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
  63 static struct genl_family dp_packet_genl_family;
  64 static struct genl_family dp_flow_genl_family;
  65 static struct genl_family dp_datapath_genl_family;
  66
  67 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
  68         .name = OVS_FLOW_MCGROUP,
  69 };
  70
  71 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  72         .name = OVS_DATAPATH_MCGROUP,
  73 };
  74
  75 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
  76         .name = OVS_VPORT_MCGROUP,
  77 };
  78
  79 /* Check if need to build a reply message.
  80  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  81 static bool ovs_must_notify(struct genl_info *info,
  82                             const struct genl_multicast_group *grp)
  83 {
  84         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  85                 netlink_has_listeners(genl_info_net(info)->genl_sock, 0);
  86 }
  87
  88 static void ovs_notify(struct genl_family *family,
  89                        struct sk_buff *skb, struct genl_info *info)
  90 {
  91         genl_notify(family, skb, genl_info_net(info), info->snd_portid,
  92                     0, info->nlhdr, GFP_KERNEL);
  93 }
  94
  95 /**
  96  * DOC: Locking:
  97  *
  98  * All writes e.g. Writes to device state (add/remove datapath, port, set
  99  * operations on vports, etc.), Writes to other state (flow table
 100  * modifications, set miscellaneous datapath parameters, etc.) are protected
 101  * by ovs_lock.
 102  *
 103  * Reads are protected by RCU.
 104  *
 105  * There are a few special cases (mostly stats) that have their own
 106  * synchronization but they nest under all of above and don't interact with
 107  * each other.
 108  *
 109  * The RTNL lock nests inside ovs_mutex.
 110  */
 111
 112 static DEFINE_MUTEX(ovs_mutex);
 113
 114 void ovs_lock(void)
 115 {
 116         mutex_lock(&ovs_mutex);
 117 }
 118
 119 void ovs_unlock(void)
 120 {
 121         mutex_unlock(&ovs_mutex);
 122 }
 123
 124 #ifdef CONFIG_LOCKDEP
 125 int lockdep_ovsl_is_held(void)
 126 {
 127         if (debug_locks)
 128                 return lockdep_is_held(&ovs_mutex);
 129         else
 130                 return 1;
 131 }
 132 #endif
 133
 134 static struct vport *new_vport(const struct vport_parms *);
 135 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 136                              const struct dp_upcall_info *);
 137 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 138                                   const struct dp_upcall_info *);
 139
 140 /* Must be called with rcu_read_lock or ovs_mutex. */
 141 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 142 {
 143         struct datapath *dp = NULL;
 144         struct net_device *dev;
 145
 146         rcu_read_lock();
 147         dev = dev_get_by_index_rcu(net, dp_ifindex);
 148         if (dev) {
 149                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 150                 if (vport)
 151                         dp = vport->dp;
 152         }
 153         rcu_read_unlock();
 154
 155         return dp;
 156 }
 157
 158 /* Must be called with rcu_read_lock or ovs_mutex. */
 159 static const char *ovs_dp_name(const struct datapath *dp)
 160 {
 161         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 162         return vport->ops->get_name(vport);
 163 }
 164
 165 static int get_dpifindex(struct datapath *dp)
 166 {
 167         struct vport *local;
 168         int ifindex;
 169
 170         rcu_read_lock();
 171
 172         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 173         if (local)
 174                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 175         else
 176                 ifindex = 0;
 177
 178         rcu_read_unlock();
 179
 180         return ifindex;
 181 }
 182
 183 static void destroy_dp_rcu(struct rcu_head *rcu)
 184 {
 185         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 186
 187         free_percpu(dp->stats_percpu);
 188         release_net(ovs_dp_get_net(dp));
 189         kfree(dp->ports);
 190         kfree(dp);
 191 }
 192
 193 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 194                                             u16 port_no)
 195 {
 196         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 197 }
 198
 199 /* Called with ovs_mutex or RCU read lock. */
 200 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 201 {
 202         struct vport *vport;
 203         struct hlist_head *head;
 204
 205         head = vport_hash_bucket(dp, port_no);
 206         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 207                 if (vport->port_no == port_no)
 208                         return vport;
 209         }
 210         return NULL;
 211 }
 212
 213 /* Called with ovs_mutex. */
 214 static struct vport *new_vport(const struct vport_parms *parms)
 215 {
 216         struct vport *vport;
 217
 218         vport = ovs_vport_add(parms);
 219         if (!IS_ERR(vport)) {
 220                 struct datapath *dp = parms->dp;
 221                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 222
 223                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 224         }
 225         return vport;
 226 }
 227
 228 void ovs_dp_detach_port(struct vport *p)
 229 {
 230         ASSERT_OVSL();
 231
 232         /* First drop references to device. */
 233         hlist_del_rcu(&p->dp_hash_node);
 234
 235         /* Then destroy it. */
 236         ovs_vport_del(p);
 237 }
 238
 239 /* Must be called with rcu_read_lock. */
 240 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 241 {
 242         struct datapath *dp = p->dp;
 243         struct sw_flow *flow;
 244         struct dp_stats_percpu *stats;
 245         struct sw_flow_key key;
 246         u64 *stats_counter;
 247         u32 n_mask_hit;
 248         int error;
 249
 250         stats = this_cpu_ptr(dp->stats_percpu);
 251
 252         /* Extract flow from 'skb' into 'key'. */
 253         error = ovs_flow_extract(skb, p->port_no, &key);
 254         if (unlikely(error)) {
 255                 kfree_skb(skb);
 256                 return;
 257         }
 258
 259         /* Look up flow. */
 260         flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
 261         if (unlikely(!flow)) {
 262                 struct dp_upcall_info upcall;
 263
 264                 upcall.cmd = OVS_PACKET_CMD_MISS;
 265                 upcall.key = &key;
 266                 upcall.userdata = NULL;
 267                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 268                 error = ovs_dp_upcall(dp, skb, &upcall);
 269                 if (unlikely(error))
 270                         kfree_skb(skb);
 271                 else
 272                         consume_skb(skb);
 273                 stats_counter = &stats->n_missed;
 274                 goto out;
 275         }
 276
 277         OVS_CB(skb)->flow = flow;
 278         OVS_CB(skb)->pkt_key = &key;
 279
 280         ovs_flow_stats_update(OVS_CB(skb)->flow, key.tp.flags, skb);
 281         ovs_execute_actions(dp, skb);
 282         stats_counter = &stats->n_hit;
 283
 284 out:
 285         /* Update datapath statistics. */
 286         u64_stats_update_begin(&stats->syncp);
 287         (*stats_counter)++;
 288         stats->n_mask_hit += n_mask_hit;
 289         u64_stats_update_end(&stats->syncp);
 290 }
 291
 292 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 293                   const struct dp_upcall_info *upcall_info)
 294 {
 295         struct dp_stats_percpu *stats;
 296         int err;
 297
 298         if (upcall_info->portid == 0) {
 299                 err = -ENOTCONN;
 300                 goto err;
 301         }
 302
 303         if (!skb_is_gso(skb))
 304                 err = queue_userspace_packet(dp, skb, upcall_info);
 305         else
 306                 err = queue_gso_packets(dp, skb, upcall_info);
 307         if (err)
 308                 goto err;
 309
 310         return 0;
 311
 312 err:
 313         stats = this_cpu_ptr(dp->stats_percpu);
 314
 315         u64_stats_update_begin(&stats->syncp);
 316         stats->n_lost++;
 317         u64_stats_update_end(&stats->syncp);
 318
 319         return err;
 320 }
 321
 322 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 323                              const struct dp_upcall_info *upcall_info)
 324 {
 325         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 326         struct dp_upcall_info later_info;
 327         struct sw_flow_key later_key;
 328         struct sk_buff *segs, *nskb;
 329         int err;
 330
 331         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 332         if (IS_ERR(segs))
 333                 return PTR_ERR(segs);
 334
 335         /* Queue all of the segments. */
 336         skb = segs;
 337         do {
 338                 err = queue_userspace_packet(dp, skb, upcall_info);
 339                 if (err)
 340                         break;
 341
 342                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 343                         /* The initial flow key extracted by ovs_flow_extract()
 344                          * in this case is for a first fragment, so we need to
 345                          * properly mark later fragments.
 346                          */
 347                         later_key = *upcall_info->key;
 348                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 349
 350                         later_info = *upcall_info;
 351                         later_info.key = &later_key;
 352                         upcall_info = &later_info;
 353                 }
 354         } while ((skb = skb->next));
 355
 356         /* Free all of the segments. */
 357         skb = segs;
 358         do {
 359                 nskb = skb->next;
 360                 if (err)
 361                         kfree_skb(skb);
 362                 else
 363                         consume_skb(skb);
 364         } while ((skb = nskb));
 365         return err;
 366 }
 367
 368 static size_t key_attr_size(void)
 369 {
 370         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 371                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 372                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
 373                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 374                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 375                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
 376                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
 377                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 378                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
 379                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 380                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 381                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 382                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 383                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 384                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 385                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 386                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 387                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 388                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 389 }
 390
 391 static size_t upcall_msg_size(const struct nlattr *userdata,
 392                               unsigned int hdrlen)
 393 {
 394         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 395                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 396                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 397
 398         /* OVS_PACKET_ATTR_USERDATA */
 399         if (userdata)
 400                 size += NLA_ALIGN(userdata->nla_len);
 401
 402         return size;
 403 }
 404
 405 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 406                                   const struct dp_upcall_info *upcall_info)
 407 {
 408         struct ovs_header *upcall;
 409         struct sk_buff *nskb = NULL;
 410         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 411         struct nlattr *nla;
 412         struct genl_info info = {
 413                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 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                 nskb->vlan_tci = 0;
 434                 skb = nskb;
 435         }
 436
 437         if (nla_attr_size(skb->len) > USHRT_MAX) {
 438                 err = -EFBIG;
 439                 goto out;
 440         }
 441
 442         /* Complete checksum if needed */
 443         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 444             (err = skb_checksum_help(skb)))
 445                 goto out;
 446
 447         /* Older versions of OVS user space enforce alignment of the last
 448          * Netlink attribute to NLA_ALIGNTO which would require extensive
 449          * padding logic. Only perform zerocopy if padding is not required.
 450          */
 451         if (dp->user_features & OVS_DP_F_UNALIGNED)
 452                 hlen = skb_zerocopy_headlen(skb);
 453         else
 454                 hlen = skb->len;
 455
 456         len = upcall_msg_size(upcall_info->userdata, hlen);
 457         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 458         if (!user_skb) {
 459                 err = -ENOMEM;
 460                 goto out;
 461         }
 462
 463         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 464                              0, upcall_info->cmd);
 465         upcall->dp_ifindex = dp_ifindex;
 466
 467         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 468         err = ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
 469         BUG_ON(err);
 470         nla_nest_end(user_skb, nla);
 471
 472         if (upcall_info->userdata)
 473                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 474                           nla_len(upcall_info->userdata),
 475                           nla_data(upcall_info->userdata));
 476
 477         /* Only reserve room for attribute header, packet data is added
 478          * in skb_zerocopy() */
 479         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 480                 err = -ENOBUFS;
 481                 goto out;
 482         }
 483         nla->nla_len = nla_attr_size(skb->len);
 484
 485         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 486         if (err)
 487                 goto out;
 488
 489         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 490         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 491                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 492
 493                 if (plen > 0)
 494                         memset(skb_put(user_skb, plen), 0, plen);
 495         }
 496
 497         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 498
 499         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 500         user_skb = NULL;
 501 out:
 502         if (err)
 503                 skb_tx_error(skb);
 504         kfree_skb(user_skb);
 505         kfree_skb(nskb);
 506         return err;
 507 }
 508
 509 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 510 {
 511         struct ovs_header *ovs_header = info->userhdr;
 512         struct nlattr **a = info->attrs;
 513         struct sw_flow_actions *acts;
 514         struct sk_buff *packet;
 515         struct sw_flow *flow;
 516         struct datapath *dp;
 517         struct ethhdr *eth;
 518         int len;
 519         int err;
 520
 521         err = -EINVAL;
 522         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 523             !a[OVS_PACKET_ATTR_ACTIONS])
 524                 goto err;
 525
 526         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 527         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 528         err = -ENOMEM;
 529         if (!packet)
 530                 goto err;
 531         skb_reserve(packet, NET_IP_ALIGN);
 532
 533         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 534
 535         skb_reset_mac_header(packet);
 536         eth = eth_hdr(packet);
 537
 538         /* Normally, setting the skb 'protocol' field would be handled by a
 539          * call to eth_type_trans(), but it assumes there's a sending
 540          * device, which we may not have. */
 541         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 542                 packet->protocol = eth->h_proto;
 543         else
 544                 packet->protocol = htons(ETH_P_802_2);
 545
 546         /* Build an sw_flow for sending this packet. */
 547         flow = ovs_flow_alloc();
 548         err = PTR_ERR(flow);
 549         if (IS_ERR(flow))
 550                 goto err_kfree_skb;
 551
 552         err = ovs_flow_extract(packet, -1, &flow->key);
 553         if (err)
 554                 goto err_flow_free;
 555
 556         err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
 557         if (err)
 558                 goto err_flow_free;
 559         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 560         err = PTR_ERR(acts);
 561         if (IS_ERR(acts))
 562                 goto err_flow_free;
 563
 564         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 565                                    &flow->key, 0, &acts);
 566         rcu_assign_pointer(flow->sf_acts, acts);
 567         if (err)
 568                 goto err_flow_free;
 569
 570         OVS_CB(packet)->flow = flow;
 571         OVS_CB(packet)->pkt_key = &flow->key;
 572         packet->priority = flow->key.phy.priority;
 573         packet->mark = flow->key.phy.skb_mark;
 574
 575         rcu_read_lock();
 576         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 577         err = -ENODEV;
 578         if (!dp)
 579                 goto err_unlock;
 580
 581         local_bh_disable();
 582         err = ovs_execute_actions(dp, packet);
 583         local_bh_enable();
 584         rcu_read_unlock();
 585
 586         ovs_flow_free(flow, false);
 587         return err;
 588
 589 err_unlock:
 590         rcu_read_unlock();
 591 err_flow_free:
 592         ovs_flow_free(flow, false);
 593 err_kfree_skb:
 594         kfree_skb(packet);
 595 err:
 596         return err;
 597 }
 598
 599 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 600         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 601         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 602         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 603 };
 604
 605 static const struct genl_ops dp_packet_genl_ops[] = {
 606         { .cmd = OVS_PACKET_CMD_EXECUTE,
 607           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 608           .policy = packet_policy,
 609           .doit = ovs_packet_cmd_execute
 610         }
 611 };
 612
 613 static struct genl_family dp_packet_genl_family = {
 614         .id = GENL_ID_GENERATE,
 615         .hdrsize = sizeof(struct ovs_header),
 616         .name = OVS_PACKET_FAMILY,
 617         .version = OVS_PACKET_VERSION,
 618         .maxattr = OVS_PACKET_ATTR_MAX,
 619         .netnsok = true,
 620         .parallel_ops = true,
 621         .ops = dp_packet_genl_ops,
 622         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 623 };
 624
 625 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
 626                          struct ovs_dp_megaflow_stats *mega_stats)
 627 {
 628         int i;
 629
 630         memset(mega_stats, 0, sizeof(*mega_stats));
 631
 632         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 633         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 634
 635         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 636
 637         for_each_possible_cpu(i) {
 638                 const struct dp_stats_percpu *percpu_stats;
 639                 struct dp_stats_percpu local_stats;
 640                 unsigned int start;
 641
 642                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 643
 644                 do {
 645                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 646                         local_stats = *percpu_stats;
 647                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 648
 649                 stats->n_hit += local_stats.n_hit;
 650                 stats->n_missed += local_stats.n_missed;
 651                 stats->n_lost += local_stats.n_lost;
 652                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 653         }
 654 }
 655
 656 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 657 {
 658         return NLMSG_ALIGN(sizeof(struct ovs_header))
 659                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 660                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 661                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 662                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 663                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 664                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 665 }
 666
 667 /* Called with ovs_mutex or RCU read lock. */
 668 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 669                                   struct sk_buff *skb, u32 portid,
 670                                   u32 seq, u32 flags, u8 cmd)
 671 {
 672         const int skb_orig_len = skb->len;
 673         struct nlattr *start;
 674         struct ovs_flow_stats stats;
 675         __be16 tcp_flags;
 676         unsigned long used;
 677         struct ovs_header *ovs_header;
 678         struct nlattr *nla;
 679         int err;
 680
 681         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 682         if (!ovs_header)
 683                 return -EMSGSIZE;
 684
 685         ovs_header->dp_ifindex = dp_ifindex;
 686
 687         /* Fill flow key. */
 688         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 689         if (!nla)
 690                 goto nla_put_failure;
 691
 692         err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
 693         if (err)
 694                 goto error;
 695         nla_nest_end(skb, nla);
 696
 697         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 698         if (!nla)
 699                 goto nla_put_failure;
 700
 701         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 702         if (err)
 703                 goto error;
 704
 705         nla_nest_end(skb, nla);
 706
 707         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 708
 709         if (used &&
 710             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 711                 goto nla_put_failure;
 712
 713         if (stats.n_packets &&
 714             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 715                 goto nla_put_failure;
 716
 717         if ((u8)ntohs(tcp_flags) &&
 718              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 719                 goto nla_put_failure;
 720
 721         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 722          * this is the first flow to be dumped into 'skb'.  This is unusual for
 723          * Netlink but individual action lists can be longer than
 724          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 725          * The userspace caller can always fetch the actions separately if it
 726          * really wants them.  (Most userspace callers in fact don't care.)
 727          *
 728          * This can only fail for dump operations because the skb is always
 729          * properly sized for single flows.
 730          */
 731         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 732         if (start) {
 733                 const struct sw_flow_actions *sf_acts;
 734
 735                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 736                 err = ovs_nla_put_actions(sf_acts->actions,
 737                                           sf_acts->actions_len, skb);
 738
 739                 if (!err)
 740                         nla_nest_end(skb, start);
 741                 else {
 742                         if (skb_orig_len)
 743                                 goto error;
 744
 745                         nla_nest_cancel(skb, start);
 746                 }
 747         } else if (skb_orig_len)
 748                 goto nla_put_failure;
 749
 750         return genlmsg_end(skb, ovs_header);
 751
 752 nla_put_failure:
 753         err = -EMSGSIZE;
 754 error:
 755         genlmsg_cancel(skb, ovs_header);
 756         return err;
 757 }
 758
 759 /* May not be called with RCU read lock. */
 760 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 761                                                struct genl_info *info,
 762                                                bool always)
 763 {
 764         struct sk_buff *skb;
 765
 766         if (!always && !ovs_must_notify(info, &ovs_dp_flow_multicast_group))
 767                 return NULL;
 768
 769         skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
 770         if (!skb)
 771                 return ERR_PTR(-ENOMEM);
 772
 773         return skb;
 774 }
 775
 776 /* Called with ovs_mutex. */
 777 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 778                                                int dp_ifindex,
 779                                                struct genl_info *info, u8 cmd,
 780                                                bool always)
 781 {
 782         struct sk_buff *skb;
 783         int retval;
 784
 785         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
 786                                       always);
 787         if (IS_ERR_OR_NULL(skb))
 788                 return skb;
 789
 790         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 791                                         info->snd_portid, info->snd_seq, 0,
 792                                         cmd);
 793         BUG_ON(retval < 0);
 794         return skb;
 795 }
 796
 797 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 798 {
 799         struct nlattr **a = info->attrs;
 800         struct ovs_header *ovs_header = info->userhdr;
 801         struct sw_flow *flow, *new_flow;
 802         struct sw_flow_mask mask;
 803         struct sk_buff *reply;
 804         struct datapath *dp;
 805         struct sw_flow_actions *acts;
 806         struct sw_flow_match match;
 807         int error;
 808
 809         /* Must have key and actions. */
 810         error = -EINVAL;
 811         if (!a[OVS_FLOW_ATTR_KEY])
 812                 goto error;
 813         if (!a[OVS_FLOW_ATTR_ACTIONS])
 814                 goto error;
 815
 816         /* Most of the time we need to allocate a new flow, do it before
 817          * locking.
 818          */
 819         new_flow = ovs_flow_alloc();
 820         if (IS_ERR(new_flow)) {
 821                 error = PTR_ERR(new_flow);
 822                 goto error;
 823         }
 824
 825         /* Extract key. */
 826         ovs_match_init(&match, &new_flow->unmasked_key, &mask);
 827         error = ovs_nla_get_match(&match,
 828                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 829         if (error)
 830                 goto err_kfree_flow;
 831
 832         ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
 833
 834         /* Validate actions. */
 835         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 836         error = PTR_ERR(acts);
 837         if (IS_ERR(acts))
 838                 goto err_kfree_flow;
 839
 840         error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
 841                                      0, &acts);
 842         if (error) {
 843                 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 844                 goto err_kfree_acts;
 845         }
 846
 847         reply = ovs_flow_cmd_alloc_info(acts, info, false);
 848         if (IS_ERR(reply)) {
 849                 error = PTR_ERR(reply);
 850                 goto err_kfree_acts;
 851         }
 852
 853         ovs_lock();
 854         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 855         if (unlikely(!dp)) {
 856                 error = -ENODEV;
 857                 goto err_unlock_ovs;
 858         }
 859         /* Check if this is a duplicate flow */
 860         flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
 861         if (likely(!flow)) {
 862                 rcu_assign_pointer(new_flow->sf_acts, acts);
 863
 864                 /* Put flow in bucket. */
 865                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 866                 if (unlikely(error)) {
 867                         acts = NULL;
 868                         goto err_unlock_ovs;
 869                 }
 870
 871                 if (unlikely(reply)) {
 872                         error = ovs_flow_cmd_fill_info(new_flow,
 873                                                        ovs_header->dp_ifindex,
 874                                                        reply, info->snd_portid,
 875                                                        info->snd_seq, 0,
 876                                                        OVS_FLOW_CMD_NEW);
 877                         BUG_ON(error < 0);
 878                 }
 879                 ovs_unlock();
 880         } else {
 881                 struct sw_flow_actions *old_acts;
 882
 883                 /* Bail out if we're not allowed to modify an existing flow.
 884                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 885                  * because Generic Netlink treats the latter as a dump
 886                  * request.  We also accept NLM_F_EXCL in case that bug ever
 887                  * gets fixed.
 888                  */
 889                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
 890                                                          | NLM_F_EXCL))) {
 891                         error = -EEXIST;
 892                         goto err_unlock_ovs;
 893                 }
 894                 /* The unmasked key has to be the same for flow updates. */
 895                 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
 896                         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
 897                         if (!flow) {
 898                                 error = -ENOENT;
 899                                 goto err_unlock_ovs;
 900                         }
 901                 }
 902                 /* Update actions. */
 903                 old_acts = ovsl_dereference(flow->sf_acts);
 904                 rcu_assign_pointer(flow->sf_acts, acts);
 905
 906                 if (unlikely(reply)) {
 907                         error = ovs_flow_cmd_fill_info(flow,
 908                                                        ovs_header->dp_ifindex,
 909                                                        reply, info->snd_portid,
 910                                                        info->snd_seq, 0,
 911                                                        OVS_FLOW_CMD_NEW);
 912                         BUG_ON(error < 0);
 913                 }
 914                 ovs_unlock();
 915
 916                 ovs_nla_free_flow_actions(old_acts);
 917                 ovs_flow_free(new_flow, false);
 918         }
 919
 920         if (reply)
 921                 ovs_notify(&dp_flow_genl_family, reply, info);
 922         return 0;
 923
 924 err_unlock_ovs:
 925         ovs_unlock();
 926         kfree_skb(reply);
 927 err_kfree_acts:
 928         kfree(acts);
 929 err_kfree_flow:
 930         ovs_flow_free(new_flow, false);
 931 error:
 932         return error;
 933 }
 934
 935 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
 936 {
 937         struct nlattr **a = info->attrs;
 938         struct ovs_header *ovs_header = info->userhdr;
 939         struct sw_flow_key key, masked_key;
 940         struct sw_flow *flow;
 941         struct sw_flow_mask mask;
 942         struct sk_buff *reply = NULL;
 943         struct datapath *dp;
 944         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
 945         struct sw_flow_match match;
 946         int error;
 947
 948         /* Extract key. */
 949         error = -EINVAL;
 950         if (!a[OVS_FLOW_ATTR_KEY])
 951                 goto error;
 952
 953         ovs_match_init(&match, &key, &mask);
 954         error = ovs_nla_get_match(&match,
 955                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 956         if (error)
 957                 goto error;
 958
 959         /* Validate actions. */
 960         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 961                 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 962                 error = PTR_ERR(acts);
 963                 if (IS_ERR(acts))
 964                         goto error;
 965
 966                 ovs_flow_mask_key(&masked_key, &key, &mask);
 967                 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
 968                                              &masked_key, 0, &acts);
 969                 if (error) {
 970                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 971                         goto err_kfree_acts;
 972                 }
 973         }
 974
 975         /* Can allocate before locking if have acts. */
 976         if (acts) {
 977                 reply = ovs_flow_cmd_alloc_info(acts, info, false);
 978                 if (IS_ERR(reply)) {
 979                         error = PTR_ERR(reply);
 980                         goto err_kfree_acts;
 981                 }
 982         }
 983
 984         ovs_lock();
 985         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 986         if (unlikely(!dp)) {
 987                 error = -ENODEV;
 988                 goto err_unlock_ovs;
 989         }
 990         /* Check that the flow exists. */
 991         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
 992         if (unlikely(!flow)) {
 993                 error = -ENOENT;
 994                 goto err_unlock_ovs;
 995         }
 996
 997         /* Update actions, if present. */
 998         if (likely(acts)) {
 999                 old_acts = ovsl_dereference(flow->sf_acts);
1000                 rcu_assign_pointer(flow->sf_acts, acts);
1001
1002                 if (unlikely(reply)) {
1003                         error = ovs_flow_cmd_fill_info(flow,
1004                                                        ovs_header->dp_ifindex,
1005                                                        reply, info->snd_portid,
1006                                                        info->snd_seq, 0,
1007                                                        OVS_FLOW_CMD_NEW);
1008                         BUG_ON(error < 0);
1009                 }
1010         } else {
1011                 /* Could not alloc without acts before locking. */
1012                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1013                                                 info, OVS_FLOW_CMD_NEW, false);
1014                 if (unlikely(IS_ERR(reply))) {
1015                         error = PTR_ERR(reply);
1016                         goto err_unlock_ovs;
1017                 }
1018         }
1019
1020         /* Clear stats. */
1021         if (a[OVS_FLOW_ATTR_CLEAR])
1022                 ovs_flow_stats_clear(flow);
1023         ovs_unlock();
1024
1025         if (reply)
1026                 ovs_notify(&dp_flow_genl_family, reply, info);
1027         if (old_acts)
1028                 ovs_nla_free_flow_actions(old_acts);
1029
1030         return 0;
1031
1032 err_unlock_ovs:
1033         ovs_unlock();
1034         kfree_skb(reply);
1035 err_kfree_acts:
1036         kfree(acts);
1037 error:
1038         return error;
1039 }
1040
1041 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1042 {
1043         struct nlattr **a = info->attrs;
1044         struct ovs_header *ovs_header = info->userhdr;
1045         struct sw_flow_key key;
1046         struct sk_buff *reply;
1047         struct sw_flow *flow;
1048         struct datapath *dp;
1049         struct sw_flow_match match;
1050         int err;
1051
1052         if (!a[OVS_FLOW_ATTR_KEY]) {
1053                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1054                 return -EINVAL;
1055         }
1056
1057         ovs_match_init(&match, &key, NULL);
1058         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1059         if (err)
1060                 return err;
1061
1062         ovs_lock();
1063         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1064         if (!dp) {
1065                 err = -ENODEV;
1066                 goto unlock;
1067         }
1068
1069         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1070         if (!flow) {
1071                 err = -ENOENT;
1072                 goto unlock;
1073         }
1074
1075         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1076                                         OVS_FLOW_CMD_NEW, true);
1077         if (IS_ERR(reply)) {
1078                 err = PTR_ERR(reply);
1079                 goto unlock;
1080         }
1081
1082         ovs_unlock();
1083         return genlmsg_reply(reply, info);
1084 unlock:
1085         ovs_unlock();
1086         return err;
1087 }
1088
1089 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1090 {
1091         struct nlattr **a = info->attrs;
1092         struct ovs_header *ovs_header = info->userhdr;
1093         struct sw_flow_key key;
1094         struct sk_buff *reply;
1095         struct sw_flow *flow;
1096         struct datapath *dp;
1097         struct sw_flow_match match;
1098         int err;
1099
1100         if (likely(a[OVS_FLOW_ATTR_KEY])) {
1101                 ovs_match_init(&match, &key, NULL);
1102                 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1103                 if (unlikely(err))
1104                         return err;
1105         }
1106
1107         ovs_lock();
1108         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1109         if (unlikely(!dp)) {
1110                 err = -ENODEV;
1111                 goto unlock;
1112         }
1113
1114         if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1115                 err = ovs_flow_tbl_flush(&dp->table);
1116                 goto unlock;
1117         }
1118
1119         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1120         if (unlikely(!flow)) {
1121                 err = -ENOENT;
1122                 goto unlock;
1123         }
1124
1125         ovs_flow_tbl_remove(&dp->table, flow);
1126         ovs_unlock();
1127
1128         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1129                                         info, false);
1130         if (likely(reply)) {
1131                 if (likely(!IS_ERR(reply))) {
1132                         rcu_read_lock();        /*To keep RCU checker happy. */
1133                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1134                                                      reply, info->snd_portid,
1135                                                      info->snd_seq, 0,
1136                                                      OVS_FLOW_CMD_DEL);
1137                         rcu_read_unlock();
1138                         BUG_ON(err < 0);
1139
1140                         ovs_notify(&dp_flow_genl_family, reply, info);
1141                 } else {
1142                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1143                 }
1144         }
1145
1146         ovs_flow_free(flow, true);
1147         return 0;
1148 unlock:
1149         ovs_unlock();
1150         return err;
1151 }
1152
1153 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1154 {
1155         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1156         struct table_instance *ti;
1157         struct datapath *dp;
1158
1159         rcu_read_lock();
1160         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1161         if (!dp) {
1162                 rcu_read_unlock();
1163                 return -ENODEV;
1164         }
1165
1166         ti = rcu_dereference(dp->table.ti);
1167         for (;;) {
1168                 struct sw_flow *flow;
1169                 u32 bucket, obj;
1170
1171                 bucket = cb->args[0];
1172                 obj = cb->args[1];
1173                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1174                 if (!flow)
1175                         break;
1176
1177                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1178                                            NETLINK_CB(cb->skb).portid,
1179                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1180                                            OVS_FLOW_CMD_NEW) < 0)
1181                         break;
1182
1183                 cb->args[0] = bucket;
1184                 cb->args[1] = obj;
1185         }
1186         rcu_read_unlock();
1187         return skb->len;
1188 }
1189
1190 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1191         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1192         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1193         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1194 };
1195
1196 static const struct genl_ops dp_flow_genl_ops[] = {
1197         { .cmd = OVS_FLOW_CMD_NEW,
1198           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1199           .policy = flow_policy,
1200           .doit = ovs_flow_cmd_new
1201         },
1202         { .cmd = OVS_FLOW_CMD_DEL,
1203           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1204           .policy = flow_policy,
1205           .doit = ovs_flow_cmd_del
1206         },
1207         { .cmd = OVS_FLOW_CMD_GET,
1208           .flags = 0,               /* OK for unprivileged users. */
1209           .policy = flow_policy,
1210           .doit = ovs_flow_cmd_get,
1211           .dumpit = ovs_flow_cmd_dump
1212         },
1213         { .cmd = OVS_FLOW_CMD_SET,
1214           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1215           .policy = flow_policy,
1216           .doit = ovs_flow_cmd_set,
1217         },
1218 };
1219
1220 static struct genl_family dp_flow_genl_family = {
1221         .id = GENL_ID_GENERATE,
1222         .hdrsize = sizeof(struct ovs_header),
1223         .name = OVS_FLOW_FAMILY,
1224         .version = OVS_FLOW_VERSION,
1225         .maxattr = OVS_FLOW_ATTR_MAX,
1226         .netnsok = true,
1227         .parallel_ops = true,
1228         .ops = dp_flow_genl_ops,
1229         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1230         .mcgrps = &ovs_dp_flow_multicast_group,
1231         .n_mcgrps = 1,
1232 };
1233
1234 static size_t ovs_dp_cmd_msg_size(void)
1235 {
1236         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1237
1238         msgsize += nla_total_size(IFNAMSIZ);
1239         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1240         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1241         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1242
1243         return msgsize;
1244 }
1245
1246 /* Called with ovs_mutex or RCU read lock. */
1247 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1248                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1249 {
1250         struct ovs_header *ovs_header;
1251         struct ovs_dp_stats dp_stats;
1252         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1253         int err;
1254
1255         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1256                                    flags, cmd);
1257         if (!ovs_header)
1258                 goto error;
1259
1260         ovs_header->dp_ifindex = get_dpifindex(dp);
1261
1262         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1263         if (err)
1264                 goto nla_put_failure;
1265
1266         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1267         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1268                         &dp_stats))
1269                 goto nla_put_failure;
1270
1271         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1272                         sizeof(struct ovs_dp_megaflow_stats),
1273                         &dp_megaflow_stats))
1274                 goto nla_put_failure;
1275
1276         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1277                 goto nla_put_failure;
1278
1279         return genlmsg_end(skb, ovs_header);
1280
1281 nla_put_failure:
1282         genlmsg_cancel(skb, ovs_header);
1283 error:
1284         return -EMSGSIZE;
1285 }
1286
1287 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1288 {
1289         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1290 }
1291
1292 /* Called with rcu_read_lock or ovs_mutex. */
1293 static struct datapath *lookup_datapath(struct net *net,
1294                                         struct ovs_header *ovs_header,
1295                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1296 {
1297         struct datapath *dp;
1298
1299         if (!a[OVS_DP_ATTR_NAME])
1300                 dp = get_dp(net, ovs_header->dp_ifindex);
1301         else {
1302                 struct vport *vport;
1303
1304                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1305                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1306         }
1307         return dp ? dp : ERR_PTR(-ENODEV);
1308 }
1309
1310 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1311 {
1312         struct datapath *dp;
1313
1314         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1315         if (IS_ERR(dp))
1316                 return;
1317
1318         WARN(dp->user_features, "Dropping previously announced user features\n");
1319         dp->user_features = 0;
1320 }
1321
1322 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1323 {
1324         if (a[OVS_DP_ATTR_USER_FEATURES])
1325                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1326 }
1327
1328 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1329 {
1330         struct nlattr **a = info->attrs;
1331         struct vport_parms parms;
1332         struct sk_buff *reply;
1333         struct datapath *dp;
1334         struct vport *vport;
1335         struct ovs_net *ovs_net;
1336         int err, i;
1337
1338         err = -EINVAL;
1339         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1340                 goto err;
1341
1342         reply = ovs_dp_cmd_alloc_info(info);
1343         if (!reply)
1344                 return -ENOMEM;
1345
1346         err = -ENOMEM;
1347         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1348         if (dp == NULL)
1349                 goto err_free_reply;
1350
1351         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1352
1353         /* Allocate table. */
1354         err = ovs_flow_tbl_init(&dp->table);
1355         if (err)
1356                 goto err_free_dp;
1357
1358         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1359         if (!dp->stats_percpu) {
1360                 err = -ENOMEM;
1361                 goto err_destroy_table;
1362         }
1363
1364         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1365                             GFP_KERNEL);
1366         if (!dp->ports) {
1367                 err = -ENOMEM;
1368                 goto err_destroy_percpu;
1369         }
1370
1371         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1372                 INIT_HLIST_HEAD(&dp->ports[i]);
1373
1374         /* Set up our datapath device. */
1375         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1376         parms.type = OVS_VPORT_TYPE_INTERNAL;
1377         parms.options = NULL;
1378         parms.dp = dp;
1379         parms.port_no = OVSP_LOCAL;
1380         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1381
1382         ovs_dp_change(dp, a);
1383
1384         /* So far only local changes have been made, now need the lock. */
1385         ovs_lock();
1386
1387         vport = new_vport(&parms);
1388         if (IS_ERR(vport)) {
1389                 err = PTR_ERR(vport);
1390                 if (err == -EBUSY)
1391                         err = -EEXIST;
1392
1393                 if (err == -EEXIST) {
1394                         /* An outdated user space instance that does not understand
1395                          * the concept of user_features has attempted to create a new
1396                          * datapath and is likely to reuse it. Drop all user features.
1397                          */
1398                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1399                                 ovs_dp_reset_user_features(skb, info);
1400                 }
1401
1402                 goto err_destroy_ports_array;
1403         }
1404
1405         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1406                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1407         BUG_ON(err < 0);
1408
1409         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1410         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1411
1412         ovs_unlock();
1413
1414         ovs_notify(&dp_datapath_genl_family, reply, info);
1415         return 0;
1416
1417 err_destroy_ports_array:
1418         ovs_unlock();
1419         kfree(dp->ports);
1420 err_destroy_percpu:
1421         free_percpu(dp->stats_percpu);
1422 err_destroy_table:
1423         ovs_flow_tbl_destroy(&dp->table, false);
1424 err_free_dp:
1425         release_net(ovs_dp_get_net(dp));
1426         kfree(dp);
1427 err_free_reply:
1428         kfree_skb(reply);
1429 err:
1430         return err;
1431 }
1432
1433 /* Called with ovs_mutex. */
1434 static void __dp_destroy(struct datapath *dp)
1435 {
1436         int i;
1437
1438         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1439                 struct vport *vport;
1440                 struct hlist_node *n;
1441
1442                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1443                         if (vport->port_no != OVSP_LOCAL)
1444                                 ovs_dp_detach_port(vport);
1445         }
1446
1447         list_del_rcu(&dp->list_node);
1448
1449         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1450          * all ports in datapath are destroyed first before freeing datapath.
1451          */
1452         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1453
1454         /* RCU destroy the flow table */
1455         ovs_flow_tbl_destroy(&dp->table, true);
1456
1457         call_rcu(&dp->rcu, destroy_dp_rcu);
1458 }
1459
1460 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1461 {
1462         struct sk_buff *reply;
1463         struct datapath *dp;
1464         int err;
1465
1466         reply = ovs_dp_cmd_alloc_info(info);
1467         if (!reply)
1468                 return -ENOMEM;
1469
1470         ovs_lock();
1471         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1472         err = PTR_ERR(dp);
1473         if (IS_ERR(dp))
1474                 goto err_unlock_free;
1475
1476         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1477                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1478         BUG_ON(err < 0);
1479
1480         __dp_destroy(dp);
1481         ovs_unlock();
1482
1483         ovs_notify(&dp_datapath_genl_family, reply, info);
1484
1485         return 0;
1486
1487 err_unlock_free:
1488         ovs_unlock();
1489         kfree_skb(reply);
1490         return err;
1491 }
1492
1493 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1494 {
1495         struct sk_buff *reply;
1496         struct datapath *dp;
1497         int err;
1498
1499         reply = ovs_dp_cmd_alloc_info(info);
1500         if (!reply)
1501                 return -ENOMEM;
1502
1503         ovs_lock();
1504         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1505         err = PTR_ERR(dp);
1506         if (IS_ERR(dp))
1507                 goto err_unlock_free;
1508
1509         ovs_dp_change(dp, info->attrs);
1510
1511         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1512                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1513         BUG_ON(err < 0);
1514
1515         ovs_unlock();
1516         ovs_notify(&dp_datapath_genl_family, reply, info);
1517
1518         return 0;
1519
1520 err_unlock_free:
1521         ovs_unlock();
1522         kfree_skb(reply);
1523         return err;
1524 }
1525
1526 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1527 {
1528         struct sk_buff *reply;
1529         struct datapath *dp;
1530         int err;
1531
1532         reply = ovs_dp_cmd_alloc_info(info);
1533         if (!reply)
1534                 return -ENOMEM;
1535
1536         rcu_read_lock();
1537         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1538         if (IS_ERR(dp)) {
1539                 err = PTR_ERR(dp);
1540                 goto err_unlock_free;
1541         }
1542         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1543                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1544         BUG_ON(err < 0);
1545         rcu_read_unlock();
1546
1547         return genlmsg_reply(reply, info);
1548
1549 err_unlock_free:
1550         rcu_read_unlock();
1551         kfree_skb(reply);
1552         return err;
1553 }
1554
1555 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1556 {
1557         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1558         struct datapath *dp;
1559         int skip = cb->args[0];
1560         int i = 0;
1561
1562         rcu_read_lock();
1563         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1564                 if (i >= skip &&
1565                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1566                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1567                                          OVS_DP_CMD_NEW) < 0)
1568                         break;
1569                 i++;
1570         }
1571         rcu_read_unlock();
1572
1573         cb->args[0] = i;
1574
1575         return skb->len;
1576 }
1577
1578 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1579         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1580         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1581         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1582 };
1583
1584 static const struct genl_ops dp_datapath_genl_ops[] = {
1585         { .cmd = OVS_DP_CMD_NEW,
1586           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1587           .policy = datapath_policy,
1588           .doit = ovs_dp_cmd_new
1589         },
1590         { .cmd = OVS_DP_CMD_DEL,
1591           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1592           .policy = datapath_policy,
1593           .doit = ovs_dp_cmd_del
1594         },
1595         { .cmd = OVS_DP_CMD_GET,
1596           .flags = 0,               /* OK for unprivileged users. */
1597           .policy = datapath_policy,
1598           .doit = ovs_dp_cmd_get,
1599           .dumpit = ovs_dp_cmd_dump
1600         },
1601         { .cmd = OVS_DP_CMD_SET,
1602           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1603           .policy = datapath_policy,
1604           .doit = ovs_dp_cmd_set,
1605         },
1606 };
1607
1608 static struct genl_family dp_datapath_genl_family = {
1609         .id = GENL_ID_GENERATE,
1610         .hdrsize = sizeof(struct ovs_header),
1611         .name = OVS_DATAPATH_FAMILY,
1612         .version = OVS_DATAPATH_VERSION,
1613         .maxattr = OVS_DP_ATTR_MAX,
1614         .netnsok = true,
1615         .parallel_ops = true,
1616         .ops = dp_datapath_genl_ops,
1617         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1618         .mcgrps = &ovs_dp_datapath_multicast_group,
1619         .n_mcgrps = 1,
1620 };
1621
1622 /* Called with ovs_mutex or RCU read lock. */
1623 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1624                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1625 {
1626         struct ovs_header *ovs_header;
1627         struct ovs_vport_stats vport_stats;
1628         int err;
1629
1630         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1631                                  flags, cmd);
1632         if (!ovs_header)
1633                 return -EMSGSIZE;
1634
1635         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1636
1637         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1638             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1639             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1640                            vport->ops->get_name(vport)))
1641                 goto nla_put_failure;
1642
1643         ovs_vport_get_stats(vport, &vport_stats);
1644         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1645                     &vport_stats))
1646                 goto nla_put_failure;
1647
1648         if (ovs_vport_get_upcall_portids(vport, skb))
1649                 goto nla_put_failure;
1650
1651         err = ovs_vport_get_options(vport, skb);
1652         if (err == -EMSGSIZE)
1653                 goto error;
1654
1655         return genlmsg_end(skb, ovs_header);
1656
1657 nla_put_failure:
1658         err = -EMSGSIZE;
1659 error:
1660         genlmsg_cancel(skb, ovs_header);
1661         return err;
1662 }
1663
1664 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1665 {
1666         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1667 }
1668
1669 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1670 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1671                                          u32 seq, u8 cmd)
1672 {
1673         struct sk_buff *skb;
1674         int retval;
1675
1676         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1677         if (!skb)
1678                 return ERR_PTR(-ENOMEM);
1679
1680         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1681         BUG_ON(retval < 0);
1682
1683         return skb;
1684 }
1685
1686 /* Called with ovs_mutex or RCU read lock. */
1687 static struct vport *lookup_vport(struct net *net,
1688                                   struct ovs_header *ovs_header,
1689                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1690 {
1691         struct datapath *dp;
1692         struct vport *vport;
1693
1694         if (a[OVS_VPORT_ATTR_NAME]) {
1695                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1696                 if (!vport)
1697                         return ERR_PTR(-ENODEV);
1698                 if (ovs_header->dp_ifindex &&
1699                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1700                         return ERR_PTR(-ENODEV);
1701                 return vport;
1702         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1703                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1704
1705                 if (port_no >= DP_MAX_PORTS)
1706                         return ERR_PTR(-EFBIG);
1707
1708                 dp = get_dp(net, ovs_header->dp_ifindex);
1709                 if (!dp)
1710                         return ERR_PTR(-ENODEV);
1711
1712                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1713                 if (!vport)
1714                         return ERR_PTR(-ENODEV);
1715                 return vport;
1716         } else
1717                 return ERR_PTR(-EINVAL);
1718 }
1719
1720 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1721 {
1722         struct nlattr **a = info->attrs;
1723         struct ovs_header *ovs_header = info->userhdr;
1724         struct vport_parms parms;
1725         struct sk_buff *reply;
1726         struct vport *vport;
1727         struct datapath *dp;
1728         u32 port_no;
1729         int err;
1730
1731         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1732             !a[OVS_VPORT_ATTR_UPCALL_PID])
1733                 return -EINVAL;
1734
1735         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1736                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1737         if (port_no >= DP_MAX_PORTS)
1738                 return -EFBIG;
1739
1740         reply = ovs_vport_cmd_alloc_info();
1741         if (!reply)
1742                 return -ENOMEM;
1743
1744         ovs_lock();
1745         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1746         err = -ENODEV;
1747         if (!dp)
1748                 goto exit_unlock_free;
1749
1750         if (port_no) {
1751                 vport = ovs_vport_ovsl(dp, port_no);
1752                 err = -EBUSY;
1753                 if (vport)
1754                         goto exit_unlock_free;
1755         } else {
1756                 for (port_no = 1; ; port_no++) {
1757                         if (port_no >= DP_MAX_PORTS) {
1758                                 err = -EFBIG;
1759                                 goto exit_unlock_free;
1760                         }
1761                         vport = ovs_vport_ovsl(dp, port_no);
1762                         if (!vport)
1763                                 break;
1764                 }
1765         }
1766
1767         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1768         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1769         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1770         parms.dp = dp;
1771         parms.port_no = port_no;
1772         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1773
1774         vport = new_vport(&parms);
1775         err = PTR_ERR(vport);
1776         if (IS_ERR(vport))
1777                 goto exit_unlock_free;
1778
1779         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1780                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1781         BUG_ON(err < 0);
1782         ovs_unlock();
1783
1784         ovs_notify(&dp_vport_genl_family, reply, info);
1785         return 0;
1786
1787 exit_unlock_free:
1788         ovs_unlock();
1789         kfree_skb(reply);
1790         return err;
1791 }
1792
1793 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1794 {
1795         struct nlattr **a = info->attrs;
1796         struct sk_buff *reply;
1797         struct vport *vport;
1798         int err;
1799
1800         reply = ovs_vport_cmd_alloc_info();
1801         if (!reply)
1802                 return -ENOMEM;
1803
1804         ovs_lock();
1805         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1806         err = PTR_ERR(vport);
1807         if (IS_ERR(vport))
1808                 goto exit_unlock_free;
1809
1810         if (a[OVS_VPORT_ATTR_TYPE] &&
1811             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1812                 err = -EINVAL;
1813                 goto exit_unlock_free;
1814         }
1815
1816         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1817                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1818                 if (err)
1819                         goto exit_unlock_free;
1820         }
1821
1822
1823         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1824                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
1825
1826                 err = ovs_vport_set_upcall_portids(vport, ids);
1827                 if (err)
1828                         goto exit_unlock_free;
1829         }
1830
1831         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1832                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1833         BUG_ON(err < 0);
1834
1835         ovs_unlock();
1836         ovs_notify(&dp_vport_genl_family, reply, info);
1837         return 0;
1838
1839 exit_unlock_free:
1840         ovs_unlock();
1841         kfree_skb(reply);
1842         return err;
1843 }
1844
1845 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1846 {
1847         struct nlattr **a = info->attrs;
1848         struct sk_buff *reply;
1849         struct vport *vport;
1850         int err;
1851
1852         reply = ovs_vport_cmd_alloc_info();
1853         if (!reply)
1854                 return -ENOMEM;
1855
1856         ovs_lock();
1857         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1858         err = PTR_ERR(vport);
1859         if (IS_ERR(vport))
1860                 goto exit_unlock_free;
1861
1862         if (vport->port_no == OVSP_LOCAL) {
1863                 err = -EINVAL;
1864                 goto exit_unlock_free;
1865         }
1866
1867         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1868                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1869         BUG_ON(err < 0);
1870         ovs_dp_detach_port(vport);
1871         ovs_unlock();
1872
1873         ovs_notify(&dp_vport_genl_family, reply, info);
1874         return 0;
1875
1876 exit_unlock_free:
1877         ovs_unlock();
1878         kfree_skb(reply);
1879         return err;
1880 }
1881
1882 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1883 {
1884         struct nlattr **a = info->attrs;
1885         struct ovs_header *ovs_header = info->userhdr;
1886         struct sk_buff *reply;
1887         struct vport *vport;
1888         int err;
1889
1890         reply = ovs_vport_cmd_alloc_info();
1891         if (!reply)
1892                 return -ENOMEM;
1893
1894         rcu_read_lock();
1895         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1896         err = PTR_ERR(vport);
1897         if (IS_ERR(vport))
1898                 goto exit_unlock_free;
1899         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1900                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1901         BUG_ON(err < 0);
1902         rcu_read_unlock();
1903
1904         return genlmsg_reply(reply, info);
1905
1906 exit_unlock_free:
1907         rcu_read_unlock();
1908         kfree_skb(reply);
1909         return err;
1910 }
1911
1912 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1913 {
1914         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1915         struct datapath *dp;
1916         int bucket = cb->args[0], skip = cb->args[1];
1917         int i, j = 0;
1918
1919         rcu_read_lock();
1920         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1921         if (!dp) {
1922                 rcu_read_unlock();
1923                 return -ENODEV;
1924         }
1925         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1926                 struct vport *vport;
1927
1928                 j = 0;
1929                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1930                         if (j >= skip &&
1931                             ovs_vport_cmd_fill_info(vport, skb,
1932                                                     NETLINK_CB(cb->skb).portid,
1933                                                     cb->nlh->nlmsg_seq,
1934                                                     NLM_F_MULTI,
1935                                                     OVS_VPORT_CMD_NEW) < 0)
1936                                 goto out;
1937
1938                         j++;
1939                 }
1940                 skip = 0;
1941         }
1942 out:
1943         rcu_read_unlock();
1944
1945         cb->args[0] = i;
1946         cb->args[1] = j;
1947
1948         return skb->len;
1949 }
1950
1951 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1952         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1953         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1954         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1955         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1956         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1957         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1958 };
1959
1960 static const struct genl_ops dp_vport_genl_ops[] = {
1961         { .cmd = OVS_VPORT_CMD_NEW,
1962           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1963           .policy = vport_policy,
1964           .doit = ovs_vport_cmd_new
1965         },
1966         { .cmd = OVS_VPORT_CMD_DEL,
1967           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1968           .policy = vport_policy,
1969           .doit = ovs_vport_cmd_del
1970         },
1971         { .cmd = OVS_VPORT_CMD_GET,
1972           .flags = 0,               /* OK for unprivileged users. */
1973           .policy = vport_policy,
1974           .doit = ovs_vport_cmd_get,
1975           .dumpit = ovs_vport_cmd_dump
1976         },
1977         { .cmd = OVS_VPORT_CMD_SET,
1978           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1979           .policy = vport_policy,
1980           .doit = ovs_vport_cmd_set,
1981         },
1982 };
1983
1984 struct genl_family dp_vport_genl_family = {
1985         .id = GENL_ID_GENERATE,
1986         .hdrsize = sizeof(struct ovs_header),
1987         .name = OVS_VPORT_FAMILY,
1988         .version = OVS_VPORT_VERSION,
1989         .maxattr = OVS_VPORT_ATTR_MAX,
1990         .netnsok = true,
1991         .parallel_ops = true,
1992         .ops = dp_vport_genl_ops,
1993         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
1994         .mcgrps = &ovs_dp_vport_multicast_group,
1995         .n_mcgrps = 1,
1996 };
1997
1998 static struct genl_family * const dp_genl_families[] = {
1999         &dp_datapath_genl_family,
2000         &dp_vport_genl_family,
2001         &dp_flow_genl_family,
2002         &dp_packet_genl_family,
2003 };
2004
2005 static void dp_unregister_genl(int n_families)
2006 {
2007         int i;
2008
2009         for (i = 0; i < n_families; i++)
2010                 genl_unregister_family(dp_genl_families[i]);
2011 }
2012
2013 static int dp_register_genl(void)
2014 {
2015         int err;
2016         int i;
2017
2018         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2019
2020                 err = genl_register_family(dp_genl_families[i]);
2021                 if (err)
2022                         goto error;
2023         }
2024
2025         return 0;
2026
2027 error:
2028         dp_unregister_genl(i);
2029         return err;
2030 }
2031
2032 static int __net_init ovs_init_net(struct net *net)
2033 {
2034         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2035
2036         INIT_LIST_HEAD(&ovs_net->dps);
2037         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2038         return 0;
2039 }
2040
2041 static void __net_exit ovs_exit_net(struct net *net)
2042 {
2043         struct datapath *dp, *dp_next;
2044         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2045
2046         ovs_lock();
2047         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2048                 __dp_destroy(dp);
2049         ovs_unlock();
2050
2051         cancel_work_sync(&ovs_net->dp_notify_work);
2052 }
2053
2054 static struct pernet_operations ovs_net_ops = {
2055         .init = ovs_init_net,
2056         .exit = ovs_exit_net,
2057         .id   = &ovs_net_id,
2058         .size = sizeof(struct ovs_net),
2059 };
2060
2061 static int __init dp_init(void)
2062 {
2063         int err;
2064
2065         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2066
2067         pr_info("Open vSwitch switching datapath\n");
2068
2069         err = ovs_internal_dev_rtnl_link_register();
2070         if (err)
2071                 goto error;
2072
2073         err = ovs_flow_init();
2074         if (err)
2075                 goto error_unreg_rtnl_link;
2076
2077         err = ovs_vport_init();
2078         if (err)
2079                 goto error_flow_exit;
2080
2081         err = register_pernet_device(&ovs_net_ops);
2082         if (err)
2083                 goto error_vport_exit;
2084
2085         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2086         if (err)
2087                 goto error_netns_exit;
2088
2089         err = dp_register_genl();
2090         if (err < 0)
2091                 goto error_unreg_notifier;
2092
2093         return 0;
2094
2095 error_unreg_notifier:
2096         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2097 error_netns_exit:
2098         unregister_pernet_device(&ovs_net_ops);
2099 error_vport_exit:
2100         ovs_vport_exit();
2101 error_flow_exit:
2102         ovs_flow_exit();
2103 error_unreg_rtnl_link:
2104         ovs_internal_dev_rtnl_link_unregister();
2105 error:
2106         return err;
2107 }
2108
2109 static void dp_cleanup(void)
2110 {
2111         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2112         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2113         unregister_pernet_device(&ovs_net_ops);
2114         rcu_barrier();
2115         ovs_vport_exit();
2116         ovs_flow_exit();
2117         ovs_internal_dev_rtnl_link_unregister();
2118 }
2119
2120 module_init(dp_init);
2121 module_exit(dp_cleanup);
2122
2123 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2124 MODULE_LICENSE("GPL");