datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007-2013 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, false);
 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
 238         stats = this_cpu_ptr(dp->stats_percpu);
 239
 240         /* Extract flow from 'skb' into 'key'. */
 241         error = ovs_flow_extract(skb, p->port_no, &key);
 242         if (unlikely(error)) {
 243                 kfree_skb(skb);
 244                 return;
 245         }
 246
 247         /* Look up flow. */
 248         flow = ovs_flow_lookup(rcu_dereference(dp->table), &key);
 249         if (unlikely(!flow)) {
 250                 struct dp_upcall_info upcall;
 251
 252                 upcall.cmd = OVS_PACKET_CMD_MISS;
 253                 upcall.key = &key;
 254                 upcall.userdata = NULL;
 255                 upcall.portid = p->upcall_portid;
 256                 ovs_dp_upcall(dp, skb, &upcall);
 257                 consume_skb(skb);
 258                 stats_counter = &stats->n_missed;
 259                 goto out;
 260         }
 261
 262         OVS_CB(skb)->flow = flow;
 263         OVS_CB(skb)->pkt_key = &key;
 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, 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_destroy(old_table, true);
 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 sw_flow_match match;
 623         struct sw_flow_key key;
 624         int err, start;
 625
 626         ovs_match_init(&match, &key, NULL);
 627         err = ipv4_tun_from_nlattr(nla_data(attr), &match, false);
 628         if (err)
 629                 return err;
 630
 631         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
 632         if (start < 0)
 633                 return start;
 634
 635         err = add_action(sfa, OVS_KEY_ATTR_IPV4_TUNNEL, &match.key->tun_key,
 636                         sizeof(match.key->tun_key));
 637         add_nested_action_end(*sfa, start);
 638
 639         return err;
 640 }
 641
 642 static int validate_set(const struct nlattr *a,
 643                         const struct sw_flow_key *flow_key,
 644                         struct sw_flow_actions **sfa,
 645                         bool *set_tun)
 646 {
 647         const struct nlattr *ovs_key = nla_data(a);
 648         int key_type = nla_type(ovs_key);
 649
 650         /* There can be only one key in a action */
 651         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 652                 return -EINVAL;
 653
 654         if (key_type > OVS_KEY_ATTR_MAX ||
 655             (ovs_key_lens[key_type] != nla_len(ovs_key) &&
 656              ovs_key_lens[key_type] != -1))
 657                 return -EINVAL;
 658
 659         switch (key_type) {
 660         const struct ovs_key_ipv4 *ipv4_key;
 661         const struct ovs_key_ipv6 *ipv6_key;
 662         int err;
 663
 664         case OVS_KEY_ATTR_PRIORITY:
 665         case OVS_KEY_ATTR_ETHERNET:
 666                 break;
 667
 668         case OVS_KEY_ATTR_SKB_MARK:
 669 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
 670                 if (nla_get_u32(ovs_key) != 0)
 671                         return -EINVAL;
 672 #endif
 673                 break;
 674
 675         case OVS_KEY_ATTR_TUNNEL:
 676                 *set_tun = true;
 677                 err = validate_and_copy_set_tun(a, sfa);
 678                 if (err)
 679                         return err;
 680                 break;
 681
 682         case OVS_KEY_ATTR_IPV4:
 683                 if (flow_key->eth.type != htons(ETH_P_IP))
 684                         return -EINVAL;
 685
 686                 if (!flow_key->ip.proto)
 687                         return -EINVAL;
 688
 689                 ipv4_key = nla_data(ovs_key);
 690                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 691                         return -EINVAL;
 692
 693                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 694                         return -EINVAL;
 695
 696                 break;
 697
 698         case OVS_KEY_ATTR_IPV6:
 699                 if (flow_key->eth.type != htons(ETH_P_IPV6))
 700                         return -EINVAL;
 701
 702                 if (!flow_key->ip.proto)
 703                         return -EINVAL;
 704
 705                 ipv6_key = nla_data(ovs_key);
 706                 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
 707                         return -EINVAL;
 708
 709                 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
 710                         return -EINVAL;
 711
 712                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
 713                         return -EINVAL;
 714
 715                 break;
 716
 717         case OVS_KEY_ATTR_TCP:
 718                 if (flow_key->ip.proto != IPPROTO_TCP)
 719                         return -EINVAL;
 720
 721                 return validate_tp_port(flow_key);
 722
 723         case OVS_KEY_ATTR_UDP:
 724                 if (flow_key->ip.proto != IPPROTO_UDP)
 725                         return -EINVAL;
 726
 727                 return validate_tp_port(flow_key);
 728
 729         default:
 730                 return -EINVAL;
 731         }
 732
 733         return 0;
 734 }
 735
 736 static int validate_userspace(const struct nlattr *attr)
 737 {
 738         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 739                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 740                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
 741         };
 742         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 743         int error;
 744
 745         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 746                                  attr, userspace_policy);
 747         if (error)
 748                 return error;
 749
 750         if (!a[OVS_USERSPACE_ATTR_PID] ||
 751             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 752                 return -EINVAL;
 753
 754         return 0;
 755 }
 756
 757 static int copy_action(const struct nlattr *from,
 758                       struct sw_flow_actions **sfa)
 759 {
 760         int totlen = NLA_ALIGN(from->nla_len);
 761         struct nlattr *to;
 762
 763         to = reserve_sfa_size(sfa, from->nla_len);
 764         if (IS_ERR(to))
 765                 return PTR_ERR(to);
 766
 767         memcpy(to, from, totlen);
 768         return 0;
 769 }
 770
 771 static int validate_and_copy_actions(const struct nlattr *attr,
 772                                 const struct sw_flow_key *key,
 773                                 int depth,
 774                                 struct sw_flow_actions **sfa)
 775 {
 776         const struct nlattr *a;
 777         int rem, err;
 778
 779         if (depth >= SAMPLE_ACTION_DEPTH)
 780                 return -EOVERFLOW;
 781
 782         nla_for_each_nested(a, attr, rem) {
 783                 /* Expected argument lengths, (u32)-1 for variable length. */
 784                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 785                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 786                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 787                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 788                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 789                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 790                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 791                 };
 792                 const struct ovs_action_push_vlan *vlan;
 793                 int type = nla_type(a);
 794                 bool skip_copy;
 795
 796                 if (type > OVS_ACTION_ATTR_MAX ||
 797                     (action_lens[type] != nla_len(a) &&
 798                      action_lens[type] != (u32)-1))
 799                         return -EINVAL;
 800
 801                 skip_copy = false;
 802                 switch (type) {
 803                 case OVS_ACTION_ATTR_UNSPEC:
 804                         return -EINVAL;
 805
 806                 case OVS_ACTION_ATTR_USERSPACE:
 807                         err = validate_userspace(a);
 808                         if (err)
 809                                 return err;
 810                         break;
 811
 812                 case OVS_ACTION_ATTR_OUTPUT:
 813                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 814                                 return -EINVAL;
 815                         break;
 816
 817
 818                 case OVS_ACTION_ATTR_POP_VLAN:
 819                         break;
 820
 821                 case OVS_ACTION_ATTR_PUSH_VLAN:
 822                         vlan = nla_data(a);
 823                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 824                                 return -EINVAL;
 825                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 826                                 return -EINVAL;
 827                         break;
 828
 829                 case OVS_ACTION_ATTR_SET:
 830                         err = validate_set(a, key, sfa, &skip_copy);
 831                         if (err)
 832                                 return err;
 833                         break;
 834
 835                 case OVS_ACTION_ATTR_SAMPLE:
 836                         err = validate_and_copy_sample(a, key, depth, sfa);
 837                         if (err)
 838                                 return err;
 839                         skip_copy = true;
 840                         break;
 841
 842                 default:
 843                         return -EINVAL;
 844                 }
 845                 if (!skip_copy) {
 846                         err = copy_action(a, sfa);
 847                         if (err)
 848                                 return err;
 849                 }
 850         }
 851
 852         if (rem > 0)
 853                 return -EINVAL;
 854
 855         return 0;
 856 }
 857
 858 static void clear_stats(struct sw_flow *flow)
 859 {
 860         flow->used = 0;
 861         flow->tcp_flags = 0;
 862         flow->packet_count = 0;
 863         flow->byte_count = 0;
 864 }
 865
 866 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 867 {
 868         struct ovs_header *ovs_header = info->userhdr;
 869         struct nlattr **a = info->attrs;
 870         struct sw_flow_actions *acts;
 871         struct sk_buff *packet;
 872         struct sw_flow *flow;
 873         struct datapath *dp;
 874         struct ethhdr *eth;
 875         int len;
 876         int err;
 877
 878         err = -EINVAL;
 879         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 880             !a[OVS_PACKET_ATTR_ACTIONS])
 881                 goto err;
 882
 883         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 884         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 885         err = -ENOMEM;
 886         if (!packet)
 887                 goto err;
 888         skb_reserve(packet, NET_IP_ALIGN);
 889
 890         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 891
 892         skb_reset_mac_header(packet);
 893         eth = eth_hdr(packet);
 894
 895         /* Normally, setting the skb 'protocol' field would be handled by a
 896          * call to eth_type_trans(), but it assumes there's a sending
 897          * device, which we may not have. */
 898         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 899                 packet->protocol = eth->h_proto;
 900         else
 901                 packet->protocol = htons(ETH_P_802_2);
 902
 903         /* Build an sw_flow for sending this packet. */
 904         flow = ovs_flow_alloc();
 905         err = PTR_ERR(flow);
 906         if (IS_ERR(flow))
 907                 goto err_kfree_skb;
 908
 909         err = ovs_flow_extract(packet, -1, &flow->key);
 910         if (err)
 911                 goto err_flow_free;
 912
 913         err = ovs_flow_metadata_from_nlattrs(flow, a[OVS_PACKET_ATTR_KEY]);
 914         if (err)
 915                 goto err_flow_free;
 916         acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 917         err = PTR_ERR(acts);
 918         if (IS_ERR(acts))
 919                 goto err_flow_free;
 920
 921         err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
 922         rcu_assign_pointer(flow->sf_acts, acts);
 923         if (err)
 924                 goto err_flow_free;
 925
 926         OVS_CB(packet)->flow = flow;
 927         OVS_CB(packet)->pkt_key = &flow->key;
 928         packet->priority = flow->key.phy.priority;
 929         skb_set_mark(packet, flow->key.phy.skb_mark);
 930
 931         rcu_read_lock();
 932         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 933         err = -ENODEV;
 934         if (!dp)
 935                 goto err_unlock;
 936
 937         local_bh_disable();
 938         err = ovs_execute_actions(dp, packet);
 939         local_bh_enable();
 940         rcu_read_unlock();
 941
 942         ovs_flow_free(flow, false);
 943         return err;
 944
 945 err_unlock:
 946         rcu_read_unlock();
 947 err_flow_free:
 948         ovs_flow_free(flow, false);
 949 err_kfree_skb:
 950         kfree_skb(packet);
 951 err:
 952         return err;
 953 }
 954
 955 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 956 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
 957         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 958 #else
 959         [OVS_PACKET_ATTR_PACKET] = { .minlen = ETH_HLEN },
 960 #endif
 961         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 962         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 963 };
 964
 965 static struct genl_ops dp_packet_genl_ops[] = {
 966         { .cmd = OVS_PACKET_CMD_EXECUTE,
 967           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 968           .policy = packet_policy,
 969           .doit = ovs_packet_cmd_execute
 970         }
 971 };
 972
 973 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 974 {
 975         struct flow_table *table;
 976         int i;
 977
 978         table = rcu_dereference_check(dp->table, lockdep_ovsl_is_held());
 979         stats->n_flows = ovs_flow_tbl_count(table);
 980
 981         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 982         for_each_possible_cpu(i) {
 983                 const struct dp_stats_percpu *percpu_stats;
 984                 struct dp_stats_percpu local_stats;
 985                 unsigned int start;
 986
 987                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 988
 989                 do {
 990                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 991                         local_stats = *percpu_stats;
 992                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 993
 994                 stats->n_hit += local_stats.n_hit;
 995                 stats->n_missed += local_stats.n_missed;
 996                 stats->n_lost += local_stats.n_lost;
 997         }
 998 }
 999
1000 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1001         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1002         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1003         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1004 };
1005
1006 static struct genl_family dp_flow_genl_family = {
1007         .id = GENL_ID_GENERATE,
1008         .hdrsize = sizeof(struct ovs_header),
1009         .name = OVS_FLOW_FAMILY,
1010         .version = OVS_FLOW_VERSION,
1011         .maxattr = OVS_FLOW_ATTR_MAX,
1012          SET_NETNSOK
1013 };
1014
1015 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
1016         .name = OVS_FLOW_MCGROUP
1017 };
1018
1019 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
1020 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1021 {
1022         const struct nlattr *a;
1023         struct nlattr *start;
1024         int err = 0, rem;
1025
1026         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1027         if (!start)
1028                 return -EMSGSIZE;
1029
1030         nla_for_each_nested(a, attr, rem) {
1031                 int type = nla_type(a);
1032                 struct nlattr *st_sample;
1033
1034                 switch (type) {
1035                 case OVS_SAMPLE_ATTR_PROBABILITY:
1036                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
1037                                 return -EMSGSIZE;
1038                         break;
1039                 case OVS_SAMPLE_ATTR_ACTIONS:
1040                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1041                         if (!st_sample)
1042                                 return -EMSGSIZE;
1043                         err = actions_to_attr(nla_data(a), nla_len(a), skb);
1044                         if (err)
1045                                 return err;
1046                         nla_nest_end(skb, st_sample);
1047                         break;
1048                 }
1049         }
1050
1051         nla_nest_end(skb, start);
1052         return err;
1053 }
1054
1055 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1056 {
1057         const struct nlattr *ovs_key = nla_data(a);
1058         int key_type = nla_type(ovs_key);
1059         struct nlattr *start;
1060         int err;
1061
1062         switch (key_type) {
1063         case OVS_KEY_ATTR_IPV4_TUNNEL:
1064                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1065                 if (!start)
1066                         return -EMSGSIZE;
1067
1068                 err = ipv4_tun_to_nlattr(skb,
1069                                 nla_data(ovs_key), nla_data(ovs_key));
1070                 if (err)
1071                         return err;
1072                 nla_nest_end(skb, start);
1073                 break;
1074         default:
1075                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1076                         return -EMSGSIZE;
1077                 break;
1078         }
1079
1080         return 0;
1081 }
1082
1083 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1084 {
1085         const struct nlattr *a;
1086         int rem, err;
1087
1088         nla_for_each_attr(a, attr, len, rem) {
1089                 int type = nla_type(a);
1090
1091                 switch (type) {
1092                 case OVS_ACTION_ATTR_SET:
1093                         err = set_action_to_attr(a, skb);
1094                         if (err)
1095                                 return err;
1096                         break;
1097
1098                 case OVS_ACTION_ATTR_SAMPLE:
1099                         err = sample_action_to_attr(a, skb);
1100                         if (err)
1101                                 return err;
1102                         break;
1103                 default:
1104                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
1105                                 return -EMSGSIZE;
1106                         break;
1107                 }
1108         }
1109
1110         return 0;
1111 }
1112
1113 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
1114 {
1115         return NLMSG_ALIGN(sizeof(struct ovs_header))
1116                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
1117                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
1118                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
1119                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
1120                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
1121                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
1122 }
1123
1124 /* Called with ovs_mutex. */
1125 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1126                                   struct sk_buff *skb, u32 portid,
1127                                   u32 seq, u32 flags, u8 cmd)
1128 {
1129         const int skb_orig_len = skb->len;
1130         struct sw_flow_mask *mask;
1131         struct nlattr *start;
1132         struct ovs_flow_stats stats;
1133         struct ovs_header *ovs_header;
1134         struct nlattr *nla;
1135         unsigned long used;
1136         u8 tcp_flags;
1137         int err;
1138
1139         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1140         if (!ovs_header)
1141                 return -EMSGSIZE;
1142
1143         ovs_header->dp_ifindex = get_dpifindex(dp);
1144
1145         /* Fill flow key. */
1146         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1147         if (!nla)
1148                 goto nla_put_failure;
1149
1150         err = ovs_flow_to_nlattrs(&flow->unmasked_key,
1151                         &flow->unmasked_key, skb);
1152         if (err)
1153                 goto error;
1154         nla_nest_end(skb, nla);
1155
1156         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
1157         if (!nla)
1158                 goto nla_put_failure;
1159
1160         mask = rcu_dereference_check(flow->mask, lockdep_ovsl_is_held());
1161         err = ovs_flow_to_nlattrs(&flow->key, &mask->key, skb);
1162         if (err)
1163                 goto error;
1164
1165         nla_nest_end(skb, nla);
1166
1167         spin_lock_bh(&flow->lock);
1168         used = flow->used;
1169         stats.n_packets = flow->packet_count;
1170         stats.n_bytes = flow->byte_count;
1171         tcp_flags = flow->tcp_flags;
1172         spin_unlock_bh(&flow->lock);
1173
1174         if (used &&
1175             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1176                 goto nla_put_failure;
1177
1178         if (stats.n_packets &&
1179             nla_put(skb, OVS_FLOW_ATTR_STATS,
1180                     sizeof(struct ovs_flow_stats), &stats))
1181                 goto nla_put_failure;
1182
1183         if (tcp_flags &&
1184             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1185                 goto nla_put_failure;
1186
1187         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1188          * this is the first flow to be dumped into 'skb'.  This is unusual for
1189          * Netlink but individual action lists can be longer than
1190          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1191          * The userspace caller can always fetch the actions separately if it
1192          * really wants them.  (Most userspace callers in fact don't care.)
1193          *
1194          * This can only fail for dump operations because the skb is always
1195          * properly sized for single flows.
1196          */
1197         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1198         if (start) {
1199                 const struct sw_flow_actions *sf_acts;
1200
1201                 sf_acts = rcu_dereference_check(flow->sf_acts,
1202                                                 lockdep_ovsl_is_held());
1203
1204                 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1205                 if (!err)
1206                         nla_nest_end(skb, start);
1207                 else {
1208                         if (skb_orig_len)
1209                                 goto error;
1210
1211                         nla_nest_cancel(skb, start);
1212                 }
1213         } else if (skb_orig_len)
1214                 goto nla_put_failure;
1215
1216         return genlmsg_end(skb, ovs_header);
1217
1218 nla_put_failure:
1219         err = -EMSGSIZE;
1220 error:
1221         genlmsg_cancel(skb, ovs_header);
1222         return err;
1223 }
1224
1225 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1226 {
1227         const struct sw_flow_actions *sf_acts;
1228
1229         sf_acts = ovsl_dereference(flow->sf_acts);
1230
1231         return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
1232 }
1233
1234 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1235                                                struct datapath *dp,
1236                                                u32 portid, u32 seq, u8 cmd)
1237 {
1238         struct sk_buff *skb;
1239         int retval;
1240
1241         skb = ovs_flow_cmd_alloc_info(flow);
1242         if (!skb)
1243                 return ERR_PTR(-ENOMEM);
1244
1245         retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1246         BUG_ON(retval < 0);
1247         return skb;
1248 }
1249
1250 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1251 {
1252         struct nlattr **a = info->attrs;
1253         struct ovs_header *ovs_header = info->userhdr;
1254         struct sw_flow_key key;
1255         struct sw_flow *flow = NULL;
1256         struct sw_flow_mask mask;
1257         struct sk_buff *reply;
1258         struct datapath *dp;
1259         struct flow_table *table;
1260         struct sw_flow_actions *acts = NULL;
1261         struct sw_flow_match match;
1262         int error;
1263
1264         /* Extract key. */
1265         error = -EINVAL;
1266         if (!a[OVS_FLOW_ATTR_KEY])
1267                 goto error;
1268
1269         ovs_match_init(&match, &key, &mask);
1270         error = ovs_match_from_nlattrs(&match,
1271                         a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
1272         if (error)
1273                 goto error;
1274
1275         /* Validate actions. */
1276         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1277                 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1278                 error = PTR_ERR(acts);
1279                 if (IS_ERR(acts))
1280                         goto error;
1281
1282                 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0, &acts);
1283                 if (error)
1284                         goto err_kfree;
1285         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1286                 error = -EINVAL;
1287                 goto error;
1288         }
1289
1290         ovs_lock();
1291         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1292         error = -ENODEV;
1293         if (!dp)
1294                 goto err_unlock_ovs;
1295
1296         table = ovsl_dereference(dp->table);
1297
1298         /* Check if this is a duplicate flow */
1299         flow = ovs_flow_lookup(table, &key);
1300         if (!flow) {
1301                 struct sw_flow_mask *mask_p;
1302                 /* Bail out if we're not allowed to create a new flow. */
1303                 error = -ENOENT;
1304                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1305                         goto err_unlock_ovs;
1306
1307                 /* Expand table, if necessary, to make room. */
1308                 if (ovs_flow_tbl_need_to_expand(table)) {
1309                         struct flow_table *new_table;
1310
1311                         new_table = ovs_flow_tbl_expand(table);
1312                         if (!IS_ERR(new_table)) {
1313                                 rcu_assign_pointer(dp->table, new_table);
1314                                 ovs_flow_tbl_destroy(table, true);
1315                                 table = ovsl_dereference(dp->table);
1316                         }
1317                 }
1318
1319                 /* Allocate flow. */
1320                 flow = ovs_flow_alloc();
1321                 if (IS_ERR(flow)) {
1322                         error = PTR_ERR(flow);
1323                         goto err_unlock_ovs;
1324                 }
1325                 clear_stats(flow);
1326
1327                 /* Make sure mask is unique in the system */
1328                 mask_p = ovs_sw_flow_mask_find(table, &mask);
1329                 if (!mask_p) {
1330                         /* Allocate a new mask if none exsits. */
1331                         mask_p = ovs_sw_flow_mask_alloc();
1332                         if (!mask_p)
1333                                 goto err_flow_free;
1334                         mask_p->key = mask.key;
1335                         mask_p->range = mask.range;
1336                         ovs_sw_flow_mask_insert(table, mask_p);
1337                 }
1338
1339                 ovs_sw_flow_mask_add_ref(mask_p);
1340                 rcu_assign_pointer(flow->mask, mask_p);
1341                 rcu_assign_pointer(flow->sf_acts, acts);
1342
1343                 /* Put flow in bucket. */
1344                 ovs_flow_insert(table, flow, &key, match.range.end);
1345
1346                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1347                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
1348         } else {
1349                 /* We found a matching flow. */
1350                 struct sw_flow_actions *old_acts;
1351
1352                 /* Bail out if we're not allowed to modify an existing flow.
1353                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1354                  * because Generic Netlink treats the latter as a dump
1355                  * request.  We also accept NLM_F_EXCL in case that bug ever
1356                  * gets fixed.
1357                  */
1358                 error = -EEXIST;
1359                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1360                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1361                         goto err_unlock_ovs;
1362
1363                 /* The unmasked key has to be the same for flow updates. */
1364                 error = -EINVAL;
1365                 if (!ovs_flow_cmp_unmasked_key(flow, &key, match.range.end)) {
1366                         OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
1367                         goto err_unlock_ovs;
1368                 }
1369
1370                 /* Update actions. */
1371                 old_acts = ovsl_dereference(flow->sf_acts);
1372                 rcu_assign_pointer(flow->sf_acts, acts);
1373                 ovs_flow_deferred_free_acts(old_acts);
1374
1375                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1376                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1377
1378                 /* Clear stats. */
1379                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1380                         spin_lock_bh(&flow->lock);
1381                         clear_stats(flow);
1382                         spin_unlock_bh(&flow->lock);
1383                 }
1384         }
1385         ovs_unlock();
1386
1387         if (!IS_ERR(reply))
1388                 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1389         else
1390                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1391                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1392         return 0;
1393
1394 err_flow_free:
1395         ovs_flow_free(flow, false);
1396 err_unlock_ovs:
1397         ovs_unlock();
1398 err_kfree:
1399         kfree(acts);
1400 error:
1401         return error;
1402 }
1403
1404 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1405 {
1406         struct nlattr **a = info->attrs;
1407         struct ovs_header *ovs_header = info->userhdr;
1408         struct sw_flow_key key;
1409         struct sk_buff *reply;
1410         struct sw_flow *flow;
1411         struct datapath *dp;
1412         struct flow_table *table;
1413         struct sw_flow_match match;
1414         int err;
1415
1416         if (!a[OVS_FLOW_ATTR_KEY]) {
1417                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1418                 return -EINVAL;
1419         }
1420
1421         ovs_match_init(&match, &key, NULL);
1422         err = ovs_match_from_nlattrs(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1423         if (err)
1424                 return err;
1425
1426         ovs_lock();
1427         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1428         if (!dp) {
1429                 err = -ENODEV;
1430                 goto unlock;
1431         }
1432
1433         table = ovsl_dereference(dp->table);
1434         flow = ovs_flow_lookup_unmasked_key(table, &match);
1435         if (!flow) {
1436                 err = -ENOENT;
1437                 goto unlock;
1438         }
1439
1440         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1441                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1442         if (IS_ERR(reply)) {
1443                 err = PTR_ERR(reply);
1444                 goto unlock;
1445         }
1446
1447         ovs_unlock();
1448         return genlmsg_reply(reply, info);
1449 unlock:
1450         ovs_unlock();
1451         return err;
1452 }
1453
1454 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1455 {
1456         struct nlattr **a = info->attrs;
1457         struct ovs_header *ovs_header = info->userhdr;
1458         struct sw_flow_key key;
1459         struct sk_buff *reply;
1460         struct sw_flow *flow;
1461         struct datapath *dp;
1462         struct flow_table *table;
1463         struct sw_flow_match match;
1464         int err;
1465
1466         ovs_lock();
1467         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1468         if (!dp) {
1469                 err = -ENODEV;
1470                 goto unlock;
1471         }
1472
1473         if (!a[OVS_FLOW_ATTR_KEY]) {
1474                 err = flush_flows(dp);
1475                 goto unlock;
1476         }
1477
1478         ovs_match_init(&match, &key, NULL);
1479         err = ovs_match_from_nlattrs(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1480         if (err)
1481                 goto unlock;
1482
1483         table = ovsl_dereference(dp->table);
1484         flow = ovs_flow_lookup_unmasked_key(table, &match);
1485         if (!flow) {
1486                 err = -ENOENT;
1487                 goto unlock;
1488         }
1489
1490         reply = ovs_flow_cmd_alloc_info(flow);
1491         if (!reply) {
1492                 err = -ENOMEM;
1493                 goto unlock;
1494         }
1495
1496         ovs_flow_remove(table, flow);
1497
1498         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1499                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1500         BUG_ON(err < 0);
1501
1502         ovs_flow_free(flow, true);
1503         ovs_unlock();
1504
1505         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1506         return 0;
1507 unlock:
1508         ovs_unlock();
1509         return err;
1510 }
1511
1512 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1513 {
1514         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1515         struct datapath *dp;
1516         struct flow_table *table;
1517
1518         rcu_read_lock();
1519         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1520         if (!dp) {
1521                 rcu_read_unlock();
1522                 return -ENODEV;
1523         }
1524
1525         table = rcu_dereference(dp->table);
1526         for (;;) {
1527                 struct sw_flow *flow;
1528                 u32 bucket, obj;
1529
1530                 bucket = cb->args[0];
1531                 obj = cb->args[1];
1532                 flow = ovs_flow_dump_next(table, &bucket, &obj);
1533                 if (!flow)
1534                         break;
1535
1536                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1537                                            NETLINK_CB(cb->skb).portid,
1538                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1539                                            OVS_FLOW_CMD_NEW) < 0)
1540                         break;
1541
1542                 cb->args[0] = bucket;
1543                 cb->args[1] = obj;
1544         }
1545         rcu_read_unlock();
1546         return skb->len;
1547 }
1548
1549 static struct genl_ops dp_flow_genl_ops[] = {
1550         { .cmd = OVS_FLOW_CMD_NEW,
1551           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1552           .policy = flow_policy,
1553           .doit = ovs_flow_cmd_new_or_set
1554         },
1555         { .cmd = OVS_FLOW_CMD_DEL,
1556           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1557           .policy = flow_policy,
1558           .doit = ovs_flow_cmd_del
1559         },
1560         { .cmd = OVS_FLOW_CMD_GET,
1561           .flags = 0,               /* OK for unprivileged users. */
1562           .policy = flow_policy,
1563           .doit = ovs_flow_cmd_get,
1564           .dumpit = ovs_flow_cmd_dump
1565         },
1566         { .cmd = OVS_FLOW_CMD_SET,
1567           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1568           .policy = flow_policy,
1569           .doit = ovs_flow_cmd_new_or_set,
1570         },
1571 };
1572
1573 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1574 #ifdef HAVE_NLA_NUL_STRING
1575         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1576 #endif
1577         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1578 };
1579
1580 static struct genl_family dp_datapath_genl_family = {
1581         .id = GENL_ID_GENERATE,
1582         .hdrsize = sizeof(struct ovs_header),
1583         .name = OVS_DATAPATH_FAMILY,
1584         .version = OVS_DATAPATH_VERSION,
1585         .maxattr = OVS_DP_ATTR_MAX,
1586          SET_NETNSOK
1587 };
1588
1589 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1590         .name = OVS_DATAPATH_MCGROUP
1591 };
1592
1593 static size_t ovs_dp_cmd_msg_size(void)
1594 {
1595         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1596
1597         msgsize += nla_total_size(IFNAMSIZ);
1598         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1599
1600         return msgsize;
1601 }
1602
1603 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1604                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1605 {
1606         struct ovs_header *ovs_header;
1607         struct ovs_dp_stats dp_stats;
1608         int err;
1609
1610         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1611                                    flags, cmd);
1612         if (!ovs_header)
1613                 goto error;
1614
1615         ovs_header->dp_ifindex = get_dpifindex(dp);
1616
1617         rcu_read_lock();
1618         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1619         rcu_read_unlock();
1620         if (err)
1621                 goto nla_put_failure;
1622
1623         get_dp_stats(dp, &dp_stats);
1624         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1625                 goto nla_put_failure;
1626
1627         return genlmsg_end(skb, ovs_header);
1628
1629 nla_put_failure:
1630         genlmsg_cancel(skb, ovs_header);
1631 error:
1632         return -EMSGSIZE;
1633 }
1634
1635 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1636                                              u32 seq, u8 cmd)
1637 {
1638         struct sk_buff *skb;
1639         int retval;
1640
1641         skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1642         if (!skb)
1643                 return ERR_PTR(-ENOMEM);
1644
1645         retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1646         if (retval < 0) {
1647                 kfree_skb(skb);
1648                 return ERR_PTR(retval);
1649         }
1650         return skb;
1651 }
1652
1653 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1654 {
1655         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1656 }
1657
1658 /* Called with ovs_mutex. */
1659 static struct datapath *lookup_datapath(struct net *net,
1660                                         struct ovs_header *ovs_header,
1661                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1662 {
1663         struct datapath *dp;
1664
1665         if (!a[OVS_DP_ATTR_NAME])
1666                 dp = get_dp(net, ovs_header->dp_ifindex);
1667         else {
1668                 struct vport *vport;
1669
1670                 rcu_read_lock();
1671                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1672                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1673                 rcu_read_unlock();
1674         }
1675         return dp ? dp : ERR_PTR(-ENODEV);
1676 }
1677
1678 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1679 {
1680         struct nlattr **a = info->attrs;
1681         struct vport_parms parms;
1682         struct sk_buff *reply;
1683         struct datapath *dp;
1684         struct vport *vport;
1685         struct ovs_net *ovs_net;
1686         int err, i;
1687
1688         err = -EINVAL;
1689         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1690                 goto err;
1691
1692         err = ovs_dp_cmd_validate(a);
1693         if (err)
1694                 goto err;
1695
1696         ovs_lock();
1697
1698         err = -ENOMEM;
1699         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1700         if (dp == NULL)
1701                 goto err_unlock_ovs;
1702
1703         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1704
1705         /* Allocate table. */
1706         err = -ENOMEM;
1707         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1708         if (!dp->table)
1709                 goto err_free_dp;
1710
1711         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1712         if (!dp->stats_percpu) {
1713                 err = -ENOMEM;
1714                 goto err_destroy_table;
1715         }
1716
1717         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1718                             GFP_KERNEL);
1719         if (!dp->ports) {
1720                 err = -ENOMEM;
1721                 goto err_destroy_percpu;
1722         }
1723
1724         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1725                 INIT_HLIST_HEAD(&dp->ports[i]);
1726
1727         /* Set up our datapath device. */
1728         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1729         parms.type = OVS_VPORT_TYPE_INTERNAL;
1730         parms.options = NULL;
1731         parms.dp = dp;
1732         parms.port_no = OVSP_LOCAL;
1733         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1734
1735         vport = new_vport(&parms);
1736         if (IS_ERR(vport)) {
1737                 err = PTR_ERR(vport);
1738                 if (err == -EBUSY)
1739                         err = -EEXIST;
1740
1741                 goto err_destroy_ports_array;
1742         }
1743
1744         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1745                                       info->snd_seq, OVS_DP_CMD_NEW);
1746         err = PTR_ERR(reply);
1747         if (IS_ERR(reply))
1748                 goto err_destroy_local_port;
1749
1750         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1751         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1752
1753         ovs_unlock();
1754
1755         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1756         return 0;
1757
1758 err_destroy_local_port:
1759         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1760 err_destroy_ports_array:
1761         kfree(dp->ports);
1762 err_destroy_percpu:
1763         free_percpu(dp->stats_percpu);
1764 err_destroy_table:
1765         ovs_flow_tbl_destroy(ovsl_dereference(dp->table), false);
1766 err_free_dp:
1767         release_net(ovs_dp_get_net(dp));
1768         kfree(dp);
1769 err_unlock_ovs:
1770         ovs_unlock();
1771 err:
1772         return err;
1773 }
1774
1775 /* Called with ovs_mutex. */
1776 static void __dp_destroy(struct datapath *dp)
1777 {
1778         int i;
1779
1780         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1781                 struct vport *vport;
1782                 struct hlist_node *n;
1783
1784                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1785                         if (vport->port_no != OVSP_LOCAL)
1786                                 ovs_dp_detach_port(vport);
1787         }
1788
1789         list_del_rcu(&dp->list_node);
1790
1791         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1792          * all port in datapath are destroyed first before freeing datapath.
1793          */
1794         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1795
1796         call_rcu(&dp->rcu, destroy_dp_rcu);
1797 }
1798
1799 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1800 {
1801         struct sk_buff *reply;
1802         struct datapath *dp;
1803         int err;
1804
1805         err = ovs_dp_cmd_validate(info->attrs);
1806         if (err)
1807                 return err;
1808
1809         ovs_lock();
1810         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1811         err = PTR_ERR(dp);
1812         if (IS_ERR(dp))
1813                 goto unlock;
1814
1815         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1816                                       info->snd_seq, OVS_DP_CMD_DEL);
1817         err = PTR_ERR(reply);
1818         if (IS_ERR(reply))
1819                 goto unlock;
1820
1821         __dp_destroy(dp);
1822         ovs_unlock();
1823
1824         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1825
1826         return 0;
1827 unlock:
1828         ovs_unlock();
1829         return err;
1830 }
1831
1832 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1833 {
1834         struct sk_buff *reply;
1835         struct datapath *dp;
1836         int err;
1837
1838         err = ovs_dp_cmd_validate(info->attrs);
1839         if (err)
1840                 return err;
1841
1842         ovs_lock();
1843         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1844         err = PTR_ERR(dp);
1845         if (IS_ERR(dp))
1846                 goto unlock;
1847
1848         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1849                                       info->snd_seq, OVS_DP_CMD_NEW);
1850         if (IS_ERR(reply)) {
1851                 err = PTR_ERR(reply);
1852                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1853                                 ovs_dp_datapath_multicast_group.id, err);
1854                 err = 0;
1855                 goto unlock;
1856         }
1857
1858         ovs_unlock();
1859         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1860
1861         return 0;
1862 unlock:
1863         ovs_unlock();
1864         return err;
1865 }
1866
1867 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1868 {
1869         struct sk_buff *reply;
1870         struct datapath *dp;
1871         int err;
1872
1873         err = ovs_dp_cmd_validate(info->attrs);
1874         if (err)
1875                 return err;
1876
1877         ovs_lock();
1878         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1879         if (IS_ERR(dp)) {
1880                 err = PTR_ERR(dp);
1881                 goto unlock;
1882         }
1883
1884         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1885                                       info->snd_seq, OVS_DP_CMD_NEW);
1886         if (IS_ERR(reply)) {
1887                 err = PTR_ERR(reply);
1888                 goto unlock;
1889         }
1890
1891         ovs_unlock();
1892         return genlmsg_reply(reply, info);
1893
1894 unlock:
1895         ovs_unlock();
1896         return err;
1897 }
1898
1899 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1900 {
1901         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1902         struct datapath *dp;
1903         int skip = cb->args[0];
1904         int i = 0;
1905
1906         rcu_read_lock();
1907         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1908                 if (i >= skip &&
1909                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1910                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1911                                          OVS_DP_CMD_NEW) < 0)
1912                         break;
1913                 i++;
1914         }
1915         rcu_read_unlock();
1916
1917         cb->args[0] = i;
1918
1919         return skb->len;
1920 }
1921
1922 static struct genl_ops dp_datapath_genl_ops[] = {
1923         { .cmd = OVS_DP_CMD_NEW,
1924           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1925           .policy = datapath_policy,
1926           .doit = ovs_dp_cmd_new
1927         },
1928         { .cmd = OVS_DP_CMD_DEL,
1929           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1930           .policy = datapath_policy,
1931           .doit = ovs_dp_cmd_del
1932         },
1933         { .cmd = OVS_DP_CMD_GET,
1934           .flags = 0,               /* OK for unprivileged users. */
1935           .policy = datapath_policy,
1936           .doit = ovs_dp_cmd_get,
1937           .dumpit = ovs_dp_cmd_dump
1938         },
1939         { .cmd = OVS_DP_CMD_SET,
1940           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1941           .policy = datapath_policy,
1942           .doit = ovs_dp_cmd_set,
1943         },
1944 };
1945
1946 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1947 #ifdef HAVE_NLA_NUL_STRING
1948         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1949         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1950 #else
1951         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1952 #endif
1953         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1954         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1955         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1956         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1957 };
1958
1959 static struct genl_family dp_vport_genl_family = {
1960         .id = GENL_ID_GENERATE,
1961         .hdrsize = sizeof(struct ovs_header),
1962         .name = OVS_VPORT_FAMILY,
1963         .version = OVS_VPORT_VERSION,
1964         .maxattr = OVS_VPORT_ATTR_MAX,
1965          SET_NETNSOK
1966 };
1967
1968 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1969         .name = OVS_VPORT_MCGROUP
1970 };
1971
1972 /* Called with ovs_mutex or RCU read lock. */
1973 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1974                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1975 {
1976         struct ovs_header *ovs_header;
1977         struct ovs_vport_stats vport_stats;
1978         int err;
1979
1980         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1981                                  flags, cmd);
1982         if (!ovs_header)
1983                 return -EMSGSIZE;
1984
1985         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1986
1987         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1988             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1989             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1990             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1991                 goto nla_put_failure;
1992
1993         ovs_vport_get_stats(vport, &vport_stats);
1994         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1995                     &vport_stats))
1996                 goto nla_put_failure;
1997
1998         err = ovs_vport_get_options(vport, skb);
1999         if (err == -EMSGSIZE)
2000                 goto error;
2001
2002         return genlmsg_end(skb, ovs_header);
2003
2004 nla_put_failure:
2005         err = -EMSGSIZE;
2006 error:
2007         genlmsg_cancel(skb, ovs_header);
2008         return err;
2009 }
2010
2011 /* Called with ovs_mutex or RCU read lock. */
2012 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
2013                                          u32 seq, u8 cmd)
2014 {
2015         struct sk_buff *skb;
2016         int retval;
2017
2018         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2019         if (!skb)
2020                 return ERR_PTR(-ENOMEM);
2021
2022         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
2023         BUG_ON(retval < 0);
2024
2025         return skb;
2026 }
2027
2028 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2029 {
2030         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
2031 }
2032
2033 /* Called with ovs_mutex or RCU read lock. */
2034 static struct vport *lookup_vport(struct net *net,
2035                                   struct ovs_header *ovs_header,
2036                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2037 {
2038         struct datapath *dp;
2039         struct vport *vport;
2040
2041         if (a[OVS_VPORT_ATTR_NAME]) {
2042                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2043                 if (!vport)
2044                         return ERR_PTR(-ENODEV);
2045                 if (ovs_header->dp_ifindex &&
2046                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2047                         return ERR_PTR(-ENODEV);
2048                 return vport;
2049         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2050                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2051
2052                 if (port_no >= DP_MAX_PORTS)
2053                         return ERR_PTR(-EFBIG);
2054
2055                 dp = get_dp(net, ovs_header->dp_ifindex);
2056                 if (!dp)
2057                         return ERR_PTR(-ENODEV);
2058
2059                 vport = ovs_vport_ovsl_rcu(dp, port_no);
2060                 if (!vport)
2061                         return ERR_PTR(-ENODEV);
2062                 return vport;
2063         } else
2064                 return ERR_PTR(-EINVAL);
2065 }
2066
2067 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2068 {
2069         struct nlattr **a = info->attrs;
2070         struct ovs_header *ovs_header = info->userhdr;
2071         struct vport_parms parms;
2072         struct sk_buff *reply;
2073         struct vport *vport;
2074         struct datapath *dp;
2075         u32 port_no;
2076         int err;
2077
2078         err = -EINVAL;
2079         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2080             !a[OVS_VPORT_ATTR_UPCALL_PID])
2081                 goto exit;
2082
2083         err = ovs_vport_cmd_validate(a);
2084         if (err)
2085                 goto exit;
2086
2087         ovs_lock();
2088         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2089         err = -ENODEV;
2090         if (!dp)
2091                 goto exit_unlock;
2092
2093         if (a[OVS_VPORT_ATTR_PORT_NO]) {
2094                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2095
2096                 err = -EFBIG;
2097                 if (port_no >= DP_MAX_PORTS)
2098                         goto exit_unlock;
2099
2100                 vport = ovs_vport_ovsl(dp, port_no);
2101                 err = -EBUSY;
2102                 if (vport)
2103                         goto exit_unlock;
2104         } else {
2105                 for (port_no = 1; ; port_no++) {
2106                         if (port_no >= DP_MAX_PORTS) {
2107                                 err = -EFBIG;
2108                                 goto exit_unlock;
2109                         }
2110                         vport = ovs_vport_ovsl(dp, port_no);
2111                         if (!vport)
2112                                 break;
2113                 }
2114         }
2115
2116         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2117         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2118         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2119         parms.dp = dp;
2120         parms.port_no = port_no;
2121         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2122
2123         vport = new_vport(&parms);
2124         err = PTR_ERR(vport);
2125         if (IS_ERR(vport))
2126                 goto exit_unlock;
2127
2128         err = 0;
2129         if (a[OVS_VPORT_ATTR_STATS])
2130                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2131
2132         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
2133                                          OVS_VPORT_CMD_NEW);
2134         if (IS_ERR(reply)) {
2135                 err = PTR_ERR(reply);
2136                 ovs_dp_detach_port(vport);
2137                 goto exit_unlock;
2138         }
2139
2140         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2141
2142 exit_unlock:
2143         ovs_unlock();
2144 exit:
2145         return err;
2146 }
2147
2148 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2149 {
2150         struct nlattr **a = info->attrs;
2151         struct sk_buff *reply;
2152         struct vport *vport;
2153         int err;
2154
2155         err = ovs_vport_cmd_validate(a);
2156         if (err)
2157                 goto exit;
2158
2159         ovs_lock();
2160         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2161         err = PTR_ERR(vport);
2162         if (IS_ERR(vport))
2163                 goto exit_unlock;
2164
2165         if (a[OVS_VPORT_ATTR_TYPE] &&
2166             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2167                 err = -EINVAL;
2168                 goto exit_unlock;
2169         }
2170
2171         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2172         if (!reply) {
2173                 err = -ENOMEM;
2174                 goto exit_unlock;
2175         }
2176
2177         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2178                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2179                 if (err)
2180                         goto exit_free;
2181         }
2182
2183         if (a[OVS_VPORT_ATTR_STATS])
2184                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2185
2186         if (a[OVS_VPORT_ATTR_UPCALL_PID])
2187                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2188
2189         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2190                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2191         BUG_ON(err < 0);
2192
2193         ovs_unlock();
2194         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2195         return 0;
2196
2197 exit_free:
2198         kfree_skb(reply);
2199 exit_unlock:
2200         ovs_unlock();
2201 exit:
2202         return err;
2203 }
2204
2205 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2206 {
2207         struct nlattr **a = info->attrs;
2208         struct sk_buff *reply;
2209         struct vport *vport;
2210         int err;
2211
2212         err = ovs_vport_cmd_validate(a);
2213         if (err)
2214                 goto exit;
2215
2216         ovs_lock();
2217         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2218         err = PTR_ERR(vport);
2219         if (IS_ERR(vport))
2220                 goto exit_unlock;
2221
2222         if (vport->port_no == OVSP_LOCAL) {
2223                 err = -EINVAL;
2224                 goto exit_unlock;
2225         }
2226
2227         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2228                                          info->snd_seq, OVS_VPORT_CMD_DEL);
2229         err = PTR_ERR(reply);
2230         if (IS_ERR(reply))
2231                 goto exit_unlock;
2232
2233         err = 0;
2234         ovs_dp_detach_port(vport);
2235
2236         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2237
2238 exit_unlock:
2239         ovs_unlock();
2240 exit:
2241         return err;
2242 }
2243
2244 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2245 {
2246         struct nlattr **a = info->attrs;
2247         struct ovs_header *ovs_header = info->userhdr;
2248         struct sk_buff *reply;
2249         struct vport *vport;
2250         int err;
2251
2252         err = ovs_vport_cmd_validate(a);
2253         if (err)
2254                 goto exit;
2255
2256         rcu_read_lock();
2257         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2258         err = PTR_ERR(vport);
2259         if (IS_ERR(vport))
2260                 goto exit_unlock;
2261
2262         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2263                                          info->snd_seq, OVS_VPORT_CMD_NEW);
2264         err = PTR_ERR(reply);
2265         if (IS_ERR(reply))
2266                 goto exit_unlock;
2267
2268         rcu_read_unlock();
2269
2270         return genlmsg_reply(reply, info);
2271
2272 exit_unlock:
2273         rcu_read_unlock();
2274 exit:
2275         return err;
2276 }
2277
2278 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2279 {
2280         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2281         struct datapath *dp;
2282         int bucket = cb->args[0], skip = cb->args[1];
2283         int i, j = 0;
2284
2285         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2286         if (!dp)
2287                 return -ENODEV;
2288
2289         rcu_read_lock();
2290         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2291                 struct vport *vport;
2292
2293                 j = 0;
2294                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2295                         if (j >= skip &&
2296                             ovs_vport_cmd_fill_info(vport, skb,
2297                                                     NETLINK_CB(cb->skb).portid,
2298                                                     cb->nlh->nlmsg_seq,
2299                                                     NLM_F_MULTI,
2300                                                     OVS_VPORT_CMD_NEW) < 0)
2301                                 goto out;
2302
2303                         j++;
2304                 }
2305                 skip = 0;
2306         }
2307 out:
2308         rcu_read_unlock();
2309
2310         cb->args[0] = i;
2311         cb->args[1] = j;
2312
2313         return skb->len;
2314 }
2315
2316 static struct genl_ops dp_vport_genl_ops[] = {
2317         { .cmd = OVS_VPORT_CMD_NEW,
2318           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2319           .policy = vport_policy,
2320           .doit = ovs_vport_cmd_new
2321         },
2322         { .cmd = OVS_VPORT_CMD_DEL,
2323           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2324           .policy = vport_policy,
2325           .doit = ovs_vport_cmd_del
2326         },
2327         { .cmd = OVS_VPORT_CMD_GET,
2328           .flags = 0,               /* OK for unprivileged users. */
2329           .policy = vport_policy,
2330           .doit = ovs_vport_cmd_get,
2331           .dumpit = ovs_vport_cmd_dump
2332         },
2333         { .cmd = OVS_VPORT_CMD_SET,
2334           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2335           .policy = vport_policy,
2336           .doit = ovs_vport_cmd_set,
2337         },
2338 };
2339
2340 struct genl_family_and_ops {
2341         struct genl_family *family;
2342         struct genl_ops *ops;
2343         int n_ops;
2344         struct genl_multicast_group *group;
2345 };
2346
2347 static const struct genl_family_and_ops dp_genl_families[] = {
2348         { &dp_datapath_genl_family,
2349           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2350           &ovs_dp_datapath_multicast_group },
2351         { &dp_vport_genl_family,
2352           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2353           &ovs_dp_vport_multicast_group },
2354         { &dp_flow_genl_family,
2355           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2356           &ovs_dp_flow_multicast_group },
2357         { &dp_packet_genl_family,
2358           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2359           NULL },
2360 };
2361
2362 static void dp_unregister_genl(int n_families)
2363 {
2364         int i;
2365
2366         for (i = 0; i < n_families; i++)
2367                 genl_unregister_family(dp_genl_families[i].family);
2368 }
2369
2370 static int dp_register_genl(void)
2371 {
2372         int n_registered;
2373         int err;
2374         int i;
2375
2376         n_registered = 0;
2377         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2378                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2379
2380                 err = genl_register_family_with_ops(f->family, f->ops,
2381                                                     f->n_ops);
2382                 if (err)
2383                         goto error;
2384                 n_registered++;
2385
2386                 if (f->group) {
2387                         err = genl_register_mc_group(f->family, f->group);
2388                         if (err)
2389                                 goto error;
2390                 }
2391         }
2392
2393         return 0;
2394
2395 error:
2396         dp_unregister_genl(n_registered);
2397         return err;
2398 }
2399
2400 static void rehash_flow_table(struct work_struct *work)
2401 {
2402         struct datapath *dp;
2403         struct net *net;
2404
2405         ovs_lock();
2406         rtnl_lock();
2407         for_each_net(net) {
2408                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2409
2410                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2411                         struct flow_table *old_table = ovsl_dereference(dp->table);
2412                         struct flow_table *new_table;
2413
2414                         new_table = ovs_flow_tbl_rehash(old_table);
2415                         if (!IS_ERR(new_table)) {
2416                                 rcu_assign_pointer(dp->table, new_table);
2417                                 ovs_flow_tbl_destroy(old_table, true);
2418                         }
2419                 }
2420         }
2421         rtnl_unlock();
2422         ovs_unlock();
2423         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2424 }
2425
2426 static int __net_init ovs_init_net(struct net *net)
2427 {
2428         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2429
2430         INIT_LIST_HEAD(&ovs_net->dps);
2431         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2432         return 0;
2433 }
2434
2435 static void __net_exit ovs_exit_net(struct net *net)
2436 {
2437         struct datapath *dp, *dp_next;
2438         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2439
2440         ovs_lock();
2441         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2442                 __dp_destroy(dp);
2443         ovs_unlock();
2444
2445         cancel_work_sync(&ovs_net->dp_notify_work);
2446 }
2447
2448 static struct pernet_operations ovs_net_ops = {
2449         .init = ovs_init_net,
2450         .exit = ovs_exit_net,
2451         .id   = &ovs_net_id,
2452         .size = sizeof(struct ovs_net),
2453 };
2454
2455 static int __init dp_init(void)
2456 {
2457         int err;
2458
2459         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2460
2461         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2462                 VERSION);
2463
2464         err = ovs_workqueues_init();
2465         if (err)
2466                 goto error;
2467
2468         err = ovs_flow_init();
2469         if (err)
2470                 goto error_wq;
2471
2472         err = ovs_vport_init();
2473         if (err)
2474                 goto error_flow_exit;
2475
2476         err = register_pernet_device(&ovs_net_ops);
2477         if (err)
2478                 goto error_vport_exit;
2479
2480         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2481         if (err)
2482                 goto error_netns_exit;
2483
2484         err = dp_register_genl();
2485         if (err < 0)
2486                 goto error_unreg_notifier;
2487
2488         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2489
2490         return 0;
2491
2492 error_unreg_notifier:
2493         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2494 error_netns_exit:
2495         unregister_pernet_device(&ovs_net_ops);
2496 error_vport_exit:
2497         ovs_vport_exit();
2498 error_flow_exit:
2499         ovs_flow_exit();
2500 error_wq:
2501         ovs_workqueues_exit();
2502 error:
2503         return err;
2504 }
2505
2506 static void dp_cleanup(void)
2507 {
2508         cancel_delayed_work_sync(&rehash_flow_wq);
2509         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2510         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2511         unregister_pernet_device(&ovs_net_ops);
2512         rcu_barrier();
2513         ovs_vport_exit();
2514         ovs_flow_exit();
2515         ovs_workqueues_exit();
2516 }
2517
2518 module_init(dp_init);
2519 module_exit(dp_cleanup);
2520
2521 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2522 MODULE_LICENSE("GPL");
2523 MODULE_VERSION(VERSION);