Pileus Git - ~andy/linux/blob - net/openvswitch/datapath.c

   1 /*
   2  * Copyright (c) 2007-2013 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/ethtool.h>
  40 #include <linux/wait.h>
  41 #include <asm/div64.h>
  42 #include <linux/highmem.h>
  43 #include <linux/netfilter_bridge.h>
  44 #include <linux/netfilter_ipv4.h>
  45 #include <linux/inetdevice.h>
  46 #include <linux/list.h>
  47 #include <linux/lockdep.h>
  48 #include <linux/openvswitch.h>
  49 #include <linux/rculist.h>
  50 #include <linux/dmi.h>
  51 #include <linux/workqueue.h>
  52 #include <net/genetlink.h>
  53 #include <net/net_namespace.h>
  54 #include <net/netns/generic.h>
  55
  56 #include "datapath.h"
  57 #include "flow.h"
  58 #include "vport-internal_dev.h"
  59 #include "vport-netdev.h"
  60
  61
  62 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
  63 static void rehash_flow_table(struct work_struct *work);
  64 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
  65
  66 int ovs_net_id __read_mostly;
  67
  68 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
  69                        struct genl_multicast_group *grp)
  70 {
  71         genl_notify(skb, genl_info_net(info), info->snd_portid,
  72                     grp->id, info->nlhdr, GFP_KERNEL);
  73 }
  74
  75 /**
  76  * DOC: Locking:
  77  *
  78  * All writes e.g. Writes to device state (add/remove datapath, port, set
  79  * operations on vports, etc.), Writes to other state (flow table
  80  * modifications, set miscellaneous datapath parameters, etc.) are protected
  81  * by ovs_lock.
  82  *
  83  * Reads are protected by RCU.
  84  *
  85  * There are a few special cases (mostly stats) that have their own
  86  * synchronization but they nest under all of above and don't interact with
  87  * each other.
  88  *
  89  * The RTNL lock nests inside ovs_mutex.
  90  */
  91
  92 static DEFINE_MUTEX(ovs_mutex);
  93
  94 void ovs_lock(void)
  95 {
  96         mutex_lock(&ovs_mutex);
  97 }
  98
  99 void ovs_unlock(void)
 100 {
 101         mutex_unlock(&ovs_mutex);
 102 }
 103
 104 #ifdef CONFIG_LOCKDEP
 105 int lockdep_ovsl_is_held(void)
 106 {
 107         if (debug_locks)
 108                 return lockdep_is_held(&ovs_mutex);
 109         else
 110                 return 1;
 111 }
 112 #endif
 113
 114 static struct vport *new_vport(const struct vport_parms *);
 115 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
 116                              const struct dp_upcall_info *);
 117 static int queue_userspace_packet(struct net *, int dp_ifindex,
 118                                   struct sk_buff *,
 119                                   const struct dp_upcall_info *);
 120
 121 /* Must be called with rcu_read_lock or ovs_mutex. */
 122 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 123 {
 124         struct datapath *dp = NULL;
 125         struct net_device *dev;
 126
 127         rcu_read_lock();
 128         dev = dev_get_by_index_rcu(net, dp_ifindex);
 129         if (dev) {
 130                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 131                 if (vport)
 132                         dp = vport->dp;
 133         }
 134         rcu_read_unlock();
 135
 136         return dp;
 137 }
 138
 139 /* Must be called with rcu_read_lock or ovs_mutex. */
 140 const char *ovs_dp_name(const struct datapath *dp)
 141 {
 142         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 143         return vport->ops->get_name(vport);
 144 }
 145
 146 static int get_dpifindex(struct datapath *dp)
 147 {
 148         struct vport *local;
 149         int ifindex;
 150
 151         rcu_read_lock();
 152
 153         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 154         if (local)
 155                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 156         else
 157                 ifindex = 0;
 158
 159         rcu_read_unlock();
 160
 161         return ifindex;
 162 }
 163
 164 static void destroy_dp_rcu(struct rcu_head *rcu)
 165 {
 166         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 167
 168         ovs_flow_tbl_destroy((__force struct flow_table *)dp->table, false);
 169         free_percpu(dp->stats_percpu);
 170         release_net(ovs_dp_get_net(dp));
 171         kfree(dp->ports);
 172         kfree(dp);
 173 }
 174
 175 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 176                                             u16 port_no)
 177 {
 178         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 179 }
 180
 181 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 182 {
 183         struct vport *vport;
 184         struct hlist_head *head;
 185
 186         head = vport_hash_bucket(dp, port_no);
 187         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 188                 if (vport->port_no == port_no)
 189                         return vport;
 190         }
 191         return NULL;
 192 }
 193
 194 /* Called with ovs_mutex. */
 195 static struct vport *new_vport(const struct vport_parms *parms)
 196 {
 197         struct vport *vport;
 198
 199         vport = ovs_vport_add(parms);
 200         if (!IS_ERR(vport)) {
 201                 struct datapath *dp = parms->dp;
 202                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 203
 204                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 205         }
 206         return vport;
 207 }
 208
 209 void ovs_dp_detach_port(struct vport *p)
 210 {
 211         ASSERT_OVSL();
 212
 213         /* First drop references to device. */
 214         hlist_del_rcu(&p->dp_hash_node);
 215
 216         /* Then destroy it. */
 217         ovs_vport_del(p);
 218 }
 219
 220 /* Must be called with rcu_read_lock. */
 221 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 222 {
 223         struct datapath *dp = p->dp;
 224         struct sw_flow *flow;
 225         struct dp_stats_percpu *stats;
 226         struct sw_flow_key key;
 227         u64 *stats_counter;
 228         int error;
 229
 230         stats = this_cpu_ptr(dp->stats_percpu);
 231
 232         /* Extract flow from 'skb' into 'key'. */
 233         error = ovs_flow_extract(skb, p->port_no, &key);
 234         if (unlikely(error)) {
 235                 kfree_skb(skb);
 236                 return;
 237         }
 238
 239         /* Look up flow. */
 240         flow = ovs_flow_lookup(rcu_dereference(dp->table), &key);
 241         if (unlikely(!flow)) {
 242                 struct dp_upcall_info upcall;
 243
 244                 upcall.cmd = OVS_PACKET_CMD_MISS;
 245                 upcall.key = &key;
 246                 upcall.userdata = NULL;
 247                 upcall.portid = p->upcall_portid;
 248                 ovs_dp_upcall(dp, skb, &upcall);
 249                 consume_skb(skb);
 250                 stats_counter = &stats->n_missed;
 251                 goto out;
 252         }
 253
 254         OVS_CB(skb)->flow = flow;
 255         OVS_CB(skb)->pkt_key = &key;
 256
 257         stats_counter = &stats->n_hit;
 258         ovs_flow_used(OVS_CB(skb)->flow, skb);
 259         ovs_execute_actions(dp, skb);
 260
 261 out:
 262         /* Update datapath statistics. */
 263         u64_stats_update_begin(&stats->sync);
 264         (*stats_counter)++;
 265         u64_stats_update_end(&stats->sync);
 266 }
 267
 268 static struct genl_family dp_packet_genl_family = {
 269         .id = GENL_ID_GENERATE,
 270         .hdrsize = sizeof(struct ovs_header),
 271         .name = OVS_PACKET_FAMILY,
 272         .version = OVS_PACKET_VERSION,
 273         .maxattr = OVS_PACKET_ATTR_MAX,
 274         .netnsok = true,
 275         .parallel_ops = true,
 276 };
 277
 278 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 279                   const struct dp_upcall_info *upcall_info)
 280 {
 281         struct dp_stats_percpu *stats;
 282         int dp_ifindex;
 283         int err;
 284
 285         if (upcall_info->portid == 0) {
 286                 err = -ENOTCONN;
 287                 goto err;
 288         }
 289
 290         dp_ifindex = get_dpifindex(dp);
 291         if (!dp_ifindex) {
 292                 err = -ENODEV;
 293                 goto err;
 294         }
 295
 296         if (!skb_is_gso(skb))
 297                 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 298         else
 299                 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 300         if (err)
 301                 goto err;
 302
 303         return 0;
 304
 305 err:
 306         stats = this_cpu_ptr(dp->stats_percpu);
 307
 308         u64_stats_update_begin(&stats->sync);
 309         stats->n_lost++;
 310         u64_stats_update_end(&stats->sync);
 311
 312         return err;
 313 }
 314
 315 static int queue_gso_packets(struct net *net, int dp_ifindex,
 316                              struct sk_buff *skb,
 317                              const struct dp_upcall_info *upcall_info)
 318 {
 319         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 320         struct dp_upcall_info later_info;
 321         struct sw_flow_key later_key;
 322         struct sk_buff *segs, *nskb;
 323         int err;
 324
 325         segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
 326         if (IS_ERR(segs))
 327                 return PTR_ERR(segs);
 328
 329         /* Queue all of the segments. */
 330         skb = segs;
 331         do {
 332                 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
 333                 if (err)
 334                         break;
 335
 336                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 337                         /* The initial flow key extracted by ovs_flow_extract()
 338                          * in this case is for a first fragment, so we need to
 339                          * properly mark later fragments.
 340                          */
 341                         later_key = *upcall_info->key;
 342                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 343
 344                         later_info = *upcall_info;
 345                         later_info.key = &later_key;
 346                         upcall_info = &later_info;
 347                 }
 348         } while ((skb = skb->next));
 349
 350         /* Free all of the segments. */
 351         skb = segs;
 352         do {
 353                 nskb = skb->next;
 354                 if (err)
 355                         kfree_skb(skb);
 356                 else
 357                         consume_skb(skb);
 358         } while ((skb = nskb));
 359         return err;
 360 }
 361
 362 static size_t key_attr_size(void)
 363 {
 364         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 365                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 366                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
 367                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 368                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 369                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
 370                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
 371                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 372                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
 373                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 374                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 375                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 376                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 377                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 378                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 379                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 380                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 381                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 382                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 383 }
 384
 385 static size_t upcall_msg_size(const struct sk_buff *skb,
 386                               const struct nlattr *userdata)
 387 {
 388         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 389                 + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
 390                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 391
 392         /* OVS_PACKET_ATTR_USERDATA */
 393         if (userdata)
 394                 size += NLA_ALIGN(userdata->nla_len);
 395
 396         return size;
 397 }
 398
 399 static int queue_userspace_packet(struct net *net, int dp_ifindex,
 400                                   struct sk_buff *skb,
 401                                   const struct dp_upcall_info *upcall_info)
 402 {
 403         struct ovs_header *upcall;
 404         struct sk_buff *nskb = NULL;
 405         struct sk_buff *user_skb; /* to be queued to userspace */
 406         struct nlattr *nla;
 407         int err;
 408
 409         if (vlan_tx_tag_present(skb)) {
 410                 nskb = skb_clone(skb, GFP_ATOMIC);
 411                 if (!nskb)
 412                         return -ENOMEM;
 413
 414                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 415                 if (!nskb)
 416                         return -ENOMEM;
 417
 418                 nskb->vlan_tci = 0;
 419                 skb = nskb;
 420         }
 421
 422         if (nla_attr_size(skb->len) > USHRT_MAX) {
 423                 err = -EFBIG;
 424                 goto out;
 425         }
 426
 427         user_skb = genlmsg_new(upcall_msg_size(skb, upcall_info->userdata), GFP_ATOMIC);
 428         if (!user_skb) {
 429                 err = -ENOMEM;
 430                 goto out;
 431         }
 432
 433         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 434                              0, upcall_info->cmd);
 435         upcall->dp_ifindex = dp_ifindex;
 436
 437         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 438         ovs_flow_to_nlattrs(upcall_info->key, upcall_info->key, user_skb);
 439         nla_nest_end(user_skb, nla);
 440
 441         if (upcall_info->userdata)
 442                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 443                           nla_len(upcall_info->userdata),
 444                           nla_data(upcall_info->userdata));
 445
 446         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 447
 448         skb_copy_and_csum_dev(skb, nla_data(nla));
 449
 450         genlmsg_end(user_skb, upcall);
 451         err = genlmsg_unicast(net, user_skb, upcall_info->portid);
 452
 453 out:
 454         kfree_skb(nskb);
 455         return err;
 456 }
 457
 458 /* Called with ovs_mutex. */
 459 static int flush_flows(struct datapath *dp)
 460 {
 461         struct flow_table *old_table;
 462         struct flow_table *new_table;
 463
 464         old_table = ovsl_dereference(dp->table);
 465         new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
 466         if (!new_table)
 467                 return -ENOMEM;
 468
 469         rcu_assign_pointer(dp->table, new_table);
 470
 471         ovs_flow_tbl_destroy(old_table, true);
 472         return 0;
 473 }
 474
 475 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa, int attr_len)
 476 {
 477
 478         struct sw_flow_actions *acts;
 479         int new_acts_size;
 480         int req_size = NLA_ALIGN(attr_len);
 481         int next_offset = offsetof(struct sw_flow_actions, actions) +
 482                                         (*sfa)->actions_len;
 483
 484         if (req_size <= (ksize(*sfa) - next_offset))
 485                 goto out;
 486
 487         new_acts_size = ksize(*sfa) * 2;
 488
 489         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
 490                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
 491                         return ERR_PTR(-EMSGSIZE);
 492                 new_acts_size = MAX_ACTIONS_BUFSIZE;
 493         }
 494
 495         acts = ovs_flow_actions_alloc(new_acts_size);
 496         if (IS_ERR(acts))
 497                 return (void *)acts;
 498
 499         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
 500         acts->actions_len = (*sfa)->actions_len;
 501         kfree(*sfa);
 502         *sfa = acts;
 503
 504 out:
 505         (*sfa)->actions_len += req_size;
 506         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
 507 }
 508
 509 static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, int len)
 510 {
 511         struct nlattr *a;
 512
 513         a = reserve_sfa_size(sfa, nla_attr_size(len));
 514         if (IS_ERR(a))
 515                 return PTR_ERR(a);
 516
 517         a->nla_type = attrtype;
 518         a->nla_len = nla_attr_size(len);
 519
 520         if (data)
 521                 memcpy(nla_data(a), data, len);
 522         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
 523
 524         return 0;
 525 }
 526
 527 static inline int add_nested_action_start(struct sw_flow_actions **sfa, int attrtype)
 528 {
 529         int used = (*sfa)->actions_len;
 530         int err;
 531
 532         err = add_action(sfa, attrtype, NULL, 0);
 533         if (err)
 534                 return err;
 535
 536         return used;
 537 }
 538
 539 static inline void add_nested_action_end(struct sw_flow_actions *sfa, int st_offset)
 540 {
 541         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions + st_offset);
 542
 543         a->nla_len = sfa->actions_len - st_offset;
 544 }
 545
 546 static int validate_and_copy_actions(const struct nlattr *attr,
 547                                      const struct sw_flow_key *key, int depth,
 548                                      struct sw_flow_actions **sfa);
 549
 550 static int validate_and_copy_sample(const struct nlattr *attr,
 551                                     const struct sw_flow_key *key, int depth,
 552                                     struct sw_flow_actions **sfa)
 553 {
 554         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 555         const struct nlattr *probability, *actions;
 556         const struct nlattr *a;
 557         int rem, start, err, st_acts;
 558
 559         memset(attrs, 0, sizeof(attrs));
 560         nla_for_each_nested(a, attr, rem) {
 561                 int type = nla_type(a);
 562                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 563                         return -EINVAL;
 564                 attrs[type] = a;
 565         }
 566         if (rem)
 567                 return -EINVAL;
 568
 569         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 570         if (!probability || nla_len(probability) != sizeof(u32))
 571                 return -EINVAL;
 572
 573         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 574         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 575                 return -EINVAL;
 576
 577         /* validation done, copy sample action. */
 578         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
 579         if (start < 0)
 580                 return start;
 581         err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY, nla_data(probability), sizeof(u32));
 582         if (err)
 583                 return err;
 584         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
 585         if (st_acts < 0)
 586                 return st_acts;
 587
 588         err = validate_and_copy_actions(actions, key, depth + 1, sfa);
 589         if (err)
 590                 return err;
 591
 592         add_nested_action_end(*sfa, st_acts);
 593         add_nested_action_end(*sfa, start);
 594
 595         return 0;
 596 }
 597
 598 static int validate_tp_port(const struct sw_flow_key *flow_key)
 599 {
 600         if (flow_key->eth.type == htons(ETH_P_IP)) {
 601                 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
 602                         return 0;
 603         } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
 604                 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
 605                         return 0;
 606         }
 607
 608         return -EINVAL;
 609 }
 610
 611 static int validate_and_copy_set_tun(const struct nlattr *attr,
 612                                      struct sw_flow_actions **sfa)
 613 {
 614         struct sw_flow_match match;
 615         struct sw_flow_key key;
 616         int err, start;
 617
 618         ovs_match_init(&match, &key, NULL);
 619         err = ovs_ipv4_tun_from_nlattr(nla_data(attr), &match, false);
 620         if (err)
 621                 return err;
 622
 623         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
 624         if (start < 0)
 625                 return start;
 626
 627         err = add_action(sfa, OVS_KEY_ATTR_IPV4_TUNNEL, &match.key->tun_key,
 628                         sizeof(match.key->tun_key));
 629         add_nested_action_end(*sfa, start);
 630
 631         return err;
 632 }
 633
 634 static int validate_set(const struct nlattr *a,
 635                         const struct sw_flow_key *flow_key,
 636                         struct sw_flow_actions **sfa,
 637                         bool *set_tun)
 638 {
 639         const struct nlattr *ovs_key = nla_data(a);
 640         int key_type = nla_type(ovs_key);
 641
 642         /* There can be only one key in a action */
 643         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 644                 return -EINVAL;
 645
 646         if (key_type > OVS_KEY_ATTR_MAX ||
 647            (ovs_key_lens[key_type] != nla_len(ovs_key) &&
 648             ovs_key_lens[key_type] != -1))
 649                 return -EINVAL;
 650
 651         switch (key_type) {
 652         const struct ovs_key_ipv4 *ipv4_key;
 653         const struct ovs_key_ipv6 *ipv6_key;
 654         int err;
 655
 656         case OVS_KEY_ATTR_PRIORITY:
 657         case OVS_KEY_ATTR_SKB_MARK:
 658         case OVS_KEY_ATTR_ETHERNET:
 659                 break;
 660
 661         case OVS_KEY_ATTR_TUNNEL:
 662                 *set_tun = true;
 663                 err = validate_and_copy_set_tun(a, sfa);
 664                 if (err)
 665                         return err;
 666                 break;
 667
 668         case OVS_KEY_ATTR_IPV4:
 669                 if (flow_key->eth.type != htons(ETH_P_IP))
 670                         return -EINVAL;
 671
 672                 if (!flow_key->ip.proto)
 673                         return -EINVAL;
 674
 675                 ipv4_key = nla_data(ovs_key);
 676                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 677                         return -EINVAL;
 678
 679                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 680                         return -EINVAL;
 681
 682                 break;
 683
 684         case OVS_KEY_ATTR_IPV6:
 685                 if (flow_key->eth.type != htons(ETH_P_IPV6))
 686                         return -EINVAL;
 687
 688                 if (!flow_key->ip.proto)
 689                         return -EINVAL;
 690
 691                 ipv6_key = nla_data(ovs_key);
 692                 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
 693                         return -EINVAL;
 694
 695                 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
 696                         return -EINVAL;
 697
 698                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
 699                         return -EINVAL;
 700
 701                 break;
 702
 703         case OVS_KEY_ATTR_TCP:
 704                 if (flow_key->ip.proto != IPPROTO_TCP)
 705                         return -EINVAL;
 706
 707                 return validate_tp_port(flow_key);
 708
 709         case OVS_KEY_ATTR_UDP:
 710                 if (flow_key->ip.proto != IPPROTO_UDP)
 711                         return -EINVAL;
 712
 713                 return validate_tp_port(flow_key);
 714
 715         case OVS_KEY_ATTR_SCTP:
 716                 if (flow_key->ip.proto != IPPROTO_SCTP)
 717                         return -EINVAL;
 718
 719                 return validate_tp_port(flow_key);
 720
 721         default:
 722                 return -EINVAL;
 723         }
 724
 725         return 0;
 726 }
 727
 728 static int validate_userspace(const struct nlattr *attr)
 729 {
 730         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 731                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 732                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
 733         };
 734         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 735         int error;
 736
 737         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 738                                  attr, userspace_policy);
 739         if (error)
 740                 return error;
 741
 742         if (!a[OVS_USERSPACE_ATTR_PID] ||
 743             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 744                 return -EINVAL;
 745
 746         return 0;
 747 }
 748
 749 static int copy_action(const struct nlattr *from,
 750                        struct sw_flow_actions **sfa)
 751 {
 752         int totlen = NLA_ALIGN(from->nla_len);
 753         struct nlattr *to;
 754
 755         to = reserve_sfa_size(sfa, from->nla_len);
 756         if (IS_ERR(to))
 757                 return PTR_ERR(to);
 758
 759         memcpy(to, from, totlen);
 760         return 0;
 761 }
 762
 763 static int validate_and_copy_actions(const struct nlattr *attr,
 764                                      const struct sw_flow_key *key,
 765                                      int depth,
 766                                      struct sw_flow_actions **sfa)
 767 {
 768         const struct nlattr *a;
 769         int rem, err;
 770
 771         if (depth >= SAMPLE_ACTION_DEPTH)
 772                 return -EOVERFLOW;
 773
 774         nla_for_each_nested(a, attr, rem) {
 775                 /* Expected argument lengths, (u32)-1 for variable length. */
 776                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 777                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 778                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 779                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 780                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 781                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 782                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 783                 };
 784                 const struct ovs_action_push_vlan *vlan;
 785                 int type = nla_type(a);
 786                 bool skip_copy;
 787
 788                 if (type > OVS_ACTION_ATTR_MAX ||
 789                     (action_lens[type] != nla_len(a) &&
 790                      action_lens[type] != (u32)-1))
 791                         return -EINVAL;
 792
 793                 skip_copy = false;
 794                 switch (type) {
 795                 case OVS_ACTION_ATTR_UNSPEC:
 796                         return -EINVAL;
 797
 798                 case OVS_ACTION_ATTR_USERSPACE:
 799                         err = validate_userspace(a);
 800                         if (err)
 801                                 return err;
 802                         break;
 803
 804                 case OVS_ACTION_ATTR_OUTPUT:
 805                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 806                                 return -EINVAL;
 807                         break;
 808
 809
 810                 case OVS_ACTION_ATTR_POP_VLAN:
 811                         break;
 812
 813                 case OVS_ACTION_ATTR_PUSH_VLAN:
 814                         vlan = nla_data(a);
 815                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 816                                 return -EINVAL;
 817                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 818                                 return -EINVAL;
 819                         break;
 820
 821                 case OVS_ACTION_ATTR_SET:
 822                         err = validate_set(a, key, sfa, &skip_copy);
 823                         if (err)
 824                                 return err;
 825                         break;
 826
 827                 case OVS_ACTION_ATTR_SAMPLE:
 828                         err = validate_and_copy_sample(a, key, depth, sfa);
 829                         if (err)
 830                                 return err;
 831                         skip_copy = true;
 832                         break;
 833
 834                 default:
 835                         return -EINVAL;
 836                 }
 837                 if (!skip_copy) {
 838                         err = copy_action(a, sfa);
 839                         if (err)
 840                                 return err;
 841                 }
 842         }
 843
 844         if (rem > 0)
 845                 return -EINVAL;
 846
 847         return 0;
 848 }
 849
 850 static void clear_stats(struct sw_flow *flow)
 851 {
 852         flow->used = 0;
 853         flow->tcp_flags = 0;
 854         flow->packet_count = 0;
 855         flow->byte_count = 0;
 856 }
 857
 858 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 859 {
 860         struct ovs_header *ovs_header = info->userhdr;
 861         struct nlattr **a = info->attrs;
 862         struct sw_flow_actions *acts;
 863         struct sk_buff *packet;
 864         struct sw_flow *flow;
 865         struct datapath *dp;
 866         struct ethhdr *eth;
 867         int len;
 868         int err;
 869
 870         err = -EINVAL;
 871         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 872             !a[OVS_PACKET_ATTR_ACTIONS])
 873                 goto err;
 874
 875         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 876         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 877         err = -ENOMEM;
 878         if (!packet)
 879                 goto err;
 880         skb_reserve(packet, NET_IP_ALIGN);
 881
 882         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 883
 884         skb_reset_mac_header(packet);
 885         eth = eth_hdr(packet);
 886
 887         /* Normally, setting the skb 'protocol' field would be handled by a
 888          * call to eth_type_trans(), but it assumes there's a sending
 889          * device, which we may not have. */
 890         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 891                 packet->protocol = eth->h_proto;
 892         else
 893                 packet->protocol = htons(ETH_P_802_2);
 894
 895         /* Build an sw_flow for sending this packet. */
 896         flow = ovs_flow_alloc();
 897         err = PTR_ERR(flow);
 898         if (IS_ERR(flow))
 899                 goto err_kfree_skb;
 900
 901         err = ovs_flow_extract(packet, -1, &flow->key);
 902         if (err)
 903                 goto err_flow_free;
 904
 905         err = ovs_flow_metadata_from_nlattrs(flow, a[OVS_PACKET_ATTR_KEY]);
 906         if (err)
 907                 goto err_flow_free;
 908         acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 909         err = PTR_ERR(acts);
 910         if (IS_ERR(acts))
 911                 goto err_flow_free;
 912
 913         err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
 914         rcu_assign_pointer(flow->sf_acts, acts);
 915         if (err)
 916                 goto err_flow_free;
 917
 918         OVS_CB(packet)->flow = flow;
 919         OVS_CB(packet)->pkt_key = &flow->key;
 920         packet->priority = flow->key.phy.priority;
 921         packet->mark = flow->key.phy.skb_mark;
 922
 923         rcu_read_lock();
 924         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 925         err = -ENODEV;
 926         if (!dp)
 927                 goto err_unlock;
 928
 929         local_bh_disable();
 930         err = ovs_execute_actions(dp, packet);
 931         local_bh_enable();
 932         rcu_read_unlock();
 933
 934         ovs_flow_free(flow, false);
 935         return err;
 936
 937 err_unlock:
 938         rcu_read_unlock();
 939 err_flow_free:
 940         ovs_flow_free(flow, false);
 941 err_kfree_skb:
 942         kfree_skb(packet);
 943 err:
 944         return err;
 945 }
 946
 947 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 948         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 949         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 950         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 951 };
 952
 953 static struct genl_ops dp_packet_genl_ops[] = {
 954         { .cmd = OVS_PACKET_CMD_EXECUTE,
 955           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 956           .policy = packet_policy,
 957           .doit = ovs_packet_cmd_execute
 958         }
 959 };
 960
 961 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 962 {
 963         struct flow_table *table;
 964         int i;
 965
 966         table = rcu_dereference_check(dp->table, lockdep_ovsl_is_held());
 967         stats->n_flows = ovs_flow_tbl_count(table);
 968
 969         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 970         for_each_possible_cpu(i) {
 971                 const struct dp_stats_percpu *percpu_stats;
 972                 struct dp_stats_percpu local_stats;
 973                 unsigned int start;
 974
 975                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 976
 977                 do {
 978                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 979                         local_stats = *percpu_stats;
 980                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 981
 982                 stats->n_hit += local_stats.n_hit;
 983                 stats->n_missed += local_stats.n_missed;
 984                 stats->n_lost += local_stats.n_lost;
 985         }
 986 }
 987
 988 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 989         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 990         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 991         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 992 };
 993
 994 static struct genl_family dp_flow_genl_family = {
 995         .id = GENL_ID_GENERATE,
 996         .hdrsize = sizeof(struct ovs_header),
 997         .name = OVS_FLOW_FAMILY,
 998         .version = OVS_FLOW_VERSION,
 999         .maxattr = OVS_FLOW_ATTR_MAX,
1000         .netnsok = true,
1001         .parallel_ops = true,
1002 };
1003
1004 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
1005         .name = OVS_FLOW_MCGROUP
1006 };
1007
1008 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
1009 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1010 {
1011         const struct nlattr *a;
1012         struct nlattr *start;
1013         int err = 0, rem;
1014
1015         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1016         if (!start)
1017                 return -EMSGSIZE;
1018
1019         nla_for_each_nested(a, attr, rem) {
1020                 int type = nla_type(a);
1021                 struct nlattr *st_sample;
1022
1023                 switch (type) {
1024                 case OVS_SAMPLE_ATTR_PROBABILITY:
1025                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
1026                                 return -EMSGSIZE;
1027                         break;
1028                 case OVS_SAMPLE_ATTR_ACTIONS:
1029                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1030                         if (!st_sample)
1031                                 return -EMSGSIZE;
1032                         err = actions_to_attr(nla_data(a), nla_len(a), skb);
1033                         if (err)
1034                                 return err;
1035                         nla_nest_end(skb, st_sample);
1036                         break;
1037                 }
1038         }
1039
1040         nla_nest_end(skb, start);
1041         return err;
1042 }
1043
1044 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1045 {
1046         const struct nlattr *ovs_key = nla_data(a);
1047         int key_type = nla_type(ovs_key);
1048         struct nlattr *start;
1049         int err;
1050
1051         switch (key_type) {
1052         case OVS_KEY_ATTR_IPV4_TUNNEL:
1053                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1054                 if (!start)
1055                         return -EMSGSIZE;
1056
1057                 err = ovs_ipv4_tun_to_nlattr(skb, nla_data(ovs_key),
1058                                              nla_data(ovs_key));
1059                 if (err)
1060                         return err;
1061                 nla_nest_end(skb, start);
1062                 break;
1063         default:
1064                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1065                         return -EMSGSIZE;
1066                 break;
1067         }
1068
1069         return 0;
1070 }
1071
1072 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1073 {
1074         const struct nlattr *a;
1075         int rem, err;
1076
1077         nla_for_each_attr(a, attr, len, rem) {
1078                 int type = nla_type(a);
1079
1080                 switch (type) {
1081                 case OVS_ACTION_ATTR_SET:
1082                         err = set_action_to_attr(a, skb);
1083                         if (err)
1084                                 return err;
1085                         break;
1086
1087                 case OVS_ACTION_ATTR_SAMPLE:
1088                         err = sample_action_to_attr(a, skb);
1089                         if (err)
1090                                 return err;
1091                         break;
1092                 default:
1093                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
1094                                 return -EMSGSIZE;
1095                         break;
1096                 }
1097         }
1098
1099         return 0;
1100 }
1101
1102 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
1103 {
1104         return NLMSG_ALIGN(sizeof(struct ovs_header))
1105                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
1106                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
1107                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
1108                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
1109                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
1110                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
1111 }
1112
1113 /* Called with ovs_mutex. */
1114 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1115                                   struct sk_buff *skb, u32 portid,
1116                                   u32 seq, u32 flags, u8 cmd)
1117 {
1118         const int skb_orig_len = skb->len;
1119         struct nlattr *start;
1120         struct ovs_flow_stats stats;
1121         struct ovs_header *ovs_header;
1122         struct nlattr *nla;
1123         unsigned long used;
1124         u8 tcp_flags;
1125         int err;
1126
1127         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1128         if (!ovs_header)
1129                 return -EMSGSIZE;
1130
1131         ovs_header->dp_ifindex = get_dpifindex(dp);
1132
1133         /* Fill flow key. */
1134         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1135         if (!nla)
1136                 goto nla_put_failure;
1137
1138         err = ovs_flow_to_nlattrs(&flow->unmasked_key,
1139                         &flow->unmasked_key, skb);
1140         if (err)
1141                 goto error;
1142         nla_nest_end(skb, nla);
1143
1144         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
1145         if (!nla)
1146                 goto nla_put_failure;
1147
1148         err = ovs_flow_to_nlattrs(&flow->key, &flow->mask->key, skb);
1149         if (err)
1150                 goto error;
1151
1152         nla_nest_end(skb, nla);
1153
1154         spin_lock_bh(&flow->lock);
1155         used = flow->used;
1156         stats.n_packets = flow->packet_count;
1157         stats.n_bytes = flow->byte_count;
1158         tcp_flags = flow->tcp_flags;
1159         spin_unlock_bh(&flow->lock);
1160
1161         if (used &&
1162             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1163                 goto nla_put_failure;
1164
1165         if (stats.n_packets &&
1166             nla_put(skb, OVS_FLOW_ATTR_STATS,
1167                     sizeof(struct ovs_flow_stats), &stats))
1168                 goto nla_put_failure;
1169
1170         if (tcp_flags &&
1171             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1172                 goto nla_put_failure;
1173
1174         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1175          * this is the first flow to be dumped into 'skb'.  This is unusual for
1176          * Netlink but individual action lists can be longer than
1177          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1178          * The userspace caller can always fetch the actions separately if it
1179          * really wants them.  (Most userspace callers in fact don't care.)
1180          *
1181          * This can only fail for dump operations because the skb is always
1182          * properly sized for single flows.
1183          */
1184         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1185         if (start) {
1186                 const struct sw_flow_actions *sf_acts;
1187
1188                 sf_acts = rcu_dereference_check(flow->sf_acts,
1189                                                 lockdep_ovsl_is_held());
1190
1191                 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1192                 if (!err)
1193                         nla_nest_end(skb, start);
1194                 else {
1195                         if (skb_orig_len)
1196                                 goto error;
1197
1198                         nla_nest_cancel(skb, start);
1199                 }
1200         } else if (skb_orig_len)
1201                 goto nla_put_failure;
1202
1203         return genlmsg_end(skb, ovs_header);
1204
1205 nla_put_failure:
1206         err = -EMSGSIZE;
1207 error:
1208         genlmsg_cancel(skb, ovs_header);
1209         return err;
1210 }
1211
1212 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1213 {
1214         const struct sw_flow_actions *sf_acts;
1215
1216         sf_acts = ovsl_dereference(flow->sf_acts);
1217
1218         return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
1219 }
1220
1221 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1222                                                struct datapath *dp,
1223                                                u32 portid, u32 seq, u8 cmd)
1224 {
1225         struct sk_buff *skb;
1226         int retval;
1227
1228         skb = ovs_flow_cmd_alloc_info(flow);
1229         if (!skb)
1230                 return ERR_PTR(-ENOMEM);
1231
1232         retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1233         BUG_ON(retval < 0);
1234         return skb;
1235 }
1236
1237 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1238 {
1239         struct nlattr **a = info->attrs;
1240         struct ovs_header *ovs_header = info->userhdr;
1241         struct sw_flow_key key, masked_key;
1242         struct sw_flow *flow = NULL;
1243         struct sw_flow_mask mask;
1244         struct sk_buff *reply;
1245         struct datapath *dp;
1246         struct flow_table *table;
1247         struct sw_flow_actions *acts = NULL;
1248         struct sw_flow_match match;
1249         int error;
1250
1251         /* Extract key. */
1252         error = -EINVAL;
1253         if (!a[OVS_FLOW_ATTR_KEY])
1254                 goto error;
1255
1256         ovs_match_init(&match, &key, &mask);
1257         error = ovs_match_from_nlattrs(&match,
1258                         a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
1259         if (error)
1260                 goto error;
1261
1262         /* Validate actions. */
1263         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1264                 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1265                 error = PTR_ERR(acts);
1266                 if (IS_ERR(acts))
1267                         goto error;
1268
1269                 ovs_flow_key_mask(&masked_key, &key, &mask);
1270                 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
1271                                                   &masked_key, 0, &acts);
1272                 if (error) {
1273                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
1274                         goto err_kfree;
1275                 }
1276         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1277                 error = -EINVAL;
1278                 goto error;
1279         }
1280
1281         ovs_lock();
1282         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1283         error = -ENODEV;
1284         if (!dp)
1285                 goto err_unlock_ovs;
1286
1287         table = ovsl_dereference(dp->table);
1288
1289         /* Check if this is a duplicate flow */
1290         flow = ovs_flow_lookup(table, &key);
1291         if (!flow) {
1292                 struct sw_flow_mask *mask_p;
1293                 /* Bail out if we're not allowed to create a new flow. */
1294                 error = -ENOENT;
1295                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1296                         goto err_unlock_ovs;
1297
1298                 /* Expand table, if necessary, to make room. */
1299                 if (ovs_flow_tbl_need_to_expand(table)) {
1300                         struct flow_table *new_table;
1301
1302                         new_table = ovs_flow_tbl_expand(table);
1303                         if (!IS_ERR(new_table)) {
1304                                 rcu_assign_pointer(dp->table, new_table);
1305                                 ovs_flow_tbl_destroy(table, true);
1306                                 table = ovsl_dereference(dp->table);
1307                         }
1308                 }
1309
1310                 /* Allocate flow. */
1311                 flow = ovs_flow_alloc();
1312                 if (IS_ERR(flow)) {
1313                         error = PTR_ERR(flow);
1314                         goto err_unlock_ovs;
1315                 }
1316                 clear_stats(flow);
1317
1318                 flow->key = masked_key;
1319                 flow->unmasked_key = key;
1320
1321                 /* Make sure mask is unique in the system */
1322                 mask_p = ovs_sw_flow_mask_find(table, &mask);
1323                 if (!mask_p) {
1324                         /* Allocate a new mask if none exsits. */
1325                         mask_p = ovs_sw_flow_mask_alloc();
1326                         if (!mask_p)
1327                                 goto err_flow_free;
1328                         mask_p->key = mask.key;
1329                         mask_p->range = mask.range;
1330                         ovs_sw_flow_mask_insert(table, mask_p);
1331                 }
1332
1333                 ovs_sw_flow_mask_add_ref(mask_p);
1334                 flow->mask = mask_p;
1335                 rcu_assign_pointer(flow->sf_acts, acts);
1336
1337                 /* Put flow in bucket. */
1338                 ovs_flow_insert(table, flow);
1339
1340                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1341                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
1342         } else {
1343                 /* We found a matching flow. */
1344                 struct sw_flow_actions *old_acts;
1345
1346                 /* Bail out if we're not allowed to modify an existing flow.
1347                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1348                  * because Generic Netlink treats the latter as a dump
1349                  * request.  We also accept NLM_F_EXCL in case that bug ever
1350                  * gets fixed.
1351                  */
1352                 error = -EEXIST;
1353                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1354                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1355                         goto err_unlock_ovs;
1356
1357                 /* The unmasked key has to be the same for flow updates. */
1358                 error = -EINVAL;
1359                 if (!ovs_flow_cmp_unmasked_key(flow, &key, match.range.end)) {
1360                         OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
1361                         goto err_unlock_ovs;
1362                 }
1363
1364                 /* Update actions. */
1365                 old_acts = ovsl_dereference(flow->sf_acts);
1366                 rcu_assign_pointer(flow->sf_acts, acts);
1367                 ovs_flow_deferred_free_acts(old_acts);
1368
1369                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1370                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1371
1372                 /* Clear stats. */
1373                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1374                         spin_lock_bh(&flow->lock);
1375                         clear_stats(flow);
1376                         spin_unlock_bh(&flow->lock);
1377                 }
1378         }
1379         ovs_unlock();
1380
1381         if (!IS_ERR(reply))
1382                 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1383         else
1384                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1385                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1386         return 0;
1387
1388 err_flow_free:
1389         ovs_flow_free(flow, false);
1390 err_unlock_ovs:
1391         ovs_unlock();
1392 err_kfree:
1393         kfree(acts);
1394 error:
1395         return error;
1396 }
1397
1398 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1399 {
1400         struct nlattr **a = info->attrs;
1401         struct ovs_header *ovs_header = info->userhdr;
1402         struct sw_flow_key key;
1403         struct sk_buff *reply;
1404         struct sw_flow *flow;
1405         struct datapath *dp;
1406         struct flow_table *table;
1407         struct sw_flow_match match;
1408         int err;
1409
1410         if (!a[OVS_FLOW_ATTR_KEY]) {
1411                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1412                 return -EINVAL;
1413         }
1414
1415         ovs_match_init(&match, &key, NULL);
1416         err = ovs_match_from_nlattrs(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1417         if (err)
1418                 return err;
1419
1420         ovs_lock();
1421         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1422         if (!dp) {
1423                 err = -ENODEV;
1424                 goto unlock;
1425         }
1426
1427         table = ovsl_dereference(dp->table);
1428         flow = ovs_flow_lookup_unmasked_key(table, &match);
1429         if (!flow) {
1430                 err = -ENOENT;
1431                 goto unlock;
1432         }
1433
1434         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1435                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1436         if (IS_ERR(reply)) {
1437                 err = PTR_ERR(reply);
1438                 goto unlock;
1439         }
1440
1441         ovs_unlock();
1442         return genlmsg_reply(reply, info);
1443 unlock:
1444         ovs_unlock();
1445         return err;
1446 }
1447
1448 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1449 {
1450         struct nlattr **a = info->attrs;
1451         struct ovs_header *ovs_header = info->userhdr;
1452         struct sw_flow_key key;
1453         struct sk_buff *reply;
1454         struct sw_flow *flow;
1455         struct datapath *dp;
1456         struct flow_table *table;
1457         struct sw_flow_match match;
1458         int err;
1459
1460         ovs_lock();
1461         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1462         if (!dp) {
1463                 err = -ENODEV;
1464                 goto unlock;
1465         }
1466
1467         if (!a[OVS_FLOW_ATTR_KEY]) {
1468                 err = flush_flows(dp);
1469                 goto unlock;
1470         }
1471
1472         ovs_match_init(&match, &key, NULL);
1473         err = ovs_match_from_nlattrs(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1474         if (err)
1475                 goto unlock;
1476
1477         table = ovsl_dereference(dp->table);
1478         flow = ovs_flow_lookup_unmasked_key(table, &match);
1479         if (!flow) {
1480                 err = -ENOENT;
1481                 goto unlock;
1482         }
1483
1484         reply = ovs_flow_cmd_alloc_info(flow);
1485         if (!reply) {
1486                 err = -ENOMEM;
1487                 goto unlock;
1488         }
1489
1490         ovs_flow_remove(table, flow);
1491
1492         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1493                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1494         BUG_ON(err < 0);
1495
1496         ovs_flow_free(flow, true);
1497         ovs_unlock();
1498
1499         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1500         return 0;
1501 unlock:
1502         ovs_unlock();
1503         return err;
1504 }
1505
1506 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1507 {
1508         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1509         struct datapath *dp;
1510         struct flow_table *table;
1511
1512         rcu_read_lock();
1513         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1514         if (!dp) {
1515                 rcu_read_unlock();
1516                 return -ENODEV;
1517         }
1518
1519         table = rcu_dereference(dp->table);
1520         for (;;) {
1521                 struct sw_flow *flow;
1522                 u32 bucket, obj;
1523
1524                 bucket = cb->args[0];
1525                 obj = cb->args[1];
1526                 flow = ovs_flow_dump_next(table, &bucket, &obj);
1527                 if (!flow)
1528                         break;
1529
1530                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1531                                            NETLINK_CB(cb->skb).portid,
1532                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1533                                            OVS_FLOW_CMD_NEW) < 0)
1534                         break;
1535
1536                 cb->args[0] = bucket;
1537                 cb->args[1] = obj;
1538         }
1539         rcu_read_unlock();
1540         return skb->len;
1541 }
1542
1543 static struct genl_ops dp_flow_genl_ops[] = {
1544         { .cmd = OVS_FLOW_CMD_NEW,
1545           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1546           .policy = flow_policy,
1547           .doit = ovs_flow_cmd_new_or_set
1548         },
1549         { .cmd = OVS_FLOW_CMD_DEL,
1550           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1551           .policy = flow_policy,
1552           .doit = ovs_flow_cmd_del
1553         },
1554         { .cmd = OVS_FLOW_CMD_GET,
1555           .flags = 0,               /* OK for unprivileged users. */
1556           .policy = flow_policy,
1557           .doit = ovs_flow_cmd_get,
1558           .dumpit = ovs_flow_cmd_dump
1559         },
1560         { .cmd = OVS_FLOW_CMD_SET,
1561           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1562           .policy = flow_policy,
1563           .doit = ovs_flow_cmd_new_or_set,
1564         },
1565 };
1566
1567 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1568         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1569         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1570 };
1571
1572 static struct genl_family dp_datapath_genl_family = {
1573         .id = GENL_ID_GENERATE,
1574         .hdrsize = sizeof(struct ovs_header),
1575         .name = OVS_DATAPATH_FAMILY,
1576         .version = OVS_DATAPATH_VERSION,
1577         .maxattr = OVS_DP_ATTR_MAX,
1578         .netnsok = true,
1579         .parallel_ops = true,
1580 };
1581
1582 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1583         .name = OVS_DATAPATH_MCGROUP
1584 };
1585
1586 static size_t ovs_dp_cmd_msg_size(void)
1587 {
1588         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1589
1590         msgsize += nla_total_size(IFNAMSIZ);
1591         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1592
1593         return msgsize;
1594 }
1595
1596 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1597                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1598 {
1599         struct ovs_header *ovs_header;
1600         struct ovs_dp_stats dp_stats;
1601         int err;
1602
1603         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1604                                    flags, cmd);
1605         if (!ovs_header)
1606                 goto error;
1607
1608         ovs_header->dp_ifindex = get_dpifindex(dp);
1609
1610         rcu_read_lock();
1611         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1612         rcu_read_unlock();
1613         if (err)
1614                 goto nla_put_failure;
1615
1616         get_dp_stats(dp, &dp_stats);
1617         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1618                 goto nla_put_failure;
1619
1620         return genlmsg_end(skb, ovs_header);
1621
1622 nla_put_failure:
1623         genlmsg_cancel(skb, ovs_header);
1624 error:
1625         return -EMSGSIZE;
1626 }
1627
1628 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1629                                              u32 seq, u8 cmd)
1630 {
1631         struct sk_buff *skb;
1632         int retval;
1633
1634         skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1635         if (!skb)
1636                 return ERR_PTR(-ENOMEM);
1637
1638         retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1639         if (retval < 0) {
1640                 kfree_skb(skb);
1641                 return ERR_PTR(retval);
1642         }
1643         return skb;
1644 }
1645
1646 /* Called with ovs_mutex. */
1647 static struct datapath *lookup_datapath(struct net *net,
1648                                         struct ovs_header *ovs_header,
1649                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1650 {
1651         struct datapath *dp;
1652
1653         if (!a[OVS_DP_ATTR_NAME])
1654                 dp = get_dp(net, ovs_header->dp_ifindex);
1655         else {
1656                 struct vport *vport;
1657
1658                 rcu_read_lock();
1659                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1660                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1661                 rcu_read_unlock();
1662         }
1663         return dp ? dp : ERR_PTR(-ENODEV);
1664 }
1665
1666 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1667 {
1668         struct nlattr **a = info->attrs;
1669         struct vport_parms parms;
1670         struct sk_buff *reply;
1671         struct datapath *dp;
1672         struct vport *vport;
1673         struct ovs_net *ovs_net;
1674         int err, i;
1675
1676         err = -EINVAL;
1677         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1678                 goto err;
1679
1680         ovs_lock();
1681
1682         err = -ENOMEM;
1683         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1684         if (dp == NULL)
1685                 goto err_unlock_ovs;
1686
1687         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1688
1689         /* Allocate table. */
1690         err = -ENOMEM;
1691         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1692         if (!dp->table)
1693                 goto err_free_dp;
1694
1695         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1696         if (!dp->stats_percpu) {
1697                 err = -ENOMEM;
1698                 goto err_destroy_table;
1699         }
1700
1701         for_each_possible_cpu(i) {
1702                 struct dp_stats_percpu *dpath_stats;
1703                 dpath_stats = per_cpu_ptr(dp->stats_percpu, i);
1704                 u64_stats_init(&dpath_stats->sync);
1705         }
1706
1707         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1708                         GFP_KERNEL);
1709         if (!dp->ports) {
1710                 err = -ENOMEM;
1711                 goto err_destroy_percpu;
1712         }
1713
1714         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1715                 INIT_HLIST_HEAD(&dp->ports[i]);
1716
1717         /* Set up our datapath device. */
1718         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1719         parms.type = OVS_VPORT_TYPE_INTERNAL;
1720         parms.options = NULL;
1721         parms.dp = dp;
1722         parms.port_no = OVSP_LOCAL;
1723         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1724
1725         vport = new_vport(&parms);
1726         if (IS_ERR(vport)) {
1727                 err = PTR_ERR(vport);
1728                 if (err == -EBUSY)
1729                         err = -EEXIST;
1730
1731                 goto err_destroy_ports_array;
1732         }
1733
1734         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1735                                       info->snd_seq, OVS_DP_CMD_NEW);
1736         err = PTR_ERR(reply);
1737         if (IS_ERR(reply))
1738                 goto err_destroy_local_port;
1739
1740         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1741         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1742
1743         ovs_unlock();
1744
1745         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1746         return 0;
1747
1748 err_destroy_local_port:
1749         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1750 err_destroy_ports_array:
1751         kfree(dp->ports);
1752 err_destroy_percpu:
1753         free_percpu(dp->stats_percpu);
1754 err_destroy_table:
1755         ovs_flow_tbl_destroy(ovsl_dereference(dp->table), false);
1756 err_free_dp:
1757         release_net(ovs_dp_get_net(dp));
1758         kfree(dp);
1759 err_unlock_ovs:
1760         ovs_unlock();
1761 err:
1762         return err;
1763 }
1764
1765 /* Called with ovs_mutex. */
1766 static void __dp_destroy(struct datapath *dp)
1767 {
1768         int i;
1769
1770         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1771                 struct vport *vport;
1772                 struct hlist_node *n;
1773
1774                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1775                         if (vport->port_no != OVSP_LOCAL)
1776                                 ovs_dp_detach_port(vport);
1777         }
1778
1779         list_del_rcu(&dp->list_node);
1780
1781         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1782          * all port in datapath are destroyed first before freeing datapath.
1783          */
1784         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1785
1786         call_rcu(&dp->rcu, destroy_dp_rcu);
1787 }
1788
1789 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1790 {
1791         struct sk_buff *reply;
1792         struct datapath *dp;
1793         int err;
1794
1795         ovs_lock();
1796         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1797         err = PTR_ERR(dp);
1798         if (IS_ERR(dp))
1799                 goto unlock;
1800
1801         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1802                                       info->snd_seq, OVS_DP_CMD_DEL);
1803         err = PTR_ERR(reply);
1804         if (IS_ERR(reply))
1805                 goto unlock;
1806
1807         __dp_destroy(dp);
1808         ovs_unlock();
1809
1810         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1811
1812         return 0;
1813 unlock:
1814         ovs_unlock();
1815         return err;
1816 }
1817
1818 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1819 {
1820         struct sk_buff *reply;
1821         struct datapath *dp;
1822         int err;
1823
1824         ovs_lock();
1825         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1826         err = PTR_ERR(dp);
1827         if (IS_ERR(dp))
1828                 goto unlock;
1829
1830         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1831                                       info->snd_seq, OVS_DP_CMD_NEW);
1832         if (IS_ERR(reply)) {
1833                 err = PTR_ERR(reply);
1834                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1835                                 ovs_dp_datapath_multicast_group.id, err);
1836                 err = 0;
1837                 goto unlock;
1838         }
1839
1840         ovs_unlock();
1841         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1842
1843         return 0;
1844 unlock:
1845         ovs_unlock();
1846         return err;
1847 }
1848
1849 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1850 {
1851         struct sk_buff *reply;
1852         struct datapath *dp;
1853         int err;
1854
1855         ovs_lock();
1856         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1857         if (IS_ERR(dp)) {
1858                 err = PTR_ERR(dp);
1859                 goto unlock;
1860         }
1861
1862         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1863                                       info->snd_seq, OVS_DP_CMD_NEW);
1864         if (IS_ERR(reply)) {
1865                 err = PTR_ERR(reply);
1866                 goto unlock;
1867         }
1868
1869         ovs_unlock();
1870         return genlmsg_reply(reply, info);
1871
1872 unlock:
1873         ovs_unlock();
1874         return err;
1875 }
1876
1877 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1878 {
1879         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1880         struct datapath *dp;
1881         int skip = cb->args[0];
1882         int i = 0;
1883
1884         rcu_read_lock();
1885         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1886                 if (i >= skip &&
1887                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1888                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1889                                          OVS_DP_CMD_NEW) < 0)
1890                         break;
1891                 i++;
1892         }
1893         rcu_read_unlock();
1894
1895         cb->args[0] = i;
1896
1897         return skb->len;
1898 }
1899
1900 static struct genl_ops dp_datapath_genl_ops[] = {
1901         { .cmd = OVS_DP_CMD_NEW,
1902           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1903           .policy = datapath_policy,
1904           .doit = ovs_dp_cmd_new
1905         },
1906         { .cmd = OVS_DP_CMD_DEL,
1907           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1908           .policy = datapath_policy,
1909           .doit = ovs_dp_cmd_del
1910         },
1911         { .cmd = OVS_DP_CMD_GET,
1912           .flags = 0,               /* OK for unprivileged users. */
1913           .policy = datapath_policy,
1914           .doit = ovs_dp_cmd_get,
1915           .dumpit = ovs_dp_cmd_dump
1916         },
1917         { .cmd = OVS_DP_CMD_SET,
1918           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1919           .policy = datapath_policy,
1920           .doit = ovs_dp_cmd_set,
1921         },
1922 };
1923
1924 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1925         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1926         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1927         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1928         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1929         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1930         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1931 };
1932
1933 static struct genl_family dp_vport_genl_family = {
1934         .id = GENL_ID_GENERATE,
1935         .hdrsize = sizeof(struct ovs_header),
1936         .name = OVS_VPORT_FAMILY,
1937         .version = OVS_VPORT_VERSION,
1938         .maxattr = OVS_VPORT_ATTR_MAX,
1939         .netnsok = true,
1940         .parallel_ops = true,
1941 };
1942
1943 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1944         .name = OVS_VPORT_MCGROUP
1945 };
1946
1947 /* Called with ovs_mutex or RCU read lock. */
1948 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1949                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1950 {
1951         struct ovs_header *ovs_header;
1952         struct ovs_vport_stats vport_stats;
1953         int err;
1954
1955         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1956                                  flags, cmd);
1957         if (!ovs_header)
1958                 return -EMSGSIZE;
1959
1960         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1961
1962         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1963             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1964             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1965             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1966                 goto nla_put_failure;
1967
1968         ovs_vport_get_stats(vport, &vport_stats);
1969         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1970                     &vport_stats))
1971                 goto nla_put_failure;
1972
1973         err = ovs_vport_get_options(vport, skb);
1974         if (err == -EMSGSIZE)
1975                 goto error;
1976
1977         return genlmsg_end(skb, ovs_header);
1978
1979 nla_put_failure:
1980         err = -EMSGSIZE;
1981 error:
1982         genlmsg_cancel(skb, ovs_header);
1983         return err;
1984 }
1985
1986 /* Called with ovs_mutex or RCU read lock. */
1987 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1988                                          u32 seq, u8 cmd)
1989 {
1990         struct sk_buff *skb;
1991         int retval;
1992
1993         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1994         if (!skb)
1995                 return ERR_PTR(-ENOMEM);
1996
1997         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1998         BUG_ON(retval < 0);
1999
2000         return skb;
2001 }
2002
2003 /* Called with ovs_mutex or RCU read lock. */
2004 static struct vport *lookup_vport(struct net *net,
2005                                   struct ovs_header *ovs_header,
2006                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2007 {
2008         struct datapath *dp;
2009         struct vport *vport;
2010
2011         if (a[OVS_VPORT_ATTR_NAME]) {
2012                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2013                 if (!vport)
2014                         return ERR_PTR(-ENODEV);
2015                 if (ovs_header->dp_ifindex &&
2016                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2017                         return ERR_PTR(-ENODEV);
2018                 return vport;
2019         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2020                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2021
2022                 if (port_no >= DP_MAX_PORTS)
2023                         return ERR_PTR(-EFBIG);
2024
2025                 dp = get_dp(net, ovs_header->dp_ifindex);
2026                 if (!dp)
2027                         return ERR_PTR(-ENODEV);
2028
2029                 vport = ovs_vport_ovsl_rcu(dp, port_no);
2030                 if (!vport)
2031                         return ERR_PTR(-ENODEV);
2032                 return vport;
2033         } else
2034                 return ERR_PTR(-EINVAL);
2035 }
2036
2037 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2038 {
2039         struct nlattr **a = info->attrs;
2040         struct ovs_header *ovs_header = info->userhdr;
2041         struct vport_parms parms;
2042         struct sk_buff *reply;
2043         struct vport *vport;
2044         struct datapath *dp;
2045         u32 port_no;
2046         int err;
2047
2048         err = -EINVAL;
2049         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2050             !a[OVS_VPORT_ATTR_UPCALL_PID])
2051                 goto exit;
2052
2053         ovs_lock();
2054         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2055         err = -ENODEV;
2056         if (!dp)
2057                 goto exit_unlock;
2058
2059         if (a[OVS_VPORT_ATTR_PORT_NO]) {
2060                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2061
2062                 err = -EFBIG;
2063                 if (port_no >= DP_MAX_PORTS)
2064                         goto exit_unlock;
2065
2066                 vport = ovs_vport_ovsl(dp, port_no);
2067                 err = -EBUSY;
2068                 if (vport)
2069                         goto exit_unlock;
2070         } else {
2071                 for (port_no = 1; ; port_no++) {
2072                         if (port_no >= DP_MAX_PORTS) {
2073                                 err = -EFBIG;
2074                                 goto exit_unlock;
2075                         }
2076                         vport = ovs_vport_ovsl(dp, port_no);
2077                         if (!vport)
2078                                 break;
2079                 }
2080         }
2081
2082         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2083         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2084         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2085         parms.dp = dp;
2086         parms.port_no = port_no;
2087         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2088
2089         vport = new_vport(&parms);
2090         err = PTR_ERR(vport);
2091         if (IS_ERR(vport))
2092                 goto exit_unlock;
2093
2094         err = 0;
2095         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
2096                                          OVS_VPORT_CMD_NEW);
2097         if (IS_ERR(reply)) {
2098                 err = PTR_ERR(reply);
2099                 ovs_dp_detach_port(vport);
2100                 goto exit_unlock;
2101         }
2102
2103         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2104
2105 exit_unlock:
2106         ovs_unlock();
2107 exit:
2108         return err;
2109 }
2110
2111 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2112 {
2113         struct nlattr **a = info->attrs;
2114         struct sk_buff *reply;
2115         struct vport *vport;
2116         int err;
2117
2118         ovs_lock();
2119         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2120         err = PTR_ERR(vport);
2121         if (IS_ERR(vport))
2122                 goto exit_unlock;
2123
2124         if (a[OVS_VPORT_ATTR_TYPE] &&
2125             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2126                 err = -EINVAL;
2127                 goto exit_unlock;
2128         }
2129
2130         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2131         if (!reply) {
2132                 err = -ENOMEM;
2133                 goto exit_unlock;
2134         }
2135
2136         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2137                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2138                 if (err)
2139                         goto exit_free;
2140         }
2141
2142         if (a[OVS_VPORT_ATTR_UPCALL_PID])
2143                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2144
2145         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2146                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2147         BUG_ON(err < 0);
2148
2149         ovs_unlock();
2150         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2151         return 0;
2152
2153 exit_free:
2154         kfree_skb(reply);
2155 exit_unlock:
2156         ovs_unlock();
2157         return err;
2158 }
2159
2160 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2161 {
2162         struct nlattr **a = info->attrs;
2163         struct sk_buff *reply;
2164         struct vport *vport;
2165         int err;
2166
2167         ovs_lock();
2168         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2169         err = PTR_ERR(vport);
2170         if (IS_ERR(vport))
2171                 goto exit_unlock;
2172
2173         if (vport->port_no == OVSP_LOCAL) {
2174                 err = -EINVAL;
2175                 goto exit_unlock;
2176         }
2177
2178         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2179                                          info->snd_seq, OVS_VPORT_CMD_DEL);
2180         err = PTR_ERR(reply);
2181         if (IS_ERR(reply))
2182                 goto exit_unlock;
2183
2184         err = 0;
2185         ovs_dp_detach_port(vport);
2186
2187         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2188
2189 exit_unlock:
2190         ovs_unlock();
2191         return err;
2192 }
2193
2194 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2195 {
2196         struct nlattr **a = info->attrs;
2197         struct ovs_header *ovs_header = info->userhdr;
2198         struct sk_buff *reply;
2199         struct vport *vport;
2200         int err;
2201
2202         rcu_read_lock();
2203         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2204         err = PTR_ERR(vport);
2205         if (IS_ERR(vport))
2206                 goto exit_unlock;
2207
2208         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2209                                          info->snd_seq, OVS_VPORT_CMD_NEW);
2210         err = PTR_ERR(reply);
2211         if (IS_ERR(reply))
2212                 goto exit_unlock;
2213
2214         rcu_read_unlock();
2215
2216         return genlmsg_reply(reply, info);
2217
2218 exit_unlock:
2219         rcu_read_unlock();
2220         return err;
2221 }
2222
2223 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2224 {
2225         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2226         struct datapath *dp;
2227         int bucket = cb->args[0], skip = cb->args[1];
2228         int i, j = 0;
2229
2230         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2231         if (!dp)
2232                 return -ENODEV;
2233
2234         rcu_read_lock();
2235         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2236                 struct vport *vport;
2237
2238                 j = 0;
2239                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2240                         if (j >= skip &&
2241                             ovs_vport_cmd_fill_info(vport, skb,
2242                                                     NETLINK_CB(cb->skb).portid,
2243                                                     cb->nlh->nlmsg_seq,
2244                                                     NLM_F_MULTI,
2245                                                     OVS_VPORT_CMD_NEW) < 0)
2246                                 goto out;
2247
2248                         j++;
2249                 }
2250                 skip = 0;
2251         }
2252 out:
2253         rcu_read_unlock();
2254
2255         cb->args[0] = i;
2256         cb->args[1] = j;
2257
2258         return skb->len;
2259 }
2260
2261 static struct genl_ops dp_vport_genl_ops[] = {
2262         { .cmd = OVS_VPORT_CMD_NEW,
2263           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2264           .policy = vport_policy,
2265           .doit = ovs_vport_cmd_new
2266         },
2267         { .cmd = OVS_VPORT_CMD_DEL,
2268           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2269           .policy = vport_policy,
2270           .doit = ovs_vport_cmd_del
2271         },
2272         { .cmd = OVS_VPORT_CMD_GET,
2273           .flags = 0,               /* OK for unprivileged users. */
2274           .policy = vport_policy,
2275           .doit = ovs_vport_cmd_get,
2276           .dumpit = ovs_vport_cmd_dump
2277         },
2278         { .cmd = OVS_VPORT_CMD_SET,
2279           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2280           .policy = vport_policy,
2281           .doit = ovs_vport_cmd_set,
2282         },
2283 };
2284
2285 struct genl_family_and_ops {
2286         struct genl_family *family;
2287         struct genl_ops *ops;
2288         int n_ops;
2289         struct genl_multicast_group *group;
2290 };
2291
2292 static const struct genl_family_and_ops dp_genl_families[] = {
2293         { &dp_datapath_genl_family,
2294           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2295           &ovs_dp_datapath_multicast_group },
2296         { &dp_vport_genl_family,
2297           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2298           &ovs_dp_vport_multicast_group },
2299         { &dp_flow_genl_family,
2300           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2301           &ovs_dp_flow_multicast_group },
2302         { &dp_packet_genl_family,
2303           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2304           NULL },
2305 };
2306
2307 static void dp_unregister_genl(int n_families)
2308 {
2309         int i;
2310
2311         for (i = 0; i < n_families; i++)
2312                 genl_unregister_family(dp_genl_families[i].family);
2313 }
2314
2315 static int dp_register_genl(void)
2316 {
2317         int n_registered;
2318         int err;
2319         int i;
2320
2321         n_registered = 0;
2322         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2323                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2324
2325                 err = genl_register_family_with_ops(f->family, f->ops,
2326                                                     f->n_ops);
2327                 if (err)
2328                         goto error;
2329                 n_registered++;
2330
2331                 if (f->group) {
2332                         err = genl_register_mc_group(f->family, f->group);
2333                         if (err)
2334                                 goto error;
2335                 }
2336         }
2337
2338         return 0;
2339
2340 error:
2341         dp_unregister_genl(n_registered);
2342         return err;
2343 }
2344
2345 static void rehash_flow_table(struct work_struct *work)
2346 {
2347         struct datapath *dp;
2348         struct net *net;
2349
2350         ovs_lock();
2351         rtnl_lock();
2352         for_each_net(net) {
2353                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2354
2355                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2356                         struct flow_table *old_table = ovsl_dereference(dp->table);
2357                         struct flow_table *new_table;
2358
2359                         new_table = ovs_flow_tbl_rehash(old_table);
2360                         if (!IS_ERR(new_table)) {
2361                                 rcu_assign_pointer(dp->table, new_table);
2362                                 ovs_flow_tbl_destroy(old_table, true);
2363                         }
2364                 }
2365         }
2366         rtnl_unlock();
2367         ovs_unlock();
2368         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2369 }
2370
2371 static int __net_init ovs_init_net(struct net *net)
2372 {
2373         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2374
2375         INIT_LIST_HEAD(&ovs_net->dps);
2376         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2377         return 0;
2378 }
2379
2380 static void __net_exit ovs_exit_net(struct net *net)
2381 {
2382         struct datapath *dp, *dp_next;
2383         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2384
2385         ovs_lock();
2386         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2387                 __dp_destroy(dp);
2388         ovs_unlock();
2389
2390         cancel_work_sync(&ovs_net->dp_notify_work);
2391 }
2392
2393 static struct pernet_operations ovs_net_ops = {
2394         .init = ovs_init_net,
2395         .exit = ovs_exit_net,
2396         .id   = &ovs_net_id,
2397         .size = sizeof(struct ovs_net),
2398 };
2399
2400 static int __init dp_init(void)
2401 {
2402         int err;
2403
2404         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2405
2406         pr_info("Open vSwitch switching datapath\n");
2407
2408         err = ovs_flow_init();
2409         if (err)
2410                 goto error;
2411
2412         err = ovs_vport_init();
2413         if (err)
2414                 goto error_flow_exit;
2415
2416         err = register_pernet_device(&ovs_net_ops);
2417         if (err)
2418                 goto error_vport_exit;
2419
2420         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2421         if (err)
2422                 goto error_netns_exit;
2423
2424         err = dp_register_genl();
2425         if (err < 0)
2426                 goto error_unreg_notifier;
2427
2428         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2429
2430         return 0;
2431
2432 error_unreg_notifier:
2433         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2434 error_netns_exit:
2435         unregister_pernet_device(&ovs_net_ops);
2436 error_vport_exit:
2437         ovs_vport_exit();
2438 error_flow_exit:
2439         ovs_flow_exit();
2440 error:
2441         return err;
2442 }
2443
2444 static void dp_cleanup(void)
2445 {
2446         cancel_delayed_work_sync(&rehash_flow_wq);
2447         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2448         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2449         unregister_pernet_device(&ovs_net_ops);
2450         rcu_barrier();
2451         ovs_vport_exit();
2452         ovs_flow_exit();
2453 }
2454
2455 module_init(dp_init);
2456 module_exit(dp_cleanup);
2457
2458 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2459 MODULE_LICENSE("GPL");