Pileus Git - ~andy/linux/blob - net/openvswitch/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/ethtool.h>
  40 #include <linux/wait.h>
  41 #include <asm/div64.h>
  42 #include <linux/highmem.h>
  43 #include <linux/netfilter_bridge.h>
  44 #include <linux/netfilter_ipv4.h>
  45 #include <linux/inetdevice.h>
  46 #include <linux/list.h>
  47 #include <linux/lockdep.h>
  48 #include <linux/openvswitch.h>
  49 #include <linux/rculist.h>
  50 #include <linux/dmi.h>
  51 #include <linux/workqueue.h>
  52 #include <net/genetlink.h>
  53 #include <net/net_namespace.h>
  54 #include <net/netns/generic.h>
  55
  56 #include "datapath.h"
  57 #include "flow.h"
  58 #include "flow_table.h"
  59 #include "flow_netlink.h"
  60 #include "vport-internal_dev.h"
  61 #include "vport-netdev.h"
  62
  63 int ovs_net_id __read_mostly;
  64
  65 static void ovs_notify(struct genl_family *family,
  66                        struct sk_buff *skb, struct genl_info *info)
  67 {
  68         genl_notify(family, skb, genl_info_net(info), info->snd_portid,
  69                     0, info->nlhdr, GFP_KERNEL);
  70 }
  71
  72 /**
  73  * DOC: Locking:
  74  *
  75  * All writes e.g. Writes to device state (add/remove datapath, port, set
  76  * operations on vports, etc.), Writes to other state (flow table
  77  * modifications, set miscellaneous datapath parameters, etc.) are protected
  78  * by ovs_lock.
  79  *
  80  * Reads are protected by RCU.
  81  *
  82  * There are a few special cases (mostly stats) that have their own
  83  * synchronization but they nest under all of above and don't interact with
  84  * each other.
  85  *
  86  * The RTNL lock nests inside ovs_mutex.
  87  */
  88
  89 static DEFINE_MUTEX(ovs_mutex);
  90
  91 void ovs_lock(void)
  92 {
  93         mutex_lock(&ovs_mutex);
  94 }
  95
  96 void ovs_unlock(void)
  97 {
  98         mutex_unlock(&ovs_mutex);
  99 }
 100
 101 #ifdef CONFIG_LOCKDEP
 102 int lockdep_ovsl_is_held(void)
 103 {
 104         if (debug_locks)
 105                 return lockdep_is_held(&ovs_mutex);
 106         else
 107                 return 1;
 108 }
 109 #endif
 110
 111 static struct vport *new_vport(const struct vport_parms *);
 112 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 113                              const struct dp_upcall_info *);
 114 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 115                                   const struct dp_upcall_info *);
 116
 117 /* Must be called with rcu_read_lock or ovs_mutex. */
 118 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 119 {
 120         struct datapath *dp = NULL;
 121         struct net_device *dev;
 122
 123         rcu_read_lock();
 124         dev = dev_get_by_index_rcu(net, dp_ifindex);
 125         if (dev) {
 126                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 127                 if (vport)
 128                         dp = vport->dp;
 129         }
 130         rcu_read_unlock();
 131
 132         return dp;
 133 }
 134
 135 /* Must be called with rcu_read_lock or ovs_mutex. */
 136 static const char *ovs_dp_name(const struct datapath *dp)
 137 {
 138         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 139         return vport->ops->get_name(vport);
 140 }
 141
 142 static int get_dpifindex(struct datapath *dp)
 143 {
 144         struct vport *local;
 145         int ifindex;
 146
 147         rcu_read_lock();
 148
 149         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 150         if (local)
 151                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 152         else
 153                 ifindex = 0;
 154
 155         rcu_read_unlock();
 156
 157         return ifindex;
 158 }
 159
 160 static void destroy_dp_rcu(struct rcu_head *rcu)
 161 {
 162         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 163
 164         free_percpu(dp->stats_percpu);
 165         release_net(ovs_dp_get_net(dp));
 166         kfree(dp->ports);
 167         kfree(dp);
 168 }
 169
 170 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 171                                             u16 port_no)
 172 {
 173         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 174 }
 175
 176 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 177 {
 178         struct vport *vport;
 179         struct hlist_head *head;
 180
 181         head = vport_hash_bucket(dp, port_no);
 182         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 183                 if (vport->port_no == port_no)
 184                         return vport;
 185         }
 186         return NULL;
 187 }
 188
 189 /* Called with ovs_mutex. */
 190 static struct vport *new_vport(const struct vport_parms *parms)
 191 {
 192         struct vport *vport;
 193
 194         vport = ovs_vport_add(parms);
 195         if (!IS_ERR(vport)) {
 196                 struct datapath *dp = parms->dp;
 197                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 198
 199                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 200         }
 201         return vport;
 202 }
 203
 204 void ovs_dp_detach_port(struct vport *p)
 205 {
 206         ASSERT_OVSL();
 207
 208         /* First drop references to device. */
 209         hlist_del_rcu(&p->dp_hash_node);
 210
 211         /* Then destroy it. */
 212         ovs_vport_del(p);
 213 }
 214
 215 /* Must be called with rcu_read_lock. */
 216 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 217 {
 218         struct datapath *dp = p->dp;
 219         struct sw_flow *flow;
 220         struct dp_stats_percpu *stats;
 221         struct sw_flow_key key;
 222         u64 *stats_counter;
 223         u32 n_mask_hit;
 224         int error;
 225
 226         stats = this_cpu_ptr(dp->stats_percpu);
 227
 228         /* Extract flow from 'skb' into 'key'. */
 229         error = ovs_flow_extract(skb, p->port_no, &key);
 230         if (unlikely(error)) {
 231                 kfree_skb(skb);
 232                 return;
 233         }
 234
 235         /* Look up flow. */
 236         flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
 237         if (unlikely(!flow)) {
 238                 struct dp_upcall_info upcall;
 239
 240                 upcall.cmd = OVS_PACKET_CMD_MISS;
 241                 upcall.key = &key;
 242                 upcall.userdata = NULL;
 243                 upcall.portid = p->upcall_portid;
 244                 ovs_dp_upcall(dp, skb, &upcall);
 245                 consume_skb(skb);
 246                 stats_counter = &stats->n_missed;
 247                 goto out;
 248         }
 249
 250         OVS_CB(skb)->flow = flow;
 251         OVS_CB(skb)->pkt_key = &key;
 252
 253         ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
 254         ovs_execute_actions(dp, skb);
 255         stats_counter = &stats->n_hit;
 256
 257 out:
 258         /* Update datapath statistics. */
 259         u64_stats_update_begin(&stats->sync);
 260         (*stats_counter)++;
 261         stats->n_mask_hit += n_mask_hit;
 262         u64_stats_update_end(&stats->sync);
 263 }
 264
 265 static struct genl_family dp_packet_genl_family = {
 266         .id = GENL_ID_GENERATE,
 267         .hdrsize = sizeof(struct ovs_header),
 268         .name = OVS_PACKET_FAMILY,
 269         .version = OVS_PACKET_VERSION,
 270         .maxattr = OVS_PACKET_ATTR_MAX,
 271         .netnsok = true,
 272         .parallel_ops = true,
 273 };
 274
 275 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 276                   const struct dp_upcall_info *upcall_info)
 277 {
 278         struct dp_stats_percpu *stats;
 279         int err;
 280
 281         if (upcall_info->portid == 0) {
 282                 err = -ENOTCONN;
 283                 goto err;
 284         }
 285
 286         if (!skb_is_gso(skb))
 287                 err = queue_userspace_packet(dp, skb, upcall_info);
 288         else
 289                 err = queue_gso_packets(dp, skb, upcall_info);
 290         if (err)
 291                 goto err;
 292
 293         return 0;
 294
 295 err:
 296         stats = this_cpu_ptr(dp->stats_percpu);
 297
 298         u64_stats_update_begin(&stats->sync);
 299         stats->n_lost++;
 300         u64_stats_update_end(&stats->sync);
 301
 302         return err;
 303 }
 304
 305 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 306                              const struct dp_upcall_info *upcall_info)
 307 {
 308         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 309         struct dp_upcall_info later_info;
 310         struct sw_flow_key later_key;
 311         struct sk_buff *segs, *nskb;
 312         int err;
 313
 314         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 315         if (IS_ERR(segs))
 316                 return PTR_ERR(segs);
 317
 318         /* Queue all of the segments. */
 319         skb = segs;
 320         do {
 321                 err = queue_userspace_packet(dp, skb, upcall_info);
 322                 if (err)
 323                         break;
 324
 325                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 326                         /* The initial flow key extracted by ovs_flow_extract()
 327                          * in this case is for a first fragment, so we need to
 328                          * properly mark later fragments.
 329                          */
 330                         later_key = *upcall_info->key;
 331                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 332
 333                         later_info = *upcall_info;
 334                         later_info.key = &later_key;
 335                         upcall_info = &later_info;
 336                 }
 337         } while ((skb = skb->next));
 338
 339         /* Free all of the segments. */
 340         skb = segs;
 341         do {
 342                 nskb = skb->next;
 343                 if (err)
 344                         kfree_skb(skb);
 345                 else
 346                         consume_skb(skb);
 347         } while ((skb = nskb));
 348         return err;
 349 }
 350
 351 static size_t key_attr_size(void)
 352 {
 353         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 354                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 355                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
 356                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 357                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 358                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
 359                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
 360                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 361                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
 362                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 363                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 364                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 365                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 366                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 367                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 368                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 369                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 370                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 371                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 372 }
 373
 374 static size_t upcall_msg_size(const struct nlattr *userdata,
 375                               unsigned int hdrlen)
 376 {
 377         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 378                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 379                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 380
 381         /* OVS_PACKET_ATTR_USERDATA */
 382         if (userdata)
 383                 size += NLA_ALIGN(userdata->nla_len);
 384
 385         return size;
 386 }
 387
 388 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 389                                   const struct dp_upcall_info *upcall_info)
 390 {
 391         struct ovs_header *upcall;
 392         struct sk_buff *nskb = NULL;
 393         struct sk_buff *user_skb; /* to be queued to userspace */
 394         struct nlattr *nla;
 395         struct genl_info info = {
 396                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 397                 .snd_portid = upcall_info->portid,
 398         };
 399         size_t len;
 400         unsigned int hlen;
 401         int err, dp_ifindex;
 402
 403         dp_ifindex = get_dpifindex(dp);
 404         if (!dp_ifindex)
 405                 return -ENODEV;
 406
 407         if (vlan_tx_tag_present(skb)) {
 408                 nskb = skb_clone(skb, GFP_ATOMIC);
 409                 if (!nskb)
 410                         return -ENOMEM;
 411
 412                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 413                 if (!nskb)
 414                         return -ENOMEM;
 415
 416                 nskb->vlan_tci = 0;
 417                 skb = nskb;
 418         }
 419
 420         if (nla_attr_size(skb->len) > USHRT_MAX) {
 421                 err = -EFBIG;
 422                 goto out;
 423         }
 424
 425         /* Complete checksum if needed */
 426         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 427             (err = skb_checksum_help(skb)))
 428                 goto out;
 429
 430         /* Older versions of OVS user space enforce alignment of the last
 431          * Netlink attribute to NLA_ALIGNTO which would require extensive
 432          * padding logic. Only perform zerocopy if padding is not required.
 433          */
 434         if (dp->user_features & OVS_DP_F_UNALIGNED)
 435                 hlen = skb_zerocopy_headlen(skb);
 436         else
 437                 hlen = skb->len;
 438
 439         len = upcall_msg_size(upcall_info->userdata, hlen);
 440         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 441         if (!user_skb) {
 442                 err = -ENOMEM;
 443                 goto out;
 444         }
 445
 446         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 447                              0, upcall_info->cmd);
 448         upcall->dp_ifindex = dp_ifindex;
 449
 450         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 451         ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
 452         nla_nest_end(user_skb, nla);
 453
 454         if (upcall_info->userdata)
 455                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 456                           nla_len(upcall_info->userdata),
 457                           nla_data(upcall_info->userdata));
 458
 459         /* Only reserve room for attribute header, packet data is added
 460          * in skb_zerocopy() */
 461         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 462                 err = -ENOBUFS;
 463                 goto out;
 464         }
 465         nla->nla_len = nla_attr_size(skb->len);
 466
 467         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 468         if (err)
 469                 goto out;
 470
 471         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 472         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 473                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 474
 475                 if (plen > 0)
 476                         memset(skb_put(user_skb, plen), 0, plen);
 477         }
 478
 479         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 480
 481         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 482 out:
 483         if (err)
 484                 skb_tx_error(skb);
 485         kfree_skb(nskb);
 486         return err;
 487 }
 488
 489 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 490 {
 491         struct ovs_header *ovs_header = info->userhdr;
 492         struct nlattr **a = info->attrs;
 493         struct sw_flow_actions *acts;
 494         struct sk_buff *packet;
 495         struct sw_flow *flow;
 496         struct datapath *dp;
 497         struct ethhdr *eth;
 498         int len;
 499         int err;
 500
 501         err = -EINVAL;
 502         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 503             !a[OVS_PACKET_ATTR_ACTIONS])
 504                 goto err;
 505
 506         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 507         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 508         err = -ENOMEM;
 509         if (!packet)
 510                 goto err;
 511         skb_reserve(packet, NET_IP_ALIGN);
 512
 513         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 514
 515         skb_reset_mac_header(packet);
 516         eth = eth_hdr(packet);
 517
 518         /* Normally, setting the skb 'protocol' field would be handled by a
 519          * call to eth_type_trans(), but it assumes there's a sending
 520          * device, which we may not have. */
 521         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 522                 packet->protocol = eth->h_proto;
 523         else
 524                 packet->protocol = htons(ETH_P_802_2);
 525
 526         /* Build an sw_flow for sending this packet. */
 527         flow = ovs_flow_alloc(false);
 528         err = PTR_ERR(flow);
 529         if (IS_ERR(flow))
 530                 goto err_kfree_skb;
 531
 532         err = ovs_flow_extract(packet, -1, &flow->key);
 533         if (err)
 534                 goto err_flow_free;
 535
 536         err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
 537         if (err)
 538                 goto err_flow_free;
 539         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 540         err = PTR_ERR(acts);
 541         if (IS_ERR(acts))
 542                 goto err_flow_free;
 543
 544         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 545                                    &flow->key, 0, &acts);
 546         rcu_assign_pointer(flow->sf_acts, acts);
 547         if (err)
 548                 goto err_flow_free;
 549
 550         OVS_CB(packet)->flow = flow;
 551         OVS_CB(packet)->pkt_key = &flow->key;
 552         packet->priority = flow->key.phy.priority;
 553         packet->mark = flow->key.phy.skb_mark;
 554
 555         rcu_read_lock();
 556         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 557         err = -ENODEV;
 558         if (!dp)
 559                 goto err_unlock;
 560
 561         local_bh_disable();
 562         err = ovs_execute_actions(dp, packet);
 563         local_bh_enable();
 564         rcu_read_unlock();
 565
 566         ovs_flow_free(flow, false);
 567         return err;
 568
 569 err_unlock:
 570         rcu_read_unlock();
 571 err_flow_free:
 572         ovs_flow_free(flow, false);
 573 err_kfree_skb:
 574         kfree_skb(packet);
 575 err:
 576         return err;
 577 }
 578
 579 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 580         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 581         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 582         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 583 };
 584
 585 static const struct genl_ops dp_packet_genl_ops[] = {
 586         { .cmd = OVS_PACKET_CMD_EXECUTE,
 587           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 588           .policy = packet_policy,
 589           .doit = ovs_packet_cmd_execute
 590         }
 591 };
 592
 593 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
 594                          struct ovs_dp_megaflow_stats *mega_stats)
 595 {
 596         int i;
 597
 598         memset(mega_stats, 0, sizeof(*mega_stats));
 599
 600         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 601         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 602
 603         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 604
 605         for_each_possible_cpu(i) {
 606                 const struct dp_stats_percpu *percpu_stats;
 607                 struct dp_stats_percpu local_stats;
 608                 unsigned int start;
 609
 610                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 611
 612                 do {
 613                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 614                         local_stats = *percpu_stats;
 615                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 616
 617                 stats->n_hit += local_stats.n_hit;
 618                 stats->n_missed += local_stats.n_missed;
 619                 stats->n_lost += local_stats.n_lost;
 620                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 621         }
 622 }
 623
 624 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 625         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 626         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 627         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 628 };
 629
 630 static struct genl_family dp_flow_genl_family = {
 631         .id = GENL_ID_GENERATE,
 632         .hdrsize = sizeof(struct ovs_header),
 633         .name = OVS_FLOW_FAMILY,
 634         .version = OVS_FLOW_VERSION,
 635         .maxattr = OVS_FLOW_ATTR_MAX,
 636         .netnsok = true,
 637         .parallel_ops = true,
 638 };
 639
 640 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 641         .name = OVS_FLOW_MCGROUP
 642 };
 643
 644 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 645 {
 646         return NLMSG_ALIGN(sizeof(struct ovs_header))
 647                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 648                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 649                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 650                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 651                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 652                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 653 }
 654
 655 /* Called with ovs_mutex. */
 656 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 657                                   struct sk_buff *skb, u32 portid,
 658                                   u32 seq, u32 flags, u8 cmd)
 659 {
 660         const int skb_orig_len = skb->len;
 661         struct nlattr *start;
 662         struct ovs_flow_stats stats;
 663         __be16 tcp_flags;
 664         unsigned long used;
 665         struct ovs_header *ovs_header;
 666         struct nlattr *nla;
 667         int err;
 668
 669         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 670         if (!ovs_header)
 671                 return -EMSGSIZE;
 672
 673         ovs_header->dp_ifindex = get_dpifindex(dp);
 674
 675         /* Fill flow key. */
 676         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 677         if (!nla)
 678                 goto nla_put_failure;
 679
 680         err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
 681         if (err)
 682                 goto error;
 683         nla_nest_end(skb, nla);
 684
 685         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 686         if (!nla)
 687                 goto nla_put_failure;
 688
 689         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 690         if (err)
 691                 goto error;
 692
 693         nla_nest_end(skb, nla);
 694
 695         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 696         if (used &&
 697             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 698                 goto nla_put_failure;
 699
 700         if (stats.n_packets &&
 701             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 702                 goto nla_put_failure;
 703
 704         if ((u8)ntohs(tcp_flags) &&
 705              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 706                 goto nla_put_failure;
 707
 708         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 709          * this is the first flow to be dumped into 'skb'.  This is unusual for
 710          * Netlink but individual action lists can be longer than
 711          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 712          * The userspace caller can always fetch the actions separately if it
 713          * really wants them.  (Most userspace callers in fact don't care.)
 714          *
 715          * This can only fail for dump operations because the skb is always
 716          * properly sized for single flows.
 717          */
 718         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 719         if (start) {
 720                 const struct sw_flow_actions *sf_acts;
 721
 722                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 723
 724                 err = ovs_nla_put_actions(sf_acts->actions,
 725                                           sf_acts->actions_len, skb);
 726                 if (!err)
 727                         nla_nest_end(skb, start);
 728                 else {
 729                         if (skb_orig_len)
 730                                 goto error;
 731
 732                         nla_nest_cancel(skb, start);
 733                 }
 734         } else if (skb_orig_len)
 735                 goto nla_put_failure;
 736
 737         return genlmsg_end(skb, ovs_header);
 738
 739 nla_put_failure:
 740         err = -EMSGSIZE;
 741 error:
 742         genlmsg_cancel(skb, ovs_header);
 743         return err;
 744 }
 745
 746 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow,
 747                                                struct genl_info *info)
 748 {
 749         size_t len;
 750
 751         len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts));
 752
 753         return genlmsg_new_unicast(len, info, GFP_KERNEL);
 754 }
 755
 756 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 757                                                struct datapath *dp,
 758                                                struct genl_info *info,
 759                                                u8 cmd)
 760 {
 761         struct sk_buff *skb;
 762         int retval;
 763
 764         skb = ovs_flow_cmd_alloc_info(flow, info);
 765         if (!skb)
 766                 return ERR_PTR(-ENOMEM);
 767
 768         retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid,
 769                                         info->snd_seq, 0, cmd);
 770         BUG_ON(retval < 0);
 771         return skb;
 772 }
 773
 774 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 775 {
 776         struct nlattr **a = info->attrs;
 777         struct ovs_header *ovs_header = info->userhdr;
 778         struct sw_flow_key key, masked_key;
 779         struct sw_flow *flow = NULL;
 780         struct sw_flow_mask mask;
 781         struct sk_buff *reply;
 782         struct datapath *dp;
 783         struct sw_flow_actions *acts = NULL;
 784         struct sw_flow_match match;
 785         bool exact_5tuple;
 786         int error;
 787
 788         /* Extract key. */
 789         error = -EINVAL;
 790         if (!a[OVS_FLOW_ATTR_KEY])
 791                 goto error;
 792
 793         ovs_match_init(&match, &key, &mask);
 794         error = ovs_nla_get_match(&match, &exact_5tuple,
 795                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 796         if (error)
 797                 goto error;
 798
 799         /* Validate actions. */
 800         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 801                 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 802                 error = PTR_ERR(acts);
 803                 if (IS_ERR(acts))
 804                         goto error;
 805
 806                 ovs_flow_mask_key(&masked_key, &key, &mask);
 807                 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
 808                                              &masked_key, 0, &acts);
 809                 if (error) {
 810                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 811                         goto err_kfree;
 812                 }
 813         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 814                 error = -EINVAL;
 815                 goto error;
 816         }
 817
 818         ovs_lock();
 819         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 820         error = -ENODEV;
 821         if (!dp)
 822                 goto err_unlock_ovs;
 823
 824         /* Check if this is a duplicate flow */
 825         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 826         if (!flow) {
 827                 /* Bail out if we're not allowed to create a new flow. */
 828                 error = -ENOENT;
 829                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
 830                         goto err_unlock_ovs;
 831
 832                 /* Allocate flow. */
 833                 flow = ovs_flow_alloc(!exact_5tuple);
 834                 if (IS_ERR(flow)) {
 835                         error = PTR_ERR(flow);
 836                         goto err_unlock_ovs;
 837                 }
 838
 839                 flow->key = masked_key;
 840                 flow->unmasked_key = key;
 841                 rcu_assign_pointer(flow->sf_acts, acts);
 842
 843                 /* Put flow in bucket. */
 844                 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
 845                 if (error) {
 846                         acts = NULL;
 847                         goto err_flow_free;
 848                 }
 849
 850                 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
 851         } else {
 852                 /* We found a matching flow. */
 853                 struct sw_flow_actions *old_acts;
 854
 855                 /* Bail out if we're not allowed to modify an existing flow.
 856                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 857                  * because Generic Netlink treats the latter as a dump
 858                  * request.  We also accept NLM_F_EXCL in case that bug ever
 859                  * gets fixed.
 860                  */
 861                 error = -EEXIST;
 862                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
 863                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
 864                         goto err_unlock_ovs;
 865
 866                 /* The unmasked key has to be the same for flow updates. */
 867                 if (!ovs_flow_cmp_unmasked_key(flow, &match))
 868                         goto err_unlock_ovs;
 869
 870                 /* Update actions. */
 871                 old_acts = ovsl_dereference(flow->sf_acts);
 872                 rcu_assign_pointer(flow->sf_acts, acts);
 873                 ovs_nla_free_flow_actions(old_acts);
 874
 875                 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
 876
 877                 /* Clear stats. */
 878                 if (a[OVS_FLOW_ATTR_CLEAR])
 879                         ovs_flow_stats_clear(flow);
 880         }
 881         ovs_unlock();
 882
 883         if (!IS_ERR(reply))
 884                 ovs_notify(&dp_flow_genl_family, reply, info);
 885         else
 886                 genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
 887                              0, PTR_ERR(reply));
 888         return 0;
 889
 890 err_flow_free:
 891         ovs_flow_free(flow, false);
 892 err_unlock_ovs:
 893         ovs_unlock();
 894 err_kfree:
 895         kfree(acts);
 896 error:
 897         return error;
 898 }
 899
 900 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
 901 {
 902         struct nlattr **a = info->attrs;
 903         struct ovs_header *ovs_header = info->userhdr;
 904         struct sw_flow_key key;
 905         struct sk_buff *reply;
 906         struct sw_flow *flow;
 907         struct datapath *dp;
 908         struct sw_flow_match match;
 909         int err;
 910
 911         if (!a[OVS_FLOW_ATTR_KEY]) {
 912                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
 913                 return -EINVAL;
 914         }
 915
 916         ovs_match_init(&match, &key, NULL);
 917         err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
 918         if (err)
 919                 return err;
 920
 921         ovs_lock();
 922         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 923         if (!dp) {
 924                 err = -ENODEV;
 925                 goto unlock;
 926         }
 927
 928         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 929         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
 930                 err = -ENOENT;
 931                 goto unlock;
 932         }
 933
 934         reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
 935         if (IS_ERR(reply)) {
 936                 err = PTR_ERR(reply);
 937                 goto unlock;
 938         }
 939
 940         ovs_unlock();
 941         return genlmsg_reply(reply, info);
 942 unlock:
 943         ovs_unlock();
 944         return err;
 945 }
 946
 947 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
 948 {
 949         struct nlattr **a = info->attrs;
 950         struct ovs_header *ovs_header = info->userhdr;
 951         struct sw_flow_key key;
 952         struct sk_buff *reply;
 953         struct sw_flow *flow;
 954         struct datapath *dp;
 955         struct sw_flow_match match;
 956         int err;
 957
 958         ovs_lock();
 959         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 960         if (!dp) {
 961                 err = -ENODEV;
 962                 goto unlock;
 963         }
 964
 965         if (!a[OVS_FLOW_ATTR_KEY]) {
 966                 err = ovs_flow_tbl_flush(&dp->table);
 967                 goto unlock;
 968         }
 969
 970         ovs_match_init(&match, &key, NULL);
 971         err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
 972         if (err)
 973                 goto unlock;
 974
 975         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 976         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
 977                 err = -ENOENT;
 978                 goto unlock;
 979         }
 980
 981         reply = ovs_flow_cmd_alloc_info(flow, info);
 982         if (!reply) {
 983                 err = -ENOMEM;
 984                 goto unlock;
 985         }
 986
 987         ovs_flow_tbl_remove(&dp->table, flow);
 988
 989         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
 990                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
 991         BUG_ON(err < 0);
 992
 993         ovs_flow_free(flow, true);
 994         ovs_unlock();
 995
 996         ovs_notify(&dp_flow_genl_family, reply, info);
 997         return 0;
 998 unlock:
 999         ovs_unlock();
1000         return err;
1001 }
1002
1003 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1004 {
1005         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1006         struct table_instance *ti;
1007         struct datapath *dp;
1008
1009         rcu_read_lock();
1010         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1011         if (!dp) {
1012                 rcu_read_unlock();
1013                 return -ENODEV;
1014         }
1015
1016         ti = rcu_dereference(dp->table.ti);
1017         for (;;) {
1018                 struct sw_flow *flow;
1019                 u32 bucket, obj;
1020
1021                 bucket = cb->args[0];
1022                 obj = cb->args[1];
1023                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1024                 if (!flow)
1025                         break;
1026
1027                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1028                                            NETLINK_CB(cb->skb).portid,
1029                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1030                                            OVS_FLOW_CMD_NEW) < 0)
1031                         break;
1032
1033                 cb->args[0] = bucket;
1034                 cb->args[1] = obj;
1035         }
1036         rcu_read_unlock();
1037         return skb->len;
1038 }
1039
1040 static const struct genl_ops dp_flow_genl_ops[] = {
1041         { .cmd = OVS_FLOW_CMD_NEW,
1042           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1043           .policy = flow_policy,
1044           .doit = ovs_flow_cmd_new_or_set
1045         },
1046         { .cmd = OVS_FLOW_CMD_DEL,
1047           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1048           .policy = flow_policy,
1049           .doit = ovs_flow_cmd_del
1050         },
1051         { .cmd = OVS_FLOW_CMD_GET,
1052           .flags = 0,               /* OK for unprivileged users. */
1053           .policy = flow_policy,
1054           .doit = ovs_flow_cmd_get,
1055           .dumpit = ovs_flow_cmd_dump
1056         },
1057         { .cmd = OVS_FLOW_CMD_SET,
1058           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1059           .policy = flow_policy,
1060           .doit = ovs_flow_cmd_new_or_set,
1061         },
1062 };
1063
1064 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1065         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1066         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1067         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1068 };
1069
1070 static struct genl_family dp_datapath_genl_family = {
1071         .id = GENL_ID_GENERATE,
1072         .hdrsize = sizeof(struct ovs_header),
1073         .name = OVS_DATAPATH_FAMILY,
1074         .version = OVS_DATAPATH_VERSION,
1075         .maxattr = OVS_DP_ATTR_MAX,
1076         .netnsok = true,
1077         .parallel_ops = true,
1078 };
1079
1080 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1081         .name = OVS_DATAPATH_MCGROUP
1082 };
1083
1084 static size_t ovs_dp_cmd_msg_size(void)
1085 {
1086         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1087
1088         msgsize += nla_total_size(IFNAMSIZ);
1089         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1090         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1091         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1092
1093         return msgsize;
1094 }
1095
1096 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1097                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1098 {
1099         struct ovs_header *ovs_header;
1100         struct ovs_dp_stats dp_stats;
1101         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1102         int err;
1103
1104         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1105                                    flags, cmd);
1106         if (!ovs_header)
1107                 goto error;
1108
1109         ovs_header->dp_ifindex = get_dpifindex(dp);
1110
1111         rcu_read_lock();
1112         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1113         rcu_read_unlock();
1114         if (err)
1115                 goto nla_put_failure;
1116
1117         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1118         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1119                         &dp_stats))
1120                 goto nla_put_failure;
1121
1122         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1123                         sizeof(struct ovs_dp_megaflow_stats),
1124                         &dp_megaflow_stats))
1125                 goto nla_put_failure;
1126
1127         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1128                 goto nla_put_failure;
1129
1130         return genlmsg_end(skb, ovs_header);
1131
1132 nla_put_failure:
1133         genlmsg_cancel(skb, ovs_header);
1134 error:
1135         return -EMSGSIZE;
1136 }
1137
1138 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp,
1139                                              struct genl_info *info, u8 cmd)
1140 {
1141         struct sk_buff *skb;
1142         int retval;
1143
1144         skb = genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1145         if (!skb)
1146                 return ERR_PTR(-ENOMEM);
1147
1148         retval = ovs_dp_cmd_fill_info(dp, skb, info->snd_portid, info->snd_seq, 0, cmd);
1149         if (retval < 0) {
1150                 kfree_skb(skb);
1151                 return ERR_PTR(retval);
1152         }
1153         return skb;
1154 }
1155
1156 /* Called with ovs_mutex. */
1157 static struct datapath *lookup_datapath(struct net *net,
1158                                         struct ovs_header *ovs_header,
1159                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1160 {
1161         struct datapath *dp;
1162
1163         if (!a[OVS_DP_ATTR_NAME])
1164                 dp = get_dp(net, ovs_header->dp_ifindex);
1165         else {
1166                 struct vport *vport;
1167
1168                 rcu_read_lock();
1169                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1170                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1171                 rcu_read_unlock();
1172         }
1173         return dp ? dp : ERR_PTR(-ENODEV);
1174 }
1175
1176 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1177 {
1178         struct datapath *dp;
1179
1180         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1181         if (IS_ERR(dp))
1182                 return;
1183
1184         WARN(dp->user_features, "Dropping previously announced user features\n");
1185         dp->user_features = 0;
1186 }
1187
1188 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1189 {
1190         if (a[OVS_DP_ATTR_USER_FEATURES])
1191                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1192 }
1193
1194 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1195 {
1196         struct nlattr **a = info->attrs;
1197         struct vport_parms parms;
1198         struct sk_buff *reply;
1199         struct datapath *dp;
1200         struct vport *vport;
1201         struct ovs_net *ovs_net;
1202         int err, i;
1203
1204         err = -EINVAL;
1205         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1206                 goto err;
1207
1208         ovs_lock();
1209
1210         err = -ENOMEM;
1211         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1212         if (dp == NULL)
1213                 goto err_unlock_ovs;
1214
1215         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1216
1217         /* Allocate table. */
1218         err = ovs_flow_tbl_init(&dp->table);
1219         if (err)
1220                 goto err_free_dp;
1221
1222         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1223         if (!dp->stats_percpu) {
1224                 err = -ENOMEM;
1225                 goto err_destroy_table;
1226         }
1227
1228         for_each_possible_cpu(i) {
1229                 struct dp_stats_percpu *dpath_stats;
1230                 dpath_stats = per_cpu_ptr(dp->stats_percpu, i);
1231                 u64_stats_init(&dpath_stats->sync);
1232         }
1233
1234         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1235                             GFP_KERNEL);
1236         if (!dp->ports) {
1237                 err = -ENOMEM;
1238                 goto err_destroy_percpu;
1239         }
1240
1241         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1242                 INIT_HLIST_HEAD(&dp->ports[i]);
1243
1244         /* Set up our datapath device. */
1245         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1246         parms.type = OVS_VPORT_TYPE_INTERNAL;
1247         parms.options = NULL;
1248         parms.dp = dp;
1249         parms.port_no = OVSP_LOCAL;
1250         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1251
1252         ovs_dp_change(dp, a);
1253
1254         vport = new_vport(&parms);
1255         if (IS_ERR(vport)) {
1256                 err = PTR_ERR(vport);
1257                 if (err == -EBUSY)
1258                         err = -EEXIST;
1259
1260                 if (err == -EEXIST) {
1261                         /* An outdated user space instance that does not understand
1262                          * the concept of user_features has attempted to create a new
1263                          * datapath and is likely to reuse it. Drop all user features.
1264                          */
1265                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1266                                 ovs_dp_reset_user_features(skb, info);
1267                 }
1268
1269                 goto err_destroy_ports_array;
1270         }
1271
1272         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1273         err = PTR_ERR(reply);
1274         if (IS_ERR(reply))
1275                 goto err_destroy_local_port;
1276
1277         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1278         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1279
1280         ovs_unlock();
1281
1282         ovs_notify(&dp_datapath_genl_family, reply, info);
1283         return 0;
1284
1285 err_destroy_local_port:
1286         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1287 err_destroy_ports_array:
1288         kfree(dp->ports);
1289 err_destroy_percpu:
1290         free_percpu(dp->stats_percpu);
1291 err_destroy_table:
1292         ovs_flow_tbl_destroy(&dp->table, false);
1293 err_free_dp:
1294         release_net(ovs_dp_get_net(dp));
1295         kfree(dp);
1296 err_unlock_ovs:
1297         ovs_unlock();
1298 err:
1299         return err;
1300 }
1301
1302 /* Called with ovs_mutex. */
1303 static void __dp_destroy(struct datapath *dp)
1304 {
1305         int i;
1306
1307         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1308                 struct vport *vport;
1309                 struct hlist_node *n;
1310
1311                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1312                         if (vport->port_no != OVSP_LOCAL)
1313                                 ovs_dp_detach_port(vport);
1314         }
1315
1316         list_del_rcu(&dp->list_node);
1317
1318         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1319          * all ports in datapath are destroyed first before freeing datapath.
1320          */
1321         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1322
1323         /* RCU destroy the flow table */
1324         ovs_flow_tbl_destroy(&dp->table, true);
1325
1326         call_rcu(&dp->rcu, destroy_dp_rcu);
1327 }
1328
1329 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1330 {
1331         struct sk_buff *reply;
1332         struct datapath *dp;
1333         int err;
1334
1335         ovs_lock();
1336         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1337         err = PTR_ERR(dp);
1338         if (IS_ERR(dp))
1339                 goto unlock;
1340
1341         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_DEL);
1342         err = PTR_ERR(reply);
1343         if (IS_ERR(reply))
1344                 goto unlock;
1345
1346         __dp_destroy(dp);
1347         ovs_unlock();
1348
1349         ovs_notify(&dp_datapath_genl_family, reply, info);
1350
1351         return 0;
1352 unlock:
1353         ovs_unlock();
1354         return err;
1355 }
1356
1357 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1358 {
1359         struct sk_buff *reply;
1360         struct datapath *dp;
1361         int err;
1362
1363         ovs_lock();
1364         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1365         err = PTR_ERR(dp);
1366         if (IS_ERR(dp))
1367                 goto unlock;
1368
1369         ovs_dp_change(dp, info->attrs);
1370
1371         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1372         if (IS_ERR(reply)) {
1373                 err = PTR_ERR(reply);
1374                 genl_set_err(&dp_datapath_genl_family, sock_net(skb->sk), 0,
1375                              0, err);
1376                 err = 0;
1377                 goto unlock;
1378         }
1379
1380         ovs_unlock();
1381         ovs_notify(&dp_datapath_genl_family, reply, info);
1382
1383         return 0;
1384 unlock:
1385         ovs_unlock();
1386         return err;
1387 }
1388
1389 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1390 {
1391         struct sk_buff *reply;
1392         struct datapath *dp;
1393         int err;
1394
1395         ovs_lock();
1396         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1397         if (IS_ERR(dp)) {
1398                 err = PTR_ERR(dp);
1399                 goto unlock;
1400         }
1401
1402         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1403         if (IS_ERR(reply)) {
1404                 err = PTR_ERR(reply);
1405                 goto unlock;
1406         }
1407
1408         ovs_unlock();
1409         return genlmsg_reply(reply, info);
1410
1411 unlock:
1412         ovs_unlock();
1413         return err;
1414 }
1415
1416 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1417 {
1418         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1419         struct datapath *dp;
1420         int skip = cb->args[0];
1421         int i = 0;
1422
1423         rcu_read_lock();
1424         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1425                 if (i >= skip &&
1426                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1427                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1428                                          OVS_DP_CMD_NEW) < 0)
1429                         break;
1430                 i++;
1431         }
1432         rcu_read_unlock();
1433
1434         cb->args[0] = i;
1435
1436         return skb->len;
1437 }
1438
1439 static const struct genl_ops dp_datapath_genl_ops[] = {
1440         { .cmd = OVS_DP_CMD_NEW,
1441           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1442           .policy = datapath_policy,
1443           .doit = ovs_dp_cmd_new
1444         },
1445         { .cmd = OVS_DP_CMD_DEL,
1446           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1447           .policy = datapath_policy,
1448           .doit = ovs_dp_cmd_del
1449         },
1450         { .cmd = OVS_DP_CMD_GET,
1451           .flags = 0,               /* OK for unprivileged users. */
1452           .policy = datapath_policy,
1453           .doit = ovs_dp_cmd_get,
1454           .dumpit = ovs_dp_cmd_dump
1455         },
1456         { .cmd = OVS_DP_CMD_SET,
1457           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1458           .policy = datapath_policy,
1459           .doit = ovs_dp_cmd_set,
1460         },
1461 };
1462
1463 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1464         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1465         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1466         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1467         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1468         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1469         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1470 };
1471
1472 struct genl_family dp_vport_genl_family = {
1473         .id = GENL_ID_GENERATE,
1474         .hdrsize = sizeof(struct ovs_header),
1475         .name = OVS_VPORT_FAMILY,
1476         .version = OVS_VPORT_VERSION,
1477         .maxattr = OVS_VPORT_ATTR_MAX,
1478         .netnsok = true,
1479         .parallel_ops = true,
1480 };
1481
1482 static struct genl_multicast_group ovs_dp_vport_multicast_group = {
1483         .name = OVS_VPORT_MCGROUP
1484 };
1485
1486 /* Called with ovs_mutex or RCU read lock. */
1487 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1488                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1489 {
1490         struct ovs_header *ovs_header;
1491         struct ovs_vport_stats vport_stats;
1492         int err;
1493
1494         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1495                                  flags, cmd);
1496         if (!ovs_header)
1497                 return -EMSGSIZE;
1498
1499         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1500
1501         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1502             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1503             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1504             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1505                 goto nla_put_failure;
1506
1507         ovs_vport_get_stats(vport, &vport_stats);
1508         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1509                     &vport_stats))
1510                 goto nla_put_failure;
1511
1512         err = ovs_vport_get_options(vport, skb);
1513         if (err == -EMSGSIZE)
1514                 goto error;
1515
1516         return genlmsg_end(skb, ovs_header);
1517
1518 nla_put_failure:
1519         err = -EMSGSIZE;
1520 error:
1521         genlmsg_cancel(skb, ovs_header);
1522         return err;
1523 }
1524
1525 /* Called with ovs_mutex or RCU read lock. */
1526 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1527                                          u32 seq, u8 cmd)
1528 {
1529         struct sk_buff *skb;
1530         int retval;
1531
1532         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1533         if (!skb)
1534                 return ERR_PTR(-ENOMEM);
1535
1536         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1537         BUG_ON(retval < 0);
1538
1539         return skb;
1540 }
1541
1542 /* Called with ovs_mutex or RCU read lock. */
1543 static struct vport *lookup_vport(struct net *net,
1544                                   struct ovs_header *ovs_header,
1545                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1546 {
1547         struct datapath *dp;
1548         struct vport *vport;
1549
1550         if (a[OVS_VPORT_ATTR_NAME]) {
1551                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1552                 if (!vport)
1553                         return ERR_PTR(-ENODEV);
1554                 if (ovs_header->dp_ifindex &&
1555                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1556                         return ERR_PTR(-ENODEV);
1557                 return vport;
1558         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1559                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1560
1561                 if (port_no >= DP_MAX_PORTS)
1562                         return ERR_PTR(-EFBIG);
1563
1564                 dp = get_dp(net, ovs_header->dp_ifindex);
1565                 if (!dp)
1566                         return ERR_PTR(-ENODEV);
1567
1568                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1569                 if (!vport)
1570                         return ERR_PTR(-ENODEV);
1571                 return vport;
1572         } else
1573                 return ERR_PTR(-EINVAL);
1574 }
1575
1576 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1577 {
1578         struct nlattr **a = info->attrs;
1579         struct ovs_header *ovs_header = info->userhdr;
1580         struct vport_parms parms;
1581         struct sk_buff *reply;
1582         struct vport *vport;
1583         struct datapath *dp;
1584         u32 port_no;
1585         int err;
1586
1587         err = -EINVAL;
1588         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1589             !a[OVS_VPORT_ATTR_UPCALL_PID])
1590                 goto exit;
1591
1592         ovs_lock();
1593         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1594         err = -ENODEV;
1595         if (!dp)
1596                 goto exit_unlock;
1597
1598         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1599                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1600
1601                 err = -EFBIG;
1602                 if (port_no >= DP_MAX_PORTS)
1603                         goto exit_unlock;
1604
1605                 vport = ovs_vport_ovsl(dp, port_no);
1606                 err = -EBUSY;
1607                 if (vport)
1608                         goto exit_unlock;
1609         } else {
1610                 for (port_no = 1; ; port_no++) {
1611                         if (port_no >= DP_MAX_PORTS) {
1612                                 err = -EFBIG;
1613                                 goto exit_unlock;
1614                         }
1615                         vport = ovs_vport_ovsl(dp, port_no);
1616                         if (!vport)
1617                                 break;
1618                 }
1619         }
1620
1621         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1622         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1623         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1624         parms.dp = dp;
1625         parms.port_no = port_no;
1626         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1627
1628         vport = new_vport(&parms);
1629         err = PTR_ERR(vport);
1630         if (IS_ERR(vport))
1631                 goto exit_unlock;
1632
1633         err = 0;
1634         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1635                                          OVS_VPORT_CMD_NEW);
1636         if (IS_ERR(reply)) {
1637                 err = PTR_ERR(reply);
1638                 ovs_dp_detach_port(vport);
1639                 goto exit_unlock;
1640         }
1641
1642         ovs_notify(&dp_vport_genl_family, reply, info);
1643
1644 exit_unlock:
1645         ovs_unlock();
1646 exit:
1647         return err;
1648 }
1649
1650 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1651 {
1652         struct nlattr **a = info->attrs;
1653         struct sk_buff *reply;
1654         struct vport *vport;
1655         int err;
1656
1657         ovs_lock();
1658         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1659         err = PTR_ERR(vport);
1660         if (IS_ERR(vport))
1661                 goto exit_unlock;
1662
1663         if (a[OVS_VPORT_ATTR_TYPE] &&
1664             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1665                 err = -EINVAL;
1666                 goto exit_unlock;
1667         }
1668
1669         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1670         if (!reply) {
1671                 err = -ENOMEM;
1672                 goto exit_unlock;
1673         }
1674
1675         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1676                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1677                 if (err)
1678                         goto exit_free;
1679         }
1680
1681         if (a[OVS_VPORT_ATTR_UPCALL_PID])
1682                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1683
1684         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1685                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1686         BUG_ON(err < 0);
1687
1688         ovs_unlock();
1689         ovs_notify(&dp_vport_genl_family, reply, info);
1690         return 0;
1691
1692 exit_free:
1693         kfree_skb(reply);
1694 exit_unlock:
1695         ovs_unlock();
1696         return err;
1697 }
1698
1699 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1700 {
1701         struct nlattr **a = info->attrs;
1702         struct sk_buff *reply;
1703         struct vport *vport;
1704         int err;
1705
1706         ovs_lock();
1707         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1708         err = PTR_ERR(vport);
1709         if (IS_ERR(vport))
1710                 goto exit_unlock;
1711
1712         if (vport->port_no == OVSP_LOCAL) {
1713                 err = -EINVAL;
1714                 goto exit_unlock;
1715         }
1716
1717         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1718                                          info->snd_seq, OVS_VPORT_CMD_DEL);
1719         err = PTR_ERR(reply);
1720         if (IS_ERR(reply))
1721                 goto exit_unlock;
1722
1723         err = 0;
1724         ovs_dp_detach_port(vport);
1725
1726         ovs_notify(&dp_vport_genl_family, reply, info);
1727
1728 exit_unlock:
1729         ovs_unlock();
1730         return err;
1731 }
1732
1733 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1734 {
1735         struct nlattr **a = info->attrs;
1736         struct ovs_header *ovs_header = info->userhdr;
1737         struct sk_buff *reply;
1738         struct vport *vport;
1739         int err;
1740
1741         rcu_read_lock();
1742         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1743         err = PTR_ERR(vport);
1744         if (IS_ERR(vport))
1745                 goto exit_unlock;
1746
1747         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1748                                          info->snd_seq, OVS_VPORT_CMD_NEW);
1749         err = PTR_ERR(reply);
1750         if (IS_ERR(reply))
1751                 goto exit_unlock;
1752
1753         rcu_read_unlock();
1754
1755         return genlmsg_reply(reply, info);
1756
1757 exit_unlock:
1758         rcu_read_unlock();
1759         return err;
1760 }
1761
1762 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1763 {
1764         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1765         struct datapath *dp;
1766         int bucket = cb->args[0], skip = cb->args[1];
1767         int i, j = 0;
1768
1769         rcu_read_lock();
1770         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1771         if (!dp) {
1772                 rcu_read_unlock();
1773                 return -ENODEV;
1774         }
1775         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1776                 struct vport *vport;
1777
1778                 j = 0;
1779                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1780                         if (j >= skip &&
1781                             ovs_vport_cmd_fill_info(vport, skb,
1782                                                     NETLINK_CB(cb->skb).portid,
1783                                                     cb->nlh->nlmsg_seq,
1784                                                     NLM_F_MULTI,
1785                                                     OVS_VPORT_CMD_NEW) < 0)
1786                                 goto out;
1787
1788                         j++;
1789                 }
1790                 skip = 0;
1791         }
1792 out:
1793         rcu_read_unlock();
1794
1795         cb->args[0] = i;
1796         cb->args[1] = j;
1797
1798         return skb->len;
1799 }
1800
1801 static const struct genl_ops dp_vport_genl_ops[] = {
1802         { .cmd = OVS_VPORT_CMD_NEW,
1803           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1804           .policy = vport_policy,
1805           .doit = ovs_vport_cmd_new
1806         },
1807         { .cmd = OVS_VPORT_CMD_DEL,
1808           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1809           .policy = vport_policy,
1810           .doit = ovs_vport_cmd_del
1811         },
1812         { .cmd = OVS_VPORT_CMD_GET,
1813           .flags = 0,               /* OK for unprivileged users. */
1814           .policy = vport_policy,
1815           .doit = ovs_vport_cmd_get,
1816           .dumpit = ovs_vport_cmd_dump
1817         },
1818         { .cmd = OVS_VPORT_CMD_SET,
1819           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1820           .policy = vport_policy,
1821           .doit = ovs_vport_cmd_set,
1822         },
1823 };
1824
1825 struct genl_family_and_ops {
1826         struct genl_family *family;
1827         const struct genl_ops *ops;
1828         int n_ops;
1829         const struct genl_multicast_group *group;
1830 };
1831
1832 static const struct genl_family_and_ops dp_genl_families[] = {
1833         { &dp_datapath_genl_family,
1834           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1835           &ovs_dp_datapath_multicast_group },
1836         { &dp_vport_genl_family,
1837           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1838           &ovs_dp_vport_multicast_group },
1839         { &dp_flow_genl_family,
1840           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1841           &ovs_dp_flow_multicast_group },
1842         { &dp_packet_genl_family,
1843           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1844           NULL },
1845 };
1846
1847 static void dp_unregister_genl(int n_families)
1848 {
1849         int i;
1850
1851         for (i = 0; i < n_families; i++)
1852                 genl_unregister_family(dp_genl_families[i].family);
1853 }
1854
1855 static int dp_register_genl(void)
1856 {
1857         int n_registered;
1858         int err;
1859         int i;
1860
1861         n_registered = 0;
1862         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1863                 const struct genl_family_and_ops *f = &dp_genl_families[i];
1864
1865                 f->family->ops = f->ops;
1866                 f->family->n_ops = f->n_ops;
1867                 f->family->mcgrps = f->group;
1868                 f->family->n_mcgrps = f->group ? 1 : 0;
1869                 err = genl_register_family(f->family);
1870                 if (err)
1871                         goto error;
1872                 n_registered++;
1873         }
1874
1875         return 0;
1876
1877 error:
1878         dp_unregister_genl(n_registered);
1879         return err;
1880 }
1881
1882 static int __net_init ovs_init_net(struct net *net)
1883 {
1884         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1885
1886         INIT_LIST_HEAD(&ovs_net->dps);
1887         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1888         return 0;
1889 }
1890
1891 static void __net_exit ovs_exit_net(struct net *net)
1892 {
1893         struct datapath *dp, *dp_next;
1894         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1895
1896         ovs_lock();
1897         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1898                 __dp_destroy(dp);
1899         ovs_unlock();
1900
1901         cancel_work_sync(&ovs_net->dp_notify_work);
1902 }
1903
1904 static struct pernet_operations ovs_net_ops = {
1905         .init = ovs_init_net,
1906         .exit = ovs_exit_net,
1907         .id   = &ovs_net_id,
1908         .size = sizeof(struct ovs_net),
1909 };
1910
1911 static int __init dp_init(void)
1912 {
1913         int err;
1914
1915         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1916
1917         pr_info("Open vSwitch switching datapath\n");
1918
1919         err = ovs_flow_init();
1920         if (err)
1921                 goto error;
1922
1923         err = ovs_vport_init();
1924         if (err)
1925                 goto error_flow_exit;
1926
1927         err = register_pernet_device(&ovs_net_ops);
1928         if (err)
1929                 goto error_vport_exit;
1930
1931         err = register_netdevice_notifier(&ovs_dp_device_notifier);
1932         if (err)
1933                 goto error_netns_exit;
1934
1935         err = dp_register_genl();
1936         if (err < 0)
1937                 goto error_unreg_notifier;
1938
1939         return 0;
1940
1941 error_unreg_notifier:
1942         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1943 error_netns_exit:
1944         unregister_pernet_device(&ovs_net_ops);
1945 error_vport_exit:
1946         ovs_vport_exit();
1947 error_flow_exit:
1948         ovs_flow_exit();
1949 error:
1950         return err;
1951 }
1952
1953 static void dp_cleanup(void)
1954 {
1955         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1956         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1957         unregister_pernet_device(&ovs_net_ops);
1958         rcu_barrier();
1959         ovs_vport_exit();
1960         ovs_flow_exit();
1961 }
1962
1963 module_init(dp_init);
1964 module_exit(dp_cleanup);
1965
1966 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1967 MODULE_LICENSE("GPL");