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(3,6,0)
  65 #error Kernels before 2.6.18 or after 3.5 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         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
 591         case OVS_KEY_ATTR_PRIORITY:
 592         case OVS_KEY_ATTR_TUN_ID:
 593         case OVS_KEY_ATTR_ETHERNET:
 594                 break;
 595
 596         case OVS_KEY_ATTR_IPV4:
 597                 if (flow_key->eth.type != htons(ETH_P_IP))
 598                         return -EINVAL;
 599
 600                 if (!flow_key->ip.proto)
 601                         return -EINVAL;
 602
 603                 ipv4_key = nla_data(ovs_key);
 604                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 605                         return -EINVAL;
 606
 607                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 608                         return -EINVAL;
 609
 610                 break;
 611
 612         case OVS_KEY_ATTR_TCP:
 613                 if (flow_key->ip.proto != IPPROTO_TCP)
 614                         return -EINVAL;
 615
 616                 return validate_tp_port(flow_key);
 617
 618         case OVS_KEY_ATTR_UDP:
 619                 if (flow_key->ip.proto != IPPROTO_UDP)
 620                         return -EINVAL;
 621
 622                 return validate_tp_port(flow_key);
 623
 624         default:
 625                 return -EINVAL;
 626         }
 627
 628         return 0;
 629 }
 630
 631 static int validate_userspace(const struct nlattr *attr)
 632 {
 633         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 634                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 635                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
 636         };
 637         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 638         int error;
 639
 640         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 641                                  attr, userspace_policy);
 642         if (error)
 643                 return error;
 644
 645         if (!a[OVS_USERSPACE_ATTR_PID] ||
 646             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 647                 return -EINVAL;
 648
 649         return 0;
 650 }
 651
 652 static int validate_actions(const struct nlattr *attr,
 653                                 const struct sw_flow_key *key,  int depth)
 654 {
 655         const struct nlattr *a;
 656         int rem, err;
 657
 658         if (depth >= SAMPLE_ACTION_DEPTH)
 659                 return -EOVERFLOW;
 660
 661         nla_for_each_nested(a, attr, rem) {
 662                 /* Expected argument lengths, (u32)-1 for variable length. */
 663                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 664                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 665                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 666                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 667                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 668                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 669                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 670                 };
 671                 const struct ovs_action_push_vlan *vlan;
 672                 int type = nla_type(a);
 673
 674                 if (type > OVS_ACTION_ATTR_MAX ||
 675                     (action_lens[type] != nla_len(a) &&
 676                      action_lens[type] != (u32)-1))
 677                         return -EINVAL;
 678
 679                 switch (type) {
 680                 case OVS_ACTION_ATTR_UNSPEC:
 681                         return -EINVAL;
 682
 683                 case OVS_ACTION_ATTR_USERSPACE:
 684                         err = validate_userspace(a);
 685                         if (err)
 686                                 return err;
 687                         break;
 688
 689                 case OVS_ACTION_ATTR_OUTPUT:
 690                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 691                                 return -EINVAL;
 692                         break;
 693
 694
 695                 case OVS_ACTION_ATTR_POP_VLAN:
 696                         break;
 697
 698                 case OVS_ACTION_ATTR_PUSH_VLAN:
 699                         vlan = nla_data(a);
 700                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 701                                 return -EINVAL;
 702                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 703                                 return -EINVAL;
 704                         break;
 705
 706                 case OVS_ACTION_ATTR_SET:
 707                         err = validate_set(a, key);
 708                         if (err)
 709                                 return err;
 710                         break;
 711
 712                 case OVS_ACTION_ATTR_SAMPLE:
 713                         err = validate_sample(a, key, depth);
 714                         if (err)
 715                                 return err;
 716                         break;
 717
 718                 default:
 719                         return -EINVAL;
 720                 }
 721         }
 722
 723         if (rem > 0)
 724                 return -EINVAL;
 725
 726         return 0;
 727 }
 728
 729 static void clear_stats(struct sw_flow *flow)
 730 {
 731         flow->used = 0;
 732         flow->tcp_flags = 0;
 733         flow->packet_count = 0;
 734         flow->byte_count = 0;
 735 }
 736
 737 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 738 {
 739         struct ovs_header *ovs_header = info->userhdr;
 740         struct nlattr **a = info->attrs;
 741         struct sw_flow_actions *acts;
 742         struct sk_buff *packet;
 743         struct sw_flow *flow;
 744         struct datapath *dp;
 745         struct ethhdr *eth;
 746         int len;
 747         int err;
 748         int key_len;
 749
 750         err = -EINVAL;
 751         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 752             !a[OVS_PACKET_ATTR_ACTIONS] ||
 753             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 754                 goto err;
 755
 756         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 757         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 758         err = -ENOMEM;
 759         if (!packet)
 760                 goto err;
 761         skb_reserve(packet, NET_IP_ALIGN);
 762
 763         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 764
 765         skb_reset_mac_header(packet);
 766         eth = eth_hdr(packet);
 767
 768         /* Normally, setting the skb 'protocol' field would be handled by a
 769          * call to eth_type_trans(), but it assumes there's a sending
 770          * device, which we may not have. */
 771         if (ntohs(eth->h_proto) >= 1536)
 772                 packet->protocol = eth->h_proto;
 773         else
 774                 packet->protocol = htons(ETH_P_802_2);
 775
 776         /* Build an sw_flow for sending this packet. */
 777         flow = ovs_flow_alloc();
 778         err = PTR_ERR(flow);
 779         if (IS_ERR(flow))
 780                 goto err_kfree_skb;
 781
 782         err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
 783         if (err)
 784                 goto err_flow_put;
 785
 786         err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
 787                                              &flow->key.phy.in_port,
 788                                              &flow->key.phy.tun_id,
 789                                              a[OVS_PACKET_ATTR_KEY]);
 790         if (err)
 791                 goto err_flow_put;
 792
 793         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
 794         if (err)
 795                 goto err_flow_put;
 796
 797         flow->hash = ovs_flow_hash(&flow->key, key_len);
 798
 799         acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 800         err = PTR_ERR(acts);
 801         if (IS_ERR(acts))
 802                 goto err_flow_put;
 803         rcu_assign_pointer(flow->sf_acts, acts);
 804
 805         OVS_CB(packet)->flow = flow;
 806         packet->priority = flow->key.phy.priority;
 807
 808         rcu_read_lock();
 809         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 810         err = -ENODEV;
 811         if (!dp)
 812                 goto err_unlock;
 813
 814         local_bh_disable();
 815         err = ovs_execute_actions(dp, packet);
 816         local_bh_enable();
 817         rcu_read_unlock();
 818
 819         ovs_flow_put(flow);
 820         return err;
 821
 822 err_unlock:
 823         rcu_read_unlock();
 824 err_flow_put:
 825         ovs_flow_put(flow);
 826 err_kfree_skb:
 827         kfree_skb(packet);
 828 err:
 829         return err;
 830 }
 831
 832 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 833         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 834         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 835         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 836 };
 837
 838 static struct genl_ops dp_packet_genl_ops[] = {
 839         { .cmd = OVS_PACKET_CMD_EXECUTE,
 840           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 841           .policy = packet_policy,
 842           .doit = ovs_packet_cmd_execute
 843         }
 844 };
 845
 846 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 847 {
 848         int i;
 849         struct flow_table *table = genl_dereference(dp->table);
 850
 851         stats->n_flows = ovs_flow_tbl_count(table);
 852
 853         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 854         for_each_possible_cpu(i) {
 855                 const struct dp_stats_percpu *percpu_stats;
 856                 struct dp_stats_percpu local_stats;
 857                 unsigned int start;
 858
 859                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 860
 861                 do {
 862                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 863                         local_stats = *percpu_stats;
 864                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 865
 866                 stats->n_hit += local_stats.n_hit;
 867                 stats->n_missed += local_stats.n_missed;
 868                 stats->n_lost += local_stats.n_lost;
 869         }
 870 }
 871
 872 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 873         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 874         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 875         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 876 };
 877
 878 static struct genl_family dp_flow_genl_family = {
 879         .id = GENL_ID_GENERATE,
 880         .hdrsize = sizeof(struct ovs_header),
 881         .name = OVS_FLOW_FAMILY,
 882         .version = OVS_FLOW_VERSION,
 883         .maxattr = OVS_FLOW_ATTR_MAX,
 884          SET_NETNSOK
 885 };
 886
 887 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 888         .name = OVS_FLOW_MCGROUP
 889 };
 890
 891 /* Called with genl_lock. */
 892 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 893                                   struct sk_buff *skb, u32 pid,
 894                                   u32 seq, u32 flags, u8 cmd)
 895 {
 896         const int skb_orig_len = skb->len;
 897         const struct sw_flow_actions *sf_acts;
 898         struct ovs_flow_stats stats;
 899         struct ovs_header *ovs_header;
 900         struct nlattr *nla;
 901         unsigned long used;
 902         u8 tcp_flags;
 903         int err;
 904
 905         sf_acts = rcu_dereference_protected(flow->sf_acts,
 906                                             lockdep_genl_is_held());
 907
 908         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 909         if (!ovs_header)
 910                 return -EMSGSIZE;
 911
 912         ovs_header->dp_ifindex = get_dpifindex(dp);
 913
 914         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 915         if (!nla)
 916                 goto nla_put_failure;
 917         err = ovs_flow_to_nlattrs(&flow->key, skb);
 918         if (err)
 919                 goto error;
 920         nla_nest_end(skb, nla);
 921
 922         spin_lock_bh(&flow->lock);
 923         used = flow->used;
 924         stats.n_packets = flow->packet_count;
 925         stats.n_bytes = flow->byte_count;
 926         tcp_flags = flow->tcp_flags;
 927         spin_unlock_bh(&flow->lock);
 928
 929         if (used &&
 930             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 931                 goto nla_put_failure;
 932
 933         if (stats.n_packets &&
 934             nla_put(skb, OVS_FLOW_ATTR_STATS,
 935                     sizeof(struct ovs_flow_stats), &stats))
 936                 goto nla_put_failure;
 937
 938         if (tcp_flags &&
 939             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
 940                 goto nla_put_failure;
 941
 942         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 943          * this is the first flow to be dumped into 'skb'.  This is unusual for
 944          * Netlink but individual action lists can be longer than
 945          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 946          * The userspace caller can always fetch the actions separately if it
 947          * really wants them.  (Most userspace callers in fact don't care.)
 948          *
 949          * This can only fail for dump operations because the skb is always
 950          * properly sized for single flows.
 951          */
 952         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 953                       sf_acts->actions);
 954         if (err < 0 && skb_orig_len)
 955                 goto error;
 956
 957         return genlmsg_end(skb, ovs_header);
 958
 959 nla_put_failure:
 960         err = -EMSGSIZE;
 961 error:
 962         genlmsg_cancel(skb, ovs_header);
 963         return err;
 964 }
 965
 966 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 967 {
 968         const struct sw_flow_actions *sf_acts;
 969         int len;
 970
 971         sf_acts = rcu_dereference_protected(flow->sf_acts,
 972                                             lockdep_genl_is_held());
 973
 974         /* OVS_FLOW_ATTR_KEY */
 975         len = nla_total_size(FLOW_BUFSIZE);
 976         /* OVS_FLOW_ATTR_ACTIONS */
 977         len += nla_total_size(sf_acts->actions_len);
 978         /* OVS_FLOW_ATTR_STATS */
 979         len += nla_total_size(sizeof(struct ovs_flow_stats));
 980         /* OVS_FLOW_ATTR_TCP_FLAGS */
 981         len += nla_total_size(1);
 982         /* OVS_FLOW_ATTR_USED */
 983         len += nla_total_size(8);
 984
 985         len += NLMSG_ALIGN(sizeof(struct ovs_header));
 986
 987         return genlmsg_new(len, GFP_KERNEL);
 988 }
 989
 990 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
 991                                                struct datapath *dp,
 992                                                u32 pid, u32 seq, u8 cmd)
 993 {
 994         struct sk_buff *skb;
 995         int retval;
 996
 997         skb = ovs_flow_cmd_alloc_info(flow);
 998         if (!skb)
 999                 return ERR_PTR(-ENOMEM);
1000
1001         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
1002         BUG_ON(retval < 0);
1003         return skb;
1004 }
1005
1006 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1007 {
1008         struct nlattr **a = info->attrs;
1009         struct ovs_header *ovs_header = info->userhdr;
1010         struct sw_flow_key key;
1011         struct sw_flow *flow;
1012         struct sk_buff *reply;
1013         struct datapath *dp;
1014         struct flow_table *table;
1015         int error;
1016         int key_len;
1017
1018         /* Extract key. */
1019         error = -EINVAL;
1020         if (!a[OVS_FLOW_ATTR_KEY])
1021                 goto error;
1022         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1023         if (error)
1024                 goto error;
1025
1026         /* Validate actions. */
1027         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1028                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
1029                 if (error)
1030                         goto error;
1031         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1032                 error = -EINVAL;
1033                 goto error;
1034         }
1035
1036         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1037         error = -ENODEV;
1038         if (!dp)
1039                 goto error;
1040
1041         table = genl_dereference(dp->table);
1042         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1043         if (!flow) {
1044                 struct sw_flow_actions *acts;
1045
1046                 /* Bail out if we're not allowed to create a new flow. */
1047                 error = -ENOENT;
1048                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1049                         goto error;
1050
1051                 /* Expand table, if necessary, to make room. */
1052                 if (ovs_flow_tbl_need_to_expand(table)) {
1053                         struct flow_table *new_table;
1054
1055                         new_table = ovs_flow_tbl_expand(table);
1056                         if (!IS_ERR(new_table)) {
1057                                 rcu_assign_pointer(dp->table, new_table);
1058                                 ovs_flow_tbl_deferred_destroy(table);
1059                                 table = genl_dereference(dp->table);
1060                         }
1061                 }
1062
1063                 /* Allocate flow. */
1064                 flow = ovs_flow_alloc();
1065                 if (IS_ERR(flow)) {
1066                         error = PTR_ERR(flow);
1067                         goto error;
1068                 }
1069                 flow->key = key;
1070                 clear_stats(flow);
1071
1072                 /* Obtain actions. */
1073                 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1074                 error = PTR_ERR(acts);
1075                 if (IS_ERR(acts))
1076                         goto error_free_flow;
1077                 rcu_assign_pointer(flow->sf_acts, acts);
1078
1079                 /* Put flow in bucket. */
1080                 flow->hash = ovs_flow_hash(&key, key_len);
1081                 ovs_flow_tbl_insert(table, flow);
1082
1083                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1084                                                 info->snd_seq,
1085                                                 OVS_FLOW_CMD_NEW);
1086         } else {
1087                 /* We found a matching flow. */
1088                 struct sw_flow_actions *old_acts;
1089                 struct nlattr *acts_attrs;
1090
1091                 /* Bail out if we're not allowed to modify an existing flow.
1092                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1093                  * because Generic Netlink treats the latter as a dump
1094                  * request.  We also accept NLM_F_EXCL in case that bug ever
1095                  * gets fixed.
1096                  */
1097                 error = -EEXIST;
1098                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1099                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1100                         goto error;
1101
1102                 /* Update actions. */
1103                 old_acts = rcu_dereference_protected(flow->sf_acts,
1104                                                      lockdep_genl_is_held());
1105                 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1106                 if (acts_attrs &&
1107                    (old_acts->actions_len != nla_len(acts_attrs) ||
1108                    memcmp(old_acts->actions, nla_data(acts_attrs),
1109                           old_acts->actions_len))) {
1110                         struct sw_flow_actions *new_acts;
1111
1112                         new_acts = ovs_flow_actions_alloc(acts_attrs);
1113                         error = PTR_ERR(new_acts);
1114                         if (IS_ERR(new_acts))
1115                                 goto error;
1116
1117                         rcu_assign_pointer(flow->sf_acts, new_acts);
1118                         ovs_flow_deferred_free_acts(old_acts);
1119                 }
1120
1121                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1122                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1123
1124                 /* Clear stats. */
1125                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1126                         spin_lock_bh(&flow->lock);
1127                         clear_stats(flow);
1128                         spin_unlock_bh(&flow->lock);
1129                 }
1130         }
1131
1132         if (!IS_ERR(reply))
1133                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1134                            ovs_dp_flow_multicast_group.id, info->nlhdr,
1135                            GFP_KERNEL);
1136         else
1137                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1138                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1139         return 0;
1140
1141 error_free_flow:
1142         ovs_flow_put(flow);
1143 error:
1144         return error;
1145 }
1146
1147 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1148 {
1149         struct nlattr **a = info->attrs;
1150         struct ovs_header *ovs_header = info->userhdr;
1151         struct sw_flow_key key;
1152         struct sk_buff *reply;
1153         struct sw_flow *flow;
1154         struct datapath *dp;
1155         struct flow_table *table;
1156         int err;
1157         int key_len;
1158
1159         if (!a[OVS_FLOW_ATTR_KEY])
1160                 return -EINVAL;
1161         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1162         if (err)
1163                 return err;
1164
1165         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1166         if (!dp)
1167                 return -ENODEV;
1168
1169         table = genl_dereference(dp->table);
1170         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1171         if (!flow)
1172                 return -ENOENT;
1173
1174         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1175                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1176         if (IS_ERR(reply))
1177                 return PTR_ERR(reply);
1178
1179         return genlmsg_reply(reply, info);
1180 }
1181
1182 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1183 {
1184         struct nlattr **a = info->attrs;
1185         struct ovs_header *ovs_header = info->userhdr;
1186         struct sw_flow_key key;
1187         struct sk_buff *reply;
1188         struct sw_flow *flow;
1189         struct datapath *dp;
1190         struct flow_table *table;
1191         int err;
1192         int key_len;
1193
1194         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1195         if (!dp)
1196                 return -ENODEV;
1197
1198         if (!a[OVS_FLOW_ATTR_KEY])
1199                 return flush_flows(dp);
1200
1201         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1202         if (err)
1203                 return err;
1204
1205         table = genl_dereference(dp->table);
1206         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1207         if (!flow)
1208                 return -ENOENT;
1209
1210         reply = ovs_flow_cmd_alloc_info(flow);
1211         if (!reply)
1212                 return -ENOMEM;
1213
1214         ovs_flow_tbl_remove(table, flow);
1215
1216         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1217                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1218         BUG_ON(err < 0);
1219
1220         ovs_flow_deferred_free(flow);
1221
1222         genl_notify(reply, genl_info_net(info), info->snd_pid,
1223                     ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1224         return 0;
1225 }
1226
1227 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1228 {
1229         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1230         struct datapath *dp;
1231         struct flow_table *table;
1232
1233         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1234         if (!dp)
1235                 return -ENODEV;
1236
1237         table = genl_dereference(dp->table);
1238
1239         for (;;) {
1240                 struct sw_flow *flow;
1241                 u32 bucket, obj;
1242
1243                 bucket = cb->args[0];
1244                 obj = cb->args[1];
1245                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1246                 if (!flow)
1247                         break;
1248
1249                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1250                                            NETLINK_CB(cb->skb).pid,
1251                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1252                                            OVS_FLOW_CMD_NEW) < 0)
1253                         break;
1254
1255                 cb->args[0] = bucket;
1256                 cb->args[1] = obj;
1257         }
1258         return skb->len;
1259 }
1260
1261 static struct genl_ops dp_flow_genl_ops[] = {
1262         { .cmd = OVS_FLOW_CMD_NEW,
1263           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1264           .policy = flow_policy,
1265           .doit = ovs_flow_cmd_new_or_set
1266         },
1267         { .cmd = OVS_FLOW_CMD_DEL,
1268           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1269           .policy = flow_policy,
1270           .doit = ovs_flow_cmd_del
1271         },
1272         { .cmd = OVS_FLOW_CMD_GET,
1273           .flags = 0,               /* OK for unprivileged users. */
1274           .policy = flow_policy,
1275           .doit = ovs_flow_cmd_get,
1276           .dumpit = ovs_flow_cmd_dump
1277         },
1278         { .cmd = OVS_FLOW_CMD_SET,
1279           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1280           .policy = flow_policy,
1281           .doit = ovs_flow_cmd_new_or_set,
1282         },
1283 };
1284
1285 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1286 #ifdef HAVE_NLA_NUL_STRING
1287         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1288 #endif
1289         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1290 };
1291
1292 static struct genl_family dp_datapath_genl_family = {
1293         .id = GENL_ID_GENERATE,
1294         .hdrsize = sizeof(struct ovs_header),
1295         .name = OVS_DATAPATH_FAMILY,
1296         .version = OVS_DATAPATH_VERSION,
1297         .maxattr = OVS_DP_ATTR_MAX,
1298          SET_NETNSOK
1299 };
1300
1301 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1302         .name = OVS_DATAPATH_MCGROUP
1303 };
1304
1305 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1306                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1307 {
1308         struct ovs_header *ovs_header;
1309         struct ovs_dp_stats dp_stats;
1310         int err;
1311
1312         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1313                                    flags, cmd);
1314         if (!ovs_header)
1315                 goto error;
1316
1317         ovs_header->dp_ifindex = get_dpifindex(dp);
1318
1319         rcu_read_lock();
1320         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1321         rcu_read_unlock();
1322         if (err)
1323                 goto nla_put_failure;
1324
1325         get_dp_stats(dp, &dp_stats);
1326         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1327                 goto nla_put_failure;
1328
1329         return genlmsg_end(skb, ovs_header);
1330
1331 nla_put_failure:
1332         genlmsg_cancel(skb, ovs_header);
1333 error:
1334         return -EMSGSIZE;
1335 }
1336
1337 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1338                                              u32 seq, u8 cmd)
1339 {
1340         struct sk_buff *skb;
1341         int retval;
1342
1343         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1344         if (!skb)
1345                 return ERR_PTR(-ENOMEM);
1346
1347         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1348         if (retval < 0) {
1349                 kfree_skb(skb);
1350                 return ERR_PTR(retval);
1351         }
1352         return skb;
1353 }
1354
1355 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1356 {
1357         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1358 }
1359
1360 /* Called with genl_mutex and optionally with RTNL lock also. */
1361 static struct datapath *lookup_datapath(struct net *net,
1362                                         struct ovs_header *ovs_header,
1363                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1364 {
1365         struct datapath *dp;
1366
1367         if (!a[OVS_DP_ATTR_NAME])
1368                 dp = get_dp(net, ovs_header->dp_ifindex);
1369         else {
1370                 struct vport *vport;
1371
1372                 rcu_read_lock();
1373                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1374                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1375                 rcu_read_unlock();
1376         }
1377         return dp ? dp : ERR_PTR(-ENODEV);
1378 }
1379
1380 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1381 {
1382         struct nlattr **a = info->attrs;
1383         struct vport_parms parms;
1384         struct sk_buff *reply;
1385         struct datapath *dp;
1386         struct vport *vport;
1387         struct ovs_net *ovs_net;
1388         int err, i;
1389
1390         err = -EINVAL;
1391         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1392                 goto err;
1393
1394         err = ovs_dp_cmd_validate(a);
1395         if (err)
1396                 goto err;
1397
1398         rtnl_lock();
1399
1400         err = -ENOMEM;
1401         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1402         if (dp == NULL)
1403                 goto err_unlock_rtnl;
1404
1405         /* Initialize kobject for bridge.  This will be added as
1406          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1407         dp->ifobj.kset = NULL;
1408         kobject_init(&dp->ifobj, &dp_ktype);
1409
1410         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1411
1412         /* Allocate table. */
1413         err = -ENOMEM;
1414         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1415         if (!dp->table)
1416                 goto err_free_dp;
1417
1418         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1419         if (!dp->stats_percpu) {
1420                 err = -ENOMEM;
1421                 goto err_destroy_table;
1422         }
1423
1424         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1425                             GFP_KERNEL);
1426         if (!dp->ports) {
1427                 err = -ENOMEM;
1428                 goto err_destroy_percpu;
1429         }
1430
1431         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1432                 INIT_HLIST_HEAD(&dp->ports[i]);
1433
1434         /* Set up our datapath device. */
1435         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1436         parms.type = OVS_VPORT_TYPE_INTERNAL;
1437         parms.options = NULL;
1438         parms.dp = dp;
1439         parms.port_no = OVSP_LOCAL;
1440         parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1441
1442         vport = new_vport(&parms);
1443         if (IS_ERR(vport)) {
1444                 err = PTR_ERR(vport);
1445                 if (err == -EBUSY)
1446                         err = -EEXIST;
1447
1448                 goto err_destroy_ports_array;
1449         }
1450
1451         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1452                                       info->snd_seq, OVS_DP_CMD_NEW);
1453         err = PTR_ERR(reply);
1454         if (IS_ERR(reply))
1455                 goto err_destroy_local_port;
1456
1457         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1458         list_add_tail(&dp->list_node, &ovs_net->dps);
1459         ovs_dp_sysfs_add_dp(dp);
1460
1461         rtnl_unlock();
1462
1463         genl_notify(reply, genl_info_net(info), info->snd_pid,
1464                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1465                     GFP_KERNEL);
1466         return 0;
1467
1468 err_destroy_local_port:
1469         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1470 err_destroy_ports_array:
1471         kfree(dp->ports);
1472 err_destroy_percpu:
1473         free_percpu(dp->stats_percpu);
1474 err_destroy_table:
1475         ovs_flow_tbl_destroy(genl_dereference(dp->table));
1476 err_free_dp:
1477         release_net(ovs_dp_get_net(dp));
1478         kfree(dp);
1479 err_unlock_rtnl:
1480         rtnl_unlock();
1481 err:
1482         return err;
1483 }
1484
1485 /* Called with genl_mutex. */
1486 static void __dp_destroy(struct datapath *dp)
1487 {
1488         int i;
1489
1490         rtnl_lock();
1491
1492         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1493                 struct vport *vport;
1494                 struct hlist_node *node, *n;
1495
1496                 hlist_for_each_entry_safe(vport, node, n, &dp->ports[i], dp_hash_node)
1497                         if (vport->port_no != OVSP_LOCAL)
1498                                 ovs_dp_detach_port(vport);
1499         }
1500
1501         ovs_dp_sysfs_del_dp(dp);
1502         list_del(&dp->list_node);
1503         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1504
1505         /* rtnl_unlock() will wait until all the references to devices that
1506          * are pending unregistration have been dropped.  We do it here to
1507          * ensure that any internal devices (which contain DP pointers) are
1508          * fully destroyed before freeing the datapath.
1509          */
1510         rtnl_unlock();
1511
1512         call_rcu(&dp->rcu, destroy_dp_rcu);
1513 }
1514
1515 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1516 {
1517         struct sk_buff *reply;
1518         struct datapath *dp;
1519         int err;
1520
1521         err = ovs_dp_cmd_validate(info->attrs);
1522         if (err)
1523                 return err;
1524
1525         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1526         err = PTR_ERR(dp);
1527         if (IS_ERR(dp))
1528                 return err;
1529
1530         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1531                                       info->snd_seq, OVS_DP_CMD_DEL);
1532         err = PTR_ERR(reply);
1533         if (IS_ERR(reply))
1534                 return err;
1535
1536         __dp_destroy(dp);
1537
1538         genl_notify(reply, genl_info_net(info), info->snd_pid,
1539                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1540                     GFP_KERNEL);
1541
1542         return 0;
1543 }
1544
1545 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1546 {
1547         struct sk_buff *reply;
1548         struct datapath *dp;
1549         int err;
1550
1551         err = ovs_dp_cmd_validate(info->attrs);
1552         if (err)
1553                 return err;
1554
1555         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1556         if (IS_ERR(dp))
1557                 return PTR_ERR(dp);
1558
1559         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1560                                       info->snd_seq, OVS_DP_CMD_NEW);
1561         if (IS_ERR(reply)) {
1562                 err = PTR_ERR(reply);
1563                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1564                                 ovs_dp_datapath_multicast_group.id, err);
1565                 return 0;
1566         }
1567
1568         genl_notify(reply, genl_info_net(info), info->snd_pid,
1569                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1570                     GFP_KERNEL);
1571
1572         return 0;
1573 }
1574
1575 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1576 {
1577         struct sk_buff *reply;
1578         struct datapath *dp;
1579         int err;
1580
1581         err = ovs_dp_cmd_validate(info->attrs);
1582         if (err)
1583                 return err;
1584
1585         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1586         if (IS_ERR(dp))
1587                 return PTR_ERR(dp);
1588
1589         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1590                                       info->snd_seq, OVS_DP_CMD_NEW);
1591         if (IS_ERR(reply))
1592                 return PTR_ERR(reply);
1593
1594         return genlmsg_reply(reply, info);
1595 }
1596
1597 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1598 {
1599         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1600         struct datapath *dp;
1601         int skip = cb->args[0];
1602         int i = 0;
1603
1604         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1605                 if (i >= skip &&
1606                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1607                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1608                                          OVS_DP_CMD_NEW) < 0)
1609                         break;
1610                 i++;
1611         }
1612
1613         cb->args[0] = i;
1614
1615         return skb->len;
1616 }
1617
1618 static struct genl_ops dp_datapath_genl_ops[] = {
1619         { .cmd = OVS_DP_CMD_NEW,
1620           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1621           .policy = datapath_policy,
1622           .doit = ovs_dp_cmd_new
1623         },
1624         { .cmd = OVS_DP_CMD_DEL,
1625           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1626           .policy = datapath_policy,
1627           .doit = ovs_dp_cmd_del
1628         },
1629         { .cmd = OVS_DP_CMD_GET,
1630           .flags = 0,               /* OK for unprivileged users. */
1631           .policy = datapath_policy,
1632           .doit = ovs_dp_cmd_get,
1633           .dumpit = ovs_dp_cmd_dump
1634         },
1635         { .cmd = OVS_DP_CMD_SET,
1636           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1637           .policy = datapath_policy,
1638           .doit = ovs_dp_cmd_set,
1639         },
1640 };
1641
1642 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1643 #ifdef HAVE_NLA_NUL_STRING
1644         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1645         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1646         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1647 #else
1648         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1649         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1650 #endif
1651         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1652         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1653         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1654         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1655 };
1656
1657 static struct genl_family dp_vport_genl_family = {
1658         .id = GENL_ID_GENERATE,
1659         .hdrsize = sizeof(struct ovs_header),
1660         .name = OVS_VPORT_FAMILY,
1661         .version = OVS_VPORT_VERSION,
1662         .maxattr = OVS_VPORT_ATTR_MAX,
1663          SET_NETNSOK
1664 };
1665
1666 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1667         .name = OVS_VPORT_MCGROUP
1668 };
1669
1670 /* Called with RTNL lock or RCU read lock. */
1671 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1672                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1673 {
1674         struct ovs_header *ovs_header;
1675         struct ovs_vport_stats vport_stats;
1676         int err;
1677
1678         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1679                                  flags, cmd);
1680         if (!ovs_header)
1681                 return -EMSGSIZE;
1682
1683         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1684
1685         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1686             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1687             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1688             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid))
1689                 goto nla_put_failure;
1690
1691         ovs_vport_get_stats(vport, &vport_stats);
1692         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1693                     &vport_stats))
1694                 goto nla_put_failure;
1695
1696         if (nla_put(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN,
1697                     vport->ops->get_addr(vport)))
1698                 goto nla_put_failure;
1699
1700         err = ovs_vport_get_options(vport, skb);
1701         if (err == -EMSGSIZE)
1702                 goto error;
1703
1704         return genlmsg_end(skb, ovs_header);
1705
1706 nla_put_failure:
1707         err = -EMSGSIZE;
1708 error:
1709         genlmsg_cancel(skb, ovs_header);
1710         return err;
1711 }
1712
1713 /* Called with RTNL lock or RCU read lock. */
1714 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1715                                          u32 seq, u8 cmd)
1716 {
1717         struct sk_buff *skb;
1718         int retval;
1719
1720         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1721         if (!skb)
1722                 return ERR_PTR(-ENOMEM);
1723
1724         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1725         if (retval < 0) {
1726                 kfree_skb(skb);
1727                 return ERR_PTR(retval);
1728         }
1729         return skb;
1730 }
1731
1732 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1733 {
1734         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1735 }
1736
1737 /* Called with RTNL lock or RCU read lock. */
1738 static struct vport *lookup_vport(struct net *net,
1739                                   struct ovs_header *ovs_header,
1740                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1741 {
1742         struct datapath *dp;
1743         struct vport *vport;
1744
1745         if (a[OVS_VPORT_ATTR_NAME]) {
1746                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1747                 if (!vport)
1748                         return ERR_PTR(-ENODEV);
1749                 if (ovs_header->dp_ifindex &&
1750                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1751                         return ERR_PTR(-ENODEV);
1752                 return vport;
1753         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1754                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1755
1756                 if (port_no >= DP_MAX_PORTS)
1757                         return ERR_PTR(-EFBIG);
1758
1759                 dp = get_dp(net, ovs_header->dp_ifindex);
1760                 if (!dp)
1761                         return ERR_PTR(-ENODEV);
1762
1763                 vport = ovs_vport_rtnl_rcu(dp, port_no);
1764                 if (!vport)
1765                         return ERR_PTR(-ENOENT);
1766                 return vport;
1767         } else
1768                 return ERR_PTR(-EINVAL);
1769 }
1770
1771 /* Called with RTNL lock. */
1772 static int change_vport(struct vport *vport,
1773                         struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1774 {
1775         int err = 0;
1776
1777         if (a[OVS_VPORT_ATTR_STATS])
1778                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1779
1780         if (a[OVS_VPORT_ATTR_ADDRESS])
1781                 err = ovs_vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1782
1783         return err;
1784 }
1785
1786 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1787 {
1788         struct nlattr **a = info->attrs;
1789         struct ovs_header *ovs_header = info->userhdr;
1790         struct vport_parms parms;
1791         struct sk_buff *reply;
1792         struct vport *vport;
1793         struct datapath *dp;
1794         u32 port_no;
1795         int err;
1796
1797         err = -EINVAL;
1798         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1799             !a[OVS_VPORT_ATTR_UPCALL_PID])
1800                 goto exit;
1801
1802         err = ovs_vport_cmd_validate(a);
1803         if (err)
1804                 goto exit;
1805
1806         rtnl_lock();
1807         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1808         err = -ENODEV;
1809         if (!dp)
1810                 goto exit_unlock;
1811
1812         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1813                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1814
1815                 err = -EFBIG;
1816                 if (port_no >= DP_MAX_PORTS)
1817                         goto exit_unlock;
1818
1819                 vport = ovs_vport_rtnl(dp, port_no);
1820                 err = -EBUSY;
1821                 if (vport)
1822                         goto exit_unlock;
1823         } else {
1824                 for (port_no = 1; ; port_no++) {
1825                         if (port_no >= DP_MAX_PORTS) {
1826                                 err = -EFBIG;
1827                                 goto exit_unlock;
1828                         }
1829                         vport = ovs_vport_rtnl(dp, port_no);
1830                         if (!vport)
1831                                 break;
1832                 }
1833         }
1834
1835         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1836         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1837         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1838         parms.dp = dp;
1839         parms.port_no = port_no;
1840         parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1841
1842         vport = new_vport(&parms);
1843         err = PTR_ERR(vport);
1844         if (IS_ERR(vport))
1845                 goto exit_unlock;
1846
1847         ovs_dp_sysfs_add_if(vport);
1848
1849         err = change_vport(vport, a);
1850         if (!err) {
1851                 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1852                                                  info->snd_seq,
1853                                                  OVS_VPORT_CMD_NEW);
1854                 if (IS_ERR(reply))
1855                         err = PTR_ERR(reply);
1856         }
1857         if (err) {
1858                 ovs_dp_detach_port(vport);
1859                 goto exit_unlock;
1860         }
1861         genl_notify(reply, genl_info_net(info), info->snd_pid,
1862                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1863
1864 exit_unlock:
1865         rtnl_unlock();
1866 exit:
1867         return err;
1868 }
1869
1870 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1871 {
1872         struct nlattr **a = info->attrs;
1873         struct sk_buff *reply;
1874         struct vport *vport;
1875         int err;
1876
1877         err = ovs_vport_cmd_validate(a);
1878         if (err)
1879                 goto exit;
1880
1881         rtnl_lock();
1882         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1883         err = PTR_ERR(vport);
1884         if (IS_ERR(vport))
1885                 goto exit_unlock;
1886
1887         err = 0;
1888         if (a[OVS_VPORT_ATTR_TYPE] &&
1889             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1890                 err = -EINVAL;
1891
1892         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1893                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1894         if (!err)
1895                 err = change_vport(vport, a);
1896         else
1897                 goto exit_unlock;
1898         if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1899                 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1900
1901         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1902                                          OVS_VPORT_CMD_NEW);
1903         if (IS_ERR(reply)) {
1904                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1905                                 ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
1906                 goto exit_unlock;
1907         }
1908
1909         genl_notify(reply, genl_info_net(info), info->snd_pid,
1910                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1911
1912 exit_unlock:
1913         rtnl_unlock();
1914 exit:
1915         return err;
1916 }
1917
1918 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1919 {
1920         struct nlattr **a = info->attrs;
1921         struct sk_buff *reply;
1922         struct vport *vport;
1923         int err;
1924
1925         err = ovs_vport_cmd_validate(a);
1926         if (err)
1927                 goto exit;
1928
1929         rtnl_lock();
1930         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1931         err = PTR_ERR(vport);
1932         if (IS_ERR(vport))
1933                 goto exit_unlock;
1934
1935         if (vport->port_no == OVSP_LOCAL) {
1936                 err = -EINVAL;
1937                 goto exit_unlock;
1938         }
1939
1940         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1941                                          OVS_VPORT_CMD_DEL);
1942         err = PTR_ERR(reply);
1943         if (IS_ERR(reply))
1944                 goto exit_unlock;
1945
1946         ovs_dp_detach_port(vport);
1947
1948         genl_notify(reply, genl_info_net(info), info->snd_pid,
1949                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1950
1951 exit_unlock:
1952         rtnl_unlock();
1953 exit:
1954         return err;
1955 }
1956
1957 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1958 {
1959         struct nlattr **a = info->attrs;
1960         struct ovs_header *ovs_header = info->userhdr;
1961         struct sk_buff *reply;
1962         struct vport *vport;
1963         int err;
1964
1965         err = ovs_vport_cmd_validate(a);
1966         if (err)
1967                 goto exit;
1968
1969         rcu_read_lock();
1970         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1971         err = PTR_ERR(vport);
1972         if (IS_ERR(vport))
1973                 goto exit_unlock;
1974
1975         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1976                                          OVS_VPORT_CMD_NEW);
1977         err = PTR_ERR(reply);
1978         if (IS_ERR(reply))
1979                 goto exit_unlock;
1980
1981         rcu_read_unlock();
1982
1983         return genlmsg_reply(reply, info);
1984
1985 exit_unlock:
1986         rcu_read_unlock();
1987 exit:
1988         return err;
1989 }
1990
1991 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1992 {
1993         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1994         struct datapath *dp;
1995         int bucket = cb->args[0], skip = cb->args[1];
1996         int i, j = 0;
1997
1998         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1999         if (!dp)
2000                 return -ENODEV;
2001
2002         rcu_read_lock();
2003         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2004                 struct vport *vport;
2005                 struct hlist_node *n;
2006
2007                 j = 0;
2008                 hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node) {
2009                         if (j >= skip &&
2010                             ovs_vport_cmd_fill_info(vport, skb,
2011                                                     NETLINK_CB(cb->skb).pid,
2012                                                     cb->nlh->nlmsg_seq,
2013                                                     NLM_F_MULTI,
2014                                                     OVS_VPORT_CMD_NEW) < 0)
2015                                 goto out;
2016
2017                         j++;
2018                 }
2019                 skip = 0;
2020         }
2021 out:
2022         rcu_read_unlock();
2023
2024         cb->args[0] = i;
2025         cb->args[1] = j;
2026
2027         return skb->len;
2028 }
2029
2030 static struct genl_ops dp_vport_genl_ops[] = {
2031         { .cmd = OVS_VPORT_CMD_NEW,
2032           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2033           .policy = vport_policy,
2034           .doit = ovs_vport_cmd_new
2035         },
2036         { .cmd = OVS_VPORT_CMD_DEL,
2037           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2038           .policy = vport_policy,
2039           .doit = ovs_vport_cmd_del
2040         },
2041         { .cmd = OVS_VPORT_CMD_GET,
2042           .flags = 0,               /* OK for unprivileged users. */
2043           .policy = vport_policy,
2044           .doit = ovs_vport_cmd_get,
2045           .dumpit = ovs_vport_cmd_dump
2046         },
2047         { .cmd = OVS_VPORT_CMD_SET,
2048           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2049           .policy = vport_policy,
2050           .doit = ovs_vport_cmd_set,
2051         },
2052 };
2053
2054 struct genl_family_and_ops {
2055         struct genl_family *family;
2056         struct genl_ops *ops;
2057         int n_ops;
2058         struct genl_multicast_group *group;
2059 };
2060
2061 static const struct genl_family_and_ops dp_genl_families[] = {
2062         { &dp_datapath_genl_family,
2063           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2064           &ovs_dp_datapath_multicast_group },
2065         { &dp_vport_genl_family,
2066           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2067           &ovs_dp_vport_multicast_group },
2068         { &dp_flow_genl_family,
2069           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2070           &ovs_dp_flow_multicast_group },
2071         { &dp_packet_genl_family,
2072           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2073           NULL },
2074 };
2075
2076 static void dp_unregister_genl(int n_families)
2077 {
2078         int i;
2079
2080         for (i = 0; i < n_families; i++)
2081                 genl_unregister_family(dp_genl_families[i].family);
2082 }
2083
2084 static int dp_register_genl(void)
2085 {
2086         int n_registered;
2087         int err;
2088         int i;
2089
2090         n_registered = 0;
2091         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2092                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2093
2094                 err = genl_register_family_with_ops(f->family, f->ops,
2095                                                     f->n_ops);
2096                 if (err)
2097                         goto error;
2098                 n_registered++;
2099
2100                 if (f->group) {
2101                         err = genl_register_mc_group(f->family, f->group);
2102                         if (err)
2103                                 goto error;
2104                 }
2105         }
2106
2107         return 0;
2108
2109 error:
2110         dp_unregister_genl(n_registered);
2111         return err;
2112 }
2113
2114 static int __rehash_flow_table(void *dummy)
2115 {
2116         struct datapath *dp;
2117         struct net *net;
2118
2119         rtnl_lock();
2120         for_each_net(net) {
2121                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2122
2123                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2124                         struct flow_table *old_table = genl_dereference(dp->table);
2125                         struct flow_table *new_table;
2126
2127                         new_table = ovs_flow_tbl_rehash(old_table);
2128                         if (!IS_ERR(new_table)) {
2129                                 rcu_assign_pointer(dp->table, new_table);
2130                                 ovs_flow_tbl_deferred_destroy(old_table);
2131                         }
2132                 }
2133         }
2134         rtnl_unlock();
2135         return 0;
2136 }
2137
2138 static void rehash_flow_table(struct work_struct *work)
2139 {
2140         genl_exec(__rehash_flow_table, NULL);
2141         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2142 }
2143
2144 static int dp_destroy_all(void *data)
2145 {
2146         struct datapath *dp, *dp_next;
2147         struct ovs_net *ovs_net = data;
2148
2149         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2150                 __dp_destroy(dp);
2151
2152         return 0;
2153 }
2154
2155 static int __net_init ovs_init_net(struct net *net)
2156 {
2157         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2158
2159         INIT_LIST_HEAD(&ovs_net->dps);
2160         return 0;
2161 }
2162
2163 static void __net_exit ovs_exit_net(struct net *net)
2164 {
2165         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2166
2167         genl_exec(dp_destroy_all, ovs_net);
2168 }
2169
2170 static struct pernet_operations ovs_net_ops = {
2171         .init = ovs_init_net,
2172         .exit = ovs_exit_net,
2173         .id   = &ovs_net_id,
2174         .size = sizeof(struct ovs_net),
2175 };
2176
2177 static int __init dp_init(void)
2178 {
2179         struct sk_buff *dummy_skb;
2180         int err;
2181
2182         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2183
2184         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2185                 VERSION);
2186
2187         err = genl_exec_init();
2188         if (err)
2189                 goto error;
2190
2191         err = ovs_workqueues_init();
2192         if (err)
2193                 goto error_genl_exec;
2194
2195         err = ovs_tnl_init();
2196         if (err)
2197                 goto error_wq;
2198
2199         err = ovs_flow_init();
2200         if (err)
2201                 goto error_tnl_exit;
2202
2203         err = ovs_vport_init();
2204         if (err)
2205                 goto error_flow_exit;
2206
2207         err = register_pernet_device(&ovs_net_ops);
2208         if (err)
2209                 goto error_vport_exit;
2210
2211         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2212         if (err)
2213                 goto error_netns_exit;
2214
2215         err = dp_register_genl();
2216         if (err < 0)
2217                 goto error_unreg_notifier;
2218
2219         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2220
2221         return 0;
2222
2223 error_unreg_notifier:
2224         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2225 error_netns_exit:
2226         unregister_pernet_device(&ovs_net_ops);
2227 error_vport_exit:
2228         ovs_vport_exit();
2229 error_flow_exit:
2230         ovs_flow_exit();
2231 error_tnl_exit:
2232         ovs_tnl_exit();
2233 error_wq:
2234         ovs_workqueues_exit();
2235 error_genl_exec:
2236         genl_exec_exit();
2237 error:
2238         return err;
2239 }
2240
2241 static void dp_cleanup(void)
2242 {
2243         cancel_delayed_work_sync(&rehash_flow_wq);
2244         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2245         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2246         unregister_pernet_device(&ovs_net_ops);
2247         rcu_barrier();
2248         ovs_vport_exit();
2249         ovs_flow_exit();
2250         ovs_tnl_exit();
2251         ovs_workqueues_exit();
2252         genl_exec_exit();
2253 }
2254
2255 module_init(dp_init);
2256 module_exit(dp_cleanup);
2257
2258 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2259 MODULE_LICENSE("GPL");