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 static void ovs_dp_masks_rebalance(struct work_struct *work);
 134
 135 /* Must be called with rcu_read_lock or ovs_mutex. */
 136 const char *ovs_dp_name(const struct datapath *dp)
 137 {
 138         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 139         return ovs_vport_name(vport);
 140 }
 141
 142 static int get_dpifindex(const struct datapath *dp)
 143 {
 144         struct vport *local;
 145         int ifindex;
 146
 147         rcu_read_lock();
 148
 149         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 150         if (local)
 151                 ifindex = local->dev->ifindex;
 152         else
 153                 ifindex = 0;
 154
 155         rcu_read_unlock();
 156
 157         return ifindex;
 158 }
 159
 160 static void destroy_dp_rcu(struct rcu_head *rcu)
 161 {
 162         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 163
 164         ovs_flow_tbl_destroy(&dp->table);
 165         free_percpu(dp->stats_percpu);
 166         kfree(dp->ports);
 167         ovs_meters_exit(dp);
 168         kfree(dp);
 169 }
 170
 171 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 172                                             u16 port_no)
 173 {
 174         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 175 }
 176
 177 /* Called with ovs_mutex or RCU read lock. */
 178 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 179 {
 180         struct vport *vport;
 181         struct hlist_head *head;
 182
 183         head = vport_hash_bucket(dp, port_no);
 184         hlist_for_each_entry_rcu(vport, head, dp_hash_node,
 185                                 lockdep_ovsl_is_held()) {
 186                 if (vport->port_no == port_no)
 187                         return vport;
 188         }
 189         return NULL;
 190 }
 191
 192 /* Called with ovs_mutex. */
 193 static struct vport *new_vport(const struct vport_parms *parms)
 194 {
 195         struct vport *vport;
 196
 197         vport = ovs_vport_add(parms);
 198         if (!IS_ERR(vport)) {
 199                 struct datapath *dp = parms->dp;
 200                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 201
 202                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 203         }
 204         return vport;
 205 }
 206
 207 void ovs_dp_detach_port(struct vport *p)
 208 {
 209         ASSERT_OVSL();
 210
 211         /* First drop references to device. */
 212         hlist_del_rcu(&p->dp_hash_node);
 213
 214         /* Then destroy it. */
 215         ovs_vport_del(p);
 216 }
 217
 218 /* Must be called with rcu_read_lock. */
 219 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 220 {
 221         const struct vport *p = OVS_CB(skb)->input_vport;
 222         struct datapath *dp = p->dp;
 223         struct sw_flow *flow;
 224         struct sw_flow_actions *sf_acts;
 225         struct dp_stats_percpu *stats;
 226         u64 *stats_counter;
 227         u32 n_mask_hit;
 228         int error;
 229
 230         stats = this_cpu_ptr(dp->stats_percpu);
 231
 232         /* Look up flow. */
 233         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
 234                                          &n_mask_hit);
 235         if (unlikely(!flow)) {
 236                 struct dp_upcall_info upcall;
 237
 238                 memset(&upcall, 0, sizeof(upcall));
 239                 upcall.cmd = OVS_PACKET_CMD_MISS;
 240                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 241                 upcall.mru = OVS_CB(skb)->mru;
 242                 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
 243                 if (unlikely(error))
 244                         kfree_skb(skb);
 245                 else
 246                         consume_skb(skb);
 247                 stats_counter = &stats->n_missed;
 248                 goto out;
 249         }
 250
 251         ovs_flow_stats_update(flow, key->tp.flags, skb);
 252         sf_acts = rcu_dereference(flow->sf_acts);
 253         error = ovs_execute_actions(dp, skb, sf_acts, key);
 254         if (unlikely(error))
 255                 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
 256                                                         ovs_dp_name(dp), error);
 257
 258         stats_counter = &stats->n_hit;
 259
 260 out:
 261         /* Update datapath statistics. */
 262         u64_stats_update_begin(&stats->syncp);
 263         (*stats_counter)++;
 264         stats->n_mask_hit += n_mask_hit;
 265         u64_stats_update_end(&stats->syncp);
 266 }
 267
 268 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 269                   const struct sw_flow_key *key,
 270                   const struct dp_upcall_info *upcall_info,
 271                   uint32_t cutlen)
 272 {
 273         struct dp_stats_percpu *stats;
 274         int err;
 275
 276         if (upcall_info->portid == 0) {
 277                 err = -ENOTCONN;
 278                 goto err;
 279         }
 280
 281         if (!skb_is_gso(skb))
 282                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 283         else
 284                 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
 285         if (err)
 286                 goto err;
 287
 288         return 0;
 289
 290 err:
 291         stats = this_cpu_ptr(dp->stats_percpu);
 292
 293         u64_stats_update_begin(&stats->syncp);
 294         stats->n_lost++;
 295         u64_stats_update_end(&stats->syncp);
 296
 297         return err;
 298 }
 299
 300 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 301                              const struct sw_flow_key *key,
 302                              const struct dp_upcall_info *upcall_info,
 303                                  uint32_t cutlen)
 304 {
 305         unsigned int gso_type = skb_shinfo(skb)->gso_type;
 306         struct sw_flow_key later_key;
 307         struct sk_buff *segs, *nskb;
 308         int err;
 309
 310         BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_GSO_CB_OFFSET);
 311         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 312         if (IS_ERR(segs))
 313                 return PTR_ERR(segs);
 314         if (segs == NULL)
 315                 return -EINVAL;
 316
 317         if (gso_type & SKB_GSO_UDP) {
 318                 /* The initial flow key extracted by ovs_flow_key_extract()
 319                  * in this case is for a first fragment, so we need to
 320                  * properly mark later fragments.
 321                  */
 322                 later_key = *key;
 323                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 324         }
 325
 326         /* Queue all of the segments. */
 327         skb_list_walk_safe(segs, skb, nskb) {
 328                 if (gso_type & SKB_GSO_UDP && skb != segs)
 329                         key = &later_key;
 330
 331                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 332                 if (err)
 333                         break;
 334
 335         }
 336
 337         /* Free all of the segments. */
 338         skb_list_walk_safe(segs, skb, nskb) {
 339                 if (err)
 340                         kfree_skb(skb);
 341                 else
 342                         consume_skb(skb);
 343         }
 344         return err;
 345 }
 346
 347 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 348                               unsigned int hdrlen, int actions_attrlen)
 349 {
 350         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 351                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 352                 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 353                 + nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
 354                 + nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
 355
 356         /* OVS_PACKET_ATTR_USERDATA */
 357         if (upcall_info->userdata)
 358                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 359
 360         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 361         if (upcall_info->egress_tun_info)
 362                 size += nla_total_size(ovs_tun_key_attr_size());
 363
 364         /* OVS_PACKET_ATTR_ACTIONS */
 365         if (upcall_info->actions_len)
 366                 size += nla_total_size(actions_attrlen);
 367
 368         /* OVS_PACKET_ATTR_MRU */
 369         if (upcall_info->mru)
 370                 size += nla_total_size(sizeof(upcall_info->mru));
 371
 372         return size;
 373 }
 374
 375 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 376 {
 377         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 378                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 379
 380                 if (plen > 0)
 381                         skb_put_zero(skb, plen);
 382         }
 383 }
 384
 385 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 386                                   const struct sw_flow_key *key,
 387                                   const struct dp_upcall_info *upcall_info,
 388                                   uint32_t cutlen)
 389 {
 390         struct ovs_header *upcall;
 391         struct sk_buff *nskb = NULL;
 392         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 393         struct nlattr *nla;
 394         size_t len;
 395         unsigned int hlen;
 396         int err, dp_ifindex;
 397         u64 hash;
 398
 399         dp_ifindex = get_dpifindex(dp);
 400         if (!dp_ifindex)
 401                 return -ENODEV;
 402
 403         if (skb_vlan_tag_present(skb)) {
 404                 nskb = skb_clone(skb, GFP_ATOMIC);
 405                 if (!nskb)
 406                         return -ENOMEM;
 407
 408                 nskb = __vlan_hwaccel_push_inside(nskb);
 409                 if (!nskb)
 410                         return -ENOMEM;
 411
 412                 skb = nskb;
 413         }
 414
 415         if (nla_attr_size(skb->len) > USHRT_MAX) {
 416                 err = -EFBIG;
 417                 goto out;
 418         }
 419
 420         /* Complete checksum if needed */
 421         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 422             (err = skb_csum_hwoffload_help(skb, 0)))
 423                 goto out;
 424
 425         /* Older versions of OVS user space enforce alignment of the last
 426          * Netlink attribute to NLA_ALIGNTO which would require extensive
 427          * padding logic. Only perform zerocopy if padding is not required.
 428          */
 429         if (dp->user_features & OVS_DP_F_UNALIGNED)
 430                 hlen = skb_zerocopy_headlen(skb);
 431         else
 432                 hlen = skb->len;
 433
 434         len = upcall_msg_size(upcall_info, hlen - cutlen,
 435                               OVS_CB(skb)->acts_origlen);
 436         user_skb = genlmsg_new(len, GFP_ATOMIC);
 437         if (!user_skb) {
 438                 err = -ENOMEM;
 439                 goto out;
 440         }
 441
 442         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 443                              0, upcall_info->cmd);
 444         if (!upcall) {
 445                 err = -EINVAL;
 446                 goto out;
 447         }
 448         upcall->dp_ifindex = dp_ifindex;
 449
 450         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 451         if (err)
 452                 goto out;
 453
 454         if (upcall_info->userdata)
 455                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 456                           nla_len(upcall_info->userdata),
 457                           nla_data(upcall_info->userdata));
 458
 459         if (upcall_info->egress_tun_info) {
 460                 nla = nla_nest_start_noflag(user_skb,
 461                                             OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 462                 if (!nla) {
 463                         err = -EMSGSIZE;
 464                         goto out;
 465                 }
 466                 err = ovs_nla_put_tunnel_info(user_skb,
 467                                               upcall_info->egress_tun_info);
 468                 if (err)
 469                         goto out;
 470
 471                 nla_nest_end(user_skb, nla);
 472         }
 473
 474         if (upcall_info->actions_len) {
 475                 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
 476                 if (!nla) {
 477                         err = -EMSGSIZE;
 478                         goto out;
 479                 }
 480                 err = ovs_nla_put_actions(upcall_info->actions,
 481                                           upcall_info->actions_len,
 482                                           user_skb);
 483                 if (!err)
 484                         nla_nest_end(user_skb, nla);
 485                 else
 486                         nla_nest_cancel(user_skb, nla);
 487         }
 488
 489         /* Add OVS_PACKET_ATTR_MRU */
 490         if (upcall_info->mru &&
 491             nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
 492                 err = -ENOBUFS;
 493                 goto out;
 494         }
 495
 496         /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 497         if (cutlen > 0 &&
 498             nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
 499                 err = -ENOBUFS;
 500                 goto out;
 501         }
 502
 503         /* Add OVS_PACKET_ATTR_HASH */
 504         hash = skb_get_hash_raw(skb);
 505         if (skb->sw_hash)
 506                 hash |= OVS_PACKET_HASH_SW_BIT;
 507
 508         if (skb->l4_hash)
 509                 hash |= OVS_PACKET_HASH_L4_BIT;
 510
 511         if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
 512                 err = -ENOBUFS;
 513                 goto out;
 514         }
 515
 516         /* Only reserve room for attribute header, packet data is added
 517          * in skb_zerocopy() */
 518         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 519                 err = -ENOBUFS;
 520                 goto out;
 521         }
 522         nla->nla_len = nla_attr_size(skb->len - cutlen);
 523
 524         err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 525         if (err)
 526                 goto out;
 527
 528         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 529         pad_packet(dp, user_skb);
 530
 531         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 532
 533         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 534         user_skb = NULL;
 535 out:
 536         if (err)
 537                 skb_tx_error(skb);
 538         kfree_skb(user_skb);
 539         kfree_skb(nskb);
 540         return err;
 541 }
 542
 543 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 544 {
 545         struct ovs_header *ovs_header = info->userhdr;
 546         struct net *net = sock_net(skb->sk);
 547         struct nlattr **a = info->attrs;
 548         struct sw_flow_actions *acts;
 549         struct sk_buff *packet;
 550         struct sw_flow *flow;
 551         struct sw_flow_actions *sf_acts;
 552         struct datapath *dp;
 553         struct vport *input_vport;
 554         u16 mru = 0;
 555         u64 hash;
 556         int len;
 557         int err;
 558         bool log = !a[OVS_PACKET_ATTR_PROBE];
 559
 560         err = -EINVAL;
 561         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 562             !a[OVS_PACKET_ATTR_ACTIONS])
 563                 goto err;
 564
 565         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 566         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 567         err = -ENOMEM;
 568         if (!packet)
 569                 goto err;
 570         skb_reserve(packet, NET_IP_ALIGN);
 571
 572         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 573
 574         /* Set packet's mru */
 575         if (a[OVS_PACKET_ATTR_MRU]) {
 576                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 577                 packet->ignore_df = 1;
 578         }
 579         OVS_CB(packet)->mru = mru;
 580
 581         if (a[OVS_PACKET_ATTR_HASH]) {
 582                 hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
 583
 584                 __skb_set_hash(packet, hash & 0xFFFFFFFFULL,
 585                                !!(hash & OVS_PACKET_HASH_SW_BIT),
 586                                !!(hash & OVS_PACKET_HASH_L4_BIT));
 587         }
 588
 589         /* Build an sw_flow for sending this packet. */
 590         flow = ovs_flow_alloc();
 591         err = PTR_ERR(flow);
 592         if (IS_ERR(flow))
 593                 goto err_kfree_skb;
 594
 595         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 596                                              packet, &flow->key, log);
 597         if (err)
 598                 goto err_flow_free;
 599
 600         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 601                                    &flow->key, &acts, log);
 602         if (err)
 603                 goto err_flow_free;
 604
 605         rcu_assign_pointer(flow->sf_acts, acts);
 606         packet->priority = flow->key.phy.priority;
 607         packet->mark = flow->key.phy.skb_mark;
 608
 609         rcu_read_lock();
 610         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 611         err = -ENODEV;
 612         if (!dp)
 613                 goto err_unlock;
 614
 615         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 616         if (!input_vport)
 617                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 618
 619         if (!input_vport)
 620                 goto err_unlock;
 621
 622         packet->dev = input_vport->dev;
 623         OVS_CB(packet)->input_vport = input_vport;
 624         sf_acts = rcu_dereference(flow->sf_acts);
 625
 626         local_bh_disable();
 627         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 628         local_bh_enable();
 629         rcu_read_unlock();
 630
 631         ovs_flow_free(flow, false);
 632         return err;
 633
 634 err_unlock:
 635         rcu_read_unlock();
 636 err_flow_free:
 637         ovs_flow_free(flow, false);
 638 err_kfree_skb:
 639         kfree_skb(packet);
 640 err:
 641         return err;
 642 }
 643
 644 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 645         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 646         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 647         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 648         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 649         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 650         [OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
 651 };
 652
 653 static const struct genl_ops dp_packet_genl_ops[] = {
 654         { .cmd = OVS_PACKET_CMD_EXECUTE,
 655           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 656           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 657           .doit = ovs_packet_cmd_execute
 658         }
 659 };
 660
 661 static struct genl_family dp_packet_genl_family __ro_after_init = {
 662         .hdrsize = sizeof(struct ovs_header),
 663         .name = OVS_PACKET_FAMILY,
 664         .version = OVS_PACKET_VERSION,
 665         .maxattr = OVS_PACKET_ATTR_MAX,
 666         .policy = packet_policy,
 667         .netnsok = true,
 668         .parallel_ops = true,
 669         .ops = dp_packet_genl_ops,
 670         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 671         .module = THIS_MODULE,
 672 };
 673
 674 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 675                          struct ovs_dp_megaflow_stats *mega_stats)
 676 {
 677         int i;
 678
 679         memset(mega_stats, 0, sizeof(*mega_stats));
 680
 681         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 682         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 683
 684         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 685
 686         for_each_possible_cpu(i) {
 687                 const struct dp_stats_percpu *percpu_stats;
 688                 struct dp_stats_percpu local_stats;
 689                 unsigned int start;
 690
 691                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 692
 693                 do {
 694                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 695                         local_stats = *percpu_stats;
 696                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 697
 698                 stats->n_hit += local_stats.n_hit;
 699                 stats->n_missed += local_stats.n_missed;
 700                 stats->n_lost += local_stats.n_lost;
 701                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 702         }
 703 }
 704
 705 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 706 {
 707         return ovs_identifier_is_ufid(sfid) &&
 708                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 709 }
 710
 711 static bool should_fill_mask(uint32_t ufid_flags)
 712 {
 713         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 714 }
 715
 716 static bool should_fill_actions(uint32_t ufid_flags)
 717 {
 718         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 719 }
 720
 721 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 722                                     const struct sw_flow_id *sfid,
 723                                     uint32_t ufid_flags)
 724 {
 725         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 726
 727         /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
 728          * see ovs_nla_put_identifier()
 729          */
 730         if (sfid && ovs_identifier_is_ufid(sfid))
 731                 len += nla_total_size(sfid->ufid_len);
 732         else
 733                 len += nla_total_size(ovs_key_attr_size());
 734
 735         /* OVS_FLOW_ATTR_KEY */
 736         if (!sfid || should_fill_key(sfid, ufid_flags))
 737                 len += nla_total_size(ovs_key_attr_size());
 738
 739         /* OVS_FLOW_ATTR_MASK */
 740         if (should_fill_mask(ufid_flags))
 741                 len += nla_total_size(ovs_key_attr_size());
 742
 743         /* OVS_FLOW_ATTR_ACTIONS */
 744         if (should_fill_actions(ufid_flags))
 745                 len += nla_total_size(acts->orig_len);
 746
 747         return len
 748                 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 749                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 750                 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 751 }
 752
 753 /* Called with ovs_mutex or RCU read lock. */
 754 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 755                                    struct sk_buff *skb)
 756 {
 757         struct ovs_flow_stats stats;
 758         __be16 tcp_flags;
 759         unsigned long used;
 760
 761         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 762
 763         if (used &&
 764             nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 765                               OVS_FLOW_ATTR_PAD))
 766                 return -EMSGSIZE;
 767
 768         if (stats.n_packets &&
 769             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 770                           sizeof(struct ovs_flow_stats), &stats,
 771                           OVS_FLOW_ATTR_PAD))
 772                 return -EMSGSIZE;
 773
 774         if ((u8)ntohs(tcp_flags) &&
 775              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 776                 return -EMSGSIZE;
 777
 778         return 0;
 779 }
 780
 781 /* Called with ovs_mutex or RCU read lock. */
 782 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 783                                      struct sk_buff *skb, int skb_orig_len)
 784 {
 785         struct nlattr *start;
 786         int err;
 787
 788         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 789          * this is the first flow to be dumped into 'skb'.  This is unusual for
 790          * Netlink but individual action lists can be longer than
 791          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 792          * The userspace caller can always fetch the actions separately if it
 793          * really wants them.  (Most userspace callers in fact don't care.)
 794          *
 795          * This can only fail for dump operations because the skb is always
 796          * properly sized for single flows.
 797          */
 798         start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
 799         if (start) {
 800                 const struct sw_flow_actions *sf_acts;
 801
 802                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 803                 err = ovs_nla_put_actions(sf_acts->actions,
 804                                           sf_acts->actions_len, skb);
 805
 806                 if (!err)
 807                         nla_nest_end(skb, start);
 808                 else {
 809                         if (skb_orig_len)
 810                                 return err;
 811
 812                         nla_nest_cancel(skb, start);
 813                 }
 814         } else if (skb_orig_len) {
 815                 return -EMSGSIZE;
 816         }
 817
 818         return 0;
 819 }
 820
 821 /* Called with ovs_mutex or RCU read lock. */
 822 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 823                                   struct sk_buff *skb, u32 portid,
 824                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 825 {
 826         const int skb_orig_len = skb->len;
 827         struct ovs_header *ovs_header;
 828         int err;
 829
 830         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 831                                  flags, cmd);
 832         if (!ovs_header)
 833                 return -EMSGSIZE;
 834
 835         ovs_header->dp_ifindex = dp_ifindex;
 836
 837         err = ovs_nla_put_identifier(flow, skb);
 838         if (err)
 839                 goto error;
 840
 841         if (should_fill_key(&flow->id, ufid_flags)) {
 842                 err = ovs_nla_put_masked_key(flow, skb);
 843                 if (err)
 844                         goto error;
 845         }
 846
 847         if (should_fill_mask(ufid_flags)) {
 848                 err = ovs_nla_put_mask(flow, skb);
 849                 if (err)
 850                         goto error;
 851         }
 852
 853         err = ovs_flow_cmd_fill_stats(flow, skb);
 854         if (err)
 855                 goto error;
 856
 857         if (should_fill_actions(ufid_flags)) {
 858                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 859                 if (err)
 860                         goto error;
 861         }
 862
 863         genlmsg_end(skb, ovs_header);
 864         return 0;
 865
 866 error:
 867         genlmsg_cancel(skb, ovs_header);
 868         return err;
 869 }
 870
 871 /* May not be called with RCU read lock. */
 872 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 873                                                const struct sw_flow_id *sfid,
 874                                                struct genl_info *info,
 875                                                bool always,
 876                                                uint32_t ufid_flags)
 877 {
 878         struct sk_buff *skb;
 879         size_t len;
 880
 881         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 882                 return NULL;
 883
 884         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 885         skb = genlmsg_new(len, GFP_KERNEL);
 886         if (!skb)
 887                 return ERR_PTR(-ENOMEM);
 888
 889         return skb;
 890 }
 891
 892 /* Called with ovs_mutex. */
 893 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 894                                                int dp_ifindex,
 895                                                struct genl_info *info, u8 cmd,
 896                                                bool always, u32 ufid_flags)
 897 {
 898         struct sk_buff *skb;
 899         int retval;
 900
 901         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 902                                       &flow->id, info, always, ufid_flags);
 903         if (IS_ERR_OR_NULL(skb))
 904                 return skb;
 905
 906         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 907                                         info->snd_portid, info->snd_seq, 0,
 908                                         cmd, ufid_flags);
 909         if (WARN_ON_ONCE(retval < 0)) {
 910                 kfree_skb(skb);
 911                 skb = ERR_PTR(retval);
 912         }
 913         return skb;
 914 }
 915
 916 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 917 {
 918         struct net *net = sock_net(skb->sk);
 919         struct nlattr **a = info->attrs;
 920         struct ovs_header *ovs_header = info->userhdr;
 921         struct sw_flow *flow = NULL, *new_flow;
 922         struct sw_flow_mask mask;
 923         struct sk_buff *reply;
 924         struct datapath *dp;
 925         struct sw_flow_actions *acts;
 926         struct sw_flow_match match;
 927         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 928         int error;
 929         bool log = !a[OVS_FLOW_ATTR_PROBE];
 930
 931         /* Must have key and actions. */
 932         error = -EINVAL;
 933         if (!a[OVS_FLOW_ATTR_KEY]) {
 934                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 935                 goto error;
 936         }
 937         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 938                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 939                 goto error;
 940         }
 941
 942         /* Most of the time we need to allocate a new flow, do it before
 943          * locking.
 944          */
 945         new_flow = ovs_flow_alloc();
 946         if (IS_ERR(new_flow)) {
 947                 error = PTR_ERR(new_flow);
 948                 goto error;
 949         }
 950
 951         /* Extract key. */
 952         ovs_match_init(&match, &new_flow->key, false, &mask);
 953         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 954                                   a[OVS_FLOW_ATTR_MASK], log);
 955         if (error)
 956                 goto err_kfree_flow;
 957
 958         /* Extract flow identifier. */
 959         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 960                                        &new_flow->key, log);
 961         if (error)
 962                 goto err_kfree_flow;
 963
 964         /* unmasked key is needed to match when ufid is not used. */
 965         if (ovs_identifier_is_key(&new_flow->id))
 966                 match.key = new_flow->id.unmasked_key;
 967
 968         ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
 969
 970         /* Validate actions. */
 971         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 972                                      &new_flow->key, &acts, log);
 973         if (error) {
 974                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 975                 goto err_kfree_flow;
 976         }
 977
 978         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 979                                         ufid_flags);
 980         if (IS_ERR(reply)) {
 981                 error = PTR_ERR(reply);
 982                 goto err_kfree_acts;
 983         }
 984
 985         ovs_lock();
 986         dp = get_dp(net, ovs_header->dp_ifindex);
 987         if (unlikely(!dp)) {
 988                 error = -ENODEV;
 989                 goto err_unlock_ovs;
 990         }
 991
 992         /* Check if this is a duplicate flow */
 993         if (ovs_identifier_is_ufid(&new_flow->id))
 994                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
 995         if (!flow)
 996                 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
 997         if (likely(!flow)) {
 998                 rcu_assign_pointer(new_flow->sf_acts, acts);
 999
1000                 /* Put flow in bucket. */
1001                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1002                 if (unlikely(error)) {
1003                         acts = NULL;
1004                         goto err_unlock_ovs;
1005                 }
1006
1007                 if (unlikely(reply)) {
1008                         error = ovs_flow_cmd_fill_info(new_flow,
1009                                                        ovs_header->dp_ifindex,
1010                                                        reply, info->snd_portid,
1011                                                        info->snd_seq, 0,
1012                                                        OVS_FLOW_CMD_NEW,
1013                                                        ufid_flags);
1014                         BUG_ON(error < 0);
1015                 }
1016                 ovs_unlock();
1017         } else {
1018                 struct sw_flow_actions *old_acts;
1019
1020                 /* Bail out if we're not allowed to modify an existing flow.
1021                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1022                  * because Generic Netlink treats the latter as a dump
1023                  * request.  We also accept NLM_F_EXCL in case that bug ever
1024                  * gets fixed.
1025                  */
1026                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1027                                                          | NLM_F_EXCL))) {
1028                         error = -EEXIST;
1029                         goto err_unlock_ovs;
1030                 }
1031                 /* The flow identifier has to be the same for flow updates.
1032                  * Look for any overlapping flow.
1033                  */
1034                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1035                         if (ovs_identifier_is_key(&flow->id))
1036                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1037                                                                  &match);
1038                         else /* UFID matches but key is different */
1039                                 flow = NULL;
1040                         if (!flow) {
1041                                 error = -ENOENT;
1042                                 goto err_unlock_ovs;
1043                         }
1044                 }
1045                 /* Update actions. */
1046                 old_acts = ovsl_dereference(flow->sf_acts);
1047                 rcu_assign_pointer(flow->sf_acts, acts);
1048
1049                 if (unlikely(reply)) {
1050                         error = ovs_flow_cmd_fill_info(flow,
1051                                                        ovs_header->dp_ifindex,
1052                                                        reply, info->snd_portid,
1053                                                        info->snd_seq, 0,
1054                                                        OVS_FLOW_CMD_NEW,
1055                                                        ufid_flags);
1056                         BUG_ON(error < 0);
1057                 }
1058                 ovs_unlock();
1059
1060                 ovs_nla_free_flow_actions_rcu(old_acts);
1061                 ovs_flow_free(new_flow, false);
1062         }
1063
1064         if (reply)
1065                 ovs_notify(&dp_flow_genl_family, reply, info);
1066         return 0;
1067
1068 err_unlock_ovs:
1069         ovs_unlock();
1070         kfree_skb(reply);
1071 err_kfree_acts:
1072         ovs_nla_free_flow_actions(acts);
1073 err_kfree_flow:
1074         ovs_flow_free(new_flow, false);
1075 error:
1076         return error;
1077 }
1078
1079 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1080 static noinline_for_stack struct sw_flow_actions *get_flow_actions(struct net *net,
1081                                                 const struct nlattr *a,
1082                                                 const struct sw_flow_key *key,
1083                                                 const struct sw_flow_mask *mask,
1084                                                 bool log)
1085 {
1086         struct sw_flow_actions *acts;
1087         struct sw_flow_key masked_key;
1088         int error;
1089
1090         ovs_flow_mask_key(&masked_key, key, true, mask);
1091         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1092         if (error) {
1093                 OVS_NLERR(log,
1094                           "Actions may not be safe on all matching packets");
1095                 return ERR_PTR(error);
1096         }
1097
1098         return acts;
1099 }
1100
1101 /* Factor out match-init and action-copy to avoid
1102  * "Wframe-larger-than=1024" warning. Because mask is only
1103  * used to get actions, we new a function to save some
1104  * stack space.
1105  *
1106  * If there are not key and action attrs, we return 0
1107  * directly. In the case, the caller will also not use the
1108  * match as before. If there is action attr, we try to get
1109  * actions and save them to *acts. Before returning from
1110  * the function, we reset the match->mask pointer. Because
1111  * we should not to return match object with dangling reference
1112  * to mask.
1113  * */
1114 static noinline_for_stack int
1115 ovs_nla_init_match_and_action(struct net *net,
1116                               struct sw_flow_match *match,
1117                               struct sw_flow_key *key,
1118                               struct nlattr **a,
1119                               struct sw_flow_actions **acts,
1120                               bool log)
1121 {
1122         struct sw_flow_mask mask;
1123         int error = 0;
1124
1125         if (a[OVS_FLOW_ATTR_KEY]) {
1126                 ovs_match_init(match, key, true, &mask);
1127                 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1128                                           a[OVS_FLOW_ATTR_MASK], log);
1129                 if (error)
1130                         goto error;
1131         }
1132
1133         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1134                 if (!a[OVS_FLOW_ATTR_KEY]) {
1135                         OVS_NLERR(log,
1136                                   "Flow key attribute not present in set flow.");
1137                         error = -EINVAL;
1138                         goto error;
1139                 }
1140
1141                 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1142                                          &mask, log);
1143                 if (IS_ERR(*acts)) {
1144                         error = PTR_ERR(*acts);
1145                         goto error;
1146                 }
1147         }
1148
1149         /* On success, error is 0. */
1150 error:
1151         match->mask = NULL;
1152         return error;
1153 }
1154
1155 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1156 {
1157         struct net *net = sock_net(skb->sk);
1158         struct nlattr **a = info->attrs;
1159         struct ovs_header *ovs_header = info->userhdr;
1160         struct sw_flow_key key;
1161         struct sw_flow *flow;
1162         struct sk_buff *reply = NULL;
1163         struct datapath *dp;
1164         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1165         struct sw_flow_match match;
1166         struct sw_flow_id sfid;
1167         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1168         int error = 0;
1169         bool log = !a[OVS_FLOW_ATTR_PROBE];
1170         bool ufid_present;
1171
1172         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1173         if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1174                 OVS_NLERR(log,
1175                           "Flow set message rejected, Key attribute missing.");
1176                 return -EINVAL;
1177         }
1178
1179         error = ovs_nla_init_match_and_action(net, &match, &key, a,
1180                                               &acts, log);
1181         if (error)
1182                 goto error;
1183
1184         if (acts) {
1185                 /* Can allocate before locking if have acts. */
1186                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1187                                                 ufid_flags);
1188                 if (IS_ERR(reply)) {
1189                         error = PTR_ERR(reply);
1190                         goto err_kfree_acts;
1191                 }
1192         }
1193
1194         ovs_lock();
1195         dp = get_dp(net, ovs_header->dp_ifindex);
1196         if (unlikely(!dp)) {
1197                 error = -ENODEV;
1198                 goto err_unlock_ovs;
1199         }
1200         /* Check that the flow exists. */
1201         if (ufid_present)
1202                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1203         else
1204                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1205         if (unlikely(!flow)) {
1206                 error = -ENOENT;
1207                 goto err_unlock_ovs;
1208         }
1209
1210         /* Update actions, if present. */
1211         if (likely(acts)) {
1212                 old_acts = ovsl_dereference(flow->sf_acts);
1213                 rcu_assign_pointer(flow->sf_acts, acts);
1214
1215                 if (unlikely(reply)) {
1216                         error = ovs_flow_cmd_fill_info(flow,
1217                                                        ovs_header->dp_ifindex,
1218                                                        reply, info->snd_portid,
1219                                                        info->snd_seq, 0,
1220                                                        OVS_FLOW_CMD_SET,
1221                                                        ufid_flags);
1222                         BUG_ON(error < 0);
1223                 }
1224         } else {
1225                 /* Could not alloc without acts before locking. */
1226                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1227                                                 info, OVS_FLOW_CMD_SET, false,
1228                                                 ufid_flags);
1229
1230                 if (IS_ERR(reply)) {
1231                         error = PTR_ERR(reply);
1232                         goto err_unlock_ovs;
1233                 }
1234         }
1235
1236         /* Clear stats. */
1237         if (a[OVS_FLOW_ATTR_CLEAR])
1238                 ovs_flow_stats_clear(flow);
1239         ovs_unlock();
1240
1241         if (reply)
1242                 ovs_notify(&dp_flow_genl_family, reply, info);
1243         if (old_acts)
1244                 ovs_nla_free_flow_actions_rcu(old_acts);
1245
1246         return 0;
1247
1248 err_unlock_ovs:
1249         ovs_unlock();
1250         kfree_skb(reply);
1251 err_kfree_acts:
1252         ovs_nla_free_flow_actions(acts);
1253 error:
1254         return error;
1255 }
1256
1257 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1258 {
1259         struct nlattr **a = info->attrs;
1260         struct ovs_header *ovs_header = info->userhdr;
1261         struct net *net = sock_net(skb->sk);
1262         struct sw_flow_key key;
1263         struct sk_buff *reply;
1264         struct sw_flow *flow;
1265         struct datapath *dp;
1266         struct sw_flow_match match;
1267         struct sw_flow_id ufid;
1268         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1269         int err = 0;
1270         bool log = !a[OVS_FLOW_ATTR_PROBE];
1271         bool ufid_present;
1272
1273         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1274         if (a[OVS_FLOW_ATTR_KEY]) {
1275                 ovs_match_init(&match, &key, true, NULL);
1276                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1277                                         log);
1278         } else if (!ufid_present) {
1279                 OVS_NLERR(log,
1280                           "Flow get message rejected, Key attribute missing.");
1281                 err = -EINVAL;
1282         }
1283         if (err)
1284                 return err;
1285
1286         ovs_lock();
1287         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1288         if (!dp) {
1289                 err = -ENODEV;
1290                 goto unlock;
1291         }
1292
1293         if (ufid_present)
1294                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1295         else
1296                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1297         if (!flow) {
1298                 err = -ENOENT;
1299                 goto unlock;
1300         }
1301
1302         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1303                                         OVS_FLOW_CMD_GET, true, ufid_flags);
1304         if (IS_ERR(reply)) {
1305                 err = PTR_ERR(reply);
1306                 goto unlock;
1307         }
1308
1309         ovs_unlock();
1310         return genlmsg_reply(reply, info);
1311 unlock:
1312         ovs_unlock();
1313         return err;
1314 }
1315
1316 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1317 {
1318         struct nlattr **a = info->attrs;
1319         struct ovs_header *ovs_header = info->userhdr;
1320         struct net *net = sock_net(skb->sk);
1321         struct sw_flow_key key;
1322         struct sk_buff *reply;
1323         struct sw_flow *flow = NULL;
1324         struct datapath *dp;
1325         struct sw_flow_match match;
1326         struct sw_flow_id ufid;
1327         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1328         int err;
1329         bool log = !a[OVS_FLOW_ATTR_PROBE];
1330         bool ufid_present;
1331
1332         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1333         if (a[OVS_FLOW_ATTR_KEY]) {
1334                 ovs_match_init(&match, &key, true, NULL);
1335                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1336                                         NULL, log);
1337                 if (unlikely(err))
1338                         return err;
1339         }
1340
1341         ovs_lock();
1342         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1343         if (unlikely(!dp)) {
1344                 err = -ENODEV;
1345                 goto unlock;
1346         }
1347
1348         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1349                 err = ovs_flow_tbl_flush(&dp->table);
1350                 goto unlock;
1351         }
1352
1353         if (ufid_present)
1354                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1355         else
1356                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1357         if (unlikely(!flow)) {
1358                 err = -ENOENT;
1359                 goto unlock;
1360         }
1361
1362         ovs_flow_tbl_remove(&dp->table, flow);
1363         ovs_unlock();
1364
1365         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1366                                         &flow->id, info, false, ufid_flags);
1367         if (likely(reply)) {
1368                 if (!IS_ERR(reply)) {
1369                         rcu_read_lock();        /*To keep RCU checker happy. */
1370                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1371                                                      reply, info->snd_portid,
1372                                                      info->snd_seq, 0,
1373                                                      OVS_FLOW_CMD_DEL,
1374                                                      ufid_flags);
1375                         rcu_read_unlock();
1376                         if (WARN_ON_ONCE(err < 0)) {
1377                                 kfree_skb(reply);
1378                                 goto out_free;
1379                         }
1380
1381                         ovs_notify(&dp_flow_genl_family, reply, info);
1382                 } else {
1383                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1384                 }
1385         }
1386
1387 out_free:
1388         ovs_flow_free(flow, true);
1389         return 0;
1390 unlock:
1391         ovs_unlock();
1392         return err;
1393 }
1394
1395 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1396 {
1397         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1398         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1399         struct table_instance *ti;
1400         struct datapath *dp;
1401         u32 ufid_flags;
1402         int err;
1403
1404         err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1405                                        OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1406         if (err)
1407                 return err;
1408         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1409
1410         rcu_read_lock();
1411         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1412         if (!dp) {
1413                 rcu_read_unlock();
1414                 return -ENODEV;
1415         }
1416
1417         ti = rcu_dereference(dp->table.ti);
1418         for (;;) {
1419                 struct sw_flow *flow;
1420                 u32 bucket, obj;
1421
1422                 bucket = cb->args[0];
1423                 obj = cb->args[1];
1424                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1425                 if (!flow)
1426                         break;
1427
1428                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1429                                            NETLINK_CB(cb->skb).portid,
1430                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1431                                            OVS_FLOW_CMD_GET, ufid_flags) < 0)
1432                         break;
1433
1434                 cb->args[0] = bucket;
1435                 cb->args[1] = obj;
1436         }
1437         rcu_read_unlock();
1438         return skb->len;
1439 }
1440
1441 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1442         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1443         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1444         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1445         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1446         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1447         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1448         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1449 };
1450
1451 static const struct genl_ops dp_flow_genl_ops[] = {
1452         { .cmd = OVS_FLOW_CMD_NEW,
1453           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1454           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1455           .doit = ovs_flow_cmd_new
1456         },
1457         { .cmd = OVS_FLOW_CMD_DEL,
1458           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1459           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1460           .doit = ovs_flow_cmd_del
1461         },
1462         { .cmd = OVS_FLOW_CMD_GET,
1463           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1464           .flags = 0,               /* OK for unprivileged users. */
1465           .doit = ovs_flow_cmd_get,
1466           .dumpit = ovs_flow_cmd_dump
1467         },
1468         { .cmd = OVS_FLOW_CMD_SET,
1469           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1470           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1471           .doit = ovs_flow_cmd_set,
1472         },
1473 };
1474
1475 static struct genl_family dp_flow_genl_family __ro_after_init = {
1476         .hdrsize = sizeof(struct ovs_header),
1477         .name = OVS_FLOW_FAMILY,
1478         .version = OVS_FLOW_VERSION,
1479         .maxattr = OVS_FLOW_ATTR_MAX,
1480         .policy = flow_policy,
1481         .netnsok = true,
1482         .parallel_ops = true,
1483         .ops = dp_flow_genl_ops,
1484         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1485         .mcgrps = &ovs_dp_flow_multicast_group,
1486         .n_mcgrps = 1,
1487         .module = THIS_MODULE,
1488 };
1489
1490 static size_t ovs_dp_cmd_msg_size(void)
1491 {
1492         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1493
1494         msgsize += nla_total_size(IFNAMSIZ);
1495         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1496         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1497         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1498
1499         return msgsize;
1500 }
1501
1502 /* Called with ovs_mutex. */
1503 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1504                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1505 {
1506         struct ovs_header *ovs_header;
1507         struct ovs_dp_stats dp_stats;
1508         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1509         int err;
1510
1511         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1512                                    flags, cmd);
1513         if (!ovs_header)
1514                 goto error;
1515
1516         ovs_header->dp_ifindex = get_dpifindex(dp);
1517
1518         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1519         if (err)
1520                 goto nla_put_failure;
1521
1522         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1523         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1524                           &dp_stats, OVS_DP_ATTR_PAD))
1525                 goto nla_put_failure;
1526
1527         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1528                           sizeof(struct ovs_dp_megaflow_stats),
1529                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1530                 goto nla_put_failure;
1531
1532         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1533                 goto nla_put_failure;
1534
1535         genlmsg_end(skb, ovs_header);
1536         return 0;
1537
1538 nla_put_failure:
1539         genlmsg_cancel(skb, ovs_header);
1540 error:
1541         return -EMSGSIZE;
1542 }
1543
1544 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1545 {
1546         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1547 }
1548
1549 /* Called with rcu_read_lock or ovs_mutex. */
1550 static struct datapath *lookup_datapath(struct net *net,
1551                                         const struct ovs_header *ovs_header,
1552                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1553 {
1554         struct datapath *dp;
1555
1556         if (!a[OVS_DP_ATTR_NAME])
1557                 dp = get_dp(net, ovs_header->dp_ifindex);
1558         else {
1559                 struct vport *vport;
1560
1561                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1562                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1563         }
1564         return dp ? dp : ERR_PTR(-ENODEV);
1565 }
1566
1567 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1568 {
1569         struct datapath *dp;
1570
1571         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1572         if (IS_ERR(dp))
1573                 return;
1574
1575         WARN(dp->user_features, "Dropping previously announced user features\n");
1576         dp->user_features = 0;
1577 }
1578
1579 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1580
1581 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1582 {
1583         u32 user_features = 0;
1584
1585         if (a[OVS_DP_ATTR_USER_FEATURES]) {
1586                 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1587
1588                 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1589                                       OVS_DP_F_UNALIGNED |
1590                                       OVS_DP_F_TC_RECIRC_SHARING))
1591                         return -EOPNOTSUPP;
1592
1593 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1594                 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1595                         return -EOPNOTSUPP;
1596 #endif
1597         }
1598
1599         dp->user_features = user_features;
1600
1601         if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1602                 static_branch_enable(&tc_recirc_sharing_support);
1603         else
1604                 static_branch_disable(&tc_recirc_sharing_support);
1605
1606         return 0;
1607 }
1608
1609 static int ovs_dp_stats_init(struct datapath *dp)
1610 {
1611         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1612         if (!dp->stats_percpu)
1613                 return -ENOMEM;
1614
1615         return 0;
1616 }
1617
1618 static int ovs_dp_vport_init(struct datapath *dp)
1619 {
1620         int i;
1621
1622         dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1623                                   sizeof(struct hlist_head),
1624                                   GFP_KERNEL);
1625         if (!dp->ports)
1626                 return -ENOMEM;
1627
1628         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1629                 INIT_HLIST_HEAD(&dp->ports[i]);
1630
1631         return 0;
1632 }
1633
1634 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1635 {
1636         struct nlattr **a = info->attrs;
1637         struct vport_parms parms;
1638         struct sk_buff *reply;
1639         struct datapath *dp;
1640         struct vport *vport;
1641         struct ovs_net *ovs_net;
1642         int err;
1643
1644         err = -EINVAL;
1645         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1646                 goto err;
1647
1648         reply = ovs_dp_cmd_alloc_info();
1649         if (!reply)
1650                 return -ENOMEM;
1651
1652         err = -ENOMEM;
1653         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1654         if (dp == NULL)
1655                 goto err_destroy_reply;
1656
1657         ovs_dp_set_net(dp, sock_net(skb->sk));
1658         INIT_DELAYED_WORK(&dp->masks_rebalance, ovs_dp_masks_rebalance);
1659
1660         /* Allocate table. */
1661         err = ovs_flow_tbl_init(&dp->table);
1662         if (err)
1663                 goto err_destroy_dp;
1664
1665         err = ovs_dp_stats_init(dp);
1666         if (err)
1667                 goto err_destroy_table;
1668
1669         err = ovs_dp_vport_init(dp);
1670         if (err)
1671                 goto err_destroy_stats;
1672
1673         err = ovs_meters_init(dp);
1674         if (err)
1675                 goto err_destroy_ports;
1676
1677         /* Set up our datapath device. */
1678         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1679         parms.type = OVS_VPORT_TYPE_INTERNAL;
1680         parms.options = NULL;
1681         parms.dp = dp;
1682         parms.port_no = OVSP_LOCAL;
1683         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1684
1685         err = ovs_dp_change(dp, a);
1686         if (err)
1687                 goto err_destroy_meters;
1688
1689         /* So far only local changes have been made, now need the lock. */
1690         ovs_lock();
1691
1692         vport = new_vport(&parms);
1693         if (IS_ERR(vport)) {
1694                 err = PTR_ERR(vport);
1695                 if (err == -EBUSY)
1696                         err = -EEXIST;
1697
1698                 if (err == -EEXIST) {
1699                         /* An outdated user space instance that does not understand
1700                          * the concept of user_features has attempted to create a new
1701                          * datapath and is likely to reuse it. Drop all user features.
1702                          */
1703                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1704                                 ovs_dp_reset_user_features(skb, info);
1705                 }
1706
1707                 ovs_unlock();
1708                 goto err_destroy_meters;
1709         }
1710
1711         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1712                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1713         BUG_ON(err < 0);
1714
1715         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1716         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1717
1718         schedule_delayed_work(&dp->masks_rebalance,
1719                               msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
1720
1721         ovs_unlock();
1722
1723         ovs_notify(&dp_datapath_genl_family, reply, info);
1724         return 0;
1725
1726 err_destroy_meters:
1727         ovs_meters_exit(dp);
1728 err_destroy_ports:
1729         kfree(dp->ports);
1730 err_destroy_stats:
1731         free_percpu(dp->stats_percpu);
1732 err_destroy_table:
1733         ovs_flow_tbl_destroy(&dp->table);
1734 err_destroy_dp:
1735         kfree(dp);
1736 err_destroy_reply:
1737         kfree_skb(reply);
1738 err:
1739         return err;
1740 }
1741
1742 /* Called with ovs_mutex. */
1743 static void __dp_destroy(struct datapath *dp)
1744 {
1745         int i;
1746
1747         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1748                 struct vport *vport;
1749                 struct hlist_node *n;
1750
1751                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1752                         if (vport->port_no != OVSP_LOCAL)
1753                                 ovs_dp_detach_port(vport);
1754         }
1755
1756         list_del_rcu(&dp->list_node);
1757
1758         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1759          * all ports in datapath are destroyed first before freeing datapath.
1760          */
1761         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1762
1763         /* RCU destroy the flow table */
1764         call_rcu(&dp->rcu, destroy_dp_rcu);
1765
1766         /* Cancel remaining work. */
1767         cancel_delayed_work_sync(&dp->masks_rebalance);
1768 }
1769
1770 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1771 {
1772         struct sk_buff *reply;
1773         struct datapath *dp;
1774         int err;
1775
1776         reply = ovs_dp_cmd_alloc_info();
1777         if (!reply)
1778                 return -ENOMEM;
1779
1780         ovs_lock();
1781         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1782         err = PTR_ERR(dp);
1783         if (IS_ERR(dp))
1784                 goto err_unlock_free;
1785
1786         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1787                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1788         BUG_ON(err < 0);
1789
1790         __dp_destroy(dp);
1791         ovs_unlock();
1792
1793         ovs_notify(&dp_datapath_genl_family, reply, info);
1794
1795         return 0;
1796
1797 err_unlock_free:
1798         ovs_unlock();
1799         kfree_skb(reply);
1800         return err;
1801 }
1802
1803 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1804 {
1805         struct sk_buff *reply;
1806         struct datapath *dp;
1807         int err;
1808
1809         reply = ovs_dp_cmd_alloc_info();
1810         if (!reply)
1811                 return -ENOMEM;
1812
1813         ovs_lock();
1814         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1815         err = PTR_ERR(dp);
1816         if (IS_ERR(dp))
1817                 goto err_unlock_free;
1818
1819         err = ovs_dp_change(dp, info->attrs);
1820         if (err)
1821                 goto err_unlock_free;
1822
1823         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1824                                    info->snd_seq, 0, OVS_DP_CMD_SET);
1825         BUG_ON(err < 0);
1826
1827         ovs_unlock();
1828         ovs_notify(&dp_datapath_genl_family, reply, info);
1829
1830         return 0;
1831
1832 err_unlock_free:
1833         ovs_unlock();
1834         kfree_skb(reply);
1835         return err;
1836 }
1837
1838 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1839 {
1840         struct sk_buff *reply;
1841         struct datapath *dp;
1842         int err;
1843
1844         reply = ovs_dp_cmd_alloc_info();
1845         if (!reply)
1846                 return -ENOMEM;
1847
1848         ovs_lock();
1849         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1850         if (IS_ERR(dp)) {
1851                 err = PTR_ERR(dp);
1852                 goto err_unlock_free;
1853         }
1854         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1855                                    info->snd_seq, 0, OVS_DP_CMD_GET);
1856         BUG_ON(err < 0);
1857         ovs_unlock();
1858
1859         return genlmsg_reply(reply, info);
1860
1861 err_unlock_free:
1862         ovs_unlock();
1863         kfree_skb(reply);
1864         return err;
1865 }
1866
1867 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1868 {
1869         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1870         struct datapath *dp;
1871         int skip = cb->args[0];
1872         int i = 0;
1873
1874         ovs_lock();
1875         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1876                 if (i >= skip &&
1877                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1878                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1879                                          OVS_DP_CMD_GET) < 0)
1880                         break;
1881                 i++;
1882         }
1883         ovs_unlock();
1884
1885         cb->args[0] = i;
1886
1887         return skb->len;
1888 }
1889
1890 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1891         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1892         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1893         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1894 };
1895
1896 static const struct genl_ops dp_datapath_genl_ops[] = {
1897         { .cmd = OVS_DP_CMD_NEW,
1898           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1899           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1900           .doit = ovs_dp_cmd_new
1901         },
1902         { .cmd = OVS_DP_CMD_DEL,
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_del
1906         },
1907         { .cmd = OVS_DP_CMD_GET,
1908           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1909           .flags = 0,               /* OK for unprivileged users. */
1910           .doit = ovs_dp_cmd_get,
1911           .dumpit = ovs_dp_cmd_dump
1912         },
1913         { .cmd = OVS_DP_CMD_SET,
1914           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1915           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1916           .doit = ovs_dp_cmd_set,
1917         },
1918 };
1919
1920 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1921         .hdrsize = sizeof(struct ovs_header),
1922         .name = OVS_DATAPATH_FAMILY,
1923         .version = OVS_DATAPATH_VERSION,
1924         .maxattr = OVS_DP_ATTR_MAX,
1925         .policy = datapath_policy,
1926         .netnsok = true,
1927         .parallel_ops = true,
1928         .ops = dp_datapath_genl_ops,
1929         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1930         .mcgrps = &ovs_dp_datapath_multicast_group,
1931         .n_mcgrps = 1,
1932         .module = THIS_MODULE,
1933 };
1934
1935 /* Called with ovs_mutex or RCU read lock. */
1936 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1937                                    struct net *net, u32 portid, u32 seq,
1938                                    u32 flags, u8 cmd, gfp_t gfp)
1939 {
1940         struct ovs_header *ovs_header;
1941         struct ovs_vport_stats vport_stats;
1942         int err;
1943
1944         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1945                                  flags, cmd);
1946         if (!ovs_header)
1947                 return -EMSGSIZE;
1948
1949         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1950
1951         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1952             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1953             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1954                            ovs_vport_name(vport)) ||
1955             nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1956                 goto nla_put_failure;
1957
1958         if (!net_eq(net, dev_net(vport->dev))) {
1959                 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
1960
1961                 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1962                         goto nla_put_failure;
1963         }
1964
1965         ovs_vport_get_stats(vport, &vport_stats);
1966         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1967                           sizeof(struct ovs_vport_stats), &vport_stats,
1968                           OVS_VPORT_ATTR_PAD))
1969                 goto nla_put_failure;
1970
1971         if (ovs_vport_get_upcall_portids(vport, skb))
1972                 goto nla_put_failure;
1973
1974         err = ovs_vport_get_options(vport, skb);
1975         if (err == -EMSGSIZE)
1976                 goto error;
1977
1978         genlmsg_end(skb, ovs_header);
1979         return 0;
1980
1981 nla_put_failure:
1982         err = -EMSGSIZE;
1983 error:
1984         genlmsg_cancel(skb, ovs_header);
1985         return err;
1986 }
1987
1988 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1989 {
1990         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1991 }
1992
1993 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1994 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1995                                          u32 portid, u32 seq, u8 cmd)
1996 {
1997         struct sk_buff *skb;
1998         int retval;
1999
2000         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2001         if (!skb)
2002                 return ERR_PTR(-ENOMEM);
2003
2004         retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2005                                          GFP_KERNEL);
2006         BUG_ON(retval < 0);
2007
2008         return skb;
2009 }
2010
2011 /* Called with ovs_mutex or RCU read lock. */
2012 static struct vport *lookup_vport(struct net *net,
2013                                   const struct ovs_header *ovs_header,
2014                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2015 {
2016         struct datapath *dp;
2017         struct vport *vport;
2018
2019         if (a[OVS_VPORT_ATTR_IFINDEX])
2020                 return ERR_PTR(-EOPNOTSUPP);
2021         if (a[OVS_VPORT_ATTR_NAME]) {
2022                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2023                 if (!vport)
2024                         return ERR_PTR(-ENODEV);
2025                 if (ovs_header->dp_ifindex &&
2026                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2027                         return ERR_PTR(-ENODEV);
2028                 return vport;
2029         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2030                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2031
2032                 if (port_no >= DP_MAX_PORTS)
2033                         return ERR_PTR(-EFBIG);
2034
2035                 dp = get_dp(net, ovs_header->dp_ifindex);
2036                 if (!dp)
2037                         return ERR_PTR(-ENODEV);
2038
2039                 vport = ovs_vport_ovsl_rcu(dp, port_no);
2040                 if (!vport)
2041                         return ERR_PTR(-ENODEV);
2042                 return vport;
2043         } else
2044                 return ERR_PTR(-EINVAL);
2045
2046 }
2047
2048 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2049 {
2050         unsigned int dev_headroom, max_headroom = 0;
2051         struct net_device *dev;
2052         struct vport *vport;
2053         int i;
2054
2055         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2056                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2057                                         lockdep_ovsl_is_held()) {
2058                         dev = vport->dev;
2059                         dev_headroom = netdev_get_fwd_headroom(dev);
2060                         if (dev_headroom > max_headroom)
2061                                 max_headroom = dev_headroom;
2062                 }
2063         }
2064
2065         return max_headroom;
2066 }
2067
2068 /* Called with ovs_mutex */
2069 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2070 {
2071         struct vport *vport;
2072         int i;
2073
2074         dp->max_headroom = new_headroom;
2075         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
2076                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2077                                         lockdep_ovsl_is_held())
2078                         netdev_set_rx_headroom(vport->dev, new_headroom);
2079 }
2080
2081 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2082 {
2083         struct nlattr **a = info->attrs;
2084         struct ovs_header *ovs_header = info->userhdr;
2085         struct vport_parms parms;
2086         struct sk_buff *reply;
2087         struct vport *vport;
2088         struct datapath *dp;
2089         unsigned int new_headroom;
2090         u32 port_no;
2091         int err;
2092
2093         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2094             !a[OVS_VPORT_ATTR_UPCALL_PID])
2095                 return -EINVAL;
2096         if (a[OVS_VPORT_ATTR_IFINDEX])
2097                 return -EOPNOTSUPP;
2098
2099         port_no = a[OVS_VPORT_ATTR_PORT_NO]
2100                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2101         if (port_no >= DP_MAX_PORTS)
2102                 return -EFBIG;
2103
2104         reply = ovs_vport_cmd_alloc_info();
2105         if (!reply)
2106                 return -ENOMEM;
2107
2108         ovs_lock();
2109 restart:
2110         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2111         err = -ENODEV;
2112         if (!dp)
2113                 goto exit_unlock_free;
2114
2115         if (port_no) {
2116                 vport = ovs_vport_ovsl(dp, port_no);
2117                 err = -EBUSY;
2118                 if (vport)
2119                         goto exit_unlock_free;
2120         } else {
2121                 for (port_no = 1; ; port_no++) {
2122                         if (port_no >= DP_MAX_PORTS) {
2123                                 err = -EFBIG;
2124                                 goto exit_unlock_free;
2125                         }
2126                         vport = ovs_vport_ovsl(dp, port_no);
2127                         if (!vport)
2128                                 break;
2129                 }
2130         }
2131
2132         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2133         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2134         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2135         parms.dp = dp;
2136         parms.port_no = port_no;
2137         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2138
2139         vport = new_vport(&parms);
2140         err = PTR_ERR(vport);
2141         if (IS_ERR(vport)) {
2142                 if (err == -EAGAIN)
2143                         goto restart;
2144                 goto exit_unlock_free;
2145         }
2146
2147         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2148                                       info->snd_portid, info->snd_seq, 0,
2149                                       OVS_VPORT_CMD_NEW, GFP_KERNEL);
2150
2151         new_headroom = netdev_get_fwd_headroom(vport->dev);
2152
2153         if (new_headroom > dp->max_headroom)
2154                 ovs_update_headroom(dp, new_headroom);
2155         else
2156                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2157
2158         BUG_ON(err < 0);
2159         ovs_unlock();
2160
2161         ovs_notify(&dp_vport_genl_family, reply, info);
2162         return 0;
2163
2164 exit_unlock_free:
2165         ovs_unlock();
2166         kfree_skb(reply);
2167         return err;
2168 }
2169
2170 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2171 {
2172         struct nlattr **a = info->attrs;
2173         struct sk_buff *reply;
2174         struct vport *vport;
2175         int err;
2176
2177         reply = ovs_vport_cmd_alloc_info();
2178         if (!reply)
2179                 return -ENOMEM;
2180
2181         ovs_lock();
2182         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2183         err = PTR_ERR(vport);
2184         if (IS_ERR(vport))
2185                 goto exit_unlock_free;
2186
2187         if (a[OVS_VPORT_ATTR_TYPE] &&
2188             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2189                 err = -EINVAL;
2190                 goto exit_unlock_free;
2191         }
2192
2193         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2194                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2195                 if (err)
2196                         goto exit_unlock_free;
2197         }
2198
2199
2200         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2201                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2202
2203                 err = ovs_vport_set_upcall_portids(vport, ids);
2204                 if (err)
2205                         goto exit_unlock_free;
2206         }
2207
2208         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2209                                       info->snd_portid, info->snd_seq, 0,
2210                                       OVS_VPORT_CMD_SET, GFP_KERNEL);
2211         BUG_ON(err < 0);
2212
2213         ovs_unlock();
2214         ovs_notify(&dp_vport_genl_family, reply, info);
2215         return 0;
2216
2217 exit_unlock_free:
2218         ovs_unlock();
2219         kfree_skb(reply);
2220         return err;
2221 }
2222
2223 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2224 {
2225         bool update_headroom = false;
2226         struct nlattr **a = info->attrs;
2227         struct sk_buff *reply;
2228         struct datapath *dp;
2229         struct vport *vport;
2230         unsigned int new_headroom;
2231         int err;
2232
2233         reply = ovs_vport_cmd_alloc_info();
2234         if (!reply)
2235                 return -ENOMEM;
2236
2237         ovs_lock();
2238         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2239         err = PTR_ERR(vport);
2240         if (IS_ERR(vport))
2241                 goto exit_unlock_free;
2242
2243         if (vport->port_no == OVSP_LOCAL) {
2244                 err = -EINVAL;
2245                 goto exit_unlock_free;
2246         }
2247
2248         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2249                                       info->snd_portid, info->snd_seq, 0,
2250                                       OVS_VPORT_CMD_DEL, GFP_KERNEL);
2251         BUG_ON(err < 0);
2252
2253         /* the vport deletion may trigger dp headroom update */
2254         dp = vport->dp;
2255         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2256                 update_headroom = true;
2257
2258         netdev_reset_rx_headroom(vport->dev);
2259         ovs_dp_detach_port(vport);
2260
2261         if (update_headroom) {
2262                 new_headroom = ovs_get_max_headroom(dp);
2263
2264                 if (new_headroom < dp->max_headroom)
2265                         ovs_update_headroom(dp, new_headroom);
2266         }
2267         ovs_unlock();
2268
2269         ovs_notify(&dp_vport_genl_family, reply, info);
2270         return 0;
2271
2272 exit_unlock_free:
2273         ovs_unlock();
2274         kfree_skb(reply);
2275         return err;
2276 }
2277
2278 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2279 {
2280         struct nlattr **a = info->attrs;
2281         struct ovs_header *ovs_header = info->userhdr;
2282         struct sk_buff *reply;
2283         struct vport *vport;
2284         int err;
2285
2286         reply = ovs_vport_cmd_alloc_info();
2287         if (!reply)
2288                 return -ENOMEM;
2289
2290         rcu_read_lock();
2291         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2292         err = PTR_ERR(vport);
2293         if (IS_ERR(vport))
2294                 goto exit_unlock_free;
2295         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2296                                       info->snd_portid, info->snd_seq, 0,
2297                                       OVS_VPORT_CMD_GET, GFP_ATOMIC);
2298         BUG_ON(err < 0);
2299         rcu_read_unlock();
2300
2301         return genlmsg_reply(reply, info);
2302
2303 exit_unlock_free:
2304         rcu_read_unlock();
2305         kfree_skb(reply);
2306         return err;
2307 }
2308
2309 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2310 {
2311         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2312         struct datapath *dp;
2313         int bucket = cb->args[0], skip = cb->args[1];
2314         int i, j = 0;
2315
2316         rcu_read_lock();
2317         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2318         if (!dp) {
2319                 rcu_read_unlock();
2320                 return -ENODEV;
2321         }
2322         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2323                 struct vport *vport;
2324
2325                 j = 0;
2326                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2327                         if (j >= skip &&
2328                             ovs_vport_cmd_fill_info(vport, skb,
2329                                                     sock_net(skb->sk),
2330                                                     NETLINK_CB(cb->skb).portid,
2331                                                     cb->nlh->nlmsg_seq,
2332                                                     NLM_F_MULTI,
2333                                                     OVS_VPORT_CMD_GET,
2334                                                     GFP_ATOMIC) < 0)
2335                                 goto out;
2336
2337                         j++;
2338                 }
2339                 skip = 0;
2340         }
2341 out:
2342         rcu_read_unlock();
2343
2344         cb->args[0] = i;
2345         cb->args[1] = j;
2346
2347         return skb->len;
2348 }
2349
2350 static void ovs_dp_masks_rebalance(struct work_struct *work)
2351 {
2352         struct datapath *dp = container_of(work, struct datapath,
2353                                            masks_rebalance.work);
2354
2355         ovs_lock();
2356         ovs_flow_masks_rebalance(&dp->table);
2357         ovs_unlock();
2358
2359         schedule_delayed_work(&dp->masks_rebalance,
2360                               msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2361 }
2362
2363 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2364         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2365         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2366         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2367         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2368         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2369         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2370         [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2371         [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2372 };
2373
2374 static const struct genl_ops dp_vport_genl_ops[] = {
2375         { .cmd = OVS_VPORT_CMD_NEW,
2376           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2377           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2378           .doit = ovs_vport_cmd_new
2379         },
2380         { .cmd = OVS_VPORT_CMD_DEL,
2381           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2382           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2383           .doit = ovs_vport_cmd_del
2384         },
2385         { .cmd = OVS_VPORT_CMD_GET,
2386           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2387           .flags = 0,               /* OK for unprivileged users. */
2388           .doit = ovs_vport_cmd_get,
2389           .dumpit = ovs_vport_cmd_dump
2390         },
2391         { .cmd = OVS_VPORT_CMD_SET,
2392           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2393           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2394           .doit = ovs_vport_cmd_set,
2395         },
2396 };
2397
2398 struct genl_family dp_vport_genl_family __ro_after_init = {
2399         .hdrsize = sizeof(struct ovs_header),
2400         .name = OVS_VPORT_FAMILY,
2401         .version = OVS_VPORT_VERSION,
2402         .maxattr = OVS_VPORT_ATTR_MAX,
2403         .policy = vport_policy,
2404         .netnsok = true,
2405         .parallel_ops = true,
2406         .ops = dp_vport_genl_ops,
2407         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2408         .mcgrps = &ovs_dp_vport_multicast_group,
2409         .n_mcgrps = 1,
2410         .module = THIS_MODULE,
2411 };
2412
2413 static struct genl_family * const dp_genl_families[] = {
2414         &dp_datapath_genl_family,
2415         &dp_vport_genl_family,
2416         &dp_flow_genl_family,
2417         &dp_packet_genl_family,
2418         &dp_meter_genl_family,
2419 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2420         &dp_ct_limit_genl_family,
2421 #endif
2422 };
2423
2424 static void dp_unregister_genl(int n_families)
2425 {
2426         int i;
2427
2428         for (i = 0; i < n_families; i++)
2429                 genl_unregister_family(dp_genl_families[i]);
2430 }
2431
2432 static int __init dp_register_genl(void)
2433 {
2434         int err;
2435         int i;
2436
2437         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2438
2439                 err = genl_register_family(dp_genl_families[i]);
2440                 if (err)
2441                         goto error;
2442         }
2443
2444         return 0;
2445
2446 error:
2447         dp_unregister_genl(i);
2448         return err;
2449 }
2450
2451 static int __net_init ovs_init_net(struct net *net)
2452 {
2453         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2454
2455         INIT_LIST_HEAD(&ovs_net->dps);
2456         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2457         return ovs_ct_init(net);
2458 }
2459
2460 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2461                                             struct list_head *head)
2462 {
2463         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2464         struct datapath *dp;
2465
2466         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2467                 int i;
2468
2469                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2470                         struct vport *vport;
2471
2472                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2473                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2474                                         continue;
2475
2476                                 if (dev_net(vport->dev) == dnet)
2477                                         list_add(&vport->detach_list, head);
2478                         }
2479                 }
2480         }
2481 }
2482
2483 static void __net_exit ovs_exit_net(struct net *dnet)
2484 {
2485         struct datapath *dp, *dp_next;
2486         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2487         struct vport *vport, *vport_next;
2488         struct net *net;
2489         LIST_HEAD(head);
2490
2491         ovs_lock();
2492
2493         ovs_ct_exit(dnet);
2494
2495         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2496                 __dp_destroy(dp);
2497
2498         down_read(&net_rwsem);
2499         for_each_net(net)
2500                 list_vports_from_net(net, dnet, &head);
2501         up_read(&net_rwsem);
2502
2503         /* Detach all vports from given namespace. */
2504         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2505                 list_del(&vport->detach_list);
2506                 ovs_dp_detach_port(vport);
2507         }
2508
2509         ovs_unlock();
2510
2511         cancel_work_sync(&ovs_net->dp_notify_work);
2512 }
2513
2514 static struct pernet_operations ovs_net_ops = {
2515         .init = ovs_init_net,
2516         .exit = ovs_exit_net,
2517         .id   = &ovs_net_id,
2518         .size = sizeof(struct ovs_net),
2519 };
2520
2521 static int __init dp_init(void)
2522 {
2523         int err;
2524
2525         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof_field(struct sk_buff, cb));
2526
2527         pr_info("Open vSwitch switching datapath\n");
2528
2529         err = action_fifos_init();
2530         if (err)
2531                 goto error;
2532
2533         err = ovs_internal_dev_rtnl_link_register();
2534         if (err)
2535                 goto error_action_fifos_exit;
2536
2537         err = ovs_flow_init();
2538         if (err)
2539                 goto error_unreg_rtnl_link;
2540
2541         err = ovs_vport_init();
2542         if (err)
2543                 goto error_flow_exit;
2544
2545         err = register_pernet_device(&ovs_net_ops);
2546         if (err)
2547                 goto error_vport_exit;
2548
2549         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2550         if (err)
2551                 goto error_netns_exit;
2552
2553         err = ovs_netdev_init();
2554         if (err)
2555                 goto error_unreg_notifier;
2556
2557         err = dp_register_genl();
2558         if (err < 0)
2559                 goto error_unreg_netdev;
2560
2561         return 0;
2562
2563 error_unreg_netdev:
2564         ovs_netdev_exit();
2565 error_unreg_notifier:
2566         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2567 error_netns_exit:
2568         unregister_pernet_device(&ovs_net_ops);
2569 error_vport_exit:
2570         ovs_vport_exit();
2571 error_flow_exit:
2572         ovs_flow_exit();
2573 error_unreg_rtnl_link:
2574         ovs_internal_dev_rtnl_link_unregister();
2575 error_action_fifos_exit:
2576         action_fifos_exit();
2577 error:
2578         return err;
2579 }
2580
2581 static void dp_cleanup(void)
2582 {
2583         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2584         ovs_netdev_exit();
2585         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2586         unregister_pernet_device(&ovs_net_ops);
2587         rcu_barrier();
2588         ovs_vport_exit();
2589         ovs_flow_exit();
2590         ovs_internal_dev_rtnl_link_unregister();
2591         action_fifos_exit();
2592 }
2593
2594 module_init(dp_init);
2595 module_exit(dp_cleanup);
2596
2597 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2598 MODULE_LICENSE("GPL");
2599 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2600 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2601 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2602 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2603 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2604 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);