1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (c) 2007-2014 Nicira, Inc.
   4 */
   5
   6#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   7
   8#include <linux/init.h>
   9#include <linux/module.h>
  10#include <linux/if_arp.h>
  11#include <linux/if_vlan.h>
  12#include <linux/in.h>
  13#include <linux/ip.h>
  14#include <linux/jhash.h>
  15#include <linux/delay.h>
  16#include <linux/time.h>
  17#include <linux/etherdevice.h>
  18#include <linux/genetlink.h>
  19#include <linux/kernel.h>
  20#include <linux/kthread.h>
  21#include <linux/mutex.h>
  22#include <linux/percpu.h>
  23#include <linux/rcupdate.h>
  24#include <linux/tcp.h>
  25#include <linux/udp.h>
  26#include <linux/ethtool.h>
  27#include <linux/wait.h>
  28#include <asm/div64.h>
  29#include <linux/highmem.h>
  30#include <linux/netfilter_bridge.h>
  31#include <linux/netfilter_ipv4.h>
  32#include <linux/inetdevice.h>
  33#include <linux/list.h>
  34#include <linux/openvswitch.h>
  35#include <linux/rculist.h>
  36#include <linux/dmi.h>
  37#include <net/genetlink.h>
  38#include <net/net_namespace.h>
  39#include <net/netns/generic.h>
  40
  41#include "datapath.h"
  42#include "flow.h"
  43#include "flow_table.h"
  44#include "flow_netlink.h"
  45#include "meter.h"
  46#include "openvswitch_trace.h"
  47#include "vport-internal_dev.h"
  48#include "vport-netdev.h"
  49
  50unsigned int ovs_net_id __read_mostly;
  51
  52static struct genl_family dp_packet_genl_family;
  53static struct genl_family dp_flow_genl_family;
  54static struct genl_family dp_datapath_genl_family;
  55
  56static const struct nla_policy flow_policy[];
  57
  58static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
  59	.name = OVS_FLOW_MCGROUP,
  60};
  61
  62static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  63	.name = OVS_DATAPATH_MCGROUP,
  64};
  65
  66static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
  67	.name = OVS_VPORT_MCGROUP,
  68};
  69
  70/* Check if need to build a reply message.
  71 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  72static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  73			    unsigned int group)
  74{
  75	return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  76	       genl_has_listeners(family, genl_info_net(info), group);
  77}
  78
  79static void ovs_notify(struct genl_family *family,
  80		       struct sk_buff *skb, struct genl_info *info)
  81{
  82	genl_notify(family, skb, info, 0, GFP_KERNEL);
  83}
  84
  85/**
  86 * DOC: Locking:
  87 *
  88 * All writes e.g. Writes to device state (add/remove datapath, port, set
  89 * operations on vports, etc.), Writes to other state (flow table
  90 * modifications, set miscellaneous datapath parameters, etc.) are protected
  91 * by ovs_lock.
  92 *
  93 * Reads are protected by RCU.
  94 *
  95 * There are a few special cases (mostly stats) that have their own
  96 * synchronization but they nest under all of above and don't interact with
  97 * each other.
  98 *
  99 * The RTNL lock nests inside ovs_mutex.
 100 */
 101
 102static DEFINE_MUTEX(ovs_mutex);
 103
 104void ovs_lock(void)
 105{
 106	mutex_lock(&ovs_mutex);
 107}
 108
 109void ovs_unlock(void)
 110{
 111	mutex_unlock(&ovs_mutex);
 112}
 113
 114#ifdef CONFIG_LOCKDEP
 115int lockdep_ovsl_is_held(void)
 116{
 117	if (debug_locks)
 118		return lockdep_is_held(&ovs_mutex);
 119	else
 120		return 1;
 121}
 122#endif
 123
 124static struct vport *new_vport(const struct vport_parms *);
 125static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 126			     const struct sw_flow_key *,
 127			     const struct dp_upcall_info *,
 128			     uint32_t cutlen);
 129static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 130				  const struct sw_flow_key *,
 131				  const struct dp_upcall_info *,
 132				  uint32_t cutlen);
 133
 134static void ovs_dp_masks_rebalance(struct work_struct *work);
 135
 136/* Must be called with rcu_read_lock or ovs_mutex. */
 137const char *ovs_dp_name(const struct datapath *dp)
 138{
 139	struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 140	return ovs_vport_name(vport);
 141}
 142
 143static int get_dpifindex(const struct datapath *dp)
 144{
 145	struct vport *local;
 146	int ifindex;
 147
 148	rcu_read_lock();
 149
 150	local = ovs_vport_rcu(dp, OVSP_LOCAL);
 151	if (local)
 152		ifindex = local->dev->ifindex;
 153	else
 154		ifindex = 0;
 155
 156	rcu_read_unlock();
 157
 158	return ifindex;
 159}
 160
 161static void destroy_dp_rcu(struct rcu_head *rcu)
 162{
 163	struct datapath *dp = container_of(rcu, struct datapath, rcu);
 164
 165	ovs_flow_tbl_destroy(&dp->table);
 166	free_percpu(dp->stats_percpu);
 167	kfree(dp->ports);
 168	ovs_meters_exit(dp);
 169	kfree(dp);
 170}
 171
 172static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 173					    u16 port_no)
 174{
 175	return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 176}
 177
 178/* Called with ovs_mutex or RCU read lock. */
 179struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 180{
 181	struct vport *vport;
 182	struct hlist_head *head;
 183
 184	head = vport_hash_bucket(dp, port_no);
 185	hlist_for_each_entry_rcu(vport, head, dp_hash_node,
 186				 lockdep_ovsl_is_held()) {
 187		if (vport->port_no == port_no)
 188			return vport;
 189	}
 190	return NULL;
 191}
 192
 193/* Called with ovs_mutex. */
 194static struct vport *new_vport(const struct vport_parms *parms)
 195{
 196	struct vport *vport;
 197
 198	vport = ovs_vport_add(parms);
 199	if (!IS_ERR(vport)) {
 200		struct datapath *dp = parms->dp;
 201		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 202
 203		hlist_add_head_rcu(&vport->dp_hash_node, head);
 204	}
 205	return vport;
 206}
 207
 208void ovs_dp_detach_port(struct vport *p)
 209{
 210	ASSERT_OVSL();
 211
 212	/* First drop references to device. */
 213	hlist_del_rcu(&p->dp_hash_node);
 214
 215	/* Then destroy it. */
 216	ovs_vport_del(p);
 217}
 218
 219/* Must be called with rcu_read_lock. */
 220void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 221{
 222	const struct vport *p = OVS_CB(skb)->input_vport;
 223	struct datapath *dp = p->dp;
 224	struct sw_flow *flow;
 225	struct sw_flow_actions *sf_acts;
 226	struct dp_stats_percpu *stats;
 227	u64 *stats_counter;
 228	u32 n_mask_hit;
 229	u32 n_cache_hit;
 230	int error;
 231
 232	stats = this_cpu_ptr(dp->stats_percpu);
 233
 234	/* Look up flow. */
 235	flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
 236					 &n_mask_hit, &n_cache_hit);
 237	if (unlikely(!flow)) {
 238		struct dp_upcall_info upcall;
 239
 240		memset(&upcall, 0, sizeof(upcall));
 241		upcall.cmd = OVS_PACKET_CMD_MISS;
 242		upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 243		upcall.mru = OVS_CB(skb)->mru;
 244		error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
 245		if (unlikely(error))
 246			kfree_skb(skb);
 247		else
 248			consume_skb(skb);
 249		stats_counter = &stats->n_missed;
 250		goto out;
 251	}
 252
 253	ovs_flow_stats_update(flow, key->tp.flags, skb);
 254	sf_acts = rcu_dereference(flow->sf_acts);
 255	error = ovs_execute_actions(dp, skb, sf_acts, key);
 256	if (unlikely(error))
 257		net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
 258				    ovs_dp_name(dp), error);
 259
 260	stats_counter = &stats->n_hit;
 261
 262out:
 263	/* Update datapath statistics. */
 264	u64_stats_update_begin(&stats->syncp);
 265	(*stats_counter)++;
 266	stats->n_mask_hit += n_mask_hit;
 267	stats->n_cache_hit += n_cache_hit;
 268	u64_stats_update_end(&stats->syncp);
 269}
 270
 271int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 272		  const struct sw_flow_key *key,
 273		  const struct dp_upcall_info *upcall_info,
 274		  uint32_t cutlen)
 275{
 276	struct dp_stats_percpu *stats;
 277	int err;
 278
 279	if (trace_ovs_dp_upcall_enabled())
 280		trace_ovs_dp_upcall(dp, skb, key, upcall_info);
 281
 282	if (upcall_info->portid == 0) {
 283		err = -ENOTCONN;
 284		goto err;
 285	}
 286
 287	if (!skb_is_gso(skb))
 288		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 289	else
 290		err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
 291	if (err)
 292		goto err;
 293
 294	return 0;
 295
 296err:
 297	stats = this_cpu_ptr(dp->stats_percpu);
 298
 299	u64_stats_update_begin(&stats->syncp);
 300	stats->n_lost++;
 301	u64_stats_update_end(&stats->syncp);
 302
 303	return err;
 304}
 305
 306static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 307			     const struct sw_flow_key *key,
 308			     const struct dp_upcall_info *upcall_info,
 309			     uint32_t cutlen)
 310{
 311	unsigned int gso_type = skb_shinfo(skb)->gso_type;
 312	struct sw_flow_key later_key;
 313	struct sk_buff *segs, *nskb;
 314	int err;
 315
 316	BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_GSO_CB_OFFSET);
 317	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 318	if (IS_ERR(segs))
 319		return PTR_ERR(segs);
 320	if (segs == NULL)
 321		return -EINVAL;
 322
 323	if (gso_type & SKB_GSO_UDP) {
 324		/* The initial flow key extracted by ovs_flow_key_extract()
 325		 * in this case is for a first fragment, so we need to
 326		 * properly mark later fragments.
 327		 */
 328		later_key = *key;
 329		later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 330	}
 331
 332	/* Queue all of the segments. */
 333	skb_list_walk_safe(segs, skb, nskb) {
 
 334		if (gso_type & SKB_GSO_UDP && skb != segs)
 335			key = &later_key;
 336
 337		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 338		if (err)
 339			break;
 340
 341	}
 342
 343	/* Free all of the segments. */
 344	skb_list_walk_safe(segs, skb, nskb) {
 
 
 345		if (err)
 346			kfree_skb(skb);
 347		else
 348			consume_skb(skb);
 349	}
 350	return err;
 351}
 352
 353static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 354			      unsigned int hdrlen, int actions_attrlen)
 355{
 356	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 357		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 358		+ nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 359		+ nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
 360		+ nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
 361
 362	/* OVS_PACKET_ATTR_USERDATA */
 363	if (upcall_info->userdata)
 364		size += NLA_ALIGN(upcall_info->userdata->nla_len);
 365
 366	/* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 367	if (upcall_info->egress_tun_info)
 368		size += nla_total_size(ovs_tun_key_attr_size());
 369
 370	/* OVS_PACKET_ATTR_ACTIONS */
 371	if (upcall_info->actions_len)
 372		size += nla_total_size(actions_attrlen);
 373
 374	/* OVS_PACKET_ATTR_MRU */
 375	if (upcall_info->mru)
 376		size += nla_total_size(sizeof(upcall_info->mru));
 377
 378	return size;
 379}
 380
 381static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 382{
 383	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 384		size_t plen = NLA_ALIGN(skb->len) - skb->len;
 385
 386		if (plen > 0)
 387			skb_put_zero(skb, plen);
 388	}
 389}
 390
 391static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 392				  const struct sw_flow_key *key,
 393				  const struct dp_upcall_info *upcall_info,
 394				  uint32_t cutlen)
 395{
 396	struct ovs_header *upcall;
 397	struct sk_buff *nskb = NULL;
 398	struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 399	struct nlattr *nla;
 400	size_t len;
 401	unsigned int hlen;
 402	int err, dp_ifindex;
 403	u64 hash;
 404
 405	dp_ifindex = get_dpifindex(dp);
 406	if (!dp_ifindex)
 407		return -ENODEV;
 408
 409	if (skb_vlan_tag_present(skb)) {
 410		nskb = skb_clone(skb, GFP_ATOMIC);
 411		if (!nskb)
 412			return -ENOMEM;
 413
 414		nskb = __vlan_hwaccel_push_inside(nskb);
 415		if (!nskb)
 416			return -ENOMEM;
 417
 418		skb = nskb;
 419	}
 420
 421	if (nla_attr_size(skb->len) > USHRT_MAX) {
 422		err = -EFBIG;
 423		goto out;
 424	}
 425
 426	/* Complete checksum if needed */
 427	if (skb->ip_summed == CHECKSUM_PARTIAL &&
 428	    (err = skb_csum_hwoffload_help(skb, 0)))
 429		goto out;
 430
 431	/* Older versions of OVS user space enforce alignment of the last
 432	 * Netlink attribute to NLA_ALIGNTO which would require extensive
 433	 * padding logic. Only perform zerocopy if padding is not required.
 434	 */
 435	if (dp->user_features & OVS_DP_F_UNALIGNED)
 436		hlen = skb_zerocopy_headlen(skb);
 437	else
 438		hlen = skb->len;
 439
 440	len = upcall_msg_size(upcall_info, hlen - cutlen,
 441			      OVS_CB(skb)->acts_origlen);
 442	user_skb = genlmsg_new(len, GFP_ATOMIC);
 443	if (!user_skb) {
 444		err = -ENOMEM;
 445		goto out;
 446	}
 447
 448	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 449			     0, upcall_info->cmd);
 450	if (!upcall) {
 451		err = -EINVAL;
 452		goto out;
 453	}
 454	upcall->dp_ifindex = dp_ifindex;
 455
 456	err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 457	if (err)
 458		goto out;
 459
 460	if (upcall_info->userdata)
 461		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 462			  nla_len(upcall_info->userdata),
 463			  nla_data(upcall_info->userdata));
 464
 465	if (upcall_info->egress_tun_info) {
 466		nla = nla_nest_start_noflag(user_skb,
 467					    OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 468		if (!nla) {
 469			err = -EMSGSIZE;
 470			goto out;
 471		}
 472		err = ovs_nla_put_tunnel_info(user_skb,
 473					      upcall_info->egress_tun_info);
 474		if (err)
 475			goto out;
 476
 477		nla_nest_end(user_skb, nla);
 478	}
 479
 480	if (upcall_info->actions_len) {
 481		nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
 482		if (!nla) {
 483			err = -EMSGSIZE;
 484			goto out;
 485		}
 486		err = ovs_nla_put_actions(upcall_info->actions,
 487					  upcall_info->actions_len,
 488					  user_skb);
 489		if (!err)
 490			nla_nest_end(user_skb, nla);
 491		else
 492			nla_nest_cancel(user_skb, nla);
 493	}
 494
 495	/* Add OVS_PACKET_ATTR_MRU */
 496	if (upcall_info->mru &&
 497	    nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
 498		err = -ENOBUFS;
 499		goto out;
 
 
 
 500	}
 501
 502	/* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 503	if (cutlen > 0 &&
 504	    nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
 505		err = -ENOBUFS;
 506		goto out;
 507	}
 508
 509	/* Add OVS_PACKET_ATTR_HASH */
 510	hash = skb_get_hash_raw(skb);
 511	if (skb->sw_hash)
 512		hash |= OVS_PACKET_HASH_SW_BIT;
 513
 514	if (skb->l4_hash)
 515		hash |= OVS_PACKET_HASH_L4_BIT;
 516
 517	if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
 518		err = -ENOBUFS;
 519		goto out;
 520	}
 521
 522	/* Only reserve room for attribute header, packet data is added
 523	 * in skb_zerocopy() */
 524	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 525		err = -ENOBUFS;
 526		goto out;
 527	}
 528	nla->nla_len = nla_attr_size(skb->len - cutlen);
 529
 530	err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 531	if (err)
 532		goto out;
 533
 534	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 535	pad_packet(dp, user_skb);
 536
 537	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 538
 539	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 540	user_skb = NULL;
 541out:
 542	if (err)
 543		skb_tx_error(skb);
 544	kfree_skb(user_skb);
 545	kfree_skb(nskb);
 546	return err;
 547}
 548
 549static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 550{
 551	struct ovs_header *ovs_header = info->userhdr;
 552	struct net *net = sock_net(skb->sk);
 553	struct nlattr **a = info->attrs;
 554	struct sw_flow_actions *acts;
 555	struct sk_buff *packet;
 556	struct sw_flow *flow;
 557	struct sw_flow_actions *sf_acts;
 558	struct datapath *dp;
 559	struct vport *input_vport;
 560	u16 mru = 0;
 561	u64 hash;
 562	int len;
 563	int err;
 564	bool log = !a[OVS_PACKET_ATTR_PROBE];
 565
 566	err = -EINVAL;
 567	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 568	    !a[OVS_PACKET_ATTR_ACTIONS])
 569		goto err;
 570
 571	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 572	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 573	err = -ENOMEM;
 574	if (!packet)
 575		goto err;
 576	skb_reserve(packet, NET_IP_ALIGN);
 577
 578	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 579
 580	/* Set packet's mru */
 581	if (a[OVS_PACKET_ATTR_MRU]) {
 582		mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 583		packet->ignore_df = 1;
 584	}
 585	OVS_CB(packet)->mru = mru;
 586
 587	if (a[OVS_PACKET_ATTR_HASH]) {
 588		hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
 589
 590		__skb_set_hash(packet, hash & 0xFFFFFFFFULL,
 591			       !!(hash & OVS_PACKET_HASH_SW_BIT),
 592			       !!(hash & OVS_PACKET_HASH_L4_BIT));
 593	}
 594
 595	/* Build an sw_flow for sending this packet. */
 596	flow = ovs_flow_alloc();
 597	err = PTR_ERR(flow);
 598	if (IS_ERR(flow))
 599		goto err_kfree_skb;
 600
 601	err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 602					     packet, &flow->key, log);
 603	if (err)
 604		goto err_flow_free;
 605
 606	err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 607				   &flow->key, &acts, log);
 608	if (err)
 609		goto err_flow_free;
 610
 611	rcu_assign_pointer(flow->sf_acts, acts);
 612	packet->priority = flow->key.phy.priority;
 613	packet->mark = flow->key.phy.skb_mark;
 614
 615	rcu_read_lock();
 616	dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 617	err = -ENODEV;
 618	if (!dp)
 619		goto err_unlock;
 620
 621	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 622	if (!input_vport)
 623		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 624
 625	if (!input_vport)
 626		goto err_unlock;
 627
 628	packet->dev = input_vport->dev;
 629	OVS_CB(packet)->input_vport = input_vport;
 630	sf_acts = rcu_dereference(flow->sf_acts);
 631
 632	local_bh_disable();
 633	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 634	local_bh_enable();
 635	rcu_read_unlock();
 636
 637	ovs_flow_free(flow, false);
 638	return err;
 639
 640err_unlock:
 641	rcu_read_unlock();
 642err_flow_free:
 643	ovs_flow_free(flow, false);
 644err_kfree_skb:
 645	kfree_skb(packet);
 646err:
 647	return err;
 648}
 649
 650static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 651	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 652	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 653	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 654	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 655	[OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 656	[OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
 657};
 658
 659static const struct genl_small_ops dp_packet_genl_ops[] = {
 660	{ .cmd = OVS_PACKET_CMD_EXECUTE,
 661	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 662	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 663	  .doit = ovs_packet_cmd_execute
 664	}
 665};
 666
 667static struct genl_family dp_packet_genl_family __ro_after_init = {
 668	.hdrsize = sizeof(struct ovs_header),
 669	.name = OVS_PACKET_FAMILY,
 670	.version = OVS_PACKET_VERSION,
 671	.maxattr = OVS_PACKET_ATTR_MAX,
 672	.policy = packet_policy,
 673	.netnsok = true,
 674	.parallel_ops = true,
 675	.small_ops = dp_packet_genl_ops,
 676	.n_small_ops = ARRAY_SIZE(dp_packet_genl_ops),
 677	.module = THIS_MODULE,
 678};
 679
 680static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 681			 struct ovs_dp_megaflow_stats *mega_stats)
 682{
 683	int i;
 684
 685	memset(mega_stats, 0, sizeof(*mega_stats));
 686
 687	stats->n_flows = ovs_flow_tbl_count(&dp->table);
 688	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 689
 690	stats->n_hit = stats->n_missed = stats->n_lost = 0;
 691
 692	for_each_possible_cpu(i) {
 693		const struct dp_stats_percpu *percpu_stats;
 694		struct dp_stats_percpu local_stats;
 695		unsigned int start;
 696
 697		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 698
 699		do {
 700			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 701			local_stats = *percpu_stats;
 702		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 703
 704		stats->n_hit += local_stats.n_hit;
 705		stats->n_missed += local_stats.n_missed;
 706		stats->n_lost += local_stats.n_lost;
 707		mega_stats->n_mask_hit += local_stats.n_mask_hit;
 708		mega_stats->n_cache_hit += local_stats.n_cache_hit;
 709	}
 710}
 711
 712static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 713{
 714	return ovs_identifier_is_ufid(sfid) &&
 715	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 716}
 717
 718static bool should_fill_mask(uint32_t ufid_flags)
 719{
 720	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 721}
 722
 723static bool should_fill_actions(uint32_t ufid_flags)
 724{
 725	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 726}
 727
 728static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 729				    const struct sw_flow_id *sfid,
 730				    uint32_t ufid_flags)
 731{
 732	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 733
 734	/* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
 735	 * see ovs_nla_put_identifier()
 736	 */
 737	if (sfid && ovs_identifier_is_ufid(sfid))
 738		len += nla_total_size(sfid->ufid_len);
 739	else
 740		len += nla_total_size(ovs_key_attr_size());
 741
 742	/* OVS_FLOW_ATTR_KEY */
 743	if (!sfid || should_fill_key(sfid, ufid_flags))
 744		len += nla_total_size(ovs_key_attr_size());
 745
 746	/* OVS_FLOW_ATTR_MASK */
 747	if (should_fill_mask(ufid_flags))
 748		len += nla_total_size(ovs_key_attr_size());
 749
 750	/* OVS_FLOW_ATTR_ACTIONS */
 751	if (should_fill_actions(ufid_flags))
 752		len += nla_total_size(acts->orig_len);
 753
 754	return len
 755		+ nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 756		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 757		+ nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 758}
 759
 760/* Called with ovs_mutex or RCU read lock. */
 761static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 762				   struct sk_buff *skb)
 763{
 764	struct ovs_flow_stats stats;
 765	__be16 tcp_flags;
 766	unsigned long used;
 767
 768	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 769
 770	if (used &&
 771	    nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 772			      OVS_FLOW_ATTR_PAD))
 773		return -EMSGSIZE;
 774
 775	if (stats.n_packets &&
 776	    nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 777			  sizeof(struct ovs_flow_stats), &stats,
 778			  OVS_FLOW_ATTR_PAD))
 779		return -EMSGSIZE;
 780
 781	if ((u8)ntohs(tcp_flags) &&
 782	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 783		return -EMSGSIZE;
 784
 785	return 0;
 786}
 787
 788/* Called with ovs_mutex or RCU read lock. */
 789static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 790				     struct sk_buff *skb, int skb_orig_len)
 791{
 792	struct nlattr *start;
 793	int err;
 794
 795	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 796	 * this is the first flow to be dumped into 'skb'.  This is unusual for
 797	 * Netlink but individual action lists can be longer than
 798	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 799	 * The userspace caller can always fetch the actions separately if it
 800	 * really wants them.  (Most userspace callers in fact don't care.)
 801	 *
 802	 * This can only fail for dump operations because the skb is always
 803	 * properly sized for single flows.
 804	 */
 805	start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
 806	if (start) {
 807		const struct sw_flow_actions *sf_acts;
 808
 809		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 810		err = ovs_nla_put_actions(sf_acts->actions,
 811					  sf_acts->actions_len, skb);
 812
 813		if (!err)
 814			nla_nest_end(skb, start);
 815		else {
 816			if (skb_orig_len)
 817				return err;
 818
 819			nla_nest_cancel(skb, start);
 820		}
 821	} else if (skb_orig_len) {
 822		return -EMSGSIZE;
 823	}
 824
 825	return 0;
 826}
 827
 828/* Called with ovs_mutex or RCU read lock. */
 829static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 830				  struct sk_buff *skb, u32 portid,
 831				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 832{
 833	const int skb_orig_len = skb->len;
 834	struct ovs_header *ovs_header;
 835	int err;
 836
 837	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 838				 flags, cmd);
 839	if (!ovs_header)
 840		return -EMSGSIZE;
 841
 842	ovs_header->dp_ifindex = dp_ifindex;
 843
 844	err = ovs_nla_put_identifier(flow, skb);
 845	if (err)
 846		goto error;
 847
 848	if (should_fill_key(&flow->id, ufid_flags)) {
 849		err = ovs_nla_put_masked_key(flow, skb);
 850		if (err)
 851			goto error;
 852	}
 853
 854	if (should_fill_mask(ufid_flags)) {
 855		err = ovs_nla_put_mask(flow, skb);
 856		if (err)
 857			goto error;
 858	}
 859
 860	err = ovs_flow_cmd_fill_stats(flow, skb);
 861	if (err)
 862		goto error;
 863
 864	if (should_fill_actions(ufid_flags)) {
 865		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 866		if (err)
 867			goto error;
 868	}
 869
 870	genlmsg_end(skb, ovs_header);
 871	return 0;
 872
 873error:
 874	genlmsg_cancel(skb, ovs_header);
 875	return err;
 876}
 877
 878/* May not be called with RCU read lock. */
 879static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 880					       const struct sw_flow_id *sfid,
 881					       struct genl_info *info,
 882					       bool always,
 883					       uint32_t ufid_flags)
 884{
 885	struct sk_buff *skb;
 886	size_t len;
 887
 888	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 889		return NULL;
 890
 891	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 892	skb = genlmsg_new(len, GFP_KERNEL);
 893	if (!skb)
 894		return ERR_PTR(-ENOMEM);
 895
 896	return skb;
 897}
 898
 899/* Called with ovs_mutex. */
 900static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 901					       int dp_ifindex,
 902					       struct genl_info *info, u8 cmd,
 903					       bool always, u32 ufid_flags)
 904{
 905	struct sk_buff *skb;
 906	int retval;
 907
 908	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 909				      &flow->id, info, always, ufid_flags);
 910	if (IS_ERR_OR_NULL(skb))
 911		return skb;
 912
 913	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 914					info->snd_portid, info->snd_seq, 0,
 915					cmd, ufid_flags);
 916	if (WARN_ON_ONCE(retval < 0)) {
 917		kfree_skb(skb);
 918		skb = ERR_PTR(retval);
 919	}
 920	return skb;
 921}
 922
 923static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 924{
 925	struct net *net = sock_net(skb->sk);
 926	struct nlattr **a = info->attrs;
 927	struct ovs_header *ovs_header = info->userhdr;
 928	struct sw_flow *flow = NULL, *new_flow;
 929	struct sw_flow_mask mask;
 930	struct sk_buff *reply;
 931	struct datapath *dp;
 932	struct sw_flow_actions *acts;
 933	struct sw_flow_match match;
 934	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 935	int error;
 936	bool log = !a[OVS_FLOW_ATTR_PROBE];
 937
 938	/* Must have key and actions. */
 939	error = -EINVAL;
 940	if (!a[OVS_FLOW_ATTR_KEY]) {
 941		OVS_NLERR(log, "Flow key attr not present in new flow.");
 942		goto error;
 943	}
 944	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 945		OVS_NLERR(log, "Flow actions attr not present in new flow.");
 946		goto error;
 947	}
 948
 949	/* Most of the time we need to allocate a new flow, do it before
 950	 * locking.
 951	 */
 952	new_flow = ovs_flow_alloc();
 953	if (IS_ERR(new_flow)) {
 954		error = PTR_ERR(new_flow);
 955		goto error;
 956	}
 957
 958	/* Extract key. */
 959	ovs_match_init(&match, &new_flow->key, false, &mask);
 960	error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 961				  a[OVS_FLOW_ATTR_MASK], log);
 962	if (error)
 963		goto err_kfree_flow;
 964
 965	/* Extract flow identifier. */
 966	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 967				       &new_flow->key, log);
 968	if (error)
 969		goto err_kfree_flow;
 970
 971	/* unmasked key is needed to match when ufid is not used. */
 972	if (ovs_identifier_is_key(&new_flow->id))
 973		match.key = new_flow->id.unmasked_key;
 974
 975	ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
 976
 977	/* Validate actions. */
 978	error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 979				     &new_flow->key, &acts, log);
 980	if (error) {
 981		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 982		goto err_kfree_flow;
 983	}
 984
 985	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 986					ufid_flags);
 987	if (IS_ERR(reply)) {
 988		error = PTR_ERR(reply);
 989		goto err_kfree_acts;
 990	}
 991
 992	ovs_lock();
 993	dp = get_dp(net, ovs_header->dp_ifindex);
 994	if (unlikely(!dp)) {
 995		error = -ENODEV;
 996		goto err_unlock_ovs;
 997	}
 998
 999	/* Check if this is a duplicate flow */
1000	if (ovs_identifier_is_ufid(&new_flow->id))
1001		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1002	if (!flow)
1003		flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
1004	if (likely(!flow)) {
1005		rcu_assign_pointer(new_flow->sf_acts, acts);
1006
1007		/* Put flow in bucket. */
1008		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1009		if (unlikely(error)) {
1010			acts = NULL;
1011			goto err_unlock_ovs;
1012		}
1013
1014		if (unlikely(reply)) {
1015			error = ovs_flow_cmd_fill_info(new_flow,
1016						       ovs_header->dp_ifindex,
1017						       reply, info->snd_portid,
1018						       info->snd_seq, 0,
1019						       OVS_FLOW_CMD_NEW,
1020						       ufid_flags);
1021			BUG_ON(error < 0);
1022		}
1023		ovs_unlock();
1024	} else {
1025		struct sw_flow_actions *old_acts;
1026
1027		/* Bail out if we're not allowed to modify an existing flow.
1028		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1029		 * because Generic Netlink treats the latter as a dump
1030		 * request.  We also accept NLM_F_EXCL in case that bug ever
1031		 * gets fixed.
1032		 */
1033		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1034							 | NLM_F_EXCL))) {
1035			error = -EEXIST;
1036			goto err_unlock_ovs;
1037		}
1038		/* The flow identifier has to be the same for flow updates.
1039		 * Look for any overlapping flow.
1040		 */
1041		if (unlikely(!ovs_flow_cmp(flow, &match))) {
1042			if (ovs_identifier_is_key(&flow->id))
1043				flow = ovs_flow_tbl_lookup_exact(&dp->table,
1044								 &match);
1045			else /* UFID matches but key is different */
1046				flow = NULL;
1047			if (!flow) {
1048				error = -ENOENT;
1049				goto err_unlock_ovs;
1050			}
1051		}
1052		/* Update actions. */
1053		old_acts = ovsl_dereference(flow->sf_acts);
1054		rcu_assign_pointer(flow->sf_acts, acts);
1055
1056		if (unlikely(reply)) {
1057			error = ovs_flow_cmd_fill_info(flow,
1058						       ovs_header->dp_ifindex,
1059						       reply, info->snd_portid,
1060						       info->snd_seq, 0,
1061						       OVS_FLOW_CMD_NEW,
1062						       ufid_flags);
1063			BUG_ON(error < 0);
1064		}
1065		ovs_unlock();
1066
1067		ovs_nla_free_flow_actions_rcu(old_acts);
1068		ovs_flow_free(new_flow, false);
1069	}
1070
1071	if (reply)
1072		ovs_notify(&dp_flow_genl_family, reply, info);
1073	return 0;
1074
1075err_unlock_ovs:
1076	ovs_unlock();
1077	kfree_skb(reply);
1078err_kfree_acts:
1079	ovs_nla_free_flow_actions(acts);
1080err_kfree_flow:
1081	ovs_flow_free(new_flow, false);
1082error:
1083	return error;
1084}
1085
1086/* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1087static noinline_for_stack
1088struct sw_flow_actions *get_flow_actions(struct net *net,
1089					 const struct nlattr *a,
1090					 const struct sw_flow_key *key,
1091					 const struct sw_flow_mask *mask,
1092					 bool log)
1093{
1094	struct sw_flow_actions *acts;
1095	struct sw_flow_key masked_key;
1096	int error;
1097
1098	ovs_flow_mask_key(&masked_key, key, true, mask);
1099	error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1100	if (error) {
1101		OVS_NLERR(log,
1102			  "Actions may not be safe on all matching packets");
1103		return ERR_PTR(error);
1104	}
1105
1106	return acts;
1107}
1108
1109/* Factor out match-init and action-copy to avoid
1110 * "Wframe-larger-than=1024" warning. Because mask is only
1111 * used to get actions, we new a function to save some
1112 * stack space.
1113 *
1114 * If there are not key and action attrs, we return 0
1115 * directly. In the case, the caller will also not use the
1116 * match as before. If there is action attr, we try to get
1117 * actions and save them to *acts. Before returning from
1118 * the function, we reset the match->mask pointer. Because
1119 * we should not to return match object with dangling reference
1120 * to mask.
1121 * */
1122static noinline_for_stack int
1123ovs_nla_init_match_and_action(struct net *net,
1124			      struct sw_flow_match *match,
1125			      struct sw_flow_key *key,
1126			      struct nlattr **a,
1127			      struct sw_flow_actions **acts,
1128			      bool log)
1129{
1130	struct sw_flow_mask mask;
1131	int error = 0;
1132
1133	if (a[OVS_FLOW_ATTR_KEY]) {
1134		ovs_match_init(match, key, true, &mask);
1135		error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1136					  a[OVS_FLOW_ATTR_MASK], log);
1137		if (error)
1138			goto error;
1139	}
1140
1141	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1142		if (!a[OVS_FLOW_ATTR_KEY]) {
1143			OVS_NLERR(log,
1144				  "Flow key attribute not present in set flow.");
1145			error = -EINVAL;
1146			goto error;
1147		}
1148
1149		*acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1150					 &mask, log);
1151		if (IS_ERR(*acts)) {
1152			error = PTR_ERR(*acts);
1153			goto error;
1154		}
1155	}
1156
1157	/* On success, error is 0. */
1158error:
1159	match->mask = NULL;
1160	return error;
1161}
1162
1163static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1164{
1165	struct net *net = sock_net(skb->sk);
1166	struct nlattr **a = info->attrs;
1167	struct ovs_header *ovs_header = info->userhdr;
1168	struct sw_flow_key key;
1169	struct sw_flow *flow;
1170	struct sk_buff *reply = NULL;
1171	struct datapath *dp;
1172	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1173	struct sw_flow_match match;
1174	struct sw_flow_id sfid;
1175	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1176	int error = 0;
1177	bool log = !a[OVS_FLOW_ATTR_PROBE];
1178	bool ufid_present;
1179
1180	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1181	if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1182		OVS_NLERR(log,
1183			  "Flow set message rejected, Key attribute missing.");
1184		return -EINVAL;
1185	}
1186
1187	error = ovs_nla_init_match_and_action(net, &match, &key, a,
1188					      &acts, log);
1189	if (error)
1190		goto error;
1191
1192	if (acts) {
1193		/* Can allocate before locking if have acts. */
1194		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1195						ufid_flags);
1196		if (IS_ERR(reply)) {
1197			error = PTR_ERR(reply);
1198			goto err_kfree_acts;
1199		}
1200	}
1201
1202	ovs_lock();
1203	dp = get_dp(net, ovs_header->dp_ifindex);
1204	if (unlikely(!dp)) {
1205		error = -ENODEV;
1206		goto err_unlock_ovs;
1207	}
1208	/* Check that the flow exists. */
1209	if (ufid_present)
1210		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1211	else
1212		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1213	if (unlikely(!flow)) {
1214		error = -ENOENT;
1215		goto err_unlock_ovs;
1216	}
1217
1218	/* Update actions, if present. */
1219	if (likely(acts)) {
1220		old_acts = ovsl_dereference(flow->sf_acts);
1221		rcu_assign_pointer(flow->sf_acts, acts);
1222
1223		if (unlikely(reply)) {
1224			error = ovs_flow_cmd_fill_info(flow,
1225						       ovs_header->dp_ifindex,
1226						       reply, info->snd_portid,
1227						       info->snd_seq, 0,
1228						       OVS_FLOW_CMD_SET,
1229						       ufid_flags);
1230			BUG_ON(error < 0);
1231		}
1232	} else {
1233		/* Could not alloc without acts before locking. */
1234		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1235						info, OVS_FLOW_CMD_SET, false,
1236						ufid_flags);
1237
1238		if (IS_ERR(reply)) {
1239			error = PTR_ERR(reply);
1240			goto err_unlock_ovs;
1241		}
1242	}
1243
1244	/* Clear stats. */
1245	if (a[OVS_FLOW_ATTR_CLEAR])
1246		ovs_flow_stats_clear(flow);
1247	ovs_unlock();
1248
1249	if (reply)
1250		ovs_notify(&dp_flow_genl_family, reply, info);
1251	if (old_acts)
1252		ovs_nla_free_flow_actions_rcu(old_acts);
1253
1254	return 0;
1255
1256err_unlock_ovs:
1257	ovs_unlock();
1258	kfree_skb(reply);
1259err_kfree_acts:
1260	ovs_nla_free_flow_actions(acts);
1261error:
1262	return error;
1263}
1264
1265static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1266{
1267	struct nlattr **a = info->attrs;
1268	struct ovs_header *ovs_header = info->userhdr;
1269	struct net *net = sock_net(skb->sk);
1270	struct sw_flow_key key;
1271	struct sk_buff *reply;
1272	struct sw_flow *flow;
1273	struct datapath *dp;
1274	struct sw_flow_match match;
1275	struct sw_flow_id ufid;
1276	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1277	int err = 0;
1278	bool log = !a[OVS_FLOW_ATTR_PROBE];
1279	bool ufid_present;
1280
1281	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1282	if (a[OVS_FLOW_ATTR_KEY]) {
1283		ovs_match_init(&match, &key, true, NULL);
1284		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1285					log);
1286	} else if (!ufid_present) {
1287		OVS_NLERR(log,
1288			  "Flow get message rejected, Key attribute missing.");
1289		err = -EINVAL;
1290	}
1291	if (err)
1292		return err;
1293
1294	ovs_lock();
1295	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1296	if (!dp) {
1297		err = -ENODEV;
1298		goto unlock;
1299	}
1300
1301	if (ufid_present)
1302		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1303	else
1304		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1305	if (!flow) {
1306		err = -ENOENT;
1307		goto unlock;
1308	}
1309
1310	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1311					OVS_FLOW_CMD_GET, true, ufid_flags);
1312	if (IS_ERR(reply)) {
1313		err = PTR_ERR(reply);
1314		goto unlock;
1315	}
1316
1317	ovs_unlock();
1318	return genlmsg_reply(reply, info);
1319unlock:
1320	ovs_unlock();
1321	return err;
1322}
1323
1324static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1325{
1326	struct nlattr **a = info->attrs;
1327	struct ovs_header *ovs_header = info->userhdr;
1328	struct net *net = sock_net(skb->sk);
1329	struct sw_flow_key key;
1330	struct sk_buff *reply;
1331	struct sw_flow *flow = NULL;
1332	struct datapath *dp;
1333	struct sw_flow_match match;
1334	struct sw_flow_id ufid;
1335	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1336	int err;
1337	bool log = !a[OVS_FLOW_ATTR_PROBE];
1338	bool ufid_present;
1339
1340	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1341	if (a[OVS_FLOW_ATTR_KEY]) {
1342		ovs_match_init(&match, &key, true, NULL);
1343		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1344					NULL, log);
1345		if (unlikely(err))
1346			return err;
1347	}
1348
1349	ovs_lock();
1350	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1351	if (unlikely(!dp)) {
1352		err = -ENODEV;
1353		goto unlock;
1354	}
1355
1356	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1357		err = ovs_flow_tbl_flush(&dp->table);
1358		goto unlock;
1359	}
1360
1361	if (ufid_present)
1362		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1363	else
1364		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1365	if (unlikely(!flow)) {
1366		err = -ENOENT;
1367		goto unlock;
1368	}
1369
1370	ovs_flow_tbl_remove(&dp->table, flow);
1371	ovs_unlock();
1372
1373	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1374					&flow->id, info, false, ufid_flags);
1375	if (likely(reply)) {
1376		if (!IS_ERR(reply)) {
1377			rcu_read_lock();	/*To keep RCU checker happy. */
1378			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1379						     reply, info->snd_portid,
1380						     info->snd_seq, 0,
1381						     OVS_FLOW_CMD_DEL,
1382						     ufid_flags);
1383			rcu_read_unlock();
1384			if (WARN_ON_ONCE(err < 0)) {
1385				kfree_skb(reply);
1386				goto out_free;
1387			}
1388
1389			ovs_notify(&dp_flow_genl_family, reply, info);
1390		} else {
1391			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1392					PTR_ERR(reply));
1393		}
1394	}
1395
1396out_free:
1397	ovs_flow_free(flow, true);
1398	return 0;
1399unlock:
1400	ovs_unlock();
1401	return err;
1402}
1403
1404static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1405{
1406	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1407	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1408	struct table_instance *ti;
1409	struct datapath *dp;
1410	u32 ufid_flags;
1411	int err;
1412
1413	err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1414				       OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1415	if (err)
1416		return err;
1417	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1418
1419	rcu_read_lock();
1420	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1421	if (!dp) {
1422		rcu_read_unlock();
1423		return -ENODEV;
1424	}
1425
1426	ti = rcu_dereference(dp->table.ti);
1427	for (;;) {
1428		struct sw_flow *flow;
1429		u32 bucket, obj;
1430
1431		bucket = cb->args[0];
1432		obj = cb->args[1];
1433		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1434		if (!flow)
1435			break;
1436
1437		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1438					   NETLINK_CB(cb->skb).portid,
1439					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1440					   OVS_FLOW_CMD_GET, ufid_flags) < 0)
1441			break;
1442
1443		cb->args[0] = bucket;
1444		cb->args[1] = obj;
1445	}
1446	rcu_read_unlock();
1447	return skb->len;
1448}
1449
1450static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1451	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1452	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1453	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1454	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1455	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1456	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1457	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1458};
1459
1460static const struct genl_small_ops dp_flow_genl_ops[] = {
1461	{ .cmd = OVS_FLOW_CMD_NEW,
1462	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1463	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1464	  .doit = ovs_flow_cmd_new
1465	},
1466	{ .cmd = OVS_FLOW_CMD_DEL,
1467	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1468	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1469	  .doit = ovs_flow_cmd_del
1470	},
1471	{ .cmd = OVS_FLOW_CMD_GET,
1472	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1473	  .flags = 0,		    /* OK for unprivileged users. */
1474	  .doit = ovs_flow_cmd_get,
1475	  .dumpit = ovs_flow_cmd_dump
1476	},
1477	{ .cmd = OVS_FLOW_CMD_SET,
1478	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1479	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1480	  .doit = ovs_flow_cmd_set,
1481	},
1482};
1483
1484static struct genl_family dp_flow_genl_family __ro_after_init = {
1485	.hdrsize = sizeof(struct ovs_header),
1486	.name = OVS_FLOW_FAMILY,
1487	.version = OVS_FLOW_VERSION,
1488	.maxattr = OVS_FLOW_ATTR_MAX,
1489	.policy = flow_policy,
1490	.netnsok = true,
1491	.parallel_ops = true,
1492	.small_ops = dp_flow_genl_ops,
1493	.n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
1494	.mcgrps = &ovs_dp_flow_multicast_group,
1495	.n_mcgrps = 1,
1496	.module = THIS_MODULE,
1497};
1498
1499static size_t ovs_dp_cmd_msg_size(void)
1500{
1501	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1502
1503	msgsize += nla_total_size(IFNAMSIZ);
1504	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1505	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1506	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1507	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1508
1509	return msgsize;
1510}
1511
1512/* Called with ovs_mutex. */
1513static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1514				u32 portid, u32 seq, u32 flags, u8 cmd)
1515{
1516	struct ovs_header *ovs_header;
1517	struct ovs_dp_stats dp_stats;
1518	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1519	int err;
1520
1521	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1522				 flags, cmd);
1523	if (!ovs_header)
1524		goto error;
1525
1526	ovs_header->dp_ifindex = get_dpifindex(dp);
1527
1528	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1529	if (err)
1530		goto nla_put_failure;
1531
1532	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1533	if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1534			  &dp_stats, OVS_DP_ATTR_PAD))
1535		goto nla_put_failure;
1536
1537	if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1538			  sizeof(struct ovs_dp_megaflow_stats),
1539			  &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1540		goto nla_put_failure;
1541
1542	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1543		goto nla_put_failure;
1544
1545	if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1546			ovs_flow_tbl_masks_cache_size(&dp->table)))
1547		goto nla_put_failure;
1548
1549	genlmsg_end(skb, ovs_header);
1550	return 0;
1551
1552nla_put_failure:
1553	genlmsg_cancel(skb, ovs_header);
1554error:
1555	return -EMSGSIZE;
1556}
1557
1558static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1559{
1560	return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1561}
1562
1563/* Called with rcu_read_lock or ovs_mutex. */
1564static struct datapath *lookup_datapath(struct net *net,
1565					const struct ovs_header *ovs_header,
1566					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1567{
1568	struct datapath *dp;
1569
1570	if (!a[OVS_DP_ATTR_NAME])
1571		dp = get_dp(net, ovs_header->dp_ifindex);
1572	else {
1573		struct vport *vport;
1574
1575		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1576		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1577	}
1578	return dp ? dp : ERR_PTR(-ENODEV);
1579}
1580
1581static void ovs_dp_reset_user_features(struct sk_buff *skb,
1582				       struct genl_info *info)
1583{
1584	struct datapath *dp;
1585
1586	dp = lookup_datapath(sock_net(skb->sk), info->userhdr,
1587			     info->attrs);
1588	if (IS_ERR(dp))
1589		return;
1590
1591	WARN(dp->user_features, "Dropping previously announced user features\n");
1592	dp->user_features = 0;
1593}
1594
1595DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1596
1597static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1598{
1599	u32 user_features = 0;
1600
1601	if (a[OVS_DP_ATTR_USER_FEATURES]) {
1602		user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1603
1604		if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1605				      OVS_DP_F_UNALIGNED |
1606				      OVS_DP_F_TC_RECIRC_SHARING))
1607			return -EOPNOTSUPP;
1608
1609#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1610		if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1611			return -EOPNOTSUPP;
1612#endif
1613	}
1614
1615	if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1616		int err;
1617		u32 cache_size;
1618
1619		cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1620		err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1621		if (err)
1622			return err;
1623	}
1624
1625	dp->user_features = user_features;
1626
1627	if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1628		static_branch_enable(&tc_recirc_sharing_support);
1629	else
1630		static_branch_disable(&tc_recirc_sharing_support);
1631
1632	return 0;
1633}
1634
1635static int ovs_dp_stats_init(struct datapath *dp)
1636{
1637	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1638	if (!dp->stats_percpu)
1639		return -ENOMEM;
1640
1641	return 0;
1642}
1643
1644static int ovs_dp_vport_init(struct datapath *dp)
1645{
1646	int i;
1647
1648	dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1649				  sizeof(struct hlist_head),
1650				  GFP_KERNEL);
1651	if (!dp->ports)
1652		return -ENOMEM;
1653
1654	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1655		INIT_HLIST_HEAD(&dp->ports[i]);
1656
1657	return 0;
1658}
1659
1660static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1661{
1662	struct nlattr **a = info->attrs;
1663	struct vport_parms parms;
1664	struct sk_buff *reply;
1665	struct datapath *dp;
1666	struct vport *vport;
1667	struct ovs_net *ovs_net;
1668	int err;
1669
1670	err = -EINVAL;
1671	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1672		goto err;
1673
1674	reply = ovs_dp_cmd_alloc_info();
1675	if (!reply)
1676		return -ENOMEM;
1677
1678	err = -ENOMEM;
1679	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1680	if (dp == NULL)
1681		goto err_destroy_reply;
1682
1683	ovs_dp_set_net(dp, sock_net(skb->sk));
1684
1685	/* Allocate table. */
1686	err = ovs_flow_tbl_init(&dp->table);
1687	if (err)
1688		goto err_destroy_dp;
1689
1690	err = ovs_dp_stats_init(dp);
1691	if (err)
 
1692		goto err_destroy_table;
 
1693
1694	err = ovs_dp_vport_init(dp);
1695	if (err)
1696		goto err_destroy_stats;
 
 
 
 
 
 
 
1697
1698	err = ovs_meters_init(dp);
1699	if (err)
1700		goto err_destroy_ports;
1701
1702	/* Set up our datapath device. */
1703	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1704	parms.type = OVS_VPORT_TYPE_INTERNAL;
1705	parms.options = NULL;
1706	parms.dp = dp;
1707	parms.port_no = OVSP_LOCAL;
1708	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1709
1710	/* So far only local changes have been made, now need the lock. */
1711	ovs_lock();
1712
1713	err = ovs_dp_change(dp, a);
1714	if (err)
1715		goto err_unlock_and_destroy_meters;
 
 
 
1716
1717	vport = new_vport(&parms);
1718	if (IS_ERR(vport)) {
1719		err = PTR_ERR(vport);
1720		if (err == -EBUSY)
1721			err = -EEXIST;
1722
1723		if (err == -EEXIST) {
1724			/* An outdated user space instance that does not understand
1725			 * the concept of user_features has attempted to create a new
1726			 * datapath and is likely to reuse it. Drop all user features.
1727			 */
1728			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1729				ovs_dp_reset_user_features(skb, info);
1730		}
1731
1732		goto err_unlock_and_destroy_meters;
1733	}
1734
1735	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1736				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1737	BUG_ON(err < 0);
1738
1739	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1740	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1741
1742	ovs_unlock();
1743
1744	ovs_notify(&dp_datapath_genl_family, reply, info);
1745	return 0;
1746
1747err_unlock_and_destroy_meters:
1748	ovs_unlock();
1749	ovs_meters_exit(dp);
1750err_destroy_ports:
1751	kfree(dp->ports);
1752err_destroy_stats:
1753	free_percpu(dp->stats_percpu);
1754err_destroy_table:
1755	ovs_flow_tbl_destroy(&dp->table);
1756err_destroy_dp:
1757	kfree(dp);
1758err_destroy_reply:
1759	kfree_skb(reply);
1760err:
1761	return err;
1762}
1763
1764/* Called with ovs_mutex. */
1765static void __dp_destroy(struct datapath *dp)
1766{
1767	struct flow_table *table = &dp->table;
1768	int i;
1769
1770	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1771		struct vport *vport;
1772		struct hlist_node *n;
1773
1774		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1775			if (vport->port_no != OVSP_LOCAL)
1776				ovs_dp_detach_port(vport);
1777	}
1778
1779	list_del_rcu(&dp->list_node);
1780
1781	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1782	 * all ports in datapath are destroyed first before freeing datapath.
1783	 */
1784	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1785
1786	/* Flush sw_flow in the tables. RCU cb only releases resource
1787	 * such as dp, ports and tables. That may avoid some issues
1788	 * such as RCU usage warning.
1789	 */
1790	table_instance_flow_flush(table, ovsl_dereference(table->ti),
1791				  ovsl_dereference(table->ufid_ti));
1792
1793	/* RCU destroy the ports, meters and flow tables. */
1794	call_rcu(&dp->rcu, destroy_dp_rcu);
1795}
1796
1797static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1798{
1799	struct sk_buff *reply;
1800	struct datapath *dp;
1801	int err;
1802
1803	reply = ovs_dp_cmd_alloc_info();
1804	if (!reply)
1805		return -ENOMEM;
1806
1807	ovs_lock();
1808	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1809	err = PTR_ERR(dp);
1810	if (IS_ERR(dp))
1811		goto err_unlock_free;
1812
1813	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1814				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1815	BUG_ON(err < 0);
1816
1817	__dp_destroy(dp);
1818	ovs_unlock();
1819
1820	ovs_notify(&dp_datapath_genl_family, reply, info);
1821
1822	return 0;
1823
1824err_unlock_free:
1825	ovs_unlock();
1826	kfree_skb(reply);
1827	return err;
1828}
1829
1830static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1831{
1832	struct sk_buff *reply;
1833	struct datapath *dp;
1834	int err;
1835
1836	reply = ovs_dp_cmd_alloc_info();
1837	if (!reply)
1838		return -ENOMEM;
1839
1840	ovs_lock();
1841	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1842	err = PTR_ERR(dp);
1843	if (IS_ERR(dp))
1844		goto err_unlock_free;
1845
1846	err = ovs_dp_change(dp, info->attrs);
1847	if (err)
1848		goto err_unlock_free;
1849
1850	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1851				   info->snd_seq, 0, OVS_DP_CMD_SET);
1852	BUG_ON(err < 0);
1853
1854	ovs_unlock();
1855	ovs_notify(&dp_datapath_genl_family, reply, info);
1856
1857	return 0;
1858
1859err_unlock_free:
1860	ovs_unlock();
1861	kfree_skb(reply);
1862	return err;
1863}
1864
1865static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1866{
1867	struct sk_buff *reply;
1868	struct datapath *dp;
1869	int err;
1870
1871	reply = ovs_dp_cmd_alloc_info();
1872	if (!reply)
1873		return -ENOMEM;
1874
1875	ovs_lock();
1876	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1877	if (IS_ERR(dp)) {
1878		err = PTR_ERR(dp);
1879		goto err_unlock_free;
1880	}
1881	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1882				   info->snd_seq, 0, OVS_DP_CMD_GET);
1883	BUG_ON(err < 0);
1884	ovs_unlock();
1885
1886	return genlmsg_reply(reply, info);
1887
1888err_unlock_free:
1889	ovs_unlock();
1890	kfree_skb(reply);
1891	return err;
1892}
1893
1894static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1895{
1896	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1897	struct datapath *dp;
1898	int skip = cb->args[0];
1899	int i = 0;
1900
1901	ovs_lock();
1902	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1903		if (i >= skip &&
1904		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1905					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1906					 OVS_DP_CMD_GET) < 0)
1907			break;
1908		i++;
1909	}
1910	ovs_unlock();
1911
1912	cb->args[0] = i;
1913
1914	return skb->len;
1915}
1916
1917static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1918	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1919	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1920	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1921	[OVS_DP_ATTR_MASKS_CACHE_SIZE] =  NLA_POLICY_RANGE(NLA_U32, 0,
1922		PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
1923};
1924
1925static const struct genl_small_ops dp_datapath_genl_ops[] = {
1926	{ .cmd = OVS_DP_CMD_NEW,
1927	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1928	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1929	  .doit = ovs_dp_cmd_new
1930	},
1931	{ .cmd = OVS_DP_CMD_DEL,
1932	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1933	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1934	  .doit = ovs_dp_cmd_del
1935	},
1936	{ .cmd = OVS_DP_CMD_GET,
1937	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1938	  .flags = 0,		    /* OK for unprivileged users. */
1939	  .doit = ovs_dp_cmd_get,
1940	  .dumpit = ovs_dp_cmd_dump
1941	},
1942	{ .cmd = OVS_DP_CMD_SET,
1943	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1944	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1945	  .doit = ovs_dp_cmd_set,
1946	},
1947};
1948
1949static struct genl_family dp_datapath_genl_family __ro_after_init = {
1950	.hdrsize = sizeof(struct ovs_header),
1951	.name = OVS_DATAPATH_FAMILY,
1952	.version = OVS_DATAPATH_VERSION,
1953	.maxattr = OVS_DP_ATTR_MAX,
1954	.policy = datapath_policy,
1955	.netnsok = true,
1956	.parallel_ops = true,
1957	.small_ops = dp_datapath_genl_ops,
1958	.n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1959	.mcgrps = &ovs_dp_datapath_multicast_group,
1960	.n_mcgrps = 1,
1961	.module = THIS_MODULE,
1962};
1963
1964/* Called with ovs_mutex or RCU read lock. */
1965static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1966				   struct net *net, u32 portid, u32 seq,
1967				   u32 flags, u8 cmd, gfp_t gfp)
1968{
1969	struct ovs_header *ovs_header;
1970	struct ovs_vport_stats vport_stats;
1971	int err;
1972
1973	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1974				 flags, cmd);
1975	if (!ovs_header)
1976		return -EMSGSIZE;
1977
1978	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1979
1980	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1981	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1982	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1983			   ovs_vport_name(vport)) ||
1984	    nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1985		goto nla_put_failure;
1986
1987	if (!net_eq(net, dev_net(vport->dev))) {
1988		int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
1989
1990		if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1991			goto nla_put_failure;
1992	}
1993
1994	ovs_vport_get_stats(vport, &vport_stats);
1995	if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1996			  sizeof(struct ovs_vport_stats), &vport_stats,
1997			  OVS_VPORT_ATTR_PAD))
1998		goto nla_put_failure;
1999
2000	if (ovs_vport_get_upcall_portids(vport, skb))
2001		goto nla_put_failure;
2002
2003	err = ovs_vport_get_options(vport, skb);
2004	if (err == -EMSGSIZE)
2005		goto error;
2006
2007	genlmsg_end(skb, ovs_header);
2008	return 0;
2009
2010nla_put_failure:
2011	err = -EMSGSIZE;
2012error:
2013	genlmsg_cancel(skb, ovs_header);
2014	return err;
2015}
2016
2017static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2018{
2019	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2020}
2021
2022/* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2023struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2024					 u32 portid, u32 seq, u8 cmd)
2025{
2026	struct sk_buff *skb;
2027	int retval;
2028
2029	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2030	if (!skb)
2031		return ERR_PTR(-ENOMEM);
2032
2033	retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2034					 GFP_KERNEL);
2035	BUG_ON(retval < 0);
2036
2037	return skb;
2038}
2039
2040/* Called with ovs_mutex or RCU read lock. */
2041static struct vport *lookup_vport(struct net *net,
2042				  const struct ovs_header *ovs_header,
2043				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2044{
2045	struct datapath *dp;
2046	struct vport *vport;
2047
2048	if (a[OVS_VPORT_ATTR_IFINDEX])
2049		return ERR_PTR(-EOPNOTSUPP);
2050	if (a[OVS_VPORT_ATTR_NAME]) {
2051		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2052		if (!vport)
2053			return ERR_PTR(-ENODEV);
2054		if (ovs_header->dp_ifindex &&
2055		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2056			return ERR_PTR(-ENODEV);
2057		return vport;
2058	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2059		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2060
2061		if (port_no >= DP_MAX_PORTS)
2062			return ERR_PTR(-EFBIG);
2063
2064		dp = get_dp(net, ovs_header->dp_ifindex);
2065		if (!dp)
2066			return ERR_PTR(-ENODEV);
2067
2068		vport = ovs_vport_ovsl_rcu(dp, port_no);
2069		if (!vport)
2070			return ERR_PTR(-ENODEV);
2071		return vport;
2072	} else
2073		return ERR_PTR(-EINVAL);
2074
2075}
2076
2077static unsigned int ovs_get_max_headroom(struct datapath *dp)
2078{
2079	unsigned int dev_headroom, max_headroom = 0;
2080	struct net_device *dev;
2081	struct vport *vport;
2082	int i;
2083
2084	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2085		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2086					 lockdep_ovsl_is_held()) {
2087			dev = vport->dev;
2088			dev_headroom = netdev_get_fwd_headroom(dev);
2089			if (dev_headroom > max_headroom)
2090				max_headroom = dev_headroom;
2091		}
2092	}
2093
2094	return max_headroom;
2095}
2096
2097/* Called with ovs_mutex */
2098static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2099{
2100	struct vport *vport;
2101	int i;
2102
2103	dp->max_headroom = new_headroom;
2104	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2105		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2106					 lockdep_ovsl_is_held())
2107			netdev_set_rx_headroom(vport->dev, new_headroom);
2108	}
2109}
2110
2111static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2112{
2113	struct nlattr **a = info->attrs;
2114	struct ovs_header *ovs_header = info->userhdr;
2115	struct vport_parms parms;
2116	struct sk_buff *reply;
2117	struct vport *vport;
2118	struct datapath *dp;
2119	unsigned int new_headroom;
2120	u32 port_no;
2121	int err;
2122
2123	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2124	    !a[OVS_VPORT_ATTR_UPCALL_PID])
2125		return -EINVAL;
2126	if (a[OVS_VPORT_ATTR_IFINDEX])
2127		return -EOPNOTSUPP;
2128
2129	port_no = a[OVS_VPORT_ATTR_PORT_NO]
2130		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2131	if (port_no >= DP_MAX_PORTS)
2132		return -EFBIG;
2133
2134	reply = ovs_vport_cmd_alloc_info();
2135	if (!reply)
2136		return -ENOMEM;
2137
2138	ovs_lock();
2139restart:
2140	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2141	err = -ENODEV;
2142	if (!dp)
2143		goto exit_unlock_free;
2144
2145	if (port_no) {
2146		vport = ovs_vport_ovsl(dp, port_no);
2147		err = -EBUSY;
2148		if (vport)
2149			goto exit_unlock_free;
2150	} else {
2151		for (port_no = 1; ; port_no++) {
2152			if (port_no >= DP_MAX_PORTS) {
2153				err = -EFBIG;
2154				goto exit_unlock_free;
2155			}
2156			vport = ovs_vport_ovsl(dp, port_no);
2157			if (!vport)
2158				break;
2159		}
2160	}
2161
2162	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2163	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2164	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2165	parms.dp = dp;
2166	parms.port_no = port_no;
2167	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2168
2169	vport = new_vport(&parms);
2170	err = PTR_ERR(vport);
2171	if (IS_ERR(vport)) {
2172		if (err == -EAGAIN)
2173			goto restart;
2174		goto exit_unlock_free;
2175	}
2176
2177	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2178				      info->snd_portid, info->snd_seq, 0,
2179				      OVS_VPORT_CMD_NEW, GFP_KERNEL);
2180
2181	new_headroom = netdev_get_fwd_headroom(vport->dev);
2182
2183	if (new_headroom > dp->max_headroom)
2184		ovs_update_headroom(dp, new_headroom);
2185	else
2186		netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2187
2188	BUG_ON(err < 0);
2189	ovs_unlock();
2190
2191	ovs_notify(&dp_vport_genl_family, reply, info);
2192	return 0;
2193
2194exit_unlock_free:
2195	ovs_unlock();
2196	kfree_skb(reply);
2197	return err;
2198}
2199
2200static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2201{
2202	struct nlattr **a = info->attrs;
2203	struct sk_buff *reply;
2204	struct vport *vport;
2205	int err;
2206
2207	reply = ovs_vport_cmd_alloc_info();
2208	if (!reply)
2209		return -ENOMEM;
2210
2211	ovs_lock();
2212	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2213	err = PTR_ERR(vport);
2214	if (IS_ERR(vport))
2215		goto exit_unlock_free;
2216
2217	if (a[OVS_VPORT_ATTR_TYPE] &&
2218	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2219		err = -EINVAL;
2220		goto exit_unlock_free;
2221	}
2222
2223	if (a[OVS_VPORT_ATTR_OPTIONS]) {
2224		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2225		if (err)
2226			goto exit_unlock_free;
2227	}
2228
2229
2230	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2231		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2232
2233		err = ovs_vport_set_upcall_portids(vport, ids);
2234		if (err)
2235			goto exit_unlock_free;
2236	}
2237
2238	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2239				      info->snd_portid, info->snd_seq, 0,
2240				      OVS_VPORT_CMD_SET, GFP_KERNEL);
2241	BUG_ON(err < 0);
2242
2243	ovs_unlock();
2244	ovs_notify(&dp_vport_genl_family, reply, info);
2245	return 0;
2246
2247exit_unlock_free:
2248	ovs_unlock();
2249	kfree_skb(reply);
2250	return err;
2251}
2252
2253static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2254{
2255	bool update_headroom = false;
2256	struct nlattr **a = info->attrs;
2257	struct sk_buff *reply;
2258	struct datapath *dp;
2259	struct vport *vport;
2260	unsigned int new_headroom;
2261	int err;
2262
2263	reply = ovs_vport_cmd_alloc_info();
2264	if (!reply)
2265		return -ENOMEM;
2266
2267	ovs_lock();
2268	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2269	err = PTR_ERR(vport);
2270	if (IS_ERR(vport))
2271		goto exit_unlock_free;
2272
2273	if (vport->port_no == OVSP_LOCAL) {
2274		err = -EINVAL;
2275		goto exit_unlock_free;
2276	}
2277
2278	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2279				      info->snd_portid, info->snd_seq, 0,
2280				      OVS_VPORT_CMD_DEL, GFP_KERNEL);
2281	BUG_ON(err < 0);
2282
2283	/* the vport deletion may trigger dp headroom update */
2284	dp = vport->dp;
2285	if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2286		update_headroom = true;
2287
2288	netdev_reset_rx_headroom(vport->dev);
2289	ovs_dp_detach_port(vport);
2290
2291	if (update_headroom) {
2292		new_headroom = ovs_get_max_headroom(dp);
2293
2294		if (new_headroom < dp->max_headroom)
2295			ovs_update_headroom(dp, new_headroom);
2296	}
2297	ovs_unlock();
2298
2299	ovs_notify(&dp_vport_genl_family, reply, info);
2300	return 0;
2301
2302exit_unlock_free:
2303	ovs_unlock();
2304	kfree_skb(reply);
2305	return err;
2306}
2307
2308static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2309{
2310	struct nlattr **a = info->attrs;
2311	struct ovs_header *ovs_header = info->userhdr;
2312	struct sk_buff *reply;
2313	struct vport *vport;
2314	int err;
2315
2316	reply = ovs_vport_cmd_alloc_info();
2317	if (!reply)
2318		return -ENOMEM;
2319
2320	rcu_read_lock();
2321	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2322	err = PTR_ERR(vport);
2323	if (IS_ERR(vport))
2324		goto exit_unlock_free;
2325	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2326				      info->snd_portid, info->snd_seq, 0,
2327				      OVS_VPORT_CMD_GET, GFP_ATOMIC);
2328	BUG_ON(err < 0);
2329	rcu_read_unlock();
2330
2331	return genlmsg_reply(reply, info);
2332
2333exit_unlock_free:
2334	rcu_read_unlock();
2335	kfree_skb(reply);
2336	return err;
2337}
2338
2339static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2340{
2341	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2342	struct datapath *dp;
2343	int bucket = cb->args[0], skip = cb->args[1];
2344	int i, j = 0;
2345
2346	rcu_read_lock();
2347	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2348	if (!dp) {
2349		rcu_read_unlock();
2350		return -ENODEV;
2351	}
2352	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2353		struct vport *vport;
2354
2355		j = 0;
2356		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2357			if (j >= skip &&
2358			    ovs_vport_cmd_fill_info(vport, skb,
2359						    sock_net(skb->sk),
2360						    NETLINK_CB(cb->skb).portid,
2361						    cb->nlh->nlmsg_seq,
2362						    NLM_F_MULTI,
2363						    OVS_VPORT_CMD_GET,
2364						    GFP_ATOMIC) < 0)
2365				goto out;
2366
2367			j++;
2368		}
2369		skip = 0;
2370	}
2371out:
2372	rcu_read_unlock();
2373
2374	cb->args[0] = i;
2375	cb->args[1] = j;
2376
2377	return skb->len;
2378}
2379
2380static void ovs_dp_masks_rebalance(struct work_struct *work)
2381{
2382	struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2383					       masks_rebalance.work);
2384	struct datapath *dp;
2385
2386	ovs_lock();
2387
2388	list_for_each_entry(dp, &ovs_net->dps, list_node)
2389		ovs_flow_masks_rebalance(&dp->table);
2390
2391	ovs_unlock();
2392
2393	schedule_delayed_work(&ovs_net->masks_rebalance,
2394			      msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2395}
2396
2397static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2398	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2399	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2400	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2401	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2402	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2403	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2404	[OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2405	[OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2406};
2407
2408static const struct genl_small_ops dp_vport_genl_ops[] = {
2409	{ .cmd = OVS_VPORT_CMD_NEW,
2410	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2411	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2412	  .doit = ovs_vport_cmd_new
2413	},
2414	{ .cmd = OVS_VPORT_CMD_DEL,
2415	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2416	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2417	  .doit = ovs_vport_cmd_del
2418	},
2419	{ .cmd = OVS_VPORT_CMD_GET,
2420	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2421	  .flags = 0,		    /* OK for unprivileged users. */
2422	  .doit = ovs_vport_cmd_get,
2423	  .dumpit = ovs_vport_cmd_dump
2424	},
2425	{ .cmd = OVS_VPORT_CMD_SET,
2426	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2427	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2428	  .doit = ovs_vport_cmd_set,
2429	},
2430};
2431
2432struct genl_family dp_vport_genl_family __ro_after_init = {
2433	.hdrsize = sizeof(struct ovs_header),
2434	.name = OVS_VPORT_FAMILY,
2435	.version = OVS_VPORT_VERSION,
2436	.maxattr = OVS_VPORT_ATTR_MAX,
2437	.policy = vport_policy,
2438	.netnsok = true,
2439	.parallel_ops = true,
2440	.small_ops = dp_vport_genl_ops,
2441	.n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
2442	.mcgrps = &ovs_dp_vport_multicast_group,
2443	.n_mcgrps = 1,
2444	.module = THIS_MODULE,
2445};
2446
2447static struct genl_family * const dp_genl_families[] = {
2448	&dp_datapath_genl_family,
2449	&dp_vport_genl_family,
2450	&dp_flow_genl_family,
2451	&dp_packet_genl_family,
2452	&dp_meter_genl_family,
2453#if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2454	&dp_ct_limit_genl_family,
2455#endif
2456};
2457
2458static void dp_unregister_genl(int n_families)
2459{
2460	int i;
2461
2462	for (i = 0; i < n_families; i++)
2463		genl_unregister_family(dp_genl_families[i]);
2464}
2465
2466static int __init dp_register_genl(void)
2467{
2468	int err;
2469	int i;
2470
2471	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2472
2473		err = genl_register_family(dp_genl_families[i]);
2474		if (err)
2475			goto error;
2476	}
2477
2478	return 0;
2479
2480error:
2481	dp_unregister_genl(i);
2482	return err;
2483}
2484
2485static int __net_init ovs_init_net(struct net *net)
2486{
2487	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2488	int err;
2489
2490	INIT_LIST_HEAD(&ovs_net->dps);
2491	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2492	INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2493
2494	err = ovs_ct_init(net);
2495	if (err)
2496		return err;
2497
2498	schedule_delayed_work(&ovs_net->masks_rebalance,
2499			      msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2500	return 0;
2501}
2502
2503static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2504					    struct list_head *head)
2505{
2506	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2507	struct datapath *dp;
2508
2509	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2510		int i;
2511
2512		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2513			struct vport *vport;
2514
2515			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2516				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2517					continue;
2518
2519				if (dev_net(vport->dev) == dnet)
2520					list_add(&vport->detach_list, head);
2521			}
2522		}
2523	}
2524}
2525
2526static void __net_exit ovs_exit_net(struct net *dnet)
2527{
2528	struct datapath *dp, *dp_next;
2529	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2530	struct vport *vport, *vport_next;
2531	struct net *net;
2532	LIST_HEAD(head);
2533
2534	ovs_lock();
2535
2536	ovs_ct_exit(dnet);
2537
2538	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2539		__dp_destroy(dp);
2540
2541	down_read(&net_rwsem);
2542	for_each_net(net)
2543		list_vports_from_net(net, dnet, &head);
2544	up_read(&net_rwsem);
2545
2546	/* Detach all vports from given namespace. */
2547	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2548		list_del(&vport->detach_list);
2549		ovs_dp_detach_port(vport);
2550	}
2551
2552	ovs_unlock();
2553
2554	cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2555	cancel_work_sync(&ovs_net->dp_notify_work);
2556}
2557
2558static struct pernet_operations ovs_net_ops = {
2559	.init = ovs_init_net,
2560	.exit = ovs_exit_net,
2561	.id   = &ovs_net_id,
2562	.size = sizeof(struct ovs_net),
2563};
2564
2565static int __init dp_init(void)
2566{
2567	int err;
2568
2569	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2570		     sizeof_field(struct sk_buff, cb));
2571
2572	pr_info("Open vSwitch switching datapath\n");
2573
2574	err = action_fifos_init();
2575	if (err)
2576		goto error;
2577
2578	err = ovs_internal_dev_rtnl_link_register();
2579	if (err)
2580		goto error_action_fifos_exit;
2581
2582	err = ovs_flow_init();
2583	if (err)
2584		goto error_unreg_rtnl_link;
2585
2586	err = ovs_vport_init();
2587	if (err)
2588		goto error_flow_exit;
2589
2590	err = register_pernet_device(&ovs_net_ops);
2591	if (err)
2592		goto error_vport_exit;
2593
2594	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2595	if (err)
2596		goto error_netns_exit;
2597
2598	err = ovs_netdev_init();
2599	if (err)
2600		goto error_unreg_notifier;
2601
2602	err = dp_register_genl();
2603	if (err < 0)
2604		goto error_unreg_netdev;
2605
2606	return 0;
2607
2608error_unreg_netdev:
2609	ovs_netdev_exit();
2610error_unreg_notifier:
2611	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2612error_netns_exit:
2613	unregister_pernet_device(&ovs_net_ops);
2614error_vport_exit:
2615	ovs_vport_exit();
2616error_flow_exit:
2617	ovs_flow_exit();
2618error_unreg_rtnl_link:
2619	ovs_internal_dev_rtnl_link_unregister();
2620error_action_fifos_exit:
2621	action_fifos_exit();
2622error:
2623	return err;
2624}
2625
2626static void dp_cleanup(void)
2627{
2628	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2629	ovs_netdev_exit();
2630	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2631	unregister_pernet_device(&ovs_net_ops);
2632	rcu_barrier();
2633	ovs_vport_exit();
2634	ovs_flow_exit();
2635	ovs_internal_dev_rtnl_link_unregister();
2636	action_fifos_exit();
2637}
2638
2639module_init(dp_init);
2640module_exit(dp_cleanup);
2641
2642MODULE_DESCRIPTION("Open vSwitch switching datapath");
2643MODULE_LICENSE("GPL");
2644MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2645MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2646MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2647MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2648MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2649MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);