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