esp6.c

   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * Copyright (C)2002 USAGI/WIDE Project
   4 *
   5 * Authors
   6 *
   7 *	Mitsuru KANDA @USAGI       : IPv6 Support
   8 *	Kazunori MIYAZAWA @USAGI   :
   9 *	Kunihiro Ishiguro <kunihiro@ipinfusion.com>
  10 *
  11 *	This file is derived from net/ipv4/esp.c
  12 */
  13
  14#define pr_fmt(fmt) "IPv6: " fmt
  15
  16#include <crypto/aead.h>
  17#include <crypto/authenc.h>
  18#include <linux/err.h>
  19#include <linux/module.h>
  20#include <net/ip.h>
  21#include <net/xfrm.h>
  22#include <net/esp.h>
  23#include <linux/scatterlist.h>
  24#include <linux/kernel.h>
  25#include <linux/pfkeyv2.h>
  26#include <linux/random.h>
  27#include <linux/slab.h>
  28#include <linux/spinlock.h>
  29#include <net/ip6_checksum.h>
  30#include <net/ip6_route.h>
  31#include <net/icmp.h>
  32#include <net/ipv6.h>
  33#include <net/protocol.h>
  34#include <net/udp.h>
  35#include <linux/icmpv6.h>
  36#include <net/tcp.h>
  37#include <net/espintcp.h>
  38#include <net/inet6_hashtables.h>
  39#include <linux/skbuff_ref.h>
  40
  41#include <linux/highmem.h>
  42
  43struct esp_skb_cb {
  44	struct xfrm_skb_cb xfrm;
  45	void *tmp;
  46};
  47
  48struct esp_output_extra {
  49	__be32 seqhi;
  50	u32 esphoff;
  51};
  52
  53#define ESP_SKB_CB(__skb) ((struct esp_skb_cb *)&((__skb)->cb[0]))
  54
  55/*
  56 * Allocate an AEAD request structure with extra space for SG and IV.
  57 *
  58 * For alignment considerations the upper 32 bits of the sequence number are
  59 * placed at the front, if present. Followed by the IV, the request and finally
  60 * the SG list.
  61 *
  62 * TODO: Use spare space in skb for this where possible.
  63 */
  64static void *esp_alloc_tmp(struct crypto_aead *aead, int nfrags, int seqihlen)
  65{
  66	unsigned int len;
  67
  68	len = seqihlen;
  69
  70	len += crypto_aead_ivsize(aead);
  71
  72	if (len) {
  73		len += crypto_aead_alignmask(aead) &
  74		       ~(crypto_tfm_ctx_alignment() - 1);
  75		len = ALIGN(len, crypto_tfm_ctx_alignment());
  76	}
  77
  78	len += sizeof(struct aead_request) + crypto_aead_reqsize(aead);
  79	len = ALIGN(len, __alignof__(struct scatterlist));
  80
  81	len += sizeof(struct scatterlist) * nfrags;
  82
  83	return kmalloc(len, GFP_ATOMIC);
  84}
  85
  86static inline void *esp_tmp_extra(void *tmp)
  87{
  88	return PTR_ALIGN(tmp, __alignof__(struct esp_output_extra));
  89}
  90
  91static inline u8 *esp_tmp_iv(struct crypto_aead *aead, void *tmp, int seqhilen)
  92{
  93	return crypto_aead_ivsize(aead) ?
  94	       PTR_ALIGN((u8 *)tmp + seqhilen,
  95			 crypto_aead_alignmask(aead) + 1) : tmp + seqhilen;
  96}
  97
  98static inline struct aead_request *esp_tmp_req(struct crypto_aead *aead, u8 *iv)
  99{
 100	struct aead_request *req;
 101
 102	req = (void *)PTR_ALIGN(iv + crypto_aead_ivsize(aead),
 103				crypto_tfm_ctx_alignment());
 104	aead_request_set_tfm(req, aead);
 105	return req;
 106}
 107
 108static inline struct scatterlist *esp_req_sg(struct crypto_aead *aead,
 109					     struct aead_request *req)
 110{
 111	return (void *)ALIGN((unsigned long)(req + 1) +
 112			     crypto_aead_reqsize(aead),
 113			     __alignof__(struct scatterlist));
 114}
 115
 116static void esp_ssg_unref(struct xfrm_state *x, void *tmp, struct sk_buff *skb)
 117{
 118	struct crypto_aead *aead = x->data;
 119	int extralen = 0;
 120	u8 *iv;
 121	struct aead_request *req;
 122	struct scatterlist *sg;
 123
 124	if (x->props.flags & XFRM_STATE_ESN)
 125		extralen += sizeof(struct esp_output_extra);
 126
 127	iv = esp_tmp_iv(aead, tmp, extralen);
 128	req = esp_tmp_req(aead, iv);
 129
 130	/* Unref skb_frag_pages in the src scatterlist if necessary.
 131	 * Skip the first sg which comes from skb->data.
 132	 */
 133	if (req->src != req->dst)
 134		for (sg = sg_next(req->src); sg; sg = sg_next(sg))
 135			skb_page_unref(page_to_netmem(sg_page(sg)),
 136				       skb->pp_recycle);
 137}
 138
 139#ifdef CONFIG_INET6_ESPINTCP
 140static struct sock *esp6_find_tcp_sk(struct xfrm_state *x)
 141{
 142	struct xfrm_encap_tmpl *encap = x->encap;
 143	struct net *net = xs_net(x);
 144	__be16 sport, dport;
 145	struct sock *sk;
 146
 147	spin_lock_bh(&x->lock);
 148	sport = encap->encap_sport;
 149	dport = encap->encap_dport;
 150	spin_unlock_bh(&x->lock);
 151
 152	sk = __inet6_lookup_established(net, &x->id.daddr.in6, dport,
 153					&x->props.saddr.in6, ntohs(sport), 0, 0);
 154	if (!sk)
 155		return ERR_PTR(-ENOENT);
 156
 157	if (!tcp_is_ulp_esp(sk)) {
 158		sock_put(sk);
 159		return ERR_PTR(-EINVAL);
 160	}
 161
 162	return sk;
 163}
 164
 165static int esp_output_tcp_finish(struct xfrm_state *x, struct sk_buff *skb)
 166{
 167	struct sock *sk;
 168	int err;
 169
 170	rcu_read_lock();
 171
 172	sk = esp6_find_tcp_sk(x);
 173	err = PTR_ERR_OR_ZERO(sk);
 174	if (err) {
 175		kfree_skb(skb);
 176		goto out;
 177	}
 178
 179	bh_lock_sock(sk);
 180	if (sock_owned_by_user(sk))
 181		err = espintcp_queue_out(sk, skb);
 182	else
 183		err = espintcp_push_skb(sk, skb);
 184	bh_unlock_sock(sk);
 185
 186	sock_put(sk);
 187
 188out:
 189	rcu_read_unlock();
 190	return err;
 191}
 192
 193static int esp_output_tcp_encap_cb(struct net *net, struct sock *sk,
 194				   struct sk_buff *skb)
 195{
 196	struct dst_entry *dst = skb_dst(skb);
 197	struct xfrm_state *x = dst->xfrm;
 198
 199	return esp_output_tcp_finish(x, skb);
 200}
 201
 202static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
 203{
 204	int err;
 205
 206	local_bh_disable();
 207	err = xfrm_trans_queue_net(xs_net(x), skb, esp_output_tcp_encap_cb);
 208	local_bh_enable();
 209
 210	/* EINPROGRESS just happens to do the right thing.  It
 211	 * actually means that the skb has been consumed and
 212	 * isn't coming back.
 213	 */
 214	return err ?: -EINPROGRESS;
 215}
 216#else
 217static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
 218{
 219	WARN_ON(1);
 220	return -EOPNOTSUPP;
 221}
 222#endif
 223
 224static void esp_output_encap_csum(struct sk_buff *skb)
 225{
 226	/* UDP encap with IPv6 requires a valid checksum */
 227	if (*skb_mac_header(skb) == IPPROTO_UDP) {
 228		struct udphdr *uh = udp_hdr(skb);
 229		struct ipv6hdr *ip6h = ipv6_hdr(skb);
 230		int len = ntohs(uh->len);
 231		unsigned int offset = skb_transport_offset(skb);
 232		__wsum csum = skb_checksum(skb, offset, skb->len - offset, 0);
 233
 234		uh->check = csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
 235					    len, IPPROTO_UDP, csum);
 236		if (uh->check == 0)
 237			uh->check = CSUM_MANGLED_0;
 238	}
 239}
 240
 241static void esp_output_done(void *data, int err)
 242{
 243	struct sk_buff *skb = data;
 244	struct xfrm_offload *xo = xfrm_offload(skb);
 245	void *tmp;
 246	struct xfrm_state *x;
 247
 248	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
 249		struct sec_path *sp = skb_sec_path(skb);
 250
 251		x = sp->xvec[sp->len - 1];
 252	} else {
 253		x = skb_dst(skb)->xfrm;
 254	}
 255
 256	tmp = ESP_SKB_CB(skb)->tmp;
 257	esp_ssg_unref(x, tmp, skb);
 258	kfree(tmp);
 259
 260	esp_output_encap_csum(skb);
 261
 262	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
 263		if (err) {
 264			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
 265			kfree_skb(skb);
 266			return;
 267		}
 268
 269		skb_push(skb, skb->data - skb_mac_header(skb));
 270		secpath_reset(skb);
 271		xfrm_dev_resume(skb);
 272	} else {
 273		if (!err &&
 274		    x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP) {
 275			err = esp_output_tail_tcp(x, skb);
 276			if (err != -EINPROGRESS)
 277				kfree_skb(skb);
 278		} else {
 279			xfrm_output_resume(skb_to_full_sk(skb), skb, err);
 280		}
 281	}
 282}
 283
 284/* Move ESP header back into place. */
 285static void esp_restore_header(struct sk_buff *skb, unsigned int offset)
 286{
 287	struct ip_esp_hdr *esph = (void *)(skb->data + offset);
 288	void *tmp = ESP_SKB_CB(skb)->tmp;
 289	__be32 *seqhi = esp_tmp_extra(tmp);
 290
 291	esph->seq_no = esph->spi;
 292	esph->spi = *seqhi;
 293}
 294
 295static void esp_output_restore_header(struct sk_buff *skb)
 296{
 297	void *tmp = ESP_SKB_CB(skb)->tmp;
 298	struct esp_output_extra *extra = esp_tmp_extra(tmp);
 299
 300	esp_restore_header(skb, skb_transport_offset(skb) + extra->esphoff -
 301				sizeof(__be32));
 302}
 303
 304static struct ip_esp_hdr *esp_output_set_esn(struct sk_buff *skb,
 305					     struct xfrm_state *x,
 306					     struct ip_esp_hdr *esph,
 307					     struct esp_output_extra *extra)
 308{
 309	/* For ESN we move the header forward by 4 bytes to
 310	 * accommodate the high bits.  We will move it back after
 311	 * encryption.
 312	 */
 313	if ((x->props.flags & XFRM_STATE_ESN)) {
 314		__u32 seqhi;
 315		struct xfrm_offload *xo = xfrm_offload(skb);
 316
 317		if (xo)
 318			seqhi = xo->seq.hi;
 319		else
 320			seqhi = XFRM_SKB_CB(skb)->seq.output.hi;
 321
 322		extra->esphoff = (unsigned char *)esph -
 323				 skb_transport_header(skb);
 324		esph = (struct ip_esp_hdr *)((unsigned char *)esph - 4);
 325		extra->seqhi = esph->spi;
 326		esph->seq_no = htonl(seqhi);
 327	}
 328
 329	esph->spi = x->id.spi;
 330
 331	return esph;
 332}
 333
 334static void esp_output_done_esn(void *data, int err)
 335{
 336	struct sk_buff *skb = data;
 337
 338	esp_output_restore_header(skb);
 339	esp_output_done(data, err);
 340}
 341
 342static struct ip_esp_hdr *esp6_output_udp_encap(struct sk_buff *skb,
 343					       int encap_type,
 344					       struct esp_info *esp,
 345					       __be16 sport,
 346					       __be16 dport)
 347{
 348	struct udphdr *uh;
 349	unsigned int len;
 350
 351	len = skb->len + esp->tailen - skb_transport_offset(skb);
 352	if (len > U16_MAX)
 353		return ERR_PTR(-EMSGSIZE);
 354
 355	uh = (struct udphdr *)esp->esph;
 356	uh->source = sport;
 357	uh->dest = dport;
 358	uh->len = htons(len);
 359	uh->check = 0;
 360
 361	*skb_mac_header(skb) = IPPROTO_UDP;
 362
 363	return (struct ip_esp_hdr *)(uh + 1);
 364}
 365
 366#ifdef CONFIG_INET6_ESPINTCP
 367static struct ip_esp_hdr *esp6_output_tcp_encap(struct xfrm_state *x,
 368						struct sk_buff *skb,
 369						struct esp_info *esp)
 370{
 371	__be16 *lenp = (void *)esp->esph;
 372	struct ip_esp_hdr *esph;
 373	unsigned int len;
 374	struct sock *sk;
 375
 376	len = skb->len + esp->tailen - skb_transport_offset(skb);
 377	if (len > IP_MAX_MTU)
 378		return ERR_PTR(-EMSGSIZE);
 379
 380	rcu_read_lock();
 381	sk = esp6_find_tcp_sk(x);
 382	rcu_read_unlock();
 383
 384	if (IS_ERR(sk))
 385		return ERR_CAST(sk);
 386
 387	sock_put(sk);
 388
 389	*lenp = htons(len);
 390	esph = (struct ip_esp_hdr *)(lenp + 1);
 391
 392	return esph;
 393}
 394#else
 395static struct ip_esp_hdr *esp6_output_tcp_encap(struct xfrm_state *x,
 396						struct sk_buff *skb,
 397						struct esp_info *esp)
 398{
 399	return ERR_PTR(-EOPNOTSUPP);
 400}
 401#endif
 402
 403static int esp6_output_encap(struct xfrm_state *x, struct sk_buff *skb,
 404			    struct esp_info *esp)
 405{
 406	struct xfrm_encap_tmpl *encap = x->encap;
 407	struct ip_esp_hdr *esph;
 408	__be16 sport, dport;
 409	int encap_type;
 410
 411	spin_lock_bh(&x->lock);
 412	sport = encap->encap_sport;
 413	dport = encap->encap_dport;
 414	encap_type = encap->encap_type;
 415	spin_unlock_bh(&x->lock);
 416
 417	switch (encap_type) {
 418	default:
 419	case UDP_ENCAP_ESPINUDP:
 420		esph = esp6_output_udp_encap(skb, encap_type, esp, sport, dport);
 421		break;
 422	case TCP_ENCAP_ESPINTCP:
 423		esph = esp6_output_tcp_encap(x, skb, esp);
 424		break;
 425	}
 426
 427	if (IS_ERR(esph))
 428		return PTR_ERR(esph);
 429
 430	esp->esph = esph;
 431
 432	return 0;
 433}
 434
 435int esp6_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
 436{
 437	u8 *tail;
 438	int nfrags;
 439	int esph_offset;
 440	struct page *page;
 441	struct sk_buff *trailer;
 442	int tailen = esp->tailen;
 443
 444	if (x->encap) {
 445		int err = esp6_output_encap(x, skb, esp);
 446
 447		if (err < 0)
 448			return err;
 449	}
 450
 451	if (ALIGN(tailen, L1_CACHE_BYTES) > PAGE_SIZE ||
 452	    ALIGN(skb->data_len, L1_CACHE_BYTES) > PAGE_SIZE)
 453		goto cow;
 454
 455	if (!skb_cloned(skb)) {
 456		if (tailen <= skb_tailroom(skb)) {
 457			nfrags = 1;
 458			trailer = skb;
 459			tail = skb_tail_pointer(trailer);
 460
 461			goto skip_cow;
 462		} else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
 463			   && !skb_has_frag_list(skb)) {
 464			int allocsize;
 465			struct sock *sk = skb->sk;
 466			struct page_frag *pfrag = &x->xfrag;
 467
 468			esp->inplace = false;
 469
 470			allocsize = ALIGN(tailen, L1_CACHE_BYTES);
 471
 472			spin_lock_bh(&x->lock);
 473
 474			if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
 475				spin_unlock_bh(&x->lock);
 476				goto cow;
 477			}
 478
 479			page = pfrag->page;
 480			get_page(page);
 481
 482			tail = page_address(page) + pfrag->offset;
 483
 484			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
 485
 486			nfrags = skb_shinfo(skb)->nr_frags;
 487
 488			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
 489					     tailen);
 490			skb_shinfo(skb)->nr_frags = ++nfrags;
 491
 492			pfrag->offset = pfrag->offset + allocsize;
 493
 494			spin_unlock_bh(&x->lock);
 495
 496			nfrags++;
 497
 498			skb->len += tailen;
 499			skb->data_len += tailen;
 500			skb->truesize += tailen;
 501			if (sk && sk_fullsock(sk))
 502				refcount_add(tailen, &sk->sk_wmem_alloc);
 503
 504			goto out;
 505		}
 506	}
 507
 508cow:
 509	esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
 510
 511	nfrags = skb_cow_data(skb, tailen, &trailer);
 512	if (nfrags < 0)
 513		goto out;
 514	tail = skb_tail_pointer(trailer);
 515	esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
 516
 517skip_cow:
 518	esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
 519	pskb_put(skb, trailer, tailen);
 520
 521out:
 522	return nfrags;
 523}
 524EXPORT_SYMBOL_GPL(esp6_output_head);
 525
 526int esp6_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
 527{
 528	u8 *iv;
 529	int alen;
 530	void *tmp;
 531	int ivlen;
 532	int assoclen;
 533	int extralen;
 534	struct page *page;
 535	struct ip_esp_hdr *esph;
 536	struct aead_request *req;
 537	struct crypto_aead *aead;
 538	struct scatterlist *sg, *dsg;
 539	struct esp_output_extra *extra;
 540	int err = -ENOMEM;
 541
 542	assoclen = sizeof(struct ip_esp_hdr);
 543	extralen = 0;
 544
 545	if (x->props.flags & XFRM_STATE_ESN) {
 546		extralen += sizeof(*extra);
 547		assoclen += sizeof(__be32);
 548	}
 549
 550	aead = x->data;
 551	alen = crypto_aead_authsize(aead);
 552	ivlen = crypto_aead_ivsize(aead);
 553
 554	tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
 555	if (!tmp)
 556		goto error;
 557
 558	extra = esp_tmp_extra(tmp);
 559	iv = esp_tmp_iv(aead, tmp, extralen);
 560	req = esp_tmp_req(aead, iv);
 561	sg = esp_req_sg(aead, req);
 562
 563	if (esp->inplace)
 564		dsg = sg;
 565	else
 566		dsg = &sg[esp->nfrags];
 567
 568	esph = esp_output_set_esn(skb, x, esp->esph, extra);
 569	esp->esph = esph;
 570
 571	sg_init_table(sg, esp->nfrags);
 572	err = skb_to_sgvec(skb, sg,
 573		           (unsigned char *)esph - skb->data,
 574		           assoclen + ivlen + esp->clen + alen);
 575	if (unlikely(err < 0))
 576		goto error_free;
 577
 578	if (!esp->inplace) {
 579		int allocsize;
 580		struct page_frag *pfrag = &x->xfrag;
 581
 582		allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
 583
 584		spin_lock_bh(&x->lock);
 585		if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
 586			spin_unlock_bh(&x->lock);
 587			goto error_free;
 588		}
 589
 590		skb_shinfo(skb)->nr_frags = 1;
 591
 592		page = pfrag->page;
 593		get_page(page);
 594		/* replace page frags in skb with new page */
 595		__skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
 596		pfrag->offset = pfrag->offset + allocsize;
 597		spin_unlock_bh(&x->lock);
 598
 599		sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
 600		err = skb_to_sgvec(skb, dsg,
 601			           (unsigned char *)esph - skb->data,
 602			           assoclen + ivlen + esp->clen + alen);
 603		if (unlikely(err < 0))
 604			goto error_free;
 605	}
 606
 607	if ((x->props.flags & XFRM_STATE_ESN))
 608		aead_request_set_callback(req, 0, esp_output_done_esn, skb);
 609	else
 610		aead_request_set_callback(req, 0, esp_output_done, skb);
 611
 612	aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
 613	aead_request_set_ad(req, assoclen);
 614
 615	memset(iv, 0, ivlen);
 616	memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
 617	       min(ivlen, 8));
 618
 619	ESP_SKB_CB(skb)->tmp = tmp;
 620	err = crypto_aead_encrypt(req);
 621
 622	switch (err) {
 623	case -EINPROGRESS:
 624		goto error;
 625
 626	case -ENOSPC:
 627		err = NET_XMIT_DROP;
 628		break;
 629
 630	case 0:
 631		if ((x->props.flags & XFRM_STATE_ESN))
 632			esp_output_restore_header(skb);
 633		esp_output_encap_csum(skb);
 634	}
 635
 636	if (sg != dsg)
 637		esp_ssg_unref(x, tmp, skb);
 638
 639	if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
 640		err = esp_output_tail_tcp(x, skb);
 641
 642error_free:
 643	kfree(tmp);
 644error:
 645	return err;
 646}
 647EXPORT_SYMBOL_GPL(esp6_output_tail);
 648
 649static int esp6_output(struct xfrm_state *x, struct sk_buff *skb)
 650{
 651	int alen;
 652	int blksize;
 653	struct ip_esp_hdr *esph;
 654	struct crypto_aead *aead;
 655	struct esp_info esp;
 656
 657	esp.inplace = true;
 658
 659	esp.proto = *skb_mac_header(skb);
 660	*skb_mac_header(skb) = IPPROTO_ESP;
 661
 662	/* skb is pure payload to encrypt */
 663
 664	aead = x->data;
 665	alen = crypto_aead_authsize(aead);
 666
 667	esp.tfclen = 0;
 668	if (x->tfcpad) {
 669		struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
 670		u32 padto;
 671
 672		padto = min(x->tfcpad, xfrm_state_mtu(x, dst->child_mtu_cached));
 673		if (skb->len < padto)
 674			esp.tfclen = padto - skb->len;
 675	}
 676	blksize = ALIGN(crypto_aead_blocksize(aead), 4);
 677	esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
 678	esp.plen = esp.clen - skb->len - esp.tfclen;
 679	esp.tailen = esp.tfclen + esp.plen + alen;
 680
 681	esp.esph = ip_esp_hdr(skb);
 682
 683	esp.nfrags = esp6_output_head(x, skb, &esp);
 684	if (esp.nfrags < 0)
 685		return esp.nfrags;
 686
 687	esph = esp.esph;
 688	esph->spi = x->id.spi;
 689
 690	esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
 691	esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
 692			    ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
 693
 694	skb_push(skb, -skb_network_offset(skb));
 695
 696	return esp6_output_tail(x, skb, &esp);
 697}
 698
 699static inline int esp_remove_trailer(struct sk_buff *skb)
 700{
 701	struct xfrm_state *x = xfrm_input_state(skb);
 702	struct crypto_aead *aead = x->data;
 703	int alen, hlen, elen;
 704	int padlen, trimlen;
 705	__wsum csumdiff;
 706	u8 nexthdr[2];
 707	int ret;
 708
 709	alen = crypto_aead_authsize(aead);
 710	hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
 711	elen = skb->len - hlen;
 712
 713	ret = skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2);
 714	BUG_ON(ret);
 715
 716	ret = -EINVAL;
 717	padlen = nexthdr[0];
 718	if (padlen + 2 + alen >= elen) {
 719		net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
 720				    padlen + 2, elen - alen);
 721		goto out;
 722	}
 723
 724	trimlen = alen + padlen + 2;
 725	if (skb->ip_summed == CHECKSUM_COMPLETE) {
 726		csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
 727		skb->csum = csum_block_sub(skb->csum, csumdiff,
 728					   skb->len - trimlen);
 729	}
 730	ret = pskb_trim(skb, skb->len - trimlen);
 731	if (unlikely(ret))
 732		return ret;
 733
 734	ret = nexthdr[1];
 735
 736out:
 737	return ret;
 738}
 739
 740int esp6_input_done2(struct sk_buff *skb, int err)
 741{
 742	struct xfrm_state *x = xfrm_input_state(skb);
 743	struct xfrm_offload *xo = xfrm_offload(skb);
 744	struct crypto_aead *aead = x->data;
 745	int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
 746	int hdr_len = skb_network_header_len(skb);
 747
 748	if (!xo || !(xo->flags & CRYPTO_DONE))
 749		kfree(ESP_SKB_CB(skb)->tmp);
 750
 751	if (unlikely(err))
 752		goto out;
 753
 754	err = esp_remove_trailer(skb);
 755	if (unlikely(err < 0))
 756		goto out;
 757
 758	if (x->encap) {
 759		const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 760		int offset = skb_network_offset(skb) + sizeof(*ip6h);
 761		struct xfrm_encap_tmpl *encap = x->encap;
 762		u8 nexthdr = ip6h->nexthdr;
 763		__be16 frag_off, source;
 764		struct udphdr *uh;
 765		struct tcphdr *th;
 766
 767		offset = ipv6_skip_exthdr(skb, offset, &nexthdr, &frag_off);
 768		if (offset == -1) {
 769			err = -EINVAL;
 770			goto out;
 771		}
 772
 773		uh = (void *)(skb->data + offset);
 774		th = (void *)(skb->data + offset);
 775		hdr_len += offset;
 776
 777		switch (x->encap->encap_type) {
 778		case TCP_ENCAP_ESPINTCP:
 779			source = th->source;
 780			break;
 781		case UDP_ENCAP_ESPINUDP:
 782			source = uh->source;
 783			break;
 784		default:
 785			WARN_ON_ONCE(1);
 786			err = -EINVAL;
 787			goto out;
 788		}
 789
 790		/*
 791		 * 1) if the NAT-T peer's IP or port changed then
 792		 *    advertise the change to the keying daemon.
 793		 *    This is an inbound SA, so just compare
 794		 *    SRC ports.
 795		 */
 796		if (!ipv6_addr_equal(&ip6h->saddr, &x->props.saddr.in6) ||
 797		    source != encap->encap_sport) {
 798			xfrm_address_t ipaddr;
 799
 800			memcpy(&ipaddr.a6, &ip6h->saddr.s6_addr, sizeof(ipaddr.a6));
 801			km_new_mapping(x, &ipaddr, source);
 802
 803			/* XXX: perhaps add an extra
 804			 * policy check here, to see
 805			 * if we should allow or
 806			 * reject a packet from a
 807			 * different source
 808			 * address/port.
 809			 */
 810		}
 811
 812		/*
 813		 * 2) ignore UDP/TCP checksums in case
 814		 *    of NAT-T in Transport Mode, or
 815		 *    perform other post-processing fixes
 816		 *    as per draft-ietf-ipsec-udp-encaps-06,
 817		 *    section 3.1.2
 818		 */
 819		if (x->props.mode == XFRM_MODE_TRANSPORT)
 820			skb->ip_summed = CHECKSUM_UNNECESSARY;
 821	}
 822
 823	skb_postpull_rcsum(skb, skb_network_header(skb),
 824			   skb_network_header_len(skb));
 825	skb_pull_rcsum(skb, hlen);
 826	if (x->props.mode == XFRM_MODE_TUNNEL ||
 827	    x->props.mode == XFRM_MODE_IPTFS)
 828		skb_reset_transport_header(skb);
 829	else
 830		skb_set_transport_header(skb, -hdr_len);
 831
 832	/* RFC4303: Drop dummy packets without any error */
 833	if (err == IPPROTO_NONE)
 834		err = -EINVAL;
 835
 836out:
 837	return err;
 838}
 839EXPORT_SYMBOL_GPL(esp6_input_done2);
 840
 841static void esp_input_done(void *data, int err)
 842{
 843	struct sk_buff *skb = data;
 844
 845	xfrm_input_resume(skb, esp6_input_done2(skb, err));
 846}
 847
 848static void esp_input_restore_header(struct sk_buff *skb)
 849{
 850	esp_restore_header(skb, 0);
 851	__skb_pull(skb, 4);
 852}
 853
 854static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
 855{
 856	struct xfrm_state *x = xfrm_input_state(skb);
 857
 858	/* For ESN we move the header forward by 4 bytes to
 859	 * accommodate the high bits.  We will move it back after
 860	 * decryption.
 861	 */
 862	if ((x->props.flags & XFRM_STATE_ESN)) {
 863		struct ip_esp_hdr *esph = skb_push(skb, 4);
 864
 865		*seqhi = esph->spi;
 866		esph->spi = esph->seq_no;
 867		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
 868	}
 869}
 870
 871static void esp_input_done_esn(void *data, int err)
 872{
 873	struct sk_buff *skb = data;
 874
 875	esp_input_restore_header(skb);
 876	esp_input_done(data, err);
 877}
 878
 879static int esp6_input(struct xfrm_state *x, struct sk_buff *skb)
 880{
 881	struct crypto_aead *aead = x->data;
 882	struct aead_request *req;
 883	struct sk_buff *trailer;
 884	int ivlen = crypto_aead_ivsize(aead);
 885	int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
 886	int nfrags;
 887	int assoclen;
 888	int seqhilen;
 889	int ret = 0;
 890	void *tmp;
 891	__be32 *seqhi;
 892	u8 *iv;
 893	struct scatterlist *sg;
 894
 895	if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen)) {
 896		ret = -EINVAL;
 897		goto out;
 898	}
 899
 900	if (elen <= 0) {
 901		ret = -EINVAL;
 902		goto out;
 903	}
 904
 905	assoclen = sizeof(struct ip_esp_hdr);
 906	seqhilen = 0;
 907
 908	if (x->props.flags & XFRM_STATE_ESN) {
 909		seqhilen += sizeof(__be32);
 910		assoclen += seqhilen;
 911	}
 912
 913	if (!skb_cloned(skb)) {
 914		if (!skb_is_nonlinear(skb)) {
 915			nfrags = 1;
 916
 917			goto skip_cow;
 918		} else if (!skb_has_frag_list(skb)) {
 919			nfrags = skb_shinfo(skb)->nr_frags;
 920			nfrags++;
 921
 922			goto skip_cow;
 923		}
 924	}
 925
 926	nfrags = skb_cow_data(skb, 0, &trailer);
 927	if (nfrags < 0) {
 928		ret = -EINVAL;
 929		goto out;
 930	}
 931
 932skip_cow:
 933	ret = -ENOMEM;
 934	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
 935	if (!tmp)
 936		goto out;
 937
 938	ESP_SKB_CB(skb)->tmp = tmp;
 939	seqhi = esp_tmp_extra(tmp);
 940	iv = esp_tmp_iv(aead, tmp, seqhilen);
 941	req = esp_tmp_req(aead, iv);
 942	sg = esp_req_sg(aead, req);
 943
 944	esp_input_set_header(skb, seqhi);
 945
 946	sg_init_table(sg, nfrags);
 947	ret = skb_to_sgvec(skb, sg, 0, skb->len);
 948	if (unlikely(ret < 0)) {
 949		kfree(tmp);
 950		goto out;
 951	}
 952
 953	skb->ip_summed = CHECKSUM_NONE;
 954
 955	if ((x->props.flags & XFRM_STATE_ESN))
 956		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
 957	else
 958		aead_request_set_callback(req, 0, esp_input_done, skb);
 959
 960	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
 961	aead_request_set_ad(req, assoclen);
 962
 963	ret = crypto_aead_decrypt(req);
 964	if (ret == -EINPROGRESS)
 965		goto out;
 966
 967	if ((x->props.flags & XFRM_STATE_ESN))
 968		esp_input_restore_header(skb);
 969
 970	ret = esp6_input_done2(skb, ret);
 971
 972out:
 973	return ret;
 974}
 975
 976static int esp6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 977		    u8 type, u8 code, int offset, __be32 info)
 978{
 979	struct net *net = dev_net(skb->dev);
 980	const struct ipv6hdr *iph = (const struct ipv6hdr *)skb->data;
 981	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data + offset);
 982	struct xfrm_state *x;
 983
 984	if (type != ICMPV6_PKT_TOOBIG &&
 985	    type != NDISC_REDIRECT)
 986		return 0;
 987
 988	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
 989			      esph->spi, IPPROTO_ESP, AF_INET6);
 990	if (!x)
 991		return 0;
 992
 993	if (type == NDISC_REDIRECT)
 994		ip6_redirect(skb, net, skb->dev->ifindex, 0,
 995			     sock_net_uid(net, NULL));
 996	else
 997		ip6_update_pmtu(skb, net, info, 0, 0, sock_net_uid(net, NULL));
 998	xfrm_state_put(x);
 999
1000	return 0;
1001}
1002
1003static void esp6_destroy(struct xfrm_state *x)
1004{
1005	struct crypto_aead *aead = x->data;
1006
1007	if (!aead)
1008		return;
1009
1010	crypto_free_aead(aead);
1011}
1012
1013static int esp_init_aead(struct xfrm_state *x, struct netlink_ext_ack *extack)
1014{
1015	char aead_name[CRYPTO_MAX_ALG_NAME];
1016	struct crypto_aead *aead;
1017	int err;
1018
1019	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1020		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME) {
1021		NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1022		return -ENAMETOOLONG;
1023	}
1024
1025	aead = crypto_alloc_aead(aead_name, 0, 0);
1026	err = PTR_ERR(aead);
1027	if (IS_ERR(aead))
1028		goto error;
1029
1030	x->data = aead;
1031
1032	err = crypto_aead_setkey(aead, x->aead->alg_key,
1033				 (x->aead->alg_key_len + 7) / 8);
1034	if (err)
1035		goto error;
1036
1037	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1038	if (err)
1039		goto error;
1040
1041	return 0;
1042
1043error:
1044	NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1045	return err;
1046}
1047
1048static int esp_init_authenc(struct xfrm_state *x,
1049			    struct netlink_ext_ack *extack)
1050{
1051	struct crypto_aead *aead;
1052	struct crypto_authenc_key_param *param;
1053	struct rtattr *rta;
1054	char *key;
1055	char *p;
1056	char authenc_name[CRYPTO_MAX_ALG_NAME];
1057	unsigned int keylen;
1058	int err;
1059
1060	err = -ENAMETOOLONG;
1061
1062	if ((x->props.flags & XFRM_STATE_ESN)) {
1063		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1064			     "%s%sauthencesn(%s,%s)%s",
1065			     x->geniv ?: "", x->geniv ? "(" : "",
1066			     x->aalg ? x->aalg->alg_name : "digest_null",
1067			     x->ealg->alg_name,
1068			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1069			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1070			goto error;
1071		}
1072	} else {
1073		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1074			     "%s%sauthenc(%s,%s)%s",
1075			     x->geniv ?: "", x->geniv ? "(" : "",
1076			     x->aalg ? x->aalg->alg_name : "digest_null",
1077			     x->ealg->alg_name,
1078			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1079			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1080			goto error;
1081		}
1082	}
1083
1084	aead = crypto_alloc_aead(authenc_name, 0, 0);
1085	err = PTR_ERR(aead);
1086	if (IS_ERR(aead)) {
1087		NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1088		goto error;
1089	}
1090
1091	x->data = aead;
1092
1093	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1094		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1095	err = -ENOMEM;
1096	key = kmalloc(keylen, GFP_KERNEL);
1097	if (!key)
1098		goto error;
1099
1100	p = key;
1101	rta = (void *)p;
1102	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1103	rta->rta_len = RTA_LENGTH(sizeof(*param));
1104	param = RTA_DATA(rta);
1105	p += RTA_SPACE(sizeof(*param));
1106
1107	if (x->aalg) {
1108		struct xfrm_algo_desc *aalg_desc;
1109
1110		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1111		p += (x->aalg->alg_key_len + 7) / 8;
1112
1113		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1114		BUG_ON(!aalg_desc);
1115
1116		err = -EINVAL;
1117		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1118		    crypto_aead_authsize(aead)) {
1119			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1120			goto free_key;
1121		}
1122
1123		err = crypto_aead_setauthsize(
1124			aead, x->aalg->alg_trunc_len / 8);
1125		if (err) {
1126			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1127			goto free_key;
1128		}
1129	}
1130
1131	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1132	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1133
1134	err = crypto_aead_setkey(aead, key, keylen);
1135
1136free_key:
1137	kfree(key);
1138
1139error:
1140	return err;
1141}
1142
1143static int esp6_init_state(struct xfrm_state *x, struct netlink_ext_ack *extack)
1144{
1145	struct crypto_aead *aead;
1146	u32 align;
1147	int err;
1148
1149	x->data = NULL;
1150
1151	if (x->aead) {
1152		err = esp_init_aead(x, extack);
1153	} else if (x->ealg) {
1154		err = esp_init_authenc(x, extack);
1155	} else {
1156		NL_SET_ERR_MSG(extack, "ESP: AEAD or CRYPT must be provided");
1157		err = -EINVAL;
1158	}
1159
1160	if (err)
1161		goto error;
1162
1163	aead = x->data;
1164
1165	x->props.header_len = sizeof(struct ip_esp_hdr) +
1166			      crypto_aead_ivsize(aead);
1167	switch (x->props.mode) {
1168	case XFRM_MODE_BEET:
1169		if (x->sel.family != AF_INET6)
1170			x->props.header_len += IPV4_BEET_PHMAXLEN +
1171					       (sizeof(struct ipv6hdr) - sizeof(struct iphdr));
1172		break;
1173	default:
1174	case XFRM_MODE_TRANSPORT:
1175		break;
1176	case XFRM_MODE_TUNNEL:
1177		x->props.header_len += sizeof(struct ipv6hdr);
1178		break;
1179	}
1180
1181	if (x->encap) {
1182		struct xfrm_encap_tmpl *encap = x->encap;
1183
1184		switch (encap->encap_type) {
1185		default:
1186			NL_SET_ERR_MSG(extack, "Unsupported encapsulation type for ESP");
1187			err = -EINVAL;
1188			goto error;
1189		case UDP_ENCAP_ESPINUDP:
1190			x->props.header_len += sizeof(struct udphdr);
1191			break;
1192#ifdef CONFIG_INET6_ESPINTCP
1193		case TCP_ENCAP_ESPINTCP:
1194			/* only the length field, TCP encap is done by
1195			 * the socket
1196			 */
1197			x->props.header_len += 2;
1198			break;
1199#endif
1200		}
1201	}
1202
1203	align = ALIGN(crypto_aead_blocksize(aead), 4);
1204	x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1205
1206error:
1207	return err;
1208}
1209
1210static int esp6_rcv_cb(struct sk_buff *skb, int err)
1211{
1212	return 0;
1213}
1214
1215static const struct xfrm_type esp6_type = {
1216	.owner		= THIS_MODULE,
1217	.proto		= IPPROTO_ESP,
1218	.flags		= XFRM_TYPE_REPLAY_PROT,
1219	.init_state	= esp6_init_state,
1220	.destructor	= esp6_destroy,
1221	.input		= esp6_input,
1222	.output		= esp6_output,
1223};
1224
1225static struct xfrm6_protocol esp6_protocol = {
1226	.handler	=	xfrm6_rcv,
1227	.input_handler	=	xfrm_input,
1228	.cb_handler	=	esp6_rcv_cb,
1229	.err_handler	=	esp6_err,
1230	.priority	=	0,
1231};
1232
1233static int __init esp6_init(void)
1234{
1235	if (xfrm_register_type(&esp6_type, AF_INET6) < 0) {
1236		pr_info("%s: can't add xfrm type\n", __func__);
1237		return -EAGAIN;
1238	}
1239	if (xfrm6_protocol_register(&esp6_protocol, IPPROTO_ESP) < 0) {
1240		pr_info("%s: can't add protocol\n", __func__);
1241		xfrm_unregister_type(&esp6_type, AF_INET6);
1242		return -EAGAIN;
1243	}
1244
1245	return 0;
1246}
1247
1248static void __exit esp6_fini(void)
1249{
1250	if (xfrm6_protocol_deregister(&esp6_protocol, IPPROTO_ESP) < 0)
1251		pr_info("%s: can't remove protocol\n", __func__);
1252	xfrm_unregister_type(&esp6_type, AF_INET6);
1253}
1254
1255module_init(esp6_init);
1256module_exit(esp6_fini);
1257
1258MODULE_DESCRIPTION("IPv6 ESP transformation helpers");
1259MODULE_LICENSE("GPL");
1260MODULE_ALIAS_XFRM_TYPE(AF_INET6, XFRM_PROTO_ESP);