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 <linux/genetlink.h>
  52 #include <net/genetlink.h>
  53 #include <net/genetlink.h>
  54 #include <net/net_namespace.h>
  55 #include <net/netns/generic.h>
  56
  57 #include "checksum.h"
  58 #include "datapath.h"
  59 #include "flow.h"
  60 #include "vlan.h"
  61 #include "tunnel.h"
  62 #include "vport-internal_dev.h"
  63 #include "vport-netdev.h"
  64
  65 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
  66     LINUX_VERSION_CODE >= KERNEL_VERSION(3,9,0)
  67 #error Kernels before 2.6.18 or after 3.8 are not supported by this version of Open vSwitch.
  68 #endif
  69
  70 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
  71 static void rehash_flow_table(struct work_struct *work);
  72 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
  73
  74 int ovs_net_id __read_mostly;
  75
  76 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
  77                        struct genl_multicast_group *grp)
  78 {
  79         genl_notify(skb, genl_info_net(info), info->snd_portid,
  80                     grp->id, info->nlhdr, GFP_KERNEL);
  81 }
  82
  83 /**
  84  * DOC: Locking:
  85  *
  86  * All writes e.g. Writes to device state (add/remove datapath, port, set
  87  * operations on vports, etc.), Writes to other state (flow table
  88  * modifications, set miscellaneous datapath parameters, etc.) are protected
  89  * by ovs_lock.
  90  *
  91  * Reads are protected by RCU.
  92  *
  93  * There are a few special cases (mostly stats) that have their own
  94  * synchronization but they nest under all of above and don't interact with
  95  * each other.
  96  *
  97  * The RTNL lock nests inside ovs_mutex.
  98  */
  99
 100 static DEFINE_MUTEX(ovs_mutex);
 101
 102 void ovs_lock(void)
 103 {
 104         mutex_lock(&ovs_mutex);
 105 }
 106
 107 void ovs_unlock(void)
 108 {
 109         mutex_unlock(&ovs_mutex);
 110 }
 111
 112 #ifdef CONFIG_LOCKDEP
 113 int lockdep_ovsl_is_held(void)
 114 {
 115         if (debug_locks)
 116                 return lockdep_is_held(&ovs_mutex);
 117         else
 118                 return 1;
 119 }
 120 #endif
 121
 122 static struct vport *new_vport(const struct vport_parms *);
 123 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
 124                              const struct dp_upcall_info *);
 125 static int queue_userspace_packet(struct net *, int dp_ifindex,
 126                                   struct sk_buff *,
 127                                   const struct dp_upcall_info *);
 128
 129 /* Must be called with rcu_read_lock or ovs_mutex. */
 130 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 131 {
 132         struct datapath *dp = NULL;
 133         struct net_device *dev;
 134
 135         rcu_read_lock();
 136         dev = dev_get_by_index_rcu(net, dp_ifindex);
 137         if (dev) {
 138                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 139                 if (vport)
 140                         dp = vport->dp;
 141         }
 142         rcu_read_unlock();
 143
 144         return dp;
 145 }
 146
 147 /* Must be called with rcu_read_lock or ovs_mutex. */
 148 const char *ovs_dp_name(const struct datapath *dp)
 149 {
 150         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 151         return vport->ops->get_name(vport);
 152 }
 153
 154 static int get_dpifindex(struct datapath *dp)
 155 {
 156         struct vport *local;
 157         int ifindex;
 158
 159         rcu_read_lock();
 160
 161         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 162         if (local)
 163                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 164         else
 165                 ifindex = 0;
 166
 167         rcu_read_unlock();
 168
 169         return ifindex;
 170 }
 171
 172 static void destroy_dp_rcu(struct rcu_head *rcu)
 173 {
 174         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 175
 176         ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
 177         free_percpu(dp->stats_percpu);
 178         release_net(ovs_dp_get_net(dp));
 179         kfree(dp->ports);
 180         kfree(dp);
 181 }
 182
 183 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 184                                             u16 port_no)
 185 {
 186         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 187 }
 188
 189 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 190 {
 191         struct vport *vport;
 192         struct hlist_head *head;
 193
 194         head = vport_hash_bucket(dp, port_no);
 195         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 196                 if (vport->port_no == port_no)
 197                         return vport;
 198         }
 199         return NULL;
 200 }
 201
 202 /* Called with ovs_mutex. */
 203 static struct vport *new_vport(const struct vport_parms *parms)
 204 {
 205         struct vport *vport;
 206
 207         vport = ovs_vport_add(parms);
 208         if (!IS_ERR(vport)) {
 209                 struct datapath *dp = parms->dp;
 210                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 211
 212                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 213         }
 214         return vport;
 215 }
 216
 217 void ovs_dp_detach_port(struct vport *p)
 218 {
 219         ASSERT_OVSL();
 220
 221         /* First drop references to device. */
 222         hlist_del_rcu(&p->dp_hash_node);
 223
 224         /* Then destroy it. */
 225         ovs_vport_del(p);
 226 }
 227
 228 /* Must be called with rcu_read_lock. */
 229 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 230 {
 231         struct datapath *dp = p->dp;
 232         struct sw_flow *flow;
 233         struct dp_stats_percpu *stats;
 234         struct sw_flow_key key;
 235         u64 *stats_counter;
 236         int error;
 237         int key_len;
 238
 239         stats = this_cpu_ptr(dp->stats_percpu);
 240
 241         /* Extract flow from 'skb' into 'key'. */
 242         error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
 243         if (unlikely(error)) {
 244                 kfree_skb(skb);
 245                 return;
 246         }
 247
 248         /* Look up flow. */
 249         flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
 250         if (unlikely(!flow)) {
 251                 struct dp_upcall_info upcall;
 252
 253                 upcall.cmd = OVS_PACKET_CMD_MISS;
 254                 upcall.key = &key;
 255                 upcall.userdata = NULL;
 256                 upcall.portid = p->upcall_portid;
 257                 ovs_dp_upcall(dp, skb, &upcall);
 258                 consume_skb(skb);
 259                 stats_counter = &stats->n_missed;
 260                 goto out;
 261         }
 262
 263         OVS_CB(skb)->flow = flow;
 264
 265         stats_counter = &stats->n_hit;
 266         ovs_flow_used(OVS_CB(skb)->flow, skb);
 267         ovs_execute_actions(dp, skb);
 268
 269 out:
 270         /* Update datapath statistics. */
 271         u64_stats_update_begin(&stats->sync);
 272         (*stats_counter)++;
 273         u64_stats_update_end(&stats->sync);
 274 }
 275
 276 static struct genl_family dp_packet_genl_family = {
 277         .id = GENL_ID_GENERATE,
 278         .hdrsize = sizeof(struct ovs_header),
 279         .name = OVS_PACKET_FAMILY,
 280         .version = OVS_PACKET_VERSION,
 281         .maxattr = OVS_PACKET_ATTR_MAX,
 282          SET_NETNSOK
 283 };
 284
 285 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 286                   const struct dp_upcall_info *upcall_info)
 287 {
 288         struct dp_stats_percpu *stats;
 289         int dp_ifindex;
 290         int err;
 291
 292         if (upcall_info->portid == 0) {
 293                 err = -ENOTCONN;
 294                 goto err;
 295         }
 296
 297         dp_ifindex = get_dpifindex(dp);
 298         if (!dp_ifindex) {
 299                 err = -ENODEV;
 300                 goto err;
 301         }
 302
 303         forward_ip_summed(skb, true);
 304
 305         if (!skb_is_gso(skb))
 306                 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 307         else
 308                 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 309         if (err)
 310                 goto err;
 311
 312         return 0;
 313
 314 err:
 315         stats = this_cpu_ptr(dp->stats_percpu);
 316
 317         u64_stats_update_begin(&stats->sync);
 318         stats->n_lost++;
 319         u64_stats_update_end(&stats->sync);
 320
 321         return err;
 322 }
 323
 324 static int queue_gso_packets(struct net *net, int dp_ifindex,
 325                              struct sk_buff *skb,
 326                              const struct dp_upcall_info *upcall_info)
 327 {
 328         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 329         struct dp_upcall_info later_info;
 330         struct sw_flow_key later_key;
 331         struct sk_buff *segs, *nskb;
 332         int err;
 333
 334         segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
 335         if (IS_ERR(segs))
 336                 return PTR_ERR(segs);
 337
 338         /* Queue all of the segments. */
 339         skb = segs;
 340         do {
 341                 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
 342                 if (err)
 343                         break;
 344
 345                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 346                         /* The initial flow key extracted by ovs_flow_extract()
 347                          * in this case is for a first fragment, so we need to
 348                          * properly mark later fragments.
 349                          */
 350                         later_key = *upcall_info->key;
 351                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 352
 353                         later_info = *upcall_info;
 354                         later_info.key = &later_key;
 355                         upcall_info = &later_info;
 356                 }
 357         } while ((skb = skb->next));
 358
 359         /* Free all of the segments. */
 360         skb = segs;
 361         do {
 362                 nskb = skb->next;
 363                 if (err)
 364                         kfree_skb(skb);
 365                 else
 366                         consume_skb(skb);
 367         } while ((skb = nskb));
 368         return err;
 369 }
 370
 371 static size_t key_attr_size(void)
 372 {
 373         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 374                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 375                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
 376                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 377                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 378                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
 379                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
 380                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 381                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
 382                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 383                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 384                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 385                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 386                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 387                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 388                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 389                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 390                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 391                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 392 }
 393
 394 static size_t upcall_msg_size(const struct sk_buff *skb,
 395                               const struct nlattr *userdata)
 396 {
 397         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 398                 + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
 399                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 400
 401         /* OVS_PACKET_ATTR_USERDATA */
 402         if (userdata)
 403                 size += NLA_ALIGN(userdata->nla_len);
 404
 405         return size;
 406 }
 407
 408 static int queue_userspace_packet(struct net *net, int dp_ifindex,
 409                                   struct sk_buff *skb,
 410                                   const struct dp_upcall_info *upcall_info)
 411 {
 412         struct ovs_header *upcall;
 413         struct sk_buff *nskb = NULL;
 414         struct sk_buff *user_skb; /* to be queued to userspace */
 415         struct nlattr *nla;
 416         int err;
 417
 418         if (vlan_tx_tag_present(skb)) {
 419                 nskb = skb_clone(skb, GFP_ATOMIC);
 420                 if (!nskb)
 421                         return -ENOMEM;
 422
 423                 err = vlan_deaccel_tag(nskb);
 424                 if (err)
 425                         return err;
 426
 427                 skb = nskb;
 428         }
 429
 430         if (nla_attr_size(skb->len) > USHRT_MAX) {
 431                 err = -EFBIG;
 432                 goto out;
 433         }
 434
 435         user_skb = genlmsg_new(upcall_msg_size(skb, upcall_info->userdata), GFP_ATOMIC);
 436         if (!user_skb) {
 437                 err = -ENOMEM;
 438                 goto out;
 439         }
 440
 441         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 442                              0, upcall_info->cmd);
 443         upcall->dp_ifindex = dp_ifindex;
 444
 445         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 446         ovs_flow_to_nlattrs(upcall_info->key, user_skb);
 447         nla_nest_end(user_skb, nla);
 448
 449         if (upcall_info->userdata)
 450                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 451                           nla_len(upcall_info->userdata),
 452                           nla_data(upcall_info->userdata));
 453
 454         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 455
 456         skb_copy_and_csum_dev(skb, nla_data(nla));
 457
 458         genlmsg_end(user_skb, upcall);
 459         err = genlmsg_unicast(net, user_skb, upcall_info->portid);
 460
 461 out:
 462         kfree_skb(nskb);
 463         return err;
 464 }
 465
 466 /* Called with ovs_mutex. */
 467 static int flush_flows(struct datapath *dp)
 468 {
 469         struct flow_table *old_table;
 470         struct flow_table *new_table;
 471
 472         old_table = ovsl_dereference(dp->table);
 473         new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
 474         if (!new_table)
 475                 return -ENOMEM;
 476
 477         rcu_assign_pointer(dp->table, new_table);
 478
 479         ovs_flow_tbl_deferred_destroy(old_table);
 480         return 0;
 481 }
 482
 483 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa, int attr_len)
 484 {
 485
 486         struct sw_flow_actions *acts;
 487         int new_acts_size;
 488         int req_size = NLA_ALIGN(attr_len);
 489         int next_offset = offsetof(struct sw_flow_actions, actions) +
 490                                         (*sfa)->actions_len;
 491
 492         if (req_size <= (ksize(*sfa) - next_offset))
 493                 goto out;
 494
 495         new_acts_size = ksize(*sfa) * 2;
 496
 497         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
 498                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
 499                         return ERR_PTR(-EMSGSIZE);
 500                 new_acts_size = MAX_ACTIONS_BUFSIZE;
 501         }
 502
 503         acts = ovs_flow_actions_alloc(new_acts_size);
 504         if (IS_ERR(acts))
 505                 return (void *)acts;
 506
 507         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
 508         acts->actions_len = (*sfa)->actions_len;
 509         kfree(*sfa);
 510         *sfa = acts;
 511
 512 out:
 513         (*sfa)->actions_len += req_size;
 514         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
 515 }
 516
 517 static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, int len)
 518 {
 519         struct nlattr *a;
 520
 521         a = reserve_sfa_size(sfa, nla_attr_size(len));
 522         if (IS_ERR(a))
 523                 return PTR_ERR(a);
 524
 525         a->nla_type = attrtype;
 526         a->nla_len = nla_attr_size(len);
 527
 528         if (data)
 529                 memcpy(nla_data(a), data, len);
 530         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
 531
 532         return 0;
 533 }
 534
 535 static inline int add_nested_action_start(struct sw_flow_actions **sfa, int attrtype)
 536 {
 537         int used = (*sfa)->actions_len;
 538         int err;
 539
 540         err = add_action(sfa, attrtype, NULL, 0);
 541         if (err)
 542                 return err;
 543
 544         return used;
 545 }
 546
 547 static inline void add_nested_action_end(struct sw_flow_actions *sfa, int st_offset)
 548 {
 549         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions + st_offset);
 550
 551         a->nla_len = sfa->actions_len - st_offset;
 552 }
 553
 554 static int validate_and_copy_actions(const struct nlattr *attr,
 555                                 const struct sw_flow_key *key, int depth,
 556                                 struct sw_flow_actions **sfa);
 557
 558 static int validate_and_copy_sample(const struct nlattr *attr,
 559                            const struct sw_flow_key *key, int depth,
 560                            struct sw_flow_actions **sfa)
 561 {
 562         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 563         const struct nlattr *probability, *actions;
 564         const struct nlattr *a;
 565         int rem, start, err, st_acts;
 566
 567         memset(attrs, 0, sizeof(attrs));
 568         nla_for_each_nested(a, attr, rem) {
 569                 int type = nla_type(a);
 570                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 571                         return -EINVAL;
 572                 attrs[type] = a;
 573         }
 574         if (rem)
 575                 return -EINVAL;
 576
 577         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 578         if (!probability || nla_len(probability) != sizeof(u32))
 579                 return -EINVAL;
 580
 581         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 582         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 583                 return -EINVAL;
 584
 585         /* validation done, copy sample action. */
 586         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
 587         if (start < 0)
 588                 return start;
 589         err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY, nla_data(probability), sizeof(u32));
 590         if (err)
 591                 return err;
 592         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
 593         if (st_acts < 0)
 594                 return st_acts;
 595
 596         err = validate_and_copy_actions(actions, key, depth + 1, sfa);
 597         if (err)
 598                 return err;
 599
 600         add_nested_action_end(*sfa, st_acts);
 601         add_nested_action_end(*sfa, start);
 602
 603         return 0;
 604 }
 605
 606 static int validate_tp_port(const struct sw_flow_key *flow_key)
 607 {
 608         if (flow_key->eth.type == htons(ETH_P_IP)) {
 609                 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
 610                         return 0;
 611         } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
 612                 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
 613                         return 0;
 614         }
 615
 616         return -EINVAL;
 617 }
 618
 619 static int validate_and_copy_set_tun(const struct nlattr *attr,
 620                                      struct sw_flow_actions **sfa)
 621 {
 622         struct ovs_key_ipv4_tunnel tun_key;
 623         int err, start;
 624
 625         err = ipv4_tun_from_nlattr(nla_data(attr), &tun_key);
 626         if (err)
 627                 return err;
 628
 629         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
 630         if (start < 0)
 631                 return start;
 632
 633         err = add_action(sfa, OVS_KEY_ATTR_IPV4_TUNNEL, &tun_key, sizeof(tun_key));
 634         add_nested_action_end(*sfa, start);
 635
 636         return err;
 637 }
 638
 639 static int validate_set(const struct nlattr *a,
 640                         const struct sw_flow_key *flow_key,
 641                         struct sw_flow_actions **sfa,
 642                         bool *set_tun)
 643 {
 644         const struct nlattr *ovs_key = nla_data(a);
 645         int key_type = nla_type(ovs_key);
 646
 647         /* There can be only one key in a action */
 648         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 649                 return -EINVAL;
 650
 651         if (key_type > OVS_KEY_ATTR_MAX ||
 652             (ovs_key_lens[key_type] != nla_len(ovs_key) &&
 653              ovs_key_lens[key_type] != -1))
 654                 return -EINVAL;
 655
 656         switch (key_type) {
 657         const struct ovs_key_ipv4 *ipv4_key;
 658         const struct ovs_key_ipv6 *ipv6_key;
 659         int err;
 660
 661         case OVS_KEY_ATTR_PRIORITY:
 662         case OVS_KEY_ATTR_ETHERNET:
 663                 break;
 664
 665         case OVS_KEY_ATTR_SKB_MARK:
 666 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
 667                 if (nla_get_u32(ovs_key) != 0)
 668                         return -EINVAL;
 669 #endif
 670                 break;
 671
 672         case OVS_KEY_ATTR_TUNNEL:
 673                 *set_tun = true;
 674                 err = validate_and_copy_set_tun(a, sfa);
 675                 if (err)
 676                         return err;
 677                 break;
 678
 679         case OVS_KEY_ATTR_IPV4:
 680                 if (flow_key->eth.type != htons(ETH_P_IP))
 681                         return -EINVAL;
 682
 683                 if (!flow_key->ip.proto)
 684                         return -EINVAL;
 685
 686                 ipv4_key = nla_data(ovs_key);
 687                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 688                         return -EINVAL;
 689
 690                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 691                         return -EINVAL;
 692
 693                 break;
 694
 695         case OVS_KEY_ATTR_IPV6:
 696                 if (flow_key->eth.type != htons(ETH_P_IPV6))
 697                         return -EINVAL;
 698
 699                 if (!flow_key->ip.proto)
 700                         return -EINVAL;
 701
 702                 ipv6_key = nla_data(ovs_key);
 703                 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
 704                         return -EINVAL;
 705
 706                 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
 707                         return -EINVAL;
 708
 709                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
 710                         return -EINVAL;
 711
 712                 break;
 713
 714         case OVS_KEY_ATTR_TCP:
 715                 if (flow_key->ip.proto != IPPROTO_TCP)
 716                         return -EINVAL;
 717
 718                 return validate_tp_port(flow_key);
 719
 720         case OVS_KEY_ATTR_UDP:
 721                 if (flow_key->ip.proto != IPPROTO_UDP)
 722                         return -EINVAL;
 723
 724                 return validate_tp_port(flow_key);
 725
 726         default:
 727                 return -EINVAL;
 728         }
 729
 730         return 0;
 731 }
 732
 733 static int validate_userspace(const struct nlattr *attr)
 734 {
 735         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 736                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 737                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
 738         };
 739         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 740         int error;
 741
 742         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 743                                  attr, userspace_policy);
 744         if (error)
 745                 return error;
 746
 747         if (!a[OVS_USERSPACE_ATTR_PID] ||
 748             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 749                 return -EINVAL;
 750
 751         return 0;
 752 }
 753
 754 static int copy_action(const struct nlattr *from,
 755                       struct sw_flow_actions **sfa)
 756 {
 757         int totlen = NLA_ALIGN(from->nla_len);
 758         struct nlattr *to;
 759
 760         to = reserve_sfa_size(sfa, from->nla_len);
 761         if (IS_ERR(to))
 762                 return PTR_ERR(to);
 763
 764         memcpy(to, from, totlen);
 765         return 0;
 766 }
 767
 768 static int validate_and_copy_actions(const struct nlattr *attr,
 769                                 const struct sw_flow_key *key,
 770                                 int depth,
 771                                 struct sw_flow_actions **sfa)
 772 {
 773         const struct nlattr *a;
 774         int rem, err;
 775
 776         if (depth >= SAMPLE_ACTION_DEPTH)
 777                 return -EOVERFLOW;
 778
 779         nla_for_each_nested(a, attr, rem) {
 780                 /* Expected argument lengths, (u32)-1 for variable length. */
 781                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 782                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 783                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 784                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 785                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 786                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 787                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 788                 };
 789                 const struct ovs_action_push_vlan *vlan;
 790                 int type = nla_type(a);
 791                 bool skip_copy;
 792
 793                 if (type > OVS_ACTION_ATTR_MAX ||
 794                     (action_lens[type] != nla_len(a) &&
 795                      action_lens[type] != (u32)-1))
 796                         return -EINVAL;
 797
 798                 skip_copy = false;
 799                 switch (type) {
 800                 case OVS_ACTION_ATTR_UNSPEC:
 801                         return -EINVAL;
 802
 803                 case OVS_ACTION_ATTR_USERSPACE:
 804                         err = validate_userspace(a);
 805                         if (err)
 806                                 return err;
 807                         break;
 808
 809                 case OVS_ACTION_ATTR_OUTPUT:
 810                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 811                                 return -EINVAL;
 812                         break;
 813
 814
 815                 case OVS_ACTION_ATTR_POP_VLAN:
 816                         break;
 817
 818                 case OVS_ACTION_ATTR_PUSH_VLAN:
 819                         vlan = nla_data(a);
 820                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 821                                 return -EINVAL;
 822                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 823                                 return -EINVAL;
 824                         break;
 825
 826                 case OVS_ACTION_ATTR_SET:
 827                         err = validate_set(a, key, sfa, &skip_copy);
 828                         if (err)
 829                                 return err;
 830                         break;
 831
 832                 case OVS_ACTION_ATTR_SAMPLE:
 833                         err = validate_and_copy_sample(a, key, depth, sfa);
 834                         if (err)
 835                                 return err;
 836                         skip_copy = true;
 837                         break;
 838
 839                 default:
 840                         return -EINVAL;
 841                 }
 842                 if (!skip_copy) {
 843                         err = copy_action(a, sfa);
 844                         if (err)
 845                                 return err;
 846                 }
 847         }
 848
 849         if (rem > 0)
 850                 return -EINVAL;
 851
 852         return 0;
 853 }
 854
 855 static void clear_stats(struct sw_flow *flow)
 856 {
 857         flow->used = 0;
 858         flow->tcp_flags = 0;
 859         flow->packet_count = 0;
 860         flow->byte_count = 0;
 861 }
 862
 863 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 864 {
 865         struct ovs_header *ovs_header = info->userhdr;
 866         struct nlattr **a = info->attrs;
 867         struct sw_flow_actions *acts;
 868         struct sk_buff *packet;
 869         struct sw_flow *flow;
 870         struct datapath *dp;
 871         struct ethhdr *eth;
 872         int len;
 873         int err;
 874         int key_len;
 875
 876         err = -EINVAL;
 877         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 878             !a[OVS_PACKET_ATTR_ACTIONS])
 879                 goto err;
 880
 881         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 882         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 883         err = -ENOMEM;
 884         if (!packet)
 885                 goto err;
 886         skb_reserve(packet, NET_IP_ALIGN);
 887
 888         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 889
 890         skb_reset_mac_header(packet);
 891         eth = eth_hdr(packet);
 892
 893         /* Normally, setting the skb 'protocol' field would be handled by a
 894          * call to eth_type_trans(), but it assumes there's a sending
 895          * device, which we may not have. */
 896         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 897                 packet->protocol = eth->h_proto;
 898         else
 899                 packet->protocol = htons(ETH_P_802_2);
 900
 901         /* Build an sw_flow for sending this packet. */
 902         flow = ovs_flow_alloc();
 903         err = PTR_ERR(flow);
 904         if (IS_ERR(flow))
 905                 goto err_kfree_skb;
 906
 907         err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
 908         if (err)
 909                 goto err_flow_free;
 910
 911         err = ovs_flow_metadata_from_nlattrs(flow, key_len, a[OVS_PACKET_ATTR_KEY]);
 912         if (err)
 913                 goto err_flow_free;
 914         acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 915         err = PTR_ERR(acts);
 916         if (IS_ERR(acts))
 917                 goto err_flow_free;
 918
 919         err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
 920         rcu_assign_pointer(flow->sf_acts, acts);
 921         if (err)
 922                 goto err_flow_free;
 923
 924         OVS_CB(packet)->flow = flow;
 925         packet->priority = flow->key.phy.priority;
 926         skb_set_mark(packet, flow->key.phy.skb_mark);
 927
 928         rcu_read_lock();
 929         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 930         err = -ENODEV;
 931         if (!dp)
 932                 goto err_unlock;
 933
 934         local_bh_disable();
 935         err = ovs_execute_actions(dp, packet);
 936         local_bh_enable();
 937         rcu_read_unlock();
 938
 939         ovs_flow_free(flow);
 940         return err;
 941
 942 err_unlock:
 943         rcu_read_unlock();
 944 err_flow_free:
 945         ovs_flow_free(flow);
 946 err_kfree_skb:
 947         kfree_skb(packet);
 948 err:
 949         return err;
 950 }
 951
 952 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 953 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
 954         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 955 #else
 956         [OVS_PACKET_ATTR_PACKET] = { .minlen = ETH_HLEN },
 957 #endif
 958         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 959         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 960 };
 961
 962 static struct genl_ops dp_packet_genl_ops[] = {
 963         { .cmd = OVS_PACKET_CMD_EXECUTE,
 964           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 965           .policy = packet_policy,
 966           .doit = ovs_packet_cmd_execute
 967         }
 968 };
 969
 970 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 971 {
 972         int i;
 973         struct flow_table *table = ovsl_dereference(dp->table);
 974
 975         stats->n_flows = ovs_flow_tbl_count(table);
 976
 977         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 978         for_each_possible_cpu(i) {
 979                 const struct dp_stats_percpu *percpu_stats;
 980                 struct dp_stats_percpu local_stats;
 981                 unsigned int start;
 982
 983                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 984
 985                 do {
 986                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 987                         local_stats = *percpu_stats;
 988                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 989
 990                 stats->n_hit += local_stats.n_hit;
 991                 stats->n_missed += local_stats.n_missed;
 992                 stats->n_lost += local_stats.n_lost;
 993         }
 994 }
 995
 996 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 997         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 998         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 999         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1000 };
1001
1002 static struct genl_family dp_flow_genl_family = {
1003         .id = GENL_ID_GENERATE,
1004         .hdrsize = sizeof(struct ovs_header),
1005         .name = OVS_FLOW_FAMILY,
1006         .version = OVS_FLOW_VERSION,
1007         .maxattr = OVS_FLOW_ATTR_MAX,
1008          SET_NETNSOK
1009 };
1010
1011 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
1012         .name = OVS_FLOW_MCGROUP
1013 };
1014
1015 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
1016 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1017 {
1018         const struct nlattr *a;
1019         struct nlattr *start;
1020         int err = 0, rem;
1021
1022         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1023         if (!start)
1024                 return -EMSGSIZE;
1025
1026         nla_for_each_nested(a, attr, rem) {
1027                 int type = nla_type(a);
1028                 struct nlattr *st_sample;
1029
1030                 switch (type) {
1031                 case OVS_SAMPLE_ATTR_PROBABILITY:
1032                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
1033                                 return -EMSGSIZE;
1034                         break;
1035                 case OVS_SAMPLE_ATTR_ACTIONS:
1036                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1037                         if (!st_sample)
1038                                 return -EMSGSIZE;
1039                         err = actions_to_attr(nla_data(a), nla_len(a), skb);
1040                         if (err)
1041                                 return err;
1042                         nla_nest_end(skb, st_sample);
1043                         break;
1044                 }
1045         }
1046
1047         nla_nest_end(skb, start);
1048         return err;
1049 }
1050
1051 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1052 {
1053         const struct nlattr *ovs_key = nla_data(a);
1054         int key_type = nla_type(ovs_key);
1055         struct nlattr *start;
1056         int err;
1057
1058         switch (key_type) {
1059         case OVS_KEY_ATTR_IPV4_TUNNEL:
1060                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1061                 if (!start)
1062                         return -EMSGSIZE;
1063
1064                 err = ipv4_tun_to_nlattr(skb, nla_data(ovs_key));
1065                 if (err)
1066                         return err;
1067                 nla_nest_end(skb, start);
1068                 break;
1069         default:
1070                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1071                         return -EMSGSIZE;
1072                 break;
1073         }
1074
1075         return 0;
1076 }
1077
1078 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1079 {
1080         const struct nlattr *a;
1081         int rem, err;
1082
1083         nla_for_each_attr(a, attr, len, rem) {
1084                 int type = nla_type(a);
1085
1086                 switch (type) {
1087                 case OVS_ACTION_ATTR_SET:
1088                         err = set_action_to_attr(a, skb);
1089                         if (err)
1090                                 return err;
1091                         break;
1092
1093                 case OVS_ACTION_ATTR_SAMPLE:
1094                         err = sample_action_to_attr(a, skb);
1095                         if (err)
1096                                 return err;
1097                         break;
1098                 default:
1099                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
1100                                 return -EMSGSIZE;
1101                         break;
1102                 }
1103         }
1104
1105         return 0;
1106 }
1107
1108 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
1109 {
1110         return NLMSG_ALIGN(sizeof(struct ovs_header))
1111                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
1112                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
1113                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
1114                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
1115                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
1116 }
1117
1118 /* Called with ovs_mutex. */
1119 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1120                                   struct sk_buff *skb, u32 portid,
1121                                   u32 seq, u32 flags, u8 cmd)
1122 {
1123         const int skb_orig_len = skb->len;
1124         const struct sw_flow_actions *sf_acts;
1125         struct nlattr *start;
1126         struct ovs_flow_stats stats;
1127         struct ovs_header *ovs_header;
1128         struct nlattr *nla;
1129         unsigned long used;
1130         u8 tcp_flags;
1131         int err;
1132
1133         sf_acts = ovsl_dereference(flow->sf_acts);
1134
1135         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1136         if (!ovs_header)
1137                 return -EMSGSIZE;
1138
1139         ovs_header->dp_ifindex = get_dpifindex(dp);
1140
1141         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1142         if (!nla)
1143                 goto nla_put_failure;
1144         err = ovs_flow_to_nlattrs(&flow->key, skb);
1145         if (err)
1146                 goto error;
1147         nla_nest_end(skb, nla);
1148
1149         spin_lock_bh(&flow->lock);
1150         used = flow->used;
1151         stats.n_packets = flow->packet_count;
1152         stats.n_bytes = flow->byte_count;
1153         tcp_flags = flow->tcp_flags;
1154         spin_unlock_bh(&flow->lock);
1155
1156         if (used &&
1157             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1158                 goto nla_put_failure;
1159
1160         if (stats.n_packets &&
1161             nla_put(skb, OVS_FLOW_ATTR_STATS,
1162                     sizeof(struct ovs_flow_stats), &stats))
1163                 goto nla_put_failure;
1164
1165         if (tcp_flags &&
1166             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1167                 goto nla_put_failure;
1168
1169         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1170          * this is the first flow to be dumped into 'skb'.  This is unusual for
1171          * Netlink but individual action lists can be longer than
1172          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1173          * The userspace caller can always fetch the actions separately if it
1174          * really wants them.  (Most userspace callers in fact don't care.)
1175          *
1176          * This can only fail for dump operations because the skb is always
1177          * properly sized for single flows.
1178          */
1179         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1180         if (start) {
1181                 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1182                 if (!err)
1183                         nla_nest_end(skb, start);
1184                 else {
1185                         if (skb_orig_len)
1186                                 goto error;
1187
1188                         nla_nest_cancel(skb, start);
1189                 }
1190         } else if (skb_orig_len)
1191                 goto nla_put_failure;
1192
1193         return genlmsg_end(skb, ovs_header);
1194
1195 nla_put_failure:
1196         err = -EMSGSIZE;
1197 error:
1198         genlmsg_cancel(skb, ovs_header);
1199         return err;
1200 }
1201
1202 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1203 {
1204         const struct sw_flow_actions *sf_acts;
1205
1206         sf_acts = ovsl_dereference(flow->sf_acts);
1207
1208         return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
1209 }
1210
1211 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1212                                                struct datapath *dp,
1213                                                u32 portid, u32 seq, u8 cmd)
1214 {
1215         struct sk_buff *skb;
1216         int retval;
1217
1218         skb = ovs_flow_cmd_alloc_info(flow);
1219         if (!skb)
1220                 return ERR_PTR(-ENOMEM);
1221
1222         retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1223         BUG_ON(retval < 0);
1224         return skb;
1225 }
1226
1227 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1228 {
1229         struct nlattr **a = info->attrs;
1230         struct ovs_header *ovs_header = info->userhdr;
1231         struct sw_flow_key key;
1232         struct sw_flow *flow;
1233         struct sk_buff *reply;
1234         struct datapath *dp;
1235         struct flow_table *table;
1236         struct sw_flow_actions *acts = NULL;
1237         int error;
1238         int key_len;
1239
1240         /* Extract key. */
1241         error = -EINVAL;
1242         if (!a[OVS_FLOW_ATTR_KEY])
1243                 goto error;
1244         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1245         if (error)
1246                 goto error;
1247
1248         /* Validate actions. */
1249         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1250                 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1251                 error = PTR_ERR(acts);
1252                 if (IS_ERR(acts))
1253                         goto error;
1254
1255                 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0, &acts);
1256                 if (error)
1257                         goto err_kfree;
1258         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1259                 error = -EINVAL;
1260                 goto error;
1261         }
1262
1263         ovs_lock();
1264         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1265         error = -ENODEV;
1266         if (!dp)
1267                 goto err_unlock_ovs;
1268
1269         table = ovsl_dereference(dp->table);
1270         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1271         if (!flow) {
1272                 /* Bail out if we're not allowed to create a new flow. */
1273                 error = -ENOENT;
1274                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1275                         goto err_unlock_ovs;
1276
1277                 /* Expand table, if necessary, to make room. */
1278                 if (ovs_flow_tbl_need_to_expand(table)) {
1279                         struct flow_table *new_table;
1280
1281                         new_table = ovs_flow_tbl_expand(table);
1282                         if (!IS_ERR(new_table)) {
1283                                 rcu_assign_pointer(dp->table, new_table);
1284                                 ovs_flow_tbl_deferred_destroy(table);
1285                                 table = ovsl_dereference(dp->table);
1286                         }
1287                 }
1288
1289                 /* Allocate flow. */
1290                 flow = ovs_flow_alloc();
1291                 if (IS_ERR(flow)) {
1292                         error = PTR_ERR(flow);
1293                         goto err_unlock_ovs;
1294                 }
1295                 clear_stats(flow);
1296
1297                 rcu_assign_pointer(flow->sf_acts, acts);
1298
1299                 /* Put flow in bucket. */
1300                 ovs_flow_tbl_insert(table, flow, &key, key_len);
1301
1302                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1303                                                 info->snd_seq,
1304                                                 OVS_FLOW_CMD_NEW);
1305         } else {
1306                 /* We found a matching flow. */
1307                 struct sw_flow_actions *old_acts;
1308
1309                 /* Bail out if we're not allowed to modify an existing flow.
1310                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1311                  * because Generic Netlink treats the latter as a dump
1312                  * request.  We also accept NLM_F_EXCL in case that bug ever
1313                  * gets fixed.
1314                  */
1315                 error = -EEXIST;
1316                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1317                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1318                         goto err_unlock_ovs;
1319
1320                 /* Update actions. */
1321                 old_acts = ovsl_dereference(flow->sf_acts);
1322                 rcu_assign_pointer(flow->sf_acts, acts);
1323                 ovs_flow_deferred_free_acts(old_acts);
1324
1325                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1326                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1327
1328                 /* Clear stats. */
1329                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1330                         spin_lock_bh(&flow->lock);
1331                         clear_stats(flow);
1332                         spin_unlock_bh(&flow->lock);
1333                 }
1334         }
1335         ovs_unlock();
1336
1337         if (!IS_ERR(reply))
1338                 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1339         else
1340                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1341                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1342         return 0;
1343
1344 err_unlock_ovs:
1345         ovs_unlock();
1346 err_kfree:
1347         kfree(acts);
1348 error:
1349         return error;
1350 }
1351
1352 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1353 {
1354         struct nlattr **a = info->attrs;
1355         struct ovs_header *ovs_header = info->userhdr;
1356         struct sw_flow_key key;
1357         struct sk_buff *reply;
1358         struct sw_flow *flow;
1359         struct datapath *dp;
1360         struct flow_table *table;
1361         int err;
1362         int key_len;
1363
1364         if (!a[OVS_FLOW_ATTR_KEY])
1365                 return -EINVAL;
1366         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1367         if (err)
1368                 return err;
1369
1370         ovs_lock();
1371         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1372         if (!dp) {
1373                 err = -ENODEV;
1374                 goto unlock;
1375         }
1376
1377         table = ovsl_dereference(dp->table);
1378         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1379         if (!flow) {
1380                 err = -ENOENT;
1381                 goto unlock;
1382         }
1383
1384         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1385                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1386         if (IS_ERR(reply)) {
1387                 err = PTR_ERR(reply);
1388                 goto unlock;
1389         }
1390
1391         ovs_unlock();
1392         return genlmsg_reply(reply, info);
1393 unlock:
1394         ovs_unlock();
1395         return err;
1396 }
1397
1398 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1399 {
1400         struct nlattr **a = info->attrs;
1401         struct ovs_header *ovs_header = info->userhdr;
1402         struct sw_flow_key key;
1403         struct sk_buff *reply;
1404         struct sw_flow *flow;
1405         struct datapath *dp;
1406         struct flow_table *table;
1407         int err;
1408         int key_len;
1409
1410         ovs_lock();
1411         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1412         if (!dp) {
1413                 err = -ENODEV;
1414                 goto unlock;
1415         }
1416
1417         if (!a[OVS_FLOW_ATTR_KEY]) {
1418                 err = flush_flows(dp);
1419                 goto unlock;
1420         }
1421         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1422         if (err)
1423                 goto unlock;
1424
1425         table = ovsl_dereference(dp->table);
1426         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1427         if (!flow) {
1428                 err = -ENOENT;
1429                 goto unlock;
1430         }
1431
1432         reply = ovs_flow_cmd_alloc_info(flow);
1433         if (!reply) {
1434                 err = -ENOMEM;
1435                 goto unlock;
1436         }
1437
1438         ovs_flow_tbl_remove(table, flow);
1439
1440         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1441                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1442         BUG_ON(err < 0);
1443
1444         ovs_flow_deferred_free(flow);
1445         ovs_unlock();
1446
1447         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1448         return 0;
1449 unlock:
1450         ovs_unlock();
1451         return err;
1452 }
1453
1454 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1455 {
1456         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1457         struct datapath *dp;
1458         struct flow_table *table;
1459
1460         ovs_lock();
1461         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1462         if (!dp) {
1463                 ovs_unlock();
1464                 return -ENODEV;
1465         }
1466
1467         table = ovsl_dereference(dp->table);
1468
1469         for (;;) {
1470                 struct sw_flow *flow;
1471                 u32 bucket, obj;
1472
1473                 bucket = cb->args[0];
1474                 obj = cb->args[1];
1475                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1476                 if (!flow)
1477                         break;
1478
1479                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1480                                            NETLINK_CB(cb->skb).portid,
1481                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1482                                            OVS_FLOW_CMD_NEW) < 0)
1483                         break;
1484
1485                 cb->args[0] = bucket;
1486                 cb->args[1] = obj;
1487         }
1488         ovs_unlock();
1489         return skb->len;
1490 }
1491
1492 static struct genl_ops dp_flow_genl_ops[] = {
1493         { .cmd = OVS_FLOW_CMD_NEW,
1494           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1495           .policy = flow_policy,
1496           .doit = ovs_flow_cmd_new_or_set
1497         },
1498         { .cmd = OVS_FLOW_CMD_DEL,
1499           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1500           .policy = flow_policy,
1501           .doit = ovs_flow_cmd_del
1502         },
1503         { .cmd = OVS_FLOW_CMD_GET,
1504           .flags = 0,               /* OK for unprivileged users. */
1505           .policy = flow_policy,
1506           .doit = ovs_flow_cmd_get,
1507           .dumpit = ovs_flow_cmd_dump
1508         },
1509         { .cmd = OVS_FLOW_CMD_SET,
1510           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1511           .policy = flow_policy,
1512           .doit = ovs_flow_cmd_new_or_set,
1513         },
1514 };
1515
1516 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1517 #ifdef HAVE_NLA_NUL_STRING
1518         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1519 #endif
1520         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1521 };
1522
1523 static struct genl_family dp_datapath_genl_family = {
1524         .id = GENL_ID_GENERATE,
1525         .hdrsize = sizeof(struct ovs_header),
1526         .name = OVS_DATAPATH_FAMILY,
1527         .version = OVS_DATAPATH_VERSION,
1528         .maxattr = OVS_DP_ATTR_MAX,
1529          SET_NETNSOK
1530 };
1531
1532 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1533         .name = OVS_DATAPATH_MCGROUP
1534 };
1535
1536 static size_t ovs_dp_cmd_msg_size(void)
1537 {
1538         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1539
1540         msgsize += nla_total_size(IFNAMSIZ);
1541         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1542
1543         return msgsize;
1544 }
1545
1546 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1547                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1548 {
1549         struct ovs_header *ovs_header;
1550         struct ovs_dp_stats dp_stats;
1551         int err;
1552
1553         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1554                                    flags, cmd);
1555         if (!ovs_header)
1556                 goto error;
1557
1558         ovs_header->dp_ifindex = get_dpifindex(dp);
1559
1560         rcu_read_lock();
1561         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1562         rcu_read_unlock();
1563         if (err)
1564                 goto nla_put_failure;
1565
1566         get_dp_stats(dp, &dp_stats);
1567         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1568                 goto nla_put_failure;
1569
1570         return genlmsg_end(skb, ovs_header);
1571
1572 nla_put_failure:
1573         genlmsg_cancel(skb, ovs_header);
1574 error:
1575         return -EMSGSIZE;
1576 }
1577
1578 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1579                                              u32 seq, u8 cmd)
1580 {
1581         struct sk_buff *skb;
1582         int retval;
1583
1584         skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1585         if (!skb)
1586                 return ERR_PTR(-ENOMEM);
1587
1588         retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1589         if (retval < 0) {
1590                 kfree_skb(skb);
1591                 return ERR_PTR(retval);
1592         }
1593         return skb;
1594 }
1595
1596 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1597 {
1598         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1599 }
1600
1601 /* Called with ovs_mutex. */
1602 static struct datapath *lookup_datapath(struct net *net,
1603                                         struct ovs_header *ovs_header,
1604                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1605 {
1606         struct datapath *dp;
1607
1608         if (!a[OVS_DP_ATTR_NAME])
1609                 dp = get_dp(net, ovs_header->dp_ifindex);
1610         else {
1611                 struct vport *vport;
1612
1613                 rcu_read_lock();
1614                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1615                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1616                 rcu_read_unlock();
1617         }
1618         return dp ? dp : ERR_PTR(-ENODEV);
1619 }
1620
1621 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1622 {
1623         struct nlattr **a = info->attrs;
1624         struct vport_parms parms;
1625         struct sk_buff *reply;
1626         struct datapath *dp;
1627         struct vport *vport;
1628         struct ovs_net *ovs_net;
1629         int err, i;
1630
1631         err = -EINVAL;
1632         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1633                 goto err;
1634
1635         err = ovs_dp_cmd_validate(a);
1636         if (err)
1637                 goto err;
1638
1639         ovs_lock();
1640
1641         err = -ENOMEM;
1642         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1643         if (dp == NULL)
1644                 goto err_unlock_ovs;
1645
1646         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1647
1648         /* Allocate table. */
1649         err = -ENOMEM;
1650         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1651         if (!dp->table)
1652                 goto err_free_dp;
1653
1654         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1655         if (!dp->stats_percpu) {
1656                 err = -ENOMEM;
1657                 goto err_destroy_table;
1658         }
1659
1660         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1661                             GFP_KERNEL);
1662         if (!dp->ports) {
1663                 err = -ENOMEM;
1664                 goto err_destroy_percpu;
1665         }
1666
1667         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1668                 INIT_HLIST_HEAD(&dp->ports[i]);
1669
1670         /* Set up our datapath device. */
1671         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1672         parms.type = OVS_VPORT_TYPE_INTERNAL;
1673         parms.options = NULL;
1674         parms.dp = dp;
1675         parms.port_no = OVSP_LOCAL;
1676         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1677
1678         vport = new_vport(&parms);
1679         if (IS_ERR(vport)) {
1680                 err = PTR_ERR(vport);
1681                 if (err == -EBUSY)
1682                         err = -EEXIST;
1683
1684                 goto err_destroy_ports_array;
1685         }
1686
1687         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1688                                       info->snd_seq, OVS_DP_CMD_NEW);
1689         err = PTR_ERR(reply);
1690         if (IS_ERR(reply))
1691                 goto err_destroy_local_port;
1692
1693         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1694         list_add_tail(&dp->list_node, &ovs_net->dps);
1695
1696         ovs_unlock();
1697
1698         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1699         return 0;
1700
1701 err_destroy_local_port:
1702         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1703 err_destroy_ports_array:
1704         kfree(dp->ports);
1705 err_destroy_percpu:
1706         free_percpu(dp->stats_percpu);
1707 err_destroy_table:
1708         ovs_flow_tbl_destroy(ovsl_dereference(dp->table));
1709 err_free_dp:
1710         release_net(ovs_dp_get_net(dp));
1711         kfree(dp);
1712 err_unlock_ovs:
1713         ovs_unlock();
1714 err:
1715         return err;
1716 }
1717
1718 /* Called with ovs_mutex. */
1719 static void __dp_destroy(struct datapath *dp)
1720 {
1721         int i;
1722
1723         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1724                 struct vport *vport;
1725                 struct hlist_node *n;
1726
1727                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1728                         if (vport->port_no != OVSP_LOCAL)
1729                                 ovs_dp_detach_port(vport);
1730         }
1731
1732         list_del(&dp->list_node);
1733
1734         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1735          * all port in datapath are destroyed first before freeing datapath.
1736          */
1737         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1738
1739         call_rcu(&dp->rcu, destroy_dp_rcu);
1740 }
1741
1742 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1743 {
1744         struct sk_buff *reply;
1745         struct datapath *dp;
1746         int err;
1747
1748         err = ovs_dp_cmd_validate(info->attrs);
1749         if (err)
1750                 return err;
1751
1752         ovs_lock();
1753         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1754         err = PTR_ERR(dp);
1755         if (IS_ERR(dp))
1756                 goto unlock;
1757
1758         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1759                                       info->snd_seq, OVS_DP_CMD_DEL);
1760         err = PTR_ERR(reply);
1761         if (IS_ERR(reply))
1762                 goto unlock;
1763
1764         __dp_destroy(dp);
1765         ovs_unlock();
1766
1767         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1768
1769         return 0;
1770 unlock:
1771         ovs_unlock();
1772         return err;
1773 }
1774
1775 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1776 {
1777         struct sk_buff *reply;
1778         struct datapath *dp;
1779         int err;
1780
1781         err = ovs_dp_cmd_validate(info->attrs);
1782         if (err)
1783                 return err;
1784
1785         ovs_lock();
1786         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1787         err = PTR_ERR(dp);
1788         if (IS_ERR(dp))
1789                 goto unlock;
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                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1796                                 ovs_dp_datapath_multicast_group.id, err);
1797                 err = 0;
1798                 goto unlock;
1799         }
1800
1801         ovs_unlock();
1802         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1803
1804         return 0;
1805 unlock:
1806         ovs_unlock();
1807         return err;
1808 }
1809
1810 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1811 {
1812         struct sk_buff *reply;
1813         struct datapath *dp;
1814         int err;
1815
1816         err = ovs_dp_cmd_validate(info->attrs);
1817         if (err)
1818                 return err;
1819
1820         ovs_lock();
1821         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1822         if (IS_ERR(dp)) {
1823                 err = PTR_ERR(dp);
1824                 goto unlock;
1825         }
1826
1827         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1828                                       info->snd_seq, OVS_DP_CMD_NEW);
1829         if (IS_ERR(reply)) {
1830                 err = PTR_ERR(reply);
1831                 goto unlock;
1832         }
1833
1834         ovs_unlock();
1835         return genlmsg_reply(reply, info);
1836
1837 unlock:
1838         ovs_unlock();
1839         return err;
1840 }
1841
1842 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1843 {
1844         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1845         struct datapath *dp;
1846         int skip = cb->args[0];
1847         int i = 0;
1848
1849         ovs_lock();
1850         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1851                 if (i >= skip &&
1852                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1853                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1854                                          OVS_DP_CMD_NEW) < 0)
1855                         break;
1856                 i++;
1857         }
1858         ovs_unlock();
1859
1860         cb->args[0] = i;
1861
1862         return skb->len;
1863 }
1864
1865 static struct genl_ops dp_datapath_genl_ops[] = {
1866         { .cmd = OVS_DP_CMD_NEW,
1867           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1868           .policy = datapath_policy,
1869           .doit = ovs_dp_cmd_new
1870         },
1871         { .cmd = OVS_DP_CMD_DEL,
1872           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1873           .policy = datapath_policy,
1874           .doit = ovs_dp_cmd_del
1875         },
1876         { .cmd = OVS_DP_CMD_GET,
1877           .flags = 0,               /* OK for unprivileged users. */
1878           .policy = datapath_policy,
1879           .doit = ovs_dp_cmd_get,
1880           .dumpit = ovs_dp_cmd_dump
1881         },
1882         { .cmd = OVS_DP_CMD_SET,
1883           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1884           .policy = datapath_policy,
1885           .doit = ovs_dp_cmd_set,
1886         },
1887 };
1888
1889 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1890 #ifdef HAVE_NLA_NUL_STRING
1891         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1892         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1893 #else
1894         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1895 #endif
1896         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1897         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1898         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1899         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1900 };
1901
1902 static struct genl_family dp_vport_genl_family = {
1903         .id = GENL_ID_GENERATE,
1904         .hdrsize = sizeof(struct ovs_header),
1905         .name = OVS_VPORT_FAMILY,
1906         .version = OVS_VPORT_VERSION,
1907         .maxattr = OVS_VPORT_ATTR_MAX,
1908          SET_NETNSOK
1909 };
1910
1911 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1912         .name = OVS_VPORT_MCGROUP
1913 };
1914
1915 /* Called with ovs_mutex or RCU read lock. */
1916 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1917                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1918 {
1919         struct ovs_header *ovs_header;
1920         struct ovs_vport_stats vport_stats;
1921         int err;
1922
1923         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1924                                  flags, cmd);
1925         if (!ovs_header)
1926                 return -EMSGSIZE;
1927
1928         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1929
1930         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1931             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1932             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1933             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1934                 goto nla_put_failure;
1935
1936         ovs_vport_get_stats(vport, &vport_stats);
1937         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1938                     &vport_stats))
1939                 goto nla_put_failure;
1940
1941         err = ovs_vport_get_options(vport, skb);
1942         if (err == -EMSGSIZE)
1943                 goto error;
1944
1945         return genlmsg_end(skb, ovs_header);
1946
1947 nla_put_failure:
1948         err = -EMSGSIZE;
1949 error:
1950         genlmsg_cancel(skb, ovs_header);
1951         return err;
1952 }
1953
1954 /* Called with ovs_mutex or RCU read lock. */
1955 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1956                                          u32 seq, u8 cmd)
1957 {
1958         struct sk_buff *skb;
1959         int retval;
1960
1961         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1962         if (!skb)
1963                 return ERR_PTR(-ENOMEM);
1964
1965         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1966         BUG_ON(retval < 0);
1967
1968         return skb;
1969 }
1970
1971 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1972 {
1973         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1974 }
1975
1976 /* Called with ovs_mutex or RCU read lock. */
1977 static struct vport *lookup_vport(struct net *net,
1978                                   struct ovs_header *ovs_header,
1979                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1980 {
1981         struct datapath *dp;
1982         struct vport *vport;
1983
1984         if (a[OVS_VPORT_ATTR_NAME]) {
1985                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1986                 if (!vport)
1987                         return ERR_PTR(-ENODEV);
1988                 if (ovs_header->dp_ifindex &&
1989                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1990                         return ERR_PTR(-ENODEV);
1991                 return vport;
1992         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1993                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1994
1995                 if (port_no >= DP_MAX_PORTS)
1996                         return ERR_PTR(-EFBIG);
1997
1998                 dp = get_dp(net, ovs_header->dp_ifindex);
1999                 if (!dp)
2000                         return ERR_PTR(-ENODEV);
2001
2002                 vport = ovs_vport_ovsl_rcu(dp, port_no);
2003                 if (!vport)
2004                         return ERR_PTR(-ENODEV);
2005                 return vport;
2006         } else
2007                 return ERR_PTR(-EINVAL);
2008 }
2009
2010 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2011 {
2012         struct nlattr **a = info->attrs;
2013         struct ovs_header *ovs_header = info->userhdr;
2014         struct vport_parms parms;
2015         struct sk_buff *reply;
2016         struct vport *vport;
2017         struct datapath *dp;
2018         u32 port_no;
2019         int err;
2020
2021         err = -EINVAL;
2022         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2023             !a[OVS_VPORT_ATTR_UPCALL_PID])
2024                 goto exit;
2025
2026         err = ovs_vport_cmd_validate(a);
2027         if (err)
2028                 goto exit;
2029
2030         ovs_lock();
2031         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2032         err = -ENODEV;
2033         if (!dp)
2034                 goto exit_unlock;
2035
2036         if (a[OVS_VPORT_ATTR_PORT_NO]) {
2037                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2038
2039                 err = -EFBIG;
2040                 if (port_no >= DP_MAX_PORTS)
2041                         goto exit_unlock;
2042
2043                 vport = ovs_vport_ovsl(dp, port_no);
2044                 err = -EBUSY;
2045                 if (vport)
2046                         goto exit_unlock;
2047         } else {
2048                 for (port_no = 1; ; port_no++) {
2049                         if (port_no >= DP_MAX_PORTS) {
2050                                 err = -EFBIG;
2051                                 goto exit_unlock;
2052                         }
2053                         vport = ovs_vport_ovsl(dp, port_no);
2054                         if (!vport)
2055                                 break;
2056                 }
2057         }
2058
2059         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2060         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2061         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2062         parms.dp = dp;
2063         parms.port_no = port_no;
2064         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2065
2066         vport = new_vport(&parms);
2067         err = PTR_ERR(vport);
2068         if (IS_ERR(vport))
2069                 goto exit_unlock;
2070
2071         err = 0;
2072         if (a[OVS_VPORT_ATTR_STATS])
2073                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2074
2075         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
2076                                          OVS_VPORT_CMD_NEW);
2077         if (IS_ERR(reply)) {
2078                 err = PTR_ERR(reply);
2079                 ovs_dp_detach_port(vport);
2080                 goto exit_unlock;
2081         }
2082
2083         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2084
2085 exit_unlock:
2086         ovs_unlock();
2087 exit:
2088         return err;
2089 }
2090
2091 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2092 {
2093         struct nlattr **a = info->attrs;
2094         struct sk_buff *reply;
2095         struct vport *vport;
2096         int err;
2097
2098         err = ovs_vport_cmd_validate(a);
2099         if (err)
2100                 goto exit;
2101
2102         ovs_lock();
2103         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2104         err = PTR_ERR(vport);
2105         if (IS_ERR(vport))
2106                 goto exit_unlock;
2107
2108         err = 0;
2109         if (a[OVS_VPORT_ATTR_TYPE] &&
2110             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
2111                 err = -EINVAL;
2112
2113         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2114         if (!reply) {
2115                 err = -ENOMEM;
2116                 goto exit_unlock;
2117         }
2118
2119         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
2120                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2121         if (err)
2122                 goto exit_free;
2123
2124         if (a[OVS_VPORT_ATTR_STATS])
2125                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2126
2127         if (a[OVS_VPORT_ATTR_UPCALL_PID])
2128                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2129
2130         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2131                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2132         BUG_ON(err < 0);
2133
2134         ovs_unlock();
2135         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2136         return 0;
2137
2138 exit_free:
2139         kfree_skb(reply);
2140 exit_unlock:
2141         ovs_unlock();
2142 exit:
2143         return err;
2144 }
2145
2146 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2147 {
2148         struct nlattr **a = info->attrs;
2149         struct sk_buff *reply;
2150         struct vport *vport;
2151         int err;
2152
2153         err = ovs_vport_cmd_validate(a);
2154         if (err)
2155                 goto exit;
2156
2157         ovs_lock();
2158         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2159         err = PTR_ERR(vport);
2160         if (IS_ERR(vport))
2161                 goto exit_unlock;
2162
2163         if (vport->port_no == OVSP_LOCAL) {
2164                 err = -EINVAL;
2165                 goto exit_unlock;
2166         }
2167
2168         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2169                                          info->snd_seq, OVS_VPORT_CMD_DEL);
2170         err = PTR_ERR(reply);
2171         if (IS_ERR(reply))
2172                 goto exit_unlock;
2173
2174         err = 0;
2175         ovs_dp_detach_port(vport);
2176
2177         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2178
2179 exit_unlock:
2180         ovs_unlock();
2181 exit:
2182         return err;
2183 }
2184
2185 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2186 {
2187         struct nlattr **a = info->attrs;
2188         struct ovs_header *ovs_header = info->userhdr;
2189         struct sk_buff *reply;
2190         struct vport *vport;
2191         int err;
2192
2193         err = ovs_vport_cmd_validate(a);
2194         if (err)
2195                 goto exit;
2196
2197         rcu_read_lock();
2198         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2199         err = PTR_ERR(vport);
2200         if (IS_ERR(vport))
2201                 goto exit_unlock;
2202
2203         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2204                                          info->snd_seq, OVS_VPORT_CMD_NEW);
2205         err = PTR_ERR(reply);
2206         if (IS_ERR(reply))
2207                 goto exit_unlock;
2208
2209         rcu_read_unlock();
2210
2211         return genlmsg_reply(reply, info);
2212
2213 exit_unlock:
2214         rcu_read_unlock();
2215 exit:
2216         return err;
2217 }
2218
2219 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2220 {
2221         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2222         struct datapath *dp;
2223         int bucket = cb->args[0], skip = cb->args[1];
2224         int i, j = 0;
2225
2226         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2227         if (!dp)
2228                 return -ENODEV;
2229
2230         rcu_read_lock();
2231         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2232                 struct vport *vport;
2233
2234                 j = 0;
2235                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2236                         if (j >= skip &&
2237                             ovs_vport_cmd_fill_info(vport, skb,
2238                                                     NETLINK_CB(cb->skb).portid,
2239                                                     cb->nlh->nlmsg_seq,
2240                                                     NLM_F_MULTI,
2241                                                     OVS_VPORT_CMD_NEW) < 0)
2242                                 goto out;
2243
2244                         j++;
2245                 }
2246                 skip = 0;
2247         }
2248 out:
2249         rcu_read_unlock();
2250
2251         cb->args[0] = i;
2252         cb->args[1] = j;
2253
2254         return skb->len;
2255 }
2256
2257 static struct genl_ops dp_vport_genl_ops[] = {
2258         { .cmd = OVS_VPORT_CMD_NEW,
2259           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2260           .policy = vport_policy,
2261           .doit = ovs_vport_cmd_new
2262         },
2263         { .cmd = OVS_VPORT_CMD_DEL,
2264           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2265           .policy = vport_policy,
2266           .doit = ovs_vport_cmd_del
2267         },
2268         { .cmd = OVS_VPORT_CMD_GET,
2269           .flags = 0,               /* OK for unprivileged users. */
2270           .policy = vport_policy,
2271           .doit = ovs_vport_cmd_get,
2272           .dumpit = ovs_vport_cmd_dump
2273         },
2274         { .cmd = OVS_VPORT_CMD_SET,
2275           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2276           .policy = vport_policy,
2277           .doit = ovs_vport_cmd_set,
2278         },
2279 };
2280
2281 struct genl_family_and_ops {
2282         struct genl_family *family;
2283         struct genl_ops *ops;
2284         int n_ops;
2285         struct genl_multicast_group *group;
2286 };
2287
2288 static const struct genl_family_and_ops dp_genl_families[] = {
2289         { &dp_datapath_genl_family,
2290           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2291           &ovs_dp_datapath_multicast_group },
2292         { &dp_vport_genl_family,
2293           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2294           &ovs_dp_vport_multicast_group },
2295         { &dp_flow_genl_family,
2296           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2297           &ovs_dp_flow_multicast_group },
2298         { &dp_packet_genl_family,
2299           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2300           NULL },
2301 };
2302
2303 static void dp_unregister_genl(int n_families)
2304 {
2305         int i;
2306
2307         for (i = 0; i < n_families; i++)
2308                 genl_unregister_family(dp_genl_families[i].family);
2309 }
2310
2311 static int dp_register_genl(void)
2312 {
2313         int n_registered;
2314         int err;
2315         int i;
2316
2317         n_registered = 0;
2318         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2319                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2320
2321                 err = genl_register_family_with_ops(f->family, f->ops,
2322                                                     f->n_ops);
2323                 if (err)
2324                         goto error;
2325                 n_registered++;
2326
2327                 if (f->group) {
2328                         err = genl_register_mc_group(f->family, f->group);
2329                         if (err)
2330                                 goto error;
2331                 }
2332         }
2333
2334         return 0;
2335
2336 error:
2337         dp_unregister_genl(n_registered);
2338         return err;
2339 }
2340
2341 static void rehash_flow_table(struct work_struct *work)
2342 {
2343         struct datapath *dp;
2344         struct net *net;
2345
2346         ovs_lock();
2347         rtnl_lock();
2348         for_each_net(net) {
2349                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2350
2351                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2352                         struct flow_table *old_table = ovsl_dereference(dp->table);
2353                         struct flow_table *new_table;
2354
2355                         new_table = ovs_flow_tbl_rehash(old_table);
2356                         if (!IS_ERR(new_table)) {
2357                                 rcu_assign_pointer(dp->table, new_table);
2358                                 ovs_flow_tbl_deferred_destroy(old_table);
2359                         }
2360                 }
2361         }
2362         rtnl_unlock();
2363         ovs_unlock();
2364         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2365 }
2366
2367 static int __net_init ovs_init_net(struct net *net)
2368 {
2369         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2370
2371         INIT_LIST_HEAD(&ovs_net->dps);
2372         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2373         return 0;
2374 }
2375
2376 static void __net_exit ovs_exit_net(struct net *net)
2377 {
2378         struct datapath *dp, *dp_next;
2379         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2380
2381         ovs_lock();
2382         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2383                 __dp_destroy(dp);
2384         ovs_unlock();
2385
2386         cancel_work_sync(&ovs_net->dp_notify_work);
2387 }
2388
2389 static struct pernet_operations ovs_net_ops = {
2390         .init = ovs_init_net,
2391         .exit = ovs_exit_net,
2392         .id   = &ovs_net_id,
2393         .size = sizeof(struct ovs_net),
2394 };
2395
2396 static int __init dp_init(void)
2397 {
2398         int err;
2399
2400         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2401
2402         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2403                 VERSION);
2404
2405         err = ovs_workqueues_init();
2406         if (err)
2407                 goto error;
2408
2409         err = ovs_flow_init();
2410         if (err)
2411                 goto error_wq;
2412
2413         err = ovs_vport_init();
2414         if (err)
2415                 goto error_flow_exit;
2416
2417         err = register_pernet_device(&ovs_net_ops);
2418         if (err)
2419                 goto error_vport_exit;
2420
2421         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2422         if (err)
2423                 goto error_netns_exit;
2424
2425         err = dp_register_genl();
2426         if (err < 0)
2427                 goto error_unreg_notifier;
2428
2429         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2430
2431         return 0;
2432
2433 error_unreg_notifier:
2434         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2435 error_netns_exit:
2436         unregister_pernet_device(&ovs_net_ops);
2437 error_vport_exit:
2438         ovs_vport_exit();
2439 error_flow_exit:
2440         ovs_flow_exit();
2441 error_wq:
2442         ovs_workqueues_exit();
2443 error:
2444         return err;
2445 }
2446
2447 static void dp_cleanup(void)
2448 {
2449         cancel_delayed_work_sync(&rehash_flow_wq);
2450         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2451         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2452         unregister_pernet_device(&ovs_net_ops);
2453         rcu_barrier();
2454         ovs_vport_exit();
2455         ovs_flow_exit();
2456         ovs_workqueues_exit();
2457 }
2458
2459 module_init(dp_init);
2460 module_exit(dp_cleanup);
2461
2462 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2463 MODULE_LICENSE("GPL");
2464 MODULE_VERSION(VERSION);