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 EXPORT_SYMBOL(ovs_net_id);
  63
  64 static struct genl_family dp_packet_genl_family;
  65 static struct genl_family dp_flow_genl_family;
  66 static struct genl_family dp_datapath_genl_family;
  67
  68 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
  69         .name = OVS_FLOW_MCGROUP,
  70 };
  71
  72 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  73         .name = OVS_DATAPATH_MCGROUP,
  74 };
  75
  76 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
  77         .name = OVS_VPORT_MCGROUP,
  78 };
  79
  80 /* Check if need to build a reply message.
  81  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  82 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  83                             unsigned int group)
  84 {
  85         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  86                genl_has_listeners(family, genl_info_net(info)->genl_sock,
  87                                   group);
  88 }
  89
  90 static void ovs_notify(struct genl_family *family,
  91                        struct sk_buff *skb, struct genl_info *info)
  92 {
  93         genl_notify(family, skb, genl_info_net(info), info->snd_portid,
  94                     0, info->nlhdr, GFP_KERNEL);
  95 }
  96
  97 /**
  98  * DOC: Locking:
  99  *
 100  * All writes e.g. Writes to device state (add/remove datapath, port, set
 101  * operations on vports, etc.), Writes to other state (flow table
 102  * modifications, set miscellaneous datapath parameters, etc.) are protected
 103  * by ovs_lock.
 104  *
 105  * Reads are protected by RCU.
 106  *
 107  * There are a few special cases (mostly stats) that have their own
 108  * synchronization but they nest under all of above and don't interact with
 109  * each other.
 110  *
 111  * The RTNL lock nests inside ovs_mutex.
 112  */
 113
 114 static DEFINE_MUTEX(ovs_mutex);
 115
 116 void ovs_lock(void)
 117 {
 118         mutex_lock(&ovs_mutex);
 119 }
 120
 121 void ovs_unlock(void)
 122 {
 123         mutex_unlock(&ovs_mutex);
 124 }
 125
 126 #ifdef CONFIG_LOCKDEP
 127 int lockdep_ovsl_is_held(void)
 128 {
 129         if (debug_locks)
 130                 return lockdep_is_held(&ovs_mutex);
 131         else
 132                 return 1;
 133 }
 134 EXPORT_SYMBOL(lockdep_ovsl_is_held);
 135 #endif
 136
 137 static struct vport *new_vport(const struct vport_parms *);
 138 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 139                              const struct dp_upcall_info *);
 140 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 141                                   const struct dp_upcall_info *);
 142
 143 /* Must be called with rcu_read_lock or ovs_mutex. */
 144 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 145 {
 146         struct datapath *dp = NULL;
 147         struct net_device *dev;
 148
 149         rcu_read_lock();
 150         dev = dev_get_by_index_rcu(net, dp_ifindex);
 151         if (dev) {
 152                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 153                 if (vport)
 154                         dp = vport->dp;
 155         }
 156         rcu_read_unlock();
 157
 158         return dp;
 159 }
 160
 161 /* Must be called with rcu_read_lock or ovs_mutex. */
 162 const char *ovs_dp_name(const struct datapath *dp)
 163 {
 164         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 165         return vport->ops->get_name(vport);
 166 }
 167
 168 static int get_dpifindex(struct datapath *dp)
 169 {
 170         struct vport *local;
 171         int ifindex;
 172
 173         rcu_read_lock();
 174
 175         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 176         if (local)
 177                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 178         else
 179                 ifindex = 0;
 180
 181         rcu_read_unlock();
 182
 183         return ifindex;
 184 }
 185
 186 static void destroy_dp_rcu(struct rcu_head *rcu)
 187 {
 188         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 189
 190         free_percpu(dp->stats_percpu);
 191         release_net(ovs_dp_get_net(dp));
 192         kfree(dp->ports);
 193         kfree(dp);
 194 }
 195
 196 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 197                                             u16 port_no)
 198 {
 199         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 200 }
 201
 202 /* Called with ovs_mutex or RCU read lock. */
 203 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 204 {
 205         struct vport *vport;
 206         struct hlist_head *head;
 207
 208         head = vport_hash_bucket(dp, port_no);
 209         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 210                 if (vport->port_no == port_no)
 211                         return vport;
 212         }
 213         return NULL;
 214 }
 215
 216 /* Called with ovs_mutex. */
 217 static struct vport *new_vport(const struct vport_parms *parms)
 218 {
 219         struct vport *vport;
 220
 221         vport = ovs_vport_add(parms);
 222         if (!IS_ERR(vport)) {
 223                 struct datapath *dp = parms->dp;
 224                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 225
 226                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 227         }
 228         return vport;
 229 }
 230
 231 void ovs_dp_detach_port(struct vport *p)
 232 {
 233         ASSERT_OVSL();
 234
 235         /* First drop references to device. */
 236         hlist_del_rcu(&p->dp_hash_node);
 237
 238         /* Then destroy it. */
 239         ovs_vport_del(p);
 240 }
 241
 242 /* Must be called with rcu_read_lock. */
 243 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 244 {
 245         const struct vport *p = OVS_CB(skb)->input_vport;
 246         struct datapath *dp = p->dp;
 247         struct sw_flow *flow;
 248         struct dp_stats_percpu *stats;
 249         u64 *stats_counter;
 250         u32 n_mask_hit;
 251
 252         stats = this_cpu_ptr(dp->stats_percpu);
 253
 254         /* Look up flow. */
 255         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
 256         if (unlikely(!flow)) {
 257                 struct dp_upcall_info upcall;
 258                 int error;
 259
 260                 upcall.cmd = OVS_PACKET_CMD_MISS;
 261                 upcall.key = key;
 262                 upcall.userdata = NULL;
 263                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 264                 error = ovs_dp_upcall(dp, skb, &upcall);
 265                 if (unlikely(error))
 266                         kfree_skb(skb);
 267                 else
 268                         consume_skb(skb);
 269                 stats_counter = &stats->n_missed;
 270                 goto out;
 271         }
 272
 273         OVS_CB(skb)->flow = flow;
 274
 275         ovs_flow_stats_update(OVS_CB(skb)->flow, key->tp.flags, skb);
 276         ovs_execute_actions(dp, skb, key);
 277         stats_counter = &stats->n_hit;
 278
 279 out:
 280         /* Update datapath statistics. */
 281         u64_stats_update_begin(&stats->syncp);
 282         (*stats_counter)++;
 283         stats->n_mask_hit += n_mask_hit;
 284         u64_stats_update_end(&stats->syncp);
 285 }
 286
 287 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 288                   const struct dp_upcall_info *upcall_info)
 289 {
 290         struct dp_stats_percpu *stats;
 291         int err;
 292
 293         if (upcall_info->portid == 0) {
 294                 err = -ENOTCONN;
 295                 goto err;
 296         }
 297
 298         if (!skb_is_gso(skb))
 299                 err = queue_userspace_packet(dp, skb, upcall_info);
 300         else
 301                 err = queue_gso_packets(dp, skb, upcall_info);
 302         if (err)
 303                 goto err;
 304
 305         return 0;
 306
 307 err:
 308         stats = this_cpu_ptr(dp->stats_percpu);
 309
 310         u64_stats_update_begin(&stats->syncp);
 311         stats->n_lost++;
 312         u64_stats_update_end(&stats->syncp);
 313
 314         return err;
 315 }
 316
 317 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 318                              const struct dp_upcall_info *upcall_info)
 319 {
 320         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 321         struct dp_upcall_info later_info;
 322         struct sw_flow_key later_key;
 323         struct sk_buff *segs, *nskb;
 324         int err;
 325
 326         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 327         if (IS_ERR(segs))
 328                 return PTR_ERR(segs);
 329         if (segs == NULL)
 330                 return -EINVAL;
 331
 332         /* Queue all of the segments. */
 333         skb = segs;
 334         do {
 335                 err = queue_userspace_packet(dp, skb, upcall_info);
 336                 if (err)
 337                         break;
 338
 339                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 340                         /* The initial flow key extracted by ovs_flow_extract()
 341                          * in this case is for a first fragment, so we need to
 342                          * properly mark later fragments.
 343                          */
 344                         later_key = *upcall_info->key;
 345                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 346
 347                         later_info = *upcall_info;
 348                         later_info.key = &later_key;
 349                         upcall_info = &later_info;
 350                 }
 351         } while ((skb = skb->next));
 352
 353         /* Free all of the segments. */
 354         skb = segs;
 355         do {
 356                 nskb = skb->next;
 357                 if (err)
 358                         kfree_skb(skb);
 359                 else
 360                         consume_skb(skb);
 361         } while ((skb = nskb));
 362         return err;
 363 }
 364
 365 static size_t key_attr_size(void)
 366 {
 367         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 368                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 369                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
 370                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 371                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 372                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
 373                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
 374                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 375                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
 376                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_OAM */
 377                   + nla_total_size(256)   /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
 378                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 379                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 380                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 381                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 382                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 383                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 384                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 385                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 386                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 387                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 388 }
 389
 390 static size_t upcall_msg_size(const struct nlattr *userdata,
 391                               unsigned int hdrlen)
 392 {
 393         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 394                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 395                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 396
 397         /* OVS_PACKET_ATTR_USERDATA */
 398         if (userdata)
 399                 size += NLA_ALIGN(userdata->nla_len);
 400
 401         return size;
 402 }
 403
 404 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 405                                   const struct dp_upcall_info *upcall_info)
 406 {
 407         struct ovs_header *upcall;
 408         struct sk_buff *nskb = NULL;
 409         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 410         struct nlattr *nla;
 411         struct genl_info info = {
 412                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 413                 .snd_portid = upcall_info->portid,
 414         };
 415         size_t len;
 416         unsigned int hlen;
 417         int err, dp_ifindex;
 418
 419         dp_ifindex = get_dpifindex(dp);
 420         if (!dp_ifindex)
 421                 return -ENODEV;
 422
 423         if (vlan_tx_tag_present(skb)) {
 424                 nskb = skb_clone(skb, GFP_ATOMIC);
 425                 if (!nskb)
 426                         return -ENOMEM;
 427
 428                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 429                 if (!nskb)
 430                         return -ENOMEM;
 431
 432                 nskb->vlan_tci = 0;
 433                 skb = nskb;
 434         }
 435
 436         if (nla_attr_size(skb->len) > USHRT_MAX) {
 437                 err = -EFBIG;
 438                 goto out;
 439         }
 440
 441         /* Complete checksum if needed */
 442         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 443             (err = skb_checksum_help(skb)))
 444                 goto out;
 445
 446         /* Older versions of OVS user space enforce alignment of the last
 447          * Netlink attribute to NLA_ALIGNTO which would require extensive
 448          * padding logic. Only perform zerocopy if padding is not required.
 449          */
 450         if (dp->user_features & OVS_DP_F_UNALIGNED)
 451                 hlen = skb_zerocopy_headlen(skb);
 452         else
 453                 hlen = skb->len;
 454
 455         len = upcall_msg_size(upcall_info->userdata, hlen);
 456         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 457         if (!user_skb) {
 458                 err = -ENOMEM;
 459                 goto out;
 460         }
 461
 462         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 463                              0, upcall_info->cmd);
 464         upcall->dp_ifindex = dp_ifindex;
 465
 466         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 467         err = ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
 468         BUG_ON(err);
 469         nla_nest_end(user_skb, nla);
 470
 471         if (upcall_info->userdata)
 472                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 473                           nla_len(upcall_info->userdata),
 474                           nla_data(upcall_info->userdata));
 475
 476         /* Only reserve room for attribute header, packet data is added
 477          * in skb_zerocopy() */
 478         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 479                 err = -ENOBUFS;
 480                 goto out;
 481         }
 482         nla->nla_len = nla_attr_size(skb->len);
 483
 484         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 485         if (err)
 486                 goto out;
 487
 488         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 489         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 490                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 491
 492                 if (plen > 0)
 493                         memset(skb_put(user_skb, plen), 0, plen);
 494         }
 495
 496         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 497
 498         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 499         user_skb = NULL;
 500 out:
 501         if (err)
 502                 skb_tx_error(skb);
 503         kfree_skb(user_skb);
 504         kfree_skb(nskb);
 505         return err;
 506 }
 507
 508 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 509 {
 510         struct ovs_header *ovs_header = info->userhdr;
 511         struct nlattr **a = info->attrs;
 512         struct sw_flow_actions *acts;
 513         struct sk_buff *packet;
 514         struct sw_flow *flow;
 515         struct datapath *dp;
 516         struct ethhdr *eth;
 517         struct vport *input_vport;
 518         int len;
 519         int err;
 520
 521         err = -EINVAL;
 522         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 523             !a[OVS_PACKET_ATTR_ACTIONS])
 524                 goto err;
 525
 526         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 527         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 528         err = -ENOMEM;
 529         if (!packet)
 530                 goto err;
 531         skb_reserve(packet, NET_IP_ALIGN);
 532
 533         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 534
 535         skb_reset_mac_header(packet);
 536         eth = eth_hdr(packet);
 537
 538         /* Normally, setting the skb 'protocol' field would be handled by a
 539          * call to eth_type_trans(), but it assumes there's a sending
 540          * device, which we may not have. */
 541         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 542                 packet->protocol = eth->h_proto;
 543         else
 544                 packet->protocol = htons(ETH_P_802_2);
 545
 546         /* Build an sw_flow for sending this packet. */
 547         flow = ovs_flow_alloc();
 548         err = PTR_ERR(flow);
 549         if (IS_ERR(flow))
 550                 goto err_kfree_skb;
 551
 552         err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
 553                                              &flow->key);
 554         if (err)
 555                 goto err_flow_free;
 556
 557         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 558         err = PTR_ERR(acts);
 559         if (IS_ERR(acts))
 560                 goto err_flow_free;
 561
 562         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 563                                    &flow->key, &acts);
 564         if (err)
 565                 goto err_flow_free;
 566
 567         rcu_assign_pointer(flow->sf_acts, acts);
 568
 569         OVS_CB(packet)->egress_tun_info = NULL;
 570         OVS_CB(packet)->flow = flow;
 571         packet->priority = flow->key.phy.priority;
 572         packet->mark = flow->key.phy.skb_mark;
 573
 574         rcu_read_lock();
 575         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 576         err = -ENODEV;
 577         if (!dp)
 578                 goto err_unlock;
 579
 580         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 581         if (!input_vport)
 582                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 583
 584         if (!input_vport)
 585                 goto err_unlock;
 586
 587         OVS_CB(packet)->input_vport = input_vport;
 588
 589         local_bh_disable();
 590         err = ovs_execute_actions(dp, packet, &flow->key);
 591         local_bh_enable();
 592         rcu_read_unlock();
 593
 594         ovs_flow_free(flow, false);
 595         return err;
 596
 597 err_unlock:
 598         rcu_read_unlock();
 599 err_flow_free:
 600         ovs_flow_free(flow, false);
 601 err_kfree_skb:
 602         kfree_skb(packet);
 603 err:
 604         return err;
 605 }
 606
 607 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 608         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 609         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 610         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 611 };
 612
 613 static const struct genl_ops dp_packet_genl_ops[] = {
 614         { .cmd = OVS_PACKET_CMD_EXECUTE,
 615           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 616           .policy = packet_policy,
 617           .doit = ovs_packet_cmd_execute
 618         }
 619 };
 620
 621 static struct genl_family dp_packet_genl_family = {
 622         .id = GENL_ID_GENERATE,
 623         .hdrsize = sizeof(struct ovs_header),
 624         .name = OVS_PACKET_FAMILY,
 625         .version = OVS_PACKET_VERSION,
 626         .maxattr = OVS_PACKET_ATTR_MAX,
 627         .netnsok = true,
 628         .parallel_ops = true,
 629         .ops = dp_packet_genl_ops,
 630         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 631 };
 632
 633 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
 634                          struct ovs_dp_megaflow_stats *mega_stats)
 635 {
 636         int i;
 637
 638         memset(mega_stats, 0, sizeof(*mega_stats));
 639
 640         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 641         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 642
 643         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 644
 645         for_each_possible_cpu(i) {
 646                 const struct dp_stats_percpu *percpu_stats;
 647                 struct dp_stats_percpu local_stats;
 648                 unsigned int start;
 649
 650                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 651
 652                 do {
 653                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 654                         local_stats = *percpu_stats;
 655                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 656
 657                 stats->n_hit += local_stats.n_hit;
 658                 stats->n_missed += local_stats.n_missed;
 659                 stats->n_lost += local_stats.n_lost;
 660                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 661         }
 662 }
 663
 664 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 665 {
 666         return NLMSG_ALIGN(sizeof(struct ovs_header))
 667                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 668                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 669                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 670                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 671                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 672                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 673 }
 674
 675 /* Called with ovs_mutex or RCU read lock. */
 676 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 677                                   struct sk_buff *skb, u32 portid,
 678                                   u32 seq, u32 flags, u8 cmd)
 679 {
 680         const int skb_orig_len = skb->len;
 681         struct nlattr *start;
 682         struct ovs_flow_stats stats;
 683         __be16 tcp_flags;
 684         unsigned long used;
 685         struct ovs_header *ovs_header;
 686         struct nlattr *nla;
 687         int err;
 688
 689         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 690         if (!ovs_header)
 691                 return -EMSGSIZE;
 692
 693         ovs_header->dp_ifindex = dp_ifindex;
 694
 695         /* Fill flow key. */
 696         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 697         if (!nla)
 698                 goto nla_put_failure;
 699
 700         err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
 701         if (err)
 702                 goto error;
 703         nla_nest_end(skb, nla);
 704
 705         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 706         if (!nla)
 707                 goto nla_put_failure;
 708
 709         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 710         if (err)
 711                 goto error;
 712
 713         nla_nest_end(skb, nla);
 714
 715         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 716
 717         if (used &&
 718             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 719                 goto nla_put_failure;
 720
 721         if (stats.n_packets &&
 722             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 723                 goto nla_put_failure;
 724
 725         if ((u8)ntohs(tcp_flags) &&
 726              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 727                 goto nla_put_failure;
 728
 729         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 730          * this is the first flow to be dumped into 'skb'.  This is unusual for
 731          * Netlink but individual action lists can be longer than
 732          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 733          * The userspace caller can always fetch the actions separately if it
 734          * really wants them.  (Most userspace callers in fact don't care.)
 735          *
 736          * This can only fail for dump operations because the skb is always
 737          * properly sized for single flows.
 738          */
 739         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 740         if (start) {
 741                 const struct sw_flow_actions *sf_acts;
 742
 743                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 744                 err = ovs_nla_put_actions(sf_acts->actions,
 745                                           sf_acts->actions_len, skb);
 746
 747                 if (!err)
 748                         nla_nest_end(skb, start);
 749                 else {
 750                         if (skb_orig_len)
 751                                 goto error;
 752
 753                         nla_nest_cancel(skb, start);
 754                 }
 755         } else if (skb_orig_len)
 756                 goto nla_put_failure;
 757
 758         return genlmsg_end(skb, ovs_header);
 759
 760 nla_put_failure:
 761         err = -EMSGSIZE;
 762 error:
 763         genlmsg_cancel(skb, ovs_header);
 764         return err;
 765 }
 766
 767 /* May not be called with RCU read lock. */
 768 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 769                                                struct genl_info *info,
 770                                                bool always)
 771 {
 772         struct sk_buff *skb;
 773
 774         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 775                 return NULL;
 776
 777         skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
 778         if (!skb)
 779                 return ERR_PTR(-ENOMEM);
 780
 781         return skb;
 782 }
 783
 784 /* Called with ovs_mutex. */
 785 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 786                                                int dp_ifindex,
 787                                                struct genl_info *info, u8 cmd,
 788                                                bool always)
 789 {
 790         struct sk_buff *skb;
 791         int retval;
 792
 793         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
 794                                       always);
 795         if (IS_ERR_OR_NULL(skb))
 796                 return skb;
 797
 798         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 799                                         info->snd_portid, info->snd_seq, 0,
 800                                         cmd);
 801         BUG_ON(retval < 0);
 802         return skb;
 803 }
 804
 805 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 806 {
 807         struct nlattr **a = info->attrs;
 808         struct ovs_header *ovs_header = info->userhdr;
 809         struct sw_flow *flow, *new_flow;
 810         struct sw_flow_mask mask;
 811         struct sk_buff *reply;
 812         struct datapath *dp;
 813         struct sw_flow_actions *acts;
 814         struct sw_flow_match match;
 815         int error;
 816
 817         /* Must have key and actions. */
 818         error = -EINVAL;
 819         if (!a[OVS_FLOW_ATTR_KEY])
 820                 goto error;
 821         if (!a[OVS_FLOW_ATTR_ACTIONS])
 822                 goto error;
 823
 824         /* Most of the time we need to allocate a new flow, do it before
 825          * locking.
 826          */
 827         new_flow = ovs_flow_alloc();
 828         if (IS_ERR(new_flow)) {
 829                 error = PTR_ERR(new_flow);
 830                 goto error;
 831         }
 832
 833         /* Extract key. */
 834         ovs_match_init(&match, &new_flow->unmasked_key, &mask);
 835         error = ovs_nla_get_match(&match,
 836                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 837         if (error)
 838                 goto err_kfree_flow;
 839
 840         ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
 841
 842         /* Validate actions. */
 843         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 844         error = PTR_ERR(acts);
 845         if (IS_ERR(acts))
 846                 goto err_kfree_flow;
 847
 848         error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
 849                                      &acts);
 850         if (error) {
 851                 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 852                 goto err_kfree_acts;
 853         }
 854
 855         reply = ovs_flow_cmd_alloc_info(acts, info, false);
 856         if (IS_ERR(reply)) {
 857                 error = PTR_ERR(reply);
 858                 goto err_kfree_acts;
 859         }
 860
 861         ovs_lock();
 862         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 863         if (unlikely(!dp)) {
 864                 error = -ENODEV;
 865                 goto err_unlock_ovs;
 866         }
 867         /* Check if this is a duplicate flow */
 868         flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
 869         if (likely(!flow)) {
 870                 rcu_assign_pointer(new_flow->sf_acts, acts);
 871
 872                 /* Put flow in bucket. */
 873                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 874                 if (unlikely(error)) {
 875                         acts = NULL;
 876                         goto err_unlock_ovs;
 877                 }
 878
 879                 if (unlikely(reply)) {
 880                         error = ovs_flow_cmd_fill_info(new_flow,
 881                                                        ovs_header->dp_ifindex,
 882                                                        reply, info->snd_portid,
 883                                                        info->snd_seq, 0,
 884                                                        OVS_FLOW_CMD_NEW);
 885                         BUG_ON(error < 0);
 886                 }
 887                 ovs_unlock();
 888         } else {
 889                 struct sw_flow_actions *old_acts;
 890
 891                 /* Bail out if we're not allowed to modify an existing flow.
 892                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 893                  * because Generic Netlink treats the latter as a dump
 894                  * request.  We also accept NLM_F_EXCL in case that bug ever
 895                  * gets fixed.
 896                  */
 897                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
 898                                                          | NLM_F_EXCL))) {
 899                         error = -EEXIST;
 900                         goto err_unlock_ovs;
 901                 }
 902                 /* The unmasked key has to be the same for flow updates. */
 903                 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
 904                         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
 905                         if (!flow) {
 906                                 error = -ENOENT;
 907                                 goto err_unlock_ovs;
 908                         }
 909                 }
 910                 /* Update actions. */
 911                 old_acts = ovsl_dereference(flow->sf_acts);
 912                 rcu_assign_pointer(flow->sf_acts, acts);
 913
 914                 if (unlikely(reply)) {
 915                         error = ovs_flow_cmd_fill_info(flow,
 916                                                        ovs_header->dp_ifindex,
 917                                                        reply, info->snd_portid,
 918                                                        info->snd_seq, 0,
 919                                                        OVS_FLOW_CMD_NEW);
 920                         BUG_ON(error < 0);
 921                 }
 922                 ovs_unlock();
 923
 924                 ovs_nla_free_flow_actions(old_acts);
 925                 ovs_flow_free(new_flow, false);
 926         }
 927
 928         if (reply)
 929                 ovs_notify(&dp_flow_genl_family, reply, info);
 930         return 0;
 931
 932 err_unlock_ovs:
 933         ovs_unlock();
 934         kfree_skb(reply);
 935 err_kfree_acts:
 936         kfree(acts);
 937 err_kfree_flow:
 938         ovs_flow_free(new_flow, false);
 939 error:
 940         return error;
 941 }
 942
 943 static struct sw_flow_actions *get_flow_actions(const struct nlattr *a,
 944                                                 const struct sw_flow_key *key,
 945                                                 const struct sw_flow_mask *mask)
 946 {
 947         struct sw_flow_actions *acts;
 948         struct sw_flow_key masked_key;
 949         int error;
 950
 951         acts = ovs_nla_alloc_flow_actions(nla_len(a));
 952         if (IS_ERR(acts))
 953                 return acts;
 954
 955         ovs_flow_mask_key(&masked_key, key, mask);
 956         error = ovs_nla_copy_actions(a, &masked_key, &acts);
 957         if (error) {
 958                 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 959                 kfree(acts);
 960                 return ERR_PTR(error);
 961         }
 962
 963         return acts;
 964 }
 965
 966 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
 967 {
 968         struct nlattr **a = info->attrs;
 969         struct ovs_header *ovs_header = info->userhdr;
 970         struct sw_flow_key key;
 971         struct sw_flow *flow;
 972         struct sw_flow_mask mask;
 973         struct sk_buff *reply = NULL;
 974         struct datapath *dp;
 975         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
 976         struct sw_flow_match match;
 977         int error;
 978
 979         /* Extract key. */
 980         error = -EINVAL;
 981         if (!a[OVS_FLOW_ATTR_KEY])
 982                 goto error;
 983
 984         ovs_match_init(&match, &key, &mask);
 985         error = ovs_nla_get_match(&match,
 986                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 987         if (error)
 988                 goto error;
 989
 990         /* Validate actions. */
 991         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 992                 acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask);
 993                 if (IS_ERR(acts)) {
 994                         error = PTR_ERR(acts);
 995                         goto error;
 996                 }
 997         }
 998
 999         /* Can allocate before locking if have acts. */
1000         if (acts) {
1001                 reply = ovs_flow_cmd_alloc_info(acts, info, false);
1002                 if (IS_ERR(reply)) {
1003                         error = PTR_ERR(reply);
1004                         goto err_kfree_acts;
1005                 }
1006         }
1007
1008         ovs_lock();
1009         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1010         if (unlikely(!dp)) {
1011                 error = -ENODEV;
1012                 goto err_unlock_ovs;
1013         }
1014         /* Check that the flow exists. */
1015         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1016         if (unlikely(!flow)) {
1017                 error = -ENOENT;
1018                 goto err_unlock_ovs;
1019         }
1020
1021         /* Update actions, if present. */
1022         if (likely(acts)) {
1023                 old_acts = ovsl_dereference(flow->sf_acts);
1024                 rcu_assign_pointer(flow->sf_acts, acts);
1025
1026                 if (unlikely(reply)) {
1027                         error = ovs_flow_cmd_fill_info(flow,
1028                                                        ovs_header->dp_ifindex,
1029                                                        reply, info->snd_portid,
1030                                                        info->snd_seq, 0,
1031                                                        OVS_FLOW_CMD_NEW);
1032                         BUG_ON(error < 0);
1033                 }
1034         } else {
1035                 /* Could not alloc without acts before locking. */
1036                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1037                                                 info, OVS_FLOW_CMD_NEW, false);
1038                 if (unlikely(IS_ERR(reply))) {
1039                         error = PTR_ERR(reply);
1040                         goto err_unlock_ovs;
1041                 }
1042         }
1043
1044         /* Clear stats. */
1045         if (a[OVS_FLOW_ATTR_CLEAR])
1046                 ovs_flow_stats_clear(flow);
1047         ovs_unlock();
1048
1049         if (reply)
1050                 ovs_notify(&dp_flow_genl_family, reply, info);
1051         if (old_acts)
1052                 ovs_nla_free_flow_actions(old_acts);
1053
1054         return 0;
1055
1056 err_unlock_ovs:
1057         ovs_unlock();
1058         kfree_skb(reply);
1059 err_kfree_acts:
1060         kfree(acts);
1061 error:
1062         return error;
1063 }
1064
1065 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1066 {
1067         struct nlattr **a = info->attrs;
1068         struct ovs_header *ovs_header = info->userhdr;
1069         struct sw_flow_key key;
1070         struct sk_buff *reply;
1071         struct sw_flow *flow;
1072         struct datapath *dp;
1073         struct sw_flow_match match;
1074         int err;
1075
1076         if (!a[OVS_FLOW_ATTR_KEY]) {
1077                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1078                 return -EINVAL;
1079         }
1080
1081         ovs_match_init(&match, &key, NULL);
1082         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1083         if (err)
1084                 return err;
1085
1086         ovs_lock();
1087         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1088         if (!dp) {
1089                 err = -ENODEV;
1090                 goto unlock;
1091         }
1092
1093         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1094         if (!flow) {
1095                 err = -ENOENT;
1096                 goto unlock;
1097         }
1098
1099         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1100                                         OVS_FLOW_CMD_NEW, true);
1101         if (IS_ERR(reply)) {
1102                 err = PTR_ERR(reply);
1103                 goto unlock;
1104         }
1105
1106         ovs_unlock();
1107         return genlmsg_reply(reply, info);
1108 unlock:
1109         ovs_unlock();
1110         return err;
1111 }
1112
1113 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1114 {
1115         struct nlattr **a = info->attrs;
1116         struct ovs_header *ovs_header = info->userhdr;
1117         struct sw_flow_key key;
1118         struct sk_buff *reply;
1119         struct sw_flow *flow;
1120         struct datapath *dp;
1121         struct sw_flow_match match;
1122         int err;
1123
1124         if (likely(a[OVS_FLOW_ATTR_KEY])) {
1125                 ovs_match_init(&match, &key, NULL);
1126                 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1127                 if (unlikely(err))
1128                         return err;
1129         }
1130
1131         ovs_lock();
1132         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1133         if (unlikely(!dp)) {
1134                 err = -ENODEV;
1135                 goto unlock;
1136         }
1137
1138         if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1139                 err = ovs_flow_tbl_flush(&dp->table);
1140                 goto unlock;
1141         }
1142
1143         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1144         if (unlikely(!flow)) {
1145                 err = -ENOENT;
1146                 goto unlock;
1147         }
1148
1149         ovs_flow_tbl_remove(&dp->table, flow);
1150         ovs_unlock();
1151
1152         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1153                                         info, false);
1154         if (likely(reply)) {
1155                 if (likely(!IS_ERR(reply))) {
1156                         rcu_read_lock();        /*To keep RCU checker happy. */
1157                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1158                                                      reply, info->snd_portid,
1159                                                      info->snd_seq, 0,
1160                                                      OVS_FLOW_CMD_DEL);
1161                         rcu_read_unlock();
1162                         BUG_ON(err < 0);
1163
1164                         ovs_notify(&dp_flow_genl_family, reply, info);
1165                 } else {
1166                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1167                 }
1168         }
1169
1170         ovs_flow_free(flow, true);
1171         return 0;
1172 unlock:
1173         ovs_unlock();
1174         return err;
1175 }
1176
1177 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1178 {
1179         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1180         struct table_instance *ti;
1181         struct datapath *dp;
1182
1183         rcu_read_lock();
1184         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1185         if (!dp) {
1186                 rcu_read_unlock();
1187                 return -ENODEV;
1188         }
1189
1190         ti = rcu_dereference(dp->table.ti);
1191         for (;;) {
1192                 struct sw_flow *flow;
1193                 u32 bucket, obj;
1194
1195                 bucket = cb->args[0];
1196                 obj = cb->args[1];
1197                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1198                 if (!flow)
1199                         break;
1200
1201                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1202                                            NETLINK_CB(cb->skb).portid,
1203                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1204                                            OVS_FLOW_CMD_NEW) < 0)
1205                         break;
1206
1207                 cb->args[0] = bucket;
1208                 cb->args[1] = obj;
1209         }
1210         rcu_read_unlock();
1211         return skb->len;
1212 }
1213
1214 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1215         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1216         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1217         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1218 };
1219
1220 static const struct genl_ops dp_flow_genl_ops[] = {
1221         { .cmd = OVS_FLOW_CMD_NEW,
1222           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1223           .policy = flow_policy,
1224           .doit = ovs_flow_cmd_new
1225         },
1226         { .cmd = OVS_FLOW_CMD_DEL,
1227           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1228           .policy = flow_policy,
1229           .doit = ovs_flow_cmd_del
1230         },
1231         { .cmd = OVS_FLOW_CMD_GET,
1232           .flags = 0,               /* OK for unprivileged users. */
1233           .policy = flow_policy,
1234           .doit = ovs_flow_cmd_get,
1235           .dumpit = ovs_flow_cmd_dump
1236         },
1237         { .cmd = OVS_FLOW_CMD_SET,
1238           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1239           .policy = flow_policy,
1240           .doit = ovs_flow_cmd_set,
1241         },
1242 };
1243
1244 static struct genl_family dp_flow_genl_family = {
1245         .id = GENL_ID_GENERATE,
1246         .hdrsize = sizeof(struct ovs_header),
1247         .name = OVS_FLOW_FAMILY,
1248         .version = OVS_FLOW_VERSION,
1249         .maxattr = OVS_FLOW_ATTR_MAX,
1250         .netnsok = true,
1251         .parallel_ops = true,
1252         .ops = dp_flow_genl_ops,
1253         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1254         .mcgrps = &ovs_dp_flow_multicast_group,
1255         .n_mcgrps = 1,
1256 };
1257
1258 static size_t ovs_dp_cmd_msg_size(void)
1259 {
1260         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1261
1262         msgsize += nla_total_size(IFNAMSIZ);
1263         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1264         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1265         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1266
1267         return msgsize;
1268 }
1269
1270 /* Called with ovs_mutex or RCU read lock. */
1271 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1272                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1273 {
1274         struct ovs_header *ovs_header;
1275         struct ovs_dp_stats dp_stats;
1276         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1277         int err;
1278
1279         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1280                                    flags, cmd);
1281         if (!ovs_header)
1282                 goto error;
1283
1284         ovs_header->dp_ifindex = get_dpifindex(dp);
1285
1286         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1287         if (err)
1288                 goto nla_put_failure;
1289
1290         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1291         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1292                         &dp_stats))
1293                 goto nla_put_failure;
1294
1295         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1296                         sizeof(struct ovs_dp_megaflow_stats),
1297                         &dp_megaflow_stats))
1298                 goto nla_put_failure;
1299
1300         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1301                 goto nla_put_failure;
1302
1303         return genlmsg_end(skb, ovs_header);
1304
1305 nla_put_failure:
1306         genlmsg_cancel(skb, ovs_header);
1307 error:
1308         return -EMSGSIZE;
1309 }
1310
1311 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1312 {
1313         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1314 }
1315
1316 /* Called with rcu_read_lock or ovs_mutex. */
1317 static struct datapath *lookup_datapath(struct net *net,
1318                                         struct ovs_header *ovs_header,
1319                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1320 {
1321         struct datapath *dp;
1322
1323         if (!a[OVS_DP_ATTR_NAME])
1324                 dp = get_dp(net, ovs_header->dp_ifindex);
1325         else {
1326                 struct vport *vport;
1327
1328                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1329                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1330         }
1331         return dp ? dp : ERR_PTR(-ENODEV);
1332 }
1333
1334 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1335 {
1336         struct datapath *dp;
1337
1338         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1339         if (IS_ERR(dp))
1340                 return;
1341
1342         WARN(dp->user_features, "Dropping previously announced user features\n");
1343         dp->user_features = 0;
1344 }
1345
1346 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1347 {
1348         if (a[OVS_DP_ATTR_USER_FEATURES])
1349                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1350 }
1351
1352 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1353 {
1354         struct nlattr **a = info->attrs;
1355         struct vport_parms parms;
1356         struct sk_buff *reply;
1357         struct datapath *dp;
1358         struct vport *vport;
1359         struct ovs_net *ovs_net;
1360         int err, i;
1361
1362         err = -EINVAL;
1363         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1364                 goto err;
1365
1366         reply = ovs_dp_cmd_alloc_info(info);
1367         if (!reply)
1368                 return -ENOMEM;
1369
1370         err = -ENOMEM;
1371         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1372         if (dp == NULL)
1373                 goto err_free_reply;
1374
1375         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1376
1377         /* Allocate table. */
1378         err = ovs_flow_tbl_init(&dp->table);
1379         if (err)
1380                 goto err_free_dp;
1381
1382         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1383         if (!dp->stats_percpu) {
1384                 err = -ENOMEM;
1385                 goto err_destroy_table;
1386         }
1387
1388         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1389                             GFP_KERNEL);
1390         if (!dp->ports) {
1391                 err = -ENOMEM;
1392                 goto err_destroy_percpu;
1393         }
1394
1395         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1396                 INIT_HLIST_HEAD(&dp->ports[i]);
1397
1398         /* Set up our datapath device. */
1399         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1400         parms.type = OVS_VPORT_TYPE_INTERNAL;
1401         parms.options = NULL;
1402         parms.dp = dp;
1403         parms.port_no = OVSP_LOCAL;
1404         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1405
1406         ovs_dp_change(dp, a);
1407
1408         /* So far only local changes have been made, now need the lock. */
1409         ovs_lock();
1410
1411         vport = new_vport(&parms);
1412         if (IS_ERR(vport)) {
1413                 err = PTR_ERR(vport);
1414                 if (err == -EBUSY)
1415                         err = -EEXIST;
1416
1417                 if (err == -EEXIST) {
1418                         /* An outdated user space instance that does not understand
1419                          * the concept of user_features has attempted to create a new
1420                          * datapath and is likely to reuse it. Drop all user features.
1421                          */
1422                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1423                                 ovs_dp_reset_user_features(skb, info);
1424                 }
1425
1426                 goto err_destroy_ports_array;
1427         }
1428
1429         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1430                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1431         BUG_ON(err < 0);
1432
1433         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1434         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1435
1436         ovs_unlock();
1437
1438         ovs_notify(&dp_datapath_genl_family, reply, info);
1439         return 0;
1440
1441 err_destroy_ports_array:
1442         ovs_unlock();
1443         kfree(dp->ports);
1444 err_destroy_percpu:
1445         free_percpu(dp->stats_percpu);
1446 err_destroy_table:
1447         ovs_flow_tbl_destroy(&dp->table, false);
1448 err_free_dp:
1449         release_net(ovs_dp_get_net(dp));
1450         kfree(dp);
1451 err_free_reply:
1452         kfree_skb(reply);
1453 err:
1454         return err;
1455 }
1456
1457 /* Called with ovs_mutex. */
1458 static void __dp_destroy(struct datapath *dp)
1459 {
1460         int i;
1461
1462         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1463                 struct vport *vport;
1464                 struct hlist_node *n;
1465
1466                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1467                         if (vport->port_no != OVSP_LOCAL)
1468                                 ovs_dp_detach_port(vport);
1469         }
1470
1471         list_del_rcu(&dp->list_node);
1472
1473         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1474          * all ports in datapath are destroyed first before freeing datapath.
1475          */
1476         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1477
1478         /* RCU destroy the flow table */
1479         ovs_flow_tbl_destroy(&dp->table, true);
1480
1481         call_rcu(&dp->rcu, destroy_dp_rcu);
1482 }
1483
1484 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1485 {
1486         struct sk_buff *reply;
1487         struct datapath *dp;
1488         int err;
1489
1490         reply = ovs_dp_cmd_alloc_info(info);
1491         if (!reply)
1492                 return -ENOMEM;
1493
1494         ovs_lock();
1495         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1496         err = PTR_ERR(dp);
1497         if (IS_ERR(dp))
1498                 goto err_unlock_free;
1499
1500         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1501                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1502         BUG_ON(err < 0);
1503
1504         __dp_destroy(dp);
1505         ovs_unlock();
1506
1507         ovs_notify(&dp_datapath_genl_family, reply, info);
1508
1509         return 0;
1510
1511 err_unlock_free:
1512         ovs_unlock();
1513         kfree_skb(reply);
1514         return err;
1515 }
1516
1517 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1518 {
1519         struct sk_buff *reply;
1520         struct datapath *dp;
1521         int err;
1522
1523         reply = ovs_dp_cmd_alloc_info(info);
1524         if (!reply)
1525                 return -ENOMEM;
1526
1527         ovs_lock();
1528         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1529         err = PTR_ERR(dp);
1530         if (IS_ERR(dp))
1531                 goto err_unlock_free;
1532
1533         ovs_dp_change(dp, info->attrs);
1534
1535         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1536                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1537         BUG_ON(err < 0);
1538
1539         ovs_unlock();
1540         ovs_notify(&dp_datapath_genl_family, reply, info);
1541
1542         return 0;
1543
1544 err_unlock_free:
1545         ovs_unlock();
1546         kfree_skb(reply);
1547         return err;
1548 }
1549
1550 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1551 {
1552         struct sk_buff *reply;
1553         struct datapath *dp;
1554         int err;
1555
1556         reply = ovs_dp_cmd_alloc_info(info);
1557         if (!reply)
1558                 return -ENOMEM;
1559
1560         rcu_read_lock();
1561         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1562         if (IS_ERR(dp)) {
1563                 err = PTR_ERR(dp);
1564                 goto err_unlock_free;
1565         }
1566         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1567                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1568         BUG_ON(err < 0);
1569         rcu_read_unlock();
1570
1571         return genlmsg_reply(reply, info);
1572
1573 err_unlock_free:
1574         rcu_read_unlock();
1575         kfree_skb(reply);
1576         return err;
1577 }
1578
1579 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1580 {
1581         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1582         struct datapath *dp;
1583         int skip = cb->args[0];
1584         int i = 0;
1585
1586         rcu_read_lock();
1587         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1588                 if (i >= skip &&
1589                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1590                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1591                                          OVS_DP_CMD_NEW) < 0)
1592                         break;
1593                 i++;
1594         }
1595         rcu_read_unlock();
1596
1597         cb->args[0] = i;
1598
1599         return skb->len;
1600 }
1601
1602 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1603         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1604         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1605         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1606 };
1607
1608 static const struct genl_ops dp_datapath_genl_ops[] = {
1609         { .cmd = OVS_DP_CMD_NEW,
1610           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1611           .policy = datapath_policy,
1612           .doit = ovs_dp_cmd_new
1613         },
1614         { .cmd = OVS_DP_CMD_DEL,
1615           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1616           .policy = datapath_policy,
1617           .doit = ovs_dp_cmd_del
1618         },
1619         { .cmd = OVS_DP_CMD_GET,
1620           .flags = 0,               /* OK for unprivileged users. */
1621           .policy = datapath_policy,
1622           .doit = ovs_dp_cmd_get,
1623           .dumpit = ovs_dp_cmd_dump
1624         },
1625         { .cmd = OVS_DP_CMD_SET,
1626           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1627           .policy = datapath_policy,
1628           .doit = ovs_dp_cmd_set,
1629         },
1630 };
1631
1632 static struct genl_family dp_datapath_genl_family = {
1633         .id = GENL_ID_GENERATE,
1634         .hdrsize = sizeof(struct ovs_header),
1635         .name = OVS_DATAPATH_FAMILY,
1636         .version = OVS_DATAPATH_VERSION,
1637         .maxattr = OVS_DP_ATTR_MAX,
1638         .netnsok = true,
1639         .parallel_ops = true,
1640         .ops = dp_datapath_genl_ops,
1641         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1642         .mcgrps = &ovs_dp_datapath_multicast_group,
1643         .n_mcgrps = 1,
1644 };
1645
1646 /* Called with ovs_mutex or RCU read lock. */
1647 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1648                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1649 {
1650         struct ovs_header *ovs_header;
1651         struct ovs_vport_stats vport_stats;
1652         int err;
1653
1654         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1655                                  flags, cmd);
1656         if (!ovs_header)
1657                 return -EMSGSIZE;
1658
1659         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1660
1661         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1662             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1663             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1664                            vport->ops->get_name(vport)))
1665                 goto nla_put_failure;
1666
1667         ovs_vport_get_stats(vport, &vport_stats);
1668         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1669                     &vport_stats))
1670                 goto nla_put_failure;
1671
1672         if (ovs_vport_get_upcall_portids(vport, skb))
1673                 goto nla_put_failure;
1674
1675         err = ovs_vport_get_options(vport, skb);
1676         if (err == -EMSGSIZE)
1677                 goto error;
1678
1679         return genlmsg_end(skb, ovs_header);
1680
1681 nla_put_failure:
1682         err = -EMSGSIZE;
1683 error:
1684         genlmsg_cancel(skb, ovs_header);
1685         return err;
1686 }
1687
1688 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1689 {
1690         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1691 }
1692
1693 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1694 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1695                                          u32 seq, u8 cmd)
1696 {
1697         struct sk_buff *skb;
1698         int retval;
1699
1700         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1701         if (!skb)
1702                 return ERR_PTR(-ENOMEM);
1703
1704         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1705         BUG_ON(retval < 0);
1706
1707         return skb;
1708 }
1709
1710 /* Called with ovs_mutex or RCU read lock. */
1711 static struct vport *lookup_vport(struct net *net,
1712                                   struct ovs_header *ovs_header,
1713                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1714 {
1715         struct datapath *dp;
1716         struct vport *vport;
1717
1718         if (a[OVS_VPORT_ATTR_NAME]) {
1719                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1720                 if (!vport)
1721                         return ERR_PTR(-ENODEV);
1722                 if (ovs_header->dp_ifindex &&
1723                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1724                         return ERR_PTR(-ENODEV);
1725                 return vport;
1726         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1727                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1728
1729                 if (port_no >= DP_MAX_PORTS)
1730                         return ERR_PTR(-EFBIG);
1731
1732                 dp = get_dp(net, ovs_header->dp_ifindex);
1733                 if (!dp)
1734                         return ERR_PTR(-ENODEV);
1735
1736                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1737                 if (!vport)
1738                         return ERR_PTR(-ENODEV);
1739                 return vport;
1740         } else
1741                 return ERR_PTR(-EINVAL);
1742 }
1743
1744 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1745 {
1746         struct nlattr **a = info->attrs;
1747         struct ovs_header *ovs_header = info->userhdr;
1748         struct vport_parms parms;
1749         struct sk_buff *reply;
1750         struct vport *vport;
1751         struct datapath *dp;
1752         u32 port_no;
1753         int err;
1754
1755         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1756             !a[OVS_VPORT_ATTR_UPCALL_PID])
1757                 return -EINVAL;
1758
1759         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1760                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1761         if (port_no >= DP_MAX_PORTS)
1762                 return -EFBIG;
1763
1764         reply = ovs_vport_cmd_alloc_info();
1765         if (!reply)
1766                 return -ENOMEM;
1767
1768         ovs_lock();
1769 restart:
1770         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1771         err = -ENODEV;
1772         if (!dp)
1773                 goto exit_unlock_free;
1774
1775         if (port_no) {
1776                 vport = ovs_vport_ovsl(dp, port_no);
1777                 err = -EBUSY;
1778                 if (vport)
1779                         goto exit_unlock_free;
1780         } else {
1781                 for (port_no = 1; ; port_no++) {
1782                         if (port_no >= DP_MAX_PORTS) {
1783                                 err = -EFBIG;
1784                                 goto exit_unlock_free;
1785                         }
1786                         vport = ovs_vport_ovsl(dp, port_no);
1787                         if (!vport)
1788                                 break;
1789                 }
1790         }
1791
1792         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1793         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1794         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1795         parms.dp = dp;
1796         parms.port_no = port_no;
1797         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1798
1799         vport = new_vport(&parms);
1800         err = PTR_ERR(vport);
1801         if (IS_ERR(vport)) {
1802                 if (err == -EAGAIN)
1803                         goto restart;
1804                 goto exit_unlock_free;
1805         }
1806
1807         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1808                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1809         BUG_ON(err < 0);
1810         ovs_unlock();
1811
1812         ovs_notify(&dp_vport_genl_family, reply, info);
1813         return 0;
1814
1815 exit_unlock_free:
1816         ovs_unlock();
1817         kfree_skb(reply);
1818         return err;
1819 }
1820
1821 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1822 {
1823         struct nlattr **a = info->attrs;
1824         struct sk_buff *reply;
1825         struct vport *vport;
1826         int err;
1827
1828         reply = ovs_vport_cmd_alloc_info();
1829         if (!reply)
1830                 return -ENOMEM;
1831
1832         ovs_lock();
1833         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1834         err = PTR_ERR(vport);
1835         if (IS_ERR(vport))
1836                 goto exit_unlock_free;
1837
1838         if (a[OVS_VPORT_ATTR_TYPE] &&
1839             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1840                 err = -EINVAL;
1841                 goto exit_unlock_free;
1842         }
1843
1844         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1845                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1846                 if (err)
1847                         goto exit_unlock_free;
1848         }
1849
1850
1851         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1852                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
1853
1854                 err = ovs_vport_set_upcall_portids(vport, ids);
1855                 if (err)
1856                         goto exit_unlock_free;
1857         }
1858
1859         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1860                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1861         BUG_ON(err < 0);
1862
1863         ovs_unlock();
1864         ovs_notify(&dp_vport_genl_family, reply, info);
1865         return 0;
1866
1867 exit_unlock_free:
1868         ovs_unlock();
1869         kfree_skb(reply);
1870         return err;
1871 }
1872
1873 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1874 {
1875         struct nlattr **a = info->attrs;
1876         struct sk_buff *reply;
1877         struct vport *vport;
1878         int err;
1879
1880         reply = ovs_vport_cmd_alloc_info();
1881         if (!reply)
1882                 return -ENOMEM;
1883
1884         ovs_lock();
1885         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1886         err = PTR_ERR(vport);
1887         if (IS_ERR(vport))
1888                 goto exit_unlock_free;
1889
1890         if (vport->port_no == OVSP_LOCAL) {
1891                 err = -EINVAL;
1892                 goto exit_unlock_free;
1893         }
1894
1895         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1896                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1897         BUG_ON(err < 0);
1898         ovs_dp_detach_port(vport);
1899         ovs_unlock();
1900
1901         ovs_notify(&dp_vport_genl_family, reply, info);
1902         return 0;
1903
1904 exit_unlock_free:
1905         ovs_unlock();
1906         kfree_skb(reply);
1907         return err;
1908 }
1909
1910 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1911 {
1912         struct nlattr **a = info->attrs;
1913         struct ovs_header *ovs_header = info->userhdr;
1914         struct sk_buff *reply;
1915         struct vport *vport;
1916         int err;
1917
1918         reply = ovs_vport_cmd_alloc_info();
1919         if (!reply)
1920                 return -ENOMEM;
1921
1922         rcu_read_lock();
1923         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1924         err = PTR_ERR(vport);
1925         if (IS_ERR(vport))
1926                 goto exit_unlock_free;
1927         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1928                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1929         BUG_ON(err < 0);
1930         rcu_read_unlock();
1931
1932         return genlmsg_reply(reply, info);
1933
1934 exit_unlock_free:
1935         rcu_read_unlock();
1936         kfree_skb(reply);
1937         return err;
1938 }
1939
1940 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1941 {
1942         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1943         struct datapath *dp;
1944         int bucket = cb->args[0], skip = cb->args[1];
1945         int i, j = 0;
1946
1947         rcu_read_lock();
1948         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1949         if (!dp) {
1950                 rcu_read_unlock();
1951                 return -ENODEV;
1952         }
1953         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1954                 struct vport *vport;
1955
1956                 j = 0;
1957                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1958                         if (j >= skip &&
1959                             ovs_vport_cmd_fill_info(vport, skb,
1960                                                     NETLINK_CB(cb->skb).portid,
1961                                                     cb->nlh->nlmsg_seq,
1962                                                     NLM_F_MULTI,
1963                                                     OVS_VPORT_CMD_NEW) < 0)
1964                                 goto out;
1965
1966                         j++;
1967                 }
1968                 skip = 0;
1969         }
1970 out:
1971         rcu_read_unlock();
1972
1973         cb->args[0] = i;
1974         cb->args[1] = j;
1975
1976         return skb->len;
1977 }
1978
1979 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1980         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1981         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1982         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1983         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1984         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1985         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1986 };
1987
1988 static const struct genl_ops dp_vport_genl_ops[] = {
1989         { .cmd = OVS_VPORT_CMD_NEW,
1990           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1991           .policy = vport_policy,
1992           .doit = ovs_vport_cmd_new
1993         },
1994         { .cmd = OVS_VPORT_CMD_DEL,
1995           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1996           .policy = vport_policy,
1997           .doit = ovs_vport_cmd_del
1998         },
1999         { .cmd = OVS_VPORT_CMD_GET,
2000           .flags = 0,               /* OK for unprivileged users. */
2001           .policy = vport_policy,
2002           .doit = ovs_vport_cmd_get,
2003           .dumpit = ovs_vport_cmd_dump
2004         },
2005         { .cmd = OVS_VPORT_CMD_SET,
2006           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2007           .policy = vport_policy,
2008           .doit = ovs_vport_cmd_set,
2009         },
2010 };
2011
2012 struct genl_family dp_vport_genl_family = {
2013         .id = GENL_ID_GENERATE,
2014         .hdrsize = sizeof(struct ovs_header),
2015         .name = OVS_VPORT_FAMILY,
2016         .version = OVS_VPORT_VERSION,
2017         .maxattr = OVS_VPORT_ATTR_MAX,
2018         .netnsok = true,
2019         .parallel_ops = true,
2020         .ops = dp_vport_genl_ops,
2021         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2022         .mcgrps = &ovs_dp_vport_multicast_group,
2023         .n_mcgrps = 1,
2024 };
2025
2026 static struct genl_family * const dp_genl_families[] = {
2027         &dp_datapath_genl_family,
2028         &dp_vport_genl_family,
2029         &dp_flow_genl_family,
2030         &dp_packet_genl_family,
2031 };
2032
2033 static void dp_unregister_genl(int n_families)
2034 {
2035         int i;
2036
2037         for (i = 0; i < n_families; i++)
2038                 genl_unregister_family(dp_genl_families[i]);
2039 }
2040
2041 static int dp_register_genl(void)
2042 {
2043         int err;
2044         int i;
2045
2046         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2047
2048                 err = genl_register_family(dp_genl_families[i]);
2049                 if (err)
2050                         goto error;
2051         }
2052
2053         return 0;
2054
2055 error:
2056         dp_unregister_genl(i);
2057         return err;
2058 }
2059
2060 static int __net_init ovs_init_net(struct net *net)
2061 {
2062         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2063
2064         INIT_LIST_HEAD(&ovs_net->dps);
2065         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2066         return 0;
2067 }
2068
2069 static void __net_exit ovs_exit_net(struct net *net)
2070 {
2071         struct datapath *dp, *dp_next;
2072         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2073
2074         ovs_lock();
2075         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2076                 __dp_destroy(dp);
2077         ovs_unlock();
2078
2079         cancel_work_sync(&ovs_net->dp_notify_work);
2080 }
2081
2082 static struct pernet_operations ovs_net_ops = {
2083         .init = ovs_init_net,
2084         .exit = ovs_exit_net,
2085         .id   = &ovs_net_id,
2086         .size = sizeof(struct ovs_net),
2087 };
2088
2089 static int __init dp_init(void)
2090 {
2091         int err;
2092
2093         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2094
2095         pr_info("Open vSwitch switching datapath\n");
2096
2097         err = action_fifos_init();
2098         if (err)
2099                 goto error;
2100
2101         err = ovs_internal_dev_rtnl_link_register();
2102         if (err)
2103                 goto error_action_fifos_exit;
2104
2105         err = ovs_flow_init();
2106         if (err)
2107                 goto error_unreg_rtnl_link;
2108
2109         err = ovs_vport_init();
2110         if (err)
2111                 goto error_flow_exit;
2112
2113         err = register_pernet_device(&ovs_net_ops);
2114         if (err)
2115                 goto error_vport_exit;
2116
2117         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2118         if (err)
2119                 goto error_netns_exit;
2120
2121         err = ovs_netdev_init();
2122         if (err)
2123                 goto error_unreg_notifier;
2124
2125         err = dp_register_genl();
2126         if (err < 0)
2127                 goto error_unreg_netdev;
2128
2129         return 0;
2130
2131 error_unreg_netdev:
2132         ovs_netdev_exit();
2133 error_unreg_notifier:
2134         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2135 error_netns_exit:
2136         unregister_pernet_device(&ovs_net_ops);
2137 error_vport_exit:
2138         ovs_vport_exit();
2139 error_flow_exit:
2140         ovs_flow_exit();
2141 error_unreg_rtnl_link:
2142         ovs_internal_dev_rtnl_link_unregister();
2143 error_action_fifos_exit:
2144         action_fifos_exit();
2145 error:
2146         return err;
2147 }
2148
2149 static void dp_cleanup(void)
2150 {
2151         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2152         ovs_netdev_exit();
2153         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2154         unregister_pernet_device(&ovs_net_ops);
2155         rcu_barrier();
2156         ovs_vport_exit();
2157         ovs_flow_exit();
2158         ovs_internal_dev_rtnl_link_unregister();
2159         action_fifos_exit();
2160 }
2161
2162 module_init(dp_init);
2163 module_exit(dp_cleanup);
2164
2165 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2166 MODULE_LICENSE("GPL");