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_family *family, struct genl_info *info,
  82                             unsigned int group)
  83 {
  84         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  85                genl_has_listeners(family, genl_info_net(info)->genl_sock,
  86                                   group);
  87 }
  88
  89 static void ovs_notify(struct genl_family *family,
  90                        struct sk_buff *skb, struct genl_info *info)
  91 {
  92         genl_notify(family, skb, genl_info_net(info), info->snd_portid,
  93                     0, info->nlhdr, GFP_KERNEL);
  94 }
  95
  96 /**
  97  * DOC: Locking:
  98  *
  99  * All writes e.g. Writes to device state (add/remove datapath, port, set
 100  * operations on vports, etc.), Writes to other state (flow table
 101  * modifications, set miscellaneous datapath parameters, etc.) are protected
 102  * by ovs_lock.
 103  *
 104  * Reads are protected by RCU.
 105  *
 106  * There are a few special cases (mostly stats) that have their own
 107  * synchronization but they nest under all of above and don't interact with
 108  * each other.
 109  *
 110  * The RTNL lock nests inside ovs_mutex.
 111  */
 112
 113 static DEFINE_MUTEX(ovs_mutex);
 114
 115 void ovs_lock(void)
 116 {
 117         mutex_lock(&ovs_mutex);
 118 }
 119
 120 void ovs_unlock(void)
 121 {
 122         mutex_unlock(&ovs_mutex);
 123 }
 124
 125 #ifdef CONFIG_LOCKDEP
 126 int lockdep_ovsl_is_held(void)
 127 {
 128         if (debug_locks)
 129                 return lockdep_is_held(&ovs_mutex);
 130         else
 131                 return 1;
 132 }
 133 #endif
 134
 135 static struct vport *new_vport(const struct vport_parms *);
 136 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 137                              const struct dp_upcall_info *);
 138 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 139                                   const struct dp_upcall_info *);
 140
 141 /* Must be called with rcu_read_lock or ovs_mutex. */
 142 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 143 {
 144         struct datapath *dp = NULL;
 145         struct net_device *dev;
 146
 147         rcu_read_lock();
 148         dev = dev_get_by_index_rcu(net, dp_ifindex);
 149         if (dev) {
 150                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 151                 if (vport)
 152                         dp = vport->dp;
 153         }
 154         rcu_read_unlock();
 155
 156         return dp;
 157 }
 158
 159 /* Must be called with rcu_read_lock or ovs_mutex. */
 160 const char *ovs_dp_name(const struct datapath *dp)
 161 {
 162         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 163         return vport->ops->get_name(vport);
 164 }
 165
 166 static int get_dpifindex(struct datapath *dp)
 167 {
 168         struct vport *local;
 169         int ifindex;
 170
 171         rcu_read_lock();
 172
 173         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 174         if (local)
 175                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 176         else
 177                 ifindex = 0;
 178
 179         rcu_read_unlock();
 180
 181         return ifindex;
 182 }
 183
 184 static void destroy_dp_rcu(struct rcu_head *rcu)
 185 {
 186         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 187
 188         free_percpu(dp->stats_percpu);
 189         release_net(ovs_dp_get_net(dp));
 190         kfree(dp->ports);
 191         kfree(dp);
 192 }
 193
 194 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 195                                             u16 port_no)
 196 {
 197         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 198 }
 199
 200 /* Called with ovs_mutex or RCU read lock. */
 201 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 202 {
 203         struct vport *vport;
 204         struct hlist_head *head;
 205
 206         head = vport_hash_bucket(dp, port_no);
 207         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 208                 if (vport->port_no == port_no)
 209                         return vport;
 210         }
 211         return NULL;
 212 }
 213
 214 /* Called with ovs_mutex. */
 215 static struct vport *new_vport(const struct vport_parms *parms)
 216 {
 217         struct vport *vport;
 218
 219         vport = ovs_vport_add(parms);
 220         if (!IS_ERR(vport)) {
 221                 struct datapath *dp = parms->dp;
 222                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 223
 224                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 225         }
 226         return vport;
 227 }
 228
 229 void ovs_dp_detach_port(struct vport *p)
 230 {
 231         ASSERT_OVSL();
 232
 233         /* First drop references to device. */
 234         hlist_del_rcu(&p->dp_hash_node);
 235
 236         /* Then destroy it. */
 237         ovs_vport_del(p);
 238 }
 239
 240 /* Must be called with rcu_read_lock. */
 241 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 242 {
 243         const struct vport *p = OVS_CB(skb)->input_vport;
 244         struct datapath *dp = p->dp;
 245         struct sw_flow *flow;
 246         struct dp_stats_percpu *stats;
 247         u64 *stats_counter;
 248         u32 n_mask_hit;
 249
 250         stats = this_cpu_ptr(dp->stats_percpu);
 251
 252         /* Look up flow. */
 253         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
 254         if (unlikely(!flow)) {
 255                 struct dp_upcall_info upcall;
 256                 int error;
 257
 258                 upcall.cmd = OVS_PACKET_CMD_MISS;
 259                 upcall.key = key;
 260                 upcall.userdata = NULL;
 261                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 262                 error = ovs_dp_upcall(dp, skb, &upcall);
 263                 if (unlikely(error))
 264                         kfree_skb(skb);
 265                 else
 266                         consume_skb(skb);
 267                 stats_counter = &stats->n_missed;
 268                 goto out;
 269         }
 270
 271         OVS_CB(skb)->flow = flow;
 272
 273         ovs_flow_stats_update(OVS_CB(skb)->flow, key->tp.flags, skb);
 274         ovs_execute_actions(dp, skb, key);
 275         stats_counter = &stats->n_hit;
 276
 277 out:
 278         /* Update datapath statistics. */
 279         u64_stats_update_begin(&stats->syncp);
 280         (*stats_counter)++;
 281         stats->n_mask_hit += n_mask_hit;
 282         u64_stats_update_end(&stats->syncp);
 283 }
 284
 285 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 286                   const struct dp_upcall_info *upcall_info)
 287 {
 288         struct dp_stats_percpu *stats;
 289         int err;
 290
 291         if (upcall_info->portid == 0) {
 292                 err = -ENOTCONN;
 293                 goto err;
 294         }
 295
 296         if (!skb_is_gso(skb))
 297                 err = queue_userspace_packet(dp, skb, upcall_info);
 298         else
 299                 err = queue_gso_packets(dp, skb, upcall_info);
 300         if (err)
 301                 goto err;
 302
 303         return 0;
 304
 305 err:
 306         stats = this_cpu_ptr(dp->stats_percpu);
 307
 308         u64_stats_update_begin(&stats->syncp);
 309         stats->n_lost++;
 310         u64_stats_update_end(&stats->syncp);
 311
 312         return err;
 313 }
 314
 315 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 316                              const struct dp_upcall_info *upcall_info)
 317 {
 318         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 319         struct dp_upcall_info later_info;
 320         struct sw_flow_key later_key;
 321         struct sk_buff *segs, *nskb;
 322         int err;
 323
 324         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 325         if (IS_ERR(segs))
 326                 return PTR_ERR(segs);
 327         if (segs == NULL)
 328                 return -EINVAL;
 329
 330         /* Queue all of the segments. */
 331         skb = segs;
 332         do {
 333                 err = queue_userspace_packet(dp, skb, upcall_info);
 334                 if (err)
 335                         break;
 336
 337                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 338                         /* The initial flow key extracted by ovs_flow_extract()
 339                          * in this case is for a first fragment, so we need to
 340                          * properly mark later fragments.
 341                          */
 342                         later_key = *upcall_info->key;
 343                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 344
 345                         later_info = *upcall_info;
 346                         later_info.key = &later_key;
 347                         upcall_info = &later_info;
 348                 }
 349         } while ((skb = skb->next));
 350
 351         /* Free all of the segments. */
 352         skb = segs;
 353         do {
 354                 nskb = skb->next;
 355                 if (err)
 356                         kfree_skb(skb);
 357                 else
 358                         consume_skb(skb);
 359         } while ((skb = nskb));
 360         return err;
 361 }
 362
 363 static size_t key_attr_size(void)
 364 {
 365         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 366                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 367                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
 368                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 369                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 370                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
 371                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
 372                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 373                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
 374                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_OAM */
 375                   + nla_total_size(256)   /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
 376                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 377                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 378                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 379                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 380                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 381                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 382                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 383                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 384                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 385                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 386 }
 387
 388 static size_t upcall_msg_size(const struct nlattr *userdata,
 389                               unsigned int hdrlen)
 390 {
 391         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 392                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 393                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 394
 395         /* OVS_PACKET_ATTR_USERDATA */
 396         if (userdata)
 397                 size += NLA_ALIGN(userdata->nla_len);
 398
 399         return size;
 400 }
 401
 402 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 403                                   const struct dp_upcall_info *upcall_info)
 404 {
 405         struct ovs_header *upcall;
 406         struct sk_buff *nskb = NULL;
 407         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 408         struct nlattr *nla;
 409         struct genl_info info = {
 410                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 411                 .snd_portid = upcall_info->portid,
 412         };
 413         size_t len;
 414         unsigned int hlen;
 415         int err, dp_ifindex;
 416
 417         dp_ifindex = get_dpifindex(dp);
 418         if (!dp_ifindex)
 419                 return -ENODEV;
 420
 421         if (vlan_tx_tag_present(skb)) {
 422                 nskb = skb_clone(skb, GFP_ATOMIC);
 423                 if (!nskb)
 424                         return -ENOMEM;
 425
 426                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 427                 if (!nskb)
 428                         return -ENOMEM;
 429
 430                 nskb->vlan_tci = 0;
 431                 skb = nskb;
 432         }
 433
 434         if (nla_attr_size(skb->len) > USHRT_MAX) {
 435                 err = -EFBIG;
 436                 goto out;
 437         }
 438
 439         /* Complete checksum if needed */
 440         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 441             (err = skb_checksum_help(skb)))
 442                 goto out;
 443
 444         /* Older versions of OVS user space enforce alignment of the last
 445          * Netlink attribute to NLA_ALIGNTO which would require extensive
 446          * padding logic. Only perform zerocopy if padding is not required.
 447          */
 448         if (dp->user_features & OVS_DP_F_UNALIGNED)
 449                 hlen = skb_zerocopy_headlen(skb);
 450         else
 451                 hlen = skb->len;
 452
 453         len = upcall_msg_size(upcall_info->userdata, hlen);
 454         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 455         if (!user_skb) {
 456                 err = -ENOMEM;
 457                 goto out;
 458         }
 459
 460         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 461                              0, upcall_info->cmd);
 462         upcall->dp_ifindex = dp_ifindex;
 463
 464         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 465         err = ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
 466         BUG_ON(err);
 467         nla_nest_end(user_skb, nla);
 468
 469         if (upcall_info->userdata)
 470                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 471                           nla_len(upcall_info->userdata),
 472                           nla_data(upcall_info->userdata));
 473
 474         /* Only reserve room for attribute header, packet data is added
 475          * in skb_zerocopy() */
 476         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 477                 err = -ENOBUFS;
 478                 goto out;
 479         }
 480         nla->nla_len = nla_attr_size(skb->len);
 481
 482         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 483         if (err)
 484                 goto out;
 485
 486         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 487         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 488                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 489
 490                 if (plen > 0)
 491                         memset(skb_put(user_skb, plen), 0, plen);
 492         }
 493
 494         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 495
 496         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 497         user_skb = NULL;
 498 out:
 499         if (err)
 500                 skb_tx_error(skb);
 501         kfree_skb(user_skb);
 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         struct vport *input_vport;
 516         int len;
 517         int err;
 518
 519         err = -EINVAL;
 520         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 521             !a[OVS_PACKET_ATTR_ACTIONS])
 522                 goto err;
 523
 524         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 525         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 526         err = -ENOMEM;
 527         if (!packet)
 528                 goto err;
 529         skb_reserve(packet, NET_IP_ALIGN);
 530
 531         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 532
 533         skb_reset_mac_header(packet);
 534         eth = eth_hdr(packet);
 535
 536         /* Normally, setting the skb 'protocol' field would be handled by a
 537          * call to eth_type_trans(), but it assumes there's a sending
 538          * device, which we may not have. */
 539         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 540                 packet->protocol = eth->h_proto;
 541         else
 542                 packet->protocol = htons(ETH_P_802_2);
 543
 544         /* Build an sw_flow for sending this packet. */
 545         flow = ovs_flow_alloc();
 546         err = PTR_ERR(flow);
 547         if (IS_ERR(flow))
 548                 goto err_kfree_skb;
 549
 550         err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
 551                                              &flow->key);
 552         if (err)
 553                 goto err_flow_free;
 554
 555         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 556         err = PTR_ERR(acts);
 557         if (IS_ERR(acts))
 558                 goto err_flow_free;
 559
 560         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 561                                    &flow->key, 0, &acts);
 562         if (err)
 563                 goto err_flow_free;
 564
 565         rcu_assign_pointer(flow->sf_acts, acts);
 566
 567         OVS_CB(packet)->egress_tun_info = NULL;
 568         OVS_CB(packet)->flow = flow;
 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         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 579         if (!input_vport)
 580                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 581
 582         if (!input_vport)
 583                 goto err_unlock;
 584
 585         OVS_CB(packet)->input_vport = input_vport;
 586
 587         local_bh_disable();
 588         err = ovs_execute_actions(dp, packet, &flow->key);
 589         local_bh_enable();
 590         rcu_read_unlock();
 591
 592         ovs_flow_free(flow, false);
 593         return err;
 594
 595 err_unlock:
 596         rcu_read_unlock();
 597 err_flow_free:
 598         ovs_flow_free(flow, false);
 599 err_kfree_skb:
 600         kfree_skb(packet);
 601 err:
 602         return err;
 603 }
 604
 605 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 606         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 607         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 608         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 609 };
 610
 611 static const struct genl_ops dp_packet_genl_ops[] = {
 612         { .cmd = OVS_PACKET_CMD_EXECUTE,
 613           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 614           .policy = packet_policy,
 615           .doit = ovs_packet_cmd_execute
 616         }
 617 };
 618
 619 static struct genl_family dp_packet_genl_family = {
 620         .id = GENL_ID_GENERATE,
 621         .hdrsize = sizeof(struct ovs_header),
 622         .name = OVS_PACKET_FAMILY,
 623         .version = OVS_PACKET_VERSION,
 624         .maxattr = OVS_PACKET_ATTR_MAX,
 625         .netnsok = true,
 626         .parallel_ops = true,
 627         .ops = dp_packet_genl_ops,
 628         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 629 };
 630
 631 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
 632                          struct ovs_dp_megaflow_stats *mega_stats)
 633 {
 634         int i;
 635
 636         memset(mega_stats, 0, sizeof(*mega_stats));
 637
 638         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 639         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 640
 641         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 642
 643         for_each_possible_cpu(i) {
 644                 const struct dp_stats_percpu *percpu_stats;
 645                 struct dp_stats_percpu local_stats;
 646                 unsigned int start;
 647
 648                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 649
 650                 do {
 651                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 652                         local_stats = *percpu_stats;
 653                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 654
 655                 stats->n_hit += local_stats.n_hit;
 656                 stats->n_missed += local_stats.n_missed;
 657                 stats->n_lost += local_stats.n_lost;
 658                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 659         }
 660 }
 661
 662 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 663 {
 664         return NLMSG_ALIGN(sizeof(struct ovs_header))
 665                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 666                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 667                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 668                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 669                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 670                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 671 }
 672
 673 /* Called with ovs_mutex or RCU read lock. */
 674 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 675                                   struct sk_buff *skb, u32 portid,
 676                                   u32 seq, u32 flags, u8 cmd)
 677 {
 678         const int skb_orig_len = skb->len;
 679         struct nlattr *start;
 680         struct ovs_flow_stats stats;
 681         __be16 tcp_flags;
 682         unsigned long used;
 683         struct ovs_header *ovs_header;
 684         struct nlattr *nla;
 685         int err;
 686
 687         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 688         if (!ovs_header)
 689                 return -EMSGSIZE;
 690
 691         ovs_header->dp_ifindex = dp_ifindex;
 692
 693         /* Fill flow key. */
 694         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 695         if (!nla)
 696                 goto nla_put_failure;
 697
 698         err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
 699         if (err)
 700                 goto error;
 701         nla_nest_end(skb, nla);
 702
 703         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 704         if (!nla)
 705                 goto nla_put_failure;
 706
 707         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 708         if (err)
 709                 goto error;
 710
 711         nla_nest_end(skb, nla);
 712
 713         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 714
 715         if (used &&
 716             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 717                 goto nla_put_failure;
 718
 719         if (stats.n_packets &&
 720             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 721                 goto nla_put_failure;
 722
 723         if ((u8)ntohs(tcp_flags) &&
 724              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 725                 goto nla_put_failure;
 726
 727         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 728          * this is the first flow to be dumped into 'skb'.  This is unusual for
 729          * Netlink but individual action lists can be longer than
 730          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 731          * The userspace caller can always fetch the actions separately if it
 732          * really wants them.  (Most userspace callers in fact don't care.)
 733          *
 734          * This can only fail for dump operations because the skb is always
 735          * properly sized for single flows.
 736          */
 737         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 738         if (start) {
 739                 const struct sw_flow_actions *sf_acts;
 740
 741                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 742                 err = ovs_nla_put_actions(sf_acts->actions,
 743                                           sf_acts->actions_len, skb);
 744
 745                 if (!err)
 746                         nla_nest_end(skb, start);
 747                 else {
 748                         if (skb_orig_len)
 749                                 goto error;
 750
 751                         nla_nest_cancel(skb, start);
 752                 }
 753         } else if (skb_orig_len)
 754                 goto nla_put_failure;
 755
 756         return genlmsg_end(skb, ovs_header);
 757
 758 nla_put_failure:
 759         err = -EMSGSIZE;
 760 error:
 761         genlmsg_cancel(skb, ovs_header);
 762         return err;
 763 }
 764
 765 /* May not be called with RCU read lock. */
 766 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 767                                                struct genl_info *info,
 768                                                bool always)
 769 {
 770         struct sk_buff *skb;
 771
 772         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 773                 return NULL;
 774
 775         skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
 776         if (!skb)
 777                 return ERR_PTR(-ENOMEM);
 778
 779         return skb;
 780 }
 781
 782 /* Called with ovs_mutex. */
 783 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 784                                                int dp_ifindex,
 785                                                struct genl_info *info, u8 cmd,
 786                                                bool always)
 787 {
 788         struct sk_buff *skb;
 789         int retval;
 790
 791         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
 792                                       always);
 793         if (IS_ERR_OR_NULL(skb))
 794                 return skb;
 795
 796         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 797                                         info->snd_portid, info->snd_seq, 0,
 798                                         cmd);
 799         BUG_ON(retval < 0);
 800         return skb;
 801 }
 802
 803 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 804 {
 805         struct nlattr **a = info->attrs;
 806         struct ovs_header *ovs_header = info->userhdr;
 807         struct sw_flow *flow, *new_flow;
 808         struct sw_flow_mask mask;
 809         struct sk_buff *reply;
 810         struct datapath *dp;
 811         struct sw_flow_actions *acts;
 812         struct sw_flow_match match;
 813         int error;
 814
 815         /* Must have key and actions. */
 816         error = -EINVAL;
 817         if (!a[OVS_FLOW_ATTR_KEY])
 818                 goto error;
 819         if (!a[OVS_FLOW_ATTR_ACTIONS])
 820                 goto error;
 821
 822         /* Most of the time we need to allocate a new flow, do it before
 823          * locking.
 824          */
 825         new_flow = ovs_flow_alloc();
 826         if (IS_ERR(new_flow)) {
 827                 error = PTR_ERR(new_flow);
 828                 goto error;
 829         }
 830
 831         /* Extract key. */
 832         ovs_match_init(&match, &new_flow->unmasked_key, &mask);
 833         error = ovs_nla_get_match(&match,
 834                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 835         if (error)
 836                 goto err_kfree_flow;
 837
 838         ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
 839
 840         /* Validate actions. */
 841         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 842         error = PTR_ERR(acts);
 843         if (IS_ERR(acts))
 844                 goto err_kfree_flow;
 845
 846         error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
 847                                      0, &acts);
 848         if (error) {
 849                 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 850                 goto err_kfree_acts;
 851         }
 852
 853         reply = ovs_flow_cmd_alloc_info(acts, info, false);
 854         if (IS_ERR(reply)) {
 855                 error = PTR_ERR(reply);
 856                 goto err_kfree_acts;
 857         }
 858
 859         ovs_lock();
 860         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 861         if (unlikely(!dp)) {
 862                 error = -ENODEV;
 863                 goto err_unlock_ovs;
 864         }
 865         /* Check if this is a duplicate flow */
 866         flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
 867         if (likely(!flow)) {
 868                 rcu_assign_pointer(new_flow->sf_acts, acts);
 869
 870                 /* Put flow in bucket. */
 871                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 872                 if (unlikely(error)) {
 873                         acts = NULL;
 874                         goto err_unlock_ovs;
 875                 }
 876
 877                 if (unlikely(reply)) {
 878                         error = ovs_flow_cmd_fill_info(new_flow,
 879                                                        ovs_header->dp_ifindex,
 880                                                        reply, info->snd_portid,
 881                                                        info->snd_seq, 0,
 882                                                        OVS_FLOW_CMD_NEW);
 883                         BUG_ON(error < 0);
 884                 }
 885                 ovs_unlock();
 886         } else {
 887                 struct sw_flow_actions *old_acts;
 888
 889                 /* Bail out if we're not allowed to modify an existing flow.
 890                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 891                  * because Generic Netlink treats the latter as a dump
 892                  * request.  We also accept NLM_F_EXCL in case that bug ever
 893                  * gets fixed.
 894                  */
 895                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
 896                                                          | NLM_F_EXCL))) {
 897                         error = -EEXIST;
 898                         goto err_unlock_ovs;
 899                 }
 900                 /* The unmasked key has to be the same for flow updates. */
 901                 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
 902                         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
 903                         if (!flow) {
 904                                 error = -ENOENT;
 905                                 goto err_unlock_ovs;
 906                         }
 907                 }
 908                 /* Update actions. */
 909                 old_acts = ovsl_dereference(flow->sf_acts);
 910                 rcu_assign_pointer(flow->sf_acts, acts);
 911
 912                 if (unlikely(reply)) {
 913                         error = ovs_flow_cmd_fill_info(flow,
 914                                                        ovs_header->dp_ifindex,
 915                                                        reply, info->snd_portid,
 916                                                        info->snd_seq, 0,
 917                                                        OVS_FLOW_CMD_NEW);
 918                         BUG_ON(error < 0);
 919                 }
 920                 ovs_unlock();
 921
 922                 ovs_nla_free_flow_actions(old_acts);
 923                 ovs_flow_free(new_flow, false);
 924         }
 925
 926         if (reply)
 927                 ovs_notify(&dp_flow_genl_family, reply, info);
 928         return 0;
 929
 930 err_unlock_ovs:
 931         ovs_unlock();
 932         kfree_skb(reply);
 933 err_kfree_acts:
 934         kfree(acts);
 935 err_kfree_flow:
 936         ovs_flow_free(new_flow, false);
 937 error:
 938         return error;
 939 }
 940
 941 static struct sw_flow_actions *get_flow_actions(const struct nlattr *a,
 942                                                 const struct sw_flow_key *key,
 943                                                 const struct sw_flow_mask *mask)
 944 {
 945         struct sw_flow_actions *acts;
 946         struct sw_flow_key masked_key;
 947         int error;
 948
 949         acts = ovs_nla_alloc_flow_actions(nla_len(a));
 950         if (IS_ERR(acts))
 951                 return acts;
 952
 953         ovs_flow_mask_key(&masked_key, key, mask);
 954         error = ovs_nla_copy_actions(a, &masked_key, 0, &acts);
 955         if (error) {
 956                 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 957                 kfree(acts);
 958                 return ERR_PTR(error);
 959         }
 960
 961         return acts;
 962 }
 963
 964 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
 965 {
 966         struct nlattr **a = info->attrs;
 967         struct ovs_header *ovs_header = info->userhdr;
 968         struct sw_flow_key key;
 969         struct sw_flow *flow;
 970         struct sw_flow_mask mask;
 971         struct sk_buff *reply = NULL;
 972         struct datapath *dp;
 973         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
 974         struct sw_flow_match match;
 975         int error;
 976
 977         /* Extract key. */
 978         error = -EINVAL;
 979         if (!a[OVS_FLOW_ATTR_KEY])
 980                 goto error;
 981
 982         ovs_match_init(&match, &key, &mask);
 983         error = ovs_nla_get_match(&match,
 984                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 985         if (error)
 986                 goto error;
 987
 988         /* Validate actions. */
 989         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 990                 acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask);
 991                 if (IS_ERR(acts)) {
 992                         error = PTR_ERR(acts);
 993                         goto error;
 994                 }
 995         }
 996
 997         /* Can allocate before locking if have acts. */
 998         if (acts) {
 999                 reply = ovs_flow_cmd_alloc_info(acts, info, false);
1000                 if (IS_ERR(reply)) {
1001                         error = PTR_ERR(reply);
1002                         goto err_kfree_acts;
1003                 }
1004         }
1005
1006         ovs_lock();
1007         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1008         if (unlikely(!dp)) {
1009                 error = -ENODEV;
1010                 goto err_unlock_ovs;
1011         }
1012         /* Check that the flow exists. */
1013         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1014         if (unlikely(!flow)) {
1015                 error = -ENOENT;
1016                 goto err_unlock_ovs;
1017         }
1018
1019         /* Update actions, if present. */
1020         if (likely(acts)) {
1021                 old_acts = ovsl_dereference(flow->sf_acts);
1022                 rcu_assign_pointer(flow->sf_acts, acts);
1023
1024                 if (unlikely(reply)) {
1025                         error = ovs_flow_cmd_fill_info(flow,
1026                                                        ovs_header->dp_ifindex,
1027                                                        reply, info->snd_portid,
1028                                                        info->snd_seq, 0,
1029                                                        OVS_FLOW_CMD_NEW);
1030                         BUG_ON(error < 0);
1031                 }
1032         } else {
1033                 /* Could not alloc without acts before locking. */
1034                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1035                                                 info, OVS_FLOW_CMD_NEW, false);
1036                 if (unlikely(IS_ERR(reply))) {
1037                         error = PTR_ERR(reply);
1038                         goto err_unlock_ovs;
1039                 }
1040         }
1041
1042         /* Clear stats. */
1043         if (a[OVS_FLOW_ATTR_CLEAR])
1044                 ovs_flow_stats_clear(flow);
1045         ovs_unlock();
1046
1047         if (reply)
1048                 ovs_notify(&dp_flow_genl_family, reply, info);
1049         if (old_acts)
1050                 ovs_nla_free_flow_actions(old_acts);
1051
1052         return 0;
1053
1054 err_unlock_ovs:
1055         ovs_unlock();
1056         kfree_skb(reply);
1057 err_kfree_acts:
1058         kfree(acts);
1059 error:
1060         return error;
1061 }
1062
1063 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1064 {
1065         struct nlattr **a = info->attrs;
1066         struct ovs_header *ovs_header = info->userhdr;
1067         struct sw_flow_key key;
1068         struct sk_buff *reply;
1069         struct sw_flow *flow;
1070         struct datapath *dp;
1071         struct sw_flow_match match;
1072         int err;
1073
1074         if (!a[OVS_FLOW_ATTR_KEY]) {
1075                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1076                 return -EINVAL;
1077         }
1078
1079         ovs_match_init(&match, &key, NULL);
1080         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1081         if (err)
1082                 return err;
1083
1084         ovs_lock();
1085         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1086         if (!dp) {
1087                 err = -ENODEV;
1088                 goto unlock;
1089         }
1090
1091         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1092         if (!flow) {
1093                 err = -ENOENT;
1094                 goto unlock;
1095         }
1096
1097         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1098                                         OVS_FLOW_CMD_NEW, true);
1099         if (IS_ERR(reply)) {
1100                 err = PTR_ERR(reply);
1101                 goto unlock;
1102         }
1103
1104         ovs_unlock();
1105         return genlmsg_reply(reply, info);
1106 unlock:
1107         ovs_unlock();
1108         return err;
1109 }
1110
1111 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1112 {
1113         struct nlattr **a = info->attrs;
1114         struct ovs_header *ovs_header = info->userhdr;
1115         struct sw_flow_key key;
1116         struct sk_buff *reply;
1117         struct sw_flow *flow;
1118         struct datapath *dp;
1119         struct sw_flow_match match;
1120         int err;
1121
1122         if (likely(a[OVS_FLOW_ATTR_KEY])) {
1123                 ovs_match_init(&match, &key, NULL);
1124                 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1125                 if (unlikely(err))
1126                         return err;
1127         }
1128
1129         ovs_lock();
1130         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1131         if (unlikely(!dp)) {
1132                 err = -ENODEV;
1133                 goto unlock;
1134         }
1135
1136         if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1137                 err = ovs_flow_tbl_flush(&dp->table);
1138                 goto unlock;
1139         }
1140
1141         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1142         if (unlikely(!flow)) {
1143                 err = -ENOENT;
1144                 goto unlock;
1145         }
1146
1147         ovs_flow_tbl_remove(&dp->table, flow);
1148         ovs_unlock();
1149
1150         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1151                                         info, false);
1152         if (likely(reply)) {
1153                 if (likely(!IS_ERR(reply))) {
1154                         rcu_read_lock();        /*To keep RCU checker happy. */
1155                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1156                                                      reply, info->snd_portid,
1157                                                      info->snd_seq, 0,
1158                                                      OVS_FLOW_CMD_DEL);
1159                         rcu_read_unlock();
1160                         BUG_ON(err < 0);
1161
1162                         ovs_notify(&dp_flow_genl_family, reply, info);
1163                 } else {
1164                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1165                 }
1166         }
1167
1168         ovs_flow_free(flow, true);
1169         return 0;
1170 unlock:
1171         ovs_unlock();
1172         return err;
1173 }
1174
1175 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1176 {
1177         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1178         struct table_instance *ti;
1179         struct datapath *dp;
1180
1181         rcu_read_lock();
1182         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1183         if (!dp) {
1184                 rcu_read_unlock();
1185                 return -ENODEV;
1186         }
1187
1188         ti = rcu_dereference(dp->table.ti);
1189         for (;;) {
1190                 struct sw_flow *flow;
1191                 u32 bucket, obj;
1192
1193                 bucket = cb->args[0];
1194                 obj = cb->args[1];
1195                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1196                 if (!flow)
1197                         break;
1198
1199                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1200                                            NETLINK_CB(cb->skb).portid,
1201                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1202                                            OVS_FLOW_CMD_NEW) < 0)
1203                         break;
1204
1205                 cb->args[0] = bucket;
1206                 cb->args[1] = obj;
1207         }
1208         rcu_read_unlock();
1209         return skb->len;
1210 }
1211
1212 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1213         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1214         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1215         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1216 };
1217
1218 static const struct genl_ops dp_flow_genl_ops[] = {
1219         { .cmd = OVS_FLOW_CMD_NEW,
1220           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1221           .policy = flow_policy,
1222           .doit = ovs_flow_cmd_new
1223         },
1224         { .cmd = OVS_FLOW_CMD_DEL,
1225           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1226           .policy = flow_policy,
1227           .doit = ovs_flow_cmd_del
1228         },
1229         { .cmd = OVS_FLOW_CMD_GET,
1230           .flags = 0,               /* OK for unprivileged users. */
1231           .policy = flow_policy,
1232           .doit = ovs_flow_cmd_get,
1233           .dumpit = ovs_flow_cmd_dump
1234         },
1235         { .cmd = OVS_FLOW_CMD_SET,
1236           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1237           .policy = flow_policy,
1238           .doit = ovs_flow_cmd_set,
1239         },
1240 };
1241
1242 static struct genl_family dp_flow_genl_family = {
1243         .id = GENL_ID_GENERATE,
1244         .hdrsize = sizeof(struct ovs_header),
1245         .name = OVS_FLOW_FAMILY,
1246         .version = OVS_FLOW_VERSION,
1247         .maxattr = OVS_FLOW_ATTR_MAX,
1248         .netnsok = true,
1249         .parallel_ops = true,
1250         .ops = dp_flow_genl_ops,
1251         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1252         .mcgrps = &ovs_dp_flow_multicast_group,
1253         .n_mcgrps = 1,
1254 };
1255
1256 static size_t ovs_dp_cmd_msg_size(void)
1257 {
1258         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1259
1260         msgsize += nla_total_size(IFNAMSIZ);
1261         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1262         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1263         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1264
1265         return msgsize;
1266 }
1267
1268 /* Called with ovs_mutex or RCU read lock. */
1269 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1270                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1271 {
1272         struct ovs_header *ovs_header;
1273         struct ovs_dp_stats dp_stats;
1274         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1275         int err;
1276
1277         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1278                                    flags, cmd);
1279         if (!ovs_header)
1280                 goto error;
1281
1282         ovs_header->dp_ifindex = get_dpifindex(dp);
1283
1284         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1285         if (err)
1286                 goto nla_put_failure;
1287
1288         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1289         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1290                         &dp_stats))
1291                 goto nla_put_failure;
1292
1293         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1294                         sizeof(struct ovs_dp_megaflow_stats),
1295                         &dp_megaflow_stats))
1296                 goto nla_put_failure;
1297
1298         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1299                 goto nla_put_failure;
1300
1301         return genlmsg_end(skb, ovs_header);
1302
1303 nla_put_failure:
1304         genlmsg_cancel(skb, ovs_header);
1305 error:
1306         return -EMSGSIZE;
1307 }
1308
1309 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1310 {
1311         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1312 }
1313
1314 /* Called with rcu_read_lock or ovs_mutex. */
1315 static struct datapath *lookup_datapath(struct net *net,
1316                                         struct ovs_header *ovs_header,
1317                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1318 {
1319         struct datapath *dp;
1320
1321         if (!a[OVS_DP_ATTR_NAME])
1322                 dp = get_dp(net, ovs_header->dp_ifindex);
1323         else {
1324                 struct vport *vport;
1325
1326                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1327                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1328         }
1329         return dp ? dp : ERR_PTR(-ENODEV);
1330 }
1331
1332 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1333 {
1334         struct datapath *dp;
1335
1336         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1337         if (IS_ERR(dp))
1338                 return;
1339
1340         WARN(dp->user_features, "Dropping previously announced user features\n");
1341         dp->user_features = 0;
1342 }
1343
1344 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1345 {
1346         if (a[OVS_DP_ATTR_USER_FEATURES])
1347                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1348 }
1349
1350 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1351 {
1352         struct nlattr **a = info->attrs;
1353         struct vport_parms parms;
1354         struct sk_buff *reply;
1355         struct datapath *dp;
1356         struct vport *vport;
1357         struct ovs_net *ovs_net;
1358         int err, i;
1359
1360         err = -EINVAL;
1361         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1362                 goto err;
1363
1364         reply = ovs_dp_cmd_alloc_info(info);
1365         if (!reply)
1366                 return -ENOMEM;
1367
1368         err = -ENOMEM;
1369         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1370         if (dp == NULL)
1371                 goto err_free_reply;
1372
1373         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1374
1375         /* Allocate table. */
1376         err = ovs_flow_tbl_init(&dp->table);
1377         if (err)
1378                 goto err_free_dp;
1379
1380         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1381         if (!dp->stats_percpu) {
1382                 err = -ENOMEM;
1383                 goto err_destroy_table;
1384         }
1385
1386         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1387                             GFP_KERNEL);
1388         if (!dp->ports) {
1389                 err = -ENOMEM;
1390                 goto err_destroy_percpu;
1391         }
1392
1393         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1394                 INIT_HLIST_HEAD(&dp->ports[i]);
1395
1396         /* Set up our datapath device. */
1397         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1398         parms.type = OVS_VPORT_TYPE_INTERNAL;
1399         parms.options = NULL;
1400         parms.dp = dp;
1401         parms.port_no = OVSP_LOCAL;
1402         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1403
1404         ovs_dp_change(dp, a);
1405
1406         /* So far only local changes have been made, now need the lock. */
1407         ovs_lock();
1408
1409         vport = new_vport(&parms);
1410         if (IS_ERR(vport)) {
1411                 err = PTR_ERR(vport);
1412                 if (err == -EBUSY)
1413                         err = -EEXIST;
1414
1415                 if (err == -EEXIST) {
1416                         /* An outdated user space instance that does not understand
1417                          * the concept of user_features has attempted to create a new
1418                          * datapath and is likely to reuse it. Drop all user features.
1419                          */
1420                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1421                                 ovs_dp_reset_user_features(skb, info);
1422                 }
1423
1424                 goto err_destroy_ports_array;
1425         }
1426
1427         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1428                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1429         BUG_ON(err < 0);
1430
1431         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1432         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1433
1434         ovs_unlock();
1435
1436         ovs_notify(&dp_datapath_genl_family, reply, info);
1437         return 0;
1438
1439 err_destroy_ports_array:
1440         ovs_unlock();
1441         kfree(dp->ports);
1442 err_destroy_percpu:
1443         free_percpu(dp->stats_percpu);
1444 err_destroy_table:
1445         ovs_flow_tbl_destroy(&dp->table, false);
1446 err_free_dp:
1447         release_net(ovs_dp_get_net(dp));
1448         kfree(dp);
1449 err_free_reply:
1450         kfree_skb(reply);
1451 err:
1452         return err;
1453 }
1454
1455 /* Called with ovs_mutex. */
1456 static void __dp_destroy(struct datapath *dp)
1457 {
1458         int i;
1459
1460         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1461                 struct vport *vport;
1462                 struct hlist_node *n;
1463
1464                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1465                         if (vport->port_no != OVSP_LOCAL)
1466                                 ovs_dp_detach_port(vport);
1467         }
1468
1469         list_del_rcu(&dp->list_node);
1470
1471         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1472          * all ports in datapath are destroyed first before freeing datapath.
1473          */
1474         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1475
1476         /* RCU destroy the flow table */
1477         ovs_flow_tbl_destroy(&dp->table, true);
1478
1479         call_rcu(&dp->rcu, destroy_dp_rcu);
1480 }
1481
1482 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1483 {
1484         struct sk_buff *reply;
1485         struct datapath *dp;
1486         int err;
1487
1488         reply = ovs_dp_cmd_alloc_info(info);
1489         if (!reply)
1490                 return -ENOMEM;
1491
1492         ovs_lock();
1493         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1494         err = PTR_ERR(dp);
1495         if (IS_ERR(dp))
1496                 goto err_unlock_free;
1497
1498         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1499                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1500         BUG_ON(err < 0);
1501
1502         __dp_destroy(dp);
1503         ovs_unlock();
1504
1505         ovs_notify(&dp_datapath_genl_family, reply, info);
1506
1507         return 0;
1508
1509 err_unlock_free:
1510         ovs_unlock();
1511         kfree_skb(reply);
1512         return err;
1513 }
1514
1515 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1516 {
1517         struct sk_buff *reply;
1518         struct datapath *dp;
1519         int err;
1520
1521         reply = ovs_dp_cmd_alloc_info(info);
1522         if (!reply)
1523                 return -ENOMEM;
1524
1525         ovs_lock();
1526         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1527         err = PTR_ERR(dp);
1528         if (IS_ERR(dp))
1529                 goto err_unlock_free;
1530
1531         ovs_dp_change(dp, info->attrs);
1532
1533         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1534                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1535         BUG_ON(err < 0);
1536
1537         ovs_unlock();
1538         ovs_notify(&dp_datapath_genl_family, reply, info);
1539
1540         return 0;
1541
1542 err_unlock_free:
1543         ovs_unlock();
1544         kfree_skb(reply);
1545         return err;
1546 }
1547
1548 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1549 {
1550         struct sk_buff *reply;
1551         struct datapath *dp;
1552         int err;
1553
1554         reply = ovs_dp_cmd_alloc_info(info);
1555         if (!reply)
1556                 return -ENOMEM;
1557
1558         rcu_read_lock();
1559         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1560         if (IS_ERR(dp)) {
1561                 err = PTR_ERR(dp);
1562                 goto err_unlock_free;
1563         }
1564         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1565                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1566         BUG_ON(err < 0);
1567         rcu_read_unlock();
1568
1569         return genlmsg_reply(reply, info);
1570
1571 err_unlock_free:
1572         rcu_read_unlock();
1573         kfree_skb(reply);
1574         return err;
1575 }
1576
1577 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1578 {
1579         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1580         struct datapath *dp;
1581         int skip = cb->args[0];
1582         int i = 0;
1583
1584         rcu_read_lock();
1585         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1586                 if (i >= skip &&
1587                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1588                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1589                                          OVS_DP_CMD_NEW) < 0)
1590                         break;
1591                 i++;
1592         }
1593         rcu_read_unlock();
1594
1595         cb->args[0] = i;
1596
1597         return skb->len;
1598 }
1599
1600 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1601         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1602         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1603         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1604 };
1605
1606 static const struct genl_ops dp_datapath_genl_ops[] = {
1607         { .cmd = OVS_DP_CMD_NEW,
1608           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1609           .policy = datapath_policy,
1610           .doit = ovs_dp_cmd_new
1611         },
1612         { .cmd = OVS_DP_CMD_DEL,
1613           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1614           .policy = datapath_policy,
1615           .doit = ovs_dp_cmd_del
1616         },
1617         { .cmd = OVS_DP_CMD_GET,
1618           .flags = 0,               /* OK for unprivileged users. */
1619           .policy = datapath_policy,
1620           .doit = ovs_dp_cmd_get,
1621           .dumpit = ovs_dp_cmd_dump
1622         },
1623         { .cmd = OVS_DP_CMD_SET,
1624           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1625           .policy = datapath_policy,
1626           .doit = ovs_dp_cmd_set,
1627         },
1628 };
1629
1630 static struct genl_family dp_datapath_genl_family = {
1631         .id = GENL_ID_GENERATE,
1632         .hdrsize = sizeof(struct ovs_header),
1633         .name = OVS_DATAPATH_FAMILY,
1634         .version = OVS_DATAPATH_VERSION,
1635         .maxattr = OVS_DP_ATTR_MAX,
1636         .netnsok = true,
1637         .parallel_ops = true,
1638         .ops = dp_datapath_genl_ops,
1639         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1640         .mcgrps = &ovs_dp_datapath_multicast_group,
1641         .n_mcgrps = 1,
1642 };
1643
1644 /* Called with ovs_mutex or RCU read lock. */
1645 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1646                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1647 {
1648         struct ovs_header *ovs_header;
1649         struct ovs_vport_stats vport_stats;
1650         int err;
1651
1652         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1653                                  flags, cmd);
1654         if (!ovs_header)
1655                 return -EMSGSIZE;
1656
1657         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1658
1659         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1660             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1661             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1662                            vport->ops->get_name(vport)))
1663                 goto nla_put_failure;
1664
1665         ovs_vport_get_stats(vport, &vport_stats);
1666         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1667                     &vport_stats))
1668                 goto nla_put_failure;
1669
1670         if (ovs_vport_get_upcall_portids(vport, skb))
1671                 goto nla_put_failure;
1672
1673         err = ovs_vport_get_options(vport, skb);
1674         if (err == -EMSGSIZE)
1675                 goto error;
1676
1677         return genlmsg_end(skb, ovs_header);
1678
1679 nla_put_failure:
1680         err = -EMSGSIZE;
1681 error:
1682         genlmsg_cancel(skb, ovs_header);
1683         return err;
1684 }
1685
1686 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1687 {
1688         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1689 }
1690
1691 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1692 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1693                                          u32 seq, u8 cmd)
1694 {
1695         struct sk_buff *skb;
1696         int retval;
1697
1698         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1699         if (!skb)
1700                 return ERR_PTR(-ENOMEM);
1701
1702         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1703         BUG_ON(retval < 0);
1704
1705         return skb;
1706 }
1707
1708 /* Called with ovs_mutex or RCU read lock. */
1709 static struct vport *lookup_vport(struct net *net,
1710                                   struct ovs_header *ovs_header,
1711                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1712 {
1713         struct datapath *dp;
1714         struct vport *vport;
1715
1716         if (a[OVS_VPORT_ATTR_NAME]) {
1717                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1718                 if (!vport)
1719                         return ERR_PTR(-ENODEV);
1720                 if (ovs_header->dp_ifindex &&
1721                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1722                         return ERR_PTR(-ENODEV);
1723                 return vport;
1724         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1725                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1726
1727                 if (port_no >= DP_MAX_PORTS)
1728                         return ERR_PTR(-EFBIG);
1729
1730                 dp = get_dp(net, ovs_header->dp_ifindex);
1731                 if (!dp)
1732                         return ERR_PTR(-ENODEV);
1733
1734                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1735                 if (!vport)
1736                         return ERR_PTR(-ENODEV);
1737                 return vport;
1738         } else
1739                 return ERR_PTR(-EINVAL);
1740 }
1741
1742 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1743 {
1744         struct nlattr **a = info->attrs;
1745         struct ovs_header *ovs_header = info->userhdr;
1746         struct vport_parms parms;
1747         struct sk_buff *reply;
1748         struct vport *vport;
1749         struct datapath *dp;
1750         u32 port_no;
1751         int err;
1752
1753         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1754             !a[OVS_VPORT_ATTR_UPCALL_PID])
1755                 return -EINVAL;
1756
1757         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1758                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1759         if (port_no >= DP_MAX_PORTS)
1760                 return -EFBIG;
1761
1762         reply = ovs_vport_cmd_alloc_info();
1763         if (!reply)
1764                 return -ENOMEM;
1765
1766         ovs_lock();
1767         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1768         err = -ENODEV;
1769         if (!dp)
1770                 goto exit_unlock_free;
1771
1772         if (port_no) {
1773                 vport = ovs_vport_ovsl(dp, port_no);
1774                 err = -EBUSY;
1775                 if (vport)
1776                         goto exit_unlock_free;
1777         } else {
1778                 for (port_no = 1; ; port_no++) {
1779                         if (port_no >= DP_MAX_PORTS) {
1780                                 err = -EFBIG;
1781                                 goto exit_unlock_free;
1782                         }
1783                         vport = ovs_vport_ovsl(dp, port_no);
1784                         if (!vport)
1785                                 break;
1786                 }
1787         }
1788
1789         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1790         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1791         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1792         parms.dp = dp;
1793         parms.port_no = port_no;
1794         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1795
1796         vport = new_vport(&parms);
1797         err = PTR_ERR(vport);
1798         if (IS_ERR(vport))
1799                 goto exit_unlock_free;
1800
1801         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1802                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1803         BUG_ON(err < 0);
1804         ovs_unlock();
1805
1806         ovs_notify(&dp_vport_genl_family, reply, info);
1807         return 0;
1808
1809 exit_unlock_free:
1810         ovs_unlock();
1811         kfree_skb(reply);
1812         return err;
1813 }
1814
1815 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1816 {
1817         struct nlattr **a = info->attrs;
1818         struct sk_buff *reply;
1819         struct vport *vport;
1820         int err;
1821
1822         reply = ovs_vport_cmd_alloc_info();
1823         if (!reply)
1824                 return -ENOMEM;
1825
1826         ovs_lock();
1827         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1828         err = PTR_ERR(vport);
1829         if (IS_ERR(vport))
1830                 goto exit_unlock_free;
1831
1832         if (a[OVS_VPORT_ATTR_TYPE] &&
1833             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1834                 err = -EINVAL;
1835                 goto exit_unlock_free;
1836         }
1837
1838         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1839                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1840                 if (err)
1841                         goto exit_unlock_free;
1842         }
1843
1844
1845         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1846                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
1847
1848                 err = ovs_vport_set_upcall_portids(vport, ids);
1849                 if (err)
1850                         goto exit_unlock_free;
1851         }
1852
1853         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1854                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1855         BUG_ON(err < 0);
1856
1857         ovs_unlock();
1858         ovs_notify(&dp_vport_genl_family, reply, info);
1859         return 0;
1860
1861 exit_unlock_free:
1862         ovs_unlock();
1863         kfree_skb(reply);
1864         return err;
1865 }
1866
1867 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1868 {
1869         struct nlattr **a = info->attrs;
1870         struct sk_buff *reply;
1871         struct vport *vport;
1872         int err;
1873
1874         reply = ovs_vport_cmd_alloc_info();
1875         if (!reply)
1876                 return -ENOMEM;
1877
1878         ovs_lock();
1879         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1880         err = PTR_ERR(vport);
1881         if (IS_ERR(vport))
1882                 goto exit_unlock_free;
1883
1884         if (vport->port_no == OVSP_LOCAL) {
1885                 err = -EINVAL;
1886                 goto exit_unlock_free;
1887         }
1888
1889         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1890                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1891         BUG_ON(err < 0);
1892         ovs_dp_detach_port(vport);
1893         ovs_unlock();
1894
1895         ovs_notify(&dp_vport_genl_family, reply, info);
1896         return 0;
1897
1898 exit_unlock_free:
1899         ovs_unlock();
1900         kfree_skb(reply);
1901         return err;
1902 }
1903
1904 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1905 {
1906         struct nlattr **a = info->attrs;
1907         struct ovs_header *ovs_header = info->userhdr;
1908         struct sk_buff *reply;
1909         struct vport *vport;
1910         int err;
1911
1912         reply = ovs_vport_cmd_alloc_info();
1913         if (!reply)
1914                 return -ENOMEM;
1915
1916         rcu_read_lock();
1917         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1918         err = PTR_ERR(vport);
1919         if (IS_ERR(vport))
1920                 goto exit_unlock_free;
1921         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1922                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1923         BUG_ON(err < 0);
1924         rcu_read_unlock();
1925
1926         return genlmsg_reply(reply, info);
1927
1928 exit_unlock_free:
1929         rcu_read_unlock();
1930         kfree_skb(reply);
1931         return err;
1932 }
1933
1934 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1935 {
1936         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1937         struct datapath *dp;
1938         int bucket = cb->args[0], skip = cb->args[1];
1939         int i, j = 0;
1940
1941         rcu_read_lock();
1942         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1943         if (!dp) {
1944                 rcu_read_unlock();
1945                 return -ENODEV;
1946         }
1947         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1948                 struct vport *vport;
1949
1950                 j = 0;
1951                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1952                         if (j >= skip &&
1953                             ovs_vport_cmd_fill_info(vport, skb,
1954                                                     NETLINK_CB(cb->skb).portid,
1955                                                     cb->nlh->nlmsg_seq,
1956                                                     NLM_F_MULTI,
1957                                                     OVS_VPORT_CMD_NEW) < 0)
1958                                 goto out;
1959
1960                         j++;
1961                 }
1962                 skip = 0;
1963         }
1964 out:
1965         rcu_read_unlock();
1966
1967         cb->args[0] = i;
1968         cb->args[1] = j;
1969
1970         return skb->len;
1971 }
1972
1973 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1974         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1975         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1976         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1977         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1978         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1979         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1980 };
1981
1982 static const struct genl_ops dp_vport_genl_ops[] = {
1983         { .cmd = OVS_VPORT_CMD_NEW,
1984           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1985           .policy = vport_policy,
1986           .doit = ovs_vport_cmd_new
1987         },
1988         { .cmd = OVS_VPORT_CMD_DEL,
1989           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1990           .policy = vport_policy,
1991           .doit = ovs_vport_cmd_del
1992         },
1993         { .cmd = OVS_VPORT_CMD_GET,
1994           .flags = 0,               /* OK for unprivileged users. */
1995           .policy = vport_policy,
1996           .doit = ovs_vport_cmd_get,
1997           .dumpit = ovs_vport_cmd_dump
1998         },
1999         { .cmd = OVS_VPORT_CMD_SET,
2000           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2001           .policy = vport_policy,
2002           .doit = ovs_vport_cmd_set,
2003         },
2004 };
2005
2006 struct genl_family dp_vport_genl_family = {
2007         .id = GENL_ID_GENERATE,
2008         .hdrsize = sizeof(struct ovs_header),
2009         .name = OVS_VPORT_FAMILY,
2010         .version = OVS_VPORT_VERSION,
2011         .maxattr = OVS_VPORT_ATTR_MAX,
2012         .netnsok = true,
2013         .parallel_ops = true,
2014         .ops = dp_vport_genl_ops,
2015         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2016         .mcgrps = &ovs_dp_vport_multicast_group,
2017         .n_mcgrps = 1,
2018 };
2019
2020 static struct genl_family * const dp_genl_families[] = {
2021         &dp_datapath_genl_family,
2022         &dp_vport_genl_family,
2023         &dp_flow_genl_family,
2024         &dp_packet_genl_family,
2025 };
2026
2027 static void dp_unregister_genl(int n_families)
2028 {
2029         int i;
2030
2031         for (i = 0; i < n_families; i++)
2032                 genl_unregister_family(dp_genl_families[i]);
2033 }
2034
2035 static int dp_register_genl(void)
2036 {
2037         int err;
2038         int i;
2039
2040         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2041
2042                 err = genl_register_family(dp_genl_families[i]);
2043                 if (err)
2044                         goto error;
2045         }
2046
2047         return 0;
2048
2049 error:
2050         dp_unregister_genl(i);
2051         return err;
2052 }
2053
2054 static int __net_init ovs_init_net(struct net *net)
2055 {
2056         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2057
2058         INIT_LIST_HEAD(&ovs_net->dps);
2059         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2060         return 0;
2061 }
2062
2063 static void __net_exit ovs_exit_net(struct net *net)
2064 {
2065         struct datapath *dp, *dp_next;
2066         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2067
2068         ovs_lock();
2069         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2070                 __dp_destroy(dp);
2071         ovs_unlock();
2072
2073         cancel_work_sync(&ovs_net->dp_notify_work);
2074 }
2075
2076 static struct pernet_operations ovs_net_ops = {
2077         .init = ovs_init_net,
2078         .exit = ovs_exit_net,
2079         .id   = &ovs_net_id,
2080         .size = sizeof(struct ovs_net),
2081 };
2082
2083 static int __init dp_init(void)
2084 {
2085         int err;
2086
2087         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2088
2089         pr_info("Open vSwitch switching datapath\n");
2090
2091         err = action_fifos_init();
2092         if (err)
2093                 goto error;
2094
2095         err = ovs_internal_dev_rtnl_link_register();
2096         if (err)
2097                 goto error_action_fifos_exit;
2098
2099         err = ovs_flow_init();
2100         if (err)
2101                 goto error_unreg_rtnl_link;
2102
2103         err = ovs_vport_init();
2104         if (err)
2105                 goto error_flow_exit;
2106
2107         err = register_pernet_device(&ovs_net_ops);
2108         if (err)
2109                 goto error_vport_exit;
2110
2111         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2112         if (err)
2113                 goto error_netns_exit;
2114
2115         err = dp_register_genl();
2116         if (err < 0)
2117                 goto error_unreg_notifier;
2118
2119         return 0;
2120
2121 error_unreg_notifier:
2122         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2123 error_netns_exit:
2124         unregister_pernet_device(&ovs_net_ops);
2125 error_vport_exit:
2126         ovs_vport_exit();
2127 error_flow_exit:
2128         ovs_flow_exit();
2129 error_unreg_rtnl_link:
2130         ovs_internal_dev_rtnl_link_unregister();
2131 error_action_fifos_exit:
2132         action_fifos_exit();
2133 error:
2134         return err;
2135 }
2136
2137 static void dp_cleanup(void)
2138 {
2139         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2140         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2141         unregister_pernet_device(&ovs_net_ops);
2142         rcu_barrier();
2143         ovs_vport_exit();
2144         ovs_flow_exit();
2145         ovs_internal_dev_rtnl_link_unregister();
2146         action_fifos_exit();
2147 }
2148
2149 module_init(dp_init);
2150 module_exit(dp_cleanup);
2151
2152 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2153 MODULE_LICENSE("GPL");