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 "meter.h"
  59 #include "vport-internal_dev.h"
  60 #include "vport-netdev.h"
  61
  62 unsigned int ovs_net_id __read_mostly;
  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 #endif
 135
 136 static struct vport *new_vport(const struct vport_parms *);
 137 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 138                              const struct sw_flow_key *,
 139                              const struct dp_upcall_info *,
 140                              uint32_t cutlen);
 141 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 142                                   const struct sw_flow_key *,
 143                                   const struct dp_upcall_info *,
 144                                   uint32_t cutlen);
 145
 146 /* Must be called with rcu_read_lock or ovs_mutex. */
 147 const char *ovs_dp_name(const struct datapath *dp)
 148 {
 149         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 150         return ovs_vport_name(vport);
 151 }
 152
 153 static int get_dpifindex(const struct datapath *dp)
 154 {
 155         struct vport *local;
 156         int ifindex;
 157
 158         rcu_read_lock();
 159
 160         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 161         if (local)
 162                 ifindex = local->dev->ifindex;
 163         else
 164                 ifindex = 0;
 165
 166         rcu_read_unlock();
 167
 168         return ifindex;
 169 }
 170
 171 static void destroy_dp_rcu(struct rcu_head *rcu)
 172 {
 173         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 174
 175         ovs_flow_tbl_destroy(&dp->table);
 176         free_percpu(dp->stats_percpu);
 177         kfree(dp->ports);
 178         ovs_meters_exit(dp);
 179         kfree(dp);
 180 }
 181
 182 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 183                                             u16 port_no)
 184 {
 185         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 186 }
 187
 188 /* Called with ovs_mutex or RCU read lock. */
 189 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 190 {
 191         struct vport *vport;
 192         struct hlist_head *head;
 193
 194         head = vport_hash_bucket(dp, port_no);
 195         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 196                 if (vport->port_no == port_no)
 197                         return vport;
 198         }
 199         return NULL;
 200 }
 201
 202 /* Called with ovs_mutex. */
 203 static struct vport *new_vport(const struct vport_parms *parms)
 204 {
 205         struct vport *vport;
 206
 207         vport = ovs_vport_add(parms);
 208         if (!IS_ERR(vport)) {
 209                 struct datapath *dp = parms->dp;
 210                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 211
 212                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 213         }
 214         return vport;
 215 }
 216
 217 void ovs_dp_detach_port(struct vport *p)
 218 {
 219         ASSERT_OVSL();
 220
 221         /* First drop references to device. */
 222         hlist_del_rcu(&p->dp_hash_node);
 223
 224         /* Then destroy it. */
 225         ovs_vport_del(p);
 226 }
 227
 228 /* Must be called with rcu_read_lock. */
 229 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 230 {
 231         const struct vport *p = OVS_CB(skb)->input_vport;
 232         struct datapath *dp = p->dp;
 233         struct sw_flow *flow;
 234         struct sw_flow_actions *sf_acts;
 235         struct dp_stats_percpu *stats;
 236         u64 *stats_counter;
 237         u32 n_mask_hit;
 238
 239         stats = this_cpu_ptr(dp->stats_percpu);
 240
 241         /* Look up flow. */
 242         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
 243         if (unlikely(!flow)) {
 244                 struct dp_upcall_info upcall;
 245                 int error;
 246
 247                 memset(&upcall, 0, sizeof(upcall));
 248                 upcall.cmd = OVS_PACKET_CMD_MISS;
 249                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 250                 upcall.mru = OVS_CB(skb)->mru;
 251                 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
 252                 if (unlikely(error))
 253                         kfree_skb(skb);
 254                 else
 255                         consume_skb(skb);
 256                 stats_counter = &stats->n_missed;
 257                 goto out;
 258         }
 259
 260         ovs_flow_stats_update(flow, key->tp.flags, skb);
 261         sf_acts = rcu_dereference(flow->sf_acts);
 262         ovs_execute_actions(dp, skb, sf_acts, key);
 263
 264         stats_counter = &stats->n_hit;
 265
 266 out:
 267         /* Update datapath statistics. */
 268         u64_stats_update_begin(&stats->syncp);
 269         (*stats_counter)++;
 270         stats->n_mask_hit += n_mask_hit;
 271         u64_stats_update_end(&stats->syncp);
 272 }
 273
 274 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 275                   const struct sw_flow_key *key,
 276                   const struct dp_upcall_info *upcall_info,
 277                   uint32_t cutlen)
 278 {
 279         struct dp_stats_percpu *stats;
 280         int err;
 281
 282         if (upcall_info->portid == 0) {
 283                 err = -ENOTCONN;
 284                 goto err;
 285         }
 286
 287         if (!skb_is_gso(skb))
 288                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 289         else
 290                 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
 291         if (err)
 292                 goto err;
 293
 294         return 0;
 295
 296 err:
 297         stats = this_cpu_ptr(dp->stats_percpu);
 298
 299         u64_stats_update_begin(&stats->syncp);
 300         stats->n_lost++;
 301         u64_stats_update_end(&stats->syncp);
 302
 303         return err;
 304 }
 305
 306 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 307                              const struct sw_flow_key *key,
 308                              const struct dp_upcall_info *upcall_info,
 309                                  uint32_t cutlen)
 310 {
 311         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 312         struct sw_flow_key later_key;
 313         struct sk_buff *segs, *nskb;
 314         int err;
 315
 316         BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
 317         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 318         if (IS_ERR(segs))
 319                 return PTR_ERR(segs);
 320         if (segs == NULL)
 321                 return -EINVAL;
 322
 323         if (gso_type & SKB_GSO_UDP) {
 324                 /* The initial flow key extracted by ovs_flow_key_extract()
 325                  * in this case is for a first fragment, so we need to
 326                  * properly mark later fragments.
 327                  */
 328                 later_key = *key;
 329                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 330         }
 331
 332         /* Queue all of the segments. */
 333         skb = segs;
 334         do {
 335                 if (gso_type & SKB_GSO_UDP && skb != segs)
 336                         key = &later_key;
 337
 338                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 339                 if (err)
 340                         break;
 341
 342         } while ((skb = skb->next));
 343
 344         /* Free all of the segments. */
 345         skb = segs;
 346         do {
 347                 nskb = skb->next;
 348                 if (err)
 349                         kfree_skb(skb);
 350                 else
 351                         consume_skb(skb);
 352         } while ((skb = nskb));
 353         return err;
 354 }
 355
 356 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 357                               unsigned int hdrlen, int actions_attrlen)
 358 {
 359         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 360                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 361                 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 362                 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
 363
 364         /* OVS_PACKET_ATTR_USERDATA */
 365         if (upcall_info->userdata)
 366                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 367
 368         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 369         if (upcall_info->egress_tun_info)
 370                 size += nla_total_size(ovs_tun_key_attr_size());
 371
 372         /* OVS_PACKET_ATTR_ACTIONS */
 373         if (upcall_info->actions_len)
 374                 size += nla_total_size(actions_attrlen);
 375
 376         /* OVS_PACKET_ATTR_MRU */
 377         if (upcall_info->mru)
 378                 size += nla_total_size(sizeof(upcall_info->mru));
 379
 380         return size;
 381 }
 382
 383 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 384 {
 385         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 386                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 387
 388                 if (plen > 0)
 389                         skb_put_zero(skb, plen);
 390         }
 391 }
 392
 393 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 394                                   const struct sw_flow_key *key,
 395                                   const struct dp_upcall_info *upcall_info,
 396                                   uint32_t cutlen)
 397 {
 398         struct ovs_header *upcall;
 399         struct sk_buff *nskb = NULL;
 400         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 401         struct nlattr *nla;
 402         size_t len;
 403         unsigned int hlen;
 404         int err, dp_ifindex;
 405
 406         dp_ifindex = get_dpifindex(dp);
 407         if (!dp_ifindex)
 408                 return -ENODEV;
 409
 410         if (skb_vlan_tag_present(skb)) {
 411                 nskb = skb_clone(skb, GFP_ATOMIC);
 412                 if (!nskb)
 413                         return -ENOMEM;
 414
 415                 nskb = __vlan_hwaccel_push_inside(nskb);
 416                 if (!nskb)
 417                         return -ENOMEM;
 418
 419                 skb = nskb;
 420         }
 421
 422         if (nla_attr_size(skb->len) > USHRT_MAX) {
 423                 err = -EFBIG;
 424                 goto out;
 425         }
 426
 427         /* Complete checksum if needed */
 428         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 429             (err = skb_csum_hwoffload_help(skb, 0)))
 430                 goto out;
 431
 432         /* Older versions of OVS user space enforce alignment of the last
 433          * Netlink attribute to NLA_ALIGNTO which would require extensive
 434          * padding logic. Only perform zerocopy if padding is not required.
 435          */
 436         if (dp->user_features & OVS_DP_F_UNALIGNED)
 437                 hlen = skb_zerocopy_headlen(skb);
 438         else
 439                 hlen = skb->len;
 440
 441         len = upcall_msg_size(upcall_info, hlen - cutlen,
 442                               OVS_CB(skb)->acts_origlen);
 443         user_skb = genlmsg_new(len, GFP_ATOMIC);
 444         if (!user_skb) {
 445                 err = -ENOMEM;
 446                 goto out;
 447         }
 448
 449         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 450                              0, upcall_info->cmd);
 451         upcall->dp_ifindex = dp_ifindex;
 452
 453         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 454         BUG_ON(err);
 455
 456         if (upcall_info->userdata)
 457                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 458                           nla_len(upcall_info->userdata),
 459                           nla_data(upcall_info->userdata));
 460
 461         if (upcall_info->egress_tun_info) {
 462                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 463                 err = ovs_nla_put_tunnel_info(user_skb,
 464                                               upcall_info->egress_tun_info);
 465                 BUG_ON(err);
 466                 nla_nest_end(user_skb, nla);
 467         }
 468
 469         if (upcall_info->actions_len) {
 470                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 471                 err = ovs_nla_put_actions(upcall_info->actions,
 472                                           upcall_info->actions_len,
 473                                           user_skb);
 474                 if (!err)
 475                         nla_nest_end(user_skb, nla);
 476                 else
 477                         nla_nest_cancel(user_skb, nla);
 478         }
 479
 480         /* Add OVS_PACKET_ATTR_MRU */
 481         if (upcall_info->mru) {
 482                 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
 483                                 upcall_info->mru)) {
 484                         err = -ENOBUFS;
 485                         goto out;
 486                 }
 487                 pad_packet(dp, user_skb);
 488         }
 489
 490         /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 491         if (cutlen > 0) {
 492                 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
 493                                 skb->len)) {
 494                         err = -ENOBUFS;
 495                         goto out;
 496                 }
 497                 pad_packet(dp, user_skb);
 498         }
 499
 500         /* Only reserve room for attribute header, packet data is added
 501          * in skb_zerocopy() */
 502         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 503                 err = -ENOBUFS;
 504                 goto out;
 505         }
 506         nla->nla_len = nla_attr_size(skb->len - cutlen);
 507
 508         err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 509         if (err)
 510                 goto out;
 511
 512         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 513         pad_packet(dp, user_skb);
 514
 515         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 516
 517         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 518         user_skb = NULL;
 519 out:
 520         if (err)
 521                 skb_tx_error(skb);
 522         kfree_skb(user_skb);
 523         kfree_skb(nskb);
 524         return err;
 525 }
 526
 527 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 528 {
 529         struct ovs_header *ovs_header = info->userhdr;
 530         struct net *net = sock_net(skb->sk);
 531         struct nlattr **a = info->attrs;
 532         struct sw_flow_actions *acts;
 533         struct sk_buff *packet;
 534         struct sw_flow *flow;
 535         struct sw_flow_actions *sf_acts;
 536         struct datapath *dp;
 537         struct vport *input_vport;
 538         u16 mru = 0;
 539         int len;
 540         int err;
 541         bool log = !a[OVS_PACKET_ATTR_PROBE];
 542
 543         err = -EINVAL;
 544         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 545             !a[OVS_PACKET_ATTR_ACTIONS])
 546                 goto err;
 547
 548         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 549         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 550         err = -ENOMEM;
 551         if (!packet)
 552                 goto err;
 553         skb_reserve(packet, NET_IP_ALIGN);
 554
 555         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 556
 557         /* Set packet's mru */
 558         if (a[OVS_PACKET_ATTR_MRU]) {
 559                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 560                 packet->ignore_df = 1;
 561         }
 562         OVS_CB(packet)->mru = mru;
 563
 564         /* Build an sw_flow for sending this packet. */
 565         flow = ovs_flow_alloc();
 566         err = PTR_ERR(flow);
 567         if (IS_ERR(flow))
 568                 goto err_kfree_skb;
 569
 570         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 571                                              packet, &flow->key, log);
 572         if (err)
 573                 goto err_flow_free;
 574
 575         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 576                                    &flow->key, &acts, log);
 577         if (err)
 578                 goto err_flow_free;
 579
 580         rcu_assign_pointer(flow->sf_acts, acts);
 581         packet->priority = flow->key.phy.priority;
 582         packet->mark = flow->key.phy.skb_mark;
 583
 584         rcu_read_lock();
 585         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 586         err = -ENODEV;
 587         if (!dp)
 588                 goto err_unlock;
 589
 590         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 591         if (!input_vport)
 592                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 593
 594         if (!input_vport)
 595                 goto err_unlock;
 596
 597         packet->dev = input_vport->dev;
 598         OVS_CB(packet)->input_vport = input_vport;
 599         sf_acts = rcu_dereference(flow->sf_acts);
 600
 601         local_bh_disable();
 602         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 603         local_bh_enable();
 604         rcu_read_unlock();
 605
 606         ovs_flow_free(flow, false);
 607         return err;
 608
 609 err_unlock:
 610         rcu_read_unlock();
 611 err_flow_free:
 612         ovs_flow_free(flow, false);
 613 err_kfree_skb:
 614         kfree_skb(packet);
 615 err:
 616         return err;
 617 }
 618
 619 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 620         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 621         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 622         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 623         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 624         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 625 };
 626
 627 static const struct genl_ops dp_packet_genl_ops[] = {
 628         { .cmd = OVS_PACKET_CMD_EXECUTE,
 629           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 630           .policy = packet_policy,
 631           .doit = ovs_packet_cmd_execute
 632         }
 633 };
 634
 635 static struct genl_family dp_packet_genl_family __ro_after_init = {
 636         .hdrsize = sizeof(struct ovs_header),
 637         .name = OVS_PACKET_FAMILY,
 638         .version = OVS_PACKET_VERSION,
 639         .maxattr = OVS_PACKET_ATTR_MAX,
 640         .netnsok = true,
 641         .parallel_ops = true,
 642         .ops = dp_packet_genl_ops,
 643         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 644         .module = THIS_MODULE,
 645 };
 646
 647 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 648                          struct ovs_dp_megaflow_stats *mega_stats)
 649 {
 650         int i;
 651
 652         memset(mega_stats, 0, sizeof(*mega_stats));
 653
 654         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 655         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 656
 657         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 658
 659         for_each_possible_cpu(i) {
 660                 const struct dp_stats_percpu *percpu_stats;
 661                 struct dp_stats_percpu local_stats;
 662                 unsigned int start;
 663
 664                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 665
 666                 do {
 667                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 668                         local_stats = *percpu_stats;
 669                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 670
 671                 stats->n_hit += local_stats.n_hit;
 672                 stats->n_missed += local_stats.n_missed;
 673                 stats->n_lost += local_stats.n_lost;
 674                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 675         }
 676 }
 677
 678 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 679 {
 680         return ovs_identifier_is_ufid(sfid) &&
 681                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 682 }
 683
 684 static bool should_fill_mask(uint32_t ufid_flags)
 685 {
 686         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 687 }
 688
 689 static bool should_fill_actions(uint32_t ufid_flags)
 690 {
 691         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 692 }
 693
 694 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 695                                     const struct sw_flow_id *sfid,
 696                                     uint32_t ufid_flags)
 697 {
 698         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 699
 700         /* OVS_FLOW_ATTR_UFID */
 701         if (sfid && ovs_identifier_is_ufid(sfid))
 702                 len += nla_total_size(sfid->ufid_len);
 703
 704         /* OVS_FLOW_ATTR_KEY */
 705         if (!sfid || should_fill_key(sfid, ufid_flags))
 706                 len += nla_total_size(ovs_key_attr_size());
 707
 708         /* OVS_FLOW_ATTR_MASK */
 709         if (should_fill_mask(ufid_flags))
 710                 len += nla_total_size(ovs_key_attr_size());
 711
 712         /* OVS_FLOW_ATTR_ACTIONS */
 713         if (should_fill_actions(ufid_flags))
 714                 len += nla_total_size(acts->orig_len);
 715
 716         return len
 717                 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 718                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 719                 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 720 }
 721
 722 /* Called with ovs_mutex or RCU read lock. */
 723 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 724                                    struct sk_buff *skb)
 725 {
 726         struct ovs_flow_stats stats;
 727         __be16 tcp_flags;
 728         unsigned long used;
 729
 730         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 731
 732         if (used &&
 733             nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 734                               OVS_FLOW_ATTR_PAD))
 735                 return -EMSGSIZE;
 736
 737         if (stats.n_packets &&
 738             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 739                           sizeof(struct ovs_flow_stats), &stats,
 740                           OVS_FLOW_ATTR_PAD))
 741                 return -EMSGSIZE;
 742
 743         if ((u8)ntohs(tcp_flags) &&
 744              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 745                 return -EMSGSIZE;
 746
 747         return 0;
 748 }
 749
 750 /* Called with ovs_mutex or RCU read lock. */
 751 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 752                                      struct sk_buff *skb, int skb_orig_len)
 753 {
 754         struct nlattr *start;
 755         int err;
 756
 757         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 758          * this is the first flow to be dumped into 'skb'.  This is unusual for
 759          * Netlink but individual action lists can be longer than
 760          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 761          * The userspace caller can always fetch the actions separately if it
 762          * really wants them.  (Most userspace callers in fact don't care.)
 763          *
 764          * This can only fail for dump operations because the skb is always
 765          * properly sized for single flows.
 766          */
 767         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 768         if (start) {
 769                 const struct sw_flow_actions *sf_acts;
 770
 771                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 772                 err = ovs_nla_put_actions(sf_acts->actions,
 773                                           sf_acts->actions_len, skb);
 774
 775                 if (!err)
 776                         nla_nest_end(skb, start);
 777                 else {
 778                         if (skb_orig_len)
 779                                 return err;
 780
 781                         nla_nest_cancel(skb, start);
 782                 }
 783         } else if (skb_orig_len) {
 784                 return -EMSGSIZE;
 785         }
 786
 787         return 0;
 788 }
 789
 790 /* Called with ovs_mutex or RCU read lock. */
 791 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 792                                   struct sk_buff *skb, u32 portid,
 793                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 794 {
 795         const int skb_orig_len = skb->len;
 796         struct ovs_header *ovs_header;
 797         int err;
 798
 799         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 800                                  flags, cmd);
 801         if (!ovs_header)
 802                 return -EMSGSIZE;
 803
 804         ovs_header->dp_ifindex = dp_ifindex;
 805
 806         err = ovs_nla_put_identifier(flow, skb);
 807         if (err)
 808                 goto error;
 809
 810         if (should_fill_key(&flow->id, ufid_flags)) {
 811                 err = ovs_nla_put_masked_key(flow, skb);
 812                 if (err)
 813                         goto error;
 814         }
 815
 816         if (should_fill_mask(ufid_flags)) {
 817                 err = ovs_nla_put_mask(flow, skb);
 818                 if (err)
 819                         goto error;
 820         }
 821
 822         err = ovs_flow_cmd_fill_stats(flow, skb);
 823         if (err)
 824                 goto error;
 825
 826         if (should_fill_actions(ufid_flags)) {
 827                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 828                 if (err)
 829                         goto error;
 830         }
 831
 832         genlmsg_end(skb, ovs_header);
 833         return 0;
 834
 835 error:
 836         genlmsg_cancel(skb, ovs_header);
 837         return err;
 838 }
 839
 840 /* May not be called with RCU read lock. */
 841 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 842                                                const struct sw_flow_id *sfid,
 843                                                struct genl_info *info,
 844                                                bool always,
 845                                                uint32_t ufid_flags)
 846 {
 847         struct sk_buff *skb;
 848         size_t len;
 849
 850         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 851                 return NULL;
 852
 853         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 854         skb = genlmsg_new(len, GFP_KERNEL);
 855         if (!skb)
 856                 return ERR_PTR(-ENOMEM);
 857
 858         return skb;
 859 }
 860
 861 /* Called with ovs_mutex. */
 862 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 863                                                int dp_ifindex,
 864                                                struct genl_info *info, u8 cmd,
 865                                                bool always, u32 ufid_flags)
 866 {
 867         struct sk_buff *skb;
 868         int retval;
 869
 870         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 871                                       &flow->id, info, always, ufid_flags);
 872         if (IS_ERR_OR_NULL(skb))
 873                 return skb;
 874
 875         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 876                                         info->snd_portid, info->snd_seq, 0,
 877                                         cmd, ufid_flags);
 878         BUG_ON(retval < 0);
 879         return skb;
 880 }
 881
 882 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 883 {
 884         struct net *net = sock_net(skb->sk);
 885         struct nlattr **a = info->attrs;
 886         struct ovs_header *ovs_header = info->userhdr;
 887         struct sw_flow *flow = NULL, *new_flow;
 888         struct sw_flow_mask mask;
 889         struct sk_buff *reply;
 890         struct datapath *dp;
 891         struct sw_flow_actions *acts;
 892         struct sw_flow_match match;
 893         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 894         int error;
 895         bool log = !a[OVS_FLOW_ATTR_PROBE];
 896
 897         /* Must have key and actions. */
 898         error = -EINVAL;
 899         if (!a[OVS_FLOW_ATTR_KEY]) {
 900                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 901                 goto error;
 902         }
 903         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 904                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 905                 goto error;
 906         }
 907
 908         /* Most of the time we need to allocate a new flow, do it before
 909          * locking.
 910          */
 911         new_flow = ovs_flow_alloc();
 912         if (IS_ERR(new_flow)) {
 913                 error = PTR_ERR(new_flow);
 914                 goto error;
 915         }
 916
 917         /* Extract key. */
 918         ovs_match_init(&match, &new_flow->key, false, &mask);
 919         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 920                                   a[OVS_FLOW_ATTR_MASK], log);
 921         if (error)
 922                 goto err_kfree_flow;
 923
 924         /* Extract flow identifier. */
 925         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 926                                        &new_flow->key, log);
 927         if (error)
 928                 goto err_kfree_flow;
 929
 930         /* unmasked key is needed to match when ufid is not used. */
 931         if (ovs_identifier_is_key(&new_flow->id))
 932                 match.key = new_flow->id.unmasked_key;
 933
 934         ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
 935
 936         /* Validate actions. */
 937         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 938                                      &new_flow->key, &acts, log);
 939         if (error) {
 940                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 941                 goto err_kfree_flow;
 942         }
 943
 944         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 945                                         ufid_flags);
 946         if (IS_ERR(reply)) {
 947                 error = PTR_ERR(reply);
 948                 goto err_kfree_acts;
 949         }
 950
 951         ovs_lock();
 952         dp = get_dp(net, ovs_header->dp_ifindex);
 953         if (unlikely(!dp)) {
 954                 error = -ENODEV;
 955                 goto err_unlock_ovs;
 956         }
 957
 958         /* Check if this is a duplicate flow */
 959         if (ovs_identifier_is_ufid(&new_flow->id))
 960                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
 961         if (!flow)
 962                 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
 963         if (likely(!flow)) {
 964                 rcu_assign_pointer(new_flow->sf_acts, acts);
 965
 966                 /* Put flow in bucket. */
 967                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 968                 if (unlikely(error)) {
 969                         acts = NULL;
 970                         goto err_unlock_ovs;
 971                 }
 972
 973                 if (unlikely(reply)) {
 974                         error = ovs_flow_cmd_fill_info(new_flow,
 975                                                        ovs_header->dp_ifindex,
 976                                                        reply, info->snd_portid,
 977                                                        info->snd_seq, 0,
 978                                                        OVS_FLOW_CMD_NEW,
 979                                                        ufid_flags);
 980                         BUG_ON(error < 0);
 981                 }
 982                 ovs_unlock();
 983         } else {
 984                 struct sw_flow_actions *old_acts;
 985
 986                 /* Bail out if we're not allowed to modify an existing flow.
 987                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 988                  * because Generic Netlink treats the latter as a dump
 989                  * request.  We also accept NLM_F_EXCL in case that bug ever
 990                  * gets fixed.
 991                  */
 992                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
 993                                                          | NLM_F_EXCL))) {
 994                         error = -EEXIST;
 995                         goto err_unlock_ovs;
 996                 }
 997                 /* The flow identifier has to be the same for flow updates.
 998                  * Look for any overlapping flow.
 999                  */
1000                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1001                         if (ovs_identifier_is_key(&flow->id))
1002                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1003                                                                  &match);
1004                         else /* UFID matches but key is different */
1005                                 flow = NULL;
1006                         if (!flow) {
1007                                 error = -ENOENT;
1008                                 goto err_unlock_ovs;
1009                         }
1010                 }
1011                 /* Update actions. */
1012                 old_acts = ovsl_dereference(flow->sf_acts);
1013                 rcu_assign_pointer(flow->sf_acts, acts);
1014
1015                 if (unlikely(reply)) {
1016                         error = ovs_flow_cmd_fill_info(flow,
1017                                                        ovs_header->dp_ifindex,
1018                                                        reply, info->snd_portid,
1019                                                        info->snd_seq, 0,
1020                                                        OVS_FLOW_CMD_NEW,
1021                                                        ufid_flags);
1022                         BUG_ON(error < 0);
1023                 }
1024                 ovs_unlock();
1025
1026                 ovs_nla_free_flow_actions_rcu(old_acts);
1027                 ovs_flow_free(new_flow, false);
1028         }
1029
1030         if (reply)
1031                 ovs_notify(&dp_flow_genl_family, reply, info);
1032         return 0;
1033
1034 err_unlock_ovs:
1035         ovs_unlock();
1036         kfree_skb(reply);
1037 err_kfree_acts:
1038         ovs_nla_free_flow_actions(acts);
1039 err_kfree_flow:
1040         ovs_flow_free(new_flow, false);
1041 error:
1042         return error;
1043 }
1044
1045 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1046 static struct sw_flow_actions *get_flow_actions(struct net *net,
1047                                                 const struct nlattr *a,
1048                                                 const struct sw_flow_key *key,
1049                                                 const struct sw_flow_mask *mask,
1050                                                 bool log)
1051 {
1052         struct sw_flow_actions *acts;
1053         struct sw_flow_key masked_key;
1054         int error;
1055
1056         ovs_flow_mask_key(&masked_key, key, true, mask);
1057         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1058         if (error) {
1059                 OVS_NLERR(log,
1060                           "Actions may not be safe on all matching packets");
1061                 return ERR_PTR(error);
1062         }
1063
1064         return acts;
1065 }
1066
1067 /* Factor out match-init and action-copy to avoid
1068  * "Wframe-larger-than=1024" warning. Because mask is only
1069  * used to get actions, we new a function to save some
1070  * stack space.
1071  *
1072  * If there are not key and action attrs, we return 0
1073  * directly. In the case, the caller will also not use the
1074  * match as before. If there is action attr, we try to get
1075  * actions and save them to *acts. Before returning from
1076  * the function, we reset the match->mask pointer. Because
1077  * we should not to return match object with dangling reference
1078  * to mask.
1079  * */
1080 static int ovs_nla_init_match_and_action(struct net *net,
1081                                          struct sw_flow_match *match,
1082                                          struct sw_flow_key *key,
1083                                          struct nlattr **a,
1084                                          struct sw_flow_actions **acts,
1085                                          bool log)
1086 {
1087         struct sw_flow_mask mask;
1088         int error = 0;
1089
1090         if (a[OVS_FLOW_ATTR_KEY]) {
1091                 ovs_match_init(match, key, true, &mask);
1092                 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1093                                           a[OVS_FLOW_ATTR_MASK], log);
1094                 if (error)
1095                         goto error;
1096         }
1097
1098         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1099                 if (!a[OVS_FLOW_ATTR_KEY]) {
1100                         OVS_NLERR(log,
1101                                   "Flow key attribute not present in set flow.");
1102                         error = -EINVAL;
1103                         goto error;
1104                 }
1105
1106                 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1107                                          &mask, log);
1108                 if (IS_ERR(*acts)) {
1109                         error = PTR_ERR(*acts);
1110                         goto error;
1111                 }
1112         }
1113
1114         /* On success, error is 0. */
1115 error:
1116         match->mask = NULL;
1117         return error;
1118 }
1119
1120 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1121 {
1122         struct net *net = sock_net(skb->sk);
1123         struct nlattr **a = info->attrs;
1124         struct ovs_header *ovs_header = info->userhdr;
1125         struct sw_flow_key key;
1126         struct sw_flow *flow;
1127         struct sk_buff *reply = NULL;
1128         struct datapath *dp;
1129         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1130         struct sw_flow_match match;
1131         struct sw_flow_id sfid;
1132         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1133         int error = 0;
1134         bool log = !a[OVS_FLOW_ATTR_PROBE];
1135         bool ufid_present;
1136
1137         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1138         if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1139                 OVS_NLERR(log,
1140                           "Flow set message rejected, Key attribute missing.");
1141                 return -EINVAL;
1142         }
1143
1144         error = ovs_nla_init_match_and_action(net, &match, &key, a,
1145                                               &acts, log);
1146         if (error)
1147                 goto error;
1148
1149         if (acts) {
1150                 /* Can allocate before locking if have acts. */
1151                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1152                                                 ufid_flags);
1153                 if (IS_ERR(reply)) {
1154                         error = PTR_ERR(reply);
1155                         goto err_kfree_acts;
1156                 }
1157         }
1158
1159         ovs_lock();
1160         dp = get_dp(net, ovs_header->dp_ifindex);
1161         if (unlikely(!dp)) {
1162                 error = -ENODEV;
1163                 goto err_unlock_ovs;
1164         }
1165         /* Check that the flow exists. */
1166         if (ufid_present)
1167                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1168         else
1169                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1170         if (unlikely(!flow)) {
1171                 error = -ENOENT;
1172                 goto err_unlock_ovs;
1173         }
1174
1175         /* Update actions, if present. */
1176         if (likely(acts)) {
1177                 old_acts = ovsl_dereference(flow->sf_acts);
1178                 rcu_assign_pointer(flow->sf_acts, acts);
1179
1180                 if (unlikely(reply)) {
1181                         error = ovs_flow_cmd_fill_info(flow,
1182                                                        ovs_header->dp_ifindex,
1183                                                        reply, info->snd_portid,
1184                                                        info->snd_seq, 0,
1185                                                        OVS_FLOW_CMD_NEW,
1186                                                        ufid_flags);
1187                         BUG_ON(error < 0);
1188                 }
1189         } else {
1190                 /* Could not alloc without acts before locking. */
1191                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1192                                                 info, OVS_FLOW_CMD_NEW, false,
1193                                                 ufid_flags);
1194
1195                 if (IS_ERR(reply)) {
1196                         error = PTR_ERR(reply);
1197                         goto err_unlock_ovs;
1198                 }
1199         }
1200
1201         /* Clear stats. */
1202         if (a[OVS_FLOW_ATTR_CLEAR])
1203                 ovs_flow_stats_clear(flow);
1204         ovs_unlock();
1205
1206         if (reply)
1207                 ovs_notify(&dp_flow_genl_family, reply, info);
1208         if (old_acts)
1209                 ovs_nla_free_flow_actions_rcu(old_acts);
1210
1211         return 0;
1212
1213 err_unlock_ovs:
1214         ovs_unlock();
1215         kfree_skb(reply);
1216 err_kfree_acts:
1217         ovs_nla_free_flow_actions(acts);
1218 error:
1219         return error;
1220 }
1221
1222 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1223 {
1224         struct nlattr **a = info->attrs;
1225         struct ovs_header *ovs_header = info->userhdr;
1226         struct net *net = sock_net(skb->sk);
1227         struct sw_flow_key key;
1228         struct sk_buff *reply;
1229         struct sw_flow *flow;
1230         struct datapath *dp;
1231         struct sw_flow_match match;
1232         struct sw_flow_id ufid;
1233         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1234         int err = 0;
1235         bool log = !a[OVS_FLOW_ATTR_PROBE];
1236         bool ufid_present;
1237
1238         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1239         if (a[OVS_FLOW_ATTR_KEY]) {
1240                 ovs_match_init(&match, &key, true, NULL);
1241                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1242                                         log);
1243         } else if (!ufid_present) {
1244                 OVS_NLERR(log,
1245                           "Flow get message rejected, Key attribute missing.");
1246                 err = -EINVAL;
1247         }
1248         if (err)
1249                 return err;
1250
1251         ovs_lock();
1252         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1253         if (!dp) {
1254                 err = -ENODEV;
1255                 goto unlock;
1256         }
1257
1258         if (ufid_present)
1259                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1260         else
1261                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1262         if (!flow) {
1263                 err = -ENOENT;
1264                 goto unlock;
1265         }
1266
1267         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1268                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1269         if (IS_ERR(reply)) {
1270                 err = PTR_ERR(reply);
1271                 goto unlock;
1272         }
1273
1274         ovs_unlock();
1275         return genlmsg_reply(reply, info);
1276 unlock:
1277         ovs_unlock();
1278         return err;
1279 }
1280
1281 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1282 {
1283         struct nlattr **a = info->attrs;
1284         struct ovs_header *ovs_header = info->userhdr;
1285         struct net *net = sock_net(skb->sk);
1286         struct sw_flow_key key;
1287         struct sk_buff *reply;
1288         struct sw_flow *flow = NULL;
1289         struct datapath *dp;
1290         struct sw_flow_match match;
1291         struct sw_flow_id ufid;
1292         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1293         int err;
1294         bool log = !a[OVS_FLOW_ATTR_PROBE];
1295         bool ufid_present;
1296
1297         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1298         if (a[OVS_FLOW_ATTR_KEY]) {
1299                 ovs_match_init(&match, &key, true, NULL);
1300                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1301                                         NULL, log);
1302                 if (unlikely(err))
1303                         return err;
1304         }
1305
1306         ovs_lock();
1307         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1308         if (unlikely(!dp)) {
1309                 err = -ENODEV;
1310                 goto unlock;
1311         }
1312
1313         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1314                 err = ovs_flow_tbl_flush(&dp->table);
1315                 goto unlock;
1316         }
1317
1318         if (ufid_present)
1319                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1320         else
1321                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1322         if (unlikely(!flow)) {
1323                 err = -ENOENT;
1324                 goto unlock;
1325         }
1326
1327         ovs_flow_tbl_remove(&dp->table, flow);
1328         ovs_unlock();
1329
1330         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1331                                         &flow->id, info, false, ufid_flags);
1332         if (likely(reply)) {
1333                 if (likely(!IS_ERR(reply))) {
1334                         rcu_read_lock();        /*To keep RCU checker happy. */
1335                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1336                                                      reply, info->snd_portid,
1337                                                      info->snd_seq, 0,
1338                                                      OVS_FLOW_CMD_DEL,
1339                                                      ufid_flags);
1340                         rcu_read_unlock();
1341                         BUG_ON(err < 0);
1342
1343                         ovs_notify(&dp_flow_genl_family, reply, info);
1344                 } else {
1345                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1346                 }
1347         }
1348
1349         ovs_flow_free(flow, true);
1350         return 0;
1351 unlock:
1352         ovs_unlock();
1353         return err;
1354 }
1355
1356 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1357 {
1358         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1359         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1360         struct table_instance *ti;
1361         struct datapath *dp;
1362         u32 ufid_flags;
1363         int err;
1364
1365         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1366                             OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1367         if (err)
1368                 return err;
1369         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1370
1371         rcu_read_lock();
1372         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1373         if (!dp) {
1374                 rcu_read_unlock();
1375                 return -ENODEV;
1376         }
1377
1378         ti = rcu_dereference(dp->table.ti);
1379         for (;;) {
1380                 struct sw_flow *flow;
1381                 u32 bucket, obj;
1382
1383                 bucket = cb->args[0];
1384                 obj = cb->args[1];
1385                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1386                 if (!flow)
1387                         break;
1388
1389                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1390                                            NETLINK_CB(cb->skb).portid,
1391                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1392                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1393                         break;
1394
1395                 cb->args[0] = bucket;
1396                 cb->args[1] = obj;
1397         }
1398         rcu_read_unlock();
1399         return skb->len;
1400 }
1401
1402 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1403         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1404         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1405         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1406         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1407         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1408         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1409         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1410 };
1411
1412 static const struct genl_ops dp_flow_genl_ops[] = {
1413         { .cmd = OVS_FLOW_CMD_NEW,
1414           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1415           .policy = flow_policy,
1416           .doit = ovs_flow_cmd_new
1417         },
1418         { .cmd = OVS_FLOW_CMD_DEL,
1419           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1420           .policy = flow_policy,
1421           .doit = ovs_flow_cmd_del
1422         },
1423         { .cmd = OVS_FLOW_CMD_GET,
1424           .flags = 0,               /* OK for unprivileged users. */
1425           .policy = flow_policy,
1426           .doit = ovs_flow_cmd_get,
1427           .dumpit = ovs_flow_cmd_dump
1428         },
1429         { .cmd = OVS_FLOW_CMD_SET,
1430           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1431           .policy = flow_policy,
1432           .doit = ovs_flow_cmd_set,
1433         },
1434 };
1435
1436 static struct genl_family dp_flow_genl_family __ro_after_init = {
1437         .hdrsize = sizeof(struct ovs_header),
1438         .name = OVS_FLOW_FAMILY,
1439         .version = OVS_FLOW_VERSION,
1440         .maxattr = OVS_FLOW_ATTR_MAX,
1441         .netnsok = true,
1442         .parallel_ops = true,
1443         .ops = dp_flow_genl_ops,
1444         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1445         .mcgrps = &ovs_dp_flow_multicast_group,
1446         .n_mcgrps = 1,
1447         .module = THIS_MODULE,
1448 };
1449
1450 static size_t ovs_dp_cmd_msg_size(void)
1451 {
1452         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1453
1454         msgsize += nla_total_size(IFNAMSIZ);
1455         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1456         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1457         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1458
1459         return msgsize;
1460 }
1461
1462 /* Called with ovs_mutex. */
1463 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1464                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1465 {
1466         struct ovs_header *ovs_header;
1467         struct ovs_dp_stats dp_stats;
1468         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1469         int err;
1470
1471         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1472                                    flags, cmd);
1473         if (!ovs_header)
1474                 goto error;
1475
1476         ovs_header->dp_ifindex = get_dpifindex(dp);
1477
1478         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1479         if (err)
1480                 goto nla_put_failure;
1481
1482         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1483         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1484                           &dp_stats, OVS_DP_ATTR_PAD))
1485                 goto nla_put_failure;
1486
1487         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1488                           sizeof(struct ovs_dp_megaflow_stats),
1489                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1490                 goto nla_put_failure;
1491
1492         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1493                 goto nla_put_failure;
1494
1495         genlmsg_end(skb, ovs_header);
1496         return 0;
1497
1498 nla_put_failure:
1499         genlmsg_cancel(skb, ovs_header);
1500 error:
1501         return -EMSGSIZE;
1502 }
1503
1504 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1505 {
1506         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1507 }
1508
1509 /* Called with rcu_read_lock or ovs_mutex. */
1510 static struct datapath *lookup_datapath(struct net *net,
1511                                         const struct ovs_header *ovs_header,
1512                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1513 {
1514         struct datapath *dp;
1515
1516         if (!a[OVS_DP_ATTR_NAME])
1517                 dp = get_dp(net, ovs_header->dp_ifindex);
1518         else {
1519                 struct vport *vport;
1520
1521                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1522                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1523         }
1524         return dp ? dp : ERR_PTR(-ENODEV);
1525 }
1526
1527 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1528 {
1529         struct datapath *dp;
1530
1531         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1532         if (IS_ERR(dp))
1533                 return;
1534
1535         WARN(dp->user_features, "Dropping previously announced user features\n");
1536         dp->user_features = 0;
1537 }
1538
1539 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1540 {
1541         if (a[OVS_DP_ATTR_USER_FEATURES])
1542                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1543 }
1544
1545 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1546 {
1547         struct nlattr **a = info->attrs;
1548         struct vport_parms parms;
1549         struct sk_buff *reply;
1550         struct datapath *dp;
1551         struct vport *vport;
1552         struct ovs_net *ovs_net;
1553         int err, i;
1554
1555         err = -EINVAL;
1556         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1557                 goto err;
1558
1559         reply = ovs_dp_cmd_alloc_info();
1560         if (!reply)
1561                 return -ENOMEM;
1562
1563         err = -ENOMEM;
1564         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1565         if (dp == NULL)
1566                 goto err_free_reply;
1567
1568         ovs_dp_set_net(dp, sock_net(skb->sk));
1569
1570         /* Allocate table. */
1571         err = ovs_flow_tbl_init(&dp->table);
1572         if (err)
1573                 goto err_free_dp;
1574
1575         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1576         if (!dp->stats_percpu) {
1577                 err = -ENOMEM;
1578                 goto err_destroy_table;
1579         }
1580
1581         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1582                             GFP_KERNEL);
1583         if (!dp->ports) {
1584                 err = -ENOMEM;
1585                 goto err_destroy_percpu;
1586         }
1587
1588         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1589                 INIT_HLIST_HEAD(&dp->ports[i]);
1590
1591         err = ovs_meters_init(dp);
1592         if (err)
1593                 goto err_destroy_ports_array;
1594
1595         /* Set up our datapath device. */
1596         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1597         parms.type = OVS_VPORT_TYPE_INTERNAL;
1598         parms.options = NULL;
1599         parms.dp = dp;
1600         parms.port_no = OVSP_LOCAL;
1601         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1602
1603         ovs_dp_change(dp, a);
1604
1605         /* So far only local changes have been made, now need the lock. */
1606         ovs_lock();
1607
1608         vport = new_vport(&parms);
1609         if (IS_ERR(vport)) {
1610                 err = PTR_ERR(vport);
1611                 if (err == -EBUSY)
1612                         err = -EEXIST;
1613
1614                 if (err == -EEXIST) {
1615                         /* An outdated user space instance that does not understand
1616                          * the concept of user_features has attempted to create a new
1617                          * datapath and is likely to reuse it. Drop all user features.
1618                          */
1619                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1620                                 ovs_dp_reset_user_features(skb, info);
1621                 }
1622
1623                 goto err_destroy_meters;
1624         }
1625
1626         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1627                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1628         BUG_ON(err < 0);
1629
1630         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1631         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1632
1633         ovs_unlock();
1634
1635         ovs_notify(&dp_datapath_genl_family, reply, info);
1636         return 0;
1637
1638 err_destroy_meters:
1639         ovs_unlock();
1640         ovs_meters_exit(dp);
1641 err_destroy_ports_array:
1642         kfree(dp->ports);
1643 err_destroy_percpu:
1644         free_percpu(dp->stats_percpu);
1645 err_destroy_table:
1646         ovs_flow_tbl_destroy(&dp->table);
1647 err_free_dp:
1648         kfree(dp);
1649 err_free_reply:
1650         kfree_skb(reply);
1651 err:
1652         return err;
1653 }
1654
1655 /* Called with ovs_mutex. */
1656 static void __dp_destroy(struct datapath *dp)
1657 {
1658         int i;
1659
1660         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1661                 struct vport *vport;
1662                 struct hlist_node *n;
1663
1664                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1665                         if (vport->port_no != OVSP_LOCAL)
1666                                 ovs_dp_detach_port(vport);
1667         }
1668
1669         list_del_rcu(&dp->list_node);
1670
1671         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1672          * all ports in datapath are destroyed first before freeing datapath.
1673          */
1674         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1675
1676         /* RCU destroy the flow table */
1677         call_rcu(&dp->rcu, destroy_dp_rcu);
1678 }
1679
1680 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1681 {
1682         struct sk_buff *reply;
1683         struct datapath *dp;
1684         int err;
1685
1686         reply = ovs_dp_cmd_alloc_info();
1687         if (!reply)
1688                 return -ENOMEM;
1689
1690         ovs_lock();
1691         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1692         err = PTR_ERR(dp);
1693         if (IS_ERR(dp))
1694                 goto err_unlock_free;
1695
1696         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1697                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1698         BUG_ON(err < 0);
1699
1700         __dp_destroy(dp);
1701         ovs_unlock();
1702
1703         ovs_notify(&dp_datapath_genl_family, reply, info);
1704
1705         return 0;
1706
1707 err_unlock_free:
1708         ovs_unlock();
1709         kfree_skb(reply);
1710         return err;
1711 }
1712
1713 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1714 {
1715         struct sk_buff *reply;
1716         struct datapath *dp;
1717         int err;
1718
1719         reply = ovs_dp_cmd_alloc_info();
1720         if (!reply)
1721                 return -ENOMEM;
1722
1723         ovs_lock();
1724         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1725         err = PTR_ERR(dp);
1726         if (IS_ERR(dp))
1727                 goto err_unlock_free;
1728
1729         ovs_dp_change(dp, info->attrs);
1730
1731         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1732                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1733         BUG_ON(err < 0);
1734
1735         ovs_unlock();
1736         ovs_notify(&dp_datapath_genl_family, reply, info);
1737
1738         return 0;
1739
1740 err_unlock_free:
1741         ovs_unlock();
1742         kfree_skb(reply);
1743         return err;
1744 }
1745
1746 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1747 {
1748         struct sk_buff *reply;
1749         struct datapath *dp;
1750         int err;
1751
1752         reply = ovs_dp_cmd_alloc_info();
1753         if (!reply)
1754                 return -ENOMEM;
1755
1756         ovs_lock();
1757         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1758         if (IS_ERR(dp)) {
1759                 err = PTR_ERR(dp);
1760                 goto err_unlock_free;
1761         }
1762         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1763                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1764         BUG_ON(err < 0);
1765         ovs_unlock();
1766
1767         return genlmsg_reply(reply, info);
1768
1769 err_unlock_free:
1770         ovs_unlock();
1771         kfree_skb(reply);
1772         return err;
1773 }
1774
1775 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1776 {
1777         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1778         struct datapath *dp;
1779         int skip = cb->args[0];
1780         int i = 0;
1781
1782         ovs_lock();
1783         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1784                 if (i >= skip &&
1785                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1786                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1787                                          OVS_DP_CMD_NEW) < 0)
1788                         break;
1789                 i++;
1790         }
1791         ovs_unlock();
1792
1793         cb->args[0] = i;
1794
1795         return skb->len;
1796 }
1797
1798 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1799         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1800         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1801         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1802 };
1803
1804 static const struct genl_ops dp_datapath_genl_ops[] = {
1805         { .cmd = OVS_DP_CMD_NEW,
1806           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1807           .policy = datapath_policy,
1808           .doit = ovs_dp_cmd_new
1809         },
1810         { .cmd = OVS_DP_CMD_DEL,
1811           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1812           .policy = datapath_policy,
1813           .doit = ovs_dp_cmd_del
1814         },
1815         { .cmd = OVS_DP_CMD_GET,
1816           .flags = 0,               /* OK for unprivileged users. */
1817           .policy = datapath_policy,
1818           .doit = ovs_dp_cmd_get,
1819           .dumpit = ovs_dp_cmd_dump
1820         },
1821         { .cmd = OVS_DP_CMD_SET,
1822           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1823           .policy = datapath_policy,
1824           .doit = ovs_dp_cmd_set,
1825         },
1826 };
1827
1828 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1829         .hdrsize = sizeof(struct ovs_header),
1830         .name = OVS_DATAPATH_FAMILY,
1831         .version = OVS_DATAPATH_VERSION,
1832         .maxattr = OVS_DP_ATTR_MAX,
1833         .netnsok = true,
1834         .parallel_ops = true,
1835         .ops = dp_datapath_genl_ops,
1836         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1837         .mcgrps = &ovs_dp_datapath_multicast_group,
1838         .n_mcgrps = 1,
1839         .module = THIS_MODULE,
1840 };
1841
1842 /* Called with ovs_mutex or RCU read lock. */
1843 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1844                                    struct net *net, u32 portid, u32 seq,
1845                                    u32 flags, u8 cmd)
1846 {
1847         struct ovs_header *ovs_header;
1848         struct ovs_vport_stats vport_stats;
1849         int err;
1850
1851         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1852                                  flags, cmd);
1853         if (!ovs_header)
1854                 return -EMSGSIZE;
1855
1856         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1857
1858         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1859             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1860             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1861                            ovs_vport_name(vport)) ||
1862             nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1863                 goto nla_put_failure;
1864
1865         if (!net_eq(net, dev_net(vport->dev))) {
1866                 int id = peernet2id_alloc(net, dev_net(vport->dev));
1867
1868                 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1869                         goto nla_put_failure;
1870         }
1871
1872         ovs_vport_get_stats(vport, &vport_stats);
1873         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1874                           sizeof(struct ovs_vport_stats), &vport_stats,
1875                           OVS_VPORT_ATTR_PAD))
1876                 goto nla_put_failure;
1877
1878         if (ovs_vport_get_upcall_portids(vport, skb))
1879                 goto nla_put_failure;
1880
1881         err = ovs_vport_get_options(vport, skb);
1882         if (err == -EMSGSIZE)
1883                 goto error;
1884
1885         genlmsg_end(skb, ovs_header);
1886         return 0;
1887
1888 nla_put_failure:
1889         err = -EMSGSIZE;
1890 error:
1891         genlmsg_cancel(skb, ovs_header);
1892         return err;
1893 }
1894
1895 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1896 {
1897         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1898 }
1899
1900 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1901 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1902                                          u32 portid, u32 seq, u8 cmd)
1903 {
1904         struct sk_buff *skb;
1905         int retval;
1906
1907         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1908         if (!skb)
1909                 return ERR_PTR(-ENOMEM);
1910
1911         retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd);
1912         BUG_ON(retval < 0);
1913
1914         return skb;
1915 }
1916
1917 /* Called with ovs_mutex or RCU read lock. */
1918 static struct vport *lookup_vport(struct net *net,
1919                                   const struct ovs_header *ovs_header,
1920                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1921 {
1922         struct datapath *dp;
1923         struct vport *vport;
1924
1925         if (a[OVS_VPORT_ATTR_IFINDEX])
1926                 return ERR_PTR(-EOPNOTSUPP);
1927         if (a[OVS_VPORT_ATTR_NAME]) {
1928                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1929                 if (!vport)
1930                         return ERR_PTR(-ENODEV);
1931                 if (ovs_header->dp_ifindex &&
1932                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1933                         return ERR_PTR(-ENODEV);
1934                 return vport;
1935         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1936                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1937
1938                 if (port_no >= DP_MAX_PORTS)
1939                         return ERR_PTR(-EFBIG);
1940
1941                 dp = get_dp(net, ovs_header->dp_ifindex);
1942                 if (!dp)
1943                         return ERR_PTR(-ENODEV);
1944
1945                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1946                 if (!vport)
1947                         return ERR_PTR(-ENODEV);
1948                 return vport;
1949         } else
1950                 return ERR_PTR(-EINVAL);
1951
1952 }
1953
1954 /* Called with ovs_mutex */
1955 static void update_headroom(struct datapath *dp)
1956 {
1957         unsigned dev_headroom, max_headroom = 0;
1958         struct net_device *dev;
1959         struct vport *vport;
1960         int i;
1961
1962         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1963                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1964                         dev = vport->dev;
1965                         dev_headroom = netdev_get_fwd_headroom(dev);
1966                         if (dev_headroom > max_headroom)
1967                                 max_headroom = dev_headroom;
1968                 }
1969         }
1970
1971         dp->max_headroom = max_headroom;
1972         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1973                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1974                         netdev_set_rx_headroom(vport->dev, max_headroom);
1975 }
1976
1977 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1978 {
1979         struct nlattr **a = info->attrs;
1980         struct ovs_header *ovs_header = info->userhdr;
1981         struct vport_parms parms;
1982         struct sk_buff *reply;
1983         struct vport *vport;
1984         struct datapath *dp;
1985         u32 port_no;
1986         int err;
1987
1988         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1989             !a[OVS_VPORT_ATTR_UPCALL_PID])
1990                 return -EINVAL;
1991         if (a[OVS_VPORT_ATTR_IFINDEX])
1992                 return -EOPNOTSUPP;
1993
1994         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1995                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1996         if (port_no >= DP_MAX_PORTS)
1997                 return -EFBIG;
1998
1999         reply = ovs_vport_cmd_alloc_info();
2000         if (!reply)
2001                 return -ENOMEM;
2002
2003         ovs_lock();
2004 restart:
2005         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2006         err = -ENODEV;
2007         if (!dp)
2008                 goto exit_unlock_free;
2009
2010         if (port_no) {
2011                 vport = ovs_vport_ovsl(dp, port_no);
2012                 err = -EBUSY;
2013                 if (vport)
2014                         goto exit_unlock_free;
2015         } else {
2016                 for (port_no = 1; ; port_no++) {
2017                         if (port_no >= DP_MAX_PORTS) {
2018                                 err = -EFBIG;
2019                                 goto exit_unlock_free;
2020                         }
2021                         vport = ovs_vport_ovsl(dp, port_no);
2022                         if (!vport)
2023                                 break;
2024                 }
2025         }
2026
2027         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2028         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2029         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2030         parms.dp = dp;
2031         parms.port_no = port_no;
2032         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2033
2034         vport = new_vport(&parms);
2035         err = PTR_ERR(vport);
2036         if (IS_ERR(vport)) {
2037                 if (err == -EAGAIN)
2038                         goto restart;
2039                 goto exit_unlock_free;
2040         }
2041
2042         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2043                                       info->snd_portid, info->snd_seq, 0,
2044                                       OVS_VPORT_CMD_NEW);
2045
2046         if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2047                 update_headroom(dp);
2048         else
2049                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2050
2051         BUG_ON(err < 0);
2052         ovs_unlock();
2053
2054         ovs_notify(&dp_vport_genl_family, reply, info);
2055         return 0;
2056
2057 exit_unlock_free:
2058         ovs_unlock();
2059         kfree_skb(reply);
2060         return err;
2061 }
2062
2063 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2064 {
2065         struct nlattr **a = info->attrs;
2066         struct sk_buff *reply;
2067         struct vport *vport;
2068         int err;
2069
2070         reply = ovs_vport_cmd_alloc_info();
2071         if (!reply)
2072                 return -ENOMEM;
2073
2074         ovs_lock();
2075         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2076         err = PTR_ERR(vport);
2077         if (IS_ERR(vport))
2078                 goto exit_unlock_free;
2079
2080         if (a[OVS_VPORT_ATTR_TYPE] &&
2081             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2082                 err = -EINVAL;
2083                 goto exit_unlock_free;
2084         }
2085
2086         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2087                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2088                 if (err)
2089                         goto exit_unlock_free;
2090         }
2091
2092
2093         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2094                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2095
2096                 err = ovs_vport_set_upcall_portids(vport, ids);
2097                 if (err)
2098                         goto exit_unlock_free;
2099         }
2100
2101         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2102                                       info->snd_portid, info->snd_seq, 0,
2103                                       OVS_VPORT_CMD_NEW);
2104         BUG_ON(err < 0);
2105
2106         ovs_unlock();
2107         ovs_notify(&dp_vport_genl_family, reply, info);
2108         return 0;
2109
2110 exit_unlock_free:
2111         ovs_unlock();
2112         kfree_skb(reply);
2113         return err;
2114 }
2115
2116 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2117 {
2118         bool must_update_headroom = false;
2119         struct nlattr **a = info->attrs;
2120         struct sk_buff *reply;
2121         struct datapath *dp;
2122         struct vport *vport;
2123         int err;
2124
2125         reply = ovs_vport_cmd_alloc_info();
2126         if (!reply)
2127                 return -ENOMEM;
2128
2129         ovs_lock();
2130         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2131         err = PTR_ERR(vport);
2132         if (IS_ERR(vport))
2133                 goto exit_unlock_free;
2134
2135         if (vport->port_no == OVSP_LOCAL) {
2136                 err = -EINVAL;
2137                 goto exit_unlock_free;
2138         }
2139
2140         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2141                                       info->snd_portid, info->snd_seq, 0,
2142                                       OVS_VPORT_CMD_DEL);
2143         BUG_ON(err < 0);
2144
2145         /* the vport deletion may trigger dp headroom update */
2146         dp = vport->dp;
2147         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2148                 must_update_headroom = true;
2149         netdev_reset_rx_headroom(vport->dev);
2150         ovs_dp_detach_port(vport);
2151
2152         if (must_update_headroom)
2153                 update_headroom(dp);
2154         ovs_unlock();
2155
2156         ovs_notify(&dp_vport_genl_family, reply, info);
2157         return 0;
2158
2159 exit_unlock_free:
2160         ovs_unlock();
2161         kfree_skb(reply);
2162         return err;
2163 }
2164
2165 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2166 {
2167         struct nlattr **a = info->attrs;
2168         struct ovs_header *ovs_header = info->userhdr;
2169         struct sk_buff *reply;
2170         struct vport *vport;
2171         int err;
2172
2173         reply = ovs_vport_cmd_alloc_info();
2174         if (!reply)
2175                 return -ENOMEM;
2176
2177         rcu_read_lock();
2178         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2179         err = PTR_ERR(vport);
2180         if (IS_ERR(vport))
2181                 goto exit_unlock_free;
2182         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2183                                       info->snd_portid, info->snd_seq, 0,
2184                                       OVS_VPORT_CMD_NEW);
2185         BUG_ON(err < 0);
2186         rcu_read_unlock();
2187
2188         return genlmsg_reply(reply, info);
2189
2190 exit_unlock_free:
2191         rcu_read_unlock();
2192         kfree_skb(reply);
2193         return err;
2194 }
2195
2196 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2197 {
2198         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2199         struct datapath *dp;
2200         int bucket = cb->args[0], skip = cb->args[1];
2201         int i, j = 0;
2202
2203         rcu_read_lock();
2204         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2205         if (!dp) {
2206                 rcu_read_unlock();
2207                 return -ENODEV;
2208         }
2209         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2210                 struct vport *vport;
2211
2212                 j = 0;
2213                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2214                         if (j >= skip &&
2215                             ovs_vport_cmd_fill_info(vport, skb,
2216                                                     sock_net(skb->sk),
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_U32 },
2242         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2243         [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2244         [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2245 };
2246
2247 static const struct genl_ops dp_vport_genl_ops[] = {
2248         { .cmd = OVS_VPORT_CMD_NEW,
2249           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2250           .policy = vport_policy,
2251           .doit = ovs_vport_cmd_new
2252         },
2253         { .cmd = OVS_VPORT_CMD_DEL,
2254           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2255           .policy = vport_policy,
2256           .doit = ovs_vport_cmd_del
2257         },
2258         { .cmd = OVS_VPORT_CMD_GET,
2259           .flags = 0,               /* OK for unprivileged users. */
2260           .policy = vport_policy,
2261           .doit = ovs_vport_cmd_get,
2262           .dumpit = ovs_vport_cmd_dump
2263         },
2264         { .cmd = OVS_VPORT_CMD_SET,
2265           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2266           .policy = vport_policy,
2267           .doit = ovs_vport_cmd_set,
2268         },
2269 };
2270
2271 struct genl_family dp_vport_genl_family __ro_after_init = {
2272         .hdrsize = sizeof(struct ovs_header),
2273         .name = OVS_VPORT_FAMILY,
2274         .version = OVS_VPORT_VERSION,
2275         .maxattr = OVS_VPORT_ATTR_MAX,
2276         .netnsok = true,
2277         .parallel_ops = true,
2278         .ops = dp_vport_genl_ops,
2279         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2280         .mcgrps = &ovs_dp_vport_multicast_group,
2281         .n_mcgrps = 1,
2282         .module = THIS_MODULE,
2283 };
2284
2285 static struct genl_family * const dp_genl_families[] = {
2286         &dp_datapath_genl_family,
2287         &dp_vport_genl_family,
2288         &dp_flow_genl_family,
2289         &dp_packet_genl_family,
2290         &dp_meter_genl_family,
2291 };
2292
2293 static void dp_unregister_genl(int n_families)
2294 {
2295         int i;
2296
2297         for (i = 0; i < n_families; i++)
2298                 genl_unregister_family(dp_genl_families[i]);
2299 }
2300
2301 static int __init dp_register_genl(void)
2302 {
2303         int err;
2304         int i;
2305
2306         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2307
2308                 err = genl_register_family(dp_genl_families[i]);
2309                 if (err)
2310                         goto error;
2311         }
2312
2313         return 0;
2314
2315 error:
2316         dp_unregister_genl(i);
2317         return err;
2318 }
2319
2320 static int __net_init ovs_init_net(struct net *net)
2321 {
2322         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2323
2324         INIT_LIST_HEAD(&ovs_net->dps);
2325         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2326         ovs_ct_init(net);
2327         return 0;
2328 }
2329
2330 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2331                                             struct list_head *head)
2332 {
2333         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2334         struct datapath *dp;
2335
2336         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2337                 int i;
2338
2339                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2340                         struct vport *vport;
2341
2342                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2343                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2344                                         continue;
2345
2346                                 if (dev_net(vport->dev) == dnet)
2347                                         list_add(&vport->detach_list, head);
2348                         }
2349                 }
2350         }
2351 }
2352
2353 static void __net_exit ovs_exit_net(struct net *dnet)
2354 {
2355         struct datapath *dp, *dp_next;
2356         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2357         struct vport *vport, *vport_next;
2358         struct net *net;
2359         LIST_HEAD(head);
2360
2361         ovs_ct_exit(dnet);
2362         ovs_lock();
2363         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2364                 __dp_destroy(dp);
2365
2366         rtnl_lock();
2367         for_each_net(net)
2368                 list_vports_from_net(net, dnet, &head);
2369         rtnl_unlock();
2370
2371         /* Detach all vports from given namespace. */
2372         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2373                 list_del(&vport->detach_list);
2374                 ovs_dp_detach_port(vport);
2375         }
2376
2377         ovs_unlock();
2378
2379         cancel_work_sync(&ovs_net->dp_notify_work);
2380 }
2381
2382 static struct pernet_operations ovs_net_ops = {
2383         .init = ovs_init_net,
2384         .exit = ovs_exit_net,
2385         .id   = &ovs_net_id,
2386         .size = sizeof(struct ovs_net),
2387 };
2388
2389 static int __init dp_init(void)
2390 {
2391         int err;
2392
2393         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2394
2395         pr_info("Open vSwitch switching datapath\n");
2396
2397         err = action_fifos_init();
2398         if (err)
2399                 goto error;
2400
2401         err = ovs_internal_dev_rtnl_link_register();
2402         if (err)
2403                 goto error_action_fifos_exit;
2404
2405         err = ovs_flow_init();
2406         if (err)
2407                 goto error_unreg_rtnl_link;
2408
2409         err = ovs_vport_init();
2410         if (err)
2411                 goto error_flow_exit;
2412
2413         err = register_pernet_device(&ovs_net_ops);
2414         if (err)
2415                 goto error_vport_exit;
2416
2417         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2418         if (err)
2419                 goto error_netns_exit;
2420
2421         err = ovs_netdev_init();
2422         if (err)
2423                 goto error_unreg_notifier;
2424
2425         err = dp_register_genl();
2426         if (err < 0)
2427                 goto error_unreg_netdev;
2428
2429         return 0;
2430
2431 error_unreg_netdev:
2432         ovs_netdev_exit();
2433 error_unreg_notifier:
2434         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2435 error_netns_exit:
2436         unregister_pernet_device(&ovs_net_ops);
2437 error_vport_exit:
2438         ovs_vport_exit();
2439 error_flow_exit:
2440         ovs_flow_exit();
2441 error_unreg_rtnl_link:
2442         ovs_internal_dev_rtnl_link_unregister();
2443 error_action_fifos_exit:
2444         action_fifos_exit();
2445 error:
2446         return err;
2447 }
2448
2449 static void dp_cleanup(void)
2450 {
2451         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2452         ovs_netdev_exit();
2453         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2454         unregister_pernet_device(&ovs_net_ops);
2455         rcu_barrier();
2456         ovs_vport_exit();
2457         ovs_flow_exit();
2458         ovs_internal_dev_rtnl_link_unregister();
2459         action_fifos_exit();
2460 }
2461
2462 module_init(dp_init);
2463 module_exit(dp_cleanup);
2464
2465 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2466 MODULE_LICENSE("GPL");
2467 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2468 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2469 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2470 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2471 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);