net/openvswitch/datapath.c

   1 /*
   2  * Copyright (c) 2007-2014 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/ethtool.h>
  40 #include <linux/wait.h>
  41 #include <asm/div64.h>
  42 #include <linux/highmem.h>
  43 #include <linux/netfilter_bridge.h>
  44 #include <linux/netfilter_ipv4.h>
  45 #include <linux/inetdevice.h>
  46 #include <linux/list.h>
  47 #include <linux/openvswitch.h>
  48 #include <linux/rculist.h>
  49 #include <linux/dmi.h>
  50 #include <net/genetlink.h>
  51 #include <net/net_namespace.h>
  52 #include <net/netns/generic.h>
  53
  54 #include "datapath.h"
  55 #include "flow.h"
  56 #include "flow_table.h"
  57 #include "flow_netlink.h"
  58 #include "vport-internal_dev.h"
  59 #include "vport-netdev.h"
  60
  61 int ovs_net_id __read_mostly;
  62 EXPORT_SYMBOL_GPL(ovs_net_id);
  63
  64 static struct genl_family dp_packet_genl_family;
  65 static struct genl_family dp_flow_genl_family;
  66 static struct genl_family dp_datapath_genl_family;
  67
  68 static const struct nla_policy flow_policy[];
  69
  70 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
  71         .name = OVS_FLOW_MCGROUP,
  72 };
  73
  74 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  75         .name = OVS_DATAPATH_MCGROUP,
  76 };
  77
  78 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
  79         .name = OVS_VPORT_MCGROUP,
  80 };
  81
  82 /* Check if need to build a reply message.
  83  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  84 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  85                             unsigned int group)
  86 {
  87         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  88                genl_has_listeners(family, genl_info_net(info), group);
  89 }
  90
  91 static void ovs_notify(struct genl_family *family,
  92                        struct sk_buff *skb, struct genl_info *info)
  93 {
  94         genl_notify(family, skb, genl_info_net(info), info->snd_portid,
  95                     0, info->nlhdr, GFP_KERNEL);
  96 }
  97
  98 /**
  99  * DOC: Locking:
 100  *
 101  * All writes e.g. Writes to device state (add/remove datapath, port, set
 102  * operations on vports, etc.), Writes to other state (flow table
 103  * modifications, set miscellaneous datapath parameters, etc.) are protected
 104  * by ovs_lock.
 105  *
 106  * Reads are protected by RCU.
 107  *
 108  * There are a few special cases (mostly stats) that have their own
 109  * synchronization but they nest under all of above and don't interact with
 110  * each other.
 111  *
 112  * The RTNL lock nests inside ovs_mutex.
 113  */
 114
 115 static DEFINE_MUTEX(ovs_mutex);
 116
 117 void ovs_lock(void)
 118 {
 119         mutex_lock(&ovs_mutex);
 120 }
 121
 122 void ovs_unlock(void)
 123 {
 124         mutex_unlock(&ovs_mutex);
 125 }
 126
 127 #ifdef CONFIG_LOCKDEP
 128 int lockdep_ovsl_is_held(void)
 129 {
 130         if (debug_locks)
 131                 return lockdep_is_held(&ovs_mutex);
 132         else
 133                 return 1;
 134 }
 135 EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held);
 136 #endif
 137
 138 static struct vport *new_vport(const struct vport_parms *);
 139 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 140                              const struct sw_flow_key *,
 141                              const struct dp_upcall_info *);
 142 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 143                                   const struct sw_flow_key *,
 144                                   const struct dp_upcall_info *);
 145
 146 /* Must be called with rcu_read_lock. */
 147 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
 148 {
 149         struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
 150
 151         if (dev) {
 152                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 153                 if (vport)
 154                         return vport->dp;
 155         }
 156
 157         return NULL;
 158 }
 159
 160 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
 161  * returned dp pointer valid.
 162  */
 163 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
 164 {
 165         struct datapath *dp;
 166
 167         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
 168         rcu_read_lock();
 169         dp = get_dp_rcu(net, dp_ifindex);
 170         rcu_read_unlock();
 171
 172         return dp;
 173 }
 174
 175 /* Must be called with rcu_read_lock or ovs_mutex. */
 176 const char *ovs_dp_name(const struct datapath *dp)
 177 {
 178         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 179         return ovs_vport_name(vport);
 180 }
 181
 182 static int get_dpifindex(const struct datapath *dp)
 183 {
 184         struct vport *local;
 185         int ifindex;
 186
 187         rcu_read_lock();
 188
 189         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 190         if (local)
 191                 ifindex = local->dev->ifindex;
 192         else
 193                 ifindex = 0;
 194
 195         rcu_read_unlock();
 196
 197         return ifindex;
 198 }
 199
 200 static void destroy_dp_rcu(struct rcu_head *rcu)
 201 {
 202         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 203
 204         ovs_flow_tbl_destroy(&dp->table);
 205         free_percpu(dp->stats_percpu);
 206         kfree(dp->ports);
 207         kfree(dp);
 208 }
 209
 210 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 211                                             u16 port_no)
 212 {
 213         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 214 }
 215
 216 /* Called with ovs_mutex or RCU read lock. */
 217 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 218 {
 219         struct vport *vport;
 220         struct hlist_head *head;
 221
 222         head = vport_hash_bucket(dp, port_no);
 223         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 224                 if (vport->port_no == port_no)
 225                         return vport;
 226         }
 227         return NULL;
 228 }
 229
 230 /* Called with ovs_mutex. */
 231 static struct vport *new_vport(const struct vport_parms *parms)
 232 {
 233         struct vport *vport;
 234
 235         vport = ovs_vport_add(parms);
 236         if (!IS_ERR(vport)) {
 237                 struct datapath *dp = parms->dp;
 238                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 239
 240                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 241         }
 242         return vport;
 243 }
 244
 245 void ovs_dp_detach_port(struct vport *p)
 246 {
 247         ASSERT_OVSL();
 248
 249         /* First drop references to device. */
 250         hlist_del_rcu(&p->dp_hash_node);
 251
 252         /* Then destroy it. */
 253         ovs_vport_del(p);
 254 }
 255
 256 /* Must be called with rcu_read_lock. */
 257 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 258 {
 259         const struct vport *p = OVS_CB(skb)->input_vport;
 260         struct datapath *dp = p->dp;
 261         struct sw_flow *flow;
 262         struct sw_flow_actions *sf_acts;
 263         struct dp_stats_percpu *stats;
 264         u64 *stats_counter;
 265         u32 n_mask_hit;
 266
 267         stats = this_cpu_ptr(dp->stats_percpu);
 268
 269         /* Look up flow. */
 270         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
 271         if (unlikely(!flow)) {
 272                 struct dp_upcall_info upcall;
 273                 int error;
 274
 275                 memset(&upcall, 0, sizeof(upcall));
 276                 upcall.cmd = OVS_PACKET_CMD_MISS;
 277                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 278                 error = ovs_dp_upcall(dp, skb, key, &upcall);
 279                 if (unlikely(error))
 280                         kfree_skb(skb);
 281                 else
 282                         consume_skb(skb);
 283                 stats_counter = &stats->n_missed;
 284                 goto out;
 285         }
 286
 287         ovs_flow_stats_update(flow, key->tp.flags, skb);
 288         sf_acts = rcu_dereference(flow->sf_acts);
 289         ovs_execute_actions(dp, skb, sf_acts, key);
 290
 291         stats_counter = &stats->n_hit;
 292
 293 out:
 294         /* Update datapath statistics. */
 295         u64_stats_update_begin(&stats->syncp);
 296         (*stats_counter)++;
 297         stats->n_mask_hit += n_mask_hit;
 298         u64_stats_update_end(&stats->syncp);
 299 }
 300
 301 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 302                   const struct sw_flow_key *key,
 303                   const struct dp_upcall_info *upcall_info)
 304 {
 305         struct dp_stats_percpu *stats;
 306         int err;
 307
 308         if (upcall_info->portid == 0) {
 309                 err = -ENOTCONN;
 310                 goto err;
 311         }
 312
 313         if (!skb_is_gso(skb))
 314                 err = queue_userspace_packet(dp, skb, key, upcall_info);
 315         else
 316                 err = queue_gso_packets(dp, skb, key, upcall_info);
 317         if (err)
 318                 goto err;
 319
 320         return 0;
 321
 322 err:
 323         stats = this_cpu_ptr(dp->stats_percpu);
 324
 325         u64_stats_update_begin(&stats->syncp);
 326         stats->n_lost++;
 327         u64_stats_update_end(&stats->syncp);
 328
 329         return err;
 330 }
 331
 332 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 333                              const struct sw_flow_key *key,
 334                              const struct dp_upcall_info *upcall_info)
 335 {
 336         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 337         struct sw_flow_key later_key;
 338         struct sk_buff *segs, *nskb;
 339         struct ovs_skb_cb ovs_cb;
 340         int err;
 341
 342         ovs_cb = *OVS_CB(skb);
 343         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 344         *OVS_CB(skb) = ovs_cb;
 345         if (IS_ERR(segs))
 346                 return PTR_ERR(segs);
 347         if (segs == NULL)
 348                 return -EINVAL;
 349
 350         if (gso_type & SKB_GSO_UDP) {
 351                 /* The initial flow key extracted by ovs_flow_key_extract()
 352                  * in this case is for a first fragment, so we need to
 353                  * properly mark later fragments.
 354                  */
 355                 later_key = *key;
 356                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 357         }
 358
 359         /* Queue all of the segments. */
 360         skb = segs;
 361         do {
 362                 *OVS_CB(skb) = ovs_cb;
 363                 if (gso_type & SKB_GSO_UDP && skb != segs)
 364                         key = &later_key;
 365
 366                 err = queue_userspace_packet(dp, skb, key, upcall_info);
 367                 if (err)
 368                         break;
 369
 370         } while ((skb = skb->next));
 371
 372         /* Free all of the segments. */
 373         skb = segs;
 374         do {
 375                 nskb = skb->next;
 376                 if (err)
 377                         kfree_skb(skb);
 378                 else
 379                         consume_skb(skb);
 380         } while ((skb = nskb));
 381         return err;
 382 }
 383
 384 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 385                               unsigned int hdrlen)
 386 {
 387         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 388                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 389                 + nla_total_size(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 390
 391         /* OVS_PACKET_ATTR_USERDATA */
 392         if (upcall_info->userdata)
 393                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 394
 395         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 396         if (upcall_info->egress_tun_info)
 397                 size += nla_total_size(ovs_tun_key_attr_size());
 398
 399         /* OVS_PACKET_ATTR_ACTIONS */
 400         if (upcall_info->actions_len)
 401                 size += nla_total_size(upcall_info->actions_len);
 402
 403         return size;
 404 }
 405
 406 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 407                                   const struct sw_flow_key *key,
 408                                   const struct dp_upcall_info *upcall_info)
 409 {
 410         struct ovs_header *upcall;
 411         struct sk_buff *nskb = NULL;
 412         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 413         struct nlattr *nla;
 414         struct genl_info info = {
 415                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 416                 .snd_portid = upcall_info->portid,
 417         };
 418         size_t len;
 419         unsigned int hlen;
 420         int err, dp_ifindex;
 421
 422         dp_ifindex = get_dpifindex(dp);
 423         if (!dp_ifindex)
 424                 return -ENODEV;
 425
 426         if (skb_vlan_tag_present(skb)) {
 427                 nskb = skb_clone(skb, GFP_ATOMIC);
 428                 if (!nskb)
 429                         return -ENOMEM;
 430
 431                 nskb = __vlan_hwaccel_push_inside(nskb);
 432                 if (!nskb)
 433                         return -ENOMEM;
 434
 435                 skb = nskb;
 436         }
 437
 438         if (nla_attr_size(skb->len) > USHRT_MAX) {
 439                 err = -EFBIG;
 440                 goto out;
 441         }
 442
 443         /* Complete checksum if needed */
 444         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 445             (err = skb_checksum_help(skb)))
 446                 goto out;
 447
 448         /* Older versions of OVS user space enforce alignment of the last
 449          * Netlink attribute to NLA_ALIGNTO which would require extensive
 450          * padding logic. Only perform zerocopy if padding is not required.
 451          */
 452         if (dp->user_features & OVS_DP_F_UNALIGNED)
 453                 hlen = skb_zerocopy_headlen(skb);
 454         else
 455                 hlen = skb->len;
 456
 457         len = upcall_msg_size(upcall_info, hlen);
 458         user_skb = genlmsg_new_unicast(len, &info, 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         upcall->dp_ifindex = dp_ifindex;
 467
 468         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 469         BUG_ON(err);
 470
 471         if (upcall_info->userdata)
 472                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 473                           nla_len(upcall_info->userdata),
 474                           nla_data(upcall_info->userdata));
 475
 476         if (upcall_info->egress_tun_info) {
 477                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 478                 err = ovs_nla_put_egress_tunnel_key(user_skb,
 479                                                     upcall_info->egress_tun_info);
 480                 BUG_ON(err);
 481                 nla_nest_end(user_skb, nla);
 482         }
 483
 484         if (upcall_info->actions_len) {
 485                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 486                 err = ovs_nla_put_actions(upcall_info->actions,
 487                                           upcall_info->actions_len,
 488                                           user_skb);
 489                 if (!err)
 490                         nla_nest_end(user_skb, nla);
 491                 else
 492                         nla_nest_cancel(user_skb, nla);
 493         }
 494
 495         /* Only reserve room for attribute header, packet data is added
 496          * in skb_zerocopy() */
 497         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 498                 err = -ENOBUFS;
 499                 goto out;
 500         }
 501         nla->nla_len = nla_attr_size(skb->len);
 502
 503         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 504         if (err)
 505                 goto out;
 506
 507         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 508         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 509                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 510
 511                 if (plen > 0)
 512                         memset(skb_put(user_skb, plen), 0, plen);
 513         }
 514
 515         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 516
 517         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 518         user_skb = NULL;
 519 out:
 520         if (err)
 521                 skb_tx_error(skb);
 522         kfree_skb(user_skb);
 523         kfree_skb(nskb);
 524         return err;
 525 }
 526
 527 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 528 {
 529         struct ovs_header *ovs_header = info->userhdr;
 530         struct nlattr **a = info->attrs;
 531         struct sw_flow_actions *acts;
 532         struct sk_buff *packet;
 533         struct sw_flow *flow;
 534         struct sw_flow_actions *sf_acts;
 535         struct datapath *dp;
 536         struct ethhdr *eth;
 537         struct vport *input_vport;
 538         int len;
 539         int err;
 540         bool log = !a[OVS_PACKET_ATTR_PROBE];
 541
 542         err = -EINVAL;
 543         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 544             !a[OVS_PACKET_ATTR_ACTIONS])
 545                 goto err;
 546
 547         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 548         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 549         err = -ENOMEM;
 550         if (!packet)
 551                 goto err;
 552         skb_reserve(packet, NET_IP_ALIGN);
 553
 554         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 555
 556         skb_reset_mac_header(packet);
 557         eth = eth_hdr(packet);
 558
 559         /* Normally, setting the skb 'protocol' field would be handled by a
 560          * call to eth_type_trans(), but it assumes there's a sending
 561          * device, which we may not have. */
 562         if (eth_proto_is_802_3(eth->h_proto))
 563                 packet->protocol = eth->h_proto;
 564         else
 565                 packet->protocol = htons(ETH_P_802_2);
 566
 567         /* Build an sw_flow for sending this packet. */
 568         flow = ovs_flow_alloc();
 569         err = PTR_ERR(flow);
 570         if (IS_ERR(flow))
 571                 goto err_kfree_skb;
 572
 573         err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
 574                                              &flow->key, log);
 575         if (err)
 576                 goto err_flow_free;
 577
 578         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 579                                    &flow->key, &acts, log);
 580         if (err)
 581                 goto err_flow_free;
 582
 583         rcu_assign_pointer(flow->sf_acts, acts);
 584         OVS_CB(packet)->egress_tun_info = NULL;
 585         packet->priority = flow->key.phy.priority;
 586         packet->mark = flow->key.phy.skb_mark;
 587
 588         rcu_read_lock();
 589         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
 590         err = -ENODEV;
 591         if (!dp)
 592                 goto err_unlock;
 593
 594         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 595         if (!input_vport)
 596                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 597
 598         if (!input_vport)
 599                 goto err_unlock;
 600
 601         OVS_CB(packet)->input_vport = input_vport;
 602         sf_acts = rcu_dereference(flow->sf_acts);
 603
 604         local_bh_disable();
 605         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 606         local_bh_enable();
 607         rcu_read_unlock();
 608
 609         ovs_flow_free(flow, false);
 610         return err;
 611
 612 err_unlock:
 613         rcu_read_unlock();
 614 err_flow_free:
 615         ovs_flow_free(flow, false);
 616 err_kfree_skb:
 617         kfree_skb(packet);
 618 err:
 619         return err;
 620 }
 621
 622 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 623         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 624         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 625         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 626         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 627 };
 628
 629 static const struct genl_ops dp_packet_genl_ops[] = {
 630         { .cmd = OVS_PACKET_CMD_EXECUTE,
 631           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 632           .policy = packet_policy,
 633           .doit = ovs_packet_cmd_execute
 634         }
 635 };
 636
 637 static struct genl_family dp_packet_genl_family = {
 638         .id = GENL_ID_GENERATE,
 639         .hdrsize = sizeof(struct ovs_header),
 640         .name = OVS_PACKET_FAMILY,
 641         .version = OVS_PACKET_VERSION,
 642         .maxattr = OVS_PACKET_ATTR_MAX,
 643         .netnsok = true,
 644         .parallel_ops = true,
 645         .ops = dp_packet_genl_ops,
 646         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 647 };
 648
 649 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 650                          struct ovs_dp_megaflow_stats *mega_stats)
 651 {
 652         int i;
 653
 654         memset(mega_stats, 0, sizeof(*mega_stats));
 655
 656         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 657         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 658
 659         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 660
 661         for_each_possible_cpu(i) {
 662                 const struct dp_stats_percpu *percpu_stats;
 663                 struct dp_stats_percpu local_stats;
 664                 unsigned int start;
 665
 666                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 667
 668                 do {
 669                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 670                         local_stats = *percpu_stats;
 671                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 672
 673                 stats->n_hit += local_stats.n_hit;
 674                 stats->n_missed += local_stats.n_missed;
 675                 stats->n_lost += local_stats.n_lost;
 676                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 677         }
 678 }
 679
 680 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 681 {
 682         return ovs_identifier_is_ufid(sfid) &&
 683                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 684 }
 685
 686 static bool should_fill_mask(uint32_t ufid_flags)
 687 {
 688         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 689 }
 690
 691 static bool should_fill_actions(uint32_t ufid_flags)
 692 {
 693         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 694 }
 695
 696 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 697                                     const struct sw_flow_id *sfid,
 698                                     uint32_t ufid_flags)
 699 {
 700         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 701
 702         /* OVS_FLOW_ATTR_UFID */
 703         if (sfid && ovs_identifier_is_ufid(sfid))
 704                 len += nla_total_size(sfid->ufid_len);
 705
 706         /* OVS_FLOW_ATTR_KEY */
 707         if (!sfid || should_fill_key(sfid, ufid_flags))
 708                 len += nla_total_size(ovs_key_attr_size());
 709
 710         /* OVS_FLOW_ATTR_MASK */
 711         if (should_fill_mask(ufid_flags))
 712                 len += nla_total_size(ovs_key_attr_size());
 713
 714         /* OVS_FLOW_ATTR_ACTIONS */
 715         if (should_fill_actions(ufid_flags))
 716                 len += nla_total_size(acts->orig_len);
 717
 718         return len
 719                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 720                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 721                 + nla_total_size(8); /* OVS_FLOW_ATTR_USED */
 722 }
 723
 724 /* Called with ovs_mutex or RCU read lock. */
 725 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 726                                    struct sk_buff *skb)
 727 {
 728         struct ovs_flow_stats stats;
 729         __be16 tcp_flags;
 730         unsigned long used;
 731
 732         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 733
 734         if (used &&
 735             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 736                 return -EMSGSIZE;
 737
 738         if (stats.n_packets &&
 739             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 740                 return -EMSGSIZE;
 741
 742         if ((u8)ntohs(tcp_flags) &&
 743              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 744                 return -EMSGSIZE;
 745
 746         return 0;
 747 }
 748
 749 /* Called with ovs_mutex or RCU read lock. */
 750 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 751                                      struct sk_buff *skb, int skb_orig_len)
 752 {
 753         struct nlattr *start;
 754         int err;
 755
 756         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 757          * this is the first flow to be dumped into 'skb'.  This is unusual for
 758          * Netlink but individual action lists can be longer than
 759          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 760          * The userspace caller can always fetch the actions separately if it
 761          * really wants them.  (Most userspace callers in fact don't care.)
 762          *
 763          * This can only fail for dump operations because the skb is always
 764          * properly sized for single flows.
 765          */
 766         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 767         if (start) {
 768                 const struct sw_flow_actions *sf_acts;
 769
 770                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 771                 err = ovs_nla_put_actions(sf_acts->actions,
 772                                           sf_acts->actions_len, skb);
 773
 774                 if (!err)
 775                         nla_nest_end(skb, start);
 776                 else {
 777                         if (skb_orig_len)
 778                                 return err;
 779
 780                         nla_nest_cancel(skb, start);
 781                 }
 782         } else if (skb_orig_len) {
 783                 return -EMSGSIZE;
 784         }
 785
 786         return 0;
 787 }
 788
 789 /* Called with ovs_mutex or RCU read lock. */
 790 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 791                                   struct sk_buff *skb, u32 portid,
 792                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 793 {
 794         const int skb_orig_len = skb->len;
 795         struct ovs_header *ovs_header;
 796         int err;
 797
 798         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 799                                  flags, cmd);
 800         if (!ovs_header)
 801                 return -EMSGSIZE;
 802
 803         ovs_header->dp_ifindex = dp_ifindex;
 804
 805         err = ovs_nla_put_identifier(flow, skb);
 806         if (err)
 807                 goto error;
 808
 809         if (should_fill_key(&flow->id, ufid_flags)) {
 810                 err = ovs_nla_put_masked_key(flow, skb);
 811                 if (err)
 812                         goto error;
 813         }
 814
 815         if (should_fill_mask(ufid_flags)) {
 816                 err = ovs_nla_put_mask(flow, skb);
 817                 if (err)
 818                         goto error;
 819         }
 820
 821         err = ovs_flow_cmd_fill_stats(flow, skb);
 822         if (err)
 823                 goto error;
 824
 825         if (should_fill_actions(ufid_flags)) {
 826                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 827                 if (err)
 828                         goto error;
 829         }
 830
 831         genlmsg_end(skb, ovs_header);
 832         return 0;
 833
 834 error:
 835         genlmsg_cancel(skb, ovs_header);
 836         return err;
 837 }
 838
 839 /* May not be called with RCU read lock. */
 840 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 841                                                const struct sw_flow_id *sfid,
 842                                                struct genl_info *info,
 843                                                bool always,
 844                                                uint32_t ufid_flags)
 845 {
 846         struct sk_buff *skb;
 847         size_t len;
 848
 849         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 850                 return NULL;
 851
 852         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 853         skb = genlmsg_new_unicast(len, info, GFP_KERNEL);
 854         if (!skb)
 855                 return ERR_PTR(-ENOMEM);
 856
 857         return skb;
 858 }
 859
 860 /* Called with ovs_mutex. */
 861 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 862                                                int dp_ifindex,
 863                                                struct genl_info *info, u8 cmd,
 864                                                bool always, u32 ufid_flags)
 865 {
 866         struct sk_buff *skb;
 867         int retval;
 868
 869         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 870                                       &flow->id, info, always, ufid_flags);
 871         if (IS_ERR_OR_NULL(skb))
 872                 return skb;
 873
 874         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 875                                         info->snd_portid, info->snd_seq, 0,
 876                                         cmd, ufid_flags);
 877         BUG_ON(retval < 0);
 878         return skb;
 879 }
 880
 881 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 882 {
 883         struct nlattr **a = info->attrs;
 884         struct ovs_header *ovs_header = info->userhdr;
 885         struct sw_flow *flow = NULL, *new_flow;
 886         struct sw_flow_mask mask;
 887         struct sk_buff *reply;
 888         struct datapath *dp;
 889         struct sw_flow_key key;
 890         struct sw_flow_actions *acts;
 891         struct sw_flow_match match;
 892         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 893         int error;
 894         bool log = !a[OVS_FLOW_ATTR_PROBE];
 895
 896         /* Must have key and actions. */
 897         error = -EINVAL;
 898         if (!a[OVS_FLOW_ATTR_KEY]) {
 899                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 900                 goto error;
 901         }
 902         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 903                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 904                 goto error;
 905         }
 906
 907         /* Most of the time we need to allocate a new flow, do it before
 908          * locking.
 909          */
 910         new_flow = ovs_flow_alloc();
 911         if (IS_ERR(new_flow)) {
 912                 error = PTR_ERR(new_flow);
 913                 goto error;
 914         }
 915
 916         /* Extract key. */
 917         ovs_match_init(&match, &key, &mask);
 918         error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
 919                                   a[OVS_FLOW_ATTR_MASK], log);
 920         if (error)
 921                 goto err_kfree_flow;
 922
 923         ovs_flow_mask_key(&new_flow->key, &key, &mask);
 924
 925         /* Extract flow identifier. */
 926         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 927                                        &key, log);
 928         if (error)
 929                 goto err_kfree_flow;
 930
 931         /* Validate actions. */
 932         error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
 933                                      &acts, log);
 934         if (error) {
 935                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 936                 goto err_kfree_flow;
 937         }
 938
 939         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 940                                         ufid_flags);
 941         if (IS_ERR(reply)) {
 942                 error = PTR_ERR(reply);
 943                 goto err_kfree_acts;
 944         }
 945
 946         ovs_lock();
 947         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 948         if (unlikely(!dp)) {
 949                 error = -ENODEV;
 950                 goto err_unlock_ovs;
 951         }
 952
 953         /* Check if this is a duplicate flow */
 954         if (ovs_identifier_is_ufid(&new_flow->id))
 955                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
 956         if (!flow)
 957                 flow = ovs_flow_tbl_lookup(&dp->table, &key);
 958         if (likely(!flow)) {
 959                 rcu_assign_pointer(new_flow->sf_acts, acts);
 960
 961                 /* Put flow in bucket. */
 962                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 963                 if (unlikely(error)) {
 964                         acts = NULL;
 965                         goto err_unlock_ovs;
 966                 }
 967
 968                 if (unlikely(reply)) {
 969                         error = ovs_flow_cmd_fill_info(new_flow,
 970                                                        ovs_header->dp_ifindex,
 971                                                        reply, info->snd_portid,
 972                                                        info->snd_seq, 0,
 973                                                        OVS_FLOW_CMD_NEW,
 974                                                        ufid_flags);
 975                         BUG_ON(error < 0);
 976                 }
 977                 ovs_unlock();
 978         } else {
 979                 struct sw_flow_actions *old_acts;
 980
 981                 /* Bail out if we're not allowed to modify an existing flow.
 982                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 983                  * because Generic Netlink treats the latter as a dump
 984                  * request.  We also accept NLM_F_EXCL in case that bug ever
 985                  * gets fixed.
 986                  */
 987                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
 988                                                          | NLM_F_EXCL))) {
 989                         error = -EEXIST;
 990                         goto err_unlock_ovs;
 991                 }
 992                 /* The flow identifier has to be the same for flow updates.
 993                  * Look for any overlapping flow.
 994                  */
 995                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
 996                         if (ovs_identifier_is_key(&flow->id))
 997                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
 998                                                                  &match);
 999                         else /* UFID matches but key is different */
1000                                 flow = NULL;
1001                         if (!flow) {
1002                                 error = -ENOENT;
1003                                 goto err_unlock_ovs;
1004                         }
1005                 }
1006                 /* Update actions. */
1007                 old_acts = ovsl_dereference(flow->sf_acts);
1008                 rcu_assign_pointer(flow->sf_acts, acts);
1009
1010                 if (unlikely(reply)) {
1011                         error = ovs_flow_cmd_fill_info(flow,
1012                                                        ovs_header->dp_ifindex,
1013                                                        reply, info->snd_portid,
1014                                                        info->snd_seq, 0,
1015                                                        OVS_FLOW_CMD_NEW,
1016                                                        ufid_flags);
1017                         BUG_ON(error < 0);
1018                 }
1019                 ovs_unlock();
1020
1021                 ovs_nla_free_flow_actions_rcu(old_acts);
1022                 ovs_flow_free(new_flow, false);
1023         }
1024
1025         if (reply)
1026                 ovs_notify(&dp_flow_genl_family, reply, info);
1027         return 0;
1028
1029 err_unlock_ovs:
1030         ovs_unlock();
1031         kfree_skb(reply);
1032 err_kfree_acts:
1033         ovs_nla_free_flow_actions(acts);
1034 err_kfree_flow:
1035         ovs_flow_free(new_flow, false);
1036 error:
1037         return error;
1038 }
1039
1040 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1041 static struct sw_flow_actions *get_flow_actions(const struct nlattr *a,
1042                                                 const struct sw_flow_key *key,
1043                                                 const struct sw_flow_mask *mask,
1044                                                 bool log)
1045 {
1046         struct sw_flow_actions *acts;
1047         struct sw_flow_key masked_key;
1048         int error;
1049
1050         ovs_flow_mask_key(&masked_key, key, mask);
1051         error = ovs_nla_copy_actions(a, &masked_key, &acts, log);
1052         if (error) {
1053                 OVS_NLERR(log,
1054                           "Actions may not be safe on all matching packets");
1055                 return ERR_PTR(error);
1056         }
1057
1058         return acts;
1059 }
1060
1061 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1062 {
1063         struct nlattr **a = info->attrs;
1064         struct ovs_header *ovs_header = info->userhdr;
1065         struct sw_flow_key key;
1066         struct sw_flow *flow;
1067         struct sw_flow_mask mask;
1068         struct sk_buff *reply = NULL;
1069         struct datapath *dp;
1070         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1071         struct sw_flow_match match;
1072         struct sw_flow_id sfid;
1073         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1074         int error;
1075         bool log = !a[OVS_FLOW_ATTR_PROBE];
1076         bool ufid_present;
1077
1078         /* Extract key. */
1079         error = -EINVAL;
1080         if (!a[OVS_FLOW_ATTR_KEY]) {
1081                 OVS_NLERR(log, "Flow key attribute not present in set flow.");
1082                 goto error;
1083         }
1084
1085         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1086         ovs_match_init(&match, &key, &mask);
1087         error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
1088                                   a[OVS_FLOW_ATTR_MASK], log);
1089         if (error)
1090                 goto error;
1091
1092         /* Validate actions. */
1093         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1094                 acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask,
1095                                         log);
1096                 if (IS_ERR(acts)) {
1097                         error = PTR_ERR(acts);
1098                         goto error;
1099                 }
1100
1101                 /* Can allocate before locking if have acts. */
1102                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1103                                                 ufid_flags);
1104                 if (IS_ERR(reply)) {
1105                         error = PTR_ERR(reply);
1106                         goto err_kfree_acts;
1107                 }
1108         }
1109
1110         ovs_lock();
1111         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1112         if (unlikely(!dp)) {
1113                 error = -ENODEV;
1114                 goto err_unlock_ovs;
1115         }
1116         /* Check that the flow exists. */
1117         if (ufid_present)
1118                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1119         else
1120                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1121         if (unlikely(!flow)) {
1122                 error = -ENOENT;
1123                 goto err_unlock_ovs;
1124         }
1125
1126         /* Update actions, if present. */
1127         if (likely(acts)) {
1128                 old_acts = ovsl_dereference(flow->sf_acts);
1129                 rcu_assign_pointer(flow->sf_acts, acts);
1130
1131                 if (unlikely(reply)) {
1132                         error = ovs_flow_cmd_fill_info(flow,
1133                                                        ovs_header->dp_ifindex,
1134                                                        reply, info->snd_portid,
1135                                                        info->snd_seq, 0,
1136                                                        OVS_FLOW_CMD_NEW,
1137                                                        ufid_flags);
1138                         BUG_ON(error < 0);
1139                 }
1140         } else {
1141                 /* Could not alloc without acts before locking. */
1142                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1143                                                 info, OVS_FLOW_CMD_NEW, false,
1144                                                 ufid_flags);
1145
1146                 if (unlikely(IS_ERR(reply))) {
1147                         error = PTR_ERR(reply);
1148                         goto err_unlock_ovs;
1149                 }
1150         }
1151
1152         /* Clear stats. */
1153         if (a[OVS_FLOW_ATTR_CLEAR])
1154                 ovs_flow_stats_clear(flow);
1155         ovs_unlock();
1156
1157         if (reply)
1158                 ovs_notify(&dp_flow_genl_family, reply, info);
1159         if (old_acts)
1160                 ovs_nla_free_flow_actions_rcu(old_acts);
1161
1162         return 0;
1163
1164 err_unlock_ovs:
1165         ovs_unlock();
1166         kfree_skb(reply);
1167 err_kfree_acts:
1168         ovs_nla_free_flow_actions(acts);
1169 error:
1170         return error;
1171 }
1172
1173 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1174 {
1175         struct nlattr **a = info->attrs;
1176         struct ovs_header *ovs_header = info->userhdr;
1177         struct sw_flow_key key;
1178         struct sk_buff *reply;
1179         struct sw_flow *flow;
1180         struct datapath *dp;
1181         struct sw_flow_match match;
1182         struct sw_flow_id ufid;
1183         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1184         int err = 0;
1185         bool log = !a[OVS_FLOW_ATTR_PROBE];
1186         bool ufid_present;
1187
1188         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1189         if (a[OVS_FLOW_ATTR_KEY]) {
1190                 ovs_match_init(&match, &key, NULL);
1191                 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL,
1192                                         log);
1193         } else if (!ufid_present) {
1194                 OVS_NLERR(log,
1195                           "Flow get message rejected, Key attribute missing.");
1196                 err = -EINVAL;
1197         }
1198         if (err)
1199                 return err;
1200
1201         ovs_lock();
1202         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1203         if (!dp) {
1204                 err = -ENODEV;
1205                 goto unlock;
1206         }
1207
1208         if (ufid_present)
1209                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1210         else
1211                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1212         if (!flow) {
1213                 err = -ENOENT;
1214                 goto unlock;
1215         }
1216
1217         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1218                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1219         if (IS_ERR(reply)) {
1220                 err = PTR_ERR(reply);
1221                 goto unlock;
1222         }
1223
1224         ovs_unlock();
1225         return genlmsg_reply(reply, info);
1226 unlock:
1227         ovs_unlock();
1228         return err;
1229 }
1230
1231 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1232 {
1233         struct nlattr **a = info->attrs;
1234         struct ovs_header *ovs_header = info->userhdr;
1235         struct sw_flow_key key;
1236         struct sk_buff *reply;
1237         struct sw_flow *flow = NULL;
1238         struct datapath *dp;
1239         struct sw_flow_match match;
1240         struct sw_flow_id ufid;
1241         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1242         int err;
1243         bool log = !a[OVS_FLOW_ATTR_PROBE];
1244         bool ufid_present;
1245
1246         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1247         if (a[OVS_FLOW_ATTR_KEY]) {
1248                 ovs_match_init(&match, &key, NULL);
1249                 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL,
1250                                         log);
1251                 if (unlikely(err))
1252                         return err;
1253         }
1254
1255         ovs_lock();
1256         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1257         if (unlikely(!dp)) {
1258                 err = -ENODEV;
1259                 goto unlock;
1260         }
1261
1262         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1263                 err = ovs_flow_tbl_flush(&dp->table);
1264                 goto unlock;
1265         }
1266
1267         if (ufid_present)
1268                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1269         else
1270                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1271         if (unlikely(!flow)) {
1272                 err = -ENOENT;
1273                 goto unlock;
1274         }
1275
1276         ovs_flow_tbl_remove(&dp->table, flow);
1277         ovs_unlock();
1278
1279         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1280                                         &flow->id, info, false, ufid_flags);
1281         if (likely(reply)) {
1282                 if (likely(!IS_ERR(reply))) {
1283                         rcu_read_lock();        /*To keep RCU checker happy. */
1284                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1285                                                      reply, info->snd_portid,
1286                                                      info->snd_seq, 0,
1287                                                      OVS_FLOW_CMD_DEL,
1288                                                      ufid_flags);
1289                         rcu_read_unlock();
1290                         BUG_ON(err < 0);
1291
1292                         ovs_notify(&dp_flow_genl_family, reply, info);
1293                 } else {
1294                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1295                 }
1296         }
1297
1298         ovs_flow_free(flow, true);
1299         return 0;
1300 unlock:
1301         ovs_unlock();
1302         return err;
1303 }
1304
1305 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1306 {
1307         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1308         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1309         struct table_instance *ti;
1310         struct datapath *dp;
1311         u32 ufid_flags;
1312         int err;
1313
1314         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1315                             OVS_FLOW_ATTR_MAX, flow_policy);
1316         if (err)
1317                 return err;
1318         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1319
1320         rcu_read_lock();
1321         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1322         if (!dp) {
1323                 rcu_read_unlock();
1324                 return -ENODEV;
1325         }
1326
1327         ti = rcu_dereference(dp->table.ti);
1328         for (;;) {
1329                 struct sw_flow *flow;
1330                 u32 bucket, obj;
1331
1332                 bucket = cb->args[0];
1333                 obj = cb->args[1];
1334                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1335                 if (!flow)
1336                         break;
1337
1338                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1339                                            NETLINK_CB(cb->skb).portid,
1340                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1341                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1342                         break;
1343
1344                 cb->args[0] = bucket;
1345                 cb->args[1] = obj;
1346         }
1347         rcu_read_unlock();
1348         return skb->len;
1349 }
1350
1351 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1352         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1353         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1354         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1355         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1356         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1357         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1358         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1359 };
1360
1361 static const struct genl_ops dp_flow_genl_ops[] = {
1362         { .cmd = OVS_FLOW_CMD_NEW,
1363           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1364           .policy = flow_policy,
1365           .doit = ovs_flow_cmd_new
1366         },
1367         { .cmd = OVS_FLOW_CMD_DEL,
1368           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1369           .policy = flow_policy,
1370           .doit = ovs_flow_cmd_del
1371         },
1372         { .cmd = OVS_FLOW_CMD_GET,
1373           .flags = 0,               /* OK for unprivileged users. */
1374           .policy = flow_policy,
1375           .doit = ovs_flow_cmd_get,
1376           .dumpit = ovs_flow_cmd_dump
1377         },
1378         { .cmd = OVS_FLOW_CMD_SET,
1379           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1380           .policy = flow_policy,
1381           .doit = ovs_flow_cmd_set,
1382         },
1383 };
1384
1385 static struct genl_family dp_flow_genl_family = {
1386         .id = GENL_ID_GENERATE,
1387         .hdrsize = sizeof(struct ovs_header),
1388         .name = OVS_FLOW_FAMILY,
1389         .version = OVS_FLOW_VERSION,
1390         .maxattr = OVS_FLOW_ATTR_MAX,
1391         .netnsok = true,
1392         .parallel_ops = true,
1393         .ops = dp_flow_genl_ops,
1394         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1395         .mcgrps = &ovs_dp_flow_multicast_group,
1396         .n_mcgrps = 1,
1397 };
1398
1399 static size_t ovs_dp_cmd_msg_size(void)
1400 {
1401         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1402
1403         msgsize += nla_total_size(IFNAMSIZ);
1404         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1405         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1406         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1407
1408         return msgsize;
1409 }
1410
1411 /* Called with ovs_mutex. */
1412 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1413                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1414 {
1415         struct ovs_header *ovs_header;
1416         struct ovs_dp_stats dp_stats;
1417         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1418         int err;
1419
1420         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1421                                    flags, cmd);
1422         if (!ovs_header)
1423                 goto error;
1424
1425         ovs_header->dp_ifindex = get_dpifindex(dp);
1426
1427         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1428         if (err)
1429                 goto nla_put_failure;
1430
1431         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1432         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1433                         &dp_stats))
1434                 goto nla_put_failure;
1435
1436         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1437                         sizeof(struct ovs_dp_megaflow_stats),
1438                         &dp_megaflow_stats))
1439                 goto nla_put_failure;
1440
1441         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1442                 goto nla_put_failure;
1443
1444         genlmsg_end(skb, ovs_header);
1445         return 0;
1446
1447 nla_put_failure:
1448         genlmsg_cancel(skb, ovs_header);
1449 error:
1450         return -EMSGSIZE;
1451 }
1452
1453 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1454 {
1455         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1456 }
1457
1458 /* Called with rcu_read_lock or ovs_mutex. */
1459 static struct datapath *lookup_datapath(struct net *net,
1460                                         const struct ovs_header *ovs_header,
1461                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1462 {
1463         struct datapath *dp;
1464
1465         if (!a[OVS_DP_ATTR_NAME])
1466                 dp = get_dp(net, ovs_header->dp_ifindex);
1467         else {
1468                 struct vport *vport;
1469
1470                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1471                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1472         }
1473         return dp ? dp : ERR_PTR(-ENODEV);
1474 }
1475
1476 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1477 {
1478         struct datapath *dp;
1479
1480         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1481         if (IS_ERR(dp))
1482                 return;
1483
1484         WARN(dp->user_features, "Dropping previously announced user features\n");
1485         dp->user_features = 0;
1486 }
1487
1488 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1489 {
1490         if (a[OVS_DP_ATTR_USER_FEATURES])
1491                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1492 }
1493
1494 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1495 {
1496         struct nlattr **a = info->attrs;
1497         struct vport_parms parms;
1498         struct sk_buff *reply;
1499         struct datapath *dp;
1500         struct vport *vport;
1501         struct ovs_net *ovs_net;
1502         int err, i;
1503
1504         err = -EINVAL;
1505         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1506                 goto err;
1507
1508         reply = ovs_dp_cmd_alloc_info(info);
1509         if (!reply)
1510                 return -ENOMEM;
1511
1512         err = -ENOMEM;
1513         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1514         if (dp == NULL)
1515                 goto err_free_reply;
1516
1517         ovs_dp_set_net(dp, sock_net(skb->sk));
1518
1519         /* Allocate table. */
1520         err = ovs_flow_tbl_init(&dp->table);
1521         if (err)
1522                 goto err_free_dp;
1523
1524         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1525         if (!dp->stats_percpu) {
1526                 err = -ENOMEM;
1527                 goto err_destroy_table;
1528         }
1529
1530         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1531                             GFP_KERNEL);
1532         if (!dp->ports) {
1533                 err = -ENOMEM;
1534                 goto err_destroy_percpu;
1535         }
1536
1537         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1538                 INIT_HLIST_HEAD(&dp->ports[i]);
1539
1540         /* Set up our datapath device. */
1541         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1542         parms.type = OVS_VPORT_TYPE_INTERNAL;
1543         parms.options = NULL;
1544         parms.dp = dp;
1545         parms.port_no = OVSP_LOCAL;
1546         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1547
1548         ovs_dp_change(dp, a);
1549
1550         /* So far only local changes have been made, now need the lock. */
1551         ovs_lock();
1552
1553         vport = new_vport(&parms);
1554         if (IS_ERR(vport)) {
1555                 err = PTR_ERR(vport);
1556                 if (err == -EBUSY)
1557                         err = -EEXIST;
1558
1559                 if (err == -EEXIST) {
1560                         /* An outdated user space instance that does not understand
1561                          * the concept of user_features has attempted to create a new
1562                          * datapath and is likely to reuse it. Drop all user features.
1563                          */
1564                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1565                                 ovs_dp_reset_user_features(skb, info);
1566                 }
1567
1568                 goto err_destroy_ports_array;
1569         }
1570
1571         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1572                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1573         BUG_ON(err < 0);
1574
1575         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1576         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1577
1578         ovs_unlock();
1579
1580         ovs_notify(&dp_datapath_genl_family, reply, info);
1581         return 0;
1582
1583 err_destroy_ports_array:
1584         ovs_unlock();
1585         kfree(dp->ports);
1586 err_destroy_percpu:
1587         free_percpu(dp->stats_percpu);
1588 err_destroy_table:
1589         ovs_flow_tbl_destroy(&dp->table);
1590 err_free_dp:
1591         kfree(dp);
1592 err_free_reply:
1593         kfree_skb(reply);
1594 err:
1595         return err;
1596 }
1597
1598 /* Called with ovs_mutex. */
1599 static void __dp_destroy(struct datapath *dp)
1600 {
1601         int i;
1602
1603         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1604                 struct vport *vport;
1605                 struct hlist_node *n;
1606
1607                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1608                         if (vport->port_no != OVSP_LOCAL)
1609                                 ovs_dp_detach_port(vport);
1610         }
1611
1612         list_del_rcu(&dp->list_node);
1613
1614         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1615          * all ports in datapath are destroyed first before freeing datapath.
1616          */
1617         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1618
1619         /* RCU destroy the flow table */
1620         call_rcu(&dp->rcu, destroy_dp_rcu);
1621 }
1622
1623 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1624 {
1625         struct sk_buff *reply;
1626         struct datapath *dp;
1627         int err;
1628
1629         reply = ovs_dp_cmd_alloc_info(info);
1630         if (!reply)
1631                 return -ENOMEM;
1632
1633         ovs_lock();
1634         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1635         err = PTR_ERR(dp);
1636         if (IS_ERR(dp))
1637                 goto err_unlock_free;
1638
1639         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1640                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1641         BUG_ON(err < 0);
1642
1643         __dp_destroy(dp);
1644         ovs_unlock();
1645
1646         ovs_notify(&dp_datapath_genl_family, reply, info);
1647
1648         return 0;
1649
1650 err_unlock_free:
1651         ovs_unlock();
1652         kfree_skb(reply);
1653         return err;
1654 }
1655
1656 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1657 {
1658         struct sk_buff *reply;
1659         struct datapath *dp;
1660         int err;
1661
1662         reply = ovs_dp_cmd_alloc_info(info);
1663         if (!reply)
1664                 return -ENOMEM;
1665
1666         ovs_lock();
1667         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1668         err = PTR_ERR(dp);
1669         if (IS_ERR(dp))
1670                 goto err_unlock_free;
1671
1672         ovs_dp_change(dp, info->attrs);
1673
1674         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1675                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1676         BUG_ON(err < 0);
1677
1678         ovs_unlock();
1679         ovs_notify(&dp_datapath_genl_family, reply, info);
1680
1681         return 0;
1682
1683 err_unlock_free:
1684         ovs_unlock();
1685         kfree_skb(reply);
1686         return err;
1687 }
1688
1689 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1690 {
1691         struct sk_buff *reply;
1692         struct datapath *dp;
1693         int err;
1694
1695         reply = ovs_dp_cmd_alloc_info(info);
1696         if (!reply)
1697                 return -ENOMEM;
1698
1699         ovs_lock();
1700         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1701         if (IS_ERR(dp)) {
1702                 err = PTR_ERR(dp);
1703                 goto err_unlock_free;
1704         }
1705         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1706                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1707         BUG_ON(err < 0);
1708         ovs_unlock();
1709
1710         return genlmsg_reply(reply, info);
1711
1712 err_unlock_free:
1713         ovs_unlock();
1714         kfree_skb(reply);
1715         return err;
1716 }
1717
1718 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1719 {
1720         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1721         struct datapath *dp;
1722         int skip = cb->args[0];
1723         int i = 0;
1724
1725         ovs_lock();
1726         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1727                 if (i >= skip &&
1728                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1729                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1730                                          OVS_DP_CMD_NEW) < 0)
1731                         break;
1732                 i++;
1733         }
1734         ovs_unlock();
1735
1736         cb->args[0] = i;
1737
1738         return skb->len;
1739 }
1740
1741 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1742         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1743         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1744         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1745 };
1746
1747 static const struct genl_ops dp_datapath_genl_ops[] = {
1748         { .cmd = OVS_DP_CMD_NEW,
1749           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1750           .policy = datapath_policy,
1751           .doit = ovs_dp_cmd_new
1752         },
1753         { .cmd = OVS_DP_CMD_DEL,
1754           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1755           .policy = datapath_policy,
1756           .doit = ovs_dp_cmd_del
1757         },
1758         { .cmd = OVS_DP_CMD_GET,
1759           .flags = 0,               /* OK for unprivileged users. */
1760           .policy = datapath_policy,
1761           .doit = ovs_dp_cmd_get,
1762           .dumpit = ovs_dp_cmd_dump
1763         },
1764         { .cmd = OVS_DP_CMD_SET,
1765           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1766           .policy = datapath_policy,
1767           .doit = ovs_dp_cmd_set,
1768         },
1769 };
1770
1771 static struct genl_family dp_datapath_genl_family = {
1772         .id = GENL_ID_GENERATE,
1773         .hdrsize = sizeof(struct ovs_header),
1774         .name = OVS_DATAPATH_FAMILY,
1775         .version = OVS_DATAPATH_VERSION,
1776         .maxattr = OVS_DP_ATTR_MAX,
1777         .netnsok = true,
1778         .parallel_ops = true,
1779         .ops = dp_datapath_genl_ops,
1780         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1781         .mcgrps = &ovs_dp_datapath_multicast_group,
1782         .n_mcgrps = 1,
1783 };
1784
1785 /* Called with ovs_mutex or RCU read lock. */
1786 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1787                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1788 {
1789         struct ovs_header *ovs_header;
1790         struct ovs_vport_stats vport_stats;
1791         int err;
1792
1793         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1794                                  flags, cmd);
1795         if (!ovs_header)
1796                 return -EMSGSIZE;
1797
1798         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1799
1800         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1801             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1802             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1803                            ovs_vport_name(vport)))
1804                 goto nla_put_failure;
1805
1806         ovs_vport_get_stats(vport, &vport_stats);
1807         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1808                     &vport_stats))
1809                 goto nla_put_failure;
1810
1811         if (ovs_vport_get_upcall_portids(vport, skb))
1812                 goto nla_put_failure;
1813
1814         err = ovs_vport_get_options(vport, skb);
1815         if (err == -EMSGSIZE)
1816                 goto error;
1817
1818         genlmsg_end(skb, ovs_header);
1819         return 0;
1820
1821 nla_put_failure:
1822         err = -EMSGSIZE;
1823 error:
1824         genlmsg_cancel(skb, ovs_header);
1825         return err;
1826 }
1827
1828 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1829 {
1830         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1831 }
1832
1833 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1834 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1835                                          u32 seq, u8 cmd)
1836 {
1837         struct sk_buff *skb;
1838         int retval;
1839
1840         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1841         if (!skb)
1842                 return ERR_PTR(-ENOMEM);
1843
1844         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1845         BUG_ON(retval < 0);
1846
1847         return skb;
1848 }
1849
1850 /* Called with ovs_mutex or RCU read lock. */
1851 static struct vport *lookup_vport(struct net *net,
1852                                   const struct ovs_header *ovs_header,
1853                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1854 {
1855         struct datapath *dp;
1856         struct vport *vport;
1857
1858         if (a[OVS_VPORT_ATTR_NAME]) {
1859                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1860                 if (!vport)
1861                         return ERR_PTR(-ENODEV);
1862                 if (ovs_header->dp_ifindex &&
1863                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1864                         return ERR_PTR(-ENODEV);
1865                 return vport;
1866         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1867                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1868
1869                 if (port_no >= DP_MAX_PORTS)
1870                         return ERR_PTR(-EFBIG);
1871
1872                 dp = get_dp(net, ovs_header->dp_ifindex);
1873                 if (!dp)
1874                         return ERR_PTR(-ENODEV);
1875
1876                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1877                 if (!vport)
1878                         return ERR_PTR(-ENODEV);
1879                 return vport;
1880         } else
1881                 return ERR_PTR(-EINVAL);
1882 }
1883
1884 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1885 {
1886         struct nlattr **a = info->attrs;
1887         struct ovs_header *ovs_header = info->userhdr;
1888         struct vport_parms parms;
1889         struct sk_buff *reply;
1890         struct vport *vport;
1891         struct datapath *dp;
1892         u32 port_no;
1893         int err;
1894
1895         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1896             !a[OVS_VPORT_ATTR_UPCALL_PID])
1897                 return -EINVAL;
1898
1899         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1900                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1901         if (port_no >= DP_MAX_PORTS)
1902                 return -EFBIG;
1903
1904         reply = ovs_vport_cmd_alloc_info();
1905         if (!reply)
1906                 return -ENOMEM;
1907
1908         ovs_lock();
1909 restart:
1910         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1911         err = -ENODEV;
1912         if (!dp)
1913                 goto exit_unlock_free;
1914
1915         if (port_no) {
1916                 vport = ovs_vport_ovsl(dp, port_no);
1917                 err = -EBUSY;
1918                 if (vport)
1919                         goto exit_unlock_free;
1920         } else {
1921                 for (port_no = 1; ; port_no++) {
1922                         if (port_no >= DP_MAX_PORTS) {
1923                                 err = -EFBIG;
1924                                 goto exit_unlock_free;
1925                         }
1926                         vport = ovs_vport_ovsl(dp, port_no);
1927                         if (!vport)
1928                                 break;
1929                 }
1930         }
1931
1932         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1933         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1934         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1935         parms.dp = dp;
1936         parms.port_no = port_no;
1937         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1938
1939         vport = new_vport(&parms);
1940         err = PTR_ERR(vport);
1941         if (IS_ERR(vport)) {
1942                 if (err == -EAGAIN)
1943                         goto restart;
1944                 goto exit_unlock_free;
1945         }
1946
1947         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1948                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1949         BUG_ON(err < 0);
1950         ovs_unlock();
1951
1952         ovs_notify(&dp_vport_genl_family, reply, info);
1953         return 0;
1954
1955 exit_unlock_free:
1956         ovs_unlock();
1957         kfree_skb(reply);
1958         return err;
1959 }
1960
1961 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1962 {
1963         struct nlattr **a = info->attrs;
1964         struct sk_buff *reply;
1965         struct vport *vport;
1966         int err;
1967
1968         reply = ovs_vport_cmd_alloc_info();
1969         if (!reply)
1970                 return -ENOMEM;
1971
1972         ovs_lock();
1973         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1974         err = PTR_ERR(vport);
1975         if (IS_ERR(vport))
1976                 goto exit_unlock_free;
1977
1978         if (a[OVS_VPORT_ATTR_TYPE] &&
1979             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1980                 err = -EINVAL;
1981                 goto exit_unlock_free;
1982         }
1983
1984         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1985                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1986                 if (err)
1987                         goto exit_unlock_free;
1988         }
1989
1990
1991         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1992                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
1993
1994                 err = ovs_vport_set_upcall_portids(vport, ids);
1995                 if (err)
1996                         goto exit_unlock_free;
1997         }
1998
1999         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2000                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2001         BUG_ON(err < 0);
2002
2003         ovs_unlock();
2004         ovs_notify(&dp_vport_genl_family, reply, info);
2005         return 0;
2006
2007 exit_unlock_free:
2008         ovs_unlock();
2009         kfree_skb(reply);
2010         return err;
2011 }
2012
2013 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2014 {
2015         struct nlattr **a = info->attrs;
2016         struct sk_buff *reply;
2017         struct vport *vport;
2018         int err;
2019
2020         reply = ovs_vport_cmd_alloc_info();
2021         if (!reply)
2022                 return -ENOMEM;
2023
2024         ovs_lock();
2025         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2026         err = PTR_ERR(vport);
2027         if (IS_ERR(vport))
2028                 goto exit_unlock_free;
2029
2030         if (vport->port_no == OVSP_LOCAL) {
2031                 err = -EINVAL;
2032                 goto exit_unlock_free;
2033         }
2034
2035         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2036                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2037         BUG_ON(err < 0);
2038         ovs_dp_detach_port(vport);
2039         ovs_unlock();
2040
2041         ovs_notify(&dp_vport_genl_family, reply, info);
2042         return 0;
2043
2044 exit_unlock_free:
2045         ovs_unlock();
2046         kfree_skb(reply);
2047         return err;
2048 }
2049
2050 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2051 {
2052         struct nlattr **a = info->attrs;
2053         struct ovs_header *ovs_header = info->userhdr;
2054         struct sk_buff *reply;
2055         struct vport *vport;
2056         int err;
2057
2058         reply = ovs_vport_cmd_alloc_info();
2059         if (!reply)
2060                 return -ENOMEM;
2061
2062         rcu_read_lock();
2063         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2064         err = PTR_ERR(vport);
2065         if (IS_ERR(vport))
2066                 goto exit_unlock_free;
2067         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2068                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2069         BUG_ON(err < 0);
2070         rcu_read_unlock();
2071
2072         return genlmsg_reply(reply, info);
2073
2074 exit_unlock_free:
2075         rcu_read_unlock();
2076         kfree_skb(reply);
2077         return err;
2078 }
2079
2080 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2081 {
2082         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2083         struct datapath *dp;
2084         int bucket = cb->args[0], skip = cb->args[1];
2085         int i, j = 0;
2086
2087         rcu_read_lock();
2088         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2089         if (!dp) {
2090                 rcu_read_unlock();
2091                 return -ENODEV;
2092         }
2093         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2094                 struct vport *vport;
2095
2096                 j = 0;
2097                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2098                         if (j >= skip &&
2099                             ovs_vport_cmd_fill_info(vport, skb,
2100                                                     NETLINK_CB(cb->skb).portid,
2101                                                     cb->nlh->nlmsg_seq,
2102                                                     NLM_F_MULTI,
2103                                                     OVS_VPORT_CMD_NEW) < 0)
2104                                 goto out;
2105
2106                         j++;
2107                 }
2108                 skip = 0;
2109         }
2110 out:
2111         rcu_read_unlock();
2112
2113         cb->args[0] = i;
2114         cb->args[1] = j;
2115
2116         return skb->len;
2117 }
2118
2119 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2120         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2121         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2122         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2123         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2124         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2125         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2126 };
2127
2128 static const struct genl_ops dp_vport_genl_ops[] = {
2129         { .cmd = OVS_VPORT_CMD_NEW,
2130           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2131           .policy = vport_policy,
2132           .doit = ovs_vport_cmd_new
2133         },
2134         { .cmd = OVS_VPORT_CMD_DEL,
2135           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2136           .policy = vport_policy,
2137           .doit = ovs_vport_cmd_del
2138         },
2139         { .cmd = OVS_VPORT_CMD_GET,
2140           .flags = 0,               /* OK for unprivileged users. */
2141           .policy = vport_policy,
2142           .doit = ovs_vport_cmd_get,
2143           .dumpit = ovs_vport_cmd_dump
2144         },
2145         { .cmd = OVS_VPORT_CMD_SET,
2146           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2147           .policy = vport_policy,
2148           .doit = ovs_vport_cmd_set,
2149         },
2150 };
2151
2152 struct genl_family dp_vport_genl_family = {
2153         .id = GENL_ID_GENERATE,
2154         .hdrsize = sizeof(struct ovs_header),
2155         .name = OVS_VPORT_FAMILY,
2156         .version = OVS_VPORT_VERSION,
2157         .maxattr = OVS_VPORT_ATTR_MAX,
2158         .netnsok = true,
2159         .parallel_ops = true,
2160         .ops = dp_vport_genl_ops,
2161         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2162         .mcgrps = &ovs_dp_vport_multicast_group,
2163         .n_mcgrps = 1,
2164 };
2165
2166 static struct genl_family * const dp_genl_families[] = {
2167         &dp_datapath_genl_family,
2168         &dp_vport_genl_family,
2169         &dp_flow_genl_family,
2170         &dp_packet_genl_family,
2171 };
2172
2173 static void dp_unregister_genl(int n_families)
2174 {
2175         int i;
2176
2177         for (i = 0; i < n_families; i++)
2178                 genl_unregister_family(dp_genl_families[i]);
2179 }
2180
2181 static int dp_register_genl(void)
2182 {
2183         int err;
2184         int i;
2185
2186         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2187
2188                 err = genl_register_family(dp_genl_families[i]);
2189                 if (err)
2190                         goto error;
2191         }
2192
2193         return 0;
2194
2195 error:
2196         dp_unregister_genl(i);
2197         return err;
2198 }
2199
2200 static int __net_init ovs_init_net(struct net *net)
2201 {
2202         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2203
2204         INIT_LIST_HEAD(&ovs_net->dps);
2205         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2206         return 0;
2207 }
2208
2209 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2210                                             struct list_head *head)
2211 {
2212         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2213         struct datapath *dp;
2214
2215         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2216                 int i;
2217
2218                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2219                         struct vport *vport;
2220
2221                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2222                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2223                                         continue;
2224
2225                                 if (dev_net(vport->dev) == dnet)
2226                                         list_add(&vport->detach_list, head);
2227                         }
2228                 }
2229         }
2230 }
2231
2232 static void __net_exit ovs_exit_net(struct net *dnet)
2233 {
2234         struct datapath *dp, *dp_next;
2235         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2236         struct vport *vport, *vport_next;
2237         struct net *net;
2238         LIST_HEAD(head);
2239
2240         ovs_lock();
2241         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2242                 __dp_destroy(dp);
2243
2244         rtnl_lock();
2245         for_each_net(net)
2246                 list_vports_from_net(net, dnet, &head);
2247         rtnl_unlock();
2248
2249         /* Detach all vports from given namespace. */
2250         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2251                 list_del(&vport->detach_list);
2252                 ovs_dp_detach_port(vport);
2253         }
2254
2255         ovs_unlock();
2256
2257         cancel_work_sync(&ovs_net->dp_notify_work);
2258 }
2259
2260 static struct pernet_operations ovs_net_ops = {
2261         .init = ovs_init_net,
2262         .exit = ovs_exit_net,
2263         .id   = &ovs_net_id,
2264         .size = sizeof(struct ovs_net),
2265 };
2266
2267 static int __init dp_init(void)
2268 {
2269         int err;
2270
2271         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2272
2273         pr_info("Open vSwitch switching datapath\n");
2274
2275         err = action_fifos_init();
2276         if (err)
2277                 goto error;
2278
2279         err = ovs_internal_dev_rtnl_link_register();
2280         if (err)
2281                 goto error_action_fifos_exit;
2282
2283         err = ovs_flow_init();
2284         if (err)
2285                 goto error_unreg_rtnl_link;
2286
2287         err = ovs_vport_init();
2288         if (err)
2289                 goto error_flow_exit;
2290
2291         err = register_pernet_device(&ovs_net_ops);
2292         if (err)
2293                 goto error_vport_exit;
2294
2295         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2296         if (err)
2297                 goto error_netns_exit;
2298
2299         err = ovs_netdev_init();
2300         if (err)
2301                 goto error_unreg_notifier;
2302
2303         err = dp_register_genl();
2304         if (err < 0)
2305                 goto error_unreg_netdev;
2306
2307         return 0;
2308
2309 error_unreg_netdev:
2310         ovs_netdev_exit();
2311 error_unreg_notifier:
2312         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2313 error_netns_exit:
2314         unregister_pernet_device(&ovs_net_ops);
2315 error_vport_exit:
2316         ovs_vport_exit();
2317 error_flow_exit:
2318         ovs_flow_exit();
2319 error_unreg_rtnl_link:
2320         ovs_internal_dev_rtnl_link_unregister();
2321 error_action_fifos_exit:
2322         action_fifos_exit();
2323 error:
2324         return err;
2325 }
2326
2327 static void dp_cleanup(void)
2328 {
2329         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2330         ovs_netdev_exit();
2331         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2332         unregister_pernet_device(&ovs_net_ops);
2333         rcu_barrier();
2334         ovs_vport_exit();
2335         ovs_flow_exit();
2336         ovs_internal_dev_rtnl_link_unregister();
2337         action_fifos_exit();
2338 }
2339
2340 module_init(dp_init);
2341 module_exit(dp_cleanup);
2342
2343 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2344 MODULE_LICENSE("GPL");