esp4.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2#define pr_fmt(fmt) "IPsec: " fmt
   3
   4#include <crypto/aead.h>
   5#include <crypto/authenc.h>
   6#include <linux/err.h>
   7#include <linux/module.h>
   8#include <net/ip.h>
   9#include <net/xfrm.h>
  10#include <net/esp.h>
  11#include <linux/scatterlist.h>
  12#include <linux/kernel.h>
  13#include <linux/pfkeyv2.h>
  14#include <linux/rtnetlink.h>
  15#include <linux/slab.h>
  16#include <linux/spinlock.h>
  17#include <linux/in6.h>
  18#include <net/icmp.h>
  19#include <net/protocol.h>
  20#include <net/udp.h>
  21#include <net/tcp.h>
  22#include <net/espintcp.h>
  23#include <linux/skbuff_ref.h>
  24
  25#include <linux/highmem.h>
  26
  27struct esp_skb_cb {
  28	struct xfrm_skb_cb xfrm;
  29	void *tmp;
  30};
  31
  32struct esp_output_extra {
  33	__be32 seqhi;
  34	u32 esphoff;
  35};
  36
  37#define ESP_SKB_CB(__skb) ((struct esp_skb_cb *)&((__skb)->cb[0]))
  38
  39/*
  40 * Allocate an AEAD request structure with extra space for SG and IV.
  41 *
  42 * For alignment considerations the IV is placed at the front, followed
  43 * by the request and finally the SG list.
  44 *
  45 * TODO: Use spare space in skb for this where possible.
  46 */
  47static void *esp_alloc_tmp(struct crypto_aead *aead, int nfrags, int extralen)
  48{
  49	unsigned int len;
  50
  51	len = extralen;
  52
  53	len += crypto_aead_ivsize(aead);
  54
  55	if (len) {
  56		len += crypto_aead_alignmask(aead) &
  57		       ~(crypto_tfm_ctx_alignment() - 1);
  58		len = ALIGN(len, crypto_tfm_ctx_alignment());
  59	}
  60
  61	len += sizeof(struct aead_request) + crypto_aead_reqsize(aead);
  62	len = ALIGN(len, __alignof__(struct scatterlist));
  63
  64	len += sizeof(struct scatterlist) * nfrags;
  65
  66	return kmalloc(len, GFP_ATOMIC);
  67}
  68
  69static inline void *esp_tmp_extra(void *tmp)
  70{
  71	return PTR_ALIGN(tmp, __alignof__(struct esp_output_extra));
  72}
  73
  74static inline u8 *esp_tmp_iv(struct crypto_aead *aead, void *tmp, int extralen)
  75{
  76	return crypto_aead_ivsize(aead) ?
  77	       PTR_ALIGN((u8 *)tmp + extralen,
  78			 crypto_aead_alignmask(aead) + 1) : tmp + extralen;
  79}
  80
  81static inline struct aead_request *esp_tmp_req(struct crypto_aead *aead, u8 *iv)
  82{
  83	struct aead_request *req;
  84
  85	req = (void *)PTR_ALIGN(iv + crypto_aead_ivsize(aead),
  86				crypto_tfm_ctx_alignment());
  87	aead_request_set_tfm(req, aead);
  88	return req;
  89}
  90
  91static inline struct scatterlist *esp_req_sg(struct crypto_aead *aead,
  92					     struct aead_request *req)
  93{
  94	return (void *)ALIGN((unsigned long)(req + 1) +
  95			     crypto_aead_reqsize(aead),
  96			     __alignof__(struct scatterlist));
  97}
  98
  99static void esp_ssg_unref(struct xfrm_state *x, void *tmp, struct sk_buff *skb)
 100{
 101	struct crypto_aead *aead = x->data;
 102	int extralen = 0;
 103	u8 *iv;
 104	struct aead_request *req;
 105	struct scatterlist *sg;
 106
 107	if (x->props.flags & XFRM_STATE_ESN)
 108		extralen += sizeof(struct esp_output_extra);
 109
 110	iv = esp_tmp_iv(aead, tmp, extralen);
 111	req = esp_tmp_req(aead, iv);
 112
 113	/* Unref skb_frag_pages in the src scatterlist if necessary.
 114	 * Skip the first sg which comes from skb->data.
 115	 */
 116	if (req->src != req->dst)
 117		for (sg = sg_next(req->src); sg; sg = sg_next(sg))
 118			skb_page_unref(page_to_netmem(sg_page(sg)),
 119				       skb->pp_recycle);
 120}
 121
 122#ifdef CONFIG_INET_ESPINTCP
 123static struct sock *esp_find_tcp_sk(struct xfrm_state *x)
 124{
 125	struct xfrm_encap_tmpl *encap = x->encap;
 126	struct net *net = xs_net(x);
 127	__be16 sport, dport;
 128	struct sock *sk;
 129
 130	spin_lock_bh(&x->lock);
 131	sport = encap->encap_sport;
 132	dport = encap->encap_dport;
 133	spin_unlock_bh(&x->lock);
 134
 135	sk = inet_lookup_established(net, x->id.daddr.a4, dport,
 136				     x->props.saddr.a4, sport, 0);
 137	if (!sk)
 138		return ERR_PTR(-ENOENT);
 139
 140	if (!tcp_is_ulp_esp(sk)) {
 141		sock_put(sk);
 142		return ERR_PTR(-EINVAL);
 143	}
 144
 145	return sk;
 146}
 147
 148static int esp_output_tcp_finish(struct xfrm_state *x, struct sk_buff *skb)
 149{
 150	struct sock *sk;
 151	int err;
 152
 153	rcu_read_lock();
 154
 155	sk = esp_find_tcp_sk(x);
 156	err = PTR_ERR_OR_ZERO(sk);
 157	if (err) {
 158		kfree_skb(skb);
 159		goto out;
 160	}
 161
 162	bh_lock_sock(sk);
 163	if (sock_owned_by_user(sk))
 164		err = espintcp_queue_out(sk, skb);
 165	else
 166		err = espintcp_push_skb(sk, skb);
 167	bh_unlock_sock(sk);
 168
 169	sock_put(sk);
 170
 171out:
 172	rcu_read_unlock();
 173	return err;
 174}
 175
 176static int esp_output_tcp_encap_cb(struct net *net, struct sock *sk,
 177				   struct sk_buff *skb)
 178{
 179	struct dst_entry *dst = skb_dst(skb);
 180	struct xfrm_state *x = dst->xfrm;
 181
 182	return esp_output_tcp_finish(x, skb);
 183}
 184
 185static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
 186{
 187	int err;
 188
 189	local_bh_disable();
 190	err = xfrm_trans_queue_net(xs_net(x), skb, esp_output_tcp_encap_cb);
 191	local_bh_enable();
 192
 193	/* EINPROGRESS just happens to do the right thing.  It
 194	 * actually means that the skb has been consumed and
 195	 * isn't coming back.
 196	 */
 197	return err ?: -EINPROGRESS;
 198}
 199#else
 200static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
 201{
 202	WARN_ON(1);
 203	return -EOPNOTSUPP;
 204}
 205#endif
 206
 207static void esp_output_done(void *data, int err)
 208{
 209	struct sk_buff *skb = data;
 210	struct xfrm_offload *xo = xfrm_offload(skb);
 211	void *tmp;
 212	struct xfrm_state *x;
 213
 214	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
 215		struct sec_path *sp = skb_sec_path(skb);
 216
 217		x = sp->xvec[sp->len - 1];
 218	} else {
 219		x = skb_dst(skb)->xfrm;
 220	}
 221
 222	tmp = ESP_SKB_CB(skb)->tmp;
 223	esp_ssg_unref(x, tmp, skb);
 224	kfree(tmp);
 225
 226	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
 227		if (err) {
 228			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
 229			kfree_skb(skb);
 230			return;
 231		}
 232
 233		skb_push(skb, skb->data - skb_mac_header(skb));
 234		secpath_reset(skb);
 235		xfrm_dev_resume(skb);
 236	} else {
 237		if (!err &&
 238		    x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP) {
 239			err = esp_output_tail_tcp(x, skb);
 240			if (err != -EINPROGRESS)
 241				kfree_skb(skb);
 242		} else {
 243			xfrm_output_resume(skb_to_full_sk(skb), skb, err);
 244		}
 245	}
 246}
 247
 248/* Move ESP header back into place. */
 249static void esp_restore_header(struct sk_buff *skb, unsigned int offset)
 250{
 251	struct ip_esp_hdr *esph = (void *)(skb->data + offset);
 252	void *tmp = ESP_SKB_CB(skb)->tmp;
 253	__be32 *seqhi = esp_tmp_extra(tmp);
 254
 255	esph->seq_no = esph->spi;
 256	esph->spi = *seqhi;
 257}
 258
 259static void esp_output_restore_header(struct sk_buff *skb)
 260{
 261	void *tmp = ESP_SKB_CB(skb)->tmp;
 262	struct esp_output_extra *extra = esp_tmp_extra(tmp);
 263
 264	esp_restore_header(skb, skb_transport_offset(skb) + extra->esphoff -
 265				sizeof(__be32));
 266}
 267
 268static struct ip_esp_hdr *esp_output_set_extra(struct sk_buff *skb,
 269					       struct xfrm_state *x,
 270					       struct ip_esp_hdr *esph,
 271					       struct esp_output_extra *extra)
 272{
 273	/* For ESN we move the header forward by 4 bytes to
 274	 * accommodate the high bits.  We will move it back after
 275	 * encryption.
 276	 */
 277	if ((x->props.flags & XFRM_STATE_ESN)) {
 278		__u32 seqhi;
 279		struct xfrm_offload *xo = xfrm_offload(skb);
 280
 281		if (xo)
 282			seqhi = xo->seq.hi;
 283		else
 284			seqhi = XFRM_SKB_CB(skb)->seq.output.hi;
 285
 286		extra->esphoff = (unsigned char *)esph -
 287				 skb_transport_header(skb);
 288		esph = (struct ip_esp_hdr *)((unsigned char *)esph - 4);
 289		extra->seqhi = esph->spi;
 290		esph->seq_no = htonl(seqhi);
 291	}
 292
 293	esph->spi = x->id.spi;
 294
 295	return esph;
 296}
 297
 298static void esp_output_done_esn(void *data, int err)
 299{
 300	struct sk_buff *skb = data;
 301
 302	esp_output_restore_header(skb);
 303	esp_output_done(data, err);
 304}
 305
 306static struct ip_esp_hdr *esp_output_udp_encap(struct sk_buff *skb,
 307					       int encap_type,
 308					       struct esp_info *esp,
 309					       __be16 sport,
 310					       __be16 dport)
 311{
 312	struct udphdr *uh;
 313	unsigned int len;
 314	struct xfrm_offload *xo = xfrm_offload(skb);
 315
 316	len = skb->len + esp->tailen - skb_transport_offset(skb);
 317	if (len + sizeof(struct iphdr) > IP_MAX_MTU)
 318		return ERR_PTR(-EMSGSIZE);
 319
 320	uh = (struct udphdr *)esp->esph;
 321	uh->source = sport;
 322	uh->dest = dport;
 323	uh->len = htons(len);
 324	uh->check = 0;
 325
 326	/* For IPv4 ESP with UDP encapsulation, if xo is not null, the skb is in the crypto offload
 327	 * data path, which means that esp_output_udp_encap is called outside of the XFRM stack.
 328	 * In this case, the mac header doesn't point to the IPv4 protocol field, so don't set it.
 329	 */
 330	if (!xo || encap_type != UDP_ENCAP_ESPINUDP)
 331		*skb_mac_header(skb) = IPPROTO_UDP;
 332
 333	return (struct ip_esp_hdr *)(uh + 1);
 334}
 335
 336#ifdef CONFIG_INET_ESPINTCP
 337static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
 338						    struct sk_buff *skb,
 339						    struct esp_info *esp)
 340{
 341	__be16 *lenp = (void *)esp->esph;
 342	struct ip_esp_hdr *esph;
 343	unsigned int len;
 344	struct sock *sk;
 345
 346	len = skb->len + esp->tailen - skb_transport_offset(skb);
 347	if (len > IP_MAX_MTU)
 348		return ERR_PTR(-EMSGSIZE);
 349
 350	rcu_read_lock();
 351	sk = esp_find_tcp_sk(x);
 352	rcu_read_unlock();
 353
 354	if (IS_ERR(sk))
 355		return ERR_CAST(sk);
 356
 357	sock_put(sk);
 358
 359	*lenp = htons(len);
 360	esph = (struct ip_esp_hdr *)(lenp + 1);
 361
 362	return esph;
 363}
 364#else
 365static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
 366						    struct sk_buff *skb,
 367						    struct esp_info *esp)
 368{
 369	return ERR_PTR(-EOPNOTSUPP);
 370}
 371#endif
 372
 373static int esp_output_encap(struct xfrm_state *x, struct sk_buff *skb,
 374			    struct esp_info *esp)
 375{
 376	struct xfrm_encap_tmpl *encap = x->encap;
 377	struct ip_esp_hdr *esph;
 378	__be16 sport, dport;
 379	int encap_type;
 380
 381	spin_lock_bh(&x->lock);
 382	sport = encap->encap_sport;
 383	dport = encap->encap_dport;
 384	encap_type = encap->encap_type;
 385	spin_unlock_bh(&x->lock);
 386
 387	switch (encap_type) {
 388	default:
 389	case UDP_ENCAP_ESPINUDP:
 390		esph = esp_output_udp_encap(skb, encap_type, esp, sport, dport);
 391		break;
 392	case TCP_ENCAP_ESPINTCP:
 393		esph = esp_output_tcp_encap(x, skb, esp);
 394		break;
 395	}
 396
 397	if (IS_ERR(esph))
 398		return PTR_ERR(esph);
 399
 400	esp->esph = esph;
 401
 402	return 0;
 403}
 404
 405int esp_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
 406{
 407	u8 *tail;
 408	int nfrags;
 409	int esph_offset;
 410	struct page *page;
 411	struct sk_buff *trailer;
 412	int tailen = esp->tailen;
 413
 414	/* this is non-NULL only with TCP/UDP Encapsulation */
 415	if (x->encap) {
 416		int err = esp_output_encap(x, skb, esp);
 417
 418		if (err < 0)
 419			return err;
 420	}
 421
 422	if (ALIGN(tailen, L1_CACHE_BYTES) > PAGE_SIZE ||
 423	    ALIGN(skb->data_len, L1_CACHE_BYTES) > PAGE_SIZE)
 424		goto cow;
 425
 426	if (!skb_cloned(skb)) {
 427		if (tailen <= skb_tailroom(skb)) {
 428			nfrags = 1;
 429			trailer = skb;
 430			tail = skb_tail_pointer(trailer);
 431
 432			goto skip_cow;
 433		} else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
 434			   && !skb_has_frag_list(skb)) {
 435			int allocsize;
 436			struct sock *sk = skb->sk;
 437			struct page_frag *pfrag = &x->xfrag;
 438
 439			esp->inplace = false;
 440
 441			allocsize = ALIGN(tailen, L1_CACHE_BYTES);
 442
 443			spin_lock_bh(&x->lock);
 444
 445			if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
 446				spin_unlock_bh(&x->lock);
 447				goto cow;
 448			}
 449
 450			page = pfrag->page;
 451			get_page(page);
 452
 453			tail = page_address(page) + pfrag->offset;
 454
 455			esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
 456
 457			nfrags = skb_shinfo(skb)->nr_frags;
 458
 459			__skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
 460					     tailen);
 461			skb_shinfo(skb)->nr_frags = ++nfrags;
 462
 463			pfrag->offset = pfrag->offset + allocsize;
 464
 465			spin_unlock_bh(&x->lock);
 466
 467			nfrags++;
 468
 469			skb_len_add(skb, tailen);
 470			if (sk && sk_fullsock(sk))
 471				refcount_add(tailen, &sk->sk_wmem_alloc);
 472
 473			goto out;
 474		}
 475	}
 476
 477cow:
 478	esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
 479
 480	nfrags = skb_cow_data(skb, tailen, &trailer);
 481	if (nfrags < 0)
 482		goto out;
 483	tail = skb_tail_pointer(trailer);
 484	esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
 485
 486skip_cow:
 487	esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
 488	pskb_put(skb, trailer, tailen);
 489
 490out:
 491	return nfrags;
 492}
 493EXPORT_SYMBOL_GPL(esp_output_head);
 494
 495int esp_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
 496{
 497	u8 *iv;
 498	int alen;
 499	void *tmp;
 500	int ivlen;
 501	int assoclen;
 502	int extralen;
 503	struct page *page;
 504	struct ip_esp_hdr *esph;
 505	struct crypto_aead *aead;
 506	struct aead_request *req;
 507	struct scatterlist *sg, *dsg;
 508	struct esp_output_extra *extra;
 509	int err = -ENOMEM;
 510
 511	assoclen = sizeof(struct ip_esp_hdr);
 512	extralen = 0;
 513
 514	if (x->props.flags & XFRM_STATE_ESN) {
 515		extralen += sizeof(*extra);
 516		assoclen += sizeof(__be32);
 517	}
 518
 519	aead = x->data;
 520	alen = crypto_aead_authsize(aead);
 521	ivlen = crypto_aead_ivsize(aead);
 522
 523	tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
 524	if (!tmp)
 525		goto error;
 526
 527	extra = esp_tmp_extra(tmp);
 528	iv = esp_tmp_iv(aead, tmp, extralen);
 529	req = esp_tmp_req(aead, iv);
 530	sg = esp_req_sg(aead, req);
 531
 532	if (esp->inplace)
 533		dsg = sg;
 534	else
 535		dsg = &sg[esp->nfrags];
 536
 537	esph = esp_output_set_extra(skb, x, esp->esph, extra);
 538	esp->esph = esph;
 539
 540	sg_init_table(sg, esp->nfrags);
 541	err = skb_to_sgvec(skb, sg,
 542		           (unsigned char *)esph - skb->data,
 543		           assoclen + ivlen + esp->clen + alen);
 544	if (unlikely(err < 0))
 545		goto error_free;
 546
 547	if (!esp->inplace) {
 548		int allocsize;
 549		struct page_frag *pfrag = &x->xfrag;
 550
 551		allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
 552
 553		spin_lock_bh(&x->lock);
 554		if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
 555			spin_unlock_bh(&x->lock);
 556			goto error_free;
 557		}
 558
 559		skb_shinfo(skb)->nr_frags = 1;
 560
 561		page = pfrag->page;
 562		get_page(page);
 563		/* replace page frags in skb with new page */
 564		__skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
 565		pfrag->offset = pfrag->offset + allocsize;
 566		spin_unlock_bh(&x->lock);
 567
 568		sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
 569		err = skb_to_sgvec(skb, dsg,
 570			           (unsigned char *)esph - skb->data,
 571			           assoclen + ivlen + esp->clen + alen);
 572		if (unlikely(err < 0))
 573			goto error_free;
 574	}
 575
 576	if ((x->props.flags & XFRM_STATE_ESN))
 577		aead_request_set_callback(req, 0, esp_output_done_esn, skb);
 578	else
 579		aead_request_set_callback(req, 0, esp_output_done, skb);
 580
 581	aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
 582	aead_request_set_ad(req, assoclen);
 583
 584	memset(iv, 0, ivlen);
 585	memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
 586	       min(ivlen, 8));
 587
 588	ESP_SKB_CB(skb)->tmp = tmp;
 589	err = crypto_aead_encrypt(req);
 590
 591	switch (err) {
 592	case -EINPROGRESS:
 593		goto error;
 594
 595	case -ENOSPC:
 596		err = NET_XMIT_DROP;
 597		break;
 598
 599	case 0:
 600		if ((x->props.flags & XFRM_STATE_ESN))
 601			esp_output_restore_header(skb);
 602	}
 603
 604	if (sg != dsg)
 605		esp_ssg_unref(x, tmp, skb);
 606
 607	if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
 608		err = esp_output_tail_tcp(x, skb);
 609
 610error_free:
 611	kfree(tmp);
 612error:
 613	return err;
 614}
 615EXPORT_SYMBOL_GPL(esp_output_tail);
 616
 617static int esp_output(struct xfrm_state *x, struct sk_buff *skb)
 618{
 619	int alen;
 620	int blksize;
 621	struct ip_esp_hdr *esph;
 622	struct crypto_aead *aead;
 623	struct esp_info esp;
 624
 625	esp.inplace = true;
 626
 627	esp.proto = *skb_mac_header(skb);
 628	*skb_mac_header(skb) = IPPROTO_ESP;
 629
 630	/* skb is pure payload to encrypt */
 631
 632	aead = x->data;
 633	alen = crypto_aead_authsize(aead);
 634
 635	esp.tfclen = 0;
 636	if (x->tfcpad) {
 637		struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
 638		u32 padto;
 639
 640		padto = min(x->tfcpad, xfrm_state_mtu(x, dst->child_mtu_cached));
 641		if (skb->len < padto)
 642			esp.tfclen = padto - skb->len;
 643	}
 644	blksize = ALIGN(crypto_aead_blocksize(aead), 4);
 645	esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
 646	esp.plen = esp.clen - skb->len - esp.tfclen;
 647	esp.tailen = esp.tfclen + esp.plen + alen;
 648
 649	esp.esph = ip_esp_hdr(skb);
 650
 651	esp.nfrags = esp_output_head(x, skb, &esp);
 652	if (esp.nfrags < 0)
 653		return esp.nfrags;
 654
 655	esph = esp.esph;
 656	esph->spi = x->id.spi;
 657
 658	esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
 659	esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
 660				 ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
 661
 662	skb_push(skb, -skb_network_offset(skb));
 663
 664	return esp_output_tail(x, skb, &esp);
 665}
 666
 667static inline int esp_remove_trailer(struct sk_buff *skb)
 668{
 669	struct xfrm_state *x = xfrm_input_state(skb);
 670	struct crypto_aead *aead = x->data;
 671	int alen, hlen, elen;
 672	int padlen, trimlen;
 673	__wsum csumdiff;
 674	u8 nexthdr[2];
 675	int ret;
 676
 677	alen = crypto_aead_authsize(aead);
 678	hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
 679	elen = skb->len - hlen;
 680
 681	if (skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2))
 682		BUG();
 683
 684	ret = -EINVAL;
 685	padlen = nexthdr[0];
 686	if (padlen + 2 + alen >= elen) {
 687		net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
 688				    padlen + 2, elen - alen);
 689		goto out;
 690	}
 691
 692	trimlen = alen + padlen + 2;
 693	if (skb->ip_summed == CHECKSUM_COMPLETE) {
 694		csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
 695		skb->csum = csum_block_sub(skb->csum, csumdiff,
 696					   skb->len - trimlen);
 697	}
 698	ret = pskb_trim(skb, skb->len - trimlen);
 699	if (unlikely(ret))
 700		return ret;
 701
 702	ret = nexthdr[1];
 703
 704out:
 705	return ret;
 706}
 707
 708int esp_input_done2(struct sk_buff *skb, int err)
 709{
 710	const struct iphdr *iph;
 711	struct xfrm_state *x = xfrm_input_state(skb);
 712	struct xfrm_offload *xo = xfrm_offload(skb);
 713	struct crypto_aead *aead = x->data;
 714	int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
 715	int ihl;
 716
 717	if (!xo || !(xo->flags & CRYPTO_DONE))
 718		kfree(ESP_SKB_CB(skb)->tmp);
 719
 720	if (unlikely(err))
 721		goto out;
 722
 723	err = esp_remove_trailer(skb);
 724	if (unlikely(err < 0))
 725		goto out;
 726
 727	iph = ip_hdr(skb);
 728	ihl = iph->ihl * 4;
 729
 730	if (x->encap) {
 731		struct xfrm_encap_tmpl *encap = x->encap;
 732		struct tcphdr *th = (void *)(skb_network_header(skb) + ihl);
 733		struct udphdr *uh = (void *)(skb_network_header(skb) + ihl);
 734		__be16 source;
 735
 736		switch (x->encap->encap_type) {
 737		case TCP_ENCAP_ESPINTCP:
 738			source = th->source;
 739			break;
 740		case UDP_ENCAP_ESPINUDP:
 741			source = uh->source;
 742			break;
 743		default:
 744			WARN_ON_ONCE(1);
 745			err = -EINVAL;
 746			goto out;
 747		}
 748
 749		/*
 750		 * 1) if the NAT-T peer's IP or port changed then
 751		 *    advertise the change to the keying daemon.
 752		 *    This is an inbound SA, so just compare
 753		 *    SRC ports.
 754		 */
 755		if (iph->saddr != x->props.saddr.a4 ||
 756		    source != encap->encap_sport) {
 757			xfrm_address_t ipaddr;
 758
 759			ipaddr.a4 = iph->saddr;
 760			km_new_mapping(x, &ipaddr, source);
 761
 762			/* XXX: perhaps add an extra
 763			 * policy check here, to see
 764			 * if we should allow or
 765			 * reject a packet from a
 766			 * different source
 767			 * address/port.
 768			 */
 769		}
 770
 771		/*
 772		 * 2) ignore UDP/TCP checksums in case
 773		 *    of NAT-T in Transport Mode, or
 774		 *    perform other post-processing fixes
 775		 *    as per draft-ietf-ipsec-udp-encaps-06,
 776		 *    section 3.1.2
 777		 */
 778		if (x->props.mode == XFRM_MODE_TRANSPORT)
 779			skb->ip_summed = CHECKSUM_UNNECESSARY;
 780	}
 781
 782	skb_pull_rcsum(skb, hlen);
 783	if (x->props.mode == XFRM_MODE_TUNNEL ||
 784	    x->props.mode == XFRM_MODE_IPTFS)
 785		skb_reset_transport_header(skb);
 786	else
 787		skb_set_transport_header(skb, -ihl);
 788
 789	/* RFC4303: Drop dummy packets without any error */
 790	if (err == IPPROTO_NONE)
 791		err = -EINVAL;
 792
 793out:
 794	return err;
 795}
 796EXPORT_SYMBOL_GPL(esp_input_done2);
 797
 798static void esp_input_done(void *data, int err)
 799{
 800	struct sk_buff *skb = data;
 801
 802	xfrm_input_resume(skb, esp_input_done2(skb, err));
 803}
 804
 805static void esp_input_restore_header(struct sk_buff *skb)
 806{
 807	esp_restore_header(skb, 0);
 808	__skb_pull(skb, 4);
 809}
 810
 811static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
 812{
 813	struct xfrm_state *x = xfrm_input_state(skb);
 814	struct ip_esp_hdr *esph;
 815
 816	/* For ESN we move the header forward by 4 bytes to
 817	 * accommodate the high bits.  We will move it back after
 818	 * decryption.
 819	 */
 820	if ((x->props.flags & XFRM_STATE_ESN)) {
 821		esph = skb_push(skb, 4);
 822		*seqhi = esph->spi;
 823		esph->spi = esph->seq_no;
 824		esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
 825	}
 826}
 827
 828static void esp_input_done_esn(void *data, int err)
 829{
 830	struct sk_buff *skb = data;
 831
 832	esp_input_restore_header(skb);
 833	esp_input_done(data, err);
 834}
 835
 836/*
 837 * Note: detecting truncated vs. non-truncated authentication data is very
 838 * expensive, so we only support truncated data, which is the recommended
 839 * and common case.
 840 */
 841static int esp_input(struct xfrm_state *x, struct sk_buff *skb)
 842{
 843	struct crypto_aead *aead = x->data;
 844	struct aead_request *req;
 845	struct sk_buff *trailer;
 846	int ivlen = crypto_aead_ivsize(aead);
 847	int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
 848	int nfrags;
 849	int assoclen;
 850	int seqhilen;
 851	__be32 *seqhi;
 852	void *tmp;
 853	u8 *iv;
 854	struct scatterlist *sg;
 855	int err = -EINVAL;
 856
 857	if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen))
 858		goto out;
 859
 860	if (elen <= 0)
 861		goto out;
 862
 863	assoclen = sizeof(struct ip_esp_hdr);
 864	seqhilen = 0;
 865
 866	if (x->props.flags & XFRM_STATE_ESN) {
 867		seqhilen += sizeof(__be32);
 868		assoclen += seqhilen;
 869	}
 870
 871	if (!skb_cloned(skb)) {
 872		if (!skb_is_nonlinear(skb)) {
 873			nfrags = 1;
 874
 875			goto skip_cow;
 876		} else if (!skb_has_frag_list(skb)) {
 877			nfrags = skb_shinfo(skb)->nr_frags;
 878			nfrags++;
 879
 880			goto skip_cow;
 881		}
 882	}
 883
 884	err = skb_cow_data(skb, 0, &trailer);
 885	if (err < 0)
 886		goto out;
 887
 888	nfrags = err;
 889
 890skip_cow:
 891	err = -ENOMEM;
 892	tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
 893	if (!tmp)
 894		goto out;
 895
 896	ESP_SKB_CB(skb)->tmp = tmp;
 897	seqhi = esp_tmp_extra(tmp);
 898	iv = esp_tmp_iv(aead, tmp, seqhilen);
 899	req = esp_tmp_req(aead, iv);
 900	sg = esp_req_sg(aead, req);
 901
 902	esp_input_set_header(skb, seqhi);
 903
 904	sg_init_table(sg, nfrags);
 905	err = skb_to_sgvec(skb, sg, 0, skb->len);
 906	if (unlikely(err < 0)) {
 907		kfree(tmp);
 908		goto out;
 909	}
 910
 911	skb->ip_summed = CHECKSUM_NONE;
 912
 913	if ((x->props.flags & XFRM_STATE_ESN))
 914		aead_request_set_callback(req, 0, esp_input_done_esn, skb);
 915	else
 916		aead_request_set_callback(req, 0, esp_input_done, skb);
 917
 918	aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
 919	aead_request_set_ad(req, assoclen);
 920
 921	err = crypto_aead_decrypt(req);
 922	if (err == -EINPROGRESS)
 923		goto out;
 924
 925	if ((x->props.flags & XFRM_STATE_ESN))
 926		esp_input_restore_header(skb);
 927
 928	err = esp_input_done2(skb, err);
 929
 930out:
 931	return err;
 932}
 933
 934static int esp4_err(struct sk_buff *skb, u32 info)
 935{
 936	struct net *net = dev_net(skb->dev);
 937	const struct iphdr *iph = (const struct iphdr *)skb->data;
 938	struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data+(iph->ihl<<2));
 939	struct xfrm_state *x;
 940
 941	switch (icmp_hdr(skb)->type) {
 942	case ICMP_DEST_UNREACH:
 943		if (icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
 944			return 0;
 945		break;
 946	case ICMP_REDIRECT:
 947		break;
 948	default:
 949		return 0;
 950	}
 951
 952	x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
 953			      esph->spi, IPPROTO_ESP, AF_INET);
 954	if (!x)
 955		return 0;
 956
 957	if (icmp_hdr(skb)->type == ICMP_DEST_UNREACH)
 958		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ESP);
 959	else
 960		ipv4_redirect(skb, net, 0, IPPROTO_ESP);
 961	xfrm_state_put(x);
 962
 963	return 0;
 964}
 965
 966static void esp_destroy(struct xfrm_state *x)
 967{
 968	struct crypto_aead *aead = x->data;
 969
 970	if (!aead)
 971		return;
 972
 973	crypto_free_aead(aead);
 974}
 975
 976static int esp_init_aead(struct xfrm_state *x, struct netlink_ext_ack *extack)
 977{
 978	char aead_name[CRYPTO_MAX_ALG_NAME];
 979	struct crypto_aead *aead;
 980	int err;
 981
 982	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
 983		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME) {
 984		NL_SET_ERR_MSG(extack, "Algorithm name is too long");
 985		return -ENAMETOOLONG;
 986	}
 987
 988	aead = crypto_alloc_aead(aead_name, 0, 0);
 989	err = PTR_ERR(aead);
 990	if (IS_ERR(aead))
 991		goto error;
 992
 993	x->data = aead;
 994
 995	err = crypto_aead_setkey(aead, x->aead->alg_key,
 996				 (x->aead->alg_key_len + 7) / 8);
 997	if (err)
 998		goto error;
 999
1000	err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1001	if (err)
1002		goto error;
1003
1004	return 0;
1005
1006error:
1007	NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1008	return err;
1009}
1010
1011static int esp_init_authenc(struct xfrm_state *x,
1012			    struct netlink_ext_ack *extack)
1013{
1014	struct crypto_aead *aead;
1015	struct crypto_authenc_key_param *param;
1016	struct rtattr *rta;
1017	char *key;
1018	char *p;
1019	char authenc_name[CRYPTO_MAX_ALG_NAME];
1020	unsigned int keylen;
1021	int err;
1022
1023	err = -ENAMETOOLONG;
1024
1025	if ((x->props.flags & XFRM_STATE_ESN)) {
1026		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1027			     "%s%sauthencesn(%s,%s)%s",
1028			     x->geniv ?: "", x->geniv ? "(" : "",
1029			     x->aalg ? x->aalg->alg_name : "digest_null",
1030			     x->ealg->alg_name,
1031			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1032			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1033			goto error;
1034		}
1035	} else {
1036		if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1037			     "%s%sauthenc(%s,%s)%s",
1038			     x->geniv ?: "", x->geniv ? "(" : "",
1039			     x->aalg ? x->aalg->alg_name : "digest_null",
1040			     x->ealg->alg_name,
1041			     x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1042			NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1043			goto error;
1044		}
1045	}
1046
1047	aead = crypto_alloc_aead(authenc_name, 0, 0);
1048	err = PTR_ERR(aead);
1049	if (IS_ERR(aead)) {
1050		NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1051		goto error;
1052	}
1053
1054	x->data = aead;
1055
1056	keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1057		 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1058	err = -ENOMEM;
1059	key = kmalloc(keylen, GFP_KERNEL);
1060	if (!key)
1061		goto error;
1062
1063	p = key;
1064	rta = (void *)p;
1065	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1066	rta->rta_len = RTA_LENGTH(sizeof(*param));
1067	param = RTA_DATA(rta);
1068	p += RTA_SPACE(sizeof(*param));
1069
1070	if (x->aalg) {
1071		struct xfrm_algo_desc *aalg_desc;
1072
1073		memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1074		p += (x->aalg->alg_key_len + 7) / 8;
1075
1076		aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1077		BUG_ON(!aalg_desc);
1078
1079		err = -EINVAL;
1080		if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1081		    crypto_aead_authsize(aead)) {
1082			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1083			goto free_key;
1084		}
1085
1086		err = crypto_aead_setauthsize(
1087			aead, x->aalg->alg_trunc_len / 8);
1088		if (err) {
1089			NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1090			goto free_key;
1091		}
1092	}
1093
1094	param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1095	memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1096
1097	err = crypto_aead_setkey(aead, key, keylen);
1098
1099free_key:
1100	kfree_sensitive(key);
1101
1102error:
1103	return err;
1104}
1105
1106static int esp_init_state(struct xfrm_state *x, struct netlink_ext_ack *extack)
1107{
1108	struct crypto_aead *aead;
1109	u32 align;
1110	int err;
1111
1112	x->data = NULL;
1113
1114	if (x->aead) {
1115		err = esp_init_aead(x, extack);
1116	} else if (x->ealg) {
1117		err = esp_init_authenc(x, extack);
1118	} else {
1119		NL_SET_ERR_MSG(extack, "ESP: AEAD or CRYPT must be provided");
1120		err = -EINVAL;
1121	}
1122
1123	if (err)
1124		goto error;
1125
1126	aead = x->data;
1127
1128	x->props.header_len = sizeof(struct ip_esp_hdr) +
1129			      crypto_aead_ivsize(aead);
1130	if (x->props.mode == XFRM_MODE_TUNNEL)
1131		x->props.header_len += sizeof(struct iphdr);
1132	else if (x->props.mode == XFRM_MODE_BEET && x->sel.family != AF_INET6)
1133		x->props.header_len += IPV4_BEET_PHMAXLEN;
1134	if (x->encap) {
1135		struct xfrm_encap_tmpl *encap = x->encap;
1136
1137		switch (encap->encap_type) {
1138		default:
1139			NL_SET_ERR_MSG(extack, "Unsupported encapsulation type for ESP");
1140			err = -EINVAL;
1141			goto error;
1142		case UDP_ENCAP_ESPINUDP:
1143			x->props.header_len += sizeof(struct udphdr);
1144			break;
1145#ifdef CONFIG_INET_ESPINTCP
1146		case TCP_ENCAP_ESPINTCP:
1147			/* only the length field, TCP encap is done by
1148			 * the socket
1149			 */
1150			x->props.header_len += 2;
1151			break;
1152#endif
1153		}
1154	}
1155
1156	align = ALIGN(crypto_aead_blocksize(aead), 4);
1157	x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1158
1159error:
1160	return err;
1161}
1162
1163static int esp4_rcv_cb(struct sk_buff *skb, int err)
1164{
1165	return 0;
1166}
1167
1168static const struct xfrm_type esp_type =
1169{
1170	.owner		= THIS_MODULE,
1171	.proto	     	= IPPROTO_ESP,
1172	.flags		= XFRM_TYPE_REPLAY_PROT,
1173	.init_state	= esp_init_state,
1174	.destructor	= esp_destroy,
1175	.input		= esp_input,
1176	.output		= esp_output,
1177};
1178
1179static struct xfrm4_protocol esp4_protocol = {
1180	.handler	=	xfrm4_rcv,
1181	.input_handler	=	xfrm_input,
1182	.cb_handler	=	esp4_rcv_cb,
1183	.err_handler	=	esp4_err,
1184	.priority	=	0,
1185};
1186
1187static int __init esp4_init(void)
1188{
1189	if (xfrm_register_type(&esp_type, AF_INET) < 0) {
1190		pr_info("%s: can't add xfrm type\n", __func__);
1191		return -EAGAIN;
1192	}
1193	if (xfrm4_protocol_register(&esp4_protocol, IPPROTO_ESP) < 0) {
1194		pr_info("%s: can't add protocol\n", __func__);
1195		xfrm_unregister_type(&esp_type, AF_INET);
1196		return -EAGAIN;
1197	}
1198	return 0;
1199}
1200
1201static void __exit esp4_fini(void)
1202{
1203	if (xfrm4_protocol_deregister(&esp4_protocol, IPPROTO_ESP) < 0)
1204		pr_info("%s: can't remove protocol\n", __func__);
1205	xfrm_unregister_type(&esp_type, AF_INET);
1206}
1207
1208module_init(esp4_init);
1209module_exit(esp4_fini);
1210MODULE_DESCRIPTION("IPv4 ESP transformation library");
1211MODULE_LICENSE("GPL");
1212MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_ESP);