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, masked_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                 ovs_flow_key_mask(&masked_key, &key, &mask);
1283                 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
1284                                                   &masked_key, 0, &acts);
1285                 if (error) {
1286                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
1287                         goto err_kfree;
1288                 }
1289         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1290                 error = -EINVAL;
1291                 goto error;
1292         }
1293
1294         ovs_lock();
1295         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1296         error = -ENODEV;
1297         if (!dp)
1298                 goto err_unlock_ovs;
1299
1300         table = ovsl_dereference(dp->table);
1301
1302         /* Check if this is a duplicate flow */
1303         flow = ovs_flow_lookup(table, &key);
1304         if (!flow) {
1305                 struct sw_flow_mask *mask_p;
1306                 /* Bail out if we're not allowed to create a new flow. */
1307                 error = -ENOENT;
1308                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1309                         goto err_unlock_ovs;
1310
1311                 /* Expand table, if necessary, to make room. */
1312                 if (ovs_flow_tbl_need_to_expand(table)) {
1313                         struct flow_table *new_table;
1314
1315                         new_table = ovs_flow_tbl_expand(table);
1316                         if (!IS_ERR(new_table)) {
1317                                 rcu_assign_pointer(dp->table, new_table);
1318                                 ovs_flow_tbl_destroy(table, true);
1319                                 table = ovsl_dereference(dp->table);
1320                         }
1321                 }
1322
1323                 /* Allocate flow. */
1324                 flow = ovs_flow_alloc();
1325                 if (IS_ERR(flow)) {
1326                         error = PTR_ERR(flow);
1327                         goto err_unlock_ovs;
1328                 }
1329                 clear_stats(flow);
1330
1331                 flow->key = masked_key;
1332                 flow->unmasked_key = key;
1333
1334                 /* Make sure mask is unique in the system */
1335                 mask_p = ovs_sw_flow_mask_find(table, &mask);
1336                 if (!mask_p) {
1337                         /* Allocate a new mask if none exsits. */
1338                         mask_p = ovs_sw_flow_mask_alloc();
1339                         if (!mask_p)
1340                                 goto err_flow_free;
1341                         mask_p->key = mask.key;
1342                         mask_p->range = mask.range;
1343                         ovs_sw_flow_mask_insert(table, mask_p);
1344                 }
1345
1346                 ovs_sw_flow_mask_add_ref(mask_p);
1347                 rcu_assign_pointer(flow->mask, mask_p);
1348                 rcu_assign_pointer(flow->sf_acts, acts);
1349
1350                 /* Put flow in bucket. */
1351                 ovs_flow_insert(table, flow);
1352
1353                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1354                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
1355         } else {
1356                 /* We found a matching flow. */
1357                 struct sw_flow_actions *old_acts;
1358
1359                 /* Bail out if we're not allowed to modify an existing flow.
1360                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1361                  * because Generic Netlink treats the latter as a dump
1362                  * request.  We also accept NLM_F_EXCL in case that bug ever
1363                  * gets fixed.
1364                  */
1365                 error = -EEXIST;
1366                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1367                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1368                         goto err_unlock_ovs;
1369
1370                 /* The unmasked key has to be the same for flow updates. */
1371                 error = -EINVAL;
1372                 if (!ovs_flow_cmp_unmasked_key(flow, &key, match.range.end)) {
1373                         OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
1374                         goto err_unlock_ovs;
1375                 }
1376
1377                 /* Update actions. */
1378                 old_acts = ovsl_dereference(flow->sf_acts);
1379                 rcu_assign_pointer(flow->sf_acts, acts);
1380                 ovs_flow_deferred_free_acts(old_acts);
1381
1382                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1383                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1384
1385                 /* Clear stats. */
1386                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1387                         spin_lock_bh(&flow->lock);
1388                         clear_stats(flow);
1389                         spin_unlock_bh(&flow->lock);
1390                 }
1391         }
1392         ovs_unlock();
1393
1394         if (!IS_ERR(reply))
1395                 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1396         else
1397                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1398                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1399         return 0;
1400
1401 err_flow_free:
1402         ovs_flow_free(flow, false);
1403 err_unlock_ovs:
1404         ovs_unlock();
1405 err_kfree:
1406         kfree(acts);
1407 error:
1408         return error;
1409 }
1410
1411 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1412 {
1413         struct nlattr **a = info->attrs;
1414         struct ovs_header *ovs_header = info->userhdr;
1415         struct sw_flow_key key;
1416         struct sk_buff *reply;
1417         struct sw_flow *flow;
1418         struct datapath *dp;
1419         struct flow_table *table;
1420         struct sw_flow_match match;
1421         int err;
1422
1423         if (!a[OVS_FLOW_ATTR_KEY]) {
1424                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1425                 return -EINVAL;
1426         }
1427
1428         ovs_match_init(&match, &key, NULL);
1429         err = ovs_match_from_nlattrs(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1430         if (err)
1431                 return err;
1432
1433         ovs_lock();
1434         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1435         if (!dp) {
1436                 err = -ENODEV;
1437                 goto unlock;
1438         }
1439
1440         table = ovsl_dereference(dp->table);
1441         flow = ovs_flow_lookup_unmasked_key(table, &match);
1442         if (!flow) {
1443                 err = -ENOENT;
1444                 goto unlock;
1445         }
1446
1447         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1448                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1449         if (IS_ERR(reply)) {
1450                 err = PTR_ERR(reply);
1451                 goto unlock;
1452         }
1453
1454         ovs_unlock();
1455         return genlmsg_reply(reply, info);
1456 unlock:
1457         ovs_unlock();
1458         return err;
1459 }
1460
1461 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1462 {
1463         struct nlattr **a = info->attrs;
1464         struct ovs_header *ovs_header = info->userhdr;
1465         struct sw_flow_key key;
1466         struct sk_buff *reply;
1467         struct sw_flow *flow;
1468         struct datapath *dp;
1469         struct flow_table *table;
1470         struct sw_flow_match match;
1471         int err;
1472
1473         ovs_lock();
1474         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1475         if (!dp) {
1476                 err = -ENODEV;
1477                 goto unlock;
1478         }
1479
1480         if (!a[OVS_FLOW_ATTR_KEY]) {
1481                 err = flush_flows(dp);
1482                 goto unlock;
1483         }
1484
1485         ovs_match_init(&match, &key, NULL);
1486         err = ovs_match_from_nlattrs(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1487         if (err)
1488                 goto unlock;
1489
1490         table = ovsl_dereference(dp->table);
1491         flow = ovs_flow_lookup_unmasked_key(table, &match);
1492         if (!flow) {
1493                 err = -ENOENT;
1494                 goto unlock;
1495         }
1496
1497         reply = ovs_flow_cmd_alloc_info(flow);
1498         if (!reply) {
1499                 err = -ENOMEM;
1500                 goto unlock;
1501         }
1502
1503         ovs_flow_remove(table, flow);
1504
1505         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1506                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1507         BUG_ON(err < 0);
1508
1509         ovs_flow_free(flow, true);
1510         ovs_unlock();
1511
1512         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1513         return 0;
1514 unlock:
1515         ovs_unlock();
1516         return err;
1517 }
1518
1519 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1520 {
1521         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1522         struct datapath *dp;
1523         struct flow_table *table;
1524
1525         rcu_read_lock();
1526         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1527         if (!dp) {
1528                 rcu_read_unlock();
1529                 return -ENODEV;
1530         }
1531
1532         table = rcu_dereference(dp->table);
1533         for (;;) {
1534                 struct sw_flow *flow;
1535                 u32 bucket, obj;
1536
1537                 bucket = cb->args[0];
1538                 obj = cb->args[1];
1539                 flow = ovs_flow_dump_next(table, &bucket, &obj);
1540                 if (!flow)
1541                         break;
1542
1543                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1544                                            NETLINK_CB(cb->skb).portid,
1545                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1546                                            OVS_FLOW_CMD_NEW) < 0)
1547                         break;
1548
1549                 cb->args[0] = bucket;
1550                 cb->args[1] = obj;
1551         }
1552         rcu_read_unlock();
1553         return skb->len;
1554 }
1555
1556 static struct genl_ops dp_flow_genl_ops[] = {
1557         { .cmd = OVS_FLOW_CMD_NEW,
1558           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1559           .policy = flow_policy,
1560           .doit = ovs_flow_cmd_new_or_set
1561         },
1562         { .cmd = OVS_FLOW_CMD_DEL,
1563           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1564           .policy = flow_policy,
1565           .doit = ovs_flow_cmd_del
1566         },
1567         { .cmd = OVS_FLOW_CMD_GET,
1568           .flags = 0,               /* OK for unprivileged users. */
1569           .policy = flow_policy,
1570           .doit = ovs_flow_cmd_get,
1571           .dumpit = ovs_flow_cmd_dump
1572         },
1573         { .cmd = OVS_FLOW_CMD_SET,
1574           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1575           .policy = flow_policy,
1576           .doit = ovs_flow_cmd_new_or_set,
1577         },
1578 };
1579
1580 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1581 #ifdef HAVE_NLA_NUL_STRING
1582         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1583 #endif
1584         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1585 };
1586
1587 static struct genl_family dp_datapath_genl_family = {
1588         .id = GENL_ID_GENERATE,
1589         .hdrsize = sizeof(struct ovs_header),
1590         .name = OVS_DATAPATH_FAMILY,
1591         .version = OVS_DATAPATH_VERSION,
1592         .maxattr = OVS_DP_ATTR_MAX,
1593          SET_NETNSOK
1594 };
1595
1596 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1597         .name = OVS_DATAPATH_MCGROUP
1598 };
1599
1600 static size_t ovs_dp_cmd_msg_size(void)
1601 {
1602         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1603
1604         msgsize += nla_total_size(IFNAMSIZ);
1605         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1606
1607         return msgsize;
1608 }
1609
1610 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1611                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1612 {
1613         struct ovs_header *ovs_header;
1614         struct ovs_dp_stats dp_stats;
1615         int err;
1616
1617         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1618                                    flags, cmd);
1619         if (!ovs_header)
1620                 goto error;
1621
1622         ovs_header->dp_ifindex = get_dpifindex(dp);
1623
1624         rcu_read_lock();
1625         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1626         rcu_read_unlock();
1627         if (err)
1628                 goto nla_put_failure;
1629
1630         get_dp_stats(dp, &dp_stats);
1631         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1632                 goto nla_put_failure;
1633
1634         return genlmsg_end(skb, ovs_header);
1635
1636 nla_put_failure:
1637         genlmsg_cancel(skb, ovs_header);
1638 error:
1639         return -EMSGSIZE;
1640 }
1641
1642 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1643                                              u32 seq, u8 cmd)
1644 {
1645         struct sk_buff *skb;
1646         int retval;
1647
1648         skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1649         if (!skb)
1650                 return ERR_PTR(-ENOMEM);
1651
1652         retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1653         if (retval < 0) {
1654                 kfree_skb(skb);
1655                 return ERR_PTR(retval);
1656         }
1657         return skb;
1658 }
1659
1660 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1661 {
1662         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1663 }
1664
1665 /* Called with ovs_mutex. */
1666 static struct datapath *lookup_datapath(struct net *net,
1667                                         struct ovs_header *ovs_header,
1668                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1669 {
1670         struct datapath *dp;
1671
1672         if (!a[OVS_DP_ATTR_NAME])
1673                 dp = get_dp(net, ovs_header->dp_ifindex);
1674         else {
1675                 struct vport *vport;
1676
1677                 rcu_read_lock();
1678                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1679                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1680                 rcu_read_unlock();
1681         }
1682         return dp ? dp : ERR_PTR(-ENODEV);
1683 }
1684
1685 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1686 {
1687         struct nlattr **a = info->attrs;
1688         struct vport_parms parms;
1689         struct sk_buff *reply;
1690         struct datapath *dp;
1691         struct vport *vport;
1692         struct ovs_net *ovs_net;
1693         int err, i;
1694
1695         err = -EINVAL;
1696         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1697                 goto err;
1698
1699         err = ovs_dp_cmd_validate(a);
1700         if (err)
1701                 goto err;
1702
1703         ovs_lock();
1704
1705         err = -ENOMEM;
1706         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1707         if (dp == NULL)
1708                 goto err_unlock_ovs;
1709
1710         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1711
1712         /* Allocate table. */
1713         err = -ENOMEM;
1714         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1715         if (!dp->table)
1716                 goto err_free_dp;
1717
1718         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1719         if (!dp->stats_percpu) {
1720                 err = -ENOMEM;
1721                 goto err_destroy_table;
1722         }
1723
1724         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1725                             GFP_KERNEL);
1726         if (!dp->ports) {
1727                 err = -ENOMEM;
1728                 goto err_destroy_percpu;
1729         }
1730
1731         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1732                 INIT_HLIST_HEAD(&dp->ports[i]);
1733
1734         /* Set up our datapath device. */
1735         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1736         parms.type = OVS_VPORT_TYPE_INTERNAL;
1737         parms.options = NULL;
1738         parms.dp = dp;
1739         parms.port_no = OVSP_LOCAL;
1740         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1741
1742         vport = new_vport(&parms);
1743         if (IS_ERR(vport)) {
1744                 err = PTR_ERR(vport);
1745                 if (err == -EBUSY)
1746                         err = -EEXIST;
1747
1748                 goto err_destroy_ports_array;
1749         }
1750
1751         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1752                                       info->snd_seq, OVS_DP_CMD_NEW);
1753         err = PTR_ERR(reply);
1754         if (IS_ERR(reply))
1755                 goto err_destroy_local_port;
1756
1757         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1758         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1759
1760         ovs_unlock();
1761
1762         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1763         return 0;
1764
1765 err_destroy_local_port:
1766         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1767 err_destroy_ports_array:
1768         kfree(dp->ports);
1769 err_destroy_percpu:
1770         free_percpu(dp->stats_percpu);
1771 err_destroy_table:
1772         ovs_flow_tbl_destroy(ovsl_dereference(dp->table), false);
1773 err_free_dp:
1774         release_net(ovs_dp_get_net(dp));
1775         kfree(dp);
1776 err_unlock_ovs:
1777         ovs_unlock();
1778 err:
1779         return err;
1780 }
1781
1782 /* Called with ovs_mutex. */
1783 static void __dp_destroy(struct datapath *dp)
1784 {
1785         int i;
1786
1787         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1788                 struct vport *vport;
1789                 struct hlist_node *n;
1790
1791                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1792                         if (vport->port_no != OVSP_LOCAL)
1793                                 ovs_dp_detach_port(vport);
1794         }
1795
1796         list_del_rcu(&dp->list_node);
1797
1798         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1799          * all port in datapath are destroyed first before freeing datapath.
1800          */
1801         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1802
1803         call_rcu(&dp->rcu, destroy_dp_rcu);
1804 }
1805
1806 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1807 {
1808         struct sk_buff *reply;
1809         struct datapath *dp;
1810         int err;
1811
1812         err = ovs_dp_cmd_validate(info->attrs);
1813         if (err)
1814                 return err;
1815
1816         ovs_lock();
1817         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1818         err = PTR_ERR(dp);
1819         if (IS_ERR(dp))
1820                 goto unlock;
1821
1822         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1823                                       info->snd_seq, OVS_DP_CMD_DEL);
1824         err = PTR_ERR(reply);
1825         if (IS_ERR(reply))
1826                 goto unlock;
1827
1828         __dp_destroy(dp);
1829         ovs_unlock();
1830
1831         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1832
1833         return 0;
1834 unlock:
1835         ovs_unlock();
1836         return err;
1837 }
1838
1839 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1840 {
1841         struct sk_buff *reply;
1842         struct datapath *dp;
1843         int err;
1844
1845         err = ovs_dp_cmd_validate(info->attrs);
1846         if (err)
1847                 return err;
1848
1849         ovs_lock();
1850         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1851         err = PTR_ERR(dp);
1852         if (IS_ERR(dp))
1853                 goto unlock;
1854
1855         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1856                                       info->snd_seq, OVS_DP_CMD_NEW);
1857         if (IS_ERR(reply)) {
1858                 err = PTR_ERR(reply);
1859                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1860                                 ovs_dp_datapath_multicast_group.id, err);
1861                 err = 0;
1862                 goto unlock;
1863         }
1864
1865         ovs_unlock();
1866         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1867
1868         return 0;
1869 unlock:
1870         ovs_unlock();
1871         return err;
1872 }
1873
1874 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1875 {
1876         struct sk_buff *reply;
1877         struct datapath *dp;
1878         int err;
1879
1880         err = ovs_dp_cmd_validate(info->attrs);
1881         if (err)
1882                 return err;
1883
1884         ovs_lock();
1885         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1886         if (IS_ERR(dp)) {
1887                 err = PTR_ERR(dp);
1888                 goto unlock;
1889         }
1890
1891         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1892                                       info->snd_seq, OVS_DP_CMD_NEW);
1893         if (IS_ERR(reply)) {
1894                 err = PTR_ERR(reply);
1895                 goto unlock;
1896         }
1897
1898         ovs_unlock();
1899         return genlmsg_reply(reply, info);
1900
1901 unlock:
1902         ovs_unlock();
1903         return err;
1904 }
1905
1906 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1907 {
1908         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1909         struct datapath *dp;
1910         int skip = cb->args[0];
1911         int i = 0;
1912
1913         rcu_read_lock();
1914         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1915                 if (i >= skip &&
1916                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1917                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1918                                          OVS_DP_CMD_NEW) < 0)
1919                         break;
1920                 i++;
1921         }
1922         rcu_read_unlock();
1923
1924         cb->args[0] = i;
1925
1926         return skb->len;
1927 }
1928
1929 static struct genl_ops dp_datapath_genl_ops[] = {
1930         { .cmd = OVS_DP_CMD_NEW,
1931           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1932           .policy = datapath_policy,
1933           .doit = ovs_dp_cmd_new
1934         },
1935         { .cmd = OVS_DP_CMD_DEL,
1936           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1937           .policy = datapath_policy,
1938           .doit = ovs_dp_cmd_del
1939         },
1940         { .cmd = OVS_DP_CMD_GET,
1941           .flags = 0,               /* OK for unprivileged users. */
1942           .policy = datapath_policy,
1943           .doit = ovs_dp_cmd_get,
1944           .dumpit = ovs_dp_cmd_dump
1945         },
1946         { .cmd = OVS_DP_CMD_SET,
1947           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1948           .policy = datapath_policy,
1949           .doit = ovs_dp_cmd_set,
1950         },
1951 };
1952
1953 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1954 #ifdef HAVE_NLA_NUL_STRING
1955         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1956         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1957 #else
1958         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1959 #endif
1960         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1961         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1962         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1963         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1964 };
1965
1966 static struct genl_family dp_vport_genl_family = {
1967         .id = GENL_ID_GENERATE,
1968         .hdrsize = sizeof(struct ovs_header),
1969         .name = OVS_VPORT_FAMILY,
1970         .version = OVS_VPORT_VERSION,
1971         .maxattr = OVS_VPORT_ATTR_MAX,
1972          SET_NETNSOK
1973 };
1974
1975 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1976         .name = OVS_VPORT_MCGROUP
1977 };
1978
1979 /* Called with ovs_mutex or RCU read lock. */
1980 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1981                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1982 {
1983         struct ovs_header *ovs_header;
1984         struct ovs_vport_stats vport_stats;
1985         int err;
1986
1987         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1988                                  flags, cmd);
1989         if (!ovs_header)
1990                 return -EMSGSIZE;
1991
1992         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1993
1994         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1995             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1996             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1997             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1998                 goto nla_put_failure;
1999
2000         ovs_vport_get_stats(vport, &vport_stats);
2001         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
2002                     &vport_stats))
2003                 goto nla_put_failure;
2004
2005         err = ovs_vport_get_options(vport, skb);
2006         if (err == -EMSGSIZE)
2007                 goto error;
2008
2009         return genlmsg_end(skb, ovs_header);
2010
2011 nla_put_failure:
2012         err = -EMSGSIZE;
2013 error:
2014         genlmsg_cancel(skb, ovs_header);
2015         return err;
2016 }
2017
2018 /* Called with ovs_mutex or RCU read lock. */
2019 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
2020                                          u32 seq, u8 cmd)
2021 {
2022         struct sk_buff *skb;
2023         int retval;
2024
2025         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2026         if (!skb)
2027                 return ERR_PTR(-ENOMEM);
2028
2029         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
2030         BUG_ON(retval < 0);
2031
2032         return skb;
2033 }
2034
2035 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2036 {
2037         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
2038 }
2039
2040 /* Called with ovs_mutex or RCU read lock. */
2041 static struct vport *lookup_vport(struct net *net,
2042                                   struct ovs_header *ovs_header,
2043                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2044 {
2045         struct datapath *dp;
2046         struct vport *vport;
2047
2048         if (a[OVS_VPORT_ATTR_NAME]) {
2049                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2050                 if (!vport)
2051                         return ERR_PTR(-ENODEV);
2052                 if (ovs_header->dp_ifindex &&
2053                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2054                         return ERR_PTR(-ENODEV);
2055                 return vport;
2056         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2057                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2058
2059                 if (port_no >= DP_MAX_PORTS)
2060                         return ERR_PTR(-EFBIG);
2061
2062                 dp = get_dp(net, ovs_header->dp_ifindex);
2063                 if (!dp)
2064                         return ERR_PTR(-ENODEV);
2065
2066                 vport = ovs_vport_ovsl_rcu(dp, port_no);
2067                 if (!vport)
2068                         return ERR_PTR(-ENODEV);
2069                 return vport;
2070         } else
2071                 return ERR_PTR(-EINVAL);
2072 }
2073
2074 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2075 {
2076         struct nlattr **a = info->attrs;
2077         struct ovs_header *ovs_header = info->userhdr;
2078         struct vport_parms parms;
2079         struct sk_buff *reply;
2080         struct vport *vport;
2081         struct datapath *dp;
2082         u32 port_no;
2083         int err;
2084
2085         err = -EINVAL;
2086         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2087             !a[OVS_VPORT_ATTR_UPCALL_PID])
2088                 goto exit;
2089
2090         err = ovs_vport_cmd_validate(a);
2091         if (err)
2092                 goto exit;
2093
2094         ovs_lock();
2095         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2096         err = -ENODEV;
2097         if (!dp)
2098                 goto exit_unlock;
2099
2100         if (a[OVS_VPORT_ATTR_PORT_NO]) {
2101                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2102
2103                 err = -EFBIG;
2104                 if (port_no >= DP_MAX_PORTS)
2105                         goto exit_unlock;
2106
2107                 vport = ovs_vport_ovsl(dp, port_no);
2108                 err = -EBUSY;
2109                 if (vport)
2110                         goto exit_unlock;
2111         } else {
2112                 for (port_no = 1; ; port_no++) {
2113                         if (port_no >= DP_MAX_PORTS) {
2114                                 err = -EFBIG;
2115                                 goto exit_unlock;
2116                         }
2117                         vport = ovs_vport_ovsl(dp, port_no);
2118                         if (!vport)
2119                                 break;
2120                 }
2121         }
2122
2123         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2124         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2125         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2126         parms.dp = dp;
2127         parms.port_no = port_no;
2128         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2129
2130         vport = new_vport(&parms);
2131         err = PTR_ERR(vport);
2132         if (IS_ERR(vport))
2133                 goto exit_unlock;
2134
2135         err = 0;
2136         if (a[OVS_VPORT_ATTR_STATS])
2137                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2138
2139         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
2140                                          OVS_VPORT_CMD_NEW);
2141         if (IS_ERR(reply)) {
2142                 err = PTR_ERR(reply);
2143                 ovs_dp_detach_port(vport);
2144                 goto exit_unlock;
2145         }
2146
2147         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2148
2149 exit_unlock:
2150         ovs_unlock();
2151 exit:
2152         return err;
2153 }
2154
2155 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2156 {
2157         struct nlattr **a = info->attrs;
2158         struct sk_buff *reply;
2159         struct vport *vport;
2160         int err;
2161
2162         err = ovs_vport_cmd_validate(a);
2163         if (err)
2164                 goto exit;
2165
2166         ovs_lock();
2167         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2168         err = PTR_ERR(vport);
2169         if (IS_ERR(vport))
2170                 goto exit_unlock;
2171
2172         if (a[OVS_VPORT_ATTR_TYPE] &&
2173             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2174                 err = -EINVAL;
2175                 goto exit_unlock;
2176         }
2177
2178         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2179         if (!reply) {
2180                 err = -ENOMEM;
2181                 goto exit_unlock;
2182         }
2183
2184         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2185                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2186                 if (err)
2187                         goto exit_free;
2188         }
2189
2190         if (a[OVS_VPORT_ATTR_STATS])
2191                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2192
2193         if (a[OVS_VPORT_ATTR_UPCALL_PID])
2194                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2195
2196         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2197                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2198         BUG_ON(err < 0);
2199
2200         ovs_unlock();
2201         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2202         return 0;
2203
2204 exit_free:
2205         kfree_skb(reply);
2206 exit_unlock:
2207         ovs_unlock();
2208 exit:
2209         return err;
2210 }
2211
2212 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2213 {
2214         struct nlattr **a = info->attrs;
2215         struct sk_buff *reply;
2216         struct vport *vport;
2217         int err;
2218
2219         err = ovs_vport_cmd_validate(a);
2220         if (err)
2221                 goto exit;
2222
2223         ovs_lock();
2224         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2225         err = PTR_ERR(vport);
2226         if (IS_ERR(vport))
2227                 goto exit_unlock;
2228
2229         if (vport->port_no == OVSP_LOCAL) {
2230                 err = -EINVAL;
2231                 goto exit_unlock;
2232         }
2233
2234         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2235                                          info->snd_seq, OVS_VPORT_CMD_DEL);
2236         err = PTR_ERR(reply);
2237         if (IS_ERR(reply))
2238                 goto exit_unlock;
2239
2240         err = 0;
2241         ovs_dp_detach_port(vport);
2242
2243         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2244
2245 exit_unlock:
2246         ovs_unlock();
2247 exit:
2248         return err;
2249 }
2250
2251 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2252 {
2253         struct nlattr **a = info->attrs;
2254         struct ovs_header *ovs_header = info->userhdr;
2255         struct sk_buff *reply;
2256         struct vport *vport;
2257         int err;
2258
2259         err = ovs_vport_cmd_validate(a);
2260         if (err)
2261                 goto exit;
2262
2263         rcu_read_lock();
2264         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2265         err = PTR_ERR(vport);
2266         if (IS_ERR(vport))
2267                 goto exit_unlock;
2268
2269         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2270                                          info->snd_seq, OVS_VPORT_CMD_NEW);
2271         err = PTR_ERR(reply);
2272         if (IS_ERR(reply))
2273                 goto exit_unlock;
2274
2275         rcu_read_unlock();
2276
2277         return genlmsg_reply(reply, info);
2278
2279 exit_unlock:
2280         rcu_read_unlock();
2281 exit:
2282         return err;
2283 }
2284
2285 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2286 {
2287         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2288         struct datapath *dp;
2289         int bucket = cb->args[0], skip = cb->args[1];
2290         int i, j = 0;
2291
2292         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2293         if (!dp)
2294                 return -ENODEV;
2295
2296         rcu_read_lock();
2297         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2298                 struct vport *vport;
2299
2300                 j = 0;
2301                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2302                         if (j >= skip &&
2303                             ovs_vport_cmd_fill_info(vport, skb,
2304                                                     NETLINK_CB(cb->skb).portid,
2305                                                     cb->nlh->nlmsg_seq,
2306                                                     NLM_F_MULTI,
2307                                                     OVS_VPORT_CMD_NEW) < 0)
2308                                 goto out;
2309
2310                         j++;
2311                 }
2312                 skip = 0;
2313         }
2314 out:
2315         rcu_read_unlock();
2316
2317         cb->args[0] = i;
2318         cb->args[1] = j;
2319
2320         return skb->len;
2321 }
2322
2323 static struct genl_ops dp_vport_genl_ops[] = {
2324         { .cmd = OVS_VPORT_CMD_NEW,
2325           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2326           .policy = vport_policy,
2327           .doit = ovs_vport_cmd_new
2328         },
2329         { .cmd = OVS_VPORT_CMD_DEL,
2330           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2331           .policy = vport_policy,
2332           .doit = ovs_vport_cmd_del
2333         },
2334         { .cmd = OVS_VPORT_CMD_GET,
2335           .flags = 0,               /* OK for unprivileged users. */
2336           .policy = vport_policy,
2337           .doit = ovs_vport_cmd_get,
2338           .dumpit = ovs_vport_cmd_dump
2339         },
2340         { .cmd = OVS_VPORT_CMD_SET,
2341           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2342           .policy = vport_policy,
2343           .doit = ovs_vport_cmd_set,
2344         },
2345 };
2346
2347 struct genl_family_and_ops {
2348         struct genl_family *family;
2349         struct genl_ops *ops;
2350         int n_ops;
2351         struct genl_multicast_group *group;
2352 };
2353
2354 static const struct genl_family_and_ops dp_genl_families[] = {
2355         { &dp_datapath_genl_family,
2356           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2357           &ovs_dp_datapath_multicast_group },
2358         { &dp_vport_genl_family,
2359           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2360           &ovs_dp_vport_multicast_group },
2361         { &dp_flow_genl_family,
2362           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2363           &ovs_dp_flow_multicast_group },
2364         { &dp_packet_genl_family,
2365           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2366           NULL },
2367 };
2368
2369 static void dp_unregister_genl(int n_families)
2370 {
2371         int i;
2372
2373         for (i = 0; i < n_families; i++)
2374                 genl_unregister_family(dp_genl_families[i].family);
2375 }
2376
2377 static int dp_register_genl(void)
2378 {
2379         int n_registered;
2380         int err;
2381         int i;
2382
2383         n_registered = 0;
2384         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2385                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2386
2387                 err = genl_register_family_with_ops(f->family, f->ops,
2388                                                     f->n_ops);
2389                 if (err)
2390                         goto error;
2391                 n_registered++;
2392
2393                 if (f->group) {
2394                         err = genl_register_mc_group(f->family, f->group);
2395                         if (err)
2396                                 goto error;
2397                 }
2398         }
2399
2400         return 0;
2401
2402 error:
2403         dp_unregister_genl(n_registered);
2404         return err;
2405 }
2406
2407 static void rehash_flow_table(struct work_struct *work)
2408 {
2409         struct datapath *dp;
2410         struct net *net;
2411
2412         ovs_lock();
2413         rtnl_lock();
2414         for_each_net(net) {
2415                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2416
2417                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2418                         struct flow_table *old_table = ovsl_dereference(dp->table);
2419                         struct flow_table *new_table;
2420
2421                         new_table = ovs_flow_tbl_rehash(old_table);
2422                         if (!IS_ERR(new_table)) {
2423                                 rcu_assign_pointer(dp->table, new_table);
2424                                 ovs_flow_tbl_destroy(old_table, true);
2425                         }
2426                 }
2427         }
2428         rtnl_unlock();
2429         ovs_unlock();
2430         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2431 }
2432
2433 static int __net_init ovs_init_net(struct net *net)
2434 {
2435         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2436
2437         INIT_LIST_HEAD(&ovs_net->dps);
2438         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2439         return 0;
2440 }
2441
2442 static void __net_exit ovs_exit_net(struct net *net)
2443 {
2444         struct datapath *dp, *dp_next;
2445         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2446
2447         ovs_lock();
2448         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2449                 __dp_destroy(dp);
2450         ovs_unlock();
2451
2452         cancel_work_sync(&ovs_net->dp_notify_work);
2453 }
2454
2455 static struct pernet_operations ovs_net_ops = {
2456         .init = ovs_init_net,
2457         .exit = ovs_exit_net,
2458         .id   = &ovs_net_id,
2459         .size = sizeof(struct ovs_net),
2460 };
2461
2462 DEFINE_COMPAT_PNET_REG_FUNC(device);
2463
2464 static int __init dp_init(void)
2465 {
2466         int err;
2467
2468         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2469
2470         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2471                 VERSION);
2472
2473         err = ovs_workqueues_init();
2474         if (err)
2475                 goto error;
2476
2477         err = ovs_flow_init();
2478         if (err)
2479                 goto error_wq;
2480
2481         err = ovs_vport_init();
2482         if (err)
2483                 goto error_flow_exit;
2484
2485         err = register_pernet_device(&ovs_net_ops);
2486         if (err)
2487                 goto error_vport_exit;
2488
2489         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2490         if (err)
2491                 goto error_netns_exit;
2492
2493         err = dp_register_genl();
2494         if (err < 0)
2495                 goto error_unreg_notifier;
2496
2497         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2498
2499         return 0;
2500
2501 error_unreg_notifier:
2502         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2503 error_netns_exit:
2504         unregister_pernet_device(&ovs_net_ops);
2505 error_vport_exit:
2506         ovs_vport_exit();
2507 error_flow_exit:
2508         ovs_flow_exit();
2509 error_wq:
2510         ovs_workqueues_exit();
2511 error:
2512         return err;
2513 }
2514
2515 static void dp_cleanup(void)
2516 {
2517         cancel_delayed_work_sync(&rehash_flow_wq);
2518         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2519         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2520         unregister_pernet_device(&ovs_net_ops);
2521         rcu_barrier();
2522         ovs_vport_exit();
2523         ovs_flow_exit();
2524         ovs_workqueues_exit();
2525 }
2526
2527 module_init(dp_init);
2528 module_exit(dp_cleanup);
2529
2530 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2531 MODULE_LICENSE("GPL");
2532 MODULE_VERSION(VERSION);