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_SKB_MARK:
 599 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
 600                 if (nla_get_u32(ovs_key) != 0)
 601                         return -EINVAL;
 602 #endif
 603                 break;
 604
 605         case OVS_KEY_ATTR_IPV4_TUNNEL:
 606                 tun_key = nla_data(ovs_key);
 607                 if (!tun_key->ipv4_dst)
 608                         return -EINVAL;
 609                 break;
 610
 611         case OVS_KEY_ATTR_IPV4:
 612                 if (flow_key->eth.type != htons(ETH_P_IP))
 613                         return -EINVAL;
 614
 615                 if (!flow_key->ip.proto)
 616                         return -EINVAL;
 617
 618                 ipv4_key = nla_data(ovs_key);
 619                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 620                         return -EINVAL;
 621
 622                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 623                         return -EINVAL;
 624
 625                 break;
 626
 627         case OVS_KEY_ATTR_IPV6:
 628                 if (flow_key->eth.type != htons(ETH_P_IPV6))
 629                         return -EINVAL;
 630
 631                 if (!flow_key->ip.proto)
 632                         return -EINVAL;
 633
 634                 ipv6_key = nla_data(ovs_key);
 635                 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
 636                         return -EINVAL;
 637
 638                 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
 639                         return -EINVAL;
 640
 641                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
 642                         return -EINVAL;
 643
 644                 break;
 645
 646         case OVS_KEY_ATTR_TCP:
 647                 if (flow_key->ip.proto != IPPROTO_TCP)
 648                         return -EINVAL;
 649
 650                 return validate_tp_port(flow_key);
 651
 652         case OVS_KEY_ATTR_UDP:
 653                 if (flow_key->ip.proto != IPPROTO_UDP)
 654                         return -EINVAL;
 655
 656                 return validate_tp_port(flow_key);
 657
 658         default:
 659                 return -EINVAL;
 660         }
 661
 662         return 0;
 663 }
 664
 665 static int validate_userspace(const struct nlattr *attr)
 666 {
 667         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 668                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 669                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
 670         };
 671         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 672         int error;
 673
 674         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 675                                  attr, userspace_policy);
 676         if (error)
 677                 return error;
 678
 679         if (!a[OVS_USERSPACE_ATTR_PID] ||
 680             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 681                 return -EINVAL;
 682
 683         return 0;
 684 }
 685
 686 static int validate_actions(const struct nlattr *attr,
 687                                 const struct sw_flow_key *key,  int depth)
 688 {
 689         const struct nlattr *a;
 690         int rem, err;
 691
 692         if (depth >= SAMPLE_ACTION_DEPTH)
 693                 return -EOVERFLOW;
 694
 695         nla_for_each_nested(a, attr, rem) {
 696                 /* Expected argument lengths, (u32)-1 for variable length. */
 697                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 698                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 699                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 700                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 701                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 702                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 703                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 704                 };
 705                 const struct ovs_action_push_vlan *vlan;
 706                 int type = nla_type(a);
 707
 708                 if (type > OVS_ACTION_ATTR_MAX ||
 709                     (action_lens[type] != nla_len(a) &&
 710                      action_lens[type] != (u32)-1))
 711                         return -EINVAL;
 712
 713                 switch (type) {
 714                 case OVS_ACTION_ATTR_UNSPEC:
 715                         return -EINVAL;
 716
 717                 case OVS_ACTION_ATTR_USERSPACE:
 718                         err = validate_userspace(a);
 719                         if (err)
 720                                 return err;
 721                         break;
 722
 723                 case OVS_ACTION_ATTR_OUTPUT:
 724                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 725                                 return -EINVAL;
 726                         break;
 727
 728
 729                 case OVS_ACTION_ATTR_POP_VLAN:
 730                         break;
 731
 732                 case OVS_ACTION_ATTR_PUSH_VLAN:
 733                         vlan = nla_data(a);
 734                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 735                                 return -EINVAL;
 736                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 737                                 return -EINVAL;
 738                         break;
 739
 740                 case OVS_ACTION_ATTR_SET:
 741                         err = validate_set(a, key);
 742                         if (err)
 743                                 return err;
 744                         break;
 745
 746                 case OVS_ACTION_ATTR_SAMPLE:
 747                         err = validate_sample(a, key, depth);
 748                         if (err)
 749                                 return err;
 750                         break;
 751
 752                 default:
 753                         return -EINVAL;
 754                 }
 755         }
 756
 757         if (rem > 0)
 758                 return -EINVAL;
 759
 760         return 0;
 761 }
 762
 763 static void clear_stats(struct sw_flow *flow)
 764 {
 765         flow->used = 0;
 766         flow->tcp_flags = 0;
 767         flow->packet_count = 0;
 768         flow->byte_count = 0;
 769 }
 770
 771 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 772 {
 773         struct ovs_header *ovs_header = info->userhdr;
 774         struct nlattr **a = info->attrs;
 775         struct sw_flow_actions *acts;
 776         struct sk_buff *packet;
 777         struct sw_flow *flow;
 778         struct datapath *dp;
 779         struct ethhdr *eth;
 780         int len;
 781         int err;
 782         int key_len;
 783
 784         err = -EINVAL;
 785         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 786             !a[OVS_PACKET_ATTR_ACTIONS] ||
 787             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 788                 goto err;
 789
 790         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 791         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 792         err = -ENOMEM;
 793         if (!packet)
 794                 goto err;
 795         skb_reserve(packet, NET_IP_ALIGN);
 796
 797         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 798
 799         skb_reset_mac_header(packet);
 800         eth = eth_hdr(packet);
 801
 802         /* Normally, setting the skb 'protocol' field would be handled by a
 803          * call to eth_type_trans(), but it assumes there's a sending
 804          * device, which we may not have. */
 805         if (ntohs(eth->h_proto) >= 1536)
 806                 packet->protocol = eth->h_proto;
 807         else
 808                 packet->protocol = htons(ETH_P_802_2);
 809
 810         /* Build an sw_flow for sending this packet. */
 811         flow = ovs_flow_alloc();
 812         err = PTR_ERR(flow);
 813         if (IS_ERR(flow))
 814                 goto err_kfree_skb;
 815
 816         err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
 817         if (err)
 818                 goto err_flow_free;
 819
 820         err = ovs_flow_metadata_from_nlattrs(flow, key_len, a[OVS_PACKET_ATTR_KEY]);
 821         if (err)
 822                 goto err_flow_free;
 823
 824         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
 825         if (err)
 826                 goto err_flow_free;
 827
 828         acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 829         err = PTR_ERR(acts);
 830         if (IS_ERR(acts))
 831                 goto err_flow_free;
 832         rcu_assign_pointer(flow->sf_acts, acts);
 833
 834         OVS_CB(packet)->flow = flow;
 835         packet->priority = flow->key.phy.priority;
 836         skb_set_mark(packet, flow->key.phy.skb_mark);
 837
 838         rcu_read_lock();
 839         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 840         err = -ENODEV;
 841         if (!dp)
 842                 goto err_unlock;
 843
 844         local_bh_disable();
 845         err = ovs_execute_actions(dp, packet);
 846         local_bh_enable();
 847         rcu_read_unlock();
 848
 849         ovs_flow_free(flow);
 850         return err;
 851
 852 err_unlock:
 853         rcu_read_unlock();
 854 err_flow_free:
 855         ovs_flow_free(flow);
 856 err_kfree_skb:
 857         kfree_skb(packet);
 858 err:
 859         return err;
 860 }
 861
 862 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 863         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 864         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 865         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 866 };
 867
 868 static struct genl_ops dp_packet_genl_ops[] = {
 869         { .cmd = OVS_PACKET_CMD_EXECUTE,
 870           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 871           .policy = packet_policy,
 872           .doit = ovs_packet_cmd_execute
 873         }
 874 };
 875
 876 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 877 {
 878         int i;
 879         struct flow_table *table = genl_dereference(dp->table);
 880
 881         stats->n_flows = ovs_flow_tbl_count(table);
 882
 883         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 884         for_each_possible_cpu(i) {
 885                 const struct dp_stats_percpu *percpu_stats;
 886                 struct dp_stats_percpu local_stats;
 887                 unsigned int start;
 888
 889                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 890
 891                 do {
 892                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 893                         local_stats = *percpu_stats;
 894                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 895
 896                 stats->n_hit += local_stats.n_hit;
 897                 stats->n_missed += local_stats.n_missed;
 898                 stats->n_lost += local_stats.n_lost;
 899         }
 900 }
 901
 902 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 903         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 904         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 905         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 906 };
 907
 908 static struct genl_family dp_flow_genl_family = {
 909         .id = GENL_ID_GENERATE,
 910         .hdrsize = sizeof(struct ovs_header),
 911         .name = OVS_FLOW_FAMILY,
 912         .version = OVS_FLOW_VERSION,
 913         .maxattr = OVS_FLOW_ATTR_MAX,
 914          SET_NETNSOK
 915 };
 916
 917 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 918         .name = OVS_FLOW_MCGROUP
 919 };
 920
 921 /* Called with genl_lock. */
 922 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 923                                   struct sk_buff *skb, u32 pid,
 924                                   u32 seq, u32 flags, u8 cmd)
 925 {
 926         const int skb_orig_len = skb->len;
 927         const struct sw_flow_actions *sf_acts;
 928         struct ovs_flow_stats stats;
 929         struct ovs_header *ovs_header;
 930         struct nlattr *nla;
 931         unsigned long used;
 932         u8 tcp_flags;
 933         int err;
 934
 935         sf_acts = rcu_dereference_protected(flow->sf_acts,
 936                                             lockdep_genl_is_held());
 937
 938         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 939         if (!ovs_header)
 940                 return -EMSGSIZE;
 941
 942         ovs_header->dp_ifindex = get_dpifindex(dp);
 943
 944         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 945         if (!nla)
 946                 goto nla_put_failure;
 947         err = ovs_flow_to_nlattrs(&flow->key, skb);
 948         if (err)
 949                 goto error;
 950         nla_nest_end(skb, nla);
 951
 952         spin_lock_bh(&flow->lock);
 953         used = flow->used;
 954         stats.n_packets = flow->packet_count;
 955         stats.n_bytes = flow->byte_count;
 956         tcp_flags = flow->tcp_flags;
 957         spin_unlock_bh(&flow->lock);
 958
 959         if (used &&
 960             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 961                 goto nla_put_failure;
 962
 963         if (stats.n_packets &&
 964             nla_put(skb, OVS_FLOW_ATTR_STATS,
 965                     sizeof(struct ovs_flow_stats), &stats))
 966                 goto nla_put_failure;
 967
 968         if (tcp_flags &&
 969             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
 970                 goto nla_put_failure;
 971
 972         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 973          * this is the first flow to be dumped into 'skb'.  This is unusual for
 974          * Netlink but individual action lists can be longer than
 975          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 976          * The userspace caller can always fetch the actions separately if it
 977          * really wants them.  (Most userspace callers in fact don't care.)
 978          *
 979          * This can only fail for dump operations because the skb is always
 980          * properly sized for single flows.
 981          */
 982         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 983                       sf_acts->actions);
 984         if (err < 0 && skb_orig_len)
 985                 goto error;
 986
 987         return genlmsg_end(skb, ovs_header);
 988
 989 nla_put_failure:
 990         err = -EMSGSIZE;
 991 error:
 992         genlmsg_cancel(skb, ovs_header);
 993         return err;
 994 }
 995
 996 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 997 {
 998         const struct sw_flow_actions *sf_acts;
 999         int len;
1000
1001         sf_acts = rcu_dereference_protected(flow->sf_acts,
1002                                             lockdep_genl_is_held());
1003
1004         /* OVS_FLOW_ATTR_KEY */
1005         len = nla_total_size(FLOW_BUFSIZE);
1006         /* OVS_FLOW_ATTR_ACTIONS */
1007         len += nla_total_size(sf_acts->actions_len);
1008         /* OVS_FLOW_ATTR_STATS */
1009         len += nla_total_size(sizeof(struct ovs_flow_stats));
1010         /* OVS_FLOW_ATTR_TCP_FLAGS */
1011         len += nla_total_size(1);
1012         /* OVS_FLOW_ATTR_USED */
1013         len += nla_total_size(8);
1014
1015         len += NLMSG_ALIGN(sizeof(struct ovs_header));
1016
1017         return genlmsg_new(len, GFP_KERNEL);
1018 }
1019
1020 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1021                                                struct datapath *dp,
1022                                                u32 pid, u32 seq, u8 cmd)
1023 {
1024         struct sk_buff *skb;
1025         int retval;
1026
1027         skb = ovs_flow_cmd_alloc_info(flow);
1028         if (!skb)
1029                 return ERR_PTR(-ENOMEM);
1030
1031         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
1032         BUG_ON(retval < 0);
1033         return skb;
1034 }
1035
1036 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1037 {
1038         struct nlattr **a = info->attrs;
1039         struct ovs_header *ovs_header = info->userhdr;
1040         struct sw_flow_key key;
1041         struct sw_flow *flow;
1042         struct sk_buff *reply;
1043         struct datapath *dp;
1044         struct flow_table *table;
1045         int error;
1046         int key_len;
1047
1048         /* Extract key. */
1049         error = -EINVAL;
1050         if (!a[OVS_FLOW_ATTR_KEY])
1051                 goto error;
1052         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1053         if (error)
1054                 goto error;
1055
1056         /* Validate actions. */
1057         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1058                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
1059                 if (error)
1060                         goto error;
1061         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1062                 error = -EINVAL;
1063                 goto error;
1064         }
1065
1066         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1067         error = -ENODEV;
1068         if (!dp)
1069                 goto error;
1070
1071         table = genl_dereference(dp->table);
1072         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1073         if (!flow) {
1074                 struct sw_flow_actions *acts;
1075
1076                 /* Bail out if we're not allowed to create a new flow. */
1077                 error = -ENOENT;
1078                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1079                         goto error;
1080
1081                 /* Expand table, if necessary, to make room. */
1082                 if (ovs_flow_tbl_need_to_expand(table)) {
1083                         struct flow_table *new_table;
1084
1085                         new_table = ovs_flow_tbl_expand(table);
1086                         if (!IS_ERR(new_table)) {
1087                                 rcu_assign_pointer(dp->table, new_table);
1088                                 ovs_flow_tbl_deferred_destroy(table);
1089                                 table = genl_dereference(dp->table);
1090                         }
1091                 }
1092
1093                 /* Allocate flow. */
1094                 flow = ovs_flow_alloc();
1095                 if (IS_ERR(flow)) {
1096                         error = PTR_ERR(flow);
1097                         goto error;
1098                 }
1099                 clear_stats(flow);
1100
1101                 /* Obtain actions. */
1102                 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1103                 error = PTR_ERR(acts);
1104                 if (IS_ERR(acts))
1105                         goto error_free_flow;
1106                 rcu_assign_pointer(flow->sf_acts, acts);
1107
1108                 /* Put flow in bucket. */
1109                 ovs_flow_tbl_insert(table, flow, &key, key_len);
1110
1111                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1112                                                 info->snd_seq,
1113                                                 OVS_FLOW_CMD_NEW);
1114         } else {
1115                 /* We found a matching flow. */
1116                 struct sw_flow_actions *old_acts;
1117                 struct nlattr *acts_attrs;
1118
1119                 /* Bail out if we're not allowed to modify an existing flow.
1120                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1121                  * because Generic Netlink treats the latter as a dump
1122                  * request.  We also accept NLM_F_EXCL in case that bug ever
1123                  * gets fixed.
1124                  */
1125                 error = -EEXIST;
1126                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1127                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1128                         goto error;
1129
1130                 /* Update actions. */
1131                 old_acts = rcu_dereference_protected(flow->sf_acts,
1132                                                      lockdep_genl_is_held());
1133                 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1134                 if (acts_attrs &&
1135                    (old_acts->actions_len != nla_len(acts_attrs) ||
1136                    memcmp(old_acts->actions, nla_data(acts_attrs),
1137                           old_acts->actions_len))) {
1138                         struct sw_flow_actions *new_acts;
1139
1140                         new_acts = ovs_flow_actions_alloc(acts_attrs);
1141                         error = PTR_ERR(new_acts);
1142                         if (IS_ERR(new_acts))
1143                                 goto error;
1144
1145                         rcu_assign_pointer(flow->sf_acts, new_acts);
1146                         ovs_flow_deferred_free_acts(old_acts);
1147                 }
1148
1149                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1150                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1151
1152                 /* Clear stats. */
1153                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1154                         spin_lock_bh(&flow->lock);
1155                         clear_stats(flow);
1156                         spin_unlock_bh(&flow->lock);
1157                 }
1158         }
1159
1160         if (!IS_ERR(reply))
1161                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1162                            ovs_dp_flow_multicast_group.id, info->nlhdr,
1163                            GFP_KERNEL);
1164         else
1165                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1166                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1167         return 0;
1168
1169 error_free_flow:
1170         ovs_flow_free(flow);
1171 error:
1172         return error;
1173 }
1174
1175 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1176 {
1177         struct nlattr **a = info->attrs;
1178         struct ovs_header *ovs_header = info->userhdr;
1179         struct sw_flow_key key;
1180         struct sk_buff *reply;
1181         struct sw_flow *flow;
1182         struct datapath *dp;
1183         struct flow_table *table;
1184         int err;
1185         int key_len;
1186
1187         if (!a[OVS_FLOW_ATTR_KEY])
1188                 return -EINVAL;
1189         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1190         if (err)
1191                 return err;
1192
1193         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1194         if (!dp)
1195                 return -ENODEV;
1196
1197         table = genl_dereference(dp->table);
1198         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1199         if (!flow)
1200                 return -ENOENT;
1201
1202         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1203                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1204         if (IS_ERR(reply))
1205                 return PTR_ERR(reply);
1206
1207         return genlmsg_reply(reply, info);
1208 }
1209
1210 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1211 {
1212         struct nlattr **a = info->attrs;
1213         struct ovs_header *ovs_header = info->userhdr;
1214         struct sw_flow_key key;
1215         struct sk_buff *reply;
1216         struct sw_flow *flow;
1217         struct datapath *dp;
1218         struct flow_table *table;
1219         int err;
1220         int key_len;
1221
1222         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1223         if (!dp)
1224                 return -ENODEV;
1225
1226         if (!a[OVS_FLOW_ATTR_KEY])
1227                 return flush_flows(dp);
1228
1229         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1230         if (err)
1231                 return err;
1232
1233         table = genl_dereference(dp->table);
1234         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1235         if (!flow)
1236                 return -ENOENT;
1237
1238         reply = ovs_flow_cmd_alloc_info(flow);
1239         if (!reply)
1240                 return -ENOMEM;
1241
1242         ovs_flow_tbl_remove(table, flow);
1243
1244         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1245                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1246         BUG_ON(err < 0);
1247
1248         ovs_flow_deferred_free(flow);
1249
1250         genl_notify(reply, genl_info_net(info), info->snd_pid,
1251                     ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1252         return 0;
1253 }
1254
1255 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1256 {
1257         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1258         struct datapath *dp;
1259         struct flow_table *table;
1260
1261         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1262         if (!dp)
1263                 return -ENODEV;
1264
1265         table = genl_dereference(dp->table);
1266
1267         for (;;) {
1268                 struct sw_flow *flow;
1269                 u32 bucket, obj;
1270
1271                 bucket = cb->args[0];
1272                 obj = cb->args[1];
1273                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1274                 if (!flow)
1275                         break;
1276
1277                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1278                                            NETLINK_CB(cb->skb).pid,
1279                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1280                                            OVS_FLOW_CMD_NEW) < 0)
1281                         break;
1282
1283                 cb->args[0] = bucket;
1284                 cb->args[1] = obj;
1285         }
1286         return skb->len;
1287 }
1288
1289 static struct genl_ops dp_flow_genl_ops[] = {
1290         { .cmd = OVS_FLOW_CMD_NEW,
1291           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1292           .policy = flow_policy,
1293           .doit = ovs_flow_cmd_new_or_set
1294         },
1295         { .cmd = OVS_FLOW_CMD_DEL,
1296           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1297           .policy = flow_policy,
1298           .doit = ovs_flow_cmd_del
1299         },
1300         { .cmd = OVS_FLOW_CMD_GET,
1301           .flags = 0,               /* OK for unprivileged users. */
1302           .policy = flow_policy,
1303           .doit = ovs_flow_cmd_get,
1304           .dumpit = ovs_flow_cmd_dump
1305         },
1306         { .cmd = OVS_FLOW_CMD_SET,
1307           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1308           .policy = flow_policy,
1309           .doit = ovs_flow_cmd_new_or_set,
1310         },
1311 };
1312
1313 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1314 #ifdef HAVE_NLA_NUL_STRING
1315         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1316 #endif
1317         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1318 };
1319
1320 static struct genl_family dp_datapath_genl_family = {
1321         .id = GENL_ID_GENERATE,
1322         .hdrsize = sizeof(struct ovs_header),
1323         .name = OVS_DATAPATH_FAMILY,
1324         .version = OVS_DATAPATH_VERSION,
1325         .maxattr = OVS_DP_ATTR_MAX,
1326          SET_NETNSOK
1327 };
1328
1329 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1330         .name = OVS_DATAPATH_MCGROUP
1331 };
1332
1333 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1334                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1335 {
1336         struct ovs_header *ovs_header;
1337         struct ovs_dp_stats dp_stats;
1338         int err;
1339
1340         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1341                                    flags, cmd);
1342         if (!ovs_header)
1343                 goto error;
1344
1345         ovs_header->dp_ifindex = get_dpifindex(dp);
1346
1347         rcu_read_lock();
1348         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1349         rcu_read_unlock();
1350         if (err)
1351                 goto nla_put_failure;
1352
1353         get_dp_stats(dp, &dp_stats);
1354         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1355                 goto nla_put_failure;
1356
1357         return genlmsg_end(skb, ovs_header);
1358
1359 nla_put_failure:
1360         genlmsg_cancel(skb, ovs_header);
1361 error:
1362         return -EMSGSIZE;
1363 }
1364
1365 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1366                                              u32 seq, u8 cmd)
1367 {
1368         struct sk_buff *skb;
1369         int retval;
1370
1371         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1372         if (!skb)
1373                 return ERR_PTR(-ENOMEM);
1374
1375         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1376         if (retval < 0) {
1377                 kfree_skb(skb);
1378                 return ERR_PTR(retval);
1379         }
1380         return skb;
1381 }
1382
1383 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1384 {
1385         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1386 }
1387
1388 /* Called with genl_mutex and optionally with RTNL lock also. */
1389 static struct datapath *lookup_datapath(struct net *net,
1390                                         struct ovs_header *ovs_header,
1391                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1392 {
1393         struct datapath *dp;
1394
1395         if (!a[OVS_DP_ATTR_NAME])
1396                 dp = get_dp(net, ovs_header->dp_ifindex);
1397         else {
1398                 struct vport *vport;
1399
1400                 rcu_read_lock();
1401                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1402                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1403                 rcu_read_unlock();
1404         }
1405         return dp ? dp : ERR_PTR(-ENODEV);
1406 }
1407
1408 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1409 {
1410         struct nlattr **a = info->attrs;
1411         struct vport_parms parms;
1412         struct sk_buff *reply;
1413         struct datapath *dp;
1414         struct vport *vport;
1415         struct ovs_net *ovs_net;
1416         int err, i;
1417
1418         err = -EINVAL;
1419         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1420                 goto err;
1421
1422         err = ovs_dp_cmd_validate(a);
1423         if (err)
1424                 goto err;
1425
1426         rtnl_lock();
1427
1428         err = -ENOMEM;
1429         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1430         if (dp == NULL)
1431                 goto err_unlock_rtnl;
1432
1433         /* Initialize kobject for bridge.  This will be added as
1434          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1435         dp->ifobj.kset = NULL;
1436         kobject_init(&dp->ifobj, &dp_ktype);
1437
1438         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1439
1440         /* Allocate table. */
1441         err = -ENOMEM;
1442         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1443         if (!dp->table)
1444                 goto err_free_dp;
1445
1446         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1447         if (!dp->stats_percpu) {
1448                 err = -ENOMEM;
1449                 goto err_destroy_table;
1450         }
1451
1452         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1453                             GFP_KERNEL);
1454         if (!dp->ports) {
1455                 err = -ENOMEM;
1456                 goto err_destroy_percpu;
1457         }
1458
1459         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1460                 INIT_HLIST_HEAD(&dp->ports[i]);
1461
1462         /* Set up our datapath device. */
1463         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1464         parms.type = OVS_VPORT_TYPE_INTERNAL;
1465         parms.options = NULL;
1466         parms.dp = dp;
1467         parms.port_no = OVSP_LOCAL;
1468         parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1469
1470         vport = new_vport(&parms);
1471         if (IS_ERR(vport)) {
1472                 err = PTR_ERR(vport);
1473                 if (err == -EBUSY)
1474                         err = -EEXIST;
1475
1476                 goto err_destroy_ports_array;
1477         }
1478
1479         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1480                                       info->snd_seq, OVS_DP_CMD_NEW);
1481         err = PTR_ERR(reply);
1482         if (IS_ERR(reply))
1483                 goto err_destroy_local_port;
1484
1485         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1486         list_add_tail(&dp->list_node, &ovs_net->dps);
1487         ovs_dp_sysfs_add_dp(dp);
1488
1489         rtnl_unlock();
1490
1491         genl_notify(reply, genl_info_net(info), info->snd_pid,
1492                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1493                     GFP_KERNEL);
1494         return 0;
1495
1496 err_destroy_local_port:
1497         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1498 err_destroy_ports_array:
1499         kfree(dp->ports);
1500 err_destroy_percpu:
1501         free_percpu(dp->stats_percpu);
1502 err_destroy_table:
1503         ovs_flow_tbl_destroy(genl_dereference(dp->table));
1504 err_free_dp:
1505         release_net(ovs_dp_get_net(dp));
1506         kfree(dp);
1507 err_unlock_rtnl:
1508         rtnl_unlock();
1509 err:
1510         return err;
1511 }
1512
1513 /* Called with genl_mutex. */
1514 static void __dp_destroy(struct datapath *dp)
1515 {
1516         int i;
1517
1518         rtnl_lock();
1519
1520         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1521                 struct vport *vport;
1522                 struct hlist_node *node, *n;
1523
1524                 hlist_for_each_entry_safe(vport, node, n, &dp->ports[i], dp_hash_node)
1525                         if (vport->port_no != OVSP_LOCAL)
1526                                 ovs_dp_detach_port(vport);
1527         }
1528
1529         ovs_dp_sysfs_del_dp(dp);
1530         list_del(&dp->list_node);
1531         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1532
1533         /* rtnl_unlock() will wait until all the references to devices that
1534          * are pending unregistration have been dropped.  We do it here to
1535          * ensure that any internal devices (which contain DP pointers) are
1536          * fully destroyed before freeing the datapath.
1537          */
1538         rtnl_unlock();
1539
1540         call_rcu(&dp->rcu, destroy_dp_rcu);
1541 }
1542
1543 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1544 {
1545         struct sk_buff *reply;
1546         struct datapath *dp;
1547         int err;
1548
1549         err = ovs_dp_cmd_validate(info->attrs);
1550         if (err)
1551                 return err;
1552
1553         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1554         err = PTR_ERR(dp);
1555         if (IS_ERR(dp))
1556                 return err;
1557
1558         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1559                                       info->snd_seq, OVS_DP_CMD_DEL);
1560         err = PTR_ERR(reply);
1561         if (IS_ERR(reply))
1562                 return err;
1563
1564         __dp_destroy(dp);
1565
1566         genl_notify(reply, genl_info_net(info), info->snd_pid,
1567                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1568                     GFP_KERNEL);
1569
1570         return 0;
1571 }
1572
1573 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1574 {
1575         struct sk_buff *reply;
1576         struct datapath *dp;
1577         int err;
1578
1579         err = ovs_dp_cmd_validate(info->attrs);
1580         if (err)
1581                 return err;
1582
1583         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1584         if (IS_ERR(dp))
1585                 return PTR_ERR(dp);
1586
1587         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1588                                       info->snd_seq, OVS_DP_CMD_NEW);
1589         if (IS_ERR(reply)) {
1590                 err = PTR_ERR(reply);
1591                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1592                                 ovs_dp_datapath_multicast_group.id, err);
1593                 return 0;
1594         }
1595
1596         genl_notify(reply, genl_info_net(info), info->snd_pid,
1597                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1598                     GFP_KERNEL);
1599
1600         return 0;
1601 }
1602
1603 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1604 {
1605         struct sk_buff *reply;
1606         struct datapath *dp;
1607         int err;
1608
1609         err = ovs_dp_cmd_validate(info->attrs);
1610         if (err)
1611                 return err;
1612
1613         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1614         if (IS_ERR(dp))
1615                 return PTR_ERR(dp);
1616
1617         reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1618                                       info->snd_seq, OVS_DP_CMD_NEW);
1619         if (IS_ERR(reply))
1620                 return PTR_ERR(reply);
1621
1622         return genlmsg_reply(reply, info);
1623 }
1624
1625 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1626 {
1627         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1628         struct datapath *dp;
1629         int skip = cb->args[0];
1630         int i = 0;
1631
1632         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1633                 if (i >= skip &&
1634                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1635                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1636                                          OVS_DP_CMD_NEW) < 0)
1637                         break;
1638                 i++;
1639         }
1640
1641         cb->args[0] = i;
1642
1643         return skb->len;
1644 }
1645
1646 static struct genl_ops dp_datapath_genl_ops[] = {
1647         { .cmd = OVS_DP_CMD_NEW,
1648           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1649           .policy = datapath_policy,
1650           .doit = ovs_dp_cmd_new
1651         },
1652         { .cmd = OVS_DP_CMD_DEL,
1653           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1654           .policy = datapath_policy,
1655           .doit = ovs_dp_cmd_del
1656         },
1657         { .cmd = OVS_DP_CMD_GET,
1658           .flags = 0,               /* OK for unprivileged users. */
1659           .policy = datapath_policy,
1660           .doit = ovs_dp_cmd_get,
1661           .dumpit = ovs_dp_cmd_dump
1662         },
1663         { .cmd = OVS_DP_CMD_SET,
1664           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1665           .policy = datapath_policy,
1666           .doit = ovs_dp_cmd_set,
1667         },
1668 };
1669
1670 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1671 #ifdef HAVE_NLA_NUL_STRING
1672         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1673         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1674         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1675 #else
1676         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1677         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1678 #endif
1679         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1680         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1681         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1682         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1683 };
1684
1685 static struct genl_family dp_vport_genl_family = {
1686         .id = GENL_ID_GENERATE,
1687         .hdrsize = sizeof(struct ovs_header),
1688         .name = OVS_VPORT_FAMILY,
1689         .version = OVS_VPORT_VERSION,
1690         .maxattr = OVS_VPORT_ATTR_MAX,
1691          SET_NETNSOK
1692 };
1693
1694 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1695         .name = OVS_VPORT_MCGROUP
1696 };
1697
1698 /* Called with RTNL lock or RCU read lock. */
1699 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1700                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1701 {
1702         struct ovs_header *ovs_header;
1703         struct ovs_vport_stats vport_stats;
1704         int err;
1705
1706         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1707                                  flags, cmd);
1708         if (!ovs_header)
1709                 return -EMSGSIZE;
1710
1711         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1712
1713         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1714             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1715             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1716             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid))
1717                 goto nla_put_failure;
1718
1719         ovs_vport_get_stats(vport, &vport_stats);
1720         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1721                     &vport_stats))
1722                 goto nla_put_failure;
1723
1724         if (nla_put(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN,
1725                     vport->ops->get_addr(vport)))
1726                 goto nla_put_failure;
1727
1728         err = ovs_vport_get_options(vport, skb);
1729         if (err == -EMSGSIZE)
1730                 goto error;
1731
1732         return genlmsg_end(skb, ovs_header);
1733
1734 nla_put_failure:
1735         err = -EMSGSIZE;
1736 error:
1737         genlmsg_cancel(skb, ovs_header);
1738         return err;
1739 }
1740
1741 /* Called with RTNL lock or RCU read lock. */
1742 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1743                                          u32 seq, u8 cmd)
1744 {
1745         struct sk_buff *skb;
1746         int retval;
1747
1748         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1749         if (!skb)
1750                 return ERR_PTR(-ENOMEM);
1751
1752         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1753         if (retval < 0) {
1754                 kfree_skb(skb);
1755                 return ERR_PTR(retval);
1756         }
1757         return skb;
1758 }
1759
1760 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1761 {
1762         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1763 }
1764
1765 /* Called with RTNL lock or RCU read lock. */
1766 static struct vport *lookup_vport(struct net *net,
1767                                   struct ovs_header *ovs_header,
1768                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1769 {
1770         struct datapath *dp;
1771         struct vport *vport;
1772
1773         if (a[OVS_VPORT_ATTR_NAME]) {
1774                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1775                 if (!vport)
1776                         return ERR_PTR(-ENODEV);
1777                 if (ovs_header->dp_ifindex &&
1778                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1779                         return ERR_PTR(-ENODEV);
1780                 return vport;
1781         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1782                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1783
1784                 if (port_no >= DP_MAX_PORTS)
1785                         return ERR_PTR(-EFBIG);
1786
1787                 dp = get_dp(net, ovs_header->dp_ifindex);
1788                 if (!dp)
1789                         return ERR_PTR(-ENODEV);
1790
1791                 vport = ovs_vport_rtnl_rcu(dp, port_no);
1792                 if (!vport)
1793                         return ERR_PTR(-ENOENT);
1794                 return vport;
1795         } else
1796                 return ERR_PTR(-EINVAL);
1797 }
1798
1799 /* Called with RTNL lock. */
1800 static int change_vport(struct vport *vport,
1801                         struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1802 {
1803         int err = 0;
1804
1805         if (a[OVS_VPORT_ATTR_STATS])
1806                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1807
1808         if (a[OVS_VPORT_ATTR_ADDRESS])
1809                 err = ovs_vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1810
1811         return err;
1812 }
1813
1814 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1815 {
1816         struct nlattr **a = info->attrs;
1817         struct ovs_header *ovs_header = info->userhdr;
1818         struct vport_parms parms;
1819         struct sk_buff *reply;
1820         struct vport *vport;
1821         struct datapath *dp;
1822         u32 port_no;
1823         int err;
1824
1825         err = -EINVAL;
1826         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1827             !a[OVS_VPORT_ATTR_UPCALL_PID])
1828                 goto exit;
1829
1830         err = ovs_vport_cmd_validate(a);
1831         if (err)
1832                 goto exit;
1833
1834         rtnl_lock();
1835         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1836         err = -ENODEV;
1837         if (!dp)
1838                 goto exit_unlock;
1839
1840         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1841                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1842
1843                 err = -EFBIG;
1844                 if (port_no >= DP_MAX_PORTS)
1845                         goto exit_unlock;
1846
1847                 vport = ovs_vport_rtnl(dp, port_no);
1848                 err = -EBUSY;
1849                 if (vport)
1850                         goto exit_unlock;
1851         } else {
1852                 for (port_no = 1; ; port_no++) {
1853                         if (port_no >= DP_MAX_PORTS) {
1854                                 err = -EFBIG;
1855                                 goto exit_unlock;
1856                         }
1857                         vport = ovs_vport_rtnl(dp, port_no);
1858                         if (!vport)
1859                                 break;
1860                 }
1861         }
1862
1863         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1864         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1865         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1866         parms.dp = dp;
1867         parms.port_no = port_no;
1868         parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1869
1870         vport = new_vport(&parms);
1871         err = PTR_ERR(vport);
1872         if (IS_ERR(vport))
1873                 goto exit_unlock;
1874
1875         ovs_dp_sysfs_add_if(vport);
1876
1877         err = change_vport(vport, a);
1878         if (!err) {
1879                 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1880                                                  info->snd_seq,
1881                                                  OVS_VPORT_CMD_NEW);
1882                 if (IS_ERR(reply))
1883                         err = PTR_ERR(reply);
1884         }
1885         if (err) {
1886                 ovs_dp_detach_port(vport);
1887                 goto exit_unlock;
1888         }
1889         genl_notify(reply, genl_info_net(info), info->snd_pid,
1890                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1891
1892 exit_unlock:
1893         rtnl_unlock();
1894 exit:
1895         return err;
1896 }
1897
1898 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1899 {
1900         struct nlattr **a = info->attrs;
1901         struct sk_buff *reply;
1902         struct vport *vport;
1903         int err;
1904
1905         err = ovs_vport_cmd_validate(a);
1906         if (err)
1907                 goto exit;
1908
1909         rtnl_lock();
1910         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1911         err = PTR_ERR(vport);
1912         if (IS_ERR(vport))
1913                 goto exit_unlock;
1914
1915         err = 0;
1916         if (a[OVS_VPORT_ATTR_TYPE] &&
1917             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1918                 err = -EINVAL;
1919
1920         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1921                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1922         if (!err)
1923                 err = change_vport(vport, a);
1924         else
1925                 goto exit_unlock;
1926         if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1927                 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1928
1929         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1930                                          OVS_VPORT_CMD_NEW);
1931         if (IS_ERR(reply)) {
1932                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1933                                 ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
1934                 goto exit_unlock;
1935         }
1936
1937         genl_notify(reply, genl_info_net(info), info->snd_pid,
1938                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1939
1940 exit_unlock:
1941         rtnl_unlock();
1942 exit:
1943         return err;
1944 }
1945
1946 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1947 {
1948         struct nlattr **a = info->attrs;
1949         struct sk_buff *reply;
1950         struct vport *vport;
1951         int err;
1952
1953         err = ovs_vport_cmd_validate(a);
1954         if (err)
1955                 goto exit;
1956
1957         rtnl_lock();
1958         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1959         err = PTR_ERR(vport);
1960         if (IS_ERR(vport))
1961                 goto exit_unlock;
1962
1963         if (vport->port_no == OVSP_LOCAL) {
1964                 err = -EINVAL;
1965                 goto exit_unlock;
1966         }
1967
1968         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1969                                          OVS_VPORT_CMD_DEL);
1970         err = PTR_ERR(reply);
1971         if (IS_ERR(reply))
1972                 goto exit_unlock;
1973
1974         ovs_dp_detach_port(vport);
1975
1976         genl_notify(reply, genl_info_net(info), info->snd_pid,
1977                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1978
1979 exit_unlock:
1980         rtnl_unlock();
1981 exit:
1982         return err;
1983 }
1984
1985 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1986 {
1987         struct nlattr **a = info->attrs;
1988         struct ovs_header *ovs_header = info->userhdr;
1989         struct sk_buff *reply;
1990         struct vport *vport;
1991         int err;
1992
1993         err = ovs_vport_cmd_validate(a);
1994         if (err)
1995                 goto exit;
1996
1997         rcu_read_lock();
1998         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1999         err = PTR_ERR(vport);
2000         if (IS_ERR(vport))
2001                 goto exit_unlock;
2002
2003         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
2004                                          OVS_VPORT_CMD_NEW);
2005         err = PTR_ERR(reply);
2006         if (IS_ERR(reply))
2007                 goto exit_unlock;
2008
2009         rcu_read_unlock();
2010
2011         return genlmsg_reply(reply, info);
2012
2013 exit_unlock:
2014         rcu_read_unlock();
2015 exit:
2016         return err;
2017 }
2018
2019 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2020 {
2021         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2022         struct datapath *dp;
2023         int bucket = cb->args[0], skip = cb->args[1];
2024         int i, j = 0;
2025
2026         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2027         if (!dp)
2028                 return -ENODEV;
2029
2030         rcu_read_lock();
2031         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2032                 struct vport *vport;
2033                 struct hlist_node *n;
2034
2035                 j = 0;
2036                 hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node) {
2037                         if (j >= skip &&
2038                             ovs_vport_cmd_fill_info(vport, skb,
2039                                                     NETLINK_CB(cb->skb).pid,
2040                                                     cb->nlh->nlmsg_seq,
2041                                                     NLM_F_MULTI,
2042                                                     OVS_VPORT_CMD_NEW) < 0)
2043                                 goto out;
2044
2045                         j++;
2046                 }
2047                 skip = 0;
2048         }
2049 out:
2050         rcu_read_unlock();
2051
2052         cb->args[0] = i;
2053         cb->args[1] = j;
2054
2055         return skb->len;
2056 }
2057
2058 static struct genl_ops dp_vport_genl_ops[] = {
2059         { .cmd = OVS_VPORT_CMD_NEW,
2060           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2061           .policy = vport_policy,
2062           .doit = ovs_vport_cmd_new
2063         },
2064         { .cmd = OVS_VPORT_CMD_DEL,
2065           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2066           .policy = vport_policy,
2067           .doit = ovs_vport_cmd_del
2068         },
2069         { .cmd = OVS_VPORT_CMD_GET,
2070           .flags = 0,               /* OK for unprivileged users. */
2071           .policy = vport_policy,
2072           .doit = ovs_vport_cmd_get,
2073           .dumpit = ovs_vport_cmd_dump
2074         },
2075         { .cmd = OVS_VPORT_CMD_SET,
2076           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2077           .policy = vport_policy,
2078           .doit = ovs_vport_cmd_set,
2079         },
2080 };
2081
2082 struct genl_family_and_ops {
2083         struct genl_family *family;
2084         struct genl_ops *ops;
2085         int n_ops;
2086         struct genl_multicast_group *group;
2087 };
2088
2089 static const struct genl_family_and_ops dp_genl_families[] = {
2090         { &dp_datapath_genl_family,
2091           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2092           &ovs_dp_datapath_multicast_group },
2093         { &dp_vport_genl_family,
2094           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2095           &ovs_dp_vport_multicast_group },
2096         { &dp_flow_genl_family,
2097           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2098           &ovs_dp_flow_multicast_group },
2099         { &dp_packet_genl_family,
2100           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2101           NULL },
2102 };
2103
2104 static void dp_unregister_genl(int n_families)
2105 {
2106         int i;
2107
2108         for (i = 0; i < n_families; i++)
2109                 genl_unregister_family(dp_genl_families[i].family);
2110 }
2111
2112 static int dp_register_genl(void)
2113 {
2114         int n_registered;
2115         int err;
2116         int i;
2117
2118         n_registered = 0;
2119         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2120                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2121
2122                 err = genl_register_family_with_ops(f->family, f->ops,
2123                                                     f->n_ops);
2124                 if (err)
2125                         goto error;
2126                 n_registered++;
2127
2128                 if (f->group) {
2129                         err = genl_register_mc_group(f->family, f->group);
2130                         if (err)
2131                                 goto error;
2132                 }
2133         }
2134
2135         return 0;
2136
2137 error:
2138         dp_unregister_genl(n_registered);
2139         return err;
2140 }
2141
2142 static int __rehash_flow_table(void *dummy)
2143 {
2144         struct datapath *dp;
2145         struct net *net;
2146
2147         rtnl_lock();
2148         for_each_net(net) {
2149                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2150
2151                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2152                         struct flow_table *old_table = genl_dereference(dp->table);
2153                         struct flow_table *new_table;
2154
2155                         new_table = ovs_flow_tbl_rehash(old_table);
2156                         if (!IS_ERR(new_table)) {
2157                                 rcu_assign_pointer(dp->table, new_table);
2158                                 ovs_flow_tbl_deferred_destroy(old_table);
2159                         }
2160                 }
2161         }
2162         rtnl_unlock();
2163         return 0;
2164 }
2165
2166 static void rehash_flow_table(struct work_struct *work)
2167 {
2168         genl_exec(__rehash_flow_table, NULL);
2169         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2170 }
2171
2172 static int dp_destroy_all(void *data)
2173 {
2174         struct datapath *dp, *dp_next;
2175         struct ovs_net *ovs_net = data;
2176
2177         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2178                 __dp_destroy(dp);
2179
2180         return 0;
2181 }
2182
2183 static int __net_init ovs_init_net(struct net *net)
2184 {
2185         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2186
2187         INIT_LIST_HEAD(&ovs_net->dps);
2188         return 0;
2189 }
2190
2191 static void __net_exit ovs_exit_net(struct net *net)
2192 {
2193         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2194
2195         genl_exec(dp_destroy_all, ovs_net);
2196 }
2197
2198 static struct pernet_operations ovs_net_ops = {
2199         .init = ovs_init_net,
2200         .exit = ovs_exit_net,
2201         .id   = &ovs_net_id,
2202         .size = sizeof(struct ovs_net),
2203 };
2204
2205 static int __init dp_init(void)
2206 {
2207         struct sk_buff *dummy_skb;
2208         int err;
2209
2210         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2211
2212         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2213                 VERSION);
2214
2215         err = genl_exec_init();
2216         if (err)
2217                 goto error;
2218
2219         err = ovs_workqueues_init();
2220         if (err)
2221                 goto error_genl_exec;
2222
2223         err = ovs_tnl_init();
2224         if (err)
2225                 goto error_wq;
2226
2227         err = ovs_flow_init();
2228         if (err)
2229                 goto error_tnl_exit;
2230
2231         err = ovs_vport_init();
2232         if (err)
2233                 goto error_flow_exit;
2234
2235         err = register_pernet_device(&ovs_net_ops);
2236         if (err)
2237                 goto error_vport_exit;
2238
2239         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2240         if (err)
2241                 goto error_netns_exit;
2242
2243         err = dp_register_genl();
2244         if (err < 0)
2245                 goto error_unreg_notifier;
2246
2247         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2248
2249         return 0;
2250
2251 error_unreg_notifier:
2252         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2253 error_netns_exit:
2254         unregister_pernet_device(&ovs_net_ops);
2255 error_vport_exit:
2256         ovs_vport_exit();
2257 error_flow_exit:
2258         ovs_flow_exit();
2259 error_tnl_exit:
2260         ovs_tnl_exit();
2261 error_wq:
2262         ovs_workqueues_exit();
2263 error_genl_exec:
2264         genl_exec_exit();
2265 error:
2266         return err;
2267 }
2268
2269 static void dp_cleanup(void)
2270 {
2271         cancel_delayed_work_sync(&rehash_flow_wq);
2272         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2273         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2274         unregister_pernet_device(&ovs_net_ops);
2275         rcu_barrier();
2276         ovs_vport_exit();
2277         ovs_flow_exit();
2278         ovs_tnl_exit();
2279         ovs_workqueues_exit();
2280         genl_exec_exit();
2281 }
2282
2283 module_init(dp_init);
2284 module_exit(dp_cleanup);
2285
2286 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2287 MODULE_LICENSE("GPL");
2288 MODULE_VERSION(VERSION);