Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * xfrm_input.c
4 *
5 * Changes:
6 * YOSHIFUJI Hideaki @USAGI
7 * Split up af-specific portion
8 *
9 */
10
11#include <linux/bottom_half.h>
12#include <linux/cache.h>
13#include <linux/interrupt.h>
14#include <linux/slab.h>
15#include <linux/module.h>
16#include <linux/netdevice.h>
17#include <linux/percpu.h>
18#include <net/dst.h>
19#include <net/ip.h>
20#include <net/xfrm.h>
21#include <net/ip_tunnels.h>
22#include <net/ip6_tunnel.h>
23#include <net/dst_metadata.h>
24#include <net/hotdata.h>
25
26#include "xfrm_inout.h"
27
28struct xfrm_trans_tasklet {
29 struct work_struct work;
30 spinlock_t queue_lock;
31 struct sk_buff_head queue;
32};
33
34struct xfrm_trans_cb {
35 union {
36 struct inet_skb_parm h4;
37#if IS_ENABLED(CONFIG_IPV6)
38 struct inet6_skb_parm h6;
39#endif
40 } header;
41 int (*finish)(struct net *net, struct sock *sk, struct sk_buff *skb);
42 struct net *net;
43};
44
45#define XFRM_TRANS_SKB_CB(__skb) ((struct xfrm_trans_cb *)&((__skb)->cb[0]))
46
47static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
48static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[2][AF_INET6 + 1];
49
50static struct gro_cells gro_cells;
51static struct net_device xfrm_napi_dev;
52
53static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
54
55int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
56{
57 int err = 0;
58
59 if (WARN_ON(afinfo->family > AF_INET6))
60 return -EAFNOSUPPORT;
61
62 spin_lock_bh(&xfrm_input_afinfo_lock);
63 if (unlikely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family]))
64 err = -EEXIST;
65 else
66 rcu_assign_pointer(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family], afinfo);
67 spin_unlock_bh(&xfrm_input_afinfo_lock);
68 return err;
69}
70EXPORT_SYMBOL(xfrm_input_register_afinfo);
71
72int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
73{
74 int err = 0;
75
76 spin_lock_bh(&xfrm_input_afinfo_lock);
77 if (likely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family])) {
78 if (unlikely(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family] != afinfo))
79 err = -EINVAL;
80 else
81 RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->is_ipip][afinfo->family], NULL);
82 }
83 spin_unlock_bh(&xfrm_input_afinfo_lock);
84 synchronize_rcu();
85 return err;
86}
87EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
88
89static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(u8 family, bool is_ipip)
90{
91 const struct xfrm_input_afinfo *afinfo;
92
93 if (WARN_ON_ONCE(family > AF_INET6))
94 return NULL;
95
96 rcu_read_lock();
97 afinfo = rcu_dereference(xfrm_input_afinfo[is_ipip][family]);
98 if (unlikely(!afinfo))
99 rcu_read_unlock();
100 return afinfo;
101}
102
103static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
104 int err)
105{
106 bool is_ipip = (protocol == IPPROTO_IPIP || protocol == IPPROTO_IPV6);
107 const struct xfrm_input_afinfo *afinfo;
108 int ret;
109
110 afinfo = xfrm_input_get_afinfo(family, is_ipip);
111 if (!afinfo)
112 return -EAFNOSUPPORT;
113
114 ret = afinfo->callback(skb, protocol, err);
115 rcu_read_unlock();
116
117 return ret;
118}
119
120struct sec_path *secpath_set(struct sk_buff *skb)
121{
122 struct sec_path *sp, *tmp = skb_ext_find(skb, SKB_EXT_SEC_PATH);
123
124 sp = skb_ext_add(skb, SKB_EXT_SEC_PATH);
125 if (!sp)
126 return NULL;
127
128 if (tmp) /* reused existing one (was COW'd if needed) */
129 return sp;
130
131 /* allocated new secpath */
132 memset(sp->ovec, 0, sizeof(sp->ovec));
133 sp->olen = 0;
134 sp->len = 0;
135 sp->verified_cnt = 0;
136
137 return sp;
138}
139EXPORT_SYMBOL(secpath_set);
140
141/* Fetch spi and seq from ipsec header */
142
143int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
144{
145 int offset, offset_seq;
146 int hlen;
147
148 switch (nexthdr) {
149 case IPPROTO_AH:
150 hlen = sizeof(struct ip_auth_hdr);
151 offset = offsetof(struct ip_auth_hdr, spi);
152 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
153 break;
154 case IPPROTO_ESP:
155 hlen = sizeof(struct ip_esp_hdr);
156 offset = offsetof(struct ip_esp_hdr, spi);
157 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
158 break;
159 case IPPROTO_COMP:
160 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
161 return -EINVAL;
162 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
163 *seq = 0;
164 return 0;
165 default:
166 return 1;
167 }
168
169 if (!pskb_may_pull(skb, hlen))
170 return -EINVAL;
171
172 *spi = *(__be32 *)(skb_transport_header(skb) + offset);
173 *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
174 return 0;
175}
176EXPORT_SYMBOL(xfrm_parse_spi);
177
178static int xfrm4_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
179{
180 struct iphdr *iph;
181 int optlen = 0;
182 int err = -EINVAL;
183
184 skb->protocol = htons(ETH_P_IP);
185
186 if (unlikely(XFRM_MODE_SKB_CB(skb)->protocol == IPPROTO_BEETPH)) {
187 struct ip_beet_phdr *ph;
188 int phlen;
189
190 if (!pskb_may_pull(skb, sizeof(*ph)))
191 goto out;
192
193 ph = (struct ip_beet_phdr *)skb->data;
194
195 phlen = sizeof(*ph) + ph->padlen;
196 optlen = ph->hdrlen * 8 + (IPV4_BEET_PHMAXLEN - phlen);
197 if (optlen < 0 || optlen & 3 || optlen > 250)
198 goto out;
199
200 XFRM_MODE_SKB_CB(skb)->protocol = ph->nexthdr;
201
202 if (!pskb_may_pull(skb, phlen))
203 goto out;
204 __skb_pull(skb, phlen);
205 }
206
207 skb_push(skb, sizeof(*iph));
208 skb_reset_network_header(skb);
209 skb_mac_header_rebuild(skb);
210
211 xfrm4_beet_make_header(skb);
212
213 iph = ip_hdr(skb);
214
215 iph->ihl += optlen / 4;
216 iph->tot_len = htons(skb->len);
217 iph->daddr = x->sel.daddr.a4;
218 iph->saddr = x->sel.saddr.a4;
219 iph->check = 0;
220 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
221 err = 0;
222out:
223 return err;
224}
225
226static void ipip_ecn_decapsulate(struct sk_buff *skb)
227{
228 struct iphdr *inner_iph = ipip_hdr(skb);
229
230 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
231 IP_ECN_set_ce(inner_iph);
232}
233
234static int xfrm4_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
235{
236 int err = -EINVAL;
237
238 skb->protocol = htons(ETH_P_IP);
239
240 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
241 goto out;
242
243 err = skb_unclone(skb, GFP_ATOMIC);
244 if (err)
245 goto out;
246
247 if (x->props.flags & XFRM_STATE_DECAP_DSCP)
248 ipv4_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipip_hdr(skb));
249 if (!(x->props.flags & XFRM_STATE_NOECN))
250 ipip_ecn_decapsulate(skb);
251
252 skb_reset_network_header(skb);
253 skb_mac_header_rebuild(skb);
254 if (skb->mac_len)
255 eth_hdr(skb)->h_proto = skb->protocol;
256
257 err = 0;
258
259out:
260 return err;
261}
262
263static void ipip6_ecn_decapsulate(struct sk_buff *skb)
264{
265 struct ipv6hdr *inner_iph = ipipv6_hdr(skb);
266
267 if (INET_ECN_is_ce(XFRM_MODE_SKB_CB(skb)->tos))
268 IP6_ECN_set_ce(skb, inner_iph);
269}
270
271static int xfrm6_remove_tunnel_encap(struct xfrm_state *x, struct sk_buff *skb)
272{
273 int err = -EINVAL;
274
275 skb->protocol = htons(ETH_P_IPV6);
276
277 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
278 goto out;
279
280 err = skb_unclone(skb, GFP_ATOMIC);
281 if (err)
282 goto out;
283
284 if (x->props.flags & XFRM_STATE_DECAP_DSCP)
285 ipv6_copy_dscp(XFRM_MODE_SKB_CB(skb)->tos, ipipv6_hdr(skb));
286 if (!(x->props.flags & XFRM_STATE_NOECN))
287 ipip6_ecn_decapsulate(skb);
288
289 skb_reset_network_header(skb);
290 skb_mac_header_rebuild(skb);
291 if (skb->mac_len)
292 eth_hdr(skb)->h_proto = skb->protocol;
293
294 err = 0;
295
296out:
297 return err;
298}
299
300static int xfrm6_remove_beet_encap(struct xfrm_state *x, struct sk_buff *skb)
301{
302 struct ipv6hdr *ip6h;
303 int size = sizeof(struct ipv6hdr);
304 int err;
305
306 skb->protocol = htons(ETH_P_IPV6);
307
308 err = skb_cow_head(skb, size + skb->mac_len);
309 if (err)
310 goto out;
311
312 __skb_push(skb, size);
313 skb_reset_network_header(skb);
314 skb_mac_header_rebuild(skb);
315
316 xfrm6_beet_make_header(skb);
317
318 ip6h = ipv6_hdr(skb);
319 ip6h->payload_len = htons(skb->len - size);
320 ip6h->daddr = x->sel.daddr.in6;
321 ip6h->saddr = x->sel.saddr.in6;
322 err = 0;
323out:
324 return err;
325}
326
327/* Remove encapsulation header.
328 *
329 * The IP header will be moved over the top of the encapsulation
330 * header.
331 *
332 * On entry, the transport header shall point to where the IP header
333 * should be and the network header shall be set to where the IP
334 * header currently is. skb->data shall point to the start of the
335 * payload.
336 */
337static int
338xfrm_inner_mode_encap_remove(struct xfrm_state *x,
339 struct sk_buff *skb)
340{
341 switch (x->props.mode) {
342 case XFRM_MODE_BEET:
343 switch (x->sel.family) {
344 case AF_INET:
345 return xfrm4_remove_beet_encap(x, skb);
346 case AF_INET6:
347 return xfrm6_remove_beet_encap(x, skb);
348 }
349 break;
350 case XFRM_MODE_TUNNEL:
351 switch (XFRM_MODE_SKB_CB(skb)->protocol) {
352 case IPPROTO_IPIP:
353 return xfrm4_remove_tunnel_encap(x, skb);
354 case IPPROTO_IPV6:
355 return xfrm6_remove_tunnel_encap(x, skb);
356 break;
357 }
358 return -EINVAL;
359 }
360
361 WARN_ON_ONCE(1);
362 return -EOPNOTSUPP;
363}
364
365static int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
366{
367 switch (x->props.family) {
368 case AF_INET:
369 xfrm4_extract_header(skb);
370 break;
371 case AF_INET6:
372 xfrm6_extract_header(skb);
373 break;
374 default:
375 WARN_ON_ONCE(1);
376 return -EAFNOSUPPORT;
377 }
378
379 return xfrm_inner_mode_encap_remove(x, skb);
380}
381
382/* Remove encapsulation header.
383 *
384 * The IP header will be moved over the top of the encapsulation header.
385 *
386 * On entry, skb_transport_header() shall point to where the IP header
387 * should be and skb_network_header() shall be set to where the IP header
388 * currently is. skb->data shall point to the start of the payload.
389 */
390static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
391{
392 struct xfrm_offload *xo = xfrm_offload(skb);
393 int ihl = skb->data - skb_transport_header(skb);
394
395 if (skb->transport_header != skb->network_header) {
396 memmove(skb_transport_header(skb),
397 skb_network_header(skb), ihl);
398 if (xo)
399 xo->orig_mac_len =
400 skb_mac_header_was_set(skb) ? skb_mac_header_len(skb) : 0;
401 skb->network_header = skb->transport_header;
402 }
403 ip_hdr(skb)->tot_len = htons(skb->len + ihl);
404 skb_reset_transport_header(skb);
405 return 0;
406}
407
408static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
409{
410#if IS_ENABLED(CONFIG_IPV6)
411 struct xfrm_offload *xo = xfrm_offload(skb);
412 int ihl = skb->data - skb_transport_header(skb);
413
414 if (skb->transport_header != skb->network_header) {
415 memmove(skb_transport_header(skb),
416 skb_network_header(skb), ihl);
417 if (xo)
418 xo->orig_mac_len =
419 skb_mac_header_was_set(skb) ? skb_mac_header_len(skb) : 0;
420 skb->network_header = skb->transport_header;
421 }
422 ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
423 sizeof(struct ipv6hdr));
424 skb_reset_transport_header(skb);
425 return 0;
426#else
427 WARN_ON_ONCE(1);
428 return -EAFNOSUPPORT;
429#endif
430}
431
432static int xfrm_inner_mode_input(struct xfrm_state *x,
433 struct sk_buff *skb)
434{
435 switch (x->props.mode) {
436 case XFRM_MODE_BEET:
437 case XFRM_MODE_TUNNEL:
438 return xfrm_prepare_input(x, skb);
439 case XFRM_MODE_TRANSPORT:
440 if (x->props.family == AF_INET)
441 return xfrm4_transport_input(x, skb);
442 if (x->props.family == AF_INET6)
443 return xfrm6_transport_input(x, skb);
444 break;
445 case XFRM_MODE_ROUTEOPTIMIZATION:
446 WARN_ON_ONCE(1);
447 break;
448 default:
449 WARN_ON_ONCE(1);
450 break;
451 }
452
453 return -EOPNOTSUPP;
454}
455
456int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
457{
458 const struct xfrm_state_afinfo *afinfo;
459 struct net *net = dev_net(skb->dev);
460 int err;
461 __be32 seq;
462 __be32 seq_hi;
463 struct xfrm_state *x = NULL;
464 xfrm_address_t *daddr;
465 u32 mark = skb->mark;
466 unsigned int family = AF_UNSPEC;
467 int decaps = 0;
468 int async = 0;
469 bool xfrm_gro = false;
470 bool crypto_done = false;
471 struct xfrm_offload *xo = xfrm_offload(skb);
472 struct sec_path *sp;
473
474 if (encap_type < 0 || (xo && (xo->flags & XFRM_GRO || encap_type == 0 ||
475 encap_type == UDP_ENCAP_ESPINUDP))) {
476 x = xfrm_input_state(skb);
477
478 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
479 if (x->km.state == XFRM_STATE_ACQ)
480 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
481 else
482 XFRM_INC_STATS(net,
483 LINUX_MIB_XFRMINSTATEINVALID);
484
485 if (encap_type == -1)
486 dev_put(skb->dev);
487 goto drop;
488 }
489
490 family = x->props.family;
491
492 /* An encap_type of -1 indicates async resumption. */
493 if (encap_type == -1) {
494 async = 1;
495 seq = XFRM_SKB_CB(skb)->seq.input.low;
496 goto resume;
497 }
498 /* GRO call */
499 seq = XFRM_SPI_SKB_CB(skb)->seq;
500
501 if (xo && (xo->flags & CRYPTO_DONE)) {
502 crypto_done = true;
503 family = XFRM_SPI_SKB_CB(skb)->family;
504
505 if (!(xo->status & CRYPTO_SUCCESS)) {
506 if (xo->status &
507 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
508 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
509 CRYPTO_TUNNEL_AH_AUTH_FAILED |
510 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
511
512 xfrm_audit_state_icvfail(x, skb,
513 x->type->proto);
514 x->stats.integrity_failed++;
515 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
516 goto drop;
517 }
518
519 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
520 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
521 goto drop;
522 }
523
524 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
525 goto drop;
526 }
527
528 if (xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
529 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
530 goto drop;
531 }
532 }
533
534 goto lock;
535 }
536
537 family = XFRM_SPI_SKB_CB(skb)->family;
538
539 /* if tunnel is present override skb->mark value with tunnel i_key */
540 switch (family) {
541 case AF_INET:
542 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
543 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
544 break;
545 case AF_INET6:
546 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
547 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
548 break;
549 }
550
551 sp = secpath_set(skb);
552 if (!sp) {
553 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
554 goto drop;
555 }
556
557 seq = 0;
558 if (!spi && xfrm_parse_spi(skb, nexthdr, &spi, &seq)) {
559 secpath_reset(skb);
560 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
561 goto drop;
562 }
563
564 daddr = (xfrm_address_t *)(skb_network_header(skb) +
565 XFRM_SPI_SKB_CB(skb)->daddroff);
566 do {
567 sp = skb_sec_path(skb);
568
569 if (sp->len == XFRM_MAX_DEPTH) {
570 secpath_reset(skb);
571 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
572 goto drop;
573 }
574
575 x = xfrm_input_state_lookup(net, mark, daddr, spi, nexthdr, family);
576 if (x == NULL) {
577 secpath_reset(skb);
578 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
579 xfrm_audit_state_notfound(skb, family, spi, seq);
580 goto drop;
581 }
582
583 if (unlikely(x->dir && x->dir != XFRM_SA_DIR_IN)) {
584 secpath_reset(skb);
585 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEDIRERROR);
586 xfrm_audit_state_notfound(skb, family, spi, seq);
587 xfrm_state_put(x);
588 x = NULL;
589 goto drop;
590 }
591
592 skb->mark = xfrm_smark_get(skb->mark, x);
593
594 sp->xvec[sp->len++] = x;
595
596 skb_dst_force(skb);
597 if (!skb_dst(skb)) {
598 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
599 goto drop;
600 }
601
602lock:
603 spin_lock(&x->lock);
604
605 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
606 if (x->km.state == XFRM_STATE_ACQ)
607 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
608 else
609 XFRM_INC_STATS(net,
610 LINUX_MIB_XFRMINSTATEINVALID);
611 goto drop_unlock;
612 }
613
614 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
615 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
616 goto drop_unlock;
617 }
618
619 if (xfrm_replay_check(x, skb, seq)) {
620 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
621 goto drop_unlock;
622 }
623
624 if (xfrm_state_check_expire(x)) {
625 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
626 goto drop_unlock;
627 }
628
629 spin_unlock(&x->lock);
630
631 if (xfrm_tunnel_check(skb, x, family)) {
632 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
633 goto drop;
634 }
635
636 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
637
638 XFRM_SKB_CB(skb)->seq.input.low = seq;
639 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
640
641 dev_hold(skb->dev);
642
643 if (crypto_done)
644 nexthdr = x->type_offload->input_tail(x, skb);
645 else
646 nexthdr = x->type->input(x, skb);
647
648 if (nexthdr == -EINPROGRESS)
649 return 0;
650resume:
651 dev_put(skb->dev);
652
653 spin_lock(&x->lock);
654 if (nexthdr < 0) {
655 if (nexthdr == -EBADMSG) {
656 xfrm_audit_state_icvfail(x, skb,
657 x->type->proto);
658 x->stats.integrity_failed++;
659 }
660 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
661 goto drop_unlock;
662 }
663
664 /* only the first xfrm gets the encap type */
665 encap_type = 0;
666
667 if (xfrm_replay_recheck(x, skb, seq)) {
668 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
669 goto drop_unlock;
670 }
671
672 xfrm_replay_advance(x, seq);
673
674 x->curlft.bytes += skb->len;
675 x->curlft.packets++;
676 x->lastused = ktime_get_real_seconds();
677
678 spin_unlock(&x->lock);
679
680 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
681
682 if (xfrm_inner_mode_input(x, skb)) {
683 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
684 goto drop;
685 }
686
687 if (x->outer_mode.flags & XFRM_MODE_FLAG_TUNNEL) {
688 decaps = 1;
689 break;
690 }
691
692 /*
693 * We need the inner address. However, we only get here for
694 * transport mode so the outer address is identical.
695 */
696 daddr = &x->id.daddr;
697 family = x->props.family;
698
699 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
700 if (err < 0) {
701 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
702 goto drop;
703 }
704 crypto_done = false;
705 } while (!err);
706
707 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
708 if (err)
709 goto drop;
710
711 nf_reset_ct(skb);
712
713 if (decaps) {
714 sp = skb_sec_path(skb);
715 if (sp)
716 sp->olen = 0;
717 if (skb_valid_dst(skb))
718 skb_dst_drop(skb);
719 gro_cells_receive(&gro_cells, skb);
720 return 0;
721 } else {
722 xo = xfrm_offload(skb);
723 if (xo)
724 xfrm_gro = xo->flags & XFRM_GRO;
725
726 err = -EAFNOSUPPORT;
727 rcu_read_lock();
728 afinfo = xfrm_state_afinfo_get_rcu(x->props.family);
729 if (likely(afinfo))
730 err = afinfo->transport_finish(skb, xfrm_gro || async);
731 rcu_read_unlock();
732 if (xfrm_gro) {
733 sp = skb_sec_path(skb);
734 if (sp)
735 sp->olen = 0;
736 if (skb_valid_dst(skb))
737 skb_dst_drop(skb);
738 gro_cells_receive(&gro_cells, skb);
739 return err;
740 }
741
742 return err;
743 }
744
745drop_unlock:
746 spin_unlock(&x->lock);
747drop:
748 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
749 kfree_skb(skb);
750 return 0;
751}
752EXPORT_SYMBOL(xfrm_input);
753
754int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
755{
756 return xfrm_input(skb, nexthdr, 0, -1);
757}
758EXPORT_SYMBOL(xfrm_input_resume);
759
760static void xfrm_trans_reinject(struct work_struct *work)
761{
762 struct xfrm_trans_tasklet *trans = container_of(work, struct xfrm_trans_tasklet, work);
763 struct sk_buff_head queue;
764 struct sk_buff *skb;
765
766 __skb_queue_head_init(&queue);
767 spin_lock_bh(&trans->queue_lock);
768 skb_queue_splice_init(&trans->queue, &queue);
769 spin_unlock_bh(&trans->queue_lock);
770
771 local_bh_disable();
772 while ((skb = __skb_dequeue(&queue)))
773 XFRM_TRANS_SKB_CB(skb)->finish(XFRM_TRANS_SKB_CB(skb)->net,
774 NULL, skb);
775 local_bh_enable();
776}
777
778int xfrm_trans_queue_net(struct net *net, struct sk_buff *skb,
779 int (*finish)(struct net *, struct sock *,
780 struct sk_buff *))
781{
782 struct xfrm_trans_tasklet *trans;
783
784 trans = this_cpu_ptr(&xfrm_trans_tasklet);
785
786 if (skb_queue_len(&trans->queue) >= READ_ONCE(net_hotdata.max_backlog))
787 return -ENOBUFS;
788
789 BUILD_BUG_ON(sizeof(struct xfrm_trans_cb) > sizeof(skb->cb));
790
791 XFRM_TRANS_SKB_CB(skb)->finish = finish;
792 XFRM_TRANS_SKB_CB(skb)->net = net;
793 spin_lock_bh(&trans->queue_lock);
794 __skb_queue_tail(&trans->queue, skb);
795 spin_unlock_bh(&trans->queue_lock);
796 schedule_work(&trans->work);
797 return 0;
798}
799EXPORT_SYMBOL(xfrm_trans_queue_net);
800
801int xfrm_trans_queue(struct sk_buff *skb,
802 int (*finish)(struct net *, struct sock *,
803 struct sk_buff *))
804{
805 return xfrm_trans_queue_net(dev_net(skb->dev), skb, finish);
806}
807EXPORT_SYMBOL(xfrm_trans_queue);
808
809void __init xfrm_input_init(void)
810{
811 int err;
812 int i;
813
814 init_dummy_netdev(&xfrm_napi_dev);
815 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
816 if (err)
817 gro_cells.cells = NULL;
818
819 for_each_possible_cpu(i) {
820 struct xfrm_trans_tasklet *trans;
821
822 trans = &per_cpu(xfrm_trans_tasklet, i);
823 spin_lock_init(&trans->queue_lock);
824 __skb_queue_head_init(&trans->queue);
825 INIT_WORK(&trans->work, xfrm_trans_reinject);
826 }
827}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * xfrm_input.c
4 *
5 * Changes:
6 * YOSHIFUJI Hideaki @USAGI
7 * Split up af-specific portion
8 *
9 */
10
11#include <linux/bottom_half.h>
12#include <linux/cache.h>
13#include <linux/interrupt.h>
14#include <linux/slab.h>
15#include <linux/module.h>
16#include <linux/netdevice.h>
17#include <linux/percpu.h>
18#include <net/dst.h>
19#include <net/ip.h>
20#include <net/xfrm.h>
21#include <net/ip_tunnels.h>
22#include <net/ip6_tunnel.h>
23
24struct xfrm_trans_tasklet {
25 struct tasklet_struct tasklet;
26 struct sk_buff_head queue;
27};
28
29struct xfrm_trans_cb {
30 union {
31 struct inet_skb_parm h4;
32#if IS_ENABLED(CONFIG_IPV6)
33 struct inet6_skb_parm h6;
34#endif
35 } header;
36 int (*finish)(struct net *net, struct sock *sk, struct sk_buff *skb);
37};
38
39#define XFRM_TRANS_SKB_CB(__skb) ((struct xfrm_trans_cb *)&((__skb)->cb[0]))
40
41static struct kmem_cache *secpath_cachep __ro_after_init;
42
43static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
44static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[AF_INET6 + 1];
45
46static struct gro_cells gro_cells;
47static struct net_device xfrm_napi_dev;
48
49static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
50
51int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
52{
53 int err = 0;
54
55 if (WARN_ON(afinfo->family >= ARRAY_SIZE(xfrm_input_afinfo)))
56 return -EAFNOSUPPORT;
57
58 spin_lock_bh(&xfrm_input_afinfo_lock);
59 if (unlikely(xfrm_input_afinfo[afinfo->family] != NULL))
60 err = -EEXIST;
61 else
62 rcu_assign_pointer(xfrm_input_afinfo[afinfo->family], afinfo);
63 spin_unlock_bh(&xfrm_input_afinfo_lock);
64 return err;
65}
66EXPORT_SYMBOL(xfrm_input_register_afinfo);
67
68int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
69{
70 int err = 0;
71
72 spin_lock_bh(&xfrm_input_afinfo_lock);
73 if (likely(xfrm_input_afinfo[afinfo->family] != NULL)) {
74 if (unlikely(xfrm_input_afinfo[afinfo->family] != afinfo))
75 err = -EINVAL;
76 else
77 RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->family], NULL);
78 }
79 spin_unlock_bh(&xfrm_input_afinfo_lock);
80 synchronize_rcu();
81 return err;
82}
83EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
84
85static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(unsigned int family)
86{
87 const struct xfrm_input_afinfo *afinfo;
88
89 if (WARN_ON_ONCE(family >= ARRAY_SIZE(xfrm_input_afinfo)))
90 return NULL;
91
92 rcu_read_lock();
93 afinfo = rcu_dereference(xfrm_input_afinfo[family]);
94 if (unlikely(!afinfo))
95 rcu_read_unlock();
96 return afinfo;
97}
98
99static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
100 int err)
101{
102 int ret;
103 const struct xfrm_input_afinfo *afinfo = xfrm_input_get_afinfo(family);
104
105 if (!afinfo)
106 return -EAFNOSUPPORT;
107
108 ret = afinfo->callback(skb, protocol, err);
109 rcu_read_unlock();
110
111 return ret;
112}
113
114void __secpath_destroy(struct sec_path *sp)
115{
116 int i;
117 for (i = 0; i < sp->len; i++)
118 xfrm_state_put(sp->xvec[i]);
119 kmem_cache_free(secpath_cachep, sp);
120}
121EXPORT_SYMBOL(__secpath_destroy);
122
123struct sec_path *secpath_dup(struct sec_path *src)
124{
125 struct sec_path *sp;
126
127 sp = kmem_cache_alloc(secpath_cachep, GFP_ATOMIC);
128 if (!sp)
129 return NULL;
130
131 sp->len = 0;
132 sp->olen = 0;
133
134 memset(sp->ovec, 0, sizeof(sp->ovec[XFRM_MAX_OFFLOAD_DEPTH]));
135
136 if (src) {
137 int i;
138
139 memcpy(sp, src, sizeof(*sp));
140 for (i = 0; i < sp->len; i++)
141 xfrm_state_hold(sp->xvec[i]);
142 }
143 refcount_set(&sp->refcnt, 1);
144 return sp;
145}
146EXPORT_SYMBOL(secpath_dup);
147
148int secpath_set(struct sk_buff *skb)
149{
150 struct sec_path *sp;
151
152 /* Allocate new secpath or COW existing one. */
153 if (!skb->sp || refcount_read(&skb->sp->refcnt) != 1) {
154 sp = secpath_dup(skb->sp);
155 if (!sp)
156 return -ENOMEM;
157
158 if (skb->sp)
159 secpath_put(skb->sp);
160 skb->sp = sp;
161 }
162 return 0;
163}
164EXPORT_SYMBOL(secpath_set);
165
166/* Fetch spi and seq from ipsec header */
167
168int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
169{
170 int offset, offset_seq;
171 int hlen;
172
173 switch (nexthdr) {
174 case IPPROTO_AH:
175 hlen = sizeof(struct ip_auth_hdr);
176 offset = offsetof(struct ip_auth_hdr, spi);
177 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
178 break;
179 case IPPROTO_ESP:
180 hlen = sizeof(struct ip_esp_hdr);
181 offset = offsetof(struct ip_esp_hdr, spi);
182 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
183 break;
184 case IPPROTO_COMP:
185 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
186 return -EINVAL;
187 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
188 *seq = 0;
189 return 0;
190 default:
191 return 1;
192 }
193
194 if (!pskb_may_pull(skb, hlen))
195 return -EINVAL;
196
197 *spi = *(__be32 *)(skb_transport_header(skb) + offset);
198 *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
199 return 0;
200}
201EXPORT_SYMBOL(xfrm_parse_spi);
202
203int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
204{
205 struct xfrm_mode *inner_mode = x->inner_mode;
206 int err;
207
208 err = x->outer_mode->afinfo->extract_input(x, skb);
209 if (err)
210 return err;
211
212 if (x->sel.family == AF_UNSPEC) {
213 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
214 if (inner_mode == NULL)
215 return -EAFNOSUPPORT;
216 }
217
218 skb->protocol = inner_mode->afinfo->eth_proto;
219 return inner_mode->input2(x, skb);
220}
221EXPORT_SYMBOL(xfrm_prepare_input);
222
223int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
224{
225 struct net *net = dev_net(skb->dev);
226 int err;
227 __be32 seq;
228 __be32 seq_hi;
229 struct xfrm_state *x = NULL;
230 xfrm_address_t *daddr;
231 struct xfrm_mode *inner_mode;
232 u32 mark = skb->mark;
233 unsigned int family = AF_UNSPEC;
234 int decaps = 0;
235 int async = 0;
236 bool xfrm_gro = false;
237 bool crypto_done = false;
238 struct xfrm_offload *xo = xfrm_offload(skb);
239
240 if (encap_type < 0) {
241 x = xfrm_input_state(skb);
242
243 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
244 if (x->km.state == XFRM_STATE_ACQ)
245 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
246 else
247 XFRM_INC_STATS(net,
248 LINUX_MIB_XFRMINSTATEINVALID);
249 goto drop;
250 }
251
252 family = x->outer_mode->afinfo->family;
253
254 /* An encap_type of -1 indicates async resumption. */
255 if (encap_type == -1) {
256 async = 1;
257 seq = XFRM_SKB_CB(skb)->seq.input.low;
258 goto resume;
259 }
260
261 /* encap_type < -1 indicates a GRO call. */
262 encap_type = 0;
263 seq = XFRM_SPI_SKB_CB(skb)->seq;
264
265 if (xo && (xo->flags & CRYPTO_DONE)) {
266 crypto_done = true;
267 family = XFRM_SPI_SKB_CB(skb)->family;
268
269 if (!(xo->status & CRYPTO_SUCCESS)) {
270 if (xo->status &
271 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
272 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
273 CRYPTO_TUNNEL_AH_AUTH_FAILED |
274 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
275
276 xfrm_audit_state_icvfail(x, skb,
277 x->type->proto);
278 x->stats.integrity_failed++;
279 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
280 goto drop;
281 }
282
283 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
284 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
285 goto drop;
286 }
287
288 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
289 goto drop;
290 }
291
292 if ((err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
293 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
294 goto drop;
295 }
296 }
297
298 goto lock;
299 }
300
301 family = XFRM_SPI_SKB_CB(skb)->family;
302
303 /* if tunnel is present override skb->mark value with tunnel i_key */
304 switch (family) {
305 case AF_INET:
306 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
307 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
308 break;
309 case AF_INET6:
310 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
311 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
312 break;
313 }
314
315 err = secpath_set(skb);
316 if (err) {
317 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
318 goto drop;
319 }
320
321 seq = 0;
322 if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
323 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
324 goto drop;
325 }
326
327 daddr = (xfrm_address_t *)(skb_network_header(skb) +
328 XFRM_SPI_SKB_CB(skb)->daddroff);
329 do {
330 if (skb->sp->len == XFRM_MAX_DEPTH) {
331 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
332 goto drop;
333 }
334
335 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
336 if (x == NULL) {
337 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
338 xfrm_audit_state_notfound(skb, family, spi, seq);
339 goto drop;
340 }
341
342 skb->sp->xvec[skb->sp->len++] = x;
343
344lock:
345 spin_lock(&x->lock);
346
347 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
348 if (x->km.state == XFRM_STATE_ACQ)
349 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
350 else
351 XFRM_INC_STATS(net,
352 LINUX_MIB_XFRMINSTATEINVALID);
353 goto drop_unlock;
354 }
355
356 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
357 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
358 goto drop_unlock;
359 }
360
361 if (x->repl->check(x, skb, seq)) {
362 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
363 goto drop_unlock;
364 }
365
366 if (xfrm_state_check_expire(x)) {
367 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
368 goto drop_unlock;
369 }
370
371 spin_unlock(&x->lock);
372
373 if (xfrm_tunnel_check(skb, x, family)) {
374 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
375 goto drop;
376 }
377
378 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
379
380 XFRM_SKB_CB(skb)->seq.input.low = seq;
381 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
382
383 skb_dst_force(skb);
384 dev_hold(skb->dev);
385
386 if (crypto_done)
387 nexthdr = x->type_offload->input_tail(x, skb);
388 else
389 nexthdr = x->type->input(x, skb);
390
391 if (nexthdr == -EINPROGRESS)
392 return 0;
393resume:
394 dev_put(skb->dev);
395
396 spin_lock(&x->lock);
397 if (nexthdr <= 0) {
398 if (nexthdr == -EBADMSG) {
399 xfrm_audit_state_icvfail(x, skb,
400 x->type->proto);
401 x->stats.integrity_failed++;
402 }
403 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
404 goto drop_unlock;
405 }
406
407 /* only the first xfrm gets the encap type */
408 encap_type = 0;
409
410 if (async && x->repl->recheck(x, skb, seq)) {
411 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
412 goto drop_unlock;
413 }
414
415 x->repl->advance(x, seq);
416
417 x->curlft.bytes += skb->len;
418 x->curlft.packets++;
419
420 spin_unlock(&x->lock);
421
422 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
423
424 inner_mode = x->inner_mode;
425
426 if (x->sel.family == AF_UNSPEC) {
427 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
428 if (inner_mode == NULL) {
429 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
430 goto drop;
431 }
432 }
433
434 if (inner_mode->input(x, skb)) {
435 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
436 goto drop;
437 }
438
439 if (x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) {
440 decaps = 1;
441 break;
442 }
443
444 /*
445 * We need the inner address. However, we only get here for
446 * transport mode so the outer address is identical.
447 */
448 daddr = &x->id.daddr;
449 family = x->outer_mode->afinfo->family;
450
451 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
452 if (err < 0) {
453 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
454 goto drop;
455 }
456 } while (!err);
457
458 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
459 if (err)
460 goto drop;
461
462 nf_reset(skb);
463
464 if (decaps) {
465 if (skb->sp)
466 skb->sp->olen = 0;
467 skb_dst_drop(skb);
468 gro_cells_receive(&gro_cells, skb);
469 return 0;
470 } else {
471 xo = xfrm_offload(skb);
472 if (xo)
473 xfrm_gro = xo->flags & XFRM_GRO;
474
475 err = x->inner_mode->afinfo->transport_finish(skb, xfrm_gro || async);
476 if (xfrm_gro) {
477 if (skb->sp)
478 skb->sp->olen = 0;
479 skb_dst_drop(skb);
480 gro_cells_receive(&gro_cells, skb);
481 return err;
482 }
483
484 return err;
485 }
486
487drop_unlock:
488 spin_unlock(&x->lock);
489drop:
490 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
491 kfree_skb(skb);
492 return 0;
493}
494EXPORT_SYMBOL(xfrm_input);
495
496int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
497{
498 return xfrm_input(skb, nexthdr, 0, -1);
499}
500EXPORT_SYMBOL(xfrm_input_resume);
501
502static void xfrm_trans_reinject(unsigned long data)
503{
504 struct xfrm_trans_tasklet *trans = (void *)data;
505 struct sk_buff_head queue;
506 struct sk_buff *skb;
507
508 __skb_queue_head_init(&queue);
509 skb_queue_splice_init(&trans->queue, &queue);
510
511 while ((skb = __skb_dequeue(&queue)))
512 XFRM_TRANS_SKB_CB(skb)->finish(dev_net(skb->dev), NULL, skb);
513}
514
515int xfrm_trans_queue(struct sk_buff *skb,
516 int (*finish)(struct net *, struct sock *,
517 struct sk_buff *))
518{
519 struct xfrm_trans_tasklet *trans;
520
521 trans = this_cpu_ptr(&xfrm_trans_tasklet);
522
523 if (skb_queue_len(&trans->queue) >= netdev_max_backlog)
524 return -ENOBUFS;
525
526 XFRM_TRANS_SKB_CB(skb)->finish = finish;
527 __skb_queue_tail(&trans->queue, skb);
528 tasklet_schedule(&trans->tasklet);
529 return 0;
530}
531EXPORT_SYMBOL(xfrm_trans_queue);
532
533void __init xfrm_input_init(void)
534{
535 int err;
536 int i;
537
538 init_dummy_netdev(&xfrm_napi_dev);
539 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
540 if (err)
541 gro_cells.cells = NULL;
542
543 secpath_cachep = kmem_cache_create("secpath_cache",
544 sizeof(struct sec_path),
545 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
546 NULL);
547
548 for_each_possible_cpu(i) {
549 struct xfrm_trans_tasklet *trans;
550
551 trans = &per_cpu(xfrm_trans_tasklet, i);
552 __skb_queue_head_init(&trans->queue);
553 tasklet_init(&trans->tasklet, xfrm_trans_reinject,
554 (unsigned long)trans);
555 }
556}