1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2017 - 2018 Covalent IO, Inc. http://covalent.io */
3
4 #include <linux/bpf.h>
5 #include <linux/btf_ids.h>
6 #include <linux/filter.h>
7 #include <linux/errno.h>
8 #include <linux/file.h>
9 #include <linux/net.h>
10 #include <linux/workqueue.h>
11 #include <linux/skmsg.h>
12 #include <linux/list.h>
13 #include <linux/jhash.h>
14 #include <linux/sock_diag.h>
15 #include <net/udp.h>
16
17 struct bpf_stab {
18 struct bpf_map map;
19 struct sock **sks;
20 struct sk_psock_progs progs;
21 raw_spinlock_t lock;
22 };
23
24 #define SOCK_CREATE_FLAG_MASK \
25 (BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY)
26
27 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
28 struct bpf_prog *old, u32 which);
29 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map);
30
sock_map_alloc(union bpf_attr * attr)31 static struct bpf_map *sock_map_alloc(union bpf_attr *attr)
32 {
33 struct bpf_stab *stab;
34
35 if (!capable(CAP_NET_ADMIN))
36 return ERR_PTR(-EPERM);
37 if (attr->max_entries == 0 ||
38 attr->key_size != 4 ||
39 (attr->value_size != sizeof(u32) &&
40 attr->value_size != sizeof(u64)) ||
41 attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
42 return ERR_PTR(-EINVAL);
43
44 stab = bpf_map_area_alloc(sizeof(*stab), NUMA_NO_NODE);
45 if (!stab)
46 return ERR_PTR(-ENOMEM);
47
48 bpf_map_init_from_attr(&stab->map, attr);
49 raw_spin_lock_init(&stab->lock);
50
51 stab->sks = bpf_map_area_alloc((u64) stab->map.max_entries *
52 sizeof(struct sock *),
53 stab->map.numa_node);
54 if (!stab->sks) {
55 bpf_map_area_free(stab);
56 return ERR_PTR(-ENOMEM);
57 }
58
59 return &stab->map;
60 }
61
sock_map_get_from_fd(const union bpf_attr * attr,struct bpf_prog * prog)62 int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog)
63 {
64 u32 ufd = attr->target_fd;
65 struct bpf_map *map;
66 struct fd f;
67 int ret;
68
69 if (attr->attach_flags || attr->replace_bpf_fd)
70 return -EINVAL;
71
72 f = fdget(ufd);
73 map = __bpf_map_get(f);
74 if (IS_ERR(map))
75 return PTR_ERR(map);
76 ret = sock_map_prog_update(map, prog, NULL, attr->attach_type);
77 fdput(f);
78 return ret;
79 }
80
sock_map_prog_detach(const union bpf_attr * attr,enum bpf_prog_type ptype)81 int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
82 {
83 u32 ufd = attr->target_fd;
84 struct bpf_prog *prog;
85 struct bpf_map *map;
86 struct fd f;
87 int ret;
88
89 if (attr->attach_flags || attr->replace_bpf_fd)
90 return -EINVAL;
91
92 f = fdget(ufd);
93 map = __bpf_map_get(f);
94 if (IS_ERR(map))
95 return PTR_ERR(map);
96
97 prog = bpf_prog_get(attr->attach_bpf_fd);
98 if (IS_ERR(prog)) {
99 ret = PTR_ERR(prog);
100 goto put_map;
101 }
102
103 if (prog->type != ptype) {
104 ret = -EINVAL;
105 goto put_prog;
106 }
107
108 ret = sock_map_prog_update(map, NULL, prog, attr->attach_type);
109 put_prog:
110 bpf_prog_put(prog);
111 put_map:
112 fdput(f);
113 return ret;
114 }
115
sock_map_sk_acquire(struct sock * sk)116 static void sock_map_sk_acquire(struct sock *sk)
117 __acquires(&sk->sk_lock.slock)
118 {
119 lock_sock(sk);
120 rcu_read_lock();
121 }
122
sock_map_sk_release(struct sock * sk)123 static void sock_map_sk_release(struct sock *sk)
124 __releases(&sk->sk_lock.slock)
125 {
126 rcu_read_unlock();
127 release_sock(sk);
128 }
129
sock_map_add_link(struct sk_psock * psock,struct sk_psock_link * link,struct bpf_map * map,void * link_raw)130 static void sock_map_add_link(struct sk_psock *psock,
131 struct sk_psock_link *link,
132 struct bpf_map *map, void *link_raw)
133 {
134 link->link_raw = link_raw;
135 link->map = map;
136 spin_lock_bh(&psock->link_lock);
137 list_add_tail(&link->list, &psock->link);
138 spin_unlock_bh(&psock->link_lock);
139 }
140
sock_map_del_link(struct sock * sk,struct sk_psock * psock,void * link_raw)141 static void sock_map_del_link(struct sock *sk,
142 struct sk_psock *psock, void *link_raw)
143 {
144 bool strp_stop = false, verdict_stop = false;
145 struct sk_psock_link *link, *tmp;
146
147 spin_lock_bh(&psock->link_lock);
148 list_for_each_entry_safe(link, tmp, &psock->link, list) {
149 if (link->link_raw == link_raw) {
150 struct bpf_map *map = link->map;
151 struct sk_psock_progs *progs = sock_map_progs(map);
152
153 if (psock->saved_data_ready && progs->stream_parser)
154 strp_stop = true;
155 if (psock->saved_data_ready && progs->stream_verdict)
156 verdict_stop = true;
157 if (psock->saved_data_ready && progs->skb_verdict)
158 verdict_stop = true;
159 list_del(&link->list);
160 sk_psock_free_link(link);
161 }
162 }
163 spin_unlock_bh(&psock->link_lock);
164 if (strp_stop || verdict_stop) {
165 write_lock_bh(&sk->sk_callback_lock);
166 if (strp_stop)
167 sk_psock_stop_strp(sk, psock);
168 if (verdict_stop)
169 sk_psock_stop_verdict(sk, psock);
170
171 if (psock->psock_update_sk_prot)
172 psock->psock_update_sk_prot(sk, psock, false);
173 write_unlock_bh(&sk->sk_callback_lock);
174 }
175 }
176
sock_map_unref(struct sock * sk,void * link_raw)177 static void sock_map_unref(struct sock *sk, void *link_raw)
178 {
179 struct sk_psock *psock = sk_psock(sk);
180
181 if (likely(psock)) {
182 sock_map_del_link(sk, psock, link_raw);
183 sk_psock_put(sk, psock);
184 }
185 }
186
sock_map_init_proto(struct sock * sk,struct sk_psock * psock)187 static int sock_map_init_proto(struct sock *sk, struct sk_psock *psock)
188 {
189 if (!sk->sk_prot->psock_update_sk_prot)
190 return -EINVAL;
191 psock->psock_update_sk_prot = sk->sk_prot->psock_update_sk_prot;
192 return sk->sk_prot->psock_update_sk_prot(sk, psock, false);
193 }
194
sock_map_psock_get_checked(struct sock * sk)195 static struct sk_psock *sock_map_psock_get_checked(struct sock *sk)
196 {
197 struct sk_psock *psock;
198
199 rcu_read_lock();
200 psock = sk_psock(sk);
201 if (psock) {
202 if (sk->sk_prot->close != sock_map_close) {
203 psock = ERR_PTR(-EBUSY);
204 goto out;
205 }
206
207 if (!refcount_inc_not_zero(&psock->refcnt))
208 psock = ERR_PTR(-EBUSY);
209 }
210 out:
211 rcu_read_unlock();
212 return psock;
213 }
214
sock_map_link(struct bpf_map * map,struct sock * sk)215 static int sock_map_link(struct bpf_map *map, struct sock *sk)
216 {
217 struct sk_psock_progs *progs = sock_map_progs(map);
218 struct bpf_prog *stream_verdict = NULL;
219 struct bpf_prog *stream_parser = NULL;
220 struct bpf_prog *skb_verdict = NULL;
221 struct bpf_prog *msg_parser = NULL;
222 struct sk_psock *psock;
223 int ret;
224
225 stream_verdict = READ_ONCE(progs->stream_verdict);
226 if (stream_verdict) {
227 stream_verdict = bpf_prog_inc_not_zero(stream_verdict);
228 if (IS_ERR(stream_verdict))
229 return PTR_ERR(stream_verdict);
230 }
231
232 stream_parser = READ_ONCE(progs->stream_parser);
233 if (stream_parser) {
234 stream_parser = bpf_prog_inc_not_zero(stream_parser);
235 if (IS_ERR(stream_parser)) {
236 ret = PTR_ERR(stream_parser);
237 goto out_put_stream_verdict;
238 }
239 }
240
241 msg_parser = READ_ONCE(progs->msg_parser);
242 if (msg_parser) {
243 msg_parser = bpf_prog_inc_not_zero(msg_parser);
244 if (IS_ERR(msg_parser)) {
245 ret = PTR_ERR(msg_parser);
246 goto out_put_stream_parser;
247 }
248 }
249
250 skb_verdict = READ_ONCE(progs->skb_verdict);
251 if (skb_verdict) {
252 skb_verdict = bpf_prog_inc_not_zero(skb_verdict);
253 if (IS_ERR(skb_verdict)) {
254 ret = PTR_ERR(skb_verdict);
255 goto out_put_msg_parser;
256 }
257 }
258
259 psock = sock_map_psock_get_checked(sk);
260 if (IS_ERR(psock)) {
261 ret = PTR_ERR(psock);
262 goto out_progs;
263 }
264
265 if (psock) {
266 if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
267 (stream_parser && READ_ONCE(psock->progs.stream_parser)) ||
268 (skb_verdict && READ_ONCE(psock->progs.skb_verdict)) ||
269 (skb_verdict && READ_ONCE(psock->progs.stream_verdict)) ||
270 (stream_verdict && READ_ONCE(psock->progs.skb_verdict)) ||
271 (stream_verdict && READ_ONCE(psock->progs.stream_verdict))) {
272 sk_psock_put(sk, psock);
273 ret = -EBUSY;
274 goto out_progs;
275 }
276 } else {
277 psock = sk_psock_init(sk, map->numa_node);
278 if (IS_ERR(psock)) {
279 ret = PTR_ERR(psock);
280 goto out_progs;
281 }
282 }
283
284 if (msg_parser)
285 psock_set_prog(&psock->progs.msg_parser, msg_parser);
286 if (stream_parser)
287 psock_set_prog(&psock->progs.stream_parser, stream_parser);
288 if (stream_verdict)
289 psock_set_prog(&psock->progs.stream_verdict, stream_verdict);
290 if (skb_verdict)
291 psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
292
293 /* msg_* and stream_* programs references tracked in psock after this
294 * point. Reference dec and cleanup will occur through psock destructor
295 */
296 ret = sock_map_init_proto(sk, psock);
297 if (ret < 0) {
298 sk_psock_put(sk, psock);
299 goto out;
300 }
301
302 write_lock_bh(&sk->sk_callback_lock);
303 if (stream_parser && stream_verdict && !psock->saved_data_ready) {
304 ret = sk_psock_init_strp(sk, psock);
305 if (ret) {
306 write_unlock_bh(&sk->sk_callback_lock);
307 sk_psock_put(sk, psock);
308 goto out;
309 }
310 sk_psock_start_strp(sk, psock);
311 } else if (!stream_parser && stream_verdict && !psock->saved_data_ready) {
312 sk_psock_start_verdict(sk,psock);
313 } else if (!stream_verdict && skb_verdict && !psock->saved_data_ready) {
314 sk_psock_start_verdict(sk, psock);
315 }
316 write_unlock_bh(&sk->sk_callback_lock);
317 return 0;
318 out_progs:
319 if (skb_verdict)
320 bpf_prog_put(skb_verdict);
321 out_put_msg_parser:
322 if (msg_parser)
323 bpf_prog_put(msg_parser);
324 out_put_stream_parser:
325 if (stream_parser)
326 bpf_prog_put(stream_parser);
327 out_put_stream_verdict:
328 if (stream_verdict)
329 bpf_prog_put(stream_verdict);
330 out:
331 return ret;
332 }
333
sock_map_free(struct bpf_map * map)334 static void sock_map_free(struct bpf_map *map)
335 {
336 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
337 int i;
338
339 /* After the sync no updates or deletes will be in-flight so it
340 * is safe to walk map and remove entries without risking a race
341 * in EEXIST update case.
342 */
343 synchronize_rcu();
344 for (i = 0; i < stab->map.max_entries; i++) {
345 struct sock **psk = &stab->sks[i];
346 struct sock *sk;
347
348 sk = xchg(psk, NULL);
349 if (sk) {
350 sock_hold(sk);
351 lock_sock(sk);
352 rcu_read_lock();
353 sock_map_unref(sk, psk);
354 rcu_read_unlock();
355 release_sock(sk);
356 sock_put(sk);
357 }
358 }
359
360 /* wait for psock readers accessing its map link */
361 synchronize_rcu();
362
363 bpf_map_area_free(stab->sks);
364 bpf_map_area_free(stab);
365 }
366
sock_map_release_progs(struct bpf_map * map)367 static void sock_map_release_progs(struct bpf_map *map)
368 {
369 psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
370 }
371
__sock_map_lookup_elem(struct bpf_map * map,u32 key)372 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
373 {
374 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
375
376 WARN_ON_ONCE(!rcu_read_lock_held());
377
378 if (unlikely(key >= map->max_entries))
379 return NULL;
380 return READ_ONCE(stab->sks[key]);
381 }
382
sock_map_lookup(struct bpf_map * map,void * key)383 static void *sock_map_lookup(struct bpf_map *map, void *key)
384 {
385 struct sock *sk;
386
387 sk = __sock_map_lookup_elem(map, *(u32 *)key);
388 if (!sk)
389 return NULL;
390 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
391 return NULL;
392 return sk;
393 }
394
sock_map_lookup_sys(struct bpf_map * map,void * key)395 static void *sock_map_lookup_sys(struct bpf_map *map, void *key)
396 {
397 struct sock *sk;
398
399 if (map->value_size != sizeof(u64))
400 return ERR_PTR(-ENOSPC);
401
402 sk = __sock_map_lookup_elem(map, *(u32 *)key);
403 if (!sk)
404 return ERR_PTR(-ENOENT);
405
406 __sock_gen_cookie(sk);
407 return &sk->sk_cookie;
408 }
409
__sock_map_delete(struct bpf_stab * stab,struct sock * sk_test,struct sock ** psk)410 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
411 struct sock **psk)
412 {
413 struct sock *sk;
414 int err = 0;
415 unsigned long flags;
416
417 raw_spin_lock_irqsave(&stab->lock, flags);
418 sk = *psk;
419 if (!sk_test || sk_test == sk)
420 sk = xchg(psk, NULL);
421
422 if (likely(sk))
423 sock_map_unref(sk, psk);
424 else
425 err = -EINVAL;
426
427 raw_spin_unlock_irqrestore(&stab->lock, flags);
428 return err;
429 }
430
sock_map_delete_from_link(struct bpf_map * map,struct sock * sk,void * link_raw)431 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
432 void *link_raw)
433 {
434 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
435
436 __sock_map_delete(stab, sk, link_raw);
437 }
438
sock_map_delete_elem(struct bpf_map * map,void * key)439 static int sock_map_delete_elem(struct bpf_map *map, void *key)
440 {
441 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
442 u32 i = *(u32 *)key;
443 struct sock **psk;
444
445 if (unlikely(i >= map->max_entries))
446 return -EINVAL;
447
448 psk = &stab->sks[i];
449 return __sock_map_delete(stab, NULL, psk);
450 }
451
sock_map_get_next_key(struct bpf_map * map,void * key,void * next)452 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
453 {
454 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
455 u32 i = key ? *(u32 *)key : U32_MAX;
456 u32 *key_next = next;
457
458 if (i == stab->map.max_entries - 1)
459 return -ENOENT;
460 if (i >= stab->map.max_entries)
461 *key_next = 0;
462 else
463 *key_next = i + 1;
464 return 0;
465 }
466
sock_map_update_common(struct bpf_map * map,u32 idx,struct sock * sk,u64 flags)467 static int sock_map_update_common(struct bpf_map *map, u32 idx,
468 struct sock *sk, u64 flags)
469 {
470 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
471 struct sk_psock_link *link;
472 struct sk_psock *psock;
473 struct sock *osk;
474 int ret;
475
476 WARN_ON_ONCE(!rcu_read_lock_held());
477 if (unlikely(flags > BPF_EXIST))
478 return -EINVAL;
479 if (unlikely(idx >= map->max_entries))
480 return -E2BIG;
481
482 link = sk_psock_init_link();
483 if (!link)
484 return -ENOMEM;
485
486 ret = sock_map_link(map, sk);
487 if (ret < 0)
488 goto out_free;
489
490 psock = sk_psock(sk);
491 WARN_ON_ONCE(!psock);
492
493 raw_spin_lock_bh(&stab->lock);
494 osk = stab->sks[idx];
495 if (osk && flags == BPF_NOEXIST) {
496 ret = -EEXIST;
497 goto out_unlock;
498 } else if (!osk && flags == BPF_EXIST) {
499 ret = -ENOENT;
500 goto out_unlock;
501 }
502
503 sock_map_add_link(psock, link, map, &stab->sks[idx]);
504 stab->sks[idx] = sk;
505 if (osk)
506 sock_map_unref(osk, &stab->sks[idx]);
507 raw_spin_unlock_bh(&stab->lock);
508 return 0;
509 out_unlock:
510 raw_spin_unlock_bh(&stab->lock);
511 if (psock)
512 sk_psock_put(sk, psock);
513 out_free:
514 sk_psock_free_link(link);
515 return ret;
516 }
517
sock_map_op_okay(const struct bpf_sock_ops_kern * ops)518 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
519 {
520 return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
521 ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
522 ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
523 }
524
sock_map_redirect_allowed(const struct sock * sk)525 static bool sock_map_redirect_allowed(const struct sock *sk)
526 {
527 if (sk_is_tcp(sk))
528 return sk->sk_state != TCP_LISTEN;
529 else
530 return sk->sk_state == TCP_ESTABLISHED;
531 }
532
sock_map_sk_is_suitable(const struct sock * sk)533 static bool sock_map_sk_is_suitable(const struct sock *sk)
534 {
535 return !!sk->sk_prot->psock_update_sk_prot;
536 }
537
sock_map_sk_state_allowed(const struct sock * sk)538 static bool sock_map_sk_state_allowed(const struct sock *sk)
539 {
540 if (sk_is_tcp(sk))
541 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
542 if (sk_is_stream_unix(sk))
543 return (1 << sk->sk_state) & TCPF_ESTABLISHED;
544 return true;
545 }
546
547 static int sock_hash_update_common(struct bpf_map *map, void *key,
548 struct sock *sk, u64 flags);
549
sock_map_update_elem_sys(struct bpf_map * map,void * key,void * value,u64 flags)550 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
551 u64 flags)
552 {
553 struct socket *sock;
554 struct sock *sk;
555 int ret;
556 u64 ufd;
557
558 if (map->value_size == sizeof(u64))
559 ufd = *(u64 *)value;
560 else
561 ufd = *(u32 *)value;
562 if (ufd > S32_MAX)
563 return -EINVAL;
564
565 sock = sockfd_lookup(ufd, &ret);
566 if (!sock)
567 return ret;
568 sk = sock->sk;
569 if (!sk) {
570 ret = -EINVAL;
571 goto out;
572 }
573 if (!sock_map_sk_is_suitable(sk)) {
574 ret = -EOPNOTSUPP;
575 goto out;
576 }
577
578 sock_map_sk_acquire(sk);
579 if (!sock_map_sk_state_allowed(sk))
580 ret = -EOPNOTSUPP;
581 else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
582 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
583 else
584 ret = sock_hash_update_common(map, key, sk, flags);
585 sock_map_sk_release(sk);
586 out:
587 sockfd_put(sock);
588 return ret;
589 }
590
sock_map_update_elem(struct bpf_map * map,void * key,void * value,u64 flags)591 static int sock_map_update_elem(struct bpf_map *map, void *key,
592 void *value, u64 flags)
593 {
594 struct sock *sk = (struct sock *)value;
595 int ret;
596
597 if (unlikely(!sk || !sk_fullsock(sk)))
598 return -EINVAL;
599
600 if (!sock_map_sk_is_suitable(sk))
601 return -EOPNOTSUPP;
602
603 local_bh_disable();
604 bh_lock_sock(sk);
605 if (!sock_map_sk_state_allowed(sk))
606 ret = -EOPNOTSUPP;
607 else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
608 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
609 else
610 ret = sock_hash_update_common(map, key, sk, flags);
611 bh_unlock_sock(sk);
612 local_bh_enable();
613 return ret;
614 }
615
BPF_CALL_4(bpf_sock_map_update,struct bpf_sock_ops_kern *,sops,struct bpf_map *,map,void *,key,u64,flags)616 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
617 struct bpf_map *, map, void *, key, u64, flags)
618 {
619 WARN_ON_ONCE(!rcu_read_lock_held());
620
621 if (likely(sock_map_sk_is_suitable(sops->sk) &&
622 sock_map_op_okay(sops)))
623 return sock_map_update_common(map, *(u32 *)key, sops->sk,
624 flags);
625 return -EOPNOTSUPP;
626 }
627
628 const struct bpf_func_proto bpf_sock_map_update_proto = {
629 .func = bpf_sock_map_update,
630 .gpl_only = false,
631 .pkt_access = true,
632 .ret_type = RET_INTEGER,
633 .arg1_type = ARG_PTR_TO_CTX,
634 .arg2_type = ARG_CONST_MAP_PTR,
635 .arg3_type = ARG_PTR_TO_MAP_KEY,
636 .arg4_type = ARG_ANYTHING,
637 };
638
BPF_CALL_4(bpf_sk_redirect_map,struct sk_buff *,skb,struct bpf_map *,map,u32,key,u64,flags)639 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
640 struct bpf_map *, map, u32, key, u64, flags)
641 {
642 struct sock *sk;
643
644 if (unlikely(flags & ~(BPF_F_INGRESS)))
645 return SK_DROP;
646
647 sk = __sock_map_lookup_elem(map, key);
648 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
649 return SK_DROP;
650
651 skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
652 return SK_PASS;
653 }
654
655 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
656 .func = bpf_sk_redirect_map,
657 .gpl_only = false,
658 .ret_type = RET_INTEGER,
659 .arg1_type = ARG_PTR_TO_CTX,
660 .arg2_type = ARG_CONST_MAP_PTR,
661 .arg3_type = ARG_ANYTHING,
662 .arg4_type = ARG_ANYTHING,
663 };
664
BPF_CALL_4(bpf_msg_redirect_map,struct sk_msg *,msg,struct bpf_map *,map,u32,key,u64,flags)665 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
666 struct bpf_map *, map, u32, key, u64, flags)
667 {
668 struct sock *sk;
669
670 if (unlikely(flags & ~(BPF_F_INGRESS)))
671 return SK_DROP;
672
673 sk = __sock_map_lookup_elem(map, key);
674 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
675 return SK_DROP;
676 if (!(flags & BPF_F_INGRESS) && !sk_is_tcp(sk))
677 return SK_DROP;
678
679 msg->flags = flags;
680 msg->sk_redir = sk;
681 return SK_PASS;
682 }
683
684 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
685 .func = bpf_msg_redirect_map,
686 .gpl_only = false,
687 .ret_type = RET_INTEGER,
688 .arg1_type = ARG_PTR_TO_CTX,
689 .arg2_type = ARG_CONST_MAP_PTR,
690 .arg3_type = ARG_ANYTHING,
691 .arg4_type = ARG_ANYTHING,
692 };
693
694 struct sock_map_seq_info {
695 struct bpf_map *map;
696 struct sock *sk;
697 u32 index;
698 };
699
700 struct bpf_iter__sockmap {
701 __bpf_md_ptr(struct bpf_iter_meta *, meta);
702 __bpf_md_ptr(struct bpf_map *, map);
703 __bpf_md_ptr(void *, key);
704 __bpf_md_ptr(struct sock *, sk);
705 };
706
DEFINE_BPF_ITER_FUNC(sockmap,struct bpf_iter_meta * meta,struct bpf_map * map,void * key,struct sock * sk)707 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
708 struct bpf_map *map, void *key,
709 struct sock *sk)
710
711 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
712 {
713 if (unlikely(info->index >= info->map->max_entries))
714 return NULL;
715
716 info->sk = __sock_map_lookup_elem(info->map, info->index);
717
718 /* can't return sk directly, since that might be NULL */
719 return info;
720 }
721
sock_map_seq_start(struct seq_file * seq,loff_t * pos)722 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
723 __acquires(rcu)
724 {
725 struct sock_map_seq_info *info = seq->private;
726
727 if (*pos == 0)
728 ++*pos;
729
730 /* pairs with sock_map_seq_stop */
731 rcu_read_lock();
732 return sock_map_seq_lookup_elem(info);
733 }
734
sock_map_seq_next(struct seq_file * seq,void * v,loff_t * pos)735 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
736 __must_hold(rcu)
737 {
738 struct sock_map_seq_info *info = seq->private;
739
740 ++*pos;
741 ++info->index;
742
743 return sock_map_seq_lookup_elem(info);
744 }
745
sock_map_seq_show(struct seq_file * seq,void * v)746 static int sock_map_seq_show(struct seq_file *seq, void *v)
747 __must_hold(rcu)
748 {
749 struct sock_map_seq_info *info = seq->private;
750 struct bpf_iter__sockmap ctx = {};
751 struct bpf_iter_meta meta;
752 struct bpf_prog *prog;
753
754 meta.seq = seq;
755 prog = bpf_iter_get_info(&meta, !v);
756 if (!prog)
757 return 0;
758
759 ctx.meta = &meta;
760 ctx.map = info->map;
761 if (v) {
762 ctx.key = &info->index;
763 ctx.sk = info->sk;
764 }
765
766 return bpf_iter_run_prog(prog, &ctx);
767 }
768
sock_map_seq_stop(struct seq_file * seq,void * v)769 static void sock_map_seq_stop(struct seq_file *seq, void *v)
770 __releases(rcu)
771 {
772 if (!v)
773 (void)sock_map_seq_show(seq, NULL);
774
775 /* pairs with sock_map_seq_start */
776 rcu_read_unlock();
777 }
778
779 static const struct seq_operations sock_map_seq_ops = {
780 .start = sock_map_seq_start,
781 .next = sock_map_seq_next,
782 .stop = sock_map_seq_stop,
783 .show = sock_map_seq_show,
784 };
785
sock_map_init_seq_private(void * priv_data,struct bpf_iter_aux_info * aux)786 static int sock_map_init_seq_private(void *priv_data,
787 struct bpf_iter_aux_info *aux)
788 {
789 struct sock_map_seq_info *info = priv_data;
790
791 bpf_map_inc_with_uref(aux->map);
792 info->map = aux->map;
793 return 0;
794 }
795
sock_map_fini_seq_private(void * priv_data)796 static void sock_map_fini_seq_private(void *priv_data)
797 {
798 struct sock_map_seq_info *info = priv_data;
799
800 bpf_map_put_with_uref(info->map);
801 }
802
803 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
804 .seq_ops = &sock_map_seq_ops,
805 .init_seq_private = sock_map_init_seq_private,
806 .fini_seq_private = sock_map_fini_seq_private,
807 .seq_priv_size = sizeof(struct sock_map_seq_info),
808 };
809
810 BTF_ID_LIST_SINGLE(sock_map_btf_ids, struct, bpf_stab)
811 const struct bpf_map_ops sock_map_ops = {
812 .map_meta_equal = bpf_map_meta_equal,
813 .map_alloc = sock_map_alloc,
814 .map_free = sock_map_free,
815 .map_get_next_key = sock_map_get_next_key,
816 .map_lookup_elem_sys_only = sock_map_lookup_sys,
817 .map_update_elem = sock_map_update_elem,
818 .map_delete_elem = sock_map_delete_elem,
819 .map_lookup_elem = sock_map_lookup,
820 .map_release_uref = sock_map_release_progs,
821 .map_check_btf = map_check_no_btf,
822 .map_btf_id = &sock_map_btf_ids[0],
823 .iter_seq_info = &sock_map_iter_seq_info,
824 };
825
826 struct bpf_shtab_elem {
827 struct rcu_head rcu;
828 u32 hash;
829 struct sock *sk;
830 struct hlist_node node;
831 u8 key[];
832 };
833
834 struct bpf_shtab_bucket {
835 struct hlist_head head;
836 raw_spinlock_t lock;
837 };
838
839 struct bpf_shtab {
840 struct bpf_map map;
841 struct bpf_shtab_bucket *buckets;
842 u32 buckets_num;
843 u32 elem_size;
844 struct sk_psock_progs progs;
845 atomic_t count;
846 };
847
sock_hash_bucket_hash(const void * key,u32 len)848 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
849 {
850 return jhash(key, len, 0);
851 }
852
sock_hash_select_bucket(struct bpf_shtab * htab,u32 hash)853 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
854 u32 hash)
855 {
856 return &htab->buckets[hash & (htab->buckets_num - 1)];
857 }
858
859 static struct bpf_shtab_elem *
sock_hash_lookup_elem_raw(struct hlist_head * head,u32 hash,void * key,u32 key_size)860 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
861 u32 key_size)
862 {
863 struct bpf_shtab_elem *elem;
864
865 hlist_for_each_entry_rcu(elem, head, node) {
866 if (elem->hash == hash &&
867 !memcmp(&elem->key, key, key_size))
868 return elem;
869 }
870
871 return NULL;
872 }
873
__sock_hash_lookup_elem(struct bpf_map * map,void * key)874 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
875 {
876 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
877 u32 key_size = map->key_size, hash;
878 struct bpf_shtab_bucket *bucket;
879 struct bpf_shtab_elem *elem;
880
881 WARN_ON_ONCE(!rcu_read_lock_held());
882
883 hash = sock_hash_bucket_hash(key, key_size);
884 bucket = sock_hash_select_bucket(htab, hash);
885 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
886
887 return elem ? elem->sk : NULL;
888 }
889
sock_hash_free_elem(struct bpf_shtab * htab,struct bpf_shtab_elem * elem)890 static void sock_hash_free_elem(struct bpf_shtab *htab,
891 struct bpf_shtab_elem *elem)
892 {
893 atomic_dec(&htab->count);
894 kfree_rcu(elem, rcu);
895 }
896
sock_hash_delete_from_link(struct bpf_map * map,struct sock * sk,void * link_raw)897 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
898 void *link_raw)
899 {
900 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
901 struct bpf_shtab_elem *elem_probe, *elem = link_raw;
902 struct bpf_shtab_bucket *bucket;
903
904 WARN_ON_ONCE(!rcu_read_lock_held());
905 bucket = sock_hash_select_bucket(htab, elem->hash);
906
907 /* elem may be deleted in parallel from the map, but access here
908 * is okay since it's going away only after RCU grace period.
909 * However, we need to check whether it's still present.
910 */
911 raw_spin_lock_bh(&bucket->lock);
912 elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
913 elem->key, map->key_size);
914 if (elem_probe && elem_probe == elem) {
915 hlist_del_rcu(&elem->node);
916 sock_map_unref(elem->sk, elem);
917 sock_hash_free_elem(htab, elem);
918 }
919 raw_spin_unlock_bh(&bucket->lock);
920 }
921
sock_hash_delete_elem(struct bpf_map * map,void * key)922 static int sock_hash_delete_elem(struct bpf_map *map, void *key)
923 {
924 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
925 u32 hash, key_size = map->key_size;
926 struct bpf_shtab_bucket *bucket;
927 struct bpf_shtab_elem *elem;
928 int ret = -ENOENT;
929 unsigned long flags;
930
931 hash = sock_hash_bucket_hash(key, key_size);
932 bucket = sock_hash_select_bucket(htab, hash);
933
934 raw_spin_lock_irqsave(&bucket->lock, flags);
935 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
936 if (elem) {
937 hlist_del_rcu(&elem->node);
938 sock_map_unref(elem->sk, elem);
939 sock_hash_free_elem(htab, elem);
940 ret = 0;
941 }
942 raw_spin_unlock_irqrestore(&bucket->lock, flags);
943 return ret;
944 }
945
sock_hash_alloc_elem(struct bpf_shtab * htab,void * key,u32 key_size,u32 hash,struct sock * sk,struct bpf_shtab_elem * old)946 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
947 void *key, u32 key_size,
948 u32 hash, struct sock *sk,
949 struct bpf_shtab_elem *old)
950 {
951 struct bpf_shtab_elem *new;
952
953 if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
954 if (!old) {
955 atomic_dec(&htab->count);
956 return ERR_PTR(-E2BIG);
957 }
958 }
959
960 new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
961 GFP_ATOMIC | __GFP_NOWARN,
962 htab->map.numa_node);
963 if (!new) {
964 atomic_dec(&htab->count);
965 return ERR_PTR(-ENOMEM);
966 }
967 memcpy(new->key, key, key_size);
968 new->sk = sk;
969 new->hash = hash;
970 return new;
971 }
972
sock_hash_update_common(struct bpf_map * map,void * key,struct sock * sk,u64 flags)973 static int sock_hash_update_common(struct bpf_map *map, void *key,
974 struct sock *sk, u64 flags)
975 {
976 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
977 u32 key_size = map->key_size, hash;
978 struct bpf_shtab_elem *elem, *elem_new;
979 struct bpf_shtab_bucket *bucket;
980 struct sk_psock_link *link;
981 struct sk_psock *psock;
982 int ret;
983
984 WARN_ON_ONCE(!rcu_read_lock_held());
985 if (unlikely(flags > BPF_EXIST))
986 return -EINVAL;
987
988 link = sk_psock_init_link();
989 if (!link)
990 return -ENOMEM;
991
992 ret = sock_map_link(map, sk);
993 if (ret < 0)
994 goto out_free;
995
996 psock = sk_psock(sk);
997 WARN_ON_ONCE(!psock);
998
999 hash = sock_hash_bucket_hash(key, key_size);
1000 bucket = sock_hash_select_bucket(htab, hash);
1001
1002 raw_spin_lock_bh(&bucket->lock);
1003 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
1004 if (elem && flags == BPF_NOEXIST) {
1005 ret = -EEXIST;
1006 goto out_unlock;
1007 } else if (!elem && flags == BPF_EXIST) {
1008 ret = -ENOENT;
1009 goto out_unlock;
1010 }
1011
1012 elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1013 if (IS_ERR(elem_new)) {
1014 ret = PTR_ERR(elem_new);
1015 goto out_unlock;
1016 }
1017
1018 sock_map_add_link(psock, link, map, elem_new);
1019 /* Add new element to the head of the list, so that
1020 * concurrent search will find it before old elem.
1021 */
1022 hlist_add_head_rcu(&elem_new->node, &bucket->head);
1023 if (elem) {
1024 hlist_del_rcu(&elem->node);
1025 sock_map_unref(elem->sk, elem);
1026 sock_hash_free_elem(htab, elem);
1027 }
1028 raw_spin_unlock_bh(&bucket->lock);
1029 return 0;
1030 out_unlock:
1031 raw_spin_unlock_bh(&bucket->lock);
1032 sk_psock_put(sk, psock);
1033 out_free:
1034 sk_psock_free_link(link);
1035 return ret;
1036 }
1037
sock_hash_get_next_key(struct bpf_map * map,void * key,void * key_next)1038 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1039 void *key_next)
1040 {
1041 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1042 struct bpf_shtab_elem *elem, *elem_next;
1043 u32 hash, key_size = map->key_size;
1044 struct hlist_head *head;
1045 int i = 0;
1046
1047 if (!key)
1048 goto find_first_elem;
1049 hash = sock_hash_bucket_hash(key, key_size);
1050 head = &sock_hash_select_bucket(htab, hash)->head;
1051 elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1052 if (!elem)
1053 goto find_first_elem;
1054
1055 elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1056 struct bpf_shtab_elem, node);
1057 if (elem_next) {
1058 memcpy(key_next, elem_next->key, key_size);
1059 return 0;
1060 }
1061
1062 i = hash & (htab->buckets_num - 1);
1063 i++;
1064 find_first_elem:
1065 for (; i < htab->buckets_num; i++) {
1066 head = &sock_hash_select_bucket(htab, i)->head;
1067 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1068 struct bpf_shtab_elem, node);
1069 if (elem_next) {
1070 memcpy(key_next, elem_next->key, key_size);
1071 return 0;
1072 }
1073 }
1074
1075 return -ENOENT;
1076 }
1077
sock_hash_alloc(union bpf_attr * attr)1078 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1079 {
1080 struct bpf_shtab *htab;
1081 int i, err;
1082
1083 if (!capable(CAP_NET_ADMIN))
1084 return ERR_PTR(-EPERM);
1085 if (attr->max_entries == 0 ||
1086 attr->key_size == 0 ||
1087 (attr->value_size != sizeof(u32) &&
1088 attr->value_size != sizeof(u64)) ||
1089 attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1090 return ERR_PTR(-EINVAL);
1091 if (attr->key_size > MAX_BPF_STACK)
1092 return ERR_PTR(-E2BIG);
1093
1094 htab = bpf_map_area_alloc(sizeof(*htab), NUMA_NO_NODE);
1095 if (!htab)
1096 return ERR_PTR(-ENOMEM);
1097
1098 bpf_map_init_from_attr(&htab->map, attr);
1099
1100 htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1101 htab->elem_size = sizeof(struct bpf_shtab_elem) +
1102 round_up(htab->map.key_size, 8);
1103 if (htab->buckets_num == 0 ||
1104 htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1105 err = -EINVAL;
1106 goto free_htab;
1107 }
1108
1109 htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1110 sizeof(struct bpf_shtab_bucket),
1111 htab->map.numa_node);
1112 if (!htab->buckets) {
1113 err = -ENOMEM;
1114 goto free_htab;
1115 }
1116
1117 for (i = 0; i < htab->buckets_num; i++) {
1118 INIT_HLIST_HEAD(&htab->buckets[i].head);
1119 raw_spin_lock_init(&htab->buckets[i].lock);
1120 }
1121
1122 return &htab->map;
1123 free_htab:
1124 bpf_map_area_free(htab);
1125 return ERR_PTR(err);
1126 }
1127
sock_hash_free(struct bpf_map * map)1128 static void sock_hash_free(struct bpf_map *map)
1129 {
1130 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1131 struct bpf_shtab_bucket *bucket;
1132 struct hlist_head unlink_list;
1133 struct bpf_shtab_elem *elem;
1134 struct hlist_node *node;
1135 int i;
1136
1137 /* After the sync no updates or deletes will be in-flight so it
1138 * is safe to walk map and remove entries without risking a race
1139 * in EEXIST update case.
1140 */
1141 synchronize_rcu();
1142 for (i = 0; i < htab->buckets_num; i++) {
1143 bucket = sock_hash_select_bucket(htab, i);
1144
1145 /* We are racing with sock_hash_delete_from_link to
1146 * enter the spin-lock critical section. Every socket on
1147 * the list is still linked to sockhash. Since link
1148 * exists, psock exists and holds a ref to socket. That
1149 * lets us to grab a socket ref too.
1150 */
1151 raw_spin_lock_bh(&bucket->lock);
1152 hlist_for_each_entry(elem, &bucket->head, node)
1153 sock_hold(elem->sk);
1154 hlist_move_list(&bucket->head, &unlink_list);
1155 raw_spin_unlock_bh(&bucket->lock);
1156
1157 /* Process removed entries out of atomic context to
1158 * block for socket lock before deleting the psock's
1159 * link to sockhash.
1160 */
1161 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1162 hlist_del(&elem->node);
1163 lock_sock(elem->sk);
1164 rcu_read_lock();
1165 sock_map_unref(elem->sk, elem);
1166 rcu_read_unlock();
1167 release_sock(elem->sk);
1168 sock_put(elem->sk);
1169 sock_hash_free_elem(htab, elem);
1170 }
1171 }
1172
1173 /* wait for psock readers accessing its map link */
1174 synchronize_rcu();
1175
1176 bpf_map_area_free(htab->buckets);
1177 bpf_map_area_free(htab);
1178 }
1179
sock_hash_lookup_sys(struct bpf_map * map,void * key)1180 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1181 {
1182 struct sock *sk;
1183
1184 if (map->value_size != sizeof(u64))
1185 return ERR_PTR(-ENOSPC);
1186
1187 sk = __sock_hash_lookup_elem(map, key);
1188 if (!sk)
1189 return ERR_PTR(-ENOENT);
1190
1191 __sock_gen_cookie(sk);
1192 return &sk->sk_cookie;
1193 }
1194
sock_hash_lookup(struct bpf_map * map,void * key)1195 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1196 {
1197 struct sock *sk;
1198
1199 sk = __sock_hash_lookup_elem(map, key);
1200 if (!sk)
1201 return NULL;
1202 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1203 return NULL;
1204 return sk;
1205 }
1206
sock_hash_release_progs(struct bpf_map * map)1207 static void sock_hash_release_progs(struct bpf_map *map)
1208 {
1209 psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1210 }
1211
BPF_CALL_4(bpf_sock_hash_update,struct bpf_sock_ops_kern *,sops,struct bpf_map *,map,void *,key,u64,flags)1212 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1213 struct bpf_map *, map, void *, key, u64, flags)
1214 {
1215 WARN_ON_ONCE(!rcu_read_lock_held());
1216
1217 if (likely(sock_map_sk_is_suitable(sops->sk) &&
1218 sock_map_op_okay(sops)))
1219 return sock_hash_update_common(map, key, sops->sk, flags);
1220 return -EOPNOTSUPP;
1221 }
1222
1223 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1224 .func = bpf_sock_hash_update,
1225 .gpl_only = false,
1226 .pkt_access = true,
1227 .ret_type = RET_INTEGER,
1228 .arg1_type = ARG_PTR_TO_CTX,
1229 .arg2_type = ARG_CONST_MAP_PTR,
1230 .arg3_type = ARG_PTR_TO_MAP_KEY,
1231 .arg4_type = ARG_ANYTHING,
1232 };
1233
BPF_CALL_4(bpf_sk_redirect_hash,struct sk_buff *,skb,struct bpf_map *,map,void *,key,u64,flags)1234 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1235 struct bpf_map *, map, void *, key, u64, flags)
1236 {
1237 struct sock *sk;
1238
1239 if (unlikely(flags & ~(BPF_F_INGRESS)))
1240 return SK_DROP;
1241
1242 sk = __sock_hash_lookup_elem(map, key);
1243 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1244 return SK_DROP;
1245
1246 skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1247 return SK_PASS;
1248 }
1249
1250 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1251 .func = bpf_sk_redirect_hash,
1252 .gpl_only = false,
1253 .ret_type = RET_INTEGER,
1254 .arg1_type = ARG_PTR_TO_CTX,
1255 .arg2_type = ARG_CONST_MAP_PTR,
1256 .arg3_type = ARG_PTR_TO_MAP_KEY,
1257 .arg4_type = ARG_ANYTHING,
1258 };
1259
BPF_CALL_4(bpf_msg_redirect_hash,struct sk_msg *,msg,struct bpf_map *,map,void *,key,u64,flags)1260 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1261 struct bpf_map *, map, void *, key, u64, flags)
1262 {
1263 struct sock *sk;
1264
1265 if (unlikely(flags & ~(BPF_F_INGRESS)))
1266 return SK_DROP;
1267
1268 sk = __sock_hash_lookup_elem(map, key);
1269 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1270 return SK_DROP;
1271 if (!(flags & BPF_F_INGRESS) && !sk_is_tcp(sk))
1272 return SK_DROP;
1273
1274 msg->flags = flags;
1275 msg->sk_redir = sk;
1276 return SK_PASS;
1277 }
1278
1279 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1280 .func = bpf_msg_redirect_hash,
1281 .gpl_only = false,
1282 .ret_type = RET_INTEGER,
1283 .arg1_type = ARG_PTR_TO_CTX,
1284 .arg2_type = ARG_CONST_MAP_PTR,
1285 .arg3_type = ARG_PTR_TO_MAP_KEY,
1286 .arg4_type = ARG_ANYTHING,
1287 };
1288
1289 struct sock_hash_seq_info {
1290 struct bpf_map *map;
1291 struct bpf_shtab *htab;
1292 u32 bucket_id;
1293 };
1294
sock_hash_seq_find_next(struct sock_hash_seq_info * info,struct bpf_shtab_elem * prev_elem)1295 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1296 struct bpf_shtab_elem *prev_elem)
1297 {
1298 const struct bpf_shtab *htab = info->htab;
1299 struct bpf_shtab_bucket *bucket;
1300 struct bpf_shtab_elem *elem;
1301 struct hlist_node *node;
1302
1303 /* try to find next elem in the same bucket */
1304 if (prev_elem) {
1305 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1306 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1307 if (elem)
1308 return elem;
1309
1310 /* no more elements, continue in the next bucket */
1311 info->bucket_id++;
1312 }
1313
1314 for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1315 bucket = &htab->buckets[info->bucket_id];
1316 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1317 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1318 if (elem)
1319 return elem;
1320 }
1321
1322 return NULL;
1323 }
1324
sock_hash_seq_start(struct seq_file * seq,loff_t * pos)1325 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1326 __acquires(rcu)
1327 {
1328 struct sock_hash_seq_info *info = seq->private;
1329
1330 if (*pos == 0)
1331 ++*pos;
1332
1333 /* pairs with sock_hash_seq_stop */
1334 rcu_read_lock();
1335 return sock_hash_seq_find_next(info, NULL);
1336 }
1337
sock_hash_seq_next(struct seq_file * seq,void * v,loff_t * pos)1338 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1339 __must_hold(rcu)
1340 {
1341 struct sock_hash_seq_info *info = seq->private;
1342
1343 ++*pos;
1344 return sock_hash_seq_find_next(info, v);
1345 }
1346
sock_hash_seq_show(struct seq_file * seq,void * v)1347 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1348 __must_hold(rcu)
1349 {
1350 struct sock_hash_seq_info *info = seq->private;
1351 struct bpf_iter__sockmap ctx = {};
1352 struct bpf_shtab_elem *elem = v;
1353 struct bpf_iter_meta meta;
1354 struct bpf_prog *prog;
1355
1356 meta.seq = seq;
1357 prog = bpf_iter_get_info(&meta, !elem);
1358 if (!prog)
1359 return 0;
1360
1361 ctx.meta = &meta;
1362 ctx.map = info->map;
1363 if (elem) {
1364 ctx.key = elem->key;
1365 ctx.sk = elem->sk;
1366 }
1367
1368 return bpf_iter_run_prog(prog, &ctx);
1369 }
1370
sock_hash_seq_stop(struct seq_file * seq,void * v)1371 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1372 __releases(rcu)
1373 {
1374 if (!v)
1375 (void)sock_hash_seq_show(seq, NULL);
1376
1377 /* pairs with sock_hash_seq_start */
1378 rcu_read_unlock();
1379 }
1380
1381 static const struct seq_operations sock_hash_seq_ops = {
1382 .start = sock_hash_seq_start,
1383 .next = sock_hash_seq_next,
1384 .stop = sock_hash_seq_stop,
1385 .show = sock_hash_seq_show,
1386 };
1387
sock_hash_init_seq_private(void * priv_data,struct bpf_iter_aux_info * aux)1388 static int sock_hash_init_seq_private(void *priv_data,
1389 struct bpf_iter_aux_info *aux)
1390 {
1391 struct sock_hash_seq_info *info = priv_data;
1392
1393 bpf_map_inc_with_uref(aux->map);
1394 info->map = aux->map;
1395 info->htab = container_of(aux->map, struct bpf_shtab, map);
1396 return 0;
1397 }
1398
sock_hash_fini_seq_private(void * priv_data)1399 static void sock_hash_fini_seq_private(void *priv_data)
1400 {
1401 struct sock_hash_seq_info *info = priv_data;
1402
1403 bpf_map_put_with_uref(info->map);
1404 }
1405
1406 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1407 .seq_ops = &sock_hash_seq_ops,
1408 .init_seq_private = sock_hash_init_seq_private,
1409 .fini_seq_private = sock_hash_fini_seq_private,
1410 .seq_priv_size = sizeof(struct sock_hash_seq_info),
1411 };
1412
1413 BTF_ID_LIST_SINGLE(sock_hash_map_btf_ids, struct, bpf_shtab)
1414 const struct bpf_map_ops sock_hash_ops = {
1415 .map_meta_equal = bpf_map_meta_equal,
1416 .map_alloc = sock_hash_alloc,
1417 .map_free = sock_hash_free,
1418 .map_get_next_key = sock_hash_get_next_key,
1419 .map_update_elem = sock_map_update_elem,
1420 .map_delete_elem = sock_hash_delete_elem,
1421 .map_lookup_elem = sock_hash_lookup,
1422 .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1423 .map_release_uref = sock_hash_release_progs,
1424 .map_check_btf = map_check_no_btf,
1425 .map_btf_id = &sock_hash_map_btf_ids[0],
1426 .iter_seq_info = &sock_hash_iter_seq_info,
1427 };
1428
sock_map_progs(struct bpf_map * map)1429 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1430 {
1431 switch (map->map_type) {
1432 case BPF_MAP_TYPE_SOCKMAP:
1433 return &container_of(map, struct bpf_stab, map)->progs;
1434 case BPF_MAP_TYPE_SOCKHASH:
1435 return &container_of(map, struct bpf_shtab, map)->progs;
1436 default:
1437 break;
1438 }
1439
1440 return NULL;
1441 }
1442
sock_map_prog_lookup(struct bpf_map * map,struct bpf_prog *** pprog,u32 which)1443 static int sock_map_prog_lookup(struct bpf_map *map, struct bpf_prog ***pprog,
1444 u32 which)
1445 {
1446 struct sk_psock_progs *progs = sock_map_progs(map);
1447
1448 if (!progs)
1449 return -EOPNOTSUPP;
1450
1451 switch (which) {
1452 case BPF_SK_MSG_VERDICT:
1453 *pprog = &progs->msg_parser;
1454 break;
1455 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1456 case BPF_SK_SKB_STREAM_PARSER:
1457 *pprog = &progs->stream_parser;
1458 break;
1459 #endif
1460 case BPF_SK_SKB_STREAM_VERDICT:
1461 if (progs->skb_verdict)
1462 return -EBUSY;
1463 *pprog = &progs->stream_verdict;
1464 break;
1465 case BPF_SK_SKB_VERDICT:
1466 if (progs->stream_verdict)
1467 return -EBUSY;
1468 *pprog = &progs->skb_verdict;
1469 break;
1470 default:
1471 return -EOPNOTSUPP;
1472 }
1473
1474 return 0;
1475 }
1476
sock_map_prog_update(struct bpf_map * map,struct bpf_prog * prog,struct bpf_prog * old,u32 which)1477 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1478 struct bpf_prog *old, u32 which)
1479 {
1480 struct bpf_prog **pprog;
1481 int ret;
1482
1483 ret = sock_map_prog_lookup(map, &pprog, which);
1484 if (ret)
1485 return ret;
1486
1487 if (old)
1488 return psock_replace_prog(pprog, prog, old);
1489
1490 psock_set_prog(pprog, prog);
1491 return 0;
1492 }
1493
sock_map_bpf_prog_query(const union bpf_attr * attr,union bpf_attr __user * uattr)1494 int sock_map_bpf_prog_query(const union bpf_attr *attr,
1495 union bpf_attr __user *uattr)
1496 {
1497 __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
1498 u32 prog_cnt = 0, flags = 0, ufd = attr->target_fd;
1499 struct bpf_prog **pprog;
1500 struct bpf_prog *prog;
1501 struct bpf_map *map;
1502 struct fd f;
1503 u32 id = 0;
1504 int ret;
1505
1506 if (attr->query.query_flags)
1507 return -EINVAL;
1508
1509 f = fdget(ufd);
1510 map = __bpf_map_get(f);
1511 if (IS_ERR(map))
1512 return PTR_ERR(map);
1513
1514 rcu_read_lock();
1515
1516 ret = sock_map_prog_lookup(map, &pprog, attr->query.attach_type);
1517 if (ret)
1518 goto end;
1519
1520 prog = *pprog;
1521 prog_cnt = !prog ? 0 : 1;
1522
1523 if (!attr->query.prog_cnt || !prog_ids || !prog_cnt)
1524 goto end;
1525
1526 /* we do not hold the refcnt, the bpf prog may be released
1527 * asynchronously and the id would be set to 0.
1528 */
1529 id = data_race(prog->aux->id);
1530 if (id == 0)
1531 prog_cnt = 0;
1532
1533 end:
1534 rcu_read_unlock();
1535
1536 if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)) ||
1537 (id != 0 && copy_to_user(prog_ids, &id, sizeof(u32))) ||
1538 copy_to_user(&uattr->query.prog_cnt, &prog_cnt, sizeof(prog_cnt)))
1539 ret = -EFAULT;
1540
1541 fdput(f);
1542 return ret;
1543 }
1544
sock_map_unlink(struct sock * sk,struct sk_psock_link * link)1545 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1546 {
1547 switch (link->map->map_type) {
1548 case BPF_MAP_TYPE_SOCKMAP:
1549 return sock_map_delete_from_link(link->map, sk,
1550 link->link_raw);
1551 case BPF_MAP_TYPE_SOCKHASH:
1552 return sock_hash_delete_from_link(link->map, sk,
1553 link->link_raw);
1554 default:
1555 break;
1556 }
1557 }
1558
sock_map_remove_links(struct sock * sk,struct sk_psock * psock)1559 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1560 {
1561 struct sk_psock_link *link;
1562
1563 while ((link = sk_psock_link_pop(psock))) {
1564 sock_map_unlink(sk, link);
1565 sk_psock_free_link(link);
1566 }
1567 }
1568
sock_map_unhash(struct sock * sk)1569 void sock_map_unhash(struct sock *sk)
1570 {
1571 void (*saved_unhash)(struct sock *sk);
1572 struct sk_psock *psock;
1573
1574 rcu_read_lock();
1575 psock = sk_psock(sk);
1576 if (unlikely(!psock)) {
1577 rcu_read_unlock();
1578 saved_unhash = READ_ONCE(sk->sk_prot)->unhash;
1579 } else {
1580 saved_unhash = psock->saved_unhash;
1581 sock_map_remove_links(sk, psock);
1582 rcu_read_unlock();
1583 }
1584 if (WARN_ON_ONCE(saved_unhash == sock_map_unhash))
1585 return;
1586 if (saved_unhash)
1587 saved_unhash(sk);
1588 }
1589 EXPORT_SYMBOL_GPL(sock_map_unhash);
1590
sock_map_destroy(struct sock * sk)1591 void sock_map_destroy(struct sock *sk)
1592 {
1593 void (*saved_destroy)(struct sock *sk);
1594 struct sk_psock *psock;
1595
1596 rcu_read_lock();
1597 psock = sk_psock_get(sk);
1598 if (unlikely(!psock)) {
1599 rcu_read_unlock();
1600 saved_destroy = READ_ONCE(sk->sk_prot)->destroy;
1601 } else {
1602 saved_destroy = psock->saved_destroy;
1603 sock_map_remove_links(sk, psock);
1604 rcu_read_unlock();
1605 sk_psock_stop(psock);
1606 sk_psock_put(sk, psock);
1607 }
1608 if (WARN_ON_ONCE(saved_destroy == sock_map_destroy))
1609 return;
1610 if (saved_destroy)
1611 saved_destroy(sk);
1612 }
1613 EXPORT_SYMBOL_GPL(sock_map_destroy);
1614
sock_map_close(struct sock * sk,long timeout)1615 void sock_map_close(struct sock *sk, long timeout)
1616 {
1617 void (*saved_close)(struct sock *sk, long timeout);
1618 struct sk_psock *psock;
1619
1620 lock_sock(sk);
1621 rcu_read_lock();
1622 psock = sk_psock_get(sk);
1623 if (unlikely(!psock)) {
1624 rcu_read_unlock();
1625 release_sock(sk);
1626 saved_close = READ_ONCE(sk->sk_prot)->close;
1627 } else {
1628 saved_close = psock->saved_close;
1629 sock_map_remove_links(sk, psock);
1630 rcu_read_unlock();
1631 sk_psock_stop(psock);
1632 release_sock(sk);
1633 cancel_delayed_work_sync(&psock->work);
1634 sk_psock_put(sk, psock);
1635 }
1636
1637 /* Make sure we do not recurse. This is a bug.
1638 * Leak the socket instead of crashing on a stack overflow.
1639 */
1640 if (WARN_ON_ONCE(saved_close == sock_map_close))
1641 return;
1642 saved_close(sk, timeout);
1643 }
1644 EXPORT_SYMBOL_GPL(sock_map_close);
1645
sock_map_iter_attach_target(struct bpf_prog * prog,union bpf_iter_link_info * linfo,struct bpf_iter_aux_info * aux)1646 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1647 union bpf_iter_link_info *linfo,
1648 struct bpf_iter_aux_info *aux)
1649 {
1650 struct bpf_map *map;
1651 int err = -EINVAL;
1652
1653 if (!linfo->map.map_fd)
1654 return -EBADF;
1655
1656 map = bpf_map_get_with_uref(linfo->map.map_fd);
1657 if (IS_ERR(map))
1658 return PTR_ERR(map);
1659
1660 if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1661 map->map_type != BPF_MAP_TYPE_SOCKHASH)
1662 goto put_map;
1663
1664 if (prog->aux->max_rdonly_access > map->key_size) {
1665 err = -EACCES;
1666 goto put_map;
1667 }
1668
1669 aux->map = map;
1670 return 0;
1671
1672 put_map:
1673 bpf_map_put_with_uref(map);
1674 return err;
1675 }
1676
sock_map_iter_detach_target(struct bpf_iter_aux_info * aux)1677 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1678 {
1679 bpf_map_put_with_uref(aux->map);
1680 }
1681
1682 static struct bpf_iter_reg sock_map_iter_reg = {
1683 .target = "sockmap",
1684 .attach_target = sock_map_iter_attach_target,
1685 .detach_target = sock_map_iter_detach_target,
1686 .show_fdinfo = bpf_iter_map_show_fdinfo,
1687 .fill_link_info = bpf_iter_map_fill_link_info,
1688 .ctx_arg_info_size = 2,
1689 .ctx_arg_info = {
1690 { offsetof(struct bpf_iter__sockmap, key),
1691 PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY },
1692 { offsetof(struct bpf_iter__sockmap, sk),
1693 PTR_TO_BTF_ID_OR_NULL },
1694 },
1695 };
1696
bpf_sockmap_iter_init(void)1697 static int __init bpf_sockmap_iter_init(void)
1698 {
1699 sock_map_iter_reg.ctx_arg_info[1].btf_id =
1700 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1701 return bpf_iter_reg_target(&sock_map_iter_reg);
1702 }
1703 late_initcall(bpf_sockmap_iter_init);
1704