Pileus Git - ~andy/linux/blob - 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, a[OVS_PACKET_ATTR_KEY]);
 898         if (err)
 899                 goto err_flow_free;
 900         flow->hash = ovs_flow_hash(&flow->key, key_len);
 901         acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 902         err = PTR_ERR(acts);
 903         if (IS_ERR(acts))
 904                 goto err_flow_free;
 905
 906         err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
 907         rcu_assign_pointer(flow->sf_acts, acts);
 908         if (err)
 909                 goto err_flow_free;
 910
 911         OVS_CB(packet)->flow = flow;
 912         packet->priority = flow->key.phy.priority;
 913         packet->mark = flow->key.phy.skb_mark;
 914
 915         rcu_read_lock();
 916         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 917         err = -ENODEV;
 918         if (!dp)
 919                 goto err_unlock;
 920
 921         local_bh_disable();
 922         err = ovs_execute_actions(dp, packet);
 923         local_bh_enable();
 924         rcu_read_unlock();
 925
 926         ovs_flow_free(flow);
 927         return err;
 928
 929 err_unlock:
 930         rcu_read_unlock();
 931 err_flow_free:
 932         ovs_flow_free(flow);
 933 err_kfree_skb:
 934         kfree_skb(packet);
 935 err:
 936         return err;
 937 }
 938
 939 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 940         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 941         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 942         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 943 };
 944
 945 static struct genl_ops dp_packet_genl_ops[] = {
 946         { .cmd = OVS_PACKET_CMD_EXECUTE,
 947           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 948           .policy = packet_policy,
 949           .doit = ovs_packet_cmd_execute
 950         }
 951 };
 952
 953 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 954 {
 955         int i;
 956         struct flow_table *table = ovsl_dereference(dp->table);
 957
 958         stats->n_flows = ovs_flow_tbl_count(table);
 959
 960         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 961         for_each_possible_cpu(i) {
 962                 const struct dp_stats_percpu *percpu_stats;
 963                 struct dp_stats_percpu local_stats;
 964                 unsigned int start;
 965
 966                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 967
 968                 do {
 969                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 970                         local_stats = *percpu_stats;
 971                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 972
 973                 stats->n_hit += local_stats.n_hit;
 974                 stats->n_missed += local_stats.n_missed;
 975                 stats->n_lost += local_stats.n_lost;
 976         }
 977 }
 978
 979 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 980         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 981         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 982         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 983 };
 984
 985 static struct genl_family dp_flow_genl_family = {
 986         .id = GENL_ID_GENERATE,
 987         .hdrsize = sizeof(struct ovs_header),
 988         .name = OVS_FLOW_FAMILY,
 989         .version = OVS_FLOW_VERSION,
 990         .maxattr = OVS_FLOW_ATTR_MAX,
 991         .netnsok = true,
 992         .parallel_ops = true,
 993 };
 994
 995 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 996         .name = OVS_FLOW_MCGROUP
 997 };
 998
 999 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
1000 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1001 {
1002         const struct nlattr *a;
1003         struct nlattr *start;
1004         int err = 0, rem;
1005
1006         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1007         if (!start)
1008                 return -EMSGSIZE;
1009
1010         nla_for_each_nested(a, attr, rem) {
1011                 int type = nla_type(a);
1012                 struct nlattr *st_sample;
1013
1014                 switch (type) {
1015                 case OVS_SAMPLE_ATTR_PROBABILITY:
1016                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
1017                                 return -EMSGSIZE;
1018                         break;
1019                 case OVS_SAMPLE_ATTR_ACTIONS:
1020                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1021                         if (!st_sample)
1022                                 return -EMSGSIZE;
1023                         err = actions_to_attr(nla_data(a), nla_len(a), skb);
1024                         if (err)
1025                                 return err;
1026                         nla_nest_end(skb, st_sample);
1027                         break;
1028                 }
1029         }
1030
1031         nla_nest_end(skb, start);
1032         return err;
1033 }
1034
1035 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1036 {
1037         const struct nlattr *ovs_key = nla_data(a);
1038         int key_type = nla_type(ovs_key);
1039         struct nlattr *start;
1040         int err;
1041
1042         switch (key_type) {
1043         case OVS_KEY_ATTR_IPV4_TUNNEL:
1044                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1045                 if (!start)
1046                         return -EMSGSIZE;
1047
1048                 err = ovs_ipv4_tun_to_nlattr(skb, nla_data(ovs_key));
1049                 if (err)
1050                         return err;
1051                 nla_nest_end(skb, start);
1052                 break;
1053         default:
1054                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1055                         return -EMSGSIZE;
1056                 break;
1057         }
1058
1059         return 0;
1060 }
1061
1062 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1063 {
1064         const struct nlattr *a;
1065         int rem, err;
1066
1067         nla_for_each_attr(a, attr, len, rem) {
1068                 int type = nla_type(a);
1069
1070                 switch (type) {
1071                 case OVS_ACTION_ATTR_SET:
1072                         err = set_action_to_attr(a, skb);
1073                         if (err)
1074                                 return err;
1075                         break;
1076
1077                 case OVS_ACTION_ATTR_SAMPLE:
1078                         err = sample_action_to_attr(a, skb);
1079                         if (err)
1080                                 return err;
1081                         break;
1082                 default:
1083                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
1084                                 return -EMSGSIZE;
1085                         break;
1086                 }
1087         }
1088
1089         return 0;
1090 }
1091
1092 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
1093 {
1094         return NLMSG_ALIGN(sizeof(struct ovs_header))
1095                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
1096                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
1097                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
1098                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
1099                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
1100 }
1101
1102 /* Called with ovs_mutex. */
1103 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1104                                   struct sk_buff *skb, u32 portid,
1105                                   u32 seq, u32 flags, u8 cmd)
1106 {
1107         const int skb_orig_len = skb->len;
1108         const struct sw_flow_actions *sf_acts;
1109         struct nlattr *start;
1110         struct ovs_flow_stats stats;
1111         struct ovs_header *ovs_header;
1112         struct nlattr *nla;
1113         unsigned long used;
1114         u8 tcp_flags;
1115         int err;
1116
1117         sf_acts = ovsl_dereference(flow->sf_acts);
1118
1119         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1120         if (!ovs_header)
1121                 return -EMSGSIZE;
1122
1123         ovs_header->dp_ifindex = get_dpifindex(dp);
1124
1125         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1126         if (!nla)
1127                 goto nla_put_failure;
1128         err = ovs_flow_to_nlattrs(&flow->key, skb);
1129         if (err)
1130                 goto error;
1131         nla_nest_end(skb, nla);
1132
1133         spin_lock_bh(&flow->lock);
1134         used = flow->used;
1135         stats.n_packets = flow->packet_count;
1136         stats.n_bytes = flow->byte_count;
1137         tcp_flags = flow->tcp_flags;
1138         spin_unlock_bh(&flow->lock);
1139
1140         if (used &&
1141             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1142                 goto nla_put_failure;
1143
1144         if (stats.n_packets &&
1145             nla_put(skb, OVS_FLOW_ATTR_STATS,
1146                     sizeof(struct ovs_flow_stats), &stats))
1147                 goto nla_put_failure;
1148
1149         if (tcp_flags &&
1150             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1151                 goto nla_put_failure;
1152
1153         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1154          * this is the first flow to be dumped into 'skb'.  This is unusual for
1155          * Netlink but individual action lists can be longer than
1156          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1157          * The userspace caller can always fetch the actions separately if it
1158          * really wants them.  (Most userspace callers in fact don't care.)
1159          *
1160          * This can only fail for dump operations because the skb is always
1161          * properly sized for single flows.
1162          */
1163         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1164         if (start) {
1165                 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1166                 if (!err)
1167                         nla_nest_end(skb, start);
1168                 else {
1169                         if (skb_orig_len)
1170                                 goto error;
1171
1172                         nla_nest_cancel(skb, start);
1173                 }
1174         } else if (skb_orig_len)
1175                 goto nla_put_failure;
1176
1177         return genlmsg_end(skb, ovs_header);
1178
1179 nla_put_failure:
1180         err = -EMSGSIZE;
1181 error:
1182         genlmsg_cancel(skb, ovs_header);
1183         return err;
1184 }
1185
1186 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1187 {
1188         const struct sw_flow_actions *sf_acts;
1189
1190         sf_acts = ovsl_dereference(flow->sf_acts);
1191
1192         return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
1193 }
1194
1195 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1196                                                struct datapath *dp,
1197                                                u32 portid, u32 seq, u8 cmd)
1198 {
1199         struct sk_buff *skb;
1200         int retval;
1201
1202         skb = ovs_flow_cmd_alloc_info(flow);
1203         if (!skb)
1204                 return ERR_PTR(-ENOMEM);
1205
1206         retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1207         BUG_ON(retval < 0);
1208         return skb;
1209 }
1210
1211 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1212 {
1213         struct nlattr **a = info->attrs;
1214         struct ovs_header *ovs_header = info->userhdr;
1215         struct sw_flow_key key;
1216         struct sw_flow *flow;
1217         struct sk_buff *reply;
1218         struct datapath *dp;
1219         struct flow_table *table;
1220         struct sw_flow_actions *acts = NULL;
1221         int error;
1222         int key_len;
1223
1224         /* Extract key. */
1225         error = -EINVAL;
1226         if (!a[OVS_FLOW_ATTR_KEY])
1227                 goto error;
1228         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1229         if (error)
1230                 goto error;
1231
1232         /* Validate actions. */
1233         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1234                 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1235                 error = PTR_ERR(acts);
1236                 if (IS_ERR(acts))
1237                         goto error;
1238
1239                 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0, &acts);
1240                 if (error)
1241                         goto err_kfree;
1242         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1243                 error = -EINVAL;
1244                 goto error;
1245         }
1246
1247         ovs_lock();
1248         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1249         error = -ENODEV;
1250         if (!dp)
1251                 goto err_unlock_ovs;
1252
1253         table = ovsl_dereference(dp->table);
1254         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1255         if (!flow) {
1256                 /* Bail out if we're not allowed to create a new flow. */
1257                 error = -ENOENT;
1258                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1259                         goto err_unlock_ovs;
1260
1261                 /* Expand table, if necessary, to make room. */
1262                 if (ovs_flow_tbl_need_to_expand(table)) {
1263                         struct flow_table *new_table;
1264
1265                         new_table = ovs_flow_tbl_expand(table);
1266                         if (!IS_ERR(new_table)) {
1267                                 rcu_assign_pointer(dp->table, new_table);
1268                                 ovs_flow_tbl_deferred_destroy(table);
1269                                 table = ovsl_dereference(dp->table);
1270                         }
1271                 }
1272
1273                 /* Allocate flow. */
1274                 flow = ovs_flow_alloc();
1275                 if (IS_ERR(flow)) {
1276                         error = PTR_ERR(flow);
1277                         goto err_unlock_ovs;
1278                 }
1279                 flow->key = key;
1280                 clear_stats(flow);
1281
1282                 rcu_assign_pointer(flow->sf_acts, acts);
1283
1284                 /* Put flow in bucket. */
1285                 flow->hash = ovs_flow_hash(&key, key_len);
1286                 ovs_flow_tbl_insert(table, flow);
1287
1288                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1289                                                 info->snd_seq,
1290                                                 OVS_FLOW_CMD_NEW);
1291         } else {
1292                 /* We found a matching flow. */
1293                 struct sw_flow_actions *old_acts;
1294
1295                 /* Bail out if we're not allowed to modify an existing flow.
1296                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1297                  * because Generic Netlink treats the latter as a dump
1298                  * request.  We also accept NLM_F_EXCL in case that bug ever
1299                  * gets fixed.
1300                  */
1301                 error = -EEXIST;
1302                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1303                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1304                         goto err_unlock_ovs;
1305
1306                 /* Update actions. */
1307                 old_acts = ovsl_dereference(flow->sf_acts);
1308                 rcu_assign_pointer(flow->sf_acts, acts);
1309                 ovs_flow_deferred_free_acts(old_acts);
1310
1311                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1312                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1313
1314                 /* Clear stats. */
1315                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1316                         spin_lock_bh(&flow->lock);
1317                         clear_stats(flow);
1318                         spin_unlock_bh(&flow->lock);
1319                 }
1320         }
1321         ovs_unlock();
1322
1323         if (!IS_ERR(reply))
1324                 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1325         else
1326                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1327                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1328         return 0;
1329
1330 err_unlock_ovs:
1331         ovs_unlock();
1332 err_kfree:
1333         kfree(acts);
1334 error:
1335         return error;
1336 }
1337
1338 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1339 {
1340         struct nlattr **a = info->attrs;
1341         struct ovs_header *ovs_header = info->userhdr;
1342         struct sw_flow_key key;
1343         struct sk_buff *reply;
1344         struct sw_flow *flow;
1345         struct datapath *dp;
1346         struct flow_table *table;
1347         int err;
1348         int key_len;
1349
1350         if (!a[OVS_FLOW_ATTR_KEY])
1351                 return -EINVAL;
1352         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1353         if (err)
1354                 return err;
1355
1356         ovs_lock();
1357         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1358         if (!dp) {
1359                 err = -ENODEV;
1360                 goto unlock;
1361         }
1362
1363         table = ovsl_dereference(dp->table);
1364         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1365         if (!flow) {
1366                 err = -ENOENT;
1367                 goto unlock;
1368         }
1369
1370         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1371                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1372         if (IS_ERR(reply)) {
1373                 err = PTR_ERR(reply);
1374                 goto unlock;
1375         }
1376
1377         ovs_unlock();
1378         return genlmsg_reply(reply, info);
1379 unlock:
1380         ovs_unlock();
1381         return err;
1382 }
1383
1384 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1385 {
1386         struct nlattr **a = info->attrs;
1387         struct ovs_header *ovs_header = info->userhdr;
1388         struct sw_flow_key key;
1389         struct sk_buff *reply;
1390         struct sw_flow *flow;
1391         struct datapath *dp;
1392         struct flow_table *table;
1393         int err;
1394         int key_len;
1395
1396         ovs_lock();
1397         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1398         if (!dp) {
1399                 err = -ENODEV;
1400                 goto unlock;
1401         }
1402
1403         if (!a[OVS_FLOW_ATTR_KEY]) {
1404                 err = flush_flows(dp);
1405                 goto unlock;
1406         }
1407         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1408         if (err)
1409                 goto unlock;
1410
1411         table = ovsl_dereference(dp->table);
1412         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1413         if (!flow) {
1414                 err = -ENOENT;
1415                 goto unlock;
1416         }
1417
1418         reply = ovs_flow_cmd_alloc_info(flow);
1419         if (!reply) {
1420                 err = -ENOMEM;
1421                 goto unlock;
1422         }
1423
1424         ovs_flow_tbl_remove(table, flow);
1425
1426         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1427                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1428         BUG_ON(err < 0);
1429
1430         ovs_flow_deferred_free(flow);
1431         ovs_unlock();
1432
1433         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1434         return 0;
1435 unlock:
1436         ovs_unlock();
1437         return err;
1438 }
1439
1440 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1441 {
1442         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1443         struct datapath *dp;
1444         struct flow_table *table;
1445
1446         ovs_lock();
1447         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1448         if (!dp) {
1449                 ovs_unlock();
1450                 return -ENODEV;
1451         }
1452
1453         table = ovsl_dereference(dp->table);
1454
1455         for (;;) {
1456                 struct sw_flow *flow;
1457                 u32 bucket, obj;
1458
1459                 bucket = cb->args[0];
1460                 obj = cb->args[1];
1461                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1462                 if (!flow)
1463                         break;
1464
1465                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1466                                            NETLINK_CB(cb->skb).portid,
1467                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1468                                            OVS_FLOW_CMD_NEW) < 0)
1469                         break;
1470
1471                 cb->args[0] = bucket;
1472                 cb->args[1] = obj;
1473         }
1474         ovs_unlock();
1475         return skb->len;
1476 }
1477
1478 static struct genl_ops dp_flow_genl_ops[] = {
1479         { .cmd = OVS_FLOW_CMD_NEW,
1480           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1481           .policy = flow_policy,
1482           .doit = ovs_flow_cmd_new_or_set
1483         },
1484         { .cmd = OVS_FLOW_CMD_DEL,
1485           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1486           .policy = flow_policy,
1487           .doit = ovs_flow_cmd_del
1488         },
1489         { .cmd = OVS_FLOW_CMD_GET,
1490           .flags = 0,               /* OK for unprivileged users. */
1491           .policy = flow_policy,
1492           .doit = ovs_flow_cmd_get,
1493           .dumpit = ovs_flow_cmd_dump
1494         },
1495         { .cmd = OVS_FLOW_CMD_SET,
1496           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1497           .policy = flow_policy,
1498           .doit = ovs_flow_cmd_new_or_set,
1499         },
1500 };
1501
1502 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1503         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1504         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1505 };
1506
1507 static struct genl_family dp_datapath_genl_family = {
1508         .id = GENL_ID_GENERATE,
1509         .hdrsize = sizeof(struct ovs_header),
1510         .name = OVS_DATAPATH_FAMILY,
1511         .version = OVS_DATAPATH_VERSION,
1512         .maxattr = OVS_DP_ATTR_MAX,
1513         .netnsok = true,
1514         .parallel_ops = true,
1515 };
1516
1517 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1518         .name = OVS_DATAPATH_MCGROUP
1519 };
1520
1521 static size_t ovs_dp_cmd_msg_size(void)
1522 {
1523         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1524
1525         msgsize += nla_total_size(IFNAMSIZ);
1526         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1527
1528         return msgsize;
1529 }
1530
1531 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1532                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1533 {
1534         struct ovs_header *ovs_header;
1535         struct ovs_dp_stats dp_stats;
1536         int err;
1537
1538         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1539                                    flags, cmd);
1540         if (!ovs_header)
1541                 goto error;
1542
1543         ovs_header->dp_ifindex = get_dpifindex(dp);
1544
1545         rcu_read_lock();
1546         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1547         rcu_read_unlock();
1548         if (err)
1549                 goto nla_put_failure;
1550
1551         get_dp_stats(dp, &dp_stats);
1552         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1553                 goto nla_put_failure;
1554
1555         return genlmsg_end(skb, ovs_header);
1556
1557 nla_put_failure:
1558         genlmsg_cancel(skb, ovs_header);
1559 error:
1560         return -EMSGSIZE;
1561 }
1562
1563 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1564                                              u32 seq, u8 cmd)
1565 {
1566         struct sk_buff *skb;
1567         int retval;
1568
1569         skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1570         if (!skb)
1571                 return ERR_PTR(-ENOMEM);
1572
1573         retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1574         if (retval < 0) {
1575                 kfree_skb(skb);
1576                 return ERR_PTR(retval);
1577         }
1578         return skb;
1579 }
1580
1581 /* Called with ovs_mutex. */
1582 static struct datapath *lookup_datapath(struct net *net,
1583                                         struct ovs_header *ovs_header,
1584                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1585 {
1586         struct datapath *dp;
1587
1588         if (!a[OVS_DP_ATTR_NAME])
1589                 dp = get_dp(net, ovs_header->dp_ifindex);
1590         else {
1591                 struct vport *vport;
1592
1593                 rcu_read_lock();
1594                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1595                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1596                 rcu_read_unlock();
1597         }
1598         return dp ? dp : ERR_PTR(-ENODEV);
1599 }
1600
1601 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1602 {
1603         struct nlattr **a = info->attrs;
1604         struct vport_parms parms;
1605         struct sk_buff *reply;
1606         struct datapath *dp;
1607         struct vport *vport;
1608         struct ovs_net *ovs_net;
1609         int err, i;
1610
1611         err = -EINVAL;
1612         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1613                 goto err;
1614
1615         ovs_lock();
1616
1617         err = -ENOMEM;
1618         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1619         if (dp == NULL)
1620                 goto err_unlock_ovs;
1621
1622         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1623
1624         /* Allocate table. */
1625         err = -ENOMEM;
1626         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1627         if (!dp->table)
1628                 goto err_free_dp;
1629
1630         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1631         if (!dp->stats_percpu) {
1632                 err = -ENOMEM;
1633                 goto err_destroy_table;
1634         }
1635
1636         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1637                         GFP_KERNEL);
1638         if (!dp->ports) {
1639                 err = -ENOMEM;
1640                 goto err_destroy_percpu;
1641         }
1642
1643         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1644                 INIT_HLIST_HEAD(&dp->ports[i]);
1645
1646         /* Set up our datapath device. */
1647         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1648         parms.type = OVS_VPORT_TYPE_INTERNAL;
1649         parms.options = NULL;
1650         parms.dp = dp;
1651         parms.port_no = OVSP_LOCAL;
1652         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1653
1654         vport = new_vport(&parms);
1655         if (IS_ERR(vport)) {
1656                 err = PTR_ERR(vport);
1657                 if (err == -EBUSY)
1658                         err = -EEXIST;
1659
1660                 goto err_destroy_ports_array;
1661         }
1662
1663         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1664                                       info->snd_seq, OVS_DP_CMD_NEW);
1665         err = PTR_ERR(reply);
1666         if (IS_ERR(reply))
1667                 goto err_destroy_local_port;
1668
1669         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1670         list_add_tail(&dp->list_node, &ovs_net->dps);
1671
1672         ovs_unlock();
1673
1674         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1675         return 0;
1676
1677 err_destroy_local_port:
1678         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1679 err_destroy_ports_array:
1680         kfree(dp->ports);
1681 err_destroy_percpu:
1682         free_percpu(dp->stats_percpu);
1683 err_destroy_table:
1684         ovs_flow_tbl_destroy(ovsl_dereference(dp->table));
1685 err_free_dp:
1686         release_net(ovs_dp_get_net(dp));
1687         kfree(dp);
1688 err_unlock_ovs:
1689         ovs_unlock();
1690 err:
1691         return err;
1692 }
1693
1694 /* Called with ovs_mutex. */
1695 static void __dp_destroy(struct datapath *dp)
1696 {
1697         int i;
1698
1699         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1700                 struct vport *vport;
1701                 struct hlist_node *n;
1702
1703                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1704                         if (vport->port_no != OVSP_LOCAL)
1705                                 ovs_dp_detach_port(vport);
1706         }
1707
1708         list_del(&dp->list_node);
1709
1710         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1711          * all port in datapath are destroyed first before freeing datapath.
1712          */
1713         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1714
1715         call_rcu(&dp->rcu, destroy_dp_rcu);
1716 }
1717
1718 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1719 {
1720         struct sk_buff *reply;
1721         struct datapath *dp;
1722         int err;
1723
1724         ovs_lock();
1725         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1726         err = PTR_ERR(dp);
1727         if (IS_ERR(dp))
1728                 goto unlock;
1729
1730         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1731                                       info->snd_seq, OVS_DP_CMD_DEL);
1732         err = PTR_ERR(reply);
1733         if (IS_ERR(reply))
1734                 goto unlock;
1735
1736         __dp_destroy(dp);
1737         ovs_unlock();
1738
1739         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1740
1741         return 0;
1742 unlock:
1743         ovs_unlock();
1744         return err;
1745 }
1746
1747 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1748 {
1749         struct sk_buff *reply;
1750         struct datapath *dp;
1751         int err;
1752
1753         ovs_lock();
1754         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1755         err = PTR_ERR(dp);
1756         if (IS_ERR(dp))
1757                 goto unlock;
1758
1759         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1760                                       info->snd_seq, OVS_DP_CMD_NEW);
1761         if (IS_ERR(reply)) {
1762                 err = PTR_ERR(reply);
1763                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1764                                 ovs_dp_datapath_multicast_group.id, err);
1765                 err = 0;
1766                 goto unlock;
1767         }
1768
1769         ovs_unlock();
1770         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1771
1772         return 0;
1773 unlock:
1774         ovs_unlock();
1775         return err;
1776 }
1777
1778 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1779 {
1780         struct sk_buff *reply;
1781         struct datapath *dp;
1782         int err;
1783
1784         ovs_lock();
1785         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1786         if (IS_ERR(dp)) {
1787                 err = PTR_ERR(dp);
1788                 goto unlock;
1789         }
1790
1791         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1792                                       info->snd_seq, OVS_DP_CMD_NEW);
1793         if (IS_ERR(reply)) {
1794                 err = PTR_ERR(reply);
1795                 goto unlock;
1796         }
1797
1798         ovs_unlock();
1799         return genlmsg_reply(reply, info);
1800
1801 unlock:
1802         ovs_unlock();
1803         return err;
1804 }
1805
1806 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1807 {
1808         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1809         struct datapath *dp;
1810         int skip = cb->args[0];
1811         int i = 0;
1812
1813         ovs_lock();
1814         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1815                 if (i >= skip &&
1816                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1817                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1818                                          OVS_DP_CMD_NEW) < 0)
1819                         break;
1820                 i++;
1821         }
1822         ovs_unlock();
1823
1824         cb->args[0] = i;
1825
1826         return skb->len;
1827 }
1828
1829 static struct genl_ops dp_datapath_genl_ops[] = {
1830         { .cmd = OVS_DP_CMD_NEW,
1831           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1832           .policy = datapath_policy,
1833           .doit = ovs_dp_cmd_new
1834         },
1835         { .cmd = OVS_DP_CMD_DEL,
1836           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1837           .policy = datapath_policy,
1838           .doit = ovs_dp_cmd_del
1839         },
1840         { .cmd = OVS_DP_CMD_GET,
1841           .flags = 0,               /* OK for unprivileged users. */
1842           .policy = datapath_policy,
1843           .doit = ovs_dp_cmd_get,
1844           .dumpit = ovs_dp_cmd_dump
1845         },
1846         { .cmd = OVS_DP_CMD_SET,
1847           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1848           .policy = datapath_policy,
1849           .doit = ovs_dp_cmd_set,
1850         },
1851 };
1852
1853 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1854         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1855         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1856         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1857         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1858         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1859         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1860 };
1861
1862 static struct genl_family dp_vport_genl_family = {
1863         .id = GENL_ID_GENERATE,
1864         .hdrsize = sizeof(struct ovs_header),
1865         .name = OVS_VPORT_FAMILY,
1866         .version = OVS_VPORT_VERSION,
1867         .maxattr = OVS_VPORT_ATTR_MAX,
1868         .netnsok = true,
1869         .parallel_ops = true,
1870 };
1871
1872 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1873         .name = OVS_VPORT_MCGROUP
1874 };
1875
1876 /* Called with ovs_mutex or RCU read lock. */
1877 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1878                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1879 {
1880         struct ovs_header *ovs_header;
1881         struct ovs_vport_stats vport_stats;
1882         int err;
1883
1884         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1885                                  flags, cmd);
1886         if (!ovs_header)
1887                 return -EMSGSIZE;
1888
1889         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1890
1891         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1892             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1893             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1894             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1895                 goto nla_put_failure;
1896
1897         ovs_vport_get_stats(vport, &vport_stats);
1898         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1899                     &vport_stats))
1900                 goto nla_put_failure;
1901
1902         err = ovs_vport_get_options(vport, skb);
1903         if (err == -EMSGSIZE)
1904                 goto error;
1905
1906         return genlmsg_end(skb, ovs_header);
1907
1908 nla_put_failure:
1909         err = -EMSGSIZE;
1910 error:
1911         genlmsg_cancel(skb, ovs_header);
1912         return err;
1913 }
1914
1915 /* Called with ovs_mutex or RCU read lock. */
1916 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1917                                          u32 seq, u8 cmd)
1918 {
1919         struct sk_buff *skb;
1920         int retval;
1921
1922         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1923         if (!skb)
1924                 return ERR_PTR(-ENOMEM);
1925
1926         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1927         BUG_ON(retval < 0);
1928
1929         return skb;
1930 }
1931
1932 /* Called with ovs_mutex or RCU read lock. */
1933 static struct vport *lookup_vport(struct net *net,
1934                                   struct ovs_header *ovs_header,
1935                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1936 {
1937         struct datapath *dp;
1938         struct vport *vport;
1939
1940         if (a[OVS_VPORT_ATTR_NAME]) {
1941                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1942                 if (!vport)
1943                         return ERR_PTR(-ENODEV);
1944                 if (ovs_header->dp_ifindex &&
1945                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1946                         return ERR_PTR(-ENODEV);
1947                 return vport;
1948         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1949                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1950
1951                 if (port_no >= DP_MAX_PORTS)
1952                         return ERR_PTR(-EFBIG);
1953
1954                 dp = get_dp(net, ovs_header->dp_ifindex);
1955                 if (!dp)
1956                         return ERR_PTR(-ENODEV);
1957
1958                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1959                 if (!vport)
1960                         return ERR_PTR(-ENODEV);
1961                 return vport;
1962         } else
1963                 return ERR_PTR(-EINVAL);
1964 }
1965
1966 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1967 {
1968         struct nlattr **a = info->attrs;
1969         struct ovs_header *ovs_header = info->userhdr;
1970         struct vport_parms parms;
1971         struct sk_buff *reply;
1972         struct vport *vport;
1973         struct datapath *dp;
1974         u32 port_no;
1975         int err;
1976
1977         err = -EINVAL;
1978         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1979             !a[OVS_VPORT_ATTR_UPCALL_PID])
1980                 goto exit;
1981
1982         ovs_lock();
1983         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1984         err = -ENODEV;
1985         if (!dp)
1986                 goto exit_unlock;
1987
1988         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1989                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1990
1991                 err = -EFBIG;
1992                 if (port_no >= DP_MAX_PORTS)
1993                         goto exit_unlock;
1994
1995                 vport = ovs_vport_ovsl(dp, port_no);
1996                 err = -EBUSY;
1997                 if (vport)
1998                         goto exit_unlock;
1999         } else {
2000                 for (port_no = 1; ; port_no++) {
2001                         if (port_no >= DP_MAX_PORTS) {
2002                                 err = -EFBIG;
2003                                 goto exit_unlock;
2004                         }
2005                         vport = ovs_vport_ovsl(dp, port_no);
2006                         if (!vport)
2007                                 break;
2008                 }
2009         }
2010
2011         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2012         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2013         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2014         parms.dp = dp;
2015         parms.port_no = port_no;
2016         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2017
2018         vport = new_vport(&parms);
2019         err = PTR_ERR(vport);
2020         if (IS_ERR(vport))
2021                 goto exit_unlock;
2022
2023         err = 0;
2024         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
2025                                          OVS_VPORT_CMD_NEW);
2026         if (IS_ERR(reply)) {
2027                 err = PTR_ERR(reply);
2028                 ovs_dp_detach_port(vport);
2029                 goto exit_unlock;
2030         }
2031
2032         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2033
2034 exit_unlock:
2035         ovs_unlock();
2036 exit:
2037         return err;
2038 }
2039
2040 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2041 {
2042         struct nlattr **a = info->attrs;
2043         struct sk_buff *reply;
2044         struct vport *vport;
2045         int err;
2046
2047         ovs_lock();
2048         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2049         err = PTR_ERR(vport);
2050         if (IS_ERR(vport))
2051                 goto exit_unlock;
2052
2053         if (a[OVS_VPORT_ATTR_TYPE] &&
2054             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2055                 err = -EINVAL;
2056                 goto exit_unlock;
2057         }
2058
2059         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2060         if (!reply) {
2061                 err = -ENOMEM;
2062                 goto exit_unlock;
2063         }
2064
2065         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2066                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2067                 if (err)
2068                         goto exit_free;
2069         }
2070
2071         if (a[OVS_VPORT_ATTR_UPCALL_PID])
2072                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2073
2074         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2075                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2076         BUG_ON(err < 0);
2077
2078         ovs_unlock();
2079         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2080         return 0;
2081
2082         rtnl_unlock();
2083         return 0;
2084
2085 exit_free:
2086         kfree_skb(reply);
2087 exit_unlock:
2088         ovs_unlock();
2089         return err;
2090 }
2091
2092 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2093 {
2094         struct nlattr **a = info->attrs;
2095         struct sk_buff *reply;
2096         struct vport *vport;
2097         int err;
2098
2099         ovs_lock();
2100         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2101         err = PTR_ERR(vport);
2102         if (IS_ERR(vport))
2103                 goto exit_unlock;
2104
2105         if (vport->port_no == OVSP_LOCAL) {
2106                 err = -EINVAL;
2107                 goto exit_unlock;
2108         }
2109
2110         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2111                                          info->snd_seq, OVS_VPORT_CMD_DEL);
2112         err = PTR_ERR(reply);
2113         if (IS_ERR(reply))
2114                 goto exit_unlock;
2115
2116         err = 0;
2117         ovs_dp_detach_port(vport);
2118
2119         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2120
2121 exit_unlock:
2122         ovs_unlock();
2123         return err;
2124 }
2125
2126 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2127 {
2128         struct nlattr **a = info->attrs;
2129         struct ovs_header *ovs_header = info->userhdr;
2130         struct sk_buff *reply;
2131         struct vport *vport;
2132         int err;
2133
2134         rcu_read_lock();
2135         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2136         err = PTR_ERR(vport);
2137         if (IS_ERR(vport))
2138                 goto exit_unlock;
2139
2140         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2141                                          info->snd_seq, OVS_VPORT_CMD_NEW);
2142         err = PTR_ERR(reply);
2143         if (IS_ERR(reply))
2144                 goto exit_unlock;
2145
2146         rcu_read_unlock();
2147
2148         return genlmsg_reply(reply, info);
2149
2150 exit_unlock:
2151         rcu_read_unlock();
2152         return err;
2153 }
2154
2155 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2156 {
2157         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2158         struct datapath *dp;
2159         int bucket = cb->args[0], skip = cb->args[1];
2160         int i, j = 0;
2161
2162         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2163         if (!dp)
2164                 return -ENODEV;
2165
2166         rcu_read_lock();
2167         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2168                 struct vport *vport;
2169
2170                 j = 0;
2171                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2172                         if (j >= skip &&
2173                             ovs_vport_cmd_fill_info(vport, skb,
2174                                                     NETLINK_CB(cb->skb).portid,
2175                                                     cb->nlh->nlmsg_seq,
2176                                                     NLM_F_MULTI,
2177                                                     OVS_VPORT_CMD_NEW) < 0)
2178                                 goto out;
2179
2180                         j++;
2181                 }
2182                 skip = 0;
2183         }
2184 out:
2185         rcu_read_unlock();
2186
2187         cb->args[0] = i;
2188         cb->args[1] = j;
2189
2190         return skb->len;
2191 }
2192
2193 static struct genl_ops dp_vport_genl_ops[] = {
2194         { .cmd = OVS_VPORT_CMD_NEW,
2195           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2196           .policy = vport_policy,
2197           .doit = ovs_vport_cmd_new
2198         },
2199         { .cmd = OVS_VPORT_CMD_DEL,
2200           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2201           .policy = vport_policy,
2202           .doit = ovs_vport_cmd_del
2203         },
2204         { .cmd = OVS_VPORT_CMD_GET,
2205           .flags = 0,               /* OK for unprivileged users. */
2206           .policy = vport_policy,
2207           .doit = ovs_vport_cmd_get,
2208           .dumpit = ovs_vport_cmd_dump
2209         },
2210         { .cmd = OVS_VPORT_CMD_SET,
2211           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2212           .policy = vport_policy,
2213           .doit = ovs_vport_cmd_set,
2214         },
2215 };
2216
2217 struct genl_family_and_ops {
2218         struct genl_family *family;
2219         struct genl_ops *ops;
2220         int n_ops;
2221         struct genl_multicast_group *group;
2222 };
2223
2224 static const struct genl_family_and_ops dp_genl_families[] = {
2225         { &dp_datapath_genl_family,
2226           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2227           &ovs_dp_datapath_multicast_group },
2228         { &dp_vport_genl_family,
2229           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2230           &ovs_dp_vport_multicast_group },
2231         { &dp_flow_genl_family,
2232           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2233           &ovs_dp_flow_multicast_group },
2234         { &dp_packet_genl_family,
2235           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2236           NULL },
2237 };
2238
2239 static void dp_unregister_genl(int n_families)
2240 {
2241         int i;
2242
2243         for (i = 0; i < n_families; i++)
2244                 genl_unregister_family(dp_genl_families[i].family);
2245 }
2246
2247 static int dp_register_genl(void)
2248 {
2249         int n_registered;
2250         int err;
2251         int i;
2252
2253         n_registered = 0;
2254         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2255                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2256
2257                 err = genl_register_family_with_ops(f->family, f->ops,
2258                                                     f->n_ops);
2259                 if (err)
2260                         goto error;
2261                 n_registered++;
2262
2263                 if (f->group) {
2264                         err = genl_register_mc_group(f->family, f->group);
2265                         if (err)
2266                                 goto error;
2267                 }
2268         }
2269
2270         return 0;
2271
2272 error:
2273         dp_unregister_genl(n_registered);
2274         return err;
2275 }
2276
2277 static void rehash_flow_table(struct work_struct *work)
2278 {
2279         struct datapath *dp;
2280         struct net *net;
2281
2282         ovs_lock();
2283         rtnl_lock();
2284         for_each_net(net) {
2285                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2286
2287                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2288                         struct flow_table *old_table = ovsl_dereference(dp->table);
2289                         struct flow_table *new_table;
2290
2291                         new_table = ovs_flow_tbl_rehash(old_table);
2292                         if (!IS_ERR(new_table)) {
2293                                 rcu_assign_pointer(dp->table, new_table);
2294                                 ovs_flow_tbl_deferred_destroy(old_table);
2295                         }
2296                 }
2297         }
2298         rtnl_unlock();
2299         ovs_unlock();
2300         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2301 }
2302
2303 static int __net_init ovs_init_net(struct net *net)
2304 {
2305         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2306
2307         INIT_LIST_HEAD(&ovs_net->dps);
2308         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2309         return 0;
2310 }
2311
2312 static void __net_exit ovs_exit_net(struct net *net)
2313 {
2314         struct datapath *dp, *dp_next;
2315         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2316
2317         ovs_lock();
2318         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2319                 __dp_destroy(dp);
2320         ovs_unlock();
2321
2322         cancel_work_sync(&ovs_net->dp_notify_work);
2323 }
2324
2325 static struct pernet_operations ovs_net_ops = {
2326         .init = ovs_init_net,
2327         .exit = ovs_exit_net,
2328         .id   = &ovs_net_id,
2329         .size = sizeof(struct ovs_net),
2330 };
2331
2332 static int __init dp_init(void)
2333 {
2334         int err;
2335
2336         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2337
2338         pr_info("Open vSwitch switching datapath\n");
2339
2340         err = ovs_flow_init();
2341         if (err)
2342                 goto error;
2343
2344         err = ovs_vport_init();
2345         if (err)
2346                 goto error_flow_exit;
2347
2348         err = register_pernet_device(&ovs_net_ops);
2349         if (err)
2350                 goto error_vport_exit;
2351
2352         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2353         if (err)
2354                 goto error_netns_exit;
2355
2356         err = dp_register_genl();
2357         if (err < 0)
2358                 goto error_unreg_notifier;
2359
2360         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2361
2362         return 0;
2363
2364 error_unreg_notifier:
2365         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2366 error_netns_exit:
2367         unregister_pernet_device(&ovs_net_ops);
2368 error_vport_exit:
2369         ovs_vport_exit();
2370 error_flow_exit:
2371         ovs_flow_exit();
2372 error:
2373         return err;
2374 }
2375
2376 static void dp_cleanup(void)
2377 {
2378         cancel_delayed_work_sync(&rehash_flow_wq);
2379         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2380         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2381         unregister_pernet_device(&ovs_net_ops);
2382         rcu_barrier();
2383         ovs_vport_exit();
2384         ovs_flow_exit();
2385 }
2386
2387 module_init(dp_init);
2388 module_exit(dp_cleanup);
2389
2390 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2391 MODULE_LICENSE("GPL");