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 "table.h"
  53 #include "vlan.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 *, 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 tbl *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[ODPP_LOCAL]));
 125 }
 126
 127 static inline size_t br_nlmsg_size(void)
 128 {
 129         return NLMSG_ALIGN(sizeof(struct ifinfomsg))
 130                + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
 131                + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
 132                + nla_total_size(4) /* IFLA_MASTER */
 133                + nla_total_size(4) /* IFLA_MTU */
 134                + nla_total_size(4) /* IFLA_LINK */
 135                + nla_total_size(1); /* IFLA_OPERSTATE */
 136 }
 137
 138 /* Caller must hold RTNL lock. */
 139 static int dp_fill_ifinfo(struct sk_buff *skb,
 140                           const struct vport *port,
 141                           int event, unsigned int flags)
 142 {
 143         struct datapath *dp = port->dp;
 144         int ifindex = vport_get_ifindex(port);
 145         int iflink = vport_get_iflink(port);
 146         struct ifinfomsg *hdr;
 147         struct nlmsghdr *nlh;
 148
 149         if (ifindex < 0)
 150                 return ifindex;
 151
 152         if (iflink < 0)
 153                 return iflink;
 154
 155         nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
 156         if (nlh == NULL)
 157                 return -EMSGSIZE;
 158
 159         hdr = nlmsg_data(nlh);
 160         hdr->ifi_family = AF_BRIDGE;
 161         hdr->__ifi_pad = 0;
 162         hdr->ifi_type = ARPHRD_ETHER;
 163         hdr->ifi_index = ifindex;
 164         hdr->ifi_flags = vport_get_flags(port);
 165         hdr->ifi_change = 0;
 166
 167         NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
 168         NLA_PUT_U32(skb, IFLA_MASTER,
 169                 vport_get_ifindex(get_vport_protected(dp, ODPP_LOCAL)));
 170         NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
 171 #ifdef IFLA_OPERSTATE
 172         NLA_PUT_U8(skb, IFLA_OPERSTATE,
 173                    vport_is_running(port)
 174                         ? vport_get_operstate(port)
 175                         : IF_OPER_DOWN);
 176 #endif
 177
 178         NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
 179
 180         if (ifindex != iflink)
 181                 NLA_PUT_U32(skb, IFLA_LINK,iflink);
 182
 183         return nlmsg_end(skb, nlh);
 184
 185 nla_put_failure:
 186         nlmsg_cancel(skb, nlh);
 187         return -EMSGSIZE;
 188 }
 189
 190 /* Caller must hold RTNL lock. */
 191 static void dp_ifinfo_notify(int event, struct vport *port)
 192 {
 193         struct sk_buff *skb;
 194         int err = -ENOBUFS;
 195
 196         skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
 197         if (skb == NULL)
 198                 goto errout;
 199
 200         err = dp_fill_ifinfo(skb, port, event, 0);
 201         if (err < 0) {
 202                 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
 203                 WARN_ON(err == -EMSGSIZE);
 204                 kfree_skb(skb);
 205                 goto errout;
 206         }
 207         rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 208         return;
 209 errout:
 210         if (err < 0)
 211                 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
 212 }
 213
 214 static void release_dp(struct kobject *kobj)
 215 {
 216         struct datapath *dp = container_of(kobj, struct datapath, ifobj);
 217         kfree(dp);
 218 }
 219
 220 static struct kobj_type dp_ktype = {
 221         .release = release_dp
 222 };
 223
 224 static void destroy_dp_rcu(struct rcu_head *rcu)
 225 {
 226         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 227
 228         tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl);
 229         free_percpu(dp->stats_percpu);
 230         kobject_put(&dp->ifobj);
 231 }
 232
 233 /* Called with RTNL lock and genl_lock. */
 234 static struct vport *new_vport(const struct vport_parms *parms)
 235 {
 236         struct vport *vport;
 237
 238         vport = vport_add(parms);
 239         if (!IS_ERR(vport)) {
 240                 struct datapath *dp = parms->dp;
 241
 242                 rcu_assign_pointer(dp->ports[parms->port_no], vport);
 243                 list_add(&vport->node, &dp->port_list);
 244
 245                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 246         }
 247
 248         return vport;
 249 }
 250
 251 /* Called with RTNL lock. */
 252 int dp_detach_port(struct vport *p)
 253 {
 254         ASSERT_RTNL();
 255
 256         if (p->port_no != ODPP_LOCAL)
 257                 dp_sysfs_del_if(p);
 258         dp_ifinfo_notify(RTM_DELLINK, p);
 259
 260         /* First drop references to device. */
 261         list_del(&p->node);
 262         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 263
 264         /* Then destroy it. */
 265         return vport_del(p);
 266 }
 267
 268 /* Must be called with rcu_read_lock. */
 269 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 270 {
 271         struct datapath *dp = p->dp;
 272         struct dp_stats_percpu *stats;
 273         int stats_counter_off;
 274         int error;
 275
 276         OVS_CB(skb)->vport = p;
 277
 278         if (!OVS_CB(skb)->flow) {
 279                 struct sw_flow_key key;
 280                 struct tbl_node *flow_node;
 281                 int key_len;
 282                 bool is_frag;
 283
 284                 /* Extract flow from 'skb' into 'key'. */
 285                 error = flow_extract(skb, p->port_no, &key, &key_len, &is_frag);
 286                 if (unlikely(error)) {
 287                         kfree_skb(skb);
 288                         return;
 289                 }
 290
 291                 if (is_frag && dp->drop_frags) {
 292                         consume_skb(skb);
 293                         stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
 294                         goto out;
 295                 }
 296
 297                 /* Look up flow. */
 298                 flow_node = tbl_lookup(rcu_dereference(dp->table), &key, key_len,
 299                                        flow_hash(&key, key_len), flow_cmp);
 300                 if (unlikely(!flow_node)) {
 301                         struct dp_upcall_info upcall;
 302
 303                         upcall.cmd = ODP_PACKET_CMD_MISS;
 304                         upcall.key = &key;
 305                         upcall.userdata = 0;
 306                         upcall.sample_pool = 0;
 307                         upcall.actions = NULL;
 308                         upcall.actions_len = 0;
 309                         dp_upcall(dp, skb, &upcall);
 310                         stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
 311                         goto out;
 312                 }
 313
 314                 OVS_CB(skb)->flow = flow_cast(flow_node);
 315         }
 316
 317         stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
 318         flow_used(OVS_CB(skb)->flow, skb);
 319         execute_actions(dp, skb);
 320
 321 out:
 322         /* Update datapath statistics. */
 323         local_bh_disable();
 324         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 325
 326         write_seqcount_begin(&stats->seqlock);
 327         (*(u64 *)((u8 *)stats + stats_counter_off))++;
 328         write_seqcount_end(&stats->seqlock);
 329
 330         local_bh_enable();
 331 }
 332
 333 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
 334 {
 335         u16 csum_start, csum_offset;
 336         __wsum csum;
 337
 338         get_skb_csum_pointers(skb, &csum_start, &csum_offset);
 339         csum_start -= skb_headroom(skb);
 340
 341         skb_copy_bits(skb, 0, to, csum_start);
 342
 343         csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
 344                                       skb->len - csum_start, 0);
 345         *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
 346 }
 347
 348 static struct genl_family dp_packet_genl_family = {
 349         .id = GENL_ID_GENERATE,
 350         .hdrsize = sizeof(struct odp_header),
 351         .name = ODP_PACKET_FAMILY,
 352         .version = 1,
 353         .maxattr = ODP_PACKET_ATTR_MAX
 354 };
 355
 356 /* Generic Netlink multicast groups for upcalls.
 357  *
 358  * We really want three unique multicast groups per datapath, but we can't even
 359  * get one, because genl_register_mc_group() takes genl_lock, which is also
 360  * held during Generic Netlink message processing, so trying to acquire
 361  * multicast groups during ODP_DP_NEW processing deadlocks.  Instead, we
 362  * preallocate a few groups and use them round-robin for datapaths.  Collision
 363  * isn't fatal--multicast listeners should check that the family is the one
 364  * that they want and discard others--but it wastes time and memory to receive
 365  * unwanted messages.
 366  */
 367 #define PACKET_N_MC_GROUPS 16
 368 static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
 369
 370 static u32 packet_mc_group(struct datapath *dp, u8 cmd)
 371 {
 372         u32 idx;
 373         BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
 374
 375         idx = jhash_2words(dp->dp_ifindex, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
 376         return packet_mc_groups[idx].id;
 377 }
 378
 379 static int packet_register_mc_groups(void)
 380 {
 381         int i;
 382
 383         for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
 384                 struct genl_multicast_group *group = &packet_mc_groups[i];
 385                 int error;
 386
 387                 sprintf(group->name, "packet%d", i);
 388                 error = genl_register_mc_group(&dp_packet_genl_family, group);
 389                 if (error)
 390                         return error;
 391         }
 392         return 0;
 393 }
 394
 395 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
 396 {
 397         struct dp_stats_percpu *stats;
 398         int err;
 399
 400         WARN_ON_ONCE(skb_shared(skb));
 401
 402         forward_ip_summed(skb, true);
 403
 404         /* Break apart GSO packets into their component pieces.  Otherwise
 405          * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
 406         if (skb_is_gso(skb)) {
 407                 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 408
 409                 if (IS_ERR(nskb)) {
 410                         kfree_skb(skb);
 411                         err = PTR_ERR(nskb);
 412                         goto err;
 413                 }
 414                 consume_skb(skb);
 415                 skb = nskb;
 416         }
 417
 418         err = queue_userspace_packets(dp, skb, upcall_info);
 419         if (err)
 420                 goto err;
 421
 422         return 0;
 423
 424 err:
 425         local_bh_disable();
 426         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 427
 428         write_seqcount_begin(&stats->seqlock);
 429         stats->n_lost++;
 430         write_seqcount_end(&stats->seqlock);
 431
 432         local_bh_enable();
 433
 434         return err;
 435 }
 436
 437 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
 438  * 'upcall_info'.  There will be only one packet unless we broke up a GSO
 439  * packet.
 440  */
 441 static int queue_userspace_packets(struct datapath *dp, struct sk_buff *skb,
 442                                  const struct dp_upcall_info *upcall_info)
 443 {
 444         u32 group = packet_mc_group(dp, upcall_info->cmd);
 445         struct sk_buff *nskb;
 446         int err;
 447
 448         do {
 449                 struct odp_header *upcall;
 450                 struct sk_buff *user_skb; /* to be queued to userspace */
 451                 struct nlattr *nla;
 452                 unsigned int len;
 453
 454                 nskb = skb->next;
 455                 skb->next = NULL;
 456
 457                 err = vlan_deaccel_tag(skb);
 458                 if (unlikely(err))
 459                         goto err_kfree_skbs;
 460
 461                 if (nla_attr_size(skb->len) > USHRT_MAX) {
 462                         err = -EFBIG;
 463                         goto err_kfree_skbs;
 464                 }
 465
 466                 len = sizeof(struct odp_header);
 467                 len += nla_total_size(skb->len);
 468                 len += nla_total_size(FLOW_BUFSIZE);
 469                 if (upcall_info->userdata)
 470                         len += nla_total_size(8);
 471                 if (upcall_info->sample_pool)
 472                         len += nla_total_size(4);
 473                 if (upcall_info->actions_len)
 474                         len += nla_total_size(upcall_info->actions_len);
 475
 476                 user_skb = genlmsg_new(len, GFP_ATOMIC);
 477                 if (!user_skb) {
 478                         netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS);
 479                         err = -ENOMEM;
 480                         goto err_kfree_skbs;
 481                 }
 482
 483                 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
 484                 upcall->dp_ifindex = dp->dp_ifindex;
 485
 486                 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY);
 487                 flow_to_nlattrs(upcall_info->key, user_skb);
 488                 nla_nest_end(user_skb, nla);
 489
 490                 if (upcall_info->userdata)
 491                         nla_put_u64(user_skb, ODP_PACKET_ATTR_USERDATA, upcall_info->userdata);
 492                 if (upcall_info->sample_pool)
 493                         nla_put_u32(user_skb, ODP_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
 494                 if (upcall_info->actions_len) {
 495                         const struct nlattr *actions = upcall_info->actions;
 496                         u32 actions_len = upcall_info->actions_len;
 497
 498                         nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_ACTIONS);
 499                         memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
 500                         nla_nest_end(user_skb, nla);
 501                 }
 502
 503                 nla = __nla_reserve(user_skb, ODP_PACKET_ATTR_PACKET, skb->len);
 504                 if (skb->ip_summed == CHECKSUM_PARTIAL)
 505                         copy_and_csum_skb(skb, nla_data(nla));
 506                 else
 507                         skb_copy_bits(skb, 0, nla_data(nla), skb->len);
 508
 509                 err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC);
 510                 if (err)
 511                         goto err_kfree_skbs;
 512
 513                 consume_skb(skb);
 514                 skb = nskb;
 515         } while (skb);
 516         return 0;
 517
 518 err_kfree_skbs:
 519         kfree_skb(skb);
 520         while ((skb = nskb) != NULL) {
 521                 nskb = skb->next;
 522                 kfree_skb(skb);
 523         }
 524         return err;
 525 }
 526
 527 /* Called with genl_mutex. */
 528 static int flush_flows(int dp_ifindex)
 529 {
 530         struct tbl *old_table;
 531         struct tbl *new_table;
 532         struct datapath *dp;
 533
 534         dp = get_dp(dp_ifindex);
 535         if (!dp)
 536                 return -ENODEV;
 537
 538         old_table = get_table_protected(dp);
 539         new_table = tbl_create(TBL_MIN_BUCKETS);
 540         if (!new_table)
 541                 return -ENOMEM;
 542
 543         rcu_assign_pointer(dp->table, new_table);
 544
 545         tbl_deferred_destroy(old_table, flow_free_tbl);
 546
 547         return 0;
 548 }
 549
 550 static int validate_actions(const struct nlattr *attr)
 551 {
 552         const struct nlattr *a;
 553         int rem;
 554
 555         nla_for_each_nested(a, attr, rem) {
 556                 static const u32 action_lens[ODP_ACTION_ATTR_MAX + 1] = {
 557                         [ODP_ACTION_ATTR_OUTPUT] = 4,
 558                         [ODP_ACTION_ATTR_USERSPACE] = 8,
 559                         [ODP_ACTION_ATTR_SET_DL_TCI] = 2,
 560                         [ODP_ACTION_ATTR_STRIP_VLAN] = 0,
 561                         [ODP_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
 562                         [ODP_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
 563                         [ODP_ACTION_ATTR_SET_NW_SRC] = 4,
 564                         [ODP_ACTION_ATTR_SET_NW_DST] = 4,
 565                         [ODP_ACTION_ATTR_SET_NW_TOS] = 1,
 566                         [ODP_ACTION_ATTR_SET_TP_SRC] = 2,
 567                         [ODP_ACTION_ATTR_SET_TP_DST] = 2,
 568                         [ODP_ACTION_ATTR_SET_TUNNEL] = 8,
 569                         [ODP_ACTION_ATTR_SET_PRIORITY] = 4,
 570                         [ODP_ACTION_ATTR_POP_PRIORITY] = 0,
 571                 };
 572                 int type = nla_type(a);
 573
 574                 if (type > ODP_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
 575                         return -EINVAL;
 576
 577                 switch (type) {
 578                 case ODP_ACTION_ATTR_UNSPEC:
 579                         return -EINVAL;
 580
 581                 case ODP_ACTION_ATTR_USERSPACE:
 582                 case ODP_ACTION_ATTR_STRIP_VLAN:
 583                 case ODP_ACTION_ATTR_SET_DL_SRC:
 584                 case ODP_ACTION_ATTR_SET_DL_DST:
 585                 case ODP_ACTION_ATTR_SET_NW_SRC:
 586                 case ODP_ACTION_ATTR_SET_NW_DST:
 587                 case ODP_ACTION_ATTR_SET_TP_SRC:
 588                 case ODP_ACTION_ATTR_SET_TP_DST:
 589                 case ODP_ACTION_ATTR_SET_TUNNEL:
 590                 case ODP_ACTION_ATTR_SET_PRIORITY:
 591                 case ODP_ACTION_ATTR_POP_PRIORITY:
 592                         /* No validation needed. */
 593                         break;
 594
 595                 case ODP_ACTION_ATTR_OUTPUT:
 596                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 597                                 return -EINVAL;
 598                         break;
 599
 600                 case ODP_ACTION_ATTR_SET_DL_TCI:
 601                         if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
 602                                 return -EINVAL;
 603                         break;
 604
 605                 case ODP_ACTION_ATTR_SET_NW_TOS:
 606                         if (nla_get_u8(a) & INET_ECN_MASK)
 607                                 return -EINVAL;
 608                         break;
 609
 610                 default:
 611                         return -EOPNOTSUPP;
 612                 }
 613         }
 614
 615         if (rem > 0)
 616                 return -EINVAL;
 617
 618         return 0;
 619 }
 620 static void clear_stats(struct sw_flow *flow)
 621 {
 622         flow->used = 0;
 623         flow->tcp_flags = 0;
 624         flow->packet_count = 0;
 625         flow->byte_count = 0;
 626 }
 627
 628 /* Called with genl_mutex. */
 629 static int expand_table(struct datapath *dp)
 630 {
 631         struct tbl *old_table = get_table_protected(dp);
 632         struct tbl *new_table;
 633
 634         new_table = tbl_expand(old_table);
 635         if (IS_ERR(new_table))
 636                 return PTR_ERR(new_table);
 637
 638         rcu_assign_pointer(dp->table, new_table);
 639         tbl_deferred_destroy(old_table, NULL);
 640
 641         return 0;
 642 }
 643
 644 static int odp_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 645 {
 646         struct odp_header *odp_header = info->userhdr;
 647         struct nlattr **a = info->attrs;
 648         struct sw_flow_actions *acts;
 649         struct sk_buff *packet;
 650         struct sw_flow *flow;
 651         struct datapath *dp;
 652         struct ethhdr *eth;
 653         bool is_frag;
 654         int len;
 655         int err;
 656         int key_len;
 657
 658         err = -EINVAL;
 659         if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_KEY] ||
 660             !a[ODP_PACKET_ATTR_ACTIONS] ||
 661             nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN)
 662                 goto err;
 663
 664         err = validate_actions(a[ODP_PACKET_ATTR_ACTIONS]);
 665         if (err)
 666                 goto err;
 667
 668         len = nla_len(a[ODP_PACKET_ATTR_PACKET]);
 669         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 670         err = -ENOMEM;
 671         if (!packet)
 672                 goto err;
 673         skb_reserve(packet, NET_IP_ALIGN);
 674
 675         memcpy(__skb_put(packet, len), nla_data(a[ODP_PACKET_ATTR_PACKET]), len);
 676
 677         skb_reset_mac_header(packet);
 678         eth = eth_hdr(packet);
 679
 680         /* Normally, setting the skb 'protocol' field would be handled by a
 681          * call to eth_type_trans(), but it assumes there's a sending
 682          * device, which we may not have. */
 683         if (ntohs(eth->h_proto) >= 1536)
 684                 packet->protocol = eth->h_proto;
 685         else
 686                 packet->protocol = htons(ETH_P_802_2);
 687
 688         /* Build an sw_flow for sending this packet. */
 689         flow = flow_alloc();
 690         err = PTR_ERR(flow);
 691         if (IS_ERR(flow))
 692                 goto err_kfree_skb;
 693
 694         err = flow_extract(packet, -1, &flow->key, &key_len, &is_frag);
 695         if (err)
 696                 goto err_flow_put;
 697
 698         err = flow_metadata_from_nlattrs(&flow->key.eth.in_port,
 699                                          &flow->key.eth.tun_id,
 700                                          a[ODP_PACKET_ATTR_KEY]);
 701         if (err)
 702                 goto err_flow_put;
 703
 704         flow->tbl_node.hash = flow_hash(&flow->key, key_len);
 705
 706         acts = flow_actions_alloc(a[ODP_PACKET_ATTR_ACTIONS]);
 707         err = PTR_ERR(acts);
 708         if (IS_ERR(acts))
 709                 goto err_flow_put;
 710         rcu_assign_pointer(flow->sf_acts, acts);
 711
 712         OVS_CB(packet)->flow = flow;
 713
 714         rcu_read_lock();
 715         dp = get_dp(odp_header->dp_ifindex);
 716         err = -ENODEV;
 717         if (!dp)
 718                 goto err_unlock;
 719
 720         if (flow->key.eth.in_port < DP_MAX_PORTS)
 721                 OVS_CB(packet)->vport = get_vport_protected(dp,
 722                                                         flow->key.eth.in_port);
 723
 724         err = execute_actions(dp, packet);
 725         rcu_read_unlock();
 726
 727         flow_put(flow);
 728         return err;
 729
 730 err_unlock:
 731         rcu_read_unlock();
 732 err_flow_put:
 733         flow_put(flow);
 734 err_kfree_skb:
 735         kfree_skb(packet);
 736 err:
 737         return err;
 738 }
 739
 740 static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = {
 741         [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 742         [ODP_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 743         [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 744 };
 745
 746 static struct genl_ops dp_packet_genl_ops[] = {
 747         { .cmd = ODP_PACKET_CMD_EXECUTE,
 748           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 749           .policy = packet_policy,
 750           .doit = odp_packet_cmd_execute
 751         }
 752 };
 753
 754 static void get_dp_stats(struct datapath *dp, struct odp_stats *stats)
 755 {
 756         int i;
 757
 758         stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
 759         for_each_possible_cpu(i) {
 760                 const struct dp_stats_percpu *percpu_stats;
 761                 struct dp_stats_percpu local_stats;
 762                 unsigned seqcount;
 763
 764                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 765
 766                 do {
 767                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 768                         local_stats = *percpu_stats;
 769                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 770
 771                 stats->n_frags += local_stats.n_frags;
 772                 stats->n_hit += local_stats.n_hit;
 773                 stats->n_missed += local_stats.n_missed;
 774                 stats->n_lost += local_stats.n_lost;
 775         }
 776 }
 777
 778 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
 779  * Called with RTNL lock.
 780  */
 781 int dp_min_mtu(const struct datapath *dp)
 782 {
 783         struct vport *p;
 784         int mtu = 0;
 785
 786         ASSERT_RTNL();
 787
 788         list_for_each_entry (p, &dp->port_list, node) {
 789                 int dev_mtu;
 790
 791                 /* Skip any internal ports, since that's what we're trying to
 792                  * set. */
 793                 if (is_internal_vport(p))
 794                         continue;
 795
 796                 dev_mtu = vport_get_mtu(p);
 797                 if (!dev_mtu)
 798                         continue;
 799                 if (!mtu || dev_mtu < mtu)
 800                         mtu = dev_mtu;
 801         }
 802
 803         return mtu ? mtu : ETH_DATA_LEN;
 804 }
 805
 806 /* Sets the MTU of all datapath devices to the minimum of the ports
 807  * Called with RTNL lock.
 808  */
 809 void set_internal_devs_mtu(const struct datapath *dp)
 810 {
 811         struct vport *p;
 812         int mtu;
 813
 814         ASSERT_RTNL();
 815
 816         mtu = dp_min_mtu(dp);
 817
 818         list_for_each_entry (p, &dp->port_list, node) {
 819                 if (is_internal_vport(p))
 820                         vport_set_mtu(p, mtu);
 821         }
 822 }
 823
 824 static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = {
 825         [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 826         [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 827         [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 828 };
 829
 830 static struct genl_family dp_flow_genl_family = {
 831         .id = GENL_ID_GENERATE,
 832         .hdrsize = sizeof(struct odp_header),
 833         .name = ODP_FLOW_FAMILY,
 834         .version = 1,
 835         .maxattr = ODP_FLOW_ATTR_MAX
 836 };
 837
 838 static struct genl_multicast_group dp_flow_multicast_group = {
 839         .name = ODP_FLOW_MCGROUP
 840 };
 841
 842 /* Called with genl_lock. */
 843 static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 844                                   struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
 845 {
 846         const int skb_orig_len = skb->len;
 847         const struct sw_flow_actions *sf_acts;
 848         struct odp_flow_stats stats;
 849         struct odp_header *odp_header;
 850         struct nlattr *nla;
 851         unsigned long used;
 852         u8 tcp_flags;
 853         int err;
 854
 855         sf_acts = rcu_dereference_protected(flow->sf_acts,
 856                                             lockdep_genl_is_held());
 857
 858         odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 859         if (!odp_header)
 860                 return -EMSGSIZE;
 861
 862         odp_header->dp_ifindex = dp->dp_ifindex;
 863
 864         nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY);
 865         if (!nla)
 866                 goto nla_put_failure;
 867         err = flow_to_nlattrs(&flow->key, skb);
 868         if (err)
 869                 goto error;
 870         nla_nest_end(skb, nla);
 871
 872         spin_lock_bh(&flow->lock);
 873         used = flow->used;
 874         stats.n_packets = flow->packet_count;
 875         stats.n_bytes = flow->byte_count;
 876         tcp_flags = flow->tcp_flags;
 877         spin_unlock_bh(&flow->lock);
 878
 879         if (used)
 880                 NLA_PUT_U64(skb, ODP_FLOW_ATTR_USED, flow_used_time(used));
 881
 882         if (stats.n_packets)
 883                 NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats);
 884
 885         if (tcp_flags)
 886                 NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 887
 888         /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 889          * this is the first flow to be dumped into 'skb'.  This is unusual for
 890          * Netlink but individual action lists can be longer than
 891          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 892          * The userspace caller can always fetch the actions separately if it
 893          * really wants them.  (Most userspace callers in fact don't care.)
 894          *
 895          * This can only fail for dump operations because the skb is always
 896          * properly sized for single flows.
 897          */
 898         err = nla_put(skb, ODP_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 899                       sf_acts->actions);
 900         if (err < 0 && skb_orig_len)
 901                 goto error;
 902
 903         return genlmsg_end(skb, odp_header);
 904
 905 nla_put_failure:
 906         err = -EMSGSIZE;
 907 error:
 908         genlmsg_cancel(skb, odp_header);
 909         return err;
 910 }
 911
 912 static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow)
 913 {
 914         const struct sw_flow_actions *sf_acts;
 915         int len;
 916
 917         sf_acts = rcu_dereference_protected(flow->sf_acts,
 918                                             lockdep_genl_is_held());
 919
 920         len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */
 921         len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */
 922         len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */
 923         len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */
 924         len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */
 925         return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL);
 926 }
 927
 928 static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
 929                                                u32 pid, u32 seq, u8 cmd)
 930 {
 931         struct sk_buff *skb;
 932         int retval;
 933
 934         skb = odp_flow_cmd_alloc_info(flow);
 935         if (!skb)
 936                 return ERR_PTR(-ENOMEM);
 937
 938         retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 939         BUG_ON(retval < 0);
 940         return skb;
 941 }
 942
 943 static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 944 {
 945         struct nlattr **a = info->attrs;
 946         struct odp_header *odp_header = info->userhdr;
 947         struct tbl_node *flow_node;
 948         struct sw_flow_key key;
 949         struct sw_flow *flow;
 950         struct sk_buff *reply;
 951         struct datapath *dp;
 952         struct tbl *table;
 953         u32 hash;
 954         int error;
 955         int key_len;
 956
 957         /* Extract key. */
 958         error = -EINVAL;
 959         if (!a[ODP_FLOW_ATTR_KEY])
 960                 goto error;
 961         error = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
 962         if (error)
 963                 goto error;
 964
 965         /* Validate actions. */
 966         if (a[ODP_FLOW_ATTR_ACTIONS]) {
 967                 error = validate_actions(a[ODP_FLOW_ATTR_ACTIONS]);
 968                 if (error)
 969                         goto error;
 970         } else if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW) {
 971                 error = -EINVAL;
 972                 goto error;
 973         }
 974
 975         dp = get_dp(odp_header->dp_ifindex);
 976         error = -ENODEV;
 977         if (!dp)
 978                 goto error;
 979
 980         hash = flow_hash(&key, key_len);
 981         table = get_table_protected(dp);
 982         flow_node = tbl_lookup(table, &key, key_len, hash, flow_cmp);
 983         if (!flow_node) {
 984                 struct sw_flow_actions *acts;
 985
 986                 /* Bail out if we're not allowed to create a new flow. */
 987                 error = -ENOENT;
 988                 if (info->genlhdr->cmd == ODP_FLOW_CMD_SET)
 989                         goto error;
 990
 991                 /* Expand table, if necessary, to make room. */
 992                 if (tbl_count(table) >= tbl_n_buckets(table)) {
 993                         error = expand_table(dp);
 994                         if (error)
 995                                 goto error;
 996                         table = get_table_protected(dp);
 997                 }
 998
 999                 /* Allocate flow. */
1000                 flow = flow_alloc();
1001                 if (IS_ERR(flow)) {
1002                         error = PTR_ERR(flow);
1003                         goto error;
1004                 }
1005                 flow->key = key;
1006                 clear_stats(flow);
1007
1008                 /* Obtain actions. */
1009                 acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1010                 error = PTR_ERR(acts);
1011                 if (IS_ERR(acts))
1012                         goto error_free_flow;
1013                 rcu_assign_pointer(flow->sf_acts, acts);
1014
1015                 /* Put flow in bucket. */
1016                 error = tbl_insert(table, &flow->tbl_node, hash);
1017                 if (error)
1018                         goto error_free_flow;
1019
1020                 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1021                                                 info->snd_seq, ODP_FLOW_CMD_NEW);
1022         } else {
1023                 /* We found a matching flow. */
1024                 struct sw_flow_actions *old_acts;
1025
1026                 /* Bail out if we're not allowed to modify an existing flow.
1027                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1028                  * because Generic Netlink treats the latter as a dump
1029                  * request.  We also accept NLM_F_EXCL in case that bug ever
1030                  * gets fixed.
1031                  */
1032                 error = -EEXIST;
1033                 if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW &&
1034                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1035                         goto error;
1036
1037                 /* Update actions. */
1038                 flow = flow_cast(flow_node);
1039                 old_acts = rcu_dereference_protected(flow->sf_acts,
1040                                                      lockdep_genl_is_held());
1041                 if (a[ODP_FLOW_ATTR_ACTIONS] &&
1042                     (old_acts->actions_len != nla_len(a[ODP_FLOW_ATTR_ACTIONS]) ||
1043                      memcmp(old_acts->actions, nla_data(a[ODP_FLOW_ATTR_ACTIONS]),
1044                             old_acts->actions_len))) {
1045                         struct sw_flow_actions *new_acts;
1046
1047                         new_acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1048                         error = PTR_ERR(new_acts);
1049                         if (IS_ERR(new_acts))
1050                                 goto error;
1051
1052                         rcu_assign_pointer(flow->sf_acts, new_acts);
1053                         flow_deferred_free_acts(old_acts);
1054                 }
1055
1056                 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1057                                                 info->snd_seq, ODP_FLOW_CMD_NEW);
1058
1059                 /* Clear stats. */
1060                 if (a[ODP_FLOW_ATTR_CLEAR]) {
1061                         spin_lock_bh(&flow->lock);
1062                         clear_stats(flow);
1063                         spin_unlock_bh(&flow->lock);
1064                 }
1065         }
1066
1067         if (!IS_ERR(reply))
1068                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1069                             dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1070         else
1071                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1072                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1073         return 0;
1074
1075 error_free_flow:
1076         flow_put(flow);
1077 error:
1078         return error;
1079 }
1080
1081 static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1082 {
1083         struct nlattr **a = info->attrs;
1084         struct odp_header *odp_header = info->userhdr;
1085         struct sw_flow_key key;
1086         struct tbl_node *flow_node;
1087         struct sk_buff *reply;
1088         struct sw_flow *flow;
1089         struct datapath *dp;
1090         struct tbl *table;
1091         int err;
1092         int key_len;
1093
1094         if (!a[ODP_FLOW_ATTR_KEY])
1095                 return -EINVAL;
1096         err = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
1097         if (err)
1098                 return err;
1099
1100         dp = get_dp(odp_header->dp_ifindex);
1101         if (!dp)
1102                 return -ENODEV;
1103
1104         table = get_table_protected(dp);
1105         flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1106                                flow_cmp);
1107         if (!flow_node)
1108                 return -ENOENT;
1109
1110         flow = flow_cast(flow_node);
1111         reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW);
1112         if (IS_ERR(reply))
1113                 return PTR_ERR(reply);
1114
1115         return genlmsg_reply(reply, info);
1116 }
1117
1118 static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1119 {
1120         struct nlattr **a = info->attrs;
1121         struct odp_header *odp_header = info->userhdr;
1122         struct sw_flow_key key;
1123         struct tbl_node *flow_node;
1124         struct sk_buff *reply;
1125         struct sw_flow *flow;
1126         struct datapath *dp;
1127         struct tbl *table;
1128         int err;
1129         int key_len;
1130
1131         if (!a[ODP_FLOW_ATTR_KEY])
1132                 return flush_flows(odp_header->dp_ifindex);
1133         err = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
1134         if (err)
1135                 return err;
1136
1137         dp = get_dp(odp_header->dp_ifindex);
1138         if (!dp)
1139                 return -ENODEV;
1140
1141         table = get_table_protected(dp);
1142         flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1143                                flow_cmp);
1144         if (!flow_node)
1145                 return -ENOENT;
1146         flow = flow_cast(flow_node);
1147
1148         reply = odp_flow_cmd_alloc_info(flow);
1149         if (!reply)
1150                 return -ENOMEM;
1151
1152         err = tbl_remove(table, flow_node);
1153         if (err) {
1154                 kfree_skb(reply);
1155                 return err;
1156         }
1157
1158         err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1159                                      info->snd_seq, 0, ODP_FLOW_CMD_DEL);
1160         BUG_ON(err < 0);
1161
1162         flow_deferred_free(flow);
1163
1164         genl_notify(reply, genl_info_net(info), info->snd_pid,
1165                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1166         return 0;
1167 }
1168
1169 static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1170 {
1171         struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1172         struct datapath *dp;
1173
1174         dp = get_dp(odp_header->dp_ifindex);
1175         if (!dp)
1176                 return -ENODEV;
1177
1178         for (;;) {
1179                 struct tbl_node *flow_node;
1180                 struct sw_flow *flow;
1181                 u32 bucket, obj;
1182
1183                 bucket = cb->args[0];
1184                 obj = cb->args[1];
1185                 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1186                 if (!flow_node)
1187                         break;
1188
1189                 flow = flow_cast(flow_node);
1190                 if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1191                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1192                                            ODP_FLOW_CMD_NEW) < 0)
1193                         break;
1194
1195                 cb->args[0] = bucket;
1196                 cb->args[1] = obj;
1197         }
1198         return skb->len;
1199 }
1200
1201 static struct genl_ops dp_flow_genl_ops[] = {
1202         { .cmd = ODP_FLOW_CMD_NEW,
1203           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1204           .policy = flow_policy,
1205           .doit = odp_flow_cmd_new_or_set
1206         },
1207         { .cmd = ODP_FLOW_CMD_DEL,
1208           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1209           .policy = flow_policy,
1210           .doit = odp_flow_cmd_del
1211         },
1212         { .cmd = ODP_FLOW_CMD_GET,
1213           .flags = 0,               /* OK for unprivileged users. */
1214           .policy = flow_policy,
1215           .doit = odp_flow_cmd_get,
1216           .dumpit = odp_flow_cmd_dump
1217         },
1218         { .cmd = ODP_FLOW_CMD_SET,
1219           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1220           .policy = flow_policy,
1221           .doit = odp_flow_cmd_new_or_set,
1222         },
1223 };
1224
1225 static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
1226 #ifdef HAVE_NLA_NUL_STRING
1227         [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1228 #endif
1229         [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1230         [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1231 };
1232
1233 static struct genl_family dp_datapath_genl_family = {
1234         .id = GENL_ID_GENERATE,
1235         .hdrsize = sizeof(struct odp_header),
1236         .name = ODP_DATAPATH_FAMILY,
1237         .version = 1,
1238         .maxattr = ODP_DP_ATTR_MAX
1239 };
1240
1241 static struct genl_multicast_group dp_datapath_multicast_group = {
1242         .name = ODP_DATAPATH_MCGROUP
1243 };
1244
1245 static int odp_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1246                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1247 {
1248         struct odp_header *odp_header;
1249         struct nlattr *nla;
1250         int err;
1251
1252         odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1253                                    flags, cmd);
1254         if (!odp_header)
1255                 goto error;
1256
1257         odp_header->dp_ifindex = dp->dp_ifindex;
1258
1259         rcu_read_lock();
1260         err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
1261         rcu_read_unlock();
1262         if (err)
1263                 goto nla_put_failure;
1264
1265         nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats));
1266         if (!nla)
1267                 goto nla_put_failure;
1268         get_dp_stats(dp, nla_data(nla));
1269
1270         NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS,
1271                     dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO);
1272
1273         if (dp->sflow_probability)
1274                 NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability);
1275
1276         nla = nla_nest_start(skb, ODP_DP_ATTR_MCGROUPS);
1277         if (!nla)
1278                 goto nla_put_failure;
1279         NLA_PUT_U32(skb, ODP_PACKET_CMD_MISS, packet_mc_group(dp, ODP_PACKET_CMD_MISS));
1280         NLA_PUT_U32(skb, ODP_PACKET_CMD_ACTION, packet_mc_group(dp, ODP_PACKET_CMD_ACTION));
1281         NLA_PUT_U32(skb, ODP_PACKET_CMD_SAMPLE, packet_mc_group(dp, ODP_PACKET_CMD_SAMPLE));
1282         nla_nest_end(skb, nla);
1283
1284         return genlmsg_end(skb, odp_header);
1285
1286 nla_put_failure:
1287         genlmsg_cancel(skb, odp_header);
1288 error:
1289         return -EMSGSIZE;
1290 }
1291
1292 static struct sk_buff *odp_dp_cmd_build_info(struct datapath *dp, u32 pid,
1293                                              u32 seq, u8 cmd)
1294 {
1295         struct sk_buff *skb;
1296         int retval;
1297
1298         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1299         if (!skb)
1300                 return ERR_PTR(-ENOMEM);
1301
1302         retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1303         if (retval < 0) {
1304                 kfree_skb(skb);
1305                 return ERR_PTR(retval);
1306         }
1307         return skb;
1308 }
1309
1310 static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1311 {
1312         if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1313                 u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1314
1315                 if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP)
1316                         return -EINVAL;
1317         }
1318
1319         return CHECK_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
1320 }
1321
1322 /* Called with genl_mutex and optionally with RTNL lock also. */
1323 static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1324 {
1325         struct datapath *dp;
1326
1327         if (!a[ODP_DP_ATTR_NAME])
1328                 dp = get_dp(odp_header->dp_ifindex);
1329         else {
1330                 struct vport *vport;
1331
1332                 rcu_read_lock();
1333                 vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
1334                 dp = vport && vport->port_no == ODPP_LOCAL ? vport->dp : NULL;
1335                 rcu_read_unlock();
1336         }
1337         return dp ? dp : ERR_PTR(-ENODEV);
1338 }
1339
1340 /* Called with genl_mutex. */
1341 static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1342 {
1343         if (a[ODP_DP_ATTR_IPV4_FRAGS])
1344                 dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP;
1345         if (a[ODP_DP_ATTR_SAMPLING])
1346                 dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]);
1347 }
1348
1349 static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1350 {
1351         struct nlattr **a = info->attrs;
1352         struct vport_parms parms;
1353         struct sk_buff *reply;
1354         struct datapath *dp;
1355         struct vport *vport;
1356         int err;
1357
1358         err = -EINVAL;
1359         if (!a[ODP_DP_ATTR_NAME])
1360                 goto err;
1361
1362         err = odp_dp_cmd_validate(a);
1363         if (err)
1364                 goto err;
1365
1366         rtnl_lock();
1367         err = -ENODEV;
1368         if (!try_module_get(THIS_MODULE))
1369                 goto err_unlock_rtnl;
1370
1371         err = -ENOMEM;
1372         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1373         if (dp == NULL)
1374                 goto err_put_module;
1375         INIT_LIST_HEAD(&dp->port_list);
1376
1377         /* Initialize kobject for bridge.  This will be added as
1378          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1379         dp->ifobj.kset = NULL;
1380         kobject_init(&dp->ifobj, &dp_ktype);
1381
1382         /* Allocate table. */
1383         err = -ENOMEM;
1384         rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1385         if (!dp->table)
1386                 goto err_free_dp;
1387
1388         /* Set up our datapath device. */
1389         parms.name = nla_data(a[ODP_DP_ATTR_NAME]);
1390         parms.type = ODP_VPORT_TYPE_INTERNAL;
1391         parms.options = NULL;
1392         parms.dp = dp;
1393         parms.port_no = ODPP_LOCAL;
1394         vport = new_vport(&parms);
1395         if (IS_ERR(vport)) {
1396                 err = PTR_ERR(vport);
1397                 if (err == -EBUSY)
1398                         err = -EEXIST;
1399
1400                 goto err_destroy_table;
1401         }
1402         dp->dp_ifindex = vport_get_ifindex(vport);
1403
1404         dp->drop_frags = 0;
1405         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1406         if (!dp->stats_percpu) {
1407                 err = -ENOMEM;
1408                 goto err_destroy_local_port;
1409         }
1410
1411         change_datapath(dp, a);
1412
1413         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1414         err = PTR_ERR(reply);
1415         if (IS_ERR(reply))
1416                 goto err_destroy_local_port;
1417
1418         list_add_tail(&dp->list_node, &dps);
1419         dp_sysfs_add_dp(dp);
1420
1421         rtnl_unlock();
1422
1423         genl_notify(reply, genl_info_net(info), info->snd_pid,
1424                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1425         return 0;
1426
1427 err_destroy_local_port:
1428         dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1429 err_destroy_table:
1430         tbl_destroy(get_table_protected(dp), NULL);
1431 err_free_dp:
1432         kfree(dp);
1433 err_put_module:
1434         module_put(THIS_MODULE);
1435 err_unlock_rtnl:
1436         rtnl_unlock();
1437 err:
1438         return err;
1439 }
1440
1441 static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1442 {
1443         struct vport *vport, *next_vport;
1444         struct sk_buff *reply;
1445         struct datapath *dp;
1446         int err;
1447
1448         err = odp_dp_cmd_validate(info->attrs);
1449         if (err)
1450                 goto exit;
1451
1452         rtnl_lock();
1453         dp = lookup_datapath(info->userhdr, info->attrs);
1454         err = PTR_ERR(dp);
1455         if (IS_ERR(dp))
1456                 goto exit_unlock;
1457
1458         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_DEL);
1459         err = PTR_ERR(reply);
1460         if (IS_ERR(reply))
1461                 goto exit_unlock;
1462
1463         list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1464                 if (vport->port_no != ODPP_LOCAL)
1465                         dp_detach_port(vport);
1466
1467         dp_sysfs_del_dp(dp);
1468         list_del(&dp->list_node);
1469         dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1470
1471         /* rtnl_unlock() will wait until all the references to devices that
1472          * are pending unregistration have been dropped.  We do it here to
1473          * ensure that any internal devices (which contain DP pointers) are
1474          * fully destroyed before freeing the datapath.
1475          */
1476         rtnl_unlock();
1477
1478         call_rcu(&dp->rcu, destroy_dp_rcu);
1479         module_put(THIS_MODULE);
1480
1481         genl_notify(reply, genl_info_net(info), info->snd_pid,
1482                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1483
1484         return 0;
1485
1486 exit_unlock:
1487         rtnl_unlock();
1488 exit:
1489         return err;
1490 }
1491
1492 static int odp_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1493 {
1494         struct sk_buff *reply;
1495         struct datapath *dp;
1496         int err;
1497
1498         err = odp_dp_cmd_validate(info->attrs);
1499         if (err)
1500                 return err;
1501
1502         dp = lookup_datapath(info->userhdr, info->attrs);
1503         if (IS_ERR(dp))
1504                 return PTR_ERR(dp);
1505
1506         change_datapath(dp, info->attrs);
1507
1508         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1509         if (IS_ERR(reply)) {
1510                 err = PTR_ERR(reply);
1511                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1512                                 dp_datapath_multicast_group.id, err);
1513                 return 0;
1514         }
1515
1516         genl_notify(reply, genl_info_net(info), info->snd_pid,
1517                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1518         return 0;
1519 }
1520
1521 static int odp_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1522 {
1523         struct sk_buff *reply;
1524         struct datapath *dp;
1525         int err;
1526
1527         err = odp_dp_cmd_validate(info->attrs);
1528         if (err)
1529                 return err;
1530
1531         dp = lookup_datapath(info->userhdr, info->attrs);
1532         if (IS_ERR(dp))
1533                 return PTR_ERR(dp);
1534
1535         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1536         if (IS_ERR(reply))
1537                 return PTR_ERR(reply);
1538
1539         return genlmsg_reply(reply, info);
1540 }
1541
1542 static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1543 {
1544         struct datapath *dp;
1545         int skip = cb->args[0];
1546         int i = 0;
1547
1548         list_for_each_entry (dp, &dps, list_node) {
1549                 if (i < skip)
1550                         continue;
1551                 if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1552                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1553                                          ODP_DP_CMD_NEW) < 0)
1554                         break;
1555                 i++;
1556         }
1557
1558         cb->args[0] = i;
1559
1560         return skb->len;
1561 }
1562
1563 static struct genl_ops dp_datapath_genl_ops[] = {
1564         { .cmd = ODP_DP_CMD_NEW,
1565           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1566           .policy = datapath_policy,
1567           .doit = odp_dp_cmd_new
1568         },
1569         { .cmd = ODP_DP_CMD_DEL,
1570           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1571           .policy = datapath_policy,
1572           .doit = odp_dp_cmd_del
1573         },
1574         { .cmd = ODP_DP_CMD_GET,
1575           .flags = 0,               /* OK for unprivileged users. */
1576           .policy = datapath_policy,
1577           .doit = odp_dp_cmd_get,
1578           .dumpit = odp_dp_cmd_dump
1579         },
1580         { .cmd = ODP_DP_CMD_SET,
1581           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1582           .policy = datapath_policy,
1583           .doit = odp_dp_cmd_set,
1584         },
1585 };
1586
1587 static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = {
1588 #ifdef HAVE_NLA_NUL_STRING
1589         [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1590         [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1591         [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1592         [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1593         [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1594 #else
1595         [ODP_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1596         [ODP_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1597 #endif
1598         [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1599         [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1600 };
1601
1602 static struct genl_family dp_vport_genl_family = {
1603         .id = GENL_ID_GENERATE,
1604         .hdrsize = sizeof(struct odp_header),
1605         .name = ODP_VPORT_FAMILY,
1606         .version = 1,
1607         .maxattr = ODP_VPORT_ATTR_MAX
1608 };
1609
1610 static struct genl_multicast_group dp_vport_multicast_group = {
1611         .name = ODP_VPORT_MCGROUP
1612 };
1613
1614 /* Called with RTNL lock or RCU read lock. */
1615 static int odp_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1616                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1617 {
1618         struct odp_header *odp_header;
1619         struct nlattr *nla;
1620         int ifindex, iflink;
1621         int mtu;
1622         int err;
1623
1624         odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1625                                  flags, cmd);
1626         if (!odp_header)
1627                 return -EMSGSIZE;
1628
1629         odp_header->dp_ifindex = vport->dp->dp_ifindex;
1630
1631         NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1632         NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport));
1633         NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport));
1634
1635         nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1636         if (!nla)
1637                 goto nla_put_failure;
1638         if (vport_get_stats(vport, nla_data(nla)))
1639                 __skb_trim(skb, skb->len - nla->nla_len);
1640
1641         NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1642
1643         mtu = vport_get_mtu(vport);
1644         if (mtu)
1645                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, mtu);
1646
1647         err = vport_get_options(vport, skb);
1648         if (err == -EMSGSIZE)
1649                 goto error;
1650
1651         ifindex = vport_get_ifindex(vport);
1652         if (ifindex > 0)
1653                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex);
1654
1655         iflink = vport_get_iflink(vport);
1656         if (iflink > 0)
1657                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink);
1658
1659         return genlmsg_end(skb, odp_header);
1660
1661 nla_put_failure:
1662         err = -EMSGSIZE;
1663 error:
1664         genlmsg_cancel(skb, odp_header);
1665         return err;
1666 }
1667
1668 /* Called with RTNL lock or RCU read lock. */
1669 static struct sk_buff *odp_vport_cmd_build_info(struct vport *vport, u32 pid,
1670                                                 u32 seq, u8 cmd)
1671 {
1672         struct sk_buff *skb;
1673         int retval;
1674
1675         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1676         if (!skb)
1677                 return ERR_PTR(-ENOMEM);
1678
1679         retval = odp_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1680         if (retval < 0) {
1681                 kfree_skb(skb);
1682                 return ERR_PTR(retval);
1683         }
1684         return skb;
1685 }
1686
1687 static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1688 {
1689         return CHECK_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1690 }
1691
1692 /* Called with RTNL lock or RCU read lock. */
1693 static struct vport *lookup_vport(struct odp_header *odp_header,
1694                                   struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1695 {
1696         struct datapath *dp;
1697         struct vport *vport;
1698
1699         if (a[ODP_VPORT_ATTR_NAME]) {
1700                 vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME]));
1701                 if (!vport)
1702                         return ERR_PTR(-ENODEV);
1703                 return vport;
1704         } else if (a[ODP_VPORT_ATTR_PORT_NO]) {
1705                 u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1706
1707                 if (port_no >= DP_MAX_PORTS)
1708                         return ERR_PTR(-EFBIG);
1709
1710                 dp = get_dp(odp_header->dp_ifindex);
1711                 if (!dp)
1712                         return ERR_PTR(-ENODEV);
1713
1714                 vport = get_vport_protected(dp, port_no);
1715                 if (!vport)
1716                         return ERR_PTR(-ENOENT);
1717                 return vport;
1718         } else
1719                 return ERR_PTR(-EINVAL);
1720 }
1721
1722 /* Called with RTNL lock. */
1723 static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1724 {
1725         int err = 0;
1726         if (a[ODP_VPORT_ATTR_STATS])
1727                 err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS]));
1728         if (!err && a[ODP_VPORT_ATTR_ADDRESS])
1729                 err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS]));
1730         if (!err && a[ODP_VPORT_ATTR_MTU])
1731                 err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU]));
1732         return err;
1733 }
1734
1735 static int odp_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1736 {
1737         struct nlattr **a = info->attrs;
1738         struct odp_header *odp_header = info->userhdr;
1739         struct vport_parms parms;
1740         struct sk_buff *reply;
1741         struct vport *vport;
1742         struct datapath *dp;
1743         u32 port_no;
1744         int err;
1745
1746         err = -EINVAL;
1747         if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE])
1748                 goto exit;
1749
1750         err = odp_vport_cmd_validate(a);
1751         if (err)
1752                 goto exit;
1753
1754         rtnl_lock();
1755         dp = get_dp(odp_header->dp_ifindex);
1756         err = -ENODEV;
1757         if (!dp)
1758                 goto exit_unlock;
1759
1760         if (a[ODP_VPORT_ATTR_PORT_NO]) {
1761                 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1762
1763                 err = -EFBIG;
1764                 if (port_no >= DP_MAX_PORTS)
1765                         goto exit_unlock;
1766
1767                 vport = get_vport_protected(dp, port_no);
1768                 err = -EBUSY;
1769                 if (vport)
1770                         goto exit_unlock;
1771         } else {
1772                 for (port_no = 1; ; port_no++) {
1773                         if (port_no >= DP_MAX_PORTS) {
1774                                 err = -EFBIG;
1775                                 goto exit_unlock;
1776                         }
1777                         vport = get_vport_protected(dp, port_no);
1778                         if (!vport)
1779                                 break;
1780                 }
1781         }
1782
1783         parms.name = nla_data(a[ODP_VPORT_ATTR_NAME]);
1784         parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1785         parms.options = a[ODP_VPORT_ATTR_OPTIONS];
1786         parms.dp = dp;
1787         parms.port_no = port_no;
1788
1789         vport = new_vport(&parms);
1790         err = PTR_ERR(vport);
1791         if (IS_ERR(vport))
1792                 goto exit_unlock;
1793
1794         set_internal_devs_mtu(dp);
1795         dp_sysfs_add_if(vport);
1796
1797         err = change_vport(vport, a);
1798         if (!err) {
1799                 reply = odp_vport_cmd_build_info(vport, info->snd_pid,
1800                                                  info->snd_seq, ODP_VPORT_CMD_NEW);
1801                 if (IS_ERR(reply))
1802                         err = PTR_ERR(reply);
1803         }
1804         if (err) {
1805                 dp_detach_port(vport);
1806                 goto exit_unlock;
1807         }
1808         genl_notify(reply, genl_info_net(info), info->snd_pid,
1809                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1810
1811
1812 exit_unlock:
1813         rtnl_unlock();
1814 exit:
1815         return err;
1816 }
1817
1818 static int odp_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1819 {
1820         struct nlattr **a = info->attrs;
1821         struct sk_buff *reply;
1822         struct vport *vport;
1823         int err;
1824
1825         err = odp_vport_cmd_validate(a);
1826         if (err)
1827                 goto exit;
1828
1829         rtnl_lock();
1830         vport = lookup_vport(info->userhdr, a);
1831         err = PTR_ERR(vport);
1832         if (IS_ERR(vport))
1833                 goto exit_unlock;
1834
1835         err = 0;
1836         if (a[ODP_VPORT_ATTR_OPTIONS])
1837                 err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]);
1838         if (!err)
1839                 err = change_vport(vport, a);
1840
1841         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1842                                          ODP_VPORT_CMD_NEW);
1843         if (IS_ERR(reply)) {
1844                 err = PTR_ERR(reply);
1845                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1846                                 dp_vport_multicast_group.id, err);
1847                 return 0;
1848         }
1849
1850         genl_notify(reply, genl_info_net(info), info->snd_pid,
1851                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1852
1853 exit_unlock:
1854         rtnl_unlock();
1855 exit:
1856         return err;
1857 }
1858
1859 static int odp_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1860 {
1861         struct nlattr **a = info->attrs;
1862         struct sk_buff *reply;
1863         struct vport *vport;
1864         int err;
1865
1866         err = odp_vport_cmd_validate(a);
1867         if (err)
1868                 goto exit;
1869
1870         rtnl_lock();
1871         vport = lookup_vport(info->userhdr, a);
1872         err = PTR_ERR(vport);
1873         if (IS_ERR(vport))
1874                 goto exit_unlock;
1875
1876         if (vport->port_no == ODPP_LOCAL) {
1877                 err = -EINVAL;
1878                 goto exit_unlock;
1879         }
1880
1881         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1882                                          ODP_VPORT_CMD_DEL);
1883         err = PTR_ERR(reply);
1884         if (IS_ERR(reply))
1885                 goto exit_unlock;
1886
1887         err = dp_detach_port(vport);
1888
1889         genl_notify(reply, genl_info_net(info), info->snd_pid,
1890                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1891
1892 exit_unlock:
1893         rtnl_unlock();
1894 exit:
1895         return err;
1896 }
1897
1898 static int odp_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1899 {
1900         struct nlattr **a = info->attrs;
1901         struct odp_header *odp_header = info->userhdr;
1902         struct sk_buff *reply;
1903         struct vport *vport;
1904         int err;
1905
1906         err = odp_vport_cmd_validate(a);
1907         if (err)
1908                 goto exit;
1909
1910         rcu_read_lock();
1911         vport = lookup_vport(odp_header, a);
1912         err = PTR_ERR(vport);
1913         if (IS_ERR(vport))
1914                 goto exit_unlock;
1915
1916         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1917                                          ODP_VPORT_CMD_NEW);
1918         err = PTR_ERR(reply);
1919         if (IS_ERR(reply))
1920                 goto exit_unlock;
1921
1922         rcu_read_unlock();
1923
1924         return genlmsg_reply(reply, info);
1925
1926 exit_unlock:
1927         rcu_read_unlock();
1928 exit:
1929         return err;
1930 }
1931
1932 static int odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1933 {
1934         struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1935         struct datapath *dp;
1936         u32 port_no;
1937         int retval;
1938
1939         dp = get_dp(odp_header->dp_ifindex);
1940         if (!dp)
1941                 return -ENODEV;
1942
1943         rcu_read_lock();
1944         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1945                 struct vport *vport;
1946
1947                 vport = get_vport_protected(dp, port_no);
1948                 if (!vport)
1949                         continue;
1950
1951                 if (odp_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1952                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1953                                             ODP_VPORT_CMD_NEW) < 0)
1954                         break;
1955         }
1956         rcu_read_unlock();
1957
1958         cb->args[0] = port_no;
1959         retval = skb->len;
1960
1961         return retval;
1962 }
1963
1964 static struct genl_ops dp_vport_genl_ops[] = {
1965         { .cmd = ODP_VPORT_CMD_NEW,
1966           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1967           .policy = vport_policy,
1968           .doit = odp_vport_cmd_new
1969         },
1970         { .cmd = ODP_VPORT_CMD_DEL,
1971           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1972           .policy = vport_policy,
1973           .doit = odp_vport_cmd_del
1974         },
1975         { .cmd = ODP_VPORT_CMD_GET,
1976           .flags = 0,               /* OK for unprivileged users. */
1977           .policy = vport_policy,
1978           .doit = odp_vport_cmd_get,
1979           .dumpit = odp_vport_cmd_dump
1980         },
1981         { .cmd = ODP_VPORT_CMD_SET,
1982           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1983           .policy = vport_policy,
1984           .doit = odp_vport_cmd_set,
1985         },
1986 };
1987
1988 struct genl_family_and_ops {
1989         struct genl_family *family;
1990         struct genl_ops *ops;
1991         int n_ops;
1992         struct genl_multicast_group *group;
1993 };
1994
1995 static const struct genl_family_and_ops dp_genl_families[] = {
1996         { &dp_datapath_genl_family,
1997           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1998           &dp_datapath_multicast_group },
1999         { &dp_vport_genl_family,
2000           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2001           &dp_vport_multicast_group },
2002         { &dp_flow_genl_family,
2003           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2004           &dp_flow_multicast_group },
2005         { &dp_packet_genl_family,
2006           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2007           NULL },
2008 };
2009
2010 static void dp_unregister_genl(int n_families)
2011 {
2012         int i;
2013
2014         for (i = 0; i < n_families; i++) {
2015                 genl_unregister_family(dp_genl_families[i].family);
2016         }
2017 }
2018
2019 static int dp_register_genl(void)
2020 {
2021         int n_registered;
2022         int err;
2023         int i;
2024
2025         n_registered = 0;
2026         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2027                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2028
2029                 err = genl_register_family_with_ops(f->family, f->ops,
2030                                                     f->n_ops);
2031                 if (err)
2032                         goto error;
2033                 n_registered++;
2034
2035                 if (f->group) {
2036                         err = genl_register_mc_group(f->family, f->group);
2037                         if (err)
2038                                 goto error;
2039                 }
2040         }
2041
2042         err = packet_register_mc_groups();
2043         if (err)
2044                 goto error;
2045         return 0;
2046
2047 error:
2048         dp_unregister_genl(n_registered);
2049         return err;
2050 }
2051
2052 static int __init dp_init(void)
2053 {
2054         struct sk_buff *dummy_skb;
2055         int err;
2056
2057         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2058
2059         printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2060
2061         err = flow_init();
2062         if (err)
2063                 goto error;
2064
2065         err = vport_init();
2066         if (err)
2067                 goto error_flow_exit;
2068
2069         err = register_netdevice_notifier(&dp_device_notifier);
2070         if (err)
2071                 goto error_vport_exit;
2072
2073         err = dp_register_genl();
2074         if (err < 0)
2075                 goto error_unreg_notifier;
2076
2077         return 0;
2078
2079 error_unreg_notifier:
2080         unregister_netdevice_notifier(&dp_device_notifier);
2081 error_vport_exit:
2082         vport_exit();
2083 error_flow_exit:
2084         flow_exit();
2085 error:
2086         return err;
2087 }
2088
2089 static void dp_cleanup(void)
2090 {
2091         rcu_barrier();
2092         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2093         unregister_netdevice_notifier(&dp_device_notifier);
2094         vport_exit();
2095         flow_exit();
2096 }
2097
2098 module_init(dp_init);
2099 module_exit(dp_cleanup);
2100
2101 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2102 MODULE_LICENSE("GPL");