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 "vlan.h"
  53 #include "tunnel.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 flow_table *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[OVSP_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(1); /* IFLA_OPERSTATE */
 135 }
 136
 137 /* Caller must hold RTNL lock. */
 138 static int dp_fill_ifinfo(struct sk_buff *skb,
 139                           const struct vport *port,
 140                           int event, unsigned int flags)
 141 {
 142         struct datapath *dp = port->dp;
 143         int ifindex = vport_get_ifindex(port);
 144         struct ifinfomsg *hdr;
 145         struct nlmsghdr *nlh;
 146
 147         if (ifindex < 0)
 148                 return ifindex;
 149
 150         nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
 151         if (nlh == NULL)
 152                 return -EMSGSIZE;
 153
 154         hdr = nlmsg_data(nlh);
 155         hdr->ifi_family = AF_BRIDGE;
 156         hdr->__ifi_pad = 0;
 157         hdr->ifi_type = ARPHRD_ETHER;
 158         hdr->ifi_index = ifindex;
 159         hdr->ifi_flags = vport_get_flags(port);
 160         hdr->ifi_change = 0;
 161
 162         NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
 163         NLA_PUT_U32(skb, IFLA_MASTER,
 164                 vport_get_ifindex(get_vport_protected(dp, OVSP_LOCAL)));
 165         NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
 166 #ifdef IFLA_OPERSTATE
 167         NLA_PUT_U8(skb, IFLA_OPERSTATE,
 168                    vport_is_running(port)
 169                         ? vport_get_operstate(port)
 170                         : IF_OPER_DOWN);
 171 #endif
 172
 173         NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
 174
 175         return nlmsg_end(skb, nlh);
 176
 177 nla_put_failure:
 178         nlmsg_cancel(skb, nlh);
 179         return -EMSGSIZE;
 180 }
 181
 182 /* Caller must hold RTNL lock. */
 183 static void dp_ifinfo_notify(int event, struct vport *port)
 184 {
 185         struct sk_buff *skb;
 186         int err = -ENOBUFS;
 187
 188         skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
 189         if (skb == NULL)
 190                 goto errout;
 191
 192         err = dp_fill_ifinfo(skb, port, event, 0);
 193         if (err < 0) {
 194                 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
 195                 WARN_ON(err == -EMSGSIZE);
 196                 kfree_skb(skb);
 197                 goto errout;
 198         }
 199         rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 200         return;
 201 errout:
 202         if (err < 0)
 203                 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
 204 }
 205
 206 static void release_dp(struct kobject *kobj)
 207 {
 208         struct datapath *dp = container_of(kobj, struct datapath, ifobj);
 209         kfree(dp);
 210 }
 211
 212 static struct kobj_type dp_ktype = {
 213         .release = release_dp
 214 };
 215
 216 static void destroy_dp_rcu(struct rcu_head *rcu)
 217 {
 218         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 219
 220         flow_tbl_destroy(dp->table);
 221         free_percpu(dp->stats_percpu);
 222         kobject_put(&dp->ifobj);
 223 }
 224
 225 /* Called with RTNL lock and genl_lock. */
 226 static struct vport *new_vport(const struct vport_parms *parms)
 227 {
 228         struct vport *vport;
 229
 230         vport = vport_add(parms);
 231         if (!IS_ERR(vport)) {
 232                 struct datapath *dp = parms->dp;
 233
 234                 rcu_assign_pointer(dp->ports[parms->port_no], vport);
 235                 list_add(&vport->node, &dp->port_list);
 236
 237                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 238         }
 239
 240         return vport;
 241 }
 242
 243 /* Called with RTNL lock. */
 244 void dp_detach_port(struct vport *p)
 245 {
 246         ASSERT_RTNL();
 247
 248         if (p->port_no != OVSP_LOCAL)
 249                 dp_sysfs_del_if(p);
 250         dp_ifinfo_notify(RTM_DELLINK, p);
 251
 252         /* First drop references to device. */
 253         list_del(&p->node);
 254         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 255
 256         /* Then destroy it. */
 257         vport_del(p);
 258 }
 259
 260 /* Must be called with rcu_read_lock. */
 261 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 262 {
 263         struct datapath *dp = p->dp;
 264         struct sw_flow *flow;
 265         struct dp_stats_percpu *stats;
 266         int stats_counter_off;
 267         int error;
 268
 269         OVS_CB(skb)->vport = p;
 270
 271         if (!OVS_CB(skb)->flow) {
 272                 struct sw_flow_key key;
 273                 int key_len;
 274                 bool is_frag;
 275
 276                 /* Extract flow from 'skb' into 'key'. */
 277                 error = flow_extract(skb, p->port_no, &key, &key_len, &is_frag);
 278                 if (unlikely(error)) {
 279                         kfree_skb(skb);
 280                         return;
 281                 }
 282
 283                 if (is_frag && dp->drop_frags) {
 284                         consume_skb(skb);
 285                         stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
 286                         goto out;
 287                 }
 288
 289                 /* Look up flow. */
 290                 flow = flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
 291                 if (unlikely(!flow)) {
 292                         struct dp_upcall_info upcall;
 293
 294                         upcall.cmd = OVS_PACKET_CMD_MISS;
 295                         upcall.key = &key;
 296                         upcall.userdata = 0;
 297                         upcall.sample_pool = 0;
 298                         upcall.actions = NULL;
 299                         upcall.actions_len = 0;
 300                         dp_upcall(dp, skb, &upcall);
 301                         stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
 302                         goto out;
 303                 }
 304
 305                 OVS_CB(skb)->flow = flow;
 306         }
 307
 308         stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
 309         flow_used(OVS_CB(skb)->flow, skb);
 310         execute_actions(dp, skb);
 311
 312 out:
 313         /* Update datapath statistics. */
 314         local_bh_disable();
 315         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 316
 317         write_seqcount_begin(&stats->seqlock);
 318         (*(u64 *)((u8 *)stats + stats_counter_off))++;
 319         write_seqcount_end(&stats->seqlock);
 320
 321         local_bh_enable();
 322 }
 323
 324 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
 325 {
 326         u16 csum_start, csum_offset;
 327         __wsum csum;
 328
 329         get_skb_csum_pointers(skb, &csum_start, &csum_offset);
 330         csum_start -= skb_headroom(skb);
 331
 332         skb_copy_bits(skb, 0, to, csum_start);
 333
 334         csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
 335                                       skb->len - csum_start, 0);
 336         *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
 337 }
 338
 339 static struct genl_family dp_packet_genl_family = {
 340         .id = GENL_ID_GENERATE,
 341         .hdrsize = sizeof(struct ovs_header),
 342         .name = OVS_PACKET_FAMILY,
 343         .version = 1,
 344         .maxattr = OVS_PACKET_ATTR_MAX
 345 };
 346
 347 /* Generic Netlink multicast groups for upcalls.
 348  *
 349  * We really want three unique multicast groups per datapath, but we can't even
 350  * get one, because genl_register_mc_group() takes genl_lock, which is also
 351  * held during Generic Netlink message processing, so trying to acquire
 352  * multicast groups during OVS_DP_NEW processing deadlocks.  Instead, we
 353  * preallocate a few groups and use them round-robin for datapaths.  Collision
 354  * isn't fatal--multicast listeners should check that the family is the one
 355  * that they want and discard others--but it wastes time and memory to receive
 356  * unwanted messages.
 357  */
 358 #define PACKET_N_MC_GROUPS 16
 359 static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
 360
 361 static u32 packet_mc_group(struct datapath *dp, u8 cmd)
 362 {
 363         u32 idx;
 364         BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
 365
 366         idx = jhash_2words(dp->dp_ifindex, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
 367         return packet_mc_groups[idx].id;
 368 }
 369
 370 static int packet_register_mc_groups(void)
 371 {
 372         int i;
 373
 374         for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
 375                 struct genl_multicast_group *group = &packet_mc_groups[i];
 376                 int error;
 377
 378                 sprintf(group->name, "packet%d", i);
 379                 error = genl_register_mc_group(&dp_packet_genl_family, group);
 380                 if (error)
 381                         return error;
 382         }
 383         return 0;
 384 }
 385
 386 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
 387 {
 388         struct dp_stats_percpu *stats;
 389         int err;
 390
 391         WARN_ON_ONCE(skb_shared(skb));
 392
 393         forward_ip_summed(skb, true);
 394
 395         /* Break apart GSO packets into their component pieces.  Otherwise
 396          * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
 397         if (skb_is_gso(skb)) {
 398                 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 399
 400                 if (IS_ERR(nskb)) {
 401                         kfree_skb(skb);
 402                         err = PTR_ERR(nskb);
 403                         goto err;
 404                 }
 405                 consume_skb(skb);
 406                 skb = nskb;
 407         }
 408
 409         err = queue_userspace_packets(dp, skb, upcall_info);
 410         if (err)
 411                 goto err;
 412
 413         return 0;
 414
 415 err:
 416         local_bh_disable();
 417         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 418
 419         write_seqcount_begin(&stats->seqlock);
 420         stats->n_lost++;
 421         write_seqcount_end(&stats->seqlock);
 422
 423         local_bh_enable();
 424
 425         return err;
 426 }
 427
 428 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
 429  * 'upcall_info'.  There will be only one packet unless we broke up a GSO
 430  * packet.
 431  */
 432 static int queue_userspace_packets(struct datapath *dp, struct sk_buff *skb,
 433                                  const struct dp_upcall_info *upcall_info)
 434 {
 435         u32 group = packet_mc_group(dp, upcall_info->cmd);
 436         struct sk_buff *nskb;
 437         int err;
 438
 439         do {
 440                 struct ovs_header *upcall;
 441                 struct sk_buff *user_skb; /* to be queued to userspace */
 442                 struct nlattr *nla;
 443                 unsigned int len;
 444
 445                 nskb = skb->next;
 446                 skb->next = NULL;
 447
 448                 err = vlan_deaccel_tag(skb);
 449                 if (unlikely(err))
 450                         goto err_kfree_skbs;
 451
 452                 if (nla_attr_size(skb->len) > USHRT_MAX)
 453                         goto err_kfree_skbs;
 454
 455                 len = sizeof(struct ovs_header);
 456                 len += nla_total_size(skb->len);
 457                 len += nla_total_size(FLOW_BUFSIZE);
 458                 if (upcall_info->userdata)
 459                         len += nla_total_size(8);
 460                 if (upcall_info->sample_pool)
 461                         len += nla_total_size(4);
 462                 if (upcall_info->actions_len)
 463                         len += nla_total_size(upcall_info->actions_len);
 464
 465                 user_skb = genlmsg_new(len, GFP_ATOMIC);
 466                 if (!user_skb) {
 467                         netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS);
 468                         goto err_kfree_skbs;
 469                 }
 470
 471                 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
 472                 upcall->dp_ifindex = dp->dp_ifindex;
 473
 474                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 475                 flow_to_nlattrs(upcall_info->key, user_skb);
 476                 nla_nest_end(user_skb, nla);
 477
 478                 if (upcall_info->userdata)
 479                         nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA, upcall_info->userdata);
 480                 if (upcall_info->sample_pool)
 481                         nla_put_u32(user_skb, OVS_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
 482                 if (upcall_info->actions_len) {
 483                         const struct nlattr *actions = upcall_info->actions;
 484                         u32 actions_len = upcall_info->actions_len;
 485
 486                         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 487                         memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
 488                         nla_nest_end(user_skb, nla);
 489                 }
 490
 491                 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 492                 if (skb->ip_summed == CHECKSUM_PARTIAL)
 493                         copy_and_csum_skb(skb, nla_data(nla));
 494                 else
 495                         skb_copy_bits(skb, 0, nla_data(nla), skb->len);
 496
 497                 err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC);
 498                 if (err)
 499                         goto err_kfree_skbs;
 500
 501                 consume_skb(skb);
 502                 skb = nskb;
 503         } while (skb);
 504         return 0;
 505
 506 err_kfree_skbs:
 507         kfree_skb(skb);
 508         while ((skb = nskb) != NULL) {
 509                 nskb = skb->next;
 510                 kfree_skb(skb);
 511         }
 512         return err;
 513 }
 514
 515 /* Called with genl_mutex. */
 516 static int flush_flows(int dp_ifindex)
 517 {
 518         struct flow_table *old_table;
 519         struct flow_table *new_table;
 520         struct datapath *dp;
 521
 522         dp = get_dp(dp_ifindex);
 523         if (!dp)
 524                 return -ENODEV;
 525
 526         old_table = get_table_protected(dp);
 527         new_table = flow_tbl_alloc(TBL_MIN_BUCKETS);
 528         if (!new_table)
 529                 return -ENOMEM;
 530
 531         rcu_assign_pointer(dp->table, new_table);
 532
 533         flow_tbl_deferred_destroy(old_table);
 534         return 0;
 535 }
 536
 537 static int validate_actions(const struct nlattr *attr)
 538 {
 539         const struct nlattr *a;
 540         int rem;
 541
 542         nla_for_each_nested(a, attr, rem) {
 543                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 544                         [OVS_ACTION_ATTR_OUTPUT] = 4,
 545                         [OVS_ACTION_ATTR_USERSPACE] = 8,
 546                         [OVS_ACTION_ATTR_PUSH_VLAN] = 2,
 547                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 548                         [OVS_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
 549                         [OVS_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
 550                         [OVS_ACTION_ATTR_SET_NW_SRC] = 4,
 551                         [OVS_ACTION_ATTR_SET_NW_DST] = 4,
 552                         [OVS_ACTION_ATTR_SET_NW_TOS] = 1,
 553                         [OVS_ACTION_ATTR_SET_TP_SRC] = 2,
 554                         [OVS_ACTION_ATTR_SET_TP_DST] = 2,
 555                         [OVS_ACTION_ATTR_SET_TUNNEL] = 8,
 556                         [OVS_ACTION_ATTR_SET_PRIORITY] = 4,
 557                         [OVS_ACTION_ATTR_POP_PRIORITY] = 0,
 558                 };
 559                 int type = nla_type(a);
 560
 561                 if (type > OVS_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
 562                         return -EINVAL;
 563
 564                 switch (type) {
 565                 case OVS_ACTION_ATTR_UNSPEC:
 566                         return -EINVAL;
 567
 568                 case OVS_ACTION_ATTR_USERSPACE:
 569                 case OVS_ACTION_ATTR_POP_VLAN:
 570                 case OVS_ACTION_ATTR_SET_DL_SRC:
 571                 case OVS_ACTION_ATTR_SET_DL_DST:
 572                 case OVS_ACTION_ATTR_SET_NW_SRC:
 573                 case OVS_ACTION_ATTR_SET_NW_DST:
 574                 case OVS_ACTION_ATTR_SET_TP_SRC:
 575                 case OVS_ACTION_ATTR_SET_TP_DST:
 576                 case OVS_ACTION_ATTR_SET_TUNNEL:
 577                 case OVS_ACTION_ATTR_SET_PRIORITY:
 578                 case OVS_ACTION_ATTR_POP_PRIORITY:
 579                         /* No validation needed. */
 580                         break;
 581
 582                 case OVS_ACTION_ATTR_OUTPUT:
 583                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 584                                 return -EINVAL;
 585                         break;
 586
 587                 case OVS_ACTION_ATTR_PUSH_VLAN:
 588                         if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
 589                                 return -EINVAL;
 590                         break;
 591
 592                 case OVS_ACTION_ATTR_SET_NW_TOS:
 593                         if (nla_get_u8(a) & INET_ECN_MASK)
 594                                 return -EINVAL;
 595                         break;
 596
 597                 default:
 598                         return -EOPNOTSUPP;
 599                 }
 600         }
 601
 602         if (rem > 0)
 603                 return -EINVAL;
 604
 605         return 0;
 606 }
 607 static void clear_stats(struct sw_flow *flow)
 608 {
 609         flow->used = 0;
 610         flow->tcp_flags = 0;
 611         flow->packet_count = 0;
 612         flow->byte_count = 0;
 613 }
 614
 615 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 616 {
 617         struct ovs_header *ovs_header = info->userhdr;
 618         struct nlattr **a = info->attrs;
 619         struct sw_flow_actions *acts;
 620         struct sk_buff *packet;
 621         struct sw_flow *flow;
 622         struct datapath *dp;
 623         struct ethhdr *eth;
 624         bool is_frag;
 625         int len;
 626         int err;
 627         int key_len;
 628
 629         err = -EINVAL;
 630         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 631             !a[OVS_PACKET_ATTR_ACTIONS] ||
 632             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 633                 goto err;
 634
 635         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS]);
 636         if (err)
 637                 goto err;
 638
 639         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 640         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 641         err = -ENOMEM;
 642         if (!packet)
 643                 goto err;
 644         skb_reserve(packet, NET_IP_ALIGN);
 645
 646         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 647
 648         skb_reset_mac_header(packet);
 649         eth = eth_hdr(packet);
 650
 651         /* Normally, setting the skb 'protocol' field would be handled by a
 652          * call to eth_type_trans(), but it assumes there's a sending
 653          * device, which we may not have. */
 654         if (ntohs(eth->h_proto) >= 1536)
 655                 packet->protocol = eth->h_proto;
 656         else
 657                 packet->protocol = htons(ETH_P_802_2);
 658
 659         /* Build an sw_flow for sending this packet. */
 660         flow = flow_alloc();
 661         err = PTR_ERR(flow);
 662         if (IS_ERR(flow))
 663                 goto err_kfree_skb;
 664
 665         err = flow_extract(packet, -1, &flow->key, &key_len, &is_frag);
 666         if (err)
 667                 goto err_flow_put;
 668
 669         err = flow_metadata_from_nlattrs(&flow->key.eth.in_port,
 670                                          &flow->key.eth.tun_id,
 671                                          a[OVS_PACKET_ATTR_KEY]);
 672         if (err)
 673                 goto err_flow_put;
 674
 675         flow->hash = flow_hash(&flow->key, key_len);
 676
 677         acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 678         err = PTR_ERR(acts);
 679         if (IS_ERR(acts))
 680                 goto err_flow_put;
 681         rcu_assign_pointer(flow->sf_acts, acts);
 682
 683         OVS_CB(packet)->flow = flow;
 684
 685         rcu_read_lock();
 686         dp = get_dp(ovs_header->dp_ifindex);
 687         err = -ENODEV;
 688         if (!dp)
 689                 goto err_unlock;
 690         err = execute_actions(dp, packet);
 691         rcu_read_unlock();
 692
 693         flow_put(flow);
 694         return err;
 695
 696 err_unlock:
 697         rcu_read_unlock();
 698 err_flow_put:
 699         flow_put(flow);
 700 err_kfree_skb:
 701         kfree_skb(packet);
 702 err:
 703         return err;
 704 }
 705
 706 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 707         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 708         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 709         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 710 };
 711
 712 static struct genl_ops dp_packet_genl_ops[] = {
 713         { .cmd = OVS_PACKET_CMD_EXECUTE,
 714           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 715           .policy = packet_policy,
 716           .doit = ovs_packet_cmd_execute
 717         }
 718 };
 719
 720 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 721 {
 722         int i;
 723         struct flow_table *table = get_table_protected(dp);
 724
 725         stats->n_flows = flow_tbl_count(table);
 726
 727         stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
 728         for_each_possible_cpu(i) {
 729                 const struct dp_stats_percpu *percpu_stats;
 730                 struct dp_stats_percpu local_stats;
 731                 unsigned seqcount;
 732
 733                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 734
 735                 do {
 736                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 737                         local_stats = *percpu_stats;
 738                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 739
 740                 stats->n_frags += local_stats.n_frags;
 741                 stats->n_hit += local_stats.n_hit;
 742                 stats->n_missed += local_stats.n_missed;
 743                 stats->n_lost += local_stats.n_lost;
 744         }
 745 }
 746
 747 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
 748  * Called with RTNL lock.
 749  */
 750 int dp_min_mtu(const struct datapath *dp)
 751 {
 752         struct vport *p;
 753         int mtu = 0;
 754
 755         ASSERT_RTNL();
 756
 757         list_for_each_entry (p, &dp->port_list, node) {
 758                 int dev_mtu;
 759
 760                 /* Skip any internal ports, since that's what we're trying to
 761                  * set. */
 762                 if (is_internal_vport(p))
 763                         continue;
 764
 765                 dev_mtu = vport_get_mtu(p);
 766                 if (!dev_mtu)
 767                         continue;
 768                 if (!mtu || dev_mtu < mtu)
 769                         mtu = dev_mtu;
 770         }
 771
 772         return mtu ? mtu : ETH_DATA_LEN;
 773 }
 774
 775 /* Sets the MTU of all datapath devices to the minimum of the ports
 776  * Called with RTNL lock.
 777  */
 778 void set_internal_devs_mtu(const struct datapath *dp)
 779 {
 780         struct vport *p;
 781         int mtu;
 782
 783         ASSERT_RTNL();
 784
 785         mtu = dp_min_mtu(dp);
 786
 787         list_for_each_entry (p, &dp->port_list, node) {
 788                 if (is_internal_vport(p))
 789                         vport_set_mtu(p, mtu);
 790         }
 791 }
 792
 793 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 794         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 795         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 796         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 797 };
 798
 799 static struct genl_family dp_flow_genl_family = {
 800         .id = GENL_ID_GENERATE,
 801         .hdrsize = sizeof(struct ovs_header),
 802         .name = OVS_FLOW_FAMILY,
 803         .version = 1,
 804         .maxattr = OVS_FLOW_ATTR_MAX
 805 };
 806
 807 static struct genl_multicast_group dp_flow_multicast_group = {
 808         .name = OVS_FLOW_MCGROUP
 809 };
 810
 811 /* Called with genl_lock. */
 812 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 813                                   struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
 814 {
 815         const int skb_orig_len = skb->len;
 816         const struct sw_flow_actions *sf_acts;
 817         struct ovs_flow_stats stats;
 818         struct ovs_header *ovs_header;
 819         struct nlattr *nla;
 820         unsigned long used;
 821         u8 tcp_flags;
 822         int err;
 823
 824         sf_acts = rcu_dereference_protected(flow->sf_acts,
 825                                             lockdep_genl_is_held());
 826
 827         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 828         if (!ovs_header)
 829                 return -EMSGSIZE;
 830
 831         ovs_header->dp_ifindex = dp->dp_ifindex;
 832
 833         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 834         if (!nla)
 835                 goto nla_put_failure;
 836         err = flow_to_nlattrs(&flow->key, skb);
 837         if (err)
 838                 goto error;
 839         nla_nest_end(skb, nla);
 840
 841         spin_lock_bh(&flow->lock);
 842         used = flow->used;
 843         stats.n_packets = flow->packet_count;
 844         stats.n_bytes = flow->byte_count;
 845         tcp_flags = flow->tcp_flags;
 846         spin_unlock_bh(&flow->lock);
 847
 848         if (used)
 849                 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
 850
 851         if (stats.n_packets)
 852                 NLA_PUT(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats);
 853
 854         if (tcp_flags)
 855                 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 856
 857         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 858          * this is the first flow to be dumped into 'skb'.  This is unusual for
 859          * Netlink but individual action lists can be longer than
 860          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 861          * The userspace caller can always fetch the actions separately if it
 862          * really wants them.  (Most userspace callers in fact don't care.)
 863          *
 864          * This can only fail for dump operations because the skb is always
 865          * properly sized for single flows.
 866          */
 867         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 868                       sf_acts->actions);
 869         if (err < 0 && skb_orig_len)
 870                 goto error;
 871
 872         return genlmsg_end(skb, ovs_header);
 873
 874 nla_put_failure:
 875         err = -EMSGSIZE;
 876 error:
 877         genlmsg_cancel(skb, ovs_header);
 878         return err;
 879 }
 880
 881 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 882 {
 883         const struct sw_flow_actions *sf_acts;
 884         int len;
 885
 886         sf_acts = rcu_dereference_protected(flow->sf_acts,
 887                                             lockdep_genl_is_held());
 888
 889         len = nla_total_size(FLOW_BUFSIZE); /* OVS_FLOW_ATTR_KEY */
 890         len += nla_total_size(sf_acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 891         len += nla_total_size(sizeof(struct ovs_flow_stats)); /* OVS_FLOW_ATTR_STATS */
 892         len += nla_total_size(1); /* OVS_FLOW_ATTR_TCP_FLAGS */
 893         len += nla_total_size(8); /* OVS_FLOW_ATTR_USED */
 894         return genlmsg_new(NLMSG_ALIGN(sizeof(struct ovs_header)) + len, GFP_KERNEL);
 895 }
 896
 897 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
 898                                                u32 pid, u32 seq, u8 cmd)
 899 {
 900         struct sk_buff *skb;
 901         int retval;
 902
 903         skb = ovs_flow_cmd_alloc_info(flow);
 904         if (!skb)
 905                 return ERR_PTR(-ENOMEM);
 906
 907         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 908         BUG_ON(retval < 0);
 909         return skb;
 910 }
 911
 912 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 913 {
 914         struct nlattr **a = info->attrs;
 915         struct ovs_header *ovs_header = info->userhdr;
 916         struct sw_flow_key key;
 917         struct sw_flow *flow;
 918         struct sk_buff *reply;
 919         struct datapath *dp;
 920         struct flow_table *table;
 921         int error;
 922         int key_len;
 923
 924         /* Extract key. */
 925         error = -EINVAL;
 926         if (!a[OVS_FLOW_ATTR_KEY])
 927                 goto error;
 928         error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
 929         if (error)
 930                 goto error;
 931
 932         /* Validate actions. */
 933         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 934                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS]);
 935                 if (error)
 936                         goto error;
 937         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 938                 error = -EINVAL;
 939                 goto error;
 940         }
 941
 942         dp = get_dp(ovs_header->dp_ifindex);
 943         error = -ENODEV;
 944         if (!dp)
 945                 goto error;
 946
 947         table = get_table_protected(dp);
 948         flow = flow_tbl_lookup(table, &key, key_len);
 949         if (!flow) {
 950                 struct sw_flow_actions *acts;
 951
 952                 /* Bail out if we're not allowed to create a new flow. */
 953                 error = -ENOENT;
 954                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
 955                         goto error;
 956
 957                 /* Expand table, if necessary, to make room. */
 958                 if (flow_tbl_need_to_expand(table)) {
 959                         struct flow_table *new_table;
 960
 961                         new_table = flow_tbl_expand(table);
 962                         if (!IS_ERR(new_table)) {
 963                                 rcu_assign_pointer(dp->table, new_table);
 964                                 flow_tbl_deferred_destroy(table);
 965                                 table = get_table_protected(dp);
 966                         }
 967                 }
 968
 969                 /* Allocate flow. */
 970                 flow = flow_alloc();
 971                 if (IS_ERR(flow)) {
 972                         error = PTR_ERR(flow);
 973                         goto error;
 974                 }
 975                 flow->key = key;
 976                 clear_stats(flow);
 977
 978                 /* Obtain actions. */
 979                 acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
 980                 error = PTR_ERR(acts);
 981                 if (IS_ERR(acts))
 982                         goto error_free_flow;
 983                 rcu_assign_pointer(flow->sf_acts, acts);
 984
 985                 /* Put flow in bucket. */
 986                 flow->hash = flow_hash(&key, key_len);
 987                 flow_tbl_insert(table, flow);
 988
 989                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
 990                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
 991         } else {
 992                 /* We found a matching flow. */
 993                 struct sw_flow_actions *old_acts;
 994
 995                 /* Bail out if we're not allowed to modify an existing flow.
 996                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 997                  * because Generic Netlink treats the latter as a dump
 998                  * request.  We also accept NLM_F_EXCL in case that bug ever
 999                  * gets fixed.
1000                  */
1001                 error = -EEXIST;
1002                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1003                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1004                         goto error;
1005
1006                 /* Update actions. */
1007                 old_acts = rcu_dereference_protected(flow->sf_acts,
1008                                                      lockdep_genl_is_held());
1009                 if (a[OVS_FLOW_ATTR_ACTIONS] &&
1010                     (old_acts->actions_len != nla_len(a[OVS_FLOW_ATTR_ACTIONS]) ||
1011                      memcmp(old_acts->actions, nla_data(a[OVS_FLOW_ATTR_ACTIONS]),
1012                             old_acts->actions_len))) {
1013                         struct sw_flow_actions *new_acts;
1014
1015                         new_acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1016                         error = PTR_ERR(new_acts);
1017                         if (IS_ERR(new_acts))
1018                                 goto error;
1019
1020                         rcu_assign_pointer(flow->sf_acts, new_acts);
1021                         flow_deferred_free_acts(old_acts);
1022                 }
1023
1024                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1025                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
1026
1027                 /* Clear stats. */
1028                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1029                         spin_lock_bh(&flow->lock);
1030                         clear_stats(flow);
1031                         spin_unlock_bh(&flow->lock);
1032                 }
1033         }
1034
1035         if (!IS_ERR(reply))
1036                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1037                             dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1038         else
1039                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1040                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1041         return 0;
1042
1043 error_free_flow:
1044         flow_put(flow);
1045 error:
1046         return error;
1047 }
1048
1049 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1050 {
1051         struct nlattr **a = info->attrs;
1052         struct ovs_header *ovs_header = info->userhdr;
1053         struct sw_flow_key key;
1054         struct sk_buff *reply;
1055         struct sw_flow *flow;
1056         struct datapath *dp;
1057         struct flow_table *table;
1058         int err;
1059         int key_len;
1060
1061         if (!a[OVS_FLOW_ATTR_KEY])
1062                 return -EINVAL;
1063         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1064         if (err)
1065                 return err;
1066
1067         dp = get_dp(ovs_header->dp_ifindex);
1068         if (!dp)
1069                 return -ENODEV;
1070
1071         table = get_table_protected(dp);
1072         flow = flow_tbl_lookup(table, &key, key_len);
1073         if (!flow)
1074                 return -ENOENT;
1075
1076         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, OVS_FLOW_CMD_NEW);
1077         if (IS_ERR(reply))
1078                 return PTR_ERR(reply);
1079
1080         return genlmsg_reply(reply, info);
1081 }
1082
1083 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1084 {
1085         struct nlattr **a = info->attrs;
1086         struct ovs_header *ovs_header = info->userhdr;
1087         struct sw_flow_key key;
1088         struct sk_buff *reply;
1089         struct sw_flow *flow;
1090         struct datapath *dp;
1091         struct flow_table *table;
1092         int err;
1093         int key_len;
1094
1095         if (!a[OVS_FLOW_ATTR_KEY])
1096                 return flush_flows(ovs_header->dp_ifindex);
1097         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1098         if (err)
1099                 return err;
1100
1101         dp = get_dp(ovs_header->dp_ifindex);
1102         if (!dp)
1103                 return -ENODEV;
1104
1105         table = get_table_protected(dp);
1106         flow = flow_tbl_lookup(table, &key, key_len);
1107         if (!flow)
1108                 return -ENOENT;
1109
1110         reply = ovs_flow_cmd_alloc_info(flow);
1111         if (!reply)
1112                 return -ENOMEM;
1113
1114         flow_tbl_remove(table, flow);
1115
1116         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1117                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1118         BUG_ON(err < 0);
1119
1120         flow_deferred_free(flow);
1121
1122         genl_notify(reply, genl_info_net(info), info->snd_pid,
1123                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1124         return 0;
1125 }
1126
1127 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1128 {
1129         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1130         struct datapath *dp;
1131
1132         dp = get_dp(ovs_header->dp_ifindex);
1133         if (!dp)
1134                 return -ENODEV;
1135
1136         for (;;) {
1137                 struct sw_flow *flow;
1138                 u32 bucket, obj;
1139
1140                 bucket = cb->args[0];
1141                 obj = cb->args[1];
1142                 flow = flow_tbl_next(get_table_protected(dp), &bucket, &obj);
1143                 if (!flow)
1144                         break;
1145
1146                 if (ovs_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1147                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1148                                            OVS_FLOW_CMD_NEW) < 0)
1149                         break;
1150
1151                 cb->args[0] = bucket;
1152                 cb->args[1] = obj;
1153         }
1154         return skb->len;
1155 }
1156
1157 static struct genl_ops dp_flow_genl_ops[] = {
1158         { .cmd = OVS_FLOW_CMD_NEW,
1159           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1160           .policy = flow_policy,
1161           .doit = ovs_flow_cmd_new_or_set
1162         },
1163         { .cmd = OVS_FLOW_CMD_DEL,
1164           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1165           .policy = flow_policy,
1166           .doit = ovs_flow_cmd_del
1167         },
1168         { .cmd = OVS_FLOW_CMD_GET,
1169           .flags = 0,               /* OK for unprivileged users. */
1170           .policy = flow_policy,
1171           .doit = ovs_flow_cmd_get,
1172           .dumpit = ovs_flow_cmd_dump
1173         },
1174         { .cmd = OVS_FLOW_CMD_SET,
1175           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1176           .policy = flow_policy,
1177           .doit = ovs_flow_cmd_new_or_set,
1178         },
1179 };
1180
1181 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1182 #ifdef HAVE_NLA_NUL_STRING
1183         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1184 #endif
1185         [OVS_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1186         [OVS_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1187 };
1188
1189 static struct genl_family dp_datapath_genl_family = {
1190         .id = GENL_ID_GENERATE,
1191         .hdrsize = sizeof(struct ovs_header),
1192         .name = OVS_DATAPATH_FAMILY,
1193         .version = 1,
1194         .maxattr = OVS_DP_ATTR_MAX
1195 };
1196
1197 static struct genl_multicast_group dp_datapath_multicast_group = {
1198         .name = OVS_DATAPATH_MCGROUP
1199 };
1200
1201 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1202                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1203 {
1204         struct ovs_header *ovs_header;
1205         struct nlattr *nla;
1206         int err;
1207
1208         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1209                                    flags, cmd);
1210         if (!ovs_header)
1211                 goto error;
1212
1213         ovs_header->dp_ifindex = dp->dp_ifindex;
1214
1215         rcu_read_lock();
1216         err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
1217         rcu_read_unlock();
1218         if (err)
1219                 goto nla_put_failure;
1220
1221         nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats));
1222         if (!nla)
1223                 goto nla_put_failure;
1224         get_dp_stats(dp, nla_data(nla));
1225
1226         NLA_PUT_U32(skb, OVS_DP_ATTR_IPV4_FRAGS,
1227                     dp->drop_frags ? OVS_DP_FRAG_DROP : OVS_DP_FRAG_ZERO);
1228
1229         if (dp->sflow_probability)
1230                 NLA_PUT_U32(skb, OVS_DP_ATTR_SAMPLING, dp->sflow_probability);
1231
1232         nla = nla_nest_start(skb, OVS_DP_ATTR_MCGROUPS);
1233         if (!nla)
1234                 goto nla_put_failure;
1235         NLA_PUT_U32(skb, OVS_PACKET_CMD_MISS, packet_mc_group(dp, OVS_PACKET_CMD_MISS));
1236         NLA_PUT_U32(skb, OVS_PACKET_CMD_ACTION, packet_mc_group(dp, OVS_PACKET_CMD_ACTION));
1237         NLA_PUT_U32(skb, OVS_PACKET_CMD_SAMPLE, packet_mc_group(dp, OVS_PACKET_CMD_SAMPLE));
1238         nla_nest_end(skb, nla);
1239
1240         return genlmsg_end(skb, ovs_header);
1241
1242 nla_put_failure:
1243         genlmsg_cancel(skb, ovs_header);
1244 error:
1245         return -EMSGSIZE;
1246 }
1247
1248 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1249                                              u32 seq, u8 cmd)
1250 {
1251         struct sk_buff *skb;
1252         int retval;
1253
1254         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1255         if (!skb)
1256                 return ERR_PTR(-ENOMEM);
1257
1258         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1259         if (retval < 0) {
1260                 kfree_skb(skb);
1261                 return ERR_PTR(retval);
1262         }
1263         return skb;
1264 }
1265
1266 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1267 {
1268         if (a[OVS_DP_ATTR_IPV4_FRAGS]) {
1269                 u32 frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]);
1270
1271                 if (frags != OVS_DP_FRAG_ZERO && frags != OVS_DP_FRAG_DROP)
1272                         return -EINVAL;
1273         }
1274
1275         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1276 }
1277
1278 /* Called with genl_mutex and optionally with RTNL lock also. */
1279 static struct datapath *lookup_datapath(struct ovs_header *ovs_header, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1280 {
1281         struct datapath *dp;
1282
1283         if (!a[OVS_DP_ATTR_NAME])
1284                 dp = get_dp(ovs_header->dp_ifindex);
1285         else {
1286                 struct vport *vport;
1287
1288                 rcu_read_lock();
1289                 vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1290                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1291                 rcu_read_unlock();
1292         }
1293         return dp ? dp : ERR_PTR(-ENODEV);
1294 }
1295
1296 /* Called with genl_mutex. */
1297 static void change_datapath(struct datapath *dp, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1298 {
1299         if (a[OVS_DP_ATTR_IPV4_FRAGS])
1300                 dp->drop_frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]) == OVS_DP_FRAG_DROP;
1301         if (a[OVS_DP_ATTR_SAMPLING])
1302                 dp->sflow_probability = nla_get_u32(a[OVS_DP_ATTR_SAMPLING]);
1303 }
1304
1305 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1306 {
1307         struct nlattr **a = info->attrs;
1308         struct vport_parms parms;
1309         struct sk_buff *reply;
1310         struct datapath *dp;
1311         struct vport *vport;
1312         int err;
1313
1314         err = -EINVAL;
1315         if (!a[OVS_DP_ATTR_NAME])
1316                 goto err;
1317
1318         err = ovs_dp_cmd_validate(a);
1319         if (err)
1320                 goto err;
1321
1322         rtnl_lock();
1323         err = -ENODEV;
1324         if (!try_module_get(THIS_MODULE))
1325                 goto err_unlock_rtnl;
1326
1327         err = -ENOMEM;
1328         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1329         if (dp == NULL)
1330                 goto err_put_module;
1331         INIT_LIST_HEAD(&dp->port_list);
1332
1333         /* Initialize kobject for bridge.  This will be added as
1334          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1335         dp->ifobj.kset = NULL;
1336         kobject_init(&dp->ifobj, &dp_ktype);
1337
1338         /* Allocate table. */
1339         err = -ENOMEM;
1340         rcu_assign_pointer(dp->table, flow_tbl_alloc(TBL_MIN_BUCKETS));
1341         if (!dp->table)
1342                 goto err_free_dp;
1343
1344         /* Set up our datapath device. */
1345         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1346         parms.type = OVS_VPORT_TYPE_INTERNAL;
1347         parms.options = NULL;
1348         parms.dp = dp;
1349         parms.port_no = OVSP_LOCAL;
1350         vport = new_vport(&parms);
1351         if (IS_ERR(vport)) {
1352                 err = PTR_ERR(vport);
1353                 if (err == -EBUSY)
1354                         err = -EEXIST;
1355
1356                 goto err_destroy_table;
1357         }
1358         dp->dp_ifindex = vport_get_ifindex(vport);
1359
1360         dp->drop_frags = 0;
1361         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1362         if (!dp->stats_percpu) {
1363                 err = -ENOMEM;
1364                 goto err_destroy_local_port;
1365         }
1366
1367         change_datapath(dp, a);
1368
1369         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1370         err = PTR_ERR(reply);
1371         if (IS_ERR(reply))
1372                 goto err_destroy_local_port;
1373
1374         list_add_tail(&dp->list_node, &dps);
1375         dp_sysfs_add_dp(dp);
1376
1377         rtnl_unlock();
1378
1379         genl_notify(reply, genl_info_net(info), info->snd_pid,
1380                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1381         return 0;
1382
1383 err_destroy_local_port:
1384         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1385 err_destroy_table:
1386         flow_tbl_destroy(get_table_protected(dp));
1387 err_free_dp:
1388         kfree(dp);
1389 err_put_module:
1390         module_put(THIS_MODULE);
1391 err_unlock_rtnl:
1392         rtnl_unlock();
1393 err:
1394         return err;
1395 }
1396
1397 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1398 {
1399         struct vport *vport, *next_vport;
1400         struct sk_buff *reply;
1401         struct datapath *dp;
1402         int err;
1403
1404         err = ovs_dp_cmd_validate(info->attrs);
1405         if (err)
1406                 goto exit;
1407
1408         rtnl_lock();
1409         dp = lookup_datapath(info->userhdr, info->attrs);
1410         err = PTR_ERR(dp);
1411         if (IS_ERR(dp))
1412                 goto exit_unlock;
1413
1414         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_DEL);
1415         err = PTR_ERR(reply);
1416         if (IS_ERR(reply))
1417                 goto exit_unlock;
1418
1419         list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1420                 if (vport->port_no != OVSP_LOCAL)
1421                         dp_detach_port(vport);
1422
1423         dp_sysfs_del_dp(dp);
1424         list_del(&dp->list_node);
1425         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1426
1427         /* rtnl_unlock() will wait until all the references to devices that
1428          * are pending unregistration have been dropped.  We do it here to
1429          * ensure that any internal devices (which contain DP pointers) are
1430          * fully destroyed before freeing the datapath.
1431          */
1432         rtnl_unlock();
1433
1434         call_rcu(&dp->rcu, destroy_dp_rcu);
1435         module_put(THIS_MODULE);
1436
1437         genl_notify(reply, genl_info_net(info), info->snd_pid,
1438                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1439
1440         return 0;
1441
1442 exit_unlock:
1443         rtnl_unlock();
1444 exit:
1445         return err;
1446 }
1447
1448 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1449 {
1450         struct sk_buff *reply;
1451         struct datapath *dp;
1452         int err;
1453
1454         err = ovs_dp_cmd_validate(info->attrs);
1455         if (err)
1456                 return err;
1457
1458         dp = lookup_datapath(info->userhdr, info->attrs);
1459         if (IS_ERR(dp))
1460                 return PTR_ERR(dp);
1461
1462         change_datapath(dp, info->attrs);
1463
1464         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1465         if (IS_ERR(reply)) {
1466                 err = PTR_ERR(reply);
1467                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1468                                 dp_datapath_multicast_group.id, err);
1469                 return 0;
1470         }
1471
1472         genl_notify(reply, genl_info_net(info), info->snd_pid,
1473                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1474         return 0;
1475 }
1476
1477 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1478 {
1479         struct sk_buff *reply;
1480         struct datapath *dp;
1481         int err;
1482
1483         err = ovs_dp_cmd_validate(info->attrs);
1484         if (err)
1485                 return err;
1486
1487         dp = lookup_datapath(info->userhdr, info->attrs);
1488         if (IS_ERR(dp))
1489                 return PTR_ERR(dp);
1490
1491         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1492         if (IS_ERR(reply))
1493                 return PTR_ERR(reply);
1494
1495         return genlmsg_reply(reply, info);
1496 }
1497
1498 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1499 {
1500         struct datapath *dp;
1501         int skip = cb->args[0];
1502         int i = 0;
1503
1504         list_for_each_entry (dp, &dps, list_node) {
1505                 if (i < skip)
1506                         continue;
1507                 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1508                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1509                                          OVS_DP_CMD_NEW) < 0)
1510                         break;
1511                 i++;
1512         }
1513
1514         cb->args[0] = i;
1515
1516         return skb->len;
1517 }
1518
1519 static struct genl_ops dp_datapath_genl_ops[] = {
1520         { .cmd = OVS_DP_CMD_NEW,
1521           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1522           .policy = datapath_policy,
1523           .doit = ovs_dp_cmd_new
1524         },
1525         { .cmd = OVS_DP_CMD_DEL,
1526           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1527           .policy = datapath_policy,
1528           .doit = ovs_dp_cmd_del
1529         },
1530         { .cmd = OVS_DP_CMD_GET,
1531           .flags = 0,               /* OK for unprivileged users. */
1532           .policy = datapath_policy,
1533           .doit = ovs_dp_cmd_get,
1534           .dumpit = ovs_dp_cmd_dump
1535         },
1536         { .cmd = OVS_DP_CMD_SET,
1537           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1538           .policy = datapath_policy,
1539           .doit = ovs_dp_cmd_set,
1540         },
1541 };
1542
1543 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1544 #ifdef HAVE_NLA_NUL_STRING
1545         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1546         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1547         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1548         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1549         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1550 #else
1551         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1552         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1553 #endif
1554         [OVS_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1555         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1556 };
1557
1558 static struct genl_family dp_vport_genl_family = {
1559         .id = GENL_ID_GENERATE,
1560         .hdrsize = sizeof(struct ovs_header),
1561         .name = OVS_VPORT_FAMILY,
1562         .version = 1,
1563         .maxattr = OVS_VPORT_ATTR_MAX
1564 };
1565
1566 struct genl_multicast_group dp_vport_multicast_group = {
1567         .name = OVS_VPORT_MCGROUP
1568 };
1569
1570 /* Called with RTNL lock or RCU read lock. */
1571 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1572                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1573 {
1574         struct ovs_header *ovs_header;
1575         struct nlattr *nla;
1576         int ifindex;
1577         int mtu;
1578         int err;
1579
1580         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1581                                  flags, cmd);
1582         if (!ovs_header)
1583                 return -EMSGSIZE;
1584
1585         ovs_header->dp_ifindex = vport->dp->dp_ifindex;
1586
1587         NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1588         NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport_get_type(vport));
1589         NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport_get_name(vport));
1590
1591         nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1592         if (!nla)
1593                 goto nla_put_failure;
1594         if (vport_get_stats(vport, nla_data(nla)))
1595                 __skb_trim(skb, skb->len - nla->nla_len);
1596
1597         NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1598
1599         mtu = vport_get_mtu(vport);
1600         if (mtu)
1601                 NLA_PUT_U32(skb, OVS_VPORT_ATTR_MTU, mtu);
1602
1603         err = vport_get_options(vport, skb);
1604         if (err == -EMSGSIZE)
1605                 goto error;
1606
1607         ifindex = vport_get_ifindex(vport);
1608         if (ifindex > 0)
1609                 NLA_PUT_U32(skb, OVS_VPORT_ATTR_IFINDEX, ifindex);
1610
1611         return genlmsg_end(skb, ovs_header);
1612
1613 nla_put_failure:
1614         err = -EMSGSIZE;
1615 error:
1616         genlmsg_cancel(skb, ovs_header);
1617         return err;
1618 }
1619
1620 /* Called with RTNL lock or RCU read lock. */
1621 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1622                                          u32 seq, u8 cmd)
1623 {
1624         struct sk_buff *skb;
1625         int retval;
1626
1627         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1628         if (!skb)
1629                 return ERR_PTR(-ENOMEM);
1630
1631         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1632         if (retval < 0) {
1633                 kfree_skb(skb);
1634                 return ERR_PTR(retval);
1635         }
1636         return skb;
1637 }
1638
1639 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1640 {
1641         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1642 }
1643
1644 /* Called with RTNL lock or RCU read lock. */
1645 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1646                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1647 {
1648         struct datapath *dp;
1649         struct vport *vport;
1650
1651         if (a[OVS_VPORT_ATTR_NAME]) {
1652                 vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1653                 if (!vport)
1654                         return ERR_PTR(-ENODEV);
1655                 return vport;
1656         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1657                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1658
1659                 if (port_no >= DP_MAX_PORTS)
1660                         return ERR_PTR(-EFBIG);
1661
1662                 dp = get_dp(ovs_header->dp_ifindex);
1663                 if (!dp)
1664                         return ERR_PTR(-ENODEV);
1665
1666                 vport = get_vport_protected(dp, port_no);
1667                 if (!vport)
1668                         return ERR_PTR(-ENOENT);
1669                 return vport;
1670         } else
1671                 return ERR_PTR(-EINVAL);
1672 }
1673
1674 /* Called with RTNL lock. */
1675 static int change_vport(struct vport *vport, struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1676 {
1677         int err = 0;
1678         if (a[OVS_VPORT_ATTR_STATS])
1679                 err = vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1680         if (!err && a[OVS_VPORT_ATTR_ADDRESS])
1681                 err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1682         if (!err && a[OVS_VPORT_ATTR_MTU])
1683                 err = vport_set_mtu(vport, nla_get_u32(a[OVS_VPORT_ATTR_MTU]));
1684         return err;
1685 }
1686
1687 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1688 {
1689         struct nlattr **a = info->attrs;
1690         struct ovs_header *ovs_header = info->userhdr;
1691         struct vport_parms parms;
1692         struct sk_buff *reply;
1693         struct vport *vport;
1694         struct datapath *dp;
1695         u32 port_no;
1696         int err;
1697
1698         err = -EINVAL;
1699         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE])
1700                 goto exit;
1701
1702         err = ovs_vport_cmd_validate(a);
1703         if (err)
1704                 goto exit;
1705
1706         rtnl_lock();
1707         dp = get_dp(ovs_header->dp_ifindex);
1708         err = -ENODEV;
1709         if (!dp)
1710                 goto exit_unlock;
1711
1712         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1713                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1714
1715                 err = -EFBIG;
1716                 if (port_no >= DP_MAX_PORTS)
1717                         goto exit_unlock;
1718
1719                 vport = get_vport_protected(dp, port_no);
1720                 err = -EBUSY;
1721                 if (vport)
1722                         goto exit_unlock;
1723         } else {
1724                 for (port_no = 1; ; port_no++) {
1725                         if (port_no >= DP_MAX_PORTS) {
1726                                 err = -EFBIG;
1727                                 goto exit_unlock;
1728                         }
1729                         vport = get_vport_protected(dp, port_no);
1730                         if (!vport)
1731                                 break;
1732                 }
1733         }
1734
1735         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1736         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1737         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1738         parms.dp = dp;
1739         parms.port_no = port_no;
1740
1741         vport = new_vport(&parms);
1742         err = PTR_ERR(vport);
1743         if (IS_ERR(vport))
1744                 goto exit_unlock;
1745
1746         set_internal_devs_mtu(dp);
1747         dp_sysfs_add_if(vport);
1748
1749         err = change_vport(vport, a);
1750         if (!err) {
1751                 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1752                                                  info->snd_seq, OVS_VPORT_CMD_NEW);
1753                 if (IS_ERR(reply))
1754                         err = PTR_ERR(reply);
1755         }
1756         if (err) {
1757                 dp_detach_port(vport);
1758                 goto exit_unlock;
1759         }
1760         genl_notify(reply, genl_info_net(info), info->snd_pid,
1761                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1762
1763
1764 exit_unlock:
1765         rtnl_unlock();
1766 exit:
1767         return err;
1768 }
1769
1770 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1771 {
1772         struct nlattr **a = info->attrs;
1773         struct sk_buff *reply;
1774         struct vport *vport;
1775         int err;
1776
1777         err = ovs_vport_cmd_validate(a);
1778         if (err)
1779                 goto exit;
1780
1781         rtnl_lock();
1782         vport = lookup_vport(info->userhdr, a);
1783         err = PTR_ERR(vport);
1784         if (IS_ERR(vport))
1785                 goto exit_unlock;
1786
1787         err = 0;
1788         if (a[OVS_VPORT_ATTR_OPTIONS])
1789                 err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1790         if (!err)
1791                 err = change_vport(vport, a);
1792
1793         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1794                                          OVS_VPORT_CMD_NEW);
1795         if (IS_ERR(reply)) {
1796                 err = PTR_ERR(reply);
1797                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1798                                 dp_vport_multicast_group.id, err);
1799                 return 0;
1800         }
1801
1802         genl_notify(reply, genl_info_net(info), info->snd_pid,
1803                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1804
1805 exit_unlock:
1806         rtnl_unlock();
1807 exit:
1808         return err;
1809 }
1810
1811 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1812 {
1813         struct nlattr **a = info->attrs;
1814         struct sk_buff *reply;
1815         struct vport *vport;
1816         int err;
1817
1818         err = ovs_vport_cmd_validate(a);
1819         if (err)
1820                 goto exit;
1821
1822         rtnl_lock();
1823         vport = lookup_vport(info->userhdr, a);
1824         err = PTR_ERR(vport);
1825         if (IS_ERR(vport))
1826                 goto exit_unlock;
1827
1828         if (vport->port_no == OVSP_LOCAL) {
1829                 err = -EINVAL;
1830                 goto exit_unlock;
1831         }
1832
1833         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1834                                          OVS_VPORT_CMD_DEL);
1835         err = PTR_ERR(reply);
1836         if (IS_ERR(reply))
1837                 goto exit_unlock;
1838
1839         dp_detach_port(vport);
1840
1841         genl_notify(reply, genl_info_net(info), info->snd_pid,
1842                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1843
1844 exit_unlock:
1845         rtnl_unlock();
1846 exit:
1847         return err;
1848 }
1849
1850 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1851 {
1852         struct nlattr **a = info->attrs;
1853         struct ovs_header *ovs_header = info->userhdr;
1854         struct sk_buff *reply;
1855         struct vport *vport;
1856         int err;
1857
1858         err = ovs_vport_cmd_validate(a);
1859         if (err)
1860                 goto exit;
1861
1862         rcu_read_lock();
1863         vport = lookup_vport(ovs_header, a);
1864         err = PTR_ERR(vport);
1865         if (IS_ERR(vport))
1866                 goto exit_unlock;
1867
1868         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1869                                          OVS_VPORT_CMD_NEW);
1870         err = PTR_ERR(reply);
1871         if (IS_ERR(reply))
1872                 goto exit_unlock;
1873
1874         rcu_read_unlock();
1875
1876         return genlmsg_reply(reply, info);
1877
1878 exit_unlock:
1879         rcu_read_unlock();
1880 exit:
1881         return err;
1882 }
1883
1884 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1885 {
1886         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1887         struct datapath *dp;
1888         u32 port_no;
1889         int retval;
1890
1891         dp = get_dp(ovs_header->dp_ifindex);
1892         if (!dp)
1893                 return -ENODEV;
1894
1895         rcu_read_lock();
1896         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1897                 struct vport *vport;
1898
1899                 vport = get_vport_protected(dp, port_no);
1900                 if (!vport)
1901                         continue;
1902
1903                 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1904                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1905                                             OVS_VPORT_CMD_NEW) < 0)
1906                         break;
1907         }
1908         rcu_read_unlock();
1909
1910         cb->args[0] = port_no;
1911         retval = skb->len;
1912
1913         return retval;
1914 }
1915
1916 static struct genl_ops dp_vport_genl_ops[] = {
1917         { .cmd = OVS_VPORT_CMD_NEW,
1918           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1919           .policy = vport_policy,
1920           .doit = ovs_vport_cmd_new
1921         },
1922         { .cmd = OVS_VPORT_CMD_DEL,
1923           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1924           .policy = vport_policy,
1925           .doit = ovs_vport_cmd_del
1926         },
1927         { .cmd = OVS_VPORT_CMD_GET,
1928           .flags = 0,               /* OK for unprivileged users. */
1929           .policy = vport_policy,
1930           .doit = ovs_vport_cmd_get,
1931           .dumpit = ovs_vport_cmd_dump
1932         },
1933         { .cmd = OVS_VPORT_CMD_SET,
1934           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1935           .policy = vport_policy,
1936           .doit = ovs_vport_cmd_set,
1937         },
1938 };
1939
1940 struct genl_family_and_ops {
1941         struct genl_family *family;
1942         struct genl_ops *ops;
1943         int n_ops;
1944         struct genl_multicast_group *group;
1945 };
1946
1947 static const struct genl_family_and_ops dp_genl_families[] = {
1948         { &dp_datapath_genl_family,
1949           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1950           &dp_datapath_multicast_group },
1951         { &dp_vport_genl_family,
1952           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1953           &dp_vport_multicast_group },
1954         { &dp_flow_genl_family,
1955           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1956           &dp_flow_multicast_group },
1957         { &dp_packet_genl_family,
1958           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1959           NULL },
1960 };
1961
1962 static void dp_unregister_genl(int n_families)
1963 {
1964         int i;
1965
1966         for (i = 0; i < n_families; i++)
1967                 genl_unregister_family(dp_genl_families[i].family);
1968 }
1969
1970 static int dp_register_genl(void)
1971 {
1972         int n_registered;
1973         int err;
1974         int i;
1975
1976         n_registered = 0;
1977         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1978                 const struct genl_family_and_ops *f = &dp_genl_families[i];
1979
1980                 err = genl_register_family_with_ops(f->family, f->ops,
1981                                                     f->n_ops);
1982                 if (err)
1983                         goto error;
1984                 n_registered++;
1985
1986                 if (f->group) {
1987                         err = genl_register_mc_group(f->family, f->group);
1988                         if (err)
1989                                 goto error;
1990                 }
1991         }
1992
1993         err = packet_register_mc_groups();
1994         if (err)
1995                 goto error;
1996         return 0;
1997
1998 error:
1999         dp_unregister_genl(n_registered);
2000         return err;
2001 }
2002
2003 static int __init dp_init(void)
2004 {
2005         struct sk_buff *dummy_skb;
2006         int err;
2007
2008         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2009
2010         printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2011
2012         err = tnl_init();
2013         if (err)
2014                 goto error;
2015
2016         err = flow_init();
2017         if (err)
2018                 goto error_tnl_exit;
2019
2020         err = vport_init();
2021         if (err)
2022                 goto error_flow_exit;
2023
2024         err = register_netdevice_notifier(&dp_device_notifier);
2025         if (err)
2026                 goto error_vport_exit;
2027
2028         err = dp_register_genl();
2029         if (err < 0)
2030                 goto error_unreg_notifier;
2031
2032         return 0;
2033
2034 error_unreg_notifier:
2035         unregister_netdevice_notifier(&dp_device_notifier);
2036 error_vport_exit:
2037         vport_exit();
2038 error_flow_exit:
2039         flow_exit();
2040 error_tnl_exit:
2041         tnl_exit();
2042 error:
2043         return err;
2044 }
2045
2046 static void dp_cleanup(void)
2047 {
2048         rcu_barrier();
2049         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2050         unregister_netdevice_notifier(&dp_device_notifier);
2051         vport_exit();
2052         flow_exit();
2053         tnl_exit();
2054 }
2055
2056 module_init(dp_init);
2057 module_exit(dp_cleanup);
2058
2059 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2060 MODULE_LICENSE("GPL");