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 int 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         if (!upcall) {
 452                 err = -EINVAL;
 453                 goto out;
 454         }
 455         upcall->dp_ifindex = dp_ifindex;
 456
 457         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 458         BUG_ON(err);
 459
 460         if (upcall_info->userdata)
 461                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 462                           nla_len(upcall_info->userdata),
 463                           nla_data(upcall_info->userdata));
 464
 465         if (upcall_info->egress_tun_info) {
 466                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 467                 if (!nla) {
 468                         err = -EMSGSIZE;
 469                         goto out;
 470                 }
 471                 err = ovs_nla_put_tunnel_info(user_skb,
 472                                               upcall_info->egress_tun_info);
 473                 BUG_ON(err);
 474                 nla_nest_end(user_skb, nla);
 475         }
 476
 477         if (upcall_info->actions_len) {
 478                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 479                 if (!nla) {
 480                         err = -EMSGSIZE;
 481                         goto out;
 482                 }
 483                 err = ovs_nla_put_actions(upcall_info->actions,
 484                                           upcall_info->actions_len,
 485                                           user_skb);
 486                 if (!err)
 487                         nla_nest_end(user_skb, nla);
 488                 else
 489                         nla_nest_cancel(user_skb, nla);
 490         }
 491
 492         /* Add OVS_PACKET_ATTR_MRU */
 493         if (upcall_info->mru) {
 494                 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
 495                                 upcall_info->mru)) {
 496                         err = -ENOBUFS;
 497                         goto out;
 498                 }
 499                 pad_packet(dp, user_skb);
 500         }
 501
 502         /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 503         if (cutlen > 0) {
 504                 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
 505                                 skb->len)) {
 506                         err = -ENOBUFS;
 507                         goto out;
 508                 }
 509                 pad_packet(dp, user_skb);
 510         }
 511
 512         /* Only reserve room for attribute header, packet data is added
 513          * in skb_zerocopy() */
 514         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 515                 err = -ENOBUFS;
 516                 goto out;
 517         }
 518         nla->nla_len = nla_attr_size(skb->len - cutlen);
 519
 520         err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 521         if (err)
 522                 goto out;
 523
 524         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 525         pad_packet(dp, user_skb);
 526
 527         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 528
 529         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 530         user_skb = NULL;
 531 out:
 532         if (err)
 533                 skb_tx_error(skb);
 534         kfree_skb(user_skb);
 535         kfree_skb(nskb);
 536         return err;
 537 }
 538
 539 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 540 {
 541         struct ovs_header *ovs_header = info->userhdr;
 542         struct net *net = sock_net(skb->sk);
 543         struct nlattr **a = info->attrs;
 544         struct sw_flow_actions *acts;
 545         struct sk_buff *packet;
 546         struct sw_flow *flow;
 547         struct sw_flow_actions *sf_acts;
 548         struct datapath *dp;
 549         struct vport *input_vport;
 550         u16 mru = 0;
 551         int len;
 552         int err;
 553         bool log = !a[OVS_PACKET_ATTR_PROBE];
 554
 555         err = -EINVAL;
 556         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 557             !a[OVS_PACKET_ATTR_ACTIONS])
 558                 goto err;
 559
 560         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 561         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 562         err = -ENOMEM;
 563         if (!packet)
 564                 goto err;
 565         skb_reserve(packet, NET_IP_ALIGN);
 566
 567         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 568
 569         /* Set packet's mru */
 570         if (a[OVS_PACKET_ATTR_MRU]) {
 571                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 572                 packet->ignore_df = 1;
 573         }
 574         OVS_CB(packet)->mru = mru;
 575
 576         /* Build an sw_flow for sending this packet. */
 577         flow = ovs_flow_alloc();
 578         err = PTR_ERR(flow);
 579         if (IS_ERR(flow))
 580                 goto err_kfree_skb;
 581
 582         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 583                                              packet, &flow->key, log);
 584         if (err)
 585                 goto err_flow_free;
 586
 587         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 588                                    &flow->key, &acts, log);
 589         if (err)
 590                 goto err_flow_free;
 591
 592         rcu_assign_pointer(flow->sf_acts, acts);
 593         packet->priority = flow->key.phy.priority;
 594         packet->mark = flow->key.phy.skb_mark;
 595
 596         rcu_read_lock();
 597         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 598         err = -ENODEV;
 599         if (!dp)
 600                 goto err_unlock;
 601
 602         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 603         if (!input_vport)
 604                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 605
 606         if (!input_vport)
 607                 goto err_unlock;
 608
 609         packet->dev = input_vport->dev;
 610         OVS_CB(packet)->input_vport = input_vport;
 611         sf_acts = rcu_dereference(flow->sf_acts);
 612
 613         local_bh_disable();
 614         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 615         local_bh_enable();
 616         rcu_read_unlock();
 617
 618         ovs_flow_free(flow, false);
 619         return err;
 620
 621 err_unlock:
 622         rcu_read_unlock();
 623 err_flow_free:
 624         ovs_flow_free(flow, false);
 625 err_kfree_skb:
 626         kfree_skb(packet);
 627 err:
 628         return err;
 629 }
 630
 631 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 632         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 633         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 634         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 635         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 636         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 637 };
 638
 639 static const struct genl_ops dp_packet_genl_ops[] = {
 640         { .cmd = OVS_PACKET_CMD_EXECUTE,
 641           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 642           .policy = packet_policy,
 643           .doit = ovs_packet_cmd_execute
 644         }
 645 };
 646
 647 static struct genl_family dp_packet_genl_family __ro_after_init = {
 648         .hdrsize = sizeof(struct ovs_header),
 649         .name = OVS_PACKET_FAMILY,
 650         .version = OVS_PACKET_VERSION,
 651         .maxattr = OVS_PACKET_ATTR_MAX,
 652         .netnsok = true,
 653         .parallel_ops = true,
 654         .ops = dp_packet_genl_ops,
 655         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 656         .module = THIS_MODULE,
 657 };
 658
 659 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 660                          struct ovs_dp_megaflow_stats *mega_stats)
 661 {
 662         int i;
 663
 664         memset(mega_stats, 0, sizeof(*mega_stats));
 665
 666         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 667         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 668
 669         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 670
 671         for_each_possible_cpu(i) {
 672                 const struct dp_stats_percpu *percpu_stats;
 673                 struct dp_stats_percpu local_stats;
 674                 unsigned int start;
 675
 676                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 677
 678                 do {
 679                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 680                         local_stats = *percpu_stats;
 681                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 682
 683                 stats->n_hit += local_stats.n_hit;
 684                 stats->n_missed += local_stats.n_missed;
 685                 stats->n_lost += local_stats.n_lost;
 686                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 687         }
 688 }
 689
 690 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 691 {
 692         return ovs_identifier_is_ufid(sfid) &&
 693                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 694 }
 695
 696 static bool should_fill_mask(uint32_t ufid_flags)
 697 {
 698         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 699 }
 700
 701 static bool should_fill_actions(uint32_t ufid_flags)
 702 {
 703         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 704 }
 705
 706 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 707                                     const struct sw_flow_id *sfid,
 708                                     uint32_t ufid_flags)
 709 {
 710         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 711
 712         /* OVS_FLOW_ATTR_UFID */
 713         if (sfid && ovs_identifier_is_ufid(sfid))
 714                 len += nla_total_size(sfid->ufid_len);
 715
 716         /* OVS_FLOW_ATTR_KEY */
 717         if (!sfid || should_fill_key(sfid, ufid_flags))
 718                 len += nla_total_size(ovs_key_attr_size());
 719
 720         /* OVS_FLOW_ATTR_MASK */
 721         if (should_fill_mask(ufid_flags))
 722                 len += nla_total_size(ovs_key_attr_size());
 723
 724         /* OVS_FLOW_ATTR_ACTIONS */
 725         if (should_fill_actions(ufid_flags))
 726                 len += nla_total_size(acts->orig_len);
 727
 728         return len
 729                 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 730                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 731                 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 732 }
 733
 734 /* Called with ovs_mutex or RCU read lock. */
 735 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 736                                    struct sk_buff *skb)
 737 {
 738         struct ovs_flow_stats stats;
 739         __be16 tcp_flags;
 740         unsigned long used;
 741
 742         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 743
 744         if (used &&
 745             nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 746                               OVS_FLOW_ATTR_PAD))
 747                 return -EMSGSIZE;
 748
 749         if (stats.n_packets &&
 750             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 751                           sizeof(struct ovs_flow_stats), &stats,
 752                           OVS_FLOW_ATTR_PAD))
 753                 return -EMSGSIZE;
 754
 755         if ((u8)ntohs(tcp_flags) &&
 756              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 757                 return -EMSGSIZE;
 758
 759         return 0;
 760 }
 761
 762 /* Called with ovs_mutex or RCU read lock. */
 763 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 764                                      struct sk_buff *skb, int skb_orig_len)
 765 {
 766         struct nlattr *start;
 767         int err;
 768
 769         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 770          * this is the first flow to be dumped into 'skb'.  This is unusual for
 771          * Netlink but individual action lists can be longer than
 772          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 773          * The userspace caller can always fetch the actions separately if it
 774          * really wants them.  (Most userspace callers in fact don't care.)
 775          *
 776          * This can only fail for dump operations because the skb is always
 777          * properly sized for single flows.
 778          */
 779         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 780         if (start) {
 781                 const struct sw_flow_actions *sf_acts;
 782
 783                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 784                 err = ovs_nla_put_actions(sf_acts->actions,
 785                                           sf_acts->actions_len, skb);
 786
 787                 if (!err)
 788                         nla_nest_end(skb, start);
 789                 else {
 790                         if (skb_orig_len)
 791                                 return err;
 792
 793                         nla_nest_cancel(skb, start);
 794                 }
 795         } else if (skb_orig_len) {
 796                 return -EMSGSIZE;
 797         }
 798
 799         return 0;
 800 }
 801
 802 /* Called with ovs_mutex or RCU read lock. */
 803 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 804                                   struct sk_buff *skb, u32 portid,
 805                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 806 {
 807         const int skb_orig_len = skb->len;
 808         struct ovs_header *ovs_header;
 809         int err;
 810
 811         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 812                                  flags, cmd);
 813         if (!ovs_header)
 814                 return -EMSGSIZE;
 815
 816         ovs_header->dp_ifindex = dp_ifindex;
 817
 818         err = ovs_nla_put_identifier(flow, skb);
 819         if (err)
 820                 goto error;
 821
 822         if (should_fill_key(&flow->id, ufid_flags)) {
 823                 err = ovs_nla_put_masked_key(flow, skb);
 824                 if (err)
 825                         goto error;
 826         }
 827
 828         if (should_fill_mask(ufid_flags)) {
 829                 err = ovs_nla_put_mask(flow, skb);
 830                 if (err)
 831                         goto error;
 832         }
 833
 834         err = ovs_flow_cmd_fill_stats(flow, skb);
 835         if (err)
 836                 goto error;
 837
 838         if (should_fill_actions(ufid_flags)) {
 839                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 840                 if (err)
 841                         goto error;
 842         }
 843
 844         genlmsg_end(skb, ovs_header);
 845         return 0;
 846
 847 error:
 848         genlmsg_cancel(skb, ovs_header);
 849         return err;
 850 }
 851
 852 /* May not be called with RCU read lock. */
 853 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 854                                                const struct sw_flow_id *sfid,
 855                                                struct genl_info *info,
 856                                                bool always,
 857                                                uint32_t ufid_flags)
 858 {
 859         struct sk_buff *skb;
 860         size_t len;
 861
 862         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 863                 return NULL;
 864
 865         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 866         skb = genlmsg_new(len, GFP_KERNEL);
 867         if (!skb)
 868                 return ERR_PTR(-ENOMEM);
 869
 870         return skb;
 871 }
 872
 873 /* Called with ovs_mutex. */
 874 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 875                                                int dp_ifindex,
 876                                                struct genl_info *info, u8 cmd,
 877                                                bool always, u32 ufid_flags)
 878 {
 879         struct sk_buff *skb;
 880         int retval;
 881
 882         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 883                                       &flow->id, info, always, ufid_flags);
 884         if (IS_ERR_OR_NULL(skb))
 885                 return skb;
 886
 887         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 888                                         info->snd_portid, info->snd_seq, 0,
 889                                         cmd, ufid_flags);
 890         BUG_ON(retval < 0);
 891         return skb;
 892 }
 893
 894 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 895 {
 896         struct net *net = sock_net(skb->sk);
 897         struct nlattr **a = info->attrs;
 898         struct ovs_header *ovs_header = info->userhdr;
 899         struct sw_flow *flow = NULL, *new_flow;
 900         struct sw_flow_mask mask;
 901         struct sk_buff *reply;
 902         struct datapath *dp;
 903         struct sw_flow_actions *acts;
 904         struct sw_flow_match match;
 905         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 906         int error;
 907         bool log = !a[OVS_FLOW_ATTR_PROBE];
 908
 909         /* Must have key and actions. */
 910         error = -EINVAL;
 911         if (!a[OVS_FLOW_ATTR_KEY]) {
 912                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 913                 goto error;
 914         }
 915         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 916                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 917                 goto error;
 918         }
 919
 920         /* Most of the time we need to allocate a new flow, do it before
 921          * locking.
 922          */
 923         new_flow = ovs_flow_alloc();
 924         if (IS_ERR(new_flow)) {
 925                 error = PTR_ERR(new_flow);
 926                 goto error;
 927         }
 928
 929         /* Extract key. */
 930         ovs_match_init(&match, &new_flow->key, false, &mask);
 931         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 932                                   a[OVS_FLOW_ATTR_MASK], log);
 933         if (error)
 934                 goto err_kfree_flow;
 935
 936         /* Extract flow identifier. */
 937         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 938                                        &new_flow->key, log);
 939         if (error)
 940                 goto err_kfree_flow;
 941
 942         /* unmasked key is needed to match when ufid is not used. */
 943         if (ovs_identifier_is_key(&new_flow->id))
 944                 match.key = new_flow->id.unmasked_key;
 945
 946         ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
 947
 948         /* Validate actions. */
 949         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 950                                      &new_flow->key, &acts, log);
 951         if (error) {
 952                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 953                 goto err_kfree_flow;
 954         }
 955
 956         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 957                                         ufid_flags);
 958         if (IS_ERR(reply)) {
 959                 error = PTR_ERR(reply);
 960                 goto err_kfree_acts;
 961         }
 962
 963         ovs_lock();
 964         dp = get_dp(net, ovs_header->dp_ifindex);
 965         if (unlikely(!dp)) {
 966                 error = -ENODEV;
 967                 goto err_unlock_ovs;
 968         }
 969
 970         /* Check if this is a duplicate flow */
 971         if (ovs_identifier_is_ufid(&new_flow->id))
 972                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
 973         if (!flow)
 974                 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
 975         if (likely(!flow)) {
 976                 rcu_assign_pointer(new_flow->sf_acts, acts);
 977
 978                 /* Put flow in bucket. */
 979                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 980                 if (unlikely(error)) {
 981                         acts = NULL;
 982                         goto err_unlock_ovs;
 983                 }
 984
 985                 if (unlikely(reply)) {
 986                         error = ovs_flow_cmd_fill_info(new_flow,
 987                                                        ovs_header->dp_ifindex,
 988                                                        reply, info->snd_portid,
 989                                                        info->snd_seq, 0,
 990                                                        OVS_FLOW_CMD_NEW,
 991                                                        ufid_flags);
 992                         BUG_ON(error < 0);
 993                 }
 994                 ovs_unlock();
 995         } else {
 996                 struct sw_flow_actions *old_acts;
 997
 998                 /* Bail out if we're not allowed to modify an existing flow.
 999                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1000                  * because Generic Netlink treats the latter as a dump
1001                  * request.  We also accept NLM_F_EXCL in case that bug ever
1002                  * gets fixed.
1003                  */
1004                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1005                                                          | NLM_F_EXCL))) {
1006                         error = -EEXIST;
1007                         goto err_unlock_ovs;
1008                 }
1009                 /* The flow identifier has to be the same for flow updates.
1010                  * Look for any overlapping flow.
1011                  */
1012                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1013                         if (ovs_identifier_is_key(&flow->id))
1014                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1015                                                                  &match);
1016                         else /* UFID matches but key is different */
1017                                 flow = NULL;
1018                         if (!flow) {
1019                                 error = -ENOENT;
1020                                 goto err_unlock_ovs;
1021                         }
1022                 }
1023                 /* Update actions. */
1024                 old_acts = ovsl_dereference(flow->sf_acts);
1025                 rcu_assign_pointer(flow->sf_acts, acts);
1026
1027                 if (unlikely(reply)) {
1028                         error = ovs_flow_cmd_fill_info(flow,
1029                                                        ovs_header->dp_ifindex,
1030                                                        reply, info->snd_portid,
1031                                                        info->snd_seq, 0,
1032                                                        OVS_FLOW_CMD_NEW,
1033                                                        ufid_flags);
1034                         BUG_ON(error < 0);
1035                 }
1036                 ovs_unlock();
1037
1038                 ovs_nla_free_flow_actions_rcu(old_acts);
1039                 ovs_flow_free(new_flow, false);
1040         }
1041
1042         if (reply)
1043                 ovs_notify(&dp_flow_genl_family, reply, info);
1044         return 0;
1045
1046 err_unlock_ovs:
1047         ovs_unlock();
1048         kfree_skb(reply);
1049 err_kfree_acts:
1050         ovs_nla_free_flow_actions(acts);
1051 err_kfree_flow:
1052         ovs_flow_free(new_flow, false);
1053 error:
1054         return error;
1055 }
1056
1057 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1058 static struct sw_flow_actions *get_flow_actions(struct net *net,
1059                                                 const struct nlattr *a,
1060                                                 const struct sw_flow_key *key,
1061                                                 const struct sw_flow_mask *mask,
1062                                                 bool log)
1063 {
1064         struct sw_flow_actions *acts;
1065         struct sw_flow_key masked_key;
1066         int error;
1067
1068         ovs_flow_mask_key(&masked_key, key, true, mask);
1069         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1070         if (error) {
1071                 OVS_NLERR(log,
1072                           "Actions may not be safe on all matching packets");
1073                 return ERR_PTR(error);
1074         }
1075
1076         return acts;
1077 }
1078
1079 /* Factor out match-init and action-copy to avoid
1080  * "Wframe-larger-than=1024" warning. Because mask is only
1081  * used to get actions, we new a function to save some
1082  * stack space.
1083  *
1084  * If there are not key and action attrs, we return 0
1085  * directly. In the case, the caller will also not use the
1086  * match as before. If there is action attr, we try to get
1087  * actions and save them to *acts. Before returning from
1088  * the function, we reset the match->mask pointer. Because
1089  * we should not to return match object with dangling reference
1090  * to mask.
1091  * */
1092 static int ovs_nla_init_match_and_action(struct net *net,
1093                                          struct sw_flow_match *match,
1094                                          struct sw_flow_key *key,
1095                                          struct nlattr **a,
1096                                          struct sw_flow_actions **acts,
1097                                          bool log)
1098 {
1099         struct sw_flow_mask mask;
1100         int error = 0;
1101
1102         if (a[OVS_FLOW_ATTR_KEY]) {
1103                 ovs_match_init(match, key, true, &mask);
1104                 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1105                                           a[OVS_FLOW_ATTR_MASK], log);
1106                 if (error)
1107                         goto error;
1108         }
1109
1110         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1111                 if (!a[OVS_FLOW_ATTR_KEY]) {
1112                         OVS_NLERR(log,
1113                                   "Flow key attribute not present in set flow.");
1114                         error = -EINVAL;
1115                         goto error;
1116                 }
1117
1118                 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1119                                          &mask, log);
1120                 if (IS_ERR(*acts)) {
1121                         error = PTR_ERR(*acts);
1122                         goto error;
1123                 }
1124         }
1125
1126         /* On success, error is 0. */
1127 error:
1128         match->mask = NULL;
1129         return error;
1130 }
1131
1132 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1133 {
1134         struct net *net = sock_net(skb->sk);
1135         struct nlattr **a = info->attrs;
1136         struct ovs_header *ovs_header = info->userhdr;
1137         struct sw_flow_key key;
1138         struct sw_flow *flow;
1139         struct sk_buff *reply = NULL;
1140         struct datapath *dp;
1141         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1142         struct sw_flow_match match;
1143         struct sw_flow_id sfid;
1144         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1145         int error = 0;
1146         bool log = !a[OVS_FLOW_ATTR_PROBE];
1147         bool ufid_present;
1148
1149         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1150         if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1151                 OVS_NLERR(log,
1152                           "Flow set message rejected, Key attribute missing.");
1153                 return -EINVAL;
1154         }
1155
1156         error = ovs_nla_init_match_and_action(net, &match, &key, a,
1157                                               &acts, log);
1158         if (error)
1159                 goto error;
1160
1161         if (acts) {
1162                 /* Can allocate before locking if have acts. */
1163                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1164                                                 ufid_flags);
1165                 if (IS_ERR(reply)) {
1166                         error = PTR_ERR(reply);
1167                         goto err_kfree_acts;
1168                 }
1169         }
1170
1171         ovs_lock();
1172         dp = get_dp(net, ovs_header->dp_ifindex);
1173         if (unlikely(!dp)) {
1174                 error = -ENODEV;
1175                 goto err_unlock_ovs;
1176         }
1177         /* Check that the flow exists. */
1178         if (ufid_present)
1179                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1180         else
1181                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1182         if (unlikely(!flow)) {
1183                 error = -ENOENT;
1184                 goto err_unlock_ovs;
1185         }
1186
1187         /* Update actions, if present. */
1188         if (likely(acts)) {
1189                 old_acts = ovsl_dereference(flow->sf_acts);
1190                 rcu_assign_pointer(flow->sf_acts, acts);
1191
1192                 if (unlikely(reply)) {
1193                         error = ovs_flow_cmd_fill_info(flow,
1194                                                        ovs_header->dp_ifindex,
1195                                                        reply, info->snd_portid,
1196                                                        info->snd_seq, 0,
1197                                                        OVS_FLOW_CMD_SET,
1198                                                        ufid_flags);
1199                         BUG_ON(error < 0);
1200                 }
1201         } else {
1202                 /* Could not alloc without acts before locking. */
1203                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1204                                                 info, OVS_FLOW_CMD_SET, false,
1205                                                 ufid_flags);
1206
1207                 if (IS_ERR(reply)) {
1208                         error = PTR_ERR(reply);
1209                         goto err_unlock_ovs;
1210                 }
1211         }
1212
1213         /* Clear stats. */
1214         if (a[OVS_FLOW_ATTR_CLEAR])
1215                 ovs_flow_stats_clear(flow);
1216         ovs_unlock();
1217
1218         if (reply)
1219                 ovs_notify(&dp_flow_genl_family, reply, info);
1220         if (old_acts)
1221                 ovs_nla_free_flow_actions_rcu(old_acts);
1222
1223         return 0;
1224
1225 err_unlock_ovs:
1226         ovs_unlock();
1227         kfree_skb(reply);
1228 err_kfree_acts:
1229         ovs_nla_free_flow_actions(acts);
1230 error:
1231         return error;
1232 }
1233
1234 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1235 {
1236         struct nlattr **a = info->attrs;
1237         struct ovs_header *ovs_header = info->userhdr;
1238         struct net *net = sock_net(skb->sk);
1239         struct sw_flow_key key;
1240         struct sk_buff *reply;
1241         struct sw_flow *flow;
1242         struct datapath *dp;
1243         struct sw_flow_match match;
1244         struct sw_flow_id ufid;
1245         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1246         int err = 0;
1247         bool log = !a[OVS_FLOW_ATTR_PROBE];
1248         bool ufid_present;
1249
1250         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1251         if (a[OVS_FLOW_ATTR_KEY]) {
1252                 ovs_match_init(&match, &key, true, NULL);
1253                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1254                                         log);
1255         } else if (!ufid_present) {
1256                 OVS_NLERR(log,
1257                           "Flow get message rejected, Key attribute missing.");
1258                 err = -EINVAL;
1259         }
1260         if (err)
1261                 return err;
1262
1263         ovs_lock();
1264         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1265         if (!dp) {
1266                 err = -ENODEV;
1267                 goto unlock;
1268         }
1269
1270         if (ufid_present)
1271                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1272         else
1273                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1274         if (!flow) {
1275                 err = -ENOENT;
1276                 goto unlock;
1277         }
1278
1279         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1280                                         OVS_FLOW_CMD_GET, true, ufid_flags);
1281         if (IS_ERR(reply)) {
1282                 err = PTR_ERR(reply);
1283                 goto unlock;
1284         }
1285
1286         ovs_unlock();
1287         return genlmsg_reply(reply, info);
1288 unlock:
1289         ovs_unlock();
1290         return err;
1291 }
1292
1293 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1294 {
1295         struct nlattr **a = info->attrs;
1296         struct ovs_header *ovs_header = info->userhdr;
1297         struct net *net = sock_net(skb->sk);
1298         struct sw_flow_key key;
1299         struct sk_buff *reply;
1300         struct sw_flow *flow = NULL;
1301         struct datapath *dp;
1302         struct sw_flow_match match;
1303         struct sw_flow_id ufid;
1304         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1305         int err;
1306         bool log = !a[OVS_FLOW_ATTR_PROBE];
1307         bool ufid_present;
1308
1309         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1310         if (a[OVS_FLOW_ATTR_KEY]) {
1311                 ovs_match_init(&match, &key, true, NULL);
1312                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1313                                         NULL, log);
1314                 if (unlikely(err))
1315                         return err;
1316         }
1317
1318         ovs_lock();
1319         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1320         if (unlikely(!dp)) {
1321                 err = -ENODEV;
1322                 goto unlock;
1323         }
1324
1325         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1326                 err = ovs_flow_tbl_flush(&dp->table);
1327                 goto unlock;
1328         }
1329
1330         if (ufid_present)
1331                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1332         else
1333                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1334         if (unlikely(!flow)) {
1335                 err = -ENOENT;
1336                 goto unlock;
1337         }
1338
1339         ovs_flow_tbl_remove(&dp->table, flow);
1340         ovs_unlock();
1341
1342         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1343                                         &flow->id, info, false, ufid_flags);
1344         if (likely(reply)) {
1345                 if (likely(!IS_ERR(reply))) {
1346                         rcu_read_lock();        /*To keep RCU checker happy. */
1347                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1348                                                      reply, info->snd_portid,
1349                                                      info->snd_seq, 0,
1350                                                      OVS_FLOW_CMD_DEL,
1351                                                      ufid_flags);
1352                         rcu_read_unlock();
1353                         BUG_ON(err < 0);
1354
1355                         ovs_notify(&dp_flow_genl_family, reply, info);
1356                 } else {
1357                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1358                 }
1359         }
1360
1361         ovs_flow_free(flow, true);
1362         return 0;
1363 unlock:
1364         ovs_unlock();
1365         return err;
1366 }
1367
1368 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1369 {
1370         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1371         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1372         struct table_instance *ti;
1373         struct datapath *dp;
1374         u32 ufid_flags;
1375         int err;
1376
1377         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1378                             OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1379         if (err)
1380                 return err;
1381         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1382
1383         rcu_read_lock();
1384         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1385         if (!dp) {
1386                 rcu_read_unlock();
1387                 return -ENODEV;
1388         }
1389
1390         ti = rcu_dereference(dp->table.ti);
1391         for (;;) {
1392                 struct sw_flow *flow;
1393                 u32 bucket, obj;
1394
1395                 bucket = cb->args[0];
1396                 obj = cb->args[1];
1397                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1398                 if (!flow)
1399                         break;
1400
1401                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1402                                            NETLINK_CB(cb->skb).portid,
1403                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1404                                            OVS_FLOW_CMD_GET, ufid_flags) < 0)
1405                         break;
1406
1407                 cb->args[0] = bucket;
1408                 cb->args[1] = obj;
1409         }
1410         rcu_read_unlock();
1411         return skb->len;
1412 }
1413
1414 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1415         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1416         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1417         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1418         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1419         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1420         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1421         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1422 };
1423
1424 static const struct genl_ops dp_flow_genl_ops[] = {
1425         { .cmd = OVS_FLOW_CMD_NEW,
1426           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1427           .policy = flow_policy,
1428           .doit = ovs_flow_cmd_new
1429         },
1430         { .cmd = OVS_FLOW_CMD_DEL,
1431           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1432           .policy = flow_policy,
1433           .doit = ovs_flow_cmd_del
1434         },
1435         { .cmd = OVS_FLOW_CMD_GET,
1436           .flags = 0,               /* OK for unprivileged users. */
1437           .policy = flow_policy,
1438           .doit = ovs_flow_cmd_get,
1439           .dumpit = ovs_flow_cmd_dump
1440         },
1441         { .cmd = OVS_FLOW_CMD_SET,
1442           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1443           .policy = flow_policy,
1444           .doit = ovs_flow_cmd_set,
1445         },
1446 };
1447
1448 static struct genl_family dp_flow_genl_family __ro_after_init = {
1449         .hdrsize = sizeof(struct ovs_header),
1450         .name = OVS_FLOW_FAMILY,
1451         .version = OVS_FLOW_VERSION,
1452         .maxattr = OVS_FLOW_ATTR_MAX,
1453         .netnsok = true,
1454         .parallel_ops = true,
1455         .ops = dp_flow_genl_ops,
1456         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1457         .mcgrps = &ovs_dp_flow_multicast_group,
1458         .n_mcgrps = 1,
1459         .module = THIS_MODULE,
1460 };
1461
1462 static size_t ovs_dp_cmd_msg_size(void)
1463 {
1464         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1465
1466         msgsize += nla_total_size(IFNAMSIZ);
1467         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1468         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1469         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1470
1471         return msgsize;
1472 }
1473
1474 /* Called with ovs_mutex. */
1475 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1476                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1477 {
1478         struct ovs_header *ovs_header;
1479         struct ovs_dp_stats dp_stats;
1480         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1481         int err;
1482
1483         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1484                                    flags, cmd);
1485         if (!ovs_header)
1486                 goto error;
1487
1488         ovs_header->dp_ifindex = get_dpifindex(dp);
1489
1490         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1491         if (err)
1492                 goto nla_put_failure;
1493
1494         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1495         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1496                           &dp_stats, OVS_DP_ATTR_PAD))
1497                 goto nla_put_failure;
1498
1499         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1500                           sizeof(struct ovs_dp_megaflow_stats),
1501                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1502                 goto nla_put_failure;
1503
1504         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1505                 goto nla_put_failure;
1506
1507         genlmsg_end(skb, ovs_header);
1508         return 0;
1509
1510 nla_put_failure:
1511         genlmsg_cancel(skb, ovs_header);
1512 error:
1513         return -EMSGSIZE;
1514 }
1515
1516 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1517 {
1518         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1519 }
1520
1521 /* Called with rcu_read_lock or ovs_mutex. */
1522 static struct datapath *lookup_datapath(struct net *net,
1523                                         const struct ovs_header *ovs_header,
1524                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1525 {
1526         struct datapath *dp;
1527
1528         if (!a[OVS_DP_ATTR_NAME])
1529                 dp = get_dp(net, ovs_header->dp_ifindex);
1530         else {
1531                 struct vport *vport;
1532
1533                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1534                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1535         }
1536         return dp ? dp : ERR_PTR(-ENODEV);
1537 }
1538
1539 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1540 {
1541         struct datapath *dp;
1542
1543         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1544         if (IS_ERR(dp))
1545                 return;
1546
1547         WARN(dp->user_features, "Dropping previously announced user features\n");
1548         dp->user_features = 0;
1549 }
1550
1551 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1552 {
1553         if (a[OVS_DP_ATTR_USER_FEATURES])
1554                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1555 }
1556
1557 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1558 {
1559         struct nlattr **a = info->attrs;
1560         struct vport_parms parms;
1561         struct sk_buff *reply;
1562         struct datapath *dp;
1563         struct vport *vport;
1564         struct ovs_net *ovs_net;
1565         int err, i;
1566
1567         err = -EINVAL;
1568         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1569                 goto err;
1570
1571         reply = ovs_dp_cmd_alloc_info();
1572         if (!reply)
1573                 return -ENOMEM;
1574
1575         err = -ENOMEM;
1576         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1577         if (dp == NULL)
1578                 goto err_free_reply;
1579
1580         ovs_dp_set_net(dp, sock_net(skb->sk));
1581
1582         /* Allocate table. */
1583         err = ovs_flow_tbl_init(&dp->table);
1584         if (err)
1585                 goto err_free_dp;
1586
1587         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1588         if (!dp->stats_percpu) {
1589                 err = -ENOMEM;
1590                 goto err_destroy_table;
1591         }
1592
1593         dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1594                                   sizeof(struct hlist_head),
1595                                   GFP_KERNEL);
1596         if (!dp->ports) {
1597                 err = -ENOMEM;
1598                 goto err_destroy_percpu;
1599         }
1600
1601         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1602                 INIT_HLIST_HEAD(&dp->ports[i]);
1603
1604         err = ovs_meters_init(dp);
1605         if (err)
1606                 goto err_destroy_ports_array;
1607
1608         /* Set up our datapath device. */
1609         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1610         parms.type = OVS_VPORT_TYPE_INTERNAL;
1611         parms.options = NULL;
1612         parms.dp = dp;
1613         parms.port_no = OVSP_LOCAL;
1614         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1615
1616         ovs_dp_change(dp, a);
1617
1618         /* So far only local changes have been made, now need the lock. */
1619         ovs_lock();
1620
1621         vport = new_vport(&parms);
1622         if (IS_ERR(vport)) {
1623                 err = PTR_ERR(vport);
1624                 if (err == -EBUSY)
1625                         err = -EEXIST;
1626
1627                 if (err == -EEXIST) {
1628                         /* An outdated user space instance that does not understand
1629                          * the concept of user_features has attempted to create a new
1630                          * datapath and is likely to reuse it. Drop all user features.
1631                          */
1632                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1633                                 ovs_dp_reset_user_features(skb, info);
1634                 }
1635
1636                 goto err_destroy_meters;
1637         }
1638
1639         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1640                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1641         BUG_ON(err < 0);
1642
1643         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1644         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1645
1646         ovs_unlock();
1647
1648         ovs_notify(&dp_datapath_genl_family, reply, info);
1649         return 0;
1650
1651 err_destroy_meters:
1652         ovs_unlock();
1653         ovs_meters_exit(dp);
1654 err_destroy_ports_array:
1655         kfree(dp->ports);
1656 err_destroy_percpu:
1657         free_percpu(dp->stats_percpu);
1658 err_destroy_table:
1659         ovs_flow_tbl_destroy(&dp->table);
1660 err_free_dp:
1661         kfree(dp);
1662 err_free_reply:
1663         kfree_skb(reply);
1664 err:
1665         return err;
1666 }
1667
1668 /* Called with ovs_mutex. */
1669 static void __dp_destroy(struct datapath *dp)
1670 {
1671         int i;
1672
1673         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1674                 struct vport *vport;
1675                 struct hlist_node *n;
1676
1677                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1678                         if (vport->port_no != OVSP_LOCAL)
1679                                 ovs_dp_detach_port(vport);
1680         }
1681
1682         list_del_rcu(&dp->list_node);
1683
1684         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1685          * all ports in datapath are destroyed first before freeing datapath.
1686          */
1687         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1688
1689         /* RCU destroy the flow table */
1690         call_rcu(&dp->rcu, destroy_dp_rcu);
1691 }
1692
1693 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1694 {
1695         struct sk_buff *reply;
1696         struct datapath *dp;
1697         int err;
1698
1699         reply = ovs_dp_cmd_alloc_info();
1700         if (!reply)
1701                 return -ENOMEM;
1702
1703         ovs_lock();
1704         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1705         err = PTR_ERR(dp);
1706         if (IS_ERR(dp))
1707                 goto err_unlock_free;
1708
1709         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1710                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1711         BUG_ON(err < 0);
1712
1713         __dp_destroy(dp);
1714         ovs_unlock();
1715
1716         ovs_notify(&dp_datapath_genl_family, reply, info);
1717
1718         return 0;
1719
1720 err_unlock_free:
1721         ovs_unlock();
1722         kfree_skb(reply);
1723         return err;
1724 }
1725
1726 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1727 {
1728         struct sk_buff *reply;
1729         struct datapath *dp;
1730         int err;
1731
1732         reply = ovs_dp_cmd_alloc_info();
1733         if (!reply)
1734                 return -ENOMEM;
1735
1736         ovs_lock();
1737         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1738         err = PTR_ERR(dp);
1739         if (IS_ERR(dp))
1740                 goto err_unlock_free;
1741
1742         ovs_dp_change(dp, info->attrs);
1743
1744         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1745                                    info->snd_seq, 0, OVS_DP_CMD_SET);
1746         BUG_ON(err < 0);
1747
1748         ovs_unlock();
1749         ovs_notify(&dp_datapath_genl_family, reply, info);
1750
1751         return 0;
1752
1753 err_unlock_free:
1754         ovs_unlock();
1755         kfree_skb(reply);
1756         return err;
1757 }
1758
1759 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1760 {
1761         struct sk_buff *reply;
1762         struct datapath *dp;
1763         int err;
1764
1765         reply = ovs_dp_cmd_alloc_info();
1766         if (!reply)
1767                 return -ENOMEM;
1768
1769         ovs_lock();
1770         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1771         if (IS_ERR(dp)) {
1772                 err = PTR_ERR(dp);
1773                 goto err_unlock_free;
1774         }
1775         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1776                                    info->snd_seq, 0, OVS_DP_CMD_GET);
1777         BUG_ON(err < 0);
1778         ovs_unlock();
1779
1780         return genlmsg_reply(reply, info);
1781
1782 err_unlock_free:
1783         ovs_unlock();
1784         kfree_skb(reply);
1785         return err;
1786 }
1787
1788 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1789 {
1790         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1791         struct datapath *dp;
1792         int skip = cb->args[0];
1793         int i = 0;
1794
1795         ovs_lock();
1796         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1797                 if (i >= skip &&
1798                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1799                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1800                                          OVS_DP_CMD_GET) < 0)
1801                         break;
1802                 i++;
1803         }
1804         ovs_unlock();
1805
1806         cb->args[0] = i;
1807
1808         return skb->len;
1809 }
1810
1811 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1812         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1813         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1814         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1815 };
1816
1817 static const struct genl_ops dp_datapath_genl_ops[] = {
1818         { .cmd = OVS_DP_CMD_NEW,
1819           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1820           .policy = datapath_policy,
1821           .doit = ovs_dp_cmd_new
1822         },
1823         { .cmd = OVS_DP_CMD_DEL,
1824           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1825           .policy = datapath_policy,
1826           .doit = ovs_dp_cmd_del
1827         },
1828         { .cmd = OVS_DP_CMD_GET,
1829           .flags = 0,               /* OK for unprivileged users. */
1830           .policy = datapath_policy,
1831           .doit = ovs_dp_cmd_get,
1832           .dumpit = ovs_dp_cmd_dump
1833         },
1834         { .cmd = OVS_DP_CMD_SET,
1835           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1836           .policy = datapath_policy,
1837           .doit = ovs_dp_cmd_set,
1838         },
1839 };
1840
1841 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1842         .hdrsize = sizeof(struct ovs_header),
1843         .name = OVS_DATAPATH_FAMILY,
1844         .version = OVS_DATAPATH_VERSION,
1845         .maxattr = OVS_DP_ATTR_MAX,
1846         .netnsok = true,
1847         .parallel_ops = true,
1848         .ops = dp_datapath_genl_ops,
1849         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1850         .mcgrps = &ovs_dp_datapath_multicast_group,
1851         .n_mcgrps = 1,
1852         .module = THIS_MODULE,
1853 };
1854
1855 /* Called with ovs_mutex or RCU read lock. */
1856 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1857                                    struct net *net, u32 portid, u32 seq,
1858                                    u32 flags, u8 cmd)
1859 {
1860         struct ovs_header *ovs_header;
1861         struct ovs_vport_stats vport_stats;
1862         int err;
1863
1864         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1865                                  flags, cmd);
1866         if (!ovs_header)
1867                 return -EMSGSIZE;
1868
1869         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1870
1871         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1872             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1873             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1874                            ovs_vport_name(vport)) ||
1875             nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1876                 goto nla_put_failure;
1877
1878         if (!net_eq(net, dev_net(vport->dev))) {
1879                 int id = peernet2id_alloc(net, dev_net(vport->dev));
1880
1881                 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1882                         goto nla_put_failure;
1883         }
1884
1885         ovs_vport_get_stats(vport, &vport_stats);
1886         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1887                           sizeof(struct ovs_vport_stats), &vport_stats,
1888                           OVS_VPORT_ATTR_PAD))
1889                 goto nla_put_failure;
1890
1891         if (ovs_vport_get_upcall_portids(vport, skb))
1892                 goto nla_put_failure;
1893
1894         err = ovs_vport_get_options(vport, skb);
1895         if (err == -EMSGSIZE)
1896                 goto error;
1897
1898         genlmsg_end(skb, ovs_header);
1899         return 0;
1900
1901 nla_put_failure:
1902         err = -EMSGSIZE;
1903 error:
1904         genlmsg_cancel(skb, ovs_header);
1905         return err;
1906 }
1907
1908 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1909 {
1910         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1911 }
1912
1913 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1914 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1915                                          u32 portid, u32 seq, u8 cmd)
1916 {
1917         struct sk_buff *skb;
1918         int retval;
1919
1920         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1921         if (!skb)
1922                 return ERR_PTR(-ENOMEM);
1923
1924         retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd);
1925         BUG_ON(retval < 0);
1926
1927         return skb;
1928 }
1929
1930 /* Called with ovs_mutex or RCU read lock. */
1931 static struct vport *lookup_vport(struct net *net,
1932                                   const struct ovs_header *ovs_header,
1933                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1934 {
1935         struct datapath *dp;
1936         struct vport *vport;
1937
1938         if (a[OVS_VPORT_ATTR_IFINDEX])
1939                 return ERR_PTR(-EOPNOTSUPP);
1940         if (a[OVS_VPORT_ATTR_NAME]) {
1941                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1942                 if (!vport)
1943                         return ERR_PTR(-ENODEV);
1944                 if (ovs_header->dp_ifindex &&
1945                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1946                         return ERR_PTR(-ENODEV);
1947                 return vport;
1948         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1949                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1950
1951                 if (port_no >= DP_MAX_PORTS)
1952                         return ERR_PTR(-EFBIG);
1953
1954                 dp = get_dp(net, ovs_header->dp_ifindex);
1955                 if (!dp)
1956                         return ERR_PTR(-ENODEV);
1957
1958                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1959                 if (!vport)
1960                         return ERR_PTR(-ENODEV);
1961                 return vport;
1962         } else
1963                 return ERR_PTR(-EINVAL);
1964
1965 }
1966
1967 /* Called with ovs_mutex */
1968 static void update_headroom(struct datapath *dp)
1969 {
1970         unsigned dev_headroom, max_headroom = 0;
1971         struct net_device *dev;
1972         struct vport *vport;
1973         int i;
1974
1975         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1976                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1977                         dev = vport->dev;
1978                         dev_headroom = netdev_get_fwd_headroom(dev);
1979                         if (dev_headroom > max_headroom)
1980                                 max_headroom = dev_headroom;
1981                 }
1982         }
1983
1984         dp->max_headroom = max_headroom;
1985         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1986                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1987                         netdev_set_rx_headroom(vport->dev, max_headroom);
1988 }
1989
1990 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1991 {
1992         struct nlattr **a = info->attrs;
1993         struct ovs_header *ovs_header = info->userhdr;
1994         struct vport_parms parms;
1995         struct sk_buff *reply;
1996         struct vport *vport;
1997         struct datapath *dp;
1998         u32 port_no;
1999         int err;
2000
2001         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2002             !a[OVS_VPORT_ATTR_UPCALL_PID])
2003                 return -EINVAL;
2004         if (a[OVS_VPORT_ATTR_IFINDEX])
2005                 return -EOPNOTSUPP;
2006
2007         port_no = a[OVS_VPORT_ATTR_PORT_NO]
2008                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2009         if (port_no >= DP_MAX_PORTS)
2010                 return -EFBIG;
2011
2012         reply = ovs_vport_cmd_alloc_info();
2013         if (!reply)
2014                 return -ENOMEM;
2015
2016         ovs_lock();
2017 restart:
2018         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2019         err = -ENODEV;
2020         if (!dp)
2021                 goto exit_unlock_free;
2022
2023         if (port_no) {
2024                 vport = ovs_vport_ovsl(dp, port_no);
2025                 err = -EBUSY;
2026                 if (vport)
2027                         goto exit_unlock_free;
2028         } else {
2029                 for (port_no = 1; ; port_no++) {
2030                         if (port_no >= DP_MAX_PORTS) {
2031                                 err = -EFBIG;
2032                                 goto exit_unlock_free;
2033                         }
2034                         vport = ovs_vport_ovsl(dp, port_no);
2035                         if (!vport)
2036                                 break;
2037                 }
2038         }
2039
2040         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2041         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2042         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2043         parms.dp = dp;
2044         parms.port_no = port_no;
2045         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2046
2047         vport = new_vport(&parms);
2048         err = PTR_ERR(vport);
2049         if (IS_ERR(vport)) {
2050                 if (err == -EAGAIN)
2051                         goto restart;
2052                 goto exit_unlock_free;
2053         }
2054
2055         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2056                                       info->snd_portid, info->snd_seq, 0,
2057                                       OVS_VPORT_CMD_NEW);
2058
2059         if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2060                 update_headroom(dp);
2061         else
2062                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2063
2064         BUG_ON(err < 0);
2065         ovs_unlock();
2066
2067         ovs_notify(&dp_vport_genl_family, reply, info);
2068         return 0;
2069
2070 exit_unlock_free:
2071         ovs_unlock();
2072         kfree_skb(reply);
2073         return err;
2074 }
2075
2076 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2077 {
2078         struct nlattr **a = info->attrs;
2079         struct sk_buff *reply;
2080         struct vport *vport;
2081         int err;
2082
2083         reply = ovs_vport_cmd_alloc_info();
2084         if (!reply)
2085                 return -ENOMEM;
2086
2087         ovs_lock();
2088         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2089         err = PTR_ERR(vport);
2090         if (IS_ERR(vport))
2091                 goto exit_unlock_free;
2092
2093         if (a[OVS_VPORT_ATTR_TYPE] &&
2094             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2095                 err = -EINVAL;
2096                 goto exit_unlock_free;
2097         }
2098
2099         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2100                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2101                 if (err)
2102                         goto exit_unlock_free;
2103         }
2104
2105
2106         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2107                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2108
2109                 err = ovs_vport_set_upcall_portids(vport, ids);
2110                 if (err)
2111                         goto exit_unlock_free;
2112         }
2113
2114         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2115                                       info->snd_portid, info->snd_seq, 0,
2116                                       OVS_VPORT_CMD_SET);
2117         BUG_ON(err < 0);
2118
2119         ovs_unlock();
2120         ovs_notify(&dp_vport_genl_family, reply, info);
2121         return 0;
2122
2123 exit_unlock_free:
2124         ovs_unlock();
2125         kfree_skb(reply);
2126         return err;
2127 }
2128
2129 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2130 {
2131         bool must_update_headroom = false;
2132         struct nlattr **a = info->attrs;
2133         struct sk_buff *reply;
2134         struct datapath *dp;
2135         struct vport *vport;
2136         int err;
2137
2138         reply = ovs_vport_cmd_alloc_info();
2139         if (!reply)
2140                 return -ENOMEM;
2141
2142         ovs_lock();
2143         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2144         err = PTR_ERR(vport);
2145         if (IS_ERR(vport))
2146                 goto exit_unlock_free;
2147
2148         if (vport->port_no == OVSP_LOCAL) {
2149                 err = -EINVAL;
2150                 goto exit_unlock_free;
2151         }
2152
2153         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2154                                       info->snd_portid, info->snd_seq, 0,
2155                                       OVS_VPORT_CMD_DEL);
2156         BUG_ON(err < 0);
2157
2158         /* the vport deletion may trigger dp headroom update */
2159         dp = vport->dp;
2160         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2161                 must_update_headroom = true;
2162         netdev_reset_rx_headroom(vport->dev);
2163         ovs_dp_detach_port(vport);
2164
2165         if (must_update_headroom)
2166                 update_headroom(dp);
2167         ovs_unlock();
2168
2169         ovs_notify(&dp_vport_genl_family, reply, info);
2170         return 0;
2171
2172 exit_unlock_free:
2173         ovs_unlock();
2174         kfree_skb(reply);
2175         return err;
2176 }
2177
2178 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2179 {
2180         struct nlattr **a = info->attrs;
2181         struct ovs_header *ovs_header = info->userhdr;
2182         struct sk_buff *reply;
2183         struct vport *vport;
2184         int err;
2185
2186         reply = ovs_vport_cmd_alloc_info();
2187         if (!reply)
2188                 return -ENOMEM;
2189
2190         rcu_read_lock();
2191         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2192         err = PTR_ERR(vport);
2193         if (IS_ERR(vport))
2194                 goto exit_unlock_free;
2195         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2196                                       info->snd_portid, info->snd_seq, 0,
2197                                       OVS_VPORT_CMD_GET);
2198         BUG_ON(err < 0);
2199         rcu_read_unlock();
2200
2201         return genlmsg_reply(reply, info);
2202
2203 exit_unlock_free:
2204         rcu_read_unlock();
2205         kfree_skb(reply);
2206         return err;
2207 }
2208
2209 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2210 {
2211         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2212         struct datapath *dp;
2213         int bucket = cb->args[0], skip = cb->args[1];
2214         int i, j = 0;
2215
2216         rcu_read_lock();
2217         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2218         if (!dp) {
2219                 rcu_read_unlock();
2220                 return -ENODEV;
2221         }
2222         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2223                 struct vport *vport;
2224
2225                 j = 0;
2226                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2227                         if (j >= skip &&
2228                             ovs_vport_cmd_fill_info(vport, skb,
2229                                                     sock_net(skb->sk),
2230                                                     NETLINK_CB(cb->skb).portid,
2231                                                     cb->nlh->nlmsg_seq,
2232                                                     NLM_F_MULTI,
2233                                                     OVS_VPORT_CMD_GET) < 0)
2234                                 goto out;
2235
2236                         j++;
2237                 }
2238                 skip = 0;
2239         }
2240 out:
2241         rcu_read_unlock();
2242
2243         cb->args[0] = i;
2244         cb->args[1] = j;
2245
2246         return skb->len;
2247 }
2248
2249 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2250         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2251         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2252         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2253         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2254         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2255         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2256         [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2257         [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2258 };
2259
2260 static const struct genl_ops dp_vport_genl_ops[] = {
2261         { .cmd = OVS_VPORT_CMD_NEW,
2262           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2263           .policy = vport_policy,
2264           .doit = ovs_vport_cmd_new
2265         },
2266         { .cmd = OVS_VPORT_CMD_DEL,
2267           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2268           .policy = vport_policy,
2269           .doit = ovs_vport_cmd_del
2270         },
2271         { .cmd = OVS_VPORT_CMD_GET,
2272           .flags = 0,               /* OK for unprivileged users. */
2273           .policy = vport_policy,
2274           .doit = ovs_vport_cmd_get,
2275           .dumpit = ovs_vport_cmd_dump
2276         },
2277         { .cmd = OVS_VPORT_CMD_SET,
2278           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2279           .policy = vport_policy,
2280           .doit = ovs_vport_cmd_set,
2281         },
2282 };
2283
2284 struct genl_family dp_vport_genl_family __ro_after_init = {
2285         .hdrsize = sizeof(struct ovs_header),
2286         .name = OVS_VPORT_FAMILY,
2287         .version = OVS_VPORT_VERSION,
2288         .maxattr = OVS_VPORT_ATTR_MAX,
2289         .netnsok = true,
2290         .parallel_ops = true,
2291         .ops = dp_vport_genl_ops,
2292         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2293         .mcgrps = &ovs_dp_vport_multicast_group,
2294         .n_mcgrps = 1,
2295         .module = THIS_MODULE,
2296 };
2297
2298 static struct genl_family * const dp_genl_families[] = {
2299         &dp_datapath_genl_family,
2300         &dp_vport_genl_family,
2301         &dp_flow_genl_family,
2302         &dp_packet_genl_family,
2303         &dp_meter_genl_family,
2304 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2305         &dp_ct_limit_genl_family,
2306 #endif
2307 };
2308
2309 static void dp_unregister_genl(int n_families)
2310 {
2311         int i;
2312
2313         for (i = 0; i < n_families; i++)
2314                 genl_unregister_family(dp_genl_families[i]);
2315 }
2316
2317 static int __init dp_register_genl(void)
2318 {
2319         int err;
2320         int i;
2321
2322         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2323
2324                 err = genl_register_family(dp_genl_families[i]);
2325                 if (err)
2326                         goto error;
2327         }
2328
2329         return 0;
2330
2331 error:
2332         dp_unregister_genl(i);
2333         return err;
2334 }
2335
2336 static int __net_init ovs_init_net(struct net *net)
2337 {
2338         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2339
2340         INIT_LIST_HEAD(&ovs_net->dps);
2341         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2342         return ovs_ct_init(net);
2343 }
2344
2345 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2346                                             struct list_head *head)
2347 {
2348         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2349         struct datapath *dp;
2350
2351         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2352                 int i;
2353
2354                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2355                         struct vport *vport;
2356
2357                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2358                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2359                                         continue;
2360
2361                                 if (dev_net(vport->dev) == dnet)
2362                                         list_add(&vport->detach_list, head);
2363                         }
2364                 }
2365         }
2366 }
2367
2368 static void __net_exit ovs_exit_net(struct net *dnet)
2369 {
2370         struct datapath *dp, *dp_next;
2371         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2372         struct vport *vport, *vport_next;
2373         struct net *net;
2374         LIST_HEAD(head);
2375
2376         ovs_ct_exit(dnet);
2377         ovs_lock();
2378         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2379                 __dp_destroy(dp);
2380
2381         down_read(&net_rwsem);
2382         for_each_net(net)
2383                 list_vports_from_net(net, dnet, &head);
2384         up_read(&net_rwsem);
2385
2386         /* Detach all vports from given namespace. */
2387         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2388                 list_del(&vport->detach_list);
2389                 ovs_dp_detach_port(vport);
2390         }
2391
2392         ovs_unlock();
2393
2394         cancel_work_sync(&ovs_net->dp_notify_work);
2395 }
2396
2397 static struct pernet_operations ovs_net_ops = {
2398         .init = ovs_init_net,
2399         .exit = ovs_exit_net,
2400         .id   = &ovs_net_id,
2401         .size = sizeof(struct ovs_net),
2402 };
2403
2404 static int __init dp_init(void)
2405 {
2406         int err;
2407
2408         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2409
2410         pr_info("Open vSwitch switching datapath\n");
2411
2412         err = action_fifos_init();
2413         if (err)
2414                 goto error;
2415
2416         err = ovs_internal_dev_rtnl_link_register();
2417         if (err)
2418                 goto error_action_fifos_exit;
2419
2420         err = ovs_flow_init();
2421         if (err)
2422                 goto error_unreg_rtnl_link;
2423
2424         err = ovs_vport_init();
2425         if (err)
2426                 goto error_flow_exit;
2427
2428         err = register_pernet_device(&ovs_net_ops);
2429         if (err)
2430                 goto error_vport_exit;
2431
2432         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2433         if (err)
2434                 goto error_netns_exit;
2435
2436         err = ovs_netdev_init();
2437         if (err)
2438                 goto error_unreg_notifier;
2439
2440         err = dp_register_genl();
2441         if (err < 0)
2442                 goto error_unreg_netdev;
2443
2444         return 0;
2445
2446 error_unreg_netdev:
2447         ovs_netdev_exit();
2448 error_unreg_notifier:
2449         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2450 error_netns_exit:
2451         unregister_pernet_device(&ovs_net_ops);
2452 error_vport_exit:
2453         ovs_vport_exit();
2454 error_flow_exit:
2455         ovs_flow_exit();
2456 error_unreg_rtnl_link:
2457         ovs_internal_dev_rtnl_link_unregister();
2458 error_action_fifos_exit:
2459         action_fifos_exit();
2460 error:
2461         return err;
2462 }
2463
2464 static void dp_cleanup(void)
2465 {
2466         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2467         ovs_netdev_exit();
2468         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2469         unregister_pernet_device(&ovs_net_ops);
2470         rcu_barrier();
2471         ovs_vport_exit();
2472         ovs_flow_exit();
2473         ovs_internal_dev_rtnl_link_unregister();
2474         action_fifos_exit();
2475 }
2476
2477 module_init(dp_init);
2478 module_exit(dp_cleanup);
2479
2480 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2481 MODULE_LICENSE("GPL");
2482 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2483 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2484 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2485 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2486 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2487 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);