1 // SPDX-License-Identifier: GPL-2.0-only
2 #define pr_fmt(fmt) "IPsec: " fmt
4 #include <crypto/aead.h>
5 #include <crypto/authenc.h>
7 #include <linux/module.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>
19 #include <net/protocol.h>
22 #include <net/espintcp.h>
24 #include <linux/highmem.h>
27 struct xfrm_skb_cb xfrm;
31 struct esp_output_extra {
36 #define ESP_SKB_CB(__skb) ((struct esp_skb_cb *)&((__skb)->cb[0]))
39 * Allocate an AEAD request structure with extra space for SG and IV.
41 * For alignment considerations the IV is placed at the front, followed
42 * by the request and finally the SG list.
44 * TODO: Use spare space in skb for this where possible.
46 static void *esp_alloc_tmp(struct crypto_aead *aead, int nfrags, int extralen)
52 len += crypto_aead_ivsize(aead);
55 len += crypto_aead_alignmask(aead) &
56 ~(crypto_tfm_ctx_alignment() - 1);
57 len = ALIGN(len, crypto_tfm_ctx_alignment());
60 len += sizeof(struct aead_request) + crypto_aead_reqsize(aead);
61 len = ALIGN(len, __alignof__(struct scatterlist));
63 len += sizeof(struct scatterlist) * nfrags;
65 return kmalloc(len, GFP_ATOMIC);
68 static inline void *esp_tmp_extra(void *tmp)
70 return PTR_ALIGN(tmp, __alignof__(struct esp_output_extra));
73 static inline u8 *esp_tmp_iv(struct crypto_aead *aead, void *tmp, int extralen)
75 return crypto_aead_ivsize(aead) ?
76 PTR_ALIGN((u8 *)tmp + extralen,
77 crypto_aead_alignmask(aead) + 1) : tmp + extralen;
80 static inline struct aead_request *esp_tmp_req(struct crypto_aead *aead, u8 *iv)
82 struct aead_request *req;
84 req = (void *)PTR_ALIGN(iv + crypto_aead_ivsize(aead),
85 crypto_tfm_ctx_alignment());
86 aead_request_set_tfm(req, aead);
90 static inline struct scatterlist *esp_req_sg(struct crypto_aead *aead,
91 struct aead_request *req)
93 return (void *)ALIGN((unsigned long)(req + 1) +
94 crypto_aead_reqsize(aead),
95 __alignof__(struct scatterlist));
98 static void esp_ssg_unref(struct xfrm_state *x, void *tmp)
100 struct crypto_aead *aead = x->data;
103 struct aead_request *req;
104 struct scatterlist *sg;
106 if (x->props.flags & XFRM_STATE_ESN)
107 extralen += sizeof(struct esp_output_extra);
109 iv = esp_tmp_iv(aead, tmp, extralen);
110 req = esp_tmp_req(aead, iv);
112 /* Unref skb_frag_pages in the src scatterlist if necessary.
113 * Skip the first sg which comes from skb->data.
115 if (req->src != req->dst)
116 for (sg = sg_next(req->src); sg; sg = sg_next(sg))
117 put_page(sg_page(sg));
120 #ifdef CONFIG_INET_ESPINTCP
126 static void esp_free_tcp_sk(struct rcu_head *head)
128 struct esp_tcp_sk *esk = container_of(head, struct esp_tcp_sk, rcu);
134 static struct sock *esp_find_tcp_sk(struct xfrm_state *x)
136 struct xfrm_encap_tmpl *encap = x->encap;
137 struct net *net = xs_net(x);
138 struct esp_tcp_sk *esk;
143 sk = rcu_dereference(x->encap_sk);
144 if (sk && sk->sk_state == TCP_ESTABLISHED)
147 spin_lock_bh(&x->lock);
148 sport = encap->encap_sport;
149 dport = encap->encap_dport;
150 nsk = rcu_dereference_protected(x->encap_sk,
151 lockdep_is_held(&x->lock));
152 if (sk && sk == nsk) {
153 esk = kmalloc(sizeof(*esk), GFP_ATOMIC);
155 spin_unlock_bh(&x->lock);
156 return ERR_PTR(-ENOMEM);
158 RCU_INIT_POINTER(x->encap_sk, NULL);
160 call_rcu(&esk->rcu, esp_free_tcp_sk);
162 spin_unlock_bh(&x->lock);
164 sk = inet_lookup_established(net, net->ipv4.tcp_death_row.hashinfo, x->id.daddr.a4,
165 dport, x->props.saddr.a4, sport, 0);
167 return ERR_PTR(-ENOENT);
169 if (!tcp_is_ulp_esp(sk)) {
171 return ERR_PTR(-EINVAL);
174 spin_lock_bh(&x->lock);
175 nsk = rcu_dereference_protected(x->encap_sk,
176 lockdep_is_held(&x->lock));
177 if (encap->encap_sport != sport ||
178 encap->encap_dport != dport) {
180 sk = nsk ?: ERR_PTR(-EREMCHG);
181 } else if (sk == nsk) {
184 rcu_assign_pointer(x->encap_sk, sk);
186 spin_unlock_bh(&x->lock);
191 static int esp_output_tcp_finish(struct xfrm_state *x, struct sk_buff *skb)
198 sk = esp_find_tcp_sk(x);
199 err = PTR_ERR_OR_ZERO(sk);
204 if (sock_owned_by_user(sk))
205 err = espintcp_queue_out(sk, skb);
207 err = espintcp_push_skb(sk, skb);
215 static int esp_output_tcp_encap_cb(struct net *net, struct sock *sk,
218 struct dst_entry *dst = skb_dst(skb);
219 struct xfrm_state *x = dst->xfrm;
221 return esp_output_tcp_finish(x, skb);
224 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
229 err = xfrm_trans_queue_net(xs_net(x), skb, esp_output_tcp_encap_cb);
232 /* EINPROGRESS just happens to do the right thing. It
233 * actually means that the skb has been consumed and
236 return err ?: -EINPROGRESS;
239 static int esp_output_tail_tcp(struct xfrm_state *x, struct sk_buff *skb)
247 static void esp_output_done(struct crypto_async_request *base, int err)
249 struct sk_buff *skb = base->data;
250 struct xfrm_offload *xo = xfrm_offload(skb);
252 struct xfrm_state *x;
254 if (xo && (xo->flags & XFRM_DEV_RESUME)) {
255 struct sec_path *sp = skb_sec_path(skb);
257 x = sp->xvec[sp->len - 1];
259 x = skb_dst(skb)->xfrm;
262 tmp = ESP_SKB_CB(skb)->tmp;
263 esp_ssg_unref(x, tmp);
266 if (xo && (xo->flags & XFRM_DEV_RESUME)) {
268 XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
273 skb_push(skb, skb->data - skb_mac_header(skb));
275 xfrm_dev_resume(skb);
278 x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
279 esp_output_tail_tcp(x, skb);
281 xfrm_output_resume(skb->sk, skb, err);
285 /* Move ESP header back into place. */
286 static void esp_restore_header(struct sk_buff *skb, unsigned int offset)
288 struct ip_esp_hdr *esph = (void *)(skb->data + offset);
289 void *tmp = ESP_SKB_CB(skb)->tmp;
290 __be32 *seqhi = esp_tmp_extra(tmp);
292 esph->seq_no = esph->spi;
296 static void esp_output_restore_header(struct sk_buff *skb)
298 void *tmp = ESP_SKB_CB(skb)->tmp;
299 struct esp_output_extra *extra = esp_tmp_extra(tmp);
301 esp_restore_header(skb, skb_transport_offset(skb) + extra->esphoff -
305 static struct ip_esp_hdr *esp_output_set_extra(struct sk_buff *skb,
306 struct xfrm_state *x,
307 struct ip_esp_hdr *esph,
308 struct esp_output_extra *extra)
310 /* For ESN we move the header forward by 4 bytes to
311 * accommodate the high bits. We will move it back after
314 if ((x->props.flags & XFRM_STATE_ESN)) {
316 struct xfrm_offload *xo = xfrm_offload(skb);
321 seqhi = XFRM_SKB_CB(skb)->seq.output.hi;
323 extra->esphoff = (unsigned char *)esph -
324 skb_transport_header(skb);
325 esph = (struct ip_esp_hdr *)((unsigned char *)esph - 4);
326 extra->seqhi = esph->spi;
327 esph->seq_no = htonl(seqhi);
330 esph->spi = x->id.spi;
335 static void esp_output_done_esn(struct crypto_async_request *base, int err)
337 struct sk_buff *skb = base->data;
339 esp_output_restore_header(skb);
340 esp_output_done(base, err);
343 static struct ip_esp_hdr *esp_output_udp_encap(struct sk_buff *skb,
345 struct esp_info *esp,
353 len = skb->len + esp->tailen - skb_transport_offset(skb);
354 if (len + sizeof(struct iphdr) > IP_MAX_MTU)
355 return ERR_PTR(-EMSGSIZE);
357 uh = (struct udphdr *)esp->esph;
360 uh->len = htons(len);
363 *skb_mac_header(skb) = IPPROTO_UDP;
365 if (encap_type == UDP_ENCAP_ESPINUDP_NON_IKE) {
366 udpdata32 = (__be32 *)(uh + 1);
367 udpdata32[0] = udpdata32[1] = 0;
368 return (struct ip_esp_hdr *)(udpdata32 + 2);
371 return (struct ip_esp_hdr *)(uh + 1);
374 #ifdef CONFIG_INET_ESPINTCP
375 static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
377 struct esp_info *esp)
379 __be16 *lenp = (void *)esp->esph;
380 struct ip_esp_hdr *esph;
384 len = skb->len + esp->tailen - skb_transport_offset(skb);
385 if (len > IP_MAX_MTU)
386 return ERR_PTR(-EMSGSIZE);
389 sk = esp_find_tcp_sk(x);
396 esph = (struct ip_esp_hdr *)(lenp + 1);
401 static struct ip_esp_hdr *esp_output_tcp_encap(struct xfrm_state *x,
403 struct esp_info *esp)
405 return ERR_PTR(-EOPNOTSUPP);
409 static int esp_output_encap(struct xfrm_state *x, struct sk_buff *skb,
410 struct esp_info *esp)
412 struct xfrm_encap_tmpl *encap = x->encap;
413 struct ip_esp_hdr *esph;
417 spin_lock_bh(&x->lock);
418 sport = encap->encap_sport;
419 dport = encap->encap_dport;
420 encap_type = encap->encap_type;
421 spin_unlock_bh(&x->lock);
423 switch (encap_type) {
425 case UDP_ENCAP_ESPINUDP:
426 case UDP_ENCAP_ESPINUDP_NON_IKE:
427 esph = esp_output_udp_encap(skb, encap_type, esp, sport, dport);
429 case TCP_ENCAP_ESPINTCP:
430 esph = esp_output_tcp_encap(x, skb, esp);
435 return PTR_ERR(esph);
442 int esp_output_head(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
448 struct sk_buff *trailer;
449 int tailen = esp->tailen;
451 /* this is non-NULL only with TCP/UDP Encapsulation */
453 int err = esp_output_encap(x, skb, esp);
459 if (ALIGN(tailen, L1_CACHE_BYTES) > PAGE_SIZE ||
460 ALIGN(skb->data_len, L1_CACHE_BYTES) > PAGE_SIZE)
463 if (!skb_cloned(skb)) {
464 if (tailen <= skb_tailroom(skb)) {
467 tail = skb_tail_pointer(trailer);
470 } else if ((skb_shinfo(skb)->nr_frags < MAX_SKB_FRAGS)
471 && !skb_has_frag_list(skb)) {
473 struct sock *sk = skb->sk;
474 struct page_frag *pfrag = &x->xfrag;
476 esp->inplace = false;
478 allocsize = ALIGN(tailen, L1_CACHE_BYTES);
480 spin_lock_bh(&x->lock);
482 if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
483 spin_unlock_bh(&x->lock);
490 tail = page_address(page) + pfrag->offset;
492 esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
494 nfrags = skb_shinfo(skb)->nr_frags;
496 __skb_fill_page_desc(skb, nfrags, page, pfrag->offset,
498 skb_shinfo(skb)->nr_frags = ++nfrags;
500 pfrag->offset = pfrag->offset + allocsize;
502 spin_unlock_bh(&x->lock);
506 skb_len_add(skb, tailen);
507 if (sk && sk_fullsock(sk))
508 refcount_add(tailen, &sk->sk_wmem_alloc);
515 esph_offset = (unsigned char *)esp->esph - skb_transport_header(skb);
517 nfrags = skb_cow_data(skb, tailen, &trailer);
520 tail = skb_tail_pointer(trailer);
521 esp->esph = (struct ip_esp_hdr *)(skb_transport_header(skb) + esph_offset);
524 esp_output_fill_trailer(tail, esp->tfclen, esp->plen, esp->proto);
525 pskb_put(skb, trailer, tailen);
530 EXPORT_SYMBOL_GPL(esp_output_head);
532 int esp_output_tail(struct xfrm_state *x, struct sk_buff *skb, struct esp_info *esp)
541 struct ip_esp_hdr *esph;
542 struct crypto_aead *aead;
543 struct aead_request *req;
544 struct scatterlist *sg, *dsg;
545 struct esp_output_extra *extra;
548 assoclen = sizeof(struct ip_esp_hdr);
551 if (x->props.flags & XFRM_STATE_ESN) {
552 extralen += sizeof(*extra);
553 assoclen += sizeof(__be32);
557 alen = crypto_aead_authsize(aead);
558 ivlen = crypto_aead_ivsize(aead);
560 tmp = esp_alloc_tmp(aead, esp->nfrags + 2, extralen);
564 extra = esp_tmp_extra(tmp);
565 iv = esp_tmp_iv(aead, tmp, extralen);
566 req = esp_tmp_req(aead, iv);
567 sg = esp_req_sg(aead, req);
572 dsg = &sg[esp->nfrags];
574 esph = esp_output_set_extra(skb, x, esp->esph, extra);
577 sg_init_table(sg, esp->nfrags);
578 err = skb_to_sgvec(skb, sg,
579 (unsigned char *)esph - skb->data,
580 assoclen + ivlen + esp->clen + alen);
581 if (unlikely(err < 0))
586 struct page_frag *pfrag = &x->xfrag;
588 allocsize = ALIGN(skb->data_len, L1_CACHE_BYTES);
590 spin_lock_bh(&x->lock);
591 if (unlikely(!skb_page_frag_refill(allocsize, pfrag, GFP_ATOMIC))) {
592 spin_unlock_bh(&x->lock);
596 skb_shinfo(skb)->nr_frags = 1;
600 /* replace page frags in skb with new page */
601 __skb_fill_page_desc(skb, 0, page, pfrag->offset, skb->data_len);
602 pfrag->offset = pfrag->offset + allocsize;
603 spin_unlock_bh(&x->lock);
605 sg_init_table(dsg, skb_shinfo(skb)->nr_frags + 1);
606 err = skb_to_sgvec(skb, dsg,
607 (unsigned char *)esph - skb->data,
608 assoclen + ivlen + esp->clen + alen);
609 if (unlikely(err < 0))
613 if ((x->props.flags & XFRM_STATE_ESN))
614 aead_request_set_callback(req, 0, esp_output_done_esn, skb);
616 aead_request_set_callback(req, 0, esp_output_done, skb);
618 aead_request_set_crypt(req, sg, dsg, ivlen + esp->clen, iv);
619 aead_request_set_ad(req, assoclen);
621 memset(iv, 0, ivlen);
622 memcpy(iv + ivlen - min(ivlen, 8), (u8 *)&esp->seqno + 8 - min(ivlen, 8),
625 ESP_SKB_CB(skb)->tmp = tmp;
626 err = crypto_aead_encrypt(req);
637 if ((x->props.flags & XFRM_STATE_ESN))
638 esp_output_restore_header(skb);
642 esp_ssg_unref(x, tmp);
644 if (!err && x->encap && x->encap->encap_type == TCP_ENCAP_ESPINTCP)
645 err = esp_output_tail_tcp(x, skb);
652 EXPORT_SYMBOL_GPL(esp_output_tail);
654 static int esp_output(struct xfrm_state *x, struct sk_buff *skb)
658 struct ip_esp_hdr *esph;
659 struct crypto_aead *aead;
664 esp.proto = *skb_mac_header(skb);
665 *skb_mac_header(skb) = IPPROTO_ESP;
667 /* skb is pure payload to encrypt */
670 alen = crypto_aead_authsize(aead);
674 struct xfrm_dst *dst = (struct xfrm_dst *)skb_dst(skb);
677 padto = min(x->tfcpad, xfrm_state_mtu(x, dst->child_mtu_cached));
678 if (skb->len < padto)
679 esp.tfclen = padto - skb->len;
681 blksize = ALIGN(crypto_aead_blocksize(aead), 4);
682 esp.clen = ALIGN(skb->len + 2 + esp.tfclen, blksize);
683 esp.plen = esp.clen - skb->len - esp.tfclen;
684 esp.tailen = esp.tfclen + esp.plen + alen;
686 esp.esph = ip_esp_hdr(skb);
688 esp.nfrags = esp_output_head(x, skb, &esp);
693 esph->spi = x->id.spi;
695 esph->seq_no = htonl(XFRM_SKB_CB(skb)->seq.output.low);
696 esp.seqno = cpu_to_be64(XFRM_SKB_CB(skb)->seq.output.low +
697 ((u64)XFRM_SKB_CB(skb)->seq.output.hi << 32));
699 skb_push(skb, -skb_network_offset(skb));
701 return esp_output_tail(x, skb, &esp);
704 static inline int esp_remove_trailer(struct sk_buff *skb)
706 struct xfrm_state *x = xfrm_input_state(skb);
707 struct crypto_aead *aead = x->data;
708 int alen, hlen, elen;
714 alen = crypto_aead_authsize(aead);
715 hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
716 elen = skb->len - hlen;
718 if (skb_copy_bits(skb, skb->len - alen - 2, nexthdr, 2))
723 if (padlen + 2 + alen >= elen) {
724 net_dbg_ratelimited("ipsec esp packet is garbage padlen=%d, elen=%d\n",
725 padlen + 2, elen - alen);
729 trimlen = alen + padlen + 2;
730 if (skb->ip_summed == CHECKSUM_COMPLETE) {
731 csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
732 skb->csum = csum_block_sub(skb->csum, csumdiff,
735 pskb_trim(skb, skb->len - trimlen);
743 int esp_input_done2(struct sk_buff *skb, int err)
745 const struct iphdr *iph;
746 struct xfrm_state *x = xfrm_input_state(skb);
747 struct xfrm_offload *xo = xfrm_offload(skb);
748 struct crypto_aead *aead = x->data;
749 int hlen = sizeof(struct ip_esp_hdr) + crypto_aead_ivsize(aead);
752 if (!xo || !(xo->flags & CRYPTO_DONE))
753 kfree(ESP_SKB_CB(skb)->tmp);
758 err = esp_remove_trailer(skb);
759 if (unlikely(err < 0))
766 struct xfrm_encap_tmpl *encap = x->encap;
767 struct tcphdr *th = (void *)(skb_network_header(skb) + ihl);
768 struct udphdr *uh = (void *)(skb_network_header(skb) + ihl);
771 switch (x->encap->encap_type) {
772 case TCP_ENCAP_ESPINTCP:
775 case UDP_ENCAP_ESPINUDP:
776 case UDP_ENCAP_ESPINUDP_NON_IKE:
786 * 1) if the NAT-T peer's IP or port changed then
787 * advertize the change to the keying daemon.
788 * This is an inbound SA, so just compare
791 if (iph->saddr != x->props.saddr.a4 ||
792 source != encap->encap_sport) {
793 xfrm_address_t ipaddr;
795 ipaddr.a4 = iph->saddr;
796 km_new_mapping(x, &ipaddr, source);
798 /* XXX: perhaps add an extra
799 * policy check here, to see
800 * if we should allow or
801 * reject a packet from a
808 * 2) ignore UDP/TCP checksums in case
809 * of NAT-T in Transport Mode, or
810 * perform other post-processing fixes
811 * as per draft-ietf-ipsec-udp-encaps-06,
814 if (x->props.mode == XFRM_MODE_TRANSPORT)
815 skb->ip_summed = CHECKSUM_UNNECESSARY;
818 skb_pull_rcsum(skb, hlen);
819 if (x->props.mode == XFRM_MODE_TUNNEL)
820 skb_reset_transport_header(skb);
822 skb_set_transport_header(skb, -ihl);
824 /* RFC4303: Drop dummy packets without any error */
825 if (err == IPPROTO_NONE)
831 EXPORT_SYMBOL_GPL(esp_input_done2);
833 static void esp_input_done(struct crypto_async_request *base, int err)
835 struct sk_buff *skb = base->data;
837 xfrm_input_resume(skb, esp_input_done2(skb, err));
840 static void esp_input_restore_header(struct sk_buff *skb)
842 esp_restore_header(skb, 0);
846 static void esp_input_set_header(struct sk_buff *skb, __be32 *seqhi)
848 struct xfrm_state *x = xfrm_input_state(skb);
849 struct ip_esp_hdr *esph;
851 /* For ESN we move the header forward by 4 bytes to
852 * accommodate the high bits. We will move it back after
855 if ((x->props.flags & XFRM_STATE_ESN)) {
856 esph = skb_push(skb, 4);
858 esph->spi = esph->seq_no;
859 esph->seq_no = XFRM_SKB_CB(skb)->seq.input.hi;
863 static void esp_input_done_esn(struct crypto_async_request *base, int err)
865 struct sk_buff *skb = base->data;
867 esp_input_restore_header(skb);
868 esp_input_done(base, err);
872 * Note: detecting truncated vs. non-truncated authentication data is very
873 * expensive, so we only support truncated data, which is the recommended
876 static int esp_input(struct xfrm_state *x, struct sk_buff *skb)
878 struct crypto_aead *aead = x->data;
879 struct aead_request *req;
880 struct sk_buff *trailer;
881 int ivlen = crypto_aead_ivsize(aead);
882 int elen = skb->len - sizeof(struct ip_esp_hdr) - ivlen;
889 struct scatterlist *sg;
892 if (!pskb_may_pull(skb, sizeof(struct ip_esp_hdr) + ivlen))
898 assoclen = sizeof(struct ip_esp_hdr);
901 if (x->props.flags & XFRM_STATE_ESN) {
902 seqhilen += sizeof(__be32);
903 assoclen += seqhilen;
906 if (!skb_cloned(skb)) {
907 if (!skb_is_nonlinear(skb)) {
911 } else if (!skb_has_frag_list(skb)) {
912 nfrags = skb_shinfo(skb)->nr_frags;
919 err = skb_cow_data(skb, 0, &trailer);
927 tmp = esp_alloc_tmp(aead, nfrags, seqhilen);
931 ESP_SKB_CB(skb)->tmp = tmp;
932 seqhi = esp_tmp_extra(tmp);
933 iv = esp_tmp_iv(aead, tmp, seqhilen);
934 req = esp_tmp_req(aead, iv);
935 sg = esp_req_sg(aead, req);
937 esp_input_set_header(skb, seqhi);
939 sg_init_table(sg, nfrags);
940 err = skb_to_sgvec(skb, sg, 0, skb->len);
941 if (unlikely(err < 0)) {
946 skb->ip_summed = CHECKSUM_NONE;
948 if ((x->props.flags & XFRM_STATE_ESN))
949 aead_request_set_callback(req, 0, esp_input_done_esn, skb);
951 aead_request_set_callback(req, 0, esp_input_done, skb);
953 aead_request_set_crypt(req, sg, sg, elen + ivlen, iv);
954 aead_request_set_ad(req, assoclen);
956 err = crypto_aead_decrypt(req);
957 if (err == -EINPROGRESS)
960 if ((x->props.flags & XFRM_STATE_ESN))
961 esp_input_restore_header(skb);
963 err = esp_input_done2(skb, err);
969 static int esp4_err(struct sk_buff *skb, u32 info)
971 struct net *net = dev_net(skb->dev);
972 const struct iphdr *iph = (const struct iphdr *)skb->data;
973 struct ip_esp_hdr *esph = (struct ip_esp_hdr *)(skb->data+(iph->ihl<<2));
974 struct xfrm_state *x;
976 switch (icmp_hdr(skb)->type) {
977 case ICMP_DEST_UNREACH:
978 if (icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
987 x = xfrm_state_lookup(net, skb->mark, (const xfrm_address_t *)&iph->daddr,
988 esph->spi, IPPROTO_ESP, AF_INET);
992 if (icmp_hdr(skb)->type == ICMP_DEST_UNREACH)
993 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ESP);
995 ipv4_redirect(skb, net, 0, IPPROTO_ESP);
1001 static void esp_destroy(struct xfrm_state *x)
1003 struct crypto_aead *aead = x->data;
1008 crypto_free_aead(aead);
1011 static int esp_init_aead(struct xfrm_state *x, struct netlink_ext_ack *extack)
1013 char aead_name[CRYPTO_MAX_ALG_NAME];
1014 struct crypto_aead *aead;
1017 if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
1018 x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME) {
1019 NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1020 return -ENAMETOOLONG;
1023 aead = crypto_alloc_aead(aead_name, 0, 0);
1024 err = PTR_ERR(aead);
1030 err = crypto_aead_setkey(aead, x->aead->alg_key,
1031 (x->aead->alg_key_len + 7) / 8);
1035 err = crypto_aead_setauthsize(aead, x->aead->alg_icv_len / 8);
1042 NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1046 static int esp_init_authenc(struct xfrm_state *x,
1047 struct netlink_ext_ack *extack)
1049 struct crypto_aead *aead;
1050 struct crypto_authenc_key_param *param;
1054 char authenc_name[CRYPTO_MAX_ALG_NAME];
1055 unsigned int keylen;
1058 err = -ENAMETOOLONG;
1060 if ((x->props.flags & XFRM_STATE_ESN)) {
1061 if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1062 "%s%sauthencesn(%s,%s)%s",
1063 x->geniv ?: "", x->geniv ? "(" : "",
1064 x->aalg ? x->aalg->alg_name : "digest_null",
1066 x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1067 NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1071 if (snprintf(authenc_name, CRYPTO_MAX_ALG_NAME,
1072 "%s%sauthenc(%s,%s)%s",
1073 x->geniv ?: "", x->geniv ? "(" : "",
1074 x->aalg ? x->aalg->alg_name : "digest_null",
1076 x->geniv ? ")" : "") >= CRYPTO_MAX_ALG_NAME) {
1077 NL_SET_ERR_MSG(extack, "Algorithm name is too long");
1082 aead = crypto_alloc_aead(authenc_name, 0, 0);
1083 err = PTR_ERR(aead);
1085 NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1091 keylen = (x->aalg ? (x->aalg->alg_key_len + 7) / 8 : 0) +
1092 (x->ealg->alg_key_len + 7) / 8 + RTA_SPACE(sizeof(*param));
1094 key = kmalloc(keylen, GFP_KERNEL);
1100 rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
1101 rta->rta_len = RTA_LENGTH(sizeof(*param));
1102 param = RTA_DATA(rta);
1103 p += RTA_SPACE(sizeof(*param));
1106 struct xfrm_algo_desc *aalg_desc;
1108 memcpy(p, x->aalg->alg_key, (x->aalg->alg_key_len + 7) / 8);
1109 p += (x->aalg->alg_key_len + 7) / 8;
1111 aalg_desc = xfrm_aalg_get_byname(x->aalg->alg_name, 0);
1115 if (aalg_desc->uinfo.auth.icv_fullbits / 8 !=
1116 crypto_aead_authsize(aead)) {
1117 NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1121 err = crypto_aead_setauthsize(
1122 aead, x->aalg->alg_trunc_len / 8);
1124 NL_SET_ERR_MSG(extack, "Kernel was unable to initialize cryptographic operations");
1129 param->enckeylen = cpu_to_be32((x->ealg->alg_key_len + 7) / 8);
1130 memcpy(p, x->ealg->alg_key, (x->ealg->alg_key_len + 7) / 8);
1132 err = crypto_aead_setkey(aead, key, keylen);
1141 static int esp_init_state(struct xfrm_state *x, struct netlink_ext_ack *extack)
1143 struct crypto_aead *aead;
1150 err = esp_init_aead(x, extack);
1151 } else if (x->ealg) {
1152 err = esp_init_authenc(x, extack);
1154 NL_SET_ERR_MSG(extack, "ESP: AEAD or CRYPT must be provided");
1163 x->props.header_len = sizeof(struct ip_esp_hdr) +
1164 crypto_aead_ivsize(aead);
1165 if (x->props.mode == XFRM_MODE_TUNNEL)
1166 x->props.header_len += sizeof(struct iphdr);
1167 else if (x->props.mode == XFRM_MODE_BEET && x->sel.family != AF_INET6)
1168 x->props.header_len += IPV4_BEET_PHMAXLEN;
1170 struct xfrm_encap_tmpl *encap = x->encap;
1172 switch (encap->encap_type) {
1174 NL_SET_ERR_MSG(extack, "Unsupported encapsulation type for ESP");
1177 case UDP_ENCAP_ESPINUDP:
1178 x->props.header_len += sizeof(struct udphdr);
1180 case UDP_ENCAP_ESPINUDP_NON_IKE:
1181 x->props.header_len += sizeof(struct udphdr) + 2 * sizeof(u32);
1183 #ifdef CONFIG_INET_ESPINTCP
1184 case TCP_ENCAP_ESPINTCP:
1185 /* only the length field, TCP encap is done by
1188 x->props.header_len += 2;
1194 align = ALIGN(crypto_aead_blocksize(aead), 4);
1195 x->props.trailer_len = align + 1 + crypto_aead_authsize(aead);
1201 static int esp4_rcv_cb(struct sk_buff *skb, int err)
1206 static const struct xfrm_type esp_type =
1208 .owner = THIS_MODULE,
1209 .proto = IPPROTO_ESP,
1210 .flags = XFRM_TYPE_REPLAY_PROT,
1211 .init_state = esp_init_state,
1212 .destructor = esp_destroy,
1214 .output = esp_output,
1217 static struct xfrm4_protocol esp4_protocol = {
1218 .handler = xfrm4_rcv,
1219 .input_handler = xfrm_input,
1220 .cb_handler = esp4_rcv_cb,
1221 .err_handler = esp4_err,
1225 static int __init esp4_init(void)
1227 if (xfrm_register_type(&esp_type, AF_INET) < 0) {
1228 pr_info("%s: can't add xfrm type\n", __func__);
1231 if (xfrm4_protocol_register(&esp4_protocol, IPPROTO_ESP) < 0) {
1232 pr_info("%s: can't add protocol\n", __func__);
1233 xfrm_unregister_type(&esp_type, AF_INET);
1239 static void __exit esp4_fini(void)
1241 if (xfrm4_protocol_deregister(&esp4_protocol, IPPROTO_ESP) < 0)
1242 pr_info("%s: can't remove protocol\n", __func__);
1243 xfrm_unregister_type(&esp_type, AF_INET);
1246 module_init(esp4_init);
1247 module_exit(esp4_fini);
1248 MODULE_LICENSE("GPL");
1249 MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_ESP);