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