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         unsigned int gso_type = skb_shinfo(skb)->gso_type;
 339         struct sw_flow_key later_key;
 340         struct sk_buff *segs, *nskb;
 341         int err;
 342
 343         BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
 344         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 345         if (IS_ERR(segs))
 346                 return PTR_ERR(segs);
 347         if (segs == NULL)
 348                 return -EINVAL;
 349
 350         if (gso_type & SKB_GSO_UDP) {
 351                 /* The initial flow key extracted by ovs_flow_key_extract()
 352                  * in this case is for a first fragment, so we need to
 353                  * properly mark later fragments.
 354                  */
 355                 later_key = *key;
 356                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 357         }
 358
 359         /* Queue all of the segments. */
 360         skb = segs;
 361         do {
 362                 if (gso_type & SKB_GSO_UDP && skb != segs)
 363                         key = &later_key;
 364
 365                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 366                 if (err)
 367                         break;
 368
 369         } while ((skb = skb->next));
 370
 371         /* Free all of the segments. */
 372         skb = segs;
 373         do {
 374                 nskb = skb->next;
 375                 if (err)
 376                         kfree_skb(skb);
 377                 else
 378                         consume_skb(skb);
 379         } while ((skb = nskb));
 380         return err;
 381 }
 382
 383 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 384                               unsigned int hdrlen, int actions_attrlen)
 385 {
 386         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 387                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 388                 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 389                 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
 390
 391         /* OVS_PACKET_ATTR_USERDATA */
 392         if (upcall_info->userdata)
 393                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 394
 395         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 396         if (upcall_info->egress_tun_info)
 397                 size += nla_total_size(ovs_tun_key_attr_size());
 398
 399         /* OVS_PACKET_ATTR_ACTIONS */
 400         if (upcall_info->actions_len)
 401                 size += nla_total_size(actions_attrlen);
 402
 403         /* OVS_PACKET_ATTR_MRU */
 404         if (upcall_info->mru)
 405                 size += nla_total_size(sizeof(upcall_info->mru));
 406
 407         return size;
 408 }
 409
 410 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 411 {
 412         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 413                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 414
 415                 if (plen > 0)
 416                         skb_put_zero(skb, plen);
 417         }
 418 }
 419
 420 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 421                                   const struct sw_flow_key *key,
 422                                   const struct dp_upcall_info *upcall_info,
 423                                   uint32_t cutlen)
 424 {
 425         struct ovs_header *upcall;
 426         struct sk_buff *nskb = NULL;
 427         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 428         struct nlattr *nla;
 429         size_t len;
 430         unsigned int hlen;
 431         int err, dp_ifindex;
 432
 433         dp_ifindex = get_dpifindex(dp);
 434         if (!dp_ifindex)
 435                 return -ENODEV;
 436
 437         if (skb_vlan_tag_present(skb)) {
 438                 nskb = skb_clone(skb, GFP_ATOMIC);
 439                 if (!nskb)
 440                         return -ENOMEM;
 441
 442                 nskb = __vlan_hwaccel_push_inside(nskb);
 443                 if (!nskb)
 444                         return -ENOMEM;
 445
 446                 skb = nskb;
 447         }
 448
 449         if (nla_attr_size(skb->len) > USHRT_MAX) {
 450                 err = -EFBIG;
 451                 goto out;
 452         }
 453
 454         /* Complete checksum if needed */
 455         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 456             (err = skb_csum_hwoffload_help(skb, 0)))
 457                 goto out;
 458
 459         /* Older versions of OVS user space enforce alignment of the last
 460          * Netlink attribute to NLA_ALIGNTO which would require extensive
 461          * padding logic. Only perform zerocopy if padding is not required.
 462          */
 463         if (dp->user_features & OVS_DP_F_UNALIGNED)
 464                 hlen = skb_zerocopy_headlen(skb);
 465         else
 466                 hlen = skb->len;
 467
 468         len = upcall_msg_size(upcall_info, hlen - cutlen,
 469                               OVS_CB(skb)->acts_origlen);
 470         user_skb = genlmsg_new(len, GFP_ATOMIC);
 471         if (!user_skb) {
 472                 err = -ENOMEM;
 473                 goto out;
 474         }
 475
 476         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 477                              0, upcall_info->cmd);
 478         upcall->dp_ifindex = dp_ifindex;
 479
 480         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 481         BUG_ON(err);
 482
 483         if (upcall_info->userdata)
 484                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 485                           nla_len(upcall_info->userdata),
 486                           nla_data(upcall_info->userdata));
 487
 488         if (upcall_info->egress_tun_info) {
 489                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 490                 err = ovs_nla_put_tunnel_info(user_skb,
 491                                               upcall_info->egress_tun_info);
 492                 BUG_ON(err);
 493                 nla_nest_end(user_skb, nla);
 494         }
 495
 496         if (upcall_info->actions_len) {
 497                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 498                 err = ovs_nla_put_actions(upcall_info->actions,
 499                                           upcall_info->actions_len,
 500                                           user_skb);
 501                 if (!err)
 502                         nla_nest_end(user_skb, nla);
 503                 else
 504                         nla_nest_cancel(user_skb, nla);
 505         }
 506
 507         /* Add OVS_PACKET_ATTR_MRU */
 508         if (upcall_info->mru) {
 509                 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
 510                                 upcall_info->mru)) {
 511                         err = -ENOBUFS;
 512                         goto out;
 513                 }
 514                 pad_packet(dp, user_skb);
 515         }
 516
 517         /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 518         if (cutlen > 0) {
 519                 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
 520                                 skb->len)) {
 521                         err = -ENOBUFS;
 522                         goto out;
 523                 }
 524                 pad_packet(dp, user_skb);
 525         }
 526
 527         /* Only reserve room for attribute header, packet data is added
 528          * in skb_zerocopy() */
 529         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 530                 err = -ENOBUFS;
 531                 goto out;
 532         }
 533         nla->nla_len = nla_attr_size(skb->len - cutlen);
 534
 535         err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 536         if (err)
 537                 goto out;
 538
 539         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 540         pad_packet(dp, user_skb);
 541
 542         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 543
 544         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 545         user_skb = NULL;
 546 out:
 547         if (err)
 548                 skb_tx_error(skb);
 549         kfree_skb(user_skb);
 550         kfree_skb(nskb);
 551         return err;
 552 }
 553
 554 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 555 {
 556         struct ovs_header *ovs_header = info->userhdr;
 557         struct net *net = sock_net(skb->sk);
 558         struct nlattr **a = info->attrs;
 559         struct sw_flow_actions *acts;
 560         struct sk_buff *packet;
 561         struct sw_flow *flow;
 562         struct sw_flow_actions *sf_acts;
 563         struct datapath *dp;
 564         struct vport *input_vport;
 565         u16 mru = 0;
 566         int len;
 567         int err;
 568         bool log = !a[OVS_PACKET_ATTR_PROBE];
 569
 570         err = -EINVAL;
 571         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 572             !a[OVS_PACKET_ATTR_ACTIONS])
 573                 goto err;
 574
 575         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 576         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 577         err = -ENOMEM;
 578         if (!packet)
 579                 goto err;
 580         skb_reserve(packet, NET_IP_ALIGN);
 581
 582         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 583
 584         /* Set packet's mru */
 585         if (a[OVS_PACKET_ATTR_MRU]) {
 586                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 587                 packet->ignore_df = 1;
 588         }
 589         OVS_CB(packet)->mru = mru;
 590
 591         /* Build an sw_flow for sending this packet. */
 592         flow = ovs_flow_alloc();
 593         err = PTR_ERR(flow);
 594         if (IS_ERR(flow))
 595                 goto err_kfree_skb;
 596
 597         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 598                                              packet, &flow->key, log);
 599         if (err)
 600                 goto err_flow_free;
 601
 602         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 603                                    &flow->key, &acts, log);
 604         if (err)
 605                 goto err_flow_free;
 606
 607         rcu_assign_pointer(flow->sf_acts, acts);
 608         packet->priority = flow->key.phy.priority;
 609         packet->mark = flow->key.phy.skb_mark;
 610
 611         rcu_read_lock();
 612         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 613         err = -ENODEV;
 614         if (!dp)
 615                 goto err_unlock;
 616
 617         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 618         if (!input_vport)
 619                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 620
 621         if (!input_vport)
 622                 goto err_unlock;
 623
 624         packet->dev = input_vport->dev;
 625         OVS_CB(packet)->input_vport = input_vport;
 626         sf_acts = rcu_dereference(flow->sf_acts);
 627
 628         local_bh_disable();
 629         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 630         local_bh_enable();
 631         rcu_read_unlock();
 632
 633         ovs_flow_free(flow, false);
 634         return err;
 635
 636 err_unlock:
 637         rcu_read_unlock();
 638 err_flow_free:
 639         ovs_flow_free(flow, false);
 640 err_kfree_skb:
 641         kfree_skb(packet);
 642 err:
 643         return err;
 644 }
 645
 646 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 647         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 648         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 649         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 650         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 651         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 652 };
 653
 654 static const struct genl_ops dp_packet_genl_ops[] = {
 655         { .cmd = OVS_PACKET_CMD_EXECUTE,
 656           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 657           .policy = packet_policy,
 658           .doit = ovs_packet_cmd_execute
 659         }
 660 };
 661
 662 static struct genl_family dp_packet_genl_family __ro_after_init = {
 663         .hdrsize = sizeof(struct ovs_header),
 664         .name = OVS_PACKET_FAMILY,
 665         .version = OVS_PACKET_VERSION,
 666         .maxattr = OVS_PACKET_ATTR_MAX,
 667         .netnsok = true,
 668         .parallel_ops = true,
 669         .ops = dp_packet_genl_ops,
 670         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 671         .module = THIS_MODULE,
 672 };
 673
 674 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 675                          struct ovs_dp_megaflow_stats *mega_stats)
 676 {
 677         int i;
 678
 679         memset(mega_stats, 0, sizeof(*mega_stats));
 680
 681         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 682         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 683
 684         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 685
 686         for_each_possible_cpu(i) {
 687                 const struct dp_stats_percpu *percpu_stats;
 688                 struct dp_stats_percpu local_stats;
 689                 unsigned int start;
 690
 691                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 692
 693                 do {
 694                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 695                         local_stats = *percpu_stats;
 696                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 697
 698                 stats->n_hit += local_stats.n_hit;
 699                 stats->n_missed += local_stats.n_missed;
 700                 stats->n_lost += local_stats.n_lost;
 701                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 702         }
 703 }
 704
 705 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 706 {
 707         return ovs_identifier_is_ufid(sfid) &&
 708                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 709 }
 710
 711 static bool should_fill_mask(uint32_t ufid_flags)
 712 {
 713         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 714 }
 715
 716 static bool should_fill_actions(uint32_t ufid_flags)
 717 {
 718         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 719 }
 720
 721 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 722                                     const struct sw_flow_id *sfid,
 723                                     uint32_t ufid_flags)
 724 {
 725         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 726
 727         /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
 728          * see ovs_nla_put_identifier()
 729          */
 730         if (sfid && ovs_identifier_is_ufid(sfid))
 731                 len += nla_total_size(sfid->ufid_len);
 732         else
 733                 len += nla_total_size(ovs_key_attr_size());
 734
 735         /* OVS_FLOW_ATTR_KEY */
 736         if (!sfid || should_fill_key(sfid, ufid_flags))
 737                 len += nla_total_size(ovs_key_attr_size());
 738
 739         /* OVS_FLOW_ATTR_MASK */
 740         if (should_fill_mask(ufid_flags))
 741                 len += nla_total_size(ovs_key_attr_size());
 742
 743         /* OVS_FLOW_ATTR_ACTIONS */
 744         if (should_fill_actions(ufid_flags))
 745                 len += nla_total_size(acts->orig_len);
 746
 747         return len
 748                 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 749                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 750                 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 751 }
 752
 753 /* Called with ovs_mutex or RCU read lock. */
 754 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 755                                    struct sk_buff *skb)
 756 {
 757         struct ovs_flow_stats stats;
 758         __be16 tcp_flags;
 759         unsigned long used;
 760
 761         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 762
 763         if (used &&
 764             nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 765                               OVS_FLOW_ATTR_PAD))
 766                 return -EMSGSIZE;
 767
 768         if (stats.n_packets &&
 769             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 770                           sizeof(struct ovs_flow_stats), &stats,
 771                           OVS_FLOW_ATTR_PAD))
 772                 return -EMSGSIZE;
 773
 774         if ((u8)ntohs(tcp_flags) &&
 775              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 776                 return -EMSGSIZE;
 777
 778         return 0;
 779 }
 780
 781 /* Called with ovs_mutex or RCU read lock. */
 782 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 783                                      struct sk_buff *skb, int skb_orig_len)
 784 {
 785         struct nlattr *start;
 786         int err;
 787
 788         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 789          * this is the first flow to be dumped into 'skb'.  This is unusual for
 790          * Netlink but individual action lists can be longer than
 791          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 792          * The userspace caller can always fetch the actions separately if it
 793          * really wants them.  (Most userspace callers in fact don't care.)
 794          *
 795          * This can only fail for dump operations because the skb is always
 796          * properly sized for single flows.
 797          */
 798         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 799         if (start) {
 800                 const struct sw_flow_actions *sf_acts;
 801
 802                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 803                 err = ovs_nla_put_actions(sf_acts->actions,
 804                                           sf_acts->actions_len, skb);
 805
 806                 if (!err)
 807                         nla_nest_end(skb, start);
 808                 else {
 809                         if (skb_orig_len)
 810                                 return err;
 811
 812                         nla_nest_cancel(skb, start);
 813                 }
 814         } else if (skb_orig_len) {
 815                 return -EMSGSIZE;
 816         }
 817
 818         return 0;
 819 }
 820
 821 /* Called with ovs_mutex or RCU read lock. */
 822 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 823                                   struct sk_buff *skb, u32 portid,
 824                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 825 {
 826         const int skb_orig_len = skb->len;
 827         struct ovs_header *ovs_header;
 828         int err;
 829
 830         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 831                                  flags, cmd);
 832         if (!ovs_header)
 833                 return -EMSGSIZE;
 834
 835         ovs_header->dp_ifindex = dp_ifindex;
 836
 837         err = ovs_nla_put_identifier(flow, skb);
 838         if (err)
 839                 goto error;
 840
 841         if (should_fill_key(&flow->id, ufid_flags)) {
 842                 err = ovs_nla_put_masked_key(flow, skb);
 843                 if (err)
 844                         goto error;
 845         }
 846
 847         if (should_fill_mask(ufid_flags)) {
 848                 err = ovs_nla_put_mask(flow, skb);
 849                 if (err)
 850                         goto error;
 851         }
 852
 853         err = ovs_flow_cmd_fill_stats(flow, skb);
 854         if (err)
 855                 goto error;
 856
 857         if (should_fill_actions(ufid_flags)) {
 858                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 859                 if (err)
 860                         goto error;
 861         }
 862
 863         genlmsg_end(skb, ovs_header);
 864         return 0;
 865
 866 error:
 867         genlmsg_cancel(skb, ovs_header);
 868         return err;
 869 }
 870
 871 /* May not be called with RCU read lock. */
 872 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 873                                                const struct sw_flow_id *sfid,
 874                                                struct genl_info *info,
 875                                                bool always,
 876                                                uint32_t ufid_flags)
 877 {
 878         struct sk_buff *skb;
 879         size_t len;
 880
 881         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 882                 return NULL;
 883
 884         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 885         skb = genlmsg_new(len, GFP_KERNEL);
 886         if (!skb)
 887                 return ERR_PTR(-ENOMEM);
 888
 889         return skb;
 890 }
 891
 892 /* Called with ovs_mutex. */
 893 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 894                                                int dp_ifindex,
 895                                                struct genl_info *info, u8 cmd,
 896                                                bool always, u32 ufid_flags)
 897 {
 898         struct sk_buff *skb;
 899         int retval;
 900
 901         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 902                                       &flow->id, info, always, ufid_flags);
 903         if (IS_ERR_OR_NULL(skb))
 904                 return skb;
 905
 906         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 907                                         info->snd_portid, info->snd_seq, 0,
 908                                         cmd, ufid_flags);
 909         if (WARN_ON_ONCE(retval < 0)) {
 910                 kfree_skb(skb);
 911                 skb = ERR_PTR(retval);
 912         }
 913         return skb;
 914 }
 915
 916 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 917 {
 918         struct net *net = sock_net(skb->sk);
 919         struct nlattr **a = info->attrs;
 920         struct ovs_header *ovs_header = info->userhdr;
 921         struct sw_flow *flow = NULL, *new_flow;
 922         struct sw_flow_mask mask;
 923         struct sk_buff *reply;
 924         struct datapath *dp;
 925         struct sw_flow_actions *acts;
 926         struct sw_flow_match match;
 927         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 928         int error;
 929         bool log = !a[OVS_FLOW_ATTR_PROBE];
 930
 931         /* Must have key and actions. */
 932         error = -EINVAL;
 933         if (!a[OVS_FLOW_ATTR_KEY]) {
 934                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 935                 goto error;
 936         }
 937         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 938                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 939                 goto error;
 940         }
 941
 942         /* Most of the time we need to allocate a new flow, do it before
 943          * locking.
 944          */
 945         new_flow = ovs_flow_alloc();
 946         if (IS_ERR(new_flow)) {
 947                 error = PTR_ERR(new_flow);
 948                 goto error;
 949         }
 950
 951         /* Extract key. */
 952         ovs_match_init(&match, &new_flow->key, false, &mask);
 953         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 954                                   a[OVS_FLOW_ATTR_MASK], log);
 955         if (error)
 956                 goto err_kfree_flow;
 957
 958         /* Extract flow identifier. */
 959         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 960                                        &new_flow->key, log);
 961         if (error)
 962                 goto err_kfree_flow;
 963
 964         /* unmasked key is needed to match when ufid is not used. */
 965         if (ovs_identifier_is_key(&new_flow->id))
 966                 match.key = new_flow->id.unmasked_key;
 967
 968         ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
 969
 970         /* Validate actions. */
 971         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 972                                      &new_flow->key, &acts, log);
 973         if (error) {
 974                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 975                 goto err_kfree_flow;
 976         }
 977
 978         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 979                                         ufid_flags);
 980         if (IS_ERR(reply)) {
 981                 error = PTR_ERR(reply);
 982                 goto err_kfree_acts;
 983         }
 984
 985         ovs_lock();
 986         dp = get_dp(net, ovs_header->dp_ifindex);
 987         if (unlikely(!dp)) {
 988                 error = -ENODEV;
 989                 goto err_unlock_ovs;
 990         }
 991
 992         /* Check if this is a duplicate flow */
 993         if (ovs_identifier_is_ufid(&new_flow->id))
 994                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
 995         if (!flow)
 996                 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
 997         if (likely(!flow)) {
 998                 rcu_assign_pointer(new_flow->sf_acts, acts);
 999
1000                 /* Put flow in bucket. */
1001                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1002                 if (unlikely(error)) {
1003                         acts = NULL;
1004                         goto err_unlock_ovs;
1005                 }
1006
1007                 if (unlikely(reply)) {
1008                         error = ovs_flow_cmd_fill_info(new_flow,
1009                                                        ovs_header->dp_ifindex,
1010                                                        reply, info->snd_portid,
1011                                                        info->snd_seq, 0,
1012                                                        OVS_FLOW_CMD_NEW,
1013                                                        ufid_flags);
1014                         BUG_ON(error < 0);
1015                 }
1016                 ovs_unlock();
1017         } else {
1018                 struct sw_flow_actions *old_acts;
1019
1020                 /* Bail out if we're not allowed to modify an existing flow.
1021                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1022                  * because Generic Netlink treats the latter as a dump
1023                  * request.  We also accept NLM_F_EXCL in case that bug ever
1024                  * gets fixed.
1025                  */
1026                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1027                                                          | NLM_F_EXCL))) {
1028                         error = -EEXIST;
1029                         goto err_unlock_ovs;
1030                 }
1031                 /* The flow identifier has to be the same for flow updates.
1032                  * Look for any overlapping flow.
1033                  */
1034                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1035                         if (ovs_identifier_is_key(&flow->id))
1036                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1037                                                                  &match);
1038                         else /* UFID matches but key is different */
1039                                 flow = NULL;
1040                         if (!flow) {
1041                                 error = -ENOENT;
1042                                 goto err_unlock_ovs;
1043                         }
1044                 }
1045                 /* Update actions. */
1046                 old_acts = ovsl_dereference(flow->sf_acts);
1047                 rcu_assign_pointer(flow->sf_acts, acts);
1048
1049                 if (unlikely(reply)) {
1050                         error = ovs_flow_cmd_fill_info(flow,
1051                                                        ovs_header->dp_ifindex,
1052                                                        reply, info->snd_portid,
1053                                                        info->snd_seq, 0,
1054                                                        OVS_FLOW_CMD_NEW,
1055                                                        ufid_flags);
1056                         BUG_ON(error < 0);
1057                 }
1058                 ovs_unlock();
1059
1060                 ovs_nla_free_flow_actions_rcu(old_acts);
1061                 ovs_flow_free(new_flow, false);
1062         }
1063
1064         if (reply)
1065                 ovs_notify(&dp_flow_genl_family, reply, info);
1066         return 0;
1067
1068 err_unlock_ovs:
1069         ovs_unlock();
1070         kfree_skb(reply);
1071 err_kfree_acts:
1072         ovs_nla_free_flow_actions(acts);
1073 err_kfree_flow:
1074         ovs_flow_free(new_flow, false);
1075 error:
1076         return error;
1077 }
1078
1079 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1080 static struct sw_flow_actions *get_flow_actions(struct net *net,
1081                                                 const struct nlattr *a,
1082                                                 const struct sw_flow_key *key,
1083                                                 const struct sw_flow_mask *mask,
1084                                                 bool log)
1085 {
1086         struct sw_flow_actions *acts;
1087         struct sw_flow_key masked_key;
1088         int error;
1089
1090         ovs_flow_mask_key(&masked_key, key, true, mask);
1091         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1092         if (error) {
1093                 OVS_NLERR(log,
1094                           "Actions may not be safe on all matching packets");
1095                 return ERR_PTR(error);
1096         }
1097
1098         return acts;
1099 }
1100
1101 /* Factor out match-init and action-copy to avoid
1102  * "Wframe-larger-than=1024" warning. Because mask is only
1103  * used to get actions, we new a function to save some
1104  * stack space.
1105  *
1106  * If there are not key and action attrs, we return 0
1107  * directly. In the case, the caller will also not use the
1108  * match as before. If there is action attr, we try to get
1109  * actions and save them to *acts. Before returning from
1110  * the function, we reset the match->mask pointer. Because
1111  * we should not to return match object with dangling reference
1112  * to mask.
1113  * */
1114 static int ovs_nla_init_match_and_action(struct net *net,
1115                                          struct sw_flow_match *match,
1116                                          struct sw_flow_key *key,
1117                                          struct nlattr **a,
1118                                          struct sw_flow_actions **acts,
1119                                          bool log)
1120 {
1121         struct sw_flow_mask mask;
1122         int error = 0;
1123
1124         if (a[OVS_FLOW_ATTR_KEY]) {
1125                 ovs_match_init(match, key, true, &mask);
1126                 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1127                                           a[OVS_FLOW_ATTR_MASK], log);
1128                 if (error)
1129                         goto error;
1130         }
1131
1132         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1133                 if (!a[OVS_FLOW_ATTR_KEY]) {
1134                         OVS_NLERR(log,
1135                                   "Flow key attribute not present in set flow.");
1136                         error = -EINVAL;
1137                         goto error;
1138                 }
1139
1140                 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1141                                          &mask, log);
1142                 if (IS_ERR(*acts)) {
1143                         error = PTR_ERR(*acts);
1144                         goto error;
1145                 }
1146         }
1147
1148         /* On success, error is 0. */
1149 error:
1150         match->mask = NULL;
1151         return error;
1152 }
1153
1154 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1155 {
1156         struct net *net = sock_net(skb->sk);
1157         struct nlattr **a = info->attrs;
1158         struct ovs_header *ovs_header = info->userhdr;
1159         struct sw_flow_key key;
1160         struct sw_flow *flow;
1161         struct sk_buff *reply = NULL;
1162         struct datapath *dp;
1163         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1164         struct sw_flow_match match;
1165         struct sw_flow_id sfid;
1166         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1167         int error = 0;
1168         bool log = !a[OVS_FLOW_ATTR_PROBE];
1169         bool ufid_present;
1170
1171         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1172         if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1173                 OVS_NLERR(log,
1174                           "Flow set message rejected, Key attribute missing.");
1175                 return -EINVAL;
1176         }
1177
1178         error = ovs_nla_init_match_and_action(net, &match, &key, a,
1179                                               &acts, log);
1180         if (error)
1181                 goto error;
1182
1183         if (acts) {
1184                 /* Can allocate before locking if have acts. */
1185                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1186                                                 ufid_flags);
1187                 if (IS_ERR(reply)) {
1188                         error = PTR_ERR(reply);
1189                         goto err_kfree_acts;
1190                 }
1191         }
1192
1193         ovs_lock();
1194         dp = get_dp(net, ovs_header->dp_ifindex);
1195         if (unlikely(!dp)) {
1196                 error = -ENODEV;
1197                 goto err_unlock_ovs;
1198         }
1199         /* Check that the flow exists. */
1200         if (ufid_present)
1201                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1202         else
1203                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1204         if (unlikely(!flow)) {
1205                 error = -ENOENT;
1206                 goto err_unlock_ovs;
1207         }
1208
1209         /* Update actions, if present. */
1210         if (likely(acts)) {
1211                 old_acts = ovsl_dereference(flow->sf_acts);
1212                 rcu_assign_pointer(flow->sf_acts, acts);
1213
1214                 if (unlikely(reply)) {
1215                         error = ovs_flow_cmd_fill_info(flow,
1216                                                        ovs_header->dp_ifindex,
1217                                                        reply, info->snd_portid,
1218                                                        info->snd_seq, 0,
1219                                                        OVS_FLOW_CMD_NEW,
1220                                                        ufid_flags);
1221                         BUG_ON(error < 0);
1222                 }
1223         } else {
1224                 /* Could not alloc without acts before locking. */
1225                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1226                                                 info, OVS_FLOW_CMD_NEW, false,
1227                                                 ufid_flags);
1228
1229                 if (IS_ERR(reply)) {
1230                         error = PTR_ERR(reply);
1231                         goto err_unlock_ovs;
1232                 }
1233         }
1234
1235         /* Clear stats. */
1236         if (a[OVS_FLOW_ATTR_CLEAR])
1237                 ovs_flow_stats_clear(flow);
1238         ovs_unlock();
1239
1240         if (reply)
1241                 ovs_notify(&dp_flow_genl_family, reply, info);
1242         if (old_acts)
1243                 ovs_nla_free_flow_actions_rcu(old_acts);
1244
1245         return 0;
1246
1247 err_unlock_ovs:
1248         ovs_unlock();
1249         kfree_skb(reply);
1250 err_kfree_acts:
1251         ovs_nla_free_flow_actions(acts);
1252 error:
1253         return error;
1254 }
1255
1256 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1257 {
1258         struct nlattr **a = info->attrs;
1259         struct ovs_header *ovs_header = info->userhdr;
1260         struct net *net = sock_net(skb->sk);
1261         struct sw_flow_key key;
1262         struct sk_buff *reply;
1263         struct sw_flow *flow;
1264         struct datapath *dp;
1265         struct sw_flow_match match;
1266         struct sw_flow_id ufid;
1267         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1268         int err = 0;
1269         bool log = !a[OVS_FLOW_ATTR_PROBE];
1270         bool ufid_present;
1271
1272         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1273         if (a[OVS_FLOW_ATTR_KEY]) {
1274                 ovs_match_init(&match, &key, true, NULL);
1275                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1276                                         log);
1277         } else if (!ufid_present) {
1278                 OVS_NLERR(log,
1279                           "Flow get message rejected, Key attribute missing.");
1280                 err = -EINVAL;
1281         }
1282         if (err)
1283                 return err;
1284
1285         ovs_lock();
1286         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1287         if (!dp) {
1288                 err = -ENODEV;
1289                 goto unlock;
1290         }
1291
1292         if (ufid_present)
1293                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1294         else
1295                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1296         if (!flow) {
1297                 err = -ENOENT;
1298                 goto unlock;
1299         }
1300
1301         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1302                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1303         if (IS_ERR(reply)) {
1304                 err = PTR_ERR(reply);
1305                 goto unlock;
1306         }
1307
1308         ovs_unlock();
1309         return genlmsg_reply(reply, info);
1310 unlock:
1311         ovs_unlock();
1312         return err;
1313 }
1314
1315 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1316 {
1317         struct nlattr **a = info->attrs;
1318         struct ovs_header *ovs_header = info->userhdr;
1319         struct net *net = sock_net(skb->sk);
1320         struct sw_flow_key key;
1321         struct sk_buff *reply;
1322         struct sw_flow *flow = NULL;
1323         struct datapath *dp;
1324         struct sw_flow_match match;
1325         struct sw_flow_id ufid;
1326         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1327         int err;
1328         bool log = !a[OVS_FLOW_ATTR_PROBE];
1329         bool ufid_present;
1330
1331         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1332         if (a[OVS_FLOW_ATTR_KEY]) {
1333                 ovs_match_init(&match, &key, true, NULL);
1334                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1335                                         NULL, log);
1336                 if (unlikely(err))
1337                         return err;
1338         }
1339
1340         ovs_lock();
1341         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1342         if (unlikely(!dp)) {
1343                 err = -ENODEV;
1344                 goto unlock;
1345         }
1346
1347         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1348                 err = ovs_flow_tbl_flush(&dp->table);
1349                 goto unlock;
1350         }
1351
1352         if (ufid_present)
1353                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1354         else
1355                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1356         if (unlikely(!flow)) {
1357                 err = -ENOENT;
1358                 goto unlock;
1359         }
1360
1361         ovs_flow_tbl_remove(&dp->table, flow);
1362         ovs_unlock();
1363
1364         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1365                                         &flow->id, info, false, ufid_flags);
1366         if (likely(reply)) {
1367                 if (likely(!IS_ERR(reply))) {
1368                         rcu_read_lock();        /*To keep RCU checker happy. */
1369                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1370                                                      reply, info->snd_portid,
1371                                                      info->snd_seq, 0,
1372                                                      OVS_FLOW_CMD_DEL,
1373                                                      ufid_flags);
1374                         rcu_read_unlock();
1375                         if (WARN_ON_ONCE(err < 0)) {
1376                                 kfree_skb(reply);
1377                                 goto out_free;
1378                         }
1379
1380                         ovs_notify(&dp_flow_genl_family, reply, info);
1381                 } else {
1382                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1383                 }
1384         }
1385
1386 out_free:
1387         ovs_flow_free(flow, true);
1388         return 0;
1389 unlock:
1390         ovs_unlock();
1391         return err;
1392 }
1393
1394 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1395 {
1396         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1397         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1398         struct table_instance *ti;
1399         struct datapath *dp;
1400         u32 ufid_flags;
1401         int err;
1402
1403         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1404                             OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1405         if (err)
1406                 return err;
1407         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1408
1409         rcu_read_lock();
1410         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1411         if (!dp) {
1412                 rcu_read_unlock();
1413                 return -ENODEV;
1414         }
1415
1416         ti = rcu_dereference(dp->table.ti);
1417         for (;;) {
1418                 struct sw_flow *flow;
1419                 u32 bucket, obj;
1420
1421                 bucket = cb->args[0];
1422                 obj = cb->args[1];
1423                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1424                 if (!flow)
1425                         break;
1426
1427                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1428                                            NETLINK_CB(cb->skb).portid,
1429                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1430                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1431                         break;
1432
1433                 cb->args[0] = bucket;
1434                 cb->args[1] = obj;
1435         }
1436         rcu_read_unlock();
1437         return skb->len;
1438 }
1439
1440 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1441         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1442         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1443         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1444         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1445         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1446         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1447         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1448 };
1449
1450 static const struct genl_ops dp_flow_genl_ops[] = {
1451         { .cmd = OVS_FLOW_CMD_NEW,
1452           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1453           .policy = flow_policy,
1454           .doit = ovs_flow_cmd_new
1455         },
1456         { .cmd = OVS_FLOW_CMD_DEL,
1457           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1458           .policy = flow_policy,
1459           .doit = ovs_flow_cmd_del
1460         },
1461         { .cmd = OVS_FLOW_CMD_GET,
1462           .flags = 0,               /* OK for unprivileged users. */
1463           .policy = flow_policy,
1464           .doit = ovs_flow_cmd_get,
1465           .dumpit = ovs_flow_cmd_dump
1466         },
1467         { .cmd = OVS_FLOW_CMD_SET,
1468           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1469           .policy = flow_policy,
1470           .doit = ovs_flow_cmd_set,
1471         },
1472 };
1473
1474 static struct genl_family dp_flow_genl_family __ro_after_init = {
1475         .hdrsize = sizeof(struct ovs_header),
1476         .name = OVS_FLOW_FAMILY,
1477         .version = OVS_FLOW_VERSION,
1478         .maxattr = OVS_FLOW_ATTR_MAX,
1479         .netnsok = true,
1480         .parallel_ops = true,
1481         .ops = dp_flow_genl_ops,
1482         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1483         .mcgrps = &ovs_dp_flow_multicast_group,
1484         .n_mcgrps = 1,
1485         .module = THIS_MODULE,
1486 };
1487
1488 static size_t ovs_dp_cmd_msg_size(void)
1489 {
1490         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1491
1492         msgsize += nla_total_size(IFNAMSIZ);
1493         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1494         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1495         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1496
1497         return msgsize;
1498 }
1499
1500 /* Called with ovs_mutex. */
1501 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1502                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1503 {
1504         struct ovs_header *ovs_header;
1505         struct ovs_dp_stats dp_stats;
1506         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1507         int err;
1508
1509         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1510                                    flags, cmd);
1511         if (!ovs_header)
1512                 goto error;
1513
1514         ovs_header->dp_ifindex = get_dpifindex(dp);
1515
1516         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1517         if (err)
1518                 goto nla_put_failure;
1519
1520         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1521         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1522                           &dp_stats, OVS_DP_ATTR_PAD))
1523                 goto nla_put_failure;
1524
1525         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1526                           sizeof(struct ovs_dp_megaflow_stats),
1527                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1528                 goto nla_put_failure;
1529
1530         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1531                 goto nla_put_failure;
1532
1533         genlmsg_end(skb, ovs_header);
1534         return 0;
1535
1536 nla_put_failure:
1537         genlmsg_cancel(skb, ovs_header);
1538 error:
1539         return -EMSGSIZE;
1540 }
1541
1542 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1543 {
1544         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1545 }
1546
1547 /* Called with rcu_read_lock or ovs_mutex. */
1548 static struct datapath *lookup_datapath(struct net *net,
1549                                         const struct ovs_header *ovs_header,
1550                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1551 {
1552         struct datapath *dp;
1553
1554         if (!a[OVS_DP_ATTR_NAME])
1555                 dp = get_dp(net, ovs_header->dp_ifindex);
1556         else {
1557                 struct vport *vport;
1558
1559                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1560                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1561         }
1562         return dp ? dp : ERR_PTR(-ENODEV);
1563 }
1564
1565 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1566 {
1567         struct datapath *dp;
1568
1569         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1570         if (IS_ERR(dp))
1571                 return;
1572
1573         WARN(dp->user_features, "Dropping previously announced user features\n");
1574         dp->user_features = 0;
1575 }
1576
1577 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1578 {
1579         if (a[OVS_DP_ATTR_USER_FEATURES])
1580                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1581 }
1582
1583 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1584 {
1585         struct nlattr **a = info->attrs;
1586         struct vport_parms parms;
1587         struct sk_buff *reply;
1588         struct datapath *dp;
1589         struct vport *vport;
1590         struct ovs_net *ovs_net;
1591         int err, i;
1592
1593         err = -EINVAL;
1594         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1595                 goto err;
1596
1597         reply = ovs_dp_cmd_alloc_info();
1598         if (!reply)
1599                 return -ENOMEM;
1600
1601         err = -ENOMEM;
1602         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1603         if (dp == NULL)
1604                 goto err_free_reply;
1605
1606         ovs_dp_set_net(dp, sock_net(skb->sk));
1607
1608         /* Allocate table. */
1609         err = ovs_flow_tbl_init(&dp->table);
1610         if (err)
1611                 goto err_free_dp;
1612
1613         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1614         if (!dp->stats_percpu) {
1615                 err = -ENOMEM;
1616                 goto err_destroy_table;
1617         }
1618
1619         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1620                             GFP_KERNEL);
1621         if (!dp->ports) {
1622                 err = -ENOMEM;
1623                 goto err_destroy_percpu;
1624         }
1625
1626         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1627                 INIT_HLIST_HEAD(&dp->ports[i]);
1628
1629         /* Set up our datapath device. */
1630         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1631         parms.type = OVS_VPORT_TYPE_INTERNAL;
1632         parms.options = NULL;
1633         parms.dp = dp;
1634         parms.port_no = OVSP_LOCAL;
1635         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1636
1637         ovs_dp_change(dp, a);
1638
1639         /* So far only local changes have been made, now need the lock. */
1640         ovs_lock();
1641
1642         vport = new_vport(&parms);
1643         if (IS_ERR(vport)) {
1644                 err = PTR_ERR(vport);
1645                 if (err == -EBUSY)
1646                         err = -EEXIST;
1647
1648                 if (err == -EEXIST) {
1649                         /* An outdated user space instance that does not understand
1650                          * the concept of user_features has attempted to create a new
1651                          * datapath and is likely to reuse it. Drop all user features.
1652                          */
1653                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1654                                 ovs_dp_reset_user_features(skb, info);
1655                 }
1656
1657                 goto err_destroy_ports_array;
1658         }
1659
1660         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1661                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1662         BUG_ON(err < 0);
1663
1664         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1665         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1666
1667         ovs_unlock();
1668
1669         ovs_notify(&dp_datapath_genl_family, reply, info);
1670         return 0;
1671
1672 err_destroy_ports_array:
1673         ovs_unlock();
1674         kfree(dp->ports);
1675 err_destroy_percpu:
1676         free_percpu(dp->stats_percpu);
1677 err_destroy_table:
1678         ovs_flow_tbl_destroy(&dp->table);
1679 err_free_dp:
1680         kfree(dp);
1681 err_free_reply:
1682         kfree_skb(reply);
1683 err:
1684         return err;
1685 }
1686
1687 /* Called with ovs_mutex. */
1688 static void __dp_destroy(struct datapath *dp)
1689 {
1690         int i;
1691
1692         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1693                 struct vport *vport;
1694                 struct hlist_node *n;
1695
1696                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1697                         if (vport->port_no != OVSP_LOCAL)
1698                                 ovs_dp_detach_port(vport);
1699         }
1700
1701         list_del_rcu(&dp->list_node);
1702
1703         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1704          * all ports in datapath are destroyed first before freeing datapath.
1705          */
1706         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1707
1708         /* RCU destroy the flow table */
1709         call_rcu(&dp->rcu, destroy_dp_rcu);
1710 }
1711
1712 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1713 {
1714         struct sk_buff *reply;
1715         struct datapath *dp;
1716         int err;
1717
1718         reply = ovs_dp_cmd_alloc_info();
1719         if (!reply)
1720                 return -ENOMEM;
1721
1722         ovs_lock();
1723         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1724         err = PTR_ERR(dp);
1725         if (IS_ERR(dp))
1726                 goto err_unlock_free;
1727
1728         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1729                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1730         BUG_ON(err < 0);
1731
1732         __dp_destroy(dp);
1733         ovs_unlock();
1734
1735         ovs_notify(&dp_datapath_genl_family, reply, info);
1736
1737         return 0;
1738
1739 err_unlock_free:
1740         ovs_unlock();
1741         kfree_skb(reply);
1742         return err;
1743 }
1744
1745 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1746 {
1747         struct sk_buff *reply;
1748         struct datapath *dp;
1749         int err;
1750
1751         reply = ovs_dp_cmd_alloc_info();
1752         if (!reply)
1753                 return -ENOMEM;
1754
1755         ovs_lock();
1756         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1757         err = PTR_ERR(dp);
1758         if (IS_ERR(dp))
1759                 goto err_unlock_free;
1760
1761         ovs_dp_change(dp, info->attrs);
1762
1763         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1764                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1765         BUG_ON(err < 0);
1766
1767         ovs_unlock();
1768         ovs_notify(&dp_datapath_genl_family, reply, info);
1769
1770         return 0;
1771
1772 err_unlock_free:
1773         ovs_unlock();
1774         kfree_skb(reply);
1775         return err;
1776 }
1777
1778 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1779 {
1780         struct sk_buff *reply;
1781         struct datapath *dp;
1782         int err;
1783
1784         reply = ovs_dp_cmd_alloc_info();
1785         if (!reply)
1786                 return -ENOMEM;
1787
1788         ovs_lock();
1789         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1790         if (IS_ERR(dp)) {
1791                 err = PTR_ERR(dp);
1792                 goto err_unlock_free;
1793         }
1794         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1795                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1796         BUG_ON(err < 0);
1797         ovs_unlock();
1798
1799         return genlmsg_reply(reply, info);
1800
1801 err_unlock_free:
1802         ovs_unlock();
1803         kfree_skb(reply);
1804         return err;
1805 }
1806
1807 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1808 {
1809         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1810         struct datapath *dp;
1811         int skip = cb->args[0];
1812         int i = 0;
1813
1814         ovs_lock();
1815         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1816                 if (i >= skip &&
1817                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1818                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1819                                          OVS_DP_CMD_NEW) < 0)
1820                         break;
1821                 i++;
1822         }
1823         ovs_unlock();
1824
1825         cb->args[0] = i;
1826
1827         return skb->len;
1828 }
1829
1830 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1831         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1832         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1833         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1834 };
1835
1836 static const struct genl_ops dp_datapath_genl_ops[] = {
1837         { .cmd = OVS_DP_CMD_NEW,
1838           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1839           .policy = datapath_policy,
1840           .doit = ovs_dp_cmd_new
1841         },
1842         { .cmd = OVS_DP_CMD_DEL,
1843           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1844           .policy = datapath_policy,
1845           .doit = ovs_dp_cmd_del
1846         },
1847         { .cmd = OVS_DP_CMD_GET,
1848           .flags = 0,               /* OK for unprivileged users. */
1849           .policy = datapath_policy,
1850           .doit = ovs_dp_cmd_get,
1851           .dumpit = ovs_dp_cmd_dump
1852         },
1853         { .cmd = OVS_DP_CMD_SET,
1854           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1855           .policy = datapath_policy,
1856           .doit = ovs_dp_cmd_set,
1857         },
1858 };
1859
1860 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1861         .hdrsize = sizeof(struct ovs_header),
1862         .name = OVS_DATAPATH_FAMILY,
1863         .version = OVS_DATAPATH_VERSION,
1864         .maxattr = OVS_DP_ATTR_MAX,
1865         .netnsok = true,
1866         .parallel_ops = true,
1867         .ops = dp_datapath_genl_ops,
1868         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1869         .mcgrps = &ovs_dp_datapath_multicast_group,
1870         .n_mcgrps = 1,
1871         .module = THIS_MODULE,
1872 };
1873
1874 /* Called with ovs_mutex or RCU read lock. */
1875 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1876                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1877 {
1878         struct ovs_header *ovs_header;
1879         struct ovs_vport_stats vport_stats;
1880         int err;
1881
1882         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1883                                  flags, cmd);
1884         if (!ovs_header)
1885                 return -EMSGSIZE;
1886
1887         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1888
1889         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1890             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1891             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1892                            ovs_vport_name(vport)))
1893                 goto nla_put_failure;
1894
1895         ovs_vport_get_stats(vport, &vport_stats);
1896         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1897                           sizeof(struct ovs_vport_stats), &vport_stats,
1898                           OVS_VPORT_ATTR_PAD))
1899                 goto nla_put_failure;
1900
1901         if (ovs_vport_get_upcall_portids(vport, skb))
1902                 goto nla_put_failure;
1903
1904         err = ovs_vport_get_options(vport, skb);
1905         if (err == -EMSGSIZE)
1906                 goto error;
1907
1908         genlmsg_end(skb, ovs_header);
1909         return 0;
1910
1911 nla_put_failure:
1912         err = -EMSGSIZE;
1913 error:
1914         genlmsg_cancel(skb, ovs_header);
1915         return err;
1916 }
1917
1918 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1919 {
1920         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1921 }
1922
1923 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1924 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1925                                          u32 seq, u8 cmd)
1926 {
1927         struct sk_buff *skb;
1928         int retval;
1929
1930         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1931         if (!skb)
1932                 return ERR_PTR(-ENOMEM);
1933
1934         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1935         BUG_ON(retval < 0);
1936
1937         return skb;
1938 }
1939
1940 /* Called with ovs_mutex or RCU read lock. */
1941 static struct vport *lookup_vport(struct net *net,
1942                                   const struct ovs_header *ovs_header,
1943                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1944 {
1945         struct datapath *dp;
1946         struct vport *vport;
1947
1948         if (a[OVS_VPORT_ATTR_NAME]) {
1949                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1950                 if (!vport)
1951                         return ERR_PTR(-ENODEV);
1952                 if (ovs_header->dp_ifindex &&
1953                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1954                         return ERR_PTR(-ENODEV);
1955                 return vport;
1956         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1957                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1958
1959                 if (port_no >= DP_MAX_PORTS)
1960                         return ERR_PTR(-EFBIG);
1961
1962                 dp = get_dp(net, ovs_header->dp_ifindex);
1963                 if (!dp)
1964                         return ERR_PTR(-ENODEV);
1965
1966                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1967                 if (!vport)
1968                         return ERR_PTR(-ENODEV);
1969                 return vport;
1970         } else
1971                 return ERR_PTR(-EINVAL);
1972 }
1973
1974 /* Called with ovs_mutex */
1975 static void update_headroom(struct datapath *dp)
1976 {
1977         unsigned dev_headroom, max_headroom = 0;
1978         struct net_device *dev;
1979         struct vport *vport;
1980         int i;
1981
1982         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1983                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1984                         dev = vport->dev;
1985                         dev_headroom = netdev_get_fwd_headroom(dev);
1986                         if (dev_headroom > max_headroom)
1987                                 max_headroom = dev_headroom;
1988                 }
1989         }
1990
1991         dp->max_headroom = max_headroom;
1992         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1993                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1994                         netdev_set_rx_headroom(vport->dev, max_headroom);
1995 }
1996
1997 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1998 {
1999         struct nlattr **a = info->attrs;
2000         struct ovs_header *ovs_header = info->userhdr;
2001         struct vport_parms parms;
2002         struct sk_buff *reply;
2003         struct vport *vport;
2004         struct datapath *dp;
2005         u32 port_no;
2006         int err;
2007
2008         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2009             !a[OVS_VPORT_ATTR_UPCALL_PID])
2010                 return -EINVAL;
2011
2012         port_no = a[OVS_VPORT_ATTR_PORT_NO]
2013                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2014         if (port_no >= DP_MAX_PORTS)
2015                 return -EFBIG;
2016
2017         reply = ovs_vport_cmd_alloc_info();
2018         if (!reply)
2019                 return -ENOMEM;
2020
2021         ovs_lock();
2022 restart:
2023         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2024         err = -ENODEV;
2025         if (!dp)
2026                 goto exit_unlock_free;
2027
2028         if (port_no) {
2029                 vport = ovs_vport_ovsl(dp, port_no);
2030                 err = -EBUSY;
2031                 if (vport)
2032                         goto exit_unlock_free;
2033         } else {
2034                 for (port_no = 1; ; port_no++) {
2035                         if (port_no >= DP_MAX_PORTS) {
2036                                 err = -EFBIG;
2037                                 goto exit_unlock_free;
2038                         }
2039                         vport = ovs_vport_ovsl(dp, port_no);
2040                         if (!vport)
2041                                 break;
2042                 }
2043         }
2044
2045         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2046         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2047         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2048         parms.dp = dp;
2049         parms.port_no = port_no;
2050         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2051
2052         vport = new_vport(&parms);
2053         err = PTR_ERR(vport);
2054         if (IS_ERR(vport)) {
2055                 if (err == -EAGAIN)
2056                         goto restart;
2057                 goto exit_unlock_free;
2058         }
2059
2060         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2061                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2062
2063         if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2064                 update_headroom(dp);
2065         else
2066                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2067
2068         BUG_ON(err < 0);
2069         ovs_unlock();
2070
2071         ovs_notify(&dp_vport_genl_family, reply, info);
2072         return 0;
2073
2074 exit_unlock_free:
2075         ovs_unlock();
2076         kfree_skb(reply);
2077         return err;
2078 }
2079
2080 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2081 {
2082         struct nlattr **a = info->attrs;
2083         struct sk_buff *reply;
2084         struct vport *vport;
2085         int err;
2086
2087         reply = ovs_vport_cmd_alloc_info();
2088         if (!reply)
2089                 return -ENOMEM;
2090
2091         ovs_lock();
2092         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2093         err = PTR_ERR(vport);
2094         if (IS_ERR(vport))
2095                 goto exit_unlock_free;
2096
2097         if (a[OVS_VPORT_ATTR_TYPE] &&
2098             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2099                 err = -EINVAL;
2100                 goto exit_unlock_free;
2101         }
2102
2103         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2104                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2105                 if (err)
2106                         goto exit_unlock_free;
2107         }
2108
2109
2110         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2111                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2112
2113                 err = ovs_vport_set_upcall_portids(vport, ids);
2114                 if (err)
2115                         goto exit_unlock_free;
2116         }
2117
2118         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2119                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2120         BUG_ON(err < 0);
2121
2122         ovs_unlock();
2123         ovs_notify(&dp_vport_genl_family, reply, info);
2124         return 0;
2125
2126 exit_unlock_free:
2127         ovs_unlock();
2128         kfree_skb(reply);
2129         return err;
2130 }
2131
2132 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2133 {
2134         bool must_update_headroom = false;
2135         struct nlattr **a = info->attrs;
2136         struct sk_buff *reply;
2137         struct datapath *dp;
2138         struct vport *vport;
2139         int err;
2140
2141         reply = ovs_vport_cmd_alloc_info();
2142         if (!reply)
2143                 return -ENOMEM;
2144
2145         ovs_lock();
2146         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2147         err = PTR_ERR(vport);
2148         if (IS_ERR(vport))
2149                 goto exit_unlock_free;
2150
2151         if (vport->port_no == OVSP_LOCAL) {
2152                 err = -EINVAL;
2153                 goto exit_unlock_free;
2154         }
2155
2156         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2157                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2158         BUG_ON(err < 0);
2159
2160         /* the vport deletion may trigger dp headroom update */
2161         dp = vport->dp;
2162         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2163                 must_update_headroom = true;
2164         netdev_reset_rx_headroom(vport->dev);
2165         ovs_dp_detach_port(vport);
2166
2167         if (must_update_headroom)
2168                 update_headroom(dp);
2169         ovs_unlock();
2170
2171         ovs_notify(&dp_vport_genl_family, reply, info);
2172         return 0;
2173
2174 exit_unlock_free:
2175         ovs_unlock();
2176         kfree_skb(reply);
2177         return err;
2178 }
2179
2180 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2181 {
2182         struct nlattr **a = info->attrs;
2183         struct ovs_header *ovs_header = info->userhdr;
2184         struct sk_buff *reply;
2185         struct vport *vport;
2186         int err;
2187
2188         reply = ovs_vport_cmd_alloc_info();
2189         if (!reply)
2190                 return -ENOMEM;
2191
2192         rcu_read_lock();
2193         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2194         err = PTR_ERR(vport);
2195         if (IS_ERR(vport))
2196                 goto exit_unlock_free;
2197         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2198                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2199         BUG_ON(err < 0);
2200         rcu_read_unlock();
2201
2202         return genlmsg_reply(reply, info);
2203
2204 exit_unlock_free:
2205         rcu_read_unlock();
2206         kfree_skb(reply);
2207         return err;
2208 }
2209
2210 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2211 {
2212         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2213         struct datapath *dp;
2214         int bucket = cb->args[0], skip = cb->args[1];
2215         int i, j = 0;
2216
2217         rcu_read_lock();
2218         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2219         if (!dp) {
2220                 rcu_read_unlock();
2221                 return -ENODEV;
2222         }
2223         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2224                 struct vport *vport;
2225
2226                 j = 0;
2227                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2228                         if (j >= skip &&
2229                             ovs_vport_cmd_fill_info(vport, skb,
2230                                                     NETLINK_CB(cb->skb).portid,
2231                                                     cb->nlh->nlmsg_seq,
2232                                                     NLM_F_MULTI,
2233                                                     OVS_VPORT_CMD_NEW) < 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_UNSPEC },
2255         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2256 };
2257
2258 static const struct genl_ops dp_vport_genl_ops[] = {
2259         { .cmd = OVS_VPORT_CMD_NEW,
2260           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2261           .policy = vport_policy,
2262           .doit = ovs_vport_cmd_new
2263         },
2264         { .cmd = OVS_VPORT_CMD_DEL,
2265           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2266           .policy = vport_policy,
2267           .doit = ovs_vport_cmd_del
2268         },
2269         { .cmd = OVS_VPORT_CMD_GET,
2270           .flags = 0,               /* OK for unprivileged users. */
2271           .policy = vport_policy,
2272           .doit = ovs_vport_cmd_get,
2273           .dumpit = ovs_vport_cmd_dump
2274         },
2275         { .cmd = OVS_VPORT_CMD_SET,
2276           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2277           .policy = vport_policy,
2278           .doit = ovs_vport_cmd_set,
2279         },
2280 };
2281
2282 struct genl_family dp_vport_genl_family __ro_after_init = {
2283         .hdrsize = sizeof(struct ovs_header),
2284         .name = OVS_VPORT_FAMILY,
2285         .version = OVS_VPORT_VERSION,
2286         .maxattr = OVS_VPORT_ATTR_MAX,
2287         .netnsok = true,
2288         .parallel_ops = true,
2289         .ops = dp_vport_genl_ops,
2290         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2291         .mcgrps = &ovs_dp_vport_multicast_group,
2292         .n_mcgrps = 1,
2293         .module = THIS_MODULE,
2294 };
2295
2296 static struct genl_family * const dp_genl_families[] = {
2297         &dp_datapath_genl_family,
2298         &dp_vport_genl_family,
2299         &dp_flow_genl_family,
2300         &dp_packet_genl_family,
2301 };
2302
2303 static void dp_unregister_genl(int n_families)
2304 {
2305         int i;
2306
2307         for (i = 0; i < n_families; i++)
2308                 genl_unregister_family(dp_genl_families[i]);
2309 }
2310
2311 static int __init dp_register_genl(void)
2312 {
2313         int err;
2314         int i;
2315
2316         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2317
2318                 err = genl_register_family(dp_genl_families[i]);
2319                 if (err)
2320                         goto error;
2321         }
2322
2323         return 0;
2324
2325 error:
2326         dp_unregister_genl(i);
2327         return err;
2328 }
2329
2330 static int __net_init ovs_init_net(struct net *net)
2331 {
2332         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2333
2334         INIT_LIST_HEAD(&ovs_net->dps);
2335         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2336         ovs_ct_init(net);
2337         return 0;
2338 }
2339
2340 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2341                                             struct list_head *head)
2342 {
2343         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2344         struct datapath *dp;
2345
2346         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2347                 int i;
2348
2349                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2350                         struct vport *vport;
2351
2352                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2353                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2354                                         continue;
2355
2356                                 if (dev_net(vport->dev) == dnet)
2357                                         list_add(&vport->detach_list, head);
2358                         }
2359                 }
2360         }
2361 }
2362
2363 static void __net_exit ovs_exit_net(struct net *dnet)
2364 {
2365         struct datapath *dp, *dp_next;
2366         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2367         struct vport *vport, *vport_next;
2368         struct net *net;
2369         LIST_HEAD(head);
2370
2371         ovs_ct_exit(dnet);
2372         ovs_lock();
2373         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2374                 __dp_destroy(dp);
2375
2376         rtnl_lock();
2377         for_each_net(net)
2378                 list_vports_from_net(net, dnet, &head);
2379         rtnl_unlock();
2380
2381         /* Detach all vports from given namespace. */
2382         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2383                 list_del(&vport->detach_list);
2384                 ovs_dp_detach_port(vport);
2385         }
2386
2387         ovs_unlock();
2388
2389         cancel_work_sync(&ovs_net->dp_notify_work);
2390 }
2391
2392 static struct pernet_operations ovs_net_ops = {
2393         .init = ovs_init_net,
2394         .exit = ovs_exit_net,
2395         .id   = &ovs_net_id,
2396         .size = sizeof(struct ovs_net),
2397 };
2398
2399 static int __init dp_init(void)
2400 {
2401         int err;
2402
2403         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2404
2405         pr_info("Open vSwitch switching datapath\n");
2406
2407         err = action_fifos_init();
2408         if (err)
2409                 goto error;
2410
2411         err = ovs_internal_dev_rtnl_link_register();
2412         if (err)
2413                 goto error_action_fifos_exit;
2414
2415         err = ovs_flow_init();
2416         if (err)
2417                 goto error_unreg_rtnl_link;
2418
2419         err = ovs_vport_init();
2420         if (err)
2421                 goto error_flow_exit;
2422
2423         err = register_pernet_device(&ovs_net_ops);
2424         if (err)
2425                 goto error_vport_exit;
2426
2427         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2428         if (err)
2429                 goto error_netns_exit;
2430
2431         err = ovs_netdev_init();
2432         if (err)
2433                 goto error_unreg_notifier;
2434
2435         err = dp_register_genl();
2436         if (err < 0)
2437                 goto error_unreg_netdev;
2438
2439         return 0;
2440
2441 error_unreg_netdev:
2442         ovs_netdev_exit();
2443 error_unreg_notifier:
2444         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2445 error_netns_exit:
2446         unregister_pernet_device(&ovs_net_ops);
2447 error_vport_exit:
2448         ovs_vport_exit();
2449 error_flow_exit:
2450         ovs_flow_exit();
2451 error_unreg_rtnl_link:
2452         ovs_internal_dev_rtnl_link_unregister();
2453 error_action_fifos_exit:
2454         action_fifos_exit();
2455 error:
2456         return err;
2457 }
2458
2459 static void dp_cleanup(void)
2460 {
2461         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2462         ovs_netdev_exit();
2463         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2464         unregister_pernet_device(&ovs_net_ops);
2465         rcu_barrier();
2466         ovs_vport_exit();
2467         ovs_flow_exit();
2468         ovs_internal_dev_rtnl_link_unregister();
2469         action_fifos_exit();
2470 }
2471
2472 module_init(dp_init);
2473 module_exit(dp_cleanup);
2474
2475 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2476 MODULE_LICENSE("GPL");
2477 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2478 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2479 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2480 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);