datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007-2012 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/version.h>
  40 #include <linux/ethtool.h>
  41 #include <linux/wait.h>
  42 #include <asm/div64.h>
  43 #include <linux/highmem.h>
  44 #include <linux/netfilter_bridge.h>
  45 #include <linux/netfilter_ipv4.h>
  46 #include <linux/inetdevice.h>
  47 #include <linux/list.h>
  48 #include <linux/openvswitch.h>
  49 #include <linux/rculist.h>
  50 #include <linux/dmi.h>
  51 #include <net/genetlink.h>
  52 #include <net/net_namespace.h>
  53 #include <net/netns/generic.h>
  54
  55 #include "checksum.h"
  56 #include "datapath.h"
  57 #include "flow.h"
  58 #include "genl_exec.h"
  59 #include "vlan.h"
  60 #include "tunnel.h"
  61 #include "vport-internal_dev.h"
  62
  63 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
  64     LINUX_VERSION_CODE >= KERNEL_VERSION(3,5,0)
  65 #error Kernels before 2.6.18 or after 3.4 are not supported by this version of Open vSwitch.
  66 #endif
  67
  68 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
  69 static void rehash_flow_table(struct work_struct *work);
  70 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
  71
  72 int ovs_net_id __read_mostly;
  73
  74 int (*ovs_dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
  75 EXPORT_SYMBOL(ovs_dp_ioctl_hook);
  76
  77 /**
  78  * DOC: Locking:
  79  *
  80  * Writes to device state (add/remove datapath, port, set operations on vports,
  81  * etc.) are protected by RTNL.
  82  *
  83  * Writes to other state (flow table modifications, set miscellaneous datapath
  84  * parameters, etc.) are protected by genl_mutex.  The RTNL lock nests inside
  85  * genl_mutex.
  86  *
  87  * Reads are protected by RCU.
  88  *
  89  * There are a few special cases (mostly stats) that have their own
  90  * synchronization but they nest under all of above and don't interact with
  91  * each other.
  92  */
  93
  94 static struct vport *new_vport(const struct vport_parms *);
  95 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
  96                              const struct dp_upcall_info *);
  97 static int queue_userspace_packet(struct net *, int dp_ifindex,
  98                                   struct sk_buff *,
  99                                   const struct dp_upcall_info *);
 100
 101 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
 102 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 103 {
 104         struct datapath *dp = NULL;
 105         struct net_device *dev;
 106
 107         rcu_read_lock();
 108         dev = dev_get_by_index_rcu(net, dp_ifindex);
 109         if (dev) {
 110                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 111                 if (vport)
 112                         dp = vport->dp;
 113         }
 114         rcu_read_unlock();
 115
 116         return dp;
 117 }
 118
 119 /* Must be called with rcu_read_lock or RTNL lock. */
 120 const char *ovs_dp_name(const struct datapath *dp)
 121 {
 122         struct vport *vport = ovs_vport_rtnl_rcu(dp, OVSP_LOCAL);
 123         return vport->ops->get_name(vport);
 124 }
 125
 126 static int get_dpifindex(struct datapath *dp)
 127 {
 128         struct vport *local;
 129         int ifindex;
 130
 131         rcu_read_lock();
 132
 133         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 134         if (local)
 135                 ifindex = local->ops->get_ifindex(local);
 136         else
 137                 ifindex = 0;
 138
 139         rcu_read_unlock();
 140
 141         return ifindex;
 142 }
 143
 144 static size_t br_nlmsg_size(void)
 145 {
 146         return NLMSG_ALIGN(sizeof(struct ifinfomsg))
 147                + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
 148                + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
 149                + nla_total_size(4) /* IFLA_MASTER */
 150                + nla_total_size(4) /* IFLA_MTU */
 151                + nla_total_size(1); /* IFLA_OPERSTATE */
 152 }
 153
 154 /* Caller must hold RTNL lock. */
 155 static int dp_fill_ifinfo(struct sk_buff *skb,
 156                           const struct vport *port,
 157                           int event, unsigned int flags)
 158 {
 159         struct datapath *dp = port->dp;
 160         struct ifinfomsg *hdr;
 161         struct nlmsghdr *nlh;
 162
 163         if (!port->ops->get_ifindex)
 164                 return -ENODEV;
 165
 166         nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
 167         if (nlh == NULL)
 168                 return -EMSGSIZE;
 169
 170         hdr = nlmsg_data(nlh);
 171         hdr->ifi_family = AF_BRIDGE;
 172         hdr->__ifi_pad = 0;
 173         hdr->ifi_type = ARPHRD_ETHER;
 174         hdr->ifi_index = port->ops->get_ifindex(port);
 175         hdr->ifi_flags = port->ops->get_dev_flags(port);
 176         hdr->ifi_change = 0;
 177
 178         if (nla_put_string(skb, IFLA_IFNAME, port->ops->get_name(port)) ||
 179             nla_put_u32(skb, IFLA_MASTER, get_dpifindex(dp)) ||
 180             nla_put_u32(skb, IFLA_MTU, port->ops->get_mtu(port)) ||
 181 #ifdef IFLA_OPERSTATE
 182             nla_put_u8(skb, IFLA_OPERSTATE,
 183                        port->ops->is_running(port) ?
 184                                 port->ops->get_operstate(port) :
 185                                 IF_OPER_DOWN) ||
 186 #endif
 187             nla_put(skb, IFLA_ADDRESS, ETH_ALEN, port->ops->get_addr(port)))
 188                 goto nla_put_failure;
 189
 190         return nlmsg_end(skb, nlh);
 191
 192 nla_put_failure:
 193         nlmsg_cancel(skb, nlh);
 194         return -EMSGSIZE;
 195 }
 196
 197 /* Caller must hold RTNL lock. */
 198 static void dp_ifinfo_notify(int event, struct vport *port)
 199 {
 200         struct sk_buff *skb;
 201         int err;
 202
 203         skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
 204         if (!skb) {
 205                 err = -ENOBUFS;
 206                 goto err;
 207         }
 208
 209         err = dp_fill_ifinfo(skb, port, event, 0);
 210         if (err < 0) {
 211                 if (err == -ENODEV) {
 212                         goto out;
 213                 } else {
 214                         /* -EMSGSIZE implies BUG in br_nlmsg_size() */
 215                         WARN_ON(err == -EMSGSIZE);
 216                         goto err;
 217                 }
 218         }
 219
 220         rtnl_notify(skb, ovs_dp_get_net(port->dp), 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 221
 222         return;
 223 err:
 224         rtnl_set_sk_err(ovs_dp_get_net(port->dp), RTNLGRP_LINK, err);
 225 out:
 226         kfree_skb(skb);
 227 }
 228
 229 static void release_dp(struct kobject *kobj)
 230 {
 231         struct datapath *dp = container_of(kobj, struct datapath, ifobj);
 232         kfree(dp);
 233 }
 234
 235 static struct kobj_type dp_ktype = {
 236         .release = release_dp
 237 };
 238
 239 static void destroy_dp_rcu(struct rcu_head *rcu)
 240 {
 241         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 242
 243         ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
 244         free_percpu(dp->stats_percpu);
 245         release_net(ovs_dp_get_net(dp));
 246         kfree(dp->ports);
 247         kobject_put(&dp->ifobj);
 248 }
 249
 250 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 251                                             u16 port_no)
 252 {
 253         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 254 }
 255
 256 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 257 {
 258         struct vport *vport;
 259         struct hlist_node *n;
 260         struct hlist_head *head;
 261
 262         head = vport_hash_bucket(dp, port_no);
 263         hlist_for_each_entry_rcu(vport, n, head, dp_hash_node) {
 264                 if (vport->port_no == port_no)
 265                         return vport;
 266         }
 267         return NULL;
 268 }
 269
 270 /* Called with RTNL lock and genl_lock. */
 271 static struct vport *new_vport(const struct vport_parms *parms)
 272 {
 273         struct vport *vport;
 274
 275         vport = ovs_vport_add(parms);
 276         if (!IS_ERR(vport)) {
 277                 struct datapath *dp = parms->dp;
 278                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 279
 280                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 281                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 282         }
 283         return vport;
 284 }
 285
 286 /* Called with RTNL lock. */
 287 void ovs_dp_detach_port(struct vport *p)
 288 {
 289         ASSERT_RTNL();
 290
 291         if (p->port_no != OVSP_LOCAL)
 292                 ovs_dp_sysfs_del_if(p);
 293
 294         dp_ifinfo_notify(RTM_DELLINK, p);
 295
 296         /* First drop references to device. */
 297         hlist_del_rcu(&p->dp_hash_node);
 298
 299         /* Then destroy it. */
 300         ovs_vport_del(p);
 301 }
 302
 303 /* Must be called with rcu_read_lock. */
 304 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 305 {
 306         struct datapath *dp = p->dp;
 307         struct sw_flow *flow;
 308         struct dp_stats_percpu *stats;
 309         u64 *stats_counter;
 310         int error;
 311
 312         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 313
 314         if (!OVS_CB(skb)->flow) {
 315                 struct sw_flow_key key;
 316                 int key_len;
 317
 318                 /* Extract flow from 'skb' into 'key'. */
 319                 error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
 320                 if (unlikely(error)) {
 321                         kfree_skb(skb);
 322                         return;
 323                 }
 324
 325                 /* Look up flow. */
 326                 flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table),
 327                                            &key, key_len);
 328                 if (unlikely(!flow)) {
 329                         struct dp_upcall_info upcall;
 330
 331                         upcall.cmd = OVS_PACKET_CMD_MISS;
 332                         upcall.key = &key;
 333                         upcall.userdata = NULL;
 334                         upcall.pid = p->upcall_pid;
 335                         ovs_dp_upcall(dp, skb, &upcall);
 336                         consume_skb(skb);
 337                         stats_counter = &stats->n_missed;
 338                         goto out;
 339                 }
 340
 341                 OVS_CB(skb)->flow = flow;
 342         }
 343
 344         stats_counter = &stats->n_hit;
 345         ovs_flow_used(OVS_CB(skb)->flow, skb);
 346         ovs_execute_actions(dp, skb);
 347
 348 out:
 349         /* Update datapath statistics. */
 350         u64_stats_update_begin(&stats->sync);
 351         (*stats_counter)++;
 352         u64_stats_update_end(&stats->sync);
 353 }
 354
 355 static struct genl_family dp_packet_genl_family = {
 356         .id = GENL_ID_GENERATE,
 357         .hdrsize = sizeof(struct ovs_header),
 358         .name = OVS_PACKET_FAMILY,
 359         .version = OVS_PACKET_VERSION,
 360         .maxattr = OVS_PACKET_ATTR_MAX,
 361          SET_NETNSOK
 362 };
 363
 364 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 365                   const struct dp_upcall_info *upcall_info)
 366 {
 367         struct dp_stats_percpu *stats;
 368         int dp_ifindex;
 369         int err;
 370
 371         if (upcall_info->pid == 0) {
 372                 err = -ENOTCONN;
 373                 goto err;
 374         }
 375
 376         dp_ifindex = get_dpifindex(dp);
 377         if (!dp_ifindex) {
 378                 err = -ENODEV;
 379                 goto err;
 380         }
 381
 382         forward_ip_summed(skb, true);
 383
 384         if (!skb_is_gso(skb))
 385                 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 386         else
 387                 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 388         if (err)
 389                 goto err;
 390
 391         return 0;
 392
 393 err:
 394         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 395
 396         u64_stats_update_begin(&stats->sync);
 397         stats->n_lost++;
 398         u64_stats_update_end(&stats->sync);
 399
 400         return err;
 401 }
 402
 403 static int queue_gso_packets(struct net *net, int dp_ifindex,
 404                              struct sk_buff *skb,
 405                              const struct dp_upcall_info *upcall_info)
 406 {
 407         struct dp_upcall_info later_info;
 408         struct sw_flow_key later_key;
 409         struct sk_buff *segs, *nskb;
 410         int err;
 411
 412         segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 413         if (IS_ERR(segs))
 414                 return PTR_ERR(segs);
 415
 416         /* Queue all of the segments. */
 417         skb = segs;
 418         do {
 419                 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
 420                 if (err)
 421                         break;
 422
 423                 if (skb == segs && skb_shinfo(skb)->gso_type & SKB_GSO_UDP) {
 424                         /* The initial flow key extracted by ovs_flow_extract()
 425                          * in this case is for a first fragment, so we need to
 426                          * properly mark later fragments.
 427                          */
 428                         later_key = *upcall_info->key;
 429                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 430
 431                         later_info = *upcall_info;
 432                         later_info.key = &later_key;
 433                         upcall_info = &later_info;
 434                 }
 435         } while ((skb = skb->next));
 436
 437         /* Free all of the segments. */
 438         skb = segs;
 439         do {
 440                 nskb = skb->next;
 441                 if (err)
 442                         kfree_skb(skb);
 443                 else
 444                         consume_skb(skb);
 445         } while ((skb = nskb));
 446         return err;
 447 }
 448
 449 static int queue_userspace_packet(struct net *net, int dp_ifindex,
 450                                   struct sk_buff *skb,
 451                                   const struct dp_upcall_info *upcall_info)
 452 {
 453         struct ovs_header *upcall;
 454         struct sk_buff *nskb = NULL;
 455         struct sk_buff *user_skb; /* to be queued to userspace */
 456         struct nlattr *nla;
 457         unsigned int len;
 458         int err;
 459
 460         if (vlan_tx_tag_present(skb)) {
 461                 nskb = skb_clone(skb, GFP_ATOMIC);
 462                 if (!nskb)
 463                         return -ENOMEM;
 464
 465                 err = vlan_deaccel_tag(nskb);
 466                 if (err)
 467                         return err;
 468
 469                 skb = nskb;
 470         }
 471
 472         if (nla_attr_size(skb->len) > USHRT_MAX) {
 473                 err = -EFBIG;
 474                 goto out;
 475         }
 476
 477         len = sizeof(struct ovs_header);
 478         len += nla_total_size(skb->len);
 479         len += nla_total_size(FLOW_BUFSIZE);
 480         if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
 481                 len += nla_total_size(8);
 482
 483         user_skb = genlmsg_new(len, GFP_ATOMIC);
 484         if (!user_skb) {
 485                 err = -ENOMEM;
 486                 goto out;
 487         }
 488
 489         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 490                              0, upcall_info->cmd);
 491         upcall->dp_ifindex = dp_ifindex;
 492
 493         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 494         ovs_flow_to_nlattrs(upcall_info->key, user_skb);
 495         nla_nest_end(user_skb, nla);
 496
 497         if (upcall_info->userdata)
 498                 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
 499                             nla_get_u64(upcall_info->userdata));
 500
 501         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 502
 503         skb_copy_and_csum_dev(skb, nla_data(nla));
 504
 505         err = genlmsg_unicast(net, user_skb, upcall_info->pid);
 506
 507 out:
 508         kfree_skb(nskb);
 509         return err;
 510 }
 511
 512 /* Called with genl_mutex. */
 513 static int flush_flows(struct datapath *dp)
 514 {
 515         struct flow_table *old_table;
 516         struct flow_table *new_table;
 517
 518         old_table = genl_dereference(dp->table);
 519         new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
 520         if (!new_table)
 521                 return -ENOMEM;
 522
 523         rcu_assign_pointer(dp->table, new_table);
 524
 525         ovs_flow_tbl_deferred_destroy(old_table);
 526         return 0;
 527 }
 528
 529 static int validate_actions(const struct nlattr *attr,
 530                                 const struct sw_flow_key *key, int depth);
 531
 532 static int validate_sample(const struct nlattr *attr,
 533                                 const struct sw_flow_key *key, int depth)
 534 {
 535         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 536         const struct nlattr *probability, *actions;
 537         const struct nlattr *a;
 538         int rem;
 539
 540         memset(attrs, 0, sizeof(attrs));
 541         nla_for_each_nested(a, attr, rem) {
 542                 int type = nla_type(a);
 543                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 544                         return -EINVAL;
 545                 attrs[type] = a;
 546         }
 547         if (rem)
 548                 return -EINVAL;
 549
 550         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 551         if (!probability || nla_len(probability) != sizeof(u32))
 552                 return -EINVAL;
 553
 554         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 555         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 556                 return -EINVAL;
 557         return validate_actions(actions, key, depth + 1);
 558 }
 559
 560 static int validate_tp_port(const struct sw_flow_key *flow_key)
 561 {
 562         if (flow_key->eth.type == htons(ETH_P_IP)) {
 563                 if (flow_key->ipv4.tp.src && flow_key->ipv4.tp.dst)
 564                         return 0;
 565         } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
 566                 if (flow_key->ipv6.tp.src && flow_key->ipv6.tp.dst)
 567                         return 0;
 568         }
 569
 570         return -EINVAL;
 571 }
 572
 573 static int validate_set(const struct nlattr *a,
 574                         const struct sw_flow_key *flow_key)
 575 {
 576         const struct nlattr *ovs_key = nla_data(a);
 577         int key_type = nla_type(ovs_key);
 578
 579         /* There can be only one key in a action */
 580         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 581                 return -EINVAL;
 582
 583         if (key_type > OVS_KEY_ATTR_MAX ||
 584             nla_len(ovs_key) != ovs_key_lens[key_type])
 585                 return -EINVAL;
 586
 587         switch (key_type) {
 588         const struct ovs_key_ipv4 *ipv4_key;
 589
 590         case OVS_KEY_ATTR_PRIORITY:
 591         case OVS_KEY_ATTR_TUN_ID:
 592         case OVS_KEY_ATTR_ETHERNET:
 593                 break;
 594
 595         case OVS_KEY_ATTR_IPV4:
 596                 if (flow_key->eth.type != htons(ETH_P_IP))
 597                         return -EINVAL;
 598
 599                 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
 600                         return -EINVAL;
 601
 602                 ipv4_key = nla_data(ovs_key);
 603                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 604                         return -EINVAL;
 605
 606                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 607                         return -EINVAL;
 608
 609                 break;
 610
 611         case OVS_KEY_ATTR_TCP:
 612                 if (flow_key->ip.proto != IPPROTO_TCP)
 613                         return -EINVAL;
 614
 615                 return validate_tp_port(flow_key);
 616
 617         case OVS_KEY_ATTR_UDP:
 618                 if (flow_key->ip.proto != IPPROTO_UDP)
 619                         return -EINVAL;
 620
 621                 return validate_tp_port(flow_key);
 622
 623         default:
 624                 return -EINVAL;
 625         }
 626
 627         return 0;
 628 }
 629
 630 static int validate_userspace(const struct nlattr *attr)
 631 {
 632         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 633                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 634                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
 635         };
 636         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 637         int error;
 638
 639         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 640                                  attr, userspace_policy);
 641         if (error)
 642                 return error;
 643
 644         if (!a[OVS_USERSPACE_ATTR_PID] ||
 645             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 646                 return -EINVAL;
 647
 648         return 0;
 649 }
 650
 651 static int validate_actions(const struct nlattr *attr,
 652                                 const struct sw_flow_key *key,  int depth)
 653 {
 654         const struct nlattr *a;
 655         int rem, err;
 656
 657         if (depth >= SAMPLE_ACTION_DEPTH)
 658                 return -EOVERFLOW;
 659
 660         nla_for_each_nested(a, attr, rem) {
 661                 /* Expected argument lengths, (u32)-1 for variable length. */
 662                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 663                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 664                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 665                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 666                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 667                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 668                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 669                 };
 670                 const struct ovs_action_push_vlan *vlan;
 671                 int type = nla_type(a);
 672
 673                 if (type > OVS_ACTION_ATTR_MAX ||
 674                     (action_lens[type] != nla_len(a) &&
 675                      action_lens[type] != (u32)-1))
 676                         return -EINVAL;
 677
 678                 switch (type) {
 679                 case OVS_ACTION_ATTR_UNSPEC:
 680                         return -EINVAL;
 681
 682                 case OVS_ACTION_ATTR_USERSPACE:
 683                         err = validate_userspace(a);
 684                         if (err)
 685                                 return err;
 686                         break;
 687
 688                 case OVS_ACTION_ATTR_OUTPUT:
 689                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 690                                 return -EINVAL;
 691                         break;
 692
 693
 694                 case OVS_ACTION_ATTR_POP_VLAN:
 695                         break;
 696
 697                 case OVS_ACTION_ATTR_PUSH_VLAN:
 698                         vlan = nla_data(a);
 699                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 700                                 return -EINVAL;
 701                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 702                                 return -EINVAL;
 703                         break;
 704
 705                 case OVS_ACTION_ATTR_SET:
 706                         err = validate_set(a, key);
 707                         if (err)
 708                                 return err;
 709                         break;
 710
 711                 case OVS_ACTION_ATTR_SAMPLE:
 712                         err = validate_sample(a, key, depth);
 713                         if (err)
 714                                 return err;
 715                         break;
 716
 717                 default:
 718                         return -EINVAL;
 719                 }
 720         }
 721
 722         if (rem > 0)
 723                 return -EINVAL;
 724
 725         return 0;
 726 }
 727
 728 static void clear_stats(struct sw_flow *flow)
 729 {
 730         flow->used = 0;
 731         flow->tcp_flags = 0;
 732         flow->packet_count = 0;
 733         flow->byte_count = 0;
 734 }
 735
 736 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 737 {
 738         struct ovs_header *ovs_header = info->userhdr;
 739         struct nlattr **a = info->attrs;
 740         struct sw_flow_actions *acts;
 741         struct sk_buff *packet;
 742         struct sw_flow *flow;
 743         struct datapath *dp;
 744         struct ethhdr *eth;
 745         int len;
 746         int err;
 747         int key_len;
 748
 749         err = -EINVAL;
 750         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 751             !a[OVS_PACKET_ATTR_ACTIONS] ||
 752             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 753                 goto err;
 754
 755         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 756         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 757         err = -ENOMEM;
 758         if (!packet)
 759                 goto err;
 760         skb_reserve(packet, NET_IP_ALIGN);
 761
 762         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 763
 764         skb_reset_mac_header(packet);
 765         eth = eth_hdr(packet);
 766
 767         /* Normally, setting the skb 'protocol' field would be handled by a
 768          * call to eth_type_trans(), but it assumes there's a sending
 769          * device, which we may not have. */
 770         if (ntohs(eth->h_proto) >= 1536)
 771                 packet->protocol = eth->h_proto;
 772         else
 773                 packet->protocol = htons(ETH_P_802_2);
 774
 775         /* Build an sw_flow for sending this packet. */
 776         flow = ovs_flow_alloc();
 777         err = PTR_ERR(flow);
 778         if (IS_ERR(flow))
 779                 goto err_kfree_skb;
 780
 781         err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
 782         if (err)
 783                 goto err_flow_put;
 784
 785         err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
 786                                              &flow->key.phy.in_port,
 787                                              &flow->key.phy.tun_id,
 788                                              a[OVS_PACKET_ATTR_KEY]);
 789         if (err)
 790                 goto err_flow_put;
 791
 792         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
 793         if (err)
 794                 goto err_flow_put;
 795
 796         flow->hash = ovs_flow_hash(&flow->key, key_len);
 797
 798         acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 799         err = PTR_ERR(acts);
 800         if (IS_ERR(acts))
 801                 goto err_flow_put;
 802         rcu_assign_pointer(flow->sf_acts, acts);
 803
 804         OVS_CB(packet)->flow = flow;
 805         packet->priority = flow->key.phy.priority;
 806
 807         rcu_read_lock();
 808         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 809         err = -ENODEV;
 810         if (!dp)
 811                 goto err_unlock;
 812
 813         local_bh_disable();
 814         err = ovs_execute_actions(dp, packet);
 815         local_bh_enable();
 816         rcu_read_unlock();
 817
 818         ovs_flow_put(flow);
 819         return err;
 820
 821 err_unlock:
 822         rcu_read_unlock();
 823 err_flow_put:
 824         ovs_flow_put(flow);
 825 err_kfree_skb:
 826         kfree_skb(packet);
 827 err:
 828         return err;
 829 }
 830
 831 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 832         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 833         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 834         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 835 };
 836
 837 static struct genl_ops dp_packet_genl_ops[] = {
 838         { .cmd = OVS_PACKET_CMD_EXECUTE,
 839           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 840           .policy = packet_policy,
 841           .doit = ovs_packet_cmd_execute
 842         }
 843 };
 844
 845 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 846 {
 847         int i;
 848         struct flow_table *table = genl_dereference(dp->table);
 849
 850         stats->n_flows = ovs_flow_tbl_count(table);
 851
 852         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 853         for_each_possible_cpu(i) {
 854                 const struct dp_stats_percpu *percpu_stats;
 855                 struct dp_stats_percpu local_stats;
 856                 unsigned int start;
 857
 858                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 859
 860                 do {
 861                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 862                         local_stats = *percpu_stats;
 863                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 864
 865                 stats->n_hit += local_stats.n_hit;
 866                 stats->n_missed += local_stats.n_missed;
 867                 stats->n_lost += local_stats.n_lost;
 868         }
 869 }
 870
 871 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 872         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 873         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 874         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 875 };
 876
 877 static struct genl_family dp_flow_genl_family = {
 878         .id = GENL_ID_GENERATE,
 879         .hdrsize = sizeof(struct ovs_header),
 880         .name = OVS_FLOW_FAMILY,
 881         .version = OVS_FLOW_VERSION,
 882         .maxattr = OVS_FLOW_ATTR_MAX,
 883          SET_NETNSOK
 884 };
 885
 886 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 887         .name = OVS_FLOW_MCGROUP
 888 };
 889
 890 /* Called with genl_lock. */
 891 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 892                                   struct sk_buff *skb, u32 pid,
 893                                   u32 seq, u32 flags, u8 cmd)
 894 {
 895         const int skb_orig_len = skb->len;
 896         const struct sw_flow_actions *sf_acts;
 897         struct ovs_flow_stats stats;
 898         struct ovs_header *ovs_header;
 899         struct nlattr *nla;
 900         unsigned long used;
 901         u8 tcp_flags;
 902         int err;
 903
 904         sf_acts = rcu_dereference_protected(flow->sf_acts,
 905                                             lockdep_genl_is_held());
 906
 907         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 908         if (!ovs_header)
 909                 return -EMSGSIZE;
 910
 911         ovs_header->dp_ifindex = get_dpifindex(dp);
 912
 913         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 914         if (!nla)
 915                 goto nla_put_failure;
 916         err = ovs_flow_to_nlattrs(&flow->key, skb);
 917         if (err)
 918                 goto error;
 919         nla_nest_end(skb, nla);
 920
 921         spin_lock_bh(&flow->lock);
 922         used = flow->used;
 923         stats.n_packets = flow->packet_count;
 924         stats.n_bytes = flow->byte_count;
 925         tcp_flags = flow->tcp_flags;
 926         spin_unlock_bh(&flow->lock);
 927
 928         if (used &&
 929             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 930                 goto nla_put_failure;
 931
 932         if (stats.n_packets &&
 933             nla_put(skb, OVS_FLOW_ATTR_STATS,
 934                     sizeof(struct ovs_flow_stats), &stats))
 935                 goto nla_put_failure;
 936
 937         if (tcp_flags &&
 938             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
 939                 goto nla_put_failure;
 940
 941         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 942          * this is the first flow to be dumped into 'skb'.  This is unusual for
 943          * Netlink but individual action lists can be longer than
 944          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 945          * The userspace caller can always fetch the actions separately if it
 946          * really wants them.  (Most userspace callers in fact don't care.)
 947          *
 948          * This can only fail for dump operations because the skb is always
 949          * properly sized for single flows.
 950          */
 951         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 952                       sf_acts->actions);
 953         if (err < 0 && skb_orig_len)
 954                 goto error;
 955
 956         return genlmsg_end(skb, ovs_header);
 957
 958 nla_put_failure:
 959         err = -EMSGSIZE;
 960 error:
 961         genlmsg_cancel(skb, ovs_header);
 962         return err;
 963 }
 964
 965 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 966 {
 967         const struct sw_flow_actions *sf_acts;
 968         int len;
 969
 970         sf_acts = rcu_dereference_protected(flow->sf_acts,
 971                                             lockdep_genl_is_held());
 972
 973         /* OVS_FLOW_ATTR_KEY */
 974         len = nla_total_size(FLOW_BUFSIZE);
 975         /* OVS_FLOW_ATTR_ACTIONS */
 976         len += nla_total_size(sf_acts->actions_len);
 977         /* OVS_FLOW_ATTR_STATS */
 978         len += nla_total_size(sizeof(struct ovs_flow_stats));
 979         /* OVS_FLOW_ATTR_TCP_FLAGS */
 980         len += nla_total_size(1);
 981         /* OVS_FLOW_ATTR_USED */
 982         len += nla_total_size(8);
 983
 984         len += NLMSG_ALIGN(sizeof(struct ovs_header));
 985
 986         return genlmsg_new(len, GFP_KERNEL);
 987 }
 988
 989 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 990                                                struct datapath *dp,
 991                                                u32 pid, u32 seq, u8 cmd)
 992 {
 993         struct sk_buff *skb;
 994         int retval;
 995
 996         skb = ovs_flow_cmd_alloc_info(flow);
 997         if (!skb)
 998                 return ERR_PTR(-ENOMEM);
 999
1000         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
1001         BUG_ON(retval < 0);
1002         return skb;
1003 }
1004
1005 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1006 {
1007         struct nlattr **a = info->attrs;
1008         struct ovs_header *ovs_header = info->userhdr;
1009         struct sw_flow_key key;
1010         struct sw_flow *flow;
1011         struct sk_buff *reply;
1012         struct datapath *dp;
1013         struct flow_table *table;
1014         int error;
1015         int key_len;
1016
1017         /* Extract key. */
1018         error = -EINVAL;
1019         if (!a[OVS_FLOW_ATTR_KEY])
1020                 goto error;
1021         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1022         if (error)
1023                 goto error;
1024
1025         /* Validate actions. */
1026         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1027                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
1028                 if (error)
1029                         goto error;
1030         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1031                 error = -EINVAL;
1032                 goto error;
1033         }
1034
1035         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1036         error = -ENODEV;
1037         if (!dp)
1038                 goto error;
1039
1040         table = genl_dereference(dp->table);
1041         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1042         if (!flow) {
1043                 struct sw_flow_actions *acts;
1044
1045                 /* Bail out if we're not allowed to create a new flow. */
1046                 error = -ENOENT;
1047                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1048                         goto error;
1049
1050                 /* Expand table, if necessary, to make room. */
1051                 if (ovs_flow_tbl_need_to_expand(table)) {
1052                         struct flow_table *new_table;
1053
1054                         new_table = ovs_flow_tbl_expand(table);
1055                         if (!IS_ERR(new_table)) {
1056                                 rcu_assign_pointer(dp->table, new_table);
1057                                 ovs_flow_tbl_deferred_destroy(table);
1058                                 table = genl_dereference(dp->table);
1059                         }
1060                 }
1061
1062                 /* Allocate flow. */
1063                 flow = ovs_flow_alloc();
1064                 if (IS_ERR(flow)) {
1065                         error = PTR_ERR(flow);
1066                         goto error;
1067                 }
1068                 flow->key = key;
1069                 clear_stats(flow);
1070
1071                 /* Obtain actions. */
1072                 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1073                 error = PTR_ERR(acts);
1074                 if (IS_ERR(acts))
1075                         goto error_free_flow;
1076                 rcu_assign_pointer(flow->sf_acts, acts);
1077
1078                 /* Put flow in bucket. */
1079                 flow->hash = ovs_flow_hash(&key, key_len);
1080                 ovs_flow_tbl_insert(table, flow);
1081
1082                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1083                                                 info->snd_seq,
1084                                                 OVS_FLOW_CMD_NEW);
1085         } else {
1086                 /* We found a matching flow. */
1087                 struct sw_flow_actions *old_acts;
1088                 struct nlattr *acts_attrs;
1089
1090                 /* Bail out if we're not allowed to modify an existing flow.
1091                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1092                  * because Generic Netlink treats the latter as a dump
1093                  * request.  We also accept NLM_F_EXCL in case that bug ever
1094                  * gets fixed.
1095                  */
1096                 error = -EEXIST;
1097                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1098                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1099                         goto error;
1100
1101                 /* Update actions. */
1102                 old_acts = rcu_dereference_protected(flow->sf_acts,
1103                                                      lockdep_genl_is_held());
1104                 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1105                 if (acts_attrs &&
1106                    (old_acts->actions_len != nla_len(acts_attrs) ||
1107                    memcmp(old_acts->actions, nla_data(acts_attrs),
1108                           old_acts->actions_len))) {
1109                         struct sw_flow_actions *new_acts;
1110
1111                         new_acts = ovs_flow_actions_alloc(acts_attrs);
1112                         error = PTR_ERR(new_acts);
1113                         if (IS_ERR(new_acts))
1114                                 goto error;
1115
1116                         rcu_assign_pointer(flow->sf_acts, new_acts);
1117                         ovs_flow_deferred_free_acts(old_acts);
1118                 }
1119
1120                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1121                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1122
1123                 /* Clear stats. */
1124                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1125                         spin_lock_bh(&flow->lock);
1126                         clear_stats(flow);
1127                         spin_unlock_bh(&flow->lock);
1128                 }
1129         }
1130
1131         if (!IS_ERR(reply))
1132                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1133                            ovs_dp_flow_multicast_group.id, info->nlhdr,
1134                            GFP_KERNEL);
1135         else
1136                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1137                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1138         return 0;
1139
1140 error_free_flow:
1141         ovs_flow_put(flow);
1142 error:
1143         return error;
1144 }
1145
1146 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1147 {
1148         struct nlattr **a = info->attrs;
1149         struct ovs_header *ovs_header = info->userhdr;
1150         struct sw_flow_key key;
1151         struct sk_buff *reply;
1152         struct sw_flow *flow;
1153         struct datapath *dp;
1154         struct flow_table *table;
1155         int err;
1156         int key_len;
1157
1158         if (!a[OVS_FLOW_ATTR_KEY])
1159                 return -EINVAL;
1160         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1161         if (err)
1162                 return err;
1163
1164         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1165         if (!dp)
1166                 return -ENODEV;
1167
1168         table = genl_dereference(dp->table);
1169         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1170         if (!flow)
1171                 return -ENOENT;
1172
1173         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1174                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1175         if (IS_ERR(reply))
1176                 return PTR_ERR(reply);
1177
1178         return genlmsg_reply(reply, info);
1179 }
1180
1181 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1182 {
1183         struct nlattr **a = info->attrs;
1184         struct ovs_header *ovs_header = info->userhdr;
1185         struct sw_flow_key key;
1186         struct sk_buff *reply;
1187         struct sw_flow *flow;
1188         struct datapath *dp;
1189         struct flow_table *table;
1190         int err;
1191         int key_len;
1192
1193         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1194         if (!dp)
1195                 return -ENODEV;
1196
1197         if (!a[OVS_FLOW_ATTR_KEY])
1198                 return flush_flows(dp);
1199
1200         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1201         if (err)
1202                 return err;
1203
1204         table = genl_dereference(dp->table);
1205         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1206         if (!flow)
1207                 return -ENOENT;
1208
1209         reply = ovs_flow_cmd_alloc_info(flow);
1210         if (!reply)
1211                 return -ENOMEM;
1212
1213         ovs_flow_tbl_remove(table, flow);
1214
1215         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1216                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1217         BUG_ON(err < 0);
1218
1219         ovs_flow_deferred_free(flow);
1220
1221         genl_notify(reply, genl_info_net(info), info->snd_pid,
1222                     ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1223         return 0;
1224 }
1225
1226 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1227 {
1228         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1229         struct datapath *dp;
1230         struct flow_table *table;
1231
1232         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1233         if (!dp)
1234                 return -ENODEV;
1235
1236         table = genl_dereference(dp->table);
1237
1238         for (;;) {
1239                 struct sw_flow *flow;
1240                 u32 bucket, obj;
1241
1242                 bucket = cb->args[0];
1243                 obj = cb->args[1];
1244                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1245                 if (!flow)
1246                         break;
1247
1248                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1249                                            NETLINK_CB(cb->skb).pid,
1250                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1251                                            OVS_FLOW_CMD_NEW) < 0)
1252                         break;
1253
1254                 cb->args[0] = bucket;
1255                 cb->args[1] = obj;
1256         }
1257         return skb->len;
1258 }
1259
1260 static struct genl_ops dp_flow_genl_ops[] = {
1261         { .cmd = OVS_FLOW_CMD_NEW,
1262           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1263           .policy = flow_policy,
1264           .doit = ovs_flow_cmd_new_or_set
1265         },
1266         { .cmd = OVS_FLOW_CMD_DEL,
1267           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1268           .policy = flow_policy,
1269           .doit = ovs_flow_cmd_del
1270         },
1271         { .cmd = OVS_FLOW_CMD_GET,
1272           .flags = 0,               /* OK for unprivileged users. */
1273           .policy = flow_policy,
1274           .doit = ovs_flow_cmd_get,
1275           .dumpit = ovs_flow_cmd_dump
1276         },
1277         { .cmd = OVS_FLOW_CMD_SET,
1278           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1279           .policy = flow_policy,
1280           .doit = ovs_flow_cmd_new_or_set,
1281         },
1282 };
1283
1284 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1285 #ifdef HAVE_NLA_NUL_STRING
1286         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1287 #endif
1288         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1289 };
1290
1291 static struct genl_family dp_datapath_genl_family = {
1292         .id = GENL_ID_GENERATE,
1293         .hdrsize = sizeof(struct ovs_header),
1294         .name = OVS_DATAPATH_FAMILY,
1295         .version = OVS_DATAPATH_VERSION,
1296         .maxattr = OVS_DP_ATTR_MAX,
1297          SET_NETNSOK
1298 };
1299
1300 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1301         .name = OVS_DATAPATH_MCGROUP
1302 };
1303
1304 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1305                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1306 {
1307         struct ovs_header *ovs_header;
1308         struct ovs_dp_stats dp_stats;
1309         int err;
1310
1311         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1312                                    flags, cmd);
1313         if (!ovs_header)
1314                 goto error;
1315
1316         ovs_header->dp_ifindex = get_dpifindex(dp);
1317
1318         rcu_read_lock();
1319         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1320         rcu_read_unlock();
1321         if (err)
1322                 goto nla_put_failure;
1323
1324         get_dp_stats(dp, &dp_stats);
1325         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1326                 goto nla_put_failure;
1327
1328         return genlmsg_end(skb, ovs_header);
1329
1330 nla_put_failure:
1331         genlmsg_cancel(skb, ovs_header);
1332 error:
1333         return -EMSGSIZE;
1334 }
1335
1336 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1337                                              u32 seq, u8 cmd)
1338 {
1339         struct sk_buff *skb;
1340         int retval;
1341
1342         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1343         if (!skb)
1344                 return ERR_PTR(-ENOMEM);
1345
1346         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1347         if (retval < 0) {
1348                 kfree_skb(skb);
1349                 return ERR_PTR(retval);
1350         }
1351         return skb;
1352 }
1353
1354 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1355 {
1356         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1357 }
1358
1359 /* Called with genl_mutex and optionally with RTNL lock also. */
1360 static struct datapath *lookup_datapath(struct net *net,
1361                                         struct ovs_header *ovs_header,
1362                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1363 {
1364         struct datapath *dp;
1365
1366         if (!a[OVS_DP_ATTR_NAME])
1367                 dp = get_dp(net, ovs_header->dp_ifindex);
1368         else {
1369                 struct vport *vport;
1370
1371                 rcu_read_lock();
1372                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1373                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1374                 rcu_read_unlock();
1375         }
1376         return dp ? dp : ERR_PTR(-ENODEV);
1377 }
1378
1379 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1380 {
1381         struct nlattr **a = info->attrs;
1382         struct vport_parms parms;
1383         struct sk_buff *reply;
1384         struct datapath *dp;
1385         struct vport *vport;
1386         struct ovs_net *ovs_net;
1387         int err, i;
1388
1389         err = -EINVAL;
1390         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1391                 goto err;
1392
1393         err = ovs_dp_cmd_validate(a);
1394         if (err)
1395                 goto err;
1396
1397         rtnl_lock();
1398
1399         err = -ENOMEM;
1400         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1401         if (dp == NULL)
1402                 goto err_unlock_rtnl;
1403
1404         /* Initialize kobject for bridge.  This will be added as
1405          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1406         dp->ifobj.kset = NULL;
1407         kobject_init(&dp->ifobj, &dp_ktype);
1408
1409         /* Allocate table. */
1410         err = -ENOMEM;
1411         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1412         if (!dp->table)
1413                 goto err_free_dp;
1414
1415         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1416         if (!dp->stats_percpu) {
1417                 err = -ENOMEM;
1418                 goto err_destroy_table;
1419         }
1420         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1421
1422         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1423                             GFP_KERNEL);
1424         if (!dp->ports) {
1425                 err = -ENOMEM;
1426                 goto err_destroy_percpu;
1427         }
1428
1429         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1430                 INIT_HLIST_HEAD(&dp->ports[i]);
1431
1432         /* Set up our datapath device. */
1433         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1434         parms.type = OVS_VPORT_TYPE_INTERNAL;
1435         parms.options = NULL;
1436         parms.dp = dp;
1437         parms.port_no = OVSP_LOCAL;
1438         parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1439
1440         vport = new_vport(&parms);
1441         if (IS_ERR(vport)) {
1442                 err = PTR_ERR(vport);
1443                 if (err == -EBUSY)
1444                         err = -EEXIST;
1445
1446                 goto err_destroy_ports_array;
1447         }
1448
1449         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1450                                       info->snd_seq, OVS_DP_CMD_NEW);
1451         err = PTR_ERR(reply);
1452         if (IS_ERR(reply))
1453                 goto err_destroy_local_port;
1454
1455         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1456         list_add_tail(&dp->list_node, &ovs_net->dps);
1457         ovs_dp_sysfs_add_dp(dp);
1458
1459         rtnl_unlock();
1460
1461         genl_notify(reply, genl_info_net(info), info->snd_pid,
1462                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1463                     GFP_KERNEL);
1464         return 0;
1465
1466 err_destroy_local_port:
1467         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1468 err_destroy_ports_array:
1469         kfree(dp->ports);
1470 err_destroy_percpu:
1471         free_percpu(dp->stats_percpu);
1472 err_destroy_table:
1473         ovs_flow_tbl_destroy(genl_dereference(dp->table));
1474 err_free_dp:
1475         kfree(dp);
1476 err_unlock_rtnl:
1477         rtnl_unlock();
1478 err:
1479         return err;
1480 }
1481
1482 /* Called with genl_mutex. */
1483 static void __dp_destroy(struct datapath *dp)
1484 {
1485         int i;
1486
1487         rtnl_lock();
1488
1489         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1490                 struct vport *vport;
1491                 struct hlist_node *node, *n;
1492
1493                 hlist_for_each_entry_safe(vport, node, n, &dp->ports[i], dp_hash_node)
1494                         if (vport->port_no != OVSP_LOCAL)
1495                                 ovs_dp_detach_port(vport);
1496         }
1497
1498         ovs_dp_sysfs_del_dp(dp);
1499         list_del(&dp->list_node);
1500         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1501
1502         /* rtnl_unlock() will wait until all the references to devices that
1503          * are pending unregistration have been dropped.  We do it here to
1504          * ensure that any internal devices (which contain DP pointers) are
1505          * fully destroyed before freeing the datapath.
1506          */
1507         rtnl_unlock();
1508
1509         call_rcu(&dp->rcu, destroy_dp_rcu);
1510 }
1511
1512 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1513 {
1514         struct sk_buff *reply;
1515         struct datapath *dp;
1516         int err;
1517
1518         err = ovs_dp_cmd_validate(info->attrs);
1519         if (err)
1520                 return err;
1521
1522         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1523         err = PTR_ERR(dp);
1524         if (IS_ERR(dp))
1525                 return err;
1526
1527         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1528                                       info->snd_seq, OVS_DP_CMD_DEL);
1529         err = PTR_ERR(reply);
1530         if (IS_ERR(reply))
1531                 return err;
1532
1533         __dp_destroy(dp);
1534
1535         genl_notify(reply, genl_info_net(info), info->snd_pid,
1536                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1537                     GFP_KERNEL);
1538
1539         return 0;
1540 }
1541
1542 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1543 {
1544         struct sk_buff *reply;
1545         struct datapath *dp;
1546         int err;
1547
1548         err = ovs_dp_cmd_validate(info->attrs);
1549         if (err)
1550                 return err;
1551
1552         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1553         if (IS_ERR(dp))
1554                 return PTR_ERR(dp);
1555
1556         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1557                                       info->snd_seq, OVS_DP_CMD_NEW);
1558         if (IS_ERR(reply)) {
1559                 err = PTR_ERR(reply);
1560                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1561                                 ovs_dp_datapath_multicast_group.id, err);
1562                 return 0;
1563         }
1564
1565         genl_notify(reply, genl_info_net(info), info->snd_pid,
1566                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1567                     GFP_KERNEL);
1568
1569         return 0;
1570 }
1571
1572 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1573 {
1574         struct sk_buff *reply;
1575         struct datapath *dp;
1576         int err;
1577
1578         err = ovs_dp_cmd_validate(info->attrs);
1579         if (err)
1580                 return err;
1581
1582         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1583         if (IS_ERR(dp))
1584                 return PTR_ERR(dp);
1585
1586         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1587                                       info->snd_seq, OVS_DP_CMD_NEW);
1588         if (IS_ERR(reply))
1589                 return PTR_ERR(reply);
1590
1591         return genlmsg_reply(reply, info);
1592 }
1593
1594 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1595 {
1596         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1597         struct datapath *dp;
1598         int skip = cb->args[0];
1599         int i = 0;
1600
1601         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1602                 if (i >= skip &&
1603                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1604                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1605                                          OVS_DP_CMD_NEW) < 0)
1606                         break;
1607                 i++;
1608         }
1609
1610         cb->args[0] = i;
1611
1612         return skb->len;
1613 }
1614
1615 static struct genl_ops dp_datapath_genl_ops[] = {
1616         { .cmd = OVS_DP_CMD_NEW,
1617           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1618           .policy = datapath_policy,
1619           .doit = ovs_dp_cmd_new
1620         },
1621         { .cmd = OVS_DP_CMD_DEL,
1622           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1623           .policy = datapath_policy,
1624           .doit = ovs_dp_cmd_del
1625         },
1626         { .cmd = OVS_DP_CMD_GET,
1627           .flags = 0,               /* OK for unprivileged users. */
1628           .policy = datapath_policy,
1629           .doit = ovs_dp_cmd_get,
1630           .dumpit = ovs_dp_cmd_dump
1631         },
1632         { .cmd = OVS_DP_CMD_SET,
1633           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1634           .policy = datapath_policy,
1635           .doit = ovs_dp_cmd_set,
1636         },
1637 };
1638
1639 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1640 #ifdef HAVE_NLA_NUL_STRING
1641         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1642         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1643         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1644 #else
1645         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1646         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1647 #endif
1648         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1649         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1650         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1651         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1652 };
1653
1654 static struct genl_family dp_vport_genl_family = {
1655         .id = GENL_ID_GENERATE,
1656         .hdrsize = sizeof(struct ovs_header),
1657         .name = OVS_VPORT_FAMILY,
1658         .version = OVS_VPORT_VERSION,
1659         .maxattr = OVS_VPORT_ATTR_MAX,
1660          SET_NETNSOK
1661 };
1662
1663 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1664         .name = OVS_VPORT_MCGROUP
1665 };
1666
1667 /* Called with RTNL lock or RCU read lock. */
1668 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1669                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1670 {
1671         struct ovs_header *ovs_header;
1672         struct ovs_vport_stats vport_stats;
1673         int err;
1674
1675         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1676                                  flags, cmd);
1677         if (!ovs_header)
1678                 return -EMSGSIZE;
1679
1680         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1681
1682         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1683             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1684             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1685             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid))
1686                 goto nla_put_failure;
1687
1688         ovs_vport_get_stats(vport, &vport_stats);
1689         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1690                     &vport_stats))
1691                 goto nla_put_failure;
1692
1693         if (nla_put(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN,
1694                     vport->ops->get_addr(vport)))
1695                 goto nla_put_failure;
1696
1697         err = ovs_vport_get_options(vport, skb);
1698         if (err == -EMSGSIZE)
1699                 goto error;
1700
1701         return genlmsg_end(skb, ovs_header);
1702
1703 nla_put_failure:
1704         err = -EMSGSIZE;
1705 error:
1706         genlmsg_cancel(skb, ovs_header);
1707         return err;
1708 }
1709
1710 /* Called with RTNL lock or RCU read lock. */
1711 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1712                                          u32 seq, u8 cmd)
1713 {
1714         struct sk_buff *skb;
1715         int retval;
1716
1717         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1718         if (!skb)
1719                 return ERR_PTR(-ENOMEM);
1720
1721         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1722         if (retval < 0) {
1723                 kfree_skb(skb);
1724                 return ERR_PTR(retval);
1725         }
1726         return skb;
1727 }
1728
1729 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1730 {
1731         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1732 }
1733
1734 /* Called with RTNL lock or RCU read lock. */
1735 static struct vport *lookup_vport(struct net *net,
1736                                   struct ovs_header *ovs_header,
1737                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1738 {
1739         struct datapath *dp;
1740         struct vport *vport;
1741
1742         if (a[OVS_VPORT_ATTR_NAME]) {
1743                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1744                 if (!vport)
1745                         return ERR_PTR(-ENODEV);
1746                 if (ovs_header->dp_ifindex &&
1747                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1748                         return ERR_PTR(-ENODEV);
1749                 return vport;
1750         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1751                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1752
1753                 if (port_no >= DP_MAX_PORTS)
1754                         return ERR_PTR(-EFBIG);
1755
1756                 dp = get_dp(net, ovs_header->dp_ifindex);
1757                 if (!dp)
1758                         return ERR_PTR(-ENODEV);
1759
1760                 vport = ovs_vport_rtnl_rcu(dp, port_no);
1761                 if (!vport)
1762                         return ERR_PTR(-ENOENT);
1763                 return vport;
1764         } else
1765                 return ERR_PTR(-EINVAL);
1766 }
1767
1768 /* Called with RTNL lock. */
1769 static int change_vport(struct vport *vport,
1770                         struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1771 {
1772         int err = 0;
1773
1774         if (a[OVS_VPORT_ATTR_STATS])
1775                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1776
1777         if (a[OVS_VPORT_ATTR_ADDRESS])
1778                 err = ovs_vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1779
1780         return err;
1781 }
1782
1783 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1784 {
1785         struct nlattr **a = info->attrs;
1786         struct ovs_header *ovs_header = info->userhdr;
1787         struct vport_parms parms;
1788         struct sk_buff *reply;
1789         struct vport *vport;
1790         struct datapath *dp;
1791         u32 port_no;
1792         int err;
1793
1794         err = -EINVAL;
1795         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1796             !a[OVS_VPORT_ATTR_UPCALL_PID])
1797                 goto exit;
1798
1799         err = ovs_vport_cmd_validate(a);
1800         if (err)
1801                 goto exit;
1802
1803         rtnl_lock();
1804         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1805         err = -ENODEV;
1806         if (!dp)
1807                 goto exit_unlock;
1808
1809         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1810                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1811
1812                 err = -EFBIG;
1813                 if (port_no >= DP_MAX_PORTS)
1814                         goto exit_unlock;
1815
1816                 vport = ovs_vport_rtnl(dp, port_no);
1817                 err = -EBUSY;
1818                 if (vport)
1819                         goto exit_unlock;
1820         } else {
1821                 for (port_no = 1; ; port_no++) {
1822                         if (port_no >= DP_MAX_PORTS) {
1823                                 err = -EFBIG;
1824                                 goto exit_unlock;
1825                         }
1826                         vport = ovs_vport_rtnl(dp, port_no);
1827                         if (!vport)
1828                                 break;
1829                 }
1830         }
1831
1832         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1833         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1834         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1835         parms.dp = dp;
1836         parms.port_no = port_no;
1837         parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1838
1839         vport = new_vport(&parms);
1840         err = PTR_ERR(vport);
1841         if (IS_ERR(vport))
1842                 goto exit_unlock;
1843
1844         ovs_dp_sysfs_add_if(vport);
1845
1846         err = change_vport(vport, a);
1847         if (!err) {
1848                 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1849                                                  info->snd_seq,
1850                                                  OVS_VPORT_CMD_NEW);
1851                 if (IS_ERR(reply))
1852                         err = PTR_ERR(reply);
1853         }
1854         if (err) {
1855                 ovs_dp_detach_port(vport);
1856                 goto exit_unlock;
1857         }
1858         genl_notify(reply, genl_info_net(info), info->snd_pid,
1859                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1860
1861 exit_unlock:
1862         rtnl_unlock();
1863 exit:
1864         return err;
1865 }
1866
1867 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1868 {
1869         struct nlattr **a = info->attrs;
1870         struct sk_buff *reply;
1871         struct vport *vport;
1872         int err;
1873
1874         err = ovs_vport_cmd_validate(a);
1875         if (err)
1876                 goto exit;
1877
1878         rtnl_lock();
1879         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1880         err = PTR_ERR(vport);
1881         if (IS_ERR(vport))
1882                 goto exit_unlock;
1883
1884         err = 0;
1885         if (a[OVS_VPORT_ATTR_TYPE] &&
1886             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1887                 err = -EINVAL;
1888
1889         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1890                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1891         if (!err)
1892                 err = change_vport(vport, a);
1893         else
1894                 goto exit_unlock;
1895         if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1896                 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1897
1898         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1899                                          OVS_VPORT_CMD_NEW);
1900         if (IS_ERR(reply)) {
1901                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1902                                 ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
1903                 goto exit_unlock;
1904         }
1905
1906         genl_notify(reply, genl_info_net(info), info->snd_pid,
1907                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1908
1909 exit_unlock:
1910         rtnl_unlock();
1911 exit:
1912         return err;
1913 }
1914
1915 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1916 {
1917         struct nlattr **a = info->attrs;
1918         struct sk_buff *reply;
1919         struct vport *vport;
1920         int err;
1921
1922         err = ovs_vport_cmd_validate(a);
1923         if (err)
1924                 goto exit;
1925
1926         rtnl_lock();
1927         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1928         err = PTR_ERR(vport);
1929         if (IS_ERR(vport))
1930                 goto exit_unlock;
1931
1932         if (vport->port_no == OVSP_LOCAL) {
1933                 err = -EINVAL;
1934                 goto exit_unlock;
1935         }
1936
1937         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1938                                          OVS_VPORT_CMD_DEL);
1939         err = PTR_ERR(reply);
1940         if (IS_ERR(reply))
1941                 goto exit_unlock;
1942
1943         ovs_dp_detach_port(vport);
1944
1945         genl_notify(reply, genl_info_net(info), info->snd_pid,
1946                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1947
1948 exit_unlock:
1949         rtnl_unlock();
1950 exit:
1951         return err;
1952 }
1953
1954 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1955 {
1956         struct nlattr **a = info->attrs;
1957         struct ovs_header *ovs_header = info->userhdr;
1958         struct sk_buff *reply;
1959         struct vport *vport;
1960         int err;
1961
1962         err = ovs_vport_cmd_validate(a);
1963         if (err)
1964                 goto exit;
1965
1966         rcu_read_lock();
1967         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1968         err = PTR_ERR(vport);
1969         if (IS_ERR(vport))
1970                 goto exit_unlock;
1971
1972         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1973                                          OVS_VPORT_CMD_NEW);
1974         err = PTR_ERR(reply);
1975         if (IS_ERR(reply))
1976                 goto exit_unlock;
1977
1978         rcu_read_unlock();
1979
1980         return genlmsg_reply(reply, info);
1981
1982 exit_unlock:
1983         rcu_read_unlock();
1984 exit:
1985         return err;
1986 }
1987
1988 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1989 {
1990         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1991         struct datapath *dp;
1992         int bucket = cb->args[0], skip = cb->args[1];
1993         int i, j = 0;
1994
1995         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1996         if (!dp)
1997                 return -ENODEV;
1998
1999         rcu_read_lock();
2000         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2001                 struct vport *vport;
2002                 struct hlist_node *n;
2003
2004                 j = 0;
2005                 hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node) {
2006                         if (j >= skip &&
2007                             ovs_vport_cmd_fill_info(vport, skb,
2008                                                     NETLINK_CB(cb->skb).pid,
2009                                                     cb->nlh->nlmsg_seq,
2010                                                     NLM_F_MULTI,
2011                                                     OVS_VPORT_CMD_NEW) < 0)
2012                                 goto out;
2013
2014                         j++;
2015                 }
2016                 skip = 0;
2017         }
2018 out:
2019         rcu_read_unlock();
2020
2021         cb->args[0] = i;
2022         cb->args[1] = j;
2023
2024         return skb->len;
2025 }
2026
2027 static struct genl_ops dp_vport_genl_ops[] = {
2028         { .cmd = OVS_VPORT_CMD_NEW,
2029           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2030           .policy = vport_policy,
2031           .doit = ovs_vport_cmd_new
2032         },
2033         { .cmd = OVS_VPORT_CMD_DEL,
2034           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2035           .policy = vport_policy,
2036           .doit = ovs_vport_cmd_del
2037         },
2038         { .cmd = OVS_VPORT_CMD_GET,
2039           .flags = 0,               /* OK for unprivileged users. */
2040           .policy = vport_policy,
2041           .doit = ovs_vport_cmd_get,
2042           .dumpit = ovs_vport_cmd_dump
2043         },
2044         { .cmd = OVS_VPORT_CMD_SET,
2045           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2046           .policy = vport_policy,
2047           .doit = ovs_vport_cmd_set,
2048         },
2049 };
2050
2051 struct genl_family_and_ops {
2052         struct genl_family *family;
2053         struct genl_ops *ops;
2054         int n_ops;
2055         struct genl_multicast_group *group;
2056 };
2057
2058 static const struct genl_family_and_ops dp_genl_families[] = {
2059         { &dp_datapath_genl_family,
2060           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2061           &ovs_dp_datapath_multicast_group },
2062         { &dp_vport_genl_family,
2063           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2064           &ovs_dp_vport_multicast_group },
2065         { &dp_flow_genl_family,
2066           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2067           &ovs_dp_flow_multicast_group },
2068         { &dp_packet_genl_family,
2069           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2070           NULL },
2071 };
2072
2073 static void dp_unregister_genl(int n_families)
2074 {
2075         int i;
2076
2077         for (i = 0; i < n_families; i++)
2078                 genl_unregister_family(dp_genl_families[i].family);
2079 }
2080
2081 static int dp_register_genl(void)
2082 {
2083         int n_registered;
2084         int err;
2085         int i;
2086
2087         n_registered = 0;
2088         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2089                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2090
2091                 err = genl_register_family_with_ops(f->family, f->ops,
2092                                                     f->n_ops);
2093                 if (err)
2094                         goto error;
2095                 n_registered++;
2096
2097                 if (f->group) {
2098                         err = genl_register_mc_group(f->family, f->group);
2099                         if (err)
2100                                 goto error;
2101                 }
2102         }
2103
2104         return 0;
2105
2106 error:
2107         dp_unregister_genl(n_registered);
2108         return err;
2109 }
2110
2111 static int __rehash_flow_table(void *dummy)
2112 {
2113         struct datapath *dp;
2114         struct net *net;
2115
2116         rtnl_lock();
2117         for_each_net(net) {
2118                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2119
2120                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2121                         struct flow_table *old_table = genl_dereference(dp->table);
2122                         struct flow_table *new_table;
2123
2124                         new_table = ovs_flow_tbl_rehash(old_table);
2125                         if (!IS_ERR(new_table)) {
2126                                 rcu_assign_pointer(dp->table, new_table);
2127                                 ovs_flow_tbl_deferred_destroy(old_table);
2128                         }
2129                 }
2130         }
2131         rtnl_unlock();
2132         return 0;
2133 }
2134
2135 static void rehash_flow_table(struct work_struct *work)
2136 {
2137         genl_exec(__rehash_flow_table, NULL);
2138         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2139 }
2140
2141 static int dp_destroy_all(void *data)
2142 {
2143         struct datapath *dp, *dp_next;
2144         struct ovs_net *ovs_net = data;
2145
2146         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2147                 __dp_destroy(dp);
2148
2149         return 0;
2150 }
2151
2152 static int __net_init ovs_init_net(struct net *net)
2153 {
2154         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2155
2156         INIT_LIST_HEAD(&ovs_net->dps);
2157         return 0;
2158 }
2159
2160 static void __net_exit ovs_exit_net(struct net *net)
2161 {
2162         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2163
2164         genl_exec(dp_destroy_all, ovs_net);
2165 }
2166
2167 static struct pernet_operations ovs_net_ops = {
2168         .init = ovs_init_net,
2169         .exit = ovs_exit_net,
2170         .id   = &ovs_net_id,
2171         .size = sizeof(struct ovs_net),
2172 };
2173
2174 static int __init dp_init(void)
2175 {
2176         struct sk_buff *dummy_skb;
2177         int err;
2178
2179         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2180
2181         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2182                 VERSION);
2183
2184         err = genl_exec_init();
2185         if (err)
2186                 goto error;
2187
2188         err = ovs_workqueues_init();
2189         if (err)
2190                 goto error_genl_exec;
2191
2192         err = ovs_tnl_init();
2193         if (err)
2194                 goto error_wq;
2195
2196         err = ovs_flow_init();
2197         if (err)
2198                 goto error_tnl_exit;
2199
2200         err = ovs_vport_init();
2201         if (err)
2202                 goto error_flow_exit;
2203
2204         err = register_pernet_device(&ovs_net_ops);
2205         if (err)
2206                 goto error_vport_exit;
2207
2208         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2209         if (err)
2210                 goto error_netns_exit;
2211
2212         err = dp_register_genl();
2213         if (err < 0)
2214                 goto error_unreg_notifier;
2215
2216         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2217
2218         return 0;
2219
2220 error_unreg_notifier:
2221         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2222 error_netns_exit:
2223         unregister_pernet_device(&ovs_net_ops);
2224 error_vport_exit:
2225         ovs_vport_exit();
2226 error_flow_exit:
2227         ovs_flow_exit();
2228 error_tnl_exit:
2229         ovs_tnl_exit();
2230 error_wq:
2231         ovs_workqueues_exit();
2232 error_genl_exec:
2233         genl_exec_exit();
2234 error:
2235         return err;
2236 }
2237
2238 static void dp_cleanup(void)
2239 {
2240         cancel_delayed_work_sync(&rehash_flow_wq);
2241         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2242         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2243         unregister_pernet_device(&ovs_net_ops);
2244         rcu_barrier();
2245         ovs_vport_exit();
2246         ovs_flow_exit();
2247         ovs_tnl_exit();
2248         ovs_workqueues_exit();
2249         genl_exec_exit();
2250 }
2251
2252 module_init(dp_init);
2253 module_exit(dp_cleanup);
2254
2255 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2256 MODULE_LICENSE("GPL");