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_actions *acts;
 937         struct sw_flow_match match;
 938         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 939         int error;
 940         bool log = !a[OVS_FLOW_ATTR_PROBE];
 941
 942         /* Must have key and actions. */
 943         error = -EINVAL;
 944         if (!a[OVS_FLOW_ATTR_KEY]) {
 945                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 946                 goto error;
 947         }
 948         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 949                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 950                 goto error;
 951         }
 952
 953         /* Most of the time we need to allocate a new flow, do it before
 954          * locking.
 955          */
 956         new_flow = ovs_flow_alloc();
 957         if (IS_ERR(new_flow)) {
 958                 error = PTR_ERR(new_flow);
 959                 goto error;
 960         }
 961
 962         /* Extract key. */
 963         ovs_match_init(&match, &new_flow->key, false, &mask);
 964         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 965                                   a[OVS_FLOW_ATTR_MASK], log);
 966         if (error)
 967                 goto err_kfree_flow;
 968
 969         /* Extract flow identifier. */
 970         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 971                                        &new_flow->key, log);
 972         if (error)
 973                 goto err_kfree_flow;
 974
 975         /* unmasked key is needed to match when ufid is not used. */
 976         if (ovs_identifier_is_key(&new_flow->id))
 977                 match.key = new_flow->id.unmasked_key;
 978
 979         ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
 980
 981         /* Validate actions. */
 982         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 983                                      &new_flow->key, &acts, log);
 984         if (error) {
 985                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 986                 goto err_kfree_flow;
 987         }
 988
 989         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 990                                         ufid_flags);
 991         if (IS_ERR(reply)) {
 992                 error = PTR_ERR(reply);
 993                 goto err_kfree_acts;
 994         }
 995
 996         ovs_lock();
 997         dp = get_dp(net, ovs_header->dp_ifindex);
 998         if (unlikely(!dp)) {
 999                 error = -ENODEV;
1000                 goto err_unlock_ovs;
1001         }
1002
1003         /* Check if this is a duplicate flow */
1004         if (ovs_identifier_is_ufid(&new_flow->id))
1005                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1006         if (!flow)
1007                 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
1008         if (likely(!flow)) {
1009                 rcu_assign_pointer(new_flow->sf_acts, acts);
1010
1011                 /* Put flow in bucket. */
1012                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1013                 if (unlikely(error)) {
1014                         acts = NULL;
1015                         goto err_unlock_ovs;
1016                 }
1017
1018                 if (unlikely(reply)) {
1019                         error = ovs_flow_cmd_fill_info(new_flow,
1020                                                        ovs_header->dp_ifindex,
1021                                                        reply, info->snd_portid,
1022                                                        info->snd_seq, 0,
1023                                                        OVS_FLOW_CMD_NEW,
1024                                                        ufid_flags);
1025                         BUG_ON(error < 0);
1026                 }
1027                 ovs_unlock();
1028         } else {
1029                 struct sw_flow_actions *old_acts;
1030
1031                 /* Bail out if we're not allowed to modify an existing flow.
1032                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1033                  * because Generic Netlink treats the latter as a dump
1034                  * request.  We also accept NLM_F_EXCL in case that bug ever
1035                  * gets fixed.
1036                  */
1037                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1038                                                          | NLM_F_EXCL))) {
1039                         error = -EEXIST;
1040                         goto err_unlock_ovs;
1041                 }
1042                 /* The flow identifier has to be the same for flow updates.
1043                  * Look for any overlapping flow.
1044                  */
1045                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1046                         if (ovs_identifier_is_key(&flow->id))
1047                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1048                                                                  &match);
1049                         else /* UFID matches but key is different */
1050                                 flow = NULL;
1051                         if (!flow) {
1052                                 error = -ENOENT;
1053                                 goto err_unlock_ovs;
1054                         }
1055                 }
1056                 /* Update actions. */
1057                 old_acts = ovsl_dereference(flow->sf_acts);
1058                 rcu_assign_pointer(flow->sf_acts, acts);
1059
1060                 if (unlikely(reply)) {
1061                         error = ovs_flow_cmd_fill_info(flow,
1062                                                        ovs_header->dp_ifindex,
1063                                                        reply, info->snd_portid,
1064                                                        info->snd_seq, 0,
1065                                                        OVS_FLOW_CMD_NEW,
1066                                                        ufid_flags);
1067                         BUG_ON(error < 0);
1068                 }
1069                 ovs_unlock();
1070
1071                 ovs_nla_free_flow_actions_rcu(old_acts);
1072                 ovs_flow_free(new_flow, false);
1073         }
1074
1075         if (reply)
1076                 ovs_notify(&dp_flow_genl_family, reply, info);
1077         return 0;
1078
1079 err_unlock_ovs:
1080         ovs_unlock();
1081         kfree_skb(reply);
1082 err_kfree_acts:
1083         ovs_nla_free_flow_actions(acts);
1084 err_kfree_flow:
1085         ovs_flow_free(new_flow, false);
1086 error:
1087         return error;
1088 }
1089
1090 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1091 static noinline_for_stack
1092 struct sw_flow_actions *get_flow_actions(struct net *net,
1093                                          const struct nlattr *a,
1094                                          const struct sw_flow_key *key,
1095                                          const struct sw_flow_mask *mask,
1096                                          bool log)
1097 {
1098         struct sw_flow_actions *acts;
1099         struct sw_flow_key masked_key;
1100         int error;
1101
1102         ovs_flow_mask_key(&masked_key, key, true, mask);
1103         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1104         if (error) {
1105                 OVS_NLERR(log,
1106                           "Actions may not be safe on all matching packets");
1107                 return ERR_PTR(error);
1108         }
1109
1110         return acts;
1111 }
1112
1113 /* Factor out match-init and action-copy to avoid
1114  * "Wframe-larger-than=1024" warning. Because mask is only
1115  * used to get actions, we new a function to save some
1116  * stack space.
1117  *
1118  * If there are not key and action attrs, we return 0
1119  * directly. In the case, the caller will also not use the
1120  * match as before. If there is action attr, we try to get
1121  * actions and save them to *acts. Before returning from
1122  * the function, we reset the match->mask pointer. Because
1123  * we should not to return match object with dangling reference
1124  * to mask.
1125  * */
1126 static noinline_for_stack int
1127 ovs_nla_init_match_and_action(struct net *net,
1128                               struct sw_flow_match *match,
1129                               struct sw_flow_key *key,
1130                               struct nlattr **a,
1131                               struct sw_flow_actions **acts,
1132                               bool log)
1133 {
1134         struct sw_flow_mask mask;
1135         int error = 0;
1136
1137         if (a[OVS_FLOW_ATTR_KEY]) {
1138                 ovs_match_init(match, key, true, &mask);
1139                 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1140                                           a[OVS_FLOW_ATTR_MASK], log);
1141                 if (error)
1142                         goto error;
1143         }
1144
1145         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1146                 if (!a[OVS_FLOW_ATTR_KEY]) {
1147                         OVS_NLERR(log,
1148                                   "Flow key attribute not present in set flow.");
1149                         error = -EINVAL;
1150                         goto error;
1151                 }
1152
1153                 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1154                                          &mask, log);
1155                 if (IS_ERR(*acts)) {
1156                         error = PTR_ERR(*acts);
1157                         goto error;
1158                 }
1159         }
1160
1161         /* On success, error is 0. */
1162 error:
1163         match->mask = NULL;
1164         return error;
1165 }
1166
1167 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1168 {
1169         struct net *net = sock_net(skb->sk);
1170         struct nlattr **a = info->attrs;
1171         struct ovs_header *ovs_header = info->userhdr;
1172         struct sw_flow_key key;
1173         struct sw_flow *flow;
1174         struct sk_buff *reply = NULL;
1175         struct datapath *dp;
1176         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1177         struct sw_flow_match match;
1178         struct sw_flow_id sfid;
1179         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1180         int error = 0;
1181         bool log = !a[OVS_FLOW_ATTR_PROBE];
1182         bool ufid_present;
1183
1184         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1185         if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1186                 OVS_NLERR(log,
1187                           "Flow set message rejected, Key attribute missing.");
1188                 return -EINVAL;
1189         }
1190
1191         error = ovs_nla_init_match_and_action(net, &match, &key, a,
1192                                               &acts, log);
1193         if (error)
1194                 goto error;
1195
1196         if (acts) {
1197                 /* Can allocate before locking if have acts. */
1198                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1199                                                 ufid_flags);
1200                 if (IS_ERR(reply)) {
1201                         error = PTR_ERR(reply);
1202                         goto err_kfree_acts;
1203                 }
1204         }
1205
1206         ovs_lock();
1207         dp = get_dp(net, ovs_header->dp_ifindex);
1208         if (unlikely(!dp)) {
1209                 error = -ENODEV;
1210                 goto err_unlock_ovs;
1211         }
1212         /* Check that the flow exists. */
1213         if (ufid_present)
1214                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1215         else
1216                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1217         if (unlikely(!flow)) {
1218                 error = -ENOENT;
1219                 goto err_unlock_ovs;
1220         }
1221
1222         /* Update actions, if present. */
1223         if (likely(acts)) {
1224                 old_acts = ovsl_dereference(flow->sf_acts);
1225                 rcu_assign_pointer(flow->sf_acts, acts);
1226
1227                 if (unlikely(reply)) {
1228                         error = ovs_flow_cmd_fill_info(flow,
1229                                                        ovs_header->dp_ifindex,
1230                                                        reply, info->snd_portid,
1231                                                        info->snd_seq, 0,
1232                                                        OVS_FLOW_CMD_SET,
1233                                                        ufid_flags);
1234                         BUG_ON(error < 0);
1235                 }
1236         } else {
1237                 /* Could not alloc without acts before locking. */
1238                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1239                                                 info, OVS_FLOW_CMD_SET, false,
1240                                                 ufid_flags);
1241
1242                 if (IS_ERR(reply)) {
1243                         error = PTR_ERR(reply);
1244                         goto err_unlock_ovs;
1245                 }
1246         }
1247
1248         /* Clear stats. */
1249         if (a[OVS_FLOW_ATTR_CLEAR])
1250                 ovs_flow_stats_clear(flow);
1251         ovs_unlock();
1252
1253         if (reply)
1254                 ovs_notify(&dp_flow_genl_family, reply, info);
1255         if (old_acts)
1256                 ovs_nla_free_flow_actions_rcu(old_acts);
1257
1258         return 0;
1259
1260 err_unlock_ovs:
1261         ovs_unlock();
1262         kfree_skb(reply);
1263 err_kfree_acts:
1264         ovs_nla_free_flow_actions(acts);
1265 error:
1266         return error;
1267 }
1268
1269 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1270 {
1271         struct nlattr **a = info->attrs;
1272         struct ovs_header *ovs_header = info->userhdr;
1273         struct net *net = sock_net(skb->sk);
1274         struct sw_flow_key key;
1275         struct sk_buff *reply;
1276         struct sw_flow *flow;
1277         struct datapath *dp;
1278         struct sw_flow_match match;
1279         struct sw_flow_id ufid;
1280         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1281         int err = 0;
1282         bool log = !a[OVS_FLOW_ATTR_PROBE];
1283         bool ufid_present;
1284
1285         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1286         if (a[OVS_FLOW_ATTR_KEY]) {
1287                 ovs_match_init(&match, &key, true, NULL);
1288                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1289                                         log);
1290         } else if (!ufid_present) {
1291                 OVS_NLERR(log,
1292                           "Flow get message rejected, Key attribute missing.");
1293                 err = -EINVAL;
1294         }
1295         if (err)
1296                 return err;
1297
1298         ovs_lock();
1299         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1300         if (!dp) {
1301                 err = -ENODEV;
1302                 goto unlock;
1303         }
1304
1305         if (ufid_present)
1306                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1307         else
1308                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1309         if (!flow) {
1310                 err = -ENOENT;
1311                 goto unlock;
1312         }
1313
1314         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1315                                         OVS_FLOW_CMD_GET, true, ufid_flags);
1316         if (IS_ERR(reply)) {
1317                 err = PTR_ERR(reply);
1318                 goto unlock;
1319         }
1320
1321         ovs_unlock();
1322         return genlmsg_reply(reply, info);
1323 unlock:
1324         ovs_unlock();
1325         return err;
1326 }
1327
1328 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1329 {
1330         struct nlattr **a = info->attrs;
1331         struct ovs_header *ovs_header = info->userhdr;
1332         struct net *net = sock_net(skb->sk);
1333         struct sw_flow_key key;
1334         struct sk_buff *reply;
1335         struct sw_flow *flow = NULL;
1336         struct datapath *dp;
1337         struct sw_flow_match match;
1338         struct sw_flow_id ufid;
1339         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1340         int err;
1341         bool log = !a[OVS_FLOW_ATTR_PROBE];
1342         bool ufid_present;
1343
1344         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1345         if (a[OVS_FLOW_ATTR_KEY]) {
1346                 ovs_match_init(&match, &key, true, NULL);
1347                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1348                                         NULL, log);
1349                 if (unlikely(err))
1350                         return err;
1351         }
1352
1353         ovs_lock();
1354         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1355         if (unlikely(!dp)) {
1356                 err = -ENODEV;
1357                 goto unlock;
1358         }
1359
1360         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1361                 err = ovs_flow_tbl_flush(&dp->table);
1362                 goto unlock;
1363         }
1364
1365         if (ufid_present)
1366                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1367         else
1368                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1369         if (unlikely(!flow)) {
1370                 err = -ENOENT;
1371                 goto unlock;
1372         }
1373
1374         ovs_flow_tbl_remove(&dp->table, flow);
1375         ovs_unlock();
1376
1377         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1378                                         &flow->id, info, false, ufid_flags);
1379         if (likely(reply)) {
1380                 if (!IS_ERR(reply)) {
1381                         rcu_read_lock();        /*To keep RCU checker happy. */
1382                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1383                                                      reply, info->snd_portid,
1384                                                      info->snd_seq, 0,
1385                                                      OVS_FLOW_CMD_DEL,
1386                                                      ufid_flags);
1387                         rcu_read_unlock();
1388                         if (WARN_ON_ONCE(err < 0)) {
1389                                 kfree_skb(reply);
1390                                 goto out_free;
1391                         }
1392
1393                         ovs_notify(&dp_flow_genl_family, reply, info);
1394                 } else {
1395                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1396                                         PTR_ERR(reply));
1397                 }
1398         }
1399
1400 out_free:
1401         ovs_flow_free(flow, true);
1402         return 0;
1403 unlock:
1404         ovs_unlock();
1405         return err;
1406 }
1407
1408 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1409 {
1410         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1411         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1412         struct table_instance *ti;
1413         struct datapath *dp;
1414         u32 ufid_flags;
1415         int err;
1416
1417         err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1418                                        OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1419         if (err)
1420                 return err;
1421         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1422
1423         rcu_read_lock();
1424         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1425         if (!dp) {
1426                 rcu_read_unlock();
1427                 return -ENODEV;
1428         }
1429
1430         ti = rcu_dereference(dp->table.ti);
1431         for (;;) {
1432                 struct sw_flow *flow;
1433                 u32 bucket, obj;
1434
1435                 bucket = cb->args[0];
1436                 obj = cb->args[1];
1437                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1438                 if (!flow)
1439                         break;
1440
1441                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1442                                            NETLINK_CB(cb->skb).portid,
1443                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1444                                            OVS_FLOW_CMD_GET, ufid_flags) < 0)
1445                         break;
1446
1447                 cb->args[0] = bucket;
1448                 cb->args[1] = obj;
1449         }
1450         rcu_read_unlock();
1451         return skb->len;
1452 }
1453
1454 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1455         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1456         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1457         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1458         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1459         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1460         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1461         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1462 };
1463
1464 static const struct genl_small_ops dp_flow_genl_ops[] = {
1465         { .cmd = OVS_FLOW_CMD_NEW,
1466           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1467           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1468           .doit = ovs_flow_cmd_new
1469         },
1470         { .cmd = OVS_FLOW_CMD_DEL,
1471           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1472           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1473           .doit = ovs_flow_cmd_del
1474         },
1475         { .cmd = OVS_FLOW_CMD_GET,
1476           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1477           .flags = 0,               /* OK for unprivileged users. */
1478           .doit = ovs_flow_cmd_get,
1479           .dumpit = ovs_flow_cmd_dump
1480         },
1481         { .cmd = OVS_FLOW_CMD_SET,
1482           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1483           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1484           .doit = ovs_flow_cmd_set,
1485         },
1486 };
1487
1488 static struct genl_family dp_flow_genl_family __ro_after_init = {
1489         .hdrsize = sizeof(struct ovs_header),
1490         .name = OVS_FLOW_FAMILY,
1491         .version = OVS_FLOW_VERSION,
1492         .maxattr = OVS_FLOW_ATTR_MAX,
1493         .policy = flow_policy,
1494         .netnsok = true,
1495         .parallel_ops = true,
1496         .small_ops = dp_flow_genl_ops,
1497         .n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
1498         .mcgrps = &ovs_dp_flow_multicast_group,
1499         .n_mcgrps = 1,
1500         .module = THIS_MODULE,
1501 };
1502
1503 static size_t ovs_dp_cmd_msg_size(void)
1504 {
1505         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1506
1507         msgsize += nla_total_size(IFNAMSIZ);
1508         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1509         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1510         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1511         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1512
1513         return msgsize;
1514 }
1515
1516 /* Called with ovs_mutex. */
1517 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1518                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1519 {
1520         struct ovs_header *ovs_header;
1521         struct ovs_dp_stats dp_stats;
1522         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1523         int err;
1524
1525         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1526                                  flags, cmd);
1527         if (!ovs_header)
1528                 goto error;
1529
1530         ovs_header->dp_ifindex = get_dpifindex(dp);
1531
1532         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1533         if (err)
1534                 goto nla_put_failure;
1535
1536         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1537         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1538                           &dp_stats, OVS_DP_ATTR_PAD))
1539                 goto nla_put_failure;
1540
1541         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1542                           sizeof(struct ovs_dp_megaflow_stats),
1543                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1544                 goto nla_put_failure;
1545
1546         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1547                 goto nla_put_failure;
1548
1549         if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1550                         ovs_flow_tbl_masks_cache_size(&dp->table)))
1551                 goto nla_put_failure;
1552
1553         genlmsg_end(skb, ovs_header);
1554         return 0;
1555
1556 nla_put_failure:
1557         genlmsg_cancel(skb, ovs_header);
1558 error:
1559         return -EMSGSIZE;
1560 }
1561
1562 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1563 {
1564         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1565 }
1566
1567 /* Called with rcu_read_lock or ovs_mutex. */
1568 static struct datapath *lookup_datapath(struct net *net,
1569                                         const struct ovs_header *ovs_header,
1570                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1571 {
1572         struct datapath *dp;
1573
1574         if (!a[OVS_DP_ATTR_NAME])
1575                 dp = get_dp(net, ovs_header->dp_ifindex);
1576         else {
1577                 struct vport *vport;
1578
1579                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1580                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1581         }
1582         return dp ? dp : ERR_PTR(-ENODEV);
1583 }
1584
1585 static void ovs_dp_reset_user_features(struct sk_buff *skb,
1586                                        struct genl_info *info)
1587 {
1588         struct datapath *dp;
1589
1590         dp = lookup_datapath(sock_net(skb->sk), info->userhdr,
1591                              info->attrs);
1592         if (IS_ERR(dp))
1593                 return;
1594
1595         pr_warn("%s: Dropping previously announced user features\n",
1596                 ovs_dp_name(dp));
1597         dp->user_features = 0;
1598 }
1599
1600 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1601
1602 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1603 {
1604         u32 user_features = 0;
1605
1606         if (a[OVS_DP_ATTR_USER_FEATURES]) {
1607                 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1608
1609                 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1610                                       OVS_DP_F_UNALIGNED |
1611                                       OVS_DP_F_TC_RECIRC_SHARING))
1612                         return -EOPNOTSUPP;
1613
1614 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1615                 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1616                         return -EOPNOTSUPP;
1617 #endif
1618         }
1619
1620         if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1621                 int err;
1622                 u32 cache_size;
1623
1624                 cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1625                 err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1626                 if (err)
1627                         return err;
1628         }
1629
1630         dp->user_features = user_features;
1631
1632         if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1633                 static_branch_enable(&tc_recirc_sharing_support);
1634         else
1635                 static_branch_disable(&tc_recirc_sharing_support);
1636
1637         return 0;
1638 }
1639
1640 static int ovs_dp_stats_init(struct datapath *dp)
1641 {
1642         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1643         if (!dp->stats_percpu)
1644                 return -ENOMEM;
1645
1646         return 0;
1647 }
1648
1649 static int ovs_dp_vport_init(struct datapath *dp)
1650 {
1651         int i;
1652
1653         dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1654                                   sizeof(struct hlist_head),
1655                                   GFP_KERNEL);
1656         if (!dp->ports)
1657                 return -ENOMEM;
1658
1659         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1660                 INIT_HLIST_HEAD(&dp->ports[i]);
1661
1662         return 0;
1663 }
1664
1665 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1666 {
1667         struct nlattr **a = info->attrs;
1668         struct vport_parms parms;
1669         struct sk_buff *reply;
1670         struct datapath *dp;
1671         struct vport *vport;
1672         struct ovs_net *ovs_net;
1673         int err;
1674
1675         err = -EINVAL;
1676         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1677                 goto err;
1678
1679         reply = ovs_dp_cmd_alloc_info();
1680         if (!reply)
1681                 return -ENOMEM;
1682
1683         err = -ENOMEM;
1684         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1685         if (dp == NULL)
1686                 goto err_destroy_reply;
1687
1688         ovs_dp_set_net(dp, sock_net(skb->sk));
1689
1690         /* Allocate table. */
1691         err = ovs_flow_tbl_init(&dp->table);
1692         if (err)
1693                 goto err_destroy_dp;
1694
1695         err = ovs_dp_stats_init(dp);
1696         if (err)
1697                 goto err_destroy_table;
1698
1699         err = ovs_dp_vport_init(dp);
1700         if (err)
1701                 goto err_destroy_stats;
1702
1703         err = ovs_meters_init(dp);
1704         if (err)
1705                 goto err_destroy_ports;
1706
1707         /* Set up our datapath device. */
1708         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1709         parms.type = OVS_VPORT_TYPE_INTERNAL;
1710         parms.options = NULL;
1711         parms.dp = dp;
1712         parms.port_no = OVSP_LOCAL;
1713         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1714
1715         /* So far only local changes have been made, now need the lock. */
1716         ovs_lock();
1717
1718         err = ovs_dp_change(dp, a);
1719         if (err)
1720                 goto err_unlock_and_destroy_meters;
1721
1722         vport = new_vport(&parms);
1723         if (IS_ERR(vport)) {
1724                 err = PTR_ERR(vport);
1725                 if (err == -EBUSY)
1726                         err = -EEXIST;
1727
1728                 if (err == -EEXIST) {
1729                         /* An outdated user space instance that does not understand
1730                          * the concept of user_features has attempted to create a new
1731                          * datapath and is likely to reuse it. Drop all user features.
1732                          */
1733                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1734                                 ovs_dp_reset_user_features(skb, info);
1735                 }
1736
1737                 goto err_unlock_and_destroy_meters;
1738         }
1739
1740         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1741                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1742         BUG_ON(err < 0);
1743
1744         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1745         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1746
1747         ovs_unlock();
1748
1749         ovs_notify(&dp_datapath_genl_family, reply, info);
1750         return 0;
1751
1752 err_unlock_and_destroy_meters:
1753         ovs_unlock();
1754         ovs_meters_exit(dp);
1755 err_destroy_ports:
1756         kfree(dp->ports);
1757 err_destroy_stats:
1758         free_percpu(dp->stats_percpu);
1759 err_destroy_table:
1760         ovs_flow_tbl_destroy(&dp->table);
1761 err_destroy_dp:
1762         kfree(dp);
1763 err_destroy_reply:
1764         kfree_skb(reply);
1765 err:
1766         return err;
1767 }
1768
1769 /* Called with ovs_mutex. */
1770 static void __dp_destroy(struct datapath *dp)
1771 {
1772         struct flow_table *table = &dp->table;
1773         int i;
1774
1775         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1776                 struct vport *vport;
1777                 struct hlist_node *n;
1778
1779                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1780                         if (vport->port_no != OVSP_LOCAL)
1781                                 ovs_dp_detach_port(vport);
1782         }
1783
1784         list_del_rcu(&dp->list_node);
1785
1786         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1787          * all ports in datapath are destroyed first before freeing datapath.
1788          */
1789         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1790
1791         /* Flush sw_flow in the tables. RCU cb only releases resource
1792          * such as dp, ports and tables. That may avoid some issues
1793          * such as RCU usage warning.
1794          */
1795         table_instance_flow_flush(table, ovsl_dereference(table->ti),
1796                                   ovsl_dereference(table->ufid_ti));
1797
1798         /* RCU destroy the ports, meters and flow tables. */
1799         call_rcu(&dp->rcu, destroy_dp_rcu);
1800 }
1801
1802 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1803 {
1804         struct sk_buff *reply;
1805         struct datapath *dp;
1806         int err;
1807
1808         reply = ovs_dp_cmd_alloc_info();
1809         if (!reply)
1810                 return -ENOMEM;
1811
1812         ovs_lock();
1813         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1814         err = PTR_ERR(dp);
1815         if (IS_ERR(dp))
1816                 goto err_unlock_free;
1817
1818         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1819                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1820         BUG_ON(err < 0);
1821
1822         __dp_destroy(dp);
1823         ovs_unlock();
1824
1825         ovs_notify(&dp_datapath_genl_family, reply, info);
1826
1827         return 0;
1828
1829 err_unlock_free:
1830         ovs_unlock();
1831         kfree_skb(reply);
1832         return err;
1833 }
1834
1835 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1836 {
1837         struct sk_buff *reply;
1838         struct datapath *dp;
1839         int err;
1840
1841         reply = ovs_dp_cmd_alloc_info();
1842         if (!reply)
1843                 return -ENOMEM;
1844
1845         ovs_lock();
1846         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1847         err = PTR_ERR(dp);
1848         if (IS_ERR(dp))
1849                 goto err_unlock_free;
1850
1851         err = ovs_dp_change(dp, info->attrs);
1852         if (err)
1853                 goto err_unlock_free;
1854
1855         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1856                                    info->snd_seq, 0, OVS_DP_CMD_SET);
1857         BUG_ON(err < 0);
1858
1859         ovs_unlock();
1860         ovs_notify(&dp_datapath_genl_family, reply, info);
1861
1862         return 0;
1863
1864 err_unlock_free:
1865         ovs_unlock();
1866         kfree_skb(reply);
1867         return err;
1868 }
1869
1870 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1871 {
1872         struct sk_buff *reply;
1873         struct datapath *dp;
1874         int err;
1875
1876         reply = ovs_dp_cmd_alloc_info();
1877         if (!reply)
1878                 return -ENOMEM;
1879
1880         ovs_lock();
1881         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1882         if (IS_ERR(dp)) {
1883                 err = PTR_ERR(dp);
1884                 goto err_unlock_free;
1885         }
1886         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1887                                    info->snd_seq, 0, OVS_DP_CMD_GET);
1888         BUG_ON(err < 0);
1889         ovs_unlock();
1890
1891         return genlmsg_reply(reply, info);
1892
1893 err_unlock_free:
1894         ovs_unlock();
1895         kfree_skb(reply);
1896         return err;
1897 }
1898
1899 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1900 {
1901         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1902         struct datapath *dp;
1903         int skip = cb->args[0];
1904         int i = 0;
1905
1906         ovs_lock();
1907         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1908                 if (i >= skip &&
1909                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1910                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1911                                          OVS_DP_CMD_GET) < 0)
1912                         break;
1913                 i++;
1914         }
1915         ovs_unlock();
1916
1917         cb->args[0] = i;
1918
1919         return skb->len;
1920 }
1921
1922 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1923         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1924         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1925         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1926         [OVS_DP_ATTR_MASKS_CACHE_SIZE] =  NLA_POLICY_RANGE(NLA_U32, 0,
1927                 PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
1928 };
1929
1930 static const struct genl_small_ops dp_datapath_genl_ops[] = {
1931         { .cmd = OVS_DP_CMD_NEW,
1932           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1933           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1934           .doit = ovs_dp_cmd_new
1935         },
1936         { .cmd = OVS_DP_CMD_DEL,
1937           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1938           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1939           .doit = ovs_dp_cmd_del
1940         },
1941         { .cmd = OVS_DP_CMD_GET,
1942           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1943           .flags = 0,               /* OK for unprivileged users. */
1944           .doit = ovs_dp_cmd_get,
1945           .dumpit = ovs_dp_cmd_dump
1946         },
1947         { .cmd = OVS_DP_CMD_SET,
1948           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1949           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1950           .doit = ovs_dp_cmd_set,
1951         },
1952 };
1953
1954 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1955         .hdrsize = sizeof(struct ovs_header),
1956         .name = OVS_DATAPATH_FAMILY,
1957         .version = OVS_DATAPATH_VERSION,
1958         .maxattr = OVS_DP_ATTR_MAX,
1959         .policy = datapath_policy,
1960         .netnsok = true,
1961         .parallel_ops = true,
1962         .small_ops = dp_datapath_genl_ops,
1963         .n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1964         .mcgrps = &ovs_dp_datapath_multicast_group,
1965         .n_mcgrps = 1,
1966         .module = THIS_MODULE,
1967 };
1968
1969 /* Called with ovs_mutex or RCU read lock. */
1970 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1971                                    struct net *net, u32 portid, u32 seq,
1972                                    u32 flags, u8 cmd, gfp_t gfp)
1973 {
1974         struct ovs_header *ovs_header;
1975         struct ovs_vport_stats vport_stats;
1976         int err;
1977
1978         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1979                                  flags, cmd);
1980         if (!ovs_header)
1981                 return -EMSGSIZE;
1982
1983         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1984
1985         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1986             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1987             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1988                            ovs_vport_name(vport)) ||
1989             nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1990                 goto nla_put_failure;
1991
1992         if (!net_eq(net, dev_net(vport->dev))) {
1993                 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
1994
1995                 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1996                         goto nla_put_failure;
1997         }
1998
1999         ovs_vport_get_stats(vport, &vport_stats);
2000         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
2001                           sizeof(struct ovs_vport_stats), &vport_stats,
2002                           OVS_VPORT_ATTR_PAD))
2003                 goto nla_put_failure;
2004
2005         if (ovs_vport_get_upcall_portids(vport, skb))
2006                 goto nla_put_failure;
2007
2008         err = ovs_vport_get_options(vport, skb);
2009         if (err == -EMSGSIZE)
2010                 goto error;
2011
2012         genlmsg_end(skb, ovs_header);
2013         return 0;
2014
2015 nla_put_failure:
2016         err = -EMSGSIZE;
2017 error:
2018         genlmsg_cancel(skb, ovs_header);
2019         return err;
2020 }
2021
2022 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2023 {
2024         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2025 }
2026
2027 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2028 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2029                                          u32 portid, u32 seq, u8 cmd)
2030 {
2031         struct sk_buff *skb;
2032         int retval;
2033
2034         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2035         if (!skb)
2036                 return ERR_PTR(-ENOMEM);
2037
2038         retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2039                                          GFP_KERNEL);
2040         BUG_ON(retval < 0);
2041
2042         return skb;
2043 }
2044
2045 /* Called with ovs_mutex or RCU read lock. */
2046 static struct vport *lookup_vport(struct net *net,
2047                                   const struct ovs_header *ovs_header,
2048                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2049 {
2050         struct datapath *dp;
2051         struct vport *vport;
2052
2053         if (a[OVS_VPORT_ATTR_IFINDEX])
2054                 return ERR_PTR(-EOPNOTSUPP);
2055         if (a[OVS_VPORT_ATTR_NAME]) {
2056                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2057                 if (!vport)
2058                         return ERR_PTR(-ENODEV);
2059                 if (ovs_header->dp_ifindex &&
2060                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2061                         return ERR_PTR(-ENODEV);
2062                 return vport;
2063         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2064                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2065
2066                 if (port_no >= DP_MAX_PORTS)
2067                         return ERR_PTR(-EFBIG);
2068
2069                 dp = get_dp(net, ovs_header->dp_ifindex);
2070                 if (!dp)
2071                         return ERR_PTR(-ENODEV);
2072
2073                 vport = ovs_vport_ovsl_rcu(dp, port_no);
2074                 if (!vport)
2075                         return ERR_PTR(-ENODEV);
2076                 return vport;
2077         } else
2078                 return ERR_PTR(-EINVAL);
2079
2080 }
2081
2082 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2083 {
2084         unsigned int dev_headroom, max_headroom = 0;
2085         struct net_device *dev;
2086         struct vport *vport;
2087         int i;
2088
2089         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2090                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2091                                          lockdep_ovsl_is_held()) {
2092                         dev = vport->dev;
2093                         dev_headroom = netdev_get_fwd_headroom(dev);
2094                         if (dev_headroom > max_headroom)
2095                                 max_headroom = dev_headroom;
2096                 }
2097         }
2098
2099         return max_headroom;
2100 }
2101
2102 /* Called with ovs_mutex */
2103 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2104 {
2105         struct vport *vport;
2106         int i;
2107
2108         dp->max_headroom = new_headroom;
2109         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2110                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2111                                          lockdep_ovsl_is_held())
2112                         netdev_set_rx_headroom(vport->dev, new_headroom);
2113         }
2114 }
2115
2116 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2117 {
2118         struct nlattr **a = info->attrs;
2119         struct ovs_header *ovs_header = info->userhdr;
2120         struct vport_parms parms;
2121         struct sk_buff *reply;
2122         struct vport *vport;
2123         struct datapath *dp;
2124         unsigned int new_headroom;
2125         u32 port_no;
2126         int err;
2127
2128         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2129             !a[OVS_VPORT_ATTR_UPCALL_PID])
2130                 return -EINVAL;
2131         if (a[OVS_VPORT_ATTR_IFINDEX])
2132                 return -EOPNOTSUPP;
2133
2134         port_no = a[OVS_VPORT_ATTR_PORT_NO]
2135                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2136         if (port_no >= DP_MAX_PORTS)
2137                 return -EFBIG;
2138
2139         reply = ovs_vport_cmd_alloc_info();
2140         if (!reply)
2141                 return -ENOMEM;
2142
2143         ovs_lock();
2144 restart:
2145         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2146         err = -ENODEV;
2147         if (!dp)
2148                 goto exit_unlock_free;
2149
2150         if (port_no) {
2151                 vport = ovs_vport_ovsl(dp, port_no);
2152                 err = -EBUSY;
2153                 if (vport)
2154                         goto exit_unlock_free;
2155         } else {
2156                 for (port_no = 1; ; port_no++) {
2157                         if (port_no >= DP_MAX_PORTS) {
2158                                 err = -EFBIG;
2159                                 goto exit_unlock_free;
2160                         }
2161                         vport = ovs_vport_ovsl(dp, port_no);
2162                         if (!vport)
2163                                 break;
2164                 }
2165         }
2166
2167         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2168         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2169         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2170         parms.dp = dp;
2171         parms.port_no = port_no;
2172         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2173
2174         vport = new_vport(&parms);
2175         err = PTR_ERR(vport);
2176         if (IS_ERR(vport)) {
2177                 if (err == -EAGAIN)
2178                         goto restart;
2179                 goto exit_unlock_free;
2180         }
2181
2182         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2183                                       info->snd_portid, info->snd_seq, 0,
2184                                       OVS_VPORT_CMD_NEW, GFP_KERNEL);
2185
2186         new_headroom = netdev_get_fwd_headroom(vport->dev);
2187
2188         if (new_headroom > dp->max_headroom)
2189                 ovs_update_headroom(dp, new_headroom);
2190         else
2191                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2192
2193         BUG_ON(err < 0);
2194         ovs_unlock();
2195
2196         ovs_notify(&dp_vport_genl_family, reply, info);
2197         return 0;
2198
2199 exit_unlock_free:
2200         ovs_unlock();
2201         kfree_skb(reply);
2202         return err;
2203 }
2204
2205 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2206 {
2207         struct nlattr **a = info->attrs;
2208         struct sk_buff *reply;
2209         struct vport *vport;
2210         int err;
2211
2212         reply = ovs_vport_cmd_alloc_info();
2213         if (!reply)
2214                 return -ENOMEM;
2215
2216         ovs_lock();
2217         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2218         err = PTR_ERR(vport);
2219         if (IS_ERR(vport))
2220                 goto exit_unlock_free;
2221
2222         if (a[OVS_VPORT_ATTR_TYPE] &&
2223             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2224                 err = -EINVAL;
2225                 goto exit_unlock_free;
2226         }
2227
2228         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2229                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2230                 if (err)
2231                         goto exit_unlock_free;
2232         }
2233
2234
2235         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2236                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2237
2238                 err = ovs_vport_set_upcall_portids(vport, ids);
2239                 if (err)
2240                         goto exit_unlock_free;
2241         }
2242
2243         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2244                                       info->snd_portid, info->snd_seq, 0,
2245                                       OVS_VPORT_CMD_SET, GFP_KERNEL);
2246         BUG_ON(err < 0);
2247
2248         ovs_unlock();
2249         ovs_notify(&dp_vport_genl_family, reply, info);
2250         return 0;
2251
2252 exit_unlock_free:
2253         ovs_unlock();
2254         kfree_skb(reply);
2255         return err;
2256 }
2257
2258 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2259 {
2260         bool update_headroom = false;
2261         struct nlattr **a = info->attrs;
2262         struct sk_buff *reply;
2263         struct datapath *dp;
2264         struct vport *vport;
2265         unsigned int new_headroom;
2266         int err;
2267
2268         reply = ovs_vport_cmd_alloc_info();
2269         if (!reply)
2270                 return -ENOMEM;
2271
2272         ovs_lock();
2273         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2274         err = PTR_ERR(vport);
2275         if (IS_ERR(vport))
2276                 goto exit_unlock_free;
2277
2278         if (vport->port_no == OVSP_LOCAL) {
2279                 err = -EINVAL;
2280                 goto exit_unlock_free;
2281         }
2282
2283         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2284                                       info->snd_portid, info->snd_seq, 0,
2285                                       OVS_VPORT_CMD_DEL, GFP_KERNEL);
2286         BUG_ON(err < 0);
2287
2288         /* the vport deletion may trigger dp headroom update */
2289         dp = vport->dp;
2290         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2291                 update_headroom = true;
2292
2293         netdev_reset_rx_headroom(vport->dev);
2294         ovs_dp_detach_port(vport);
2295
2296         if (update_headroom) {
2297                 new_headroom = ovs_get_max_headroom(dp);
2298
2299                 if (new_headroom < dp->max_headroom)
2300                         ovs_update_headroom(dp, new_headroom);
2301         }
2302         ovs_unlock();
2303
2304         ovs_notify(&dp_vport_genl_family, reply, info);
2305         return 0;
2306
2307 exit_unlock_free:
2308         ovs_unlock();
2309         kfree_skb(reply);
2310         return err;
2311 }
2312
2313 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2314 {
2315         struct nlattr **a = info->attrs;
2316         struct ovs_header *ovs_header = info->userhdr;
2317         struct sk_buff *reply;
2318         struct vport *vport;
2319         int err;
2320
2321         reply = ovs_vport_cmd_alloc_info();
2322         if (!reply)
2323                 return -ENOMEM;
2324
2325         rcu_read_lock();
2326         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2327         err = PTR_ERR(vport);
2328         if (IS_ERR(vport))
2329                 goto exit_unlock_free;
2330         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2331                                       info->snd_portid, info->snd_seq, 0,
2332                                       OVS_VPORT_CMD_GET, GFP_ATOMIC);
2333         BUG_ON(err < 0);
2334         rcu_read_unlock();
2335
2336         return genlmsg_reply(reply, info);
2337
2338 exit_unlock_free:
2339         rcu_read_unlock();
2340         kfree_skb(reply);
2341         return err;
2342 }
2343
2344 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2345 {
2346         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2347         struct datapath *dp;
2348         int bucket = cb->args[0], skip = cb->args[1];
2349         int i, j = 0;
2350
2351         rcu_read_lock();
2352         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2353         if (!dp) {
2354                 rcu_read_unlock();
2355                 return -ENODEV;
2356         }
2357         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2358                 struct vport *vport;
2359
2360                 j = 0;
2361                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2362                         if (j >= skip &&
2363                             ovs_vport_cmd_fill_info(vport, skb,
2364                                                     sock_net(skb->sk),
2365                                                     NETLINK_CB(cb->skb).portid,
2366                                                     cb->nlh->nlmsg_seq,
2367                                                     NLM_F_MULTI,
2368                                                     OVS_VPORT_CMD_GET,
2369                                                     GFP_ATOMIC) < 0)
2370                                 goto out;
2371
2372                         j++;
2373                 }
2374                 skip = 0;
2375         }
2376 out:
2377         rcu_read_unlock();
2378
2379         cb->args[0] = i;
2380         cb->args[1] = j;
2381
2382         return skb->len;
2383 }
2384
2385 static void ovs_dp_masks_rebalance(struct work_struct *work)
2386 {
2387         struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2388                                                masks_rebalance.work);
2389         struct datapath *dp;
2390
2391         ovs_lock();
2392
2393         list_for_each_entry(dp, &ovs_net->dps, list_node)
2394                 ovs_flow_masks_rebalance(&dp->table);
2395
2396         ovs_unlock();
2397
2398         schedule_delayed_work(&ovs_net->masks_rebalance,
2399                               msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2400 }
2401
2402 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2403         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2404         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2405         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2406         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2407         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2408         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2409         [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2410         [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2411 };
2412
2413 static const struct genl_small_ops dp_vport_genl_ops[] = {
2414         { .cmd = OVS_VPORT_CMD_NEW,
2415           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2416           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2417           .doit = ovs_vport_cmd_new
2418         },
2419         { .cmd = OVS_VPORT_CMD_DEL,
2420           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2421           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2422           .doit = ovs_vport_cmd_del
2423         },
2424         { .cmd = OVS_VPORT_CMD_GET,
2425           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2426           .flags = 0,               /* OK for unprivileged users. */
2427           .doit = ovs_vport_cmd_get,
2428           .dumpit = ovs_vport_cmd_dump
2429         },
2430         { .cmd = OVS_VPORT_CMD_SET,
2431           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2432           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2433           .doit = ovs_vport_cmd_set,
2434         },
2435 };
2436
2437 struct genl_family dp_vport_genl_family __ro_after_init = {
2438         .hdrsize = sizeof(struct ovs_header),
2439         .name = OVS_VPORT_FAMILY,
2440         .version = OVS_VPORT_VERSION,
2441         .maxattr = OVS_VPORT_ATTR_MAX,
2442         .policy = vport_policy,
2443         .netnsok = true,
2444         .parallel_ops = true,
2445         .small_ops = dp_vport_genl_ops,
2446         .n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
2447         .mcgrps = &ovs_dp_vport_multicast_group,
2448         .n_mcgrps = 1,
2449         .module = THIS_MODULE,
2450 };
2451
2452 static struct genl_family * const dp_genl_families[] = {
2453         &dp_datapath_genl_family,
2454         &dp_vport_genl_family,
2455         &dp_flow_genl_family,
2456         &dp_packet_genl_family,
2457         &dp_meter_genl_family,
2458 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2459         &dp_ct_limit_genl_family,
2460 #endif
2461 };
2462
2463 static void dp_unregister_genl(int n_families)
2464 {
2465         int i;
2466
2467         for (i = 0; i < n_families; i++)
2468                 genl_unregister_family(dp_genl_families[i]);
2469 }
2470
2471 static int __init dp_register_genl(void)
2472 {
2473         int err;
2474         int i;
2475
2476         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2477
2478                 err = genl_register_family(dp_genl_families[i]);
2479                 if (err)
2480                         goto error;
2481         }
2482
2483         return 0;
2484
2485 error:
2486         dp_unregister_genl(i);
2487         return err;
2488 }
2489
2490 static int __net_init ovs_init_net(struct net *net)
2491 {
2492         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2493         int err;
2494
2495         INIT_LIST_HEAD(&ovs_net->dps);
2496         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2497         INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2498
2499         err = ovs_ct_init(net);
2500         if (err)
2501                 return err;
2502
2503         schedule_delayed_work(&ovs_net->masks_rebalance,
2504                               msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2505         return 0;
2506 }
2507
2508 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2509                                             struct list_head *head)
2510 {
2511         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2512         struct datapath *dp;
2513
2514         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2515                 int i;
2516
2517                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2518                         struct vport *vport;
2519
2520                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2521                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2522                                         continue;
2523
2524                                 if (dev_net(vport->dev) == dnet)
2525                                         list_add(&vport->detach_list, head);
2526                         }
2527                 }
2528         }
2529 }
2530
2531 static void __net_exit ovs_exit_net(struct net *dnet)
2532 {
2533         struct datapath *dp, *dp_next;
2534         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2535         struct vport *vport, *vport_next;
2536         struct net *net;
2537         LIST_HEAD(head);
2538
2539         ovs_lock();
2540
2541         ovs_ct_exit(dnet);
2542
2543         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2544                 __dp_destroy(dp);
2545
2546         down_read(&net_rwsem);
2547         for_each_net(net)
2548                 list_vports_from_net(net, dnet, &head);
2549         up_read(&net_rwsem);
2550
2551         /* Detach all vports from given namespace. */
2552         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2553                 list_del(&vport->detach_list);
2554                 ovs_dp_detach_port(vport);
2555         }
2556
2557         ovs_unlock();
2558
2559         cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2560         cancel_work_sync(&ovs_net->dp_notify_work);
2561 }
2562
2563 static struct pernet_operations ovs_net_ops = {
2564         .init = ovs_init_net,
2565         .exit = ovs_exit_net,
2566         .id   = &ovs_net_id,
2567         .size = sizeof(struct ovs_net),
2568 };
2569
2570 static int __init dp_init(void)
2571 {
2572         int err;
2573
2574         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2575                      sizeof_field(struct sk_buff, cb));
2576
2577         pr_info("Open vSwitch switching datapath\n");
2578
2579         err = action_fifos_init();
2580         if (err)
2581                 goto error;
2582
2583         err = ovs_internal_dev_rtnl_link_register();
2584         if (err)
2585                 goto error_action_fifos_exit;
2586
2587         err = ovs_flow_init();
2588         if (err)
2589                 goto error_unreg_rtnl_link;
2590
2591         err = ovs_vport_init();
2592         if (err)
2593                 goto error_flow_exit;
2594
2595         err = register_pernet_device(&ovs_net_ops);
2596         if (err)
2597                 goto error_vport_exit;
2598
2599         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2600         if (err)
2601                 goto error_netns_exit;
2602
2603         err = ovs_netdev_init();
2604         if (err)
2605                 goto error_unreg_notifier;
2606
2607         err = dp_register_genl();
2608         if (err < 0)
2609                 goto error_unreg_netdev;
2610
2611         return 0;
2612
2613 error_unreg_netdev:
2614         ovs_netdev_exit();
2615 error_unreg_notifier:
2616         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2617 error_netns_exit:
2618         unregister_pernet_device(&ovs_net_ops);
2619 error_vport_exit:
2620         ovs_vport_exit();
2621 error_flow_exit:
2622         ovs_flow_exit();
2623 error_unreg_rtnl_link:
2624         ovs_internal_dev_rtnl_link_unregister();
2625 error_action_fifos_exit:
2626         action_fifos_exit();
2627 error:
2628         return err;
2629 }
2630
2631 static void dp_cleanup(void)
2632 {
2633         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2634         ovs_netdev_exit();
2635         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2636         unregister_pernet_device(&ovs_net_ops);
2637         rcu_barrier();
2638         ovs_vport_exit();
2639         ovs_flow_exit();
2640         ovs_internal_dev_rtnl_link_unregister();
2641         action_fifos_exit();
2642 }
2643
2644 module_init(dp_init);
2645 module_exit(dp_cleanup);
2646
2647 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2648 MODULE_LICENSE("GPL");
2649 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2650 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2651 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2652 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2653 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2654 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);