datapath/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/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 "checksum.h"
  56 #include "datapath.h"
  57 #include "flow.h"
  58 #include "genl_exec.h"
  59 #include "vlan.h"
  60 #include "tunnel.h"
  61 #include "vport-internal_dev.h"
  62
  63 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
  64     LINUX_VERSION_CODE >= KERNEL_VERSION(3,9,0)
  65 #error Kernels before 2.6.18 or after 3.8 are not supported by this version of Open vSwitch.
  66 #endif
  67
  68 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
  69 static void rehash_flow_table(struct work_struct *work);
  70 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
  71
  72 int ovs_net_id __read_mostly;
  73
  74 /**
  75  * DOC: Locking:
  76  *
  77  * Writes to device state (add/remove datapath, port, set operations on vports,
  78  * etc.) are protected by RTNL.
  79  *
  80  * Writes to other state (flow table modifications, set miscellaneous datapath
  81  * parameters, etc.) are protected by genl_mutex.  The RTNL lock nests inside
  82  * genl_mutex.
  83  *
  84  * Reads are protected by RCU.
  85  *
  86  * There are a few special cases (mostly stats) that have their own
  87  * synchronization but they nest under all of above and don't interact with
  88  * each other.
  89  */
  90
  91 static struct vport *new_vport(const struct vport_parms *);
  92 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
  93                              const struct dp_upcall_info *);
  94 static int queue_userspace_packet(struct net *, int dp_ifindex,
  95                                   struct sk_buff *,
  96                                   const struct dp_upcall_info *);
  97
  98 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  99 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 100 {
 101         struct datapath *dp = NULL;
 102         struct net_device *dev;
 103
 104         rcu_read_lock();
 105         dev = dev_get_by_index_rcu(net, dp_ifindex);
 106         if (dev) {
 107                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 108                 if (vport)
 109                         dp = vport->dp;
 110         }
 111         rcu_read_unlock();
 112
 113         return dp;
 114 }
 115
 116 /* Must be called with rcu_read_lock or RTNL lock. */
 117 const char *ovs_dp_name(const struct datapath *dp)
 118 {
 119         struct vport *vport = ovs_vport_rtnl_rcu(dp, OVSP_LOCAL);
 120         return vport->ops->get_name(vport);
 121 }
 122
 123 static int get_dpifindex(struct datapath *dp)
 124 {
 125         struct vport *local;
 126         int ifindex;
 127
 128         rcu_read_lock();
 129
 130         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 131         if (local)
 132                 ifindex = local->ops->get_ifindex(local);
 133         else
 134                 ifindex = 0;
 135
 136         rcu_read_unlock();
 137
 138         return ifindex;
 139 }
 140
 141 static void destroy_dp_rcu(struct rcu_head *rcu)
 142 {
 143         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 144
 145         ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
 146         free_percpu(dp->stats_percpu);
 147         release_net(ovs_dp_get_net(dp));
 148         kfree(dp->ports);
 149         kfree(dp);
 150 }
 151
 152 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 153                                             u16 port_no)
 154 {
 155         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 156 }
 157
 158 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 159 {
 160         struct vport *vport;
 161         struct hlist_node *n;
 162         struct hlist_head *head;
 163
 164         head = vport_hash_bucket(dp, port_no);
 165         hlist_for_each_entry_rcu(vport, n, head, dp_hash_node) {
 166                 if (vport->port_no == port_no)
 167                         return vport;
 168         }
 169         return NULL;
 170 }
 171
 172 /* Called with RTNL lock and genl_lock. */
 173 static struct vport *new_vport(const struct vport_parms *parms)
 174 {
 175         struct vport *vport;
 176
 177         vport = ovs_vport_add(parms);
 178         if (!IS_ERR(vport)) {
 179                 struct datapath *dp = parms->dp;
 180                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 181
 182                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 183         }
 184         return vport;
 185 }
 186
 187 /* Called with RTNL lock. */
 188 void ovs_dp_detach_port(struct vport *p)
 189 {
 190         ASSERT_RTNL();
 191
 192         /* First drop references to device. */
 193         hlist_del_rcu(&p->dp_hash_node);
 194
 195         /* Then destroy it. */
 196         ovs_vport_del(p);
 197 }
 198
 199 /* Must be called with rcu_read_lock. */
 200 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 201 {
 202         struct datapath *dp = p->dp;
 203         struct sw_flow *flow;
 204         struct dp_stats_percpu *stats;
 205         u64 *stats_counter;
 206         int error;
 207
 208         stats = this_cpu_ptr(dp->stats_percpu);
 209
 210         if (!OVS_CB(skb)->flow) {
 211                 struct sw_flow_key key;
 212                 int key_len;
 213
 214                 /* Extract flow from 'skb' into 'key'. */
 215                 error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
 216                 if (unlikely(error)) {
 217                         kfree_skb(skb);
 218                         return;
 219                 }
 220
 221                 /* Look up flow. */
 222                 flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table),
 223                                            &key, key_len);
 224                 if (unlikely(!flow)) {
 225                         struct dp_upcall_info upcall;
 226
 227                         upcall.cmd = OVS_PACKET_CMD_MISS;
 228                         upcall.key = &key;
 229                         upcall.userdata = NULL;
 230                         upcall.portid = p->upcall_portid;
 231                         ovs_dp_upcall(dp, skb, &upcall);
 232                         consume_skb(skb);
 233                         stats_counter = &stats->n_missed;
 234                         goto out;
 235                 }
 236
 237                 OVS_CB(skb)->flow = flow;
 238         }
 239
 240         stats_counter = &stats->n_hit;
 241         ovs_flow_used(OVS_CB(skb)->flow, skb);
 242         ovs_execute_actions(dp, skb);
 243
 244 out:
 245         /* Update datapath statistics. */
 246         u64_stats_update_begin(&stats->sync);
 247         (*stats_counter)++;
 248         u64_stats_update_end(&stats->sync);
 249 }
 250
 251 static struct genl_family dp_packet_genl_family = {
 252         .id = GENL_ID_GENERATE,
 253         .hdrsize = sizeof(struct ovs_header),
 254         .name = OVS_PACKET_FAMILY,
 255         .version = OVS_PACKET_VERSION,
 256         .maxattr = OVS_PACKET_ATTR_MAX,
 257          SET_NETNSOK
 258 };
 259
 260 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 261                   const struct dp_upcall_info *upcall_info)
 262 {
 263         struct dp_stats_percpu *stats;
 264         int dp_ifindex;
 265         int err;
 266
 267         if (upcall_info->portid == 0) {
 268                 err = -ENOTCONN;
 269                 goto err;
 270         }
 271
 272         dp_ifindex = get_dpifindex(dp);
 273         if (!dp_ifindex) {
 274                 err = -ENODEV;
 275                 goto err;
 276         }
 277
 278         forward_ip_summed(skb, true);
 279
 280         if (!skb_is_gso(skb))
 281                 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 282         else
 283                 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 284         if (err)
 285                 goto err;
 286
 287         return 0;
 288
 289 err:
 290         stats = this_cpu_ptr(dp->stats_percpu);
 291
 292         u64_stats_update_begin(&stats->sync);
 293         stats->n_lost++;
 294         u64_stats_update_end(&stats->sync);
 295
 296         return err;
 297 }
 298
 299 static int queue_gso_packets(struct net *net, int dp_ifindex,
 300                              struct sk_buff *skb,
 301                              const struct dp_upcall_info *upcall_info)
 302 {
 303         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 304         struct dp_upcall_info later_info;
 305         struct sw_flow_key later_key;
 306         struct sk_buff *segs, *nskb;
 307         int err;
 308
 309         segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 310         if (IS_ERR(segs))
 311                 return PTR_ERR(segs);
 312         if (!segs)
 313                 return queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
 314
 315         /* Queue all of the segments. */
 316         skb = segs;
 317         do {
 318                 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
 319                 if (err)
 320                         break;
 321
 322                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 323                         /* The initial flow key extracted by ovs_flow_extract()
 324                          * in this case is for a first fragment, so we need to
 325                          * properly mark later fragments.
 326                          */
 327                         later_key = *upcall_info->key;
 328                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 329
 330                         later_info = *upcall_info;
 331                         later_info.key = &later_key;
 332                         upcall_info = &later_info;
 333                 }
 334         } while ((skb = skb->next));
 335
 336         /* Free all of the segments. */
 337         skb = segs;
 338         do {
 339                 nskb = skb->next;
 340                 if (err)
 341                         kfree_skb(skb);
 342                 else
 343                         consume_skb(skb);
 344         } while ((skb = nskb));
 345         return err;
 346 }
 347
 348 static int queue_userspace_packet(struct net *net, int dp_ifindex,
 349                                   struct sk_buff *skb,
 350                                   const struct dp_upcall_info *upcall_info)
 351 {
 352         struct ovs_header *upcall;
 353         struct sk_buff *nskb = NULL;
 354         struct sk_buff *user_skb; /* to be queued to userspace */
 355         struct nlattr *nla;
 356         unsigned int len;
 357         int err;
 358
 359         if (vlan_tx_tag_present(skb)) {
 360                 nskb = skb_clone(skb, GFP_ATOMIC);
 361                 if (!nskb)
 362                         return -ENOMEM;
 363
 364                 err = vlan_deaccel_tag(nskb);
 365                 if (err)
 366                         return err;
 367
 368                 skb = nskb;
 369         }
 370
 371         if (nla_attr_size(skb->len) > USHRT_MAX) {
 372                 err = -EFBIG;
 373                 goto out;
 374         }
 375
 376         len = sizeof(struct ovs_header);
 377         len += nla_total_size(skb->len);
 378         len += nla_total_size(FLOW_BUFSIZE);
 379         if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
 380                 len += nla_total_size(8);
 381
 382         user_skb = genlmsg_new(len, GFP_ATOMIC);
 383         if (!user_skb) {
 384                 err = -ENOMEM;
 385                 goto out;
 386         }
 387
 388         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 389                              0, upcall_info->cmd);
 390         upcall->dp_ifindex = dp_ifindex;
 391
 392         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 393         ovs_flow_to_nlattrs(upcall_info->key, user_skb);
 394         nla_nest_end(user_skb, nla);
 395
 396         if (upcall_info->userdata)
 397                 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
 398                             nla_get_u64(upcall_info->userdata));
 399
 400         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 401
 402         skb_copy_and_csum_dev(skb, nla_data(nla));
 403
 404         err = genlmsg_unicast(net, user_skb, upcall_info->portid);
 405
 406 out:
 407         kfree_skb(nskb);
 408         return err;
 409 }
 410
 411 /* Called with genl_mutex. */
 412 static int flush_flows(struct datapath *dp)
 413 {
 414         struct flow_table *old_table;
 415         struct flow_table *new_table;
 416
 417         old_table = genl_dereference(dp->table);
 418         new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
 419         if (!new_table)
 420                 return -ENOMEM;
 421
 422         rcu_assign_pointer(dp->table, new_table);
 423
 424         ovs_flow_tbl_deferred_destroy(old_table);
 425         return 0;
 426 }
 427
 428 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa, int attr_len)
 429 {
 430
 431         struct sw_flow_actions *acts;
 432         int new_acts_size;
 433         int req_size = NLA_ALIGN(attr_len);
 434         int next_offset = offsetof(struct sw_flow_actions, actions) +
 435                                         (*sfa)->actions_len;
 436
 437         if (req_size <= (ksize(*sfa) - next_offset))
 438                 goto out;
 439
 440         new_acts_size = ksize(*sfa) * 2;
 441
 442         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
 443                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
 444                         return ERR_PTR(-EMSGSIZE);
 445                 new_acts_size = MAX_ACTIONS_BUFSIZE;
 446         }
 447
 448         acts = ovs_flow_actions_alloc(new_acts_size);
 449         if (IS_ERR(acts))
 450                 return (void *)acts;
 451
 452         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
 453         acts->actions_len = (*sfa)->actions_len;
 454         kfree(*sfa);
 455         *sfa = acts;
 456
 457 out:
 458         (*sfa)->actions_len += req_size;
 459         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
 460 }
 461
 462 static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, int len)
 463 {
 464         struct nlattr *a;
 465
 466         a = reserve_sfa_size(sfa, nla_attr_size(len));
 467         if (IS_ERR(a))
 468                 return PTR_ERR(a);
 469
 470         a->nla_type = attrtype;
 471         a->nla_len = nla_attr_size(len);
 472
 473         if (data)
 474                 memcpy(nla_data(a), data, len);
 475         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
 476
 477         return 0;
 478 }
 479
 480 static inline int add_nested_action_start(struct sw_flow_actions **sfa, int attrtype)
 481 {
 482         int used = (*sfa)->actions_len;
 483         int err;
 484
 485         err = add_action(sfa, attrtype, NULL, 0);
 486         if (err)
 487                 return err;
 488
 489         return used;
 490 }
 491
 492 static inline void add_nested_action_end(struct sw_flow_actions *sfa, int st_offset)
 493 {
 494         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions + st_offset);
 495
 496         a->nla_len = sfa->actions_len - st_offset;
 497 }
 498
 499 static int validate_and_copy_actions(const struct nlattr *attr,
 500                                 const struct sw_flow_key *key, int depth,
 501                                 struct sw_flow_actions **sfa);
 502
 503 static int validate_and_copy_sample(const struct nlattr *attr,
 504                            const struct sw_flow_key *key, int depth,
 505                            struct sw_flow_actions **sfa)
 506 {
 507         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 508         const struct nlattr *probability, *actions;
 509         const struct nlattr *a;
 510         int rem, start, err, st_acts;
 511
 512         memset(attrs, 0, sizeof(attrs));
 513         nla_for_each_nested(a, attr, rem) {
 514                 int type = nla_type(a);
 515                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 516                         return -EINVAL;
 517                 attrs[type] = a;
 518         }
 519         if (rem)
 520                 return -EINVAL;
 521
 522         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 523         if (!probability || nla_len(probability) != sizeof(u32))
 524                 return -EINVAL;
 525
 526         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 527         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 528                 return -EINVAL;
 529
 530         /* validation done, copy sample action. */
 531         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
 532         if (start < 0)
 533                 return start;
 534         err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY, nla_data(probability), sizeof(u32));
 535         if (err)
 536                 return err;
 537         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
 538         if (st_acts < 0)
 539                 return st_acts;
 540
 541         err = validate_and_copy_actions(actions, key, depth + 1, sfa);
 542         if (err)
 543                 return err;
 544
 545         add_nested_action_end(*sfa, st_acts);
 546         add_nested_action_end(*sfa, start);
 547
 548         return 0;
 549 }
 550
 551 static int validate_tp_port(const struct sw_flow_key *flow_key)
 552 {
 553         if (flow_key->eth.type == htons(ETH_P_IP)) {
 554                 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
 555                         return 0;
 556         } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
 557                 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
 558                         return 0;
 559         }
 560
 561         return -EINVAL;
 562 }
 563
 564 static int validate_and_copy_set_tun(const struct nlattr *attr,
 565                                      struct sw_flow_actions **sfa)
 566 {
 567         struct ovs_key_ipv4_tunnel tun_key;
 568         int err, start;
 569
 570         err = ipv4_tun_from_nlattr(nla_data(attr), &tun_key);
 571         if (err)
 572                 return err;
 573
 574         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
 575         if (start < 0)
 576                 return start;
 577
 578         err = add_action(sfa, OVS_KEY_ATTR_IPV4_TUNNEL, &tun_key, sizeof(tun_key));
 579         add_nested_action_end(*sfa, start);
 580
 581         return err;
 582 }
 583
 584 static int validate_set(const struct nlattr *a,
 585                         const struct sw_flow_key *flow_key,
 586                         struct sw_flow_actions **sfa,
 587                         bool *set_tun)
 588 {
 589         const struct nlattr *ovs_key = nla_data(a);
 590         int key_type = nla_type(ovs_key);
 591
 592         /* There can be only one key in a action */
 593         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 594                 return -EINVAL;
 595
 596         if (key_type > OVS_KEY_ATTR_MAX ||
 597             (ovs_key_lens[key_type] != nla_len(ovs_key) &&
 598              ovs_key_lens[key_type] != -1))
 599                 return -EINVAL;
 600
 601         switch (key_type) {
 602         const struct ovs_key_ipv4 *ipv4_key;
 603         const struct ovs_key_ipv6 *ipv6_key;
 604         int err;
 605
 606         case OVS_KEY_ATTR_PRIORITY:
 607         case OVS_KEY_ATTR_TUN_ID:
 608         case OVS_KEY_ATTR_ETHERNET:
 609                 break;
 610
 611         case OVS_KEY_ATTR_SKB_MARK:
 612 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
 613                 if (nla_get_u32(ovs_key) != 0)
 614                         return -EINVAL;
 615 #endif
 616                 break;
 617
 618         case OVS_KEY_ATTR_TUNNEL:
 619                 *set_tun = true;
 620                 err = validate_and_copy_set_tun(a, sfa);
 621                 if (err)
 622                         return err;
 623                 break;
 624
 625         case OVS_KEY_ATTR_IPV4:
 626                 if (flow_key->eth.type != htons(ETH_P_IP))
 627                         return -EINVAL;
 628
 629                 if (!flow_key->ip.proto)
 630                         return -EINVAL;
 631
 632                 ipv4_key = nla_data(ovs_key);
 633                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 634                         return -EINVAL;
 635
 636                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 637                         return -EINVAL;
 638
 639                 break;
 640
 641         case OVS_KEY_ATTR_IPV6:
 642                 if (flow_key->eth.type != htons(ETH_P_IPV6))
 643                         return -EINVAL;
 644
 645                 if (!flow_key->ip.proto)
 646                         return -EINVAL;
 647
 648                 ipv6_key = nla_data(ovs_key);
 649                 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
 650                         return -EINVAL;
 651
 652                 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
 653                         return -EINVAL;
 654
 655                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
 656                         return -EINVAL;
 657
 658                 break;
 659
 660         case OVS_KEY_ATTR_TCP:
 661                 if (flow_key->ip.proto != IPPROTO_TCP)
 662                         return -EINVAL;
 663
 664                 return validate_tp_port(flow_key);
 665
 666         case OVS_KEY_ATTR_UDP:
 667                 if (flow_key->ip.proto != IPPROTO_UDP)
 668                         return -EINVAL;
 669
 670                 return validate_tp_port(flow_key);
 671
 672         default:
 673                 return -EINVAL;
 674         }
 675
 676         return 0;
 677 }
 678
 679 static int validate_userspace(const struct nlattr *attr)
 680 {
 681         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 682                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 683                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
 684         };
 685         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 686         int error;
 687
 688         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 689                                  attr, userspace_policy);
 690         if (error)
 691                 return error;
 692
 693         if (!a[OVS_USERSPACE_ATTR_PID] ||
 694             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 695                 return -EINVAL;
 696
 697         return 0;
 698 }
 699
 700 static int copy_action(const struct nlattr *from,
 701                       struct sw_flow_actions **sfa)
 702 {
 703         int totlen = NLA_ALIGN(from->nla_len);
 704         struct nlattr *to;
 705
 706         to = reserve_sfa_size(sfa, from->nla_len);
 707         if (IS_ERR(to))
 708                 return PTR_ERR(to);
 709
 710         memcpy(to, from, totlen);
 711         return 0;
 712 }
 713
 714 static int validate_and_copy_actions(const struct nlattr *attr,
 715                                 const struct sw_flow_key *key,
 716                                 int depth,
 717                                 struct sw_flow_actions **sfa)
 718 {
 719         const struct nlattr *a;
 720         int rem, err;
 721
 722         if (depth >= SAMPLE_ACTION_DEPTH)
 723                 return -EOVERFLOW;
 724
 725         nla_for_each_nested(a, attr, rem) {
 726                 /* Expected argument lengths, (u32)-1 for variable length. */
 727                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 728                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 729                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 730                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 731                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 732                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 733                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 734                 };
 735                 const struct ovs_action_push_vlan *vlan;
 736                 int type = nla_type(a);
 737                 bool skip_copy;
 738
 739                 if (type > OVS_ACTION_ATTR_MAX ||
 740                     (action_lens[type] != nla_len(a) &&
 741                      action_lens[type] != (u32)-1))
 742                         return -EINVAL;
 743
 744                 skip_copy = false;
 745                 switch (type) {
 746                 case OVS_ACTION_ATTR_UNSPEC:
 747                         return -EINVAL;
 748
 749                 case OVS_ACTION_ATTR_USERSPACE:
 750                         err = validate_userspace(a);
 751                         if (err)
 752                                 return err;
 753                         break;
 754
 755                 case OVS_ACTION_ATTR_OUTPUT:
 756                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 757                                 return -EINVAL;
 758                         break;
 759
 760
 761                 case OVS_ACTION_ATTR_POP_VLAN:
 762                         break;
 763
 764                 case OVS_ACTION_ATTR_PUSH_VLAN:
 765                         vlan = nla_data(a);
 766                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 767                                 return -EINVAL;
 768                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 769                                 return -EINVAL;
 770                         break;
 771
 772                 case OVS_ACTION_ATTR_SET:
 773                         err = validate_set(a, key, sfa, &skip_copy);
 774                         if (err)
 775                                 return err;
 776                         break;
 777
 778                 case OVS_ACTION_ATTR_SAMPLE:
 779                         err = validate_and_copy_sample(a, key, depth, sfa);
 780                         if (err)
 781                                 return err;
 782                         skip_copy = true;
 783                         break;
 784
 785                 default:
 786                         return -EINVAL;
 787                 }
 788                 if (!skip_copy) {
 789                         err = copy_action(a, sfa);
 790                         if (err)
 791                                 return err;
 792                 }
 793         }
 794
 795         if (rem > 0)
 796                 return -EINVAL;
 797
 798         return 0;
 799 }
 800
 801 static void clear_stats(struct sw_flow *flow)
 802 {
 803         flow->used = 0;
 804         flow->tcp_flags = 0;
 805         flow->packet_count = 0;
 806         flow->byte_count = 0;
 807 }
 808
 809 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 810 {
 811         struct ovs_header *ovs_header = info->userhdr;
 812         struct nlattr **a = info->attrs;
 813         struct sw_flow_actions *acts;
 814         struct sk_buff *packet;
 815         struct sw_flow *flow;
 816         struct datapath *dp;
 817         struct ethhdr *eth;
 818         int len;
 819         int err;
 820         int key_len;
 821
 822         err = -EINVAL;
 823         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 824             !a[OVS_PACKET_ATTR_ACTIONS] ||
 825             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 826                 goto err;
 827
 828         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 829         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 830         err = -ENOMEM;
 831         if (!packet)
 832                 goto err;
 833         skb_reserve(packet, NET_IP_ALIGN);
 834
 835         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 836
 837         skb_reset_mac_header(packet);
 838         eth = eth_hdr(packet);
 839
 840         /* Normally, setting the skb 'protocol' field would be handled by a
 841          * call to eth_type_trans(), but it assumes there's a sending
 842          * device, which we may not have. */
 843         if (ntohs(eth->h_proto) >= 1536)
 844                 packet->protocol = eth->h_proto;
 845         else
 846                 packet->protocol = htons(ETH_P_802_2);
 847
 848         /* Build an sw_flow for sending this packet. */
 849         flow = ovs_flow_alloc();
 850         err = PTR_ERR(flow);
 851         if (IS_ERR(flow))
 852                 goto err_kfree_skb;
 853
 854         err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
 855         if (err)
 856                 goto err_flow_free;
 857
 858         err = ovs_flow_metadata_from_nlattrs(flow, key_len, a[OVS_PACKET_ATTR_KEY]);
 859         if (err)
 860                 goto err_flow_free;
 861         acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 862         err = PTR_ERR(acts);
 863         if (IS_ERR(acts))
 864                 goto err_flow_free;
 865
 866         err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
 867         rcu_assign_pointer(flow->sf_acts, acts);
 868         if (err)
 869                 goto err_flow_free;
 870
 871         OVS_CB(packet)->flow = flow;
 872         packet->priority = flow->key.phy.priority;
 873         skb_set_mark(packet, flow->key.phy.skb_mark);
 874
 875         rcu_read_lock();
 876         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 877         err = -ENODEV;
 878         if (!dp)
 879                 goto err_unlock;
 880
 881         local_bh_disable();
 882         err = ovs_execute_actions(dp, packet);
 883         local_bh_enable();
 884         rcu_read_unlock();
 885
 886         ovs_flow_free(flow);
 887         return err;
 888
 889 err_unlock:
 890         rcu_read_unlock();
 891 err_flow_free:
 892         ovs_flow_free(flow);
 893 err_kfree_skb:
 894         kfree_skb(packet);
 895 err:
 896         return err;
 897 }
 898
 899 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 900         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 901         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 902         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 903 };
 904
 905 static struct genl_ops dp_packet_genl_ops[] = {
 906         { .cmd = OVS_PACKET_CMD_EXECUTE,
 907           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 908           .policy = packet_policy,
 909           .doit = ovs_packet_cmd_execute
 910         }
 911 };
 912
 913 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 914 {
 915         int i;
 916         struct flow_table *table = genl_dereference(dp->table);
 917
 918         stats->n_flows = ovs_flow_tbl_count(table);
 919
 920         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 921         for_each_possible_cpu(i) {
 922                 const struct dp_stats_percpu *percpu_stats;
 923                 struct dp_stats_percpu local_stats;
 924                 unsigned int start;
 925
 926                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 927
 928                 do {
 929                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 930                         local_stats = *percpu_stats;
 931                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 932
 933                 stats->n_hit += local_stats.n_hit;
 934                 stats->n_missed += local_stats.n_missed;
 935                 stats->n_lost += local_stats.n_lost;
 936         }
 937 }
 938
 939 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 940         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 941         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 942         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 943 };
 944
 945 static struct genl_family dp_flow_genl_family = {
 946         .id = GENL_ID_GENERATE,
 947         .hdrsize = sizeof(struct ovs_header),
 948         .name = OVS_FLOW_FAMILY,
 949         .version = OVS_FLOW_VERSION,
 950         .maxattr = OVS_FLOW_ATTR_MAX,
 951          SET_NETNSOK
 952 };
 953
 954 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 955         .name = OVS_FLOW_MCGROUP
 956 };
 957
 958 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
 959 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
 960 {
 961         const struct nlattr *a;
 962         struct nlattr *start;
 963         int err = 0, rem;
 964
 965         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
 966         if (!start)
 967                 return -EMSGSIZE;
 968
 969         nla_for_each_nested(a, attr, rem) {
 970                 int type = nla_type(a);
 971                 struct nlattr *st_sample;
 972
 973                 switch (type) {
 974                 case OVS_SAMPLE_ATTR_PROBABILITY:
 975                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
 976                                 return -EMSGSIZE;
 977                         break;
 978                 case OVS_SAMPLE_ATTR_ACTIONS:
 979                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
 980                         if (!st_sample)
 981                                 return -EMSGSIZE;
 982                         err = actions_to_attr(nla_data(a), nla_len(a), skb);
 983                         if (err)
 984                                 return err;
 985                         nla_nest_end(skb, st_sample);
 986                         break;
 987                 }
 988         }
 989
 990         nla_nest_end(skb, start);
 991         return err;
 992 }
 993
 994 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
 995 {
 996         const struct nlattr *ovs_key = nla_data(a);
 997         int key_type = nla_type(ovs_key);
 998         struct nlattr *start;
 999         int err;
1000
1001         switch (key_type) {
1002         case OVS_KEY_ATTR_IPV4_TUNNEL:
1003                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1004                 if (!start)
1005                         return -EMSGSIZE;
1006
1007                 err = ipv4_tun_to_nlattr(skb, nla_data(ovs_key));
1008                 if (err)
1009                         return err;
1010                 nla_nest_end(skb, start);
1011                 break;
1012         default:
1013                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1014                         return -EMSGSIZE;
1015                 break;
1016         }
1017
1018         return 0;
1019 }
1020
1021 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1022 {
1023         const struct nlattr *a;
1024         int rem, err;
1025
1026         nla_for_each_attr(a, attr, len, rem) {
1027                 int type = nla_type(a);
1028
1029                 switch (type) {
1030                 case OVS_ACTION_ATTR_SET:
1031                         err = set_action_to_attr(a, skb);
1032                         if (err)
1033                                 return err;
1034                         break;
1035
1036                 case OVS_ACTION_ATTR_SAMPLE:
1037                         err = sample_action_to_attr(a, skb);
1038                         if (err)
1039                                 return err;
1040                         break;
1041                 default:
1042                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
1043                                 return -EMSGSIZE;
1044                         break;
1045                 }
1046         }
1047
1048         return 0;
1049 }
1050
1051 /* Called with genl_lock. */
1052 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1053                                   struct sk_buff *skb, u32 portid,
1054                                   u32 seq, u32 flags, u8 cmd)
1055 {
1056         const int skb_orig_len = skb->len;
1057         const struct sw_flow_actions *sf_acts;
1058         struct nlattr *start;
1059         struct ovs_flow_stats stats;
1060         struct ovs_header *ovs_header;
1061         struct nlattr *nla;
1062         unsigned long used;
1063         u8 tcp_flags;
1064         int err;
1065
1066         sf_acts = rcu_dereference_protected(flow->sf_acts,
1067                                             lockdep_genl_is_held());
1068
1069         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1070         if (!ovs_header)
1071                 return -EMSGSIZE;
1072
1073         ovs_header->dp_ifindex = get_dpifindex(dp);
1074
1075         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1076         if (!nla)
1077                 goto nla_put_failure;
1078         err = ovs_flow_to_nlattrs(&flow->key, skb);
1079         if (err)
1080                 goto error;
1081         nla_nest_end(skb, nla);
1082
1083         spin_lock_bh(&flow->lock);
1084         used = flow->used;
1085         stats.n_packets = flow->packet_count;
1086         stats.n_bytes = flow->byte_count;
1087         tcp_flags = flow->tcp_flags;
1088         spin_unlock_bh(&flow->lock);
1089
1090         if (used &&
1091             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1092                 goto nla_put_failure;
1093
1094         if (stats.n_packets &&
1095             nla_put(skb, OVS_FLOW_ATTR_STATS,
1096                     sizeof(struct ovs_flow_stats), &stats))
1097                 goto nla_put_failure;
1098
1099         if (tcp_flags &&
1100             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1101                 goto nla_put_failure;
1102
1103         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1104          * this is the first flow to be dumped into 'skb'.  This is unusual for
1105          * Netlink but individual action lists can be longer than
1106          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1107          * The userspace caller can always fetch the actions separately if it
1108          * really wants them.  (Most userspace callers in fact don't care.)
1109          *
1110          * This can only fail for dump operations because the skb is always
1111          * properly sized for single flows.
1112          */
1113         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1114         if (start) {
1115                 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1116                 if (err < 0 && skb_orig_len)
1117                         goto error;
1118                 nla_nest_end(skb, start);
1119         } else if (skb_orig_len) {
1120                 err = -ENOMEM;
1121                 goto error;
1122         }
1123
1124         return genlmsg_end(skb, ovs_header);
1125
1126 nla_put_failure:
1127         err = -EMSGSIZE;
1128 error:
1129         genlmsg_cancel(skb, ovs_header);
1130         return err;
1131 }
1132
1133 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1134 {
1135         const struct sw_flow_actions *sf_acts;
1136         int len;
1137
1138         sf_acts = rcu_dereference_protected(flow->sf_acts,
1139                                             lockdep_genl_is_held());
1140
1141         /* OVS_FLOW_ATTR_KEY */
1142         len = nla_total_size(FLOW_BUFSIZE);
1143         /* OVS_FLOW_ATTR_ACTIONS */
1144         len += nla_total_size(sf_acts->actions_len);
1145         /* OVS_FLOW_ATTR_STATS */
1146         len += nla_total_size(sizeof(struct ovs_flow_stats));
1147         /* OVS_FLOW_ATTR_TCP_FLAGS */
1148         len += nla_total_size(1);
1149         /* OVS_FLOW_ATTR_USED */
1150         len += nla_total_size(8);
1151
1152         len += NLMSG_ALIGN(sizeof(struct ovs_header));
1153
1154         return genlmsg_new(len, GFP_KERNEL);
1155 }
1156
1157 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1158                                                struct datapath *dp,
1159                                                u32 portid, u32 seq, u8 cmd)
1160 {
1161         struct sk_buff *skb;
1162         int retval;
1163
1164         skb = ovs_flow_cmd_alloc_info(flow);
1165         if (!skb)
1166                 return ERR_PTR(-ENOMEM);
1167
1168         retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1169         BUG_ON(retval < 0);
1170         return skb;
1171 }
1172
1173 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1174 {
1175         struct nlattr **a = info->attrs;
1176         struct ovs_header *ovs_header = info->userhdr;
1177         struct sw_flow_key key;
1178         struct sw_flow *flow;
1179         struct sk_buff *reply;
1180         struct datapath *dp;
1181         struct flow_table *table;
1182         struct sw_flow_actions *acts = NULL;
1183         int error;
1184         int key_len;
1185
1186         /* Extract key. */
1187         error = -EINVAL;
1188         if (!a[OVS_FLOW_ATTR_KEY])
1189                 goto error;
1190         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1191         if (error)
1192                 goto error;
1193
1194         /* Validate actions. */
1195         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1196                 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1197                 error = PTR_ERR(acts);
1198                 if (IS_ERR(acts))
1199                         goto error;
1200
1201                 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0, &acts);
1202                 if (error)
1203                         goto err_kfree;
1204         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1205                 error = -EINVAL;
1206                 goto error;
1207         }
1208
1209         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1210         error = -ENODEV;
1211         if (!dp)
1212                 goto err_kfree;
1213
1214         table = genl_dereference(dp->table);
1215         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1216         if (!flow) {
1217                 /* Bail out if we're not allowed to create a new flow. */
1218                 error = -ENOENT;
1219                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1220                         goto err_kfree;
1221
1222                 /* Expand table, if necessary, to make room. */
1223                 if (ovs_flow_tbl_need_to_expand(table)) {
1224                         struct flow_table *new_table;
1225
1226                         new_table = ovs_flow_tbl_expand(table);
1227                         if (!IS_ERR(new_table)) {
1228                                 rcu_assign_pointer(dp->table, new_table);
1229                                 ovs_flow_tbl_deferred_destroy(table);
1230                                 table = genl_dereference(dp->table);
1231                         }
1232                 }
1233
1234                 /* Allocate flow. */
1235                 flow = ovs_flow_alloc();
1236                 if (IS_ERR(flow)) {
1237                         error = PTR_ERR(flow);
1238                         goto err_kfree;
1239                 }
1240                 clear_stats(flow);
1241
1242                 rcu_assign_pointer(flow->sf_acts, acts);
1243
1244                 /* Put flow in bucket. */
1245                 ovs_flow_tbl_insert(table, flow, &key, key_len);
1246
1247                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1248                                                 info->snd_seq,
1249                                                 OVS_FLOW_CMD_NEW);
1250         } else {
1251                 /* We found a matching flow. */
1252                 struct sw_flow_actions *old_acts;
1253
1254                 /* Bail out if we're not allowed to modify an existing flow.
1255                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1256                  * because Generic Netlink treats the latter as a dump
1257                  * request.  We also accept NLM_F_EXCL in case that bug ever
1258                  * gets fixed.
1259                  */
1260                 error = -EEXIST;
1261                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1262                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1263                         goto err_kfree;
1264
1265                 /* Update actions. */
1266                 old_acts = rcu_dereference_protected(flow->sf_acts,
1267                                                      lockdep_genl_is_held());
1268                 rcu_assign_pointer(flow->sf_acts, acts);
1269                 ovs_flow_deferred_free_acts(old_acts);
1270
1271                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1272                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1273
1274                 /* Clear stats. */
1275                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1276                         spin_lock_bh(&flow->lock);
1277                         clear_stats(flow);
1278                         spin_unlock_bh(&flow->lock);
1279                 }
1280         }
1281
1282         if (!IS_ERR(reply))
1283                 genl_notify(reply, genl_info_net(info), info->snd_portid,
1284                            ovs_dp_flow_multicast_group.id, info->nlhdr,
1285                            GFP_KERNEL);
1286         else
1287                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1288                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1289         return 0;
1290
1291 err_kfree:
1292         kfree(acts);
1293 error:
1294         return error;
1295 }
1296
1297 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1298 {
1299         struct nlattr **a = info->attrs;
1300         struct ovs_header *ovs_header = info->userhdr;
1301         struct sw_flow_key key;
1302         struct sk_buff *reply;
1303         struct sw_flow *flow;
1304         struct datapath *dp;
1305         struct flow_table *table;
1306         int err;
1307         int key_len;
1308
1309         if (!a[OVS_FLOW_ATTR_KEY])
1310                 return -EINVAL;
1311         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1312         if (err)
1313                 return err;
1314
1315         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1316         if (!dp)
1317                 return -ENODEV;
1318
1319         table = genl_dereference(dp->table);
1320         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1321         if (!flow)
1322                 return -ENOENT;
1323
1324         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1325                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1326         if (IS_ERR(reply))
1327                 return PTR_ERR(reply);
1328
1329         return genlmsg_reply(reply, info);
1330 }
1331
1332 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1333 {
1334         struct nlattr **a = info->attrs;
1335         struct ovs_header *ovs_header = info->userhdr;
1336         struct sw_flow_key key;
1337         struct sk_buff *reply;
1338         struct sw_flow *flow;
1339         struct datapath *dp;
1340         struct flow_table *table;
1341         int err;
1342         int key_len;
1343
1344         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1345         if (!dp)
1346                 return -ENODEV;
1347
1348         if (!a[OVS_FLOW_ATTR_KEY])
1349                 return flush_flows(dp);
1350
1351         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1352         if (err)
1353                 return err;
1354
1355         table = genl_dereference(dp->table);
1356         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1357         if (!flow)
1358                 return -ENOENT;
1359
1360         reply = ovs_flow_cmd_alloc_info(flow);
1361         if (!reply)
1362                 return -ENOMEM;
1363
1364         ovs_flow_tbl_remove(table, flow);
1365
1366         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1367                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1368         BUG_ON(err < 0);
1369
1370         ovs_flow_deferred_free(flow);
1371
1372         genl_notify(reply, genl_info_net(info), info->snd_portid,
1373                     ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1374         return 0;
1375 }
1376
1377 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1378 {
1379         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1380         struct datapath *dp;
1381         struct flow_table *table;
1382
1383         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1384         if (!dp)
1385                 return -ENODEV;
1386
1387         table = genl_dereference(dp->table);
1388
1389         for (;;) {
1390                 struct sw_flow *flow;
1391                 u32 bucket, obj;
1392
1393                 bucket = cb->args[0];
1394                 obj = cb->args[1];
1395                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1396                 if (!flow)
1397                         break;
1398
1399                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1400                                            NETLINK_CB(cb->skb).portid,
1401                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1402                                            OVS_FLOW_CMD_NEW) < 0)
1403                         break;
1404
1405                 cb->args[0] = bucket;
1406                 cb->args[1] = obj;
1407         }
1408         return skb->len;
1409 }
1410
1411 static struct genl_ops dp_flow_genl_ops[] = {
1412         { .cmd = OVS_FLOW_CMD_NEW,
1413           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1414           .policy = flow_policy,
1415           .doit = ovs_flow_cmd_new_or_set
1416         },
1417         { .cmd = OVS_FLOW_CMD_DEL,
1418           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1419           .policy = flow_policy,
1420           .doit = ovs_flow_cmd_del
1421         },
1422         { .cmd = OVS_FLOW_CMD_GET,
1423           .flags = 0,               /* OK for unprivileged users. */
1424           .policy = flow_policy,
1425           .doit = ovs_flow_cmd_get,
1426           .dumpit = ovs_flow_cmd_dump
1427         },
1428         { .cmd = OVS_FLOW_CMD_SET,
1429           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1430           .policy = flow_policy,
1431           .doit = ovs_flow_cmd_new_or_set,
1432         },
1433 };
1434
1435 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1436 #ifdef HAVE_NLA_NUL_STRING
1437         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1438 #endif
1439         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1440 };
1441
1442 static struct genl_family dp_datapath_genl_family = {
1443         .id = GENL_ID_GENERATE,
1444         .hdrsize = sizeof(struct ovs_header),
1445         .name = OVS_DATAPATH_FAMILY,
1446         .version = OVS_DATAPATH_VERSION,
1447         .maxattr = OVS_DP_ATTR_MAX,
1448          SET_NETNSOK
1449 };
1450
1451 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1452         .name = OVS_DATAPATH_MCGROUP
1453 };
1454
1455 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1456                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1457 {
1458         struct ovs_header *ovs_header;
1459         struct ovs_dp_stats dp_stats;
1460         int err;
1461
1462         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1463                                    flags, cmd);
1464         if (!ovs_header)
1465                 goto error;
1466
1467         ovs_header->dp_ifindex = get_dpifindex(dp);
1468
1469         rcu_read_lock();
1470         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1471         rcu_read_unlock();
1472         if (err)
1473                 goto nla_put_failure;
1474
1475         get_dp_stats(dp, &dp_stats);
1476         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1477                 goto nla_put_failure;
1478
1479         return genlmsg_end(skb, ovs_header);
1480
1481 nla_put_failure:
1482         genlmsg_cancel(skb, ovs_header);
1483 error:
1484         return -EMSGSIZE;
1485 }
1486
1487 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1488                                              u32 seq, u8 cmd)
1489 {
1490         struct sk_buff *skb;
1491         int retval;
1492
1493         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1494         if (!skb)
1495                 return ERR_PTR(-ENOMEM);
1496
1497         retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1498         if (retval < 0) {
1499                 kfree_skb(skb);
1500                 return ERR_PTR(retval);
1501         }
1502         return skb;
1503 }
1504
1505 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1506 {
1507         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1508 }
1509
1510 /* Called with genl_mutex and optionally with RTNL lock also. */
1511 static struct datapath *lookup_datapath(struct net *net,
1512                                         struct ovs_header *ovs_header,
1513                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1514 {
1515         struct datapath *dp;
1516
1517         if (!a[OVS_DP_ATTR_NAME])
1518                 dp = get_dp(net, ovs_header->dp_ifindex);
1519         else {
1520                 struct vport *vport;
1521
1522                 rcu_read_lock();
1523                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1524                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1525                 rcu_read_unlock();
1526         }
1527         return dp ? dp : ERR_PTR(-ENODEV);
1528 }
1529
1530 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1531 {
1532         struct nlattr **a = info->attrs;
1533         struct vport_parms parms;
1534         struct sk_buff *reply;
1535         struct datapath *dp;
1536         struct vport *vport;
1537         struct ovs_net *ovs_net;
1538         int err, i;
1539
1540         err = -EINVAL;
1541         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1542                 goto err;
1543
1544         err = ovs_dp_cmd_validate(a);
1545         if (err)
1546                 goto err;
1547
1548         rtnl_lock();
1549
1550         err = -ENOMEM;
1551         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1552         if (dp == NULL)
1553                 goto err_unlock_rtnl;
1554
1555         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1556
1557         /* Allocate table. */
1558         err = -ENOMEM;
1559         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1560         if (!dp->table)
1561                 goto err_free_dp;
1562
1563         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1564         if (!dp->stats_percpu) {
1565                 err = -ENOMEM;
1566                 goto err_destroy_table;
1567         }
1568
1569         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1570                             GFP_KERNEL);
1571         if (!dp->ports) {
1572                 err = -ENOMEM;
1573                 goto err_destroy_percpu;
1574         }
1575
1576         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1577                 INIT_HLIST_HEAD(&dp->ports[i]);
1578
1579         /* Set up our datapath device. */
1580         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1581         parms.type = OVS_VPORT_TYPE_INTERNAL;
1582         parms.options = NULL;
1583         parms.dp = dp;
1584         parms.port_no = OVSP_LOCAL;
1585         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1586
1587         vport = new_vport(&parms);
1588         if (IS_ERR(vport)) {
1589                 err = PTR_ERR(vport);
1590                 if (err == -EBUSY)
1591                         err = -EEXIST;
1592
1593                 goto err_destroy_ports_array;
1594         }
1595
1596         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1597                                       info->snd_seq, OVS_DP_CMD_NEW);
1598         err = PTR_ERR(reply);
1599         if (IS_ERR(reply))
1600                 goto err_destroy_local_port;
1601
1602         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1603         list_add_tail(&dp->list_node, &ovs_net->dps);
1604
1605         rtnl_unlock();
1606
1607         genl_notify(reply, genl_info_net(info), info->snd_portid,
1608                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1609                     GFP_KERNEL);
1610         return 0;
1611
1612 err_destroy_local_port:
1613         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1614 err_destroy_ports_array:
1615         kfree(dp->ports);
1616 err_destroy_percpu:
1617         free_percpu(dp->stats_percpu);
1618 err_destroy_table:
1619         ovs_flow_tbl_destroy(genl_dereference(dp->table));
1620 err_free_dp:
1621         release_net(ovs_dp_get_net(dp));
1622         kfree(dp);
1623 err_unlock_rtnl:
1624         rtnl_unlock();
1625 err:
1626         return err;
1627 }
1628
1629 /* Called with genl_mutex. */
1630 static void __dp_destroy(struct datapath *dp)
1631 {
1632         int i;
1633
1634         rtnl_lock();
1635
1636         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1637                 struct vport *vport;
1638                 struct hlist_node *node, *n;
1639
1640                 hlist_for_each_entry_safe(vport, node, n, &dp->ports[i], dp_hash_node)
1641                         if (vport->port_no != OVSP_LOCAL)
1642                                 ovs_dp_detach_port(vport);
1643         }
1644
1645         list_del(&dp->list_node);
1646         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1647
1648         /* rtnl_unlock() will wait until all the references to devices that
1649          * are pending unregistration have been dropped.  We do it here to
1650          * ensure that any internal devices (which contain DP pointers) are
1651          * fully destroyed before freeing the datapath.
1652          */
1653         rtnl_unlock();
1654
1655         call_rcu(&dp->rcu, destroy_dp_rcu);
1656 }
1657
1658 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1659 {
1660         struct sk_buff *reply;
1661         struct datapath *dp;
1662         int err;
1663
1664         err = ovs_dp_cmd_validate(info->attrs);
1665         if (err)
1666                 return err;
1667
1668         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1669         err = PTR_ERR(dp);
1670         if (IS_ERR(dp))
1671                 return err;
1672
1673         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1674                                       info->snd_seq, OVS_DP_CMD_DEL);
1675         err = PTR_ERR(reply);
1676         if (IS_ERR(reply))
1677                 return err;
1678
1679         __dp_destroy(dp);
1680
1681         genl_notify(reply, genl_info_net(info), info->snd_portid,
1682                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1683                     GFP_KERNEL);
1684
1685         return 0;
1686 }
1687
1688 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1689 {
1690         struct sk_buff *reply;
1691         struct datapath *dp;
1692         int err;
1693
1694         err = ovs_dp_cmd_validate(info->attrs);
1695         if (err)
1696                 return err;
1697
1698         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1699         if (IS_ERR(dp))
1700                 return PTR_ERR(dp);
1701
1702         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1703                                       info->snd_seq, OVS_DP_CMD_NEW);
1704         if (IS_ERR(reply)) {
1705                 err = PTR_ERR(reply);
1706                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1707                                 ovs_dp_datapath_multicast_group.id, err);
1708                 return 0;
1709         }
1710
1711         genl_notify(reply, genl_info_net(info), info->snd_portid,
1712                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1713                     GFP_KERNEL);
1714
1715         return 0;
1716 }
1717
1718 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1719 {
1720         struct sk_buff *reply;
1721         struct datapath *dp;
1722         int err;
1723
1724         err = ovs_dp_cmd_validate(info->attrs);
1725         if (err)
1726                 return err;
1727
1728         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1729         if (IS_ERR(dp))
1730                 return PTR_ERR(dp);
1731
1732         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1733                                       info->snd_seq, OVS_DP_CMD_NEW);
1734         if (IS_ERR(reply))
1735                 return PTR_ERR(reply);
1736
1737         return genlmsg_reply(reply, info);
1738 }
1739
1740 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1741 {
1742         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1743         struct datapath *dp;
1744         int skip = cb->args[0];
1745         int i = 0;
1746
1747         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1748                 if (i >= skip &&
1749                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1750                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1751                                          OVS_DP_CMD_NEW) < 0)
1752                         break;
1753                 i++;
1754         }
1755
1756         cb->args[0] = i;
1757
1758         return skb->len;
1759 }
1760
1761 static struct genl_ops dp_datapath_genl_ops[] = {
1762         { .cmd = OVS_DP_CMD_NEW,
1763           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1764           .policy = datapath_policy,
1765           .doit = ovs_dp_cmd_new
1766         },
1767         { .cmd = OVS_DP_CMD_DEL,
1768           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1769           .policy = datapath_policy,
1770           .doit = ovs_dp_cmd_del
1771         },
1772         { .cmd = OVS_DP_CMD_GET,
1773           .flags = 0,               /* OK for unprivileged users. */
1774           .policy = datapath_policy,
1775           .doit = ovs_dp_cmd_get,
1776           .dumpit = ovs_dp_cmd_dump
1777         },
1778         { .cmd = OVS_DP_CMD_SET,
1779           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1780           .policy = datapath_policy,
1781           .doit = ovs_dp_cmd_set,
1782         },
1783 };
1784
1785 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1786 #ifdef HAVE_NLA_NUL_STRING
1787         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1788         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1789         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1790 #else
1791         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1792         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1793 #endif
1794         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1795         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1796         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1797         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1798 };
1799
1800 static struct genl_family dp_vport_genl_family = {
1801         .id = GENL_ID_GENERATE,
1802         .hdrsize = sizeof(struct ovs_header),
1803         .name = OVS_VPORT_FAMILY,
1804         .version = OVS_VPORT_VERSION,
1805         .maxattr = OVS_VPORT_ATTR_MAX,
1806          SET_NETNSOK
1807 };
1808
1809 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1810         .name = OVS_VPORT_MCGROUP
1811 };
1812
1813 /* Called with RTNL lock or RCU read lock. */
1814 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1815                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1816 {
1817         struct ovs_header *ovs_header;
1818         struct ovs_vport_stats vport_stats;
1819         int err;
1820
1821         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1822                                  flags, cmd);
1823         if (!ovs_header)
1824                 return -EMSGSIZE;
1825
1826         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1827
1828         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1829             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1830             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1831             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1832                 goto nla_put_failure;
1833
1834         ovs_vport_get_stats(vport, &vport_stats);
1835         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1836                     &vport_stats))
1837                 goto nla_put_failure;
1838
1839         if (nla_put(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN,
1840                     vport->ops->get_addr(vport)))
1841                 goto nla_put_failure;
1842
1843         err = ovs_vport_get_options(vport, skb);
1844         if (err == -EMSGSIZE)
1845                 goto error;
1846
1847         return genlmsg_end(skb, ovs_header);
1848
1849 nla_put_failure:
1850         err = -EMSGSIZE;
1851 error:
1852         genlmsg_cancel(skb, ovs_header);
1853         return err;
1854 }
1855
1856 /* Called with RTNL lock or RCU read lock. */
1857 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1858                                          u32 seq, u8 cmd)
1859 {
1860         struct sk_buff *skb;
1861         int retval;
1862
1863         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1864         if (!skb)
1865                 return ERR_PTR(-ENOMEM);
1866
1867         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1868         if (retval < 0) {
1869                 kfree_skb(skb);
1870                 return ERR_PTR(retval);
1871         }
1872         return skb;
1873 }
1874
1875 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1876 {
1877         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1878 }
1879
1880 /* Called with RTNL lock or RCU read lock. */
1881 static struct vport *lookup_vport(struct net *net,
1882                                   struct ovs_header *ovs_header,
1883                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1884 {
1885         struct datapath *dp;
1886         struct vport *vport;
1887
1888         if (a[OVS_VPORT_ATTR_NAME]) {
1889                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1890                 if (!vport)
1891                         return ERR_PTR(-ENODEV);
1892                 if (ovs_header->dp_ifindex &&
1893                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1894                         return ERR_PTR(-ENODEV);
1895                 return vport;
1896         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1897                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1898
1899                 if (port_no >= DP_MAX_PORTS)
1900                         return ERR_PTR(-EFBIG);
1901
1902                 dp = get_dp(net, ovs_header->dp_ifindex);
1903                 if (!dp)
1904                         return ERR_PTR(-ENODEV);
1905
1906                 vport = ovs_vport_rtnl_rcu(dp, port_no);
1907                 if (!vport)
1908                         return ERR_PTR(-ENODEV);
1909                 return vport;
1910         } else
1911                 return ERR_PTR(-EINVAL);
1912 }
1913
1914 /* Called with RTNL lock. */
1915 static int change_vport(struct vport *vport,
1916                         struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1917 {
1918         int err = 0;
1919
1920         if (a[OVS_VPORT_ATTR_STATS])
1921                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1922
1923         if (a[OVS_VPORT_ATTR_ADDRESS])
1924                 err = ovs_vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1925
1926         return err;
1927 }
1928
1929 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1930 {
1931         struct nlattr **a = info->attrs;
1932         struct ovs_header *ovs_header = info->userhdr;
1933         struct vport_parms parms;
1934         struct sk_buff *reply;
1935         struct vport *vport;
1936         struct datapath *dp;
1937         u32 port_no;
1938         int err;
1939
1940         err = -EINVAL;
1941         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1942             !a[OVS_VPORT_ATTR_UPCALL_PID])
1943                 goto exit;
1944
1945         err = ovs_vport_cmd_validate(a);
1946         if (err)
1947                 goto exit;
1948
1949         rtnl_lock();
1950         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1951         err = -ENODEV;
1952         if (!dp)
1953                 goto exit_unlock;
1954
1955         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1956                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1957
1958                 err = -EFBIG;
1959                 if (port_no >= DP_MAX_PORTS)
1960                         goto exit_unlock;
1961
1962                 vport = ovs_vport_rtnl(dp, port_no);
1963                 err = -EBUSY;
1964                 if (vport)
1965                         goto exit_unlock;
1966         } else {
1967                 for (port_no = 1; ; port_no++) {
1968                         if (port_no >= DP_MAX_PORTS) {
1969                                 err = -EFBIG;
1970                                 goto exit_unlock;
1971                         }
1972                         vport = ovs_vport_rtnl(dp, port_no);
1973                         if (!vport)
1974                                 break;
1975                 }
1976         }
1977
1978         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1979         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1980         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1981         parms.dp = dp;
1982         parms.port_no = port_no;
1983         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1984
1985         vport = new_vport(&parms);
1986         err = PTR_ERR(vport);
1987         if (IS_ERR(vport))
1988                 goto exit_unlock;
1989
1990         err = change_vport(vport, a);
1991         if (!err) {
1992                 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1993                                                  info->snd_seq,
1994                                                  OVS_VPORT_CMD_NEW);
1995                 if (IS_ERR(reply))
1996                         err = PTR_ERR(reply);
1997         }
1998         if (err) {
1999                 ovs_dp_detach_port(vport);
2000                 goto exit_unlock;
2001         }
2002         genl_notify(reply, genl_info_net(info), info->snd_portid,
2003                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2004
2005 exit_unlock:
2006         rtnl_unlock();
2007 exit:
2008         return err;
2009 }
2010
2011 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2012 {
2013         struct nlattr **a = info->attrs;
2014         struct sk_buff *reply;
2015         struct vport *vport;
2016         int err;
2017
2018         err = ovs_vport_cmd_validate(a);
2019         if (err)
2020                 goto exit;
2021
2022         rtnl_lock();
2023         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2024         err = PTR_ERR(vport);
2025         if (IS_ERR(vport))
2026                 goto exit_unlock;
2027
2028         err = 0;
2029         if (a[OVS_VPORT_ATTR_TYPE] &&
2030             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
2031                 err = -EINVAL;
2032
2033         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
2034                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2035         if (!err)
2036                 err = change_vport(vport, a);
2037         else
2038                 goto exit_unlock;
2039         if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
2040                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2041
2042         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2043                                          info->snd_seq, OVS_VPORT_CMD_NEW);
2044         if (IS_ERR(reply)) {
2045                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
2046                                 ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
2047                 goto exit_unlock;
2048         }
2049
2050         genl_notify(reply, genl_info_net(info), info->snd_portid,
2051                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2052
2053 exit_unlock:
2054         rtnl_unlock();
2055 exit:
2056         return err;
2057 }
2058
2059 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2060 {
2061         struct nlattr **a = info->attrs;
2062         struct sk_buff *reply;
2063         struct vport *vport;
2064         int err;
2065
2066         err = ovs_vport_cmd_validate(a);
2067         if (err)
2068                 goto exit;
2069
2070         rtnl_lock();
2071         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2072         err = PTR_ERR(vport);
2073         if (IS_ERR(vport))
2074                 goto exit_unlock;
2075
2076         if (vport->port_no == OVSP_LOCAL) {
2077                 err = -EINVAL;
2078                 goto exit_unlock;
2079         }
2080
2081         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2082                                          info->snd_seq, OVS_VPORT_CMD_DEL);
2083         err = PTR_ERR(reply);
2084         if (IS_ERR(reply))
2085                 goto exit_unlock;
2086
2087         ovs_dp_detach_port(vport);
2088
2089         genl_notify(reply, genl_info_net(info), info->snd_portid,
2090                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2091
2092 exit_unlock:
2093         rtnl_unlock();
2094 exit:
2095         return err;
2096 }
2097
2098 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2099 {
2100         struct nlattr **a = info->attrs;
2101         struct ovs_header *ovs_header = info->userhdr;
2102         struct sk_buff *reply;
2103         struct vport *vport;
2104         int err;
2105
2106         err = ovs_vport_cmd_validate(a);
2107         if (err)
2108                 goto exit;
2109
2110         rcu_read_lock();
2111         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2112         err = PTR_ERR(vport);
2113         if (IS_ERR(vport))
2114                 goto exit_unlock;
2115
2116         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2117                                          info->snd_seq, OVS_VPORT_CMD_NEW);
2118         err = PTR_ERR(reply);
2119         if (IS_ERR(reply))
2120                 goto exit_unlock;
2121
2122         rcu_read_unlock();
2123
2124         return genlmsg_reply(reply, info);
2125
2126 exit_unlock:
2127         rcu_read_unlock();
2128 exit:
2129         return err;
2130 }
2131
2132 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2133 {
2134         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2135         struct datapath *dp;
2136         int bucket = cb->args[0], skip = cb->args[1];
2137         int i, j = 0;
2138
2139         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2140         if (!dp)
2141                 return -ENODEV;
2142
2143         rcu_read_lock();
2144         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2145                 struct vport *vport;
2146                 struct hlist_node *n;
2147
2148                 j = 0;
2149                 hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node) {
2150                         if (j >= skip &&
2151                             ovs_vport_cmd_fill_info(vport, skb,
2152                                                     NETLINK_CB(cb->skb).portid,
2153                                                     cb->nlh->nlmsg_seq,
2154                                                     NLM_F_MULTI,
2155                                                     OVS_VPORT_CMD_NEW) < 0)
2156                                 goto out;
2157
2158                         j++;
2159                 }
2160                 skip = 0;
2161         }
2162 out:
2163         rcu_read_unlock();
2164
2165         cb->args[0] = i;
2166         cb->args[1] = j;
2167
2168         return skb->len;
2169 }
2170
2171 static struct genl_ops dp_vport_genl_ops[] = {
2172         { .cmd = OVS_VPORT_CMD_NEW,
2173           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2174           .policy = vport_policy,
2175           .doit = ovs_vport_cmd_new
2176         },
2177         { .cmd = OVS_VPORT_CMD_DEL,
2178           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2179           .policy = vport_policy,
2180           .doit = ovs_vport_cmd_del
2181         },
2182         { .cmd = OVS_VPORT_CMD_GET,
2183           .flags = 0,               /* OK for unprivileged users. */
2184           .policy = vport_policy,
2185           .doit = ovs_vport_cmd_get,
2186           .dumpit = ovs_vport_cmd_dump
2187         },
2188         { .cmd = OVS_VPORT_CMD_SET,
2189           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2190           .policy = vport_policy,
2191           .doit = ovs_vport_cmd_set,
2192         },
2193 };
2194
2195 struct genl_family_and_ops {
2196         struct genl_family *family;
2197         struct genl_ops *ops;
2198         int n_ops;
2199         struct genl_multicast_group *group;
2200 };
2201
2202 static const struct genl_family_and_ops dp_genl_families[] = {
2203         { &dp_datapath_genl_family,
2204           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2205           &ovs_dp_datapath_multicast_group },
2206         { &dp_vport_genl_family,
2207           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2208           &ovs_dp_vport_multicast_group },
2209         { &dp_flow_genl_family,
2210           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2211           &ovs_dp_flow_multicast_group },
2212         { &dp_packet_genl_family,
2213           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2214           NULL },
2215 };
2216
2217 static void dp_unregister_genl(int n_families)
2218 {
2219         int i;
2220
2221         for (i = 0; i < n_families; i++)
2222                 genl_unregister_family(dp_genl_families[i].family);
2223 }
2224
2225 static int dp_register_genl(void)
2226 {
2227         int n_registered;
2228         int err;
2229         int i;
2230
2231         n_registered = 0;
2232         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2233                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2234
2235                 err = genl_register_family_with_ops(f->family, f->ops,
2236                                                     f->n_ops);
2237                 if (err)
2238                         goto error;
2239                 n_registered++;
2240
2241                 if (f->group) {
2242                         err = genl_register_mc_group(f->family, f->group);
2243                         if (err)
2244                                 goto error;
2245                 }
2246         }
2247
2248         return 0;
2249
2250 error:
2251         dp_unregister_genl(n_registered);
2252         return err;
2253 }
2254
2255 static int __rehash_flow_table(void *dummy)
2256 {
2257         struct datapath *dp;
2258         struct net *net;
2259
2260         rtnl_lock();
2261         for_each_net(net) {
2262                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2263
2264                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2265                         struct flow_table *old_table = genl_dereference(dp->table);
2266                         struct flow_table *new_table;
2267
2268                         new_table = ovs_flow_tbl_rehash(old_table);
2269                         if (!IS_ERR(new_table)) {
2270                                 rcu_assign_pointer(dp->table, new_table);
2271                                 ovs_flow_tbl_deferred_destroy(old_table);
2272                         }
2273                 }
2274         }
2275         rtnl_unlock();
2276         return 0;
2277 }
2278
2279 static void rehash_flow_table(struct work_struct *work)
2280 {
2281         genl_exec(__rehash_flow_table, NULL);
2282         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2283 }
2284
2285 static int dp_destroy_all(void *data)
2286 {
2287         struct datapath *dp, *dp_next;
2288         struct ovs_net *ovs_net = data;
2289
2290         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2291                 __dp_destroy(dp);
2292
2293         return 0;
2294 }
2295
2296 static int __net_init ovs_init_net(struct net *net)
2297 {
2298         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2299
2300         INIT_LIST_HEAD(&ovs_net->dps);
2301         return 0;
2302 }
2303
2304 static void __net_exit ovs_exit_net(struct net *net)
2305 {
2306         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2307
2308         genl_exec(dp_destroy_all, ovs_net);
2309 }
2310
2311 static struct pernet_operations ovs_net_ops = {
2312         .init = ovs_init_net,
2313         .exit = ovs_exit_net,
2314         .id   = &ovs_net_id,
2315         .size = sizeof(struct ovs_net),
2316 };
2317
2318 static int __init dp_init(void)
2319 {
2320         int err;
2321
2322         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2323
2324         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2325                 VERSION);
2326
2327         err = genl_exec_init();
2328         if (err)
2329                 goto error;
2330
2331         err = ovs_workqueues_init();
2332         if (err)
2333                 goto error_genl_exec;
2334
2335         err = ovs_tnl_init();
2336         if (err)
2337                 goto error_wq;
2338
2339         err = ovs_flow_init();
2340         if (err)
2341                 goto error_tnl_exit;
2342
2343         err = ovs_vport_init();
2344         if (err)
2345                 goto error_flow_exit;
2346
2347         err = register_pernet_device(&ovs_net_ops);
2348         if (err)
2349                 goto error_vport_exit;
2350
2351         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2352         if (err)
2353                 goto error_netns_exit;
2354
2355         err = dp_register_genl();
2356         if (err < 0)
2357                 goto error_unreg_notifier;
2358
2359         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2360
2361         return 0;
2362
2363 error_unreg_notifier:
2364         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2365 error_netns_exit:
2366         unregister_pernet_device(&ovs_net_ops);
2367 error_vport_exit:
2368         ovs_vport_exit();
2369 error_flow_exit:
2370         ovs_flow_exit();
2371 error_tnl_exit:
2372         ovs_tnl_exit();
2373 error_wq:
2374         ovs_workqueues_exit();
2375 error_genl_exec:
2376         genl_exec_exit();
2377 error:
2378         return err;
2379 }
2380
2381 static void dp_cleanup(void)
2382 {
2383         cancel_delayed_work_sync(&rehash_flow_wq);
2384         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2385         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2386         unregister_pernet_device(&ovs_net_ops);
2387         rcu_barrier();
2388         ovs_vport_exit();
2389         ovs_flow_exit();
2390         ovs_tnl_exit();
2391         ovs_workqueues_exit();
2392         genl_exec_exit();
2393 }
2394
2395 module_init(dp_init);
2396 module_exit(dp_cleanup);
2397
2398 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2399 MODULE_LICENSE("GPL");
2400 MODULE_VERSION(VERSION);