net/openvswitch/datapath.c

   1 /*
   2  * Copyright (c) 2007-2014 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/openvswitch.h>
  48 #include <linux/rculist.h>
  49 #include <linux/dmi.h>
  50 #include <net/genetlink.h>
  51 #include <net/net_namespace.h>
  52 #include <net/netns/generic.h>
  53
  54 #include "datapath.h"
  55 #include "flow.h"
  56 #include "flow_table.h"
  57 #include "flow_netlink.h"
  58 #include "vport-internal_dev.h"
  59 #include "vport-netdev.h"
  60
  61 int ovs_net_id __read_mostly;
  62 EXPORT_SYMBOL_GPL(ovs_net_id);
  63
  64 static struct genl_family dp_packet_genl_family;
  65 static struct genl_family dp_flow_genl_family;
  66 static struct genl_family dp_datapath_genl_family;
  67
  68 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
  69         .name = OVS_FLOW_MCGROUP,
  70 };
  71
  72 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  73         .name = OVS_DATAPATH_MCGROUP,
  74 };
  75
  76 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
  77         .name = OVS_VPORT_MCGROUP,
  78 };
  79
  80 /* Check if need to build a reply message.
  81  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  82 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  83                             unsigned int group)
  84 {
  85         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  86                genl_has_listeners(family, genl_info_net(info), group);
  87 }
  88
  89 static void ovs_notify(struct genl_family *family,
  90                        struct sk_buff *skb, struct genl_info *info)
  91 {
  92         genl_notify(family, skb, genl_info_net(info), info->snd_portid,
  93                     0, info->nlhdr, GFP_KERNEL);
  94 }
  95
  96 /**
  97  * DOC: Locking:
  98  *
  99  * All writes e.g. Writes to device state (add/remove datapath, port, set
 100  * operations on vports, etc.), Writes to other state (flow table
 101  * modifications, set miscellaneous datapath parameters, etc.) are protected
 102  * by ovs_lock.
 103  *
 104  * Reads are protected by RCU.
 105  *
 106  * There are a few special cases (mostly stats) that have their own
 107  * synchronization but they nest under all of above and don't interact with
 108  * each other.
 109  *
 110  * The RTNL lock nests inside ovs_mutex.
 111  */
 112
 113 static DEFINE_MUTEX(ovs_mutex);
 114
 115 void ovs_lock(void)
 116 {
 117         mutex_lock(&ovs_mutex);
 118 }
 119
 120 void ovs_unlock(void)
 121 {
 122         mutex_unlock(&ovs_mutex);
 123 }
 124
 125 #ifdef CONFIG_LOCKDEP
 126 int lockdep_ovsl_is_held(void)
 127 {
 128         if (debug_locks)
 129                 return lockdep_is_held(&ovs_mutex);
 130         else
 131                 return 1;
 132 }
 133 EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held);
 134 #endif
 135
 136 static struct vport *new_vport(const struct vport_parms *);
 137 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 138                              const struct sw_flow_key *,
 139                              const struct dp_upcall_info *);
 140 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 141                                   const struct sw_flow_key *,
 142                                   const struct dp_upcall_info *);
 143
 144 /* Must be called with rcu_read_lock. */
 145 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
 146 {
 147         struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
 148
 149         if (dev) {
 150                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 151                 if (vport)
 152                         return vport->dp;
 153         }
 154
 155         return NULL;
 156 }
 157
 158 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
 159  * returned dp pointer valid.
 160  */
 161 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
 162 {
 163         struct datapath *dp;
 164
 165         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
 166         rcu_read_lock();
 167         dp = get_dp_rcu(net, dp_ifindex);
 168         rcu_read_unlock();
 169
 170         return dp;
 171 }
 172
 173 /* Must be called with rcu_read_lock or ovs_mutex. */
 174 const char *ovs_dp_name(const struct datapath *dp)
 175 {
 176         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 177         return vport->ops->get_name(vport);
 178 }
 179
 180 static int get_dpifindex(const struct datapath *dp)
 181 {
 182         struct vport *local;
 183         int ifindex;
 184
 185         rcu_read_lock();
 186
 187         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 188         if (local)
 189                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 190         else
 191                 ifindex = 0;
 192
 193         rcu_read_unlock();
 194
 195         return ifindex;
 196 }
 197
 198 static void destroy_dp_rcu(struct rcu_head *rcu)
 199 {
 200         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 201
 202         ovs_flow_tbl_destroy(&dp->table);
 203         free_percpu(dp->stats_percpu);
 204         release_net(ovs_dp_get_net(dp));
 205         kfree(dp->ports);
 206         kfree(dp);
 207 }
 208
 209 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 210                                             u16 port_no)
 211 {
 212         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 213 }
 214
 215 /* Called with ovs_mutex or RCU read lock. */
 216 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 217 {
 218         struct vport *vport;
 219         struct hlist_head *head;
 220
 221         head = vport_hash_bucket(dp, port_no);
 222         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 223                 if (vport->port_no == port_no)
 224                         return vport;
 225         }
 226         return NULL;
 227 }
 228
 229 /* Called with ovs_mutex. */
 230 static struct vport *new_vport(const struct vport_parms *parms)
 231 {
 232         struct vport *vport;
 233
 234         vport = ovs_vport_add(parms);
 235         if (!IS_ERR(vport)) {
 236                 struct datapath *dp = parms->dp;
 237                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 238
 239                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 240         }
 241         return vport;
 242 }
 243
 244 void ovs_dp_detach_port(struct vport *p)
 245 {
 246         ASSERT_OVSL();
 247
 248         /* First drop references to device. */
 249         hlist_del_rcu(&p->dp_hash_node);
 250
 251         /* Then destroy it. */
 252         ovs_vport_del(p);
 253 }
 254
 255 /* Must be called with rcu_read_lock. */
 256 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 257 {
 258         const struct vport *p = OVS_CB(skb)->input_vport;
 259         struct datapath *dp = p->dp;
 260         struct sw_flow *flow;
 261         struct sw_flow_actions *sf_acts;
 262         struct dp_stats_percpu *stats;
 263         u64 *stats_counter;
 264         u32 n_mask_hit;
 265
 266         stats = this_cpu_ptr(dp->stats_percpu);
 267
 268         /* Look up flow. */
 269         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
 270         if (unlikely(!flow)) {
 271                 struct dp_upcall_info upcall;
 272                 int error;
 273
 274                 upcall.cmd = OVS_PACKET_CMD_MISS;
 275                 upcall.userdata = NULL;
 276                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 277                 upcall.egress_tun_info = NULL;
 278                 error = ovs_dp_upcall(dp, skb, key, &upcall);
 279                 if (unlikely(error))
 280                         kfree_skb(skb);
 281                 else
 282                         consume_skb(skb);
 283                 stats_counter = &stats->n_missed;
 284                 goto out;
 285         }
 286
 287         ovs_flow_stats_update(flow, key->tp.flags, skb);
 288         sf_acts = rcu_dereference(flow->sf_acts);
 289         ovs_execute_actions(dp, skb, sf_acts, key);
 290
 291         stats_counter = &stats->n_hit;
 292
 293 out:
 294         /* Update datapath statistics. */
 295         u64_stats_update_begin(&stats->syncp);
 296         (*stats_counter)++;
 297         stats->n_mask_hit += n_mask_hit;
 298         u64_stats_update_end(&stats->syncp);
 299 }
 300
 301 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 302                   const struct sw_flow_key *key,
 303                   const struct dp_upcall_info *upcall_info)
 304 {
 305         struct dp_stats_percpu *stats;
 306         int err;
 307
 308         if (upcall_info->portid == 0) {
 309                 err = -ENOTCONN;
 310                 goto err;
 311         }
 312
 313         if (!skb_is_gso(skb))
 314                 err = queue_userspace_packet(dp, skb, key, upcall_info);
 315         else
 316                 err = queue_gso_packets(dp, skb, key, upcall_info);
 317         if (err)
 318                 goto err;
 319
 320         return 0;
 321
 322 err:
 323         stats = this_cpu_ptr(dp->stats_percpu);
 324
 325         u64_stats_update_begin(&stats->syncp);
 326         stats->n_lost++;
 327         u64_stats_update_end(&stats->syncp);
 328
 329         return err;
 330 }
 331
 332 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 333                              const struct sw_flow_key *key,
 334                              const struct dp_upcall_info *upcall_info)
 335 {
 336         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 337         struct sw_flow_key later_key;
 338         struct sk_buff *segs, *nskb;
 339         struct ovs_skb_cb ovs_cb;
 340         int err;
 341
 342         ovs_cb = *OVS_CB(skb);
 343         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 344         *OVS_CB(skb) = ovs_cb;
 345         if (IS_ERR(segs))
 346                 return PTR_ERR(segs);
 347         if (segs == NULL)
 348                 return -EINVAL;
 349
 350         if (gso_type & SKB_GSO_UDP) {
 351                 /* The initial flow key extracted by ovs_flow_key_extract()
 352                  * in this case is for a first fragment, so we need to
 353                  * properly mark later fragments.
 354                  */
 355                 later_key = *key;
 356                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 357         }
 358
 359         /* Queue all of the segments. */
 360         skb = segs;
 361         do {
 362                 *OVS_CB(skb) = ovs_cb;
 363                 if (gso_type & SKB_GSO_UDP && skb != segs)
 364                         key = &later_key;
 365
 366                 err = queue_userspace_packet(dp, skb, key, upcall_info);
 367                 if (err)
 368                         break;
 369
 370         } while ((skb = skb->next));
 371
 372         /* Free all of the segments. */
 373         skb = segs;
 374         do {
 375                 nskb = skb->next;
 376                 if (err)
 377                         kfree_skb(skb);
 378                 else
 379                         consume_skb(skb);
 380         } while ((skb = nskb));
 381         return err;
 382 }
 383
 384 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 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(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 390
 391         /* OVS_PACKET_ATTR_USERDATA */
 392         if (upcall_info->userdata)
 393                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 394
 395         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 396         if (upcall_info->egress_tun_info)
 397                 size += nla_total_size(ovs_tun_key_attr_size());
 398
 399         return size;
 400 }
 401
 402 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 403                                   const struct sw_flow_key *key,
 404                                   const struct dp_upcall_info *upcall_info)
 405 {
 406         struct ovs_header *upcall;
 407         struct sk_buff *nskb = NULL;
 408         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 409         struct nlattr *nla;
 410         struct genl_info info = {
 411                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 412                 .snd_portid = upcall_info->portid,
 413         };
 414         size_t len;
 415         unsigned int hlen;
 416         int err, dp_ifindex;
 417
 418         dp_ifindex = get_dpifindex(dp);
 419         if (!dp_ifindex)
 420                 return -ENODEV;
 421
 422         if (skb_vlan_tag_present(skb)) {
 423                 nskb = skb_clone(skb, GFP_ATOMIC);
 424                 if (!nskb)
 425                         return -ENOMEM;
 426
 427                 nskb = __vlan_hwaccel_push_inside(nskb);
 428                 if (!nskb)
 429                         return -ENOMEM;
 430
 431                 skb = nskb;
 432         }
 433
 434         if (nla_attr_size(skb->len) > USHRT_MAX) {
 435                 err = -EFBIG;
 436                 goto out;
 437         }
 438
 439         /* Complete checksum if needed */
 440         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 441             (err = skb_checksum_help(skb)))
 442                 goto out;
 443
 444         /* Older versions of OVS user space enforce alignment of the last
 445          * Netlink attribute to NLA_ALIGNTO which would require extensive
 446          * padding logic. Only perform zerocopy if padding is not required.
 447          */
 448         if (dp->user_features & OVS_DP_F_UNALIGNED)
 449                 hlen = skb_zerocopy_headlen(skb);
 450         else
 451                 hlen = skb->len;
 452
 453         len = upcall_msg_size(upcall_info, hlen);
 454         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 455         if (!user_skb) {
 456                 err = -ENOMEM;
 457                 goto out;
 458         }
 459
 460         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 461                              0, upcall_info->cmd);
 462         upcall->dp_ifindex = dp_ifindex;
 463
 464         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 465         err = ovs_nla_put_flow(key, key, user_skb);
 466         BUG_ON(err);
 467         nla_nest_end(user_skb, nla);
 468
 469         if (upcall_info->userdata)
 470                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 471                           nla_len(upcall_info->userdata),
 472                           nla_data(upcall_info->userdata));
 473
 474         if (upcall_info->egress_tun_info) {
 475                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 476                 err = ovs_nla_put_egress_tunnel_key(user_skb,
 477                                                     upcall_info->egress_tun_info);
 478                 BUG_ON(err);
 479                 nla_nest_end(user_skb, nla);
 480         }
 481
 482         /* Only reserve room for attribute header, packet data is added
 483          * in skb_zerocopy() */
 484         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 485                 err = -ENOBUFS;
 486                 goto out;
 487         }
 488         nla->nla_len = nla_attr_size(skb->len);
 489
 490         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 491         if (err)
 492                 goto out;
 493
 494         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 495         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 496                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 497
 498                 if (plen > 0)
 499                         memset(skb_put(user_skb, plen), 0, plen);
 500         }
 501
 502         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 503
 504         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 505         user_skb = NULL;
 506 out:
 507         if (err)
 508                 skb_tx_error(skb);
 509         kfree_skb(user_skb);
 510         kfree_skb(nskb);
 511         return err;
 512 }
 513
 514 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 515 {
 516         struct ovs_header *ovs_header = info->userhdr;
 517         struct nlattr **a = info->attrs;
 518         struct sw_flow_actions *acts;
 519         struct sk_buff *packet;
 520         struct sw_flow *flow;
 521         struct sw_flow_actions *sf_acts;
 522         struct datapath *dp;
 523         struct ethhdr *eth;
 524         struct vport *input_vport;
 525         int len;
 526         int err;
 527         bool log = !a[OVS_PACKET_ATTR_PROBE];
 528
 529         err = -EINVAL;
 530         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 531             !a[OVS_PACKET_ATTR_ACTIONS])
 532                 goto err;
 533
 534         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 535         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 536         err = -ENOMEM;
 537         if (!packet)
 538                 goto err;
 539         skb_reserve(packet, NET_IP_ALIGN);
 540
 541         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 542
 543         skb_reset_mac_header(packet);
 544         eth = eth_hdr(packet);
 545
 546         /* Normally, setting the skb 'protocol' field would be handled by a
 547          * call to eth_type_trans(), but it assumes there's a sending
 548          * device, which we may not have. */
 549         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 550                 packet->protocol = eth->h_proto;
 551         else
 552                 packet->protocol = htons(ETH_P_802_2);
 553
 554         /* Build an sw_flow for sending this packet. */
 555         flow = ovs_flow_alloc();
 556         err = PTR_ERR(flow);
 557         if (IS_ERR(flow))
 558                 goto err_kfree_skb;
 559
 560         err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
 561                                              &flow->key, log);
 562         if (err)
 563                 goto err_flow_free;
 564
 565         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 566                                    &flow->key, &acts, log);
 567         if (err)
 568                 goto err_flow_free;
 569
 570         rcu_assign_pointer(flow->sf_acts, acts);
 571         OVS_CB(packet)->egress_tun_info = NULL;
 572         packet->priority = flow->key.phy.priority;
 573         packet->mark = flow->key.phy.skb_mark;
 574
 575         rcu_read_lock();
 576         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
 577         err = -ENODEV;
 578         if (!dp)
 579                 goto err_unlock;
 580
 581         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 582         if (!input_vport)
 583                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 584
 585         if (!input_vport)
 586                 goto err_unlock;
 587
 588         OVS_CB(packet)->input_vport = input_vport;
 589         sf_acts = rcu_dereference(flow->sf_acts);
 590
 591         local_bh_disable();
 592         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 593         local_bh_enable();
 594         rcu_read_unlock();
 595
 596         ovs_flow_free(flow, false);
 597         return err;
 598
 599 err_unlock:
 600         rcu_read_unlock();
 601 err_flow_free:
 602         ovs_flow_free(flow, false);
 603 err_kfree_skb:
 604         kfree_skb(packet);
 605 err:
 606         return err;
 607 }
 608
 609 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 610         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 611         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 612         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 613         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 614 };
 615
 616 static const struct genl_ops dp_packet_genl_ops[] = {
 617         { .cmd = OVS_PACKET_CMD_EXECUTE,
 618           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 619           .policy = packet_policy,
 620           .doit = ovs_packet_cmd_execute
 621         }
 622 };
 623
 624 static struct genl_family dp_packet_genl_family = {
 625         .id = GENL_ID_GENERATE,
 626         .hdrsize = sizeof(struct ovs_header),
 627         .name = OVS_PACKET_FAMILY,
 628         .version = OVS_PACKET_VERSION,
 629         .maxattr = OVS_PACKET_ATTR_MAX,
 630         .netnsok = true,
 631         .parallel_ops = true,
 632         .ops = dp_packet_genl_ops,
 633         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 634 };
 635
 636 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 637                          struct ovs_dp_megaflow_stats *mega_stats)
 638 {
 639         int i;
 640
 641         memset(mega_stats, 0, sizeof(*mega_stats));
 642
 643         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 644         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 645
 646         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 647
 648         for_each_possible_cpu(i) {
 649                 const struct dp_stats_percpu *percpu_stats;
 650                 struct dp_stats_percpu local_stats;
 651                 unsigned int start;
 652
 653                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 654
 655                 do {
 656                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 657                         local_stats = *percpu_stats;
 658                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 659
 660                 stats->n_hit += local_stats.n_hit;
 661                 stats->n_missed += local_stats.n_missed;
 662                 stats->n_lost += local_stats.n_lost;
 663                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 664         }
 665 }
 666
 667 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 668 {
 669         return NLMSG_ALIGN(sizeof(struct ovs_header))
 670                 + nla_total_size(ovs_key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 671                 + nla_total_size(ovs_key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 672                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 673                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 674                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 675                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 676 }
 677
 678 /* Called with ovs_mutex or RCU read lock. */
 679 static int ovs_flow_cmd_fill_match(const struct sw_flow *flow,
 680                                    struct sk_buff *skb)
 681 {
 682         struct nlattr *nla;
 683         int err;
 684
 685         /* Fill flow key. */
 686         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 687         if (!nla)
 688                 return -EMSGSIZE;
 689
 690         err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
 691         if (err)
 692                 return err;
 693
 694         nla_nest_end(skb, nla);
 695
 696         /* Fill flow mask. */
 697         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 698         if (!nla)
 699                 return -EMSGSIZE;
 700
 701         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 702         if (err)
 703                 return err;
 704
 705         nla_nest_end(skb, nla);
 706         return 0;
 707 }
 708
 709 /* Called with ovs_mutex or RCU read lock. */
 710 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 711                                    struct sk_buff *skb)
 712 {
 713         struct ovs_flow_stats stats;
 714         __be16 tcp_flags;
 715         unsigned long used;
 716
 717         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 718
 719         if (used &&
 720             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 721                 return -EMSGSIZE;
 722
 723         if (stats.n_packets &&
 724             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 725                 return -EMSGSIZE;
 726
 727         if ((u8)ntohs(tcp_flags) &&
 728              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 729                 return -EMSGSIZE;
 730
 731         return 0;
 732 }
 733
 734 /* Called with ovs_mutex or RCU read lock. */
 735 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 736                                      struct sk_buff *skb, int skb_orig_len)
 737 {
 738         struct nlattr *start;
 739         int err;
 740
 741         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 742          * this is the first flow to be dumped into 'skb'.  This is unusual for
 743          * Netlink but individual action lists can be longer than
 744          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 745          * The userspace caller can always fetch the actions separately if it
 746          * really wants them.  (Most userspace callers in fact don't care.)
 747          *
 748          * This can only fail for dump operations because the skb is always
 749          * properly sized for single flows.
 750          */
 751         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 752         if (start) {
 753                 const struct sw_flow_actions *sf_acts;
 754
 755                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 756                 err = ovs_nla_put_actions(sf_acts->actions,
 757                                           sf_acts->actions_len, skb);
 758
 759                 if (!err)
 760                         nla_nest_end(skb, start);
 761                 else {
 762                         if (skb_orig_len)
 763                                 return err;
 764
 765                         nla_nest_cancel(skb, start);
 766                 }
 767         } else if (skb_orig_len) {
 768                 return -EMSGSIZE;
 769         }
 770
 771         return 0;
 772 }
 773
 774 /* Called with ovs_mutex or RCU read lock. */
 775 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 776                                   struct sk_buff *skb, u32 portid,
 777                                   u32 seq, u32 flags, u8 cmd)
 778 {
 779         const int skb_orig_len = skb->len;
 780         struct ovs_header *ovs_header;
 781         int err;
 782
 783         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 784                                  flags, cmd);
 785         if (!ovs_header)
 786                 return -EMSGSIZE;
 787
 788         ovs_header->dp_ifindex = dp_ifindex;
 789
 790         err = ovs_flow_cmd_fill_match(flow, skb);
 791         if (err)
 792                 goto error;
 793
 794         err = ovs_flow_cmd_fill_stats(flow, skb);
 795         if (err)
 796                 goto error;
 797
 798         err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 799         if (err)
 800                 goto error;
 801
 802         genlmsg_end(skb, ovs_header);
 803         return 0;
 804
 805 error:
 806         genlmsg_cancel(skb, ovs_header);
 807         return err;
 808 }
 809
 810 /* May not be called with RCU read lock. */
 811 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 812                                                struct genl_info *info,
 813                                                bool always)
 814 {
 815         struct sk_buff *skb;
 816
 817         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 818                 return NULL;
 819
 820         skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
 821         if (!skb)
 822                 return ERR_PTR(-ENOMEM);
 823
 824         return skb;
 825 }
 826
 827 /* Called with ovs_mutex. */
 828 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 829                                                int dp_ifindex,
 830                                                struct genl_info *info, u8 cmd,
 831                                                bool always)
 832 {
 833         struct sk_buff *skb;
 834         int retval;
 835
 836         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
 837                                       always);
 838         if (IS_ERR_OR_NULL(skb))
 839                 return skb;
 840
 841         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 842                                         info->snd_portid, info->snd_seq, 0,
 843                                         cmd);
 844         BUG_ON(retval < 0);
 845         return skb;
 846 }
 847
 848 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 849 {
 850         struct nlattr **a = info->attrs;
 851         struct ovs_header *ovs_header = info->userhdr;
 852         struct sw_flow *flow, *new_flow;
 853         struct sw_flow_mask mask;
 854         struct sk_buff *reply;
 855         struct datapath *dp;
 856         struct sw_flow_actions *acts;
 857         struct sw_flow_match match;
 858         int error;
 859         bool log = !a[OVS_FLOW_ATTR_PROBE];
 860
 861         /* Must have key and actions. */
 862         error = -EINVAL;
 863         if (!a[OVS_FLOW_ATTR_KEY]) {
 864                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 865                 goto error;
 866         }
 867         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 868                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 869                 goto error;
 870         }
 871
 872         /* Most of the time we need to allocate a new flow, do it before
 873          * locking.
 874          */
 875         new_flow = ovs_flow_alloc();
 876         if (IS_ERR(new_flow)) {
 877                 error = PTR_ERR(new_flow);
 878                 goto error;
 879         }
 880
 881         /* Extract key. */
 882         ovs_match_init(&match, &new_flow->unmasked_key, &mask);
 883         error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
 884                                   a[OVS_FLOW_ATTR_MASK], log);
 885         if (error)
 886                 goto err_kfree_flow;
 887
 888         ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
 889
 890         /* Validate actions. */
 891         error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
 892                                      &acts, log);
 893         if (error) {
 894                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 895                 goto err_kfree_flow;
 896         }
 897
 898         reply = ovs_flow_cmd_alloc_info(acts, info, false);
 899         if (IS_ERR(reply)) {
 900                 error = PTR_ERR(reply);
 901                 goto err_kfree_acts;
 902         }
 903
 904         ovs_lock();
 905         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 906         if (unlikely(!dp)) {
 907                 error = -ENODEV;
 908                 goto err_unlock_ovs;
 909         }
 910         /* Check if this is a duplicate flow */
 911         flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
 912         if (likely(!flow)) {
 913                 rcu_assign_pointer(new_flow->sf_acts, acts);
 914
 915                 /* Put flow in bucket. */
 916                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 917                 if (unlikely(error)) {
 918                         acts = NULL;
 919                         goto err_unlock_ovs;
 920                 }
 921
 922                 if (unlikely(reply)) {
 923                         error = ovs_flow_cmd_fill_info(new_flow,
 924                                                        ovs_header->dp_ifindex,
 925                                                        reply, info->snd_portid,
 926                                                        info->snd_seq, 0,
 927                                                        OVS_FLOW_CMD_NEW);
 928                         BUG_ON(error < 0);
 929                 }
 930                 ovs_unlock();
 931         } else {
 932                 struct sw_flow_actions *old_acts;
 933
 934                 /* Bail out if we're not allowed to modify an existing flow.
 935                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 936                  * because Generic Netlink treats the latter as a dump
 937                  * request.  We also accept NLM_F_EXCL in case that bug ever
 938                  * gets fixed.
 939                  */
 940                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
 941                                                          | NLM_F_EXCL))) {
 942                         error = -EEXIST;
 943                         goto err_unlock_ovs;
 944                 }
 945                 /* The unmasked key has to be the same for flow updates. */
 946                 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
 947                         /* Look for any overlapping flow. */
 948                         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
 949                         if (!flow) {
 950                                 error = -ENOENT;
 951                                 goto err_unlock_ovs;
 952                         }
 953                 }
 954                 /* Update actions. */
 955                 old_acts = ovsl_dereference(flow->sf_acts);
 956                 rcu_assign_pointer(flow->sf_acts, acts);
 957
 958                 if (unlikely(reply)) {
 959                         error = ovs_flow_cmd_fill_info(flow,
 960                                                        ovs_header->dp_ifindex,
 961                                                        reply, info->snd_portid,
 962                                                        info->snd_seq, 0,
 963                                                        OVS_FLOW_CMD_NEW);
 964                         BUG_ON(error < 0);
 965                 }
 966                 ovs_unlock();
 967
 968                 ovs_nla_free_flow_actions(old_acts);
 969                 ovs_flow_free(new_flow, false);
 970         }
 971
 972         if (reply)
 973                 ovs_notify(&dp_flow_genl_family, reply, info);
 974         return 0;
 975
 976 err_unlock_ovs:
 977         ovs_unlock();
 978         kfree_skb(reply);
 979 err_kfree_acts:
 980         kfree(acts);
 981 err_kfree_flow:
 982         ovs_flow_free(new_flow, false);
 983 error:
 984         return error;
 985 }
 986
 987 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
 988 static struct sw_flow_actions *get_flow_actions(const struct nlattr *a,
 989                                                 const struct sw_flow_key *key,
 990                                                 const struct sw_flow_mask *mask,
 991                                                 bool log)
 992 {
 993         struct sw_flow_actions *acts;
 994         struct sw_flow_key masked_key;
 995         int error;
 996
 997         ovs_flow_mask_key(&masked_key, key, mask);
 998         error = ovs_nla_copy_actions(a, &masked_key, &acts, log);
 999         if (error) {
1000                 OVS_NLERR(log,
1001                           "Actions may not be safe on all matching packets");
1002                 return ERR_PTR(error);
1003         }
1004
1005         return acts;
1006 }
1007
1008 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1009 {
1010         struct nlattr **a = info->attrs;
1011         struct ovs_header *ovs_header = info->userhdr;
1012         struct sw_flow_key key;
1013         struct sw_flow *flow;
1014         struct sw_flow_mask mask;
1015         struct sk_buff *reply = NULL;
1016         struct datapath *dp;
1017         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1018         struct sw_flow_match match;
1019         int error;
1020         bool log = !a[OVS_FLOW_ATTR_PROBE];
1021
1022         /* Extract key. */
1023         error = -EINVAL;
1024         if (!a[OVS_FLOW_ATTR_KEY]) {
1025                 OVS_NLERR(log, "Flow key attribute not present in set flow.");
1026                 goto error;
1027         }
1028
1029         ovs_match_init(&match, &key, &mask);
1030         error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
1031                                   a[OVS_FLOW_ATTR_MASK], log);
1032         if (error)
1033                 goto error;
1034
1035         /* Validate actions. */
1036         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1037                 acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask,
1038                                         log);
1039                 if (IS_ERR(acts)) {
1040                         error = PTR_ERR(acts);
1041                         goto error;
1042                 }
1043
1044                 /* Can allocate before locking if have acts. */
1045                 reply = ovs_flow_cmd_alloc_info(acts, info, false);
1046                 if (IS_ERR(reply)) {
1047                         error = PTR_ERR(reply);
1048                         goto err_kfree_acts;
1049                 }
1050         }
1051
1052         ovs_lock();
1053         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1054         if (unlikely(!dp)) {
1055                 error = -ENODEV;
1056                 goto err_unlock_ovs;
1057         }
1058         /* Check that the flow exists. */
1059         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1060         if (unlikely(!flow)) {
1061                 error = -ENOENT;
1062                 goto err_unlock_ovs;
1063         }
1064
1065         /* Update actions, if present. */
1066         if (likely(acts)) {
1067                 old_acts = ovsl_dereference(flow->sf_acts);
1068                 rcu_assign_pointer(flow->sf_acts, acts);
1069
1070                 if (unlikely(reply)) {
1071                         error = ovs_flow_cmd_fill_info(flow,
1072                                                        ovs_header->dp_ifindex,
1073                                                        reply, info->snd_portid,
1074                                                        info->snd_seq, 0,
1075                                                        OVS_FLOW_CMD_NEW);
1076                         BUG_ON(error < 0);
1077                 }
1078         } else {
1079                 /* Could not alloc without acts before locking. */
1080                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1081                                                 info, OVS_FLOW_CMD_NEW, false);
1082                 if (unlikely(IS_ERR(reply))) {
1083                         error = PTR_ERR(reply);
1084                         goto err_unlock_ovs;
1085                 }
1086         }
1087
1088         /* Clear stats. */
1089         if (a[OVS_FLOW_ATTR_CLEAR])
1090                 ovs_flow_stats_clear(flow);
1091         ovs_unlock();
1092
1093         if (reply)
1094                 ovs_notify(&dp_flow_genl_family, reply, info);
1095         if (old_acts)
1096                 ovs_nla_free_flow_actions(old_acts);
1097
1098         return 0;
1099
1100 err_unlock_ovs:
1101         ovs_unlock();
1102         kfree_skb(reply);
1103 err_kfree_acts:
1104         kfree(acts);
1105 error:
1106         return error;
1107 }
1108
1109 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1110 {
1111         struct nlattr **a = info->attrs;
1112         struct ovs_header *ovs_header = info->userhdr;
1113         struct sw_flow_key key;
1114         struct sk_buff *reply;
1115         struct sw_flow *flow;
1116         struct datapath *dp;
1117         struct sw_flow_match match;
1118         int err;
1119         bool log = !a[OVS_FLOW_ATTR_PROBE];
1120
1121         if (!a[OVS_FLOW_ATTR_KEY]) {
1122                 OVS_NLERR(log,
1123                           "Flow get message rejected, Key attribute missing.");
1124                 return -EINVAL;
1125         }
1126
1127         ovs_match_init(&match, &key, NULL);
1128         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL, log);
1129         if (err)
1130                 return err;
1131
1132         ovs_lock();
1133         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1134         if (!dp) {
1135                 err = -ENODEV;
1136                 goto unlock;
1137         }
1138
1139         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1140         if (!flow) {
1141                 err = -ENOENT;
1142                 goto unlock;
1143         }
1144
1145         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1146                                         OVS_FLOW_CMD_NEW, true);
1147         if (IS_ERR(reply)) {
1148                 err = PTR_ERR(reply);
1149                 goto unlock;
1150         }
1151
1152         ovs_unlock();
1153         return genlmsg_reply(reply, info);
1154 unlock:
1155         ovs_unlock();
1156         return err;
1157 }
1158
1159 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1160 {
1161         struct nlattr **a = info->attrs;
1162         struct ovs_header *ovs_header = info->userhdr;
1163         struct sw_flow_key key;
1164         struct sk_buff *reply;
1165         struct sw_flow *flow;
1166         struct datapath *dp;
1167         struct sw_flow_match match;
1168         int err;
1169         bool log = !a[OVS_FLOW_ATTR_PROBE];
1170
1171         if (likely(a[OVS_FLOW_ATTR_KEY])) {
1172                 ovs_match_init(&match, &key, NULL);
1173                 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL,
1174                                         log);
1175                 if (unlikely(err))
1176                         return err;
1177         }
1178
1179         ovs_lock();
1180         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1181         if (unlikely(!dp)) {
1182                 err = -ENODEV;
1183                 goto unlock;
1184         }
1185
1186         if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1187                 err = ovs_flow_tbl_flush(&dp->table);
1188                 goto unlock;
1189         }
1190
1191         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1192         if (unlikely(!flow)) {
1193                 err = -ENOENT;
1194                 goto unlock;
1195         }
1196
1197         ovs_flow_tbl_remove(&dp->table, flow);
1198         ovs_unlock();
1199
1200         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1201                                         info, false);
1202         if (likely(reply)) {
1203                 if (likely(!IS_ERR(reply))) {
1204                         rcu_read_lock();        /*To keep RCU checker happy. */
1205                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1206                                                      reply, info->snd_portid,
1207                                                      info->snd_seq, 0,
1208                                                      OVS_FLOW_CMD_DEL);
1209                         rcu_read_unlock();
1210                         BUG_ON(err < 0);
1211
1212                         ovs_notify(&dp_flow_genl_family, reply, info);
1213                 } else {
1214                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1215                 }
1216         }
1217
1218         ovs_flow_free(flow, true);
1219         return 0;
1220 unlock:
1221         ovs_unlock();
1222         return err;
1223 }
1224
1225 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1226 {
1227         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1228         struct table_instance *ti;
1229         struct datapath *dp;
1230
1231         rcu_read_lock();
1232         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1233         if (!dp) {
1234                 rcu_read_unlock();
1235                 return -ENODEV;
1236         }
1237
1238         ti = rcu_dereference(dp->table.ti);
1239         for (;;) {
1240                 struct sw_flow *flow;
1241                 u32 bucket, obj;
1242
1243                 bucket = cb->args[0];
1244                 obj = cb->args[1];
1245                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1246                 if (!flow)
1247                         break;
1248
1249                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1250                                            NETLINK_CB(cb->skb).portid,
1251                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1252                                            OVS_FLOW_CMD_NEW) < 0)
1253                         break;
1254
1255                 cb->args[0] = bucket;
1256                 cb->args[1] = obj;
1257         }
1258         rcu_read_unlock();
1259         return skb->len;
1260 }
1261
1262 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1263         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1264         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1265         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1266         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1267         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1268 };
1269
1270 static const struct genl_ops dp_flow_genl_ops[] = {
1271         { .cmd = OVS_FLOW_CMD_NEW,
1272           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1273           .policy = flow_policy,
1274           .doit = ovs_flow_cmd_new
1275         },
1276         { .cmd = OVS_FLOW_CMD_DEL,
1277           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1278           .policy = flow_policy,
1279           .doit = ovs_flow_cmd_del
1280         },
1281         { .cmd = OVS_FLOW_CMD_GET,
1282           .flags = 0,               /* OK for unprivileged users. */
1283           .policy = flow_policy,
1284           .doit = ovs_flow_cmd_get,
1285           .dumpit = ovs_flow_cmd_dump
1286         },
1287         { .cmd = OVS_FLOW_CMD_SET,
1288           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1289           .policy = flow_policy,
1290           .doit = ovs_flow_cmd_set,
1291         },
1292 };
1293
1294 static struct genl_family dp_flow_genl_family = {
1295         .id = GENL_ID_GENERATE,
1296         .hdrsize = sizeof(struct ovs_header),
1297         .name = OVS_FLOW_FAMILY,
1298         .version = OVS_FLOW_VERSION,
1299         .maxattr = OVS_FLOW_ATTR_MAX,
1300         .netnsok = true,
1301         .parallel_ops = true,
1302         .ops = dp_flow_genl_ops,
1303         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1304         .mcgrps = &ovs_dp_flow_multicast_group,
1305         .n_mcgrps = 1,
1306 };
1307
1308 static size_t ovs_dp_cmd_msg_size(void)
1309 {
1310         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1311
1312         msgsize += nla_total_size(IFNAMSIZ);
1313         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1314         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1315         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1316
1317         return msgsize;
1318 }
1319
1320 /* Called with ovs_mutex. */
1321 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1322                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1323 {
1324         struct ovs_header *ovs_header;
1325         struct ovs_dp_stats dp_stats;
1326         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1327         int err;
1328
1329         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1330                                    flags, cmd);
1331         if (!ovs_header)
1332                 goto error;
1333
1334         ovs_header->dp_ifindex = get_dpifindex(dp);
1335
1336         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1337         if (err)
1338                 goto nla_put_failure;
1339
1340         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1341         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1342                         &dp_stats))
1343                 goto nla_put_failure;
1344
1345         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1346                         sizeof(struct ovs_dp_megaflow_stats),
1347                         &dp_megaflow_stats))
1348                 goto nla_put_failure;
1349
1350         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1351                 goto nla_put_failure;
1352
1353         genlmsg_end(skb, ovs_header);
1354         return 0;
1355
1356 nla_put_failure:
1357         genlmsg_cancel(skb, ovs_header);
1358 error:
1359         return -EMSGSIZE;
1360 }
1361
1362 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1363 {
1364         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1365 }
1366
1367 /* Called with rcu_read_lock or ovs_mutex. */
1368 static struct datapath *lookup_datapath(struct net *net,
1369                                         const struct ovs_header *ovs_header,
1370                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1371 {
1372         struct datapath *dp;
1373
1374         if (!a[OVS_DP_ATTR_NAME])
1375                 dp = get_dp(net, ovs_header->dp_ifindex);
1376         else {
1377                 struct vport *vport;
1378
1379                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1380                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1381         }
1382         return dp ? dp : ERR_PTR(-ENODEV);
1383 }
1384
1385 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1386 {
1387         struct datapath *dp;
1388
1389         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1390         if (IS_ERR(dp))
1391                 return;
1392
1393         WARN(dp->user_features, "Dropping previously announced user features\n");
1394         dp->user_features = 0;
1395 }
1396
1397 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1398 {
1399         if (a[OVS_DP_ATTR_USER_FEATURES])
1400                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1401 }
1402
1403 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1404 {
1405         struct nlattr **a = info->attrs;
1406         struct vport_parms parms;
1407         struct sk_buff *reply;
1408         struct datapath *dp;
1409         struct vport *vport;
1410         struct ovs_net *ovs_net;
1411         int err, i;
1412
1413         err = -EINVAL;
1414         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1415                 goto err;
1416
1417         reply = ovs_dp_cmd_alloc_info(info);
1418         if (!reply)
1419                 return -ENOMEM;
1420
1421         err = -ENOMEM;
1422         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1423         if (dp == NULL)
1424                 goto err_free_reply;
1425
1426         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1427
1428         /* Allocate table. */
1429         err = ovs_flow_tbl_init(&dp->table);
1430         if (err)
1431                 goto err_free_dp;
1432
1433         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1434         if (!dp->stats_percpu) {
1435                 err = -ENOMEM;
1436                 goto err_destroy_table;
1437         }
1438
1439         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1440                             GFP_KERNEL);
1441         if (!dp->ports) {
1442                 err = -ENOMEM;
1443                 goto err_destroy_percpu;
1444         }
1445
1446         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1447                 INIT_HLIST_HEAD(&dp->ports[i]);
1448
1449         /* Set up our datapath device. */
1450         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1451         parms.type = OVS_VPORT_TYPE_INTERNAL;
1452         parms.options = NULL;
1453         parms.dp = dp;
1454         parms.port_no = OVSP_LOCAL;
1455         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1456
1457         ovs_dp_change(dp, a);
1458
1459         /* So far only local changes have been made, now need the lock. */
1460         ovs_lock();
1461
1462         vport = new_vport(&parms);
1463         if (IS_ERR(vport)) {
1464                 err = PTR_ERR(vport);
1465                 if (err == -EBUSY)
1466                         err = -EEXIST;
1467
1468                 if (err == -EEXIST) {
1469                         /* An outdated user space instance that does not understand
1470                          * the concept of user_features has attempted to create a new
1471                          * datapath and is likely to reuse it. Drop all user features.
1472                          */
1473                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1474                                 ovs_dp_reset_user_features(skb, info);
1475                 }
1476
1477                 goto err_destroy_ports_array;
1478         }
1479
1480         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1481                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1482         BUG_ON(err < 0);
1483
1484         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1485         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1486
1487         ovs_unlock();
1488
1489         ovs_notify(&dp_datapath_genl_family, reply, info);
1490         return 0;
1491
1492 err_destroy_ports_array:
1493         ovs_unlock();
1494         kfree(dp->ports);
1495 err_destroy_percpu:
1496         free_percpu(dp->stats_percpu);
1497 err_destroy_table:
1498         ovs_flow_tbl_destroy(&dp->table);
1499 err_free_dp:
1500         release_net(ovs_dp_get_net(dp));
1501         kfree(dp);
1502 err_free_reply:
1503         kfree_skb(reply);
1504 err:
1505         return err;
1506 }
1507
1508 /* Called with ovs_mutex. */
1509 static void __dp_destroy(struct datapath *dp)
1510 {
1511         int i;
1512
1513         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1514                 struct vport *vport;
1515                 struct hlist_node *n;
1516
1517                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1518                         if (vport->port_no != OVSP_LOCAL)
1519                                 ovs_dp_detach_port(vport);
1520         }
1521
1522         list_del_rcu(&dp->list_node);
1523
1524         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1525          * all ports in datapath are destroyed first before freeing datapath.
1526          */
1527         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1528
1529         /* RCU destroy the flow table */
1530         call_rcu(&dp->rcu, destroy_dp_rcu);
1531 }
1532
1533 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1534 {
1535         struct sk_buff *reply;
1536         struct datapath *dp;
1537         int err;
1538
1539         reply = ovs_dp_cmd_alloc_info(info);
1540         if (!reply)
1541                 return -ENOMEM;
1542
1543         ovs_lock();
1544         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1545         err = PTR_ERR(dp);
1546         if (IS_ERR(dp))
1547                 goto err_unlock_free;
1548
1549         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1550                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1551         BUG_ON(err < 0);
1552
1553         __dp_destroy(dp);
1554         ovs_unlock();
1555
1556         ovs_notify(&dp_datapath_genl_family, reply, info);
1557
1558         return 0;
1559
1560 err_unlock_free:
1561         ovs_unlock();
1562         kfree_skb(reply);
1563         return err;
1564 }
1565
1566 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1567 {
1568         struct sk_buff *reply;
1569         struct datapath *dp;
1570         int err;
1571
1572         reply = ovs_dp_cmd_alloc_info(info);
1573         if (!reply)
1574                 return -ENOMEM;
1575
1576         ovs_lock();
1577         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1578         err = PTR_ERR(dp);
1579         if (IS_ERR(dp))
1580                 goto err_unlock_free;
1581
1582         ovs_dp_change(dp, info->attrs);
1583
1584         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1585                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1586         BUG_ON(err < 0);
1587
1588         ovs_unlock();
1589         ovs_notify(&dp_datapath_genl_family, reply, info);
1590
1591         return 0;
1592
1593 err_unlock_free:
1594         ovs_unlock();
1595         kfree_skb(reply);
1596         return err;
1597 }
1598
1599 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1600 {
1601         struct sk_buff *reply;
1602         struct datapath *dp;
1603         int err;
1604
1605         reply = ovs_dp_cmd_alloc_info(info);
1606         if (!reply)
1607                 return -ENOMEM;
1608
1609         ovs_lock();
1610         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1611         if (IS_ERR(dp)) {
1612                 err = PTR_ERR(dp);
1613                 goto err_unlock_free;
1614         }
1615         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1616                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1617         BUG_ON(err < 0);
1618         ovs_unlock();
1619
1620         return genlmsg_reply(reply, info);
1621
1622 err_unlock_free:
1623         ovs_unlock();
1624         kfree_skb(reply);
1625         return err;
1626 }
1627
1628 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1629 {
1630         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1631         struct datapath *dp;
1632         int skip = cb->args[0];
1633         int i = 0;
1634
1635         ovs_lock();
1636         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1637                 if (i >= skip &&
1638                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1639                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1640                                          OVS_DP_CMD_NEW) < 0)
1641                         break;
1642                 i++;
1643         }
1644         ovs_unlock();
1645
1646         cb->args[0] = i;
1647
1648         return skb->len;
1649 }
1650
1651 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1652         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1653         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1654         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1655 };
1656
1657 static const struct genl_ops dp_datapath_genl_ops[] = {
1658         { .cmd = OVS_DP_CMD_NEW,
1659           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1660           .policy = datapath_policy,
1661           .doit = ovs_dp_cmd_new
1662         },
1663         { .cmd = OVS_DP_CMD_DEL,
1664           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1665           .policy = datapath_policy,
1666           .doit = ovs_dp_cmd_del
1667         },
1668         { .cmd = OVS_DP_CMD_GET,
1669           .flags = 0,               /* OK for unprivileged users. */
1670           .policy = datapath_policy,
1671           .doit = ovs_dp_cmd_get,
1672           .dumpit = ovs_dp_cmd_dump
1673         },
1674         { .cmd = OVS_DP_CMD_SET,
1675           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1676           .policy = datapath_policy,
1677           .doit = ovs_dp_cmd_set,
1678         },
1679 };
1680
1681 static struct genl_family dp_datapath_genl_family = {
1682         .id = GENL_ID_GENERATE,
1683         .hdrsize = sizeof(struct ovs_header),
1684         .name = OVS_DATAPATH_FAMILY,
1685         .version = OVS_DATAPATH_VERSION,
1686         .maxattr = OVS_DP_ATTR_MAX,
1687         .netnsok = true,
1688         .parallel_ops = true,
1689         .ops = dp_datapath_genl_ops,
1690         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1691         .mcgrps = &ovs_dp_datapath_multicast_group,
1692         .n_mcgrps = 1,
1693 };
1694
1695 /* Called with ovs_mutex or RCU read lock. */
1696 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1697                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1698 {
1699         struct ovs_header *ovs_header;
1700         struct ovs_vport_stats vport_stats;
1701         int err;
1702
1703         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1704                                  flags, cmd);
1705         if (!ovs_header)
1706                 return -EMSGSIZE;
1707
1708         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1709
1710         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1711             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1712             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1713                            vport->ops->get_name(vport)))
1714                 goto nla_put_failure;
1715
1716         ovs_vport_get_stats(vport, &vport_stats);
1717         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1718                     &vport_stats))
1719                 goto nla_put_failure;
1720
1721         if (ovs_vport_get_upcall_portids(vport, skb))
1722                 goto nla_put_failure;
1723
1724         err = ovs_vport_get_options(vport, skb);
1725         if (err == -EMSGSIZE)
1726                 goto error;
1727
1728         genlmsg_end(skb, ovs_header);
1729         return 0;
1730
1731 nla_put_failure:
1732         err = -EMSGSIZE;
1733 error:
1734         genlmsg_cancel(skb, ovs_header);
1735         return err;
1736 }
1737
1738 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1739 {
1740         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1741 }
1742
1743 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1744 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1745                                          u32 seq, u8 cmd)
1746 {
1747         struct sk_buff *skb;
1748         int retval;
1749
1750         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1751         if (!skb)
1752                 return ERR_PTR(-ENOMEM);
1753
1754         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1755         BUG_ON(retval < 0);
1756
1757         return skb;
1758 }
1759
1760 /* Called with ovs_mutex or RCU read lock. */
1761 static struct vport *lookup_vport(struct net *net,
1762                                   const struct ovs_header *ovs_header,
1763                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1764 {
1765         struct datapath *dp;
1766         struct vport *vport;
1767
1768         if (a[OVS_VPORT_ATTR_NAME]) {
1769                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1770                 if (!vport)
1771                         return ERR_PTR(-ENODEV);
1772                 if (ovs_header->dp_ifindex &&
1773                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1774                         return ERR_PTR(-ENODEV);
1775                 return vport;
1776         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1777                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1778
1779                 if (port_no >= DP_MAX_PORTS)
1780                         return ERR_PTR(-EFBIG);
1781
1782                 dp = get_dp(net, ovs_header->dp_ifindex);
1783                 if (!dp)
1784                         return ERR_PTR(-ENODEV);
1785
1786                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1787                 if (!vport)
1788                         return ERR_PTR(-ENODEV);
1789                 return vport;
1790         } else
1791                 return ERR_PTR(-EINVAL);
1792 }
1793
1794 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1795 {
1796         struct nlattr **a = info->attrs;
1797         struct ovs_header *ovs_header = info->userhdr;
1798         struct vport_parms parms;
1799         struct sk_buff *reply;
1800         struct vport *vport;
1801         struct datapath *dp;
1802         u32 port_no;
1803         int err;
1804
1805         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1806             !a[OVS_VPORT_ATTR_UPCALL_PID])
1807                 return -EINVAL;
1808
1809         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1810                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1811         if (port_no >= DP_MAX_PORTS)
1812                 return -EFBIG;
1813
1814         reply = ovs_vport_cmd_alloc_info();
1815         if (!reply)
1816                 return -ENOMEM;
1817
1818         ovs_lock();
1819 restart:
1820         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1821         err = -ENODEV;
1822         if (!dp)
1823                 goto exit_unlock_free;
1824
1825         if (port_no) {
1826                 vport = ovs_vport_ovsl(dp, port_no);
1827                 err = -EBUSY;
1828                 if (vport)
1829                         goto exit_unlock_free;
1830         } else {
1831                 for (port_no = 1; ; port_no++) {
1832                         if (port_no >= DP_MAX_PORTS) {
1833                                 err = -EFBIG;
1834                                 goto exit_unlock_free;
1835                         }
1836                         vport = ovs_vport_ovsl(dp, port_no);
1837                         if (!vport)
1838                                 break;
1839                 }
1840         }
1841
1842         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1843         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1844         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1845         parms.dp = dp;
1846         parms.port_no = port_no;
1847         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1848
1849         vport = new_vport(&parms);
1850         err = PTR_ERR(vport);
1851         if (IS_ERR(vport)) {
1852                 if (err == -EAGAIN)
1853                         goto restart;
1854                 goto exit_unlock_free;
1855         }
1856
1857         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1858                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1859         BUG_ON(err < 0);
1860         ovs_unlock();
1861
1862         ovs_notify(&dp_vport_genl_family, reply, info);
1863         return 0;
1864
1865 exit_unlock_free:
1866         ovs_unlock();
1867         kfree_skb(reply);
1868         return err;
1869 }
1870
1871 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1872 {
1873         struct nlattr **a = info->attrs;
1874         struct sk_buff *reply;
1875         struct vport *vport;
1876         int err;
1877
1878         reply = ovs_vport_cmd_alloc_info();
1879         if (!reply)
1880                 return -ENOMEM;
1881
1882         ovs_lock();
1883         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1884         err = PTR_ERR(vport);
1885         if (IS_ERR(vport))
1886                 goto exit_unlock_free;
1887
1888         if (a[OVS_VPORT_ATTR_TYPE] &&
1889             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1890                 err = -EINVAL;
1891                 goto exit_unlock_free;
1892         }
1893
1894         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1895                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1896                 if (err)
1897                         goto exit_unlock_free;
1898         }
1899
1900
1901         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1902                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
1903
1904                 err = ovs_vport_set_upcall_portids(vport, ids);
1905                 if (err)
1906                         goto exit_unlock_free;
1907         }
1908
1909         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1910                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1911         BUG_ON(err < 0);
1912
1913         ovs_unlock();
1914         ovs_notify(&dp_vport_genl_family, reply, info);
1915         return 0;
1916
1917 exit_unlock_free:
1918         ovs_unlock();
1919         kfree_skb(reply);
1920         return err;
1921 }
1922
1923 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1924 {
1925         struct nlattr **a = info->attrs;
1926         struct sk_buff *reply;
1927         struct vport *vport;
1928         int err;
1929
1930         reply = ovs_vport_cmd_alloc_info();
1931         if (!reply)
1932                 return -ENOMEM;
1933
1934         ovs_lock();
1935         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1936         err = PTR_ERR(vport);
1937         if (IS_ERR(vport))
1938                 goto exit_unlock_free;
1939
1940         if (vport->port_no == OVSP_LOCAL) {
1941                 err = -EINVAL;
1942                 goto exit_unlock_free;
1943         }
1944
1945         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1946                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1947         BUG_ON(err < 0);
1948         ovs_dp_detach_port(vport);
1949         ovs_unlock();
1950
1951         ovs_notify(&dp_vport_genl_family, reply, info);
1952         return 0;
1953
1954 exit_unlock_free:
1955         ovs_unlock();
1956         kfree_skb(reply);
1957         return err;
1958 }
1959
1960 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1961 {
1962         struct nlattr **a = info->attrs;
1963         struct ovs_header *ovs_header = info->userhdr;
1964         struct sk_buff *reply;
1965         struct vport *vport;
1966         int err;
1967
1968         reply = ovs_vport_cmd_alloc_info();
1969         if (!reply)
1970                 return -ENOMEM;
1971
1972         rcu_read_lock();
1973         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1974         err = PTR_ERR(vport);
1975         if (IS_ERR(vport))
1976                 goto exit_unlock_free;
1977         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1978                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1979         BUG_ON(err < 0);
1980         rcu_read_unlock();
1981
1982         return genlmsg_reply(reply, info);
1983
1984 exit_unlock_free:
1985         rcu_read_unlock();
1986         kfree_skb(reply);
1987         return err;
1988 }
1989
1990 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1991 {
1992         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1993         struct datapath *dp;
1994         int bucket = cb->args[0], skip = cb->args[1];
1995         int i, j = 0;
1996
1997         rcu_read_lock();
1998         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1999         if (!dp) {
2000                 rcu_read_unlock();
2001                 return -ENODEV;
2002         }
2003         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2004                 struct vport *vport;
2005
2006                 j = 0;
2007                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2008                         if (j >= skip &&
2009                             ovs_vport_cmd_fill_info(vport, skb,
2010                                                     NETLINK_CB(cb->skb).portid,
2011                                                     cb->nlh->nlmsg_seq,
2012                                                     NLM_F_MULTI,
2013                                                     OVS_VPORT_CMD_NEW) < 0)
2014                                 goto out;
2015
2016                         j++;
2017                 }
2018                 skip = 0;
2019         }
2020 out:
2021         rcu_read_unlock();
2022
2023         cb->args[0] = i;
2024         cb->args[1] = j;
2025
2026         return skb->len;
2027 }
2028
2029 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2030         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2031         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2032         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2033         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2034         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2035         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2036 };
2037
2038 static const struct genl_ops dp_vport_genl_ops[] = {
2039         { .cmd = OVS_VPORT_CMD_NEW,
2040           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2041           .policy = vport_policy,
2042           .doit = ovs_vport_cmd_new
2043         },
2044         { .cmd = OVS_VPORT_CMD_DEL,
2045           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2046           .policy = vport_policy,
2047           .doit = ovs_vport_cmd_del
2048         },
2049         { .cmd = OVS_VPORT_CMD_GET,
2050           .flags = 0,               /* OK for unprivileged users. */
2051           .policy = vport_policy,
2052           .doit = ovs_vport_cmd_get,
2053           .dumpit = ovs_vport_cmd_dump
2054         },
2055         { .cmd = OVS_VPORT_CMD_SET,
2056           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2057           .policy = vport_policy,
2058           .doit = ovs_vport_cmd_set,
2059         },
2060 };
2061
2062 struct genl_family dp_vport_genl_family = {
2063         .id = GENL_ID_GENERATE,
2064         .hdrsize = sizeof(struct ovs_header),
2065         .name = OVS_VPORT_FAMILY,
2066         .version = OVS_VPORT_VERSION,
2067         .maxattr = OVS_VPORT_ATTR_MAX,
2068         .netnsok = true,
2069         .parallel_ops = true,
2070         .ops = dp_vport_genl_ops,
2071         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2072         .mcgrps = &ovs_dp_vport_multicast_group,
2073         .n_mcgrps = 1,
2074 };
2075
2076 static struct genl_family * const dp_genl_families[] = {
2077         &dp_datapath_genl_family,
2078         &dp_vport_genl_family,
2079         &dp_flow_genl_family,
2080         &dp_packet_genl_family,
2081 };
2082
2083 static void dp_unregister_genl(int n_families)
2084 {
2085         int i;
2086
2087         for (i = 0; i < n_families; i++)
2088                 genl_unregister_family(dp_genl_families[i]);
2089 }
2090
2091 static int dp_register_genl(void)
2092 {
2093         int err;
2094         int i;
2095
2096         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2097
2098                 err = genl_register_family(dp_genl_families[i]);
2099                 if (err)
2100                         goto error;
2101         }
2102
2103         return 0;
2104
2105 error:
2106         dp_unregister_genl(i);
2107         return err;
2108 }
2109
2110 static int __net_init ovs_init_net(struct net *net)
2111 {
2112         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2113
2114         INIT_LIST_HEAD(&ovs_net->dps);
2115         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2116         return 0;
2117 }
2118
2119 static void __net_exit ovs_exit_net(struct net *net)
2120 {
2121         struct datapath *dp, *dp_next;
2122         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2123
2124         ovs_lock();
2125         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2126                 __dp_destroy(dp);
2127         ovs_unlock();
2128
2129         cancel_work_sync(&ovs_net->dp_notify_work);
2130 }
2131
2132 static struct pernet_operations ovs_net_ops = {
2133         .init = ovs_init_net,
2134         .exit = ovs_exit_net,
2135         .id   = &ovs_net_id,
2136         .size = sizeof(struct ovs_net),
2137 };
2138
2139 static int __init dp_init(void)
2140 {
2141         int err;
2142
2143         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2144
2145         pr_info("Open vSwitch switching datapath\n");
2146
2147         err = action_fifos_init();
2148         if (err)
2149                 goto error;
2150
2151         err = ovs_internal_dev_rtnl_link_register();
2152         if (err)
2153                 goto error_action_fifos_exit;
2154
2155         err = ovs_flow_init();
2156         if (err)
2157                 goto error_unreg_rtnl_link;
2158
2159         err = ovs_vport_init();
2160         if (err)
2161                 goto error_flow_exit;
2162
2163         err = register_pernet_device(&ovs_net_ops);
2164         if (err)
2165                 goto error_vport_exit;
2166
2167         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2168         if (err)
2169                 goto error_netns_exit;
2170
2171         err = ovs_netdev_init();
2172         if (err)
2173                 goto error_unreg_notifier;
2174
2175         err = dp_register_genl();
2176         if (err < 0)
2177                 goto error_unreg_netdev;
2178
2179         return 0;
2180
2181 error_unreg_netdev:
2182         ovs_netdev_exit();
2183 error_unreg_notifier:
2184         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2185 error_netns_exit:
2186         unregister_pernet_device(&ovs_net_ops);
2187 error_vport_exit:
2188         ovs_vport_exit();
2189 error_flow_exit:
2190         ovs_flow_exit();
2191 error_unreg_rtnl_link:
2192         ovs_internal_dev_rtnl_link_unregister();
2193 error_action_fifos_exit:
2194         action_fifos_exit();
2195 error:
2196         return err;
2197 }
2198
2199 static void dp_cleanup(void)
2200 {
2201         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2202         ovs_netdev_exit();
2203         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2204         unregister_pernet_device(&ovs_net_ops);
2205         rcu_barrier();
2206         ovs_vport_exit();
2207         ovs_flow_exit();
2208         ovs_internal_dev_rtnl_link_unregister();
2209         action_fifos_exit();
2210 }
2211
2212 module_init(dp_init);
2213 module_exit(dp_cleanup);
2214
2215 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2216 MODULE_LICENSE("GPL");