datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
   3  * Distributed under the terms of the GNU GPL version 2.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 /* Functions for managing the dp interface/device. */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/init.h>
  14 #include <linux/module.h>
  15 #include <linux/if_arp.h>
  16 #include <linux/if_vlan.h>
  17 #include <linux/in.h>
  18 #include <linux/ip.h>
  19 #include <linux/jhash.h>
  20 #include <linux/delay.h>
  21 #include <linux/time.h>
  22 #include <linux/etherdevice.h>
  23 #include <linux/genetlink.h>
  24 #include <linux/kernel.h>
  25 #include <linux/kthread.h>
  26 #include <linux/mutex.h>
  27 #include <linux/percpu.h>
  28 #include <linux/rcupdate.h>
  29 #include <linux/tcp.h>
  30 #include <linux/udp.h>
  31 #include <linux/version.h>
  32 #include <linux/ethtool.h>
  33 #include <linux/wait.h>
  34 #include <asm/system.h>
  35 #include <asm/div64.h>
  36 #include <asm/bug.h>
  37 #include <linux/highmem.h>
  38 #include <linux/netfilter_bridge.h>
  39 #include <linux/netfilter_ipv4.h>
  40 #include <linux/inetdevice.h>
  41 #include <linux/list.h>
  42 #include <linux/rculist.h>
  43 #include <linux/dmi.h>
  44 #include <net/inet_ecn.h>
  45 #include <net/genetlink.h>
  46
  47 #include "openvswitch/datapath-protocol.h"
  48 #include "checksum.h"
  49 #include "datapath.h"
  50 #include "actions.h"
  51 #include "flow.h"
  52 #include "table.h"
  53 #include "vlan.h"
  54 #include "vport-internal_dev.h"
  55
  56 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
  57     LINUX_VERSION_CODE >= KERNEL_VERSION(3,1,0)
  58 #error Kernels before 2.6.18 or after 3.0 are not supported by this version of Open vSwitch.
  59 #endif
  60
  61 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
  62 EXPORT_SYMBOL(dp_ioctl_hook);
  63
  64 /**
  65  * DOC: Locking:
  66  *
  67  * Writes to device state (add/remove datapath, port, set operations on vports,
  68  * etc.) are protected by RTNL.
  69  *
  70  * Writes to other state (flow table modifications, set miscellaneous datapath
  71  * parameters such as drop frags, etc.) are protected by genl_mutex.  The RTNL
  72  * lock nests inside genl_mutex.
  73  *
  74  * Reads are protected by RCU.
  75  *
  76  * There are a few special cases (mostly stats) that have their own
  77  * synchronization but they nest under all of above and don't interact with
  78  * each other.
  79  */
  80
  81 /* Global list of datapaths to enable dumping them all out.
  82  * Protected by genl_mutex.
  83  */
  84 static LIST_HEAD(dps);
  85
  86 static struct vport *new_vport(const struct vport_parms *);
  87 static int queue_userspace_packets(struct datapath *, struct sk_buff *,
  88                                  const struct dp_upcall_info *);
  89
  90 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  91 struct datapath *get_dp(int dp_ifindex)
  92 {
  93         struct datapath *dp = NULL;
  94         struct net_device *dev;
  95
  96         rcu_read_lock();
  97         dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
  98         if (dev) {
  99                 struct vport *vport = internal_dev_get_vport(dev);
 100                 if (vport)
 101                         dp = vport->dp;
 102         }
 103         rcu_read_unlock();
 104
 105         return dp;
 106 }
 107 EXPORT_SYMBOL_GPL(get_dp);
 108
 109 /* Must be called with genl_mutex. */
 110 static struct tbl *get_table_protected(struct datapath *dp)
 111 {
 112         return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
 113 }
 114
 115 /* Must be called with rcu_read_lock or RTNL lock. */
 116 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
 117 {
 118         return rcu_dereference_rtnl(dp->ports[port_no]);
 119 }
 120
 121 /* Must be called with rcu_read_lock or RTNL lock. */
 122 const char *dp_name(const struct datapath *dp)
 123 {
 124         return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL]));
 125 }
 126
 127 static inline size_t br_nlmsg_size(void)
 128 {
 129         return NLMSG_ALIGN(sizeof(struct ifinfomsg))
 130                + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
 131                + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
 132                + nla_total_size(4) /* IFLA_MASTER */
 133                + nla_total_size(4) /* IFLA_MTU */
 134                + nla_total_size(4) /* IFLA_LINK */
 135                + nla_total_size(1); /* IFLA_OPERSTATE */
 136 }
 137
 138 /* Caller must hold RTNL lock. */
 139 static int dp_fill_ifinfo(struct sk_buff *skb,
 140                           const struct vport *port,
 141                           int event, unsigned int flags)
 142 {
 143         struct datapath *dp = port->dp;
 144         int ifindex = vport_get_ifindex(port);
 145         int iflink = vport_get_iflink(port);
 146         struct ifinfomsg *hdr;
 147         struct nlmsghdr *nlh;
 148
 149         if (ifindex < 0)
 150                 return ifindex;
 151
 152         if (iflink < 0)
 153                 return iflink;
 154
 155         nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
 156         if (nlh == NULL)
 157                 return -EMSGSIZE;
 158
 159         hdr = nlmsg_data(nlh);
 160         hdr->ifi_family = AF_BRIDGE;
 161         hdr->__ifi_pad = 0;
 162         hdr->ifi_type = ARPHRD_ETHER;
 163         hdr->ifi_index = ifindex;
 164         hdr->ifi_flags = vport_get_flags(port);
 165         hdr->ifi_change = 0;
 166
 167         NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
 168         NLA_PUT_U32(skb, IFLA_MASTER,
 169                 vport_get_ifindex(get_vport_protected(dp, ODPP_LOCAL)));
 170         NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
 171 #ifdef IFLA_OPERSTATE
 172         NLA_PUT_U8(skb, IFLA_OPERSTATE,
 173                    vport_is_running(port)
 174                         ? vport_get_operstate(port)
 175                         : IF_OPER_DOWN);
 176 #endif
 177
 178         NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
 179
 180         if (ifindex != iflink)
 181                 NLA_PUT_U32(skb, IFLA_LINK,iflink);
 182
 183         return nlmsg_end(skb, nlh);
 184
 185 nla_put_failure:
 186         nlmsg_cancel(skb, nlh);
 187         return -EMSGSIZE;
 188 }
 189
 190 /* Caller must hold RTNL lock. */
 191 static void dp_ifinfo_notify(int event, struct vport *port)
 192 {
 193         struct sk_buff *skb;
 194         int err = -ENOBUFS;
 195
 196         skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
 197         if (skb == NULL)
 198                 goto errout;
 199
 200         err = dp_fill_ifinfo(skb, port, event, 0);
 201         if (err < 0) {
 202                 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
 203                 WARN_ON(err == -EMSGSIZE);
 204                 kfree_skb(skb);
 205                 goto errout;
 206         }
 207         rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 208         return;
 209 errout:
 210         if (err < 0)
 211                 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
 212 }
 213
 214 static void release_dp(struct kobject *kobj)
 215 {
 216         struct datapath *dp = container_of(kobj, struct datapath, ifobj);
 217         kfree(dp);
 218 }
 219
 220 static struct kobj_type dp_ktype = {
 221         .release = release_dp
 222 };
 223
 224 static void destroy_dp_rcu(struct rcu_head *rcu)
 225 {
 226         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 227
 228         tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl);
 229         free_percpu(dp->stats_percpu);
 230         kobject_put(&dp->ifobj);
 231 }
 232
 233 /* Called with RTNL lock and genl_lock. */
 234 static struct vport *new_vport(const struct vport_parms *parms)
 235 {
 236         struct vport *vport;
 237
 238         vport = vport_add(parms);
 239         if (!IS_ERR(vport)) {
 240                 struct datapath *dp = parms->dp;
 241
 242                 rcu_assign_pointer(dp->ports[parms->port_no], vport);
 243                 list_add(&vport->node, &dp->port_list);
 244
 245                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 246         }
 247
 248         return vport;
 249 }
 250
 251 /* Called with RTNL lock. */
 252 int dp_detach_port(struct vport *p)
 253 {
 254         ASSERT_RTNL();
 255
 256         if (p->port_no != ODPP_LOCAL)
 257                 dp_sysfs_del_if(p);
 258         dp_ifinfo_notify(RTM_DELLINK, p);
 259
 260         /* First drop references to device. */
 261         list_del(&p->node);
 262         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 263
 264         /* Then destroy it. */
 265         return vport_del(p);
 266 }
 267
 268 /* Must be called with rcu_read_lock. */
 269 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 270 {
 271         struct datapath *dp = p->dp;
 272         struct dp_stats_percpu *stats;
 273         int stats_counter_off;
 274         int error;
 275
 276         OVS_CB(skb)->vport = p;
 277
 278         if (!OVS_CB(skb)->flow) {
 279                 struct sw_flow_key key;
 280                 struct tbl_node *flow_node;
 281                 int key_len;
 282                 bool is_frag;
 283
 284                 /* Extract flow from 'skb' into 'key'. */
 285                 error = flow_extract(skb, p->port_no, &key, &key_len, &is_frag);
 286                 if (unlikely(error)) {
 287                         kfree_skb(skb);
 288                         return;
 289                 }
 290
 291                 if (is_frag && dp->drop_frags) {
 292                         consume_skb(skb);
 293                         stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
 294                         goto out;
 295                 }
 296
 297                 /* Look up flow. */
 298                 flow_node = tbl_lookup(rcu_dereference(dp->table), &key, key_len,
 299                                        flow_hash(&key, key_len), flow_cmp);
 300                 if (unlikely(!flow_node)) {
 301                         struct dp_upcall_info upcall;
 302
 303                         upcall.cmd = ODP_PACKET_CMD_MISS;
 304                         upcall.key = &key;
 305                         upcall.userdata = 0;
 306                         upcall.sample_pool = 0;
 307                         upcall.actions = NULL;
 308                         upcall.actions_len = 0;
 309                         dp_upcall(dp, skb, &upcall);
 310                         stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
 311                         goto out;
 312                 }
 313
 314                 OVS_CB(skb)->flow = flow_cast(flow_node);
 315         }
 316
 317         stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
 318         flow_used(OVS_CB(skb)->flow, skb);
 319         execute_actions(dp, skb);
 320
 321 out:
 322         /* Update datapath statistics. */
 323         local_bh_disable();
 324         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 325
 326         write_seqcount_begin(&stats->seqlock);
 327         (*(u64 *)((u8 *)stats + stats_counter_off))++;
 328         write_seqcount_end(&stats->seqlock);
 329
 330         local_bh_enable();
 331 }
 332
 333 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
 334 {
 335         u16 csum_start, csum_offset;
 336         __wsum csum;
 337
 338         get_skb_csum_pointers(skb, &csum_start, &csum_offset);
 339         csum_start -= skb_headroom(skb);
 340
 341         skb_copy_bits(skb, 0, to, csum_start);
 342
 343         csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
 344                                       skb->len - csum_start, 0);
 345         *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
 346 }
 347
 348 static struct genl_family dp_packet_genl_family = {
 349         .id = GENL_ID_GENERATE,
 350         .hdrsize = sizeof(struct odp_header),
 351         .name = ODP_PACKET_FAMILY,
 352         .version = 1,
 353         .maxattr = ODP_PACKET_ATTR_MAX
 354 };
 355
 356 /* Generic Netlink multicast groups for upcalls.
 357  *
 358  * We really want three unique multicast groups per datapath, but we can't even
 359  * get one, because genl_register_mc_group() takes genl_lock, which is also
 360  * held during Generic Netlink message processing, so trying to acquire
 361  * multicast groups during ODP_DP_NEW processing deadlocks.  Instead, we
 362  * preallocate a few groups and use them round-robin for datapaths.  Collision
 363  * isn't fatal--multicast listeners should check that the family is the one
 364  * that they want and discard others--but it wastes time and memory to receive
 365  * unwanted messages.
 366  */
 367 #define PACKET_N_MC_GROUPS 16
 368 static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
 369
 370 static u32 packet_mc_group(struct datapath *dp, u8 cmd)
 371 {
 372         u32 idx;
 373         BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
 374
 375         idx = jhash_2words(dp->dp_ifindex, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
 376         return packet_mc_groups[idx].id;
 377 }
 378
 379 static int packet_register_mc_groups(void)
 380 {
 381         int i;
 382
 383         for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
 384                 struct genl_multicast_group *group = &packet_mc_groups[i];
 385                 int error;
 386
 387                 sprintf(group->name, "packet%d", i);
 388                 error = genl_register_mc_group(&dp_packet_genl_family, group);
 389                 if (error)
 390                         return error;
 391         }
 392         return 0;
 393 }
 394
 395 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
 396 {
 397         struct dp_stats_percpu *stats;
 398         int err;
 399
 400         WARN_ON_ONCE(skb_shared(skb));
 401
 402         forward_ip_summed(skb, true);
 403
 404         /* Break apart GSO packets into their component pieces.  Otherwise
 405          * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
 406         if (skb_is_gso(skb)) {
 407                 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 408
 409                 if (IS_ERR(nskb)) {
 410                         kfree_skb(skb);
 411                         err = PTR_ERR(nskb);
 412                         goto err;
 413                 }
 414                 consume_skb(skb);
 415                 skb = nskb;
 416         }
 417
 418         err = queue_userspace_packets(dp, skb, upcall_info);
 419         if (err)
 420                 goto err;
 421
 422         return 0;
 423
 424 err:
 425         local_bh_disable();
 426         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 427
 428         write_seqcount_begin(&stats->seqlock);
 429         stats->n_lost++;
 430         write_seqcount_end(&stats->seqlock);
 431
 432         local_bh_enable();
 433
 434         return err;
 435 }
 436
 437 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
 438  * 'upcall_info'.  There will be only one packet unless we broke up a GSO
 439  * packet.
 440  */
 441 static int queue_userspace_packets(struct datapath *dp, struct sk_buff *skb,
 442                                  const struct dp_upcall_info *upcall_info)
 443 {
 444         u32 group = packet_mc_group(dp, upcall_info->cmd);
 445         struct sk_buff *nskb;
 446         int err;
 447
 448         do {
 449                 struct odp_header *upcall;
 450                 struct sk_buff *user_skb; /* to be queued to userspace */
 451                 struct nlattr *nla;
 452                 unsigned int len;
 453
 454                 nskb = skb->next;
 455                 skb->next = NULL;
 456
 457                 err = vlan_deaccel_tag(skb);
 458                 if (unlikely(err))
 459                         goto err_kfree_skbs;
 460
 461                 if (nla_attr_size(skb->len) > USHRT_MAX)
 462                         goto err_kfree_skbs;
 463
 464                 len = sizeof(struct odp_header);
 465                 len += nla_total_size(skb->len);
 466                 len += nla_total_size(FLOW_BUFSIZE);
 467                 if (upcall_info->userdata)
 468                         len += nla_total_size(8);
 469                 if (upcall_info->sample_pool)
 470                         len += nla_total_size(4);
 471                 if (upcall_info->actions_len)
 472                         len += nla_total_size(upcall_info->actions_len);
 473
 474                 user_skb = genlmsg_new(len, GFP_ATOMIC);
 475                 if (!user_skb) {
 476                         netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS);
 477                         goto err_kfree_skbs;
 478                 }
 479
 480                 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
 481                 upcall->dp_ifindex = dp->dp_ifindex;
 482
 483                 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY);
 484                 flow_to_nlattrs(upcall_info->key, user_skb);
 485                 nla_nest_end(user_skb, nla);
 486
 487                 if (upcall_info->userdata)
 488                         nla_put_u64(user_skb, ODP_PACKET_ATTR_USERDATA, upcall_info->userdata);
 489                 if (upcall_info->sample_pool)
 490                         nla_put_u32(user_skb, ODP_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
 491                 if (upcall_info->actions_len) {
 492                         const struct nlattr *actions = upcall_info->actions;
 493                         u32 actions_len = upcall_info->actions_len;
 494
 495                         nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_ACTIONS);
 496                         memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
 497                         nla_nest_end(user_skb, nla);
 498                 }
 499
 500                 nla = __nla_reserve(user_skb, ODP_PACKET_ATTR_PACKET, skb->len);
 501                 if (skb->ip_summed == CHECKSUM_PARTIAL)
 502                         copy_and_csum_skb(skb, nla_data(nla));
 503                 else
 504                         skb_copy_bits(skb, 0, nla_data(nla), skb->len);
 505
 506                 err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC);
 507                 if (err)
 508                         goto err_kfree_skbs;
 509
 510                 consume_skb(skb);
 511                 skb = nskb;
 512         } while (skb);
 513         return 0;
 514
 515 err_kfree_skbs:
 516         kfree_skb(skb);
 517         while ((skb = nskb) != NULL) {
 518                 nskb = skb->next;
 519                 kfree_skb(skb);
 520         }
 521         return err;
 522 }
 523
 524 /* Called with genl_mutex. */
 525 static int flush_flows(int dp_ifindex)
 526 {
 527         struct tbl *old_table;
 528         struct tbl *new_table;
 529         struct datapath *dp;
 530
 531         dp = get_dp(dp_ifindex);
 532         if (!dp)
 533                 return -ENODEV;
 534
 535         old_table = get_table_protected(dp);
 536         new_table = tbl_create(TBL_MIN_BUCKETS);
 537         if (!new_table)
 538                 return -ENOMEM;
 539
 540         rcu_assign_pointer(dp->table, new_table);
 541
 542         tbl_deferred_destroy(old_table, flow_free_tbl);
 543
 544         return 0;
 545 }
 546
 547 static int validate_actions(const struct nlattr *attr)
 548 {
 549         const struct nlattr *a;
 550         int rem;
 551
 552         nla_for_each_nested(a, attr, rem) {
 553                 static const u32 action_lens[ODP_ACTION_ATTR_MAX + 1] = {
 554                         [ODP_ACTION_ATTR_OUTPUT] = 4,
 555                         [ODP_ACTION_ATTR_USERSPACE] = 8,
 556                         [ODP_ACTION_ATTR_SET_DL_TCI] = 2,
 557                         [ODP_ACTION_ATTR_STRIP_VLAN] = 0,
 558                         [ODP_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
 559                         [ODP_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
 560                         [ODP_ACTION_ATTR_SET_NW_SRC] = 4,
 561                         [ODP_ACTION_ATTR_SET_NW_DST] = 4,
 562                         [ODP_ACTION_ATTR_SET_NW_TOS] = 1,
 563                         [ODP_ACTION_ATTR_SET_TP_SRC] = 2,
 564                         [ODP_ACTION_ATTR_SET_TP_DST] = 2,
 565                         [ODP_ACTION_ATTR_SET_TUNNEL] = 8,
 566                         [ODP_ACTION_ATTR_SET_PRIORITY] = 4,
 567                         [ODP_ACTION_ATTR_POP_PRIORITY] = 0,
 568                 };
 569                 int type = nla_type(a);
 570
 571                 if (type > ODP_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
 572                         return -EINVAL;
 573
 574                 switch (type) {
 575                 case ODP_ACTION_ATTR_UNSPEC:
 576                         return -EINVAL;
 577
 578                 case ODP_ACTION_ATTR_USERSPACE:
 579                 case ODP_ACTION_ATTR_STRIP_VLAN:
 580                 case ODP_ACTION_ATTR_SET_DL_SRC:
 581                 case ODP_ACTION_ATTR_SET_DL_DST:
 582                 case ODP_ACTION_ATTR_SET_NW_SRC:
 583                 case ODP_ACTION_ATTR_SET_NW_DST:
 584                 case ODP_ACTION_ATTR_SET_TP_SRC:
 585                 case ODP_ACTION_ATTR_SET_TP_DST:
 586                 case ODP_ACTION_ATTR_SET_TUNNEL:
 587                 case ODP_ACTION_ATTR_SET_PRIORITY:
 588                 case ODP_ACTION_ATTR_POP_PRIORITY:
 589                         /* No validation needed. */
 590                         break;
 591
 592                 case ODP_ACTION_ATTR_OUTPUT:
 593                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 594                                 return -EINVAL;
 595                         break;
 596
 597                 case ODP_ACTION_ATTR_SET_DL_TCI:
 598                         if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
 599                                 return -EINVAL;
 600                         break;
 601
 602                 case ODP_ACTION_ATTR_SET_NW_TOS:
 603                         if (nla_get_u8(a) & INET_ECN_MASK)
 604                                 return -EINVAL;
 605                         break;
 606
 607                 default:
 608                         return -EOPNOTSUPP;
 609                 }
 610         }
 611
 612         if (rem > 0)
 613                 return -EINVAL;
 614
 615         return 0;
 616 }
 617 static void clear_stats(struct sw_flow *flow)
 618 {
 619         flow->used = 0;
 620         flow->tcp_flags = 0;
 621         flow->packet_count = 0;
 622         flow->byte_count = 0;
 623 }
 624
 625 /* Called with genl_mutex. */
 626 static int expand_table(struct datapath *dp)
 627 {
 628         struct tbl *old_table = get_table_protected(dp);
 629         struct tbl *new_table;
 630
 631         new_table = tbl_expand(old_table);
 632         if (IS_ERR(new_table))
 633                 return PTR_ERR(new_table);
 634
 635         rcu_assign_pointer(dp->table, new_table);
 636         tbl_deferred_destroy(old_table, NULL);
 637
 638         return 0;
 639 }
 640
 641 static int odp_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 642 {
 643         struct odp_header *odp_header = info->userhdr;
 644         struct nlattr **a = info->attrs;
 645         struct sw_flow_actions *acts;
 646         struct sk_buff *packet;
 647         struct sw_flow *flow;
 648         struct datapath *dp;
 649         struct ethhdr *eth;
 650         bool is_frag;
 651         int len;
 652         int err;
 653         int key_len;
 654
 655         err = -EINVAL;
 656         if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_KEY] ||
 657             !a[ODP_PACKET_ATTR_ACTIONS] ||
 658             nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN)
 659                 goto err;
 660
 661         err = validate_actions(a[ODP_PACKET_ATTR_ACTIONS]);
 662         if (err)
 663                 goto err;
 664
 665         len = nla_len(a[ODP_PACKET_ATTR_PACKET]);
 666         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 667         err = -ENOMEM;
 668         if (!packet)
 669                 goto err;
 670         skb_reserve(packet, NET_IP_ALIGN);
 671
 672         memcpy(__skb_put(packet, len), nla_data(a[ODP_PACKET_ATTR_PACKET]), len);
 673
 674         skb_reset_mac_header(packet);
 675         eth = eth_hdr(packet);
 676
 677         /* Normally, setting the skb 'protocol' field would be handled by a
 678          * call to eth_type_trans(), but it assumes there's a sending
 679          * device, which we may not have. */
 680         if (ntohs(eth->h_proto) >= 1536)
 681                 packet->protocol = eth->h_proto;
 682         else
 683                 packet->protocol = htons(ETH_P_802_2);
 684
 685         /* Build an sw_flow for sending this packet. */
 686         flow = flow_alloc();
 687         err = PTR_ERR(flow);
 688         if (IS_ERR(flow))
 689                 goto err_kfree_skb;
 690
 691         err = flow_extract(packet, -1, &flow->key, &key_len, &is_frag);
 692         if (err)
 693                 goto err_flow_put;
 694
 695         err = flow_metadata_from_nlattrs(&flow->key.eth.in_port,
 696                                          &flow->key.eth.tun_id,
 697                                          a[ODP_PACKET_ATTR_KEY]);
 698         if (err)
 699                 goto err_flow_put;
 700
 701         flow->tbl_node.hash = flow_hash(&flow->key, key_len);
 702
 703         acts = flow_actions_alloc(a[ODP_PACKET_ATTR_ACTIONS]);
 704         err = PTR_ERR(acts);
 705         if (IS_ERR(acts))
 706                 goto err_flow_put;
 707         rcu_assign_pointer(flow->sf_acts, acts);
 708
 709         OVS_CB(packet)->flow = flow;
 710
 711         rcu_read_lock();
 712         dp = get_dp(odp_header->dp_ifindex);
 713         err = -ENODEV;
 714         if (!dp)
 715                 goto err_unlock;
 716         err = execute_actions(dp, packet);
 717         rcu_read_unlock();
 718
 719         flow_put(flow);
 720         return err;
 721
 722 err_unlock:
 723         rcu_read_unlock();
 724 err_flow_put:
 725         flow_put(flow);
 726 err_kfree_skb:
 727         kfree_skb(packet);
 728 err:
 729         return err;
 730 }
 731
 732 static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = {
 733         [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 734         [ODP_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 735         [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 736 };
 737
 738 static struct genl_ops dp_packet_genl_ops[] = {
 739         { .cmd = ODP_PACKET_CMD_EXECUTE,
 740           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 741           .policy = packet_policy,
 742           .doit = odp_packet_cmd_execute
 743         }
 744 };
 745
 746 static void get_dp_stats(struct datapath *dp, struct odp_stats *stats)
 747 {
 748         int i;
 749
 750         stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
 751         for_each_possible_cpu(i) {
 752                 const struct dp_stats_percpu *percpu_stats;
 753                 struct dp_stats_percpu local_stats;
 754                 unsigned seqcount;
 755
 756                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 757
 758                 do {
 759                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 760                         local_stats = *percpu_stats;
 761                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 762
 763                 stats->n_frags += local_stats.n_frags;
 764                 stats->n_hit += local_stats.n_hit;
 765                 stats->n_missed += local_stats.n_missed;
 766                 stats->n_lost += local_stats.n_lost;
 767         }
 768 }
 769
 770 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
 771  * Called with RTNL lock.
 772  */
 773 int dp_min_mtu(const struct datapath *dp)
 774 {
 775         struct vport *p;
 776         int mtu = 0;
 777
 778         ASSERT_RTNL();
 779
 780         list_for_each_entry (p, &dp->port_list, node) {
 781                 int dev_mtu;
 782
 783                 /* Skip any internal ports, since that's what we're trying to
 784                  * set. */
 785                 if (is_internal_vport(p))
 786                         continue;
 787
 788                 dev_mtu = vport_get_mtu(p);
 789                 if (!dev_mtu)
 790                         continue;
 791                 if (!mtu || dev_mtu < mtu)
 792                         mtu = dev_mtu;
 793         }
 794
 795         return mtu ? mtu : ETH_DATA_LEN;
 796 }
 797
 798 /* Sets the MTU of all datapath devices to the minimum of the ports
 799  * Called with RTNL lock.
 800  */
 801 void set_internal_devs_mtu(const struct datapath *dp)
 802 {
 803         struct vport *p;
 804         int mtu;
 805
 806         ASSERT_RTNL();
 807
 808         mtu = dp_min_mtu(dp);
 809
 810         list_for_each_entry (p, &dp->port_list, node) {
 811                 if (is_internal_vport(p))
 812                         vport_set_mtu(p, mtu);
 813         }
 814 }
 815
 816 static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = {
 817         [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 818         [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 819         [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 820 };
 821
 822 static struct genl_family dp_flow_genl_family = {
 823         .id = GENL_ID_GENERATE,
 824         .hdrsize = sizeof(struct odp_header),
 825         .name = ODP_FLOW_FAMILY,
 826         .version = 1,
 827         .maxattr = ODP_FLOW_ATTR_MAX
 828 };
 829
 830 static struct genl_multicast_group dp_flow_multicast_group = {
 831         .name = ODP_FLOW_MCGROUP
 832 };
 833
 834 /* Called with genl_lock. */
 835 static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 836                                   struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
 837 {
 838         const int skb_orig_len = skb->len;
 839         const struct sw_flow_actions *sf_acts;
 840         struct odp_flow_stats stats;
 841         struct odp_header *odp_header;
 842         struct nlattr *nla;
 843         unsigned long used;
 844         u8 tcp_flags;
 845         int err;
 846
 847         sf_acts = rcu_dereference_protected(flow->sf_acts,
 848                                             lockdep_genl_is_held());
 849
 850         odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 851         if (!odp_header)
 852                 return -EMSGSIZE;
 853
 854         odp_header->dp_ifindex = dp->dp_ifindex;
 855
 856         nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY);
 857         if (!nla)
 858                 goto nla_put_failure;
 859         err = flow_to_nlattrs(&flow->key, skb);
 860         if (err)
 861                 goto error;
 862         nla_nest_end(skb, nla);
 863
 864         spin_lock_bh(&flow->lock);
 865         used = flow->used;
 866         stats.n_packets = flow->packet_count;
 867         stats.n_bytes = flow->byte_count;
 868         tcp_flags = flow->tcp_flags;
 869         spin_unlock_bh(&flow->lock);
 870
 871         if (used)
 872                 NLA_PUT_U64(skb, ODP_FLOW_ATTR_USED, flow_used_time(used));
 873
 874         if (stats.n_packets)
 875                 NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats);
 876
 877         if (tcp_flags)
 878                 NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 879
 880         /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 881          * this is the first flow to be dumped into 'skb'.  This is unusual for
 882          * Netlink but individual action lists can be longer than
 883          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 884          * The userspace caller can always fetch the actions separately if it
 885          * really wants them.  (Most userspace callers in fact don't care.)
 886          *
 887          * This can only fail for dump operations because the skb is always
 888          * properly sized for single flows.
 889          */
 890         err = nla_put(skb, ODP_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 891                       sf_acts->actions);
 892         if (err < 0 && skb_orig_len)
 893                 goto error;
 894
 895         return genlmsg_end(skb, odp_header);
 896
 897 nla_put_failure:
 898         err = -EMSGSIZE;
 899 error:
 900         genlmsg_cancel(skb, odp_header);
 901         return err;
 902 }
 903
 904 static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow)
 905 {
 906         const struct sw_flow_actions *sf_acts;
 907         int len;
 908
 909         sf_acts = rcu_dereference_protected(flow->sf_acts,
 910                                             lockdep_genl_is_held());
 911
 912         len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */
 913         len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */
 914         len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */
 915         len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */
 916         len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */
 917         return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL);
 918 }
 919
 920 static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
 921                                                u32 pid, u32 seq, u8 cmd)
 922 {
 923         struct sk_buff *skb;
 924         int retval;
 925
 926         skb = odp_flow_cmd_alloc_info(flow);
 927         if (!skb)
 928                 return ERR_PTR(-ENOMEM);
 929
 930         retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 931         BUG_ON(retval < 0);
 932         return skb;
 933 }
 934
 935 static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 936 {
 937         struct nlattr **a = info->attrs;
 938         struct odp_header *odp_header = info->userhdr;
 939         struct tbl_node *flow_node;
 940         struct sw_flow_key key;
 941         struct sw_flow *flow;
 942         struct sk_buff *reply;
 943         struct datapath *dp;
 944         struct tbl *table;
 945         u32 hash;
 946         int error;
 947         int key_len;
 948
 949         /* Extract key. */
 950         error = -EINVAL;
 951         if (!a[ODP_FLOW_ATTR_KEY])
 952                 goto error;
 953         error = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
 954         if (error)
 955                 goto error;
 956
 957         /* Validate actions. */
 958         if (a[ODP_FLOW_ATTR_ACTIONS]) {
 959                 error = validate_actions(a[ODP_FLOW_ATTR_ACTIONS]);
 960                 if (error)
 961                         goto error;
 962         } else if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW) {
 963                 error = -EINVAL;
 964                 goto error;
 965         }
 966
 967         dp = get_dp(odp_header->dp_ifindex);
 968         error = -ENODEV;
 969         if (!dp)
 970                 goto error;
 971
 972         hash = flow_hash(&key, key_len);
 973         table = get_table_protected(dp);
 974         flow_node = tbl_lookup(table, &key, key_len, hash, flow_cmp);
 975         if (!flow_node) {
 976                 struct sw_flow_actions *acts;
 977
 978                 /* Bail out if we're not allowed to create a new flow. */
 979                 error = -ENOENT;
 980                 if (info->genlhdr->cmd == ODP_FLOW_CMD_SET)
 981                         goto error;
 982
 983                 /* Expand table, if necessary, to make room. */
 984                 if (tbl_count(table) >= tbl_n_buckets(table)) {
 985                         error = expand_table(dp);
 986                         if (error)
 987                                 goto error;
 988                         table = get_table_protected(dp);
 989                 }
 990
 991                 /* Allocate flow. */
 992                 flow = flow_alloc();
 993                 if (IS_ERR(flow)) {
 994                         error = PTR_ERR(flow);
 995                         goto error;
 996                 }
 997                 flow->key = key;
 998                 clear_stats(flow);
 999
1000                 /* Obtain actions. */
1001                 acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1002                 error = PTR_ERR(acts);
1003                 if (IS_ERR(acts))
1004                         goto error_free_flow;
1005                 rcu_assign_pointer(flow->sf_acts, acts);
1006
1007                 /* Put flow in bucket. */
1008                 error = tbl_insert(table, &flow->tbl_node, hash);
1009                 if (error)
1010                         goto error_free_flow;
1011
1012                 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1013                                                 info->snd_seq, ODP_FLOW_CMD_NEW);
1014         } else {
1015                 /* We found a matching flow. */
1016                 struct sw_flow_actions *old_acts;
1017
1018                 /* Bail out if we're not allowed to modify an existing flow.
1019                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1020                  * because Generic Netlink treats the latter as a dump
1021                  * request.  We also accept NLM_F_EXCL in case that bug ever
1022                  * gets fixed.
1023                  */
1024                 error = -EEXIST;
1025                 if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW &&
1026                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1027                         goto error;
1028
1029                 /* Update actions. */
1030                 flow = flow_cast(flow_node);
1031                 old_acts = rcu_dereference_protected(flow->sf_acts,
1032                                                      lockdep_genl_is_held());
1033                 if (a[ODP_FLOW_ATTR_ACTIONS] &&
1034                     (old_acts->actions_len != nla_len(a[ODP_FLOW_ATTR_ACTIONS]) ||
1035                      memcmp(old_acts->actions, nla_data(a[ODP_FLOW_ATTR_ACTIONS]),
1036                             old_acts->actions_len))) {
1037                         struct sw_flow_actions *new_acts;
1038
1039                         new_acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1040                         error = PTR_ERR(new_acts);
1041                         if (IS_ERR(new_acts))
1042                                 goto error;
1043
1044                         rcu_assign_pointer(flow->sf_acts, new_acts);
1045                         flow_deferred_free_acts(old_acts);
1046                 }
1047
1048                 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1049                                                 info->snd_seq, ODP_FLOW_CMD_NEW);
1050
1051                 /* Clear stats. */
1052                 if (a[ODP_FLOW_ATTR_CLEAR]) {
1053                         spin_lock_bh(&flow->lock);
1054                         clear_stats(flow);
1055                         spin_unlock_bh(&flow->lock);
1056                 }
1057         }
1058
1059         if (!IS_ERR(reply))
1060                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1061                             dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1062         else
1063                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1064                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1065         return 0;
1066
1067 error_free_flow:
1068         flow_put(flow);
1069 error:
1070         return error;
1071 }
1072
1073 static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1074 {
1075         struct nlattr **a = info->attrs;
1076         struct odp_header *odp_header = info->userhdr;
1077         struct sw_flow_key key;
1078         struct tbl_node *flow_node;
1079         struct sk_buff *reply;
1080         struct sw_flow *flow;
1081         struct datapath *dp;
1082         struct tbl *table;
1083         int err;
1084         int key_len;
1085
1086         if (!a[ODP_FLOW_ATTR_KEY])
1087                 return -EINVAL;
1088         err = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
1089         if (err)
1090                 return err;
1091
1092         dp = get_dp(odp_header->dp_ifindex);
1093         if (!dp)
1094                 return -ENODEV;
1095
1096         table = get_table_protected(dp);
1097         flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1098                                flow_cmp);
1099         if (!flow_node)
1100                 return -ENOENT;
1101
1102         flow = flow_cast(flow_node);
1103         reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW);
1104         if (IS_ERR(reply))
1105                 return PTR_ERR(reply);
1106
1107         return genlmsg_reply(reply, info);
1108 }
1109
1110 static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1111 {
1112         struct nlattr **a = info->attrs;
1113         struct odp_header *odp_header = info->userhdr;
1114         struct sw_flow_key key;
1115         struct tbl_node *flow_node;
1116         struct sk_buff *reply;
1117         struct sw_flow *flow;
1118         struct datapath *dp;
1119         struct tbl *table;
1120         int err;
1121         int key_len;
1122
1123         if (!a[ODP_FLOW_ATTR_KEY])
1124                 return flush_flows(odp_header->dp_ifindex);
1125         err = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
1126         if (err)
1127                 return err;
1128
1129         dp = get_dp(odp_header->dp_ifindex);
1130         if (!dp)
1131                 return -ENODEV;
1132
1133         table = get_table_protected(dp);
1134         flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1135                                flow_cmp);
1136         if (!flow_node)
1137                 return -ENOENT;
1138         flow = flow_cast(flow_node);
1139
1140         reply = odp_flow_cmd_alloc_info(flow);
1141         if (!reply)
1142                 return -ENOMEM;
1143
1144         err = tbl_remove(table, flow_node);
1145         if (err) {
1146                 kfree_skb(reply);
1147                 return err;
1148         }
1149
1150         err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1151                                      info->snd_seq, 0, ODP_FLOW_CMD_DEL);
1152         BUG_ON(err < 0);
1153
1154         flow_deferred_free(flow);
1155
1156         genl_notify(reply, genl_info_net(info), info->snd_pid,
1157                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1158         return 0;
1159 }
1160
1161 static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1162 {
1163         struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1164         struct datapath *dp;
1165
1166         dp = get_dp(odp_header->dp_ifindex);
1167         if (!dp)
1168                 return -ENODEV;
1169
1170         for (;;) {
1171                 struct tbl_node *flow_node;
1172                 struct sw_flow *flow;
1173                 u32 bucket, obj;
1174
1175                 bucket = cb->args[0];
1176                 obj = cb->args[1];
1177                 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1178                 if (!flow_node)
1179                         break;
1180
1181                 flow = flow_cast(flow_node);
1182                 if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1183                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1184                                            ODP_FLOW_CMD_NEW) < 0)
1185                         break;
1186
1187                 cb->args[0] = bucket;
1188                 cb->args[1] = obj;
1189         }
1190         return skb->len;
1191 }
1192
1193 static struct genl_ops dp_flow_genl_ops[] = {
1194         { .cmd = ODP_FLOW_CMD_NEW,
1195           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1196           .policy = flow_policy,
1197           .doit = odp_flow_cmd_new_or_set
1198         },
1199         { .cmd = ODP_FLOW_CMD_DEL,
1200           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1201           .policy = flow_policy,
1202           .doit = odp_flow_cmd_del
1203         },
1204         { .cmd = ODP_FLOW_CMD_GET,
1205           .flags = 0,               /* OK for unprivileged users. */
1206           .policy = flow_policy,
1207           .doit = odp_flow_cmd_get,
1208           .dumpit = odp_flow_cmd_dump
1209         },
1210         { .cmd = ODP_FLOW_CMD_SET,
1211           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1212           .policy = flow_policy,
1213           .doit = odp_flow_cmd_new_or_set,
1214         },
1215 };
1216
1217 static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
1218 #ifdef HAVE_NLA_NUL_STRING
1219         [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1220 #endif
1221         [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1222         [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1223 };
1224
1225 static struct genl_family dp_datapath_genl_family = {
1226         .id = GENL_ID_GENERATE,
1227         .hdrsize = sizeof(struct odp_header),
1228         .name = ODP_DATAPATH_FAMILY,
1229         .version = 1,
1230         .maxattr = ODP_DP_ATTR_MAX
1231 };
1232
1233 static struct genl_multicast_group dp_datapath_multicast_group = {
1234         .name = ODP_DATAPATH_MCGROUP
1235 };
1236
1237 static int odp_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1238                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1239 {
1240         struct odp_header *odp_header;
1241         struct nlattr *nla;
1242         int err;
1243
1244         odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1245                                    flags, cmd);
1246         if (!odp_header)
1247                 goto error;
1248
1249         odp_header->dp_ifindex = dp->dp_ifindex;
1250
1251         rcu_read_lock();
1252         err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
1253         rcu_read_unlock();
1254         if (err)
1255                 goto nla_put_failure;
1256
1257         nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats));
1258         if (!nla)
1259                 goto nla_put_failure;
1260         get_dp_stats(dp, nla_data(nla));
1261
1262         NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS,
1263                     dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO);
1264
1265         if (dp->sflow_probability)
1266                 NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability);
1267
1268         nla = nla_nest_start(skb, ODP_DP_ATTR_MCGROUPS);
1269         if (!nla)
1270                 goto nla_put_failure;
1271         NLA_PUT_U32(skb, ODP_PACKET_CMD_MISS, packet_mc_group(dp, ODP_PACKET_CMD_MISS));
1272         NLA_PUT_U32(skb, ODP_PACKET_CMD_ACTION, packet_mc_group(dp, ODP_PACKET_CMD_ACTION));
1273         NLA_PUT_U32(skb, ODP_PACKET_CMD_SAMPLE, packet_mc_group(dp, ODP_PACKET_CMD_SAMPLE));
1274         nla_nest_end(skb, nla);
1275
1276         return genlmsg_end(skb, odp_header);
1277
1278 nla_put_failure:
1279         genlmsg_cancel(skb, odp_header);
1280 error:
1281         return -EMSGSIZE;
1282 }
1283
1284 static struct sk_buff *odp_dp_cmd_build_info(struct datapath *dp, u32 pid,
1285                                              u32 seq, u8 cmd)
1286 {
1287         struct sk_buff *skb;
1288         int retval;
1289
1290         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1291         if (!skb)
1292                 return ERR_PTR(-ENOMEM);
1293
1294         retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1295         if (retval < 0) {
1296                 kfree_skb(skb);
1297                 return ERR_PTR(retval);
1298         }
1299         return skb;
1300 }
1301
1302 static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1303 {
1304         if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1305                 u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1306
1307                 if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP)
1308                         return -EINVAL;
1309         }
1310
1311         return CHECK_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
1312 }
1313
1314 /* Called with genl_mutex and optionally with RTNL lock also. */
1315 static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1316 {
1317         struct datapath *dp;
1318
1319         if (!a[ODP_DP_ATTR_NAME])
1320                 dp = get_dp(odp_header->dp_ifindex);
1321         else {
1322                 struct vport *vport;
1323
1324                 rcu_read_lock();
1325                 vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
1326                 dp = vport && vport->port_no == ODPP_LOCAL ? vport->dp : NULL;
1327                 rcu_read_unlock();
1328         }
1329         return dp ? dp : ERR_PTR(-ENODEV);
1330 }
1331
1332 /* Called with genl_mutex. */
1333 static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1334 {
1335         if (a[ODP_DP_ATTR_IPV4_FRAGS])
1336                 dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP;
1337         if (a[ODP_DP_ATTR_SAMPLING])
1338                 dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]);
1339 }
1340
1341 static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1342 {
1343         struct nlattr **a = info->attrs;
1344         struct vport_parms parms;
1345         struct sk_buff *reply;
1346         struct datapath *dp;
1347         struct vport *vport;
1348         int err;
1349
1350         err = -EINVAL;
1351         if (!a[ODP_DP_ATTR_NAME])
1352                 goto err;
1353
1354         err = odp_dp_cmd_validate(a);
1355         if (err)
1356                 goto err;
1357
1358         rtnl_lock();
1359         err = -ENODEV;
1360         if (!try_module_get(THIS_MODULE))
1361                 goto err_unlock_rtnl;
1362
1363         err = -ENOMEM;
1364         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1365         if (dp == NULL)
1366                 goto err_put_module;
1367         INIT_LIST_HEAD(&dp->port_list);
1368
1369         /* Initialize kobject for bridge.  This will be added as
1370          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1371         dp->ifobj.kset = NULL;
1372         kobject_init(&dp->ifobj, &dp_ktype);
1373
1374         /* Allocate table. */
1375         err = -ENOMEM;
1376         rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1377         if (!dp->table)
1378                 goto err_free_dp;
1379
1380         /* Set up our datapath device. */
1381         parms.name = nla_data(a[ODP_DP_ATTR_NAME]);
1382         parms.type = ODP_VPORT_TYPE_INTERNAL;
1383         parms.options = NULL;
1384         parms.dp = dp;
1385         parms.port_no = ODPP_LOCAL;
1386         vport = new_vport(&parms);
1387         if (IS_ERR(vport)) {
1388                 err = PTR_ERR(vport);
1389                 if (err == -EBUSY)
1390                         err = -EEXIST;
1391
1392                 goto err_destroy_table;
1393         }
1394         dp->dp_ifindex = vport_get_ifindex(vport);
1395
1396         dp->drop_frags = 0;
1397         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1398         if (!dp->stats_percpu) {
1399                 err = -ENOMEM;
1400                 goto err_destroy_local_port;
1401         }
1402
1403         change_datapath(dp, a);
1404
1405         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1406         err = PTR_ERR(reply);
1407         if (IS_ERR(reply))
1408                 goto err_destroy_local_port;
1409
1410         list_add_tail(&dp->list_node, &dps);
1411         dp_sysfs_add_dp(dp);
1412
1413         rtnl_unlock();
1414
1415         genl_notify(reply, genl_info_net(info), info->snd_pid,
1416                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1417         return 0;
1418
1419 err_destroy_local_port:
1420         dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1421 err_destroy_table:
1422         tbl_destroy(get_table_protected(dp), NULL);
1423 err_free_dp:
1424         kfree(dp);
1425 err_put_module:
1426         module_put(THIS_MODULE);
1427 err_unlock_rtnl:
1428         rtnl_unlock();
1429 err:
1430         return err;
1431 }
1432
1433 static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1434 {
1435         struct vport *vport, *next_vport;
1436         struct sk_buff *reply;
1437         struct datapath *dp;
1438         int err;
1439
1440         err = odp_dp_cmd_validate(info->attrs);
1441         if (err)
1442                 goto exit;
1443
1444         rtnl_lock();
1445         dp = lookup_datapath(info->userhdr, info->attrs);
1446         err = PTR_ERR(dp);
1447         if (IS_ERR(dp))
1448                 goto exit_unlock;
1449
1450         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_DEL);
1451         err = PTR_ERR(reply);
1452         if (IS_ERR(reply))
1453                 goto exit_unlock;
1454
1455         list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1456                 if (vport->port_no != ODPP_LOCAL)
1457                         dp_detach_port(vport);
1458
1459         dp_sysfs_del_dp(dp);
1460         list_del(&dp->list_node);
1461         dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1462
1463         /* rtnl_unlock() will wait until all the references to devices that
1464          * are pending unregistration have been dropped.  We do it here to
1465          * ensure that any internal devices (which contain DP pointers) are
1466          * fully destroyed before freeing the datapath.
1467          */
1468         rtnl_unlock();
1469
1470         call_rcu(&dp->rcu, destroy_dp_rcu);
1471         module_put(THIS_MODULE);
1472
1473         genl_notify(reply, genl_info_net(info), info->snd_pid,
1474                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1475
1476         return 0;
1477
1478 exit_unlock:
1479         rtnl_unlock();
1480 exit:
1481         return err;
1482 }
1483
1484 static int odp_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1485 {
1486         struct sk_buff *reply;
1487         struct datapath *dp;
1488         int err;
1489
1490         err = odp_dp_cmd_validate(info->attrs);
1491         if (err)
1492                 return err;
1493
1494         dp = lookup_datapath(info->userhdr, info->attrs);
1495         if (IS_ERR(dp))
1496                 return PTR_ERR(dp);
1497
1498         change_datapath(dp, info->attrs);
1499
1500         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1501         if (IS_ERR(reply)) {
1502                 err = PTR_ERR(reply);
1503                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1504                                 dp_datapath_multicast_group.id, err);
1505                 return 0;
1506         }
1507
1508         genl_notify(reply, genl_info_net(info), info->snd_pid,
1509                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1510         return 0;
1511 }
1512
1513 static int odp_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1514 {
1515         struct sk_buff *reply;
1516         struct datapath *dp;
1517         int err;
1518
1519         err = odp_dp_cmd_validate(info->attrs);
1520         if (err)
1521                 return err;
1522
1523         dp = lookup_datapath(info->userhdr, info->attrs);
1524         if (IS_ERR(dp))
1525                 return PTR_ERR(dp);
1526
1527         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1528         if (IS_ERR(reply))
1529                 return PTR_ERR(reply);
1530
1531         return genlmsg_reply(reply, info);
1532 }
1533
1534 static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1535 {
1536         struct datapath *dp;
1537         int skip = cb->args[0];
1538         int i = 0;
1539
1540         list_for_each_entry (dp, &dps, list_node) {
1541                 if (i < skip)
1542                         continue;
1543                 if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1544                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1545                                          ODP_DP_CMD_NEW) < 0)
1546                         break;
1547                 i++;
1548         }
1549
1550         cb->args[0] = i;
1551
1552         return skb->len;
1553 }
1554
1555 static struct genl_ops dp_datapath_genl_ops[] = {
1556         { .cmd = ODP_DP_CMD_NEW,
1557           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1558           .policy = datapath_policy,
1559           .doit = odp_dp_cmd_new
1560         },
1561         { .cmd = ODP_DP_CMD_DEL,
1562           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1563           .policy = datapath_policy,
1564           .doit = odp_dp_cmd_del
1565         },
1566         { .cmd = ODP_DP_CMD_GET,
1567           .flags = 0,               /* OK for unprivileged users. */
1568           .policy = datapath_policy,
1569           .doit = odp_dp_cmd_get,
1570           .dumpit = odp_dp_cmd_dump
1571         },
1572         { .cmd = ODP_DP_CMD_SET,
1573           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1574           .policy = datapath_policy,
1575           .doit = odp_dp_cmd_set,
1576         },
1577 };
1578
1579 static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = {
1580 #ifdef HAVE_NLA_NUL_STRING
1581         [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1582         [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1583         [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1584         [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1585         [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1586 #else
1587         [ODP_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1588         [ODP_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1589 #endif
1590         [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1591         [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1592 };
1593
1594 static struct genl_family dp_vport_genl_family = {
1595         .id = GENL_ID_GENERATE,
1596         .hdrsize = sizeof(struct odp_header),
1597         .name = ODP_VPORT_FAMILY,
1598         .version = 1,
1599         .maxattr = ODP_VPORT_ATTR_MAX
1600 };
1601
1602 static struct genl_multicast_group dp_vport_multicast_group = {
1603         .name = ODP_VPORT_MCGROUP
1604 };
1605
1606 /* Called with RTNL lock or RCU read lock. */
1607 static int odp_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1608                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1609 {
1610         struct odp_header *odp_header;
1611         struct nlattr *nla;
1612         int ifindex, iflink;
1613         int mtu;
1614         int err;
1615
1616         odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1617                                  flags, cmd);
1618         if (!odp_header)
1619                 return -EMSGSIZE;
1620
1621         odp_header->dp_ifindex = vport->dp->dp_ifindex;
1622
1623         NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1624         NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport));
1625         NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport));
1626
1627         nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1628         if (!nla)
1629                 goto nla_put_failure;
1630         if (vport_get_stats(vport, nla_data(nla)))
1631                 __skb_trim(skb, skb->len - nla->nla_len);
1632
1633         NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1634
1635         mtu = vport_get_mtu(vport);
1636         if (mtu)
1637                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, mtu);
1638
1639         err = vport_get_options(vport, skb);
1640         if (err == -EMSGSIZE)
1641                 goto error;
1642
1643         ifindex = vport_get_ifindex(vport);
1644         if (ifindex > 0)
1645                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex);
1646
1647         iflink = vport_get_iflink(vport);
1648         if (iflink > 0)
1649                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink);
1650
1651         return genlmsg_end(skb, odp_header);
1652
1653 nla_put_failure:
1654         err = -EMSGSIZE;
1655 error:
1656         genlmsg_cancel(skb, odp_header);
1657         return err;
1658 }
1659
1660 /* Called with RTNL lock or RCU read lock. */
1661 static struct sk_buff *odp_vport_cmd_build_info(struct vport *vport, u32 pid,
1662                                                 u32 seq, u8 cmd)
1663 {
1664         struct sk_buff *skb;
1665         int retval;
1666
1667         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1668         if (!skb)
1669                 return ERR_PTR(-ENOMEM);
1670
1671         retval = odp_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1672         if (retval < 0) {
1673                 kfree_skb(skb);
1674                 return ERR_PTR(retval);
1675         }
1676         return skb;
1677 }
1678
1679 static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1680 {
1681         return CHECK_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1682 }
1683
1684 /* Called with RTNL lock or RCU read lock. */
1685 static struct vport *lookup_vport(struct odp_header *odp_header,
1686                                   struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1687 {
1688         struct datapath *dp;
1689         struct vport *vport;
1690
1691         if (a[ODP_VPORT_ATTR_NAME]) {
1692                 vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME]));
1693                 if (!vport)
1694                         return ERR_PTR(-ENODEV);
1695                 return vport;
1696         } else if (a[ODP_VPORT_ATTR_PORT_NO]) {
1697                 u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1698
1699                 if (port_no >= DP_MAX_PORTS)
1700                         return ERR_PTR(-EFBIG);
1701
1702                 dp = get_dp(odp_header->dp_ifindex);
1703                 if (!dp)
1704                         return ERR_PTR(-ENODEV);
1705
1706                 vport = get_vport_protected(dp, port_no);
1707                 if (!vport)
1708                         return ERR_PTR(-ENOENT);
1709                 return vport;
1710         } else
1711                 return ERR_PTR(-EINVAL);
1712 }
1713
1714 /* Called with RTNL lock. */
1715 static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1716 {
1717         int err = 0;
1718         if (a[ODP_VPORT_ATTR_STATS])
1719                 err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS]));
1720         if (!err && a[ODP_VPORT_ATTR_ADDRESS])
1721                 err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS]));
1722         if (!err && a[ODP_VPORT_ATTR_MTU])
1723                 err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU]));
1724         return err;
1725 }
1726
1727 static int odp_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1728 {
1729         struct nlattr **a = info->attrs;
1730         struct odp_header *odp_header = info->userhdr;
1731         struct vport_parms parms;
1732         struct sk_buff *reply;
1733         struct vport *vport;
1734         struct datapath *dp;
1735         u32 port_no;
1736         int err;
1737
1738         err = -EINVAL;
1739         if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE])
1740                 goto exit;
1741
1742         err = odp_vport_cmd_validate(a);
1743         if (err)
1744                 goto exit;
1745
1746         rtnl_lock();
1747         dp = get_dp(odp_header->dp_ifindex);
1748         err = -ENODEV;
1749         if (!dp)
1750                 goto exit_unlock;
1751
1752         if (a[ODP_VPORT_ATTR_PORT_NO]) {
1753                 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1754
1755                 err = -EFBIG;
1756                 if (port_no >= DP_MAX_PORTS)
1757                         goto exit_unlock;
1758
1759                 vport = get_vport_protected(dp, port_no);
1760                 err = -EBUSY;
1761                 if (vport)
1762                         goto exit_unlock;
1763         } else {
1764                 for (port_no = 1; ; port_no++) {
1765                         if (port_no >= DP_MAX_PORTS) {
1766                                 err = -EFBIG;
1767                                 goto exit_unlock;
1768                         }
1769                         vport = get_vport_protected(dp, port_no);
1770                         if (!vport)
1771                                 break;
1772                 }
1773         }
1774
1775         parms.name = nla_data(a[ODP_VPORT_ATTR_NAME]);
1776         parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1777         parms.options = a[ODP_VPORT_ATTR_OPTIONS];
1778         parms.dp = dp;
1779         parms.port_no = port_no;
1780
1781         vport = new_vport(&parms);
1782         err = PTR_ERR(vport);
1783         if (IS_ERR(vport))
1784                 goto exit_unlock;
1785
1786         set_internal_devs_mtu(dp);
1787         dp_sysfs_add_if(vport);
1788
1789         err = change_vport(vport, a);
1790         if (!err) {
1791                 reply = odp_vport_cmd_build_info(vport, info->snd_pid,
1792                                                  info->snd_seq, ODP_VPORT_CMD_NEW);
1793                 if (IS_ERR(reply))
1794                         err = PTR_ERR(reply);
1795         }
1796         if (err) {
1797                 dp_detach_port(vport);
1798                 goto exit_unlock;
1799         }
1800         genl_notify(reply, genl_info_net(info), info->snd_pid,
1801                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1802
1803
1804 exit_unlock:
1805         rtnl_unlock();
1806 exit:
1807         return err;
1808 }
1809
1810 static int odp_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1811 {
1812         struct nlattr **a = info->attrs;
1813         struct sk_buff *reply;
1814         struct vport *vport;
1815         int err;
1816
1817         err = odp_vport_cmd_validate(a);
1818         if (err)
1819                 goto exit;
1820
1821         rtnl_lock();
1822         vport = lookup_vport(info->userhdr, a);
1823         err = PTR_ERR(vport);
1824         if (IS_ERR(vport))
1825                 goto exit_unlock;
1826
1827         err = 0;
1828         if (a[ODP_VPORT_ATTR_OPTIONS])
1829                 err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]);
1830         if (!err)
1831                 err = change_vport(vport, a);
1832
1833         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1834                                          ODP_VPORT_CMD_NEW);
1835         if (IS_ERR(reply)) {
1836                 err = PTR_ERR(reply);
1837                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1838                                 dp_vport_multicast_group.id, err);
1839                 return 0;
1840         }
1841
1842         genl_notify(reply, genl_info_net(info), info->snd_pid,
1843                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1844
1845 exit_unlock:
1846         rtnl_unlock();
1847 exit:
1848         return err;
1849 }
1850
1851 static int odp_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1852 {
1853         struct nlattr **a = info->attrs;
1854         struct sk_buff *reply;
1855         struct vport *vport;
1856         int err;
1857
1858         err = odp_vport_cmd_validate(a);
1859         if (err)
1860                 goto exit;
1861
1862         rtnl_lock();
1863         vport = lookup_vport(info->userhdr, a);
1864         err = PTR_ERR(vport);
1865         if (IS_ERR(vport))
1866                 goto exit_unlock;
1867
1868         if (vport->port_no == ODPP_LOCAL) {
1869                 err = -EINVAL;
1870                 goto exit_unlock;
1871         }
1872
1873         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1874                                          ODP_VPORT_CMD_DEL);
1875         err = PTR_ERR(reply);
1876         if (IS_ERR(reply))
1877                 goto exit_unlock;
1878
1879         err = dp_detach_port(vport);
1880
1881         genl_notify(reply, genl_info_net(info), info->snd_pid,
1882                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1883
1884 exit_unlock:
1885         rtnl_unlock();
1886 exit:
1887         return err;
1888 }
1889
1890 static int odp_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1891 {
1892         struct nlattr **a = info->attrs;
1893         struct odp_header *odp_header = info->userhdr;
1894         struct sk_buff *reply;
1895         struct vport *vport;
1896         int err;
1897
1898         err = odp_vport_cmd_validate(a);
1899         if (err)
1900                 goto exit;
1901
1902         rcu_read_lock();
1903         vport = lookup_vport(odp_header, a);
1904         err = PTR_ERR(vport);
1905         if (IS_ERR(vport))
1906                 goto exit_unlock;
1907
1908         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1909                                          ODP_VPORT_CMD_NEW);
1910         err = PTR_ERR(reply);
1911         if (IS_ERR(reply))
1912                 goto exit_unlock;
1913
1914         rcu_read_unlock();
1915
1916         return genlmsg_reply(reply, info);
1917
1918 exit_unlock:
1919         rcu_read_unlock();
1920 exit:
1921         return err;
1922 }
1923
1924 static int odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1925 {
1926         struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1927         struct datapath *dp;
1928         u32 port_no;
1929         int retval;
1930
1931         dp = get_dp(odp_header->dp_ifindex);
1932         if (!dp)
1933                 return -ENODEV;
1934
1935         rcu_read_lock();
1936         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1937                 struct vport *vport;
1938
1939                 vport = get_vport_protected(dp, port_no);
1940                 if (!vport)
1941                         continue;
1942
1943                 if (odp_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1944                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1945                                             ODP_VPORT_CMD_NEW) < 0)
1946                         break;
1947         }
1948         rcu_read_unlock();
1949
1950         cb->args[0] = port_no;
1951         retval = skb->len;
1952
1953         return retval;
1954 }
1955
1956 static struct genl_ops dp_vport_genl_ops[] = {
1957         { .cmd = ODP_VPORT_CMD_NEW,
1958           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1959           .policy = vport_policy,
1960           .doit = odp_vport_cmd_new
1961         },
1962         { .cmd = ODP_VPORT_CMD_DEL,
1963           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1964           .policy = vport_policy,
1965           .doit = odp_vport_cmd_del
1966         },
1967         { .cmd = ODP_VPORT_CMD_GET,
1968           .flags = 0,               /* OK for unprivileged users. */
1969           .policy = vport_policy,
1970           .doit = odp_vport_cmd_get,
1971           .dumpit = odp_vport_cmd_dump
1972         },
1973         { .cmd = ODP_VPORT_CMD_SET,
1974           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1975           .policy = vport_policy,
1976           .doit = odp_vport_cmd_set,
1977         },
1978 };
1979
1980 struct genl_family_and_ops {
1981         struct genl_family *family;
1982         struct genl_ops *ops;
1983         int n_ops;
1984         struct genl_multicast_group *group;
1985 };
1986
1987 static const struct genl_family_and_ops dp_genl_families[] = {
1988         { &dp_datapath_genl_family,
1989           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1990           &dp_datapath_multicast_group },
1991         { &dp_vport_genl_family,
1992           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1993           &dp_vport_multicast_group },
1994         { &dp_flow_genl_family,
1995           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1996           &dp_flow_multicast_group },
1997         { &dp_packet_genl_family,
1998           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1999           NULL },
2000 };
2001
2002 static void dp_unregister_genl(int n_families)
2003 {
2004         int i;
2005
2006         for (i = 0; i < n_families; i++) {
2007                 genl_unregister_family(dp_genl_families[i].family);
2008         }
2009 }
2010
2011 static int dp_register_genl(void)
2012 {
2013         int n_registered;
2014         int err;
2015         int i;
2016
2017         n_registered = 0;
2018         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2019                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2020
2021                 err = genl_register_family_with_ops(f->family, f->ops,
2022                                                     f->n_ops);
2023                 if (err)
2024                         goto error;
2025                 n_registered++;
2026
2027                 if (f->group) {
2028                         err = genl_register_mc_group(f->family, f->group);
2029                         if (err)
2030                                 goto error;
2031                 }
2032         }
2033
2034         err = packet_register_mc_groups();
2035         if (err)
2036                 goto error;
2037         return 0;
2038
2039 error:
2040         dp_unregister_genl(n_registered);
2041         return err;
2042 }
2043
2044 static int __init dp_init(void)
2045 {
2046         struct sk_buff *dummy_skb;
2047         int err;
2048
2049         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2050
2051         printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2052
2053         err = flow_init();
2054         if (err)
2055                 goto error;
2056
2057         err = vport_init();
2058         if (err)
2059                 goto error_flow_exit;
2060
2061         err = register_netdevice_notifier(&dp_device_notifier);
2062         if (err)
2063                 goto error_vport_exit;
2064
2065         err = dp_register_genl();
2066         if (err < 0)
2067                 goto error_unreg_notifier;
2068
2069         return 0;
2070
2071 error_unreg_notifier:
2072         unregister_netdevice_notifier(&dp_device_notifier);
2073 error_vport_exit:
2074         vport_exit();
2075 error_flow_exit:
2076         flow_exit();
2077 error:
2078         return err;
2079 }
2080
2081 static void dp_cleanup(void)
2082 {
2083         rcu_barrier();
2084         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2085         unregister_netdevice_notifier(&dp_device_notifier);
2086         vport_exit();
2087         flow_exit();
2088 }
2089
2090 module_init(dp_init);
2091 module_exit(dp_cleanup);
2092
2093 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2094 MODULE_LICENSE("GPL");