datapath/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/version.h>
  40 #include <linux/ethtool.h>
  41 #include <linux/wait.h>
  42 #include <asm/div64.h>
  43 #include <linux/highmem.h>
  44 #include <linux/netfilter_bridge.h>
  45 #include <linux/netfilter_ipv4.h>
  46 #include <linux/inetdevice.h>
  47 #include <linux/list.h>
  48 #include <linux/openvswitch.h>
  49 #include <linux/rculist.h>
  50 #include <linux/dmi.h>
  51 #include <net/genetlink.h>
  52 #include <net/net_namespace.h>
  53 #include <net/netns/generic.h>
  54
  55 #include "datapath.h"
  56 #include "flow.h"
  57 #include "flow_table.h"
  58 #include "flow_netlink.h"
  59 #include "vlan.h"
  60 #include "vport-internal_dev.h"
  61 #include "vport-netdev.h"
  62
  63 int ovs_net_id __read_mostly;
  64
  65 static struct genl_family dp_packet_genl_family;
  66 static struct genl_family dp_flow_genl_family;
  67 static struct genl_family dp_datapath_genl_family;
  68
  69 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
  70         .name = OVS_FLOW_MCGROUP
  71 };
  72
  73 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  74         .name = OVS_DATAPATH_MCGROUP
  75 };
  76
  77 struct genl_multicast_group ovs_dp_vport_multicast_group = {
  78         .name = OVS_VPORT_MCGROUP
  79 };
  80
  81 /* Check if need to build a reply message.
  82  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply.
  83  */
  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, struct genl_multicast_group *grp,
  92                        struct sk_buff *skb, struct genl_info *info)
  93 {
  94         genl_notify(family, skb, genl_info_net(info),
  95                     info->snd_portid, GROUP_ID(grp), 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 #endif
 136
 137 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 138                              const struct sw_flow_key *,
 139                              const struct dp_upcall_info *);
 140 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 141                                   const struct sw_flow_key *,
 142                                   const struct dp_upcall_info *);
 143
 144 /* Must be called with rcu_read_lock. */
 145 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
 146 {
 147         struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
 148
 149         if (dev) {
 150                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 151                 if (vport)
 152                         return vport->dp;
 153         }
 154
 155         return NULL;
 156 }
 157
 158 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
 159  * returned dp pointer valid.
 160  */
 161 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
 162 {
 163         struct datapath *dp;
 164
 165         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
 166         rcu_read_lock();
 167         dp = get_dp_rcu(net, dp_ifindex);
 168         rcu_read_unlock();
 169
 170         return dp;
 171 }
 172
 173 /* Must be called with rcu_read_lock or ovs_mutex. */
 174 const char *ovs_dp_name(const struct datapath *dp)
 175 {
 176         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 177         return vport->ops->get_name(vport);
 178 }
 179
 180 static int get_dpifindex(const struct datapath *dp)
 181 {
 182         struct vport *local;
 183         int ifindex;
 184
 185         rcu_read_lock();
 186
 187         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 188         if (local)
 189                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 190         else
 191                 ifindex = 0;
 192
 193         rcu_read_unlock();
 194
 195         return ifindex;
 196 }
 197
 198 static void destroy_dp_rcu(struct rcu_head *rcu)
 199 {
 200         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 201
 202         ovs_flow_tbl_destroy(&dp->table);
 203         free_percpu(dp->stats_percpu);
 204         release_net(ovs_dp_get_net(dp));
 205         kfree(dp->ports);
 206         kfree(dp);
 207 }
 208
 209 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 210                                             u16 port_no)
 211 {
 212         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 213 }
 214
 215 /* Called with ovs_mutex or RCU read lock. */
 216 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 217 {
 218         struct vport *vport;
 219         struct hlist_head *head;
 220
 221         head = vport_hash_bucket(dp, port_no);
 222         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 223                 if (vport->port_no == port_no)
 224                         return vport;
 225         }
 226         return NULL;
 227 }
 228
 229 /* Called with ovs_mutex. */
 230 static struct vport *new_vport(const struct vport_parms *parms)
 231 {
 232         struct vport *vport;
 233
 234         vport = ovs_vport_add(parms);
 235         if (!IS_ERR(vport)) {
 236                 struct datapath *dp = parms->dp;
 237                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 238
 239                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 240         }
 241         return vport;
 242 }
 243
 244 void ovs_dp_detach_port(struct vport *p)
 245 {
 246         ASSERT_OVSL();
 247
 248         /* First drop references to device. */
 249         hlist_del_rcu(&p->dp_hash_node);
 250
 251         /* Then destroy it. */
 252         ovs_vport_del(p);
 253 }
 254
 255 /* Must be called with rcu_read_lock. */
 256 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 257 {
 258         const struct vport *p = OVS_CB(skb)->input_vport;
 259         struct datapath *dp = p->dp;
 260         struct sw_flow *flow;
 261         struct sw_flow_actions *sf_acts;
 262         struct dp_stats_percpu *stats;
 263         u64 *stats_counter;
 264         u32 n_mask_hit;
 265
 266         stats = this_cpu_ptr(dp->stats_percpu);
 267
 268         /* Look up flow. */
 269         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
 270                                          &n_mask_hit);
 271         if (unlikely(!flow)) {
 272                 struct dp_upcall_info upcall;
 273                 int error;
 274
 275                 upcall.cmd = OVS_PACKET_CMD_MISS;
 276                 upcall.userdata = NULL;
 277                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 278                 upcall.egress_tun_info = NULL;
 279                 error = ovs_dp_upcall(dp, skb, key, &upcall);
 280                 if (unlikely(error))
 281                         kfree_skb(skb);
 282                 else
 283                         consume_skb(skb);
 284                 stats_counter = &stats->n_missed;
 285                 goto out;
 286         }
 287
 288         ovs_flow_stats_update(flow, key->tp.flags, skb);
 289         sf_acts = rcu_dereference(flow->sf_acts);
 290         ovs_execute_actions(dp, skb, sf_acts, key);
 291
 292         stats_counter = &stats->n_hit;
 293
 294 out:
 295         /* Update datapath statistics. */
 296         u64_stats_update_begin(&stats->syncp);
 297         (*stats_counter)++;
 298         stats->n_mask_hit += n_mask_hit;
 299         u64_stats_update_end(&stats->syncp);
 300 }
 301
 302 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 303                   const struct sw_flow_key *key,
 304                   const struct dp_upcall_info *upcall_info)
 305 {
 306         struct dp_stats_percpu *stats;
 307         int err;
 308
 309         if (upcall_info->portid == 0) {
 310                 err = -ENOTCONN;
 311                 goto err;
 312         }
 313
 314         if (!skb_is_gso(skb))
 315                 err = queue_userspace_packet(dp, skb, key, upcall_info);
 316         else
 317                 err = queue_gso_packets(dp, skb, key, upcall_info);
 318         if (err)
 319                 goto err;
 320
 321         return 0;
 322
 323 err:
 324         stats = this_cpu_ptr(dp->stats_percpu);
 325
 326         u64_stats_update_begin(&stats->syncp);
 327         stats->n_lost++;
 328         u64_stats_update_end(&stats->syncp);
 329
 330         return err;
 331 }
 332
 333 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 334                              const struct sw_flow_key *key,
 335                              const struct dp_upcall_info *upcall_info)
 336 {
 337         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 338         struct sw_flow_key later_key;
 339         struct sk_buff *segs, *nskb;
 340         struct ovs_skb_cb ovs_cb;
 341         int err;
 342
 343         ovs_cb = *OVS_CB(skb);
 344         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 345         *OVS_CB(skb) = ovs_cb;
 346         if (IS_ERR(segs))
 347                 return PTR_ERR(segs);
 348         if (segs == NULL)
 349                 return -EINVAL;
 350
 351         if (gso_type & SKB_GSO_UDP) {
 352                 /* The initial flow key extracted by ovs_flow_key_extract()
 353                  * in this case is for a first fragment, so we need to
 354                  * properly mark later fragments.
 355                  */
 356                 later_key = *key;
 357                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 358         }
 359
 360         /* Queue all of the segments. */
 361         skb = segs;
 362         do {
 363                 *OVS_CB(skb) = ovs_cb;
 364                 if (gso_type & SKB_GSO_UDP && skb != segs)
 365                         key = &later_key;
 366
 367                 err = queue_userspace_packet(dp, skb, key, upcall_info);
 368                 if (err)
 369                         break;
 370
 371         } while ((skb = skb->next));
 372
 373         /* Free all of the segments. */
 374         skb = segs;
 375         do {
 376                 nskb = skb->next;
 377                 if (err)
 378                         kfree_skb(skb);
 379                 else
 380                         consume_skb(skb);
 381         } while ((skb = nskb));
 382         return err;
 383 }
 384
 385 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 386                               unsigned int hdrlen)
 387 {
 388         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 389                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 390                 + nla_total_size(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 391
 392         /* OVS_PACKET_ATTR_USERDATA */
 393         if (upcall_info->userdata)
 394                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 395
 396         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 397         if (upcall_info->egress_tun_info)
 398                 size += nla_total_size(ovs_tun_key_attr_size());
 399
 400         return size;
 401 }
 402
 403 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 404                                   const struct sw_flow_key *key,
 405                                   const struct dp_upcall_info *upcall_info)
 406 {
 407         struct ovs_header *upcall;
 408         struct sk_buff *nskb = NULL;
 409         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 410         struct nlattr *nla;
 411         struct genl_info info = {
 412 #ifdef HAVE_GENLMSG_NEW_UNICAST
 413                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 414 #endif
 415                 .snd_portid = upcall_info->portid,
 416         };
 417         size_t len;
 418         unsigned int hlen;
 419         int err, dp_ifindex;
 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_insert_tag_set_proto(nskb, nskb->vlan_proto, skb_vlan_tag_get(nskb));
 431                 if (!nskb)
 432                         return -ENOMEM;
 433
 434                 vlan_set_tci(nskb, 0);
 435
 436                 skb = nskb;
 437         }
 438
 439         if (nla_attr_size(skb->len) > USHRT_MAX) {
 440                 err = -EFBIG;
 441                 goto out;
 442         }
 443
 444         /* Complete checksum if needed */
 445         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 446             (err = skb_checksum_help(skb)))
 447                 goto out;
 448
 449         /* Older versions of OVS user space enforce alignment of the last
 450          * Netlink attribute to NLA_ALIGNTO which would require extensive
 451          * padding logic. Only perform zerocopy if padding is not required.
 452          */
 453         if (dp->user_features & OVS_DP_F_UNALIGNED)
 454                 hlen = skb_zerocopy_headlen(skb);
 455         else
 456                 hlen = skb->len;
 457
 458         len = upcall_msg_size(upcall_info, hlen);
 459         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 460         if (!user_skb) {
 461                 err = -ENOMEM;
 462                 goto out;
 463         }
 464
 465         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 466                              0, upcall_info->cmd);
 467         upcall->dp_ifindex = dp_ifindex;
 468
 469         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 470         err = ovs_nla_put_flow(key, key, user_skb);
 471         BUG_ON(err);
 472         nla_nest_end(user_skb, nla);
 473
 474         if (upcall_info->userdata)
 475                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 476                           nla_len(upcall_info->userdata),
 477                           nla_data(upcall_info->userdata));
 478
 479         if (upcall_info->egress_tun_info) {
 480                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 481                 err = ovs_nla_put_egress_tunnel_key(user_skb,
 482                                                     upcall_info->egress_tun_info);
 483                 BUG_ON(err);
 484                 nla_nest_end(user_skb, nla);
 485         }
 486
 487         /* Only reserve room for attribute header, packet data is added
 488          * in skb_zerocopy()
 489          */
 490         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 491                 err = -ENOBUFS;
 492                 goto out;
 493         }
 494         nla->nla_len = nla_attr_size(skb->len);
 495
 496         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 497         if (err)
 498                 goto out;
 499
 500         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 501         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 502                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 503
 504                 if (plen > 0)
 505                         memset(skb_put(user_skb, plen), 0, plen);
 506         }
 507
 508         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 509
 510         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 511         user_skb = NULL;
 512 out:
 513         if (err)
 514                 skb_tx_error(skb);
 515         kfree_skb(user_skb);
 516         kfree_skb(nskb);
 517         return err;
 518 }
 519
 520 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 521 {
 522         struct ovs_header *ovs_header = info->userhdr;
 523         struct nlattr **a = info->attrs;
 524         struct sw_flow_actions *acts;
 525         struct sk_buff *packet;
 526         struct sw_flow *flow;
 527         struct sw_flow_actions *sf_acts;
 528         struct datapath *dp;
 529         struct ethhdr *eth;
 530         struct vport *input_vport;
 531         int len;
 532         int err;
 533         bool log = !a[OVS_PACKET_ATTR_PROBE];
 534
 535         err = -EINVAL;
 536         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 537             !a[OVS_PACKET_ATTR_ACTIONS])
 538                 goto err;
 539
 540         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 541         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 542         err = -ENOMEM;
 543         if (!packet)
 544                 goto err;
 545         skb_reserve(packet, NET_IP_ALIGN);
 546
 547         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 548
 549         skb_reset_mac_header(packet);
 550         eth = eth_hdr(packet);
 551
 552         /* Normally, setting the skb 'protocol' field would be handled by a
 553          * call to eth_type_trans(), but it assumes there's a sending
 554          * device, which we may not have.
 555          */
 556         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 557                 packet->protocol = eth->h_proto;
 558         else
 559                 packet->protocol = htons(ETH_P_802_2);
 560
 561         /* Build an sw_flow for sending this packet. */
 562         flow = ovs_flow_alloc();
 563         err = PTR_ERR(flow);
 564         if (IS_ERR(flow))
 565                 goto err_kfree_skb;
 566
 567         err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
 568                                              &flow->key, log);
 569         if (err)
 570                 goto err_flow_free;
 571
 572         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 573                                    &flow->key, &acts, log);
 574         if (err)
 575                 goto err_flow_free;
 576
 577         rcu_assign_pointer(flow->sf_acts, acts);
 578         OVS_CB(packet)->egress_tun_info = NULL;
 579         packet->priority = flow->key.phy.priority;
 580         packet->mark = flow->key.phy.skb_mark;
 581
 582         rcu_read_lock();
 583         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
 584         err = -ENODEV;
 585         if (!dp)
 586                 goto err_unlock;
 587
 588         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 589         if (!input_vport)
 590                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 591
 592         if (!input_vport)
 593                 goto err_unlock;
 594
 595         OVS_CB(packet)->input_vport = input_vport;
 596         sf_acts = rcu_dereference(flow->sf_acts);
 597
 598         local_bh_disable();
 599         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 600         local_bh_enable();
 601         rcu_read_unlock();
 602
 603         ovs_flow_free(flow, false);
 604         return err;
 605
 606 err_unlock:
 607         rcu_read_unlock();
 608 err_flow_free:
 609         ovs_flow_free(flow, false);
 610 err_kfree_skb:
 611         kfree_skb(packet);
 612 err:
 613         return err;
 614 }
 615
 616 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 617         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 618         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 619         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 620         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 621 };
 622
 623 static struct genl_ops dp_packet_genl_ops[] = {
 624         { .cmd = OVS_PACKET_CMD_EXECUTE,
 625           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 626           .policy = packet_policy,
 627           .doit = ovs_packet_cmd_execute
 628         }
 629 };
 630
 631 static struct genl_family dp_packet_genl_family = {
 632         .id = GENL_ID_GENERATE,
 633         .hdrsize = sizeof(struct ovs_header),
 634         .name = OVS_PACKET_FAMILY,
 635         .version = OVS_PACKET_VERSION,
 636         .maxattr = OVS_PACKET_ATTR_MAX,
 637         .netnsok = true,
 638         .parallel_ops = true,
 639         .ops = dp_packet_genl_ops,
 640         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 641 };
 642
 643 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 644                          struct ovs_dp_megaflow_stats *mega_stats)
 645 {
 646         int i;
 647
 648         memset(mega_stats, 0, sizeof(*mega_stats));
 649
 650         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 651         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 652
 653         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 654
 655         for_each_possible_cpu(i) {
 656                 const struct dp_stats_percpu *percpu_stats;
 657                 struct dp_stats_percpu local_stats;
 658                 unsigned int start;
 659
 660                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 661
 662                 do {
 663                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 664                         local_stats = *percpu_stats;
 665                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 666
 667                 stats->n_hit += local_stats.n_hit;
 668                 stats->n_missed += local_stats.n_missed;
 669                 stats->n_lost += local_stats.n_lost;
 670                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 671         }
 672 }
 673
 674 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 675 {
 676         return NLMSG_ALIGN(sizeof(struct ovs_header))
 677                 + nla_total_size(ovs_key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 678                 + nla_total_size(ovs_key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 679                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 680                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 681                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 682                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 683 }
 684
 685 /* Called with ovs_mutex or RCU read lock. */
 686 static int ovs_flow_cmd_fill_match(const struct sw_flow *flow,
 687                                    struct sk_buff *skb)
 688 {
 689         struct nlattr *nla;
 690         int err;
 691
 692         /* Fill flow key. */
 693         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 694         if (!nla)
 695                 return -EMSGSIZE;
 696
 697         err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
 698         if (err)
 699                 return err;
 700
 701         nla_nest_end(skb, nla);
 702
 703         /* Fill flow mask. */
 704         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 705         if (!nla)
 706                 return -EMSGSIZE;
 707
 708         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 709         if (err)
 710                 return err;
 711
 712         nla_nest_end(skb, nla);
 713         return 0;
 714 }
 715
 716 /* Called with ovs_mutex or RCU read lock. */
 717 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 718                                    struct sk_buff *skb)
 719 {
 720         struct ovs_flow_stats stats;
 721         __be16 tcp_flags;
 722         unsigned long used;
 723
 724         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 725
 726         if (used &&
 727             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 728                 return -EMSGSIZE;
 729
 730         if (stats.n_packets &&
 731             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 732                 return -EMSGSIZE;
 733
 734         if ((u8)ntohs(tcp_flags) &&
 735              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 736                 return -EMSGSIZE;
 737
 738         return 0;
 739 }
 740
 741 /* Called with ovs_mutex or RCU read lock. */
 742 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 743                                      struct sk_buff *skb, int skb_orig_len)
 744 {
 745         struct nlattr *start;
 746         int err;
 747
 748         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 749          * this is the first flow to be dumped into 'skb'.  This is unusual for
 750          * Netlink but individual action lists can be longer than
 751          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 752          * The userspace caller can always fetch the actions separately if it
 753          * really wants them.  (Most userspace callers in fact don't care.)
 754          *
 755          * This can only fail for dump operations because the skb is always
 756          * properly sized for single flows.
 757          */
 758         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 759         if (start) {
 760                 const struct sw_flow_actions *sf_acts;
 761
 762                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 763                 err = ovs_nla_put_actions(sf_acts->actions,
 764                                           sf_acts->actions_len, skb);
 765
 766                 if (!err)
 767                         nla_nest_end(skb, start);
 768                 else {
 769                         if (skb_orig_len)
 770                                 return err;
 771
 772                         nla_nest_cancel(skb, start);
 773                 }
 774         } else if (skb_orig_len) {
 775                 return -EMSGSIZE;
 776         }
 777
 778         return 0;
 779 }
 780
 781 /* Called with ovs_mutex or RCU read lock. */
 782 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 783                                   struct sk_buff *skb, u32 portid,
 784                                   u32 seq, u32 flags, u8 cmd)
 785 {
 786         const int skb_orig_len = skb->len;
 787         struct ovs_header *ovs_header;
 788         int err;
 789
 790         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 791                                  flags, cmd);
 792         if (!ovs_header)
 793                 return -EMSGSIZE;
 794
 795         ovs_header->dp_ifindex = dp_ifindex;
 796
 797         err = ovs_flow_cmd_fill_match(flow, skb);
 798         if (err)
 799                 goto error;
 800
 801         err = ovs_flow_cmd_fill_stats(flow, skb);
 802         if (err)
 803                 goto error;
 804
 805         err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 806         if (err)
 807                 goto error;
 808
 809         genlmsg_end(skb, ovs_header);
 810         return 0;
 811
 812 error:
 813         genlmsg_cancel(skb, ovs_header);
 814         return err;
 815 }
 816
 817 /* May not be called with RCU read lock. */
 818 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 819                                                struct genl_info *info,
 820                                                bool always)
 821 {
 822         struct sk_buff *skb;
 823
 824         if (!always && !ovs_must_notify(&dp_flow_genl_family, info,
 825                                         GROUP_ID(&ovs_dp_flow_multicast_group)))
 826                 return NULL;
 827
 828         skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
 829         if (!skb)
 830                 return ERR_PTR(-ENOMEM);
 831
 832         return skb;
 833 }
 834
 835 /* Called with ovs_mutex. */
 836 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 837                                                int dp_ifindex,
 838                                                struct genl_info *info, u8 cmd,
 839                                                bool always)
 840 {
 841         struct sk_buff *skb;
 842         int retval;
 843
 844         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
 845                                       always);
 846         if (IS_ERR_OR_NULL(skb))
 847                 return skb;
 848
 849         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 850                                         info->snd_portid, info->snd_seq, 0,
 851                                         cmd);
 852         BUG_ON(retval < 0);
 853         return skb;
 854 }
 855
 856 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 857 {
 858         struct nlattr **a = info->attrs;
 859         struct ovs_header *ovs_header = info->userhdr;
 860         struct sw_flow *flow, *new_flow;
 861         struct sw_flow_mask mask;
 862         struct sk_buff *reply;
 863         struct datapath *dp;
 864         struct sw_flow_actions *acts;
 865         struct sw_flow_match match;
 866         int error;
 867         bool log = !a[OVS_FLOW_ATTR_PROBE];
 868
 869         /* Must have key and actions. */
 870         error = -EINVAL;
 871         if (!a[OVS_FLOW_ATTR_KEY]) {
 872                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 873                 goto error;
 874         }
 875         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 876                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 877                 goto error;
 878         }
 879
 880         /* Most of the time we need to allocate a new flow, do it before
 881          * locking.
 882          */
 883         new_flow = ovs_flow_alloc();
 884         if (IS_ERR(new_flow)) {
 885                 error = PTR_ERR(new_flow);
 886                 goto error;
 887         }
 888
 889         /* Extract key. */
 890         ovs_match_init(&match, &new_flow->unmasked_key, &mask);
 891         error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
 892                                   a[OVS_FLOW_ATTR_MASK], log);
 893         if (error)
 894                 goto err_kfree_flow;
 895
 896         ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
 897
 898         /* Validate actions. */
 899         error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
 900                                      &acts, log);
 901         if (error) {
 902                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 903                 goto err_kfree_flow;
 904         }
 905
 906         reply = ovs_flow_cmd_alloc_info(acts, info, false);
 907         if (IS_ERR(reply)) {
 908                 error = PTR_ERR(reply);
 909                 goto err_kfree_acts;
 910         }
 911
 912         ovs_lock();
 913         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 914         if (unlikely(!dp)) {
 915                 error = -ENODEV;
 916                 goto err_unlock_ovs;
 917         }
 918         /* Check if this is a duplicate flow */
 919         flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
 920         if (likely(!flow)) {
 921                 rcu_assign_pointer(new_flow->sf_acts, acts);
 922
 923                 /* Put flow in bucket. */
 924                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 925                 if (unlikely(error)) {
 926                         acts = NULL;
 927                         goto err_unlock_ovs;
 928                 }
 929
 930                 if (unlikely(reply)) {
 931                         error = ovs_flow_cmd_fill_info(new_flow,
 932                                                        ovs_header->dp_ifindex,
 933                                                        reply, info->snd_portid,
 934                                                        info->snd_seq, 0,
 935                                                        OVS_FLOW_CMD_NEW);
 936                         BUG_ON(error < 0);
 937                 }
 938                 ovs_unlock();
 939         } else {
 940                 struct sw_flow_actions *old_acts;
 941
 942                 /* Bail out if we're not allowed to modify an existing flow.
 943                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 944                  * because Generic Netlink treats the latter as a dump
 945                  * request.  We also accept NLM_F_EXCL in case that bug ever
 946                  * gets fixed.
 947                  */
 948                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
 949                                                          | NLM_F_EXCL))) {
 950                         error = -EEXIST;
 951                         goto err_unlock_ovs;
 952                 }
 953                 /* The unmasked key has to be the same for flow updates. */
 954                 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
 955                         /* Look for any overlapping flow. */
 956                         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
 957                         if (!flow) {
 958                                 error = -ENOENT;
 959                                 goto err_unlock_ovs;
 960                         }
 961                 }
 962                 /* Update actions. */
 963                 old_acts = ovsl_dereference(flow->sf_acts);
 964                 rcu_assign_pointer(flow->sf_acts, acts);
 965
 966                 if (unlikely(reply)) {
 967                         error = ovs_flow_cmd_fill_info(flow,
 968                                                        ovs_header->dp_ifindex,
 969                                                        reply, info->snd_portid,
 970                                                        info->snd_seq, 0,
 971                                                        OVS_FLOW_CMD_NEW);
 972                         BUG_ON(error < 0);
 973                 }
 974                 ovs_unlock();
 975
 976                 ovs_nla_free_flow_actions(old_acts);
 977                 ovs_flow_free(new_flow, false);
 978         }
 979
 980         if (reply)
 981                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
 982         return 0;
 983
 984 err_unlock_ovs:
 985         ovs_unlock();
 986         kfree_skb(reply);
 987 err_kfree_acts:
 988         kfree(acts);
 989 err_kfree_flow:
 990         ovs_flow_free(new_flow, false);
 991 error:
 992         return error;
 993 }
 994
 995 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
 996 static struct sw_flow_actions *get_flow_actions(const struct nlattr *a,
 997                                                 const struct sw_flow_key *key,
 998                                                 const struct sw_flow_mask *mask,
 999                                                 bool log)
1000 {
1001         struct sw_flow_actions *acts;
1002         struct sw_flow_key masked_key;
1003         int error;
1004
1005         ovs_flow_mask_key(&masked_key, key, mask);
1006         error = ovs_nla_copy_actions(a, &masked_key, &acts, log);
1007         if (error) {
1008                 OVS_NLERR(log,
1009                           "Actions may not be safe on all matching packets");
1010                 return ERR_PTR(error);
1011         }
1012
1013         return acts;
1014 }
1015
1016 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1017 {
1018         struct nlattr **a = info->attrs;
1019         struct ovs_header *ovs_header = info->userhdr;
1020         struct sw_flow_key key;
1021         struct sw_flow *flow;
1022         struct sw_flow_mask mask;
1023         struct sk_buff *reply = NULL;
1024         struct datapath *dp;
1025         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1026         struct sw_flow_match match;
1027         int error;
1028         bool log = !a[OVS_FLOW_ATTR_PROBE];
1029
1030         /* Extract key. */
1031         error = -EINVAL;
1032         if (!a[OVS_FLOW_ATTR_KEY]) {
1033                 OVS_NLERR(log, "Flow key attribute not present in set flow.");
1034                 goto error;
1035         }
1036
1037         ovs_match_init(&match, &key, &mask);
1038         error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
1039                                   a[OVS_FLOW_ATTR_MASK], log);
1040         if (error)
1041                 goto error;
1042
1043         /* Validate actions. */
1044         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1045                 acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask,
1046                                         log);
1047                 if (IS_ERR(acts)) {
1048                         error = PTR_ERR(acts);
1049                         goto error;
1050                 }
1051
1052                 /* Can allocate before locking if have acts. */
1053                 reply = ovs_flow_cmd_alloc_info(acts, info, false);
1054                 if (IS_ERR(reply)) {
1055                         error = PTR_ERR(reply);
1056                         goto err_kfree_acts;
1057                 }
1058         }
1059
1060         ovs_lock();
1061         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1062         if (unlikely(!dp)) {
1063                 error = -ENODEV;
1064                 goto err_unlock_ovs;
1065         }
1066         /* Check that the flow exists. */
1067         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1068         if (unlikely(!flow)) {
1069                 error = -ENOENT;
1070                 goto err_unlock_ovs;
1071         }
1072
1073         /* Update actions, if present. */
1074         if (likely(acts)) {
1075                 old_acts = ovsl_dereference(flow->sf_acts);
1076                 rcu_assign_pointer(flow->sf_acts, acts);
1077
1078                 if (unlikely(reply)) {
1079                         error = ovs_flow_cmd_fill_info(flow,
1080                                                        ovs_header->dp_ifindex,
1081                                                        reply, info->snd_portid,
1082                                                        info->snd_seq, 0,
1083                                                        OVS_FLOW_CMD_NEW);
1084                         BUG_ON(error < 0);
1085                 }
1086         } else {
1087                 /* Could not alloc without acts before locking. */
1088                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1089                                                 info, OVS_FLOW_CMD_NEW, false);
1090                 if (unlikely(IS_ERR(reply))) {
1091                         error = PTR_ERR(reply);
1092                         goto err_unlock_ovs;
1093                 }
1094         }
1095
1096         /* Clear stats. */
1097         if (a[OVS_FLOW_ATTR_CLEAR])
1098                 ovs_flow_stats_clear(flow);
1099         ovs_unlock();
1100
1101         if (reply)
1102                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1103         if (old_acts)
1104                 ovs_nla_free_flow_actions(old_acts);
1105
1106         return 0;
1107
1108 err_unlock_ovs:
1109         ovs_unlock();
1110         kfree_skb(reply);
1111 err_kfree_acts:
1112         kfree(acts);
1113 error:
1114         return error;
1115 }
1116
1117 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1118 {
1119         struct nlattr **a = info->attrs;
1120         struct ovs_header *ovs_header = info->userhdr;
1121         struct sw_flow_key key;
1122         struct sk_buff *reply;
1123         struct sw_flow *flow;
1124         struct datapath *dp;
1125         struct sw_flow_match match;
1126         int err;
1127         bool log = !a[OVS_FLOW_ATTR_PROBE];
1128
1129         if (!a[OVS_FLOW_ATTR_KEY]) {
1130                 OVS_NLERR(log,
1131                           "Flow get message rejected, Key attribute missing.");
1132                 return -EINVAL;
1133         }
1134
1135         ovs_match_init(&match, &key, NULL);
1136         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL, log);
1137         if (err)
1138                 return err;
1139
1140         ovs_lock();
1141         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1142         if (!dp) {
1143                 err = -ENODEV;
1144                 goto unlock;
1145         }
1146
1147         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1148         if (!flow) {
1149                 err = -ENOENT;
1150                 goto unlock;
1151         }
1152
1153         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1154                                         OVS_FLOW_CMD_NEW, true);
1155         if (IS_ERR(reply)) {
1156                 err = PTR_ERR(reply);
1157                 goto unlock;
1158         }
1159
1160         ovs_unlock();
1161         return genlmsg_reply(reply, info);
1162 unlock:
1163         ovs_unlock();
1164         return err;
1165 }
1166
1167 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1168 {
1169         struct nlattr **a = info->attrs;
1170         struct ovs_header *ovs_header = info->userhdr;
1171         struct sw_flow_key key;
1172         struct sk_buff *reply;
1173         struct sw_flow *flow;
1174         struct datapath *dp;
1175         struct sw_flow_match match;
1176         int err;
1177         bool log = !a[OVS_FLOW_ATTR_PROBE];
1178
1179         if (likely(a[OVS_FLOW_ATTR_KEY])) {
1180                 ovs_match_init(&match, &key, NULL);
1181                 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL,
1182                                         log);
1183                 if (unlikely(err))
1184                         return err;
1185         }
1186
1187         ovs_lock();
1188         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1189         if (unlikely(!dp)) {
1190                 err = -ENODEV;
1191                 goto unlock;
1192         }
1193
1194         if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1195                 err = ovs_flow_tbl_flush(&dp->table);
1196                 goto unlock;
1197         }
1198
1199         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1200         if (unlikely(!flow)) {
1201                 err = -ENOENT;
1202                 goto unlock;
1203         }
1204
1205         ovs_flow_tbl_remove(&dp->table, flow);
1206         ovs_unlock();
1207
1208         reply = ovs_flow_cmd_alloc_info(rcu_dereference_raw(flow->sf_acts),
1209                                         info, false);
1210
1211         if (likely(reply)) {
1212                 if (likely(!IS_ERR(reply))) {
1213                         rcu_read_lock();        /*To keep RCU checker happy. */
1214                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1215                                                      reply, info->snd_portid,
1216                                                      info->snd_seq, 0,
1217                                                      OVS_FLOW_CMD_DEL);
1218                         rcu_read_unlock();
1219                         BUG_ON(err < 0);
1220                         ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1221                 } else {
1222                         genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1223                                      GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1224
1225                 }
1226         }
1227
1228         ovs_flow_free(flow, true);
1229         return 0;
1230 unlock:
1231         ovs_unlock();
1232         return err;
1233 }
1234
1235 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1236 {
1237         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1238         struct table_instance *ti;
1239         struct datapath *dp;
1240
1241         rcu_read_lock();
1242         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1243         if (!dp) {
1244                 rcu_read_unlock();
1245                 return -ENODEV;
1246         }
1247
1248         ti = rcu_dereference(dp->table.ti);
1249         for (;;) {
1250                 struct sw_flow *flow;
1251                 u32 bucket, obj;
1252
1253                 bucket = cb->args[0];
1254                 obj = cb->args[1];
1255                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1256                 if (!flow)
1257                         break;
1258
1259                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1260                                            NETLINK_CB(cb->skb).portid,
1261                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1262                                            OVS_FLOW_CMD_NEW) < 0)
1263                         break;
1264
1265                 cb->args[0] = bucket;
1266                 cb->args[1] = obj;
1267         }
1268         rcu_read_unlock();
1269         return skb->len;
1270 }
1271
1272 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1273         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1274         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1275         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1276         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1277         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1278 };
1279
1280 static struct genl_ops dp_flow_genl_ops[] = {
1281         { .cmd = OVS_FLOW_CMD_NEW,
1282           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1283           .policy = flow_policy,
1284           .doit = ovs_flow_cmd_new
1285         },
1286         { .cmd = OVS_FLOW_CMD_DEL,
1287           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1288           .policy = flow_policy,
1289           .doit = ovs_flow_cmd_del
1290         },
1291         { .cmd = OVS_FLOW_CMD_GET,
1292           .flags = 0,               /* OK for unprivileged users. */
1293           .policy = flow_policy,
1294           .doit = ovs_flow_cmd_get,
1295           .dumpit = ovs_flow_cmd_dump
1296         },
1297         { .cmd = OVS_FLOW_CMD_SET,
1298           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1299           .policy = flow_policy,
1300           .doit = ovs_flow_cmd_set,
1301         },
1302 };
1303
1304 static struct genl_family dp_flow_genl_family = {
1305         .id = GENL_ID_GENERATE,
1306         .hdrsize = sizeof(struct ovs_header),
1307         .name = OVS_FLOW_FAMILY,
1308         .version = OVS_FLOW_VERSION,
1309         .maxattr = OVS_FLOW_ATTR_MAX,
1310         .netnsok = true,
1311         .parallel_ops = true,
1312         .ops = dp_flow_genl_ops,
1313         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1314         .mcgrps = &ovs_dp_flow_multicast_group,
1315         .n_mcgrps = 1,
1316 };
1317
1318 static size_t ovs_dp_cmd_msg_size(void)
1319 {
1320         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1321
1322         msgsize += nla_total_size(IFNAMSIZ);
1323         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1324         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1325         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1326
1327         return msgsize;
1328 }
1329
1330 /* Called with ovs_mutex. */
1331 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1332                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1333 {
1334         struct ovs_header *ovs_header;
1335         struct ovs_dp_stats dp_stats;
1336         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1337         int err;
1338
1339         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1340                                    flags, cmd);
1341         if (!ovs_header)
1342                 goto error;
1343
1344         ovs_header->dp_ifindex = get_dpifindex(dp);
1345
1346         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1347         if (err)
1348                 goto nla_put_failure;
1349
1350         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1351         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1352                         &dp_stats))
1353                 goto nla_put_failure;
1354
1355         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1356                         sizeof(struct ovs_dp_megaflow_stats),
1357                         &dp_megaflow_stats))
1358                 goto nla_put_failure;
1359
1360         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1361                 goto nla_put_failure;
1362
1363         genlmsg_end(skb, ovs_header);
1364         return 0;
1365
1366 nla_put_failure:
1367         genlmsg_cancel(skb, ovs_header);
1368 error:
1369         return -EMSGSIZE;
1370 }
1371
1372 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1373 {
1374         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1375 }
1376
1377 /* Called with rcu_read_lock or ovs_mutex. */
1378 static struct datapath *lookup_datapath(struct net *net,
1379                                         const struct ovs_header *ovs_header,
1380                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1381 {
1382         struct datapath *dp;
1383
1384         if (!a[OVS_DP_ATTR_NAME])
1385                 dp = get_dp(net, ovs_header->dp_ifindex);
1386         else {
1387                 struct vport *vport;
1388
1389                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1390                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1391         }
1392         return dp ? dp : ERR_PTR(-ENODEV);
1393 }
1394
1395 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1396 {
1397         struct datapath *dp;
1398
1399         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1400         if (IS_ERR(dp))
1401                 return;
1402
1403         WARN(dp->user_features, "Dropping previously announced user features\n");
1404         dp->user_features = 0;
1405 }
1406
1407 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1408 {
1409         if (a[OVS_DP_ATTR_USER_FEATURES])
1410                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1411 }
1412
1413 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1414 {
1415         struct nlattr **a = info->attrs;
1416         struct vport_parms parms;
1417         struct sk_buff *reply;
1418         struct datapath *dp;
1419         struct vport *vport;
1420         struct ovs_net *ovs_net;
1421         int err, i;
1422
1423         err = -EINVAL;
1424         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1425                 goto err;
1426
1427         reply = ovs_dp_cmd_alloc_info(info);
1428         if (!reply)
1429                 return -ENOMEM;
1430
1431         err = -ENOMEM;
1432         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1433         if (dp == NULL)
1434                 goto err_free_reply;
1435
1436         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1437
1438         /* Allocate table. */
1439         err = ovs_flow_tbl_init(&dp->table);
1440         if (err)
1441                 goto err_free_dp;
1442
1443         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1444         if (!dp->stats_percpu) {
1445                 err = -ENOMEM;
1446                 goto err_destroy_table;
1447         }
1448
1449         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1450                             GFP_KERNEL);
1451         if (!dp->ports) {
1452                 err = -ENOMEM;
1453                 goto err_destroy_percpu;
1454         }
1455
1456         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1457                 INIT_HLIST_HEAD(&dp->ports[i]);
1458
1459         /* Set up our datapath device. */
1460         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1461         parms.type = OVS_VPORT_TYPE_INTERNAL;
1462         parms.options = NULL;
1463         parms.dp = dp;
1464         parms.port_no = OVSP_LOCAL;
1465         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1466
1467         ovs_dp_change(dp, a);
1468
1469         /* So far only local changes have been made, now need the lock. */
1470         ovs_lock();
1471
1472         vport = new_vport(&parms);
1473         if (IS_ERR(vport)) {
1474                 err = PTR_ERR(vport);
1475                 if (err == -EBUSY)
1476                         err = -EEXIST;
1477
1478                 if (err == -EEXIST) {
1479                         /* An outdated user space instance that does not understand
1480                          * the concept of user_features has attempted to create a new
1481                          * datapath and is likely to reuse it. Drop all user features.
1482                          */
1483                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1484                                 ovs_dp_reset_user_features(skb, info);
1485                 }
1486
1487                 goto err_destroy_ports_array;
1488         }
1489
1490         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1491                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1492         BUG_ON(err < 0);
1493
1494         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1495         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1496         ovs_unlock();
1497
1498         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1499         return 0;
1500
1501 err_destroy_ports_array:
1502         ovs_unlock();
1503         kfree(dp->ports);
1504 err_destroy_percpu:
1505         free_percpu(dp->stats_percpu);
1506 err_destroy_table:
1507         ovs_flow_tbl_destroy(&dp->table);
1508 err_free_dp:
1509         release_net(ovs_dp_get_net(dp));
1510         kfree(dp);
1511 err_free_reply:
1512         kfree_skb(reply);
1513 err:
1514         return err;
1515 }
1516
1517 /* Called with ovs_mutex. */
1518 static void __dp_destroy(struct datapath *dp)
1519 {
1520         int i;
1521
1522         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1523                 struct vport *vport;
1524                 struct hlist_node *n;
1525
1526                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1527                         if (vport->port_no != OVSP_LOCAL)
1528                                 ovs_dp_detach_port(vport);
1529         }
1530
1531         list_del_rcu(&dp->list_node);
1532
1533         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1534          * all ports in datapath are destroyed first before freeing datapath.
1535          */
1536         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1537
1538         /* RCU destroy the flow table */
1539         call_rcu(&dp->rcu, destroy_dp_rcu);
1540 }
1541
1542 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1543 {
1544         struct sk_buff *reply;
1545         struct datapath *dp;
1546         int err;
1547
1548         reply = ovs_dp_cmd_alloc_info(info);
1549         if (!reply)
1550                 return -ENOMEM;
1551
1552         ovs_lock();
1553         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1554         err = PTR_ERR(dp);
1555         if (IS_ERR(dp))
1556                 goto err_unlock_free;
1557
1558         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1559                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1560         BUG_ON(err < 0);
1561
1562         __dp_destroy(dp);
1563         ovs_unlock();
1564
1565         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1566         return 0;
1567
1568 err_unlock_free:
1569         ovs_unlock();
1570         kfree_skb(reply);
1571         return err;
1572 }
1573
1574 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1575 {
1576         struct sk_buff *reply;
1577         struct datapath *dp;
1578         int err;
1579
1580         reply = ovs_dp_cmd_alloc_info(info);
1581         if (!reply)
1582                 return -ENOMEM;
1583
1584         ovs_lock();
1585         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1586         err = PTR_ERR(dp);
1587         if (IS_ERR(dp))
1588                 goto err_unlock_free;
1589
1590         ovs_dp_change(dp, info->attrs);
1591
1592         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1593                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1594         BUG_ON(err < 0);
1595         ovs_unlock();
1596
1597         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1598         return 0;
1599
1600 err_unlock_free:
1601         ovs_unlock();
1602         kfree_skb(reply);
1603         return err;
1604 }
1605
1606 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1607 {
1608         struct sk_buff *reply;
1609         struct datapath *dp;
1610         int err;
1611
1612         reply = ovs_dp_cmd_alloc_info(info);
1613         if (!reply)
1614                 return -ENOMEM;
1615
1616         ovs_lock();
1617         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1618         if (IS_ERR(dp)) {
1619                 err = PTR_ERR(dp);
1620                 goto err_unlock_free;
1621         }
1622         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1623                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1624         BUG_ON(err < 0);
1625         ovs_unlock();
1626
1627         return genlmsg_reply(reply, info);
1628
1629 err_unlock_free:
1630         ovs_unlock();
1631         kfree_skb(reply);
1632         return err;
1633 }
1634
1635 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1636 {
1637         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1638         struct datapath *dp;
1639         int skip = cb->args[0];
1640         int i = 0;
1641
1642         ovs_lock();
1643         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1644                 if (i >= skip &&
1645                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1646                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1647                                          OVS_DP_CMD_NEW) < 0)
1648                         break;
1649                 i++;
1650         }
1651         ovs_unlock();
1652
1653         cb->args[0] = i;
1654
1655         return skb->len;
1656 }
1657
1658 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1659         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1660         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1661         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1662 };
1663
1664 static struct genl_ops dp_datapath_genl_ops[] = {
1665         { .cmd = OVS_DP_CMD_NEW,
1666           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1667           .policy = datapath_policy,
1668           .doit = ovs_dp_cmd_new
1669         },
1670         { .cmd = OVS_DP_CMD_DEL,
1671           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1672           .policy = datapath_policy,
1673           .doit = ovs_dp_cmd_del
1674         },
1675         { .cmd = OVS_DP_CMD_GET,
1676           .flags = 0,               /* OK for unprivileged users. */
1677           .policy = datapath_policy,
1678           .doit = ovs_dp_cmd_get,
1679           .dumpit = ovs_dp_cmd_dump
1680         },
1681         { .cmd = OVS_DP_CMD_SET,
1682           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1683           .policy = datapath_policy,
1684           .doit = ovs_dp_cmd_set,
1685         },
1686 };
1687
1688 static struct genl_family dp_datapath_genl_family = {
1689         .id = GENL_ID_GENERATE,
1690         .hdrsize = sizeof(struct ovs_header),
1691         .name = OVS_DATAPATH_FAMILY,
1692         .version = OVS_DATAPATH_VERSION,
1693         .maxattr = OVS_DP_ATTR_MAX,
1694         .netnsok = true,
1695         .parallel_ops = true,
1696         .ops = dp_datapath_genl_ops,
1697         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1698         .mcgrps = &ovs_dp_datapath_multicast_group,
1699         .n_mcgrps = 1,
1700 };
1701
1702 /* Called with ovs_mutex or RCU read lock. */
1703 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1704                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1705 {
1706         struct ovs_header *ovs_header;
1707         struct ovs_vport_stats vport_stats;
1708         int err;
1709
1710         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1711                                  flags, cmd);
1712         if (!ovs_header)
1713                 return -EMSGSIZE;
1714
1715         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1716
1717         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1718             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1719             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)))
1720                 goto nla_put_failure;
1721
1722         ovs_vport_get_stats(vport, &vport_stats);
1723         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1724                     &vport_stats))
1725                 goto nla_put_failure;
1726
1727         if (ovs_vport_get_upcall_portids(vport, skb))
1728                 goto nla_put_failure;
1729
1730         err = ovs_vport_get_options(vport, skb);
1731         if (err == -EMSGSIZE)
1732                 goto error;
1733
1734         genlmsg_end(skb, ovs_header);
1735         return 0;
1736
1737 nla_put_failure:
1738         err = -EMSGSIZE;
1739 error:
1740         genlmsg_cancel(skb, ovs_header);
1741         return err;
1742 }
1743
1744 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1745 {
1746         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1747 }
1748
1749 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1750 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1751                                          u32 seq, u8 cmd)
1752 {
1753         struct sk_buff *skb;
1754         int retval;
1755
1756         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1757         if (!skb)
1758                 return ERR_PTR(-ENOMEM);
1759
1760         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1761         BUG_ON(retval < 0);
1762
1763         return skb;
1764 }
1765
1766 /* Called with ovs_mutex or RCU read lock. */
1767 static struct vport *lookup_vport(struct net *net,
1768                                   const struct ovs_header *ovs_header,
1769                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1770 {
1771         struct datapath *dp;
1772         struct vport *vport;
1773
1774         if (a[OVS_VPORT_ATTR_NAME]) {
1775                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1776                 if (!vport)
1777                         return ERR_PTR(-ENODEV);
1778                 if (ovs_header->dp_ifindex &&
1779                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1780                         return ERR_PTR(-ENODEV);
1781                 return vport;
1782         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1783                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1784
1785                 if (port_no >= DP_MAX_PORTS)
1786                         return ERR_PTR(-EFBIG);
1787
1788                 dp = get_dp(net, ovs_header->dp_ifindex);
1789                 if (!dp)
1790                         return ERR_PTR(-ENODEV);
1791
1792                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1793                 if (!vport)
1794                         return ERR_PTR(-ENODEV);
1795                 return vport;
1796         } else
1797                 return ERR_PTR(-EINVAL);
1798 }
1799
1800 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1801 {
1802         struct nlattr **a = info->attrs;
1803         struct ovs_header *ovs_header = info->userhdr;
1804         struct vport_parms parms;
1805         struct sk_buff *reply;
1806         struct vport *vport;
1807         struct datapath *dp;
1808         u32 port_no;
1809         int err;
1810
1811         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1812             !a[OVS_VPORT_ATTR_UPCALL_PID])
1813                 return -EINVAL;
1814
1815         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1816                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1817         if (port_no >= DP_MAX_PORTS)
1818                 return -EFBIG;
1819
1820         reply = ovs_vport_cmd_alloc_info();
1821         if (!reply)
1822                 return -ENOMEM;
1823
1824         ovs_lock();
1825         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1826         err = -ENODEV;
1827         if (!dp)
1828                 goto exit_unlock_free;
1829
1830         if (port_no) {
1831                 vport = ovs_vport_ovsl(dp, port_no);
1832                 err = -EBUSY;
1833                 if (vport)
1834                         goto exit_unlock_free;
1835         } else {
1836                 for (port_no = 1; ; port_no++) {
1837                         if (port_no >= DP_MAX_PORTS) {
1838                                 err = -EFBIG;
1839                                 goto exit_unlock_free;
1840                         }
1841                         vport = ovs_vport_ovsl(dp, port_no);
1842                         if (!vport)
1843                                 break;
1844                 }
1845         }
1846
1847         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1848         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1849         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1850         parms.dp = dp;
1851         parms.port_no = port_no;
1852         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1853
1854         vport = new_vport(&parms);
1855         err = PTR_ERR(vport);
1856         if (IS_ERR(vport))
1857                 goto exit_unlock_free;
1858
1859         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1860                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1861         BUG_ON(err < 0);
1862         ovs_unlock();
1863
1864         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1865         return 0;
1866
1867 exit_unlock_free:
1868         ovs_unlock();
1869         kfree_skb(reply);
1870         return err;
1871 }
1872
1873 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1874 {
1875         struct nlattr **a = info->attrs;
1876         struct sk_buff *reply;
1877         struct vport *vport;
1878         int err;
1879
1880         reply = ovs_vport_cmd_alloc_info();
1881         if (!reply)
1882                 return -ENOMEM;
1883
1884         ovs_lock();
1885         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1886         err = PTR_ERR(vport);
1887         if (IS_ERR(vport))
1888                 goto exit_unlock_free;
1889
1890         if (a[OVS_VPORT_ATTR_TYPE] &&
1891             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1892                 err = -EINVAL;
1893                 goto exit_unlock_free;
1894         }
1895
1896         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1897                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1898                 if (err)
1899                         goto exit_unlock_free;
1900         }
1901
1902         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1903                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
1904
1905                 err = ovs_vport_set_upcall_portids(vport, ids);
1906                 if (err)
1907                         goto exit_unlock_free;
1908         }
1909
1910         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1911                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1912         BUG_ON(err < 0);
1913         ovs_unlock();
1914
1915         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1916         return 0;
1917
1918 exit_unlock_free:
1919         ovs_unlock();
1920         kfree_skb(reply);
1921         return err;
1922 }
1923
1924 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1925 {
1926         struct nlattr **a = info->attrs;
1927         struct sk_buff *reply;
1928         struct vport *vport;
1929         int err;
1930
1931         reply = ovs_vport_cmd_alloc_info();
1932         if (!reply)
1933                 return -ENOMEM;
1934
1935         ovs_lock();
1936         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1937         err = PTR_ERR(vport);
1938         if (IS_ERR(vport))
1939                 goto exit_unlock_free;
1940
1941         if (vport->port_no == OVSP_LOCAL) {
1942                 err = -EINVAL;
1943                 goto exit_unlock_free;
1944         }
1945
1946         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1947                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1948         BUG_ON(err < 0);
1949         ovs_dp_detach_port(vport);
1950         ovs_unlock();
1951
1952         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, 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_get(struct sk_buff *skb, struct genl_info *info)
1962 {
1963         struct nlattr **a = info->attrs;
1964         struct ovs_header *ovs_header = info->userhdr;
1965         struct sk_buff *reply;
1966         struct vport *vport;
1967         int err;
1968
1969         reply = ovs_vport_cmd_alloc_info();
1970         if (!reply)
1971                 return -ENOMEM;
1972
1973         rcu_read_lock();
1974         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1975         err = PTR_ERR(vport);
1976         if (IS_ERR(vport))
1977                 goto exit_unlock_free;
1978         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1979                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1980         BUG_ON(err < 0);
1981         rcu_read_unlock();
1982
1983         return genlmsg_reply(reply, info);
1984
1985 exit_unlock_free:
1986         rcu_read_unlock();
1987         kfree_skb(reply);
1988         return err;
1989 }
1990
1991 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1992 {
1993         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1994         struct datapath *dp;
1995         int bucket = cb->args[0], skip = cb->args[1];
1996         int i, j = 0;
1997
1998         rcu_read_lock();
1999         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2000         if (!dp) {
2001                 rcu_read_unlock();
2002                 return -ENODEV;
2003         }
2004         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2005                 struct vport *vport;
2006
2007                 j = 0;
2008                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2009                         if (j >= skip &&
2010                             ovs_vport_cmd_fill_info(vport, skb,
2011                                                     NETLINK_CB(cb->skb).portid,
2012                                                     cb->nlh->nlmsg_seq,
2013                                                     NLM_F_MULTI,
2014                                                     OVS_VPORT_CMD_NEW) < 0)
2015                                 goto out;
2016
2017                         j++;
2018                 }
2019                 skip = 0;
2020         }
2021 out:
2022         rcu_read_unlock();
2023
2024         cb->args[0] = i;
2025         cb->args[1] = j;
2026
2027         return skb->len;
2028 }
2029
2030 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2031         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2032         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2033         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2034         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2035         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2036         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2037 };
2038
2039 static struct genl_ops dp_vport_genl_ops[] = {
2040         { .cmd = OVS_VPORT_CMD_NEW,
2041           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2042           .policy = vport_policy,
2043           .doit = ovs_vport_cmd_new
2044         },
2045         { .cmd = OVS_VPORT_CMD_DEL,
2046           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2047           .policy = vport_policy,
2048           .doit = ovs_vport_cmd_del
2049         },
2050         { .cmd = OVS_VPORT_CMD_GET,
2051           .flags = 0,               /* OK for unprivileged users. */
2052           .policy = vport_policy,
2053           .doit = ovs_vport_cmd_get,
2054           .dumpit = ovs_vport_cmd_dump
2055         },
2056         { .cmd = OVS_VPORT_CMD_SET,
2057           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2058           .policy = vport_policy,
2059           .doit = ovs_vport_cmd_set,
2060         },
2061 };
2062
2063 struct genl_family dp_vport_genl_family = {
2064         .id = GENL_ID_GENERATE,
2065         .hdrsize = sizeof(struct ovs_header),
2066         .name = OVS_VPORT_FAMILY,
2067         .version = OVS_VPORT_VERSION,
2068         .maxattr = OVS_VPORT_ATTR_MAX,
2069         .netnsok = true,
2070         .parallel_ops = true,
2071         .ops = dp_vport_genl_ops,
2072         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2073         .mcgrps = &ovs_dp_vport_multicast_group,
2074         .n_mcgrps = 1,
2075 };
2076
2077 static struct genl_family *dp_genl_families[] = {
2078         &dp_datapath_genl_family,
2079         &dp_vport_genl_family,
2080         &dp_flow_genl_family,
2081         &dp_packet_genl_family,
2082 };
2083
2084 static void dp_unregister_genl(int n_families)
2085 {
2086         int i;
2087
2088         for (i = 0; i < n_families; i++)
2089                 genl_unregister_family(dp_genl_families[i]);
2090 }
2091
2092 static int dp_register_genl(void)
2093 {
2094         int err;
2095         int i;
2096
2097         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2098
2099                 err = genl_register_family(dp_genl_families[i]);
2100                 if (err)
2101                         goto error;
2102         }
2103
2104         return 0;
2105
2106 error:
2107         dp_unregister_genl(i);
2108         return err;
2109 }
2110
2111 static int __net_init ovs_init_net(struct net *net)
2112 {
2113         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2114
2115         INIT_LIST_HEAD(&ovs_net->dps);
2116         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2117         return 0;
2118 }
2119
2120 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2121                                             struct list_head *head)
2122 {
2123         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2124         struct datapath *dp;
2125
2126         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2127                 int i;
2128
2129                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2130                         struct vport *vport;
2131
2132                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2133                                 struct netdev_vport *netdev_vport;
2134
2135                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2136                                         continue;
2137
2138                                 netdev_vport = netdev_vport_priv(vport);
2139                                 if (dev_net(netdev_vport->dev) == dnet)
2140                                         list_add(&vport->detach_list, head);
2141                         }
2142                 }
2143         }
2144 }
2145
2146 static void __net_exit ovs_exit_net(struct net *dnet)
2147 {
2148         struct datapath *dp, *dp_next;
2149         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2150         struct vport *vport, *vport_next;
2151         struct net *net;
2152         LIST_HEAD(head);
2153
2154         ovs_lock();
2155         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2156                 __dp_destroy(dp);
2157
2158         rtnl_lock();
2159         for_each_net(net)
2160                 list_vports_from_net(net, dnet, &head);
2161         rtnl_unlock();
2162
2163         /* Detach all vports from given namespace. */
2164         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2165                 list_del(&vport->detach_list);
2166                 ovs_dp_detach_port(vport);
2167         }
2168
2169         ovs_unlock();
2170
2171         cancel_work_sync(&ovs_net->dp_notify_work);
2172 }
2173
2174 static struct pernet_operations ovs_net_ops = {
2175         .init = ovs_init_net,
2176         .exit = ovs_exit_net,
2177         .id   = &ovs_net_id,
2178         .size = sizeof(struct ovs_net),
2179 };
2180
2181 DEFINE_COMPAT_PNET_REG_FUNC(device);
2182
2183 static int __init dp_init(void)
2184 {
2185         int err;
2186
2187         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2188
2189         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2190                 VERSION);
2191
2192         err = action_fifos_init();
2193         if (err)
2194                 goto error;
2195
2196         err = ovs_internal_dev_rtnl_link_register();
2197         if (err)
2198                 goto error_action_fifos_exit;
2199
2200         err = ovs_flow_init();
2201         if (err)
2202                 goto error_unreg_rtnl_link;
2203
2204         err = ovs_vport_init();
2205         if (err)
2206                 goto error_flow_exit;
2207
2208         err = register_pernet_device(&ovs_net_ops);
2209         if (err)
2210                 goto error_vport_exit;
2211
2212         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2213         if (err)
2214                 goto error_netns_exit;
2215
2216         err = dp_register_genl();
2217         if (err < 0)
2218                 goto error_unreg_notifier;
2219
2220         return 0;
2221
2222 error_unreg_notifier:
2223         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2224 error_netns_exit:
2225         unregister_pernet_device(&ovs_net_ops);
2226 error_vport_exit:
2227         ovs_vport_exit();
2228 error_flow_exit:
2229         ovs_flow_exit();
2230 error_unreg_rtnl_link:
2231         ovs_internal_dev_rtnl_link_unregister();
2232 error_action_fifos_exit:
2233         action_fifos_exit();
2234 error:
2235         return err;
2236 }
2237
2238 static void dp_cleanup(void)
2239 {
2240         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2241         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2242         unregister_pernet_device(&ovs_net_ops);
2243         rcu_barrier();
2244         ovs_vport_exit();
2245         ovs_flow_exit();
2246         ovs_internal_dev_rtnl_link_unregister();
2247         action_fifos_exit();
2248 }
2249
2250 module_init(dp_init);
2251 module_exit(dp_cleanup);
2252
2253 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2254 MODULE_LICENSE("GPL");
2255 MODULE_VERSION(VERSION);