net/openvswitch/datapath.c

   1 /*
   2  * Copyright (c) 2007-2012 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/lockdep.h>
  48 #include <linux/openvswitch.h>
  49 #include <linux/rculist.h>
  50 #include <linux/dmi.h>
  51 #include <linux/workqueue.h>
  52 #include <net/genetlink.h>
  53 #include <net/net_namespace.h>
  54 #include <net/netns/generic.h>
  55
  56 #include "datapath.h"
  57 #include "flow.h"
  58 #include "vport-internal_dev.h"
  59 #include "vport-netdev.h"
  60
  61
  62 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
  63 static void rehash_flow_table(struct work_struct *work);
  64 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
  65
  66 int ovs_net_id __read_mostly;
  67
  68 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
  69                        struct genl_multicast_group *grp)
  70 {
  71         genl_notify(skb, genl_info_net(info), info->snd_portid,
  72                     grp->id, info->nlhdr, GFP_KERNEL);
  73 }
  74
  75 /**
  76  * DOC: Locking:
  77  *
  78  * All writes e.g. Writes to device state (add/remove datapath, port, set
  79  * operations on vports, etc.), Writes to other state (flow table
  80  * modifications, set miscellaneous datapath parameters, etc.) are protected
  81  * by ovs_lock.
  82  *
  83  * Reads are protected by RCU.
  84  *
  85  * There are a few special cases (mostly stats) that have their own
  86  * synchronization but they nest under all of above and don't interact with
  87  * each other.
  88  *
  89  * The RTNL lock nests inside ovs_mutex.
  90  */
  91
  92 static DEFINE_MUTEX(ovs_mutex);
  93
  94 void ovs_lock(void)
  95 {
  96         mutex_lock(&ovs_mutex);
  97 }
  98
  99 void ovs_unlock(void)
 100 {
 101         mutex_unlock(&ovs_mutex);
 102 }
 103
 104 #ifdef CONFIG_LOCKDEP
 105 int lockdep_ovsl_is_held(void)
 106 {
 107         if (debug_locks)
 108                 return lockdep_is_held(&ovs_mutex);
 109         else
 110                 return 1;
 111 }
 112 #endif
 113
 114 static struct vport *new_vport(const struct vport_parms *);
 115 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
 116                              const struct dp_upcall_info *);
 117 static int queue_userspace_packet(struct net *, int dp_ifindex,
 118                                   struct sk_buff *,
 119                                   const struct dp_upcall_info *);
 120
 121 /* Must be called with rcu_read_lock or ovs_mutex. */
 122 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 123 {
 124         struct datapath *dp = NULL;
 125         struct net_device *dev;
 126
 127         rcu_read_lock();
 128         dev = dev_get_by_index_rcu(net, dp_ifindex);
 129         if (dev) {
 130                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 131                 if (vport)
 132                         dp = vport->dp;
 133         }
 134         rcu_read_unlock();
 135
 136         return dp;
 137 }
 138
 139 /* Must be called with rcu_read_lock or ovs_mutex. */
 140 const char *ovs_dp_name(const struct datapath *dp)
 141 {
 142         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 143         return vport->ops->get_name(vport);
 144 }
 145
 146 static int get_dpifindex(struct datapath *dp)
 147 {
 148         struct vport *local;
 149         int ifindex;
 150
 151         rcu_read_lock();
 152
 153         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 154         if (local)
 155                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 156         else
 157                 ifindex = 0;
 158
 159         rcu_read_unlock();
 160
 161         return ifindex;
 162 }
 163
 164 static void destroy_dp_rcu(struct rcu_head *rcu)
 165 {
 166         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 167
 168         ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
 169         free_percpu(dp->stats_percpu);
 170         release_net(ovs_dp_get_net(dp));
 171         kfree(dp->ports);
 172         kfree(dp);
 173 }
 174
 175 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 176                                             u16 port_no)
 177 {
 178         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 179 }
 180
 181 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 182 {
 183         struct vport *vport;
 184         struct hlist_head *head;
 185
 186         head = vport_hash_bucket(dp, port_no);
 187         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 188                 if (vport->port_no == port_no)
 189                         return vport;
 190         }
 191         return NULL;
 192 }
 193
 194 /* Called with ovs_mutex. */
 195 static struct vport *new_vport(const struct vport_parms *parms)
 196 {
 197         struct vport *vport;
 198
 199         vport = ovs_vport_add(parms);
 200         if (!IS_ERR(vport)) {
 201                 struct datapath *dp = parms->dp;
 202                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 203
 204                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 205         }
 206         return vport;
 207 }
 208
 209 void ovs_dp_detach_port(struct vport *p)
 210 {
 211         ASSERT_OVSL();
 212
 213         /* First drop references to device. */
 214         hlist_del_rcu(&p->dp_hash_node);
 215
 216         /* Then destroy it. */
 217         ovs_vport_del(p);
 218 }
 219
 220 /* Must be called with rcu_read_lock. */
 221 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 222 {
 223         struct datapath *dp = p->dp;
 224         struct sw_flow *flow;
 225         struct dp_stats_percpu *stats;
 226         struct sw_flow_key key;
 227         u64 *stats_counter;
 228         int error;
 229         int key_len;
 230
 231         stats = this_cpu_ptr(dp->stats_percpu);
 232
 233         /* Extract flow from 'skb' into 'key'. */
 234         error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
 235         if (unlikely(error)) {
 236                 kfree_skb(skb);
 237                 return;
 238         }
 239
 240         /* Look up flow. */
 241         flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
 242         if (unlikely(!flow)) {
 243                 struct dp_upcall_info upcall;
 244
 245                 upcall.cmd = OVS_PACKET_CMD_MISS;
 246                 upcall.key = &key;
 247                 upcall.userdata = NULL;
 248                 upcall.portid = p->upcall_portid;
 249                 ovs_dp_upcall(dp, skb, &upcall);
 250                 consume_skb(skb);
 251                 stats_counter = &stats->n_missed;
 252                 goto out;
 253         }
 254
 255         OVS_CB(skb)->flow = flow;
 256
 257         stats_counter = &stats->n_hit;
 258         ovs_flow_used(OVS_CB(skb)->flow, skb);
 259         ovs_execute_actions(dp, skb);
 260
 261 out:
 262         /* Update datapath statistics. */
 263         u64_stats_update_begin(&stats->sync);
 264         (*stats_counter)++;
 265         u64_stats_update_end(&stats->sync);
 266 }
 267
 268 static struct genl_family dp_packet_genl_family = {
 269         .id = GENL_ID_GENERATE,
 270         .hdrsize = sizeof(struct ovs_header),
 271         .name = OVS_PACKET_FAMILY,
 272         .version = OVS_PACKET_VERSION,
 273         .maxattr = OVS_PACKET_ATTR_MAX,
 274         .netnsok = true,
 275         .parallel_ops = true,
 276 };
 277
 278 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 279                   const struct dp_upcall_info *upcall_info)
 280 {
 281         struct dp_stats_percpu *stats;
 282         int dp_ifindex;
 283         int err;
 284
 285         if (upcall_info->portid == 0) {
 286                 err = -ENOTCONN;
 287                 goto err;
 288         }
 289
 290         dp_ifindex = get_dpifindex(dp);
 291         if (!dp_ifindex) {
 292                 err = -ENODEV;
 293                 goto err;
 294         }
 295
 296         if (!skb_is_gso(skb))
 297                 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 298         else
 299                 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 300         if (err)
 301                 goto err;
 302
 303         return 0;
 304
 305 err:
 306         stats = this_cpu_ptr(dp->stats_percpu);
 307
 308         u64_stats_update_begin(&stats->sync);
 309         stats->n_lost++;
 310         u64_stats_update_end(&stats->sync);
 311
 312         return err;
 313 }
 314
 315 static int queue_gso_packets(struct net *net, int dp_ifindex,
 316                              struct sk_buff *skb,
 317                              const struct dp_upcall_info *upcall_info)
 318 {
 319         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 320         struct dp_upcall_info later_info;
 321         struct sw_flow_key later_key;
 322         struct sk_buff *segs, *nskb;
 323         int err;
 324
 325         segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
 326         if (IS_ERR(segs))
 327                 return PTR_ERR(segs);
 328
 329         /* Queue all of the segments. */
 330         skb = segs;
 331         do {
 332                 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
 333                 if (err)
 334                         break;
 335
 336                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 337                         /* The initial flow key extracted by ovs_flow_extract()
 338                          * in this case is for a first fragment, so we need to
 339                          * properly mark later fragments.
 340                          */
 341                         later_key = *upcall_info->key;
 342                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 343
 344                         later_info = *upcall_info;
 345                         later_info.key = &later_key;
 346                         upcall_info = &later_info;
 347                 }
 348         } while ((skb = skb->next));
 349
 350         /* Free all of the segments. */
 351         skb = segs;
 352         do {
 353                 nskb = skb->next;
 354                 if (err)
 355                         kfree_skb(skb);
 356                 else
 357                         consume_skb(skb);
 358         } while ((skb = nskb));
 359         return err;
 360 }
 361
 362 static size_t key_attr_size(void)
 363 {
 364         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 365                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 366                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
 367                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 368                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 369                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
 370                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
 371                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 372                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
 373                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 374                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 375                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 376                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 377                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 378                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 379                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 380                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 381                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 382                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 383 }
 384
 385 static size_t upcall_msg_size(const struct sk_buff *skb,
 386                               const struct nlattr *userdata)
 387 {
 388         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 389                 + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
 390                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 391
 392         /* OVS_PACKET_ATTR_USERDATA */
 393         if (userdata)
 394                 size += NLA_ALIGN(userdata->nla_len);
 395
 396         return size;
 397 }
 398
 399 static int queue_userspace_packet(struct net *net, int dp_ifindex,
 400                                   struct sk_buff *skb,
 401                                   const struct dp_upcall_info *upcall_info)
 402 {
 403         struct ovs_header *upcall;
 404         struct sk_buff *nskb = NULL;
 405         struct sk_buff *user_skb; /* to be queued to userspace */
 406         struct nlattr *nla;
 407         int err;
 408
 409         if (vlan_tx_tag_present(skb)) {
 410                 nskb = skb_clone(skb, GFP_ATOMIC);
 411                 if (!nskb)
 412                         return -ENOMEM;
 413
 414                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 415                 if (!nskb)
 416                         return -ENOMEM;
 417
 418                 nskb->vlan_tci = 0;
 419                 skb = nskb;
 420         }
 421
 422         if (nla_attr_size(skb->len) > USHRT_MAX) {
 423                 err = -EFBIG;
 424                 goto out;
 425         }
 426
 427         user_skb = genlmsg_new(upcall_msg_size(skb, upcall_info->userdata), GFP_ATOMIC);
 428         if (!user_skb) {
 429                 err = -ENOMEM;
 430                 goto out;
 431         }
 432
 433         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 434                              0, upcall_info->cmd);
 435         upcall->dp_ifindex = dp_ifindex;
 436
 437         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 438         ovs_flow_to_nlattrs(upcall_info->key, user_skb);
 439         nla_nest_end(user_skb, nla);
 440
 441         if (upcall_info->userdata)
 442                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 443                           nla_len(upcall_info->userdata),
 444                           nla_data(upcall_info->userdata));
 445
 446         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 447
 448         skb_copy_and_csum_dev(skb, nla_data(nla));
 449
 450         genlmsg_end(user_skb, upcall);
 451         err = genlmsg_unicast(net, user_skb, upcall_info->portid);
 452
 453 out:
 454         kfree_skb(nskb);
 455         return err;
 456 }
 457
 458 /* Called with ovs_mutex. */
 459 static int flush_flows(struct datapath *dp)
 460 {
 461         struct flow_table *old_table;
 462         struct flow_table *new_table;
 463
 464         old_table = ovsl_dereference(dp->table);
 465         new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
 466         if (!new_table)
 467                 return -ENOMEM;
 468
 469         rcu_assign_pointer(dp->table, new_table);
 470
 471         ovs_flow_tbl_deferred_destroy(old_table);
 472         return 0;
 473 }
 474
 475 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa, int attr_len)
 476 {
 477
 478         struct sw_flow_actions *acts;
 479         int new_acts_size;
 480         int req_size = NLA_ALIGN(attr_len);
 481         int next_offset = offsetof(struct sw_flow_actions, actions) +
 482                                         (*sfa)->actions_len;
 483
 484         if (req_size <= (ksize(*sfa) - next_offset))
 485                 goto out;
 486
 487         new_acts_size = ksize(*sfa) * 2;
 488
 489         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
 490                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
 491                         return ERR_PTR(-EMSGSIZE);
 492                 new_acts_size = MAX_ACTIONS_BUFSIZE;
 493         }
 494
 495         acts = ovs_flow_actions_alloc(new_acts_size);
 496         if (IS_ERR(acts))
 497                 return (void *)acts;
 498
 499         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
 500         acts->actions_len = (*sfa)->actions_len;
 501         kfree(*sfa);
 502         *sfa = acts;
 503
 504 out:
 505         (*sfa)->actions_len += req_size;
 506         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
 507 }
 508
 509 static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, int len)
 510 {
 511         struct nlattr *a;
 512
 513         a = reserve_sfa_size(sfa, nla_attr_size(len));
 514         if (IS_ERR(a))
 515                 return PTR_ERR(a);
 516
 517         a->nla_type = attrtype;
 518         a->nla_len = nla_attr_size(len);
 519
 520         if (data)
 521                 memcpy(nla_data(a), data, len);
 522         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
 523
 524         return 0;
 525 }
 526
 527 static inline int add_nested_action_start(struct sw_flow_actions **sfa, int attrtype)
 528 {
 529         int used = (*sfa)->actions_len;
 530         int err;
 531
 532         err = add_action(sfa, attrtype, NULL, 0);
 533         if (err)
 534                 return err;
 535
 536         return used;
 537 }
 538
 539 static inline void add_nested_action_end(struct sw_flow_actions *sfa, int st_offset)
 540 {
 541         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions + st_offset);
 542
 543         a->nla_len = sfa->actions_len - st_offset;
 544 }
 545
 546 static int validate_and_copy_actions(const struct nlattr *attr,
 547                                      const struct sw_flow_key *key, int depth,
 548                                      struct sw_flow_actions **sfa);
 549
 550 static int validate_and_copy_sample(const struct nlattr *attr,
 551                                     const struct sw_flow_key *key, int depth,
 552                                     struct sw_flow_actions **sfa)
 553 {
 554         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 555         const struct nlattr *probability, *actions;
 556         const struct nlattr *a;
 557         int rem, start, err, st_acts;
 558
 559         memset(attrs, 0, sizeof(attrs));
 560         nla_for_each_nested(a, attr, rem) {
 561                 int type = nla_type(a);
 562                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 563                         return -EINVAL;
 564                 attrs[type] = a;
 565         }
 566         if (rem)
 567                 return -EINVAL;
 568
 569         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 570         if (!probability || nla_len(probability) != sizeof(u32))
 571                 return -EINVAL;
 572
 573         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 574         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 575                 return -EINVAL;
 576
 577         /* validation done, copy sample action. */
 578         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
 579         if (start < 0)
 580                 return start;
 581         err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY, nla_data(probability), sizeof(u32));
 582         if (err)
 583                 return err;
 584         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
 585         if (st_acts < 0)
 586                 return st_acts;
 587
 588         err = validate_and_copy_actions(actions, key, depth + 1, sfa);
 589         if (err)
 590                 return err;
 591
 592         add_nested_action_end(*sfa, st_acts);
 593         add_nested_action_end(*sfa, start);
 594
 595         return 0;
 596 }
 597
 598 static int validate_tp_port(const struct sw_flow_key *flow_key)
 599 {
 600         if (flow_key->eth.type == htons(ETH_P_IP)) {
 601                 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
 602                         return 0;
 603         } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
 604                 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
 605                         return 0;
 606         }
 607
 608         return -EINVAL;
 609 }
 610
 611 static int validate_and_copy_set_tun(const struct nlattr *attr,
 612                                      struct sw_flow_actions **sfa)
 613 {
 614         struct ovs_key_ipv4_tunnel tun_key;
 615         int err, start;
 616
 617         err = ovs_ipv4_tun_from_nlattr(nla_data(attr), &tun_key);
 618         if (err)
 619                 return err;
 620
 621         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
 622         if (start < 0)
 623                 return start;
 624
 625         err = add_action(sfa, OVS_KEY_ATTR_IPV4_TUNNEL, &tun_key, sizeof(tun_key));
 626         add_nested_action_end(*sfa, start);
 627
 628         return err;
 629 }
 630
 631 static int validate_set(const struct nlattr *a,
 632                         const struct sw_flow_key *flow_key,
 633                         struct sw_flow_actions **sfa,
 634                         bool *set_tun)
 635 {
 636         const struct nlattr *ovs_key = nla_data(a);
 637         int key_type = nla_type(ovs_key);
 638
 639         /* There can be only one key in a action */
 640         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 641                 return -EINVAL;
 642
 643         if (key_type > OVS_KEY_ATTR_MAX ||
 644            (ovs_key_lens[key_type] != nla_len(ovs_key) &&
 645             ovs_key_lens[key_type] != -1))
 646                 return -EINVAL;
 647
 648         switch (key_type) {
 649         const struct ovs_key_ipv4 *ipv4_key;
 650         const struct ovs_key_ipv6 *ipv6_key;
 651         int err;
 652
 653         case OVS_KEY_ATTR_PRIORITY:
 654         case OVS_KEY_ATTR_SKB_MARK:
 655         case OVS_KEY_ATTR_ETHERNET:
 656                 break;
 657
 658         case OVS_KEY_ATTR_TUNNEL:
 659                 *set_tun = true;
 660                 err = validate_and_copy_set_tun(a, sfa);
 661                 if (err)
 662                         return err;
 663                 break;
 664
 665         case OVS_KEY_ATTR_IPV4:
 666                 if (flow_key->eth.type != htons(ETH_P_IP))
 667                         return -EINVAL;
 668
 669                 if (!flow_key->ip.proto)
 670                         return -EINVAL;
 671
 672                 ipv4_key = nla_data(ovs_key);
 673                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 674                         return -EINVAL;
 675
 676                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 677                         return -EINVAL;
 678
 679                 break;
 680
 681         case OVS_KEY_ATTR_IPV6:
 682                 if (flow_key->eth.type != htons(ETH_P_IPV6))
 683                         return -EINVAL;
 684
 685                 if (!flow_key->ip.proto)
 686                         return -EINVAL;
 687
 688                 ipv6_key = nla_data(ovs_key);
 689                 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
 690                         return -EINVAL;
 691
 692                 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
 693                         return -EINVAL;
 694
 695                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
 696                         return -EINVAL;
 697
 698                 break;
 699
 700         case OVS_KEY_ATTR_TCP:
 701                 if (flow_key->ip.proto != IPPROTO_TCP)
 702                         return -EINVAL;
 703
 704                 return validate_tp_port(flow_key);
 705
 706         case OVS_KEY_ATTR_UDP:
 707                 if (flow_key->ip.proto != IPPROTO_UDP)
 708                         return -EINVAL;
 709
 710                 return validate_tp_port(flow_key);
 711
 712         default:
 713                 return -EINVAL;
 714         }
 715
 716         return 0;
 717 }
 718
 719 static int validate_userspace(const struct nlattr *attr)
 720 {
 721         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 722                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 723                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
 724         };
 725         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 726         int error;
 727
 728         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 729                                  attr, userspace_policy);
 730         if (error)
 731                 return error;
 732
 733         if (!a[OVS_USERSPACE_ATTR_PID] ||
 734             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 735                 return -EINVAL;
 736
 737         return 0;
 738 }
 739
 740 static int copy_action(const struct nlattr *from,
 741                        struct sw_flow_actions **sfa)
 742 {
 743         int totlen = NLA_ALIGN(from->nla_len);
 744         struct nlattr *to;
 745
 746         to = reserve_sfa_size(sfa, from->nla_len);
 747         if (IS_ERR(to))
 748                 return PTR_ERR(to);
 749
 750         memcpy(to, from, totlen);
 751         return 0;
 752 }
 753
 754 static int validate_and_copy_actions(const struct nlattr *attr,
 755                                      const struct sw_flow_key *key,
 756                                      int depth,
 757                                      struct sw_flow_actions **sfa)
 758 {
 759         const struct nlattr *a;
 760         int rem, err;
 761
 762         if (depth >= SAMPLE_ACTION_DEPTH)
 763                 return -EOVERFLOW;
 764
 765         nla_for_each_nested(a, attr, rem) {
 766                 /* Expected argument lengths, (u32)-1 for variable length. */
 767                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 768                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 769                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 770                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 771                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 772                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 773                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 774                 };
 775                 const struct ovs_action_push_vlan *vlan;
 776                 int type = nla_type(a);
 777                 bool skip_copy;
 778
 779                 if (type > OVS_ACTION_ATTR_MAX ||
 780                     (action_lens[type] != nla_len(a) &&
 781                      action_lens[type] != (u32)-1))
 782                         return -EINVAL;
 783
 784                 skip_copy = false;
 785                 switch (type) {
 786                 case OVS_ACTION_ATTR_UNSPEC:
 787                         return -EINVAL;
 788
 789                 case OVS_ACTION_ATTR_USERSPACE:
 790                         err = validate_userspace(a);
 791                         if (err)
 792                                 return err;
 793                         break;
 794
 795                 case OVS_ACTION_ATTR_OUTPUT:
 796                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 797                                 return -EINVAL;
 798                         break;
 799
 800
 801                 case OVS_ACTION_ATTR_POP_VLAN:
 802                         break;
 803
 804                 case OVS_ACTION_ATTR_PUSH_VLAN:
 805                         vlan = nla_data(a);
 806                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 807                                 return -EINVAL;
 808                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 809                                 return -EINVAL;
 810                         break;
 811
 812                 case OVS_ACTION_ATTR_SET:
 813                         err = validate_set(a, key, sfa, &skip_copy);
 814                         if (err)
 815                                 return err;
 816                         break;
 817
 818                 case OVS_ACTION_ATTR_SAMPLE:
 819                         err = validate_and_copy_sample(a, key, depth, sfa);
 820                         if (err)
 821                                 return err;
 822                         skip_copy = true;
 823                         break;
 824
 825                 default:
 826                         return -EINVAL;
 827                 }
 828                 if (!skip_copy) {
 829                         err = copy_action(a, sfa);
 830                         if (err)
 831                                 return err;
 832                 }
 833         }
 834
 835         if (rem > 0)
 836                 return -EINVAL;
 837
 838         return 0;
 839 }
 840
 841 static void clear_stats(struct sw_flow *flow)
 842 {
 843         flow->used = 0;
 844         flow->tcp_flags = 0;
 845         flow->packet_count = 0;
 846         flow->byte_count = 0;
 847 }
 848
 849 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 850 {
 851         struct ovs_header *ovs_header = info->userhdr;
 852         struct nlattr **a = info->attrs;
 853         struct sw_flow_actions *acts;
 854         struct sk_buff *packet;
 855         struct sw_flow *flow;
 856         struct datapath *dp;
 857         struct ethhdr *eth;
 858         int len;
 859         int err;
 860         int key_len;
 861
 862         err = -EINVAL;
 863         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 864             !a[OVS_PACKET_ATTR_ACTIONS])
 865                 goto err;
 866
 867         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 868         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 869         err = -ENOMEM;
 870         if (!packet)
 871                 goto err;
 872         skb_reserve(packet, NET_IP_ALIGN);
 873
 874         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 875
 876         skb_reset_mac_header(packet);
 877         eth = eth_hdr(packet);
 878
 879         /* Normally, setting the skb 'protocol' field would be handled by a
 880          * call to eth_type_trans(), but it assumes there's a sending
 881          * device, which we may not have. */
 882         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 883                 packet->protocol = eth->h_proto;
 884         else
 885                 packet->protocol = htons(ETH_P_802_2);
 886
 887         /* Build an sw_flow for sending this packet. */
 888         flow = ovs_flow_alloc();
 889         err = PTR_ERR(flow);
 890         if (IS_ERR(flow))
 891                 goto err_kfree_skb;
 892
 893         err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
 894         if (err)
 895                 goto err_flow_free;
 896
 897         err = ovs_flow_metadata_from_nlattrs(flow, key_len, a[OVS_PACKET_ATTR_KEY]);
 898         if (err)
 899                 goto err_flow_free;
 900         acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 901         err = PTR_ERR(acts);
 902         if (IS_ERR(acts))
 903                 goto err_flow_free;
 904
 905         err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
 906         rcu_assign_pointer(flow->sf_acts, acts);
 907         if (err)
 908                 goto err_flow_free;
 909
 910         OVS_CB(packet)->flow = flow;
 911         packet->priority = flow->key.phy.priority;
 912         packet->mark = flow->key.phy.skb_mark;
 913
 914         rcu_read_lock();
 915         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 916         err = -ENODEV;
 917         if (!dp)
 918                 goto err_unlock;
 919
 920         local_bh_disable();
 921         err = ovs_execute_actions(dp, packet);
 922         local_bh_enable();
 923         rcu_read_unlock();
 924
 925         ovs_flow_free(flow);
 926         return err;
 927
 928 err_unlock:
 929         rcu_read_unlock();
 930 err_flow_free:
 931         ovs_flow_free(flow);
 932 err_kfree_skb:
 933         kfree_skb(packet);
 934 err:
 935         return err;
 936 }
 937
 938 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 939         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 940         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 941         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 942 };
 943
 944 static struct genl_ops dp_packet_genl_ops[] = {
 945         { .cmd = OVS_PACKET_CMD_EXECUTE,
 946           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 947           .policy = packet_policy,
 948           .doit = ovs_packet_cmd_execute
 949         }
 950 };
 951
 952 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 953 {
 954         int i;
 955         struct flow_table *table = ovsl_dereference(dp->table);
 956
 957         stats->n_flows = ovs_flow_tbl_count(table);
 958
 959         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 960         for_each_possible_cpu(i) {
 961                 const struct dp_stats_percpu *percpu_stats;
 962                 struct dp_stats_percpu local_stats;
 963                 unsigned int start;
 964
 965                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 966
 967                 do {
 968                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 969                         local_stats = *percpu_stats;
 970                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 971
 972                 stats->n_hit += local_stats.n_hit;
 973                 stats->n_missed += local_stats.n_missed;
 974                 stats->n_lost += local_stats.n_lost;
 975         }
 976 }
 977
 978 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 979         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 980         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 981         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 982 };
 983
 984 static struct genl_family dp_flow_genl_family = {
 985         .id = GENL_ID_GENERATE,
 986         .hdrsize = sizeof(struct ovs_header),
 987         .name = OVS_FLOW_FAMILY,
 988         .version = OVS_FLOW_VERSION,
 989         .maxattr = OVS_FLOW_ATTR_MAX,
 990         .netnsok = true,
 991         .parallel_ops = true,
 992 };
 993
 994 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 995         .name = OVS_FLOW_MCGROUP
 996 };
 997
 998 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
 999 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1000 {
1001         const struct nlattr *a;
1002         struct nlattr *start;
1003         int err = 0, rem;
1004
1005         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1006         if (!start)
1007                 return -EMSGSIZE;
1008
1009         nla_for_each_nested(a, attr, rem) {
1010                 int type = nla_type(a);
1011                 struct nlattr *st_sample;
1012
1013                 switch (type) {
1014                 case OVS_SAMPLE_ATTR_PROBABILITY:
1015                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
1016                                 return -EMSGSIZE;
1017                         break;
1018                 case OVS_SAMPLE_ATTR_ACTIONS:
1019                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1020                         if (!st_sample)
1021                                 return -EMSGSIZE;
1022                         err = actions_to_attr(nla_data(a), nla_len(a), skb);
1023                         if (err)
1024                                 return err;
1025                         nla_nest_end(skb, st_sample);
1026                         break;
1027                 }
1028         }
1029
1030         nla_nest_end(skb, start);
1031         return err;
1032 }
1033
1034 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1035 {
1036         const struct nlattr *ovs_key = nla_data(a);
1037         int key_type = nla_type(ovs_key);
1038         struct nlattr *start;
1039         int err;
1040
1041         switch (key_type) {
1042         case OVS_KEY_ATTR_IPV4_TUNNEL:
1043                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1044                 if (!start)
1045                         return -EMSGSIZE;
1046
1047                 err = ovs_ipv4_tun_to_nlattr(skb, nla_data(ovs_key));
1048                 if (err)
1049                         return err;
1050                 nla_nest_end(skb, start);
1051                 break;
1052         default:
1053                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1054                         return -EMSGSIZE;
1055                 break;
1056         }
1057
1058         return 0;
1059 }
1060
1061 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1062 {
1063         const struct nlattr *a;
1064         int rem, err;
1065
1066         nla_for_each_attr(a, attr, len, rem) {
1067                 int type = nla_type(a);
1068
1069                 switch (type) {
1070                 case OVS_ACTION_ATTR_SET:
1071                         err = set_action_to_attr(a, skb);
1072                         if (err)
1073                                 return err;
1074                         break;
1075
1076                 case OVS_ACTION_ATTR_SAMPLE:
1077                         err = sample_action_to_attr(a, skb);
1078                         if (err)
1079                                 return err;
1080                         break;
1081                 default:
1082                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
1083                                 return -EMSGSIZE;
1084                         break;
1085                 }
1086         }
1087
1088         return 0;
1089 }
1090
1091 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
1092 {
1093         return NLMSG_ALIGN(sizeof(struct ovs_header))
1094                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
1095                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
1096                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
1097                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
1098                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
1099 }
1100
1101 /* Called with ovs_mutex. */
1102 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1103                                   struct sk_buff *skb, u32 portid,
1104                                   u32 seq, u32 flags, u8 cmd)
1105 {
1106         const int skb_orig_len = skb->len;
1107         const struct sw_flow_actions *sf_acts;
1108         struct nlattr *start;
1109         struct ovs_flow_stats stats;
1110         struct ovs_header *ovs_header;
1111         struct nlattr *nla;
1112         unsigned long used;
1113         u8 tcp_flags;
1114         int err;
1115
1116         sf_acts = ovsl_dereference(flow->sf_acts);
1117
1118         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1119         if (!ovs_header)
1120                 return -EMSGSIZE;
1121
1122         ovs_header->dp_ifindex = get_dpifindex(dp);
1123
1124         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1125         if (!nla)
1126                 goto nla_put_failure;
1127         err = ovs_flow_to_nlattrs(&flow->key, skb);
1128         if (err)
1129                 goto error;
1130         nla_nest_end(skb, nla);
1131
1132         spin_lock_bh(&flow->lock);
1133         used = flow->used;
1134         stats.n_packets = flow->packet_count;
1135         stats.n_bytes = flow->byte_count;
1136         tcp_flags = flow->tcp_flags;
1137         spin_unlock_bh(&flow->lock);
1138
1139         if (used &&
1140             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1141                 goto nla_put_failure;
1142
1143         if (stats.n_packets &&
1144             nla_put(skb, OVS_FLOW_ATTR_STATS,
1145                     sizeof(struct ovs_flow_stats), &stats))
1146                 goto nla_put_failure;
1147
1148         if (tcp_flags &&
1149             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1150                 goto nla_put_failure;
1151
1152         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1153          * this is the first flow to be dumped into 'skb'.  This is unusual for
1154          * Netlink but individual action lists can be longer than
1155          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1156          * The userspace caller can always fetch the actions separately if it
1157          * really wants them.  (Most userspace callers in fact don't care.)
1158          *
1159          * This can only fail for dump operations because the skb is always
1160          * properly sized for single flows.
1161          */
1162         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1163         if (start) {
1164                 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1165                 if (!err)
1166                         nla_nest_end(skb, start);
1167                 else {
1168                         if (skb_orig_len)
1169                                 goto error;
1170
1171                         nla_nest_cancel(skb, start);
1172                 }
1173         } else if (skb_orig_len)
1174                 goto nla_put_failure;
1175
1176         return genlmsg_end(skb, ovs_header);
1177
1178 nla_put_failure:
1179         err = -EMSGSIZE;
1180 error:
1181         genlmsg_cancel(skb, ovs_header);
1182         return err;
1183 }
1184
1185 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1186 {
1187         const struct sw_flow_actions *sf_acts;
1188
1189         sf_acts = ovsl_dereference(flow->sf_acts);
1190
1191         return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
1192 }
1193
1194 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1195                                                struct datapath *dp,
1196                                                u32 portid, u32 seq, u8 cmd)
1197 {
1198         struct sk_buff *skb;
1199         int retval;
1200
1201         skb = ovs_flow_cmd_alloc_info(flow);
1202         if (!skb)
1203                 return ERR_PTR(-ENOMEM);
1204
1205         retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1206         BUG_ON(retval < 0);
1207         return skb;
1208 }
1209
1210 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1211 {
1212         struct nlattr **a = info->attrs;
1213         struct ovs_header *ovs_header = info->userhdr;
1214         struct sw_flow_key key;
1215         struct sw_flow *flow;
1216         struct sk_buff *reply;
1217         struct datapath *dp;
1218         struct flow_table *table;
1219         struct sw_flow_actions *acts = NULL;
1220         int error;
1221         int key_len;
1222
1223         /* Extract key. */
1224         error = -EINVAL;
1225         if (!a[OVS_FLOW_ATTR_KEY])
1226                 goto error;
1227         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1228         if (error)
1229                 goto error;
1230
1231         /* Validate actions. */
1232         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1233                 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1234                 error = PTR_ERR(acts);
1235                 if (IS_ERR(acts))
1236                         goto error;
1237
1238                 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0, &acts);
1239                 if (error)
1240                         goto err_kfree;
1241         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1242                 error = -EINVAL;
1243                 goto error;
1244         }
1245
1246         ovs_lock();
1247         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1248         error = -ENODEV;
1249         if (!dp)
1250                 goto err_unlock_ovs;
1251
1252         table = ovsl_dereference(dp->table);
1253         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1254         if (!flow) {
1255                 /* Bail out if we're not allowed to create a new flow. */
1256                 error = -ENOENT;
1257                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1258                         goto err_unlock_ovs;
1259
1260                 /* Expand table, if necessary, to make room. */
1261                 if (ovs_flow_tbl_need_to_expand(table)) {
1262                         struct flow_table *new_table;
1263
1264                         new_table = ovs_flow_tbl_expand(table);
1265                         if (!IS_ERR(new_table)) {
1266                                 rcu_assign_pointer(dp->table, new_table);
1267                                 ovs_flow_tbl_deferred_destroy(table);
1268                                 table = ovsl_dereference(dp->table);
1269                         }
1270                 }
1271
1272                 /* Allocate flow. */
1273                 flow = ovs_flow_alloc();
1274                 if (IS_ERR(flow)) {
1275                         error = PTR_ERR(flow);
1276                         goto err_unlock_ovs;
1277                 }
1278                 clear_stats(flow);
1279
1280                 rcu_assign_pointer(flow->sf_acts, acts);
1281
1282                 /* Put flow in bucket. */
1283                 ovs_flow_tbl_insert(table, flow, &key, key_len);
1284
1285                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1286                                                 info->snd_seq,
1287                                                 OVS_FLOW_CMD_NEW);
1288         } else {
1289                 /* We found a matching flow. */
1290                 struct sw_flow_actions *old_acts;
1291
1292                 /* Bail out if we're not allowed to modify an existing flow.
1293                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1294                  * because Generic Netlink treats the latter as a dump
1295                  * request.  We also accept NLM_F_EXCL in case that bug ever
1296                  * gets fixed.
1297                  */
1298                 error = -EEXIST;
1299                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1300                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1301                         goto err_unlock_ovs;
1302
1303                 /* Update actions. */
1304                 old_acts = ovsl_dereference(flow->sf_acts);
1305                 rcu_assign_pointer(flow->sf_acts, acts);
1306                 ovs_flow_deferred_free_acts(old_acts);
1307
1308                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1309                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1310
1311                 /* Clear stats. */
1312                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1313                         spin_lock_bh(&flow->lock);
1314                         clear_stats(flow);
1315                         spin_unlock_bh(&flow->lock);
1316                 }
1317         }
1318         ovs_unlock();
1319
1320         if (!IS_ERR(reply))
1321                 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1322         else
1323                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1324                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1325         return 0;
1326
1327 err_unlock_ovs:
1328         ovs_unlock();
1329 err_kfree:
1330         kfree(acts);
1331 error:
1332         return error;
1333 }
1334
1335 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1336 {
1337         struct nlattr **a = info->attrs;
1338         struct ovs_header *ovs_header = info->userhdr;
1339         struct sw_flow_key key;
1340         struct sk_buff *reply;
1341         struct sw_flow *flow;
1342         struct datapath *dp;
1343         struct flow_table *table;
1344         int err;
1345         int key_len;
1346
1347         if (!a[OVS_FLOW_ATTR_KEY])
1348                 return -EINVAL;
1349         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1350         if (err)
1351                 return err;
1352
1353         ovs_lock();
1354         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1355         if (!dp) {
1356                 err = -ENODEV;
1357                 goto unlock;
1358         }
1359
1360         table = ovsl_dereference(dp->table);
1361         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1362         if (!flow) {
1363                 err = -ENOENT;
1364                 goto unlock;
1365         }
1366
1367         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1368                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1369         if (IS_ERR(reply)) {
1370                 err = PTR_ERR(reply);
1371                 goto unlock;
1372         }
1373
1374         ovs_unlock();
1375         return genlmsg_reply(reply, info);
1376 unlock:
1377         ovs_unlock();
1378         return err;
1379 }
1380
1381 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1382 {
1383         struct nlattr **a = info->attrs;
1384         struct ovs_header *ovs_header = info->userhdr;
1385         struct sw_flow_key key;
1386         struct sk_buff *reply;
1387         struct sw_flow *flow;
1388         struct datapath *dp;
1389         struct flow_table *table;
1390         int err;
1391         int key_len;
1392
1393         ovs_lock();
1394         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1395         if (!dp) {
1396                 err = -ENODEV;
1397                 goto unlock;
1398         }
1399
1400         if (!a[OVS_FLOW_ATTR_KEY]) {
1401                 err = flush_flows(dp);
1402                 goto unlock;
1403         }
1404         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1405         if (err)
1406                 goto unlock;
1407
1408         table = ovsl_dereference(dp->table);
1409         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1410         if (!flow) {
1411                 err = -ENOENT;
1412                 goto unlock;
1413         }
1414
1415         reply = ovs_flow_cmd_alloc_info(flow);
1416         if (!reply) {
1417                 err = -ENOMEM;
1418                 goto unlock;
1419         }
1420
1421         ovs_flow_tbl_remove(table, flow);
1422
1423         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1424                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1425         BUG_ON(err < 0);
1426
1427         ovs_flow_deferred_free(flow);
1428         ovs_unlock();
1429
1430         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1431         return 0;
1432 unlock:
1433         ovs_unlock();
1434         return err;
1435 }
1436
1437 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1438 {
1439         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1440         struct datapath *dp;
1441         struct flow_table *table;
1442
1443         ovs_lock();
1444         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1445         if (!dp) {
1446                 ovs_unlock();
1447                 return -ENODEV;
1448         }
1449
1450         table = ovsl_dereference(dp->table);
1451
1452         for (;;) {
1453                 struct sw_flow *flow;
1454                 u32 bucket, obj;
1455
1456                 bucket = cb->args[0];
1457                 obj = cb->args[1];
1458                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1459                 if (!flow)
1460                         break;
1461
1462                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1463                                            NETLINK_CB(cb->skb).portid,
1464                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1465                                            OVS_FLOW_CMD_NEW) < 0)
1466                         break;
1467
1468                 cb->args[0] = bucket;
1469                 cb->args[1] = obj;
1470         }
1471         ovs_unlock();
1472         return skb->len;
1473 }
1474
1475 static struct genl_ops dp_flow_genl_ops[] = {
1476         { .cmd = OVS_FLOW_CMD_NEW,
1477           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1478           .policy = flow_policy,
1479           .doit = ovs_flow_cmd_new_or_set
1480         },
1481         { .cmd = OVS_FLOW_CMD_DEL,
1482           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1483           .policy = flow_policy,
1484           .doit = ovs_flow_cmd_del
1485         },
1486         { .cmd = OVS_FLOW_CMD_GET,
1487           .flags = 0,               /* OK for unprivileged users. */
1488           .policy = flow_policy,
1489           .doit = ovs_flow_cmd_get,
1490           .dumpit = ovs_flow_cmd_dump
1491         },
1492         { .cmd = OVS_FLOW_CMD_SET,
1493           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1494           .policy = flow_policy,
1495           .doit = ovs_flow_cmd_new_or_set,
1496         },
1497 };
1498
1499 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1500         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1501         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1502 };
1503
1504 static struct genl_family dp_datapath_genl_family = {
1505         .id = GENL_ID_GENERATE,
1506         .hdrsize = sizeof(struct ovs_header),
1507         .name = OVS_DATAPATH_FAMILY,
1508         .version = OVS_DATAPATH_VERSION,
1509         .maxattr = OVS_DP_ATTR_MAX,
1510         .netnsok = true,
1511         .parallel_ops = true,
1512 };
1513
1514 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1515         .name = OVS_DATAPATH_MCGROUP
1516 };
1517
1518 static size_t ovs_dp_cmd_msg_size(void)
1519 {
1520         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1521
1522         msgsize += nla_total_size(IFNAMSIZ);
1523         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1524
1525         return msgsize;
1526 }
1527
1528 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1529                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1530 {
1531         struct ovs_header *ovs_header;
1532         struct ovs_dp_stats dp_stats;
1533         int err;
1534
1535         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1536                                    flags, cmd);
1537         if (!ovs_header)
1538                 goto error;
1539
1540         ovs_header->dp_ifindex = get_dpifindex(dp);
1541
1542         rcu_read_lock();
1543         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1544         rcu_read_unlock();
1545         if (err)
1546                 goto nla_put_failure;
1547
1548         get_dp_stats(dp, &dp_stats);
1549         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1550                 goto nla_put_failure;
1551
1552         return genlmsg_end(skb, ovs_header);
1553
1554 nla_put_failure:
1555         genlmsg_cancel(skb, ovs_header);
1556 error:
1557         return -EMSGSIZE;
1558 }
1559
1560 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1561                                              u32 seq, u8 cmd)
1562 {
1563         struct sk_buff *skb;
1564         int retval;
1565
1566         skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1567         if (!skb)
1568                 return ERR_PTR(-ENOMEM);
1569
1570         retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1571         if (retval < 0) {
1572                 kfree_skb(skb);
1573                 return ERR_PTR(retval);
1574         }
1575         return skb;
1576 }
1577
1578 /* Called with ovs_mutex. */
1579 static struct datapath *lookup_datapath(struct net *net,
1580                                         struct ovs_header *ovs_header,
1581                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1582 {
1583         struct datapath *dp;
1584
1585         if (!a[OVS_DP_ATTR_NAME])
1586                 dp = get_dp(net, ovs_header->dp_ifindex);
1587         else {
1588                 struct vport *vport;
1589
1590                 rcu_read_lock();
1591                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1592                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1593                 rcu_read_unlock();
1594         }
1595         return dp ? dp : ERR_PTR(-ENODEV);
1596 }
1597
1598 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1599 {
1600         struct nlattr **a = info->attrs;
1601         struct vport_parms parms;
1602         struct sk_buff *reply;
1603         struct datapath *dp;
1604         struct vport *vport;
1605         struct ovs_net *ovs_net;
1606         int err, i;
1607
1608         err = -EINVAL;
1609         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1610                 goto err;
1611
1612         ovs_lock();
1613
1614         err = -ENOMEM;
1615         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1616         if (dp == NULL)
1617                 goto err_unlock_ovs;
1618
1619         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1620
1621         /* Allocate table. */
1622         err = -ENOMEM;
1623         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1624         if (!dp->table)
1625                 goto err_free_dp;
1626
1627         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1628         if (!dp->stats_percpu) {
1629                 err = -ENOMEM;
1630                 goto err_destroy_table;
1631         }
1632
1633         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1634                         GFP_KERNEL);
1635         if (!dp->ports) {
1636                 err = -ENOMEM;
1637                 goto err_destroy_percpu;
1638         }
1639
1640         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1641                 INIT_HLIST_HEAD(&dp->ports[i]);
1642
1643         /* Set up our datapath device. */
1644         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1645         parms.type = OVS_VPORT_TYPE_INTERNAL;
1646         parms.options = NULL;
1647         parms.dp = dp;
1648         parms.port_no = OVSP_LOCAL;
1649         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1650
1651         vport = new_vport(&parms);
1652         if (IS_ERR(vport)) {
1653                 err = PTR_ERR(vport);
1654                 if (err == -EBUSY)
1655                         err = -EEXIST;
1656
1657                 goto err_destroy_ports_array;
1658         }
1659
1660         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1661                                       info->snd_seq, OVS_DP_CMD_NEW);
1662         err = PTR_ERR(reply);
1663         if (IS_ERR(reply))
1664                 goto err_destroy_local_port;
1665
1666         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1667         list_add_tail(&dp->list_node, &ovs_net->dps);
1668
1669         ovs_unlock();
1670
1671         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1672         return 0;
1673
1674 err_destroy_local_port:
1675         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1676 err_destroy_ports_array:
1677         kfree(dp->ports);
1678 err_destroy_percpu:
1679         free_percpu(dp->stats_percpu);
1680 err_destroy_table:
1681         ovs_flow_tbl_destroy(ovsl_dereference(dp->table));
1682 err_free_dp:
1683         release_net(ovs_dp_get_net(dp));
1684         kfree(dp);
1685 err_unlock_ovs:
1686         ovs_unlock();
1687 err:
1688         return err;
1689 }
1690
1691 /* Called with ovs_mutex. */
1692 static void __dp_destroy(struct datapath *dp)
1693 {
1694         int i;
1695
1696         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1697                 struct vport *vport;
1698                 struct hlist_node *n;
1699
1700                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1701                         if (vport->port_no != OVSP_LOCAL)
1702                                 ovs_dp_detach_port(vport);
1703         }
1704
1705         list_del(&dp->list_node);
1706
1707         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1708          * all port in datapath are destroyed first before freeing datapath.
1709          */
1710         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1711
1712         call_rcu(&dp->rcu, destroy_dp_rcu);
1713 }
1714
1715 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1716 {
1717         struct sk_buff *reply;
1718         struct datapath *dp;
1719         int err;
1720
1721         ovs_lock();
1722         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1723         err = PTR_ERR(dp);
1724         if (IS_ERR(dp))
1725                 goto unlock;
1726
1727         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1728                                       info->snd_seq, OVS_DP_CMD_DEL);
1729         err = PTR_ERR(reply);
1730         if (IS_ERR(reply))
1731                 goto unlock;
1732
1733         __dp_destroy(dp);
1734         ovs_unlock();
1735
1736         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1737
1738         return 0;
1739 unlock:
1740         ovs_unlock();
1741         return err;
1742 }
1743
1744 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1745 {
1746         struct sk_buff *reply;
1747         struct datapath *dp;
1748         int err;
1749
1750         ovs_lock();
1751         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1752         err = PTR_ERR(dp);
1753         if (IS_ERR(dp))
1754                 goto unlock;
1755
1756         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1757                                       info->snd_seq, OVS_DP_CMD_NEW);
1758         if (IS_ERR(reply)) {
1759                 err = PTR_ERR(reply);
1760                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1761                                 ovs_dp_datapath_multicast_group.id, err);
1762                 err = 0;
1763                 goto unlock;
1764         }
1765
1766         ovs_unlock();
1767         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1768
1769         return 0;
1770 unlock:
1771         ovs_unlock();
1772         return err;
1773 }
1774
1775 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1776 {
1777         struct sk_buff *reply;
1778         struct datapath *dp;
1779         int err;
1780
1781         ovs_lock();
1782         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1783         if (IS_ERR(dp)) {
1784                 err = PTR_ERR(dp);
1785                 goto unlock;
1786         }
1787
1788         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1789                                       info->snd_seq, OVS_DP_CMD_NEW);
1790         if (IS_ERR(reply)) {
1791                 err = PTR_ERR(reply);
1792                 goto unlock;
1793         }
1794
1795         ovs_unlock();
1796         return genlmsg_reply(reply, info);
1797
1798 unlock:
1799         ovs_unlock();
1800         return err;
1801 }
1802
1803 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1804 {
1805         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1806         struct datapath *dp;
1807         int skip = cb->args[0];
1808         int i = 0;
1809
1810         ovs_lock();
1811         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1812                 if (i >= skip &&
1813                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1814                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1815                                          OVS_DP_CMD_NEW) < 0)
1816                         break;
1817                 i++;
1818         }
1819         ovs_unlock();
1820
1821         cb->args[0] = i;
1822
1823         return skb->len;
1824 }
1825
1826 static struct genl_ops dp_datapath_genl_ops[] = {
1827         { .cmd = OVS_DP_CMD_NEW,
1828           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1829           .policy = datapath_policy,
1830           .doit = ovs_dp_cmd_new
1831         },
1832         { .cmd = OVS_DP_CMD_DEL,
1833           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1834           .policy = datapath_policy,
1835           .doit = ovs_dp_cmd_del
1836         },
1837         { .cmd = OVS_DP_CMD_GET,
1838           .flags = 0,               /* OK for unprivileged users. */
1839           .policy = datapath_policy,
1840           .doit = ovs_dp_cmd_get,
1841           .dumpit = ovs_dp_cmd_dump
1842         },
1843         { .cmd = OVS_DP_CMD_SET,
1844           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1845           .policy = datapath_policy,
1846           .doit = ovs_dp_cmd_set,
1847         },
1848 };
1849
1850 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1851         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1852         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1853         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1854         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1855         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1856         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1857 };
1858
1859 static struct genl_family dp_vport_genl_family = {
1860         .id = GENL_ID_GENERATE,
1861         .hdrsize = sizeof(struct ovs_header),
1862         .name = OVS_VPORT_FAMILY,
1863         .version = OVS_VPORT_VERSION,
1864         .maxattr = OVS_VPORT_ATTR_MAX,
1865         .netnsok = true,
1866         .parallel_ops = true,
1867 };
1868
1869 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1870         .name = OVS_VPORT_MCGROUP
1871 };
1872
1873 /* Called with ovs_mutex or RCU read lock. */
1874 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1875                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1876 {
1877         struct ovs_header *ovs_header;
1878         struct ovs_vport_stats vport_stats;
1879         int err;
1880
1881         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1882                                  flags, cmd);
1883         if (!ovs_header)
1884                 return -EMSGSIZE;
1885
1886         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1887
1888         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1889             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1890             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1891             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1892                 goto nla_put_failure;
1893
1894         ovs_vport_get_stats(vport, &vport_stats);
1895         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1896                     &vport_stats))
1897                 goto nla_put_failure;
1898
1899         err = ovs_vport_get_options(vport, skb);
1900         if (err == -EMSGSIZE)
1901                 goto error;
1902
1903         return genlmsg_end(skb, ovs_header);
1904
1905 nla_put_failure:
1906         err = -EMSGSIZE;
1907 error:
1908         genlmsg_cancel(skb, ovs_header);
1909         return err;
1910 }
1911
1912 /* Called with ovs_mutex or RCU read lock. */
1913 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1914                                          u32 seq, u8 cmd)
1915 {
1916         struct sk_buff *skb;
1917         int retval;
1918
1919         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1920         if (!skb)
1921                 return ERR_PTR(-ENOMEM);
1922
1923         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1924         BUG_ON(retval < 0);
1925
1926         return skb;
1927 }
1928
1929 /* Called with ovs_mutex or RCU read lock. */
1930 static struct vport *lookup_vport(struct net *net,
1931                                   struct ovs_header *ovs_header,
1932                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1933 {
1934         struct datapath *dp;
1935         struct vport *vport;
1936
1937         if (a[OVS_VPORT_ATTR_NAME]) {
1938                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1939                 if (!vport)
1940                         return ERR_PTR(-ENODEV);
1941                 if (ovs_header->dp_ifindex &&
1942                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1943                         return ERR_PTR(-ENODEV);
1944                 return vport;
1945         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1946                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1947
1948                 if (port_no >= DP_MAX_PORTS)
1949                         return ERR_PTR(-EFBIG);
1950
1951                 dp = get_dp(net, ovs_header->dp_ifindex);
1952                 if (!dp)
1953                         return ERR_PTR(-ENODEV);
1954
1955                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1956                 if (!vport)
1957                         return ERR_PTR(-ENODEV);
1958                 return vport;
1959         } else
1960                 return ERR_PTR(-EINVAL);
1961 }
1962
1963 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1964 {
1965         struct nlattr **a = info->attrs;
1966         struct ovs_header *ovs_header = info->userhdr;
1967         struct vport_parms parms;
1968         struct sk_buff *reply;
1969         struct vport *vport;
1970         struct datapath *dp;
1971         u32 port_no;
1972         int err;
1973
1974         err = -EINVAL;
1975         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1976             !a[OVS_VPORT_ATTR_UPCALL_PID])
1977                 goto exit;
1978
1979         ovs_lock();
1980         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1981         err = -ENODEV;
1982         if (!dp)
1983                 goto exit_unlock;
1984
1985         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1986                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1987
1988                 err = -EFBIG;
1989                 if (port_no >= DP_MAX_PORTS)
1990                         goto exit_unlock;
1991
1992                 vport = ovs_vport_ovsl(dp, port_no);
1993                 err = -EBUSY;
1994                 if (vport)
1995                         goto exit_unlock;
1996         } else {
1997                 for (port_no = 1; ; port_no++) {
1998                         if (port_no >= DP_MAX_PORTS) {
1999                                 err = -EFBIG;
2000                                 goto exit_unlock;
2001                         }
2002                         vport = ovs_vport_ovsl(dp, port_no);
2003                         if (!vport)
2004                                 break;
2005                 }
2006         }
2007
2008         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2009         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2010         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2011         parms.dp = dp;
2012         parms.port_no = port_no;
2013         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2014
2015         vport = new_vport(&parms);
2016         err = PTR_ERR(vport);
2017         if (IS_ERR(vport))
2018                 goto exit_unlock;
2019
2020         err = 0;
2021         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
2022                                          OVS_VPORT_CMD_NEW);
2023         if (IS_ERR(reply)) {
2024                 err = PTR_ERR(reply);
2025                 ovs_dp_detach_port(vport);
2026                 goto exit_unlock;
2027         }
2028
2029         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2030
2031 exit_unlock:
2032         ovs_unlock();
2033 exit:
2034         return err;
2035 }
2036
2037 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2038 {
2039         struct nlattr **a = info->attrs;
2040         struct sk_buff *reply;
2041         struct vport *vport;
2042         int err;
2043
2044         ovs_lock();
2045         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2046         err = PTR_ERR(vport);
2047         if (IS_ERR(vport))
2048                 goto exit_unlock;
2049
2050         if (a[OVS_VPORT_ATTR_TYPE] &&
2051             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2052                 err = -EINVAL;
2053                 goto exit_unlock;
2054         }
2055
2056         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2057         if (!reply) {
2058                 err = -ENOMEM;
2059                 goto exit_unlock;
2060         }
2061
2062         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2063                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2064                 if (err)
2065                         goto exit_free;
2066         }
2067
2068         if (a[OVS_VPORT_ATTR_UPCALL_PID])
2069                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2070
2071         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2072                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2073         BUG_ON(err < 0);
2074
2075         ovs_unlock();
2076         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2077         return 0;
2078
2079 exit_free:
2080         kfree_skb(reply);
2081 exit_unlock:
2082         ovs_unlock();
2083         return err;
2084 }
2085
2086 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2087 {
2088         struct nlattr **a = info->attrs;
2089         struct sk_buff *reply;
2090         struct vport *vport;
2091         int err;
2092
2093         ovs_lock();
2094         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2095         err = PTR_ERR(vport);
2096         if (IS_ERR(vport))
2097                 goto exit_unlock;
2098
2099         if (vport->port_no == OVSP_LOCAL) {
2100                 err = -EINVAL;
2101                 goto exit_unlock;
2102         }
2103
2104         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2105                                          info->snd_seq, OVS_VPORT_CMD_DEL);
2106         err = PTR_ERR(reply);
2107         if (IS_ERR(reply))
2108                 goto exit_unlock;
2109
2110         err = 0;
2111         ovs_dp_detach_port(vport);
2112
2113         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2114
2115 exit_unlock:
2116         ovs_unlock();
2117         return err;
2118 }
2119
2120 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2121 {
2122         struct nlattr **a = info->attrs;
2123         struct ovs_header *ovs_header = info->userhdr;
2124         struct sk_buff *reply;
2125         struct vport *vport;
2126         int err;
2127
2128         rcu_read_lock();
2129         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2130         err = PTR_ERR(vport);
2131         if (IS_ERR(vport))
2132                 goto exit_unlock;
2133
2134         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2135                                          info->snd_seq, OVS_VPORT_CMD_NEW);
2136         err = PTR_ERR(reply);
2137         if (IS_ERR(reply))
2138                 goto exit_unlock;
2139
2140         rcu_read_unlock();
2141
2142         return genlmsg_reply(reply, info);
2143
2144 exit_unlock:
2145         rcu_read_unlock();
2146         return err;
2147 }
2148
2149 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2150 {
2151         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2152         struct datapath *dp;
2153         int bucket = cb->args[0], skip = cb->args[1];
2154         int i, j = 0;
2155
2156         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2157         if (!dp)
2158                 return -ENODEV;
2159
2160         rcu_read_lock();
2161         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2162                 struct vport *vport;
2163
2164                 j = 0;
2165                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2166                         if (j >= skip &&
2167                             ovs_vport_cmd_fill_info(vport, skb,
2168                                                     NETLINK_CB(cb->skb).portid,
2169                                                     cb->nlh->nlmsg_seq,
2170                                                     NLM_F_MULTI,
2171                                                     OVS_VPORT_CMD_NEW) < 0)
2172                                 goto out;
2173
2174                         j++;
2175                 }
2176                 skip = 0;
2177         }
2178 out:
2179         rcu_read_unlock();
2180
2181         cb->args[0] = i;
2182         cb->args[1] = j;
2183
2184         return skb->len;
2185 }
2186
2187 static struct genl_ops dp_vport_genl_ops[] = {
2188         { .cmd = OVS_VPORT_CMD_NEW,
2189           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2190           .policy = vport_policy,
2191           .doit = ovs_vport_cmd_new
2192         },
2193         { .cmd = OVS_VPORT_CMD_DEL,
2194           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2195           .policy = vport_policy,
2196           .doit = ovs_vport_cmd_del
2197         },
2198         { .cmd = OVS_VPORT_CMD_GET,
2199           .flags = 0,               /* OK for unprivileged users. */
2200           .policy = vport_policy,
2201           .doit = ovs_vport_cmd_get,
2202           .dumpit = ovs_vport_cmd_dump
2203         },
2204         { .cmd = OVS_VPORT_CMD_SET,
2205           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2206           .policy = vport_policy,
2207           .doit = ovs_vport_cmd_set,
2208         },
2209 };
2210
2211 struct genl_family_and_ops {
2212         struct genl_family *family;
2213         struct genl_ops *ops;
2214         int n_ops;
2215         struct genl_multicast_group *group;
2216 };
2217
2218 static const struct genl_family_and_ops dp_genl_families[] = {
2219         { &dp_datapath_genl_family,
2220           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2221           &ovs_dp_datapath_multicast_group },
2222         { &dp_vport_genl_family,
2223           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2224           &ovs_dp_vport_multicast_group },
2225         { &dp_flow_genl_family,
2226           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2227           &ovs_dp_flow_multicast_group },
2228         { &dp_packet_genl_family,
2229           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2230           NULL },
2231 };
2232
2233 static void dp_unregister_genl(int n_families)
2234 {
2235         int i;
2236
2237         for (i = 0; i < n_families; i++)
2238                 genl_unregister_family(dp_genl_families[i].family);
2239 }
2240
2241 static int dp_register_genl(void)
2242 {
2243         int n_registered;
2244         int err;
2245         int i;
2246
2247         n_registered = 0;
2248         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2249                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2250
2251                 err = genl_register_family_with_ops(f->family, f->ops,
2252                                                     f->n_ops);
2253                 if (err)
2254                         goto error;
2255                 n_registered++;
2256
2257                 if (f->group) {
2258                         err = genl_register_mc_group(f->family, f->group);
2259                         if (err)
2260                                 goto error;
2261                 }
2262         }
2263
2264         return 0;
2265
2266 error:
2267         dp_unregister_genl(n_registered);
2268         return err;
2269 }
2270
2271 static void rehash_flow_table(struct work_struct *work)
2272 {
2273         struct datapath *dp;
2274         struct net *net;
2275
2276         ovs_lock();
2277         rtnl_lock();
2278         for_each_net(net) {
2279                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2280
2281                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2282                         struct flow_table *old_table = ovsl_dereference(dp->table);
2283                         struct flow_table *new_table;
2284
2285                         new_table = ovs_flow_tbl_rehash(old_table);
2286                         if (!IS_ERR(new_table)) {
2287                                 rcu_assign_pointer(dp->table, new_table);
2288                                 ovs_flow_tbl_deferred_destroy(old_table);
2289                         }
2290                 }
2291         }
2292         rtnl_unlock();
2293         ovs_unlock();
2294         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2295 }
2296
2297 static int __net_init ovs_init_net(struct net *net)
2298 {
2299         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2300
2301         INIT_LIST_HEAD(&ovs_net->dps);
2302         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2303         return 0;
2304 }
2305
2306 static void __net_exit ovs_exit_net(struct net *net)
2307 {
2308         struct datapath *dp, *dp_next;
2309         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2310
2311         ovs_lock();
2312         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2313                 __dp_destroy(dp);
2314         ovs_unlock();
2315
2316         cancel_work_sync(&ovs_net->dp_notify_work);
2317 }
2318
2319 static struct pernet_operations ovs_net_ops = {
2320         .init = ovs_init_net,
2321         .exit = ovs_exit_net,
2322         .id   = &ovs_net_id,
2323         .size = sizeof(struct ovs_net),
2324 };
2325
2326 static int __init dp_init(void)
2327 {
2328         int err;
2329
2330         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2331
2332         pr_info("Open vSwitch switching datapath\n");
2333
2334         err = ovs_flow_init();
2335         if (err)
2336                 goto error;
2337
2338         err = ovs_vport_init();
2339         if (err)
2340                 goto error_flow_exit;
2341
2342         err = register_pernet_device(&ovs_net_ops);
2343         if (err)
2344                 goto error_vport_exit;
2345
2346         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2347         if (err)
2348                 goto error_netns_exit;
2349
2350         err = dp_register_genl();
2351         if (err < 0)
2352                 goto error_unreg_notifier;
2353
2354         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2355
2356         return 0;
2357
2358 error_unreg_notifier:
2359         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2360 error_netns_exit:
2361         unregister_pernet_device(&ovs_net_ops);
2362 error_vport_exit:
2363         ovs_vport_exit();
2364 error_flow_exit:
2365         ovs_flow_exit();
2366 error:
2367         return err;
2368 }
2369
2370 static void dp_cleanup(void)
2371 {
2372         cancel_delayed_work_sync(&rehash_flow_wq);
2373         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2374         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2375         unregister_pernet_device(&ovs_net_ops);
2376         rcu_barrier();
2377         ovs_vport_exit();
2378         ovs_flow_exit();
2379 }
2380
2381 module_init(dp_init);
2382 module_exit(dp_cleanup);
2383
2384 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2385 MODULE_LICENSE("GPL");