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
24 struct xfrm_trans_tasklet {
25 struct tasklet_struct tasklet;
26 struct sk_buff_head queue;
27 };
28
29 struct 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
41 static struct kmem_cache *secpath_cachep __ro_after_init;
42
43 static DEFINE_SPINLOCK(xfrm_input_afinfo_lock);
44 static struct xfrm_input_afinfo const __rcu *xfrm_input_afinfo[AF_INET6 + 1];
45
46 static struct gro_cells gro_cells;
47 static struct net_device xfrm_napi_dev;
48
49 static DEFINE_PER_CPU(struct xfrm_trans_tasklet, xfrm_trans_tasklet);
50
xfrm_input_register_afinfo(const struct xfrm_input_afinfo * afinfo)51 int 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 }
66 EXPORT_SYMBOL(xfrm_input_register_afinfo);
67
xfrm_input_unregister_afinfo(const struct xfrm_input_afinfo * afinfo)68 int 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 }
83 EXPORT_SYMBOL(xfrm_input_unregister_afinfo);
84
xfrm_input_get_afinfo(unsigned int family)85 static 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
xfrm_rcv_cb(struct sk_buff * skb,unsigned int family,u8 protocol,int err)99 static 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
__secpath_destroy(struct sec_path * sp)114 void __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 }
121 EXPORT_SYMBOL(__secpath_destroy);
122
secpath_dup(struct sec_path * src)123 struct 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 }
146 EXPORT_SYMBOL(secpath_dup);
147
secpath_set(struct sk_buff * skb)148 int 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 }
164 EXPORT_SYMBOL(secpath_set);
165
166 /* Fetch spi and seq from ipsec header */
167
xfrm_parse_spi(struct sk_buff * skb,u8 nexthdr,__be32 * spi,__be32 * seq)168 int 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 }
201 EXPORT_SYMBOL(xfrm_parse_spi);
202
xfrm_prepare_input(struct xfrm_state * x,struct sk_buff * skb)203 int 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 }
221 EXPORT_SYMBOL(xfrm_prepare_input);
222
xfrm_input(struct sk_buff * skb,int nexthdr,__be32 spi,int encap_type)223 int 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 secpath_reset(skb);
324 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
325 goto drop;
326 }
327
328 daddr = (xfrm_address_t *)(skb_network_header(skb) +
329 XFRM_SPI_SKB_CB(skb)->daddroff);
330 do {
331 if (skb->sp->len == XFRM_MAX_DEPTH) {
332 secpath_reset(skb);
333 XFRM_INC_STATS(net, LINUX_MIB_XFRMINBUFFERERROR);
334 goto drop;
335 }
336
337 x = xfrm_state_lookup(net, mark, daddr, spi, nexthdr, family);
338 if (x == NULL) {
339 secpath_reset(skb);
340 XFRM_INC_STATS(net, LINUX_MIB_XFRMINNOSTATES);
341 xfrm_audit_state_notfound(skb, family, spi, seq);
342 goto drop;
343 }
344
345 skb->mark = xfrm_smark_get(skb->mark, x);
346
347 skb->sp->xvec[skb->sp->len++] = x;
348
349 lock:
350 spin_lock(&x->lock);
351
352 if (unlikely(x->km.state != XFRM_STATE_VALID)) {
353 if (x->km.state == XFRM_STATE_ACQ)
354 XFRM_INC_STATS(net, LINUX_MIB_XFRMACQUIREERROR);
355 else
356 XFRM_INC_STATS(net,
357 LINUX_MIB_XFRMINSTATEINVALID);
358 goto drop_unlock;
359 }
360
361 if ((x->encap ? x->encap->encap_type : 0) != encap_type) {
362 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMISMATCH);
363 goto drop_unlock;
364 }
365
366 if (x->repl->check(x, skb, seq)) {
367 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
368 goto drop_unlock;
369 }
370
371 if (xfrm_state_check_expire(x)) {
372 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEEXPIRED);
373 goto drop_unlock;
374 }
375
376 spin_unlock(&x->lock);
377
378 if (xfrm_tunnel_check(skb, x, family)) {
379 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
380 goto drop;
381 }
382
383 seq_hi = htonl(xfrm_replay_seqhi(x, seq));
384
385 XFRM_SKB_CB(skb)->seq.input.low = seq;
386 XFRM_SKB_CB(skb)->seq.input.hi = seq_hi;
387
388 skb_dst_force(skb);
389 dev_hold(skb->dev);
390
391 if (crypto_done)
392 nexthdr = x->type_offload->input_tail(x, skb);
393 else
394 nexthdr = x->type->input(x, skb);
395
396 if (nexthdr == -EINPROGRESS)
397 return 0;
398 resume:
399 dev_put(skb->dev);
400
401 spin_lock(&x->lock);
402 if (nexthdr <= 0) {
403 if (nexthdr == -EBADMSG) {
404 xfrm_audit_state_icvfail(x, skb,
405 x->type->proto);
406 x->stats.integrity_failed++;
407 }
408 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEPROTOERROR);
409 goto drop_unlock;
410 }
411
412 /* only the first xfrm gets the encap type */
413 encap_type = 0;
414
415 if (async && x->repl->recheck(x, skb, seq)) {
416 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATESEQERROR);
417 goto drop_unlock;
418 }
419
420 x->repl->advance(x, seq);
421
422 x->curlft.bytes += skb->len;
423 x->curlft.packets++;
424
425 spin_unlock(&x->lock);
426
427 XFRM_MODE_SKB_CB(skb)->protocol = nexthdr;
428
429 inner_mode = x->inner_mode;
430
431 if (x->sel.family == AF_UNSPEC) {
432 inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
433 if (inner_mode == NULL) {
434 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
435 goto drop;
436 }
437 }
438
439 if (inner_mode->input(x, skb)) {
440 XFRM_INC_STATS(net, LINUX_MIB_XFRMINSTATEMODEERROR);
441 goto drop;
442 }
443
444 if (x->outer_mode->flags & XFRM_MODE_FLAG_TUNNEL) {
445 decaps = 1;
446 break;
447 }
448
449 /*
450 * We need the inner address. However, we only get here for
451 * transport mode so the outer address is identical.
452 */
453 daddr = &x->id.daddr;
454 family = x->outer_mode->afinfo->family;
455
456 err = xfrm_parse_spi(skb, nexthdr, &spi, &seq);
457 if (err < 0) {
458 XFRM_INC_STATS(net, LINUX_MIB_XFRMINHDRERROR);
459 goto drop;
460 }
461 crypto_done = false;
462 } while (!err);
463
464 err = xfrm_rcv_cb(skb, family, x->type->proto, 0);
465 if (err)
466 goto drop;
467
468 nf_reset(skb);
469
470 if (decaps) {
471 if (skb->sp)
472 skb->sp->olen = 0;
473 skb_dst_drop(skb);
474 gro_cells_receive(&gro_cells, skb);
475 return 0;
476 } else {
477 xo = xfrm_offload(skb);
478 if (xo)
479 xfrm_gro = xo->flags & XFRM_GRO;
480
481 err = x->inner_mode->afinfo->transport_finish(skb, xfrm_gro || async);
482 if (xfrm_gro) {
483 if (skb->sp)
484 skb->sp->olen = 0;
485 skb_dst_drop(skb);
486 gro_cells_receive(&gro_cells, skb);
487 return err;
488 }
489
490 return err;
491 }
492
493 drop_unlock:
494 spin_unlock(&x->lock);
495 drop:
496 xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
497 kfree_skb(skb);
498 return 0;
499 }
500 EXPORT_SYMBOL(xfrm_input);
501
xfrm_input_resume(struct sk_buff * skb,int nexthdr)502 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
503 {
504 return xfrm_input(skb, nexthdr, 0, -1);
505 }
506 EXPORT_SYMBOL(xfrm_input_resume);
507
xfrm_trans_reinject(unsigned long data)508 static void xfrm_trans_reinject(unsigned long data)
509 {
510 struct xfrm_trans_tasklet *trans = (void *)data;
511 struct sk_buff_head queue;
512 struct sk_buff *skb;
513
514 __skb_queue_head_init(&queue);
515 skb_queue_splice_init(&trans->queue, &queue);
516
517 while ((skb = __skb_dequeue(&queue)))
518 XFRM_TRANS_SKB_CB(skb)->finish(dev_net(skb->dev), NULL, skb);
519 }
520
xfrm_trans_queue(struct sk_buff * skb,int (* finish)(struct net *,struct sock *,struct sk_buff *))521 int xfrm_trans_queue(struct sk_buff *skb,
522 int (*finish)(struct net *, struct sock *,
523 struct sk_buff *))
524 {
525 struct xfrm_trans_tasklet *trans;
526
527 trans = this_cpu_ptr(&xfrm_trans_tasklet);
528
529 if (skb_queue_len(&trans->queue) >= netdev_max_backlog)
530 return -ENOBUFS;
531
532 XFRM_TRANS_SKB_CB(skb)->finish = finish;
533 __skb_queue_tail(&trans->queue, skb);
534 tasklet_schedule(&trans->tasklet);
535 return 0;
536 }
537 EXPORT_SYMBOL(xfrm_trans_queue);
538
xfrm_input_init(void)539 void __init xfrm_input_init(void)
540 {
541 int err;
542 int i;
543
544 init_dummy_netdev(&xfrm_napi_dev);
545 err = gro_cells_init(&gro_cells, &xfrm_napi_dev);
546 if (err)
547 gro_cells.cells = NULL;
548
549 secpath_cachep = kmem_cache_create("secpath_cache",
550 sizeof(struct sec_path),
551 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC,
552 NULL);
553
554 for_each_possible_cpu(i) {
555 struct xfrm_trans_tasklet *trans;
556
557 trans = &per_cpu(xfrm_trans_tasklet, i);
558 __skb_queue_head_init(&trans->queue);
559 tasklet_init(&trans->tasklet, xfrm_trans_reinject,
560 (unsigned long)trans);
561 }
562 }
563
