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         default:
 716                 return -EINVAL;
 717         }
 718
 719         return 0;
 720 }
 721
 722 static int validate_userspace(const struct nlattr *attr)
 723 {
 724         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 725                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 726                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
 727         };
 728         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 729         int error;
 730
 731         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 732                                  attr, userspace_policy);
 733         if (error)
 734                 return error;
 735
 736         if (!a[OVS_USERSPACE_ATTR_PID] ||
 737             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 738                 return -EINVAL;
 739
 740         return 0;
 741 }
 742
 743 static int copy_action(const struct nlattr *from,
 744                        struct sw_flow_actions **sfa)
 745 {
 746         int totlen = NLA_ALIGN(from->nla_len);
 747         struct nlattr *to;
 748
 749         to = reserve_sfa_size(sfa, from->nla_len);
 750         if (IS_ERR(to))
 751                 return PTR_ERR(to);
 752
 753         memcpy(to, from, totlen);
 754         return 0;
 755 }
 756
 757 static int validate_and_copy_actions(const struct nlattr *attr,
 758                                      const struct sw_flow_key *key,
 759                                      int depth,
 760                                      struct sw_flow_actions **sfa)
 761 {
 762         const struct nlattr *a;
 763         int rem, err;
 764
 765         if (depth >= SAMPLE_ACTION_DEPTH)
 766                 return -EOVERFLOW;
 767
 768         nla_for_each_nested(a, attr, rem) {
 769                 /* Expected argument lengths, (u32)-1 for variable length. */
 770                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 771                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 772                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 773                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 774                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 775                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 776                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 777                 };
 778                 const struct ovs_action_push_vlan *vlan;
 779                 int type = nla_type(a);
 780                 bool skip_copy;
 781
 782                 if (type > OVS_ACTION_ATTR_MAX ||
 783                     (action_lens[type] != nla_len(a) &&
 784                      action_lens[type] != (u32)-1))
 785                         return -EINVAL;
 786
 787                 skip_copy = false;
 788                 switch (type) {
 789                 case OVS_ACTION_ATTR_UNSPEC:
 790                         return -EINVAL;
 791
 792                 case OVS_ACTION_ATTR_USERSPACE:
 793                         err = validate_userspace(a);
 794                         if (err)
 795                                 return err;
 796                         break;
 797
 798                 case OVS_ACTION_ATTR_OUTPUT:
 799                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 800                                 return -EINVAL;
 801                         break;
 802
 803
 804                 case OVS_ACTION_ATTR_POP_VLAN:
 805                         break;
 806
 807                 case OVS_ACTION_ATTR_PUSH_VLAN:
 808                         vlan = nla_data(a);
 809                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 810                                 return -EINVAL;
 811                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 812                                 return -EINVAL;
 813                         break;
 814
 815                 case OVS_ACTION_ATTR_SET:
 816                         err = validate_set(a, key, sfa, &skip_copy);
 817                         if (err)
 818                                 return err;
 819                         break;
 820
 821                 case OVS_ACTION_ATTR_SAMPLE:
 822                         err = validate_and_copy_sample(a, key, depth, sfa);
 823                         if (err)
 824                                 return err;
 825                         skip_copy = true;
 826                         break;
 827
 828                 default:
 829                         return -EINVAL;
 830                 }
 831                 if (!skip_copy) {
 832                         err = copy_action(a, sfa);
 833                         if (err)
 834                                 return err;
 835                 }
 836         }
 837
 838         if (rem > 0)
 839                 return -EINVAL;
 840
 841         return 0;
 842 }
 843
 844 static void clear_stats(struct sw_flow *flow)
 845 {
 846         flow->used = 0;
 847         flow->tcp_flags = 0;
 848         flow->packet_count = 0;
 849         flow->byte_count = 0;
 850 }
 851
 852 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 853 {
 854         struct ovs_header *ovs_header = info->userhdr;
 855         struct nlattr **a = info->attrs;
 856         struct sw_flow_actions *acts;
 857         struct sk_buff *packet;
 858         struct sw_flow *flow;
 859         struct datapath *dp;
 860         struct ethhdr *eth;
 861         int len;
 862         int err;
 863
 864         err = -EINVAL;
 865         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 866             !a[OVS_PACKET_ATTR_ACTIONS])
 867                 goto err;
 868
 869         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 870         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 871         err = -ENOMEM;
 872         if (!packet)
 873                 goto err;
 874         skb_reserve(packet, NET_IP_ALIGN);
 875
 876         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 877
 878         skb_reset_mac_header(packet);
 879         eth = eth_hdr(packet);
 880
 881         /* Normally, setting the skb 'protocol' field would be handled by a
 882          * call to eth_type_trans(), but it assumes there's a sending
 883          * device, which we may not have. */
 884         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 885                 packet->protocol = eth->h_proto;
 886         else
 887                 packet->protocol = htons(ETH_P_802_2);
 888
 889         /* Build an sw_flow for sending this packet. */
 890         flow = ovs_flow_alloc();
 891         err = PTR_ERR(flow);
 892         if (IS_ERR(flow))
 893                 goto err_kfree_skb;
 894
 895         err = ovs_flow_extract(packet, -1, &flow->key);
 896         if (err)
 897                 goto err_flow_free;
 898
 899         err = ovs_flow_metadata_from_nlattrs(flow, a[OVS_PACKET_ATTR_KEY]);
 900         if (err)
 901                 goto err_flow_free;
 902         acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 903         err = PTR_ERR(acts);
 904         if (IS_ERR(acts))
 905                 goto err_flow_free;
 906
 907         err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
 908         rcu_assign_pointer(flow->sf_acts, acts);
 909         if (err)
 910                 goto err_flow_free;
 911
 912         OVS_CB(packet)->flow = flow;
 913         OVS_CB(packet)->pkt_key = &flow->key;
 914         packet->priority = flow->key.phy.priority;
 915         packet->mark = flow->key.phy.skb_mark;
 916
 917         rcu_read_lock();
 918         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 919         err = -ENODEV;
 920         if (!dp)
 921                 goto err_unlock;
 922
 923         local_bh_disable();
 924         err = ovs_execute_actions(dp, packet);
 925         local_bh_enable();
 926         rcu_read_unlock();
 927
 928         ovs_flow_free(flow, false);
 929         return err;
 930
 931 err_unlock:
 932         rcu_read_unlock();
 933 err_flow_free:
 934         ovs_flow_free(flow, false);
 935 err_kfree_skb:
 936         kfree_skb(packet);
 937 err:
 938         return err;
 939 }
 940
 941 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 942         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 943         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 944         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 945 };
 946
 947 static struct genl_ops dp_packet_genl_ops[] = {
 948         { .cmd = OVS_PACKET_CMD_EXECUTE,
 949           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 950           .policy = packet_policy,
 951           .doit = ovs_packet_cmd_execute
 952         }
 953 };
 954
 955 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 956 {
 957         struct flow_table *table;
 958         int i;
 959
 960         table = rcu_dereference_check(dp->table, lockdep_ovsl_is_held());
 961         stats->n_flows = ovs_flow_tbl_count(table);
 962
 963         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 964         for_each_possible_cpu(i) {
 965                 const struct dp_stats_percpu *percpu_stats;
 966                 struct dp_stats_percpu local_stats;
 967                 unsigned int start;
 968
 969                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 970
 971                 do {
 972                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 973                         local_stats = *percpu_stats;
 974                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 975
 976                 stats->n_hit += local_stats.n_hit;
 977                 stats->n_missed += local_stats.n_missed;
 978                 stats->n_lost += local_stats.n_lost;
 979         }
 980 }
 981
 982 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 983         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 984         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 985         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 986 };
 987
 988 static struct genl_family dp_flow_genl_family = {
 989         .id = GENL_ID_GENERATE,
 990         .hdrsize = sizeof(struct ovs_header),
 991         .name = OVS_FLOW_FAMILY,
 992         .version = OVS_FLOW_VERSION,
 993         .maxattr = OVS_FLOW_ATTR_MAX,
 994         .netnsok = true,
 995         .parallel_ops = true,
 996 };
 997
 998 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 999         .name = OVS_FLOW_MCGROUP
1000 };
1001
1002 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
1003 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1004 {
1005         const struct nlattr *a;
1006         struct nlattr *start;
1007         int err = 0, rem;
1008
1009         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1010         if (!start)
1011                 return -EMSGSIZE;
1012
1013         nla_for_each_nested(a, attr, rem) {
1014                 int type = nla_type(a);
1015                 struct nlattr *st_sample;
1016
1017                 switch (type) {
1018                 case OVS_SAMPLE_ATTR_PROBABILITY:
1019                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
1020                                 return -EMSGSIZE;
1021                         break;
1022                 case OVS_SAMPLE_ATTR_ACTIONS:
1023                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1024                         if (!st_sample)
1025                                 return -EMSGSIZE;
1026                         err = actions_to_attr(nla_data(a), nla_len(a), skb);
1027                         if (err)
1028                                 return err;
1029                         nla_nest_end(skb, st_sample);
1030                         break;
1031                 }
1032         }
1033
1034         nla_nest_end(skb, start);
1035         return err;
1036 }
1037
1038 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1039 {
1040         const struct nlattr *ovs_key = nla_data(a);
1041         int key_type = nla_type(ovs_key);
1042         struct nlattr *start;
1043         int err;
1044
1045         switch (key_type) {
1046         case OVS_KEY_ATTR_IPV4_TUNNEL:
1047                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1048                 if (!start)
1049                         return -EMSGSIZE;
1050
1051                 err = ovs_ipv4_tun_to_nlattr(skb, nla_data(ovs_key),
1052                                              nla_data(ovs_key));
1053                 if (err)
1054                         return err;
1055                 nla_nest_end(skb, start);
1056                 break;
1057         default:
1058                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1059                         return -EMSGSIZE;
1060                 break;
1061         }
1062
1063         return 0;
1064 }
1065
1066 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1067 {
1068         const struct nlattr *a;
1069         int rem, err;
1070
1071         nla_for_each_attr(a, attr, len, rem) {
1072                 int type = nla_type(a);
1073
1074                 switch (type) {
1075                 case OVS_ACTION_ATTR_SET:
1076                         err = set_action_to_attr(a, skb);
1077                         if (err)
1078                                 return err;
1079                         break;
1080
1081                 case OVS_ACTION_ATTR_SAMPLE:
1082                         err = sample_action_to_attr(a, skb);
1083                         if (err)
1084                                 return err;
1085                         break;
1086                 default:
1087                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
1088                                 return -EMSGSIZE;
1089                         break;
1090                 }
1091         }
1092
1093         return 0;
1094 }
1095
1096 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
1097 {
1098         return NLMSG_ALIGN(sizeof(struct ovs_header))
1099                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
1100                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
1101                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
1102                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
1103                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
1104                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
1105 }
1106
1107 /* Called with ovs_mutex. */
1108 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1109                                   struct sk_buff *skb, u32 portid,
1110                                   u32 seq, u32 flags, u8 cmd)
1111 {
1112         const int skb_orig_len = skb->len;
1113         struct nlattr *start;
1114         struct ovs_flow_stats stats;
1115         struct ovs_header *ovs_header;
1116         struct nlattr *nla;
1117         unsigned long used;
1118         u8 tcp_flags;
1119         int err;
1120
1121         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1122         if (!ovs_header)
1123                 return -EMSGSIZE;
1124
1125         ovs_header->dp_ifindex = get_dpifindex(dp);
1126
1127         /* Fill flow key. */
1128         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1129         if (!nla)
1130                 goto nla_put_failure;
1131
1132         err = ovs_flow_to_nlattrs(&flow->unmasked_key,
1133                         &flow->unmasked_key, skb);
1134         if (err)
1135                 goto error;
1136         nla_nest_end(skb, nla);
1137
1138         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
1139         if (!nla)
1140                 goto nla_put_failure;
1141
1142         err = ovs_flow_to_nlattrs(&flow->key, &flow->mask->key, skb);
1143         if (err)
1144                 goto error;
1145
1146         nla_nest_end(skb, nla);
1147
1148         spin_lock_bh(&flow->lock);
1149         used = flow->used;
1150         stats.n_packets = flow->packet_count;
1151         stats.n_bytes = flow->byte_count;
1152         tcp_flags = flow->tcp_flags;
1153         spin_unlock_bh(&flow->lock);
1154
1155         if (used &&
1156             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1157                 goto nla_put_failure;
1158
1159         if (stats.n_packets &&
1160             nla_put(skb, OVS_FLOW_ATTR_STATS,
1161                     sizeof(struct ovs_flow_stats), &stats))
1162                 goto nla_put_failure;
1163
1164         if (tcp_flags &&
1165             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1166                 goto nla_put_failure;
1167
1168         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1169          * this is the first flow to be dumped into 'skb'.  This is unusual for
1170          * Netlink but individual action lists can be longer than
1171          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1172          * The userspace caller can always fetch the actions separately if it
1173          * really wants them.  (Most userspace callers in fact don't care.)
1174          *
1175          * This can only fail for dump operations because the skb is always
1176          * properly sized for single flows.
1177          */
1178         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1179         if (start) {
1180                 const struct sw_flow_actions *sf_acts;
1181
1182                 sf_acts = rcu_dereference_check(flow->sf_acts,
1183                                                 lockdep_ovsl_is_held());
1184
1185                 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1186                 if (!err)
1187                         nla_nest_end(skb, start);
1188                 else {
1189                         if (skb_orig_len)
1190                                 goto error;
1191
1192                         nla_nest_cancel(skb, start);
1193                 }
1194         } else if (skb_orig_len)
1195                 goto nla_put_failure;
1196
1197         return genlmsg_end(skb, ovs_header);
1198
1199 nla_put_failure:
1200         err = -EMSGSIZE;
1201 error:
1202         genlmsg_cancel(skb, ovs_header);
1203         return err;
1204 }
1205
1206 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1207 {
1208         const struct sw_flow_actions *sf_acts;
1209
1210         sf_acts = ovsl_dereference(flow->sf_acts);
1211
1212         return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
1213 }
1214
1215 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1216                                                struct datapath *dp,
1217                                                u32 portid, u32 seq, u8 cmd)
1218 {
1219         struct sk_buff *skb;
1220         int retval;
1221
1222         skb = ovs_flow_cmd_alloc_info(flow);
1223         if (!skb)
1224                 return ERR_PTR(-ENOMEM);
1225
1226         retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1227         BUG_ON(retval < 0);
1228         return skb;
1229 }
1230
1231 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1232 {
1233         struct nlattr **a = info->attrs;
1234         struct ovs_header *ovs_header = info->userhdr;
1235         struct sw_flow_key key, masked_key;
1236         struct sw_flow *flow = NULL;
1237         struct sw_flow_mask mask;
1238         struct sk_buff *reply;
1239         struct datapath *dp;
1240         struct flow_table *table;
1241         struct sw_flow_actions *acts = NULL;
1242         struct sw_flow_match match;
1243         int error;
1244
1245         /* Extract key. */
1246         error = -EINVAL;
1247         if (!a[OVS_FLOW_ATTR_KEY])
1248                 goto error;
1249
1250         ovs_match_init(&match, &key, &mask);
1251         error = ovs_match_from_nlattrs(&match,
1252                         a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
1253         if (error)
1254                 goto error;
1255
1256         /* Validate actions. */
1257         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1258                 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1259                 error = PTR_ERR(acts);
1260                 if (IS_ERR(acts))
1261                         goto error;
1262
1263                 ovs_flow_key_mask(&masked_key, &key, &mask);
1264                 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
1265                                                   &masked_key, 0, &acts);
1266                 if (error) {
1267                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
1268                         goto err_kfree;
1269                 }
1270         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1271                 error = -EINVAL;
1272                 goto error;
1273         }
1274
1275         ovs_lock();
1276         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1277         error = -ENODEV;
1278         if (!dp)
1279                 goto err_unlock_ovs;
1280
1281         table = ovsl_dereference(dp->table);
1282
1283         /* Check if this is a duplicate flow */
1284         flow = ovs_flow_lookup(table, &key);
1285         if (!flow) {
1286                 struct sw_flow_mask *mask_p;
1287                 /* Bail out if we're not allowed to create a new flow. */
1288                 error = -ENOENT;
1289                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1290                         goto err_unlock_ovs;
1291
1292                 /* Expand table, if necessary, to make room. */
1293                 if (ovs_flow_tbl_need_to_expand(table)) {
1294                         struct flow_table *new_table;
1295
1296                         new_table = ovs_flow_tbl_expand(table);
1297                         if (!IS_ERR(new_table)) {
1298                                 rcu_assign_pointer(dp->table, new_table);
1299                                 ovs_flow_tbl_destroy(table, true);
1300                                 table = ovsl_dereference(dp->table);
1301                         }
1302                 }
1303
1304                 /* Allocate flow. */
1305                 flow = ovs_flow_alloc();
1306                 if (IS_ERR(flow)) {
1307                         error = PTR_ERR(flow);
1308                         goto err_unlock_ovs;
1309                 }
1310                 clear_stats(flow);
1311
1312                 flow->key = masked_key;
1313                 flow->unmasked_key = key;
1314
1315                 /* Make sure mask is unique in the system */
1316                 mask_p = ovs_sw_flow_mask_find(table, &mask);
1317                 if (!mask_p) {
1318                         /* Allocate a new mask if none exsits. */
1319                         mask_p = ovs_sw_flow_mask_alloc();
1320                         if (!mask_p)
1321                                 goto err_flow_free;
1322                         mask_p->key = mask.key;
1323                         mask_p->range = mask.range;
1324                         ovs_sw_flow_mask_insert(table, mask_p);
1325                 }
1326
1327                 ovs_sw_flow_mask_add_ref(mask_p);
1328                 flow->mask = mask_p;
1329                 rcu_assign_pointer(flow->sf_acts, acts);
1330
1331                 /* Put flow in bucket. */
1332                 ovs_flow_insert(table, flow);
1333
1334                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1335                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
1336         } else {
1337                 /* We found a matching flow. */
1338                 struct sw_flow_actions *old_acts;
1339
1340                 /* Bail out if we're not allowed to modify an existing flow.
1341                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1342                  * because Generic Netlink treats the latter as a dump
1343                  * request.  We also accept NLM_F_EXCL in case that bug ever
1344                  * gets fixed.
1345                  */
1346                 error = -EEXIST;
1347                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1348                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1349                         goto err_unlock_ovs;
1350
1351                 /* The unmasked key has to be the same for flow updates. */
1352                 error = -EINVAL;
1353                 if (!ovs_flow_cmp_unmasked_key(flow, &key, match.range.end)) {
1354                         OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
1355                         goto err_unlock_ovs;
1356                 }
1357
1358                 /* Update actions. */
1359                 old_acts = ovsl_dereference(flow->sf_acts);
1360                 rcu_assign_pointer(flow->sf_acts, acts);
1361                 ovs_flow_deferred_free_acts(old_acts);
1362
1363                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1364                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1365
1366                 /* Clear stats. */
1367                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1368                         spin_lock_bh(&flow->lock);
1369                         clear_stats(flow);
1370                         spin_unlock_bh(&flow->lock);
1371                 }
1372         }
1373         ovs_unlock();
1374
1375         if (!IS_ERR(reply))
1376                 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1377         else
1378                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1379                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1380         return 0;
1381
1382 err_flow_free:
1383         ovs_flow_free(flow, false);
1384 err_unlock_ovs:
1385         ovs_unlock();
1386 err_kfree:
1387         kfree(acts);
1388 error:
1389         return error;
1390 }
1391
1392 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1393 {
1394         struct nlattr **a = info->attrs;
1395         struct ovs_header *ovs_header = info->userhdr;
1396         struct sw_flow_key key;
1397         struct sk_buff *reply;
1398         struct sw_flow *flow;
1399         struct datapath *dp;
1400         struct flow_table *table;
1401         struct sw_flow_match match;
1402         int err;
1403
1404         if (!a[OVS_FLOW_ATTR_KEY]) {
1405                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1406                 return -EINVAL;
1407         }
1408
1409         ovs_match_init(&match, &key, NULL);
1410         err = ovs_match_from_nlattrs(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1411         if (err)
1412                 return err;
1413
1414         ovs_lock();
1415         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1416         if (!dp) {
1417                 err = -ENODEV;
1418                 goto unlock;
1419         }
1420
1421         table = ovsl_dereference(dp->table);
1422         flow = ovs_flow_lookup_unmasked_key(table, &match);
1423         if (!flow) {
1424                 err = -ENOENT;
1425                 goto unlock;
1426         }
1427
1428         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1429                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1430         if (IS_ERR(reply)) {
1431                 err = PTR_ERR(reply);
1432                 goto unlock;
1433         }
1434
1435         ovs_unlock();
1436         return genlmsg_reply(reply, info);
1437 unlock:
1438         ovs_unlock();
1439         return err;
1440 }
1441
1442 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1443 {
1444         struct nlattr **a = info->attrs;
1445         struct ovs_header *ovs_header = info->userhdr;
1446         struct sw_flow_key key;
1447         struct sk_buff *reply;
1448         struct sw_flow *flow;
1449         struct datapath *dp;
1450         struct flow_table *table;
1451         struct sw_flow_match match;
1452         int err;
1453
1454         ovs_lock();
1455         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1456         if (!dp) {
1457                 err = -ENODEV;
1458                 goto unlock;
1459         }
1460
1461         if (!a[OVS_FLOW_ATTR_KEY]) {
1462                 err = flush_flows(dp);
1463                 goto unlock;
1464         }
1465
1466         ovs_match_init(&match, &key, NULL);
1467         err = ovs_match_from_nlattrs(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1468         if (err)
1469                 goto unlock;
1470
1471         table = ovsl_dereference(dp->table);
1472         flow = ovs_flow_lookup_unmasked_key(table, &match);
1473         if (!flow) {
1474                 err = -ENOENT;
1475                 goto unlock;
1476         }
1477
1478         reply = ovs_flow_cmd_alloc_info(flow);
1479         if (!reply) {
1480                 err = -ENOMEM;
1481                 goto unlock;
1482         }
1483
1484         ovs_flow_remove(table, flow);
1485
1486         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1487                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1488         BUG_ON(err < 0);
1489
1490         ovs_flow_free(flow, true);
1491         ovs_unlock();
1492
1493         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1494         return 0;
1495 unlock:
1496         ovs_unlock();
1497         return err;
1498 }
1499
1500 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1501 {
1502         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1503         struct datapath *dp;
1504         struct flow_table *table;
1505
1506         rcu_read_lock();
1507         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1508         if (!dp) {
1509                 rcu_read_unlock();
1510                 return -ENODEV;
1511         }
1512
1513         table = rcu_dereference(dp->table);
1514         for (;;) {
1515                 struct sw_flow *flow;
1516                 u32 bucket, obj;
1517
1518                 bucket = cb->args[0];
1519                 obj = cb->args[1];
1520                 flow = ovs_flow_dump_next(table, &bucket, &obj);
1521                 if (!flow)
1522                         break;
1523
1524                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1525                                            NETLINK_CB(cb->skb).portid,
1526                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1527                                            OVS_FLOW_CMD_NEW) < 0)
1528                         break;
1529
1530                 cb->args[0] = bucket;
1531                 cb->args[1] = obj;
1532         }
1533         rcu_read_unlock();
1534         return skb->len;
1535 }
1536
1537 static struct genl_ops dp_flow_genl_ops[] = {
1538         { .cmd = OVS_FLOW_CMD_NEW,
1539           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1540           .policy = flow_policy,
1541           .doit = ovs_flow_cmd_new_or_set
1542         },
1543         { .cmd = OVS_FLOW_CMD_DEL,
1544           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1545           .policy = flow_policy,
1546           .doit = ovs_flow_cmd_del
1547         },
1548         { .cmd = OVS_FLOW_CMD_GET,
1549           .flags = 0,               /* OK for unprivileged users. */
1550           .policy = flow_policy,
1551           .doit = ovs_flow_cmd_get,
1552           .dumpit = ovs_flow_cmd_dump
1553         },
1554         { .cmd = OVS_FLOW_CMD_SET,
1555           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1556           .policy = flow_policy,
1557           .doit = ovs_flow_cmd_new_or_set,
1558         },
1559 };
1560
1561 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1562         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1563         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1564 };
1565
1566 static struct genl_family dp_datapath_genl_family = {
1567         .id = GENL_ID_GENERATE,
1568         .hdrsize = sizeof(struct ovs_header),
1569         .name = OVS_DATAPATH_FAMILY,
1570         .version = OVS_DATAPATH_VERSION,
1571         .maxattr = OVS_DP_ATTR_MAX,
1572         .netnsok = true,
1573         .parallel_ops = true,
1574 };
1575
1576 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1577         .name = OVS_DATAPATH_MCGROUP
1578 };
1579
1580 static size_t ovs_dp_cmd_msg_size(void)
1581 {
1582         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1583
1584         msgsize += nla_total_size(IFNAMSIZ);
1585         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1586
1587         return msgsize;
1588 }
1589
1590 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1591                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1592 {
1593         struct ovs_header *ovs_header;
1594         struct ovs_dp_stats dp_stats;
1595         int err;
1596
1597         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1598                                    flags, cmd);
1599         if (!ovs_header)
1600                 goto error;
1601
1602         ovs_header->dp_ifindex = get_dpifindex(dp);
1603
1604         rcu_read_lock();
1605         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1606         rcu_read_unlock();
1607         if (err)
1608                 goto nla_put_failure;
1609
1610         get_dp_stats(dp, &dp_stats);
1611         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1612                 goto nla_put_failure;
1613
1614         return genlmsg_end(skb, ovs_header);
1615
1616 nla_put_failure:
1617         genlmsg_cancel(skb, ovs_header);
1618 error:
1619         return -EMSGSIZE;
1620 }
1621
1622 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1623                                              u32 seq, u8 cmd)
1624 {
1625         struct sk_buff *skb;
1626         int retval;
1627
1628         skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1629         if (!skb)
1630                 return ERR_PTR(-ENOMEM);
1631
1632         retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1633         if (retval < 0) {
1634                 kfree_skb(skb);
1635                 return ERR_PTR(retval);
1636         }
1637         return skb;
1638 }
1639
1640 /* Called with ovs_mutex. */
1641 static struct datapath *lookup_datapath(struct net *net,
1642                                         struct ovs_header *ovs_header,
1643                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1644 {
1645         struct datapath *dp;
1646
1647         if (!a[OVS_DP_ATTR_NAME])
1648                 dp = get_dp(net, ovs_header->dp_ifindex);
1649         else {
1650                 struct vport *vport;
1651
1652                 rcu_read_lock();
1653                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1654                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1655                 rcu_read_unlock();
1656         }
1657         return dp ? dp : ERR_PTR(-ENODEV);
1658 }
1659
1660 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1661 {
1662         struct nlattr **a = info->attrs;
1663         struct vport_parms parms;
1664         struct sk_buff *reply;
1665         struct datapath *dp;
1666         struct vport *vport;
1667         struct ovs_net *ovs_net;
1668         int err, i;
1669
1670         err = -EINVAL;
1671         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1672                 goto err;
1673
1674         ovs_lock();
1675
1676         err = -ENOMEM;
1677         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1678         if (dp == NULL)
1679                 goto err_unlock_ovs;
1680
1681         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1682
1683         /* Allocate table. */
1684         err = -ENOMEM;
1685         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1686         if (!dp->table)
1687                 goto err_free_dp;
1688
1689         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1690         if (!dp->stats_percpu) {
1691                 err = -ENOMEM;
1692                 goto err_destroy_table;
1693         }
1694
1695         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1696                         GFP_KERNEL);
1697         if (!dp->ports) {
1698                 err = -ENOMEM;
1699                 goto err_destroy_percpu;
1700         }
1701
1702         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1703                 INIT_HLIST_HEAD(&dp->ports[i]);
1704
1705         /* Set up our datapath device. */
1706         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1707         parms.type = OVS_VPORT_TYPE_INTERNAL;
1708         parms.options = NULL;
1709         parms.dp = dp;
1710         parms.port_no = OVSP_LOCAL;
1711         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1712
1713         vport = new_vport(&parms);
1714         if (IS_ERR(vport)) {
1715                 err = PTR_ERR(vport);
1716                 if (err == -EBUSY)
1717                         err = -EEXIST;
1718
1719                 goto err_destroy_ports_array;
1720         }
1721
1722         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1723                                       info->snd_seq, OVS_DP_CMD_NEW);
1724         err = PTR_ERR(reply);
1725         if (IS_ERR(reply))
1726                 goto err_destroy_local_port;
1727
1728         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1729         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1730
1731         ovs_unlock();
1732
1733         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1734         return 0;
1735
1736 err_destroy_local_port:
1737         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1738 err_destroy_ports_array:
1739         kfree(dp->ports);
1740 err_destroy_percpu:
1741         free_percpu(dp->stats_percpu);
1742 err_destroy_table:
1743         ovs_flow_tbl_destroy(ovsl_dereference(dp->table), false);
1744 err_free_dp:
1745         release_net(ovs_dp_get_net(dp));
1746         kfree(dp);
1747 err_unlock_ovs:
1748         ovs_unlock();
1749 err:
1750         return err;
1751 }
1752
1753 /* Called with ovs_mutex. */
1754 static void __dp_destroy(struct datapath *dp)
1755 {
1756         int i;
1757
1758         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1759                 struct vport *vport;
1760                 struct hlist_node *n;
1761
1762                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1763                         if (vport->port_no != OVSP_LOCAL)
1764                                 ovs_dp_detach_port(vport);
1765         }
1766
1767         list_del_rcu(&dp->list_node);
1768
1769         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1770          * all port in datapath are destroyed first before freeing datapath.
1771          */
1772         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1773
1774         call_rcu(&dp->rcu, destroy_dp_rcu);
1775 }
1776
1777 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1778 {
1779         struct sk_buff *reply;
1780         struct datapath *dp;
1781         int err;
1782
1783         ovs_lock();
1784         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1785         err = PTR_ERR(dp);
1786         if (IS_ERR(dp))
1787                 goto unlock;
1788
1789         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1790                                       info->snd_seq, OVS_DP_CMD_DEL);
1791         err = PTR_ERR(reply);
1792         if (IS_ERR(reply))
1793                 goto unlock;
1794
1795         __dp_destroy(dp);
1796         ovs_unlock();
1797
1798         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1799
1800         return 0;
1801 unlock:
1802         ovs_unlock();
1803         return err;
1804 }
1805
1806 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1807 {
1808         struct sk_buff *reply;
1809         struct datapath *dp;
1810         int err;
1811
1812         ovs_lock();
1813         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1814         err = PTR_ERR(dp);
1815         if (IS_ERR(dp))
1816                 goto unlock;
1817
1818         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1819                                       info->snd_seq, OVS_DP_CMD_NEW);
1820         if (IS_ERR(reply)) {
1821                 err = PTR_ERR(reply);
1822                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1823                                 ovs_dp_datapath_multicast_group.id, err);
1824                 err = 0;
1825                 goto unlock;
1826         }
1827
1828         ovs_unlock();
1829         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1830
1831         return 0;
1832 unlock:
1833         ovs_unlock();
1834         return err;
1835 }
1836
1837 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1838 {
1839         struct sk_buff *reply;
1840         struct datapath *dp;
1841         int err;
1842
1843         ovs_lock();
1844         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1845         if (IS_ERR(dp)) {
1846                 err = PTR_ERR(dp);
1847                 goto unlock;
1848         }
1849
1850         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1851                                       info->snd_seq, OVS_DP_CMD_NEW);
1852         if (IS_ERR(reply)) {
1853                 err = PTR_ERR(reply);
1854                 goto unlock;
1855         }
1856
1857         ovs_unlock();
1858         return genlmsg_reply(reply, info);
1859
1860 unlock:
1861         ovs_unlock();
1862         return err;
1863 }
1864
1865 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1866 {
1867         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1868         struct datapath *dp;
1869         int skip = cb->args[0];
1870         int i = 0;
1871
1872         rcu_read_lock();
1873         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1874                 if (i >= skip &&
1875                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1876                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1877                                          OVS_DP_CMD_NEW) < 0)
1878                         break;
1879                 i++;
1880         }
1881         rcu_read_unlock();
1882
1883         cb->args[0] = i;
1884
1885         return skb->len;
1886 }
1887
1888 static struct genl_ops dp_datapath_genl_ops[] = {
1889         { .cmd = OVS_DP_CMD_NEW,
1890           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1891           .policy = datapath_policy,
1892           .doit = ovs_dp_cmd_new
1893         },
1894         { .cmd = OVS_DP_CMD_DEL,
1895           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1896           .policy = datapath_policy,
1897           .doit = ovs_dp_cmd_del
1898         },
1899         { .cmd = OVS_DP_CMD_GET,
1900           .flags = 0,               /* OK for unprivileged users. */
1901           .policy = datapath_policy,
1902           .doit = ovs_dp_cmd_get,
1903           .dumpit = ovs_dp_cmd_dump
1904         },
1905         { .cmd = OVS_DP_CMD_SET,
1906           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1907           .policy = datapath_policy,
1908           .doit = ovs_dp_cmd_set,
1909         },
1910 };
1911
1912 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1913         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1914         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1915         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1916         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1917         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1918         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1919 };
1920
1921 static struct genl_family dp_vport_genl_family = {
1922         .id = GENL_ID_GENERATE,
1923         .hdrsize = sizeof(struct ovs_header),
1924         .name = OVS_VPORT_FAMILY,
1925         .version = OVS_VPORT_VERSION,
1926         .maxattr = OVS_VPORT_ATTR_MAX,
1927         .netnsok = true,
1928         .parallel_ops = true,
1929 };
1930
1931 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1932         .name = OVS_VPORT_MCGROUP
1933 };
1934
1935 /* Called with ovs_mutex or RCU read lock. */
1936 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1937                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1938 {
1939         struct ovs_header *ovs_header;
1940         struct ovs_vport_stats vport_stats;
1941         int err;
1942
1943         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1944                                  flags, cmd);
1945         if (!ovs_header)
1946                 return -EMSGSIZE;
1947
1948         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1949
1950         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1951             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1952             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1953             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1954                 goto nla_put_failure;
1955
1956         ovs_vport_get_stats(vport, &vport_stats);
1957         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1958                     &vport_stats))
1959                 goto nla_put_failure;
1960
1961         err = ovs_vport_get_options(vport, skb);
1962         if (err == -EMSGSIZE)
1963                 goto error;
1964
1965         return genlmsg_end(skb, ovs_header);
1966
1967 nla_put_failure:
1968         err = -EMSGSIZE;
1969 error:
1970         genlmsg_cancel(skb, ovs_header);
1971         return err;
1972 }
1973
1974 /* Called with ovs_mutex or RCU read lock. */
1975 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1976                                          u32 seq, u8 cmd)
1977 {
1978         struct sk_buff *skb;
1979         int retval;
1980
1981         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1982         if (!skb)
1983                 return ERR_PTR(-ENOMEM);
1984
1985         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1986         BUG_ON(retval < 0);
1987
1988         return skb;
1989 }
1990
1991 /* Called with ovs_mutex or RCU read lock. */
1992 static struct vport *lookup_vport(struct net *net,
1993                                   struct ovs_header *ovs_header,
1994                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1995 {
1996         struct datapath *dp;
1997         struct vport *vport;
1998
1999         if (a[OVS_VPORT_ATTR_NAME]) {
2000                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2001                 if (!vport)
2002                         return ERR_PTR(-ENODEV);
2003                 if (ovs_header->dp_ifindex &&
2004                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2005                         return ERR_PTR(-ENODEV);
2006                 return vport;
2007         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2008                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2009
2010                 if (port_no >= DP_MAX_PORTS)
2011                         return ERR_PTR(-EFBIG);
2012
2013                 dp = get_dp(net, ovs_header->dp_ifindex);
2014                 if (!dp)
2015                         return ERR_PTR(-ENODEV);
2016
2017                 vport = ovs_vport_ovsl_rcu(dp, port_no);
2018                 if (!vport)
2019                         return ERR_PTR(-ENODEV);
2020                 return vport;
2021         } else
2022                 return ERR_PTR(-EINVAL);
2023 }
2024
2025 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2026 {
2027         struct nlattr **a = info->attrs;
2028         struct ovs_header *ovs_header = info->userhdr;
2029         struct vport_parms parms;
2030         struct sk_buff *reply;
2031         struct vport *vport;
2032         struct datapath *dp;
2033         u32 port_no;
2034         int err;
2035
2036         err = -EINVAL;
2037         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2038             !a[OVS_VPORT_ATTR_UPCALL_PID])
2039                 goto exit;
2040
2041         ovs_lock();
2042         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2043         err = -ENODEV;
2044         if (!dp)
2045                 goto exit_unlock;
2046
2047         if (a[OVS_VPORT_ATTR_PORT_NO]) {
2048                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2049
2050                 err = -EFBIG;
2051                 if (port_no >= DP_MAX_PORTS)
2052                         goto exit_unlock;
2053
2054                 vport = ovs_vport_ovsl(dp, port_no);
2055                 err = -EBUSY;
2056                 if (vport)
2057                         goto exit_unlock;
2058         } else {
2059                 for (port_no = 1; ; port_no++) {
2060                         if (port_no >= DP_MAX_PORTS) {
2061                                 err = -EFBIG;
2062                                 goto exit_unlock;
2063                         }
2064                         vport = ovs_vport_ovsl(dp, port_no);
2065                         if (!vport)
2066                                 break;
2067                 }
2068         }
2069
2070         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2071         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2072         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2073         parms.dp = dp;
2074         parms.port_no = port_no;
2075         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2076
2077         vport = new_vport(&parms);
2078         err = PTR_ERR(vport);
2079         if (IS_ERR(vport))
2080                 goto exit_unlock;
2081
2082         err = 0;
2083         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
2084                                          OVS_VPORT_CMD_NEW);
2085         if (IS_ERR(reply)) {
2086                 err = PTR_ERR(reply);
2087                 ovs_dp_detach_port(vport);
2088                 goto exit_unlock;
2089         }
2090
2091         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2092
2093 exit_unlock:
2094         ovs_unlock();
2095 exit:
2096         return err;
2097 }
2098
2099 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2100 {
2101         struct nlattr **a = info->attrs;
2102         struct sk_buff *reply;
2103         struct vport *vport;
2104         int err;
2105
2106         ovs_lock();
2107         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2108         err = PTR_ERR(vport);
2109         if (IS_ERR(vport))
2110                 goto exit_unlock;
2111
2112         if (a[OVS_VPORT_ATTR_TYPE] &&
2113             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2114                 err = -EINVAL;
2115                 goto exit_unlock;
2116         }
2117
2118         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2119         if (!reply) {
2120                 err = -ENOMEM;
2121                 goto exit_unlock;
2122         }
2123
2124         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2125                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2126                 if (err)
2127                         goto exit_free;
2128         }
2129
2130         if (a[OVS_VPORT_ATTR_UPCALL_PID])
2131                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2132
2133         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2134                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2135         BUG_ON(err < 0);
2136
2137         ovs_unlock();
2138         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2139         return 0;
2140
2141 exit_free:
2142         kfree_skb(reply);
2143 exit_unlock:
2144         ovs_unlock();
2145         return err;
2146 }
2147
2148 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2149 {
2150         struct nlattr **a = info->attrs;
2151         struct sk_buff *reply;
2152         struct vport *vport;
2153         int err;
2154
2155         ovs_lock();
2156         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2157         err = PTR_ERR(vport);
2158         if (IS_ERR(vport))
2159                 goto exit_unlock;
2160
2161         if (vport->port_no == OVSP_LOCAL) {
2162                 err = -EINVAL;
2163                 goto exit_unlock;
2164         }
2165
2166         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2167                                          info->snd_seq, OVS_VPORT_CMD_DEL);
2168         err = PTR_ERR(reply);
2169         if (IS_ERR(reply))
2170                 goto exit_unlock;
2171
2172         err = 0;
2173         ovs_dp_detach_port(vport);
2174
2175         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2176
2177 exit_unlock:
2178         ovs_unlock();
2179         return err;
2180 }
2181
2182 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2183 {
2184         struct nlattr **a = info->attrs;
2185         struct ovs_header *ovs_header = info->userhdr;
2186         struct sk_buff *reply;
2187         struct vport *vport;
2188         int err;
2189
2190         rcu_read_lock();
2191         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2192         err = PTR_ERR(vport);
2193         if (IS_ERR(vport))
2194                 goto exit_unlock;
2195
2196         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2197                                          info->snd_seq, OVS_VPORT_CMD_NEW);
2198         err = PTR_ERR(reply);
2199         if (IS_ERR(reply))
2200                 goto exit_unlock;
2201
2202         rcu_read_unlock();
2203
2204         return genlmsg_reply(reply, info);
2205
2206 exit_unlock:
2207         rcu_read_unlock();
2208         return err;
2209 }
2210
2211 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2212 {
2213         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2214         struct datapath *dp;
2215         int bucket = cb->args[0], skip = cb->args[1];
2216         int i, j = 0;
2217
2218         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2219         if (!dp)
2220                 return -ENODEV;
2221
2222         rcu_read_lock();
2223         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2224                 struct vport *vport;
2225
2226                 j = 0;
2227                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2228                         if (j >= skip &&
2229                             ovs_vport_cmd_fill_info(vport, skb,
2230                                                     NETLINK_CB(cb->skb).portid,
2231                                                     cb->nlh->nlmsg_seq,
2232                                                     NLM_F_MULTI,
2233                                                     OVS_VPORT_CMD_NEW) < 0)
2234                                 goto out;
2235
2236                         j++;
2237                 }
2238                 skip = 0;
2239         }
2240 out:
2241         rcu_read_unlock();
2242
2243         cb->args[0] = i;
2244         cb->args[1] = j;
2245
2246         return skb->len;
2247 }
2248
2249 static struct genl_ops dp_vport_genl_ops[] = {
2250         { .cmd = OVS_VPORT_CMD_NEW,
2251           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2252           .policy = vport_policy,
2253           .doit = ovs_vport_cmd_new
2254         },
2255         { .cmd = OVS_VPORT_CMD_DEL,
2256           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2257           .policy = vport_policy,
2258           .doit = ovs_vport_cmd_del
2259         },
2260         { .cmd = OVS_VPORT_CMD_GET,
2261           .flags = 0,               /* OK for unprivileged users. */
2262           .policy = vport_policy,
2263           .doit = ovs_vport_cmd_get,
2264           .dumpit = ovs_vport_cmd_dump
2265         },
2266         { .cmd = OVS_VPORT_CMD_SET,
2267           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2268           .policy = vport_policy,
2269           .doit = ovs_vport_cmd_set,
2270         },
2271 };
2272
2273 struct genl_family_and_ops {
2274         struct genl_family *family;
2275         struct genl_ops *ops;
2276         int n_ops;
2277         struct genl_multicast_group *group;
2278 };
2279
2280 static const struct genl_family_and_ops dp_genl_families[] = {
2281         { &dp_datapath_genl_family,
2282           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2283           &ovs_dp_datapath_multicast_group },
2284         { &dp_vport_genl_family,
2285           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2286           &ovs_dp_vport_multicast_group },
2287         { &dp_flow_genl_family,
2288           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2289           &ovs_dp_flow_multicast_group },
2290         { &dp_packet_genl_family,
2291           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2292           NULL },
2293 };
2294
2295 static void dp_unregister_genl(int n_families)
2296 {
2297         int i;
2298
2299         for (i = 0; i < n_families; i++)
2300                 genl_unregister_family(dp_genl_families[i].family);
2301 }
2302
2303 static int dp_register_genl(void)
2304 {
2305         int n_registered;
2306         int err;
2307         int i;
2308
2309         n_registered = 0;
2310         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2311                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2312
2313                 err = genl_register_family_with_ops(f->family, f->ops,
2314                                                     f->n_ops);
2315                 if (err)
2316                         goto error;
2317                 n_registered++;
2318
2319                 if (f->group) {
2320                         err = genl_register_mc_group(f->family, f->group);
2321                         if (err)
2322                                 goto error;
2323                 }
2324         }
2325
2326         return 0;
2327
2328 error:
2329         dp_unregister_genl(n_registered);
2330         return err;
2331 }
2332
2333 static void rehash_flow_table(struct work_struct *work)
2334 {
2335         struct datapath *dp;
2336         struct net *net;
2337
2338         ovs_lock();
2339         rtnl_lock();
2340         for_each_net(net) {
2341                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2342
2343                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2344                         struct flow_table *old_table = ovsl_dereference(dp->table);
2345                         struct flow_table *new_table;
2346
2347                         new_table = ovs_flow_tbl_rehash(old_table);
2348                         if (!IS_ERR(new_table)) {
2349                                 rcu_assign_pointer(dp->table, new_table);
2350                                 ovs_flow_tbl_destroy(old_table, true);
2351                         }
2352                 }
2353         }
2354         rtnl_unlock();
2355         ovs_unlock();
2356         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2357 }
2358
2359 static int __net_init ovs_init_net(struct net *net)
2360 {
2361         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2362
2363         INIT_LIST_HEAD(&ovs_net->dps);
2364         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2365         return 0;
2366 }
2367
2368 static void __net_exit ovs_exit_net(struct net *net)
2369 {
2370         struct datapath *dp, *dp_next;
2371         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2372
2373         ovs_lock();
2374         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2375                 __dp_destroy(dp);
2376         ovs_unlock();
2377
2378         cancel_work_sync(&ovs_net->dp_notify_work);
2379 }
2380
2381 static struct pernet_operations ovs_net_ops = {
2382         .init = ovs_init_net,
2383         .exit = ovs_exit_net,
2384         .id   = &ovs_net_id,
2385         .size = sizeof(struct ovs_net),
2386 };
2387
2388 static int __init dp_init(void)
2389 {
2390         int err;
2391
2392         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2393
2394         pr_info("Open vSwitch switching datapath\n");
2395
2396         err = ovs_flow_init();
2397         if (err)
2398                 goto error;
2399
2400         err = ovs_vport_init();
2401         if (err)
2402                 goto error_flow_exit;
2403
2404         err = register_pernet_device(&ovs_net_ops);
2405         if (err)
2406                 goto error_vport_exit;
2407
2408         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2409         if (err)
2410                 goto error_netns_exit;
2411
2412         err = dp_register_genl();
2413         if (err < 0)
2414                 goto error_unreg_notifier;
2415
2416         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2417
2418         return 0;
2419
2420 error_unreg_notifier:
2421         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2422 error_netns_exit:
2423         unregister_pernet_device(&ovs_net_ops);
2424 error_vport_exit:
2425         ovs_vport_exit();
2426 error_flow_exit:
2427         ovs_flow_exit();
2428 error:
2429         return err;
2430 }
2431
2432 static void dp_cleanup(void)
2433 {
2434         cancel_delayed_work_sync(&rehash_flow_wq);
2435         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2436         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2437         unregister_pernet_device(&ovs_net_ops);
2438         rcu_barrier();
2439         ovs_vport_exit();
2440         ovs_flow_exit();
2441 }
2442
2443 module_init(dp_init);
2444 module_exit(dp_cleanup);
2445
2446 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2447 MODULE_LICENSE("GPL");