datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007-2012 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/version.h>
  40 #include <linux/ethtool.h>
  41 #include <linux/wait.h>
  42 #include <asm/div64.h>
  43 #include <linux/highmem.h>
  44 #include <linux/netfilter_bridge.h>
  45 #include <linux/netfilter_ipv4.h>
  46 #include <linux/inetdevice.h>
  47 #include <linux/list.h>
  48 #include <linux/openvswitch.h>
  49 #include <linux/rculist.h>
  50 #include <linux/dmi.h>
  51 #include <net/genetlink.h>
  52 #include <net/net_namespace.h>
  53 #include <net/netns/generic.h>
  54
  55 #include "checksum.h"
  56 #include "datapath.h"
  57 #include "flow.h"
  58 #include "genl_exec.h"
  59 #include "vlan.h"
  60 #include "tunnel.h"
  61 #include "vport-internal_dev.h"
  62
  63 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
  64     LINUX_VERSION_CODE >= KERNEL_VERSION(3,9,0)
  65 #error Kernels before 2.6.18 or after 3.8 are not supported by this version of Open vSwitch.
  66 #endif
  67
  68 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
  69 static void rehash_flow_table(struct work_struct *work);
  70 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
  71
  72 int ovs_net_id __read_mostly;
  73
  74 /**
  75  * DOC: Locking:
  76  *
  77  * Writes to device state (add/remove datapath, port, set operations on vports,
  78  * etc.) are protected by RTNL.
  79  *
  80  * Writes to other state (flow table modifications, set miscellaneous datapath
  81  * parameters, etc.) are protected by genl_mutex.  The RTNL lock nests inside
  82  * genl_mutex.
  83  *
  84  * Reads are protected by RCU.
  85  *
  86  * There are a few special cases (mostly stats) that have their own
  87  * synchronization but they nest under all of above and don't interact with
  88  * each other.
  89  */
  90
  91 static struct vport *new_vport(const struct vport_parms *);
  92 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
  93                              const struct dp_upcall_info *);
  94 static int queue_userspace_packet(struct net *, int dp_ifindex,
  95                                   struct sk_buff *,
  96                                   const struct dp_upcall_info *);
  97
  98 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  99 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 100 {
 101         struct datapath *dp = NULL;
 102         struct net_device *dev;
 103
 104         rcu_read_lock();
 105         dev = dev_get_by_index_rcu(net, dp_ifindex);
 106         if (dev) {
 107                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 108                 if (vport)
 109                         dp = vport->dp;
 110         }
 111         rcu_read_unlock();
 112
 113         return dp;
 114 }
 115
 116 /* Must be called with rcu_read_lock or RTNL lock. */
 117 const char *ovs_dp_name(const struct datapath *dp)
 118 {
 119         struct vport *vport = ovs_vport_rtnl_rcu(dp, OVSP_LOCAL);
 120         return vport->ops->get_name(vport);
 121 }
 122
 123 static int get_dpifindex(struct datapath *dp)
 124 {
 125         struct vport *local;
 126         int ifindex;
 127
 128         rcu_read_lock();
 129
 130         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 131         if (local)
 132                 ifindex = local->ops->get_ifindex(local);
 133         else
 134                 ifindex = 0;
 135
 136         rcu_read_unlock();
 137
 138         return ifindex;
 139 }
 140
 141 static void destroy_dp_rcu(struct rcu_head *rcu)
 142 {
 143         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 144
 145         ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
 146         free_percpu(dp->stats_percpu);
 147         release_net(ovs_dp_get_net(dp));
 148         kfree(dp->ports);
 149         kfree(dp);
 150 }
 151
 152 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 153                                             u16 port_no)
 154 {
 155         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 156 }
 157
 158 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 159 {
 160         struct vport *vport;
 161         struct hlist_node *n;
 162         struct hlist_head *head;
 163
 164         head = vport_hash_bucket(dp, port_no);
 165         hlist_for_each_entry_rcu(vport, n, head, dp_hash_node) {
 166                 if (vport->port_no == port_no)
 167                         return vport;
 168         }
 169         return NULL;
 170 }
 171
 172 /* Called with RTNL lock and genl_lock. */
 173 static struct vport *new_vport(const struct vport_parms *parms)
 174 {
 175         struct vport *vport;
 176
 177         vport = ovs_vport_add(parms);
 178         if (!IS_ERR(vport)) {
 179                 struct datapath *dp = parms->dp;
 180                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 181
 182                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 183         }
 184         return vport;
 185 }
 186
 187 /* Called with RTNL lock. */
 188 void ovs_dp_detach_port(struct vport *p)
 189 {
 190         ASSERT_RTNL();
 191
 192         /* First drop references to device. */
 193         hlist_del_rcu(&p->dp_hash_node);
 194
 195         /* Then destroy it. */
 196         ovs_vport_del(p);
 197 }
 198
 199 /* Must be called with rcu_read_lock. */
 200 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 201 {
 202         struct datapath *dp = p->dp;
 203         struct sw_flow *flow;
 204         struct dp_stats_percpu *stats;
 205         u64 *stats_counter;
 206         int error;
 207
 208         stats = this_cpu_ptr(dp->stats_percpu);
 209
 210         if (!OVS_CB(skb)->flow) {
 211                 struct sw_flow_key key;
 212                 int key_len;
 213
 214                 /* Extract flow from 'skb' into 'key'. */
 215                 error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
 216                 if (unlikely(error)) {
 217                         kfree_skb(skb);
 218                         return;
 219                 }
 220
 221                 /* Look up flow. */
 222                 flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table),
 223                                            &key, key_len);
 224                 if (unlikely(!flow)) {
 225                         struct dp_upcall_info upcall;
 226
 227                         upcall.cmd = OVS_PACKET_CMD_MISS;
 228                         upcall.key = &key;
 229                         upcall.userdata = NULL;
 230                         upcall.portid = p->upcall_portid;
 231                         ovs_dp_upcall(dp, skb, &upcall);
 232                         consume_skb(skb);
 233                         stats_counter = &stats->n_missed;
 234                         goto out;
 235                 }
 236
 237                 OVS_CB(skb)->flow = flow;
 238         }
 239
 240         stats_counter = &stats->n_hit;
 241         ovs_flow_used(OVS_CB(skb)->flow, skb);
 242         ovs_execute_actions(dp, skb);
 243
 244 out:
 245         /* Update datapath statistics. */
 246         u64_stats_update_begin(&stats->sync);
 247         (*stats_counter)++;
 248         u64_stats_update_end(&stats->sync);
 249 }
 250
 251 static struct genl_family dp_packet_genl_family = {
 252         .id = GENL_ID_GENERATE,
 253         .hdrsize = sizeof(struct ovs_header),
 254         .name = OVS_PACKET_FAMILY,
 255         .version = OVS_PACKET_VERSION,
 256         .maxattr = OVS_PACKET_ATTR_MAX,
 257          SET_NETNSOK
 258 };
 259
 260 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 261                   const struct dp_upcall_info *upcall_info)
 262 {
 263         struct dp_stats_percpu *stats;
 264         int dp_ifindex;
 265         int err;
 266
 267         if (upcall_info->portid == 0) {
 268                 err = -ENOTCONN;
 269                 goto err;
 270         }
 271
 272         dp_ifindex = get_dpifindex(dp);
 273         if (!dp_ifindex) {
 274                 err = -ENODEV;
 275                 goto err;
 276         }
 277
 278         forward_ip_summed(skb, true);
 279
 280         if (!skb_is_gso(skb))
 281                 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 282         else
 283                 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 284         if (err)
 285                 goto err;
 286
 287         return 0;
 288
 289 err:
 290         stats = this_cpu_ptr(dp->stats_percpu);
 291
 292         u64_stats_update_begin(&stats->sync);
 293         stats->n_lost++;
 294         u64_stats_update_end(&stats->sync);
 295
 296         return err;
 297 }
 298
 299 static int queue_gso_packets(struct net *net, int dp_ifindex,
 300                              struct sk_buff *skb,
 301                              const struct dp_upcall_info *upcall_info)
 302 {
 303         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 304         struct dp_upcall_info later_info;
 305         struct sw_flow_key later_key;
 306         struct sk_buff *segs, *nskb;
 307         int err;
 308
 309         segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 310         if (IS_ERR(segs))
 311                 return PTR_ERR(segs);
 312
 313         /* Queue all of the segments. */
 314         skb = segs;
 315         do {
 316                 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
 317                 if (err)
 318                         break;
 319
 320                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 321                         /* The initial flow key extracted by ovs_flow_extract()
 322                          * in this case is for a first fragment, so we need to
 323                          * properly mark later fragments.
 324                          */
 325                         later_key = *upcall_info->key;
 326                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 327
 328                         later_info = *upcall_info;
 329                         later_info.key = &later_key;
 330                         upcall_info = &later_info;
 331                 }
 332         } while ((skb = skb->next));
 333
 334         /* Free all of the segments. */
 335         skb = segs;
 336         do {
 337                 nskb = skb->next;
 338                 if (err)
 339                         kfree_skb(skb);
 340                 else
 341                         consume_skb(skb);
 342         } while ((skb = nskb));
 343         return err;
 344 }
 345
 346 static int queue_userspace_packet(struct net *net, int dp_ifindex,
 347                                   struct sk_buff *skb,
 348                                   const struct dp_upcall_info *upcall_info)
 349 {
 350         struct ovs_header *upcall;
 351         struct sk_buff *nskb = NULL;
 352         struct sk_buff *user_skb; /* to be queued to userspace */
 353         struct nlattr *nla;
 354         unsigned int len;
 355         int err;
 356
 357         if (vlan_tx_tag_present(skb)) {
 358                 nskb = skb_clone(skb, GFP_ATOMIC);
 359                 if (!nskb)
 360                         return -ENOMEM;
 361
 362                 err = vlan_deaccel_tag(nskb);
 363                 if (err)
 364                         return err;
 365
 366                 skb = nskb;
 367         }
 368
 369         if (nla_attr_size(skb->len) > USHRT_MAX) {
 370                 err = -EFBIG;
 371                 goto out;
 372         }
 373
 374         len = sizeof(struct ovs_header);
 375         len += nla_total_size(skb->len);
 376         len += nla_total_size(FLOW_BUFSIZE);
 377         if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
 378                 len += nla_total_size(8);
 379
 380         user_skb = genlmsg_new(len, GFP_ATOMIC);
 381         if (!user_skb) {
 382                 err = -ENOMEM;
 383                 goto out;
 384         }
 385
 386         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 387                              0, upcall_info->cmd);
 388         upcall->dp_ifindex = dp_ifindex;
 389
 390         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 391         ovs_flow_to_nlattrs(upcall_info->key, user_skb);
 392         nla_nest_end(user_skb, nla);
 393
 394         if (upcall_info->userdata)
 395                 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
 396                             nla_get_u64(upcall_info->userdata));
 397
 398         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 399
 400         skb_copy_and_csum_dev(skb, nla_data(nla));
 401
 402         err = genlmsg_unicast(net, user_skb, upcall_info->portid);
 403
 404 out:
 405         kfree_skb(nskb);
 406         return err;
 407 }
 408
 409 /* Called with genl_mutex. */
 410 static int flush_flows(struct datapath *dp)
 411 {
 412         struct flow_table *old_table;
 413         struct flow_table *new_table;
 414
 415         old_table = genl_dereference(dp->table);
 416         new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
 417         if (!new_table)
 418                 return -ENOMEM;
 419
 420         rcu_assign_pointer(dp->table, new_table);
 421
 422         ovs_flow_tbl_deferred_destroy(old_table);
 423         return 0;
 424 }
 425
 426 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa, int attr_len)
 427 {
 428
 429         struct sw_flow_actions *acts;
 430         int new_acts_size;
 431         int req_size = NLA_ALIGN(attr_len);
 432         int next_offset = offsetof(struct sw_flow_actions, actions) +
 433                                         (*sfa)->actions_len;
 434
 435         if (req_size <= (ksize(*sfa) - next_offset))
 436                 goto out;
 437
 438         new_acts_size = ksize(*sfa) * 2;
 439
 440         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
 441                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
 442                         return ERR_PTR(-EMSGSIZE);
 443                 new_acts_size = MAX_ACTIONS_BUFSIZE;
 444         }
 445
 446         acts = ovs_flow_actions_alloc(new_acts_size);
 447         if (IS_ERR(acts))
 448                 return (void *)acts;
 449
 450         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
 451         acts->actions_len = (*sfa)->actions_len;
 452         kfree(*sfa);
 453         *sfa = acts;
 454
 455 out:
 456         (*sfa)->actions_len += req_size;
 457         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
 458 }
 459
 460 static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, int len)
 461 {
 462         struct nlattr *a;
 463
 464         a = reserve_sfa_size(sfa, nla_attr_size(len));
 465         if (IS_ERR(a))
 466                 return PTR_ERR(a);
 467
 468         a->nla_type = attrtype;
 469         a->nla_len = nla_attr_size(len);
 470
 471         if (data)
 472                 memcpy(nla_data(a), data, len);
 473         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
 474
 475         return 0;
 476 }
 477
 478 static inline int add_nested_action_start(struct sw_flow_actions **sfa, int attrtype)
 479 {
 480         int used = (*sfa)->actions_len;
 481         int err;
 482
 483         err = add_action(sfa, attrtype, NULL, 0);
 484         if (err)
 485                 return err;
 486
 487         return used;
 488 }
 489
 490 static inline void add_nested_action_end(struct sw_flow_actions *sfa, int st_offset)
 491 {
 492         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions + st_offset);
 493
 494         a->nla_len = sfa->actions_len - st_offset;
 495 }
 496
 497 static int validate_and_copy_actions(const struct nlattr *attr,
 498                                 const struct sw_flow_key *key, int depth,
 499                                 struct sw_flow_actions **sfa);
 500
 501 static int validate_and_copy_sample(const struct nlattr *attr,
 502                            const struct sw_flow_key *key, int depth,
 503                            struct sw_flow_actions **sfa)
 504 {
 505         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 506         const struct nlattr *probability, *actions;
 507         const struct nlattr *a;
 508         int rem, start, err, st_acts;
 509
 510         memset(attrs, 0, sizeof(attrs));
 511         nla_for_each_nested(a, attr, rem) {
 512                 int type = nla_type(a);
 513                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 514                         return -EINVAL;
 515                 attrs[type] = a;
 516         }
 517         if (rem)
 518                 return -EINVAL;
 519
 520         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 521         if (!probability || nla_len(probability) != sizeof(u32))
 522                 return -EINVAL;
 523
 524         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 525         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 526                 return -EINVAL;
 527
 528         /* validation done, copy sample action. */
 529         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
 530         if (start < 0)
 531                 return start;
 532         err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY, nla_data(probability), sizeof(u32));
 533         if (err)
 534                 return err;
 535         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
 536         if (st_acts < 0)
 537                 return st_acts;
 538
 539         err = validate_and_copy_actions(actions, key, depth + 1, sfa);
 540         if (err)
 541                 return err;
 542
 543         add_nested_action_end(*sfa, st_acts);
 544         add_nested_action_end(*sfa, start);
 545
 546         return 0;
 547 }
 548
 549 static int validate_tp_port(const struct sw_flow_key *flow_key)
 550 {
 551         if (flow_key->eth.type == htons(ETH_P_IP)) {
 552                 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
 553                         return 0;
 554         } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
 555                 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
 556                         return 0;
 557         }
 558
 559         return -EINVAL;
 560 }
 561
 562 static int validate_and_copy_set_tun(const struct nlattr *attr,
 563                                      struct sw_flow_actions **sfa)
 564 {
 565         struct ovs_key_ipv4_tunnel tun_key;
 566         int err, start;
 567
 568         err = ipv4_tun_from_nlattr(nla_data(attr), &tun_key);
 569         if (err)
 570                 return err;
 571
 572         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
 573         if (start < 0)
 574                 return start;
 575
 576         err = add_action(sfa, OVS_KEY_ATTR_IPV4_TUNNEL, &tun_key, sizeof(tun_key));
 577         add_nested_action_end(*sfa, start);
 578
 579         return err;
 580 }
 581
 582 static int validate_set(const struct nlattr *a,
 583                         const struct sw_flow_key *flow_key,
 584                         struct sw_flow_actions **sfa,
 585                         bool *set_tun)
 586 {
 587         const struct nlattr *ovs_key = nla_data(a);
 588         int key_type = nla_type(ovs_key);
 589
 590         /* There can be only one key in a action */
 591         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 592                 return -EINVAL;
 593
 594         if (key_type > OVS_KEY_ATTR_MAX ||
 595             (ovs_key_lens[key_type] != nla_len(ovs_key) &&
 596              ovs_key_lens[key_type] != -1))
 597                 return -EINVAL;
 598
 599         switch (key_type) {
 600         const struct ovs_key_ipv4 *ipv4_key;
 601         const struct ovs_key_ipv6 *ipv6_key;
 602         int err;
 603
 604         case OVS_KEY_ATTR_PRIORITY:
 605         case OVS_KEY_ATTR_TUN_ID:
 606         case OVS_KEY_ATTR_ETHERNET:
 607                 break;
 608
 609         case OVS_KEY_ATTR_SKB_MARK:
 610 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
 611                 if (nla_get_u32(ovs_key) != 0)
 612                         return -EINVAL;
 613 #endif
 614                 break;
 615
 616         case OVS_KEY_ATTR_TUNNEL:
 617                 *set_tun = true;
 618                 err = validate_and_copy_set_tun(a, sfa);
 619                 if (err)
 620                         return err;
 621                 break;
 622
 623         case OVS_KEY_ATTR_IPV4:
 624                 if (flow_key->eth.type != htons(ETH_P_IP))
 625                         return -EINVAL;
 626
 627                 if (!flow_key->ip.proto)
 628                         return -EINVAL;
 629
 630                 ipv4_key = nla_data(ovs_key);
 631                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 632                         return -EINVAL;
 633
 634                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 635                         return -EINVAL;
 636
 637                 break;
 638
 639         case OVS_KEY_ATTR_IPV6:
 640                 if (flow_key->eth.type != htons(ETH_P_IPV6))
 641                         return -EINVAL;
 642
 643                 if (!flow_key->ip.proto)
 644                         return -EINVAL;
 645
 646                 ipv6_key = nla_data(ovs_key);
 647                 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
 648                         return -EINVAL;
 649
 650                 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
 651                         return -EINVAL;
 652
 653                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
 654                         return -EINVAL;
 655
 656                 break;
 657
 658         case OVS_KEY_ATTR_TCP:
 659                 if (flow_key->ip.proto != IPPROTO_TCP)
 660                         return -EINVAL;
 661
 662                 return validate_tp_port(flow_key);
 663
 664         case OVS_KEY_ATTR_UDP:
 665                 if (flow_key->ip.proto != IPPROTO_UDP)
 666                         return -EINVAL;
 667
 668                 return validate_tp_port(flow_key);
 669
 670         default:
 671                 return -EINVAL;
 672         }
 673
 674         return 0;
 675 }
 676
 677 static int validate_userspace(const struct nlattr *attr)
 678 {
 679         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 680                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 681                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
 682         };
 683         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 684         int error;
 685
 686         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 687                                  attr, userspace_policy);
 688         if (error)
 689                 return error;
 690
 691         if (!a[OVS_USERSPACE_ATTR_PID] ||
 692             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 693                 return -EINVAL;
 694
 695         return 0;
 696 }
 697
 698 static int copy_action(const struct nlattr *from,
 699                       struct sw_flow_actions **sfa)
 700 {
 701         int totlen = NLA_ALIGN(from->nla_len);
 702         struct nlattr *to;
 703
 704         to = reserve_sfa_size(sfa, from->nla_len);
 705         if (IS_ERR(to))
 706                 return PTR_ERR(to);
 707
 708         memcpy(to, from, totlen);
 709         return 0;
 710 }
 711
 712 static int validate_and_copy_actions(const struct nlattr *attr,
 713                                 const struct sw_flow_key *key,
 714                                 int depth,
 715                                 struct sw_flow_actions **sfa)
 716 {
 717         const struct nlattr *a;
 718         int rem, err;
 719
 720         if (depth >= SAMPLE_ACTION_DEPTH)
 721                 return -EOVERFLOW;
 722
 723         nla_for_each_nested(a, attr, rem) {
 724                 /* Expected argument lengths, (u32)-1 for variable length. */
 725                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 726                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 727                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 728                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 729                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 730                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 731                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 732                 };
 733                 const struct ovs_action_push_vlan *vlan;
 734                 int type = nla_type(a);
 735                 bool skip_copy;
 736
 737                 if (type > OVS_ACTION_ATTR_MAX ||
 738                     (action_lens[type] != nla_len(a) &&
 739                      action_lens[type] != (u32)-1))
 740                         return -EINVAL;
 741
 742                 skip_copy = false;
 743                 switch (type) {
 744                 case OVS_ACTION_ATTR_UNSPEC:
 745                         return -EINVAL;
 746
 747                 case OVS_ACTION_ATTR_USERSPACE:
 748                         err = validate_userspace(a);
 749                         if (err)
 750                                 return err;
 751                         break;
 752
 753                 case OVS_ACTION_ATTR_OUTPUT:
 754                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 755                                 return -EINVAL;
 756                         break;
 757
 758
 759                 case OVS_ACTION_ATTR_POP_VLAN:
 760                         break;
 761
 762                 case OVS_ACTION_ATTR_PUSH_VLAN:
 763                         vlan = nla_data(a);
 764                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 765                                 return -EINVAL;
 766                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 767                                 return -EINVAL;
 768                         break;
 769
 770                 case OVS_ACTION_ATTR_SET:
 771                         err = validate_set(a, key, sfa, &skip_copy);
 772                         if (err)
 773                                 return err;
 774                         break;
 775
 776                 case OVS_ACTION_ATTR_SAMPLE:
 777                         err = validate_and_copy_sample(a, key, depth, sfa);
 778                         if (err)
 779                                 return err;
 780                         skip_copy = true;
 781                         break;
 782
 783                 default:
 784                         return -EINVAL;
 785                 }
 786                 if (!skip_copy) {
 787                         err = copy_action(a, sfa);
 788                         if (err)
 789                                 return err;
 790                 }
 791         }
 792
 793         if (rem > 0)
 794                 return -EINVAL;
 795
 796         return 0;
 797 }
 798
 799 static void clear_stats(struct sw_flow *flow)
 800 {
 801         flow->used = 0;
 802         flow->tcp_flags = 0;
 803         flow->packet_count = 0;
 804         flow->byte_count = 0;
 805 }
 806
 807 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 808 {
 809         struct ovs_header *ovs_header = info->userhdr;
 810         struct nlattr **a = info->attrs;
 811         struct sw_flow_actions *acts;
 812         struct sk_buff *packet;
 813         struct sw_flow *flow;
 814         struct datapath *dp;
 815         struct ethhdr *eth;
 816         int len;
 817         int err;
 818         int key_len;
 819
 820         err = -EINVAL;
 821         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 822             !a[OVS_PACKET_ATTR_ACTIONS] ||
 823             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 824                 goto err;
 825
 826         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 827         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 828         err = -ENOMEM;
 829         if (!packet)
 830                 goto err;
 831         skb_reserve(packet, NET_IP_ALIGN);
 832
 833         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 834
 835         skb_reset_mac_header(packet);
 836         eth = eth_hdr(packet);
 837
 838         /* Normally, setting the skb 'protocol' field would be handled by a
 839          * call to eth_type_trans(), but it assumes there's a sending
 840          * device, which we may not have. */
 841         if (ntohs(eth->h_proto) >= 1536)
 842                 packet->protocol = eth->h_proto;
 843         else
 844                 packet->protocol = htons(ETH_P_802_2);
 845
 846         /* Build an sw_flow for sending this packet. */
 847         flow = ovs_flow_alloc();
 848         err = PTR_ERR(flow);
 849         if (IS_ERR(flow))
 850                 goto err_kfree_skb;
 851
 852         err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
 853         if (err)
 854                 goto err_flow_free;
 855
 856         err = ovs_flow_metadata_from_nlattrs(flow, key_len, a[OVS_PACKET_ATTR_KEY]);
 857         if (err)
 858                 goto err_flow_free;
 859         acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 860         err = PTR_ERR(acts);
 861         if (IS_ERR(acts))
 862                 goto err_flow_free;
 863
 864         err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
 865         rcu_assign_pointer(flow->sf_acts, acts);
 866         if (err)
 867                 goto err_flow_free;
 868
 869         OVS_CB(packet)->flow = flow;
 870         packet->priority = flow->key.phy.priority;
 871         skb_set_mark(packet, flow->key.phy.skb_mark);
 872
 873         rcu_read_lock();
 874         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 875         err = -ENODEV;
 876         if (!dp)
 877                 goto err_unlock;
 878
 879         local_bh_disable();
 880         err = ovs_execute_actions(dp, packet);
 881         local_bh_enable();
 882         rcu_read_unlock();
 883
 884         ovs_flow_free(flow);
 885         return err;
 886
 887 err_unlock:
 888         rcu_read_unlock();
 889 err_flow_free:
 890         ovs_flow_free(flow);
 891 err_kfree_skb:
 892         kfree_skb(packet);
 893 err:
 894         return err;
 895 }
 896
 897 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 898         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 899         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 900         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 901 };
 902
 903 static struct genl_ops dp_packet_genl_ops[] = {
 904         { .cmd = OVS_PACKET_CMD_EXECUTE,
 905           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 906           .policy = packet_policy,
 907           .doit = ovs_packet_cmd_execute
 908         }
 909 };
 910
 911 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 912 {
 913         int i;
 914         struct flow_table *table = genl_dereference(dp->table);
 915
 916         stats->n_flows = ovs_flow_tbl_count(table);
 917
 918         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 919         for_each_possible_cpu(i) {
 920                 const struct dp_stats_percpu *percpu_stats;
 921                 struct dp_stats_percpu local_stats;
 922                 unsigned int start;
 923
 924                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 925
 926                 do {
 927                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 928                         local_stats = *percpu_stats;
 929                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 930
 931                 stats->n_hit += local_stats.n_hit;
 932                 stats->n_missed += local_stats.n_missed;
 933                 stats->n_lost += local_stats.n_lost;
 934         }
 935 }
 936
 937 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 938         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 939         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 940         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 941 };
 942
 943 static struct genl_family dp_flow_genl_family = {
 944         .id = GENL_ID_GENERATE,
 945         .hdrsize = sizeof(struct ovs_header),
 946         .name = OVS_FLOW_FAMILY,
 947         .version = OVS_FLOW_VERSION,
 948         .maxattr = OVS_FLOW_ATTR_MAX,
 949          SET_NETNSOK
 950 };
 951
 952 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 953         .name = OVS_FLOW_MCGROUP
 954 };
 955
 956 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
 957 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
 958 {
 959         const struct nlattr *a;
 960         struct nlattr *start;
 961         int err = 0, rem;
 962
 963         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
 964         if (!start)
 965                 return -EMSGSIZE;
 966
 967         nla_for_each_nested(a, attr, rem) {
 968                 int type = nla_type(a);
 969                 struct nlattr *st_sample;
 970
 971                 switch (type) {
 972                 case OVS_SAMPLE_ATTR_PROBABILITY:
 973                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
 974                                 return -EMSGSIZE;
 975                         break;
 976                 case OVS_SAMPLE_ATTR_ACTIONS:
 977                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
 978                         if (!st_sample)
 979                                 return -EMSGSIZE;
 980                         err = actions_to_attr(nla_data(a), nla_len(a), skb);
 981                         if (err)
 982                                 return err;
 983                         nla_nest_end(skb, st_sample);
 984                         break;
 985                 }
 986         }
 987
 988         nla_nest_end(skb, start);
 989         return err;
 990 }
 991
 992 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
 993 {
 994         const struct nlattr *ovs_key = nla_data(a);
 995         int key_type = nla_type(ovs_key);
 996         struct nlattr *start;
 997         int err;
 998
 999         switch (key_type) {
1000         case OVS_KEY_ATTR_IPV4_TUNNEL:
1001                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1002                 if (!start)
1003                         return -EMSGSIZE;
1004
1005                 err = ipv4_tun_to_nlattr(skb, nla_data(ovs_key));
1006                 if (err)
1007                         return err;
1008                 nla_nest_end(skb, start);
1009                 break;
1010         default:
1011                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1012                         return -EMSGSIZE;
1013                 break;
1014         }
1015
1016         return 0;
1017 }
1018
1019 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1020 {
1021         const struct nlattr *a;
1022         int rem, err;
1023
1024         nla_for_each_attr(a, attr, len, rem) {
1025                 int type = nla_type(a);
1026
1027                 switch (type) {
1028                 case OVS_ACTION_ATTR_SET:
1029                         err = set_action_to_attr(a, skb);
1030                         if (err)
1031                                 return err;
1032                         break;
1033
1034                 case OVS_ACTION_ATTR_SAMPLE:
1035                         err = sample_action_to_attr(a, skb);
1036                         if (err)
1037                                 return err;
1038                         break;
1039                 default:
1040                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
1041                                 return -EMSGSIZE;
1042                         break;
1043                 }
1044         }
1045
1046         return 0;
1047 }
1048
1049 /* Called with genl_lock. */
1050 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1051                                   struct sk_buff *skb, u32 portid,
1052                                   u32 seq, u32 flags, u8 cmd)
1053 {
1054         const int skb_orig_len = skb->len;
1055         const struct sw_flow_actions *sf_acts;
1056         struct nlattr *start;
1057         struct ovs_flow_stats stats;
1058         struct ovs_header *ovs_header;
1059         struct nlattr *nla;
1060         unsigned long used;
1061         u8 tcp_flags;
1062         int err;
1063
1064         sf_acts = rcu_dereference_protected(flow->sf_acts,
1065                                             lockdep_genl_is_held());
1066
1067         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1068         if (!ovs_header)
1069                 return -EMSGSIZE;
1070
1071         ovs_header->dp_ifindex = get_dpifindex(dp);
1072
1073         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1074         if (!nla)
1075                 goto nla_put_failure;
1076         err = ovs_flow_to_nlattrs(&flow->key, skb);
1077         if (err)
1078                 goto error;
1079         nla_nest_end(skb, nla);
1080
1081         spin_lock_bh(&flow->lock);
1082         used = flow->used;
1083         stats.n_packets = flow->packet_count;
1084         stats.n_bytes = flow->byte_count;
1085         tcp_flags = flow->tcp_flags;
1086         spin_unlock_bh(&flow->lock);
1087
1088         if (used &&
1089             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1090                 goto nla_put_failure;
1091
1092         if (stats.n_packets &&
1093             nla_put(skb, OVS_FLOW_ATTR_STATS,
1094                     sizeof(struct ovs_flow_stats), &stats))
1095                 goto nla_put_failure;
1096
1097         if (tcp_flags &&
1098             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1099                 goto nla_put_failure;
1100
1101         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1102          * this is the first flow to be dumped into 'skb'.  This is unusual for
1103          * Netlink but individual action lists can be longer than
1104          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1105          * The userspace caller can always fetch the actions separately if it
1106          * really wants them.  (Most userspace callers in fact don't care.)
1107          *
1108          * This can only fail for dump operations because the skb is always
1109          * properly sized for single flows.
1110          */
1111         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1112         if (start) {
1113                 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1114                 if (!err)
1115                         nla_nest_end(skb, start);
1116                 else {
1117                         if (skb_orig_len)
1118                                 goto error;
1119
1120                         nla_nest_cancel(skb, start);
1121                 }
1122         } else if (skb_orig_len) {
1123                 err = -ENOMEM;
1124                 goto error;
1125         }
1126
1127         return genlmsg_end(skb, ovs_header);
1128
1129 nla_put_failure:
1130         err = -EMSGSIZE;
1131 error:
1132         genlmsg_cancel(skb, ovs_header);
1133         return err;
1134 }
1135
1136 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1137 {
1138         const struct sw_flow_actions *sf_acts;
1139         int len;
1140
1141         sf_acts = rcu_dereference_protected(flow->sf_acts,
1142                                             lockdep_genl_is_held());
1143
1144         /* OVS_FLOW_ATTR_KEY */
1145         len = nla_total_size(FLOW_BUFSIZE);
1146         /* OVS_FLOW_ATTR_ACTIONS */
1147         len += nla_total_size(sf_acts->actions_len);
1148         /* OVS_FLOW_ATTR_STATS */
1149         len += nla_total_size(sizeof(struct ovs_flow_stats));
1150         /* OVS_FLOW_ATTR_TCP_FLAGS */
1151         len += nla_total_size(1);
1152         /* OVS_FLOW_ATTR_USED */
1153         len += nla_total_size(8);
1154
1155         len += NLMSG_ALIGN(sizeof(struct ovs_header));
1156
1157         return genlmsg_new(len, GFP_KERNEL);
1158 }
1159
1160 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1161                                                struct datapath *dp,
1162                                                u32 portid, u32 seq, u8 cmd)
1163 {
1164         struct sk_buff *skb;
1165         int retval;
1166
1167         skb = ovs_flow_cmd_alloc_info(flow);
1168         if (!skb)
1169                 return ERR_PTR(-ENOMEM);
1170
1171         retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1172         BUG_ON(retval < 0);
1173         return skb;
1174 }
1175
1176 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1177 {
1178         struct nlattr **a = info->attrs;
1179         struct ovs_header *ovs_header = info->userhdr;
1180         struct sw_flow_key key;
1181         struct sw_flow *flow;
1182         struct sk_buff *reply;
1183         struct datapath *dp;
1184         struct flow_table *table;
1185         struct sw_flow_actions *acts = NULL;
1186         int error;
1187         int key_len;
1188
1189         /* Extract key. */
1190         error = -EINVAL;
1191         if (!a[OVS_FLOW_ATTR_KEY])
1192                 goto error;
1193         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1194         if (error)
1195                 goto error;
1196
1197         /* Validate actions. */
1198         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1199                 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1200                 error = PTR_ERR(acts);
1201                 if (IS_ERR(acts))
1202                         goto error;
1203
1204                 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0, &acts);
1205                 if (error)
1206                         goto err_kfree;
1207         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1208                 error = -EINVAL;
1209                 goto error;
1210         }
1211
1212         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1213         error = -ENODEV;
1214         if (!dp)
1215                 goto err_kfree;
1216
1217         table = genl_dereference(dp->table);
1218         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1219         if (!flow) {
1220                 /* Bail out if we're not allowed to create a new flow. */
1221                 error = -ENOENT;
1222                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1223                         goto err_kfree;
1224
1225                 /* Expand table, if necessary, to make room. */
1226                 if (ovs_flow_tbl_need_to_expand(table)) {
1227                         struct flow_table *new_table;
1228
1229                         new_table = ovs_flow_tbl_expand(table);
1230                         if (!IS_ERR(new_table)) {
1231                                 rcu_assign_pointer(dp->table, new_table);
1232                                 ovs_flow_tbl_deferred_destroy(table);
1233                                 table = genl_dereference(dp->table);
1234                         }
1235                 }
1236
1237                 /* Allocate flow. */
1238                 flow = ovs_flow_alloc();
1239                 if (IS_ERR(flow)) {
1240                         error = PTR_ERR(flow);
1241                         goto err_kfree;
1242                 }
1243                 clear_stats(flow);
1244
1245                 rcu_assign_pointer(flow->sf_acts, acts);
1246
1247                 /* Put flow in bucket. */
1248                 ovs_flow_tbl_insert(table, flow, &key, key_len);
1249
1250                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1251                                                 info->snd_seq,
1252                                                 OVS_FLOW_CMD_NEW);
1253         } else {
1254                 /* We found a matching flow. */
1255                 struct sw_flow_actions *old_acts;
1256
1257                 /* Bail out if we're not allowed to modify an existing flow.
1258                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1259                  * because Generic Netlink treats the latter as a dump
1260                  * request.  We also accept NLM_F_EXCL in case that bug ever
1261                  * gets fixed.
1262                  */
1263                 error = -EEXIST;
1264                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1265                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1266                         goto err_kfree;
1267
1268                 /* Update actions. */
1269                 old_acts = rcu_dereference_protected(flow->sf_acts,
1270                                                      lockdep_genl_is_held());
1271                 rcu_assign_pointer(flow->sf_acts, acts);
1272                 ovs_flow_deferred_free_acts(old_acts);
1273
1274                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1275                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1276
1277                 /* Clear stats. */
1278                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1279                         spin_lock_bh(&flow->lock);
1280                         clear_stats(flow);
1281                         spin_unlock_bh(&flow->lock);
1282                 }
1283         }
1284
1285         if (!IS_ERR(reply))
1286                 genl_notify(reply, genl_info_net(info), info->snd_portid,
1287                            ovs_dp_flow_multicast_group.id, info->nlhdr,
1288                            GFP_KERNEL);
1289         else
1290                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1291                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1292         return 0;
1293
1294 err_kfree:
1295         kfree(acts);
1296 error:
1297         return error;
1298 }
1299
1300 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1301 {
1302         struct nlattr **a = info->attrs;
1303         struct ovs_header *ovs_header = info->userhdr;
1304         struct sw_flow_key key;
1305         struct sk_buff *reply;
1306         struct sw_flow *flow;
1307         struct datapath *dp;
1308         struct flow_table *table;
1309         int err;
1310         int key_len;
1311
1312         if (!a[OVS_FLOW_ATTR_KEY])
1313                 return -EINVAL;
1314         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1315         if (err)
1316                 return err;
1317
1318         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1319         if (!dp)
1320                 return -ENODEV;
1321
1322         table = genl_dereference(dp->table);
1323         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1324         if (!flow)
1325                 return -ENOENT;
1326
1327         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1328                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1329         if (IS_ERR(reply))
1330                 return PTR_ERR(reply);
1331
1332         return genlmsg_reply(reply, info);
1333 }
1334
1335 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1336 {
1337         struct nlattr **a = info->attrs;
1338         struct ovs_header *ovs_header = info->userhdr;
1339         struct sw_flow_key key;
1340         struct sk_buff *reply;
1341         struct sw_flow *flow;
1342         struct datapath *dp;
1343         struct flow_table *table;
1344         int err;
1345         int key_len;
1346
1347         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1348         if (!dp)
1349                 return -ENODEV;
1350
1351         if (!a[OVS_FLOW_ATTR_KEY])
1352                 return flush_flows(dp);
1353
1354         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1355         if (err)
1356                 return err;
1357
1358         table = genl_dereference(dp->table);
1359         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1360         if (!flow)
1361                 return -ENOENT;
1362
1363         reply = ovs_flow_cmd_alloc_info(flow);
1364         if (!reply)
1365                 return -ENOMEM;
1366
1367         ovs_flow_tbl_remove(table, flow);
1368
1369         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1370                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1371         BUG_ON(err < 0);
1372
1373         ovs_flow_deferred_free(flow);
1374
1375         genl_notify(reply, genl_info_net(info), info->snd_portid,
1376                     ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1377         return 0;
1378 }
1379
1380 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1381 {
1382         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1383         struct datapath *dp;
1384         struct flow_table *table;
1385
1386         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1387         if (!dp)
1388                 return -ENODEV;
1389
1390         table = genl_dereference(dp->table);
1391
1392         for (;;) {
1393                 struct sw_flow *flow;
1394                 u32 bucket, obj;
1395
1396                 bucket = cb->args[0];
1397                 obj = cb->args[1];
1398                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1399                 if (!flow)
1400                         break;
1401
1402                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1403                                            NETLINK_CB(cb->skb).portid,
1404                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1405                                            OVS_FLOW_CMD_NEW) < 0)
1406                         break;
1407
1408                 cb->args[0] = bucket;
1409                 cb->args[1] = obj;
1410         }
1411         return skb->len;
1412 }
1413
1414 static struct genl_ops dp_flow_genl_ops[] = {
1415         { .cmd = OVS_FLOW_CMD_NEW,
1416           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1417           .policy = flow_policy,
1418           .doit = ovs_flow_cmd_new_or_set
1419         },
1420         { .cmd = OVS_FLOW_CMD_DEL,
1421           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1422           .policy = flow_policy,
1423           .doit = ovs_flow_cmd_del
1424         },
1425         { .cmd = OVS_FLOW_CMD_GET,
1426           .flags = 0,               /* OK for unprivileged users. */
1427           .policy = flow_policy,
1428           .doit = ovs_flow_cmd_get,
1429           .dumpit = ovs_flow_cmd_dump
1430         },
1431         { .cmd = OVS_FLOW_CMD_SET,
1432           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1433           .policy = flow_policy,
1434           .doit = ovs_flow_cmd_new_or_set,
1435         },
1436 };
1437
1438 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1439 #ifdef HAVE_NLA_NUL_STRING
1440         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1441 #endif
1442         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1443 };
1444
1445 static struct genl_family dp_datapath_genl_family = {
1446         .id = GENL_ID_GENERATE,
1447         .hdrsize = sizeof(struct ovs_header),
1448         .name = OVS_DATAPATH_FAMILY,
1449         .version = OVS_DATAPATH_VERSION,
1450         .maxattr = OVS_DP_ATTR_MAX,
1451          SET_NETNSOK
1452 };
1453
1454 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1455         .name = OVS_DATAPATH_MCGROUP
1456 };
1457
1458 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1459                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1460 {
1461         struct ovs_header *ovs_header;
1462         struct ovs_dp_stats dp_stats;
1463         int err;
1464
1465         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1466                                    flags, cmd);
1467         if (!ovs_header)
1468                 goto error;
1469
1470         ovs_header->dp_ifindex = get_dpifindex(dp);
1471
1472         rcu_read_lock();
1473         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1474         rcu_read_unlock();
1475         if (err)
1476                 goto nla_put_failure;
1477
1478         get_dp_stats(dp, &dp_stats);
1479         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1480                 goto nla_put_failure;
1481
1482         return genlmsg_end(skb, ovs_header);
1483
1484 nla_put_failure:
1485         genlmsg_cancel(skb, ovs_header);
1486 error:
1487         return -EMSGSIZE;
1488 }
1489
1490 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1491                                              u32 seq, u8 cmd)
1492 {
1493         struct sk_buff *skb;
1494         int retval;
1495
1496         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1497         if (!skb)
1498                 return ERR_PTR(-ENOMEM);
1499
1500         retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1501         if (retval < 0) {
1502                 kfree_skb(skb);
1503                 return ERR_PTR(retval);
1504         }
1505         return skb;
1506 }
1507
1508 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1509 {
1510         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1511 }
1512
1513 /* Called with genl_mutex and optionally with RTNL lock also. */
1514 static struct datapath *lookup_datapath(struct net *net,
1515                                         struct ovs_header *ovs_header,
1516                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1517 {
1518         struct datapath *dp;
1519
1520         if (!a[OVS_DP_ATTR_NAME])
1521                 dp = get_dp(net, ovs_header->dp_ifindex);
1522         else {
1523                 struct vport *vport;
1524
1525                 rcu_read_lock();
1526                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1527                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1528                 rcu_read_unlock();
1529         }
1530         return dp ? dp : ERR_PTR(-ENODEV);
1531 }
1532
1533 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1534 {
1535         struct nlattr **a = info->attrs;
1536         struct vport_parms parms;
1537         struct sk_buff *reply;
1538         struct datapath *dp;
1539         struct vport *vport;
1540         struct ovs_net *ovs_net;
1541         int err, i;
1542
1543         err = -EINVAL;
1544         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1545                 goto err;
1546
1547         err = ovs_dp_cmd_validate(a);
1548         if (err)
1549                 goto err;
1550
1551         rtnl_lock();
1552
1553         err = -ENOMEM;
1554         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1555         if (dp == NULL)
1556                 goto err_unlock_rtnl;
1557
1558         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1559
1560         /* Allocate table. */
1561         err = -ENOMEM;
1562         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1563         if (!dp->table)
1564                 goto err_free_dp;
1565
1566         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1567         if (!dp->stats_percpu) {
1568                 err = -ENOMEM;
1569                 goto err_destroy_table;
1570         }
1571
1572         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1573                             GFP_KERNEL);
1574         if (!dp->ports) {
1575                 err = -ENOMEM;
1576                 goto err_destroy_percpu;
1577         }
1578
1579         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1580                 INIT_HLIST_HEAD(&dp->ports[i]);
1581
1582         /* Set up our datapath device. */
1583         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1584         parms.type = OVS_VPORT_TYPE_INTERNAL;
1585         parms.options = NULL;
1586         parms.dp = dp;
1587         parms.port_no = OVSP_LOCAL;
1588         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1589
1590         vport = new_vport(&parms);
1591         if (IS_ERR(vport)) {
1592                 err = PTR_ERR(vport);
1593                 if (err == -EBUSY)
1594                         err = -EEXIST;
1595
1596                 goto err_destroy_ports_array;
1597         }
1598
1599         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1600                                       info->snd_seq, OVS_DP_CMD_NEW);
1601         err = PTR_ERR(reply);
1602         if (IS_ERR(reply))
1603                 goto err_destroy_local_port;
1604
1605         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1606         list_add_tail(&dp->list_node, &ovs_net->dps);
1607
1608         rtnl_unlock();
1609
1610         genl_notify(reply, genl_info_net(info), info->snd_portid,
1611                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1612                     GFP_KERNEL);
1613         return 0;
1614
1615 err_destroy_local_port:
1616         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1617 err_destroy_ports_array:
1618         kfree(dp->ports);
1619 err_destroy_percpu:
1620         free_percpu(dp->stats_percpu);
1621 err_destroy_table:
1622         ovs_flow_tbl_destroy(genl_dereference(dp->table));
1623 err_free_dp:
1624         release_net(ovs_dp_get_net(dp));
1625         kfree(dp);
1626 err_unlock_rtnl:
1627         rtnl_unlock();
1628 err:
1629         return err;
1630 }
1631
1632 /* Called with genl_mutex. */
1633 static void __dp_destroy(struct datapath *dp)
1634 {
1635         int i;
1636
1637         rtnl_lock();
1638
1639         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1640                 struct vport *vport;
1641                 struct hlist_node *node, *n;
1642
1643                 hlist_for_each_entry_safe(vport, node, n, &dp->ports[i], dp_hash_node)
1644                         if (vport->port_no != OVSP_LOCAL)
1645                                 ovs_dp_detach_port(vport);
1646         }
1647
1648         list_del(&dp->list_node);
1649         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1650
1651         /* rtnl_unlock() will wait until all the references to devices that
1652          * are pending unregistration have been dropped.  We do it here to
1653          * ensure that any internal devices (which contain DP pointers) are
1654          * fully destroyed before freeing the datapath.
1655          */
1656         rtnl_unlock();
1657
1658         call_rcu(&dp->rcu, destroy_dp_rcu);
1659 }
1660
1661 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1662 {
1663         struct sk_buff *reply;
1664         struct datapath *dp;
1665         int err;
1666
1667         err = ovs_dp_cmd_validate(info->attrs);
1668         if (err)
1669                 return err;
1670
1671         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1672         err = PTR_ERR(dp);
1673         if (IS_ERR(dp))
1674                 return err;
1675
1676         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1677                                       info->snd_seq, OVS_DP_CMD_DEL);
1678         err = PTR_ERR(reply);
1679         if (IS_ERR(reply))
1680                 return err;
1681
1682         __dp_destroy(dp);
1683
1684         genl_notify(reply, genl_info_net(info), info->snd_portid,
1685                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1686                     GFP_KERNEL);
1687
1688         return 0;
1689 }
1690
1691 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1692 {
1693         struct sk_buff *reply;
1694         struct datapath *dp;
1695         int err;
1696
1697         err = ovs_dp_cmd_validate(info->attrs);
1698         if (err)
1699                 return err;
1700
1701         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1702         if (IS_ERR(dp))
1703                 return PTR_ERR(dp);
1704
1705         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1706                                       info->snd_seq, OVS_DP_CMD_NEW);
1707         if (IS_ERR(reply)) {
1708                 err = PTR_ERR(reply);
1709                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1710                                 ovs_dp_datapath_multicast_group.id, err);
1711                 return 0;
1712         }
1713
1714         genl_notify(reply, genl_info_net(info), info->snd_portid,
1715                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1716                     GFP_KERNEL);
1717
1718         return 0;
1719 }
1720
1721 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1722 {
1723         struct sk_buff *reply;
1724         struct datapath *dp;
1725         int err;
1726
1727         err = ovs_dp_cmd_validate(info->attrs);
1728         if (err)
1729                 return err;
1730
1731         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1732         if (IS_ERR(dp))
1733                 return PTR_ERR(dp);
1734
1735         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1736                                       info->snd_seq, OVS_DP_CMD_NEW);
1737         if (IS_ERR(reply))
1738                 return PTR_ERR(reply);
1739
1740         return genlmsg_reply(reply, info);
1741 }
1742
1743 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1744 {
1745         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1746         struct datapath *dp;
1747         int skip = cb->args[0];
1748         int i = 0;
1749
1750         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1751                 if (i >= skip &&
1752                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1753                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1754                                          OVS_DP_CMD_NEW) < 0)
1755                         break;
1756                 i++;
1757         }
1758
1759         cb->args[0] = i;
1760
1761         return skb->len;
1762 }
1763
1764 static struct genl_ops dp_datapath_genl_ops[] = {
1765         { .cmd = OVS_DP_CMD_NEW,
1766           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1767           .policy = datapath_policy,
1768           .doit = ovs_dp_cmd_new
1769         },
1770         { .cmd = OVS_DP_CMD_DEL,
1771           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1772           .policy = datapath_policy,
1773           .doit = ovs_dp_cmd_del
1774         },
1775         { .cmd = OVS_DP_CMD_GET,
1776           .flags = 0,               /* OK for unprivileged users. */
1777           .policy = datapath_policy,
1778           .doit = ovs_dp_cmd_get,
1779           .dumpit = ovs_dp_cmd_dump
1780         },
1781         { .cmd = OVS_DP_CMD_SET,
1782           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1783           .policy = datapath_policy,
1784           .doit = ovs_dp_cmd_set,
1785         },
1786 };
1787
1788 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1789 #ifdef HAVE_NLA_NUL_STRING
1790         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1791         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1792         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1793 #else
1794         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1795         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1796 #endif
1797         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1798         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1799         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1800         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1801 };
1802
1803 static struct genl_family dp_vport_genl_family = {
1804         .id = GENL_ID_GENERATE,
1805         .hdrsize = sizeof(struct ovs_header),
1806         .name = OVS_VPORT_FAMILY,
1807         .version = OVS_VPORT_VERSION,
1808         .maxattr = OVS_VPORT_ATTR_MAX,
1809          SET_NETNSOK
1810 };
1811
1812 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1813         .name = OVS_VPORT_MCGROUP
1814 };
1815
1816 /* Called with RTNL lock or RCU read lock. */
1817 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1818                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1819 {
1820         struct ovs_header *ovs_header;
1821         struct ovs_vport_stats vport_stats;
1822         int err;
1823
1824         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1825                                  flags, cmd);
1826         if (!ovs_header)
1827                 return -EMSGSIZE;
1828
1829         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1830
1831         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1832             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1833             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1834             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1835                 goto nla_put_failure;
1836
1837         ovs_vport_get_stats(vport, &vport_stats);
1838         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1839                     &vport_stats))
1840                 goto nla_put_failure;
1841
1842         if (nla_put(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN,
1843                     vport->ops->get_addr(vport)))
1844                 goto nla_put_failure;
1845
1846         err = ovs_vport_get_options(vport, skb);
1847         if (err == -EMSGSIZE)
1848                 goto error;
1849
1850         return genlmsg_end(skb, ovs_header);
1851
1852 nla_put_failure:
1853         err = -EMSGSIZE;
1854 error:
1855         genlmsg_cancel(skb, ovs_header);
1856         return err;
1857 }
1858
1859 /* Called with RTNL lock or RCU read lock. */
1860 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1861                                          u32 seq, u8 cmd)
1862 {
1863         struct sk_buff *skb;
1864         int retval;
1865
1866         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1867         if (!skb)
1868                 return ERR_PTR(-ENOMEM);
1869
1870         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1871         if (retval < 0) {
1872                 kfree_skb(skb);
1873                 return ERR_PTR(retval);
1874         }
1875         return skb;
1876 }
1877
1878 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1879 {
1880         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1881 }
1882
1883 /* Called with RTNL lock or RCU read lock. */
1884 static struct vport *lookup_vport(struct net *net,
1885                                   struct ovs_header *ovs_header,
1886                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1887 {
1888         struct datapath *dp;
1889         struct vport *vport;
1890
1891         if (a[OVS_VPORT_ATTR_NAME]) {
1892                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1893                 if (!vport)
1894                         return ERR_PTR(-ENODEV);
1895                 if (ovs_header->dp_ifindex &&
1896                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1897                         return ERR_PTR(-ENODEV);
1898                 return vport;
1899         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1900                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1901
1902                 if (port_no >= DP_MAX_PORTS)
1903                         return ERR_PTR(-EFBIG);
1904
1905                 dp = get_dp(net, ovs_header->dp_ifindex);
1906                 if (!dp)
1907                         return ERR_PTR(-ENODEV);
1908
1909                 vport = ovs_vport_rtnl_rcu(dp, port_no);
1910                 if (!vport)
1911                         return ERR_PTR(-ENODEV);
1912                 return vport;
1913         } else
1914                 return ERR_PTR(-EINVAL);
1915 }
1916
1917 /* Called with RTNL lock. */
1918 static int change_vport(struct vport *vport,
1919                         struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1920 {
1921         int err = 0;
1922
1923         if (a[OVS_VPORT_ATTR_STATS])
1924                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1925
1926         if (a[OVS_VPORT_ATTR_ADDRESS])
1927                 err = ovs_vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1928
1929         return err;
1930 }
1931
1932 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1933 {
1934         struct nlattr **a = info->attrs;
1935         struct ovs_header *ovs_header = info->userhdr;
1936         struct vport_parms parms;
1937         struct sk_buff *reply;
1938         struct vport *vport;
1939         struct datapath *dp;
1940         u32 port_no;
1941         int err;
1942
1943         err = -EINVAL;
1944         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1945             !a[OVS_VPORT_ATTR_UPCALL_PID])
1946                 goto exit;
1947
1948         err = ovs_vport_cmd_validate(a);
1949         if (err)
1950                 goto exit;
1951
1952         rtnl_lock();
1953         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1954         err = -ENODEV;
1955         if (!dp)
1956                 goto exit_unlock;
1957
1958         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1959                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1960
1961                 err = -EFBIG;
1962                 if (port_no >= DP_MAX_PORTS)
1963                         goto exit_unlock;
1964
1965                 vport = ovs_vport_rtnl(dp, port_no);
1966                 err = -EBUSY;
1967                 if (vport)
1968                         goto exit_unlock;
1969         } else {
1970                 for (port_no = 1; ; port_no++) {
1971                         if (port_no >= DP_MAX_PORTS) {
1972                                 err = -EFBIG;
1973                                 goto exit_unlock;
1974                         }
1975                         vport = ovs_vport_rtnl(dp, port_no);
1976                         if (!vport)
1977                                 break;
1978                 }
1979         }
1980
1981         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1982         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1983         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1984         parms.dp = dp;
1985         parms.port_no = port_no;
1986         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1987
1988         vport = new_vport(&parms);
1989         err = PTR_ERR(vport);
1990         if (IS_ERR(vport))
1991                 goto exit_unlock;
1992
1993         err = change_vport(vport, a);
1994         if (!err) {
1995                 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1996                                                  info->snd_seq,
1997                                                  OVS_VPORT_CMD_NEW);
1998                 if (IS_ERR(reply))
1999                         err = PTR_ERR(reply);
2000         }
2001         if (err) {
2002                 ovs_dp_detach_port(vport);
2003                 goto exit_unlock;
2004         }
2005         genl_notify(reply, genl_info_net(info), info->snd_portid,
2006                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2007
2008 exit_unlock:
2009         rtnl_unlock();
2010 exit:
2011         return err;
2012 }
2013
2014 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2015 {
2016         struct nlattr **a = info->attrs;
2017         struct sk_buff *reply;
2018         struct vport *vport;
2019         int err;
2020
2021         err = ovs_vport_cmd_validate(a);
2022         if (err)
2023                 goto exit;
2024
2025         rtnl_lock();
2026         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2027         err = PTR_ERR(vport);
2028         if (IS_ERR(vport))
2029                 goto exit_unlock;
2030
2031         err = 0;
2032         if (a[OVS_VPORT_ATTR_TYPE] &&
2033             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
2034                 err = -EINVAL;
2035
2036         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
2037                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2038         if (!err)
2039                 err = change_vport(vport, a);
2040         else
2041                 goto exit_unlock;
2042         if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
2043                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2044
2045         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2046                                          info->snd_seq, OVS_VPORT_CMD_NEW);
2047         if (IS_ERR(reply)) {
2048                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
2049                                 ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
2050                 goto exit_unlock;
2051         }
2052
2053         genl_notify(reply, genl_info_net(info), info->snd_portid,
2054                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2055
2056 exit_unlock:
2057         rtnl_unlock();
2058 exit:
2059         return err;
2060 }
2061
2062 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2063 {
2064         struct nlattr **a = info->attrs;
2065         struct sk_buff *reply;
2066         struct vport *vport;
2067         int err;
2068
2069         err = ovs_vport_cmd_validate(a);
2070         if (err)
2071                 goto exit;
2072
2073         rtnl_lock();
2074         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2075         err = PTR_ERR(vport);
2076         if (IS_ERR(vport))
2077                 goto exit_unlock;
2078
2079         if (vport->port_no == OVSP_LOCAL) {
2080                 err = -EINVAL;
2081                 goto exit_unlock;
2082         }
2083
2084         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2085                                          info->snd_seq, OVS_VPORT_CMD_DEL);
2086         err = PTR_ERR(reply);
2087         if (IS_ERR(reply))
2088                 goto exit_unlock;
2089
2090         ovs_dp_detach_port(vport);
2091
2092         genl_notify(reply, genl_info_net(info), info->snd_portid,
2093                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2094
2095 exit_unlock:
2096         rtnl_unlock();
2097 exit:
2098         return err;
2099 }
2100
2101 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2102 {
2103         struct nlattr **a = info->attrs;
2104         struct ovs_header *ovs_header = info->userhdr;
2105         struct sk_buff *reply;
2106         struct vport *vport;
2107         int err;
2108
2109         err = ovs_vport_cmd_validate(a);
2110         if (err)
2111                 goto exit;
2112
2113         rcu_read_lock();
2114         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2115         err = PTR_ERR(vport);
2116         if (IS_ERR(vport))
2117                 goto exit_unlock;
2118
2119         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2120                                          info->snd_seq, OVS_VPORT_CMD_NEW);
2121         err = PTR_ERR(reply);
2122         if (IS_ERR(reply))
2123                 goto exit_unlock;
2124
2125         rcu_read_unlock();
2126
2127         return genlmsg_reply(reply, info);
2128
2129 exit_unlock:
2130         rcu_read_unlock();
2131 exit:
2132         return err;
2133 }
2134
2135 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2136 {
2137         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2138         struct datapath *dp;
2139         int bucket = cb->args[0], skip = cb->args[1];
2140         int i, j = 0;
2141
2142         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2143         if (!dp)
2144                 return -ENODEV;
2145
2146         rcu_read_lock();
2147         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2148                 struct vport *vport;
2149                 struct hlist_node *n;
2150
2151                 j = 0;
2152                 hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node) {
2153                         if (j >= skip &&
2154                             ovs_vport_cmd_fill_info(vport, skb,
2155                                                     NETLINK_CB(cb->skb).portid,
2156                                                     cb->nlh->nlmsg_seq,
2157                                                     NLM_F_MULTI,
2158                                                     OVS_VPORT_CMD_NEW) < 0)
2159                                 goto out;
2160
2161                         j++;
2162                 }
2163                 skip = 0;
2164         }
2165 out:
2166         rcu_read_unlock();
2167
2168         cb->args[0] = i;
2169         cb->args[1] = j;
2170
2171         return skb->len;
2172 }
2173
2174 static struct genl_ops dp_vport_genl_ops[] = {
2175         { .cmd = OVS_VPORT_CMD_NEW,
2176           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2177           .policy = vport_policy,
2178           .doit = ovs_vport_cmd_new
2179         },
2180         { .cmd = OVS_VPORT_CMD_DEL,
2181           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2182           .policy = vport_policy,
2183           .doit = ovs_vport_cmd_del
2184         },
2185         { .cmd = OVS_VPORT_CMD_GET,
2186           .flags = 0,               /* OK for unprivileged users. */
2187           .policy = vport_policy,
2188           .doit = ovs_vport_cmd_get,
2189           .dumpit = ovs_vport_cmd_dump
2190         },
2191         { .cmd = OVS_VPORT_CMD_SET,
2192           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2193           .policy = vport_policy,
2194           .doit = ovs_vport_cmd_set,
2195         },
2196 };
2197
2198 struct genl_family_and_ops {
2199         struct genl_family *family;
2200         struct genl_ops *ops;
2201         int n_ops;
2202         struct genl_multicast_group *group;
2203 };
2204
2205 static const struct genl_family_and_ops dp_genl_families[] = {
2206         { &dp_datapath_genl_family,
2207           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2208           &ovs_dp_datapath_multicast_group },
2209         { &dp_vport_genl_family,
2210           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2211           &ovs_dp_vport_multicast_group },
2212         { &dp_flow_genl_family,
2213           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2214           &ovs_dp_flow_multicast_group },
2215         { &dp_packet_genl_family,
2216           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2217           NULL },
2218 };
2219
2220 static void dp_unregister_genl(int n_families)
2221 {
2222         int i;
2223
2224         for (i = 0; i < n_families; i++)
2225                 genl_unregister_family(dp_genl_families[i].family);
2226 }
2227
2228 static int dp_register_genl(void)
2229 {
2230         int n_registered;
2231         int err;
2232         int i;
2233
2234         n_registered = 0;
2235         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2236                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2237
2238                 err = genl_register_family_with_ops(f->family, f->ops,
2239                                                     f->n_ops);
2240                 if (err)
2241                         goto error;
2242                 n_registered++;
2243
2244                 if (f->group) {
2245                         err = genl_register_mc_group(f->family, f->group);
2246                         if (err)
2247                                 goto error;
2248                 }
2249         }
2250
2251         return 0;
2252
2253 error:
2254         dp_unregister_genl(n_registered);
2255         return err;
2256 }
2257
2258 static int __rehash_flow_table(void *dummy)
2259 {
2260         struct datapath *dp;
2261         struct net *net;
2262
2263         rtnl_lock();
2264         for_each_net(net) {
2265                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2266
2267                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2268                         struct flow_table *old_table = genl_dereference(dp->table);
2269                         struct flow_table *new_table;
2270
2271                         new_table = ovs_flow_tbl_rehash(old_table);
2272                         if (!IS_ERR(new_table)) {
2273                                 rcu_assign_pointer(dp->table, new_table);
2274                                 ovs_flow_tbl_deferred_destroy(old_table);
2275                         }
2276                 }
2277         }
2278         rtnl_unlock();
2279         return 0;
2280 }
2281
2282 static void rehash_flow_table(struct work_struct *work)
2283 {
2284         genl_exec(__rehash_flow_table, NULL);
2285         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2286 }
2287
2288 static int dp_destroy_all(void *data)
2289 {
2290         struct datapath *dp, *dp_next;
2291         struct ovs_net *ovs_net = data;
2292
2293         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2294                 __dp_destroy(dp);
2295
2296         return 0;
2297 }
2298
2299 static int __net_init ovs_init_net(struct net *net)
2300 {
2301         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2302
2303         INIT_LIST_HEAD(&ovs_net->dps);
2304         return 0;
2305 }
2306
2307 static void __net_exit ovs_exit_net(struct net *net)
2308 {
2309         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2310
2311         genl_exec(dp_destroy_all, ovs_net);
2312 }
2313
2314 static struct pernet_operations ovs_net_ops = {
2315         .init = ovs_init_net,
2316         .exit = ovs_exit_net,
2317         .id   = &ovs_net_id,
2318         .size = sizeof(struct ovs_net),
2319 };
2320
2321 static int __init dp_init(void)
2322 {
2323         int err;
2324
2325         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2326
2327         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2328                 VERSION);
2329
2330         err = genl_exec_init();
2331         if (err)
2332                 goto error;
2333
2334         err = ovs_workqueues_init();
2335         if (err)
2336                 goto error_genl_exec;
2337
2338         err = ovs_tnl_init();
2339         if (err)
2340                 goto error_wq;
2341
2342         err = ovs_flow_init();
2343         if (err)
2344                 goto error_tnl_exit;
2345
2346         err = ovs_vport_init();
2347         if (err)
2348                 goto error_flow_exit;
2349
2350         err = register_pernet_device(&ovs_net_ops);
2351         if (err)
2352                 goto error_vport_exit;
2353
2354         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2355         if (err)
2356                 goto error_netns_exit;
2357
2358         err = dp_register_genl();
2359         if (err < 0)
2360                 goto error_unreg_notifier;
2361
2362         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2363
2364         return 0;
2365
2366 error_unreg_notifier:
2367         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2368 error_netns_exit:
2369         unregister_pernet_device(&ovs_net_ops);
2370 error_vport_exit:
2371         ovs_vport_exit();
2372 error_flow_exit:
2373         ovs_flow_exit();
2374 error_tnl_exit:
2375         ovs_tnl_exit();
2376 error_wq:
2377         ovs_workqueues_exit();
2378 error_genl_exec:
2379         genl_exec_exit();
2380 error:
2381         return err;
2382 }
2383
2384 static void dp_cleanup(void)
2385 {
2386         cancel_delayed_work_sync(&rehash_flow_wq);
2387         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2388         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2389         unregister_pernet_device(&ovs_net_ops);
2390         rcu_barrier();
2391         ovs_vport_exit();
2392         ovs_flow_exit();
2393         ovs_tnl_exit();
2394         ovs_workqueues_exit();
2395         genl_exec_exit();
2396 }
2397
2398 module_init(dp_init);
2399 module_exit(dp_cleanup);
2400
2401 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2402 MODULE_LICENSE("GPL");
2403 MODULE_VERSION(VERSION);