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, skb_get_hash(skb),
 231                                          &n_mask_hit);
 232         if (unlikely(!flow)) {
 233                 struct dp_upcall_info upcall;
 234
 235                 memset(&upcall, 0, sizeof(upcall));
 236                 upcall.cmd = OVS_PACKET_CMD_MISS;
 237                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 238                 upcall.mru = OVS_CB(skb)->mru;
 239                 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
 240                 if (unlikely(error))
 241                         kfree_skb(skb);
 242                 else
 243                         consume_skb(skb);
 244                 stats_counter = &stats->n_missed;
 245                 goto out;
 246         }
 247
 248         ovs_flow_stats_update(flow, key->tp.flags, skb);
 249         sf_acts = rcu_dereference(flow->sf_acts);
 250         error = ovs_execute_actions(dp, skb, sf_acts, key);
 251         if (unlikely(error))
 252                 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
 253                                                         ovs_dp_name(dp), error);
 254
 255         stats_counter = &stats->n_hit;
 256
 257 out:
 258         /* Update datapath statistics. */
 259         u64_stats_update_begin(&stats->syncp);
 260         (*stats_counter)++;
 261         stats->n_mask_hit += n_mask_hit;
 262         u64_stats_update_end(&stats->syncp);
 263 }
 264
 265 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 266                   const struct sw_flow_key *key,
 267                   const struct dp_upcall_info *upcall_info,
 268                   uint32_t cutlen)
 269 {
 270         struct dp_stats_percpu *stats;
 271         int err;
 272
 273         if (upcall_info->portid == 0) {
 274                 err = -ENOTCONN;
 275                 goto err;
 276         }
 277
 278         if (!skb_is_gso(skb))
 279                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 280         else
 281                 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
 282         if (err)
 283                 goto err;
 284
 285         return 0;
 286
 287 err:
 288         stats = this_cpu_ptr(dp->stats_percpu);
 289
 290         u64_stats_update_begin(&stats->syncp);
 291         stats->n_lost++;
 292         u64_stats_update_end(&stats->syncp);
 293
 294         return err;
 295 }
 296
 297 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 298                              const struct sw_flow_key *key,
 299                              const struct dp_upcall_info *upcall_info,
 300                                  uint32_t cutlen)
 301 {
 302         unsigned int gso_type = skb_shinfo(skb)->gso_type;
 303         struct sw_flow_key later_key;
 304         struct sk_buff *segs, *nskb;
 305         int err;
 306
 307         BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
 308         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 309         if (IS_ERR(segs))
 310                 return PTR_ERR(segs);
 311         if (segs == NULL)
 312                 return -EINVAL;
 313
 314         if (gso_type & SKB_GSO_UDP) {
 315                 /* The initial flow key extracted by ovs_flow_key_extract()
 316                  * in this case is for a first fragment, so we need to
 317                  * properly mark later fragments.
 318                  */
 319                 later_key = *key;
 320                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 321         }
 322
 323         /* Queue all of the segments. */
 324         skb = segs;
 325         do {
 326                 if (gso_type & SKB_GSO_UDP && skb != segs)
 327                         key = &later_key;
 328
 329                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 330                 if (err)
 331                         break;
 332
 333         } while ((skb = skb->next));
 334
 335         /* Free all of the segments. */
 336         skb = segs;
 337         do {
 338                 nskb = skb->next;
 339                 if (err)
 340                         kfree_skb(skb);
 341                 else
 342                         consume_skb(skb);
 343         } while ((skb = nskb));
 344         return err;
 345 }
 346
 347 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 348                               unsigned int hdrlen, int actions_attrlen)
 349 {
 350         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 351                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 352                 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 353                 + nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
 354                 + nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
 355
 356         /* OVS_PACKET_ATTR_USERDATA */
 357         if (upcall_info->userdata)
 358                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 359
 360         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 361         if (upcall_info->egress_tun_info)
 362                 size += nla_total_size(ovs_tun_key_attr_size());
 363
 364         /* OVS_PACKET_ATTR_ACTIONS */
 365         if (upcall_info->actions_len)
 366                 size += nla_total_size(actions_attrlen);
 367
 368         /* OVS_PACKET_ATTR_MRU */
 369         if (upcall_info->mru)
 370                 size += nla_total_size(sizeof(upcall_info->mru));
 371
 372         return size;
 373 }
 374
 375 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 376 {
 377         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 378                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 379
 380                 if (plen > 0)
 381                         skb_put_zero(skb, plen);
 382         }
 383 }
 384
 385 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 386                                   const struct sw_flow_key *key,
 387                                   const struct dp_upcall_info *upcall_info,
 388                                   uint32_t cutlen)
 389 {
 390         struct ovs_header *upcall;
 391         struct sk_buff *nskb = NULL;
 392         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 393         struct nlattr *nla;
 394         size_t len;
 395         unsigned int hlen;
 396         int err, dp_ifindex;
 397         u64 hash;
 398
 399         dp_ifindex = get_dpifindex(dp);
 400         if (!dp_ifindex)
 401                 return -ENODEV;
 402
 403         if (skb_vlan_tag_present(skb)) {
 404                 nskb = skb_clone(skb, GFP_ATOMIC);
 405                 if (!nskb)
 406                         return -ENOMEM;
 407
 408                 nskb = __vlan_hwaccel_push_inside(nskb);
 409                 if (!nskb)
 410                         return -ENOMEM;
 411
 412                 skb = nskb;
 413         }
 414
 415         if (nla_attr_size(skb->len) > USHRT_MAX) {
 416                 err = -EFBIG;
 417                 goto out;
 418         }
 419
 420         /* Complete checksum if needed */
 421         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 422             (err = skb_csum_hwoffload_help(skb, 0)))
 423                 goto out;
 424
 425         /* Older versions of OVS user space enforce alignment of the last
 426          * Netlink attribute to NLA_ALIGNTO which would require extensive
 427          * padding logic. Only perform zerocopy if padding is not required.
 428          */
 429         if (dp->user_features & OVS_DP_F_UNALIGNED)
 430                 hlen = skb_zerocopy_headlen(skb);
 431         else
 432                 hlen = skb->len;
 433
 434         len = upcall_msg_size(upcall_info, hlen - cutlen,
 435                               OVS_CB(skb)->acts_origlen);
 436         user_skb = genlmsg_new(len, GFP_ATOMIC);
 437         if (!user_skb) {
 438                 err = -ENOMEM;
 439                 goto out;
 440         }
 441
 442         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 443                              0, upcall_info->cmd);
 444         if (!upcall) {
 445                 err = -EINVAL;
 446                 goto out;
 447         }
 448         upcall->dp_ifindex = dp_ifindex;
 449
 450         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 451         if (err)
 452                 goto out;
 453
 454         if (upcall_info->userdata)
 455                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 456                           nla_len(upcall_info->userdata),
 457                           nla_data(upcall_info->userdata));
 458
 459         if (upcall_info->egress_tun_info) {
 460                 nla = nla_nest_start_noflag(user_skb,
 461                                             OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 462                 if (!nla) {
 463                         err = -EMSGSIZE;
 464                         goto out;
 465                 }
 466                 err = ovs_nla_put_tunnel_info(user_skb,
 467                                               upcall_info->egress_tun_info);
 468                 if (err)
 469                         goto out;
 470
 471                 nla_nest_end(user_skb, nla);
 472         }
 473
 474         if (upcall_info->actions_len) {
 475                 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
 476                 if (!nla) {
 477                         err = -EMSGSIZE;
 478                         goto out;
 479                 }
 480                 err = ovs_nla_put_actions(upcall_info->actions,
 481                                           upcall_info->actions_len,
 482                                           user_skb);
 483                 if (!err)
 484                         nla_nest_end(user_skb, nla);
 485                 else
 486                         nla_nest_cancel(user_skb, nla);
 487         }
 488
 489         /* Add OVS_PACKET_ATTR_MRU */
 490         if (upcall_info->mru &&
 491             nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
 492                 err = -ENOBUFS;
 493                 goto out;
 494         }
 495
 496         /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 497         if (cutlen > 0 &&
 498             nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
 499                 err = -ENOBUFS;
 500                 goto out;
 501         }
 502
 503         /* Add OVS_PACKET_ATTR_HASH */
 504         hash = skb_get_hash_raw(skb);
 505         if (skb->sw_hash)
 506                 hash |= OVS_PACKET_HASH_SW_BIT;
 507
 508         if (skb->l4_hash)
 509                 hash |= OVS_PACKET_HASH_L4_BIT;
 510
 511         if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
 512                 err = -ENOBUFS;
 513                 goto out;
 514         }
 515
 516         /* Only reserve room for attribute header, packet data is added
 517          * in skb_zerocopy() */
 518         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 519                 err = -ENOBUFS;
 520                 goto out;
 521         }
 522         nla->nla_len = nla_attr_size(skb->len - cutlen);
 523
 524         err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 525         if (err)
 526                 goto out;
 527
 528         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 529         pad_packet(dp, user_skb);
 530
 531         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 532
 533         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 534         user_skb = NULL;
 535 out:
 536         if (err)
 537                 skb_tx_error(skb);
 538         kfree_skb(user_skb);
 539         kfree_skb(nskb);
 540         return err;
 541 }
 542
 543 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 544 {
 545         struct ovs_header *ovs_header = info->userhdr;
 546         struct net *net = sock_net(skb->sk);
 547         struct nlattr **a = info->attrs;
 548         struct sw_flow_actions *acts;
 549         struct sk_buff *packet;
 550         struct sw_flow *flow;
 551         struct sw_flow_actions *sf_acts;
 552         struct datapath *dp;
 553         struct vport *input_vport;
 554         u16 mru = 0;
 555         u64 hash;
 556         int len;
 557         int err;
 558         bool log = !a[OVS_PACKET_ATTR_PROBE];
 559
 560         err = -EINVAL;
 561         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 562             !a[OVS_PACKET_ATTR_ACTIONS])
 563                 goto err;
 564
 565         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 566         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 567         err = -ENOMEM;
 568         if (!packet)
 569                 goto err;
 570         skb_reserve(packet, NET_IP_ALIGN);
 571
 572         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 573
 574         /* Set packet's mru */
 575         if (a[OVS_PACKET_ATTR_MRU]) {
 576                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 577                 packet->ignore_df = 1;
 578         }
 579         OVS_CB(packet)->mru = mru;
 580
 581         if (a[OVS_PACKET_ATTR_HASH]) {
 582                 hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
 583
 584                 __skb_set_hash(packet, hash & 0xFFFFFFFFULL,
 585                                !!(hash & OVS_PACKET_HASH_SW_BIT),
 586                                !!(hash & OVS_PACKET_HASH_L4_BIT));
 587         }
 588
 589         /* Build an sw_flow for sending this packet. */
 590         flow = ovs_flow_alloc();
 591         err = PTR_ERR(flow);
 592         if (IS_ERR(flow))
 593                 goto err_kfree_skb;
 594
 595         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 596                                              packet, &flow->key, log);
 597         if (err)
 598                 goto err_flow_free;
 599
 600         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 601                                    &flow->key, &acts, log);
 602         if (err)
 603                 goto err_flow_free;
 604
 605         rcu_assign_pointer(flow->sf_acts, acts);
 606         packet->priority = flow->key.phy.priority;
 607         packet->mark = flow->key.phy.skb_mark;
 608
 609         rcu_read_lock();
 610         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 611         err = -ENODEV;
 612         if (!dp)
 613                 goto err_unlock;
 614
 615         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 616         if (!input_vport)
 617                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 618
 619         if (!input_vport)
 620                 goto err_unlock;
 621
 622         packet->dev = input_vport->dev;
 623         OVS_CB(packet)->input_vport = input_vport;
 624         sf_acts = rcu_dereference(flow->sf_acts);
 625
 626         local_bh_disable();
 627         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 628         local_bh_enable();
 629         rcu_read_unlock();
 630
 631         ovs_flow_free(flow, false);
 632         return err;
 633
 634 err_unlock:
 635         rcu_read_unlock();
 636 err_flow_free:
 637         ovs_flow_free(flow, false);
 638 err_kfree_skb:
 639         kfree_skb(packet);
 640 err:
 641         return err;
 642 }
 643
 644 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 645         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 646         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 647         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 648         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 649         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 650 };
 651
 652 static const struct genl_ops dp_packet_genl_ops[] = {
 653         { .cmd = OVS_PACKET_CMD_EXECUTE,
 654           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 655           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 656           .doit = ovs_packet_cmd_execute
 657         }
 658 };
 659
 660 static struct genl_family dp_packet_genl_family __ro_after_init = {
 661         .hdrsize = sizeof(struct ovs_header),
 662         .name = OVS_PACKET_FAMILY,
 663         .version = OVS_PACKET_VERSION,
 664         .maxattr = OVS_PACKET_ATTR_MAX,
 665         .policy = packet_policy,
 666         .netnsok = true,
 667         .parallel_ops = true,
 668         .ops = dp_packet_genl_ops,
 669         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 670         .module = THIS_MODULE,
 671 };
 672
 673 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 674                          struct ovs_dp_megaflow_stats *mega_stats)
 675 {
 676         int i;
 677
 678         memset(mega_stats, 0, sizeof(*mega_stats));
 679
 680         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 681         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 682
 683         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 684
 685         for_each_possible_cpu(i) {
 686                 const struct dp_stats_percpu *percpu_stats;
 687                 struct dp_stats_percpu local_stats;
 688                 unsigned int start;
 689
 690                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 691
 692                 do {
 693                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 694                         local_stats = *percpu_stats;
 695                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 696
 697                 stats->n_hit += local_stats.n_hit;
 698                 stats->n_missed += local_stats.n_missed;
 699                 stats->n_lost += local_stats.n_lost;
 700                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 701         }
 702 }
 703
 704 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 705 {
 706         return ovs_identifier_is_ufid(sfid) &&
 707                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 708 }
 709
 710 static bool should_fill_mask(uint32_t ufid_flags)
 711 {
 712         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 713 }
 714
 715 static bool should_fill_actions(uint32_t ufid_flags)
 716 {
 717         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 718 }
 719
 720 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 721                                     const struct sw_flow_id *sfid,
 722                                     uint32_t ufid_flags)
 723 {
 724         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 725
 726         /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
 727          * see ovs_nla_put_identifier()
 728          */
 729         if (sfid && ovs_identifier_is_ufid(sfid))
 730                 len += nla_total_size(sfid->ufid_len);
 731         else
 732                 len += nla_total_size(ovs_key_attr_size());
 733
 734         /* OVS_FLOW_ATTR_KEY */
 735         if (!sfid || should_fill_key(sfid, ufid_flags))
 736                 len += nla_total_size(ovs_key_attr_size());
 737
 738         /* OVS_FLOW_ATTR_MASK */
 739         if (should_fill_mask(ufid_flags))
 740                 len += nla_total_size(ovs_key_attr_size());
 741
 742         /* OVS_FLOW_ATTR_ACTIONS */
 743         if (should_fill_actions(ufid_flags))
 744                 len += nla_total_size(acts->orig_len);
 745
 746         return len
 747                 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 748                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 749                 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 750 }
 751
 752 /* Called with ovs_mutex or RCU read lock. */
 753 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 754                                    struct sk_buff *skb)
 755 {
 756         struct ovs_flow_stats stats;
 757         __be16 tcp_flags;
 758         unsigned long used;
 759
 760         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 761
 762         if (used &&
 763             nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 764                               OVS_FLOW_ATTR_PAD))
 765                 return -EMSGSIZE;
 766
 767         if (stats.n_packets &&
 768             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 769                           sizeof(struct ovs_flow_stats), &stats,
 770                           OVS_FLOW_ATTR_PAD))
 771                 return -EMSGSIZE;
 772
 773         if ((u8)ntohs(tcp_flags) &&
 774              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 775                 return -EMSGSIZE;
 776
 777         return 0;
 778 }
 779
 780 /* Called with ovs_mutex or RCU read lock. */
 781 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 782                                      struct sk_buff *skb, int skb_orig_len)
 783 {
 784         struct nlattr *start;
 785         int err;
 786
 787         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 788          * this is the first flow to be dumped into 'skb'.  This is unusual for
 789          * Netlink but individual action lists can be longer than
 790          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 791          * The userspace caller can always fetch the actions separately if it
 792          * really wants them.  (Most userspace callers in fact don't care.)
 793          *
 794          * This can only fail for dump operations because the skb is always
 795          * properly sized for single flows.
 796          */
 797         start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
 798         if (start) {
 799                 const struct sw_flow_actions *sf_acts;
 800
 801                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 802                 err = ovs_nla_put_actions(sf_acts->actions,
 803                                           sf_acts->actions_len, skb);
 804
 805                 if (!err)
 806                         nla_nest_end(skb, start);
 807                 else {
 808                         if (skb_orig_len)
 809                                 return err;
 810
 811                         nla_nest_cancel(skb, start);
 812                 }
 813         } else if (skb_orig_len) {
 814                 return -EMSGSIZE;
 815         }
 816
 817         return 0;
 818 }
 819
 820 /* Called with ovs_mutex or RCU read lock. */
 821 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 822                                   struct sk_buff *skb, u32 portid,
 823                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 824 {
 825         const int skb_orig_len = skb->len;
 826         struct ovs_header *ovs_header;
 827         int err;
 828
 829         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 830                                  flags, cmd);
 831         if (!ovs_header)
 832                 return -EMSGSIZE;
 833
 834         ovs_header->dp_ifindex = dp_ifindex;
 835
 836         err = ovs_nla_put_identifier(flow, skb);
 837         if (err)
 838                 goto error;
 839
 840         if (should_fill_key(&flow->id, ufid_flags)) {
 841                 err = ovs_nla_put_masked_key(flow, skb);
 842                 if (err)
 843                         goto error;
 844         }
 845
 846         if (should_fill_mask(ufid_flags)) {
 847                 err = ovs_nla_put_mask(flow, skb);
 848                 if (err)
 849                         goto error;
 850         }
 851
 852         err = ovs_flow_cmd_fill_stats(flow, skb);
 853         if (err)
 854                 goto error;
 855
 856         if (should_fill_actions(ufid_flags)) {
 857                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 858                 if (err)
 859                         goto error;
 860         }
 861
 862         genlmsg_end(skb, ovs_header);
 863         return 0;
 864
 865 error:
 866         genlmsg_cancel(skb, ovs_header);
 867         return err;
 868 }
 869
 870 /* May not be called with RCU read lock. */
 871 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 872                                                const struct sw_flow_id *sfid,
 873                                                struct genl_info *info,
 874                                                bool always,
 875                                                uint32_t ufid_flags)
 876 {
 877         struct sk_buff *skb;
 878         size_t len;
 879
 880         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 881                 return NULL;
 882
 883         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 884         skb = genlmsg_new(len, GFP_KERNEL);
 885         if (!skb)
 886                 return ERR_PTR(-ENOMEM);
 887
 888         return skb;
 889 }
 890
 891 /* Called with ovs_mutex. */
 892 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 893                                                int dp_ifindex,
 894                                                struct genl_info *info, u8 cmd,
 895                                                bool always, u32 ufid_flags)
 896 {
 897         struct sk_buff *skb;
 898         int retval;
 899
 900         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 901                                       &flow->id, info, always, ufid_flags);
 902         if (IS_ERR_OR_NULL(skb))
 903                 return skb;
 904
 905         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 906                                         info->snd_portid, info->snd_seq, 0,
 907                                         cmd, ufid_flags);
 908         if (WARN_ON_ONCE(retval < 0)) {
 909                 kfree_skb(skb);
 910                 skb = ERR_PTR(retval);
 911         }
 912         return skb;
 913 }
 914
 915 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 916 {
 917         struct net *net = sock_net(skb->sk);
 918         struct nlattr **a = info->attrs;
 919         struct ovs_header *ovs_header = info->userhdr;
 920         struct sw_flow *flow = NULL, *new_flow;
 921         struct sw_flow_mask mask;
 922         struct sk_buff *reply;
 923         struct datapath *dp;
 924         struct sw_flow_actions *acts;
 925         struct sw_flow_match match;
 926         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 927         int error;
 928         bool log = !a[OVS_FLOW_ATTR_PROBE];
 929
 930         /* Must have key and actions. */
 931         error = -EINVAL;
 932         if (!a[OVS_FLOW_ATTR_KEY]) {
 933                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 934                 goto error;
 935         }
 936         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 937                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 938                 goto error;
 939         }
 940
 941         /* Most of the time we need to allocate a new flow, do it before
 942          * locking.
 943          */
 944         new_flow = ovs_flow_alloc();
 945         if (IS_ERR(new_flow)) {
 946                 error = PTR_ERR(new_flow);
 947                 goto error;
 948         }
 949
 950         /* Extract key. */
 951         ovs_match_init(&match, &new_flow->key, false, &mask);
 952         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 953                                   a[OVS_FLOW_ATTR_MASK], log);
 954         if (error)
 955                 goto err_kfree_flow;
 956
 957         /* Extract flow identifier. */
 958         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 959                                        &new_flow->key, log);
 960         if (error)
 961                 goto err_kfree_flow;
 962
 963         /* unmasked key is needed to match when ufid is not used. */
 964         if (ovs_identifier_is_key(&new_flow->id))
 965                 match.key = new_flow->id.unmasked_key;
 966
 967         ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
 968
 969         /* Validate actions. */
 970         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 971                                      &new_flow->key, &acts, log);
 972         if (error) {
 973                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 974                 goto err_kfree_flow;
 975         }
 976
 977         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 978                                         ufid_flags);
 979         if (IS_ERR(reply)) {
 980                 error = PTR_ERR(reply);
 981                 goto err_kfree_acts;
 982         }
 983
 984         ovs_lock();
 985         dp = get_dp(net, ovs_header->dp_ifindex);
 986         if (unlikely(!dp)) {
 987                 error = -ENODEV;
 988                 goto err_unlock_ovs;
 989         }
 990
 991         /* Check if this is a duplicate flow */
 992         if (ovs_identifier_is_ufid(&new_flow->id))
 993                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
 994         if (!flow)
 995                 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
 996         if (likely(!flow)) {
 997                 rcu_assign_pointer(new_flow->sf_acts, acts);
 998
 999                 /* Put flow in bucket. */
1000                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1001                 if (unlikely(error)) {
1002                         acts = NULL;
1003                         goto err_unlock_ovs;
1004                 }
1005
1006                 if (unlikely(reply)) {
1007                         error = ovs_flow_cmd_fill_info(new_flow,
1008                                                        ovs_header->dp_ifindex,
1009                                                        reply, info->snd_portid,
1010                                                        info->snd_seq, 0,
1011                                                        OVS_FLOW_CMD_NEW,
1012                                                        ufid_flags);
1013                         BUG_ON(error < 0);
1014                 }
1015                 ovs_unlock();
1016         } else {
1017                 struct sw_flow_actions *old_acts;
1018
1019                 /* Bail out if we're not allowed to modify an existing flow.
1020                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1021                  * because Generic Netlink treats the latter as a dump
1022                  * request.  We also accept NLM_F_EXCL in case that bug ever
1023                  * gets fixed.
1024                  */
1025                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1026                                                          | NLM_F_EXCL))) {
1027                         error = -EEXIST;
1028                         goto err_unlock_ovs;
1029                 }
1030                 /* The flow identifier has to be the same for flow updates.
1031                  * Look for any overlapping flow.
1032                  */
1033                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1034                         if (ovs_identifier_is_key(&flow->id))
1035                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1036                                                                  &match);
1037                         else /* UFID matches but key is different */
1038                                 flow = NULL;
1039                         if (!flow) {
1040                                 error = -ENOENT;
1041                                 goto err_unlock_ovs;
1042                         }
1043                 }
1044                 /* Update actions. */
1045                 old_acts = ovsl_dereference(flow->sf_acts);
1046                 rcu_assign_pointer(flow->sf_acts, acts);
1047
1048                 if (unlikely(reply)) {
1049                         error = ovs_flow_cmd_fill_info(flow,
1050                                                        ovs_header->dp_ifindex,
1051                                                        reply, info->snd_portid,
1052                                                        info->snd_seq, 0,
1053                                                        OVS_FLOW_CMD_NEW,
1054                                                        ufid_flags);
1055                         BUG_ON(error < 0);
1056                 }
1057                 ovs_unlock();
1058
1059                 ovs_nla_free_flow_actions_rcu(old_acts);
1060                 ovs_flow_free(new_flow, false);
1061         }
1062
1063         if (reply)
1064                 ovs_notify(&dp_flow_genl_family, reply, info);
1065         return 0;
1066
1067 err_unlock_ovs:
1068         ovs_unlock();
1069         kfree_skb(reply);
1070 err_kfree_acts:
1071         ovs_nla_free_flow_actions(acts);
1072 err_kfree_flow:
1073         ovs_flow_free(new_flow, false);
1074 error:
1075         return error;
1076 }
1077
1078 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1079 static noinline_for_stack struct sw_flow_actions *get_flow_actions(struct net *net,
1080                                                 const struct nlattr *a,
1081                                                 const struct sw_flow_key *key,
1082                                                 const struct sw_flow_mask *mask,
1083                                                 bool log)
1084 {
1085         struct sw_flow_actions *acts;
1086         struct sw_flow_key masked_key;
1087         int error;
1088
1089         ovs_flow_mask_key(&masked_key, key, true, mask);
1090         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1091         if (error) {
1092                 OVS_NLERR(log,
1093                           "Actions may not be safe on all matching packets");
1094                 return ERR_PTR(error);
1095         }
1096
1097         return acts;
1098 }
1099
1100 /* Factor out match-init and action-copy to avoid
1101  * "Wframe-larger-than=1024" warning. Because mask is only
1102  * used to get actions, we new a function to save some
1103  * stack space.
1104  *
1105  * If there are not key and action attrs, we return 0
1106  * directly. In the case, the caller will also not use the
1107  * match as before. If there is action attr, we try to get
1108  * actions and save them to *acts. Before returning from
1109  * the function, we reset the match->mask pointer. Because
1110  * we should not to return match object with dangling reference
1111  * to mask.
1112  * */
1113 static noinline_for_stack int
1114 ovs_nla_init_match_and_action(struct net *net,
1115                               struct sw_flow_match *match,
1116                               struct sw_flow_key *key,
1117                               struct nlattr **a,
1118                               struct sw_flow_actions **acts,
1119                               bool log)
1120 {
1121         struct sw_flow_mask mask;
1122         int error = 0;
1123
1124         if (a[OVS_FLOW_ATTR_KEY]) {
1125                 ovs_match_init(match, key, true, &mask);
1126                 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1127                                           a[OVS_FLOW_ATTR_MASK], log);
1128                 if (error)
1129                         goto error;
1130         }
1131
1132         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1133                 if (!a[OVS_FLOW_ATTR_KEY]) {
1134                         OVS_NLERR(log,
1135                                   "Flow key attribute not present in set flow.");
1136                         error = -EINVAL;
1137                         goto error;
1138                 }
1139
1140                 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1141                                          &mask, log);
1142                 if (IS_ERR(*acts)) {
1143                         error = PTR_ERR(*acts);
1144                         goto error;
1145                 }
1146         }
1147
1148         /* On success, error is 0. */
1149 error:
1150         match->mask = NULL;
1151         return error;
1152 }
1153
1154 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1155 {
1156         struct net *net = sock_net(skb->sk);
1157         struct nlattr **a = info->attrs;
1158         struct ovs_header *ovs_header = info->userhdr;
1159         struct sw_flow_key key;
1160         struct sw_flow *flow;
1161         struct sk_buff *reply = NULL;
1162         struct datapath *dp;
1163         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1164         struct sw_flow_match match;
1165         struct sw_flow_id sfid;
1166         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1167         int error = 0;
1168         bool log = !a[OVS_FLOW_ATTR_PROBE];
1169         bool ufid_present;
1170
1171         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1172         if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1173                 OVS_NLERR(log,
1174                           "Flow set message rejected, Key attribute missing.");
1175                 return -EINVAL;
1176         }
1177
1178         error = ovs_nla_init_match_and_action(net, &match, &key, a,
1179                                               &acts, log);
1180         if (error)
1181                 goto error;
1182
1183         if (acts) {
1184                 /* Can allocate before locking if have acts. */
1185                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1186                                                 ufid_flags);
1187                 if (IS_ERR(reply)) {
1188                         error = PTR_ERR(reply);
1189                         goto err_kfree_acts;
1190                 }
1191         }
1192
1193         ovs_lock();
1194         dp = get_dp(net, ovs_header->dp_ifindex);
1195         if (unlikely(!dp)) {
1196                 error = -ENODEV;
1197                 goto err_unlock_ovs;
1198         }
1199         /* Check that the flow exists. */
1200         if (ufid_present)
1201                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1202         else
1203                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1204         if (unlikely(!flow)) {
1205                 error = -ENOENT;
1206                 goto err_unlock_ovs;
1207         }
1208
1209         /* Update actions, if present. */
1210         if (likely(acts)) {
1211                 old_acts = ovsl_dereference(flow->sf_acts);
1212                 rcu_assign_pointer(flow->sf_acts, acts);
1213
1214                 if (unlikely(reply)) {
1215                         error = ovs_flow_cmd_fill_info(flow,
1216                                                        ovs_header->dp_ifindex,
1217                                                        reply, info->snd_portid,
1218                                                        info->snd_seq, 0,
1219                                                        OVS_FLOW_CMD_SET,
1220                                                        ufid_flags);
1221                         BUG_ON(error < 0);
1222                 }
1223         } else {
1224                 /* Could not alloc without acts before locking. */
1225                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1226                                                 info, OVS_FLOW_CMD_SET, false,
1227                                                 ufid_flags);
1228
1229                 if (IS_ERR(reply)) {
1230                         error = PTR_ERR(reply);
1231                         goto err_unlock_ovs;
1232                 }
1233         }
1234
1235         /* Clear stats. */
1236         if (a[OVS_FLOW_ATTR_CLEAR])
1237                 ovs_flow_stats_clear(flow);
1238         ovs_unlock();
1239
1240         if (reply)
1241                 ovs_notify(&dp_flow_genl_family, reply, info);
1242         if (old_acts)
1243                 ovs_nla_free_flow_actions_rcu(old_acts);
1244
1245         return 0;
1246
1247 err_unlock_ovs:
1248         ovs_unlock();
1249         kfree_skb(reply);
1250 err_kfree_acts:
1251         ovs_nla_free_flow_actions(acts);
1252 error:
1253         return error;
1254 }
1255
1256 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1257 {
1258         struct nlattr **a = info->attrs;
1259         struct ovs_header *ovs_header = info->userhdr;
1260         struct net *net = sock_net(skb->sk);
1261         struct sw_flow_key key;
1262         struct sk_buff *reply;
1263         struct sw_flow *flow;
1264         struct datapath *dp;
1265         struct sw_flow_match match;
1266         struct sw_flow_id ufid;
1267         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1268         int err = 0;
1269         bool log = !a[OVS_FLOW_ATTR_PROBE];
1270         bool ufid_present;
1271
1272         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1273         if (a[OVS_FLOW_ATTR_KEY]) {
1274                 ovs_match_init(&match, &key, true, NULL);
1275                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1276                                         log);
1277         } else if (!ufid_present) {
1278                 OVS_NLERR(log,
1279                           "Flow get message rejected, Key attribute missing.");
1280                 err = -EINVAL;
1281         }
1282         if (err)
1283                 return err;
1284
1285         ovs_lock();
1286         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1287         if (!dp) {
1288                 err = -ENODEV;
1289                 goto unlock;
1290         }
1291
1292         if (ufid_present)
1293                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1294         else
1295                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1296         if (!flow) {
1297                 err = -ENOENT;
1298                 goto unlock;
1299         }
1300
1301         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1302                                         OVS_FLOW_CMD_GET, true, ufid_flags);
1303         if (IS_ERR(reply)) {
1304                 err = PTR_ERR(reply);
1305                 goto unlock;
1306         }
1307
1308         ovs_unlock();
1309         return genlmsg_reply(reply, info);
1310 unlock:
1311         ovs_unlock();
1312         return err;
1313 }
1314
1315 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1316 {
1317         struct nlattr **a = info->attrs;
1318         struct ovs_header *ovs_header = info->userhdr;
1319         struct net *net = sock_net(skb->sk);
1320         struct sw_flow_key key;
1321         struct sk_buff *reply;
1322         struct sw_flow *flow = NULL;
1323         struct datapath *dp;
1324         struct sw_flow_match match;
1325         struct sw_flow_id ufid;
1326         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1327         int err;
1328         bool log = !a[OVS_FLOW_ATTR_PROBE];
1329         bool ufid_present;
1330
1331         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1332         if (a[OVS_FLOW_ATTR_KEY]) {
1333                 ovs_match_init(&match, &key, true, NULL);
1334                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1335                                         NULL, log);
1336                 if (unlikely(err))
1337                         return err;
1338         }
1339
1340         ovs_lock();
1341         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1342         if (unlikely(!dp)) {
1343                 err = -ENODEV;
1344                 goto unlock;
1345         }
1346
1347         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1348                 err = ovs_flow_tbl_flush(&dp->table);
1349                 goto unlock;
1350         }
1351
1352         if (ufid_present)
1353                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1354         else
1355                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1356         if (unlikely(!flow)) {
1357                 err = -ENOENT;
1358                 goto unlock;
1359         }
1360
1361         ovs_flow_tbl_remove(&dp->table, flow);
1362         ovs_unlock();
1363
1364         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1365                                         &flow->id, info, false, ufid_flags);
1366         if (likely(reply)) {
1367                 if (!IS_ERR(reply)) {
1368                         rcu_read_lock();        /*To keep RCU checker happy. */
1369                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1370                                                      reply, info->snd_portid,
1371                                                      info->snd_seq, 0,
1372                                                      OVS_FLOW_CMD_DEL,
1373                                                      ufid_flags);
1374                         rcu_read_unlock();
1375                         if (WARN_ON_ONCE(err < 0)) {
1376                                 kfree_skb(reply);
1377                                 goto out_free;
1378                         }
1379
1380                         ovs_notify(&dp_flow_genl_family, reply, info);
1381                 } else {
1382                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1383                 }
1384         }
1385
1386 out_free:
1387         ovs_flow_free(flow, true);
1388         return 0;
1389 unlock:
1390         ovs_unlock();
1391         return err;
1392 }
1393
1394 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1395 {
1396         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1397         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1398         struct table_instance *ti;
1399         struct datapath *dp;
1400         u32 ufid_flags;
1401         int err;
1402
1403         err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1404                                        OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1405         if (err)
1406                 return err;
1407         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1408
1409         rcu_read_lock();
1410         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1411         if (!dp) {
1412                 rcu_read_unlock();
1413                 return -ENODEV;
1414         }
1415
1416         ti = rcu_dereference(dp->table.ti);
1417         for (;;) {
1418                 struct sw_flow *flow;
1419                 u32 bucket, obj;
1420
1421                 bucket = cb->args[0];
1422                 obj = cb->args[1];
1423                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1424                 if (!flow)
1425                         break;
1426
1427                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1428                                            NETLINK_CB(cb->skb).portid,
1429                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1430                                            OVS_FLOW_CMD_GET, ufid_flags) < 0)
1431                         break;
1432
1433                 cb->args[0] = bucket;
1434                 cb->args[1] = obj;
1435         }
1436         rcu_read_unlock();
1437         return skb->len;
1438 }
1439
1440 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1441         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1442         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1443         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1444         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1445         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1446         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1447         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1448 };
1449
1450 static const struct genl_ops dp_flow_genl_ops[] = {
1451         { .cmd = OVS_FLOW_CMD_NEW,
1452           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1453           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1454           .doit = ovs_flow_cmd_new
1455         },
1456         { .cmd = OVS_FLOW_CMD_DEL,
1457           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1458           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1459           .doit = ovs_flow_cmd_del
1460         },
1461         { .cmd = OVS_FLOW_CMD_GET,
1462           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1463           .flags = 0,               /* OK for unprivileged users. */
1464           .doit = ovs_flow_cmd_get,
1465           .dumpit = ovs_flow_cmd_dump
1466         },
1467         { .cmd = OVS_FLOW_CMD_SET,
1468           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1469           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1470           .doit = ovs_flow_cmd_set,
1471         },
1472 };
1473
1474 static struct genl_family dp_flow_genl_family __ro_after_init = {
1475         .hdrsize = sizeof(struct ovs_header),
1476         .name = OVS_FLOW_FAMILY,
1477         .version = OVS_FLOW_VERSION,
1478         .maxattr = OVS_FLOW_ATTR_MAX,
1479         .policy = flow_policy,
1480         .netnsok = true,
1481         .parallel_ops = true,
1482         .ops = dp_flow_genl_ops,
1483         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1484         .mcgrps = &ovs_dp_flow_multicast_group,
1485         .n_mcgrps = 1,
1486         .module = THIS_MODULE,
1487 };
1488
1489 static size_t ovs_dp_cmd_msg_size(void)
1490 {
1491         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1492
1493         msgsize += nla_total_size(IFNAMSIZ);
1494         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1495         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1496         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1497
1498         return msgsize;
1499 }
1500
1501 /* Called with ovs_mutex. */
1502 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1503                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1504 {
1505         struct ovs_header *ovs_header;
1506         struct ovs_dp_stats dp_stats;
1507         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1508         int err;
1509
1510         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1511                                    flags, cmd);
1512         if (!ovs_header)
1513                 goto error;
1514
1515         ovs_header->dp_ifindex = get_dpifindex(dp);
1516
1517         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1518         if (err)
1519                 goto nla_put_failure;
1520
1521         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1522         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1523                           &dp_stats, OVS_DP_ATTR_PAD))
1524                 goto nla_put_failure;
1525
1526         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1527                           sizeof(struct ovs_dp_megaflow_stats),
1528                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1529                 goto nla_put_failure;
1530
1531         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1532                 goto nla_put_failure;
1533
1534         genlmsg_end(skb, ovs_header);
1535         return 0;
1536
1537 nla_put_failure:
1538         genlmsg_cancel(skb, ovs_header);
1539 error:
1540         return -EMSGSIZE;
1541 }
1542
1543 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1544 {
1545         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1546 }
1547
1548 /* Called with rcu_read_lock or ovs_mutex. */
1549 static struct datapath *lookup_datapath(struct net *net,
1550                                         const struct ovs_header *ovs_header,
1551                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1552 {
1553         struct datapath *dp;
1554
1555         if (!a[OVS_DP_ATTR_NAME])
1556                 dp = get_dp(net, ovs_header->dp_ifindex);
1557         else {
1558                 struct vport *vport;
1559
1560                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1561                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1562         }
1563         return dp ? dp : ERR_PTR(-ENODEV);
1564 }
1565
1566 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1567 {
1568         struct datapath *dp;
1569
1570         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1571         if (IS_ERR(dp))
1572                 return;
1573
1574         WARN(dp->user_features, "Dropping previously announced user features\n");
1575         dp->user_features = 0;
1576 }
1577
1578 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1579
1580 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1581 {
1582         u32 user_features = 0;
1583
1584         if (a[OVS_DP_ATTR_USER_FEATURES]) {
1585                 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1586
1587                 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1588                                       OVS_DP_F_UNALIGNED |
1589                                       OVS_DP_F_TC_RECIRC_SHARING))
1590                         return -EOPNOTSUPP;
1591
1592 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1593                 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1594                         return -EOPNOTSUPP;
1595 #endif
1596         }
1597
1598         dp->user_features = user_features;
1599
1600         if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1601                 static_branch_enable(&tc_recirc_sharing_support);
1602         else
1603                 static_branch_disable(&tc_recirc_sharing_support);
1604
1605         return 0;
1606 }
1607
1608 static int ovs_dp_stats_init(struct datapath *dp)
1609 {
1610         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1611         if (!dp->stats_percpu)
1612                 return -ENOMEM;
1613
1614         return 0;
1615 }
1616
1617 static int ovs_dp_vport_init(struct datapath *dp)
1618 {
1619         int i;
1620
1621         dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1622                                   sizeof(struct hlist_head),
1623                                   GFP_KERNEL);
1624         if (!dp->ports)
1625                 return -ENOMEM;
1626
1627         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1628                 INIT_HLIST_HEAD(&dp->ports[i]);
1629
1630         return 0;
1631 }
1632
1633 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1634 {
1635         struct nlattr **a = info->attrs;
1636         struct vport_parms parms;
1637         struct sk_buff *reply;
1638         struct datapath *dp;
1639         struct vport *vport;
1640         struct ovs_net *ovs_net;
1641         int err;
1642
1643         err = -EINVAL;
1644         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1645                 goto err;
1646
1647         reply = ovs_dp_cmd_alloc_info();
1648         if (!reply)
1649                 return -ENOMEM;
1650
1651         err = -ENOMEM;
1652         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1653         if (dp == NULL)
1654                 goto err_destroy_reply;
1655
1656         ovs_dp_set_net(dp, sock_net(skb->sk));
1657
1658         /* Allocate table. */
1659         err = ovs_flow_tbl_init(&dp->table);
1660         if (err)
1661                 goto err_destroy_dp;
1662
1663         err = ovs_dp_stats_init(dp);
1664         if (err)
1665                 goto err_destroy_table;
1666
1667         err = ovs_dp_vport_init(dp);
1668         if (err)
1669                 goto err_destroy_stats;
1670
1671         err = ovs_meters_init(dp);
1672         if (err)
1673                 goto err_destroy_ports;
1674
1675         /* Set up our datapath device. */
1676         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1677         parms.type = OVS_VPORT_TYPE_INTERNAL;
1678         parms.options = NULL;
1679         parms.dp = dp;
1680         parms.port_no = OVSP_LOCAL;
1681         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1682
1683         err = ovs_dp_change(dp, a);
1684         if (err)
1685                 goto err_destroy_meters;
1686
1687         /* So far only local changes have been made, now need the lock. */
1688         ovs_lock();
1689
1690         vport = new_vport(&parms);
1691         if (IS_ERR(vport)) {
1692                 err = PTR_ERR(vport);
1693                 if (err == -EBUSY)
1694                         err = -EEXIST;
1695
1696                 if (err == -EEXIST) {
1697                         /* An outdated user space instance that does not understand
1698                          * the concept of user_features has attempted to create a new
1699                          * datapath and is likely to reuse it. Drop all user features.
1700                          */
1701                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1702                                 ovs_dp_reset_user_features(skb, info);
1703                 }
1704
1705                 ovs_unlock();
1706                 goto err_destroy_meters;
1707         }
1708
1709         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1710                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1711         BUG_ON(err < 0);
1712
1713         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1714         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1715
1716         ovs_unlock();
1717
1718         ovs_notify(&dp_datapath_genl_family, reply, info);
1719         return 0;
1720
1721 err_destroy_meters:
1722         ovs_meters_exit(dp);
1723 err_destroy_ports:
1724         kfree(dp->ports);
1725 err_destroy_stats:
1726         free_percpu(dp->stats_percpu);
1727 err_destroy_table:
1728         ovs_flow_tbl_destroy(&dp->table);
1729 err_destroy_dp:
1730         kfree(dp);
1731 err_destroy_reply:
1732         kfree_skb(reply);
1733 err:
1734         return err;
1735 }
1736
1737 /* Called with ovs_mutex. */
1738 static void __dp_destroy(struct datapath *dp)
1739 {
1740         int i;
1741
1742         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1743                 struct vport *vport;
1744                 struct hlist_node *n;
1745
1746                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1747                         if (vport->port_no != OVSP_LOCAL)
1748                                 ovs_dp_detach_port(vport);
1749         }
1750
1751         list_del_rcu(&dp->list_node);
1752
1753         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1754          * all ports in datapath are destroyed first before freeing datapath.
1755          */
1756         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1757
1758         /* RCU destroy the flow table */
1759         call_rcu(&dp->rcu, destroy_dp_rcu);
1760 }
1761
1762 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1763 {
1764         struct sk_buff *reply;
1765         struct datapath *dp;
1766         int err;
1767
1768         reply = ovs_dp_cmd_alloc_info();
1769         if (!reply)
1770                 return -ENOMEM;
1771
1772         ovs_lock();
1773         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1774         err = PTR_ERR(dp);
1775         if (IS_ERR(dp))
1776                 goto err_unlock_free;
1777
1778         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1779                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1780         BUG_ON(err < 0);
1781
1782         __dp_destroy(dp);
1783         ovs_unlock();
1784
1785         ovs_notify(&dp_datapath_genl_family, reply, info);
1786
1787         return 0;
1788
1789 err_unlock_free:
1790         ovs_unlock();
1791         kfree_skb(reply);
1792         return err;
1793 }
1794
1795 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1796 {
1797         struct sk_buff *reply;
1798         struct datapath *dp;
1799         int err;
1800
1801         reply = ovs_dp_cmd_alloc_info();
1802         if (!reply)
1803                 return -ENOMEM;
1804
1805         ovs_lock();
1806         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1807         err = PTR_ERR(dp);
1808         if (IS_ERR(dp))
1809                 goto err_unlock_free;
1810
1811         err = ovs_dp_change(dp, info->attrs);
1812         if (err)
1813                 goto err_unlock_free;
1814
1815         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1816                                    info->snd_seq, 0, OVS_DP_CMD_SET);
1817         BUG_ON(err < 0);
1818
1819         ovs_unlock();
1820         ovs_notify(&dp_datapath_genl_family, reply, info);
1821
1822         return 0;
1823
1824 err_unlock_free:
1825         ovs_unlock();
1826         kfree_skb(reply);
1827         return err;
1828 }
1829
1830 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1831 {
1832         struct sk_buff *reply;
1833         struct datapath *dp;
1834         int err;
1835
1836         reply = ovs_dp_cmd_alloc_info();
1837         if (!reply)
1838                 return -ENOMEM;
1839
1840         ovs_lock();
1841         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1842         if (IS_ERR(dp)) {
1843                 err = PTR_ERR(dp);
1844                 goto err_unlock_free;
1845         }
1846         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1847                                    info->snd_seq, 0, OVS_DP_CMD_GET);
1848         BUG_ON(err < 0);
1849         ovs_unlock();
1850
1851         return genlmsg_reply(reply, info);
1852
1853 err_unlock_free:
1854         ovs_unlock();
1855         kfree_skb(reply);
1856         return err;
1857 }
1858
1859 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1860 {
1861         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1862         struct datapath *dp;
1863         int skip = cb->args[0];
1864         int i = 0;
1865
1866         ovs_lock();
1867         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1868                 if (i >= skip &&
1869                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1870                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1871                                          OVS_DP_CMD_GET) < 0)
1872                         break;
1873                 i++;
1874         }
1875         ovs_unlock();
1876
1877         cb->args[0] = i;
1878
1879         return skb->len;
1880 }
1881
1882 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1883         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1884         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1885         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1886 };
1887
1888 static const struct genl_ops dp_datapath_genl_ops[] = {
1889         { .cmd = OVS_DP_CMD_NEW,
1890           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1891           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1892           .doit = ovs_dp_cmd_new
1893         },
1894         { .cmd = OVS_DP_CMD_DEL,
1895           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1896           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1897           .doit = ovs_dp_cmd_del
1898         },
1899         { .cmd = OVS_DP_CMD_GET,
1900           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1901           .flags = 0,               /* OK for unprivileged users. */
1902           .doit = ovs_dp_cmd_get,
1903           .dumpit = ovs_dp_cmd_dump
1904         },
1905         { .cmd = OVS_DP_CMD_SET,
1906           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1907           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1908           .doit = ovs_dp_cmd_set,
1909         },
1910 };
1911
1912 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1913         .hdrsize = sizeof(struct ovs_header),
1914         .name = OVS_DATAPATH_FAMILY,
1915         .version = OVS_DATAPATH_VERSION,
1916         .maxattr = OVS_DP_ATTR_MAX,
1917         .policy = datapath_policy,
1918         .netnsok = true,
1919         .parallel_ops = true,
1920         .ops = dp_datapath_genl_ops,
1921         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1922         .mcgrps = &ovs_dp_datapath_multicast_group,
1923         .n_mcgrps = 1,
1924         .module = THIS_MODULE,
1925 };
1926
1927 /* Called with ovs_mutex or RCU read lock. */
1928 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1929                                    struct net *net, u32 portid, u32 seq,
1930                                    u32 flags, u8 cmd, gfp_t gfp)
1931 {
1932         struct ovs_header *ovs_header;
1933         struct ovs_vport_stats vport_stats;
1934         int err;
1935
1936         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1937                                  flags, cmd);
1938         if (!ovs_header)
1939                 return -EMSGSIZE;
1940
1941         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1942
1943         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1944             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1945             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1946                            ovs_vport_name(vport)) ||
1947             nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1948                 goto nla_put_failure;
1949
1950         if (!net_eq(net, dev_net(vport->dev))) {
1951                 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
1952
1953                 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1954                         goto nla_put_failure;
1955         }
1956
1957         ovs_vport_get_stats(vport, &vport_stats);
1958         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1959                           sizeof(struct ovs_vport_stats), &vport_stats,
1960                           OVS_VPORT_ATTR_PAD))
1961                 goto nla_put_failure;
1962
1963         if (ovs_vport_get_upcall_portids(vport, skb))
1964                 goto nla_put_failure;
1965
1966         err = ovs_vport_get_options(vport, skb);
1967         if (err == -EMSGSIZE)
1968                 goto error;
1969
1970         genlmsg_end(skb, ovs_header);
1971         return 0;
1972
1973 nla_put_failure:
1974         err = -EMSGSIZE;
1975 error:
1976         genlmsg_cancel(skb, ovs_header);
1977         return err;
1978 }
1979
1980 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1981 {
1982         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1983 }
1984
1985 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1986 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1987                                          u32 portid, u32 seq, u8 cmd)
1988 {
1989         struct sk_buff *skb;
1990         int retval;
1991
1992         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1993         if (!skb)
1994                 return ERR_PTR(-ENOMEM);
1995
1996         retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
1997                                          GFP_KERNEL);
1998         BUG_ON(retval < 0);
1999
2000         return skb;
2001 }
2002
2003 /* Called with ovs_mutex or RCU read lock. */
2004 static struct vport *lookup_vport(struct net *net,
2005                                   const struct ovs_header *ovs_header,
2006                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2007 {
2008         struct datapath *dp;
2009         struct vport *vport;
2010
2011         if (a[OVS_VPORT_ATTR_IFINDEX])
2012                 return ERR_PTR(-EOPNOTSUPP);
2013         if (a[OVS_VPORT_ATTR_NAME]) {
2014                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2015                 if (!vport)
2016                         return ERR_PTR(-ENODEV);
2017                 if (ovs_header->dp_ifindex &&
2018                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2019                         return ERR_PTR(-ENODEV);
2020                 return vport;
2021         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2022                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2023
2024                 if (port_no >= DP_MAX_PORTS)
2025                         return ERR_PTR(-EFBIG);
2026
2027                 dp = get_dp(net, ovs_header->dp_ifindex);
2028                 if (!dp)
2029                         return ERR_PTR(-ENODEV);
2030
2031                 vport = ovs_vport_ovsl_rcu(dp, port_no);
2032                 if (!vport)
2033                         return ERR_PTR(-ENODEV);
2034                 return vport;
2035         } else
2036                 return ERR_PTR(-EINVAL);
2037
2038 }
2039
2040 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2041 {
2042         unsigned int dev_headroom, max_headroom = 0;
2043         struct net_device *dev;
2044         struct vport *vport;
2045         int i;
2046
2047         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2048                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2049                         dev = vport->dev;
2050                         dev_headroom = netdev_get_fwd_headroom(dev);
2051                         if (dev_headroom > max_headroom)
2052                                 max_headroom = dev_headroom;
2053                 }
2054         }
2055
2056         return max_headroom;
2057 }
2058
2059 /* Called with ovs_mutex */
2060 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2061 {
2062         struct vport *vport;
2063         int i;
2064
2065         dp->max_headroom = new_headroom;
2066         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
2067                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
2068                         netdev_set_rx_headroom(vport->dev, new_headroom);
2069 }
2070
2071 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2072 {
2073         struct nlattr **a = info->attrs;
2074         struct ovs_header *ovs_header = info->userhdr;
2075         struct vport_parms parms;
2076         struct sk_buff *reply;
2077         struct vport *vport;
2078         struct datapath *dp;
2079         unsigned int new_headroom;
2080         u32 port_no;
2081         int err;
2082
2083         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2084             !a[OVS_VPORT_ATTR_UPCALL_PID])
2085                 return -EINVAL;
2086         if (a[OVS_VPORT_ATTR_IFINDEX])
2087                 return -EOPNOTSUPP;
2088
2089         port_no = a[OVS_VPORT_ATTR_PORT_NO]
2090                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2091         if (port_no >= DP_MAX_PORTS)
2092                 return -EFBIG;
2093
2094         reply = ovs_vport_cmd_alloc_info();
2095         if (!reply)
2096                 return -ENOMEM;
2097
2098         ovs_lock();
2099 restart:
2100         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2101         err = -ENODEV;
2102         if (!dp)
2103                 goto exit_unlock_free;
2104
2105         if (port_no) {
2106                 vport = ovs_vport_ovsl(dp, port_no);
2107                 err = -EBUSY;
2108                 if (vport)
2109                         goto exit_unlock_free;
2110         } else {
2111                 for (port_no = 1; ; port_no++) {
2112                         if (port_no >= DP_MAX_PORTS) {
2113                                 err = -EFBIG;
2114                                 goto exit_unlock_free;
2115                         }
2116                         vport = ovs_vport_ovsl(dp, port_no);
2117                         if (!vport)
2118                                 break;
2119                 }
2120         }
2121
2122         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2123         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2124         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2125         parms.dp = dp;
2126         parms.port_no = port_no;
2127         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2128
2129         vport = new_vport(&parms);
2130         err = PTR_ERR(vport);
2131         if (IS_ERR(vport)) {
2132                 if (err == -EAGAIN)
2133                         goto restart;
2134                 goto exit_unlock_free;
2135         }
2136
2137         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2138                                       info->snd_portid, info->snd_seq, 0,
2139                                       OVS_VPORT_CMD_NEW, GFP_KERNEL);
2140
2141         new_headroom = netdev_get_fwd_headroom(vport->dev);
2142
2143         if (new_headroom > dp->max_headroom)
2144                 ovs_update_headroom(dp, new_headroom);
2145         else
2146                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2147
2148         BUG_ON(err < 0);
2149         ovs_unlock();
2150
2151         ovs_notify(&dp_vport_genl_family, reply, info);
2152         return 0;
2153
2154 exit_unlock_free:
2155         ovs_unlock();
2156         kfree_skb(reply);
2157         return err;
2158 }
2159
2160 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2161 {
2162         struct nlattr **a = info->attrs;
2163         struct sk_buff *reply;
2164         struct vport *vport;
2165         int err;
2166
2167         reply = ovs_vport_cmd_alloc_info();
2168         if (!reply)
2169                 return -ENOMEM;
2170
2171         ovs_lock();
2172         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2173         err = PTR_ERR(vport);
2174         if (IS_ERR(vport))
2175                 goto exit_unlock_free;
2176
2177         if (a[OVS_VPORT_ATTR_TYPE] &&
2178             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2179                 err = -EINVAL;
2180                 goto exit_unlock_free;
2181         }
2182
2183         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2184                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2185                 if (err)
2186                         goto exit_unlock_free;
2187         }
2188
2189
2190         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2191                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2192
2193                 err = ovs_vport_set_upcall_portids(vport, ids);
2194                 if (err)
2195                         goto exit_unlock_free;
2196         }
2197
2198         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2199                                       info->snd_portid, info->snd_seq, 0,
2200                                       OVS_VPORT_CMD_SET, GFP_KERNEL);
2201         BUG_ON(err < 0);
2202
2203         ovs_unlock();
2204         ovs_notify(&dp_vport_genl_family, reply, info);
2205         return 0;
2206
2207 exit_unlock_free:
2208         ovs_unlock();
2209         kfree_skb(reply);
2210         return err;
2211 }
2212
2213 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2214 {
2215         bool update_headroom = false;
2216         struct nlattr **a = info->attrs;
2217         struct sk_buff *reply;
2218         struct datapath *dp;
2219         struct vport *vport;
2220         unsigned int new_headroom;
2221         int err;
2222
2223         reply = ovs_vport_cmd_alloc_info();
2224         if (!reply)
2225                 return -ENOMEM;
2226
2227         ovs_lock();
2228         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2229         err = PTR_ERR(vport);
2230         if (IS_ERR(vport))
2231                 goto exit_unlock_free;
2232
2233         if (vport->port_no == OVSP_LOCAL) {
2234                 err = -EINVAL;
2235                 goto exit_unlock_free;
2236         }
2237
2238         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2239                                       info->snd_portid, info->snd_seq, 0,
2240                                       OVS_VPORT_CMD_DEL, GFP_KERNEL);
2241         BUG_ON(err < 0);
2242
2243         /* the vport deletion may trigger dp headroom update */
2244         dp = vport->dp;
2245         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2246                 update_headroom = true;
2247
2248         netdev_reset_rx_headroom(vport->dev);
2249         ovs_dp_detach_port(vport);
2250
2251         if (update_headroom) {
2252                 new_headroom = ovs_get_max_headroom(dp);
2253
2254                 if (new_headroom < dp->max_headroom)
2255                         ovs_update_headroom(dp, new_headroom);
2256         }
2257         ovs_unlock();
2258
2259         ovs_notify(&dp_vport_genl_family, reply, info);
2260         return 0;
2261
2262 exit_unlock_free:
2263         ovs_unlock();
2264         kfree_skb(reply);
2265         return err;
2266 }
2267
2268 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2269 {
2270         struct nlattr **a = info->attrs;
2271         struct ovs_header *ovs_header = info->userhdr;
2272         struct sk_buff *reply;
2273         struct vport *vport;
2274         int err;
2275
2276         reply = ovs_vport_cmd_alloc_info();
2277         if (!reply)
2278                 return -ENOMEM;
2279
2280         rcu_read_lock();
2281         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2282         err = PTR_ERR(vport);
2283         if (IS_ERR(vport))
2284                 goto exit_unlock_free;
2285         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2286                                       info->snd_portid, info->snd_seq, 0,
2287                                       OVS_VPORT_CMD_GET, GFP_ATOMIC);
2288         BUG_ON(err < 0);
2289         rcu_read_unlock();
2290
2291         return genlmsg_reply(reply, info);
2292
2293 exit_unlock_free:
2294         rcu_read_unlock();
2295         kfree_skb(reply);
2296         return err;
2297 }
2298
2299 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2300 {
2301         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2302         struct datapath *dp;
2303         int bucket = cb->args[0], skip = cb->args[1];
2304         int i, j = 0;
2305
2306         rcu_read_lock();
2307         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2308         if (!dp) {
2309                 rcu_read_unlock();
2310                 return -ENODEV;
2311         }
2312         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2313                 struct vport *vport;
2314
2315                 j = 0;
2316                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2317                         if (j >= skip &&
2318                             ovs_vport_cmd_fill_info(vport, skb,
2319                                                     sock_net(skb->sk),
2320                                                     NETLINK_CB(cb->skb).portid,
2321                                                     cb->nlh->nlmsg_seq,
2322                                                     NLM_F_MULTI,
2323                                                     OVS_VPORT_CMD_GET,
2324                                                     GFP_ATOMIC) < 0)
2325                                 goto out;
2326
2327                         j++;
2328                 }
2329                 skip = 0;
2330         }
2331 out:
2332         rcu_read_unlock();
2333
2334         cb->args[0] = i;
2335         cb->args[1] = j;
2336
2337         return skb->len;
2338 }
2339
2340 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2341         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2342         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2343         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2344         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2345         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2346         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2347         [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2348         [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2349 };
2350
2351 static const struct genl_ops dp_vport_genl_ops[] = {
2352         { .cmd = OVS_VPORT_CMD_NEW,
2353           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2354           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2355           .doit = ovs_vport_cmd_new
2356         },
2357         { .cmd = OVS_VPORT_CMD_DEL,
2358           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2359           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2360           .doit = ovs_vport_cmd_del
2361         },
2362         { .cmd = OVS_VPORT_CMD_GET,
2363           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2364           .flags = 0,               /* OK for unprivileged users. */
2365           .doit = ovs_vport_cmd_get,
2366           .dumpit = ovs_vport_cmd_dump
2367         },
2368         { .cmd = OVS_VPORT_CMD_SET,
2369           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2370           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2371           .doit = ovs_vport_cmd_set,
2372         },
2373 };
2374
2375 struct genl_family dp_vport_genl_family __ro_after_init = {
2376         .hdrsize = sizeof(struct ovs_header),
2377         .name = OVS_VPORT_FAMILY,
2378         .version = OVS_VPORT_VERSION,
2379         .maxattr = OVS_VPORT_ATTR_MAX,
2380         .policy = vport_policy,
2381         .netnsok = true,
2382         .parallel_ops = true,
2383         .ops = dp_vport_genl_ops,
2384         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2385         .mcgrps = &ovs_dp_vport_multicast_group,
2386         .n_mcgrps = 1,
2387         .module = THIS_MODULE,
2388 };
2389
2390 static struct genl_family * const dp_genl_families[] = {
2391         &dp_datapath_genl_family,
2392         &dp_vport_genl_family,
2393         &dp_flow_genl_family,
2394         &dp_packet_genl_family,
2395         &dp_meter_genl_family,
2396 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2397         &dp_ct_limit_genl_family,
2398 #endif
2399 };
2400
2401 static void dp_unregister_genl(int n_families)
2402 {
2403         int i;
2404
2405         for (i = 0; i < n_families; i++)
2406                 genl_unregister_family(dp_genl_families[i]);
2407 }
2408
2409 static int __init dp_register_genl(void)
2410 {
2411         int err;
2412         int i;
2413
2414         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2415
2416                 err = genl_register_family(dp_genl_families[i]);
2417                 if (err)
2418                         goto error;
2419         }
2420
2421         return 0;
2422
2423 error:
2424         dp_unregister_genl(i);
2425         return err;
2426 }
2427
2428 static int __net_init ovs_init_net(struct net *net)
2429 {
2430         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2431
2432         INIT_LIST_HEAD(&ovs_net->dps);
2433         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2434         return ovs_ct_init(net);
2435 }
2436
2437 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2438                                             struct list_head *head)
2439 {
2440         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2441         struct datapath *dp;
2442
2443         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2444                 int i;
2445
2446                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2447                         struct vport *vport;
2448
2449                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2450                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2451                                         continue;
2452
2453                                 if (dev_net(vport->dev) == dnet)
2454                                         list_add(&vport->detach_list, head);
2455                         }
2456                 }
2457         }
2458 }
2459
2460 static void __net_exit ovs_exit_net(struct net *dnet)
2461 {
2462         struct datapath *dp, *dp_next;
2463         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2464         struct vport *vport, *vport_next;
2465         struct net *net;
2466         LIST_HEAD(head);
2467
2468         ovs_ct_exit(dnet);
2469         ovs_lock();
2470         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2471                 __dp_destroy(dp);
2472
2473         down_read(&net_rwsem);
2474         for_each_net(net)
2475                 list_vports_from_net(net, dnet, &head);
2476         up_read(&net_rwsem);
2477
2478         /* Detach all vports from given namespace. */
2479         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2480                 list_del(&vport->detach_list);
2481                 ovs_dp_detach_port(vport);
2482         }
2483
2484         ovs_unlock();
2485
2486         cancel_work_sync(&ovs_net->dp_notify_work);
2487 }
2488
2489 static struct pernet_operations ovs_net_ops = {
2490         .init = ovs_init_net,
2491         .exit = ovs_exit_net,
2492         .id   = &ovs_net_id,
2493         .size = sizeof(struct ovs_net),
2494 };
2495
2496 static int __init dp_init(void)
2497 {
2498         int err;
2499
2500         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2501
2502         pr_info("Open vSwitch switching datapath\n");
2503
2504         err = action_fifos_init();
2505         if (err)
2506                 goto error;
2507
2508         err = ovs_internal_dev_rtnl_link_register();
2509         if (err)
2510                 goto error_action_fifos_exit;
2511
2512         err = ovs_flow_init();
2513         if (err)
2514                 goto error_unreg_rtnl_link;
2515
2516         err = ovs_vport_init();
2517         if (err)
2518                 goto error_flow_exit;
2519
2520         err = register_pernet_device(&ovs_net_ops);
2521         if (err)
2522                 goto error_vport_exit;
2523
2524         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2525         if (err)
2526                 goto error_netns_exit;
2527
2528         err = ovs_netdev_init();
2529         if (err)
2530                 goto error_unreg_notifier;
2531
2532         err = dp_register_genl();
2533         if (err < 0)
2534                 goto error_unreg_netdev;
2535
2536         return 0;
2537
2538 error_unreg_netdev:
2539         ovs_netdev_exit();
2540 error_unreg_notifier:
2541         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2542 error_netns_exit:
2543         unregister_pernet_device(&ovs_net_ops);
2544 error_vport_exit:
2545         ovs_vport_exit();
2546 error_flow_exit:
2547         ovs_flow_exit();
2548 error_unreg_rtnl_link:
2549         ovs_internal_dev_rtnl_link_unregister();
2550 error_action_fifos_exit:
2551         action_fifos_exit();
2552 error:
2553         return err;
2554 }
2555
2556 static void dp_cleanup(void)
2557 {
2558         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2559         ovs_netdev_exit();
2560         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2561         unregister_pernet_device(&ovs_net_ops);
2562         rcu_barrier();
2563         ovs_vport_exit();
2564         ovs_flow_exit();
2565         ovs_internal_dev_rtnl_link_unregister();
2566         action_fifos_exit();
2567 }
2568
2569 module_init(dp_init);
2570 module_exit(dp_cleanup);
2571
2572 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2573 MODULE_LICENSE("GPL");
2574 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2575 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2576 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2577 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2578 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2579 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);