datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
   3  * Distributed under the terms of the GNU GPL version 2.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 /* Functions for managing the dp interface/device. */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/init.h>
  14 #include <linux/module.h>
  15 #include <linux/if_arp.h>
  16 #include <linux/if_vlan.h>
  17 #include <linux/in.h>
  18 #include <linux/ip.h>
  19 #include <linux/jhash.h>
  20 #include <linux/delay.h>
  21 #include <linux/time.h>
  22 #include <linux/etherdevice.h>
  23 #include <linux/genetlink.h>
  24 #include <linux/kernel.h>
  25 #include <linux/kthread.h>
  26 #include <linux/mutex.h>
  27 #include <linux/percpu.h>
  28 #include <linux/rcupdate.h>
  29 #include <linux/tcp.h>
  30 #include <linux/udp.h>
  31 #include <linux/version.h>
  32 #include <linux/ethtool.h>
  33 #include <linux/wait.h>
  34 #include <asm/system.h>
  35 #include <asm/div64.h>
  36 #include <asm/bug.h>
  37 #include <linux/highmem.h>
  38 #include <linux/netfilter_bridge.h>
  39 #include <linux/netfilter_ipv4.h>
  40 #include <linux/inetdevice.h>
  41 #include <linux/list.h>
  42 #include <linux/openvswitch.h>
  43 #include <linux/rculist.h>
  44 #include <linux/dmi.h>
  45 #include <net/inet_ecn.h>
  46 #include <net/genetlink.h>
  47
  48 #include "checksum.h"
  49 #include "datapath.h"
  50 #include "actions.h"
  51 #include "flow.h"
  52 #include "vlan.h"
  53 #include "tunnel.h"
  54 #include "vport-internal_dev.h"
  55
  56 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
  57     LINUX_VERSION_CODE >= KERNEL_VERSION(3,2,0)
  58 #error Kernels before 2.6.18 or after 3.1 are not supported by this version of Open vSwitch.
  59 #endif
  60
  61 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
  62 EXPORT_SYMBOL(dp_ioctl_hook);
  63
  64 /**
  65  * DOC: Locking:
  66  *
  67  * Writes to device state (add/remove datapath, port, set operations on vports,
  68  * etc.) are protected by RTNL.
  69  *
  70  * Writes to other state (flow table modifications, set miscellaneous datapath
  71  * parameters, etc.) are protected by genl_mutex.  The RTNL lock nests inside
  72  * genl_mutex.
  73  *
  74  * Reads are protected by RCU.
  75  *
  76  * There are a few special cases (mostly stats) that have their own
  77  * synchronization but they nest under all of above and don't interact with
  78  * each other.
  79  */
  80
  81 /* Global list of datapaths to enable dumping them all out.
  82  * Protected by genl_mutex.
  83  */
  84 static LIST_HEAD(dps);
  85
  86 static struct vport *new_vport(const struct vport_parms *);
  87 static int queue_gso_packets(int dp_ifindex, struct sk_buff *,
  88                              const struct dp_upcall_info *);
  89 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *,
  90                                   const struct dp_upcall_info *);
  91
  92 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  93 struct datapath *get_dp(int dp_ifindex)
  94 {
  95         struct datapath *dp = NULL;
  96         struct net_device *dev;
  97
  98         rcu_read_lock();
  99         dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
 100         if (dev) {
 101                 struct vport *vport = internal_dev_get_vport(dev);
 102                 if (vport)
 103                         dp = vport->dp;
 104         }
 105         rcu_read_unlock();
 106
 107         return dp;
 108 }
 109 EXPORT_SYMBOL_GPL(get_dp);
 110
 111 /* Must be called with genl_mutex. */
 112 static struct flow_table *get_table_protected(struct datapath *dp)
 113 {
 114         return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
 115 }
 116
 117 /* Must be called with rcu_read_lock or RTNL lock. */
 118 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
 119 {
 120         return rcu_dereference_rtnl(dp->ports[port_no]);
 121 }
 122
 123 /* Must be called with rcu_read_lock or RTNL lock. */
 124 const char *dp_name(const struct datapath *dp)
 125 {
 126         return vport_get_name(rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]));
 127 }
 128
 129 static int get_dpifindex(struct datapath *dp)
 130 {
 131         struct vport *local;
 132         int ifindex;
 133
 134         rcu_read_lock();
 135
 136         local = get_vport_protected(dp, OVSP_LOCAL);
 137         if (local)
 138                 ifindex = vport_get_ifindex(local);
 139         else
 140                 ifindex = 0;
 141
 142         rcu_read_unlock();
 143
 144         return ifindex;
 145 }
 146
 147 static inline size_t br_nlmsg_size(void)
 148 {
 149         return NLMSG_ALIGN(sizeof(struct ifinfomsg))
 150                + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
 151                + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
 152                + nla_total_size(4) /* IFLA_MASTER */
 153                + nla_total_size(4) /* IFLA_MTU */
 154                + nla_total_size(1); /* IFLA_OPERSTATE */
 155 }
 156
 157 /* Caller must hold RTNL lock. */
 158 static int dp_fill_ifinfo(struct sk_buff *skb,
 159                           const struct vport *port,
 160                           int event, unsigned int flags)
 161 {
 162         struct datapath *dp = port->dp;
 163         int ifindex = vport_get_ifindex(port);
 164         struct ifinfomsg *hdr;
 165         struct nlmsghdr *nlh;
 166
 167         if (ifindex < 0)
 168                 return ifindex;
 169
 170         nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
 171         if (nlh == NULL)
 172                 return -EMSGSIZE;
 173
 174         hdr = nlmsg_data(nlh);
 175         hdr->ifi_family = AF_BRIDGE;
 176         hdr->__ifi_pad = 0;
 177         hdr->ifi_type = ARPHRD_ETHER;
 178         hdr->ifi_index = ifindex;
 179         hdr->ifi_flags = vport_get_flags(port);
 180         hdr->ifi_change = 0;
 181
 182         NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
 183         NLA_PUT_U32(skb, IFLA_MASTER, get_dpifindex(dp));
 184         NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
 185 #ifdef IFLA_OPERSTATE
 186         NLA_PUT_U8(skb, IFLA_OPERSTATE,
 187                    vport_is_running(port)
 188                         ? vport_get_operstate(port)
 189                         : IF_OPER_DOWN);
 190 #endif
 191
 192         NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
 193
 194         return nlmsg_end(skb, nlh);
 195
 196 nla_put_failure:
 197         nlmsg_cancel(skb, nlh);
 198         return -EMSGSIZE;
 199 }
 200
 201 /* Caller must hold RTNL lock. */
 202 static void dp_ifinfo_notify(int event, struct vport *port)
 203 {
 204         struct sk_buff *skb;
 205         int err = -ENOBUFS;
 206
 207         skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
 208         if (skb == NULL)
 209                 goto errout;
 210
 211         err = dp_fill_ifinfo(skb, port, event, 0);
 212         if (err < 0) {
 213                 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
 214                 WARN_ON(err == -EMSGSIZE);
 215                 kfree_skb(skb);
 216                 goto errout;
 217         }
 218         rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 219         return;
 220 errout:
 221         if (err < 0)
 222                 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
 223 }
 224
 225 static void release_dp(struct kobject *kobj)
 226 {
 227         struct datapath *dp = container_of(kobj, struct datapath, ifobj);
 228         kfree(dp);
 229 }
 230
 231 static struct kobj_type dp_ktype = {
 232         .release = release_dp
 233 };
 234
 235 static void destroy_dp_rcu(struct rcu_head *rcu)
 236 {
 237         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 238
 239         flow_tbl_destroy(dp->table);
 240         free_percpu(dp->stats_percpu);
 241         kobject_put(&dp->ifobj);
 242 }
 243
 244 /* Called with RTNL lock and genl_lock. */
 245 static struct vport *new_vport(const struct vport_parms *parms)
 246 {
 247         struct vport *vport;
 248
 249         vport = vport_add(parms);
 250         if (!IS_ERR(vport)) {
 251                 struct datapath *dp = parms->dp;
 252
 253                 rcu_assign_pointer(dp->ports[parms->port_no], vport);
 254                 list_add(&vport->node, &dp->port_list);
 255
 256                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 257         }
 258
 259         return vport;
 260 }
 261
 262 /* Called with RTNL lock. */
 263 void dp_detach_port(struct vport *p)
 264 {
 265         ASSERT_RTNL();
 266
 267         if (p->port_no != OVSP_LOCAL)
 268                 dp_sysfs_del_if(p);
 269         dp_ifinfo_notify(RTM_DELLINK, p);
 270
 271         /* First drop references to device. */
 272         list_del(&p->node);
 273         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 274
 275         /* Then destroy it. */
 276         vport_del(p);
 277 }
 278
 279 /* Must be called with rcu_read_lock. */
 280 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 281 {
 282         struct datapath *dp = p->dp;
 283         struct sw_flow *flow;
 284         struct dp_stats_percpu *stats;
 285         u64 *stats_counter;
 286         int error;
 287
 288         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 289         OVS_CB(skb)->vport = p;
 290
 291         if (!OVS_CB(skb)->flow) {
 292                 struct sw_flow_key key;
 293                 int key_len;
 294
 295                 /* Extract flow from 'skb' into 'key'. */
 296                 error = flow_extract(skb, p->port_no, &key, &key_len);
 297                 if (unlikely(error)) {
 298                         kfree_skb(skb);
 299                         return;
 300                 }
 301
 302                 /* Look up flow. */
 303                 flow = flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
 304                 if (unlikely(!flow)) {
 305                         struct dp_upcall_info upcall;
 306
 307                         upcall.cmd = OVS_PACKET_CMD_MISS;
 308                         upcall.key = &key;
 309                         upcall.userdata = NULL;
 310                         upcall.pid = p->upcall_pid;
 311                         dp_upcall(dp, skb, &upcall);
 312                         consume_skb(skb);
 313                         stats_counter = &stats->n_missed;
 314                         goto out;
 315                 }
 316
 317                 OVS_CB(skb)->flow = flow;
 318         }
 319
 320         stats_counter = &stats->n_hit;
 321         flow_used(OVS_CB(skb)->flow, skb);
 322         execute_actions(dp, skb);
 323
 324 out:
 325         /* Update datapath statistics. */
 326
 327         write_seqcount_begin(&stats->seqlock);
 328         (*stats_counter)++;
 329         write_seqcount_end(&stats->seqlock);
 330 }
 331
 332 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
 333 {
 334         u16 csum_start, csum_offset;
 335         __wsum csum;
 336
 337         get_skb_csum_pointers(skb, &csum_start, &csum_offset);
 338         csum_start -= skb_headroom(skb);
 339
 340         skb_copy_bits(skb, 0, to, csum_start);
 341
 342         csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
 343                                       skb->len - csum_start, 0);
 344         *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
 345 }
 346
 347 static struct genl_family dp_packet_genl_family = {
 348         .id = GENL_ID_GENERATE,
 349         .hdrsize = sizeof(struct ovs_header),
 350         .name = OVS_PACKET_FAMILY,
 351         .version = OVS_PACKET_VERSION,
 352         .maxattr = OVS_PACKET_ATTR_MAX
 353 };
 354
 355 int dp_upcall(struct datapath *dp, struct sk_buff *skb,
 356               const struct dp_upcall_info *upcall_info)
 357 {
 358         struct dp_stats_percpu *stats;
 359         int dp_ifindex;
 360         int err;
 361
 362         if (upcall_info->pid == 0) {
 363                 err = -ENOTCONN;
 364                 goto err;
 365         }
 366
 367         dp_ifindex = get_dpifindex(dp);
 368         if (!dp_ifindex) {
 369                 err = -ENODEV;
 370                 goto err;
 371         }
 372
 373         forward_ip_summed(skb, true);
 374
 375         if (!skb_is_gso(skb))
 376                 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
 377         else
 378                 err = queue_gso_packets(dp_ifindex, skb, upcall_info);
 379         if (err)
 380                 goto err;
 381
 382         return 0;
 383
 384 err:
 385         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 386
 387         write_seqcount_begin(&stats->seqlock);
 388         stats->n_lost++;
 389         write_seqcount_end(&stats->seqlock);
 390
 391         return err;
 392 }
 393
 394 static int queue_gso_packets(int dp_ifindex, struct sk_buff *skb,
 395                              const struct dp_upcall_info *upcall_info)
 396 {
 397         struct dp_upcall_info later_info;
 398         struct sw_flow_key later_key;
 399         struct sk_buff *segs, *nskb;
 400         int err;
 401
 402         segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 403         if (IS_ERR(skb))
 404                 return PTR_ERR(skb);
 405
 406         /* Queue all of the segments. */
 407         skb = segs;
 408         do {
 409                 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
 410                 if (err)
 411                         break;
 412
 413                 if (skb == segs && skb_shinfo(skb)->gso_type & SKB_GSO_UDP) {
 414                         /* The initial flow key extracted by flow_extract() in
 415                          * this case is for a first fragment, so we need to
 416                          * properly mark later fragments.
 417                          */
 418                         later_key = *upcall_info->key;
 419                         later_key.ip.tos_frag &= ~OVS_FRAG_TYPE_MASK;
 420                         later_key.ip.tos_frag |= OVS_FRAG_TYPE_LATER;
 421
 422                         later_info = *upcall_info;
 423                         later_info.key = &later_key;
 424                         upcall_info = &later_info;
 425                 }
 426         } while ((skb = skb->next));
 427
 428         /* Free all of the segments. */
 429         skb = segs;
 430         do {
 431                 nskb = skb->next;
 432                 if (err)
 433                         kfree_skb(skb);
 434                 else
 435                         consume_skb(skb);
 436         } while ((skb = nskb));
 437         return err;
 438 }
 439
 440 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *skb,
 441                                   const struct dp_upcall_info *upcall_info)
 442 {
 443         struct ovs_header *upcall;
 444         struct sk_buff *user_skb; /* to be queued to userspace */
 445         struct nlattr *nla;
 446         unsigned int len;
 447         int err;
 448
 449         err = vlan_deaccel_tag(skb);
 450         if (unlikely(err))
 451                 return err;
 452
 453         if (nla_attr_size(skb->len) > USHRT_MAX)
 454                 return -EFBIG;
 455
 456         len = sizeof(struct ovs_header);
 457         len += nla_total_size(skb->len);
 458         len += nla_total_size(FLOW_BUFSIZE);
 459         if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
 460                 len += nla_total_size(8);
 461
 462         user_skb = genlmsg_new(len, GFP_ATOMIC);
 463         if (!user_skb)
 464                 return -ENOMEM;
 465
 466         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 467                              0, upcall_info->cmd);
 468         upcall->dp_ifindex = dp_ifindex;
 469
 470         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 471         flow_to_nlattrs(upcall_info->key, user_skb);
 472         nla_nest_end(user_skb, nla);
 473
 474         if (upcall_info->userdata)
 475                 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
 476                             nla_get_u64(upcall_info->userdata));
 477
 478         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 479         if (skb->ip_summed == CHECKSUM_PARTIAL)
 480                 copy_and_csum_skb(skb, nla_data(nla));
 481         else
 482                 skb_copy_bits(skb, 0, nla_data(nla), skb->len);
 483
 484         return genlmsg_unicast(&init_net, user_skb, upcall_info->pid);
 485 }
 486
 487 /* Called with genl_mutex. */
 488 static int flush_flows(int dp_ifindex)
 489 {
 490         struct flow_table *old_table;
 491         struct flow_table *new_table;
 492         struct datapath *dp;
 493
 494         dp = get_dp(dp_ifindex);
 495         if (!dp)
 496                 return -ENODEV;
 497
 498         old_table = get_table_protected(dp);
 499         new_table = flow_tbl_alloc(TBL_MIN_BUCKETS);
 500         if (!new_table)
 501                 return -ENOMEM;
 502
 503         rcu_assign_pointer(dp->table, new_table);
 504
 505         flow_tbl_deferred_destroy(old_table);
 506         return 0;
 507 }
 508
 509 static int validate_actions(const struct nlattr *attr,
 510                                 const struct sw_flow_key *key, int depth);
 511
 512 static int validate_sample(const struct nlattr *attr,
 513                                 const struct sw_flow_key *key, int depth)
 514 {
 515         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 516         const struct nlattr *probability, *actions;
 517         const struct nlattr *a;
 518         int rem;
 519
 520         memset(attrs, 0, sizeof(attrs));
 521         nla_for_each_nested (a, attr, rem) {
 522                 int type = nla_type(a);
 523                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 524                         return -EINVAL;
 525                 attrs[type] = a;
 526         }
 527         if (rem)
 528                 return -EINVAL;
 529
 530         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 531         if (!probability || nla_len(probability) != sizeof(u32))
 532                 return -EINVAL;
 533
 534         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 535         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 536                 return -EINVAL;
 537         return validate_actions(actions, key, depth + 1);
 538 }
 539
 540 static int validate_action_key(const struct nlattr *a,
 541                                 const struct sw_flow_key *flow_key)
 542 {
 543         int act_type = nla_type(a);
 544         const struct nlattr *ovs_key = nla_data(a);
 545         int key_type = nla_type(ovs_key);
 546
 547         /* There can be only one key in a action */
 548         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 549                 return -EINVAL;
 550
 551         if (key_type > OVS_KEY_ATTR_MAX ||
 552             nla_len(ovs_key) != ovs_key_lens[key_type])
 553                 return -EINVAL;
 554
 555 #define ACTION(act, key)        ((act << 8) | key)
 556
 557         switch(ACTION(act_type, key_type)) {
 558         const struct ovs_key_ipv4 *ipv4_key;
 559         const struct ovs_key_8021q *q_key;
 560
 561         case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_TUN_ID):
 562         case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_ETHERNET):
 563                 break;
 564
 565         case ACTION(OVS_ACTION_ATTR_PUSH, OVS_KEY_ATTR_8021Q):
 566                 q_key = nla_data(ovs_key);
 567                 if (q_key->q_tpid != htons(ETH_P_8021Q))
 568                         return -EINVAL;
 569
 570                 if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
 571                         return -EINVAL;
 572                 break;
 573
 574         case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_IPV4):
 575                 if (flow_key->eth.type != htons(ETH_P_IP))
 576                         return -EINVAL;
 577
 578                 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
 579                         return -EINVAL;
 580
 581                 ipv4_key = nla_data(ovs_key);
 582                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 583                         return -EINVAL;
 584
 585                 if (ipv4_key->ipv4_tos & INET_ECN_MASK)
 586                         return -EINVAL;
 587
 588                 if (ipv4_key->ipv4_frag !=
 589                     (flow_key->ip.tos_frag & OVS_FRAG_TYPE_MASK))
 590                         return -EINVAL;
 591
 592                 break;
 593
 594         case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_TCP):
 595                 if (flow_key->ip.proto != IPPROTO_TCP)
 596                         return -EINVAL;
 597
 598                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 599                         return -EINVAL;
 600
 601                 break;
 602
 603         case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_UDP):
 604                 if (flow_key->ip.proto != IPPROTO_UDP)
 605                         return -EINVAL;
 606
 607                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 608                         return -EINVAL;
 609                 break;
 610
 611         default:
 612                 return -EINVAL;
 613         }
 614 #undef ACTION
 615         return 0;
 616 }
 617
 618 static int validate_userspace(const struct nlattr *attr)
 619 {
 620         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =
 621         {
 622                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 623                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
 624         };
 625         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 626         int error;
 627
 628         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX, attr, userspace_policy);
 629         if (error)
 630                 return error;
 631
 632         if (!a[OVS_USERSPACE_ATTR_PID] || !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 633                 return -EINVAL;
 634
 635         return 0;
 636 }
 637
 638 static int validate_actions(const struct nlattr *attr,
 639                                 const struct sw_flow_key *key,  int depth)
 640 {
 641         const struct nlattr *a;
 642         int rem, err;
 643
 644         if (depth >= SAMPLE_ACTION_DEPTH)
 645                 return -EOVERFLOW;
 646
 647         nla_for_each_nested(a, attr, rem) {
 648                 /* Expected argument lengths, (u32)-1 for variable length. */
 649                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 650                         [OVS_ACTION_ATTR_OUTPUT] = 4,
 651                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 652                         [OVS_ACTION_ATTR_PUSH] = (u32)-1,
 653                         [OVS_ACTION_ATTR_POP] = 2,
 654                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 655                         [OVS_ACTION_ATTR_SET_PRIORITY] = 4,
 656                         [OVS_ACTION_ATTR_POP_PRIORITY] = 0,
 657                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 658                 };
 659                 int type = nla_type(a);
 660
 661                 if (type > OVS_ACTION_ATTR_MAX ||
 662                     (action_lens[type] != nla_len(a) &&
 663                      action_lens[type] != (u32)-1))
 664                         return -EINVAL;
 665
 666                 switch (type) {
 667                 case OVS_ACTION_ATTR_UNSPEC:
 668                         return -EINVAL;
 669
 670                 case OVS_ACTION_ATTR_SET_PRIORITY:
 671                 case OVS_ACTION_ATTR_POP_PRIORITY:
 672                         /* No validation needed. */
 673                         break;
 674
 675                 case OVS_ACTION_ATTR_USERSPACE:
 676                         err = validate_userspace(a);
 677                         if (err)
 678                                 return err;
 679                         break;
 680
 681                 case OVS_ACTION_ATTR_OUTPUT:
 682                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 683                                 return -EINVAL;
 684                         break;
 685
 686
 687                 case OVS_ACTION_ATTR_POP:
 688                         if (nla_get_u16(a) != OVS_KEY_ATTR_8021Q)
 689                                 return -EINVAL;
 690                         break;
 691
 692                 case OVS_ACTION_ATTR_SET:
 693                 case OVS_ACTION_ATTR_PUSH:
 694                         err = validate_action_key(a, key);
 695                         if (err)
 696                                 return err;
 697                         break;
 698
 699                 case OVS_ACTION_ATTR_SAMPLE:
 700                         err = validate_sample(a, key, depth);
 701                         if (err)
 702                                 return err;
 703                         break;
 704
 705                 default:
 706                         return -EINVAL;
 707                 }
 708         }
 709
 710         if (rem > 0)
 711                 return -EINVAL;
 712
 713         return 0;
 714 }
 715
 716 static void clear_stats(struct sw_flow *flow)
 717 {
 718         flow->used = 0;
 719         flow->tcp_flags = 0;
 720         flow->packet_count = 0;
 721         flow->byte_count = 0;
 722 }
 723
 724 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 725 {
 726         struct ovs_header *ovs_header = info->userhdr;
 727         struct nlattr **a = info->attrs;
 728         struct sw_flow_actions *acts;
 729         struct sk_buff *packet;
 730         struct sw_flow *flow;
 731         struct datapath *dp;
 732         struct ethhdr *eth;
 733         int len;
 734         int err;
 735         int key_len;
 736
 737         err = -EINVAL;
 738         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 739             !a[OVS_PACKET_ATTR_ACTIONS] ||
 740             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 741                 goto err;
 742
 743         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 744         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 745         err = -ENOMEM;
 746         if (!packet)
 747                 goto err;
 748         skb_reserve(packet, NET_IP_ALIGN);
 749
 750         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 751
 752         skb_reset_mac_header(packet);
 753         eth = eth_hdr(packet);
 754
 755         /* Normally, setting the skb 'protocol' field would be handled by a
 756          * call to eth_type_trans(), but it assumes there's a sending
 757          * device, which we may not have. */
 758         if (ntohs(eth->h_proto) >= 1536)
 759                 packet->protocol = eth->h_proto;
 760         else
 761                 packet->protocol = htons(ETH_P_802_2);
 762
 763         /* Build an sw_flow for sending this packet. */
 764         flow = flow_alloc();
 765         err = PTR_ERR(flow);
 766         if (IS_ERR(flow))
 767                 goto err_kfree_skb;
 768
 769         err = flow_extract(packet, -1, &flow->key, &key_len);
 770         if (err)
 771                 goto err_flow_put;
 772
 773         err = flow_metadata_from_nlattrs(&flow->key.eth.in_port,
 774                                          &flow->key.eth.tun_id,
 775                                          a[OVS_PACKET_ATTR_KEY]);
 776         if (err)
 777                 goto err_flow_put;
 778
 779         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
 780         if (err)
 781                 goto err_flow_put;
 782
 783         flow->hash = flow_hash(&flow->key, key_len);
 784
 785         acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 786         err = PTR_ERR(acts);
 787         if (IS_ERR(acts))
 788                 goto err_flow_put;
 789         rcu_assign_pointer(flow->sf_acts, acts);
 790
 791         OVS_CB(packet)->flow = flow;
 792
 793         rcu_read_lock();
 794         dp = get_dp(ovs_header->dp_ifindex);
 795         err = -ENODEV;
 796         if (!dp)
 797                 goto err_unlock;
 798
 799         if (flow->key.eth.in_port < DP_MAX_PORTS)
 800                 OVS_CB(packet)->vport = get_vport_protected(dp,
 801                                                         flow->key.eth.in_port);
 802
 803         local_bh_disable();
 804         err = execute_actions(dp, packet);
 805         local_bh_enable();
 806         rcu_read_unlock();
 807
 808         flow_put(flow);
 809         return err;
 810
 811 err_unlock:
 812         rcu_read_unlock();
 813 err_flow_put:
 814         flow_put(flow);
 815 err_kfree_skb:
 816         kfree_skb(packet);
 817 err:
 818         return err;
 819 }
 820
 821 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 822         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 823         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 824         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 825 };
 826
 827 static struct genl_ops dp_packet_genl_ops[] = {
 828         { .cmd = OVS_PACKET_CMD_EXECUTE,
 829           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 830           .policy = packet_policy,
 831           .doit = ovs_packet_cmd_execute
 832         }
 833 };
 834
 835 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 836 {
 837         int i;
 838         struct flow_table *table = get_table_protected(dp);
 839
 840         stats->n_flows = flow_tbl_count(table);
 841
 842         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 843         for_each_possible_cpu(i) {
 844                 const struct dp_stats_percpu *percpu_stats;
 845                 struct dp_stats_percpu local_stats;
 846                 unsigned seqcount;
 847
 848                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 849
 850                 do {
 851                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 852                         local_stats = *percpu_stats;
 853                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 854
 855                 stats->n_hit += local_stats.n_hit;
 856                 stats->n_missed += local_stats.n_missed;
 857                 stats->n_lost += local_stats.n_lost;
 858         }
 859 }
 860
 861 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 862         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 863         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 864         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 865 };
 866
 867 static struct genl_family dp_flow_genl_family = {
 868         .id = GENL_ID_GENERATE,
 869         .hdrsize = sizeof(struct ovs_header),
 870         .name = OVS_FLOW_FAMILY,
 871         .version = OVS_FLOW_VERSION,
 872         .maxattr = OVS_FLOW_ATTR_MAX
 873 };
 874
 875 static struct genl_multicast_group dp_flow_multicast_group = {
 876         .name = OVS_FLOW_MCGROUP
 877 };
 878
 879 /* Called with genl_lock. */
 880 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 881                                   struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
 882 {
 883         const int skb_orig_len = skb->len;
 884         const struct sw_flow_actions *sf_acts;
 885         struct ovs_flow_stats stats;
 886         struct ovs_header *ovs_header;
 887         struct nlattr *nla;
 888         unsigned long used;
 889         u8 tcp_flags;
 890         int err;
 891
 892         sf_acts = rcu_dereference_protected(flow->sf_acts,
 893                                             lockdep_genl_is_held());
 894
 895         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 896         if (!ovs_header)
 897                 return -EMSGSIZE;
 898
 899         ovs_header->dp_ifindex = get_dpifindex(dp);
 900
 901         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 902         if (!nla)
 903                 goto nla_put_failure;
 904         err = flow_to_nlattrs(&flow->key, skb);
 905         if (err)
 906                 goto error;
 907         nla_nest_end(skb, nla);
 908
 909         spin_lock_bh(&flow->lock);
 910         used = flow->used;
 911         stats.n_packets = flow->packet_count;
 912         stats.n_bytes = flow->byte_count;
 913         tcp_flags = flow->tcp_flags;
 914         spin_unlock_bh(&flow->lock);
 915
 916         if (used)
 917                 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
 918
 919         if (stats.n_packets)
 920                 NLA_PUT(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats);
 921
 922         if (tcp_flags)
 923                 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 924
 925         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 926          * this is the first flow to be dumped into 'skb'.  This is unusual for
 927          * Netlink but individual action lists can be longer than
 928          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 929          * The userspace caller can always fetch the actions separately if it
 930          * really wants them.  (Most userspace callers in fact don't care.)
 931          *
 932          * This can only fail for dump operations because the skb is always
 933          * properly sized for single flows.
 934          */
 935         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 936                       sf_acts->actions);
 937         if (err < 0 && skb_orig_len)
 938                 goto error;
 939
 940         return genlmsg_end(skb, ovs_header);
 941
 942 nla_put_failure:
 943         err = -EMSGSIZE;
 944 error:
 945         genlmsg_cancel(skb, ovs_header);
 946         return err;
 947 }
 948
 949 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 950 {
 951         const struct sw_flow_actions *sf_acts;
 952         int len;
 953
 954         sf_acts = rcu_dereference_protected(flow->sf_acts,
 955                                             lockdep_genl_is_held());
 956
 957         len = nla_total_size(FLOW_BUFSIZE); /* OVS_FLOW_ATTR_KEY */
 958         len += nla_total_size(sf_acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 959         len += nla_total_size(sizeof(struct ovs_flow_stats)); /* OVS_FLOW_ATTR_STATS */
 960         len += nla_total_size(1); /* OVS_FLOW_ATTR_TCP_FLAGS */
 961         len += nla_total_size(8); /* OVS_FLOW_ATTR_USED */
 962         return genlmsg_new(NLMSG_ALIGN(sizeof(struct ovs_header)) + len, GFP_KERNEL);
 963 }
 964
 965 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
 966                                                u32 pid, u32 seq, u8 cmd)
 967 {
 968         struct sk_buff *skb;
 969         int retval;
 970
 971         skb = ovs_flow_cmd_alloc_info(flow);
 972         if (!skb)
 973                 return ERR_PTR(-ENOMEM);
 974
 975         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 976         BUG_ON(retval < 0);
 977         return skb;
 978 }
 979
 980 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 981 {
 982         struct nlattr **a = info->attrs;
 983         struct ovs_header *ovs_header = info->userhdr;
 984         struct sw_flow_key key;
 985         struct sw_flow *flow;
 986         struct sk_buff *reply;
 987         struct datapath *dp;
 988         struct flow_table *table;
 989         int error;
 990         int key_len;
 991
 992         /* Extract key. */
 993         error = -EINVAL;
 994         if (!a[OVS_FLOW_ATTR_KEY])
 995                 goto error;
 996         error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
 997         if (error)
 998                 goto error;
 999
1000         /* Validate actions. */
1001         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1002                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
1003                 if (error)
1004                         goto error;
1005         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1006                 error = -EINVAL;
1007                 goto error;
1008         }
1009
1010         dp = get_dp(ovs_header->dp_ifindex);
1011         error = -ENODEV;
1012         if (!dp)
1013                 goto error;
1014
1015         table = get_table_protected(dp);
1016         flow = flow_tbl_lookup(table, &key, key_len);
1017         if (!flow) {
1018                 struct sw_flow_actions *acts;
1019
1020                 /* Bail out if we're not allowed to create a new flow. */
1021                 error = -ENOENT;
1022                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1023                         goto error;
1024
1025                 /* Expand table, if necessary, to make room. */
1026                 if (flow_tbl_need_to_expand(table)) {
1027                         struct flow_table *new_table;
1028
1029                         new_table = flow_tbl_expand(table);
1030                         if (!IS_ERR(new_table)) {
1031                                 rcu_assign_pointer(dp->table, new_table);
1032                                 flow_tbl_deferred_destroy(table);
1033                                 table = get_table_protected(dp);
1034                         }
1035                 }
1036
1037                 /* Allocate flow. */
1038                 flow = flow_alloc();
1039                 if (IS_ERR(flow)) {
1040                         error = PTR_ERR(flow);
1041                         goto error;
1042                 }
1043                 flow->key = key;
1044                 clear_stats(flow);
1045
1046                 /* Obtain actions. */
1047                 acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1048                 error = PTR_ERR(acts);
1049                 if (IS_ERR(acts))
1050                         goto error_free_flow;
1051                 rcu_assign_pointer(flow->sf_acts, acts);
1052
1053                 /* Put flow in bucket. */
1054                 flow->hash = flow_hash(&key, key_len);
1055                 flow_tbl_insert(table, flow);
1056
1057                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1058                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
1059         } else {
1060                 /* We found a matching flow. */
1061                 struct sw_flow_actions *old_acts;
1062
1063                 /* Bail out if we're not allowed to modify an existing flow.
1064                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1065                  * because Generic Netlink treats the latter as a dump
1066                  * request.  We also accept NLM_F_EXCL in case that bug ever
1067                  * gets fixed.
1068                  */
1069                 error = -EEXIST;
1070                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1071                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1072                         goto error;
1073
1074                 /* Update actions. */
1075                 old_acts = rcu_dereference_protected(flow->sf_acts,
1076                                                      lockdep_genl_is_held());
1077                 if (a[OVS_FLOW_ATTR_ACTIONS] &&
1078                     (old_acts->actions_len != nla_len(a[OVS_FLOW_ATTR_ACTIONS]) ||
1079                      memcmp(old_acts->actions, nla_data(a[OVS_FLOW_ATTR_ACTIONS]),
1080                             old_acts->actions_len))) {
1081                         struct sw_flow_actions *new_acts;
1082
1083                         new_acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1084                         error = PTR_ERR(new_acts);
1085                         if (IS_ERR(new_acts))
1086                                 goto error;
1087
1088                         rcu_assign_pointer(flow->sf_acts, new_acts);
1089                         flow_deferred_free_acts(old_acts);
1090                 }
1091
1092                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1093                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
1094
1095                 /* Clear stats. */
1096                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1097                         spin_lock_bh(&flow->lock);
1098                         clear_stats(flow);
1099                         spin_unlock_bh(&flow->lock);
1100                 }
1101         }
1102
1103         if (!IS_ERR(reply))
1104                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1105                             dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1106         else
1107                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1108                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1109         return 0;
1110
1111 error_free_flow:
1112         flow_put(flow);
1113 error:
1114         return error;
1115 }
1116
1117 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1118 {
1119         struct nlattr **a = info->attrs;
1120         struct ovs_header *ovs_header = info->userhdr;
1121         struct sw_flow_key key;
1122         struct sk_buff *reply;
1123         struct sw_flow *flow;
1124         struct datapath *dp;
1125         struct flow_table *table;
1126         int err;
1127         int key_len;
1128
1129         if (!a[OVS_FLOW_ATTR_KEY])
1130                 return -EINVAL;
1131         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1132         if (err)
1133                 return err;
1134
1135         dp = get_dp(ovs_header->dp_ifindex);
1136         if (!dp)
1137                 return -ENODEV;
1138
1139         table = get_table_protected(dp);
1140         flow = flow_tbl_lookup(table, &key, key_len);
1141         if (!flow)
1142                 return -ENOENT;
1143
1144         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, OVS_FLOW_CMD_NEW);
1145         if (IS_ERR(reply))
1146                 return PTR_ERR(reply);
1147
1148         return genlmsg_reply(reply, info);
1149 }
1150
1151 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1152 {
1153         struct nlattr **a = info->attrs;
1154         struct ovs_header *ovs_header = info->userhdr;
1155         struct sw_flow_key key;
1156         struct sk_buff *reply;
1157         struct sw_flow *flow;
1158         struct datapath *dp;
1159         struct flow_table *table;
1160         int err;
1161         int key_len;
1162
1163         if (!a[OVS_FLOW_ATTR_KEY])
1164                 return flush_flows(ovs_header->dp_ifindex);
1165         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1166         if (err)
1167                 return err;
1168
1169         dp = get_dp(ovs_header->dp_ifindex);
1170         if (!dp)
1171                 return -ENODEV;
1172
1173         table = get_table_protected(dp);
1174         flow = flow_tbl_lookup(table, &key, key_len);
1175         if (!flow)
1176                 return -ENOENT;
1177
1178         reply = ovs_flow_cmd_alloc_info(flow);
1179         if (!reply)
1180                 return -ENOMEM;
1181
1182         flow_tbl_remove(table, flow);
1183
1184         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1185                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1186         BUG_ON(err < 0);
1187
1188         flow_deferred_free(flow);
1189
1190         genl_notify(reply, genl_info_net(info), info->snd_pid,
1191                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1192         return 0;
1193 }
1194
1195 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1196 {
1197         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1198         struct datapath *dp;
1199
1200         dp = get_dp(ovs_header->dp_ifindex);
1201         if (!dp)
1202                 return -ENODEV;
1203
1204         for (;;) {
1205                 struct sw_flow *flow;
1206                 u32 bucket, obj;
1207
1208                 bucket = cb->args[0];
1209                 obj = cb->args[1];
1210                 flow = flow_tbl_next(get_table_protected(dp), &bucket, &obj);
1211                 if (!flow)
1212                         break;
1213
1214                 if (ovs_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1215                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1216                                            OVS_FLOW_CMD_NEW) < 0)
1217                         break;
1218
1219                 cb->args[0] = bucket;
1220                 cb->args[1] = obj;
1221         }
1222         return skb->len;
1223 }
1224
1225 static struct genl_ops dp_flow_genl_ops[] = {
1226         { .cmd = OVS_FLOW_CMD_NEW,
1227           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1228           .policy = flow_policy,
1229           .doit = ovs_flow_cmd_new_or_set
1230         },
1231         { .cmd = OVS_FLOW_CMD_DEL,
1232           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1233           .policy = flow_policy,
1234           .doit = ovs_flow_cmd_del
1235         },
1236         { .cmd = OVS_FLOW_CMD_GET,
1237           .flags = 0,               /* OK for unprivileged users. */
1238           .policy = flow_policy,
1239           .doit = ovs_flow_cmd_get,
1240           .dumpit = ovs_flow_cmd_dump
1241         },
1242         { .cmd = OVS_FLOW_CMD_SET,
1243           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1244           .policy = flow_policy,
1245           .doit = ovs_flow_cmd_new_or_set,
1246         },
1247 };
1248
1249 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1250 #ifdef HAVE_NLA_NUL_STRING
1251         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1252 #endif
1253         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1254 };
1255
1256 static struct genl_family dp_datapath_genl_family = {
1257         .id = GENL_ID_GENERATE,
1258         .hdrsize = sizeof(struct ovs_header),
1259         .name = OVS_DATAPATH_FAMILY,
1260         .version = OVS_DATAPATH_VERSION,
1261         .maxattr = OVS_DP_ATTR_MAX
1262 };
1263
1264 static struct genl_multicast_group dp_datapath_multicast_group = {
1265         .name = OVS_DATAPATH_MCGROUP
1266 };
1267
1268 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1269                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1270 {
1271         struct ovs_header *ovs_header;
1272         struct nlattr *nla;
1273         int err;
1274
1275         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1276                                    flags, cmd);
1277         if (!ovs_header)
1278                 goto error;
1279
1280         ovs_header->dp_ifindex = get_dpifindex(dp);
1281
1282         rcu_read_lock();
1283         err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
1284         rcu_read_unlock();
1285         if (err)
1286                 goto nla_put_failure;
1287
1288         nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats));
1289         if (!nla)
1290                 goto nla_put_failure;
1291         get_dp_stats(dp, nla_data(nla));
1292
1293         return genlmsg_end(skb, ovs_header);
1294
1295 nla_put_failure:
1296         genlmsg_cancel(skb, ovs_header);
1297 error:
1298         return -EMSGSIZE;
1299 }
1300
1301 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1302                                              u32 seq, u8 cmd)
1303 {
1304         struct sk_buff *skb;
1305         int retval;
1306
1307         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1308         if (!skb)
1309                 return ERR_PTR(-ENOMEM);
1310
1311         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1312         if (retval < 0) {
1313                 kfree_skb(skb);
1314                 return ERR_PTR(retval);
1315         }
1316         return skb;
1317 }
1318
1319 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1320 {
1321         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1322 }
1323
1324 /* Called with genl_mutex and optionally with RTNL lock also. */
1325 static struct datapath *lookup_datapath(struct ovs_header *ovs_header, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1326 {
1327         struct datapath *dp;
1328
1329         if (!a[OVS_DP_ATTR_NAME])
1330                 dp = get_dp(ovs_header->dp_ifindex);
1331         else {
1332                 struct vport *vport;
1333
1334                 rcu_read_lock();
1335                 vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1336                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1337                 rcu_read_unlock();
1338         }
1339         return dp ? dp : ERR_PTR(-ENODEV);
1340 }
1341
1342 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1343 {
1344         struct nlattr **a = info->attrs;
1345         struct vport_parms parms;
1346         struct sk_buff *reply;
1347         struct datapath *dp;
1348         struct vport *vport;
1349         int err;
1350
1351         err = -EINVAL;
1352         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1353                 goto err;
1354
1355         err = ovs_dp_cmd_validate(a);
1356         if (err)
1357                 goto err;
1358
1359         rtnl_lock();
1360         err = -ENODEV;
1361         if (!try_module_get(THIS_MODULE))
1362                 goto err_unlock_rtnl;
1363
1364         err = -ENOMEM;
1365         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1366         if (dp == NULL)
1367                 goto err_put_module;
1368         INIT_LIST_HEAD(&dp->port_list);
1369
1370         /* Initialize kobject for bridge.  This will be added as
1371          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1372         dp->ifobj.kset = NULL;
1373         kobject_init(&dp->ifobj, &dp_ktype);
1374
1375         /* Allocate table. */
1376         err = -ENOMEM;
1377         rcu_assign_pointer(dp->table, flow_tbl_alloc(TBL_MIN_BUCKETS));
1378         if (!dp->table)
1379                 goto err_free_dp;
1380
1381         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1382         if (!dp->stats_percpu) {
1383                 err = -ENOMEM;
1384                 goto err_destroy_table;
1385         }
1386
1387         /* Set up our datapath device. */
1388         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1389         parms.type = OVS_VPORT_TYPE_INTERNAL;
1390         parms.options = NULL;
1391         parms.dp = dp;
1392         parms.port_no = OVSP_LOCAL;
1393         parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1394
1395         vport = new_vport(&parms);
1396         if (IS_ERR(vport)) {
1397                 err = PTR_ERR(vport);
1398                 if (err == -EBUSY)
1399                         err = -EEXIST;
1400
1401                 goto err_destroy_percpu;
1402         }
1403
1404         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1405         err = PTR_ERR(reply);
1406         if (IS_ERR(reply))
1407                 goto err_destroy_local_port;
1408
1409         list_add_tail(&dp->list_node, &dps);
1410         dp_sysfs_add_dp(dp);
1411
1412         rtnl_unlock();
1413
1414         genl_notify(reply, genl_info_net(info), info->snd_pid,
1415                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1416         return 0;
1417
1418 err_destroy_local_port:
1419         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1420 err_destroy_percpu:
1421         free_percpu(dp->stats_percpu);
1422 err_destroy_table:
1423         flow_tbl_destroy(get_table_protected(dp));
1424 err_free_dp:
1425         kfree(dp);
1426 err_put_module:
1427         module_put(THIS_MODULE);
1428 err_unlock_rtnl:
1429         rtnl_unlock();
1430 err:
1431         return err;
1432 }
1433
1434 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1435 {
1436         struct vport *vport, *next_vport;
1437         struct sk_buff *reply;
1438         struct datapath *dp;
1439         int err;
1440
1441         err = ovs_dp_cmd_validate(info->attrs);
1442         if (err)
1443                 goto exit;
1444
1445         rtnl_lock();
1446         dp = lookup_datapath(info->userhdr, info->attrs);
1447         err = PTR_ERR(dp);
1448         if (IS_ERR(dp))
1449                 goto exit_unlock;
1450
1451         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_DEL);
1452         err = PTR_ERR(reply);
1453         if (IS_ERR(reply))
1454                 goto exit_unlock;
1455
1456         list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1457                 if (vport->port_no != OVSP_LOCAL)
1458                         dp_detach_port(vport);
1459
1460         dp_sysfs_del_dp(dp);
1461         list_del(&dp->list_node);
1462         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1463
1464         /* rtnl_unlock() will wait until all the references to devices that
1465          * are pending unregistration have been dropped.  We do it here to
1466          * ensure that any internal devices (which contain DP pointers) are
1467          * fully destroyed before freeing the datapath.
1468          */
1469         rtnl_unlock();
1470
1471         call_rcu(&dp->rcu, destroy_dp_rcu);
1472         module_put(THIS_MODULE);
1473
1474         genl_notify(reply, genl_info_net(info), info->snd_pid,
1475                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1476
1477         return 0;
1478
1479 exit_unlock:
1480         rtnl_unlock();
1481 exit:
1482         return err;
1483 }
1484
1485 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1486 {
1487         struct sk_buff *reply;
1488         struct datapath *dp;
1489         int err;
1490
1491         err = ovs_dp_cmd_validate(info->attrs);
1492         if (err)
1493                 return err;
1494
1495         dp = lookup_datapath(info->userhdr, info->attrs);
1496         if (IS_ERR(dp))
1497                 return PTR_ERR(dp);
1498
1499         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1500         if (IS_ERR(reply)) {
1501                 err = PTR_ERR(reply);
1502                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1503                                 dp_datapath_multicast_group.id, err);
1504                 return 0;
1505         }
1506
1507         genl_notify(reply, genl_info_net(info), info->snd_pid,
1508                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1509         return 0;
1510 }
1511
1512 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1513 {
1514         struct sk_buff *reply;
1515         struct datapath *dp;
1516         int err;
1517
1518         err = ovs_dp_cmd_validate(info->attrs);
1519         if (err)
1520                 return err;
1521
1522         dp = lookup_datapath(info->userhdr, info->attrs);
1523         if (IS_ERR(dp))
1524                 return PTR_ERR(dp);
1525
1526         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1527         if (IS_ERR(reply))
1528                 return PTR_ERR(reply);
1529
1530         return genlmsg_reply(reply, info);
1531 }
1532
1533 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1534 {
1535         struct datapath *dp;
1536         int skip = cb->args[0];
1537         int i = 0;
1538
1539         list_for_each_entry (dp, &dps, list_node) {
1540                 if (i < skip)
1541                         continue;
1542                 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1543                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1544                                          OVS_DP_CMD_NEW) < 0)
1545                         break;
1546                 i++;
1547         }
1548
1549         cb->args[0] = i;
1550
1551         return skb->len;
1552 }
1553
1554 static struct genl_ops dp_datapath_genl_ops[] = {
1555         { .cmd = OVS_DP_CMD_NEW,
1556           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1557           .policy = datapath_policy,
1558           .doit = ovs_dp_cmd_new
1559         },
1560         { .cmd = OVS_DP_CMD_DEL,
1561           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1562           .policy = datapath_policy,
1563           .doit = ovs_dp_cmd_del
1564         },
1565         { .cmd = OVS_DP_CMD_GET,
1566           .flags = 0,               /* OK for unprivileged users. */
1567           .policy = datapath_policy,
1568           .doit = ovs_dp_cmd_get,
1569           .dumpit = ovs_dp_cmd_dump
1570         },
1571         { .cmd = OVS_DP_CMD_SET,
1572           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1573           .policy = datapath_policy,
1574           .doit = ovs_dp_cmd_set,
1575         },
1576 };
1577
1578 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1579 #ifdef HAVE_NLA_NUL_STRING
1580         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1581         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1582         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1583 #else
1584         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1585         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1586 #endif
1587         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1588         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1589         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1590         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1591 };
1592
1593 static struct genl_family dp_vport_genl_family = {
1594         .id = GENL_ID_GENERATE,
1595         .hdrsize = sizeof(struct ovs_header),
1596         .name = OVS_VPORT_FAMILY,
1597         .version = OVS_VPORT_VERSION,
1598         .maxattr = OVS_VPORT_ATTR_MAX
1599 };
1600
1601 struct genl_multicast_group dp_vport_multicast_group = {
1602         .name = OVS_VPORT_MCGROUP
1603 };
1604
1605 /* Called with RTNL lock or RCU read lock. */
1606 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1607                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1608 {
1609         struct ovs_header *ovs_header;
1610         struct nlattr *nla;
1611         int err;
1612
1613         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1614                                  flags, cmd);
1615         if (!ovs_header)
1616                 return -EMSGSIZE;
1617
1618         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1619
1620         NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1621         NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport_get_type(vport));
1622         NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport_get_name(vport));
1623         NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1624
1625         nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats));
1626         if (!nla)
1627                 goto nla_put_failure;
1628
1629         vport_get_stats(vport, nla_data(nla));
1630
1631         NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1632
1633         err = vport_get_options(vport, skb);
1634         if (err == -EMSGSIZE)
1635                 goto error;
1636
1637         return genlmsg_end(skb, ovs_header);
1638
1639 nla_put_failure:
1640         err = -EMSGSIZE;
1641 error:
1642         genlmsg_cancel(skb, ovs_header);
1643         return err;
1644 }
1645
1646 /* Called with RTNL lock or RCU read lock. */
1647 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1648                                          u32 seq, u8 cmd)
1649 {
1650         struct sk_buff *skb;
1651         int retval;
1652
1653         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1654         if (!skb)
1655                 return ERR_PTR(-ENOMEM);
1656
1657         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1658         if (retval < 0) {
1659                 kfree_skb(skb);
1660                 return ERR_PTR(retval);
1661         }
1662         return skb;
1663 }
1664
1665 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1666 {
1667         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1668 }
1669
1670 /* Called with RTNL lock or RCU read lock. */
1671 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1672                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1673 {
1674         struct datapath *dp;
1675         struct vport *vport;
1676
1677         if (a[OVS_VPORT_ATTR_NAME]) {
1678                 vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1679                 if (!vport)
1680                         return ERR_PTR(-ENODEV);
1681                 return vport;
1682         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1683                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1684
1685                 if (port_no >= DP_MAX_PORTS)
1686                         return ERR_PTR(-EFBIG);
1687
1688                 dp = get_dp(ovs_header->dp_ifindex);
1689                 if (!dp)
1690                         return ERR_PTR(-ENODEV);
1691
1692                 vport = get_vport_protected(dp, port_no);
1693                 if (!vport)
1694                         return ERR_PTR(-ENOENT);
1695                 return vport;
1696         } else
1697                 return ERR_PTR(-EINVAL);
1698 }
1699
1700 /* Called with RTNL lock. */
1701 static int change_vport(struct vport *vport, struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1702 {
1703         int err = 0;
1704
1705         if (a[OVS_VPORT_ATTR_STATS])
1706                 vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1707
1708         if (a[OVS_VPORT_ATTR_ADDRESS])
1709                 err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1710
1711         return err;
1712 }
1713
1714 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1715 {
1716         struct nlattr **a = info->attrs;
1717         struct ovs_header *ovs_header = info->userhdr;
1718         struct vport_parms parms;
1719         struct sk_buff *reply;
1720         struct vport *vport;
1721         struct datapath *dp;
1722         u32 port_no;
1723         int err;
1724
1725         err = -EINVAL;
1726         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1727             !a[OVS_VPORT_ATTR_UPCALL_PID])
1728                 goto exit;
1729
1730         err = ovs_vport_cmd_validate(a);
1731         if (err)
1732                 goto exit;
1733
1734         rtnl_lock();
1735         dp = get_dp(ovs_header->dp_ifindex);
1736         err = -ENODEV;
1737         if (!dp)
1738                 goto exit_unlock;
1739
1740         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1741                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1742
1743                 err = -EFBIG;
1744                 if (port_no >= DP_MAX_PORTS)
1745                         goto exit_unlock;
1746
1747                 vport = get_vport_protected(dp, port_no);
1748                 err = -EBUSY;
1749                 if (vport)
1750                         goto exit_unlock;
1751         } else {
1752                 for (port_no = 1; ; port_no++) {
1753                         if (port_no >= DP_MAX_PORTS) {
1754                                 err = -EFBIG;
1755                                 goto exit_unlock;
1756                         }
1757                         vport = get_vport_protected(dp, port_no);
1758                         if (!vport)
1759                                 break;
1760                 }
1761         }
1762
1763         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1764         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1765         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1766         parms.dp = dp;
1767         parms.port_no = port_no;
1768         parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1769
1770         vport = new_vport(&parms);
1771         err = PTR_ERR(vport);
1772         if (IS_ERR(vport))
1773                 goto exit_unlock;
1774
1775         dp_sysfs_add_if(vport);
1776
1777         err = change_vport(vport, a);
1778         if (!err) {
1779                 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1780                                                  info->snd_seq, OVS_VPORT_CMD_NEW);
1781                 if (IS_ERR(reply))
1782                         err = PTR_ERR(reply);
1783         }
1784         if (err) {
1785                 dp_detach_port(vport);
1786                 goto exit_unlock;
1787         }
1788         genl_notify(reply, genl_info_net(info), info->snd_pid,
1789                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1790
1791
1792 exit_unlock:
1793         rtnl_unlock();
1794 exit:
1795         return err;
1796 }
1797
1798 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1799 {
1800         struct nlattr **a = info->attrs;
1801         struct sk_buff *reply;
1802         struct vport *vport;
1803         int err;
1804
1805         err = ovs_vport_cmd_validate(a);
1806         if (err)
1807                 goto exit;
1808
1809         rtnl_lock();
1810         vport = lookup_vport(info->userhdr, a);
1811         err = PTR_ERR(vport);
1812         if (IS_ERR(vport))
1813                 goto exit_unlock;
1814
1815         err = 0;
1816         if (a[OVS_VPORT_ATTR_TYPE] && nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport_get_type(vport))
1817                 err = -EINVAL;
1818         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1819                 err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1820         if (!err)
1821                 err = change_vport(vport, a);
1822         if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1823                 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1824
1825         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1826                                          OVS_VPORT_CMD_NEW);
1827         if (IS_ERR(reply)) {
1828                 err = PTR_ERR(reply);
1829                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1830                                 dp_vport_multicast_group.id, err);
1831                 return 0;
1832         }
1833
1834         genl_notify(reply, genl_info_net(info), info->snd_pid,
1835                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1836
1837 exit_unlock:
1838         rtnl_unlock();
1839 exit:
1840         return err;
1841 }
1842
1843 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1844 {
1845         struct nlattr **a = info->attrs;
1846         struct sk_buff *reply;
1847         struct vport *vport;
1848         int err;
1849
1850         err = ovs_vport_cmd_validate(a);
1851         if (err)
1852                 goto exit;
1853
1854         rtnl_lock();
1855         vport = lookup_vport(info->userhdr, a);
1856         err = PTR_ERR(vport);
1857         if (IS_ERR(vport))
1858                 goto exit_unlock;
1859
1860         if (vport->port_no == OVSP_LOCAL) {
1861                 err = -EINVAL;
1862                 goto exit_unlock;
1863         }
1864
1865         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1866                                          OVS_VPORT_CMD_DEL);
1867         err = PTR_ERR(reply);
1868         if (IS_ERR(reply))
1869                 goto exit_unlock;
1870
1871         dp_detach_port(vport);
1872
1873         genl_notify(reply, genl_info_net(info), info->snd_pid,
1874                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1875
1876 exit_unlock:
1877         rtnl_unlock();
1878 exit:
1879         return err;
1880 }
1881
1882 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1883 {
1884         struct nlattr **a = info->attrs;
1885         struct ovs_header *ovs_header = info->userhdr;
1886         struct sk_buff *reply;
1887         struct vport *vport;
1888         int err;
1889
1890         err = ovs_vport_cmd_validate(a);
1891         if (err)
1892                 goto exit;
1893
1894         rcu_read_lock();
1895         vport = lookup_vport(ovs_header, a);
1896         err = PTR_ERR(vport);
1897         if (IS_ERR(vport))
1898                 goto exit_unlock;
1899
1900         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1901                                          OVS_VPORT_CMD_NEW);
1902         err = PTR_ERR(reply);
1903         if (IS_ERR(reply))
1904                 goto exit_unlock;
1905
1906         rcu_read_unlock();
1907
1908         return genlmsg_reply(reply, info);
1909
1910 exit_unlock:
1911         rcu_read_unlock();
1912 exit:
1913         return err;
1914 }
1915
1916 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1917 {
1918         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1919         struct datapath *dp;
1920         u32 port_no;
1921         int retval;
1922
1923         dp = get_dp(ovs_header->dp_ifindex);
1924         if (!dp)
1925                 return -ENODEV;
1926
1927         rcu_read_lock();
1928         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1929                 struct vport *vport;
1930
1931                 vport = get_vport_protected(dp, port_no);
1932                 if (!vport)
1933                         continue;
1934
1935                 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1936                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1937                                             OVS_VPORT_CMD_NEW) < 0)
1938                         break;
1939         }
1940         rcu_read_unlock();
1941
1942         cb->args[0] = port_no;
1943         retval = skb->len;
1944
1945         return retval;
1946 }
1947
1948 static struct genl_ops dp_vport_genl_ops[] = {
1949         { .cmd = OVS_VPORT_CMD_NEW,
1950           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1951           .policy = vport_policy,
1952           .doit = ovs_vport_cmd_new
1953         },
1954         { .cmd = OVS_VPORT_CMD_DEL,
1955           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1956           .policy = vport_policy,
1957           .doit = ovs_vport_cmd_del
1958         },
1959         { .cmd = OVS_VPORT_CMD_GET,
1960           .flags = 0,               /* OK for unprivileged users. */
1961           .policy = vport_policy,
1962           .doit = ovs_vport_cmd_get,
1963           .dumpit = ovs_vport_cmd_dump
1964         },
1965         { .cmd = OVS_VPORT_CMD_SET,
1966           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1967           .policy = vport_policy,
1968           .doit = ovs_vport_cmd_set,
1969         },
1970 };
1971
1972 struct genl_family_and_ops {
1973         struct genl_family *family;
1974         struct genl_ops *ops;
1975         int n_ops;
1976         struct genl_multicast_group *group;
1977 };
1978
1979 static const struct genl_family_and_ops dp_genl_families[] = {
1980         { &dp_datapath_genl_family,
1981           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1982           &dp_datapath_multicast_group },
1983         { &dp_vport_genl_family,
1984           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1985           &dp_vport_multicast_group },
1986         { &dp_flow_genl_family,
1987           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1988           &dp_flow_multicast_group },
1989         { &dp_packet_genl_family,
1990           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1991           NULL },
1992 };
1993
1994 static void dp_unregister_genl(int n_families)
1995 {
1996         int i;
1997
1998         for (i = 0; i < n_families; i++)
1999                 genl_unregister_family(dp_genl_families[i].family);
2000 }
2001
2002 static int dp_register_genl(void)
2003 {
2004         int n_registered;
2005         int err;
2006         int i;
2007
2008         n_registered = 0;
2009         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2010                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2011
2012                 err = genl_register_family_with_ops(f->family, f->ops,
2013                                                     f->n_ops);
2014                 if (err)
2015                         goto error;
2016                 n_registered++;
2017
2018                 if (f->group) {
2019                         err = genl_register_mc_group(f->family, f->group);
2020                         if (err)
2021                                 goto error;
2022                 }
2023         }
2024
2025         return 0;
2026
2027 error:
2028         dp_unregister_genl(n_registered);
2029         return err;
2030 }
2031
2032 static int __init dp_init(void)
2033 {
2034         struct sk_buff *dummy_skb;
2035         int err;
2036
2037         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2038
2039         printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2040
2041         err = tnl_init();
2042         if (err)
2043                 goto error;
2044
2045         err = flow_init();
2046         if (err)
2047                 goto error_tnl_exit;
2048
2049         err = vport_init();
2050         if (err)
2051                 goto error_flow_exit;
2052
2053         err = register_netdevice_notifier(&dp_device_notifier);
2054         if (err)
2055                 goto error_vport_exit;
2056
2057         err = dp_register_genl();
2058         if (err < 0)
2059                 goto error_unreg_notifier;
2060
2061         return 0;
2062
2063 error_unreg_notifier:
2064         unregister_netdevice_notifier(&dp_device_notifier);
2065 error_vport_exit:
2066         vport_exit();
2067 error_flow_exit:
2068         flow_exit();
2069 error_tnl_exit:
2070         tnl_exit();
2071 error:
2072         return err;
2073 }
2074
2075 static void dp_cleanup(void)
2076 {
2077         rcu_barrier();
2078         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2079         unregister_netdevice_notifier(&dp_device_notifier);
2080         vport_exit();
2081         flow_exit();
2082         tnl_exit();
2083 }
2084
2085 module_init(dp_init);
2086 module_exit(dp_cleanup);
2087
2088 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2089 MODULE_LICENSE("GPL");