Linux Audio

Check our new training course

Loading...
v5.14.15
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (c) 2007-2017 Nicira, Inc.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
   4 */
   5
   6#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   7
   8#include <linux/skbuff.h>
   9#include <linux/in.h>
  10#include <linux/ip.h>
  11#include <linux/openvswitch.h>
 
  12#include <linux/sctp.h>
  13#include <linux/tcp.h>
  14#include <linux/udp.h>
  15#include <linux/in6.h>
  16#include <linux/if_arp.h>
  17#include <linux/if_vlan.h>
  18
  19#include <net/dst.h>
  20#include <net/ip.h>
  21#include <net/ipv6.h>
  22#include <net/ip6_fib.h>
  23#include <net/checksum.h>
  24#include <net/dsfield.h>
  25#include <net/mpls.h>
  26#include <net/sctp/checksum.h>
  27
  28#include "datapath.h"
  29#include "flow.h"
  30#include "conntrack.h"
  31#include "vport.h"
  32#include "flow_netlink.h"
  33#include "openvswitch_trace.h"
  34
  35struct deferred_action {
  36	struct sk_buff *skb;
  37	const struct nlattr *actions;
  38	int actions_len;
  39
  40	/* Store pkt_key clone when creating deferred action. */
  41	struct sw_flow_key pkt_key;
  42};
  43
  44#define MAX_L2_LEN	(VLAN_ETH_HLEN + 3 * MPLS_HLEN)
  45struct ovs_frag_data {
  46	unsigned long dst;
  47	struct vport *vport;
  48	struct ovs_skb_cb cb;
  49	__be16 inner_protocol;
  50	u16 network_offset;	/* valid only for MPLS */
  51	u16 vlan_tci;
  52	__be16 vlan_proto;
  53	unsigned int l2_len;
  54	u8 mac_proto;
  55	u8 l2_data[MAX_L2_LEN];
  56};
  57
  58static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
  59
  60#define DEFERRED_ACTION_FIFO_SIZE 10
  61#define OVS_RECURSION_LIMIT 5
  62#define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
  63struct action_fifo {
  64	int head;
  65	int tail;
  66	/* Deferred action fifo queue storage. */
  67	struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
  68};
  69
  70struct action_flow_keys {
  71	struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
  72};
  73
  74static struct action_fifo __percpu *action_fifos;
  75static struct action_flow_keys __percpu *flow_keys;
  76static DEFINE_PER_CPU(int, exec_actions_level);
  77
  78/* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
  79 * space. Return NULL if out of key spaces.
  80 */
  81static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
  82{
  83	struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
  84	int level = this_cpu_read(exec_actions_level);
  85	struct sw_flow_key *key = NULL;
  86
  87	if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
  88		key = &keys->key[level - 1];
  89		*key = *key_;
  90	}
  91
  92	return key;
  93}
  94
  95static void action_fifo_init(struct action_fifo *fifo)
  96{
  97	fifo->head = 0;
  98	fifo->tail = 0;
  99}
 100
 101static bool action_fifo_is_empty(const struct action_fifo *fifo)
 102{
 103	return (fifo->head == fifo->tail);
 104}
 105
 106static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
 107{
 108	if (action_fifo_is_empty(fifo))
 109		return NULL;
 110
 111	return &fifo->fifo[fifo->tail++];
 112}
 113
 114static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
 115{
 116	if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
 117		return NULL;
 118
 119	return &fifo->fifo[fifo->head++];
 120}
 121
 122/* Return true if fifo is not full */
 123static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
 124				    const struct sw_flow_key *key,
 125				    const struct nlattr *actions,
 126				    const int actions_len)
 127{
 128	struct action_fifo *fifo;
 129	struct deferred_action *da;
 130
 131	fifo = this_cpu_ptr(action_fifos);
 132	da = action_fifo_put(fifo);
 133	if (da) {
 134		da->skb = skb;
 135		da->actions = actions;
 136		da->actions_len = actions_len;
 137		da->pkt_key = *key;
 138	}
 139
 140	return da;
 141}
 142
 143static void invalidate_flow_key(struct sw_flow_key *key)
 144{
 145	key->mac_proto |= SW_FLOW_KEY_INVALID;
 146}
 147
 148static bool is_flow_key_valid(const struct sw_flow_key *key)
 149{
 150	return !(key->mac_proto & SW_FLOW_KEY_INVALID);
 151}
 152
 153static int clone_execute(struct datapath *dp, struct sk_buff *skb,
 154			 struct sw_flow_key *key,
 155			 u32 recirc_id,
 156			 const struct nlattr *actions, int len,
 157			 bool last, bool clone_flow_key);
 158
 159static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 160			      struct sw_flow_key *key,
 161			      const struct nlattr *attr, int len);
 
 
 
 
 
 
 
 
 
 162
 163static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 164		     __be32 mpls_lse, __be16 mpls_ethertype, __u16 mac_len)
 165{
 166	int err;
 167
 168	err = skb_mpls_push(skb, mpls_lse, mpls_ethertype, mac_len, !!mac_len);
 169	if (err)
 170		return err;
 171
 172	if (!mac_len)
 173		key->mac_proto = MAC_PROTO_NONE;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 174
 175	invalidate_flow_key(key);
 176	return 0;
 177}
 178
 179static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 180		    const __be16 ethertype)
 181{
 182	int err;
 183
 184	err = skb_mpls_pop(skb, ethertype, skb->mac_len,
 185			   ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET);
 186	if (err)
 187		return err;
 188
 189	if (ethertype == htons(ETH_P_TEB))
 190		key->mac_proto = MAC_PROTO_ETHERNET;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 191
 192	invalidate_flow_key(key);
 193	return 0;
 194}
 195
 196static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
 197		    const __be32 *mpls_lse, const __be32 *mask)
 198{
 199	struct mpls_shim_hdr *stack;
 200	__be32 lse;
 201	int err;
 202
 203	if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
 204		return -ENOMEM;
 
 205
 206	stack = mpls_hdr(skb);
 207	lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
 208	err = skb_mpls_update_lse(skb, lse);
 209	if (err)
 210		return err;
 211
 212	flow_key->mpls.lse[0] = lse;
 
 
 
 
 
 213	return 0;
 214}
 215
 216static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 217{
 218	int err;
 219
 220	err = skb_vlan_pop(skb);
 221	if (skb_vlan_tag_present(skb)) {
 222		invalidate_flow_key(key);
 223	} else {
 224		key->eth.vlan.tci = 0;
 225		key->eth.vlan.tpid = 0;
 226	}
 227	return err;
 228}
 229
 230static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
 231		     const struct ovs_action_push_vlan *vlan)
 232{
 233	if (skb_vlan_tag_present(skb)) {
 234		invalidate_flow_key(key);
 235	} else {
 236		key->eth.vlan.tci = vlan->vlan_tci;
 237		key->eth.vlan.tpid = vlan->vlan_tpid;
 238	}
 239	return skb_vlan_push(skb, vlan->vlan_tpid,
 240			     ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
 241}
 242
 243/* 'src' is already properly masked. */
 244static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
 245{
 246	u16 *dst = (u16 *)dst_;
 247	const u16 *src = (const u16 *)src_;
 248	const u16 *mask = (const u16 *)mask_;
 249
 250	OVS_SET_MASKED(dst[0], src[0], mask[0]);
 251	OVS_SET_MASKED(dst[1], src[1], mask[1]);
 252	OVS_SET_MASKED(dst[2], src[2], mask[2]);
 253}
 254
 255static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
 256			const struct ovs_key_ethernet *key,
 257			const struct ovs_key_ethernet *mask)
 258{
 259	int err;
 260
 261	err = skb_ensure_writable(skb, ETH_HLEN);
 262	if (unlikely(err))
 263		return err;
 264
 265	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 266
 267	ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
 268			       mask->eth_src);
 269	ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
 270			       mask->eth_dst);
 271
 272	skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 273
 274	ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
 275	ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
 276	return 0;
 277}
 278
 279/* pop_eth does not support VLAN packets as this action is never called
 280 * for them.
 281 */
 282static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
 283{
 284	int err;
 285
 286	err = skb_eth_pop(skb);
 287	if (err)
 288		return err;
 289
 290	/* safe right before invalidate_flow_key */
 291	key->mac_proto = MAC_PROTO_NONE;
 292	invalidate_flow_key(key);
 293	return 0;
 294}
 295
 296static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
 297		    const struct ovs_action_push_eth *ethh)
 298{
 299	int err;
 300
 301	err = skb_eth_push(skb, ethh->addresses.eth_dst,
 302			   ethh->addresses.eth_src);
 303	if (err)
 304		return err;
 
 
 
 
 
 
 
 
 
 
 305
 306	/* safe right before invalidate_flow_key */
 307	key->mac_proto = MAC_PROTO_ETHERNET;
 308	invalidate_flow_key(key);
 309	return 0;
 310}
 311
 312static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
 313		    const struct nshhdr *nh)
 314{
 315	int err;
 316
 317	err = nsh_push(skb, nh);
 318	if (err)
 319		return err;
 320
 321	/* safe right before invalidate_flow_key */
 322	key->mac_proto = MAC_PROTO_NONE;
 323	invalidate_flow_key(key);
 324	return 0;
 325}
 326
 327static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
 328{
 329	int err;
 330
 331	err = nsh_pop(skb);
 332	if (err)
 333		return err;
 334
 335	/* safe right before invalidate_flow_key */
 336	if (skb->protocol == htons(ETH_P_TEB))
 337		key->mac_proto = MAC_PROTO_ETHERNET;
 338	else
 339		key->mac_proto = MAC_PROTO_NONE;
 340	invalidate_flow_key(key);
 341	return 0;
 342}
 343
 344static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
 345				  __be32 addr, __be32 new_addr)
 346{
 347	int transport_len = skb->len - skb_transport_offset(skb);
 348
 349	if (nh->frag_off & htons(IP_OFFSET))
 350		return;
 351
 352	if (nh->protocol == IPPROTO_TCP) {
 353		if (likely(transport_len >= sizeof(struct tcphdr)))
 354			inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 355						 addr, new_addr, true);
 356	} else if (nh->protocol == IPPROTO_UDP) {
 357		if (likely(transport_len >= sizeof(struct udphdr))) {
 358			struct udphdr *uh = udp_hdr(skb);
 359
 360			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 361				inet_proto_csum_replace4(&uh->check, skb,
 362							 addr, new_addr, true);
 363				if (!uh->check)
 364					uh->check = CSUM_MANGLED_0;
 365			}
 366		}
 367	}
 368}
 369
 370static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 371			__be32 *addr, __be32 new_addr)
 372{
 373	update_ip_l4_checksum(skb, nh, *addr, new_addr);
 374	csum_replace4(&nh->check, *addr, new_addr);
 375	skb_clear_hash(skb);
 376	*addr = new_addr;
 377}
 378
 379static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 380				 __be32 addr[4], const __be32 new_addr[4])
 381{
 382	int transport_len = skb->len - skb_transport_offset(skb);
 383
 384	if (l4_proto == NEXTHDR_TCP) {
 385		if (likely(transport_len >= sizeof(struct tcphdr)))
 386			inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 387						  addr, new_addr, true);
 388	} else if (l4_proto == NEXTHDR_UDP) {
 389		if (likely(transport_len >= sizeof(struct udphdr))) {
 390			struct udphdr *uh = udp_hdr(skb);
 391
 392			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 393				inet_proto_csum_replace16(&uh->check, skb,
 394							  addr, new_addr, true);
 395				if (!uh->check)
 396					uh->check = CSUM_MANGLED_0;
 397			}
 398		}
 399	} else if (l4_proto == NEXTHDR_ICMP) {
 400		if (likely(transport_len >= sizeof(struct icmp6hdr)))
 401			inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 402						  skb, addr, new_addr, true);
 403	}
 404}
 405
 406static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
 407			   const __be32 mask[4], __be32 masked[4])
 408{
 409	masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
 410	masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
 411	masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
 412	masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
 413}
 414
 415static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 416			  __be32 addr[4], const __be32 new_addr[4],
 417			  bool recalculate_csum)
 418{
 419	if (recalculate_csum)
 420		update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 421
 422	skb_clear_hash(skb);
 423	memcpy(addr, new_addr, sizeof(__be32[4]));
 424}
 425
 426static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
 427{
 428	/* Bits 21-24 are always unmasked, so this retains their values. */
 429	OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
 430	OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
 431	OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
 432}
 433
 434static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
 435		       u8 mask)
 436{
 437	new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
 438
 439	csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 440	nh->ttl = new_ttl;
 441}
 442
 443static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
 444		    const struct ovs_key_ipv4 *key,
 445		    const struct ovs_key_ipv4 *mask)
 446{
 447	struct iphdr *nh;
 448	__be32 new_addr;
 449	int err;
 450
 451	err = skb_ensure_writable(skb, skb_network_offset(skb) +
 452				  sizeof(struct iphdr));
 453	if (unlikely(err))
 454		return err;
 455
 456	nh = ip_hdr(skb);
 457
 458	/* Setting an IP addresses is typically only a side effect of
 459	 * matching on them in the current userspace implementation, so it
 460	 * makes sense to check if the value actually changed.
 461	 */
 462	if (mask->ipv4_src) {
 463		new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
 464
 465		if (unlikely(new_addr != nh->saddr)) {
 466			set_ip_addr(skb, nh, &nh->saddr, new_addr);
 467			flow_key->ipv4.addr.src = new_addr;
 468		}
 469	}
 470	if (mask->ipv4_dst) {
 471		new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
 472
 473		if (unlikely(new_addr != nh->daddr)) {
 474			set_ip_addr(skb, nh, &nh->daddr, new_addr);
 475			flow_key->ipv4.addr.dst = new_addr;
 476		}
 477	}
 478	if (mask->ipv4_tos) {
 479		ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
 480		flow_key->ip.tos = nh->tos;
 481	}
 482	if (mask->ipv4_ttl) {
 483		set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
 484		flow_key->ip.ttl = nh->ttl;
 485	}
 486
 487	return 0;
 488}
 489
 490static bool is_ipv6_mask_nonzero(const __be32 addr[4])
 491{
 492	return !!(addr[0] | addr[1] | addr[2] | addr[3]);
 493}
 494
 495static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
 496		    const struct ovs_key_ipv6 *key,
 497		    const struct ovs_key_ipv6 *mask)
 498{
 499	struct ipv6hdr *nh;
 500	int err;
 501
 502	err = skb_ensure_writable(skb, skb_network_offset(skb) +
 503				  sizeof(struct ipv6hdr));
 504	if (unlikely(err))
 505		return err;
 506
 507	nh = ipv6_hdr(skb);
 508
 509	/* Setting an IP addresses is typically only a side effect of
 510	 * matching on them in the current userspace implementation, so it
 511	 * makes sense to check if the value actually changed.
 512	 */
 513	if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
 514		__be32 *saddr = (__be32 *)&nh->saddr;
 515		__be32 masked[4];
 516
 517		mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
 518
 519		if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
 520			set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
 521				      true);
 522			memcpy(&flow_key->ipv6.addr.src, masked,
 523			       sizeof(flow_key->ipv6.addr.src));
 524		}
 525	}
 526	if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
 527		unsigned int offset = 0;
 528		int flags = IP6_FH_F_SKIP_RH;
 529		bool recalc_csum = true;
 530		__be32 *daddr = (__be32 *)&nh->daddr;
 531		__be32 masked[4];
 532
 533		mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
 534
 535		if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
 536			if (ipv6_ext_hdr(nh->nexthdr))
 537				recalc_csum = (ipv6_find_hdr(skb, &offset,
 538							     NEXTHDR_ROUTING,
 539							     NULL, &flags)
 540					       != NEXTHDR_ROUTING);
 541
 542			set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
 543				      recalc_csum);
 544			memcpy(&flow_key->ipv6.addr.dst, masked,
 545			       sizeof(flow_key->ipv6.addr.dst));
 546		}
 547	}
 548	if (mask->ipv6_tclass) {
 549		ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
 550		flow_key->ip.tos = ipv6_get_dsfield(nh);
 551	}
 552	if (mask->ipv6_label) {
 553		set_ipv6_fl(nh, ntohl(key->ipv6_label),
 554			    ntohl(mask->ipv6_label));
 555		flow_key->ipv6.label =
 556		    *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 557	}
 558	if (mask->ipv6_hlimit) {
 559		OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
 560			       mask->ipv6_hlimit);
 561		flow_key->ip.ttl = nh->hop_limit;
 562	}
 563	return 0;
 564}
 565
 566static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
 567		   const struct nlattr *a)
 568{
 569	struct nshhdr *nh;
 570	size_t length;
 571	int err;
 572	u8 flags;
 573	u8 ttl;
 574	int i;
 575
 576	struct ovs_key_nsh key;
 577	struct ovs_key_nsh mask;
 578
 579	err = nsh_key_from_nlattr(a, &key, &mask);
 580	if (err)
 581		return err;
 582
 583	/* Make sure the NSH base header is there */
 584	if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
 585		return -ENOMEM;
 586
 587	nh = nsh_hdr(skb);
 588	length = nsh_hdr_len(nh);
 589
 590	/* Make sure the whole NSH header is there */
 591	err = skb_ensure_writable(skb, skb_network_offset(skb) +
 592				       length);
 593	if (unlikely(err))
 594		return err;
 595
 596	nh = nsh_hdr(skb);
 597	skb_postpull_rcsum(skb, nh, length);
 598	flags = nsh_get_flags(nh);
 599	flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
 600	flow_key->nsh.base.flags = flags;
 601	ttl = nsh_get_ttl(nh);
 602	ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
 603	flow_key->nsh.base.ttl = ttl;
 604	nsh_set_flags_and_ttl(nh, flags, ttl);
 605	nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
 606				  mask.base.path_hdr);
 607	flow_key->nsh.base.path_hdr = nh->path_hdr;
 608	switch (nh->mdtype) {
 609	case NSH_M_TYPE1:
 610		for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
 611			nh->md1.context[i] =
 612			    OVS_MASKED(nh->md1.context[i], key.context[i],
 613				       mask.context[i]);
 614		}
 615		memcpy(flow_key->nsh.context, nh->md1.context,
 616		       sizeof(nh->md1.context));
 617		break;
 618	case NSH_M_TYPE2:
 619		memset(flow_key->nsh.context, 0,
 620		       sizeof(flow_key->nsh.context));
 621		break;
 622	default:
 623		return -EINVAL;
 624	}
 625	skb_postpush_rcsum(skb, nh, length);
 626	return 0;
 627}
 628
 629/* Must follow skb_ensure_writable() since that can move the skb data. */
 630static void set_tp_port(struct sk_buff *skb, __be16 *port,
 631			__be16 new_port, __sum16 *check)
 632{
 633	inet_proto_csum_replace2(check, skb, *port, new_port, false);
 634	*port = new_port;
 635}
 636
 637static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 638		   const struct ovs_key_udp *key,
 639		   const struct ovs_key_udp *mask)
 640{
 641	struct udphdr *uh;
 642	__be16 src, dst;
 643	int err;
 644
 645	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 646				  sizeof(struct udphdr));
 647	if (unlikely(err))
 648		return err;
 649
 650	uh = udp_hdr(skb);
 651	/* Either of the masks is non-zero, so do not bother checking them. */
 652	src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
 653	dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
 654
 655	if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 656		if (likely(src != uh->source)) {
 657			set_tp_port(skb, &uh->source, src, &uh->check);
 658			flow_key->tp.src = src;
 659		}
 660		if (likely(dst != uh->dest)) {
 661			set_tp_port(skb, &uh->dest, dst, &uh->check);
 662			flow_key->tp.dst = dst;
 663		}
 664
 665		if (unlikely(!uh->check))
 666			uh->check = CSUM_MANGLED_0;
 667	} else {
 668		uh->source = src;
 669		uh->dest = dst;
 670		flow_key->tp.src = src;
 671		flow_key->tp.dst = dst;
 672	}
 673
 674	skb_clear_hash(skb);
 675
 676	return 0;
 677}
 678
 679static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 680		   const struct ovs_key_tcp *key,
 681		   const struct ovs_key_tcp *mask)
 682{
 683	struct tcphdr *th;
 684	__be16 src, dst;
 685	int err;
 686
 687	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 688				  sizeof(struct tcphdr));
 689	if (unlikely(err))
 690		return err;
 691
 692	th = tcp_hdr(skb);
 693	src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
 694	if (likely(src != th->source)) {
 695		set_tp_port(skb, &th->source, src, &th->check);
 696		flow_key->tp.src = src;
 697	}
 698	dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
 699	if (likely(dst != th->dest)) {
 700		set_tp_port(skb, &th->dest, dst, &th->check);
 701		flow_key->tp.dst = dst;
 702	}
 703	skb_clear_hash(skb);
 704
 705	return 0;
 706}
 707
 708static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 709		    const struct ovs_key_sctp *key,
 710		    const struct ovs_key_sctp *mask)
 711{
 712	unsigned int sctphoff = skb_transport_offset(skb);
 713	struct sctphdr *sh;
 714	__le32 old_correct_csum, new_csum, old_csum;
 715	int err;
 716
 717	err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
 718	if (unlikely(err))
 719		return err;
 720
 721	sh = sctp_hdr(skb);
 722	old_csum = sh->checksum;
 723	old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 724
 725	sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
 726	sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
 727
 728	new_csum = sctp_compute_cksum(skb, sctphoff);
 729
 730	/* Carry any checksum errors through. */
 731	sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 732
 733	skb_clear_hash(skb);
 734	flow_key->tp.src = sh->source;
 735	flow_key->tp.dst = sh->dest;
 736
 737	return 0;
 738}
 739
 740static int ovs_vport_output(struct net *net, struct sock *sk,
 741			    struct sk_buff *skb)
 742{
 743	struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
 744	struct vport *vport = data->vport;
 745
 746	if (skb_cow_head(skb, data->l2_len) < 0) {
 747		kfree_skb(skb);
 748		return -ENOMEM;
 749	}
 750
 751	__skb_dst_copy(skb, data->dst);
 752	*OVS_CB(skb) = data->cb;
 753	skb->inner_protocol = data->inner_protocol;
 754	if (data->vlan_tci & VLAN_CFI_MASK)
 755		__vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
 756	else
 757		__vlan_hwaccel_clear_tag(skb);
 758
 759	/* Reconstruct the MAC header.  */
 760	skb_push(skb, data->l2_len);
 761	memcpy(skb->data, &data->l2_data, data->l2_len);
 762	skb_postpush_rcsum(skb, skb->data, data->l2_len);
 763	skb_reset_mac_header(skb);
 764
 765	if (eth_p_mpls(skb->protocol)) {
 766		skb->inner_network_header = skb->network_header;
 767		skb_set_network_header(skb, data->network_offset);
 768		skb_reset_mac_len(skb);
 769	}
 770
 771	ovs_vport_send(vport, skb, data->mac_proto);
 772	return 0;
 773}
 774
 775static unsigned int
 776ovs_dst_get_mtu(const struct dst_entry *dst)
 777{
 778	return dst->dev->mtu;
 779}
 780
 781static struct dst_ops ovs_dst_ops = {
 782	.family = AF_UNSPEC,
 783	.mtu = ovs_dst_get_mtu,
 784};
 785
 786/* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
 787 * ovs_vport_output(), which is called once per fragmented packet.
 788 */
 789static void prepare_frag(struct vport *vport, struct sk_buff *skb,
 790			 u16 orig_network_offset, u8 mac_proto)
 791{
 792	unsigned int hlen = skb_network_offset(skb);
 793	struct ovs_frag_data *data;
 794
 795	data = this_cpu_ptr(&ovs_frag_data_storage);
 796	data->dst = skb->_skb_refdst;
 797	data->vport = vport;
 798	data->cb = *OVS_CB(skb);
 799	data->inner_protocol = skb->inner_protocol;
 800	data->network_offset = orig_network_offset;
 801	if (skb_vlan_tag_present(skb))
 802		data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
 803	else
 804		data->vlan_tci = 0;
 805	data->vlan_proto = skb->vlan_proto;
 806	data->mac_proto = mac_proto;
 807	data->l2_len = hlen;
 808	memcpy(&data->l2_data, skb->data, hlen);
 809
 810	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 811	skb_pull(skb, hlen);
 812}
 813
 814static void ovs_fragment(struct net *net, struct vport *vport,
 815			 struct sk_buff *skb, u16 mru,
 816			 struct sw_flow_key *key)
 817{
 818	u16 orig_network_offset = 0;
 819
 820	if (eth_p_mpls(skb->protocol)) {
 821		orig_network_offset = skb_network_offset(skb);
 822		skb->network_header = skb->inner_network_header;
 823	}
 824
 825	if (skb_network_offset(skb) > MAX_L2_LEN) {
 826		OVS_NLERR(1, "L2 header too long to fragment");
 827		goto err;
 828	}
 829
 830	if (key->eth.type == htons(ETH_P_IP)) {
 831		struct rtable ovs_rt = { 0 };
 832		unsigned long orig_dst;
 833
 834		prepare_frag(vport, skb, orig_network_offset,
 835			     ovs_key_mac_proto(key));
 836		dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
 837			 DST_OBSOLETE_NONE, DST_NOCOUNT);
 838		ovs_rt.dst.dev = vport->dev;
 839
 840		orig_dst = skb->_skb_refdst;
 841		skb_dst_set_noref(skb, &ovs_rt.dst);
 842		IPCB(skb)->frag_max_size = mru;
 843
 844		ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
 845		refdst_drop(orig_dst);
 846	} else if (key->eth.type == htons(ETH_P_IPV6)) {
 
 847		unsigned long orig_dst;
 848		struct rt6_info ovs_rt;
 849
 
 
 
 850		prepare_frag(vport, skb, orig_network_offset,
 851			     ovs_key_mac_proto(key));
 852		memset(&ovs_rt, 0, sizeof(ovs_rt));
 853		dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
 854			 DST_OBSOLETE_NONE, DST_NOCOUNT);
 855		ovs_rt.dst.dev = vport->dev;
 856
 857		orig_dst = skb->_skb_refdst;
 858		skb_dst_set_noref(skb, &ovs_rt.dst);
 859		IP6CB(skb)->frag_max_size = mru;
 860
 861		ipv6_stub->ipv6_fragment(net, skb->sk, skb, ovs_vport_output);
 862		refdst_drop(orig_dst);
 863	} else {
 864		WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
 865			  ovs_vport_name(vport), ntohs(key->eth.type), mru,
 866			  vport->dev->mtu);
 867		goto err;
 868	}
 869
 870	return;
 871err:
 872	kfree_skb(skb);
 873}
 874
 875static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
 876		      struct sw_flow_key *key)
 877{
 878	struct vport *vport = ovs_vport_rcu(dp, out_port);
 879
 880	if (likely(vport)) {
 881		u16 mru = OVS_CB(skb)->mru;
 882		u32 cutlen = OVS_CB(skb)->cutlen;
 883
 884		if (unlikely(cutlen > 0)) {
 885			if (skb->len - cutlen > ovs_mac_header_len(key))
 886				pskb_trim(skb, skb->len - cutlen);
 887			else
 888				pskb_trim(skb, ovs_mac_header_len(key));
 889		}
 890
 891		if (likely(!mru ||
 892		           (skb->len <= mru + vport->dev->hard_header_len))) {
 893			ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
 894		} else if (mru <= vport->dev->mtu) {
 895			struct net *net = read_pnet(&dp->net);
 896
 897			ovs_fragment(net, vport, skb, mru, key);
 898		} else {
 899			kfree_skb(skb);
 900		}
 901	} else {
 902		kfree_skb(skb);
 903	}
 904}
 905
 906static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 907			    struct sw_flow_key *key, const struct nlattr *attr,
 908			    const struct nlattr *actions, int actions_len,
 909			    uint32_t cutlen)
 910{
 911	struct dp_upcall_info upcall;
 912	const struct nlattr *a;
 913	int rem;
 914
 915	memset(&upcall, 0, sizeof(upcall));
 916	upcall.cmd = OVS_PACKET_CMD_ACTION;
 917	upcall.mru = OVS_CB(skb)->mru;
 918
 919	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 920	     a = nla_next(a, &rem)) {
 921		switch (nla_type(a)) {
 922		case OVS_USERSPACE_ATTR_USERDATA:
 923			upcall.userdata = a;
 924			break;
 925
 926		case OVS_USERSPACE_ATTR_PID:
 927			upcall.portid = nla_get_u32(a);
 928			break;
 929
 930		case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
 931			/* Get out tunnel info. */
 932			struct vport *vport;
 933
 934			vport = ovs_vport_rcu(dp, nla_get_u32(a));
 935			if (vport) {
 936				int err;
 937
 938				err = dev_fill_metadata_dst(vport->dev, skb);
 939				if (!err)
 940					upcall.egress_tun_info = skb_tunnel_info(skb);
 941			}
 942
 943			break;
 944		}
 945
 946		case OVS_USERSPACE_ATTR_ACTIONS: {
 947			/* Include actions. */
 948			upcall.actions = actions;
 949			upcall.actions_len = actions_len;
 950			break;
 951		}
 952
 953		} /* End of switch. */
 954	}
 955
 956	return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
 957}
 958
 959static int dec_ttl_exception_handler(struct datapath *dp, struct sk_buff *skb,
 960				     struct sw_flow_key *key,
 961				     const struct nlattr *attr)
 962{
 963	/* The first attribute is always 'OVS_DEC_TTL_ATTR_ACTION'. */
 964	struct nlattr *actions = nla_data(attr);
 965
 966	if (nla_len(actions))
 967		return clone_execute(dp, skb, key, 0, nla_data(actions),
 968				     nla_len(actions), true, false);
 969
 970	consume_skb(skb);
 971	return 0;
 972}
 973
 974/* When 'last' is true, sample() should always consume the 'skb'.
 975 * Otherwise, sample() should keep 'skb' intact regardless what
 976 * actions are executed within sample().
 977 */
 978static int sample(struct datapath *dp, struct sk_buff *skb,
 979		  struct sw_flow_key *key, const struct nlattr *attr,
 980		  bool last)
 981{
 982	struct nlattr *actions;
 983	struct nlattr *sample_arg;
 984	int rem = nla_len(attr);
 985	const struct sample_arg *arg;
 986	bool clone_flow_key;
 987
 988	/* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
 989	sample_arg = nla_data(attr);
 990	arg = nla_data(sample_arg);
 991	actions = nla_next(sample_arg, &rem);
 992
 993	if ((arg->probability != U32_MAX) &&
 994	    (!arg->probability || prandom_u32() > arg->probability)) {
 995		if (last)
 996			consume_skb(skb);
 997		return 0;
 998	}
 999
1000	clone_flow_key = !arg->exec;
1001	return clone_execute(dp, skb, key, 0, actions, rem, last,
1002			     clone_flow_key);
1003}
1004
1005/* When 'last' is true, clone() should always consume the 'skb'.
1006 * Otherwise, clone() should keep 'skb' intact regardless what
1007 * actions are executed within clone().
1008 */
1009static int clone(struct datapath *dp, struct sk_buff *skb,
1010		 struct sw_flow_key *key, const struct nlattr *attr,
1011		 bool last)
1012{
1013	struct nlattr *actions;
1014	struct nlattr *clone_arg;
1015	int rem = nla_len(attr);
1016	bool dont_clone_flow_key;
1017
1018	/* The first action is always 'OVS_CLONE_ATTR_ARG'. */
1019	clone_arg = nla_data(attr);
1020	dont_clone_flow_key = nla_get_u32(clone_arg);
1021	actions = nla_next(clone_arg, &rem);
1022
1023	return clone_execute(dp, skb, key, 0, actions, rem, last,
1024			     !dont_clone_flow_key);
1025}
1026
1027static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1028			 const struct nlattr *attr)
1029{
1030	struct ovs_action_hash *hash_act = nla_data(attr);
1031	u32 hash = 0;
1032
1033	/* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
1034	hash = skb_get_hash(skb);
1035	hash = jhash_1word(hash, hash_act->hash_basis);
1036	if (!hash)
1037		hash = 0x1;
1038
1039	key->ovs_flow_hash = hash;
1040}
1041
1042static int execute_set_action(struct sk_buff *skb,
1043			      struct sw_flow_key *flow_key,
1044			      const struct nlattr *a)
1045{
1046	/* Only tunnel set execution is supported without a mask. */
1047	if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1048		struct ovs_tunnel_info *tun = nla_data(a);
1049
1050		skb_dst_drop(skb);
1051		dst_hold((struct dst_entry *)tun->tun_dst);
1052		skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1053		return 0;
1054	}
1055
1056	return -EINVAL;
1057}
1058
1059/* Mask is at the midpoint of the data. */
1060#define get_mask(a, type) ((const type)nla_data(a) + 1)
1061
1062static int execute_masked_set_action(struct sk_buff *skb,
1063				     struct sw_flow_key *flow_key,
1064				     const struct nlattr *a)
1065{
1066	int err = 0;
1067
1068	switch (nla_type(a)) {
1069	case OVS_KEY_ATTR_PRIORITY:
1070		OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1071			       *get_mask(a, u32 *));
1072		flow_key->phy.priority = skb->priority;
1073		break;
1074
1075	case OVS_KEY_ATTR_SKB_MARK:
1076		OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1077		flow_key->phy.skb_mark = skb->mark;
1078		break;
1079
1080	case OVS_KEY_ATTR_TUNNEL_INFO:
1081		/* Masked data not supported for tunnel. */
1082		err = -EINVAL;
1083		break;
1084
1085	case OVS_KEY_ATTR_ETHERNET:
1086		err = set_eth_addr(skb, flow_key, nla_data(a),
1087				   get_mask(a, struct ovs_key_ethernet *));
1088		break;
1089
1090	case OVS_KEY_ATTR_NSH:
1091		err = set_nsh(skb, flow_key, a);
1092		break;
1093
1094	case OVS_KEY_ATTR_IPV4:
1095		err = set_ipv4(skb, flow_key, nla_data(a),
1096			       get_mask(a, struct ovs_key_ipv4 *));
1097		break;
1098
1099	case OVS_KEY_ATTR_IPV6:
1100		err = set_ipv6(skb, flow_key, nla_data(a),
1101			       get_mask(a, struct ovs_key_ipv6 *));
1102		break;
1103
1104	case OVS_KEY_ATTR_TCP:
1105		err = set_tcp(skb, flow_key, nla_data(a),
1106			      get_mask(a, struct ovs_key_tcp *));
1107		break;
1108
1109	case OVS_KEY_ATTR_UDP:
1110		err = set_udp(skb, flow_key, nla_data(a),
1111			      get_mask(a, struct ovs_key_udp *));
1112		break;
1113
1114	case OVS_KEY_ATTR_SCTP:
1115		err = set_sctp(skb, flow_key, nla_data(a),
1116			       get_mask(a, struct ovs_key_sctp *));
1117		break;
1118
1119	case OVS_KEY_ATTR_MPLS:
1120		err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1121								    __be32 *));
1122		break;
1123
1124	case OVS_KEY_ATTR_CT_STATE:
1125	case OVS_KEY_ATTR_CT_ZONE:
1126	case OVS_KEY_ATTR_CT_MARK:
1127	case OVS_KEY_ATTR_CT_LABELS:
1128	case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1129	case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1130		err = -EINVAL;
1131		break;
1132	}
1133
1134	return err;
1135}
1136
1137static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1138			  struct sw_flow_key *key,
1139			  const struct nlattr *a, bool last)
1140{
1141	u32 recirc_id;
1142
1143	if (!is_flow_key_valid(key)) {
1144		int err;
1145
1146		err = ovs_flow_key_update(skb, key);
1147		if (err)
1148			return err;
1149	}
1150	BUG_ON(!is_flow_key_valid(key));
1151
1152	recirc_id = nla_get_u32(a);
1153	return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1154}
1155
1156static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1157				 struct sw_flow_key *key,
1158				 const struct nlattr *attr, bool last)
1159{
1160	struct ovs_skb_cb *ovs_cb = OVS_CB(skb);
1161	const struct nlattr *actions, *cpl_arg;
1162	int len, max_len, rem = nla_len(attr);
1163	const struct check_pkt_len_arg *arg;
1164	bool clone_flow_key;
1165
1166	/* The first netlink attribute in 'attr' is always
1167	 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1168	 */
1169	cpl_arg = nla_data(attr);
1170	arg = nla_data(cpl_arg);
1171
1172	len = ovs_cb->mru ? ovs_cb->mru + skb->mac_len : skb->len;
1173	max_len = arg->pkt_len;
1174
1175	if ((skb_is_gso(skb) && skb_gso_validate_mac_len(skb, max_len)) ||
1176	    len <= max_len) {
1177		/* Second netlink attribute in 'attr' is always
1178		 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1179		 */
1180		actions = nla_next(cpl_arg, &rem);
1181		clone_flow_key = !arg->exec_for_lesser_equal;
1182	} else {
1183		/* Third netlink attribute in 'attr' is always
1184		 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1185		 */
1186		actions = nla_next(cpl_arg, &rem);
1187		actions = nla_next(actions, &rem);
1188		clone_flow_key = !arg->exec_for_greater;
1189	}
1190
1191	return clone_execute(dp, skb, key, 0, nla_data(actions),
1192			     nla_len(actions), last, clone_flow_key);
1193}
1194
1195static int execute_dec_ttl(struct sk_buff *skb, struct sw_flow_key *key)
1196{
1197	int err;
1198
1199	if (skb->protocol == htons(ETH_P_IPV6)) {
1200		struct ipv6hdr *nh;
1201
1202		err = skb_ensure_writable(skb, skb_network_offset(skb) +
1203					  sizeof(*nh));
1204		if (unlikely(err))
1205			return err;
1206
1207		nh = ipv6_hdr(skb);
1208
1209		if (nh->hop_limit <= 1)
1210			return -EHOSTUNREACH;
1211
1212		key->ip.ttl = --nh->hop_limit;
1213	} else if (skb->protocol == htons(ETH_P_IP)) {
1214		struct iphdr *nh;
1215		u8 old_ttl;
1216
1217		err = skb_ensure_writable(skb, skb_network_offset(skb) +
1218					  sizeof(*nh));
1219		if (unlikely(err))
1220			return err;
1221
1222		nh = ip_hdr(skb);
1223		if (nh->ttl <= 1)
1224			return -EHOSTUNREACH;
1225
1226		old_ttl = nh->ttl--;
1227		csum_replace2(&nh->check, htons(old_ttl << 8),
1228			      htons(nh->ttl << 8));
1229		key->ip.ttl = nh->ttl;
1230	}
1231	return 0;
1232}
1233
1234/* Execute a list of actions against 'skb'. */
1235static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1236			      struct sw_flow_key *key,
1237			      const struct nlattr *attr, int len)
1238{
1239	const struct nlattr *a;
1240	int rem;
1241
1242	for (a = attr, rem = len; rem > 0;
1243	     a = nla_next(a, &rem)) {
1244		int err = 0;
1245
1246		if (trace_ovs_do_execute_action_enabled())
1247			trace_ovs_do_execute_action(dp, skb, key, a, rem);
1248
1249		switch (nla_type(a)) {
1250		case OVS_ACTION_ATTR_OUTPUT: {
1251			int port = nla_get_u32(a);
1252			struct sk_buff *clone;
1253
1254			/* Every output action needs a separate clone
1255			 * of 'skb', In case the output action is the
1256			 * last action, cloning can be avoided.
1257			 */
1258			if (nla_is_last(a, rem)) {
1259				do_output(dp, skb, port, key);
1260				/* 'skb' has been used for output.
1261				 */
1262				return 0;
1263			}
1264
1265			clone = skb_clone(skb, GFP_ATOMIC);
1266			if (clone)
1267				do_output(dp, clone, port, key);
1268			OVS_CB(skb)->cutlen = 0;
1269			break;
1270		}
1271
1272		case OVS_ACTION_ATTR_TRUNC: {
1273			struct ovs_action_trunc *trunc = nla_data(a);
1274
1275			if (skb->len > trunc->max_len)
1276				OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1277			break;
1278		}
1279
1280		case OVS_ACTION_ATTR_USERSPACE:
1281			output_userspace(dp, skb, key, a, attr,
1282						     len, OVS_CB(skb)->cutlen);
1283			OVS_CB(skb)->cutlen = 0;
1284			break;
1285
1286		case OVS_ACTION_ATTR_HASH:
1287			execute_hash(skb, key, a);
1288			break;
1289
1290		case OVS_ACTION_ATTR_PUSH_MPLS: {
1291			struct ovs_action_push_mpls *mpls = nla_data(a);
1292
1293			err = push_mpls(skb, key, mpls->mpls_lse,
1294					mpls->mpls_ethertype, skb->mac_len);
1295			break;
1296		}
1297		case OVS_ACTION_ATTR_ADD_MPLS: {
1298			struct ovs_action_add_mpls *mpls = nla_data(a);
1299			__u16 mac_len = 0;
1300
1301			if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK)
1302				mac_len = skb->mac_len;
1303
1304			err = push_mpls(skb, key, mpls->mpls_lse,
1305					mpls->mpls_ethertype, mac_len);
1306			break;
1307		}
1308		case OVS_ACTION_ATTR_POP_MPLS:
1309			err = pop_mpls(skb, key, nla_get_be16(a));
1310			break;
1311
1312		case OVS_ACTION_ATTR_PUSH_VLAN:
1313			err = push_vlan(skb, key, nla_data(a));
1314			break;
1315
1316		case OVS_ACTION_ATTR_POP_VLAN:
1317			err = pop_vlan(skb, key);
1318			break;
1319
1320		case OVS_ACTION_ATTR_RECIRC: {
1321			bool last = nla_is_last(a, rem);
1322
1323			err = execute_recirc(dp, skb, key, a, last);
1324			if (last) {
1325				/* If this is the last action, the skb has
1326				 * been consumed or freed.
1327				 * Return immediately.
1328				 */
1329				return err;
1330			}
1331			break;
1332		}
1333
1334		case OVS_ACTION_ATTR_SET:
1335			err = execute_set_action(skb, key, nla_data(a));
1336			break;
1337
1338		case OVS_ACTION_ATTR_SET_MASKED:
1339		case OVS_ACTION_ATTR_SET_TO_MASKED:
1340			err = execute_masked_set_action(skb, key, nla_data(a));
1341			break;
1342
1343		case OVS_ACTION_ATTR_SAMPLE: {
1344			bool last = nla_is_last(a, rem);
1345
1346			err = sample(dp, skb, key, a, last);
1347			if (last)
1348				return err;
1349
1350			break;
1351		}
1352
1353		case OVS_ACTION_ATTR_CT:
1354			if (!is_flow_key_valid(key)) {
1355				err = ovs_flow_key_update(skb, key);
1356				if (err)
1357					return err;
1358			}
1359
1360			err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1361					     nla_data(a));
1362
1363			/* Hide stolen IP fragments from user space. */
1364			if (err)
1365				return err == -EINPROGRESS ? 0 : err;
1366			break;
1367
1368		case OVS_ACTION_ATTR_CT_CLEAR:
1369			err = ovs_ct_clear(skb, key);
1370			break;
1371
1372		case OVS_ACTION_ATTR_PUSH_ETH:
1373			err = push_eth(skb, key, nla_data(a));
1374			break;
1375
1376		case OVS_ACTION_ATTR_POP_ETH:
1377			err = pop_eth(skb, key);
1378			break;
1379
1380		case OVS_ACTION_ATTR_PUSH_NSH: {
1381			u8 buffer[NSH_HDR_MAX_LEN];
1382			struct nshhdr *nh = (struct nshhdr *)buffer;
1383
1384			err = nsh_hdr_from_nlattr(nla_data(a), nh,
1385						  NSH_HDR_MAX_LEN);
1386			if (unlikely(err))
1387				break;
1388			err = push_nsh(skb, key, nh);
1389			break;
1390		}
1391
1392		case OVS_ACTION_ATTR_POP_NSH:
1393			err = pop_nsh(skb, key);
1394			break;
1395
1396		case OVS_ACTION_ATTR_METER:
1397			if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1398				consume_skb(skb);
1399				return 0;
1400			}
1401			break;
1402
1403		case OVS_ACTION_ATTR_CLONE: {
1404			bool last = nla_is_last(a, rem);
1405
1406			err = clone(dp, skb, key, a, last);
1407			if (last)
1408				return err;
1409
1410			break;
1411		}
1412
1413		case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1414			bool last = nla_is_last(a, rem);
1415
1416			err = execute_check_pkt_len(dp, skb, key, a, last);
1417			if (last)
1418				return err;
1419
1420			break;
1421		}
1422
1423		case OVS_ACTION_ATTR_DEC_TTL:
1424			err = execute_dec_ttl(skb, key);
1425			if (err == -EHOSTUNREACH)
1426				return dec_ttl_exception_handler(dp, skb,
1427								 key, a);
1428			break;
1429		}
1430
1431		if (unlikely(err)) {
1432			kfree_skb(skb);
1433			return err;
1434		}
1435	}
1436
1437	consume_skb(skb);
1438	return 0;
1439}
1440
1441/* Execute the actions on the clone of the packet. The effect of the
1442 * execution does not affect the original 'skb' nor the original 'key'.
1443 *
1444 * The execution may be deferred in case the actions can not be executed
1445 * immediately.
1446 */
1447static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1448			 struct sw_flow_key *key, u32 recirc_id,
1449			 const struct nlattr *actions, int len,
1450			 bool last, bool clone_flow_key)
1451{
1452	struct deferred_action *da;
1453	struct sw_flow_key *clone;
1454
1455	skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1456	if (!skb) {
1457		/* Out of memory, skip this action.
1458		 */
1459		return 0;
1460	}
1461
1462	/* When clone_flow_key is false, the 'key' will not be change
1463	 * by the actions, then the 'key' can be used directly.
1464	 * Otherwise, try to clone key from the next recursion level of
1465	 * 'flow_keys'. If clone is successful, execute the actions
1466	 * without deferring.
1467	 */
1468	clone = clone_flow_key ? clone_key(key) : key;
1469	if (clone) {
1470		int err = 0;
1471
1472		if (actions) { /* Sample action */
1473			if (clone_flow_key)
1474				__this_cpu_inc(exec_actions_level);
1475
1476			err = do_execute_actions(dp, skb, clone,
1477						 actions, len);
1478
1479			if (clone_flow_key)
1480				__this_cpu_dec(exec_actions_level);
1481		} else { /* Recirc action */
1482			clone->recirc_id = recirc_id;
1483			ovs_dp_process_packet(skb, clone);
1484		}
1485		return err;
1486	}
1487
1488	/* Out of 'flow_keys' space. Defer actions */
1489	da = add_deferred_actions(skb, key, actions, len);
1490	if (da) {
1491		if (!actions) { /* Recirc action */
1492			key = &da->pkt_key;
1493			key->recirc_id = recirc_id;
1494		}
1495	} else {
1496		/* Out of per CPU action FIFO space. Drop the 'skb' and
1497		 * log an error.
1498		 */
1499		kfree_skb(skb);
1500
1501		if (net_ratelimit()) {
1502			if (actions) { /* Sample action */
1503				pr_warn("%s: deferred action limit reached, drop sample action\n",
1504					ovs_dp_name(dp));
1505			} else {  /* Recirc action */
1506				pr_warn("%s: deferred action limit reached, drop recirc action\n",
1507					ovs_dp_name(dp));
1508			}
1509		}
1510	}
1511	return 0;
1512}
1513
1514static void process_deferred_actions(struct datapath *dp)
1515{
1516	struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1517
1518	/* Do not touch the FIFO in case there is no deferred actions. */
1519	if (action_fifo_is_empty(fifo))
1520		return;
1521
1522	/* Finishing executing all deferred actions. */
1523	do {
1524		struct deferred_action *da = action_fifo_get(fifo);
1525		struct sk_buff *skb = da->skb;
1526		struct sw_flow_key *key = &da->pkt_key;
1527		const struct nlattr *actions = da->actions;
1528		int actions_len = da->actions_len;
1529
1530		if (actions)
1531			do_execute_actions(dp, skb, key, actions, actions_len);
1532		else
1533			ovs_dp_process_packet(skb, key);
1534	} while (!action_fifo_is_empty(fifo));
1535
1536	/* Reset FIFO for the next packet.  */
1537	action_fifo_init(fifo);
1538}
1539
1540/* Execute a list of actions against 'skb'. */
1541int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1542			const struct sw_flow_actions *acts,
1543			struct sw_flow_key *key)
1544{
1545	int err, level;
1546
1547	level = __this_cpu_inc_return(exec_actions_level);
1548	if (unlikely(level > OVS_RECURSION_LIMIT)) {
1549		net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1550				     ovs_dp_name(dp));
1551		kfree_skb(skb);
1552		err = -ENETDOWN;
1553		goto out;
1554	}
1555
1556	OVS_CB(skb)->acts_origlen = acts->orig_len;
1557	err = do_execute_actions(dp, skb, key,
1558				 acts->actions, acts->actions_len);
1559
1560	if (level == 1)
1561		process_deferred_actions(dp);
1562
1563out:
1564	__this_cpu_dec(exec_actions_level);
1565	return err;
1566}
1567
1568int action_fifos_init(void)
1569{
1570	action_fifos = alloc_percpu(struct action_fifo);
1571	if (!action_fifos)
1572		return -ENOMEM;
1573
1574	flow_keys = alloc_percpu(struct action_flow_keys);
1575	if (!flow_keys) {
1576		free_percpu(action_fifos);
1577		return -ENOMEM;
1578	}
1579
1580	return 0;
1581}
1582
1583void action_fifos_exit(void)
1584{
1585	free_percpu(action_fifos);
1586	free_percpu(flow_keys);
1587}
v4.17
 
   1/*
   2 * Copyright (c) 2007-2017 Nicira, Inc.
   3 *
   4 * This program is free software; you can redistribute it and/or
   5 * modify it under the terms of version 2 of the GNU General Public
   6 * License as published by the Free Software Foundation.
   7 *
   8 * This program is distributed in the hope that it will be useful, but
   9 * WITHOUT ANY WARRANTY; without even the implied warranty of
  10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11 * General Public License for more details.
  12 *
  13 * You should have received a copy of the GNU General Public License
  14 * along with this program; if not, write to the Free Software
  15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16 * 02110-1301, USA
  17 */
  18
  19#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21#include <linux/skbuff.h>
  22#include <linux/in.h>
  23#include <linux/ip.h>
  24#include <linux/openvswitch.h>
  25#include <linux/netfilter_ipv6.h>
  26#include <linux/sctp.h>
  27#include <linux/tcp.h>
  28#include <linux/udp.h>
  29#include <linux/in6.h>
  30#include <linux/if_arp.h>
  31#include <linux/if_vlan.h>
  32
  33#include <net/dst.h>
  34#include <net/ip.h>
  35#include <net/ipv6.h>
  36#include <net/ip6_fib.h>
  37#include <net/checksum.h>
  38#include <net/dsfield.h>
  39#include <net/mpls.h>
  40#include <net/sctp/checksum.h>
  41
  42#include "datapath.h"
  43#include "flow.h"
  44#include "conntrack.h"
  45#include "vport.h"
  46#include "flow_netlink.h"
 
  47
  48struct deferred_action {
  49	struct sk_buff *skb;
  50	const struct nlattr *actions;
  51	int actions_len;
  52
  53	/* Store pkt_key clone when creating deferred action. */
  54	struct sw_flow_key pkt_key;
  55};
  56
  57#define MAX_L2_LEN	(VLAN_ETH_HLEN + 3 * MPLS_HLEN)
  58struct ovs_frag_data {
  59	unsigned long dst;
  60	struct vport *vport;
  61	struct ovs_skb_cb cb;
  62	__be16 inner_protocol;
  63	u16 network_offset;	/* valid only for MPLS */
  64	u16 vlan_tci;
  65	__be16 vlan_proto;
  66	unsigned int l2_len;
  67	u8 mac_proto;
  68	u8 l2_data[MAX_L2_LEN];
  69};
  70
  71static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
  72
  73#define DEFERRED_ACTION_FIFO_SIZE 10
  74#define OVS_RECURSION_LIMIT 5
  75#define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
  76struct action_fifo {
  77	int head;
  78	int tail;
  79	/* Deferred action fifo queue storage. */
  80	struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
  81};
  82
  83struct action_flow_keys {
  84	struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
  85};
  86
  87static struct action_fifo __percpu *action_fifos;
  88static struct action_flow_keys __percpu *flow_keys;
  89static DEFINE_PER_CPU(int, exec_actions_level);
  90
  91/* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
  92 * space. Return NULL if out of key spaces.
  93 */
  94static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
  95{
  96	struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
  97	int level = this_cpu_read(exec_actions_level);
  98	struct sw_flow_key *key = NULL;
  99
 100	if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
 101		key = &keys->key[level - 1];
 102		*key = *key_;
 103	}
 104
 105	return key;
 106}
 107
 108static void action_fifo_init(struct action_fifo *fifo)
 109{
 110	fifo->head = 0;
 111	fifo->tail = 0;
 112}
 113
 114static bool action_fifo_is_empty(const struct action_fifo *fifo)
 115{
 116	return (fifo->head == fifo->tail);
 117}
 118
 119static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
 120{
 121	if (action_fifo_is_empty(fifo))
 122		return NULL;
 123
 124	return &fifo->fifo[fifo->tail++];
 125}
 126
 127static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
 128{
 129	if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
 130		return NULL;
 131
 132	return &fifo->fifo[fifo->head++];
 133}
 134
 135/* Return true if fifo is not full */
 136static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
 137				    const struct sw_flow_key *key,
 138				    const struct nlattr *actions,
 139				    const int actions_len)
 140{
 141	struct action_fifo *fifo;
 142	struct deferred_action *da;
 143
 144	fifo = this_cpu_ptr(action_fifos);
 145	da = action_fifo_put(fifo);
 146	if (da) {
 147		da->skb = skb;
 148		da->actions = actions;
 149		da->actions_len = actions_len;
 150		da->pkt_key = *key;
 151	}
 152
 153	return da;
 154}
 155
 156static void invalidate_flow_key(struct sw_flow_key *key)
 157{
 158	key->mac_proto |= SW_FLOW_KEY_INVALID;
 159}
 160
 161static bool is_flow_key_valid(const struct sw_flow_key *key)
 162{
 163	return !(key->mac_proto & SW_FLOW_KEY_INVALID);
 164}
 165
 166static int clone_execute(struct datapath *dp, struct sk_buff *skb,
 167			 struct sw_flow_key *key,
 168			 u32 recirc_id,
 169			 const struct nlattr *actions, int len,
 170			 bool last, bool clone_flow_key);
 171
 172static void update_ethertype(struct sk_buff *skb, struct ethhdr *hdr,
 173			     __be16 ethertype)
 174{
 175	if (skb->ip_summed == CHECKSUM_COMPLETE) {
 176		__be16 diff[] = { ~(hdr->h_proto), ethertype };
 177
 178		skb->csum = ~csum_partial((char *)diff, sizeof(diff),
 179					~skb->csum);
 180	}
 181
 182	hdr->h_proto = ethertype;
 183}
 184
 185static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 186		     const struct ovs_action_push_mpls *mpls)
 187{
 188	struct mpls_shim_hdr *new_mpls_lse;
 189
 190	/* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
 191	if (skb->encapsulation)
 192		return -ENOTSUPP;
 193
 194	if (skb_cow_head(skb, MPLS_HLEN) < 0)
 195		return -ENOMEM;
 196
 197	if (!skb->inner_protocol) {
 198		skb_set_inner_network_header(skb, skb->mac_len);
 199		skb_set_inner_protocol(skb, skb->protocol);
 200	}
 201
 202	skb_push(skb, MPLS_HLEN);
 203	memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
 204		skb->mac_len);
 205	skb_reset_mac_header(skb);
 206	skb_set_network_header(skb, skb->mac_len);
 207
 208	new_mpls_lse = mpls_hdr(skb);
 209	new_mpls_lse->label_stack_entry = mpls->mpls_lse;
 210
 211	skb_postpush_rcsum(skb, new_mpls_lse, MPLS_HLEN);
 212
 213	if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET)
 214		update_ethertype(skb, eth_hdr(skb), mpls->mpls_ethertype);
 215	skb->protocol = mpls->mpls_ethertype;
 216
 217	invalidate_flow_key(key);
 218	return 0;
 219}
 220
 221static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 222		    const __be16 ethertype)
 223{
 224	int err;
 225
 226	err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
 227	if (unlikely(err))
 
 228		return err;
 229
 230	skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
 231
 232	memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
 233		skb->mac_len);
 234
 235	__skb_pull(skb, MPLS_HLEN);
 236	skb_reset_mac_header(skb);
 237	skb_set_network_header(skb, skb->mac_len);
 238
 239	if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET) {
 240		struct ethhdr *hdr;
 241
 242		/* mpls_hdr() is used to locate the ethertype field correctly in the
 243		 * presence of VLAN tags.
 244		 */
 245		hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
 246		update_ethertype(skb, hdr, ethertype);
 247	}
 248	if (eth_p_mpls(skb->protocol))
 249		skb->protocol = ethertype;
 250
 251	invalidate_flow_key(key);
 252	return 0;
 253}
 254
 255static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
 256		    const __be32 *mpls_lse, const __be32 *mask)
 257{
 258	struct mpls_shim_hdr *stack;
 259	__be32 lse;
 260	int err;
 261
 262	err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
 263	if (unlikely(err))
 264		return err;
 265
 266	stack = mpls_hdr(skb);
 267	lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
 268	if (skb->ip_summed == CHECKSUM_COMPLETE) {
 269		__be32 diff[] = { ~(stack->label_stack_entry), lse };
 
 270
 271		skb->csum = ~csum_partial((char *)diff, sizeof(diff),
 272					  ~skb->csum);
 273	}
 274
 275	stack->label_stack_entry = lse;
 276	flow_key->mpls.top_lse = lse;
 277	return 0;
 278}
 279
 280static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 281{
 282	int err;
 283
 284	err = skb_vlan_pop(skb);
 285	if (skb_vlan_tag_present(skb)) {
 286		invalidate_flow_key(key);
 287	} else {
 288		key->eth.vlan.tci = 0;
 289		key->eth.vlan.tpid = 0;
 290	}
 291	return err;
 292}
 293
 294static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
 295		     const struct ovs_action_push_vlan *vlan)
 296{
 297	if (skb_vlan_tag_present(skb)) {
 298		invalidate_flow_key(key);
 299	} else {
 300		key->eth.vlan.tci = vlan->vlan_tci;
 301		key->eth.vlan.tpid = vlan->vlan_tpid;
 302	}
 303	return skb_vlan_push(skb, vlan->vlan_tpid,
 304			     ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
 305}
 306
 307/* 'src' is already properly masked. */
 308static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
 309{
 310	u16 *dst = (u16 *)dst_;
 311	const u16 *src = (const u16 *)src_;
 312	const u16 *mask = (const u16 *)mask_;
 313
 314	OVS_SET_MASKED(dst[0], src[0], mask[0]);
 315	OVS_SET_MASKED(dst[1], src[1], mask[1]);
 316	OVS_SET_MASKED(dst[2], src[2], mask[2]);
 317}
 318
 319static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
 320			const struct ovs_key_ethernet *key,
 321			const struct ovs_key_ethernet *mask)
 322{
 323	int err;
 324
 325	err = skb_ensure_writable(skb, ETH_HLEN);
 326	if (unlikely(err))
 327		return err;
 328
 329	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 330
 331	ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
 332			       mask->eth_src);
 333	ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
 334			       mask->eth_dst);
 335
 336	skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 337
 338	ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
 339	ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
 340	return 0;
 341}
 342
 343/* pop_eth does not support VLAN packets as this action is never called
 344 * for them.
 345 */
 346static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
 347{
 348	skb_pull_rcsum(skb, ETH_HLEN);
 349	skb_reset_mac_header(skb);
 350	skb_reset_mac_len(skb);
 
 
 351
 352	/* safe right before invalidate_flow_key */
 353	key->mac_proto = MAC_PROTO_NONE;
 354	invalidate_flow_key(key);
 355	return 0;
 356}
 357
 358static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
 359		    const struct ovs_action_push_eth *ethh)
 360{
 361	struct ethhdr *hdr;
 362
 363	/* Add the new Ethernet header */
 364	if (skb_cow_head(skb, ETH_HLEN) < 0)
 365		return -ENOMEM;
 366
 367	skb_push(skb, ETH_HLEN);
 368	skb_reset_mac_header(skb);
 369	skb_reset_mac_len(skb);
 370
 371	hdr = eth_hdr(skb);
 372	ether_addr_copy(hdr->h_source, ethh->addresses.eth_src);
 373	ether_addr_copy(hdr->h_dest, ethh->addresses.eth_dst);
 374	hdr->h_proto = skb->protocol;
 375
 376	skb_postpush_rcsum(skb, hdr, ETH_HLEN);
 377
 378	/* safe right before invalidate_flow_key */
 379	key->mac_proto = MAC_PROTO_ETHERNET;
 380	invalidate_flow_key(key);
 381	return 0;
 382}
 383
 384static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
 385		    const struct nshhdr *nh)
 386{
 387	int err;
 388
 389	err = nsh_push(skb, nh);
 390	if (err)
 391		return err;
 392
 393	/* safe right before invalidate_flow_key */
 394	key->mac_proto = MAC_PROTO_NONE;
 395	invalidate_flow_key(key);
 396	return 0;
 397}
 398
 399static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
 400{
 401	int err;
 402
 403	err = nsh_pop(skb);
 404	if (err)
 405		return err;
 406
 407	/* safe right before invalidate_flow_key */
 408	if (skb->protocol == htons(ETH_P_TEB))
 409		key->mac_proto = MAC_PROTO_ETHERNET;
 410	else
 411		key->mac_proto = MAC_PROTO_NONE;
 412	invalidate_flow_key(key);
 413	return 0;
 414}
 415
 416static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
 417				  __be32 addr, __be32 new_addr)
 418{
 419	int transport_len = skb->len - skb_transport_offset(skb);
 420
 421	if (nh->frag_off & htons(IP_OFFSET))
 422		return;
 423
 424	if (nh->protocol == IPPROTO_TCP) {
 425		if (likely(transport_len >= sizeof(struct tcphdr)))
 426			inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 427						 addr, new_addr, true);
 428	} else if (nh->protocol == IPPROTO_UDP) {
 429		if (likely(transport_len >= sizeof(struct udphdr))) {
 430			struct udphdr *uh = udp_hdr(skb);
 431
 432			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 433				inet_proto_csum_replace4(&uh->check, skb,
 434							 addr, new_addr, true);
 435				if (!uh->check)
 436					uh->check = CSUM_MANGLED_0;
 437			}
 438		}
 439	}
 440}
 441
 442static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 443			__be32 *addr, __be32 new_addr)
 444{
 445	update_ip_l4_checksum(skb, nh, *addr, new_addr);
 446	csum_replace4(&nh->check, *addr, new_addr);
 447	skb_clear_hash(skb);
 448	*addr = new_addr;
 449}
 450
 451static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 452				 __be32 addr[4], const __be32 new_addr[4])
 453{
 454	int transport_len = skb->len - skb_transport_offset(skb);
 455
 456	if (l4_proto == NEXTHDR_TCP) {
 457		if (likely(transport_len >= sizeof(struct tcphdr)))
 458			inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 459						  addr, new_addr, true);
 460	} else if (l4_proto == NEXTHDR_UDP) {
 461		if (likely(transport_len >= sizeof(struct udphdr))) {
 462			struct udphdr *uh = udp_hdr(skb);
 463
 464			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 465				inet_proto_csum_replace16(&uh->check, skb,
 466							  addr, new_addr, true);
 467				if (!uh->check)
 468					uh->check = CSUM_MANGLED_0;
 469			}
 470		}
 471	} else if (l4_proto == NEXTHDR_ICMP) {
 472		if (likely(transport_len >= sizeof(struct icmp6hdr)))
 473			inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 474						  skb, addr, new_addr, true);
 475	}
 476}
 477
 478static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
 479			   const __be32 mask[4], __be32 masked[4])
 480{
 481	masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
 482	masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
 483	masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
 484	masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
 485}
 486
 487static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 488			  __be32 addr[4], const __be32 new_addr[4],
 489			  bool recalculate_csum)
 490{
 491	if (recalculate_csum)
 492		update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 493
 494	skb_clear_hash(skb);
 495	memcpy(addr, new_addr, sizeof(__be32[4]));
 496}
 497
 498static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
 499{
 500	/* Bits 21-24 are always unmasked, so this retains their values. */
 501	OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
 502	OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
 503	OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
 504}
 505
 506static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
 507		       u8 mask)
 508{
 509	new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
 510
 511	csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 512	nh->ttl = new_ttl;
 513}
 514
 515static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
 516		    const struct ovs_key_ipv4 *key,
 517		    const struct ovs_key_ipv4 *mask)
 518{
 519	struct iphdr *nh;
 520	__be32 new_addr;
 521	int err;
 522
 523	err = skb_ensure_writable(skb, skb_network_offset(skb) +
 524				  sizeof(struct iphdr));
 525	if (unlikely(err))
 526		return err;
 527
 528	nh = ip_hdr(skb);
 529
 530	/* Setting an IP addresses is typically only a side effect of
 531	 * matching on them in the current userspace implementation, so it
 532	 * makes sense to check if the value actually changed.
 533	 */
 534	if (mask->ipv4_src) {
 535		new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
 536
 537		if (unlikely(new_addr != nh->saddr)) {
 538			set_ip_addr(skb, nh, &nh->saddr, new_addr);
 539			flow_key->ipv4.addr.src = new_addr;
 540		}
 541	}
 542	if (mask->ipv4_dst) {
 543		new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
 544
 545		if (unlikely(new_addr != nh->daddr)) {
 546			set_ip_addr(skb, nh, &nh->daddr, new_addr);
 547			flow_key->ipv4.addr.dst = new_addr;
 548		}
 549	}
 550	if (mask->ipv4_tos) {
 551		ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
 552		flow_key->ip.tos = nh->tos;
 553	}
 554	if (mask->ipv4_ttl) {
 555		set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
 556		flow_key->ip.ttl = nh->ttl;
 557	}
 558
 559	return 0;
 560}
 561
 562static bool is_ipv6_mask_nonzero(const __be32 addr[4])
 563{
 564	return !!(addr[0] | addr[1] | addr[2] | addr[3]);
 565}
 566
 567static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
 568		    const struct ovs_key_ipv6 *key,
 569		    const struct ovs_key_ipv6 *mask)
 570{
 571	struct ipv6hdr *nh;
 572	int err;
 573
 574	err = skb_ensure_writable(skb, skb_network_offset(skb) +
 575				  sizeof(struct ipv6hdr));
 576	if (unlikely(err))
 577		return err;
 578
 579	nh = ipv6_hdr(skb);
 580
 581	/* Setting an IP addresses is typically only a side effect of
 582	 * matching on them in the current userspace implementation, so it
 583	 * makes sense to check if the value actually changed.
 584	 */
 585	if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
 586		__be32 *saddr = (__be32 *)&nh->saddr;
 587		__be32 masked[4];
 588
 589		mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
 590
 591		if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
 592			set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
 593				      true);
 594			memcpy(&flow_key->ipv6.addr.src, masked,
 595			       sizeof(flow_key->ipv6.addr.src));
 596		}
 597	}
 598	if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
 599		unsigned int offset = 0;
 600		int flags = IP6_FH_F_SKIP_RH;
 601		bool recalc_csum = true;
 602		__be32 *daddr = (__be32 *)&nh->daddr;
 603		__be32 masked[4];
 604
 605		mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
 606
 607		if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
 608			if (ipv6_ext_hdr(nh->nexthdr))
 609				recalc_csum = (ipv6_find_hdr(skb, &offset,
 610							     NEXTHDR_ROUTING,
 611							     NULL, &flags)
 612					       != NEXTHDR_ROUTING);
 613
 614			set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
 615				      recalc_csum);
 616			memcpy(&flow_key->ipv6.addr.dst, masked,
 617			       sizeof(flow_key->ipv6.addr.dst));
 618		}
 619	}
 620	if (mask->ipv6_tclass) {
 621		ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
 622		flow_key->ip.tos = ipv6_get_dsfield(nh);
 623	}
 624	if (mask->ipv6_label) {
 625		set_ipv6_fl(nh, ntohl(key->ipv6_label),
 626			    ntohl(mask->ipv6_label));
 627		flow_key->ipv6.label =
 628		    *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 629	}
 630	if (mask->ipv6_hlimit) {
 631		OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
 632			       mask->ipv6_hlimit);
 633		flow_key->ip.ttl = nh->hop_limit;
 634	}
 635	return 0;
 636}
 637
 638static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
 639		   const struct nlattr *a)
 640{
 641	struct nshhdr *nh;
 642	size_t length;
 643	int err;
 644	u8 flags;
 645	u8 ttl;
 646	int i;
 647
 648	struct ovs_key_nsh key;
 649	struct ovs_key_nsh mask;
 650
 651	err = nsh_key_from_nlattr(a, &key, &mask);
 652	if (err)
 653		return err;
 654
 655	/* Make sure the NSH base header is there */
 656	if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
 657		return -ENOMEM;
 658
 659	nh = nsh_hdr(skb);
 660	length = nsh_hdr_len(nh);
 661
 662	/* Make sure the whole NSH header is there */
 663	err = skb_ensure_writable(skb, skb_network_offset(skb) +
 664				       length);
 665	if (unlikely(err))
 666		return err;
 667
 668	nh = nsh_hdr(skb);
 669	skb_postpull_rcsum(skb, nh, length);
 670	flags = nsh_get_flags(nh);
 671	flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
 672	flow_key->nsh.base.flags = flags;
 673	ttl = nsh_get_ttl(nh);
 674	ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
 675	flow_key->nsh.base.ttl = ttl;
 676	nsh_set_flags_and_ttl(nh, flags, ttl);
 677	nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
 678				  mask.base.path_hdr);
 679	flow_key->nsh.base.path_hdr = nh->path_hdr;
 680	switch (nh->mdtype) {
 681	case NSH_M_TYPE1:
 682		for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
 683			nh->md1.context[i] =
 684			    OVS_MASKED(nh->md1.context[i], key.context[i],
 685				       mask.context[i]);
 686		}
 687		memcpy(flow_key->nsh.context, nh->md1.context,
 688		       sizeof(nh->md1.context));
 689		break;
 690	case NSH_M_TYPE2:
 691		memset(flow_key->nsh.context, 0,
 692		       sizeof(flow_key->nsh.context));
 693		break;
 694	default:
 695		return -EINVAL;
 696	}
 697	skb_postpush_rcsum(skb, nh, length);
 698	return 0;
 699}
 700
 701/* Must follow skb_ensure_writable() since that can move the skb data. */
 702static void set_tp_port(struct sk_buff *skb, __be16 *port,
 703			__be16 new_port, __sum16 *check)
 704{
 705	inet_proto_csum_replace2(check, skb, *port, new_port, false);
 706	*port = new_port;
 707}
 708
 709static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 710		   const struct ovs_key_udp *key,
 711		   const struct ovs_key_udp *mask)
 712{
 713	struct udphdr *uh;
 714	__be16 src, dst;
 715	int err;
 716
 717	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 718				  sizeof(struct udphdr));
 719	if (unlikely(err))
 720		return err;
 721
 722	uh = udp_hdr(skb);
 723	/* Either of the masks is non-zero, so do not bother checking them. */
 724	src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
 725	dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
 726
 727	if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 728		if (likely(src != uh->source)) {
 729			set_tp_port(skb, &uh->source, src, &uh->check);
 730			flow_key->tp.src = src;
 731		}
 732		if (likely(dst != uh->dest)) {
 733			set_tp_port(skb, &uh->dest, dst, &uh->check);
 734			flow_key->tp.dst = dst;
 735		}
 736
 737		if (unlikely(!uh->check))
 738			uh->check = CSUM_MANGLED_0;
 739	} else {
 740		uh->source = src;
 741		uh->dest = dst;
 742		flow_key->tp.src = src;
 743		flow_key->tp.dst = dst;
 744	}
 745
 746	skb_clear_hash(skb);
 747
 748	return 0;
 749}
 750
 751static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 752		   const struct ovs_key_tcp *key,
 753		   const struct ovs_key_tcp *mask)
 754{
 755	struct tcphdr *th;
 756	__be16 src, dst;
 757	int err;
 758
 759	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 760				  sizeof(struct tcphdr));
 761	if (unlikely(err))
 762		return err;
 763
 764	th = tcp_hdr(skb);
 765	src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
 766	if (likely(src != th->source)) {
 767		set_tp_port(skb, &th->source, src, &th->check);
 768		flow_key->tp.src = src;
 769	}
 770	dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
 771	if (likely(dst != th->dest)) {
 772		set_tp_port(skb, &th->dest, dst, &th->check);
 773		flow_key->tp.dst = dst;
 774	}
 775	skb_clear_hash(skb);
 776
 777	return 0;
 778}
 779
 780static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 781		    const struct ovs_key_sctp *key,
 782		    const struct ovs_key_sctp *mask)
 783{
 784	unsigned int sctphoff = skb_transport_offset(skb);
 785	struct sctphdr *sh;
 786	__le32 old_correct_csum, new_csum, old_csum;
 787	int err;
 788
 789	err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
 790	if (unlikely(err))
 791		return err;
 792
 793	sh = sctp_hdr(skb);
 794	old_csum = sh->checksum;
 795	old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 796
 797	sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
 798	sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
 799
 800	new_csum = sctp_compute_cksum(skb, sctphoff);
 801
 802	/* Carry any checksum errors through. */
 803	sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 804
 805	skb_clear_hash(skb);
 806	flow_key->tp.src = sh->source;
 807	flow_key->tp.dst = sh->dest;
 808
 809	return 0;
 810}
 811
 812static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 
 813{
 814	struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
 815	struct vport *vport = data->vport;
 816
 817	if (skb_cow_head(skb, data->l2_len) < 0) {
 818		kfree_skb(skb);
 819		return -ENOMEM;
 820	}
 821
 822	__skb_dst_copy(skb, data->dst);
 823	*OVS_CB(skb) = data->cb;
 824	skb->inner_protocol = data->inner_protocol;
 825	skb->vlan_tci = data->vlan_tci;
 826	skb->vlan_proto = data->vlan_proto;
 
 
 827
 828	/* Reconstruct the MAC header.  */
 829	skb_push(skb, data->l2_len);
 830	memcpy(skb->data, &data->l2_data, data->l2_len);
 831	skb_postpush_rcsum(skb, skb->data, data->l2_len);
 832	skb_reset_mac_header(skb);
 833
 834	if (eth_p_mpls(skb->protocol)) {
 835		skb->inner_network_header = skb->network_header;
 836		skb_set_network_header(skb, data->network_offset);
 837		skb_reset_mac_len(skb);
 838	}
 839
 840	ovs_vport_send(vport, skb, data->mac_proto);
 841	return 0;
 842}
 843
 844static unsigned int
 845ovs_dst_get_mtu(const struct dst_entry *dst)
 846{
 847	return dst->dev->mtu;
 848}
 849
 850static struct dst_ops ovs_dst_ops = {
 851	.family = AF_UNSPEC,
 852	.mtu = ovs_dst_get_mtu,
 853};
 854
 855/* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
 856 * ovs_vport_output(), which is called once per fragmented packet.
 857 */
 858static void prepare_frag(struct vport *vport, struct sk_buff *skb,
 859			 u16 orig_network_offset, u8 mac_proto)
 860{
 861	unsigned int hlen = skb_network_offset(skb);
 862	struct ovs_frag_data *data;
 863
 864	data = this_cpu_ptr(&ovs_frag_data_storage);
 865	data->dst = skb->_skb_refdst;
 866	data->vport = vport;
 867	data->cb = *OVS_CB(skb);
 868	data->inner_protocol = skb->inner_protocol;
 869	data->network_offset = orig_network_offset;
 870	data->vlan_tci = skb->vlan_tci;
 
 
 
 871	data->vlan_proto = skb->vlan_proto;
 872	data->mac_proto = mac_proto;
 873	data->l2_len = hlen;
 874	memcpy(&data->l2_data, skb->data, hlen);
 875
 876	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 877	skb_pull(skb, hlen);
 878}
 879
 880static void ovs_fragment(struct net *net, struct vport *vport,
 881			 struct sk_buff *skb, u16 mru,
 882			 struct sw_flow_key *key)
 883{
 884	u16 orig_network_offset = 0;
 885
 886	if (eth_p_mpls(skb->protocol)) {
 887		orig_network_offset = skb_network_offset(skb);
 888		skb->network_header = skb->inner_network_header;
 889	}
 890
 891	if (skb_network_offset(skb) > MAX_L2_LEN) {
 892		OVS_NLERR(1, "L2 header too long to fragment");
 893		goto err;
 894	}
 895
 896	if (key->eth.type == htons(ETH_P_IP)) {
 897		struct dst_entry ovs_dst;
 898		unsigned long orig_dst;
 899
 900		prepare_frag(vport, skb, orig_network_offset,
 901			     ovs_key_mac_proto(key));
 902		dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
 903			 DST_OBSOLETE_NONE, DST_NOCOUNT);
 904		ovs_dst.dev = vport->dev;
 905
 906		orig_dst = skb->_skb_refdst;
 907		skb_dst_set_noref(skb, &ovs_dst);
 908		IPCB(skb)->frag_max_size = mru;
 909
 910		ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
 911		refdst_drop(orig_dst);
 912	} else if (key->eth.type == htons(ETH_P_IPV6)) {
 913		const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
 914		unsigned long orig_dst;
 915		struct rt6_info ovs_rt;
 916
 917		if (!v6ops)
 918			goto err;
 919
 920		prepare_frag(vport, skb, orig_network_offset,
 921			     ovs_key_mac_proto(key));
 922		memset(&ovs_rt, 0, sizeof(ovs_rt));
 923		dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
 924			 DST_OBSOLETE_NONE, DST_NOCOUNT);
 925		ovs_rt.dst.dev = vport->dev;
 926
 927		orig_dst = skb->_skb_refdst;
 928		skb_dst_set_noref(skb, &ovs_rt.dst);
 929		IP6CB(skb)->frag_max_size = mru;
 930
 931		v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
 932		refdst_drop(orig_dst);
 933	} else {
 934		WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
 935			  ovs_vport_name(vport), ntohs(key->eth.type), mru,
 936			  vport->dev->mtu);
 937		goto err;
 938	}
 939
 940	return;
 941err:
 942	kfree_skb(skb);
 943}
 944
 945static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
 946		      struct sw_flow_key *key)
 947{
 948	struct vport *vport = ovs_vport_rcu(dp, out_port);
 949
 950	if (likely(vport)) {
 951		u16 mru = OVS_CB(skb)->mru;
 952		u32 cutlen = OVS_CB(skb)->cutlen;
 953
 954		if (unlikely(cutlen > 0)) {
 955			if (skb->len - cutlen > ovs_mac_header_len(key))
 956				pskb_trim(skb, skb->len - cutlen);
 957			else
 958				pskb_trim(skb, ovs_mac_header_len(key));
 959		}
 960
 961		if (likely(!mru ||
 962		           (skb->len <= mru + vport->dev->hard_header_len))) {
 963			ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
 964		} else if (mru <= vport->dev->mtu) {
 965			struct net *net = read_pnet(&dp->net);
 966
 967			ovs_fragment(net, vport, skb, mru, key);
 968		} else {
 969			kfree_skb(skb);
 970		}
 971	} else {
 972		kfree_skb(skb);
 973	}
 974}
 975
 976static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 977			    struct sw_flow_key *key, const struct nlattr *attr,
 978			    const struct nlattr *actions, int actions_len,
 979			    uint32_t cutlen)
 980{
 981	struct dp_upcall_info upcall;
 982	const struct nlattr *a;
 983	int rem;
 984
 985	memset(&upcall, 0, sizeof(upcall));
 986	upcall.cmd = OVS_PACKET_CMD_ACTION;
 987	upcall.mru = OVS_CB(skb)->mru;
 988
 989	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 990		 a = nla_next(a, &rem)) {
 991		switch (nla_type(a)) {
 992		case OVS_USERSPACE_ATTR_USERDATA:
 993			upcall.userdata = a;
 994			break;
 995
 996		case OVS_USERSPACE_ATTR_PID:
 997			upcall.portid = nla_get_u32(a);
 998			break;
 999
1000		case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
1001			/* Get out tunnel info. */
1002			struct vport *vport;
1003
1004			vport = ovs_vport_rcu(dp, nla_get_u32(a));
1005			if (vport) {
1006				int err;
1007
1008				err = dev_fill_metadata_dst(vport->dev, skb);
1009				if (!err)
1010					upcall.egress_tun_info = skb_tunnel_info(skb);
1011			}
1012
1013			break;
1014		}
1015
1016		case OVS_USERSPACE_ATTR_ACTIONS: {
1017			/* Include actions. */
1018			upcall.actions = actions;
1019			upcall.actions_len = actions_len;
1020			break;
1021		}
1022
1023		} /* End of switch. */
1024	}
1025
1026	return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
1027}
1028
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1029/* When 'last' is true, sample() should always consume the 'skb'.
1030 * Otherwise, sample() should keep 'skb' intact regardless what
1031 * actions are executed within sample().
1032 */
1033static int sample(struct datapath *dp, struct sk_buff *skb,
1034		  struct sw_flow_key *key, const struct nlattr *attr,
1035		  bool last)
1036{
1037	struct nlattr *actions;
1038	struct nlattr *sample_arg;
1039	int rem = nla_len(attr);
1040	const struct sample_arg *arg;
1041	bool clone_flow_key;
1042
1043	/* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
1044	sample_arg = nla_data(attr);
1045	arg = nla_data(sample_arg);
1046	actions = nla_next(sample_arg, &rem);
1047
1048	if ((arg->probability != U32_MAX) &&
1049	    (!arg->probability || prandom_u32() > arg->probability)) {
1050		if (last)
1051			consume_skb(skb);
1052		return 0;
1053	}
1054
1055	clone_flow_key = !arg->exec;
1056	return clone_execute(dp, skb, key, 0, actions, rem, last,
1057			     clone_flow_key);
1058}
1059
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1060static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1061			 const struct nlattr *attr)
1062{
1063	struct ovs_action_hash *hash_act = nla_data(attr);
1064	u32 hash = 0;
1065
1066	/* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
1067	hash = skb_get_hash(skb);
1068	hash = jhash_1word(hash, hash_act->hash_basis);
1069	if (!hash)
1070		hash = 0x1;
1071
1072	key->ovs_flow_hash = hash;
1073}
1074
1075static int execute_set_action(struct sk_buff *skb,
1076			      struct sw_flow_key *flow_key,
1077			      const struct nlattr *a)
1078{
1079	/* Only tunnel set execution is supported without a mask. */
1080	if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1081		struct ovs_tunnel_info *tun = nla_data(a);
1082
1083		skb_dst_drop(skb);
1084		dst_hold((struct dst_entry *)tun->tun_dst);
1085		skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1086		return 0;
1087	}
1088
1089	return -EINVAL;
1090}
1091
1092/* Mask is at the midpoint of the data. */
1093#define get_mask(a, type) ((const type)nla_data(a) + 1)
1094
1095static int execute_masked_set_action(struct sk_buff *skb,
1096				     struct sw_flow_key *flow_key,
1097				     const struct nlattr *a)
1098{
1099	int err = 0;
1100
1101	switch (nla_type(a)) {
1102	case OVS_KEY_ATTR_PRIORITY:
1103		OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1104			       *get_mask(a, u32 *));
1105		flow_key->phy.priority = skb->priority;
1106		break;
1107
1108	case OVS_KEY_ATTR_SKB_MARK:
1109		OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1110		flow_key->phy.skb_mark = skb->mark;
1111		break;
1112
1113	case OVS_KEY_ATTR_TUNNEL_INFO:
1114		/* Masked data not supported for tunnel. */
1115		err = -EINVAL;
1116		break;
1117
1118	case OVS_KEY_ATTR_ETHERNET:
1119		err = set_eth_addr(skb, flow_key, nla_data(a),
1120				   get_mask(a, struct ovs_key_ethernet *));
1121		break;
1122
1123	case OVS_KEY_ATTR_NSH:
1124		err = set_nsh(skb, flow_key, a);
1125		break;
1126
1127	case OVS_KEY_ATTR_IPV4:
1128		err = set_ipv4(skb, flow_key, nla_data(a),
1129			       get_mask(a, struct ovs_key_ipv4 *));
1130		break;
1131
1132	case OVS_KEY_ATTR_IPV6:
1133		err = set_ipv6(skb, flow_key, nla_data(a),
1134			       get_mask(a, struct ovs_key_ipv6 *));
1135		break;
1136
1137	case OVS_KEY_ATTR_TCP:
1138		err = set_tcp(skb, flow_key, nla_data(a),
1139			      get_mask(a, struct ovs_key_tcp *));
1140		break;
1141
1142	case OVS_KEY_ATTR_UDP:
1143		err = set_udp(skb, flow_key, nla_data(a),
1144			      get_mask(a, struct ovs_key_udp *));
1145		break;
1146
1147	case OVS_KEY_ATTR_SCTP:
1148		err = set_sctp(skb, flow_key, nla_data(a),
1149			       get_mask(a, struct ovs_key_sctp *));
1150		break;
1151
1152	case OVS_KEY_ATTR_MPLS:
1153		err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1154								    __be32 *));
1155		break;
1156
1157	case OVS_KEY_ATTR_CT_STATE:
1158	case OVS_KEY_ATTR_CT_ZONE:
1159	case OVS_KEY_ATTR_CT_MARK:
1160	case OVS_KEY_ATTR_CT_LABELS:
1161	case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1162	case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1163		err = -EINVAL;
1164		break;
1165	}
1166
1167	return err;
1168}
1169
1170static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1171			  struct sw_flow_key *key,
1172			  const struct nlattr *a, bool last)
1173{
1174	u32 recirc_id;
1175
1176	if (!is_flow_key_valid(key)) {
1177		int err;
1178
1179		err = ovs_flow_key_update(skb, key);
1180		if (err)
1181			return err;
1182	}
1183	BUG_ON(!is_flow_key_valid(key));
1184
1185	recirc_id = nla_get_u32(a);
1186	return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1187}
1188
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1189/* Execute a list of actions against 'skb'. */
1190static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1191			      struct sw_flow_key *key,
1192			      const struct nlattr *attr, int len)
1193{
1194	const struct nlattr *a;
1195	int rem;
1196
1197	for (a = attr, rem = len; rem > 0;
1198	     a = nla_next(a, &rem)) {
1199		int err = 0;
1200
 
 
 
1201		switch (nla_type(a)) {
1202		case OVS_ACTION_ATTR_OUTPUT: {
1203			int port = nla_get_u32(a);
1204			struct sk_buff *clone;
1205
1206			/* Every output action needs a separate clone
1207			 * of 'skb', In case the output action is the
1208			 * last action, cloning can be avoided.
1209			 */
1210			if (nla_is_last(a, rem)) {
1211				do_output(dp, skb, port, key);
1212				/* 'skb' has been used for output.
1213				 */
1214				return 0;
1215			}
1216
1217			clone = skb_clone(skb, GFP_ATOMIC);
1218			if (clone)
1219				do_output(dp, clone, port, key);
1220			OVS_CB(skb)->cutlen = 0;
1221			break;
1222		}
1223
1224		case OVS_ACTION_ATTR_TRUNC: {
1225			struct ovs_action_trunc *trunc = nla_data(a);
1226
1227			if (skb->len > trunc->max_len)
1228				OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1229			break;
1230		}
1231
1232		case OVS_ACTION_ATTR_USERSPACE:
1233			output_userspace(dp, skb, key, a, attr,
1234						     len, OVS_CB(skb)->cutlen);
1235			OVS_CB(skb)->cutlen = 0;
1236			break;
1237
1238		case OVS_ACTION_ATTR_HASH:
1239			execute_hash(skb, key, a);
1240			break;
1241
1242		case OVS_ACTION_ATTR_PUSH_MPLS:
1243			err = push_mpls(skb, key, nla_data(a));
 
 
 
1244			break;
 
 
 
 
1245
 
 
 
 
 
 
 
1246		case OVS_ACTION_ATTR_POP_MPLS:
1247			err = pop_mpls(skb, key, nla_get_be16(a));
1248			break;
1249
1250		case OVS_ACTION_ATTR_PUSH_VLAN:
1251			err = push_vlan(skb, key, nla_data(a));
1252			break;
1253
1254		case OVS_ACTION_ATTR_POP_VLAN:
1255			err = pop_vlan(skb, key);
1256			break;
1257
1258		case OVS_ACTION_ATTR_RECIRC: {
1259			bool last = nla_is_last(a, rem);
1260
1261			err = execute_recirc(dp, skb, key, a, last);
1262			if (last) {
1263				/* If this is the last action, the skb has
1264				 * been consumed or freed.
1265				 * Return immediately.
1266				 */
1267				return err;
1268			}
1269			break;
1270		}
1271
1272		case OVS_ACTION_ATTR_SET:
1273			err = execute_set_action(skb, key, nla_data(a));
1274			break;
1275
1276		case OVS_ACTION_ATTR_SET_MASKED:
1277		case OVS_ACTION_ATTR_SET_TO_MASKED:
1278			err = execute_masked_set_action(skb, key, nla_data(a));
1279			break;
1280
1281		case OVS_ACTION_ATTR_SAMPLE: {
1282			bool last = nla_is_last(a, rem);
1283
1284			err = sample(dp, skb, key, a, last);
1285			if (last)
1286				return err;
1287
1288			break;
1289		}
1290
1291		case OVS_ACTION_ATTR_CT:
1292			if (!is_flow_key_valid(key)) {
1293				err = ovs_flow_key_update(skb, key);
1294				if (err)
1295					return err;
1296			}
1297
1298			err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1299					     nla_data(a));
1300
1301			/* Hide stolen IP fragments from user space. */
1302			if (err)
1303				return err == -EINPROGRESS ? 0 : err;
1304			break;
1305
1306		case OVS_ACTION_ATTR_CT_CLEAR:
1307			err = ovs_ct_clear(skb, key);
1308			break;
1309
1310		case OVS_ACTION_ATTR_PUSH_ETH:
1311			err = push_eth(skb, key, nla_data(a));
1312			break;
1313
1314		case OVS_ACTION_ATTR_POP_ETH:
1315			err = pop_eth(skb, key);
1316			break;
1317
1318		case OVS_ACTION_ATTR_PUSH_NSH: {
1319			u8 buffer[NSH_HDR_MAX_LEN];
1320			struct nshhdr *nh = (struct nshhdr *)buffer;
1321
1322			err = nsh_hdr_from_nlattr(nla_data(a), nh,
1323						  NSH_HDR_MAX_LEN);
1324			if (unlikely(err))
1325				break;
1326			err = push_nsh(skb, key, nh);
1327			break;
1328		}
1329
1330		case OVS_ACTION_ATTR_POP_NSH:
1331			err = pop_nsh(skb, key);
1332			break;
1333
1334		case OVS_ACTION_ATTR_METER:
1335			if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1336				consume_skb(skb);
1337				return 0;
1338			}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1339		}
1340
1341		if (unlikely(err)) {
1342			kfree_skb(skb);
1343			return err;
1344		}
1345	}
1346
1347	consume_skb(skb);
1348	return 0;
1349}
1350
1351/* Execute the actions on the clone of the packet. The effect of the
1352 * execution does not affect the original 'skb' nor the original 'key'.
1353 *
1354 * The execution may be deferred in case the actions can not be executed
1355 * immediately.
1356 */
1357static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1358			 struct sw_flow_key *key, u32 recirc_id,
1359			 const struct nlattr *actions, int len,
1360			 bool last, bool clone_flow_key)
1361{
1362	struct deferred_action *da;
1363	struct sw_flow_key *clone;
1364
1365	skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1366	if (!skb) {
1367		/* Out of memory, skip this action.
1368		 */
1369		return 0;
1370	}
1371
1372	/* When clone_flow_key is false, the 'key' will not be change
1373	 * by the actions, then the 'key' can be used directly.
1374	 * Otherwise, try to clone key from the next recursion level of
1375	 * 'flow_keys'. If clone is successful, execute the actions
1376	 * without deferring.
1377	 */
1378	clone = clone_flow_key ? clone_key(key) : key;
1379	if (clone) {
1380		int err = 0;
1381
1382		if (actions) { /* Sample action */
1383			if (clone_flow_key)
1384				__this_cpu_inc(exec_actions_level);
1385
1386			err = do_execute_actions(dp, skb, clone,
1387						 actions, len);
1388
1389			if (clone_flow_key)
1390				__this_cpu_dec(exec_actions_level);
1391		} else { /* Recirc action */
1392			clone->recirc_id = recirc_id;
1393			ovs_dp_process_packet(skb, clone);
1394		}
1395		return err;
1396	}
1397
1398	/* Out of 'flow_keys' space. Defer actions */
1399	da = add_deferred_actions(skb, key, actions, len);
1400	if (da) {
1401		if (!actions) { /* Recirc action */
1402			key = &da->pkt_key;
1403			key->recirc_id = recirc_id;
1404		}
1405	} else {
1406		/* Out of per CPU action FIFO space. Drop the 'skb' and
1407		 * log an error.
1408		 */
1409		kfree_skb(skb);
1410
1411		if (net_ratelimit()) {
1412			if (actions) { /* Sample action */
1413				pr_warn("%s: deferred action limit reached, drop sample action\n",
1414					ovs_dp_name(dp));
1415			} else {  /* Recirc action */
1416				pr_warn("%s: deferred action limit reached, drop recirc action\n",
1417					ovs_dp_name(dp));
1418			}
1419		}
1420	}
1421	return 0;
1422}
1423
1424static void process_deferred_actions(struct datapath *dp)
1425{
1426	struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1427
1428	/* Do not touch the FIFO in case there is no deferred actions. */
1429	if (action_fifo_is_empty(fifo))
1430		return;
1431
1432	/* Finishing executing all deferred actions. */
1433	do {
1434		struct deferred_action *da = action_fifo_get(fifo);
1435		struct sk_buff *skb = da->skb;
1436		struct sw_flow_key *key = &da->pkt_key;
1437		const struct nlattr *actions = da->actions;
1438		int actions_len = da->actions_len;
1439
1440		if (actions)
1441			do_execute_actions(dp, skb, key, actions, actions_len);
1442		else
1443			ovs_dp_process_packet(skb, key);
1444	} while (!action_fifo_is_empty(fifo));
1445
1446	/* Reset FIFO for the next packet.  */
1447	action_fifo_init(fifo);
1448}
1449
1450/* Execute a list of actions against 'skb'. */
1451int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1452			const struct sw_flow_actions *acts,
1453			struct sw_flow_key *key)
1454{
1455	int err, level;
1456
1457	level = __this_cpu_inc_return(exec_actions_level);
1458	if (unlikely(level > OVS_RECURSION_LIMIT)) {
1459		net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1460				     ovs_dp_name(dp));
1461		kfree_skb(skb);
1462		err = -ENETDOWN;
1463		goto out;
1464	}
1465
1466	OVS_CB(skb)->acts_origlen = acts->orig_len;
1467	err = do_execute_actions(dp, skb, key,
1468				 acts->actions, acts->actions_len);
1469
1470	if (level == 1)
1471		process_deferred_actions(dp);
1472
1473out:
1474	__this_cpu_dec(exec_actions_level);
1475	return err;
1476}
1477
1478int action_fifos_init(void)
1479{
1480	action_fifos = alloc_percpu(struct action_fifo);
1481	if (!action_fifos)
1482		return -ENOMEM;
1483
1484	flow_keys = alloc_percpu(struct action_flow_keys);
1485	if (!flow_keys) {
1486		free_percpu(action_fifos);
1487		return -ENOMEM;
1488	}
1489
1490	return 0;
1491}
1492
1493void action_fifos_exit(void)
1494{
1495	free_percpu(action_fifos);
1496	free_percpu(flow_keys);
1497}