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