net/openvswitch/datapath.c

   1 /*
   2  * Copyright (c) 2007-2014 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/ethtool.h>
  40 #include <linux/wait.h>
  41 #include <asm/div64.h>
  42 #include <linux/highmem.h>
  43 #include <linux/netfilter_bridge.h>
  44 #include <linux/netfilter_ipv4.h>
  45 #include <linux/inetdevice.h>
  46 #include <linux/list.h>
  47 #include <linux/openvswitch.h>
  48 #include <linux/rculist.h>
  49 #include <linux/dmi.h>
  50 #include <net/genetlink.h>
  51 #include <net/net_namespace.h>
  52 #include <net/netns/generic.h>
  53
  54 #include "datapath.h"
  55 #include "flow.h"
  56 #include "flow_table.h"
  57 #include "flow_netlink.h"
  58 #include "vport-internal_dev.h"
  59 #include "vport-netdev.h"
  60
  61 unsigned int ovs_net_id __read_mostly;
  62
  63 static struct genl_family dp_packet_genl_family;
  64 static struct genl_family dp_flow_genl_family;
  65 static struct genl_family dp_datapath_genl_family;
  66
  67 static const struct nla_policy flow_policy[];
  68
  69 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
  70         .name = OVS_FLOW_MCGROUP,
  71 };
  72
  73 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  74         .name = OVS_DATAPATH_MCGROUP,
  75 };
  76
  77 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
  78         .name = OVS_VPORT_MCGROUP,
  79 };
  80
  81 /* Check if need to build a reply message.
  82  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  83 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  84                             unsigned int group)
  85 {
  86         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  87                genl_has_listeners(family, genl_info_net(info), group);
  88 }
  89
  90 static void ovs_notify(struct genl_family *family,
  91                        struct sk_buff *skb, struct genl_info *info)
  92 {
  93         genl_notify(family, skb, info, 0, GFP_KERNEL);
  94 }
  95
  96 /**
  97  * DOC: Locking:
  98  *
  99  * All writes e.g. Writes to device state (add/remove datapath, port, set
 100  * operations on vports, etc.), Writes to other state (flow table
 101  * modifications, set miscellaneous datapath parameters, etc.) are protected
 102  * by ovs_lock.
 103  *
 104  * Reads are protected by RCU.
 105  *
 106  * There are a few special cases (mostly stats) that have their own
 107  * synchronization but they nest under all of above and don't interact with
 108  * each other.
 109  *
 110  * The RTNL lock nests inside ovs_mutex.
 111  */
 112
 113 static DEFINE_MUTEX(ovs_mutex);
 114
 115 void ovs_lock(void)
 116 {
 117         mutex_lock(&ovs_mutex);
 118 }
 119
 120 void ovs_unlock(void)
 121 {
 122         mutex_unlock(&ovs_mutex);
 123 }
 124
 125 #ifdef CONFIG_LOCKDEP
 126 int lockdep_ovsl_is_held(void)
 127 {
 128         if (debug_locks)
 129                 return lockdep_is_held(&ovs_mutex);
 130         else
 131                 return 1;
 132 }
 133 #endif
 134
 135 static struct vport *new_vport(const struct vport_parms *);
 136 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 137                              const struct sw_flow_key *,
 138                              const struct dp_upcall_info *,
 139                              uint32_t cutlen);
 140 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 141                                   const struct sw_flow_key *,
 142                                   const struct dp_upcall_info *,
 143                                   uint32_t cutlen);
 144
 145 /* Must be called with rcu_read_lock. */
 146 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
 147 {
 148         struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
 149
 150         if (dev) {
 151                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 152                 if (vport)
 153                         return vport->dp;
 154         }
 155
 156         return NULL;
 157 }
 158
 159 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
 160  * returned dp pointer valid.
 161  */
 162 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
 163 {
 164         struct datapath *dp;
 165
 166         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
 167         rcu_read_lock();
 168         dp = get_dp_rcu(net, dp_ifindex);
 169         rcu_read_unlock();
 170
 171         return dp;
 172 }
 173
 174 /* Must be called with rcu_read_lock or ovs_mutex. */
 175 const char *ovs_dp_name(const struct datapath *dp)
 176 {
 177         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 178         return ovs_vport_name(vport);
 179 }
 180
 181 static int get_dpifindex(const struct datapath *dp)
 182 {
 183         struct vport *local;
 184         int ifindex;
 185
 186         rcu_read_lock();
 187
 188         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 189         if (local)
 190                 ifindex = local->dev->ifindex;
 191         else
 192                 ifindex = 0;
 193
 194         rcu_read_unlock();
 195
 196         return ifindex;
 197 }
 198
 199 static void destroy_dp_rcu(struct rcu_head *rcu)
 200 {
 201         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 202
 203         ovs_flow_tbl_destroy(&dp->table);
 204         free_percpu(dp->stats_percpu);
 205         kfree(dp->ports);
 206         kfree(dp);
 207 }
 208
 209 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 210                                             u16 port_no)
 211 {
 212         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 213 }
 214
 215 /* Called with ovs_mutex or RCU read lock. */
 216 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 217 {
 218         struct vport *vport;
 219         struct hlist_head *head;
 220
 221         head = vport_hash_bucket(dp, port_no);
 222         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 223                 if (vport->port_no == port_no)
 224                         return vport;
 225         }
 226         return NULL;
 227 }
 228
 229 /* Called with ovs_mutex. */
 230 static struct vport *new_vport(const struct vport_parms *parms)
 231 {
 232         struct vport *vport;
 233
 234         vport = ovs_vport_add(parms);
 235         if (!IS_ERR(vport)) {
 236                 struct datapath *dp = parms->dp;
 237                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 238
 239                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 240         }
 241         return vport;
 242 }
 243
 244 void ovs_dp_detach_port(struct vport *p)
 245 {
 246         ASSERT_OVSL();
 247
 248         /* First drop references to device. */
 249         hlist_del_rcu(&p->dp_hash_node);
 250
 251         /* Then destroy it. */
 252         ovs_vport_del(p);
 253 }
 254
 255 /* Must be called with rcu_read_lock. */
 256 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 257 {
 258         const struct vport *p = OVS_CB(skb)->input_vport;
 259         struct datapath *dp = p->dp;
 260         struct sw_flow *flow;
 261         struct sw_flow_actions *sf_acts;
 262         struct dp_stats_percpu *stats;
 263         u64 *stats_counter;
 264         u32 n_mask_hit;
 265
 266         stats = this_cpu_ptr(dp->stats_percpu);
 267
 268         /* Look up flow. */
 269         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
 270         if (unlikely(!flow)) {
 271                 struct dp_upcall_info upcall;
 272                 int error;
 273
 274                 memset(&upcall, 0, sizeof(upcall));
 275                 upcall.cmd = OVS_PACKET_CMD_MISS;
 276                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 277                 upcall.mru = OVS_CB(skb)->mru;
 278                 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
 279                 if (unlikely(error))
 280                         kfree_skb(skb);
 281                 else
 282                         consume_skb(skb);
 283                 stats_counter = &stats->n_missed;
 284                 goto out;
 285         }
 286
 287         ovs_flow_stats_update(flow, key->tp.flags, skb);
 288         sf_acts = rcu_dereference(flow->sf_acts);
 289         ovs_execute_actions(dp, skb, sf_acts, key);
 290
 291         stats_counter = &stats->n_hit;
 292
 293 out:
 294         /* Update datapath statistics. */
 295         u64_stats_update_begin(&stats->syncp);
 296         (*stats_counter)++;
 297         stats->n_mask_hit += n_mask_hit;
 298         u64_stats_update_end(&stats->syncp);
 299 }
 300
 301 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 302                   const struct sw_flow_key *key,
 303                   const struct dp_upcall_info *upcall_info,
 304                   uint32_t cutlen)
 305 {
 306         struct dp_stats_percpu *stats;
 307         int err;
 308
 309         if (upcall_info->portid == 0) {
 310                 err = -ENOTCONN;
 311                 goto err;
 312         }
 313
 314         if (!skb_is_gso(skb))
 315                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 316         else
 317                 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
 318         if (err)
 319                 goto err;
 320
 321         return 0;
 322
 323 err:
 324         stats = this_cpu_ptr(dp->stats_percpu);
 325
 326         u64_stats_update_begin(&stats->syncp);
 327         stats->n_lost++;
 328         u64_stats_update_end(&stats->syncp);
 329
 330         return err;
 331 }
 332
 333 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 334                              const struct sw_flow_key *key,
 335                              const struct dp_upcall_info *upcall_info,
 336                                  uint32_t cutlen)
 337 {
 338         struct sk_buff *segs, *nskb;
 339         int err;
 340
 341         BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
 342         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 343         if (IS_ERR(segs))
 344                 return PTR_ERR(segs);
 345         if (segs == NULL)
 346                 return -EINVAL;
 347
 348         /* Queue all of the segments. */
 349         skb = segs;
 350         do {
 351                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 352                 if (err)
 353                         break;
 354
 355         } while ((skb = skb->next));
 356
 357         /* Free all of the segments. */
 358         skb = segs;
 359         do {
 360                 nskb = skb->next;
 361                 if (err)
 362                         kfree_skb(skb);
 363                 else
 364                         consume_skb(skb);
 365         } while ((skb = nskb));
 366         return err;
 367 }
 368
 369 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 370                               unsigned int hdrlen, int actions_attrlen)
 371 {
 372         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 373                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 374                 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 375                 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
 376
 377         /* OVS_PACKET_ATTR_USERDATA */
 378         if (upcall_info->userdata)
 379                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 380
 381         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 382         if (upcall_info->egress_tun_info)
 383                 size += nla_total_size(ovs_tun_key_attr_size());
 384
 385         /* OVS_PACKET_ATTR_ACTIONS */
 386         if (upcall_info->actions_len)
 387                 size += nla_total_size(actions_attrlen);
 388
 389         /* OVS_PACKET_ATTR_MRU */
 390         if (upcall_info->mru)
 391                 size += nla_total_size(sizeof(upcall_info->mru));
 392
 393         return size;
 394 }
 395
 396 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 397 {
 398         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 399                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 400
 401                 if (plen > 0)
 402                         skb_put_zero(skb, plen);
 403         }
 404 }
 405
 406 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 407                                   const struct sw_flow_key *key,
 408                                   const struct dp_upcall_info *upcall_info,
 409                                   uint32_t cutlen)
 410 {
 411         struct ovs_header *upcall;
 412         struct sk_buff *nskb = NULL;
 413         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 414         struct nlattr *nla;
 415         size_t len;
 416         unsigned int hlen;
 417         int err, dp_ifindex;
 418
 419         dp_ifindex = get_dpifindex(dp);
 420         if (!dp_ifindex)
 421                 return -ENODEV;
 422
 423         if (skb_vlan_tag_present(skb)) {
 424                 nskb = skb_clone(skb, GFP_ATOMIC);
 425                 if (!nskb)
 426                         return -ENOMEM;
 427
 428                 nskb = __vlan_hwaccel_push_inside(nskb);
 429                 if (!nskb)
 430                         return -ENOMEM;
 431
 432                 skb = nskb;
 433         }
 434
 435         if (nla_attr_size(skb->len) > USHRT_MAX) {
 436                 err = -EFBIG;
 437                 goto out;
 438         }
 439
 440         /* Complete checksum if needed */
 441         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 442             (err = skb_csum_hwoffload_help(skb, 0)))
 443                 goto out;
 444
 445         /* Older versions of OVS user space enforce alignment of the last
 446          * Netlink attribute to NLA_ALIGNTO which would require extensive
 447          * padding logic. Only perform zerocopy if padding is not required.
 448          */
 449         if (dp->user_features & OVS_DP_F_UNALIGNED)
 450                 hlen = skb_zerocopy_headlen(skb);
 451         else
 452                 hlen = skb->len;
 453
 454         len = upcall_msg_size(upcall_info, hlen - cutlen,
 455                               OVS_CB(skb)->acts_origlen);
 456         user_skb = genlmsg_new(len, GFP_ATOMIC);
 457         if (!user_skb) {
 458                 err = -ENOMEM;
 459                 goto out;
 460         }
 461
 462         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 463                              0, upcall_info->cmd);
 464         upcall->dp_ifindex = dp_ifindex;
 465
 466         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 467         BUG_ON(err);
 468
 469         if (upcall_info->userdata)
 470                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 471                           nla_len(upcall_info->userdata),
 472                           nla_data(upcall_info->userdata));
 473
 474         if (upcall_info->egress_tun_info) {
 475                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 476                 err = ovs_nla_put_tunnel_info(user_skb,
 477                                               upcall_info->egress_tun_info);
 478                 BUG_ON(err);
 479                 nla_nest_end(user_skb, nla);
 480         }
 481
 482         if (upcall_info->actions_len) {
 483                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 484                 err = ovs_nla_put_actions(upcall_info->actions,
 485                                           upcall_info->actions_len,
 486                                           user_skb);
 487                 if (!err)
 488                         nla_nest_end(user_skb, nla);
 489                 else
 490                         nla_nest_cancel(user_skb, nla);
 491         }
 492
 493         /* Add OVS_PACKET_ATTR_MRU */
 494         if (upcall_info->mru) {
 495                 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
 496                                 upcall_info->mru)) {
 497                         err = -ENOBUFS;
 498                         goto out;
 499                 }
 500                 pad_packet(dp, user_skb);
 501         }
 502
 503         /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 504         if (cutlen > 0) {
 505                 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
 506                                 skb->len)) {
 507                         err = -ENOBUFS;
 508                         goto out;
 509                 }
 510                 pad_packet(dp, user_skb);
 511         }
 512
 513         /* Only reserve room for attribute header, packet data is added
 514          * in skb_zerocopy() */
 515         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 516                 err = -ENOBUFS;
 517                 goto out;
 518         }
 519         nla->nla_len = nla_attr_size(skb->len - cutlen);
 520
 521         err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 522         if (err)
 523                 goto out;
 524
 525         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 526         pad_packet(dp, user_skb);
 527
 528         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 529
 530         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 531         user_skb = NULL;
 532 out:
 533         if (err)
 534                 skb_tx_error(skb);
 535         kfree_skb(user_skb);
 536         kfree_skb(nskb);
 537         return err;
 538 }
 539
 540 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 541 {
 542         struct ovs_header *ovs_header = info->userhdr;
 543         struct net *net = sock_net(skb->sk);
 544         struct nlattr **a = info->attrs;
 545         struct sw_flow_actions *acts;
 546         struct sk_buff *packet;
 547         struct sw_flow *flow;
 548         struct sw_flow_actions *sf_acts;
 549         struct datapath *dp;
 550         struct vport *input_vport;
 551         u16 mru = 0;
 552         int len;
 553         int err;
 554         bool log = !a[OVS_PACKET_ATTR_PROBE];
 555
 556         err = -EINVAL;
 557         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 558             !a[OVS_PACKET_ATTR_ACTIONS])
 559                 goto err;
 560
 561         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 562         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 563         err = -ENOMEM;
 564         if (!packet)
 565                 goto err;
 566         skb_reserve(packet, NET_IP_ALIGN);
 567
 568         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 569
 570         /* Set packet's mru */
 571         if (a[OVS_PACKET_ATTR_MRU]) {
 572                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 573                 packet->ignore_df = 1;
 574         }
 575         OVS_CB(packet)->mru = mru;
 576
 577         /* Build an sw_flow for sending this packet. */
 578         flow = ovs_flow_alloc();
 579         err = PTR_ERR(flow);
 580         if (IS_ERR(flow))
 581                 goto err_kfree_skb;
 582
 583         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 584                                              packet, &flow->key, log);
 585         if (err)
 586                 goto err_flow_free;
 587
 588         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 589                                    &flow->key, &acts, log);
 590         if (err)
 591                 goto err_flow_free;
 592
 593         rcu_assign_pointer(flow->sf_acts, acts);
 594         packet->priority = flow->key.phy.priority;
 595         packet->mark = flow->key.phy.skb_mark;
 596
 597         rcu_read_lock();
 598         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 599         err = -ENODEV;
 600         if (!dp)
 601                 goto err_unlock;
 602
 603         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 604         if (!input_vport)
 605                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 606
 607         if (!input_vport)
 608                 goto err_unlock;
 609
 610         packet->dev = input_vport->dev;
 611         OVS_CB(packet)->input_vport = input_vport;
 612         sf_acts = rcu_dereference(flow->sf_acts);
 613
 614         local_bh_disable();
 615         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 616         local_bh_enable();
 617         rcu_read_unlock();
 618
 619         ovs_flow_free(flow, false);
 620         return err;
 621
 622 err_unlock:
 623         rcu_read_unlock();
 624 err_flow_free:
 625         ovs_flow_free(flow, false);
 626 err_kfree_skb:
 627         kfree_skb(packet);
 628 err:
 629         return err;
 630 }
 631
 632 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 633         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 634         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 635         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 636         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 637         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 638 };
 639
 640 static const struct genl_ops dp_packet_genl_ops[] = {
 641         { .cmd = OVS_PACKET_CMD_EXECUTE,
 642           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 643           .policy = packet_policy,
 644           .doit = ovs_packet_cmd_execute
 645         }
 646 };
 647
 648 static struct genl_family dp_packet_genl_family __ro_after_init = {
 649         .hdrsize = sizeof(struct ovs_header),
 650         .name = OVS_PACKET_FAMILY,
 651         .version = OVS_PACKET_VERSION,
 652         .maxattr = OVS_PACKET_ATTR_MAX,
 653         .netnsok = true,
 654         .parallel_ops = true,
 655         .ops = dp_packet_genl_ops,
 656         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 657         .module = THIS_MODULE,
 658 };
 659
 660 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 661                          struct ovs_dp_megaflow_stats *mega_stats)
 662 {
 663         int i;
 664
 665         memset(mega_stats, 0, sizeof(*mega_stats));
 666
 667         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 668         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 669
 670         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 671
 672         for_each_possible_cpu(i) {
 673                 const struct dp_stats_percpu *percpu_stats;
 674                 struct dp_stats_percpu local_stats;
 675                 unsigned int start;
 676
 677                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 678
 679                 do {
 680                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 681                         local_stats = *percpu_stats;
 682                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 683
 684                 stats->n_hit += local_stats.n_hit;
 685                 stats->n_missed += local_stats.n_missed;
 686                 stats->n_lost += local_stats.n_lost;
 687                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 688         }
 689 }
 690
 691 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 692 {
 693         return ovs_identifier_is_ufid(sfid) &&
 694                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 695 }
 696
 697 static bool should_fill_mask(uint32_t ufid_flags)
 698 {
 699         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 700 }
 701
 702 static bool should_fill_actions(uint32_t ufid_flags)
 703 {
 704         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 705 }
 706
 707 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 708                                     const struct sw_flow_id *sfid,
 709                                     uint32_t ufid_flags)
 710 {
 711         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 712
 713         /* OVS_FLOW_ATTR_UFID */
 714         if (sfid && ovs_identifier_is_ufid(sfid))
 715                 len += nla_total_size(sfid->ufid_len);
 716
 717         /* OVS_FLOW_ATTR_KEY */
 718         if (!sfid || should_fill_key(sfid, ufid_flags))
 719                 len += nla_total_size(ovs_key_attr_size());
 720
 721         /* OVS_FLOW_ATTR_MASK */
 722         if (should_fill_mask(ufid_flags))
 723                 len += nla_total_size(ovs_key_attr_size());
 724
 725         /* OVS_FLOW_ATTR_ACTIONS */
 726         if (should_fill_actions(ufid_flags))
 727                 len += nla_total_size(acts->orig_len);
 728
 729         return len
 730                 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 731                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 732                 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 733 }
 734
 735 /* Called with ovs_mutex or RCU read lock. */
 736 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 737                                    struct sk_buff *skb)
 738 {
 739         struct ovs_flow_stats stats;
 740         __be16 tcp_flags;
 741         unsigned long used;
 742
 743         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 744
 745         if (used &&
 746             nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 747                               OVS_FLOW_ATTR_PAD))
 748                 return -EMSGSIZE;
 749
 750         if (stats.n_packets &&
 751             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 752                           sizeof(struct ovs_flow_stats), &stats,
 753                           OVS_FLOW_ATTR_PAD))
 754                 return -EMSGSIZE;
 755
 756         if ((u8)ntohs(tcp_flags) &&
 757              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 758                 return -EMSGSIZE;
 759
 760         return 0;
 761 }
 762
 763 /* Called with ovs_mutex or RCU read lock. */
 764 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 765                                      struct sk_buff *skb, int skb_orig_len)
 766 {
 767         struct nlattr *start;
 768         int err;
 769
 770         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 771          * this is the first flow to be dumped into 'skb'.  This is unusual for
 772          * Netlink but individual action lists can be longer than
 773          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 774          * The userspace caller can always fetch the actions separately if it
 775          * really wants them.  (Most userspace callers in fact don't care.)
 776          *
 777          * This can only fail for dump operations because the skb is always
 778          * properly sized for single flows.
 779          */
 780         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 781         if (start) {
 782                 const struct sw_flow_actions *sf_acts;
 783
 784                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 785                 err = ovs_nla_put_actions(sf_acts->actions,
 786                                           sf_acts->actions_len, skb);
 787
 788                 if (!err)
 789                         nla_nest_end(skb, start);
 790                 else {
 791                         if (skb_orig_len)
 792                                 return err;
 793
 794                         nla_nest_cancel(skb, start);
 795                 }
 796         } else if (skb_orig_len) {
 797                 return -EMSGSIZE;
 798         }
 799
 800         return 0;
 801 }
 802
 803 /* Called with ovs_mutex or RCU read lock. */
 804 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 805                                   struct sk_buff *skb, u32 portid,
 806                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 807 {
 808         const int skb_orig_len = skb->len;
 809         struct ovs_header *ovs_header;
 810         int err;
 811
 812         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 813                                  flags, cmd);
 814         if (!ovs_header)
 815                 return -EMSGSIZE;
 816
 817         ovs_header->dp_ifindex = dp_ifindex;
 818
 819         err = ovs_nla_put_identifier(flow, skb);
 820         if (err)
 821                 goto error;
 822
 823         if (should_fill_key(&flow->id, ufid_flags)) {
 824                 err = ovs_nla_put_masked_key(flow, skb);
 825                 if (err)
 826                         goto error;
 827         }
 828
 829         if (should_fill_mask(ufid_flags)) {
 830                 err = ovs_nla_put_mask(flow, skb);
 831                 if (err)
 832                         goto error;
 833         }
 834
 835         err = ovs_flow_cmd_fill_stats(flow, skb);
 836         if (err)
 837                 goto error;
 838
 839         if (should_fill_actions(ufid_flags)) {
 840                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 841                 if (err)
 842                         goto error;
 843         }
 844
 845         genlmsg_end(skb, ovs_header);
 846         return 0;
 847
 848 error:
 849         genlmsg_cancel(skb, ovs_header);
 850         return err;
 851 }
 852
 853 /* May not be called with RCU read lock. */
 854 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 855                                                const struct sw_flow_id *sfid,
 856                                                struct genl_info *info,
 857                                                bool always,
 858                                                uint32_t ufid_flags)
 859 {
 860         struct sk_buff *skb;
 861         size_t len;
 862
 863         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 864                 return NULL;
 865
 866         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 867         skb = genlmsg_new(len, GFP_KERNEL);
 868         if (!skb)
 869                 return ERR_PTR(-ENOMEM);
 870
 871         return skb;
 872 }
 873
 874 /* Called with ovs_mutex. */
 875 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 876                                                int dp_ifindex,
 877                                                struct genl_info *info, u8 cmd,
 878                                                bool always, u32 ufid_flags)
 879 {
 880         struct sk_buff *skb;
 881         int retval;
 882
 883         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 884                                       &flow->id, info, always, ufid_flags);
 885         if (IS_ERR_OR_NULL(skb))
 886                 return skb;
 887
 888         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 889                                         info->snd_portid, info->snd_seq, 0,
 890                                         cmd, ufid_flags);
 891         BUG_ON(retval < 0);
 892         return skb;
 893 }
 894
 895 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 896 {
 897         struct net *net = sock_net(skb->sk);
 898         struct nlattr **a = info->attrs;
 899         struct ovs_header *ovs_header = info->userhdr;
 900         struct sw_flow *flow = NULL, *new_flow;
 901         struct sw_flow_mask mask;
 902         struct sk_buff *reply;
 903         struct datapath *dp;
 904         struct sw_flow_actions *acts;
 905         struct sw_flow_match match;
 906         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 907         int error;
 908         bool log = !a[OVS_FLOW_ATTR_PROBE];
 909
 910         /* Must have key and actions. */
 911         error = -EINVAL;
 912         if (!a[OVS_FLOW_ATTR_KEY]) {
 913                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 914                 goto error;
 915         }
 916         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 917                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 918                 goto error;
 919         }
 920
 921         /* Most of the time we need to allocate a new flow, do it before
 922          * locking.
 923          */
 924         new_flow = ovs_flow_alloc();
 925         if (IS_ERR(new_flow)) {
 926                 error = PTR_ERR(new_flow);
 927                 goto error;
 928         }
 929
 930         /* Extract key. */
 931         ovs_match_init(&match, &new_flow->key, false, &mask);
 932         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 933                                   a[OVS_FLOW_ATTR_MASK], log);
 934         if (error)
 935                 goto err_kfree_flow;
 936
 937         /* Extract flow identifier. */
 938         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 939                                        &new_flow->key, log);
 940         if (error)
 941                 goto err_kfree_flow;
 942
 943         /* unmasked key is needed to match when ufid is not used. */
 944         if (ovs_identifier_is_key(&new_flow->id))
 945                 match.key = new_flow->id.unmasked_key;
 946
 947         ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
 948
 949         /* Validate actions. */
 950         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 951                                      &new_flow->key, &acts, log);
 952         if (error) {
 953                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 954                 goto err_kfree_flow;
 955         }
 956
 957         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 958                                         ufid_flags);
 959         if (IS_ERR(reply)) {
 960                 error = PTR_ERR(reply);
 961                 goto err_kfree_acts;
 962         }
 963
 964         ovs_lock();
 965         dp = get_dp(net, ovs_header->dp_ifindex);
 966         if (unlikely(!dp)) {
 967                 error = -ENODEV;
 968                 goto err_unlock_ovs;
 969         }
 970
 971         /* Check if this is a duplicate flow */
 972         if (ovs_identifier_is_ufid(&new_flow->id))
 973                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
 974         if (!flow)
 975                 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
 976         if (likely(!flow)) {
 977                 rcu_assign_pointer(new_flow->sf_acts, acts);
 978
 979                 /* Put flow in bucket. */
 980                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 981                 if (unlikely(error)) {
 982                         acts = NULL;
 983                         goto err_unlock_ovs;
 984                 }
 985
 986                 if (unlikely(reply)) {
 987                         error = ovs_flow_cmd_fill_info(new_flow,
 988                                                        ovs_header->dp_ifindex,
 989                                                        reply, info->snd_portid,
 990                                                        info->snd_seq, 0,
 991                                                        OVS_FLOW_CMD_NEW,
 992                                                        ufid_flags);
 993                         BUG_ON(error < 0);
 994                 }
 995                 ovs_unlock();
 996         } else {
 997                 struct sw_flow_actions *old_acts;
 998
 999                 /* Bail out if we're not allowed to modify an existing flow.
1000                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1001                  * because Generic Netlink treats the latter as a dump
1002                  * request.  We also accept NLM_F_EXCL in case that bug ever
1003                  * gets fixed.
1004                  */
1005                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1006                                                          | NLM_F_EXCL))) {
1007                         error = -EEXIST;
1008                         goto err_unlock_ovs;
1009                 }
1010                 /* The flow identifier has to be the same for flow updates.
1011                  * Look for any overlapping flow.
1012                  */
1013                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1014                         if (ovs_identifier_is_key(&flow->id))
1015                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1016                                                                  &match);
1017                         else /* UFID matches but key is different */
1018                                 flow = NULL;
1019                         if (!flow) {
1020                                 error = -ENOENT;
1021                                 goto err_unlock_ovs;
1022                         }
1023                 }
1024                 /* Update actions. */
1025                 old_acts = ovsl_dereference(flow->sf_acts);
1026                 rcu_assign_pointer(flow->sf_acts, acts);
1027
1028                 if (unlikely(reply)) {
1029                         error = ovs_flow_cmd_fill_info(flow,
1030                                                        ovs_header->dp_ifindex,
1031                                                        reply, info->snd_portid,
1032                                                        info->snd_seq, 0,
1033                                                        OVS_FLOW_CMD_NEW,
1034                                                        ufid_flags);
1035                         BUG_ON(error < 0);
1036                 }
1037                 ovs_unlock();
1038
1039                 ovs_nla_free_flow_actions_rcu(old_acts);
1040                 ovs_flow_free(new_flow, false);
1041         }
1042
1043         if (reply)
1044                 ovs_notify(&dp_flow_genl_family, reply, info);
1045         return 0;
1046
1047 err_unlock_ovs:
1048         ovs_unlock();
1049         kfree_skb(reply);
1050 err_kfree_acts:
1051         ovs_nla_free_flow_actions(acts);
1052 err_kfree_flow:
1053         ovs_flow_free(new_flow, false);
1054 error:
1055         return error;
1056 }
1057
1058 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1059 static struct sw_flow_actions *get_flow_actions(struct net *net,
1060                                                 const struct nlattr *a,
1061                                                 const struct sw_flow_key *key,
1062                                                 const struct sw_flow_mask *mask,
1063                                                 bool log)
1064 {
1065         struct sw_flow_actions *acts;
1066         struct sw_flow_key masked_key;
1067         int error;
1068
1069         ovs_flow_mask_key(&masked_key, key, true, mask);
1070         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1071         if (error) {
1072                 OVS_NLERR(log,
1073                           "Actions may not be safe on all matching packets");
1074                 return ERR_PTR(error);
1075         }
1076
1077         return acts;
1078 }
1079
1080 /* Factor out match-init and action-copy to avoid
1081  * "Wframe-larger-than=1024" warning. Because mask is only
1082  * used to get actions, we new a function to save some
1083  * stack space.
1084  *
1085  * If there are not key and action attrs, we return 0
1086  * directly. In the case, the caller will also not use the
1087  * match as before. If there is action attr, we try to get
1088  * actions and save them to *acts. Before returning from
1089  * the function, we reset the match->mask pointer. Because
1090  * we should not to return match object with dangling reference
1091  * to mask.
1092  * */
1093 static int ovs_nla_init_match_and_action(struct net *net,
1094                                          struct sw_flow_match *match,
1095                                          struct sw_flow_key *key,
1096                                          struct nlattr **a,
1097                                          struct sw_flow_actions **acts,
1098                                          bool log)
1099 {
1100         struct sw_flow_mask mask;
1101         int error = 0;
1102
1103         if (a[OVS_FLOW_ATTR_KEY]) {
1104                 ovs_match_init(match, key, true, &mask);
1105                 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1106                                           a[OVS_FLOW_ATTR_MASK], log);
1107                 if (error)
1108                         goto error;
1109         }
1110
1111         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1112                 if (!a[OVS_FLOW_ATTR_KEY]) {
1113                         OVS_NLERR(log,
1114                                   "Flow key attribute not present in set flow.");
1115                         error = -EINVAL;
1116                         goto error;
1117                 }
1118
1119                 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1120                                          &mask, log);
1121                 if (IS_ERR(*acts)) {
1122                         error = PTR_ERR(*acts);
1123                         goto error;
1124                 }
1125         }
1126
1127         /* On success, error is 0. */
1128 error:
1129         match->mask = NULL;
1130         return error;
1131 }
1132
1133 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1134 {
1135         struct net *net = sock_net(skb->sk);
1136         struct nlattr **a = info->attrs;
1137         struct ovs_header *ovs_header = info->userhdr;
1138         struct sw_flow_key key;
1139         struct sw_flow *flow;
1140         struct sk_buff *reply = NULL;
1141         struct datapath *dp;
1142         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1143         struct sw_flow_match match;
1144         struct sw_flow_id sfid;
1145         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1146         int error = 0;
1147         bool log = !a[OVS_FLOW_ATTR_PROBE];
1148         bool ufid_present;
1149
1150         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1151         if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1152                 OVS_NLERR(log,
1153                           "Flow set message rejected, Key attribute missing.");
1154                 return -EINVAL;
1155         }
1156
1157         error = ovs_nla_init_match_and_action(net, &match, &key, a,
1158                                               &acts, log);
1159         if (error)
1160                 goto error;
1161
1162         if (acts) {
1163                 /* Can allocate before locking if have acts. */
1164                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1165                                                 ufid_flags);
1166                 if (IS_ERR(reply)) {
1167                         error = PTR_ERR(reply);
1168                         goto err_kfree_acts;
1169                 }
1170         }
1171
1172         ovs_lock();
1173         dp = get_dp(net, ovs_header->dp_ifindex);
1174         if (unlikely(!dp)) {
1175                 error = -ENODEV;
1176                 goto err_unlock_ovs;
1177         }
1178         /* Check that the flow exists. */
1179         if (ufid_present)
1180                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1181         else
1182                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1183         if (unlikely(!flow)) {
1184                 error = -ENOENT;
1185                 goto err_unlock_ovs;
1186         }
1187
1188         /* Update actions, if present. */
1189         if (likely(acts)) {
1190                 old_acts = ovsl_dereference(flow->sf_acts);
1191                 rcu_assign_pointer(flow->sf_acts, acts);
1192
1193                 if (unlikely(reply)) {
1194                         error = ovs_flow_cmd_fill_info(flow,
1195                                                        ovs_header->dp_ifindex,
1196                                                        reply, info->snd_portid,
1197                                                        info->snd_seq, 0,
1198                                                        OVS_FLOW_CMD_NEW,
1199                                                        ufid_flags);
1200                         BUG_ON(error < 0);
1201                 }
1202         } else {
1203                 /* Could not alloc without acts before locking. */
1204                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1205                                                 info, OVS_FLOW_CMD_NEW, false,
1206                                                 ufid_flags);
1207
1208                 if (IS_ERR(reply)) {
1209                         error = PTR_ERR(reply);
1210                         goto err_unlock_ovs;
1211                 }
1212         }
1213
1214         /* Clear stats. */
1215         if (a[OVS_FLOW_ATTR_CLEAR])
1216                 ovs_flow_stats_clear(flow);
1217         ovs_unlock();
1218
1219         if (reply)
1220                 ovs_notify(&dp_flow_genl_family, reply, info);
1221         if (old_acts)
1222                 ovs_nla_free_flow_actions_rcu(old_acts);
1223
1224         return 0;
1225
1226 err_unlock_ovs:
1227         ovs_unlock();
1228         kfree_skb(reply);
1229 err_kfree_acts:
1230         ovs_nla_free_flow_actions(acts);
1231 error:
1232         return error;
1233 }
1234
1235 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1236 {
1237         struct nlattr **a = info->attrs;
1238         struct ovs_header *ovs_header = info->userhdr;
1239         struct net *net = sock_net(skb->sk);
1240         struct sw_flow_key key;
1241         struct sk_buff *reply;
1242         struct sw_flow *flow;
1243         struct datapath *dp;
1244         struct sw_flow_match match;
1245         struct sw_flow_id ufid;
1246         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1247         int err = 0;
1248         bool log = !a[OVS_FLOW_ATTR_PROBE];
1249         bool ufid_present;
1250
1251         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1252         if (a[OVS_FLOW_ATTR_KEY]) {
1253                 ovs_match_init(&match, &key, true, NULL);
1254                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1255                                         log);
1256         } else if (!ufid_present) {
1257                 OVS_NLERR(log,
1258                           "Flow get message rejected, Key attribute missing.");
1259                 err = -EINVAL;
1260         }
1261         if (err)
1262                 return err;
1263
1264         ovs_lock();
1265         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1266         if (!dp) {
1267                 err = -ENODEV;
1268                 goto unlock;
1269         }
1270
1271         if (ufid_present)
1272                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1273         else
1274                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1275         if (!flow) {
1276                 err = -ENOENT;
1277                 goto unlock;
1278         }
1279
1280         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1281                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1282         if (IS_ERR(reply)) {
1283                 err = PTR_ERR(reply);
1284                 goto unlock;
1285         }
1286
1287         ovs_unlock();
1288         return genlmsg_reply(reply, info);
1289 unlock:
1290         ovs_unlock();
1291         return err;
1292 }
1293
1294 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1295 {
1296         struct nlattr **a = info->attrs;
1297         struct ovs_header *ovs_header = info->userhdr;
1298         struct net *net = sock_net(skb->sk);
1299         struct sw_flow_key key;
1300         struct sk_buff *reply;
1301         struct sw_flow *flow = NULL;
1302         struct datapath *dp;
1303         struct sw_flow_match match;
1304         struct sw_flow_id ufid;
1305         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1306         int err;
1307         bool log = !a[OVS_FLOW_ATTR_PROBE];
1308         bool ufid_present;
1309
1310         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1311         if (a[OVS_FLOW_ATTR_KEY]) {
1312                 ovs_match_init(&match, &key, true, NULL);
1313                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1314                                         NULL, log);
1315                 if (unlikely(err))
1316                         return err;
1317         }
1318
1319         ovs_lock();
1320         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1321         if (unlikely(!dp)) {
1322                 err = -ENODEV;
1323                 goto unlock;
1324         }
1325
1326         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1327                 err = ovs_flow_tbl_flush(&dp->table);
1328                 goto unlock;
1329         }
1330
1331         if (ufid_present)
1332                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1333         else
1334                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1335         if (unlikely(!flow)) {
1336                 err = -ENOENT;
1337                 goto unlock;
1338         }
1339
1340         ovs_flow_tbl_remove(&dp->table, flow);
1341         ovs_unlock();
1342
1343         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1344                                         &flow->id, info, false, ufid_flags);
1345         if (likely(reply)) {
1346                 if (likely(!IS_ERR(reply))) {
1347                         rcu_read_lock();        /*To keep RCU checker happy. */
1348                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1349                                                      reply, info->snd_portid,
1350                                                      info->snd_seq, 0,
1351                                                      OVS_FLOW_CMD_DEL,
1352                                                      ufid_flags);
1353                         rcu_read_unlock();
1354                         BUG_ON(err < 0);
1355
1356                         ovs_notify(&dp_flow_genl_family, reply, info);
1357                 } else {
1358                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1359                 }
1360         }
1361
1362         ovs_flow_free(flow, true);
1363         return 0;
1364 unlock:
1365         ovs_unlock();
1366         return err;
1367 }
1368
1369 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1370 {
1371         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1372         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1373         struct table_instance *ti;
1374         struct datapath *dp;
1375         u32 ufid_flags;
1376         int err;
1377
1378         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1379                             OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1380         if (err)
1381                 return err;
1382         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1383
1384         rcu_read_lock();
1385         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1386         if (!dp) {
1387                 rcu_read_unlock();
1388                 return -ENODEV;
1389         }
1390
1391         ti = rcu_dereference(dp->table.ti);
1392         for (;;) {
1393                 struct sw_flow *flow;
1394                 u32 bucket, obj;
1395
1396                 bucket = cb->args[0];
1397                 obj = cb->args[1];
1398                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1399                 if (!flow)
1400                         break;
1401
1402                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1403                                            NETLINK_CB(cb->skb).portid,
1404                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1405                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1406                         break;
1407
1408                 cb->args[0] = bucket;
1409                 cb->args[1] = obj;
1410         }
1411         rcu_read_unlock();
1412         return skb->len;
1413 }
1414
1415 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1416         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1417         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1418         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1419         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1420         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1421         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1422         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1423 };
1424
1425 static const struct genl_ops dp_flow_genl_ops[] = {
1426         { .cmd = OVS_FLOW_CMD_NEW,
1427           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1428           .policy = flow_policy,
1429           .doit = ovs_flow_cmd_new
1430         },
1431         { .cmd = OVS_FLOW_CMD_DEL,
1432           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1433           .policy = flow_policy,
1434           .doit = ovs_flow_cmd_del
1435         },
1436         { .cmd = OVS_FLOW_CMD_GET,
1437           .flags = 0,               /* OK for unprivileged users. */
1438           .policy = flow_policy,
1439           .doit = ovs_flow_cmd_get,
1440           .dumpit = ovs_flow_cmd_dump
1441         },
1442         { .cmd = OVS_FLOW_CMD_SET,
1443           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1444           .policy = flow_policy,
1445           .doit = ovs_flow_cmd_set,
1446         },
1447 };
1448
1449 static struct genl_family dp_flow_genl_family __ro_after_init = {
1450         .hdrsize = sizeof(struct ovs_header),
1451         .name = OVS_FLOW_FAMILY,
1452         .version = OVS_FLOW_VERSION,
1453         .maxattr = OVS_FLOW_ATTR_MAX,
1454         .netnsok = true,
1455         .parallel_ops = true,
1456         .ops = dp_flow_genl_ops,
1457         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1458         .mcgrps = &ovs_dp_flow_multicast_group,
1459         .n_mcgrps = 1,
1460         .module = THIS_MODULE,
1461 };
1462
1463 static size_t ovs_dp_cmd_msg_size(void)
1464 {
1465         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1466
1467         msgsize += nla_total_size(IFNAMSIZ);
1468         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1469         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1470         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1471
1472         return msgsize;
1473 }
1474
1475 /* Called with ovs_mutex. */
1476 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1477                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1478 {
1479         struct ovs_header *ovs_header;
1480         struct ovs_dp_stats dp_stats;
1481         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1482         int err;
1483
1484         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1485                                    flags, cmd);
1486         if (!ovs_header)
1487                 goto error;
1488
1489         ovs_header->dp_ifindex = get_dpifindex(dp);
1490
1491         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1492         if (err)
1493                 goto nla_put_failure;
1494
1495         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1496         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1497                           &dp_stats, OVS_DP_ATTR_PAD))
1498                 goto nla_put_failure;
1499
1500         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1501                           sizeof(struct ovs_dp_megaflow_stats),
1502                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1503                 goto nla_put_failure;
1504
1505         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1506                 goto nla_put_failure;
1507
1508         genlmsg_end(skb, ovs_header);
1509         return 0;
1510
1511 nla_put_failure:
1512         genlmsg_cancel(skb, ovs_header);
1513 error:
1514         return -EMSGSIZE;
1515 }
1516
1517 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1518 {
1519         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1520 }
1521
1522 /* Called with rcu_read_lock or ovs_mutex. */
1523 static struct datapath *lookup_datapath(struct net *net,
1524                                         const struct ovs_header *ovs_header,
1525                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1526 {
1527         struct datapath *dp;
1528
1529         if (!a[OVS_DP_ATTR_NAME])
1530                 dp = get_dp(net, ovs_header->dp_ifindex);
1531         else {
1532                 struct vport *vport;
1533
1534                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1535                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1536         }
1537         return dp ? dp : ERR_PTR(-ENODEV);
1538 }
1539
1540 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1541 {
1542         struct datapath *dp;
1543
1544         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1545         if (IS_ERR(dp))
1546                 return;
1547
1548         WARN(dp->user_features, "Dropping previously announced user features\n");
1549         dp->user_features = 0;
1550 }
1551
1552 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1553 {
1554         if (a[OVS_DP_ATTR_USER_FEATURES])
1555                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1556 }
1557
1558 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1559 {
1560         struct nlattr **a = info->attrs;
1561         struct vport_parms parms;
1562         struct sk_buff *reply;
1563         struct datapath *dp;
1564         struct vport *vport;
1565         struct ovs_net *ovs_net;
1566         int err, i;
1567
1568         err = -EINVAL;
1569         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1570                 goto err;
1571
1572         reply = ovs_dp_cmd_alloc_info();
1573         if (!reply)
1574                 return -ENOMEM;
1575
1576         err = -ENOMEM;
1577         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1578         if (dp == NULL)
1579                 goto err_free_reply;
1580
1581         ovs_dp_set_net(dp, sock_net(skb->sk));
1582
1583         /* Allocate table. */
1584         err = ovs_flow_tbl_init(&dp->table);
1585         if (err)
1586                 goto err_free_dp;
1587
1588         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1589         if (!dp->stats_percpu) {
1590                 err = -ENOMEM;
1591                 goto err_destroy_table;
1592         }
1593
1594         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * 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         /* Set up our datapath device. */
1605         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1606         parms.type = OVS_VPORT_TYPE_INTERNAL;
1607         parms.options = NULL;
1608         parms.dp = dp;
1609         parms.port_no = OVSP_LOCAL;
1610         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1611
1612         ovs_dp_change(dp, a);
1613
1614         /* So far only local changes have been made, now need the lock. */
1615         ovs_lock();
1616
1617         vport = new_vport(&parms);
1618         if (IS_ERR(vport)) {
1619                 err = PTR_ERR(vport);
1620                 if (err == -EBUSY)
1621                         err = -EEXIST;
1622
1623                 if (err == -EEXIST) {
1624                         /* An outdated user space instance that does not understand
1625                          * the concept of user_features has attempted to create a new
1626                          * datapath and is likely to reuse it. Drop all user features.
1627                          */
1628                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1629                                 ovs_dp_reset_user_features(skb, info);
1630                 }
1631
1632                 goto err_destroy_ports_array;
1633         }
1634
1635         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1636                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1637         BUG_ON(err < 0);
1638
1639         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1640         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1641
1642         ovs_unlock();
1643
1644         ovs_notify(&dp_datapath_genl_family, reply, info);
1645         return 0;
1646
1647 err_destroy_ports_array:
1648         ovs_unlock();
1649         kfree(dp->ports);
1650 err_destroy_percpu:
1651         free_percpu(dp->stats_percpu);
1652 err_destroy_table:
1653         ovs_flow_tbl_destroy(&dp->table);
1654 err_free_dp:
1655         kfree(dp);
1656 err_free_reply:
1657         kfree_skb(reply);
1658 err:
1659         return err;
1660 }
1661
1662 /* Called with ovs_mutex. */
1663 static void __dp_destroy(struct datapath *dp)
1664 {
1665         int i;
1666
1667         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1668                 struct vport *vport;
1669                 struct hlist_node *n;
1670
1671                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1672                         if (vport->port_no != OVSP_LOCAL)
1673                                 ovs_dp_detach_port(vport);
1674         }
1675
1676         list_del_rcu(&dp->list_node);
1677
1678         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1679          * all ports in datapath are destroyed first before freeing datapath.
1680          */
1681         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1682
1683         /* RCU destroy the flow table */
1684         call_rcu(&dp->rcu, destroy_dp_rcu);
1685 }
1686
1687 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1688 {
1689         struct sk_buff *reply;
1690         struct datapath *dp;
1691         int err;
1692
1693         reply = ovs_dp_cmd_alloc_info();
1694         if (!reply)
1695                 return -ENOMEM;
1696
1697         ovs_lock();
1698         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1699         err = PTR_ERR(dp);
1700         if (IS_ERR(dp))
1701                 goto err_unlock_free;
1702
1703         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1704                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1705         BUG_ON(err < 0);
1706
1707         __dp_destroy(dp);
1708         ovs_unlock();
1709
1710         ovs_notify(&dp_datapath_genl_family, reply, info);
1711
1712         return 0;
1713
1714 err_unlock_free:
1715         ovs_unlock();
1716         kfree_skb(reply);
1717         return err;
1718 }
1719
1720 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1721 {
1722         struct sk_buff *reply;
1723         struct datapath *dp;
1724         int err;
1725
1726         reply = ovs_dp_cmd_alloc_info();
1727         if (!reply)
1728                 return -ENOMEM;
1729
1730         ovs_lock();
1731         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1732         err = PTR_ERR(dp);
1733         if (IS_ERR(dp))
1734                 goto err_unlock_free;
1735
1736         ovs_dp_change(dp, info->attrs);
1737
1738         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1739                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1740         BUG_ON(err < 0);
1741
1742         ovs_unlock();
1743         ovs_notify(&dp_datapath_genl_family, reply, info);
1744
1745         return 0;
1746
1747 err_unlock_free:
1748         ovs_unlock();
1749         kfree_skb(reply);
1750         return err;
1751 }
1752
1753 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1754 {
1755         struct sk_buff *reply;
1756         struct datapath *dp;
1757         int err;
1758
1759         reply = ovs_dp_cmd_alloc_info();
1760         if (!reply)
1761                 return -ENOMEM;
1762
1763         ovs_lock();
1764         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1765         if (IS_ERR(dp)) {
1766                 err = PTR_ERR(dp);
1767                 goto err_unlock_free;
1768         }
1769         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1770                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1771         BUG_ON(err < 0);
1772         ovs_unlock();
1773
1774         return genlmsg_reply(reply, info);
1775
1776 err_unlock_free:
1777         ovs_unlock();
1778         kfree_skb(reply);
1779         return err;
1780 }
1781
1782 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1783 {
1784         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1785         struct datapath *dp;
1786         int skip = cb->args[0];
1787         int i = 0;
1788
1789         ovs_lock();
1790         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1791                 if (i >= skip &&
1792                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1793                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1794                                          OVS_DP_CMD_NEW) < 0)
1795                         break;
1796                 i++;
1797         }
1798         ovs_unlock();
1799
1800         cb->args[0] = i;
1801
1802         return skb->len;
1803 }
1804
1805 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1806         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1807         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1808         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1809 };
1810
1811 static const struct genl_ops dp_datapath_genl_ops[] = {
1812         { .cmd = OVS_DP_CMD_NEW,
1813           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1814           .policy = datapath_policy,
1815           .doit = ovs_dp_cmd_new
1816         },
1817         { .cmd = OVS_DP_CMD_DEL,
1818           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1819           .policy = datapath_policy,
1820           .doit = ovs_dp_cmd_del
1821         },
1822         { .cmd = OVS_DP_CMD_GET,
1823           .flags = 0,               /* OK for unprivileged users. */
1824           .policy = datapath_policy,
1825           .doit = ovs_dp_cmd_get,
1826           .dumpit = ovs_dp_cmd_dump
1827         },
1828         { .cmd = OVS_DP_CMD_SET,
1829           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1830           .policy = datapath_policy,
1831           .doit = ovs_dp_cmd_set,
1832         },
1833 };
1834
1835 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1836         .hdrsize = sizeof(struct ovs_header),
1837         .name = OVS_DATAPATH_FAMILY,
1838         .version = OVS_DATAPATH_VERSION,
1839         .maxattr = OVS_DP_ATTR_MAX,
1840         .netnsok = true,
1841         .parallel_ops = true,
1842         .ops = dp_datapath_genl_ops,
1843         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1844         .mcgrps = &ovs_dp_datapath_multicast_group,
1845         .n_mcgrps = 1,
1846         .module = THIS_MODULE,
1847 };
1848
1849 /* Called with ovs_mutex or RCU read lock. */
1850 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1851                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1852 {
1853         struct ovs_header *ovs_header;
1854         struct ovs_vport_stats vport_stats;
1855         int err;
1856
1857         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1858                                  flags, cmd);
1859         if (!ovs_header)
1860                 return -EMSGSIZE;
1861
1862         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1863
1864         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1865             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1866             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1867                            ovs_vport_name(vport)))
1868                 goto nla_put_failure;
1869
1870         ovs_vport_get_stats(vport, &vport_stats);
1871         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1872                           sizeof(struct ovs_vport_stats), &vport_stats,
1873                           OVS_VPORT_ATTR_PAD))
1874                 goto nla_put_failure;
1875
1876         if (ovs_vport_get_upcall_portids(vport, skb))
1877                 goto nla_put_failure;
1878
1879         err = ovs_vport_get_options(vport, skb);
1880         if (err == -EMSGSIZE)
1881                 goto error;
1882
1883         genlmsg_end(skb, ovs_header);
1884         return 0;
1885
1886 nla_put_failure:
1887         err = -EMSGSIZE;
1888 error:
1889         genlmsg_cancel(skb, ovs_header);
1890         return err;
1891 }
1892
1893 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1894 {
1895         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1896 }
1897
1898 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1899 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1900                                          u32 seq, u8 cmd)
1901 {
1902         struct sk_buff *skb;
1903         int retval;
1904
1905         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1906         if (!skb)
1907                 return ERR_PTR(-ENOMEM);
1908
1909         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1910         BUG_ON(retval < 0);
1911
1912         return skb;
1913 }
1914
1915 /* Called with ovs_mutex or RCU read lock. */
1916 static struct vport *lookup_vport(struct net *net,
1917                                   const struct ovs_header *ovs_header,
1918                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1919 {
1920         struct datapath *dp;
1921         struct vport *vport;
1922
1923         if (a[OVS_VPORT_ATTR_NAME]) {
1924                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1925                 if (!vport)
1926                         return ERR_PTR(-ENODEV);
1927                 if (ovs_header->dp_ifindex &&
1928                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1929                         return ERR_PTR(-ENODEV);
1930                 return vport;
1931         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1932                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1933
1934                 if (port_no >= DP_MAX_PORTS)
1935                         return ERR_PTR(-EFBIG);
1936
1937                 dp = get_dp(net, ovs_header->dp_ifindex);
1938                 if (!dp)
1939                         return ERR_PTR(-ENODEV);
1940
1941                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1942                 if (!vport)
1943                         return ERR_PTR(-ENODEV);
1944                 return vport;
1945         } else
1946                 return ERR_PTR(-EINVAL);
1947 }
1948
1949 /* Called with ovs_mutex */
1950 static void update_headroom(struct datapath *dp)
1951 {
1952         unsigned dev_headroom, max_headroom = 0;
1953         struct net_device *dev;
1954         struct vport *vport;
1955         int i;
1956
1957         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1958                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1959                         dev = vport->dev;
1960                         dev_headroom = netdev_get_fwd_headroom(dev);
1961                         if (dev_headroom > max_headroom)
1962                                 max_headroom = dev_headroom;
1963                 }
1964         }
1965
1966         dp->max_headroom = max_headroom;
1967         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1968                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1969                         netdev_set_rx_headroom(vport->dev, max_headroom);
1970 }
1971
1972 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1973 {
1974         struct nlattr **a = info->attrs;
1975         struct ovs_header *ovs_header = info->userhdr;
1976         struct vport_parms parms;
1977         struct sk_buff *reply;
1978         struct vport *vport;
1979         struct datapath *dp;
1980         u32 port_no;
1981         int err;
1982
1983         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1984             !a[OVS_VPORT_ATTR_UPCALL_PID])
1985                 return -EINVAL;
1986
1987         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1988                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1989         if (port_no >= DP_MAX_PORTS)
1990                 return -EFBIG;
1991
1992         reply = ovs_vport_cmd_alloc_info();
1993         if (!reply)
1994                 return -ENOMEM;
1995
1996         ovs_lock();
1997 restart:
1998         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1999         err = -ENODEV;
2000         if (!dp)
2001                 goto exit_unlock_free;
2002
2003         if (port_no) {
2004                 vport = ovs_vport_ovsl(dp, port_no);
2005                 err = -EBUSY;
2006                 if (vport)
2007                         goto exit_unlock_free;
2008         } else {
2009                 for (port_no = 1; ; port_no++) {
2010                         if (port_no >= DP_MAX_PORTS) {
2011                                 err = -EFBIG;
2012                                 goto exit_unlock_free;
2013                         }
2014                         vport = ovs_vport_ovsl(dp, port_no);
2015                         if (!vport)
2016                                 break;
2017                 }
2018         }
2019
2020         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2021         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2022         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2023         parms.dp = dp;
2024         parms.port_no = port_no;
2025         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2026
2027         vport = new_vport(&parms);
2028         err = PTR_ERR(vport);
2029         if (IS_ERR(vport)) {
2030                 if (err == -EAGAIN)
2031                         goto restart;
2032                 goto exit_unlock_free;
2033         }
2034
2035         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2036                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2037
2038         if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2039                 update_headroom(dp);
2040         else
2041                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2042
2043         BUG_ON(err < 0);
2044         ovs_unlock();
2045
2046         ovs_notify(&dp_vport_genl_family, reply, info);
2047         return 0;
2048
2049 exit_unlock_free:
2050         ovs_unlock();
2051         kfree_skb(reply);
2052         return err;
2053 }
2054
2055 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2056 {
2057         struct nlattr **a = info->attrs;
2058         struct sk_buff *reply;
2059         struct vport *vport;
2060         int err;
2061
2062         reply = ovs_vport_cmd_alloc_info();
2063         if (!reply)
2064                 return -ENOMEM;
2065
2066         ovs_lock();
2067         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2068         err = PTR_ERR(vport);
2069         if (IS_ERR(vport))
2070                 goto exit_unlock_free;
2071
2072         if (a[OVS_VPORT_ATTR_TYPE] &&
2073             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2074                 err = -EINVAL;
2075                 goto exit_unlock_free;
2076         }
2077
2078         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2079                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2080                 if (err)
2081                         goto exit_unlock_free;
2082         }
2083
2084
2085         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2086                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2087
2088                 err = ovs_vport_set_upcall_portids(vport, ids);
2089                 if (err)
2090                         goto exit_unlock_free;
2091         }
2092
2093         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2094                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2095         BUG_ON(err < 0);
2096
2097         ovs_unlock();
2098         ovs_notify(&dp_vport_genl_family, reply, info);
2099         return 0;
2100
2101 exit_unlock_free:
2102         ovs_unlock();
2103         kfree_skb(reply);
2104         return err;
2105 }
2106
2107 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2108 {
2109         bool must_update_headroom = false;
2110         struct nlattr **a = info->attrs;
2111         struct sk_buff *reply;
2112         struct datapath *dp;
2113         struct vport *vport;
2114         int err;
2115
2116         reply = ovs_vport_cmd_alloc_info();
2117         if (!reply)
2118                 return -ENOMEM;
2119
2120         ovs_lock();
2121         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2122         err = PTR_ERR(vport);
2123         if (IS_ERR(vport))
2124                 goto exit_unlock_free;
2125
2126         if (vport->port_no == OVSP_LOCAL) {
2127                 err = -EINVAL;
2128                 goto exit_unlock_free;
2129         }
2130
2131         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2132                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2133         BUG_ON(err < 0);
2134
2135         /* the vport deletion may trigger dp headroom update */
2136         dp = vport->dp;
2137         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2138                 must_update_headroom = true;
2139         netdev_reset_rx_headroom(vport->dev);
2140         ovs_dp_detach_port(vport);
2141
2142         if (must_update_headroom)
2143                 update_headroom(dp);
2144         ovs_unlock();
2145
2146         ovs_notify(&dp_vport_genl_family, reply, info);
2147         return 0;
2148
2149 exit_unlock_free:
2150         ovs_unlock();
2151         kfree_skb(reply);
2152         return err;
2153 }
2154
2155 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2156 {
2157         struct nlattr **a = info->attrs;
2158         struct ovs_header *ovs_header = info->userhdr;
2159         struct sk_buff *reply;
2160         struct vport *vport;
2161         int err;
2162
2163         reply = ovs_vport_cmd_alloc_info();
2164         if (!reply)
2165                 return -ENOMEM;
2166
2167         rcu_read_lock();
2168         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2169         err = PTR_ERR(vport);
2170         if (IS_ERR(vport))
2171                 goto exit_unlock_free;
2172         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2173                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2174         BUG_ON(err < 0);
2175         rcu_read_unlock();
2176
2177         return genlmsg_reply(reply, info);
2178
2179 exit_unlock_free:
2180         rcu_read_unlock();
2181         kfree_skb(reply);
2182         return err;
2183 }
2184
2185 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2186 {
2187         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2188         struct datapath *dp;
2189         int bucket = cb->args[0], skip = cb->args[1];
2190         int i, j = 0;
2191
2192         rcu_read_lock();
2193         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2194         if (!dp) {
2195                 rcu_read_unlock();
2196                 return -ENODEV;
2197         }
2198         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2199                 struct vport *vport;
2200
2201                 j = 0;
2202                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2203                         if (j >= skip &&
2204                             ovs_vport_cmd_fill_info(vport, skb,
2205                                                     NETLINK_CB(cb->skb).portid,
2206                                                     cb->nlh->nlmsg_seq,
2207                                                     NLM_F_MULTI,
2208                                                     OVS_VPORT_CMD_NEW) < 0)
2209                                 goto out;
2210
2211                         j++;
2212                 }
2213                 skip = 0;
2214         }
2215 out:
2216         rcu_read_unlock();
2217
2218         cb->args[0] = i;
2219         cb->args[1] = j;
2220
2221         return skb->len;
2222 }
2223
2224 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2225         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2226         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2227         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2228         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2229         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2230         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2231 };
2232
2233 static const struct genl_ops dp_vport_genl_ops[] = {
2234         { .cmd = OVS_VPORT_CMD_NEW,
2235           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2236           .policy = vport_policy,
2237           .doit = ovs_vport_cmd_new
2238         },
2239         { .cmd = OVS_VPORT_CMD_DEL,
2240           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2241           .policy = vport_policy,
2242           .doit = ovs_vport_cmd_del
2243         },
2244         { .cmd = OVS_VPORT_CMD_GET,
2245           .flags = 0,               /* OK for unprivileged users. */
2246           .policy = vport_policy,
2247           .doit = ovs_vport_cmd_get,
2248           .dumpit = ovs_vport_cmd_dump
2249         },
2250         { .cmd = OVS_VPORT_CMD_SET,
2251           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2252           .policy = vport_policy,
2253           .doit = ovs_vport_cmd_set,
2254         },
2255 };
2256
2257 struct genl_family dp_vport_genl_family __ro_after_init = {
2258         .hdrsize = sizeof(struct ovs_header),
2259         .name = OVS_VPORT_FAMILY,
2260         .version = OVS_VPORT_VERSION,
2261         .maxattr = OVS_VPORT_ATTR_MAX,
2262         .netnsok = true,
2263         .parallel_ops = true,
2264         .ops = dp_vport_genl_ops,
2265         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2266         .mcgrps = &ovs_dp_vport_multicast_group,
2267         .n_mcgrps = 1,
2268         .module = THIS_MODULE,
2269 };
2270
2271 static struct genl_family * const dp_genl_families[] = {
2272         &dp_datapath_genl_family,
2273         &dp_vport_genl_family,
2274         &dp_flow_genl_family,
2275         &dp_packet_genl_family,
2276 };
2277
2278 static void dp_unregister_genl(int n_families)
2279 {
2280         int i;
2281
2282         for (i = 0; i < n_families; i++)
2283                 genl_unregister_family(dp_genl_families[i]);
2284 }
2285
2286 static int __init dp_register_genl(void)
2287 {
2288         int err;
2289         int i;
2290
2291         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2292
2293                 err = genl_register_family(dp_genl_families[i]);
2294                 if (err)
2295                         goto error;
2296         }
2297
2298         return 0;
2299
2300 error:
2301         dp_unregister_genl(i);
2302         return err;
2303 }
2304
2305 static int __net_init ovs_init_net(struct net *net)
2306 {
2307         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2308
2309         INIT_LIST_HEAD(&ovs_net->dps);
2310         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2311         ovs_ct_init(net);
2312         return 0;
2313 }
2314
2315 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2316                                             struct list_head *head)
2317 {
2318         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2319         struct datapath *dp;
2320
2321         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2322                 int i;
2323
2324                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2325                         struct vport *vport;
2326
2327                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2328                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2329                                         continue;
2330
2331                                 if (dev_net(vport->dev) == dnet)
2332                                         list_add(&vport->detach_list, head);
2333                         }
2334                 }
2335         }
2336 }
2337
2338 static void __net_exit ovs_exit_net(struct net *dnet)
2339 {
2340         struct datapath *dp, *dp_next;
2341         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2342         struct vport *vport, *vport_next;
2343         struct net *net;
2344         LIST_HEAD(head);
2345
2346         ovs_ct_exit(dnet);
2347         ovs_lock();
2348         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2349                 __dp_destroy(dp);
2350
2351         rtnl_lock();
2352         for_each_net(net)
2353                 list_vports_from_net(net, dnet, &head);
2354         rtnl_unlock();
2355
2356         /* Detach all vports from given namespace. */
2357         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2358                 list_del(&vport->detach_list);
2359                 ovs_dp_detach_port(vport);
2360         }
2361
2362         ovs_unlock();
2363
2364         cancel_work_sync(&ovs_net->dp_notify_work);
2365 }
2366
2367 static struct pernet_operations ovs_net_ops = {
2368         .init = ovs_init_net,
2369         .exit = ovs_exit_net,
2370         .id   = &ovs_net_id,
2371         .size = sizeof(struct ovs_net),
2372 };
2373
2374 static int __init dp_init(void)
2375 {
2376         int err;
2377
2378         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2379
2380         pr_info("Open vSwitch switching datapath\n");
2381
2382         err = action_fifos_init();
2383         if (err)
2384                 goto error;
2385
2386         err = ovs_internal_dev_rtnl_link_register();
2387         if (err)
2388                 goto error_action_fifos_exit;
2389
2390         err = ovs_flow_init();
2391         if (err)
2392                 goto error_unreg_rtnl_link;
2393
2394         err = ovs_vport_init();
2395         if (err)
2396                 goto error_flow_exit;
2397
2398         err = register_pernet_device(&ovs_net_ops);
2399         if (err)
2400                 goto error_vport_exit;
2401
2402         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2403         if (err)
2404                 goto error_netns_exit;
2405
2406         err = ovs_netdev_init();
2407         if (err)
2408                 goto error_unreg_notifier;
2409
2410         err = dp_register_genl();
2411         if (err < 0)
2412                 goto error_unreg_netdev;
2413
2414         return 0;
2415
2416 error_unreg_netdev:
2417         ovs_netdev_exit();
2418 error_unreg_notifier:
2419         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2420 error_netns_exit:
2421         unregister_pernet_device(&ovs_net_ops);
2422 error_vport_exit:
2423         ovs_vport_exit();
2424 error_flow_exit:
2425         ovs_flow_exit();
2426 error_unreg_rtnl_link:
2427         ovs_internal_dev_rtnl_link_unregister();
2428 error_action_fifos_exit:
2429         action_fifos_exit();
2430 error:
2431         return err;
2432 }
2433
2434 static void dp_cleanup(void)
2435 {
2436         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2437         ovs_netdev_exit();
2438         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2439         unregister_pernet_device(&ovs_net_ops);
2440         rcu_barrier();
2441         ovs_vport_exit();
2442         ovs_flow_exit();
2443         ovs_internal_dev_rtnl_link_unregister();
2444         action_fifos_exit();
2445 }
2446
2447 module_init(dp_init);
2448 module_exit(dp_cleanup);
2449
2450 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2451 MODULE_LICENSE("GPL");
2452 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2453 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2454 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2455 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);