Loading...
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
17struct bpf_stab {
18 struct bpf_map map;
19 struct sock **sks;
20 struct sk_psock_progs progs;
21 spinlock_t lock;
22};
23
24#define SOCK_CREATE_FLAG_MASK \
25 (BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY)
26
27/* This mutex is used to
28 * - protect race between prog/link attach/detach and link prog update, and
29 * - protect race between releasing and accessing map in bpf_link.
30 * A single global mutex lock is used since it is expected contention is low.
31 */
32static DEFINE_MUTEX(sockmap_mutex);
33
34static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
35 struct bpf_prog *old, struct bpf_link *link,
36 u32 which);
37static struct sk_psock_progs *sock_map_progs(struct bpf_map *map);
38
39static struct bpf_map *sock_map_alloc(union bpf_attr *attr)
40{
41 struct bpf_stab *stab;
42
43 if (attr->max_entries == 0 ||
44 attr->key_size != 4 ||
45 (attr->value_size != sizeof(u32) &&
46 attr->value_size != sizeof(u64)) ||
47 attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
48 return ERR_PTR(-EINVAL);
49
50 stab = bpf_map_area_alloc(sizeof(*stab), NUMA_NO_NODE);
51 if (!stab)
52 return ERR_PTR(-ENOMEM);
53
54 bpf_map_init_from_attr(&stab->map, attr);
55 spin_lock_init(&stab->lock);
56
57 stab->sks = bpf_map_area_alloc((u64) stab->map.max_entries *
58 sizeof(struct sock *),
59 stab->map.numa_node);
60 if (!stab->sks) {
61 bpf_map_area_free(stab);
62 return ERR_PTR(-ENOMEM);
63 }
64
65 return &stab->map;
66}
67
68int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog)
69{
70 struct bpf_map *map;
71 int ret;
72
73 if (attr->attach_flags || attr->replace_bpf_fd)
74 return -EINVAL;
75
76 CLASS(fd, f)(attr->target_fd);
77 map = __bpf_map_get(f);
78 if (IS_ERR(map))
79 return PTR_ERR(map);
80 mutex_lock(&sockmap_mutex);
81 ret = sock_map_prog_update(map, prog, NULL, NULL, attr->attach_type);
82 mutex_unlock(&sockmap_mutex);
83 return ret;
84}
85
86int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
87{
88 struct bpf_prog *prog;
89 struct bpf_map *map;
90 int ret;
91
92 if (attr->attach_flags || attr->replace_bpf_fd)
93 return -EINVAL;
94
95 CLASS(fd, f)(attr->target_fd);
96 map = __bpf_map_get(f);
97 if (IS_ERR(map))
98 return PTR_ERR(map);
99
100 prog = bpf_prog_get(attr->attach_bpf_fd);
101 if (IS_ERR(prog))
102 return PTR_ERR(prog);
103
104 if (prog->type != ptype) {
105 ret = -EINVAL;
106 goto put_prog;
107 }
108
109 mutex_lock(&sockmap_mutex);
110 ret = sock_map_prog_update(map, NULL, prog, NULL, attr->attach_type);
111 mutex_unlock(&sockmap_mutex);
112put_prog:
113 bpf_prog_put(prog);
114 return ret;
115}
116
117static void sock_map_sk_acquire(struct sock *sk)
118 __acquires(&sk->sk_lock.slock)
119{
120 lock_sock(sk);
121 rcu_read_lock();
122}
123
124static void sock_map_sk_release(struct sock *sk)
125 __releases(&sk->sk_lock.slock)
126{
127 rcu_read_unlock();
128 release_sock(sk);
129}
130
131static void sock_map_add_link(struct sk_psock *psock,
132 struct sk_psock_link *link,
133 struct bpf_map *map, void *link_raw)
134{
135 link->link_raw = link_raw;
136 link->map = map;
137 spin_lock_bh(&psock->link_lock);
138 list_add_tail(&link->list, &psock->link);
139 spin_unlock_bh(&psock->link_lock);
140}
141
142static void sock_map_del_link(struct sock *sk,
143 struct sk_psock *psock, void *link_raw)
144{
145 bool strp_stop = false, verdict_stop = false;
146 struct sk_psock_link *link, *tmp;
147
148 spin_lock_bh(&psock->link_lock);
149 list_for_each_entry_safe(link, tmp, &psock->link, list) {
150 if (link->link_raw == link_raw) {
151 struct bpf_map *map = link->map;
152 struct sk_psock_progs *progs = sock_map_progs(map);
153
154 if (psock->saved_data_ready && progs->stream_parser)
155 strp_stop = true;
156 if (psock->saved_data_ready && progs->stream_verdict)
157 verdict_stop = true;
158 if (psock->saved_data_ready && progs->skb_verdict)
159 verdict_stop = true;
160 list_del(&link->list);
161 sk_psock_free_link(link);
162 break;
163 }
164 }
165 spin_unlock_bh(&psock->link_lock);
166 if (strp_stop || verdict_stop) {
167 write_lock_bh(&sk->sk_callback_lock);
168 if (strp_stop)
169 sk_psock_stop_strp(sk, psock);
170 if (verdict_stop)
171 sk_psock_stop_verdict(sk, psock);
172
173 if (psock->psock_update_sk_prot)
174 psock->psock_update_sk_prot(sk, psock, false);
175 write_unlock_bh(&sk->sk_callback_lock);
176 }
177}
178
179static void sock_map_unref(struct sock *sk, void *link_raw)
180{
181 struct sk_psock *psock = sk_psock(sk);
182
183 if (likely(psock)) {
184 sock_map_del_link(sk, psock, link_raw);
185 sk_psock_put(sk, psock);
186 }
187}
188
189static int sock_map_init_proto(struct sock *sk, struct sk_psock *psock)
190{
191 if (!sk->sk_prot->psock_update_sk_prot)
192 return -EINVAL;
193 psock->psock_update_sk_prot = sk->sk_prot->psock_update_sk_prot;
194 return sk->sk_prot->psock_update_sk_prot(sk, psock, false);
195}
196
197static struct sk_psock *sock_map_psock_get_checked(struct sock *sk)
198{
199 struct sk_psock *psock;
200
201 rcu_read_lock();
202 psock = sk_psock(sk);
203 if (psock) {
204 if (sk->sk_prot->close != sock_map_close) {
205 psock = ERR_PTR(-EBUSY);
206 goto out;
207 }
208
209 if (!refcount_inc_not_zero(&psock->refcnt))
210 psock = ERR_PTR(-EBUSY);
211 }
212out:
213 rcu_read_unlock();
214 return psock;
215}
216
217static int sock_map_link(struct bpf_map *map, struct sock *sk)
218{
219 struct sk_psock_progs *progs = sock_map_progs(map);
220 struct bpf_prog *stream_verdict = NULL;
221 struct bpf_prog *stream_parser = NULL;
222 struct bpf_prog *skb_verdict = NULL;
223 struct bpf_prog *msg_parser = NULL;
224 struct sk_psock *psock;
225 int ret;
226
227 stream_verdict = READ_ONCE(progs->stream_verdict);
228 if (stream_verdict) {
229 stream_verdict = bpf_prog_inc_not_zero(stream_verdict);
230 if (IS_ERR(stream_verdict))
231 return PTR_ERR(stream_verdict);
232 }
233
234 stream_parser = READ_ONCE(progs->stream_parser);
235 if (stream_parser) {
236 stream_parser = bpf_prog_inc_not_zero(stream_parser);
237 if (IS_ERR(stream_parser)) {
238 ret = PTR_ERR(stream_parser);
239 goto out_put_stream_verdict;
240 }
241 }
242
243 msg_parser = READ_ONCE(progs->msg_parser);
244 if (msg_parser) {
245 msg_parser = bpf_prog_inc_not_zero(msg_parser);
246 if (IS_ERR(msg_parser)) {
247 ret = PTR_ERR(msg_parser);
248 goto out_put_stream_parser;
249 }
250 }
251
252 skb_verdict = READ_ONCE(progs->skb_verdict);
253 if (skb_verdict) {
254 skb_verdict = bpf_prog_inc_not_zero(skb_verdict);
255 if (IS_ERR(skb_verdict)) {
256 ret = PTR_ERR(skb_verdict);
257 goto out_put_msg_parser;
258 }
259 }
260
261 psock = sock_map_psock_get_checked(sk);
262 if (IS_ERR(psock)) {
263 ret = PTR_ERR(psock);
264 goto out_progs;
265 }
266
267 if (psock) {
268 if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
269 (stream_parser && READ_ONCE(psock->progs.stream_parser)) ||
270 (skb_verdict && READ_ONCE(psock->progs.skb_verdict)) ||
271 (skb_verdict && READ_ONCE(psock->progs.stream_verdict)) ||
272 (stream_verdict && READ_ONCE(psock->progs.skb_verdict)) ||
273 (stream_verdict && READ_ONCE(psock->progs.stream_verdict))) {
274 sk_psock_put(sk, psock);
275 ret = -EBUSY;
276 goto out_progs;
277 }
278 } else {
279 psock = sk_psock_init(sk, map->numa_node);
280 if (IS_ERR(psock)) {
281 ret = PTR_ERR(psock);
282 goto out_progs;
283 }
284 }
285
286 if (msg_parser)
287 psock_set_prog(&psock->progs.msg_parser, msg_parser);
288 if (stream_parser)
289 psock_set_prog(&psock->progs.stream_parser, stream_parser);
290 if (stream_verdict)
291 psock_set_prog(&psock->progs.stream_verdict, stream_verdict);
292 if (skb_verdict)
293 psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
294
295 /* msg_* and stream_* programs references tracked in psock after this
296 * point. Reference dec and cleanup will occur through psock destructor
297 */
298 ret = sock_map_init_proto(sk, psock);
299 if (ret < 0) {
300 sk_psock_put(sk, psock);
301 goto out;
302 }
303
304 write_lock_bh(&sk->sk_callback_lock);
305 if (stream_parser && stream_verdict && !psock->saved_data_ready) {
306 if (sk_is_tcp(sk))
307 ret = sk_psock_init_strp(sk, psock);
308 else
309 ret = -EOPNOTSUPP;
310 if (ret) {
311 write_unlock_bh(&sk->sk_callback_lock);
312 sk_psock_put(sk, psock);
313 goto out;
314 }
315 sk_psock_start_strp(sk, psock);
316 } else if (!stream_parser && stream_verdict && !psock->saved_data_ready) {
317 sk_psock_start_verdict(sk,psock);
318 } else if (!stream_verdict && skb_verdict && !psock->saved_data_ready) {
319 sk_psock_start_verdict(sk, psock);
320 }
321 write_unlock_bh(&sk->sk_callback_lock);
322 return 0;
323out_progs:
324 if (skb_verdict)
325 bpf_prog_put(skb_verdict);
326out_put_msg_parser:
327 if (msg_parser)
328 bpf_prog_put(msg_parser);
329out_put_stream_parser:
330 if (stream_parser)
331 bpf_prog_put(stream_parser);
332out_put_stream_verdict:
333 if (stream_verdict)
334 bpf_prog_put(stream_verdict);
335out:
336 return ret;
337}
338
339static void sock_map_free(struct bpf_map *map)
340{
341 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
342 int i;
343
344 /* After the sync no updates or deletes will be in-flight so it
345 * is safe to walk map and remove entries without risking a race
346 * in EEXIST update case.
347 */
348 synchronize_rcu();
349 for (i = 0; i < stab->map.max_entries; i++) {
350 struct sock **psk = &stab->sks[i];
351 struct sock *sk;
352
353 sk = xchg(psk, NULL);
354 if (sk) {
355 sock_hold(sk);
356 lock_sock(sk);
357 rcu_read_lock();
358 sock_map_unref(sk, psk);
359 rcu_read_unlock();
360 release_sock(sk);
361 sock_put(sk);
362 }
363 }
364
365 /* wait for psock readers accessing its map link */
366 synchronize_rcu();
367
368 bpf_map_area_free(stab->sks);
369 bpf_map_area_free(stab);
370}
371
372static void sock_map_release_progs(struct bpf_map *map)
373{
374 psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
375}
376
377static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
378{
379 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
380
381 WARN_ON_ONCE(!rcu_read_lock_held());
382
383 if (unlikely(key >= map->max_entries))
384 return NULL;
385 return READ_ONCE(stab->sks[key]);
386}
387
388static void *sock_map_lookup(struct bpf_map *map, void *key)
389{
390 struct sock *sk;
391
392 sk = __sock_map_lookup_elem(map, *(u32 *)key);
393 if (!sk)
394 return NULL;
395 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
396 return NULL;
397 return sk;
398}
399
400static void *sock_map_lookup_sys(struct bpf_map *map, void *key)
401{
402 struct sock *sk;
403
404 if (map->value_size != sizeof(u64))
405 return ERR_PTR(-ENOSPC);
406
407 sk = __sock_map_lookup_elem(map, *(u32 *)key);
408 if (!sk)
409 return ERR_PTR(-ENOENT);
410
411 __sock_gen_cookie(sk);
412 return &sk->sk_cookie;
413}
414
415static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
416 struct sock **psk)
417{
418 struct sock *sk = NULL;
419 int err = 0;
420
421 spin_lock_bh(&stab->lock);
422 if (!sk_test || sk_test == *psk)
423 sk = xchg(psk, NULL);
424
425 if (likely(sk))
426 sock_map_unref(sk, psk);
427 else
428 err = -EINVAL;
429
430 spin_unlock_bh(&stab->lock);
431 return err;
432}
433
434static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
435 void *link_raw)
436{
437 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
438
439 __sock_map_delete(stab, sk, link_raw);
440}
441
442static long sock_map_delete_elem(struct bpf_map *map, void *key)
443{
444 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
445 u32 i = *(u32 *)key;
446 struct sock **psk;
447
448 if (unlikely(i >= map->max_entries))
449 return -EINVAL;
450
451 psk = &stab->sks[i];
452 return __sock_map_delete(stab, NULL, psk);
453}
454
455static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
456{
457 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
458 u32 i = key ? *(u32 *)key : U32_MAX;
459 u32 *key_next = next;
460
461 if (i == stab->map.max_entries - 1)
462 return -ENOENT;
463 if (i >= stab->map.max_entries)
464 *key_next = 0;
465 else
466 *key_next = i + 1;
467 return 0;
468}
469
470static int sock_map_update_common(struct bpf_map *map, u32 idx,
471 struct sock *sk, u64 flags)
472{
473 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
474 struct sk_psock_link *link;
475 struct sk_psock *psock;
476 struct sock *osk;
477 int ret;
478
479 WARN_ON_ONCE(!rcu_read_lock_held());
480 if (unlikely(flags > BPF_EXIST))
481 return -EINVAL;
482 if (unlikely(idx >= map->max_entries))
483 return -E2BIG;
484
485 link = sk_psock_init_link();
486 if (!link)
487 return -ENOMEM;
488
489 ret = sock_map_link(map, sk);
490 if (ret < 0)
491 goto out_free;
492
493 psock = sk_psock(sk);
494 WARN_ON_ONCE(!psock);
495
496 spin_lock_bh(&stab->lock);
497 osk = stab->sks[idx];
498 if (osk && flags == BPF_NOEXIST) {
499 ret = -EEXIST;
500 goto out_unlock;
501 } else if (!osk && flags == BPF_EXIST) {
502 ret = -ENOENT;
503 goto out_unlock;
504 }
505
506 sock_map_add_link(psock, link, map, &stab->sks[idx]);
507 stab->sks[idx] = sk;
508 if (osk)
509 sock_map_unref(osk, &stab->sks[idx]);
510 spin_unlock_bh(&stab->lock);
511 return 0;
512out_unlock:
513 spin_unlock_bh(&stab->lock);
514 if (psock)
515 sk_psock_put(sk, psock);
516out_free:
517 sk_psock_free_link(link);
518 return ret;
519}
520
521static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
522{
523 return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
524 ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
525 ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
526}
527
528static bool sock_map_redirect_allowed(const struct sock *sk)
529{
530 if (sk_is_tcp(sk))
531 return sk->sk_state != TCP_LISTEN;
532 else
533 return sk->sk_state == TCP_ESTABLISHED;
534}
535
536static bool sock_map_sk_is_suitable(const struct sock *sk)
537{
538 return !!sk->sk_prot->psock_update_sk_prot;
539}
540
541static bool sock_map_sk_state_allowed(const struct sock *sk)
542{
543 if (sk_is_tcp(sk))
544 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
545 if (sk_is_stream_unix(sk))
546 return (1 << sk->sk_state) & TCPF_ESTABLISHED;
547 if (sk_is_vsock(sk) &&
548 (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET))
549 return (1 << sk->sk_state) & TCPF_ESTABLISHED;
550 return true;
551}
552
553static int sock_hash_update_common(struct bpf_map *map, void *key,
554 struct sock *sk, u64 flags);
555
556int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
557 u64 flags)
558{
559 struct socket *sock;
560 struct sock *sk;
561 int ret;
562 u64 ufd;
563
564 if (map->value_size == sizeof(u64))
565 ufd = *(u64 *)value;
566 else
567 ufd = *(u32 *)value;
568 if (ufd > S32_MAX)
569 return -EINVAL;
570
571 sock = sockfd_lookup(ufd, &ret);
572 if (!sock)
573 return ret;
574 sk = sock->sk;
575 if (!sk) {
576 ret = -EINVAL;
577 goto out;
578 }
579 if (!sock_map_sk_is_suitable(sk)) {
580 ret = -EOPNOTSUPP;
581 goto out;
582 }
583
584 sock_map_sk_acquire(sk);
585 if (!sock_map_sk_state_allowed(sk))
586 ret = -EOPNOTSUPP;
587 else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
588 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
589 else
590 ret = sock_hash_update_common(map, key, sk, flags);
591 sock_map_sk_release(sk);
592out:
593 sockfd_put(sock);
594 return ret;
595}
596
597static long sock_map_update_elem(struct bpf_map *map, void *key,
598 void *value, u64 flags)
599{
600 struct sock *sk = (struct sock *)value;
601 int ret;
602
603 if (unlikely(!sk || !sk_fullsock(sk)))
604 return -EINVAL;
605
606 if (!sock_map_sk_is_suitable(sk))
607 return -EOPNOTSUPP;
608
609 local_bh_disable();
610 bh_lock_sock(sk);
611 if (!sock_map_sk_state_allowed(sk))
612 ret = -EOPNOTSUPP;
613 else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
614 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
615 else
616 ret = sock_hash_update_common(map, key, sk, flags);
617 bh_unlock_sock(sk);
618 local_bh_enable();
619 return ret;
620}
621
622BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
623 struct bpf_map *, map, void *, key, u64, flags)
624{
625 WARN_ON_ONCE(!rcu_read_lock_held());
626
627 if (likely(sock_map_sk_is_suitable(sops->sk) &&
628 sock_map_op_okay(sops)))
629 return sock_map_update_common(map, *(u32 *)key, sops->sk,
630 flags);
631 return -EOPNOTSUPP;
632}
633
634const struct bpf_func_proto bpf_sock_map_update_proto = {
635 .func = bpf_sock_map_update,
636 .gpl_only = false,
637 .pkt_access = true,
638 .ret_type = RET_INTEGER,
639 .arg1_type = ARG_PTR_TO_CTX,
640 .arg2_type = ARG_CONST_MAP_PTR,
641 .arg3_type = ARG_PTR_TO_MAP_KEY,
642 .arg4_type = ARG_ANYTHING,
643};
644
645BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
646 struct bpf_map *, map, u32, key, u64, flags)
647{
648 struct sock *sk;
649
650 if (unlikely(flags & ~(BPF_F_INGRESS)))
651 return SK_DROP;
652
653 sk = __sock_map_lookup_elem(map, key);
654 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
655 return SK_DROP;
656 if ((flags & BPF_F_INGRESS) && sk_is_vsock(sk))
657 return SK_DROP;
658
659 skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
660 return SK_PASS;
661}
662
663const struct bpf_func_proto bpf_sk_redirect_map_proto = {
664 .func = bpf_sk_redirect_map,
665 .gpl_only = false,
666 .ret_type = RET_INTEGER,
667 .arg1_type = ARG_PTR_TO_CTX,
668 .arg2_type = ARG_CONST_MAP_PTR,
669 .arg3_type = ARG_ANYTHING,
670 .arg4_type = ARG_ANYTHING,
671};
672
673BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
674 struct bpf_map *, map, u32, key, u64, flags)
675{
676 struct sock *sk;
677
678 if (unlikely(flags & ~(BPF_F_INGRESS)))
679 return SK_DROP;
680
681 sk = __sock_map_lookup_elem(map, key);
682 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
683 return SK_DROP;
684 if (!(flags & BPF_F_INGRESS) && !sk_is_tcp(sk))
685 return SK_DROP;
686 if (sk_is_vsock(sk))
687 return SK_DROP;
688
689 msg->flags = flags;
690 msg->sk_redir = sk;
691 return SK_PASS;
692}
693
694const struct bpf_func_proto bpf_msg_redirect_map_proto = {
695 .func = bpf_msg_redirect_map,
696 .gpl_only = false,
697 .ret_type = RET_INTEGER,
698 .arg1_type = ARG_PTR_TO_CTX,
699 .arg2_type = ARG_CONST_MAP_PTR,
700 .arg3_type = ARG_ANYTHING,
701 .arg4_type = ARG_ANYTHING,
702};
703
704struct sock_map_seq_info {
705 struct bpf_map *map;
706 struct sock *sk;
707 u32 index;
708};
709
710struct bpf_iter__sockmap {
711 __bpf_md_ptr(struct bpf_iter_meta *, meta);
712 __bpf_md_ptr(struct bpf_map *, map);
713 __bpf_md_ptr(void *, key);
714 __bpf_md_ptr(struct sock *, sk);
715};
716
717DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
718 struct bpf_map *map, void *key,
719 struct sock *sk)
720
721static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
722{
723 if (unlikely(info->index >= info->map->max_entries))
724 return NULL;
725
726 info->sk = __sock_map_lookup_elem(info->map, info->index);
727
728 /* can't return sk directly, since that might be NULL */
729 return info;
730}
731
732static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
733 __acquires(rcu)
734{
735 struct sock_map_seq_info *info = seq->private;
736
737 if (*pos == 0)
738 ++*pos;
739
740 /* pairs with sock_map_seq_stop */
741 rcu_read_lock();
742 return sock_map_seq_lookup_elem(info);
743}
744
745static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
746 __must_hold(rcu)
747{
748 struct sock_map_seq_info *info = seq->private;
749
750 ++*pos;
751 ++info->index;
752
753 return sock_map_seq_lookup_elem(info);
754}
755
756static int sock_map_seq_show(struct seq_file *seq, void *v)
757 __must_hold(rcu)
758{
759 struct sock_map_seq_info *info = seq->private;
760 struct bpf_iter__sockmap ctx = {};
761 struct bpf_iter_meta meta;
762 struct bpf_prog *prog;
763
764 meta.seq = seq;
765 prog = bpf_iter_get_info(&meta, !v);
766 if (!prog)
767 return 0;
768
769 ctx.meta = &meta;
770 ctx.map = info->map;
771 if (v) {
772 ctx.key = &info->index;
773 ctx.sk = info->sk;
774 }
775
776 return bpf_iter_run_prog(prog, &ctx);
777}
778
779static void sock_map_seq_stop(struct seq_file *seq, void *v)
780 __releases(rcu)
781{
782 if (!v)
783 (void)sock_map_seq_show(seq, NULL);
784
785 /* pairs with sock_map_seq_start */
786 rcu_read_unlock();
787}
788
789static const struct seq_operations sock_map_seq_ops = {
790 .start = sock_map_seq_start,
791 .next = sock_map_seq_next,
792 .stop = sock_map_seq_stop,
793 .show = sock_map_seq_show,
794};
795
796static int sock_map_init_seq_private(void *priv_data,
797 struct bpf_iter_aux_info *aux)
798{
799 struct sock_map_seq_info *info = priv_data;
800
801 bpf_map_inc_with_uref(aux->map);
802 info->map = aux->map;
803 return 0;
804}
805
806static void sock_map_fini_seq_private(void *priv_data)
807{
808 struct sock_map_seq_info *info = priv_data;
809
810 bpf_map_put_with_uref(info->map);
811}
812
813static u64 sock_map_mem_usage(const struct bpf_map *map)
814{
815 u64 usage = sizeof(struct bpf_stab);
816
817 usage += (u64)map->max_entries * sizeof(struct sock *);
818 return usage;
819}
820
821static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
822 .seq_ops = &sock_map_seq_ops,
823 .init_seq_private = sock_map_init_seq_private,
824 .fini_seq_private = sock_map_fini_seq_private,
825 .seq_priv_size = sizeof(struct sock_map_seq_info),
826};
827
828BTF_ID_LIST_SINGLE(sock_map_btf_ids, struct, bpf_stab)
829const struct bpf_map_ops sock_map_ops = {
830 .map_meta_equal = bpf_map_meta_equal,
831 .map_alloc = sock_map_alloc,
832 .map_free = sock_map_free,
833 .map_get_next_key = sock_map_get_next_key,
834 .map_lookup_elem_sys_only = sock_map_lookup_sys,
835 .map_update_elem = sock_map_update_elem,
836 .map_delete_elem = sock_map_delete_elem,
837 .map_lookup_elem = sock_map_lookup,
838 .map_release_uref = sock_map_release_progs,
839 .map_check_btf = map_check_no_btf,
840 .map_mem_usage = sock_map_mem_usage,
841 .map_btf_id = &sock_map_btf_ids[0],
842 .iter_seq_info = &sock_map_iter_seq_info,
843};
844
845struct bpf_shtab_elem {
846 struct rcu_head rcu;
847 u32 hash;
848 struct sock *sk;
849 struct hlist_node node;
850 u8 key[];
851};
852
853struct bpf_shtab_bucket {
854 struct hlist_head head;
855 spinlock_t lock;
856};
857
858struct bpf_shtab {
859 struct bpf_map map;
860 struct bpf_shtab_bucket *buckets;
861 u32 buckets_num;
862 u32 elem_size;
863 struct sk_psock_progs progs;
864 atomic_t count;
865};
866
867static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
868{
869 return jhash(key, len, 0);
870}
871
872static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
873 u32 hash)
874{
875 return &htab->buckets[hash & (htab->buckets_num - 1)];
876}
877
878static struct bpf_shtab_elem *
879sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
880 u32 key_size)
881{
882 struct bpf_shtab_elem *elem;
883
884 hlist_for_each_entry_rcu(elem, head, node) {
885 if (elem->hash == hash &&
886 !memcmp(&elem->key, key, key_size))
887 return elem;
888 }
889
890 return NULL;
891}
892
893static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
894{
895 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
896 u32 key_size = map->key_size, hash;
897 struct bpf_shtab_bucket *bucket;
898 struct bpf_shtab_elem *elem;
899
900 WARN_ON_ONCE(!rcu_read_lock_held());
901
902 hash = sock_hash_bucket_hash(key, key_size);
903 bucket = sock_hash_select_bucket(htab, hash);
904 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
905
906 return elem ? elem->sk : NULL;
907}
908
909static void sock_hash_free_elem(struct bpf_shtab *htab,
910 struct bpf_shtab_elem *elem)
911{
912 atomic_dec(&htab->count);
913 kfree_rcu(elem, rcu);
914}
915
916static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
917 void *link_raw)
918{
919 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
920 struct bpf_shtab_elem *elem_probe, *elem = link_raw;
921 struct bpf_shtab_bucket *bucket;
922
923 WARN_ON_ONCE(!rcu_read_lock_held());
924 bucket = sock_hash_select_bucket(htab, elem->hash);
925
926 /* elem may be deleted in parallel from the map, but access here
927 * is okay since it's going away only after RCU grace period.
928 * However, we need to check whether it's still present.
929 */
930 spin_lock_bh(&bucket->lock);
931 elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
932 elem->key, map->key_size);
933 if (elem_probe && elem_probe == elem) {
934 hlist_del_rcu(&elem->node);
935 sock_map_unref(elem->sk, elem);
936 sock_hash_free_elem(htab, elem);
937 }
938 spin_unlock_bh(&bucket->lock);
939}
940
941static long sock_hash_delete_elem(struct bpf_map *map, void *key)
942{
943 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
944 u32 hash, key_size = map->key_size;
945 struct bpf_shtab_bucket *bucket;
946 struct bpf_shtab_elem *elem;
947 int ret = -ENOENT;
948
949 hash = sock_hash_bucket_hash(key, key_size);
950 bucket = sock_hash_select_bucket(htab, hash);
951
952 spin_lock_bh(&bucket->lock);
953 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
954 if (elem) {
955 hlist_del_rcu(&elem->node);
956 sock_map_unref(elem->sk, elem);
957 sock_hash_free_elem(htab, elem);
958 ret = 0;
959 }
960 spin_unlock_bh(&bucket->lock);
961 return ret;
962}
963
964static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
965 void *key, u32 key_size,
966 u32 hash, struct sock *sk,
967 struct bpf_shtab_elem *old)
968{
969 struct bpf_shtab_elem *new;
970
971 if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
972 if (!old) {
973 atomic_dec(&htab->count);
974 return ERR_PTR(-E2BIG);
975 }
976 }
977
978 new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
979 GFP_ATOMIC | __GFP_NOWARN,
980 htab->map.numa_node);
981 if (!new) {
982 atomic_dec(&htab->count);
983 return ERR_PTR(-ENOMEM);
984 }
985 memcpy(new->key, key, key_size);
986 new->sk = sk;
987 new->hash = hash;
988 return new;
989}
990
991static int sock_hash_update_common(struct bpf_map *map, void *key,
992 struct sock *sk, u64 flags)
993{
994 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
995 u32 key_size = map->key_size, hash;
996 struct bpf_shtab_elem *elem, *elem_new;
997 struct bpf_shtab_bucket *bucket;
998 struct sk_psock_link *link;
999 struct sk_psock *psock;
1000 int ret;
1001
1002 WARN_ON_ONCE(!rcu_read_lock_held());
1003 if (unlikely(flags > BPF_EXIST))
1004 return -EINVAL;
1005
1006 link = sk_psock_init_link();
1007 if (!link)
1008 return -ENOMEM;
1009
1010 ret = sock_map_link(map, sk);
1011 if (ret < 0)
1012 goto out_free;
1013
1014 psock = sk_psock(sk);
1015 WARN_ON_ONCE(!psock);
1016
1017 hash = sock_hash_bucket_hash(key, key_size);
1018 bucket = sock_hash_select_bucket(htab, hash);
1019
1020 spin_lock_bh(&bucket->lock);
1021 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
1022 if (elem && flags == BPF_NOEXIST) {
1023 ret = -EEXIST;
1024 goto out_unlock;
1025 } else if (!elem && flags == BPF_EXIST) {
1026 ret = -ENOENT;
1027 goto out_unlock;
1028 }
1029
1030 elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1031 if (IS_ERR(elem_new)) {
1032 ret = PTR_ERR(elem_new);
1033 goto out_unlock;
1034 }
1035
1036 sock_map_add_link(psock, link, map, elem_new);
1037 /* Add new element to the head of the list, so that
1038 * concurrent search will find it before old elem.
1039 */
1040 hlist_add_head_rcu(&elem_new->node, &bucket->head);
1041 if (elem) {
1042 hlist_del_rcu(&elem->node);
1043 sock_map_unref(elem->sk, elem);
1044 sock_hash_free_elem(htab, elem);
1045 }
1046 spin_unlock_bh(&bucket->lock);
1047 return 0;
1048out_unlock:
1049 spin_unlock_bh(&bucket->lock);
1050 sk_psock_put(sk, psock);
1051out_free:
1052 sk_psock_free_link(link);
1053 return ret;
1054}
1055
1056static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1057 void *key_next)
1058{
1059 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1060 struct bpf_shtab_elem *elem, *elem_next;
1061 u32 hash, key_size = map->key_size;
1062 struct hlist_head *head;
1063 int i = 0;
1064
1065 if (!key)
1066 goto find_first_elem;
1067 hash = sock_hash_bucket_hash(key, key_size);
1068 head = &sock_hash_select_bucket(htab, hash)->head;
1069 elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1070 if (!elem)
1071 goto find_first_elem;
1072
1073 elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1074 struct bpf_shtab_elem, node);
1075 if (elem_next) {
1076 memcpy(key_next, elem_next->key, key_size);
1077 return 0;
1078 }
1079
1080 i = hash & (htab->buckets_num - 1);
1081 i++;
1082find_first_elem:
1083 for (; i < htab->buckets_num; i++) {
1084 head = &sock_hash_select_bucket(htab, i)->head;
1085 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1086 struct bpf_shtab_elem, node);
1087 if (elem_next) {
1088 memcpy(key_next, elem_next->key, key_size);
1089 return 0;
1090 }
1091 }
1092
1093 return -ENOENT;
1094}
1095
1096static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1097{
1098 struct bpf_shtab *htab;
1099 int i, err;
1100
1101 if (attr->max_entries == 0 ||
1102 attr->key_size == 0 ||
1103 (attr->value_size != sizeof(u32) &&
1104 attr->value_size != sizeof(u64)) ||
1105 attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1106 return ERR_PTR(-EINVAL);
1107 if (attr->key_size > MAX_BPF_STACK)
1108 return ERR_PTR(-E2BIG);
1109
1110 htab = bpf_map_area_alloc(sizeof(*htab), NUMA_NO_NODE);
1111 if (!htab)
1112 return ERR_PTR(-ENOMEM);
1113
1114 bpf_map_init_from_attr(&htab->map, attr);
1115
1116 htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1117 htab->elem_size = sizeof(struct bpf_shtab_elem) +
1118 round_up(htab->map.key_size, 8);
1119 if (htab->buckets_num == 0 ||
1120 htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1121 err = -EINVAL;
1122 goto free_htab;
1123 }
1124
1125 htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1126 sizeof(struct bpf_shtab_bucket),
1127 htab->map.numa_node);
1128 if (!htab->buckets) {
1129 err = -ENOMEM;
1130 goto free_htab;
1131 }
1132
1133 for (i = 0; i < htab->buckets_num; i++) {
1134 INIT_HLIST_HEAD(&htab->buckets[i].head);
1135 spin_lock_init(&htab->buckets[i].lock);
1136 }
1137
1138 return &htab->map;
1139free_htab:
1140 bpf_map_area_free(htab);
1141 return ERR_PTR(err);
1142}
1143
1144static void sock_hash_free(struct bpf_map *map)
1145{
1146 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1147 struct bpf_shtab_bucket *bucket;
1148 struct hlist_head unlink_list;
1149 struct bpf_shtab_elem *elem;
1150 struct hlist_node *node;
1151 int i;
1152
1153 /* After the sync no updates or deletes will be in-flight so it
1154 * is safe to walk map and remove entries without risking a race
1155 * in EEXIST update case.
1156 */
1157 synchronize_rcu();
1158 for (i = 0; i < htab->buckets_num; i++) {
1159 bucket = sock_hash_select_bucket(htab, i);
1160
1161 /* We are racing with sock_hash_delete_from_link to
1162 * enter the spin-lock critical section. Every socket on
1163 * the list is still linked to sockhash. Since link
1164 * exists, psock exists and holds a ref to socket. That
1165 * lets us to grab a socket ref too.
1166 */
1167 spin_lock_bh(&bucket->lock);
1168 hlist_for_each_entry(elem, &bucket->head, node)
1169 sock_hold(elem->sk);
1170 hlist_move_list(&bucket->head, &unlink_list);
1171 spin_unlock_bh(&bucket->lock);
1172
1173 /* Process removed entries out of atomic context to
1174 * block for socket lock before deleting the psock's
1175 * link to sockhash.
1176 */
1177 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1178 hlist_del(&elem->node);
1179 lock_sock(elem->sk);
1180 rcu_read_lock();
1181 sock_map_unref(elem->sk, elem);
1182 rcu_read_unlock();
1183 release_sock(elem->sk);
1184 sock_put(elem->sk);
1185 sock_hash_free_elem(htab, elem);
1186 }
1187 cond_resched();
1188 }
1189
1190 /* wait for psock readers accessing its map link */
1191 synchronize_rcu();
1192
1193 bpf_map_area_free(htab->buckets);
1194 bpf_map_area_free(htab);
1195}
1196
1197static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1198{
1199 struct sock *sk;
1200
1201 if (map->value_size != sizeof(u64))
1202 return ERR_PTR(-ENOSPC);
1203
1204 sk = __sock_hash_lookup_elem(map, key);
1205 if (!sk)
1206 return ERR_PTR(-ENOENT);
1207
1208 __sock_gen_cookie(sk);
1209 return &sk->sk_cookie;
1210}
1211
1212static void *sock_hash_lookup(struct bpf_map *map, void *key)
1213{
1214 struct sock *sk;
1215
1216 sk = __sock_hash_lookup_elem(map, key);
1217 if (!sk)
1218 return NULL;
1219 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1220 return NULL;
1221 return sk;
1222}
1223
1224static void sock_hash_release_progs(struct bpf_map *map)
1225{
1226 psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1227}
1228
1229BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1230 struct bpf_map *, map, void *, key, u64, flags)
1231{
1232 WARN_ON_ONCE(!rcu_read_lock_held());
1233
1234 if (likely(sock_map_sk_is_suitable(sops->sk) &&
1235 sock_map_op_okay(sops)))
1236 return sock_hash_update_common(map, key, sops->sk, flags);
1237 return -EOPNOTSUPP;
1238}
1239
1240const struct bpf_func_proto bpf_sock_hash_update_proto = {
1241 .func = bpf_sock_hash_update,
1242 .gpl_only = false,
1243 .pkt_access = true,
1244 .ret_type = RET_INTEGER,
1245 .arg1_type = ARG_PTR_TO_CTX,
1246 .arg2_type = ARG_CONST_MAP_PTR,
1247 .arg3_type = ARG_PTR_TO_MAP_KEY,
1248 .arg4_type = ARG_ANYTHING,
1249};
1250
1251BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1252 struct bpf_map *, map, void *, key, u64, flags)
1253{
1254 struct sock *sk;
1255
1256 if (unlikely(flags & ~(BPF_F_INGRESS)))
1257 return SK_DROP;
1258
1259 sk = __sock_hash_lookup_elem(map, key);
1260 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1261 return SK_DROP;
1262 if ((flags & BPF_F_INGRESS) && sk_is_vsock(sk))
1263 return SK_DROP;
1264
1265 skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1266 return SK_PASS;
1267}
1268
1269const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1270 .func = bpf_sk_redirect_hash,
1271 .gpl_only = false,
1272 .ret_type = RET_INTEGER,
1273 .arg1_type = ARG_PTR_TO_CTX,
1274 .arg2_type = ARG_CONST_MAP_PTR,
1275 .arg3_type = ARG_PTR_TO_MAP_KEY,
1276 .arg4_type = ARG_ANYTHING,
1277};
1278
1279BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1280 struct bpf_map *, map, void *, key, u64, flags)
1281{
1282 struct sock *sk;
1283
1284 if (unlikely(flags & ~(BPF_F_INGRESS)))
1285 return SK_DROP;
1286
1287 sk = __sock_hash_lookup_elem(map, key);
1288 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1289 return SK_DROP;
1290 if (!(flags & BPF_F_INGRESS) && !sk_is_tcp(sk))
1291 return SK_DROP;
1292 if (sk_is_vsock(sk))
1293 return SK_DROP;
1294
1295 msg->flags = flags;
1296 msg->sk_redir = sk;
1297 return SK_PASS;
1298}
1299
1300const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1301 .func = bpf_msg_redirect_hash,
1302 .gpl_only = false,
1303 .ret_type = RET_INTEGER,
1304 .arg1_type = ARG_PTR_TO_CTX,
1305 .arg2_type = ARG_CONST_MAP_PTR,
1306 .arg3_type = ARG_PTR_TO_MAP_KEY,
1307 .arg4_type = ARG_ANYTHING,
1308};
1309
1310struct sock_hash_seq_info {
1311 struct bpf_map *map;
1312 struct bpf_shtab *htab;
1313 u32 bucket_id;
1314};
1315
1316static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1317 struct bpf_shtab_elem *prev_elem)
1318{
1319 const struct bpf_shtab *htab = info->htab;
1320 struct bpf_shtab_bucket *bucket;
1321 struct bpf_shtab_elem *elem;
1322 struct hlist_node *node;
1323
1324 /* try to find next elem in the same bucket */
1325 if (prev_elem) {
1326 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1327 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1328 if (elem)
1329 return elem;
1330
1331 /* no more elements, continue in the next bucket */
1332 info->bucket_id++;
1333 }
1334
1335 for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1336 bucket = &htab->buckets[info->bucket_id];
1337 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1338 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1339 if (elem)
1340 return elem;
1341 }
1342
1343 return NULL;
1344}
1345
1346static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1347 __acquires(rcu)
1348{
1349 struct sock_hash_seq_info *info = seq->private;
1350
1351 if (*pos == 0)
1352 ++*pos;
1353
1354 /* pairs with sock_hash_seq_stop */
1355 rcu_read_lock();
1356 return sock_hash_seq_find_next(info, NULL);
1357}
1358
1359static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1360 __must_hold(rcu)
1361{
1362 struct sock_hash_seq_info *info = seq->private;
1363
1364 ++*pos;
1365 return sock_hash_seq_find_next(info, v);
1366}
1367
1368static int sock_hash_seq_show(struct seq_file *seq, void *v)
1369 __must_hold(rcu)
1370{
1371 struct sock_hash_seq_info *info = seq->private;
1372 struct bpf_iter__sockmap ctx = {};
1373 struct bpf_shtab_elem *elem = v;
1374 struct bpf_iter_meta meta;
1375 struct bpf_prog *prog;
1376
1377 meta.seq = seq;
1378 prog = bpf_iter_get_info(&meta, !elem);
1379 if (!prog)
1380 return 0;
1381
1382 ctx.meta = &meta;
1383 ctx.map = info->map;
1384 if (elem) {
1385 ctx.key = elem->key;
1386 ctx.sk = elem->sk;
1387 }
1388
1389 return bpf_iter_run_prog(prog, &ctx);
1390}
1391
1392static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1393 __releases(rcu)
1394{
1395 if (!v)
1396 (void)sock_hash_seq_show(seq, NULL);
1397
1398 /* pairs with sock_hash_seq_start */
1399 rcu_read_unlock();
1400}
1401
1402static const struct seq_operations sock_hash_seq_ops = {
1403 .start = sock_hash_seq_start,
1404 .next = sock_hash_seq_next,
1405 .stop = sock_hash_seq_stop,
1406 .show = sock_hash_seq_show,
1407};
1408
1409static int sock_hash_init_seq_private(void *priv_data,
1410 struct bpf_iter_aux_info *aux)
1411{
1412 struct sock_hash_seq_info *info = priv_data;
1413
1414 bpf_map_inc_with_uref(aux->map);
1415 info->map = aux->map;
1416 info->htab = container_of(aux->map, struct bpf_shtab, map);
1417 return 0;
1418}
1419
1420static void sock_hash_fini_seq_private(void *priv_data)
1421{
1422 struct sock_hash_seq_info *info = priv_data;
1423
1424 bpf_map_put_with_uref(info->map);
1425}
1426
1427static u64 sock_hash_mem_usage(const struct bpf_map *map)
1428{
1429 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1430 u64 usage = sizeof(*htab);
1431
1432 usage += htab->buckets_num * sizeof(struct bpf_shtab_bucket);
1433 usage += atomic_read(&htab->count) * (u64)htab->elem_size;
1434 return usage;
1435}
1436
1437static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1438 .seq_ops = &sock_hash_seq_ops,
1439 .init_seq_private = sock_hash_init_seq_private,
1440 .fini_seq_private = sock_hash_fini_seq_private,
1441 .seq_priv_size = sizeof(struct sock_hash_seq_info),
1442};
1443
1444BTF_ID_LIST_SINGLE(sock_hash_map_btf_ids, struct, bpf_shtab)
1445const struct bpf_map_ops sock_hash_ops = {
1446 .map_meta_equal = bpf_map_meta_equal,
1447 .map_alloc = sock_hash_alloc,
1448 .map_free = sock_hash_free,
1449 .map_get_next_key = sock_hash_get_next_key,
1450 .map_update_elem = sock_map_update_elem,
1451 .map_delete_elem = sock_hash_delete_elem,
1452 .map_lookup_elem = sock_hash_lookup,
1453 .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1454 .map_release_uref = sock_hash_release_progs,
1455 .map_check_btf = map_check_no_btf,
1456 .map_mem_usage = sock_hash_mem_usage,
1457 .map_btf_id = &sock_hash_map_btf_ids[0],
1458 .iter_seq_info = &sock_hash_iter_seq_info,
1459};
1460
1461static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1462{
1463 switch (map->map_type) {
1464 case BPF_MAP_TYPE_SOCKMAP:
1465 return &container_of(map, struct bpf_stab, map)->progs;
1466 case BPF_MAP_TYPE_SOCKHASH:
1467 return &container_of(map, struct bpf_shtab, map)->progs;
1468 default:
1469 break;
1470 }
1471
1472 return NULL;
1473}
1474
1475static int sock_map_prog_link_lookup(struct bpf_map *map, struct bpf_prog ***pprog,
1476 struct bpf_link ***plink, u32 which)
1477{
1478 struct sk_psock_progs *progs = sock_map_progs(map);
1479 struct bpf_prog **cur_pprog;
1480 struct bpf_link **cur_plink;
1481
1482 if (!progs)
1483 return -EOPNOTSUPP;
1484
1485 switch (which) {
1486 case BPF_SK_MSG_VERDICT:
1487 cur_pprog = &progs->msg_parser;
1488 cur_plink = &progs->msg_parser_link;
1489 break;
1490#if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1491 case BPF_SK_SKB_STREAM_PARSER:
1492 cur_pprog = &progs->stream_parser;
1493 cur_plink = &progs->stream_parser_link;
1494 break;
1495#endif
1496 case BPF_SK_SKB_STREAM_VERDICT:
1497 if (progs->skb_verdict)
1498 return -EBUSY;
1499 cur_pprog = &progs->stream_verdict;
1500 cur_plink = &progs->stream_verdict_link;
1501 break;
1502 case BPF_SK_SKB_VERDICT:
1503 if (progs->stream_verdict)
1504 return -EBUSY;
1505 cur_pprog = &progs->skb_verdict;
1506 cur_plink = &progs->skb_verdict_link;
1507 break;
1508 default:
1509 return -EOPNOTSUPP;
1510 }
1511
1512 *pprog = cur_pprog;
1513 if (plink)
1514 *plink = cur_plink;
1515 return 0;
1516}
1517
1518/* Handle the following four cases:
1519 * prog_attach: prog != NULL, old == NULL, link == NULL
1520 * prog_detach: prog == NULL, old != NULL, link == NULL
1521 * link_attach: prog != NULL, old == NULL, link != NULL
1522 * link_detach: prog == NULL, old != NULL, link != NULL
1523 */
1524static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1525 struct bpf_prog *old, struct bpf_link *link,
1526 u32 which)
1527{
1528 struct bpf_prog **pprog;
1529 struct bpf_link **plink;
1530 int ret;
1531
1532 ret = sock_map_prog_link_lookup(map, &pprog, &plink, which);
1533 if (ret)
1534 return ret;
1535
1536 /* for prog_attach/prog_detach/link_attach, return error if a bpf_link
1537 * exists for that prog.
1538 */
1539 if ((!link || prog) && *plink)
1540 return -EBUSY;
1541
1542 if (old) {
1543 ret = psock_replace_prog(pprog, prog, old);
1544 if (!ret)
1545 *plink = NULL;
1546 } else {
1547 psock_set_prog(pprog, prog);
1548 if (link)
1549 *plink = link;
1550 }
1551
1552 return ret;
1553}
1554
1555int sock_map_bpf_prog_query(const union bpf_attr *attr,
1556 union bpf_attr __user *uattr)
1557{
1558 __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
1559 u32 prog_cnt = 0, flags = 0;
1560 struct bpf_prog **pprog;
1561 struct bpf_prog *prog;
1562 struct bpf_map *map;
1563 u32 id = 0;
1564 int ret;
1565
1566 if (attr->query.query_flags)
1567 return -EINVAL;
1568
1569 CLASS(fd, f)(attr->target_fd);
1570 map = __bpf_map_get(f);
1571 if (IS_ERR(map))
1572 return PTR_ERR(map);
1573
1574 rcu_read_lock();
1575
1576 ret = sock_map_prog_link_lookup(map, &pprog, NULL, attr->query.attach_type);
1577 if (ret)
1578 goto end;
1579
1580 prog = *pprog;
1581 prog_cnt = !prog ? 0 : 1;
1582
1583 if (!attr->query.prog_cnt || !prog_ids || !prog_cnt)
1584 goto end;
1585
1586 /* we do not hold the refcnt, the bpf prog may be released
1587 * asynchronously and the id would be set to 0.
1588 */
1589 id = data_race(prog->aux->id);
1590 if (id == 0)
1591 prog_cnt = 0;
1592
1593end:
1594 rcu_read_unlock();
1595
1596 if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)) ||
1597 (id != 0 && copy_to_user(prog_ids, &id, sizeof(u32))) ||
1598 copy_to_user(&uattr->query.prog_cnt, &prog_cnt, sizeof(prog_cnt)))
1599 ret = -EFAULT;
1600
1601 return ret;
1602}
1603
1604static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1605{
1606 switch (link->map->map_type) {
1607 case BPF_MAP_TYPE_SOCKMAP:
1608 return sock_map_delete_from_link(link->map, sk,
1609 link->link_raw);
1610 case BPF_MAP_TYPE_SOCKHASH:
1611 return sock_hash_delete_from_link(link->map, sk,
1612 link->link_raw);
1613 default:
1614 break;
1615 }
1616}
1617
1618static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1619{
1620 struct sk_psock_link *link;
1621
1622 while ((link = sk_psock_link_pop(psock))) {
1623 sock_map_unlink(sk, link);
1624 sk_psock_free_link(link);
1625 }
1626}
1627
1628void sock_map_unhash(struct sock *sk)
1629{
1630 void (*saved_unhash)(struct sock *sk);
1631 struct sk_psock *psock;
1632
1633 rcu_read_lock();
1634 psock = sk_psock(sk);
1635 if (unlikely(!psock)) {
1636 rcu_read_unlock();
1637 saved_unhash = READ_ONCE(sk->sk_prot)->unhash;
1638 } else {
1639 saved_unhash = psock->saved_unhash;
1640 sock_map_remove_links(sk, psock);
1641 rcu_read_unlock();
1642 }
1643 if (WARN_ON_ONCE(saved_unhash == sock_map_unhash))
1644 return;
1645 if (saved_unhash)
1646 saved_unhash(sk);
1647}
1648EXPORT_SYMBOL_GPL(sock_map_unhash);
1649
1650void sock_map_destroy(struct sock *sk)
1651{
1652 void (*saved_destroy)(struct sock *sk);
1653 struct sk_psock *psock;
1654
1655 rcu_read_lock();
1656 psock = sk_psock_get(sk);
1657 if (unlikely(!psock)) {
1658 rcu_read_unlock();
1659 saved_destroy = READ_ONCE(sk->sk_prot)->destroy;
1660 } else {
1661 saved_destroy = psock->saved_destroy;
1662 sock_map_remove_links(sk, psock);
1663 rcu_read_unlock();
1664 sk_psock_stop(psock);
1665 sk_psock_put(sk, psock);
1666 }
1667 if (WARN_ON_ONCE(saved_destroy == sock_map_destroy))
1668 return;
1669 if (saved_destroy)
1670 saved_destroy(sk);
1671}
1672EXPORT_SYMBOL_GPL(sock_map_destroy);
1673
1674void sock_map_close(struct sock *sk, long timeout)
1675{
1676 void (*saved_close)(struct sock *sk, long timeout);
1677 struct sk_psock *psock;
1678
1679 lock_sock(sk);
1680 rcu_read_lock();
1681 psock = sk_psock(sk);
1682 if (likely(psock)) {
1683 saved_close = psock->saved_close;
1684 sock_map_remove_links(sk, psock);
1685 psock = sk_psock_get(sk);
1686 if (unlikely(!psock))
1687 goto no_psock;
1688 rcu_read_unlock();
1689 sk_psock_stop(psock);
1690 release_sock(sk);
1691 cancel_delayed_work_sync(&psock->work);
1692 sk_psock_put(sk, psock);
1693 } else {
1694 saved_close = READ_ONCE(sk->sk_prot)->close;
1695no_psock:
1696 rcu_read_unlock();
1697 release_sock(sk);
1698 }
1699
1700 /* Make sure we do not recurse. This is a bug.
1701 * Leak the socket instead of crashing on a stack overflow.
1702 */
1703 if (WARN_ON_ONCE(saved_close == sock_map_close))
1704 return;
1705 saved_close(sk, timeout);
1706}
1707EXPORT_SYMBOL_GPL(sock_map_close);
1708
1709struct sockmap_link {
1710 struct bpf_link link;
1711 struct bpf_map *map;
1712 enum bpf_attach_type attach_type;
1713};
1714
1715static void sock_map_link_release(struct bpf_link *link)
1716{
1717 struct sockmap_link *sockmap_link = container_of(link, struct sockmap_link, link);
1718
1719 mutex_lock(&sockmap_mutex);
1720 if (!sockmap_link->map)
1721 goto out;
1722
1723 WARN_ON_ONCE(sock_map_prog_update(sockmap_link->map, NULL, link->prog, link,
1724 sockmap_link->attach_type));
1725
1726 bpf_map_put_with_uref(sockmap_link->map);
1727 sockmap_link->map = NULL;
1728out:
1729 mutex_unlock(&sockmap_mutex);
1730}
1731
1732static int sock_map_link_detach(struct bpf_link *link)
1733{
1734 sock_map_link_release(link);
1735 return 0;
1736}
1737
1738static void sock_map_link_dealloc(struct bpf_link *link)
1739{
1740 kfree(link);
1741}
1742
1743/* Handle the following two cases:
1744 * case 1: link != NULL, prog != NULL, old != NULL
1745 * case 2: link != NULL, prog != NULL, old == NULL
1746 */
1747static int sock_map_link_update_prog(struct bpf_link *link,
1748 struct bpf_prog *prog,
1749 struct bpf_prog *old)
1750{
1751 const struct sockmap_link *sockmap_link = container_of(link, struct sockmap_link, link);
1752 struct bpf_prog **pprog, *old_link_prog;
1753 struct bpf_link **plink;
1754 int ret = 0;
1755
1756 mutex_lock(&sockmap_mutex);
1757
1758 /* If old prog is not NULL, ensure old prog is the same as link->prog. */
1759 if (old && link->prog != old) {
1760 ret = -EPERM;
1761 goto out;
1762 }
1763 /* Ensure link->prog has the same type/attach_type as the new prog. */
1764 if (link->prog->type != prog->type ||
1765 link->prog->expected_attach_type != prog->expected_attach_type) {
1766 ret = -EINVAL;
1767 goto out;
1768 }
1769 if (!sockmap_link->map) {
1770 ret = -ENOLINK;
1771 goto out;
1772 }
1773
1774 ret = sock_map_prog_link_lookup(sockmap_link->map, &pprog, &plink,
1775 sockmap_link->attach_type);
1776 if (ret)
1777 goto out;
1778
1779 /* return error if the stored bpf_link does not match the incoming bpf_link. */
1780 if (link != *plink) {
1781 ret = -EBUSY;
1782 goto out;
1783 }
1784
1785 if (old) {
1786 ret = psock_replace_prog(pprog, prog, old);
1787 if (ret)
1788 goto out;
1789 } else {
1790 psock_set_prog(pprog, prog);
1791 }
1792
1793 bpf_prog_inc(prog);
1794 old_link_prog = xchg(&link->prog, prog);
1795 bpf_prog_put(old_link_prog);
1796
1797out:
1798 mutex_unlock(&sockmap_mutex);
1799 return ret;
1800}
1801
1802static u32 sock_map_link_get_map_id(const struct sockmap_link *sockmap_link)
1803{
1804 u32 map_id = 0;
1805
1806 mutex_lock(&sockmap_mutex);
1807 if (sockmap_link->map)
1808 map_id = sockmap_link->map->id;
1809 mutex_unlock(&sockmap_mutex);
1810 return map_id;
1811}
1812
1813static int sock_map_link_fill_info(const struct bpf_link *link,
1814 struct bpf_link_info *info)
1815{
1816 const struct sockmap_link *sockmap_link = container_of(link, struct sockmap_link, link);
1817 u32 map_id = sock_map_link_get_map_id(sockmap_link);
1818
1819 info->sockmap.map_id = map_id;
1820 info->sockmap.attach_type = sockmap_link->attach_type;
1821 return 0;
1822}
1823
1824static void sock_map_link_show_fdinfo(const struct bpf_link *link,
1825 struct seq_file *seq)
1826{
1827 const struct sockmap_link *sockmap_link = container_of(link, struct sockmap_link, link);
1828 u32 map_id = sock_map_link_get_map_id(sockmap_link);
1829
1830 seq_printf(seq, "map_id:\t%u\n", map_id);
1831 seq_printf(seq, "attach_type:\t%u\n", sockmap_link->attach_type);
1832}
1833
1834static const struct bpf_link_ops sock_map_link_ops = {
1835 .release = sock_map_link_release,
1836 .dealloc = sock_map_link_dealloc,
1837 .detach = sock_map_link_detach,
1838 .update_prog = sock_map_link_update_prog,
1839 .fill_link_info = sock_map_link_fill_info,
1840 .show_fdinfo = sock_map_link_show_fdinfo,
1841};
1842
1843int sock_map_link_create(const union bpf_attr *attr, struct bpf_prog *prog)
1844{
1845 struct bpf_link_primer link_primer;
1846 struct sockmap_link *sockmap_link;
1847 enum bpf_attach_type attach_type;
1848 struct bpf_map *map;
1849 int ret;
1850
1851 if (attr->link_create.flags)
1852 return -EINVAL;
1853
1854 map = bpf_map_get_with_uref(attr->link_create.target_fd);
1855 if (IS_ERR(map))
1856 return PTR_ERR(map);
1857 if (map->map_type != BPF_MAP_TYPE_SOCKMAP && map->map_type != BPF_MAP_TYPE_SOCKHASH) {
1858 ret = -EINVAL;
1859 goto out;
1860 }
1861
1862 sockmap_link = kzalloc(sizeof(*sockmap_link), GFP_USER);
1863 if (!sockmap_link) {
1864 ret = -ENOMEM;
1865 goto out;
1866 }
1867
1868 attach_type = attr->link_create.attach_type;
1869 bpf_link_init(&sockmap_link->link, BPF_LINK_TYPE_SOCKMAP, &sock_map_link_ops, prog);
1870 sockmap_link->map = map;
1871 sockmap_link->attach_type = attach_type;
1872
1873 ret = bpf_link_prime(&sockmap_link->link, &link_primer);
1874 if (ret) {
1875 kfree(sockmap_link);
1876 goto out;
1877 }
1878
1879 mutex_lock(&sockmap_mutex);
1880 ret = sock_map_prog_update(map, prog, NULL, &sockmap_link->link, attach_type);
1881 mutex_unlock(&sockmap_mutex);
1882 if (ret) {
1883 bpf_link_cleanup(&link_primer);
1884 goto out;
1885 }
1886
1887 /* Increase refcnt for the prog since when old prog is replaced with
1888 * psock_replace_prog() and psock_set_prog() its refcnt will be decreased.
1889 *
1890 * Actually, we do not need to increase refcnt for the prog since bpf_link
1891 * will hold a reference. But in order to have less complexity w.r.t.
1892 * replacing/setting prog, let us increase the refcnt to make things simpler.
1893 */
1894 bpf_prog_inc(prog);
1895
1896 return bpf_link_settle(&link_primer);
1897
1898out:
1899 bpf_map_put_with_uref(map);
1900 return ret;
1901}
1902
1903static int sock_map_iter_attach_target(struct bpf_prog *prog,
1904 union bpf_iter_link_info *linfo,
1905 struct bpf_iter_aux_info *aux)
1906{
1907 struct bpf_map *map;
1908 int err = -EINVAL;
1909
1910 if (!linfo->map.map_fd)
1911 return -EBADF;
1912
1913 map = bpf_map_get_with_uref(linfo->map.map_fd);
1914 if (IS_ERR(map))
1915 return PTR_ERR(map);
1916
1917 if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1918 map->map_type != BPF_MAP_TYPE_SOCKHASH)
1919 goto put_map;
1920
1921 if (prog->aux->max_rdonly_access > map->key_size) {
1922 err = -EACCES;
1923 goto put_map;
1924 }
1925
1926 aux->map = map;
1927 return 0;
1928
1929put_map:
1930 bpf_map_put_with_uref(map);
1931 return err;
1932}
1933
1934static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1935{
1936 bpf_map_put_with_uref(aux->map);
1937}
1938
1939static struct bpf_iter_reg sock_map_iter_reg = {
1940 .target = "sockmap",
1941 .attach_target = sock_map_iter_attach_target,
1942 .detach_target = sock_map_iter_detach_target,
1943 .show_fdinfo = bpf_iter_map_show_fdinfo,
1944 .fill_link_info = bpf_iter_map_fill_link_info,
1945 .ctx_arg_info_size = 2,
1946 .ctx_arg_info = {
1947 { offsetof(struct bpf_iter__sockmap, key),
1948 PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY },
1949 { offsetof(struct bpf_iter__sockmap, sk),
1950 PTR_TO_BTF_ID_OR_NULL },
1951 },
1952};
1953
1954static int __init bpf_sockmap_iter_init(void)
1955{
1956 sock_map_iter_reg.ctx_arg_info[1].btf_id =
1957 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1958 return bpf_iter_reg_target(&sock_map_iter_reg);
1959}
1960late_initcall(bpf_sockmap_iter_init);
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/filter.h>
6#include <linux/errno.h>
7#include <linux/file.h>
8#include <linux/net.h>
9#include <linux/workqueue.h>
10#include <linux/skmsg.h>
11#include <linux/list.h>
12#include <linux/jhash.h>
13#include <linux/sock_diag.h>
14#include <net/udp.h>
15
16struct bpf_stab {
17 struct bpf_map map;
18 struct sock **sks;
19 struct sk_psock_progs progs;
20 raw_spinlock_t lock;
21};
22
23#define SOCK_CREATE_FLAG_MASK \
24 (BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY)
25
26static struct bpf_map *sock_map_alloc(union bpf_attr *attr)
27{
28 struct bpf_stab *stab;
29 u64 cost;
30 int err;
31
32 if (!capable(CAP_NET_ADMIN))
33 return ERR_PTR(-EPERM);
34 if (attr->max_entries == 0 ||
35 attr->key_size != 4 ||
36 (attr->value_size != sizeof(u32) &&
37 attr->value_size != sizeof(u64)) ||
38 attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
39 return ERR_PTR(-EINVAL);
40
41 stab = kzalloc(sizeof(*stab), GFP_USER);
42 if (!stab)
43 return ERR_PTR(-ENOMEM);
44
45 bpf_map_init_from_attr(&stab->map, attr);
46 raw_spin_lock_init(&stab->lock);
47
48 /* Make sure page count doesn't overflow. */
49 cost = (u64) stab->map.max_entries * sizeof(struct sock *);
50 err = bpf_map_charge_init(&stab->map.memory, cost);
51 if (err)
52 goto free_stab;
53
54 stab->sks = bpf_map_area_alloc(stab->map.max_entries *
55 sizeof(struct sock *),
56 stab->map.numa_node);
57 if (stab->sks)
58 return &stab->map;
59 err = -ENOMEM;
60 bpf_map_charge_finish(&stab->map.memory);
61free_stab:
62 kfree(stab);
63 return ERR_PTR(err);
64}
65
66int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog)
67{
68 u32 ufd = attr->target_fd;
69 struct bpf_map *map;
70 struct fd f;
71 int ret;
72
73 if (attr->attach_flags || attr->replace_bpf_fd)
74 return -EINVAL;
75
76 f = fdget(ufd);
77 map = __bpf_map_get(f);
78 if (IS_ERR(map))
79 return PTR_ERR(map);
80 ret = sock_map_prog_update(map, prog, NULL, attr->attach_type);
81 fdput(f);
82 return ret;
83}
84
85int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
86{
87 u32 ufd = attr->target_fd;
88 struct bpf_prog *prog;
89 struct bpf_map *map;
90 struct fd f;
91 int ret;
92
93 if (attr->attach_flags || attr->replace_bpf_fd)
94 return -EINVAL;
95
96 f = fdget(ufd);
97 map = __bpf_map_get(f);
98 if (IS_ERR(map))
99 return PTR_ERR(map);
100
101 prog = bpf_prog_get(attr->attach_bpf_fd);
102 if (IS_ERR(prog)) {
103 ret = PTR_ERR(prog);
104 goto put_map;
105 }
106
107 if (prog->type != ptype) {
108 ret = -EINVAL;
109 goto put_prog;
110 }
111
112 ret = sock_map_prog_update(map, NULL, prog, attr->attach_type);
113put_prog:
114 bpf_prog_put(prog);
115put_map:
116 fdput(f);
117 return ret;
118}
119
120static void sock_map_sk_acquire(struct sock *sk)
121 __acquires(&sk->sk_lock.slock)
122{
123 lock_sock(sk);
124 preempt_disable();
125 rcu_read_lock();
126}
127
128static void sock_map_sk_release(struct sock *sk)
129 __releases(&sk->sk_lock.slock)
130{
131 rcu_read_unlock();
132 preempt_enable();
133 release_sock(sk);
134}
135
136static void sock_map_add_link(struct sk_psock *psock,
137 struct sk_psock_link *link,
138 struct bpf_map *map, void *link_raw)
139{
140 link->link_raw = link_raw;
141 link->map = map;
142 spin_lock_bh(&psock->link_lock);
143 list_add_tail(&link->list, &psock->link);
144 spin_unlock_bh(&psock->link_lock);
145}
146
147static void sock_map_del_link(struct sock *sk,
148 struct sk_psock *psock, void *link_raw)
149{
150 struct sk_psock_link *link, *tmp;
151 bool strp_stop = false;
152
153 spin_lock_bh(&psock->link_lock);
154 list_for_each_entry_safe(link, tmp, &psock->link, list) {
155 if (link->link_raw == link_raw) {
156 struct bpf_map *map = link->map;
157 struct bpf_stab *stab = container_of(map, struct bpf_stab,
158 map);
159 if (psock->parser.enabled && stab->progs.skb_parser)
160 strp_stop = true;
161 list_del(&link->list);
162 sk_psock_free_link(link);
163 }
164 }
165 spin_unlock_bh(&psock->link_lock);
166 if (strp_stop) {
167 write_lock_bh(&sk->sk_callback_lock);
168 sk_psock_stop_strp(sk, psock);
169 write_unlock_bh(&sk->sk_callback_lock);
170 }
171}
172
173static void sock_map_unref(struct sock *sk, void *link_raw)
174{
175 struct sk_psock *psock = sk_psock(sk);
176
177 if (likely(psock)) {
178 sock_map_del_link(sk, psock, link_raw);
179 sk_psock_put(sk, psock);
180 }
181}
182
183static int sock_map_init_proto(struct sock *sk, struct sk_psock *psock)
184{
185 struct proto *prot;
186
187 sock_owned_by_me(sk);
188
189 switch (sk->sk_type) {
190 case SOCK_STREAM:
191 prot = tcp_bpf_get_proto(sk, psock);
192 break;
193
194 case SOCK_DGRAM:
195 prot = udp_bpf_get_proto(sk, psock);
196 break;
197
198 default:
199 return -EINVAL;
200 }
201
202 if (IS_ERR(prot))
203 return PTR_ERR(prot);
204
205 sk_psock_update_proto(sk, psock, prot);
206 return 0;
207}
208
209static struct sk_psock *sock_map_psock_get_checked(struct sock *sk)
210{
211 struct sk_psock *psock;
212
213 rcu_read_lock();
214 psock = sk_psock(sk);
215 if (psock) {
216 if (sk->sk_prot->close != sock_map_close) {
217 psock = ERR_PTR(-EBUSY);
218 goto out;
219 }
220
221 if (!refcount_inc_not_zero(&psock->refcnt))
222 psock = ERR_PTR(-EBUSY);
223 }
224out:
225 rcu_read_unlock();
226 return psock;
227}
228
229static int sock_map_link(struct bpf_map *map, struct sk_psock_progs *progs,
230 struct sock *sk)
231{
232 struct bpf_prog *msg_parser, *skb_parser, *skb_verdict;
233 struct sk_psock *psock;
234 bool skb_progs;
235 int ret;
236
237 skb_verdict = READ_ONCE(progs->skb_verdict);
238 skb_parser = READ_ONCE(progs->skb_parser);
239 skb_progs = skb_parser && skb_verdict;
240 if (skb_progs) {
241 skb_verdict = bpf_prog_inc_not_zero(skb_verdict);
242 if (IS_ERR(skb_verdict))
243 return PTR_ERR(skb_verdict);
244 skb_parser = bpf_prog_inc_not_zero(skb_parser);
245 if (IS_ERR(skb_parser)) {
246 bpf_prog_put(skb_verdict);
247 return PTR_ERR(skb_parser);
248 }
249 }
250
251 msg_parser = READ_ONCE(progs->msg_parser);
252 if (msg_parser) {
253 msg_parser = bpf_prog_inc_not_zero(msg_parser);
254 if (IS_ERR(msg_parser)) {
255 ret = PTR_ERR(msg_parser);
256 goto out;
257 }
258 }
259
260 psock = sock_map_psock_get_checked(sk);
261 if (IS_ERR(psock)) {
262 ret = PTR_ERR(psock);
263 goto out_progs;
264 }
265
266 if (psock) {
267 if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
268 (skb_progs && READ_ONCE(psock->progs.skb_parser))) {
269 sk_psock_put(sk, psock);
270 ret = -EBUSY;
271 goto out_progs;
272 }
273 } else {
274 psock = sk_psock_init(sk, map->numa_node);
275 if (!psock) {
276 ret = -ENOMEM;
277 goto out_progs;
278 }
279 }
280
281 if (msg_parser)
282 psock_set_prog(&psock->progs.msg_parser, msg_parser);
283
284 ret = sock_map_init_proto(sk, psock);
285 if (ret < 0)
286 goto out_drop;
287
288 write_lock_bh(&sk->sk_callback_lock);
289 if (skb_progs && !psock->parser.enabled) {
290 ret = sk_psock_init_strp(sk, psock);
291 if (ret) {
292 write_unlock_bh(&sk->sk_callback_lock);
293 goto out_drop;
294 }
295 psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
296 psock_set_prog(&psock->progs.skb_parser, skb_parser);
297 sk_psock_start_strp(sk, psock);
298 }
299 write_unlock_bh(&sk->sk_callback_lock);
300 return 0;
301out_drop:
302 sk_psock_put(sk, psock);
303out_progs:
304 if (msg_parser)
305 bpf_prog_put(msg_parser);
306out:
307 if (skb_progs) {
308 bpf_prog_put(skb_verdict);
309 bpf_prog_put(skb_parser);
310 }
311 return ret;
312}
313
314static int sock_map_link_no_progs(struct bpf_map *map, struct sock *sk)
315{
316 struct sk_psock *psock;
317 int ret;
318
319 psock = sock_map_psock_get_checked(sk);
320 if (IS_ERR(psock))
321 return PTR_ERR(psock);
322
323 if (!psock) {
324 psock = sk_psock_init(sk, map->numa_node);
325 if (!psock)
326 return -ENOMEM;
327 }
328
329 ret = sock_map_init_proto(sk, psock);
330 if (ret < 0)
331 sk_psock_put(sk, psock);
332 return ret;
333}
334
335static void sock_map_free(struct bpf_map *map)
336{
337 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
338 int i;
339
340 /* After the sync no updates or deletes will be in-flight so it
341 * is safe to walk map and remove entries without risking a race
342 * in EEXIST update case.
343 */
344 synchronize_rcu();
345 for (i = 0; i < stab->map.max_entries; i++) {
346 struct sock **psk = &stab->sks[i];
347 struct sock *sk;
348
349 sk = xchg(psk, NULL);
350 if (sk) {
351 lock_sock(sk);
352 rcu_read_lock();
353 sock_map_unref(sk, psk);
354 rcu_read_unlock();
355 release_sock(sk);
356 }
357 }
358
359 /* wait for psock readers accessing its map link */
360 synchronize_rcu();
361
362 bpf_map_area_free(stab->sks);
363 kfree(stab);
364}
365
366static void sock_map_release_progs(struct bpf_map *map)
367{
368 psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
369}
370
371static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
372{
373 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
374
375 WARN_ON_ONCE(!rcu_read_lock_held());
376
377 if (unlikely(key >= map->max_entries))
378 return NULL;
379 return READ_ONCE(stab->sks[key]);
380}
381
382static void *sock_map_lookup(struct bpf_map *map, void *key)
383{
384 struct sock *sk;
385
386 sk = __sock_map_lookup_elem(map, *(u32 *)key);
387 if (!sk || !sk_fullsock(sk))
388 return NULL;
389 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
390 return NULL;
391 return sk;
392}
393
394static void *sock_map_lookup_sys(struct bpf_map *map, void *key)
395{
396 struct sock *sk;
397
398 if (map->value_size != sizeof(u64))
399 return ERR_PTR(-ENOSPC);
400
401 sk = __sock_map_lookup_elem(map, *(u32 *)key);
402 if (!sk)
403 return ERR_PTR(-ENOENT);
404
405 sock_gen_cookie(sk);
406 return &sk->sk_cookie;
407}
408
409static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
410 struct sock **psk)
411{
412 struct sock *sk;
413 int err = 0;
414
415 raw_spin_lock_bh(&stab->lock);
416 sk = *psk;
417 if (!sk_test || sk_test == sk)
418 sk = xchg(psk, NULL);
419
420 if (likely(sk))
421 sock_map_unref(sk, psk);
422 else
423 err = -EINVAL;
424
425 raw_spin_unlock_bh(&stab->lock);
426 return err;
427}
428
429static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
430 void *link_raw)
431{
432 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
433
434 __sock_map_delete(stab, sk, link_raw);
435}
436
437static int sock_map_delete_elem(struct bpf_map *map, void *key)
438{
439 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
440 u32 i = *(u32 *)key;
441 struct sock **psk;
442
443 if (unlikely(i >= map->max_entries))
444 return -EINVAL;
445
446 psk = &stab->sks[i];
447 return __sock_map_delete(stab, NULL, psk);
448}
449
450static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
451{
452 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
453 u32 i = key ? *(u32 *)key : U32_MAX;
454 u32 *key_next = next;
455
456 if (i == stab->map.max_entries - 1)
457 return -ENOENT;
458 if (i >= stab->map.max_entries)
459 *key_next = 0;
460 else
461 *key_next = i + 1;
462 return 0;
463}
464
465static bool sock_map_redirect_allowed(const struct sock *sk);
466
467static 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 if (inet_csk_has_ulp(sk))
482 return -EINVAL;
483
484 link = sk_psock_init_link();
485 if (!link)
486 return -ENOMEM;
487
488 /* Only sockets we can redirect into/from in BPF need to hold
489 * refs to parser/verdict progs and have their sk_data_ready
490 * and sk_write_space callbacks overridden.
491 */
492 if (sock_map_redirect_allowed(sk))
493 ret = sock_map_link(map, &stab->progs, sk);
494 else
495 ret = sock_map_link_no_progs(map, sk);
496 if (ret < 0)
497 goto out_free;
498
499 psock = sk_psock(sk);
500 WARN_ON_ONCE(!psock);
501
502 raw_spin_lock_bh(&stab->lock);
503 osk = stab->sks[idx];
504 if (osk && flags == BPF_NOEXIST) {
505 ret = -EEXIST;
506 goto out_unlock;
507 } else if (!osk && flags == BPF_EXIST) {
508 ret = -ENOENT;
509 goto out_unlock;
510 }
511
512 sock_map_add_link(psock, link, map, &stab->sks[idx]);
513 stab->sks[idx] = sk;
514 if (osk)
515 sock_map_unref(osk, &stab->sks[idx]);
516 raw_spin_unlock_bh(&stab->lock);
517 return 0;
518out_unlock:
519 raw_spin_unlock_bh(&stab->lock);
520 if (psock)
521 sk_psock_put(sk, psock);
522out_free:
523 sk_psock_free_link(link);
524 return ret;
525}
526
527static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
528{
529 return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
530 ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
531 ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
532}
533
534static bool sk_is_tcp(const struct sock *sk)
535{
536 return sk->sk_type == SOCK_STREAM &&
537 sk->sk_protocol == IPPROTO_TCP;
538}
539
540static bool sk_is_udp(const struct sock *sk)
541{
542 return sk->sk_type == SOCK_DGRAM &&
543 sk->sk_protocol == IPPROTO_UDP;
544}
545
546static bool sock_map_redirect_allowed(const struct sock *sk)
547{
548 return sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN;
549}
550
551static bool sock_map_sk_is_suitable(const struct sock *sk)
552{
553 return sk_is_tcp(sk) || sk_is_udp(sk);
554}
555
556static bool sock_map_sk_state_allowed(const struct sock *sk)
557{
558 if (sk_is_tcp(sk))
559 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
560 else if (sk_is_udp(sk))
561 return sk_hashed(sk);
562
563 return false;
564}
565
566static int sock_map_update_elem(struct bpf_map *map, void *key,
567 void *value, u64 flags)
568{
569 u32 idx = *(u32 *)key;
570 struct socket *sock;
571 struct sock *sk;
572 int ret;
573 u64 ufd;
574
575 if (map->value_size == sizeof(u64))
576 ufd = *(u64 *)value;
577 else
578 ufd = *(u32 *)value;
579 if (ufd > S32_MAX)
580 return -EINVAL;
581
582 sock = sockfd_lookup(ufd, &ret);
583 if (!sock)
584 return ret;
585 sk = sock->sk;
586 if (!sk) {
587 ret = -EINVAL;
588 goto out;
589 }
590 if (!sock_map_sk_is_suitable(sk)) {
591 ret = -EOPNOTSUPP;
592 goto out;
593 }
594
595 sock_map_sk_acquire(sk);
596 if (!sock_map_sk_state_allowed(sk))
597 ret = -EOPNOTSUPP;
598 else
599 ret = sock_map_update_common(map, idx, sk, flags);
600 sock_map_sk_release(sk);
601out:
602 fput(sock->file);
603 return ret;
604}
605
606BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
607 struct bpf_map *, map, void *, key, u64, flags)
608{
609 WARN_ON_ONCE(!rcu_read_lock_held());
610
611 if (likely(sock_map_sk_is_suitable(sops->sk) &&
612 sock_map_op_okay(sops)))
613 return sock_map_update_common(map, *(u32 *)key, sops->sk,
614 flags);
615 return -EOPNOTSUPP;
616}
617
618const struct bpf_func_proto bpf_sock_map_update_proto = {
619 .func = bpf_sock_map_update,
620 .gpl_only = false,
621 .pkt_access = true,
622 .ret_type = RET_INTEGER,
623 .arg1_type = ARG_PTR_TO_CTX,
624 .arg2_type = ARG_CONST_MAP_PTR,
625 .arg3_type = ARG_PTR_TO_MAP_KEY,
626 .arg4_type = ARG_ANYTHING,
627};
628
629BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
630 struct bpf_map *, map, u32, key, u64, flags)
631{
632 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
633 struct sock *sk;
634
635 if (unlikely(flags & ~(BPF_F_INGRESS)))
636 return SK_DROP;
637
638 sk = __sock_map_lookup_elem(map, key);
639 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
640 return SK_DROP;
641
642 tcb->bpf.flags = flags;
643 tcb->bpf.sk_redir = sk;
644 return SK_PASS;
645}
646
647const struct bpf_func_proto bpf_sk_redirect_map_proto = {
648 .func = bpf_sk_redirect_map,
649 .gpl_only = false,
650 .ret_type = RET_INTEGER,
651 .arg1_type = ARG_PTR_TO_CTX,
652 .arg2_type = ARG_CONST_MAP_PTR,
653 .arg3_type = ARG_ANYTHING,
654 .arg4_type = ARG_ANYTHING,
655};
656
657BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
658 struct bpf_map *, map, u32, key, u64, flags)
659{
660 struct sock *sk;
661
662 if (unlikely(flags & ~(BPF_F_INGRESS)))
663 return SK_DROP;
664
665 sk = __sock_map_lookup_elem(map, key);
666 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
667 return SK_DROP;
668
669 msg->flags = flags;
670 msg->sk_redir = sk;
671 return SK_PASS;
672}
673
674const struct bpf_func_proto bpf_msg_redirect_map_proto = {
675 .func = bpf_msg_redirect_map,
676 .gpl_only = false,
677 .ret_type = RET_INTEGER,
678 .arg1_type = ARG_PTR_TO_CTX,
679 .arg2_type = ARG_CONST_MAP_PTR,
680 .arg3_type = ARG_ANYTHING,
681 .arg4_type = ARG_ANYTHING,
682};
683
684static int sock_map_btf_id;
685const struct bpf_map_ops sock_map_ops = {
686 .map_alloc = sock_map_alloc,
687 .map_free = sock_map_free,
688 .map_get_next_key = sock_map_get_next_key,
689 .map_lookup_elem_sys_only = sock_map_lookup_sys,
690 .map_update_elem = sock_map_update_elem,
691 .map_delete_elem = sock_map_delete_elem,
692 .map_lookup_elem = sock_map_lookup,
693 .map_release_uref = sock_map_release_progs,
694 .map_check_btf = map_check_no_btf,
695 .map_btf_name = "bpf_stab",
696 .map_btf_id = &sock_map_btf_id,
697};
698
699struct bpf_shtab_elem {
700 struct rcu_head rcu;
701 u32 hash;
702 struct sock *sk;
703 struct hlist_node node;
704 u8 key[];
705};
706
707struct bpf_shtab_bucket {
708 struct hlist_head head;
709 raw_spinlock_t lock;
710};
711
712struct bpf_shtab {
713 struct bpf_map map;
714 struct bpf_shtab_bucket *buckets;
715 u32 buckets_num;
716 u32 elem_size;
717 struct sk_psock_progs progs;
718 atomic_t count;
719};
720
721static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
722{
723 return jhash(key, len, 0);
724}
725
726static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
727 u32 hash)
728{
729 return &htab->buckets[hash & (htab->buckets_num - 1)];
730}
731
732static struct bpf_shtab_elem *
733sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
734 u32 key_size)
735{
736 struct bpf_shtab_elem *elem;
737
738 hlist_for_each_entry_rcu(elem, head, node) {
739 if (elem->hash == hash &&
740 !memcmp(&elem->key, key, key_size))
741 return elem;
742 }
743
744 return NULL;
745}
746
747static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
748{
749 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
750 u32 key_size = map->key_size, hash;
751 struct bpf_shtab_bucket *bucket;
752 struct bpf_shtab_elem *elem;
753
754 WARN_ON_ONCE(!rcu_read_lock_held());
755
756 hash = sock_hash_bucket_hash(key, key_size);
757 bucket = sock_hash_select_bucket(htab, hash);
758 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
759
760 return elem ? elem->sk : NULL;
761}
762
763static void sock_hash_free_elem(struct bpf_shtab *htab,
764 struct bpf_shtab_elem *elem)
765{
766 atomic_dec(&htab->count);
767 kfree_rcu(elem, rcu);
768}
769
770static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
771 void *link_raw)
772{
773 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
774 struct bpf_shtab_elem *elem_probe, *elem = link_raw;
775 struct bpf_shtab_bucket *bucket;
776
777 WARN_ON_ONCE(!rcu_read_lock_held());
778 bucket = sock_hash_select_bucket(htab, elem->hash);
779
780 /* elem may be deleted in parallel from the map, but access here
781 * is okay since it's going away only after RCU grace period.
782 * However, we need to check whether it's still present.
783 */
784 raw_spin_lock_bh(&bucket->lock);
785 elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
786 elem->key, map->key_size);
787 if (elem_probe && elem_probe == elem) {
788 hlist_del_rcu(&elem->node);
789 sock_map_unref(elem->sk, elem);
790 sock_hash_free_elem(htab, elem);
791 }
792 raw_spin_unlock_bh(&bucket->lock);
793}
794
795static int sock_hash_delete_elem(struct bpf_map *map, void *key)
796{
797 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
798 u32 hash, key_size = map->key_size;
799 struct bpf_shtab_bucket *bucket;
800 struct bpf_shtab_elem *elem;
801 int ret = -ENOENT;
802
803 hash = sock_hash_bucket_hash(key, key_size);
804 bucket = sock_hash_select_bucket(htab, hash);
805
806 raw_spin_lock_bh(&bucket->lock);
807 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
808 if (elem) {
809 hlist_del_rcu(&elem->node);
810 sock_map_unref(elem->sk, elem);
811 sock_hash_free_elem(htab, elem);
812 ret = 0;
813 }
814 raw_spin_unlock_bh(&bucket->lock);
815 return ret;
816}
817
818static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
819 void *key, u32 key_size,
820 u32 hash, struct sock *sk,
821 struct bpf_shtab_elem *old)
822{
823 struct bpf_shtab_elem *new;
824
825 if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
826 if (!old) {
827 atomic_dec(&htab->count);
828 return ERR_PTR(-E2BIG);
829 }
830 }
831
832 new = kmalloc_node(htab->elem_size, GFP_ATOMIC | __GFP_NOWARN,
833 htab->map.numa_node);
834 if (!new) {
835 atomic_dec(&htab->count);
836 return ERR_PTR(-ENOMEM);
837 }
838 memcpy(new->key, key, key_size);
839 new->sk = sk;
840 new->hash = hash;
841 return new;
842}
843
844static int sock_hash_update_common(struct bpf_map *map, void *key,
845 struct sock *sk, u64 flags)
846{
847 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
848 u32 key_size = map->key_size, hash;
849 struct bpf_shtab_elem *elem, *elem_new;
850 struct bpf_shtab_bucket *bucket;
851 struct sk_psock_link *link;
852 struct sk_psock *psock;
853 int ret;
854
855 WARN_ON_ONCE(!rcu_read_lock_held());
856 if (unlikely(flags > BPF_EXIST))
857 return -EINVAL;
858 if (inet_csk_has_ulp(sk))
859 return -EINVAL;
860
861 link = sk_psock_init_link();
862 if (!link)
863 return -ENOMEM;
864
865 /* Only sockets we can redirect into/from in BPF need to hold
866 * refs to parser/verdict progs and have their sk_data_ready
867 * and sk_write_space callbacks overridden.
868 */
869 if (sock_map_redirect_allowed(sk))
870 ret = sock_map_link(map, &htab->progs, sk);
871 else
872 ret = sock_map_link_no_progs(map, sk);
873 if (ret < 0)
874 goto out_free;
875
876 psock = sk_psock(sk);
877 WARN_ON_ONCE(!psock);
878
879 hash = sock_hash_bucket_hash(key, key_size);
880 bucket = sock_hash_select_bucket(htab, hash);
881
882 raw_spin_lock_bh(&bucket->lock);
883 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
884 if (elem && flags == BPF_NOEXIST) {
885 ret = -EEXIST;
886 goto out_unlock;
887 } else if (!elem && flags == BPF_EXIST) {
888 ret = -ENOENT;
889 goto out_unlock;
890 }
891
892 elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
893 if (IS_ERR(elem_new)) {
894 ret = PTR_ERR(elem_new);
895 goto out_unlock;
896 }
897
898 sock_map_add_link(psock, link, map, elem_new);
899 /* Add new element to the head of the list, so that
900 * concurrent search will find it before old elem.
901 */
902 hlist_add_head_rcu(&elem_new->node, &bucket->head);
903 if (elem) {
904 hlist_del_rcu(&elem->node);
905 sock_map_unref(elem->sk, elem);
906 sock_hash_free_elem(htab, elem);
907 }
908 raw_spin_unlock_bh(&bucket->lock);
909 return 0;
910out_unlock:
911 raw_spin_unlock_bh(&bucket->lock);
912 sk_psock_put(sk, psock);
913out_free:
914 sk_psock_free_link(link);
915 return ret;
916}
917
918static int sock_hash_update_elem(struct bpf_map *map, void *key,
919 void *value, u64 flags)
920{
921 struct socket *sock;
922 struct sock *sk;
923 int ret;
924 u64 ufd;
925
926 if (map->value_size == sizeof(u64))
927 ufd = *(u64 *)value;
928 else
929 ufd = *(u32 *)value;
930 if (ufd > S32_MAX)
931 return -EINVAL;
932
933 sock = sockfd_lookup(ufd, &ret);
934 if (!sock)
935 return ret;
936 sk = sock->sk;
937 if (!sk) {
938 ret = -EINVAL;
939 goto out;
940 }
941 if (!sock_map_sk_is_suitable(sk)) {
942 ret = -EOPNOTSUPP;
943 goto out;
944 }
945
946 sock_map_sk_acquire(sk);
947 if (!sock_map_sk_state_allowed(sk))
948 ret = -EOPNOTSUPP;
949 else
950 ret = sock_hash_update_common(map, key, sk, flags);
951 sock_map_sk_release(sk);
952out:
953 fput(sock->file);
954 return ret;
955}
956
957static int sock_hash_get_next_key(struct bpf_map *map, void *key,
958 void *key_next)
959{
960 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
961 struct bpf_shtab_elem *elem, *elem_next;
962 u32 hash, key_size = map->key_size;
963 struct hlist_head *head;
964 int i = 0;
965
966 if (!key)
967 goto find_first_elem;
968 hash = sock_hash_bucket_hash(key, key_size);
969 head = &sock_hash_select_bucket(htab, hash)->head;
970 elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
971 if (!elem)
972 goto find_first_elem;
973
974 elem_next = hlist_entry_safe(rcu_dereference_raw(hlist_next_rcu(&elem->node)),
975 struct bpf_shtab_elem, node);
976 if (elem_next) {
977 memcpy(key_next, elem_next->key, key_size);
978 return 0;
979 }
980
981 i = hash & (htab->buckets_num - 1);
982 i++;
983find_first_elem:
984 for (; i < htab->buckets_num; i++) {
985 head = &sock_hash_select_bucket(htab, i)->head;
986 elem_next = hlist_entry_safe(rcu_dereference_raw(hlist_first_rcu(head)),
987 struct bpf_shtab_elem, node);
988 if (elem_next) {
989 memcpy(key_next, elem_next->key, key_size);
990 return 0;
991 }
992 }
993
994 return -ENOENT;
995}
996
997static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
998{
999 struct bpf_shtab *htab;
1000 int i, err;
1001 u64 cost;
1002
1003 if (!capable(CAP_NET_ADMIN))
1004 return ERR_PTR(-EPERM);
1005 if (attr->max_entries == 0 ||
1006 attr->key_size == 0 ||
1007 (attr->value_size != sizeof(u32) &&
1008 attr->value_size != sizeof(u64)) ||
1009 attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1010 return ERR_PTR(-EINVAL);
1011 if (attr->key_size > MAX_BPF_STACK)
1012 return ERR_PTR(-E2BIG);
1013
1014 htab = kzalloc(sizeof(*htab), GFP_USER);
1015 if (!htab)
1016 return ERR_PTR(-ENOMEM);
1017
1018 bpf_map_init_from_attr(&htab->map, attr);
1019
1020 htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1021 htab->elem_size = sizeof(struct bpf_shtab_elem) +
1022 round_up(htab->map.key_size, 8);
1023 if (htab->buckets_num == 0 ||
1024 htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1025 err = -EINVAL;
1026 goto free_htab;
1027 }
1028
1029 cost = (u64) htab->buckets_num * sizeof(struct bpf_shtab_bucket) +
1030 (u64) htab->elem_size * htab->map.max_entries;
1031 if (cost >= U32_MAX - PAGE_SIZE) {
1032 err = -EINVAL;
1033 goto free_htab;
1034 }
1035 err = bpf_map_charge_init(&htab->map.memory, cost);
1036 if (err)
1037 goto free_htab;
1038
1039 htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1040 sizeof(struct bpf_shtab_bucket),
1041 htab->map.numa_node);
1042 if (!htab->buckets) {
1043 bpf_map_charge_finish(&htab->map.memory);
1044 err = -ENOMEM;
1045 goto free_htab;
1046 }
1047
1048 for (i = 0; i < htab->buckets_num; i++) {
1049 INIT_HLIST_HEAD(&htab->buckets[i].head);
1050 raw_spin_lock_init(&htab->buckets[i].lock);
1051 }
1052
1053 return &htab->map;
1054free_htab:
1055 kfree(htab);
1056 return ERR_PTR(err);
1057}
1058
1059static void sock_hash_free(struct bpf_map *map)
1060{
1061 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1062 struct bpf_shtab_bucket *bucket;
1063 struct hlist_head unlink_list;
1064 struct bpf_shtab_elem *elem;
1065 struct hlist_node *node;
1066 int i;
1067
1068 /* After the sync no updates or deletes will be in-flight so it
1069 * is safe to walk map and remove entries without risking a race
1070 * in EEXIST update case.
1071 */
1072 synchronize_rcu();
1073 for (i = 0; i < htab->buckets_num; i++) {
1074 bucket = sock_hash_select_bucket(htab, i);
1075
1076 /* We are racing with sock_hash_delete_from_link to
1077 * enter the spin-lock critical section. Every socket on
1078 * the list is still linked to sockhash. Since link
1079 * exists, psock exists and holds a ref to socket. That
1080 * lets us to grab a socket ref too.
1081 */
1082 raw_spin_lock_bh(&bucket->lock);
1083 hlist_for_each_entry(elem, &bucket->head, node)
1084 sock_hold(elem->sk);
1085 hlist_move_list(&bucket->head, &unlink_list);
1086 raw_spin_unlock_bh(&bucket->lock);
1087
1088 /* Process removed entries out of atomic context to
1089 * block for socket lock before deleting the psock's
1090 * link to sockhash.
1091 */
1092 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1093 hlist_del(&elem->node);
1094 lock_sock(elem->sk);
1095 rcu_read_lock();
1096 sock_map_unref(elem->sk, elem);
1097 rcu_read_unlock();
1098 release_sock(elem->sk);
1099 sock_put(elem->sk);
1100 sock_hash_free_elem(htab, elem);
1101 }
1102 }
1103
1104 /* wait for psock readers accessing its map link */
1105 synchronize_rcu();
1106
1107 bpf_map_area_free(htab->buckets);
1108 kfree(htab);
1109}
1110
1111static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1112{
1113 struct sock *sk;
1114
1115 if (map->value_size != sizeof(u64))
1116 return ERR_PTR(-ENOSPC);
1117
1118 sk = __sock_hash_lookup_elem(map, key);
1119 if (!sk)
1120 return ERR_PTR(-ENOENT);
1121
1122 sock_gen_cookie(sk);
1123 return &sk->sk_cookie;
1124}
1125
1126static void *sock_hash_lookup(struct bpf_map *map, void *key)
1127{
1128 struct sock *sk;
1129
1130 sk = __sock_hash_lookup_elem(map, key);
1131 if (!sk || !sk_fullsock(sk))
1132 return NULL;
1133 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1134 return NULL;
1135 return sk;
1136}
1137
1138static void sock_hash_release_progs(struct bpf_map *map)
1139{
1140 psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1141}
1142
1143BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1144 struct bpf_map *, map, void *, key, u64, flags)
1145{
1146 WARN_ON_ONCE(!rcu_read_lock_held());
1147
1148 if (likely(sock_map_sk_is_suitable(sops->sk) &&
1149 sock_map_op_okay(sops)))
1150 return sock_hash_update_common(map, key, sops->sk, flags);
1151 return -EOPNOTSUPP;
1152}
1153
1154const struct bpf_func_proto bpf_sock_hash_update_proto = {
1155 .func = bpf_sock_hash_update,
1156 .gpl_only = false,
1157 .pkt_access = true,
1158 .ret_type = RET_INTEGER,
1159 .arg1_type = ARG_PTR_TO_CTX,
1160 .arg2_type = ARG_CONST_MAP_PTR,
1161 .arg3_type = ARG_PTR_TO_MAP_KEY,
1162 .arg4_type = ARG_ANYTHING,
1163};
1164
1165BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1166 struct bpf_map *, map, void *, key, u64, flags)
1167{
1168 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
1169 struct sock *sk;
1170
1171 if (unlikely(flags & ~(BPF_F_INGRESS)))
1172 return SK_DROP;
1173
1174 sk = __sock_hash_lookup_elem(map, key);
1175 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1176 return SK_DROP;
1177
1178 tcb->bpf.flags = flags;
1179 tcb->bpf.sk_redir = sk;
1180 return SK_PASS;
1181}
1182
1183const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1184 .func = bpf_sk_redirect_hash,
1185 .gpl_only = false,
1186 .ret_type = RET_INTEGER,
1187 .arg1_type = ARG_PTR_TO_CTX,
1188 .arg2_type = ARG_CONST_MAP_PTR,
1189 .arg3_type = ARG_PTR_TO_MAP_KEY,
1190 .arg4_type = ARG_ANYTHING,
1191};
1192
1193BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1194 struct bpf_map *, map, void *, key, u64, flags)
1195{
1196 struct sock *sk;
1197
1198 if (unlikely(flags & ~(BPF_F_INGRESS)))
1199 return SK_DROP;
1200
1201 sk = __sock_hash_lookup_elem(map, key);
1202 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1203 return SK_DROP;
1204
1205 msg->flags = flags;
1206 msg->sk_redir = sk;
1207 return SK_PASS;
1208}
1209
1210const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1211 .func = bpf_msg_redirect_hash,
1212 .gpl_only = false,
1213 .ret_type = RET_INTEGER,
1214 .arg1_type = ARG_PTR_TO_CTX,
1215 .arg2_type = ARG_CONST_MAP_PTR,
1216 .arg3_type = ARG_PTR_TO_MAP_KEY,
1217 .arg4_type = ARG_ANYTHING,
1218};
1219
1220static int sock_hash_map_btf_id;
1221const struct bpf_map_ops sock_hash_ops = {
1222 .map_alloc = sock_hash_alloc,
1223 .map_free = sock_hash_free,
1224 .map_get_next_key = sock_hash_get_next_key,
1225 .map_update_elem = sock_hash_update_elem,
1226 .map_delete_elem = sock_hash_delete_elem,
1227 .map_lookup_elem = sock_hash_lookup,
1228 .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1229 .map_release_uref = sock_hash_release_progs,
1230 .map_check_btf = map_check_no_btf,
1231 .map_btf_name = "bpf_shtab",
1232 .map_btf_id = &sock_hash_map_btf_id,
1233};
1234
1235static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1236{
1237 switch (map->map_type) {
1238 case BPF_MAP_TYPE_SOCKMAP:
1239 return &container_of(map, struct bpf_stab, map)->progs;
1240 case BPF_MAP_TYPE_SOCKHASH:
1241 return &container_of(map, struct bpf_shtab, map)->progs;
1242 default:
1243 break;
1244 }
1245
1246 return NULL;
1247}
1248
1249int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1250 struct bpf_prog *old, u32 which)
1251{
1252 struct sk_psock_progs *progs = sock_map_progs(map);
1253 struct bpf_prog **pprog;
1254
1255 if (!progs)
1256 return -EOPNOTSUPP;
1257
1258 switch (which) {
1259 case BPF_SK_MSG_VERDICT:
1260 pprog = &progs->msg_parser;
1261 break;
1262 case BPF_SK_SKB_STREAM_PARSER:
1263 pprog = &progs->skb_parser;
1264 break;
1265 case BPF_SK_SKB_STREAM_VERDICT:
1266 pprog = &progs->skb_verdict;
1267 break;
1268 default:
1269 return -EOPNOTSUPP;
1270 }
1271
1272 if (old)
1273 return psock_replace_prog(pprog, prog, old);
1274
1275 psock_set_prog(pprog, prog);
1276 return 0;
1277}
1278
1279static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1280{
1281 switch (link->map->map_type) {
1282 case BPF_MAP_TYPE_SOCKMAP:
1283 return sock_map_delete_from_link(link->map, sk,
1284 link->link_raw);
1285 case BPF_MAP_TYPE_SOCKHASH:
1286 return sock_hash_delete_from_link(link->map, sk,
1287 link->link_raw);
1288 default:
1289 break;
1290 }
1291}
1292
1293static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1294{
1295 struct sk_psock_link *link;
1296
1297 while ((link = sk_psock_link_pop(psock))) {
1298 sock_map_unlink(sk, link);
1299 sk_psock_free_link(link);
1300 }
1301}
1302
1303void sock_map_unhash(struct sock *sk)
1304{
1305 void (*saved_unhash)(struct sock *sk);
1306 struct sk_psock *psock;
1307
1308 rcu_read_lock();
1309 psock = sk_psock(sk);
1310 if (unlikely(!psock)) {
1311 rcu_read_unlock();
1312 if (sk->sk_prot->unhash)
1313 sk->sk_prot->unhash(sk);
1314 return;
1315 }
1316
1317 saved_unhash = psock->saved_unhash;
1318 sock_map_remove_links(sk, psock);
1319 rcu_read_unlock();
1320 saved_unhash(sk);
1321}
1322
1323void sock_map_close(struct sock *sk, long timeout)
1324{
1325 void (*saved_close)(struct sock *sk, long timeout);
1326 struct sk_psock *psock;
1327
1328 lock_sock(sk);
1329 rcu_read_lock();
1330 psock = sk_psock(sk);
1331 if (unlikely(!psock)) {
1332 rcu_read_unlock();
1333 release_sock(sk);
1334 return sk->sk_prot->close(sk, timeout);
1335 }
1336
1337 saved_close = psock->saved_close;
1338 sock_map_remove_links(sk, psock);
1339 rcu_read_unlock();
1340 release_sock(sk);
1341 saved_close(sk, timeout);
1342}