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 <linux/genetlink.h>
  52 #include <net/genetlink.h>
  53 #include <net/genetlink.h>
  54 #include <net/net_namespace.h>
  55 #include <net/netns/generic.h>
  56
  57 #include "datapath.h"
  58 #include "flow.h"
  59 #include "flow_table.h"
  60 #include "flow_netlink.h"
  61 #include "vlan.h"
  62 #include "vport-internal_dev.h"
  63 #include "vport-netdev.h"
  64
  65 int ovs_net_id __read_mostly;
  66
  67 static struct genl_family dp_packet_genl_family;
  68 static struct genl_family dp_flow_genl_family;
  69 static struct genl_family dp_datapath_genl_family;
  70
  71 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
  72         .name = OVS_FLOW_MCGROUP
  73 };
  74
  75 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  76         .name = OVS_DATAPATH_MCGROUP
  77 };
  78
  79 struct genl_multicast_group ovs_dp_vport_multicast_group = {
  80         .name = OVS_VPORT_MCGROUP
  81 };
  82
  83 /* Check if need to build a reply message.
  84  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  85 static bool ovs_must_notify(struct genl_info *info,
  86                             const struct genl_multicast_group *grp)
  87 {
  88         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  89                 netlink_has_listeners(genl_info_net(info)->genl_sock, GROUP_ID(grp));
  90 }
  91
  92 static void ovs_notify(struct genl_family *family, struct genl_multicast_group *grp,
  93                        struct sk_buff *skb, struct genl_info *info)
  94 {
  95         genl_notify(family, skb, genl_info_net(info),
  96                     info->snd_portid, GROUP_ID(grp), info->nlhdr, GFP_KERNEL);
  97 }
  98
  99 /**
 100  * DOC: Locking:
 101  *
 102  * All writes e.g. Writes to device state (add/remove datapath, port, set
 103  * operations on vports, etc.), Writes to other state (flow table
 104  * modifications, set miscellaneous datapath parameters, etc.) are protected
 105  * by ovs_lock.
 106  *
 107  * Reads are protected by RCU.
 108  *
 109  * There are a few special cases (mostly stats) that have their own
 110  * synchronization but they nest under all of above and don't interact with
 111  * each other.
 112  *
 113  * The RTNL lock nests inside ovs_mutex.
 114  */
 115
 116 static DEFINE_MUTEX(ovs_mutex);
 117
 118 void ovs_lock(void)
 119 {
 120         mutex_lock(&ovs_mutex);
 121 }
 122
 123 void ovs_unlock(void)
 124 {
 125         mutex_unlock(&ovs_mutex);
 126 }
 127
 128 #ifdef CONFIG_LOCKDEP
 129 int lockdep_ovsl_is_held(void)
 130 {
 131         if (debug_locks)
 132                 return lockdep_is_held(&ovs_mutex);
 133         else
 134                 return 1;
 135 }
 136 #endif
 137
 138 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 139                              const struct dp_upcall_info *);
 140 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 141                                   const struct dp_upcall_info *);
 142
 143 /* Must be called with rcu_read_lock or ovs_mutex. */
 144 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 145 {
 146         struct datapath *dp = NULL;
 147         struct net_device *dev;
 148
 149         rcu_read_lock();
 150         dev = dev_get_by_index_rcu(net, dp_ifindex);
 151         if (dev) {
 152                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 153                 if (vport)
 154                         dp = vport->dp;
 155         }
 156         rcu_read_unlock();
 157
 158         return dp;
 159 }
 160
 161 /* Must be called with rcu_read_lock or ovs_mutex. */
 162 const char *ovs_dp_name(const struct datapath *dp)
 163 {
 164         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 165         return vport->ops->get_name(vport);
 166 }
 167
 168 static int get_dpifindex(struct datapath *dp)
 169 {
 170         struct vport *local;
 171         int ifindex;
 172
 173         rcu_read_lock();
 174
 175         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 176         if (local)
 177                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 178         else
 179                 ifindex = 0;
 180
 181         rcu_read_unlock();
 182
 183         return ifindex;
 184 }
 185
 186 static void destroy_dp_rcu(struct rcu_head *rcu)
 187 {
 188         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 189
 190         free_percpu(dp->stats_percpu);
 191         release_net(ovs_dp_get_net(dp));
 192         kfree(dp->ports);
 193         kfree(dp);
 194 }
 195
 196 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 197                                             u16 port_no)
 198 {
 199         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 200 }
 201
 202 /* Called with ovs_mutex or RCU read lock. */
 203 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 204 {
 205         struct vport *vport;
 206         struct hlist_head *head;
 207
 208         head = vport_hash_bucket(dp, port_no);
 209         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 210                 if (vport->port_no == port_no)
 211                         return vport;
 212         }
 213         return NULL;
 214 }
 215
 216 /* Called with ovs_mutex. */
 217 static struct vport *new_vport(const struct vport_parms *parms)
 218 {
 219         struct vport *vport;
 220
 221         vport = ovs_vport_add(parms);
 222         if (!IS_ERR(vport)) {
 223                 struct datapath *dp = parms->dp;
 224                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 225
 226                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 227         }
 228         return vport;
 229 }
 230
 231 void ovs_dp_detach_port(struct vport *p)
 232 {
 233         ASSERT_OVSL();
 234
 235         /* First drop references to device. */
 236         hlist_del_rcu(&p->dp_hash_node);
 237
 238         /* Then destroy it. */
 239         ovs_vport_del(p);
 240 }
 241
 242 /* Must be called with rcu_read_lock. */
 243 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 244 {
 245         struct datapath *dp = p->dp;
 246         struct sw_flow *flow;
 247         struct dp_stats_percpu *stats;
 248         struct sw_flow_key key;
 249         u64 *stats_counter;
 250         u32 n_mask_hit;
 251         int error;
 252
 253         stats = this_cpu_ptr(dp->stats_percpu);
 254
 255         /* Extract flow from 'skb' into 'key'. */
 256         error = ovs_flow_extract(skb, p->port_no, &key);
 257         if (unlikely(error)) {
 258                 kfree_skb(skb);
 259                 return;
 260         }
 261
 262         /* Look up flow. */
 263         flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
 264         if (unlikely(!flow)) {
 265                 struct dp_upcall_info upcall;
 266
 267                 upcall.cmd = OVS_PACKET_CMD_MISS;
 268                 upcall.key = &key;
 269                 upcall.userdata = NULL;
 270                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 271                 ovs_dp_upcall(dp, skb, &upcall);
 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->sync);
 287         (*stats_counter)++;
 288         stats->n_mask_hit += n_mask_hit;
 289         u64_stats_update_end(&stats->sync);
 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->sync);
 316         stats->n_lost++;
 317         u64_stats_update_end(&stats->sync);
 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; /* to be queued to userspace */
 411         struct nlattr *nla;
 412         struct genl_info info = {
 413 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,14,0)
 414                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 415 #endif
 416                 .snd_portid = upcall_info->portid,
 417         };
 418         size_t len;
 419         unsigned int hlen;
 420         int err, dp_ifindex;
 421
 422         dp_ifindex = get_dpifindex(dp);
 423         if (!dp_ifindex)
 424                 return -ENODEV;
 425
 426         if (vlan_tx_tag_present(skb)) {
 427                 nskb = skb_clone(skb, GFP_ATOMIC);
 428                 if (!nskb)
 429                         return -ENOMEM;
 430
 431                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 432                 if (!nskb)
 433                         return -ENOMEM;
 434
 435                 vlan_set_tci(nskb, 0);
 436
 437                 skb = nskb;
 438         }
 439
 440         if (nla_attr_size(skb->len) > USHRT_MAX) {
 441                 err = -EFBIG;
 442                 goto out;
 443         }
 444
 445         /* Complete checksum if needed */
 446         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 447             (err = skb_checksum_help(skb)))
 448                 goto out;
 449
 450         /* Older versions of OVS user space enforce alignment of the last
 451          * Netlink attribute to NLA_ALIGNTO which would require extensive
 452          * padding logic. Only perform zerocopy if padding is not required.
 453          */
 454         if (dp->user_features & OVS_DP_F_UNALIGNED)
 455                 hlen = skb_zerocopy_headlen(skb);
 456         else
 457                 hlen = skb->len;
 458
 459         len = upcall_msg_size(upcall_info->userdata, hlen);
 460         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 461         if (!user_skb) {
 462                 err = -ENOMEM;
 463                 goto out;
 464         }
 465
 466         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 467                              0, upcall_info->cmd);
 468         upcall->dp_ifindex = dp_ifindex;
 469
 470         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 471         ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
 472         nla_nest_end(user_skb, nla);
 473
 474         if (upcall_info->userdata)
 475                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 476                           nla_len(upcall_info->userdata),
 477                           nla_data(upcall_info->userdata));
 478
 479         /* Only reserve room for attribute header, packet data is added
 480          * in skb_zerocopy() */
 481         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 482                 err = -ENOBUFS;
 483                 goto out;
 484         }
 485         nla->nla_len = nla_attr_size(skb->len);
 486
 487         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 488         if (err)
 489                 goto out;
 490
 491         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 492         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 493                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 494
 495                 if (plen > 0)
 496                         memset(skb_put(user_skb, plen), 0, plen);
 497         }
 498
 499         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 500
 501         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 502 out:
 503         if (err)
 504                 skb_tx_error(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 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_bh(&percpu_stats->sync);
 646                         local_stats = *percpu_stats;
 647                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, 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
 771         if (!skb)
 772                 return ERR_PTR(-ENOMEM);
 773
 774         return skb;
 775 }
 776
 777 /* Called with ovs_mutex. */
 778 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 779                                                int dp_ifindex,
 780                                                struct genl_info *info, u8 cmd,
 781                                                bool always)
 782 {
 783         struct sk_buff *skb;
 784         int retval;
 785
 786         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
 787                                       always);
 788         if (!skb || IS_ERR(skb))
 789                 return skb;
 790
 791         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 792                                         info->snd_portid, info->snd_seq, 0,
 793                                         cmd);
 794         BUG_ON(retval < 0);
 795         return skb;
 796 }
 797
 798 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 799 {
 800         struct nlattr **a = info->attrs;
 801         struct ovs_header *ovs_header = info->userhdr;
 802         struct sw_flow *flow, *new_flow;
 803         struct sw_flow_mask mask;
 804         struct sk_buff *reply;
 805         struct datapath *dp;
 806         struct sw_flow_actions *acts;
 807         struct sw_flow_match match;
 808         int error;
 809
 810         /* Must have key and actions. */
 811         error = -EINVAL;
 812         if (!a[OVS_FLOW_ATTR_KEY])
 813                 goto error;
 814         if (!a[OVS_FLOW_ATTR_ACTIONS])
 815                 goto error;
 816
 817         /* Most of the time we need to allocate a new flow, do it before
 818          * locking. */
 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                         error = -EEXIST;
 897                         goto err_unlock_ovs;
 898                 }
 899                 /* Update actions. */
 900                 old_acts = ovsl_dereference(flow->sf_acts);
 901                 rcu_assign_pointer(flow->sf_acts, acts);
 902
 903                 if (unlikely(reply)) {
 904                         error = ovs_flow_cmd_fill_info(flow,
 905                                                        ovs_header->dp_ifindex,
 906                                                        reply, info->snd_portid,
 907                                                        info->snd_seq, 0,
 908                                                        OVS_FLOW_CMD_NEW);
 909                         BUG_ON(error < 0);
 910                 }
 911                 ovs_unlock();
 912
 913                 ovs_nla_free_flow_actions(old_acts);
 914                 ovs_flow_free(new_flow, false);
 915         }
 916
 917         if (reply)
 918                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
 919         return 0;
 920
 921 err_unlock_ovs:
 922         ovs_unlock();
 923         kfree_skb(reply);
 924 err_kfree_acts:
 925         kfree(acts);
 926 err_kfree_flow:
 927         ovs_flow_free(new_flow, false);
 928 error:
 929         return error;
 930 }
 931
 932 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
 933 {
 934         struct nlattr **a = info->attrs;
 935         struct ovs_header *ovs_header = info->userhdr;
 936         struct sw_flow_key key, masked_key;
 937         struct sw_flow *flow;
 938         struct sw_flow_mask mask;
 939         struct sk_buff *reply = NULL;
 940         struct datapath *dp;
 941         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
 942         struct sw_flow_match match;
 943         int error;
 944
 945         /* Extract key. */
 946         error = -EINVAL;
 947         if (!a[OVS_FLOW_ATTR_KEY])
 948                 goto error;
 949
 950         ovs_match_init(&match, &key, &mask);
 951         error = ovs_nla_get_match(&match,
 952                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 953         if (error)
 954                 goto error;
 955
 956         /* Validate actions. */
 957         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 958                 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 959                 error = PTR_ERR(acts);
 960                 if (IS_ERR(acts))
 961                         goto error;
 962
 963                 ovs_flow_mask_key(&masked_key, &key, &mask);
 964                 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
 965                                              &masked_key, 0, &acts);
 966                 if (error) {
 967                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 968                         goto err_kfree_acts;
 969                 }
 970         }
 971
 972         /* Can allocate before locking if have acts. */
 973         if (acts) {
 974                 reply = ovs_flow_cmd_alloc_info(acts, info, false);
 975                 if (IS_ERR(reply)) {
 976                         error = PTR_ERR(reply);
 977                         goto err_kfree_acts;
 978                 }
 979         }
 980
 981         ovs_lock();
 982         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 983         if (unlikely(!dp)) {
 984                 error = -ENODEV;
 985                 goto err_unlock_ovs;
 986         }
 987         /* Check that the flow exists. */
 988         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 989         if (unlikely(!flow)) {
 990                 error = -ENOENT;
 991                 goto err_unlock_ovs;
 992         }
 993         /* The unmasked key has to be the same for flow updates. */
 994         if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
 995                 error = -EEXIST;
 996                 goto err_unlock_ovs;
 997         }
 998         /* Update actions, if present. */
 999         if (likely(acts)) {
1000                 old_acts = ovsl_dereference(flow->sf_acts);
1001                 rcu_assign_pointer(flow->sf_acts, acts);
1002
1003                 if (unlikely(reply)) {
1004                         error = ovs_flow_cmd_fill_info(flow,
1005                                                        ovs_header->dp_ifindex,
1006                                                        reply, info->snd_portid,
1007                                                        info->snd_seq, 0,
1008                                                        OVS_FLOW_CMD_NEW);
1009                         BUG_ON(error < 0);
1010                 }
1011         } else {
1012                 /* Could not alloc without acts before locking. */
1013                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1014                                                 info, OVS_FLOW_CMD_NEW, false);
1015                 if (unlikely(IS_ERR(reply))) {
1016                         error = PTR_ERR(reply);
1017                         goto err_unlock_ovs;
1018                 }
1019         }
1020
1021         /* Clear stats. */
1022         if (a[OVS_FLOW_ATTR_CLEAR])
1023                 ovs_flow_stats_clear(flow);
1024         ovs_unlock();
1025
1026         if (reply)
1027                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1028         if (old_acts)
1029                 ovs_nla_free_flow_actions(old_acts);
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(&dp->table, &key);
1070         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
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         if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1114                 err = ovs_flow_tbl_flush(&dp->table);
1115                 goto unlock;
1116         }
1117         flow = ovs_flow_tbl_lookup(&dp->table, &key);
1118         if (unlikely(!flow || !ovs_flow_cmp_unmasked_key(flow, &match))) {
1119                 err = -ENOENT;
1120                 goto unlock;
1121         }
1122
1123         ovs_flow_tbl_remove(&dp->table, flow);
1124         ovs_unlock();
1125
1126         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *)flow->sf_acts,
1127                                         info, false);
1128
1129         if (likely(reply)) {
1130                 if (likely(!IS_ERR(reply))) {
1131                         rcu_read_lock(); /* Keep RCU checker happy. */
1132                         err = ovs_flow_cmd_fill_info(flow,
1133                                                      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                         ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1140                 } else {
1141                         genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1142                                      GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1143
1144                 }
1145         }
1146
1147         ovs_flow_free(flow, true);
1148         return 0;
1149 unlock:
1150         ovs_unlock();
1151         return err;
1152 }
1153
1154 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1155 {
1156         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1157         struct table_instance *ti;
1158         struct datapath *dp;
1159
1160         rcu_read_lock();
1161         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1162         if (!dp) {
1163                 rcu_read_unlock();
1164                 return -ENODEV;
1165         }
1166
1167         ti = rcu_dereference(dp->table.ti);
1168         for (;;) {
1169                 struct sw_flow *flow;
1170                 u32 bucket, obj;
1171
1172                 bucket = cb->args[0];
1173                 obj = cb->args[1];
1174                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1175                 if (!flow)
1176                         break;
1177
1178                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1179                                            NETLINK_CB(cb->skb).portid,
1180                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1181                                            OVS_FLOW_CMD_NEW) < 0)
1182                         break;
1183
1184                 cb->args[0] = bucket;
1185                 cb->args[1] = obj;
1186         }
1187         rcu_read_unlock();
1188         return skb->len;
1189 }
1190
1191 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1192         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1193         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1194         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1195 };
1196
1197 static struct genl_ops dp_flow_genl_ops[] = {
1198         { .cmd = OVS_FLOW_CMD_NEW,
1199           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1200           .policy = flow_policy,
1201           .doit = ovs_flow_cmd_new
1202         },
1203         { .cmd = OVS_FLOW_CMD_DEL,
1204           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1205           .policy = flow_policy,
1206           .doit = ovs_flow_cmd_del
1207         },
1208         { .cmd = OVS_FLOW_CMD_GET,
1209           .flags = 0,               /* OK for unprivileged users. */
1210           .policy = flow_policy,
1211           .doit = ovs_flow_cmd_get,
1212           .dumpit = ovs_flow_cmd_dump
1213         },
1214         { .cmd = OVS_FLOW_CMD_SET,
1215           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1216           .policy = flow_policy,
1217           .doit = ovs_flow_cmd_set,
1218         },
1219 };
1220
1221 static struct genl_family dp_flow_genl_family = {
1222         .id = GENL_ID_GENERATE,
1223         .hdrsize = sizeof(struct ovs_header),
1224         .name = OVS_FLOW_FAMILY,
1225         .version = OVS_FLOW_VERSION,
1226         .maxattr = OVS_FLOW_ATTR_MAX,
1227         .netnsok = true,
1228         .parallel_ops = true,
1229         .ops = dp_flow_genl_ops,
1230         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1231         .mcgrps = &ovs_dp_flow_multicast_group,
1232         .n_mcgrps = 1,
1233 };
1234
1235 static size_t ovs_dp_cmd_msg_size(void)
1236 {
1237         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1238
1239         msgsize += nla_total_size(IFNAMSIZ);
1240         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1241         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1242         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1243
1244         return msgsize;
1245 }
1246
1247 /* Called with ovs_mutex or RCU read lock. */
1248 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1249                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1250 {
1251         struct ovs_header *ovs_header;
1252         struct ovs_dp_stats dp_stats;
1253         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1254         int err;
1255
1256         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1257                                    flags, cmd);
1258         if (!ovs_header)
1259                 goto error;
1260
1261         ovs_header->dp_ifindex = get_dpifindex(dp);
1262
1263         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1264         if (err)
1265                 goto nla_put_failure;
1266
1267         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1268         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1269                         &dp_stats))
1270                 goto nla_put_failure;
1271
1272         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1273                         sizeof(struct ovs_dp_megaflow_stats),
1274                         &dp_megaflow_stats))
1275                 goto nla_put_failure;
1276
1277         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1278                 goto nla_put_failure;
1279
1280         return genlmsg_end(skb, ovs_header);
1281
1282 nla_put_failure:
1283         genlmsg_cancel(skb, ovs_header);
1284 error:
1285         return -EMSGSIZE;
1286 }
1287
1288 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1289 {
1290         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1291 }
1292
1293 /* Called with rcu_read_lock or ovs_mutex. */
1294 static struct datapath *lookup_datapath(struct net *net,
1295                                         struct ovs_header *ovs_header,
1296                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1297 {
1298         struct datapath *dp;
1299
1300         if (!a[OVS_DP_ATTR_NAME])
1301                 dp = get_dp(net, ovs_header->dp_ifindex);
1302         else {
1303                 struct vport *vport;
1304
1305                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1306                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1307         }
1308         return dp ? dp : ERR_PTR(-ENODEV);
1309 }
1310
1311 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1312 {
1313         struct datapath *dp;
1314
1315         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1316         if (IS_ERR(dp))
1317                 return;
1318
1319         WARN(dp->user_features, "Dropping previously announced user features\n");
1320         dp->user_features = 0;
1321 }
1322
1323 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1324 {
1325         if (a[OVS_DP_ATTR_USER_FEATURES])
1326                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1327 }
1328
1329 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1330 {
1331         struct nlattr **a = info->attrs;
1332         struct vport_parms parms;
1333         struct sk_buff *reply;
1334         struct datapath *dp;
1335         struct vport *vport;
1336         struct ovs_net *ovs_net;
1337         int err, i;
1338
1339         err = -EINVAL;
1340         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1341                 goto err;
1342
1343         reply = ovs_dp_cmd_alloc_info(info);
1344         if (!reply)
1345                 return -ENOMEM;
1346
1347         err = -ENOMEM;
1348         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1349         if (dp == NULL)
1350                 goto err_free_reply;
1351
1352         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1353
1354         /* Allocate table. */
1355         err = ovs_flow_tbl_init(&dp->table);
1356         if (err)
1357                 goto err_free_dp;
1358
1359         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1360         if (!dp->stats_percpu) {
1361                 err = -ENOMEM;
1362                 goto err_destroy_table;
1363         }
1364
1365         for_each_possible_cpu(i) {
1366                 struct dp_stats_percpu *dpath_stats;
1367                 dpath_stats = per_cpu_ptr(dp->stats_percpu, i);
1368                 u64_stats_init(&dpath_stats->sync);
1369         }
1370
1371         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1372                             GFP_KERNEL);
1373         if (!dp->ports) {
1374                 err = -ENOMEM;
1375                 goto err_destroy_percpu;
1376         }
1377
1378         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1379                 INIT_HLIST_HEAD(&dp->ports[i]);
1380
1381         /* Set up our datapath device. */
1382         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1383         parms.type = OVS_VPORT_TYPE_INTERNAL;
1384         parms.options = NULL;
1385         parms.dp = dp;
1386         parms.port_no = OVSP_LOCAL;
1387         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1388
1389         ovs_dp_change(dp, a);
1390
1391         /* So far only local changes have been made, now need the lock. */
1392         ovs_lock();
1393
1394         vport = new_vport(&parms);
1395         if (IS_ERR(vport)) {
1396                 err = PTR_ERR(vport);
1397                 if (err == -EBUSY)
1398                         err = -EEXIST;
1399
1400                 if (err == -EEXIST) {
1401                         /* An outdated user space instance that does not understand
1402                          * the concept of user_features has attempted to create a new
1403                          * datapath and is likely to reuse it. Drop all user features.
1404                          */
1405                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1406                                 ovs_dp_reset_user_features(skb, info);
1407                 }
1408
1409                 goto err_destroy_ports_array;
1410         }
1411
1412         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1413                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1414         BUG_ON(err < 0);
1415
1416         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1417         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1418
1419         ovs_unlock();
1420
1421         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1422         return 0;
1423
1424 err_destroy_ports_array:
1425         ovs_unlock();
1426         kfree(dp->ports);
1427 err_destroy_percpu:
1428         free_percpu(dp->stats_percpu);
1429 err_destroy_table:
1430         ovs_flow_tbl_destroy(&dp->table, false);
1431 err_free_dp:
1432         release_net(ovs_dp_get_net(dp));
1433         kfree(dp);
1434 err_free_reply:
1435         kfree_skb(reply);
1436 err:
1437         return err;
1438 }
1439
1440 /* Called with ovs_mutex. */
1441 static void __dp_destroy(struct datapath *dp)
1442 {
1443         int i;
1444
1445         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1446                 struct vport *vport;
1447                 struct hlist_node *n;
1448
1449                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1450                         if (vport->port_no != OVSP_LOCAL)
1451                                 ovs_dp_detach_port(vport);
1452         }
1453
1454         list_del_rcu(&dp->list_node);
1455
1456         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1457          * all ports in datapath are destroyed first before freeing datapath.
1458          */
1459         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1460
1461         /* RCU destroy the flow table */
1462         ovs_flow_tbl_destroy(&dp->table, true);
1463
1464         call_rcu(&dp->rcu, destroy_dp_rcu);
1465 }
1466
1467 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1468 {
1469         struct sk_buff *reply;
1470         struct datapath *dp;
1471         int err;
1472
1473         reply = ovs_dp_cmd_alloc_info(info);
1474         if (!reply)
1475                 return -ENOMEM;
1476
1477         ovs_lock();
1478         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1479         err = PTR_ERR(dp);
1480         if (IS_ERR(dp))
1481                 goto err_unlock_free;
1482
1483         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1484                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1485         BUG_ON(err < 0);
1486
1487         __dp_destroy(dp);
1488
1489         ovs_unlock();
1490         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1491         return 0;
1492
1493 err_unlock_free:
1494         ovs_unlock();
1495         kfree_skb(reply);
1496         return err;
1497 }
1498
1499 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1500 {
1501         struct sk_buff *reply;
1502         struct datapath *dp;
1503         int err;
1504
1505         reply = ovs_dp_cmd_alloc_info(info);
1506         if (!reply)
1507                 return -ENOMEM;
1508
1509         ovs_lock();
1510         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1511         err = PTR_ERR(dp);
1512         if (IS_ERR(dp))
1513                 goto err_unlock_free;
1514
1515         ovs_dp_change(dp, info->attrs);
1516
1517         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1518                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1519         BUG_ON(err < 0);
1520
1521         ovs_unlock();
1522         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1523         return 0;
1524
1525 err_unlock_free:
1526         ovs_unlock();
1527         kfree_skb(reply);
1528         return err;
1529 }
1530
1531 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1532 {
1533         struct sk_buff *reply;
1534         struct datapath *dp;
1535         int err;
1536
1537         reply = ovs_dp_cmd_alloc_info(info);
1538         if (!reply)
1539                 return -ENOMEM;
1540
1541         rcu_read_lock();
1542         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1543         if (IS_ERR(dp)) {
1544                 err = PTR_ERR(dp);
1545                 goto err_unlock_free;
1546         }
1547         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1548                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1549         BUG_ON(err < 0);
1550         rcu_read_unlock();
1551
1552         return genlmsg_reply(reply, info);
1553
1554 err_unlock_free:
1555         rcu_read_unlock();
1556         kfree_skb(reply);
1557         return err;
1558 }
1559
1560 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1561 {
1562         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1563         struct datapath *dp;
1564         int skip = cb->args[0];
1565         int i = 0;
1566
1567         rcu_read_lock();
1568         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1569                 if (i >= skip &&
1570                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1571                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1572                                          OVS_DP_CMD_NEW) < 0)
1573                         break;
1574                 i++;
1575         }
1576         rcu_read_unlock();
1577
1578         cb->args[0] = i;
1579
1580         return skb->len;
1581 }
1582
1583 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1584         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1585         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1586         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1587 };
1588
1589 static struct genl_ops dp_datapath_genl_ops[] = {
1590         { .cmd = OVS_DP_CMD_NEW,
1591           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1592           .policy = datapath_policy,
1593           .doit = ovs_dp_cmd_new
1594         },
1595         { .cmd = OVS_DP_CMD_DEL,
1596           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1597           .policy = datapath_policy,
1598           .doit = ovs_dp_cmd_del
1599         },
1600         { .cmd = OVS_DP_CMD_GET,
1601           .flags = 0,               /* OK for unprivileged users. */
1602           .policy = datapath_policy,
1603           .doit = ovs_dp_cmd_get,
1604           .dumpit = ovs_dp_cmd_dump
1605         },
1606         { .cmd = OVS_DP_CMD_SET,
1607           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1608           .policy = datapath_policy,
1609           .doit = ovs_dp_cmd_set,
1610         },
1611 };
1612
1613 static struct genl_family dp_datapath_genl_family = {
1614         .id = GENL_ID_GENERATE,
1615         .hdrsize = sizeof(struct ovs_header),
1616         .name = OVS_DATAPATH_FAMILY,
1617         .version = OVS_DATAPATH_VERSION,
1618         .maxattr = OVS_DP_ATTR_MAX,
1619         .netnsok = true,
1620         .parallel_ops = true,
1621         .ops = dp_datapath_genl_ops,
1622         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1623         .mcgrps = &ovs_dp_datapath_multicast_group,
1624         .n_mcgrps = 1,
1625 };
1626
1627 /* Called with ovs_mutex or RCU read lock. */
1628 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1629                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1630 {
1631         struct ovs_header *ovs_header;
1632         struct ovs_vport_stats vport_stats;
1633         int err;
1634
1635         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1636                                  flags, cmd);
1637         if (!ovs_header)
1638                 return -EMSGSIZE;
1639
1640         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1641
1642         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1643             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1644             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)))
1645                 goto nla_put_failure;
1646
1647         ovs_vport_get_stats(vport, &vport_stats);
1648         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1649                     &vport_stats))
1650                 goto nla_put_failure;
1651
1652         if (ovs_vport_get_upcall_portids(vport, skb))
1653                 goto nla_put_failure;
1654
1655         err = ovs_vport_get_options(vport, skb);
1656         if (err == -EMSGSIZE)
1657                 goto error;
1658
1659         return genlmsg_end(skb, ovs_header);
1660
1661 nla_put_failure:
1662         err = -EMSGSIZE;
1663 error:
1664         genlmsg_cancel(skb, ovs_header);
1665         return err;
1666 }
1667
1668 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1669 {
1670         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1671 }
1672
1673 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1674 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1675                                          u32 seq, u8 cmd)
1676 {
1677         struct sk_buff *skb;
1678         int retval;
1679
1680         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1681         if (!skb)
1682                 return ERR_PTR(-ENOMEM);
1683
1684         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1685         BUG_ON(retval < 0);
1686
1687         return skb;
1688 }
1689
1690 /* Called with ovs_mutex or RCU read lock. */
1691 static struct vport *lookup_vport(struct net *net,
1692                                   struct ovs_header *ovs_header,
1693                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1694 {
1695         struct datapath *dp;
1696         struct vport *vport;
1697
1698         if (a[OVS_VPORT_ATTR_NAME]) {
1699                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1700                 if (!vport)
1701                         return ERR_PTR(-ENODEV);
1702                 if (ovs_header->dp_ifindex &&
1703                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1704                         return ERR_PTR(-ENODEV);
1705                 return vport;
1706         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1707                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1708
1709                 if (port_no >= DP_MAX_PORTS)
1710                         return ERR_PTR(-EFBIG);
1711
1712                 dp = get_dp(net, ovs_header->dp_ifindex);
1713                 if (!dp)
1714                         return ERR_PTR(-ENODEV);
1715
1716                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1717                 if (!vport)
1718                         return ERR_PTR(-ENODEV);
1719                 return vport;
1720         } else
1721                 return ERR_PTR(-EINVAL);
1722 }
1723
1724 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1725 {
1726         struct nlattr **a = info->attrs;
1727         struct ovs_header *ovs_header = info->userhdr;
1728         struct vport_parms parms;
1729         struct sk_buff *reply;
1730         struct vport *vport;
1731         struct datapath *dp;
1732         u32 port_no;
1733         int err;
1734
1735         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1736             !a[OVS_VPORT_ATTR_UPCALL_PID])
1737                 return -EINVAL;
1738
1739         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1740                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1741         if (port_no >= DP_MAX_PORTS)
1742                 return -EFBIG;
1743
1744         reply = ovs_vport_cmd_alloc_info();
1745         if (!reply)
1746                 return -ENOMEM;
1747
1748         ovs_lock();
1749         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1750         err = -ENODEV;
1751         if (!dp)
1752                 goto exit_unlock_free;
1753
1754         if (port_no) {
1755                 vport = ovs_vport_ovsl(dp, port_no);
1756                 err = -EBUSY;
1757                 if (vport)
1758                         goto exit_unlock_free;
1759         } else {
1760                 for (port_no = 1; ; port_no++) {
1761                         if (port_no >= DP_MAX_PORTS) {
1762                                 err = -EFBIG;
1763                                 goto exit_unlock_free;
1764                         }
1765                         vport = ovs_vport_ovsl(dp, port_no);
1766                         if (!vport)
1767                                 break;
1768                 }
1769         }
1770
1771         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1772         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1773         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1774         parms.dp = dp;
1775         parms.port_no = port_no;
1776         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1777
1778         vport = new_vport(&parms);
1779         err = PTR_ERR(vport);
1780         if (IS_ERR(vport))
1781                 goto exit_unlock_free;
1782
1783         err = 0;
1784         if (a[OVS_VPORT_ATTR_STATS])
1785                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1786
1787         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1788                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1789         BUG_ON(err < 0);
1790         ovs_unlock();
1791
1792         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1793         return 0;
1794
1795 exit_unlock_free:
1796         ovs_unlock();
1797         kfree_skb(reply);
1798         return err;
1799 }
1800
1801 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1802 {
1803         struct nlattr **a = info->attrs;
1804         struct sk_buff *reply;
1805         struct vport *vport;
1806         int err;
1807
1808         reply = ovs_vport_cmd_alloc_info();
1809         if (!reply)
1810                 return -ENOMEM;
1811
1812         ovs_lock();
1813         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1814         err = PTR_ERR(vport);
1815         if (IS_ERR(vport))
1816                 goto exit_unlock_free;
1817
1818         if (a[OVS_VPORT_ATTR_TYPE] &&
1819             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1820                 err = -EINVAL;
1821                 goto exit_unlock_free;
1822         }
1823
1824         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1825                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1826                 if (err)
1827                         goto exit_unlock_free;
1828         }
1829
1830         if (a[OVS_VPORT_ATTR_STATS])
1831                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1832
1833
1834         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1835                 err = ovs_vport_set_upcall_portids(vport,
1836                                                    a[OVS_VPORT_ATTR_UPCALL_PID]);
1837                 if (err)
1838                         goto exit_unlock_free;
1839         }
1840
1841         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1842                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1843         BUG_ON(err < 0);
1844         ovs_unlock();
1845
1846         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1847         return 0;
1848
1849 exit_unlock_free:
1850         ovs_unlock();
1851         kfree_skb(reply);
1852         return err;
1853 }
1854
1855 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1856 {
1857         struct nlattr **a = info->attrs;
1858         struct sk_buff *reply;
1859         struct vport *vport;
1860         int err;
1861
1862         reply = ovs_vport_cmd_alloc_info();
1863         if (!reply)
1864                 return -ENOMEM;
1865
1866         ovs_lock();
1867         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1868         err = PTR_ERR(vport);
1869         if (IS_ERR(vport))
1870                 goto exit_unlock_free;
1871
1872         if (vport->port_no == OVSP_LOCAL) {
1873                 err = -EINVAL;
1874                 goto exit_unlock_free;
1875         }
1876
1877         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1878                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1879         BUG_ON(err < 0);
1880         ovs_dp_detach_port(vport);
1881         ovs_unlock();
1882
1883         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1884         return 0;
1885
1886 exit_unlock_free:
1887         ovs_unlock();
1888         kfree_skb(reply);
1889         return err;
1890 }
1891
1892 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1893 {
1894         struct nlattr **a = info->attrs;
1895         struct ovs_header *ovs_header = info->userhdr;
1896         struct sk_buff *reply;
1897         struct vport *vport;
1898         int err;
1899
1900         reply = ovs_vport_cmd_alloc_info();
1901         if (!reply)
1902                 return -ENOMEM;
1903
1904         rcu_read_lock();
1905         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1906         err = PTR_ERR(vport);
1907         if (IS_ERR(vport))
1908                 goto exit_unlock_free;
1909         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1910                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1911         BUG_ON(err < 0);
1912         rcu_read_unlock();
1913
1914         return genlmsg_reply(reply, info);
1915
1916 exit_unlock_free:
1917         rcu_read_unlock();
1918         kfree_skb(reply);
1919         return err;
1920 }
1921
1922 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1923 {
1924         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1925         struct datapath *dp;
1926         int bucket = cb->args[0], skip = cb->args[1];
1927         int i, j = 0;
1928
1929         rcu_read_lock();
1930         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1931         if (!dp) {
1932                 rcu_read_unlock();
1933                 return -ENODEV;
1934         }
1935         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1936                 struct vport *vport;
1937
1938                 j = 0;
1939                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1940                         if (j >= skip &&
1941                             ovs_vport_cmd_fill_info(vport, skb,
1942                                                     NETLINK_CB(cb->skb).portid,
1943                                                     cb->nlh->nlmsg_seq,
1944                                                     NLM_F_MULTI,
1945                                                     OVS_VPORT_CMD_NEW) < 0)
1946                                 goto out;
1947
1948                         j++;
1949                 }
1950                 skip = 0;
1951         }
1952 out:
1953         rcu_read_unlock();
1954
1955         cb->args[0] = i;
1956         cb->args[1] = j;
1957
1958         return skb->len;
1959 }
1960
1961 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1962         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1963         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1964         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1965         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1966         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1967         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1968 };
1969
1970 static struct genl_ops dp_vport_genl_ops[] = {
1971         { .cmd = OVS_VPORT_CMD_NEW,
1972           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1973           .policy = vport_policy,
1974           .doit = ovs_vport_cmd_new
1975         },
1976         { .cmd = OVS_VPORT_CMD_DEL,
1977           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1978           .policy = vport_policy,
1979           .doit = ovs_vport_cmd_del
1980         },
1981         { .cmd = OVS_VPORT_CMD_GET,
1982           .flags = 0,               /* OK for unprivileged users. */
1983           .policy = vport_policy,
1984           .doit = ovs_vport_cmd_get,
1985           .dumpit = ovs_vport_cmd_dump
1986         },
1987         { .cmd = OVS_VPORT_CMD_SET,
1988           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1989           .policy = vport_policy,
1990           .doit = ovs_vport_cmd_set,
1991         },
1992 };
1993
1994 struct genl_family dp_vport_genl_family = {
1995         .id = GENL_ID_GENERATE,
1996         .hdrsize = sizeof(struct ovs_header),
1997         .name = OVS_VPORT_FAMILY,
1998         .version = OVS_VPORT_VERSION,
1999         .maxattr = OVS_VPORT_ATTR_MAX,
2000         .netnsok = true,
2001         .parallel_ops = true,
2002         .ops = dp_vport_genl_ops,
2003         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2004         .mcgrps = &ovs_dp_vport_multicast_group,
2005         .n_mcgrps = 1,
2006 };
2007
2008 static struct genl_family *dp_genl_families[] = {
2009         &dp_datapath_genl_family,
2010         &dp_vport_genl_family,
2011         &dp_flow_genl_family,
2012         &dp_packet_genl_family,
2013 };
2014
2015 static void dp_unregister_genl(int n_families)
2016 {
2017         int i;
2018
2019         for (i = 0; i < n_families; i++)
2020                 genl_unregister_family(dp_genl_families[i]);
2021 }
2022
2023 static int dp_register_genl(void)
2024 {
2025         int err;
2026         int i;
2027
2028         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2029
2030                 err = genl_register_family(dp_genl_families[i]);
2031                 if (err)
2032                         goto error;
2033         }
2034
2035         return 0;
2036
2037 error:
2038         dp_unregister_genl(i);
2039         return err;
2040 }
2041
2042 static int __net_init ovs_init_net(struct net *net)
2043 {
2044         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2045
2046         INIT_LIST_HEAD(&ovs_net->dps);
2047         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2048         return 0;
2049 }
2050
2051 static void __net_exit ovs_exit_net(struct net *net)
2052 {
2053         struct datapath *dp, *dp_next;
2054         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2055
2056         ovs_lock();
2057         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2058                 __dp_destroy(dp);
2059         ovs_unlock();
2060
2061         cancel_work_sync(&ovs_net->dp_notify_work);
2062 }
2063
2064 static struct pernet_operations ovs_net_ops = {
2065         .init = ovs_init_net,
2066         .exit = ovs_exit_net,
2067         .id   = &ovs_net_id,
2068         .size = sizeof(struct ovs_net),
2069 };
2070
2071 DEFINE_COMPAT_PNET_REG_FUNC(device);
2072
2073 static int __init dp_init(void)
2074 {
2075         int err;
2076
2077         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2078
2079         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2080                 VERSION);
2081
2082         err = ovs_flow_init();
2083         if (err)
2084                 goto error;
2085
2086         err = ovs_vport_init();
2087         if (err)
2088                 goto error_flow_exit;
2089
2090         err = register_pernet_device(&ovs_net_ops);
2091         if (err)
2092                 goto error_vport_exit;
2093
2094         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2095         if (err)
2096                 goto error_netns_exit;
2097
2098         err = dp_register_genl();
2099         if (err < 0)
2100                 goto error_unreg_notifier;
2101
2102         return 0;
2103
2104 error_unreg_notifier:
2105         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2106 error_netns_exit:
2107         unregister_pernet_device(&ovs_net_ops);
2108 error_vport_exit:
2109         ovs_vport_exit();
2110 error_flow_exit:
2111         ovs_flow_exit();
2112 error:
2113         return err;
2114 }
2115
2116 static void dp_cleanup(void)
2117 {
2118         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2119         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2120         unregister_pernet_device(&ovs_net_ops);
2121         rcu_barrier();
2122         ovs_vport_exit();
2123         ovs_flow_exit();
2124 }
2125
2126 module_init(dp_init);
2127 module_exit(dp_cleanup);
2128
2129 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2130 MODULE_LICENSE("GPL");
2131 MODULE_VERSION(VERSION);