datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007-2012 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/version.h>
  40 #include <linux/ethtool.h>
  41 #include <linux/wait.h>
  42 #include <asm/div64.h>
  43 #include <linux/highmem.h>
  44 #include <linux/netfilter_bridge.h>
  45 #include <linux/netfilter_ipv4.h>
  46 #include <linux/inetdevice.h>
  47 #include <linux/list.h>
  48 #include <linux/openvswitch.h>
  49 #include <linux/rculist.h>
  50 #include <linux/dmi.h>
  51 #include <net/genetlink.h>
  52 #include <net/net_namespace.h>
  53 #include <net/netns/generic.h>
  54
  55 #include "checksum.h"
  56 #include "datapath.h"
  57 #include "flow.h"
  58 #include "genl_exec.h"
  59 #include "vlan.h"
  60 #include "tunnel.h"
  61 #include "vport-internal_dev.h"
  62
  63 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
  64     LINUX_VERSION_CODE >= KERNEL_VERSION(3,9,0)
  65 #error Kernels before 2.6.18 or after 3.8 are not supported by this version of Open vSwitch.
  66 #endif
  67
  68 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
  69 static void rehash_flow_table(struct work_struct *work);
  70 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
  71
  72 int ovs_net_id __read_mostly;
  73
  74 /**
  75  * DOC: Locking:
  76  *
  77  * Writes to device state (add/remove datapath, port, set operations on vports,
  78  * etc.) are protected by RTNL.
  79  *
  80  * Writes to other state (flow table modifications, set miscellaneous datapath
  81  * parameters, etc.) are protected by genl_mutex.  The RTNL lock nests inside
  82  * genl_mutex.
  83  *
  84  * Reads are protected by RCU.
  85  *
  86  * There are a few special cases (mostly stats) that have their own
  87  * synchronization but they nest under all of above and don't interact with
  88  * each other.
  89  */
  90
  91 static struct vport *new_vport(const struct vport_parms *);
  92 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
  93                              const struct dp_upcall_info *);
  94 static int queue_userspace_packet(struct net *, int dp_ifindex,
  95                                   struct sk_buff *,
  96                                   const struct dp_upcall_info *);
  97
  98 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  99 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 100 {
 101         struct datapath *dp = NULL;
 102         struct net_device *dev;
 103
 104         rcu_read_lock();
 105         dev = dev_get_by_index_rcu(net, dp_ifindex);
 106         if (dev) {
 107                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 108                 if (vport)
 109                         dp = vport->dp;
 110         }
 111         rcu_read_unlock();
 112
 113         return dp;
 114 }
 115
 116 /* Must be called with rcu_read_lock or RTNL lock. */
 117 const char *ovs_dp_name(const struct datapath *dp)
 118 {
 119         struct vport *vport = ovs_vport_rtnl_rcu(dp, OVSP_LOCAL);
 120         return vport->ops->get_name(vport);
 121 }
 122
 123 static int get_dpifindex(struct datapath *dp)
 124 {
 125         struct vport *local;
 126         int ifindex;
 127
 128         rcu_read_lock();
 129
 130         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 131         if (local)
 132                 ifindex = local->ops->get_ifindex(local);
 133         else
 134                 ifindex = 0;
 135
 136         rcu_read_unlock();
 137
 138         return ifindex;
 139 }
 140
 141 static void destroy_dp_rcu(struct rcu_head *rcu)
 142 {
 143         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 144
 145         ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
 146         free_percpu(dp->stats_percpu);
 147         release_net(ovs_dp_get_net(dp));
 148         kfree(dp->ports);
 149         kfree(dp);
 150 }
 151
 152 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 153                                             u16 port_no)
 154 {
 155         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 156 }
 157
 158 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 159 {
 160         struct vport *vport;
 161         struct hlist_head *head;
 162
 163         head = vport_hash_bucket(dp, port_no);
 164         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 165                 if (vport->port_no == port_no)
 166                         return vport;
 167         }
 168         return NULL;
 169 }
 170
 171 /* Called with RTNL lock and genl_lock. */
 172 static struct vport *new_vport(const struct vport_parms *parms)
 173 {
 174         struct vport *vport;
 175
 176         vport = ovs_vport_add(parms);
 177         if (!IS_ERR(vport)) {
 178                 struct datapath *dp = parms->dp;
 179                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 180
 181                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 182         }
 183         return vport;
 184 }
 185
 186 /* Called with RTNL lock. */
 187 void ovs_dp_detach_port(struct vport *p)
 188 {
 189         ASSERT_RTNL();
 190
 191         /* First drop references to device. */
 192         hlist_del_rcu(&p->dp_hash_node);
 193
 194         /* Then destroy it. */
 195         ovs_vport_del(p);
 196 }
 197
 198 /* Must be called with rcu_read_lock. */
 199 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 200 {
 201         struct datapath *dp = p->dp;
 202         struct sw_flow *flow;
 203         struct dp_stats_percpu *stats;
 204         struct sw_flow_key key;
 205         u64 *stats_counter;
 206         int error;
 207         int key_len;
 208
 209         stats = this_cpu_ptr(dp->stats_percpu);
 210
 211         /* Extract flow from 'skb' into 'key'. */
 212         error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
 213         if (unlikely(error)) {
 214                 kfree_skb(skb);
 215                 return;
 216         }
 217
 218         /* Look up flow. */
 219         flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
 220         if (unlikely(!flow)) {
 221                 struct dp_upcall_info upcall;
 222
 223                 upcall.cmd = OVS_PACKET_CMD_MISS;
 224                 upcall.key = &key;
 225                 upcall.userdata = NULL;
 226                 upcall.portid = p->upcall_portid;
 227                 ovs_dp_upcall(dp, skb, &upcall);
 228                 consume_skb(skb);
 229                 stats_counter = &stats->n_missed;
 230                 goto out;
 231         }
 232
 233         OVS_CB(skb)->flow = flow;
 234
 235         stats_counter = &stats->n_hit;
 236         ovs_flow_used(OVS_CB(skb)->flow, skb);
 237         ovs_execute_actions(dp, skb);
 238
 239 out:
 240         /* Update datapath statistics. */
 241         u64_stats_update_begin(&stats->sync);
 242         (*stats_counter)++;
 243         u64_stats_update_end(&stats->sync);
 244 }
 245
 246 static struct genl_family dp_packet_genl_family = {
 247         .id = GENL_ID_GENERATE,
 248         .hdrsize = sizeof(struct ovs_header),
 249         .name = OVS_PACKET_FAMILY,
 250         .version = OVS_PACKET_VERSION,
 251         .maxattr = OVS_PACKET_ATTR_MAX,
 252          SET_NETNSOK
 253 };
 254
 255 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 256                   const struct dp_upcall_info *upcall_info)
 257 {
 258         struct dp_stats_percpu *stats;
 259         int dp_ifindex;
 260         int err;
 261
 262         if (upcall_info->portid == 0) {
 263                 err = -ENOTCONN;
 264                 goto err;
 265         }
 266
 267         dp_ifindex = get_dpifindex(dp);
 268         if (!dp_ifindex) {
 269                 err = -ENODEV;
 270                 goto err;
 271         }
 272
 273         forward_ip_summed(skb, true);
 274
 275         if (!skb_is_gso(skb))
 276                 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 277         else
 278                 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
 279         if (err)
 280                 goto err;
 281
 282         return 0;
 283
 284 err:
 285         stats = this_cpu_ptr(dp->stats_percpu);
 286
 287         u64_stats_update_begin(&stats->sync);
 288         stats->n_lost++;
 289         u64_stats_update_end(&stats->sync);
 290
 291         return err;
 292 }
 293
 294 static int queue_gso_packets(struct net *net, int dp_ifindex,
 295                              struct sk_buff *skb,
 296                              const struct dp_upcall_info *upcall_info)
 297 {
 298         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 299         struct dp_upcall_info later_info;
 300         struct sw_flow_key later_key;
 301         struct sk_buff *segs, *nskb;
 302         int err;
 303
 304         segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
 305         if (IS_ERR(segs))
 306                 return PTR_ERR(segs);
 307
 308         /* Queue all of the segments. */
 309         skb = segs;
 310         do {
 311                 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
 312                 if (err)
 313                         break;
 314
 315                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 316                         /* The initial flow key extracted by ovs_flow_extract()
 317                          * in this case is for a first fragment, so we need to
 318                          * properly mark later fragments.
 319                          */
 320                         later_key = *upcall_info->key;
 321                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 322
 323                         later_info = *upcall_info;
 324                         later_info.key = &later_key;
 325                         upcall_info = &later_info;
 326                 }
 327         } while ((skb = skb->next));
 328
 329         /* Free all of the segments. */
 330         skb = segs;
 331         do {
 332                 nskb = skb->next;
 333                 if (err)
 334                         kfree_skb(skb);
 335                 else
 336                         consume_skb(skb);
 337         } while ((skb = nskb));
 338         return err;
 339 }
 340
 341 static int queue_userspace_packet(struct net *net, int dp_ifindex,
 342                                   struct sk_buff *skb,
 343                                   const struct dp_upcall_info *upcall_info)
 344 {
 345         struct ovs_header *upcall;
 346         struct sk_buff *nskb = NULL;
 347         struct sk_buff *user_skb; /* to be queued to userspace */
 348         struct nlattr *nla;
 349         unsigned int len;
 350         int err;
 351
 352         if (vlan_tx_tag_present(skb)) {
 353                 nskb = skb_clone(skb, GFP_ATOMIC);
 354                 if (!nskb)
 355                         return -ENOMEM;
 356
 357                 err = vlan_deaccel_tag(nskb);
 358                 if (err)
 359                         return err;
 360
 361                 skb = nskb;
 362         }
 363
 364         if (nla_attr_size(skb->len) > USHRT_MAX) {
 365                 err = -EFBIG;
 366                 goto out;
 367         }
 368
 369         len = sizeof(struct ovs_header);
 370         len += nla_total_size(skb->len);
 371         len += nla_total_size(FLOW_BUFSIZE);
 372         if (upcall_info->userdata)
 373                 len += NLA_ALIGN(upcall_info->userdata->nla_len);
 374
 375         user_skb = genlmsg_new(len, GFP_ATOMIC);
 376         if (!user_skb) {
 377                 err = -ENOMEM;
 378                 goto out;
 379         }
 380
 381         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 382                              0, upcall_info->cmd);
 383         upcall->dp_ifindex = dp_ifindex;
 384
 385         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 386         ovs_flow_to_nlattrs(upcall_info->key, user_skb);
 387         nla_nest_end(user_skb, nla);
 388
 389         if (upcall_info->userdata)
 390                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 391                           nla_len(upcall_info->userdata),
 392                           nla_data(upcall_info->userdata));
 393
 394         nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 395
 396         skb_copy_and_csum_dev(skb, nla_data(nla));
 397
 398         genlmsg_end(user_skb, upcall);
 399         err = genlmsg_unicast(net, user_skb, upcall_info->portid);
 400
 401 out:
 402         kfree_skb(nskb);
 403         return err;
 404 }
 405
 406 /* Called with genl_mutex. */
 407 static int flush_flows(struct datapath *dp)
 408 {
 409         struct flow_table *old_table;
 410         struct flow_table *new_table;
 411
 412         old_table = genl_dereference(dp->table);
 413         new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
 414         if (!new_table)
 415                 return -ENOMEM;
 416
 417         rcu_assign_pointer(dp->table, new_table);
 418
 419         ovs_flow_tbl_deferred_destroy(old_table);
 420         return 0;
 421 }
 422
 423 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa, int attr_len)
 424 {
 425
 426         struct sw_flow_actions *acts;
 427         int new_acts_size;
 428         int req_size = NLA_ALIGN(attr_len);
 429         int next_offset = offsetof(struct sw_flow_actions, actions) +
 430                                         (*sfa)->actions_len;
 431
 432         if (req_size <= (ksize(*sfa) - next_offset))
 433                 goto out;
 434
 435         new_acts_size = ksize(*sfa) * 2;
 436
 437         if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
 438                 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
 439                         return ERR_PTR(-EMSGSIZE);
 440                 new_acts_size = MAX_ACTIONS_BUFSIZE;
 441         }
 442
 443         acts = ovs_flow_actions_alloc(new_acts_size);
 444         if (IS_ERR(acts))
 445                 return (void *)acts;
 446
 447         memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
 448         acts->actions_len = (*sfa)->actions_len;
 449         kfree(*sfa);
 450         *sfa = acts;
 451
 452 out:
 453         (*sfa)->actions_len += req_size;
 454         return  (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
 455 }
 456
 457 static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, int len)
 458 {
 459         struct nlattr *a;
 460
 461         a = reserve_sfa_size(sfa, nla_attr_size(len));
 462         if (IS_ERR(a))
 463                 return PTR_ERR(a);
 464
 465         a->nla_type = attrtype;
 466         a->nla_len = nla_attr_size(len);
 467
 468         if (data)
 469                 memcpy(nla_data(a), data, len);
 470         memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
 471
 472         return 0;
 473 }
 474
 475 static inline int add_nested_action_start(struct sw_flow_actions **sfa, int attrtype)
 476 {
 477         int used = (*sfa)->actions_len;
 478         int err;
 479
 480         err = add_action(sfa, attrtype, NULL, 0);
 481         if (err)
 482                 return err;
 483
 484         return used;
 485 }
 486
 487 static inline void add_nested_action_end(struct sw_flow_actions *sfa, int st_offset)
 488 {
 489         struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions + st_offset);
 490
 491         a->nla_len = sfa->actions_len - st_offset;
 492 }
 493
 494 static int validate_and_copy_actions(const struct nlattr *attr,
 495                                 const struct sw_flow_key *key, int depth,
 496                                 struct sw_flow_actions **sfa);
 497
 498 static int validate_and_copy_sample(const struct nlattr *attr,
 499                            const struct sw_flow_key *key, int depth,
 500                            struct sw_flow_actions **sfa)
 501 {
 502         const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
 503         const struct nlattr *probability, *actions;
 504         const struct nlattr *a;
 505         int rem, start, err, st_acts;
 506
 507         memset(attrs, 0, sizeof(attrs));
 508         nla_for_each_nested(a, attr, rem) {
 509                 int type = nla_type(a);
 510                 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
 511                         return -EINVAL;
 512                 attrs[type] = a;
 513         }
 514         if (rem)
 515                 return -EINVAL;
 516
 517         probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
 518         if (!probability || nla_len(probability) != sizeof(u32))
 519                 return -EINVAL;
 520
 521         actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
 522         if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
 523                 return -EINVAL;
 524
 525         /* validation done, copy sample action. */
 526         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
 527         if (start < 0)
 528                 return start;
 529         err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY, nla_data(probability), sizeof(u32));
 530         if (err)
 531                 return err;
 532         st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
 533         if (st_acts < 0)
 534                 return st_acts;
 535
 536         err = validate_and_copy_actions(actions, key, depth + 1, sfa);
 537         if (err)
 538                 return err;
 539
 540         add_nested_action_end(*sfa, st_acts);
 541         add_nested_action_end(*sfa, start);
 542
 543         return 0;
 544 }
 545
 546 static int validate_tp_port(const struct sw_flow_key *flow_key)
 547 {
 548         if (flow_key->eth.type == htons(ETH_P_IP)) {
 549                 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
 550                         return 0;
 551         } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
 552                 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
 553                         return 0;
 554         }
 555
 556         return -EINVAL;
 557 }
 558
 559 static int validate_and_copy_set_tun(const struct nlattr *attr,
 560                                      struct sw_flow_actions **sfa)
 561 {
 562         struct ovs_key_ipv4_tunnel tun_key;
 563         int err, start;
 564
 565         err = ipv4_tun_from_nlattr(nla_data(attr), &tun_key);
 566         if (err)
 567                 return err;
 568
 569         start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
 570         if (start < 0)
 571                 return start;
 572
 573         err = add_action(sfa, OVS_KEY_ATTR_IPV4_TUNNEL, &tun_key, sizeof(tun_key));
 574         add_nested_action_end(*sfa, start);
 575
 576         return err;
 577 }
 578
 579 static int validate_set(const struct nlattr *a,
 580                         const struct sw_flow_key *flow_key,
 581                         struct sw_flow_actions **sfa,
 582                         bool *set_tun)
 583 {
 584         const struct nlattr *ovs_key = nla_data(a);
 585         int key_type = nla_type(ovs_key);
 586
 587         /* There can be only one key in a action */
 588         if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
 589                 return -EINVAL;
 590
 591         if (key_type > OVS_KEY_ATTR_MAX ||
 592             (ovs_key_lens[key_type] != nla_len(ovs_key) &&
 593              ovs_key_lens[key_type] != -1))
 594                 return -EINVAL;
 595
 596         switch (key_type) {
 597         const struct ovs_key_ipv4 *ipv4_key;
 598         const struct ovs_key_ipv6 *ipv6_key;
 599         int err;
 600
 601         case OVS_KEY_ATTR_PRIORITY:
 602         case OVS_KEY_ATTR_ETHERNET:
 603                 break;
 604
 605         case OVS_KEY_ATTR_SKB_MARK:
 606 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
 607                 if (nla_get_u32(ovs_key) != 0)
 608                         return -EINVAL;
 609 #endif
 610                 break;
 611
 612         case OVS_KEY_ATTR_TUNNEL:
 613                 *set_tun = true;
 614                 err = validate_and_copy_set_tun(a, sfa);
 615                 if (err)
 616                         return err;
 617                 break;
 618
 619         case OVS_KEY_ATTR_IPV4:
 620                 if (flow_key->eth.type != htons(ETH_P_IP))
 621                         return -EINVAL;
 622
 623                 if (!flow_key->ip.proto)
 624                         return -EINVAL;
 625
 626                 ipv4_key = nla_data(ovs_key);
 627                 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
 628                         return -EINVAL;
 629
 630                 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
 631                         return -EINVAL;
 632
 633                 break;
 634
 635         case OVS_KEY_ATTR_IPV6:
 636                 if (flow_key->eth.type != htons(ETH_P_IPV6))
 637                         return -EINVAL;
 638
 639                 if (!flow_key->ip.proto)
 640                         return -EINVAL;
 641
 642                 ipv6_key = nla_data(ovs_key);
 643                 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
 644                         return -EINVAL;
 645
 646                 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
 647                         return -EINVAL;
 648
 649                 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
 650                         return -EINVAL;
 651
 652                 break;
 653
 654         case OVS_KEY_ATTR_TCP:
 655                 if (flow_key->ip.proto != IPPROTO_TCP)
 656                         return -EINVAL;
 657
 658                 return validate_tp_port(flow_key);
 659
 660         case OVS_KEY_ATTR_UDP:
 661                 if (flow_key->ip.proto != IPPROTO_UDP)
 662                         return -EINVAL;
 663
 664                 return validate_tp_port(flow_key);
 665
 666         default:
 667                 return -EINVAL;
 668         }
 669
 670         return 0;
 671 }
 672
 673 static int validate_userspace(const struct nlattr *attr)
 674 {
 675         static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =   {
 676                 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
 677                 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
 678         };
 679         struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
 680         int error;
 681
 682         error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
 683                                  attr, userspace_policy);
 684         if (error)
 685                 return error;
 686
 687         if (!a[OVS_USERSPACE_ATTR_PID] ||
 688             !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
 689                 return -EINVAL;
 690
 691         return 0;
 692 }
 693
 694 static int copy_action(const struct nlattr *from,
 695                       struct sw_flow_actions **sfa)
 696 {
 697         int totlen = NLA_ALIGN(from->nla_len);
 698         struct nlattr *to;
 699
 700         to = reserve_sfa_size(sfa, from->nla_len);
 701         if (IS_ERR(to))
 702                 return PTR_ERR(to);
 703
 704         memcpy(to, from, totlen);
 705         return 0;
 706 }
 707
 708 static int validate_and_copy_actions(const struct nlattr *attr,
 709                                 const struct sw_flow_key *key,
 710                                 int depth,
 711                                 struct sw_flow_actions **sfa)
 712 {
 713         const struct nlattr *a;
 714         int rem, err;
 715
 716         if (depth >= SAMPLE_ACTION_DEPTH)
 717                 return -EOVERFLOW;
 718
 719         nla_for_each_nested(a, attr, rem) {
 720                 /* Expected argument lengths, (u32)-1 for variable length. */
 721                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 722                         [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
 723                         [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
 724                         [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
 725                         [OVS_ACTION_ATTR_POP_VLAN] = 0,
 726                         [OVS_ACTION_ATTR_SET] = (u32)-1,
 727                         [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
 728                 };
 729                 const struct ovs_action_push_vlan *vlan;
 730                 int type = nla_type(a);
 731                 bool skip_copy;
 732
 733                 if (type > OVS_ACTION_ATTR_MAX ||
 734                     (action_lens[type] != nla_len(a) &&
 735                      action_lens[type] != (u32)-1))
 736                         return -EINVAL;
 737
 738                 skip_copy = false;
 739                 switch (type) {
 740                 case OVS_ACTION_ATTR_UNSPEC:
 741                         return -EINVAL;
 742
 743                 case OVS_ACTION_ATTR_USERSPACE:
 744                         err = validate_userspace(a);
 745                         if (err)
 746                                 return err;
 747                         break;
 748
 749                 case OVS_ACTION_ATTR_OUTPUT:
 750                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 751                                 return -EINVAL;
 752                         break;
 753
 754
 755                 case OVS_ACTION_ATTR_POP_VLAN:
 756                         break;
 757
 758                 case OVS_ACTION_ATTR_PUSH_VLAN:
 759                         vlan = nla_data(a);
 760                         if (vlan->vlan_tpid != htons(ETH_P_8021Q))
 761                                 return -EINVAL;
 762                         if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
 763                                 return -EINVAL;
 764                         break;
 765
 766                 case OVS_ACTION_ATTR_SET:
 767                         err = validate_set(a, key, sfa, &skip_copy);
 768                         if (err)
 769                                 return err;
 770                         break;
 771
 772                 case OVS_ACTION_ATTR_SAMPLE:
 773                         err = validate_and_copy_sample(a, key, depth, sfa);
 774                         if (err)
 775                                 return err;
 776                         skip_copy = true;
 777                         break;
 778
 779                 default:
 780                         return -EINVAL;
 781                 }
 782                 if (!skip_copy) {
 783                         err = copy_action(a, sfa);
 784                         if (err)
 785                                 return err;
 786                 }
 787         }
 788
 789         if (rem > 0)
 790                 return -EINVAL;
 791
 792         return 0;
 793 }
 794
 795 static void clear_stats(struct sw_flow *flow)
 796 {
 797         flow->used = 0;
 798         flow->tcp_flags = 0;
 799         flow->packet_count = 0;
 800         flow->byte_count = 0;
 801 }
 802
 803 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 804 {
 805         struct ovs_header *ovs_header = info->userhdr;
 806         struct nlattr **a = info->attrs;
 807         struct sw_flow_actions *acts;
 808         struct sk_buff *packet;
 809         struct sw_flow *flow;
 810         struct datapath *dp;
 811         struct ethhdr *eth;
 812         int len;
 813         int err;
 814         int key_len;
 815
 816         err = -EINVAL;
 817         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 818             !a[OVS_PACKET_ATTR_ACTIONS])
 819                 goto err;
 820
 821         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 822         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 823         err = -ENOMEM;
 824         if (!packet)
 825                 goto err;
 826         skb_reserve(packet, NET_IP_ALIGN);
 827
 828         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 829
 830         skb_reset_mac_header(packet);
 831         eth = eth_hdr(packet);
 832
 833         /* Normally, setting the skb 'protocol' field would be handled by a
 834          * call to eth_type_trans(), but it assumes there's a sending
 835          * device, which we may not have. */
 836         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 837                 packet->protocol = eth->h_proto;
 838         else
 839                 packet->protocol = htons(ETH_P_802_2);
 840
 841         /* Build an sw_flow for sending this packet. */
 842         flow = ovs_flow_alloc();
 843         err = PTR_ERR(flow);
 844         if (IS_ERR(flow))
 845                 goto err_kfree_skb;
 846
 847         err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
 848         if (err)
 849                 goto err_flow_free;
 850
 851         err = ovs_flow_metadata_from_nlattrs(flow, key_len, a[OVS_PACKET_ATTR_KEY]);
 852         if (err)
 853                 goto err_flow_free;
 854         acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 855         err = PTR_ERR(acts);
 856         if (IS_ERR(acts))
 857                 goto err_flow_free;
 858
 859         err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
 860         rcu_assign_pointer(flow->sf_acts, acts);
 861         if (err)
 862                 goto err_flow_free;
 863
 864         OVS_CB(packet)->flow = flow;
 865         packet->priority = flow->key.phy.priority;
 866         skb_set_mark(packet, flow->key.phy.skb_mark);
 867
 868         rcu_read_lock();
 869         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 870         err = -ENODEV;
 871         if (!dp)
 872                 goto err_unlock;
 873
 874         local_bh_disable();
 875         err = ovs_execute_actions(dp, packet);
 876         local_bh_enable();
 877         rcu_read_unlock();
 878
 879         ovs_flow_free(flow);
 880         return err;
 881
 882 err_unlock:
 883         rcu_read_unlock();
 884 err_flow_free:
 885         ovs_flow_free(flow);
 886 err_kfree_skb:
 887         kfree_skb(packet);
 888 err:
 889         return err;
 890 }
 891
 892 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 893 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
 894         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 895 #else
 896         [OVS_PACKET_ATTR_PACKET] = { .minlen = ETH_HLEN },
 897 #endif
 898         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 899         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 900 };
 901
 902 static struct genl_ops dp_packet_genl_ops[] = {
 903         { .cmd = OVS_PACKET_CMD_EXECUTE,
 904           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 905           .policy = packet_policy,
 906           .doit = ovs_packet_cmd_execute
 907         }
 908 };
 909
 910 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 911 {
 912         int i;
 913         struct flow_table *table = genl_dereference(dp->table);
 914
 915         stats->n_flows = ovs_flow_tbl_count(table);
 916
 917         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 918         for_each_possible_cpu(i) {
 919                 const struct dp_stats_percpu *percpu_stats;
 920                 struct dp_stats_percpu local_stats;
 921                 unsigned int start;
 922
 923                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 924
 925                 do {
 926                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 927                         local_stats = *percpu_stats;
 928                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 929
 930                 stats->n_hit += local_stats.n_hit;
 931                 stats->n_missed += local_stats.n_missed;
 932                 stats->n_lost += local_stats.n_lost;
 933         }
 934 }
 935
 936 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 937         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 938         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 939         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 940 };
 941
 942 static struct genl_family dp_flow_genl_family = {
 943         .id = GENL_ID_GENERATE,
 944         .hdrsize = sizeof(struct ovs_header),
 945         .name = OVS_FLOW_FAMILY,
 946         .version = OVS_FLOW_VERSION,
 947         .maxattr = OVS_FLOW_ATTR_MAX,
 948          SET_NETNSOK
 949 };
 950
 951 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 952         .name = OVS_FLOW_MCGROUP
 953 };
 954
 955 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
 956 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
 957 {
 958         const struct nlattr *a;
 959         struct nlattr *start;
 960         int err = 0, rem;
 961
 962         start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
 963         if (!start)
 964                 return -EMSGSIZE;
 965
 966         nla_for_each_nested(a, attr, rem) {
 967                 int type = nla_type(a);
 968                 struct nlattr *st_sample;
 969
 970                 switch (type) {
 971                 case OVS_SAMPLE_ATTR_PROBABILITY:
 972                         if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
 973                                 return -EMSGSIZE;
 974                         break;
 975                 case OVS_SAMPLE_ATTR_ACTIONS:
 976                         st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
 977                         if (!st_sample)
 978                                 return -EMSGSIZE;
 979                         err = actions_to_attr(nla_data(a), nla_len(a), skb);
 980                         if (err)
 981                                 return err;
 982                         nla_nest_end(skb, st_sample);
 983                         break;
 984                 }
 985         }
 986
 987         nla_nest_end(skb, start);
 988         return err;
 989 }
 990
 991 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
 992 {
 993         const struct nlattr *ovs_key = nla_data(a);
 994         int key_type = nla_type(ovs_key);
 995         struct nlattr *start;
 996         int err;
 997
 998         switch (key_type) {
 999         case OVS_KEY_ATTR_IPV4_TUNNEL:
1000                 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1001                 if (!start)
1002                         return -EMSGSIZE;
1003
1004                 err = ipv4_tun_to_nlattr(skb, nla_data(ovs_key));
1005                 if (err)
1006                         return err;
1007                 nla_nest_end(skb, start);
1008                 break;
1009         default:
1010                 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1011                         return -EMSGSIZE;
1012                 break;
1013         }
1014
1015         return 0;
1016 }
1017
1018 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1019 {
1020         const struct nlattr *a;
1021         int rem, err;
1022
1023         nla_for_each_attr(a, attr, len, rem) {
1024                 int type = nla_type(a);
1025
1026                 switch (type) {
1027                 case OVS_ACTION_ATTR_SET:
1028                         err = set_action_to_attr(a, skb);
1029                         if (err)
1030                                 return err;
1031                         break;
1032
1033                 case OVS_ACTION_ATTR_SAMPLE:
1034                         err = sample_action_to_attr(a, skb);
1035                         if (err)
1036                                 return err;
1037                         break;
1038                 default:
1039                         if (nla_put(skb, type, nla_len(a), nla_data(a)))
1040                                 return -EMSGSIZE;
1041                         break;
1042                 }
1043         }
1044
1045         return 0;
1046 }
1047
1048 /* Called with genl_lock. */
1049 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1050                                   struct sk_buff *skb, u32 portid,
1051                                   u32 seq, u32 flags, u8 cmd)
1052 {
1053         const int skb_orig_len = skb->len;
1054         const struct sw_flow_actions *sf_acts;
1055         struct nlattr *start;
1056         struct ovs_flow_stats stats;
1057         struct ovs_header *ovs_header;
1058         struct nlattr *nla;
1059         unsigned long used;
1060         u8 tcp_flags;
1061         int err;
1062
1063         sf_acts = rcu_dereference_protected(flow->sf_acts,
1064                                             lockdep_genl_is_held());
1065
1066         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1067         if (!ovs_header)
1068                 return -EMSGSIZE;
1069
1070         ovs_header->dp_ifindex = get_dpifindex(dp);
1071
1072         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1073         if (!nla)
1074                 goto nla_put_failure;
1075         err = ovs_flow_to_nlattrs(&flow->key, skb);
1076         if (err)
1077                 goto error;
1078         nla_nest_end(skb, nla);
1079
1080         spin_lock_bh(&flow->lock);
1081         used = flow->used;
1082         stats.n_packets = flow->packet_count;
1083         stats.n_bytes = flow->byte_count;
1084         tcp_flags = flow->tcp_flags;
1085         spin_unlock_bh(&flow->lock);
1086
1087         if (used &&
1088             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1089                 goto nla_put_failure;
1090
1091         if (stats.n_packets &&
1092             nla_put(skb, OVS_FLOW_ATTR_STATS,
1093                     sizeof(struct ovs_flow_stats), &stats))
1094                 goto nla_put_failure;
1095
1096         if (tcp_flags &&
1097             nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1098                 goto nla_put_failure;
1099
1100         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1101          * this is the first flow to be dumped into 'skb'.  This is unusual for
1102          * Netlink but individual action lists can be longer than
1103          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1104          * The userspace caller can always fetch the actions separately if it
1105          * really wants them.  (Most userspace callers in fact don't care.)
1106          *
1107          * This can only fail for dump operations because the skb is always
1108          * properly sized for single flows.
1109          */
1110         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1111         if (start) {
1112                 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1113                 if (!err)
1114                         nla_nest_end(skb, start);
1115                 else {
1116                         if (skb_orig_len)
1117                                 goto error;
1118
1119                         nla_nest_cancel(skb, start);
1120                 }
1121         } else if (skb_orig_len)
1122                 goto nla_put_failure;
1123
1124         return genlmsg_end(skb, ovs_header);
1125
1126 nla_put_failure:
1127         err = -EMSGSIZE;
1128 error:
1129         genlmsg_cancel(skb, ovs_header);
1130         return err;
1131 }
1132
1133 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1134 {
1135         const struct sw_flow_actions *sf_acts;
1136         int len;
1137
1138         sf_acts = rcu_dereference_protected(flow->sf_acts,
1139                                             lockdep_genl_is_held());
1140
1141         /* OVS_FLOW_ATTR_KEY */
1142         len = nla_total_size(FLOW_BUFSIZE);
1143         /* OVS_FLOW_ATTR_ACTIONS */
1144         len += nla_total_size(sf_acts->actions_len);
1145         /* OVS_FLOW_ATTR_STATS */
1146         len += nla_total_size(sizeof(struct ovs_flow_stats));
1147         /* OVS_FLOW_ATTR_TCP_FLAGS */
1148         len += nla_total_size(1);
1149         /* OVS_FLOW_ATTR_USED */
1150         len += nla_total_size(8);
1151
1152         len += NLMSG_ALIGN(sizeof(struct ovs_header));
1153
1154         return genlmsg_new(len, GFP_KERNEL);
1155 }
1156
1157 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1158                                                struct datapath *dp,
1159                                                u32 portid, u32 seq, u8 cmd)
1160 {
1161         struct sk_buff *skb;
1162         int retval;
1163
1164         skb = ovs_flow_cmd_alloc_info(flow);
1165         if (!skb)
1166                 return ERR_PTR(-ENOMEM);
1167
1168         retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1169         BUG_ON(retval < 0);
1170         return skb;
1171 }
1172
1173 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1174 {
1175         struct nlattr **a = info->attrs;
1176         struct ovs_header *ovs_header = info->userhdr;
1177         struct sw_flow_key key;
1178         struct sw_flow *flow;
1179         struct sk_buff *reply;
1180         struct datapath *dp;
1181         struct flow_table *table;
1182         struct sw_flow_actions *acts = NULL;
1183         int error;
1184         int key_len;
1185
1186         /* Extract key. */
1187         error = -EINVAL;
1188         if (!a[OVS_FLOW_ATTR_KEY])
1189                 goto error;
1190         error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1191         if (error)
1192                 goto error;
1193
1194         /* Validate actions. */
1195         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1196                 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1197                 error = PTR_ERR(acts);
1198                 if (IS_ERR(acts))
1199                         goto error;
1200
1201                 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0, &acts);
1202                 if (error)
1203                         goto err_kfree;
1204         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1205                 error = -EINVAL;
1206                 goto error;
1207         }
1208
1209         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1210         error = -ENODEV;
1211         if (!dp)
1212                 goto err_kfree;
1213
1214         table = genl_dereference(dp->table);
1215         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1216         if (!flow) {
1217                 /* Bail out if we're not allowed to create a new flow. */
1218                 error = -ENOENT;
1219                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1220                         goto err_kfree;
1221
1222                 /* Expand table, if necessary, to make room. */
1223                 if (ovs_flow_tbl_need_to_expand(table)) {
1224                         struct flow_table *new_table;
1225
1226                         new_table = ovs_flow_tbl_expand(table);
1227                         if (!IS_ERR(new_table)) {
1228                                 rcu_assign_pointer(dp->table, new_table);
1229                                 ovs_flow_tbl_deferred_destroy(table);
1230                                 table = genl_dereference(dp->table);
1231                         }
1232                 }
1233
1234                 /* Allocate flow. */
1235                 flow = ovs_flow_alloc();
1236                 if (IS_ERR(flow)) {
1237                         error = PTR_ERR(flow);
1238                         goto err_kfree;
1239                 }
1240                 clear_stats(flow);
1241
1242                 rcu_assign_pointer(flow->sf_acts, acts);
1243
1244                 /* Put flow in bucket. */
1245                 ovs_flow_tbl_insert(table, flow, &key, key_len);
1246
1247                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1248                                                 info->snd_seq,
1249                                                 OVS_FLOW_CMD_NEW);
1250         } else {
1251                 /* We found a matching flow. */
1252                 struct sw_flow_actions *old_acts;
1253
1254                 /* Bail out if we're not allowed to modify an existing flow.
1255                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1256                  * because Generic Netlink treats the latter as a dump
1257                  * request.  We also accept NLM_F_EXCL in case that bug ever
1258                  * gets fixed.
1259                  */
1260                 error = -EEXIST;
1261                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1262                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1263                         goto err_kfree;
1264
1265                 /* Update actions. */
1266                 old_acts = rcu_dereference_protected(flow->sf_acts,
1267                                                      lockdep_genl_is_held());
1268                 rcu_assign_pointer(flow->sf_acts, acts);
1269                 ovs_flow_deferred_free_acts(old_acts);
1270
1271                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1272                                                info->snd_seq, OVS_FLOW_CMD_NEW);
1273
1274                 /* Clear stats. */
1275                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1276                         spin_lock_bh(&flow->lock);
1277                         clear_stats(flow);
1278                         spin_unlock_bh(&flow->lock);
1279                 }
1280         }
1281
1282         if (!IS_ERR(reply))
1283                 genl_notify(reply, genl_info_net(info), info->snd_portid,
1284                            ovs_dp_flow_multicast_group.id, info->nlhdr,
1285                            GFP_KERNEL);
1286         else
1287                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1288                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1289         return 0;
1290
1291 err_kfree:
1292         kfree(acts);
1293 error:
1294         return error;
1295 }
1296
1297 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1298 {
1299         struct nlattr **a = info->attrs;
1300         struct ovs_header *ovs_header = info->userhdr;
1301         struct sw_flow_key key;
1302         struct sk_buff *reply;
1303         struct sw_flow *flow;
1304         struct datapath *dp;
1305         struct flow_table *table;
1306         int err;
1307         int key_len;
1308
1309         if (!a[OVS_FLOW_ATTR_KEY])
1310                 return -EINVAL;
1311         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1312         if (err)
1313                 return err;
1314
1315         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1316         if (!dp)
1317                 return -ENODEV;
1318
1319         table = genl_dereference(dp->table);
1320         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1321         if (!flow)
1322                 return -ENOENT;
1323
1324         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1325                                         info->snd_seq, OVS_FLOW_CMD_NEW);
1326         if (IS_ERR(reply))
1327                 return PTR_ERR(reply);
1328
1329         return genlmsg_reply(reply, info);
1330 }
1331
1332 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1333 {
1334         struct nlattr **a = info->attrs;
1335         struct ovs_header *ovs_header = info->userhdr;
1336         struct sw_flow_key key;
1337         struct sk_buff *reply;
1338         struct sw_flow *flow;
1339         struct datapath *dp;
1340         struct flow_table *table;
1341         int err;
1342         int key_len;
1343
1344         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1345         if (!dp)
1346                 return -ENODEV;
1347
1348         if (!a[OVS_FLOW_ATTR_KEY])
1349                 return flush_flows(dp);
1350
1351         err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1352         if (err)
1353                 return err;
1354
1355         table = genl_dereference(dp->table);
1356         flow = ovs_flow_tbl_lookup(table, &key, key_len);
1357         if (!flow)
1358                 return -ENOENT;
1359
1360         reply = ovs_flow_cmd_alloc_info(flow);
1361         if (!reply)
1362                 return -ENOMEM;
1363
1364         ovs_flow_tbl_remove(table, flow);
1365
1366         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1367                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1368         BUG_ON(err < 0);
1369
1370         ovs_flow_deferred_free(flow);
1371
1372         genl_notify(reply, genl_info_net(info), info->snd_portid,
1373                     ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1374         return 0;
1375 }
1376
1377 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1378 {
1379         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1380         struct datapath *dp;
1381         struct flow_table *table;
1382
1383         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1384         if (!dp)
1385                 return -ENODEV;
1386
1387         table = genl_dereference(dp->table);
1388
1389         for (;;) {
1390                 struct sw_flow *flow;
1391                 u32 bucket, obj;
1392
1393                 bucket = cb->args[0];
1394                 obj = cb->args[1];
1395                 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1396                 if (!flow)
1397                         break;
1398
1399                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1400                                            NETLINK_CB(cb->skb).portid,
1401                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1402                                            OVS_FLOW_CMD_NEW) < 0)
1403                         break;
1404
1405                 cb->args[0] = bucket;
1406                 cb->args[1] = obj;
1407         }
1408         return skb->len;
1409 }
1410
1411 static struct genl_ops dp_flow_genl_ops[] = {
1412         { .cmd = OVS_FLOW_CMD_NEW,
1413           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1414           .policy = flow_policy,
1415           .doit = ovs_flow_cmd_new_or_set
1416         },
1417         { .cmd = OVS_FLOW_CMD_DEL,
1418           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1419           .policy = flow_policy,
1420           .doit = ovs_flow_cmd_del
1421         },
1422         { .cmd = OVS_FLOW_CMD_GET,
1423           .flags = 0,               /* OK for unprivileged users. */
1424           .policy = flow_policy,
1425           .doit = ovs_flow_cmd_get,
1426           .dumpit = ovs_flow_cmd_dump
1427         },
1428         { .cmd = OVS_FLOW_CMD_SET,
1429           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1430           .policy = flow_policy,
1431           .doit = ovs_flow_cmd_new_or_set,
1432         },
1433 };
1434
1435 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1436 #ifdef HAVE_NLA_NUL_STRING
1437         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1438 #endif
1439         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1440 };
1441
1442 static struct genl_family dp_datapath_genl_family = {
1443         .id = GENL_ID_GENERATE,
1444         .hdrsize = sizeof(struct ovs_header),
1445         .name = OVS_DATAPATH_FAMILY,
1446         .version = OVS_DATAPATH_VERSION,
1447         .maxattr = OVS_DP_ATTR_MAX,
1448          SET_NETNSOK
1449 };
1450
1451 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1452         .name = OVS_DATAPATH_MCGROUP
1453 };
1454
1455 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1456                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1457 {
1458         struct ovs_header *ovs_header;
1459         struct ovs_dp_stats dp_stats;
1460         int err;
1461
1462         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1463                                    flags, cmd);
1464         if (!ovs_header)
1465                 goto error;
1466
1467         ovs_header->dp_ifindex = get_dpifindex(dp);
1468
1469         rcu_read_lock();
1470         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1471         rcu_read_unlock();
1472         if (err)
1473                 goto nla_put_failure;
1474
1475         get_dp_stats(dp, &dp_stats);
1476         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1477                 goto nla_put_failure;
1478
1479         return genlmsg_end(skb, ovs_header);
1480
1481 nla_put_failure:
1482         genlmsg_cancel(skb, ovs_header);
1483 error:
1484         return -EMSGSIZE;
1485 }
1486
1487 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1488                                              u32 seq, u8 cmd)
1489 {
1490         struct sk_buff *skb;
1491         int retval;
1492
1493         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1494         if (!skb)
1495                 return ERR_PTR(-ENOMEM);
1496
1497         retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1498         if (retval < 0) {
1499                 kfree_skb(skb);
1500                 return ERR_PTR(retval);
1501         }
1502         return skb;
1503 }
1504
1505 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1506 {
1507         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1508 }
1509
1510 /* Called with genl_mutex and optionally with RTNL lock also. */
1511 static struct datapath *lookup_datapath(struct net *net,
1512                                         struct ovs_header *ovs_header,
1513                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1514 {
1515         struct datapath *dp;
1516
1517         if (!a[OVS_DP_ATTR_NAME])
1518                 dp = get_dp(net, ovs_header->dp_ifindex);
1519         else {
1520                 struct vport *vport;
1521
1522                 rcu_read_lock();
1523                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1524                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1525                 rcu_read_unlock();
1526         }
1527         return dp ? dp : ERR_PTR(-ENODEV);
1528 }
1529
1530 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1531 {
1532         struct nlattr **a = info->attrs;
1533         struct vport_parms parms;
1534         struct sk_buff *reply;
1535         struct datapath *dp;
1536         struct vport *vport;
1537         struct ovs_net *ovs_net;
1538         int err, i;
1539
1540         err = -EINVAL;
1541         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1542                 goto err;
1543
1544         err = ovs_dp_cmd_validate(a);
1545         if (err)
1546                 goto err;
1547
1548         rtnl_lock();
1549
1550         err = -ENOMEM;
1551         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1552         if (dp == NULL)
1553                 goto err_unlock_rtnl;
1554
1555         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1556
1557         /* Allocate table. */
1558         err = -ENOMEM;
1559         rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1560         if (!dp->table)
1561                 goto err_free_dp;
1562
1563         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1564         if (!dp->stats_percpu) {
1565                 err = -ENOMEM;
1566                 goto err_destroy_table;
1567         }
1568
1569         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1570                             GFP_KERNEL);
1571         if (!dp->ports) {
1572                 err = -ENOMEM;
1573                 goto err_destroy_percpu;
1574         }
1575
1576         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1577                 INIT_HLIST_HEAD(&dp->ports[i]);
1578
1579         /* Set up our datapath device. */
1580         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1581         parms.type = OVS_VPORT_TYPE_INTERNAL;
1582         parms.options = NULL;
1583         parms.dp = dp;
1584         parms.port_no = OVSP_LOCAL;
1585         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1586
1587         vport = new_vport(&parms);
1588         if (IS_ERR(vport)) {
1589                 err = PTR_ERR(vport);
1590                 if (err == -EBUSY)
1591                         err = -EEXIST;
1592
1593                 goto err_destroy_ports_array;
1594         }
1595
1596         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1597                                       info->snd_seq, OVS_DP_CMD_NEW);
1598         err = PTR_ERR(reply);
1599         if (IS_ERR(reply))
1600                 goto err_destroy_local_port;
1601
1602         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1603         list_add_tail(&dp->list_node, &ovs_net->dps);
1604
1605         rtnl_unlock();
1606
1607         genl_notify(reply, genl_info_net(info), info->snd_portid,
1608                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1609                     GFP_KERNEL);
1610         return 0;
1611
1612 err_destroy_local_port:
1613         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1614 err_destroy_ports_array:
1615         kfree(dp->ports);
1616 err_destroy_percpu:
1617         free_percpu(dp->stats_percpu);
1618 err_destroy_table:
1619         ovs_flow_tbl_destroy(genl_dereference(dp->table));
1620 err_free_dp:
1621         release_net(ovs_dp_get_net(dp));
1622         kfree(dp);
1623 err_unlock_rtnl:
1624         rtnl_unlock();
1625 err:
1626         return err;
1627 }
1628
1629 /* Called with genl_mutex. */
1630 static void __dp_destroy(struct datapath *dp)
1631 {
1632         int i;
1633
1634         rtnl_lock();
1635
1636         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1637                 struct vport *vport;
1638                 struct hlist_node *n;
1639
1640                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1641                         if (vport->port_no != OVSP_LOCAL)
1642                                 ovs_dp_detach_port(vport);
1643         }
1644
1645         list_del(&dp->list_node);
1646         ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1647
1648         /* rtnl_unlock() will wait until all the references to devices that
1649          * are pending unregistration have been dropped.  We do it here to
1650          * ensure that any internal devices (which contain DP pointers) are
1651          * fully destroyed before freeing the datapath.
1652          */
1653         rtnl_unlock();
1654
1655         call_rcu(&dp->rcu, destroy_dp_rcu);
1656 }
1657
1658 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1659 {
1660         struct sk_buff *reply;
1661         struct datapath *dp;
1662         int err;
1663
1664         err = ovs_dp_cmd_validate(info->attrs);
1665         if (err)
1666                 return err;
1667
1668         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1669         err = PTR_ERR(dp);
1670         if (IS_ERR(dp))
1671                 return err;
1672
1673         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1674                                       info->snd_seq, OVS_DP_CMD_DEL);
1675         err = PTR_ERR(reply);
1676         if (IS_ERR(reply))
1677                 return err;
1678
1679         __dp_destroy(dp);
1680
1681         genl_notify(reply, genl_info_net(info), info->snd_portid,
1682                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1683                     GFP_KERNEL);
1684
1685         return 0;
1686 }
1687
1688 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1689 {
1690         struct sk_buff *reply;
1691         struct datapath *dp;
1692         int err;
1693
1694         err = ovs_dp_cmd_validate(info->attrs);
1695         if (err)
1696                 return err;
1697
1698         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1699         if (IS_ERR(dp))
1700                 return PTR_ERR(dp);
1701
1702         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1703                                       info->snd_seq, OVS_DP_CMD_NEW);
1704         if (IS_ERR(reply)) {
1705                 err = PTR_ERR(reply);
1706                 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1707                                 ovs_dp_datapath_multicast_group.id, err);
1708                 return 0;
1709         }
1710
1711         genl_notify(reply, genl_info_net(info), info->snd_portid,
1712                     ovs_dp_datapath_multicast_group.id, info->nlhdr,
1713                     GFP_KERNEL);
1714
1715         return 0;
1716 }
1717
1718 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1719 {
1720         struct sk_buff *reply;
1721         struct datapath *dp;
1722         int err;
1723
1724         err = ovs_dp_cmd_validate(info->attrs);
1725         if (err)
1726                 return err;
1727
1728         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1729         if (IS_ERR(dp))
1730                 return PTR_ERR(dp);
1731
1732         reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1733                                       info->snd_seq, OVS_DP_CMD_NEW);
1734         if (IS_ERR(reply))
1735                 return PTR_ERR(reply);
1736
1737         return genlmsg_reply(reply, info);
1738 }
1739
1740 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1741 {
1742         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1743         struct datapath *dp;
1744         int skip = cb->args[0];
1745         int i = 0;
1746
1747         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1748                 if (i >= skip &&
1749                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1750                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1751                                          OVS_DP_CMD_NEW) < 0)
1752                         break;
1753                 i++;
1754         }
1755
1756         cb->args[0] = i;
1757
1758         return skb->len;
1759 }
1760
1761 static struct genl_ops dp_datapath_genl_ops[] = {
1762         { .cmd = OVS_DP_CMD_NEW,
1763           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1764           .policy = datapath_policy,
1765           .doit = ovs_dp_cmd_new
1766         },
1767         { .cmd = OVS_DP_CMD_DEL,
1768           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1769           .policy = datapath_policy,
1770           .doit = ovs_dp_cmd_del
1771         },
1772         { .cmd = OVS_DP_CMD_GET,
1773           .flags = 0,               /* OK for unprivileged users. */
1774           .policy = datapath_policy,
1775           .doit = ovs_dp_cmd_get,
1776           .dumpit = ovs_dp_cmd_dump
1777         },
1778         { .cmd = OVS_DP_CMD_SET,
1779           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1780           .policy = datapath_policy,
1781           .doit = ovs_dp_cmd_set,
1782         },
1783 };
1784
1785 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1786 #ifdef HAVE_NLA_NUL_STRING
1787         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1788         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1789 #else
1790         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1791 #endif
1792         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1793         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1794         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1795         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1796 };
1797
1798 static struct genl_family dp_vport_genl_family = {
1799         .id = GENL_ID_GENERATE,
1800         .hdrsize = sizeof(struct ovs_header),
1801         .name = OVS_VPORT_FAMILY,
1802         .version = OVS_VPORT_VERSION,
1803         .maxattr = OVS_VPORT_ATTR_MAX,
1804          SET_NETNSOK
1805 };
1806
1807 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1808         .name = OVS_VPORT_MCGROUP
1809 };
1810
1811 /* Called with RTNL lock or RCU read lock. */
1812 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1813                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1814 {
1815         struct ovs_header *ovs_header;
1816         struct ovs_vport_stats vport_stats;
1817         int err;
1818
1819         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1820                                  flags, cmd);
1821         if (!ovs_header)
1822                 return -EMSGSIZE;
1823
1824         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1825
1826         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1827             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1828             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1829             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1830                 goto nla_put_failure;
1831
1832         ovs_vport_get_stats(vport, &vport_stats);
1833         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1834                     &vport_stats))
1835                 goto nla_put_failure;
1836
1837         err = ovs_vport_get_options(vport, skb);
1838         if (err == -EMSGSIZE)
1839                 goto error;
1840
1841         return genlmsg_end(skb, ovs_header);
1842
1843 nla_put_failure:
1844         err = -EMSGSIZE;
1845 error:
1846         genlmsg_cancel(skb, ovs_header);
1847         return err;
1848 }
1849
1850 /* Called with RTNL lock or RCU read lock. */
1851 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1852                                          u32 seq, u8 cmd)
1853 {
1854         struct sk_buff *skb;
1855         int retval;
1856
1857         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1858         if (!skb)
1859                 return ERR_PTR(-ENOMEM);
1860
1861         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1862         BUG_ON(retval < 0);
1863
1864         return skb;
1865 }
1866
1867 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1868 {
1869         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1870 }
1871
1872 /* Called with RTNL lock or RCU read lock. */
1873 static struct vport *lookup_vport(struct net *net,
1874                                   struct ovs_header *ovs_header,
1875                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1876 {
1877         struct datapath *dp;
1878         struct vport *vport;
1879
1880         if (a[OVS_VPORT_ATTR_NAME]) {
1881                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1882                 if (!vport)
1883                         return ERR_PTR(-ENODEV);
1884                 if (ovs_header->dp_ifindex &&
1885                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1886                         return ERR_PTR(-ENODEV);
1887                 return vport;
1888         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1889                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1890
1891                 if (port_no >= DP_MAX_PORTS)
1892                         return ERR_PTR(-EFBIG);
1893
1894                 dp = get_dp(net, ovs_header->dp_ifindex);
1895                 if (!dp)
1896                         return ERR_PTR(-ENODEV);
1897
1898                 vport = ovs_vport_rtnl_rcu(dp, port_no);
1899                 if (!vport)
1900                         return ERR_PTR(-ENODEV);
1901                 return vport;
1902         } else
1903                 return ERR_PTR(-EINVAL);
1904 }
1905
1906 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1907 {
1908         struct nlattr **a = info->attrs;
1909         struct ovs_header *ovs_header = info->userhdr;
1910         struct vport_parms parms;
1911         struct sk_buff *reply;
1912         struct vport *vport;
1913         struct datapath *dp;
1914         u32 port_no;
1915         int err;
1916
1917         err = -EINVAL;
1918         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1919             !a[OVS_VPORT_ATTR_UPCALL_PID])
1920                 goto exit;
1921
1922         err = ovs_vport_cmd_validate(a);
1923         if (err)
1924                 goto exit;
1925
1926         rtnl_lock();
1927         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1928         err = -ENODEV;
1929         if (!dp)
1930                 goto exit_unlock;
1931
1932         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1933                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1934
1935                 err = -EFBIG;
1936                 if (port_no >= DP_MAX_PORTS)
1937                         goto exit_unlock;
1938
1939                 vport = ovs_vport_rtnl(dp, port_no);
1940                 err = -EBUSY;
1941                 if (vport)
1942                         goto exit_unlock;
1943         } else {
1944                 for (port_no = 1; ; port_no++) {
1945                         if (port_no >= DP_MAX_PORTS) {
1946                                 err = -EFBIG;
1947                                 goto exit_unlock;
1948                         }
1949                         vport = ovs_vport_rtnl(dp, port_no);
1950                         if (!vport)
1951                                 break;
1952                 }
1953         }
1954
1955         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1956         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1957         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1958         parms.dp = dp;
1959         parms.port_no = port_no;
1960         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1961
1962         vport = new_vport(&parms);
1963         err = PTR_ERR(vport);
1964         if (IS_ERR(vport))
1965                 goto exit_unlock;
1966
1967         err = 0;
1968         if (a[OVS_VPORT_ATTR_STATS])
1969                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1970
1971         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1972                                          OVS_VPORT_CMD_NEW);
1973         if (IS_ERR(reply)) {
1974                 err = PTR_ERR(reply);
1975                 ovs_dp_detach_port(vport);
1976                 goto exit_unlock;
1977         }
1978         genl_notify(reply, genl_info_net(info), info->snd_portid,
1979                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1980
1981 exit_unlock:
1982         rtnl_unlock();
1983 exit:
1984         return err;
1985 }
1986
1987 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1988 {
1989         struct nlattr **a = info->attrs;
1990         struct sk_buff *reply;
1991         struct vport *vport;
1992         int err;
1993
1994         err = ovs_vport_cmd_validate(a);
1995         if (err)
1996                 goto exit;
1997
1998         rtnl_lock();
1999         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2000         err = PTR_ERR(vport);
2001         if (IS_ERR(vport))
2002                 goto exit_unlock;
2003
2004         err = 0;
2005         if (a[OVS_VPORT_ATTR_TYPE] &&
2006             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
2007                 err = -EINVAL;
2008
2009         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2010         if (!reply) {
2011                 err = -ENOMEM;
2012                 goto exit_unlock;
2013         }
2014
2015         if (!err && a[OVS_VPORT_ATTR_OPTIONS])
2016                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2017         if (err)
2018                 goto exit_free;
2019
2020         if (a[OVS_VPORT_ATTR_STATS])
2021                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2022
2023         if (a[OVS_VPORT_ATTR_UPCALL_PID])
2024                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2025
2026         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2027                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2028         BUG_ON(err < 0);
2029
2030         genl_notify(reply, genl_info_net(info), info->snd_portid,
2031                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2032
2033         rtnl_unlock();
2034         return 0;
2035
2036 exit_free:
2037         kfree_skb(reply);
2038 exit_unlock:
2039         rtnl_unlock();
2040 exit:
2041         return err;
2042 }
2043
2044 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2045 {
2046         struct nlattr **a = info->attrs;
2047         struct sk_buff *reply;
2048         struct vport *vport;
2049         int err;
2050
2051         err = ovs_vport_cmd_validate(a);
2052         if (err)
2053                 goto exit;
2054
2055         rtnl_lock();
2056         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2057         err = PTR_ERR(vport);
2058         if (IS_ERR(vport))
2059                 goto exit_unlock;
2060
2061         if (vport->port_no == OVSP_LOCAL) {
2062                 err = -EINVAL;
2063                 goto exit_unlock;
2064         }
2065
2066         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2067                                          info->snd_seq, OVS_VPORT_CMD_DEL);
2068         err = PTR_ERR(reply);
2069         if (IS_ERR(reply))
2070                 goto exit_unlock;
2071
2072         err = 0;
2073         ovs_dp_detach_port(vport);
2074
2075         genl_notify(reply, genl_info_net(info), info->snd_portid,
2076                     ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2077
2078 exit_unlock:
2079         rtnl_unlock();
2080 exit:
2081         return err;
2082 }
2083
2084 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2085 {
2086         struct nlattr **a = info->attrs;
2087         struct ovs_header *ovs_header = info->userhdr;
2088         struct sk_buff *reply;
2089         struct vport *vport;
2090         int err;
2091
2092         err = ovs_vport_cmd_validate(a);
2093         if (err)
2094                 goto exit;
2095
2096         rcu_read_lock();
2097         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2098         err = PTR_ERR(vport);
2099         if (IS_ERR(vport))
2100                 goto exit_unlock;
2101
2102         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2103                                          info->snd_seq, OVS_VPORT_CMD_NEW);
2104         err = PTR_ERR(reply);
2105         if (IS_ERR(reply))
2106                 goto exit_unlock;
2107
2108         rcu_read_unlock();
2109
2110         return genlmsg_reply(reply, info);
2111
2112 exit_unlock:
2113         rcu_read_unlock();
2114 exit:
2115         return err;
2116 }
2117
2118 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2119 {
2120         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2121         struct datapath *dp;
2122         int bucket = cb->args[0], skip = cb->args[1];
2123         int i, j = 0;
2124
2125         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2126         if (!dp)
2127                 return -ENODEV;
2128
2129         rcu_read_lock();
2130         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2131                 struct vport *vport;
2132
2133                 j = 0;
2134                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2135                         if (j >= skip &&
2136                             ovs_vport_cmd_fill_info(vport, skb,
2137                                                     NETLINK_CB(cb->skb).portid,
2138                                                     cb->nlh->nlmsg_seq,
2139                                                     NLM_F_MULTI,
2140                                                     OVS_VPORT_CMD_NEW) < 0)
2141                                 goto out;
2142
2143                         j++;
2144                 }
2145                 skip = 0;
2146         }
2147 out:
2148         rcu_read_unlock();
2149
2150         cb->args[0] = i;
2151         cb->args[1] = j;
2152
2153         return skb->len;
2154 }
2155
2156 static struct genl_ops dp_vport_genl_ops[] = {
2157         { .cmd = OVS_VPORT_CMD_NEW,
2158           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2159           .policy = vport_policy,
2160           .doit = ovs_vport_cmd_new
2161         },
2162         { .cmd = OVS_VPORT_CMD_DEL,
2163           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2164           .policy = vport_policy,
2165           .doit = ovs_vport_cmd_del
2166         },
2167         { .cmd = OVS_VPORT_CMD_GET,
2168           .flags = 0,               /* OK for unprivileged users. */
2169           .policy = vport_policy,
2170           .doit = ovs_vport_cmd_get,
2171           .dumpit = ovs_vport_cmd_dump
2172         },
2173         { .cmd = OVS_VPORT_CMD_SET,
2174           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2175           .policy = vport_policy,
2176           .doit = ovs_vport_cmd_set,
2177         },
2178 };
2179
2180 struct genl_family_and_ops {
2181         struct genl_family *family;
2182         struct genl_ops *ops;
2183         int n_ops;
2184         struct genl_multicast_group *group;
2185 };
2186
2187 static const struct genl_family_and_ops dp_genl_families[] = {
2188         { &dp_datapath_genl_family,
2189           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2190           &ovs_dp_datapath_multicast_group },
2191         { &dp_vport_genl_family,
2192           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2193           &ovs_dp_vport_multicast_group },
2194         { &dp_flow_genl_family,
2195           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2196           &ovs_dp_flow_multicast_group },
2197         { &dp_packet_genl_family,
2198           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2199           NULL },
2200 };
2201
2202 static void dp_unregister_genl(int n_families)
2203 {
2204         int i;
2205
2206         for (i = 0; i < n_families; i++)
2207                 genl_unregister_family(dp_genl_families[i].family);
2208 }
2209
2210 static int dp_register_genl(void)
2211 {
2212         int n_registered;
2213         int err;
2214         int i;
2215
2216         n_registered = 0;
2217         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2218                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2219
2220                 err = genl_register_family_with_ops(f->family, f->ops,
2221                                                     f->n_ops);
2222                 if (err)
2223                         goto error;
2224                 n_registered++;
2225
2226                 if (f->group) {
2227                         err = genl_register_mc_group(f->family, f->group);
2228                         if (err)
2229                                 goto error;
2230                 }
2231         }
2232
2233         return 0;
2234
2235 error:
2236         dp_unregister_genl(n_registered);
2237         return err;
2238 }
2239
2240 static int __rehash_flow_table(void *dummy)
2241 {
2242         struct datapath *dp;
2243         struct net *net;
2244
2245         rtnl_lock();
2246         for_each_net(net) {
2247                 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2248
2249                 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2250                         struct flow_table *old_table = genl_dereference(dp->table);
2251                         struct flow_table *new_table;
2252
2253                         new_table = ovs_flow_tbl_rehash(old_table);
2254                         if (!IS_ERR(new_table)) {
2255                                 rcu_assign_pointer(dp->table, new_table);
2256                                 ovs_flow_tbl_deferred_destroy(old_table);
2257                         }
2258                 }
2259         }
2260         rtnl_unlock();
2261         return 0;
2262 }
2263
2264 static void rehash_flow_table(struct work_struct *work)
2265 {
2266         genl_exec(__rehash_flow_table, NULL);
2267         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2268 }
2269
2270 static int dp_destroy_all(void *data)
2271 {
2272         struct datapath *dp, *dp_next;
2273         struct ovs_net *ovs_net = data;
2274
2275         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2276                 __dp_destroy(dp);
2277
2278         return 0;
2279 }
2280
2281 static int __net_init ovs_init_net(struct net *net)
2282 {
2283         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2284
2285         INIT_LIST_HEAD(&ovs_net->dps);
2286         return 0;
2287 }
2288
2289 static void __net_exit ovs_exit_net(struct net *net)
2290 {
2291         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2292
2293         genl_exec(dp_destroy_all, ovs_net);
2294 }
2295
2296 static struct pernet_operations ovs_net_ops = {
2297         .init = ovs_init_net,
2298         .exit = ovs_exit_net,
2299         .id   = &ovs_net_id,
2300         .size = sizeof(struct ovs_net),
2301 };
2302
2303 static int __init dp_init(void)
2304 {
2305         int err;
2306
2307         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2308
2309         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2310                 VERSION);
2311
2312         err = genl_exec_init();
2313         if (err)
2314                 goto error;
2315
2316         err = ovs_workqueues_init();
2317         if (err)
2318                 goto error_genl_exec;
2319
2320         err = ovs_flow_init();
2321         if (err)
2322                 goto error_wq;
2323
2324         err = ovs_vport_init();
2325         if (err)
2326                 goto error_flow_exit;
2327
2328         err = register_pernet_device(&ovs_net_ops);
2329         if (err)
2330                 goto error_vport_exit;
2331
2332         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2333         if (err)
2334                 goto error_netns_exit;
2335
2336         err = dp_register_genl();
2337         if (err < 0)
2338                 goto error_unreg_notifier;
2339
2340         schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2341
2342         return 0;
2343
2344 error_unreg_notifier:
2345         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2346 error_netns_exit:
2347         unregister_pernet_device(&ovs_net_ops);
2348 error_vport_exit:
2349         ovs_vport_exit();
2350 error_flow_exit:
2351         ovs_flow_exit();
2352 error_wq:
2353         ovs_workqueues_exit();
2354 error_genl_exec:
2355         genl_exec_exit();
2356 error:
2357         return err;
2358 }
2359
2360 static void dp_cleanup(void)
2361 {
2362         cancel_delayed_work_sync(&rehash_flow_wq);
2363         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2364         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2365         unregister_pernet_device(&ovs_net_ops);
2366         rcu_barrier();
2367         ovs_vport_exit();
2368         ovs_flow_exit();
2369         ovs_workqueues_exit();
2370         genl_exec_exit();
2371 }
2372
2373 module_init(dp_init);
2374 module_exit(dp_cleanup);
2375
2376 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2377 MODULE_LICENSE("GPL");
2378 MODULE_VERSION(VERSION);