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 #include <net/pkt_cls.h>
  41
  42 #include "datapath.h"
  43 #include "flow.h"
  44 #include "flow_table.h"
  45 #include "flow_netlink.h"
  46 #include "meter.h"
  47 #include "openvswitch_trace.h"
  48 #include "vport-internal_dev.h"
  49 #include "vport-netdev.h"
  50
  51 unsigned int ovs_net_id __read_mostly;
  52
  53 static struct genl_family dp_packet_genl_family;
  54 static struct genl_family dp_flow_genl_family;
  55 static struct genl_family dp_datapath_genl_family;
  56
  57 static const struct nla_policy flow_policy[];
  58
  59 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
  60         .name = OVS_FLOW_MCGROUP,
  61 };
  62
  63 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  64         .name = OVS_DATAPATH_MCGROUP,
  65 };
  66
  67 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
  68         .name = OVS_VPORT_MCGROUP,
  69 };
  70
  71 /* Check if need to build a reply message.
  72  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  73 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  74                             unsigned int group)
  75 {
  76         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  77                genl_has_listeners(family, genl_info_net(info), group);
  78 }
  79
  80 static void ovs_notify(struct genl_family *family,
  81                        struct sk_buff *skb, struct genl_info *info)
  82 {
  83         genl_notify(family, skb, info, 0, GFP_KERNEL);
  84 }
  85
  86 /**
  87  * DOC: Locking:
  88  *
  89  * All writes e.g. Writes to device state (add/remove datapath, port, set
  90  * operations on vports, etc.), Writes to other state (flow table
  91  * modifications, set miscellaneous datapath parameters, etc.) are protected
  92  * by ovs_lock.
  93  *
  94  * Reads are protected by RCU.
  95  *
  96  * There are a few special cases (mostly stats) that have their own
  97  * synchronization but they nest under all of above and don't interact with
  98  * each other.
  99  *
 100  * The RTNL lock nests inside ovs_mutex.
 101  */
 102
 103 static DEFINE_MUTEX(ovs_mutex);
 104
 105 void ovs_lock(void)
 106 {
 107         mutex_lock(&ovs_mutex);
 108 }
 109
 110 void ovs_unlock(void)
 111 {
 112         mutex_unlock(&ovs_mutex);
 113 }
 114
 115 #ifdef CONFIG_LOCKDEP
 116 int lockdep_ovsl_is_held(void)
 117 {
 118         if (debug_locks)
 119                 return lockdep_is_held(&ovs_mutex);
 120         else
 121                 return 1;
 122 }
 123 #endif
 124
 125 static struct vport *new_vport(const struct vport_parms *);
 126 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 127                              const struct sw_flow_key *,
 128                              const struct dp_upcall_info *,
 129                              uint32_t cutlen);
 130 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 131                                   const struct sw_flow_key *,
 132                                   const struct dp_upcall_info *,
 133                                   uint32_t cutlen);
 134
 135 static void ovs_dp_masks_rebalance(struct work_struct *work);
 136
 137 static int ovs_dp_set_upcall_portids(struct datapath *, const struct nlattr *);
 138
 139 /* Must be called with rcu_read_lock or ovs_mutex. */
 140 const char *ovs_dp_name(const struct datapath *dp)
 141 {
 142         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 143         return ovs_vport_name(vport);
 144 }
 145
 146 static int get_dpifindex(const struct datapath *dp)
 147 {
 148         struct vport *local;
 149         int ifindex;
 150
 151         rcu_read_lock();
 152
 153         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 154         if (local)
 155                 ifindex = local->dev->ifindex;
 156         else
 157                 ifindex = 0;
 158
 159         rcu_read_unlock();
 160
 161         return ifindex;
 162 }
 163
 164 static void destroy_dp_rcu(struct rcu_head *rcu)
 165 {
 166         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 167
 168         ovs_flow_tbl_destroy(&dp->table);
 169         free_percpu(dp->stats_percpu);
 170         kfree(dp->ports);
 171         ovs_meters_exit(dp);
 172         kfree(rcu_dereference_raw(dp->upcall_portids));
 173         kfree(dp);
 174 }
 175
 176 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 177                                             u16 port_no)
 178 {
 179         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 180 }
 181
 182 /* Called with ovs_mutex or RCU read lock. */
 183 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 184 {
 185         struct vport *vport;
 186         struct hlist_head *head;
 187
 188         head = vport_hash_bucket(dp, port_no);
 189         hlist_for_each_entry_rcu(vport, head, dp_hash_node,
 190                                  lockdep_ovsl_is_held()) {
 191                 if (vport->port_no == port_no)
 192                         return vport;
 193         }
 194         return NULL;
 195 }
 196
 197 /* Called with ovs_mutex. */
 198 static struct vport *new_vport(const struct vport_parms *parms)
 199 {
 200         struct vport *vport;
 201
 202         vport = ovs_vport_add(parms);
 203         if (!IS_ERR(vport)) {
 204                 struct datapath *dp = parms->dp;
 205                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 206
 207                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 208         }
 209         return vport;
 210 }
 211
 212 void ovs_dp_detach_port(struct vport *p)
 213 {
 214         ASSERT_OVSL();
 215
 216         /* First drop references to device. */
 217         hlist_del_rcu(&p->dp_hash_node);
 218
 219         /* Then destroy it. */
 220         ovs_vport_del(p);
 221 }
 222
 223 /* Must be called with rcu_read_lock. */
 224 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 225 {
 226         const struct vport *p = OVS_CB(skb)->input_vport;
 227         struct datapath *dp = p->dp;
 228         struct sw_flow *flow;
 229         struct sw_flow_actions *sf_acts;
 230         struct dp_stats_percpu *stats;
 231         u64 *stats_counter;
 232         u32 n_mask_hit;
 233         u32 n_cache_hit;
 234         int error;
 235
 236         stats = this_cpu_ptr(dp->stats_percpu);
 237
 238         /* Look up flow. */
 239         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
 240                                          &n_mask_hit, &n_cache_hit);
 241         if (unlikely(!flow)) {
 242                 struct dp_upcall_info upcall;
 243
 244                 memset(&upcall, 0, sizeof(upcall));
 245                 upcall.cmd = OVS_PACKET_CMD_MISS;
 246
 247                 if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
 248                         upcall.portid =
 249                             ovs_dp_get_upcall_portid(dp, smp_processor_id());
 250                 else
 251                         upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 252
 253                 upcall.mru = OVS_CB(skb)->mru;
 254                 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
 255                 switch (error) {
 256                 case 0:
 257                 case -EAGAIN:
 258                 case -ERESTARTSYS:
 259                 case -EINTR:
 260                         consume_skb(skb);
 261                         break;
 262                 default:
 263                         kfree_skb(skb);
 264                         break;
 265                 }
 266                 stats_counter = &stats->n_missed;
 267                 goto out;
 268         }
 269
 270         ovs_flow_stats_update(flow, key->tp.flags, skb);
 271         sf_acts = rcu_dereference(flow->sf_acts);
 272         error = ovs_execute_actions(dp, skb, sf_acts, key);
 273         if (unlikely(error))
 274                 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
 275                                     ovs_dp_name(dp), error);
 276
 277         stats_counter = &stats->n_hit;
 278
 279 out:
 280         /* Update datapath statistics. */
 281         u64_stats_update_begin(&stats->syncp);
 282         (*stats_counter)++;
 283         stats->n_mask_hit += n_mask_hit;
 284         stats->n_cache_hit += n_cache_hit;
 285         u64_stats_update_end(&stats->syncp);
 286 }
 287
 288 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 289                   const struct sw_flow_key *key,
 290                   const struct dp_upcall_info *upcall_info,
 291                   uint32_t cutlen)
 292 {
 293         struct dp_stats_percpu *stats;
 294         int err;
 295
 296         if (trace_ovs_dp_upcall_enabled())
 297                 trace_ovs_dp_upcall(dp, skb, key, upcall_info);
 298
 299         if (upcall_info->portid == 0) {
 300                 err = -ENOTCONN;
 301                 goto err;
 302         }
 303
 304         if (!skb_is_gso(skb))
 305                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 306         else
 307                 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
 308         if (err)
 309                 goto err;
 310
 311         return 0;
 312
 313 err:
 314         stats = this_cpu_ptr(dp->stats_percpu);
 315
 316         u64_stats_update_begin(&stats->syncp);
 317         stats->n_lost++;
 318         u64_stats_update_end(&stats->syncp);
 319
 320         return err;
 321 }
 322
 323 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 324                              const struct sw_flow_key *key,
 325                              const struct dp_upcall_info *upcall_info,
 326                              uint32_t cutlen)
 327 {
 328         unsigned int gso_type = skb_shinfo(skb)->gso_type;
 329         struct sw_flow_key later_key;
 330         struct sk_buff *segs, *nskb;
 331         int err;
 332
 333         BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_GSO_CB_OFFSET);
 334         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 335         if (IS_ERR(segs))
 336                 return PTR_ERR(segs);
 337         if (segs == NULL)
 338                 return -EINVAL;
 339
 340         if (gso_type & SKB_GSO_UDP) {
 341                 /* The initial flow key extracted by ovs_flow_key_extract()
 342                  * in this case is for a first fragment, so we need to
 343                  * properly mark later fragments.
 344                  */
 345                 later_key = *key;
 346                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 347         }
 348
 349         /* Queue all of the segments. */
 350         skb_list_walk_safe(segs, skb, nskb) {
 351                 if (gso_type & SKB_GSO_UDP && skb != segs)
 352                         key = &later_key;
 353
 354                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 355                 if (err)
 356                         break;
 357
 358         }
 359
 360         /* Free all of the segments. */
 361         skb_list_walk_safe(segs, skb, nskb) {
 362                 if (err)
 363                         kfree_skb(skb);
 364                 else
 365                         consume_skb(skb);
 366         }
 367         return err;
 368 }
 369
 370 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 371                               unsigned int hdrlen, int actions_attrlen)
 372 {
 373         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 374                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 375                 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 376                 + nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
 377                 + nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
 378
 379         /* OVS_PACKET_ATTR_USERDATA */
 380         if (upcall_info->userdata)
 381                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 382
 383         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 384         if (upcall_info->egress_tun_info)
 385                 size += nla_total_size(ovs_tun_key_attr_size());
 386
 387         /* OVS_PACKET_ATTR_ACTIONS */
 388         if (upcall_info->actions_len)
 389                 size += nla_total_size(actions_attrlen);
 390
 391         /* OVS_PACKET_ATTR_MRU */
 392         if (upcall_info->mru)
 393                 size += nla_total_size(sizeof(upcall_info->mru));
 394
 395         return size;
 396 }
 397
 398 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 399 {
 400         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 401                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 402
 403                 if (plen > 0)
 404                         skb_put_zero(skb, plen);
 405         }
 406 }
 407
 408 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 409                                   const struct sw_flow_key *key,
 410                                   const struct dp_upcall_info *upcall_info,
 411                                   uint32_t cutlen)
 412 {
 413         struct ovs_header *upcall;
 414         struct sk_buff *nskb = NULL;
 415         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 416         struct nlattr *nla;
 417         size_t len;
 418         unsigned int hlen;
 419         int err, dp_ifindex;
 420         u64 hash;
 421
 422         dp_ifindex = get_dpifindex(dp);
 423         if (!dp_ifindex)
 424                 return -ENODEV;
 425
 426         if (skb_vlan_tag_present(skb)) {
 427                 nskb = skb_clone(skb, GFP_ATOMIC);
 428                 if (!nskb)
 429                         return -ENOMEM;
 430
 431                 nskb = __vlan_hwaccel_push_inside(nskb);
 432                 if (!nskb)
 433                         return -ENOMEM;
 434
 435                 skb = nskb;
 436         }
 437
 438         if (nla_attr_size(skb->len) > USHRT_MAX) {
 439                 err = -EFBIG;
 440                 goto out;
 441         }
 442
 443         /* Complete checksum if needed */
 444         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 445             (err = skb_csum_hwoffload_help(skb, 0)))
 446                 goto out;
 447
 448         /* Older versions of OVS user space enforce alignment of the last
 449          * Netlink attribute to NLA_ALIGNTO which would require extensive
 450          * padding logic. Only perform zerocopy if padding is not required.
 451          */
 452         if (dp->user_features & OVS_DP_F_UNALIGNED)
 453                 hlen = skb_zerocopy_headlen(skb);
 454         else
 455                 hlen = skb->len;
 456
 457         len = upcall_msg_size(upcall_info, hlen - cutlen,
 458                               OVS_CB(skb)->acts_origlen);
 459         user_skb = genlmsg_new(len, GFP_ATOMIC);
 460         if (!user_skb) {
 461                 err = -ENOMEM;
 462                 goto out;
 463         }
 464
 465         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 466                              0, upcall_info->cmd);
 467         if (!upcall) {
 468                 err = -EINVAL;
 469                 goto out;
 470         }
 471         upcall->dp_ifindex = dp_ifindex;
 472
 473         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 474         if (err)
 475                 goto out;
 476
 477         if (upcall_info->userdata)
 478                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 479                           nla_len(upcall_info->userdata),
 480                           nla_data(upcall_info->userdata));
 481
 482         if (upcall_info->egress_tun_info) {
 483                 nla = nla_nest_start_noflag(user_skb,
 484                                             OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 485                 if (!nla) {
 486                         err = -EMSGSIZE;
 487                         goto out;
 488                 }
 489                 err = ovs_nla_put_tunnel_info(user_skb,
 490                                               upcall_info->egress_tun_info);
 491                 if (err)
 492                         goto out;
 493
 494                 nla_nest_end(user_skb, nla);
 495         }
 496
 497         if (upcall_info->actions_len) {
 498                 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
 499                 if (!nla) {
 500                         err = -EMSGSIZE;
 501                         goto out;
 502                 }
 503                 err = ovs_nla_put_actions(upcall_info->actions,
 504                                           upcall_info->actions_len,
 505                                           user_skb);
 506                 if (!err)
 507                         nla_nest_end(user_skb, nla);
 508                 else
 509                         nla_nest_cancel(user_skb, nla);
 510         }
 511
 512         /* Add OVS_PACKET_ATTR_MRU */
 513         if (upcall_info->mru &&
 514             nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
 515                 err = -ENOBUFS;
 516                 goto out;
 517         }
 518
 519         /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 520         if (cutlen > 0 &&
 521             nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
 522                 err = -ENOBUFS;
 523                 goto out;
 524         }
 525
 526         /* Add OVS_PACKET_ATTR_HASH */
 527         hash = skb_get_hash_raw(skb);
 528         if (skb->sw_hash)
 529                 hash |= OVS_PACKET_HASH_SW_BIT;
 530
 531         if (skb->l4_hash)
 532                 hash |= OVS_PACKET_HASH_L4_BIT;
 533
 534         if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
 535                 err = -ENOBUFS;
 536                 goto out;
 537         }
 538
 539         /* Only reserve room for attribute header, packet data is added
 540          * in skb_zerocopy() */
 541         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 542                 err = -ENOBUFS;
 543                 goto out;
 544         }
 545         nla->nla_len = nla_attr_size(skb->len - cutlen);
 546
 547         err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 548         if (err)
 549                 goto out;
 550
 551         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 552         pad_packet(dp, user_skb);
 553
 554         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 555
 556         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 557         user_skb = NULL;
 558 out:
 559         if (err)
 560                 skb_tx_error(skb);
 561         consume_skb(user_skb);
 562         consume_skb(nskb);
 563
 564         return err;
 565 }
 566
 567 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 568 {
 569         struct ovs_header *ovs_header = info->userhdr;
 570         struct net *net = sock_net(skb->sk);
 571         struct nlattr **a = info->attrs;
 572         struct sw_flow_actions *acts;
 573         struct sk_buff *packet;
 574         struct sw_flow *flow;
 575         struct sw_flow_actions *sf_acts;
 576         struct datapath *dp;
 577         struct vport *input_vport;
 578         u16 mru = 0;
 579         u64 hash;
 580         int len;
 581         int err;
 582         bool log = !a[OVS_PACKET_ATTR_PROBE];
 583
 584         err = -EINVAL;
 585         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 586             !a[OVS_PACKET_ATTR_ACTIONS])
 587                 goto err;
 588
 589         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 590         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 591         err = -ENOMEM;
 592         if (!packet)
 593                 goto err;
 594         skb_reserve(packet, NET_IP_ALIGN);
 595
 596         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 597
 598         /* Set packet's mru */
 599         if (a[OVS_PACKET_ATTR_MRU]) {
 600                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 601                 packet->ignore_df = 1;
 602         }
 603         OVS_CB(packet)->mru = mru;
 604
 605         if (a[OVS_PACKET_ATTR_HASH]) {
 606                 hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
 607
 608                 __skb_set_hash(packet, hash & 0xFFFFFFFFULL,
 609                                !!(hash & OVS_PACKET_HASH_SW_BIT),
 610                                !!(hash & OVS_PACKET_HASH_L4_BIT));
 611         }
 612
 613         /* Build an sw_flow for sending this packet. */
 614         flow = ovs_flow_alloc();
 615         err = PTR_ERR(flow);
 616         if (IS_ERR(flow))
 617                 goto err_kfree_skb;
 618
 619         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 620                                              packet, &flow->key, log);
 621         if (err)
 622                 goto err_flow_free;
 623
 624         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 625                                    &flow->key, &acts, log);
 626         if (err)
 627                 goto err_flow_free;
 628
 629         rcu_assign_pointer(flow->sf_acts, acts);
 630         packet->priority = flow->key.phy.priority;
 631         packet->mark = flow->key.phy.skb_mark;
 632
 633         rcu_read_lock();
 634         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 635         err = -ENODEV;
 636         if (!dp)
 637                 goto err_unlock;
 638
 639         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 640         if (!input_vport)
 641                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 642
 643         if (!input_vport)
 644                 goto err_unlock;
 645
 646         packet->dev = input_vport->dev;
 647         OVS_CB(packet)->input_vport = input_vport;
 648         sf_acts = rcu_dereference(flow->sf_acts);
 649
 650         local_bh_disable();
 651         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 652         local_bh_enable();
 653         rcu_read_unlock();
 654
 655         ovs_flow_free(flow, false);
 656         return err;
 657
 658 err_unlock:
 659         rcu_read_unlock();
 660 err_flow_free:
 661         ovs_flow_free(flow, false);
 662 err_kfree_skb:
 663         kfree_skb(packet);
 664 err:
 665         return err;
 666 }
 667
 668 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 669         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 670         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 671         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 672         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 673         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 674         [OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
 675 };
 676
 677 static const struct genl_small_ops dp_packet_genl_ops[] = {
 678         { .cmd = OVS_PACKET_CMD_EXECUTE,
 679           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 680           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 681           .doit = ovs_packet_cmd_execute
 682         }
 683 };
 684
 685 static struct genl_family dp_packet_genl_family __ro_after_init = {
 686         .hdrsize = sizeof(struct ovs_header),
 687         .name = OVS_PACKET_FAMILY,
 688         .version = OVS_PACKET_VERSION,
 689         .maxattr = OVS_PACKET_ATTR_MAX,
 690         .policy = packet_policy,
 691         .netnsok = true,
 692         .parallel_ops = true,
 693         .small_ops = dp_packet_genl_ops,
 694         .n_small_ops = ARRAY_SIZE(dp_packet_genl_ops),
 695         .resv_start_op = OVS_PACKET_CMD_EXECUTE + 1,
 696         .module = THIS_MODULE,
 697 };
 698
 699 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 700                          struct ovs_dp_megaflow_stats *mega_stats)
 701 {
 702         int i;
 703
 704         memset(mega_stats, 0, sizeof(*mega_stats));
 705
 706         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 707         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 708
 709         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 710
 711         for_each_possible_cpu(i) {
 712                 const struct dp_stats_percpu *percpu_stats;
 713                 struct dp_stats_percpu local_stats;
 714                 unsigned int start;
 715
 716                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 717
 718                 do {
 719                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 720                         local_stats = *percpu_stats;
 721                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 722
 723                 stats->n_hit += local_stats.n_hit;
 724                 stats->n_missed += local_stats.n_missed;
 725                 stats->n_lost += local_stats.n_lost;
 726                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 727                 mega_stats->n_cache_hit += local_stats.n_cache_hit;
 728         }
 729 }
 730
 731 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 732 {
 733         return ovs_identifier_is_ufid(sfid) &&
 734                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 735 }
 736
 737 static bool should_fill_mask(uint32_t ufid_flags)
 738 {
 739         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 740 }
 741
 742 static bool should_fill_actions(uint32_t ufid_flags)
 743 {
 744         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 745 }
 746
 747 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 748                                     const struct sw_flow_id *sfid,
 749                                     uint32_t ufid_flags)
 750 {
 751         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 752
 753         /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
 754          * see ovs_nla_put_identifier()
 755          */
 756         if (sfid && ovs_identifier_is_ufid(sfid))
 757                 len += nla_total_size(sfid->ufid_len);
 758         else
 759                 len += nla_total_size(ovs_key_attr_size());
 760
 761         /* OVS_FLOW_ATTR_KEY */
 762         if (!sfid || should_fill_key(sfid, ufid_flags))
 763                 len += nla_total_size(ovs_key_attr_size());
 764
 765         /* OVS_FLOW_ATTR_MASK */
 766         if (should_fill_mask(ufid_flags))
 767                 len += nla_total_size(ovs_key_attr_size());
 768
 769         /* OVS_FLOW_ATTR_ACTIONS */
 770         if (should_fill_actions(ufid_flags))
 771                 len += nla_total_size(acts->orig_len);
 772
 773         return len
 774                 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 775                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 776                 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 777 }
 778
 779 /* Called with ovs_mutex or RCU read lock. */
 780 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 781                                    struct sk_buff *skb)
 782 {
 783         struct ovs_flow_stats stats;
 784         __be16 tcp_flags;
 785         unsigned long used;
 786
 787         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 788
 789         if (used &&
 790             nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 791                               OVS_FLOW_ATTR_PAD))
 792                 return -EMSGSIZE;
 793
 794         if (stats.n_packets &&
 795             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 796                           sizeof(struct ovs_flow_stats), &stats,
 797                           OVS_FLOW_ATTR_PAD))
 798                 return -EMSGSIZE;
 799
 800         if ((u8)ntohs(tcp_flags) &&
 801              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 802                 return -EMSGSIZE;
 803
 804         return 0;
 805 }
 806
 807 /* Called with ovs_mutex or RCU read lock. */
 808 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 809                                      struct sk_buff *skb, int skb_orig_len)
 810 {
 811         struct nlattr *start;
 812         int err;
 813
 814         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 815          * this is the first flow to be dumped into 'skb'.  This is unusual for
 816          * Netlink but individual action lists can be longer than
 817          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 818          * The userspace caller can always fetch the actions separately if it
 819          * really wants them.  (Most userspace callers in fact don't care.)
 820          *
 821          * This can only fail for dump operations because the skb is always
 822          * properly sized for single flows.
 823          */
 824         start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
 825         if (start) {
 826                 const struct sw_flow_actions *sf_acts;
 827
 828                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 829                 err = ovs_nla_put_actions(sf_acts->actions,
 830                                           sf_acts->actions_len, skb);
 831
 832                 if (!err)
 833                         nla_nest_end(skb, start);
 834                 else {
 835                         if (skb_orig_len)
 836                                 return err;
 837
 838                         nla_nest_cancel(skb, start);
 839                 }
 840         } else if (skb_orig_len) {
 841                 return -EMSGSIZE;
 842         }
 843
 844         return 0;
 845 }
 846
 847 /* Called with ovs_mutex or RCU read lock. */
 848 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 849                                   struct sk_buff *skb, u32 portid,
 850                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 851 {
 852         const int skb_orig_len = skb->len;
 853         struct ovs_header *ovs_header;
 854         int err;
 855
 856         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 857                                  flags, cmd);
 858         if (!ovs_header)
 859                 return -EMSGSIZE;
 860
 861         ovs_header->dp_ifindex = dp_ifindex;
 862
 863         err = ovs_nla_put_identifier(flow, skb);
 864         if (err)
 865                 goto error;
 866
 867         if (should_fill_key(&flow->id, ufid_flags)) {
 868                 err = ovs_nla_put_masked_key(flow, skb);
 869                 if (err)
 870                         goto error;
 871         }
 872
 873         if (should_fill_mask(ufid_flags)) {
 874                 err = ovs_nla_put_mask(flow, skb);
 875                 if (err)
 876                         goto error;
 877         }
 878
 879         err = ovs_flow_cmd_fill_stats(flow, skb);
 880         if (err)
 881                 goto error;
 882
 883         if (should_fill_actions(ufid_flags)) {
 884                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 885                 if (err)
 886                         goto error;
 887         }
 888
 889         genlmsg_end(skb, ovs_header);
 890         return 0;
 891
 892 error:
 893         genlmsg_cancel(skb, ovs_header);
 894         return err;
 895 }
 896
 897 /* May not be called with RCU read lock. */
 898 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 899                                                const struct sw_flow_id *sfid,
 900                                                struct genl_info *info,
 901                                                bool always,
 902                                                uint32_t ufid_flags)
 903 {
 904         struct sk_buff *skb;
 905         size_t len;
 906
 907         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 908                 return NULL;
 909
 910         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 911         skb = genlmsg_new(len, GFP_KERNEL);
 912         if (!skb)
 913                 return ERR_PTR(-ENOMEM);
 914
 915         return skb;
 916 }
 917
 918 /* Called with ovs_mutex. */
 919 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 920                                                int dp_ifindex,
 921                                                struct genl_info *info, u8 cmd,
 922                                                bool always, u32 ufid_flags)
 923 {
 924         struct sk_buff *skb;
 925         int retval;
 926
 927         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 928                                       &flow->id, info, always, ufid_flags);
 929         if (IS_ERR_OR_NULL(skb))
 930                 return skb;
 931
 932         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 933                                         info->snd_portid, info->snd_seq, 0,
 934                                         cmd, ufid_flags);
 935         if (WARN_ON_ONCE(retval < 0)) {
 936                 kfree_skb(skb);
 937                 skb = ERR_PTR(retval);
 938         }
 939         return skb;
 940 }
 941
 942 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 943 {
 944         struct net *net = sock_net(skb->sk);
 945         struct nlattr **a = info->attrs;
 946         struct ovs_header *ovs_header = info->userhdr;
 947         struct sw_flow *flow = NULL, *new_flow;
 948         struct sw_flow_mask mask;
 949         struct sk_buff *reply;
 950         struct datapath *dp;
 951         struct sw_flow_actions *acts;
 952         struct sw_flow_match match;
 953         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 954         int error;
 955         bool log = !a[OVS_FLOW_ATTR_PROBE];
 956
 957         /* Must have key and actions. */
 958         error = -EINVAL;
 959         if (!a[OVS_FLOW_ATTR_KEY]) {
 960                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 961                 goto error;
 962         }
 963         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 964                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 965                 goto error;
 966         }
 967
 968         /* Most of the time we need to allocate a new flow, do it before
 969          * locking.
 970          */
 971         new_flow = ovs_flow_alloc();
 972         if (IS_ERR(new_flow)) {
 973                 error = PTR_ERR(new_flow);
 974                 goto error;
 975         }
 976
 977         /* Extract key. */
 978         ovs_match_init(&match, &new_flow->key, false, &mask);
 979         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 980                                   a[OVS_FLOW_ATTR_MASK], log);
 981         if (error)
 982                 goto err_kfree_flow;
 983
 984         /* Extract flow identifier. */
 985         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 986                                        &new_flow->key, log);
 987         if (error)
 988                 goto err_kfree_flow;
 989
 990         /* unmasked key is needed to match when ufid is not used. */
 991         if (ovs_identifier_is_key(&new_flow->id))
 992                 match.key = new_flow->id.unmasked_key;
 993
 994         ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
 995
 996         /* Validate actions. */
 997         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 998                                      &new_flow->key, &acts, log);
 999         if (error) {
1000                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
1001                 goto err_kfree_flow;
1002         }
1003
1004         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
1005                                         ufid_flags);
1006         if (IS_ERR(reply)) {
1007                 error = PTR_ERR(reply);
1008                 goto err_kfree_acts;
1009         }
1010
1011         ovs_lock();
1012         dp = get_dp(net, ovs_header->dp_ifindex);
1013         if (unlikely(!dp)) {
1014                 error = -ENODEV;
1015                 goto err_unlock_ovs;
1016         }
1017
1018         /* Check if this is a duplicate flow */
1019         if (ovs_identifier_is_ufid(&new_flow->id))
1020                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1021         if (!flow)
1022                 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
1023         if (likely(!flow)) {
1024                 rcu_assign_pointer(new_flow->sf_acts, acts);
1025
1026                 /* Put flow in bucket. */
1027                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1028                 if (unlikely(error)) {
1029                         acts = NULL;
1030                         goto err_unlock_ovs;
1031                 }
1032
1033                 if (unlikely(reply)) {
1034                         error = ovs_flow_cmd_fill_info(new_flow,
1035                                                        ovs_header->dp_ifindex,
1036                                                        reply, info->snd_portid,
1037                                                        info->snd_seq, 0,
1038                                                        OVS_FLOW_CMD_NEW,
1039                                                        ufid_flags);
1040                         BUG_ON(error < 0);
1041                 }
1042                 ovs_unlock();
1043         } else {
1044                 struct sw_flow_actions *old_acts;
1045
1046                 /* Bail out if we're not allowed to modify an existing flow.
1047                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1048                  * because Generic Netlink treats the latter as a dump
1049                  * request.  We also accept NLM_F_EXCL in case that bug ever
1050                  * gets fixed.
1051                  */
1052                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1053                                                          | NLM_F_EXCL))) {
1054                         error = -EEXIST;
1055                         goto err_unlock_ovs;
1056                 }
1057                 /* The flow identifier has to be the same for flow updates.
1058                  * Look for any overlapping flow.
1059                  */
1060                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1061                         if (ovs_identifier_is_key(&flow->id))
1062                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1063                                                                  &match);
1064                         else /* UFID matches but key is different */
1065                                 flow = NULL;
1066                         if (!flow) {
1067                                 error = -ENOENT;
1068                                 goto err_unlock_ovs;
1069                         }
1070                 }
1071                 /* Update actions. */
1072                 old_acts = ovsl_dereference(flow->sf_acts);
1073                 rcu_assign_pointer(flow->sf_acts, acts);
1074
1075                 if (unlikely(reply)) {
1076                         error = ovs_flow_cmd_fill_info(flow,
1077                                                        ovs_header->dp_ifindex,
1078                                                        reply, info->snd_portid,
1079                                                        info->snd_seq, 0,
1080                                                        OVS_FLOW_CMD_NEW,
1081                                                        ufid_flags);
1082                         BUG_ON(error < 0);
1083                 }
1084                 ovs_unlock();
1085
1086                 ovs_nla_free_flow_actions_rcu(old_acts);
1087                 ovs_flow_free(new_flow, false);
1088         }
1089
1090         if (reply)
1091                 ovs_notify(&dp_flow_genl_family, reply, info);
1092         return 0;
1093
1094 err_unlock_ovs:
1095         ovs_unlock();
1096         kfree_skb(reply);
1097 err_kfree_acts:
1098         ovs_nla_free_flow_actions(acts);
1099 err_kfree_flow:
1100         ovs_flow_free(new_flow, false);
1101 error:
1102         return error;
1103 }
1104
1105 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1106 static noinline_for_stack
1107 struct sw_flow_actions *get_flow_actions(struct net *net,
1108                                          const struct nlattr *a,
1109                                          const struct sw_flow_key *key,
1110                                          const struct sw_flow_mask *mask,
1111                                          bool log)
1112 {
1113         struct sw_flow_actions *acts;
1114         struct sw_flow_key masked_key;
1115         int error;
1116
1117         ovs_flow_mask_key(&masked_key, key, true, mask);
1118         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1119         if (error) {
1120                 OVS_NLERR(log,
1121                           "Actions may not be safe on all matching packets");
1122                 return ERR_PTR(error);
1123         }
1124
1125         return acts;
1126 }
1127
1128 /* Factor out match-init and action-copy to avoid
1129  * "Wframe-larger-than=1024" warning. Because mask is only
1130  * used to get actions, we new a function to save some
1131  * stack space.
1132  *
1133  * If there are not key and action attrs, we return 0
1134  * directly. In the case, the caller will also not use the
1135  * match as before. If there is action attr, we try to get
1136  * actions and save them to *acts. Before returning from
1137  * the function, we reset the match->mask pointer. Because
1138  * we should not to return match object with dangling reference
1139  * to mask.
1140  * */
1141 static noinline_for_stack int
1142 ovs_nla_init_match_and_action(struct net *net,
1143                               struct sw_flow_match *match,
1144                               struct sw_flow_key *key,
1145                               struct nlattr **a,
1146                               struct sw_flow_actions **acts,
1147                               bool log)
1148 {
1149         struct sw_flow_mask mask;
1150         int error = 0;
1151
1152         if (a[OVS_FLOW_ATTR_KEY]) {
1153                 ovs_match_init(match, key, true, &mask);
1154                 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1155                                           a[OVS_FLOW_ATTR_MASK], log);
1156                 if (error)
1157                         goto error;
1158         }
1159
1160         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1161                 if (!a[OVS_FLOW_ATTR_KEY]) {
1162                         OVS_NLERR(log,
1163                                   "Flow key attribute not present in set flow.");
1164                         error = -EINVAL;
1165                         goto error;
1166                 }
1167
1168                 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1169                                          &mask, log);
1170                 if (IS_ERR(*acts)) {
1171                         error = PTR_ERR(*acts);
1172                         goto error;
1173                 }
1174         }
1175
1176         /* On success, error is 0. */
1177 error:
1178         match->mask = NULL;
1179         return error;
1180 }
1181
1182 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1183 {
1184         struct net *net = sock_net(skb->sk);
1185         struct nlattr **a = info->attrs;
1186         struct ovs_header *ovs_header = info->userhdr;
1187         struct sw_flow_key key;
1188         struct sw_flow *flow;
1189         struct sk_buff *reply = NULL;
1190         struct datapath *dp;
1191         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1192         struct sw_flow_match match;
1193         struct sw_flow_id sfid;
1194         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1195         int error = 0;
1196         bool log = !a[OVS_FLOW_ATTR_PROBE];
1197         bool ufid_present;
1198
1199         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1200         if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1201                 OVS_NLERR(log,
1202                           "Flow set message rejected, Key attribute missing.");
1203                 return -EINVAL;
1204         }
1205
1206         error = ovs_nla_init_match_and_action(net, &match, &key, a,
1207                                               &acts, log);
1208         if (error)
1209                 goto error;
1210
1211         if (acts) {
1212                 /* Can allocate before locking if have acts. */
1213                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1214                                                 ufid_flags);
1215                 if (IS_ERR(reply)) {
1216                         error = PTR_ERR(reply);
1217                         goto err_kfree_acts;
1218                 }
1219         }
1220
1221         ovs_lock();
1222         dp = get_dp(net, ovs_header->dp_ifindex);
1223         if (unlikely(!dp)) {
1224                 error = -ENODEV;
1225                 goto err_unlock_ovs;
1226         }
1227         /* Check that the flow exists. */
1228         if (ufid_present)
1229                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1230         else
1231                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1232         if (unlikely(!flow)) {
1233                 error = -ENOENT;
1234                 goto err_unlock_ovs;
1235         }
1236
1237         /* Update actions, if present. */
1238         if (likely(acts)) {
1239                 old_acts = ovsl_dereference(flow->sf_acts);
1240                 rcu_assign_pointer(flow->sf_acts, acts);
1241
1242                 if (unlikely(reply)) {
1243                         error = ovs_flow_cmd_fill_info(flow,
1244                                                        ovs_header->dp_ifindex,
1245                                                        reply, info->snd_portid,
1246                                                        info->snd_seq, 0,
1247                                                        OVS_FLOW_CMD_SET,
1248                                                        ufid_flags);
1249                         BUG_ON(error < 0);
1250                 }
1251         } else {
1252                 /* Could not alloc without acts before locking. */
1253                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1254                                                 info, OVS_FLOW_CMD_SET, false,
1255                                                 ufid_flags);
1256
1257                 if (IS_ERR(reply)) {
1258                         error = PTR_ERR(reply);
1259                         goto err_unlock_ovs;
1260                 }
1261         }
1262
1263         /* Clear stats. */
1264         if (a[OVS_FLOW_ATTR_CLEAR])
1265                 ovs_flow_stats_clear(flow);
1266         ovs_unlock();
1267
1268         if (reply)
1269                 ovs_notify(&dp_flow_genl_family, reply, info);
1270         if (old_acts)
1271                 ovs_nla_free_flow_actions_rcu(old_acts);
1272
1273         return 0;
1274
1275 err_unlock_ovs:
1276         ovs_unlock();
1277         kfree_skb(reply);
1278 err_kfree_acts:
1279         ovs_nla_free_flow_actions(acts);
1280 error:
1281         return error;
1282 }
1283
1284 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1285 {
1286         struct nlattr **a = info->attrs;
1287         struct ovs_header *ovs_header = info->userhdr;
1288         struct net *net = sock_net(skb->sk);
1289         struct sw_flow_key key;
1290         struct sk_buff *reply;
1291         struct sw_flow *flow;
1292         struct datapath *dp;
1293         struct sw_flow_match match;
1294         struct sw_flow_id ufid;
1295         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1296         int err = 0;
1297         bool log = !a[OVS_FLOW_ATTR_PROBE];
1298         bool ufid_present;
1299
1300         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1301         if (a[OVS_FLOW_ATTR_KEY]) {
1302                 ovs_match_init(&match, &key, true, NULL);
1303                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1304                                         log);
1305         } else if (!ufid_present) {
1306                 OVS_NLERR(log,
1307                           "Flow get message rejected, Key attribute missing.");
1308                 err = -EINVAL;
1309         }
1310         if (err)
1311                 return err;
1312
1313         ovs_lock();
1314         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1315         if (!dp) {
1316                 err = -ENODEV;
1317                 goto unlock;
1318         }
1319
1320         if (ufid_present)
1321                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1322         else
1323                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1324         if (!flow) {
1325                 err = -ENOENT;
1326                 goto unlock;
1327         }
1328
1329         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1330                                         OVS_FLOW_CMD_GET, true, ufid_flags);
1331         if (IS_ERR(reply)) {
1332                 err = PTR_ERR(reply);
1333                 goto unlock;
1334         }
1335
1336         ovs_unlock();
1337         return genlmsg_reply(reply, info);
1338 unlock:
1339         ovs_unlock();
1340         return err;
1341 }
1342
1343 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1344 {
1345         struct nlattr **a = info->attrs;
1346         struct ovs_header *ovs_header = info->userhdr;
1347         struct net *net = sock_net(skb->sk);
1348         struct sw_flow_key key;
1349         struct sk_buff *reply;
1350         struct sw_flow *flow = NULL;
1351         struct datapath *dp;
1352         struct sw_flow_match match;
1353         struct sw_flow_id ufid;
1354         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1355         int err;
1356         bool log = !a[OVS_FLOW_ATTR_PROBE];
1357         bool ufid_present;
1358
1359         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1360         if (a[OVS_FLOW_ATTR_KEY]) {
1361                 ovs_match_init(&match, &key, true, NULL);
1362                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1363                                         NULL, log);
1364                 if (unlikely(err))
1365                         return err;
1366         }
1367
1368         ovs_lock();
1369         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1370         if (unlikely(!dp)) {
1371                 err = -ENODEV;
1372                 goto unlock;
1373         }
1374
1375         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1376                 err = ovs_flow_tbl_flush(&dp->table);
1377                 goto unlock;
1378         }
1379
1380         if (ufid_present)
1381                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1382         else
1383                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1384         if (unlikely(!flow)) {
1385                 err = -ENOENT;
1386                 goto unlock;
1387         }
1388
1389         ovs_flow_tbl_remove(&dp->table, flow);
1390         ovs_unlock();
1391
1392         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1393                                         &flow->id, info, false, ufid_flags);
1394         if (likely(reply)) {
1395                 if (!IS_ERR(reply)) {
1396                         rcu_read_lock();        /*To keep RCU checker happy. */
1397                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1398                                                      reply, info->snd_portid,
1399                                                      info->snd_seq, 0,
1400                                                      OVS_FLOW_CMD_DEL,
1401                                                      ufid_flags);
1402                         rcu_read_unlock();
1403                         if (WARN_ON_ONCE(err < 0)) {
1404                                 kfree_skb(reply);
1405                                 goto out_free;
1406                         }
1407
1408                         ovs_notify(&dp_flow_genl_family, reply, info);
1409                 } else {
1410                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1411                                         PTR_ERR(reply));
1412                 }
1413         }
1414
1415 out_free:
1416         ovs_flow_free(flow, true);
1417         return 0;
1418 unlock:
1419         ovs_unlock();
1420         return err;
1421 }
1422
1423 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1424 {
1425         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1426         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1427         struct table_instance *ti;
1428         struct datapath *dp;
1429         u32 ufid_flags;
1430         int err;
1431
1432         err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1433                                        OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1434         if (err)
1435                 return err;
1436         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1437
1438         rcu_read_lock();
1439         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1440         if (!dp) {
1441                 rcu_read_unlock();
1442                 return -ENODEV;
1443         }
1444
1445         ti = rcu_dereference(dp->table.ti);
1446         for (;;) {
1447                 struct sw_flow *flow;
1448                 u32 bucket, obj;
1449
1450                 bucket = cb->args[0];
1451                 obj = cb->args[1];
1452                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1453                 if (!flow)
1454                         break;
1455
1456                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1457                                            NETLINK_CB(cb->skb).portid,
1458                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1459                                            OVS_FLOW_CMD_GET, ufid_flags) < 0)
1460                         break;
1461
1462                 cb->args[0] = bucket;
1463                 cb->args[1] = obj;
1464         }
1465         rcu_read_unlock();
1466         return skb->len;
1467 }
1468
1469 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1470         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1471         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1472         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1473         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1474         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1475         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1476         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1477 };
1478
1479 static const struct genl_small_ops dp_flow_genl_ops[] = {
1480         { .cmd = OVS_FLOW_CMD_NEW,
1481           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1482           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1483           .doit = ovs_flow_cmd_new
1484         },
1485         { .cmd = OVS_FLOW_CMD_DEL,
1486           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1487           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1488           .doit = ovs_flow_cmd_del
1489         },
1490         { .cmd = OVS_FLOW_CMD_GET,
1491           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1492           .flags = 0,               /* OK for unprivileged users. */
1493           .doit = ovs_flow_cmd_get,
1494           .dumpit = ovs_flow_cmd_dump
1495         },
1496         { .cmd = OVS_FLOW_CMD_SET,
1497           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1498           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1499           .doit = ovs_flow_cmd_set,
1500         },
1501 };
1502
1503 static struct genl_family dp_flow_genl_family __ro_after_init = {
1504         .hdrsize = sizeof(struct ovs_header),
1505         .name = OVS_FLOW_FAMILY,
1506         .version = OVS_FLOW_VERSION,
1507         .maxattr = OVS_FLOW_ATTR_MAX,
1508         .policy = flow_policy,
1509         .netnsok = true,
1510         .parallel_ops = true,
1511         .small_ops = dp_flow_genl_ops,
1512         .n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
1513         .resv_start_op = OVS_FLOW_CMD_SET + 1,
1514         .mcgrps = &ovs_dp_flow_multicast_group,
1515         .n_mcgrps = 1,
1516         .module = THIS_MODULE,
1517 };
1518
1519 static size_t ovs_dp_cmd_msg_size(void)
1520 {
1521         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1522
1523         msgsize += nla_total_size(IFNAMSIZ);
1524         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1525         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1526         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1527         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1528         msgsize += nla_total_size(sizeof(u32) * nr_cpu_ids); /* OVS_DP_ATTR_PER_CPU_PIDS */
1529
1530         return msgsize;
1531 }
1532
1533 /* Called with ovs_mutex. */
1534 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1535                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1536 {
1537         struct ovs_header *ovs_header;
1538         struct ovs_dp_stats dp_stats;
1539         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1540         struct dp_nlsk_pids *pids = ovsl_dereference(dp->upcall_portids);
1541         int err, pids_len;
1542
1543         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1544                                  flags, cmd);
1545         if (!ovs_header)
1546                 goto error;
1547
1548         ovs_header->dp_ifindex = get_dpifindex(dp);
1549
1550         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1551         if (err)
1552                 goto nla_put_failure;
1553
1554         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1555         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1556                           &dp_stats, OVS_DP_ATTR_PAD))
1557                 goto nla_put_failure;
1558
1559         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1560                           sizeof(struct ovs_dp_megaflow_stats),
1561                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1562                 goto nla_put_failure;
1563
1564         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1565                 goto nla_put_failure;
1566
1567         if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1568                         ovs_flow_tbl_masks_cache_size(&dp->table)))
1569                 goto nla_put_failure;
1570
1571         if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU && pids) {
1572                 pids_len = min(pids->n_pids, nr_cpu_ids) * sizeof(u32);
1573                 if (nla_put(skb, OVS_DP_ATTR_PER_CPU_PIDS, pids_len, &pids->pids))
1574                         goto nla_put_failure;
1575         }
1576
1577         genlmsg_end(skb, ovs_header);
1578         return 0;
1579
1580 nla_put_failure:
1581         genlmsg_cancel(skb, ovs_header);
1582 error:
1583         return -EMSGSIZE;
1584 }
1585
1586 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1587 {
1588         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1589 }
1590
1591 /* Called with rcu_read_lock or ovs_mutex. */
1592 static struct datapath *lookup_datapath(struct net *net,
1593                                         const struct ovs_header *ovs_header,
1594                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1595 {
1596         struct datapath *dp;
1597
1598         if (!a[OVS_DP_ATTR_NAME])
1599                 dp = get_dp(net, ovs_header->dp_ifindex);
1600         else {
1601                 struct vport *vport;
1602
1603                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1604                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1605         }
1606         return dp ? dp : ERR_PTR(-ENODEV);
1607 }
1608
1609 static void ovs_dp_reset_user_features(struct sk_buff *skb,
1610                                        struct genl_info *info)
1611 {
1612         struct datapath *dp;
1613
1614         dp = lookup_datapath(sock_net(skb->sk), info->userhdr,
1615                              info->attrs);
1616         if (IS_ERR(dp))
1617                 return;
1618
1619         pr_warn("%s: Dropping previously announced user features\n",
1620                 ovs_dp_name(dp));
1621         dp->user_features = 0;
1622 }
1623
1624 static int ovs_dp_set_upcall_portids(struct datapath *dp,
1625                               const struct nlattr *ids)
1626 {
1627         struct dp_nlsk_pids *old, *dp_nlsk_pids;
1628
1629         if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
1630                 return -EINVAL;
1631
1632         old = ovsl_dereference(dp->upcall_portids);
1633
1634         dp_nlsk_pids = kmalloc(sizeof(*dp_nlsk_pids) + nla_len(ids),
1635                                GFP_KERNEL);
1636         if (!dp_nlsk_pids)
1637                 return -ENOMEM;
1638
1639         dp_nlsk_pids->n_pids = nla_len(ids) / sizeof(u32);
1640         nla_memcpy(dp_nlsk_pids->pids, ids, nla_len(ids));
1641
1642         rcu_assign_pointer(dp->upcall_portids, dp_nlsk_pids);
1643
1644         kfree_rcu(old, rcu);
1645
1646         return 0;
1647 }
1648
1649 u32 ovs_dp_get_upcall_portid(const struct datapath *dp, uint32_t cpu_id)
1650 {
1651         struct dp_nlsk_pids *dp_nlsk_pids;
1652
1653         dp_nlsk_pids = rcu_dereference(dp->upcall_portids);
1654
1655         if (dp_nlsk_pids) {
1656                 if (cpu_id < dp_nlsk_pids->n_pids) {
1657                         return dp_nlsk_pids->pids[cpu_id];
1658                 } else if (dp_nlsk_pids->n_pids > 0 &&
1659                            cpu_id >= dp_nlsk_pids->n_pids) {
1660                         /* If the number of netlink PIDs is mismatched with
1661                          * the number of CPUs as seen by the kernel, log this
1662                          * and send the upcall to an arbitrary socket (0) in
1663                          * order to not drop packets
1664                          */
1665                         pr_info_ratelimited("cpu_id mismatch with handler threads");
1666                         return dp_nlsk_pids->pids[cpu_id %
1667                                                   dp_nlsk_pids->n_pids];
1668                 } else {
1669                         return 0;
1670                 }
1671         } else {
1672                 return 0;
1673         }
1674 }
1675
1676 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1677 {
1678         u32 user_features = 0, old_features = dp->user_features;
1679         int err;
1680
1681         if (a[OVS_DP_ATTR_USER_FEATURES]) {
1682                 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1683
1684                 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1685                                       OVS_DP_F_UNALIGNED |
1686                                       OVS_DP_F_TC_RECIRC_SHARING |
1687                                       OVS_DP_F_DISPATCH_UPCALL_PER_CPU))
1688                         return -EOPNOTSUPP;
1689
1690 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1691                 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1692                         return -EOPNOTSUPP;
1693 #endif
1694         }
1695
1696         if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1697                 int err;
1698                 u32 cache_size;
1699
1700                 cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1701                 err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1702                 if (err)
1703                         return err;
1704         }
1705
1706         dp->user_features = user_features;
1707
1708         if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU &&
1709             a[OVS_DP_ATTR_PER_CPU_PIDS]) {
1710                 /* Upcall Netlink Port IDs have been updated */
1711                 err = ovs_dp_set_upcall_portids(dp,
1712                                                 a[OVS_DP_ATTR_PER_CPU_PIDS]);
1713                 if (err)
1714                         return err;
1715         }
1716
1717         if ((dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1718             !(old_features & OVS_DP_F_TC_RECIRC_SHARING))
1719                 tc_skb_ext_tc_enable();
1720         else if (!(dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1721                  (old_features & OVS_DP_F_TC_RECIRC_SHARING))
1722                 tc_skb_ext_tc_disable();
1723
1724         return 0;
1725 }
1726
1727 static int ovs_dp_stats_init(struct datapath *dp)
1728 {
1729         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1730         if (!dp->stats_percpu)
1731                 return -ENOMEM;
1732
1733         return 0;
1734 }
1735
1736 static int ovs_dp_vport_init(struct datapath *dp)
1737 {
1738         int i;
1739
1740         dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1741                                   sizeof(struct hlist_head),
1742                                   GFP_KERNEL);
1743         if (!dp->ports)
1744                 return -ENOMEM;
1745
1746         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1747                 INIT_HLIST_HEAD(&dp->ports[i]);
1748
1749         return 0;
1750 }
1751
1752 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1753 {
1754         struct nlattr **a = info->attrs;
1755         struct vport_parms parms;
1756         struct sk_buff *reply;
1757         struct datapath *dp;
1758         struct vport *vport;
1759         struct ovs_net *ovs_net;
1760         int err;
1761
1762         err = -EINVAL;
1763         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1764                 goto err;
1765
1766         reply = ovs_dp_cmd_alloc_info();
1767         if (!reply)
1768                 return -ENOMEM;
1769
1770         err = -ENOMEM;
1771         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1772         if (dp == NULL)
1773                 goto err_destroy_reply;
1774
1775         ovs_dp_set_net(dp, sock_net(skb->sk));
1776
1777         /* Allocate table. */
1778         err = ovs_flow_tbl_init(&dp->table);
1779         if (err)
1780                 goto err_destroy_dp;
1781
1782         err = ovs_dp_stats_init(dp);
1783         if (err)
1784                 goto err_destroy_table;
1785
1786         err = ovs_dp_vport_init(dp);
1787         if (err)
1788                 goto err_destroy_stats;
1789
1790         err = ovs_meters_init(dp);
1791         if (err)
1792                 goto err_destroy_ports;
1793
1794         /* Set up our datapath device. */
1795         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1796         parms.type = OVS_VPORT_TYPE_INTERNAL;
1797         parms.options = NULL;
1798         parms.dp = dp;
1799         parms.port_no = OVSP_LOCAL;
1800         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1801         parms.desired_ifindex = a[OVS_DP_ATTR_IFINDEX]
1802                 ? nla_get_u32(a[OVS_DP_ATTR_IFINDEX]) : 0;
1803
1804         /* So far only local changes have been made, now need the lock. */
1805         ovs_lock();
1806
1807         err = ovs_dp_change(dp, a);
1808         if (err)
1809                 goto err_unlock_and_destroy_meters;
1810
1811         vport = new_vport(&parms);
1812         if (IS_ERR(vport)) {
1813                 err = PTR_ERR(vport);
1814                 if (err == -EBUSY)
1815                         err = -EEXIST;
1816
1817                 if (err == -EEXIST) {
1818                         /* An outdated user space instance that does not understand
1819                          * the concept of user_features has attempted to create a new
1820                          * datapath and is likely to reuse it. Drop all user features.
1821                          */
1822                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1823                                 ovs_dp_reset_user_features(skb, info);
1824                 }
1825
1826                 goto err_destroy_portids;
1827         }
1828
1829         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1830                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1831         BUG_ON(err < 0);
1832
1833         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1834         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1835
1836         ovs_unlock();
1837
1838         ovs_notify(&dp_datapath_genl_family, reply, info);
1839         return 0;
1840
1841 err_destroy_portids:
1842         kfree(rcu_dereference_raw(dp->upcall_portids));
1843 err_unlock_and_destroy_meters:
1844         ovs_unlock();
1845         ovs_meters_exit(dp);
1846 err_destroy_ports:
1847         kfree(dp->ports);
1848 err_destroy_stats:
1849         free_percpu(dp->stats_percpu);
1850 err_destroy_table:
1851         ovs_flow_tbl_destroy(&dp->table);
1852 err_destroy_dp:
1853         kfree(dp);
1854 err_destroy_reply:
1855         kfree_skb(reply);
1856 err:
1857         return err;
1858 }
1859
1860 /* Called with ovs_mutex. */
1861 static void __dp_destroy(struct datapath *dp)
1862 {
1863         struct flow_table *table = &dp->table;
1864         int i;
1865
1866         if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1867                 tc_skb_ext_tc_disable();
1868
1869         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1870                 struct vport *vport;
1871                 struct hlist_node *n;
1872
1873                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1874                         if (vport->port_no != OVSP_LOCAL)
1875                                 ovs_dp_detach_port(vport);
1876         }
1877
1878         list_del_rcu(&dp->list_node);
1879
1880         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1881          * all ports in datapath are destroyed first before freeing datapath.
1882          */
1883         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1884
1885         /* Flush sw_flow in the tables. RCU cb only releases resource
1886          * such as dp, ports and tables. That may avoid some issues
1887          * such as RCU usage warning.
1888          */
1889         table_instance_flow_flush(table, ovsl_dereference(table->ti),
1890                                   ovsl_dereference(table->ufid_ti));
1891
1892         /* RCU destroy the ports, meters and flow tables. */
1893         call_rcu(&dp->rcu, destroy_dp_rcu);
1894 }
1895
1896 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1897 {
1898         struct sk_buff *reply;
1899         struct datapath *dp;
1900         int err;
1901
1902         reply = ovs_dp_cmd_alloc_info();
1903         if (!reply)
1904                 return -ENOMEM;
1905
1906         ovs_lock();
1907         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1908         err = PTR_ERR(dp);
1909         if (IS_ERR(dp))
1910                 goto err_unlock_free;
1911
1912         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1913                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1914         BUG_ON(err < 0);
1915
1916         __dp_destroy(dp);
1917         ovs_unlock();
1918
1919         ovs_notify(&dp_datapath_genl_family, reply, info);
1920
1921         return 0;
1922
1923 err_unlock_free:
1924         ovs_unlock();
1925         kfree_skb(reply);
1926         return err;
1927 }
1928
1929 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1930 {
1931         struct sk_buff *reply;
1932         struct datapath *dp;
1933         int err;
1934
1935         reply = ovs_dp_cmd_alloc_info();
1936         if (!reply)
1937                 return -ENOMEM;
1938
1939         ovs_lock();
1940         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1941         err = PTR_ERR(dp);
1942         if (IS_ERR(dp))
1943                 goto err_unlock_free;
1944
1945         err = ovs_dp_change(dp, info->attrs);
1946         if (err)
1947                 goto err_unlock_free;
1948
1949         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1950                                    info->snd_seq, 0, OVS_DP_CMD_SET);
1951         BUG_ON(err < 0);
1952
1953         ovs_unlock();
1954         ovs_notify(&dp_datapath_genl_family, reply, info);
1955
1956         return 0;
1957
1958 err_unlock_free:
1959         ovs_unlock();
1960         kfree_skb(reply);
1961         return err;
1962 }
1963
1964 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1965 {
1966         struct sk_buff *reply;
1967         struct datapath *dp;
1968         int err;
1969
1970         reply = ovs_dp_cmd_alloc_info();
1971         if (!reply)
1972                 return -ENOMEM;
1973
1974         ovs_lock();
1975         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1976         if (IS_ERR(dp)) {
1977                 err = PTR_ERR(dp);
1978                 goto err_unlock_free;
1979         }
1980         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1981                                    info->snd_seq, 0, OVS_DP_CMD_GET);
1982         BUG_ON(err < 0);
1983         ovs_unlock();
1984
1985         return genlmsg_reply(reply, info);
1986
1987 err_unlock_free:
1988         ovs_unlock();
1989         kfree_skb(reply);
1990         return err;
1991 }
1992
1993 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1994 {
1995         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1996         struct datapath *dp;
1997         int skip = cb->args[0];
1998         int i = 0;
1999
2000         ovs_lock();
2001         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2002                 if (i >= skip &&
2003                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
2004                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
2005                                          OVS_DP_CMD_GET) < 0)
2006                         break;
2007                 i++;
2008         }
2009         ovs_unlock();
2010
2011         cb->args[0] = i;
2012
2013         return skb->len;
2014 }
2015
2016 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
2017         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2018         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2019         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
2020         [OVS_DP_ATTR_MASKS_CACHE_SIZE] =  NLA_POLICY_RANGE(NLA_U32, 0,
2021                 PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
2022         [OVS_DP_ATTR_IFINDEX] = {.type = NLA_U32 },
2023 };
2024
2025 static const struct genl_small_ops dp_datapath_genl_ops[] = {
2026         { .cmd = OVS_DP_CMD_NEW,
2027           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2028           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2029           .doit = ovs_dp_cmd_new
2030         },
2031         { .cmd = OVS_DP_CMD_DEL,
2032           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2033           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2034           .doit = ovs_dp_cmd_del
2035         },
2036         { .cmd = OVS_DP_CMD_GET,
2037           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2038           .flags = 0,               /* OK for unprivileged users. */
2039           .doit = ovs_dp_cmd_get,
2040           .dumpit = ovs_dp_cmd_dump
2041         },
2042         { .cmd = OVS_DP_CMD_SET,
2043           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2044           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2045           .doit = ovs_dp_cmd_set,
2046         },
2047 };
2048
2049 static struct genl_family dp_datapath_genl_family __ro_after_init = {
2050         .hdrsize = sizeof(struct ovs_header),
2051         .name = OVS_DATAPATH_FAMILY,
2052         .version = OVS_DATAPATH_VERSION,
2053         .maxattr = OVS_DP_ATTR_MAX,
2054         .policy = datapath_policy,
2055         .netnsok = true,
2056         .parallel_ops = true,
2057         .small_ops = dp_datapath_genl_ops,
2058         .n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
2059         .resv_start_op = OVS_DP_CMD_SET + 1,
2060         .mcgrps = &ovs_dp_datapath_multicast_group,
2061         .n_mcgrps = 1,
2062         .module = THIS_MODULE,
2063 };
2064
2065 /* Called with ovs_mutex or RCU read lock. */
2066 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
2067                                    struct net *net, u32 portid, u32 seq,
2068                                    u32 flags, u8 cmd, gfp_t gfp)
2069 {
2070         struct ovs_header *ovs_header;
2071         struct ovs_vport_stats vport_stats;
2072         int err;
2073
2074         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
2075                                  flags, cmd);
2076         if (!ovs_header)
2077                 return -EMSGSIZE;
2078
2079         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
2080
2081         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
2082             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
2083             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
2084                            ovs_vport_name(vport)) ||
2085             nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
2086                 goto nla_put_failure;
2087
2088         if (!net_eq(net, dev_net(vport->dev))) {
2089                 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
2090
2091                 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
2092                         goto nla_put_failure;
2093         }
2094
2095         ovs_vport_get_stats(vport, &vport_stats);
2096         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
2097                           sizeof(struct ovs_vport_stats), &vport_stats,
2098                           OVS_VPORT_ATTR_PAD))
2099                 goto nla_put_failure;
2100
2101         if (ovs_vport_get_upcall_portids(vport, skb))
2102                 goto nla_put_failure;
2103
2104         err = ovs_vport_get_options(vport, skb);
2105         if (err == -EMSGSIZE)
2106                 goto error;
2107
2108         genlmsg_end(skb, ovs_header);
2109         return 0;
2110
2111 nla_put_failure:
2112         err = -EMSGSIZE;
2113 error:
2114         genlmsg_cancel(skb, ovs_header);
2115         return err;
2116 }
2117
2118 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2119 {
2120         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2121 }
2122
2123 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2124 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2125                                          u32 portid, u32 seq, u8 cmd)
2126 {
2127         struct sk_buff *skb;
2128         int retval;
2129
2130         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2131         if (!skb)
2132                 return ERR_PTR(-ENOMEM);
2133
2134         retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2135                                          GFP_KERNEL);
2136         BUG_ON(retval < 0);
2137
2138         return skb;
2139 }
2140
2141 /* Called with ovs_mutex or RCU read lock. */
2142 static struct vport *lookup_vport(struct net *net,
2143                                   const struct ovs_header *ovs_header,
2144                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2145 {
2146         struct datapath *dp;
2147         struct vport *vport;
2148
2149         if (a[OVS_VPORT_ATTR_IFINDEX])
2150                 return ERR_PTR(-EOPNOTSUPP);
2151         if (a[OVS_VPORT_ATTR_NAME]) {
2152                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2153                 if (!vport)
2154                         return ERR_PTR(-ENODEV);
2155                 if (ovs_header->dp_ifindex &&
2156                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2157                         return ERR_PTR(-ENODEV);
2158                 return vport;
2159         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2160                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2161
2162                 if (port_no >= DP_MAX_PORTS)
2163                         return ERR_PTR(-EFBIG);
2164
2165                 dp = get_dp(net, ovs_header->dp_ifindex);
2166                 if (!dp)
2167                         return ERR_PTR(-ENODEV);
2168
2169                 vport = ovs_vport_ovsl_rcu(dp, port_no);
2170                 if (!vport)
2171                         return ERR_PTR(-ENODEV);
2172                 return vport;
2173         } else
2174                 return ERR_PTR(-EINVAL);
2175
2176 }
2177
2178 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2179 {
2180         unsigned int dev_headroom, max_headroom = 0;
2181         struct net_device *dev;
2182         struct vport *vport;
2183         int i;
2184
2185         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2186                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2187                                          lockdep_ovsl_is_held()) {
2188                         dev = vport->dev;
2189                         dev_headroom = netdev_get_fwd_headroom(dev);
2190                         if (dev_headroom > max_headroom)
2191                                 max_headroom = dev_headroom;
2192                 }
2193         }
2194
2195         return max_headroom;
2196 }
2197
2198 /* Called with ovs_mutex */
2199 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2200 {
2201         struct vport *vport;
2202         int i;
2203
2204         dp->max_headroom = new_headroom;
2205         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2206                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2207                                          lockdep_ovsl_is_held())
2208                         netdev_set_rx_headroom(vport->dev, new_headroom);
2209         }
2210 }
2211
2212 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2213 {
2214         struct nlattr **a = info->attrs;
2215         struct ovs_header *ovs_header = info->userhdr;
2216         struct vport_parms parms;
2217         struct sk_buff *reply;
2218         struct vport *vport;
2219         struct datapath *dp;
2220         unsigned int new_headroom;
2221         u32 port_no;
2222         int err;
2223
2224         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2225             !a[OVS_VPORT_ATTR_UPCALL_PID])
2226                 return -EINVAL;
2227
2228         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2229
2230         if (a[OVS_VPORT_ATTR_IFINDEX] && parms.type != OVS_VPORT_TYPE_INTERNAL)
2231                 return -EOPNOTSUPP;
2232
2233         port_no = a[OVS_VPORT_ATTR_PORT_NO]
2234                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2235         if (port_no >= DP_MAX_PORTS)
2236                 return -EFBIG;
2237
2238         reply = ovs_vport_cmd_alloc_info();
2239         if (!reply)
2240                 return -ENOMEM;
2241
2242         ovs_lock();
2243 restart:
2244         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2245         err = -ENODEV;
2246         if (!dp)
2247                 goto exit_unlock_free;
2248
2249         if (port_no) {
2250                 vport = ovs_vport_ovsl(dp, port_no);
2251                 err = -EBUSY;
2252                 if (vport)
2253                         goto exit_unlock_free;
2254         } else {
2255                 for (port_no = 1; ; port_no++) {
2256                         if (port_no >= DP_MAX_PORTS) {
2257                                 err = -EFBIG;
2258                                 goto exit_unlock_free;
2259                         }
2260                         vport = ovs_vport_ovsl(dp, port_no);
2261                         if (!vport)
2262                                 break;
2263                 }
2264         }
2265
2266         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2267         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2268         parms.dp = dp;
2269         parms.port_no = port_no;
2270         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2271         parms.desired_ifindex = a[OVS_VPORT_ATTR_IFINDEX]
2272                 ? nla_get_u32(a[OVS_VPORT_ATTR_IFINDEX]) : 0;
2273
2274         vport = new_vport(&parms);
2275         err = PTR_ERR(vport);
2276         if (IS_ERR(vport)) {
2277                 if (err == -EAGAIN)
2278                         goto restart;
2279                 goto exit_unlock_free;
2280         }
2281
2282         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2283                                       info->snd_portid, info->snd_seq, 0,
2284                                       OVS_VPORT_CMD_NEW, GFP_KERNEL);
2285
2286         new_headroom = netdev_get_fwd_headroom(vport->dev);
2287
2288         if (new_headroom > dp->max_headroom)
2289                 ovs_update_headroom(dp, new_headroom);
2290         else
2291                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2292
2293         BUG_ON(err < 0);
2294         ovs_unlock();
2295
2296         ovs_notify(&dp_vport_genl_family, reply, info);
2297         return 0;
2298
2299 exit_unlock_free:
2300         ovs_unlock();
2301         kfree_skb(reply);
2302         return err;
2303 }
2304
2305 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2306 {
2307         struct nlattr **a = info->attrs;
2308         struct sk_buff *reply;
2309         struct vport *vport;
2310         int err;
2311
2312         reply = ovs_vport_cmd_alloc_info();
2313         if (!reply)
2314                 return -ENOMEM;
2315
2316         ovs_lock();
2317         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2318         err = PTR_ERR(vport);
2319         if (IS_ERR(vport))
2320                 goto exit_unlock_free;
2321
2322         if (a[OVS_VPORT_ATTR_TYPE] &&
2323             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2324                 err = -EINVAL;
2325                 goto exit_unlock_free;
2326         }
2327
2328         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2329                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2330                 if (err)
2331                         goto exit_unlock_free;
2332         }
2333
2334
2335         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2336                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2337
2338                 err = ovs_vport_set_upcall_portids(vport, ids);
2339                 if (err)
2340                         goto exit_unlock_free;
2341         }
2342
2343         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2344                                       info->snd_portid, info->snd_seq, 0,
2345                                       OVS_VPORT_CMD_SET, GFP_KERNEL);
2346         BUG_ON(err < 0);
2347
2348         ovs_unlock();
2349         ovs_notify(&dp_vport_genl_family, reply, info);
2350         return 0;
2351
2352 exit_unlock_free:
2353         ovs_unlock();
2354         kfree_skb(reply);
2355         return err;
2356 }
2357
2358 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2359 {
2360         bool update_headroom = false;
2361         struct nlattr **a = info->attrs;
2362         struct sk_buff *reply;
2363         struct datapath *dp;
2364         struct vport *vport;
2365         unsigned int new_headroom;
2366         int err;
2367
2368         reply = ovs_vport_cmd_alloc_info();
2369         if (!reply)
2370                 return -ENOMEM;
2371
2372         ovs_lock();
2373         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2374         err = PTR_ERR(vport);
2375         if (IS_ERR(vport))
2376                 goto exit_unlock_free;
2377
2378         if (vport->port_no == OVSP_LOCAL) {
2379                 err = -EINVAL;
2380                 goto exit_unlock_free;
2381         }
2382
2383         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2384                                       info->snd_portid, info->snd_seq, 0,
2385                                       OVS_VPORT_CMD_DEL, GFP_KERNEL);
2386         BUG_ON(err < 0);
2387
2388         /* the vport deletion may trigger dp headroom update */
2389         dp = vport->dp;
2390         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2391                 update_headroom = true;
2392
2393         netdev_reset_rx_headroom(vport->dev);
2394         ovs_dp_detach_port(vport);
2395
2396         if (update_headroom) {
2397                 new_headroom = ovs_get_max_headroom(dp);
2398
2399                 if (new_headroom < dp->max_headroom)
2400                         ovs_update_headroom(dp, new_headroom);
2401         }
2402         ovs_unlock();
2403
2404         ovs_notify(&dp_vport_genl_family, reply, info);
2405         return 0;
2406
2407 exit_unlock_free:
2408         ovs_unlock();
2409         kfree_skb(reply);
2410         return err;
2411 }
2412
2413 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2414 {
2415         struct nlattr **a = info->attrs;
2416         struct ovs_header *ovs_header = info->userhdr;
2417         struct sk_buff *reply;
2418         struct vport *vport;
2419         int err;
2420
2421         reply = ovs_vport_cmd_alloc_info();
2422         if (!reply)
2423                 return -ENOMEM;
2424
2425         rcu_read_lock();
2426         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2427         err = PTR_ERR(vport);
2428         if (IS_ERR(vport))
2429                 goto exit_unlock_free;
2430         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2431                                       info->snd_portid, info->snd_seq, 0,
2432                                       OVS_VPORT_CMD_GET, GFP_ATOMIC);
2433         BUG_ON(err < 0);
2434         rcu_read_unlock();
2435
2436         return genlmsg_reply(reply, info);
2437
2438 exit_unlock_free:
2439         rcu_read_unlock();
2440         kfree_skb(reply);
2441         return err;
2442 }
2443
2444 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2445 {
2446         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2447         struct datapath *dp;
2448         int bucket = cb->args[0], skip = cb->args[1];
2449         int i, j = 0;
2450
2451         rcu_read_lock();
2452         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2453         if (!dp) {
2454                 rcu_read_unlock();
2455                 return -ENODEV;
2456         }
2457         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2458                 struct vport *vport;
2459
2460                 j = 0;
2461                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2462                         if (j >= skip &&
2463                             ovs_vport_cmd_fill_info(vport, skb,
2464                                                     sock_net(skb->sk),
2465                                                     NETLINK_CB(cb->skb).portid,
2466                                                     cb->nlh->nlmsg_seq,
2467                                                     NLM_F_MULTI,
2468                                                     OVS_VPORT_CMD_GET,
2469                                                     GFP_ATOMIC) < 0)
2470                                 goto out;
2471
2472                         j++;
2473                 }
2474                 skip = 0;
2475         }
2476 out:
2477         rcu_read_unlock();
2478
2479         cb->args[0] = i;
2480         cb->args[1] = j;
2481
2482         return skb->len;
2483 }
2484
2485 static void ovs_dp_masks_rebalance(struct work_struct *work)
2486 {
2487         struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2488                                                masks_rebalance.work);
2489         struct datapath *dp;
2490
2491         ovs_lock();
2492
2493         list_for_each_entry(dp, &ovs_net->dps, list_node)
2494                 ovs_flow_masks_rebalance(&dp->table);
2495
2496         ovs_unlock();
2497
2498         schedule_delayed_work(&ovs_net->masks_rebalance,
2499                               msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2500 }
2501
2502 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2503         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2504         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2505         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2506         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2507         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2508         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2509         [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2510         [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2511 };
2512
2513 static const struct genl_small_ops dp_vport_genl_ops[] = {
2514         { .cmd = OVS_VPORT_CMD_NEW,
2515           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2516           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2517           .doit = ovs_vport_cmd_new
2518         },
2519         { .cmd = OVS_VPORT_CMD_DEL,
2520           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2521           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2522           .doit = ovs_vport_cmd_del
2523         },
2524         { .cmd = OVS_VPORT_CMD_GET,
2525           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2526           .flags = 0,               /* OK for unprivileged users. */
2527           .doit = ovs_vport_cmd_get,
2528           .dumpit = ovs_vport_cmd_dump
2529         },
2530         { .cmd = OVS_VPORT_CMD_SET,
2531           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2532           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2533           .doit = ovs_vport_cmd_set,
2534         },
2535 };
2536
2537 struct genl_family dp_vport_genl_family __ro_after_init = {
2538         .hdrsize = sizeof(struct ovs_header),
2539         .name = OVS_VPORT_FAMILY,
2540         .version = OVS_VPORT_VERSION,
2541         .maxattr = OVS_VPORT_ATTR_MAX,
2542         .policy = vport_policy,
2543         .netnsok = true,
2544         .parallel_ops = true,
2545         .small_ops = dp_vport_genl_ops,
2546         .n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
2547         .resv_start_op = OVS_VPORT_CMD_SET + 1,
2548         .mcgrps = &ovs_dp_vport_multicast_group,
2549         .n_mcgrps = 1,
2550         .module = THIS_MODULE,
2551 };
2552
2553 static struct genl_family * const dp_genl_families[] = {
2554         &dp_datapath_genl_family,
2555         &dp_vport_genl_family,
2556         &dp_flow_genl_family,
2557         &dp_packet_genl_family,
2558         &dp_meter_genl_family,
2559 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2560         &dp_ct_limit_genl_family,
2561 #endif
2562 };
2563
2564 static void dp_unregister_genl(int n_families)
2565 {
2566         int i;
2567
2568         for (i = 0; i < n_families; i++)
2569                 genl_unregister_family(dp_genl_families[i]);
2570 }
2571
2572 static int __init dp_register_genl(void)
2573 {
2574         int err;
2575         int i;
2576
2577         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2578
2579                 err = genl_register_family(dp_genl_families[i]);
2580                 if (err)
2581                         goto error;
2582         }
2583
2584         return 0;
2585
2586 error:
2587         dp_unregister_genl(i);
2588         return err;
2589 }
2590
2591 static int __net_init ovs_init_net(struct net *net)
2592 {
2593         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2594         int err;
2595
2596         INIT_LIST_HEAD(&ovs_net->dps);
2597         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2598         INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2599
2600         err = ovs_ct_init(net);
2601         if (err)
2602                 return err;
2603
2604         schedule_delayed_work(&ovs_net->masks_rebalance,
2605                               msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2606         return 0;
2607 }
2608
2609 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2610                                             struct list_head *head)
2611 {
2612         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2613         struct datapath *dp;
2614
2615         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2616                 int i;
2617
2618                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2619                         struct vport *vport;
2620
2621                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2622                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2623                                         continue;
2624
2625                                 if (dev_net(vport->dev) == dnet)
2626                                         list_add(&vport->detach_list, head);
2627                         }
2628                 }
2629         }
2630 }
2631
2632 static void __net_exit ovs_exit_net(struct net *dnet)
2633 {
2634         struct datapath *dp, *dp_next;
2635         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2636         struct vport *vport, *vport_next;
2637         struct net *net;
2638         LIST_HEAD(head);
2639
2640         ovs_lock();
2641
2642         ovs_ct_exit(dnet);
2643
2644         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2645                 __dp_destroy(dp);
2646
2647         down_read(&net_rwsem);
2648         for_each_net(net)
2649                 list_vports_from_net(net, dnet, &head);
2650         up_read(&net_rwsem);
2651
2652         /* Detach all vports from given namespace. */
2653         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2654                 list_del(&vport->detach_list);
2655                 ovs_dp_detach_port(vport);
2656         }
2657
2658         ovs_unlock();
2659
2660         cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2661         cancel_work_sync(&ovs_net->dp_notify_work);
2662 }
2663
2664 static struct pernet_operations ovs_net_ops = {
2665         .init = ovs_init_net,
2666         .exit = ovs_exit_net,
2667         .id   = &ovs_net_id,
2668         .size = sizeof(struct ovs_net),
2669 };
2670
2671 static int __init dp_init(void)
2672 {
2673         int err;
2674
2675         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2676                      sizeof_field(struct sk_buff, cb));
2677
2678         pr_info("Open vSwitch switching datapath\n");
2679
2680         err = action_fifos_init();
2681         if (err)
2682                 goto error;
2683
2684         err = ovs_internal_dev_rtnl_link_register();
2685         if (err)
2686                 goto error_action_fifos_exit;
2687
2688         err = ovs_flow_init();
2689         if (err)
2690                 goto error_unreg_rtnl_link;
2691
2692         err = ovs_vport_init();
2693         if (err)
2694                 goto error_flow_exit;
2695
2696         err = register_pernet_device(&ovs_net_ops);
2697         if (err)
2698                 goto error_vport_exit;
2699
2700         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2701         if (err)
2702                 goto error_netns_exit;
2703
2704         err = ovs_netdev_init();
2705         if (err)
2706                 goto error_unreg_notifier;
2707
2708         err = dp_register_genl();
2709         if (err < 0)
2710                 goto error_unreg_netdev;
2711
2712         return 0;
2713
2714 error_unreg_netdev:
2715         ovs_netdev_exit();
2716 error_unreg_notifier:
2717         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2718 error_netns_exit:
2719         unregister_pernet_device(&ovs_net_ops);
2720 error_vport_exit:
2721         ovs_vport_exit();
2722 error_flow_exit:
2723         ovs_flow_exit();
2724 error_unreg_rtnl_link:
2725         ovs_internal_dev_rtnl_link_unregister();
2726 error_action_fifos_exit:
2727         action_fifos_exit();
2728 error:
2729         return err;
2730 }
2731
2732 static void dp_cleanup(void)
2733 {
2734         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2735         ovs_netdev_exit();
2736         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2737         unregister_pernet_device(&ovs_net_ops);
2738         rcu_barrier();
2739         ovs_vport_exit();
2740         ovs_flow_exit();
2741         ovs_internal_dev_rtnl_link_unregister();
2742         action_fifos_exit();
2743 }
2744
2745 module_init(dp_init);
2746 module_exit(dp_cleanup);
2747
2748 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2749 MODULE_LICENSE("GPL");
2750 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2751 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2752 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2753 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2754 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2755 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);