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         u32 n_cache_hit;
 229         int error;
 230
 231         stats = this_cpu_ptr(dp->stats_percpu);
 232
 233         /* Look up flow. */
 234         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
 235                                          &n_mask_hit, &n_cache_hit);
 236         if (unlikely(!flow)) {
 237                 struct dp_upcall_info upcall;
 238
 239                 memset(&upcall, 0, sizeof(upcall));
 240                 upcall.cmd = OVS_PACKET_CMD_MISS;
 241                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 242                 upcall.mru = OVS_CB(skb)->mru;
 243                 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
 244                 switch (error) {
 245                 case 0:
 246                 case -EAGAIN:
 247                 case -ERESTARTSYS:
 248                 case -EINTR:
 249                         consume_skb(skb);
 250                         break;
 251                 default:
 252                         kfree_skb(skb);
 253                         break;
 254                 }
 255                 stats_counter = &stats->n_missed;
 256                 goto out;
 257         }
 258
 259         ovs_flow_stats_update(flow, key->tp.flags, skb);
 260         sf_acts = rcu_dereference(flow->sf_acts);
 261         error = ovs_execute_actions(dp, skb, sf_acts, key);
 262         if (unlikely(error))
 263                 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
 264                                     ovs_dp_name(dp), error);
 265
 266         stats_counter = &stats->n_hit;
 267
 268 out:
 269         /* Update datapath statistics. */
 270         u64_stats_update_begin(&stats->syncp);
 271         (*stats_counter)++;
 272         stats->n_mask_hit += n_mask_hit;
 273         stats->n_cache_hit += n_cache_hit;
 274         u64_stats_update_end(&stats->syncp);
 275 }
 276
 277 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 278                   const struct sw_flow_key *key,
 279                   const struct dp_upcall_info *upcall_info,
 280                   uint32_t cutlen)
 281 {
 282         struct dp_stats_percpu *stats;
 283         int err;
 284
 285         if (upcall_info->portid == 0) {
 286                 err = -ENOTCONN;
 287                 goto err;
 288         }
 289
 290         if (!skb_is_gso(skb))
 291                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 292         else
 293                 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
 294         if (err)
 295                 goto err;
 296
 297         return 0;
 298
 299 err:
 300         stats = this_cpu_ptr(dp->stats_percpu);
 301
 302         u64_stats_update_begin(&stats->syncp);
 303         stats->n_lost++;
 304         u64_stats_update_end(&stats->syncp);
 305
 306         return err;
 307 }
 308
 309 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 310                              const struct sw_flow_key *key,
 311                              const struct dp_upcall_info *upcall_info,
 312                              uint32_t cutlen)
 313 {
 314         unsigned int gso_type = skb_shinfo(skb)->gso_type;
 315         struct sw_flow_key later_key;
 316         struct sk_buff *segs, *nskb;
 317         int err;
 318
 319         BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_GSO_CB_OFFSET);
 320         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 321         if (IS_ERR(segs))
 322                 return PTR_ERR(segs);
 323         if (segs == NULL)
 324                 return -EINVAL;
 325
 326         if (gso_type & SKB_GSO_UDP) {
 327                 /* The initial flow key extracted by ovs_flow_key_extract()
 328                  * in this case is for a first fragment, so we need to
 329                  * properly mark later fragments.
 330                  */
 331                 later_key = *key;
 332                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 333         }
 334
 335         /* Queue all of the segments. */
 336         skb_list_walk_safe(segs, skb, nskb) {
 337                 if (gso_type & SKB_GSO_UDP && skb != segs)
 338                         key = &later_key;
 339
 340                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 341                 if (err)
 342                         break;
 343
 344         }
 345
 346         /* Free all of the segments. */
 347         skb_list_walk_safe(segs, skb, nskb) {
 348                 if (err)
 349                         kfree_skb(skb);
 350                 else
 351                         consume_skb(skb);
 352         }
 353         return err;
 354 }
 355
 356 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 357                               unsigned int hdrlen, int actions_attrlen)
 358 {
 359         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 360                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 361                 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 362                 + nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
 363                 + nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
 364
 365         /* OVS_PACKET_ATTR_USERDATA */
 366         if (upcall_info->userdata)
 367                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 368
 369         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 370         if (upcall_info->egress_tun_info)
 371                 size += nla_total_size(ovs_tun_key_attr_size());
 372
 373         /* OVS_PACKET_ATTR_ACTIONS */
 374         if (upcall_info->actions_len)
 375                 size += nla_total_size(actions_attrlen);
 376
 377         /* OVS_PACKET_ATTR_MRU */
 378         if (upcall_info->mru)
 379                 size += nla_total_size(sizeof(upcall_info->mru));
 380
 381         return size;
 382 }
 383
 384 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 385 {
 386         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 387                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 388
 389                 if (plen > 0)
 390                         skb_put_zero(skb, plen);
 391         }
 392 }
 393
 394 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 395                                   const struct sw_flow_key *key,
 396                                   const struct dp_upcall_info *upcall_info,
 397                                   uint32_t cutlen)
 398 {
 399         struct ovs_header *upcall;
 400         struct sk_buff *nskb = NULL;
 401         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 402         struct nlattr *nla;
 403         size_t len;
 404         unsigned int hlen;
 405         int err, dp_ifindex;
 406         u64 hash;
 407
 408         dp_ifindex = get_dpifindex(dp);
 409         if (!dp_ifindex)
 410                 return -ENODEV;
 411
 412         if (skb_vlan_tag_present(skb)) {
 413                 nskb = skb_clone(skb, GFP_ATOMIC);
 414                 if (!nskb)
 415                         return -ENOMEM;
 416
 417                 nskb = __vlan_hwaccel_push_inside(nskb);
 418                 if (!nskb)
 419                         return -ENOMEM;
 420
 421                 skb = nskb;
 422         }
 423
 424         if (nla_attr_size(skb->len) > USHRT_MAX) {
 425                 err = -EFBIG;
 426                 goto out;
 427         }
 428
 429         /* Complete checksum if needed */
 430         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 431             (err = skb_csum_hwoffload_help(skb, 0)))
 432                 goto out;
 433
 434         /* Older versions of OVS user space enforce alignment of the last
 435          * Netlink attribute to NLA_ALIGNTO which would require extensive
 436          * padding logic. Only perform zerocopy if padding is not required.
 437          */
 438         if (dp->user_features & OVS_DP_F_UNALIGNED)
 439                 hlen = skb_zerocopy_headlen(skb);
 440         else
 441                 hlen = skb->len;
 442
 443         len = upcall_msg_size(upcall_info, hlen - cutlen,
 444                               OVS_CB(skb)->acts_origlen);
 445         user_skb = genlmsg_new(len, GFP_ATOMIC);
 446         if (!user_skb) {
 447                 err = -ENOMEM;
 448                 goto out;
 449         }
 450
 451         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 452                              0, upcall_info->cmd);
 453         if (!upcall) {
 454                 err = -EINVAL;
 455                 goto out;
 456         }
 457         upcall->dp_ifindex = dp_ifindex;
 458
 459         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 460         if (err)
 461                 goto out;
 462
 463         if (upcall_info->userdata)
 464                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 465                           nla_len(upcall_info->userdata),
 466                           nla_data(upcall_info->userdata));
 467
 468         if (upcall_info->egress_tun_info) {
 469                 nla = nla_nest_start_noflag(user_skb,
 470                                             OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 471                 if (!nla) {
 472                         err = -EMSGSIZE;
 473                         goto out;
 474                 }
 475                 err = ovs_nla_put_tunnel_info(user_skb,
 476                                               upcall_info->egress_tun_info);
 477                 if (err)
 478                         goto out;
 479
 480                 nla_nest_end(user_skb, nla);
 481         }
 482
 483         if (upcall_info->actions_len) {
 484                 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
 485                 if (!nla) {
 486                         err = -EMSGSIZE;
 487                         goto out;
 488                 }
 489                 err = ovs_nla_put_actions(upcall_info->actions,
 490                                           upcall_info->actions_len,
 491                                           user_skb);
 492                 if (!err)
 493                         nla_nest_end(user_skb, nla);
 494                 else
 495                         nla_nest_cancel(user_skb, nla);
 496         }
 497
 498         /* Add OVS_PACKET_ATTR_MRU */
 499         if (upcall_info->mru &&
 500             nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
 501                 err = -ENOBUFS;
 502                 goto out;
 503         }
 504
 505         /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 506         if (cutlen > 0 &&
 507             nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
 508                 err = -ENOBUFS;
 509                 goto out;
 510         }
 511
 512         /* Add OVS_PACKET_ATTR_HASH */
 513         hash = skb_get_hash_raw(skb);
 514         if (skb->sw_hash)
 515                 hash |= OVS_PACKET_HASH_SW_BIT;
 516
 517         if (skb->l4_hash)
 518                 hash |= OVS_PACKET_HASH_L4_BIT;
 519
 520         if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
 521                 err = -ENOBUFS;
 522                 goto out;
 523         }
 524
 525         /* Only reserve room for attribute header, packet data is added
 526          * in skb_zerocopy() */
 527         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 528                 err = -ENOBUFS;
 529                 goto out;
 530         }
 531         nla->nla_len = nla_attr_size(skb->len - cutlen);
 532
 533         err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 534         if (err)
 535                 goto out;
 536
 537         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 538         pad_packet(dp, user_skb);
 539
 540         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 541
 542         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 543         user_skb = NULL;
 544 out:
 545         if (err)
 546                 skb_tx_error(skb);
 547         consume_skb(user_skb);
 548         consume_skb(nskb);
 549
 550         return err;
 551 }
 552
 553 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 554 {
 555         struct ovs_header *ovs_header = info->userhdr;
 556         struct net *net = sock_net(skb->sk);
 557         struct nlattr **a = info->attrs;
 558         struct sw_flow_actions *acts;
 559         struct sk_buff *packet;
 560         struct sw_flow *flow;
 561         struct sw_flow_actions *sf_acts;
 562         struct datapath *dp;
 563         struct vport *input_vport;
 564         u16 mru = 0;
 565         u64 hash;
 566         int len;
 567         int err;
 568         bool log = !a[OVS_PACKET_ATTR_PROBE];
 569
 570         err = -EINVAL;
 571         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 572             !a[OVS_PACKET_ATTR_ACTIONS])
 573                 goto err;
 574
 575         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 576         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 577         err = -ENOMEM;
 578         if (!packet)
 579                 goto err;
 580         skb_reserve(packet, NET_IP_ALIGN);
 581
 582         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 583
 584         /* Set packet's mru */
 585         if (a[OVS_PACKET_ATTR_MRU]) {
 586                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 587                 packet->ignore_df = 1;
 588         }
 589         OVS_CB(packet)->mru = mru;
 590
 591         if (a[OVS_PACKET_ATTR_HASH]) {
 592                 hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
 593
 594                 __skb_set_hash(packet, hash & 0xFFFFFFFFULL,
 595                                !!(hash & OVS_PACKET_HASH_SW_BIT),
 596                                !!(hash & OVS_PACKET_HASH_L4_BIT));
 597         }
 598
 599         /* Build an sw_flow for sending this packet. */
 600         flow = ovs_flow_alloc();
 601         err = PTR_ERR(flow);
 602         if (IS_ERR(flow))
 603                 goto err_kfree_skb;
 604
 605         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 606                                              packet, &flow->key, log);
 607         if (err)
 608                 goto err_flow_free;
 609
 610         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 611                                    &flow->key, &acts, log);
 612         if (err)
 613                 goto err_flow_free;
 614
 615         rcu_assign_pointer(flow->sf_acts, acts);
 616         packet->priority = flow->key.phy.priority;
 617         packet->mark = flow->key.phy.skb_mark;
 618
 619         rcu_read_lock();
 620         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 621         err = -ENODEV;
 622         if (!dp)
 623                 goto err_unlock;
 624
 625         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 626         if (!input_vport)
 627                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 628
 629         if (!input_vport)
 630                 goto err_unlock;
 631
 632         packet->dev = input_vport->dev;
 633         OVS_CB(packet)->input_vport = input_vport;
 634         sf_acts = rcu_dereference(flow->sf_acts);
 635
 636         local_bh_disable();
 637         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 638         local_bh_enable();
 639         rcu_read_unlock();
 640
 641         ovs_flow_free(flow, false);
 642         return err;
 643
 644 err_unlock:
 645         rcu_read_unlock();
 646 err_flow_free:
 647         ovs_flow_free(flow, false);
 648 err_kfree_skb:
 649         kfree_skb(packet);
 650 err:
 651         return err;
 652 }
 653
 654 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 655         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 656         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 657         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 658         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 659         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 660         [OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
 661 };
 662
 663 static const struct genl_small_ops dp_packet_genl_ops[] = {
 664         { .cmd = OVS_PACKET_CMD_EXECUTE,
 665           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 666           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 667           .doit = ovs_packet_cmd_execute
 668         }
 669 };
 670
 671 static struct genl_family dp_packet_genl_family __ro_after_init = {
 672         .hdrsize = sizeof(struct ovs_header),
 673         .name = OVS_PACKET_FAMILY,
 674         .version = OVS_PACKET_VERSION,
 675         .maxattr = OVS_PACKET_ATTR_MAX,
 676         .policy = packet_policy,
 677         .netnsok = true,
 678         .parallel_ops = true,
 679         .small_ops = dp_packet_genl_ops,
 680         .n_small_ops = ARRAY_SIZE(dp_packet_genl_ops),
 681         .module = THIS_MODULE,
 682 };
 683
 684 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 685                          struct ovs_dp_megaflow_stats *mega_stats)
 686 {
 687         int i;
 688
 689         memset(mega_stats, 0, sizeof(*mega_stats));
 690
 691         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 692         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 693
 694         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 695
 696         for_each_possible_cpu(i) {
 697                 const struct dp_stats_percpu *percpu_stats;
 698                 struct dp_stats_percpu local_stats;
 699                 unsigned int start;
 700
 701                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 702
 703                 do {
 704                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 705                         local_stats = *percpu_stats;
 706                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 707
 708                 stats->n_hit += local_stats.n_hit;
 709                 stats->n_missed += local_stats.n_missed;
 710                 stats->n_lost += local_stats.n_lost;
 711                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 712                 mega_stats->n_cache_hit += local_stats.n_cache_hit;
 713         }
 714 }
 715
 716 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 717 {
 718         return ovs_identifier_is_ufid(sfid) &&
 719                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 720 }
 721
 722 static bool should_fill_mask(uint32_t ufid_flags)
 723 {
 724         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 725 }
 726
 727 static bool should_fill_actions(uint32_t ufid_flags)
 728 {
 729         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 730 }
 731
 732 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 733                                     const struct sw_flow_id *sfid,
 734                                     uint32_t ufid_flags)
 735 {
 736         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 737
 738         /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
 739          * see ovs_nla_put_identifier()
 740          */
 741         if (sfid && ovs_identifier_is_ufid(sfid))
 742                 len += nla_total_size(sfid->ufid_len);
 743         else
 744                 len += nla_total_size(ovs_key_attr_size());
 745
 746         /* OVS_FLOW_ATTR_KEY */
 747         if (!sfid || should_fill_key(sfid, ufid_flags))
 748                 len += nla_total_size(ovs_key_attr_size());
 749
 750         /* OVS_FLOW_ATTR_MASK */
 751         if (should_fill_mask(ufid_flags))
 752                 len += nla_total_size(ovs_key_attr_size());
 753
 754         /* OVS_FLOW_ATTR_ACTIONS */
 755         if (should_fill_actions(ufid_flags))
 756                 len += nla_total_size(acts->orig_len);
 757
 758         return len
 759                 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 760                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 761                 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 762 }
 763
 764 /* Called with ovs_mutex or RCU read lock. */
 765 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 766                                    struct sk_buff *skb)
 767 {
 768         struct ovs_flow_stats stats;
 769         __be16 tcp_flags;
 770         unsigned long used;
 771
 772         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 773
 774         if (used &&
 775             nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 776                               OVS_FLOW_ATTR_PAD))
 777                 return -EMSGSIZE;
 778
 779         if (stats.n_packets &&
 780             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 781                           sizeof(struct ovs_flow_stats), &stats,
 782                           OVS_FLOW_ATTR_PAD))
 783                 return -EMSGSIZE;
 784
 785         if ((u8)ntohs(tcp_flags) &&
 786              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 787                 return -EMSGSIZE;
 788
 789         return 0;
 790 }
 791
 792 /* Called with ovs_mutex or RCU read lock. */
 793 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 794                                      struct sk_buff *skb, int skb_orig_len)
 795 {
 796         struct nlattr *start;
 797         int err;
 798
 799         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 800          * this is the first flow to be dumped into 'skb'.  This is unusual for
 801          * Netlink but individual action lists can be longer than
 802          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 803          * The userspace caller can always fetch the actions separately if it
 804          * really wants them.  (Most userspace callers in fact don't care.)
 805          *
 806          * This can only fail for dump operations because the skb is always
 807          * properly sized for single flows.
 808          */
 809         start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
 810         if (start) {
 811                 const struct sw_flow_actions *sf_acts;
 812
 813                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 814                 err = ovs_nla_put_actions(sf_acts->actions,
 815                                           sf_acts->actions_len, skb);
 816
 817                 if (!err)
 818                         nla_nest_end(skb, start);
 819                 else {
 820                         if (skb_orig_len)
 821                                 return err;
 822
 823                         nla_nest_cancel(skb, start);
 824                 }
 825         } else if (skb_orig_len) {
 826                 return -EMSGSIZE;
 827         }
 828
 829         return 0;
 830 }
 831
 832 /* Called with ovs_mutex or RCU read lock. */
 833 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 834                                   struct sk_buff *skb, u32 portid,
 835                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 836 {
 837         const int skb_orig_len = skb->len;
 838         struct ovs_header *ovs_header;
 839         int err;
 840
 841         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 842                                  flags, cmd);
 843         if (!ovs_header)
 844                 return -EMSGSIZE;
 845
 846         ovs_header->dp_ifindex = dp_ifindex;
 847
 848         err = ovs_nla_put_identifier(flow, skb);
 849         if (err)
 850                 goto error;
 851
 852         if (should_fill_key(&flow->id, ufid_flags)) {
 853                 err = ovs_nla_put_masked_key(flow, skb);
 854                 if (err)
 855                         goto error;
 856         }
 857
 858         if (should_fill_mask(ufid_flags)) {
 859                 err = ovs_nla_put_mask(flow, skb);
 860                 if (err)
 861                         goto error;
 862         }
 863
 864         err = ovs_flow_cmd_fill_stats(flow, skb);
 865         if (err)
 866                 goto error;
 867
 868         if (should_fill_actions(ufid_flags)) {
 869                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 870                 if (err)
 871                         goto error;
 872         }
 873
 874         genlmsg_end(skb, ovs_header);
 875         return 0;
 876
 877 error:
 878         genlmsg_cancel(skb, ovs_header);
 879         return err;
 880 }
 881
 882 /* May not be called with RCU read lock. */
 883 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 884                                                const struct sw_flow_id *sfid,
 885                                                struct genl_info *info,
 886                                                bool always,
 887                                                uint32_t ufid_flags)
 888 {
 889         struct sk_buff *skb;
 890         size_t len;
 891
 892         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 893                 return NULL;
 894
 895         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 896         skb = genlmsg_new(len, GFP_KERNEL);
 897         if (!skb)
 898                 return ERR_PTR(-ENOMEM);
 899
 900         return skb;
 901 }
 902
 903 /* Called with ovs_mutex. */
 904 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 905                                                int dp_ifindex,
 906                                                struct genl_info *info, u8 cmd,
 907                                                bool always, u32 ufid_flags)
 908 {
 909         struct sk_buff *skb;
 910         int retval;
 911
 912         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 913                                       &flow->id, info, always, ufid_flags);
 914         if (IS_ERR_OR_NULL(skb))
 915                 return skb;
 916
 917         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 918                                         info->snd_portid, info->snd_seq, 0,
 919                                         cmd, ufid_flags);
 920         if (WARN_ON_ONCE(retval < 0)) {
 921                 kfree_skb(skb);
 922                 skb = ERR_PTR(retval);
 923         }
 924         return skb;
 925 }
 926
 927 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 928 {
 929         struct net *net = sock_net(skb->sk);
 930         struct nlattr **a = info->attrs;
 931         struct ovs_header *ovs_header = info->userhdr;
 932         struct sw_flow *flow = NULL, *new_flow;
 933         struct sw_flow_mask mask;
 934         struct sk_buff *reply;
 935         struct datapath *dp;
 936         struct sw_flow_key *key;
 937         struct sw_flow_actions *acts;
 938         struct sw_flow_match match;
 939         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 940         int error;
 941         bool log = !a[OVS_FLOW_ATTR_PROBE];
 942
 943         /* Must have key and actions. */
 944         error = -EINVAL;
 945         if (!a[OVS_FLOW_ATTR_KEY]) {
 946                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 947                 goto error;
 948         }
 949         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 950                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 951                 goto error;
 952         }
 953
 954         /* Most of the time we need to allocate a new flow, do it before
 955          * locking.
 956          */
 957         new_flow = ovs_flow_alloc();
 958         if (IS_ERR(new_flow)) {
 959                 error = PTR_ERR(new_flow);
 960                 goto error;
 961         }
 962
 963         /* Extract key. */
 964         key = kzalloc(sizeof(*key), GFP_KERNEL);
 965         if (!key) {
 966                 error = -ENOMEM;
 967                 goto err_kfree_flow;
 968         }
 969
 970         ovs_match_init(&match, key, false, &mask);
 971         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 972                                   a[OVS_FLOW_ATTR_MASK], log);
 973         if (error)
 974                 goto err_kfree_key;
 975
 976         ovs_flow_mask_key(&new_flow->key, key, true, &mask);
 977
 978         /* Extract flow identifier. */
 979         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 980                                        key, log);
 981         if (error)
 982                 goto err_kfree_key;
 983
 984         /* Validate actions. */
 985         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 986                                      &new_flow->key, &acts, log);
 987         if (error) {
 988                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 989                 goto err_kfree_key;
 990         }
 991
 992         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 993                                         ufid_flags);
 994         if (IS_ERR(reply)) {
 995                 error = PTR_ERR(reply);
 996                 goto err_kfree_acts;
 997         }
 998
 999         ovs_lock();
1000         dp = get_dp(net, ovs_header->dp_ifindex);
1001         if (unlikely(!dp)) {
1002                 error = -ENODEV;
1003                 goto err_unlock_ovs;
1004         }
1005
1006         /* Check if this is a duplicate flow */
1007         if (ovs_identifier_is_ufid(&new_flow->id))
1008                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1009         if (!flow)
1010                 flow = ovs_flow_tbl_lookup(&dp->table, key);
1011         if (likely(!flow)) {
1012                 rcu_assign_pointer(new_flow->sf_acts, acts);
1013
1014                 /* Put flow in bucket. */
1015                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1016                 if (unlikely(error)) {
1017                         acts = NULL;
1018                         goto err_unlock_ovs;
1019                 }
1020
1021                 if (unlikely(reply)) {
1022                         error = ovs_flow_cmd_fill_info(new_flow,
1023                                                        ovs_header->dp_ifindex,
1024                                                        reply, info->snd_portid,
1025                                                        info->snd_seq, 0,
1026                                                        OVS_FLOW_CMD_NEW,
1027                                                        ufid_flags);
1028                         BUG_ON(error < 0);
1029                 }
1030                 ovs_unlock();
1031         } else {
1032                 struct sw_flow_actions *old_acts;
1033
1034                 /* Bail out if we're not allowed to modify an existing flow.
1035                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1036                  * because Generic Netlink treats the latter as a dump
1037                  * request.  We also accept NLM_F_EXCL in case that bug ever
1038                  * gets fixed.
1039                  */
1040                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1041                                                          | NLM_F_EXCL))) {
1042                         error = -EEXIST;
1043                         goto err_unlock_ovs;
1044                 }
1045                 /* The flow identifier has to be the same for flow updates.
1046                  * Look for any overlapping flow.
1047                  */
1048                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1049                         if (ovs_identifier_is_key(&flow->id))
1050                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1051                                                                  &match);
1052                         else /* UFID matches but key is different */
1053                                 flow = NULL;
1054                         if (!flow) {
1055                                 error = -ENOENT;
1056                                 goto err_unlock_ovs;
1057                         }
1058                 }
1059                 /* Update actions. */
1060                 old_acts = ovsl_dereference(flow->sf_acts);
1061                 rcu_assign_pointer(flow->sf_acts, acts);
1062
1063                 if (unlikely(reply)) {
1064                         error = ovs_flow_cmd_fill_info(flow,
1065                                                        ovs_header->dp_ifindex,
1066                                                        reply, info->snd_portid,
1067                                                        info->snd_seq, 0,
1068                                                        OVS_FLOW_CMD_NEW,
1069                                                        ufid_flags);
1070                         BUG_ON(error < 0);
1071                 }
1072                 ovs_unlock();
1073
1074                 ovs_nla_free_flow_actions_rcu(old_acts);
1075                 ovs_flow_free(new_flow, false);
1076         }
1077
1078         if (reply)
1079                 ovs_notify(&dp_flow_genl_family, reply, info);
1080
1081         kfree(key);
1082         return 0;
1083
1084 err_unlock_ovs:
1085         ovs_unlock();
1086         kfree_skb(reply);
1087 err_kfree_acts:
1088         ovs_nla_free_flow_actions(acts);
1089 err_kfree_key:
1090         kfree(key);
1091 err_kfree_flow:
1092         ovs_flow_free(new_flow, false);
1093 error:
1094         return error;
1095 }
1096
1097 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1098 static noinline_for_stack
1099 struct sw_flow_actions *get_flow_actions(struct net *net,
1100                                          const struct nlattr *a,
1101                                          const struct sw_flow_key *key,
1102                                          const struct sw_flow_mask *mask,
1103                                          bool log)
1104 {
1105         struct sw_flow_actions *acts;
1106         struct sw_flow_key masked_key;
1107         int error;
1108
1109         ovs_flow_mask_key(&masked_key, key, true, mask);
1110         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1111         if (error) {
1112                 OVS_NLERR(log,
1113                           "Actions may not be safe on all matching packets");
1114                 return ERR_PTR(error);
1115         }
1116
1117         return acts;
1118 }
1119
1120 /* Factor out match-init and action-copy to avoid
1121  * "Wframe-larger-than=1024" warning. Because mask is only
1122  * used to get actions, we new a function to save some
1123  * stack space.
1124  *
1125  * If there are not key and action attrs, we return 0
1126  * directly. In the case, the caller will also not use the
1127  * match as before. If there is action attr, we try to get
1128  * actions and save them to *acts. Before returning from
1129  * the function, we reset the match->mask pointer. Because
1130  * we should not to return match object with dangling reference
1131  * to mask.
1132  * */
1133 static noinline_for_stack int
1134 ovs_nla_init_match_and_action(struct net *net,
1135                               struct sw_flow_match *match,
1136                               struct sw_flow_key *key,
1137                               struct nlattr **a,
1138                               struct sw_flow_actions **acts,
1139                               bool log)
1140 {
1141         struct sw_flow_mask mask;
1142         int error = 0;
1143
1144         if (a[OVS_FLOW_ATTR_KEY]) {
1145                 ovs_match_init(match, key, true, &mask);
1146                 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1147                                           a[OVS_FLOW_ATTR_MASK], log);
1148                 if (error)
1149                         goto error;
1150         }
1151
1152         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1153                 if (!a[OVS_FLOW_ATTR_KEY]) {
1154                         OVS_NLERR(log,
1155                                   "Flow key attribute not present in set flow.");
1156                         error = -EINVAL;
1157                         goto error;
1158                 }
1159
1160                 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1161                                          &mask, log);
1162                 if (IS_ERR(*acts)) {
1163                         error = PTR_ERR(*acts);
1164                         goto error;
1165                 }
1166         }
1167
1168         /* On success, error is 0. */
1169 error:
1170         match->mask = NULL;
1171         return error;
1172 }
1173
1174 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1175 {
1176         struct net *net = sock_net(skb->sk);
1177         struct nlattr **a = info->attrs;
1178         struct ovs_header *ovs_header = info->userhdr;
1179         struct sw_flow_key key;
1180         struct sw_flow *flow;
1181         struct sk_buff *reply = NULL;
1182         struct datapath *dp;
1183         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1184         struct sw_flow_match match;
1185         struct sw_flow_id sfid;
1186         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1187         int error = 0;
1188         bool log = !a[OVS_FLOW_ATTR_PROBE];
1189         bool ufid_present;
1190
1191         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1192         if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1193                 OVS_NLERR(log,
1194                           "Flow set message rejected, Key attribute missing.");
1195                 return -EINVAL;
1196         }
1197
1198         error = ovs_nla_init_match_and_action(net, &match, &key, a,
1199                                               &acts, log);
1200         if (error)
1201                 goto error;
1202
1203         if (acts) {
1204                 /* Can allocate before locking if have acts. */
1205                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1206                                                 ufid_flags);
1207                 if (IS_ERR(reply)) {
1208                         error = PTR_ERR(reply);
1209                         goto err_kfree_acts;
1210                 }
1211         }
1212
1213         ovs_lock();
1214         dp = get_dp(net, ovs_header->dp_ifindex);
1215         if (unlikely(!dp)) {
1216                 error = -ENODEV;
1217                 goto err_unlock_ovs;
1218         }
1219         /* Check that the flow exists. */
1220         if (ufid_present)
1221                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1222         else
1223                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1224         if (unlikely(!flow)) {
1225                 error = -ENOENT;
1226                 goto err_unlock_ovs;
1227         }
1228
1229         /* Update actions, if present. */
1230         if (likely(acts)) {
1231                 old_acts = ovsl_dereference(flow->sf_acts);
1232                 rcu_assign_pointer(flow->sf_acts, acts);
1233
1234                 if (unlikely(reply)) {
1235                         error = ovs_flow_cmd_fill_info(flow,
1236                                                        ovs_header->dp_ifindex,
1237                                                        reply, info->snd_portid,
1238                                                        info->snd_seq, 0,
1239                                                        OVS_FLOW_CMD_SET,
1240                                                        ufid_flags);
1241                         BUG_ON(error < 0);
1242                 }
1243         } else {
1244                 /* Could not alloc without acts before locking. */
1245                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1246                                                 info, OVS_FLOW_CMD_SET, false,
1247                                                 ufid_flags);
1248
1249                 if (IS_ERR(reply)) {
1250                         error = PTR_ERR(reply);
1251                         goto err_unlock_ovs;
1252                 }
1253         }
1254
1255         /* Clear stats. */
1256         if (a[OVS_FLOW_ATTR_CLEAR])
1257                 ovs_flow_stats_clear(flow);
1258         ovs_unlock();
1259
1260         if (reply)
1261                 ovs_notify(&dp_flow_genl_family, reply, info);
1262         if (old_acts)
1263                 ovs_nla_free_flow_actions_rcu(old_acts);
1264
1265         return 0;
1266
1267 err_unlock_ovs:
1268         ovs_unlock();
1269         kfree_skb(reply);
1270 err_kfree_acts:
1271         ovs_nla_free_flow_actions(acts);
1272 error:
1273         return error;
1274 }
1275
1276 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1277 {
1278         struct nlattr **a = info->attrs;
1279         struct ovs_header *ovs_header = info->userhdr;
1280         struct net *net = sock_net(skb->sk);
1281         struct sw_flow_key key;
1282         struct sk_buff *reply;
1283         struct sw_flow *flow;
1284         struct datapath *dp;
1285         struct sw_flow_match match;
1286         struct sw_flow_id ufid;
1287         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1288         int err = 0;
1289         bool log = !a[OVS_FLOW_ATTR_PROBE];
1290         bool ufid_present;
1291
1292         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1293         if (a[OVS_FLOW_ATTR_KEY]) {
1294                 ovs_match_init(&match, &key, true, NULL);
1295                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1296                                         log);
1297         } else if (!ufid_present) {
1298                 OVS_NLERR(log,
1299                           "Flow get message rejected, Key attribute missing.");
1300                 err = -EINVAL;
1301         }
1302         if (err)
1303                 return err;
1304
1305         ovs_lock();
1306         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1307         if (!dp) {
1308                 err = -ENODEV;
1309                 goto unlock;
1310         }
1311
1312         if (ufid_present)
1313                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1314         else
1315                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1316         if (!flow) {
1317                 err = -ENOENT;
1318                 goto unlock;
1319         }
1320
1321         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1322                                         OVS_FLOW_CMD_GET, true, ufid_flags);
1323         if (IS_ERR(reply)) {
1324                 err = PTR_ERR(reply);
1325                 goto unlock;
1326         }
1327
1328         ovs_unlock();
1329         return genlmsg_reply(reply, info);
1330 unlock:
1331         ovs_unlock();
1332         return err;
1333 }
1334
1335 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1336 {
1337         struct nlattr **a = info->attrs;
1338         struct ovs_header *ovs_header = info->userhdr;
1339         struct net *net = sock_net(skb->sk);
1340         struct sw_flow_key key;
1341         struct sk_buff *reply;
1342         struct sw_flow *flow = NULL;
1343         struct datapath *dp;
1344         struct sw_flow_match match;
1345         struct sw_flow_id ufid;
1346         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1347         int err;
1348         bool log = !a[OVS_FLOW_ATTR_PROBE];
1349         bool ufid_present;
1350
1351         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1352         if (a[OVS_FLOW_ATTR_KEY]) {
1353                 ovs_match_init(&match, &key, true, NULL);
1354                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1355                                         NULL, log);
1356                 if (unlikely(err))
1357                         return err;
1358         }
1359
1360         ovs_lock();
1361         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1362         if (unlikely(!dp)) {
1363                 err = -ENODEV;
1364                 goto unlock;
1365         }
1366
1367         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1368                 err = ovs_flow_tbl_flush(&dp->table);
1369                 goto unlock;
1370         }
1371
1372         if (ufid_present)
1373                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1374         else
1375                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1376         if (unlikely(!flow)) {
1377                 err = -ENOENT;
1378                 goto unlock;
1379         }
1380
1381         ovs_flow_tbl_remove(&dp->table, flow);
1382         ovs_unlock();
1383
1384         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1385                                         &flow->id, info, false, ufid_flags);
1386         if (likely(reply)) {
1387                 if (!IS_ERR(reply)) {
1388                         rcu_read_lock();        /*To keep RCU checker happy. */
1389                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1390                                                      reply, info->snd_portid,
1391                                                      info->snd_seq, 0,
1392                                                      OVS_FLOW_CMD_DEL,
1393                                                      ufid_flags);
1394                         rcu_read_unlock();
1395                         if (WARN_ON_ONCE(err < 0)) {
1396                                 kfree_skb(reply);
1397                                 goto out_free;
1398                         }
1399
1400                         ovs_notify(&dp_flow_genl_family, reply, info);
1401                 } else {
1402                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1403                                         PTR_ERR(reply));
1404                 }
1405         }
1406
1407 out_free:
1408         ovs_flow_free(flow, true);
1409         return 0;
1410 unlock:
1411         ovs_unlock();
1412         return err;
1413 }
1414
1415 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1416 {
1417         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1418         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1419         struct table_instance *ti;
1420         struct datapath *dp;
1421         u32 ufid_flags;
1422         int err;
1423
1424         err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1425                                        OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1426         if (err)
1427                 return err;
1428         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1429
1430         rcu_read_lock();
1431         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1432         if (!dp) {
1433                 rcu_read_unlock();
1434                 return -ENODEV;
1435         }
1436
1437         ti = rcu_dereference(dp->table.ti);
1438         for (;;) {
1439                 struct sw_flow *flow;
1440                 u32 bucket, obj;
1441
1442                 bucket = cb->args[0];
1443                 obj = cb->args[1];
1444                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1445                 if (!flow)
1446                         break;
1447
1448                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1449                                            NETLINK_CB(cb->skb).portid,
1450                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1451                                            OVS_FLOW_CMD_GET, ufid_flags) < 0)
1452                         break;
1453
1454                 cb->args[0] = bucket;
1455                 cb->args[1] = obj;
1456         }
1457         rcu_read_unlock();
1458         return skb->len;
1459 }
1460
1461 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1462         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1463         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1464         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1465         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1466         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1467         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1468         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1469 };
1470
1471 static const struct genl_small_ops dp_flow_genl_ops[] = {
1472         { .cmd = OVS_FLOW_CMD_NEW,
1473           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1474           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1475           .doit = ovs_flow_cmd_new
1476         },
1477         { .cmd = OVS_FLOW_CMD_DEL,
1478           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1479           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1480           .doit = ovs_flow_cmd_del
1481         },
1482         { .cmd = OVS_FLOW_CMD_GET,
1483           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1484           .flags = 0,               /* OK for unprivileged users. */
1485           .doit = ovs_flow_cmd_get,
1486           .dumpit = ovs_flow_cmd_dump
1487         },
1488         { .cmd = OVS_FLOW_CMD_SET,
1489           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1490           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1491           .doit = ovs_flow_cmd_set,
1492         },
1493 };
1494
1495 static struct genl_family dp_flow_genl_family __ro_after_init = {
1496         .hdrsize = sizeof(struct ovs_header),
1497         .name = OVS_FLOW_FAMILY,
1498         .version = OVS_FLOW_VERSION,
1499         .maxattr = OVS_FLOW_ATTR_MAX,
1500         .policy = flow_policy,
1501         .netnsok = true,
1502         .parallel_ops = true,
1503         .small_ops = dp_flow_genl_ops,
1504         .n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
1505         .mcgrps = &ovs_dp_flow_multicast_group,
1506         .n_mcgrps = 1,
1507         .module = THIS_MODULE,
1508 };
1509
1510 static size_t ovs_dp_cmd_msg_size(void)
1511 {
1512         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1513
1514         msgsize += nla_total_size(IFNAMSIZ);
1515         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1516         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1517         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1518         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1519
1520         return msgsize;
1521 }
1522
1523 /* Called with ovs_mutex. */
1524 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1525                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1526 {
1527         struct ovs_header *ovs_header;
1528         struct ovs_dp_stats dp_stats;
1529         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1530         int err;
1531
1532         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1533                                  flags, cmd);
1534         if (!ovs_header)
1535                 goto error;
1536
1537         ovs_header->dp_ifindex = get_dpifindex(dp);
1538
1539         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1540         if (err)
1541                 goto nla_put_failure;
1542
1543         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1544         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1545                           &dp_stats, OVS_DP_ATTR_PAD))
1546                 goto nla_put_failure;
1547
1548         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1549                           sizeof(struct ovs_dp_megaflow_stats),
1550                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1551                 goto nla_put_failure;
1552
1553         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1554                 goto nla_put_failure;
1555
1556         if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1557                         ovs_flow_tbl_masks_cache_size(&dp->table)))
1558                 goto nla_put_failure;
1559
1560         genlmsg_end(skb, ovs_header);
1561         return 0;
1562
1563 nla_put_failure:
1564         genlmsg_cancel(skb, ovs_header);
1565 error:
1566         return -EMSGSIZE;
1567 }
1568
1569 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1570 {
1571         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1572 }
1573
1574 /* Called with rcu_read_lock or ovs_mutex. */
1575 static struct datapath *lookup_datapath(struct net *net,
1576                                         const struct ovs_header *ovs_header,
1577                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1578 {
1579         struct datapath *dp;
1580
1581         if (!a[OVS_DP_ATTR_NAME])
1582                 dp = get_dp(net, ovs_header->dp_ifindex);
1583         else {
1584                 struct vport *vport;
1585
1586                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1587                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1588         }
1589         return dp ? dp : ERR_PTR(-ENODEV);
1590 }
1591
1592 static void ovs_dp_reset_user_features(struct sk_buff *skb,
1593                                        struct genl_info *info)
1594 {
1595         struct datapath *dp;
1596
1597         dp = lookup_datapath(sock_net(skb->sk), info->userhdr,
1598                              info->attrs);
1599         if (IS_ERR(dp))
1600                 return;
1601
1602         pr_warn("%s: Dropping previously announced user features\n",
1603                 ovs_dp_name(dp));
1604         dp->user_features = 0;
1605 }
1606
1607 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1608
1609 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1610 {
1611         u32 user_features = 0;
1612
1613         if (a[OVS_DP_ATTR_USER_FEATURES]) {
1614                 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1615
1616                 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1617                                       OVS_DP_F_UNALIGNED |
1618                                       OVS_DP_F_TC_RECIRC_SHARING))
1619                         return -EOPNOTSUPP;
1620
1621 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1622                 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1623                         return -EOPNOTSUPP;
1624 #endif
1625         }
1626
1627         if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1628                 int err;
1629                 u32 cache_size;
1630
1631                 cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1632                 err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1633                 if (err)
1634                         return err;
1635         }
1636
1637         dp->user_features = user_features;
1638
1639         if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1640                 static_branch_enable(&tc_recirc_sharing_support);
1641         else
1642                 static_branch_disable(&tc_recirc_sharing_support);
1643
1644         return 0;
1645 }
1646
1647 static int ovs_dp_stats_init(struct datapath *dp)
1648 {
1649         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1650         if (!dp->stats_percpu)
1651                 return -ENOMEM;
1652
1653         return 0;
1654 }
1655
1656 static int ovs_dp_vport_init(struct datapath *dp)
1657 {
1658         int i;
1659
1660         dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1661                                   sizeof(struct hlist_head),
1662                                   GFP_KERNEL);
1663         if (!dp->ports)
1664                 return -ENOMEM;
1665
1666         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1667                 INIT_HLIST_HEAD(&dp->ports[i]);
1668
1669         return 0;
1670 }
1671
1672 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1673 {
1674         struct nlattr **a = info->attrs;
1675         struct vport_parms parms;
1676         struct sk_buff *reply;
1677         struct datapath *dp;
1678         struct vport *vport;
1679         struct ovs_net *ovs_net;
1680         int err;
1681
1682         err = -EINVAL;
1683         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1684                 goto err;
1685
1686         reply = ovs_dp_cmd_alloc_info();
1687         if (!reply)
1688                 return -ENOMEM;
1689
1690         err = -ENOMEM;
1691         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1692         if (dp == NULL)
1693                 goto err_destroy_reply;
1694
1695         ovs_dp_set_net(dp, sock_net(skb->sk));
1696
1697         /* Allocate table. */
1698         err = ovs_flow_tbl_init(&dp->table);
1699         if (err)
1700                 goto err_destroy_dp;
1701
1702         err = ovs_dp_stats_init(dp);
1703         if (err)
1704                 goto err_destroy_table;
1705
1706         err = ovs_dp_vport_init(dp);
1707         if (err)
1708                 goto err_destroy_stats;
1709
1710         err = ovs_meters_init(dp);
1711         if (err)
1712                 goto err_destroy_ports;
1713
1714         /* Set up our datapath device. */
1715         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1716         parms.type = OVS_VPORT_TYPE_INTERNAL;
1717         parms.options = NULL;
1718         parms.dp = dp;
1719         parms.port_no = OVSP_LOCAL;
1720         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1721
1722         /* So far only local changes have been made, now need the lock. */
1723         ovs_lock();
1724
1725         err = ovs_dp_change(dp, a);
1726         if (err)
1727                 goto err_unlock_and_destroy_meters;
1728
1729         vport = new_vport(&parms);
1730         if (IS_ERR(vport)) {
1731                 err = PTR_ERR(vport);
1732                 if (err == -EBUSY)
1733                         err = -EEXIST;
1734
1735                 if (err == -EEXIST) {
1736                         /* An outdated user space instance that does not understand
1737                          * the concept of user_features has attempted to create a new
1738                          * datapath and is likely to reuse it. Drop all user features.
1739                          */
1740                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1741                                 ovs_dp_reset_user_features(skb, info);
1742                 }
1743
1744                 goto err_unlock_and_destroy_meters;
1745         }
1746
1747         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1748                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1749         BUG_ON(err < 0);
1750
1751         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1752         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1753
1754         ovs_unlock();
1755
1756         ovs_notify(&dp_datapath_genl_family, reply, info);
1757         return 0;
1758
1759 err_unlock_and_destroy_meters:
1760         ovs_unlock();
1761         ovs_meters_exit(dp);
1762 err_destroy_ports:
1763         kfree(dp->ports);
1764 err_destroy_stats:
1765         free_percpu(dp->stats_percpu);
1766 err_destroy_table:
1767         ovs_flow_tbl_destroy(&dp->table);
1768 err_destroy_dp:
1769         kfree(dp);
1770 err_destroy_reply:
1771         kfree_skb(reply);
1772 err:
1773         return err;
1774 }
1775
1776 /* Called with ovs_mutex. */
1777 static void __dp_destroy(struct datapath *dp)
1778 {
1779         struct flow_table *table = &dp->table;
1780         int i;
1781
1782         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1783                 struct vport *vport;
1784                 struct hlist_node *n;
1785
1786                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1787                         if (vport->port_no != OVSP_LOCAL)
1788                                 ovs_dp_detach_port(vport);
1789         }
1790
1791         list_del_rcu(&dp->list_node);
1792
1793         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1794          * all ports in datapath are destroyed first before freeing datapath.
1795          */
1796         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1797
1798         /* Flush sw_flow in the tables. RCU cb only releases resource
1799          * such as dp, ports and tables. That may avoid some issues
1800          * such as RCU usage warning.
1801          */
1802         table_instance_flow_flush(table, ovsl_dereference(table->ti),
1803                                   ovsl_dereference(table->ufid_ti));
1804
1805         /* RCU destroy the ports, meters and flow tables. */
1806         call_rcu(&dp->rcu, destroy_dp_rcu);
1807 }
1808
1809 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1810 {
1811         struct sk_buff *reply;
1812         struct datapath *dp;
1813         int err;
1814
1815         reply = ovs_dp_cmd_alloc_info();
1816         if (!reply)
1817                 return -ENOMEM;
1818
1819         ovs_lock();
1820         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1821         err = PTR_ERR(dp);
1822         if (IS_ERR(dp))
1823                 goto err_unlock_free;
1824
1825         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1826                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1827         BUG_ON(err < 0);
1828
1829         __dp_destroy(dp);
1830         ovs_unlock();
1831
1832         ovs_notify(&dp_datapath_genl_family, reply, info);
1833
1834         return 0;
1835
1836 err_unlock_free:
1837         ovs_unlock();
1838         kfree_skb(reply);
1839         return err;
1840 }
1841
1842 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1843 {
1844         struct sk_buff *reply;
1845         struct datapath *dp;
1846         int err;
1847
1848         reply = ovs_dp_cmd_alloc_info();
1849         if (!reply)
1850                 return -ENOMEM;
1851
1852         ovs_lock();
1853         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1854         err = PTR_ERR(dp);
1855         if (IS_ERR(dp))
1856                 goto err_unlock_free;
1857
1858         err = ovs_dp_change(dp, info->attrs);
1859         if (err)
1860                 goto err_unlock_free;
1861
1862         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1863                                    info->snd_seq, 0, OVS_DP_CMD_SET);
1864         BUG_ON(err < 0);
1865
1866         ovs_unlock();
1867         ovs_notify(&dp_datapath_genl_family, reply, info);
1868
1869         return 0;
1870
1871 err_unlock_free:
1872         ovs_unlock();
1873         kfree_skb(reply);
1874         return err;
1875 }
1876
1877 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1878 {
1879         struct sk_buff *reply;
1880         struct datapath *dp;
1881         int err;
1882
1883         reply = ovs_dp_cmd_alloc_info();
1884         if (!reply)
1885                 return -ENOMEM;
1886
1887         ovs_lock();
1888         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1889         if (IS_ERR(dp)) {
1890                 err = PTR_ERR(dp);
1891                 goto err_unlock_free;
1892         }
1893         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1894                                    info->snd_seq, 0, OVS_DP_CMD_GET);
1895         BUG_ON(err < 0);
1896         ovs_unlock();
1897
1898         return genlmsg_reply(reply, info);
1899
1900 err_unlock_free:
1901         ovs_unlock();
1902         kfree_skb(reply);
1903         return err;
1904 }
1905
1906 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1907 {
1908         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1909         struct datapath *dp;
1910         int skip = cb->args[0];
1911         int i = 0;
1912
1913         ovs_lock();
1914         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1915                 if (i >= skip &&
1916                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1917                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1918                                          OVS_DP_CMD_GET) < 0)
1919                         break;
1920                 i++;
1921         }
1922         ovs_unlock();
1923
1924         cb->args[0] = i;
1925
1926         return skb->len;
1927 }
1928
1929 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1930         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1931         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1932         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1933         [OVS_DP_ATTR_MASKS_CACHE_SIZE] =  NLA_POLICY_RANGE(NLA_U32, 0,
1934                 PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
1935 };
1936
1937 static const struct genl_small_ops dp_datapath_genl_ops[] = {
1938         { .cmd = OVS_DP_CMD_NEW,
1939           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1940           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1941           .doit = ovs_dp_cmd_new
1942         },
1943         { .cmd = OVS_DP_CMD_DEL,
1944           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1945           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1946           .doit = ovs_dp_cmd_del
1947         },
1948         { .cmd = OVS_DP_CMD_GET,
1949           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1950           .flags = 0,               /* OK for unprivileged users. */
1951           .doit = ovs_dp_cmd_get,
1952           .dumpit = ovs_dp_cmd_dump
1953         },
1954         { .cmd = OVS_DP_CMD_SET,
1955           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1956           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1957           .doit = ovs_dp_cmd_set,
1958         },
1959 };
1960
1961 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1962         .hdrsize = sizeof(struct ovs_header),
1963         .name = OVS_DATAPATH_FAMILY,
1964         .version = OVS_DATAPATH_VERSION,
1965         .maxattr = OVS_DP_ATTR_MAX,
1966         .policy = datapath_policy,
1967         .netnsok = true,
1968         .parallel_ops = true,
1969         .small_ops = dp_datapath_genl_ops,
1970         .n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1971         .mcgrps = &ovs_dp_datapath_multicast_group,
1972         .n_mcgrps = 1,
1973         .module = THIS_MODULE,
1974 };
1975
1976 /* Called with ovs_mutex or RCU read lock. */
1977 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1978                                    struct net *net, u32 portid, u32 seq,
1979                                    u32 flags, u8 cmd, gfp_t gfp)
1980 {
1981         struct ovs_header *ovs_header;
1982         struct ovs_vport_stats vport_stats;
1983         int err;
1984
1985         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1986                                  flags, cmd);
1987         if (!ovs_header)
1988                 return -EMSGSIZE;
1989
1990         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1991
1992         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1993             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1994             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1995                            ovs_vport_name(vport)) ||
1996             nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1997                 goto nla_put_failure;
1998
1999         if (!net_eq(net, dev_net(vport->dev))) {
2000                 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
2001
2002                 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
2003                         goto nla_put_failure;
2004         }
2005
2006         ovs_vport_get_stats(vport, &vport_stats);
2007         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
2008                           sizeof(struct ovs_vport_stats), &vport_stats,
2009                           OVS_VPORT_ATTR_PAD))
2010                 goto nla_put_failure;
2011
2012         if (ovs_vport_get_upcall_portids(vport, skb))
2013                 goto nla_put_failure;
2014
2015         err = ovs_vport_get_options(vport, skb);
2016         if (err == -EMSGSIZE)
2017                 goto error;
2018
2019         genlmsg_end(skb, ovs_header);
2020         return 0;
2021
2022 nla_put_failure:
2023         err = -EMSGSIZE;
2024 error:
2025         genlmsg_cancel(skb, ovs_header);
2026         return err;
2027 }
2028
2029 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2030 {
2031         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2032 }
2033
2034 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2035 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2036                                          u32 portid, u32 seq, u8 cmd)
2037 {
2038         struct sk_buff *skb;
2039         int retval;
2040
2041         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2042         if (!skb)
2043                 return ERR_PTR(-ENOMEM);
2044
2045         retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2046                                          GFP_KERNEL);
2047         BUG_ON(retval < 0);
2048
2049         return skb;
2050 }
2051
2052 /* Called with ovs_mutex or RCU read lock. */
2053 static struct vport *lookup_vport(struct net *net,
2054                                   const struct ovs_header *ovs_header,
2055                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2056 {
2057         struct datapath *dp;
2058         struct vport *vport;
2059
2060         if (a[OVS_VPORT_ATTR_IFINDEX])
2061                 return ERR_PTR(-EOPNOTSUPP);
2062         if (a[OVS_VPORT_ATTR_NAME]) {
2063                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2064                 if (!vport)
2065                         return ERR_PTR(-ENODEV);
2066                 if (ovs_header->dp_ifindex &&
2067                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2068                         return ERR_PTR(-ENODEV);
2069                 return vport;
2070         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2071                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2072
2073                 if (port_no >= DP_MAX_PORTS)
2074                         return ERR_PTR(-EFBIG);
2075
2076                 dp = get_dp(net, ovs_header->dp_ifindex);
2077                 if (!dp)
2078                         return ERR_PTR(-ENODEV);
2079
2080                 vport = ovs_vport_ovsl_rcu(dp, port_no);
2081                 if (!vport)
2082                         return ERR_PTR(-ENODEV);
2083                 return vport;
2084         } else
2085                 return ERR_PTR(-EINVAL);
2086
2087 }
2088
2089 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2090 {
2091         unsigned int dev_headroom, max_headroom = 0;
2092         struct net_device *dev;
2093         struct vport *vport;
2094         int i;
2095
2096         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2097                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2098                                          lockdep_ovsl_is_held()) {
2099                         dev = vport->dev;
2100                         dev_headroom = netdev_get_fwd_headroom(dev);
2101                         if (dev_headroom > max_headroom)
2102                                 max_headroom = dev_headroom;
2103                 }
2104         }
2105
2106         return max_headroom;
2107 }
2108
2109 /* Called with ovs_mutex */
2110 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2111 {
2112         struct vport *vport;
2113         int i;
2114
2115         dp->max_headroom = new_headroom;
2116         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2117                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2118                                          lockdep_ovsl_is_held())
2119                         netdev_set_rx_headroom(vport->dev, new_headroom);
2120         }
2121 }
2122
2123 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2124 {
2125         struct nlattr **a = info->attrs;
2126         struct ovs_header *ovs_header = info->userhdr;
2127         struct vport_parms parms;
2128         struct sk_buff *reply;
2129         struct vport *vport;
2130         struct datapath *dp;
2131         unsigned int new_headroom;
2132         u32 port_no;
2133         int err;
2134
2135         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2136             !a[OVS_VPORT_ATTR_UPCALL_PID])
2137                 return -EINVAL;
2138         if (a[OVS_VPORT_ATTR_IFINDEX])
2139                 return -EOPNOTSUPP;
2140
2141         port_no = a[OVS_VPORT_ATTR_PORT_NO]
2142                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2143         if (port_no >= DP_MAX_PORTS)
2144                 return -EFBIG;
2145
2146         reply = ovs_vport_cmd_alloc_info();
2147         if (!reply)
2148                 return -ENOMEM;
2149
2150         ovs_lock();
2151 restart:
2152         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2153         err = -ENODEV;
2154         if (!dp)
2155                 goto exit_unlock_free;
2156
2157         if (port_no) {
2158                 vport = ovs_vport_ovsl(dp, port_no);
2159                 err = -EBUSY;
2160                 if (vport)
2161                         goto exit_unlock_free;
2162         } else {
2163                 for (port_no = 1; ; port_no++) {
2164                         if (port_no >= DP_MAX_PORTS) {
2165                                 err = -EFBIG;
2166                                 goto exit_unlock_free;
2167                         }
2168                         vport = ovs_vport_ovsl(dp, port_no);
2169                         if (!vport)
2170                                 break;
2171                 }
2172         }
2173
2174         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2175         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2176         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2177         parms.dp = dp;
2178         parms.port_no = port_no;
2179         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2180
2181         vport = new_vport(&parms);
2182         err = PTR_ERR(vport);
2183         if (IS_ERR(vport)) {
2184                 if (err == -EAGAIN)
2185                         goto restart;
2186                 goto exit_unlock_free;
2187         }
2188
2189         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2190                                       info->snd_portid, info->snd_seq, 0,
2191                                       OVS_VPORT_CMD_NEW, GFP_KERNEL);
2192
2193         new_headroom = netdev_get_fwd_headroom(vport->dev);
2194
2195         if (new_headroom > dp->max_headroom)
2196                 ovs_update_headroom(dp, new_headroom);
2197         else
2198                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2199
2200         BUG_ON(err < 0);
2201         ovs_unlock();
2202
2203         ovs_notify(&dp_vport_genl_family, reply, info);
2204         return 0;
2205
2206 exit_unlock_free:
2207         ovs_unlock();
2208         kfree_skb(reply);
2209         return err;
2210 }
2211
2212 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2213 {
2214         struct nlattr **a = info->attrs;
2215         struct sk_buff *reply;
2216         struct vport *vport;
2217         int err;
2218
2219         reply = ovs_vport_cmd_alloc_info();
2220         if (!reply)
2221                 return -ENOMEM;
2222
2223         ovs_lock();
2224         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2225         err = PTR_ERR(vport);
2226         if (IS_ERR(vport))
2227                 goto exit_unlock_free;
2228
2229         if (a[OVS_VPORT_ATTR_TYPE] &&
2230             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2231                 err = -EINVAL;
2232                 goto exit_unlock_free;
2233         }
2234
2235         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2236                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2237                 if (err)
2238                         goto exit_unlock_free;
2239         }
2240
2241
2242         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2243                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2244
2245                 err = ovs_vport_set_upcall_portids(vport, ids);
2246                 if (err)
2247                         goto exit_unlock_free;
2248         }
2249
2250         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2251                                       info->snd_portid, info->snd_seq, 0,
2252                                       OVS_VPORT_CMD_SET, GFP_KERNEL);
2253         BUG_ON(err < 0);
2254
2255         ovs_unlock();
2256         ovs_notify(&dp_vport_genl_family, reply, info);
2257         return 0;
2258
2259 exit_unlock_free:
2260         ovs_unlock();
2261         kfree_skb(reply);
2262         return err;
2263 }
2264
2265 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2266 {
2267         bool update_headroom = false;
2268         struct nlattr **a = info->attrs;
2269         struct sk_buff *reply;
2270         struct datapath *dp;
2271         struct vport *vport;
2272         unsigned int new_headroom;
2273         int err;
2274
2275         reply = ovs_vport_cmd_alloc_info();
2276         if (!reply)
2277                 return -ENOMEM;
2278
2279         ovs_lock();
2280         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2281         err = PTR_ERR(vport);
2282         if (IS_ERR(vport))
2283                 goto exit_unlock_free;
2284
2285         if (vport->port_no == OVSP_LOCAL) {
2286                 err = -EINVAL;
2287                 goto exit_unlock_free;
2288         }
2289
2290         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2291                                       info->snd_portid, info->snd_seq, 0,
2292                                       OVS_VPORT_CMD_DEL, GFP_KERNEL);
2293         BUG_ON(err < 0);
2294
2295         /* the vport deletion may trigger dp headroom update */
2296         dp = vport->dp;
2297         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2298                 update_headroom = true;
2299
2300         netdev_reset_rx_headroom(vport->dev);
2301         ovs_dp_detach_port(vport);
2302
2303         if (update_headroom) {
2304                 new_headroom = ovs_get_max_headroom(dp);
2305
2306                 if (new_headroom < dp->max_headroom)
2307                         ovs_update_headroom(dp, new_headroom);
2308         }
2309         ovs_unlock();
2310
2311         ovs_notify(&dp_vport_genl_family, reply, info);
2312         return 0;
2313
2314 exit_unlock_free:
2315         ovs_unlock();
2316         kfree_skb(reply);
2317         return err;
2318 }
2319
2320 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2321 {
2322         struct nlattr **a = info->attrs;
2323         struct ovs_header *ovs_header = info->userhdr;
2324         struct sk_buff *reply;
2325         struct vport *vport;
2326         int err;
2327
2328         reply = ovs_vport_cmd_alloc_info();
2329         if (!reply)
2330                 return -ENOMEM;
2331
2332         rcu_read_lock();
2333         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2334         err = PTR_ERR(vport);
2335         if (IS_ERR(vport))
2336                 goto exit_unlock_free;
2337         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2338                                       info->snd_portid, info->snd_seq, 0,
2339                                       OVS_VPORT_CMD_GET, GFP_ATOMIC);
2340         BUG_ON(err < 0);
2341         rcu_read_unlock();
2342
2343         return genlmsg_reply(reply, info);
2344
2345 exit_unlock_free:
2346         rcu_read_unlock();
2347         kfree_skb(reply);
2348         return err;
2349 }
2350
2351 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2352 {
2353         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2354         struct datapath *dp;
2355         int bucket = cb->args[0], skip = cb->args[1];
2356         int i, j = 0;
2357
2358         rcu_read_lock();
2359         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2360         if (!dp) {
2361                 rcu_read_unlock();
2362                 return -ENODEV;
2363         }
2364         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2365                 struct vport *vport;
2366
2367                 j = 0;
2368                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2369                         if (j >= skip &&
2370                             ovs_vport_cmd_fill_info(vport, skb,
2371                                                     sock_net(skb->sk),
2372                                                     NETLINK_CB(cb->skb).portid,
2373                                                     cb->nlh->nlmsg_seq,
2374                                                     NLM_F_MULTI,
2375                                                     OVS_VPORT_CMD_GET,
2376                                                     GFP_ATOMIC) < 0)
2377                                 goto out;
2378
2379                         j++;
2380                 }
2381                 skip = 0;
2382         }
2383 out:
2384         rcu_read_unlock();
2385
2386         cb->args[0] = i;
2387         cb->args[1] = j;
2388
2389         return skb->len;
2390 }
2391
2392 static void ovs_dp_masks_rebalance(struct work_struct *work)
2393 {
2394         struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2395                                                masks_rebalance.work);
2396         struct datapath *dp;
2397
2398         ovs_lock();
2399
2400         list_for_each_entry(dp, &ovs_net->dps, list_node)
2401                 ovs_flow_masks_rebalance(&dp->table);
2402
2403         ovs_unlock();
2404
2405         schedule_delayed_work(&ovs_net->masks_rebalance,
2406                               msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2407 }
2408
2409 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2410         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2411         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2412         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2413         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2414         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2415         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2416         [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2417         [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2418 };
2419
2420 static const struct genl_small_ops dp_vport_genl_ops[] = {
2421         { .cmd = OVS_VPORT_CMD_NEW,
2422           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2423           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2424           .doit = ovs_vport_cmd_new
2425         },
2426         { .cmd = OVS_VPORT_CMD_DEL,
2427           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2428           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2429           .doit = ovs_vport_cmd_del
2430         },
2431         { .cmd = OVS_VPORT_CMD_GET,
2432           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2433           .flags = 0,               /* OK for unprivileged users. */
2434           .doit = ovs_vport_cmd_get,
2435           .dumpit = ovs_vport_cmd_dump
2436         },
2437         { .cmd = OVS_VPORT_CMD_SET,
2438           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2439           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2440           .doit = ovs_vport_cmd_set,
2441         },
2442 };
2443
2444 struct genl_family dp_vport_genl_family __ro_after_init = {
2445         .hdrsize = sizeof(struct ovs_header),
2446         .name = OVS_VPORT_FAMILY,
2447         .version = OVS_VPORT_VERSION,
2448         .maxattr = OVS_VPORT_ATTR_MAX,
2449         .policy = vport_policy,
2450         .netnsok = true,
2451         .parallel_ops = true,
2452         .small_ops = dp_vport_genl_ops,
2453         .n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
2454         .mcgrps = &ovs_dp_vport_multicast_group,
2455         .n_mcgrps = 1,
2456         .module = THIS_MODULE,
2457 };
2458
2459 static struct genl_family * const dp_genl_families[] = {
2460         &dp_datapath_genl_family,
2461         &dp_vport_genl_family,
2462         &dp_flow_genl_family,
2463         &dp_packet_genl_family,
2464         &dp_meter_genl_family,
2465 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2466         &dp_ct_limit_genl_family,
2467 #endif
2468 };
2469
2470 static void dp_unregister_genl(int n_families)
2471 {
2472         int i;
2473
2474         for (i = 0; i < n_families; i++)
2475                 genl_unregister_family(dp_genl_families[i]);
2476 }
2477
2478 static int __init dp_register_genl(void)
2479 {
2480         int err;
2481         int i;
2482
2483         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2484
2485                 err = genl_register_family(dp_genl_families[i]);
2486                 if (err)
2487                         goto error;
2488         }
2489
2490         return 0;
2491
2492 error:
2493         dp_unregister_genl(i);
2494         return err;
2495 }
2496
2497 static int __net_init ovs_init_net(struct net *net)
2498 {
2499         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2500         int err;
2501
2502         INIT_LIST_HEAD(&ovs_net->dps);
2503         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2504         INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2505
2506         err = ovs_ct_init(net);
2507         if (err)
2508                 return err;
2509
2510         schedule_delayed_work(&ovs_net->masks_rebalance,
2511                               msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2512         return 0;
2513 }
2514
2515 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2516                                             struct list_head *head)
2517 {
2518         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2519         struct datapath *dp;
2520
2521         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2522                 int i;
2523
2524                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2525                         struct vport *vport;
2526
2527                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2528                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2529                                         continue;
2530
2531                                 if (dev_net(vport->dev) == dnet)
2532                                         list_add(&vport->detach_list, head);
2533                         }
2534                 }
2535         }
2536 }
2537
2538 static void __net_exit ovs_exit_net(struct net *dnet)
2539 {
2540         struct datapath *dp, *dp_next;
2541         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2542         struct vport *vport, *vport_next;
2543         struct net *net;
2544         LIST_HEAD(head);
2545
2546         ovs_lock();
2547
2548         ovs_ct_exit(dnet);
2549
2550         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2551                 __dp_destroy(dp);
2552
2553         down_read(&net_rwsem);
2554         for_each_net(net)
2555                 list_vports_from_net(net, dnet, &head);
2556         up_read(&net_rwsem);
2557
2558         /* Detach all vports from given namespace. */
2559         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2560                 list_del(&vport->detach_list);
2561                 ovs_dp_detach_port(vport);
2562         }
2563
2564         ovs_unlock();
2565
2566         cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2567         cancel_work_sync(&ovs_net->dp_notify_work);
2568 }
2569
2570 static struct pernet_operations ovs_net_ops = {
2571         .init = ovs_init_net,
2572         .exit = ovs_exit_net,
2573         .id   = &ovs_net_id,
2574         .size = sizeof(struct ovs_net),
2575 };
2576
2577 static int __init dp_init(void)
2578 {
2579         int err;
2580
2581         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2582                      sizeof_field(struct sk_buff, cb));
2583
2584         pr_info("Open vSwitch switching datapath\n");
2585
2586         err = action_fifos_init();
2587         if (err)
2588                 goto error;
2589
2590         err = ovs_internal_dev_rtnl_link_register();
2591         if (err)
2592                 goto error_action_fifos_exit;
2593
2594         err = ovs_flow_init();
2595         if (err)
2596                 goto error_unreg_rtnl_link;
2597
2598         err = ovs_vport_init();
2599         if (err)
2600                 goto error_flow_exit;
2601
2602         err = register_pernet_device(&ovs_net_ops);
2603         if (err)
2604                 goto error_vport_exit;
2605
2606         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2607         if (err)
2608                 goto error_netns_exit;
2609
2610         err = ovs_netdev_init();
2611         if (err)
2612                 goto error_unreg_notifier;
2613
2614         err = dp_register_genl();
2615         if (err < 0)
2616                 goto error_unreg_netdev;
2617
2618         return 0;
2619
2620 error_unreg_netdev:
2621         ovs_netdev_exit();
2622 error_unreg_notifier:
2623         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2624 error_netns_exit:
2625         unregister_pernet_device(&ovs_net_ops);
2626 error_vport_exit:
2627         ovs_vport_exit();
2628 error_flow_exit:
2629         ovs_flow_exit();
2630 error_unreg_rtnl_link:
2631         ovs_internal_dev_rtnl_link_unregister();
2632 error_action_fifos_exit:
2633         action_fifos_exit();
2634 error:
2635         return err;
2636 }
2637
2638 static void dp_cleanup(void)
2639 {
2640         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2641         ovs_netdev_exit();
2642         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2643         unregister_pernet_device(&ovs_net_ops);
2644         rcu_barrier();
2645         ovs_vport_exit();
2646         ovs_flow_exit();
2647         ovs_internal_dev_rtnl_link_unregister();
2648         action_fifos_exit();
2649 }
2650
2651 module_init(dp_init);
2652 module_exit(dp_cleanup);
2653
2654 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2655 MODULE_LICENSE("GPL");
2656 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2657 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2658 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2659 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2660 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2661 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);