datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
   3  * Distributed under the terms of the GNU GPL version 2.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 /* Functions for managing the dp interface/device. */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/init.h>
  14 #include <linux/module.h>
  15 #include <linux/if_arp.h>
  16 #include <linux/if_vlan.h>
  17 #include <linux/in.h>
  18 #include <linux/ip.h>
  19 #include <linux/jhash.h>
  20 #include <linux/delay.h>
  21 #include <linux/time.h>
  22 #include <linux/etherdevice.h>
  23 #include <linux/genetlink.h>
  24 #include <linux/kernel.h>
  25 #include <linux/kthread.h>
  26 #include <linux/mutex.h>
  27 #include <linux/percpu.h>
  28 #include <linux/rcupdate.h>
  29 #include <linux/tcp.h>
  30 #include <linux/udp.h>
  31 #include <linux/version.h>
  32 #include <linux/ethtool.h>
  33 #include <linux/wait.h>
  34 #include <asm/system.h>
  35 #include <asm/div64.h>
  36 #include <asm/bug.h>
  37 #include <linux/highmem.h>
  38 #include <linux/netfilter_bridge.h>
  39 #include <linux/netfilter_ipv4.h>
  40 #include <linux/inetdevice.h>
  41 #include <linux/list.h>
  42 #include <linux/openvswitch.h>
  43 #include <linux/rculist.h>
  44 #include <linux/dmi.h>
  45 #include <net/inet_ecn.h>
  46 #include <net/genetlink.h>
  47
  48 #include "checksum.h"
  49 #include "datapath.h"
  50 #include "actions.h"
  51 #include "flow.h"
  52 #include "vlan.h"
  53 #include "tunnel.h"
  54 #include "vport-internal_dev.h"
  55
  56 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
  57     LINUX_VERSION_CODE >= KERNEL_VERSION(3,2,0)
  58 #error Kernels before 2.6.18 or after 3.1 are not supported by this version of Open vSwitch.
  59 #endif
  60
  61 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
  62 EXPORT_SYMBOL(dp_ioctl_hook);
  63
  64 /**
  65  * DOC: Locking:
  66  *
  67  * Writes to device state (add/remove datapath, port, set operations on vports,
  68  * etc.) are protected by RTNL.
  69  *
  70  * Writes to other state (flow table modifications, set miscellaneous datapath
  71  * parameters such as drop frags, etc.) are protected by genl_mutex.  The RTNL
  72  * lock nests inside genl_mutex.
  73  *
  74  * Reads are protected by RCU.
  75  *
  76  * There are a few special cases (mostly stats) that have their own
  77  * synchronization but they nest under all of above and don't interact with
  78  * each other.
  79  */
  80
  81 /* Global list of datapaths to enable dumping them all out.
  82  * Protected by genl_mutex.
  83  */
  84 static LIST_HEAD(dps);
  85
  86 static struct vport *new_vport(const struct vport_parms *);
  87 static int queue_userspace_packets(struct datapath *, struct sk_buff *,
  88                                  const struct dp_upcall_info *);
  89
  90 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  91 struct datapath *get_dp(int dp_ifindex)
  92 {
  93         struct datapath *dp = NULL;
  94         struct net_device *dev;
  95
  96         rcu_read_lock();
  97         dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
  98         if (dev) {
  99                 struct vport *vport = internal_dev_get_vport(dev);
 100                 if (vport)
 101                         dp = vport->dp;
 102         }
 103         rcu_read_unlock();
 104
 105         return dp;
 106 }
 107 EXPORT_SYMBOL_GPL(get_dp);
 108
 109 /* Must be called with genl_mutex. */
 110 static struct flow_table *get_table_protected(struct datapath *dp)
 111 {
 112         return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
 113 }
 114
 115 /* Must be called with rcu_read_lock or RTNL lock. */
 116 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
 117 {
 118         return rcu_dereference_rtnl(dp->ports[port_no]);
 119 }
 120
 121 /* Must be called with rcu_read_lock or RTNL lock. */
 122 const char *dp_name(const struct datapath *dp)
 123 {
 124         return vport_get_name(rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]));
 125 }
 126
 127 static int get_dpifindex(struct datapath *dp)
 128 {
 129         struct vport *local;
 130         int ifindex;
 131
 132         rcu_read_lock();
 133
 134         local = get_vport_protected(dp, OVSP_LOCAL);
 135         if (local)
 136                 ifindex = vport_get_ifindex(local);
 137         else
 138                 ifindex = 0;
 139
 140         rcu_read_unlock();
 141
 142         return ifindex;
 143 }
 144
 145 static inline size_t br_nlmsg_size(void)
 146 {
 147         return NLMSG_ALIGN(sizeof(struct ifinfomsg))
 148                + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
 149                + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
 150                + nla_total_size(4) /* IFLA_MASTER */
 151                + nla_total_size(4) /* IFLA_MTU */
 152                + nla_total_size(1); /* IFLA_OPERSTATE */
 153 }
 154
 155 /* Caller must hold RTNL lock. */
 156 static int dp_fill_ifinfo(struct sk_buff *skb,
 157                           const struct vport *port,
 158                           int event, unsigned int flags)
 159 {
 160         struct datapath *dp = port->dp;
 161         int ifindex = vport_get_ifindex(port);
 162         struct ifinfomsg *hdr;
 163         struct nlmsghdr *nlh;
 164
 165         if (ifindex < 0)
 166                 return ifindex;
 167
 168         nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
 169         if (nlh == NULL)
 170                 return -EMSGSIZE;
 171
 172         hdr = nlmsg_data(nlh);
 173         hdr->ifi_family = AF_BRIDGE;
 174         hdr->__ifi_pad = 0;
 175         hdr->ifi_type = ARPHRD_ETHER;
 176         hdr->ifi_index = ifindex;
 177         hdr->ifi_flags = vport_get_flags(port);
 178         hdr->ifi_change = 0;
 179
 180         NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
 181         NLA_PUT_U32(skb, IFLA_MASTER, get_dpifindex(dp));
 182         NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
 183 #ifdef IFLA_OPERSTATE
 184         NLA_PUT_U8(skb, IFLA_OPERSTATE,
 185                    vport_is_running(port)
 186                         ? vport_get_operstate(port)
 187                         : IF_OPER_DOWN);
 188 #endif
 189
 190         NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
 191
 192         return nlmsg_end(skb, nlh);
 193
 194 nla_put_failure:
 195         nlmsg_cancel(skb, nlh);
 196         return -EMSGSIZE;
 197 }
 198
 199 /* Caller must hold RTNL lock. */
 200 static void dp_ifinfo_notify(int event, struct vport *port)
 201 {
 202         struct sk_buff *skb;
 203         int err = -ENOBUFS;
 204
 205         skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
 206         if (skb == NULL)
 207                 goto errout;
 208
 209         err = dp_fill_ifinfo(skb, port, event, 0);
 210         if (err < 0) {
 211                 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
 212                 WARN_ON(err == -EMSGSIZE);
 213                 kfree_skb(skb);
 214                 goto errout;
 215         }
 216         rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 217         return;
 218 errout:
 219         if (err < 0)
 220                 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
 221 }
 222
 223 static void release_dp(struct kobject *kobj)
 224 {
 225         struct datapath *dp = container_of(kobj, struct datapath, ifobj);
 226         kfree(dp);
 227 }
 228
 229 static struct kobj_type dp_ktype = {
 230         .release = release_dp
 231 };
 232
 233 static void destroy_dp_rcu(struct rcu_head *rcu)
 234 {
 235         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 236
 237         flow_tbl_destroy(dp->table);
 238         free_percpu(dp->stats_percpu);
 239         kobject_put(&dp->ifobj);
 240 }
 241
 242 /* Called with RTNL lock and genl_lock. */
 243 static struct vport *new_vport(const struct vport_parms *parms)
 244 {
 245         struct vport *vport;
 246
 247         vport = vport_add(parms);
 248         if (!IS_ERR(vport)) {
 249                 struct datapath *dp = parms->dp;
 250
 251                 rcu_assign_pointer(dp->ports[parms->port_no], vport);
 252                 list_add(&vport->node, &dp->port_list);
 253
 254                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 255         }
 256
 257         return vport;
 258 }
 259
 260 /* Called with RTNL lock. */
 261 void dp_detach_port(struct vport *p)
 262 {
 263         ASSERT_RTNL();
 264
 265         if (p->port_no != OVSP_LOCAL)
 266                 dp_sysfs_del_if(p);
 267         dp_ifinfo_notify(RTM_DELLINK, p);
 268
 269         /* First drop references to device. */
 270         list_del(&p->node);
 271         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 272
 273         /* Then destroy it. */
 274         vport_del(p);
 275 }
 276
 277 /* Must be called with rcu_read_lock. */
 278 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 279 {
 280         struct datapath *dp = p->dp;
 281         struct sw_flow *flow;
 282         struct dp_stats_percpu *stats;
 283         u64 *stats_counter;
 284         int error;
 285
 286         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 287         OVS_CB(skb)->vport = p;
 288
 289         if (!OVS_CB(skb)->flow) {
 290                 struct sw_flow_key key;
 291                 int key_len;
 292                 bool is_frag;
 293
 294                 /* Extract flow from 'skb' into 'key'. */
 295                 error = flow_extract(skb, p->port_no, &key, &key_len, &is_frag);
 296                 if (unlikely(error)) {
 297                         kfree_skb(skb);
 298                         return;
 299                 }
 300
 301                 if (is_frag && dp->drop_frags) {
 302                         consume_skb(skb);
 303                         stats_counter = &stats->n_frags;
 304                         goto out;
 305                 }
 306
 307                 /* Look up flow. */
 308                 flow = flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
 309                 if (unlikely(!flow)) {
 310                         struct dp_upcall_info upcall;
 311
 312                         upcall.cmd = OVS_PACKET_CMD_MISS;
 313                         upcall.key = &key;
 314                         upcall.userdata = NULL;
 315                         upcall.pid = p->upcall_pid;
 316                         dp_upcall(dp, skb, &upcall);
 317                         kfree_skb(skb);
 318                         stats_counter = &stats->n_missed;
 319                         goto out;
 320                 }
 321
 322                 OVS_CB(skb)->flow = flow;
 323         }
 324
 325         stats_counter = &stats->n_hit;
 326         flow_used(OVS_CB(skb)->flow, skb);
 327         execute_actions(dp, skb);
 328
 329 out:
 330         /* Update datapath statistics. */
 331
 332         write_seqcount_begin(&stats->seqlock);
 333         (*stats_counter)++;
 334         write_seqcount_end(&stats->seqlock);
 335 }
 336
 337 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
 338 {
 339         u16 csum_start, csum_offset;
 340         __wsum csum;
 341
 342         get_skb_csum_pointers(skb, &csum_start, &csum_offset);
 343         csum_start -= skb_headroom(skb);
 344
 345         skb_copy_bits(skb, 0, to, csum_start);
 346
 347         csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
 348                                       skb->len - csum_start, 0);
 349         *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
 350 }
 351
 352 static struct genl_family dp_packet_genl_family = {
 353         .id = GENL_ID_GENERATE,
 354         .hdrsize = sizeof(struct ovs_header),
 355         .name = OVS_PACKET_FAMILY,
 356         .version = 1,
 357         .maxattr = OVS_PACKET_ATTR_MAX
 358 };
 359
 360 int dp_upcall(struct datapath *dp, struct sk_buff *skb,
 361               const struct dp_upcall_info *upcall_info)
 362 {
 363         struct sk_buff *segs = NULL;
 364         struct dp_stats_percpu *stats;
 365         int err;
 366
 367         if (upcall_info->pid == 0) {
 368                 err = -ENOTCONN;
 369                 goto err;
 370         }
 371
 372         forward_ip_summed(skb, true);
 373
 374         /* Break apart GSO packets into their component pieces.  Otherwise
 375          * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
 376         if (skb_is_gso(skb)) {
 377                 segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 378
 379                 if (IS_ERR(segs)) {
 380                         err = PTR_ERR(segs);
 381                         goto err;
 382                 }
 383                 skb = segs;
 384         }
 385
 386         err = queue_userspace_packets(dp, skb, upcall_info);
 387         if (segs) {
 388                 struct sk_buff *next;
 389                 /* Free GSO-segments */
 390                 do {
 391                         next = segs->next;
 392                         kfree_skb(segs);
 393                 } while ((segs = next) != NULL);
 394         }
 395
 396         if (err)
 397                 goto err;
 398
 399         return 0;
 400
 401 err:
 402         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 403
 404         write_seqcount_begin(&stats->seqlock);
 405         stats->n_lost++;
 406         write_seqcount_end(&stats->seqlock);
 407
 408         return err;
 409 }
 410
 411 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
 412  * 'upcall_info'.  There will be only one packet unless we broke up a GSO
 413  * packet.
 414  */
 415 static int queue_userspace_packets(struct datapath *dp, struct sk_buff *skb,
 416                                    const struct dp_upcall_info *upcall_info)
 417 {
 418         int dp_ifindex;
 419
 420         dp_ifindex = get_dpifindex(dp);
 421         if (!dp_ifindex)
 422                 return -ENODEV;
 423
 424         do {
 425                 struct ovs_header *upcall;
 426                 struct sk_buff *user_skb; /* to be queued to userspace */
 427                 struct nlattr *nla;
 428                 unsigned int len;
 429                 int err;
 430
 431                 err = vlan_deaccel_tag(skb);
 432                 if (unlikely(err))
 433                         return err;
 434
 435                 if (nla_attr_size(skb->len) > USHRT_MAX)
 436                         return -EFBIG;
 437
 438                 len = sizeof(struct ovs_header);
 439                 len += nla_total_size(skb->len);
 440                 len += nla_total_size(FLOW_BUFSIZE);
 441                 if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
 442                         len += nla_total_size(8);
 443
 444                 user_skb = genlmsg_new(len, GFP_ATOMIC);
 445                 if (!user_skb)
 446                         return -ENOMEM;
 447
 448                 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 449                                          0, upcall_info->cmd);
 450                 upcall->dp_ifindex = dp_ifindex;
 451
 452                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 453                 flow_to_nlattrs(upcall_info->key, user_skb);
 454                 nla_nest_end(user_skb, nla);
 455
 456                 if (upcall_info->userdata)
 457                         nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
 458                                     nla_get_u64(upcall_info->userdata));
 459
 460                 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 461                 if (skb->ip_summed == CHECKSUM_PARTIAL)
 462                         copy_and_csum_skb(skb, nla_data(nla));
 463                 else
 464                         skb_copy_bits(skb, 0, nla_data(nla), skb->len);
 465
 466                 err = genlmsg_unicast(&init_net, user_skb, upcall_info->pid);
 467                 if (err)
 468                         return err;
 469
 470         } while ((skb = skb->next));
 471
 472         return 0;
 473 }
 474
 475 /* Called with genl_mutex. */
 476 static int flush_flows(int dp_ifindex)
 477 {
 478         struct flow_table *old_table;
 479         struct flow_table *new_table;
 480         struct datapath *dp;
 481
 482         dp = get_dp(dp_ifindex);
 483         if (!dp)
 484                 return -ENODEV;
 485
 486         old_table = get_table_protected(dp);
 487         new_table = flow_tbl_alloc(TBL_MIN_BUCKETS);
 488         if (!new_table)
 489                 return -ENOMEM;
 490
 491         rcu_assign_pointer(dp->table, new_table);
 492
 493         flow_tbl_deferred_destroy(old_table);
 494         return 0;
 495 }
 496
 497 static int validate_actions(const struct nlattr *attr,
 498                                 const struct sw_flow_key *key, int depth);
 499
 500 static int validate_sample(const struct nlattr *attr,
 501                                 const struct sw_flow_key *key, int depth)
 502 {
 503         static const struct nla_policy sample_policy[OVS_SAMPLE_ATTR_MAX + 1] =
 504         {
 505                 [OVS_SAMPLE_ATTR_PROBABILITY] = {.type = NLA_U32 },
 506                 [OVS_SAMPLE_ATTR_ACTIONS] = {.type = NLA_UNSPEC },
 507         };
 508         struct nlattr *a[OVS_SAMPLE_ATTR_MAX + 1];
 509         int error;
 510
 511         error = nla_parse_nested(a, OVS_SAMPLE_ATTR_MAX, attr, sample_policy);
 512         if (error)
 513                 return error;
 514
 515         if (!a[OVS_SAMPLE_ATTR_PROBABILITY])
 516                 return -EINVAL;
 517         if (!a[OVS_SAMPLE_ATTR_ACTIONS])
 518                 return -EINVAL;
 519
 520         return validate_actions(a[OVS_SAMPLE_ATTR_ACTIONS], key, (depth + 1));
 521 }
 522
 523 static int validate_action_key(const struct nlattr *a,
 524                                 const struct sw_flow_key *flow_key)
 525 {
 526         int act_type = nla_type(a);
 527         const struct nlattr *ovs_key = nla_data(a);
 528         int key_type = nla_type(ovs_key);
 529
 530         /* There can be only one key in a action */
 531         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 532                 return -EINVAL;
 533
 534         if (key_type > OVS_KEY_ATTR_MAX ||
 535             nla_len(ovs_key) != ovs_key_lens[key_type])
 536                 return -EINVAL;
 537
 538 #define ACTION(act, key)        ((act << 8) | key)
 539
 540         switch(ACTION(act_type, key_type)) {
 541         const struct ovs_key_ipv4 *ipv4_key;
 542         const struct ovs_key_8021q *q_key;
 543
 544         case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_TUN_ID):
 545         case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_ETHERNET):
 546                         break;
 547
 548         case ACTION(OVS_ACTION_ATTR_PUSH, OVS_KEY_ATTR_8021Q):
 549                 q_key = nla_data(ovs_key);
 550                 if (q_key->q_tpid != htons(ETH_P_8021Q))
 551                         return -EINVAL;
 552
 553                 if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
 554                         return -EINVAL;
 555                 break;
 556
 557         case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_IPV4):
 558                 if (flow_key->eth.type != htons(ETH_P_IP))
 559                         return -EINVAL;
 560
 561                 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
 562                         return -EINVAL;
 563
 564                 ipv4_key = nla_data(ovs_key);
 565                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 566                         return -EINVAL;
 567
 568                 if (ipv4_key->ipv4_tos & INET_ECN_MASK)
 569                         return -EINVAL;
 570                 break;
 571
 572         case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_TCP):
 573                 if (flow_key->ip.proto != IPPROTO_TCP)
 574                         return -EINVAL;
 575
 576                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 577                         return -EINVAL;
 578
 579                 break;
 580
 581         case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_UDP):
 582                 if (flow_key->ip.proto != IPPROTO_UDP)
 583                         return -EINVAL;
 584
 585                 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
 586                         return -EINVAL;
 587                 break;
 588
 589         default:
 590                 return -EINVAL;
 591         }
 592 #undef ACTION
 593         return 0;
 594 }
 595
 596 static int validate_userspace(const struct nlattr *attr)
 597 {
 598         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =
 599         {
 600                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 601                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
 602         };
 603         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 604         int error;
 605
 606         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX, attr, userspace_policy);
 607         if (error)
 608                 return error;
 609
 610         if (!a[OVS_USERSPACE_ATTR_PID] || !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 611                 return -EINVAL;
 612
 613         return 0;
 614 }
 615
 616 static int validate_actions(const struct nlattr *attr,
 617                                 const struct sw_flow_key *key,  int depth)
 618 {
 619         const struct nlattr *a;
 620         int rem, err;
 621
 622         if (depth >= SAMPLE_ACTION_DEPTH)
 623                 return -EOVERFLOW;
 624
 625         nla_for_each_nested(a, attr, rem) {
 626                 /* Expected argument lengths, (u32)-1 for variable length. */
 627                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 628                         [OVS_ACTION_ATTR_OUTPUT] = 4,
 629                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 630                         [OVS_ACTION_ATTR_PUSH] = (u32)-1,
 631                         [OVS_ACTION_ATTR_POP] = 2,
 632                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 633                         [OVS_ACTION_ATTR_SET_PRIORITY] = 4,
 634                         [OVS_ACTION_ATTR_POP_PRIORITY] = 0,
 635                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 636                 };
 637                 int type = nla_type(a);
 638
 639                 if (type > OVS_ACTION_ATTR_MAX ||
 640                     (action_lens[type] != nla_len(a) &&
 641                      action_lens[type] != (u32)-1))
 642                         return -EINVAL;
 643
 644                 switch (type) {
 645                 case OVS_ACTION_ATTR_UNSPEC:
 646                         return -EINVAL;
 647
 648                 case OVS_ACTION_ATTR_SET_PRIORITY:
 649                 case OVS_ACTION_ATTR_POP_PRIORITY:
 650                         /* No validation needed. */
 651                         break;
 652
 653                 case OVS_ACTION_ATTR_USERSPACE:
 654                         err = validate_userspace(a);
 655                         if (err)
 656                                 return err;
 657                         break;
 658
 659                 case OVS_ACTION_ATTR_OUTPUT:
 660                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 661                                 return -EINVAL;
 662                         break;
 663
 664
 665                 case OVS_ACTION_ATTR_POP:
 666                         if (nla_get_u16(a) != OVS_KEY_ATTR_8021Q)
 667                                 return -EINVAL;
 668                         break;
 669
 670                 case OVS_ACTION_ATTR_SET:
 671                 case OVS_ACTION_ATTR_PUSH:
 672                         err = validate_action_key(a, key);
 673                         if (err)
 674                                 return err;
 675                         break;
 676
 677                 case OVS_ACTION_ATTR_SAMPLE:
 678                         err = validate_sample(a, key, depth);
 679                         if (err)
 680                                 return err;
 681                         break;
 682
 683                 default:
 684                         return -EINVAL;
 685                 }
 686         }
 687
 688         if (rem > 0)
 689                 return -EINVAL;
 690
 691         return 0;
 692 }
 693
 694 static void clear_stats(struct sw_flow *flow)
 695 {
 696         flow->used = 0;
 697         flow->tcp_flags = 0;
 698         flow->packet_count = 0;
 699         flow->byte_count = 0;
 700 }
 701
 702 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 703 {
 704         struct ovs_header *ovs_header = info->userhdr;
 705         struct nlattr **a = info->attrs;
 706         struct sw_flow_actions *acts;
 707         struct sk_buff *packet;
 708         struct sw_flow *flow;
 709         struct datapath *dp;
 710         struct ethhdr *eth;
 711         bool is_frag;
 712         int len;
 713         int err;
 714         int key_len;
 715
 716         err = -EINVAL;
 717         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 718             !a[OVS_PACKET_ATTR_ACTIONS] ||
 719             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 720                 goto err;
 721
 722         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 723         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 724         err = -ENOMEM;
 725         if (!packet)
 726                 goto err;
 727         skb_reserve(packet, NET_IP_ALIGN);
 728
 729         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 730
 731         skb_reset_mac_header(packet);
 732         eth = eth_hdr(packet);
 733
 734         /* Normally, setting the skb 'protocol' field would be handled by a
 735          * call to eth_type_trans(), but it assumes there's a sending
 736          * device, which we may not have. */
 737         if (ntohs(eth->h_proto) >= 1536)
 738                 packet->protocol = eth->h_proto;
 739         else
 740                 packet->protocol = htons(ETH_P_802_2);
 741
 742         /* Build an sw_flow for sending this packet. */
 743         flow = flow_alloc();
 744         err = PTR_ERR(flow);
 745         if (IS_ERR(flow))
 746                 goto err_kfree_skb;
 747
 748         err = flow_extract(packet, -1, &flow->key, &key_len, &is_frag);
 749         if (err)
 750                 goto err_flow_put;
 751
 752         err = flow_metadata_from_nlattrs(&flow->key.eth.in_port,
 753                                          &flow->key.eth.tun_id,
 754                                          a[OVS_PACKET_ATTR_KEY]);
 755         if (err)
 756                 goto err_flow_put;
 757
 758         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
 759         if (err)
 760                 goto err_flow_put;
 761
 762         flow->hash = flow_hash(&flow->key, key_len);
 763
 764         acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 765         err = PTR_ERR(acts);
 766         if (IS_ERR(acts))
 767                 goto err_flow_put;
 768         rcu_assign_pointer(flow->sf_acts, acts);
 769
 770         OVS_CB(packet)->flow = flow;
 771
 772         rcu_read_lock();
 773         dp = get_dp(ovs_header->dp_ifindex);
 774         err = -ENODEV;
 775         if (!dp)
 776                 goto err_unlock;
 777
 778         if (flow->key.eth.in_port < DP_MAX_PORTS)
 779                 OVS_CB(packet)->vport = get_vport_protected(dp,
 780                                                         flow->key.eth.in_port);
 781
 782         local_bh_disable();
 783         err = execute_actions(dp, packet);
 784         local_bh_enable();
 785         rcu_read_unlock();
 786
 787         flow_put(flow);
 788         return err;
 789
 790 err_unlock:
 791         rcu_read_unlock();
 792 err_flow_put:
 793         flow_put(flow);
 794 err_kfree_skb:
 795         kfree_skb(packet);
 796 err:
 797         return err;
 798 }
 799
 800 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 801         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 802         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 803         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 804 };
 805
 806 static struct genl_ops dp_packet_genl_ops[] = {
 807         { .cmd = OVS_PACKET_CMD_EXECUTE,
 808           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 809           .policy = packet_policy,
 810           .doit = ovs_packet_cmd_execute
 811         }
 812 };
 813
 814 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 815 {
 816         int i;
 817         struct flow_table *table = get_table_protected(dp);
 818
 819         stats->n_flows = flow_tbl_count(table);
 820
 821         stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
 822         for_each_possible_cpu(i) {
 823                 const struct dp_stats_percpu *percpu_stats;
 824                 struct dp_stats_percpu local_stats;
 825                 unsigned seqcount;
 826
 827                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 828
 829                 do {
 830                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 831                         local_stats = *percpu_stats;
 832                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 833
 834                 stats->n_frags += local_stats.n_frags;
 835                 stats->n_hit += local_stats.n_hit;
 836                 stats->n_missed += local_stats.n_missed;
 837                 stats->n_lost += local_stats.n_lost;
 838         }
 839 }
 840
 841 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 842         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 843         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 844         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 845 };
 846
 847 static struct genl_family dp_flow_genl_family = {
 848         .id = GENL_ID_GENERATE,
 849         .hdrsize = sizeof(struct ovs_header),
 850         .name = OVS_FLOW_FAMILY,
 851         .version = 1,
 852         .maxattr = OVS_FLOW_ATTR_MAX
 853 };
 854
 855 static struct genl_multicast_group dp_flow_multicast_group = {
 856         .name = OVS_FLOW_MCGROUP
 857 };
 858
 859 /* Called with genl_lock. */
 860 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 861                                   struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
 862 {
 863         const int skb_orig_len = skb->len;
 864         const struct sw_flow_actions *sf_acts;
 865         struct ovs_flow_stats stats;
 866         struct ovs_header *ovs_header;
 867         struct nlattr *nla;
 868         unsigned long used;
 869         u8 tcp_flags;
 870         int err;
 871
 872         sf_acts = rcu_dereference_protected(flow->sf_acts,
 873                                             lockdep_genl_is_held());
 874
 875         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 876         if (!ovs_header)
 877                 return -EMSGSIZE;
 878
 879         ovs_header->dp_ifindex = get_dpifindex(dp);
 880
 881         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 882         if (!nla)
 883                 goto nla_put_failure;
 884         err = flow_to_nlattrs(&flow->key, skb);
 885         if (err)
 886                 goto error;
 887         nla_nest_end(skb, nla);
 888
 889         spin_lock_bh(&flow->lock);
 890         used = flow->used;
 891         stats.n_packets = flow->packet_count;
 892         stats.n_bytes = flow->byte_count;
 893         tcp_flags = flow->tcp_flags;
 894         spin_unlock_bh(&flow->lock);
 895
 896         if (used)
 897                 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
 898
 899         if (stats.n_packets)
 900                 NLA_PUT(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats);
 901
 902         if (tcp_flags)
 903                 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 904
 905         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 906          * this is the first flow to be dumped into 'skb'.  This is unusual for
 907          * Netlink but individual action lists can be longer than
 908          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 909          * The userspace caller can always fetch the actions separately if it
 910          * really wants them.  (Most userspace callers in fact don't care.)
 911          *
 912          * This can only fail for dump operations because the skb is always
 913          * properly sized for single flows.
 914          */
 915         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 916                       sf_acts->actions);
 917         if (err < 0 && skb_orig_len)
 918                 goto error;
 919
 920         return genlmsg_end(skb, ovs_header);
 921
 922 nla_put_failure:
 923         err = -EMSGSIZE;
 924 error:
 925         genlmsg_cancel(skb, ovs_header);
 926         return err;
 927 }
 928
 929 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 930 {
 931         const struct sw_flow_actions *sf_acts;
 932         int len;
 933
 934         sf_acts = rcu_dereference_protected(flow->sf_acts,
 935                                             lockdep_genl_is_held());
 936
 937         len = nla_total_size(FLOW_BUFSIZE); /* OVS_FLOW_ATTR_KEY */
 938         len += nla_total_size(sf_acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 939         len += nla_total_size(sizeof(struct ovs_flow_stats)); /* OVS_FLOW_ATTR_STATS */
 940         len += nla_total_size(1); /* OVS_FLOW_ATTR_TCP_FLAGS */
 941         len += nla_total_size(8); /* OVS_FLOW_ATTR_USED */
 942         return genlmsg_new(NLMSG_ALIGN(sizeof(struct ovs_header)) + len, GFP_KERNEL);
 943 }
 944
 945 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
 946                                                u32 pid, u32 seq, u8 cmd)
 947 {
 948         struct sk_buff *skb;
 949         int retval;
 950
 951         skb = ovs_flow_cmd_alloc_info(flow);
 952         if (!skb)
 953                 return ERR_PTR(-ENOMEM);
 954
 955         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 956         BUG_ON(retval < 0);
 957         return skb;
 958 }
 959
 960 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 961 {
 962         struct nlattr **a = info->attrs;
 963         struct ovs_header *ovs_header = info->userhdr;
 964         struct sw_flow_key key;
 965         struct sw_flow *flow;
 966         struct sk_buff *reply;
 967         struct datapath *dp;
 968         struct flow_table *table;
 969         int error;
 970         int key_len;
 971
 972         /* Extract key. */
 973         error = -EINVAL;
 974         if (!a[OVS_FLOW_ATTR_KEY])
 975                 goto error;
 976         error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
 977         if (error)
 978                 goto error;
 979
 980         /* Validate actions. */
 981         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 982                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
 983                 if (error)
 984                         goto error;
 985         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 986                 error = -EINVAL;
 987                 goto error;
 988         }
 989
 990         dp = get_dp(ovs_header->dp_ifindex);
 991         error = -ENODEV;
 992         if (!dp)
 993                 goto error;
 994
 995         table = get_table_protected(dp);
 996         flow = flow_tbl_lookup(table, &key, key_len);
 997         if (!flow) {
 998                 struct sw_flow_actions *acts;
 999
1000                 /* Bail out if we're not allowed to create a new flow. */
1001                 error = -ENOENT;
1002                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1003                         goto error;
1004
1005                 /* Expand table, if necessary, to make room. */
1006                 if (flow_tbl_need_to_expand(table)) {
1007                         struct flow_table *new_table;
1008
1009                         new_table = flow_tbl_expand(table);
1010                         if (!IS_ERR(new_table)) {
1011                                 rcu_assign_pointer(dp->table, new_table);
1012                                 flow_tbl_deferred_destroy(table);
1013                                 table = get_table_protected(dp);
1014                         }
1015                 }
1016
1017                 /* Allocate flow. */
1018                 flow = flow_alloc();
1019                 if (IS_ERR(flow)) {
1020                         error = PTR_ERR(flow);
1021                         goto error;
1022                 }
1023                 flow->key = key;
1024                 clear_stats(flow);
1025
1026                 /* Obtain actions. */
1027                 acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1028                 error = PTR_ERR(acts);
1029                 if (IS_ERR(acts))
1030                         goto error_free_flow;
1031                 rcu_assign_pointer(flow->sf_acts, acts);
1032
1033                 /* Put flow in bucket. */
1034                 flow->hash = flow_hash(&key, key_len);
1035                 flow_tbl_insert(table, flow);
1036
1037                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1038                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
1039         } else {
1040                 /* We found a matching flow. */
1041                 struct sw_flow_actions *old_acts;
1042
1043                 /* Bail out if we're not allowed to modify an existing flow.
1044                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1045                  * because Generic Netlink treats the latter as a dump
1046                  * request.  We also accept NLM_F_EXCL in case that bug ever
1047                  * gets fixed.
1048                  */
1049                 error = -EEXIST;
1050                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1051                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1052                         goto error;
1053
1054                 /* Update actions. */
1055                 old_acts = rcu_dereference_protected(flow->sf_acts,
1056                                                      lockdep_genl_is_held());
1057                 if (a[OVS_FLOW_ATTR_ACTIONS] &&
1058                     (old_acts->actions_len != nla_len(a[OVS_FLOW_ATTR_ACTIONS]) ||
1059                      memcmp(old_acts->actions, nla_data(a[OVS_FLOW_ATTR_ACTIONS]),
1060                             old_acts->actions_len))) {
1061                         struct sw_flow_actions *new_acts;
1062
1063                         new_acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1064                         error = PTR_ERR(new_acts);
1065                         if (IS_ERR(new_acts))
1066                                 goto error;
1067
1068                         rcu_assign_pointer(flow->sf_acts, new_acts);
1069                         flow_deferred_free_acts(old_acts);
1070                 }
1071
1072                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1073                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
1074
1075                 /* Clear stats. */
1076                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1077                         spin_lock_bh(&flow->lock);
1078                         clear_stats(flow);
1079                         spin_unlock_bh(&flow->lock);
1080                 }
1081         }
1082
1083         if (!IS_ERR(reply))
1084                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1085                             dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1086         else
1087                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1088                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1089         return 0;
1090
1091 error_free_flow:
1092         flow_put(flow);
1093 error:
1094         return error;
1095 }
1096
1097 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1098 {
1099         struct nlattr **a = info->attrs;
1100         struct ovs_header *ovs_header = info->userhdr;
1101         struct sw_flow_key key;
1102         struct sk_buff *reply;
1103         struct sw_flow *flow;
1104         struct datapath *dp;
1105         struct flow_table *table;
1106         int err;
1107         int key_len;
1108
1109         if (!a[OVS_FLOW_ATTR_KEY])
1110                 return -EINVAL;
1111         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1112         if (err)
1113                 return err;
1114
1115         dp = get_dp(ovs_header->dp_ifindex);
1116         if (!dp)
1117                 return -ENODEV;
1118
1119         table = get_table_protected(dp);
1120         flow = flow_tbl_lookup(table, &key, key_len);
1121         if (!flow)
1122                 return -ENOENT;
1123
1124         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, OVS_FLOW_CMD_NEW);
1125         if (IS_ERR(reply))
1126                 return PTR_ERR(reply);
1127
1128         return genlmsg_reply(reply, info);
1129 }
1130
1131 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1132 {
1133         struct nlattr **a = info->attrs;
1134         struct ovs_header *ovs_header = info->userhdr;
1135         struct sw_flow_key key;
1136         struct sk_buff *reply;
1137         struct sw_flow *flow;
1138         struct datapath *dp;
1139         struct flow_table *table;
1140         int err;
1141         int key_len;
1142
1143         if (!a[OVS_FLOW_ATTR_KEY])
1144                 return flush_flows(ovs_header->dp_ifindex);
1145         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1146         if (err)
1147                 return err;
1148
1149         dp = get_dp(ovs_header->dp_ifindex);
1150         if (!dp)
1151                 return -ENODEV;
1152
1153         table = get_table_protected(dp);
1154         flow = flow_tbl_lookup(table, &key, key_len);
1155         if (!flow)
1156                 return -ENOENT;
1157
1158         reply = ovs_flow_cmd_alloc_info(flow);
1159         if (!reply)
1160                 return -ENOMEM;
1161
1162         flow_tbl_remove(table, flow);
1163
1164         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1165                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1166         BUG_ON(err < 0);
1167
1168         flow_deferred_free(flow);
1169
1170         genl_notify(reply, genl_info_net(info), info->snd_pid,
1171                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1172         return 0;
1173 }
1174
1175 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1176 {
1177         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1178         struct datapath *dp;
1179
1180         dp = get_dp(ovs_header->dp_ifindex);
1181         if (!dp)
1182                 return -ENODEV;
1183
1184         for (;;) {
1185                 struct sw_flow *flow;
1186                 u32 bucket, obj;
1187
1188                 bucket = cb->args[0];
1189                 obj = cb->args[1];
1190                 flow = flow_tbl_next(get_table_protected(dp), &bucket, &obj);
1191                 if (!flow)
1192                         break;
1193
1194                 if (ovs_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1195                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1196                                            OVS_FLOW_CMD_NEW) < 0)
1197                         break;
1198
1199                 cb->args[0] = bucket;
1200                 cb->args[1] = obj;
1201         }
1202         return skb->len;
1203 }
1204
1205 static struct genl_ops dp_flow_genl_ops[] = {
1206         { .cmd = OVS_FLOW_CMD_NEW,
1207           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1208           .policy = flow_policy,
1209           .doit = ovs_flow_cmd_new_or_set
1210         },
1211         { .cmd = OVS_FLOW_CMD_DEL,
1212           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1213           .policy = flow_policy,
1214           .doit = ovs_flow_cmd_del
1215         },
1216         { .cmd = OVS_FLOW_CMD_GET,
1217           .flags = 0,               /* OK for unprivileged users. */
1218           .policy = flow_policy,
1219           .doit = ovs_flow_cmd_get,
1220           .dumpit = ovs_flow_cmd_dump
1221         },
1222         { .cmd = OVS_FLOW_CMD_SET,
1223           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1224           .policy = flow_policy,
1225           .doit = ovs_flow_cmd_new_or_set,
1226         },
1227 };
1228
1229 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1230 #ifdef HAVE_NLA_NUL_STRING
1231         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1232 #endif
1233         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1234         [OVS_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1235 };
1236
1237 static struct genl_family dp_datapath_genl_family = {
1238         .id = GENL_ID_GENERATE,
1239         .hdrsize = sizeof(struct ovs_header),
1240         .name = OVS_DATAPATH_FAMILY,
1241         .version = 1,
1242         .maxattr = OVS_DP_ATTR_MAX
1243 };
1244
1245 static struct genl_multicast_group dp_datapath_multicast_group = {
1246         .name = OVS_DATAPATH_MCGROUP
1247 };
1248
1249 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1250                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1251 {
1252         struct ovs_header *ovs_header;
1253         struct nlattr *nla;
1254         int err;
1255
1256         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1257                                    flags, cmd);
1258         if (!ovs_header)
1259                 goto error;
1260
1261         ovs_header->dp_ifindex = get_dpifindex(dp);
1262
1263         rcu_read_lock();
1264         err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
1265         rcu_read_unlock();
1266         if (err)
1267                 goto nla_put_failure;
1268
1269         nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats));
1270         if (!nla)
1271                 goto nla_put_failure;
1272         get_dp_stats(dp, nla_data(nla));
1273
1274         NLA_PUT_U32(skb, OVS_DP_ATTR_IPV4_FRAGS,
1275                     dp->drop_frags ? OVS_DP_FRAG_DROP : OVS_DP_FRAG_ZERO);
1276
1277         return genlmsg_end(skb, ovs_header);
1278
1279 nla_put_failure:
1280         genlmsg_cancel(skb, ovs_header);
1281 error:
1282         return -EMSGSIZE;
1283 }
1284
1285 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1286                                              u32 seq, u8 cmd)
1287 {
1288         struct sk_buff *skb;
1289         int retval;
1290
1291         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1292         if (!skb)
1293                 return ERR_PTR(-ENOMEM);
1294
1295         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1296         if (retval < 0) {
1297                 kfree_skb(skb);
1298                 return ERR_PTR(retval);
1299         }
1300         return skb;
1301 }
1302
1303 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1304 {
1305         if (a[OVS_DP_ATTR_IPV4_FRAGS]) {
1306                 u32 frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]);
1307
1308                 if (frags != OVS_DP_FRAG_ZERO && frags != OVS_DP_FRAG_DROP)
1309                         return -EINVAL;
1310         }
1311
1312         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1313 }
1314
1315 /* Called with genl_mutex and optionally with RTNL lock also. */
1316 static struct datapath *lookup_datapath(struct ovs_header *ovs_header, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1317 {
1318         struct datapath *dp;
1319
1320         if (!a[OVS_DP_ATTR_NAME])
1321                 dp = get_dp(ovs_header->dp_ifindex);
1322         else {
1323                 struct vport *vport;
1324
1325                 rcu_read_lock();
1326                 vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1327                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1328                 rcu_read_unlock();
1329         }
1330         return dp ? dp : ERR_PTR(-ENODEV);
1331 }
1332
1333 /* Called with genl_mutex. */
1334 static void change_datapath(struct datapath *dp, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1335 {
1336         if (a[OVS_DP_ATTR_IPV4_FRAGS])
1337                 dp->drop_frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]) == OVS_DP_FRAG_DROP;
1338 }
1339
1340 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1341 {
1342         struct nlattr **a = info->attrs;
1343         struct vport_parms parms;
1344         struct sk_buff *reply;
1345         struct datapath *dp;
1346         struct vport *vport;
1347         int err;
1348
1349         err = -EINVAL;
1350         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1351                 goto err;
1352
1353         err = ovs_dp_cmd_validate(a);
1354         if (err)
1355                 goto err;
1356
1357         rtnl_lock();
1358         err = -ENODEV;
1359         if (!try_module_get(THIS_MODULE))
1360                 goto err_unlock_rtnl;
1361
1362         err = -ENOMEM;
1363         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1364         if (dp == NULL)
1365                 goto err_put_module;
1366         INIT_LIST_HEAD(&dp->port_list);
1367
1368         /* Initialize kobject for bridge.  This will be added as
1369          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1370         dp->ifobj.kset = NULL;
1371         kobject_init(&dp->ifobj, &dp_ktype);
1372
1373         /* Allocate table. */
1374         err = -ENOMEM;
1375         rcu_assign_pointer(dp->table, flow_tbl_alloc(TBL_MIN_BUCKETS));
1376         if (!dp->table)
1377                 goto err_free_dp;
1378
1379         dp->drop_frags = 0;
1380         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1381         if (!dp->stats_percpu) {
1382                 err = -ENOMEM;
1383                 goto err_destroy_table;
1384         }
1385
1386         change_datapath(dp, a);
1387
1388         /* Set up our datapath device. */
1389         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1390         parms.type = OVS_VPORT_TYPE_INTERNAL;
1391         parms.options = NULL;
1392         parms.dp = dp;
1393         parms.port_no = OVSP_LOCAL;
1394         parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1395
1396         vport = new_vport(&parms);
1397         if (IS_ERR(vport)) {
1398                 err = PTR_ERR(vport);
1399                 if (err == -EBUSY)
1400                         err = -EEXIST;
1401
1402                 goto err_destroy_percpu;
1403         }
1404
1405         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1406         err = PTR_ERR(reply);
1407         if (IS_ERR(reply))
1408                 goto err_destroy_local_port;
1409
1410         list_add_tail(&dp->list_node, &dps);
1411         dp_sysfs_add_dp(dp);
1412
1413         rtnl_unlock();
1414
1415         genl_notify(reply, genl_info_net(info), info->snd_pid,
1416                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1417         return 0;
1418
1419 err_destroy_local_port:
1420         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1421 err_destroy_percpu:
1422         free_percpu(dp->stats_percpu);
1423 err_destroy_table:
1424         flow_tbl_destroy(get_table_protected(dp));
1425 err_free_dp:
1426         kfree(dp);
1427 err_put_module:
1428         module_put(THIS_MODULE);
1429 err_unlock_rtnl:
1430         rtnl_unlock();
1431 err:
1432         return err;
1433 }
1434
1435 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1436 {
1437         struct vport *vport, *next_vport;
1438         struct sk_buff *reply;
1439         struct datapath *dp;
1440         int err;
1441
1442         err = ovs_dp_cmd_validate(info->attrs);
1443         if (err)
1444                 goto exit;
1445
1446         rtnl_lock();
1447         dp = lookup_datapath(info->userhdr, info->attrs);
1448         err = PTR_ERR(dp);
1449         if (IS_ERR(dp))
1450                 goto exit_unlock;
1451
1452         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_DEL);
1453         err = PTR_ERR(reply);
1454         if (IS_ERR(reply))
1455                 goto exit_unlock;
1456
1457         list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1458                 if (vport->port_no != OVSP_LOCAL)
1459                         dp_detach_port(vport);
1460
1461         dp_sysfs_del_dp(dp);
1462         list_del(&dp->list_node);
1463         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1464
1465         /* rtnl_unlock() will wait until all the references to devices that
1466          * are pending unregistration have been dropped.  We do it here to
1467          * ensure that any internal devices (which contain DP pointers) are
1468          * fully destroyed before freeing the datapath.
1469          */
1470         rtnl_unlock();
1471
1472         call_rcu(&dp->rcu, destroy_dp_rcu);
1473         module_put(THIS_MODULE);
1474
1475         genl_notify(reply, genl_info_net(info), info->snd_pid,
1476                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1477
1478         return 0;
1479
1480 exit_unlock:
1481         rtnl_unlock();
1482 exit:
1483         return err;
1484 }
1485
1486 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1487 {
1488         struct sk_buff *reply;
1489         struct datapath *dp;
1490         int err;
1491
1492         err = ovs_dp_cmd_validate(info->attrs);
1493         if (err)
1494                 return err;
1495
1496         dp = lookup_datapath(info->userhdr, info->attrs);
1497         if (IS_ERR(dp))
1498                 return PTR_ERR(dp);
1499
1500         change_datapath(dp, info->attrs);
1501
1502         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1503         if (IS_ERR(reply)) {
1504                 err = PTR_ERR(reply);
1505                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1506                                 dp_datapath_multicast_group.id, err);
1507                 return 0;
1508         }
1509
1510         genl_notify(reply, genl_info_net(info), info->snd_pid,
1511                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1512         return 0;
1513 }
1514
1515 static int ovs_dp_cmd_get(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(info->userhdr, info->attrs);
1526         if (IS_ERR(dp))
1527                 return PTR_ERR(dp);
1528
1529         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1530         if (IS_ERR(reply))
1531                 return PTR_ERR(reply);
1532
1533         return genlmsg_reply(reply, info);
1534 }
1535
1536 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1537 {
1538         struct datapath *dp;
1539         int skip = cb->args[0];
1540         int i = 0;
1541
1542         list_for_each_entry (dp, &dps, list_node) {
1543                 if (i < skip)
1544                         continue;
1545                 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1546                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1547                                          OVS_DP_CMD_NEW) < 0)
1548                         break;
1549                 i++;
1550         }
1551
1552         cb->args[0] = i;
1553
1554         return skb->len;
1555 }
1556
1557 static struct genl_ops dp_datapath_genl_ops[] = {
1558         { .cmd = OVS_DP_CMD_NEW,
1559           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1560           .policy = datapath_policy,
1561           .doit = ovs_dp_cmd_new
1562         },
1563         { .cmd = OVS_DP_CMD_DEL,
1564           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1565           .policy = datapath_policy,
1566           .doit = ovs_dp_cmd_del
1567         },
1568         { .cmd = OVS_DP_CMD_GET,
1569           .flags = 0,               /* OK for unprivileged users. */
1570           .policy = datapath_policy,
1571           .doit = ovs_dp_cmd_get,
1572           .dumpit = ovs_dp_cmd_dump
1573         },
1574         { .cmd = OVS_DP_CMD_SET,
1575           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1576           .policy = datapath_policy,
1577           .doit = ovs_dp_cmd_set,
1578         },
1579 };
1580
1581 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1582 #ifdef HAVE_NLA_NUL_STRING
1583         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1584         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1585         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1586 #else
1587         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1588         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1589 #endif
1590         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1591         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1592         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1593         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1594 };
1595
1596 static struct genl_family dp_vport_genl_family = {
1597         .id = GENL_ID_GENERATE,
1598         .hdrsize = sizeof(struct ovs_header),
1599         .name = OVS_VPORT_FAMILY,
1600         .version = 1,
1601         .maxattr = OVS_VPORT_ATTR_MAX
1602 };
1603
1604 struct genl_multicast_group dp_vport_multicast_group = {
1605         .name = OVS_VPORT_MCGROUP
1606 };
1607
1608 /* Called with RTNL lock or RCU read lock. */
1609 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1610                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1611 {
1612         struct ovs_header *ovs_header;
1613         struct nlattr *nla;
1614         int err;
1615
1616         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1617                                  flags, cmd);
1618         if (!ovs_header)
1619                 return -EMSGSIZE;
1620
1621         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1622
1623         NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1624         NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport_get_type(vport));
1625         NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport_get_name(vport));
1626         NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1627
1628         nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats));
1629         if (!nla)
1630                 goto nla_put_failure;
1631
1632         vport_get_stats(vport, nla_data(nla));
1633
1634         NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1635
1636         err = vport_get_options(vport, skb);
1637         if (err == -EMSGSIZE)
1638                 goto error;
1639
1640         return genlmsg_end(skb, ovs_header);
1641
1642 nla_put_failure:
1643         err = -EMSGSIZE;
1644 error:
1645         genlmsg_cancel(skb, ovs_header);
1646         return err;
1647 }
1648
1649 /* Called with RTNL lock or RCU read lock. */
1650 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1651                                          u32 seq, u8 cmd)
1652 {
1653         struct sk_buff *skb;
1654         int retval;
1655
1656         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1657         if (!skb)
1658                 return ERR_PTR(-ENOMEM);
1659
1660         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1661         if (retval < 0) {
1662                 kfree_skb(skb);
1663                 return ERR_PTR(retval);
1664         }
1665         return skb;
1666 }
1667
1668 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1669 {
1670         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1671 }
1672
1673 /* Called with RTNL lock or RCU read lock. */
1674 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1675                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1676 {
1677         struct datapath *dp;
1678         struct vport *vport;
1679
1680         if (a[OVS_VPORT_ATTR_NAME]) {
1681                 vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1682                 if (!vport)
1683                         return ERR_PTR(-ENODEV);
1684                 return vport;
1685         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1686                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1687
1688                 if (port_no >= DP_MAX_PORTS)
1689                         return ERR_PTR(-EFBIG);
1690
1691                 dp = get_dp(ovs_header->dp_ifindex);
1692                 if (!dp)
1693                         return ERR_PTR(-ENODEV);
1694
1695                 vport = get_vport_protected(dp, port_no);
1696                 if (!vport)
1697                         return ERR_PTR(-ENOENT);
1698                 return vport;
1699         } else
1700                 return ERR_PTR(-EINVAL);
1701 }
1702
1703 /* Called with RTNL lock. */
1704 static int change_vport(struct vport *vport, struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1705 {
1706         int err = 0;
1707
1708         if (a[OVS_VPORT_ATTR_STATS])
1709                 vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1710
1711         if (a[OVS_VPORT_ATTR_ADDRESS])
1712                 err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1713
1714         return err;
1715 }
1716
1717 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1718 {
1719         struct nlattr **a = info->attrs;
1720         struct ovs_header *ovs_header = info->userhdr;
1721         struct vport_parms parms;
1722         struct sk_buff *reply;
1723         struct vport *vport;
1724         struct datapath *dp;
1725         u32 port_no;
1726         int err;
1727
1728         err = -EINVAL;
1729         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1730             !a[OVS_VPORT_ATTR_UPCALL_PID])
1731                 goto exit;
1732
1733         err = ovs_vport_cmd_validate(a);
1734         if (err)
1735                 goto exit;
1736
1737         rtnl_lock();
1738         dp = get_dp(ovs_header->dp_ifindex);
1739         err = -ENODEV;
1740         if (!dp)
1741                 goto exit_unlock;
1742
1743         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1744                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1745
1746                 err = -EFBIG;
1747                 if (port_no >= DP_MAX_PORTS)
1748                         goto exit_unlock;
1749
1750                 vport = get_vport_protected(dp, port_no);
1751                 err = -EBUSY;
1752                 if (vport)
1753                         goto exit_unlock;
1754         } else {
1755                 for (port_no = 1; ; port_no++) {
1756                         if (port_no >= DP_MAX_PORTS) {
1757                                 err = -EFBIG;
1758                                 goto exit_unlock;
1759                         }
1760                         vport = get_vport_protected(dp, port_no);
1761                         if (!vport)
1762                                 break;
1763                 }
1764         }
1765
1766         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1767         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1768         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1769         parms.dp = dp;
1770         parms.port_no = port_no;
1771         parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1772
1773         vport = new_vport(&parms);
1774         err = PTR_ERR(vport);
1775         if (IS_ERR(vport))
1776                 goto exit_unlock;
1777
1778         dp_sysfs_add_if(vport);
1779
1780         err = change_vport(vport, a);
1781         if (!err) {
1782                 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1783                                                  info->snd_seq, OVS_VPORT_CMD_NEW);
1784                 if (IS_ERR(reply))
1785                         err = PTR_ERR(reply);
1786         }
1787         if (err) {
1788                 dp_detach_port(vport);
1789                 goto exit_unlock;
1790         }
1791         genl_notify(reply, genl_info_net(info), info->snd_pid,
1792                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1793
1794
1795 exit_unlock:
1796         rtnl_unlock();
1797 exit:
1798         return err;
1799 }
1800
1801 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1802 {
1803         struct nlattr **a = info->attrs;
1804         struct sk_buff *reply;
1805         struct vport *vport;
1806         int err;
1807
1808         err = ovs_vport_cmd_validate(a);
1809         if (err)
1810                 goto exit;
1811
1812         rtnl_lock();
1813         vport = lookup_vport(info->userhdr, a);
1814         err = PTR_ERR(vport);
1815         if (IS_ERR(vport))
1816                 goto exit_unlock;
1817
1818         err = 0;
1819         if (a[OVS_VPORT_ATTR_TYPE] && nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport_get_type(vport))
1820                 err = -EINVAL;
1821         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1822                 err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1823         if (!err)
1824                 err = change_vport(vport, a);
1825         if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1826                 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1827
1828         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1829                                          OVS_VPORT_CMD_NEW);
1830         if (IS_ERR(reply)) {
1831                 err = PTR_ERR(reply);
1832                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1833                                 dp_vport_multicast_group.id, err);
1834                 return 0;
1835         }
1836
1837         genl_notify(reply, genl_info_net(info), info->snd_pid,
1838                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1839
1840 exit_unlock:
1841         rtnl_unlock();
1842 exit:
1843         return err;
1844 }
1845
1846 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1847 {
1848         struct nlattr **a = info->attrs;
1849         struct sk_buff *reply;
1850         struct vport *vport;
1851         int err;
1852
1853         err = ovs_vport_cmd_validate(a);
1854         if (err)
1855                 goto exit;
1856
1857         rtnl_lock();
1858         vport = lookup_vport(info->userhdr, a);
1859         err = PTR_ERR(vport);
1860         if (IS_ERR(vport))
1861                 goto exit_unlock;
1862
1863         if (vport->port_no == OVSP_LOCAL) {
1864                 err = -EINVAL;
1865                 goto exit_unlock;
1866         }
1867
1868         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1869                                          OVS_VPORT_CMD_DEL);
1870         err = PTR_ERR(reply);
1871         if (IS_ERR(reply))
1872                 goto exit_unlock;
1873
1874         dp_detach_port(vport);
1875
1876         genl_notify(reply, genl_info_net(info), info->snd_pid,
1877                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1878
1879 exit_unlock:
1880         rtnl_unlock();
1881 exit:
1882         return err;
1883 }
1884
1885 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1886 {
1887         struct nlattr **a = info->attrs;
1888         struct ovs_header *ovs_header = info->userhdr;
1889         struct sk_buff *reply;
1890         struct vport *vport;
1891         int err;
1892
1893         err = ovs_vport_cmd_validate(a);
1894         if (err)
1895                 goto exit;
1896
1897         rcu_read_lock();
1898         vport = lookup_vport(ovs_header, a);
1899         err = PTR_ERR(vport);
1900         if (IS_ERR(vport))
1901                 goto exit_unlock;
1902
1903         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1904                                          OVS_VPORT_CMD_NEW);
1905         err = PTR_ERR(reply);
1906         if (IS_ERR(reply))
1907                 goto exit_unlock;
1908
1909         rcu_read_unlock();
1910
1911         return genlmsg_reply(reply, info);
1912
1913 exit_unlock:
1914         rcu_read_unlock();
1915 exit:
1916         return err;
1917 }
1918
1919 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1920 {
1921         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1922         struct datapath *dp;
1923         u32 port_no;
1924         int retval;
1925
1926         dp = get_dp(ovs_header->dp_ifindex);
1927         if (!dp)
1928                 return -ENODEV;
1929
1930         rcu_read_lock();
1931         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1932                 struct vport *vport;
1933
1934                 vport = get_vport_protected(dp, port_no);
1935                 if (!vport)
1936                         continue;
1937
1938                 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1939                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1940                                             OVS_VPORT_CMD_NEW) < 0)
1941                         break;
1942         }
1943         rcu_read_unlock();
1944
1945         cb->args[0] = port_no;
1946         retval = skb->len;
1947
1948         return retval;
1949 }
1950
1951 static struct genl_ops dp_vport_genl_ops[] = {
1952         { .cmd = OVS_VPORT_CMD_NEW,
1953           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1954           .policy = vport_policy,
1955           .doit = ovs_vport_cmd_new
1956         },
1957         { .cmd = OVS_VPORT_CMD_DEL,
1958           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1959           .policy = vport_policy,
1960           .doit = ovs_vport_cmd_del
1961         },
1962         { .cmd = OVS_VPORT_CMD_GET,
1963           .flags = 0,               /* OK for unprivileged users. */
1964           .policy = vport_policy,
1965           .doit = ovs_vport_cmd_get,
1966           .dumpit = ovs_vport_cmd_dump
1967         },
1968         { .cmd = OVS_VPORT_CMD_SET,
1969           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1970           .policy = vport_policy,
1971           .doit = ovs_vport_cmd_set,
1972         },
1973 };
1974
1975 struct genl_family_and_ops {
1976         struct genl_family *family;
1977         struct genl_ops *ops;
1978         int n_ops;
1979         struct genl_multicast_group *group;
1980 };
1981
1982 static const struct genl_family_and_ops dp_genl_families[] = {
1983         { &dp_datapath_genl_family,
1984           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1985           &dp_datapath_multicast_group },
1986         { &dp_vport_genl_family,
1987           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1988           &dp_vport_multicast_group },
1989         { &dp_flow_genl_family,
1990           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1991           &dp_flow_multicast_group },
1992         { &dp_packet_genl_family,
1993           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1994           NULL },
1995 };
1996
1997 static void dp_unregister_genl(int n_families)
1998 {
1999         int i;
2000
2001         for (i = 0; i < n_families; i++)
2002                 genl_unregister_family(dp_genl_families[i].family);
2003 }
2004
2005 static int dp_register_genl(void)
2006 {
2007         int n_registered;
2008         int err;
2009         int i;
2010
2011         n_registered = 0;
2012         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2013                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2014
2015                 err = genl_register_family_with_ops(f->family, f->ops,
2016                                                     f->n_ops);
2017                 if (err)
2018                         goto error;
2019                 n_registered++;
2020
2021                 if (f->group) {
2022                         err = genl_register_mc_group(f->family, f->group);
2023                         if (err)
2024                                 goto error;
2025                 }
2026         }
2027
2028         return 0;
2029
2030 error:
2031         dp_unregister_genl(n_registered);
2032         return err;
2033 }
2034
2035 static int __init dp_init(void)
2036 {
2037         struct sk_buff *dummy_skb;
2038         int err;
2039
2040         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2041
2042         printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2043
2044         err = tnl_init();
2045         if (err)
2046                 goto error;
2047
2048         err = flow_init();
2049         if (err)
2050                 goto error_tnl_exit;
2051
2052         err = vport_init();
2053         if (err)
2054                 goto error_flow_exit;
2055
2056         err = register_netdevice_notifier(&dp_device_notifier);
2057         if (err)
2058                 goto error_vport_exit;
2059
2060         err = dp_register_genl();
2061         if (err < 0)
2062                 goto error_unreg_notifier;
2063
2064         return 0;
2065
2066 error_unreg_notifier:
2067         unregister_netdevice_notifier(&dp_device_notifier);
2068 error_vport_exit:
2069         vport_exit();
2070 error_flow_exit:
2071         flow_exit();
2072 error_tnl_exit:
2073         tnl_exit();
2074 error:
2075         return err;
2076 }
2077
2078 static void dp_cleanup(void)
2079 {
2080         rcu_barrier();
2081         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2082         unregister_netdevice_notifier(&dp_device_notifier);
2083         vport_exit();
2084         flow_exit();
2085         tnl_exit();
2086 }
2087
2088 module_init(dp_init);
2089 module_exit(dp_cleanup);
2090
2091 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2092 MODULE_LICENSE("GPL");