net/openvswitch/datapath.c

   1 /*
   2  * Copyright (c) 2007-2014 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/openvswitch.h>
  48 #include <linux/rculist.h>
  49 #include <linux/dmi.h>
  50 #include <net/genetlink.h>
  51 #include <net/net_namespace.h>
  52 #include <net/netns/generic.h>
  53
  54 #include "datapath.h"
  55 #include "flow.h"
  56 #include "flow_table.h"
  57 #include "flow_netlink.h"
  58 #include "vport-internal_dev.h"
  59 #include "vport-netdev.h"
  60
  61 int ovs_net_id __read_mostly;
  62 EXPORT_SYMBOL_GPL(ovs_net_id);
  63
  64 static struct genl_family dp_packet_genl_family;
  65 static struct genl_family dp_flow_genl_family;
  66 static struct genl_family dp_datapath_genl_family;
  67
  68 static const struct nla_policy flow_policy[];
  69
  70 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
  71         .name = OVS_FLOW_MCGROUP,
  72 };
  73
  74 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  75         .name = OVS_DATAPATH_MCGROUP,
  76 };
  77
  78 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
  79         .name = OVS_VPORT_MCGROUP,
  80 };
  81
  82 /* Check if need to build a reply message.
  83  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  84 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  85                             unsigned int group)
  86 {
  87         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  88                genl_has_listeners(family, genl_info_net(info), group);
  89 }
  90
  91 static void ovs_notify(struct genl_family *family,
  92                        struct sk_buff *skb, struct genl_info *info)
  93 {
  94         genl_notify(family, skb, info, 0, GFP_KERNEL);
  95 }
  96
  97 /**
  98  * DOC: Locking:
  99  *
 100  * All writes e.g. Writes to device state (add/remove datapath, port, set
 101  * operations on vports, etc.), Writes to other state (flow table
 102  * modifications, set miscellaneous datapath parameters, etc.) are protected
 103  * by ovs_lock.
 104  *
 105  * Reads are protected by RCU.
 106  *
 107  * There are a few special cases (mostly stats) that have their own
 108  * synchronization but they nest under all of above and don't interact with
 109  * each other.
 110  *
 111  * The RTNL lock nests inside ovs_mutex.
 112  */
 113
 114 static DEFINE_MUTEX(ovs_mutex);
 115
 116 void ovs_lock(void)
 117 {
 118         mutex_lock(&ovs_mutex);
 119 }
 120
 121 void ovs_unlock(void)
 122 {
 123         mutex_unlock(&ovs_mutex);
 124 }
 125
 126 #ifdef CONFIG_LOCKDEP
 127 int lockdep_ovsl_is_held(void)
 128 {
 129         if (debug_locks)
 130                 return lockdep_is_held(&ovs_mutex);
 131         else
 132                 return 1;
 133 }
 134 EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held);
 135 #endif
 136
 137 static struct vport *new_vport(const struct vport_parms *);
 138 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 139                              const struct sw_flow_key *,
 140                              const struct dp_upcall_info *,
 141                              uint32_t cutlen);
 142 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 143                                   const struct sw_flow_key *,
 144                                   const struct dp_upcall_info *,
 145                                   uint32_t cutlen);
 146
 147 /* Must be called with rcu_read_lock. */
 148 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
 149 {
 150         struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
 151
 152         if (dev) {
 153                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 154                 if (vport)
 155                         return vport->dp;
 156         }
 157
 158         return NULL;
 159 }
 160
 161 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
 162  * returned dp pointer valid.
 163  */
 164 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
 165 {
 166         struct datapath *dp;
 167
 168         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
 169         rcu_read_lock();
 170         dp = get_dp_rcu(net, dp_ifindex);
 171         rcu_read_unlock();
 172
 173         return dp;
 174 }
 175
 176 /* Must be called with rcu_read_lock or ovs_mutex. */
 177 const char *ovs_dp_name(const struct datapath *dp)
 178 {
 179         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 180         return ovs_vport_name(vport);
 181 }
 182
 183 static int get_dpifindex(const struct datapath *dp)
 184 {
 185         struct vport *local;
 186         int ifindex;
 187
 188         rcu_read_lock();
 189
 190         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 191         if (local)
 192                 ifindex = local->dev->ifindex;
 193         else
 194                 ifindex = 0;
 195
 196         rcu_read_unlock();
 197
 198         return ifindex;
 199 }
 200
 201 static void destroy_dp_rcu(struct rcu_head *rcu)
 202 {
 203         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 204
 205         ovs_flow_tbl_destroy(&dp->table);
 206         free_percpu(dp->stats_percpu);
 207         kfree(dp->ports);
 208         kfree(dp);
 209 }
 210
 211 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 212                                             u16 port_no)
 213 {
 214         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 215 }
 216
 217 /* Called with ovs_mutex or RCU read lock. */
 218 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 219 {
 220         struct vport *vport;
 221         struct hlist_head *head;
 222
 223         head = vport_hash_bucket(dp, port_no);
 224         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 225                 if (vport->port_no == port_no)
 226                         return vport;
 227         }
 228         return NULL;
 229 }
 230
 231 /* Called with ovs_mutex. */
 232 static struct vport *new_vport(const struct vport_parms *parms)
 233 {
 234         struct vport *vport;
 235
 236         vport = ovs_vport_add(parms);
 237         if (!IS_ERR(vport)) {
 238                 struct datapath *dp = parms->dp;
 239                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 240
 241                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 242         }
 243         return vport;
 244 }
 245
 246 void ovs_dp_detach_port(struct vport *p)
 247 {
 248         ASSERT_OVSL();
 249
 250         /* First drop references to device. */
 251         hlist_del_rcu(&p->dp_hash_node);
 252
 253         /* Then destroy it. */
 254         ovs_vport_del(p);
 255 }
 256
 257 /* Must be called with rcu_read_lock. */
 258 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 259 {
 260         const struct vport *p = OVS_CB(skb)->input_vport;
 261         struct datapath *dp = p->dp;
 262         struct sw_flow *flow;
 263         struct sw_flow_actions *sf_acts;
 264         struct dp_stats_percpu *stats;
 265         u64 *stats_counter;
 266         u32 n_mask_hit;
 267
 268         stats = this_cpu_ptr(dp->stats_percpu);
 269
 270         /* Look up flow. */
 271         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
 272         if (unlikely(!flow)) {
 273                 struct dp_upcall_info upcall;
 274                 int error;
 275
 276                 memset(&upcall, 0, sizeof(upcall));
 277                 upcall.cmd = OVS_PACKET_CMD_MISS;
 278                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 279                 upcall.mru = OVS_CB(skb)->mru;
 280                 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
 281                 switch (error) {
 282                 case 0:
 283                 case -EAGAIN:
 284                 case -ERESTARTSYS:
 285                 case -EINTR:
 286                         consume_skb(skb);
 287                         break;
 288                 default:
 289                         kfree_skb(skb);
 290                         break;
 291                 }
 292                 stats_counter = &stats->n_missed;
 293                 goto out;
 294         }
 295
 296         ovs_flow_stats_update(flow, key->tp.flags, skb);
 297         sf_acts = rcu_dereference(flow->sf_acts);
 298         ovs_execute_actions(dp, skb, sf_acts, key);
 299
 300         stats_counter = &stats->n_hit;
 301
 302 out:
 303         /* Update datapath statistics. */
 304         u64_stats_update_begin(&stats->syncp);
 305         (*stats_counter)++;
 306         stats->n_mask_hit += n_mask_hit;
 307         u64_stats_update_end(&stats->syncp);
 308 }
 309
 310 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 311                   const struct sw_flow_key *key,
 312                   const struct dp_upcall_info *upcall_info,
 313                   uint32_t cutlen)
 314 {
 315         struct dp_stats_percpu *stats;
 316         int err;
 317
 318         if (upcall_info->portid == 0) {
 319                 err = -ENOTCONN;
 320                 goto err;
 321         }
 322
 323         if (!skb_is_gso(skb))
 324                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 325         else
 326                 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
 327         if (err)
 328                 goto err;
 329
 330         return 0;
 331
 332 err:
 333         stats = this_cpu_ptr(dp->stats_percpu);
 334
 335         u64_stats_update_begin(&stats->syncp);
 336         stats->n_lost++;
 337         u64_stats_update_end(&stats->syncp);
 338
 339         return err;
 340 }
 341
 342 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 343                              const struct sw_flow_key *key,
 344                              const struct dp_upcall_info *upcall_info,
 345                                  uint32_t cutlen)
 346 {
 347         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 348         struct sw_flow_key later_key;
 349         struct sk_buff *segs, *nskb;
 350         int err;
 351
 352         BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
 353         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 354         if (IS_ERR(segs))
 355                 return PTR_ERR(segs);
 356         if (segs == NULL)
 357                 return -EINVAL;
 358
 359         if (gso_type & SKB_GSO_UDP) {
 360                 /* The initial flow key extracted by ovs_flow_key_extract()
 361                  * in this case is for a first fragment, so we need to
 362                  * properly mark later fragments.
 363                  */
 364                 later_key = *key;
 365                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 366         }
 367
 368         /* Queue all of the segments. */
 369         skb = segs;
 370         do {
 371                 if (gso_type & SKB_GSO_UDP && skb != segs)
 372                         key = &later_key;
 373
 374                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 375                 if (err)
 376                         break;
 377
 378         } while ((skb = skb->next));
 379
 380         /* Free all of the segments. */
 381         skb = segs;
 382         do {
 383                 nskb = skb->next;
 384                 if (err)
 385                         kfree_skb(skb);
 386                 else
 387                         consume_skb(skb);
 388         } while ((skb = nskb));
 389         return err;
 390 }
 391
 392 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 393                               unsigned int hdrlen, int actions_attrlen)
 394 {
 395         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 396                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 397                 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 398                 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
 399
 400         /* OVS_PACKET_ATTR_USERDATA */
 401         if (upcall_info->userdata)
 402                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 403
 404         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 405         if (upcall_info->egress_tun_info)
 406                 size += nla_total_size(ovs_tun_key_attr_size());
 407
 408         /* OVS_PACKET_ATTR_ACTIONS */
 409         if (upcall_info->actions_len)
 410                 size += nla_total_size(actions_attrlen);
 411
 412         /* OVS_PACKET_ATTR_MRU */
 413         if (upcall_info->mru)
 414                 size += nla_total_size(sizeof(upcall_info->mru));
 415
 416         return size;
 417 }
 418
 419 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 420 {
 421         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 422                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 423
 424                 if (plen > 0)
 425                         memset(skb_put(skb, plen), 0, plen);
 426         }
 427 }
 428
 429 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 430                                   const struct sw_flow_key *key,
 431                                   const struct dp_upcall_info *upcall_info,
 432                                   uint32_t cutlen)
 433 {
 434         struct ovs_header *upcall;
 435         struct sk_buff *nskb = NULL;
 436         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 437         struct nlattr *nla;
 438         size_t len;
 439         unsigned int hlen;
 440         int err, dp_ifindex;
 441
 442         dp_ifindex = get_dpifindex(dp);
 443         if (!dp_ifindex)
 444                 return -ENODEV;
 445
 446         if (skb_vlan_tag_present(skb)) {
 447                 nskb = skb_clone(skb, GFP_ATOMIC);
 448                 if (!nskb)
 449                         return -ENOMEM;
 450
 451                 nskb = __vlan_hwaccel_push_inside(nskb);
 452                 if (!nskb)
 453                         return -ENOMEM;
 454
 455                 skb = nskb;
 456         }
 457
 458         if (nla_attr_size(skb->len) > USHRT_MAX) {
 459                 err = -EFBIG;
 460                 goto out;
 461         }
 462
 463         /* Complete checksum if needed */
 464         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 465             (err = skb_checksum_help(skb)))
 466                 goto out;
 467
 468         /* Older versions of OVS user space enforce alignment of the last
 469          * Netlink attribute to NLA_ALIGNTO which would require extensive
 470          * padding logic. Only perform zerocopy if padding is not required.
 471          */
 472         if (dp->user_features & OVS_DP_F_UNALIGNED)
 473                 hlen = skb_zerocopy_headlen(skb);
 474         else
 475                 hlen = skb->len;
 476
 477         len = upcall_msg_size(upcall_info, hlen - cutlen,
 478                               OVS_CB(skb)->acts_origlen);
 479         user_skb = genlmsg_new(len, GFP_ATOMIC);
 480         if (!user_skb) {
 481                 err = -ENOMEM;
 482                 goto out;
 483         }
 484
 485         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 486                              0, upcall_info->cmd);
 487         upcall->dp_ifindex = dp_ifindex;
 488
 489         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 490         BUG_ON(err);
 491
 492         if (upcall_info->userdata)
 493                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 494                           nla_len(upcall_info->userdata),
 495                           nla_data(upcall_info->userdata));
 496
 497         if (upcall_info->egress_tun_info) {
 498                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 499                 err = ovs_nla_put_tunnel_info(user_skb,
 500                                               upcall_info->egress_tun_info);
 501                 BUG_ON(err);
 502                 nla_nest_end(user_skb, nla);
 503         }
 504
 505         if (upcall_info->actions_len) {
 506                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 507                 err = ovs_nla_put_actions(upcall_info->actions,
 508                                           upcall_info->actions_len,
 509                                           user_skb);
 510                 if (!err)
 511                         nla_nest_end(user_skb, nla);
 512                 else
 513                         nla_nest_cancel(user_skb, nla);
 514         }
 515
 516         /* Add OVS_PACKET_ATTR_MRU */
 517         if (upcall_info->mru) {
 518                 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
 519                                 upcall_info->mru)) {
 520                         err = -ENOBUFS;
 521                         goto out;
 522                 }
 523                 pad_packet(dp, user_skb);
 524         }
 525
 526         /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 527         if (cutlen > 0) {
 528                 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
 529                                 skb->len)) {
 530                         err = -ENOBUFS;
 531                         goto out;
 532                 }
 533                 pad_packet(dp, user_skb);
 534         }
 535
 536         /* Only reserve room for attribute header, packet data is added
 537          * in skb_zerocopy() */
 538         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 539                 err = -ENOBUFS;
 540                 goto out;
 541         }
 542         nla->nla_len = nla_attr_size(skb->len - cutlen);
 543
 544         err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 545         if (err)
 546                 goto out;
 547
 548         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 549         pad_packet(dp, user_skb);
 550
 551         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 552
 553         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 554         user_skb = NULL;
 555 out:
 556         if (err)
 557                 skb_tx_error(skb);
 558         consume_skb(user_skb);
 559         consume_skb(nskb);
 560
 561         return err;
 562 }
 563
 564 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 565 {
 566         struct ovs_header *ovs_header = info->userhdr;
 567         struct net *net = sock_net(skb->sk);
 568         struct nlattr **a = info->attrs;
 569         struct sw_flow_actions *acts;
 570         struct sk_buff *packet;
 571         struct sw_flow *flow;
 572         struct sw_flow_actions *sf_acts;
 573         struct datapath *dp;
 574         struct ethhdr *eth;
 575         struct vport *input_vport;
 576         u16 mru = 0;
 577         int len;
 578         int err;
 579         bool log = !a[OVS_PACKET_ATTR_PROBE];
 580
 581         err = -EINVAL;
 582         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 583             !a[OVS_PACKET_ATTR_ACTIONS])
 584                 goto err;
 585
 586         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 587         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 588         err = -ENOMEM;
 589         if (!packet)
 590                 goto err;
 591         skb_reserve(packet, NET_IP_ALIGN);
 592
 593         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 594
 595         skb_reset_mac_header(packet);
 596         eth = eth_hdr(packet);
 597
 598         /* Normally, setting the skb 'protocol' field would be handled by a
 599          * call to eth_type_trans(), but it assumes there's a sending
 600          * device, which we may not have. */
 601         if (eth_proto_is_802_3(eth->h_proto))
 602                 packet->protocol = eth->h_proto;
 603         else
 604                 packet->protocol = htons(ETH_P_802_2);
 605
 606         /* Set packet's mru */
 607         if (a[OVS_PACKET_ATTR_MRU]) {
 608                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 609                 packet->ignore_df = 1;
 610         }
 611         OVS_CB(packet)->mru = mru;
 612
 613         /* Build an sw_flow for sending this packet. */
 614         flow = ovs_flow_alloc();
 615         err = PTR_ERR(flow);
 616         if (IS_ERR(flow))
 617                 goto err_kfree_skb;
 618
 619         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 620                                              packet, &flow->key, log);
 621         if (err)
 622                 goto err_flow_free;
 623
 624         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 625                                    &flow->key, &acts, log);
 626         if (err)
 627                 goto err_flow_free;
 628
 629         rcu_assign_pointer(flow->sf_acts, acts);
 630         packet->priority = flow->key.phy.priority;
 631         packet->mark = flow->key.phy.skb_mark;
 632
 633         rcu_read_lock();
 634         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 635         err = -ENODEV;
 636         if (!dp)
 637                 goto err_unlock;
 638
 639         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 640         if (!input_vport)
 641                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 642
 643         if (!input_vport)
 644                 goto err_unlock;
 645
 646         packet->dev = input_vport->dev;
 647         OVS_CB(packet)->input_vport = input_vport;
 648         sf_acts = rcu_dereference(flow->sf_acts);
 649
 650         local_bh_disable();
 651         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 652         local_bh_enable();
 653         rcu_read_unlock();
 654
 655         ovs_flow_free(flow, false);
 656         return err;
 657
 658 err_unlock:
 659         rcu_read_unlock();
 660 err_flow_free:
 661         ovs_flow_free(flow, false);
 662 err_kfree_skb:
 663         kfree_skb(packet);
 664 err:
 665         return err;
 666 }
 667
 668 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 669         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 670         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 671         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 672         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 673         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 674 };
 675
 676 static const struct genl_ops dp_packet_genl_ops[] = {
 677         { .cmd = OVS_PACKET_CMD_EXECUTE,
 678           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 679           .policy = packet_policy,
 680           .doit = ovs_packet_cmd_execute
 681         }
 682 };
 683
 684 static struct genl_family dp_packet_genl_family = {
 685         .id = GENL_ID_GENERATE,
 686         .hdrsize = sizeof(struct ovs_header),
 687         .name = OVS_PACKET_FAMILY,
 688         .version = OVS_PACKET_VERSION,
 689         .maxattr = OVS_PACKET_ATTR_MAX,
 690         .netnsok = true,
 691         .parallel_ops = true,
 692         .ops = dp_packet_genl_ops,
 693         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 694 };
 695
 696 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 697                          struct ovs_dp_megaflow_stats *mega_stats)
 698 {
 699         int i;
 700
 701         memset(mega_stats, 0, sizeof(*mega_stats));
 702
 703         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 704         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 705
 706         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 707
 708         for_each_possible_cpu(i) {
 709                 const struct dp_stats_percpu *percpu_stats;
 710                 struct dp_stats_percpu local_stats;
 711                 unsigned int start;
 712
 713                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 714
 715                 do {
 716                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 717                         local_stats = *percpu_stats;
 718                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 719
 720                 stats->n_hit += local_stats.n_hit;
 721                 stats->n_missed += local_stats.n_missed;
 722                 stats->n_lost += local_stats.n_lost;
 723                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 724         }
 725 }
 726
 727 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 728 {
 729         return ovs_identifier_is_ufid(sfid) &&
 730                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 731 }
 732
 733 static bool should_fill_mask(uint32_t ufid_flags)
 734 {
 735         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 736 }
 737
 738 static bool should_fill_actions(uint32_t ufid_flags)
 739 {
 740         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 741 }
 742
 743 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 744                                     const struct sw_flow_id *sfid,
 745                                     uint32_t ufid_flags)
 746 {
 747         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 748
 749         /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
 750          * see ovs_nla_put_identifier()
 751          */
 752         if (sfid && ovs_identifier_is_ufid(sfid))
 753                 len += nla_total_size(sfid->ufid_len);
 754         else
 755                 len += nla_total_size(ovs_key_attr_size());
 756
 757         /* OVS_FLOW_ATTR_KEY */
 758         if (!sfid || should_fill_key(sfid, ufid_flags))
 759                 len += nla_total_size(ovs_key_attr_size());
 760
 761         /* OVS_FLOW_ATTR_MASK */
 762         if (should_fill_mask(ufid_flags))
 763                 len += nla_total_size(ovs_key_attr_size());
 764
 765         /* OVS_FLOW_ATTR_ACTIONS */
 766         if (should_fill_actions(ufid_flags))
 767                 len += nla_total_size(acts->orig_len);
 768
 769         return len
 770                 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 771                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 772                 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 773 }
 774
 775 /* Called with ovs_mutex or RCU read lock. */
 776 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 777                                    struct sk_buff *skb)
 778 {
 779         struct ovs_flow_stats stats;
 780         __be16 tcp_flags;
 781         unsigned long used;
 782
 783         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 784
 785         if (used &&
 786             nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 787                               OVS_FLOW_ATTR_PAD))
 788                 return -EMSGSIZE;
 789
 790         if (stats.n_packets &&
 791             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 792                           sizeof(struct ovs_flow_stats), &stats,
 793                           OVS_FLOW_ATTR_PAD))
 794                 return -EMSGSIZE;
 795
 796         if ((u8)ntohs(tcp_flags) &&
 797              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 798                 return -EMSGSIZE;
 799
 800         return 0;
 801 }
 802
 803 /* Called with ovs_mutex or RCU read lock. */
 804 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 805                                      struct sk_buff *skb, int skb_orig_len)
 806 {
 807         struct nlattr *start;
 808         int err;
 809
 810         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 811          * this is the first flow to be dumped into 'skb'.  This is unusual for
 812          * Netlink but individual action lists can be longer than
 813          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 814          * The userspace caller can always fetch the actions separately if it
 815          * really wants them.  (Most userspace callers in fact don't care.)
 816          *
 817          * This can only fail for dump operations because the skb is always
 818          * properly sized for single flows.
 819          */
 820         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 821         if (start) {
 822                 const struct sw_flow_actions *sf_acts;
 823
 824                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 825                 err = ovs_nla_put_actions(sf_acts->actions,
 826                                           sf_acts->actions_len, skb);
 827
 828                 if (!err)
 829                         nla_nest_end(skb, start);
 830                 else {
 831                         if (skb_orig_len)
 832                                 return err;
 833
 834                         nla_nest_cancel(skb, start);
 835                 }
 836         } else if (skb_orig_len) {
 837                 return -EMSGSIZE;
 838         }
 839
 840         return 0;
 841 }
 842
 843 /* Called with ovs_mutex or RCU read lock. */
 844 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 845                                   struct sk_buff *skb, u32 portid,
 846                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 847 {
 848         const int skb_orig_len = skb->len;
 849         struct ovs_header *ovs_header;
 850         int err;
 851
 852         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 853                                  flags, cmd);
 854         if (!ovs_header)
 855                 return -EMSGSIZE;
 856
 857         ovs_header->dp_ifindex = dp_ifindex;
 858
 859         err = ovs_nla_put_identifier(flow, skb);
 860         if (err)
 861                 goto error;
 862
 863         if (should_fill_key(&flow->id, ufid_flags)) {
 864                 err = ovs_nla_put_masked_key(flow, skb);
 865                 if (err)
 866                         goto error;
 867         }
 868
 869         if (should_fill_mask(ufid_flags)) {
 870                 err = ovs_nla_put_mask(flow, skb);
 871                 if (err)
 872                         goto error;
 873         }
 874
 875         err = ovs_flow_cmd_fill_stats(flow, skb);
 876         if (err)
 877                 goto error;
 878
 879         if (should_fill_actions(ufid_flags)) {
 880                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 881                 if (err)
 882                         goto error;
 883         }
 884
 885         genlmsg_end(skb, ovs_header);
 886         return 0;
 887
 888 error:
 889         genlmsg_cancel(skb, ovs_header);
 890         return err;
 891 }
 892
 893 /* May not be called with RCU read lock. */
 894 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 895                                                const struct sw_flow_id *sfid,
 896                                                struct genl_info *info,
 897                                                bool always,
 898                                                uint32_t ufid_flags)
 899 {
 900         struct sk_buff *skb;
 901         size_t len;
 902
 903         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 904                 return NULL;
 905
 906         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 907         skb = genlmsg_new(len, GFP_KERNEL);
 908         if (!skb)
 909                 return ERR_PTR(-ENOMEM);
 910
 911         return skb;
 912 }
 913
 914 /* Called with ovs_mutex. */
 915 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 916                                                int dp_ifindex,
 917                                                struct genl_info *info, u8 cmd,
 918                                                bool always, u32 ufid_flags)
 919 {
 920         struct sk_buff *skb;
 921         int retval;
 922
 923         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 924                                       &flow->id, info, always, ufid_flags);
 925         if (IS_ERR_OR_NULL(skb))
 926                 return skb;
 927
 928         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 929                                         info->snd_portid, info->snd_seq, 0,
 930                                         cmd, ufid_flags);
 931         if (WARN_ON_ONCE(retval < 0)) {
 932                 kfree_skb(skb);
 933                 skb = ERR_PTR(retval);
 934         }
 935         return skb;
 936 }
 937
 938 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 939 {
 940         struct net *net = sock_net(skb->sk);
 941         struct nlattr **a = info->attrs;
 942         struct ovs_header *ovs_header = info->userhdr;
 943         struct sw_flow *flow = NULL, *new_flow;
 944         struct sw_flow_mask mask;
 945         struct sk_buff *reply;
 946         struct datapath *dp;
 947         struct sw_flow_actions *acts;
 948         struct sw_flow_match match;
 949         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 950         int error;
 951         bool log = !a[OVS_FLOW_ATTR_PROBE];
 952
 953         /* Must have key and actions. */
 954         error = -EINVAL;
 955         if (!a[OVS_FLOW_ATTR_KEY]) {
 956                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 957                 goto error;
 958         }
 959         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 960                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 961                 goto error;
 962         }
 963
 964         /* Most of the time we need to allocate a new flow, do it before
 965          * locking.
 966          */
 967         new_flow = ovs_flow_alloc();
 968         if (IS_ERR(new_flow)) {
 969                 error = PTR_ERR(new_flow);
 970                 goto error;
 971         }
 972
 973         /* Extract key. */
 974         ovs_match_init(&match, &new_flow->key, false, &mask);
 975         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 976                                   a[OVS_FLOW_ATTR_MASK], log);
 977         if (error)
 978                 goto err_kfree_flow;
 979
 980         /* Extract flow identifier. */
 981         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 982                                        &new_flow->key, log);
 983         if (error)
 984                 goto err_kfree_flow;
 985
 986         /* unmasked key is needed to match when ufid is not used. */
 987         if (ovs_identifier_is_key(&new_flow->id))
 988                 match.key = new_flow->id.unmasked_key;
 989
 990         ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
 991
 992         /* Validate actions. */
 993         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 994                                      &new_flow->key, &acts, log);
 995         if (error) {
 996                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 997                 goto err_kfree_flow;
 998         }
 999
1000         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
1001                                         ufid_flags);
1002         if (IS_ERR(reply)) {
1003                 error = PTR_ERR(reply);
1004                 goto err_kfree_acts;
1005         }
1006
1007         ovs_lock();
1008         dp = get_dp(net, ovs_header->dp_ifindex);
1009         if (unlikely(!dp)) {
1010                 error = -ENODEV;
1011                 goto err_unlock_ovs;
1012         }
1013
1014         /* Check if this is a duplicate flow */
1015         if (ovs_identifier_is_ufid(&new_flow->id))
1016                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1017         if (!flow)
1018                 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
1019         if (likely(!flow)) {
1020                 rcu_assign_pointer(new_flow->sf_acts, acts);
1021
1022                 /* Put flow in bucket. */
1023                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1024                 if (unlikely(error)) {
1025                         acts = NULL;
1026                         goto err_unlock_ovs;
1027                 }
1028
1029                 if (unlikely(reply)) {
1030                         error = ovs_flow_cmd_fill_info(new_flow,
1031                                                        ovs_header->dp_ifindex,
1032                                                        reply, info->snd_portid,
1033                                                        info->snd_seq, 0,
1034                                                        OVS_FLOW_CMD_NEW,
1035                                                        ufid_flags);
1036                         BUG_ON(error < 0);
1037                 }
1038                 ovs_unlock();
1039         } else {
1040                 struct sw_flow_actions *old_acts;
1041
1042                 /* Bail out if we're not allowed to modify an existing flow.
1043                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1044                  * because Generic Netlink treats the latter as a dump
1045                  * request.  We also accept NLM_F_EXCL in case that bug ever
1046                  * gets fixed.
1047                  */
1048                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1049                                                          | NLM_F_EXCL))) {
1050                         error = -EEXIST;
1051                         goto err_unlock_ovs;
1052                 }
1053                 /* The flow identifier has to be the same for flow updates.
1054                  * Look for any overlapping flow.
1055                  */
1056                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1057                         if (ovs_identifier_is_key(&flow->id))
1058                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1059                                                                  &match);
1060                         else /* UFID matches but key is different */
1061                                 flow = NULL;
1062                         if (!flow) {
1063                                 error = -ENOENT;
1064                                 goto err_unlock_ovs;
1065                         }
1066                 }
1067                 /* Update actions. */
1068                 old_acts = ovsl_dereference(flow->sf_acts);
1069                 rcu_assign_pointer(flow->sf_acts, acts);
1070
1071                 if (unlikely(reply)) {
1072                         error = ovs_flow_cmd_fill_info(flow,
1073                                                        ovs_header->dp_ifindex,
1074                                                        reply, info->snd_portid,
1075                                                        info->snd_seq, 0,
1076                                                        OVS_FLOW_CMD_NEW,
1077                                                        ufid_flags);
1078                         BUG_ON(error < 0);
1079                 }
1080                 ovs_unlock();
1081
1082                 ovs_nla_free_flow_actions_rcu(old_acts);
1083                 ovs_flow_free(new_flow, false);
1084         }
1085
1086         if (reply)
1087                 ovs_notify(&dp_flow_genl_family, reply, info);
1088         return 0;
1089
1090 err_unlock_ovs:
1091         ovs_unlock();
1092         kfree_skb(reply);
1093 err_kfree_acts:
1094         ovs_nla_free_flow_actions(acts);
1095 err_kfree_flow:
1096         ovs_flow_free(new_flow, false);
1097 error:
1098         return error;
1099 }
1100
1101 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1102 static struct sw_flow_actions *get_flow_actions(struct net *net,
1103                                                 const struct nlattr *a,
1104                                                 const struct sw_flow_key *key,
1105                                                 const struct sw_flow_mask *mask,
1106                                                 bool log)
1107 {
1108         struct sw_flow_actions *acts;
1109         struct sw_flow_key masked_key;
1110         int error;
1111
1112         ovs_flow_mask_key(&masked_key, key, true, mask);
1113         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1114         if (error) {
1115                 OVS_NLERR(log,
1116                           "Actions may not be safe on all matching packets");
1117                 return ERR_PTR(error);
1118         }
1119
1120         return acts;
1121 }
1122
1123 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1124 {
1125         struct net *net = sock_net(skb->sk);
1126         struct nlattr **a = info->attrs;
1127         struct ovs_header *ovs_header = info->userhdr;
1128         struct sw_flow_key key;
1129         struct sw_flow *flow;
1130         struct sw_flow_mask mask;
1131         struct sk_buff *reply = NULL;
1132         struct datapath *dp;
1133         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1134         struct sw_flow_match match;
1135         struct sw_flow_id sfid;
1136         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1137         int error = 0;
1138         bool log = !a[OVS_FLOW_ATTR_PROBE];
1139         bool ufid_present;
1140
1141         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1142         if (a[OVS_FLOW_ATTR_KEY]) {
1143                 ovs_match_init(&match, &key, true, &mask);
1144                 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1145                                           a[OVS_FLOW_ATTR_MASK], log);
1146         } else if (!ufid_present) {
1147                 OVS_NLERR(log,
1148                           "Flow set message rejected, Key attribute missing.");
1149                 error = -EINVAL;
1150         }
1151         if (error)
1152                 goto error;
1153
1154         /* Validate actions. */
1155         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1156                 if (!a[OVS_FLOW_ATTR_KEY]) {
1157                         OVS_NLERR(log,
1158                                   "Flow key attribute not present in set flow.");
1159                         error = -EINVAL;
1160                         goto error;
1161                 }
1162
1163                 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1164                                         &mask, log);
1165                 if (IS_ERR(acts)) {
1166                         error = PTR_ERR(acts);
1167                         goto error;
1168                 }
1169
1170                 /* Can allocate before locking if have acts. */
1171                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1172                                                 ufid_flags);
1173                 if (IS_ERR(reply)) {
1174                         error = PTR_ERR(reply);
1175                         goto err_kfree_acts;
1176                 }
1177         }
1178
1179         ovs_lock();
1180         dp = get_dp(net, ovs_header->dp_ifindex);
1181         if (unlikely(!dp)) {
1182                 error = -ENODEV;
1183                 goto err_unlock_ovs;
1184         }
1185         /* Check that the flow exists. */
1186         if (ufid_present)
1187                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1188         else
1189                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1190         if (unlikely(!flow)) {
1191                 error = -ENOENT;
1192                 goto err_unlock_ovs;
1193         }
1194
1195         /* Update actions, if present. */
1196         if (likely(acts)) {
1197                 old_acts = ovsl_dereference(flow->sf_acts);
1198                 rcu_assign_pointer(flow->sf_acts, acts);
1199
1200                 if (unlikely(reply)) {
1201                         error = ovs_flow_cmd_fill_info(flow,
1202                                                        ovs_header->dp_ifindex,
1203                                                        reply, info->snd_portid,
1204                                                        info->snd_seq, 0,
1205                                                        OVS_FLOW_CMD_NEW,
1206                                                        ufid_flags);
1207                         BUG_ON(error < 0);
1208                 }
1209         } else {
1210                 /* Could not alloc without acts before locking. */
1211                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1212                                                 info, OVS_FLOW_CMD_NEW, false,
1213                                                 ufid_flags);
1214
1215                 if (IS_ERR(reply)) {
1216                         error = PTR_ERR(reply);
1217                         goto err_unlock_ovs;
1218                 }
1219         }
1220
1221         /* Clear stats. */
1222         if (a[OVS_FLOW_ATTR_CLEAR])
1223                 ovs_flow_stats_clear(flow);
1224         ovs_unlock();
1225
1226         if (reply)
1227                 ovs_notify(&dp_flow_genl_family, reply, info);
1228         if (old_acts)
1229                 ovs_nla_free_flow_actions_rcu(old_acts);
1230
1231         return 0;
1232
1233 err_unlock_ovs:
1234         ovs_unlock();
1235         kfree_skb(reply);
1236 err_kfree_acts:
1237         ovs_nla_free_flow_actions(acts);
1238 error:
1239         return error;
1240 }
1241
1242 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1243 {
1244         struct nlattr **a = info->attrs;
1245         struct ovs_header *ovs_header = info->userhdr;
1246         struct net *net = sock_net(skb->sk);
1247         struct sw_flow_key key;
1248         struct sk_buff *reply;
1249         struct sw_flow *flow;
1250         struct datapath *dp;
1251         struct sw_flow_match match;
1252         struct sw_flow_id ufid;
1253         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1254         int err = 0;
1255         bool log = !a[OVS_FLOW_ATTR_PROBE];
1256         bool ufid_present;
1257
1258         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1259         if (a[OVS_FLOW_ATTR_KEY]) {
1260                 ovs_match_init(&match, &key, true, NULL);
1261                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1262                                         log);
1263         } else if (!ufid_present) {
1264                 OVS_NLERR(log,
1265                           "Flow get message rejected, Key attribute missing.");
1266                 err = -EINVAL;
1267         }
1268         if (err)
1269                 return err;
1270
1271         ovs_lock();
1272         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1273         if (!dp) {
1274                 err = -ENODEV;
1275                 goto unlock;
1276         }
1277
1278         if (ufid_present)
1279                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1280         else
1281                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1282         if (!flow) {
1283                 err = -ENOENT;
1284                 goto unlock;
1285         }
1286
1287         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1288                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1289         if (IS_ERR(reply)) {
1290                 err = PTR_ERR(reply);
1291                 goto unlock;
1292         }
1293
1294         ovs_unlock();
1295         return genlmsg_reply(reply, info);
1296 unlock:
1297         ovs_unlock();
1298         return err;
1299 }
1300
1301 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1302 {
1303         struct nlattr **a = info->attrs;
1304         struct ovs_header *ovs_header = info->userhdr;
1305         struct net *net = sock_net(skb->sk);
1306         struct sw_flow_key key;
1307         struct sk_buff *reply;
1308         struct sw_flow *flow = NULL;
1309         struct datapath *dp;
1310         struct sw_flow_match match;
1311         struct sw_flow_id ufid;
1312         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1313         int err;
1314         bool log = !a[OVS_FLOW_ATTR_PROBE];
1315         bool ufid_present;
1316
1317         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1318         if (a[OVS_FLOW_ATTR_KEY]) {
1319                 ovs_match_init(&match, &key, true, NULL);
1320                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1321                                         NULL, log);
1322                 if (unlikely(err))
1323                         return err;
1324         }
1325
1326         ovs_lock();
1327         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1328         if (unlikely(!dp)) {
1329                 err = -ENODEV;
1330                 goto unlock;
1331         }
1332
1333         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1334                 err = ovs_flow_tbl_flush(&dp->table);
1335                 goto unlock;
1336         }
1337
1338         if (ufid_present)
1339                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1340         else
1341                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1342         if (unlikely(!flow)) {
1343                 err = -ENOENT;
1344                 goto unlock;
1345         }
1346
1347         ovs_flow_tbl_remove(&dp->table, flow);
1348         ovs_unlock();
1349
1350         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1351                                         &flow->id, info, false, ufid_flags);
1352         if (likely(reply)) {
1353                 if (likely(!IS_ERR(reply))) {
1354                         rcu_read_lock();        /*To keep RCU checker happy. */
1355                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1356                                                      reply, info->snd_portid,
1357                                                      info->snd_seq, 0,
1358                                                      OVS_FLOW_CMD_DEL,
1359                                                      ufid_flags);
1360                         rcu_read_unlock();
1361                         if (WARN_ON_ONCE(err < 0)) {
1362                                 kfree_skb(reply);
1363                                 goto out_free;
1364                         }
1365
1366                         ovs_notify(&dp_flow_genl_family, reply, info);
1367                 } else {
1368                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1369                 }
1370         }
1371
1372 out_free:
1373         ovs_flow_free(flow, true);
1374         return 0;
1375 unlock:
1376         ovs_unlock();
1377         return err;
1378 }
1379
1380 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1381 {
1382         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1383         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1384         struct table_instance *ti;
1385         struct datapath *dp;
1386         u32 ufid_flags;
1387         int err;
1388
1389         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1390                             OVS_FLOW_ATTR_MAX, flow_policy);
1391         if (err)
1392                 return err;
1393         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1394
1395         rcu_read_lock();
1396         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1397         if (!dp) {
1398                 rcu_read_unlock();
1399                 return -ENODEV;
1400         }
1401
1402         ti = rcu_dereference(dp->table.ti);
1403         for (;;) {
1404                 struct sw_flow *flow;
1405                 u32 bucket, obj;
1406
1407                 bucket = cb->args[0];
1408                 obj = cb->args[1];
1409                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1410                 if (!flow)
1411                         break;
1412
1413                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1414                                            NETLINK_CB(cb->skb).portid,
1415                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1416                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1417                         break;
1418
1419                 cb->args[0] = bucket;
1420                 cb->args[1] = obj;
1421         }
1422         rcu_read_unlock();
1423         return skb->len;
1424 }
1425
1426 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1427         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1428         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1429         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1430         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1431         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1432         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1433         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1434 };
1435
1436 static const struct genl_ops dp_flow_genl_ops[] = {
1437         { .cmd = OVS_FLOW_CMD_NEW,
1438           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1439           .policy = flow_policy,
1440           .doit = ovs_flow_cmd_new
1441         },
1442         { .cmd = OVS_FLOW_CMD_DEL,
1443           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1444           .policy = flow_policy,
1445           .doit = ovs_flow_cmd_del
1446         },
1447         { .cmd = OVS_FLOW_CMD_GET,
1448           .flags = 0,               /* OK for unprivileged users. */
1449           .policy = flow_policy,
1450           .doit = ovs_flow_cmd_get,
1451           .dumpit = ovs_flow_cmd_dump
1452         },
1453         { .cmd = OVS_FLOW_CMD_SET,
1454           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1455           .policy = flow_policy,
1456           .doit = ovs_flow_cmd_set,
1457         },
1458 };
1459
1460 static struct genl_family dp_flow_genl_family = {
1461         .id = GENL_ID_GENERATE,
1462         .hdrsize = sizeof(struct ovs_header),
1463         .name = OVS_FLOW_FAMILY,
1464         .version = OVS_FLOW_VERSION,
1465         .maxattr = OVS_FLOW_ATTR_MAX,
1466         .netnsok = true,
1467         .parallel_ops = true,
1468         .ops = dp_flow_genl_ops,
1469         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1470         .mcgrps = &ovs_dp_flow_multicast_group,
1471         .n_mcgrps = 1,
1472 };
1473
1474 static size_t ovs_dp_cmd_msg_size(void)
1475 {
1476         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1477
1478         msgsize += nla_total_size(IFNAMSIZ);
1479         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1480         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1481         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1482
1483         return msgsize;
1484 }
1485
1486 /* Called with ovs_mutex. */
1487 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1488                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1489 {
1490         struct ovs_header *ovs_header;
1491         struct ovs_dp_stats dp_stats;
1492         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1493         int err;
1494
1495         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1496                                    flags, cmd);
1497         if (!ovs_header)
1498                 goto error;
1499
1500         ovs_header->dp_ifindex = get_dpifindex(dp);
1501
1502         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1503         if (err)
1504                 goto nla_put_failure;
1505
1506         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1507         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1508                           &dp_stats, OVS_DP_ATTR_PAD))
1509                 goto nla_put_failure;
1510
1511         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1512                           sizeof(struct ovs_dp_megaflow_stats),
1513                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1514                 goto nla_put_failure;
1515
1516         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1517                 goto nla_put_failure;
1518
1519         genlmsg_end(skb, ovs_header);
1520         return 0;
1521
1522 nla_put_failure:
1523         genlmsg_cancel(skb, ovs_header);
1524 error:
1525         return -EMSGSIZE;
1526 }
1527
1528 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1529 {
1530         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1531 }
1532
1533 /* Called with rcu_read_lock or ovs_mutex. */
1534 static struct datapath *lookup_datapath(struct net *net,
1535                                         const struct ovs_header *ovs_header,
1536                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1537 {
1538         struct datapath *dp;
1539
1540         if (!a[OVS_DP_ATTR_NAME])
1541                 dp = get_dp(net, ovs_header->dp_ifindex);
1542         else {
1543                 struct vport *vport;
1544
1545                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1546                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1547         }
1548         return dp ? dp : ERR_PTR(-ENODEV);
1549 }
1550
1551 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1552 {
1553         struct datapath *dp;
1554
1555         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1556         if (IS_ERR(dp))
1557                 return;
1558
1559         pr_warn("%s: Dropping previously announced user features\n",
1560                 ovs_dp_name(dp));
1561         dp->user_features = 0;
1562 }
1563
1564 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1565 {
1566         if (a[OVS_DP_ATTR_USER_FEATURES])
1567                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1568 }
1569
1570 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1571 {
1572         struct nlattr **a = info->attrs;
1573         struct vport_parms parms;
1574         struct sk_buff *reply;
1575         struct datapath *dp;
1576         struct vport *vport;
1577         struct ovs_net *ovs_net;
1578         int err, i;
1579
1580         err = -EINVAL;
1581         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1582                 goto err;
1583
1584         reply = ovs_dp_cmd_alloc_info();
1585         if (!reply)
1586                 return -ENOMEM;
1587
1588         err = -ENOMEM;
1589         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1590         if (dp == NULL)
1591                 goto err_free_reply;
1592
1593         ovs_dp_set_net(dp, sock_net(skb->sk));
1594
1595         /* Allocate table. */
1596         err = ovs_flow_tbl_init(&dp->table);
1597         if (err)
1598                 goto err_free_dp;
1599
1600         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1601         if (!dp->stats_percpu) {
1602                 err = -ENOMEM;
1603                 goto err_destroy_table;
1604         }
1605
1606         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1607                             GFP_KERNEL);
1608         if (!dp->ports) {
1609                 err = -ENOMEM;
1610                 goto err_destroy_percpu;
1611         }
1612
1613         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1614                 INIT_HLIST_HEAD(&dp->ports[i]);
1615
1616         /* Set up our datapath device. */
1617         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1618         parms.type = OVS_VPORT_TYPE_INTERNAL;
1619         parms.options = NULL;
1620         parms.dp = dp;
1621         parms.port_no = OVSP_LOCAL;
1622         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1623
1624         ovs_dp_change(dp, a);
1625
1626         /* So far only local changes have been made, now need the lock. */
1627         ovs_lock();
1628
1629         vport = new_vport(&parms);
1630         if (IS_ERR(vport)) {
1631                 err = PTR_ERR(vport);
1632                 if (err == -EBUSY)
1633                         err = -EEXIST;
1634
1635                 if (err == -EEXIST) {
1636                         /* An outdated user space instance that does not understand
1637                          * the concept of user_features has attempted to create a new
1638                          * datapath and is likely to reuse it. Drop all user features.
1639                          */
1640                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1641                                 ovs_dp_reset_user_features(skb, info);
1642                 }
1643
1644                 goto err_destroy_ports_array;
1645         }
1646
1647         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1648                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1649         BUG_ON(err < 0);
1650
1651         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1652         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1653
1654         ovs_unlock();
1655
1656         ovs_notify(&dp_datapath_genl_family, reply, info);
1657         return 0;
1658
1659 err_destroy_ports_array:
1660         ovs_unlock();
1661         kfree(dp->ports);
1662 err_destroy_percpu:
1663         free_percpu(dp->stats_percpu);
1664 err_destroy_table:
1665         ovs_flow_tbl_destroy(&dp->table);
1666 err_free_dp:
1667         kfree(dp);
1668 err_free_reply:
1669         kfree_skb(reply);
1670 err:
1671         return err;
1672 }
1673
1674 /* Called with ovs_mutex. */
1675 static void __dp_destroy(struct datapath *dp)
1676 {
1677         int i;
1678
1679         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1680                 struct vport *vport;
1681                 struct hlist_node *n;
1682
1683                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1684                         if (vport->port_no != OVSP_LOCAL)
1685                                 ovs_dp_detach_port(vport);
1686         }
1687
1688         list_del_rcu(&dp->list_node);
1689
1690         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1691          * all ports in datapath are destroyed first before freeing datapath.
1692          */
1693         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1694
1695         /* RCU destroy the flow table */
1696         call_rcu(&dp->rcu, destroy_dp_rcu);
1697 }
1698
1699 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1700 {
1701         struct sk_buff *reply;
1702         struct datapath *dp;
1703         int err;
1704
1705         reply = ovs_dp_cmd_alloc_info();
1706         if (!reply)
1707                 return -ENOMEM;
1708
1709         ovs_lock();
1710         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1711         err = PTR_ERR(dp);
1712         if (IS_ERR(dp))
1713                 goto err_unlock_free;
1714
1715         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1716                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1717         BUG_ON(err < 0);
1718
1719         __dp_destroy(dp);
1720         ovs_unlock();
1721
1722         ovs_notify(&dp_datapath_genl_family, reply, info);
1723
1724         return 0;
1725
1726 err_unlock_free:
1727         ovs_unlock();
1728         kfree_skb(reply);
1729         return err;
1730 }
1731
1732 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1733 {
1734         struct sk_buff *reply;
1735         struct datapath *dp;
1736         int err;
1737
1738         reply = ovs_dp_cmd_alloc_info();
1739         if (!reply)
1740                 return -ENOMEM;
1741
1742         ovs_lock();
1743         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1744         err = PTR_ERR(dp);
1745         if (IS_ERR(dp))
1746                 goto err_unlock_free;
1747
1748         ovs_dp_change(dp, info->attrs);
1749
1750         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1751                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1752         BUG_ON(err < 0);
1753
1754         ovs_unlock();
1755         ovs_notify(&dp_datapath_genl_family, reply, info);
1756
1757         return 0;
1758
1759 err_unlock_free:
1760         ovs_unlock();
1761         kfree_skb(reply);
1762         return err;
1763 }
1764
1765 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1766 {
1767         struct sk_buff *reply;
1768         struct datapath *dp;
1769         int err;
1770
1771         reply = ovs_dp_cmd_alloc_info();
1772         if (!reply)
1773                 return -ENOMEM;
1774
1775         ovs_lock();
1776         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1777         if (IS_ERR(dp)) {
1778                 err = PTR_ERR(dp);
1779                 goto err_unlock_free;
1780         }
1781         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1782                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1783         BUG_ON(err < 0);
1784         ovs_unlock();
1785
1786         return genlmsg_reply(reply, info);
1787
1788 err_unlock_free:
1789         ovs_unlock();
1790         kfree_skb(reply);
1791         return err;
1792 }
1793
1794 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1795 {
1796         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1797         struct datapath *dp;
1798         int skip = cb->args[0];
1799         int i = 0;
1800
1801         ovs_lock();
1802         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1803                 if (i >= skip &&
1804                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1805                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1806                                          OVS_DP_CMD_NEW) < 0)
1807                         break;
1808                 i++;
1809         }
1810         ovs_unlock();
1811
1812         cb->args[0] = i;
1813
1814         return skb->len;
1815 }
1816
1817 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1818         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1819         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1820         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1821 };
1822
1823 static const struct genl_ops dp_datapath_genl_ops[] = {
1824         { .cmd = OVS_DP_CMD_NEW,
1825           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1826           .policy = datapath_policy,
1827           .doit = ovs_dp_cmd_new
1828         },
1829         { .cmd = OVS_DP_CMD_DEL,
1830           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1831           .policy = datapath_policy,
1832           .doit = ovs_dp_cmd_del
1833         },
1834         { .cmd = OVS_DP_CMD_GET,
1835           .flags = 0,               /* OK for unprivileged users. */
1836           .policy = datapath_policy,
1837           .doit = ovs_dp_cmd_get,
1838           .dumpit = ovs_dp_cmd_dump
1839         },
1840         { .cmd = OVS_DP_CMD_SET,
1841           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1842           .policy = datapath_policy,
1843           .doit = ovs_dp_cmd_set,
1844         },
1845 };
1846
1847 static struct genl_family dp_datapath_genl_family = {
1848         .id = GENL_ID_GENERATE,
1849         .hdrsize = sizeof(struct ovs_header),
1850         .name = OVS_DATAPATH_FAMILY,
1851         .version = OVS_DATAPATH_VERSION,
1852         .maxattr = OVS_DP_ATTR_MAX,
1853         .netnsok = true,
1854         .parallel_ops = true,
1855         .ops = dp_datapath_genl_ops,
1856         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1857         .mcgrps = &ovs_dp_datapath_multicast_group,
1858         .n_mcgrps = 1,
1859 };
1860
1861 /* Called with ovs_mutex or RCU read lock. */
1862 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1863                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1864 {
1865         struct ovs_header *ovs_header;
1866         struct ovs_vport_stats vport_stats;
1867         int err;
1868
1869         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1870                                  flags, cmd);
1871         if (!ovs_header)
1872                 return -EMSGSIZE;
1873
1874         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1875
1876         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1877             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1878             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1879                            ovs_vport_name(vport)))
1880                 goto nla_put_failure;
1881
1882         ovs_vport_get_stats(vport, &vport_stats);
1883         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1884                           sizeof(struct ovs_vport_stats), &vport_stats,
1885                           OVS_VPORT_ATTR_PAD))
1886                 goto nla_put_failure;
1887
1888         if (ovs_vport_get_upcall_portids(vport, skb))
1889                 goto nla_put_failure;
1890
1891         err = ovs_vport_get_options(vport, skb);
1892         if (err == -EMSGSIZE)
1893                 goto error;
1894
1895         genlmsg_end(skb, ovs_header);
1896         return 0;
1897
1898 nla_put_failure:
1899         err = -EMSGSIZE;
1900 error:
1901         genlmsg_cancel(skb, ovs_header);
1902         return err;
1903 }
1904
1905 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1906 {
1907         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1908 }
1909
1910 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1911 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1912                                          u32 seq, u8 cmd)
1913 {
1914         struct sk_buff *skb;
1915         int retval;
1916
1917         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1918         if (!skb)
1919                 return ERR_PTR(-ENOMEM);
1920
1921         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1922         BUG_ON(retval < 0);
1923
1924         return skb;
1925 }
1926
1927 /* Called with ovs_mutex or RCU read lock. */
1928 static struct vport *lookup_vport(struct net *net,
1929                                   const struct ovs_header *ovs_header,
1930                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1931 {
1932         struct datapath *dp;
1933         struct vport *vport;
1934
1935         if (a[OVS_VPORT_ATTR_NAME]) {
1936                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1937                 if (!vport)
1938                         return ERR_PTR(-ENODEV);
1939                 if (ovs_header->dp_ifindex &&
1940                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1941                         return ERR_PTR(-ENODEV);
1942                 return vport;
1943         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1944                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1945
1946                 if (port_no >= DP_MAX_PORTS)
1947                         return ERR_PTR(-EFBIG);
1948
1949                 dp = get_dp(net, ovs_header->dp_ifindex);
1950                 if (!dp)
1951                         return ERR_PTR(-ENODEV);
1952
1953                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1954                 if (!vport)
1955                         return ERR_PTR(-ENODEV);
1956                 return vport;
1957         } else
1958                 return ERR_PTR(-EINVAL);
1959 }
1960
1961 /* Called with ovs_mutex */
1962 static void update_headroom(struct datapath *dp)
1963 {
1964         unsigned dev_headroom, max_headroom = 0;
1965         struct net_device *dev;
1966         struct vport *vport;
1967         int i;
1968
1969         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1970                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1971                         dev = vport->dev;
1972                         dev_headroom = netdev_get_fwd_headroom(dev);
1973                         if (dev_headroom > max_headroom)
1974                                 max_headroom = dev_headroom;
1975                 }
1976         }
1977
1978         dp->max_headroom = max_headroom;
1979         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1980                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1981                         netdev_set_rx_headroom(vport->dev, max_headroom);
1982 }
1983
1984 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1985 {
1986         struct nlattr **a = info->attrs;
1987         struct ovs_header *ovs_header = info->userhdr;
1988         struct vport_parms parms;
1989         struct sk_buff *reply;
1990         struct vport *vport;
1991         struct datapath *dp;
1992         u32 port_no;
1993         int err;
1994
1995         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1996             !a[OVS_VPORT_ATTR_UPCALL_PID])
1997                 return -EINVAL;
1998
1999         port_no = a[OVS_VPORT_ATTR_PORT_NO]
2000                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2001         if (port_no >= DP_MAX_PORTS)
2002                 return -EFBIG;
2003
2004         reply = ovs_vport_cmd_alloc_info();
2005         if (!reply)
2006                 return -ENOMEM;
2007
2008         ovs_lock();
2009 restart:
2010         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2011         err = -ENODEV;
2012         if (!dp)
2013                 goto exit_unlock_free;
2014
2015         if (port_no) {
2016                 vport = ovs_vport_ovsl(dp, port_no);
2017                 err = -EBUSY;
2018                 if (vport)
2019                         goto exit_unlock_free;
2020         } else {
2021                 for (port_no = 1; ; port_no++) {
2022                         if (port_no >= DP_MAX_PORTS) {
2023                                 err = -EFBIG;
2024                                 goto exit_unlock_free;
2025                         }
2026                         vport = ovs_vport_ovsl(dp, port_no);
2027                         if (!vport)
2028                                 break;
2029                 }
2030         }
2031
2032         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2033         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2034         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2035         parms.dp = dp;
2036         parms.port_no = port_no;
2037         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2038
2039         vport = new_vport(&parms);
2040         err = PTR_ERR(vport);
2041         if (IS_ERR(vport)) {
2042                 if (err == -EAGAIN)
2043                         goto restart;
2044                 goto exit_unlock_free;
2045         }
2046
2047         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2048                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2049
2050         if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2051                 update_headroom(dp);
2052         else
2053                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2054
2055         BUG_ON(err < 0);
2056         ovs_unlock();
2057
2058         ovs_notify(&dp_vport_genl_family, reply, info);
2059         return 0;
2060
2061 exit_unlock_free:
2062         ovs_unlock();
2063         kfree_skb(reply);
2064         return err;
2065 }
2066
2067 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2068 {
2069         struct nlattr **a = info->attrs;
2070         struct sk_buff *reply;
2071         struct vport *vport;
2072         int err;
2073
2074         reply = ovs_vport_cmd_alloc_info();
2075         if (!reply)
2076                 return -ENOMEM;
2077
2078         ovs_lock();
2079         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2080         err = PTR_ERR(vport);
2081         if (IS_ERR(vport))
2082                 goto exit_unlock_free;
2083
2084         if (a[OVS_VPORT_ATTR_TYPE] &&
2085             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2086                 err = -EINVAL;
2087                 goto exit_unlock_free;
2088         }
2089
2090         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2091                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2092                 if (err)
2093                         goto exit_unlock_free;
2094         }
2095
2096
2097         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2098                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2099
2100                 err = ovs_vport_set_upcall_portids(vport, ids);
2101                 if (err)
2102                         goto exit_unlock_free;
2103         }
2104
2105         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2106                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2107         BUG_ON(err < 0);
2108
2109         ovs_unlock();
2110         ovs_notify(&dp_vport_genl_family, reply, info);
2111         return 0;
2112
2113 exit_unlock_free:
2114         ovs_unlock();
2115         kfree_skb(reply);
2116         return err;
2117 }
2118
2119 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2120 {
2121         bool must_update_headroom = false;
2122         struct nlattr **a = info->attrs;
2123         struct sk_buff *reply;
2124         struct datapath *dp;
2125         struct vport *vport;
2126         int err;
2127
2128         reply = ovs_vport_cmd_alloc_info();
2129         if (!reply)
2130                 return -ENOMEM;
2131
2132         ovs_lock();
2133         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2134         err = PTR_ERR(vport);
2135         if (IS_ERR(vport))
2136                 goto exit_unlock_free;
2137
2138         if (vport->port_no == OVSP_LOCAL) {
2139                 err = -EINVAL;
2140                 goto exit_unlock_free;
2141         }
2142
2143         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2144                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2145         BUG_ON(err < 0);
2146
2147         /* the vport deletion may trigger dp headroom update */
2148         dp = vport->dp;
2149         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2150                 must_update_headroom = true;
2151         netdev_reset_rx_headroom(vport->dev);
2152         ovs_dp_detach_port(vport);
2153
2154         if (must_update_headroom)
2155                 update_headroom(dp);
2156         ovs_unlock();
2157
2158         ovs_notify(&dp_vport_genl_family, reply, info);
2159         return 0;
2160
2161 exit_unlock_free:
2162         ovs_unlock();
2163         kfree_skb(reply);
2164         return err;
2165 }
2166
2167 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2168 {
2169         struct nlattr **a = info->attrs;
2170         struct ovs_header *ovs_header = info->userhdr;
2171         struct sk_buff *reply;
2172         struct vport *vport;
2173         int err;
2174
2175         reply = ovs_vport_cmd_alloc_info();
2176         if (!reply)
2177                 return -ENOMEM;
2178
2179         rcu_read_lock();
2180         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2181         err = PTR_ERR(vport);
2182         if (IS_ERR(vport))
2183                 goto exit_unlock_free;
2184         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2185                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2186         BUG_ON(err < 0);
2187         rcu_read_unlock();
2188
2189         return genlmsg_reply(reply, info);
2190
2191 exit_unlock_free:
2192         rcu_read_unlock();
2193         kfree_skb(reply);
2194         return err;
2195 }
2196
2197 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2198 {
2199         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2200         struct datapath *dp;
2201         int bucket = cb->args[0], skip = cb->args[1];
2202         int i, j = 0;
2203
2204         rcu_read_lock();
2205         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2206         if (!dp) {
2207                 rcu_read_unlock();
2208                 return -ENODEV;
2209         }
2210         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2211                 struct vport *vport;
2212
2213                 j = 0;
2214                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2215                         if (j >= skip &&
2216                             ovs_vport_cmd_fill_info(vport, skb,
2217                                                     NETLINK_CB(cb->skb).portid,
2218                                                     cb->nlh->nlmsg_seq,
2219                                                     NLM_F_MULTI,
2220                                                     OVS_VPORT_CMD_NEW) < 0)
2221                                 goto out;
2222
2223                         j++;
2224                 }
2225                 skip = 0;
2226         }
2227 out:
2228         rcu_read_unlock();
2229
2230         cb->args[0] = i;
2231         cb->args[1] = j;
2232
2233         return skb->len;
2234 }
2235
2236 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2237         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2238         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2239         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2240         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2241         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2242         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2243 };
2244
2245 static const struct genl_ops dp_vport_genl_ops[] = {
2246         { .cmd = OVS_VPORT_CMD_NEW,
2247           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2248           .policy = vport_policy,
2249           .doit = ovs_vport_cmd_new
2250         },
2251         { .cmd = OVS_VPORT_CMD_DEL,
2252           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2253           .policy = vport_policy,
2254           .doit = ovs_vport_cmd_del
2255         },
2256         { .cmd = OVS_VPORT_CMD_GET,
2257           .flags = 0,               /* OK for unprivileged users. */
2258           .policy = vport_policy,
2259           .doit = ovs_vport_cmd_get,
2260           .dumpit = ovs_vport_cmd_dump
2261         },
2262         { .cmd = OVS_VPORT_CMD_SET,
2263           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2264           .policy = vport_policy,
2265           .doit = ovs_vport_cmd_set,
2266         },
2267 };
2268
2269 struct genl_family dp_vport_genl_family = {
2270         .id = GENL_ID_GENERATE,
2271         .hdrsize = sizeof(struct ovs_header),
2272         .name = OVS_VPORT_FAMILY,
2273         .version = OVS_VPORT_VERSION,
2274         .maxattr = OVS_VPORT_ATTR_MAX,
2275         .netnsok = true,
2276         .parallel_ops = true,
2277         .ops = dp_vport_genl_ops,
2278         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2279         .mcgrps = &ovs_dp_vport_multicast_group,
2280         .n_mcgrps = 1,
2281 };
2282
2283 static struct genl_family * const dp_genl_families[] = {
2284         &dp_datapath_genl_family,
2285         &dp_vport_genl_family,
2286         &dp_flow_genl_family,
2287         &dp_packet_genl_family,
2288 };
2289
2290 static void dp_unregister_genl(int n_families)
2291 {
2292         int i;
2293
2294         for (i = 0; i < n_families; i++)
2295                 genl_unregister_family(dp_genl_families[i]);
2296 }
2297
2298 static int dp_register_genl(void)
2299 {
2300         int err;
2301         int i;
2302
2303         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2304
2305                 err = genl_register_family(dp_genl_families[i]);
2306                 if (err)
2307                         goto error;
2308         }
2309
2310         return 0;
2311
2312 error:
2313         dp_unregister_genl(i);
2314         return err;
2315 }
2316
2317 static int __net_init ovs_init_net(struct net *net)
2318 {
2319         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2320
2321         INIT_LIST_HEAD(&ovs_net->dps);
2322         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2323         ovs_ct_init(net);
2324         return 0;
2325 }
2326
2327 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2328                                             struct list_head *head)
2329 {
2330         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2331         struct datapath *dp;
2332
2333         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2334                 int i;
2335
2336                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2337                         struct vport *vport;
2338
2339                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2340                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2341                                         continue;
2342
2343                                 if (dev_net(vport->dev) == dnet)
2344                                         list_add(&vport->detach_list, head);
2345                         }
2346                 }
2347         }
2348 }
2349
2350 static void __net_exit ovs_exit_net(struct net *dnet)
2351 {
2352         struct datapath *dp, *dp_next;
2353         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2354         struct vport *vport, *vport_next;
2355         struct net *net;
2356         LIST_HEAD(head);
2357
2358         ovs_ct_exit(dnet);
2359         ovs_lock();
2360         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2361                 __dp_destroy(dp);
2362
2363         rtnl_lock();
2364         for_each_net(net)
2365                 list_vports_from_net(net, dnet, &head);
2366         rtnl_unlock();
2367
2368         /* Detach all vports from given namespace. */
2369         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2370                 list_del(&vport->detach_list);
2371                 ovs_dp_detach_port(vport);
2372         }
2373
2374         ovs_unlock();
2375
2376         cancel_work_sync(&ovs_net->dp_notify_work);
2377 }
2378
2379 static struct pernet_operations ovs_net_ops = {
2380         .init = ovs_init_net,
2381         .exit = ovs_exit_net,
2382         .id   = &ovs_net_id,
2383         .size = sizeof(struct ovs_net),
2384 };
2385
2386 static int __init dp_init(void)
2387 {
2388         int err;
2389
2390         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2391
2392         pr_info("Open vSwitch switching datapath\n");
2393
2394         err = action_fifos_init();
2395         if (err)
2396                 goto error;
2397
2398         err = ovs_internal_dev_rtnl_link_register();
2399         if (err)
2400                 goto error_action_fifos_exit;
2401
2402         err = ovs_flow_init();
2403         if (err)
2404                 goto error_unreg_rtnl_link;
2405
2406         err = ovs_vport_init();
2407         if (err)
2408                 goto error_flow_exit;
2409
2410         err = register_pernet_device(&ovs_net_ops);
2411         if (err)
2412                 goto error_vport_exit;
2413
2414         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2415         if (err)
2416                 goto error_netns_exit;
2417
2418         err = ovs_netdev_init();
2419         if (err)
2420                 goto error_unreg_notifier;
2421
2422         err = dp_register_genl();
2423         if (err < 0)
2424                 goto error_unreg_netdev;
2425
2426         return 0;
2427
2428 error_unreg_netdev:
2429         ovs_netdev_exit();
2430 error_unreg_notifier:
2431         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2432 error_netns_exit:
2433         unregister_pernet_device(&ovs_net_ops);
2434 error_vport_exit:
2435         ovs_vport_exit();
2436 error_flow_exit:
2437         ovs_flow_exit();
2438 error_unreg_rtnl_link:
2439         ovs_internal_dev_rtnl_link_unregister();
2440 error_action_fifos_exit:
2441         action_fifos_exit();
2442 error:
2443         return err;
2444 }
2445
2446 static void dp_cleanup(void)
2447 {
2448         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2449         ovs_netdev_exit();
2450         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2451         unregister_pernet_device(&ovs_net_ops);
2452         rcu_barrier();
2453         ovs_vport_exit();
2454         ovs_flow_exit();
2455         ovs_internal_dev_rtnl_link_unregister();
2456         action_fifos_exit();
2457 }
2458
2459 module_init(dp_init);
2460 module_exit(dp_cleanup);
2461
2462 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2463 MODULE_LICENSE("GPL");
2464 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2465 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2466 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2467 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);