drm/amdgpu/vg20:Restruct uvd to support multiple uvds
[muen/linux.git] / net / xfrm / xfrm_input.c
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
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
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
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
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
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
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
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
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
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
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                 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
344 lock:
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;
393 resume:
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
487 drop_unlock:
488         spin_unlock(&x->lock);
489 drop:
490         xfrm_rcv_cb(skb, family, x && x->type ? x->type->proto : nexthdr, -1);
491         kfree_skb(skb);
492         return 0;
493 }
494 EXPORT_SYMBOL(xfrm_input);
495
496 int xfrm_input_resume(struct sk_buff *skb, int nexthdr)
497 {
498         return xfrm_input(skb, nexthdr, 0, -1);
499 }
500 EXPORT_SYMBOL(xfrm_input_resume);
501
502 static 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
515 int 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 }
531 EXPORT_SYMBOL(xfrm_trans_queue);
532
533 void __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 }