net/openvswitch/datapath.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2007-2014 Nicira, Inc.
   4  */
   5
   6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   7
   8 #include <linux/init.h>
   9 #include <linux/module.h>
  10 #include <linux/if_arp.h>
  11 #include <linux/if_vlan.h>
  12 #include <linux/in.h>
  13 #include <linux/ip.h>
  14 #include <linux/jhash.h>
  15 #include <linux/delay.h>
  16 #include <linux/time.h>
  17 #include <linux/etherdevice.h>
  18 #include <linux/genetlink.h>
  19 #include <linux/kernel.h>
  20 #include <linux/kthread.h>
  21 #include <linux/mutex.h>
  22 #include <linux/percpu.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/tcp.h>
  25 #include <linux/udp.h>
  26 #include <linux/ethtool.h>
  27 #include <linux/wait.h>
  28 #include <asm/div64.h>
  29 #include <linux/highmem.h>
  30 #include <linux/netfilter_bridge.h>
  31 #include <linux/netfilter_ipv4.h>
  32 #include <linux/inetdevice.h>
  33 #include <linux/list.h>
  34 #include <linux/openvswitch.h>
  35 #include <linux/rculist.h>
  36 #include <linux/dmi.h>
  37 #include <net/genetlink.h>
  38 #include <net/net_namespace.h>
  39 #include <net/netns/generic.h>
  40
  41 #include "datapath.h"
  42 #include "flow.h"
  43 #include "flow_table.h"
  44 #include "flow_netlink.h"
  45 #include "meter.h"
  46 #include "vport-internal_dev.h"
  47 #include "vport-netdev.h"
  48
  49 unsigned int ovs_net_id __read_mostly;
  50
  51 static struct genl_family dp_packet_genl_family;
  52 static struct genl_family dp_flow_genl_family;
  53 static struct genl_family dp_datapath_genl_family;
  54
  55 static const struct nla_policy flow_policy[];
  56
  57 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
  58         .name = OVS_FLOW_MCGROUP,
  59 };
  60
  61 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  62         .name = OVS_DATAPATH_MCGROUP,
  63 };
  64
  65 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
  66         .name = OVS_VPORT_MCGROUP,
  67 };
  68
  69 /* Check if need to build a reply message.
  70  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  71 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  72                             unsigned int group)
  73 {
  74         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  75                genl_has_listeners(family, genl_info_net(info), group);
  76 }
  77
  78 static void ovs_notify(struct genl_family *family,
  79                        struct sk_buff *skb, struct genl_info *info)
  80 {
  81         genl_notify(family, skb, info, 0, GFP_KERNEL);
  82 }
  83
  84 /**
  85  * DOC: Locking:
  86  *
  87  * All writes e.g. Writes to device state (add/remove datapath, port, set
  88  * operations on vports, etc.), Writes to other state (flow table
  89  * modifications, set miscellaneous datapath parameters, etc.) are protected
  90  * by ovs_lock.
  91  *
  92  * Reads are protected by RCU.
  93  *
  94  * There are a few special cases (mostly stats) that have their own
  95  * synchronization but they nest under all of above and don't interact with
  96  * each other.
  97  *
  98  * The RTNL lock nests inside ovs_mutex.
  99  */
 100
 101 static DEFINE_MUTEX(ovs_mutex);
 102
 103 void ovs_lock(void)
 104 {
 105         mutex_lock(&ovs_mutex);
 106 }
 107
 108 void ovs_unlock(void)
 109 {
 110         mutex_unlock(&ovs_mutex);
 111 }
 112
 113 #ifdef CONFIG_LOCKDEP
 114 int lockdep_ovsl_is_held(void)
 115 {
 116         if (debug_locks)
 117                 return lockdep_is_held(&ovs_mutex);
 118         else
 119                 return 1;
 120 }
 121 #endif
 122
 123 static struct vport *new_vport(const struct vport_parms *);
 124 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 125                              const struct sw_flow_key *,
 126                              const struct dp_upcall_info *,
 127                              uint32_t cutlen);
 128 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 129                                   const struct sw_flow_key *,
 130                                   const struct dp_upcall_info *,
 131                                   uint32_t cutlen);
 132
 133 /* Must be called with rcu_read_lock or ovs_mutex. */
 134 const char *ovs_dp_name(const struct datapath *dp)
 135 {
 136         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 137         return ovs_vport_name(vport);
 138 }
 139
 140 static int get_dpifindex(const struct datapath *dp)
 141 {
 142         struct vport *local;
 143         int ifindex;
 144
 145         rcu_read_lock();
 146
 147         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 148         if (local)
 149                 ifindex = local->dev->ifindex;
 150         else
 151                 ifindex = 0;
 152
 153         rcu_read_unlock();
 154
 155         return ifindex;
 156 }
 157
 158 static void destroy_dp_rcu(struct rcu_head *rcu)
 159 {
 160         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 161
 162         ovs_flow_tbl_destroy(&dp->table);
 163         free_percpu(dp->stats_percpu);
 164         kfree(dp->ports);
 165         ovs_meters_exit(dp);
 166         kfree(dp);
 167 }
 168
 169 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 170                                             u16 port_no)
 171 {
 172         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 173 }
 174
 175 /* Called with ovs_mutex or RCU read lock. */
 176 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 177 {
 178         struct vport *vport;
 179         struct hlist_head *head;
 180
 181         head = vport_hash_bucket(dp, port_no);
 182         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 183                 if (vport->port_no == port_no)
 184                         return vport;
 185         }
 186         return NULL;
 187 }
 188
 189 /* Called with ovs_mutex. */
 190 static struct vport *new_vport(const struct vport_parms *parms)
 191 {
 192         struct vport *vport;
 193
 194         vport = ovs_vport_add(parms);
 195         if (!IS_ERR(vport)) {
 196                 struct datapath *dp = parms->dp;
 197                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 198
 199                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 200         }
 201         return vport;
 202 }
 203
 204 void ovs_dp_detach_port(struct vport *p)
 205 {
 206         ASSERT_OVSL();
 207
 208         /* First drop references to device. */
 209         hlist_del_rcu(&p->dp_hash_node);
 210
 211         /* Then destroy it. */
 212         ovs_vport_del(p);
 213 }
 214
 215 /* Must be called with rcu_read_lock. */
 216 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 217 {
 218         const struct vport *p = OVS_CB(skb)->input_vport;
 219         struct datapath *dp = p->dp;
 220         struct sw_flow *flow;
 221         struct sw_flow_actions *sf_acts;
 222         struct dp_stats_percpu *stats;
 223         u64 *stats_counter;
 224         u32 n_mask_hit;
 225         int error;
 226
 227         stats = this_cpu_ptr(dp->stats_percpu);
 228
 229         /* Look up flow. */
 230         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
 231         if (unlikely(!flow)) {
 232                 struct dp_upcall_info upcall;
 233
 234                 memset(&upcall, 0, sizeof(upcall));
 235                 upcall.cmd = OVS_PACKET_CMD_MISS;
 236                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 237                 upcall.mru = OVS_CB(skb)->mru;
 238                 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
 239                 switch (error) {
 240                 case 0:
 241                 case -EAGAIN:
 242                 case -ERESTARTSYS:
 243                 case -EINTR:
 244                         consume_skb(skb);
 245                         break;
 246                 default:
 247                         kfree_skb(skb);
 248                         break;
 249                 }
 250                 stats_counter = &stats->n_missed;
 251                 goto out;
 252         }
 253
 254         ovs_flow_stats_update(flow, key->tp.flags, skb);
 255         sf_acts = rcu_dereference(flow->sf_acts);
 256         error = ovs_execute_actions(dp, skb, sf_acts, key);
 257         if (unlikely(error))
 258                 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
 259                                                         ovs_dp_name(dp), error);
 260
 261         stats_counter = &stats->n_hit;
 262
 263 out:
 264         /* Update datapath statistics. */
 265         u64_stats_update_begin(&stats->syncp);
 266         (*stats_counter)++;
 267         stats->n_mask_hit += n_mask_hit;
 268         u64_stats_update_end(&stats->syncp);
 269 }
 270
 271 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 272                   const struct sw_flow_key *key,
 273                   const struct dp_upcall_info *upcall_info,
 274                   uint32_t cutlen)
 275 {
 276         struct dp_stats_percpu *stats;
 277         int err;
 278
 279         if (upcall_info->portid == 0) {
 280                 err = -ENOTCONN;
 281                 goto err;
 282         }
 283
 284         if (!skb_is_gso(skb))
 285                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 286         else
 287                 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
 288         if (err)
 289                 goto err;
 290
 291         return 0;
 292
 293 err:
 294         stats = this_cpu_ptr(dp->stats_percpu);
 295
 296         u64_stats_update_begin(&stats->syncp);
 297         stats->n_lost++;
 298         u64_stats_update_end(&stats->syncp);
 299
 300         return err;
 301 }
 302
 303 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 304                              const struct sw_flow_key *key,
 305                              const struct dp_upcall_info *upcall_info,
 306                                  uint32_t cutlen)
 307 {
 308         unsigned int gso_type = skb_shinfo(skb)->gso_type;
 309         struct sw_flow_key later_key;
 310         struct sk_buff *segs, *nskb;
 311         int err;
 312
 313         BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
 314         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 315         if (IS_ERR(segs))
 316                 return PTR_ERR(segs);
 317         if (segs == NULL)
 318                 return -EINVAL;
 319
 320         if (gso_type & SKB_GSO_UDP) {
 321                 /* The initial flow key extracted by ovs_flow_key_extract()
 322                  * in this case is for a first fragment, so we need to
 323                  * properly mark later fragments.
 324                  */
 325                 later_key = *key;
 326                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 327         }
 328
 329         /* Queue all of the segments. */
 330         skb = segs;
 331         do {
 332                 if (gso_type & SKB_GSO_UDP && skb != segs)
 333                         key = &later_key;
 334
 335                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 336                 if (err)
 337                         break;
 338
 339         } while ((skb = skb->next));
 340
 341         /* Free all of the segments. */
 342         skb = segs;
 343         do {
 344                 nskb = skb->next;
 345                 if (err)
 346                         kfree_skb(skb);
 347                 else
 348                         consume_skb(skb);
 349         } while ((skb = nskb));
 350         return err;
 351 }
 352
 353 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 354                               unsigned int hdrlen, int actions_attrlen)
 355 {
 356         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 357                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 358                 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 359                 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
 360
 361         /* OVS_PACKET_ATTR_USERDATA */
 362         if (upcall_info->userdata)
 363                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 364
 365         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 366         if (upcall_info->egress_tun_info)
 367                 size += nla_total_size(ovs_tun_key_attr_size());
 368
 369         /* OVS_PACKET_ATTR_ACTIONS */
 370         if (upcall_info->actions_len)
 371                 size += nla_total_size(actions_attrlen);
 372
 373         /* OVS_PACKET_ATTR_MRU */
 374         if (upcall_info->mru)
 375                 size += nla_total_size(sizeof(upcall_info->mru));
 376
 377         return size;
 378 }
 379
 380 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 381 {
 382         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 383                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 384
 385                 if (plen > 0)
 386                         skb_put_zero(skb, plen);
 387         }
 388 }
 389
 390 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 391                                   const struct sw_flow_key *key,
 392                                   const struct dp_upcall_info *upcall_info,
 393                                   uint32_t cutlen)
 394 {
 395         struct ovs_header *upcall;
 396         struct sk_buff *nskb = NULL;
 397         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 398         struct nlattr *nla;
 399         size_t len;
 400         unsigned int hlen;
 401         int err, dp_ifindex;
 402
 403         dp_ifindex = get_dpifindex(dp);
 404         if (!dp_ifindex)
 405                 return -ENODEV;
 406
 407         if (skb_vlan_tag_present(skb)) {
 408                 nskb = skb_clone(skb, GFP_ATOMIC);
 409                 if (!nskb)
 410                         return -ENOMEM;
 411
 412                 nskb = __vlan_hwaccel_push_inside(nskb);
 413                 if (!nskb)
 414                         return -ENOMEM;
 415
 416                 skb = nskb;
 417         }
 418
 419         if (nla_attr_size(skb->len) > USHRT_MAX) {
 420                 err = -EFBIG;
 421                 goto out;
 422         }
 423
 424         /* Complete checksum if needed */
 425         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 426             (err = skb_csum_hwoffload_help(skb, 0)))
 427                 goto out;
 428
 429         /* Older versions of OVS user space enforce alignment of the last
 430          * Netlink attribute to NLA_ALIGNTO which would require extensive
 431          * padding logic. Only perform zerocopy if padding is not required.
 432          */
 433         if (dp->user_features & OVS_DP_F_UNALIGNED)
 434                 hlen = skb_zerocopy_headlen(skb);
 435         else
 436                 hlen = skb->len;
 437
 438         len = upcall_msg_size(upcall_info, hlen - cutlen,
 439                               OVS_CB(skb)->acts_origlen);
 440         user_skb = genlmsg_new(len, GFP_ATOMIC);
 441         if (!user_skb) {
 442                 err = -ENOMEM;
 443                 goto out;
 444         }
 445
 446         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 447                              0, upcall_info->cmd);
 448         if (!upcall) {
 449                 err = -EINVAL;
 450                 goto out;
 451         }
 452         upcall->dp_ifindex = dp_ifindex;
 453
 454         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 455         if (err)
 456                 goto out;
 457
 458         if (upcall_info->userdata)
 459                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 460                           nla_len(upcall_info->userdata),
 461                           nla_data(upcall_info->userdata));
 462
 463         if (upcall_info->egress_tun_info) {
 464                 nla = nla_nest_start_noflag(user_skb,
 465                                             OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 466                 if (!nla) {
 467                         err = -EMSGSIZE;
 468                         goto out;
 469                 }
 470                 err = ovs_nla_put_tunnel_info(user_skb,
 471                                               upcall_info->egress_tun_info);
 472                 if (err)
 473                         goto out;
 474
 475                 nla_nest_end(user_skb, nla);
 476         }
 477
 478         if (upcall_info->actions_len) {
 479                 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
 480                 if (!nla) {
 481                         err = -EMSGSIZE;
 482                         goto out;
 483                 }
 484                 err = ovs_nla_put_actions(upcall_info->actions,
 485                                           upcall_info->actions_len,
 486                                           user_skb);
 487                 if (!err)
 488                         nla_nest_end(user_skb, nla);
 489                 else
 490                         nla_nest_cancel(user_skb, nla);
 491         }
 492
 493         /* Add OVS_PACKET_ATTR_MRU */
 494         if (upcall_info->mru) {
 495                 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
 496                                 upcall_info->mru)) {
 497                         err = -ENOBUFS;
 498                         goto out;
 499                 }
 500                 pad_packet(dp, user_skb);
 501         }
 502
 503         /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 504         if (cutlen > 0) {
 505                 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
 506                                 skb->len)) {
 507                         err = -ENOBUFS;
 508                         goto out;
 509                 }
 510                 pad_packet(dp, user_skb);
 511         }
 512
 513         /* Only reserve room for attribute header, packet data is added
 514          * in skb_zerocopy() */
 515         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 516                 err = -ENOBUFS;
 517                 goto out;
 518         }
 519         nla->nla_len = nla_attr_size(skb->len - cutlen);
 520
 521         err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 522         if (err)
 523                 goto out;
 524
 525         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 526         pad_packet(dp, user_skb);
 527
 528         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 529
 530         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 531         user_skb = NULL;
 532 out:
 533         if (err)
 534                 skb_tx_error(skb);
 535         consume_skb(user_skb);
 536         consume_skb(nskb);
 537
 538         return err;
 539 }
 540
 541 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 542 {
 543         struct ovs_header *ovs_header = info->userhdr;
 544         struct net *net = sock_net(skb->sk);
 545         struct nlattr **a = info->attrs;
 546         struct sw_flow_actions *acts;
 547         struct sk_buff *packet;
 548         struct sw_flow *flow;
 549         struct sw_flow_actions *sf_acts;
 550         struct datapath *dp;
 551         struct vport *input_vport;
 552         u16 mru = 0;
 553         int len;
 554         int err;
 555         bool log = !a[OVS_PACKET_ATTR_PROBE];
 556
 557         err = -EINVAL;
 558         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 559             !a[OVS_PACKET_ATTR_ACTIONS])
 560                 goto err;
 561
 562         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 563         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 564         err = -ENOMEM;
 565         if (!packet)
 566                 goto err;
 567         skb_reserve(packet, NET_IP_ALIGN);
 568
 569         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 570
 571         /* Set packet's mru */
 572         if (a[OVS_PACKET_ATTR_MRU]) {
 573                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 574                 packet->ignore_df = 1;
 575         }
 576         OVS_CB(packet)->mru = mru;
 577
 578         /* Build an sw_flow for sending this packet. */
 579         flow = ovs_flow_alloc();
 580         err = PTR_ERR(flow);
 581         if (IS_ERR(flow))
 582                 goto err_kfree_skb;
 583
 584         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 585                                              packet, &flow->key, log);
 586         if (err)
 587                 goto err_flow_free;
 588
 589         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 590                                    &flow->key, &acts, log);
 591         if (err)
 592                 goto err_flow_free;
 593
 594         rcu_assign_pointer(flow->sf_acts, acts);
 595         packet->priority = flow->key.phy.priority;
 596         packet->mark = flow->key.phy.skb_mark;
 597
 598         rcu_read_lock();
 599         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 600         err = -ENODEV;
 601         if (!dp)
 602                 goto err_unlock;
 603
 604         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 605         if (!input_vport)
 606                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 607
 608         if (!input_vport)
 609                 goto err_unlock;
 610
 611         packet->dev = input_vport->dev;
 612         OVS_CB(packet)->input_vport = input_vport;
 613         sf_acts = rcu_dereference(flow->sf_acts);
 614
 615         local_bh_disable();
 616         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 617         local_bh_enable();
 618         rcu_read_unlock();
 619
 620         ovs_flow_free(flow, false);
 621         return err;
 622
 623 err_unlock:
 624         rcu_read_unlock();
 625 err_flow_free:
 626         ovs_flow_free(flow, false);
 627 err_kfree_skb:
 628         kfree_skb(packet);
 629 err:
 630         return err;
 631 }
 632
 633 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 634         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 635         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 636         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 637         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 638         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 639 };
 640
 641 static const struct genl_ops dp_packet_genl_ops[] = {
 642         { .cmd = OVS_PACKET_CMD_EXECUTE,
 643           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 644           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 645           .doit = ovs_packet_cmd_execute
 646         }
 647 };
 648
 649 static struct genl_family dp_packet_genl_family __ro_after_init = {
 650         .hdrsize = sizeof(struct ovs_header),
 651         .name = OVS_PACKET_FAMILY,
 652         .version = OVS_PACKET_VERSION,
 653         .maxattr = OVS_PACKET_ATTR_MAX,
 654         .policy = packet_policy,
 655         .netnsok = true,
 656         .parallel_ops = true,
 657         .ops = dp_packet_genl_ops,
 658         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 659         .module = THIS_MODULE,
 660 };
 661
 662 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 663                          struct ovs_dp_megaflow_stats *mega_stats)
 664 {
 665         int i;
 666
 667         memset(mega_stats, 0, sizeof(*mega_stats));
 668
 669         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 670         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 671
 672         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 673
 674         for_each_possible_cpu(i) {
 675                 const struct dp_stats_percpu *percpu_stats;
 676                 struct dp_stats_percpu local_stats;
 677                 unsigned int start;
 678
 679                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 680
 681                 do {
 682                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 683                         local_stats = *percpu_stats;
 684                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 685
 686                 stats->n_hit += local_stats.n_hit;
 687                 stats->n_missed += local_stats.n_missed;
 688                 stats->n_lost += local_stats.n_lost;
 689                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 690         }
 691 }
 692
 693 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 694 {
 695         return ovs_identifier_is_ufid(sfid) &&
 696                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 697 }
 698
 699 static bool should_fill_mask(uint32_t ufid_flags)
 700 {
 701         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 702 }
 703
 704 static bool should_fill_actions(uint32_t ufid_flags)
 705 {
 706         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 707 }
 708
 709 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 710                                     const struct sw_flow_id *sfid,
 711                                     uint32_t ufid_flags)
 712 {
 713         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 714
 715         /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
 716          * see ovs_nla_put_identifier()
 717          */
 718         if (sfid && ovs_identifier_is_ufid(sfid))
 719                 len += nla_total_size(sfid->ufid_len);
 720         else
 721                 len += nla_total_size(ovs_key_attr_size());
 722
 723         /* OVS_FLOW_ATTR_KEY */
 724         if (!sfid || should_fill_key(sfid, ufid_flags))
 725                 len += nla_total_size(ovs_key_attr_size());
 726
 727         /* OVS_FLOW_ATTR_MASK */
 728         if (should_fill_mask(ufid_flags))
 729                 len += nla_total_size(ovs_key_attr_size());
 730
 731         /* OVS_FLOW_ATTR_ACTIONS */
 732         if (should_fill_actions(ufid_flags))
 733                 len += nla_total_size(acts->orig_len);
 734
 735         return len
 736                 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 737                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 738                 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 739 }
 740
 741 /* Called with ovs_mutex or RCU read lock. */
 742 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 743                                    struct sk_buff *skb)
 744 {
 745         struct ovs_flow_stats stats;
 746         __be16 tcp_flags;
 747         unsigned long used;
 748
 749         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 750
 751         if (used &&
 752             nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 753                               OVS_FLOW_ATTR_PAD))
 754                 return -EMSGSIZE;
 755
 756         if (stats.n_packets &&
 757             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 758                           sizeof(struct ovs_flow_stats), &stats,
 759                           OVS_FLOW_ATTR_PAD))
 760                 return -EMSGSIZE;
 761
 762         if ((u8)ntohs(tcp_flags) &&
 763              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 764                 return -EMSGSIZE;
 765
 766         return 0;
 767 }
 768
 769 /* Called with ovs_mutex or RCU read lock. */
 770 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 771                                      struct sk_buff *skb, int skb_orig_len)
 772 {
 773         struct nlattr *start;
 774         int err;
 775
 776         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 777          * this is the first flow to be dumped into 'skb'.  This is unusual for
 778          * Netlink but individual action lists can be longer than
 779          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 780          * The userspace caller can always fetch the actions separately if it
 781          * really wants them.  (Most userspace callers in fact don't care.)
 782          *
 783          * This can only fail for dump operations because the skb is always
 784          * properly sized for single flows.
 785          */
 786         start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
 787         if (start) {
 788                 const struct sw_flow_actions *sf_acts;
 789
 790                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 791                 err = ovs_nla_put_actions(sf_acts->actions,
 792                                           sf_acts->actions_len, skb);
 793
 794                 if (!err)
 795                         nla_nest_end(skb, start);
 796                 else {
 797                         if (skb_orig_len)
 798                                 return err;
 799
 800                         nla_nest_cancel(skb, start);
 801                 }
 802         } else if (skb_orig_len) {
 803                 return -EMSGSIZE;
 804         }
 805
 806         return 0;
 807 }
 808
 809 /* Called with ovs_mutex or RCU read lock. */
 810 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 811                                   struct sk_buff *skb, u32 portid,
 812                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 813 {
 814         const int skb_orig_len = skb->len;
 815         struct ovs_header *ovs_header;
 816         int err;
 817
 818         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 819                                  flags, cmd);
 820         if (!ovs_header)
 821                 return -EMSGSIZE;
 822
 823         ovs_header->dp_ifindex = dp_ifindex;
 824
 825         err = ovs_nla_put_identifier(flow, skb);
 826         if (err)
 827                 goto error;
 828
 829         if (should_fill_key(&flow->id, ufid_flags)) {
 830                 err = ovs_nla_put_masked_key(flow, skb);
 831                 if (err)
 832                         goto error;
 833         }
 834
 835         if (should_fill_mask(ufid_flags)) {
 836                 err = ovs_nla_put_mask(flow, skb);
 837                 if (err)
 838                         goto error;
 839         }
 840
 841         err = ovs_flow_cmd_fill_stats(flow, skb);
 842         if (err)
 843                 goto error;
 844
 845         if (should_fill_actions(ufid_flags)) {
 846                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 847                 if (err)
 848                         goto error;
 849         }
 850
 851         genlmsg_end(skb, ovs_header);
 852         return 0;
 853
 854 error:
 855         genlmsg_cancel(skb, ovs_header);
 856         return err;
 857 }
 858
 859 /* May not be called with RCU read lock. */
 860 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 861                                                const struct sw_flow_id *sfid,
 862                                                struct genl_info *info,
 863                                                bool always,
 864                                                uint32_t ufid_flags)
 865 {
 866         struct sk_buff *skb;
 867         size_t len;
 868
 869         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 870                 return NULL;
 871
 872         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 873         skb = genlmsg_new(len, GFP_KERNEL);
 874         if (!skb)
 875                 return ERR_PTR(-ENOMEM);
 876
 877         return skb;
 878 }
 879
 880 /* Called with ovs_mutex. */
 881 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 882                                                int dp_ifindex,
 883                                                struct genl_info *info, u8 cmd,
 884                                                bool always, u32 ufid_flags)
 885 {
 886         struct sk_buff *skb;
 887         int retval;
 888
 889         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 890                                       &flow->id, info, always, ufid_flags);
 891         if (IS_ERR_OR_NULL(skb))
 892                 return skb;
 893
 894         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 895                                         info->snd_portid, info->snd_seq, 0,
 896                                         cmd, ufid_flags);
 897         if (WARN_ON_ONCE(retval < 0)) {
 898                 kfree_skb(skb);
 899                 skb = ERR_PTR(retval);
 900         }
 901         return skb;
 902 }
 903
 904 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 905 {
 906         struct net *net = sock_net(skb->sk);
 907         struct nlattr **a = info->attrs;
 908         struct ovs_header *ovs_header = info->userhdr;
 909         struct sw_flow *flow = NULL, *new_flow;
 910         struct sw_flow_mask mask;
 911         struct sk_buff *reply;
 912         struct datapath *dp;
 913         struct sw_flow_key *key;
 914         struct sw_flow_actions *acts;
 915         struct sw_flow_match match;
 916         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 917         int error;
 918         bool log = !a[OVS_FLOW_ATTR_PROBE];
 919
 920         /* Must have key and actions. */
 921         error = -EINVAL;
 922         if (!a[OVS_FLOW_ATTR_KEY]) {
 923                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 924                 goto error;
 925         }
 926         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 927                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 928                 goto error;
 929         }
 930
 931         /* Most of the time we need to allocate a new flow, do it before
 932          * locking.
 933          */
 934         new_flow = ovs_flow_alloc();
 935         if (IS_ERR(new_flow)) {
 936                 error = PTR_ERR(new_flow);
 937                 goto error;
 938         }
 939
 940         /* Extract key. */
 941         key = kzalloc(sizeof(*key), GFP_KERNEL);
 942         if (!key) {
 943                 error = -ENOMEM;
 944                 goto err_kfree_flow;
 945         }
 946
 947         ovs_match_init(&match, key, false, &mask);
 948         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 949                                   a[OVS_FLOW_ATTR_MASK], log);
 950         if (error)
 951                 goto err_kfree_key;
 952
 953         ovs_flow_mask_key(&new_flow->key, key, true, &mask);
 954
 955         /* Extract flow identifier. */
 956         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 957                                        key, log);
 958         if (error)
 959                 goto err_kfree_key;
 960
 961         /* Validate actions. */
 962         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 963                                      &new_flow->key, &acts, log);
 964         if (error) {
 965                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 966                 goto err_kfree_key;
 967         }
 968
 969         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 970                                         ufid_flags);
 971         if (IS_ERR(reply)) {
 972                 error = PTR_ERR(reply);
 973                 goto err_kfree_acts;
 974         }
 975
 976         ovs_lock();
 977         dp = get_dp(net, ovs_header->dp_ifindex);
 978         if (unlikely(!dp)) {
 979                 error = -ENODEV;
 980                 goto err_unlock_ovs;
 981         }
 982
 983         /* Check if this is a duplicate flow */
 984         if (ovs_identifier_is_ufid(&new_flow->id))
 985                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
 986         if (!flow)
 987                 flow = ovs_flow_tbl_lookup(&dp->table, key);
 988         if (likely(!flow)) {
 989                 rcu_assign_pointer(new_flow->sf_acts, acts);
 990
 991                 /* Put flow in bucket. */
 992                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 993                 if (unlikely(error)) {
 994                         acts = NULL;
 995                         goto err_unlock_ovs;
 996                 }
 997
 998                 if (unlikely(reply)) {
 999                         error = ovs_flow_cmd_fill_info(new_flow,
1000                                                        ovs_header->dp_ifindex,
1001                                                        reply, info->snd_portid,
1002                                                        info->snd_seq, 0,
1003                                                        OVS_FLOW_CMD_NEW,
1004                                                        ufid_flags);
1005                         BUG_ON(error < 0);
1006                 }
1007                 ovs_unlock();
1008         } else {
1009                 struct sw_flow_actions *old_acts;
1010
1011                 /* Bail out if we're not allowed to modify an existing flow.
1012                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1013                  * because Generic Netlink treats the latter as a dump
1014                  * request.  We also accept NLM_F_EXCL in case that bug ever
1015                  * gets fixed.
1016                  */
1017                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1018                                                          | NLM_F_EXCL))) {
1019                         error = -EEXIST;
1020                         goto err_unlock_ovs;
1021                 }
1022                 /* The flow identifier has to be the same for flow updates.
1023                  * Look for any overlapping flow.
1024                  */
1025                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1026                         if (ovs_identifier_is_key(&flow->id))
1027                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1028                                                                  &match);
1029                         else /* UFID matches but key is different */
1030                                 flow = NULL;
1031                         if (!flow) {
1032                                 error = -ENOENT;
1033                                 goto err_unlock_ovs;
1034                         }
1035                 }
1036                 /* Update actions. */
1037                 old_acts = ovsl_dereference(flow->sf_acts);
1038                 rcu_assign_pointer(flow->sf_acts, acts);
1039
1040                 if (unlikely(reply)) {
1041                         error = ovs_flow_cmd_fill_info(flow,
1042                                                        ovs_header->dp_ifindex,
1043                                                        reply, info->snd_portid,
1044                                                        info->snd_seq, 0,
1045                                                        OVS_FLOW_CMD_NEW,
1046                                                        ufid_flags);
1047                         BUG_ON(error < 0);
1048                 }
1049                 ovs_unlock();
1050
1051                 ovs_nla_free_flow_actions_rcu(old_acts);
1052                 ovs_flow_free(new_flow, false);
1053         }
1054
1055         if (reply)
1056                 ovs_notify(&dp_flow_genl_family, reply, info);
1057
1058         kfree(key);
1059         return 0;
1060
1061 err_unlock_ovs:
1062         ovs_unlock();
1063         kfree_skb(reply);
1064 err_kfree_acts:
1065         ovs_nla_free_flow_actions(acts);
1066 err_kfree_key:
1067         kfree(key);
1068 err_kfree_flow:
1069         ovs_flow_free(new_flow, false);
1070 error:
1071         return error;
1072 }
1073
1074 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1075 static noinline_for_stack struct sw_flow_actions *get_flow_actions(struct net *net,
1076                                                 const struct nlattr *a,
1077                                                 const struct sw_flow_key *key,
1078                                                 const struct sw_flow_mask *mask,
1079                                                 bool log)
1080 {
1081         struct sw_flow_actions *acts;
1082         struct sw_flow_key masked_key;
1083         int error;
1084
1085         ovs_flow_mask_key(&masked_key, key, true, mask);
1086         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1087         if (error) {
1088                 OVS_NLERR(log,
1089                           "Actions may not be safe on all matching packets");
1090                 return ERR_PTR(error);
1091         }
1092
1093         return acts;
1094 }
1095
1096 /* Factor out match-init and action-copy to avoid
1097  * "Wframe-larger-than=1024" warning. Because mask is only
1098  * used to get actions, we new a function to save some
1099  * stack space.
1100  *
1101  * If there are not key and action attrs, we return 0
1102  * directly. In the case, the caller will also not use the
1103  * match as before. If there is action attr, we try to get
1104  * actions and save them to *acts. Before returning from
1105  * the function, we reset the match->mask pointer. Because
1106  * we should not to return match object with dangling reference
1107  * to mask.
1108  * */
1109 static noinline_for_stack int
1110 ovs_nla_init_match_and_action(struct net *net,
1111                               struct sw_flow_match *match,
1112                               struct sw_flow_key *key,
1113                               struct nlattr **a,
1114                               struct sw_flow_actions **acts,
1115                               bool log)
1116 {
1117         struct sw_flow_mask mask;
1118         int error = 0;
1119
1120         if (a[OVS_FLOW_ATTR_KEY]) {
1121                 ovs_match_init(match, key, true, &mask);
1122                 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1123                                           a[OVS_FLOW_ATTR_MASK], log);
1124                 if (error)
1125                         goto error;
1126         }
1127
1128         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1129                 if (!a[OVS_FLOW_ATTR_KEY]) {
1130                         OVS_NLERR(log,
1131                                   "Flow key attribute not present in set flow.");
1132                         error = -EINVAL;
1133                         goto error;
1134                 }
1135
1136                 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1137                                          &mask, log);
1138                 if (IS_ERR(*acts)) {
1139                         error = PTR_ERR(*acts);
1140                         goto error;
1141                 }
1142         }
1143
1144         /* On success, error is 0. */
1145 error:
1146         match->mask = NULL;
1147         return error;
1148 }
1149
1150 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1151 {
1152         struct net *net = sock_net(skb->sk);
1153         struct nlattr **a = info->attrs;
1154         struct ovs_header *ovs_header = info->userhdr;
1155         struct sw_flow_key key;
1156         struct sw_flow *flow;
1157         struct sk_buff *reply = NULL;
1158         struct datapath *dp;
1159         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1160         struct sw_flow_match match;
1161         struct sw_flow_id sfid;
1162         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1163         int error = 0;
1164         bool log = !a[OVS_FLOW_ATTR_PROBE];
1165         bool ufid_present;
1166
1167         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1168         if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1169                 OVS_NLERR(log,
1170                           "Flow set message rejected, Key attribute missing.");
1171                 return -EINVAL;
1172         }
1173
1174         error = ovs_nla_init_match_and_action(net, &match, &key, a,
1175                                               &acts, log);
1176         if (error)
1177                 goto error;
1178
1179         if (acts) {
1180                 /* Can allocate before locking if have acts. */
1181                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1182                                                 ufid_flags);
1183                 if (IS_ERR(reply)) {
1184                         error = PTR_ERR(reply);
1185                         goto err_kfree_acts;
1186                 }
1187         }
1188
1189         ovs_lock();
1190         dp = get_dp(net, ovs_header->dp_ifindex);
1191         if (unlikely(!dp)) {
1192                 error = -ENODEV;
1193                 goto err_unlock_ovs;
1194         }
1195         /* Check that the flow exists. */
1196         if (ufid_present)
1197                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1198         else
1199                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1200         if (unlikely(!flow)) {
1201                 error = -ENOENT;
1202                 goto err_unlock_ovs;
1203         }
1204
1205         /* Update actions, if present. */
1206         if (likely(acts)) {
1207                 old_acts = ovsl_dereference(flow->sf_acts);
1208                 rcu_assign_pointer(flow->sf_acts, acts);
1209
1210                 if (unlikely(reply)) {
1211                         error = ovs_flow_cmd_fill_info(flow,
1212                                                        ovs_header->dp_ifindex,
1213                                                        reply, info->snd_portid,
1214                                                        info->snd_seq, 0,
1215                                                        OVS_FLOW_CMD_SET,
1216                                                        ufid_flags);
1217                         BUG_ON(error < 0);
1218                 }
1219         } else {
1220                 /* Could not alloc without acts before locking. */
1221                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1222                                                 info, OVS_FLOW_CMD_SET, false,
1223                                                 ufid_flags);
1224
1225                 if (IS_ERR(reply)) {
1226                         error = PTR_ERR(reply);
1227                         goto err_unlock_ovs;
1228                 }
1229         }
1230
1231         /* Clear stats. */
1232         if (a[OVS_FLOW_ATTR_CLEAR])
1233                 ovs_flow_stats_clear(flow);
1234         ovs_unlock();
1235
1236         if (reply)
1237                 ovs_notify(&dp_flow_genl_family, reply, info);
1238         if (old_acts)
1239                 ovs_nla_free_flow_actions_rcu(old_acts);
1240
1241         return 0;
1242
1243 err_unlock_ovs:
1244         ovs_unlock();
1245         kfree_skb(reply);
1246 err_kfree_acts:
1247         ovs_nla_free_flow_actions(acts);
1248 error:
1249         return error;
1250 }
1251
1252 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1253 {
1254         struct nlattr **a = info->attrs;
1255         struct ovs_header *ovs_header = info->userhdr;
1256         struct net *net = sock_net(skb->sk);
1257         struct sw_flow_key key;
1258         struct sk_buff *reply;
1259         struct sw_flow *flow;
1260         struct datapath *dp;
1261         struct sw_flow_match match;
1262         struct sw_flow_id ufid;
1263         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1264         int err = 0;
1265         bool log = !a[OVS_FLOW_ATTR_PROBE];
1266         bool ufid_present;
1267
1268         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1269         if (a[OVS_FLOW_ATTR_KEY]) {
1270                 ovs_match_init(&match, &key, true, NULL);
1271                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1272                                         log);
1273         } else if (!ufid_present) {
1274                 OVS_NLERR(log,
1275                           "Flow get message rejected, Key attribute missing.");
1276                 err = -EINVAL;
1277         }
1278         if (err)
1279                 return err;
1280
1281         ovs_lock();
1282         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1283         if (!dp) {
1284                 err = -ENODEV;
1285                 goto unlock;
1286         }
1287
1288         if (ufid_present)
1289                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1290         else
1291                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1292         if (!flow) {
1293                 err = -ENOENT;
1294                 goto unlock;
1295         }
1296
1297         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1298                                         OVS_FLOW_CMD_GET, true, ufid_flags);
1299         if (IS_ERR(reply)) {
1300                 err = PTR_ERR(reply);
1301                 goto unlock;
1302         }
1303
1304         ovs_unlock();
1305         return genlmsg_reply(reply, info);
1306 unlock:
1307         ovs_unlock();
1308         return err;
1309 }
1310
1311 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1312 {
1313         struct nlattr **a = info->attrs;
1314         struct ovs_header *ovs_header = info->userhdr;
1315         struct net *net = sock_net(skb->sk);
1316         struct sw_flow_key key;
1317         struct sk_buff *reply;
1318         struct sw_flow *flow = NULL;
1319         struct datapath *dp;
1320         struct sw_flow_match match;
1321         struct sw_flow_id ufid;
1322         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1323         int err;
1324         bool log = !a[OVS_FLOW_ATTR_PROBE];
1325         bool ufid_present;
1326
1327         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1328         if (a[OVS_FLOW_ATTR_KEY]) {
1329                 ovs_match_init(&match, &key, true, NULL);
1330                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1331                                         NULL, log);
1332                 if (unlikely(err))
1333                         return err;
1334         }
1335
1336         ovs_lock();
1337         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1338         if (unlikely(!dp)) {
1339                 err = -ENODEV;
1340                 goto unlock;
1341         }
1342
1343         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1344                 err = ovs_flow_tbl_flush(&dp->table);
1345                 goto unlock;
1346         }
1347
1348         if (ufid_present)
1349                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1350         else
1351                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1352         if (unlikely(!flow)) {
1353                 err = -ENOENT;
1354                 goto unlock;
1355         }
1356
1357         ovs_flow_tbl_remove(&dp->table, flow);
1358         ovs_unlock();
1359
1360         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1361                                         &flow->id, info, false, ufid_flags);
1362         if (likely(reply)) {
1363                 if (!IS_ERR(reply)) {
1364                         rcu_read_lock();        /*To keep RCU checker happy. */
1365                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1366                                                      reply, info->snd_portid,
1367                                                      info->snd_seq, 0,
1368                                                      OVS_FLOW_CMD_DEL,
1369                                                      ufid_flags);
1370                         rcu_read_unlock();
1371                         if (WARN_ON_ONCE(err < 0)) {
1372                                 kfree_skb(reply);
1373                                 goto out_free;
1374                         }
1375
1376                         ovs_notify(&dp_flow_genl_family, reply, info);
1377                 } else {
1378                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1379                 }
1380         }
1381
1382 out_free:
1383         ovs_flow_free(flow, true);
1384         return 0;
1385 unlock:
1386         ovs_unlock();
1387         return err;
1388 }
1389
1390 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1391 {
1392         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1393         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1394         struct table_instance *ti;
1395         struct datapath *dp;
1396         u32 ufid_flags;
1397         int err;
1398
1399         err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1400                                        OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1401         if (err)
1402                 return err;
1403         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1404
1405         rcu_read_lock();
1406         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1407         if (!dp) {
1408                 rcu_read_unlock();
1409                 return -ENODEV;
1410         }
1411
1412         ti = rcu_dereference(dp->table.ti);
1413         for (;;) {
1414                 struct sw_flow *flow;
1415                 u32 bucket, obj;
1416
1417                 bucket = cb->args[0];
1418                 obj = cb->args[1];
1419                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1420                 if (!flow)
1421                         break;
1422
1423                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1424                                            NETLINK_CB(cb->skb).portid,
1425                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1426                                            OVS_FLOW_CMD_GET, ufid_flags) < 0)
1427                         break;
1428
1429                 cb->args[0] = bucket;
1430                 cb->args[1] = obj;
1431         }
1432         rcu_read_unlock();
1433         return skb->len;
1434 }
1435
1436 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1437         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1438         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1439         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1440         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1441         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1442         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1443         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1444 };
1445
1446 static const struct genl_ops dp_flow_genl_ops[] = {
1447         { .cmd = OVS_FLOW_CMD_NEW,
1448           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1449           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1450           .doit = ovs_flow_cmd_new
1451         },
1452         { .cmd = OVS_FLOW_CMD_DEL,
1453           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1454           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1455           .doit = ovs_flow_cmd_del
1456         },
1457         { .cmd = OVS_FLOW_CMD_GET,
1458           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1459           .flags = 0,               /* OK for unprivileged users. */
1460           .doit = ovs_flow_cmd_get,
1461           .dumpit = ovs_flow_cmd_dump
1462         },
1463         { .cmd = OVS_FLOW_CMD_SET,
1464           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1465           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1466           .doit = ovs_flow_cmd_set,
1467         },
1468 };
1469
1470 static struct genl_family dp_flow_genl_family __ro_after_init = {
1471         .hdrsize = sizeof(struct ovs_header),
1472         .name = OVS_FLOW_FAMILY,
1473         .version = OVS_FLOW_VERSION,
1474         .maxattr = OVS_FLOW_ATTR_MAX,
1475         .policy = flow_policy,
1476         .netnsok = true,
1477         .parallel_ops = true,
1478         .ops = dp_flow_genl_ops,
1479         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1480         .mcgrps = &ovs_dp_flow_multicast_group,
1481         .n_mcgrps = 1,
1482         .module = THIS_MODULE,
1483 };
1484
1485 static size_t ovs_dp_cmd_msg_size(void)
1486 {
1487         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1488
1489         msgsize += nla_total_size(IFNAMSIZ);
1490         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1491         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1492         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1493
1494         return msgsize;
1495 }
1496
1497 /* Called with ovs_mutex. */
1498 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1499                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1500 {
1501         struct ovs_header *ovs_header;
1502         struct ovs_dp_stats dp_stats;
1503         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1504         int err;
1505
1506         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1507                                    flags, cmd);
1508         if (!ovs_header)
1509                 goto error;
1510
1511         ovs_header->dp_ifindex = get_dpifindex(dp);
1512
1513         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1514         if (err)
1515                 goto nla_put_failure;
1516
1517         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1518         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1519                           &dp_stats, OVS_DP_ATTR_PAD))
1520                 goto nla_put_failure;
1521
1522         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1523                           sizeof(struct ovs_dp_megaflow_stats),
1524                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1525                 goto nla_put_failure;
1526
1527         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1528                 goto nla_put_failure;
1529
1530         genlmsg_end(skb, ovs_header);
1531         return 0;
1532
1533 nla_put_failure:
1534         genlmsg_cancel(skb, ovs_header);
1535 error:
1536         return -EMSGSIZE;
1537 }
1538
1539 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1540 {
1541         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1542 }
1543
1544 /* Called with rcu_read_lock or ovs_mutex. */
1545 static struct datapath *lookup_datapath(struct net *net,
1546                                         const struct ovs_header *ovs_header,
1547                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1548 {
1549         struct datapath *dp;
1550
1551         if (!a[OVS_DP_ATTR_NAME])
1552                 dp = get_dp(net, ovs_header->dp_ifindex);
1553         else {
1554                 struct vport *vport;
1555
1556                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1557                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1558         }
1559         return dp ? dp : ERR_PTR(-ENODEV);
1560 }
1561
1562 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1563 {
1564         struct datapath *dp;
1565
1566         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1567         if (IS_ERR(dp))
1568                 return;
1569
1570         pr_warn("%s: Dropping previously announced user features\n",
1571                 ovs_dp_name(dp));
1572         dp->user_features = 0;
1573 }
1574
1575 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1576
1577 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1578 {
1579         u32 user_features = 0;
1580
1581         if (a[OVS_DP_ATTR_USER_FEATURES]) {
1582                 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1583
1584                 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1585                                       OVS_DP_F_UNALIGNED |
1586                                       OVS_DP_F_TC_RECIRC_SHARING))
1587                         return -EOPNOTSUPP;
1588
1589 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1590                 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1591                         return -EOPNOTSUPP;
1592 #endif
1593         }
1594
1595         dp->user_features = user_features;
1596
1597         if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1598                 static_branch_enable(&tc_recirc_sharing_support);
1599         else
1600                 static_branch_disable(&tc_recirc_sharing_support);
1601
1602         return 0;
1603 }
1604
1605 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1606 {
1607         struct nlattr **a = info->attrs;
1608         struct vport_parms parms;
1609         struct sk_buff *reply;
1610         struct datapath *dp;
1611         struct vport *vport;
1612         struct ovs_net *ovs_net;
1613         int err, i;
1614
1615         err = -EINVAL;
1616         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1617                 goto err;
1618
1619         reply = ovs_dp_cmd_alloc_info();
1620         if (!reply)
1621                 return -ENOMEM;
1622
1623         err = -ENOMEM;
1624         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1625         if (dp == NULL)
1626                 goto err_free_reply;
1627
1628         ovs_dp_set_net(dp, sock_net(skb->sk));
1629
1630         /* Allocate table. */
1631         err = ovs_flow_tbl_init(&dp->table);
1632         if (err)
1633                 goto err_free_dp;
1634
1635         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1636         if (!dp->stats_percpu) {
1637                 err = -ENOMEM;
1638                 goto err_destroy_table;
1639         }
1640
1641         dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1642                                   sizeof(struct hlist_head),
1643                                   GFP_KERNEL);
1644         if (!dp->ports) {
1645                 err = -ENOMEM;
1646                 goto err_destroy_percpu;
1647         }
1648
1649         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1650                 INIT_HLIST_HEAD(&dp->ports[i]);
1651
1652         err = ovs_meters_init(dp);
1653         if (err)
1654                 goto err_destroy_ports_array;
1655
1656         /* Set up our datapath device. */
1657         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1658         parms.type = OVS_VPORT_TYPE_INTERNAL;
1659         parms.options = NULL;
1660         parms.dp = dp;
1661         parms.port_no = OVSP_LOCAL;
1662         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1663
1664         err = ovs_dp_change(dp, a);
1665         if (err)
1666                 goto err_destroy_meters;
1667
1668         /* So far only local changes have been made, now need the lock. */
1669         ovs_lock();
1670
1671         vport = new_vport(&parms);
1672         if (IS_ERR(vport)) {
1673                 err = PTR_ERR(vport);
1674                 if (err == -EBUSY)
1675                         err = -EEXIST;
1676
1677                 if (err == -EEXIST) {
1678                         /* An outdated user space instance that does not understand
1679                          * the concept of user_features has attempted to create a new
1680                          * datapath and is likely to reuse it. Drop all user features.
1681                          */
1682                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1683                                 ovs_dp_reset_user_features(skb, info);
1684                 }
1685
1686                 ovs_unlock();
1687                 goto err_destroy_meters;
1688         }
1689
1690         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1691                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1692         BUG_ON(err < 0);
1693
1694         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1695         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1696
1697         ovs_unlock();
1698
1699         ovs_notify(&dp_datapath_genl_family, reply, info);
1700         return 0;
1701
1702 err_destroy_meters:
1703         ovs_meters_exit(dp);
1704 err_destroy_ports_array:
1705         kfree(dp->ports);
1706 err_destroy_percpu:
1707         free_percpu(dp->stats_percpu);
1708 err_destroy_table:
1709         ovs_flow_tbl_destroy(&dp->table);
1710 err_free_dp:
1711         kfree(dp);
1712 err_free_reply:
1713         kfree_skb(reply);
1714 err:
1715         return err;
1716 }
1717
1718 /* Called with ovs_mutex. */
1719 static void __dp_destroy(struct datapath *dp)
1720 {
1721         int i;
1722
1723         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1724                 struct vport *vport;
1725                 struct hlist_node *n;
1726
1727                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1728                         if (vport->port_no != OVSP_LOCAL)
1729                                 ovs_dp_detach_port(vport);
1730         }
1731
1732         list_del_rcu(&dp->list_node);
1733
1734         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1735          * all ports in datapath are destroyed first before freeing datapath.
1736          */
1737         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1738
1739         /* RCU destroy the flow table */
1740         call_rcu(&dp->rcu, destroy_dp_rcu);
1741 }
1742
1743 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1744 {
1745         struct sk_buff *reply;
1746         struct datapath *dp;
1747         int err;
1748
1749         reply = ovs_dp_cmd_alloc_info();
1750         if (!reply)
1751                 return -ENOMEM;
1752
1753         ovs_lock();
1754         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1755         err = PTR_ERR(dp);
1756         if (IS_ERR(dp))
1757                 goto err_unlock_free;
1758
1759         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1760                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1761         BUG_ON(err < 0);
1762
1763         __dp_destroy(dp);
1764         ovs_unlock();
1765
1766         ovs_notify(&dp_datapath_genl_family, reply, info);
1767
1768         return 0;
1769
1770 err_unlock_free:
1771         ovs_unlock();
1772         kfree_skb(reply);
1773         return err;
1774 }
1775
1776 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1777 {
1778         struct sk_buff *reply;
1779         struct datapath *dp;
1780         int err;
1781
1782         reply = ovs_dp_cmd_alloc_info();
1783         if (!reply)
1784                 return -ENOMEM;
1785
1786         ovs_lock();
1787         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1788         err = PTR_ERR(dp);
1789         if (IS_ERR(dp))
1790                 goto err_unlock_free;
1791
1792         err = ovs_dp_change(dp, info->attrs);
1793         if (err)
1794                 goto err_unlock_free;
1795
1796         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1797                                    info->snd_seq, 0, OVS_DP_CMD_SET);
1798         BUG_ON(err < 0);
1799
1800         ovs_unlock();
1801         ovs_notify(&dp_datapath_genl_family, reply, info);
1802
1803         return 0;
1804
1805 err_unlock_free:
1806         ovs_unlock();
1807         kfree_skb(reply);
1808         return err;
1809 }
1810
1811 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1812 {
1813         struct sk_buff *reply;
1814         struct datapath *dp;
1815         int err;
1816
1817         reply = ovs_dp_cmd_alloc_info();
1818         if (!reply)
1819                 return -ENOMEM;
1820
1821         ovs_lock();
1822         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1823         if (IS_ERR(dp)) {
1824                 err = PTR_ERR(dp);
1825                 goto err_unlock_free;
1826         }
1827         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1828                                    info->snd_seq, 0, OVS_DP_CMD_GET);
1829         BUG_ON(err < 0);
1830         ovs_unlock();
1831
1832         return genlmsg_reply(reply, info);
1833
1834 err_unlock_free:
1835         ovs_unlock();
1836         kfree_skb(reply);
1837         return err;
1838 }
1839
1840 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1841 {
1842         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1843         struct datapath *dp;
1844         int skip = cb->args[0];
1845         int i = 0;
1846
1847         ovs_lock();
1848         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1849                 if (i >= skip &&
1850                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1851                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1852                                          OVS_DP_CMD_GET) < 0)
1853                         break;
1854                 i++;
1855         }
1856         ovs_unlock();
1857
1858         cb->args[0] = i;
1859
1860         return skb->len;
1861 }
1862
1863 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1864         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1865         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1866         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1867 };
1868
1869 static const struct genl_ops dp_datapath_genl_ops[] = {
1870         { .cmd = OVS_DP_CMD_NEW,
1871           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1872           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1873           .doit = ovs_dp_cmd_new
1874         },
1875         { .cmd = OVS_DP_CMD_DEL,
1876           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1877           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1878           .doit = ovs_dp_cmd_del
1879         },
1880         { .cmd = OVS_DP_CMD_GET,
1881           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1882           .flags = 0,               /* OK for unprivileged users. */
1883           .doit = ovs_dp_cmd_get,
1884           .dumpit = ovs_dp_cmd_dump
1885         },
1886         { .cmd = OVS_DP_CMD_SET,
1887           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1888           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1889           .doit = ovs_dp_cmd_set,
1890         },
1891 };
1892
1893 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1894         .hdrsize = sizeof(struct ovs_header),
1895         .name = OVS_DATAPATH_FAMILY,
1896         .version = OVS_DATAPATH_VERSION,
1897         .maxattr = OVS_DP_ATTR_MAX,
1898         .policy = datapath_policy,
1899         .netnsok = true,
1900         .parallel_ops = true,
1901         .ops = dp_datapath_genl_ops,
1902         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1903         .mcgrps = &ovs_dp_datapath_multicast_group,
1904         .n_mcgrps = 1,
1905         .module = THIS_MODULE,
1906 };
1907
1908 /* Called with ovs_mutex or RCU read lock. */
1909 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1910                                    struct net *net, u32 portid, u32 seq,
1911                                    u32 flags, u8 cmd, gfp_t gfp)
1912 {
1913         struct ovs_header *ovs_header;
1914         struct ovs_vport_stats vport_stats;
1915         int err;
1916
1917         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1918                                  flags, cmd);
1919         if (!ovs_header)
1920                 return -EMSGSIZE;
1921
1922         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1923
1924         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1925             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1926             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1927                            ovs_vport_name(vport)) ||
1928             nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1929                 goto nla_put_failure;
1930
1931         if (!net_eq(net, dev_net(vport->dev))) {
1932                 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
1933
1934                 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1935                         goto nla_put_failure;
1936         }
1937
1938         ovs_vport_get_stats(vport, &vport_stats);
1939         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1940                           sizeof(struct ovs_vport_stats), &vport_stats,
1941                           OVS_VPORT_ATTR_PAD))
1942                 goto nla_put_failure;
1943
1944         if (ovs_vport_get_upcall_portids(vport, skb))
1945                 goto nla_put_failure;
1946
1947         err = ovs_vport_get_options(vport, skb);
1948         if (err == -EMSGSIZE)
1949                 goto error;
1950
1951         genlmsg_end(skb, ovs_header);
1952         return 0;
1953
1954 nla_put_failure:
1955         err = -EMSGSIZE;
1956 error:
1957         genlmsg_cancel(skb, ovs_header);
1958         return err;
1959 }
1960
1961 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1962 {
1963         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1964 }
1965
1966 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1967 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1968                                          u32 portid, u32 seq, u8 cmd)
1969 {
1970         struct sk_buff *skb;
1971         int retval;
1972
1973         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1974         if (!skb)
1975                 return ERR_PTR(-ENOMEM);
1976
1977         retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
1978                                          GFP_KERNEL);
1979         BUG_ON(retval < 0);
1980
1981         return skb;
1982 }
1983
1984 /* Called with ovs_mutex or RCU read lock. */
1985 static struct vport *lookup_vport(struct net *net,
1986                                   const struct ovs_header *ovs_header,
1987                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1988 {
1989         struct datapath *dp;
1990         struct vport *vport;
1991
1992         if (a[OVS_VPORT_ATTR_IFINDEX])
1993                 return ERR_PTR(-EOPNOTSUPP);
1994         if (a[OVS_VPORT_ATTR_NAME]) {
1995                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1996                 if (!vport)
1997                         return ERR_PTR(-ENODEV);
1998                 if (ovs_header->dp_ifindex &&
1999                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2000                         return ERR_PTR(-ENODEV);
2001                 return vport;
2002         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2003                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2004
2005                 if (port_no >= DP_MAX_PORTS)
2006                         return ERR_PTR(-EFBIG);
2007
2008                 dp = get_dp(net, ovs_header->dp_ifindex);
2009                 if (!dp)
2010                         return ERR_PTR(-ENODEV);
2011
2012                 vport = ovs_vport_ovsl_rcu(dp, port_no);
2013                 if (!vport)
2014                         return ERR_PTR(-ENODEV);
2015                 return vport;
2016         } else
2017                 return ERR_PTR(-EINVAL);
2018
2019 }
2020
2021 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2022 {
2023         unsigned int dev_headroom, max_headroom = 0;
2024         struct net_device *dev;
2025         struct vport *vport;
2026         int i;
2027
2028         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2029                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2030                         dev = vport->dev;
2031                         dev_headroom = netdev_get_fwd_headroom(dev);
2032                         if (dev_headroom > max_headroom)
2033                                 max_headroom = dev_headroom;
2034                 }
2035         }
2036
2037         return max_headroom;
2038 }
2039
2040 /* Called with ovs_mutex */
2041 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2042 {
2043         struct vport *vport;
2044         int i;
2045
2046         dp->max_headroom = new_headroom;
2047         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
2048                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
2049                         netdev_set_rx_headroom(vport->dev, new_headroom);
2050 }
2051
2052 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2053 {
2054         struct nlattr **a = info->attrs;
2055         struct ovs_header *ovs_header = info->userhdr;
2056         struct vport_parms parms;
2057         struct sk_buff *reply;
2058         struct vport *vport;
2059         struct datapath *dp;
2060         unsigned int new_headroom;
2061         u32 port_no;
2062         int err;
2063
2064         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2065             !a[OVS_VPORT_ATTR_UPCALL_PID])
2066                 return -EINVAL;
2067         if (a[OVS_VPORT_ATTR_IFINDEX])
2068                 return -EOPNOTSUPP;
2069
2070         port_no = a[OVS_VPORT_ATTR_PORT_NO]
2071                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2072         if (port_no >= DP_MAX_PORTS)
2073                 return -EFBIG;
2074
2075         reply = ovs_vport_cmd_alloc_info();
2076         if (!reply)
2077                 return -ENOMEM;
2078
2079         ovs_lock();
2080 restart:
2081         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2082         err = -ENODEV;
2083         if (!dp)
2084                 goto exit_unlock_free;
2085
2086         if (port_no) {
2087                 vport = ovs_vport_ovsl(dp, port_no);
2088                 err = -EBUSY;
2089                 if (vport)
2090                         goto exit_unlock_free;
2091         } else {
2092                 for (port_no = 1; ; port_no++) {
2093                         if (port_no >= DP_MAX_PORTS) {
2094                                 err = -EFBIG;
2095                                 goto exit_unlock_free;
2096                         }
2097                         vport = ovs_vport_ovsl(dp, port_no);
2098                         if (!vport)
2099                                 break;
2100                 }
2101         }
2102
2103         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2104         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2105         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2106         parms.dp = dp;
2107         parms.port_no = port_no;
2108         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2109
2110         vport = new_vport(&parms);
2111         err = PTR_ERR(vport);
2112         if (IS_ERR(vport)) {
2113                 if (err == -EAGAIN)
2114                         goto restart;
2115                 goto exit_unlock_free;
2116         }
2117
2118         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2119                                       info->snd_portid, info->snd_seq, 0,
2120                                       OVS_VPORT_CMD_NEW, GFP_KERNEL);
2121
2122         new_headroom = netdev_get_fwd_headroom(vport->dev);
2123
2124         if (new_headroom > dp->max_headroom)
2125                 ovs_update_headroom(dp, new_headroom);
2126         else
2127                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2128
2129         BUG_ON(err < 0);
2130         ovs_unlock();
2131
2132         ovs_notify(&dp_vport_genl_family, reply, info);
2133         return 0;
2134
2135 exit_unlock_free:
2136         ovs_unlock();
2137         kfree_skb(reply);
2138         return err;
2139 }
2140
2141 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2142 {
2143         struct nlattr **a = info->attrs;
2144         struct sk_buff *reply;
2145         struct vport *vport;
2146         int err;
2147
2148         reply = ovs_vport_cmd_alloc_info();
2149         if (!reply)
2150                 return -ENOMEM;
2151
2152         ovs_lock();
2153         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2154         err = PTR_ERR(vport);
2155         if (IS_ERR(vport))
2156                 goto exit_unlock_free;
2157
2158         if (a[OVS_VPORT_ATTR_TYPE] &&
2159             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2160                 err = -EINVAL;
2161                 goto exit_unlock_free;
2162         }
2163
2164         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2165                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2166                 if (err)
2167                         goto exit_unlock_free;
2168         }
2169
2170
2171         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2172                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2173
2174                 err = ovs_vport_set_upcall_portids(vport, ids);
2175                 if (err)
2176                         goto exit_unlock_free;
2177         }
2178
2179         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2180                                       info->snd_portid, info->snd_seq, 0,
2181                                       OVS_VPORT_CMD_SET, GFP_KERNEL);
2182         BUG_ON(err < 0);
2183
2184         ovs_unlock();
2185         ovs_notify(&dp_vport_genl_family, reply, info);
2186         return 0;
2187
2188 exit_unlock_free:
2189         ovs_unlock();
2190         kfree_skb(reply);
2191         return err;
2192 }
2193
2194 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2195 {
2196         bool update_headroom = false;
2197         struct nlattr **a = info->attrs;
2198         struct sk_buff *reply;
2199         struct datapath *dp;
2200         struct vport *vport;
2201         unsigned int new_headroom;
2202         int err;
2203
2204         reply = ovs_vport_cmd_alloc_info();
2205         if (!reply)
2206                 return -ENOMEM;
2207
2208         ovs_lock();
2209         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2210         err = PTR_ERR(vport);
2211         if (IS_ERR(vport))
2212                 goto exit_unlock_free;
2213
2214         if (vport->port_no == OVSP_LOCAL) {
2215                 err = -EINVAL;
2216                 goto exit_unlock_free;
2217         }
2218
2219         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2220                                       info->snd_portid, info->snd_seq, 0,
2221                                       OVS_VPORT_CMD_DEL, GFP_KERNEL);
2222         BUG_ON(err < 0);
2223
2224         /* the vport deletion may trigger dp headroom update */
2225         dp = vport->dp;
2226         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2227                 update_headroom = true;
2228
2229         netdev_reset_rx_headroom(vport->dev);
2230         ovs_dp_detach_port(vport);
2231
2232         if (update_headroom) {
2233                 new_headroom = ovs_get_max_headroom(dp);
2234
2235                 if (new_headroom < dp->max_headroom)
2236                         ovs_update_headroom(dp, new_headroom);
2237         }
2238         ovs_unlock();
2239
2240         ovs_notify(&dp_vport_genl_family, reply, info);
2241         return 0;
2242
2243 exit_unlock_free:
2244         ovs_unlock();
2245         kfree_skb(reply);
2246         return err;
2247 }
2248
2249 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2250 {
2251         struct nlattr **a = info->attrs;
2252         struct ovs_header *ovs_header = info->userhdr;
2253         struct sk_buff *reply;
2254         struct vport *vport;
2255         int err;
2256
2257         reply = ovs_vport_cmd_alloc_info();
2258         if (!reply)
2259                 return -ENOMEM;
2260
2261         rcu_read_lock();
2262         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2263         err = PTR_ERR(vport);
2264         if (IS_ERR(vport))
2265                 goto exit_unlock_free;
2266         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2267                                       info->snd_portid, info->snd_seq, 0,
2268                                       OVS_VPORT_CMD_GET, GFP_ATOMIC);
2269         BUG_ON(err < 0);
2270         rcu_read_unlock();
2271
2272         return genlmsg_reply(reply, info);
2273
2274 exit_unlock_free:
2275         rcu_read_unlock();
2276         kfree_skb(reply);
2277         return err;
2278 }
2279
2280 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2281 {
2282         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2283         struct datapath *dp;
2284         int bucket = cb->args[0], skip = cb->args[1];
2285         int i, j = 0;
2286
2287         rcu_read_lock();
2288         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2289         if (!dp) {
2290                 rcu_read_unlock();
2291                 return -ENODEV;
2292         }
2293         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2294                 struct vport *vport;
2295
2296                 j = 0;
2297                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2298                         if (j >= skip &&
2299                             ovs_vport_cmd_fill_info(vport, skb,
2300                                                     sock_net(skb->sk),
2301                                                     NETLINK_CB(cb->skb).portid,
2302                                                     cb->nlh->nlmsg_seq,
2303                                                     NLM_F_MULTI,
2304                                                     OVS_VPORT_CMD_GET,
2305                                                     GFP_ATOMIC) < 0)
2306                                 goto out;
2307
2308                         j++;
2309                 }
2310                 skip = 0;
2311         }
2312 out:
2313         rcu_read_unlock();
2314
2315         cb->args[0] = i;
2316         cb->args[1] = j;
2317
2318         return skb->len;
2319 }
2320
2321 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2322         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2323         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2324         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2325         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2326         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2327         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2328         [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2329         [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2330 };
2331
2332 static const struct genl_ops dp_vport_genl_ops[] = {
2333         { .cmd = OVS_VPORT_CMD_NEW,
2334           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2335           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2336           .doit = ovs_vport_cmd_new
2337         },
2338         { .cmd = OVS_VPORT_CMD_DEL,
2339           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2340           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2341           .doit = ovs_vport_cmd_del
2342         },
2343         { .cmd = OVS_VPORT_CMD_GET,
2344           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2345           .flags = 0,               /* OK for unprivileged users. */
2346           .doit = ovs_vport_cmd_get,
2347           .dumpit = ovs_vport_cmd_dump
2348         },
2349         { .cmd = OVS_VPORT_CMD_SET,
2350           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2351           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2352           .doit = ovs_vport_cmd_set,
2353         },
2354 };
2355
2356 struct genl_family dp_vport_genl_family __ro_after_init = {
2357         .hdrsize = sizeof(struct ovs_header),
2358         .name = OVS_VPORT_FAMILY,
2359         .version = OVS_VPORT_VERSION,
2360         .maxattr = OVS_VPORT_ATTR_MAX,
2361         .policy = vport_policy,
2362         .netnsok = true,
2363         .parallel_ops = true,
2364         .ops = dp_vport_genl_ops,
2365         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2366         .mcgrps = &ovs_dp_vport_multicast_group,
2367         .n_mcgrps = 1,
2368         .module = THIS_MODULE,
2369 };
2370
2371 static struct genl_family * const dp_genl_families[] = {
2372         &dp_datapath_genl_family,
2373         &dp_vport_genl_family,
2374         &dp_flow_genl_family,
2375         &dp_packet_genl_family,
2376         &dp_meter_genl_family,
2377 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2378         &dp_ct_limit_genl_family,
2379 #endif
2380 };
2381
2382 static void dp_unregister_genl(int n_families)
2383 {
2384         int i;
2385
2386         for (i = 0; i < n_families; i++)
2387                 genl_unregister_family(dp_genl_families[i]);
2388 }
2389
2390 static int __init dp_register_genl(void)
2391 {
2392         int err;
2393         int i;
2394
2395         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2396
2397                 err = genl_register_family(dp_genl_families[i]);
2398                 if (err)
2399                         goto error;
2400         }
2401
2402         return 0;
2403
2404 error:
2405         dp_unregister_genl(i);
2406         return err;
2407 }
2408
2409 static int __net_init ovs_init_net(struct net *net)
2410 {
2411         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2412
2413         INIT_LIST_HEAD(&ovs_net->dps);
2414         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2415         return ovs_ct_init(net);
2416 }
2417
2418 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2419                                             struct list_head *head)
2420 {
2421         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2422         struct datapath *dp;
2423
2424         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2425                 int i;
2426
2427                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2428                         struct vport *vport;
2429
2430                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2431                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2432                                         continue;
2433
2434                                 if (dev_net(vport->dev) == dnet)
2435                                         list_add(&vport->detach_list, head);
2436                         }
2437                 }
2438         }
2439 }
2440
2441 static void __net_exit ovs_exit_net(struct net *dnet)
2442 {
2443         struct datapath *dp, *dp_next;
2444         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2445         struct vport *vport, *vport_next;
2446         struct net *net;
2447         LIST_HEAD(head);
2448
2449         ovs_lock();
2450
2451         ovs_ct_exit(dnet);
2452
2453         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2454                 __dp_destroy(dp);
2455
2456         down_read(&net_rwsem);
2457         for_each_net(net)
2458                 list_vports_from_net(net, dnet, &head);
2459         up_read(&net_rwsem);
2460
2461         /* Detach all vports from given namespace. */
2462         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2463                 list_del(&vport->detach_list);
2464                 ovs_dp_detach_port(vport);
2465         }
2466
2467         ovs_unlock();
2468
2469         cancel_work_sync(&ovs_net->dp_notify_work);
2470 }
2471
2472 static struct pernet_operations ovs_net_ops = {
2473         .init = ovs_init_net,
2474         .exit = ovs_exit_net,
2475         .id   = &ovs_net_id,
2476         .size = sizeof(struct ovs_net),
2477 };
2478
2479 static int __init dp_init(void)
2480 {
2481         int err;
2482
2483         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2484
2485         pr_info("Open vSwitch switching datapath\n");
2486
2487         err = action_fifos_init();
2488         if (err)
2489                 goto error;
2490
2491         err = ovs_internal_dev_rtnl_link_register();
2492         if (err)
2493                 goto error_action_fifos_exit;
2494
2495         err = ovs_flow_init();
2496         if (err)
2497                 goto error_unreg_rtnl_link;
2498
2499         err = ovs_vport_init();
2500         if (err)
2501                 goto error_flow_exit;
2502
2503         err = register_pernet_device(&ovs_net_ops);
2504         if (err)
2505                 goto error_vport_exit;
2506
2507         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2508         if (err)
2509                 goto error_netns_exit;
2510
2511         err = ovs_netdev_init();
2512         if (err)
2513                 goto error_unreg_notifier;
2514
2515         err = dp_register_genl();
2516         if (err < 0)
2517                 goto error_unreg_netdev;
2518
2519         return 0;
2520
2521 error_unreg_netdev:
2522         ovs_netdev_exit();
2523 error_unreg_notifier:
2524         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2525 error_netns_exit:
2526         unregister_pernet_device(&ovs_net_ops);
2527 error_vport_exit:
2528         ovs_vport_exit();
2529 error_flow_exit:
2530         ovs_flow_exit();
2531 error_unreg_rtnl_link:
2532         ovs_internal_dev_rtnl_link_unregister();
2533 error_action_fifos_exit:
2534         action_fifos_exit();
2535 error:
2536         return err;
2537 }
2538
2539 static void dp_cleanup(void)
2540 {
2541         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2542         ovs_netdev_exit();
2543         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2544         unregister_pernet_device(&ovs_net_ops);
2545         rcu_barrier();
2546         ovs_vport_exit();
2547         ovs_flow_exit();
2548         ovs_internal_dev_rtnl_link_unregister();
2549         action_fifos_exit();
2550 }
2551
2552 module_init(dp_init);
2553 module_exit(dp_cleanup);
2554
2555 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2556 MODULE_LICENSE("GPL");
2557 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2558 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2559 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2560 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2561 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2562 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);