net/openvswitch/datapath.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2007-2014 Nicira, Inc.
   4  */
   5
   6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   7
   8 #include <linux/init.h>
   9 #include <linux/module.h>
  10 #include <linux/if_arp.h>
  11 #include <linux/if_vlan.h>
  12 #include <linux/in.h>
  13 #include <linux/ip.h>
  14 #include <linux/jhash.h>
  15 #include <linux/delay.h>
  16 #include <linux/time.h>
  17 #include <linux/etherdevice.h>
  18 #include <linux/genetlink.h>
  19 #include <linux/kernel.h>
  20 #include <linux/kthread.h>
  21 #include <linux/mutex.h>
  22 #include <linux/percpu.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/tcp.h>
  25 #include <linux/udp.h>
  26 #include <linux/ethtool.h>
  27 #include <linux/wait.h>
  28 #include <asm/div64.h>
  29 #include <linux/highmem.h>
  30 #include <linux/netfilter_bridge.h>
  31 #include <linux/netfilter_ipv4.h>
  32 #include <linux/inetdevice.h>
  33 #include <linux/list.h>
  34 #include <linux/openvswitch.h>
  35 #include <linux/rculist.h>
  36 #include <linux/dmi.h>
  37 #include <net/genetlink.h>
  38 #include <net/net_namespace.h>
  39 #include <net/netns/generic.h>
  40
  41 #include "datapath.h"
  42 #include "flow.h"
  43 #include "flow_table.h"
  44 #include "flow_netlink.h"
  45 #include "meter.h"
  46 #include "openvswitch_trace.h"
  47 #include "vport-internal_dev.h"
  48 #include "vport-netdev.h"
  49
  50 unsigned int ovs_net_id __read_mostly;
  51
  52 static struct genl_family dp_packet_genl_family;
  53 static struct genl_family dp_flow_genl_family;
  54 static struct genl_family dp_datapath_genl_family;
  55
  56 static const struct nla_policy flow_policy[];
  57
  58 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
  59         .name = OVS_FLOW_MCGROUP,
  60 };
  61
  62 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  63         .name = OVS_DATAPATH_MCGROUP,
  64 };
  65
  66 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
  67         .name = OVS_VPORT_MCGROUP,
  68 };
  69
  70 /* Check if need to build a reply message.
  71  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  72 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  73                             unsigned int group)
  74 {
  75         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  76                genl_has_listeners(family, genl_info_net(info), group);
  77 }
  78
  79 static void ovs_notify(struct genl_family *family,
  80                        struct sk_buff *skb, struct genl_info *info)
  81 {
  82         genl_notify(family, skb, info, 0, GFP_KERNEL);
  83 }
  84
  85 /**
  86  * DOC: Locking:
  87  *
  88  * All writes e.g. Writes to device state (add/remove datapath, port, set
  89  * operations on vports, etc.), Writes to other state (flow table
  90  * modifications, set miscellaneous datapath parameters, etc.) are protected
  91  * by ovs_lock.
  92  *
  93  * Reads are protected by RCU.
  94  *
  95  * There are a few special cases (mostly stats) that have their own
  96  * synchronization but they nest under all of above and don't interact with
  97  * each other.
  98  *
  99  * The RTNL lock nests inside ovs_mutex.
 100  */
 101
 102 static DEFINE_MUTEX(ovs_mutex);
 103
 104 void ovs_lock(void)
 105 {
 106         mutex_lock(&ovs_mutex);
 107 }
 108
 109 void ovs_unlock(void)
 110 {
 111         mutex_unlock(&ovs_mutex);
 112 }
 113
 114 #ifdef CONFIG_LOCKDEP
 115 int lockdep_ovsl_is_held(void)
 116 {
 117         if (debug_locks)
 118                 return lockdep_is_held(&ovs_mutex);
 119         else
 120                 return 1;
 121 }
 122 #endif
 123
 124 static struct vport *new_vport(const struct vport_parms *);
 125 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 126                              const struct sw_flow_key *,
 127                              const struct dp_upcall_info *,
 128                              uint32_t cutlen);
 129 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 130                                   const struct sw_flow_key *,
 131                                   const struct dp_upcall_info *,
 132                                   uint32_t cutlen);
 133
 134 static void ovs_dp_masks_rebalance(struct work_struct *work);
 135
 136 static int ovs_dp_set_upcall_portids(struct datapath *, const struct nlattr *);
 137
 138 /* Must be called with rcu_read_lock or ovs_mutex. */
 139 const char *ovs_dp_name(const struct datapath *dp)
 140 {
 141         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 142         return ovs_vport_name(vport);
 143 }
 144
 145 static int get_dpifindex(const struct datapath *dp)
 146 {
 147         struct vport *local;
 148         int ifindex;
 149
 150         rcu_read_lock();
 151
 152         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 153         if (local)
 154                 ifindex = local->dev->ifindex;
 155         else
 156                 ifindex = 0;
 157
 158         rcu_read_unlock();
 159
 160         return ifindex;
 161 }
 162
 163 static void destroy_dp_rcu(struct rcu_head *rcu)
 164 {
 165         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 166
 167         ovs_flow_tbl_destroy(&dp->table);
 168         free_percpu(dp->stats_percpu);
 169         kfree(dp->ports);
 170         ovs_meters_exit(dp);
 171         kfree(rcu_dereference_raw(dp->upcall_portids));
 172         kfree(dp);
 173 }
 174
 175 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 176                                             u16 port_no)
 177 {
 178         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 179 }
 180
 181 /* Called with ovs_mutex or RCU read lock. */
 182 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 183 {
 184         struct vport *vport;
 185         struct hlist_head *head;
 186
 187         head = vport_hash_bucket(dp, port_no);
 188         hlist_for_each_entry_rcu(vport, head, dp_hash_node,
 189                                  lockdep_ovsl_is_held()) {
 190                 if (vport->port_no == port_no)
 191                         return vport;
 192         }
 193         return NULL;
 194 }
 195
 196 /* Called with ovs_mutex. */
 197 static struct vport *new_vport(const struct vport_parms *parms)
 198 {
 199         struct vport *vport;
 200
 201         vport = ovs_vport_add(parms);
 202         if (!IS_ERR(vport)) {
 203                 struct datapath *dp = parms->dp;
 204                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 205
 206                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 207         }
 208         return vport;
 209 }
 210
 211 void ovs_dp_detach_port(struct vport *p)
 212 {
 213         ASSERT_OVSL();
 214
 215         /* First drop references to device. */
 216         hlist_del_rcu(&p->dp_hash_node);
 217
 218         /* Then destroy it. */
 219         ovs_vport_del(p);
 220 }
 221
 222 /* Must be called with rcu_read_lock. */
 223 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 224 {
 225         const struct vport *p = OVS_CB(skb)->input_vport;
 226         struct datapath *dp = p->dp;
 227         struct sw_flow *flow;
 228         struct sw_flow_actions *sf_acts;
 229         struct dp_stats_percpu *stats;
 230         u64 *stats_counter;
 231         u32 n_mask_hit;
 232         u32 n_cache_hit;
 233         int error;
 234
 235         stats = this_cpu_ptr(dp->stats_percpu);
 236
 237         /* Look up flow. */
 238         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
 239                                          &n_mask_hit, &n_cache_hit);
 240         if (unlikely(!flow)) {
 241                 struct dp_upcall_info upcall;
 242
 243                 memset(&upcall, 0, sizeof(upcall));
 244                 upcall.cmd = OVS_PACKET_CMD_MISS;
 245
 246                 if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
 247                         upcall.portid =
 248                             ovs_dp_get_upcall_portid(dp, smp_processor_id());
 249                 else
 250                         upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 251
 252                 upcall.mru = OVS_CB(skb)->mru;
 253                 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
 254                 switch (error) {
 255                 case 0:
 256                 case -EAGAIN:
 257                 case -ERESTARTSYS:
 258                 case -EINTR:
 259                         consume_skb(skb);
 260                         break;
 261                 default:
 262                         kfree_skb(skb);
 263                         break;
 264                 }
 265                 stats_counter = &stats->n_missed;
 266                 goto out;
 267         }
 268
 269         ovs_flow_stats_update(flow, key->tp.flags, skb);
 270         sf_acts = rcu_dereference(flow->sf_acts);
 271         error = ovs_execute_actions(dp, skb, sf_acts, key);
 272         if (unlikely(error))
 273                 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
 274                                     ovs_dp_name(dp), error);
 275
 276         stats_counter = &stats->n_hit;
 277
 278 out:
 279         /* Update datapath statistics. */
 280         u64_stats_update_begin(&stats->syncp);
 281         (*stats_counter)++;
 282         stats->n_mask_hit += n_mask_hit;
 283         stats->n_cache_hit += n_cache_hit;
 284         u64_stats_update_end(&stats->syncp);
 285 }
 286
 287 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 288                   const struct sw_flow_key *key,
 289                   const struct dp_upcall_info *upcall_info,
 290                   uint32_t cutlen)
 291 {
 292         struct dp_stats_percpu *stats;
 293         int err;
 294
 295         if (trace_ovs_dp_upcall_enabled())
 296                 trace_ovs_dp_upcall(dp, skb, key, upcall_info);
 297
 298         if (upcall_info->portid == 0) {
 299                 err = -ENOTCONN;
 300                 goto err;
 301         }
 302
 303         if (!skb_is_gso(skb))
 304                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 305         else
 306                 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
 307         if (err)
 308                 goto err;
 309
 310         return 0;
 311
 312 err:
 313         stats = this_cpu_ptr(dp->stats_percpu);
 314
 315         u64_stats_update_begin(&stats->syncp);
 316         stats->n_lost++;
 317         u64_stats_update_end(&stats->syncp);
 318
 319         return err;
 320 }
 321
 322 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 323                              const struct sw_flow_key *key,
 324                              const struct dp_upcall_info *upcall_info,
 325                              uint32_t cutlen)
 326 {
 327         unsigned int gso_type = skb_shinfo(skb)->gso_type;
 328         struct sw_flow_key later_key;
 329         struct sk_buff *segs, *nskb;
 330         int err;
 331
 332         BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_GSO_CB_OFFSET);
 333         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 334         if (IS_ERR(segs))
 335                 return PTR_ERR(segs);
 336         if (segs == NULL)
 337                 return -EINVAL;
 338
 339         if (gso_type & SKB_GSO_UDP) {
 340                 /* The initial flow key extracted by ovs_flow_key_extract()
 341                  * in this case is for a first fragment, so we need to
 342                  * properly mark later fragments.
 343                  */
 344                 later_key = *key;
 345                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 346         }
 347
 348         /* Queue all of the segments. */
 349         skb_list_walk_safe(segs, skb, nskb) {
 350                 if (gso_type & SKB_GSO_UDP && skb != segs)
 351                         key = &later_key;
 352
 353                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 354                 if (err)
 355                         break;
 356
 357         }
 358
 359         /* Free all of the segments. */
 360         skb_list_walk_safe(segs, skb, nskb) {
 361                 if (err)
 362                         kfree_skb(skb);
 363                 else
 364                         consume_skb(skb);
 365         }
 366         return err;
 367 }
 368
 369 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 370                               unsigned int hdrlen, int actions_attrlen)
 371 {
 372         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 373                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 374                 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 375                 + nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
 376                 + nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
 377
 378         /* OVS_PACKET_ATTR_USERDATA */
 379         if (upcall_info->userdata)
 380                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 381
 382         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 383         if (upcall_info->egress_tun_info)
 384                 size += nla_total_size(ovs_tun_key_attr_size());
 385
 386         /* OVS_PACKET_ATTR_ACTIONS */
 387         if (upcall_info->actions_len)
 388                 size += nla_total_size(actions_attrlen);
 389
 390         /* OVS_PACKET_ATTR_MRU */
 391         if (upcall_info->mru)
 392                 size += nla_total_size(sizeof(upcall_info->mru));
 393
 394         return size;
 395 }
 396
 397 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 398 {
 399         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 400                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 401
 402                 if (plen > 0)
 403                         skb_put_zero(skb, plen);
 404         }
 405 }
 406
 407 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 408                                   const struct sw_flow_key *key,
 409                                   const struct dp_upcall_info *upcall_info,
 410                                   uint32_t cutlen)
 411 {
 412         struct ovs_header *upcall;
 413         struct sk_buff *nskb = NULL;
 414         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 415         struct nlattr *nla;
 416         size_t len;
 417         unsigned int hlen;
 418         int err, dp_ifindex;
 419         u64 hash;
 420
 421         dp_ifindex = get_dpifindex(dp);
 422         if (!dp_ifindex)
 423                 return -ENODEV;
 424
 425         if (skb_vlan_tag_present(skb)) {
 426                 nskb = skb_clone(skb, GFP_ATOMIC);
 427                 if (!nskb)
 428                         return -ENOMEM;
 429
 430                 nskb = __vlan_hwaccel_push_inside(nskb);
 431                 if (!nskb)
 432                         return -ENOMEM;
 433
 434                 skb = nskb;
 435         }
 436
 437         if (nla_attr_size(skb->len) > USHRT_MAX) {
 438                 err = -EFBIG;
 439                 goto out;
 440         }
 441
 442         /* Complete checksum if needed */
 443         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 444             (err = skb_csum_hwoffload_help(skb, 0)))
 445                 goto out;
 446
 447         /* Older versions of OVS user space enforce alignment of the last
 448          * Netlink attribute to NLA_ALIGNTO which would require extensive
 449          * padding logic. Only perform zerocopy if padding is not required.
 450          */
 451         if (dp->user_features & OVS_DP_F_UNALIGNED)
 452                 hlen = skb_zerocopy_headlen(skb);
 453         else
 454                 hlen = skb->len;
 455
 456         len = upcall_msg_size(upcall_info, hlen - cutlen,
 457                               OVS_CB(skb)->acts_origlen);
 458         user_skb = genlmsg_new(len, GFP_ATOMIC);
 459         if (!user_skb) {
 460                 err = -ENOMEM;
 461                 goto out;
 462         }
 463
 464         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 465                              0, upcall_info->cmd);
 466         if (!upcall) {
 467                 err = -EINVAL;
 468                 goto out;
 469         }
 470         upcall->dp_ifindex = dp_ifindex;
 471
 472         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 473         if (err)
 474                 goto out;
 475
 476         if (upcall_info->userdata)
 477                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 478                           nla_len(upcall_info->userdata),
 479                           nla_data(upcall_info->userdata));
 480
 481         if (upcall_info->egress_tun_info) {
 482                 nla = nla_nest_start_noflag(user_skb,
 483                                             OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 484                 if (!nla) {
 485                         err = -EMSGSIZE;
 486                         goto out;
 487                 }
 488                 err = ovs_nla_put_tunnel_info(user_skb,
 489                                               upcall_info->egress_tun_info);
 490                 if (err)
 491                         goto out;
 492
 493                 nla_nest_end(user_skb, nla);
 494         }
 495
 496         if (upcall_info->actions_len) {
 497                 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
 498                 if (!nla) {
 499                         err = -EMSGSIZE;
 500                         goto out;
 501                 }
 502                 err = ovs_nla_put_actions(upcall_info->actions,
 503                                           upcall_info->actions_len,
 504                                           user_skb);
 505                 if (!err)
 506                         nla_nest_end(user_skb, nla);
 507                 else
 508                         nla_nest_cancel(user_skb, nla);
 509         }
 510
 511         /* Add OVS_PACKET_ATTR_MRU */
 512         if (upcall_info->mru &&
 513             nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
 514                 err = -ENOBUFS;
 515                 goto out;
 516         }
 517
 518         /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 519         if (cutlen > 0 &&
 520             nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
 521                 err = -ENOBUFS;
 522                 goto out;
 523         }
 524
 525         /* Add OVS_PACKET_ATTR_HASH */
 526         hash = skb_get_hash_raw(skb);
 527         if (skb->sw_hash)
 528                 hash |= OVS_PACKET_HASH_SW_BIT;
 529
 530         if (skb->l4_hash)
 531                 hash |= OVS_PACKET_HASH_L4_BIT;
 532
 533         if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
 534                 err = -ENOBUFS;
 535                 goto out;
 536         }
 537
 538         /* Only reserve room for attribute header, packet data is added
 539          * in skb_zerocopy() */
 540         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 541                 err = -ENOBUFS;
 542                 goto out;
 543         }
 544         nla->nla_len = nla_attr_size(skb->len - cutlen);
 545
 546         err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 547         if (err)
 548                 goto out;
 549
 550         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 551         pad_packet(dp, user_skb);
 552
 553         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 554
 555         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 556         user_skb = NULL;
 557 out:
 558         if (err)
 559                 skb_tx_error(skb);
 560         consume_skb(user_skb);
 561         consume_skb(nskb);
 562
 563         return err;
 564 }
 565
 566 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 567 {
 568         struct ovs_header *ovs_header = info->userhdr;
 569         struct net *net = sock_net(skb->sk);
 570         struct nlattr **a = info->attrs;
 571         struct sw_flow_actions *acts;
 572         struct sk_buff *packet;
 573         struct sw_flow *flow;
 574         struct sw_flow_actions *sf_acts;
 575         struct datapath *dp;
 576         struct vport *input_vport;
 577         u16 mru = 0;
 578         u64 hash;
 579         int len;
 580         int err;
 581         bool log = !a[OVS_PACKET_ATTR_PROBE];
 582
 583         err = -EINVAL;
 584         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 585             !a[OVS_PACKET_ATTR_ACTIONS])
 586                 goto err;
 587
 588         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 589         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 590         err = -ENOMEM;
 591         if (!packet)
 592                 goto err;
 593         skb_reserve(packet, NET_IP_ALIGN);
 594
 595         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 596
 597         /* Set packet's mru */
 598         if (a[OVS_PACKET_ATTR_MRU]) {
 599                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 600                 packet->ignore_df = 1;
 601         }
 602         OVS_CB(packet)->mru = mru;
 603
 604         if (a[OVS_PACKET_ATTR_HASH]) {
 605                 hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
 606
 607                 __skb_set_hash(packet, hash & 0xFFFFFFFFULL,
 608                                !!(hash & OVS_PACKET_HASH_SW_BIT),
 609                                !!(hash & OVS_PACKET_HASH_L4_BIT));
 610         }
 611
 612         /* Build an sw_flow for sending this packet. */
 613         flow = ovs_flow_alloc();
 614         err = PTR_ERR(flow);
 615         if (IS_ERR(flow))
 616                 goto err_kfree_skb;
 617
 618         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 619                                              packet, &flow->key, log);
 620         if (err)
 621                 goto err_flow_free;
 622
 623         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 624                                    &flow->key, &acts, log);
 625         if (err)
 626                 goto err_flow_free;
 627
 628         rcu_assign_pointer(flow->sf_acts, acts);
 629         packet->priority = flow->key.phy.priority;
 630         packet->mark = flow->key.phy.skb_mark;
 631
 632         rcu_read_lock();
 633         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 634         err = -ENODEV;
 635         if (!dp)
 636                 goto err_unlock;
 637
 638         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 639         if (!input_vport)
 640                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 641
 642         if (!input_vport)
 643                 goto err_unlock;
 644
 645         packet->dev = input_vport->dev;
 646         OVS_CB(packet)->input_vport = input_vport;
 647         sf_acts = rcu_dereference(flow->sf_acts);
 648
 649         local_bh_disable();
 650         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 651         local_bh_enable();
 652         rcu_read_unlock();
 653
 654         ovs_flow_free(flow, false);
 655         return err;
 656
 657 err_unlock:
 658         rcu_read_unlock();
 659 err_flow_free:
 660         ovs_flow_free(flow, false);
 661 err_kfree_skb:
 662         kfree_skb(packet);
 663 err:
 664         return err;
 665 }
 666
 667 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 668         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 669         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 670         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 671         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 672         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 673         [OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
 674 };
 675
 676 static const struct genl_small_ops dp_packet_genl_ops[] = {
 677         { .cmd = OVS_PACKET_CMD_EXECUTE,
 678           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 679           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 680           .doit = ovs_packet_cmd_execute
 681         }
 682 };
 683
 684 static struct genl_family dp_packet_genl_family __ro_after_init = {
 685         .hdrsize = sizeof(struct ovs_header),
 686         .name = OVS_PACKET_FAMILY,
 687         .version = OVS_PACKET_VERSION,
 688         .maxattr = OVS_PACKET_ATTR_MAX,
 689         .policy = packet_policy,
 690         .netnsok = true,
 691         .parallel_ops = true,
 692         .small_ops = dp_packet_genl_ops,
 693         .n_small_ops = ARRAY_SIZE(dp_packet_genl_ops),
 694         .module = THIS_MODULE,
 695 };
 696
 697 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 698                          struct ovs_dp_megaflow_stats *mega_stats)
 699 {
 700         int i;
 701
 702         memset(mega_stats, 0, sizeof(*mega_stats));
 703
 704         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 705         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 706
 707         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 708
 709         for_each_possible_cpu(i) {
 710                 const struct dp_stats_percpu *percpu_stats;
 711                 struct dp_stats_percpu local_stats;
 712                 unsigned int start;
 713
 714                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 715
 716                 do {
 717                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 718                         local_stats = *percpu_stats;
 719                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 720
 721                 stats->n_hit += local_stats.n_hit;
 722                 stats->n_missed += local_stats.n_missed;
 723                 stats->n_lost += local_stats.n_lost;
 724                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 725                 mega_stats->n_cache_hit += local_stats.n_cache_hit;
 726         }
 727 }
 728
 729 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 730 {
 731         return ovs_identifier_is_ufid(sfid) &&
 732                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 733 }
 734
 735 static bool should_fill_mask(uint32_t ufid_flags)
 736 {
 737         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 738 }
 739
 740 static bool should_fill_actions(uint32_t ufid_flags)
 741 {
 742         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 743 }
 744
 745 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 746                                     const struct sw_flow_id *sfid,
 747                                     uint32_t ufid_flags)
 748 {
 749         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 750
 751         /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
 752          * see ovs_nla_put_identifier()
 753          */
 754         if (sfid && ovs_identifier_is_ufid(sfid))
 755                 len += nla_total_size(sfid->ufid_len);
 756         else
 757                 len += nla_total_size(ovs_key_attr_size());
 758
 759         /* OVS_FLOW_ATTR_KEY */
 760         if (!sfid || should_fill_key(sfid, ufid_flags))
 761                 len += nla_total_size(ovs_key_attr_size());
 762
 763         /* OVS_FLOW_ATTR_MASK */
 764         if (should_fill_mask(ufid_flags))
 765                 len += nla_total_size(ovs_key_attr_size());
 766
 767         /* OVS_FLOW_ATTR_ACTIONS */
 768         if (should_fill_actions(ufid_flags))
 769                 len += nla_total_size(acts->orig_len);
 770
 771         return len
 772                 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 773                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 774                 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 775 }
 776
 777 /* Called with ovs_mutex or RCU read lock. */
 778 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 779                                    struct sk_buff *skb)
 780 {
 781         struct ovs_flow_stats stats;
 782         __be16 tcp_flags;
 783         unsigned long used;
 784
 785         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 786
 787         if (used &&
 788             nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 789                               OVS_FLOW_ATTR_PAD))
 790                 return -EMSGSIZE;
 791
 792         if (stats.n_packets &&
 793             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 794                           sizeof(struct ovs_flow_stats), &stats,
 795                           OVS_FLOW_ATTR_PAD))
 796                 return -EMSGSIZE;
 797
 798         if ((u8)ntohs(tcp_flags) &&
 799              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 800                 return -EMSGSIZE;
 801
 802         return 0;
 803 }
 804
 805 /* Called with ovs_mutex or RCU read lock. */
 806 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 807                                      struct sk_buff *skb, int skb_orig_len)
 808 {
 809         struct nlattr *start;
 810         int err;
 811
 812         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 813          * this is the first flow to be dumped into 'skb'.  This is unusual for
 814          * Netlink but individual action lists can be longer than
 815          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 816          * The userspace caller can always fetch the actions separately if it
 817          * really wants them.  (Most userspace callers in fact don't care.)
 818          *
 819          * This can only fail for dump operations because the skb is always
 820          * properly sized for single flows.
 821          */
 822         start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
 823         if (start) {
 824                 const struct sw_flow_actions *sf_acts;
 825
 826                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 827                 err = ovs_nla_put_actions(sf_acts->actions,
 828                                           sf_acts->actions_len, skb);
 829
 830                 if (!err)
 831                         nla_nest_end(skb, start);
 832                 else {
 833                         if (skb_orig_len)
 834                                 return err;
 835
 836                         nla_nest_cancel(skb, start);
 837                 }
 838         } else if (skb_orig_len) {
 839                 return -EMSGSIZE;
 840         }
 841
 842         return 0;
 843 }
 844
 845 /* Called with ovs_mutex or RCU read lock. */
 846 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 847                                   struct sk_buff *skb, u32 portid,
 848                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 849 {
 850         const int skb_orig_len = skb->len;
 851         struct ovs_header *ovs_header;
 852         int err;
 853
 854         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 855                                  flags, cmd);
 856         if (!ovs_header)
 857                 return -EMSGSIZE;
 858
 859         ovs_header->dp_ifindex = dp_ifindex;
 860
 861         err = ovs_nla_put_identifier(flow, skb);
 862         if (err)
 863                 goto error;
 864
 865         if (should_fill_key(&flow->id, ufid_flags)) {
 866                 err = ovs_nla_put_masked_key(flow, skb);
 867                 if (err)
 868                         goto error;
 869         }
 870
 871         if (should_fill_mask(ufid_flags)) {
 872                 err = ovs_nla_put_mask(flow, skb);
 873                 if (err)
 874                         goto error;
 875         }
 876
 877         err = ovs_flow_cmd_fill_stats(flow, skb);
 878         if (err)
 879                 goto error;
 880
 881         if (should_fill_actions(ufid_flags)) {
 882                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 883                 if (err)
 884                         goto error;
 885         }
 886
 887         genlmsg_end(skb, ovs_header);
 888         return 0;
 889
 890 error:
 891         genlmsg_cancel(skb, ovs_header);
 892         return err;
 893 }
 894
 895 /* May not be called with RCU read lock. */
 896 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 897                                                const struct sw_flow_id *sfid,
 898                                                struct genl_info *info,
 899                                                bool always,
 900                                                uint32_t ufid_flags)
 901 {
 902         struct sk_buff *skb;
 903         size_t len;
 904
 905         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 906                 return NULL;
 907
 908         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 909         skb = genlmsg_new(len, GFP_KERNEL);
 910         if (!skb)
 911                 return ERR_PTR(-ENOMEM);
 912
 913         return skb;
 914 }
 915
 916 /* Called with ovs_mutex. */
 917 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 918                                                int dp_ifindex,
 919                                                struct genl_info *info, u8 cmd,
 920                                                bool always, u32 ufid_flags)
 921 {
 922         struct sk_buff *skb;
 923         int retval;
 924
 925         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 926                                       &flow->id, info, always, ufid_flags);
 927         if (IS_ERR_OR_NULL(skb))
 928                 return skb;
 929
 930         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 931                                         info->snd_portid, info->snd_seq, 0,
 932                                         cmd, ufid_flags);
 933         if (WARN_ON_ONCE(retval < 0)) {
 934                 kfree_skb(skb);
 935                 skb = ERR_PTR(retval);
 936         }
 937         return skb;
 938 }
 939
 940 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 941 {
 942         struct net *net = sock_net(skb->sk);
 943         struct nlattr **a = info->attrs;
 944         struct ovs_header *ovs_header = info->userhdr;
 945         struct sw_flow *flow = NULL, *new_flow;
 946         struct sw_flow_mask mask;
 947         struct sk_buff *reply;
 948         struct datapath *dp;
 949         struct sw_flow_key *key;
 950         struct sw_flow_actions *acts;
 951         struct sw_flow_match match;
 952         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 953         int error;
 954         bool log = !a[OVS_FLOW_ATTR_PROBE];
 955
 956         /* Must have key and actions. */
 957         error = -EINVAL;
 958         if (!a[OVS_FLOW_ATTR_KEY]) {
 959                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 960                 goto error;
 961         }
 962         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 963                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 964                 goto error;
 965         }
 966
 967         /* Most of the time we need to allocate a new flow, do it before
 968          * locking.
 969          */
 970         new_flow = ovs_flow_alloc();
 971         if (IS_ERR(new_flow)) {
 972                 error = PTR_ERR(new_flow);
 973                 goto error;
 974         }
 975
 976         /* Extract key. */
 977         key = kzalloc(sizeof(*key), GFP_KERNEL);
 978         if (!key) {
 979                 error = -ENOMEM;
 980                 goto err_kfree_key;
 981         }
 982
 983         ovs_match_init(&match, key, false, &mask);
 984         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 985                                   a[OVS_FLOW_ATTR_MASK], log);
 986         if (error)
 987                 goto err_kfree_flow;
 988
 989         ovs_flow_mask_key(&new_flow->key, key, true, &mask);
 990
 991         /* Extract flow identifier. */
 992         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 993                                        key, log);
 994         if (error)
 995                 goto err_kfree_flow;
 996
 997         /* Validate actions. */
 998         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 999                                      &new_flow->key, &acts, log);
1000         if (error) {
1001                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
1002                 goto err_kfree_flow;
1003         }
1004
1005         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
1006                                         ufid_flags);
1007         if (IS_ERR(reply)) {
1008                 error = PTR_ERR(reply);
1009                 goto err_kfree_acts;
1010         }
1011
1012         ovs_lock();
1013         dp = get_dp(net, ovs_header->dp_ifindex);
1014         if (unlikely(!dp)) {
1015                 error = -ENODEV;
1016                 goto err_unlock_ovs;
1017         }
1018
1019         /* Check if this is a duplicate flow */
1020         if (ovs_identifier_is_ufid(&new_flow->id))
1021                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1022         if (!flow)
1023                 flow = ovs_flow_tbl_lookup(&dp->table, key);
1024         if (likely(!flow)) {
1025                 rcu_assign_pointer(new_flow->sf_acts, acts);
1026
1027                 /* Put flow in bucket. */
1028                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1029                 if (unlikely(error)) {
1030                         acts = NULL;
1031                         goto err_unlock_ovs;
1032                 }
1033
1034                 if (unlikely(reply)) {
1035                         error = ovs_flow_cmd_fill_info(new_flow,
1036                                                        ovs_header->dp_ifindex,
1037                                                        reply, info->snd_portid,
1038                                                        info->snd_seq, 0,
1039                                                        OVS_FLOW_CMD_NEW,
1040                                                        ufid_flags);
1041                         BUG_ON(error < 0);
1042                 }
1043                 ovs_unlock();
1044         } else {
1045                 struct sw_flow_actions *old_acts;
1046
1047                 /* Bail out if we're not allowed to modify an existing flow.
1048                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1049                  * because Generic Netlink treats the latter as a dump
1050                  * request.  We also accept NLM_F_EXCL in case that bug ever
1051                  * gets fixed.
1052                  */
1053                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1054                                                          | NLM_F_EXCL))) {
1055                         error = -EEXIST;
1056                         goto err_unlock_ovs;
1057                 }
1058                 /* The flow identifier has to be the same for flow updates.
1059                  * Look for any overlapping flow.
1060                  */
1061                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1062                         if (ovs_identifier_is_key(&flow->id))
1063                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1064                                                                  &match);
1065                         else /* UFID matches but key is different */
1066                                 flow = NULL;
1067                         if (!flow) {
1068                                 error = -ENOENT;
1069                                 goto err_unlock_ovs;
1070                         }
1071                 }
1072                 /* Update actions. */
1073                 old_acts = ovsl_dereference(flow->sf_acts);
1074                 rcu_assign_pointer(flow->sf_acts, acts);
1075
1076                 if (unlikely(reply)) {
1077                         error = ovs_flow_cmd_fill_info(flow,
1078                                                        ovs_header->dp_ifindex,
1079                                                        reply, info->snd_portid,
1080                                                        info->snd_seq, 0,
1081                                                        OVS_FLOW_CMD_NEW,
1082                                                        ufid_flags);
1083                         BUG_ON(error < 0);
1084                 }
1085                 ovs_unlock();
1086
1087                 ovs_nla_free_flow_actions_rcu(old_acts);
1088                 ovs_flow_free(new_flow, false);
1089         }
1090
1091         if (reply)
1092                 ovs_notify(&dp_flow_genl_family, reply, info);
1093
1094         kfree(key);
1095         return 0;
1096
1097 err_unlock_ovs:
1098         ovs_unlock();
1099         kfree_skb(reply);
1100 err_kfree_acts:
1101         ovs_nla_free_flow_actions(acts);
1102 err_kfree_flow:
1103         ovs_flow_free(new_flow, false);
1104 err_kfree_key:
1105         kfree(key);
1106 error:
1107         return error;
1108 }
1109
1110 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1111 static noinline_for_stack
1112 struct sw_flow_actions *get_flow_actions(struct net *net,
1113                                          const struct nlattr *a,
1114                                          const struct sw_flow_key *key,
1115                                          const struct sw_flow_mask *mask,
1116                                          bool log)
1117 {
1118         struct sw_flow_actions *acts;
1119         struct sw_flow_key masked_key;
1120         int error;
1121
1122         ovs_flow_mask_key(&masked_key, key, true, mask);
1123         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1124         if (error) {
1125                 OVS_NLERR(log,
1126                           "Actions may not be safe on all matching packets");
1127                 return ERR_PTR(error);
1128         }
1129
1130         return acts;
1131 }
1132
1133 /* Factor out match-init and action-copy to avoid
1134  * "Wframe-larger-than=1024" warning. Because mask is only
1135  * used to get actions, we new a function to save some
1136  * stack space.
1137  *
1138  * If there are not key and action attrs, we return 0
1139  * directly. In the case, the caller will also not use the
1140  * match as before. If there is action attr, we try to get
1141  * actions and save them to *acts. Before returning from
1142  * the function, we reset the match->mask pointer. Because
1143  * we should not to return match object with dangling reference
1144  * to mask.
1145  * */
1146 static noinline_for_stack int
1147 ovs_nla_init_match_and_action(struct net *net,
1148                               struct sw_flow_match *match,
1149                               struct sw_flow_key *key,
1150                               struct nlattr **a,
1151                               struct sw_flow_actions **acts,
1152                               bool log)
1153 {
1154         struct sw_flow_mask mask;
1155         int error = 0;
1156
1157         if (a[OVS_FLOW_ATTR_KEY]) {
1158                 ovs_match_init(match, key, true, &mask);
1159                 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1160                                           a[OVS_FLOW_ATTR_MASK], log);
1161                 if (error)
1162                         goto error;
1163         }
1164
1165         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1166                 if (!a[OVS_FLOW_ATTR_KEY]) {
1167                         OVS_NLERR(log,
1168                                   "Flow key attribute not present in set flow.");
1169                         error = -EINVAL;
1170                         goto error;
1171                 }
1172
1173                 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1174                                          &mask, log);
1175                 if (IS_ERR(*acts)) {
1176                         error = PTR_ERR(*acts);
1177                         goto error;
1178                 }
1179         }
1180
1181         /* On success, error is 0. */
1182 error:
1183         match->mask = NULL;
1184         return error;
1185 }
1186
1187 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1188 {
1189         struct net *net = sock_net(skb->sk);
1190         struct nlattr **a = info->attrs;
1191         struct ovs_header *ovs_header = info->userhdr;
1192         struct sw_flow_key key;
1193         struct sw_flow *flow;
1194         struct sk_buff *reply = NULL;
1195         struct datapath *dp;
1196         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1197         struct sw_flow_match match;
1198         struct sw_flow_id sfid;
1199         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1200         int error = 0;
1201         bool log = !a[OVS_FLOW_ATTR_PROBE];
1202         bool ufid_present;
1203
1204         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1205         if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1206                 OVS_NLERR(log,
1207                           "Flow set message rejected, Key attribute missing.");
1208                 return -EINVAL;
1209         }
1210
1211         error = ovs_nla_init_match_and_action(net, &match, &key, a,
1212                                               &acts, log);
1213         if (error)
1214                 goto error;
1215
1216         if (acts) {
1217                 /* Can allocate before locking if have acts. */
1218                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1219                                                 ufid_flags);
1220                 if (IS_ERR(reply)) {
1221                         error = PTR_ERR(reply);
1222                         goto err_kfree_acts;
1223                 }
1224         }
1225
1226         ovs_lock();
1227         dp = get_dp(net, ovs_header->dp_ifindex);
1228         if (unlikely(!dp)) {
1229                 error = -ENODEV;
1230                 goto err_unlock_ovs;
1231         }
1232         /* Check that the flow exists. */
1233         if (ufid_present)
1234                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1235         else
1236                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1237         if (unlikely(!flow)) {
1238                 error = -ENOENT;
1239                 goto err_unlock_ovs;
1240         }
1241
1242         /* Update actions, if present. */
1243         if (likely(acts)) {
1244                 old_acts = ovsl_dereference(flow->sf_acts);
1245                 rcu_assign_pointer(flow->sf_acts, acts);
1246
1247                 if (unlikely(reply)) {
1248                         error = ovs_flow_cmd_fill_info(flow,
1249                                                        ovs_header->dp_ifindex,
1250                                                        reply, info->snd_portid,
1251                                                        info->snd_seq, 0,
1252                                                        OVS_FLOW_CMD_SET,
1253                                                        ufid_flags);
1254                         BUG_ON(error < 0);
1255                 }
1256         } else {
1257                 /* Could not alloc without acts before locking. */
1258                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1259                                                 info, OVS_FLOW_CMD_SET, false,
1260                                                 ufid_flags);
1261
1262                 if (IS_ERR(reply)) {
1263                         error = PTR_ERR(reply);
1264                         goto err_unlock_ovs;
1265                 }
1266         }
1267
1268         /* Clear stats. */
1269         if (a[OVS_FLOW_ATTR_CLEAR])
1270                 ovs_flow_stats_clear(flow);
1271         ovs_unlock();
1272
1273         if (reply)
1274                 ovs_notify(&dp_flow_genl_family, reply, info);
1275         if (old_acts)
1276                 ovs_nla_free_flow_actions_rcu(old_acts);
1277
1278         return 0;
1279
1280 err_unlock_ovs:
1281         ovs_unlock();
1282         kfree_skb(reply);
1283 err_kfree_acts:
1284         ovs_nla_free_flow_actions(acts);
1285 error:
1286         return error;
1287 }
1288
1289 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1290 {
1291         struct nlattr **a = info->attrs;
1292         struct ovs_header *ovs_header = info->userhdr;
1293         struct net *net = sock_net(skb->sk);
1294         struct sw_flow_key key;
1295         struct sk_buff *reply;
1296         struct sw_flow *flow;
1297         struct datapath *dp;
1298         struct sw_flow_match match;
1299         struct sw_flow_id ufid;
1300         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1301         int err = 0;
1302         bool log = !a[OVS_FLOW_ATTR_PROBE];
1303         bool ufid_present;
1304
1305         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1306         if (a[OVS_FLOW_ATTR_KEY]) {
1307                 ovs_match_init(&match, &key, true, NULL);
1308                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1309                                         log);
1310         } else if (!ufid_present) {
1311                 OVS_NLERR(log,
1312                           "Flow get message rejected, Key attribute missing.");
1313                 err = -EINVAL;
1314         }
1315         if (err)
1316                 return err;
1317
1318         ovs_lock();
1319         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1320         if (!dp) {
1321                 err = -ENODEV;
1322                 goto unlock;
1323         }
1324
1325         if (ufid_present)
1326                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1327         else
1328                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1329         if (!flow) {
1330                 err = -ENOENT;
1331                 goto unlock;
1332         }
1333
1334         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1335                                         OVS_FLOW_CMD_GET, true, ufid_flags);
1336         if (IS_ERR(reply)) {
1337                 err = PTR_ERR(reply);
1338                 goto unlock;
1339         }
1340
1341         ovs_unlock();
1342         return genlmsg_reply(reply, info);
1343 unlock:
1344         ovs_unlock();
1345         return err;
1346 }
1347
1348 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1349 {
1350         struct nlattr **a = info->attrs;
1351         struct ovs_header *ovs_header = info->userhdr;
1352         struct net *net = sock_net(skb->sk);
1353         struct sw_flow_key key;
1354         struct sk_buff *reply;
1355         struct sw_flow *flow = NULL;
1356         struct datapath *dp;
1357         struct sw_flow_match match;
1358         struct sw_flow_id ufid;
1359         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1360         int err;
1361         bool log = !a[OVS_FLOW_ATTR_PROBE];
1362         bool ufid_present;
1363
1364         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1365         if (a[OVS_FLOW_ATTR_KEY]) {
1366                 ovs_match_init(&match, &key, true, NULL);
1367                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1368                                         NULL, log);
1369                 if (unlikely(err))
1370                         return err;
1371         }
1372
1373         ovs_lock();
1374         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1375         if (unlikely(!dp)) {
1376                 err = -ENODEV;
1377                 goto unlock;
1378         }
1379
1380         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1381                 err = ovs_flow_tbl_flush(&dp->table);
1382                 goto unlock;
1383         }
1384
1385         if (ufid_present)
1386                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1387         else
1388                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1389         if (unlikely(!flow)) {
1390                 err = -ENOENT;
1391                 goto unlock;
1392         }
1393
1394         ovs_flow_tbl_remove(&dp->table, flow);
1395         ovs_unlock();
1396
1397         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1398                                         &flow->id, info, false, ufid_flags);
1399         if (likely(reply)) {
1400                 if (!IS_ERR(reply)) {
1401                         rcu_read_lock();        /*To keep RCU checker happy. */
1402                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1403                                                      reply, info->snd_portid,
1404                                                      info->snd_seq, 0,
1405                                                      OVS_FLOW_CMD_DEL,
1406                                                      ufid_flags);
1407                         rcu_read_unlock();
1408                         if (WARN_ON_ONCE(err < 0)) {
1409                                 kfree_skb(reply);
1410                                 goto out_free;
1411                         }
1412
1413                         ovs_notify(&dp_flow_genl_family, reply, info);
1414                 } else {
1415                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1416                                         PTR_ERR(reply));
1417                 }
1418         }
1419
1420 out_free:
1421         ovs_flow_free(flow, true);
1422         return 0;
1423 unlock:
1424         ovs_unlock();
1425         return err;
1426 }
1427
1428 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1429 {
1430         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1431         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1432         struct table_instance *ti;
1433         struct datapath *dp;
1434         u32 ufid_flags;
1435         int err;
1436
1437         err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1438                                        OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1439         if (err)
1440                 return err;
1441         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1442
1443         rcu_read_lock();
1444         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1445         if (!dp) {
1446                 rcu_read_unlock();
1447                 return -ENODEV;
1448         }
1449
1450         ti = rcu_dereference(dp->table.ti);
1451         for (;;) {
1452                 struct sw_flow *flow;
1453                 u32 bucket, obj;
1454
1455                 bucket = cb->args[0];
1456                 obj = cb->args[1];
1457                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1458                 if (!flow)
1459                         break;
1460
1461                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1462                                            NETLINK_CB(cb->skb).portid,
1463                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1464                                            OVS_FLOW_CMD_GET, ufid_flags) < 0)
1465                         break;
1466
1467                 cb->args[0] = bucket;
1468                 cb->args[1] = obj;
1469         }
1470         rcu_read_unlock();
1471         return skb->len;
1472 }
1473
1474 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1475         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1476         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1477         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1478         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1479         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1480         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1481         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1482 };
1483
1484 static const struct genl_small_ops dp_flow_genl_ops[] = {
1485         { .cmd = OVS_FLOW_CMD_NEW,
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_new
1489         },
1490         { .cmd = OVS_FLOW_CMD_DEL,
1491           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1492           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1493           .doit = ovs_flow_cmd_del
1494         },
1495         { .cmd = OVS_FLOW_CMD_GET,
1496           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1497           .flags = 0,               /* OK for unprivileged users. */
1498           .doit = ovs_flow_cmd_get,
1499           .dumpit = ovs_flow_cmd_dump
1500         },
1501         { .cmd = OVS_FLOW_CMD_SET,
1502           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1503           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1504           .doit = ovs_flow_cmd_set,
1505         },
1506 };
1507
1508 static struct genl_family dp_flow_genl_family __ro_after_init = {
1509         .hdrsize = sizeof(struct ovs_header),
1510         .name = OVS_FLOW_FAMILY,
1511         .version = OVS_FLOW_VERSION,
1512         .maxattr = OVS_FLOW_ATTR_MAX,
1513         .policy = flow_policy,
1514         .netnsok = true,
1515         .parallel_ops = true,
1516         .small_ops = dp_flow_genl_ops,
1517         .n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
1518         .mcgrps = &ovs_dp_flow_multicast_group,
1519         .n_mcgrps = 1,
1520         .module = THIS_MODULE,
1521 };
1522
1523 static size_t ovs_dp_cmd_msg_size(void)
1524 {
1525         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1526
1527         msgsize += nla_total_size(IFNAMSIZ);
1528         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1529         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1530         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1531         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1532
1533         return msgsize;
1534 }
1535
1536 /* Called with ovs_mutex. */
1537 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1538                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1539 {
1540         struct ovs_header *ovs_header;
1541         struct ovs_dp_stats dp_stats;
1542         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1543         int err;
1544
1545         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1546                                  flags, cmd);
1547         if (!ovs_header)
1548                 goto error;
1549
1550         ovs_header->dp_ifindex = get_dpifindex(dp);
1551
1552         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1553         if (err)
1554                 goto nla_put_failure;
1555
1556         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1557         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1558                           &dp_stats, OVS_DP_ATTR_PAD))
1559                 goto nla_put_failure;
1560
1561         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1562                           sizeof(struct ovs_dp_megaflow_stats),
1563                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1564                 goto nla_put_failure;
1565
1566         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1567                 goto nla_put_failure;
1568
1569         if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1570                         ovs_flow_tbl_masks_cache_size(&dp->table)))
1571                 goto nla_put_failure;
1572
1573         genlmsg_end(skb, ovs_header);
1574         return 0;
1575
1576 nla_put_failure:
1577         genlmsg_cancel(skb, ovs_header);
1578 error:
1579         return -EMSGSIZE;
1580 }
1581
1582 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1583 {
1584         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1585 }
1586
1587 /* Called with rcu_read_lock or ovs_mutex. */
1588 static struct datapath *lookup_datapath(struct net *net,
1589                                         const struct ovs_header *ovs_header,
1590                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1591 {
1592         struct datapath *dp;
1593
1594         if (!a[OVS_DP_ATTR_NAME])
1595                 dp = get_dp(net, ovs_header->dp_ifindex);
1596         else {
1597                 struct vport *vport;
1598
1599                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1600                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1601         }
1602         return dp ? dp : ERR_PTR(-ENODEV);
1603 }
1604
1605 static void ovs_dp_reset_user_features(struct sk_buff *skb,
1606                                        struct genl_info *info)
1607 {
1608         struct datapath *dp;
1609
1610         dp = lookup_datapath(sock_net(skb->sk), info->userhdr,
1611                              info->attrs);
1612         if (IS_ERR(dp))
1613                 return;
1614
1615         pr_warn("%s: Dropping previously announced user features\n",
1616                 ovs_dp_name(dp));
1617         dp->user_features = 0;
1618 }
1619
1620 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1621
1622 static int ovs_dp_set_upcall_portids(struct datapath *dp,
1623                               const struct nlattr *ids)
1624 {
1625         struct dp_nlsk_pids *old, *dp_nlsk_pids;
1626
1627         if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
1628                 return -EINVAL;
1629
1630         old = ovsl_dereference(dp->upcall_portids);
1631
1632         dp_nlsk_pids = kmalloc(sizeof(*dp_nlsk_pids) + nla_len(ids),
1633                                GFP_KERNEL);
1634         if (!dp_nlsk_pids)
1635                 return -ENOMEM;
1636
1637         dp_nlsk_pids->n_pids = nla_len(ids) / sizeof(u32);
1638         nla_memcpy(dp_nlsk_pids->pids, ids, nla_len(ids));
1639
1640         rcu_assign_pointer(dp->upcall_portids, dp_nlsk_pids);
1641
1642         kfree_rcu(old, rcu);
1643
1644         return 0;
1645 }
1646
1647 u32 ovs_dp_get_upcall_portid(const struct datapath *dp, uint32_t cpu_id)
1648 {
1649         struct dp_nlsk_pids *dp_nlsk_pids;
1650
1651         dp_nlsk_pids = rcu_dereference(dp->upcall_portids);
1652
1653         if (dp_nlsk_pids) {
1654                 if (cpu_id < dp_nlsk_pids->n_pids) {
1655                         return dp_nlsk_pids->pids[cpu_id];
1656                 } else if (dp_nlsk_pids->n_pids > 0 &&
1657                            cpu_id >= dp_nlsk_pids->n_pids) {
1658                         /* If the number of netlink PIDs is mismatched with
1659                          * the number of CPUs as seen by the kernel, log this
1660                          * and send the upcall to an arbitrary socket (0) in
1661                          * order to not drop packets
1662                          */
1663                         pr_info_ratelimited("cpu_id mismatch with handler threads");
1664                         return dp_nlsk_pids->pids[cpu_id %
1665                                                   dp_nlsk_pids->n_pids];
1666                 } else {
1667                         return 0;
1668                 }
1669         } else {
1670                 return 0;
1671         }
1672 }
1673
1674 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1675 {
1676         u32 user_features = 0;
1677         int err;
1678
1679         if (a[OVS_DP_ATTR_USER_FEATURES]) {
1680                 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1681
1682                 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1683                                       OVS_DP_F_UNALIGNED |
1684                                       OVS_DP_F_TC_RECIRC_SHARING |
1685                                       OVS_DP_F_DISPATCH_UPCALL_PER_CPU))
1686                         return -EOPNOTSUPP;
1687
1688 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1689                 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1690                         return -EOPNOTSUPP;
1691 #endif
1692         }
1693
1694         if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1695                 int err;
1696                 u32 cache_size;
1697
1698                 cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1699                 err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1700                 if (err)
1701                         return err;
1702         }
1703
1704         dp->user_features = user_features;
1705
1706         if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU &&
1707             a[OVS_DP_ATTR_PER_CPU_PIDS]) {
1708                 /* Upcall Netlink Port IDs have been updated */
1709                 err = ovs_dp_set_upcall_portids(dp,
1710                                                 a[OVS_DP_ATTR_PER_CPU_PIDS]);
1711                 if (err)
1712                         return err;
1713         }
1714
1715         if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1716                 static_branch_enable(&tc_recirc_sharing_support);
1717         else
1718                 static_branch_disable(&tc_recirc_sharing_support);
1719
1720         return 0;
1721 }
1722
1723 static int ovs_dp_stats_init(struct datapath *dp)
1724 {
1725         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1726         if (!dp->stats_percpu)
1727                 return -ENOMEM;
1728
1729         return 0;
1730 }
1731
1732 static int ovs_dp_vport_init(struct datapath *dp)
1733 {
1734         int i;
1735
1736         dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1737                                   sizeof(struct hlist_head),
1738                                   GFP_KERNEL);
1739         if (!dp->ports)
1740                 return -ENOMEM;
1741
1742         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1743                 INIT_HLIST_HEAD(&dp->ports[i]);
1744
1745         return 0;
1746 }
1747
1748 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1749 {
1750         struct nlattr **a = info->attrs;
1751         struct vport_parms parms;
1752         struct sk_buff *reply;
1753         struct datapath *dp;
1754         struct vport *vport;
1755         struct ovs_net *ovs_net;
1756         int err;
1757
1758         err = -EINVAL;
1759         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1760                 goto err;
1761
1762         reply = ovs_dp_cmd_alloc_info();
1763         if (!reply)
1764                 return -ENOMEM;
1765
1766         err = -ENOMEM;
1767         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1768         if (dp == NULL)
1769                 goto err_destroy_reply;
1770
1771         ovs_dp_set_net(dp, sock_net(skb->sk));
1772
1773         /* Allocate table. */
1774         err = ovs_flow_tbl_init(&dp->table);
1775         if (err)
1776                 goto err_destroy_dp;
1777
1778         err = ovs_dp_stats_init(dp);
1779         if (err)
1780                 goto err_destroy_table;
1781
1782         err = ovs_dp_vport_init(dp);
1783         if (err)
1784                 goto err_destroy_stats;
1785
1786         err = ovs_meters_init(dp);
1787         if (err)
1788                 goto err_destroy_ports;
1789
1790         /* Set up our datapath device. */
1791         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1792         parms.type = OVS_VPORT_TYPE_INTERNAL;
1793         parms.options = NULL;
1794         parms.dp = dp;
1795         parms.port_no = OVSP_LOCAL;
1796         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1797
1798         /* So far only local changes have been made, now need the lock. */
1799         ovs_lock();
1800
1801         err = ovs_dp_change(dp, a);
1802         if (err)
1803                 goto err_unlock_and_destroy_meters;
1804
1805         vport = new_vport(&parms);
1806         if (IS_ERR(vport)) {
1807                 err = PTR_ERR(vport);
1808                 if (err == -EBUSY)
1809                         err = -EEXIST;
1810
1811                 if (err == -EEXIST) {
1812                         /* An outdated user space instance that does not understand
1813                          * the concept of user_features has attempted to create a new
1814                          * datapath and is likely to reuse it. Drop all user features.
1815                          */
1816                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1817                                 ovs_dp_reset_user_features(skb, info);
1818                 }
1819
1820                 goto err_destroy_portids;
1821         }
1822
1823         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1824                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1825         BUG_ON(err < 0);
1826
1827         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1828         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1829
1830         ovs_unlock();
1831
1832         ovs_notify(&dp_datapath_genl_family, reply, info);
1833         return 0;
1834
1835 err_destroy_portids:
1836         kfree(rcu_dereference_raw(dp->upcall_portids));
1837 err_unlock_and_destroy_meters:
1838         ovs_unlock();
1839         ovs_meters_exit(dp);
1840 err_destroy_ports:
1841         kfree(dp->ports);
1842 err_destroy_stats:
1843         free_percpu(dp->stats_percpu);
1844 err_destroy_table:
1845         ovs_flow_tbl_destroy(&dp->table);
1846 err_destroy_dp:
1847         kfree(dp);
1848 err_destroy_reply:
1849         kfree_skb(reply);
1850 err:
1851         return err;
1852 }
1853
1854 /* Called with ovs_mutex. */
1855 static void __dp_destroy(struct datapath *dp)
1856 {
1857         struct flow_table *table = &dp->table;
1858         int i;
1859
1860         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1861                 struct vport *vport;
1862                 struct hlist_node *n;
1863
1864                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1865                         if (vport->port_no != OVSP_LOCAL)
1866                                 ovs_dp_detach_port(vport);
1867         }
1868
1869         list_del_rcu(&dp->list_node);
1870
1871         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1872          * all ports in datapath are destroyed first before freeing datapath.
1873          */
1874         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1875
1876         /* Flush sw_flow in the tables. RCU cb only releases resource
1877          * such as dp, ports and tables. That may avoid some issues
1878          * such as RCU usage warning.
1879          */
1880         table_instance_flow_flush(table, ovsl_dereference(table->ti),
1881                                   ovsl_dereference(table->ufid_ti));
1882
1883         /* RCU destroy the ports, meters and flow tables. */
1884         call_rcu(&dp->rcu, destroy_dp_rcu);
1885 }
1886
1887 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1888 {
1889         struct sk_buff *reply;
1890         struct datapath *dp;
1891         int err;
1892
1893         reply = ovs_dp_cmd_alloc_info();
1894         if (!reply)
1895                 return -ENOMEM;
1896
1897         ovs_lock();
1898         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1899         err = PTR_ERR(dp);
1900         if (IS_ERR(dp))
1901                 goto err_unlock_free;
1902
1903         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1904                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1905         BUG_ON(err < 0);
1906
1907         __dp_destroy(dp);
1908         ovs_unlock();
1909
1910         ovs_notify(&dp_datapath_genl_family, reply, info);
1911
1912         return 0;
1913
1914 err_unlock_free:
1915         ovs_unlock();
1916         kfree_skb(reply);
1917         return err;
1918 }
1919
1920 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1921 {
1922         struct sk_buff *reply;
1923         struct datapath *dp;
1924         int err;
1925
1926         reply = ovs_dp_cmd_alloc_info();
1927         if (!reply)
1928                 return -ENOMEM;
1929
1930         ovs_lock();
1931         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1932         err = PTR_ERR(dp);
1933         if (IS_ERR(dp))
1934                 goto err_unlock_free;
1935
1936         err = ovs_dp_change(dp, info->attrs);
1937         if (err)
1938                 goto err_unlock_free;
1939
1940         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1941                                    info->snd_seq, 0, OVS_DP_CMD_SET);
1942         BUG_ON(err < 0);
1943
1944         ovs_unlock();
1945         ovs_notify(&dp_datapath_genl_family, reply, info);
1946
1947         return 0;
1948
1949 err_unlock_free:
1950         ovs_unlock();
1951         kfree_skb(reply);
1952         return err;
1953 }
1954
1955 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1956 {
1957         struct sk_buff *reply;
1958         struct datapath *dp;
1959         int err;
1960
1961         reply = ovs_dp_cmd_alloc_info();
1962         if (!reply)
1963                 return -ENOMEM;
1964
1965         ovs_lock();
1966         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1967         if (IS_ERR(dp)) {
1968                 err = PTR_ERR(dp);
1969                 goto err_unlock_free;
1970         }
1971         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1972                                    info->snd_seq, 0, OVS_DP_CMD_GET);
1973         BUG_ON(err < 0);
1974         ovs_unlock();
1975
1976         return genlmsg_reply(reply, info);
1977
1978 err_unlock_free:
1979         ovs_unlock();
1980         kfree_skb(reply);
1981         return err;
1982 }
1983
1984 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1985 {
1986         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1987         struct datapath *dp;
1988         int skip = cb->args[0];
1989         int i = 0;
1990
1991         ovs_lock();
1992         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1993                 if (i >= skip &&
1994                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1995                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1996                                          OVS_DP_CMD_GET) < 0)
1997                         break;
1998                 i++;
1999         }
2000         ovs_unlock();
2001
2002         cb->args[0] = i;
2003
2004         return skb->len;
2005 }
2006
2007 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
2008         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2009         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2010         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
2011         [OVS_DP_ATTR_MASKS_CACHE_SIZE] =  NLA_POLICY_RANGE(NLA_U32, 0,
2012                 PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
2013 };
2014
2015 static const struct genl_small_ops dp_datapath_genl_ops[] = {
2016         { .cmd = OVS_DP_CMD_NEW,
2017           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2018           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2019           .doit = ovs_dp_cmd_new
2020         },
2021         { .cmd = OVS_DP_CMD_DEL,
2022           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2023           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2024           .doit = ovs_dp_cmd_del
2025         },
2026         { .cmd = OVS_DP_CMD_GET,
2027           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2028           .flags = 0,               /* OK for unprivileged users. */
2029           .doit = ovs_dp_cmd_get,
2030           .dumpit = ovs_dp_cmd_dump
2031         },
2032         { .cmd = OVS_DP_CMD_SET,
2033           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2034           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2035           .doit = ovs_dp_cmd_set,
2036         },
2037 };
2038
2039 static struct genl_family dp_datapath_genl_family __ro_after_init = {
2040         .hdrsize = sizeof(struct ovs_header),
2041         .name = OVS_DATAPATH_FAMILY,
2042         .version = OVS_DATAPATH_VERSION,
2043         .maxattr = OVS_DP_ATTR_MAX,
2044         .policy = datapath_policy,
2045         .netnsok = true,
2046         .parallel_ops = true,
2047         .small_ops = dp_datapath_genl_ops,
2048         .n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
2049         .mcgrps = &ovs_dp_datapath_multicast_group,
2050         .n_mcgrps = 1,
2051         .module = THIS_MODULE,
2052 };
2053
2054 /* Called with ovs_mutex or RCU read lock. */
2055 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
2056                                    struct net *net, u32 portid, u32 seq,
2057                                    u32 flags, u8 cmd, gfp_t gfp)
2058 {
2059         struct ovs_header *ovs_header;
2060         struct ovs_vport_stats vport_stats;
2061         int err;
2062
2063         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
2064                                  flags, cmd);
2065         if (!ovs_header)
2066                 return -EMSGSIZE;
2067
2068         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
2069
2070         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
2071             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
2072             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
2073                            ovs_vport_name(vport)) ||
2074             nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
2075                 goto nla_put_failure;
2076
2077         if (!net_eq(net, dev_net(vport->dev))) {
2078                 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
2079
2080                 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
2081                         goto nla_put_failure;
2082         }
2083
2084         ovs_vport_get_stats(vport, &vport_stats);
2085         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
2086                           sizeof(struct ovs_vport_stats), &vport_stats,
2087                           OVS_VPORT_ATTR_PAD))
2088                 goto nla_put_failure;
2089
2090         if (ovs_vport_get_upcall_portids(vport, skb))
2091                 goto nla_put_failure;
2092
2093         err = ovs_vport_get_options(vport, skb);
2094         if (err == -EMSGSIZE)
2095                 goto error;
2096
2097         genlmsg_end(skb, ovs_header);
2098         return 0;
2099
2100 nla_put_failure:
2101         err = -EMSGSIZE;
2102 error:
2103         genlmsg_cancel(skb, ovs_header);
2104         return err;
2105 }
2106
2107 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2108 {
2109         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2110 }
2111
2112 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2113 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2114                                          u32 portid, u32 seq, u8 cmd)
2115 {
2116         struct sk_buff *skb;
2117         int retval;
2118
2119         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2120         if (!skb)
2121                 return ERR_PTR(-ENOMEM);
2122
2123         retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2124                                          GFP_KERNEL);
2125         BUG_ON(retval < 0);
2126
2127         return skb;
2128 }
2129
2130 /* Called with ovs_mutex or RCU read lock. */
2131 static struct vport *lookup_vport(struct net *net,
2132                                   const struct ovs_header *ovs_header,
2133                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2134 {
2135         struct datapath *dp;
2136         struct vport *vport;
2137
2138         if (a[OVS_VPORT_ATTR_IFINDEX])
2139                 return ERR_PTR(-EOPNOTSUPP);
2140         if (a[OVS_VPORT_ATTR_NAME]) {
2141                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2142                 if (!vport)
2143                         return ERR_PTR(-ENODEV);
2144                 if (ovs_header->dp_ifindex &&
2145                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2146                         return ERR_PTR(-ENODEV);
2147                 return vport;
2148         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2149                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2150
2151                 if (port_no >= DP_MAX_PORTS)
2152                         return ERR_PTR(-EFBIG);
2153
2154                 dp = get_dp(net, ovs_header->dp_ifindex);
2155                 if (!dp)
2156                         return ERR_PTR(-ENODEV);
2157
2158                 vport = ovs_vport_ovsl_rcu(dp, port_no);
2159                 if (!vport)
2160                         return ERR_PTR(-ENODEV);
2161                 return vport;
2162         } else
2163                 return ERR_PTR(-EINVAL);
2164
2165 }
2166
2167 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2168 {
2169         unsigned int dev_headroom, max_headroom = 0;
2170         struct net_device *dev;
2171         struct vport *vport;
2172         int i;
2173
2174         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2175                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2176                                          lockdep_ovsl_is_held()) {
2177                         dev = vport->dev;
2178                         dev_headroom = netdev_get_fwd_headroom(dev);
2179                         if (dev_headroom > max_headroom)
2180                                 max_headroom = dev_headroom;
2181                 }
2182         }
2183
2184         return max_headroom;
2185 }
2186
2187 /* Called with ovs_mutex */
2188 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2189 {
2190         struct vport *vport;
2191         int i;
2192
2193         dp->max_headroom = new_headroom;
2194         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2195                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2196                                          lockdep_ovsl_is_held())
2197                         netdev_set_rx_headroom(vport->dev, new_headroom);
2198         }
2199 }
2200
2201 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2202 {
2203         struct nlattr **a = info->attrs;
2204         struct ovs_header *ovs_header = info->userhdr;
2205         struct vport_parms parms;
2206         struct sk_buff *reply;
2207         struct vport *vport;
2208         struct datapath *dp;
2209         unsigned int new_headroom;
2210         u32 port_no;
2211         int err;
2212
2213         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2214             !a[OVS_VPORT_ATTR_UPCALL_PID])
2215                 return -EINVAL;
2216         if (a[OVS_VPORT_ATTR_IFINDEX])
2217                 return -EOPNOTSUPP;
2218
2219         port_no = a[OVS_VPORT_ATTR_PORT_NO]
2220                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2221         if (port_no >= DP_MAX_PORTS)
2222                 return -EFBIG;
2223
2224         reply = ovs_vport_cmd_alloc_info();
2225         if (!reply)
2226                 return -ENOMEM;
2227
2228         ovs_lock();
2229 restart:
2230         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2231         err = -ENODEV;
2232         if (!dp)
2233                 goto exit_unlock_free;
2234
2235         if (port_no) {
2236                 vport = ovs_vport_ovsl(dp, port_no);
2237                 err = -EBUSY;
2238                 if (vport)
2239                         goto exit_unlock_free;
2240         } else {
2241                 for (port_no = 1; ; port_no++) {
2242                         if (port_no >= DP_MAX_PORTS) {
2243                                 err = -EFBIG;
2244                                 goto exit_unlock_free;
2245                         }
2246                         vport = ovs_vport_ovsl(dp, port_no);
2247                         if (!vport)
2248                                 break;
2249                 }
2250         }
2251
2252         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2253         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2254         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2255         parms.dp = dp;
2256         parms.port_no = port_no;
2257         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2258
2259         vport = new_vport(&parms);
2260         err = PTR_ERR(vport);
2261         if (IS_ERR(vport)) {
2262                 if (err == -EAGAIN)
2263                         goto restart;
2264                 goto exit_unlock_free;
2265         }
2266
2267         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2268                                       info->snd_portid, info->snd_seq, 0,
2269                                       OVS_VPORT_CMD_NEW, GFP_KERNEL);
2270
2271         new_headroom = netdev_get_fwd_headroom(vport->dev);
2272
2273         if (new_headroom > dp->max_headroom)
2274                 ovs_update_headroom(dp, new_headroom);
2275         else
2276                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2277
2278         BUG_ON(err < 0);
2279         ovs_unlock();
2280
2281         ovs_notify(&dp_vport_genl_family, reply, info);
2282         return 0;
2283
2284 exit_unlock_free:
2285         ovs_unlock();
2286         kfree_skb(reply);
2287         return err;
2288 }
2289
2290 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2291 {
2292         struct nlattr **a = info->attrs;
2293         struct sk_buff *reply;
2294         struct vport *vport;
2295         int err;
2296
2297         reply = ovs_vport_cmd_alloc_info();
2298         if (!reply)
2299                 return -ENOMEM;
2300
2301         ovs_lock();
2302         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2303         err = PTR_ERR(vport);
2304         if (IS_ERR(vport))
2305                 goto exit_unlock_free;
2306
2307         if (a[OVS_VPORT_ATTR_TYPE] &&
2308             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2309                 err = -EINVAL;
2310                 goto exit_unlock_free;
2311         }
2312
2313         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2314                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2315                 if (err)
2316                         goto exit_unlock_free;
2317         }
2318
2319
2320         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2321                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2322
2323                 err = ovs_vport_set_upcall_portids(vport, ids);
2324                 if (err)
2325                         goto exit_unlock_free;
2326         }
2327
2328         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2329                                       info->snd_portid, info->snd_seq, 0,
2330                                       OVS_VPORT_CMD_SET, GFP_KERNEL);
2331         BUG_ON(err < 0);
2332
2333         ovs_unlock();
2334         ovs_notify(&dp_vport_genl_family, reply, info);
2335         return 0;
2336
2337 exit_unlock_free:
2338         ovs_unlock();
2339         kfree_skb(reply);
2340         return err;
2341 }
2342
2343 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2344 {
2345         bool update_headroom = false;
2346         struct nlattr **a = info->attrs;
2347         struct sk_buff *reply;
2348         struct datapath *dp;
2349         struct vport *vport;
2350         unsigned int new_headroom;
2351         int err;
2352
2353         reply = ovs_vport_cmd_alloc_info();
2354         if (!reply)
2355                 return -ENOMEM;
2356
2357         ovs_lock();
2358         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2359         err = PTR_ERR(vport);
2360         if (IS_ERR(vport))
2361                 goto exit_unlock_free;
2362
2363         if (vport->port_no == OVSP_LOCAL) {
2364                 err = -EINVAL;
2365                 goto exit_unlock_free;
2366         }
2367
2368         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2369                                       info->snd_portid, info->snd_seq, 0,
2370                                       OVS_VPORT_CMD_DEL, GFP_KERNEL);
2371         BUG_ON(err < 0);
2372
2373         /* the vport deletion may trigger dp headroom update */
2374         dp = vport->dp;
2375         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2376                 update_headroom = true;
2377
2378         netdev_reset_rx_headroom(vport->dev);
2379         ovs_dp_detach_port(vport);
2380
2381         if (update_headroom) {
2382                 new_headroom = ovs_get_max_headroom(dp);
2383
2384                 if (new_headroom < dp->max_headroom)
2385                         ovs_update_headroom(dp, new_headroom);
2386         }
2387         ovs_unlock();
2388
2389         ovs_notify(&dp_vport_genl_family, reply, info);
2390         return 0;
2391
2392 exit_unlock_free:
2393         ovs_unlock();
2394         kfree_skb(reply);
2395         return err;
2396 }
2397
2398 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2399 {
2400         struct nlattr **a = info->attrs;
2401         struct ovs_header *ovs_header = info->userhdr;
2402         struct sk_buff *reply;
2403         struct vport *vport;
2404         int err;
2405
2406         reply = ovs_vport_cmd_alloc_info();
2407         if (!reply)
2408                 return -ENOMEM;
2409
2410         rcu_read_lock();
2411         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2412         err = PTR_ERR(vport);
2413         if (IS_ERR(vport))
2414                 goto exit_unlock_free;
2415         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2416                                       info->snd_portid, info->snd_seq, 0,
2417                                       OVS_VPORT_CMD_GET, GFP_ATOMIC);
2418         BUG_ON(err < 0);
2419         rcu_read_unlock();
2420
2421         return genlmsg_reply(reply, info);
2422
2423 exit_unlock_free:
2424         rcu_read_unlock();
2425         kfree_skb(reply);
2426         return err;
2427 }
2428
2429 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2430 {
2431         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2432         struct datapath *dp;
2433         int bucket = cb->args[0], skip = cb->args[1];
2434         int i, j = 0;
2435
2436         rcu_read_lock();
2437         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2438         if (!dp) {
2439                 rcu_read_unlock();
2440                 return -ENODEV;
2441         }
2442         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2443                 struct vport *vport;
2444
2445                 j = 0;
2446                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2447                         if (j >= skip &&
2448                             ovs_vport_cmd_fill_info(vport, skb,
2449                                                     sock_net(skb->sk),
2450                                                     NETLINK_CB(cb->skb).portid,
2451                                                     cb->nlh->nlmsg_seq,
2452                                                     NLM_F_MULTI,
2453                                                     OVS_VPORT_CMD_GET,
2454                                                     GFP_ATOMIC) < 0)
2455                                 goto out;
2456
2457                         j++;
2458                 }
2459                 skip = 0;
2460         }
2461 out:
2462         rcu_read_unlock();
2463
2464         cb->args[0] = i;
2465         cb->args[1] = j;
2466
2467         return skb->len;
2468 }
2469
2470 static void ovs_dp_masks_rebalance(struct work_struct *work)
2471 {
2472         struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2473                                                masks_rebalance.work);
2474         struct datapath *dp;
2475
2476         ovs_lock();
2477
2478         list_for_each_entry(dp, &ovs_net->dps, list_node)
2479                 ovs_flow_masks_rebalance(&dp->table);
2480
2481         ovs_unlock();
2482
2483         schedule_delayed_work(&ovs_net->masks_rebalance,
2484                               msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2485 }
2486
2487 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2488         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2489         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2490         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2491         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2492         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2493         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2494         [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2495         [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2496 };
2497
2498 static const struct genl_small_ops dp_vport_genl_ops[] = {
2499         { .cmd = OVS_VPORT_CMD_NEW,
2500           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2501           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2502           .doit = ovs_vport_cmd_new
2503         },
2504         { .cmd = OVS_VPORT_CMD_DEL,
2505           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2506           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2507           .doit = ovs_vport_cmd_del
2508         },
2509         { .cmd = OVS_VPORT_CMD_GET,
2510           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2511           .flags = 0,               /* OK for unprivileged users. */
2512           .doit = ovs_vport_cmd_get,
2513           .dumpit = ovs_vport_cmd_dump
2514         },
2515         { .cmd = OVS_VPORT_CMD_SET,
2516           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2517           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2518           .doit = ovs_vport_cmd_set,
2519         },
2520 };
2521
2522 struct genl_family dp_vport_genl_family __ro_after_init = {
2523         .hdrsize = sizeof(struct ovs_header),
2524         .name = OVS_VPORT_FAMILY,
2525         .version = OVS_VPORT_VERSION,
2526         .maxattr = OVS_VPORT_ATTR_MAX,
2527         .policy = vport_policy,
2528         .netnsok = true,
2529         .parallel_ops = true,
2530         .small_ops = dp_vport_genl_ops,
2531         .n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
2532         .mcgrps = &ovs_dp_vport_multicast_group,
2533         .n_mcgrps = 1,
2534         .module = THIS_MODULE,
2535 };
2536
2537 static struct genl_family * const dp_genl_families[] = {
2538         &dp_datapath_genl_family,
2539         &dp_vport_genl_family,
2540         &dp_flow_genl_family,
2541         &dp_packet_genl_family,
2542         &dp_meter_genl_family,
2543 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2544         &dp_ct_limit_genl_family,
2545 #endif
2546 };
2547
2548 static void dp_unregister_genl(int n_families)
2549 {
2550         int i;
2551
2552         for (i = 0; i < n_families; i++)
2553                 genl_unregister_family(dp_genl_families[i]);
2554 }
2555
2556 static int __init dp_register_genl(void)
2557 {
2558         int err;
2559         int i;
2560
2561         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2562
2563                 err = genl_register_family(dp_genl_families[i]);
2564                 if (err)
2565                         goto error;
2566         }
2567
2568         return 0;
2569
2570 error:
2571         dp_unregister_genl(i);
2572         return err;
2573 }
2574
2575 static int __net_init ovs_init_net(struct net *net)
2576 {
2577         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2578         int err;
2579
2580         INIT_LIST_HEAD(&ovs_net->dps);
2581         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2582         INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2583
2584         err = ovs_ct_init(net);
2585         if (err)
2586                 return err;
2587
2588         schedule_delayed_work(&ovs_net->masks_rebalance,
2589                               msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2590         return 0;
2591 }
2592
2593 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2594                                             struct list_head *head)
2595 {
2596         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2597         struct datapath *dp;
2598
2599         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2600                 int i;
2601
2602                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2603                         struct vport *vport;
2604
2605                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2606                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2607                                         continue;
2608
2609                                 if (dev_net(vport->dev) == dnet)
2610                                         list_add(&vport->detach_list, head);
2611                         }
2612                 }
2613         }
2614 }
2615
2616 static void __net_exit ovs_exit_net(struct net *dnet)
2617 {
2618         struct datapath *dp, *dp_next;
2619         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2620         struct vport *vport, *vport_next;
2621         struct net *net;
2622         LIST_HEAD(head);
2623
2624         ovs_lock();
2625
2626         ovs_ct_exit(dnet);
2627
2628         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2629                 __dp_destroy(dp);
2630
2631         down_read(&net_rwsem);
2632         for_each_net(net)
2633                 list_vports_from_net(net, dnet, &head);
2634         up_read(&net_rwsem);
2635
2636         /* Detach all vports from given namespace. */
2637         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2638                 list_del(&vport->detach_list);
2639                 ovs_dp_detach_port(vport);
2640         }
2641
2642         ovs_unlock();
2643
2644         cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2645         cancel_work_sync(&ovs_net->dp_notify_work);
2646 }
2647
2648 static struct pernet_operations ovs_net_ops = {
2649         .init = ovs_init_net,
2650         .exit = ovs_exit_net,
2651         .id   = &ovs_net_id,
2652         .size = sizeof(struct ovs_net),
2653 };
2654
2655 static int __init dp_init(void)
2656 {
2657         int err;
2658
2659         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2660                      sizeof_field(struct sk_buff, cb));
2661
2662         pr_info("Open vSwitch switching datapath\n");
2663
2664         err = action_fifos_init();
2665         if (err)
2666                 goto error;
2667
2668         err = ovs_internal_dev_rtnl_link_register();
2669         if (err)
2670                 goto error_action_fifos_exit;
2671
2672         err = ovs_flow_init();
2673         if (err)
2674                 goto error_unreg_rtnl_link;
2675
2676         err = ovs_vport_init();
2677         if (err)
2678                 goto error_flow_exit;
2679
2680         err = register_pernet_device(&ovs_net_ops);
2681         if (err)
2682                 goto error_vport_exit;
2683
2684         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2685         if (err)
2686                 goto error_netns_exit;
2687
2688         err = ovs_netdev_init();
2689         if (err)
2690                 goto error_unreg_notifier;
2691
2692         err = dp_register_genl();
2693         if (err < 0)
2694                 goto error_unreg_netdev;
2695
2696         return 0;
2697
2698 error_unreg_netdev:
2699         ovs_netdev_exit();
2700 error_unreg_notifier:
2701         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2702 error_netns_exit:
2703         unregister_pernet_device(&ovs_net_ops);
2704 error_vport_exit:
2705         ovs_vport_exit();
2706 error_flow_exit:
2707         ovs_flow_exit();
2708 error_unreg_rtnl_link:
2709         ovs_internal_dev_rtnl_link_unregister();
2710 error_action_fifos_exit:
2711         action_fifos_exit();
2712 error:
2713         return err;
2714 }
2715
2716 static void dp_cleanup(void)
2717 {
2718         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2719         ovs_netdev_exit();
2720         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2721         unregister_pernet_device(&ovs_net_ops);
2722         rcu_barrier();
2723         ovs_vport_exit();
2724         ovs_flow_exit();
2725         ovs_internal_dev_rtnl_link_unregister();
2726         action_fifos_exit();
2727 }
2728
2729 module_init(dp_init);
2730 module_exit(dp_cleanup);
2731
2732 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2733 MODULE_LICENSE("GPL");
2734 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2735 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2736 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2737 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2738 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2739 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);