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 "vport-internal_dev.h"
  55
  56 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
  57 EXPORT_SYMBOL(dp_ioctl_hook);
  58
  59 /**
  60  * DOC: Locking:
  61  *
  62  * Writes to device state (add/remove datapath, port, set operations on vports,
  63  * etc.) are protected by RTNL.
  64  *
  65  * Writes to other state (flow table modifications, set miscellaneous datapath
  66  * parameters such as drop frags, etc.) are protected by genl_mutex.  The RTNL
  67  * lock nests inside genl_mutex.
  68  *
  69  * Reads are protected by RCU.
  70  *
  71  * There are a few special cases (mostly stats) that have their own
  72  * synchronization but they nest under all of above and don't interact with
  73  * each other.
  74  */
  75
  76 /* Protected by genl_mutex. */
  77 static struct datapath __rcu *dps[256];
  78
  79 static struct vport *new_vport(const struct vport_parms *);
  80
  81 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  82 struct datapath *get_dp(int dp_idx)
  83 {
  84         if (dp_idx < 0 || dp_idx >= ARRAY_SIZE(dps))
  85                 return NULL;
  86
  87         return rcu_dereference_check(dps[dp_idx], rcu_read_lock_held() ||
  88                                          lockdep_rtnl_is_held() ||
  89                                          lockdep_genl_is_held());
  90 }
  91 EXPORT_SYMBOL_GPL(get_dp);
  92
  93 /* Must be called with genl_mutex. */
  94 static struct tbl *get_table_protected(struct datapath *dp)
  95 {
  96         return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
  97 }
  98
  99 /* Must be called with rcu_read_lock or RTNL lock. */
 100 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
 101 {
 102         return rcu_dereference_rtnl(dp->ports[port_no]);
 103 }
 104
 105 /* Must be called with rcu_read_lock or RTNL lock. */
 106 const char *dp_name(const struct datapath *dp)
 107 {
 108         return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL]));
 109 }
 110
 111 static inline size_t br_nlmsg_size(void)
 112 {
 113         return NLMSG_ALIGN(sizeof(struct ifinfomsg))
 114                + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
 115                + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
 116                + nla_total_size(4) /* IFLA_MASTER */
 117                + nla_total_size(4) /* IFLA_MTU */
 118                + nla_total_size(4) /* IFLA_LINK */
 119                + nla_total_size(1); /* IFLA_OPERSTATE */
 120 }
 121
 122 /* Caller must hold RTNL lock. */
 123 static int dp_fill_ifinfo(struct sk_buff *skb,
 124                           const struct vport *port,
 125                           int event, unsigned int flags)
 126 {
 127         struct datapath *dp = port->dp;
 128         int ifindex = vport_get_ifindex(port);
 129         int iflink = vport_get_iflink(port);
 130         struct ifinfomsg *hdr;
 131         struct nlmsghdr *nlh;
 132
 133         if (ifindex < 0)
 134                 return ifindex;
 135
 136         if (iflink < 0)
 137                 return iflink;
 138
 139         nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
 140         if (nlh == NULL)
 141                 return -EMSGSIZE;
 142
 143         hdr = nlmsg_data(nlh);
 144         hdr->ifi_family = AF_BRIDGE;
 145         hdr->__ifi_pad = 0;
 146         hdr->ifi_type = ARPHRD_ETHER;
 147         hdr->ifi_index = ifindex;
 148         hdr->ifi_flags = vport_get_flags(port);
 149         hdr->ifi_change = 0;
 150
 151         NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
 152         NLA_PUT_U32(skb, IFLA_MASTER,
 153                 vport_get_ifindex(get_vport_protected(dp, ODPP_LOCAL)));
 154         NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
 155 #ifdef IFLA_OPERSTATE
 156         NLA_PUT_U8(skb, IFLA_OPERSTATE,
 157                    vport_is_running(port)
 158                         ? vport_get_operstate(port)
 159                         : IF_OPER_DOWN);
 160 #endif
 161
 162         NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
 163
 164         if (ifindex != iflink)
 165                 NLA_PUT_U32(skb, IFLA_LINK,iflink);
 166
 167         return nlmsg_end(skb, nlh);
 168
 169 nla_put_failure:
 170         nlmsg_cancel(skb, nlh);
 171         return -EMSGSIZE;
 172 }
 173
 174 /* Caller must hold RTNL lock. */
 175 static void dp_ifinfo_notify(int event, struct vport *port)
 176 {
 177         struct sk_buff *skb;
 178         int err = -ENOBUFS;
 179
 180         skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
 181         if (skb == NULL)
 182                 goto errout;
 183
 184         err = dp_fill_ifinfo(skb, port, event, 0);
 185         if (err < 0) {
 186                 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
 187                 WARN_ON(err == -EMSGSIZE);
 188                 kfree_skb(skb);
 189                 goto errout;
 190         }
 191         rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 192         return;
 193 errout:
 194         if (err < 0)
 195                 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
 196 }
 197
 198 static void release_dp(struct kobject *kobj)
 199 {
 200         struct datapath *dp = container_of(kobj, struct datapath, ifobj);
 201         kfree(dp);
 202 }
 203
 204 static struct kobj_type dp_ktype = {
 205         .release = release_dp
 206 };
 207
 208 static void destroy_dp_rcu(struct rcu_head *rcu)
 209 {
 210         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 211
 212         tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl);
 213         free_percpu(dp->stats_percpu);
 214         kobject_put(&dp->ifobj);
 215 }
 216
 217 /* Called with RTNL lock and genl_lock. */
 218 static struct vport *new_vport(const struct vport_parms *parms)
 219 {
 220         struct vport *vport;
 221
 222         vport = vport_add(parms);
 223         if (!IS_ERR(vport)) {
 224                 struct datapath *dp = parms->dp;
 225
 226                 rcu_assign_pointer(dp->ports[parms->port_no], vport);
 227                 list_add(&vport->node, &dp->port_list);
 228
 229                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 230         }
 231
 232         return vport;
 233 }
 234
 235 /* Called with RTNL lock. */
 236 int dp_detach_port(struct vport *p)
 237 {
 238         ASSERT_RTNL();
 239
 240         if (p->port_no != ODPP_LOCAL)
 241                 dp_sysfs_del_if(p);
 242         dp_ifinfo_notify(RTM_DELLINK, p);
 243
 244         /* First drop references to device. */
 245         list_del(&p->node);
 246         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 247
 248         /* Then destroy it. */
 249         return vport_del(p);
 250 }
 251
 252 /* Must be called with rcu_read_lock. */
 253 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 254 {
 255         struct datapath *dp = p->dp;
 256         struct dp_stats_percpu *stats;
 257         int stats_counter_off;
 258         struct sw_flow_actions *acts;
 259         struct loop_counter *loop;
 260         int error;
 261
 262         OVS_CB(skb)->vport = p;
 263
 264         if (!OVS_CB(skb)->flow) {
 265                 struct sw_flow_key key;
 266                 struct tbl_node *flow_node;
 267                 bool is_frag;
 268
 269                 /* Extract flow from 'skb' into 'key'. */
 270                 error = flow_extract(skb, p->port_no, &key, &is_frag);
 271                 if (unlikely(error)) {
 272                         kfree_skb(skb);
 273                         return;
 274                 }
 275
 276                 if (is_frag && dp->drop_frags) {
 277                         kfree_skb(skb);
 278                         stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
 279                         goto out;
 280                 }
 281
 282                 /* Look up flow. */
 283                 flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
 284                                         flow_hash(&key), flow_cmp);
 285                 if (unlikely(!flow_node)) {
 286                         struct dp_upcall_info upcall;
 287
 288                         upcall.cmd = ODP_PACKET_CMD_MISS;
 289                         upcall.key = &key;
 290                         upcall.userdata = 0;
 291                         upcall.sample_pool = 0;
 292                         upcall.actions = NULL;
 293                         upcall.actions_len = 0;
 294                         dp_upcall(dp, skb, &upcall);
 295                         stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
 296                         goto out;
 297                 }
 298
 299                 OVS_CB(skb)->flow = flow_cast(flow_node);
 300         }
 301
 302         stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
 303         flow_used(OVS_CB(skb)->flow, skb);
 304
 305         acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
 306
 307         /* Check whether we've looped too much. */
 308         loop = loop_get_counter();
 309         if (unlikely(++loop->count > MAX_LOOPS))
 310                 loop->looping = true;
 311         if (unlikely(loop->looping)) {
 312                 loop_suppress(dp, acts);
 313                 kfree_skb(skb);
 314                 goto out_loop;
 315         }
 316
 317         /* Execute actions. */
 318         execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
 319                         acts->actions_len);
 320
 321         /* Check whether sub-actions looped too much. */
 322         if (unlikely(loop->looping))
 323                 loop_suppress(dp, acts);
 324
 325 out_loop:
 326         /* Decrement loop counter. */
 327         if (!--loop->count)
 328                 loop->looping = false;
 329         loop_put_counter();
 330
 331 out:
 332         /* Update datapath statistics. */
 333         local_bh_disable();
 334         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 335
 336         write_seqcount_begin(&stats->seqlock);
 337         (*(u64 *)((u8 *)stats + stats_counter_off))++;
 338         write_seqcount_end(&stats->seqlock);
 339
 340         local_bh_enable();
 341 }
 342
 343 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
 344 {
 345         u16 csum_start, csum_offset;
 346         __wsum csum;
 347
 348         get_skb_csum_pointers(skb, &csum_start, &csum_offset);
 349         csum_start -= skb_headroom(skb);
 350         BUG_ON(csum_start >= skb_headlen(skb));
 351
 352         skb_copy_bits(skb, 0, to, csum_start);
 353
 354         csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
 355                                       skb->len - csum_start, 0);
 356         *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
 357 }
 358
 359 static struct genl_family dp_packet_genl_family;
 360 #define PACKET_N_MC_GROUPS 16
 361
 362 static int packet_mc_group(struct datapath *dp, u8 cmd)
 363 {
 364         BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
 365         return jhash_2words(dp->dp_idx, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
 366 }
 367
 368 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
 369  * 'upcall_info'.  There will be only one packet unless we broke up a GSO
 370  * packet.
 371  */
 372 static int queue_control_packets(struct datapath *dp, struct sk_buff *skb,
 373                                  const struct dp_upcall_info *upcall_info)
 374 {
 375         u32 group = packet_mc_group(dp, upcall_info->cmd);
 376         struct sk_buff *nskb;
 377         int port_no;
 378         int err;
 379
 380         if (OVS_CB(skb)->vport)
 381                 port_no = OVS_CB(skb)->vport->port_no;
 382         else
 383                 port_no = ODPP_LOCAL;
 384
 385         do {
 386                 struct odp_header *upcall;
 387                 struct sk_buff *user_skb; /* to be queued to userspace */
 388                 struct nlattr *nla;
 389                 unsigned int len;
 390
 391                 nskb = skb->next;
 392                 skb->next = NULL;
 393
 394                 len = sizeof(struct odp_header);
 395                 len += nla_total_size(4); /* ODP_PACKET_ATTR_TYPE. */
 396                 len += nla_total_size(skb->len);
 397                 len += nla_total_size(FLOW_BUFSIZE);
 398                 if (upcall_info->userdata)
 399                         len += nla_total_size(8);
 400                 if (upcall_info->sample_pool)
 401                         len += nla_total_size(4);
 402                 if (upcall_info->actions_len)
 403                         len += nla_total_size(upcall_info->actions_len);
 404
 405                 user_skb = genlmsg_new(len, GFP_ATOMIC);
 406                 if (!user_skb) {
 407                         netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS);
 408                         goto err_kfree_skbs;
 409                 }
 410
 411                 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
 412                 upcall->dp_idx = dp->dp_idx;
 413
 414                 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY);
 415                 flow_to_nlattrs(upcall_info->key, user_skb);
 416                 nla_nest_end(user_skb, nla);
 417
 418                 if (upcall_info->userdata)
 419                         nla_put_u64(user_skb, ODP_PACKET_ATTR_USERDATA, upcall_info->userdata);
 420                 if (upcall_info->sample_pool)
 421                         nla_put_u32(user_skb, ODP_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
 422                 if (upcall_info->actions_len) {
 423                         const struct nlattr *actions = upcall_info->actions;
 424                         u32 actions_len = upcall_info->actions_len;
 425
 426                         nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_ACTIONS);
 427                         memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
 428                         nla_nest_end(user_skb, nla);
 429                 }
 430
 431                 nla = __nla_reserve(user_skb, ODP_PACKET_ATTR_PACKET, skb->len);
 432                 if (skb->ip_summed == CHECKSUM_PARTIAL)
 433                         copy_and_csum_skb(skb, nla_data(nla));
 434                 else
 435                         skb_copy_bits(skb, 0, nla_data(nla), skb->len);
 436
 437                 err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC);
 438                 if (err)
 439                         goto err_kfree_skbs;
 440
 441                 kfree_skb(skb);
 442                 skb = nskb;
 443         } while (skb);
 444         return 0;
 445
 446 err_kfree_skbs:
 447         kfree_skb(skb);
 448         while ((skb = nskb) != NULL) {
 449                 nskb = skb->next;
 450                 kfree_skb(skb);
 451         }
 452         return err;
 453 }
 454
 455 /* Generic Netlink multicast groups for upcalls.
 456  *
 457  * We really want three unique multicast groups per datapath, but we can't even
 458  * get one, because genl_register_mc_group() takes genl_lock, which is also
 459  * held during Generic Netlink message processing, so trying to acquire
 460  * multicast groups during ODP_DP_NEW processing deadlocks.  Instead, we
 461  * preallocate a few groups and use them round-robin for datapaths.  Collision
 462  * isn't fatal--multicast listeners should check that the family is the one
 463  * that they want and discard others--but it wastes time and memory to receive
 464  * unwanted messages.
 465  */
 466 static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
 467
 468 static struct genl_family dp_packet_genl_family = {
 469         .id = GENL_ID_GENERATE,
 470         .hdrsize = sizeof(struct odp_header),
 471         .name = ODP_PACKET_FAMILY,
 472         .version = 1,
 473         .maxattr = ODP_PACKET_ATTR_MAX
 474 };
 475
 476 static int packet_register_mc_groups(void)
 477 {
 478         int i;
 479
 480         for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
 481                 struct genl_multicast_group *group = &packet_mc_groups[i];
 482                 int error;
 483
 484                 sprintf(group->name, "packet%d", i);
 485                 error = genl_register_mc_group(&dp_packet_genl_family, group);
 486                 if (error)
 487                         return error;
 488         }
 489         return 0;
 490 }
 491
 492 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
 493 {
 494         struct dp_stats_percpu *stats;
 495         int err;
 496
 497         WARN_ON_ONCE(skb_shared(skb));
 498
 499         forward_ip_summed(skb);
 500
 501         err = vswitch_skb_checksum_setup(skb);
 502         if (err)
 503                 goto err_kfree_skb;
 504
 505         /* Break apart GSO packets into their component pieces.  Otherwise
 506          * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
 507         if (skb_is_gso(skb)) {
 508                 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 509
 510                 kfree_skb(skb);
 511                 skb = nskb;
 512                 if (IS_ERR(skb)) {
 513                         err = PTR_ERR(skb);
 514                         goto err;
 515                 }
 516         }
 517
 518         return queue_control_packets(dp, skb, upcall_info);
 519
 520 err_kfree_skb:
 521         kfree_skb(skb);
 522 err:
 523         local_bh_disable();
 524         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 525
 526         write_seqcount_begin(&stats->seqlock);
 527         stats->n_lost++;
 528         write_seqcount_end(&stats->seqlock);
 529
 530         local_bh_enable();
 531
 532         return err;
 533 }
 534
 535 /* Called with genl_mutex. */
 536 static int flush_flows(int dp_idx)
 537 {
 538         struct tbl *old_table;
 539         struct tbl *new_table;
 540         struct datapath *dp;
 541
 542         dp = get_dp(dp_idx);
 543         if (!dp)
 544                 return -ENODEV;
 545
 546         old_table = get_table_protected(dp);
 547         new_table = tbl_create(TBL_MIN_BUCKETS);
 548         if (!new_table)
 549                 return -ENOMEM;
 550
 551         rcu_assign_pointer(dp->table, new_table);
 552
 553         tbl_deferred_destroy(old_table, flow_free_tbl);
 554
 555         return 0;
 556 }
 557
 558 static int validate_actions(const struct nlattr *attr)
 559 {
 560         const struct nlattr *a;
 561         int rem;
 562
 563         nla_for_each_nested(a, attr, rem) {
 564                 static const u32 action_lens[ODPAT_MAX + 1] = {
 565                         [ODPAT_OUTPUT] = 4,
 566                         [ODPAT_CONTROLLER] = 8,
 567                         [ODPAT_SET_DL_TCI] = 2,
 568                         [ODPAT_STRIP_VLAN] = 0,
 569                         [ODPAT_SET_DL_SRC] = ETH_ALEN,
 570                         [ODPAT_SET_DL_DST] = ETH_ALEN,
 571                         [ODPAT_SET_NW_SRC] = 4,
 572                         [ODPAT_SET_NW_DST] = 4,
 573                         [ODPAT_SET_NW_TOS] = 1,
 574                         [ODPAT_SET_TP_SRC] = 2,
 575                         [ODPAT_SET_TP_DST] = 2,
 576                         [ODPAT_SET_TUNNEL] = 8,
 577                         [ODPAT_SET_PRIORITY] = 4,
 578                         [ODPAT_POP_PRIORITY] = 0,
 579                         [ODPAT_DROP_SPOOFED_ARP] = 0,
 580                 };
 581                 int type = nla_type(a);
 582
 583                 if (type > ODPAT_MAX || nla_len(a) != action_lens[type])
 584                         return -EINVAL;
 585
 586                 switch (type) {
 587                 case ODPAT_UNSPEC:
 588                         return -EINVAL;
 589
 590                 case ODPAT_CONTROLLER:
 591                 case ODPAT_STRIP_VLAN:
 592                 case ODPAT_SET_DL_SRC:
 593                 case ODPAT_SET_DL_DST:
 594                 case ODPAT_SET_NW_SRC:
 595                 case ODPAT_SET_NW_DST:
 596                 case ODPAT_SET_TP_SRC:
 597                 case ODPAT_SET_TP_DST:
 598                 case ODPAT_SET_TUNNEL:
 599                 case ODPAT_SET_PRIORITY:
 600                 case ODPAT_POP_PRIORITY:
 601                 case ODPAT_DROP_SPOOFED_ARP:
 602                         /* No validation needed. */
 603                         break;
 604
 605                 case ODPAT_OUTPUT:
 606                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 607                                 return -EINVAL;
 608                         break;
 609
 610                 case ODPAT_SET_DL_TCI:
 611                         if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
 612                                 return -EINVAL;
 613                         break;
 614
 615                 case ODPAT_SET_NW_TOS:
 616                         if (nla_get_u8(a) & INET_ECN_MASK)
 617                                 return -EINVAL;
 618                         break;
 619
 620                 default:
 621                         return -EOPNOTSUPP;
 622                 }
 623         }
 624
 625         if (rem > 0)
 626                 return -EINVAL;
 627
 628         return 0;
 629 }
 630 static void clear_stats(struct sw_flow *flow)
 631 {
 632         flow->used = 0;
 633         flow->tcp_flags = 0;
 634         flow->packet_count = 0;
 635         flow->byte_count = 0;
 636 }
 637
 638 /* Called with genl_mutex. */
 639 static int expand_table(struct datapath *dp)
 640 {
 641         struct tbl *old_table = get_table_protected(dp);
 642         struct tbl *new_table;
 643
 644         new_table = tbl_expand(old_table);
 645         if (IS_ERR(new_table))
 646                 return PTR_ERR(new_table);
 647
 648         rcu_assign_pointer(dp->table, new_table);
 649         tbl_deferred_destroy(old_table, NULL);
 650
 651         return 0;
 652 }
 653
 654 static int odp_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 655 {
 656         struct odp_header *odp_header = info->userhdr;
 657         struct nlattr **a = info->attrs;
 658         struct sk_buff *packet;
 659         struct sw_flow_key key;
 660         struct datapath *dp;
 661         struct ethhdr *eth;
 662         bool is_frag;
 663         int err;
 664
 665         err = -EINVAL;
 666         if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_ACTIONS] ||
 667             nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN)
 668                 goto exit;
 669
 670         err = validate_actions(a[ODP_PACKET_ATTR_ACTIONS]);
 671         if (err)
 672                 goto exit;
 673
 674         packet = skb_clone(skb, GFP_KERNEL);
 675         err = -ENOMEM;
 676         if (!packet)
 677                 goto exit;
 678         packet->data = nla_data(a[ODP_PACKET_ATTR_PACKET]);
 679         packet->len = nla_len(a[ODP_PACKET_ATTR_PACKET]);
 680
 681         skb_reset_mac_header(packet);
 682         eth = eth_hdr(packet);
 683
 684         /* Normally, setting the skb 'protocol' field would be handled by a
 685          * call to eth_type_trans(), but it assumes there's a sending
 686          * device, which we may not have. */
 687         if (ntohs(eth->h_proto) >= 1536)
 688                 packet->protocol = eth->h_proto;
 689         else
 690                 packet->protocol = htons(ETH_P_802_2);
 691
 692         err = flow_extract(packet, -1, &key, &is_frag);
 693         if (err)
 694                 goto exit;
 695
 696         rcu_read_lock();
 697         dp = get_dp(odp_header->dp_idx);
 698         err = -ENODEV;
 699         if (dp)
 700                 err = execute_actions(dp, packet, &key,
 701                                       nla_data(a[ODP_PACKET_ATTR_ACTIONS]),
 702                                       nla_len(a[ODP_PACKET_ATTR_ACTIONS]));
 703         rcu_read_unlock();
 704
 705 exit:
 706         return err;
 707 }
 708
 709 static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = {
 710         [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 711         [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 712 };
 713
 714 static struct genl_ops dp_packet_genl_ops[] = {
 715         { .cmd = ODP_PACKET_CMD_EXECUTE,
 716           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 717           .policy = packet_policy,
 718           .doit = odp_packet_cmd_execute
 719         }
 720 };
 721
 722 static void get_dp_stats(struct datapath *dp, struct odp_stats *stats)
 723 {
 724         int i;
 725
 726         stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
 727         for_each_possible_cpu(i) {
 728                 const struct dp_stats_percpu *percpu_stats;
 729                 struct dp_stats_percpu local_stats;
 730                 unsigned seqcount;
 731
 732                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 733
 734                 do {
 735                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 736                         local_stats = *percpu_stats;
 737                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 738
 739                 stats->n_frags += local_stats.n_frags;
 740                 stats->n_hit += local_stats.n_hit;
 741                 stats->n_missed += local_stats.n_missed;
 742                 stats->n_lost += local_stats.n_lost;
 743         }
 744 }
 745
 746 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
 747  * Called with RTNL lock.
 748  */
 749 int dp_min_mtu(const struct datapath *dp)
 750 {
 751         struct vport *p;
 752         int mtu = 0;
 753
 754         ASSERT_RTNL();
 755
 756         list_for_each_entry (p, &dp->port_list, node) {
 757                 int dev_mtu;
 758
 759                 /* Skip any internal ports, since that's what we're trying to
 760                  * set. */
 761                 if (is_internal_vport(p))
 762                         continue;
 763
 764                 dev_mtu = vport_get_mtu(p);
 765                 if (!mtu || dev_mtu < mtu)
 766                         mtu = dev_mtu;
 767         }
 768
 769         return mtu ? mtu : ETH_DATA_LEN;
 770 }
 771
 772 /* Sets the MTU of all datapath devices to the minimum of the ports
 773  * Called with RTNL lock.
 774  */
 775 void set_internal_devs_mtu(const struct datapath *dp)
 776 {
 777         struct vport *p;
 778         int mtu;
 779
 780         ASSERT_RTNL();
 781
 782         mtu = dp_min_mtu(dp);
 783
 784         list_for_each_entry (p, &dp->port_list, node) {
 785                 if (is_internal_vport(p))
 786                         vport_set_mtu(p, mtu);
 787         }
 788 }
 789
 790 static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = {
 791         [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 792         [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 793         [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 794 };
 795
 796 static struct genl_family dp_flow_genl_family = {
 797         .id = GENL_ID_GENERATE,
 798         .hdrsize = sizeof(struct odp_header),
 799         .name = ODP_FLOW_FAMILY,
 800         .version = 1,
 801         .maxattr = ODP_FLOW_ATTR_MAX
 802 };
 803
 804 static struct genl_multicast_group dp_flow_multicast_group = {
 805         .name = ODP_FLOW_MCGROUP
 806 };
 807
 808 /* Called with genl_lock. */
 809 static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 810                                   struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
 811 {
 812         const int skb_orig_len = skb->len;
 813         const struct sw_flow_actions *sf_acts;
 814         struct odp_flow_stats stats;
 815         struct odp_header *odp_header;
 816         struct nlattr *nla;
 817         unsigned long used;
 818         u8 tcp_flags;
 819         int nla_len;
 820         int err;
 821
 822         sf_acts = rcu_dereference_protected(flow->sf_acts,
 823                                             lockdep_genl_is_held());
 824
 825         odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 826         if (!odp_header)
 827                 return -EMSGSIZE;
 828
 829         odp_header->dp_idx = dp->dp_idx;
 830
 831         nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY);
 832         if (!nla)
 833                 goto nla_put_failure;
 834         err = flow_to_nlattrs(&flow->key, skb);
 835         if (err)
 836                 goto error;
 837         nla_nest_end(skb, nla);
 838
 839         spin_lock_bh(&flow->lock);
 840         used = flow->used;
 841         stats.n_packets = flow->packet_count;
 842         stats.n_bytes = flow->byte_count;
 843         tcp_flags = flow->tcp_flags;
 844         spin_unlock_bh(&flow->lock);
 845
 846         if (used)
 847                 NLA_PUT_MSECS(skb, ODP_FLOW_ATTR_USED, used);
 848
 849         if (stats.n_packets)
 850                 NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats);
 851
 852         if (tcp_flags)
 853                 NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 854
 855         /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, and this is the first flow to
 856          * be dumped into 'skb', then expand the skb.  This is unusual for
 857          * Netlink but individual action lists can be longer than a page and
 858          * thus entirely undumpable if we didn't do this. */
 859         nla_len = nla_total_size(sf_acts->actions_len);
 860         if (nla_len > skb_tailroom(skb) && !skb_orig_len) {
 861                 int hdr_off = (unsigned char *)odp_header - skb->data;
 862
 863                 err = pskb_expand_head(skb, 0, nla_len - skb_tailroom(skb), GFP_KERNEL);
 864                 if (err)
 865                         goto error;
 866
 867                 odp_header = (struct odp_header *)(skb->data + hdr_off);
 868         }
 869         nla = nla_nest_start(skb, ODP_FLOW_ATTR_ACTIONS);
 870         memcpy(__skb_put(skb, sf_acts->actions_len), sf_acts->actions, sf_acts->actions_len);
 871         nla_nest_end(skb, nla);
 872
 873         return genlmsg_end(skb, odp_header);
 874
 875 nla_put_failure:
 876         err = -EMSGSIZE;
 877 error:
 878         genlmsg_cancel(skb, odp_header);
 879         return err;
 880 }
 881
 882 static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow)
 883 {
 884         const struct sw_flow_actions *sf_acts;
 885         int len;
 886
 887         sf_acts = rcu_dereference_protected(flow->sf_acts,
 888                                             lockdep_genl_is_held());
 889
 890         len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */
 891         len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */
 892         len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */
 893         len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */
 894         len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */
 895         return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL);
 896 }
 897
 898 static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
 899                                                u32 pid, u32 seq, u8 cmd)
 900 {
 901         struct sk_buff *skb;
 902         int retval;
 903
 904         skb = odp_flow_cmd_alloc_info(flow);
 905         if (!skb)
 906                 return ERR_PTR(-ENOMEM);
 907
 908         retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 909         BUG_ON(retval < 0);
 910         return skb;
 911 }
 912
 913 static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 914 {
 915         struct nlattr **a = info->attrs;
 916         struct odp_header *odp_header = info->userhdr;
 917         struct tbl_node *flow_node;
 918         struct sw_flow_key key;
 919         struct sw_flow *flow;
 920         struct sk_buff *reply;
 921         struct datapath *dp;
 922         struct tbl *table;
 923         u32 hash;
 924         int error;
 925
 926         /* Extract key. */
 927         error = -EINVAL;
 928         if (!a[ODP_FLOW_ATTR_KEY])
 929                 goto error;
 930         error = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
 931         if (error)
 932                 goto error;
 933
 934         /* Validate actions. */
 935         if (a[ODP_FLOW_ATTR_ACTIONS]) {
 936                 error = validate_actions(a[ODP_FLOW_ATTR_ACTIONS]);
 937                 if (error)
 938                         goto error;
 939         } else if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW) {
 940                 error = -EINVAL;
 941                 goto error;
 942         }
 943
 944         dp = get_dp(odp_header->dp_idx);
 945         error = -ENODEV;
 946         if (!dp)
 947                 goto error;
 948
 949         hash = flow_hash(&key);
 950         table = get_table_protected(dp);
 951         flow_node = tbl_lookup(table, &key, hash, flow_cmp);
 952         if (!flow_node) {
 953                 struct sw_flow_actions *acts;
 954
 955                 /* Bail out if we're not allowed to create a new flow. */
 956                 error = -ENOENT;
 957                 if (info->genlhdr->cmd == ODP_FLOW_CMD_SET)
 958                         goto error;
 959
 960                 /* Expand table, if necessary, to make room. */
 961                 if (tbl_count(table) >= tbl_n_buckets(table)) {
 962                         error = expand_table(dp);
 963                         if (error)
 964                                 goto error;
 965                         table = get_table_protected(dp);
 966                 }
 967
 968                 /* Allocate flow. */
 969                 flow = flow_alloc();
 970                 if (IS_ERR(flow)) {
 971                         error = PTR_ERR(flow);
 972                         goto error;
 973                 }
 974                 flow->key = key;
 975                 clear_stats(flow);
 976
 977                 /* Obtain actions. */
 978                 acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
 979                 error = PTR_ERR(acts);
 980                 if (IS_ERR(acts))
 981                         goto error_free_flow;
 982                 rcu_assign_pointer(flow->sf_acts, acts);
 983
 984                 /* Put flow in bucket. */
 985                 error = tbl_insert(table, &flow->tbl_node, hash);
 986                 if (error)
 987                         goto error_free_flow;
 988
 989                 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
 990                                                 info->snd_seq, ODP_FLOW_CMD_NEW);
 991         } else {
 992                 /* We found a matching flow. */
 993                 struct sw_flow_actions *old_acts;
 994
 995                 /* Bail out if we're not allowed to modify an existing flow.
 996                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 997                  * because Generic Netlink treats the latter as a dump
 998                  * request.  We also accept NLM_F_EXCL in case that bug ever
 999                  * gets fixed.
1000                  */
1001                 error = -EEXIST;
1002                 if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW &&
1003                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1004                         goto error;
1005
1006                 /* Update actions. */
1007                 flow = flow_cast(flow_node);
1008                 old_acts = rcu_dereference_protected(flow->sf_acts,
1009                                                      lockdep_genl_is_held());
1010                 if (a[ODP_FLOW_ATTR_ACTIONS] &&
1011                     (old_acts->actions_len != nla_len(a[ODP_FLOW_ATTR_ACTIONS]) ||
1012                      memcmp(old_acts->actions, nla_data(a[ODP_FLOW_ATTR_ACTIONS]),
1013                             old_acts->actions_len))) {
1014                         struct sw_flow_actions *new_acts;
1015
1016                         new_acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1017                         error = PTR_ERR(new_acts);
1018                         if (IS_ERR(new_acts))
1019                                 goto error;
1020
1021                         rcu_assign_pointer(flow->sf_acts, new_acts);
1022                         flow_deferred_free_acts(old_acts);
1023                 }
1024
1025                 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1026                                                 info->snd_seq, ODP_FLOW_CMD_NEW);
1027
1028                 /* Clear stats. */
1029                 if (a[ODP_FLOW_ATTR_CLEAR]) {
1030                         spin_lock_bh(&flow->lock);
1031                         clear_stats(flow);
1032                         spin_unlock_bh(&flow->lock);
1033                 }
1034         }
1035
1036         if (!IS_ERR(reply))
1037                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1038                             dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1039         else
1040                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1041                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1042         return 0;
1043
1044 error_free_flow:
1045         flow_put(flow);
1046 error:
1047         return error;
1048 }
1049
1050 static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1051 {
1052         struct nlattr **a = info->attrs;
1053         struct odp_header *odp_header = info->userhdr;
1054         struct sw_flow_key key;
1055         struct tbl_node *flow_node;
1056         struct sk_buff *reply;
1057         struct sw_flow *flow;
1058         struct datapath *dp;
1059         struct tbl *table;
1060         int err;
1061
1062         if (!a[ODP_FLOW_ATTR_KEY])
1063                 return -EINVAL;
1064         err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1065         if (err)
1066                 return err;
1067
1068         dp = get_dp(odp_header->dp_idx);
1069         if (!dp)
1070                 return -ENODEV;
1071
1072         table = get_table_protected(dp);
1073         flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1074         if (!flow_node)
1075                 return -ENOENT;
1076
1077         flow = flow_cast(flow_node);
1078         reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW);
1079         if (IS_ERR(reply))
1080                 return PTR_ERR(reply);
1081
1082         return genlmsg_reply(reply, info);
1083 }
1084
1085 static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1086 {
1087         struct nlattr **a = info->attrs;
1088         struct odp_header *odp_header = info->userhdr;
1089         struct sw_flow_key key;
1090         struct tbl_node *flow_node;
1091         struct sk_buff *reply;
1092         struct sw_flow *flow;
1093         struct datapath *dp;
1094         struct tbl *table;
1095         int err;
1096
1097         if (!a[ODP_FLOW_ATTR_KEY])
1098                 return flush_flows(odp_header->dp_idx);
1099         err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1100         if (err)
1101                 return err;
1102
1103         dp = get_dp(odp_header->dp_idx);
1104         if (!dp)
1105                 return -ENODEV;
1106
1107         table = get_table_protected(dp);
1108         flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1109         if (!flow_node)
1110                 return -ENOENT;
1111         flow = flow_cast(flow_node);
1112
1113         reply = odp_flow_cmd_alloc_info(flow);
1114         if (!reply)
1115                 return -ENOMEM;
1116
1117         err = tbl_remove(table, flow_node);
1118         if (err) {
1119                 kfree_skb(reply);
1120                 return err;
1121         }
1122
1123         err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1124                                      info->snd_seq, 0, ODP_FLOW_CMD_DEL);
1125         BUG_ON(err < 0);
1126
1127         flow_deferred_free(flow);
1128
1129         genl_notify(reply, genl_info_net(info), info->snd_pid,
1130                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1131         return 0;
1132 }
1133
1134 static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1135 {
1136         struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1137         struct datapath *dp;
1138
1139         dp = get_dp(odp_header->dp_idx);
1140         if (!dp)
1141                 return -ENODEV;
1142
1143         for (;;) {
1144                 struct tbl_node *flow_node;
1145                 struct sw_flow *flow;
1146                 u32 bucket, obj;
1147
1148                 bucket = cb->args[0];
1149                 obj = cb->args[1];
1150                 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1151                 if (!flow_node)
1152                         break;
1153
1154                 flow = flow_cast(flow_node);
1155                 if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1156                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1157                                            ODP_FLOW_CMD_NEW) < 0)
1158                         break;
1159
1160                 cb->args[0] = bucket;
1161                 cb->args[1] = obj;
1162         }
1163         return skb->len;
1164 }
1165
1166 static struct genl_ops dp_flow_genl_ops[] = {
1167         { .cmd = ODP_FLOW_CMD_NEW,
1168           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1169           .policy = flow_policy,
1170           .doit = odp_flow_cmd_new_or_set
1171         },
1172         { .cmd = ODP_FLOW_CMD_DEL,
1173           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1174           .policy = flow_policy,
1175           .doit = odp_flow_cmd_del
1176         },
1177         { .cmd = ODP_FLOW_CMD_GET,
1178           .flags = 0,               /* OK for unprivileged users. */
1179           .policy = flow_policy,
1180           .doit = odp_flow_cmd_get,
1181           .dumpit = odp_flow_cmd_dump
1182         },
1183         { .cmd = ODP_FLOW_CMD_SET,
1184           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1185           .policy = flow_policy,
1186           .doit = odp_flow_cmd_new_or_set,
1187         },
1188 };
1189
1190 static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
1191 #ifdef HAVE_NLA_NUL_STRING
1192         [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1193 #endif
1194         [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1195         [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1196 };
1197
1198 static struct genl_family dp_datapath_genl_family = {
1199         .id = GENL_ID_GENERATE,
1200         .hdrsize = sizeof(struct odp_header),
1201         .name = ODP_DATAPATH_FAMILY,
1202         .version = 1,
1203         .maxattr = ODP_DP_ATTR_MAX
1204 };
1205
1206 static struct genl_multicast_group dp_datapath_multicast_group = {
1207         .name = ODP_DATAPATH_MCGROUP
1208 };
1209
1210 static int odp_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1211                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1212 {
1213         struct odp_header *odp_header;
1214         struct nlattr *nla;
1215         int err;
1216
1217         odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1218                                    flags, cmd);
1219         if (!odp_header)
1220                 goto error;
1221
1222         odp_header->dp_idx = dp->dp_idx;
1223
1224         rcu_read_lock();
1225         err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
1226         rcu_read_unlock();
1227         if (err)
1228                 goto nla_put_failure;
1229
1230         nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats));
1231         if (!nla)
1232                 goto nla_put_failure;
1233         get_dp_stats(dp, nla_data(nla));
1234
1235         NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS,
1236                     dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO);
1237
1238         if (dp->sflow_probability)
1239                 NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability);
1240
1241         nla = nla_nest_start(skb, ODP_DP_ATTR_MCGROUPS);
1242         if (!nla)
1243                 goto nla_put_failure;
1244         NLA_PUT_U32(skb, ODP_PACKET_CMD_MISS, packet_mc_group(dp, ODP_PACKET_CMD_MISS));
1245         NLA_PUT_U32(skb, ODP_PACKET_CMD_ACTION, packet_mc_group(dp, ODP_PACKET_CMD_ACTION));
1246         NLA_PUT_U32(skb, ODP_PACKET_CMD_SAMPLE, packet_mc_group(dp, ODP_PACKET_CMD_SAMPLE));
1247         nla_nest_end(skb, nla);
1248
1249         return genlmsg_end(skb, odp_header);
1250
1251 nla_put_failure:
1252         genlmsg_cancel(skb, odp_header);
1253 error:
1254         return -EMSGSIZE;
1255 }
1256
1257 static struct sk_buff *odp_dp_cmd_build_info(struct datapath *dp, u32 pid,
1258                                              u32 seq, u8 cmd)
1259 {
1260         struct sk_buff *skb;
1261         int retval;
1262
1263         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1264         if (!skb)
1265                 return ERR_PTR(-ENOMEM);
1266
1267         retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1268         if (retval < 0) {
1269                 kfree_skb(skb);
1270                 return ERR_PTR(retval);
1271         }
1272         return skb;
1273 }
1274
1275 static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1276 {
1277         if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1278                 u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1279
1280                 if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP)
1281                         return -EINVAL;
1282         }
1283
1284         return VERIFY_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
1285 }
1286
1287 /* Called with genl_mutex and optionally with RTNL lock also. */
1288 static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1289 {
1290         if (!a[ODP_DP_ATTR_NAME]) {
1291                 struct datapath *dp = get_dp(odp_header->dp_idx);
1292                 if (!dp)
1293                         return ERR_PTR(-ENODEV);
1294                 return dp;
1295         } else {
1296                 struct vport *vport;
1297                 int dp_idx;
1298
1299                 rcu_read_lock();
1300                 vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
1301                 dp_idx = vport && vport->port_no == ODPP_LOCAL ? vport->dp->dp_idx : -1;
1302                 rcu_read_unlock();
1303
1304                 if (dp_idx < 0)
1305                         return ERR_PTR(-ENODEV);
1306                 return vport->dp;
1307         }
1308 }
1309
1310 /* Called with genl_mutex. */
1311 static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1312 {
1313         if (a[ODP_DP_ATTR_IPV4_FRAGS])
1314                 dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP;
1315         if (a[ODP_DP_ATTR_SAMPLING])
1316                 dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]);
1317 }
1318
1319 static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1320 {
1321         struct nlattr **a = info->attrs;
1322         struct odp_header *odp_header = info->userhdr;
1323         struct vport_parms parms;
1324         struct sk_buff *reply;
1325         struct datapath *dp;
1326         struct vport *vport;
1327         int dp_idx;
1328         int err;
1329
1330         err = -EINVAL;
1331         if (!a[ODP_DP_ATTR_NAME])
1332                 goto err;
1333
1334         err = odp_dp_cmd_validate(a);
1335         if (err)
1336                 goto err;
1337
1338         rtnl_lock();
1339         err = -ENODEV;
1340         if (!try_module_get(THIS_MODULE))
1341                 goto err_unlock_rtnl;
1342
1343         dp_idx = odp_header->dp_idx;
1344         if (dp_idx < 0) {
1345                 err = -EFBIG;
1346                 for (dp_idx = 0; dp_idx < ARRAY_SIZE(dps); dp_idx++) {
1347                         if (get_dp(dp_idx))
1348                                 continue;
1349                         err = 0;
1350                         break;
1351                 }
1352         } else if (dp_idx < ARRAY_SIZE(dps))
1353                 err = get_dp(dp_idx) ? -EBUSY : 0;
1354         else
1355                 err = -EINVAL;
1356         if (err)
1357                 goto err_put_module;
1358
1359         err = -ENOMEM;
1360         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1361         if (dp == NULL)
1362                 goto err_put_module;
1363         INIT_LIST_HEAD(&dp->port_list);
1364         dp->dp_idx = dp_idx;
1365
1366         /* Initialize kobject for bridge.  This will be added as
1367          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1368         dp->ifobj.kset = NULL;
1369         kobject_init(&dp->ifobj, &dp_ktype);
1370
1371         /* Allocate table. */
1372         err = -ENOMEM;
1373         rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1374         if (!dp->table)
1375                 goto err_free_dp;
1376
1377         /* Set up our datapath device. */
1378         parms.name = nla_data(a[ODP_DP_ATTR_NAME]);
1379         parms.type = ODP_VPORT_TYPE_INTERNAL;
1380         parms.options = NULL;
1381         parms.dp = dp;
1382         parms.port_no = ODPP_LOCAL;
1383         vport = new_vport(&parms);
1384         if (IS_ERR(vport)) {
1385                 err = PTR_ERR(vport);
1386                 if (err == -EBUSY)
1387                         err = -EEXIST;
1388
1389                 goto err_destroy_table;
1390         }
1391
1392         dp->drop_frags = 0;
1393         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1394         if (!dp->stats_percpu) {
1395                 err = -ENOMEM;
1396                 goto err_destroy_local_port;
1397         }
1398
1399         change_datapath(dp, a);
1400
1401         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1402         err = PTR_ERR(reply);
1403         if (IS_ERR(reply))
1404                 goto err_destroy_local_port;
1405
1406         rcu_assign_pointer(dps[dp_idx], dp);
1407         dp_sysfs_add_dp(dp);
1408
1409         rtnl_unlock();
1410
1411         genl_notify(reply, genl_info_net(info), info->snd_pid,
1412                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1413         return 0;
1414
1415 err_destroy_local_port:
1416         dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1417 err_destroy_table:
1418         tbl_destroy(get_table_protected(dp), NULL);
1419 err_free_dp:
1420         kfree(dp);
1421 err_put_module:
1422         module_put(THIS_MODULE);
1423 err_unlock_rtnl:
1424         rtnl_unlock();
1425 err:
1426         return err;
1427 }
1428
1429 static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1430 {
1431         struct vport *vport, *next_vport;
1432         struct sk_buff *reply;
1433         struct datapath *dp;
1434         int err;
1435
1436         err = odp_dp_cmd_validate(info->attrs);
1437         if (err)
1438                 goto exit;
1439
1440         rtnl_lock();
1441         dp = lookup_datapath(info->userhdr, info->attrs);
1442         err = PTR_ERR(dp);
1443         if (IS_ERR(dp))
1444                 goto exit_unlock;
1445
1446         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_DEL);
1447         err = PTR_ERR(reply);
1448         if (IS_ERR(reply))
1449                 goto exit_unlock;
1450
1451         list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1452                 if (vport->port_no != ODPP_LOCAL)
1453                         dp_detach_port(vport);
1454
1455         dp_sysfs_del_dp(dp);
1456         rcu_assign_pointer(dps[dp->dp_idx], NULL);
1457         dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1458
1459         call_rcu(&dp->rcu, destroy_dp_rcu);
1460         module_put(THIS_MODULE);
1461
1462         genl_notify(reply, genl_info_net(info), info->snd_pid,
1463                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1464         err = 0;
1465
1466 exit_unlock:
1467         rtnl_unlock();
1468 exit:
1469         return err;
1470 }
1471
1472 static int odp_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1473 {
1474         struct sk_buff *reply;
1475         struct datapath *dp;
1476         int err;
1477
1478         err = odp_dp_cmd_validate(info->attrs);
1479         if (err)
1480                 return err;
1481
1482         dp = lookup_datapath(info->userhdr, info->attrs);
1483         if (IS_ERR(dp))
1484                 return PTR_ERR(dp);
1485
1486         change_datapath(dp, info->attrs);
1487
1488         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1489         if (IS_ERR(reply)) {
1490                 err = PTR_ERR(reply);
1491                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1492                                 dp_datapath_multicast_group.id, err);
1493                 return 0;
1494         }
1495
1496         genl_notify(reply, genl_info_net(info), info->snd_pid,
1497                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1498         return 0;
1499 }
1500
1501 static int odp_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1502 {
1503         struct sk_buff *reply;
1504         struct datapath *dp;
1505         int err;
1506
1507         err = odp_dp_cmd_validate(info->attrs);
1508         if (err)
1509                 return err;
1510
1511         dp = lookup_datapath(info->userhdr, info->attrs);
1512         if (IS_ERR(dp))
1513                 return PTR_ERR(dp);
1514
1515         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1516         if (IS_ERR(reply))
1517                 return PTR_ERR(reply);
1518
1519         return genlmsg_reply(reply, info);
1520 }
1521
1522 static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1523 {
1524         u32 dp_idx;
1525
1526         for (dp_idx = cb->args[0]; dp_idx < ARRAY_SIZE(dps); dp_idx++) {
1527                 struct datapath *dp = get_dp(dp_idx);
1528                 if (!dp)
1529                         continue;
1530                 if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1531                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1532                                          ODP_DP_CMD_NEW) < 0)
1533                         break;
1534         }
1535
1536         cb->args[0] = dp_idx;
1537         return skb->len;
1538 }
1539
1540 static struct genl_ops dp_datapath_genl_ops[] = {
1541         { .cmd = ODP_DP_CMD_NEW,
1542           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1543           .policy = datapath_policy,
1544           .doit = odp_dp_cmd_new
1545         },
1546         { .cmd = ODP_DP_CMD_DEL,
1547           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1548           .policy = datapath_policy,
1549           .doit = odp_dp_cmd_del
1550         },
1551         { .cmd = ODP_DP_CMD_GET,
1552           .flags = 0,               /* OK for unprivileged users. */
1553           .policy = datapath_policy,
1554           .doit = odp_dp_cmd_get,
1555           .dumpit = odp_dp_cmd_dump
1556         },
1557         { .cmd = ODP_DP_CMD_SET,
1558           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1559           .policy = datapath_policy,
1560           .doit = odp_dp_cmd_set,
1561         },
1562 };
1563
1564 static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = {
1565 #ifdef HAVE_NLA_NUL_STRING
1566         [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1567         [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1568         [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1569         [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1570         [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1571 #else
1572         [ODP_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1573         [ODP_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1574 #endif
1575         [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1576         [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1577 };
1578
1579 static struct genl_family dp_vport_genl_family = {
1580         .id = GENL_ID_GENERATE,
1581         .hdrsize = sizeof(struct odp_header),
1582         .name = ODP_VPORT_FAMILY,
1583         .version = 1,
1584         .maxattr = ODP_VPORT_ATTR_MAX
1585 };
1586
1587 static struct genl_multicast_group dp_vport_multicast_group = {
1588         .name = ODP_VPORT_MCGROUP
1589 };
1590
1591 /* Called with RTNL lock or RCU read lock. */
1592 static int odp_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1593                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1594 {
1595         struct odp_header *odp_header;
1596         struct nlattr *nla;
1597         int ifindex, iflink;
1598         int err;
1599
1600         odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1601                                  flags, cmd);
1602         if (!odp_header)
1603                 return -EMSGSIZE;
1604
1605         odp_header->dp_idx = vport->dp->dp_idx;
1606
1607         NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1608         NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport));
1609         NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport));
1610
1611         nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1612         if (!nla)
1613                 goto nla_put_failure;
1614         if (vport_get_stats(vport, nla_data(nla)))
1615                 __skb_trim(skb, skb->len - nla->nla_len);
1616
1617         NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1618
1619         NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, vport_get_mtu(vport));
1620
1621         err = vport_get_options(vport, skb);
1622         if (err == -EMSGSIZE)
1623                 goto error;
1624
1625         ifindex = vport_get_ifindex(vport);
1626         if (ifindex > 0)
1627                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex);
1628
1629         iflink = vport_get_iflink(vport);
1630         if (iflink > 0)
1631                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink);
1632
1633         return genlmsg_end(skb, odp_header);
1634
1635 nla_put_failure:
1636         err = -EMSGSIZE;
1637 error:
1638         genlmsg_cancel(skb, odp_header);
1639         return err;
1640 }
1641
1642 /* Called with RTNL lock or RCU read lock. */
1643 static struct sk_buff *odp_vport_cmd_build_info(struct vport *vport, u32 pid,
1644                                                 u32 seq, u8 cmd)
1645 {
1646         struct sk_buff *skb;
1647         int retval;
1648
1649         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1650         if (!skb)
1651                 return ERR_PTR(-ENOMEM);
1652
1653         retval = odp_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1654         if (retval < 0) {
1655                 kfree_skb(skb);
1656                 return ERR_PTR(retval);
1657         }
1658         return skb;
1659 }
1660
1661 static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1662 {
1663         return VERIFY_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1664 }
1665
1666 /* Called with RTNL lock or RCU read lock. */
1667 static struct vport *lookup_vport(struct odp_header *odp_header,
1668                                   struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1669 {
1670         struct datapath *dp;
1671         struct vport *vport;
1672
1673         if (a[ODP_VPORT_ATTR_NAME]) {
1674                 vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME]));
1675                 if (!vport)
1676                         return ERR_PTR(-ENODEV);
1677                 return vport;
1678         } else if (a[ODP_VPORT_ATTR_PORT_NO]) {
1679                 u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1680
1681                 if (port_no >= DP_MAX_PORTS)
1682                         return ERR_PTR(-EFBIG);
1683
1684                 dp = get_dp(odp_header->dp_idx);
1685                 if (!dp)
1686                         return ERR_PTR(-ENODEV);
1687
1688                 vport = get_vport_protected(dp, port_no);
1689                 if (!vport)
1690                         return ERR_PTR(-ENOENT);
1691                 return vport;
1692         } else
1693                 return ERR_PTR(-EINVAL);
1694 }
1695
1696 /* Called with RTNL lock. */
1697 static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1698 {
1699         int err = 0;
1700         if (a[ODP_VPORT_ATTR_STATS])
1701                 err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS]));
1702         if (!err && a[ODP_VPORT_ATTR_ADDRESS])
1703                 err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS]));
1704         if (!err && a[ODP_VPORT_ATTR_MTU])
1705                 err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU]));
1706         return err;
1707 }
1708
1709 static int odp_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1710 {
1711         struct nlattr **a = info->attrs;
1712         struct odp_header *odp_header = info->userhdr;
1713         struct vport_parms parms;
1714         struct sk_buff *reply;
1715         struct vport *vport;
1716         struct datapath *dp;
1717         u32 port_no;
1718         int err;
1719
1720         err = -EINVAL;
1721         if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE])
1722                 goto exit;
1723
1724         err = odp_vport_cmd_validate(a);
1725         if (err)
1726                 goto exit;
1727
1728         rtnl_lock();
1729         dp = get_dp(odp_header->dp_idx);
1730         err = -ENODEV;
1731         if (!dp)
1732                 goto exit_unlock;
1733
1734         if (a[ODP_VPORT_ATTR_PORT_NO]) {
1735                 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1736
1737                 err = -EFBIG;
1738                 if (port_no >= DP_MAX_PORTS)
1739                         goto exit_unlock;
1740
1741                 vport = get_vport_protected(dp, port_no);
1742                 err = -EBUSY;
1743                 if (vport)
1744                         goto exit_unlock;
1745         } else {
1746                 for (port_no = 1; ; port_no++) {
1747                         if (port_no >= DP_MAX_PORTS) {
1748                                 err = -EFBIG;
1749                                 goto exit_unlock;
1750                         }
1751                         vport = get_vport_protected(dp, port_no);
1752                         if (!vport)
1753                                 break;
1754                 }
1755         }
1756
1757         parms.name = nla_data(a[ODP_VPORT_ATTR_NAME]);
1758         parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1759         parms.options = a[ODP_VPORT_ATTR_OPTIONS];
1760         parms.dp = dp;
1761         parms.port_no = port_no;
1762
1763         vport = new_vport(&parms);
1764         err = PTR_ERR(vport);
1765         if (IS_ERR(vport))
1766                 goto exit_unlock;
1767
1768         set_internal_devs_mtu(dp);
1769         dp_sysfs_add_if(vport);
1770
1771         err = change_vport(vport, a);
1772         if (!err) {
1773                 reply = odp_vport_cmd_build_info(vport, info->snd_pid,
1774                                                  info->snd_seq, ODP_VPORT_CMD_NEW);
1775                 if (IS_ERR(reply))
1776                         err = PTR_ERR(reply);
1777         }
1778         if (err) {
1779                 dp_detach_port(vport);
1780                 goto exit_unlock;
1781         }
1782         genl_notify(reply, genl_info_net(info), info->snd_pid,
1783                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1784
1785
1786 exit_unlock:
1787         rtnl_unlock();
1788 exit:
1789         return err;
1790 }
1791
1792 static int odp_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1793 {
1794         struct nlattr **a = info->attrs;
1795         struct sk_buff *reply;
1796         struct vport *vport;
1797         int err;
1798
1799         err = odp_vport_cmd_validate(a);
1800         if (err)
1801                 goto exit;
1802
1803         rtnl_lock();
1804         vport = lookup_vport(info->userhdr, a);
1805         err = PTR_ERR(vport);
1806         if (IS_ERR(vport))
1807                 goto exit_unlock;
1808
1809         err = 0;
1810         if (a[ODP_VPORT_ATTR_OPTIONS])
1811                 err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]);
1812         if (!err)
1813                 err = change_vport(vport, a);
1814
1815         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1816                                          ODP_VPORT_CMD_NEW);
1817         if (IS_ERR(reply)) {
1818                 err = PTR_ERR(reply);
1819                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1820                                 dp_vport_multicast_group.id, err);
1821                 return 0;
1822         }
1823
1824         genl_notify(reply, genl_info_net(info), info->snd_pid,
1825                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1826
1827 exit_unlock:
1828         rtnl_unlock();
1829 exit:
1830         return err;
1831 }
1832
1833 static int odp_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1834 {
1835         struct nlattr **a = info->attrs;
1836         struct sk_buff *reply;
1837         struct vport *vport;
1838         int err;
1839
1840         err = odp_vport_cmd_validate(a);
1841         if (err)
1842                 goto exit;
1843
1844         rtnl_lock();
1845         vport = lookup_vport(info->userhdr, a);
1846         err = PTR_ERR(vport);
1847         if (IS_ERR(vport))
1848                 goto exit_unlock;
1849
1850         if (vport->port_no == ODPP_LOCAL) {
1851                 err = -EINVAL;
1852                 goto exit_unlock;
1853         }
1854
1855         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1856                                          ODP_VPORT_CMD_DEL);
1857         err = PTR_ERR(reply);
1858         if (IS_ERR(reply))
1859                 goto exit_unlock;
1860
1861         err = dp_detach_port(vport);
1862
1863         genl_notify(reply, genl_info_net(info), info->snd_pid,
1864                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1865
1866 exit_unlock:
1867         rtnl_unlock();
1868 exit:
1869         return err;
1870 }
1871
1872 static int odp_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1873 {
1874         struct nlattr **a = info->attrs;
1875         struct odp_header *odp_header = info->userhdr;
1876         struct sk_buff *reply;
1877         struct vport *vport;
1878         int err;
1879
1880         err = odp_vport_cmd_validate(a);
1881         if (err)
1882                 goto exit;
1883
1884         rcu_read_lock();
1885         vport = lookup_vport(odp_header, a);
1886         err = PTR_ERR(vport);
1887         if (IS_ERR(vport))
1888                 goto exit_unlock;
1889
1890         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1891                                          ODP_VPORT_CMD_NEW);
1892         err = PTR_ERR(reply);
1893         if (IS_ERR(reply))
1894                 goto exit_unlock;
1895
1896         err = genlmsg_reply(reply, info);
1897
1898 exit_unlock:
1899         rcu_read_unlock();
1900 exit:
1901         return err;
1902 }
1903
1904 static int odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1905 {
1906         struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1907         struct datapath *dp;
1908         u32 port_no;
1909         int retval;
1910
1911         dp = get_dp(odp_header->dp_idx);
1912         if (!dp)
1913                 return -ENODEV;
1914
1915         rcu_read_lock();
1916         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1917                 struct vport *vport;
1918
1919                 vport = get_vport_protected(dp, port_no);
1920                 if (!vport)
1921                         continue;
1922
1923                 if (odp_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1924                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1925                                             ODP_VPORT_CMD_NEW) < 0)
1926                         break;
1927         }
1928         rcu_read_unlock();
1929
1930         cb->args[0] = port_no;
1931         retval = skb->len;
1932
1933         return retval;
1934 }
1935
1936 static struct genl_ops dp_vport_genl_ops[] = {
1937         { .cmd = ODP_VPORT_CMD_NEW,
1938           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1939           .policy = vport_policy,
1940           .doit = odp_vport_cmd_new
1941         },
1942         { .cmd = ODP_VPORT_CMD_DEL,
1943           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1944           .policy = vport_policy,
1945           .doit = odp_vport_cmd_del
1946         },
1947         { .cmd = ODP_VPORT_CMD_GET,
1948           .flags = 0,               /* OK for unprivileged users. */
1949           .policy = vport_policy,
1950           .doit = odp_vport_cmd_get,
1951           .dumpit = odp_vport_cmd_dump
1952         },
1953         { .cmd = ODP_VPORT_CMD_SET,
1954           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1955           .policy = vport_policy,
1956           .doit = odp_vport_cmd_set,
1957         },
1958 };
1959
1960 struct genl_family_and_ops {
1961         struct genl_family *family;
1962         struct genl_ops *ops;
1963         int n_ops;
1964         struct genl_multicast_group *group;
1965 };
1966
1967 static const struct genl_family_and_ops dp_genl_families[] = {
1968         { &dp_datapath_genl_family,
1969           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1970           &dp_datapath_multicast_group },
1971         { &dp_vport_genl_family,
1972           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1973           &dp_vport_multicast_group },
1974         { &dp_flow_genl_family,
1975           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1976           &dp_flow_multicast_group },
1977         { &dp_packet_genl_family,
1978           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1979           NULL },
1980 };
1981
1982 static void dp_unregister_genl(int n_families)
1983 {
1984         int i;
1985
1986         for (i = 0; i < n_families; i++) {
1987                 genl_unregister_family(dp_genl_families[i].family);
1988         }
1989 }
1990
1991 static int dp_register_genl(void)
1992 {
1993         int n_registered;
1994         int err;
1995         int i;
1996
1997         n_registered = 0;
1998         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1999                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2000
2001                 err = genl_register_family_with_ops(f->family, f->ops,
2002                                                     f->n_ops);
2003                 if (err)
2004                         goto error;
2005                 n_registered++;
2006
2007                 if (f->group) {
2008                         err = genl_register_mc_group(f->family, f->group);
2009                         if (err)
2010                                 goto error;
2011                 }
2012         }
2013
2014         err = packet_register_mc_groups();
2015         if (err)
2016                 goto error;
2017         return 0;
2018
2019 error:
2020         dp_unregister_genl(n_registered);
2021         return err;
2022 }
2023
2024 static int __init dp_init(void)
2025 {
2026         struct sk_buff *dummy_skb;
2027         int err;
2028
2029         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2030
2031         printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2032
2033         err = flow_init();
2034         if (err)
2035                 goto error;
2036
2037         err = vport_init();
2038         if (err)
2039                 goto error_flow_exit;
2040
2041         err = register_netdevice_notifier(&dp_device_notifier);
2042         if (err)
2043                 goto error_vport_exit;
2044
2045         err = dp_register_genl();
2046         if (err < 0)
2047                 goto error_unreg_notifier;
2048
2049         return 0;
2050
2051 error_unreg_notifier:
2052         unregister_netdevice_notifier(&dp_device_notifier);
2053 error_vport_exit:
2054         vport_exit();
2055 error_flow_exit:
2056         flow_exit();
2057 error:
2058         return err;
2059 }
2060
2061 static void dp_cleanup(void)
2062 {
2063         rcu_barrier();
2064         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2065         unregister_netdevice_notifier(&dp_device_notifier);
2066         vport_exit();
2067         flow_exit();
2068 }
2069
2070 module_init(dp_init);
2071 module_exit(dp_cleanup);
2072
2073 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2074 MODULE_LICENSE("GPL");