Linux Audio

Check our new training course

Loading...
v5.4
   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/netfilter_ipv6.h>
  13#include <linux/sctp.h>
  14#include <linux/tcp.h>
  15#include <linux/udp.h>
  16#include <linux/in6.h>
  17#include <linux/if_arp.h>
  18#include <linux/if_vlan.h>
  19
  20#include <net/dst.h>
  21#include <net/ip.h>
  22#include <net/ipv6.h>
  23#include <net/ip6_fib.h>
  24#include <net/checksum.h>
  25#include <net/dsfield.h>
  26#include <net/mpls.h>
  27#include <net/sctp/checksum.h>
  28
  29#include "datapath.h"
  30#include "flow.h"
  31#include "conntrack.h"
  32#include "vport.h"
  33#include "flow_netlink.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		     const struct ovs_action_push_mpls *mpls)
 165{
 166	int err;
 167
 168	err = skb_mpls_push(skb, mpls->mpls_lse, mpls->mpls_ethertype,
 169			    skb->mac_len);
 170	if (err)
 171		return err;
 172
 
 
 
 173	invalidate_flow_key(key);
 174	return 0;
 175}
 176
 177static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
 178		    const __be16 ethertype)
 179{
 180	int err;
 181
 182	err = skb_mpls_pop(skb, ethertype, skb->mac_len);
 
 183	if (err)
 184		return err;
 185
 
 
 
 186	invalidate_flow_key(key);
 187	return 0;
 188}
 189
 190static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
 191		    const __be32 *mpls_lse, const __be32 *mask)
 192{
 193	struct mpls_shim_hdr *stack;
 194	__be32 lse;
 195	int err;
 196
 197	stack = mpls_hdr(skb);
 198	lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
 199	err = skb_mpls_update_lse(skb, lse);
 200	if (err)
 201		return err;
 202
 203	flow_key->mpls.top_lse = lse;
 204	return 0;
 205}
 206
 207static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 208{
 209	int err;
 210
 211	err = skb_vlan_pop(skb);
 212	if (skb_vlan_tag_present(skb)) {
 213		invalidate_flow_key(key);
 214	} else {
 215		key->eth.vlan.tci = 0;
 216		key->eth.vlan.tpid = 0;
 217	}
 218	return err;
 219}
 220
 221static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
 222		     const struct ovs_action_push_vlan *vlan)
 223{
 224	if (skb_vlan_tag_present(skb)) {
 225		invalidate_flow_key(key);
 226	} else {
 227		key->eth.vlan.tci = vlan->vlan_tci;
 228		key->eth.vlan.tpid = vlan->vlan_tpid;
 229	}
 230	return skb_vlan_push(skb, vlan->vlan_tpid,
 231			     ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
 232}
 233
 234/* 'src' is already properly masked. */
 235static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
 236{
 237	u16 *dst = (u16 *)dst_;
 238	const u16 *src = (const u16 *)src_;
 239	const u16 *mask = (const u16 *)mask_;
 240
 241	OVS_SET_MASKED(dst[0], src[0], mask[0]);
 242	OVS_SET_MASKED(dst[1], src[1], mask[1]);
 243	OVS_SET_MASKED(dst[2], src[2], mask[2]);
 244}
 245
 246static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
 247			const struct ovs_key_ethernet *key,
 248			const struct ovs_key_ethernet *mask)
 249{
 250	int err;
 251
 252	err = skb_ensure_writable(skb, ETH_HLEN);
 253	if (unlikely(err))
 254		return err;
 255
 256	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 257
 258	ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
 259			       mask->eth_src);
 260	ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
 261			       mask->eth_dst);
 262
 263	skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 264
 265	ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
 266	ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
 267	return 0;
 268}
 269
 270/* pop_eth does not support VLAN packets as this action is never called
 271 * for them.
 272 */
 273static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
 274{
 275	skb_pull_rcsum(skb, ETH_HLEN);
 276	skb_reset_mac_header(skb);
 277	skb_reset_mac_len(skb);
 278
 279	/* safe right before invalidate_flow_key */
 280	key->mac_proto = MAC_PROTO_NONE;
 281	invalidate_flow_key(key);
 282	return 0;
 283}
 284
 285static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
 286		    const struct ovs_action_push_eth *ethh)
 287{
 288	struct ethhdr *hdr;
 289
 290	/* Add the new Ethernet header */
 291	if (skb_cow_head(skb, ETH_HLEN) < 0)
 292		return -ENOMEM;
 293
 294	skb_push(skb, ETH_HLEN);
 295	skb_reset_mac_header(skb);
 296	skb_reset_mac_len(skb);
 297
 298	hdr = eth_hdr(skb);
 299	ether_addr_copy(hdr->h_source, ethh->addresses.eth_src);
 300	ether_addr_copy(hdr->h_dest, ethh->addresses.eth_dst);
 301	hdr->h_proto = skb->protocol;
 302
 303	skb_postpush_rcsum(skb, hdr, ETH_HLEN);
 304
 305	/* safe right before invalidate_flow_key */
 306	key->mac_proto = MAC_PROTO_ETHERNET;
 307	invalidate_flow_key(key);
 308	return 0;
 309}
 310
 311static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
 312		    const struct nshhdr *nh)
 313{
 314	int err;
 315
 316	err = nsh_push(skb, nh);
 317	if (err)
 318		return err;
 319
 320	/* safe right before invalidate_flow_key */
 321	key->mac_proto = MAC_PROTO_NONE;
 322	invalidate_flow_key(key);
 323	return 0;
 324}
 325
 326static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
 327{
 328	int err;
 329
 330	err = nsh_pop(skb);
 331	if (err)
 332		return err;
 333
 334	/* safe right before invalidate_flow_key */
 335	if (skb->protocol == htons(ETH_P_TEB))
 336		key->mac_proto = MAC_PROTO_ETHERNET;
 337	else
 338		key->mac_proto = MAC_PROTO_NONE;
 339	invalidate_flow_key(key);
 340	return 0;
 341}
 342
 343static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
 344				  __be32 addr, __be32 new_addr)
 345{
 346	int transport_len = skb->len - skb_transport_offset(skb);
 347
 348	if (nh->frag_off & htons(IP_OFFSET))
 349		return;
 350
 351	if (nh->protocol == IPPROTO_TCP) {
 352		if (likely(transport_len >= sizeof(struct tcphdr)))
 353			inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 354						 addr, new_addr, true);
 355	} else if (nh->protocol == IPPROTO_UDP) {
 356		if (likely(transport_len >= sizeof(struct udphdr))) {
 357			struct udphdr *uh = udp_hdr(skb);
 358
 359			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 360				inet_proto_csum_replace4(&uh->check, skb,
 361							 addr, new_addr, true);
 362				if (!uh->check)
 363					uh->check = CSUM_MANGLED_0;
 364			}
 365		}
 366	}
 367}
 368
 369static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 370			__be32 *addr, __be32 new_addr)
 371{
 372	update_ip_l4_checksum(skb, nh, *addr, new_addr);
 373	csum_replace4(&nh->check, *addr, new_addr);
 374	skb_clear_hash(skb);
 375	*addr = new_addr;
 376}
 377
 378static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 379				 __be32 addr[4], const __be32 new_addr[4])
 380{
 381	int transport_len = skb->len - skb_transport_offset(skb);
 382
 383	if (l4_proto == NEXTHDR_TCP) {
 384		if (likely(transport_len >= sizeof(struct tcphdr)))
 385			inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 386						  addr, new_addr, true);
 387	} else if (l4_proto == NEXTHDR_UDP) {
 388		if (likely(transport_len >= sizeof(struct udphdr))) {
 389			struct udphdr *uh = udp_hdr(skb);
 390
 391			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 392				inet_proto_csum_replace16(&uh->check, skb,
 393							  addr, new_addr, true);
 394				if (!uh->check)
 395					uh->check = CSUM_MANGLED_0;
 396			}
 397		}
 398	} else if (l4_proto == NEXTHDR_ICMP) {
 399		if (likely(transport_len >= sizeof(struct icmp6hdr)))
 400			inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 401						  skb, addr, new_addr, true);
 402	}
 403}
 404
 405static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
 406			   const __be32 mask[4], __be32 masked[4])
 407{
 408	masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
 409	masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
 410	masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
 411	masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
 412}
 413
 414static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 415			  __be32 addr[4], const __be32 new_addr[4],
 416			  bool recalculate_csum)
 417{
 418	if (recalculate_csum)
 419		update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 420
 421	skb_clear_hash(skb);
 422	memcpy(addr, new_addr, sizeof(__be32[4]));
 423}
 424
 425static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
 426{
 427	/* Bits 21-24 are always unmasked, so this retains their values. */
 428	OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
 429	OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
 430	OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
 431}
 432
 433static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
 434		       u8 mask)
 435{
 436	new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
 437
 438	csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 439	nh->ttl = new_ttl;
 440}
 441
 442static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
 443		    const struct ovs_key_ipv4 *key,
 444		    const struct ovs_key_ipv4 *mask)
 445{
 446	struct iphdr *nh;
 447	__be32 new_addr;
 448	int err;
 449
 450	err = skb_ensure_writable(skb, skb_network_offset(skb) +
 451				  sizeof(struct iphdr));
 452	if (unlikely(err))
 453		return err;
 454
 455	nh = ip_hdr(skb);
 456
 457	/* Setting an IP addresses is typically only a side effect of
 458	 * matching on them in the current userspace implementation, so it
 459	 * makes sense to check if the value actually changed.
 460	 */
 461	if (mask->ipv4_src) {
 462		new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
 463
 464		if (unlikely(new_addr != nh->saddr)) {
 465			set_ip_addr(skb, nh, &nh->saddr, new_addr);
 466			flow_key->ipv4.addr.src = new_addr;
 467		}
 468	}
 469	if (mask->ipv4_dst) {
 470		new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
 471
 472		if (unlikely(new_addr != nh->daddr)) {
 473			set_ip_addr(skb, nh, &nh->daddr, new_addr);
 474			flow_key->ipv4.addr.dst = new_addr;
 475		}
 476	}
 477	if (mask->ipv4_tos) {
 478		ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
 479		flow_key->ip.tos = nh->tos;
 480	}
 481	if (mask->ipv4_ttl) {
 482		set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
 483		flow_key->ip.ttl = nh->ttl;
 484	}
 485
 486	return 0;
 487}
 488
 489static bool is_ipv6_mask_nonzero(const __be32 addr[4])
 490{
 491	return !!(addr[0] | addr[1] | addr[2] | addr[3]);
 492}
 493
 494static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
 495		    const struct ovs_key_ipv6 *key,
 496		    const struct ovs_key_ipv6 *mask)
 497{
 498	struct ipv6hdr *nh;
 499	int err;
 500
 501	err = skb_ensure_writable(skb, skb_network_offset(skb) +
 502				  sizeof(struct ipv6hdr));
 503	if (unlikely(err))
 504		return err;
 505
 506	nh = ipv6_hdr(skb);
 507
 508	/* Setting an IP addresses is typically only a side effect of
 509	 * matching on them in the current userspace implementation, so it
 510	 * makes sense to check if the value actually changed.
 511	 */
 512	if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
 513		__be32 *saddr = (__be32 *)&nh->saddr;
 514		__be32 masked[4];
 515
 516		mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
 517
 518		if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
 519			set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
 520				      true);
 521			memcpy(&flow_key->ipv6.addr.src, masked,
 522			       sizeof(flow_key->ipv6.addr.src));
 523		}
 524	}
 525	if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
 526		unsigned int offset = 0;
 527		int flags = IP6_FH_F_SKIP_RH;
 528		bool recalc_csum = true;
 529		__be32 *daddr = (__be32 *)&nh->daddr;
 530		__be32 masked[4];
 531
 532		mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
 533
 534		if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
 535			if (ipv6_ext_hdr(nh->nexthdr))
 536				recalc_csum = (ipv6_find_hdr(skb, &offset,
 537							     NEXTHDR_ROUTING,
 538							     NULL, &flags)
 539					       != NEXTHDR_ROUTING);
 540
 541			set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
 542				      recalc_csum);
 543			memcpy(&flow_key->ipv6.addr.dst, masked,
 544			       sizeof(flow_key->ipv6.addr.dst));
 545		}
 546	}
 547	if (mask->ipv6_tclass) {
 548		ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
 549		flow_key->ip.tos = ipv6_get_dsfield(nh);
 550	}
 551	if (mask->ipv6_label) {
 552		set_ipv6_fl(nh, ntohl(key->ipv6_label),
 553			    ntohl(mask->ipv6_label));
 554		flow_key->ipv6.label =
 555		    *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 556	}
 557	if (mask->ipv6_hlimit) {
 558		OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
 559			       mask->ipv6_hlimit);
 560		flow_key->ip.ttl = nh->hop_limit;
 561	}
 562	return 0;
 563}
 564
 565static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
 566		   const struct nlattr *a)
 567{
 568	struct nshhdr *nh;
 569	size_t length;
 570	int err;
 571	u8 flags;
 572	u8 ttl;
 573	int i;
 574
 575	struct ovs_key_nsh key;
 576	struct ovs_key_nsh mask;
 577
 578	err = nsh_key_from_nlattr(a, &key, &mask);
 579	if (err)
 580		return err;
 581
 582	/* Make sure the NSH base header is there */
 583	if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
 584		return -ENOMEM;
 585
 586	nh = nsh_hdr(skb);
 587	length = nsh_hdr_len(nh);
 588
 589	/* Make sure the whole NSH header is there */
 590	err = skb_ensure_writable(skb, skb_network_offset(skb) +
 591				       length);
 592	if (unlikely(err))
 593		return err;
 594
 595	nh = nsh_hdr(skb);
 596	skb_postpull_rcsum(skb, nh, length);
 597	flags = nsh_get_flags(nh);
 598	flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
 599	flow_key->nsh.base.flags = flags;
 600	ttl = nsh_get_ttl(nh);
 601	ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
 602	flow_key->nsh.base.ttl = ttl;
 603	nsh_set_flags_and_ttl(nh, flags, ttl);
 604	nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
 605				  mask.base.path_hdr);
 606	flow_key->nsh.base.path_hdr = nh->path_hdr;
 607	switch (nh->mdtype) {
 608	case NSH_M_TYPE1:
 609		for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
 610			nh->md1.context[i] =
 611			    OVS_MASKED(nh->md1.context[i], key.context[i],
 612				       mask.context[i]);
 613		}
 614		memcpy(flow_key->nsh.context, nh->md1.context,
 615		       sizeof(nh->md1.context));
 616		break;
 617	case NSH_M_TYPE2:
 618		memset(flow_key->nsh.context, 0,
 619		       sizeof(flow_key->nsh.context));
 620		break;
 621	default:
 622		return -EINVAL;
 623	}
 624	skb_postpush_rcsum(skb, nh, length);
 625	return 0;
 626}
 627
 628/* Must follow skb_ensure_writable() since that can move the skb data. */
 629static void set_tp_port(struct sk_buff *skb, __be16 *port,
 630			__be16 new_port, __sum16 *check)
 631{
 632	inet_proto_csum_replace2(check, skb, *port, new_port, false);
 633	*port = new_port;
 634}
 635
 636static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 637		   const struct ovs_key_udp *key,
 638		   const struct ovs_key_udp *mask)
 639{
 640	struct udphdr *uh;
 641	__be16 src, dst;
 642	int err;
 643
 644	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 645				  sizeof(struct udphdr));
 646	if (unlikely(err))
 647		return err;
 648
 649	uh = udp_hdr(skb);
 650	/* Either of the masks is non-zero, so do not bother checking them. */
 651	src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
 652	dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
 653
 654	if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 655		if (likely(src != uh->source)) {
 656			set_tp_port(skb, &uh->source, src, &uh->check);
 657			flow_key->tp.src = src;
 658		}
 659		if (likely(dst != uh->dest)) {
 660			set_tp_port(skb, &uh->dest, dst, &uh->check);
 661			flow_key->tp.dst = dst;
 662		}
 663
 664		if (unlikely(!uh->check))
 665			uh->check = CSUM_MANGLED_0;
 666	} else {
 667		uh->source = src;
 668		uh->dest = dst;
 669		flow_key->tp.src = src;
 670		flow_key->tp.dst = dst;
 671	}
 672
 673	skb_clear_hash(skb);
 674
 675	return 0;
 676}
 677
 678static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 679		   const struct ovs_key_tcp *key,
 680		   const struct ovs_key_tcp *mask)
 681{
 682	struct tcphdr *th;
 683	__be16 src, dst;
 684	int err;
 685
 686	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 687				  sizeof(struct tcphdr));
 688	if (unlikely(err))
 689		return err;
 690
 691	th = tcp_hdr(skb);
 692	src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
 693	if (likely(src != th->source)) {
 694		set_tp_port(skb, &th->source, src, &th->check);
 695		flow_key->tp.src = src;
 696	}
 697	dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
 698	if (likely(dst != th->dest)) {
 699		set_tp_port(skb, &th->dest, dst, &th->check);
 700		flow_key->tp.dst = dst;
 701	}
 702	skb_clear_hash(skb);
 703
 704	return 0;
 705}
 706
 707static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 708		    const struct ovs_key_sctp *key,
 709		    const struct ovs_key_sctp *mask)
 710{
 711	unsigned int sctphoff = skb_transport_offset(skb);
 712	struct sctphdr *sh;
 713	__le32 old_correct_csum, new_csum, old_csum;
 714	int err;
 715
 716	err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
 717	if (unlikely(err))
 718		return err;
 719
 720	sh = sctp_hdr(skb);
 721	old_csum = sh->checksum;
 722	old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 723
 724	sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
 725	sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
 726
 727	new_csum = sctp_compute_cksum(skb, sctphoff);
 728
 729	/* Carry any checksum errors through. */
 730	sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 731
 732	skb_clear_hash(skb);
 733	flow_key->tp.src = sh->source;
 734	flow_key->tp.dst = sh->dest;
 735
 736	return 0;
 737}
 738
 739static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 740{
 741	struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
 742	struct vport *vport = data->vport;
 743
 744	if (skb_cow_head(skb, data->l2_len) < 0) {
 745		kfree_skb(skb);
 746		return -ENOMEM;
 747	}
 748
 749	__skb_dst_copy(skb, data->dst);
 750	*OVS_CB(skb) = data->cb;
 751	skb->inner_protocol = data->inner_protocol;
 752	if (data->vlan_tci & VLAN_CFI_MASK)
 753		__vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
 754	else
 755		__vlan_hwaccel_clear_tag(skb);
 756
 757	/* Reconstruct the MAC header.  */
 758	skb_push(skb, data->l2_len);
 759	memcpy(skb->data, &data->l2_data, data->l2_len);
 760	skb_postpush_rcsum(skb, skb->data, data->l2_len);
 761	skb_reset_mac_header(skb);
 762
 763	if (eth_p_mpls(skb->protocol)) {
 764		skb->inner_network_header = skb->network_header;
 765		skb_set_network_header(skb, data->network_offset);
 766		skb_reset_mac_len(skb);
 767	}
 768
 769	ovs_vport_send(vport, skb, data->mac_proto);
 770	return 0;
 771}
 772
 773static unsigned int
 774ovs_dst_get_mtu(const struct dst_entry *dst)
 775{
 776	return dst->dev->mtu;
 777}
 778
 779static struct dst_ops ovs_dst_ops = {
 780	.family = AF_UNSPEC,
 781	.mtu = ovs_dst_get_mtu,
 782};
 783
 784/* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
 785 * ovs_vport_output(), which is called once per fragmented packet.
 786 */
 787static void prepare_frag(struct vport *vport, struct sk_buff *skb,
 788			 u16 orig_network_offset, u8 mac_proto)
 789{
 790	unsigned int hlen = skb_network_offset(skb);
 791	struct ovs_frag_data *data;
 792
 793	data = this_cpu_ptr(&ovs_frag_data_storage);
 794	data->dst = skb->_skb_refdst;
 795	data->vport = vport;
 796	data->cb = *OVS_CB(skb);
 797	data->inner_protocol = skb->inner_protocol;
 798	data->network_offset = orig_network_offset;
 799	if (skb_vlan_tag_present(skb))
 800		data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
 801	else
 802		data->vlan_tci = 0;
 803	data->vlan_proto = skb->vlan_proto;
 804	data->mac_proto = mac_proto;
 805	data->l2_len = hlen;
 806	memcpy(&data->l2_data, skb->data, hlen);
 807
 808	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 809	skb_pull(skb, hlen);
 810}
 811
 812static void ovs_fragment(struct net *net, struct vport *vport,
 813			 struct sk_buff *skb, u16 mru,
 814			 struct sw_flow_key *key)
 815{
 816	u16 orig_network_offset = 0;
 817
 818	if (eth_p_mpls(skb->protocol)) {
 819		orig_network_offset = skb_network_offset(skb);
 820		skb->network_header = skb->inner_network_header;
 821	}
 822
 823	if (skb_network_offset(skb) > MAX_L2_LEN) {
 824		OVS_NLERR(1, "L2 header too long to fragment");
 825		goto err;
 826	}
 827
 828	if (key->eth.type == htons(ETH_P_IP)) {
 829		struct dst_entry ovs_dst;
 830		unsigned long orig_dst;
 831
 832		prepare_frag(vport, skb, orig_network_offset,
 833			     ovs_key_mac_proto(key));
 834		dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
 835			 DST_OBSOLETE_NONE, DST_NOCOUNT);
 836		ovs_dst.dev = vport->dev;
 837
 838		orig_dst = skb->_skb_refdst;
 839		skb_dst_set_noref(skb, &ovs_dst);
 840		IPCB(skb)->frag_max_size = mru;
 841
 842		ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
 843		refdst_drop(orig_dst);
 844	} else if (key->eth.type == htons(ETH_P_IPV6)) {
 845		const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
 846		unsigned long orig_dst;
 847		struct rt6_info ovs_rt;
 848
 849		if (!v6ops)
 850			goto err;
 851
 852		prepare_frag(vport, skb, orig_network_offset,
 853			     ovs_key_mac_proto(key));
 854		memset(&ovs_rt, 0, sizeof(ovs_rt));
 855		dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
 856			 DST_OBSOLETE_NONE, DST_NOCOUNT);
 857		ovs_rt.dst.dev = vport->dev;
 858
 859		orig_dst = skb->_skb_refdst;
 860		skb_dst_set_noref(skb, &ovs_rt.dst);
 861		IP6CB(skb)->frag_max_size = mru;
 862
 863		v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
 864		refdst_drop(orig_dst);
 865	} else {
 866		WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
 867			  ovs_vport_name(vport), ntohs(key->eth.type), mru,
 868			  vport->dev->mtu);
 869		goto err;
 870	}
 871
 872	return;
 873err:
 874	kfree_skb(skb);
 875}
 876
 877static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
 878		      struct sw_flow_key *key)
 879{
 880	struct vport *vport = ovs_vport_rcu(dp, out_port);
 881
 882	if (likely(vport)) {
 883		u16 mru = OVS_CB(skb)->mru;
 884		u32 cutlen = OVS_CB(skb)->cutlen;
 885
 886		if (unlikely(cutlen > 0)) {
 887			if (skb->len - cutlen > ovs_mac_header_len(key))
 888				pskb_trim(skb, skb->len - cutlen);
 889			else
 890				pskb_trim(skb, ovs_mac_header_len(key));
 891		}
 892
 893		if (likely(!mru ||
 894		           (skb->len <= mru + vport->dev->hard_header_len))) {
 895			ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
 896		} else if (mru <= vport->dev->mtu) {
 897			struct net *net = read_pnet(&dp->net);
 898
 899			ovs_fragment(net, vport, skb, mru, key);
 900		} else {
 901			kfree_skb(skb);
 902		}
 903	} else {
 904		kfree_skb(skb);
 905	}
 906}
 907
 908static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 909			    struct sw_flow_key *key, const struct nlattr *attr,
 910			    const struct nlattr *actions, int actions_len,
 911			    uint32_t cutlen)
 912{
 913	struct dp_upcall_info upcall;
 914	const struct nlattr *a;
 915	int rem;
 916
 917	memset(&upcall, 0, sizeof(upcall));
 918	upcall.cmd = OVS_PACKET_CMD_ACTION;
 919	upcall.mru = OVS_CB(skb)->mru;
 920
 921	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 922		 a = nla_next(a, &rem)) {
 923		switch (nla_type(a)) {
 924		case OVS_USERSPACE_ATTR_USERDATA:
 925			upcall.userdata = a;
 926			break;
 927
 928		case OVS_USERSPACE_ATTR_PID:
 929			upcall.portid = nla_get_u32(a);
 930			break;
 931
 932		case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
 933			/* Get out tunnel info. */
 934			struct vport *vport;
 935
 936			vport = ovs_vport_rcu(dp, nla_get_u32(a));
 937			if (vport) {
 938				int err;
 939
 940				err = dev_fill_metadata_dst(vport->dev, skb);
 941				if (!err)
 942					upcall.egress_tun_info = skb_tunnel_info(skb);
 943			}
 944
 945			break;
 946		}
 947
 948		case OVS_USERSPACE_ATTR_ACTIONS: {
 949			/* Include actions. */
 950			upcall.actions = actions;
 951			upcall.actions_len = actions_len;
 952			break;
 953		}
 954
 955		} /* End of switch. */
 956	}
 957
 958	return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
 959}
 960
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 961/* When 'last' is true, sample() should always consume the 'skb'.
 962 * Otherwise, sample() should keep 'skb' intact regardless what
 963 * actions are executed within sample().
 964 */
 965static int sample(struct datapath *dp, struct sk_buff *skb,
 966		  struct sw_flow_key *key, const struct nlattr *attr,
 967		  bool last)
 968{
 969	struct nlattr *actions;
 970	struct nlattr *sample_arg;
 971	int rem = nla_len(attr);
 972	const struct sample_arg *arg;
 973	bool clone_flow_key;
 974
 975	/* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
 976	sample_arg = nla_data(attr);
 977	arg = nla_data(sample_arg);
 978	actions = nla_next(sample_arg, &rem);
 979
 980	if ((arg->probability != U32_MAX) &&
 981	    (!arg->probability || prandom_u32() > arg->probability)) {
 982		if (last)
 983			consume_skb(skb);
 984		return 0;
 985	}
 986
 987	clone_flow_key = !arg->exec;
 988	return clone_execute(dp, skb, key, 0, actions, rem, last,
 989			     clone_flow_key);
 990}
 991
 992/* When 'last' is true, clone() should always consume the 'skb'.
 993 * Otherwise, clone() should keep 'skb' intact regardless what
 994 * actions are executed within clone().
 995 */
 996static int clone(struct datapath *dp, struct sk_buff *skb,
 997		 struct sw_flow_key *key, const struct nlattr *attr,
 998		 bool last)
 999{
1000	struct nlattr *actions;
1001	struct nlattr *clone_arg;
1002	int rem = nla_len(attr);
1003	bool dont_clone_flow_key;
1004
1005	/* The first action is always 'OVS_CLONE_ATTR_ARG'. */
1006	clone_arg = nla_data(attr);
1007	dont_clone_flow_key = nla_get_u32(clone_arg);
1008	actions = nla_next(clone_arg, &rem);
1009
1010	return clone_execute(dp, skb, key, 0, actions, rem, last,
1011			     !dont_clone_flow_key);
1012}
1013
1014static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1015			 const struct nlattr *attr)
1016{
1017	struct ovs_action_hash *hash_act = nla_data(attr);
1018	u32 hash = 0;
1019
1020	/* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
1021	hash = skb_get_hash(skb);
1022	hash = jhash_1word(hash, hash_act->hash_basis);
1023	if (!hash)
1024		hash = 0x1;
1025
1026	key->ovs_flow_hash = hash;
1027}
1028
1029static int execute_set_action(struct sk_buff *skb,
1030			      struct sw_flow_key *flow_key,
1031			      const struct nlattr *a)
1032{
1033	/* Only tunnel set execution is supported without a mask. */
1034	if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1035		struct ovs_tunnel_info *tun = nla_data(a);
1036
1037		skb_dst_drop(skb);
1038		dst_hold((struct dst_entry *)tun->tun_dst);
1039		skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1040		return 0;
1041	}
1042
1043	return -EINVAL;
1044}
1045
1046/* Mask is at the midpoint of the data. */
1047#define get_mask(a, type) ((const type)nla_data(a) + 1)
1048
1049static int execute_masked_set_action(struct sk_buff *skb,
1050				     struct sw_flow_key *flow_key,
1051				     const struct nlattr *a)
1052{
1053	int err = 0;
1054
1055	switch (nla_type(a)) {
1056	case OVS_KEY_ATTR_PRIORITY:
1057		OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1058			       *get_mask(a, u32 *));
1059		flow_key->phy.priority = skb->priority;
1060		break;
1061
1062	case OVS_KEY_ATTR_SKB_MARK:
1063		OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1064		flow_key->phy.skb_mark = skb->mark;
1065		break;
1066
1067	case OVS_KEY_ATTR_TUNNEL_INFO:
1068		/* Masked data not supported for tunnel. */
1069		err = -EINVAL;
1070		break;
1071
1072	case OVS_KEY_ATTR_ETHERNET:
1073		err = set_eth_addr(skb, flow_key, nla_data(a),
1074				   get_mask(a, struct ovs_key_ethernet *));
1075		break;
1076
1077	case OVS_KEY_ATTR_NSH:
1078		err = set_nsh(skb, flow_key, a);
1079		break;
1080
1081	case OVS_KEY_ATTR_IPV4:
1082		err = set_ipv4(skb, flow_key, nla_data(a),
1083			       get_mask(a, struct ovs_key_ipv4 *));
1084		break;
1085
1086	case OVS_KEY_ATTR_IPV6:
1087		err = set_ipv6(skb, flow_key, nla_data(a),
1088			       get_mask(a, struct ovs_key_ipv6 *));
1089		break;
1090
1091	case OVS_KEY_ATTR_TCP:
1092		err = set_tcp(skb, flow_key, nla_data(a),
1093			      get_mask(a, struct ovs_key_tcp *));
1094		break;
1095
1096	case OVS_KEY_ATTR_UDP:
1097		err = set_udp(skb, flow_key, nla_data(a),
1098			      get_mask(a, struct ovs_key_udp *));
1099		break;
1100
1101	case OVS_KEY_ATTR_SCTP:
1102		err = set_sctp(skb, flow_key, nla_data(a),
1103			       get_mask(a, struct ovs_key_sctp *));
1104		break;
1105
1106	case OVS_KEY_ATTR_MPLS:
1107		err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1108								    __be32 *));
1109		break;
1110
1111	case OVS_KEY_ATTR_CT_STATE:
1112	case OVS_KEY_ATTR_CT_ZONE:
1113	case OVS_KEY_ATTR_CT_MARK:
1114	case OVS_KEY_ATTR_CT_LABELS:
1115	case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1116	case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1117		err = -EINVAL;
1118		break;
1119	}
1120
1121	return err;
1122}
1123
1124static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1125			  struct sw_flow_key *key,
1126			  const struct nlattr *a, bool last)
1127{
1128	u32 recirc_id;
1129
1130	if (!is_flow_key_valid(key)) {
1131		int err;
1132
1133		err = ovs_flow_key_update(skb, key);
1134		if (err)
1135			return err;
1136	}
1137	BUG_ON(!is_flow_key_valid(key));
1138
1139	recirc_id = nla_get_u32(a);
1140	return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1141}
1142
1143static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1144				 struct sw_flow_key *key,
1145				 const struct nlattr *attr, bool last)
1146{
 
1147	const struct nlattr *actions, *cpl_arg;
 
1148	const struct check_pkt_len_arg *arg;
1149	int rem = nla_len(attr);
1150	bool clone_flow_key;
1151
1152	/* The first netlink attribute in 'attr' is always
1153	 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1154	 */
1155	cpl_arg = nla_data(attr);
1156	arg = nla_data(cpl_arg);
1157
1158	if (skb->len <= arg->pkt_len) {
 
 
 
 
1159		/* Second netlink attribute in 'attr' is always
1160		 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1161		 */
1162		actions = nla_next(cpl_arg, &rem);
1163		clone_flow_key = !arg->exec_for_lesser_equal;
1164	} else {
1165		/* Third netlink attribute in 'attr' is always
1166		 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1167		 */
1168		actions = nla_next(cpl_arg, &rem);
1169		actions = nla_next(actions, &rem);
1170		clone_flow_key = !arg->exec_for_greater;
1171	}
1172
1173	return clone_execute(dp, skb, key, 0, nla_data(actions),
1174			     nla_len(actions), last, clone_flow_key);
1175}
1176
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1177/* Execute a list of actions against 'skb'. */
1178static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1179			      struct sw_flow_key *key,
1180			      const struct nlattr *attr, int len)
1181{
1182	const struct nlattr *a;
1183	int rem;
1184
1185	for (a = attr, rem = len; rem > 0;
1186	     a = nla_next(a, &rem)) {
1187		int err = 0;
1188
1189		switch (nla_type(a)) {
1190		case OVS_ACTION_ATTR_OUTPUT: {
1191			int port = nla_get_u32(a);
1192			struct sk_buff *clone;
1193
1194			/* Every output action needs a separate clone
1195			 * of 'skb', In case the output action is the
1196			 * last action, cloning can be avoided.
1197			 */
1198			if (nla_is_last(a, rem)) {
1199				do_output(dp, skb, port, key);
1200				/* 'skb' has been used for output.
1201				 */
1202				return 0;
1203			}
1204
1205			clone = skb_clone(skb, GFP_ATOMIC);
1206			if (clone)
1207				do_output(dp, clone, port, key);
1208			OVS_CB(skb)->cutlen = 0;
1209			break;
1210		}
1211
1212		case OVS_ACTION_ATTR_TRUNC: {
1213			struct ovs_action_trunc *trunc = nla_data(a);
1214
1215			if (skb->len > trunc->max_len)
1216				OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1217			break;
1218		}
1219
1220		case OVS_ACTION_ATTR_USERSPACE:
1221			output_userspace(dp, skb, key, a, attr,
1222						     len, OVS_CB(skb)->cutlen);
1223			OVS_CB(skb)->cutlen = 0;
1224			break;
1225
1226		case OVS_ACTION_ATTR_HASH:
1227			execute_hash(skb, key, a);
1228			break;
1229
1230		case OVS_ACTION_ATTR_PUSH_MPLS:
1231			err = push_mpls(skb, key, nla_data(a));
 
 
 
1232			break;
 
 
 
 
 
 
 
1233
 
 
 
 
1234		case OVS_ACTION_ATTR_POP_MPLS:
1235			err = pop_mpls(skb, key, nla_get_be16(a));
1236			break;
1237
1238		case OVS_ACTION_ATTR_PUSH_VLAN:
1239			err = push_vlan(skb, key, nla_data(a));
1240			break;
1241
1242		case OVS_ACTION_ATTR_POP_VLAN:
1243			err = pop_vlan(skb, key);
1244			break;
1245
1246		case OVS_ACTION_ATTR_RECIRC: {
1247			bool last = nla_is_last(a, rem);
1248
1249			err = execute_recirc(dp, skb, key, a, last);
1250			if (last) {
1251				/* If this is the last action, the skb has
1252				 * been consumed or freed.
1253				 * Return immediately.
1254				 */
1255				return err;
1256			}
1257			break;
1258		}
1259
1260		case OVS_ACTION_ATTR_SET:
1261			err = execute_set_action(skb, key, nla_data(a));
1262			break;
1263
1264		case OVS_ACTION_ATTR_SET_MASKED:
1265		case OVS_ACTION_ATTR_SET_TO_MASKED:
1266			err = execute_masked_set_action(skb, key, nla_data(a));
1267			break;
1268
1269		case OVS_ACTION_ATTR_SAMPLE: {
1270			bool last = nla_is_last(a, rem);
1271
1272			err = sample(dp, skb, key, a, last);
1273			if (last)
1274				return err;
1275
1276			break;
1277		}
1278
1279		case OVS_ACTION_ATTR_CT:
1280			if (!is_flow_key_valid(key)) {
1281				err = ovs_flow_key_update(skb, key);
1282				if (err)
1283					return err;
1284			}
1285
1286			err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1287					     nla_data(a));
1288
1289			/* Hide stolen IP fragments from user space. */
1290			if (err)
1291				return err == -EINPROGRESS ? 0 : err;
1292			break;
1293
1294		case OVS_ACTION_ATTR_CT_CLEAR:
1295			err = ovs_ct_clear(skb, key);
1296			break;
1297
1298		case OVS_ACTION_ATTR_PUSH_ETH:
1299			err = push_eth(skb, key, nla_data(a));
1300			break;
1301
1302		case OVS_ACTION_ATTR_POP_ETH:
1303			err = pop_eth(skb, key);
1304			break;
1305
1306		case OVS_ACTION_ATTR_PUSH_NSH: {
1307			u8 buffer[NSH_HDR_MAX_LEN];
1308			struct nshhdr *nh = (struct nshhdr *)buffer;
1309
1310			err = nsh_hdr_from_nlattr(nla_data(a), nh,
1311						  NSH_HDR_MAX_LEN);
1312			if (unlikely(err))
1313				break;
1314			err = push_nsh(skb, key, nh);
1315			break;
1316		}
1317
1318		case OVS_ACTION_ATTR_POP_NSH:
1319			err = pop_nsh(skb, key);
1320			break;
1321
1322		case OVS_ACTION_ATTR_METER:
1323			if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1324				consume_skb(skb);
1325				return 0;
1326			}
1327			break;
1328
1329		case OVS_ACTION_ATTR_CLONE: {
1330			bool last = nla_is_last(a, rem);
1331
1332			err = clone(dp, skb, key, a, last);
1333			if (last)
1334				return err;
1335
1336			break;
1337		}
1338
1339		case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1340			bool last = nla_is_last(a, rem);
1341
1342			err = execute_check_pkt_len(dp, skb, key, a, last);
1343			if (last)
1344				return err;
1345
1346			break;
1347		}
 
 
 
 
 
 
 
 
 
1348		}
1349
1350		if (unlikely(err)) {
1351			kfree_skb(skb);
1352			return err;
1353		}
1354	}
1355
1356	consume_skb(skb);
1357	return 0;
1358}
1359
1360/* Execute the actions on the clone of the packet. The effect of the
1361 * execution does not affect the original 'skb' nor the original 'key'.
1362 *
1363 * The execution may be deferred in case the actions can not be executed
1364 * immediately.
1365 */
1366static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1367			 struct sw_flow_key *key, u32 recirc_id,
1368			 const struct nlattr *actions, int len,
1369			 bool last, bool clone_flow_key)
1370{
1371	struct deferred_action *da;
1372	struct sw_flow_key *clone;
1373
1374	skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1375	if (!skb) {
1376		/* Out of memory, skip this action.
1377		 */
1378		return 0;
1379	}
1380
1381	/* When clone_flow_key is false, the 'key' will not be change
1382	 * by the actions, then the 'key' can be used directly.
1383	 * Otherwise, try to clone key from the next recursion level of
1384	 * 'flow_keys'. If clone is successful, execute the actions
1385	 * without deferring.
1386	 */
1387	clone = clone_flow_key ? clone_key(key) : key;
1388	if (clone) {
1389		int err = 0;
1390
1391		if (actions) { /* Sample action */
1392			if (clone_flow_key)
1393				__this_cpu_inc(exec_actions_level);
1394
1395			err = do_execute_actions(dp, skb, clone,
1396						 actions, len);
1397
1398			if (clone_flow_key)
1399				__this_cpu_dec(exec_actions_level);
1400		} else { /* Recirc action */
1401			clone->recirc_id = recirc_id;
1402			ovs_dp_process_packet(skb, clone);
1403		}
1404		return err;
1405	}
1406
1407	/* Out of 'flow_keys' space. Defer actions */
1408	da = add_deferred_actions(skb, key, actions, len);
1409	if (da) {
1410		if (!actions) { /* Recirc action */
1411			key = &da->pkt_key;
1412			key->recirc_id = recirc_id;
1413		}
1414	} else {
1415		/* Out of per CPU action FIFO space. Drop the 'skb' and
1416		 * log an error.
1417		 */
1418		kfree_skb(skb);
1419
1420		if (net_ratelimit()) {
1421			if (actions) { /* Sample action */
1422				pr_warn("%s: deferred action limit reached, drop sample action\n",
1423					ovs_dp_name(dp));
1424			} else {  /* Recirc action */
1425				pr_warn("%s: deferred action limit reached, drop recirc action\n",
1426					ovs_dp_name(dp));
1427			}
1428		}
1429	}
1430	return 0;
1431}
1432
1433static void process_deferred_actions(struct datapath *dp)
1434{
1435	struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1436
1437	/* Do not touch the FIFO in case there is no deferred actions. */
1438	if (action_fifo_is_empty(fifo))
1439		return;
1440
1441	/* Finishing executing all deferred actions. */
1442	do {
1443		struct deferred_action *da = action_fifo_get(fifo);
1444		struct sk_buff *skb = da->skb;
1445		struct sw_flow_key *key = &da->pkt_key;
1446		const struct nlattr *actions = da->actions;
1447		int actions_len = da->actions_len;
1448
1449		if (actions)
1450			do_execute_actions(dp, skb, key, actions, actions_len);
1451		else
1452			ovs_dp_process_packet(skb, key);
1453	} while (!action_fifo_is_empty(fifo));
1454
1455	/* Reset FIFO for the next packet.  */
1456	action_fifo_init(fifo);
1457}
1458
1459/* Execute a list of actions against 'skb'. */
1460int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1461			const struct sw_flow_actions *acts,
1462			struct sw_flow_key *key)
1463{
1464	int err, level;
1465
1466	level = __this_cpu_inc_return(exec_actions_level);
1467	if (unlikely(level > OVS_RECURSION_LIMIT)) {
1468		net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1469				     ovs_dp_name(dp));
1470		kfree_skb(skb);
1471		err = -ENETDOWN;
1472		goto out;
1473	}
1474
1475	OVS_CB(skb)->acts_origlen = acts->orig_len;
1476	err = do_execute_actions(dp, skb, key,
1477				 acts->actions, acts->actions_len);
1478
1479	if (level == 1)
1480		process_deferred_actions(dp);
1481
1482out:
1483	__this_cpu_dec(exec_actions_level);
1484	return err;
1485}
1486
1487int action_fifos_init(void)
1488{
1489	action_fifos = alloc_percpu(struct action_fifo);
1490	if (!action_fifos)
1491		return -ENOMEM;
1492
1493	flow_keys = alloc_percpu(struct action_flow_keys);
1494	if (!flow_keys) {
1495		free_percpu(action_fifos);
1496		return -ENOMEM;
1497	}
1498
1499	return 0;
1500}
1501
1502void action_fifos_exit(void)
1503{
1504	free_percpu(action_fifos);
1505	free_percpu(flow_keys);
1506}
v5.9
   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/netfilter_ipv6.h>
  13#include <linux/sctp.h>
  14#include <linux/tcp.h>
  15#include <linux/udp.h>
  16#include <linux/in6.h>
  17#include <linux/if_arp.h>
  18#include <linux/if_vlan.h>
  19
  20#include <net/dst.h>
  21#include <net/ip.h>
  22#include <net/ipv6.h>
  23#include <net/ip6_fib.h>
  24#include <net/checksum.h>
  25#include <net/dsfield.h>
  26#include <net/mpls.h>
  27#include <net/sctp/checksum.h>
  28
  29#include "datapath.h"
  30#include "flow.h"
  31#include "conntrack.h"
  32#include "vport.h"
  33#include "flow_netlink.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	stack = mpls_hdr(skb);
 204	lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
 205	err = skb_mpls_update_lse(skb, lse);
 206	if (err)
 207		return err;
 208
 209	flow_key->mpls.lse[0] = lse;
 210	return 0;
 211}
 212
 213static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 214{
 215	int err;
 216
 217	err = skb_vlan_pop(skb);
 218	if (skb_vlan_tag_present(skb)) {
 219		invalidate_flow_key(key);
 220	} else {
 221		key->eth.vlan.tci = 0;
 222		key->eth.vlan.tpid = 0;
 223	}
 224	return err;
 225}
 226
 227static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
 228		     const struct ovs_action_push_vlan *vlan)
 229{
 230	if (skb_vlan_tag_present(skb)) {
 231		invalidate_flow_key(key);
 232	} else {
 233		key->eth.vlan.tci = vlan->vlan_tci;
 234		key->eth.vlan.tpid = vlan->vlan_tpid;
 235	}
 236	return skb_vlan_push(skb, vlan->vlan_tpid,
 237			     ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
 238}
 239
 240/* 'src' is already properly masked. */
 241static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
 242{
 243	u16 *dst = (u16 *)dst_;
 244	const u16 *src = (const u16 *)src_;
 245	const u16 *mask = (const u16 *)mask_;
 246
 247	OVS_SET_MASKED(dst[0], src[0], mask[0]);
 248	OVS_SET_MASKED(dst[1], src[1], mask[1]);
 249	OVS_SET_MASKED(dst[2], src[2], mask[2]);
 250}
 251
 252static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
 253			const struct ovs_key_ethernet *key,
 254			const struct ovs_key_ethernet *mask)
 255{
 256	int err;
 257
 258	err = skb_ensure_writable(skb, ETH_HLEN);
 259	if (unlikely(err))
 260		return err;
 261
 262	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 263
 264	ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
 265			       mask->eth_src);
 266	ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
 267			       mask->eth_dst);
 268
 269	skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
 270
 271	ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
 272	ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
 273	return 0;
 274}
 275
 276/* pop_eth does not support VLAN packets as this action is never called
 277 * for them.
 278 */
 279static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
 280{
 281	skb_pull_rcsum(skb, ETH_HLEN);
 282	skb_reset_mac_header(skb);
 283	skb_reset_mac_len(skb);
 284
 285	/* safe right before invalidate_flow_key */
 286	key->mac_proto = MAC_PROTO_NONE;
 287	invalidate_flow_key(key);
 288	return 0;
 289}
 290
 291static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
 292		    const struct ovs_action_push_eth *ethh)
 293{
 294	struct ethhdr *hdr;
 295
 296	/* Add the new Ethernet header */
 297	if (skb_cow_head(skb, ETH_HLEN) < 0)
 298		return -ENOMEM;
 299
 300	skb_push(skb, ETH_HLEN);
 301	skb_reset_mac_header(skb);
 302	skb_reset_mac_len(skb);
 303
 304	hdr = eth_hdr(skb);
 305	ether_addr_copy(hdr->h_source, ethh->addresses.eth_src);
 306	ether_addr_copy(hdr->h_dest, ethh->addresses.eth_dst);
 307	hdr->h_proto = skb->protocol;
 308
 309	skb_postpush_rcsum(skb, hdr, ETH_HLEN);
 310
 311	/* safe right before invalidate_flow_key */
 312	key->mac_proto = MAC_PROTO_ETHERNET;
 313	invalidate_flow_key(key);
 314	return 0;
 315}
 316
 317static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
 318		    const struct nshhdr *nh)
 319{
 320	int err;
 321
 322	err = nsh_push(skb, nh);
 323	if (err)
 324		return err;
 325
 326	/* safe right before invalidate_flow_key */
 327	key->mac_proto = MAC_PROTO_NONE;
 328	invalidate_flow_key(key);
 329	return 0;
 330}
 331
 332static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
 333{
 334	int err;
 335
 336	err = nsh_pop(skb);
 337	if (err)
 338		return err;
 339
 340	/* safe right before invalidate_flow_key */
 341	if (skb->protocol == htons(ETH_P_TEB))
 342		key->mac_proto = MAC_PROTO_ETHERNET;
 343	else
 344		key->mac_proto = MAC_PROTO_NONE;
 345	invalidate_flow_key(key);
 346	return 0;
 347}
 348
 349static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
 350				  __be32 addr, __be32 new_addr)
 351{
 352	int transport_len = skb->len - skb_transport_offset(skb);
 353
 354	if (nh->frag_off & htons(IP_OFFSET))
 355		return;
 356
 357	if (nh->protocol == IPPROTO_TCP) {
 358		if (likely(transport_len >= sizeof(struct tcphdr)))
 359			inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
 360						 addr, new_addr, true);
 361	} else if (nh->protocol == IPPROTO_UDP) {
 362		if (likely(transport_len >= sizeof(struct udphdr))) {
 363			struct udphdr *uh = udp_hdr(skb);
 364
 365			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 366				inet_proto_csum_replace4(&uh->check, skb,
 367							 addr, new_addr, true);
 368				if (!uh->check)
 369					uh->check = CSUM_MANGLED_0;
 370			}
 371		}
 372	}
 373}
 374
 375static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
 376			__be32 *addr, __be32 new_addr)
 377{
 378	update_ip_l4_checksum(skb, nh, *addr, new_addr);
 379	csum_replace4(&nh->check, *addr, new_addr);
 380	skb_clear_hash(skb);
 381	*addr = new_addr;
 382}
 383
 384static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
 385				 __be32 addr[4], const __be32 new_addr[4])
 386{
 387	int transport_len = skb->len - skb_transport_offset(skb);
 388
 389	if (l4_proto == NEXTHDR_TCP) {
 390		if (likely(transport_len >= sizeof(struct tcphdr)))
 391			inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
 392						  addr, new_addr, true);
 393	} else if (l4_proto == NEXTHDR_UDP) {
 394		if (likely(transport_len >= sizeof(struct udphdr))) {
 395			struct udphdr *uh = udp_hdr(skb);
 396
 397			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 398				inet_proto_csum_replace16(&uh->check, skb,
 399							  addr, new_addr, true);
 400				if (!uh->check)
 401					uh->check = CSUM_MANGLED_0;
 402			}
 403		}
 404	} else if (l4_proto == NEXTHDR_ICMP) {
 405		if (likely(transport_len >= sizeof(struct icmp6hdr)))
 406			inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
 407						  skb, addr, new_addr, true);
 408	}
 409}
 410
 411static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
 412			   const __be32 mask[4], __be32 masked[4])
 413{
 414	masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
 415	masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
 416	masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
 417	masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
 418}
 419
 420static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
 421			  __be32 addr[4], const __be32 new_addr[4],
 422			  bool recalculate_csum)
 423{
 424	if (recalculate_csum)
 425		update_ipv6_checksum(skb, l4_proto, addr, new_addr);
 426
 427	skb_clear_hash(skb);
 428	memcpy(addr, new_addr, sizeof(__be32[4]));
 429}
 430
 431static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
 432{
 433	/* Bits 21-24 are always unmasked, so this retains their values. */
 434	OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
 435	OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
 436	OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
 437}
 438
 439static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
 440		       u8 mask)
 441{
 442	new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
 443
 444	csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
 445	nh->ttl = new_ttl;
 446}
 447
 448static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
 449		    const struct ovs_key_ipv4 *key,
 450		    const struct ovs_key_ipv4 *mask)
 451{
 452	struct iphdr *nh;
 453	__be32 new_addr;
 454	int err;
 455
 456	err = skb_ensure_writable(skb, skb_network_offset(skb) +
 457				  sizeof(struct iphdr));
 458	if (unlikely(err))
 459		return err;
 460
 461	nh = ip_hdr(skb);
 462
 463	/* Setting an IP addresses is typically only a side effect of
 464	 * matching on them in the current userspace implementation, so it
 465	 * makes sense to check if the value actually changed.
 466	 */
 467	if (mask->ipv4_src) {
 468		new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
 469
 470		if (unlikely(new_addr != nh->saddr)) {
 471			set_ip_addr(skb, nh, &nh->saddr, new_addr);
 472			flow_key->ipv4.addr.src = new_addr;
 473		}
 474	}
 475	if (mask->ipv4_dst) {
 476		new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
 477
 478		if (unlikely(new_addr != nh->daddr)) {
 479			set_ip_addr(skb, nh, &nh->daddr, new_addr);
 480			flow_key->ipv4.addr.dst = new_addr;
 481		}
 482	}
 483	if (mask->ipv4_tos) {
 484		ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
 485		flow_key->ip.tos = nh->tos;
 486	}
 487	if (mask->ipv4_ttl) {
 488		set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
 489		flow_key->ip.ttl = nh->ttl;
 490	}
 491
 492	return 0;
 493}
 494
 495static bool is_ipv6_mask_nonzero(const __be32 addr[4])
 496{
 497	return !!(addr[0] | addr[1] | addr[2] | addr[3]);
 498}
 499
 500static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
 501		    const struct ovs_key_ipv6 *key,
 502		    const struct ovs_key_ipv6 *mask)
 503{
 504	struct ipv6hdr *nh;
 505	int err;
 506
 507	err = skb_ensure_writable(skb, skb_network_offset(skb) +
 508				  sizeof(struct ipv6hdr));
 509	if (unlikely(err))
 510		return err;
 511
 512	nh = ipv6_hdr(skb);
 513
 514	/* Setting an IP addresses is typically only a side effect of
 515	 * matching on them in the current userspace implementation, so it
 516	 * makes sense to check if the value actually changed.
 517	 */
 518	if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
 519		__be32 *saddr = (__be32 *)&nh->saddr;
 520		__be32 masked[4];
 521
 522		mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
 523
 524		if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
 525			set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
 526				      true);
 527			memcpy(&flow_key->ipv6.addr.src, masked,
 528			       sizeof(flow_key->ipv6.addr.src));
 529		}
 530	}
 531	if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
 532		unsigned int offset = 0;
 533		int flags = IP6_FH_F_SKIP_RH;
 534		bool recalc_csum = true;
 535		__be32 *daddr = (__be32 *)&nh->daddr;
 536		__be32 masked[4];
 537
 538		mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
 539
 540		if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
 541			if (ipv6_ext_hdr(nh->nexthdr))
 542				recalc_csum = (ipv6_find_hdr(skb, &offset,
 543							     NEXTHDR_ROUTING,
 544							     NULL, &flags)
 545					       != NEXTHDR_ROUTING);
 546
 547			set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
 548				      recalc_csum);
 549			memcpy(&flow_key->ipv6.addr.dst, masked,
 550			       sizeof(flow_key->ipv6.addr.dst));
 551		}
 552	}
 553	if (mask->ipv6_tclass) {
 554		ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
 555		flow_key->ip.tos = ipv6_get_dsfield(nh);
 556	}
 557	if (mask->ipv6_label) {
 558		set_ipv6_fl(nh, ntohl(key->ipv6_label),
 559			    ntohl(mask->ipv6_label));
 560		flow_key->ipv6.label =
 561		    *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 562	}
 563	if (mask->ipv6_hlimit) {
 564		OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
 565			       mask->ipv6_hlimit);
 566		flow_key->ip.ttl = nh->hop_limit;
 567	}
 568	return 0;
 569}
 570
 571static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
 572		   const struct nlattr *a)
 573{
 574	struct nshhdr *nh;
 575	size_t length;
 576	int err;
 577	u8 flags;
 578	u8 ttl;
 579	int i;
 580
 581	struct ovs_key_nsh key;
 582	struct ovs_key_nsh mask;
 583
 584	err = nsh_key_from_nlattr(a, &key, &mask);
 585	if (err)
 586		return err;
 587
 588	/* Make sure the NSH base header is there */
 589	if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
 590		return -ENOMEM;
 591
 592	nh = nsh_hdr(skb);
 593	length = nsh_hdr_len(nh);
 594
 595	/* Make sure the whole NSH header is there */
 596	err = skb_ensure_writable(skb, skb_network_offset(skb) +
 597				       length);
 598	if (unlikely(err))
 599		return err;
 600
 601	nh = nsh_hdr(skb);
 602	skb_postpull_rcsum(skb, nh, length);
 603	flags = nsh_get_flags(nh);
 604	flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
 605	flow_key->nsh.base.flags = flags;
 606	ttl = nsh_get_ttl(nh);
 607	ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
 608	flow_key->nsh.base.ttl = ttl;
 609	nsh_set_flags_and_ttl(nh, flags, ttl);
 610	nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
 611				  mask.base.path_hdr);
 612	flow_key->nsh.base.path_hdr = nh->path_hdr;
 613	switch (nh->mdtype) {
 614	case NSH_M_TYPE1:
 615		for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
 616			nh->md1.context[i] =
 617			    OVS_MASKED(nh->md1.context[i], key.context[i],
 618				       mask.context[i]);
 619		}
 620		memcpy(flow_key->nsh.context, nh->md1.context,
 621		       sizeof(nh->md1.context));
 622		break;
 623	case NSH_M_TYPE2:
 624		memset(flow_key->nsh.context, 0,
 625		       sizeof(flow_key->nsh.context));
 626		break;
 627	default:
 628		return -EINVAL;
 629	}
 630	skb_postpush_rcsum(skb, nh, length);
 631	return 0;
 632}
 633
 634/* Must follow skb_ensure_writable() since that can move the skb data. */
 635static void set_tp_port(struct sk_buff *skb, __be16 *port,
 636			__be16 new_port, __sum16 *check)
 637{
 638	inet_proto_csum_replace2(check, skb, *port, new_port, false);
 639	*port = new_port;
 640}
 641
 642static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 643		   const struct ovs_key_udp *key,
 644		   const struct ovs_key_udp *mask)
 645{
 646	struct udphdr *uh;
 647	__be16 src, dst;
 648	int err;
 649
 650	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 651				  sizeof(struct udphdr));
 652	if (unlikely(err))
 653		return err;
 654
 655	uh = udp_hdr(skb);
 656	/* Either of the masks is non-zero, so do not bother checking them. */
 657	src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
 658	dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
 659
 660	if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
 661		if (likely(src != uh->source)) {
 662			set_tp_port(skb, &uh->source, src, &uh->check);
 663			flow_key->tp.src = src;
 664		}
 665		if (likely(dst != uh->dest)) {
 666			set_tp_port(skb, &uh->dest, dst, &uh->check);
 667			flow_key->tp.dst = dst;
 668		}
 669
 670		if (unlikely(!uh->check))
 671			uh->check = CSUM_MANGLED_0;
 672	} else {
 673		uh->source = src;
 674		uh->dest = dst;
 675		flow_key->tp.src = src;
 676		flow_key->tp.dst = dst;
 677	}
 678
 679	skb_clear_hash(skb);
 680
 681	return 0;
 682}
 683
 684static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 685		   const struct ovs_key_tcp *key,
 686		   const struct ovs_key_tcp *mask)
 687{
 688	struct tcphdr *th;
 689	__be16 src, dst;
 690	int err;
 691
 692	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
 693				  sizeof(struct tcphdr));
 694	if (unlikely(err))
 695		return err;
 696
 697	th = tcp_hdr(skb);
 698	src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
 699	if (likely(src != th->source)) {
 700		set_tp_port(skb, &th->source, src, &th->check);
 701		flow_key->tp.src = src;
 702	}
 703	dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
 704	if (likely(dst != th->dest)) {
 705		set_tp_port(skb, &th->dest, dst, &th->check);
 706		flow_key->tp.dst = dst;
 707	}
 708	skb_clear_hash(skb);
 709
 710	return 0;
 711}
 712
 713static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
 714		    const struct ovs_key_sctp *key,
 715		    const struct ovs_key_sctp *mask)
 716{
 717	unsigned int sctphoff = skb_transport_offset(skb);
 718	struct sctphdr *sh;
 719	__le32 old_correct_csum, new_csum, old_csum;
 720	int err;
 721
 722	err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
 723	if (unlikely(err))
 724		return err;
 725
 726	sh = sctp_hdr(skb);
 727	old_csum = sh->checksum;
 728	old_correct_csum = sctp_compute_cksum(skb, sctphoff);
 729
 730	sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
 731	sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
 732
 733	new_csum = sctp_compute_cksum(skb, sctphoff);
 734
 735	/* Carry any checksum errors through. */
 736	sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
 737
 738	skb_clear_hash(skb);
 739	flow_key->tp.src = sh->source;
 740	flow_key->tp.dst = sh->dest;
 741
 742	return 0;
 743}
 744
 745static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 746{
 747	struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
 748	struct vport *vport = data->vport;
 749
 750	if (skb_cow_head(skb, data->l2_len) < 0) {
 751		kfree_skb(skb);
 752		return -ENOMEM;
 753	}
 754
 755	__skb_dst_copy(skb, data->dst);
 756	*OVS_CB(skb) = data->cb;
 757	skb->inner_protocol = data->inner_protocol;
 758	if (data->vlan_tci & VLAN_CFI_MASK)
 759		__vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
 760	else
 761		__vlan_hwaccel_clear_tag(skb);
 762
 763	/* Reconstruct the MAC header.  */
 764	skb_push(skb, data->l2_len);
 765	memcpy(skb->data, &data->l2_data, data->l2_len);
 766	skb_postpush_rcsum(skb, skb->data, data->l2_len);
 767	skb_reset_mac_header(skb);
 768
 769	if (eth_p_mpls(skb->protocol)) {
 770		skb->inner_network_header = skb->network_header;
 771		skb_set_network_header(skb, data->network_offset);
 772		skb_reset_mac_len(skb);
 773	}
 774
 775	ovs_vport_send(vport, skb, data->mac_proto);
 776	return 0;
 777}
 778
 779static unsigned int
 780ovs_dst_get_mtu(const struct dst_entry *dst)
 781{
 782	return dst->dev->mtu;
 783}
 784
 785static struct dst_ops ovs_dst_ops = {
 786	.family = AF_UNSPEC,
 787	.mtu = ovs_dst_get_mtu,
 788};
 789
 790/* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
 791 * ovs_vport_output(), which is called once per fragmented packet.
 792 */
 793static void prepare_frag(struct vport *vport, struct sk_buff *skb,
 794			 u16 orig_network_offset, u8 mac_proto)
 795{
 796	unsigned int hlen = skb_network_offset(skb);
 797	struct ovs_frag_data *data;
 798
 799	data = this_cpu_ptr(&ovs_frag_data_storage);
 800	data->dst = skb->_skb_refdst;
 801	data->vport = vport;
 802	data->cb = *OVS_CB(skb);
 803	data->inner_protocol = skb->inner_protocol;
 804	data->network_offset = orig_network_offset;
 805	if (skb_vlan_tag_present(skb))
 806		data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
 807	else
 808		data->vlan_tci = 0;
 809	data->vlan_proto = skb->vlan_proto;
 810	data->mac_proto = mac_proto;
 811	data->l2_len = hlen;
 812	memcpy(&data->l2_data, skb->data, hlen);
 813
 814	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 815	skb_pull(skb, hlen);
 816}
 817
 818static void ovs_fragment(struct net *net, struct vport *vport,
 819			 struct sk_buff *skb, u16 mru,
 820			 struct sw_flow_key *key)
 821{
 822	u16 orig_network_offset = 0;
 823
 824	if (eth_p_mpls(skb->protocol)) {
 825		orig_network_offset = skb_network_offset(skb);
 826		skb->network_header = skb->inner_network_header;
 827	}
 828
 829	if (skb_network_offset(skb) > MAX_L2_LEN) {
 830		OVS_NLERR(1, "L2 header too long to fragment");
 831		goto err;
 832	}
 833
 834	if (key->eth.type == htons(ETH_P_IP)) {
 835		struct dst_entry ovs_dst;
 836		unsigned long orig_dst;
 837
 838		prepare_frag(vport, skb, orig_network_offset,
 839			     ovs_key_mac_proto(key));
 840		dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
 841			 DST_OBSOLETE_NONE, DST_NOCOUNT);
 842		ovs_dst.dev = vport->dev;
 843
 844		orig_dst = skb->_skb_refdst;
 845		skb_dst_set_noref(skb, &ovs_dst);
 846		IPCB(skb)->frag_max_size = mru;
 847
 848		ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
 849		refdst_drop(orig_dst);
 850	} else if (key->eth.type == htons(ETH_P_IPV6)) {
 851		const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
 852		unsigned long orig_dst;
 853		struct rt6_info ovs_rt;
 854
 855		if (!v6ops)
 856			goto err;
 857
 858		prepare_frag(vport, skb, orig_network_offset,
 859			     ovs_key_mac_proto(key));
 860		memset(&ovs_rt, 0, sizeof(ovs_rt));
 861		dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
 862			 DST_OBSOLETE_NONE, DST_NOCOUNT);
 863		ovs_rt.dst.dev = vport->dev;
 864
 865		orig_dst = skb->_skb_refdst;
 866		skb_dst_set_noref(skb, &ovs_rt.dst);
 867		IP6CB(skb)->frag_max_size = mru;
 868
 869		v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
 870		refdst_drop(orig_dst);
 871	} else {
 872		WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
 873			  ovs_vport_name(vport), ntohs(key->eth.type), mru,
 874			  vport->dev->mtu);
 875		goto err;
 876	}
 877
 878	return;
 879err:
 880	kfree_skb(skb);
 881}
 882
 883static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
 884		      struct sw_flow_key *key)
 885{
 886	struct vport *vport = ovs_vport_rcu(dp, out_port);
 887
 888	if (likely(vport)) {
 889		u16 mru = OVS_CB(skb)->mru;
 890		u32 cutlen = OVS_CB(skb)->cutlen;
 891
 892		if (unlikely(cutlen > 0)) {
 893			if (skb->len - cutlen > ovs_mac_header_len(key))
 894				pskb_trim(skb, skb->len - cutlen);
 895			else
 896				pskb_trim(skb, ovs_mac_header_len(key));
 897		}
 898
 899		if (likely(!mru ||
 900		           (skb->len <= mru + vport->dev->hard_header_len))) {
 901			ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
 902		} else if (mru <= vport->dev->mtu) {
 903			struct net *net = read_pnet(&dp->net);
 904
 905			ovs_fragment(net, vport, skb, mru, key);
 906		} else {
 907			kfree_skb(skb);
 908		}
 909	} else {
 910		kfree_skb(skb);
 911	}
 912}
 913
 914static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 915			    struct sw_flow_key *key, const struct nlattr *attr,
 916			    const struct nlattr *actions, int actions_len,
 917			    uint32_t cutlen)
 918{
 919	struct dp_upcall_info upcall;
 920	const struct nlattr *a;
 921	int rem;
 922
 923	memset(&upcall, 0, sizeof(upcall));
 924	upcall.cmd = OVS_PACKET_CMD_ACTION;
 925	upcall.mru = OVS_CB(skb)->mru;
 926
 927	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 928		 a = nla_next(a, &rem)) {
 929		switch (nla_type(a)) {
 930		case OVS_USERSPACE_ATTR_USERDATA:
 931			upcall.userdata = a;
 932			break;
 933
 934		case OVS_USERSPACE_ATTR_PID:
 935			upcall.portid = nla_get_u32(a);
 936			break;
 937
 938		case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
 939			/* Get out tunnel info. */
 940			struct vport *vport;
 941
 942			vport = ovs_vport_rcu(dp, nla_get_u32(a));
 943			if (vport) {
 944				int err;
 945
 946				err = dev_fill_metadata_dst(vport->dev, skb);
 947				if (!err)
 948					upcall.egress_tun_info = skb_tunnel_info(skb);
 949			}
 950
 951			break;
 952		}
 953
 954		case OVS_USERSPACE_ATTR_ACTIONS: {
 955			/* Include actions. */
 956			upcall.actions = actions;
 957			upcall.actions_len = actions_len;
 958			break;
 959		}
 960
 961		} /* End of switch. */
 962	}
 963
 964	return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
 965}
 966
 967static int dec_ttl_exception_handler(struct datapath *dp, struct sk_buff *skb,
 968				     struct sw_flow_key *key,
 969				     const struct nlattr *attr, bool last)
 970{
 971	/* The first action is always 'OVS_DEC_TTL_ATTR_ARG'. */
 972	struct nlattr *dec_ttl_arg = nla_data(attr);
 973	int rem = nla_len(attr);
 974
 975	if (nla_len(dec_ttl_arg)) {
 976		struct nlattr *actions = nla_next(dec_ttl_arg, &rem);
 977
 978		if (actions)
 979			return clone_execute(dp, skb, key, 0, actions, rem,
 980					     last, false);
 981	}
 982	consume_skb(skb);
 983	return 0;
 984}
 985
 986/* When 'last' is true, sample() should always consume the 'skb'.
 987 * Otherwise, sample() should keep 'skb' intact regardless what
 988 * actions are executed within sample().
 989 */
 990static int sample(struct datapath *dp, struct sk_buff *skb,
 991		  struct sw_flow_key *key, const struct nlattr *attr,
 992		  bool last)
 993{
 994	struct nlattr *actions;
 995	struct nlattr *sample_arg;
 996	int rem = nla_len(attr);
 997	const struct sample_arg *arg;
 998	bool clone_flow_key;
 999
1000	/* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
1001	sample_arg = nla_data(attr);
1002	arg = nla_data(sample_arg);
1003	actions = nla_next(sample_arg, &rem);
1004
1005	if ((arg->probability != U32_MAX) &&
1006	    (!arg->probability || prandom_u32() > arg->probability)) {
1007		if (last)
1008			consume_skb(skb);
1009		return 0;
1010	}
1011
1012	clone_flow_key = !arg->exec;
1013	return clone_execute(dp, skb, key, 0, actions, rem, last,
1014			     clone_flow_key);
1015}
1016
1017/* When 'last' is true, clone() should always consume the 'skb'.
1018 * Otherwise, clone() should keep 'skb' intact regardless what
1019 * actions are executed within clone().
1020 */
1021static int clone(struct datapath *dp, struct sk_buff *skb,
1022		 struct sw_flow_key *key, const struct nlattr *attr,
1023		 bool last)
1024{
1025	struct nlattr *actions;
1026	struct nlattr *clone_arg;
1027	int rem = nla_len(attr);
1028	bool dont_clone_flow_key;
1029
1030	/* The first action is always 'OVS_CLONE_ATTR_ARG'. */
1031	clone_arg = nla_data(attr);
1032	dont_clone_flow_key = nla_get_u32(clone_arg);
1033	actions = nla_next(clone_arg, &rem);
1034
1035	return clone_execute(dp, skb, key, 0, actions, rem, last,
1036			     !dont_clone_flow_key);
1037}
1038
1039static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1040			 const struct nlattr *attr)
1041{
1042	struct ovs_action_hash *hash_act = nla_data(attr);
1043	u32 hash = 0;
1044
1045	/* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
1046	hash = skb_get_hash(skb);
1047	hash = jhash_1word(hash, hash_act->hash_basis);
1048	if (!hash)
1049		hash = 0x1;
1050
1051	key->ovs_flow_hash = hash;
1052}
1053
1054static int execute_set_action(struct sk_buff *skb,
1055			      struct sw_flow_key *flow_key,
1056			      const struct nlattr *a)
1057{
1058	/* Only tunnel set execution is supported without a mask. */
1059	if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1060		struct ovs_tunnel_info *tun = nla_data(a);
1061
1062		skb_dst_drop(skb);
1063		dst_hold((struct dst_entry *)tun->tun_dst);
1064		skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1065		return 0;
1066	}
1067
1068	return -EINVAL;
1069}
1070
1071/* Mask is at the midpoint of the data. */
1072#define get_mask(a, type) ((const type)nla_data(a) + 1)
1073
1074static int execute_masked_set_action(struct sk_buff *skb,
1075				     struct sw_flow_key *flow_key,
1076				     const struct nlattr *a)
1077{
1078	int err = 0;
1079
1080	switch (nla_type(a)) {
1081	case OVS_KEY_ATTR_PRIORITY:
1082		OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1083			       *get_mask(a, u32 *));
1084		flow_key->phy.priority = skb->priority;
1085		break;
1086
1087	case OVS_KEY_ATTR_SKB_MARK:
1088		OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1089		flow_key->phy.skb_mark = skb->mark;
1090		break;
1091
1092	case OVS_KEY_ATTR_TUNNEL_INFO:
1093		/* Masked data not supported for tunnel. */
1094		err = -EINVAL;
1095		break;
1096
1097	case OVS_KEY_ATTR_ETHERNET:
1098		err = set_eth_addr(skb, flow_key, nla_data(a),
1099				   get_mask(a, struct ovs_key_ethernet *));
1100		break;
1101
1102	case OVS_KEY_ATTR_NSH:
1103		err = set_nsh(skb, flow_key, a);
1104		break;
1105
1106	case OVS_KEY_ATTR_IPV4:
1107		err = set_ipv4(skb, flow_key, nla_data(a),
1108			       get_mask(a, struct ovs_key_ipv4 *));
1109		break;
1110
1111	case OVS_KEY_ATTR_IPV6:
1112		err = set_ipv6(skb, flow_key, nla_data(a),
1113			       get_mask(a, struct ovs_key_ipv6 *));
1114		break;
1115
1116	case OVS_KEY_ATTR_TCP:
1117		err = set_tcp(skb, flow_key, nla_data(a),
1118			      get_mask(a, struct ovs_key_tcp *));
1119		break;
1120
1121	case OVS_KEY_ATTR_UDP:
1122		err = set_udp(skb, flow_key, nla_data(a),
1123			      get_mask(a, struct ovs_key_udp *));
1124		break;
1125
1126	case OVS_KEY_ATTR_SCTP:
1127		err = set_sctp(skb, flow_key, nla_data(a),
1128			       get_mask(a, struct ovs_key_sctp *));
1129		break;
1130
1131	case OVS_KEY_ATTR_MPLS:
1132		err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1133								    __be32 *));
1134		break;
1135
1136	case OVS_KEY_ATTR_CT_STATE:
1137	case OVS_KEY_ATTR_CT_ZONE:
1138	case OVS_KEY_ATTR_CT_MARK:
1139	case OVS_KEY_ATTR_CT_LABELS:
1140	case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1141	case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1142		err = -EINVAL;
1143		break;
1144	}
1145
1146	return err;
1147}
1148
1149static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1150			  struct sw_flow_key *key,
1151			  const struct nlattr *a, bool last)
1152{
1153	u32 recirc_id;
1154
1155	if (!is_flow_key_valid(key)) {
1156		int err;
1157
1158		err = ovs_flow_key_update(skb, key);
1159		if (err)
1160			return err;
1161	}
1162	BUG_ON(!is_flow_key_valid(key));
1163
1164	recirc_id = nla_get_u32(a);
1165	return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1166}
1167
1168static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1169				 struct sw_flow_key *key,
1170				 const struct nlattr *attr, bool last)
1171{
1172	struct ovs_skb_cb *ovs_cb = OVS_CB(skb);
1173	const struct nlattr *actions, *cpl_arg;
1174	int len, max_len, rem = nla_len(attr);
1175	const struct check_pkt_len_arg *arg;
 
1176	bool clone_flow_key;
1177
1178	/* The first netlink attribute in 'attr' is always
1179	 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1180	 */
1181	cpl_arg = nla_data(attr);
1182	arg = nla_data(cpl_arg);
1183
1184	len = ovs_cb->mru ? ovs_cb->mru + skb->mac_len : skb->len;
1185	max_len = arg->pkt_len;
1186
1187	if ((skb_is_gso(skb) && skb_gso_validate_mac_len(skb, max_len)) ||
1188	    len <= max_len) {
1189		/* Second netlink attribute in 'attr' is always
1190		 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1191		 */
1192		actions = nla_next(cpl_arg, &rem);
1193		clone_flow_key = !arg->exec_for_lesser_equal;
1194	} else {
1195		/* Third netlink attribute in 'attr' is always
1196		 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1197		 */
1198		actions = nla_next(cpl_arg, &rem);
1199		actions = nla_next(actions, &rem);
1200		clone_flow_key = !arg->exec_for_greater;
1201	}
1202
1203	return clone_execute(dp, skb, key, 0, nla_data(actions),
1204			     nla_len(actions), last, clone_flow_key);
1205}
1206
1207static int execute_dec_ttl(struct sk_buff *skb, struct sw_flow_key *key)
1208{
1209	int err;
1210
1211	if (skb->protocol == htons(ETH_P_IPV6)) {
1212		struct ipv6hdr *nh;
1213
1214		err = skb_ensure_writable(skb, skb_network_offset(skb) +
1215					  sizeof(*nh));
1216		if (unlikely(err))
1217			return err;
1218
1219		nh = ipv6_hdr(skb);
1220
1221		if (nh->hop_limit <= 1)
1222			return -EHOSTUNREACH;
1223
1224		key->ip.ttl = --nh->hop_limit;
1225	} else {
1226		struct iphdr *nh;
1227		u8 old_ttl;
1228
1229		err = skb_ensure_writable(skb, skb_network_offset(skb) +
1230					  sizeof(*nh));
1231		if (unlikely(err))
1232			return err;
1233
1234		nh = ip_hdr(skb);
1235		if (nh->ttl <= 1)
1236			return -EHOSTUNREACH;
1237
1238		old_ttl = nh->ttl--;
1239		csum_replace2(&nh->check, htons(old_ttl << 8),
1240			      htons(nh->ttl << 8));
1241		key->ip.ttl = nh->ttl;
1242	}
1243	return 0;
1244}
1245
1246/* Execute a list of actions against 'skb'. */
1247static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1248			      struct sw_flow_key *key,
1249			      const struct nlattr *attr, int len)
1250{
1251	const struct nlattr *a;
1252	int rem;
1253
1254	for (a = attr, rem = len; rem > 0;
1255	     a = nla_next(a, &rem)) {
1256		int err = 0;
1257
1258		switch (nla_type(a)) {
1259		case OVS_ACTION_ATTR_OUTPUT: {
1260			int port = nla_get_u32(a);
1261			struct sk_buff *clone;
1262
1263			/* Every output action needs a separate clone
1264			 * of 'skb', In case the output action is the
1265			 * last action, cloning can be avoided.
1266			 */
1267			if (nla_is_last(a, rem)) {
1268				do_output(dp, skb, port, key);
1269				/* 'skb' has been used for output.
1270				 */
1271				return 0;
1272			}
1273
1274			clone = skb_clone(skb, GFP_ATOMIC);
1275			if (clone)
1276				do_output(dp, clone, port, key);
1277			OVS_CB(skb)->cutlen = 0;
1278			break;
1279		}
1280
1281		case OVS_ACTION_ATTR_TRUNC: {
1282			struct ovs_action_trunc *trunc = nla_data(a);
1283
1284			if (skb->len > trunc->max_len)
1285				OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1286			break;
1287		}
1288
1289		case OVS_ACTION_ATTR_USERSPACE:
1290			output_userspace(dp, skb, key, a, attr,
1291						     len, OVS_CB(skb)->cutlen);
1292			OVS_CB(skb)->cutlen = 0;
1293			break;
1294
1295		case OVS_ACTION_ATTR_HASH:
1296			execute_hash(skb, key, a);
1297			break;
1298
1299		case OVS_ACTION_ATTR_PUSH_MPLS: {
1300			struct ovs_action_push_mpls *mpls = nla_data(a);
1301
1302			err = push_mpls(skb, key, mpls->mpls_lse,
1303					mpls->mpls_ethertype, skb->mac_len);
1304			break;
1305		}
1306		case OVS_ACTION_ATTR_ADD_MPLS: {
1307			struct ovs_action_add_mpls *mpls = nla_data(a);
1308			__u16 mac_len = 0;
1309
1310			if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK)
1311				mac_len = skb->mac_len;
1312
1313			err = push_mpls(skb, key, mpls->mpls_lse,
1314					mpls->mpls_ethertype, mac_len);
1315			break;
1316		}
1317		case OVS_ACTION_ATTR_POP_MPLS:
1318			err = pop_mpls(skb, key, nla_get_be16(a));
1319			break;
1320
1321		case OVS_ACTION_ATTR_PUSH_VLAN:
1322			err = push_vlan(skb, key, nla_data(a));
1323			break;
1324
1325		case OVS_ACTION_ATTR_POP_VLAN:
1326			err = pop_vlan(skb, key);
1327			break;
1328
1329		case OVS_ACTION_ATTR_RECIRC: {
1330			bool last = nla_is_last(a, rem);
1331
1332			err = execute_recirc(dp, skb, key, a, last);
1333			if (last) {
1334				/* If this is the last action, the skb has
1335				 * been consumed or freed.
1336				 * Return immediately.
1337				 */
1338				return err;
1339			}
1340			break;
1341		}
1342
1343		case OVS_ACTION_ATTR_SET:
1344			err = execute_set_action(skb, key, nla_data(a));
1345			break;
1346
1347		case OVS_ACTION_ATTR_SET_MASKED:
1348		case OVS_ACTION_ATTR_SET_TO_MASKED:
1349			err = execute_masked_set_action(skb, key, nla_data(a));
1350			break;
1351
1352		case OVS_ACTION_ATTR_SAMPLE: {
1353			bool last = nla_is_last(a, rem);
1354
1355			err = sample(dp, skb, key, a, last);
1356			if (last)
1357				return err;
1358
1359			break;
1360		}
1361
1362		case OVS_ACTION_ATTR_CT:
1363			if (!is_flow_key_valid(key)) {
1364				err = ovs_flow_key_update(skb, key);
1365				if (err)
1366					return err;
1367			}
1368
1369			err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1370					     nla_data(a));
1371
1372			/* Hide stolen IP fragments from user space. */
1373			if (err)
1374				return err == -EINPROGRESS ? 0 : err;
1375			break;
1376
1377		case OVS_ACTION_ATTR_CT_CLEAR:
1378			err = ovs_ct_clear(skb, key);
1379			break;
1380
1381		case OVS_ACTION_ATTR_PUSH_ETH:
1382			err = push_eth(skb, key, nla_data(a));
1383			break;
1384
1385		case OVS_ACTION_ATTR_POP_ETH:
1386			err = pop_eth(skb, key);
1387			break;
1388
1389		case OVS_ACTION_ATTR_PUSH_NSH: {
1390			u8 buffer[NSH_HDR_MAX_LEN];
1391			struct nshhdr *nh = (struct nshhdr *)buffer;
1392
1393			err = nsh_hdr_from_nlattr(nla_data(a), nh,
1394						  NSH_HDR_MAX_LEN);
1395			if (unlikely(err))
1396				break;
1397			err = push_nsh(skb, key, nh);
1398			break;
1399		}
1400
1401		case OVS_ACTION_ATTR_POP_NSH:
1402			err = pop_nsh(skb, key);
1403			break;
1404
1405		case OVS_ACTION_ATTR_METER:
1406			if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1407				consume_skb(skb);
1408				return 0;
1409			}
1410			break;
1411
1412		case OVS_ACTION_ATTR_CLONE: {
1413			bool last = nla_is_last(a, rem);
1414
1415			err = clone(dp, skb, key, a, last);
1416			if (last)
1417				return err;
1418
1419			break;
1420		}
1421
1422		case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1423			bool last = nla_is_last(a, rem);
1424
1425			err = execute_check_pkt_len(dp, skb, key, a, last);
1426			if (last)
1427				return err;
1428
1429			break;
1430		}
1431
1432		case OVS_ACTION_ATTR_DEC_TTL:
1433			err = execute_dec_ttl(skb, key);
1434			if (err == -EHOSTUNREACH) {
1435				err = dec_ttl_exception_handler(dp, skb, key,
1436								a, true);
1437				return err;
1438			}
1439			break;
1440		}
1441
1442		if (unlikely(err)) {
1443			kfree_skb(skb);
1444			return err;
1445		}
1446	}
1447
1448	consume_skb(skb);
1449	return 0;
1450}
1451
1452/* Execute the actions on the clone of the packet. The effect of the
1453 * execution does not affect the original 'skb' nor the original 'key'.
1454 *
1455 * The execution may be deferred in case the actions can not be executed
1456 * immediately.
1457 */
1458static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1459			 struct sw_flow_key *key, u32 recirc_id,
1460			 const struct nlattr *actions, int len,
1461			 bool last, bool clone_flow_key)
1462{
1463	struct deferred_action *da;
1464	struct sw_flow_key *clone;
1465
1466	skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1467	if (!skb) {
1468		/* Out of memory, skip this action.
1469		 */
1470		return 0;
1471	}
1472
1473	/* When clone_flow_key is false, the 'key' will not be change
1474	 * by the actions, then the 'key' can be used directly.
1475	 * Otherwise, try to clone key from the next recursion level of
1476	 * 'flow_keys'. If clone is successful, execute the actions
1477	 * without deferring.
1478	 */
1479	clone = clone_flow_key ? clone_key(key) : key;
1480	if (clone) {
1481		int err = 0;
1482
1483		if (actions) { /* Sample action */
1484			if (clone_flow_key)
1485				__this_cpu_inc(exec_actions_level);
1486
1487			err = do_execute_actions(dp, skb, clone,
1488						 actions, len);
1489
1490			if (clone_flow_key)
1491				__this_cpu_dec(exec_actions_level);
1492		} else { /* Recirc action */
1493			clone->recirc_id = recirc_id;
1494			ovs_dp_process_packet(skb, clone);
1495		}
1496		return err;
1497	}
1498
1499	/* Out of 'flow_keys' space. Defer actions */
1500	da = add_deferred_actions(skb, key, actions, len);
1501	if (da) {
1502		if (!actions) { /* Recirc action */
1503			key = &da->pkt_key;
1504			key->recirc_id = recirc_id;
1505		}
1506	} else {
1507		/* Out of per CPU action FIFO space. Drop the 'skb' and
1508		 * log an error.
1509		 */
1510		kfree_skb(skb);
1511
1512		if (net_ratelimit()) {
1513			if (actions) { /* Sample action */
1514				pr_warn("%s: deferred action limit reached, drop sample action\n",
1515					ovs_dp_name(dp));
1516			} else {  /* Recirc action */
1517				pr_warn("%s: deferred action limit reached, drop recirc action\n",
1518					ovs_dp_name(dp));
1519			}
1520		}
1521	}
1522	return 0;
1523}
1524
1525static void process_deferred_actions(struct datapath *dp)
1526{
1527	struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1528
1529	/* Do not touch the FIFO in case there is no deferred actions. */
1530	if (action_fifo_is_empty(fifo))
1531		return;
1532
1533	/* Finishing executing all deferred actions. */
1534	do {
1535		struct deferred_action *da = action_fifo_get(fifo);
1536		struct sk_buff *skb = da->skb;
1537		struct sw_flow_key *key = &da->pkt_key;
1538		const struct nlattr *actions = da->actions;
1539		int actions_len = da->actions_len;
1540
1541		if (actions)
1542			do_execute_actions(dp, skb, key, actions, actions_len);
1543		else
1544			ovs_dp_process_packet(skb, key);
1545	} while (!action_fifo_is_empty(fifo));
1546
1547	/* Reset FIFO for the next packet.  */
1548	action_fifo_init(fifo);
1549}
1550
1551/* Execute a list of actions against 'skb'. */
1552int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1553			const struct sw_flow_actions *acts,
1554			struct sw_flow_key *key)
1555{
1556	int err, level;
1557
1558	level = __this_cpu_inc_return(exec_actions_level);
1559	if (unlikely(level > OVS_RECURSION_LIMIT)) {
1560		net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1561				     ovs_dp_name(dp));
1562		kfree_skb(skb);
1563		err = -ENETDOWN;
1564		goto out;
1565	}
1566
1567	OVS_CB(skb)->acts_origlen = acts->orig_len;
1568	err = do_execute_actions(dp, skb, key,
1569				 acts->actions, acts->actions_len);
1570
1571	if (level == 1)
1572		process_deferred_actions(dp);
1573
1574out:
1575	__this_cpu_dec(exec_actions_level);
1576	return err;
1577}
1578
1579int action_fifos_init(void)
1580{
1581	action_fifos = alloc_percpu(struct action_fifo);
1582	if (!action_fifos)
1583		return -ENOMEM;
1584
1585	flow_keys = alloc_percpu(struct action_flow_keys);
1586	if (!flow_keys) {
1587		free_percpu(action_fifos);
1588		return -ENOMEM;
1589	}
1590
1591	return 0;
1592}
1593
1594void action_fifos_exit(void)
1595{
1596	free_percpu(action_fifos);
1597	free_percpu(flow_keys);
1598}