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/rculist.h>
  43 #include <linux/dmi.h>
  44 #include <net/inet_ecn.h>
  45 #include <net/genetlink.h>
  46
  47 #include "openvswitch/datapath-protocol.h"
  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(3,1,0)
  58 #error Kernels before 2.6.18 or after 3.0 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 such as drop frags, etc.) are protected by genl_mutex.  The RTNL
  72  * lock nests inside 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_userspace_packets(struct datapath *, u32 pid, struct sk_buff *,
  88                                  const struct dp_upcall_info *);
  89
  90 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  91 struct datapath *get_dp(int dp_ifindex)
  92 {
  93         struct datapath *dp = NULL;
  94         struct net_device *dev;
  95
  96         rcu_read_lock();
  97         dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
  98         if (dev) {
  99                 struct vport *vport = internal_dev_get_vport(dev);
 100                 if (vport)
 101                         dp = vport->dp;
 102         }
 103         rcu_read_unlock();
 104
 105         return dp;
 106 }
 107 EXPORT_SYMBOL_GPL(get_dp);
 108
 109 /* Must be called with genl_mutex. */
 110 static struct flow_table *get_table_protected(struct datapath *dp)
 111 {
 112         return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
 113 }
 114
 115 /* Must be called with rcu_read_lock or RTNL lock. */
 116 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
 117 {
 118         return rcu_dereference_rtnl(dp->ports[port_no]);
 119 }
 120
 121 /* Must be called with rcu_read_lock or RTNL lock. */
 122 const char *dp_name(const struct datapath *dp)
 123 {
 124         return vport_get_name(rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]));
 125 }
 126
 127 static int get_dpifindex(struct datapath *dp)
 128 {
 129         struct vport *local;
 130         int ifindex;
 131
 132         rcu_read_lock();
 133
 134         local = get_vport_protected(dp, OVSP_LOCAL);
 135         if (local)
 136                 ifindex = vport_get_ifindex(local);
 137         else
 138                 ifindex = 0;
 139
 140         rcu_read_unlock();
 141
 142         return ifindex;
 143 }
 144
 145 static inline size_t br_nlmsg_size(void)
 146 {
 147         return NLMSG_ALIGN(sizeof(struct ifinfomsg))
 148                + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
 149                + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
 150                + nla_total_size(4) /* IFLA_MASTER */
 151                + nla_total_size(4) /* IFLA_MTU */
 152                + nla_total_size(1); /* IFLA_OPERSTATE */
 153 }
 154
 155 /* Caller must hold RTNL lock. */
 156 static int dp_fill_ifinfo(struct sk_buff *skb,
 157                           const struct vport *port,
 158                           int event, unsigned int flags)
 159 {
 160         struct datapath *dp = port->dp;
 161         int ifindex = vport_get_ifindex(port);
 162         struct ifinfomsg *hdr;
 163         struct nlmsghdr *nlh;
 164
 165         if (ifindex < 0)
 166                 return ifindex;
 167
 168         nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
 169         if (nlh == NULL)
 170                 return -EMSGSIZE;
 171
 172         hdr = nlmsg_data(nlh);
 173         hdr->ifi_family = AF_BRIDGE;
 174         hdr->__ifi_pad = 0;
 175         hdr->ifi_type = ARPHRD_ETHER;
 176         hdr->ifi_index = ifindex;
 177         hdr->ifi_flags = vport_get_flags(port);
 178         hdr->ifi_change = 0;
 179
 180         NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
 181         NLA_PUT_U32(skb, IFLA_MASTER, get_dpifindex(dp));
 182         NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
 183 #ifdef IFLA_OPERSTATE
 184         NLA_PUT_U8(skb, IFLA_OPERSTATE,
 185                    vport_is_running(port)
 186                         ? vport_get_operstate(port)
 187                         : IF_OPER_DOWN);
 188 #endif
 189
 190         NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
 191
 192         return nlmsg_end(skb, nlh);
 193
 194 nla_put_failure:
 195         nlmsg_cancel(skb, nlh);
 196         return -EMSGSIZE;
 197 }
 198
 199 /* Caller must hold RTNL lock. */
 200 static void dp_ifinfo_notify(int event, struct vport *port)
 201 {
 202         struct sk_buff *skb;
 203         int err = -ENOBUFS;
 204
 205         skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
 206         if (skb == NULL)
 207                 goto errout;
 208
 209         err = dp_fill_ifinfo(skb, port, event, 0);
 210         if (err < 0) {
 211                 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
 212                 WARN_ON(err == -EMSGSIZE);
 213                 kfree_skb(skb);
 214                 goto errout;
 215         }
 216         rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 217         return;
 218 errout:
 219         if (err < 0)
 220                 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
 221 }
 222
 223 static void release_dp(struct kobject *kobj)
 224 {
 225         struct datapath *dp = container_of(kobj, struct datapath, ifobj);
 226         kfree(dp);
 227 }
 228
 229 static struct kobj_type dp_ktype = {
 230         .release = release_dp
 231 };
 232
 233 static void destroy_dp_rcu(struct rcu_head *rcu)
 234 {
 235         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 236
 237         flow_tbl_destroy(dp->table);
 238         free_percpu(dp->stats_percpu);
 239         kobject_put(&dp->ifobj);
 240 }
 241
 242 /* Called with RTNL lock and genl_lock. */
 243 static struct vport *new_vport(const struct vport_parms *parms)
 244 {
 245         struct vport *vport;
 246
 247         vport = vport_add(parms);
 248         if (!IS_ERR(vport)) {
 249                 struct datapath *dp = parms->dp;
 250
 251                 rcu_assign_pointer(dp->ports[parms->port_no], vport);
 252                 list_add(&vport->node, &dp->port_list);
 253
 254                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 255         }
 256
 257         return vport;
 258 }
 259
 260 /* Called with RTNL lock. */
 261 void dp_detach_port(struct vport *p)
 262 {
 263         ASSERT_RTNL();
 264
 265         if (p->port_no != OVSP_LOCAL)
 266                 dp_sysfs_del_if(p);
 267         dp_ifinfo_notify(RTM_DELLINK, p);
 268
 269         /* First drop references to device. */
 270         list_del(&p->node);
 271         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 272
 273         /* Then destroy it. */
 274         vport_del(p);
 275 }
 276
 277 /* Must be called with rcu_read_lock. */
 278 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 279 {
 280         struct datapath *dp = p->dp;
 281         struct sw_flow *flow;
 282         struct dp_stats_percpu *stats;
 283         int stats_counter_off;
 284         int error;
 285
 286         OVS_CB(skb)->vport = p;
 287
 288         if (!OVS_CB(skb)->flow) {
 289                 struct sw_flow_key key;
 290                 int key_len;
 291                 bool is_frag;
 292
 293                 /* Extract flow from 'skb' into 'key'. */
 294                 error = flow_extract(skb, p->port_no, &key, &key_len, &is_frag);
 295                 if (unlikely(error)) {
 296                         kfree_skb(skb);
 297                         return;
 298                 }
 299
 300                 if (is_frag && dp->drop_frags) {
 301                         consume_skb(skb);
 302                         stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
 303                         goto out;
 304                 }
 305
 306                 /* Look up flow. */
 307                 flow = flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
 308                 if (unlikely(!flow)) {
 309                         struct dp_upcall_info upcall;
 310
 311                         upcall.cmd = OVS_PACKET_CMD_MISS;
 312                         upcall.key = &key;
 313                         dp_upcall(dp, skb, &upcall);
 314                         kfree_skb(skb);
 315                         stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
 316                         goto out;
 317                 }
 318
 319                 OVS_CB(skb)->flow = flow;
 320         }
 321
 322         stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
 323         flow_used(OVS_CB(skb)->flow, skb);
 324         execute_actions(dp, skb);
 325
 326 out:
 327         /* Update datapath statistics. */
 328         local_bh_disable();
 329         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 330
 331         write_seqcount_begin(&stats->seqlock);
 332         (*(u64 *)((u8 *)stats + stats_counter_off))++;
 333         write_seqcount_end(&stats->seqlock);
 334
 335         local_bh_enable();
 336 }
 337
 338 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
 339 {
 340         u16 csum_start, csum_offset;
 341         __wsum csum;
 342
 343         get_skb_csum_pointers(skb, &csum_start, &csum_offset);
 344         csum_start -= skb_headroom(skb);
 345
 346         skb_copy_bits(skb, 0, to, csum_start);
 347
 348         csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
 349                                       skb->len - csum_start, 0);
 350         *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
 351 }
 352
 353 static struct genl_family dp_packet_genl_family = {
 354         .id = GENL_ID_GENERATE,
 355         .hdrsize = sizeof(struct ovs_header),
 356         .name = OVS_PACKET_FAMILY,
 357         .version = 1,
 358         .maxattr = OVS_PACKET_ATTR_MAX
 359 };
 360
 361 int dp_upcall(struct datapath *dp, struct sk_buff *skb,
 362               const struct dp_upcall_info *upcall_info)
 363 {
 364         struct sk_buff *segs = NULL;
 365         struct dp_stats_percpu *stats;
 366         u32 pid;
 367         int err;
 368
 369         if (OVS_CB(skb)->flow)
 370                 pid = OVS_CB(skb)->flow->upcall_pid;
 371         else
 372                 pid = OVS_CB(skb)->vport->upcall_pid;
 373
 374         if (pid == 0) {
 375                 err = -ENOTCONN;
 376                 goto err;
 377         }
 378
 379         forward_ip_summed(skb, true);
 380
 381         /* Break apart GSO packets into their component pieces.  Otherwise
 382          * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
 383         if (skb_is_gso(skb)) {
 384                 segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 385
 386                 if (IS_ERR(segs)) {
 387                         err = PTR_ERR(segs);
 388                         goto err;
 389                 }
 390                 skb = segs;
 391         }
 392
 393         err = queue_userspace_packets(dp, pid, skb, upcall_info);
 394         if (segs) {
 395                 struct sk_buff *next;
 396                 /* Free GSO-segments */
 397                 do {
 398                         next = segs->next;
 399                         kfree_skb(segs);
 400                 } while ((segs = next) != NULL);
 401         }
 402
 403         if (err)
 404                 goto err;
 405
 406         return 0;
 407
 408 err:
 409         local_bh_disable();
 410         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 411
 412         write_seqcount_begin(&stats->seqlock);
 413         stats->n_lost++;
 414         write_seqcount_end(&stats->seqlock);
 415
 416         local_bh_enable();
 417
 418         return err;
 419 }
 420
 421 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
 422  * 'upcall_info'.  There will be only one packet unless we broke up a GSO
 423  * packet.
 424  */
 425 static int queue_userspace_packets(struct datapath *dp, u32 pid,
 426                                    struct sk_buff *skb,
 427                                    const struct dp_upcall_info *upcall_info)
 428 {
 429         int dp_ifindex;
 430
 431         dp_ifindex = get_dpifindex(dp);
 432         if (!dp_ifindex)
 433                 return -ENODEV;
 434
 435         do {
 436                 struct ovs_header *upcall;
 437                 struct sk_buff *user_skb; /* to be queued to userspace */
 438                 struct nlattr *nla;
 439                 unsigned int len;
 440                 int err;
 441
 442                 err = vlan_deaccel_tag(skb);
 443                 if (unlikely(err))
 444                         return err;
 445
 446                 if (nla_attr_size(skb->len) > USHRT_MAX)
 447                         return -EFBIG;
 448
 449                 len = sizeof(struct ovs_header);
 450                 len += nla_total_size(skb->len);
 451                 len += nla_total_size(FLOW_BUFSIZE);
 452                 if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
 453                         len += nla_total_size(8);
 454
 455                 user_skb = genlmsg_new(len, GFP_ATOMIC);
 456                 if (!user_skb)
 457                         return -ENOMEM;
 458
 459                 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 460                                          0, upcall_info->cmd);
 461                 upcall->dp_ifindex = dp_ifindex;
 462
 463                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 464                 flow_to_nlattrs(upcall_info->key, user_skb);
 465                 nla_nest_end(user_skb, nla);
 466
 467                 if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
 468                         nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
 469                                                 upcall_info->userdata);
 470
 471                 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 472                 if (skb->ip_summed == CHECKSUM_PARTIAL)
 473                         copy_and_csum_skb(skb, nla_data(nla));
 474                 else
 475                         skb_copy_bits(skb, 0, nla_data(nla), skb->len);
 476
 477                 err = genlmsg_unicast(&init_net, user_skb, pid);
 478                 if (err)
 479                         return err;
 480
 481         } while ((skb = skb->next));
 482
 483         return 0;
 484 }
 485
 486 /* Called with genl_mutex. */
 487 static int flush_flows(int dp_ifindex)
 488 {
 489         struct flow_table *old_table;
 490         struct flow_table *new_table;
 491         struct datapath *dp;
 492
 493         dp = get_dp(dp_ifindex);
 494         if (!dp)
 495                 return -ENODEV;
 496
 497         old_table = get_table_protected(dp);
 498         new_table = flow_tbl_alloc(TBL_MIN_BUCKETS);
 499         if (!new_table)
 500                 return -ENOMEM;
 501
 502         rcu_assign_pointer(dp->table, new_table);
 503
 504         flow_tbl_deferred_destroy(old_table);
 505         return 0;
 506 }
 507
 508 static int validate_actions(const struct nlattr *attr, int depth);
 509
 510 static int validate_sample(const struct nlattr *attr, int depth)
 511 {
 512         static const struct nla_policy sample_policy[OVS_SAMPLE_ATTR_MAX + 1] =
 513         {
 514                 [OVS_SAMPLE_ATTR_PROBABILITY] = {.type = NLA_U32 },
 515                 [OVS_SAMPLE_ATTR_ACTIONS] = {.type = NLA_UNSPEC },
 516         };
 517         struct nlattr *a[OVS_SAMPLE_ATTR_MAX + 1];
 518         int error;
 519
 520         error = nla_parse_nested(a, OVS_SAMPLE_ATTR_MAX, attr, sample_policy);
 521         if (error)
 522                 return error;
 523
 524         if (!a[OVS_SAMPLE_ATTR_PROBABILITY])
 525                 return -EINVAL;
 526         if (!a[OVS_SAMPLE_ATTR_ACTIONS])
 527                 return -EINVAL;
 528
 529         return validate_actions(a[OVS_SAMPLE_ATTR_ACTIONS], (depth + 1));
 530 }
 531
 532 static int validate_actions(const struct nlattr *attr, int depth)
 533 {
 534         const struct nlattr *a;
 535         int rem, err;
 536
 537         if (depth >= SAMPLE_ACTION_DEPTH)
 538                 return -EOVERFLOW;
 539
 540         nla_for_each_nested(a, attr, rem) {
 541                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 542                         [OVS_ACTION_ATTR_OUTPUT] = 4,
 543                         [OVS_ACTION_ATTR_USERSPACE] = 8,
 544                         [OVS_ACTION_ATTR_PUSH_VLAN] = 2,
 545                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 546                         [OVS_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
 547                         [OVS_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
 548                         [OVS_ACTION_ATTR_SET_NW_SRC] = 4,
 549                         [OVS_ACTION_ATTR_SET_NW_DST] = 4,
 550                         [OVS_ACTION_ATTR_SET_NW_TOS] = 1,
 551                         [OVS_ACTION_ATTR_SET_TP_SRC] = 2,
 552                         [OVS_ACTION_ATTR_SET_TP_DST] = 2,
 553                         [OVS_ACTION_ATTR_SET_TUNNEL] = 8,
 554                         [OVS_ACTION_ATTR_SET_PRIORITY] = 4,
 555                         [OVS_ACTION_ATTR_POP_PRIORITY] = 0,
 556                 };
 557                 int type = nla_type(a);
 558
 559                 /* Match expected attr len for given attr len except for
 560                  * OVS_ACTION_ATTR_SAMPLE, as it has nested actions list which
 561                  * is variable size. */
 562                 if (type > OVS_ACTION_ATTR_MAX ||
 563                    (nla_len(a) != action_lens[type] &&
 564                           type != OVS_ACTION_ATTR_SAMPLE))
 565                         return -EINVAL;
 566
 567                 switch (type) {
 568                 case OVS_ACTION_ATTR_UNSPEC:
 569                         return -EINVAL;
 570
 571                 case OVS_ACTION_ATTR_USERSPACE:
 572                 case OVS_ACTION_ATTR_POP_VLAN:
 573                 case OVS_ACTION_ATTR_SET_DL_SRC:
 574                 case OVS_ACTION_ATTR_SET_DL_DST:
 575                 case OVS_ACTION_ATTR_SET_NW_SRC:
 576                 case OVS_ACTION_ATTR_SET_NW_DST:
 577                 case OVS_ACTION_ATTR_SET_TP_SRC:
 578                 case OVS_ACTION_ATTR_SET_TP_DST:
 579                 case OVS_ACTION_ATTR_SET_TUNNEL:
 580                 case OVS_ACTION_ATTR_SET_PRIORITY:
 581                 case OVS_ACTION_ATTR_POP_PRIORITY:
 582                         /* No validation needed. */
 583                         break;
 584
 585                 case OVS_ACTION_ATTR_OUTPUT:
 586                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 587                                 return -EINVAL;
 588                         break;
 589
 590                 case OVS_ACTION_ATTR_PUSH_VLAN:
 591                         if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
 592                                 return -EINVAL;
 593                         break;
 594
 595                 case OVS_ACTION_ATTR_SET_NW_TOS:
 596                         if (nla_get_u8(a) & INET_ECN_MASK)
 597                                 return -EINVAL;
 598                         break;
 599
 600                 case OVS_ACTION_ATTR_SAMPLE:
 601                         err = validate_sample(a, depth);
 602                         if (err)
 603                                 return err;
 604                         break;
 605
 606                 default:
 607                         return -EOPNOTSUPP;
 608                 }
 609         }
 610
 611         if (rem > 0)
 612                 return -EINVAL;
 613
 614         return 0;
 615 }
 616 static void clear_stats(struct sw_flow *flow)
 617 {
 618         flow->used = 0;
 619         flow->tcp_flags = 0;
 620         flow->packet_count = 0;
 621         flow->byte_count = 0;
 622 }
 623
 624 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 625 {
 626         struct ovs_header *ovs_header = info->userhdr;
 627         struct nlattr **a = info->attrs;
 628         struct sw_flow_actions *acts;
 629         struct sk_buff *packet;
 630         struct sw_flow *flow;
 631         struct datapath *dp;
 632         struct ethhdr *eth;
 633         bool is_frag;
 634         int len;
 635         int err;
 636         int key_len;
 637
 638         err = -EINVAL;
 639         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 640             !a[OVS_PACKET_ATTR_ACTIONS] ||
 641             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 642                 goto err;
 643
 644         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], 0);
 645         if (err)
 646                 goto err;
 647
 648         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 649         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 650         err = -ENOMEM;
 651         if (!packet)
 652                 goto err;
 653         skb_reserve(packet, NET_IP_ALIGN);
 654
 655         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 656
 657         skb_reset_mac_header(packet);
 658         eth = eth_hdr(packet);
 659
 660         /* Normally, setting the skb 'protocol' field would be handled by a
 661          * call to eth_type_trans(), but it assumes there's a sending
 662          * device, which we may not have. */
 663         if (ntohs(eth->h_proto) >= 1536)
 664                 packet->protocol = eth->h_proto;
 665         else
 666                 packet->protocol = htons(ETH_P_802_2);
 667
 668         /* Build an sw_flow for sending this packet. */
 669         flow = flow_alloc();
 670         err = PTR_ERR(flow);
 671         if (IS_ERR(flow))
 672                 goto err_kfree_skb;
 673
 674         err = flow_extract(packet, -1, &flow->key, &key_len, &is_frag);
 675         if (err)
 676                 goto err_flow_put;
 677
 678         err = flow_metadata_from_nlattrs(&flow->key.eth.in_port,
 679                                          &flow->key.eth.tun_id,
 680                                          a[OVS_PACKET_ATTR_KEY]);
 681         if (err)
 682                 goto err_flow_put;
 683
 684         flow->hash = flow_hash(&flow->key, key_len);
 685
 686         if (a[OVS_PACKET_ATTR_UPCALL_PID])
 687                 flow->upcall_pid = nla_get_u32(a[OVS_PACKET_ATTR_UPCALL_PID]);
 688         else
 689                 flow->upcall_pid = NETLINK_CB(skb).pid;
 690
 691         acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 692         err = PTR_ERR(acts);
 693         if (IS_ERR(acts))
 694                 goto err_flow_put;
 695         rcu_assign_pointer(flow->sf_acts, acts);
 696
 697         OVS_CB(packet)->flow = flow;
 698
 699         rcu_read_lock();
 700         dp = get_dp(ovs_header->dp_ifindex);
 701         err = -ENODEV;
 702         if (!dp)
 703                 goto err_unlock;
 704
 705         if (flow->key.eth.in_port < DP_MAX_PORTS)
 706                 OVS_CB(packet)->vport = get_vport_protected(dp,
 707                                                         flow->key.eth.in_port);
 708
 709         err = execute_actions(dp, packet);
 710         rcu_read_unlock();
 711
 712         flow_put(flow);
 713         return err;
 714
 715 err_unlock:
 716         rcu_read_unlock();
 717 err_flow_put:
 718         flow_put(flow);
 719 err_kfree_skb:
 720         kfree_skb(packet);
 721 err:
 722         return err;
 723 }
 724
 725 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 726         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 727         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 728         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 729         [OVS_PACKET_ATTR_UPCALL_PID] = { .type = NLA_U32 },
 730 };
 731
 732 static struct genl_ops dp_packet_genl_ops[] = {
 733         { .cmd = OVS_PACKET_CMD_EXECUTE,
 734           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 735           .policy = packet_policy,
 736           .doit = ovs_packet_cmd_execute
 737         }
 738 };
 739
 740 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 741 {
 742         int i;
 743         struct flow_table *table = get_table_protected(dp);
 744
 745         stats->n_flows = flow_tbl_count(table);
 746
 747         stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
 748         for_each_possible_cpu(i) {
 749                 const struct dp_stats_percpu *percpu_stats;
 750                 struct dp_stats_percpu local_stats;
 751                 unsigned seqcount;
 752
 753                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 754
 755                 do {
 756                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 757                         local_stats = *percpu_stats;
 758                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 759
 760                 stats->n_frags += local_stats.n_frags;
 761                 stats->n_hit += local_stats.n_hit;
 762                 stats->n_missed += local_stats.n_missed;
 763                 stats->n_lost += local_stats.n_lost;
 764         }
 765 }
 766
 767 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 768         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 769         [OVS_FLOW_ATTR_UPCALL_PID] = { .type = NLA_U32 },
 770         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 771         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 772 };
 773
 774 static struct genl_family dp_flow_genl_family = {
 775         .id = GENL_ID_GENERATE,
 776         .hdrsize = sizeof(struct ovs_header),
 777         .name = OVS_FLOW_FAMILY,
 778         .version = 1,
 779         .maxattr = OVS_FLOW_ATTR_MAX
 780 };
 781
 782 static struct genl_multicast_group dp_flow_multicast_group = {
 783         .name = OVS_FLOW_MCGROUP
 784 };
 785
 786 /* Called with genl_lock. */
 787 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 788                                   struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
 789 {
 790         const int skb_orig_len = skb->len;
 791         const struct sw_flow_actions *sf_acts;
 792         struct ovs_flow_stats stats;
 793         struct ovs_header *ovs_header;
 794         struct nlattr *nla;
 795         unsigned long used;
 796         u8 tcp_flags;
 797         int err;
 798
 799         sf_acts = rcu_dereference_protected(flow->sf_acts,
 800                                             lockdep_genl_is_held());
 801
 802         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 803         if (!ovs_header)
 804                 return -EMSGSIZE;
 805
 806         ovs_header->dp_ifindex = get_dpifindex(dp);
 807
 808         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 809         if (!nla)
 810                 goto nla_put_failure;
 811         err = flow_to_nlattrs(&flow->key, skb);
 812         if (err)
 813                 goto error;
 814         nla_nest_end(skb, nla);
 815
 816         NLA_PUT_U32(skb, OVS_FLOW_ATTR_UPCALL_PID, flow->upcall_pid);
 817
 818         spin_lock_bh(&flow->lock);
 819         used = flow->used;
 820         stats.n_packets = flow->packet_count;
 821         stats.n_bytes = flow->byte_count;
 822         tcp_flags = flow->tcp_flags;
 823         spin_unlock_bh(&flow->lock);
 824
 825         if (used)
 826                 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
 827
 828         if (stats.n_packets)
 829                 NLA_PUT(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats);
 830
 831         if (tcp_flags)
 832                 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 833
 834         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 835          * this is the first flow to be dumped into 'skb'.  This is unusual for
 836          * Netlink but individual action lists can be longer than
 837          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 838          * The userspace caller can always fetch the actions separately if it
 839          * really wants them.  (Most userspace callers in fact don't care.)
 840          *
 841          * This can only fail for dump operations because the skb is always
 842          * properly sized for single flows.
 843          */
 844         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 845                       sf_acts->actions);
 846         if (err < 0 && skb_orig_len)
 847                 goto error;
 848
 849         return genlmsg_end(skb, ovs_header);
 850
 851 nla_put_failure:
 852         err = -EMSGSIZE;
 853 error:
 854         genlmsg_cancel(skb, ovs_header);
 855         return err;
 856 }
 857
 858 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 859 {
 860         const struct sw_flow_actions *sf_acts;
 861         int len;
 862
 863         sf_acts = rcu_dereference_protected(flow->sf_acts,
 864                                             lockdep_genl_is_held());
 865
 866         len = nla_total_size(FLOW_BUFSIZE); /* OVS_FLOW_ATTR_KEY */
 867         len += nla_total_size(sf_acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 868         len += nla_total_size(sizeof(struct ovs_flow_stats)); /* OVS_FLOW_ATTR_STATS */
 869         len += nla_total_size(1); /* OVS_FLOW_ATTR_TCP_FLAGS */
 870         len += nla_total_size(8); /* OVS_FLOW_ATTR_USED */
 871         return genlmsg_new(NLMSG_ALIGN(sizeof(struct ovs_header)) + len, GFP_KERNEL);
 872 }
 873
 874 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
 875                                                u32 pid, u32 seq, u8 cmd)
 876 {
 877         struct sk_buff *skb;
 878         int retval;
 879
 880         skb = ovs_flow_cmd_alloc_info(flow);
 881         if (!skb)
 882                 return ERR_PTR(-ENOMEM);
 883
 884         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 885         BUG_ON(retval < 0);
 886         return skb;
 887 }
 888
 889 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 890 {
 891         struct nlattr **a = info->attrs;
 892         struct ovs_header *ovs_header = info->userhdr;
 893         struct sw_flow_key key;
 894         struct sw_flow *flow;
 895         struct sk_buff *reply;
 896         struct datapath *dp;
 897         struct flow_table *table;
 898         int error;
 899         int key_len;
 900
 901         /* Extract key. */
 902         error = -EINVAL;
 903         if (!a[OVS_FLOW_ATTR_KEY])
 904                 goto error;
 905         error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
 906         if (error)
 907                 goto error;
 908
 909         /* Validate actions. */
 910         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 911                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], 0);
 912                 if (error)
 913                         goto error;
 914         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 915                 error = -EINVAL;
 916                 goto error;
 917         }
 918
 919         dp = get_dp(ovs_header->dp_ifindex);
 920         error = -ENODEV;
 921         if (!dp)
 922                 goto error;
 923
 924         table = get_table_protected(dp);
 925         flow = flow_tbl_lookup(table, &key, key_len);
 926         if (!flow) {
 927                 struct sw_flow_actions *acts;
 928
 929                 /* Bail out if we're not allowed to create a new flow. */
 930                 error = -ENOENT;
 931                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
 932                         goto error;
 933
 934                 /* Expand table, if necessary, to make room. */
 935                 if (flow_tbl_need_to_expand(table)) {
 936                         struct flow_table *new_table;
 937
 938                         new_table = flow_tbl_expand(table);
 939                         if (!IS_ERR(new_table)) {
 940                                 rcu_assign_pointer(dp->table, new_table);
 941                                 flow_tbl_deferred_destroy(table);
 942                                 table = get_table_protected(dp);
 943                         }
 944                 }
 945
 946                 /* Allocate flow. */
 947                 flow = flow_alloc();
 948                 if (IS_ERR(flow)) {
 949                         error = PTR_ERR(flow);
 950                         goto error;
 951                 }
 952                 flow->key = key;
 953                 clear_stats(flow);
 954
 955                 if (a[OVS_FLOW_ATTR_UPCALL_PID])
 956                         flow->upcall_pid = nla_get_u32(a[OVS_FLOW_ATTR_UPCALL_PID]);
 957                 else
 958                         flow->upcall_pid = NETLINK_CB(skb).pid;
 959
 960                 /* Obtain actions. */
 961                 acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
 962                 error = PTR_ERR(acts);
 963                 if (IS_ERR(acts))
 964                         goto error_free_flow;
 965                 rcu_assign_pointer(flow->sf_acts, acts);
 966
 967                 /* Put flow in bucket. */
 968                 flow->hash = flow_hash(&key, key_len);
 969                 flow_tbl_insert(table, flow);
 970
 971                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
 972                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
 973         } else {
 974                 /* We found a matching flow. */
 975                 struct sw_flow_actions *old_acts;
 976
 977                 /* Bail out if we're not allowed to modify an existing flow.
 978                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 979                  * because Generic Netlink treats the latter as a dump
 980                  * request.  We also accept NLM_F_EXCL in case that bug ever
 981                  * gets fixed.
 982                  */
 983                 error = -EEXIST;
 984                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
 985                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
 986                         goto error;
 987
 988                 /* Update actions. */
 989                 old_acts = rcu_dereference_protected(flow->sf_acts,
 990                                                      lockdep_genl_is_held());
 991                 if (a[OVS_FLOW_ATTR_ACTIONS] &&
 992                     (old_acts->actions_len != nla_len(a[OVS_FLOW_ATTR_ACTIONS]) ||
 993                      memcmp(old_acts->actions, nla_data(a[OVS_FLOW_ATTR_ACTIONS]),
 994                             old_acts->actions_len))) {
 995                         struct sw_flow_actions *new_acts;
 996
 997                         new_acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
 998                         error = PTR_ERR(new_acts);
 999                         if (IS_ERR(new_acts))
1000                                 goto error;
1001
1002                         rcu_assign_pointer(flow->sf_acts, new_acts);
1003                         flow_deferred_free_acts(old_acts);
1004                 }
1005
1006                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1007                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
1008
1009                 if (a[OVS_FLOW_ATTR_UPCALL_PID])
1010                         flow->upcall_pid = nla_get_u32(a[OVS_FLOW_ATTR_UPCALL_PID]);
1011
1012                 /* Clear stats. */
1013                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1014                         spin_lock_bh(&flow->lock);
1015                         clear_stats(flow);
1016                         spin_unlock_bh(&flow->lock);
1017                 }
1018         }
1019
1020         if (!IS_ERR(reply))
1021                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1022                             dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1023         else
1024                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1025                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1026         return 0;
1027
1028 error_free_flow:
1029         flow_put(flow);
1030 error:
1031         return error;
1032 }
1033
1034 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1035 {
1036         struct nlattr **a = info->attrs;
1037         struct ovs_header *ovs_header = info->userhdr;
1038         struct sw_flow_key key;
1039         struct sk_buff *reply;
1040         struct sw_flow *flow;
1041         struct datapath *dp;
1042         struct flow_table *table;
1043         int err;
1044         int key_len;
1045
1046         if (!a[OVS_FLOW_ATTR_KEY])
1047                 return -EINVAL;
1048         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1049         if (err)
1050                 return err;
1051
1052         dp = get_dp(ovs_header->dp_ifindex);
1053         if (!dp)
1054                 return -ENODEV;
1055
1056         table = get_table_protected(dp);
1057         flow = flow_tbl_lookup(table, &key, key_len);
1058         if (!flow)
1059                 return -ENOENT;
1060
1061         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, OVS_FLOW_CMD_NEW);
1062         if (IS_ERR(reply))
1063                 return PTR_ERR(reply);
1064
1065         return genlmsg_reply(reply, info);
1066 }
1067
1068 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1069 {
1070         struct nlattr **a = info->attrs;
1071         struct ovs_header *ovs_header = info->userhdr;
1072         struct sw_flow_key key;
1073         struct sk_buff *reply;
1074         struct sw_flow *flow;
1075         struct datapath *dp;
1076         struct flow_table *table;
1077         int err;
1078         int key_len;
1079
1080         if (!a[OVS_FLOW_ATTR_KEY])
1081                 return flush_flows(ovs_header->dp_ifindex);
1082         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1083         if (err)
1084                 return err;
1085
1086         dp = get_dp(ovs_header->dp_ifindex);
1087         if (!dp)
1088                 return -ENODEV;
1089
1090         table = get_table_protected(dp);
1091         flow = flow_tbl_lookup(table, &key, key_len);
1092         if (!flow)
1093                 return -ENOENT;
1094
1095         reply = ovs_flow_cmd_alloc_info(flow);
1096         if (!reply)
1097                 return -ENOMEM;
1098
1099         flow_tbl_remove(table, flow);
1100
1101         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1102                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1103         BUG_ON(err < 0);
1104
1105         flow_deferred_free(flow);
1106
1107         genl_notify(reply, genl_info_net(info), info->snd_pid,
1108                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1109         return 0;
1110 }
1111
1112 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1113 {
1114         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1115         struct datapath *dp;
1116
1117         dp = get_dp(ovs_header->dp_ifindex);
1118         if (!dp)
1119                 return -ENODEV;
1120
1121         for (;;) {
1122                 struct sw_flow *flow;
1123                 u32 bucket, obj;
1124
1125                 bucket = cb->args[0];
1126                 obj = cb->args[1];
1127                 flow = flow_tbl_next(get_table_protected(dp), &bucket, &obj);
1128                 if (!flow)
1129                         break;
1130
1131                 if (ovs_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1132                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1133                                            OVS_FLOW_CMD_NEW) < 0)
1134                         break;
1135
1136                 cb->args[0] = bucket;
1137                 cb->args[1] = obj;
1138         }
1139         return skb->len;
1140 }
1141
1142 static struct genl_ops dp_flow_genl_ops[] = {
1143         { .cmd = OVS_FLOW_CMD_NEW,
1144           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1145           .policy = flow_policy,
1146           .doit = ovs_flow_cmd_new_or_set
1147         },
1148         { .cmd = OVS_FLOW_CMD_DEL,
1149           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1150           .policy = flow_policy,
1151           .doit = ovs_flow_cmd_del
1152         },
1153         { .cmd = OVS_FLOW_CMD_GET,
1154           .flags = 0,               /* OK for unprivileged users. */
1155           .policy = flow_policy,
1156           .doit = ovs_flow_cmd_get,
1157           .dumpit = ovs_flow_cmd_dump
1158         },
1159         { .cmd = OVS_FLOW_CMD_SET,
1160           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1161           .policy = flow_policy,
1162           .doit = ovs_flow_cmd_new_or_set,
1163         },
1164 };
1165
1166 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1167 #ifdef HAVE_NLA_NUL_STRING
1168         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1169 #endif
1170         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1171         [OVS_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1172 };
1173
1174 static struct genl_family dp_datapath_genl_family = {
1175         .id = GENL_ID_GENERATE,
1176         .hdrsize = sizeof(struct ovs_header),
1177         .name = OVS_DATAPATH_FAMILY,
1178         .version = 1,
1179         .maxattr = OVS_DP_ATTR_MAX
1180 };
1181
1182 static struct genl_multicast_group dp_datapath_multicast_group = {
1183         .name = OVS_DATAPATH_MCGROUP
1184 };
1185
1186 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1187                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1188 {
1189         struct ovs_header *ovs_header;
1190         struct nlattr *nla;
1191         int err;
1192
1193         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1194                                    flags, cmd);
1195         if (!ovs_header)
1196                 goto error;
1197
1198         ovs_header->dp_ifindex = get_dpifindex(dp);
1199
1200         rcu_read_lock();
1201         err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
1202         rcu_read_unlock();
1203         if (err)
1204                 goto nla_put_failure;
1205
1206         nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats));
1207         if (!nla)
1208                 goto nla_put_failure;
1209         get_dp_stats(dp, nla_data(nla));
1210
1211         NLA_PUT_U32(skb, OVS_DP_ATTR_IPV4_FRAGS,
1212                     dp->drop_frags ? OVS_DP_FRAG_DROP : OVS_DP_FRAG_ZERO);
1213
1214         return genlmsg_end(skb, ovs_header);
1215
1216 nla_put_failure:
1217         genlmsg_cancel(skb, ovs_header);
1218 error:
1219         return -EMSGSIZE;
1220 }
1221
1222 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1223                                              u32 seq, u8 cmd)
1224 {
1225         struct sk_buff *skb;
1226         int retval;
1227
1228         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1229         if (!skb)
1230                 return ERR_PTR(-ENOMEM);
1231
1232         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1233         if (retval < 0) {
1234                 kfree_skb(skb);
1235                 return ERR_PTR(retval);
1236         }
1237         return skb;
1238 }
1239
1240 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1241 {
1242         if (a[OVS_DP_ATTR_IPV4_FRAGS]) {
1243                 u32 frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]);
1244
1245                 if (frags != OVS_DP_FRAG_ZERO && frags != OVS_DP_FRAG_DROP)
1246                         return -EINVAL;
1247         }
1248
1249         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1250 }
1251
1252 /* Called with genl_mutex and optionally with RTNL lock also. */
1253 static struct datapath *lookup_datapath(struct ovs_header *ovs_header, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1254 {
1255         struct datapath *dp;
1256
1257         if (!a[OVS_DP_ATTR_NAME])
1258                 dp = get_dp(ovs_header->dp_ifindex);
1259         else {
1260                 struct vport *vport;
1261
1262                 rcu_read_lock();
1263                 vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1264                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1265                 rcu_read_unlock();
1266         }
1267         return dp ? dp : ERR_PTR(-ENODEV);
1268 }
1269
1270 /* Called with genl_mutex. */
1271 static void change_datapath(struct datapath *dp, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1272 {
1273         if (a[OVS_DP_ATTR_IPV4_FRAGS])
1274                 dp->drop_frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]) == OVS_DP_FRAG_DROP;
1275 }
1276
1277 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1278 {
1279         struct nlattr **a = info->attrs;
1280         struct vport_parms parms;
1281         struct sk_buff *reply;
1282         struct datapath *dp;
1283         struct vport *vport;
1284         int err;
1285
1286         err = -EINVAL;
1287         if (!a[OVS_DP_ATTR_NAME])
1288                 goto err;
1289
1290         err = ovs_dp_cmd_validate(a);
1291         if (err)
1292                 goto err;
1293
1294         rtnl_lock();
1295         err = -ENODEV;
1296         if (!try_module_get(THIS_MODULE))
1297                 goto err_unlock_rtnl;
1298
1299         err = -ENOMEM;
1300         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1301         if (dp == NULL)
1302                 goto err_put_module;
1303         INIT_LIST_HEAD(&dp->port_list);
1304
1305         /* Initialize kobject for bridge.  This will be added as
1306          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1307         dp->ifobj.kset = NULL;
1308         kobject_init(&dp->ifobj, &dp_ktype);
1309
1310         /* Allocate table. */
1311         err = -ENOMEM;
1312         rcu_assign_pointer(dp->table, flow_tbl_alloc(TBL_MIN_BUCKETS));
1313         if (!dp->table)
1314                 goto err_free_dp;
1315
1316         dp->drop_frags = 0;
1317         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1318         if (!dp->stats_percpu) {
1319                 err = -ENOMEM;
1320                 goto err_destroy_table;
1321         }
1322
1323         change_datapath(dp, a);
1324
1325         /* Set up our datapath device. */
1326         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1327         parms.type = OVS_VPORT_TYPE_INTERNAL;
1328         parms.options = NULL;
1329         parms.dp = dp;
1330         parms.port_no = OVSP_LOCAL;
1331         if (a[OVS_DP_ATTR_UPCALL_PID])
1332                 parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1333         else
1334                 parms.upcall_pid = NETLINK_CB(skb).pid;
1335
1336         vport = new_vport(&parms);
1337         if (IS_ERR(vport)) {
1338                 err = PTR_ERR(vport);
1339                 if (err == -EBUSY)
1340                         err = -EEXIST;
1341
1342                 goto err_destroy_percpu;
1343         }
1344
1345         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1346         err = PTR_ERR(reply);
1347         if (IS_ERR(reply))
1348                 goto err_destroy_local_port;
1349
1350         list_add_tail(&dp->list_node, &dps);
1351         dp_sysfs_add_dp(dp);
1352
1353         rtnl_unlock();
1354
1355         genl_notify(reply, genl_info_net(info), info->snd_pid,
1356                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1357         return 0;
1358
1359 err_destroy_local_port:
1360         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1361 err_destroy_percpu:
1362         free_percpu(dp->stats_percpu);
1363 err_destroy_table:
1364         flow_tbl_destroy(get_table_protected(dp));
1365 err_free_dp:
1366         kfree(dp);
1367 err_put_module:
1368         module_put(THIS_MODULE);
1369 err_unlock_rtnl:
1370         rtnl_unlock();
1371 err:
1372         return err;
1373 }
1374
1375 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1376 {
1377         struct vport *vport, *next_vport;
1378         struct sk_buff *reply;
1379         struct datapath *dp;
1380         int err;
1381
1382         err = ovs_dp_cmd_validate(info->attrs);
1383         if (err)
1384                 goto exit;
1385
1386         rtnl_lock();
1387         dp = lookup_datapath(info->userhdr, info->attrs);
1388         err = PTR_ERR(dp);
1389         if (IS_ERR(dp))
1390                 goto exit_unlock;
1391
1392         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_DEL);
1393         err = PTR_ERR(reply);
1394         if (IS_ERR(reply))
1395                 goto exit_unlock;
1396
1397         list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1398                 if (vport->port_no != OVSP_LOCAL)
1399                         dp_detach_port(vport);
1400
1401         dp_sysfs_del_dp(dp);
1402         list_del(&dp->list_node);
1403         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1404
1405         /* rtnl_unlock() will wait until all the references to devices that
1406          * are pending unregistration have been dropped.  We do it here to
1407          * ensure that any internal devices (which contain DP pointers) are
1408          * fully destroyed before freeing the datapath.
1409          */
1410         rtnl_unlock();
1411
1412         call_rcu(&dp->rcu, destroy_dp_rcu);
1413         module_put(THIS_MODULE);
1414
1415         genl_notify(reply, genl_info_net(info), info->snd_pid,
1416                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1417
1418         return 0;
1419
1420 exit_unlock:
1421         rtnl_unlock();
1422 exit:
1423         return err;
1424 }
1425
1426 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1427 {
1428         struct sk_buff *reply;
1429         struct datapath *dp;
1430         int err;
1431
1432         err = ovs_dp_cmd_validate(info->attrs);
1433         if (err)
1434                 return err;
1435
1436         dp = lookup_datapath(info->userhdr, info->attrs);
1437         if (IS_ERR(dp))
1438                 return PTR_ERR(dp);
1439
1440         change_datapath(dp, info->attrs);
1441
1442         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1443         if (IS_ERR(reply)) {
1444                 err = PTR_ERR(reply);
1445                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1446                                 dp_datapath_multicast_group.id, err);
1447                 return 0;
1448         }
1449
1450         genl_notify(reply, genl_info_net(info), info->snd_pid,
1451                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1452         return 0;
1453 }
1454
1455 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1456 {
1457         struct sk_buff *reply;
1458         struct datapath *dp;
1459         int err;
1460
1461         err = ovs_dp_cmd_validate(info->attrs);
1462         if (err)
1463                 return err;
1464
1465         dp = lookup_datapath(info->userhdr, info->attrs);
1466         if (IS_ERR(dp))
1467                 return PTR_ERR(dp);
1468
1469         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1470         if (IS_ERR(reply))
1471                 return PTR_ERR(reply);
1472
1473         return genlmsg_reply(reply, info);
1474 }
1475
1476 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1477 {
1478         struct datapath *dp;
1479         int skip = cb->args[0];
1480         int i = 0;
1481
1482         list_for_each_entry (dp, &dps, list_node) {
1483                 if (i < skip)
1484                         continue;
1485                 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1486                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1487                                          OVS_DP_CMD_NEW) < 0)
1488                         break;
1489                 i++;
1490         }
1491
1492         cb->args[0] = i;
1493
1494         return skb->len;
1495 }
1496
1497 static struct genl_ops dp_datapath_genl_ops[] = {
1498         { .cmd = OVS_DP_CMD_NEW,
1499           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1500           .policy = datapath_policy,
1501           .doit = ovs_dp_cmd_new
1502         },
1503         { .cmd = OVS_DP_CMD_DEL,
1504           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1505           .policy = datapath_policy,
1506           .doit = ovs_dp_cmd_del
1507         },
1508         { .cmd = OVS_DP_CMD_GET,
1509           .flags = 0,               /* OK for unprivileged users. */
1510           .policy = datapath_policy,
1511           .doit = ovs_dp_cmd_get,
1512           .dumpit = ovs_dp_cmd_dump
1513         },
1514         { .cmd = OVS_DP_CMD_SET,
1515           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1516           .policy = datapath_policy,
1517           .doit = ovs_dp_cmd_set,
1518         },
1519 };
1520
1521 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1522 #ifdef HAVE_NLA_NUL_STRING
1523         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1524         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1525         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1526 #else
1527         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1528         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1529 #endif
1530         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1531         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1532         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1533         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1534 };
1535
1536 static struct genl_family dp_vport_genl_family = {
1537         .id = GENL_ID_GENERATE,
1538         .hdrsize = sizeof(struct ovs_header),
1539         .name = OVS_VPORT_FAMILY,
1540         .version = 1,
1541         .maxattr = OVS_VPORT_ATTR_MAX
1542 };
1543
1544 struct genl_multicast_group dp_vport_multicast_group = {
1545         .name = OVS_VPORT_MCGROUP
1546 };
1547
1548 /* Called with RTNL lock or RCU read lock. */
1549 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1550                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1551 {
1552         struct ovs_header *ovs_header;
1553         struct nlattr *nla;
1554         int err;
1555
1556         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1557                                  flags, cmd);
1558         if (!ovs_header)
1559                 return -EMSGSIZE;
1560
1561         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1562
1563         NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1564         NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport_get_type(vport));
1565         NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport_get_name(vport));
1566         NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1567
1568         nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats));
1569         if (!nla)
1570                 goto nla_put_failure;
1571
1572         vport_get_stats(vport, nla_data(nla));
1573
1574         NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1575
1576         err = vport_get_options(vport, skb);
1577         if (err == -EMSGSIZE)
1578                 goto error;
1579
1580         return genlmsg_end(skb, ovs_header);
1581
1582 nla_put_failure:
1583         err = -EMSGSIZE;
1584 error:
1585         genlmsg_cancel(skb, ovs_header);
1586         return err;
1587 }
1588
1589 /* Called with RTNL lock or RCU read lock. */
1590 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1591                                          u32 seq, u8 cmd)
1592 {
1593         struct sk_buff *skb;
1594         int retval;
1595
1596         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1597         if (!skb)
1598                 return ERR_PTR(-ENOMEM);
1599
1600         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1601         if (retval < 0) {
1602                 kfree_skb(skb);
1603                 return ERR_PTR(retval);
1604         }
1605         return skb;
1606 }
1607
1608 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1609 {
1610         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1611 }
1612
1613 /* Called with RTNL lock or RCU read lock. */
1614 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1615                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1616 {
1617         struct datapath *dp;
1618         struct vport *vport;
1619
1620         if (a[OVS_VPORT_ATTR_NAME]) {
1621                 vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1622                 if (!vport)
1623                         return ERR_PTR(-ENODEV);
1624                 return vport;
1625         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1626                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1627
1628                 if (port_no >= DP_MAX_PORTS)
1629                         return ERR_PTR(-EFBIG);
1630
1631                 dp = get_dp(ovs_header->dp_ifindex);
1632                 if (!dp)
1633                         return ERR_PTR(-ENODEV);
1634
1635                 vport = get_vport_protected(dp, port_no);
1636                 if (!vport)
1637                         return ERR_PTR(-ENOENT);
1638                 return vport;
1639         } else
1640                 return ERR_PTR(-EINVAL);
1641 }
1642
1643 /* Called with RTNL lock. */
1644 static int change_vport(struct vport *vport, struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1645 {
1646         int err = 0;
1647
1648         if (a[OVS_VPORT_ATTR_STATS])
1649                 vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1650
1651         if (a[OVS_VPORT_ATTR_ADDRESS])
1652                 err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1653
1654         return err;
1655 }
1656
1657 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1658 {
1659         struct nlattr **a = info->attrs;
1660         struct ovs_header *ovs_header = info->userhdr;
1661         struct vport_parms parms;
1662         struct sk_buff *reply;
1663         struct vport *vport;
1664         struct datapath *dp;
1665         u32 port_no;
1666         int err;
1667
1668         err = -EINVAL;
1669         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE])
1670                 goto exit;
1671
1672         err = ovs_vport_cmd_validate(a);
1673         if (err)
1674                 goto exit;
1675
1676         rtnl_lock();
1677         dp = get_dp(ovs_header->dp_ifindex);
1678         err = -ENODEV;
1679         if (!dp)
1680                 goto exit_unlock;
1681
1682         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1683                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1684
1685                 err = -EFBIG;
1686                 if (port_no >= DP_MAX_PORTS)
1687                         goto exit_unlock;
1688
1689                 vport = get_vport_protected(dp, port_no);
1690                 err = -EBUSY;
1691                 if (vport)
1692                         goto exit_unlock;
1693         } else {
1694                 for (port_no = 1; ; port_no++) {
1695                         if (port_no >= DP_MAX_PORTS) {
1696                                 err = -EFBIG;
1697                                 goto exit_unlock;
1698                         }
1699                         vport = get_vport_protected(dp, port_no);
1700                         if (!vport)
1701                                 break;
1702                 }
1703         }
1704
1705         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1706         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1707         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1708         parms.dp = dp;
1709         parms.port_no = port_no;
1710         if (a[OVS_VPORT_ATTR_UPCALL_PID])
1711                 parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1712         else
1713                 parms.upcall_pid = NETLINK_CB(skb).pid;
1714
1715         vport = new_vport(&parms);
1716         err = PTR_ERR(vport);
1717         if (IS_ERR(vport))
1718                 goto exit_unlock;
1719
1720         dp_sysfs_add_if(vport);
1721
1722         err = change_vport(vport, a);
1723         if (!err) {
1724                 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1725                                                  info->snd_seq, OVS_VPORT_CMD_NEW);
1726                 if (IS_ERR(reply))
1727                         err = PTR_ERR(reply);
1728         }
1729         if (err) {
1730                 dp_detach_port(vport);
1731                 goto exit_unlock;
1732         }
1733         genl_notify(reply, genl_info_net(info), info->snd_pid,
1734                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1735
1736
1737 exit_unlock:
1738         rtnl_unlock();
1739 exit:
1740         return err;
1741 }
1742
1743 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1744 {
1745         struct nlattr **a = info->attrs;
1746         struct sk_buff *reply;
1747         struct vport *vport;
1748         int err;
1749
1750         err = ovs_vport_cmd_validate(a);
1751         if (err)
1752                 goto exit;
1753
1754         rtnl_lock();
1755         vport = lookup_vport(info->userhdr, a);
1756         err = PTR_ERR(vport);
1757         if (IS_ERR(vport))
1758                 goto exit_unlock;
1759
1760         err = 0;
1761         if (a[OVS_VPORT_ATTR_OPTIONS])
1762                 err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1763         if (!err)
1764                 err = change_vport(vport, a);
1765         if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1766                 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1767
1768         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1769                                          OVS_VPORT_CMD_NEW);
1770         if (IS_ERR(reply)) {
1771                 err = PTR_ERR(reply);
1772                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1773                                 dp_vport_multicast_group.id, err);
1774                 return 0;
1775         }
1776
1777         genl_notify(reply, genl_info_net(info), info->snd_pid,
1778                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1779
1780 exit_unlock:
1781         rtnl_unlock();
1782 exit:
1783         return err;
1784 }
1785
1786 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1787 {
1788         struct nlattr **a = info->attrs;
1789         struct sk_buff *reply;
1790         struct vport *vport;
1791         int err;
1792
1793         err = ovs_vport_cmd_validate(a);
1794         if (err)
1795                 goto exit;
1796
1797         rtnl_lock();
1798         vport = lookup_vport(info->userhdr, a);
1799         err = PTR_ERR(vport);
1800         if (IS_ERR(vport))
1801                 goto exit_unlock;
1802
1803         if (vport->port_no == OVSP_LOCAL) {
1804                 err = -EINVAL;
1805                 goto exit_unlock;
1806         }
1807
1808         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1809                                          OVS_VPORT_CMD_DEL);
1810         err = PTR_ERR(reply);
1811         if (IS_ERR(reply))
1812                 goto exit_unlock;
1813
1814         dp_detach_port(vport);
1815
1816         genl_notify(reply, genl_info_net(info), info->snd_pid,
1817                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1818
1819 exit_unlock:
1820         rtnl_unlock();
1821 exit:
1822         return err;
1823 }
1824
1825 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1826 {
1827         struct nlattr **a = info->attrs;
1828         struct ovs_header *ovs_header = info->userhdr;
1829         struct sk_buff *reply;
1830         struct vport *vport;
1831         int err;
1832
1833         err = ovs_vport_cmd_validate(a);
1834         if (err)
1835                 goto exit;
1836
1837         rcu_read_lock();
1838         vport = lookup_vport(ovs_header, a);
1839         err = PTR_ERR(vport);
1840         if (IS_ERR(vport))
1841                 goto exit_unlock;
1842
1843         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1844                                          OVS_VPORT_CMD_NEW);
1845         err = PTR_ERR(reply);
1846         if (IS_ERR(reply))
1847                 goto exit_unlock;
1848
1849         rcu_read_unlock();
1850
1851         return genlmsg_reply(reply, info);
1852
1853 exit_unlock:
1854         rcu_read_unlock();
1855 exit:
1856         return err;
1857 }
1858
1859 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1860 {
1861         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1862         struct datapath *dp;
1863         u32 port_no;
1864         int retval;
1865
1866         dp = get_dp(ovs_header->dp_ifindex);
1867         if (!dp)
1868                 return -ENODEV;
1869
1870         rcu_read_lock();
1871         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1872                 struct vport *vport;
1873
1874                 vport = get_vport_protected(dp, port_no);
1875                 if (!vport)
1876                         continue;
1877
1878                 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1879                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1880                                             OVS_VPORT_CMD_NEW) < 0)
1881                         break;
1882         }
1883         rcu_read_unlock();
1884
1885         cb->args[0] = port_no;
1886         retval = skb->len;
1887
1888         return retval;
1889 }
1890
1891 static struct genl_ops dp_vport_genl_ops[] = {
1892         { .cmd = OVS_VPORT_CMD_NEW,
1893           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1894           .policy = vport_policy,
1895           .doit = ovs_vport_cmd_new
1896         },
1897         { .cmd = OVS_VPORT_CMD_DEL,
1898           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1899           .policy = vport_policy,
1900           .doit = ovs_vport_cmd_del
1901         },
1902         { .cmd = OVS_VPORT_CMD_GET,
1903           .flags = 0,               /* OK for unprivileged users. */
1904           .policy = vport_policy,
1905           .doit = ovs_vport_cmd_get,
1906           .dumpit = ovs_vport_cmd_dump
1907         },
1908         { .cmd = OVS_VPORT_CMD_SET,
1909           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1910           .policy = vport_policy,
1911           .doit = ovs_vport_cmd_set,
1912         },
1913 };
1914
1915 struct genl_family_and_ops {
1916         struct genl_family *family;
1917         struct genl_ops *ops;
1918         int n_ops;
1919         struct genl_multicast_group *group;
1920 };
1921
1922 static const struct genl_family_and_ops dp_genl_families[] = {
1923         { &dp_datapath_genl_family,
1924           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1925           &dp_datapath_multicast_group },
1926         { &dp_vport_genl_family,
1927           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1928           &dp_vport_multicast_group },
1929         { &dp_flow_genl_family,
1930           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1931           &dp_flow_multicast_group },
1932         { &dp_packet_genl_family,
1933           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1934           NULL },
1935 };
1936
1937 static void dp_unregister_genl(int n_families)
1938 {
1939         int i;
1940
1941         for (i = 0; i < n_families; i++)
1942                 genl_unregister_family(dp_genl_families[i].family);
1943 }
1944
1945 static int dp_register_genl(void)
1946 {
1947         int n_registered;
1948         int err;
1949         int i;
1950
1951         n_registered = 0;
1952         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1953                 const struct genl_family_and_ops *f = &dp_genl_families[i];
1954
1955                 err = genl_register_family_with_ops(f->family, f->ops,
1956                                                     f->n_ops);
1957                 if (err)
1958                         goto error;
1959                 n_registered++;
1960
1961                 if (f->group) {
1962                         err = genl_register_mc_group(f->family, f->group);
1963                         if (err)
1964                                 goto error;
1965                 }
1966         }
1967
1968         return 0;
1969
1970 error:
1971         dp_unregister_genl(n_registered);
1972         return err;
1973 }
1974
1975 static int __init dp_init(void)
1976 {
1977         struct sk_buff *dummy_skb;
1978         int err;
1979
1980         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
1981
1982         printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
1983
1984         err = tnl_init();
1985         if (err)
1986                 goto error;
1987
1988         err = flow_init();
1989         if (err)
1990                 goto error_tnl_exit;
1991
1992         err = vport_init();
1993         if (err)
1994                 goto error_flow_exit;
1995
1996         err = register_netdevice_notifier(&dp_device_notifier);
1997         if (err)
1998                 goto error_vport_exit;
1999
2000         err = dp_register_genl();
2001         if (err < 0)
2002                 goto error_unreg_notifier;
2003
2004         return 0;
2005
2006 error_unreg_notifier:
2007         unregister_netdevice_notifier(&dp_device_notifier);
2008 error_vport_exit:
2009         vport_exit();
2010 error_flow_exit:
2011         flow_exit();
2012 error_tnl_exit:
2013         tnl_exit();
2014 error:
2015         return err;
2016 }
2017
2018 static void dp_cleanup(void)
2019 {
2020         rcu_barrier();
2021         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2022         unregister_netdevice_notifier(&dp_device_notifier);
2023         vport_exit();
2024         flow_exit();
2025         tnl_exit();
2026 }
2027
2028 module_init(dp_init);
2029 module_exit(dp_cleanup);
2030
2031 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2032 MODULE_LICENSE("GPL");