 718         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 719          * this is the first flow to be dumped into 'skb'.  This is unusual for
 720          * Netlink but individual action lists can be longer than
 721          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 722          * The userspace caller can always fetch the actions separately if it
 723          * really wants them.  (Most userspace callers in fact don't care.)
 724          *
 725          * This can only fail for dump operations because the skb is always
 726          * properly sized for single flows.
 727          */
 728         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 729         if (start) {
 730                 const struct sw_flow_actions *sf_acts;
 731
 732                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 733
 734                 err = ovs_nla_put_actions(sf_acts->actions,
 735                                           sf_acts->actions_len, skb);
 736                 if (!err)
 737                         nla_nest_end(skb, start);
 738                 else {
 739                         if (skb_orig_len)
 740                                 goto error;
 741
 742                         nla_nest_cancel(skb, start);
 743                 }
 744         } else if (skb_orig_len)
 745                 goto nla_put_failure;
 746
 747         return genlmsg_end(skb, ovs_header);
 748
 749 nla_put_failure:
 750         err = -EMSGSIZE;
 751 error:
 752         genlmsg_cancel(skb, ovs_header);
 753         return err;
 754 }
 755
 756 /* Must be called with ovs_mutex. */
 757 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow,
 758                                                struct genl_info *info,
 759                                                bool always)
 760 {
 761         struct sk_buff *skb;
 762         size_t len;
 763
 764         if (!always && !ovs_must_notify(info, &ovs_dp_flow_multicast_group))
 765                 return NULL;
 766
 767         len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts));
 768
 769         skb = genlmsg_new_unicast(len, info, GFP_KERNEL);
 770         if (!skb)
 771                 return ERR_PTR(-ENOMEM);
 772
 773         return skb;
 774 }
 775
 776 /* Must be called with ovs_mutex. */
 777 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 778                                                struct datapath *dp,
 779                                                struct genl_info *info,
 780                                                u8 cmd, bool always)
 781 {
 782         struct sk_buff *skb;
 783         int retval;
 784
 785         skb = ovs_flow_cmd_alloc_info(flow, info, always);
 786         if (!skb || IS_ERR(skb))
 787                 return skb;
 788
 789         retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid,
 790                                         info->snd_seq, 0, cmd);
 791         BUG_ON(retval < 0);
 792         return skb;
 793 }
 794
 795 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 796 {
 797         struct nlattr **a = info->attrs;
 798         struct ovs_header *ovs_header = info->userhdr;
 799         struct sw_flow_key key, masked_key;
 800         struct sw_flow *flow = NULL;
 801         struct sw_flow_mask mask;
 802         struct sk_buff *reply;
 803         struct datapath *dp;
 804         struct sw_flow_actions *acts = NULL;
 805         struct sw_flow_match match;
 806         int error;
 807
 808         /* Extract key. */
 809         error = -EINVAL;
 810         if (!a[OVS_FLOW_ATTR_KEY])
 811                 goto error;
 812
 813         ovs_match_init(&match, &key, &mask);
 814         error = ovs_nla_get_match(&match,
 815                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 816         if (error)
 817                 goto error;
 818
 819         /* Validate actions. */
 820         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 821                 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 822                 error = PTR_ERR(acts);
 823                 if (IS_ERR(acts))
 824                         goto error;
 825
 826                 ovs_flow_mask_key(&masked_key, &key, &mask);
 827                 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
 828                                              &masked_key, 0, &acts);
 829                 if (error) {
 830                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 831                         goto err_kfree;
 832                 }
 833         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 834                 /* OVS_FLOW_CMD_NEW must have actions. */
 835                 error = -EINVAL;
 836                 goto error;
 837         }
 838
 839         ovs_lock();
 840         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 841         error = -ENODEV;
 842         if (!dp)
 843                 goto err_unlock_ovs;
 844
 845         /* Check if this is a duplicate flow */
 846         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 847         if (!flow) {
 848                 /* Bail out if we're not allowed to create a new flow. */
 849                 error = -ENOENT;
 850                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
 851                         goto err_unlock_ovs;
 852
 853                 /* Allocate flow. */
 854                 flow = ovs_flow_alloc();
 855                 if (IS_ERR(flow)) {
 856                         error = PTR_ERR(flow);
 857                         goto err_unlock_ovs;
 858                 }
 859
 860                 flow->key = masked_key;
 861                 flow->unmasked_key = key;
 862                 rcu_assign_pointer(flow->sf_acts, acts);
 863
 864                 /* Put flow in bucket. */
 865                 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
 866                 if (error) {
 867                         acts = NULL;
 868                         goto err_flow_free;
 869                 }
 870
 871                 reply = ovs_flow_cmd_build_info(flow, dp, info,
 872                                                 OVS_FLOW_CMD_NEW, false);
 873         } else {
 874                 /* We found a matching flow. */
 875                 /* Bail out if we're not allowed to modify an existing flow.
 876                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 877                  * because Generic Netlink treats the latter as a dump
 878                  * request.  We also accept NLM_F_EXCL in case that bug ever
 879                  * gets fixed.
 880                  */
 881                 error = -EEXIST;
 882                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
 883                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
 884                         goto err_unlock_ovs;
 885
 886                 /* The unmasked key has to be the same for flow updates. */
 887                 if (!ovs_flow_cmp_unmasked_key(flow, &match))
 888                         goto err_unlock_ovs;
 889
 890                 /* Update actions, if present. */
 891                 if (acts) {
 892                         struct sw_flow_actions *old_acts;
 893
 894                         old_acts = ovsl_dereference(flow->sf_acts);
 895                         rcu_assign_pointer(flow->sf_acts, acts);
 896                         ovs_nla_free_flow_actions(old_acts);
 897                 }
 898                 reply = ovs_flow_cmd_build_info(flow, dp, info,
 899                                                 OVS_FLOW_CMD_NEW, false);
 900
 901                 /* Clear stats. */
 902                 if (a[OVS_FLOW_ATTR_CLEAR])
 903                         ovs_flow_stats_clear(flow);
 904         }
 905         ovs_unlock();
 906
 907         if (reply) {
 908                 if (!IS_ERR(reply))
 909                         ovs_notify(&dp_flow_genl_family, reply, info);
 910                 else
 911                         genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
 912                                      0, PTR_ERR(reply));
 913         }
 914
 915         return 0;
 916
 917 err_flow_free:
 918         ovs_flow_free(flow, false);
 919 err_unlock_ovs:
 920         ovs_unlock();
 921 err_kfree:
 922         kfree(acts);
 923 error:
 924         return error;
 925 }
 926
 927 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
 928 {
 929         struct nlattr **a = info->attrs;
 930         struct ovs_header *ovs_header = info->userhdr;
 931         struct sw_flow_key key;
 932         struct sk_buff *reply;
 933         struct sw_flow *flow;
 934         struct datapath *dp;
 935         struct sw_flow_match match;
 936         int err;
 937
 938         if (!a[OVS_FLOW_ATTR_KEY]) {
 939                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
 940                 return -EINVAL;
 941         }
 942
 943         ovs_match_init(&match, &key, NULL);
 944         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
 945         if (err)
 946                 return err;
 947
 948         ovs_lock();
 949         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 950         if (!dp) {
 951                 err = -ENODEV;
 952                 goto unlock;
 953         }
 954
 955         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 956         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
 957                 err = -ENOENT;
 958                 goto unlock;
 959         }
 960
 961         reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW, true);
 962         if (IS_ERR(reply)) {
 963                 err = PTR_ERR(reply);
 964                 goto unlock;
 965         }
 966
 967         ovs_unlock();
 968         return genlmsg_reply(reply, info);
 969 unlock:
 970         ovs_unlock();
 971         return err;
 972 }
 973
 974 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
 975 {
 976         struct nlattr **a = info->attrs;
 977         struct ovs_header *ovs_header = info->userhdr;
 978         struct sw_flow_key key;
 979         struct sk_buff *reply;
 980         struct sw_flow *flow;
 981         struct datapath *dp;
 982         struct sw_flow_match match;
 983         int err;
 984
 985         ovs_lock();
 986         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 987         if (!dp) {
 988                 err = -ENODEV;
 989                 goto unlock;
 990         }
 991
 992         if (!a[OVS_FLOW_ATTR_KEY]) {
 993                 err = ovs_flow_tbl_flush(&dp->table);
 994                 goto unlock;
 995         }
 996
 997         ovs_match_init(&match, &key, NULL);
 998         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
 999         if (err)
1000                 goto unlock;
1001
1002         flow = ovs_flow_tbl_lookup(&dp->table, &key);
1003         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
1004                 err = -ENOENT;
1005                 goto unlock;
1006         }
1007
1008         reply = ovs_flow_cmd_alloc_info(flow, info, false);
1009         if (IS_ERR(reply)) {
1010                 err = PTR_ERR(reply);
1011                 goto unlock;
1012         }
1013
1014         ovs_flow_tbl_remove(&dp->table, flow);
1015
1016         if (reply) {
1017                 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1018                                              info->snd_seq, 0,
1019                                              OVS_FLOW_CMD_DEL);
1020                 BUG_ON(err < 0);
1021         }
1022         ovs_flow_free(flow, true);
1023         ovs_unlock();
1024
1025         if (reply)
1026                 ovs_notify(&dp_flow_genl_family, reply, info);
1027         return 0;
1028 unlock:
1029         ovs_unlock();
1030         return err;
1031 }
1032
1033 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1034 {
1035         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1036         struct table_instance *ti;
1037         struct datapath *dp;
1038
1039         rcu_read_lock();
1040         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1041         if (!dp) {
1042                 rcu_read_unlock();
1043                 return -ENODEV;
1044         }
1045
1046         ti = rcu_dereference(dp->table.ti);
1047         for (;;) {
1048                 struct sw_flow *flow;
1049                 u32 bucket, obj;
1050
1051                 bucket = cb->args[0];
1052                 obj = cb->args[1];
1053                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1054                 if (!flow)
1055                         break;
1056
1057                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1058                                            NETLINK_CB(cb->skb).portid,
1059                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1060                                            OVS_FLOW_CMD_NEW) < 0)
1061                         break;
1062
1063                 cb->args[0] = bucket;
1064                 cb->args[1] = obj;
1065         }
1066         rcu_read_unlock();
1067         return skb->len;
1068 }
1069
1070 static const struct genl_ops dp_flow_genl_ops[] = {
1071         { .cmd = OVS_FLOW_CMD_NEW,
1072           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1073           .policy = flow_policy,
1074           .doit = ovs_flow_cmd_new_or_set
1075         },
1076         { .cmd = OVS_FLOW_CMD_DEL,
1077           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1078           .policy = flow_policy,
1079           .doit = ovs_flow_cmd_del
1080         },
1081         { .cmd = OVS_FLOW_CMD_GET,
1082           .flags = 0,               /* OK for unprivileged users. */
1083           .policy = flow_policy,
1084           .doit = ovs_flow_cmd_get,
1085           .dumpit = ovs_flow_cmd_dump
1086         },
1087         { .cmd = OVS_FLOW_CMD_SET,
1088           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1089           .policy = flow_policy,
1090           .doit = ovs_flow_cmd_new_or_set,
1091         },
1092 };
1093
1094 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1095         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1096         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1097         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1098 };
1099
1100 static struct genl_family dp_datapath_genl_family = {
1101         .id = GENL_ID_GENERATE,
1102         .hdrsize = sizeof(struct ovs_header),
1103         .name = OVS_DATAPATH_FAMILY,
1104         .version = OVS_DATAPATH_VERSION,
1105         .maxattr = OVS_DP_ATTR_MAX,
1106         .netnsok = true,
1107         .parallel_ops = true,
1108 };
1109
1110 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1111         .name = OVS_DATAPATH_MCGROUP
1112 };
1113
1114 static size_t ovs_dp_cmd_msg_size(void)
1115 {
1116         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1117
1118         msgsize += nla_total_size(IFNAMSIZ);
1119         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1120         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1121         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1122
1123         return msgsize;
1124 }
1125
1126 /* Called with ovs_mutex or RCU read lock. */
1127 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1128                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1129 {
1130         struct ovs_header *ovs_header;
1131         struct ovs_dp_stats dp_stats;
1132         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1133         int err;
1134
1135         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1136                                    flags, cmd);
1137         if (!ovs_header)
1138                 goto error;
1139
1140         ovs_header->dp_ifindex = get_dpifindex(dp);
1141
1142         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1143         if (err)
1144                 goto nla_put_failure;
1145
1146         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1147         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1148                         &dp_stats))
1149                 goto nla_put_failure;
1150
1151         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1152                         sizeof(struct ovs_dp_megaflow_stats),
1153                         &dp_megaflow_stats))
1154                 goto nla_put_failure;
1155
1156         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1157                 goto nla_put_failure;
1158
1159         return genlmsg_end(skb, ovs_header);
1160
1161 nla_put_failure:
1162         genlmsg_cancel(skb, ovs_header);
1163 error:
1164         return -EMSGSIZE;
1165 }
1166
1167 /* Must be called with ovs_mutex. */
1168 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp,
1169                                              struct genl_info *info, u8 cmd)
1170 {
1171         struct sk_buff *skb;
1172         int retval;
1173
1174         skb = genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1175         if (!skb)
1176                 return ERR_PTR(-ENOMEM);
1177
1178         retval = ovs_dp_cmd_fill_info(dp, skb, info->snd_portid, info->snd_seq, 0, cmd);
1179         if (retval < 0) {
1180                 kfree_skb(skb);
1181                 return ERR_PTR(retval);
1182         }
1183         return skb;
1184 }
1185
1186 /* Called with rcu_read_lock or ovs_mutex. */
1187 static struct datapath *lookup_datapath(struct net *net,
1188                                         struct ovs_header *ovs_header,
1189                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1190 {
1191         struct datapath *dp;
1192
1193         if (!a[OVS_DP_ATTR_NAME])
1194                 dp = get_dp(net, ovs_header->dp_ifindex);
1195         else {
1196                 struct vport *vport;
1197
1198                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1199                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1200         }
1201         return dp ? dp : ERR_PTR(-ENODEV);
1202 }
1203
1204 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1205 {
1206         struct datapath *dp;
1207
1208         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1209         if (IS_ERR(dp))
1210                 return;
1211
1212         WARN(dp->user_features, "Dropping previously announced user features\n");
1213         dp->user_features = 0;
1214 }
1215
1216 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1217 {
1218         if (a[OVS_DP_ATTR_USER_FEATURES])
1219                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1220 }
1221
1222 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1223 {
1224         struct nlattr **a = info->attrs;
1225         struct vport_parms parms;
1226         struct sk_buff *reply;
1227         struct datapath *dp;
1228         struct vport *vport;
1229         struct ovs_net *ovs_net;
1230         int err, i;
1231
1232         err = -EINVAL;
1233         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1234                 goto err;
1235
1236         ovs_lock();
1237
1238         err = -ENOMEM;
1239         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1240         if (dp == NULL)
1241                 goto err_unlock_ovs;
1242
1243         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1244
1245         /* Allocate table. */
1246         err = ovs_flow_tbl_init(&dp->table);
1247         if (err)
1248                 goto err_free_dp;
1249
1250         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1251         if (!dp->stats_percpu) {
1252                 err = -ENOMEM;
1253                 goto err_destroy_table;
1254         }
1255
1256         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1257                             GFP_KERNEL);
1258         if (!dp->ports) {
1259                 err = -ENOMEM;
1260                 goto err_destroy_percpu;
1261         }
1262
1263         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1264                 INIT_HLIST_HEAD(&dp->ports[i]);
1265
1266         /* Set up our datapath device. */
1267         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1268         parms.type = OVS_VPORT_TYPE_INTERNAL;
1269         parms.options = NULL;
1270         parms.dp = dp;
1271         parms.port_no = OVSP_LOCAL;
1272         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1273
1274         ovs_dp_change(dp, a);
1275
1276         vport = new_vport(&parms);
1277         if (IS_ERR(vport)) {
1278                 err = PTR_ERR(vport);
1279                 if (err == -EBUSY)
1280                         err = -EEXIST;
1281
1282                 if (err == -EEXIST) {
1283                         /* An outdated user space instance that does not understand
1284                          * the concept of user_features has attempted to create a new
1285                          * datapath and is likely to reuse it. Drop all user features.
1286                          */
1287                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1288                                 ovs_dp_reset_user_features(skb, info);
1289                 }
1290
1291                 goto err_destroy_ports_array;
1292         }
1293
1294         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1295         err = PTR_ERR(reply);
1296         if (IS_ERR(reply))
1297                 goto err_destroy_local_port;
1298
1299         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1300         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1301
1302         ovs_unlock();
1303
1304         ovs_notify(&dp_datapath_genl_family, reply, info);
1305         return 0;
1306
1307 err_destroy_local_port:
1308         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1309 err_destroy_ports_array:
1310         kfree(dp->ports);
1311 err_destroy_percpu:
1312         free_percpu(dp->stats_percpu);
1313 err_destroy_table:
1314         ovs_flow_tbl_destroy(&dp->table, false);
1315 err_free_dp:
1316         release_net(ovs_dp_get_net(dp));
1317         kfree(dp);
1318 err_unlock_ovs:
1319         ovs_unlock();
1320 err:
1321         return err;
1322 }
1323
1324 /* Called with ovs_mutex. */
1325 static void __dp_destroy(struct datapath *dp)
1326 {
1327         int i;
1328
1329         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1330                 struct vport *vport;
1331                 struct hlist_node *n;
1332
1333                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1334                         if (vport->port_no != OVSP_LOCAL)
1335                                 ovs_dp_detach_port(vport);
1336         }
1337
1338         list_del_rcu(&dp->list_node);
1339
1340         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1341          * all ports in datapath are destroyed first before freeing datapath.
1342          */
1343         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1344
1345         /* RCU destroy the flow table */
1346         ovs_flow_tbl_destroy(&dp->table, true);
1347
1348         call_rcu(&dp->rcu, destroy_dp_rcu);
1349 }
1350
1351 static int ovs_dp_cmd_del(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         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_DEL);
1364         err = PTR_ERR(reply);
1365         if (IS_ERR(reply))
1366                 goto unlock;
1367
1368         __dp_destroy(dp);
1369         ovs_unlock();
1370
1371         ovs_notify(&dp_datapath_genl_family, reply, info);
1372
1373         return 0;
1374 unlock:
1375         ovs_unlock();
1376         return err;
1377 }
1378
1379 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1380 {
1381         struct sk_buff *reply;
1382         struct datapath *dp;
1383         int err;
1384
1385         ovs_lock();
1386         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1387         err = PTR_ERR(dp);
1388         if (IS_ERR(dp))
1389                 goto unlock;
1390
1391         ovs_dp_change(dp, info->attrs);
1392
1393         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1394         if (IS_ERR(reply)) {
1395                 err = PTR_ERR(reply);
1396                 genl_set_err(&dp_datapath_genl_family, sock_net(skb->sk), 0,
1397                              0, err);
1398                 err = 0;
1399                 goto unlock;
1400         }
1401
1402         ovs_unlock();
1403         ovs_notify(&dp_datapath_genl_family, reply, info);
1404
1405         return 0;
1406 unlock:
1407         ovs_unlock();
1408         return err;
1409 }
1410
1411 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1412 {
1413         struct sk_buff *reply;
1414         struct datapath *dp;
1415         int err;
1416
1417         ovs_lock();
1418         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1419         if (IS_ERR(dp)) {
1420                 err = PTR_ERR(dp);
1421                 goto unlock;
1422         }
1423
1424         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1425         if (IS_ERR(reply)) {
1426                 err = PTR_ERR(reply);
1427                 goto unlock;
1428         }
1429
1430         ovs_unlock();
1431         return genlmsg_reply(reply, info);
1432
1433 unlock:
1434         ovs_unlock();
1435         return err;
1436 }
1437
1438 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1439 {
1440         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1441         struct datapath *dp;
1442         int skip = cb->args[0];
1443         int i = 0;
1444
1445         rcu_read_lock();
1446         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1447                 if (i >= skip &&
1448                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1449                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1450                                          OVS_DP_CMD_NEW) < 0)
1451                         break;
1452                 i++;
1453         }
1454         rcu_read_unlock();
1455
1456         cb->args[0] = i;
1457
1458         return skb->len;
1459 }
1460
1461 static const struct genl_ops dp_datapath_genl_ops[] = {
1462         { .cmd = OVS_DP_CMD_NEW,
1463           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1464           .policy = datapath_policy,
1465           .doit = ovs_dp_cmd_new
1466         },
1467         { .cmd = OVS_DP_CMD_DEL,
1468           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1469           .policy = datapath_policy,
1470           .doit = ovs_dp_cmd_del
1471         },
1472         { .cmd = OVS_DP_CMD_GET,
1473           .flags = 0,               /* OK for unprivileged users. */
1474           .policy = datapath_policy,
1475           .doit = ovs_dp_cmd_get,
1476           .dumpit = ovs_dp_cmd_dump
1477         },
1478         { .cmd = OVS_DP_CMD_SET,
1479           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1480           .policy = datapath_policy,
1481           .doit = ovs_dp_cmd_set,
1482         },
1483 };
1484
1485 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1486         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1487         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1488         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1489         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1490         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1491         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1492 };
1493
1494 struct genl_family dp_vport_genl_family = {
1495         .id = GENL_ID_GENERATE,
1496         .hdrsize = sizeof(struct ovs_header),
1497         .name = OVS_VPORT_FAMILY,
1498         .version = OVS_VPORT_VERSION,
1499         .maxattr = OVS_VPORT_ATTR_MAX,
1500         .netnsok = true,
1501         .parallel_ops = true,
1502 };
1503
1504 static struct genl_multicast_group ovs_dp_vport_multicast_group = {
1505         .name = OVS_VPORT_MCGROUP
1506 };
1507
1508 /* Called with ovs_mutex or RCU read lock. */
1509 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1510                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1511 {
1512         struct ovs_header *ovs_header;
1513         struct ovs_vport_stats vport_stats;
1514         int err;
1515
1516         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1517                                  flags, cmd);
1518         if (!ovs_header)
1519                 return -EMSGSIZE;
1520
1521         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1522
1523         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1524             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1525             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1526             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1527                 goto nla_put_failure;
1528
1529         ovs_vport_get_stats(vport, &vport_stats);
1530         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1531                     &vport_stats))
1532                 goto nla_put_failure;
1533
1534         err = ovs_vport_get_options(vport, skb);
1535         if (err == -EMSGSIZE)
1536                 goto error;
1537
1538         return genlmsg_end(skb, ovs_header);
1539
1540 nla_put_failure:
1541         err = -EMSGSIZE;
1542 error:
1543         genlmsg_cancel(skb, ovs_header);
1544         return err;
1545 }
1546
1547 /* Called with ovs_mutex or RCU read lock. */
1548 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1549                                          u32 seq, u8 cmd)
1550 {
1551         struct sk_buff *skb;
1552         int retval;
1553
1554         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1555         if (!skb)
1556                 return ERR_PTR(-ENOMEM);
1557
1558         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1559         BUG_ON(retval < 0);
1560
1561         return skb;
1562 }
1563
1564 /* Called with ovs_mutex or RCU read lock. */
1565 static struct vport *lookup_vport(struct net *net,
1566                                   struct ovs_header *ovs_header,
1567                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1568 {
1569         struct datapath *dp;
1570         struct vport *vport;
1571
1572         if (a[OVS_VPORT_ATTR_NAME]) {
1573                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1574                 if (!vport)
1575                         return ERR_PTR(-ENODEV);
1576                 if (ovs_header->dp_ifindex &&
1577                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1578                         return ERR_PTR(-ENODEV);
1579                 return vport;
1580         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1581                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1582
1583                 if (port_no >= DP_MAX_PORTS)
1584                         return ERR_PTR(-EFBIG);
1585
1586                 dp = get_dp(net, ovs_header->dp_ifindex);
1587                 if (!dp)
1588                         return ERR_PTR(-ENODEV);
1589
1590                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1591                 if (!vport)
1592                         return ERR_PTR(-ENODEV);
1593                 return vport;
1594         } else
1595                 return ERR_PTR(-EINVAL);
1596 }
1597
1598 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1599 {
1600         struct nlattr **a = info->attrs;
1601         struct ovs_header *ovs_header = info->userhdr;
1602         struct vport_parms parms;
1603         struct sk_buff *reply;
1604         struct vport *vport;
1605         struct datapath *dp;
1606         u32 port_no;
1607         int err;
1608
1609         err = -EINVAL;
1610         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1611             !a[OVS_VPORT_ATTR_UPCALL_PID])
1612                 goto exit;
1613
1614         ovs_lock();
1615         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1616         err = -ENODEV;
1617         if (!dp)
1618                 goto exit_unlock;
1619
1620         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1621                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1622
1623                 err = -EFBIG;
1624                 if (port_no >= DP_MAX_PORTS)
1625                         goto exit_unlock;
1626
1627                 vport = ovs_vport_ovsl(dp, port_no);
1628                 err = -EBUSY;
1629                 if (vport)
1630                         goto exit_unlock;
1631         } else {
1632                 for (port_no = 1; ; port_no++) {
1633                         if (port_no >= DP_MAX_PORTS) {
1634                                 err = -EFBIG;
1635                                 goto exit_unlock;
1636                         }
1637                         vport = ovs_vport_ovsl(dp, port_no);
1638                         if (!vport)
1639                                 break;
1640                 }
1641         }
1642
1643         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1644         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1645         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1646         parms.dp = dp;
1647         parms.port_no = port_no;
1648         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1649
1650         vport = new_vport(&parms);
1651         err = PTR_ERR(vport);
1652         if (IS_ERR(vport))
1653                 goto exit_unlock;
1654
1655         err = 0;
1656         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1657                                          OVS_VPORT_CMD_NEW);
1658         if (IS_ERR(reply)) {
1659                 err = PTR_ERR(reply);
1660                 ovs_dp_detach_port(vport);
1661                 goto exit_unlock;
1662         }
1663
1664         ovs_notify(&dp_vport_genl_family, reply, info);
1665
1666 exit_unlock:
1667         ovs_unlock();
1668 exit:
1669         return err;
1670 }
1671
1672 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1673 {
1674         struct nlattr **a = info->attrs;
1675         struct sk_buff *reply;
1676         struct vport *vport;
1677         int err;
1678
1679         ovs_lock();
1680         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1681         err = PTR_ERR(vport);
1682         if (IS_ERR(vport))
1683                 goto exit_unlock;
1684
1685         if (a[OVS_VPORT_ATTR_TYPE] &&
1686             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1687                 err = -EINVAL;
1688                 goto exit_unlock;
1689         }
1690
1691         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1692         if (!reply) {
1693                 err = -ENOMEM;
1694                 goto exit_unlock;
1695         }
1696
1697         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1698                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1699                 if (err)
1700                         goto exit_free;
1701         }
1702
1703         if (a[OVS_VPORT_ATTR_UPCALL_PID])
1704                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1705
1706         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1707                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1708         BUG_ON(err < 0);
1709
1710         ovs_unlock();
1711         ovs_notify(&dp_vport_genl_family, reply, info);
1712         return 0;
1713
1714 exit_free:
1715         kfree_skb(reply);
1716 exit_unlock:
1717         ovs_unlock();
1718         return err;
1719 }
1720
1721 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1722 {
1723         struct nlattr **a = info->attrs;
1724         struct sk_buff *reply;
1725         struct vport *vport;
1726         int err;
1727
1728         ovs_lock();
1729         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1730         err = PTR_ERR(vport);
1731         if (IS_ERR(vport))
1732                 goto exit_unlock;
1733
1734         if (vport->port_no == OVSP_LOCAL) {
1735                 err = -EINVAL;
1736                 goto exit_unlock;
1737         }
1738
1739         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1740                                          info->snd_seq, OVS_VPORT_CMD_DEL);
1741         err = PTR_ERR(reply);
1742         if (IS_ERR(reply))
1743                 goto exit_unlock;
1744
1745         err = 0;
1746         ovs_dp_detach_port(vport);
1747
1748         ovs_notify(&dp_vport_genl_family, reply, info);
1749
1750 exit_unlock:
1751         ovs_unlock();
1752         return err;
1753 }
1754
1755 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1756 {
1757         struct nlattr **a = info->attrs;
1758         struct ovs_header *ovs_header = info->userhdr;
1759         struct sk_buff *reply;
1760         struct vport *vport;
1761         int err;
1762
1763         rcu_read_lock();
1764         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1765         err = PTR_ERR(vport);
1766         if (IS_ERR(vport))
1767                 goto exit_unlock;
1768
1769         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1770                                          info->snd_seq, OVS_VPORT_CMD_NEW);
1771         err = PTR_ERR(reply);
1772         if (IS_ERR(reply))
1773                 goto exit_unlock;
1774
1775         rcu_read_unlock();
1776
1777         return genlmsg_reply(reply, info);
1778
1779 exit_unlock:
1780         rcu_read_unlock();
1781         return err;
1782 }
1783
1784 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1785 {
1786         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1787         struct datapath *dp;
1788         int bucket = cb->args[0], skip = cb->args[1];
1789         int i, j = 0;
1790
1791         rcu_read_lock();
1792         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1793         if (!dp) {
1794                 rcu_read_unlock();
1795                 return -ENODEV;
1796         }
1797         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1798                 struct vport *vport;
1799
1800                 j = 0;
1801                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1802                         if (j >= skip &&
1803                             ovs_vport_cmd_fill_info(vport, skb,
1804                                                     NETLINK_CB(cb->skb).portid,
1805                                                     cb->nlh->nlmsg_seq,
1806                                                     NLM_F_MULTI,
1807                                                     OVS_VPORT_CMD_NEW) < 0)
1808                                 goto out;
1809
1810                         j++;
1811                 }
1812                 skip = 0;
1813         }
1814 out:
1815         rcu_read_unlock();
1816
1817         cb->args[0] = i;
1818         cb->args[1] = j;
1819
1820         return skb->len;
1821 }
1822
1823 static const struct genl_ops dp_vport_genl_ops[] = {
1824         { .cmd = OVS_VPORT_CMD_NEW,
1825           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1826           .policy = vport_policy,
1827           .doit = ovs_vport_cmd_new
1828         },
1829         { .cmd = OVS_VPORT_CMD_DEL,
1830           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1831           .policy = vport_policy,
1832           .doit = ovs_vport_cmd_del
1833         },
1834         { .cmd = OVS_VPORT_CMD_GET,
1835           .flags = 0,               /* OK for unprivileged users. */
1836           .policy = vport_policy,
1837           .doit = ovs_vport_cmd_get,
1838           .dumpit = ovs_vport_cmd_dump
1839         },
1840         { .cmd = OVS_VPORT_CMD_SET,
1841           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1842           .policy = vport_policy,
1843           .doit = ovs_vport_cmd_set,
1844         },
1845 };
1846
1847 struct genl_family_and_ops {
1848         struct genl_family *family;
1849         const struct genl_ops *ops;
1850         int n_ops;
1851         const struct genl_multicast_group *group;
1852 };
1853
1854 static const struct genl_family_and_ops dp_genl_families[] = {
1855         { &dp_datapath_genl_family,
1856           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1857           &ovs_dp_datapath_multicast_group },
1858         { &dp_vport_genl_family,
1859           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1860           &ovs_dp_vport_multicast_group },
1861         { &dp_flow_genl_family,
1862           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1863           &ovs_dp_flow_multicast_group },
1864         { &dp_packet_genl_family,
1865           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1866           NULL },
1867 };
1868
1869 static void dp_unregister_genl(int n_families)
1870 {
1871         int i;
1872
1873         for (i = 0; i < n_families; i++)
1874                 genl_unregister_family(dp_genl_families[i].family);
1875 }
1876
1877 static int dp_register_genl(void)
1878 {
1879         int n_registered;
1880         int err;
1881         int i;
1882
1883         n_registered = 0;
1884         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1885                 const struct genl_family_and_ops *f = &dp_genl_families[i];
1886
1887                 f->family->ops = f->ops;
1888                 f->family->n_ops = f->n_ops;
1889                 f->family->mcgrps = f->group;
1890                 f->family->n_mcgrps = f->group ? 1 : 0;
1891                 err = genl_register_family(f->family);
1892                 if (err)
1893                         goto error;
1894                 n_registered++;
1895         }
1896
1897         return 0;
1898
1899 error:
1900         dp_unregister_genl(n_registered);
1901         return err;
1902 }
1903
1904 static int __net_init ovs_init_net(struct net *net)
1905 {
1906         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1907
1908         INIT_LIST_HEAD(&ovs_net->dps);
1909         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1910         return 0;
1911 }
1912
1913 static void __net_exit ovs_exit_net(struct net *net)
1914 {
1915         struct datapath *dp, *dp_next;
1916         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1917
1918         ovs_lock();
1919         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1920                 __dp_destroy(dp);
1921         ovs_unlock();
1922
1923         cancel_work_sync(&ovs_net->dp_notify_work);
1924 }
1925
1926 static struct pernet_operations ovs_net_ops = {
1927         .init = ovs_init_net,
1928         .exit = ovs_exit_net,
1929         .id   = &ovs_net_id,
1930         .size = sizeof(struct ovs_net),
1931 };
1932
1933 static int __init dp_init(void)
1934 {
1935         int err;
1936
1937         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1938
1939         pr_info("Open vSwitch switching datapath\n");
1940
1941         err = ovs_flow_init();
1942         if (err)
1943                 goto error;
1944
1945         err = ovs_vport_init();
1946         if (err)
1947                 goto error_flow_exit;
1948
1949         err = register_pernet_device(&ovs_net_ops);
1950         if (err)
1951                 goto error_vport_exit;
1952
1953         err = register_netdevice_notifier(&ovs_dp_device_notifier);
1954         if (err)
1955                 goto error_netns_exit;
1956
1957         err = dp_register_genl();
1958         if (err < 0)
1959                 goto error_unreg_notifier;
1960
1961         return 0;
1962
1963 error_unreg_notifier:
1964         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1965 error_netns_exit:
1966         unregister_pernet_device(&ovs_net_ops);
1967 error_vport_exit:
1968         ovs_vport_exit();
1969 error_flow_exit:
1970         ovs_flow_exit();
1971 error:
1972         return err;
1973 }
1974
1975 static void dp_cleanup(void)
1976 {
1977         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1978         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1979         unregister_pernet_device(&ovs_net_ops);
1980         rcu_barrier();
1981         ovs_vport_exit();
1982         ovs_flow_exit();
1983 }
1984
1985 module_init(dp_init);
1986 module_exit(dp_cleanup);
1987
1988 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1989 MODULE_LICENSE("GPL");