datapath/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/version.h>
  40 #include <linux/ethtool.h>
  41 #include <linux/wait.h>
  42 #include <asm/div64.h>
  43 #include <linux/highmem.h>
  44 #include <linux/netfilter_bridge.h>
  45 #include <linux/netfilter_ipv4.h>
  46 #include <linux/inetdevice.h>
  47 #include <linux/list.h>
  48 #include <linux/openvswitch.h>
  49 #include <linux/rculist.h>
  50 #include <linux/dmi.h>
  51 #include <net/genetlink.h>
  52 #include <net/net_namespace.h>
  53 #include <net/netns/generic.h>
  54
  55 #include "datapath.h"
  56 #include "flow.h"
  57 #include "flow_table.h"
  58 #include "flow_netlink.h"
  59 #include "vlan.h"
  60 #include "vport-internal_dev.h"
  61 #include "vport-netdev.h"
  62
  63 int ovs_net_id __read_mostly;
  64
  65 static struct genl_family dp_packet_genl_family;
  66 static struct genl_family dp_flow_genl_family;
  67 static struct genl_family dp_datapath_genl_family;
  68
  69 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
  70         .name = OVS_FLOW_MCGROUP
  71 };
  72
  73 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  74         .name = OVS_DATAPATH_MCGROUP
  75 };
  76
  77 struct genl_multicast_group ovs_dp_vport_multicast_group = {
  78         .name = OVS_VPORT_MCGROUP
  79 };
  80
  81 /* Check if need to build a reply message.
  82  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  83 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  84                             unsigned int group)
  85 {
  86         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  87                genl_has_listeners(family, genl_info_net(info)->genl_sock,
  88                                   group);
  89 }
  90
  91 static void ovs_notify(struct genl_family *family, struct genl_multicast_group *grp,
  92                        struct sk_buff *skb, struct genl_info *info)
  93 {
  94         genl_notify(family, skb, genl_info_net(info),
  95                     info->snd_portid, GROUP_ID(grp), info->nlhdr, GFP_KERNEL);
  96 }
  97
  98 /**
  99  * DOC: Locking:
 100  *
 101  * All writes e.g. Writes to device state (add/remove datapath, port, set
 102  * operations on vports, etc.), Writes to other state (flow table
 103  * modifications, set miscellaneous datapath parameters, etc.) are protected
 104  * by ovs_lock.
 105  *
 106  * Reads are protected by RCU.
 107  *
 108  * There are a few special cases (mostly stats) that have their own
 109  * synchronization but they nest under all of above and don't interact with
 110  * each other.
 111  *
 112  * The RTNL lock nests inside ovs_mutex.
 113  */
 114
 115 static DEFINE_MUTEX(ovs_mutex);
 116
 117 void ovs_lock(void)
 118 {
 119         mutex_lock(&ovs_mutex);
 120 }
 121
 122 void ovs_unlock(void)
 123 {
 124         mutex_unlock(&ovs_mutex);
 125 }
 126
 127 #ifdef CONFIG_LOCKDEP
 128 int lockdep_ovsl_is_held(void)
 129 {
 130         if (debug_locks)
 131                 return lockdep_is_held(&ovs_mutex);
 132         else
 133                 return 1;
 134 }
 135 #endif
 136
 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 *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 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
 161 {
 162         struct datapath *dp;
 163
 164         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
 165         rcu_read_lock();
 166         dp = get_dp_rcu(net, dp_ifindex);
 167         rcu_read_unlock();
 168
 169         return dp;
 170 }
 171
 172 /* Must be called with rcu_read_lock or ovs_mutex. */
 173 const char *ovs_dp_name(const struct datapath *dp)
 174 {
 175         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 176         return vport->ops->get_name(vport);
 177 }
 178
 179 static int get_dpifindex(const struct datapath *dp)
 180 {
 181         struct vport *local;
 182         int ifindex;
 183
 184         rcu_read_lock();
 185
 186         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 187         if (local)
 188                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 189         else
 190                 ifindex = 0;
 191
 192         rcu_read_unlock();
 193
 194         return ifindex;
 195 }
 196
 197 static void destroy_dp_rcu(struct rcu_head *rcu)
 198 {
 199         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 200
 201         ovs_flow_tbl_destroy(&dp->table);
 202         free_percpu(dp->stats_percpu);
 203         release_net(ovs_dp_get_net(dp));
 204         kfree(dp->ports);
 205         kfree(dp);
 206 }
 207
 208 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 209                                             u16 port_no)
 210 {
 211         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 212 }
 213
 214 /* Called with ovs_mutex or RCU read lock. */
 215 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 216 {
 217         struct vport *vport;
 218         struct hlist_head *head;
 219
 220         head = vport_hash_bucket(dp, port_no);
 221         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 222                 if (vport->port_no == port_no)
 223                         return vport;
 224         }
 225         return NULL;
 226 }
 227
 228 /* Called with ovs_mutex. */
 229 static struct vport *new_vport(const struct vport_parms *parms)
 230 {
 231         struct vport *vport;
 232
 233         vport = ovs_vport_add(parms);
 234         if (!IS_ERR(vport)) {
 235                 struct datapath *dp = parms->dp;
 236                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 237
 238                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 239         }
 240         return vport;
 241 }
 242
 243 void ovs_dp_detach_port(struct vport *p)
 244 {
 245         ASSERT_OVSL();
 246
 247         /* First drop references to device. */
 248         hlist_del_rcu(&p->dp_hash_node);
 249
 250         /* Then destroy it. */
 251         ovs_vport_del(p);
 252 }
 253
 254 /* Must be called with rcu_read_lock. */
 255 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 256 {
 257         const struct vport *p = OVS_CB(skb)->input_vport;
 258         struct datapath *dp = p->dp;
 259         struct sw_flow *flow;
 260         struct sw_flow_actions *sf_acts;
 261         struct dp_stats_percpu *stats;
 262         u64 *stats_counter;
 263         u32 n_mask_hit;
 264
 265         stats = this_cpu_ptr(dp->stats_percpu);
 266
 267         /* Look up flow. */
 268         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
 269                                          &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
 279                 error = ovs_dp_upcall(dp, skb, key, &upcall);
 280                 if (unlikely(error))
 281                         kfree_skb(skb);
 282                 else
 283                         consume_skb(skb);
 284
 285                 stats_counter = &stats->n_missed;
 286                 goto out;
 287         }
 288
 289         ovs_flow_stats_update(flow, key->tp.flags, skb);
 290
 291         sf_acts = rcu_dereference(flow->sf_acts);
 292         ovs_execute_actions(dp, skb, key, sf_acts);
 293         stats_counter = &stats->n_hit;
 294
 295 out:
 296         /* Update datapath statistics. */
 297         u64_stats_update_begin(&stats->syncp);
 298         (*stats_counter)++;
 299         stats->n_mask_hit += n_mask_hit;
 300         u64_stats_update_end(&stats->syncp);
 301 }
 302
 303 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 304                   const struct sw_flow_key *key,
 305                   const struct dp_upcall_info *upcall_info)
 306 {
 307         struct dp_stats_percpu *stats;
 308         int err;
 309
 310         if (upcall_info->portid == 0) {
 311                 err = -ENOTCONN;
 312                 goto err;
 313         }
 314
 315         if (!skb_is_gso(skb))
 316                 err = queue_userspace_packet(dp, skb, key, upcall_info);
 317         else
 318                 err = queue_gso_packets(dp, skb, key, upcall_info);
 319         if (err)
 320                 goto err;
 321
 322         return 0;
 323
 324 err:
 325         stats = this_cpu_ptr(dp->stats_percpu);
 326
 327         u64_stats_update_begin(&stats->syncp);
 328         stats->n_lost++;
 329         u64_stats_update_end(&stats->syncp);
 330
 331         return err;
 332 }
 333
 334 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 335                              const struct sw_flow_key *key,
 336                              const struct dp_upcall_info *upcall_info)
 337 {
 338         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 339         struct sw_flow_key later_key;
 340         struct sk_buff *segs, *nskb;
 341         struct ovs_skb_cb ovs_cb;
 342         int err;
 343
 344         ovs_cb = *OVS_CB(skb);
 345         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 346         *OVS_CB(skb) = ovs_cb;
 347         if (IS_ERR(segs))
 348                 return PTR_ERR(segs);
 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 #ifdef HAVE_GENLMSG_NEW_UNICAST
 412                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 413 #endif
 414                 .snd_portid = upcall_info->portid,
 415         };
 416         size_t len;
 417         unsigned int hlen;
 418         int err, dp_ifindex;
 419
 420         dp_ifindex = get_dpifindex(dp);
 421         if (!dp_ifindex)
 422                 return -ENODEV;
 423
 424         if (vlan_tx_tag_present(skb)) {
 425                 nskb = skb_clone(skb, GFP_ATOMIC);
 426                 if (!nskb)
 427                         return -ENOMEM;
 428
 429                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 430                 if (!nskb)
 431                         return -ENOMEM;
 432
 433                 vlan_set_tci(nskb, 0);
 434
 435                 skb = nskb;
 436         }
 437
 438         if (nla_attr_size(skb->len) > USHRT_MAX) {
 439                 err = -EFBIG;
 440                 goto out;
 441         }
 442
 443         /* Complete checksum if needed */
 444         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 445             (err = skb_checksum_help(skb)))
 446                 goto out;
 447
 448         /* Older versions of OVS user space enforce alignment of the last
 449          * Netlink attribute to NLA_ALIGNTO which would require extensive
 450          * padding logic. Only perform zerocopy if padding is not required.
 451          */
 452         if (dp->user_features & OVS_DP_F_UNALIGNED)
 453                 hlen = skb_zerocopy_headlen(skb);
 454         else
 455                 hlen = skb->len;
 456
 457         len = upcall_msg_size(upcall_info, hlen);
 458         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 459         if (!user_skb) {
 460                 err = -ENOMEM;
 461                 goto out;
 462         }
 463
 464         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 465                              0, upcall_info->cmd);
 466         upcall->dp_ifindex = dp_ifindex;
 467
 468         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 469         err = ovs_nla_put_flow(key, key, user_skb);
 470         BUG_ON(err);
 471         nla_nest_end(user_skb, nla);
 472
 473         if (upcall_info->userdata)
 474                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 475                           nla_len(upcall_info->userdata),
 476                           nla_data(upcall_info->userdata));
 477
 478         if (upcall_info->egress_tun_info) {
 479                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 480                 err = ovs_nla_put_egress_tunnel_key(user_skb,
 481                                                     upcall_info->egress_tun_info);
 482                 BUG_ON(err);
 483                 nla_nest_end(user_skb, nla);
 484         }
 485
 486         /* Only reserve room for attribute header, packet data is added
 487          * in skb_zerocopy() */
 488         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 489                 err = -ENOBUFS;
 490                 goto out;
 491         }
 492         nla->nla_len = nla_attr_size(skb->len);
 493
 494         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 495         if (err)
 496                 goto out;
 497
 498         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 499         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 500                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 501
 502                 if (plen > 0)
 503                         memset(skb_put(user_skb, plen), 0, plen);
 504         }
 505
 506         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 507
 508         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 509         user_skb = NULL;
 510 out:
 511         if (err)
 512                 skb_tx_error(skb);
 513
 514         kfree_skb(user_skb);
 515         kfree_skb(nskb);
 516         return err;
 517 }
 518
 519 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 520 {
 521         struct ovs_header *ovs_header = info->userhdr;
 522         struct nlattr **a = info->attrs;
 523         struct sw_flow_actions *acts;
 524         struct sk_buff *packet;
 525         struct sw_flow *flow;
 526         struct sw_flow_actions *sf_acts;
 527         struct datapath *dp;
 528         struct ethhdr *eth;
 529         struct vport *input_vport;
 530         int len;
 531         int err;
 532         bool log = !a[OVS_FLOW_ATTR_PROBE];
 533
 534         err = -EINVAL;
 535         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 536             !a[OVS_PACKET_ATTR_ACTIONS])
 537                 goto err;
 538
 539         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 540         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 541         err = -ENOMEM;
 542         if (!packet)
 543                 goto err;
 544         skb_reserve(packet, NET_IP_ALIGN);
 545
 546         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 547
 548         skb_reset_mac_header(packet);
 549         eth = eth_hdr(packet);
 550
 551         /* Normally, setting the skb 'protocol' field would be handled by a
 552          * call to eth_type_trans(), but it assumes there's a sending
 553          * device, which we may not have. */
 554         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 555                 packet->protocol = eth->h_proto;
 556         else
 557                 packet->protocol = htons(ETH_P_802_2);
 558
 559         /* Build an sw_flow for sending this packet. */
 560         flow = ovs_flow_alloc();
 561         err = PTR_ERR(flow);
 562         if (IS_ERR(flow))
 563                 goto err_kfree_skb;
 564
 565         err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
 566                                              &flow->key, log);
 567         if (err)
 568                 goto err_flow_free;
 569
 570         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 571                                    &flow->key, &acts, log);
 572         if (err)
 573                 goto err_flow_free;
 574
 575         rcu_assign_pointer(flow->sf_acts, acts);
 576         OVS_CB(packet)->egress_tun_info = NULL;
 577         packet->priority = flow->key.phy.priority;
 578         packet->mark = flow->key.phy.skb_mark;
 579
 580         rcu_read_lock();
 581         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
 582         err = -ENODEV;
 583         if (!dp)
 584                 goto err_unlock;
 585
 586         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 587         if (!input_vport)
 588                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 589
 590         if (!input_vport)
 591                 goto err_unlock;
 592
 593         OVS_CB(packet)->input_vport = input_vport;
 594         sf_acts = rcu_dereference(flow->sf_acts);
 595
 596         local_bh_disable();
 597         err = ovs_execute_actions(dp, packet, &flow->key, sf_acts);
 598         local_bh_enable();
 599         rcu_read_unlock();
 600
 601         ovs_flow_free(flow, false);
 602         return err;
 603
 604 err_unlock:
 605         rcu_read_unlock();
 606 err_flow_free:
 607         ovs_flow_free(flow, false);
 608 err_kfree_skb:
 609         kfree_skb(packet);
 610 err:
 611         return err;
 612 }
 613
 614 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 615         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 616         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 617         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 618 };
 619
 620 static struct genl_ops dp_packet_genl_ops[] = {
 621         { .cmd = OVS_PACKET_CMD_EXECUTE,
 622           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 623           .policy = packet_policy,
 624           .doit = ovs_packet_cmd_execute
 625         }
 626 };
 627
 628 static struct genl_family dp_packet_genl_family = {
 629         .id = GENL_ID_GENERATE,
 630         .hdrsize = sizeof(struct ovs_header),
 631         .name = OVS_PACKET_FAMILY,
 632         .version = OVS_PACKET_VERSION,
 633         .maxattr = OVS_PACKET_ATTR_MAX,
 634         .netnsok = true,
 635         .parallel_ops = true,
 636         .ops = dp_packet_genl_ops,
 637         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 638 };
 639
 640 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 641                          struct ovs_dp_megaflow_stats *mega_stats)
 642 {
 643         int i;
 644
 645         memset(mega_stats, 0, sizeof(*mega_stats));
 646
 647         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 648         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 649
 650         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 651
 652         for_each_possible_cpu(i) {
 653                 const struct dp_stats_percpu *percpu_stats;
 654                 struct dp_stats_percpu local_stats;
 655                 unsigned int start;
 656
 657                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 658
 659                 do {
 660                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 661                         local_stats = *percpu_stats;
 662                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 663
 664                 stats->n_hit += local_stats.n_hit;
 665                 stats->n_missed += local_stats.n_missed;
 666                 stats->n_lost += local_stats.n_lost;
 667                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 668         }
 669 }
 670
 671 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 672 {
 673         return NLMSG_ALIGN(sizeof(struct ovs_header))
 674                 + nla_total_size(ovs_key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 675                 + nla_total_size(ovs_key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 676                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 677                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 678                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 679                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 680 }
 681
 682 /* Called with ovs_mutex or RCU read lock. */
 683 static int ovs_flow_cmd_fill_match(const struct sw_flow *flow,
 684                                    struct sk_buff *skb)
 685 {
 686         struct nlattr *nla;
 687         int err;
 688
 689         /* Fill flow key. */
 690         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 691         if (!nla)
 692                 return -EMSGSIZE;
 693
 694         err = ovs_nla_put_flow(&flow->unmasked_key,
 695                                &flow->unmasked_key, skb);
 696         if (err)
 697                 return err;
 698         nla_nest_end(skb, nla);
 699
 700         /* Fill flow mask. */
 701         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 702         if (!nla)
 703                 return -EMSGSIZE;
 704
 705         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 706         if (err)
 707                 return err;
 708         nla_nest_end(skb, nla);
 709
 710         return 0;
 711 }
 712
 713 /* Called with ovs_mutex or RCU read lock. */
 714 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 715                                    struct sk_buff *skb)
 716 {
 717         struct ovs_flow_stats stats;
 718         __be16 tcp_flags;
 719         unsigned long used;
 720
 721         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 722
 723         if (used &&
 724             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 725                 return -EMSGSIZE;
 726
 727         if (stats.n_packets &&
 728             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 729                 return -EMSGSIZE;
 730
 731         if ((u8)ntohs(tcp_flags) &&
 732              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 733                 return -EMSGSIZE;
 734
 735         return 0;
 736 }
 737
 738 /* Called with ovs_mutex or RCU read lock. */
 739 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 740                                      struct sk_buff *skb, int skb_orig_len)
 741 {
 742         struct nlattr *start;
 743         int err;
 744
 745         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 746          * this is the first flow to be dumped into 'skb'.  This is unusual for
 747          * Netlink but individual action lists can be longer than
 748          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 749          * The userspace caller can always fetch the actions separately if it
 750          * really wants them.  (Most userspace callers in fact don't care.)
 751          *
 752          * This can only fail for dump operations because the skb is always
 753          * properly sized for single flows.
 754          */
 755         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 756         if (start) {
 757                 const struct sw_flow_actions *sf_acts;
 758
 759                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 760                 err = ovs_nla_put_actions(sf_acts->actions,
 761                                           sf_acts->actions_len, skb);
 762
 763                 if (!err)
 764                         nla_nest_end(skb, start);
 765                 else {
 766                         if (skb_orig_len)
 767                                 return err;
 768
 769                         nla_nest_cancel(skb, start);
 770                 }
 771         } else if (skb_orig_len) {
 772                 return -EMSGSIZE;
 773         }
 774
 775         return 0;
 776 }
 777
 778 /* Called with ovs_mutex or RCU read lock. */
 779 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 780                                   struct sk_buff *skb, u32 portid,
 781                                   u32 seq, u32 flags, u8 cmd)
 782 {
 783         const int skb_orig_len = skb->len;
 784         struct ovs_header *ovs_header;
 785         int err;
 786
 787         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 788         if (!ovs_header)
 789                 return -EMSGSIZE;
 790         ovs_header->dp_ifindex = dp_ifindex;
 791
 792         err = ovs_flow_cmd_fill_match(flow, skb);
 793         if (err)
 794                 goto error;
 795
 796         err = ovs_flow_cmd_fill_stats(flow, skb);
 797         if (err)
 798                 goto error;
 799
 800         err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 801         if (err)
 802                 goto error;
 803
 804         return genlmsg_end(skb, ovs_header);
 805
 806 error:
 807         genlmsg_cancel(skb, ovs_header);
 808         return err;
 809 }
 810
 811 /* May not be called with RCU read lock. */
 812 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 813                                                struct genl_info *info,
 814                                                bool always)
 815 {
 816         struct sk_buff *skb;
 817
 818         if (!always && !ovs_must_notify(&dp_flow_genl_family, info,
 819                                         GROUP_ID(&ovs_dp_flow_multicast_group)))
 820                 return NULL;
 821
 822         skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
 823
 824         if (!skb)
 825                 return ERR_PTR(-ENOMEM);
 826
 827         return skb;
 828 }
 829
 830 /* Called with ovs_mutex. */
 831 static struct sk_buff *ovs_flow_cmd_build_info(struct datapath *dp,
 832                                                const struct sw_flow *flow,
 833                                                int dp_ifindex,
 834                                                struct genl_info *info, u8 cmd,
 835                                                bool always)
 836 {
 837         struct sk_buff *skb;
 838         int retval;
 839
 840         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
 841                                       always);
 842         if (IS_ERR_OR_NULL(skb))
 843                 return skb;
 844
 845         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 846                                         info->snd_portid, info->snd_seq, 0,
 847                                         cmd);
 848         BUG_ON(retval < 0);
 849         return skb;
 850 }
 851
 852 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 853 {
 854         struct nlattr **a = info->attrs;
 855         struct ovs_header *ovs_header = info->userhdr;
 856         struct sw_flow *flow, *new_flow;
 857         struct sw_flow_mask mask;
 858         struct sk_buff *reply;
 859         struct datapath *dp;
 860         struct sw_flow_actions *acts;
 861         struct sw_flow_match match;
 862         int error;
 863         bool log = !a[OVS_FLOW_ATTR_PROBE];
 864
 865         /* Must have key and actions. */
 866         error = -EINVAL;
 867         if (!a[OVS_FLOW_ATTR_KEY]) {
 868                 OVS_NLERR(log, "Flow key attribute not present in new flow.");
 869                 goto error;
 870         }
 871         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 872                 OVS_NLERR(log,
 873                           "Flow actions attribute not present in new flow.");
 874                 goto error;
 875         }
 876
 877         /* Most of the time we need to allocate a new flow, do it before
 878          * locking. */
 879         new_flow = ovs_flow_alloc();
 880         if (IS_ERR(new_flow)) {
 881                 error = PTR_ERR(new_flow);
 882                 goto error;
 883         }
 884
 885         /* Extract key. */
 886         ovs_match_init(&match, &new_flow->unmasked_key, &mask);
 887         error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
 888                                   a[OVS_FLOW_ATTR_MASK], log);
 889         if (error)
 890                 goto err_kfree_flow;
 891
 892         ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
 893
 894         /* Validate actions. */
 895         error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
 896                                      &acts, log);
 897         if (error) {
 898                 OVS_NLERR(
 899                         log,
 900                         "Flow actions may not be safe on all matching packets."
 901                         );
 902                 goto err_kfree_flow;
 903         }
 904
 905         reply = ovs_flow_cmd_alloc_info(acts, info, false);
 906         if (IS_ERR(reply)) {
 907                 error = PTR_ERR(reply);
 908                 goto err_kfree_acts;
 909         }
 910
 911         ovs_lock();
 912         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 913         if (unlikely(!dp)) {
 914                 error = -ENODEV;
 915                 goto err_unlock_ovs;
 916         }
 917         /* Check if this is a duplicate flow */
 918         flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
 919         if (likely(!flow)) {
 920                 rcu_assign_pointer(new_flow->sf_acts, acts);
 921
 922                 /* Put flow in bucket. */
 923                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 924                 if (unlikely(error)) {
 925                         acts = NULL;
 926                         goto err_unlock_ovs;
 927                 }
 928
 929                 if (unlikely(reply)) {
 930                         error = ovs_flow_cmd_fill_info(new_flow,
 931                                                        ovs_header->dp_ifindex,
 932                                                        reply, info->snd_portid,
 933                                                        info->snd_seq, 0,
 934                                                        OVS_FLOW_CMD_NEW);
 935                         BUG_ON(error < 0);
 936                 }
 937                 ovs_unlock();
 938         } else {
 939                 struct sw_flow_actions *old_acts;
 940
 941                 /* Bail out if we're not allowed to modify an existing flow.
 942                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 943                  * because Generic Netlink treats the latter as a dump
 944                  * request.  We also accept NLM_F_EXCL in case that bug ever
 945                  * gets fixed.
 946                  */
 947                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
 948                                                          | NLM_F_EXCL))) {
 949                         error = -EEXIST;
 950                         goto err_unlock_ovs;
 951                 }
 952                 /* The unmasked key has to be the same for flow updates. */
 953                 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
 954                         /* Look for any overlapping flow. */
 955                         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
 956                         if (!flow) {
 957                                 error = -ENOENT;
 958                                 goto err_unlock_ovs;
 959                         }
 960                 }
 961                 /* Update actions. */
 962                 old_acts = ovsl_dereference(flow->sf_acts);
 963                 rcu_assign_pointer(flow->sf_acts, acts);
 964
 965                 if (unlikely(reply)) {
 966                         error = ovs_flow_cmd_fill_info(flow,
 967                                                        ovs_header->dp_ifindex,
 968                                                        reply, info->snd_portid,
 969                                                        info->snd_seq, 0,
 970                                                        OVS_FLOW_CMD_NEW);
 971                         BUG_ON(error < 0);
 972                 }
 973                 ovs_unlock();
 974
 975                 ovs_nla_free_flow_actions(old_acts);
 976                 ovs_flow_free(new_flow, false);
 977         }
 978
 979         if (reply)
 980                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
 981         return 0;
 982
 983 err_unlock_ovs:
 984         ovs_unlock();
 985         kfree_skb(reply);
 986 err_kfree_acts:
 987         kfree(acts);
 988 err_kfree_flow:
 989         ovs_flow_free(new_flow, false);
 990 error:
 991         return error;
 992 }
 993
 994 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
 995 static struct sw_flow_actions *get_flow_actions(const struct nlattr *a,
 996                                                 const struct sw_flow_key *key,
 997                                                 const struct sw_flow_mask *mask,
 998                                                 bool log)
 999 {
1000         struct sw_flow_actions *acts;
1001         struct sw_flow_key masked_key;
1002         int error;
1003
1004         ovs_flow_mask_key(&masked_key, key, mask);
1005         error = ovs_nla_copy_actions(a, &masked_key, &acts, log);
1006         if (error) {
1007                 OVS_NLERR(log,
1008                           "Actions may not be safe on all matching packets.");
1009                 return ERR_PTR(error);
1010         }
1011
1012         return acts;
1013 }
1014
1015 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1016 {
1017         struct nlattr **a = info->attrs;
1018         struct ovs_header *ovs_header = info->userhdr;
1019         struct sw_flow_key key;
1020         struct sw_flow *flow;
1021         struct sw_flow_mask mask;
1022         struct sk_buff *reply = NULL;
1023         struct datapath *dp;
1024         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1025         struct sw_flow_match match;
1026         int error;
1027         bool log = !a[OVS_FLOW_ATTR_PROBE];
1028
1029         /* Extract key. */
1030         error = -EINVAL;
1031         if (!a[OVS_FLOW_ATTR_KEY]) {
1032                 OVS_NLERR(log, "Flow key attribute not present in set flow.");
1033                 goto error;
1034         }
1035
1036         ovs_match_init(&match, &key, &mask);
1037         error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
1038                                   a[OVS_FLOW_ATTR_MASK], log);
1039         if (error)
1040                 goto error;
1041
1042         /* Validate actions. */
1043         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1044                 acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask,
1045                                         log);
1046                 if (IS_ERR(acts)) {
1047                         error = PTR_ERR(acts);
1048                         goto error;
1049                 }
1050
1051                 /* Can allocate before locking if have acts. */
1052                 reply = ovs_flow_cmd_alloc_info(acts, info, false);
1053                 if (IS_ERR(reply)) {
1054                         error = PTR_ERR(reply);
1055                         goto err_kfree_acts;
1056                 }
1057         }
1058
1059         ovs_lock();
1060         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1061         if (unlikely(!dp)) {
1062                 error = -ENODEV;
1063                 goto err_unlock_ovs;
1064         }
1065         /* Check that the flow exists. */
1066         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1067         if (unlikely(!flow)) {
1068                 error = -ENOENT;
1069                 goto err_unlock_ovs;
1070         }
1071
1072         /* Update actions, if present. */
1073         if (likely(acts)) {
1074                 old_acts = ovsl_dereference(flow->sf_acts);
1075                 rcu_assign_pointer(flow->sf_acts, acts);
1076
1077                 if (unlikely(reply)) {
1078                         error = ovs_flow_cmd_fill_info(flow,
1079                                                        ovs_header->dp_ifindex,
1080                                                        reply, info->snd_portid,
1081                                                        info->snd_seq, 0,
1082                                                        OVS_FLOW_CMD_NEW);
1083                         BUG_ON(error < 0);
1084                 }
1085         } else {
1086                 /* Could not alloc without acts before locking. */
1087                 reply = ovs_flow_cmd_build_info(dp, flow,
1088                                                 ovs_header->dp_ifindex,
1089                                                 info, OVS_FLOW_CMD_NEW, false);
1090                 if (unlikely(IS_ERR(reply))) {
1091                         error = PTR_ERR(reply);
1092                         goto err_unlock_ovs;
1093                 }
1094         }
1095
1096         /* Clear stats. */
1097         if (a[OVS_FLOW_ATTR_CLEAR])
1098                 ovs_flow_stats_clear(flow);
1099         ovs_unlock();
1100
1101         if (reply)
1102                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1103         if (old_acts)
1104                 ovs_nla_free_flow_actions(old_acts);
1105         return 0;
1106
1107 err_unlock_ovs:
1108         ovs_unlock();
1109         kfree_skb(reply);
1110 err_kfree_acts:
1111         kfree(acts);
1112 error:
1113         return error;
1114 }
1115
1116 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1117 {
1118         struct nlattr **a = info->attrs;
1119         struct ovs_header *ovs_header = info->userhdr;
1120         struct sw_flow_key key;
1121         struct sk_buff *reply;
1122         struct sw_flow *flow;
1123         struct datapath *dp;
1124         struct sw_flow_match match;
1125         int err;
1126         bool log = !a[OVS_FLOW_ATTR_PROBE];
1127
1128         if (!a[OVS_FLOW_ATTR_KEY]) {
1129                 OVS_NLERR(log,
1130                           "Flow get message rejected, Key attribute missing.");
1131                 return -EINVAL;
1132         }
1133
1134         ovs_match_init(&match, &key, NULL);
1135         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL, log);
1136         if (err)
1137                 return err;
1138
1139         ovs_lock();
1140         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1141         if (!dp) {
1142                 err = -ENODEV;
1143                 goto unlock;
1144         }
1145
1146         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1147         if (!flow) {
1148                 err = -ENOENT;
1149                 goto unlock;
1150         }
1151
1152         reply = ovs_flow_cmd_build_info(dp, flow, ovs_header->dp_ifindex, info,
1153                                         OVS_FLOW_CMD_NEW, true);
1154         if (IS_ERR(reply)) {
1155                 err = PTR_ERR(reply);
1156                 goto unlock;
1157         }
1158
1159         ovs_unlock();
1160         return genlmsg_reply(reply, info);
1161 unlock:
1162         ovs_unlock();
1163         return err;
1164 }
1165
1166 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1167 {
1168         struct nlattr **a = info->attrs;
1169         struct ovs_header *ovs_header = info->userhdr;
1170         struct sw_flow_key key;
1171         struct sk_buff *reply;
1172         struct sw_flow *flow;
1173         struct datapath *dp;
1174         struct sw_flow_match match;
1175         int err;
1176         bool log = !a[OVS_FLOW_ATTR_PROBE];
1177
1178         if (likely(a[OVS_FLOW_ATTR_KEY])) {
1179                 ovs_match_init(&match, &key, NULL);
1180                 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL,
1181                                         log);
1182                 if (unlikely(err))
1183                         return err;
1184         }
1185
1186         ovs_lock();
1187         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1188         if (unlikely(!dp)) {
1189                 err = -ENODEV;
1190                 goto unlock;
1191         }
1192         if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1193                 err = ovs_flow_tbl_flush(&dp->table);
1194                 goto unlock;
1195         }
1196         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1197         if (unlikely(!flow)) {
1198                 err = -ENOENT;
1199                 goto unlock;
1200         }
1201
1202         ovs_flow_tbl_remove(&dp->table, flow);
1203         ovs_unlock();
1204
1205         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *)flow->sf_acts,
1206                                         info, false);
1207
1208         if (likely(reply)) {
1209                 if (likely(!IS_ERR(reply))) {
1210                         rcu_read_lock(); /* Keep RCU checker happy. */
1211                         err = ovs_flow_cmd_fill_info(flow,
1212                                                      ovs_header->dp_ifindex,
1213                                                      reply, info->snd_portid,
1214                                                      info->snd_seq, 0,
1215                                                      OVS_FLOW_CMD_DEL);
1216                         rcu_read_unlock();
1217                         BUG_ON(err < 0);
1218                         ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1219                 } else {
1220                         genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1221                                      GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1222
1223                 }
1224         }
1225
1226         ovs_flow_free(flow, true);
1227         return 0;
1228 unlock:
1229         ovs_unlock();
1230         return err;
1231 }
1232
1233 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1234 {
1235         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1236         struct table_instance *ti;
1237         struct datapath *dp;
1238
1239         rcu_read_lock();
1240         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1241         if (!dp) {
1242                 rcu_read_unlock();
1243                 return -ENODEV;
1244         }
1245
1246         ti = rcu_dereference(dp->table.ti);
1247         for (;;) {
1248                 struct sw_flow *flow;
1249                 u32 bucket, obj;
1250
1251                 bucket = cb->args[0];
1252                 obj = cb->args[1];
1253                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1254                 if (!flow)
1255                         break;
1256
1257                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1258                                            NETLINK_CB(cb->skb).portid,
1259                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1260                                            OVS_FLOW_CMD_NEW) < 0)
1261                         break;
1262
1263                 cb->args[0] = bucket;
1264                 cb->args[1] = obj;
1265         }
1266         rcu_read_unlock();
1267         return skb->len;
1268 }
1269
1270 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1271         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1272         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1273         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1274         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1275         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1276 };
1277
1278 static struct genl_ops dp_flow_genl_ops[] = {
1279         { .cmd = OVS_FLOW_CMD_NEW,
1280           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1281           .policy = flow_policy,
1282           .doit = ovs_flow_cmd_new
1283         },
1284         { .cmd = OVS_FLOW_CMD_DEL,
1285           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1286           .policy = flow_policy,
1287           .doit = ovs_flow_cmd_del
1288         },
1289         { .cmd = OVS_FLOW_CMD_GET,
1290           .flags = 0,               /* OK for unprivileged users. */
1291           .policy = flow_policy,
1292           .doit = ovs_flow_cmd_get,
1293           .dumpit = ovs_flow_cmd_dump
1294         },
1295         { .cmd = OVS_FLOW_CMD_SET,
1296           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1297           .policy = flow_policy,
1298           .doit = ovs_flow_cmd_set,
1299         },
1300 };
1301
1302 static struct genl_family dp_flow_genl_family = {
1303         .id = GENL_ID_GENERATE,
1304         .hdrsize = sizeof(struct ovs_header),
1305         .name = OVS_FLOW_FAMILY,
1306         .version = OVS_FLOW_VERSION,
1307         .maxattr = OVS_FLOW_ATTR_MAX,
1308         .netnsok = true,
1309         .parallel_ops = true,
1310         .ops = dp_flow_genl_ops,
1311         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1312         .mcgrps = &ovs_dp_flow_multicast_group,
1313         .n_mcgrps = 1,
1314 };
1315
1316 static size_t ovs_dp_cmd_msg_size(void)
1317 {
1318         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1319
1320         msgsize += nla_total_size(IFNAMSIZ);
1321         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1322         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1323         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1324
1325         return msgsize;
1326 }
1327
1328 /* Called with ovs_mutex or RCU read lock. */
1329 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1330                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1331 {
1332         struct ovs_header *ovs_header;
1333         struct ovs_dp_stats dp_stats;
1334         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1335         int err;
1336
1337         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1338                                    flags, cmd);
1339         if (!ovs_header)
1340                 goto error;
1341
1342         ovs_header->dp_ifindex = get_dpifindex(dp);
1343
1344         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1345         if (err)
1346                 goto nla_put_failure;
1347
1348         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1349         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1350                         &dp_stats))
1351                 goto nla_put_failure;
1352
1353         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1354                         sizeof(struct ovs_dp_megaflow_stats),
1355                         &dp_megaflow_stats))
1356                 goto nla_put_failure;
1357
1358         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1359                 goto nla_put_failure;
1360
1361         return genlmsg_end(skb, ovs_header);
1362
1363 nla_put_failure:
1364         genlmsg_cancel(skb, ovs_header);
1365 error:
1366         return -EMSGSIZE;
1367 }
1368
1369 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1370 {
1371         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1372 }
1373
1374 /* Called with rcu_read_lock or ovs_mutex. */
1375 static struct datapath *lookup_datapath(struct net *net,
1376                                         const struct ovs_header *ovs_header,
1377                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1378 {
1379         struct datapath *dp;
1380
1381         if (!a[OVS_DP_ATTR_NAME])
1382                 dp = get_dp(net, ovs_header->dp_ifindex);
1383         else {
1384                 struct vport *vport;
1385
1386                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1387                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1388         }
1389         return dp ? dp : ERR_PTR(-ENODEV);
1390 }
1391
1392 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1393 {
1394         struct datapath *dp;
1395
1396         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1397         if (IS_ERR(dp))
1398                 return;
1399
1400         WARN(dp->user_features, "Dropping previously announced user features\n");
1401         dp->user_features = 0;
1402 }
1403
1404 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1405 {
1406         if (a[OVS_DP_ATTR_USER_FEATURES])
1407                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1408 }
1409
1410 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1411 {
1412         struct nlattr **a = info->attrs;
1413         struct vport_parms parms;
1414         struct sk_buff *reply;
1415         struct datapath *dp;
1416         struct vport *vport;
1417         struct ovs_net *ovs_net;
1418         int err, i;
1419
1420         err = -EINVAL;
1421         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1422                 goto err;
1423
1424         reply = ovs_dp_cmd_alloc_info(info);
1425         if (!reply)
1426                 return -ENOMEM;
1427
1428         err = -ENOMEM;
1429         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1430         if (dp == NULL)
1431                 goto err_free_reply;
1432
1433         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1434
1435         /* Allocate table. */
1436         err = ovs_flow_tbl_init(&dp->table);
1437         if (err)
1438                 goto err_free_dp;
1439
1440         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1441         if (!dp->stats_percpu) {
1442                 err = -ENOMEM;
1443                 goto err_destroy_table;
1444         }
1445
1446         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1447                             GFP_KERNEL);
1448         if (!dp->ports) {
1449                 err = -ENOMEM;
1450                 goto err_destroy_percpu;
1451         }
1452
1453         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1454                 INIT_HLIST_HEAD(&dp->ports[i]);
1455
1456         /* Set up our datapath device. */
1457         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1458         parms.type = OVS_VPORT_TYPE_INTERNAL;
1459         parms.options = NULL;
1460         parms.dp = dp;
1461         parms.port_no = OVSP_LOCAL;
1462         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1463
1464         ovs_dp_change(dp, a);
1465
1466         /* So far only local changes have been made, now need the lock. */
1467         ovs_lock();
1468
1469         vport = new_vport(&parms);
1470         if (IS_ERR(vport)) {
1471                 err = PTR_ERR(vport);
1472                 if (err == -EBUSY)
1473                         err = -EEXIST;
1474
1475                 if (err == -EEXIST) {
1476                         /* An outdated user space instance that does not understand
1477                          * the concept of user_features has attempted to create a new
1478                          * datapath and is likely to reuse it. Drop all user features.
1479                          */
1480                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1481                                 ovs_dp_reset_user_features(skb, info);
1482                 }
1483
1484                 goto err_destroy_ports_array;
1485         }
1486
1487         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1488                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1489         BUG_ON(err < 0);
1490
1491         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1492         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1493
1494         ovs_unlock();
1495
1496         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1497         return 0;
1498
1499 err_destroy_ports_array:
1500         ovs_unlock();
1501         kfree(dp->ports);
1502 err_destroy_percpu:
1503         free_percpu(dp->stats_percpu);
1504 err_destroy_table:
1505         ovs_flow_tbl_destroy(&dp->table);
1506 err_free_dp:
1507         release_net(ovs_dp_get_net(dp));
1508         kfree(dp);
1509 err_free_reply:
1510         kfree_skb(reply);
1511 err:
1512         return err;
1513 }
1514
1515 /* Called with ovs_mutex. */
1516 static void __dp_destroy(struct datapath *dp)
1517 {
1518         int i;
1519
1520         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1521                 struct vport *vport;
1522                 struct hlist_node *n;
1523
1524                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1525                         if (vport->port_no != OVSP_LOCAL)
1526                                 ovs_dp_detach_port(vport);
1527         }
1528
1529         list_del_rcu(&dp->list_node);
1530
1531         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1532          * all ports in datapath are destroyed first before freeing datapath.
1533          */
1534         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1535
1536         /* RCU destroy the flow table */
1537         call_rcu(&dp->rcu, destroy_dp_rcu);
1538 }
1539
1540 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1541 {
1542         struct sk_buff *reply;
1543         struct datapath *dp;
1544         int err;
1545
1546         reply = ovs_dp_cmd_alloc_info(info);
1547         if (!reply)
1548                 return -ENOMEM;
1549
1550         ovs_lock();
1551         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1552         err = PTR_ERR(dp);
1553         if (IS_ERR(dp))
1554                 goto err_unlock_free;
1555
1556         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1557                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1558         BUG_ON(err < 0);
1559
1560         __dp_destroy(dp);
1561
1562         ovs_unlock();
1563         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1564         return 0;
1565
1566 err_unlock_free:
1567         ovs_unlock();
1568         kfree_skb(reply);
1569         return err;
1570 }
1571
1572 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1573 {
1574         struct sk_buff *reply;
1575         struct datapath *dp;
1576         int err;
1577
1578         reply = ovs_dp_cmd_alloc_info(info);
1579         if (!reply)
1580                 return -ENOMEM;
1581
1582         ovs_lock();
1583         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1584         err = PTR_ERR(dp);
1585         if (IS_ERR(dp))
1586                 goto err_unlock_free;
1587
1588         ovs_dp_change(dp, info->attrs);
1589
1590         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1591                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1592         BUG_ON(err < 0);
1593
1594         ovs_unlock();
1595         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1596         return 0;
1597
1598 err_unlock_free:
1599         ovs_unlock();
1600         kfree_skb(reply);
1601         return err;
1602 }
1603
1604 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1605 {
1606         struct sk_buff *reply;
1607         struct datapath *dp;
1608         int err;
1609
1610         reply = ovs_dp_cmd_alloc_info(info);
1611         if (!reply)
1612                 return -ENOMEM;
1613
1614         rcu_read_lock();
1615         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1616         if (IS_ERR(dp)) {
1617                 err = PTR_ERR(dp);
1618                 goto err_unlock_free;
1619         }
1620         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1621                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1622         BUG_ON(err < 0);
1623         rcu_read_unlock();
1624
1625         return genlmsg_reply(reply, info);
1626
1627 err_unlock_free:
1628         rcu_read_unlock();
1629         kfree_skb(reply);
1630         return err;
1631 }
1632
1633 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1634 {
1635         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1636         struct datapath *dp;
1637         int skip = cb->args[0];
1638         int i = 0;
1639
1640         rcu_read_lock();
1641         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1642                 if (i >= skip &&
1643                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1644                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1645                                          OVS_DP_CMD_NEW) < 0)
1646                         break;
1647                 i++;
1648         }
1649         rcu_read_unlock();
1650
1651         cb->args[0] = i;
1652
1653         return skb->len;
1654 }
1655
1656 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1657         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1658         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1659         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1660 };
1661
1662 static struct genl_ops dp_datapath_genl_ops[] = {
1663         { .cmd = OVS_DP_CMD_NEW,
1664           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1665           .policy = datapath_policy,
1666           .doit = ovs_dp_cmd_new
1667         },
1668         { .cmd = OVS_DP_CMD_DEL,
1669           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1670           .policy = datapath_policy,
1671           .doit = ovs_dp_cmd_del
1672         },
1673         { .cmd = OVS_DP_CMD_GET,
1674           .flags = 0,               /* OK for unprivileged users. */
1675           .policy = datapath_policy,
1676           .doit = ovs_dp_cmd_get,
1677           .dumpit = ovs_dp_cmd_dump
1678         },
1679         { .cmd = OVS_DP_CMD_SET,
1680           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1681           .policy = datapath_policy,
1682           .doit = ovs_dp_cmd_set,
1683         },
1684 };
1685
1686 static struct genl_family dp_datapath_genl_family = {
1687         .id = GENL_ID_GENERATE,
1688         .hdrsize = sizeof(struct ovs_header),
1689         .name = OVS_DATAPATH_FAMILY,
1690         .version = OVS_DATAPATH_VERSION,
1691         .maxattr = OVS_DP_ATTR_MAX,
1692         .netnsok = true,
1693         .parallel_ops = true,
1694         .ops = dp_datapath_genl_ops,
1695         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1696         .mcgrps = &ovs_dp_datapath_multicast_group,
1697         .n_mcgrps = 1,
1698 };
1699
1700 /* Called with ovs_mutex or RCU read lock. */
1701 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1702                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1703 {
1704         struct ovs_header *ovs_header;
1705         struct ovs_vport_stats vport_stats;
1706         int err;
1707
1708         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1709                                  flags, cmd);
1710         if (!ovs_header)
1711                 return -EMSGSIZE;
1712
1713         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1714
1715         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1716             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1717             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)))
1718                 goto nla_put_failure;
1719
1720         ovs_vport_get_stats(vport, &vport_stats);
1721         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1722                     &vport_stats))
1723                 goto nla_put_failure;
1724
1725         if (ovs_vport_get_upcall_portids(vport, skb))
1726                 goto nla_put_failure;
1727
1728         err = ovs_vport_get_options(vport, skb);
1729         if (err == -EMSGSIZE)
1730                 goto error;
1731
1732         return genlmsg_end(skb, ovs_header);
1733
1734 nla_put_failure:
1735         err = -EMSGSIZE;
1736 error:
1737         genlmsg_cancel(skb, ovs_header);
1738         return err;
1739 }
1740
1741 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1742 {
1743         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1744 }
1745
1746 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1747 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1748                                          u32 seq, u8 cmd)
1749 {
1750         struct sk_buff *skb;
1751         int retval;
1752
1753         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1754         if (!skb)
1755                 return ERR_PTR(-ENOMEM);
1756
1757         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1758         BUG_ON(retval < 0);
1759
1760         return skb;
1761 }
1762
1763 /* Called with ovs_mutex or RCU read lock. */
1764 static struct vport *lookup_vport(struct net *net,
1765                                   const struct ovs_header *ovs_header,
1766                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1767 {
1768         struct datapath *dp;
1769         struct vport *vport;
1770
1771         if (a[OVS_VPORT_ATTR_NAME]) {
1772                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1773                 if (!vport)
1774                         return ERR_PTR(-ENODEV);
1775                 if (ovs_header->dp_ifindex &&
1776                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1777                         return ERR_PTR(-ENODEV);
1778                 return vport;
1779         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1780                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1781
1782                 if (port_no >= DP_MAX_PORTS)
1783                         return ERR_PTR(-EFBIG);
1784
1785                 dp = get_dp(net, ovs_header->dp_ifindex);
1786                 if (!dp)
1787                         return ERR_PTR(-ENODEV);
1788
1789                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1790                 if (!vport)
1791                         return ERR_PTR(-ENODEV);
1792                 return vport;
1793         } else
1794                 return ERR_PTR(-EINVAL);
1795 }
1796
1797 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1798 {
1799         struct nlattr **a = info->attrs;
1800         struct ovs_header *ovs_header = info->userhdr;
1801         struct vport_parms parms;
1802         struct sk_buff *reply;
1803         struct vport *vport;
1804         struct datapath *dp;
1805         u32 port_no;
1806         int err;
1807
1808         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1809             !a[OVS_VPORT_ATTR_UPCALL_PID])
1810                 return -EINVAL;
1811
1812         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1813                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1814         if (port_no >= DP_MAX_PORTS)
1815                 return -EFBIG;
1816
1817         reply = ovs_vport_cmd_alloc_info();
1818         if (!reply)
1819                 return -ENOMEM;
1820
1821         ovs_lock();
1822         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1823         err = -ENODEV;
1824         if (!dp)
1825                 goto exit_unlock_free;
1826
1827         if (port_no) {
1828                 vport = ovs_vport_ovsl(dp, port_no);
1829                 err = -EBUSY;
1830                 if (vport)
1831                         goto exit_unlock_free;
1832         } else {
1833                 for (port_no = 1; ; port_no++) {
1834                         if (port_no >= DP_MAX_PORTS) {
1835                                 err = -EFBIG;
1836                                 goto exit_unlock_free;
1837                         }
1838                         vport = ovs_vport_ovsl(dp, port_no);
1839                         if (!vport)
1840                                 break;
1841                 }
1842         }
1843
1844         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1845         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1846         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1847         parms.dp = dp;
1848         parms.port_no = port_no;
1849         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1850
1851         vport = new_vport(&parms);
1852         err = PTR_ERR(vport);
1853         if (IS_ERR(vport))
1854                 goto exit_unlock_free;
1855
1856         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1857                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1858         BUG_ON(err < 0);
1859         ovs_unlock();
1860
1861         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1862         return 0;
1863
1864 exit_unlock_free:
1865         ovs_unlock();
1866         kfree_skb(reply);
1867         return err;
1868 }
1869
1870 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1871 {
1872         struct nlattr **a = info->attrs;
1873         struct sk_buff *reply;
1874         struct vport *vport;
1875         int err;
1876
1877         reply = ovs_vport_cmd_alloc_info();
1878         if (!reply)
1879                 return -ENOMEM;
1880
1881         ovs_lock();
1882         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1883         err = PTR_ERR(vport);
1884         if (IS_ERR(vport))
1885                 goto exit_unlock_free;
1886
1887         if (a[OVS_VPORT_ATTR_TYPE] &&
1888             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1889                 err = -EINVAL;
1890                 goto exit_unlock_free;
1891         }
1892
1893         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1894                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1895                 if (err)
1896                         goto exit_unlock_free;
1897         }
1898
1899         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1900                 err = ovs_vport_set_upcall_portids(vport,
1901                                                    a[OVS_VPORT_ATTR_UPCALL_PID]);
1902                 if (err)
1903                         goto exit_unlock_free;
1904         }
1905
1906         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1907                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1908         BUG_ON(err < 0);
1909         ovs_unlock();
1910
1911         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1912         return 0;
1913
1914 exit_unlock_free:
1915         ovs_unlock();
1916         kfree_skb(reply);
1917         return err;
1918 }
1919
1920 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1921 {
1922         struct nlattr **a = info->attrs;
1923         struct sk_buff *reply;
1924         struct vport *vport;
1925         int err;
1926
1927         reply = ovs_vport_cmd_alloc_info();
1928         if (!reply)
1929                 return -ENOMEM;
1930
1931         ovs_lock();
1932         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1933         err = PTR_ERR(vport);
1934         if (IS_ERR(vport))
1935                 goto exit_unlock_free;
1936
1937         if (vport->port_no == OVSP_LOCAL) {
1938                 err = -EINVAL;
1939                 goto exit_unlock_free;
1940         }
1941
1942         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1943                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1944         BUG_ON(err < 0);
1945         ovs_dp_detach_port(vport);
1946         ovs_unlock();
1947
1948         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1949         return 0;
1950
1951 exit_unlock_free:
1952         ovs_unlock();
1953         kfree_skb(reply);
1954         return err;
1955 }
1956
1957 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1958 {
1959         struct nlattr **a = info->attrs;
1960         struct ovs_header *ovs_header = info->userhdr;
1961         struct sk_buff *reply;
1962         struct vport *vport;
1963         int err;
1964
1965         reply = ovs_vport_cmd_alloc_info();
1966         if (!reply)
1967                 return -ENOMEM;
1968
1969         rcu_read_lock();
1970         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1971         err = PTR_ERR(vport);
1972         if (IS_ERR(vport))
1973                 goto exit_unlock_free;
1974         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1975                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1976         BUG_ON(err < 0);
1977         rcu_read_unlock();
1978
1979         return genlmsg_reply(reply, info);
1980
1981 exit_unlock_free:
1982         rcu_read_unlock();
1983         kfree_skb(reply);
1984         return err;
1985 }
1986
1987 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1988 {
1989         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1990         struct datapath *dp;
1991         int bucket = cb->args[0], skip = cb->args[1];
1992         int i, j = 0;
1993
1994         rcu_read_lock();
1995         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1996         if (!dp) {
1997                 rcu_read_unlock();
1998                 return -ENODEV;
1999         }
2000         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2001                 struct vport *vport;
2002
2003                 j = 0;
2004                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2005                         if (j >= skip &&
2006                             ovs_vport_cmd_fill_info(vport, skb,
2007                                                     NETLINK_CB(cb->skb).portid,
2008                                                     cb->nlh->nlmsg_seq,
2009                                                     NLM_F_MULTI,
2010                                                     OVS_VPORT_CMD_NEW) < 0)
2011                                 goto out;
2012
2013                         j++;
2014                 }
2015                 skip = 0;
2016         }
2017 out:
2018         rcu_read_unlock();
2019
2020         cb->args[0] = i;
2021         cb->args[1] = j;
2022
2023         return skb->len;
2024 }
2025
2026 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2027         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2028         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2029         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2030         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2031         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2032         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2033 };
2034
2035 static struct genl_ops dp_vport_genl_ops[] = {
2036         { .cmd = OVS_VPORT_CMD_NEW,
2037           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2038           .policy = vport_policy,
2039           .doit = ovs_vport_cmd_new
2040         },
2041         { .cmd = OVS_VPORT_CMD_DEL,
2042           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2043           .policy = vport_policy,
2044           .doit = ovs_vport_cmd_del
2045         },
2046         { .cmd = OVS_VPORT_CMD_GET,
2047           .flags = 0,               /* OK for unprivileged users. */
2048           .policy = vport_policy,
2049           .doit = ovs_vport_cmd_get,
2050           .dumpit = ovs_vport_cmd_dump
2051         },
2052         { .cmd = OVS_VPORT_CMD_SET,
2053           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2054           .policy = vport_policy,
2055           .doit = ovs_vport_cmd_set,
2056         },
2057 };
2058
2059 struct genl_family dp_vport_genl_family = {
2060         .id = GENL_ID_GENERATE,
2061         .hdrsize = sizeof(struct ovs_header),
2062         .name = OVS_VPORT_FAMILY,
2063         .version = OVS_VPORT_VERSION,
2064         .maxattr = OVS_VPORT_ATTR_MAX,
2065         .netnsok = true,
2066         .parallel_ops = true,
2067         .ops = dp_vport_genl_ops,
2068         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2069         .mcgrps = &ovs_dp_vport_multicast_group,
2070         .n_mcgrps = 1,
2071 };
2072
2073 static struct genl_family *dp_genl_families[] = {
2074         &dp_datapath_genl_family,
2075         &dp_vport_genl_family,
2076         &dp_flow_genl_family,
2077         &dp_packet_genl_family,
2078 };
2079
2080 static void dp_unregister_genl(int n_families)
2081 {
2082         int i;
2083
2084         for (i = 0; i < n_families; i++)
2085                 genl_unregister_family(dp_genl_families[i]);
2086 }
2087
2088 static int dp_register_genl(void)
2089 {
2090         int err;
2091         int i;
2092
2093         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2094
2095                 err = genl_register_family(dp_genl_families[i]);
2096                 if (err)
2097                         goto error;
2098         }
2099
2100         return 0;
2101
2102 error:
2103         dp_unregister_genl(i);
2104         return err;
2105 }
2106
2107 static int __net_init ovs_init_net(struct net *net)
2108 {
2109         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2110
2111         INIT_LIST_HEAD(&ovs_net->dps);
2112         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2113         return 0;
2114 }
2115
2116 static void __net_exit ovs_exit_net(struct net *net)
2117 {
2118         struct datapath *dp, *dp_next;
2119         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2120
2121         ovs_lock();
2122         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2123                 __dp_destroy(dp);
2124         ovs_unlock();
2125
2126         cancel_work_sync(&ovs_net->dp_notify_work);
2127 }
2128
2129 static struct pernet_operations ovs_net_ops = {
2130         .init = ovs_init_net,
2131         .exit = ovs_exit_net,
2132         .id   = &ovs_net_id,
2133         .size = sizeof(struct ovs_net),
2134 };
2135
2136 DEFINE_COMPAT_PNET_REG_FUNC(device);
2137
2138 static int __init dp_init(void)
2139 {
2140         int err;
2141
2142         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2143
2144         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2145                 VERSION);
2146
2147         err = action_fifos_init();
2148         if (err)
2149                 goto error;
2150
2151         err = ovs_flow_init();
2152         if (err)
2153                 goto error_action_fifos_exit;
2154
2155         err = ovs_vport_init();
2156         if (err)
2157                 goto error_flow_exit;
2158
2159         err = register_pernet_device(&ovs_net_ops);
2160         if (err)
2161                 goto error_vport_exit;
2162
2163         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2164         if (err)
2165                 goto error_netns_exit;
2166
2167         err = dp_register_genl();
2168         if (err < 0)
2169                 goto error_unreg_notifier;
2170
2171         return 0;
2172
2173 error_unreg_notifier:
2174         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2175 error_netns_exit:
2176         unregister_pernet_device(&ovs_net_ops);
2177 error_vport_exit:
2178         ovs_vport_exit();
2179 error_flow_exit:
2180         ovs_flow_exit();
2181 error_action_fifos_exit:
2182         action_fifos_exit();
2183 error:
2184         return err;
2185 }
2186
2187 static void dp_cleanup(void)
2188 {
2189         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2190         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2191         unregister_pernet_device(&ovs_net_ops);
2192         rcu_barrier();
2193         ovs_vport_exit();
2194         ovs_flow_exit();
2195         action_fifos_exit();
2196 }
2197
2198 module_init(dp_init);
2199 module_exit(dp_cleanup);
2200
2201 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2202 MODULE_LICENSE("GPL");
2203 MODULE_VERSION(VERSION);