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 "loop_counter.h"
  53 #include "table.h"
  54 #include "vlan.h"
  55 #include "vport-internal_dev.h"
  56
  57 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
  58 EXPORT_SYMBOL(dp_ioctl_hook);
  59
  60 /**
  61  * DOC: Locking:
  62  *
  63  * Writes to device state (add/remove datapath, port, set operations on vports,
  64  * etc.) are protected by RTNL.
  65  *
  66  * Writes to other state (flow table modifications, set miscellaneous datapath
  67  * parameters such as drop frags, etc.) are protected by genl_mutex.  The RTNL
  68  * lock nests inside genl_mutex.
  69  *
  70  * Reads are protected by RCU.
  71  *
  72  * There are a few special cases (mostly stats) that have their own
  73  * synchronization but they nest under all of above and don't interact with
  74  * each other.
  75  */
  76
  77 /* Global list of datapaths to enable dumping them all out.
  78  * Protected by genl_mutex.
  79  */
  80 static LIST_HEAD(dps);
  81
  82 static struct vport *new_vport(const struct vport_parms *);
  83 static int queue_control_packets(struct datapath *, struct sk_buff *,
  84                                  const struct dp_upcall_info *);
  85
  86 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  87 struct datapath *get_dp(int dp_ifindex)
  88 {
  89         struct datapath *dp = NULL;
  90         struct net_device *dev;
  91
  92         rcu_read_lock();
  93         dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
  94         if (dev) {
  95                 struct vport *vport = internal_dev_get_vport(dev);
  96                 if (vport)
  97                         dp = vport->dp;
  98         }
  99         rcu_read_unlock();
 100
 101         return dp;
 102 }
 103 EXPORT_SYMBOL_GPL(get_dp);
 104
 105 /* Must be called with genl_mutex. */
 106 static struct tbl *get_table_protected(struct datapath *dp)
 107 {
 108         return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
 109 }
 110
 111 /* Must be called with rcu_read_lock or RTNL lock. */
 112 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
 113 {
 114         return rcu_dereference_rtnl(dp->ports[port_no]);
 115 }
 116
 117 /* Must be called with rcu_read_lock or RTNL lock. */
 118 const char *dp_name(const struct datapath *dp)
 119 {
 120         return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL]));
 121 }
 122
 123 static inline size_t br_nlmsg_size(void)
 124 {
 125         return NLMSG_ALIGN(sizeof(struct ifinfomsg))
 126                + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
 127                + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
 128                + nla_total_size(4) /* IFLA_MASTER */
 129                + nla_total_size(4) /* IFLA_MTU */
 130                + nla_total_size(4) /* IFLA_LINK */
 131                + nla_total_size(1); /* IFLA_OPERSTATE */
 132 }
 133
 134 /* Caller must hold RTNL lock. */
 135 static int dp_fill_ifinfo(struct sk_buff *skb,
 136                           const struct vport *port,
 137                           int event, unsigned int flags)
 138 {
 139         struct datapath *dp = port->dp;
 140         int ifindex = vport_get_ifindex(port);
 141         int iflink = vport_get_iflink(port);
 142         struct ifinfomsg *hdr;
 143         struct nlmsghdr *nlh;
 144
 145         if (ifindex < 0)
 146                 return ifindex;
 147
 148         if (iflink < 0)
 149                 return iflink;
 150
 151         nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
 152         if (nlh == NULL)
 153                 return -EMSGSIZE;
 154
 155         hdr = nlmsg_data(nlh);
 156         hdr->ifi_family = AF_BRIDGE;
 157         hdr->__ifi_pad = 0;
 158         hdr->ifi_type = ARPHRD_ETHER;
 159         hdr->ifi_index = ifindex;
 160         hdr->ifi_flags = vport_get_flags(port);
 161         hdr->ifi_change = 0;
 162
 163         NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
 164         NLA_PUT_U32(skb, IFLA_MASTER,
 165                 vport_get_ifindex(get_vport_protected(dp, ODPP_LOCAL)));
 166         NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
 167 #ifdef IFLA_OPERSTATE
 168         NLA_PUT_U8(skb, IFLA_OPERSTATE,
 169                    vport_is_running(port)
 170                         ? vport_get_operstate(port)
 171                         : IF_OPER_DOWN);
 172 #endif
 173
 174         NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
 175
 176         if (ifindex != iflink)
 177                 NLA_PUT_U32(skb, IFLA_LINK,iflink);
 178
 179         return nlmsg_end(skb, nlh);
 180
 181 nla_put_failure:
 182         nlmsg_cancel(skb, nlh);
 183         return -EMSGSIZE;
 184 }
 185
 186 /* Caller must hold RTNL lock. */
 187 static void dp_ifinfo_notify(int event, struct vport *port)
 188 {
 189         struct sk_buff *skb;
 190         int err = -ENOBUFS;
 191
 192         skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
 193         if (skb == NULL)
 194                 goto errout;
 195
 196         err = dp_fill_ifinfo(skb, port, event, 0);
 197         if (err < 0) {
 198                 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
 199                 WARN_ON(err == -EMSGSIZE);
 200                 kfree_skb(skb);
 201                 goto errout;
 202         }
 203         rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 204         return;
 205 errout:
 206         if (err < 0)
 207                 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
 208 }
 209
 210 static void release_dp(struct kobject *kobj)
 211 {
 212         struct datapath *dp = container_of(kobj, struct datapath, ifobj);
 213         kfree(dp);
 214 }
 215
 216 static struct kobj_type dp_ktype = {
 217         .release = release_dp
 218 };
 219
 220 static void destroy_dp_rcu(struct rcu_head *rcu)
 221 {
 222         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 223
 224         tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl);
 225         free_percpu(dp->stats_percpu);
 226         kobject_put(&dp->ifobj);
 227 }
 228
 229 /* Called with RTNL lock and genl_lock. */
 230 static struct vport *new_vport(const struct vport_parms *parms)
 231 {
 232         struct vport *vport;
 233
 234         vport = vport_add(parms);
 235         if (!IS_ERR(vport)) {
 236                 struct datapath *dp = parms->dp;
 237
 238                 rcu_assign_pointer(dp->ports[parms->port_no], vport);
 239                 list_add(&vport->node, &dp->port_list);
 240
 241                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 242         }
 243
 244         return vport;
 245 }
 246
 247 /* Called with RTNL lock. */
 248 int dp_detach_port(struct vport *p)
 249 {
 250         ASSERT_RTNL();
 251
 252         if (p->port_no != ODPP_LOCAL)
 253                 dp_sysfs_del_if(p);
 254         dp_ifinfo_notify(RTM_DELLINK, p);
 255
 256         /* First drop references to device. */
 257         list_del(&p->node);
 258         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 259
 260         /* Then destroy it. */
 261         return vport_del(p);
 262 }
 263
 264 /* Must be called with rcu_read_lock. */
 265 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 266 {
 267         struct datapath *dp = p->dp;
 268         struct dp_stats_percpu *stats;
 269         int stats_counter_off;
 270         struct sw_flow_actions *acts;
 271         struct loop_counter *loop;
 272         int error;
 273
 274         OVS_CB(skb)->vport = p;
 275
 276         if (!OVS_CB(skb)->flow) {
 277                 struct sw_flow_key key;
 278                 struct tbl_node *flow_node;
 279                 bool is_frag;
 280
 281                 /* Extract flow from 'skb' into 'key'. */
 282                 error = flow_extract(skb, p->port_no, &key, &is_frag);
 283                 if (unlikely(error)) {
 284                         kfree_skb(skb);
 285                         return;
 286                 }
 287
 288                 if (is_frag && dp->drop_frags) {
 289                         kfree_skb(skb);
 290                         stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
 291                         goto out;
 292                 }
 293
 294                 /* Look up flow. */
 295                 flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
 296                                         flow_hash(&key), flow_cmp);
 297                 if (unlikely(!flow_node)) {
 298                         struct dp_upcall_info upcall;
 299
 300                         upcall.cmd = ODP_PACKET_CMD_MISS;
 301                         upcall.key = &key;
 302                         upcall.userdata = 0;
 303                         upcall.sample_pool = 0;
 304                         upcall.actions = NULL;
 305                         upcall.actions_len = 0;
 306                         dp_upcall(dp, skb, &upcall);
 307                         stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
 308                         goto out;
 309                 }
 310
 311                 OVS_CB(skb)->flow = flow_cast(flow_node);
 312         }
 313
 314         stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
 315         flow_used(OVS_CB(skb)->flow, skb);
 316
 317         acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
 318
 319         /* Check whether we've looped too much. */
 320         loop = loop_get_counter();
 321         if (unlikely(++loop->count > MAX_LOOPS))
 322                 loop->looping = true;
 323         if (unlikely(loop->looping)) {
 324                 loop_suppress(dp, acts);
 325                 kfree_skb(skb);
 326                 goto out_loop;
 327         }
 328
 329         /* Execute actions. */
 330         execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
 331                         acts->actions_len);
 332
 333         /* Check whether sub-actions looped too much. */
 334         if (unlikely(loop->looping))
 335                 loop_suppress(dp, acts);
 336
 337 out_loop:
 338         /* Decrement loop counter. */
 339         if (!--loop->count)
 340                 loop->looping = false;
 341         loop_put_counter();
 342
 343 out:
 344         /* Update datapath statistics. */
 345         local_bh_disable();
 346         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 347
 348         write_seqcount_begin(&stats->seqlock);
 349         (*(u64 *)((u8 *)stats + stats_counter_off))++;
 350         write_seqcount_end(&stats->seqlock);
 351
 352         local_bh_enable();
 353 }
 354
 355 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
 356 {
 357         u16 csum_start, csum_offset;
 358         __wsum csum;
 359
 360         get_skb_csum_pointers(skb, &csum_start, &csum_offset);
 361         csum_start -= skb_headroom(skb);
 362         BUG_ON(csum_start >= skb_headlen(skb));
 363
 364         skb_copy_bits(skb, 0, to, csum_start);
 365
 366         csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
 367                                       skb->len - csum_start, 0);
 368         *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
 369 }
 370
 371 static struct genl_family dp_packet_genl_family = {
 372         .id = GENL_ID_GENERATE,
 373         .hdrsize = sizeof(struct odp_header),
 374         .name = ODP_PACKET_FAMILY,
 375         .version = 1,
 376         .maxattr = ODP_PACKET_ATTR_MAX
 377 };
 378
 379 /* Generic Netlink multicast groups for upcalls.
 380  *
 381  * We really want three unique multicast groups per datapath, but we can't even
 382  * get one, because genl_register_mc_group() takes genl_lock, which is also
 383  * held during Generic Netlink message processing, so trying to acquire
 384  * multicast groups during ODP_DP_NEW processing deadlocks.  Instead, we
 385  * preallocate a few groups and use them round-robin for datapaths.  Collision
 386  * isn't fatal--multicast listeners should check that the family is the one
 387  * that they want and discard others--but it wastes time and memory to receive
 388  * unwanted messages.
 389  */
 390 #define PACKET_N_MC_GROUPS 16
 391 static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
 392
 393 static u32 packet_mc_group(struct datapath *dp, u8 cmd)
 394 {
 395         u32 idx;
 396         BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
 397
 398         idx = jhash_2words(dp->dp_ifindex, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
 399         return packet_mc_groups[idx].id;
 400 }
 401
 402 static int packet_register_mc_groups(void)
 403 {
 404         int i;
 405
 406         for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
 407                 struct genl_multicast_group *group = &packet_mc_groups[i];
 408                 int error;
 409
 410                 sprintf(group->name, "packet%d", i);
 411                 error = genl_register_mc_group(&dp_packet_genl_family, group);
 412                 if (error)
 413                         return error;
 414         }
 415         return 0;
 416 }
 417
 418 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
 419 {
 420         struct dp_stats_percpu *stats;
 421         int err;
 422
 423         WARN_ON_ONCE(skb_shared(skb));
 424
 425         forward_ip_summed(skb);
 426
 427         err = vswitch_skb_checksum_setup(skb);
 428         if (err)
 429                 goto err_kfree_skb;
 430
 431         /* Break apart GSO packets into their component pieces.  Otherwise
 432          * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
 433         if (skb_is_gso(skb)) {
 434                 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 435
 436                 kfree_skb(skb);
 437                 skb = nskb;
 438                 if (IS_ERR(skb)) {
 439                         err = PTR_ERR(skb);
 440                         goto err;
 441                 }
 442         }
 443
 444         return queue_control_packets(dp, skb, upcall_info);
 445
 446 err_kfree_skb:
 447         kfree_skb(skb);
 448 err:
 449         local_bh_disable();
 450         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 451
 452         write_seqcount_begin(&stats->seqlock);
 453         stats->n_lost++;
 454         write_seqcount_end(&stats->seqlock);
 455
 456         local_bh_enable();
 457
 458         return err;
 459 }
 460
 461 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
 462  * 'upcall_info'.  There will be only one packet unless we broke up a GSO
 463  * packet.
 464  */
 465 static int queue_control_packets(struct datapath *dp, struct sk_buff *skb,
 466                                  const struct dp_upcall_info *upcall_info)
 467 {
 468         u32 group = packet_mc_group(dp, upcall_info->cmd);
 469         struct sk_buff *nskb;
 470         int port_no;
 471         int err;
 472
 473         if (OVS_CB(skb)->vport)
 474                 port_no = OVS_CB(skb)->vport->port_no;
 475         else
 476                 port_no = ODPP_LOCAL;
 477
 478         do {
 479                 struct odp_header *upcall;
 480                 struct sk_buff *user_skb; /* to be queued to userspace */
 481                 struct nlattr *nla;
 482                 unsigned int len;
 483
 484                 nskb = skb->next;
 485                 skb->next = NULL;
 486
 487                 err = vlan_deaccel_tag(skb);
 488                 if (unlikely(err))
 489                         goto err_kfree_skbs;
 490
 491                 len = sizeof(struct odp_header);
 492                 len += nla_total_size(skb->len);
 493                 len += nla_total_size(FLOW_BUFSIZE);
 494                 if (upcall_info->userdata)
 495                         len += nla_total_size(8);
 496                 if (upcall_info->sample_pool)
 497                         len += nla_total_size(4);
 498                 if (upcall_info->actions_len)
 499                         len += nla_total_size(upcall_info->actions_len);
 500
 501                 user_skb = genlmsg_new(len, GFP_ATOMIC);
 502                 if (!user_skb) {
 503                         netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS);
 504                         goto err_kfree_skbs;
 505                 }
 506
 507                 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
 508                 upcall->dp_ifindex = dp->dp_ifindex;
 509
 510                 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY);
 511                 flow_to_nlattrs(upcall_info->key, user_skb);
 512                 nla_nest_end(user_skb, nla);
 513
 514                 if (upcall_info->userdata)
 515                         nla_put_u64(user_skb, ODP_PACKET_ATTR_USERDATA, upcall_info->userdata);
 516                 if (upcall_info->sample_pool)
 517                         nla_put_u32(user_skb, ODP_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
 518                 if (upcall_info->actions_len) {
 519                         const struct nlattr *actions = upcall_info->actions;
 520                         u32 actions_len = upcall_info->actions_len;
 521
 522                         nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_ACTIONS);
 523                         memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
 524                         nla_nest_end(user_skb, nla);
 525                 }
 526
 527                 nla = __nla_reserve(user_skb, ODP_PACKET_ATTR_PACKET, skb->len);
 528                 if (skb->ip_summed == CHECKSUM_PARTIAL)
 529                         copy_and_csum_skb(skb, nla_data(nla));
 530                 else
 531                         skb_copy_bits(skb, 0, nla_data(nla), skb->len);
 532
 533                 err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC);
 534                 if (err)
 535                         goto err_kfree_skbs;
 536
 537                 kfree_skb(skb);
 538                 skb = nskb;
 539         } while (skb);
 540         return 0;
 541
 542 err_kfree_skbs:
 543         kfree_skb(skb);
 544         while ((skb = nskb) != NULL) {
 545                 nskb = skb->next;
 546                 kfree_skb(skb);
 547         }
 548         return err;
 549 }
 550
 551 /* Called with genl_mutex. */
 552 static int flush_flows(int dp_ifindex)
 553 {
 554         struct tbl *old_table;
 555         struct tbl *new_table;
 556         struct datapath *dp;
 557
 558         dp = get_dp(dp_ifindex);
 559         if (!dp)
 560                 return -ENODEV;
 561
 562         old_table = get_table_protected(dp);
 563         new_table = tbl_create(TBL_MIN_BUCKETS);
 564         if (!new_table)
 565                 return -ENOMEM;
 566
 567         rcu_assign_pointer(dp->table, new_table);
 568
 569         tbl_deferred_destroy(old_table, flow_free_tbl);
 570
 571         return 0;
 572 }
 573
 574 static int validate_actions(const struct nlattr *attr)
 575 {
 576         const struct nlattr *a;
 577         int rem;
 578
 579         nla_for_each_nested(a, attr, rem) {
 580                 static const u32 action_lens[ODP_ACTION_ATTR_MAX + 1] = {
 581                         [ODP_ACTION_ATTR_OUTPUT] = 4,
 582                         [ODP_ACTION_ATTR_CONTROLLER] = 8,
 583                         [ODP_ACTION_ATTR_SET_DL_TCI] = 2,
 584                         [ODP_ACTION_ATTR_STRIP_VLAN] = 0,
 585                         [ODP_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
 586                         [ODP_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
 587                         [ODP_ACTION_ATTR_SET_NW_SRC] = 4,
 588                         [ODP_ACTION_ATTR_SET_NW_DST] = 4,
 589                         [ODP_ACTION_ATTR_SET_NW_TOS] = 1,
 590                         [ODP_ACTION_ATTR_SET_TP_SRC] = 2,
 591                         [ODP_ACTION_ATTR_SET_TP_DST] = 2,
 592                         [ODP_ACTION_ATTR_SET_TUNNEL] = 8,
 593                         [ODP_ACTION_ATTR_SET_PRIORITY] = 4,
 594                         [ODP_ACTION_ATTR_POP_PRIORITY] = 0,
 595                         [ODP_ACTION_ATTR_DROP_SPOOFED_ARP] = 0,
 596                 };
 597                 int type = nla_type(a);
 598
 599                 if (type > ODP_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
 600                         return -EINVAL;
 601
 602                 switch (type) {
 603                 case ODP_ACTION_ATTR_UNSPEC:
 604                         return -EINVAL;
 605
 606                 case ODP_ACTION_ATTR_CONTROLLER:
 607                 case ODP_ACTION_ATTR_STRIP_VLAN:
 608                 case ODP_ACTION_ATTR_SET_DL_SRC:
 609                 case ODP_ACTION_ATTR_SET_DL_DST:
 610                 case ODP_ACTION_ATTR_SET_NW_SRC:
 611                 case ODP_ACTION_ATTR_SET_NW_DST:
 612                 case ODP_ACTION_ATTR_SET_TP_SRC:
 613                 case ODP_ACTION_ATTR_SET_TP_DST:
 614                 case ODP_ACTION_ATTR_SET_TUNNEL:
 615                 case ODP_ACTION_ATTR_SET_PRIORITY:
 616                 case ODP_ACTION_ATTR_POP_PRIORITY:
 617                 case ODP_ACTION_ATTR_DROP_SPOOFED_ARP:
 618                         /* No validation needed. */
 619                         break;
 620
 621                 case ODP_ACTION_ATTR_OUTPUT:
 622                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 623                                 return -EINVAL;
 624                         break;
 625
 626                 case ODP_ACTION_ATTR_SET_DL_TCI:
 627                         if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
 628                                 return -EINVAL;
 629                         break;
 630
 631                 case ODP_ACTION_ATTR_SET_NW_TOS:
 632                         if (nla_get_u8(a) & INET_ECN_MASK)
 633                                 return -EINVAL;
 634                         break;
 635
 636                 default:
 637                         return -EOPNOTSUPP;
 638                 }
 639         }
 640
 641         if (rem > 0)
 642                 return -EINVAL;
 643
 644         return 0;
 645 }
 646 static void clear_stats(struct sw_flow *flow)
 647 {
 648         flow->used = 0;
 649         flow->tcp_flags = 0;
 650         flow->packet_count = 0;
 651         flow->byte_count = 0;
 652 }
 653
 654 /* Called with genl_mutex. */
 655 static int expand_table(struct datapath *dp)
 656 {
 657         struct tbl *old_table = get_table_protected(dp);
 658         struct tbl *new_table;
 659
 660         new_table = tbl_expand(old_table);
 661         if (IS_ERR(new_table))
 662                 return PTR_ERR(new_table);
 663
 664         rcu_assign_pointer(dp->table, new_table);
 665         tbl_deferred_destroy(old_table, NULL);
 666
 667         return 0;
 668 }
 669
 670 static int odp_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 671 {
 672         struct odp_header *odp_header = info->userhdr;
 673         struct nlattr **a = info->attrs;
 674         struct sk_buff *packet;
 675         struct sw_flow_key key;
 676         struct datapath *dp;
 677         struct ethhdr *eth;
 678         bool is_frag;
 679         int err;
 680
 681         err = -EINVAL;
 682         if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_ACTIONS] ||
 683             nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN)
 684                 goto exit;
 685
 686         err = validate_actions(a[ODP_PACKET_ATTR_ACTIONS]);
 687         if (err)
 688                 goto exit;
 689
 690         packet = skb_clone(skb, GFP_KERNEL);
 691         err = -ENOMEM;
 692         if (!packet)
 693                 goto exit;
 694         packet->data = nla_data(a[ODP_PACKET_ATTR_PACKET]);
 695         packet->len = nla_len(a[ODP_PACKET_ATTR_PACKET]);
 696
 697         skb_reset_mac_header(packet);
 698         eth = eth_hdr(packet);
 699
 700         /* Normally, setting the skb 'protocol' field would be handled by a
 701          * call to eth_type_trans(), but it assumes there's a sending
 702          * device, which we may not have. */
 703         if (ntohs(eth->h_proto) >= 1536)
 704                 packet->protocol = eth->h_proto;
 705         else
 706                 packet->protocol = htons(ETH_P_802_2);
 707
 708         err = flow_extract(packet, -1, &key, &is_frag);
 709         if (err)
 710                 goto exit;
 711
 712         rcu_read_lock();
 713         dp = get_dp(odp_header->dp_ifindex);
 714         err = -ENODEV;
 715         if (dp)
 716                 err = execute_actions(dp, packet, &key,
 717                                       nla_data(a[ODP_PACKET_ATTR_ACTIONS]),
 718                                       nla_len(a[ODP_PACKET_ATTR_ACTIONS]));
 719         rcu_read_unlock();
 720
 721 exit:
 722         return err;
 723 }
 724
 725 static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = {
 726         [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 727         [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 728 };
 729
 730 static struct genl_ops dp_packet_genl_ops[] = {
 731         { .cmd = ODP_PACKET_CMD_EXECUTE,
 732           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 733           .policy = packet_policy,
 734           .doit = odp_packet_cmd_execute
 735         }
 736 };
 737
 738 static void get_dp_stats(struct datapath *dp, struct odp_stats *stats)
 739 {
 740         int i;
 741
 742         stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
 743         for_each_possible_cpu(i) {
 744                 const struct dp_stats_percpu *percpu_stats;
 745                 struct dp_stats_percpu local_stats;
 746                 unsigned seqcount;
 747
 748                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 749
 750                 do {
 751                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 752                         local_stats = *percpu_stats;
 753                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 754
 755                 stats->n_frags += local_stats.n_frags;
 756                 stats->n_hit += local_stats.n_hit;
 757                 stats->n_missed += local_stats.n_missed;
 758                 stats->n_lost += local_stats.n_lost;
 759         }
 760 }
 761
 762 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
 763  * Called with RTNL lock.
 764  */
 765 int dp_min_mtu(const struct datapath *dp)
 766 {
 767         struct vport *p;
 768         int mtu = 0;
 769
 770         ASSERT_RTNL();
 771
 772         list_for_each_entry (p, &dp->port_list, node) {
 773                 int dev_mtu;
 774
 775                 /* Skip any internal ports, since that's what we're trying to
 776                  * set. */
 777                 if (is_internal_vport(p))
 778                         continue;
 779
 780                 dev_mtu = vport_get_mtu(p);
 781                 if (!dev_mtu)
 782                         continue;
 783                 if (!mtu || dev_mtu < mtu)
 784                         mtu = dev_mtu;
 785         }
 786
 787         return mtu ? mtu : ETH_DATA_LEN;
 788 }
 789
 790 /* Sets the MTU of all datapath devices to the minimum of the ports
 791  * Called with RTNL lock.
 792  */
 793 void set_internal_devs_mtu(const struct datapath *dp)
 794 {
 795         struct vport *p;
 796         int mtu;
 797
 798         ASSERT_RTNL();
 799
 800         mtu = dp_min_mtu(dp);
 801
 802         list_for_each_entry (p, &dp->port_list, node) {
 803                 if (is_internal_vport(p))
 804                         vport_set_mtu(p, mtu);
 805         }
 806 }
 807
 808 static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = {
 809         [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 810         [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 811         [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 812 };
 813
 814 static struct genl_family dp_flow_genl_family = {
 815         .id = GENL_ID_GENERATE,
 816         .hdrsize = sizeof(struct odp_header),
 817         .name = ODP_FLOW_FAMILY,
 818         .version = 1,
 819         .maxattr = ODP_FLOW_ATTR_MAX
 820 };
 821
 822 static struct genl_multicast_group dp_flow_multicast_group = {
 823         .name = ODP_FLOW_MCGROUP
 824 };
 825
 826 /* Called with genl_lock. */
 827 static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 828                                   struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
 829 {
 830         const int skb_orig_len = skb->len;
 831         const struct sw_flow_actions *sf_acts;
 832         struct odp_flow_stats stats;
 833         struct odp_header *odp_header;
 834         struct nlattr *nla;
 835         unsigned long used;
 836         u8 tcp_flags;
 837         int err;
 838
 839         sf_acts = rcu_dereference_protected(flow->sf_acts,
 840                                             lockdep_genl_is_held());
 841
 842         odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 843         if (!odp_header)
 844                 return -EMSGSIZE;
 845
 846         odp_header->dp_ifindex = dp->dp_ifindex;
 847
 848         nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY);
 849         if (!nla)
 850                 goto nla_put_failure;
 851         err = flow_to_nlattrs(&flow->key, skb);
 852         if (err)
 853                 goto error;
 854         nla_nest_end(skb, nla);
 855
 856         spin_lock_bh(&flow->lock);
 857         used = flow->used;
 858         stats.n_packets = flow->packet_count;
 859         stats.n_bytes = flow->byte_count;
 860         tcp_flags = flow->tcp_flags;
 861         spin_unlock_bh(&flow->lock);
 862
 863         if (used)
 864                 NLA_PUT_U64(skb, ODP_FLOW_ATTR_USED, flow_used_time(used));
 865
 866         if (stats.n_packets)
 867                 NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats);
 868
 869         if (tcp_flags)
 870                 NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 871
 872         /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 873          * this is the first flow to be dumped into 'skb'.  This is unusual for
 874          * Netlink but individual action lists can be longer than
 875          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 876          * The userspace caller can always fetch the actions separately if it
 877          * really wants them.  (Most userspace callers in fact don't care.)
 878          *
 879          * This can only fail for dump operations because the skb is always
 880          * properly sized for single flows.
 881          */
 882         err = nla_put(skb, ODP_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 883                       sf_acts->actions);
 884         if (err < 0 && skb_orig_len)
 885                 goto error;
 886
 887         return genlmsg_end(skb, odp_header);
 888
 889 nla_put_failure:
 890         err = -EMSGSIZE;
 891 error:
 892         genlmsg_cancel(skb, odp_header);
 893         return err;
 894 }
 895
 896 static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow)
 897 {
 898         const struct sw_flow_actions *sf_acts;
 899         int len;
 900
 901         sf_acts = rcu_dereference_protected(flow->sf_acts,
 902                                             lockdep_genl_is_held());
 903
 904         len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */
 905         len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */
 906         len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */
 907         len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */
 908         len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */
 909         return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL);
 910 }
 911
 912 static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
 913                                                u32 pid, u32 seq, u8 cmd)
 914 {
 915         struct sk_buff *skb;
 916         int retval;
 917
 918         skb = odp_flow_cmd_alloc_info(flow);
 919         if (!skb)
 920                 return ERR_PTR(-ENOMEM);
 921
 922         retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 923         BUG_ON(retval < 0);
 924         return skb;
 925 }
 926
 927 static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 928 {
 929         struct nlattr **a = info->attrs;
 930         struct odp_header *odp_header = info->userhdr;
 931         struct tbl_node *flow_node;
 932         struct sw_flow_key key;
 933         struct sw_flow *flow;
 934         struct sk_buff *reply;
 935         struct datapath *dp;
 936         struct tbl *table;
 937         u32 hash;
 938         int error;
 939
 940         /* Extract key. */
 941         error = -EINVAL;
 942         if (!a[ODP_FLOW_ATTR_KEY])
 943                 goto error;
 944         error = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
 945         if (error)
 946                 goto error;
 947
 948         /* Validate actions. */
 949         if (a[ODP_FLOW_ATTR_ACTIONS]) {
 950                 error = validate_actions(a[ODP_FLOW_ATTR_ACTIONS]);
 951                 if (error)
 952                         goto error;
 953         } else if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW) {
 954                 error = -EINVAL;
 955                 goto error;
 956         }
 957
 958         dp = get_dp(odp_header->dp_ifindex);
 959         error = -ENODEV;
 960         if (!dp)
 961                 goto error;
 962
 963         hash = flow_hash(&key);
 964         table = get_table_protected(dp);
 965         flow_node = tbl_lookup(table, &key, hash, flow_cmp);
 966         if (!flow_node) {
 967                 struct sw_flow_actions *acts;
 968
 969                 /* Bail out if we're not allowed to create a new flow. */
 970                 error = -ENOENT;
 971                 if (info->genlhdr->cmd == ODP_FLOW_CMD_SET)
 972                         goto error;
 973
 974                 /* Expand table, if necessary, to make room. */
 975                 if (tbl_count(table) >= tbl_n_buckets(table)) {
 976                         error = expand_table(dp);
 977                         if (error)
 978                                 goto error;
 979                         table = get_table_protected(dp);
 980                 }
 981
 982                 /* Allocate flow. */
 983                 flow = flow_alloc();
 984                 if (IS_ERR(flow)) {
 985                         error = PTR_ERR(flow);
 986                         goto error;
 987                 }
 988                 flow->key = key;
 989                 clear_stats(flow);
 990
 991                 /* Obtain actions. */
 992                 acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
 993                 error = PTR_ERR(acts);
 994                 if (IS_ERR(acts))
 995                         goto error_free_flow;
 996                 rcu_assign_pointer(flow->sf_acts, acts);
 997
 998                 /* Put flow in bucket. */
 999                 error = tbl_insert(table, &flow->tbl_node, hash);
1000                 if (error)
1001                         goto error_free_flow;
1002
1003                 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1004                                                 info->snd_seq, ODP_FLOW_CMD_NEW);
1005         } else {
1006                 /* We found a matching flow. */
1007                 struct sw_flow_actions *old_acts;
1008
1009                 /* Bail out if we're not allowed to modify an existing flow.
1010                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1011                  * because Generic Netlink treats the latter as a dump
1012                  * request.  We also accept NLM_F_EXCL in case that bug ever
1013                  * gets fixed.
1014                  */
1015                 error = -EEXIST;
1016                 if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW &&
1017                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1018                         goto error;
1019
1020                 /* Update actions. */
1021                 flow = flow_cast(flow_node);
1022                 old_acts = rcu_dereference_protected(flow->sf_acts,
1023                                                      lockdep_genl_is_held());
1024                 if (a[ODP_FLOW_ATTR_ACTIONS] &&
1025                     (old_acts->actions_len != nla_len(a[ODP_FLOW_ATTR_ACTIONS]) ||
1026                      memcmp(old_acts->actions, nla_data(a[ODP_FLOW_ATTR_ACTIONS]),
1027                             old_acts->actions_len))) {
1028                         struct sw_flow_actions *new_acts;
1029
1030                         new_acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1031                         error = PTR_ERR(new_acts);
1032                         if (IS_ERR(new_acts))
1033                                 goto error;
1034
1035                         rcu_assign_pointer(flow->sf_acts, new_acts);
1036                         flow_deferred_free_acts(old_acts);
1037                 }
1038
1039                 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1040                                                 info->snd_seq, ODP_FLOW_CMD_NEW);
1041
1042                 /* Clear stats. */
1043                 if (a[ODP_FLOW_ATTR_CLEAR]) {
1044                         spin_lock_bh(&flow->lock);
1045                         clear_stats(flow);
1046                         spin_unlock_bh(&flow->lock);
1047                 }
1048         }
1049
1050         if (!IS_ERR(reply))
1051                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1052                             dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1053         else
1054                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1055                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1056         return 0;
1057
1058 error_free_flow:
1059         flow_put(flow);
1060 error:
1061         return error;
1062 }
1063
1064 static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1065 {
1066         struct nlattr **a = info->attrs;
1067         struct odp_header *odp_header = info->userhdr;
1068         struct sw_flow_key key;
1069         struct tbl_node *flow_node;
1070         struct sk_buff *reply;
1071         struct sw_flow *flow;
1072         struct datapath *dp;
1073         struct tbl *table;
1074         int err;
1075
1076         if (!a[ODP_FLOW_ATTR_KEY])
1077                 return -EINVAL;
1078         err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1079         if (err)
1080                 return err;
1081
1082         dp = get_dp(odp_header->dp_ifindex);
1083         if (!dp)
1084                 return -ENODEV;
1085
1086         table = get_table_protected(dp);
1087         flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1088         if (!flow_node)
1089                 return -ENOENT;
1090
1091         flow = flow_cast(flow_node);
1092         reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW);
1093         if (IS_ERR(reply))
1094                 return PTR_ERR(reply);
1095
1096         return genlmsg_reply(reply, info);
1097 }
1098
1099 static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1100 {
1101         struct nlattr **a = info->attrs;
1102         struct odp_header *odp_header = info->userhdr;
1103         struct sw_flow_key key;
1104         struct tbl_node *flow_node;
1105         struct sk_buff *reply;
1106         struct sw_flow *flow;
1107         struct datapath *dp;
1108         struct tbl *table;
1109         int err;
1110
1111         if (!a[ODP_FLOW_ATTR_KEY])
1112                 return flush_flows(odp_header->dp_ifindex);
1113         err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1114         if (err)
1115                 return err;
1116
1117         dp = get_dp(odp_header->dp_ifindex);
1118         if (!dp)
1119                 return -ENODEV;
1120
1121         table = get_table_protected(dp);
1122         flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1123         if (!flow_node)
1124                 return -ENOENT;
1125         flow = flow_cast(flow_node);
1126
1127         reply = odp_flow_cmd_alloc_info(flow);
1128         if (!reply)
1129                 return -ENOMEM;
1130
1131         err = tbl_remove(table, flow_node);
1132         if (err) {
1133                 kfree_skb(reply);
1134                 return err;
1135         }
1136
1137         err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1138                                      info->snd_seq, 0, ODP_FLOW_CMD_DEL);
1139         BUG_ON(err < 0);
1140
1141         flow_deferred_free(flow);
1142
1143         genl_notify(reply, genl_info_net(info), info->snd_pid,
1144                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1145         return 0;
1146 }
1147
1148 static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1149 {
1150         struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1151         struct datapath *dp;
1152
1153         dp = get_dp(odp_header->dp_ifindex);
1154         if (!dp)
1155                 return -ENODEV;
1156
1157         for (;;) {
1158                 struct tbl_node *flow_node;
1159                 struct sw_flow *flow;
1160                 u32 bucket, obj;
1161
1162                 bucket = cb->args[0];
1163                 obj = cb->args[1];
1164                 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1165                 if (!flow_node)
1166                         break;
1167
1168                 flow = flow_cast(flow_node);
1169                 if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1170                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1171                                            ODP_FLOW_CMD_NEW) < 0)
1172                         break;
1173
1174                 cb->args[0] = bucket;
1175                 cb->args[1] = obj;
1176         }
1177         return skb->len;
1178 }
1179
1180 static struct genl_ops dp_flow_genl_ops[] = {
1181         { .cmd = ODP_FLOW_CMD_NEW,
1182           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1183           .policy = flow_policy,
1184           .doit = odp_flow_cmd_new_or_set
1185         },
1186         { .cmd = ODP_FLOW_CMD_DEL,
1187           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1188           .policy = flow_policy,
1189           .doit = odp_flow_cmd_del
1190         },
1191         { .cmd = ODP_FLOW_CMD_GET,
1192           .flags = 0,               /* OK for unprivileged users. */
1193           .policy = flow_policy,
1194           .doit = odp_flow_cmd_get,
1195           .dumpit = odp_flow_cmd_dump
1196         },
1197         { .cmd = ODP_FLOW_CMD_SET,
1198           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1199           .policy = flow_policy,
1200           .doit = odp_flow_cmd_new_or_set,
1201         },
1202 };
1203
1204 static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
1205 #ifdef HAVE_NLA_NUL_STRING
1206         [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1207 #endif
1208         [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1209         [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1210 };
1211
1212 static struct genl_family dp_datapath_genl_family = {
1213         .id = GENL_ID_GENERATE,
1214         .hdrsize = sizeof(struct odp_header),
1215         .name = ODP_DATAPATH_FAMILY,
1216         .version = 1,
1217         .maxattr = ODP_DP_ATTR_MAX
1218 };
1219
1220 static struct genl_multicast_group dp_datapath_multicast_group = {
1221         .name = ODP_DATAPATH_MCGROUP
1222 };
1223
1224 static int odp_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1225                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1226 {
1227         struct odp_header *odp_header;
1228         struct nlattr *nla;
1229         int err;
1230
1231         odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1232                                    flags, cmd);
1233         if (!odp_header)
1234                 goto error;
1235
1236         odp_header->dp_ifindex = dp->dp_ifindex;
1237
1238         rcu_read_lock();
1239         err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
1240         rcu_read_unlock();
1241         if (err)
1242                 goto nla_put_failure;
1243
1244         nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats));
1245         if (!nla)
1246                 goto nla_put_failure;
1247         get_dp_stats(dp, nla_data(nla));
1248
1249         NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS,
1250                     dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO);
1251
1252         if (dp->sflow_probability)
1253                 NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability);
1254
1255         nla = nla_nest_start(skb, ODP_DP_ATTR_MCGROUPS);
1256         if (!nla)
1257                 goto nla_put_failure;
1258         NLA_PUT_U32(skb, ODP_PACKET_CMD_MISS, packet_mc_group(dp, ODP_PACKET_CMD_MISS));
1259         NLA_PUT_U32(skb, ODP_PACKET_CMD_ACTION, packet_mc_group(dp, ODP_PACKET_CMD_ACTION));
1260         NLA_PUT_U32(skb, ODP_PACKET_CMD_SAMPLE, packet_mc_group(dp, ODP_PACKET_CMD_SAMPLE));
1261         nla_nest_end(skb, nla);
1262
1263         return genlmsg_end(skb, odp_header);
1264
1265 nla_put_failure:
1266         genlmsg_cancel(skb, odp_header);
1267 error:
1268         return -EMSGSIZE;
1269 }
1270
1271 static struct sk_buff *odp_dp_cmd_build_info(struct datapath *dp, u32 pid,
1272                                              u32 seq, u8 cmd)
1273 {
1274         struct sk_buff *skb;
1275         int retval;
1276
1277         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1278         if (!skb)
1279                 return ERR_PTR(-ENOMEM);
1280
1281         retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1282         if (retval < 0) {
1283                 kfree_skb(skb);
1284                 return ERR_PTR(retval);
1285         }
1286         return skb;
1287 }
1288
1289 static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1290 {
1291         if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1292                 u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1293
1294                 if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP)
1295                         return -EINVAL;
1296         }
1297
1298         return VERIFY_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
1299 }
1300
1301 /* Called with genl_mutex and optionally with RTNL lock also. */
1302 static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1303 {
1304         struct datapath *dp;
1305
1306         if (!a[ODP_DP_ATTR_NAME])
1307                 dp = get_dp(odp_header->dp_ifindex);
1308         else {
1309                 struct vport *vport;
1310
1311                 rcu_read_lock();
1312                 vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
1313                 dp = vport && vport->port_no == ODPP_LOCAL ? vport->dp : NULL;
1314                 rcu_read_unlock();
1315         }
1316         return dp ? dp : ERR_PTR(-ENODEV);
1317 }
1318
1319 /* Called with genl_mutex. */
1320 static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1321 {
1322         if (a[ODP_DP_ATTR_IPV4_FRAGS])
1323                 dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP;
1324         if (a[ODP_DP_ATTR_SAMPLING])
1325                 dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]);
1326 }
1327
1328 static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1329 {
1330         struct nlattr **a = info->attrs;
1331         struct vport_parms parms;
1332         struct sk_buff *reply;
1333         struct datapath *dp;
1334         struct vport *vport;
1335         int err;
1336
1337         err = -EINVAL;
1338         if (!a[ODP_DP_ATTR_NAME])
1339                 goto err;
1340
1341         err = odp_dp_cmd_validate(a);
1342         if (err)
1343                 goto err;
1344
1345         rtnl_lock();
1346         err = -ENODEV;
1347         if (!try_module_get(THIS_MODULE))
1348                 goto err_unlock_rtnl;
1349
1350         err = -ENOMEM;
1351         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1352         if (dp == NULL)
1353                 goto err_put_module;
1354         INIT_LIST_HEAD(&dp->port_list);
1355
1356         /* Initialize kobject for bridge.  This will be added as
1357          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1358         dp->ifobj.kset = NULL;
1359         kobject_init(&dp->ifobj, &dp_ktype);
1360
1361         /* Allocate table. */
1362         err = -ENOMEM;
1363         rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1364         if (!dp->table)
1365                 goto err_free_dp;
1366
1367         /* Set up our datapath device. */
1368         parms.name = nla_data(a[ODP_DP_ATTR_NAME]);
1369         parms.type = ODP_VPORT_TYPE_INTERNAL;
1370         parms.options = NULL;
1371         parms.dp = dp;
1372         parms.port_no = ODPP_LOCAL;
1373         vport = new_vport(&parms);
1374         if (IS_ERR(vport)) {
1375                 err = PTR_ERR(vport);
1376                 if (err == -EBUSY)
1377                         err = -EEXIST;
1378
1379                 goto err_destroy_table;
1380         }
1381         dp->dp_ifindex = vport_get_ifindex(vport);
1382
1383         dp->drop_frags = 0;
1384         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1385         if (!dp->stats_percpu) {
1386                 err = -ENOMEM;
1387                 goto err_destroy_local_port;
1388         }
1389
1390         change_datapath(dp, a);
1391
1392         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1393         err = PTR_ERR(reply);
1394         if (IS_ERR(reply))
1395                 goto err_destroy_local_port;
1396
1397         list_add_tail(&dp->list_node, &dps);
1398         dp_sysfs_add_dp(dp);
1399
1400         rtnl_unlock();
1401
1402         genl_notify(reply, genl_info_net(info), info->snd_pid,
1403                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1404         return 0;
1405
1406 err_destroy_local_port:
1407         dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1408 err_destroy_table:
1409         tbl_destroy(get_table_protected(dp), NULL);
1410 err_free_dp:
1411         kfree(dp);
1412 err_put_module:
1413         module_put(THIS_MODULE);
1414 err_unlock_rtnl:
1415         rtnl_unlock();
1416 err:
1417         return err;
1418 }
1419
1420 static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1421 {
1422         struct vport *vport, *next_vport;
1423         struct sk_buff *reply;
1424         struct datapath *dp;
1425         int err;
1426
1427         err = odp_dp_cmd_validate(info->attrs);
1428         if (err)
1429                 goto exit;
1430
1431         rtnl_lock();
1432         dp = lookup_datapath(info->userhdr, info->attrs);
1433         err = PTR_ERR(dp);
1434         if (IS_ERR(dp))
1435                 goto exit_unlock;
1436
1437         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_DEL);
1438         err = PTR_ERR(reply);
1439         if (IS_ERR(reply))
1440                 goto exit_unlock;
1441
1442         list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1443                 if (vport->port_no != ODPP_LOCAL)
1444                         dp_detach_port(vport);
1445
1446         dp_sysfs_del_dp(dp);
1447         list_del(&dp->list_node);
1448         dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1449
1450         call_rcu(&dp->rcu, destroy_dp_rcu);
1451         module_put(THIS_MODULE);
1452
1453         genl_notify(reply, genl_info_net(info), info->snd_pid,
1454                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1455         err = 0;
1456
1457 exit_unlock:
1458         rtnl_unlock();
1459 exit:
1460         return err;
1461 }
1462
1463 static int odp_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1464 {
1465         struct sk_buff *reply;
1466         struct datapath *dp;
1467         int err;
1468
1469         err = odp_dp_cmd_validate(info->attrs);
1470         if (err)
1471                 return err;
1472
1473         dp = lookup_datapath(info->userhdr, info->attrs);
1474         if (IS_ERR(dp))
1475                 return PTR_ERR(dp);
1476
1477         change_datapath(dp, info->attrs);
1478
1479         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1480         if (IS_ERR(reply)) {
1481                 err = PTR_ERR(reply);
1482                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1483                                 dp_datapath_multicast_group.id, err);
1484                 return 0;
1485         }
1486
1487         genl_notify(reply, genl_info_net(info), info->snd_pid,
1488                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1489         return 0;
1490 }
1491
1492 static int odp_dp_cmd_get(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         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1507         if (IS_ERR(reply))
1508                 return PTR_ERR(reply);
1509
1510         return genlmsg_reply(reply, info);
1511 }
1512
1513 static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1514 {
1515         struct datapath *dp;
1516         int skip = cb->args[0];
1517         int i = 0;
1518
1519         list_for_each_entry (dp, &dps, list_node) {
1520                 if (i < skip)
1521                         continue;
1522                 if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1523                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1524                                          ODP_DP_CMD_NEW) < 0)
1525                         break;
1526                 i++;
1527         }
1528
1529         cb->args[0] = i;
1530
1531         return skb->len;
1532 }
1533
1534 static struct genl_ops dp_datapath_genl_ops[] = {
1535         { .cmd = ODP_DP_CMD_NEW,
1536           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1537           .policy = datapath_policy,
1538           .doit = odp_dp_cmd_new
1539         },
1540         { .cmd = ODP_DP_CMD_DEL,
1541           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1542           .policy = datapath_policy,
1543           .doit = odp_dp_cmd_del
1544         },
1545         { .cmd = ODP_DP_CMD_GET,
1546           .flags = 0,               /* OK for unprivileged users. */
1547           .policy = datapath_policy,
1548           .doit = odp_dp_cmd_get,
1549           .dumpit = odp_dp_cmd_dump
1550         },
1551         { .cmd = ODP_DP_CMD_SET,
1552           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1553           .policy = datapath_policy,
1554           .doit = odp_dp_cmd_set,
1555         },
1556 };
1557
1558 static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = {
1559 #ifdef HAVE_NLA_NUL_STRING
1560         [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1561         [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1562         [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1563         [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1564         [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1565 #else
1566         [ODP_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1567         [ODP_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1568 #endif
1569         [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1570         [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1571 };
1572
1573 static struct genl_family dp_vport_genl_family = {
1574         .id = GENL_ID_GENERATE,
1575         .hdrsize = sizeof(struct odp_header),
1576         .name = ODP_VPORT_FAMILY,
1577         .version = 1,
1578         .maxattr = ODP_VPORT_ATTR_MAX
1579 };
1580
1581 static struct genl_multicast_group dp_vport_multicast_group = {
1582         .name = ODP_VPORT_MCGROUP
1583 };
1584
1585 /* Called with RTNL lock or RCU read lock. */
1586 static int odp_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1587                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1588 {
1589         struct odp_header *odp_header;
1590         struct nlattr *nla;
1591         int ifindex, iflink;
1592         int mtu;
1593         int err;
1594
1595         odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1596                                  flags, cmd);
1597         if (!odp_header)
1598                 return -EMSGSIZE;
1599
1600         odp_header->dp_ifindex = vport->dp->dp_ifindex;
1601
1602         NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1603         NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport));
1604         NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport));
1605
1606         nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1607         if (!nla)
1608                 goto nla_put_failure;
1609         if (vport_get_stats(vport, nla_data(nla)))
1610                 __skb_trim(skb, skb->len - nla->nla_len);
1611
1612         NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1613
1614         mtu = vport_get_mtu(vport);
1615         if (mtu)
1616                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, mtu);
1617
1618         err = vport_get_options(vport, skb);
1619         if (err == -EMSGSIZE)
1620                 goto error;
1621
1622         ifindex = vport_get_ifindex(vport);
1623         if (ifindex > 0)
1624                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex);
1625
1626         iflink = vport_get_iflink(vport);
1627         if (iflink > 0)
1628                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink);
1629
1630         return genlmsg_end(skb, odp_header);
1631
1632 nla_put_failure:
1633         err = -EMSGSIZE;
1634 error:
1635         genlmsg_cancel(skb, odp_header);
1636         return err;
1637 }
1638
1639 /* Called with RTNL lock or RCU read lock. */
1640 static struct sk_buff *odp_vport_cmd_build_info(struct vport *vport, u32 pid,
1641                                                 u32 seq, u8 cmd)
1642 {
1643         struct sk_buff *skb;
1644         int retval;
1645
1646         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1647         if (!skb)
1648                 return ERR_PTR(-ENOMEM);
1649
1650         retval = odp_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1651         if (retval < 0) {
1652                 kfree_skb(skb);
1653                 return ERR_PTR(retval);
1654         }
1655         return skb;
1656 }
1657
1658 static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1659 {
1660         return VERIFY_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1661 }
1662
1663 /* Called with RTNL lock or RCU read lock. */
1664 static struct vport *lookup_vport(struct odp_header *odp_header,
1665                                   struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1666 {
1667         struct datapath *dp;
1668         struct vport *vport;
1669
1670         if (a[ODP_VPORT_ATTR_NAME]) {
1671                 vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME]));
1672                 if (!vport)
1673                         return ERR_PTR(-ENODEV);
1674                 return vport;
1675         } else if (a[ODP_VPORT_ATTR_PORT_NO]) {
1676                 u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1677
1678                 if (port_no >= DP_MAX_PORTS)
1679                         return ERR_PTR(-EFBIG);
1680
1681                 dp = get_dp(odp_header->dp_ifindex);
1682                 if (!dp)
1683                         return ERR_PTR(-ENODEV);
1684
1685                 vport = get_vport_protected(dp, port_no);
1686                 if (!vport)
1687                         return ERR_PTR(-ENOENT);
1688                 return vport;
1689         } else
1690                 return ERR_PTR(-EINVAL);
1691 }
1692
1693 /* Called with RTNL lock. */
1694 static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1695 {
1696         int err = 0;
1697         if (a[ODP_VPORT_ATTR_STATS])
1698                 err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS]));
1699         if (!err && a[ODP_VPORT_ATTR_ADDRESS])
1700                 err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS]));
1701         if (!err && a[ODP_VPORT_ATTR_MTU])
1702                 err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU]));
1703         return err;
1704 }
1705
1706 static int odp_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1707 {
1708         struct nlattr **a = info->attrs;
1709         struct odp_header *odp_header = info->userhdr;
1710         struct vport_parms parms;
1711         struct sk_buff *reply;
1712         struct vport *vport;
1713         struct datapath *dp;
1714         u32 port_no;
1715         int err;
1716
1717         err = -EINVAL;
1718         if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE])
1719                 goto exit;
1720
1721         err = odp_vport_cmd_validate(a);
1722         if (err)
1723                 goto exit;
1724
1725         rtnl_lock();
1726         dp = get_dp(odp_header->dp_ifindex);
1727         err = -ENODEV;
1728         if (!dp)
1729                 goto exit_unlock;
1730
1731         if (a[ODP_VPORT_ATTR_PORT_NO]) {
1732                 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1733
1734                 err = -EFBIG;
1735                 if (port_no >= DP_MAX_PORTS)
1736                         goto exit_unlock;
1737
1738                 vport = get_vport_protected(dp, port_no);
1739                 err = -EBUSY;
1740                 if (vport)
1741                         goto exit_unlock;
1742         } else {
1743                 for (port_no = 1; ; port_no++) {
1744                         if (port_no >= DP_MAX_PORTS) {
1745                                 err = -EFBIG;
1746                                 goto exit_unlock;
1747                         }
1748                         vport = get_vport_protected(dp, port_no);
1749                         if (!vport)
1750                                 break;
1751                 }
1752         }
1753
1754         parms.name = nla_data(a[ODP_VPORT_ATTR_NAME]);
1755         parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1756         parms.options = a[ODP_VPORT_ATTR_OPTIONS];
1757         parms.dp = dp;
1758         parms.port_no = port_no;
1759
1760         vport = new_vport(&parms);
1761         err = PTR_ERR(vport);
1762         if (IS_ERR(vport))
1763                 goto exit_unlock;
1764
1765         set_internal_devs_mtu(dp);
1766         dp_sysfs_add_if(vport);
1767
1768         err = change_vport(vport, a);
1769         if (!err) {
1770                 reply = odp_vport_cmd_build_info(vport, info->snd_pid,
1771                                                  info->snd_seq, ODP_VPORT_CMD_NEW);
1772                 if (IS_ERR(reply))
1773                         err = PTR_ERR(reply);
1774         }
1775         if (err) {
1776                 dp_detach_port(vport);
1777                 goto exit_unlock;
1778         }
1779         genl_notify(reply, genl_info_net(info), info->snd_pid,
1780                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1781
1782
1783 exit_unlock:
1784         rtnl_unlock();
1785 exit:
1786         return err;
1787 }
1788
1789 static int odp_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1790 {
1791         struct nlattr **a = info->attrs;
1792         struct sk_buff *reply;
1793         struct vport *vport;
1794         int err;
1795
1796         err = odp_vport_cmd_validate(a);
1797         if (err)
1798                 goto exit;
1799
1800         rtnl_lock();
1801         vport = lookup_vport(info->userhdr, a);
1802         err = PTR_ERR(vport);
1803         if (IS_ERR(vport))
1804                 goto exit_unlock;
1805
1806         err = 0;
1807         if (a[ODP_VPORT_ATTR_OPTIONS])
1808                 err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]);
1809         if (!err)
1810                 err = change_vport(vport, a);
1811
1812         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1813                                          ODP_VPORT_CMD_NEW);
1814         if (IS_ERR(reply)) {
1815                 err = PTR_ERR(reply);
1816                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1817                                 dp_vport_multicast_group.id, err);
1818                 return 0;
1819         }
1820
1821         genl_notify(reply, genl_info_net(info), info->snd_pid,
1822                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1823
1824 exit_unlock:
1825         rtnl_unlock();
1826 exit:
1827         return err;
1828 }
1829
1830 static int odp_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1831 {
1832         struct nlattr **a = info->attrs;
1833         struct sk_buff *reply;
1834         struct vport *vport;
1835         int err;
1836
1837         err = odp_vport_cmd_validate(a);
1838         if (err)
1839                 goto exit;
1840
1841         rtnl_lock();
1842         vport = lookup_vport(info->userhdr, a);
1843         err = PTR_ERR(vport);
1844         if (IS_ERR(vport))
1845                 goto exit_unlock;
1846
1847         if (vport->port_no == ODPP_LOCAL) {
1848                 err = -EINVAL;
1849                 goto exit_unlock;
1850         }
1851
1852         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1853                                          ODP_VPORT_CMD_DEL);
1854         err = PTR_ERR(reply);
1855         if (IS_ERR(reply))
1856                 goto exit_unlock;
1857
1858         err = dp_detach_port(vport);
1859
1860         genl_notify(reply, genl_info_net(info), info->snd_pid,
1861                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1862
1863 exit_unlock:
1864         rtnl_unlock();
1865 exit:
1866         return err;
1867 }
1868
1869 static int odp_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1870 {
1871         struct nlattr **a = info->attrs;
1872         struct odp_header *odp_header = info->userhdr;
1873         struct sk_buff *reply;
1874         struct vport *vport;
1875         int err;
1876
1877         err = odp_vport_cmd_validate(a);
1878         if (err)
1879                 goto exit;
1880
1881         rcu_read_lock();
1882         vport = lookup_vport(odp_header, a);
1883         err = PTR_ERR(vport);
1884         if (IS_ERR(vport))
1885                 goto exit_unlock;
1886
1887         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1888                                          ODP_VPORT_CMD_NEW);
1889         err = PTR_ERR(reply);
1890         if (IS_ERR(reply))
1891                 goto exit_unlock;
1892
1893         err = genlmsg_reply(reply, info);
1894
1895 exit_unlock:
1896         rcu_read_unlock();
1897 exit:
1898         return err;
1899 }
1900
1901 static int odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1902 {
1903         struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1904         struct datapath *dp;
1905         u32 port_no;
1906         int retval;
1907
1908         dp = get_dp(odp_header->dp_ifindex);
1909         if (!dp)
1910                 return -ENODEV;
1911
1912         rcu_read_lock();
1913         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1914                 struct vport *vport;
1915
1916                 vport = get_vport_protected(dp, port_no);
1917                 if (!vport)
1918                         continue;
1919
1920                 if (odp_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1921                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1922                                             ODP_VPORT_CMD_NEW) < 0)
1923                         break;
1924         }
1925         rcu_read_unlock();
1926
1927         cb->args[0] = port_no;
1928         retval = skb->len;
1929
1930         return retval;
1931 }
1932
1933 static struct genl_ops dp_vport_genl_ops[] = {
1934         { .cmd = ODP_VPORT_CMD_NEW,
1935           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1936           .policy = vport_policy,
1937           .doit = odp_vport_cmd_new
1938         },
1939         { .cmd = ODP_VPORT_CMD_DEL,
1940           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1941           .policy = vport_policy,
1942           .doit = odp_vport_cmd_del
1943         },
1944         { .cmd = ODP_VPORT_CMD_GET,
1945           .flags = 0,               /* OK for unprivileged users. */
1946           .policy = vport_policy,
1947           .doit = odp_vport_cmd_get,
1948           .dumpit = odp_vport_cmd_dump
1949         },
1950         { .cmd = ODP_VPORT_CMD_SET,
1951           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1952           .policy = vport_policy,
1953           .doit = odp_vport_cmd_set,
1954         },
1955 };
1956
1957 struct genl_family_and_ops {
1958         struct genl_family *family;
1959         struct genl_ops *ops;
1960         int n_ops;
1961         struct genl_multicast_group *group;
1962 };
1963
1964 static const struct genl_family_and_ops dp_genl_families[] = {
1965         { &dp_datapath_genl_family,
1966           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1967           &dp_datapath_multicast_group },
1968         { &dp_vport_genl_family,
1969           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1970           &dp_vport_multicast_group },
1971         { &dp_flow_genl_family,
1972           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1973           &dp_flow_multicast_group },
1974         { &dp_packet_genl_family,
1975           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1976           NULL },
1977 };
1978
1979 static void dp_unregister_genl(int n_families)
1980 {
1981         int i;
1982
1983         for (i = 0; i < n_families; i++) {
1984                 genl_unregister_family(dp_genl_families[i].family);
1985         }
1986 }
1987
1988 static int dp_register_genl(void)
1989 {
1990         int n_registered;
1991         int err;
1992         int i;
1993
1994         n_registered = 0;
1995         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1996                 const struct genl_family_and_ops *f = &dp_genl_families[i];
1997
1998                 err = genl_register_family_with_ops(f->family, f->ops,
1999                                                     f->n_ops);
2000                 if (err)
2001                         goto error;
2002                 n_registered++;
2003
2004                 if (f->group) {
2005                         err = genl_register_mc_group(f->family, f->group);
2006                         if (err)
2007                                 goto error;
2008                 }
2009         }
2010
2011         err = packet_register_mc_groups();
2012         if (err)
2013                 goto error;
2014         return 0;
2015
2016 error:
2017         dp_unregister_genl(n_registered);
2018         return err;
2019 }
2020
2021 static int __init dp_init(void)
2022 {
2023         struct sk_buff *dummy_skb;
2024         int err;
2025
2026         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2027
2028         printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2029
2030         err = flow_init();
2031         if (err)
2032                 goto error;
2033
2034         err = vport_init();
2035         if (err)
2036                 goto error_flow_exit;
2037
2038         err = register_netdevice_notifier(&dp_device_notifier);
2039         if (err)
2040                 goto error_vport_exit;
2041
2042         err = dp_register_genl();
2043         if (err < 0)
2044                 goto error_unreg_notifier;
2045
2046         return 0;
2047
2048 error_unreg_notifier:
2049         unregister_netdevice_notifier(&dp_device_notifier);
2050 error_vport_exit:
2051         vport_exit();
2052 error_flow_exit:
2053         flow_exit();
2054 error:
2055         return err;
2056 }
2057
2058 static void dp_cleanup(void)
2059 {
2060         rcu_barrier();
2061         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2062         unregister_netdevice_notifier(&dp_device_notifier);
2063         vport_exit();
2064         flow_exit();
2065 }
2066
2067 module_init(dp_init);
2068 module_exit(dp_cleanup);
2069
2070 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2071 MODULE_LICENSE("GPL");