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(3,7,0)
  65 #error Kernels before 2.6.18 or after 3.6 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 = this_cpu_ptr(dp->stats_percpu);
 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 = this_cpu_ptr(dp->stats_percpu);
 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         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 408         struct dp_upcall_info later_info;
 409         struct sw_flow_key later_key;
 410         struct sk_buff *segs, *nskb;
 411         int err;
 412
 413         segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 414         if (IS_ERR(segs))
 415                 return PTR_ERR(segs);
 416
 417         /* Queue all of the segments. */
 418         skb = segs;
 419         do {
 420                 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
 421                 if (err)
 422                         break;
 423
 424                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 425                         /* The initial flow key extracted by ovs_flow_extract()
 426                          * in this case is for a first fragment, so we need to
 427                          * properly mark later fragments.
 428                          */
 429                         later_key = *upcall_info->key;
 430                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 431
 432                         later_info = *upcall_info;
 433                         later_info.key = &later_key;
 434                         upcall_info = &later_info;
 435                 }
 436         } while ((skb = skb->next));
 437
 438         /* Free all of the segments. */
 439         skb = segs;
 440         do {
 441                 nskb = skb->next;
 442                 if (err)
 443                         kfree_skb(skb);
 444                 else
 445                         consume_skb(skb);
 446         } while ((skb = nskb));
 447         return err;
 448 }
 449
 450 static int queue_userspace_packet(struct net *net, int dp_ifindex,
 451                                   struct sk_buff *skb,
 452                                   const struct dp_upcall_info *upcall_info)
 453 {
 454         struct ovs_header *upcall;
 455         struct sk_buff *nskb = NULL;
 456         struct sk_buff *user_skb; /* to be queued to userspace */
 457         struct nlattr *nla;
 458         unsigned int len;
 459         int err;
 460
 461         if (vlan_tx_tag_present(skb)) {
 462                 nskb = skb_clone(skb, GFP_ATOMIC);
 463                 if (!nskb)
 464                         return -ENOMEM;
 465
 466                 err = vlan_deaccel_tag(nskb);
 467                 if (err)
 468                         return err;
 469
 470                 skb = nskb;
 471         }
 472
 473         if (nla_attr_size(skb->len) > USHRT_MAX) {
 474                 err = -EFBIG;
 475                 goto out;
 476         }
 477
 478         len = sizeof(struct ovs_header);
 479         len += nla_total_size(skb->len);
 480         len += nla_total_size(FLOW_BUFSIZE);
 481         if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
 482                 len += nla_total_size(8);
 483
 484         user_skb = genlmsg_new(len, GFP_ATOMIC);
 485         if (!user_skb) {
 486                 err = -ENOMEM;
 487                 goto out;
 488         }
 489
 490         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 491                              0, upcall_info->cmd);
 492         upcall->dp_ifindex = dp_ifindex;
 493
 494         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 495         ovs_flow_to_nlattrs(upcall_info->key, user_skb);
 496         nla_nest_end(user_skb, nla);
 497
 498         if (upcall_info->userdata)
 499                 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
 500                             nla_get_u64(upcall_info->userdata));
 501
 502         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 503
 504         skb_copy_and_csum_dev(skb, nla_data(nla));
 505
 506         err = genlmsg_unicast(net, user_skb, upcall_info->pid);
 507
 508 out:
 509         kfree_skb(nskb);
 510         return err;
 511 }
 512
 513 /* Called with genl_mutex. */
 514 static int flush_flows(struct datapath *dp)
 515 {
 516         struct flow_table *old_table;
 517         struct flow_table *new_table;
 518
 519         old_table = genl_dereference(dp->table);
 520         new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
 521         if (!new_table)
 522                 return -ENOMEM;
 523
 524         rcu_assign_pointer(dp->table, new_table);
 525
 526         ovs_flow_tbl_deferred_destroy(old_table);
 527         return 0;
 528 }
 529
 530 static int validate_actions(const struct nlattr *attr,
 531                                 const struct sw_flow_key *key, int depth);
 532
 533 static int validate_sample(const struct nlattr *attr,
 534                                 const struct sw_flow_key *key, int depth)
 535 {
 536         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 537         const struct nlattr *probability, *actions;
 538         const struct nlattr *a;
 539         int rem;
 540
 541         memset(attrs, 0, sizeof(attrs));
 542         nla_for_each_nested(a, attr, rem) {
 543                 int type = nla_type(a);
 544                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 545                         return -EINVAL;
 546                 attrs[type] = a;
 547         }
 548         if (rem)
 549                 return -EINVAL;
 550
 551         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 552         if (!probability || nla_len(probability) != sizeof(u32))
 553                 return -EINVAL;
 554
 555         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 556         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 557                 return -EINVAL;
 558         return validate_actions(actions, key, depth + 1);
 559 }
 560
 561 static int validate_tp_port(const struct sw_flow_key *flow_key)
 562 {
 563         if (flow_key->eth.type == htons(ETH_P_IP)) {
 564                 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
 565                         return 0;
 566         } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
 567                 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
 568                         return 0;
 569         }
 570
 571         return -EINVAL;
 572 }
 573
 574 static int validate_set(const struct nlattr *a,
 575                         const struct sw_flow_key *flow_key)
 576 {
 577         const struct nlattr *ovs_key = nla_data(a);
 578         int key_type = nla_type(ovs_key);
 579
 580         /* There can be only one key in a action */
 581         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 582                 return -EINVAL;
 583
 584         if (key_type > OVS_KEY_ATTR_MAX ||
 585             nla_len(ovs_key) != ovs_key_lens[key_type])
 586                 return -EINVAL;
 587
 588         switch (key_type) {
 589         const struct ovs_key_ipv4 *ipv4_key;
 590         const struct ovs_key_ipv4_tunnel *tun_key;
 591         const struct ovs_key_ipv6 *ipv6_key;
 592
 593         case OVS_KEY_ATTR_PRIORITY:
 594         case OVS_KEY_ATTR_TUN_ID:
 595         case OVS_KEY_ATTR_ETHERNET:
 596                 break;
 597
 598         case OVS_KEY_ATTR_IPV4_TUNNEL:
 599                 tun_key = nla_data(ovs_key);
 600                 if (!tun_key->ipv4_dst)
 601                         return -EINVAL;
 602                 break;
 603
 604         case OVS_KEY_ATTR_IPV4:
 605                 if (flow_key->eth.type != htons(ETH_P_IP))
 606                         return -EINVAL;
 607
 608                 if (!flow_key->ip.proto)
 609                         return -EINVAL;
 610
 611                 ipv4_key = nla_data(ovs_key);
 612                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 613                         return -EINVAL;
 614
 615                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 616                         return -EINVAL;
 617
 618                 break;
 619
 620         case OVS_KEY_ATTR_IPV6:
 621                 if (flow_key->eth.type != htons(ETH_P_IPV6))
 622                         return -EINVAL;
 623
 624                 if (!flow_key->ip.proto)
 625                         return -EINVAL;
 626
 627                 ipv6_key = nla_data(ovs_key);
 628                 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
 629                         return -EINVAL;
 630
 631                 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
 632                         return -EINVAL;
 633
 634                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
 635                         return -EINVAL;
 636
 637                 break;
 638
 639         case OVS_KEY_ATTR_TCP:
 640                 if (flow_key->ip.proto != IPPROTO_TCP)
 641                         return -EINVAL;
 642
 643                 return validate_tp_port(flow_key);
 644
 645         case OVS_KEY_ATTR_UDP:
 646                 if (flow_key->ip.proto != IPPROTO_UDP)
 647                         return -EINVAL;
 648
 649                 return validate_tp_port(flow_key);
 650
 651         default:
 652                 return -EINVAL;
 653         }
 654
 655         return 0;
 656 }
 657
 658 static int validate_userspace(const struct nlattr *attr)
 659 {
 660         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 661                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 662                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
 663         };
 664         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 665         int error;
 666
 667         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 668                                  attr, userspace_policy);
 669         if (error)
 670                 return error;
 671
 672         if (!a[OVS_USERSPACE_ATTR_PID] ||
 673             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 674                 return -EINVAL;
 675
 676         return 0;
 677 }
 678
 679 static int validate_actions(const struct nlattr *attr,
 680                                 const struct sw_flow_key *key,  int depth)
 681 {
 682         const struct nlattr *a;
 683         int rem, err;
 684
 685         if (depth >= SAMPLE_ACTION_DEPTH)
 686                 return -EOVERFLOW;
 687
 688         nla_for_each_nested(a, attr, rem) {
 689                 /* Expected argument lengths, (u32)-1 for variable length. */
 690                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 691                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 692                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 693                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 694                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 695                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 696                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 697                 };
 698                 const struct ovs_action_push_vlan *vlan;
 699                 int type = nla_type(a);
 700
 701                 if (type > OVS_ACTION_ATTR_MAX ||
 702                     (action_lens[type] != nla_len(a) &&
 703                      action_lens[type] != (u32)-1))
 704                         return -EINVAL;
 705
 706                 switch (type) {
 707                 case OVS_ACTION_ATTR_UNSPEC:
 708                         return -EINVAL;
 709
 710                 case OVS_ACTION_ATTR_USERSPACE:
 711                         err = validate_userspace(a);
 712                         if (err)
 713                                 return err;
 714                         break;
 715
 716                 case OVS_ACTION_ATTR_OUTPUT:
 717                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 718                                 return -EINVAL;
 719                         break;
 720
 721
 722                 case OVS_ACTION_ATTR_POP_VLAN:
 723                         break;
 724
 725                 case OVS_ACTION_ATTR_PUSH_VLAN:
 726                         vlan = nla_data(a);
 727                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 728                                 return -EINVAL;
 729                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 730                                 return -EINVAL;
 731                         break;
 732
 733                 case OVS_ACTION_ATTR_SET:
 734                         err = validate_set(a, key);
 735                         if (err)
 736                                 return err;
 737                         break;
 738
 739                 case OVS_ACTION_ATTR_SAMPLE:
 740                         err = validate_sample(a, key, depth);
 741                         if (err)
 742                                 return err;
 743                         break;
 744
 745                 default:
 746                         return -EINVAL;
 747                 }
 748         }
 749
 750         if (rem > 0)
 751                 return -EINVAL;
 752
 753         return 0;
 754 }
 755
 756 static void clear_stats(struct sw_flow *flow)
 757 {
 758         flow->used = 0;
 759         flow->tcp_flags = 0;
 760         flow->packet_count = 0;
 761         flow->byte_count = 0;
 762 }
 763
 764 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 765 {
 766         struct ovs_header *ovs_header = info->userhdr;
 767         struct nlattr **a = info->attrs;
 768         struct sw_flow_actions *acts;
 769         struct sk_buff *packet;
 770         struct sw_flow *flow;
 771         struct datapath *dp;
 772         struct ethhdr *eth;
 773         int len;
 774         int err;
 775         int key_len;
 776
 777         err = -EINVAL;
 778         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 779             !a[OVS_PACKET_ATTR_ACTIONS] ||
 780             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 781                 goto err;
 782
 783         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 784         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 785         err = -ENOMEM;
 786         if (!packet)
 787                 goto err;
 788         skb_reserve(packet, NET_IP_ALIGN);
 789
 790         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 791
 792         skb_reset_mac_header(packet);
 793         eth = eth_hdr(packet);
 794
 795         /* Normally, setting the skb 'protocol' field would be handled by a
 796          * call to eth_type_trans(), but it assumes there's a sending
 797          * device, which we may not have. */
 798         if (ntohs(eth->h_proto) >= 1536)
 799                 packet->protocol = eth->h_proto;
 800         else
 801                 packet->protocol = htons(ETH_P_802_2);
 802
 803         /* Build an sw_flow for sending this packet. */
 804         flow = ovs_flow_alloc();
 805         err = PTR_ERR(flow);
 806         if (IS_ERR(flow))
 807                 goto err_kfree_skb;
 808
 809         err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
 810         if (err)
 811                 goto err_flow_free;
 812
 813         err = ovs_flow_metadata_from_nlattrs(flow, key_len, a[OVS_PACKET_ATTR_KEY]);
 814         if (err)
 815                 goto err_flow_free;
 816
 817         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
 818         if (err)
 819                 goto err_flow_free;
 820
 821         acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 822         err = PTR_ERR(acts);
 823         if (IS_ERR(acts))
 824                 goto err_flow_free;
 825         rcu_assign_pointer(flow->sf_acts, acts);
 826
 827         OVS_CB(packet)->flow = flow;
 828         packet->priority = flow->key.phy.priority;
 829
 830         rcu_read_lock();
 831         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 832         err = -ENODEV;
 833         if (!dp)
 834                 goto err_unlock;
 835
 836         local_bh_disable();
 837         err = ovs_execute_actions(dp, packet);
 838         local_bh_enable();
 839         rcu_read_unlock();
 840
 841         ovs_flow_free(flow);
 842         return err;
 843
 844 err_unlock:
 845         rcu_read_unlock();
 846 err_flow_free:
 847         ovs_flow_free(flow);
 848 err_kfree_skb:
 849         kfree_skb(packet);
 850 err:
 851         return err;
 852 }
 853
 854 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 855         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 856         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 857         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 858 };
 859
 860 static struct genl_ops dp_packet_genl_ops[] = {
 861         { .cmd = OVS_PACKET_CMD_EXECUTE,
 862           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 863           .policy = packet_policy,
 864           .doit = ovs_packet_cmd_execute
 865         }
 866 };
 867
 868 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 869 {
 870         int i;
 871         struct flow_table *table = genl_dereference(dp->table);
 872
 873         stats->n_flows = ovs_flow_tbl_count(table);
 874
 875         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 876         for_each_possible_cpu(i) {
 877                 const struct dp_stats_percpu *percpu_stats;
 878                 struct dp_stats_percpu local_stats;
 879                 unsigned int start;
 880
 881                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 882
 883                 do {
 884                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 885                         local_stats = *percpu_stats;
 886                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 887
 888                 stats->n_hit += local_stats.n_hit;
 889                 stats->n_missed += local_stats.n_missed;
 890                 stats->n_lost += local_stats.n_lost;
 891         }
 892 }
 893
 894 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 895         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 896         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 897         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 898 };
 899
 900 static struct genl_family dp_flow_genl_family = {
 901         .id = GENL_ID_GENERATE,
 902         .hdrsize = sizeof(struct ovs_header),
 903         .name = OVS_FLOW_FAMILY,
 904         .version = OVS_FLOW_VERSION,
 905         .maxattr = OVS_FLOW_ATTR_MAX,
 906          SET_NETNSOK
 907 };
 908
 909 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 910         .name = OVS_FLOW_MCGROUP
 911 };
 912
 913 /* Called with genl_lock. */
 914 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 915                                   struct sk_buff *skb, u32 pid,
 916                                   u32 seq, u32 flags, u8 cmd)
 917 {
 918         const int skb_orig_len = skb->len;
 919         const struct sw_flow_actions *sf_acts;
 920         struct ovs_flow_stats stats;
 921         struct ovs_header *ovs_header;
 922         struct nlattr *nla;
 923         unsigned long used;
 924         u8 tcp_flags;
 925         int err;
 926
 927         sf_acts = rcu_dereference_protected(flow->sf_acts,
 928                                             lockdep_genl_is_held());
 929
 930         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 931         if (!ovs_header)
 932                 return -EMSGSIZE;
 933
 934         ovs_header->dp_ifindex = get_dpifindex(dp);
 935
 936         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 937         if (!nla)
 938                 goto nla_put_failure;
 939         err = ovs_flow_to_nlattrs(&flow->key, skb);
 940         if (err)
 941                 goto error;
 942         nla_nest_end(skb, nla);
 943
 944         spin_lock_bh(&flow->lock);
 945         used = flow->used;
 946         stats.n_packets = flow->packet_count;
 947         stats.n_bytes = flow->byte_count;
 948         tcp_flags = flow->tcp_flags;
 949         spin_unlock_bh(&flow->lock);
 950
 951         if (used &&
 952             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 953                 goto nla_put_failure;
 954
 955         if (stats.n_packets &&
 956             nla_put(skb, OVS_FLOW_ATTR_STATS,
 957                     sizeof(struct ovs_flow_stats), &stats))
 958                 goto nla_put_failure;
 959
 960         if (tcp_flags &&
 961             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
 962                 goto nla_put_failure;
 963
 964         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 965          * this is the first flow to be dumped into 'skb'.  This is unusual for
 966          * Netlink but individual action lists can be longer than
 967          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 968          * The userspace caller can always fetch the actions separately if it
 969          * really wants them.  (Most userspace callers in fact don't care.)
 970          *
 971          * This can only fail for dump operations because the skb is always
 972          * properly sized for single flows.
 973          */
 974         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 975                       sf_acts->actions);
 976         if (err < 0 && skb_orig_len)
 977                 goto error;
 978
 979         return genlmsg_end(skb, ovs_header);
 980
 981 nla_put_failure:
 982         err = -EMSGSIZE;
 983 error:
 984         genlmsg_cancel(skb, ovs_header);
 985         return err;
 986 }
 987
 988 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 989 {
 990         const struct sw_flow_actions *sf_acts;
 991         int len;
 992
 993         sf_acts = rcu_dereference_protected(flow->sf_acts,
 994                                             lockdep_genl_is_held());
 995
 996         /* OVS_FLOW_ATTR_KEY */
 997         len = nla_total_size(FLOW_BUFSIZE);
 998         /* OVS_FLOW_ATTR_ACTIONS */
 999         len += nla_total_size(sf_acts->actions_len);
1000         /* OVS_FLOW_ATTR_STATS */
1001         len += nla_total_size(sizeof(struct ovs_flow_stats));
1002         /* OVS_FLOW_ATTR_TCP_FLAGS */
1003         len += nla_total_size(1);
1004         /* OVS_FLOW_ATTR_USED */
1005         len += nla_total_size(8);
1006
1007         len += NLMSG_ALIGN(sizeof(struct ovs_header));
1008
1009         return genlmsg_new(len, GFP_KERNEL);
1010 }
1011
1012 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1013                                                struct datapath *dp,
1014                                                u32 pid, u32 seq, u8 cmd)
1015 {
1016         struct sk_buff *skb;
1017         int retval;
1018
1019         skb = ovs_flow_cmd_alloc_info(flow);
1020         if (!skb)
1021                 return ERR_PTR(-ENOMEM);
1022
1023         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
1024         BUG_ON(retval < 0);
1025         return skb;
1026 }
1027
1028 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1029 {
1030         struct nlattr **a = info->attrs;
1031         struct ovs_header *ovs_header = info->userhdr;
1032         struct sw_flow_key key;
1033         struct sw_flow *flow;
1034         struct sk_buff *reply;
1035         struct datapath *dp;
1036         struct flow_table *table;
1037         int error;
1038         int key_len;
1039
1040         /* Extract key. */
1041         error = -EINVAL;
1042         if (!a[OVS_FLOW_ATTR_KEY])
1043                 goto error;
1044         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1045         if (error)
1046                 goto error;
1047
1048         /* Validate actions. */
1049         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1050                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
1051                 if (error)
1052                         goto error;
1053         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1054                 error = -EINVAL;
1055                 goto error;
1056         }
1057
1058         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1059         error = -ENODEV;
1060         if (!dp)
1061                 goto error;
1062
1063         table = genl_dereference(dp->table);
1064         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1065         if (!flow) {
1066                 struct sw_flow_actions *acts;
1067
1068                 /* Bail out if we're not allowed to create a new flow. */
1069                 error = -ENOENT;
1070                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1071                         goto error;
1072
1073                 /* Expand table, if necessary, to make room. */
1074                 if (ovs_flow_tbl_need_to_expand(table)) {
1075                         struct flow_table *new_table;
1076
1077                         new_table = ovs_flow_tbl_expand(table);
1078                         if (!IS_ERR(new_table)) {
1079                                 rcu_assign_pointer(dp->table, new_table);
1080                                 ovs_flow_tbl_deferred_destroy(table);
1081                                 table = genl_dereference(dp->table);
1082                         }
1083                 }
1084
1085                 /* Allocate flow. */
1086                 flow = ovs_flow_alloc();
1087                 if (IS_ERR(flow)) {
1088                         error = PTR_ERR(flow);
1089                         goto error;
1090                 }
1091                 clear_stats(flow);
1092
1093                 /* Obtain actions. */
1094                 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1095                 error = PTR_ERR(acts);
1096                 if (IS_ERR(acts))
1097                         goto error_free_flow;
1098                 rcu_assign_pointer(flow->sf_acts, acts);
1099
1100                 /* Put flow in bucket. */
1101                 ovs_flow_tbl_insert(table, flow, &key, key_len);
1102
1103                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1104                                                 info->snd_seq,
1105                                                 OVS_FLOW_CMD_NEW);
1106         } else {
1107                 /* We found a matching flow. */
1108                 struct sw_flow_actions *old_acts;
1109                 struct nlattr *acts_attrs;
1110
1111                 /* Bail out if we're not allowed to modify an existing flow.
1112                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1113                  * because Generic Netlink treats the latter as a dump
1114                  * request.  We also accept NLM_F_EXCL in case that bug ever
1115                  * gets fixed.
1116                  */
1117                 error = -EEXIST;
1118                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1119                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1120                         goto error;
1121
1122                 /* Update actions. */
1123                 old_acts = rcu_dereference_protected(flow->sf_acts,
1124                                                      lockdep_genl_is_held());
1125                 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1126                 if (acts_attrs &&
1127                    (old_acts->actions_len != nla_len(acts_attrs) ||
1128                    memcmp(old_acts->actions, nla_data(acts_attrs),
1129                           old_acts->actions_len))) {
1130                         struct sw_flow_actions *new_acts;
1131
1132                         new_acts = ovs_flow_actions_alloc(acts_attrs);
1133                         error = PTR_ERR(new_acts);
1134                         if (IS_ERR(new_acts))
1135                                 goto error;
1136
1137                         rcu_assign_pointer(flow->sf_acts, new_acts);
1138                         ovs_flow_deferred_free_acts(old_acts);
1139                 }
1140
1141                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1142                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1143
1144                 /* Clear stats. */
1145                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1146                         spin_lock_bh(&flow->lock);
1147                         clear_stats(flow);
1148                         spin_unlock_bh(&flow->lock);
1149                 }
1150         }
1151
1152         if (!IS_ERR(reply))
1153                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1154                            ovs_dp_flow_multicast_group.id, info->nlhdr,
1155                            GFP_KERNEL);
1156         else
1157                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1158                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1159         return 0;
1160
1161 error_free_flow:
1162         ovs_flow_free(flow);
1163 error:
1164         return error;
1165 }
1166
1167 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1168 {
1169         struct nlattr **a = info->attrs;
1170         struct ovs_header *ovs_header = info->userhdr;
1171         struct sw_flow_key key;
1172         struct sk_buff *reply;
1173         struct sw_flow *flow;
1174         struct datapath *dp;
1175         struct flow_table *table;
1176         int err;
1177         int key_len;
1178
1179         if (!a[OVS_FLOW_ATTR_KEY])
1180                 return -EINVAL;
1181         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1182         if (err)
1183                 return err;
1184
1185         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1186         if (!dp)
1187                 return -ENODEV;
1188
1189         table = genl_dereference(dp->table);
1190         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1191         if (!flow)
1192                 return -ENOENT;
1193
1194         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1195                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1196         if (IS_ERR(reply))
1197                 return PTR_ERR(reply);
1198
1199         return genlmsg_reply(reply, info);
1200 }
1201
1202 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1203 {
1204         struct nlattr **a = info->attrs;
1205         struct ovs_header *ovs_header = info->userhdr;
1206         struct sw_flow_key key;
1207         struct sk_buff *reply;
1208         struct sw_flow *flow;
1209         struct datapath *dp;
1210         struct flow_table *table;
1211         int err;
1212         int key_len;
1213
1214         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1215         if (!dp)
1216                 return -ENODEV;
1217
1218         if (!a[OVS_FLOW_ATTR_KEY])
1219                 return flush_flows(dp);
1220
1221         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1222         if (err)
1223                 return err;
1224
1225         table = genl_dereference(dp->table);
1226         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1227         if (!flow)
1228                 return -ENOENT;
1229
1230         reply = ovs_flow_cmd_alloc_info(flow);
1231         if (!reply)
1232                 return -ENOMEM;
1233
1234         ovs_flow_tbl_remove(table, flow);
1235
1236         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1237                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1238         BUG_ON(err < 0);
1239
1240         ovs_flow_deferred_free(flow);
1241
1242         genl_notify(reply, genl_info_net(info), info->snd_pid,
1243                     ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1244         return 0;
1245 }
1246
1247 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1248 {
1249         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1250         struct datapath *dp;
1251         struct flow_table *table;
1252
1253         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1254         if (!dp)
1255                 return -ENODEV;
1256
1257         table = genl_dereference(dp->table);
1258
1259         for (;;) {
1260                 struct sw_flow *flow;
1261                 u32 bucket, obj;
1262
1263                 bucket = cb->args[0];
1264                 obj = cb->args[1];
1265                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1266                 if (!flow)
1267                         break;
1268
1269                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1270                                            NETLINK_CB(cb->skb).pid,
1271                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1272                                            OVS_FLOW_CMD_NEW) < 0)
1273                         break;
1274
1275                 cb->args[0] = bucket;
1276                 cb->args[1] = obj;
1277         }
1278         return skb->len;
1279 }
1280
1281 static struct genl_ops dp_flow_genl_ops[] = {
1282         { .cmd = OVS_FLOW_CMD_NEW,
1283           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1284           .policy = flow_policy,
1285           .doit = ovs_flow_cmd_new_or_set
1286         },
1287         { .cmd = OVS_FLOW_CMD_DEL,
1288           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1289           .policy = flow_policy,
1290           .doit = ovs_flow_cmd_del
1291         },
1292         { .cmd = OVS_FLOW_CMD_GET,
1293           .flags = 0,               /* OK for unprivileged users. */
1294           .policy = flow_policy,
1295           .doit = ovs_flow_cmd_get,
1296           .dumpit = ovs_flow_cmd_dump
1297         },
1298         { .cmd = OVS_FLOW_CMD_SET,
1299           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1300           .policy = flow_policy,
1301           .doit = ovs_flow_cmd_new_or_set,
1302         },
1303 };
1304
1305 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1306 #ifdef HAVE_NLA_NUL_STRING
1307         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1308 #endif
1309         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1310 };
1311
1312 static struct genl_family dp_datapath_genl_family = {
1313         .id = GENL_ID_GENERATE,
1314         .hdrsize = sizeof(struct ovs_header),
1315         .name = OVS_DATAPATH_FAMILY,
1316         .version = OVS_DATAPATH_VERSION,
1317         .maxattr = OVS_DP_ATTR_MAX,
1318          SET_NETNSOK
1319 };
1320
1321 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1322         .name = OVS_DATAPATH_MCGROUP
1323 };
1324
1325 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1326                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1327 {
1328         struct ovs_header *ovs_header;
1329         struct ovs_dp_stats dp_stats;
1330         int err;
1331
1332         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1333                                    flags, cmd);
1334         if (!ovs_header)
1335                 goto error;
1336
1337         ovs_header->dp_ifindex = get_dpifindex(dp);
1338
1339         rcu_read_lock();
1340         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1341         rcu_read_unlock();
1342         if (err)
1343                 goto nla_put_failure;
1344
1345         get_dp_stats(dp, &dp_stats);
1346         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1347                 goto nla_put_failure;
1348
1349         return genlmsg_end(skb, ovs_header);
1350
1351 nla_put_failure:
1352         genlmsg_cancel(skb, ovs_header);
1353 error:
1354         return -EMSGSIZE;
1355 }
1356
1357 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1358                                              u32 seq, u8 cmd)
1359 {
1360         struct sk_buff *skb;
1361         int retval;
1362
1363         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1364         if (!skb)
1365                 return ERR_PTR(-ENOMEM);
1366
1367         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1368         if (retval < 0) {
1369                 kfree_skb(skb);
1370                 return ERR_PTR(retval);
1371         }
1372         return skb;
1373 }
1374
1375 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1376 {
1377         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1378 }
1379
1380 /* Called with genl_mutex and optionally with RTNL lock also. */
1381 static struct datapath *lookup_datapath(struct net *net,
1382                                         struct ovs_header *ovs_header,
1383                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1384 {
1385         struct datapath *dp;
1386
1387         if (!a[OVS_DP_ATTR_NAME])
1388                 dp = get_dp(net, ovs_header->dp_ifindex);
1389         else {
1390                 struct vport *vport;
1391
1392                 rcu_read_lock();
1393                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1394                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1395                 rcu_read_unlock();
1396         }
1397         return dp ? dp : ERR_PTR(-ENODEV);
1398 }
1399
1400 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1401 {
1402         struct nlattr **a = info->attrs;
1403         struct vport_parms parms;
1404         struct sk_buff *reply;
1405         struct datapath *dp;
1406         struct vport *vport;
1407         struct ovs_net *ovs_net;
1408         int err, i;
1409
1410         err = -EINVAL;
1411         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1412                 goto err;
1413
1414         err = ovs_dp_cmd_validate(a);
1415         if (err)
1416                 goto err;
1417
1418         rtnl_lock();
1419
1420         err = -ENOMEM;
1421         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1422         if (dp == NULL)
1423                 goto err_unlock_rtnl;
1424
1425         /* Initialize kobject for bridge.  This will be added as
1426          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1427         dp->ifobj.kset = NULL;
1428         kobject_init(&dp->ifobj, &dp_ktype);
1429
1430         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1431
1432         /* Allocate table. */
1433         err = -ENOMEM;
1434         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1435         if (!dp->table)
1436                 goto err_free_dp;
1437
1438         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1439         if (!dp->stats_percpu) {
1440                 err = -ENOMEM;
1441                 goto err_destroy_table;
1442         }
1443
1444         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1445                             GFP_KERNEL);
1446         if (!dp->ports) {
1447                 err = -ENOMEM;
1448                 goto err_destroy_percpu;
1449         }
1450
1451         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1452                 INIT_HLIST_HEAD(&dp->ports[i]);
1453
1454         /* Set up our datapath device. */
1455         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1456         parms.type = OVS_VPORT_TYPE_INTERNAL;
1457         parms.options = NULL;
1458         parms.dp = dp;
1459         parms.port_no = OVSP_LOCAL;
1460         parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1461
1462         vport = new_vport(&parms);
1463         if (IS_ERR(vport)) {
1464                 err = PTR_ERR(vport);
1465                 if (err == -EBUSY)
1466                         err = -EEXIST;
1467
1468                 goto err_destroy_ports_array;
1469         }
1470
1471         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1472                                       info->snd_seq, OVS_DP_CMD_NEW);
1473         err = PTR_ERR(reply);
1474         if (IS_ERR(reply))
1475                 goto err_destroy_local_port;
1476
1477         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1478         list_add_tail(&dp->list_node, &ovs_net->dps);
1479         ovs_dp_sysfs_add_dp(dp);
1480
1481         rtnl_unlock();
1482
1483         genl_notify(reply, genl_info_net(info), info->snd_pid,
1484                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1485                     GFP_KERNEL);
1486         return 0;
1487
1488 err_destroy_local_port:
1489         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1490 err_destroy_ports_array:
1491         kfree(dp->ports);
1492 err_destroy_percpu:
1493         free_percpu(dp->stats_percpu);
1494 err_destroy_table:
1495         ovs_flow_tbl_destroy(genl_dereference(dp->table));
1496 err_free_dp:
1497         release_net(ovs_dp_get_net(dp));
1498         kfree(dp);
1499 err_unlock_rtnl:
1500         rtnl_unlock();
1501 err:
1502         return err;
1503 }
1504
1505 /* Called with genl_mutex. */
1506 static void __dp_destroy(struct datapath *dp)
1507 {
1508         int i;
1509
1510         rtnl_lock();
1511
1512         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1513                 struct vport *vport;
1514                 struct hlist_node *node, *n;
1515
1516                 hlist_for_each_entry_safe(vport, node, n, &dp->ports[i], dp_hash_node)
1517                         if (vport->port_no != OVSP_LOCAL)
1518                                 ovs_dp_detach_port(vport);
1519         }
1520
1521         ovs_dp_sysfs_del_dp(dp);
1522         list_del(&dp->list_node);
1523         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1524
1525         /* rtnl_unlock() will wait until all the references to devices that
1526          * are pending unregistration have been dropped.  We do it here to
1527          * ensure that any internal devices (which contain DP pointers) are
1528          * fully destroyed before freeing the datapath.
1529          */
1530         rtnl_unlock();
1531
1532         call_rcu(&dp->rcu, destroy_dp_rcu);
1533 }
1534
1535 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1536 {
1537         struct sk_buff *reply;
1538         struct datapath *dp;
1539         int err;
1540
1541         err = ovs_dp_cmd_validate(info->attrs);
1542         if (err)
1543                 return err;
1544
1545         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1546         err = PTR_ERR(dp);
1547         if (IS_ERR(dp))
1548                 return err;
1549
1550         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1551                                       info->snd_seq, OVS_DP_CMD_DEL);
1552         err = PTR_ERR(reply);
1553         if (IS_ERR(reply))
1554                 return err;
1555
1556         __dp_destroy(dp);
1557
1558         genl_notify(reply, genl_info_net(info), info->snd_pid,
1559                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1560                     GFP_KERNEL);
1561
1562         return 0;
1563 }
1564
1565 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1566 {
1567         struct sk_buff *reply;
1568         struct datapath *dp;
1569         int err;
1570
1571         err = ovs_dp_cmd_validate(info->attrs);
1572         if (err)
1573                 return err;
1574
1575         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1576         if (IS_ERR(dp))
1577                 return PTR_ERR(dp);
1578
1579         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1580                                       info->snd_seq, OVS_DP_CMD_NEW);
1581         if (IS_ERR(reply)) {
1582                 err = PTR_ERR(reply);
1583                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1584                                 ovs_dp_datapath_multicast_group.id, err);
1585                 return 0;
1586         }
1587
1588         genl_notify(reply, genl_info_net(info), info->snd_pid,
1589                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1590                     GFP_KERNEL);
1591
1592         return 0;
1593 }
1594
1595 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1596 {
1597         struct sk_buff *reply;
1598         struct datapath *dp;
1599         int err;
1600
1601         err = ovs_dp_cmd_validate(info->attrs);
1602         if (err)
1603                 return err;
1604
1605         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1606         if (IS_ERR(dp))
1607                 return PTR_ERR(dp);
1608
1609         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1610                                       info->snd_seq, OVS_DP_CMD_NEW);
1611         if (IS_ERR(reply))
1612                 return PTR_ERR(reply);
1613
1614         return genlmsg_reply(reply, info);
1615 }
1616
1617 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1618 {
1619         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1620         struct datapath *dp;
1621         int skip = cb->args[0];
1622         int i = 0;
1623
1624         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1625                 if (i >= skip &&
1626                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1627                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1628                                          OVS_DP_CMD_NEW) < 0)
1629                         break;
1630                 i++;
1631         }
1632
1633         cb->args[0] = i;
1634
1635         return skb->len;
1636 }
1637
1638 static struct genl_ops dp_datapath_genl_ops[] = {
1639         { .cmd = OVS_DP_CMD_NEW,
1640           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1641           .policy = datapath_policy,
1642           .doit = ovs_dp_cmd_new
1643         },
1644         { .cmd = OVS_DP_CMD_DEL,
1645           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1646           .policy = datapath_policy,
1647           .doit = ovs_dp_cmd_del
1648         },
1649         { .cmd = OVS_DP_CMD_GET,
1650           .flags = 0,               /* OK for unprivileged users. */
1651           .policy = datapath_policy,
1652           .doit = ovs_dp_cmd_get,
1653           .dumpit = ovs_dp_cmd_dump
1654         },
1655         { .cmd = OVS_DP_CMD_SET,
1656           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1657           .policy = datapath_policy,
1658           .doit = ovs_dp_cmd_set,
1659         },
1660 };
1661
1662 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1663 #ifdef HAVE_NLA_NUL_STRING
1664         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1665         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1666         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1667 #else
1668         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1669         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1670 #endif
1671         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1672         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1673         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1674         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1675 };
1676
1677 static struct genl_family dp_vport_genl_family = {
1678         .id = GENL_ID_GENERATE,
1679         .hdrsize = sizeof(struct ovs_header),
1680         .name = OVS_VPORT_FAMILY,
1681         .version = OVS_VPORT_VERSION,
1682         .maxattr = OVS_VPORT_ATTR_MAX,
1683          SET_NETNSOK
1684 };
1685
1686 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1687         .name = OVS_VPORT_MCGROUP
1688 };
1689
1690 /* Called with RTNL lock or RCU read lock. */
1691 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1692                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1693 {
1694         struct ovs_header *ovs_header;
1695         struct ovs_vport_stats vport_stats;
1696         int err;
1697
1698         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1699                                  flags, cmd);
1700         if (!ovs_header)
1701                 return -EMSGSIZE;
1702
1703         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1704
1705         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1706             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1707             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1708             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid))
1709                 goto nla_put_failure;
1710
1711         ovs_vport_get_stats(vport, &vport_stats);
1712         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1713                     &vport_stats))
1714                 goto nla_put_failure;
1715
1716         if (nla_put(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN,
1717                     vport->ops->get_addr(vport)))
1718                 goto nla_put_failure;
1719
1720         err = ovs_vport_get_options(vport, skb);
1721         if (err == -EMSGSIZE)
1722                 goto error;
1723
1724         return genlmsg_end(skb, ovs_header);
1725
1726 nla_put_failure:
1727         err = -EMSGSIZE;
1728 error:
1729         genlmsg_cancel(skb, ovs_header);
1730         return err;
1731 }
1732
1733 /* Called with RTNL lock or RCU read lock. */
1734 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1735                                          u32 seq, u8 cmd)
1736 {
1737         struct sk_buff *skb;
1738         int retval;
1739
1740         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1741         if (!skb)
1742                 return ERR_PTR(-ENOMEM);
1743
1744         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1745         if (retval < 0) {
1746                 kfree_skb(skb);
1747                 return ERR_PTR(retval);
1748         }
1749         return skb;
1750 }
1751
1752 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1753 {
1754         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1755 }
1756
1757 /* Called with RTNL lock or RCU read lock. */
1758 static struct vport *lookup_vport(struct net *net,
1759                                   struct ovs_header *ovs_header,
1760                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1761 {
1762         struct datapath *dp;
1763         struct vport *vport;
1764
1765         if (a[OVS_VPORT_ATTR_NAME]) {
1766                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1767                 if (!vport)
1768                         return ERR_PTR(-ENODEV);
1769                 if (ovs_header->dp_ifindex &&
1770                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1771                         return ERR_PTR(-ENODEV);
1772                 return vport;
1773         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1774                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1775
1776                 if (port_no >= DP_MAX_PORTS)
1777                         return ERR_PTR(-EFBIG);
1778
1779                 dp = get_dp(net, ovs_header->dp_ifindex);
1780                 if (!dp)
1781                         return ERR_PTR(-ENODEV);
1782
1783                 vport = ovs_vport_rtnl_rcu(dp, port_no);
1784                 if (!vport)
1785                         return ERR_PTR(-ENOENT);
1786                 return vport;
1787         } else
1788                 return ERR_PTR(-EINVAL);
1789 }
1790
1791 /* Called with RTNL lock. */
1792 static int change_vport(struct vport *vport,
1793                         struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1794 {
1795         int err = 0;
1796
1797         if (a[OVS_VPORT_ATTR_STATS])
1798                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1799
1800         if (a[OVS_VPORT_ATTR_ADDRESS])
1801                 err = ovs_vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1802
1803         return err;
1804 }
1805
1806 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1807 {
1808         struct nlattr **a = info->attrs;
1809         struct ovs_header *ovs_header = info->userhdr;
1810         struct vport_parms parms;
1811         struct sk_buff *reply;
1812         struct vport *vport;
1813         struct datapath *dp;
1814         u32 port_no;
1815         int err;
1816
1817         err = -EINVAL;
1818         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1819             !a[OVS_VPORT_ATTR_UPCALL_PID])
1820                 goto exit;
1821
1822         err = ovs_vport_cmd_validate(a);
1823         if (err)
1824                 goto exit;
1825
1826         rtnl_lock();
1827         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1828         err = -ENODEV;
1829         if (!dp)
1830                 goto exit_unlock;
1831
1832         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1833                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1834
1835                 err = -EFBIG;
1836                 if (port_no >= DP_MAX_PORTS)
1837                         goto exit_unlock;
1838
1839                 vport = ovs_vport_rtnl(dp, port_no);
1840                 err = -EBUSY;
1841                 if (vport)
1842                         goto exit_unlock;
1843         } else {
1844                 for (port_no = 1; ; port_no++) {
1845                         if (port_no >= DP_MAX_PORTS) {
1846                                 err = -EFBIG;
1847                                 goto exit_unlock;
1848                         }
1849                         vport = ovs_vport_rtnl(dp, port_no);
1850                         if (!vport)
1851                                 break;
1852                 }
1853         }
1854
1855         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1856         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1857         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1858         parms.dp = dp;
1859         parms.port_no = port_no;
1860         parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1861
1862         vport = new_vport(&parms);
1863         err = PTR_ERR(vport);
1864         if (IS_ERR(vport))
1865                 goto exit_unlock;
1866
1867         ovs_dp_sysfs_add_if(vport);
1868
1869         err = change_vport(vport, a);
1870         if (!err) {
1871                 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1872                                                  info->snd_seq,
1873                                                  OVS_VPORT_CMD_NEW);
1874                 if (IS_ERR(reply))
1875                         err = PTR_ERR(reply);
1876         }
1877         if (err) {
1878                 ovs_dp_detach_port(vport);
1879                 goto exit_unlock;
1880         }
1881         genl_notify(reply, genl_info_net(info), info->snd_pid,
1882                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1883
1884 exit_unlock:
1885         rtnl_unlock();
1886 exit:
1887         return err;
1888 }
1889
1890 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1891 {
1892         struct nlattr **a = info->attrs;
1893         struct sk_buff *reply;
1894         struct vport *vport;
1895         int err;
1896
1897         err = ovs_vport_cmd_validate(a);
1898         if (err)
1899                 goto exit;
1900
1901         rtnl_lock();
1902         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1903         err = PTR_ERR(vport);
1904         if (IS_ERR(vport))
1905                 goto exit_unlock;
1906
1907         err = 0;
1908         if (a[OVS_VPORT_ATTR_TYPE] &&
1909             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1910                 err = -EINVAL;
1911
1912         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1913                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1914         if (!err)
1915                 err = change_vport(vport, a);
1916         else
1917                 goto exit_unlock;
1918         if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1919                 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1920
1921         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1922                                          OVS_VPORT_CMD_NEW);
1923         if (IS_ERR(reply)) {
1924                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1925                                 ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
1926                 goto exit_unlock;
1927         }
1928
1929         genl_notify(reply, genl_info_net(info), info->snd_pid,
1930                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1931
1932 exit_unlock:
1933         rtnl_unlock();
1934 exit:
1935         return err;
1936 }
1937
1938 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1939 {
1940         struct nlattr **a = info->attrs;
1941         struct sk_buff *reply;
1942         struct vport *vport;
1943         int err;
1944
1945         err = ovs_vport_cmd_validate(a);
1946         if (err)
1947                 goto exit;
1948
1949         rtnl_lock();
1950         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1951         err = PTR_ERR(vport);
1952         if (IS_ERR(vport))
1953                 goto exit_unlock;
1954
1955         if (vport->port_no == OVSP_LOCAL) {
1956                 err = -EINVAL;
1957                 goto exit_unlock;
1958         }
1959
1960         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1961                                          OVS_VPORT_CMD_DEL);
1962         err = PTR_ERR(reply);
1963         if (IS_ERR(reply))
1964                 goto exit_unlock;
1965
1966         ovs_dp_detach_port(vport);
1967
1968         genl_notify(reply, genl_info_net(info), info->snd_pid,
1969                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1970
1971 exit_unlock:
1972         rtnl_unlock();
1973 exit:
1974         return err;
1975 }
1976
1977 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1978 {
1979         struct nlattr **a = info->attrs;
1980         struct ovs_header *ovs_header = info->userhdr;
1981         struct sk_buff *reply;
1982         struct vport *vport;
1983         int err;
1984
1985         err = ovs_vport_cmd_validate(a);
1986         if (err)
1987                 goto exit;
1988
1989         rcu_read_lock();
1990         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1991         err = PTR_ERR(vport);
1992         if (IS_ERR(vport))
1993                 goto exit_unlock;
1994
1995         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1996                                          OVS_VPORT_CMD_NEW);
1997         err = PTR_ERR(reply);
1998         if (IS_ERR(reply))
1999                 goto exit_unlock;
2000
2001         rcu_read_unlock();
2002
2003         return genlmsg_reply(reply, info);
2004
2005 exit_unlock:
2006         rcu_read_unlock();
2007 exit:
2008         return err;
2009 }
2010
2011 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2012 {
2013         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2014         struct datapath *dp;
2015         int bucket = cb->args[0], skip = cb->args[1];
2016         int i, j = 0;
2017
2018         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2019         if (!dp)
2020                 return -ENODEV;
2021
2022         rcu_read_lock();
2023         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2024                 struct vport *vport;
2025                 struct hlist_node *n;
2026
2027                 j = 0;
2028                 hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node) {
2029                         if (j >= skip &&
2030                             ovs_vport_cmd_fill_info(vport, skb,
2031                                                     NETLINK_CB(cb->skb).pid,
2032                                                     cb->nlh->nlmsg_seq,
2033                                                     NLM_F_MULTI,
2034                                                     OVS_VPORT_CMD_NEW) < 0)
2035                                 goto out;
2036
2037                         j++;
2038                 }
2039                 skip = 0;
2040         }
2041 out:
2042         rcu_read_unlock();
2043
2044         cb->args[0] = i;
2045         cb->args[1] = j;
2046
2047         return skb->len;
2048 }
2049
2050 static struct genl_ops dp_vport_genl_ops[] = {
2051         { .cmd = OVS_VPORT_CMD_NEW,
2052           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2053           .policy = vport_policy,
2054           .doit = ovs_vport_cmd_new
2055         },
2056         { .cmd = OVS_VPORT_CMD_DEL,
2057           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2058           .policy = vport_policy,
2059           .doit = ovs_vport_cmd_del
2060         },
2061         { .cmd = OVS_VPORT_CMD_GET,
2062           .flags = 0,               /* OK for unprivileged users. */
2063           .policy = vport_policy,
2064           .doit = ovs_vport_cmd_get,
2065           .dumpit = ovs_vport_cmd_dump
2066         },
2067         { .cmd = OVS_VPORT_CMD_SET,
2068           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2069           .policy = vport_policy,
2070           .doit = ovs_vport_cmd_set,
2071         },
2072 };
2073
2074 struct genl_family_and_ops {
2075         struct genl_family *family;
2076         struct genl_ops *ops;
2077         int n_ops;
2078         struct genl_multicast_group *group;
2079 };
2080
2081 static const struct genl_family_and_ops dp_genl_families[] = {
2082         { &dp_datapath_genl_family,
2083           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2084           &ovs_dp_datapath_multicast_group },
2085         { &dp_vport_genl_family,
2086           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2087           &ovs_dp_vport_multicast_group },
2088         { &dp_flow_genl_family,
2089           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2090           &ovs_dp_flow_multicast_group },
2091         { &dp_packet_genl_family,
2092           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2093           NULL },
2094 };
2095
2096 static void dp_unregister_genl(int n_families)
2097 {
2098         int i;
2099
2100         for (i = 0; i < n_families; i++)
2101                 genl_unregister_family(dp_genl_families[i].family);
2102 }
2103
2104 static int dp_register_genl(void)
2105 {
2106         int n_registered;
2107         int err;
2108         int i;
2109
2110         n_registered = 0;
2111         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2112                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2113
2114                 err = genl_register_family_with_ops(f->family, f->ops,
2115                                                     f->n_ops);
2116                 if (err)
2117                         goto error;
2118                 n_registered++;
2119
2120                 if (f->group) {
2121                         err = genl_register_mc_group(f->family, f->group);
2122                         if (err)
2123                                 goto error;
2124                 }
2125         }
2126
2127         return 0;
2128
2129 error:
2130         dp_unregister_genl(n_registered);
2131         return err;
2132 }
2133
2134 static int __rehash_flow_table(void *dummy)
2135 {
2136         struct datapath *dp;
2137         struct net *net;
2138
2139         rtnl_lock();
2140         for_each_net(net) {
2141                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2142
2143                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2144                         struct flow_table *old_table = genl_dereference(dp->table);
2145                         struct flow_table *new_table;
2146
2147                         new_table = ovs_flow_tbl_rehash(old_table);
2148                         if (!IS_ERR(new_table)) {
2149                                 rcu_assign_pointer(dp->table, new_table);
2150                                 ovs_flow_tbl_deferred_destroy(old_table);
2151                         }
2152                 }
2153         }
2154         rtnl_unlock();
2155         return 0;
2156 }
2157
2158 static void rehash_flow_table(struct work_struct *work)
2159 {
2160         genl_exec(__rehash_flow_table, NULL);
2161         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2162 }
2163
2164 static int dp_destroy_all(void *data)
2165 {
2166         struct datapath *dp, *dp_next;
2167         struct ovs_net *ovs_net = data;
2168
2169         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2170                 __dp_destroy(dp);
2171
2172         return 0;
2173 }
2174
2175 static int __net_init ovs_init_net(struct net *net)
2176 {
2177         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2178
2179         INIT_LIST_HEAD(&ovs_net->dps);
2180         return 0;
2181 }
2182
2183 static void __net_exit ovs_exit_net(struct net *net)
2184 {
2185         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2186
2187         genl_exec(dp_destroy_all, ovs_net);
2188 }
2189
2190 static struct pernet_operations ovs_net_ops = {
2191         .init = ovs_init_net,
2192         .exit = ovs_exit_net,
2193         .id   = &ovs_net_id,
2194         .size = sizeof(struct ovs_net),
2195 };
2196
2197 static int __init dp_init(void)
2198 {
2199         struct sk_buff *dummy_skb;
2200         int err;
2201
2202         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2203
2204         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2205                 VERSION);
2206
2207         err = genl_exec_init();
2208         if (err)
2209                 goto error;
2210
2211         err = ovs_workqueues_init();
2212         if (err)
2213                 goto error_genl_exec;
2214
2215         err = ovs_tnl_init();
2216         if (err)
2217                 goto error_wq;
2218
2219         err = ovs_flow_init();
2220         if (err)
2221                 goto error_tnl_exit;
2222
2223         err = ovs_vport_init();
2224         if (err)
2225                 goto error_flow_exit;
2226
2227         err = register_pernet_device(&ovs_net_ops);
2228         if (err)
2229                 goto error_vport_exit;
2230
2231         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2232         if (err)
2233                 goto error_netns_exit;
2234
2235         err = dp_register_genl();
2236         if (err < 0)
2237                 goto error_unreg_notifier;
2238
2239         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2240
2241         return 0;
2242
2243 error_unreg_notifier:
2244         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2245 error_netns_exit:
2246         unregister_pernet_device(&ovs_net_ops);
2247 error_vport_exit:
2248         ovs_vport_exit();
2249 error_flow_exit:
2250         ovs_flow_exit();
2251 error_tnl_exit:
2252         ovs_tnl_exit();
2253 error_wq:
2254         ovs_workqueues_exit();
2255 error_genl_exec:
2256         genl_exec_exit();
2257 error:
2258         return err;
2259 }
2260
2261 static void dp_cleanup(void)
2262 {
2263         cancel_delayed_work_sync(&rehash_flow_wq);
2264         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2265         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2266         unregister_pernet_device(&ovs_net_ops);
2267         rcu_barrier();
2268         ovs_vport_exit();
2269         ovs_flow_exit();
2270         ovs_tnl_exit();
2271         ovs_workqueues_exit();
2272         genl_exec_exit();
2273 }
2274
2275 module_init(dp_init);
2276 module_exit(dp_cleanup);
2277
2278 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2279 MODULE_LICENSE("GPL");
2280 MODULE_VERSION(VERSION);