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/interrupt.h>
13 #include <linux/slab.h>
14 #include <linux/module.h>
15 #include <linux/netdevice.h>
16 #include <linux/percpu.h>
20 #include <net/ip_tunnels.h>
21 #include <net/ip6_tunnel.h>
23 struct xfrm_trans_tasklet {
24 struct tasklet_struct tasklet;
25 struct sk_buff_head queue;
28 struct xfrm_trans_cb {
30 struct inet_skb_parm h4;
31 #if IS_ENABLED(CONFIG_IPV6)
32 struct inet6_skb_parm h6;
35 int (*finish)(struct net *net, struct sock *sk, struct sk_buff *skb);
38 #define XFRM_TRANS_SKB_CB(__skb) ((struct xfrm_trans_cb *)&((__skb)->cb[0]))
40 static struct kmem_cache *secpath_cachep __read_mostly;
42 static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
43 static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[AF_INET6 + 1];
45 static struct gro_cells gro_cells;
46 static struct net_device xfrm_napi_dev;
48 static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
50 int xfrm_input_register_afinfo(const struct xfrm_input_afinfo *afinfo)
54 if (WARN_ON(afinfo->family >= ARRAY_SIZE(xfrm_input_afinfo)))
57 spin_lock_bh(&xfrm_input_afinfo_lock);
58 if (unlikely(xfrm_input_afinfo[afinfo->family] != NULL))
61 rcu_assign_pointer(xfrm_input_afinfo[afinfo->family], afinfo);
62 spin_unlock_bh(&xfrm_input_afinfo_lock);
65 EXPORT_SYMBOL(xfrm_input_register_afinfo);
67 int xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo *afinfo)
71 spin_lock_bh(&xfrm_input_afinfo_lock);
72 if (likely(xfrm_input_afinfo[afinfo->family] != NULL)) {
73 if (unlikely(xfrm_input_afinfo[afinfo->family] != afinfo))
76 RCU_INIT_POINTER(xfrm_input_afinfo[afinfo->family], NULL);
78 spin_unlock_bh(&xfrm_input_afinfo_lock);
82 EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
84 static const struct xfrm_input_afinfo *xfrm_input_get_afinfo(unsigned int family)
86 const struct xfrm_input_afinfo *afinfo;
88 if (WARN_ON_ONCE(family >= ARRAY_SIZE(xfrm_input_afinfo)))
92 afinfo = rcu_dereference(xfrm_input_afinfo[family]);
93 if (unlikely(!afinfo))
98 static int xfrm_rcv_cb(struct sk_buff *skb, unsigned int family, u8 protocol,
102 const struct xfrm_input_afinfo *afinfo = xfrm_input_get_afinfo(family);
105 return -EAFNOSUPPORT;
107 ret = afinfo->callback(skb, protocol, err);
113 void __secpath_destroy(struct sec_path *sp)
116 for (i = 0; i < sp->len; i++)
117 xfrm_state_put(sp->xvec[i]);
118 kmem_cache_free(secpath_cachep, sp);
120 EXPORT_SYMBOL(__secpath_destroy);
122 struct sec_path *secpath_dup(struct sec_path *src)
126 sp = kmem_cache_alloc(secpath_cachep, GFP_ATOMIC);
133 memset(sp->ovec, 0, sizeof(sp->ovec));
138 memcpy(sp, src, sizeof(*sp));
139 for (i = 0; i < sp->len; i++)
140 xfrm_state_hold(sp->xvec[i]);
142 refcount_set(&sp->refcnt, 1);
145 EXPORT_SYMBOL(secpath_dup);
147 int secpath_set(struct sk_buff *skb)
151 /* Allocate new secpath or COW existing one. */
152 if (!skb->sp || refcount_read(&skb->sp->refcnt) != 1) {
153 sp = secpath_dup(skb->sp);
158 secpath_put(skb->sp);
163 EXPORT_SYMBOL(secpath_set);
165 /* Fetch spi and seq from ipsec header */
167 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
169 int offset, offset_seq;
174 hlen = sizeof(struct ip_auth_hdr);
175 offset = offsetof(struct ip_auth_hdr, spi);
176 offset_seq = offsetof(struct ip_auth_hdr, seq_no);
179 hlen = sizeof(struct ip_esp_hdr);
180 offset = offsetof(struct ip_esp_hdr, spi);
181 offset_seq = offsetof(struct ip_esp_hdr, seq_no);
184 if (!pskb_may_pull(skb, sizeof(struct ip_comp_hdr)))
186 *spi = htonl(ntohs(*(__be16 *)(skb_transport_header(skb) + 2)));
193 if (!pskb_may_pull(skb, hlen))
196 *spi = *(__be32 *)(skb_transport_header(skb) + offset);
197 *seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
200 EXPORT_SYMBOL(xfrm_parse_spi);
202 int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
204 struct xfrm_mode *inner_mode = x->inner_mode;
207 err = x->outer_mode->afinfo->extract_input(x, skb);
211 if (x->sel.family == AF_UNSPEC) {
212 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
213 if (inner_mode == NULL)
214 return -EAFNOSUPPORT;
217 skb->protocol = inner_mode->afinfo->eth_proto;
218 return inner_mode->input2(x, skb);
220 EXPORT_SYMBOL(xfrm_prepare_input);
222 int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
224 struct net *net = dev_net(skb->dev);
228 struct xfrm_state *x = NULL;
229 xfrm_address_t *daddr;
230 struct xfrm_mode *inner_mode;
231 u32 mark = skb->mark;
232 unsigned int family = AF_UNSPEC;
235 bool xfrm_gro = false;
236 bool crypto_done = false;
237 struct xfrm_offload *xo = xfrm_offload(skb);
239 if (encap_type < 0) {
240 x = xfrm_input_state(skb);
242 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
243 if (x->km.state == XFRM_STATE_ACQ)
244 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
247 LINUX_MIB_XFRMINSTATEINVALID);
249 if (encap_type == -1)
254 family = x->outer_mode->afinfo->family;
256 /* An encap_type of -1 indicates async resumption. */
257 if (encap_type == -1) {
259 seq = XFRM_SKB_CB(skb)->seq.input.low;
263 /* encap_type < -1 indicates a GRO call. */
265 seq = XFRM_SPI_SKB_CB(skb)->seq;
267 if (xo && (xo->flags & CRYPTO_DONE)) {
269 x = xfrm_input_state(skb);
270 family = XFRM_SPI_SKB_CB(skb)->family;
272 if (!(xo->status & CRYPTO_SUCCESS)) {
274 (CRYPTO_TRANSPORT_AH_AUTH_FAILED |
275 CRYPTO_TRANSPORT_ESP_AUTH_FAILED |
276 CRYPTO_TUNNEL_AH_AUTH_FAILED |
277 CRYPTO_TUNNEL_ESP_AUTH_FAILED)) {
279 xfrm_audit_state_icvfail(x, skb,
281 x->stats.integrity_failed++;
282 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
286 if (xo->status & CRYPTO_INVALID_PROTOCOL) {
287 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
291 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
295 if ((err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
296 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
304 family = XFRM_SPI_SKB_CB(skb)->family;
306 /* if tunnel is present override skb->mark value with tunnel i_key */
309 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4)
310 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4->parms.i_key);
313 if (XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6)
314 mark = be32_to_cpu(XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6->parms.i_key);
318 err = secpath_set(skb);
320 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
325 if (!spi && (err = xfrm_parse_spi(skb, nexthdr, &spi, &seq)) != 0) {
326 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
330 daddr = (xfrm_address_t *)(skb_network_header(skb) +
331 XFRM_SPI_SKB_CB(skb)->daddroff);
333 if (skb->sp->len == XFRM_MAX_DEPTH) {
334 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
338 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
340 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
341 xfrm_audit_state_notfound(skb, family, spi, seq);
345 skb->sp->xvec[skb->sp->len++] = x;
349 XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
356 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
357 if (x->km.state == XFRM_STATE_ACQ)
358 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
361 LINUX_MIB_XFRMINSTATEINVALID);
365 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
366 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
370 if (x->repl->check(x, skb, seq)) {
371 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
375 if (xfrm_state_check_expire(x)) {
376 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
380 spin_unlock(&x->lock);
382 if (xfrm_tunnel_check(skb, x, family)) {
383 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
387 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
389 XFRM_SKB_CB(skb)->seq.input.low = seq;
390 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
395 nexthdr = x->type_offload->input_tail(x, skb);
397 nexthdr = x->type->input(x, skb);
399 if (nexthdr == -EINPROGRESS)
406 if (nexthdr == -EBADMSG) {
407 xfrm_audit_state_icvfail(x, skb,
409 x->stats.integrity_failed++;
411 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
415 /* only the first xfrm gets the encap type */
418 if (x->repl->recheck(x, skb, seq)) {
419 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
423 x->repl->advance(x, seq);
425 x->curlft.bytes += skb->len;
428 spin_unlock(&x->lock);
430 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
432 inner_mode = x->inner_mode;
434 if (x->sel.family == AF_UNSPEC) {
435 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
436 if (inner_mode == NULL) {
437 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
442 if (inner_mode->input(x, skb)) {
443 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
447 if (x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) {
453 * We need the inner address. However, we only get here for
454 * transport mode so the outer address is identical.
456 daddr = &x->id.daddr;
457 family = x->outer_mode->afinfo->family;
459 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
461 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
467 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
477 gro_cells_receive(&gro_cells, skb);
480 xo = xfrm_offload(skb);
482 xfrm_gro = xo->flags & XFRM_GRO;
484 err = x->inner_mode->afinfo->transport_finish(skb, xfrm_gro || async);
489 gro_cells_receive(&gro_cells, skb);
497 spin_unlock(&x->lock);
499 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
503 EXPORT_SYMBOL(xfrm_input);
505 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
507 return xfrm_input(skb, nexthdr, 0, -1);
509 EXPORT_SYMBOL(xfrm_input_resume);
511 static void xfrm_trans_reinject(unsigned long data)
513 struct xfrm_trans_tasklet *trans = (void *)data;
514 struct sk_buff_head queue;
517 __skb_queue_head_init(&queue);
518 skb_queue_splice_init(&trans->queue, &queue);
520 while ((skb = __skb_dequeue(&queue)))
521 XFRM_TRANS_SKB_CB(skb)->finish(dev_net(skb->dev), NULL, skb);
524 int xfrm_trans_queue(struct sk_buff *skb,
525 int (*finish)(struct net *, struct sock *,
528 struct xfrm_trans_tasklet *trans;
530 trans = this_cpu_ptr(&xfrm_trans_tasklet);
532 if (skb_queue_len(&trans->queue) >= netdev_max_backlog)
535 XFRM_TRANS_SKB_CB(skb)->finish = finish;
536 __skb_queue_tail(&trans->queue, skb);
537 tasklet_schedule(&trans->tasklet);
540 EXPORT_SYMBOL(xfrm_trans_queue);
542 void __init xfrm_input_init(void)
547 init_dummy_netdev(&xfrm_napi_dev);
548 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
550 gro_cells.cells = NULL;
552 secpath_cachep = kmem_cache_create("secpath_cache",
553 sizeof(struct sec_path),
554 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
557 for_each_possible_cpu(i) {
558 struct xfrm_trans_tasklet *trans;
560 trans = &per_cpu(xfrm_trans_tasklet, i);
561 __skb_queue_head_init(&trans->queue);
562 tasklet_init(&trans->tasklet, xfrm_trans_reinject,
563 (unsigned long)trans);