Linux Audio

Check our new training course

Loading...
v5.4
   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 "vport-internal_dev.h"
  47#include "vport-netdev.h"
  48
  49unsigned int ovs_net_id __read_mostly;
  50
  51static struct genl_family dp_packet_genl_family;
  52static struct genl_family dp_flow_genl_family;
  53static struct genl_family dp_datapath_genl_family;
  54
  55static const struct nla_policy flow_policy[];
  56
  57static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
  58	.name = OVS_FLOW_MCGROUP,
  59};
  60
  61static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  62	.name = OVS_DATAPATH_MCGROUP,
  63};
  64
  65static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
  66	.name = OVS_VPORT_MCGROUP,
  67};
  68
  69/* Check if need to build a reply message.
  70 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  71static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  72			    unsigned int group)
  73{
  74	return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  75	       genl_has_listeners(family, genl_info_net(info), group);
  76}
  77
  78static void ovs_notify(struct genl_family *family,
  79		       struct sk_buff *skb, struct genl_info *info)
  80{
  81	genl_notify(family, skb, info, 0, GFP_KERNEL);
  82}
  83
  84/**
  85 * DOC: Locking:
  86 *
  87 * All writes e.g. Writes to device state (add/remove datapath, port, set
  88 * operations on vports, etc.), Writes to other state (flow table
  89 * modifications, set miscellaneous datapath parameters, etc.) are protected
  90 * by ovs_lock.
  91 *
  92 * Reads are protected by RCU.
  93 *
  94 * There are a few special cases (mostly stats) that have their own
  95 * synchronization but they nest under all of above and don't interact with
  96 * each other.
  97 *
  98 * The RTNL lock nests inside ovs_mutex.
  99 */
 100
 101static DEFINE_MUTEX(ovs_mutex);
 102
 103void ovs_lock(void)
 104{
 105	mutex_lock(&ovs_mutex);
 106}
 107
 108void ovs_unlock(void)
 109{
 110	mutex_unlock(&ovs_mutex);
 111}
 112
 113#ifdef CONFIG_LOCKDEP
 114int lockdep_ovsl_is_held(void)
 115{
 116	if (debug_locks)
 117		return lockdep_is_held(&ovs_mutex);
 118	else
 119		return 1;
 120}
 121#endif
 122
 123static struct vport *new_vport(const struct vport_parms *);
 124static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 125			     const struct sw_flow_key *,
 126			     const struct dp_upcall_info *,
 127			     uint32_t cutlen);
 128static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 129				  const struct sw_flow_key *,
 130				  const struct dp_upcall_info *,
 131				  uint32_t cutlen);
 132
 
 
 
 
 133/* Must be called with rcu_read_lock or ovs_mutex. */
 134const char *ovs_dp_name(const struct datapath *dp)
 135{
 136	struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 137	return ovs_vport_name(vport);
 138}
 139
 140static int get_dpifindex(const struct datapath *dp)
 141{
 142	struct vport *local;
 143	int ifindex;
 144
 145	rcu_read_lock();
 146
 147	local = ovs_vport_rcu(dp, OVSP_LOCAL);
 148	if (local)
 149		ifindex = local->dev->ifindex;
 150	else
 151		ifindex = 0;
 152
 153	rcu_read_unlock();
 154
 155	return ifindex;
 156}
 157
 158static void destroy_dp_rcu(struct rcu_head *rcu)
 159{
 160	struct datapath *dp = container_of(rcu, struct datapath, rcu);
 161
 162	ovs_flow_tbl_destroy(&dp->table);
 163	free_percpu(dp->stats_percpu);
 164	kfree(dp->ports);
 165	ovs_meters_exit(dp);
 
 166	kfree(dp);
 167}
 168
 169static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 170					    u16 port_no)
 171{
 172	return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 173}
 174
 175/* Called with ovs_mutex or RCU read lock. */
 176struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 177{
 178	struct vport *vport;
 179	struct hlist_head *head;
 180
 181	head = vport_hash_bucket(dp, port_no);
 182	hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 
 183		if (vport->port_no == port_no)
 184			return vport;
 185	}
 186	return NULL;
 187}
 188
 189/* Called with ovs_mutex. */
 190static struct vport *new_vport(const struct vport_parms *parms)
 191{
 192	struct vport *vport;
 193
 194	vport = ovs_vport_add(parms);
 195	if (!IS_ERR(vport)) {
 196		struct datapath *dp = parms->dp;
 197		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 198
 199		hlist_add_head_rcu(&vport->dp_hash_node, head);
 200	}
 201	return vport;
 202}
 203
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 204void ovs_dp_detach_port(struct vport *p)
 205{
 206	ASSERT_OVSL();
 207
 208	/* First drop references to device. */
 209	hlist_del_rcu(&p->dp_hash_node);
 210
 211	/* Then destroy it. */
 212	ovs_vport_del(p);
 213}
 214
 215/* Must be called with rcu_read_lock. */
 216void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 217{
 218	const struct vport *p = OVS_CB(skb)->input_vport;
 219	struct datapath *dp = p->dp;
 220	struct sw_flow *flow;
 221	struct sw_flow_actions *sf_acts;
 222	struct dp_stats_percpu *stats;
 223	u64 *stats_counter;
 224	u32 n_mask_hit;
 
 225	int error;
 226
 227	stats = this_cpu_ptr(dp->stats_percpu);
 228
 229	/* Look up flow. */
 230	flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
 
 231	if (unlikely(!flow)) {
 232		struct dp_upcall_info upcall;
 233
 234		memset(&upcall, 0, sizeof(upcall));
 235		upcall.cmd = OVS_PACKET_CMD_MISS;
 236		upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 
 
 
 
 
 
 237		upcall.mru = OVS_CB(skb)->mru;
 238		error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
 239		if (unlikely(error))
 240			kfree_skb(skb);
 241		else
 
 
 242			consume_skb(skb);
 
 
 
 
 
 243		stats_counter = &stats->n_missed;
 244		goto out;
 245	}
 246
 247	ovs_flow_stats_update(flow, key->tp.flags, skb);
 248	sf_acts = rcu_dereference(flow->sf_acts);
 249	error = ovs_execute_actions(dp, skb, sf_acts, key);
 250	if (unlikely(error))
 251		net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
 252							ovs_dp_name(dp), error);
 253
 254	stats_counter = &stats->n_hit;
 255
 256out:
 257	/* Update datapath statistics. */
 258	u64_stats_update_begin(&stats->syncp);
 259	(*stats_counter)++;
 260	stats->n_mask_hit += n_mask_hit;
 
 261	u64_stats_update_end(&stats->syncp);
 262}
 263
 264int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 265		  const struct sw_flow_key *key,
 266		  const struct dp_upcall_info *upcall_info,
 267		  uint32_t cutlen)
 268{
 269	struct dp_stats_percpu *stats;
 270	int err;
 271
 
 
 
 272	if (upcall_info->portid == 0) {
 273		err = -ENOTCONN;
 274		goto err;
 275	}
 276
 277	if (!skb_is_gso(skb))
 278		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 279	else
 280		err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
 
 
 281	if (err)
 282		goto err;
 283
 284	return 0;
 285
 286err:
 287	stats = this_cpu_ptr(dp->stats_percpu);
 288
 289	u64_stats_update_begin(&stats->syncp);
 290	stats->n_lost++;
 291	u64_stats_update_end(&stats->syncp);
 292
 293	return err;
 294}
 295
 296static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 297			     const struct sw_flow_key *key,
 298			     const struct dp_upcall_info *upcall_info,
 299				 uint32_t cutlen)
 300{
 301	unsigned int gso_type = skb_shinfo(skb)->gso_type;
 302	struct sw_flow_key later_key;
 303	struct sk_buff *segs, *nskb;
 304	int err;
 305
 306	BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
 307	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 308	if (IS_ERR(segs))
 309		return PTR_ERR(segs);
 310	if (segs == NULL)
 311		return -EINVAL;
 312
 313	if (gso_type & SKB_GSO_UDP) {
 314		/* The initial flow key extracted by ovs_flow_key_extract()
 315		 * in this case is for a first fragment, so we need to
 316		 * properly mark later fragments.
 317		 */
 318		later_key = *key;
 319		later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 320	}
 321
 322	/* Queue all of the segments. */
 323	skb = segs;
 324	do {
 325		if (gso_type & SKB_GSO_UDP && skb != segs)
 326			key = &later_key;
 327
 328		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 329		if (err)
 330			break;
 331
 332	} while ((skb = skb->next));
 333
 334	/* Free all of the segments. */
 335	skb = segs;
 336	do {
 337		nskb = skb->next;
 338		if (err)
 339			kfree_skb(skb);
 340		else
 341			consume_skb(skb);
 342	} while ((skb = nskb));
 343	return err;
 344}
 345
 346static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 347			      unsigned int hdrlen, int actions_attrlen)
 348{
 349	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 350		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 351		+ nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 352		+ nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
 
 353
 354	/* OVS_PACKET_ATTR_USERDATA */
 355	if (upcall_info->userdata)
 356		size += NLA_ALIGN(upcall_info->userdata->nla_len);
 357
 358	/* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 359	if (upcall_info->egress_tun_info)
 360		size += nla_total_size(ovs_tun_key_attr_size());
 361
 362	/* OVS_PACKET_ATTR_ACTIONS */
 363	if (upcall_info->actions_len)
 364		size += nla_total_size(actions_attrlen);
 365
 366	/* OVS_PACKET_ATTR_MRU */
 367	if (upcall_info->mru)
 368		size += nla_total_size(sizeof(upcall_info->mru));
 369
 370	return size;
 371}
 372
 373static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 374{
 375	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 376		size_t plen = NLA_ALIGN(skb->len) - skb->len;
 377
 378		if (plen > 0)
 379			skb_put_zero(skb, plen);
 380	}
 381}
 382
 383static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 384				  const struct sw_flow_key *key,
 385				  const struct dp_upcall_info *upcall_info,
 386				  uint32_t cutlen)
 387{
 388	struct ovs_header *upcall;
 389	struct sk_buff *nskb = NULL;
 390	struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 391	struct nlattr *nla;
 392	size_t len;
 393	unsigned int hlen;
 394	int err, dp_ifindex;
 
 395
 396	dp_ifindex = get_dpifindex(dp);
 397	if (!dp_ifindex)
 398		return -ENODEV;
 399
 400	if (skb_vlan_tag_present(skb)) {
 401		nskb = skb_clone(skb, GFP_ATOMIC);
 402		if (!nskb)
 403			return -ENOMEM;
 404
 405		nskb = __vlan_hwaccel_push_inside(nskb);
 406		if (!nskb)
 407			return -ENOMEM;
 408
 409		skb = nskb;
 410	}
 411
 412	if (nla_attr_size(skb->len) > USHRT_MAX) {
 413		err = -EFBIG;
 414		goto out;
 415	}
 416
 417	/* Complete checksum if needed */
 418	if (skb->ip_summed == CHECKSUM_PARTIAL &&
 419	    (err = skb_csum_hwoffload_help(skb, 0)))
 420		goto out;
 421
 422	/* Older versions of OVS user space enforce alignment of the last
 423	 * Netlink attribute to NLA_ALIGNTO which would require extensive
 424	 * padding logic. Only perform zerocopy if padding is not required.
 425	 */
 426	if (dp->user_features & OVS_DP_F_UNALIGNED)
 427		hlen = skb_zerocopy_headlen(skb);
 428	else
 429		hlen = skb->len;
 430
 431	len = upcall_msg_size(upcall_info, hlen - cutlen,
 432			      OVS_CB(skb)->acts_origlen);
 433	user_skb = genlmsg_new(len, GFP_ATOMIC);
 434	if (!user_skb) {
 435		err = -ENOMEM;
 436		goto out;
 437	}
 438
 439	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 440			     0, upcall_info->cmd);
 441	if (!upcall) {
 442		err = -EINVAL;
 443		goto out;
 444	}
 445	upcall->dp_ifindex = dp_ifindex;
 446
 447	err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 448	if (err)
 449		goto out;
 450
 451	if (upcall_info->userdata)
 452		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 453			  nla_len(upcall_info->userdata),
 454			  nla_data(upcall_info->userdata));
 455
 456	if (upcall_info->egress_tun_info) {
 457		nla = nla_nest_start_noflag(user_skb,
 458					    OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 459		if (!nla) {
 460			err = -EMSGSIZE;
 461			goto out;
 462		}
 463		err = ovs_nla_put_tunnel_info(user_skb,
 464					      upcall_info->egress_tun_info);
 465		if (err)
 466			goto out;
 467
 468		nla_nest_end(user_skb, nla);
 469	}
 470
 471	if (upcall_info->actions_len) {
 472		nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
 473		if (!nla) {
 474			err = -EMSGSIZE;
 475			goto out;
 476		}
 477		err = ovs_nla_put_actions(upcall_info->actions,
 478					  upcall_info->actions_len,
 479					  user_skb);
 480		if (!err)
 481			nla_nest_end(user_skb, nla);
 482		else
 483			nla_nest_cancel(user_skb, nla);
 484	}
 485
 486	/* Add OVS_PACKET_ATTR_MRU */
 487	if (upcall_info->mru) {
 488		if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
 489				upcall_info->mru)) {
 490			err = -ENOBUFS;
 491			goto out;
 492		}
 493		pad_packet(dp, user_skb);
 494	}
 495
 496	/* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 497	if (cutlen > 0) {
 498		if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
 499				skb->len)) {
 500			err = -ENOBUFS;
 501			goto out;
 502		}
 503		pad_packet(dp, user_skb);
 
 
 
 
 
 
 
 
 
 
 504	}
 505
 506	/* Only reserve room for attribute header, packet data is added
 507	 * in skb_zerocopy() */
 508	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 509		err = -ENOBUFS;
 510		goto out;
 511	}
 512	nla->nla_len = nla_attr_size(skb->len - cutlen);
 513
 514	err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 515	if (err)
 516		goto out;
 517
 518	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 519	pad_packet(dp, user_skb);
 520
 521	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 522
 523	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 524	user_skb = NULL;
 525out:
 526	if (err)
 527		skb_tx_error(skb);
 528	kfree_skb(user_skb);
 529	kfree_skb(nskb);
 
 530	return err;
 531}
 532
 533static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 534{
 535	struct ovs_header *ovs_header = info->userhdr;
 536	struct net *net = sock_net(skb->sk);
 537	struct nlattr **a = info->attrs;
 538	struct sw_flow_actions *acts;
 539	struct sk_buff *packet;
 540	struct sw_flow *flow;
 541	struct sw_flow_actions *sf_acts;
 542	struct datapath *dp;
 543	struct vport *input_vport;
 544	u16 mru = 0;
 
 545	int len;
 546	int err;
 547	bool log = !a[OVS_PACKET_ATTR_PROBE];
 548
 549	err = -EINVAL;
 550	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 551	    !a[OVS_PACKET_ATTR_ACTIONS])
 552		goto err;
 553
 554	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 555	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 556	err = -ENOMEM;
 557	if (!packet)
 558		goto err;
 559	skb_reserve(packet, NET_IP_ALIGN);
 560
 561	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 562
 563	/* Set packet's mru */
 564	if (a[OVS_PACKET_ATTR_MRU]) {
 565		mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 566		packet->ignore_df = 1;
 567	}
 568	OVS_CB(packet)->mru = mru;
 569
 
 
 
 
 
 
 
 
 570	/* Build an sw_flow for sending this packet. */
 571	flow = ovs_flow_alloc();
 572	err = PTR_ERR(flow);
 573	if (IS_ERR(flow))
 574		goto err_kfree_skb;
 575
 576	err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 577					     packet, &flow->key, log);
 578	if (err)
 579		goto err_flow_free;
 580
 581	err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 582				   &flow->key, &acts, log);
 583	if (err)
 584		goto err_flow_free;
 585
 586	rcu_assign_pointer(flow->sf_acts, acts);
 587	packet->priority = flow->key.phy.priority;
 588	packet->mark = flow->key.phy.skb_mark;
 589
 590	rcu_read_lock();
 591	dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 592	err = -ENODEV;
 593	if (!dp)
 594		goto err_unlock;
 595
 596	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 597	if (!input_vport)
 598		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 599
 600	if (!input_vport)
 601		goto err_unlock;
 602
 603	packet->dev = input_vport->dev;
 604	OVS_CB(packet)->input_vport = input_vport;
 605	sf_acts = rcu_dereference(flow->sf_acts);
 606
 607	local_bh_disable();
 608	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 609	local_bh_enable();
 610	rcu_read_unlock();
 611
 612	ovs_flow_free(flow, false);
 613	return err;
 614
 615err_unlock:
 616	rcu_read_unlock();
 617err_flow_free:
 618	ovs_flow_free(flow, false);
 619err_kfree_skb:
 620	kfree_skb(packet);
 621err:
 622	return err;
 623}
 624
 625static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 626	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 627	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 628	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 629	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 630	[OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 
 631};
 632
 633static const struct genl_ops dp_packet_genl_ops[] = {
 634	{ .cmd = OVS_PACKET_CMD_EXECUTE,
 635	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 636	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 637	  .doit = ovs_packet_cmd_execute
 638	}
 639};
 640
 641static struct genl_family dp_packet_genl_family __ro_after_init = {
 642	.hdrsize = sizeof(struct ovs_header),
 643	.name = OVS_PACKET_FAMILY,
 644	.version = OVS_PACKET_VERSION,
 645	.maxattr = OVS_PACKET_ATTR_MAX,
 646	.policy = packet_policy,
 647	.netnsok = true,
 648	.parallel_ops = true,
 649	.ops = dp_packet_genl_ops,
 650	.n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 
 651	.module = THIS_MODULE,
 652};
 653
 654static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 655			 struct ovs_dp_megaflow_stats *mega_stats)
 656{
 657	int i;
 658
 659	memset(mega_stats, 0, sizeof(*mega_stats));
 660
 661	stats->n_flows = ovs_flow_tbl_count(&dp->table);
 662	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 663
 664	stats->n_hit = stats->n_missed = stats->n_lost = 0;
 665
 666	for_each_possible_cpu(i) {
 667		const struct dp_stats_percpu *percpu_stats;
 668		struct dp_stats_percpu local_stats;
 669		unsigned int start;
 670
 671		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 672
 673		do {
 674			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 675			local_stats = *percpu_stats;
 676		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 677
 678		stats->n_hit += local_stats.n_hit;
 679		stats->n_missed += local_stats.n_missed;
 680		stats->n_lost += local_stats.n_lost;
 681		mega_stats->n_mask_hit += local_stats.n_mask_hit;
 
 682	}
 683}
 684
 685static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 686{
 687	return ovs_identifier_is_ufid(sfid) &&
 688	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 689}
 690
 691static bool should_fill_mask(uint32_t ufid_flags)
 692{
 693	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 694}
 695
 696static bool should_fill_actions(uint32_t ufid_flags)
 697{
 698	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 699}
 700
 701static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 702				    const struct sw_flow_id *sfid,
 703				    uint32_t ufid_flags)
 704{
 705	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 706
 707	/* OVS_FLOW_ATTR_UFID */
 
 
 708	if (sfid && ovs_identifier_is_ufid(sfid))
 709		len += nla_total_size(sfid->ufid_len);
 
 
 710
 711	/* OVS_FLOW_ATTR_KEY */
 712	if (!sfid || should_fill_key(sfid, ufid_flags))
 713		len += nla_total_size(ovs_key_attr_size());
 714
 715	/* OVS_FLOW_ATTR_MASK */
 716	if (should_fill_mask(ufid_flags))
 717		len += nla_total_size(ovs_key_attr_size());
 718
 719	/* OVS_FLOW_ATTR_ACTIONS */
 720	if (should_fill_actions(ufid_flags))
 721		len += nla_total_size(acts->orig_len);
 722
 723	return len
 724		+ nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 725		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 726		+ nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 727}
 728
 729/* Called with ovs_mutex or RCU read lock. */
 730static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 731				   struct sk_buff *skb)
 732{
 733	struct ovs_flow_stats stats;
 734	__be16 tcp_flags;
 735	unsigned long used;
 736
 737	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 738
 739	if (used &&
 740	    nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 741			      OVS_FLOW_ATTR_PAD))
 742		return -EMSGSIZE;
 743
 744	if (stats.n_packets &&
 745	    nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 746			  sizeof(struct ovs_flow_stats), &stats,
 747			  OVS_FLOW_ATTR_PAD))
 748		return -EMSGSIZE;
 749
 750	if ((u8)ntohs(tcp_flags) &&
 751	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 752		return -EMSGSIZE;
 753
 754	return 0;
 755}
 756
 757/* Called with ovs_mutex or RCU read lock. */
 758static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 759				     struct sk_buff *skb, int skb_orig_len)
 760{
 761	struct nlattr *start;
 762	int err;
 763
 764	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 765	 * this is the first flow to be dumped into 'skb'.  This is unusual for
 766	 * Netlink but individual action lists can be longer than
 767	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 768	 * The userspace caller can always fetch the actions separately if it
 769	 * really wants them.  (Most userspace callers in fact don't care.)
 770	 *
 771	 * This can only fail for dump operations because the skb is always
 772	 * properly sized for single flows.
 773	 */
 774	start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
 775	if (start) {
 776		const struct sw_flow_actions *sf_acts;
 777
 778		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 779		err = ovs_nla_put_actions(sf_acts->actions,
 780					  sf_acts->actions_len, skb);
 781
 782		if (!err)
 783			nla_nest_end(skb, start);
 784		else {
 785			if (skb_orig_len)
 786				return err;
 787
 788			nla_nest_cancel(skb, start);
 789		}
 790	} else if (skb_orig_len) {
 791		return -EMSGSIZE;
 792	}
 793
 794	return 0;
 795}
 796
 797/* Called with ovs_mutex or RCU read lock. */
 798static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 799				  struct sk_buff *skb, u32 portid,
 800				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 801{
 802	const int skb_orig_len = skb->len;
 803	struct ovs_header *ovs_header;
 804	int err;
 805
 806	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 807				 flags, cmd);
 808	if (!ovs_header)
 809		return -EMSGSIZE;
 810
 811	ovs_header->dp_ifindex = dp_ifindex;
 812
 813	err = ovs_nla_put_identifier(flow, skb);
 814	if (err)
 815		goto error;
 816
 817	if (should_fill_key(&flow->id, ufid_flags)) {
 818		err = ovs_nla_put_masked_key(flow, skb);
 819		if (err)
 820			goto error;
 821	}
 822
 823	if (should_fill_mask(ufid_flags)) {
 824		err = ovs_nla_put_mask(flow, skb);
 825		if (err)
 826			goto error;
 827	}
 828
 829	err = ovs_flow_cmd_fill_stats(flow, skb);
 830	if (err)
 831		goto error;
 832
 833	if (should_fill_actions(ufid_flags)) {
 834		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 835		if (err)
 836			goto error;
 837	}
 838
 839	genlmsg_end(skb, ovs_header);
 840	return 0;
 841
 842error:
 843	genlmsg_cancel(skb, ovs_header);
 844	return err;
 845}
 846
 847/* May not be called with RCU read lock. */
 848static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 849					       const struct sw_flow_id *sfid,
 850					       struct genl_info *info,
 851					       bool always,
 852					       uint32_t ufid_flags)
 853{
 854	struct sk_buff *skb;
 855	size_t len;
 856
 857	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 858		return NULL;
 859
 860	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 861	skb = genlmsg_new(len, GFP_KERNEL);
 862	if (!skb)
 863		return ERR_PTR(-ENOMEM);
 864
 865	return skb;
 866}
 867
 868/* Called with ovs_mutex. */
 869static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 870					       int dp_ifindex,
 871					       struct genl_info *info, u8 cmd,
 872					       bool always, u32 ufid_flags)
 873{
 874	struct sk_buff *skb;
 875	int retval;
 876
 877	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 878				      &flow->id, info, always, ufid_flags);
 879	if (IS_ERR_OR_NULL(skb))
 880		return skb;
 881
 882	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 883					info->snd_portid, info->snd_seq, 0,
 884					cmd, ufid_flags);
 885	BUG_ON(retval < 0);
 
 
 
 886	return skb;
 887}
 888
 889static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 890{
 891	struct net *net = sock_net(skb->sk);
 892	struct nlattr **a = info->attrs;
 893	struct ovs_header *ovs_header = info->userhdr;
 894	struct sw_flow *flow = NULL, *new_flow;
 895	struct sw_flow_mask mask;
 896	struct sk_buff *reply;
 897	struct datapath *dp;
 
 898	struct sw_flow_actions *acts;
 899	struct sw_flow_match match;
 900	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 901	int error;
 902	bool log = !a[OVS_FLOW_ATTR_PROBE];
 903
 904	/* Must have key and actions. */
 905	error = -EINVAL;
 906	if (!a[OVS_FLOW_ATTR_KEY]) {
 907		OVS_NLERR(log, "Flow key attr not present in new flow.");
 908		goto error;
 909	}
 910	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 911		OVS_NLERR(log, "Flow actions attr not present in new flow.");
 912		goto error;
 913	}
 914
 915	/* Most of the time we need to allocate a new flow, do it before
 916	 * locking.
 917	 */
 918	new_flow = ovs_flow_alloc();
 919	if (IS_ERR(new_flow)) {
 920		error = PTR_ERR(new_flow);
 921		goto error;
 922	}
 923
 924	/* Extract key. */
 925	ovs_match_init(&match, &new_flow->key, false, &mask);
 
 
 
 
 
 
 926	error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 927				  a[OVS_FLOW_ATTR_MASK], log);
 928	if (error)
 929		goto err_kfree_flow;
 
 
 930
 931	/* Extract flow identifier. */
 932	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 933				       &new_flow->key, log);
 934	if (error)
 935		goto err_kfree_flow;
 936
 937	/* unmasked key is needed to match when ufid is not used. */
 938	if (ovs_identifier_is_key(&new_flow->id))
 939		match.key = new_flow->id.unmasked_key;
 940
 941	ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
 942
 943	/* Validate actions. */
 944	error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 945				     &new_flow->key, &acts, log);
 946	if (error) {
 947		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 948		goto err_kfree_flow;
 949	}
 950
 951	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 952					ufid_flags);
 953	if (IS_ERR(reply)) {
 954		error = PTR_ERR(reply);
 955		goto err_kfree_acts;
 956	}
 957
 958	ovs_lock();
 959	dp = get_dp(net, ovs_header->dp_ifindex);
 960	if (unlikely(!dp)) {
 961		error = -ENODEV;
 962		goto err_unlock_ovs;
 963	}
 964
 965	/* Check if this is a duplicate flow */
 966	if (ovs_identifier_is_ufid(&new_flow->id))
 967		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
 968	if (!flow)
 969		flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
 970	if (likely(!flow)) {
 971		rcu_assign_pointer(new_flow->sf_acts, acts);
 972
 973		/* Put flow in bucket. */
 974		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 975		if (unlikely(error)) {
 976			acts = NULL;
 977			goto err_unlock_ovs;
 978		}
 979
 980		if (unlikely(reply)) {
 981			error = ovs_flow_cmd_fill_info(new_flow,
 982						       ovs_header->dp_ifindex,
 983						       reply, info->snd_portid,
 984						       info->snd_seq, 0,
 985						       OVS_FLOW_CMD_NEW,
 986						       ufid_flags);
 987			BUG_ON(error < 0);
 988		}
 989		ovs_unlock();
 990	} else {
 991		struct sw_flow_actions *old_acts;
 992
 993		/* Bail out if we're not allowed to modify an existing flow.
 994		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 995		 * because Generic Netlink treats the latter as a dump
 996		 * request.  We also accept NLM_F_EXCL in case that bug ever
 997		 * gets fixed.
 998		 */
 999		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1000							 | NLM_F_EXCL))) {
1001			error = -EEXIST;
1002			goto err_unlock_ovs;
1003		}
1004		/* The flow identifier has to be the same for flow updates.
1005		 * Look for any overlapping flow.
1006		 */
1007		if (unlikely(!ovs_flow_cmp(flow, &match))) {
1008			if (ovs_identifier_is_key(&flow->id))
1009				flow = ovs_flow_tbl_lookup_exact(&dp->table,
1010								 &match);
1011			else /* UFID matches but key is different */
1012				flow = NULL;
1013			if (!flow) {
1014				error = -ENOENT;
1015				goto err_unlock_ovs;
1016			}
1017		}
1018		/* Update actions. */
1019		old_acts = ovsl_dereference(flow->sf_acts);
1020		rcu_assign_pointer(flow->sf_acts, acts);
1021
1022		if (unlikely(reply)) {
1023			error = ovs_flow_cmd_fill_info(flow,
1024						       ovs_header->dp_ifindex,
1025						       reply, info->snd_portid,
1026						       info->snd_seq, 0,
1027						       OVS_FLOW_CMD_NEW,
1028						       ufid_flags);
1029			BUG_ON(error < 0);
1030		}
1031		ovs_unlock();
1032
1033		ovs_nla_free_flow_actions_rcu(old_acts);
1034		ovs_flow_free(new_flow, false);
1035	}
1036
1037	if (reply)
1038		ovs_notify(&dp_flow_genl_family, reply, info);
 
 
1039	return 0;
1040
1041err_unlock_ovs:
1042	ovs_unlock();
1043	kfree_skb(reply);
1044err_kfree_acts:
1045	ovs_nla_free_flow_actions(acts);
 
 
1046err_kfree_flow:
1047	ovs_flow_free(new_flow, false);
1048error:
1049	return error;
1050}
1051
1052/* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1053static noinline_for_stack struct sw_flow_actions *get_flow_actions(struct net *net,
1054						const struct nlattr *a,
1055						const struct sw_flow_key *key,
1056						const struct sw_flow_mask *mask,
1057						bool log)
 
1058{
1059	struct sw_flow_actions *acts;
1060	struct sw_flow_key masked_key;
1061	int error;
1062
1063	ovs_flow_mask_key(&masked_key, key, true, mask);
1064	error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1065	if (error) {
1066		OVS_NLERR(log,
1067			  "Actions may not be safe on all matching packets");
1068		return ERR_PTR(error);
1069	}
1070
1071	return acts;
1072}
1073
1074/* Factor out match-init and action-copy to avoid
1075 * "Wframe-larger-than=1024" warning. Because mask is only
1076 * used to get actions, we new a function to save some
1077 * stack space.
1078 *
1079 * If there are not key and action attrs, we return 0
1080 * directly. In the case, the caller will also not use the
1081 * match as before. If there is action attr, we try to get
1082 * actions and save them to *acts. Before returning from
1083 * the function, we reset the match->mask pointer. Because
1084 * we should not to return match object with dangling reference
1085 * to mask.
1086 * */
1087static noinline_for_stack int
1088ovs_nla_init_match_and_action(struct net *net,
1089			      struct sw_flow_match *match,
1090			      struct sw_flow_key *key,
1091			      struct nlattr **a,
1092			      struct sw_flow_actions **acts,
1093			      bool log)
1094{
1095	struct sw_flow_mask mask;
1096	int error = 0;
1097
1098	if (a[OVS_FLOW_ATTR_KEY]) {
1099		ovs_match_init(match, key, true, &mask);
1100		error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1101					  a[OVS_FLOW_ATTR_MASK], log);
1102		if (error)
1103			goto error;
1104	}
1105
1106	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1107		if (!a[OVS_FLOW_ATTR_KEY]) {
1108			OVS_NLERR(log,
1109				  "Flow key attribute not present in set flow.");
1110			error = -EINVAL;
1111			goto error;
1112		}
1113
1114		*acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1115					 &mask, log);
1116		if (IS_ERR(*acts)) {
1117			error = PTR_ERR(*acts);
1118			goto error;
1119		}
1120	}
1121
1122	/* On success, error is 0. */
1123error:
1124	match->mask = NULL;
1125	return error;
1126}
1127
1128static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1129{
1130	struct net *net = sock_net(skb->sk);
1131	struct nlattr **a = info->attrs;
1132	struct ovs_header *ovs_header = info->userhdr;
1133	struct sw_flow_key key;
1134	struct sw_flow *flow;
1135	struct sk_buff *reply = NULL;
1136	struct datapath *dp;
1137	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1138	struct sw_flow_match match;
1139	struct sw_flow_id sfid;
1140	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1141	int error = 0;
1142	bool log = !a[OVS_FLOW_ATTR_PROBE];
1143	bool ufid_present;
1144
1145	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1146	if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1147		OVS_NLERR(log,
1148			  "Flow set message rejected, Key attribute missing.");
1149		return -EINVAL;
1150	}
1151
1152	error = ovs_nla_init_match_and_action(net, &match, &key, a,
1153					      &acts, log);
1154	if (error)
1155		goto error;
1156
1157	if (acts) {
1158		/* Can allocate before locking if have acts. */
1159		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1160						ufid_flags);
1161		if (IS_ERR(reply)) {
1162			error = PTR_ERR(reply);
1163			goto err_kfree_acts;
1164		}
1165	}
1166
1167	ovs_lock();
1168	dp = get_dp(net, ovs_header->dp_ifindex);
1169	if (unlikely(!dp)) {
1170		error = -ENODEV;
1171		goto err_unlock_ovs;
1172	}
1173	/* Check that the flow exists. */
1174	if (ufid_present)
1175		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1176	else
1177		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1178	if (unlikely(!flow)) {
1179		error = -ENOENT;
1180		goto err_unlock_ovs;
1181	}
1182
1183	/* Update actions, if present. */
1184	if (likely(acts)) {
1185		old_acts = ovsl_dereference(flow->sf_acts);
1186		rcu_assign_pointer(flow->sf_acts, acts);
1187
1188		if (unlikely(reply)) {
1189			error = ovs_flow_cmd_fill_info(flow,
1190						       ovs_header->dp_ifindex,
1191						       reply, info->snd_portid,
1192						       info->snd_seq, 0,
1193						       OVS_FLOW_CMD_SET,
1194						       ufid_flags);
1195			BUG_ON(error < 0);
1196		}
1197	} else {
1198		/* Could not alloc without acts before locking. */
1199		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1200						info, OVS_FLOW_CMD_SET, false,
1201						ufid_flags);
1202
1203		if (IS_ERR(reply)) {
1204			error = PTR_ERR(reply);
1205			goto err_unlock_ovs;
1206		}
1207	}
1208
1209	/* Clear stats. */
1210	if (a[OVS_FLOW_ATTR_CLEAR])
1211		ovs_flow_stats_clear(flow);
1212	ovs_unlock();
1213
1214	if (reply)
1215		ovs_notify(&dp_flow_genl_family, reply, info);
1216	if (old_acts)
1217		ovs_nla_free_flow_actions_rcu(old_acts);
1218
1219	return 0;
1220
1221err_unlock_ovs:
1222	ovs_unlock();
1223	kfree_skb(reply);
1224err_kfree_acts:
1225	ovs_nla_free_flow_actions(acts);
1226error:
1227	return error;
1228}
1229
1230static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1231{
1232	struct nlattr **a = info->attrs;
1233	struct ovs_header *ovs_header = info->userhdr;
1234	struct net *net = sock_net(skb->sk);
1235	struct sw_flow_key key;
1236	struct sk_buff *reply;
1237	struct sw_flow *flow;
1238	struct datapath *dp;
1239	struct sw_flow_match match;
1240	struct sw_flow_id ufid;
1241	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1242	int err = 0;
1243	bool log = !a[OVS_FLOW_ATTR_PROBE];
1244	bool ufid_present;
1245
1246	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1247	if (a[OVS_FLOW_ATTR_KEY]) {
1248		ovs_match_init(&match, &key, true, NULL);
1249		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1250					log);
1251	} else if (!ufid_present) {
1252		OVS_NLERR(log,
1253			  "Flow get message rejected, Key attribute missing.");
1254		err = -EINVAL;
1255	}
1256	if (err)
1257		return err;
1258
1259	ovs_lock();
1260	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1261	if (!dp) {
1262		err = -ENODEV;
1263		goto unlock;
1264	}
1265
1266	if (ufid_present)
1267		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1268	else
1269		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1270	if (!flow) {
1271		err = -ENOENT;
1272		goto unlock;
1273	}
1274
1275	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1276					OVS_FLOW_CMD_GET, true, ufid_flags);
1277	if (IS_ERR(reply)) {
1278		err = PTR_ERR(reply);
1279		goto unlock;
1280	}
1281
1282	ovs_unlock();
1283	return genlmsg_reply(reply, info);
1284unlock:
1285	ovs_unlock();
1286	return err;
1287}
1288
1289static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1290{
1291	struct nlattr **a = info->attrs;
1292	struct ovs_header *ovs_header = info->userhdr;
1293	struct net *net = sock_net(skb->sk);
1294	struct sw_flow_key key;
1295	struct sk_buff *reply;
1296	struct sw_flow *flow = NULL;
1297	struct datapath *dp;
1298	struct sw_flow_match match;
1299	struct sw_flow_id ufid;
1300	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1301	int err;
1302	bool log = !a[OVS_FLOW_ATTR_PROBE];
1303	bool ufid_present;
1304
1305	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1306	if (a[OVS_FLOW_ATTR_KEY]) {
1307		ovs_match_init(&match, &key, true, NULL);
1308		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1309					NULL, log);
1310		if (unlikely(err))
1311			return err;
1312	}
1313
1314	ovs_lock();
1315	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1316	if (unlikely(!dp)) {
1317		err = -ENODEV;
1318		goto unlock;
1319	}
1320
1321	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1322		err = ovs_flow_tbl_flush(&dp->table);
1323		goto unlock;
1324	}
1325
1326	if (ufid_present)
1327		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1328	else
1329		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1330	if (unlikely(!flow)) {
1331		err = -ENOENT;
1332		goto unlock;
1333	}
1334
1335	ovs_flow_tbl_remove(&dp->table, flow);
1336	ovs_unlock();
1337
1338	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1339					&flow->id, info, false, ufid_flags);
1340	if (likely(reply)) {
1341		if (!IS_ERR(reply)) {
1342			rcu_read_lock();	/*To keep RCU checker happy. */
1343			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1344						     reply, info->snd_portid,
1345						     info->snd_seq, 0,
1346						     OVS_FLOW_CMD_DEL,
1347						     ufid_flags);
1348			rcu_read_unlock();
1349			BUG_ON(err < 0);
 
 
 
1350
1351			ovs_notify(&dp_flow_genl_family, reply, info);
1352		} else {
1353			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
 
1354		}
1355	}
1356
 
1357	ovs_flow_free(flow, true);
1358	return 0;
1359unlock:
1360	ovs_unlock();
1361	return err;
1362}
1363
1364static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1365{
1366	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1367	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1368	struct table_instance *ti;
1369	struct datapath *dp;
1370	u32 ufid_flags;
1371	int err;
1372
1373	err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1374				       OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1375	if (err)
1376		return err;
1377	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1378
1379	rcu_read_lock();
1380	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1381	if (!dp) {
1382		rcu_read_unlock();
1383		return -ENODEV;
1384	}
1385
1386	ti = rcu_dereference(dp->table.ti);
1387	for (;;) {
1388		struct sw_flow *flow;
1389		u32 bucket, obj;
1390
1391		bucket = cb->args[0];
1392		obj = cb->args[1];
1393		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1394		if (!flow)
1395			break;
1396
1397		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1398					   NETLINK_CB(cb->skb).portid,
1399					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1400					   OVS_FLOW_CMD_GET, ufid_flags) < 0)
1401			break;
1402
1403		cb->args[0] = bucket;
1404		cb->args[1] = obj;
1405	}
1406	rcu_read_unlock();
1407	return skb->len;
1408}
1409
1410static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1411	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1412	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1413	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1414	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1415	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1416	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1417	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1418};
1419
1420static const struct genl_ops dp_flow_genl_ops[] = {
1421	{ .cmd = OVS_FLOW_CMD_NEW,
1422	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1423	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1424	  .doit = ovs_flow_cmd_new
1425	},
1426	{ .cmd = OVS_FLOW_CMD_DEL,
1427	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1428	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1429	  .doit = ovs_flow_cmd_del
1430	},
1431	{ .cmd = OVS_FLOW_CMD_GET,
1432	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1433	  .flags = 0,		    /* OK for unprivileged users. */
1434	  .doit = ovs_flow_cmd_get,
1435	  .dumpit = ovs_flow_cmd_dump
1436	},
1437	{ .cmd = OVS_FLOW_CMD_SET,
1438	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1439	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1440	  .doit = ovs_flow_cmd_set,
1441	},
1442};
1443
1444static struct genl_family dp_flow_genl_family __ro_after_init = {
1445	.hdrsize = sizeof(struct ovs_header),
1446	.name = OVS_FLOW_FAMILY,
1447	.version = OVS_FLOW_VERSION,
1448	.maxattr = OVS_FLOW_ATTR_MAX,
1449	.policy = flow_policy,
1450	.netnsok = true,
1451	.parallel_ops = true,
1452	.ops = dp_flow_genl_ops,
1453	.n_ops = ARRAY_SIZE(dp_flow_genl_ops),
 
1454	.mcgrps = &ovs_dp_flow_multicast_group,
1455	.n_mcgrps = 1,
1456	.module = THIS_MODULE,
1457};
1458
1459static size_t ovs_dp_cmd_msg_size(void)
1460{
1461	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1462
1463	msgsize += nla_total_size(IFNAMSIZ);
1464	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1465	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1466	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
 
 
1467
1468	return msgsize;
1469}
1470
1471/* Called with ovs_mutex. */
1472static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1473				u32 portid, u32 seq, u32 flags, u8 cmd)
1474{
1475	struct ovs_header *ovs_header;
1476	struct ovs_dp_stats dp_stats;
1477	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1478	int err;
 
1479
1480	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1481				   flags, cmd);
1482	if (!ovs_header)
1483		goto error;
1484
1485	ovs_header->dp_ifindex = get_dpifindex(dp);
1486
1487	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1488	if (err)
1489		goto nla_put_failure;
1490
1491	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1492	if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1493			  &dp_stats, OVS_DP_ATTR_PAD))
1494		goto nla_put_failure;
1495
1496	if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1497			  sizeof(struct ovs_dp_megaflow_stats),
1498			  &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1499		goto nla_put_failure;
1500
1501	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1502		goto nla_put_failure;
1503
 
 
 
 
 
 
 
 
 
 
1504	genlmsg_end(skb, ovs_header);
1505	return 0;
1506
1507nla_put_failure:
1508	genlmsg_cancel(skb, ovs_header);
1509error:
1510	return -EMSGSIZE;
1511}
1512
1513static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1514{
1515	return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1516}
1517
1518/* Called with rcu_read_lock or ovs_mutex. */
1519static struct datapath *lookup_datapath(struct net *net,
1520					const struct ovs_header *ovs_header,
1521					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1522{
1523	struct datapath *dp;
1524
1525	if (!a[OVS_DP_ATTR_NAME])
1526		dp = get_dp(net, ovs_header->dp_ifindex);
1527	else {
1528		struct vport *vport;
1529
1530		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1531		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1532	}
1533	return dp ? dp : ERR_PTR(-ENODEV);
1534}
1535
1536static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
 
1537{
1538	struct datapath *dp;
1539
1540	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
 
1541	if (IS_ERR(dp))
1542		return;
1543
1544	WARN(dp->user_features, "Dropping previously announced user features\n");
 
1545	dp->user_features = 0;
1546}
1547
1548DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1549
1550static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1551{
1552	u32 user_features = 0;
 
1553
1554	if (a[OVS_DP_ATTR_USER_FEATURES]) {
1555		user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1556
1557		if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1558				      OVS_DP_F_UNALIGNED |
1559				      OVS_DP_F_TC_RECIRC_SHARING))
 
1560			return -EOPNOTSUPP;
1561
1562#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1563		if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1564			return -EOPNOTSUPP;
1565#endif
1566	}
1567
 
 
 
 
 
 
 
 
 
 
1568	dp->user_features = user_features;
1569
1570	if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1571		static_branch_enable(&tc_recirc_sharing_support);
1572	else
1573		static_branch_disable(&tc_recirc_sharing_support);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1574
1575	return 0;
1576}
1577
1578static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1579{
1580	struct nlattr **a = info->attrs;
1581	struct vport_parms parms;
1582	struct sk_buff *reply;
1583	struct datapath *dp;
1584	struct vport *vport;
1585	struct ovs_net *ovs_net;
1586	int err, i;
1587
1588	err = -EINVAL;
1589	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1590		goto err;
1591
1592	reply = ovs_dp_cmd_alloc_info();
1593	if (!reply)
1594		return -ENOMEM;
1595
1596	err = -ENOMEM;
1597	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1598	if (dp == NULL)
1599		goto err_free_reply;
1600
1601	ovs_dp_set_net(dp, sock_net(skb->sk));
1602
1603	/* Allocate table. */
1604	err = ovs_flow_tbl_init(&dp->table);
1605	if (err)
1606		goto err_free_dp;
1607
1608	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1609	if (!dp->stats_percpu) {
1610		err = -ENOMEM;
1611		goto err_destroy_table;
1612	}
1613
1614	dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1615				  sizeof(struct hlist_head),
1616				  GFP_KERNEL);
1617	if (!dp->ports) {
1618		err = -ENOMEM;
1619		goto err_destroy_percpu;
1620	}
1621
1622	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1623		INIT_HLIST_HEAD(&dp->ports[i]);
1624
1625	err = ovs_meters_init(dp);
1626	if (err)
1627		goto err_destroy_ports_array;
1628
1629	/* Set up our datapath device. */
1630	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1631	parms.type = OVS_VPORT_TYPE_INTERNAL;
1632	parms.options = NULL;
1633	parms.dp = dp;
1634	parms.port_no = OVSP_LOCAL;
1635	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1636
1637	err = ovs_dp_change(dp, a);
1638	if (err)
1639		goto err_destroy_meters;
1640
1641	/* So far only local changes have been made, now need the lock. */
1642	ovs_lock();
1643
 
 
 
 
1644	vport = new_vport(&parms);
1645	if (IS_ERR(vport)) {
1646		err = PTR_ERR(vport);
1647		if (err == -EBUSY)
1648			err = -EEXIST;
1649
1650		if (err == -EEXIST) {
1651			/* An outdated user space instance that does not understand
1652			 * the concept of user_features has attempted to create a new
1653			 * datapath and is likely to reuse it. Drop all user features.
1654			 */
1655			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1656				ovs_dp_reset_user_features(skb, info);
1657		}
1658
1659		goto err_destroy_meters;
1660	}
1661
1662	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1663				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1664	BUG_ON(err < 0);
1665
1666	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1667	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1668
1669	ovs_unlock();
1670
1671	ovs_notify(&dp_datapath_genl_family, reply, info);
1672	return 0;
1673
1674err_destroy_meters:
 
 
1675	ovs_unlock();
1676	ovs_meters_exit(dp);
1677err_destroy_ports_array:
1678	kfree(dp->ports);
1679err_destroy_percpu:
1680	free_percpu(dp->stats_percpu);
1681err_destroy_table:
1682	ovs_flow_tbl_destroy(&dp->table);
1683err_free_dp:
1684	kfree(dp);
1685err_free_reply:
1686	kfree_skb(reply);
1687err:
1688	return err;
1689}
1690
1691/* Called with ovs_mutex. */
1692static void __dp_destroy(struct datapath *dp)
1693{
 
1694	int i;
1695
 
 
 
1696	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1697		struct vport *vport;
1698		struct hlist_node *n;
1699
1700		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1701			if (vport->port_no != OVSP_LOCAL)
1702				ovs_dp_detach_port(vport);
1703	}
1704
1705	list_del_rcu(&dp->list_node);
1706
1707	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1708	 * all ports in datapath are destroyed first before freeing datapath.
1709	 */
1710	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1711
1712	/* RCU destroy the flow table */
 
 
 
 
 
 
 
1713	call_rcu(&dp->rcu, destroy_dp_rcu);
1714}
1715
1716static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1717{
1718	struct sk_buff *reply;
1719	struct datapath *dp;
1720	int err;
1721
1722	reply = ovs_dp_cmd_alloc_info();
1723	if (!reply)
1724		return -ENOMEM;
1725
1726	ovs_lock();
1727	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
 
1728	err = PTR_ERR(dp);
1729	if (IS_ERR(dp))
1730		goto err_unlock_free;
1731
1732	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1733				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1734	BUG_ON(err < 0);
1735
1736	__dp_destroy(dp);
1737	ovs_unlock();
1738
1739	ovs_notify(&dp_datapath_genl_family, reply, info);
1740
1741	return 0;
1742
1743err_unlock_free:
1744	ovs_unlock();
1745	kfree_skb(reply);
1746	return err;
1747}
1748
1749static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1750{
1751	struct sk_buff *reply;
1752	struct datapath *dp;
1753	int err;
1754
1755	reply = ovs_dp_cmd_alloc_info();
1756	if (!reply)
1757		return -ENOMEM;
1758
1759	ovs_lock();
1760	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
 
1761	err = PTR_ERR(dp);
1762	if (IS_ERR(dp))
1763		goto err_unlock_free;
1764
1765	err = ovs_dp_change(dp, info->attrs);
1766	if (err)
1767		goto err_unlock_free;
1768
1769	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1770				   info->snd_seq, 0, OVS_DP_CMD_SET);
1771	BUG_ON(err < 0);
1772
1773	ovs_unlock();
1774	ovs_notify(&dp_datapath_genl_family, reply, info);
1775
1776	return 0;
1777
1778err_unlock_free:
1779	ovs_unlock();
1780	kfree_skb(reply);
1781	return err;
1782}
1783
1784static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1785{
1786	struct sk_buff *reply;
1787	struct datapath *dp;
1788	int err;
1789
1790	reply = ovs_dp_cmd_alloc_info();
1791	if (!reply)
1792		return -ENOMEM;
1793
1794	ovs_lock();
1795	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
 
1796	if (IS_ERR(dp)) {
1797		err = PTR_ERR(dp);
1798		goto err_unlock_free;
1799	}
1800	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1801				   info->snd_seq, 0, OVS_DP_CMD_GET);
1802	BUG_ON(err < 0);
1803	ovs_unlock();
1804
1805	return genlmsg_reply(reply, info);
1806
1807err_unlock_free:
1808	ovs_unlock();
1809	kfree_skb(reply);
1810	return err;
1811}
1812
1813static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1814{
1815	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1816	struct datapath *dp;
1817	int skip = cb->args[0];
1818	int i = 0;
1819
1820	ovs_lock();
1821	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1822		if (i >= skip &&
1823		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1824					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1825					 OVS_DP_CMD_GET) < 0)
1826			break;
1827		i++;
1828	}
1829	ovs_unlock();
1830
1831	cb->args[0] = i;
1832
1833	return skb->len;
1834}
1835
1836static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1837	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1838	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1839	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
 
 
 
1840};
1841
1842static const struct genl_ops dp_datapath_genl_ops[] = {
1843	{ .cmd = OVS_DP_CMD_NEW,
1844	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1845	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1846	  .doit = ovs_dp_cmd_new
1847	},
1848	{ .cmd = OVS_DP_CMD_DEL,
1849	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1850	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1851	  .doit = ovs_dp_cmd_del
1852	},
1853	{ .cmd = OVS_DP_CMD_GET,
1854	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1855	  .flags = 0,		    /* OK for unprivileged users. */
1856	  .doit = ovs_dp_cmd_get,
1857	  .dumpit = ovs_dp_cmd_dump
1858	},
1859	{ .cmd = OVS_DP_CMD_SET,
1860	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1861	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1862	  .doit = ovs_dp_cmd_set,
1863	},
1864};
1865
1866static struct genl_family dp_datapath_genl_family __ro_after_init = {
1867	.hdrsize = sizeof(struct ovs_header),
1868	.name = OVS_DATAPATH_FAMILY,
1869	.version = OVS_DATAPATH_VERSION,
1870	.maxattr = OVS_DP_ATTR_MAX,
1871	.policy = datapath_policy,
1872	.netnsok = true,
1873	.parallel_ops = true,
1874	.ops = dp_datapath_genl_ops,
1875	.n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
 
1876	.mcgrps = &ovs_dp_datapath_multicast_group,
1877	.n_mcgrps = 1,
1878	.module = THIS_MODULE,
1879};
1880
1881/* Called with ovs_mutex or RCU read lock. */
1882static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1883				   struct net *net, u32 portid, u32 seq,
1884				   u32 flags, u8 cmd, gfp_t gfp)
1885{
1886	struct ovs_header *ovs_header;
1887	struct ovs_vport_stats vport_stats;
 
1888	int err;
1889
1890	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1891				 flags, cmd);
1892	if (!ovs_header)
1893		return -EMSGSIZE;
1894
1895	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1896
1897	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1898	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1899	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1900			   ovs_vport_name(vport)) ||
1901	    nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1902		goto nla_put_failure;
1903
1904	if (!net_eq(net, dev_net(vport->dev))) {
1905		int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
 
 
1906
1907		if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1908			goto nla_put_failure;
1909	}
 
1910
1911	ovs_vport_get_stats(vport, &vport_stats);
1912	if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1913			  sizeof(struct ovs_vport_stats), &vport_stats,
1914			  OVS_VPORT_ATTR_PAD))
1915		goto nla_put_failure;
1916
 
 
 
1917	if (ovs_vport_get_upcall_portids(vport, skb))
1918		goto nla_put_failure;
1919
1920	err = ovs_vport_get_options(vport, skb);
1921	if (err == -EMSGSIZE)
1922		goto error;
1923
1924	genlmsg_end(skb, ovs_header);
1925	return 0;
1926
 
 
1927nla_put_failure:
1928	err = -EMSGSIZE;
1929error:
1930	genlmsg_cancel(skb, ovs_header);
1931	return err;
1932}
1933
1934static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1935{
1936	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1937}
1938
1939/* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1940struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1941					 u32 portid, u32 seq, u8 cmd)
1942{
1943	struct sk_buff *skb;
1944	int retval;
1945
1946	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1947	if (!skb)
1948		return ERR_PTR(-ENOMEM);
1949
1950	retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
1951					 GFP_KERNEL);
1952	BUG_ON(retval < 0);
1953
1954	return skb;
1955}
1956
1957/* Called with ovs_mutex or RCU read lock. */
1958static struct vport *lookup_vport(struct net *net,
1959				  const struct ovs_header *ovs_header,
1960				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1961{
1962	struct datapath *dp;
1963	struct vport *vport;
1964
1965	if (a[OVS_VPORT_ATTR_IFINDEX])
1966		return ERR_PTR(-EOPNOTSUPP);
1967	if (a[OVS_VPORT_ATTR_NAME]) {
1968		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1969		if (!vport)
1970			return ERR_PTR(-ENODEV);
1971		if (ovs_header->dp_ifindex &&
1972		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1973			return ERR_PTR(-ENODEV);
1974		return vport;
1975	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1976		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1977
1978		if (port_no >= DP_MAX_PORTS)
1979			return ERR_PTR(-EFBIG);
1980
1981		dp = get_dp(net, ovs_header->dp_ifindex);
1982		if (!dp)
1983			return ERR_PTR(-ENODEV);
1984
1985		vport = ovs_vport_ovsl_rcu(dp, port_no);
1986		if (!vport)
1987			return ERR_PTR(-ENODEV);
1988		return vport;
1989	} else
1990		return ERR_PTR(-EINVAL);
1991
1992}
1993
1994static unsigned int ovs_get_max_headroom(struct datapath *dp)
1995{
1996	unsigned int dev_headroom, max_headroom = 0;
1997	struct net_device *dev;
1998	struct vport *vport;
1999	int i;
2000
2001	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2002		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
 
2003			dev = vport->dev;
2004			dev_headroom = netdev_get_fwd_headroom(dev);
2005			if (dev_headroom > max_headroom)
2006				max_headroom = dev_headroom;
2007		}
2008	}
2009
2010	return max_headroom;
2011}
2012
2013/* Called with ovs_mutex */
2014static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2015{
2016	struct vport *vport;
2017	int i;
2018
2019	dp->max_headroom = new_headroom;
2020	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
2021		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
 
2022			netdev_set_rx_headroom(vport->dev, new_headroom);
 
2023}
2024
2025static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2026{
2027	struct nlattr **a = info->attrs;
2028	struct ovs_header *ovs_header = info->userhdr;
2029	struct vport_parms parms;
2030	struct sk_buff *reply;
2031	struct vport *vport;
2032	struct datapath *dp;
2033	unsigned int new_headroom;
2034	u32 port_no;
2035	int err;
2036
2037	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2038	    !a[OVS_VPORT_ATTR_UPCALL_PID])
2039		return -EINVAL;
2040	if (a[OVS_VPORT_ATTR_IFINDEX])
 
 
 
2041		return -EOPNOTSUPP;
2042
2043	port_no = a[OVS_VPORT_ATTR_PORT_NO]
2044		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2045	if (port_no >= DP_MAX_PORTS)
2046		return -EFBIG;
2047
2048	reply = ovs_vport_cmd_alloc_info();
2049	if (!reply)
2050		return -ENOMEM;
2051
2052	ovs_lock();
2053restart:
2054	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2055	err = -ENODEV;
2056	if (!dp)
2057		goto exit_unlock_free;
2058
2059	if (port_no) {
2060		vport = ovs_vport_ovsl(dp, port_no);
2061		err = -EBUSY;
2062		if (vport)
2063			goto exit_unlock_free;
2064	} else {
2065		for (port_no = 1; ; port_no++) {
2066			if (port_no >= DP_MAX_PORTS) {
2067				err = -EFBIG;
2068				goto exit_unlock_free;
2069			}
2070			vport = ovs_vport_ovsl(dp, port_no);
2071			if (!vport)
2072				break;
2073		}
2074	}
2075
2076	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2077	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2078	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2079	parms.dp = dp;
2080	parms.port_no = port_no;
2081	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
 
 
2082
2083	vport = new_vport(&parms);
2084	err = PTR_ERR(vport);
2085	if (IS_ERR(vport)) {
2086		if (err == -EAGAIN)
2087			goto restart;
2088		goto exit_unlock_free;
2089	}
2090
2091	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2092				      info->snd_portid, info->snd_seq, 0,
2093				      OVS_VPORT_CMD_NEW, GFP_KERNEL);
2094
2095	new_headroom = netdev_get_fwd_headroom(vport->dev);
2096
2097	if (new_headroom > dp->max_headroom)
2098		ovs_update_headroom(dp, new_headroom);
2099	else
2100		netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2101
2102	BUG_ON(err < 0);
2103	ovs_unlock();
2104
2105	ovs_notify(&dp_vport_genl_family, reply, info);
2106	return 0;
2107
2108exit_unlock_free:
2109	ovs_unlock();
2110	kfree_skb(reply);
2111	return err;
2112}
2113
2114static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2115{
2116	struct nlattr **a = info->attrs;
2117	struct sk_buff *reply;
2118	struct vport *vport;
2119	int err;
2120
2121	reply = ovs_vport_cmd_alloc_info();
2122	if (!reply)
2123		return -ENOMEM;
2124
2125	ovs_lock();
2126	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2127	err = PTR_ERR(vport);
2128	if (IS_ERR(vport))
2129		goto exit_unlock_free;
2130
2131	if (a[OVS_VPORT_ATTR_TYPE] &&
2132	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2133		err = -EINVAL;
2134		goto exit_unlock_free;
2135	}
2136
2137	if (a[OVS_VPORT_ATTR_OPTIONS]) {
2138		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2139		if (err)
2140			goto exit_unlock_free;
2141	}
2142
2143
2144	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2145		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2146
2147		err = ovs_vport_set_upcall_portids(vport, ids);
2148		if (err)
2149			goto exit_unlock_free;
2150	}
2151
2152	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2153				      info->snd_portid, info->snd_seq, 0,
2154				      OVS_VPORT_CMD_SET, GFP_KERNEL);
2155	BUG_ON(err < 0);
2156
2157	ovs_unlock();
2158	ovs_notify(&dp_vport_genl_family, reply, info);
2159	return 0;
2160
2161exit_unlock_free:
2162	ovs_unlock();
2163	kfree_skb(reply);
2164	return err;
2165}
2166
2167static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2168{
2169	bool update_headroom = false;
2170	struct nlattr **a = info->attrs;
2171	struct sk_buff *reply;
2172	struct datapath *dp;
2173	struct vport *vport;
2174	unsigned int new_headroom;
2175	int err;
2176
2177	reply = ovs_vport_cmd_alloc_info();
2178	if (!reply)
2179		return -ENOMEM;
2180
2181	ovs_lock();
2182	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2183	err = PTR_ERR(vport);
2184	if (IS_ERR(vport))
2185		goto exit_unlock_free;
2186
2187	if (vport->port_no == OVSP_LOCAL) {
2188		err = -EINVAL;
2189		goto exit_unlock_free;
2190	}
2191
2192	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2193				      info->snd_portid, info->snd_seq, 0,
2194				      OVS_VPORT_CMD_DEL, GFP_KERNEL);
2195	BUG_ON(err < 0);
2196
2197	/* the vport deletion may trigger dp headroom update */
2198	dp = vport->dp;
2199	if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2200		update_headroom = true;
2201
2202	netdev_reset_rx_headroom(vport->dev);
2203	ovs_dp_detach_port(vport);
2204
2205	if (update_headroom) {
2206		new_headroom = ovs_get_max_headroom(dp);
2207
2208		if (new_headroom < dp->max_headroom)
2209			ovs_update_headroom(dp, new_headroom);
2210	}
2211	ovs_unlock();
2212
2213	ovs_notify(&dp_vport_genl_family, reply, info);
2214	return 0;
2215
2216exit_unlock_free:
2217	ovs_unlock();
2218	kfree_skb(reply);
2219	return err;
2220}
2221
2222static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2223{
2224	struct nlattr **a = info->attrs;
2225	struct ovs_header *ovs_header = info->userhdr;
2226	struct sk_buff *reply;
2227	struct vport *vport;
2228	int err;
2229
2230	reply = ovs_vport_cmd_alloc_info();
2231	if (!reply)
2232		return -ENOMEM;
2233
2234	rcu_read_lock();
2235	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2236	err = PTR_ERR(vport);
2237	if (IS_ERR(vport))
2238		goto exit_unlock_free;
2239	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2240				      info->snd_portid, info->snd_seq, 0,
2241				      OVS_VPORT_CMD_GET, GFP_ATOMIC);
2242	BUG_ON(err < 0);
2243	rcu_read_unlock();
2244
2245	return genlmsg_reply(reply, info);
2246
2247exit_unlock_free:
2248	rcu_read_unlock();
2249	kfree_skb(reply);
2250	return err;
2251}
2252
2253static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2254{
2255	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2256	struct datapath *dp;
2257	int bucket = cb->args[0], skip = cb->args[1];
2258	int i, j = 0;
2259
2260	rcu_read_lock();
2261	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2262	if (!dp) {
2263		rcu_read_unlock();
2264		return -ENODEV;
2265	}
2266	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2267		struct vport *vport;
2268
2269		j = 0;
2270		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2271			if (j >= skip &&
2272			    ovs_vport_cmd_fill_info(vport, skb,
2273						    sock_net(skb->sk),
2274						    NETLINK_CB(cb->skb).portid,
2275						    cb->nlh->nlmsg_seq,
2276						    NLM_F_MULTI,
2277						    OVS_VPORT_CMD_GET,
2278						    GFP_ATOMIC) < 0)
2279				goto out;
2280
2281			j++;
2282		}
2283		skip = 0;
2284	}
2285out:
2286	rcu_read_unlock();
2287
2288	cb->args[0] = i;
2289	cb->args[1] = j;
2290
2291	return skb->len;
2292}
2293
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2294static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2295	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2296	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2297	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2298	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2299	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2300	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2301	[OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2302	[OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
 
2303};
2304
2305static const struct genl_ops dp_vport_genl_ops[] = {
2306	{ .cmd = OVS_VPORT_CMD_NEW,
2307	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2308	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2309	  .doit = ovs_vport_cmd_new
2310	},
2311	{ .cmd = OVS_VPORT_CMD_DEL,
2312	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2313	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2314	  .doit = ovs_vport_cmd_del
2315	},
2316	{ .cmd = OVS_VPORT_CMD_GET,
2317	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2318	  .flags = 0,		    /* OK for unprivileged users. */
2319	  .doit = ovs_vport_cmd_get,
2320	  .dumpit = ovs_vport_cmd_dump
2321	},
2322	{ .cmd = OVS_VPORT_CMD_SET,
2323	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2324	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2325	  .doit = ovs_vport_cmd_set,
2326	},
2327};
2328
2329struct genl_family dp_vport_genl_family __ro_after_init = {
2330	.hdrsize = sizeof(struct ovs_header),
2331	.name = OVS_VPORT_FAMILY,
2332	.version = OVS_VPORT_VERSION,
2333	.maxattr = OVS_VPORT_ATTR_MAX,
2334	.policy = vport_policy,
2335	.netnsok = true,
2336	.parallel_ops = true,
2337	.ops = dp_vport_genl_ops,
2338	.n_ops = ARRAY_SIZE(dp_vport_genl_ops),
 
2339	.mcgrps = &ovs_dp_vport_multicast_group,
2340	.n_mcgrps = 1,
2341	.module = THIS_MODULE,
2342};
2343
2344static struct genl_family * const dp_genl_families[] = {
2345	&dp_datapath_genl_family,
2346	&dp_vport_genl_family,
2347	&dp_flow_genl_family,
2348	&dp_packet_genl_family,
2349	&dp_meter_genl_family,
2350#if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2351	&dp_ct_limit_genl_family,
2352#endif
2353};
2354
2355static void dp_unregister_genl(int n_families)
2356{
2357	int i;
2358
2359	for (i = 0; i < n_families; i++)
2360		genl_unregister_family(dp_genl_families[i]);
2361}
2362
2363static int __init dp_register_genl(void)
2364{
2365	int err;
2366	int i;
2367
2368	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2369
2370		err = genl_register_family(dp_genl_families[i]);
2371		if (err)
2372			goto error;
2373	}
2374
2375	return 0;
2376
2377error:
2378	dp_unregister_genl(i);
2379	return err;
2380}
2381
2382static int __net_init ovs_init_net(struct net *net)
2383{
2384	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 
2385
2386	INIT_LIST_HEAD(&ovs_net->dps);
2387	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2388	return ovs_ct_init(net);
 
 
 
 
 
 
 
 
2389}
2390
2391static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2392					    struct list_head *head)
2393{
2394	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2395	struct datapath *dp;
2396
2397	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2398		int i;
2399
2400		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2401			struct vport *vport;
2402
2403			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2404				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2405					continue;
2406
2407				if (dev_net(vport->dev) == dnet)
2408					list_add(&vport->detach_list, head);
2409			}
2410		}
2411	}
2412}
2413
2414static void __net_exit ovs_exit_net(struct net *dnet)
2415{
2416	struct datapath *dp, *dp_next;
2417	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2418	struct vport *vport, *vport_next;
2419	struct net *net;
2420	LIST_HEAD(head);
2421
2422	ovs_ct_exit(dnet);
2423	ovs_lock();
 
 
 
2424	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2425		__dp_destroy(dp);
2426
2427	down_read(&net_rwsem);
2428	for_each_net(net)
2429		list_vports_from_net(net, dnet, &head);
2430	up_read(&net_rwsem);
2431
2432	/* Detach all vports from given namespace. */
2433	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2434		list_del(&vport->detach_list);
2435		ovs_dp_detach_port(vport);
2436	}
2437
2438	ovs_unlock();
2439
 
2440	cancel_work_sync(&ovs_net->dp_notify_work);
2441}
2442
2443static struct pernet_operations ovs_net_ops = {
2444	.init = ovs_init_net,
2445	.exit = ovs_exit_net,
2446	.id   = &ovs_net_id,
2447	.size = sizeof(struct ovs_net),
2448};
2449
 
 
 
 
 
 
 
 
 
 
 
2450static int __init dp_init(void)
2451{
2452	int err;
2453
2454	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
 
2455
2456	pr_info("Open vSwitch switching datapath\n");
2457
2458	err = action_fifos_init();
2459	if (err)
2460		goto error;
2461
2462	err = ovs_internal_dev_rtnl_link_register();
2463	if (err)
2464		goto error_action_fifos_exit;
2465
2466	err = ovs_flow_init();
2467	if (err)
2468		goto error_unreg_rtnl_link;
2469
2470	err = ovs_vport_init();
2471	if (err)
2472		goto error_flow_exit;
2473
2474	err = register_pernet_device(&ovs_net_ops);
2475	if (err)
2476		goto error_vport_exit;
2477
2478	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2479	if (err)
2480		goto error_netns_exit;
2481
2482	err = ovs_netdev_init();
2483	if (err)
2484		goto error_unreg_notifier;
2485
2486	err = dp_register_genl();
2487	if (err < 0)
2488		goto error_unreg_netdev;
2489
 
 
 
2490	return 0;
2491
2492error_unreg_netdev:
2493	ovs_netdev_exit();
2494error_unreg_notifier:
2495	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2496error_netns_exit:
2497	unregister_pernet_device(&ovs_net_ops);
2498error_vport_exit:
2499	ovs_vport_exit();
2500error_flow_exit:
2501	ovs_flow_exit();
2502error_unreg_rtnl_link:
2503	ovs_internal_dev_rtnl_link_unregister();
2504error_action_fifos_exit:
2505	action_fifos_exit();
2506error:
2507	return err;
2508}
2509
2510static void dp_cleanup(void)
2511{
2512	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2513	ovs_netdev_exit();
2514	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2515	unregister_pernet_device(&ovs_net_ops);
 
2516	rcu_barrier();
2517	ovs_vport_exit();
2518	ovs_flow_exit();
2519	ovs_internal_dev_rtnl_link_unregister();
2520	action_fifos_exit();
2521}
2522
2523module_init(dp_init);
2524module_exit(dp_cleanup);
2525
2526MODULE_DESCRIPTION("Open vSwitch switching datapath");
2527MODULE_LICENSE("GPL");
2528MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2529MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2530MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2531MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2532MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2533MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);
v6.13.7
   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/kernel.h>
  19#include <linux/kthread.h>
  20#include <linux/mutex.h>
  21#include <linux/percpu.h>
  22#include <linux/rcupdate.h>
  23#include <linux/tcp.h>
  24#include <linux/udp.h>
  25#include <linux/ethtool.h>
  26#include <linux/wait.h>
  27#include <asm/div64.h>
  28#include <linux/highmem.h>
  29#include <linux/netfilter_bridge.h>
  30#include <linux/netfilter_ipv4.h>
  31#include <linux/inetdevice.h>
  32#include <linux/list.h>
  33#include <linux/openvswitch.h>
  34#include <linux/rculist.h>
  35#include <linux/dmi.h>
  36#include <net/genetlink.h>
  37#include <net/gso.h>
  38#include <net/net_namespace.h>
  39#include <net/netns/generic.h>
  40#include <net/pkt_cls.h>
  41
  42#include "datapath.h"
  43#include "drop.h"
  44#include "flow.h"
  45#include "flow_table.h"
  46#include "flow_netlink.h"
  47#include "meter.h"
  48#include "openvswitch_trace.h"
  49#include "vport-internal_dev.h"
  50#include "vport-netdev.h"
  51
  52unsigned int ovs_net_id __read_mostly;
  53
  54static struct genl_family dp_packet_genl_family;
  55static struct genl_family dp_flow_genl_family;
  56static struct genl_family dp_datapath_genl_family;
  57
  58static const struct nla_policy flow_policy[];
  59
  60static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
  61	.name = OVS_FLOW_MCGROUP,
  62};
  63
  64static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  65	.name = OVS_DATAPATH_MCGROUP,
  66};
  67
  68static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
  69	.name = OVS_VPORT_MCGROUP,
  70};
  71
  72/* Check if need to build a reply message.
  73 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  74static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  75			    unsigned int group)
  76{
  77	return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  78	       genl_has_listeners(family, genl_info_net(info), group);
  79}
  80
  81static void ovs_notify(struct genl_family *family,
  82		       struct sk_buff *skb, struct genl_info *info)
  83{
  84	genl_notify(family, skb, info, 0, GFP_KERNEL);
  85}
  86
  87/**
  88 * DOC: Locking:
  89 *
  90 * All writes e.g. Writes to device state (add/remove datapath, port, set
  91 * operations on vports, etc.), Writes to other state (flow table
  92 * modifications, set miscellaneous datapath parameters, etc.) are protected
  93 * by ovs_lock.
  94 *
  95 * Reads are protected by RCU.
  96 *
  97 * There are a few special cases (mostly stats) that have their own
  98 * synchronization but they nest under all of above and don't interact with
  99 * each other.
 100 *
 101 * The RTNL lock nests inside ovs_mutex.
 102 */
 103
 104static DEFINE_MUTEX(ovs_mutex);
 105
 106void ovs_lock(void)
 107{
 108	mutex_lock(&ovs_mutex);
 109}
 110
 111void ovs_unlock(void)
 112{
 113	mutex_unlock(&ovs_mutex);
 114}
 115
 116#ifdef CONFIG_LOCKDEP
 117int lockdep_ovsl_is_held(void)
 118{
 119	if (debug_locks)
 120		return lockdep_is_held(&ovs_mutex);
 121	else
 122		return 1;
 123}
 124#endif
 125
 126static struct vport *new_vport(const struct vport_parms *);
 127static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 128			     const struct sw_flow_key *,
 129			     const struct dp_upcall_info *,
 130			     uint32_t cutlen);
 131static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 132				  const struct sw_flow_key *,
 133				  const struct dp_upcall_info *,
 134				  uint32_t cutlen);
 135
 136static void ovs_dp_masks_rebalance(struct work_struct *work);
 137
 138static int ovs_dp_set_upcall_portids(struct datapath *, const struct nlattr *);
 139
 140/* Must be called with rcu_read_lock or ovs_mutex. */
 141const char *ovs_dp_name(const struct datapath *dp)
 142{
 143	struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 144	return ovs_vport_name(vport);
 145}
 146
 147static int get_dpifindex(const struct datapath *dp)
 148{
 149	struct vport *local;
 150	int ifindex;
 151
 152	rcu_read_lock();
 153
 154	local = ovs_vport_rcu(dp, OVSP_LOCAL);
 155	if (local)
 156		ifindex = local->dev->ifindex;
 157	else
 158		ifindex = 0;
 159
 160	rcu_read_unlock();
 161
 162	return ifindex;
 163}
 164
 165static void destroy_dp_rcu(struct rcu_head *rcu)
 166{
 167	struct datapath *dp = container_of(rcu, struct datapath, rcu);
 168
 169	ovs_flow_tbl_destroy(&dp->table);
 170	free_percpu(dp->stats_percpu);
 171	kfree(dp->ports);
 172	ovs_meters_exit(dp);
 173	kfree(rcu_dereference_raw(dp->upcall_portids));
 174	kfree(dp);
 175}
 176
 177static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 178					    u16 port_no)
 179{
 180	return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 181}
 182
 183/* Called with ovs_mutex or RCU read lock. */
 184struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 185{
 186	struct vport *vport;
 187	struct hlist_head *head;
 188
 189	head = vport_hash_bucket(dp, port_no);
 190	hlist_for_each_entry_rcu(vport, head, dp_hash_node,
 191				 lockdep_ovsl_is_held()) {
 192		if (vport->port_no == port_no)
 193			return vport;
 194	}
 195	return NULL;
 196}
 197
 198/* Called with ovs_mutex. */
 199static struct vport *new_vport(const struct vport_parms *parms)
 200{
 201	struct vport *vport;
 202
 203	vport = ovs_vport_add(parms);
 204	if (!IS_ERR(vport)) {
 205		struct datapath *dp = parms->dp;
 206		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 207
 208		hlist_add_head_rcu(&vport->dp_hash_node, head);
 209	}
 210	return vport;
 211}
 212
 213static void ovs_vport_update_upcall_stats(struct sk_buff *skb,
 214					  const struct dp_upcall_info *upcall_info,
 215					  bool upcall_result)
 216{
 217	struct vport *p = OVS_CB(skb)->input_vport;
 218	struct vport_upcall_stats_percpu *stats;
 219
 220	if (upcall_info->cmd != OVS_PACKET_CMD_MISS &&
 221	    upcall_info->cmd != OVS_PACKET_CMD_ACTION)
 222		return;
 223
 224	stats = this_cpu_ptr(p->upcall_stats);
 225	u64_stats_update_begin(&stats->syncp);
 226	if (upcall_result)
 227		u64_stats_inc(&stats->n_success);
 228	else
 229		u64_stats_inc(&stats->n_fail);
 230	u64_stats_update_end(&stats->syncp);
 231}
 232
 233void ovs_dp_detach_port(struct vport *p)
 234{
 235	ASSERT_OVSL();
 236
 237	/* First drop references to device. */
 238	hlist_del_rcu(&p->dp_hash_node);
 239
 240	/* Then destroy it. */
 241	ovs_vport_del(p);
 242}
 243
 244/* Must be called with rcu_read_lock. */
 245void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 246{
 247	const struct vport *p = OVS_CB(skb)->input_vport;
 248	struct datapath *dp = p->dp;
 249	struct sw_flow *flow;
 250	struct sw_flow_actions *sf_acts;
 251	struct dp_stats_percpu *stats;
 252	u64 *stats_counter;
 253	u32 n_mask_hit;
 254	u32 n_cache_hit;
 255	int error;
 256
 257	stats = this_cpu_ptr(dp->stats_percpu);
 258
 259	/* Look up flow. */
 260	flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
 261					 &n_mask_hit, &n_cache_hit);
 262	if (unlikely(!flow)) {
 263		struct dp_upcall_info upcall;
 264
 265		memset(&upcall, 0, sizeof(upcall));
 266		upcall.cmd = OVS_PACKET_CMD_MISS;
 267
 268		if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
 269			upcall.portid =
 270			    ovs_dp_get_upcall_portid(dp, smp_processor_id());
 271		else
 272			upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 273
 274		upcall.mru = OVS_CB(skb)->mru;
 275		error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
 276		switch (error) {
 277		case 0:
 278		case -EAGAIN:
 279		case -ERESTARTSYS:
 280		case -EINTR:
 281			consume_skb(skb);
 282			break;
 283		default:
 284			kfree_skb(skb);
 285			break;
 286		}
 287		stats_counter = &stats->n_missed;
 288		goto out;
 289	}
 290
 291	ovs_flow_stats_update(flow, key->tp.flags, skb);
 292	sf_acts = rcu_dereference(flow->sf_acts);
 293	error = ovs_execute_actions(dp, skb, sf_acts, key);
 294	if (unlikely(error))
 295		net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
 296				    ovs_dp_name(dp), error);
 297
 298	stats_counter = &stats->n_hit;
 299
 300out:
 301	/* Update datapath statistics. */
 302	u64_stats_update_begin(&stats->syncp);
 303	(*stats_counter)++;
 304	stats->n_mask_hit += n_mask_hit;
 305	stats->n_cache_hit += n_cache_hit;
 306	u64_stats_update_end(&stats->syncp);
 307}
 308
 309int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 310		  const struct sw_flow_key *key,
 311		  const struct dp_upcall_info *upcall_info,
 312		  uint32_t cutlen)
 313{
 314	struct dp_stats_percpu *stats;
 315	int err;
 316
 317	if (trace_ovs_dp_upcall_enabled())
 318		trace_ovs_dp_upcall(dp, skb, key, upcall_info);
 319
 320	if (upcall_info->portid == 0) {
 321		err = -ENOTCONN;
 322		goto err;
 323	}
 324
 325	if (!skb_is_gso(skb))
 326		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 327	else
 328		err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
 329
 330	ovs_vport_update_upcall_stats(skb, upcall_info, !err);
 331	if (err)
 332		goto err;
 333
 334	return 0;
 335
 336err:
 337	stats = this_cpu_ptr(dp->stats_percpu);
 338
 339	u64_stats_update_begin(&stats->syncp);
 340	stats->n_lost++;
 341	u64_stats_update_end(&stats->syncp);
 342
 343	return err;
 344}
 345
 346static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 347			     const struct sw_flow_key *key,
 348			     const struct dp_upcall_info *upcall_info,
 349			     uint32_t cutlen)
 350{
 351	unsigned int gso_type = skb_shinfo(skb)->gso_type;
 352	struct sw_flow_key later_key;
 353	struct sk_buff *segs, *nskb;
 354	int err;
 355
 356	BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_GSO_CB_OFFSET);
 357	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 358	if (IS_ERR(segs))
 359		return PTR_ERR(segs);
 360	if (segs == NULL)
 361		return -EINVAL;
 362
 363	if (gso_type & SKB_GSO_UDP) {
 364		/* The initial flow key extracted by ovs_flow_key_extract()
 365		 * in this case is for a first fragment, so we need to
 366		 * properly mark later fragments.
 367		 */
 368		later_key = *key;
 369		later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 370	}
 371
 372	/* Queue all of the segments. */
 373	skb_list_walk_safe(segs, skb, nskb) {
 
 374		if (gso_type & SKB_GSO_UDP && skb != segs)
 375			key = &later_key;
 376
 377		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 378		if (err)
 379			break;
 380
 381	}
 382
 383	/* Free all of the segments. */
 384	skb_list_walk_safe(segs, skb, nskb) {
 
 
 385		if (err)
 386			kfree_skb(skb);
 387		else
 388			consume_skb(skb);
 389	}
 390	return err;
 391}
 392
 393static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 394			      unsigned int hdrlen, int actions_attrlen)
 395{
 396	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 397		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 398		+ nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 399		+ nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
 400		+ nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
 401
 402	/* OVS_PACKET_ATTR_USERDATA */
 403	if (upcall_info->userdata)
 404		size += NLA_ALIGN(upcall_info->userdata->nla_len);
 405
 406	/* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 407	if (upcall_info->egress_tun_info)
 408		size += nla_total_size(ovs_tun_key_attr_size());
 409
 410	/* OVS_PACKET_ATTR_ACTIONS */
 411	if (upcall_info->actions_len)
 412		size += nla_total_size(actions_attrlen);
 413
 414	/* OVS_PACKET_ATTR_MRU */
 415	if (upcall_info->mru)
 416		size += nla_total_size(sizeof(upcall_info->mru));
 417
 418	return size;
 419}
 420
 421static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 422{
 423	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 424		size_t plen = NLA_ALIGN(skb->len) - skb->len;
 425
 426		if (plen > 0)
 427			skb_put_zero(skb, plen);
 428	}
 429}
 430
 431static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 432				  const struct sw_flow_key *key,
 433				  const struct dp_upcall_info *upcall_info,
 434				  uint32_t cutlen)
 435{
 436	struct ovs_header *upcall;
 437	struct sk_buff *nskb = NULL;
 438	struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 439	struct nlattr *nla;
 440	size_t len;
 441	unsigned int hlen;
 442	int err, dp_ifindex;
 443	u64 hash;
 444
 445	dp_ifindex = get_dpifindex(dp);
 446	if (!dp_ifindex)
 447		return -ENODEV;
 448
 449	if (skb_vlan_tag_present(skb)) {
 450		nskb = skb_clone(skb, GFP_ATOMIC);
 451		if (!nskb)
 452			return -ENOMEM;
 453
 454		nskb = __vlan_hwaccel_push_inside(nskb);
 455		if (!nskb)
 456			return -ENOMEM;
 457
 458		skb = nskb;
 459	}
 460
 461	if (nla_attr_size(skb->len) > USHRT_MAX) {
 462		err = -EFBIG;
 463		goto out;
 464	}
 465
 466	/* Complete checksum if needed */
 467	if (skb->ip_summed == CHECKSUM_PARTIAL &&
 468	    (err = skb_csum_hwoffload_help(skb, 0)))
 469		goto out;
 470
 471	/* Older versions of OVS user space enforce alignment of the last
 472	 * Netlink attribute to NLA_ALIGNTO which would require extensive
 473	 * padding logic. Only perform zerocopy if padding is not required.
 474	 */
 475	if (dp->user_features & OVS_DP_F_UNALIGNED)
 476		hlen = skb_zerocopy_headlen(skb);
 477	else
 478		hlen = skb->len;
 479
 480	len = upcall_msg_size(upcall_info, hlen - cutlen,
 481			      OVS_CB(skb)->acts_origlen);
 482	user_skb = genlmsg_new(len, GFP_ATOMIC);
 483	if (!user_skb) {
 484		err = -ENOMEM;
 485		goto out;
 486	}
 487
 488	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 489			     0, upcall_info->cmd);
 490	if (!upcall) {
 491		err = -EINVAL;
 492		goto out;
 493	}
 494	upcall->dp_ifindex = dp_ifindex;
 495
 496	err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 497	if (err)
 498		goto out;
 499
 500	if (upcall_info->userdata)
 501		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 502			  nla_len(upcall_info->userdata),
 503			  nla_data(upcall_info->userdata));
 504
 505	if (upcall_info->egress_tun_info) {
 506		nla = nla_nest_start_noflag(user_skb,
 507					    OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 508		if (!nla) {
 509			err = -EMSGSIZE;
 510			goto out;
 511		}
 512		err = ovs_nla_put_tunnel_info(user_skb,
 513					      upcall_info->egress_tun_info);
 514		if (err)
 515			goto out;
 516
 517		nla_nest_end(user_skb, nla);
 518	}
 519
 520	if (upcall_info->actions_len) {
 521		nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
 522		if (!nla) {
 523			err = -EMSGSIZE;
 524			goto out;
 525		}
 526		err = ovs_nla_put_actions(upcall_info->actions,
 527					  upcall_info->actions_len,
 528					  user_skb);
 529		if (!err)
 530			nla_nest_end(user_skb, nla);
 531		else
 532			nla_nest_cancel(user_skb, nla);
 533	}
 534
 535	/* Add OVS_PACKET_ATTR_MRU */
 536	if (upcall_info->mru &&
 537	    nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
 538		err = -ENOBUFS;
 539		goto out;
 
 
 
 540	}
 541
 542	/* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 543	if (cutlen > 0 &&
 544	    nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
 545		err = -ENOBUFS;
 546		goto out;
 547	}
 548
 549	/* Add OVS_PACKET_ATTR_HASH */
 550	hash = skb_get_hash_raw(skb);
 551	if (skb->sw_hash)
 552		hash |= OVS_PACKET_HASH_SW_BIT;
 553
 554	if (skb->l4_hash)
 555		hash |= OVS_PACKET_HASH_L4_BIT;
 556
 557	if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
 558		err = -ENOBUFS;
 559		goto out;
 560	}
 561
 562	/* Only reserve room for attribute header, packet data is added
 563	 * in skb_zerocopy() */
 564	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 565		err = -ENOBUFS;
 566		goto out;
 567	}
 568	nla->nla_len = nla_attr_size(skb->len - cutlen);
 569
 570	err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 571	if (err)
 572		goto out;
 573
 574	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 575	pad_packet(dp, user_skb);
 576
 577	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 578
 579	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 580	user_skb = NULL;
 581out:
 582	if (err)
 583		skb_tx_error(skb);
 584	consume_skb(user_skb);
 585	consume_skb(nskb);
 586
 587	return err;
 588}
 589
 590static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 591{
 592	struct ovs_header *ovs_header = genl_info_userhdr(info);
 593	struct net *net = sock_net(skb->sk);
 594	struct nlattr **a = info->attrs;
 595	struct sw_flow_actions *acts;
 596	struct sk_buff *packet;
 597	struct sw_flow *flow;
 598	struct sw_flow_actions *sf_acts;
 599	struct datapath *dp;
 600	struct vport *input_vport;
 601	u16 mru = 0;
 602	u64 hash;
 603	int len;
 604	int err;
 605	bool log = !a[OVS_PACKET_ATTR_PROBE];
 606
 607	err = -EINVAL;
 608	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 609	    !a[OVS_PACKET_ATTR_ACTIONS])
 610		goto err;
 611
 612	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 613	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 614	err = -ENOMEM;
 615	if (!packet)
 616		goto err;
 617	skb_reserve(packet, NET_IP_ALIGN);
 618
 619	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 620
 621	/* Set packet's mru */
 622	if (a[OVS_PACKET_ATTR_MRU]) {
 623		mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 624		packet->ignore_df = 1;
 625	}
 626	OVS_CB(packet)->mru = mru;
 627
 628	if (a[OVS_PACKET_ATTR_HASH]) {
 629		hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
 630
 631		__skb_set_hash(packet, hash & 0xFFFFFFFFULL,
 632			       !!(hash & OVS_PACKET_HASH_SW_BIT),
 633			       !!(hash & OVS_PACKET_HASH_L4_BIT));
 634	}
 635
 636	/* Build an sw_flow for sending this packet. */
 637	flow = ovs_flow_alloc();
 638	err = PTR_ERR(flow);
 639	if (IS_ERR(flow))
 640		goto err_kfree_skb;
 641
 642	err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 643					     packet, &flow->key, log);
 644	if (err)
 645		goto err_flow_free;
 646
 647	err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 648				   &flow->key, &acts, log);
 649	if (err)
 650		goto err_flow_free;
 651
 652	rcu_assign_pointer(flow->sf_acts, acts);
 653	packet->priority = flow->key.phy.priority;
 654	packet->mark = flow->key.phy.skb_mark;
 655
 656	rcu_read_lock();
 657	dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 658	err = -ENODEV;
 659	if (!dp)
 660		goto err_unlock;
 661
 662	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 663	if (!input_vport)
 664		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 665
 666	if (!input_vport)
 667		goto err_unlock;
 668
 669	packet->dev = input_vport->dev;
 670	OVS_CB(packet)->input_vport = input_vport;
 671	sf_acts = rcu_dereference(flow->sf_acts);
 672
 673	local_bh_disable();
 674	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 675	local_bh_enable();
 676	rcu_read_unlock();
 677
 678	ovs_flow_free(flow, false);
 679	return err;
 680
 681err_unlock:
 682	rcu_read_unlock();
 683err_flow_free:
 684	ovs_flow_free(flow, false);
 685err_kfree_skb:
 686	kfree_skb(packet);
 687err:
 688	return err;
 689}
 690
 691static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 692	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 693	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 694	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 695	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 696	[OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 697	[OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
 698};
 699
 700static const struct genl_small_ops dp_packet_genl_ops[] = {
 701	{ .cmd = OVS_PACKET_CMD_EXECUTE,
 702	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 703	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 704	  .doit = ovs_packet_cmd_execute
 705	}
 706};
 707
 708static struct genl_family dp_packet_genl_family __ro_after_init = {
 709	.hdrsize = sizeof(struct ovs_header),
 710	.name = OVS_PACKET_FAMILY,
 711	.version = OVS_PACKET_VERSION,
 712	.maxattr = OVS_PACKET_ATTR_MAX,
 713	.policy = packet_policy,
 714	.netnsok = true,
 715	.parallel_ops = true,
 716	.small_ops = dp_packet_genl_ops,
 717	.n_small_ops = ARRAY_SIZE(dp_packet_genl_ops),
 718	.resv_start_op = OVS_PACKET_CMD_EXECUTE + 1,
 719	.module = THIS_MODULE,
 720};
 721
 722static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 723			 struct ovs_dp_megaflow_stats *mega_stats)
 724{
 725	int i;
 726
 727	memset(mega_stats, 0, sizeof(*mega_stats));
 728
 729	stats->n_flows = ovs_flow_tbl_count(&dp->table);
 730	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 731
 732	stats->n_hit = stats->n_missed = stats->n_lost = 0;
 733
 734	for_each_possible_cpu(i) {
 735		const struct dp_stats_percpu *percpu_stats;
 736		struct dp_stats_percpu local_stats;
 737		unsigned int start;
 738
 739		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 740
 741		do {
 742			start = u64_stats_fetch_begin(&percpu_stats->syncp);
 743			local_stats = *percpu_stats;
 744		} while (u64_stats_fetch_retry(&percpu_stats->syncp, start));
 745
 746		stats->n_hit += local_stats.n_hit;
 747		stats->n_missed += local_stats.n_missed;
 748		stats->n_lost += local_stats.n_lost;
 749		mega_stats->n_mask_hit += local_stats.n_mask_hit;
 750		mega_stats->n_cache_hit += local_stats.n_cache_hit;
 751	}
 752}
 753
 754static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 755{
 756	return ovs_identifier_is_ufid(sfid) &&
 757	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 758}
 759
 760static bool should_fill_mask(uint32_t ufid_flags)
 761{
 762	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 763}
 764
 765static bool should_fill_actions(uint32_t ufid_flags)
 766{
 767	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 768}
 769
 770static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 771				    const struct sw_flow_id *sfid,
 772				    uint32_t ufid_flags)
 773{
 774	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 775
 776	/* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
 777	 * see ovs_nla_put_identifier()
 778	 */
 779	if (sfid && ovs_identifier_is_ufid(sfid))
 780		len += nla_total_size(sfid->ufid_len);
 781	else
 782		len += nla_total_size(ovs_key_attr_size());
 783
 784	/* OVS_FLOW_ATTR_KEY */
 785	if (!sfid || should_fill_key(sfid, ufid_flags))
 786		len += nla_total_size(ovs_key_attr_size());
 787
 788	/* OVS_FLOW_ATTR_MASK */
 789	if (should_fill_mask(ufid_flags))
 790		len += nla_total_size(ovs_key_attr_size());
 791
 792	/* OVS_FLOW_ATTR_ACTIONS */
 793	if (should_fill_actions(ufid_flags))
 794		len += nla_total_size(acts->orig_len);
 795
 796	return len
 797		+ nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 798		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 799		+ nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 800}
 801
 802/* Called with ovs_mutex or RCU read lock. */
 803static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 804				   struct sk_buff *skb)
 805{
 806	struct ovs_flow_stats stats;
 807	__be16 tcp_flags;
 808	unsigned long used;
 809
 810	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 811
 812	if (used &&
 813	    nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 814			      OVS_FLOW_ATTR_PAD))
 815		return -EMSGSIZE;
 816
 817	if (stats.n_packets &&
 818	    nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 819			  sizeof(struct ovs_flow_stats), &stats,
 820			  OVS_FLOW_ATTR_PAD))
 821		return -EMSGSIZE;
 822
 823	if ((u8)ntohs(tcp_flags) &&
 824	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 825		return -EMSGSIZE;
 826
 827	return 0;
 828}
 829
 830/* Called with ovs_mutex or RCU read lock. */
 831static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 832				     struct sk_buff *skb, int skb_orig_len)
 833{
 834	struct nlattr *start;
 835	int err;
 836
 837	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 838	 * this is the first flow to be dumped into 'skb'.  This is unusual for
 839	 * Netlink but individual action lists can be longer than
 840	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 841	 * The userspace caller can always fetch the actions separately if it
 842	 * really wants them.  (Most userspace callers in fact don't care.)
 843	 *
 844	 * This can only fail for dump operations because the skb is always
 845	 * properly sized for single flows.
 846	 */
 847	start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
 848	if (start) {
 849		const struct sw_flow_actions *sf_acts;
 850
 851		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 852		err = ovs_nla_put_actions(sf_acts->actions,
 853					  sf_acts->actions_len, skb);
 854
 855		if (!err)
 856			nla_nest_end(skb, start);
 857		else {
 858			if (skb_orig_len)
 859				return err;
 860
 861			nla_nest_cancel(skb, start);
 862		}
 863	} else if (skb_orig_len) {
 864		return -EMSGSIZE;
 865	}
 866
 867	return 0;
 868}
 869
 870/* Called with ovs_mutex or RCU read lock. */
 871static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 872				  struct sk_buff *skb, u32 portid,
 873				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 874{
 875	const int skb_orig_len = skb->len;
 876	struct ovs_header *ovs_header;
 877	int err;
 878
 879	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 880				 flags, cmd);
 881	if (!ovs_header)
 882		return -EMSGSIZE;
 883
 884	ovs_header->dp_ifindex = dp_ifindex;
 885
 886	err = ovs_nla_put_identifier(flow, skb);
 887	if (err)
 888		goto error;
 889
 890	if (should_fill_key(&flow->id, ufid_flags)) {
 891		err = ovs_nla_put_masked_key(flow, skb);
 892		if (err)
 893			goto error;
 894	}
 895
 896	if (should_fill_mask(ufid_flags)) {
 897		err = ovs_nla_put_mask(flow, skb);
 898		if (err)
 899			goto error;
 900	}
 901
 902	err = ovs_flow_cmd_fill_stats(flow, skb);
 903	if (err)
 904		goto error;
 905
 906	if (should_fill_actions(ufid_flags)) {
 907		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 908		if (err)
 909			goto error;
 910	}
 911
 912	genlmsg_end(skb, ovs_header);
 913	return 0;
 914
 915error:
 916	genlmsg_cancel(skb, ovs_header);
 917	return err;
 918}
 919
 920/* May not be called with RCU read lock. */
 921static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 922					       const struct sw_flow_id *sfid,
 923					       struct genl_info *info,
 924					       bool always,
 925					       uint32_t ufid_flags)
 926{
 927	struct sk_buff *skb;
 928	size_t len;
 929
 930	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 931		return NULL;
 932
 933	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 934	skb = genlmsg_new(len, GFP_KERNEL);
 935	if (!skb)
 936		return ERR_PTR(-ENOMEM);
 937
 938	return skb;
 939}
 940
 941/* Called with ovs_mutex. */
 942static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 943					       int dp_ifindex,
 944					       struct genl_info *info, u8 cmd,
 945					       bool always, u32 ufid_flags)
 946{
 947	struct sk_buff *skb;
 948	int retval;
 949
 950	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 951				      &flow->id, info, always, ufid_flags);
 952	if (IS_ERR_OR_NULL(skb))
 953		return skb;
 954
 955	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 956					info->snd_portid, info->snd_seq, 0,
 957					cmd, ufid_flags);
 958	if (WARN_ON_ONCE(retval < 0)) {
 959		kfree_skb(skb);
 960		skb = ERR_PTR(retval);
 961	}
 962	return skb;
 963}
 964
 965static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 966{
 967	struct net *net = sock_net(skb->sk);
 968	struct nlattr **a = info->attrs;
 969	struct ovs_header *ovs_header = genl_info_userhdr(info);
 970	struct sw_flow *flow = NULL, *new_flow;
 971	struct sw_flow_mask mask;
 972	struct sk_buff *reply;
 973	struct datapath *dp;
 974	struct sw_flow_key *key;
 975	struct sw_flow_actions *acts;
 976	struct sw_flow_match match;
 977	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 978	int error;
 979	bool log = !a[OVS_FLOW_ATTR_PROBE];
 980
 981	/* Must have key and actions. */
 982	error = -EINVAL;
 983	if (!a[OVS_FLOW_ATTR_KEY]) {
 984		OVS_NLERR(log, "Flow key attr not present in new flow.");
 985		goto error;
 986	}
 987	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 988		OVS_NLERR(log, "Flow actions attr not present in new flow.");
 989		goto error;
 990	}
 991
 992	/* Most of the time we need to allocate a new flow, do it before
 993	 * locking.
 994	 */
 995	new_flow = ovs_flow_alloc();
 996	if (IS_ERR(new_flow)) {
 997		error = PTR_ERR(new_flow);
 998		goto error;
 999	}
1000
1001	/* Extract key. */
1002	key = kzalloc(sizeof(*key), GFP_KERNEL);
1003	if (!key) {
1004		error = -ENOMEM;
1005		goto err_kfree_flow;
1006	}
1007
1008	ovs_match_init(&match, key, false, &mask);
1009	error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1010				  a[OVS_FLOW_ATTR_MASK], log);
1011	if (error)
1012		goto err_kfree_key;
1013
1014	ovs_flow_mask_key(&new_flow->key, key, true, &mask);
1015
1016	/* Extract flow identifier. */
1017	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
1018				       key, log);
1019	if (error)
1020		goto err_kfree_key;
 
 
 
 
 
 
1021
1022	/* Validate actions. */
1023	error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
1024				     &new_flow->key, &acts, log);
1025	if (error) {
1026		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
1027		goto err_kfree_key;
1028	}
1029
1030	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
1031					ufid_flags);
1032	if (IS_ERR(reply)) {
1033		error = PTR_ERR(reply);
1034		goto err_kfree_acts;
1035	}
1036
1037	ovs_lock();
1038	dp = get_dp(net, ovs_header->dp_ifindex);
1039	if (unlikely(!dp)) {
1040		error = -ENODEV;
1041		goto err_unlock_ovs;
1042	}
1043
1044	/* Check if this is a duplicate flow */
1045	if (ovs_identifier_is_ufid(&new_flow->id))
1046		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1047	if (!flow)
1048		flow = ovs_flow_tbl_lookup(&dp->table, key);
1049	if (likely(!flow)) {
1050		rcu_assign_pointer(new_flow->sf_acts, acts);
1051
1052		/* Put flow in bucket. */
1053		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1054		if (unlikely(error)) {
1055			acts = NULL;
1056			goto err_unlock_ovs;
1057		}
1058
1059		if (unlikely(reply)) {
1060			error = ovs_flow_cmd_fill_info(new_flow,
1061						       ovs_header->dp_ifindex,
1062						       reply, info->snd_portid,
1063						       info->snd_seq, 0,
1064						       OVS_FLOW_CMD_NEW,
1065						       ufid_flags);
1066			BUG_ON(error < 0);
1067		}
1068		ovs_unlock();
1069	} else {
1070		struct sw_flow_actions *old_acts;
1071
1072		/* Bail out if we're not allowed to modify an existing flow.
1073		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1074		 * because Generic Netlink treats the latter as a dump
1075		 * request.  We also accept NLM_F_EXCL in case that bug ever
1076		 * gets fixed.
1077		 */
1078		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1079							 | NLM_F_EXCL))) {
1080			error = -EEXIST;
1081			goto err_unlock_ovs;
1082		}
1083		/* The flow identifier has to be the same for flow updates.
1084		 * Look for any overlapping flow.
1085		 */
1086		if (unlikely(!ovs_flow_cmp(flow, &match))) {
1087			if (ovs_identifier_is_key(&flow->id))
1088				flow = ovs_flow_tbl_lookup_exact(&dp->table,
1089								 &match);
1090			else /* UFID matches but key is different */
1091				flow = NULL;
1092			if (!flow) {
1093				error = -ENOENT;
1094				goto err_unlock_ovs;
1095			}
1096		}
1097		/* Update actions. */
1098		old_acts = ovsl_dereference(flow->sf_acts);
1099		rcu_assign_pointer(flow->sf_acts, acts);
1100
1101		if (unlikely(reply)) {
1102			error = ovs_flow_cmd_fill_info(flow,
1103						       ovs_header->dp_ifindex,
1104						       reply, info->snd_portid,
1105						       info->snd_seq, 0,
1106						       OVS_FLOW_CMD_NEW,
1107						       ufid_flags);
1108			BUG_ON(error < 0);
1109		}
1110		ovs_unlock();
1111
1112		ovs_nla_free_flow_actions_rcu(old_acts);
1113		ovs_flow_free(new_flow, false);
1114	}
1115
1116	if (reply)
1117		ovs_notify(&dp_flow_genl_family, reply, info);
1118
1119	kfree(key);
1120	return 0;
1121
1122err_unlock_ovs:
1123	ovs_unlock();
1124	kfree_skb(reply);
1125err_kfree_acts:
1126	ovs_nla_free_flow_actions(acts);
1127err_kfree_key:
1128	kfree(key);
1129err_kfree_flow:
1130	ovs_flow_free(new_flow, false);
1131error:
1132	return error;
1133}
1134
1135/* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1136static noinline_for_stack
1137struct sw_flow_actions *get_flow_actions(struct net *net,
1138					 const struct nlattr *a,
1139					 const struct sw_flow_key *key,
1140					 const struct sw_flow_mask *mask,
1141					 bool log)
1142{
1143	struct sw_flow_actions *acts;
1144	struct sw_flow_key masked_key;
1145	int error;
1146
1147	ovs_flow_mask_key(&masked_key, key, true, mask);
1148	error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1149	if (error) {
1150		OVS_NLERR(log,
1151			  "Actions may not be safe on all matching packets");
1152		return ERR_PTR(error);
1153	}
1154
1155	return acts;
1156}
1157
1158/* Factor out match-init and action-copy to avoid
1159 * "Wframe-larger-than=1024" warning. Because mask is only
1160 * used to get actions, we new a function to save some
1161 * stack space.
1162 *
1163 * If there are not key and action attrs, we return 0
1164 * directly. In the case, the caller will also not use the
1165 * match as before. If there is action attr, we try to get
1166 * actions and save them to *acts. Before returning from
1167 * the function, we reset the match->mask pointer. Because
1168 * we should not to return match object with dangling reference
1169 * to mask.
1170 * */
1171static noinline_for_stack int
1172ovs_nla_init_match_and_action(struct net *net,
1173			      struct sw_flow_match *match,
1174			      struct sw_flow_key *key,
1175			      struct nlattr **a,
1176			      struct sw_flow_actions **acts,
1177			      bool log)
1178{
1179	struct sw_flow_mask mask;
1180	int error = 0;
1181
1182	if (a[OVS_FLOW_ATTR_KEY]) {
1183		ovs_match_init(match, key, true, &mask);
1184		error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1185					  a[OVS_FLOW_ATTR_MASK], log);
1186		if (error)
1187			goto error;
1188	}
1189
1190	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1191		if (!a[OVS_FLOW_ATTR_KEY]) {
1192			OVS_NLERR(log,
1193				  "Flow key attribute not present in set flow.");
1194			error = -EINVAL;
1195			goto error;
1196		}
1197
1198		*acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1199					 &mask, log);
1200		if (IS_ERR(*acts)) {
1201			error = PTR_ERR(*acts);
1202			goto error;
1203		}
1204	}
1205
1206	/* On success, error is 0. */
1207error:
1208	match->mask = NULL;
1209	return error;
1210}
1211
1212static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1213{
1214	struct net *net = sock_net(skb->sk);
1215	struct nlattr **a = info->attrs;
1216	struct ovs_header *ovs_header = genl_info_userhdr(info);
1217	struct sw_flow_key key;
1218	struct sw_flow *flow;
1219	struct sk_buff *reply = NULL;
1220	struct datapath *dp;
1221	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1222	struct sw_flow_match match;
1223	struct sw_flow_id sfid;
1224	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1225	int error = 0;
1226	bool log = !a[OVS_FLOW_ATTR_PROBE];
1227	bool ufid_present;
1228
1229	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1230	if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1231		OVS_NLERR(log,
1232			  "Flow set message rejected, Key attribute missing.");
1233		return -EINVAL;
1234	}
1235
1236	error = ovs_nla_init_match_and_action(net, &match, &key, a,
1237					      &acts, log);
1238	if (error)
1239		goto error;
1240
1241	if (acts) {
1242		/* Can allocate before locking if have acts. */
1243		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1244						ufid_flags);
1245		if (IS_ERR(reply)) {
1246			error = PTR_ERR(reply);
1247			goto err_kfree_acts;
1248		}
1249	}
1250
1251	ovs_lock();
1252	dp = get_dp(net, ovs_header->dp_ifindex);
1253	if (unlikely(!dp)) {
1254		error = -ENODEV;
1255		goto err_unlock_ovs;
1256	}
1257	/* Check that the flow exists. */
1258	if (ufid_present)
1259		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1260	else
1261		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1262	if (unlikely(!flow)) {
1263		error = -ENOENT;
1264		goto err_unlock_ovs;
1265	}
1266
1267	/* Update actions, if present. */
1268	if (likely(acts)) {
1269		old_acts = ovsl_dereference(flow->sf_acts);
1270		rcu_assign_pointer(flow->sf_acts, acts);
1271
1272		if (unlikely(reply)) {
1273			error = ovs_flow_cmd_fill_info(flow,
1274						       ovs_header->dp_ifindex,
1275						       reply, info->snd_portid,
1276						       info->snd_seq, 0,
1277						       OVS_FLOW_CMD_SET,
1278						       ufid_flags);
1279			BUG_ON(error < 0);
1280		}
1281	} else {
1282		/* Could not alloc without acts before locking. */
1283		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1284						info, OVS_FLOW_CMD_SET, false,
1285						ufid_flags);
1286
1287		if (IS_ERR(reply)) {
1288			error = PTR_ERR(reply);
1289			goto err_unlock_ovs;
1290		}
1291	}
1292
1293	/* Clear stats. */
1294	if (a[OVS_FLOW_ATTR_CLEAR])
1295		ovs_flow_stats_clear(flow);
1296	ovs_unlock();
1297
1298	if (reply)
1299		ovs_notify(&dp_flow_genl_family, reply, info);
1300	if (old_acts)
1301		ovs_nla_free_flow_actions_rcu(old_acts);
1302
1303	return 0;
1304
1305err_unlock_ovs:
1306	ovs_unlock();
1307	kfree_skb(reply);
1308err_kfree_acts:
1309	ovs_nla_free_flow_actions(acts);
1310error:
1311	return error;
1312}
1313
1314static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1315{
1316	struct nlattr **a = info->attrs;
1317	struct ovs_header *ovs_header = genl_info_userhdr(info);
1318	struct net *net = sock_net(skb->sk);
1319	struct sw_flow_key key;
1320	struct sk_buff *reply;
1321	struct sw_flow *flow;
1322	struct datapath *dp;
1323	struct sw_flow_match match;
1324	struct sw_flow_id ufid;
1325	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1326	int err = 0;
1327	bool log = !a[OVS_FLOW_ATTR_PROBE];
1328	bool ufid_present;
1329
1330	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1331	if (a[OVS_FLOW_ATTR_KEY]) {
1332		ovs_match_init(&match, &key, true, NULL);
1333		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1334					log);
1335	} else if (!ufid_present) {
1336		OVS_NLERR(log,
1337			  "Flow get message rejected, Key attribute missing.");
1338		err = -EINVAL;
1339	}
1340	if (err)
1341		return err;
1342
1343	ovs_lock();
1344	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1345	if (!dp) {
1346		err = -ENODEV;
1347		goto unlock;
1348	}
1349
1350	if (ufid_present)
1351		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1352	else
1353		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1354	if (!flow) {
1355		err = -ENOENT;
1356		goto unlock;
1357	}
1358
1359	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1360					OVS_FLOW_CMD_GET, true, ufid_flags);
1361	if (IS_ERR(reply)) {
1362		err = PTR_ERR(reply);
1363		goto unlock;
1364	}
1365
1366	ovs_unlock();
1367	return genlmsg_reply(reply, info);
1368unlock:
1369	ovs_unlock();
1370	return err;
1371}
1372
1373static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1374{
1375	struct nlattr **a = info->attrs;
1376	struct ovs_header *ovs_header = genl_info_userhdr(info);
1377	struct net *net = sock_net(skb->sk);
1378	struct sw_flow_key key;
1379	struct sk_buff *reply;
1380	struct sw_flow *flow = NULL;
1381	struct datapath *dp;
1382	struct sw_flow_match match;
1383	struct sw_flow_id ufid;
1384	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1385	int err;
1386	bool log = !a[OVS_FLOW_ATTR_PROBE];
1387	bool ufid_present;
1388
1389	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1390	if (a[OVS_FLOW_ATTR_KEY]) {
1391		ovs_match_init(&match, &key, true, NULL);
1392		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1393					NULL, log);
1394		if (unlikely(err))
1395			return err;
1396	}
1397
1398	ovs_lock();
1399	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1400	if (unlikely(!dp)) {
1401		err = -ENODEV;
1402		goto unlock;
1403	}
1404
1405	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1406		err = ovs_flow_tbl_flush(&dp->table);
1407		goto unlock;
1408	}
1409
1410	if (ufid_present)
1411		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1412	else
1413		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1414	if (unlikely(!flow)) {
1415		err = -ENOENT;
1416		goto unlock;
1417	}
1418
1419	ovs_flow_tbl_remove(&dp->table, flow);
1420	ovs_unlock();
1421
1422	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1423					&flow->id, info, false, ufid_flags);
1424	if (likely(reply)) {
1425		if (!IS_ERR(reply)) {
1426			rcu_read_lock();	/*To keep RCU checker happy. */
1427			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1428						     reply, info->snd_portid,
1429						     info->snd_seq, 0,
1430						     OVS_FLOW_CMD_DEL,
1431						     ufid_flags);
1432			rcu_read_unlock();
1433			if (WARN_ON_ONCE(err < 0)) {
1434				kfree_skb(reply);
1435				goto out_free;
1436			}
1437
1438			ovs_notify(&dp_flow_genl_family, reply, info);
1439		} else {
1440			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1441					PTR_ERR(reply));
1442		}
1443	}
1444
1445out_free:
1446	ovs_flow_free(flow, true);
1447	return 0;
1448unlock:
1449	ovs_unlock();
1450	return err;
1451}
1452
1453static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1454{
1455	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1456	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1457	struct table_instance *ti;
1458	struct datapath *dp;
1459	u32 ufid_flags;
1460	int err;
1461
1462	err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1463				       OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1464	if (err)
1465		return err;
1466	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1467
1468	rcu_read_lock();
1469	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1470	if (!dp) {
1471		rcu_read_unlock();
1472		return -ENODEV;
1473	}
1474
1475	ti = rcu_dereference(dp->table.ti);
1476	for (;;) {
1477		struct sw_flow *flow;
1478		u32 bucket, obj;
1479
1480		bucket = cb->args[0];
1481		obj = cb->args[1];
1482		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1483		if (!flow)
1484			break;
1485
1486		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1487					   NETLINK_CB(cb->skb).portid,
1488					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1489					   OVS_FLOW_CMD_GET, ufid_flags) < 0)
1490			break;
1491
1492		cb->args[0] = bucket;
1493		cb->args[1] = obj;
1494	}
1495	rcu_read_unlock();
1496	return skb->len;
1497}
1498
1499static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1500	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1501	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1502	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1503	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1504	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1505	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1506	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1507};
1508
1509static const struct genl_small_ops dp_flow_genl_ops[] = {
1510	{ .cmd = OVS_FLOW_CMD_NEW,
1511	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1512	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1513	  .doit = ovs_flow_cmd_new
1514	},
1515	{ .cmd = OVS_FLOW_CMD_DEL,
1516	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1517	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1518	  .doit = ovs_flow_cmd_del
1519	},
1520	{ .cmd = OVS_FLOW_CMD_GET,
1521	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1522	  .flags = 0,		    /* OK for unprivileged users. */
1523	  .doit = ovs_flow_cmd_get,
1524	  .dumpit = ovs_flow_cmd_dump
1525	},
1526	{ .cmd = OVS_FLOW_CMD_SET,
1527	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1528	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1529	  .doit = ovs_flow_cmd_set,
1530	},
1531};
1532
1533static struct genl_family dp_flow_genl_family __ro_after_init = {
1534	.hdrsize = sizeof(struct ovs_header),
1535	.name = OVS_FLOW_FAMILY,
1536	.version = OVS_FLOW_VERSION,
1537	.maxattr = OVS_FLOW_ATTR_MAX,
1538	.policy = flow_policy,
1539	.netnsok = true,
1540	.parallel_ops = true,
1541	.small_ops = dp_flow_genl_ops,
1542	.n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
1543	.resv_start_op = OVS_FLOW_CMD_SET + 1,
1544	.mcgrps = &ovs_dp_flow_multicast_group,
1545	.n_mcgrps = 1,
1546	.module = THIS_MODULE,
1547};
1548
1549static size_t ovs_dp_cmd_msg_size(void)
1550{
1551	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1552
1553	msgsize += nla_total_size(IFNAMSIZ);
1554	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1555	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1556	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1557	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1558	msgsize += nla_total_size(sizeof(u32) * nr_cpu_ids); /* OVS_DP_ATTR_PER_CPU_PIDS */
1559
1560	return msgsize;
1561}
1562
1563/* Called with ovs_mutex. */
1564static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1565				u32 portid, u32 seq, u32 flags, u8 cmd)
1566{
1567	struct ovs_header *ovs_header;
1568	struct ovs_dp_stats dp_stats;
1569	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1570	struct dp_nlsk_pids *pids = ovsl_dereference(dp->upcall_portids);
1571	int err, pids_len;
1572
1573	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1574				 flags, cmd);
1575	if (!ovs_header)
1576		goto error;
1577
1578	ovs_header->dp_ifindex = get_dpifindex(dp);
1579
1580	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1581	if (err)
1582		goto nla_put_failure;
1583
1584	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1585	if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1586			  &dp_stats, OVS_DP_ATTR_PAD))
1587		goto nla_put_failure;
1588
1589	if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1590			  sizeof(struct ovs_dp_megaflow_stats),
1591			  &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1592		goto nla_put_failure;
1593
1594	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1595		goto nla_put_failure;
1596
1597	if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1598			ovs_flow_tbl_masks_cache_size(&dp->table)))
1599		goto nla_put_failure;
1600
1601	if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU && pids) {
1602		pids_len = min(pids->n_pids, nr_cpu_ids) * sizeof(u32);
1603		if (nla_put(skb, OVS_DP_ATTR_PER_CPU_PIDS, pids_len, &pids->pids))
1604			goto nla_put_failure;
1605	}
1606
1607	genlmsg_end(skb, ovs_header);
1608	return 0;
1609
1610nla_put_failure:
1611	genlmsg_cancel(skb, ovs_header);
1612error:
1613	return -EMSGSIZE;
1614}
1615
1616static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1617{
1618	return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1619}
1620
1621/* Called with rcu_read_lock or ovs_mutex. */
1622static struct datapath *lookup_datapath(struct net *net,
1623					const struct ovs_header *ovs_header,
1624					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1625{
1626	struct datapath *dp;
1627
1628	if (!a[OVS_DP_ATTR_NAME])
1629		dp = get_dp(net, ovs_header->dp_ifindex);
1630	else {
1631		struct vport *vport;
1632
1633		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1634		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1635	}
1636	return dp ? dp : ERR_PTR(-ENODEV);
1637}
1638
1639static void ovs_dp_reset_user_features(struct sk_buff *skb,
1640				       struct genl_info *info)
1641{
1642	struct datapath *dp;
1643
1644	dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
1645			     info->attrs);
1646	if (IS_ERR(dp))
1647		return;
1648
1649	pr_warn("%s: Dropping previously announced user features\n",
1650		ovs_dp_name(dp));
1651	dp->user_features = 0;
1652}
1653
1654static int ovs_dp_set_upcall_portids(struct datapath *dp,
1655			      const struct nlattr *ids)
1656{
1657	struct dp_nlsk_pids *old, *dp_nlsk_pids;
1658
1659	if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
1660		return -EINVAL;
1661
1662	old = ovsl_dereference(dp->upcall_portids);
1663
1664	dp_nlsk_pids = kmalloc(sizeof(*dp_nlsk_pids) + nla_len(ids),
1665			       GFP_KERNEL);
1666	if (!dp_nlsk_pids)
1667		return -ENOMEM;
1668
1669	dp_nlsk_pids->n_pids = nla_len(ids) / sizeof(u32);
1670	nla_memcpy(dp_nlsk_pids->pids, ids, nla_len(ids));
1671
1672	rcu_assign_pointer(dp->upcall_portids, dp_nlsk_pids);
1673
1674	kfree_rcu(old, rcu);
1675
1676	return 0;
1677}
1678
1679u32 ovs_dp_get_upcall_portid(const struct datapath *dp, uint32_t cpu_id)
1680{
1681	struct dp_nlsk_pids *dp_nlsk_pids;
1682
1683	dp_nlsk_pids = rcu_dereference(dp->upcall_portids);
1684
1685	if (dp_nlsk_pids) {
1686		if (cpu_id < dp_nlsk_pids->n_pids) {
1687			return dp_nlsk_pids->pids[cpu_id];
1688		} else if (dp_nlsk_pids->n_pids > 0 &&
1689			   cpu_id >= dp_nlsk_pids->n_pids) {
1690			/* If the number of netlink PIDs is mismatched with
1691			 * the number of CPUs as seen by the kernel, log this
1692			 * and send the upcall to an arbitrary socket (0) in
1693			 * order to not drop packets
1694			 */
1695			pr_info_ratelimited("cpu_id mismatch with handler threads");
1696			return dp_nlsk_pids->pids[cpu_id %
1697						  dp_nlsk_pids->n_pids];
1698		} else {
1699			return 0;
1700		}
1701	} else {
1702		return 0;
1703	}
1704}
1705
1706static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1707{
1708	u32 user_features = 0, old_features = dp->user_features;
1709	int err;
1710
1711	if (a[OVS_DP_ATTR_USER_FEATURES]) {
1712		user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1713
1714		if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1715				      OVS_DP_F_UNALIGNED |
1716				      OVS_DP_F_TC_RECIRC_SHARING |
1717				      OVS_DP_F_DISPATCH_UPCALL_PER_CPU))
1718			return -EOPNOTSUPP;
1719
1720#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1721		if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1722			return -EOPNOTSUPP;
1723#endif
1724	}
1725
1726	if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1727		int err;
1728		u32 cache_size;
1729
1730		cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1731		err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1732		if (err)
1733			return err;
1734	}
1735
1736	dp->user_features = user_features;
1737
1738	if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU &&
1739	    a[OVS_DP_ATTR_PER_CPU_PIDS]) {
1740		/* Upcall Netlink Port IDs have been updated */
1741		err = ovs_dp_set_upcall_portids(dp,
1742						a[OVS_DP_ATTR_PER_CPU_PIDS]);
1743		if (err)
1744			return err;
1745	}
1746
1747	if ((dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1748	    !(old_features & OVS_DP_F_TC_RECIRC_SHARING))
1749		tc_skb_ext_tc_enable();
1750	else if (!(dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1751		 (old_features & OVS_DP_F_TC_RECIRC_SHARING))
1752		tc_skb_ext_tc_disable();
1753
1754	return 0;
1755}
1756
1757static int ovs_dp_stats_init(struct datapath *dp)
1758{
1759	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1760	if (!dp->stats_percpu)
1761		return -ENOMEM;
1762
1763	return 0;
1764}
1765
1766static int ovs_dp_vport_init(struct datapath *dp)
1767{
1768	int i;
1769
1770	dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1771				  sizeof(struct hlist_head),
1772				  GFP_KERNEL);
1773	if (!dp->ports)
1774		return -ENOMEM;
1775
1776	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1777		INIT_HLIST_HEAD(&dp->ports[i]);
1778
1779	return 0;
1780}
1781
1782static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1783{
1784	struct nlattr **a = info->attrs;
1785	struct vport_parms parms;
1786	struct sk_buff *reply;
1787	struct datapath *dp;
1788	struct vport *vport;
1789	struct ovs_net *ovs_net;
1790	int err;
1791
1792	err = -EINVAL;
1793	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1794		goto err;
1795
1796	reply = ovs_dp_cmd_alloc_info();
1797	if (!reply)
1798		return -ENOMEM;
1799
1800	err = -ENOMEM;
1801	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1802	if (dp == NULL)
1803		goto err_destroy_reply;
1804
1805	ovs_dp_set_net(dp, sock_net(skb->sk));
1806
1807	/* Allocate table. */
1808	err = ovs_flow_tbl_init(&dp->table);
1809	if (err)
1810		goto err_destroy_dp;
1811
1812	err = ovs_dp_stats_init(dp);
1813	if (err)
 
1814		goto err_destroy_table;
 
1815
1816	err = ovs_dp_vport_init(dp);
1817	if (err)
1818		goto err_destroy_stats;
 
 
 
 
 
 
 
1819
1820	err = ovs_meters_init(dp);
1821	if (err)
1822		goto err_destroy_ports;
1823
1824	/* Set up our datapath device. */
1825	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1826	parms.type = OVS_VPORT_TYPE_INTERNAL;
1827	parms.options = NULL;
1828	parms.dp = dp;
1829	parms.port_no = OVSP_LOCAL;
1830	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1831	parms.desired_ifindex = nla_get_s32_default(a[OVS_DP_ATTR_IFINDEX], 0);
 
 
 
1832
1833	/* So far only local changes have been made, now need the lock. */
1834	ovs_lock();
1835
1836	err = ovs_dp_change(dp, a);
1837	if (err)
1838		goto err_unlock_and_destroy_meters;
1839
1840	vport = new_vport(&parms);
1841	if (IS_ERR(vport)) {
1842		err = PTR_ERR(vport);
1843		if (err == -EBUSY)
1844			err = -EEXIST;
1845
1846		if (err == -EEXIST) {
1847			/* An outdated user space instance that does not understand
1848			 * the concept of user_features has attempted to create a new
1849			 * datapath and is likely to reuse it. Drop all user features.
1850			 */
1851			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1852				ovs_dp_reset_user_features(skb, info);
1853		}
1854
1855		goto err_destroy_portids;
1856	}
1857
1858	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1859				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1860	BUG_ON(err < 0);
1861
1862	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1863	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1864
1865	ovs_unlock();
1866
1867	ovs_notify(&dp_datapath_genl_family, reply, info);
1868	return 0;
1869
1870err_destroy_portids:
1871	kfree(rcu_dereference_raw(dp->upcall_portids));
1872err_unlock_and_destroy_meters:
1873	ovs_unlock();
1874	ovs_meters_exit(dp);
1875err_destroy_ports:
1876	kfree(dp->ports);
1877err_destroy_stats:
1878	free_percpu(dp->stats_percpu);
1879err_destroy_table:
1880	ovs_flow_tbl_destroy(&dp->table);
1881err_destroy_dp:
1882	kfree(dp);
1883err_destroy_reply:
1884	kfree_skb(reply);
1885err:
1886	return err;
1887}
1888
1889/* Called with ovs_mutex. */
1890static void __dp_destroy(struct datapath *dp)
1891{
1892	struct flow_table *table = &dp->table;
1893	int i;
1894
1895	if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1896		tc_skb_ext_tc_disable();
1897
1898	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1899		struct vport *vport;
1900		struct hlist_node *n;
1901
1902		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1903			if (vport->port_no != OVSP_LOCAL)
1904				ovs_dp_detach_port(vport);
1905	}
1906
1907	list_del_rcu(&dp->list_node);
1908
1909	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1910	 * all ports in datapath are destroyed first before freeing datapath.
1911	 */
1912	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1913
1914	/* Flush sw_flow in the tables. RCU cb only releases resource
1915	 * such as dp, ports and tables. That may avoid some issues
1916	 * such as RCU usage warning.
1917	 */
1918	table_instance_flow_flush(table, ovsl_dereference(table->ti),
1919				  ovsl_dereference(table->ufid_ti));
1920
1921	/* RCU destroy the ports, meters and flow tables. */
1922	call_rcu(&dp->rcu, destroy_dp_rcu);
1923}
1924
1925static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1926{
1927	struct sk_buff *reply;
1928	struct datapath *dp;
1929	int err;
1930
1931	reply = ovs_dp_cmd_alloc_info();
1932	if (!reply)
1933		return -ENOMEM;
1934
1935	ovs_lock();
1936	dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
1937			     info->attrs);
1938	err = PTR_ERR(dp);
1939	if (IS_ERR(dp))
1940		goto err_unlock_free;
1941
1942	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1943				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1944	BUG_ON(err < 0);
1945
1946	__dp_destroy(dp);
1947	ovs_unlock();
1948
1949	ovs_notify(&dp_datapath_genl_family, reply, info);
1950
1951	return 0;
1952
1953err_unlock_free:
1954	ovs_unlock();
1955	kfree_skb(reply);
1956	return err;
1957}
1958
1959static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1960{
1961	struct sk_buff *reply;
1962	struct datapath *dp;
1963	int err;
1964
1965	reply = ovs_dp_cmd_alloc_info();
1966	if (!reply)
1967		return -ENOMEM;
1968
1969	ovs_lock();
1970	dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
1971			     info->attrs);
1972	err = PTR_ERR(dp);
1973	if (IS_ERR(dp))
1974		goto err_unlock_free;
1975
1976	err = ovs_dp_change(dp, info->attrs);
1977	if (err)
1978		goto err_unlock_free;
1979
1980	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1981				   info->snd_seq, 0, OVS_DP_CMD_SET);
1982	BUG_ON(err < 0);
1983
1984	ovs_unlock();
1985	ovs_notify(&dp_datapath_genl_family, reply, info);
1986
1987	return 0;
1988
1989err_unlock_free:
1990	ovs_unlock();
1991	kfree_skb(reply);
1992	return err;
1993}
1994
1995static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1996{
1997	struct sk_buff *reply;
1998	struct datapath *dp;
1999	int err;
2000
2001	reply = ovs_dp_cmd_alloc_info();
2002	if (!reply)
2003		return -ENOMEM;
2004
2005	ovs_lock();
2006	dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
2007			     info->attrs);
2008	if (IS_ERR(dp)) {
2009		err = PTR_ERR(dp);
2010		goto err_unlock_free;
2011	}
2012	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
2013				   info->snd_seq, 0, OVS_DP_CMD_GET);
2014	BUG_ON(err < 0);
2015	ovs_unlock();
2016
2017	return genlmsg_reply(reply, info);
2018
2019err_unlock_free:
2020	ovs_unlock();
2021	kfree_skb(reply);
2022	return err;
2023}
2024
2025static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2026{
2027	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2028	struct datapath *dp;
2029	int skip = cb->args[0];
2030	int i = 0;
2031
2032	ovs_lock();
2033	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2034		if (i >= skip &&
2035		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
2036					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2037					 OVS_DP_CMD_GET) < 0)
2038			break;
2039		i++;
2040	}
2041	ovs_unlock();
2042
2043	cb->args[0] = i;
2044
2045	return skb->len;
2046}
2047
2048static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
2049	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2050	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2051	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
2052	[OVS_DP_ATTR_MASKS_CACHE_SIZE] =  NLA_POLICY_RANGE(NLA_U32, 0,
2053		PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
2054	[OVS_DP_ATTR_IFINDEX] = NLA_POLICY_MIN(NLA_S32, 0),
2055};
2056
2057static const struct genl_small_ops dp_datapath_genl_ops[] = {
2058	{ .cmd = OVS_DP_CMD_NEW,
2059	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2060	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2061	  .doit = ovs_dp_cmd_new
2062	},
2063	{ .cmd = OVS_DP_CMD_DEL,
2064	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2065	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2066	  .doit = ovs_dp_cmd_del
2067	},
2068	{ .cmd = OVS_DP_CMD_GET,
2069	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2070	  .flags = 0,		    /* OK for unprivileged users. */
2071	  .doit = ovs_dp_cmd_get,
2072	  .dumpit = ovs_dp_cmd_dump
2073	},
2074	{ .cmd = OVS_DP_CMD_SET,
2075	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2076	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2077	  .doit = ovs_dp_cmd_set,
2078	},
2079};
2080
2081static struct genl_family dp_datapath_genl_family __ro_after_init = {
2082	.hdrsize = sizeof(struct ovs_header),
2083	.name = OVS_DATAPATH_FAMILY,
2084	.version = OVS_DATAPATH_VERSION,
2085	.maxattr = OVS_DP_ATTR_MAX,
2086	.policy = datapath_policy,
2087	.netnsok = true,
2088	.parallel_ops = true,
2089	.small_ops = dp_datapath_genl_ops,
2090	.n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
2091	.resv_start_op = OVS_DP_CMD_SET + 1,
2092	.mcgrps = &ovs_dp_datapath_multicast_group,
2093	.n_mcgrps = 1,
2094	.module = THIS_MODULE,
2095};
2096
2097/* Called with ovs_mutex or RCU read lock. */
2098static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
2099				   struct net *net, u32 portid, u32 seq,
2100				   u32 flags, u8 cmd, gfp_t gfp)
2101{
2102	struct ovs_header *ovs_header;
2103	struct ovs_vport_stats vport_stats;
2104	struct net *net_vport;
2105	int err;
2106
2107	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
2108				 flags, cmd);
2109	if (!ovs_header)
2110		return -EMSGSIZE;
2111
2112	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
2113
2114	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
2115	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
2116	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
2117			   ovs_vport_name(vport)) ||
2118	    nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
2119		goto nla_put_failure;
2120
2121	rcu_read_lock();
2122	net_vport = dev_net_rcu(vport->dev);
2123	if (!net_eq(net, net_vport)) {
2124		int id = peernet2id_alloc(net, net_vport, GFP_ATOMIC);
2125
2126		if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
2127			goto nla_put_failure_unlock;
2128	}
2129	rcu_read_unlock();
2130
2131	ovs_vport_get_stats(vport, &vport_stats);
2132	if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
2133			  sizeof(struct ovs_vport_stats), &vport_stats,
2134			  OVS_VPORT_ATTR_PAD))
2135		goto nla_put_failure;
2136
2137	if (ovs_vport_get_upcall_stats(vport, skb))
2138		goto nla_put_failure;
2139
2140	if (ovs_vport_get_upcall_portids(vport, skb))
2141		goto nla_put_failure;
2142
2143	err = ovs_vport_get_options(vport, skb);
2144	if (err == -EMSGSIZE)
2145		goto error;
2146
2147	genlmsg_end(skb, ovs_header);
2148	return 0;
2149
2150nla_put_failure_unlock:
2151	rcu_read_unlock();
2152nla_put_failure:
2153	err = -EMSGSIZE;
2154error:
2155	genlmsg_cancel(skb, ovs_header);
2156	return err;
2157}
2158
2159static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2160{
2161	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2162}
2163
2164/* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2165struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2166					 u32 portid, u32 seq, u8 cmd)
2167{
2168	struct sk_buff *skb;
2169	int retval;
2170
2171	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2172	if (!skb)
2173		return ERR_PTR(-ENOMEM);
2174
2175	retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2176					 GFP_KERNEL);
2177	BUG_ON(retval < 0);
2178
2179	return skb;
2180}
2181
2182/* Called with ovs_mutex or RCU read lock. */
2183static struct vport *lookup_vport(struct net *net,
2184				  const struct ovs_header *ovs_header,
2185				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2186{
2187	struct datapath *dp;
2188	struct vport *vport;
2189
2190	if (a[OVS_VPORT_ATTR_IFINDEX])
2191		return ERR_PTR(-EOPNOTSUPP);
2192	if (a[OVS_VPORT_ATTR_NAME]) {
2193		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2194		if (!vport)
2195			return ERR_PTR(-ENODEV);
2196		if (ovs_header->dp_ifindex &&
2197		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2198			return ERR_PTR(-ENODEV);
2199		return vport;
2200	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2201		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2202
2203		if (port_no >= DP_MAX_PORTS)
2204			return ERR_PTR(-EFBIG);
2205
2206		dp = get_dp(net, ovs_header->dp_ifindex);
2207		if (!dp)
2208			return ERR_PTR(-ENODEV);
2209
2210		vport = ovs_vport_ovsl_rcu(dp, port_no);
2211		if (!vport)
2212			return ERR_PTR(-ENODEV);
2213		return vport;
2214	} else
2215		return ERR_PTR(-EINVAL);
2216
2217}
2218
2219static unsigned int ovs_get_max_headroom(struct datapath *dp)
2220{
2221	unsigned int dev_headroom, max_headroom = 0;
2222	struct net_device *dev;
2223	struct vport *vport;
2224	int i;
2225
2226	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2227		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2228					 lockdep_ovsl_is_held()) {
2229			dev = vport->dev;
2230			dev_headroom = netdev_get_fwd_headroom(dev);
2231			if (dev_headroom > max_headroom)
2232				max_headroom = dev_headroom;
2233		}
2234	}
2235
2236	return max_headroom;
2237}
2238
2239/* Called with ovs_mutex */
2240static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2241{
2242	struct vport *vport;
2243	int i;
2244
2245	dp->max_headroom = new_headroom;
2246	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2247		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2248					 lockdep_ovsl_is_held())
2249			netdev_set_rx_headroom(vport->dev, new_headroom);
2250	}
2251}
2252
2253static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2254{
2255	struct nlattr **a = info->attrs;
2256	struct ovs_header *ovs_header = genl_info_userhdr(info);
2257	struct vport_parms parms;
2258	struct sk_buff *reply;
2259	struct vport *vport;
2260	struct datapath *dp;
2261	unsigned int new_headroom;
2262	u32 port_no;
2263	int err;
2264
2265	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2266	    !a[OVS_VPORT_ATTR_UPCALL_PID])
2267		return -EINVAL;
2268
2269	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2270
2271	if (a[OVS_VPORT_ATTR_IFINDEX] && parms.type != OVS_VPORT_TYPE_INTERNAL)
2272		return -EOPNOTSUPP;
2273
2274	port_no = nla_get_u32_default(a[OVS_VPORT_ATTR_PORT_NO], 0);
 
2275	if (port_no >= DP_MAX_PORTS)
2276		return -EFBIG;
2277
2278	reply = ovs_vport_cmd_alloc_info();
2279	if (!reply)
2280		return -ENOMEM;
2281
2282	ovs_lock();
2283restart:
2284	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2285	err = -ENODEV;
2286	if (!dp)
2287		goto exit_unlock_free;
2288
2289	if (port_no) {
2290		vport = ovs_vport_ovsl(dp, port_no);
2291		err = -EBUSY;
2292		if (vport)
2293			goto exit_unlock_free;
2294	} else {
2295		for (port_no = 1; ; port_no++) {
2296			if (port_no >= DP_MAX_PORTS) {
2297				err = -EFBIG;
2298				goto exit_unlock_free;
2299			}
2300			vport = ovs_vport_ovsl(dp, port_no);
2301			if (!vport)
2302				break;
2303		}
2304	}
2305
2306	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
 
2307	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2308	parms.dp = dp;
2309	parms.port_no = port_no;
2310	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2311	parms.desired_ifindex = nla_get_s32_default(a[OVS_VPORT_ATTR_IFINDEX],
2312						    0);
2313
2314	vport = new_vport(&parms);
2315	err = PTR_ERR(vport);
2316	if (IS_ERR(vport)) {
2317		if (err == -EAGAIN)
2318			goto restart;
2319		goto exit_unlock_free;
2320	}
2321
2322	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2323				      info->snd_portid, info->snd_seq, 0,
2324				      OVS_VPORT_CMD_NEW, GFP_KERNEL);
2325
2326	new_headroom = netdev_get_fwd_headroom(vport->dev);
2327
2328	if (new_headroom > dp->max_headroom)
2329		ovs_update_headroom(dp, new_headroom);
2330	else
2331		netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2332
2333	BUG_ON(err < 0);
2334	ovs_unlock();
2335
2336	ovs_notify(&dp_vport_genl_family, reply, info);
2337	return 0;
2338
2339exit_unlock_free:
2340	ovs_unlock();
2341	kfree_skb(reply);
2342	return err;
2343}
2344
2345static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2346{
2347	struct nlattr **a = info->attrs;
2348	struct sk_buff *reply;
2349	struct vport *vport;
2350	int err;
2351
2352	reply = ovs_vport_cmd_alloc_info();
2353	if (!reply)
2354		return -ENOMEM;
2355
2356	ovs_lock();
2357	vport = lookup_vport(sock_net(skb->sk), genl_info_userhdr(info), a);
2358	err = PTR_ERR(vport);
2359	if (IS_ERR(vport))
2360		goto exit_unlock_free;
2361
2362	if (a[OVS_VPORT_ATTR_TYPE] &&
2363	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2364		err = -EINVAL;
2365		goto exit_unlock_free;
2366	}
2367
2368	if (a[OVS_VPORT_ATTR_OPTIONS]) {
2369		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2370		if (err)
2371			goto exit_unlock_free;
2372	}
2373
2374
2375	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2376		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2377
2378		err = ovs_vport_set_upcall_portids(vport, ids);
2379		if (err)
2380			goto exit_unlock_free;
2381	}
2382
2383	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2384				      info->snd_portid, info->snd_seq, 0,
2385				      OVS_VPORT_CMD_SET, GFP_KERNEL);
2386	BUG_ON(err < 0);
2387
2388	ovs_unlock();
2389	ovs_notify(&dp_vport_genl_family, reply, info);
2390	return 0;
2391
2392exit_unlock_free:
2393	ovs_unlock();
2394	kfree_skb(reply);
2395	return err;
2396}
2397
2398static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2399{
2400	bool update_headroom = false;
2401	struct nlattr **a = info->attrs;
2402	struct sk_buff *reply;
2403	struct datapath *dp;
2404	struct vport *vport;
2405	unsigned int new_headroom;
2406	int err;
2407
2408	reply = ovs_vport_cmd_alloc_info();
2409	if (!reply)
2410		return -ENOMEM;
2411
2412	ovs_lock();
2413	vport = lookup_vport(sock_net(skb->sk), genl_info_userhdr(info), a);
2414	err = PTR_ERR(vport);
2415	if (IS_ERR(vport))
2416		goto exit_unlock_free;
2417
2418	if (vport->port_no == OVSP_LOCAL) {
2419		err = -EINVAL;
2420		goto exit_unlock_free;
2421	}
2422
2423	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2424				      info->snd_portid, info->snd_seq, 0,
2425				      OVS_VPORT_CMD_DEL, GFP_KERNEL);
2426	BUG_ON(err < 0);
2427
2428	/* the vport deletion may trigger dp headroom update */
2429	dp = vport->dp;
2430	if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2431		update_headroom = true;
2432
2433	netdev_reset_rx_headroom(vport->dev);
2434	ovs_dp_detach_port(vport);
2435
2436	if (update_headroom) {
2437		new_headroom = ovs_get_max_headroom(dp);
2438
2439		if (new_headroom < dp->max_headroom)
2440			ovs_update_headroom(dp, new_headroom);
2441	}
2442	ovs_unlock();
2443
2444	ovs_notify(&dp_vport_genl_family, reply, info);
2445	return 0;
2446
2447exit_unlock_free:
2448	ovs_unlock();
2449	kfree_skb(reply);
2450	return err;
2451}
2452
2453static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2454{
2455	struct nlattr **a = info->attrs;
2456	struct ovs_header *ovs_header = genl_info_userhdr(info);
2457	struct sk_buff *reply;
2458	struct vport *vport;
2459	int err;
2460
2461	reply = ovs_vport_cmd_alloc_info();
2462	if (!reply)
2463		return -ENOMEM;
2464
2465	rcu_read_lock();
2466	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2467	err = PTR_ERR(vport);
2468	if (IS_ERR(vport))
2469		goto exit_unlock_free;
2470	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2471				      info->snd_portid, info->snd_seq, 0,
2472				      OVS_VPORT_CMD_GET, GFP_ATOMIC);
2473	BUG_ON(err < 0);
2474	rcu_read_unlock();
2475
2476	return genlmsg_reply(reply, info);
2477
2478exit_unlock_free:
2479	rcu_read_unlock();
2480	kfree_skb(reply);
2481	return err;
2482}
2483
2484static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2485{
2486	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2487	struct datapath *dp;
2488	int bucket = cb->args[0], skip = cb->args[1];
2489	int i, j = 0;
2490
2491	rcu_read_lock();
2492	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2493	if (!dp) {
2494		rcu_read_unlock();
2495		return -ENODEV;
2496	}
2497	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2498		struct vport *vport;
2499
2500		j = 0;
2501		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2502			if (j >= skip &&
2503			    ovs_vport_cmd_fill_info(vport, skb,
2504						    sock_net(skb->sk),
2505						    NETLINK_CB(cb->skb).portid,
2506						    cb->nlh->nlmsg_seq,
2507						    NLM_F_MULTI,
2508						    OVS_VPORT_CMD_GET,
2509						    GFP_ATOMIC) < 0)
2510				goto out;
2511
2512			j++;
2513		}
2514		skip = 0;
2515	}
2516out:
2517	rcu_read_unlock();
2518
2519	cb->args[0] = i;
2520	cb->args[1] = j;
2521
2522	return skb->len;
2523}
2524
2525static void ovs_dp_masks_rebalance(struct work_struct *work)
2526{
2527	struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2528					       masks_rebalance.work);
2529	struct datapath *dp;
2530
2531	ovs_lock();
2532
2533	list_for_each_entry(dp, &ovs_net->dps, list_node)
2534		ovs_flow_masks_rebalance(&dp->table);
2535
2536	ovs_unlock();
2537
2538	schedule_delayed_work(&ovs_net->masks_rebalance,
2539			      msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2540}
2541
2542static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2543	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2544	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2545	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2546	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2547	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2548	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2549	[OVS_VPORT_ATTR_IFINDEX] = NLA_POLICY_MIN(NLA_S32, 0),
2550	[OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2551	[OVS_VPORT_ATTR_UPCALL_STATS] = { .type = NLA_NESTED },
2552};
2553
2554static const struct genl_small_ops dp_vport_genl_ops[] = {
2555	{ .cmd = OVS_VPORT_CMD_NEW,
2556	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2557	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2558	  .doit = ovs_vport_cmd_new
2559	},
2560	{ .cmd = OVS_VPORT_CMD_DEL,
2561	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2562	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2563	  .doit = ovs_vport_cmd_del
2564	},
2565	{ .cmd = OVS_VPORT_CMD_GET,
2566	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2567	  .flags = 0,		    /* OK for unprivileged users. */
2568	  .doit = ovs_vport_cmd_get,
2569	  .dumpit = ovs_vport_cmd_dump
2570	},
2571	{ .cmd = OVS_VPORT_CMD_SET,
2572	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2573	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2574	  .doit = ovs_vport_cmd_set,
2575	},
2576};
2577
2578struct genl_family dp_vport_genl_family __ro_after_init = {
2579	.hdrsize = sizeof(struct ovs_header),
2580	.name = OVS_VPORT_FAMILY,
2581	.version = OVS_VPORT_VERSION,
2582	.maxattr = OVS_VPORT_ATTR_MAX,
2583	.policy = vport_policy,
2584	.netnsok = true,
2585	.parallel_ops = true,
2586	.small_ops = dp_vport_genl_ops,
2587	.n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
2588	.resv_start_op = OVS_VPORT_CMD_SET + 1,
2589	.mcgrps = &ovs_dp_vport_multicast_group,
2590	.n_mcgrps = 1,
2591	.module = THIS_MODULE,
2592};
2593
2594static struct genl_family * const dp_genl_families[] = {
2595	&dp_datapath_genl_family,
2596	&dp_vport_genl_family,
2597	&dp_flow_genl_family,
2598	&dp_packet_genl_family,
2599	&dp_meter_genl_family,
2600#if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2601	&dp_ct_limit_genl_family,
2602#endif
2603};
2604
2605static void dp_unregister_genl(int n_families)
2606{
2607	int i;
2608
2609	for (i = 0; i < n_families; i++)
2610		genl_unregister_family(dp_genl_families[i]);
2611}
2612
2613static int __init dp_register_genl(void)
2614{
2615	int err;
2616	int i;
2617
2618	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2619
2620		err = genl_register_family(dp_genl_families[i]);
2621		if (err)
2622			goto error;
2623	}
2624
2625	return 0;
2626
2627error:
2628	dp_unregister_genl(i);
2629	return err;
2630}
2631
2632static int __net_init ovs_init_net(struct net *net)
2633{
2634	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2635	int err;
2636
2637	INIT_LIST_HEAD(&ovs_net->dps);
2638	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2639	INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2640
2641	err = ovs_ct_init(net);
2642	if (err)
2643		return err;
2644
2645	schedule_delayed_work(&ovs_net->masks_rebalance,
2646			      msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2647	return 0;
2648}
2649
2650static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2651					    struct list_head *head)
2652{
2653	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2654	struct datapath *dp;
2655
2656	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2657		int i;
2658
2659		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2660			struct vport *vport;
2661
2662			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2663				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2664					continue;
2665
2666				if (dev_net(vport->dev) == dnet)
2667					list_add(&vport->detach_list, head);
2668			}
2669		}
2670	}
2671}
2672
2673static void __net_exit ovs_exit_net(struct net *dnet)
2674{
2675	struct datapath *dp, *dp_next;
2676	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2677	struct vport *vport, *vport_next;
2678	struct net *net;
2679	LIST_HEAD(head);
2680
 
2681	ovs_lock();
2682
2683	ovs_ct_exit(dnet);
2684
2685	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2686		__dp_destroy(dp);
2687
2688	down_read(&net_rwsem);
2689	for_each_net(net)
2690		list_vports_from_net(net, dnet, &head);
2691	up_read(&net_rwsem);
2692
2693	/* Detach all vports from given namespace. */
2694	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2695		list_del(&vport->detach_list);
2696		ovs_dp_detach_port(vport);
2697	}
2698
2699	ovs_unlock();
2700
2701	cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2702	cancel_work_sync(&ovs_net->dp_notify_work);
2703}
2704
2705static struct pernet_operations ovs_net_ops = {
2706	.init = ovs_init_net,
2707	.exit = ovs_exit_net,
2708	.id   = &ovs_net_id,
2709	.size = sizeof(struct ovs_net),
2710};
2711
2712static const char * const ovs_drop_reasons[] = {
2713#define S(x) [(x) & ~SKB_DROP_REASON_SUBSYS_MASK] = (#x),
2714	OVS_DROP_REASONS(S)
2715#undef S
2716};
2717
2718static struct drop_reason_list drop_reason_list_ovs = {
2719	.reasons = ovs_drop_reasons,
2720	.n_reasons = ARRAY_SIZE(ovs_drop_reasons),
2721};
2722
2723static int __init dp_init(void)
2724{
2725	int err;
2726
2727	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2728		     sizeof_field(struct sk_buff, cb));
2729
2730	pr_info("Open vSwitch switching datapath\n");
2731
2732	err = action_fifos_init();
2733	if (err)
2734		goto error;
2735
2736	err = ovs_internal_dev_rtnl_link_register();
2737	if (err)
2738		goto error_action_fifos_exit;
2739
2740	err = ovs_flow_init();
2741	if (err)
2742		goto error_unreg_rtnl_link;
2743
2744	err = ovs_vport_init();
2745	if (err)
2746		goto error_flow_exit;
2747
2748	err = register_pernet_device(&ovs_net_ops);
2749	if (err)
2750		goto error_vport_exit;
2751
2752	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2753	if (err)
2754		goto error_netns_exit;
2755
2756	err = ovs_netdev_init();
2757	if (err)
2758		goto error_unreg_notifier;
2759
2760	err = dp_register_genl();
2761	if (err < 0)
2762		goto error_unreg_netdev;
2763
2764	drop_reasons_register_subsys(SKB_DROP_REASON_SUBSYS_OPENVSWITCH,
2765				     &drop_reason_list_ovs);
2766
2767	return 0;
2768
2769error_unreg_netdev:
2770	ovs_netdev_exit();
2771error_unreg_notifier:
2772	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2773error_netns_exit:
2774	unregister_pernet_device(&ovs_net_ops);
2775error_vport_exit:
2776	ovs_vport_exit();
2777error_flow_exit:
2778	ovs_flow_exit();
2779error_unreg_rtnl_link:
2780	ovs_internal_dev_rtnl_link_unregister();
2781error_action_fifos_exit:
2782	action_fifos_exit();
2783error:
2784	return err;
2785}
2786
2787static void dp_cleanup(void)
2788{
2789	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2790	ovs_netdev_exit();
2791	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2792	unregister_pernet_device(&ovs_net_ops);
2793	drop_reasons_unregister_subsys(SKB_DROP_REASON_SUBSYS_OPENVSWITCH);
2794	rcu_barrier();
2795	ovs_vport_exit();
2796	ovs_flow_exit();
2797	ovs_internal_dev_rtnl_link_unregister();
2798	action_fifos_exit();
2799}
2800
2801module_init(dp_init);
2802module_exit(dp_cleanup);
2803
2804MODULE_DESCRIPTION("Open vSwitch switching datapath");
2805MODULE_LICENSE("GPL");
2806MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2807MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2808MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2809MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2810MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2811MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);