net/openvswitch/datapath.c

   1 /*
   2  * Copyright (c) 2007-2014 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/ethtool.h>
  40 #include <linux/wait.h>
  41 #include <asm/div64.h>
  42 #include <linux/highmem.h>
  43 #include <linux/netfilter_bridge.h>
  44 #include <linux/netfilter_ipv4.h>
  45 #include <linux/inetdevice.h>
  46 #include <linux/list.h>
  47 #include <linux/openvswitch.h>
  48 #include <linux/rculist.h>
  49 #include <linux/dmi.h>
  50 #include <net/genetlink.h>
  51 #include <net/net_namespace.h>
  52 #include <net/netns/generic.h>
  53
  54 #include "datapath.h"
  55 #include "flow.h"
  56 #include "flow_table.h"
  57 #include "flow_netlink.h"
  58 #include "vport-internal_dev.h"
  59 #include "vport-netdev.h"
  60
  61 int ovs_net_id __read_mostly;
  62
  63 static struct genl_family dp_packet_genl_family;
  64 static struct genl_family dp_flow_genl_family;
  65 static struct genl_family dp_datapath_genl_family;
  66
  67 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
  68         .name = OVS_FLOW_MCGROUP,
  69 };
  70
  71 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  72         .name = OVS_DATAPATH_MCGROUP,
  73 };
  74
  75 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
  76         .name = OVS_VPORT_MCGROUP,
  77 };
  78
  79 /* Check if need to build a reply message.
  80  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  81 static bool ovs_must_notify(struct genl_info *info,
  82                             const struct genl_multicast_group *grp)
  83 {
  84         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  85                 netlink_has_listeners(genl_info_net(info)->genl_sock, 0);
  86 }
  87
  88 static void ovs_notify(struct genl_family *family,
  89                        struct sk_buff *skb, struct genl_info *info)
  90 {
  91         genl_notify(family, skb, genl_info_net(info), info->snd_portid,
  92                     0, info->nlhdr, GFP_KERNEL);
  93 }
  94
  95 /**
  96  * DOC: Locking:
  97  *
  98  * All writes e.g. Writes to device state (add/remove datapath, port, set
  99  * operations on vports, etc.), Writes to other state (flow table
 100  * modifications, set miscellaneous datapath parameters, etc.) are protected
 101  * by ovs_lock.
 102  *
 103  * Reads are protected by RCU.
 104  *
 105  * There are a few special cases (mostly stats) that have their own
 106  * synchronization but they nest under all of above and don't interact with
 107  * each other.
 108  *
 109  * The RTNL lock nests inside ovs_mutex.
 110  */
 111
 112 static DEFINE_MUTEX(ovs_mutex);
 113
 114 void ovs_lock(void)
 115 {
 116         mutex_lock(&ovs_mutex);
 117 }
 118
 119 void ovs_unlock(void)
 120 {
 121         mutex_unlock(&ovs_mutex);
 122 }
 123
 124 #ifdef CONFIG_LOCKDEP
 125 int lockdep_ovsl_is_held(void)
 126 {
 127         if (debug_locks)
 128                 return lockdep_is_held(&ovs_mutex);
 129         else
 130                 return 1;
 131 }
 132 #endif
 133
 134 static struct vport *new_vport(const struct vport_parms *);
 135 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 136                              const struct dp_upcall_info *);
 137 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 138                                   const struct dp_upcall_info *);
 139
 140 /* Must be called with rcu_read_lock or ovs_mutex. */
 141 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 142 {
 143         struct datapath *dp = NULL;
 144         struct net_device *dev;
 145
 146         rcu_read_lock();
 147         dev = dev_get_by_index_rcu(net, dp_ifindex);
 148         if (dev) {
 149                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 150                 if (vport)
 151                         dp = vport->dp;
 152         }
 153         rcu_read_unlock();
 154
 155         return dp;
 156 }
 157
 158 /* Must be called with rcu_read_lock or ovs_mutex. */
 159 static const char *ovs_dp_name(const struct datapath *dp)
 160 {
 161         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 162         return vport->ops->get_name(vport);
 163 }
 164
 165 static int get_dpifindex(struct datapath *dp)
 166 {
 167         struct vport *local;
 168         int ifindex;
 169
 170         rcu_read_lock();
 171
 172         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 173         if (local)
 174                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 175         else
 176                 ifindex = 0;
 177
 178         rcu_read_unlock();
 179
 180         return ifindex;
 181 }
 182
 183 static void destroy_dp_rcu(struct rcu_head *rcu)
 184 {
 185         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 186
 187         free_percpu(dp->stats_percpu);
 188         release_net(ovs_dp_get_net(dp));
 189         kfree(dp->ports);
 190         kfree(dp);
 191 }
 192
 193 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 194                                             u16 port_no)
 195 {
 196         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 197 }
 198
 199 /* Called with ovs_mutex or RCU read lock. */
 200 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 201 {
 202         struct vport *vport;
 203         struct hlist_head *head;
 204
 205         head = vport_hash_bucket(dp, port_no);
 206         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 207                 if (vport->port_no == port_no)
 208                         return vport;
 209         }
 210         return NULL;
 211 }
 212
 213 /* Called with ovs_mutex. */
 214 static struct vport *new_vport(const struct vport_parms *parms)
 215 {
 216         struct vport *vport;
 217
 218         vport = ovs_vport_add(parms);
 219         if (!IS_ERR(vport)) {
 220                 struct datapath *dp = parms->dp;
 221                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 222
 223                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 224         }
 225         return vport;
 226 }
 227
 228 void ovs_dp_detach_port(struct vport *p)
 229 {
 230         ASSERT_OVSL();
 231
 232         /* First drop references to device. */
 233         hlist_del_rcu(&p->dp_hash_node);
 234
 235         /* Then destroy it. */
 236         ovs_vport_del(p);
 237 }
 238
 239 /* Must be called with rcu_read_lock. */
 240 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 241 {
 242         const struct vport *p = OVS_CB(skb)->input_vport;
 243         struct datapath *dp = p->dp;
 244         struct sw_flow *flow;
 245         struct dp_stats_percpu *stats;
 246         u64 *stats_counter;
 247         u32 n_mask_hit;
 248
 249         stats = this_cpu_ptr(dp->stats_percpu);
 250
 251         /* Look up flow. */
 252         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
 253         if (unlikely(!flow)) {
 254                 struct dp_upcall_info upcall;
 255                 int error;
 256
 257                 upcall.cmd = OVS_PACKET_CMD_MISS;
 258                 upcall.key = key;
 259                 upcall.userdata = NULL;
 260                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 261                 error = ovs_dp_upcall(dp, skb, &upcall);
 262                 if (unlikely(error))
 263                         kfree_skb(skb);
 264                 else
 265                         consume_skb(skb);
 266                 stats_counter = &stats->n_missed;
 267                 goto out;
 268         }
 269
 270         OVS_CB(skb)->flow = flow;
 271
 272         ovs_flow_stats_update(OVS_CB(skb)->flow, key->tp.flags, skb);
 273         ovs_execute_actions(dp, skb, key);
 274         stats_counter = &stats->n_hit;
 275
 276 out:
 277         /* Update datapath statistics. */
 278         u64_stats_update_begin(&stats->syncp);
 279         (*stats_counter)++;
 280         stats->n_mask_hit += n_mask_hit;
 281         u64_stats_update_end(&stats->syncp);
 282 }
 283
 284 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 285                   const struct dp_upcall_info *upcall_info)
 286 {
 287         struct dp_stats_percpu *stats;
 288         int err;
 289
 290         if (upcall_info->portid == 0) {
 291                 err = -ENOTCONN;
 292                 goto err;
 293         }
 294
 295         if (!skb_is_gso(skb))
 296                 err = queue_userspace_packet(dp, skb, upcall_info);
 297         else
 298                 err = queue_gso_packets(dp, skb, upcall_info);
 299         if (err)
 300                 goto err;
 301
 302         return 0;
 303
 304 err:
 305         stats = this_cpu_ptr(dp->stats_percpu);
 306
 307         u64_stats_update_begin(&stats->syncp);
 308         stats->n_lost++;
 309         u64_stats_update_end(&stats->syncp);
 310
 311         return err;
 312 }
 313
 314 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 315                              const struct dp_upcall_info *upcall_info)
 316 {
 317         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 318         struct dp_upcall_info later_info;
 319         struct sw_flow_key later_key;
 320         struct sk_buff *segs, *nskb;
 321         int err;
 322
 323         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 324         if (IS_ERR(segs))
 325                 return PTR_ERR(segs);
 326
 327         /* Queue all of the segments. */
 328         skb = segs;
 329         do {
 330                 err = queue_userspace_packet(dp, skb, upcall_info);
 331                 if (err)
 332                         break;
 333
 334                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 335                         /* The initial flow key extracted by ovs_flow_extract()
 336                          * in this case is for a first fragment, so we need to
 337                          * properly mark later fragments.
 338                          */
 339                         later_key = *upcall_info->key;
 340                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 341
 342                         later_info = *upcall_info;
 343                         later_info.key = &later_key;
 344                         upcall_info = &later_info;
 345                 }
 346         } while ((skb = skb->next));
 347
 348         /* Free all of the segments. */
 349         skb = segs;
 350         do {
 351                 nskb = skb->next;
 352                 if (err)
 353                         kfree_skb(skb);
 354                 else
 355                         consume_skb(skb);
 356         } while ((skb = nskb));
 357         return err;
 358 }
 359
 360 static size_t key_attr_size(void)
 361 {
 362         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 363                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 364                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
 365                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 366                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 367                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
 368                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
 369                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 370                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
 371                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 372                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 373                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 374                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 375                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 376                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 377                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 378                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 379                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 380                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 381 }
 382
 383 static size_t upcall_msg_size(const struct nlattr *userdata,
 384                               unsigned int hdrlen)
 385 {
 386         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 387                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 388                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 389
 390         /* OVS_PACKET_ATTR_USERDATA */
 391         if (userdata)
 392                 size += NLA_ALIGN(userdata->nla_len);
 393
 394         return size;
 395 }
 396
 397 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 398                                   const struct dp_upcall_info *upcall_info)
 399 {
 400         struct ovs_header *upcall;
 401         struct sk_buff *nskb = NULL;
 402         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 403         struct nlattr *nla;
 404         struct genl_info info = {
 405                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 406                 .snd_portid = upcall_info->portid,
 407         };
 408         size_t len;
 409         unsigned int hlen;
 410         int err, dp_ifindex;
 411
 412         dp_ifindex = get_dpifindex(dp);
 413         if (!dp_ifindex)
 414                 return -ENODEV;
 415
 416         if (vlan_tx_tag_present(skb)) {
 417                 nskb = skb_clone(skb, GFP_ATOMIC);
 418                 if (!nskb)
 419                         return -ENOMEM;
 420
 421                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 422                 if (!nskb)
 423                         return -ENOMEM;
 424
 425                 nskb->vlan_tci = 0;
 426                 skb = nskb;
 427         }
 428
 429         if (nla_attr_size(skb->len) > USHRT_MAX) {
 430                 err = -EFBIG;
 431                 goto out;
 432         }
 433
 434         /* Complete checksum if needed */
 435         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 436             (err = skb_checksum_help(skb)))
 437                 goto out;
 438
 439         /* Older versions of OVS user space enforce alignment of the last
 440          * Netlink attribute to NLA_ALIGNTO which would require extensive
 441          * padding logic. Only perform zerocopy if padding is not required.
 442          */
 443         if (dp->user_features & OVS_DP_F_UNALIGNED)
 444                 hlen = skb_zerocopy_headlen(skb);
 445         else
 446                 hlen = skb->len;
 447
 448         len = upcall_msg_size(upcall_info->userdata, hlen);
 449         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 450         if (!user_skb) {
 451                 err = -ENOMEM;
 452                 goto out;
 453         }
 454
 455         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 456                              0, upcall_info->cmd);
 457         upcall->dp_ifindex = dp_ifindex;
 458
 459         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 460         err = ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
 461         BUG_ON(err);
 462         nla_nest_end(user_skb, nla);
 463
 464         if (upcall_info->userdata)
 465                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 466                           nla_len(upcall_info->userdata),
 467                           nla_data(upcall_info->userdata));
 468
 469         /* Only reserve room for attribute header, packet data is added
 470          * in skb_zerocopy() */
 471         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 472                 err = -ENOBUFS;
 473                 goto out;
 474         }
 475         nla->nla_len = nla_attr_size(skb->len);
 476
 477         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 478         if (err)
 479                 goto out;
 480
 481         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 482         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 483                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 484
 485                 if (plen > 0)
 486                         memset(skb_put(user_skb, plen), 0, plen);
 487         }
 488
 489         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 490
 491         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 492         user_skb = NULL;
 493 out:
 494         if (err)
 495                 skb_tx_error(skb);
 496         kfree_skb(user_skb);
 497         kfree_skb(nskb);
 498         return err;
 499 }
 500
 501 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 502 {
 503         struct ovs_header *ovs_header = info->userhdr;
 504         struct nlattr **a = info->attrs;
 505         struct sw_flow_actions *acts;
 506         struct sk_buff *packet;
 507         struct sw_flow *flow;
 508         struct datapath *dp;
 509         struct ethhdr *eth;
 510         struct vport *input_vport;
 511         int len;
 512         int err;
 513
 514         err = -EINVAL;
 515         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 516             !a[OVS_PACKET_ATTR_ACTIONS])
 517                 goto err;
 518
 519         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 520         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 521         err = -ENOMEM;
 522         if (!packet)
 523                 goto err;
 524         skb_reserve(packet, NET_IP_ALIGN);
 525
 526         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 527
 528         skb_reset_mac_header(packet);
 529         eth = eth_hdr(packet);
 530
 531         /* Normally, setting the skb 'protocol' field would be handled by a
 532          * call to eth_type_trans(), but it assumes there's a sending
 533          * device, which we may not have. */
 534         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 535                 packet->protocol = eth->h_proto;
 536         else
 537                 packet->protocol = htons(ETH_P_802_2);
 538
 539         /* Build an sw_flow for sending this packet. */
 540         flow = ovs_flow_alloc();
 541         err = PTR_ERR(flow);
 542         if (IS_ERR(flow))
 543                 goto err_kfree_skb;
 544
 545         err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
 546                                              &flow->key);
 547         if (err)
 548                 goto err_flow_free;
 549
 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         packet->priority = flow->key.phy.priority;
 563         packet->mark = flow->key.phy.skb_mark;
 564
 565         rcu_read_lock();
 566         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 567         err = -ENODEV;
 568         if (!dp)
 569                 goto err_unlock;
 570
 571         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 572         if (!input_vport)
 573                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 574
 575         if (!input_vport)
 576                 goto err_unlock;
 577
 578         OVS_CB(packet)->input_vport = input_vport;
 579
 580         local_bh_disable();
 581         err = ovs_execute_actions(dp, packet, &flow->key);
 582         local_bh_enable();
 583         rcu_read_unlock();
 584
 585         ovs_flow_free(flow, false);
 586         return err;
 587
 588 err_unlock:
 589         rcu_read_unlock();
 590 err_flow_free:
 591         ovs_flow_free(flow, false);
 592 err_kfree_skb:
 593         kfree_skb(packet);
 594 err:
 595         return err;
 596 }
 597
 598 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 599         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 600         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 601         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 602 };
 603
 604 static const struct genl_ops dp_packet_genl_ops[] = {
 605         { .cmd = OVS_PACKET_CMD_EXECUTE,
 606           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 607           .policy = packet_policy,
 608           .doit = ovs_packet_cmd_execute
 609         }
 610 };
 611
 612 static struct genl_family dp_packet_genl_family = {
 613         .id = GENL_ID_GENERATE,
 614         .hdrsize = sizeof(struct ovs_header),
 615         .name = OVS_PACKET_FAMILY,
 616         .version = OVS_PACKET_VERSION,
 617         .maxattr = OVS_PACKET_ATTR_MAX,
 618         .netnsok = true,
 619         .parallel_ops = true,
 620         .ops = dp_packet_genl_ops,
 621         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 622 };
 623
 624 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
 625                          struct ovs_dp_megaflow_stats *mega_stats)
 626 {
 627         int i;
 628
 629         memset(mega_stats, 0, sizeof(*mega_stats));
 630
 631         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 632         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 633
 634         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 635
 636         for_each_possible_cpu(i) {
 637                 const struct dp_stats_percpu *percpu_stats;
 638                 struct dp_stats_percpu local_stats;
 639                 unsigned int start;
 640
 641                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 642
 643                 do {
 644                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 645                         local_stats = *percpu_stats;
 646                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 647
 648                 stats->n_hit += local_stats.n_hit;
 649                 stats->n_missed += local_stats.n_missed;
 650                 stats->n_lost += local_stats.n_lost;
 651                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 652         }
 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         if (!skb)
 770                 return ERR_PTR(-ENOMEM);
 771
 772         return skb;
 773 }
 774
 775 /* Called with ovs_mutex. */
 776 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 777                                                int dp_ifindex,
 778                                                struct genl_info *info, u8 cmd,
 779                                                bool always)
 780 {
 781         struct sk_buff *skb;
 782         int retval;
 783
 784         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
 785                                       always);
 786         if (IS_ERR_OR_NULL(skb))
 787                 return skb;
 788
 789         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 790                                         info->snd_portid, info->snd_seq, 0,
 791                                         cmd);
 792         BUG_ON(retval < 0);
 793         return skb;
 794 }
 795
 796 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 797 {
 798         struct nlattr **a = info->attrs;
 799         struct ovs_header *ovs_header = info->userhdr;
 800         struct sw_flow *flow, *new_flow;
 801         struct sw_flow_mask mask;
 802         struct sk_buff *reply;
 803         struct datapath *dp;
 804         struct sw_flow_actions *acts;
 805         struct sw_flow_match match;
 806         int error;
 807
 808         /* Must have key and actions. */
 809         error = -EINVAL;
 810         if (!a[OVS_FLOW_ATTR_KEY])
 811                 goto error;
 812         if (!a[OVS_FLOW_ATTR_ACTIONS])
 813                 goto error;
 814
 815         /* Most of the time we need to allocate a new flow, do it before
 816          * locking.
 817          */
 818         new_flow = ovs_flow_alloc();
 819         if (IS_ERR(new_flow)) {
 820                 error = PTR_ERR(new_flow);
 821                 goto error;
 822         }
 823
 824         /* Extract key. */
 825         ovs_match_init(&match, &new_flow->unmasked_key, &mask);
 826         error = ovs_nla_get_match(&match,
 827                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 828         if (error)
 829                 goto err_kfree_flow;
 830
 831         ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
 832
 833         /* Validate actions. */
 834         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 835         error = PTR_ERR(acts);
 836         if (IS_ERR(acts))
 837                 goto err_kfree_flow;
 838
 839         error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
 840                                      0, &acts);
 841         if (error) {
 842                 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 843                 goto err_kfree_acts;
 844         }
 845
 846         reply = ovs_flow_cmd_alloc_info(acts, info, false);
 847         if (IS_ERR(reply)) {
 848                 error = PTR_ERR(reply);
 849                 goto err_kfree_acts;
 850         }
 851
 852         ovs_lock();
 853         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 854         if (unlikely(!dp)) {
 855                 error = -ENODEV;
 856                 goto err_unlock_ovs;
 857         }
 858         /* Check if this is a duplicate flow */
 859         flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
 860         if (likely(!flow)) {
 861                 rcu_assign_pointer(new_flow->sf_acts, acts);
 862
 863                 /* Put flow in bucket. */
 864                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 865                 if (unlikely(error)) {
 866                         acts = NULL;
 867                         goto err_unlock_ovs;
 868                 }
 869
 870                 if (unlikely(reply)) {
 871                         error = ovs_flow_cmd_fill_info(new_flow,
 872                                                        ovs_header->dp_ifindex,
 873                                                        reply, info->snd_portid,
 874                                                        info->snd_seq, 0,
 875                                                        OVS_FLOW_CMD_NEW);
 876                         BUG_ON(error < 0);
 877                 }
 878                 ovs_unlock();
 879         } else {
 880                 struct sw_flow_actions *old_acts;
 881
 882                 /* Bail out if we're not allowed to modify an existing flow.
 883                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 884                  * because Generic Netlink treats the latter as a dump
 885                  * request.  We also accept NLM_F_EXCL in case that bug ever
 886                  * gets fixed.
 887                  */
 888                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
 889                                                          | NLM_F_EXCL))) {
 890                         error = -EEXIST;
 891                         goto err_unlock_ovs;
 892                 }
 893                 /* The unmasked key has to be the same for flow updates. */
 894                 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
 895                         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
 896                         if (!flow) {
 897                                 error = -ENOENT;
 898                                 goto err_unlock_ovs;
 899                         }
 900                 }
 901                 /* Update actions. */
 902                 old_acts = ovsl_dereference(flow->sf_acts);
 903                 rcu_assign_pointer(flow->sf_acts, acts);
 904
 905                 if (unlikely(reply)) {
 906                         error = ovs_flow_cmd_fill_info(flow,
 907                                                        ovs_header->dp_ifindex,
 908                                                        reply, info->snd_portid,
 909                                                        info->snd_seq, 0,
 910                                                        OVS_FLOW_CMD_NEW);
 911                         BUG_ON(error < 0);
 912                 }
 913                 ovs_unlock();
 914
 915                 ovs_nla_free_flow_actions(old_acts);
 916                 ovs_flow_free(new_flow, false);
 917         }
 918
 919         if (reply)
 920                 ovs_notify(&dp_flow_genl_family, reply, info);
 921         return 0;
 922
 923 err_unlock_ovs:
 924         ovs_unlock();
 925         kfree_skb(reply);
 926 err_kfree_acts:
 927         kfree(acts);
 928 err_kfree_flow:
 929         ovs_flow_free(new_flow, false);
 930 error:
 931         return error;
 932 }
 933
 934 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
 935 {
 936         struct nlattr **a = info->attrs;
 937         struct ovs_header *ovs_header = info->userhdr;
 938         struct sw_flow_key key, masked_key;
 939         struct sw_flow *flow;
 940         struct sw_flow_mask mask;
 941         struct sk_buff *reply = NULL;
 942         struct datapath *dp;
 943         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
 944         struct sw_flow_match match;
 945         int error;
 946
 947         /* Extract key. */
 948         error = -EINVAL;
 949         if (!a[OVS_FLOW_ATTR_KEY])
 950                 goto error;
 951
 952         ovs_match_init(&match, &key, &mask);
 953         error = ovs_nla_get_match(&match,
 954                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 955         if (error)
 956                 goto error;
 957
 958         /* Validate actions. */
 959         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 960                 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 961                 error = PTR_ERR(acts);
 962                 if (IS_ERR(acts))
 963                         goto error;
 964
 965                 ovs_flow_mask_key(&masked_key, &key, &mask);
 966                 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
 967                                              &masked_key, 0, &acts);
 968                 if (error) {
 969                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 970                         goto err_kfree_acts;
 971                 }
 972         }
 973
 974         /* Can allocate before locking if have acts. */
 975         if (acts) {
 976                 reply = ovs_flow_cmd_alloc_info(acts, info, false);
 977                 if (IS_ERR(reply)) {
 978                         error = PTR_ERR(reply);
 979                         goto err_kfree_acts;
 980                 }
 981         }
 982
 983         ovs_lock();
 984         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 985         if (unlikely(!dp)) {
 986                 error = -ENODEV;
 987                 goto err_unlock_ovs;
 988         }
 989         /* Check that the flow exists. */
 990         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
 991         if (unlikely(!flow)) {
 992                 error = -ENOENT;
 993                 goto err_unlock_ovs;
 994         }
 995
 996         /* Update actions, if present. */
 997         if (likely(acts)) {
 998                 old_acts = ovsl_dereference(flow->sf_acts);
 999                 rcu_assign_pointer(flow->sf_acts, acts);
1000
1001                 if (unlikely(reply)) {
1002                         error = ovs_flow_cmd_fill_info(flow,
1003                                                        ovs_header->dp_ifindex,
1004                                                        reply, info->snd_portid,
1005                                                        info->snd_seq, 0,
1006                                                        OVS_FLOW_CMD_NEW);
1007                         BUG_ON(error < 0);
1008                 }
1009         } else {
1010                 /* Could not alloc without acts before locking. */
1011                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1012                                                 info, OVS_FLOW_CMD_NEW, false);
1013                 if (unlikely(IS_ERR(reply))) {
1014                         error = PTR_ERR(reply);
1015                         goto err_unlock_ovs;
1016                 }
1017         }
1018
1019         /* Clear stats. */
1020         if (a[OVS_FLOW_ATTR_CLEAR])
1021                 ovs_flow_stats_clear(flow);
1022         ovs_unlock();
1023
1024         if (reply)
1025                 ovs_notify(&dp_flow_genl_family, reply, info);
1026         if (old_acts)
1027                 ovs_nla_free_flow_actions(old_acts);
1028
1029         return 0;
1030
1031 err_unlock_ovs:
1032         ovs_unlock();
1033         kfree_skb(reply);
1034 err_kfree_acts:
1035         kfree(acts);
1036 error:
1037         return error;
1038 }
1039
1040 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1041 {
1042         struct nlattr **a = info->attrs;
1043         struct ovs_header *ovs_header = info->userhdr;
1044         struct sw_flow_key key;
1045         struct sk_buff *reply;
1046         struct sw_flow *flow;
1047         struct datapath *dp;
1048         struct sw_flow_match match;
1049         int err;
1050
1051         if (!a[OVS_FLOW_ATTR_KEY]) {
1052                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1053                 return -EINVAL;
1054         }
1055
1056         ovs_match_init(&match, &key, NULL);
1057         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1058         if (err)
1059                 return err;
1060
1061         ovs_lock();
1062         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1063         if (!dp) {
1064                 err = -ENODEV;
1065                 goto unlock;
1066         }
1067
1068         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1069         if (!flow) {
1070                 err = -ENOENT;
1071                 goto unlock;
1072         }
1073
1074         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1075                                         OVS_FLOW_CMD_NEW, true);
1076         if (IS_ERR(reply)) {
1077                 err = PTR_ERR(reply);
1078                 goto unlock;
1079         }
1080
1081         ovs_unlock();
1082         return genlmsg_reply(reply, info);
1083 unlock:
1084         ovs_unlock();
1085         return err;
1086 }
1087
1088 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1089 {
1090         struct nlattr **a = info->attrs;
1091         struct ovs_header *ovs_header = info->userhdr;
1092         struct sw_flow_key key;
1093         struct sk_buff *reply;
1094         struct sw_flow *flow;
1095         struct datapath *dp;
1096         struct sw_flow_match match;
1097         int err;
1098
1099         if (likely(a[OVS_FLOW_ATTR_KEY])) {
1100                 ovs_match_init(&match, &key, NULL);
1101                 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1102                 if (unlikely(err))
1103                         return err;
1104         }
1105
1106         ovs_lock();
1107         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1108         if (unlikely(!dp)) {
1109                 err = -ENODEV;
1110                 goto unlock;
1111         }
1112
1113         if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1114                 err = ovs_flow_tbl_flush(&dp->table);
1115                 goto unlock;
1116         }
1117
1118         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1119         if (unlikely(!flow)) {
1120                 err = -ENOENT;
1121                 goto unlock;
1122         }
1123
1124         ovs_flow_tbl_remove(&dp->table, flow);
1125         ovs_unlock();
1126
1127         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1128                                         info, false);
1129         if (likely(reply)) {
1130                 if (likely(!IS_ERR(reply))) {
1131                         rcu_read_lock();        /*To keep RCU checker happy. */
1132                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1133                                                      reply, info->snd_portid,
1134                                                      info->snd_seq, 0,
1135                                                      OVS_FLOW_CMD_DEL);
1136                         rcu_read_unlock();
1137                         BUG_ON(err < 0);
1138
1139                         ovs_notify(&dp_flow_genl_family, reply, info);
1140                 } else {
1141                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1142                 }
1143         }
1144
1145         ovs_flow_free(flow, true);
1146         return 0;
1147 unlock:
1148         ovs_unlock();
1149         return err;
1150 }
1151
1152 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1153 {
1154         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1155         struct table_instance *ti;
1156         struct datapath *dp;
1157
1158         rcu_read_lock();
1159         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1160         if (!dp) {
1161                 rcu_read_unlock();
1162                 return -ENODEV;
1163         }
1164
1165         ti = rcu_dereference(dp->table.ti);
1166         for (;;) {
1167                 struct sw_flow *flow;
1168                 u32 bucket, obj;
1169
1170                 bucket = cb->args[0];
1171                 obj = cb->args[1];
1172                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1173                 if (!flow)
1174                         break;
1175
1176                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1177                                            NETLINK_CB(cb->skb).portid,
1178                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1179                                            OVS_FLOW_CMD_NEW) < 0)
1180                         break;
1181
1182                 cb->args[0] = bucket;
1183                 cb->args[1] = obj;
1184         }
1185         rcu_read_unlock();
1186         return skb->len;
1187 }
1188
1189 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1190         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1191         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1192         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1193 };
1194
1195 static const struct genl_ops dp_flow_genl_ops[] = {
1196         { .cmd = OVS_FLOW_CMD_NEW,
1197           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1198           .policy = flow_policy,
1199           .doit = ovs_flow_cmd_new
1200         },
1201         { .cmd = OVS_FLOW_CMD_DEL,
1202           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1203           .policy = flow_policy,
1204           .doit = ovs_flow_cmd_del
1205         },
1206         { .cmd = OVS_FLOW_CMD_GET,
1207           .flags = 0,               /* OK for unprivileged users. */
1208           .policy = flow_policy,
1209           .doit = ovs_flow_cmd_get,
1210           .dumpit = ovs_flow_cmd_dump
1211         },
1212         { .cmd = OVS_FLOW_CMD_SET,
1213           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1214           .policy = flow_policy,
1215           .doit = ovs_flow_cmd_set,
1216         },
1217 };
1218
1219 static struct genl_family dp_flow_genl_family = {
1220         .id = GENL_ID_GENERATE,
1221         .hdrsize = sizeof(struct ovs_header),
1222         .name = OVS_FLOW_FAMILY,
1223         .version = OVS_FLOW_VERSION,
1224         .maxattr = OVS_FLOW_ATTR_MAX,
1225         .netnsok = true,
1226         .parallel_ops = true,
1227         .ops = dp_flow_genl_ops,
1228         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1229         .mcgrps = &ovs_dp_flow_multicast_group,
1230         .n_mcgrps = 1,
1231 };
1232
1233 static size_t ovs_dp_cmd_msg_size(void)
1234 {
1235         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1236
1237         msgsize += nla_total_size(IFNAMSIZ);
1238         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1239         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1240         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1241
1242         return msgsize;
1243 }
1244
1245 /* Called with ovs_mutex or RCU read lock. */
1246 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1247                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1248 {
1249         struct ovs_header *ovs_header;
1250         struct ovs_dp_stats dp_stats;
1251         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1252         int err;
1253
1254         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1255                                    flags, cmd);
1256         if (!ovs_header)
1257                 goto error;
1258
1259         ovs_header->dp_ifindex = get_dpifindex(dp);
1260
1261         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1262         if (err)
1263                 goto nla_put_failure;
1264
1265         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1266         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1267                         &dp_stats))
1268                 goto nla_put_failure;
1269
1270         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1271                         sizeof(struct ovs_dp_megaflow_stats),
1272                         &dp_megaflow_stats))
1273                 goto nla_put_failure;
1274
1275         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1276                 goto nla_put_failure;
1277
1278         return genlmsg_end(skb, ovs_header);
1279
1280 nla_put_failure:
1281         genlmsg_cancel(skb, ovs_header);
1282 error:
1283         return -EMSGSIZE;
1284 }
1285
1286 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1287 {
1288         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1289 }
1290
1291 /* Called with rcu_read_lock or ovs_mutex. */
1292 static struct datapath *lookup_datapath(struct net *net,
1293                                         struct ovs_header *ovs_header,
1294                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1295 {
1296         struct datapath *dp;
1297
1298         if (!a[OVS_DP_ATTR_NAME])
1299                 dp = get_dp(net, ovs_header->dp_ifindex);
1300         else {
1301                 struct vport *vport;
1302
1303                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1304                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1305         }
1306         return dp ? dp : ERR_PTR(-ENODEV);
1307 }
1308
1309 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1310 {
1311         struct datapath *dp;
1312
1313         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1314         if (IS_ERR(dp))
1315                 return;
1316
1317         WARN(dp->user_features, "Dropping previously announced user features\n");
1318         dp->user_features = 0;
1319 }
1320
1321 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1322 {
1323         if (a[OVS_DP_ATTR_USER_FEATURES])
1324                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1325 }
1326
1327 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1328 {
1329         struct nlattr **a = info->attrs;
1330         struct vport_parms parms;
1331         struct sk_buff *reply;
1332         struct datapath *dp;
1333         struct vport *vport;
1334         struct ovs_net *ovs_net;
1335         int err, i;
1336
1337         err = -EINVAL;
1338         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1339                 goto err;
1340
1341         reply = ovs_dp_cmd_alloc_info(info);
1342         if (!reply)
1343                 return -ENOMEM;
1344
1345         err = -ENOMEM;
1346         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1347         if (dp == NULL)
1348                 goto err_free_reply;
1349
1350         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1351
1352         /* Allocate table. */
1353         err = ovs_flow_tbl_init(&dp->table);
1354         if (err)
1355                 goto err_free_dp;
1356
1357         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1358         if (!dp->stats_percpu) {
1359                 err = -ENOMEM;
1360                 goto err_destroy_table;
1361         }
1362
1363         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1364                             GFP_KERNEL);
1365         if (!dp->ports) {
1366                 err = -ENOMEM;
1367                 goto err_destroy_percpu;
1368         }
1369
1370         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1371                 INIT_HLIST_HEAD(&dp->ports[i]);
1372
1373         /* Set up our datapath device. */
1374         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1375         parms.type = OVS_VPORT_TYPE_INTERNAL;
1376         parms.options = NULL;
1377         parms.dp = dp;
1378         parms.port_no = OVSP_LOCAL;
1379         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1380
1381         ovs_dp_change(dp, a);
1382
1383         /* So far only local changes have been made, now need the lock. */
1384         ovs_lock();
1385
1386         vport = new_vport(&parms);
1387         if (IS_ERR(vport)) {
1388                 err = PTR_ERR(vport);
1389                 if (err == -EBUSY)
1390                         err = -EEXIST;
1391
1392                 if (err == -EEXIST) {
1393                         /* An outdated user space instance that does not understand
1394                          * the concept of user_features has attempted to create a new
1395                          * datapath and is likely to reuse it. Drop all user features.
1396                          */
1397                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1398                                 ovs_dp_reset_user_features(skb, info);
1399                 }
1400
1401                 goto err_destroy_ports_array;
1402         }
1403
1404         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1405                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1406         BUG_ON(err < 0);
1407
1408         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1409         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1410
1411         ovs_unlock();
1412
1413         ovs_notify(&dp_datapath_genl_family, reply, info);
1414         return 0;
1415
1416 err_destroy_ports_array:
1417         ovs_unlock();
1418         kfree(dp->ports);
1419 err_destroy_percpu:
1420         free_percpu(dp->stats_percpu);
1421 err_destroy_table:
1422         ovs_flow_tbl_destroy(&dp->table, false);
1423 err_free_dp:
1424         release_net(ovs_dp_get_net(dp));
1425         kfree(dp);
1426 err_free_reply:
1427         kfree_skb(reply);
1428 err:
1429         return err;
1430 }
1431
1432 /* Called with ovs_mutex. */
1433 static void __dp_destroy(struct datapath *dp)
1434 {
1435         int i;
1436
1437         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1438                 struct vport *vport;
1439                 struct hlist_node *n;
1440
1441                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1442                         if (vport->port_no != OVSP_LOCAL)
1443                                 ovs_dp_detach_port(vport);
1444         }
1445
1446         list_del_rcu(&dp->list_node);
1447
1448         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1449          * all ports in datapath are destroyed first before freeing datapath.
1450          */
1451         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1452
1453         /* RCU destroy the flow table */
1454         ovs_flow_tbl_destroy(&dp->table, true);
1455
1456         call_rcu(&dp->rcu, destroy_dp_rcu);
1457 }
1458
1459 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1460 {
1461         struct sk_buff *reply;
1462         struct datapath *dp;
1463         int err;
1464
1465         reply = ovs_dp_cmd_alloc_info(info);
1466         if (!reply)
1467                 return -ENOMEM;
1468
1469         ovs_lock();
1470         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1471         err = PTR_ERR(dp);
1472         if (IS_ERR(dp))
1473                 goto err_unlock_free;
1474
1475         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1476                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1477         BUG_ON(err < 0);
1478
1479         __dp_destroy(dp);
1480         ovs_unlock();
1481
1482         ovs_notify(&dp_datapath_genl_family, reply, info);
1483
1484         return 0;
1485
1486 err_unlock_free:
1487         ovs_unlock();
1488         kfree_skb(reply);
1489         return err;
1490 }
1491
1492 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1493 {
1494         struct sk_buff *reply;
1495         struct datapath *dp;
1496         int err;
1497
1498         reply = ovs_dp_cmd_alloc_info(info);
1499         if (!reply)
1500                 return -ENOMEM;
1501
1502         ovs_lock();
1503         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1504         err = PTR_ERR(dp);
1505         if (IS_ERR(dp))
1506                 goto err_unlock_free;
1507
1508         ovs_dp_change(dp, info->attrs);
1509
1510         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1511                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1512         BUG_ON(err < 0);
1513
1514         ovs_unlock();
1515         ovs_notify(&dp_datapath_genl_family, reply, info);
1516
1517         return 0;
1518
1519 err_unlock_free:
1520         ovs_unlock();
1521         kfree_skb(reply);
1522         return err;
1523 }
1524
1525 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1526 {
1527         struct sk_buff *reply;
1528         struct datapath *dp;
1529         int err;
1530
1531         reply = ovs_dp_cmd_alloc_info(info);
1532         if (!reply)
1533                 return -ENOMEM;
1534
1535         rcu_read_lock();
1536         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1537         if (IS_ERR(dp)) {
1538                 err = PTR_ERR(dp);
1539                 goto err_unlock_free;
1540         }
1541         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1542                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1543         BUG_ON(err < 0);
1544         rcu_read_unlock();
1545
1546         return genlmsg_reply(reply, info);
1547
1548 err_unlock_free:
1549         rcu_read_unlock();
1550         kfree_skb(reply);
1551         return err;
1552 }
1553
1554 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1555 {
1556         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1557         struct datapath *dp;
1558         int skip = cb->args[0];
1559         int i = 0;
1560
1561         rcu_read_lock();
1562         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1563                 if (i >= skip &&
1564                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1565                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1566                                          OVS_DP_CMD_NEW) < 0)
1567                         break;
1568                 i++;
1569         }
1570         rcu_read_unlock();
1571
1572         cb->args[0] = i;
1573
1574         return skb->len;
1575 }
1576
1577 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1578         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1579         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1580         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1581 };
1582
1583 static const struct genl_ops dp_datapath_genl_ops[] = {
1584         { .cmd = OVS_DP_CMD_NEW,
1585           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1586           .policy = datapath_policy,
1587           .doit = ovs_dp_cmd_new
1588         },
1589         { .cmd = OVS_DP_CMD_DEL,
1590           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1591           .policy = datapath_policy,
1592           .doit = ovs_dp_cmd_del
1593         },
1594         { .cmd = OVS_DP_CMD_GET,
1595           .flags = 0,               /* OK for unprivileged users. */
1596           .policy = datapath_policy,
1597           .doit = ovs_dp_cmd_get,
1598           .dumpit = ovs_dp_cmd_dump
1599         },
1600         { .cmd = OVS_DP_CMD_SET,
1601           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1602           .policy = datapath_policy,
1603           .doit = ovs_dp_cmd_set,
1604         },
1605 };
1606
1607 static struct genl_family dp_datapath_genl_family = {
1608         .id = GENL_ID_GENERATE,
1609         .hdrsize = sizeof(struct ovs_header),
1610         .name = OVS_DATAPATH_FAMILY,
1611         .version = OVS_DATAPATH_VERSION,
1612         .maxattr = OVS_DP_ATTR_MAX,
1613         .netnsok = true,
1614         .parallel_ops = true,
1615         .ops = dp_datapath_genl_ops,
1616         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1617         .mcgrps = &ovs_dp_datapath_multicast_group,
1618         .n_mcgrps = 1,
1619 };
1620
1621 /* Called with ovs_mutex or RCU read lock. */
1622 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1623                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1624 {
1625         struct ovs_header *ovs_header;
1626         struct ovs_vport_stats vport_stats;
1627         int err;
1628
1629         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1630                                  flags, cmd);
1631         if (!ovs_header)
1632                 return -EMSGSIZE;
1633
1634         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1635
1636         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1637             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1638             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1639                            vport->ops->get_name(vport)))
1640                 goto nla_put_failure;
1641
1642         ovs_vport_get_stats(vport, &vport_stats);
1643         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1644                     &vport_stats))
1645                 goto nla_put_failure;
1646
1647         if (ovs_vport_get_upcall_portids(vport, skb))
1648                 goto nla_put_failure;
1649
1650         err = ovs_vport_get_options(vport, skb);
1651         if (err == -EMSGSIZE)
1652                 goto error;
1653
1654         return genlmsg_end(skb, ovs_header);
1655
1656 nla_put_failure:
1657         err = -EMSGSIZE;
1658 error:
1659         genlmsg_cancel(skb, ovs_header);
1660         return err;
1661 }
1662
1663 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1664 {
1665         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1666 }
1667
1668 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1669 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1670                                          u32 seq, u8 cmd)
1671 {
1672         struct sk_buff *skb;
1673         int retval;
1674
1675         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1676         if (!skb)
1677                 return ERR_PTR(-ENOMEM);
1678
1679         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1680         BUG_ON(retval < 0);
1681
1682         return skb;
1683 }
1684
1685 /* Called with ovs_mutex or RCU read lock. */
1686 static struct vport *lookup_vport(struct net *net,
1687                                   struct ovs_header *ovs_header,
1688                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1689 {
1690         struct datapath *dp;
1691         struct vport *vport;
1692
1693         if (a[OVS_VPORT_ATTR_NAME]) {
1694                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1695                 if (!vport)
1696                         return ERR_PTR(-ENODEV);
1697                 if (ovs_header->dp_ifindex &&
1698                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1699                         return ERR_PTR(-ENODEV);
1700                 return vport;
1701         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1702                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1703
1704                 if (port_no >= DP_MAX_PORTS)
1705                         return ERR_PTR(-EFBIG);
1706
1707                 dp = get_dp(net, ovs_header->dp_ifindex);
1708                 if (!dp)
1709                         return ERR_PTR(-ENODEV);
1710
1711                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1712                 if (!vport)
1713                         return ERR_PTR(-ENODEV);
1714                 return vport;
1715         } else
1716                 return ERR_PTR(-EINVAL);
1717 }
1718
1719 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1720 {
1721         struct nlattr **a = info->attrs;
1722         struct ovs_header *ovs_header = info->userhdr;
1723         struct vport_parms parms;
1724         struct sk_buff *reply;
1725         struct vport *vport;
1726         struct datapath *dp;
1727         u32 port_no;
1728         int err;
1729
1730         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1731             !a[OVS_VPORT_ATTR_UPCALL_PID])
1732                 return -EINVAL;
1733
1734         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1735                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1736         if (port_no >= DP_MAX_PORTS)
1737                 return -EFBIG;
1738
1739         reply = ovs_vport_cmd_alloc_info();
1740         if (!reply)
1741                 return -ENOMEM;
1742
1743         ovs_lock();
1744         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1745         err = -ENODEV;
1746         if (!dp)
1747                 goto exit_unlock_free;
1748
1749         if (port_no) {
1750                 vport = ovs_vport_ovsl(dp, port_no);
1751                 err = -EBUSY;
1752                 if (vport)
1753                         goto exit_unlock_free;
1754         } else {
1755                 for (port_no = 1; ; port_no++) {
1756                         if (port_no >= DP_MAX_PORTS) {
1757                                 err = -EFBIG;
1758                                 goto exit_unlock_free;
1759                         }
1760                         vport = ovs_vport_ovsl(dp, port_no);
1761                         if (!vport)
1762                                 break;
1763                 }
1764         }
1765
1766         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1767         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1768         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1769         parms.dp = dp;
1770         parms.port_no = port_no;
1771         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1772
1773         vport = new_vport(&parms);
1774         err = PTR_ERR(vport);
1775         if (IS_ERR(vport))
1776                 goto exit_unlock_free;
1777
1778         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1779                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1780         BUG_ON(err < 0);
1781         ovs_unlock();
1782
1783         ovs_notify(&dp_vport_genl_family, reply, info);
1784         return 0;
1785
1786 exit_unlock_free:
1787         ovs_unlock();
1788         kfree_skb(reply);
1789         return err;
1790 }
1791
1792 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1793 {
1794         struct nlattr **a = info->attrs;
1795         struct sk_buff *reply;
1796         struct vport *vport;
1797         int err;
1798
1799         reply = ovs_vport_cmd_alloc_info();
1800         if (!reply)
1801                 return -ENOMEM;
1802
1803         ovs_lock();
1804         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1805         err = PTR_ERR(vport);
1806         if (IS_ERR(vport))
1807                 goto exit_unlock_free;
1808
1809         if (a[OVS_VPORT_ATTR_TYPE] &&
1810             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1811                 err = -EINVAL;
1812                 goto exit_unlock_free;
1813         }
1814
1815         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1816                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1817                 if (err)
1818                         goto exit_unlock_free;
1819         }
1820
1821
1822         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1823                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
1824
1825                 err = ovs_vport_set_upcall_portids(vport, ids);
1826                 if (err)
1827                         goto exit_unlock_free;
1828         }
1829
1830         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1831                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1832         BUG_ON(err < 0);
1833
1834         ovs_unlock();
1835         ovs_notify(&dp_vport_genl_family, reply, info);
1836         return 0;
1837
1838 exit_unlock_free:
1839         ovs_unlock();
1840         kfree_skb(reply);
1841         return err;
1842 }
1843
1844 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1845 {
1846         struct nlattr **a = info->attrs;
1847         struct sk_buff *reply;
1848         struct vport *vport;
1849         int err;
1850
1851         reply = ovs_vport_cmd_alloc_info();
1852         if (!reply)
1853                 return -ENOMEM;
1854
1855         ovs_lock();
1856         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1857         err = PTR_ERR(vport);
1858         if (IS_ERR(vport))
1859                 goto exit_unlock_free;
1860
1861         if (vport->port_no == OVSP_LOCAL) {
1862                 err = -EINVAL;
1863                 goto exit_unlock_free;
1864         }
1865
1866         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1867                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1868         BUG_ON(err < 0);
1869         ovs_dp_detach_port(vport);
1870         ovs_unlock();
1871
1872         ovs_notify(&dp_vport_genl_family, reply, info);
1873         return 0;
1874
1875 exit_unlock_free:
1876         ovs_unlock();
1877         kfree_skb(reply);
1878         return err;
1879 }
1880
1881 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1882 {
1883         struct nlattr **a = info->attrs;
1884         struct ovs_header *ovs_header = info->userhdr;
1885         struct sk_buff *reply;
1886         struct vport *vport;
1887         int err;
1888
1889         reply = ovs_vport_cmd_alloc_info();
1890         if (!reply)
1891                 return -ENOMEM;
1892
1893         rcu_read_lock();
1894         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1895         err = PTR_ERR(vport);
1896         if (IS_ERR(vport))
1897                 goto exit_unlock_free;
1898         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1899                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1900         BUG_ON(err < 0);
1901         rcu_read_unlock();
1902
1903         return genlmsg_reply(reply, info);
1904
1905 exit_unlock_free:
1906         rcu_read_unlock();
1907         kfree_skb(reply);
1908         return err;
1909 }
1910
1911 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1912 {
1913         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1914         struct datapath *dp;
1915         int bucket = cb->args[0], skip = cb->args[1];
1916         int i, j = 0;
1917
1918         rcu_read_lock();
1919         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1920         if (!dp) {
1921                 rcu_read_unlock();
1922                 return -ENODEV;
1923         }
1924         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1925                 struct vport *vport;
1926
1927                 j = 0;
1928                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1929                         if (j >= skip &&
1930                             ovs_vport_cmd_fill_info(vport, skb,
1931                                                     NETLINK_CB(cb->skb).portid,
1932                                                     cb->nlh->nlmsg_seq,
1933                                                     NLM_F_MULTI,
1934                                                     OVS_VPORT_CMD_NEW) < 0)
1935                                 goto out;
1936
1937                         j++;
1938                 }
1939                 skip = 0;
1940         }
1941 out:
1942         rcu_read_unlock();
1943
1944         cb->args[0] = i;
1945         cb->args[1] = j;
1946
1947         return skb->len;
1948 }
1949
1950 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1951         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1952         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1953         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1954         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1955         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1956         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1957 };
1958
1959 static const struct genl_ops dp_vport_genl_ops[] = {
1960         { .cmd = OVS_VPORT_CMD_NEW,
1961           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1962           .policy = vport_policy,
1963           .doit = ovs_vport_cmd_new
1964         },
1965         { .cmd = OVS_VPORT_CMD_DEL,
1966           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1967           .policy = vport_policy,
1968           .doit = ovs_vport_cmd_del
1969         },
1970         { .cmd = OVS_VPORT_CMD_GET,
1971           .flags = 0,               /* OK for unprivileged users. */
1972           .policy = vport_policy,
1973           .doit = ovs_vport_cmd_get,
1974           .dumpit = ovs_vport_cmd_dump
1975         },
1976         { .cmd = OVS_VPORT_CMD_SET,
1977           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1978           .policy = vport_policy,
1979           .doit = ovs_vport_cmd_set,
1980         },
1981 };
1982
1983 struct genl_family dp_vport_genl_family = {
1984         .id = GENL_ID_GENERATE,
1985         .hdrsize = sizeof(struct ovs_header),
1986         .name = OVS_VPORT_FAMILY,
1987         .version = OVS_VPORT_VERSION,
1988         .maxattr = OVS_VPORT_ATTR_MAX,
1989         .netnsok = true,
1990         .parallel_ops = true,
1991         .ops = dp_vport_genl_ops,
1992         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
1993         .mcgrps = &ovs_dp_vport_multicast_group,
1994         .n_mcgrps = 1,
1995 };
1996
1997 static struct genl_family * const dp_genl_families[] = {
1998         &dp_datapath_genl_family,
1999         &dp_vport_genl_family,
2000         &dp_flow_genl_family,
2001         &dp_packet_genl_family,
2002 };
2003
2004 static void dp_unregister_genl(int n_families)
2005 {
2006         int i;
2007
2008         for (i = 0; i < n_families; i++)
2009                 genl_unregister_family(dp_genl_families[i]);
2010 }
2011
2012 static int dp_register_genl(void)
2013 {
2014         int err;
2015         int i;
2016
2017         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2018
2019                 err = genl_register_family(dp_genl_families[i]);
2020                 if (err)
2021                         goto error;
2022         }
2023
2024         return 0;
2025
2026 error:
2027         dp_unregister_genl(i);
2028         return err;
2029 }
2030
2031 static int __net_init ovs_init_net(struct net *net)
2032 {
2033         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2034
2035         INIT_LIST_HEAD(&ovs_net->dps);
2036         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2037         return 0;
2038 }
2039
2040 static void __net_exit ovs_exit_net(struct net *net)
2041 {
2042         struct datapath *dp, *dp_next;
2043         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2044
2045         ovs_lock();
2046         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2047                 __dp_destroy(dp);
2048         ovs_unlock();
2049
2050         cancel_work_sync(&ovs_net->dp_notify_work);
2051 }
2052
2053 static struct pernet_operations ovs_net_ops = {
2054         .init = ovs_init_net,
2055         .exit = ovs_exit_net,
2056         .id   = &ovs_net_id,
2057         .size = sizeof(struct ovs_net),
2058 };
2059
2060 static int __init dp_init(void)
2061 {
2062         int err;
2063
2064         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2065
2066         pr_info("Open vSwitch switching datapath\n");
2067
2068         err = ovs_internal_dev_rtnl_link_register();
2069         if (err)
2070                 goto error;
2071
2072         err = ovs_flow_init();
2073         if (err)
2074                 goto error_unreg_rtnl_link;
2075
2076         err = ovs_vport_init();
2077         if (err)
2078                 goto error_flow_exit;
2079
2080         err = register_pernet_device(&ovs_net_ops);
2081         if (err)
2082                 goto error_vport_exit;
2083
2084         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2085         if (err)
2086                 goto error_netns_exit;
2087
2088         err = dp_register_genl();
2089         if (err < 0)
2090                 goto error_unreg_notifier;
2091
2092         return 0;
2093
2094 error_unreg_notifier:
2095         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2096 error_netns_exit:
2097         unregister_pernet_device(&ovs_net_ops);
2098 error_vport_exit:
2099         ovs_vport_exit();
2100 error_flow_exit:
2101         ovs_flow_exit();
2102 error_unreg_rtnl_link:
2103         ovs_internal_dev_rtnl_link_unregister();
2104 error:
2105         return err;
2106 }
2107
2108 static void dp_cleanup(void)
2109 {
2110         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2111         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2112         unregister_pernet_device(&ovs_net_ops);
2113         rcu_barrier();
2114         ovs_vport_exit();
2115         ovs_flow_exit();
2116         ovs_internal_dev_rtnl_link_unregister();
2117 }
2118
2119 module_init(dp_init);
2120 module_exit(dp_cleanup);
2121
2122 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2123 MODULE_LICENSE("GPL");