1 // SPDX-License-Identifier: GPL-2.0
6 * YOSHIFUJI Hideaki @USAGI
7 * Split up af-specific portion
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>
21 #include <net/ip_tunnels.h>
22 #include <net/ip6_tunnel.h>
24 struct xfrm_trans_tasklet {
25 struct tasklet_struct tasklet;
26 struct sk_buff_head queue;
29 struct xfrm_trans_cb {
31 struct inet_skb_parm h4;
32 #if IS_ENABLED(CONFIG_IPV6)
33 struct inet6_skb_parm h6;
36 int (*finish)(struct net *net, struct sock *sk, struct sk_buff *skb);
39 #define XFRM_TRANS_SKB_CB(__skb) ((struct xfrm_trans_cb *)&((__skb)->cb[0]))
41 static struct kmem_cache *secpath_cachep __ro_after_init;
43 static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
44 static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[AF_INET6 + 1];
46 static struct gro_cells gro_cells;
47 static struct net_device xfrm_napi_dev;
49 static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
51 int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
55 if (WARN_ON(afinfo->family >= ARRAY_SIZE(xfrm_input_afinfo)))
58 spin_lock_bh(&xfrm_input_afinfo_lock);
59 if (unlikely(xfrm_input_afinfo[afinfo->family] != NULL))
62 rcu_assign_pointer(xfrm_input_afinfo[afinfo->family], afinfo);
63 spin_unlock_bh(&xfrm_input_afinfo_lock);
66 EXPORT_SYMBOL(xfrm_input_register_afinfo);
68 int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
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))
77 RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->family], NULL);
79 spin_unlock_bh(&xfrm_input_afinfo_lock);
83 EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
85 static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(unsigned int family)
87 const struct xfrm_input_afinfo *afinfo;
89 if (WARN_ON_ONCE(family >= ARRAY_SIZE(xfrm_input_afinfo)))
93 afinfo = rcu_dereference(xfrm_input_afinfo[family]);
94 if (unlikely(!afinfo))
99 static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
103 const struct xfrm_input_afinfo *afinfo = xfrm_input_get_afinfo(family);
106 return -EAFNOSUPPORT;
108 ret = afinfo->callback(skb, protocol, err);
114 void __secpath_destroy(struct sec_path *sp)
117 for (i = 0; i < sp->len; i++)
118 xfrm_state_put(sp->xvec[i]);
119 kmem_cache_free(secpath_cachep, sp);
121 EXPORT_SYMBOL(__secpath_destroy);
123 struct sec_path *secpath_dup(struct sec_path *src)
127 sp = kmem_cache_alloc(secpath_cachep, GFP_ATOMIC);
134 memset(sp->ovec, 0, sizeof(sp->ovec[XFRM_MAX_OFFLOAD_DEPTH]));
139 memcpy(sp, src, sizeof(*sp));
140 for (i = 0; i < sp->len; i++)
141 xfrm_state_hold(sp->xvec[i]);
143 refcount_set(&sp->refcnt, 1);
146 EXPORT_SYMBOL(secpath_dup);
148 int secpath_set(struct sk_buff *skb)
152 /* Allocate new secpath or COW existing one. */
153 if (!skb->sp || refcount_read(&skb->sp->refcnt) != 1) {
154 sp = secpath_dup(skb->sp);
159 secpath_put(skb->sp);
164 EXPORT_SYMBOL(secpath_set);
166 /* Fetch spi and seq from ipsec header */
168 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
170 int offset, offset_seq;
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);
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);
185 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
187 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
194 if (!pskb_may_pull(skb, hlen))
197 *spi = *(__be32 *)(skb_transport_header(skb) + offset);
198 *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
201 EXPORT_SYMBOL(xfrm_parse_spi);
203 int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
205 struct xfrm_mode *inner_mode = x->inner_mode;
208 err = x->outer_mode->afinfo->extract_input(x, skb);
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;
218 skb->protocol = inner_mode->afinfo->eth_proto;
219 return inner_mode->input2(x, skb);
221 EXPORT_SYMBOL(xfrm_prepare_input);
223 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
225 struct net *net = dev_net(skb->dev);
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;
236 bool xfrm_gro = false;
237 bool crypto_done = false;
238 struct xfrm_offload *xo = xfrm_offload(skb);
240 if (encap_type < 0) {
241 x = xfrm_input_state(skb);
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);
248 LINUX_MIB_XFRMINSTATEINVALID);
252 family = x->outer_mode->afinfo->family;
254 /* An encap_type of -1 indicates async resumption. */
255 if (encap_type == -1) {
257 seq = XFRM_SKB_CB(skb)->seq.input.low;
261 /* encap_type < -1 indicates a GRO call. */
263 seq = XFRM_SPI_SKB_CB(skb)->seq;
265 if (xo && (xo->flags & CRYPTO_DONE)) {
267 family = XFRM_SPI_SKB_CB(skb)->family;
269 if (!(xo->status & CRYPTO_SUCCESS)) {
271 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
272 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
273 CRYPTO_TUNNEL_AH_AUTH_FAILED |
274 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
276 xfrm_audit_state_icvfail(x, skb,
278 x->stats.integrity_failed++;
279 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
283 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
284 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
288 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
292 if ((err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
293 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
301 family = XFRM_SPI_SKB_CB(skb)->family;
303 /* if tunnel is present override skb->mark value with tunnel i_key */
306 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
307 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
310 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
311 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
315 err = secpath_set(skb);
317 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
322 if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
323 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
327 daddr = (xfrm_address_t *)(skb_network_header(skb) +
328 XFRM_SPI_SKB_CB(skb)->daddroff);
330 if (skb->sp->len == XFRM_MAX_DEPTH) {
331 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
335 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
337 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
338 xfrm_audit_state_notfound(skb, family, spi, seq);
342 skb->sp->xvec[skb->sp->len++] = x;
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);
352 LINUX_MIB_XFRMINSTATEINVALID);
356 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
357 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
361 if (x->repl->check(x, skb, seq)) {
362 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
366 if (xfrm_state_check_expire(x)) {
367 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
371 spin_unlock(&x->lock);
373 if (xfrm_tunnel_check(skb, x, family)) {
374 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
378 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
380 XFRM_SKB_CB(skb)->seq.input.low = seq;
381 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
387 nexthdr = x->type_offload->input_tail(x, skb);
389 nexthdr = x->type->input(x, skb);
391 if (nexthdr == -EINPROGRESS)
398 if (nexthdr == -EBADMSG) {
399 xfrm_audit_state_icvfail(x, skb,
401 x->stats.integrity_failed++;
403 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
407 /* only the first xfrm gets the encap type */
410 if (async && x->repl->recheck(x, skb, seq)) {
411 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
415 x->repl->advance(x, seq);
417 x->curlft.bytes += skb->len;
420 spin_unlock(&x->lock);
422 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
424 inner_mode = x->inner_mode;
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);
434 if (inner_mode->input(x, skb)) {
435 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
439 if (x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) {
445 * We need the inner address. However, we only get here for
446 * transport mode so the outer address is identical.
448 daddr = &x->id.daddr;
449 family = x->outer_mode->afinfo->family;
451 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
453 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
458 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
468 gro_cells_receive(&gro_cells, skb);
471 xo = xfrm_offload(skb);
473 xfrm_gro = xo->flags & XFRM_GRO;
475 err = x->inner_mode->afinfo->transport_finish(skb, xfrm_gro || async);
480 gro_cells_receive(&gro_cells, skb);
488 spin_unlock(&x->lock);
490 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
494 EXPORT_SYMBOL(xfrm_input);
496 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
498 return xfrm_input(skb, nexthdr, 0, -1);
500 EXPORT_SYMBOL(xfrm_input_resume);
502 static void xfrm_trans_reinject(unsigned long data)
504 struct xfrm_trans_tasklet *trans = (void *)data;
505 struct sk_buff_head queue;
508 __skb_queue_head_init(&queue);
509 skb_queue_splice_init(&trans->queue, &queue);
511 while ((skb = __skb_dequeue(&queue)))
512 XFRM_TRANS_SKB_CB(skb)->finish(dev_net(skb->dev), NULL, skb);
515 int xfrm_trans_queue(struct sk_buff *skb,
516 int (*finish)(struct net *, struct sock *,
519 struct xfrm_trans_tasklet *trans;
521 trans = this_cpu_ptr(&xfrm_trans_tasklet);
523 if (skb_queue_len(&trans->queue) >= netdev_max_backlog)
526 XFRM_TRANS_SKB_CB(skb)->finish = finish;
527 __skb_queue_tail(&trans->queue, skb);
528 tasklet_schedule(&trans->tasklet);
531 EXPORT_SYMBOL(xfrm_trans_queue);
533 void __init xfrm_input_init(void)
538 init_dummy_netdev(&xfrm_napi_dev);
539 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
541 gro_cells.cells = NULL;
543 secpath_cachep = kmem_cache_create("secpath_cache",
544 sizeof(struct sec_path),
545 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
548 for_each_possible_cpu(i) {
549 struct xfrm_trans_tasklet *trans;
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);