datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007-2014 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/version.h>
  40 #include <linux/ethtool.h>
  41 #include <linux/wait.h>
  42 #include <asm/div64.h>
  43 #include <linux/highmem.h>
  44 #include <linux/netfilter_bridge.h>
  45 #include <linux/netfilter_ipv4.h>
  46 #include <linux/inetdevice.h>
  47 #include <linux/list.h>
  48 #include <linux/openvswitch.h>
  49 #include <linux/rculist.h>
  50 #include <linux/dmi.h>
  51 #include <linux/genetlink.h>
  52 #include <net/genetlink.h>
  53 #include <net/genetlink.h>
  54 #include <net/net_namespace.h>
  55 #include <net/netns/generic.h>
  56
  57 #include "datapath.h"
  58 #include "flow.h"
  59 #include "flow_table.h"
  60 #include "flow_netlink.h"
  61 #include "vlan.h"
  62 #include "vport-internal_dev.h"
  63 #include "vport-netdev.h"
  64
  65 int ovs_net_id __read_mostly;
  66
  67 static struct genl_family dp_packet_genl_family;
  68 static struct genl_family dp_flow_genl_family;
  69 static struct genl_family dp_datapath_genl_family;
  70
  71 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
  72         .name = OVS_FLOW_MCGROUP
  73 };
  74
  75 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  76         .name = OVS_DATAPATH_MCGROUP
  77 };
  78
  79 struct genl_multicast_group ovs_dp_vport_multicast_group = {
  80         .name = OVS_VPORT_MCGROUP
  81 };
  82
  83 /* Check if need to build a reply message.
  84  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  85 static bool ovs_must_notify(struct genl_info *info,
  86                             const struct genl_multicast_group *grp)
  87 {
  88         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  89                 netlink_has_listeners(genl_info_net(info)->genl_sock, GROUP_ID(grp));
  90 }
  91
  92 static void ovs_notify(struct genl_family *family, struct genl_multicast_group *grp,
  93                        struct sk_buff *skb, struct genl_info *info)
  94 {
  95         genl_notify(family, skb, genl_info_net(info),
  96                     info->snd_portid, GROUP_ID(grp), info->nlhdr, GFP_KERNEL);
  97 }
  98
  99 /**
 100  * DOC: Locking:
 101  *
 102  * All writes e.g. Writes to device state (add/remove datapath, port, set
 103  * operations on vports, etc.), Writes to other state (flow table
 104  * modifications, set miscellaneous datapath parameters, etc.) are protected
 105  * by ovs_lock.
 106  *
 107  * Reads are protected by RCU.
 108  *
 109  * There are a few special cases (mostly stats) that have their own
 110  * synchronization but they nest under all of above and don't interact with
 111  * each other.
 112  *
 113  * The RTNL lock nests inside ovs_mutex.
 114  */
 115
 116 static DEFINE_MUTEX(ovs_mutex);
 117
 118 void ovs_lock(void)
 119 {
 120         mutex_lock(&ovs_mutex);
 121 }
 122
 123 void ovs_unlock(void)
 124 {
 125         mutex_unlock(&ovs_mutex);
 126 }
 127
 128 #ifdef CONFIG_LOCKDEP
 129 int lockdep_ovsl_is_held(void)
 130 {
 131         if (debug_locks)
 132                 return lockdep_is_held(&ovs_mutex);
 133         else
 134                 return 1;
 135 }
 136 #endif
 137
 138 static struct vport *new_vport(const struct vport_parms *);
 139 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 140                              const struct dp_upcall_info *);
 141 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 142                                   const struct dp_upcall_info *);
 143
 144 /* Must be called with rcu_read_lock or ovs_mutex. */
 145 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 146 {
 147         struct datapath *dp = NULL;
 148         struct net_device *dev;
 149
 150         rcu_read_lock();
 151         dev = dev_get_by_index_rcu(net, dp_ifindex);
 152         if (dev) {
 153                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 154                 if (vport)
 155                         dp = vport->dp;
 156         }
 157         rcu_read_unlock();
 158
 159         return dp;
 160 }
 161
 162 /* Must be called with rcu_read_lock or ovs_mutex. */
 163 const char *ovs_dp_name(const struct datapath *dp)
 164 {
 165         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 166         return vport->ops->get_name(vport);
 167 }
 168
 169 static int get_dpifindex(struct datapath *dp)
 170 {
 171         struct vport *local;
 172         int ifindex;
 173
 174         rcu_read_lock();
 175
 176         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 177         if (local)
 178                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 179         else
 180                 ifindex = 0;
 181
 182         rcu_read_unlock();
 183
 184         return ifindex;
 185 }
 186
 187 static void destroy_dp_rcu(struct rcu_head *rcu)
 188 {
 189         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 190
 191         free_percpu(dp->stats_percpu);
 192         release_net(ovs_dp_get_net(dp));
 193         kfree(dp->ports);
 194         kfree(dp);
 195 }
 196
 197 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 198                                             u16 port_no)
 199 {
 200         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 201 }
 202
 203 /* Called with ovs_mutex or RCU read lock. */
 204 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 205 {
 206         struct vport *vport;
 207         struct hlist_head *head;
 208
 209         head = vport_hash_bucket(dp, port_no);
 210         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 211                 if (vport->port_no == port_no)
 212                         return vport;
 213         }
 214         return NULL;
 215 }
 216
 217 /* Called with ovs_mutex. */
 218 static struct vport *new_vport(const struct vport_parms *parms)
 219 {
 220         struct vport *vport;
 221
 222         vport = ovs_vport_add(parms);
 223         if (!IS_ERR(vport)) {
 224                 struct datapath *dp = parms->dp;
 225                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 226
 227                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 228         }
 229         return vport;
 230 }
 231
 232 void ovs_dp_detach_port(struct vport *p)
 233 {
 234         ASSERT_OVSL();
 235
 236         /* First drop references to device. */
 237         hlist_del_rcu(&p->dp_hash_node);
 238
 239         /* Then destroy it. */
 240         ovs_vport_del(p);
 241 }
 242
 243 /* Must be called with rcu_read_lock. */
 244 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 245 {
 246         struct datapath *dp = p->dp;
 247         struct sw_flow *flow;
 248         struct dp_stats_percpu *stats;
 249         struct sw_flow_key key;
 250         u64 *stats_counter;
 251         u32 n_mask_hit;
 252         int error;
 253
 254         stats = this_cpu_ptr(dp->stats_percpu);
 255
 256         /* Extract flow from 'skb' into 'key'. */
 257         error = ovs_flow_extract(skb, p->port_no, &key);
 258         if (unlikely(error)) {
 259                 kfree_skb(skb);
 260                 return;
 261         }
 262
 263         /* Look up flow. */
 264         flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
 265         if (unlikely(!flow)) {
 266                 struct dp_upcall_info upcall;
 267
 268                 upcall.cmd = OVS_PACKET_CMD_MISS;
 269                 upcall.key = &key;
 270                 upcall.userdata = NULL;
 271                 upcall.portid = p->upcall_portid;
 272                 ovs_dp_upcall(dp, skb, &upcall);
 273                 consume_skb(skb);
 274                 stats_counter = &stats->n_missed;
 275                 goto out;
 276         }
 277
 278         OVS_CB(skb)->flow = flow;
 279         OVS_CB(skb)->pkt_key = &key;
 280
 281         ovs_flow_stats_update(OVS_CB(skb)->flow, key.tp.flags, skb);
 282         ovs_execute_actions(dp, skb);
 283         stats_counter = &stats->n_hit;
 284
 285 out:
 286         /* Update datapath statistics. */
 287         u64_stats_update_begin(&stats->sync);
 288         (*stats_counter)++;
 289         stats->n_mask_hit += n_mask_hit;
 290         u64_stats_update_end(&stats->sync);
 291 }
 292
 293 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 294                   const struct dp_upcall_info *upcall_info)
 295 {
 296         struct dp_stats_percpu *stats;
 297         int err;
 298
 299         if (upcall_info->portid == 0) {
 300                 err = -ENOTCONN;
 301                 goto err;
 302         }
 303
 304         if (!skb_is_gso(skb))
 305                 err = queue_userspace_packet(dp, skb, upcall_info);
 306         else
 307                 err = queue_gso_packets(dp, skb, upcall_info);
 308         if (err)
 309                 goto err;
 310
 311         return 0;
 312
 313 err:
 314         stats = this_cpu_ptr(dp->stats_percpu);
 315
 316         u64_stats_update_begin(&stats->sync);
 317         stats->n_lost++;
 318         u64_stats_update_end(&stats->sync);
 319
 320         return err;
 321 }
 322
 323 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 324                              const struct dp_upcall_info *upcall_info)
 325 {
 326         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 327         struct dp_upcall_info later_info;
 328         struct sw_flow_key later_key;
 329         struct sk_buff *segs, *nskb;
 330         int err;
 331
 332         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 333         if (IS_ERR(segs))
 334                 return PTR_ERR(segs);
 335
 336         /* Queue all of the segments. */
 337         skb = segs;
 338         do {
 339                 err = queue_userspace_packet(dp, skb, upcall_info);
 340                 if (err)
 341                         break;
 342
 343                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 344                         /* The initial flow key extracted by ovs_flow_extract()
 345                          * in this case is for a first fragment, so we need to
 346                          * properly mark later fragments.
 347                          */
 348                         later_key = *upcall_info->key;
 349                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 350
 351                         later_info = *upcall_info;
 352                         later_info.key = &later_key;
 353                         upcall_info = &later_info;
 354                 }
 355         } while ((skb = skb->next));
 356
 357         /* Free all of the segments. */
 358         skb = segs;
 359         do {
 360                 nskb = skb->next;
 361                 if (err)
 362                         kfree_skb(skb);
 363                 else
 364                         consume_skb(skb);
 365         } while ((skb = nskb));
 366         return err;
 367 }
 368
 369 static size_t key_attr_size(void)
 370 {
 371         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 372                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 373                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
 374                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 375                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 376                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
 377                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
 378                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 379                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
 380                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 381                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 382                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 383                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 384                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 385                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 386                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 387                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 388                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 389                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 390 }
 391
 392 static size_t upcall_msg_size(const struct nlattr *userdata,
 393                               unsigned int hdrlen)
 394 {
 395         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 396                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 397                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 398
 399         /* OVS_PACKET_ATTR_USERDATA */
 400         if (userdata)
 401                 size += NLA_ALIGN(userdata->nla_len);
 402
 403         return size;
 404 }
 405
 406 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 407                                   const struct dp_upcall_info *upcall_info)
 408 {
 409         struct ovs_header *upcall;
 410         struct sk_buff *nskb = NULL;
 411         struct sk_buff *user_skb; /* to be queued to userspace */
 412         struct nlattr *nla;
 413         struct genl_info info = {
 414 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,14,0)
 415                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 416 #endif
 417                 .snd_portid = upcall_info->portid,
 418         };
 419         size_t len;
 420         unsigned int hlen;
 421         int err, dp_ifindex;
 422
 423         dp_ifindex = get_dpifindex(dp);
 424         if (!dp_ifindex)
 425                 return -ENODEV;
 426
 427         if (vlan_tx_tag_present(skb)) {
 428                 nskb = skb_clone(skb, GFP_ATOMIC);
 429                 if (!nskb)
 430                         return -ENOMEM;
 431
 432                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 433                 if (!nskb)
 434                         return -ENOMEM;
 435
 436                 vlan_set_tci(nskb, 0);
 437
 438                 skb = nskb;
 439         }
 440
 441         if (nla_attr_size(skb->len) > USHRT_MAX) {
 442                 err = -EFBIG;
 443                 goto out;
 444         }
 445
 446         /* Complete checksum if needed */
 447         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 448             (err = skb_checksum_help(skb)))
 449                 goto out;
 450
 451         /* Older versions of OVS user space enforce alignment of the last
 452          * Netlink attribute to NLA_ALIGNTO which would require extensive
 453          * padding logic. Only perform zerocopy if padding is not required.
 454          */
 455         if (dp->user_features & OVS_DP_F_UNALIGNED)
 456                 hlen = skb_zerocopy_headlen(skb);
 457         else
 458                 hlen = skb->len;
 459
 460         len = upcall_msg_size(upcall_info->userdata, hlen);
 461         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 462         if (!user_skb) {
 463                 err = -ENOMEM;
 464                 goto out;
 465         }
 466
 467         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 468                              0, upcall_info->cmd);
 469         upcall->dp_ifindex = dp_ifindex;
 470
 471         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 472         ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
 473         nla_nest_end(user_skb, nla);
 474
 475         if (upcall_info->userdata)
 476                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 477                           nla_len(upcall_info->userdata),
 478                           nla_data(upcall_info->userdata));
 479
 480         /* Only reserve room for attribute header, packet data is added
 481          * in skb_zerocopy() */
 482         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 483                 err = -ENOBUFS;
 484                 goto out;
 485         }
 486         nla->nla_len = nla_attr_size(skb->len);
 487
 488         skb_zerocopy(user_skb, skb, skb->len, hlen);
 489
 490         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 491         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 492                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 493
 494                 if (plen > 0)
 495                         memset(skb_put(user_skb, plen), 0, plen);
 496         }
 497
 498         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 499
 500         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 501 out:
 502         kfree_skb(nskb);
 503         return err;
 504 }
 505
 506 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 507 {
 508         struct ovs_header *ovs_header = info->userhdr;
 509         struct nlattr **a = info->attrs;
 510         struct sw_flow_actions *acts;
 511         struct sk_buff *packet;
 512         struct sw_flow *flow;
 513         struct datapath *dp;
 514         struct ethhdr *eth;
 515         int len;
 516         int err;
 517
 518         err = -EINVAL;
 519         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 520             !a[OVS_PACKET_ATTR_ACTIONS])
 521                 goto err;
 522
 523         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 524         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 525         err = -ENOMEM;
 526         if (!packet)
 527                 goto err;
 528         skb_reserve(packet, NET_IP_ALIGN);
 529
 530         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 531
 532         skb_reset_mac_header(packet);
 533         eth = eth_hdr(packet);
 534
 535         /* Normally, setting the skb 'protocol' field would be handled by a
 536          * call to eth_type_trans(), but it assumes there's a sending
 537          * device, which we may not have. */
 538         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 539                 packet->protocol = eth->h_proto;
 540         else
 541                 packet->protocol = htons(ETH_P_802_2);
 542
 543         /* Build an sw_flow for sending this packet. */
 544         flow = ovs_flow_alloc();
 545         err = PTR_ERR(flow);
 546         if (IS_ERR(flow))
 547                 goto err_kfree_skb;
 548
 549         err = ovs_flow_extract(packet, -1, &flow->key);
 550         if (err)
 551                 goto err_flow_free;
 552
 553         err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
 554         if (err)
 555                 goto err_flow_free;
 556         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 557         err = PTR_ERR(acts);
 558         if (IS_ERR(acts))
 559                 goto err_flow_free;
 560
 561         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 562                                    &flow->key, 0, &acts);
 563         rcu_assign_pointer(flow->sf_acts, acts);
 564         if (err)
 565                 goto err_flow_free;
 566
 567         OVS_CB(packet)->flow = flow;
 568         OVS_CB(packet)->pkt_key = &flow->key;
 569         packet->priority = flow->key.phy.priority;
 570         packet->mark = flow->key.phy.skb_mark;
 571
 572         rcu_read_lock();
 573         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 574         err = -ENODEV;
 575         if (!dp)
 576                 goto err_unlock;
 577
 578         local_bh_disable();
 579         err = ovs_execute_actions(dp, packet);
 580         local_bh_enable();
 581         rcu_read_unlock();
 582
 583         ovs_flow_free(flow, false);
 584         return err;
 585
 586 err_unlock:
 587         rcu_read_unlock();
 588 err_flow_free:
 589         ovs_flow_free(flow, false);
 590 err_kfree_skb:
 591         kfree_skb(packet);
 592 err:
 593         return err;
 594 }
 595
 596 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 597         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 598         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 599         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 600 };
 601
 602 static struct genl_ops dp_packet_genl_ops[] = {
 603         { .cmd = OVS_PACKET_CMD_EXECUTE,
 604           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 605           .policy = packet_policy,
 606           .doit = ovs_packet_cmd_execute
 607         }
 608 };
 609
 610 static struct genl_family dp_packet_genl_family = {
 611         .id = GENL_ID_GENERATE,
 612         .hdrsize = sizeof(struct ovs_header),
 613         .name = OVS_PACKET_FAMILY,
 614         .version = OVS_PACKET_VERSION,
 615         .maxattr = OVS_PACKET_ATTR_MAX,
 616         .netnsok = true,
 617         .parallel_ops = true,
 618         .ops = dp_packet_genl_ops,
 619         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 620 };
 621
 622 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
 623                          struct ovs_dp_megaflow_stats *mega_stats)
 624 {
 625         int i;
 626
 627         memset(mega_stats, 0, sizeof(*mega_stats));
 628
 629         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 630         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 631
 632         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 633
 634         for_each_possible_cpu(i) {
 635                 const struct dp_stats_percpu *percpu_stats;
 636                 struct dp_stats_percpu local_stats;
 637                 unsigned int start;
 638
 639                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 640
 641                 do {
 642                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 643                         local_stats = *percpu_stats;
 644                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 645
 646                 stats->n_hit += local_stats.n_hit;
 647                 stats->n_missed += local_stats.n_missed;
 648                 stats->n_lost += local_stats.n_lost;
 649                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 650         }
 651 }
 652
 653 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 654 {
 655         return NLMSG_ALIGN(sizeof(struct ovs_header))
 656                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 657                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 658                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 659                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 660                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 661                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 662 }
 663
 664 /* Called with ovs_mutex or RCU read lock. */
 665 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 666                                   struct sk_buff *skb, u32 portid,
 667                                   u32 seq, u32 flags, u8 cmd)
 668 {
 669         const int skb_orig_len = skb->len;
 670         struct nlattr *start;
 671         struct ovs_flow_stats stats;
 672         __be16 tcp_flags;
 673         unsigned long used;
 674         struct ovs_header *ovs_header;
 675         struct nlattr *nla;
 676         int err;
 677
 678         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 679         if (!ovs_header)
 680                 return -EMSGSIZE;
 681
 682         ovs_header->dp_ifindex = dp_ifindex;
 683
 684         /* Fill flow key. */
 685         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 686         if (!nla)
 687                 goto nla_put_failure;
 688
 689         err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
 690         if (err)
 691                 goto error;
 692         nla_nest_end(skb, nla);
 693
 694         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 695         if (!nla)
 696                 goto nla_put_failure;
 697
 698         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 699         if (err)
 700                 goto error;
 701
 702         nla_nest_end(skb, nla);
 703
 704         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 705
 706         if (used &&
 707             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 708                 goto nla_put_failure;
 709
 710         if (stats.n_packets &&
 711             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 712                 goto nla_put_failure;
 713
 714         if ((u8)ntohs(tcp_flags) &&
 715              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 716                 goto nla_put_failure;
 717
 718         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 719          * this is the first flow to be dumped into 'skb'.  This is unusual for
 720          * Netlink but individual action lists can be longer than
 721          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 722          * The userspace caller can always fetch the actions separately if it
 723          * really wants them.  (Most userspace callers in fact don't care.)
 724          *
 725          * This can only fail for dump operations because the skb is always
 726          * properly sized for single flows.
 727          */
 728         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 729         if (start) {
 730                 const struct sw_flow_actions *sf_acts;
 731
 732                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 733                 err = ovs_nla_put_actions(sf_acts->actions,
 734                                           sf_acts->actions_len, skb);
 735
 736                 if (!err)
 737                         nla_nest_end(skb, start);
 738                 else {
 739                         if (skb_orig_len)
 740                                 goto error;
 741
 742                         nla_nest_cancel(skb, start);
 743                 }
 744         } else if (skb_orig_len)
 745                 goto nla_put_failure;
 746
 747         return genlmsg_end(skb, ovs_header);
 748
 749 nla_put_failure:
 750         err = -EMSGSIZE;
 751 error:
 752         genlmsg_cancel(skb, ovs_header);
 753         return err;
 754 }
 755
 756 /* May not be called with RCU read lock. */
 757 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 758                                                struct genl_info *info,
 759                                                bool always)
 760 {
 761         struct sk_buff *skb;
 762
 763         if (!always && !ovs_must_notify(info, &ovs_dp_flow_multicast_group))
 764                 return NULL;
 765
 766         skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
 767
 768         if (!skb)
 769                 return ERR_PTR(-ENOMEM);
 770
 771         return skb;
 772 }
 773
 774 /* Called with ovs_mutex. */
 775 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 776                                                int dp_ifindex,
 777                                                struct genl_info *info, u8 cmd,
 778                                                bool always)
 779 {
 780         struct sk_buff *skb;
 781         int retval;
 782
 783         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
 784                                       always);
 785         if (!skb || IS_ERR(skb))
 786                 return skb;
 787
 788         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 789                                         info->snd_portid, info->snd_seq, 0,
 790                                         cmd);
 791         BUG_ON(retval < 0);
 792         return skb;
 793 }
 794
 795 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 796 {
 797         struct nlattr **a = info->attrs;
 798         struct ovs_header *ovs_header = info->userhdr;
 799         struct sw_flow *flow, *new_flow;
 800         struct sw_flow_mask mask;
 801         struct sk_buff *reply;
 802         struct datapath *dp;
 803         struct sw_flow_actions *acts;
 804         struct sw_flow_match match;
 805         int error;
 806
 807         /* Must have key and actions. */
 808         error = -EINVAL;
 809         if (!a[OVS_FLOW_ATTR_KEY])
 810                 goto error;
 811         if (!a[OVS_FLOW_ATTR_ACTIONS])
 812                 goto error;
 813
 814         /* Most of the time we need to allocate a new flow, do it before
 815          * locking. */
 816         new_flow = ovs_flow_alloc();
 817         if (IS_ERR(new_flow)) {
 818                 error = PTR_ERR(new_flow);
 819                 goto error;
 820         }
 821
 822         /* Extract key. */
 823         ovs_match_init(&match, &new_flow->unmasked_key, &mask);
 824         error = ovs_nla_get_match(&match,
 825                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 826         if (error)
 827                 goto err_kfree_flow;
 828
 829         ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
 830
 831         /* Validate actions. */
 832         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 833         error = PTR_ERR(acts);
 834         if (IS_ERR(acts))
 835                 goto err_kfree_flow;
 836
 837         error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
 838                                      0, &acts);
 839         if (error) {
 840                 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 841                 goto err_kfree_acts;
 842         }
 843
 844         reply = ovs_flow_cmd_alloc_info(acts, info, false);
 845         if (IS_ERR(reply)) {
 846                 error = PTR_ERR(reply);
 847                 goto err_kfree_acts;
 848         }
 849
 850         ovs_lock();
 851         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 852         if (unlikely(!dp)) {
 853                 error = -ENODEV;
 854                 goto err_unlock_ovs;
 855         }
 856         /* Check if this is a duplicate flow */
 857         flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
 858         if (likely(!flow)) {
 859                 rcu_assign_pointer(new_flow->sf_acts, acts);
 860
 861                 /* Put flow in bucket. */
 862                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 863                 if (unlikely(error)) {
 864                         acts = NULL;
 865                         goto err_unlock_ovs;
 866                 }
 867
 868                 if (unlikely(reply)) {
 869                         error = ovs_flow_cmd_fill_info(new_flow,
 870                                                        ovs_header->dp_ifindex,
 871                                                        reply, info->snd_portid,
 872                                                        info->snd_seq, 0,
 873                                                        OVS_FLOW_CMD_NEW);
 874                         BUG_ON(error < 0);
 875                 }
 876                 ovs_unlock();
 877         } else {
 878                 struct sw_flow_actions *old_acts;
 879
 880                 /* Bail out if we're not allowed to modify an existing flow.
 881                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 882                  * because Generic Netlink treats the latter as a dump
 883                  * request.  We also accept NLM_F_EXCL in case that bug ever
 884                  * gets fixed.
 885                  */
 886                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
 887                                                          | NLM_F_EXCL))) {
 888                         error = -EEXIST;
 889                         goto err_unlock_ovs;
 890                 }
 891                 /* The unmasked key has to be the same for flow updates. */
 892                 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
 893                         error = -EEXIST;
 894                         goto err_unlock_ovs;
 895                 }
 896                 /* Update actions. */
 897                 old_acts = ovsl_dereference(flow->sf_acts);
 898                 rcu_assign_pointer(flow->sf_acts, acts);
 899
 900                 if (unlikely(reply)) {
 901                         error = ovs_flow_cmd_fill_info(flow,
 902                                                        ovs_header->dp_ifindex,
 903                                                        reply, info->snd_portid,
 904                                                        info->snd_seq, 0,
 905                                                        OVS_FLOW_CMD_NEW);
 906                         BUG_ON(error < 0);
 907                 }
 908                 ovs_unlock();
 909
 910                 ovs_nla_free_flow_actions(old_acts);
 911                 ovs_flow_free(new_flow, false);
 912         }
 913
 914         if (reply)
 915                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
 916         return 0;
 917
 918 err_unlock_ovs:
 919         ovs_unlock();
 920         kfree_skb(reply);
 921 err_kfree_acts:
 922         kfree(acts);
 923 err_kfree_flow:
 924         ovs_flow_free(new_flow, false);
 925 error:
 926         return error;
 927 }
 928
 929 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
 930 {
 931         struct nlattr **a = info->attrs;
 932         struct ovs_header *ovs_header = info->userhdr;
 933         struct sw_flow_key key, masked_key;
 934         struct sw_flow *flow;
 935         struct sw_flow_mask mask;
 936         struct sk_buff *reply = NULL;
 937         struct datapath *dp;
 938         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
 939         struct sw_flow_match match;
 940         int error;
 941
 942         /* Extract key. */
 943         error = -EINVAL;
 944         if (!a[OVS_FLOW_ATTR_KEY])
 945                 goto error;
 946
 947         ovs_match_init(&match, &key, &mask);
 948         error = ovs_nla_get_match(&match,
 949                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 950         if (error)
 951                 goto error;
 952
 953         /* Validate actions. */
 954         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 955                 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 956                 error = PTR_ERR(acts);
 957                 if (IS_ERR(acts))
 958                         goto error;
 959
 960                 ovs_flow_mask_key(&masked_key, &key, &mask);
 961                 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
 962                                              &masked_key, 0, &acts);
 963                 if (error) {
 964                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 965                         goto err_kfree_acts;
 966                 }
 967         }
 968
 969         /* Can allocate before locking if have acts. */
 970         if (acts) {
 971                 reply = ovs_flow_cmd_alloc_info(acts, info, false);
 972                 if (IS_ERR(reply)) {
 973                         error = PTR_ERR(reply);
 974                         goto err_kfree_acts;
 975                 }
 976         }
 977
 978         ovs_lock();
 979         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 980         if (unlikely(!dp)) {
 981                 error = -ENODEV;
 982                 goto err_unlock_ovs;
 983         }
 984         /* Check that the flow exists. */
 985         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 986         if (unlikely(!flow)) {
 987                 error = -ENOENT;
 988                 goto err_unlock_ovs;
 989         }
 990         /* The unmasked key has to be the same for flow updates. */
 991         if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
 992                 error = -EEXIST;
 993                 goto err_unlock_ovs;
 994         }
 995         /* Update actions, if present. */
 996         if (likely(acts)) {
 997                 old_acts = ovsl_dereference(flow->sf_acts);
 998                 rcu_assign_pointer(flow->sf_acts, acts);
 999
1000                 if (unlikely(reply)) {
1001                         error = ovs_flow_cmd_fill_info(flow,
1002                                                        ovs_header->dp_ifindex,
1003                                                        reply, info->snd_portid,
1004                                                        info->snd_seq, 0,
1005                                                        OVS_FLOW_CMD_NEW);
1006                         BUG_ON(error < 0);
1007                 }
1008         } else {
1009                 /* Could not alloc without acts before locking. */
1010                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1011                                                 info, OVS_FLOW_CMD_NEW, false);
1012                 if (unlikely(IS_ERR(reply))) {
1013                         error = PTR_ERR(reply);
1014                         goto err_unlock_ovs;
1015                 }
1016         }
1017
1018         /* Clear stats. */
1019         if (a[OVS_FLOW_ATTR_CLEAR])
1020                 ovs_flow_stats_clear(flow);
1021         ovs_unlock();
1022
1023         if (reply)
1024                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1025         if (old_acts)
1026                 ovs_nla_free_flow_actions(old_acts);
1027         return 0;
1028
1029 err_unlock_ovs:
1030         ovs_unlock();
1031         kfree_skb(reply);
1032 err_kfree_acts:
1033         kfree(acts);
1034 error:
1035         return error;
1036 }
1037
1038 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1039 {
1040         struct nlattr **a = info->attrs;
1041         struct ovs_header *ovs_header = info->userhdr;
1042         struct sw_flow_key key;
1043         struct sk_buff *reply;
1044         struct sw_flow *flow;
1045         struct datapath *dp;
1046         struct sw_flow_match match;
1047         int err;
1048
1049         if (!a[OVS_FLOW_ATTR_KEY]) {
1050                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1051                 return -EINVAL;
1052         }
1053
1054         ovs_match_init(&match, &key, NULL);
1055         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1056         if (err)
1057                 return err;
1058
1059         ovs_lock();
1060         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1061         if (!dp) {
1062                 err = -ENODEV;
1063                 goto unlock;
1064         }
1065
1066         flow = ovs_flow_tbl_lookup(&dp->table, &key);
1067         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
1068                 err = -ENOENT;
1069                 goto unlock;
1070         }
1071
1072         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1073                                         OVS_FLOW_CMD_NEW, true);
1074         if (IS_ERR(reply)) {
1075                 err = PTR_ERR(reply);
1076                 goto unlock;
1077         }
1078
1079         ovs_unlock();
1080         return genlmsg_reply(reply, info);
1081 unlock:
1082         ovs_unlock();
1083         return err;
1084 }
1085
1086 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1087 {
1088         struct nlattr **a = info->attrs;
1089         struct ovs_header *ovs_header = info->userhdr;
1090         struct sw_flow_key key;
1091         struct sk_buff *reply;
1092         struct sw_flow *flow;
1093         struct datapath *dp;
1094         struct sw_flow_match match;
1095         int err;
1096
1097         if (likely(a[OVS_FLOW_ATTR_KEY])) {
1098                 ovs_match_init(&match, &key, NULL);
1099                 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1100                 if (unlikely(err))
1101                         return err;
1102         }
1103
1104         ovs_lock();
1105         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1106         if (unlikely(!dp)) {
1107                 err = -ENODEV;
1108                 goto unlock;
1109         }
1110         if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1111                 err = ovs_flow_tbl_flush(&dp->table);
1112                 goto unlock;
1113         }
1114         flow = ovs_flow_tbl_lookup(&dp->table, &key);
1115         if (unlikely(!flow || !ovs_flow_cmp_unmasked_key(flow, &match))) {
1116                 err = -ENOENT;
1117                 goto unlock;
1118         }
1119
1120         ovs_flow_tbl_remove(&dp->table, flow);
1121         ovs_unlock();
1122
1123         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *)flow->sf_acts,
1124                                         info, false);
1125
1126         if (likely(reply)) {
1127                 if (likely(!IS_ERR(reply))) {
1128                         rcu_read_lock(); /* Keep RCU checker happy. */
1129                         err = ovs_flow_cmd_fill_info(flow,
1130                                                      ovs_header->dp_ifindex,
1131                                                      reply, info->snd_portid,
1132                                                      info->snd_seq, 0,
1133                                                      OVS_FLOW_CMD_DEL);
1134                         rcu_read_unlock();
1135                         BUG_ON(err < 0);
1136                         ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1137                 } else {
1138                         genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1139                                      GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1140
1141                 }
1142         }
1143
1144         ovs_flow_free(flow, true);
1145         return 0;
1146 unlock:
1147         ovs_unlock();
1148         return err;
1149 }
1150
1151 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1152 {
1153         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1154         struct table_instance *ti;
1155         struct datapath *dp;
1156
1157         rcu_read_lock();
1158         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1159         if (!dp) {
1160                 rcu_read_unlock();
1161                 return -ENODEV;
1162         }
1163
1164         ti = rcu_dereference(dp->table.ti);
1165         for (;;) {
1166                 struct sw_flow *flow;
1167                 u32 bucket, obj;
1168
1169                 bucket = cb->args[0];
1170                 obj = cb->args[1];
1171                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1172                 if (!flow)
1173                         break;
1174
1175                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1176                                            NETLINK_CB(cb->skb).portid,
1177                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1178                                            OVS_FLOW_CMD_NEW) < 0)
1179                         break;
1180
1181                 cb->args[0] = bucket;
1182                 cb->args[1] = obj;
1183         }
1184         rcu_read_unlock();
1185         return skb->len;
1186 }
1187
1188 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1189         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1190         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1191         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1192 };
1193
1194 static struct genl_ops dp_flow_genl_ops[] = {
1195         { .cmd = OVS_FLOW_CMD_NEW,
1196           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1197           .policy = flow_policy,
1198           .doit = ovs_flow_cmd_new
1199         },
1200         { .cmd = OVS_FLOW_CMD_DEL,
1201           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1202           .policy = flow_policy,
1203           .doit = ovs_flow_cmd_del
1204         },
1205         { .cmd = OVS_FLOW_CMD_GET,
1206           .flags = 0,               /* OK for unprivileged users. */
1207           .policy = flow_policy,
1208           .doit = ovs_flow_cmd_get,
1209           .dumpit = ovs_flow_cmd_dump
1210         },
1211         { .cmd = OVS_FLOW_CMD_SET,
1212           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1213           .policy = flow_policy,
1214           .doit = ovs_flow_cmd_set,
1215         },
1216 };
1217
1218 static struct genl_family dp_flow_genl_family = {
1219         .id = GENL_ID_GENERATE,
1220         .hdrsize = sizeof(struct ovs_header),
1221         .name = OVS_FLOW_FAMILY,
1222         .version = OVS_FLOW_VERSION,
1223         .maxattr = OVS_FLOW_ATTR_MAX,
1224         .netnsok = true,
1225         .parallel_ops = true,
1226         .ops = dp_flow_genl_ops,
1227         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1228         .mcgrps = &ovs_dp_flow_multicast_group,
1229         .n_mcgrps = 1,
1230 };
1231
1232 static size_t ovs_dp_cmd_msg_size(void)
1233 {
1234         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1235
1236         msgsize += nla_total_size(IFNAMSIZ);
1237         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1238         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1239         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1240
1241         return msgsize;
1242 }
1243
1244 /* Called with ovs_mutex or RCU read lock. */
1245 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1246                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1247 {
1248         struct ovs_header *ovs_header;
1249         struct ovs_dp_stats dp_stats;
1250         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1251         int err;
1252
1253         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1254                                    flags, cmd);
1255         if (!ovs_header)
1256                 goto error;
1257
1258         ovs_header->dp_ifindex = get_dpifindex(dp);
1259
1260         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1261         if (err)
1262                 goto nla_put_failure;
1263
1264         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1265         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1266                         &dp_stats))
1267                 goto nla_put_failure;
1268
1269         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1270                         sizeof(struct ovs_dp_megaflow_stats),
1271                         &dp_megaflow_stats))
1272                 goto nla_put_failure;
1273
1274         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1275                 goto nla_put_failure;
1276
1277         return genlmsg_end(skb, ovs_header);
1278
1279 nla_put_failure:
1280         genlmsg_cancel(skb, ovs_header);
1281 error:
1282         return -EMSGSIZE;
1283 }
1284
1285 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1286 {
1287         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1288 }
1289
1290 /* Called with rcu_read_lock or ovs_mutex. */
1291 static struct datapath *lookup_datapath(struct net *net,
1292                                         struct ovs_header *ovs_header,
1293                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1294 {
1295         struct datapath *dp;
1296
1297         if (!a[OVS_DP_ATTR_NAME])
1298                 dp = get_dp(net, ovs_header->dp_ifindex);
1299         else {
1300                 struct vport *vport;
1301
1302                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1303                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1304         }
1305         return dp ? dp : ERR_PTR(-ENODEV);
1306 }
1307
1308 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1309 {
1310         struct datapath *dp;
1311
1312         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1313         if (IS_ERR(dp))
1314                 return;
1315
1316         WARN(dp->user_features, "Dropping previously announced user features\n");
1317         dp->user_features = 0;
1318 }
1319
1320 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1321 {
1322         if (a[OVS_DP_ATTR_USER_FEATURES])
1323                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1324 }
1325
1326 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1327 {
1328         struct nlattr **a = info->attrs;
1329         struct vport_parms parms;
1330         struct sk_buff *reply;
1331         struct datapath *dp;
1332         struct vport *vport;
1333         struct ovs_net *ovs_net;
1334         int err, i;
1335
1336         err = -EINVAL;
1337         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1338                 goto err;
1339
1340         reply = ovs_dp_cmd_alloc_info(info);
1341         if (!reply)
1342                 return -ENOMEM;
1343
1344         err = -ENOMEM;
1345         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1346         if (dp == NULL)
1347                 goto err_free_reply;
1348
1349         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1350
1351         /* Allocate table. */
1352         err = ovs_flow_tbl_init(&dp->table);
1353         if (err)
1354                 goto err_free_dp;
1355
1356         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1357         if (!dp->stats_percpu) {
1358                 err = -ENOMEM;
1359                 goto err_destroy_table;
1360         }
1361
1362         for_each_possible_cpu(i) {
1363                 struct dp_stats_percpu *dpath_stats;
1364                 dpath_stats = per_cpu_ptr(dp->stats_percpu, i);
1365                 u64_stats_init(&dpath_stats->sync);
1366         }
1367
1368         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1369                             GFP_KERNEL);
1370         if (!dp->ports) {
1371                 err = -ENOMEM;
1372                 goto err_destroy_percpu;
1373         }
1374
1375         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1376                 INIT_HLIST_HEAD(&dp->ports[i]);
1377
1378         /* Set up our datapath device. */
1379         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1380         parms.type = OVS_VPORT_TYPE_INTERNAL;
1381         parms.options = NULL;
1382         parms.dp = dp;
1383         parms.port_no = OVSP_LOCAL;
1384         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1385
1386         ovs_dp_change(dp, a);
1387
1388         /* So far only local changes have been made, now need the lock. */
1389         ovs_lock();
1390
1391         vport = new_vport(&parms);
1392         if (IS_ERR(vport)) {
1393                 err = PTR_ERR(vport);
1394                 if (err == -EBUSY)
1395                         err = -EEXIST;
1396
1397                 if (err == -EEXIST) {
1398                         /* An outdated user space instance that does not understand
1399                          * the concept of user_features has attempted to create a new
1400                          * datapath and is likely to reuse it. Drop all user features.
1401                          */
1402                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1403                                 ovs_dp_reset_user_features(skb, info);
1404                 }
1405
1406                 goto err_destroy_ports_array;
1407         }
1408
1409         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1410                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1411         BUG_ON(err < 0);
1412
1413         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1414         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1415
1416         ovs_unlock();
1417
1418         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1419         return 0;
1420
1421 err_destroy_ports_array:
1422         ovs_unlock();
1423         kfree(dp->ports);
1424 err_destroy_percpu:
1425         free_percpu(dp->stats_percpu);
1426 err_destroy_table:
1427         ovs_flow_tbl_destroy(&dp->table, false);
1428 err_free_dp:
1429         release_net(ovs_dp_get_net(dp));
1430         kfree(dp);
1431 err_free_reply:
1432         kfree_skb(reply);
1433 err:
1434         return err;
1435 }
1436
1437 /* Called with ovs_mutex. */
1438 static void __dp_destroy(struct datapath *dp)
1439 {
1440         int i;
1441
1442         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1443                 struct vport *vport;
1444                 struct hlist_node *n;
1445
1446                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1447                         if (vport->port_no != OVSP_LOCAL)
1448                                 ovs_dp_detach_port(vport);
1449         }
1450
1451         list_del_rcu(&dp->list_node);
1452
1453         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1454          * all ports in datapath are destroyed first before freeing datapath.
1455          */
1456         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1457
1458         /* RCU destroy the flow table */
1459         ovs_flow_tbl_destroy(&dp->table, true);
1460
1461         call_rcu(&dp->rcu, destroy_dp_rcu);
1462 }
1463
1464 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1465 {
1466         struct sk_buff *reply;
1467         struct datapath *dp;
1468         int err;
1469
1470         reply = ovs_dp_cmd_alloc_info(info);
1471         if (!reply)
1472                 return -ENOMEM;
1473
1474         ovs_lock();
1475         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1476         err = PTR_ERR(dp);
1477         if (IS_ERR(dp))
1478                 goto err_unlock_free;
1479
1480         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1481                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1482         BUG_ON(err < 0);
1483
1484         __dp_destroy(dp);
1485
1486         ovs_unlock();
1487         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1488         return 0;
1489
1490 err_unlock_free:
1491         ovs_unlock();
1492         kfree_skb(reply);
1493         return err;
1494 }
1495
1496 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1497 {
1498         struct sk_buff *reply;
1499         struct datapath *dp;
1500         int err;
1501
1502         reply = ovs_dp_cmd_alloc_info(info);
1503         if (!reply)
1504                 return -ENOMEM;
1505
1506         ovs_lock();
1507         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1508         err = PTR_ERR(dp);
1509         if (IS_ERR(dp))
1510                 goto err_unlock_free;
1511
1512         ovs_dp_change(dp, info->attrs);
1513
1514         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1515                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1516         BUG_ON(err < 0);
1517
1518         ovs_unlock();
1519         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1520         return 0;
1521
1522 err_unlock_free:
1523         ovs_unlock();
1524         kfree_skb(reply);
1525         return err;
1526 }
1527
1528 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1529 {
1530         struct sk_buff *reply;
1531         struct datapath *dp;
1532         int err;
1533
1534         reply = ovs_dp_cmd_alloc_info(info);
1535         if (!reply)
1536                 return -ENOMEM;
1537
1538         rcu_read_lock();
1539         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1540         if (IS_ERR(dp)) {
1541                 err = PTR_ERR(dp);
1542                 goto err_unlock_free;
1543         }
1544         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1545                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1546         BUG_ON(err < 0);
1547         rcu_read_unlock();
1548
1549         return genlmsg_reply(reply, info);
1550
1551 err_unlock_free:
1552         rcu_read_unlock();
1553         kfree_skb(reply);
1554         return err;
1555 }
1556
1557 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1558 {
1559         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1560         struct datapath *dp;
1561         int skip = cb->args[0];
1562         int i = 0;
1563
1564         rcu_read_lock();
1565         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1566                 if (i >= skip &&
1567                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1568                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1569                                          OVS_DP_CMD_NEW) < 0)
1570                         break;
1571                 i++;
1572         }
1573         rcu_read_unlock();
1574
1575         cb->args[0] = i;
1576
1577         return skb->len;
1578 }
1579
1580 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1581         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1582         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1583         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1584 };
1585
1586 static struct genl_ops dp_datapath_genl_ops[] = {
1587         { .cmd = OVS_DP_CMD_NEW,
1588           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1589           .policy = datapath_policy,
1590           .doit = ovs_dp_cmd_new
1591         },
1592         { .cmd = OVS_DP_CMD_DEL,
1593           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1594           .policy = datapath_policy,
1595           .doit = ovs_dp_cmd_del
1596         },
1597         { .cmd = OVS_DP_CMD_GET,
1598           .flags = 0,               /* OK for unprivileged users. */
1599           .policy = datapath_policy,
1600           .doit = ovs_dp_cmd_get,
1601           .dumpit = ovs_dp_cmd_dump
1602         },
1603         { .cmd = OVS_DP_CMD_SET,
1604           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1605           .policy = datapath_policy,
1606           .doit = ovs_dp_cmd_set,
1607         },
1608 };
1609
1610 static struct genl_family dp_datapath_genl_family = {
1611         .id = GENL_ID_GENERATE,
1612         .hdrsize = sizeof(struct ovs_header),
1613         .name = OVS_DATAPATH_FAMILY,
1614         .version = OVS_DATAPATH_VERSION,
1615         .maxattr = OVS_DP_ATTR_MAX,
1616         .netnsok = true,
1617         .parallel_ops = true,
1618         .ops = dp_datapath_genl_ops,
1619         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1620         .mcgrps = &ovs_dp_datapath_multicast_group,
1621         .n_mcgrps = 1,
1622 };
1623
1624 /* Called with ovs_mutex or RCU read lock. */
1625 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1626                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1627 {
1628         struct ovs_header *ovs_header;
1629         struct ovs_vport_stats vport_stats;
1630         int err;
1631
1632         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1633                                  flags, cmd);
1634         if (!ovs_header)
1635                 return -EMSGSIZE;
1636
1637         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1638
1639         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1640             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1641             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1642             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1643                 goto nla_put_failure;
1644
1645         ovs_vport_get_stats(vport, &vport_stats);
1646         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1647                     &vport_stats))
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_portid = nla_get_u32(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 = 0;
1779         if (a[OVS_VPORT_ATTR_STATS])
1780                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1781
1782         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1783                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1784         BUG_ON(err < 0);
1785         ovs_unlock();
1786
1787         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1788         return 0;
1789
1790 exit_unlock_free:
1791         ovs_unlock();
1792         kfree_skb(reply);
1793         return err;
1794 }
1795
1796 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1797 {
1798         struct nlattr **a = info->attrs;
1799         struct sk_buff *reply;
1800         struct vport *vport;
1801         int err;
1802
1803         reply = ovs_vport_cmd_alloc_info();
1804         if (!reply)
1805                 return -ENOMEM;
1806
1807         ovs_lock();
1808         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1809         err = PTR_ERR(vport);
1810         if (IS_ERR(vport))
1811                 goto exit_unlock_free;
1812
1813         if (a[OVS_VPORT_ATTR_TYPE] &&
1814             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1815                 err = -EINVAL;
1816                 goto exit_unlock_free;
1817         }
1818
1819         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1820                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1821                 if (err)
1822                         goto exit_unlock_free;
1823         }
1824
1825         if (a[OVS_VPORT_ATTR_STATS])
1826                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1827
1828         if (a[OVS_VPORT_ATTR_UPCALL_PID])
1829                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1830
1831         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1832                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1833         BUG_ON(err < 0);
1834         ovs_unlock();
1835
1836         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1837         return 0;
1838
1839 exit_unlock_free:
1840         ovs_unlock();
1841         kfree_skb(reply);
1842         return err;
1843 }
1844
1845 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1846 {
1847         struct nlattr **a = info->attrs;
1848         struct sk_buff *reply;
1849         struct vport *vport;
1850         int err;
1851
1852         reply = ovs_vport_cmd_alloc_info();
1853         if (!reply)
1854                 return -ENOMEM;
1855
1856         ovs_lock();
1857         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1858         err = PTR_ERR(vport);
1859         if (IS_ERR(vport))
1860                 goto exit_unlock_free;
1861
1862         if (vport->port_no == OVSP_LOCAL) {
1863                 err = -EINVAL;
1864                 goto exit_unlock_free;
1865         }
1866
1867         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1868                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1869         BUG_ON(err < 0);
1870         ovs_dp_detach_port(vport);
1871         ovs_unlock();
1872
1873         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1874         return 0;
1875
1876 exit_unlock_free:
1877         ovs_unlock();
1878         kfree_skb(reply);
1879         return err;
1880 }
1881
1882 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1883 {
1884         struct nlattr **a = info->attrs;
1885         struct ovs_header *ovs_header = info->userhdr;
1886         struct sk_buff *reply;
1887         struct vport *vport;
1888         int err;
1889
1890         reply = ovs_vport_cmd_alloc_info();
1891         if (!reply)
1892                 return -ENOMEM;
1893
1894         rcu_read_lock();
1895         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1896         err = PTR_ERR(vport);
1897         if (IS_ERR(vport))
1898                 goto exit_unlock_free;
1899         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1900                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1901         BUG_ON(err < 0);
1902         rcu_read_unlock();
1903
1904         return genlmsg_reply(reply, info);
1905
1906 exit_unlock_free:
1907         rcu_read_unlock();
1908         kfree_skb(reply);
1909         return err;
1910 }
1911
1912 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1913 {
1914         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1915         struct datapath *dp;
1916         int bucket = cb->args[0], skip = cb->args[1];
1917         int i, j = 0;
1918
1919         rcu_read_lock();
1920         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1921         if (!dp) {
1922                 rcu_read_unlock();
1923                 return -ENODEV;
1924         }
1925         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1926                 struct vport *vport;
1927
1928                 j = 0;
1929                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1930                         if (j >= skip &&
1931                             ovs_vport_cmd_fill_info(vport, skb,
1932                                                     NETLINK_CB(cb->skb).portid,
1933                                                     cb->nlh->nlmsg_seq,
1934                                                     NLM_F_MULTI,
1935                                                     OVS_VPORT_CMD_NEW) < 0)
1936                                 goto out;
1937
1938                         j++;
1939                 }
1940                 skip = 0;
1941         }
1942 out:
1943         rcu_read_unlock();
1944
1945         cb->args[0] = i;
1946         cb->args[1] = j;
1947
1948         return skb->len;
1949 }
1950
1951 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1952         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1953         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1954         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1955         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1956         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1957         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1958 };
1959
1960 static struct genl_ops dp_vport_genl_ops[] = {
1961         { .cmd = OVS_VPORT_CMD_NEW,
1962           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1963           .policy = vport_policy,
1964           .doit = ovs_vport_cmd_new
1965         },
1966         { .cmd = OVS_VPORT_CMD_DEL,
1967           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1968           .policy = vport_policy,
1969           .doit = ovs_vport_cmd_del
1970         },
1971         { .cmd = OVS_VPORT_CMD_GET,
1972           .flags = 0,               /* OK for unprivileged users. */
1973           .policy = vport_policy,
1974           .doit = ovs_vport_cmd_get,
1975           .dumpit = ovs_vport_cmd_dump
1976         },
1977         { .cmd = OVS_VPORT_CMD_SET,
1978           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1979           .policy = vport_policy,
1980           .doit = ovs_vport_cmd_set,
1981         },
1982 };
1983
1984 struct genl_family dp_vport_genl_family = {
1985         .id = GENL_ID_GENERATE,
1986         .hdrsize = sizeof(struct ovs_header),
1987         .name = OVS_VPORT_FAMILY,
1988         .version = OVS_VPORT_VERSION,
1989         .maxattr = OVS_VPORT_ATTR_MAX,
1990         .netnsok = true,
1991         .parallel_ops = true,
1992         .ops = dp_vport_genl_ops,
1993         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
1994         .mcgrps = &ovs_dp_vport_multicast_group,
1995         .n_mcgrps = 1,
1996 };
1997
1998 static struct genl_family *dp_genl_families[] = {
1999         &dp_datapath_genl_family,
2000         &dp_vport_genl_family,
2001         &dp_flow_genl_family,
2002         &dp_packet_genl_family,
2003 };
2004
2005 static void dp_unregister_genl(int n_families)
2006 {
2007         int i;
2008
2009         for (i = 0; i < n_families; i++)
2010                 genl_unregister_family(dp_genl_families[i]);
2011 }
2012
2013 static int dp_register_genl(void)
2014 {
2015         int err;
2016         int i;
2017
2018         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2019
2020                 err = genl_register_family(dp_genl_families[i]);
2021                 if (err)
2022                         goto error;
2023         }
2024
2025         return 0;
2026
2027 error:
2028         dp_unregister_genl(i);
2029         return err;
2030 }
2031
2032 static int __net_init ovs_init_net(struct net *net)
2033 {
2034         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2035
2036         INIT_LIST_HEAD(&ovs_net->dps);
2037         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2038         return 0;
2039 }
2040
2041 static void __net_exit ovs_exit_net(struct net *net)
2042 {
2043         struct datapath *dp, *dp_next;
2044         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2045
2046         ovs_lock();
2047         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2048                 __dp_destroy(dp);
2049         ovs_unlock();
2050
2051         cancel_work_sync(&ovs_net->dp_notify_work);
2052 }
2053
2054 static struct pernet_operations ovs_net_ops = {
2055         .init = ovs_init_net,
2056         .exit = ovs_exit_net,
2057         .id   = &ovs_net_id,
2058         .size = sizeof(struct ovs_net),
2059 };
2060
2061 DEFINE_COMPAT_PNET_REG_FUNC(device);
2062
2063 static int __init dp_init(void)
2064 {
2065         int err;
2066
2067         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2068
2069         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2070                 VERSION);
2071
2072         err = ovs_flow_init();
2073         if (err)
2074                 goto error;
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:
2103         return err;
2104 }
2105
2106 static void dp_cleanup(void)
2107 {
2108         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2109         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2110         unregister_pernet_device(&ovs_net_ops);
2111         rcu_barrier();
2112         ovs_vport_exit();
2113         ovs_flow_exit();
2114 }
2115
2116 module_init(dp_init);
2117 module_exit(dp_cleanup);
2118
2119 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2120 MODULE_LICENSE("GPL");
2121 MODULE_VERSION(VERSION);