datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007-2013 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/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 /* Check if need to build a reply message.
  68  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  69 static bool ovs_must_notify(struct genl_info *info,
  70                             const struct genl_multicast_group *grp)
  71 {
  72         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  73                 netlink_has_listeners(genl_info_net(info)->genl_sock, grp->id);
  74 }
  75
  76 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
  77                        struct genl_multicast_group *grp)
  78 {
  79         genl_notify(skb, genl_info_net(info), info->snd_portid,
  80                     grp->id, info->nlhdr, GFP_KERNEL);
  81 }
  82
  83 /**
  84  * DOC: Locking:
  85  *
  86  * All writes e.g. Writes to device state (add/remove datapath, port, set
  87  * operations on vports, etc.), Writes to other state (flow table
  88  * modifications, set miscellaneous datapath parameters, etc.) are protected
  89  * by ovs_lock.
  90  *
  91  * Reads are protected by RCU.
  92  *
  93  * There are a few special cases (mostly stats) that have their own
  94  * synchronization but they nest under all of above and don't interact with
  95  * each other.
  96  *
  97  * The RTNL lock nests inside ovs_mutex.
  98  */
  99
 100 static DEFINE_MUTEX(ovs_mutex);
 101
 102 void ovs_lock(void)
 103 {
 104         mutex_lock(&ovs_mutex);
 105 }
 106
 107 void ovs_unlock(void)
 108 {
 109         mutex_unlock(&ovs_mutex);
 110 }
 111
 112 #ifdef CONFIG_LOCKDEP
 113 int lockdep_ovsl_is_held(void)
 114 {
 115         if (debug_locks)
 116                 return lockdep_is_held(&ovs_mutex);
 117         else
 118                 return 1;
 119 }
 120 #endif
 121
 122 static struct vport *new_vport(const struct vport_parms *);
 123 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 124                              const struct dp_upcall_info *);
 125 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 126                                   const struct dp_upcall_info *);
 127
 128 /* Must be called with rcu_read_lock or ovs_mutex. */
 129 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 130 {
 131         struct datapath *dp = NULL;
 132         struct net_device *dev;
 133
 134         rcu_read_lock();
 135         dev = dev_get_by_index_rcu(net, dp_ifindex);
 136         if (dev) {
 137                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 138                 if (vport)
 139                         dp = vport->dp;
 140         }
 141         rcu_read_unlock();
 142
 143         return dp;
 144 }
 145
 146 /* Must be called with rcu_read_lock or ovs_mutex. */
 147 const char *ovs_dp_name(const struct datapath *dp)
 148 {
 149         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 150         return vport->ops->get_name(vport);
 151 }
 152
 153 static int get_dpifindex(struct datapath *dp)
 154 {
 155         struct vport *local;
 156         int ifindex;
 157
 158         rcu_read_lock();
 159
 160         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 161         if (local)
 162                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 163         else
 164                 ifindex = 0;
 165
 166         rcu_read_unlock();
 167
 168         return ifindex;
 169 }
 170
 171 static void destroy_dp_rcu(struct rcu_head *rcu)
 172 {
 173         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 174
 175         free_percpu(dp->stats_percpu);
 176         release_net(ovs_dp_get_net(dp));
 177         kfree(dp->ports);
 178         kfree(dp);
 179 }
 180
 181 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 182                                             u16 port_no)
 183 {
 184         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 185 }
 186
 187 /* Called with ovs_mutex or RCU read lock. */
 188 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 189 {
 190         struct vport *vport;
 191         struct hlist_head *head;
 192
 193         head = vport_hash_bucket(dp, port_no);
 194         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 195                 if (vport->port_no == port_no)
 196                         return vport;
 197         }
 198         return NULL;
 199 }
 200
 201 /* Called with ovs_mutex. */
 202 static struct vport *new_vport(const struct vport_parms *parms)
 203 {
 204         struct vport *vport;
 205
 206         vport = ovs_vport_add(parms);
 207         if (!IS_ERR(vport)) {
 208                 struct datapath *dp = parms->dp;
 209                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 210
 211                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 212         }
 213         return vport;
 214 }
 215
 216 void ovs_dp_detach_port(struct vport *p)
 217 {
 218         ASSERT_OVSL();
 219
 220         /* First drop references to device. */
 221         hlist_del_rcu(&p->dp_hash_node);
 222
 223         /* Then destroy it. */
 224         ovs_vport_del(p);
 225 }
 226
 227 /* Must be called with rcu_read_lock. */
 228 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 229 {
 230         struct datapath *dp = p->dp;
 231         struct sw_flow *flow;
 232         struct dp_stats_percpu *stats;
 233         struct sw_flow_key key;
 234         u64 *stats_counter;
 235         u32 n_mask_hit;
 236         int error;
 237
 238         stats = this_cpu_ptr(dp->stats_percpu);
 239
 240         /* Extract flow from 'skb' into 'key'. */
 241         error = ovs_flow_extract(skb, p->port_no, &key);
 242         if (unlikely(error)) {
 243                 kfree_skb(skb);
 244                 return;
 245         }
 246
 247         /* Look up flow. */
 248         flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
 249         if (unlikely(!flow)) {
 250                 struct dp_upcall_info upcall;
 251
 252                 upcall.cmd = OVS_PACKET_CMD_MISS;
 253                 upcall.key = &key;
 254                 upcall.userdata = NULL;
 255                 upcall.portid = p->upcall_portid;
 256                 ovs_dp_upcall(dp, skb, &upcall);
 257                 consume_skb(skb);
 258                 stats_counter = &stats->n_missed;
 259                 goto out;
 260         }
 261
 262         OVS_CB(skb)->flow = flow;
 263         OVS_CB(skb)->pkt_key = &key;
 264
 265         ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
 266         ovs_execute_actions(dp, skb);
 267         stats_counter = &stats->n_hit;
 268
 269 out:
 270         /* Update datapath statistics. */
 271         u64_stats_update_begin(&stats->sync);
 272         (*stats_counter)++;
 273         stats->n_mask_hit += n_mask_hit;
 274         u64_stats_update_end(&stats->sync);
 275 }
 276
 277 static struct genl_family dp_packet_genl_family = {
 278         .id = GENL_ID_GENERATE,
 279         .hdrsize = sizeof(struct ovs_header),
 280         .name = OVS_PACKET_FAMILY,
 281         .version = OVS_PACKET_VERSION,
 282         .maxattr = OVS_PACKET_ATTR_MAX,
 283         .netnsok = true,
 284          SET_PARALLEL_OPS
 285 };
 286
 287 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 288                   const struct dp_upcall_info *upcall_info)
 289 {
 290         struct dp_stats_percpu *stats;
 291         int err;
 292
 293         if (upcall_info->portid == 0) {
 294                 err = -ENOTCONN;
 295                 goto err;
 296         }
 297
 298         if (!skb_is_gso(skb))
 299                 err = queue_userspace_packet(dp, skb, upcall_info);
 300         else
 301                 err = queue_gso_packets(dp, skb, upcall_info);
 302         if (err)
 303                 goto err;
 304
 305         return 0;
 306
 307 err:
 308         stats = this_cpu_ptr(dp->stats_percpu);
 309
 310         u64_stats_update_begin(&stats->sync);
 311         stats->n_lost++;
 312         u64_stats_update_end(&stats->sync);
 313
 314         return err;
 315 }
 316
 317 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 318                              const struct dp_upcall_info *upcall_info)
 319 {
 320         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 321         struct dp_upcall_info later_info;
 322         struct sw_flow_key later_key;
 323         struct sk_buff *segs, *nskb;
 324         int err;
 325
 326         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 327         if (IS_ERR(segs))
 328                 return PTR_ERR(segs);
 329
 330         /* Queue all of the segments. */
 331         skb = segs;
 332         do {
 333                 err = queue_userspace_packet(dp, skb, upcall_info);
 334                 if (err)
 335                         break;
 336
 337                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 338                         /* The initial flow key extracted by ovs_flow_extract()
 339                          * in this case is for a first fragment, so we need to
 340                          * properly mark later fragments.
 341                          */
 342                         later_key = *upcall_info->key;
 343                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 344
 345                         later_info = *upcall_info;
 346                         later_info.key = &later_key;
 347                         upcall_info = &later_info;
 348                 }
 349         } while ((skb = skb->next));
 350
 351         /* Free all of the segments. */
 352         skb = segs;
 353         do {
 354                 nskb = skb->next;
 355                 if (err)
 356                         kfree_skb(skb);
 357                 else
 358                         consume_skb(skb);
 359         } while ((skb = nskb));
 360         return err;
 361 }
 362
 363 static size_t key_attr_size(void)
 364 {
 365         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 366                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 367                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
 368                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 369                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 370                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
 371                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
 372                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 373                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
 374                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 375                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 376                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 377                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 378                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 379                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 380                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 381                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 382                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 383                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 384 }
 385
 386 static size_t upcall_msg_size(const struct nlattr *userdata,
 387                               unsigned int hdrlen)
 388 {
 389         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 390                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 391                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 392
 393         /* OVS_PACKET_ATTR_USERDATA */
 394         if (userdata)
 395                 size += NLA_ALIGN(userdata->nla_len);
 396
 397         return size;
 398 }
 399
 400 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 401                                   const struct dp_upcall_info *upcall_info)
 402 {
 403         struct ovs_header *upcall;
 404         struct sk_buff *nskb = NULL;
 405         struct sk_buff *user_skb; /* to be queued to userspace */
 406         struct nlattr *nla;
 407         struct genl_info info = {
 408 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,14,0)
 409                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 410 #endif
 411                 .snd_portid = upcall_info->portid,
 412         };
 413         size_t len;
 414         unsigned int hlen;
 415         int err, dp_ifindex;
 416
 417         dp_ifindex = get_dpifindex(dp);
 418         if (!dp_ifindex)
 419                 return -ENODEV;
 420
 421         if (vlan_tx_tag_present(skb)) {
 422                 nskb = skb_clone(skb, GFP_ATOMIC);
 423                 if (!nskb)
 424                         return -ENOMEM;
 425
 426                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 427                 if (!nskb)
 428                         return -ENOMEM;
 429
 430                 vlan_set_tci(nskb, 0);
 431
 432                 skb = nskb;
 433         }
 434
 435         if (nla_attr_size(skb->len) > USHRT_MAX) {
 436                 err = -EFBIG;
 437                 goto out;
 438         }
 439
 440         /* Complete checksum if needed */
 441         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 442             (err = skb_checksum_help(skb)))
 443                 goto out;
 444
 445         /* Older versions of OVS user space enforce alignment of the last
 446          * Netlink attribute to NLA_ALIGNTO which would require extensive
 447          * padding logic. Only perform zerocopy if padding is not required.
 448          */
 449         if (dp->user_features & OVS_DP_F_UNALIGNED)
 450                 hlen = skb_zerocopy_headlen(skb);
 451         else
 452                 hlen = skb->len;
 453
 454         len = upcall_msg_size(upcall_info->userdata, hlen);
 455         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 456         if (!user_skb) {
 457                 err = -ENOMEM;
 458                 goto out;
 459         }
 460
 461         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 462                              0, upcall_info->cmd);
 463         upcall->dp_ifindex = dp_ifindex;
 464
 465         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 466         ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
 467         nla_nest_end(user_skb, nla);
 468
 469         if (upcall_info->userdata)
 470                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 471                           nla_len(upcall_info->userdata),
 472                           nla_data(upcall_info->userdata));
 473
 474         /* Only reserve room for attribute header, packet data is added
 475          * in skb_zerocopy() */
 476         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 477                 err = -ENOBUFS;
 478                 goto out;
 479         }
 480         nla->nla_len = nla_attr_size(skb->len);
 481
 482         skb_zerocopy(user_skb, skb, skb->len, hlen);
 483
 484         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 485         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 486                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 487
 488                 if (plen > 0)
 489                         memset(skb_put(user_skb, plen), 0, plen);
 490         }
 491
 492         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 493
 494         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 495 out:
 496         kfree_skb(nskb);
 497         return err;
 498 }
 499
 500 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 501 {
 502         struct ovs_header *ovs_header = info->userhdr;
 503         struct nlattr **a = info->attrs;
 504         struct sw_flow_actions *acts;
 505         struct sk_buff *packet;
 506         struct sw_flow *flow;
 507         struct datapath *dp;
 508         struct ethhdr *eth;
 509         int len;
 510         int err;
 511
 512         err = -EINVAL;
 513         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 514             !a[OVS_PACKET_ATTR_ACTIONS])
 515                 goto err;
 516
 517         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 518         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 519         err = -ENOMEM;
 520         if (!packet)
 521                 goto err;
 522         skb_reserve(packet, NET_IP_ALIGN);
 523
 524         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 525
 526         skb_reset_mac_header(packet);
 527         eth = eth_hdr(packet);
 528
 529         /* Normally, setting the skb 'protocol' field would be handled by a
 530          * call to eth_type_trans(), but it assumes there's a sending
 531          * device, which we may not have. */
 532         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 533                 packet->protocol = eth->h_proto;
 534         else
 535                 packet->protocol = htons(ETH_P_802_2);
 536
 537         /* Build an sw_flow for sending this packet. */
 538         flow = ovs_flow_alloc();
 539         err = PTR_ERR(flow);
 540         if (IS_ERR(flow))
 541                 goto err_kfree_skb;
 542
 543         err = ovs_flow_extract(packet, -1, &flow->key);
 544         if (err)
 545                 goto err_flow_free;
 546
 547         err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
 548         if (err)
 549                 goto err_flow_free;
 550         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 551         err = PTR_ERR(acts);
 552         if (IS_ERR(acts))
 553                 goto err_flow_free;
 554
 555         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 556                                    &flow->key, 0, &acts);
 557         rcu_assign_pointer(flow->sf_acts, acts);
 558         if (err)
 559                 goto err_flow_free;
 560
 561         OVS_CB(packet)->flow = flow;
 562         OVS_CB(packet)->pkt_key = &flow->key;
 563         packet->priority = flow->key.phy.priority;
 564         packet->mark = flow->key.phy.skb_mark;
 565
 566         rcu_read_lock();
 567         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 568         err = -ENODEV;
 569         if (!dp)
 570                 goto err_unlock;
 571
 572         local_bh_disable();
 573         err = ovs_execute_actions(dp, packet);
 574         local_bh_enable();
 575         rcu_read_unlock();
 576
 577         ovs_flow_free(flow, false);
 578         return err;
 579
 580 err_unlock:
 581         rcu_read_unlock();
 582 err_flow_free:
 583         ovs_flow_free(flow, false);
 584 err_kfree_skb:
 585         kfree_skb(packet);
 586 err:
 587         return err;
 588 }
 589
 590 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 591         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 592         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 593         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 594 };
 595
 596 static struct genl_ops dp_packet_genl_ops[] = {
 597         { .cmd = OVS_PACKET_CMD_EXECUTE,
 598           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 599           .policy = packet_policy,
 600           .doit = ovs_packet_cmd_execute
 601         }
 602 };
 603
 604 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
 605                          struct ovs_dp_megaflow_stats *mega_stats)
 606 {
 607         int i;
 608
 609         memset(mega_stats, 0, sizeof(*mega_stats));
 610
 611         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 612         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 613
 614         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 615
 616         for_each_possible_cpu(i) {
 617                 const struct dp_stats_percpu *percpu_stats;
 618                 struct dp_stats_percpu local_stats;
 619                 unsigned int start;
 620
 621                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 622
 623                 do {
 624                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 625                         local_stats = *percpu_stats;
 626                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 627
 628                 stats->n_hit += local_stats.n_hit;
 629                 stats->n_missed += local_stats.n_missed;
 630                 stats->n_lost += local_stats.n_lost;
 631                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 632         }
 633 }
 634
 635 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 636         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 637         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 638         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 639 };
 640
 641 static struct genl_family dp_flow_genl_family = {
 642         .id = GENL_ID_GENERATE,
 643         .hdrsize = sizeof(struct ovs_header),
 644         .name = OVS_FLOW_FAMILY,
 645         .version = OVS_FLOW_VERSION,
 646         .maxattr = OVS_FLOW_ATTR_MAX,
 647         .netnsok = true,
 648          SET_PARALLEL_OPS
 649 };
 650
 651 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 652         .name = OVS_FLOW_MCGROUP
 653 };
 654
 655 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 656 {
 657         return NLMSG_ALIGN(sizeof(struct ovs_header))
 658                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 659                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 660                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 661                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 662                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 663                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 664 }
 665
 666 /* Called with ovs_mutex or RCU read lock. */
 667 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 668                                   struct sk_buff *skb, u32 portid,
 669                                   u32 seq, u32 flags, u8 cmd)
 670 {
 671         const int skb_orig_len = skb->len;
 672         struct nlattr *start;
 673         struct ovs_flow_stats stats;
 674         __be16 tcp_flags;
 675         unsigned long used;
 676         struct ovs_header *ovs_header;
 677         struct nlattr *nla;
 678         int err;
 679
 680         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 681         if (!ovs_header)
 682                 return -EMSGSIZE;
 683
 684         ovs_header->dp_ifindex = dp_ifindex;
 685
 686         /* Fill flow key. */
 687         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 688         if (!nla)
 689                 goto nla_put_failure;
 690
 691         err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
 692         if (err)
 693                 goto error;
 694         nla_nest_end(skb, nla);
 695
 696         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 697         if (!nla)
 698                 goto nla_put_failure;
 699
 700         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 701         if (err)
 702                 goto error;
 703
 704         nla_nest_end(skb, nla);
 705
 706         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 707
 708         if (used &&
 709             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 710                 goto nla_put_failure;
 711
 712         if (stats.n_packets &&
 713             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 714                 goto nla_put_failure;
 715
 716         if ((u8)ntohs(tcp_flags) &&
 717              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 718                 goto nla_put_failure;
 719
 720         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 721          * this is the first flow to be dumped into 'skb'.  This is unusual for
 722          * Netlink but individual action lists can be longer than
 723          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 724          * The userspace caller can always fetch the actions separately if it
 725          * really wants them.  (Most userspace callers in fact don't care.)
 726          *
 727          * This can only fail for dump operations because the skb is always
 728          * properly sized for single flows.
 729          */
 730         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 731         if (start) {
 732                 const struct sw_flow_actions *sf_acts;
 733
 734                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 735                 err = ovs_nla_put_actions(sf_acts->actions,
 736                                           sf_acts->actions_len, skb);
 737
 738                 if (!err)
 739                         nla_nest_end(skb, start);
 740                 else {
 741                         if (skb_orig_len)
 742                                 goto error;
 743
 744                         nla_nest_cancel(skb, start);
 745                 }
 746         } else if (skb_orig_len)
 747                 goto nla_put_failure;
 748
 749         return genlmsg_end(skb, ovs_header);
 750
 751 nla_put_failure:
 752         err = -EMSGSIZE;
 753 error:
 754         genlmsg_cancel(skb, ovs_header);
 755         return err;
 756 }
 757
 758 /* May not be called with RCU read lock. */
 759 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 760                                                struct genl_info *info,
 761                                                bool always)
 762 {
 763         struct sk_buff *skb;
 764
 765         if (!always && !ovs_must_notify(info, &ovs_dp_flow_multicast_group))
 766                 return NULL;
 767
 768         skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
 769
 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 (!skb || IS_ERR(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_or_set(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_key key, masked_key;
 802         struct sw_flow *flow = NULL;
 803         struct sw_flow_mask mask;
 804         struct sk_buff *reply;
 805         struct datapath *dp;
 806         struct sw_flow_actions *acts = NULL;
 807         struct sw_flow_match match;
 808         int error;
 809
 810         /* Extract key. */
 811         error = -EINVAL;
 812         if (!a[OVS_FLOW_ATTR_KEY])
 813                 goto error;
 814
 815         ovs_match_init(&match, &key, &mask);
 816         error = ovs_nla_get_match(&match,
 817                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 818         if (error)
 819                 goto error;
 820
 821         /* Validate actions. */
 822         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 823                 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 824                 error = PTR_ERR(acts);
 825                 if (IS_ERR(acts))
 826                         goto error;
 827
 828                 ovs_flow_mask_key(&masked_key, &key, &mask);
 829                 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
 830                                              &masked_key, 0, &acts);
 831                 if (error) {
 832                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 833                         goto err_kfree;
 834                 }
 835         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 836                 /* OVS_FLOW_CMD_NEW must have actions. */
 837                 error = -EINVAL;
 838                 goto error;
 839         }
 840
 841         ovs_lock();
 842         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 843         error = -ENODEV;
 844         if (!dp)
 845                 goto err_unlock_ovs;
 846
 847         /* Check if this is a duplicate flow */
 848         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 849         if (!flow) {
 850                 /* Bail out if we're not allowed to create a new flow. */
 851                 error = -ENOENT;
 852                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
 853                         goto err_unlock_ovs;
 854
 855                 /* Allocate flow. */
 856                 flow = ovs_flow_alloc();
 857                 if (IS_ERR(flow)) {
 858                         error = PTR_ERR(flow);
 859                         goto err_unlock_ovs;
 860                 }
 861
 862                 flow->key = masked_key;
 863                 flow->unmasked_key = key;
 864                 rcu_assign_pointer(flow->sf_acts, acts);
 865
 866                 /* Put flow in bucket. */
 867                 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
 868                 if (error) {
 869                         acts = NULL;
 870                         goto err_flow_free;
 871                 }
 872
 873                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
 874                                                 info, OVS_FLOW_CMD_NEW, false);
 875         } else {
 876                 /* We found a matching flow. */
 877                 /* Bail out if we're not allowed to modify an existing flow.
 878                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 879                  * because Generic Netlink treats the latter as a dump
 880                  * request.  We also accept NLM_F_EXCL in case that bug ever
 881                  * gets fixed.
 882                  */
 883                 error = -EEXIST;
 884                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
 885                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
 886                         goto err_unlock_ovs;
 887
 888                 /* The unmasked key has to be the same for flow updates. */
 889                 if (!ovs_flow_cmp_unmasked_key(flow, &match))
 890                         goto err_unlock_ovs;
 891
 892                 /* Update actions, if present. */
 893                 if (acts) {
 894                         struct sw_flow_actions *old_acts;
 895
 896                         old_acts = ovsl_dereference(flow->sf_acts);
 897                         rcu_assign_pointer(flow->sf_acts, acts);
 898                         ovs_nla_free_flow_actions(old_acts);
 899                 }
 900
 901                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
 902                                                 info, OVS_FLOW_CMD_NEW, false);
 903
 904                 /* Clear stats. */
 905                 if (a[OVS_FLOW_ATTR_CLEAR])
 906                         ovs_flow_stats_clear(flow);
 907         }
 908         ovs_unlock();
 909
 910         if (reply) {
 911                 if (!IS_ERR(reply))
 912                         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
 913                 else
 914                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
 915                                         ovs_dp_flow_multicast_group.id,
 916                                         PTR_ERR(reply));
 917         }
 918         return 0;
 919
 920 err_flow_free:
 921         ovs_flow_free(flow, false);
 922 err_unlock_ovs:
 923         ovs_unlock();
 924 err_kfree:
 925         kfree(acts);
 926 error:
 927         return error;
 928 }
 929
 930 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
 931 {
 932         struct nlattr **a = info->attrs;
 933         struct ovs_header *ovs_header = info->userhdr;
 934         struct sw_flow_key key;
 935         struct sk_buff *reply;
 936         struct sw_flow *flow;
 937         struct datapath *dp;
 938         struct sw_flow_match match;
 939         int err;
 940
 941         if (!a[OVS_FLOW_ATTR_KEY]) {
 942                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
 943                 return -EINVAL;
 944         }
 945
 946         ovs_match_init(&match, &key, NULL);
 947         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
 948         if (err)
 949                 return err;
 950
 951         ovs_lock();
 952         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 953         if (!dp) {
 954                 err = -ENODEV;
 955                 goto unlock;
 956         }
 957
 958         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 959         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
 960                 err = -ENOENT;
 961                 goto unlock;
 962         }
 963
 964         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
 965                                         OVS_FLOW_CMD_NEW, true);
 966         if (IS_ERR(reply)) {
 967                 err = PTR_ERR(reply);
 968                 goto unlock;
 969         }
 970
 971         ovs_unlock();
 972         return genlmsg_reply(reply, info);
 973 unlock:
 974         ovs_unlock();
 975         return err;
 976 }
 977
 978 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
 979 {
 980         struct nlattr **a = info->attrs;
 981         struct ovs_header *ovs_header = info->userhdr;
 982         struct sw_flow_key key;
 983         struct sk_buff *reply;
 984         struct sw_flow *flow;
 985         struct datapath *dp;
 986         struct sw_flow_match match;
 987         int err;
 988
 989         ovs_lock();
 990         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 991         if (!dp) {
 992                 err = -ENODEV;
 993                 goto unlock;
 994         }
 995
 996         if (!a[OVS_FLOW_ATTR_KEY]) {
 997                 err = ovs_flow_tbl_flush(&dp->table);
 998                 goto unlock;
 999         }
1000
1001         ovs_match_init(&match, &key, NULL);
1002         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1003         if (err)
1004                 goto unlock;
1005
1006         flow = ovs_flow_tbl_lookup(&dp->table, &key);
1007         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
1008                 err = -ENOENT;
1009                 goto unlock;
1010         }
1011
1012         reply = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
1013                                         false);
1014         if (IS_ERR(reply)) {
1015                 err = PTR_ERR(reply);
1016                 goto unlock;
1017         }
1018
1019         ovs_flow_tbl_remove(&dp->table, flow);
1020
1021         if (reply) {
1022                 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1023                                              reply, info->snd_portid,
1024                                              info->snd_seq, 0,
1025                                              OVS_FLOW_CMD_DEL);
1026                 BUG_ON(err < 0);
1027         }
1028
1029         ovs_flow_free(flow, true);
1030         ovs_unlock();
1031
1032         if (reply)
1033                 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1034         return 0;
1035 unlock:
1036         ovs_unlock();
1037         return err;
1038 }
1039
1040 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1041 {
1042         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1043         struct table_instance *ti;
1044         struct datapath *dp;
1045
1046         rcu_read_lock();
1047         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1048         if (!dp) {
1049                 rcu_read_unlock();
1050                 return -ENODEV;
1051         }
1052
1053         ti = rcu_dereference(dp->table.ti);
1054         for (;;) {
1055                 struct sw_flow *flow;
1056                 u32 bucket, obj;
1057
1058                 bucket = cb->args[0];
1059                 obj = cb->args[1];
1060                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1061                 if (!flow)
1062                         break;
1063
1064                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1065                                            NETLINK_CB(cb->skb).portid,
1066                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1067                                            OVS_FLOW_CMD_NEW) < 0)
1068                         break;
1069
1070                 cb->args[0] = bucket;
1071                 cb->args[1] = obj;
1072         }
1073         rcu_read_unlock();
1074         return skb->len;
1075 }
1076
1077 static struct genl_ops dp_flow_genl_ops[] = {
1078         { .cmd = OVS_FLOW_CMD_NEW,
1079           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1080           .policy = flow_policy,
1081           .doit = ovs_flow_cmd_new_or_set
1082         },
1083         { .cmd = OVS_FLOW_CMD_DEL,
1084           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1085           .policy = flow_policy,
1086           .doit = ovs_flow_cmd_del
1087         },
1088         { .cmd = OVS_FLOW_CMD_GET,
1089           .flags = 0,               /* OK for unprivileged users. */
1090           .policy = flow_policy,
1091           .doit = ovs_flow_cmd_get,
1092           .dumpit = ovs_flow_cmd_dump
1093         },
1094         { .cmd = OVS_FLOW_CMD_SET,
1095           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1096           .policy = flow_policy,
1097           .doit = ovs_flow_cmd_new_or_set,
1098         },
1099 };
1100
1101 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1102         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1103         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1104         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1105 };
1106
1107 static struct genl_family dp_datapath_genl_family = {
1108         .id = GENL_ID_GENERATE,
1109         .hdrsize = sizeof(struct ovs_header),
1110         .name = OVS_DATAPATH_FAMILY,
1111         .version = OVS_DATAPATH_VERSION,
1112         .maxattr = OVS_DP_ATTR_MAX,
1113         .netnsok = true,
1114          SET_PARALLEL_OPS
1115 };
1116
1117 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1118         .name = OVS_DATAPATH_MCGROUP
1119 };
1120
1121 static size_t ovs_dp_cmd_msg_size(void)
1122 {
1123         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1124
1125         msgsize += nla_total_size(IFNAMSIZ);
1126         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1127         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1128         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1129
1130         return msgsize;
1131 }
1132
1133 /* Called with ovs_mutex or RCU read lock. */
1134 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1135                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1136 {
1137         struct ovs_header *ovs_header;
1138         struct ovs_dp_stats dp_stats;
1139         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1140         int err;
1141
1142         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1143                                    flags, cmd);
1144         if (!ovs_header)
1145                 goto error;
1146
1147         ovs_header->dp_ifindex = get_dpifindex(dp);
1148
1149         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1150         if (err)
1151                 goto nla_put_failure;
1152
1153         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1154         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1155                         &dp_stats))
1156                 goto nla_put_failure;
1157
1158         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1159                         sizeof(struct ovs_dp_megaflow_stats),
1160                         &dp_megaflow_stats))
1161                 goto nla_put_failure;
1162
1163         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1164                 goto nla_put_failure;
1165
1166         return genlmsg_end(skb, ovs_header);
1167
1168 nla_put_failure:
1169         genlmsg_cancel(skb, ovs_header);
1170 error:
1171         return -EMSGSIZE;
1172 }
1173
1174 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1175 {
1176         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1177 }
1178
1179 /* Called with rcu_read_lock or ovs_mutex. */
1180 static struct datapath *lookup_datapath(struct net *net,
1181                                         struct ovs_header *ovs_header,
1182                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1183 {
1184         struct datapath *dp;
1185
1186         if (!a[OVS_DP_ATTR_NAME])
1187                 dp = get_dp(net, ovs_header->dp_ifindex);
1188         else {
1189                 struct vport *vport;
1190
1191                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1192                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1193         }
1194         return dp ? dp : ERR_PTR(-ENODEV);
1195 }
1196
1197 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1198 {
1199         struct datapath *dp;
1200
1201         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1202         if (IS_ERR(dp))
1203                 return;
1204
1205         WARN(dp->user_features, "Dropping previously announced user features\n");
1206         dp->user_features = 0;
1207 }
1208
1209 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1210 {
1211         if (a[OVS_DP_ATTR_USER_FEATURES])
1212                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1213 }
1214
1215 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1216 {
1217         struct nlattr **a = info->attrs;
1218         struct vport_parms parms;
1219         struct sk_buff *reply;
1220         struct datapath *dp;
1221         struct vport *vport;
1222         struct ovs_net *ovs_net;
1223         int err, i;
1224
1225         err = -EINVAL;
1226         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1227                 goto err;
1228
1229         reply = ovs_dp_cmd_alloc_info(info);
1230         if (!reply)
1231                 return -ENOMEM;
1232
1233         err = -ENOMEM;
1234         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1235         if (dp == NULL)
1236                 goto err_free_reply;
1237
1238         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1239
1240         /* Allocate table. */
1241         err = ovs_flow_tbl_init(&dp->table);
1242         if (err)
1243                 goto err_free_dp;
1244
1245         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1246         if (!dp->stats_percpu) {
1247                 err = -ENOMEM;
1248                 goto err_destroy_table;
1249         }
1250
1251         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1252                             GFP_KERNEL);
1253         if (!dp->ports) {
1254                 err = -ENOMEM;
1255                 goto err_destroy_percpu;
1256         }
1257
1258         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1259                 INIT_HLIST_HEAD(&dp->ports[i]);
1260
1261         /* Set up our datapath device. */
1262         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1263         parms.type = OVS_VPORT_TYPE_INTERNAL;
1264         parms.options = NULL;
1265         parms.dp = dp;
1266         parms.port_no = OVSP_LOCAL;
1267         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1268
1269         ovs_dp_change(dp, a);
1270
1271         /* So far only local changes have been made, now need the lock. */
1272         ovs_lock();
1273
1274         vport = new_vport(&parms);
1275         if (IS_ERR(vport)) {
1276                 err = PTR_ERR(vport);
1277                 if (err == -EBUSY)
1278                         err = -EEXIST;
1279
1280                 if (err == -EEXIST) {
1281                         /* An outdated user space instance that does not understand
1282                          * the concept of user_features has attempted to create a new
1283                          * datapath and is likely to reuse it. Drop all user features.
1284                          */
1285                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1286                                 ovs_dp_reset_user_features(skb, info);
1287                 }
1288
1289                 goto err_destroy_ports_array;
1290         }
1291
1292         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1293                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1294         BUG_ON(err < 0);
1295
1296         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1297         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1298
1299         ovs_unlock();
1300
1301         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1302         return 0;
1303
1304 err_destroy_ports_array:
1305         ovs_unlock();
1306         kfree(dp->ports);
1307 err_destroy_percpu:
1308         free_percpu(dp->stats_percpu);
1309 err_destroy_table:
1310         ovs_flow_tbl_destroy(&dp->table, false);
1311 err_free_dp:
1312         release_net(ovs_dp_get_net(dp));
1313         kfree(dp);
1314 err_free_reply:
1315         kfree_skb(reply);
1316 err:
1317         return err;
1318 }
1319
1320 /* Called with ovs_mutex. */
1321 static void __dp_destroy(struct datapath *dp)
1322 {
1323         int i;
1324
1325         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1326                 struct vport *vport;
1327                 struct hlist_node *n;
1328
1329                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1330                         if (vport->port_no != OVSP_LOCAL)
1331                                 ovs_dp_detach_port(vport);
1332         }
1333
1334         list_del_rcu(&dp->list_node);
1335
1336         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1337          * all ports in datapath are destroyed first before freeing datapath.
1338          */
1339         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1340
1341         /* RCU destroy the flow table */
1342         ovs_flow_tbl_destroy(&dp->table, true);
1343
1344         call_rcu(&dp->rcu, destroy_dp_rcu);
1345 }
1346
1347 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1348 {
1349         struct sk_buff *reply;
1350         struct datapath *dp;
1351         int err;
1352
1353         reply = ovs_dp_cmd_alloc_info(info);
1354         if (!reply)
1355                 return -ENOMEM;
1356
1357         ovs_lock();
1358         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1359         err = PTR_ERR(dp);
1360         if (IS_ERR(dp))
1361                 goto err_unlock_free;
1362
1363         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1364                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1365         BUG_ON(err < 0);
1366
1367         __dp_destroy(dp);
1368
1369         ovs_unlock();
1370         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1371         return 0;
1372
1373 err_unlock_free:
1374         ovs_unlock();
1375         kfree_skb(reply);
1376         return err;
1377 }
1378
1379 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1380 {
1381         struct sk_buff *reply;
1382         struct datapath *dp;
1383         int err;
1384
1385         reply = ovs_dp_cmd_alloc_info(info);
1386         if (!reply)
1387                 return -ENOMEM;
1388
1389         ovs_lock();
1390         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1391         err = PTR_ERR(dp);
1392         if (IS_ERR(dp))
1393                 goto err_unlock_free;
1394
1395         ovs_dp_change(dp, info->attrs);
1396
1397         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1398                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1399         BUG_ON(err < 0);
1400
1401         ovs_unlock();
1402         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1403         return 0;
1404
1405 err_unlock_free:
1406         ovs_unlock();
1407         kfree_skb(reply);
1408         return err;
1409 }
1410
1411 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1412 {
1413         struct sk_buff *reply;
1414         struct datapath *dp;
1415         int err;
1416
1417         reply = ovs_dp_cmd_alloc_info(info);
1418         if (!reply)
1419                 return -ENOMEM;
1420
1421         rcu_read_lock();
1422         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1423         if (IS_ERR(dp)) {
1424                 err = PTR_ERR(dp);
1425                 goto err_unlock_free;
1426         }
1427         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1428                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1429         BUG_ON(err < 0);
1430         rcu_read_unlock();
1431
1432         return genlmsg_reply(reply, info);
1433
1434 err_unlock_free:
1435         rcu_read_unlock();
1436         kfree_skb(reply);
1437         return err;
1438 }
1439
1440 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1441 {
1442         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1443         struct datapath *dp;
1444         int skip = cb->args[0];
1445         int i = 0;
1446
1447         rcu_read_lock();
1448         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1449                 if (i >= skip &&
1450                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1451                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1452                                          OVS_DP_CMD_NEW) < 0)
1453                         break;
1454                 i++;
1455         }
1456         rcu_read_unlock();
1457
1458         cb->args[0] = i;
1459
1460         return skb->len;
1461 }
1462
1463 static struct genl_ops dp_datapath_genl_ops[] = {
1464         { .cmd = OVS_DP_CMD_NEW,
1465           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1466           .policy = datapath_policy,
1467           .doit = ovs_dp_cmd_new
1468         },
1469         { .cmd = OVS_DP_CMD_DEL,
1470           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1471           .policy = datapath_policy,
1472           .doit = ovs_dp_cmd_del
1473         },
1474         { .cmd = OVS_DP_CMD_GET,
1475           .flags = 0,               /* OK for unprivileged users. */
1476           .policy = datapath_policy,
1477           .doit = ovs_dp_cmd_get,
1478           .dumpit = ovs_dp_cmd_dump
1479         },
1480         { .cmd = OVS_DP_CMD_SET,
1481           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1482           .policy = datapath_policy,
1483           .doit = ovs_dp_cmd_set,
1484         },
1485 };
1486
1487 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1488         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1489         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1490         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1491         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1492         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1493         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1494 };
1495
1496 static struct genl_family dp_vport_genl_family = {
1497         .id = GENL_ID_GENERATE,
1498         .hdrsize = sizeof(struct ovs_header),
1499         .name = OVS_VPORT_FAMILY,
1500         .version = OVS_VPORT_VERSION,
1501         .maxattr = OVS_VPORT_ATTR_MAX,
1502         .netnsok = true,
1503          SET_PARALLEL_OPS
1504 };
1505
1506 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1507         .name = OVS_VPORT_MCGROUP
1508 };
1509
1510 /* Called with ovs_mutex or RCU read lock. */
1511 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1512                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1513 {
1514         struct ovs_header *ovs_header;
1515         struct ovs_vport_stats vport_stats;
1516         int err;
1517
1518         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1519                                  flags, cmd);
1520         if (!ovs_header)
1521                 return -EMSGSIZE;
1522
1523         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1524
1525         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1526             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1527             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1528             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1529                 goto nla_put_failure;
1530
1531         ovs_vport_get_stats(vport, &vport_stats);
1532         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1533                     &vport_stats))
1534                 goto nla_put_failure;
1535
1536         err = ovs_vport_get_options(vport, skb);
1537         if (err == -EMSGSIZE)
1538                 goto error;
1539
1540         return genlmsg_end(skb, ovs_header);
1541
1542 nla_put_failure:
1543         err = -EMSGSIZE;
1544 error:
1545         genlmsg_cancel(skb, ovs_header);
1546         return err;
1547 }
1548
1549 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1550 {
1551         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1552 }
1553
1554 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1555 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1556                                          u32 seq, u8 cmd)
1557 {
1558         struct sk_buff *skb;
1559         int retval;
1560
1561         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1562         if (!skb)
1563                 return ERR_PTR(-ENOMEM);
1564
1565         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1566         BUG_ON(retval < 0);
1567
1568         return skb;
1569 }
1570
1571 /* Called with ovs_mutex or RCU read lock. */
1572 static struct vport *lookup_vport(struct net *net,
1573                                   struct ovs_header *ovs_header,
1574                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1575 {
1576         struct datapath *dp;
1577         struct vport *vport;
1578
1579         if (a[OVS_VPORT_ATTR_NAME]) {
1580                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1581                 if (!vport)
1582                         return ERR_PTR(-ENODEV);
1583                 if (ovs_header->dp_ifindex &&
1584                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1585                         return ERR_PTR(-ENODEV);
1586                 return vport;
1587         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1588                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1589
1590                 if (port_no >= DP_MAX_PORTS)
1591                         return ERR_PTR(-EFBIG);
1592
1593                 dp = get_dp(net, ovs_header->dp_ifindex);
1594                 if (!dp)
1595                         return ERR_PTR(-ENODEV);
1596
1597                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1598                 if (!vport)
1599                         return ERR_PTR(-ENODEV);
1600                 return vport;
1601         } else
1602                 return ERR_PTR(-EINVAL);
1603 }
1604
1605 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1606 {
1607         struct nlattr **a = info->attrs;
1608         struct ovs_header *ovs_header = info->userhdr;
1609         struct vport_parms parms;
1610         struct sk_buff *reply;
1611         struct vport *vport;
1612         struct datapath *dp;
1613         u32 port_no;
1614         int err;
1615
1616         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1617             !a[OVS_VPORT_ATTR_UPCALL_PID])
1618                 return -EINVAL;
1619
1620         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1621                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1622         if (port_no >= DP_MAX_PORTS)
1623                 return -EFBIG;
1624
1625         reply = ovs_vport_cmd_alloc_info();
1626         if (!reply)
1627                 return -ENOMEM;
1628
1629         ovs_lock();
1630         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1631         err = -ENODEV;
1632         if (!dp)
1633                 goto exit_unlock_free;
1634
1635         if (port_no) {
1636                 vport = ovs_vport_ovsl(dp, port_no);
1637                 err = -EBUSY;
1638                 if (vport)
1639                         goto exit_unlock_free;
1640         } else {
1641                 for (port_no = 1; ; port_no++) {
1642                         if (port_no >= DP_MAX_PORTS) {
1643                                 err = -EFBIG;
1644                                 goto exit_unlock_free;
1645                         }
1646                         vport = ovs_vport_ovsl(dp, port_no);
1647                         if (!vport)
1648                                 break;
1649                 }
1650         }
1651
1652         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1653         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1654         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1655         parms.dp = dp;
1656         parms.port_no = port_no;
1657         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1658
1659         vport = new_vport(&parms);
1660         err = PTR_ERR(vport);
1661         if (IS_ERR(vport))
1662                 goto exit_unlock_free;
1663
1664         err = 0;
1665         if (a[OVS_VPORT_ATTR_STATS])
1666                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1667
1668         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1669                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1670         BUG_ON(err < 0);
1671         ovs_unlock();
1672
1673         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1674         return 0;
1675
1676 exit_unlock_free:
1677         ovs_unlock();
1678         kfree_skb(reply);
1679         return err;
1680 }
1681
1682 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1683 {
1684         struct nlattr **a = info->attrs;
1685         struct sk_buff *reply;
1686         struct vport *vport;
1687         int err;
1688
1689         reply = ovs_vport_cmd_alloc_info();
1690         if (!reply)
1691                 return -ENOMEM;
1692
1693         ovs_lock();
1694         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1695         err = PTR_ERR(vport);
1696         if (IS_ERR(vport))
1697                 goto exit_unlock_free;
1698
1699         if (a[OVS_VPORT_ATTR_TYPE] &&
1700             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1701                 err = -EINVAL;
1702                 goto exit_unlock_free;
1703         }
1704
1705         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1706                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1707                 if (err)
1708                         goto exit_unlock_free;
1709         }
1710
1711         if (a[OVS_VPORT_ATTR_STATS])
1712                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1713
1714         if (a[OVS_VPORT_ATTR_UPCALL_PID])
1715                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1716
1717         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1718                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1719         BUG_ON(err < 0);
1720         ovs_unlock();
1721
1722         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1723         return 0;
1724
1725 exit_unlock_free:
1726         ovs_unlock();
1727         kfree_skb(reply);
1728         return err;
1729 }
1730
1731 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1732 {
1733         struct nlattr **a = info->attrs;
1734         struct sk_buff *reply;
1735         struct vport *vport;
1736         int err;
1737
1738         reply = ovs_vport_cmd_alloc_info();
1739         if (!reply)
1740                 return -ENOMEM;
1741
1742         ovs_lock();
1743         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1744         err = PTR_ERR(vport);
1745         if (IS_ERR(vport))
1746                 goto exit_unlock_free;
1747
1748         if (vport->port_no == OVSP_LOCAL) {
1749                 err = -EINVAL;
1750                 goto exit_unlock_free;
1751         }
1752
1753         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1754                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1755         BUG_ON(err < 0);
1756         ovs_dp_detach_port(vport);
1757         ovs_unlock();
1758
1759         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1760         return 0;
1761
1762 exit_unlock_free:
1763         ovs_unlock();
1764         kfree_skb(reply);
1765         return err;
1766 }
1767
1768 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1769 {
1770         struct nlattr **a = info->attrs;
1771         struct ovs_header *ovs_header = info->userhdr;
1772         struct sk_buff *reply;
1773         struct vport *vport;
1774         int err;
1775
1776         reply = ovs_vport_cmd_alloc_info();
1777         if (!reply)
1778                 return -ENOMEM;
1779
1780         rcu_read_lock();
1781         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1782         err = PTR_ERR(vport);
1783         if (IS_ERR(vport))
1784                 goto exit_unlock_free;
1785         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1786                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1787         BUG_ON(err < 0);
1788         rcu_read_unlock();
1789
1790         return genlmsg_reply(reply, info);
1791
1792 exit_unlock_free:
1793         rcu_read_unlock();
1794         kfree_skb(reply);
1795         return err;
1796 }
1797
1798 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1799 {
1800         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1801         struct datapath *dp;
1802         int bucket = cb->args[0], skip = cb->args[1];
1803         int i, j = 0;
1804
1805         rcu_read_lock();
1806         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1807         if (!dp) {
1808                 rcu_read_unlock();
1809                 return -ENODEV;
1810         }
1811         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1812                 struct vport *vport;
1813
1814                 j = 0;
1815                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1816                         if (j >= skip &&
1817                             ovs_vport_cmd_fill_info(vport, skb,
1818                                                     NETLINK_CB(cb->skb).portid,
1819                                                     cb->nlh->nlmsg_seq,
1820                                                     NLM_F_MULTI,
1821                                                     OVS_VPORT_CMD_NEW) < 0)
1822                                 goto out;
1823
1824                         j++;
1825                 }
1826                 skip = 0;
1827         }
1828 out:
1829         rcu_read_unlock();
1830
1831         cb->args[0] = i;
1832         cb->args[1] = j;
1833
1834         return skb->len;
1835 }
1836
1837 static struct genl_ops dp_vport_genl_ops[] = {
1838         { .cmd = OVS_VPORT_CMD_NEW,
1839           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1840           .policy = vport_policy,
1841           .doit = ovs_vport_cmd_new
1842         },
1843         { .cmd = OVS_VPORT_CMD_DEL,
1844           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1845           .policy = vport_policy,
1846           .doit = ovs_vport_cmd_del
1847         },
1848         { .cmd = OVS_VPORT_CMD_GET,
1849           .flags = 0,               /* OK for unprivileged users. */
1850           .policy = vport_policy,
1851           .doit = ovs_vport_cmd_get,
1852           .dumpit = ovs_vport_cmd_dump
1853         },
1854         { .cmd = OVS_VPORT_CMD_SET,
1855           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1856           .policy = vport_policy,
1857           .doit = ovs_vport_cmd_set,
1858         },
1859 };
1860
1861 struct genl_family_and_ops {
1862         struct genl_family *family;
1863         struct genl_ops *ops;
1864         int n_ops;
1865         struct genl_multicast_group *group;
1866 };
1867
1868 static const struct genl_family_and_ops dp_genl_families[] = {
1869         { &dp_datapath_genl_family,
1870           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1871           &ovs_dp_datapath_multicast_group },
1872         { &dp_vport_genl_family,
1873           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1874           &ovs_dp_vport_multicast_group },
1875         { &dp_flow_genl_family,
1876           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1877           &ovs_dp_flow_multicast_group },
1878         { &dp_packet_genl_family,
1879           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1880           NULL },
1881 };
1882
1883 static void dp_unregister_genl(int n_families)
1884 {
1885         int i;
1886
1887         for (i = 0; i < n_families; i++)
1888                 genl_unregister_family(dp_genl_families[i].family);
1889 }
1890
1891 static int dp_register_genl(void)
1892 {
1893         int n_registered;
1894         int err;
1895         int i;
1896
1897         n_registered = 0;
1898         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1899                 const struct genl_family_and_ops *f = &dp_genl_families[i];
1900
1901                 err = genl_register_family_with_ops(f->family, f->ops,
1902                                                     f->n_ops);
1903                 if (err)
1904                         goto error;
1905                 n_registered++;
1906
1907                 if (f->group) {
1908                         err = genl_register_mc_group(f->family, f->group);
1909                         if (err)
1910                                 goto error;
1911                 }
1912         }
1913
1914         return 0;
1915
1916 error:
1917         dp_unregister_genl(n_registered);
1918         return err;
1919 }
1920
1921 static int __net_init ovs_init_net(struct net *net)
1922 {
1923         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1924
1925         INIT_LIST_HEAD(&ovs_net->dps);
1926         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1927         return 0;
1928 }
1929
1930 static void __net_exit ovs_exit_net(struct net *net)
1931 {
1932         struct datapath *dp, *dp_next;
1933         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1934
1935         ovs_lock();
1936         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1937                 __dp_destroy(dp);
1938         ovs_unlock();
1939
1940         cancel_work_sync(&ovs_net->dp_notify_work);
1941 }
1942
1943 static struct pernet_operations ovs_net_ops = {
1944         .init = ovs_init_net,
1945         .exit = ovs_exit_net,
1946         .id   = &ovs_net_id,
1947         .size = sizeof(struct ovs_net),
1948 };
1949
1950 DEFINE_COMPAT_PNET_REG_FUNC(device);
1951
1952 static int __init dp_init(void)
1953 {
1954         int err;
1955
1956         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1957
1958         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
1959                 VERSION);
1960
1961         err = ovs_flow_init();
1962         if (err)
1963                 goto error;
1964
1965         err = ovs_vport_init();
1966         if (err)
1967                 goto error_flow_exit;
1968
1969         err = register_pernet_device(&ovs_net_ops);
1970         if (err)
1971                 goto error_vport_exit;
1972
1973         err = register_netdevice_notifier(&ovs_dp_device_notifier);
1974         if (err)
1975                 goto error_netns_exit;
1976
1977         err = dp_register_genl();
1978         if (err < 0)
1979                 goto error_unreg_notifier;
1980
1981         return 0;
1982
1983 error_unreg_notifier:
1984         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1985 error_netns_exit:
1986         unregister_pernet_device(&ovs_net_ops);
1987 error_vport_exit:
1988         ovs_vport_exit();
1989 error_flow_exit:
1990         ovs_flow_exit();
1991 error:
1992         return err;
1993 }
1994
1995 static void dp_cleanup(void)
1996 {
1997         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1998         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1999         unregister_pernet_device(&ovs_net_ops);
2000         rcu_barrier();
2001         ovs_vport_exit();
2002         ovs_flow_exit();
2003 }
2004
2005 module_init(dp_init);
2006 module_exit(dp_cleanup);
2007
2008 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2009 MODULE_LICENSE("GPL");
2010 MODULE_VERSION(VERSION);