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
 717         if (flow->key.eth.in_port < DP_MAX_PORTS)
 718                 OVS_CB(packet)->vport = get_vport_protected(dp,
 719                                                         flow->key.eth.in_port);
 720
 721         err = execute_actions(dp, packet);
 722         rcu_read_unlock();
 723
 724         flow_put(flow);
 725         return err;
 726
 727 err_unlock:
 728         rcu_read_unlock();
 729 err_flow_put:
 730         flow_put(flow);
 731 err_kfree_skb:
 732         kfree_skb(packet);
 733 err:
 734         return err;
 735 }
 736
 737 static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = {
 738         [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 739         [ODP_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 740         [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 741 };
 742
 743 static struct genl_ops dp_packet_genl_ops[] = {
 744         { .cmd = ODP_PACKET_CMD_EXECUTE,
 745           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 746           .policy = packet_policy,
 747           .doit = odp_packet_cmd_execute
 748         }
 749 };
 750
 751 static void get_dp_stats(struct datapath *dp, struct odp_stats *stats)
 752 {
 753         int i;
 754
 755         stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
 756         for_each_possible_cpu(i) {
 757                 const struct dp_stats_percpu *percpu_stats;
 758                 struct dp_stats_percpu local_stats;
 759                 unsigned seqcount;
 760
 761                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 762
 763                 do {
 764                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 765                         local_stats = *percpu_stats;
 766                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 767
 768                 stats->n_frags += local_stats.n_frags;
 769                 stats->n_hit += local_stats.n_hit;
 770                 stats->n_missed += local_stats.n_missed;
 771                 stats->n_lost += local_stats.n_lost;
 772         }
 773 }
 774
 775 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
 776  * Called with RTNL lock.
 777  */
 778 int dp_min_mtu(const struct datapath *dp)
 779 {
 780         struct vport *p;
 781         int mtu = 0;
 782
 783         ASSERT_RTNL();
 784
 785         list_for_each_entry (p, &dp->port_list, node) {
 786                 int dev_mtu;
 787
 788                 /* Skip any internal ports, since that's what we're trying to
 789                  * set. */
 790                 if (is_internal_vport(p))
 791                         continue;
 792
 793                 dev_mtu = vport_get_mtu(p);
 794                 if (!dev_mtu)
 795                         continue;
 796                 if (!mtu || dev_mtu < mtu)
 797                         mtu = dev_mtu;
 798         }
 799
 800         return mtu ? mtu : ETH_DATA_LEN;
 801 }
 802
 803 /* Sets the MTU of all datapath devices to the minimum of the ports
 804  * Called with RTNL lock.
 805  */
 806 void set_internal_devs_mtu(const struct datapath *dp)
 807 {
 808         struct vport *p;
 809         int mtu;
 810
 811         ASSERT_RTNL();
 812
 813         mtu = dp_min_mtu(dp);
 814
 815         list_for_each_entry (p, &dp->port_list, node) {
 816                 if (is_internal_vport(p))
 817                         vport_set_mtu(p, mtu);
 818         }
 819 }
 820
 821 static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = {
 822         [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 823         [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 824         [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 825 };
 826
 827 static struct genl_family dp_flow_genl_family = {
 828         .id = GENL_ID_GENERATE,
 829         .hdrsize = sizeof(struct odp_header),
 830         .name = ODP_FLOW_FAMILY,
 831         .version = 1,
 832         .maxattr = ODP_FLOW_ATTR_MAX
 833 };
 834
 835 static struct genl_multicast_group dp_flow_multicast_group = {
 836         .name = ODP_FLOW_MCGROUP
 837 };
 838
 839 /* Called with genl_lock. */
 840 static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 841                                   struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
 842 {
 843         const int skb_orig_len = skb->len;
 844         const struct sw_flow_actions *sf_acts;
 845         struct odp_flow_stats stats;
 846         struct odp_header *odp_header;
 847         struct nlattr *nla;
 848         unsigned long used;
 849         u8 tcp_flags;
 850         int err;
 851
 852         sf_acts = rcu_dereference_protected(flow->sf_acts,
 853                                             lockdep_genl_is_held());
 854
 855         odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 856         if (!odp_header)
 857                 return -EMSGSIZE;
 858
 859         odp_header->dp_ifindex = dp->dp_ifindex;
 860
 861         nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY);
 862         if (!nla)
 863                 goto nla_put_failure;
 864         err = flow_to_nlattrs(&flow->key, skb);
 865         if (err)
 866                 goto error;
 867         nla_nest_end(skb, nla);
 868
 869         spin_lock_bh(&flow->lock);
 870         used = flow->used;
 871         stats.n_packets = flow->packet_count;
 872         stats.n_bytes = flow->byte_count;
 873         tcp_flags = flow->tcp_flags;
 874         spin_unlock_bh(&flow->lock);
 875
 876         if (used)
 877                 NLA_PUT_U64(skb, ODP_FLOW_ATTR_USED, flow_used_time(used));
 878
 879         if (stats.n_packets)
 880                 NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats);
 881
 882         if (tcp_flags)
 883                 NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 884
 885         /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 886          * this is the first flow to be dumped into 'skb'.  This is unusual for
 887          * Netlink but individual action lists can be longer than
 888          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 889          * The userspace caller can always fetch the actions separately if it
 890          * really wants them.  (Most userspace callers in fact don't care.)
 891          *
 892          * This can only fail for dump operations because the skb is always
 893          * properly sized for single flows.
 894          */
 895         err = nla_put(skb, ODP_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 896                       sf_acts->actions);
 897         if (err < 0 && skb_orig_len)
 898                 goto error;
 899
 900         return genlmsg_end(skb, odp_header);
 901
 902 nla_put_failure:
 903         err = -EMSGSIZE;
 904 error:
 905         genlmsg_cancel(skb, odp_header);
 906         return err;
 907 }
 908
 909 static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow)
 910 {
 911         const struct sw_flow_actions *sf_acts;
 912         int len;
 913
 914         sf_acts = rcu_dereference_protected(flow->sf_acts,
 915                                             lockdep_genl_is_held());
 916
 917         len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */
 918         len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */
 919         len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */
 920         len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */
 921         len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */
 922         return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL);
 923 }
 924
 925 static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
 926                                                u32 pid, u32 seq, u8 cmd)
 927 {
 928         struct sk_buff *skb;
 929         int retval;
 930
 931         skb = odp_flow_cmd_alloc_info(flow);
 932         if (!skb)
 933                 return ERR_PTR(-ENOMEM);
 934
 935         retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 936         BUG_ON(retval < 0);
 937         return skb;
 938 }
 939
 940 static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 941 {
 942         struct nlattr **a = info->attrs;
 943         struct odp_header *odp_header = info->userhdr;
 944         struct tbl_node *flow_node;
 945         struct sw_flow_key key;
 946         struct sw_flow *flow;
 947         struct sk_buff *reply;
 948         struct datapath *dp;
 949         struct tbl *table;
 950         u32 hash;
 951         int error;
 952         int key_len;
 953
 954         /* Extract key. */
 955         error = -EINVAL;
 956         if (!a[ODP_FLOW_ATTR_KEY])
 957                 goto error;
 958         error = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
 959         if (error)
 960                 goto error;
 961
 962         /* Validate actions. */
 963         if (a[ODP_FLOW_ATTR_ACTIONS]) {
 964                 error = validate_actions(a[ODP_FLOW_ATTR_ACTIONS]);
 965                 if (error)
 966                         goto error;
 967         } else if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW) {
 968                 error = -EINVAL;
 969                 goto error;
 970         }
 971
 972         dp = get_dp(odp_header->dp_ifindex);
 973         error = -ENODEV;
 974         if (!dp)
 975                 goto error;
 976
 977         hash = flow_hash(&key, key_len);
 978         table = get_table_protected(dp);
 979         flow_node = tbl_lookup(table, &key, key_len, hash, flow_cmp);
 980         if (!flow_node) {
 981                 struct sw_flow_actions *acts;
 982
 983                 /* Bail out if we're not allowed to create a new flow. */
 984                 error = -ENOENT;
 985                 if (info->genlhdr->cmd == ODP_FLOW_CMD_SET)
 986                         goto error;
 987
 988                 /* Expand table, if necessary, to make room. */
 989                 if (tbl_count(table) >= tbl_n_buckets(table)) {
 990                         error = expand_table(dp);
 991                         if (error)
 992                                 goto error;
 993                         table = get_table_protected(dp);
 994                 }
 995
 996                 /* Allocate flow. */
 997                 flow = flow_alloc();
 998                 if (IS_ERR(flow)) {
 999                         error = PTR_ERR(flow);
1000                         goto error;
1001                 }
1002                 flow->key = key;
1003                 clear_stats(flow);
1004
1005                 /* Obtain actions. */
1006                 acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1007                 error = PTR_ERR(acts);
1008                 if (IS_ERR(acts))
1009                         goto error_free_flow;
1010                 rcu_assign_pointer(flow->sf_acts, acts);
1011
1012                 /* Put flow in bucket. */
1013                 error = tbl_insert(table, &flow->tbl_node, hash);
1014                 if (error)
1015                         goto error_free_flow;
1016
1017                 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1018                                                 info->snd_seq, ODP_FLOW_CMD_NEW);
1019         } else {
1020                 /* We found a matching flow. */
1021                 struct sw_flow_actions *old_acts;
1022
1023                 /* Bail out if we're not allowed to modify an existing flow.
1024                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1025                  * because Generic Netlink treats the latter as a dump
1026                  * request.  We also accept NLM_F_EXCL in case that bug ever
1027                  * gets fixed.
1028                  */
1029                 error = -EEXIST;
1030                 if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW &&
1031                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1032                         goto error;
1033
1034                 /* Update actions. */
1035                 flow = flow_cast(flow_node);
1036                 old_acts = rcu_dereference_protected(flow->sf_acts,
1037                                                      lockdep_genl_is_held());
1038                 if (a[ODP_FLOW_ATTR_ACTIONS] &&
1039                     (old_acts->actions_len != nla_len(a[ODP_FLOW_ATTR_ACTIONS]) ||
1040                      memcmp(old_acts->actions, nla_data(a[ODP_FLOW_ATTR_ACTIONS]),
1041                             old_acts->actions_len))) {
1042                         struct sw_flow_actions *new_acts;
1043
1044                         new_acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1045                         error = PTR_ERR(new_acts);
1046                         if (IS_ERR(new_acts))
1047                                 goto error;
1048
1049                         rcu_assign_pointer(flow->sf_acts, new_acts);
1050                         flow_deferred_free_acts(old_acts);
1051                 }
1052
1053                 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1054                                                 info->snd_seq, ODP_FLOW_CMD_NEW);
1055
1056                 /* Clear stats. */
1057                 if (a[ODP_FLOW_ATTR_CLEAR]) {
1058                         spin_lock_bh(&flow->lock);
1059                         clear_stats(flow);
1060                         spin_unlock_bh(&flow->lock);
1061                 }
1062         }
1063
1064         if (!IS_ERR(reply))
1065                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1066                             dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1067         else
1068                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1069                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1070         return 0;
1071
1072 error_free_flow:
1073         flow_put(flow);
1074 error:
1075         return error;
1076 }
1077
1078 static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1079 {
1080         struct nlattr **a = info->attrs;
1081         struct odp_header *odp_header = info->userhdr;
1082         struct sw_flow_key key;
1083         struct tbl_node *flow_node;
1084         struct sk_buff *reply;
1085         struct sw_flow *flow;
1086         struct datapath *dp;
1087         struct tbl *table;
1088         int err;
1089         int key_len;
1090
1091         if (!a[ODP_FLOW_ATTR_KEY])
1092                 return -EINVAL;
1093         err = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
1094         if (err)
1095                 return err;
1096
1097         dp = get_dp(odp_header->dp_ifindex);
1098         if (!dp)
1099                 return -ENODEV;
1100
1101         table = get_table_protected(dp);
1102         flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1103                                flow_cmp);
1104         if (!flow_node)
1105                 return -ENOENT;
1106
1107         flow = flow_cast(flow_node);
1108         reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW);
1109         if (IS_ERR(reply))
1110                 return PTR_ERR(reply);
1111
1112         return genlmsg_reply(reply, info);
1113 }
1114
1115 static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1116 {
1117         struct nlattr **a = info->attrs;
1118         struct odp_header *odp_header = info->userhdr;
1119         struct sw_flow_key key;
1120         struct tbl_node *flow_node;
1121         struct sk_buff *reply;
1122         struct sw_flow *flow;
1123         struct datapath *dp;
1124         struct tbl *table;
1125         int err;
1126         int key_len;
1127
1128         if (!a[ODP_FLOW_ATTR_KEY])
1129                 return flush_flows(odp_header->dp_ifindex);
1130         err = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
1131         if (err)
1132                 return err;
1133
1134         dp = get_dp(odp_header->dp_ifindex);
1135         if (!dp)
1136                 return -ENODEV;
1137
1138         table = get_table_protected(dp);
1139         flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1140                                flow_cmp);
1141         if (!flow_node)
1142                 return -ENOENT;
1143         flow = flow_cast(flow_node);
1144
1145         reply = odp_flow_cmd_alloc_info(flow);
1146         if (!reply)
1147                 return -ENOMEM;
1148
1149         err = tbl_remove(table, flow_node);
1150         if (err) {
1151                 kfree_skb(reply);
1152                 return err;
1153         }
1154
1155         err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1156                                      info->snd_seq, 0, ODP_FLOW_CMD_DEL);
1157         BUG_ON(err < 0);
1158
1159         flow_deferred_free(flow);
1160
1161         genl_notify(reply, genl_info_net(info), info->snd_pid,
1162                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1163         return 0;
1164 }
1165
1166 static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1167 {
1168         struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1169         struct datapath *dp;
1170
1171         dp = get_dp(odp_header->dp_ifindex);
1172         if (!dp)
1173                 return -ENODEV;
1174
1175         for (;;) {
1176                 struct tbl_node *flow_node;
1177                 struct sw_flow *flow;
1178                 u32 bucket, obj;
1179
1180                 bucket = cb->args[0];
1181                 obj = cb->args[1];
1182                 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1183                 if (!flow_node)
1184                         break;
1185
1186                 flow = flow_cast(flow_node);
1187                 if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1188                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1189                                            ODP_FLOW_CMD_NEW) < 0)
1190                         break;
1191
1192                 cb->args[0] = bucket;
1193                 cb->args[1] = obj;
1194         }
1195         return skb->len;
1196 }
1197
1198 static struct genl_ops dp_flow_genl_ops[] = {
1199         { .cmd = ODP_FLOW_CMD_NEW,
1200           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1201           .policy = flow_policy,
1202           .doit = odp_flow_cmd_new_or_set
1203         },
1204         { .cmd = ODP_FLOW_CMD_DEL,
1205           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1206           .policy = flow_policy,
1207           .doit = odp_flow_cmd_del
1208         },
1209         { .cmd = ODP_FLOW_CMD_GET,
1210           .flags = 0,               /* OK for unprivileged users. */
1211           .policy = flow_policy,
1212           .doit = odp_flow_cmd_get,
1213           .dumpit = odp_flow_cmd_dump
1214         },
1215         { .cmd = ODP_FLOW_CMD_SET,
1216           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1217           .policy = flow_policy,
1218           .doit = odp_flow_cmd_new_or_set,
1219         },
1220 };
1221
1222 static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
1223 #ifdef HAVE_NLA_NUL_STRING
1224         [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1225 #endif
1226         [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1227         [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1228 };
1229
1230 static struct genl_family dp_datapath_genl_family = {
1231         .id = GENL_ID_GENERATE,
1232         .hdrsize = sizeof(struct odp_header),
1233         .name = ODP_DATAPATH_FAMILY,
1234         .version = 1,
1235         .maxattr = ODP_DP_ATTR_MAX
1236 };
1237
1238 static struct genl_multicast_group dp_datapath_multicast_group = {
1239         .name = ODP_DATAPATH_MCGROUP
1240 };
1241
1242 static int odp_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1243                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1244 {
1245         struct odp_header *odp_header;
1246         struct nlattr *nla;
1247         int err;
1248
1249         odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1250                                    flags, cmd);
1251         if (!odp_header)
1252                 goto error;
1253
1254         odp_header->dp_ifindex = dp->dp_ifindex;
1255
1256         rcu_read_lock();
1257         err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
1258         rcu_read_unlock();
1259         if (err)
1260                 goto nla_put_failure;
1261
1262         nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats));
1263         if (!nla)
1264                 goto nla_put_failure;
1265         get_dp_stats(dp, nla_data(nla));
1266
1267         NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS,
1268                     dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO);
1269
1270         if (dp->sflow_probability)
1271                 NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability);
1272
1273         nla = nla_nest_start(skb, ODP_DP_ATTR_MCGROUPS);
1274         if (!nla)
1275                 goto nla_put_failure;
1276         NLA_PUT_U32(skb, ODP_PACKET_CMD_MISS, packet_mc_group(dp, ODP_PACKET_CMD_MISS));
1277         NLA_PUT_U32(skb, ODP_PACKET_CMD_ACTION, packet_mc_group(dp, ODP_PACKET_CMD_ACTION));
1278         NLA_PUT_U32(skb, ODP_PACKET_CMD_SAMPLE, packet_mc_group(dp, ODP_PACKET_CMD_SAMPLE));
1279         nla_nest_end(skb, nla);
1280
1281         return genlmsg_end(skb, odp_header);
1282
1283 nla_put_failure:
1284         genlmsg_cancel(skb, odp_header);
1285 error:
1286         return -EMSGSIZE;
1287 }
1288
1289 static struct sk_buff *odp_dp_cmd_build_info(struct datapath *dp, u32 pid,
1290                                              u32 seq, u8 cmd)
1291 {
1292         struct sk_buff *skb;
1293         int retval;
1294
1295         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1296         if (!skb)
1297                 return ERR_PTR(-ENOMEM);
1298
1299         retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1300         if (retval < 0) {
1301                 kfree_skb(skb);
1302                 return ERR_PTR(retval);
1303         }
1304         return skb;
1305 }
1306
1307 static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1308 {
1309         if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1310                 u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1311
1312                 if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP)
1313                         return -EINVAL;
1314         }
1315
1316         return CHECK_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
1317 }
1318
1319 /* Called with genl_mutex and optionally with RTNL lock also. */
1320 static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1321 {
1322         struct datapath *dp;
1323
1324         if (!a[ODP_DP_ATTR_NAME])
1325                 dp = get_dp(odp_header->dp_ifindex);
1326         else {
1327                 struct vport *vport;
1328
1329                 rcu_read_lock();
1330                 vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
1331                 dp = vport && vport->port_no == ODPP_LOCAL ? vport->dp : NULL;
1332                 rcu_read_unlock();
1333         }
1334         return dp ? dp : ERR_PTR(-ENODEV);
1335 }
1336
1337 /* Called with genl_mutex. */
1338 static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1339 {
1340         if (a[ODP_DP_ATTR_IPV4_FRAGS])
1341                 dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP;
1342         if (a[ODP_DP_ATTR_SAMPLING])
1343                 dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]);
1344 }
1345
1346 static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1347 {
1348         struct nlattr **a = info->attrs;
1349         struct vport_parms parms;
1350         struct sk_buff *reply;
1351         struct datapath *dp;
1352         struct vport *vport;
1353         int err;
1354
1355         err = -EINVAL;
1356         if (!a[ODP_DP_ATTR_NAME])
1357                 goto err;
1358
1359         err = odp_dp_cmd_validate(a);
1360         if (err)
1361                 goto err;
1362
1363         rtnl_lock();
1364         err = -ENODEV;
1365         if (!try_module_get(THIS_MODULE))
1366                 goto err_unlock_rtnl;
1367
1368         err = -ENOMEM;
1369         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1370         if (dp == NULL)
1371                 goto err_put_module;
1372         INIT_LIST_HEAD(&dp->port_list);
1373
1374         /* Initialize kobject for bridge.  This will be added as
1375          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1376         dp->ifobj.kset = NULL;
1377         kobject_init(&dp->ifobj, &dp_ktype);
1378
1379         /* Allocate table. */
1380         err = -ENOMEM;
1381         rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1382         if (!dp->table)
1383                 goto err_free_dp;
1384
1385         /* Set up our datapath device. */
1386         parms.name = nla_data(a[ODP_DP_ATTR_NAME]);
1387         parms.type = ODP_VPORT_TYPE_INTERNAL;
1388         parms.options = NULL;
1389         parms.dp = dp;
1390         parms.port_no = ODPP_LOCAL;
1391         vport = new_vport(&parms);
1392         if (IS_ERR(vport)) {
1393                 err = PTR_ERR(vport);
1394                 if (err == -EBUSY)
1395                         err = -EEXIST;
1396
1397                 goto err_destroy_table;
1398         }
1399         dp->dp_ifindex = vport_get_ifindex(vport);
1400
1401         dp->drop_frags = 0;
1402         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1403         if (!dp->stats_percpu) {
1404                 err = -ENOMEM;
1405                 goto err_destroy_local_port;
1406         }
1407
1408         change_datapath(dp, a);
1409
1410         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1411         err = PTR_ERR(reply);
1412         if (IS_ERR(reply))
1413                 goto err_destroy_local_port;
1414
1415         list_add_tail(&dp->list_node, &dps);
1416         dp_sysfs_add_dp(dp);
1417
1418         rtnl_unlock();
1419
1420         genl_notify(reply, genl_info_net(info), info->snd_pid,
1421                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1422         return 0;
1423
1424 err_destroy_local_port:
1425         dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1426 err_destroy_table:
1427         tbl_destroy(get_table_protected(dp), NULL);
1428 err_free_dp:
1429         kfree(dp);
1430 err_put_module:
1431         module_put(THIS_MODULE);
1432 err_unlock_rtnl:
1433         rtnl_unlock();
1434 err:
1435         return err;
1436 }
1437
1438 static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1439 {
1440         struct vport *vport, *next_vport;
1441         struct sk_buff *reply;
1442         struct datapath *dp;
1443         int err;
1444
1445         err = odp_dp_cmd_validate(info->attrs);
1446         if (err)
1447                 goto exit;
1448
1449         rtnl_lock();
1450         dp = lookup_datapath(info->userhdr, info->attrs);
1451         err = PTR_ERR(dp);
1452         if (IS_ERR(dp))
1453                 goto exit_unlock;
1454
1455         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_DEL);
1456         err = PTR_ERR(reply);
1457         if (IS_ERR(reply))
1458                 goto exit_unlock;
1459
1460         list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1461                 if (vport->port_no != ODPP_LOCAL)
1462                         dp_detach_port(vport);
1463
1464         dp_sysfs_del_dp(dp);
1465         list_del(&dp->list_node);
1466         dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1467
1468         /* rtnl_unlock() will wait until all the references to devices that
1469          * are pending unregistration have been dropped.  We do it here to
1470          * ensure that any internal devices (which contain DP pointers) are
1471          * fully destroyed before freeing the datapath.
1472          */
1473         rtnl_unlock();
1474
1475         call_rcu(&dp->rcu, destroy_dp_rcu);
1476         module_put(THIS_MODULE);
1477
1478         genl_notify(reply, genl_info_net(info), info->snd_pid,
1479                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1480
1481         return 0;
1482
1483 exit_unlock:
1484         rtnl_unlock();
1485 exit:
1486         return err;
1487 }
1488
1489 static int odp_dp_cmd_set(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         change_datapath(dp, info->attrs);
1504
1505         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1506         if (IS_ERR(reply)) {
1507                 err = PTR_ERR(reply);
1508                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1509                                 dp_datapath_multicast_group.id, err);
1510                 return 0;
1511         }
1512
1513         genl_notify(reply, genl_info_net(info), info->snd_pid,
1514                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1515         return 0;
1516 }
1517
1518 static int odp_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1519 {
1520         struct sk_buff *reply;
1521         struct datapath *dp;
1522         int err;
1523
1524         err = odp_dp_cmd_validate(info->attrs);
1525         if (err)
1526                 return err;
1527
1528         dp = lookup_datapath(info->userhdr, info->attrs);
1529         if (IS_ERR(dp))
1530                 return PTR_ERR(dp);
1531
1532         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1533         if (IS_ERR(reply))
1534                 return PTR_ERR(reply);
1535
1536         return genlmsg_reply(reply, info);
1537 }
1538
1539 static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1540 {
1541         struct datapath *dp;
1542         int skip = cb->args[0];
1543         int i = 0;
1544
1545         list_for_each_entry (dp, &dps, list_node) {
1546                 if (i < skip)
1547                         continue;
1548                 if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1549                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1550                                          ODP_DP_CMD_NEW) < 0)
1551                         break;
1552                 i++;
1553         }
1554
1555         cb->args[0] = i;
1556
1557         return skb->len;
1558 }
1559
1560 static struct genl_ops dp_datapath_genl_ops[] = {
1561         { .cmd = ODP_DP_CMD_NEW,
1562           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1563           .policy = datapath_policy,
1564           .doit = odp_dp_cmd_new
1565         },
1566         { .cmd = ODP_DP_CMD_DEL,
1567           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1568           .policy = datapath_policy,
1569           .doit = odp_dp_cmd_del
1570         },
1571         { .cmd = ODP_DP_CMD_GET,
1572           .flags = 0,               /* OK for unprivileged users. */
1573           .policy = datapath_policy,
1574           .doit = odp_dp_cmd_get,
1575           .dumpit = odp_dp_cmd_dump
1576         },
1577         { .cmd = ODP_DP_CMD_SET,
1578           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1579           .policy = datapath_policy,
1580           .doit = odp_dp_cmd_set,
1581         },
1582 };
1583
1584 static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = {
1585 #ifdef HAVE_NLA_NUL_STRING
1586         [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1587         [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1588         [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1589         [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1590         [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1591 #else
1592         [ODP_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1593         [ODP_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1594 #endif
1595         [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1596         [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1597 };
1598
1599 static struct genl_family dp_vport_genl_family = {
1600         .id = GENL_ID_GENERATE,
1601         .hdrsize = sizeof(struct odp_header),
1602         .name = ODP_VPORT_FAMILY,
1603         .version = 1,
1604         .maxattr = ODP_VPORT_ATTR_MAX
1605 };
1606
1607 static struct genl_multicast_group dp_vport_multicast_group = {
1608         .name = ODP_VPORT_MCGROUP
1609 };
1610
1611 /* Called with RTNL lock or RCU read lock. */
1612 static int odp_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1613                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1614 {
1615         struct odp_header *odp_header;
1616         struct nlattr *nla;
1617         int ifindex, iflink;
1618         int mtu;
1619         int err;
1620
1621         odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1622                                  flags, cmd);
1623         if (!odp_header)
1624                 return -EMSGSIZE;
1625
1626         odp_header->dp_ifindex = vport->dp->dp_ifindex;
1627
1628         NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1629         NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport));
1630         NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport));
1631
1632         nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1633         if (!nla)
1634                 goto nla_put_failure;
1635         if (vport_get_stats(vport, nla_data(nla)))
1636                 __skb_trim(skb, skb->len - nla->nla_len);
1637
1638         NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1639
1640         mtu = vport_get_mtu(vport);
1641         if (mtu)
1642                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, mtu);
1643
1644         err = vport_get_options(vport, skb);
1645         if (err == -EMSGSIZE)
1646                 goto error;
1647
1648         ifindex = vport_get_ifindex(vport);
1649         if (ifindex > 0)
1650                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex);
1651
1652         iflink = vport_get_iflink(vport);
1653         if (iflink > 0)
1654                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink);
1655
1656         return genlmsg_end(skb, odp_header);
1657
1658 nla_put_failure:
1659         err = -EMSGSIZE;
1660 error:
1661         genlmsg_cancel(skb, odp_header);
1662         return err;
1663 }
1664
1665 /* Called with RTNL lock or RCU read lock. */
1666 static struct sk_buff *odp_vport_cmd_build_info(struct vport *vport, u32 pid,
1667                                                 u32 seq, u8 cmd)
1668 {
1669         struct sk_buff *skb;
1670         int retval;
1671
1672         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1673         if (!skb)
1674                 return ERR_PTR(-ENOMEM);
1675
1676         retval = odp_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1677         if (retval < 0) {
1678                 kfree_skb(skb);
1679                 return ERR_PTR(retval);
1680         }
1681         return skb;
1682 }
1683
1684 static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1685 {
1686         return CHECK_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1687 }
1688
1689 /* Called with RTNL lock or RCU read lock. */
1690 static struct vport *lookup_vport(struct odp_header *odp_header,
1691                                   struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1692 {
1693         struct datapath *dp;
1694         struct vport *vport;
1695
1696         if (a[ODP_VPORT_ATTR_NAME]) {
1697                 vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME]));
1698                 if (!vport)
1699                         return ERR_PTR(-ENODEV);
1700                 return vport;
1701         } else if (a[ODP_VPORT_ATTR_PORT_NO]) {
1702                 u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1703
1704                 if (port_no >= DP_MAX_PORTS)
1705                         return ERR_PTR(-EFBIG);
1706
1707                 dp = get_dp(odp_header->dp_ifindex);
1708                 if (!dp)
1709                         return ERR_PTR(-ENODEV);
1710
1711                 vport = get_vport_protected(dp, port_no);
1712                 if (!vport)
1713                         return ERR_PTR(-ENOENT);
1714                 return vport;
1715         } else
1716                 return ERR_PTR(-EINVAL);
1717 }
1718
1719 /* Called with RTNL lock. */
1720 static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1721 {
1722         int err = 0;
1723         if (a[ODP_VPORT_ATTR_STATS])
1724                 err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS]));
1725         if (!err && a[ODP_VPORT_ATTR_ADDRESS])
1726                 err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS]));
1727         if (!err && a[ODP_VPORT_ATTR_MTU])
1728                 err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU]));
1729         return err;
1730 }
1731
1732 static int odp_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1733 {
1734         struct nlattr **a = info->attrs;
1735         struct odp_header *odp_header = info->userhdr;
1736         struct vport_parms parms;
1737         struct sk_buff *reply;
1738         struct vport *vport;
1739         struct datapath *dp;
1740         u32 port_no;
1741         int err;
1742
1743         err = -EINVAL;
1744         if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE])
1745                 goto exit;
1746
1747         err = odp_vport_cmd_validate(a);
1748         if (err)
1749                 goto exit;
1750
1751         rtnl_lock();
1752         dp = get_dp(odp_header->dp_ifindex);
1753         err = -ENODEV;
1754         if (!dp)
1755                 goto exit_unlock;
1756
1757         if (a[ODP_VPORT_ATTR_PORT_NO]) {
1758                 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1759
1760                 err = -EFBIG;
1761                 if (port_no >= DP_MAX_PORTS)
1762                         goto exit_unlock;
1763
1764                 vport = get_vport_protected(dp, port_no);
1765                 err = -EBUSY;
1766                 if (vport)
1767                         goto exit_unlock;
1768         } else {
1769                 for (port_no = 1; ; port_no++) {
1770                         if (port_no >= DP_MAX_PORTS) {
1771                                 err = -EFBIG;
1772                                 goto exit_unlock;
1773                         }
1774                         vport = get_vport_protected(dp, port_no);
1775                         if (!vport)
1776                                 break;
1777                 }
1778         }
1779
1780         parms.name = nla_data(a[ODP_VPORT_ATTR_NAME]);
1781         parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1782         parms.options = a[ODP_VPORT_ATTR_OPTIONS];
1783         parms.dp = dp;
1784         parms.port_no = port_no;
1785
1786         vport = new_vport(&parms);
1787         err = PTR_ERR(vport);
1788         if (IS_ERR(vport))
1789                 goto exit_unlock;
1790
1791         set_internal_devs_mtu(dp);
1792         dp_sysfs_add_if(vport);
1793
1794         err = change_vport(vport, a);
1795         if (!err) {
1796                 reply = odp_vport_cmd_build_info(vport, info->snd_pid,
1797                                                  info->snd_seq, ODP_VPORT_CMD_NEW);
1798                 if (IS_ERR(reply))
1799                         err = PTR_ERR(reply);
1800         }
1801         if (err) {
1802                 dp_detach_port(vport);
1803                 goto exit_unlock;
1804         }
1805         genl_notify(reply, genl_info_net(info), info->snd_pid,
1806                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1807
1808
1809 exit_unlock:
1810         rtnl_unlock();
1811 exit:
1812         return err;
1813 }
1814
1815 static int odp_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1816 {
1817         struct nlattr **a = info->attrs;
1818         struct sk_buff *reply;
1819         struct vport *vport;
1820         int err;
1821
1822         err = odp_vport_cmd_validate(a);
1823         if (err)
1824                 goto exit;
1825
1826         rtnl_lock();
1827         vport = lookup_vport(info->userhdr, a);
1828         err = PTR_ERR(vport);
1829         if (IS_ERR(vport))
1830                 goto exit_unlock;
1831
1832         err = 0;
1833         if (a[ODP_VPORT_ATTR_OPTIONS])
1834                 err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]);
1835         if (!err)
1836                 err = change_vport(vport, a);
1837
1838         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1839                                          ODP_VPORT_CMD_NEW);
1840         if (IS_ERR(reply)) {
1841                 err = PTR_ERR(reply);
1842                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1843                                 dp_vport_multicast_group.id, err);
1844                 return 0;
1845         }
1846
1847         genl_notify(reply, genl_info_net(info), info->snd_pid,
1848                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1849
1850 exit_unlock:
1851         rtnl_unlock();
1852 exit:
1853         return err;
1854 }
1855
1856 static int odp_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1857 {
1858         struct nlattr **a = info->attrs;
1859         struct sk_buff *reply;
1860         struct vport *vport;
1861         int err;
1862
1863         err = odp_vport_cmd_validate(a);
1864         if (err)
1865                 goto exit;
1866
1867         rtnl_lock();
1868         vport = lookup_vport(info->userhdr, a);
1869         err = PTR_ERR(vport);
1870         if (IS_ERR(vport))
1871                 goto exit_unlock;
1872
1873         if (vport->port_no == ODPP_LOCAL) {
1874                 err = -EINVAL;
1875                 goto exit_unlock;
1876         }
1877
1878         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1879                                          ODP_VPORT_CMD_DEL);
1880         err = PTR_ERR(reply);
1881         if (IS_ERR(reply))
1882                 goto exit_unlock;
1883
1884         err = dp_detach_port(vport);
1885
1886         genl_notify(reply, genl_info_net(info), info->snd_pid,
1887                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1888
1889 exit_unlock:
1890         rtnl_unlock();
1891 exit:
1892         return err;
1893 }
1894
1895 static int odp_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1896 {
1897         struct nlattr **a = info->attrs;
1898         struct odp_header *odp_header = info->userhdr;
1899         struct sk_buff *reply;
1900         struct vport *vport;
1901         int err;
1902
1903         err = odp_vport_cmd_validate(a);
1904         if (err)
1905                 goto exit;
1906
1907         rcu_read_lock();
1908         vport = lookup_vport(odp_header, a);
1909         err = PTR_ERR(vport);
1910         if (IS_ERR(vport))
1911                 goto exit_unlock;
1912
1913         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1914                                          ODP_VPORT_CMD_NEW);
1915         err = PTR_ERR(reply);
1916         if (IS_ERR(reply))
1917                 goto exit_unlock;
1918
1919         rcu_read_unlock();
1920
1921         return genlmsg_reply(reply, info);
1922
1923 exit_unlock:
1924         rcu_read_unlock();
1925 exit:
1926         return err;
1927 }
1928
1929 static int odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1930 {
1931         struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1932         struct datapath *dp;
1933         u32 port_no;
1934         int retval;
1935
1936         dp = get_dp(odp_header->dp_ifindex);
1937         if (!dp)
1938                 return -ENODEV;
1939
1940         rcu_read_lock();
1941         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1942                 struct vport *vport;
1943
1944                 vport = get_vport_protected(dp, port_no);
1945                 if (!vport)
1946                         continue;
1947
1948                 if (odp_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1949                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1950                                             ODP_VPORT_CMD_NEW) < 0)
1951                         break;
1952         }
1953         rcu_read_unlock();
1954
1955         cb->args[0] = port_no;
1956         retval = skb->len;
1957
1958         return retval;
1959 }
1960
1961 static struct genl_ops dp_vport_genl_ops[] = {
1962         { .cmd = ODP_VPORT_CMD_NEW,
1963           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1964           .policy = vport_policy,
1965           .doit = odp_vport_cmd_new
1966         },
1967         { .cmd = ODP_VPORT_CMD_DEL,
1968           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1969           .policy = vport_policy,
1970           .doit = odp_vport_cmd_del
1971         },
1972         { .cmd = ODP_VPORT_CMD_GET,
1973           .flags = 0,               /* OK for unprivileged users. */
1974           .policy = vport_policy,
1975           .doit = odp_vport_cmd_get,
1976           .dumpit = odp_vport_cmd_dump
1977         },
1978         { .cmd = ODP_VPORT_CMD_SET,
1979           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1980           .policy = vport_policy,
1981           .doit = odp_vport_cmd_set,
1982         },
1983 };
1984
1985 struct genl_family_and_ops {
1986         struct genl_family *family;
1987         struct genl_ops *ops;
1988         int n_ops;
1989         struct genl_multicast_group *group;
1990 };
1991
1992 static const struct genl_family_and_ops dp_genl_families[] = {
1993         { &dp_datapath_genl_family,
1994           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1995           &dp_datapath_multicast_group },
1996         { &dp_vport_genl_family,
1997           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1998           &dp_vport_multicast_group },
1999         { &dp_flow_genl_family,
2000           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2001           &dp_flow_multicast_group },
2002         { &dp_packet_genl_family,
2003           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2004           NULL },
2005 };
2006
2007 static void dp_unregister_genl(int n_families)
2008 {
2009         int i;
2010
2011         for (i = 0; i < n_families; i++) {
2012                 genl_unregister_family(dp_genl_families[i].family);
2013         }
2014 }
2015
2016 static int dp_register_genl(void)
2017 {
2018         int n_registered;
2019         int err;
2020         int i;
2021
2022         n_registered = 0;
2023         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2024                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2025
2026                 err = genl_register_family_with_ops(f->family, f->ops,
2027                                                     f->n_ops);
2028                 if (err)
2029                         goto error;
2030                 n_registered++;
2031
2032                 if (f->group) {
2033                         err = genl_register_mc_group(f->family, f->group);
2034                         if (err)
2035                                 goto error;
2036                 }
2037         }
2038
2039         err = packet_register_mc_groups();
2040         if (err)
2041                 goto error;
2042         return 0;
2043
2044 error:
2045         dp_unregister_genl(n_registered);
2046         return err;
2047 }
2048
2049 static int __init dp_init(void)
2050 {
2051         struct sk_buff *dummy_skb;
2052         int err;
2053
2054         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2055
2056         printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2057
2058         err = flow_init();
2059         if (err)
2060                 goto error;
2061
2062         err = vport_init();
2063         if (err)
2064                 goto error_flow_exit;
2065
2066         err = register_netdevice_notifier(&dp_device_notifier);
2067         if (err)
2068                 goto error_vport_exit;
2069
2070         err = dp_register_genl();
2071         if (err < 0)
2072                 goto error_unreg_notifier;
2073
2074         return 0;
2075
2076 error_unreg_notifier:
2077         unregister_netdevice_notifier(&dp_device_notifier);
2078 error_vport_exit:
2079         vport_exit();
2080 error_flow_exit:
2081         flow_exit();
2082 error:
2083         return err;
2084 }
2085
2086 static void dp_cleanup(void)
2087 {
2088         rcu_barrier();
2089         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2090         unregister_netdevice_notifier(&dp_device_notifier);
2091         vport_exit();
2092         flow_exit();
2093 }
2094
2095 module_init(dp_init);
2096 module_exit(dp_cleanup);
2097
2098 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2099 MODULE_LICENSE("GPL");