]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - net/core/sock_map.c
projects / mirror_ubuntu-jammy-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
UBUNTU: Ubuntu-5.15.0-39.42
[mirror_ubuntu-jammy-kernel.git] / net / core / sock_map.c
CommitLineData
604326b4
DB		 1// SPDX-License-Identifier: GPL-2.0
2/* Copyright (c) 2017 - 2018 Covalent IO, Inc. http://covalent.io */
3
4#include <linux/bpf.h>
03653515 5#include <linux/btf_ids.h>
604326b4
DB		 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>
c1cdf65d 14#include <linux/sock_diag.h>
7b98cd42 15#include <net/udp.h>
604326b4
DB		 16
17struct 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
4675e234
CW		 27static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
28 struct bpf_prog *old, u32 which);
2004fdbd 29static struct sk_psock_progs *sock_map_progs(struct bpf_map *map);
4675e234 30
604326b4
DB		 31static struct bpf_map *sock_map_alloc(union bpf_attr *attr)
32{
33 struct bpf_stab *stab;
604326b4
DB		 34
35 if (!capable(CAP_NET_ADMIN))
36 return ERR_PTR(-EPERM);
37 if (attr->max_entries == 0 ||
38 attr->key_size != 4 ||
c1cdf65d
JS		 39 (attr->value_size != sizeof(u32) &&
40 attr->value_size != sizeof(u64)) ||
604326b4
DB		 41 attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
42 return ERR_PTR(-EINVAL);
43
7846dd9f 44 stab = kzalloc(sizeof(*stab), GFP_USER | __GFP_ACCOUNT);
604326b4
DB		 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
7dd5d437 51 stab->sks = bpf_map_area_alloc((u64) stab->map.max_entries *
604326b4
DB		 52 sizeof(struct sock *),
53 stab->map.numa_node);
0d2c4f96
RG		 54 if (!stab->sks) {
55 kfree(stab);
56 return ERR_PTR(-ENOMEM);
57 }
58
59 return &stab->map;
604326b4
DB		 60}
61
62int 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
9b2b0971
LB		 69 if (attr->attach_flags || attr->replace_bpf_fd)
70 return -EINVAL;
71
604326b4
DB		 72 f = fdget(ufd);
73 map = __bpf_map_get(f);
74 if (IS_ERR(map))
75 return PTR_ERR(map);
bb0de313
LB		 76 ret = sock_map_prog_update(map, prog, NULL, attr->attach_type);
77 fdput(f);
78 return ret;
79}
80
81int 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);
109put_prog:
110 bpf_prog_put(prog);
111put_map:
604326b4
DB		 112 fdput(f);
113 return ret;
114}
115
116static void sock_map_sk_acquire(struct sock *sk)
117 __acquires(&sk->sk_lock.slock)
118{
119 lock_sock(sk);
120 preempt_disable();
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 preempt_enable();
129 release_sock(sk);
130}
131
132static void sock_map_add_link(struct sk_psock *psock,
133 struct sk_psock_link *link,
134 struct bpf_map *map, void *link_raw)
135{
136 link->link_raw = link_raw;
137 link->map = map;
138 spin_lock_bh(&psock->link_lock);
139 list_add_tail(&link->list, &psock->link);
140 spin_unlock_bh(&psock->link_lock);
141}
142
143static void sock_map_del_link(struct sock *sk,
144 struct sk_psock *psock, void *link_raw)
145{
ef565928 146 bool strp_stop = false, verdict_stop = false;
604326b4 147 struct sk_psock_link *link, *tmp;
604326b4
DB		 148
149 spin_lock_bh(&psock->link_lock);
150 list_for_each_entry_safe(link, tmp, &psock->link, list) {
151 if (link->link_raw == link_raw) {
152 struct bpf_map *map = link->map;
153 struct bpf_stab *stab = container_of(map, struct bpf_stab,
154 map);
ae8b8332 155 if (psock->saved_data_ready && stab->progs.stream_parser)
604326b4 156 strp_stop = true;
ae8b8332 157 if (psock->saved_data_ready && stab->progs.stream_verdict)
ef565928 158 verdict_stop = true;
a7ba4558
CW		 159 if (psock->saved_data_ready && stab->progs.skb_verdict)
160 verdict_stop = true;
604326b4
DB		 161 list_del(&link->list);
162 sk_psock_free_link(link);
163 }
164 }
165 spin_unlock_bh(&psock->link_lock);
ef565928 166 if (strp_stop || verdict_stop) {
604326b4 167 write_lock_bh(&sk->sk_callback_lock);
ef565928
JF		 168 if (strp_stop)
169 sk_psock_stop_strp(sk, psock);
bfec6d03 170 if (verdict_stop)
ef565928 171 sk_psock_stop_verdict(sk, psock);
bfec6d03
JF		 172
173 if (psock->psock_update_sk_prot)
174 psock->psock_update_sk_prot(sk, psock, false);
604326b4
DB		 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
cb21802b 189static int sock_map_init_proto(struct sock *sk, struct sk_psock *psock)
f747632b 190{
8a59f9d1 191 if (!sk->sk_prot->psock_update_sk_prot)
7b98cd42 192 return -EINVAL;
8a59f9d1 193 psock->psock_update_sk_prot = sk->sk_prot->psock_update_sk_prot;
51e0158a 194 return sk->sk_prot->psock_update_sk_prot(sk, psock, false);
f747632b
LB		 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) {
7b98cd42 204 if (sk->sk_prot->close != sock_map_close) {
f747632b
LB		 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
2004fdbd 217static int sock_map_link(struct bpf_map *map, struct sock *sk)
604326b4 218{
2004fdbd 219 struct sk_psock_progs *progs = sock_map_progs(map);
b0170552
CW		 220 struct bpf_prog *stream_verdict = NULL;
221 struct bpf_prog *stream_parser = NULL;
a7ba4558 222 struct bpf_prog *skb_verdict = NULL;
b0170552 223 struct bpf_prog *msg_parser = NULL;
604326b4
DB		 224 struct sk_psock *psock;
225 int ret;
226
ae8b8332
CW		 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);
743df8b7 232 }
83c11c17 233
ae8b8332
CW		 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;
604326b4
DB		 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);
ae8b8332 248 goto out_put_stream_parser;
604326b4
DB		 249 }
250 }
251
a7ba4558
CW		 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
f747632b 261 psock = sock_map_psock_get_checked(sk);
5032d079
JF		 262 if (IS_ERR(psock)) {
263 ret = PTR_ERR(psock);
264 goto out_progs;
265 }
266
604326b4 267 if (psock) {
604326b4 268 if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
ae8b8332 269 (stream_parser && READ_ONCE(psock->progs.stream_parser)) ||
a7ba4558
CW		 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)) ||
ae8b8332 273 (stream_verdict && READ_ONCE(psock->progs.stream_verdict))) {
604326b4
DB		 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);
7b219da4
LB		 280 if (IS_ERR(psock)) {
281 ret = PTR_ERR(psock);
604326b4
DB		 282 goto out_progs;
283 }
604326b4
DB		 284 }
285
286 if (msg_parser)
287 psock_set_prog(&psock->progs.msg_parser, msg_parser);
1a04edc7
JF		 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);
d19da360 294
08aaba7e
JF		 295 /* msg_* and stream_* programs references tracked in psock after this
296 * point. Reference dec and cleanup will occur through psock destructor
297 */
cb21802b 298 ret = sock_map_init_proto(sk, psock);
08aaba7e
JF		 299 if (ret < 0) {
300 sk_psock_put(sk, psock);
301 goto out;
302 }
604326b4
DB		 303
304 write_lock_bh(&sk->sk_callback_lock);
ae8b8332 305 if (stream_parser && stream_verdict && !psock->saved_data_ready) {
604326b4 306 ret = sk_psock_init_strp(sk, psock);
08aaba7e
JF		 307 if (ret) {
308 write_unlock_bh(&sk->sk_callback_lock);
309 sk_psock_put(sk, psock);
310 goto out;
311 }
604326b4 312 sk_psock_start_strp(sk, psock);
ae8b8332 313 } else if (!stream_parser && stream_verdict && !psock->saved_data_ready) {
ef565928 314 sk_psock_start_verdict(sk,psock);
a7ba4558 315 } else if (!stream_verdict && skb_verdict && !psock->saved_data_ready) {
a7ba4558 316 sk_psock_start_verdict(sk, psock);
604326b4
DB		 317 }
318 write_unlock_bh(&sk->sk_callback_lock);
319 return 0;
604326b4 320out_progs:
a7ba4558
CW		 321 if (skb_verdict)
322 bpf_prog_put(skb_verdict);
323out_put_msg_parser:
604326b4
DB		 324 if (msg_parser)
325 bpf_prog_put(msg_parser);
ae8b8332
CW		 326out_put_stream_parser:
327 if (stream_parser)
328 bpf_prog_put(stream_parser);
329out_put_stream_verdict:
330 if (stream_verdict)
331 bpf_prog_put(stream_verdict);
08aaba7e 332out:
604326b4
DB		 333 return ret;
334}
335
336static void sock_map_free(struct bpf_map *map)
337{
338 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
339 int i;
340
90db6d77
JF		 341 /* After the sync no updates or deletes will be in-flight so it
342 * is safe to walk map and remove entries without risking a race
343 * in EEXIST update case.
344 */
604326b4 345 synchronize_rcu();
604326b4
DB		 346 for (i = 0; i < stab->map.max_entries; i++) {
347 struct sock **psk = &stab->sks[i];
348 struct sock *sk;
349
350 sk = xchg(psk, NULL);
7e81a353
JF		 351 if (sk) {
352 lock_sock(sk);
db6a5018 353 rcu_read_lock();
604326b4 354 sock_map_unref(sk, psk);
db6a5018 355 rcu_read_unlock();
7e81a353
JF		 356 release_sock(sk);
357 }
604326b4 358 }
604326b4 359
0b2dc839 360 /* wait for psock readers accessing its map link */
2bb90e5c
JF		 361 synchronize_rcu();
362
604326b4
DB		 363 bpf_map_area_free(stab->sks);
364 kfree(stab);
365}
366
367static void sock_map_release_progs(struct bpf_map *map)
368{
369 psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
370}
371
372static 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
383static void *sock_map_lookup(struct bpf_map *map, void *key)
384{
64d85290
JS		 385 struct sock *sk;
386
387 sk = __sock_map_lookup_elem(map, *(u32 *)key);
654785a1 388 if (!sk)
64d85290
JS		 389 return NULL;
390 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
391 return NULL;
392 return sk;
604326b4
DB		 393}
394
c1cdf65d
JS		 395static 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
92acdc58 406 __sock_gen_cookie(sk);
c1cdf65d
JS		 407 return &sk->sk_cookie;
408}
409
604326b4
DB		 410static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
411 struct sock **psk)
412{
413 struct sock *sk;
45a4521d 414 int err = 0;
604326b4
DB		 415
416 raw_spin_lock_bh(&stab->lock);
417 sk = *psk;
418 if (!sk_test || sk_test == sk)
45a4521d
JF		 419 sk = xchg(psk, NULL);
420
421 if (likely(sk))
422 sock_map_unref(sk, psk);
423 else
424 err = -EINVAL;
425
604326b4 426 raw_spin_unlock_bh(&stab->lock);
45a4521d 427 return err;
604326b4
DB		 428}
429
430static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
431 void *link_raw)
432{
433 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
434
435 __sock_map_delete(stab, sk, link_raw);
436}
437
438static int sock_map_delete_elem(struct bpf_map *map, void *key)
439{
440 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
441 u32 i = *(u32 *)key;
442 struct sock **psk;
443
444 if (unlikely(i >= map->max_entries))
445 return -EINVAL;
446
447 psk = &stab->sks[i];
448 return __sock_map_delete(stab, NULL, psk);
449}
450
451static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
452{
453 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
454 u32 i = key ? *(u32 *)key : U32_MAX;
455 u32 *key_next = next;
456
457 if (i == stab->map.max_entries - 1)
458 return -ENOENT;
459 if (i >= stab->map.max_entries)
460 *key_next = 0;
461 else
462 *key_next = i + 1;
463 return 0;
464}
465
466static int sock_map_update_common(struct bpf_map *map, u32 idx,
467 struct sock *sk, u64 flags)
468{
469 struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
470 struct sk_psock_link *link;
471 struct sk_psock *psock;
472 struct sock *osk;
473 int ret;
474
475 WARN_ON_ONCE(!rcu_read_lock_held());
476 if (unlikely(flags > BPF_EXIST))
477 return -EINVAL;
478 if (unlikely(idx >= map->max_entries))
479 return -E2BIG;
480
481 link = sk_psock_init_link();
482 if (!link)
483 return -ENOMEM;
484
b0170552 485 ret = sock_map_link(map, sk);
604326b4
DB		 486 if (ret < 0)
487 goto out_free;
488
489 psock = sk_psock(sk);
490 WARN_ON_ONCE(!psock);
491
492 raw_spin_lock_bh(&stab->lock);
493 osk = stab->sks[idx];
494 if (osk && flags == BPF_NOEXIST) {
495 ret = -EEXIST;
496 goto out_unlock;
497 } else if (!osk && flags == BPF_EXIST) {
498 ret = -ENOENT;
499 goto out_unlock;
500 }
501
502 sock_map_add_link(psock, link, map, &stab->sks[idx]);
503 stab->sks[idx] = sk;
504 if (osk)
505 sock_map_unref(osk, &stab->sks[idx]);
506 raw_spin_unlock_bh(&stab->lock);
507 return 0;
508out_unlock:
509 raw_spin_unlock_bh(&stab->lock);
510 if (psock)
511 sk_psock_put(sk, psock);
512out_free:
513 sk_psock_free_link(link);
514 return ret;
515}
516
517static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
518{
519 return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
8ca30379
JS		 520 ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
521 ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
604326b4
DB		 522}
523
7b98cd42 524static bool sk_is_tcp(const struct sock *sk)
604326b4
DB		 525{
526 return sk->sk_type == SOCK_STREAM &&
527 sk->sk_protocol == IPPROTO_TCP;
528}
529
f6fede85
LB		 530static bool sock_map_redirect_allowed(const struct sock *sk)
531{
122e6c79
CW		 532 if (sk_is_tcp(sk))
533 return sk->sk_state != TCP_LISTEN;
534 else
535 return sk->sk_state == TCP_ESTABLISHED;
f6fede85
LB		 536}
537
7b98cd42
LB		 538static bool sock_map_sk_is_suitable(const struct sock *sk)
539{
8a59f9d1 540 return !!sk->sk_prot->psock_update_sk_prot;
7b98cd42
LB		 541}
542
8ca30379
JS		 543static bool sock_map_sk_state_allowed(const struct sock *sk)
544{
7b98cd42
LB		 545 if (sk_is_tcp(sk))
546 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
0c48eefa 547 return true;
8ca30379
JS		 548}
549
38e12f90
LB		 550static int sock_hash_update_common(struct bpf_map *map, void *key,
551 struct sock *sk, u64 flags);
552
13b79d3f
LB		 553int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
554 u64 flags)
604326b4 555{
604326b4
DB		 556 struct socket *sock;
557 struct sock *sk;
558 int ret;
c1cdf65d
JS		 559 u64 ufd;
560
561 if (map->value_size == sizeof(u64))
562 ufd = *(u64 *)value;
563 else
564 ufd = *(u32 *)value;
565 if (ufd > S32_MAX)
566 return -EINVAL;
604326b4
DB		 567
568 sock = sockfd_lookup(ufd, &ret);
569 if (!sock)
570 return ret;
571 sk = sock->sk;
572 if (!sk) {
573 ret = -EINVAL;
574 goto out;
575 }
85b8ac01 576 if (!sock_map_sk_is_suitable(sk)) {
604326b4
DB		 577 ret = -EOPNOTSUPP;
578 goto out;
579 }
580
581 sock_map_sk_acquire(sk);
8ca30379 582 if (!sock_map_sk_state_allowed(sk))
85b8ac01 583 ret = -EOPNOTSUPP;
38e12f90
LB		 584 else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
585 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
85b8ac01 586 else
38e12f90 587 ret = sock_hash_update_common(map, key, sk, flags);
604326b4
DB		 588 sock_map_sk_release(sk);
589out:
ec24e11e 590 sockfd_put(sock);
604326b4
DB		 591 return ret;
592}
593
0126240f
LB		 594static int sock_map_update_elem(struct bpf_map *map, void *key,
595 void *value, u64 flags)
596{
597 struct sock *sk = (struct sock *)value;
598 int ret;
599
6550f2dd
LB		 600 if (unlikely(!sk || !sk_fullsock(sk)))
601 return -EINVAL;
602
0126240f
LB		 603 if (!sock_map_sk_is_suitable(sk))
604 return -EOPNOTSUPP;
605
606 local_bh_disable();
607 bh_lock_sock(sk);
608 if (!sock_map_sk_state_allowed(sk))
609 ret = -EOPNOTSUPP;
610 else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
611 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
612 else
613 ret = sock_hash_update_common(map, key, sk, flags);
614 bh_unlock_sock(sk);
615 local_bh_enable();
616 return ret;
617}
618
604326b4
DB		 619BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
620 struct bpf_map *, map, void *, key, u64, flags)
621{
622 WARN_ON_ONCE(!rcu_read_lock_held());
623
624 if (likely(sock_map_sk_is_suitable(sops->sk) &&
625 sock_map_op_okay(sops)))
626 return sock_map_update_common(map, *(u32 *)key, sops->sk,
627 flags);
628 return -EOPNOTSUPP;
629}
630
631const struct bpf_func_proto bpf_sock_map_update_proto = {
632 .func = bpf_sock_map_update,
633 .gpl_only = false,
634 .pkt_access = true,
635 .ret_type = RET_INTEGER,
636 .arg1_type = ARG_PTR_TO_CTX,
637 .arg2_type = ARG_CONST_MAP_PTR,
638 .arg3_type = ARG_PTR_TO_MAP_KEY,
639 .arg4_type = ARG_ANYTHING,
640};
641
642BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
643 struct bpf_map *, map, u32, key, u64, flags)
644{
8ca30379 645 struct sock *sk;
604326b4
DB		 646
647 if (unlikely(flags & ~(BPF_F_INGRESS)))
648 return SK_DROP;
8ca30379
JS		 649
650 sk = __sock_map_lookup_elem(map, key);
651 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
604326b4 652 return SK_DROP;
8ca30379 653
e3526bb9 654 skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
604326b4
DB		 655 return SK_PASS;
656}
657
658const struct bpf_func_proto bpf_sk_redirect_map_proto = {
659 .func = bpf_sk_redirect_map,
660 .gpl_only = false,
661 .ret_type = RET_INTEGER,
662 .arg1_type = ARG_PTR_TO_CTX,
663 .arg2_type = ARG_CONST_MAP_PTR,
664 .arg3_type = ARG_ANYTHING,
665 .arg4_type = ARG_ANYTHING,
666};
667
668BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
669 struct bpf_map *, map, u32, key, u64, flags)
670{
8ca30379
JS		 671 struct sock *sk;
672
604326b4
DB		 673 if (unlikely(flags & ~(BPF_F_INGRESS)))
674 return SK_DROP;
8ca30379
JS		 675
676 sk = __sock_map_lookup_elem(map, key);
677 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
604326b4 678 return SK_DROP;
8ca30379
JS		 679
680 msg->flags = flags;
681 msg->sk_redir = sk;
604326b4
DB		 682 return SK_PASS;
683}
684
685const struct bpf_func_proto bpf_msg_redirect_map_proto = {
686 .func = bpf_msg_redirect_map,
687 .gpl_only = false,
688 .ret_type = RET_INTEGER,
689 .arg1_type = ARG_PTR_TO_CTX,
690 .arg2_type = ARG_CONST_MAP_PTR,
691 .arg3_type = ARG_ANYTHING,
692 .arg4_type = ARG_ANYTHING,
693};
694
03653515
LB		 695struct sock_map_seq_info {
696 struct bpf_map *map;
697 struct sock *sk;
698 u32 index;
699};
700
701struct bpf_iter__sockmap {
702 __bpf_md_ptr(struct bpf_iter_meta *, meta);
703 __bpf_md_ptr(struct bpf_map *, map);
704 __bpf_md_ptr(void *, key);
705 __bpf_md_ptr(struct sock *, sk);
706};
707
708DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
709 struct bpf_map *map, void *key,
710 struct sock *sk)
711
712static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
713{
714 if (unlikely(info->index >= info->map->max_entries))
715 return NULL;
716
717 info->sk = __sock_map_lookup_elem(info->map, info->index);
718
719 /* can't return sk directly, since that might be NULL */
720 return info;
721}
722
723static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
f58423ae 724 __acquires(rcu)
03653515
LB		 725{
726 struct sock_map_seq_info *info = seq->private;
727
728 if (*pos == 0)
729 ++*pos;
730
731 /* pairs with sock_map_seq_stop */
732 rcu_read_lock();
733 return sock_map_seq_lookup_elem(info);
734}
735
736static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
f58423ae 737 __must_hold(rcu)
03653515
LB		 738{
739 struct sock_map_seq_info *info = seq->private;
740
741 ++*pos;
742 ++info->index;
743
744 return sock_map_seq_lookup_elem(info);
745}
746
747static int sock_map_seq_show(struct seq_file *seq, void *v)
f58423ae 748 __must_hold(rcu)
03653515
LB		 749{
750 struct sock_map_seq_info *info = seq->private;
751 struct bpf_iter__sockmap ctx = {};
752 struct bpf_iter_meta meta;
753 struct bpf_prog *prog;
754
755 meta.seq = seq;
756 prog = bpf_iter_get_info(&meta, !v);
757 if (!prog)
758 return 0;
759
760 ctx.meta = &meta;
761 ctx.map = info->map;
762 if (v) {
763 ctx.key = &info->index;
764 ctx.sk = info->sk;
765 }
766
767 return bpf_iter_run_prog(prog, &ctx);
768}
769
770static void sock_map_seq_stop(struct seq_file *seq, void *v)
f58423ae 771 __releases(rcu)
03653515
LB		 772{
773 if (!v)
774 (void)sock_map_seq_show(seq, NULL);
775
776 /* pairs with sock_map_seq_start */
777 rcu_read_unlock();
778}
779
780static const struct seq_operations sock_map_seq_ops = {
781 .start = sock_map_seq_start,
782 .next = sock_map_seq_next,
783 .stop = sock_map_seq_stop,
784 .show = sock_map_seq_show,
785};
786
787static int sock_map_init_seq_private(void *priv_data,
788 struct bpf_iter_aux_info *aux)
789{
790 struct sock_map_seq_info *info = priv_data;
791
792 info->map = aux->map;
793 return 0;
794}
795
796static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
797 .seq_ops = &sock_map_seq_ops,
798 .init_seq_private = sock_map_init_seq_private,
799 .seq_priv_size = sizeof(struct sock_map_seq_info),
800};
801
2872e9ac 802static int sock_map_btf_id;
604326b4 803const struct bpf_map_ops sock_map_ops = {
f4d05259 804 .map_meta_equal = bpf_map_meta_equal,
604326b4
DB		 805 .map_alloc = sock_map_alloc,
806 .map_free = sock_map_free,
807 .map_get_next_key = sock_map_get_next_key,
c1cdf65d 808 .map_lookup_elem_sys_only = sock_map_lookup_sys,
0126240f 809 .map_update_elem = sock_map_update_elem,
604326b4
DB		 810 .map_delete_elem = sock_map_delete_elem,
811 .map_lookup_elem = sock_map_lookup,
812 .map_release_uref = sock_map_release_progs,
813 .map_check_btf = map_check_no_btf,
2872e9ac
AI		 814 .map_btf_name = "bpf_stab",
815 .map_btf_id = &sock_map_btf_id,
03653515 816 .iter_seq_info = &sock_map_iter_seq_info,
604326b4
DB		 817};
818
032a6b35 819struct bpf_shtab_elem {
604326b4
DB		 820 struct rcu_head rcu;
821 u32 hash;
822 struct sock *sk;
823 struct hlist_node node;
45a4296b 824 u8 key[];
604326b4
DB		 825};
826
032a6b35 827struct bpf_shtab_bucket {
604326b4
DB		 828 struct hlist_head head;
829 raw_spinlock_t lock;
830};
831
032a6b35 832struct bpf_shtab {
604326b4 833 struct bpf_map map;
032a6b35 834 struct bpf_shtab_bucket *buckets;
604326b4
DB		 835 u32 buckets_num;
836 u32 elem_size;
837 struct sk_psock_progs progs;
838 atomic_t count;
839};
840
841static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
842{
843 return jhash(key, len, 0);
844}
845
032a6b35
AI		 846static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
847 u32 hash)
604326b4
DB		 848{
849 return &htab->buckets[hash & (htab->buckets_num - 1)];
850}
851
032a6b35 852static struct bpf_shtab_elem *
604326b4
DB		 853sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
854 u32 key_size)
855{
032a6b35 856 struct bpf_shtab_elem *elem;
604326b4
DB		 857
858 hlist_for_each_entry_rcu(elem, head, node) {
859 if (elem->hash == hash &&
860 !memcmp(&elem->key, key, key_size))
861 return elem;
862 }
863
864 return NULL;
865}
866
867static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
868{
032a6b35 869 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
604326b4 870 u32 key_size = map->key_size, hash;
032a6b35
AI		 871 struct bpf_shtab_bucket *bucket;
872 struct bpf_shtab_elem *elem;
604326b4
DB		 873
874 WARN_ON_ONCE(!rcu_read_lock_held());
875
876 hash = sock_hash_bucket_hash(key, key_size);
877 bucket = sock_hash_select_bucket(htab, hash);
878 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
879
880 return elem ? elem->sk : NULL;
881}
882
032a6b35
AI		 883static void sock_hash_free_elem(struct bpf_shtab *htab,
884 struct bpf_shtab_elem *elem)
604326b4
DB		 885{
886 atomic_dec(&htab->count);
887 kfree_rcu(elem, rcu);
888}
889
890static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
891 void *link_raw)
892{
032a6b35
AI		 893 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
894 struct bpf_shtab_elem *elem_probe, *elem = link_raw;
895 struct bpf_shtab_bucket *bucket;
604326b4
DB		 896
897 WARN_ON_ONCE(!rcu_read_lock_held());
898 bucket = sock_hash_select_bucket(htab, elem->hash);
899
900 /* elem may be deleted in parallel from the map, but access here
901 * is okay since it's going away only after RCU grace period.
902 * However, we need to check whether it's still present.
903 */
904 raw_spin_lock_bh(&bucket->lock);
905 elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
906 elem->key, map->key_size);
907 if (elem_probe && elem_probe == elem) {
908 hlist_del_rcu(&elem->node);
909 sock_map_unref(elem->sk, elem);
910 sock_hash_free_elem(htab, elem);
911 }
912 raw_spin_unlock_bh(&bucket->lock);
913}
914
915static int sock_hash_delete_elem(struct bpf_map *map, void *key)
916{
032a6b35 917 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
604326b4 918 u32 hash, key_size = map->key_size;
032a6b35
AI		 919 struct bpf_shtab_bucket *bucket;
920 struct bpf_shtab_elem *elem;
604326b4
DB		 921 int ret = -ENOENT;
922
923 hash = sock_hash_bucket_hash(key, key_size);
924 bucket = sock_hash_select_bucket(htab, hash);
925
926 raw_spin_lock_bh(&bucket->lock);
927 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
928 if (elem) {
929 hlist_del_rcu(&elem->node);
930 sock_map_unref(elem->sk, elem);
931 sock_hash_free_elem(htab, elem);
932 ret = 0;
933 }
934 raw_spin_unlock_bh(&bucket->lock);
935 return ret;
936}
937
032a6b35
AI		 938static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
939 void *key, u32 key_size,
940 u32 hash, struct sock *sk,
941 struct bpf_shtab_elem *old)
604326b4 942{
032a6b35 943 struct bpf_shtab_elem *new;
604326b4
DB		 944
945 if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
946 if (!old) {
947 atomic_dec(&htab->count);
948 return ERR_PTR(-E2BIG);
949 }
950 }
951
7846dd9f
RG		 952 new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
953 GFP_ATOMIC | __GFP_NOWARN,
954 htab->map.numa_node);
604326b4
DB		 955 if (!new) {
956 atomic_dec(&htab->count);
957 return ERR_PTR(-ENOMEM);
958 }
959 memcpy(new->key, key, key_size);
960 new->sk = sk;
961 new->hash = hash;
962 return new;
963}
964
965static int sock_hash_update_common(struct bpf_map *map, void *key,
966 struct sock *sk, u64 flags)
967{
032a6b35 968 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
604326b4 969 u32 key_size = map->key_size, hash;
032a6b35
AI		 970 struct bpf_shtab_elem *elem, *elem_new;
971 struct bpf_shtab_bucket *bucket;
604326b4
DB		 972 struct sk_psock_link *link;
973 struct sk_psock *psock;
974 int ret;
975
976 WARN_ON_ONCE(!rcu_read_lock_held());
977 if (unlikely(flags > BPF_EXIST))
978 return -EINVAL;
979
980 link = sk_psock_init_link();
981 if (!link)
982 return -ENOMEM;
983
b0170552 984 ret = sock_map_link(map, sk);
604326b4
DB		 985 if (ret < 0)
986 goto out_free;
987
988 psock = sk_psock(sk);
989 WARN_ON_ONCE(!psock);
990
991 hash = sock_hash_bucket_hash(key, key_size);
992 bucket = sock_hash_select_bucket(htab, hash);
993
994 raw_spin_lock_bh(&bucket->lock);
995 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
996 if (elem && flags == BPF_NOEXIST) {
997 ret = -EEXIST;
998 goto out_unlock;
999 } else if (!elem && flags == BPF_EXIST) {
1000 ret = -ENOENT;
1001 goto out_unlock;
1002 }
1003
1004 elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1005 if (IS_ERR(elem_new)) {
1006 ret = PTR_ERR(elem_new);
1007 goto out_unlock;
1008 }
1009
1010 sock_map_add_link(psock, link, map, elem_new);
1011 /* Add new element to the head of the list, so that
1012 * concurrent search will find it before old elem.
1013 */
1014 hlist_add_head_rcu(&elem_new->node, &bucket->head);
1015 if (elem) {
1016 hlist_del_rcu(&elem->node);
1017 sock_map_unref(elem->sk, elem);
1018 sock_hash_free_elem(htab, elem);
1019 }
1020 raw_spin_unlock_bh(&bucket->lock);
1021 return 0;
1022out_unlock:
1023 raw_spin_unlock_bh(&bucket->lock);
1024 sk_psock_put(sk, psock);
1025out_free:
1026 sk_psock_free_link(link);
1027 return ret;
1028}
1029
604326b4
DB		 1030static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1031 void *key_next)
1032{
032a6b35
AI		 1033 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1034 struct bpf_shtab_elem *elem, *elem_next;
604326b4
DB		 1035 u32 hash, key_size = map->key_size;
1036 struct hlist_head *head;
1037 int i = 0;
1038
1039 if (!key)
1040 goto find_first_elem;
1041 hash = sock_hash_bucket_hash(key, key_size);
1042 head = &sock_hash_select_bucket(htab, hash)->head;
1043 elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1044 if (!elem)
1045 goto find_first_elem;
1046
03653515 1047 elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
032a6b35 1048 struct bpf_shtab_elem, node);
604326b4
DB		 1049 if (elem_next) {
1050 memcpy(key_next, elem_next->key, key_size);
1051 return 0;
1052 }
1053
1054 i = hash & (htab->buckets_num - 1);
1055 i++;
1056find_first_elem:
1057 for (; i < htab->buckets_num; i++) {
1058 head = &sock_hash_select_bucket(htab, i)->head;
03653515 1059 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
032a6b35 1060 struct bpf_shtab_elem, node);
604326b4
DB		 1061 if (elem_next) {
1062 memcpy(key_next, elem_next->key, key_size);
1063 return 0;
1064 }
1065 }
1066
1067 return -ENOENT;
1068}
1069
1070static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1071{
032a6b35 1072 struct bpf_shtab *htab;
604326b4 1073 int i, err;
604326b4
DB		 1074
1075 if (!capable(CAP_NET_ADMIN))
1076 return ERR_PTR(-EPERM);
1077 if (attr->max_entries == 0 ||
1078 attr->key_size == 0 ||
c1cdf65d
JS		 1079 (attr->value_size != sizeof(u32) &&
1080 attr->value_size != sizeof(u64)) ||
604326b4
DB		 1081 attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1082 return ERR_PTR(-EINVAL);
1083 if (attr->key_size > MAX_BPF_STACK)
1084 return ERR_PTR(-E2BIG);
1085
7846dd9f 1086 htab = kzalloc(sizeof(*htab), GFP_USER | __GFP_ACCOUNT);
604326b4
DB		 1087 if (!htab)
1088 return ERR_PTR(-ENOMEM);
1089
1090 bpf_map_init_from_attr(&htab->map, attr);
1091
1092 htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
032a6b35 1093 htab->elem_size = sizeof(struct bpf_shtab_elem) +
604326b4
DB		 1094 round_up(htab->map.key_size, 8);
1095 if (htab->buckets_num == 0 ||
032a6b35 1096 htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
604326b4
DB		 1097 err = -EINVAL;
1098 goto free_htab;
1099 }
1100
604326b4 1101 htab->buckets = bpf_map_area_alloc(htab->buckets_num *
032a6b35 1102 sizeof(struct bpf_shtab_bucket),
604326b4
DB		 1103 htab->map.numa_node);
1104 if (!htab->buckets) {
1105 err = -ENOMEM;
1106 goto free_htab;
1107 }
1108
1109 for (i = 0; i < htab->buckets_num; i++) {
1110 INIT_HLIST_HEAD(&htab->buckets[i].head);
1111 raw_spin_lock_init(&htab->buckets[i].lock);
1112 }
1113
1114 return &htab->map;
1115free_htab:
1116 kfree(htab);
1117 return ERR_PTR(err);
1118}
1119
1120static void sock_hash_free(struct bpf_map *map)
1121{
032a6b35
AI		 1122 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1123 struct bpf_shtab_bucket *bucket;
75e68e5b 1124 struct hlist_head unlink_list;
032a6b35 1125 struct bpf_shtab_elem *elem;
604326b4
DB		 1126 struct hlist_node *node;
1127 int i;
1128
90db6d77
JF		 1129 /* After the sync no updates or deletes will be in-flight so it
1130 * is safe to walk map and remove entries without risking a race
1131 * in EEXIST update case.
1132 */
604326b4 1133 synchronize_rcu();
604326b4
DB		 1134 for (i = 0; i < htab->buckets_num; i++) {
1135 bucket = sock_hash_select_bucket(htab, i);
75e68e5b
JS		 1136
1137 /* We are racing with sock_hash_delete_from_link to
1138 * enter the spin-lock critical section. Every socket on
1139 * the list is still linked to sockhash. Since link
1140 * exists, psock exists and holds a ref to socket. That
1141 * lets us to grab a socket ref too.
1142 */
1143 raw_spin_lock_bh(&bucket->lock);
1144 hlist_for_each_entry(elem, &bucket->head, node)
1145 sock_hold(elem->sk);
1146 hlist_move_list(&bucket->head, &unlink_list);
1147 raw_spin_unlock_bh(&bucket->lock);
1148
1149 /* Process removed entries out of atomic context to
1150 * block for socket lock before deleting the psock's
1151 * link to sockhash.
1152 */
1153 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1154 hlist_del(&elem->node);
7e81a353 1155 lock_sock(elem->sk);
db6a5018 1156 rcu_read_lock();
604326b4 1157 sock_map_unref(elem->sk, elem);
db6a5018 1158 rcu_read_unlock();
7e81a353 1159 release_sock(elem->sk);
75e68e5b 1160 sock_put(elem->sk);
33a7c831 1161 sock_hash_free_elem(htab, elem);
604326b4 1162 }
604326b4 1163 }
604326b4 1164
0b2dc839
JS		 1165 /* wait for psock readers accessing its map link */
1166 synchronize_rcu();
1167
604326b4
DB		 1168 bpf_map_area_free(htab->buckets);
1169 kfree(htab);
1170}
1171
c1cdf65d
JS		 1172static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1173{
1174 struct sock *sk;
1175
1176 if (map->value_size != sizeof(u64))
1177 return ERR_PTR(-ENOSPC);
1178
1179 sk = __sock_hash_lookup_elem(map, key);
1180 if (!sk)
1181 return ERR_PTR(-ENOENT);
1182
92acdc58 1183 __sock_gen_cookie(sk);
c1cdf65d
JS		 1184 return &sk->sk_cookie;
1185}
1186
1d59f3bc
JS		 1187static void *sock_hash_lookup(struct bpf_map *map, void *key)
1188{
64d85290
JS		 1189 struct sock *sk;
1190
1191 sk = __sock_hash_lookup_elem(map, key);
654785a1 1192 if (!sk)
64d85290
JS		 1193 return NULL;
1194 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1195 return NULL;
1196 return sk;
1d59f3bc
JS		 1197}
1198
604326b4
DB		 1199static void sock_hash_release_progs(struct bpf_map *map)
1200{
032a6b35 1201 psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
604326b4
DB		 1202}
1203
1204BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1205 struct bpf_map *, map, void *, key, u64, flags)
1206{
1207 WARN_ON_ONCE(!rcu_read_lock_held());
1208
1209 if (likely(sock_map_sk_is_suitable(sops->sk) &&
1210 sock_map_op_okay(sops)))
1211 return sock_hash_update_common(map, key, sops->sk, flags);
1212 return -EOPNOTSUPP;
1213}
1214
1215const struct bpf_func_proto bpf_sock_hash_update_proto = {
1216 .func = bpf_sock_hash_update,
1217 .gpl_only = false,
1218 .pkt_access = true,
1219 .ret_type = RET_INTEGER,
1220 .arg1_type = ARG_PTR_TO_CTX,
1221 .arg2_type = ARG_CONST_MAP_PTR,
1222 .arg3_type = ARG_PTR_TO_MAP_KEY,
1223 .arg4_type = ARG_ANYTHING,
1224};
1225
1226BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1227 struct bpf_map *, map, void *, key, u64, flags)
1228{
8ca30379 1229 struct sock *sk;
604326b4
DB		 1230
1231 if (unlikely(flags & ~(BPF_F_INGRESS)))
1232 return SK_DROP;
8ca30379
JS		 1233
1234 sk = __sock_hash_lookup_elem(map, key);
1235 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
604326b4 1236 return SK_DROP;
8ca30379 1237
e3526bb9 1238 skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
604326b4
DB		 1239 return SK_PASS;
1240}
1241
1242const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1243 .func = bpf_sk_redirect_hash,
1244 .gpl_only = false,
1245 .ret_type = RET_INTEGER,
1246 .arg1_type = ARG_PTR_TO_CTX,
1247 .arg2_type = ARG_CONST_MAP_PTR,
1248 .arg3_type = ARG_PTR_TO_MAP_KEY,
1249 .arg4_type = ARG_ANYTHING,
1250};
1251
1252BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1253 struct bpf_map *, map, void *, key, u64, flags)
1254{
8ca30379
JS		 1255 struct sock *sk;
1256
604326b4
DB		 1257 if (unlikely(flags & ~(BPF_F_INGRESS)))
1258 return SK_DROP;
8ca30379
JS		 1259
1260 sk = __sock_hash_lookup_elem(map, key);
1261 if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
604326b4 1262 return SK_DROP;
8ca30379
JS		 1263
1264 msg->flags = flags;
1265 msg->sk_redir = sk;
604326b4
DB		 1266 return SK_PASS;
1267}
1268
1269const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1270 .func = bpf_msg_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
03653515
LB		 1279struct sock_hash_seq_info {
1280 struct bpf_map *map;
1281 struct bpf_shtab *htab;
1282 u32 bucket_id;
1283};
1284
1285static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1286 struct bpf_shtab_elem *prev_elem)
1287{
1288 const struct bpf_shtab *htab = info->htab;
1289 struct bpf_shtab_bucket *bucket;
1290 struct bpf_shtab_elem *elem;
1291 struct hlist_node *node;
1292
1293 /* try to find next elem in the same bucket */
1294 if (prev_elem) {
1295 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1296 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1297 if (elem)
1298 return elem;
1299
1300 /* no more elements, continue in the next bucket */
1301 info->bucket_id++;
1302 }
1303
1304 for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1305 bucket = &htab->buckets[info->bucket_id];
1306 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1307 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1308 if (elem)
1309 return elem;
1310 }
1311
1312 return NULL;
1313}
1314
1315static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
f58423ae 1316 __acquires(rcu)
03653515
LB		 1317{
1318 struct sock_hash_seq_info *info = seq->private;
1319
1320 if (*pos == 0)
1321 ++*pos;
1322
1323 /* pairs with sock_hash_seq_stop */
1324 rcu_read_lock();
1325 return sock_hash_seq_find_next(info, NULL);
1326}
1327
1328static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
f58423ae 1329 __must_hold(rcu)
03653515
LB		 1330{
1331 struct sock_hash_seq_info *info = seq->private;
1332
1333 ++*pos;
1334 return sock_hash_seq_find_next(info, v);
1335}
1336
1337static int sock_hash_seq_show(struct seq_file *seq, void *v)
f58423ae 1338 __must_hold(rcu)
03653515
LB		 1339{
1340 struct sock_hash_seq_info *info = seq->private;
1341 struct bpf_iter__sockmap ctx = {};
1342 struct bpf_shtab_elem *elem = v;
1343 struct bpf_iter_meta meta;
1344 struct bpf_prog *prog;
1345
1346 meta.seq = seq;
1347 prog = bpf_iter_get_info(&meta, !elem);
1348 if (!prog)
1349 return 0;
1350
1351 ctx.meta = &meta;
1352 ctx.map = info->map;
1353 if (elem) {
1354 ctx.key = elem->key;
1355 ctx.sk = elem->sk;
1356 }
1357
1358 return bpf_iter_run_prog(prog, &ctx);
1359}
1360
1361static void sock_hash_seq_stop(struct seq_file *seq, void *v)
f58423ae 1362 __releases(rcu)
03653515
LB		 1363{
1364 if (!v)
1365 (void)sock_hash_seq_show(seq, NULL);
1366
1367 /* pairs with sock_hash_seq_start */
1368 rcu_read_unlock();
1369}
1370
1371static const struct seq_operations sock_hash_seq_ops = {
1372 .start = sock_hash_seq_start,
1373 .next = sock_hash_seq_next,
1374 .stop = sock_hash_seq_stop,
1375 .show = sock_hash_seq_show,
1376};
1377
1378static int sock_hash_init_seq_private(void *priv_data,
1379 struct bpf_iter_aux_info *aux)
1380{
1381 struct sock_hash_seq_info *info = priv_data;
1382
1383 info->map = aux->map;
1384 info->htab = container_of(aux->map, struct bpf_shtab, map);
1385 return 0;
1386}
1387
1388static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1389 .seq_ops = &sock_hash_seq_ops,
1390 .init_seq_private = sock_hash_init_seq_private,
1391 .seq_priv_size = sizeof(struct sock_hash_seq_info),
1392};
1393
2872e9ac 1394static int sock_hash_map_btf_id;
604326b4 1395const struct bpf_map_ops sock_hash_ops = {
f4d05259 1396 .map_meta_equal = bpf_map_meta_equal,
604326b4
DB		 1397 .map_alloc = sock_hash_alloc,
1398 .map_free = sock_hash_free,
1399 .map_get_next_key = sock_hash_get_next_key,
0126240f 1400 .map_update_elem = sock_map_update_elem,
604326b4 1401 .map_delete_elem = sock_hash_delete_elem,
1d59f3bc 1402 .map_lookup_elem = sock_hash_lookup,
c1cdf65d 1403 .map_lookup_elem_sys_only = sock_hash_lookup_sys,
604326b4
DB		 1404 .map_release_uref = sock_hash_release_progs,
1405 .map_check_btf = map_check_no_btf,
2872e9ac
AI		 1406 .map_btf_name = "bpf_shtab",
1407 .map_btf_id = &sock_hash_map_btf_id,
03653515 1408 .iter_seq_info = &sock_hash_iter_seq_info,
604326b4
DB		 1409};
1410
1411static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1412{
1413 switch (map->map_type) {
1414 case BPF_MAP_TYPE_SOCKMAP:
1415 return &container_of(map, struct bpf_stab, map)->progs;
1416 case BPF_MAP_TYPE_SOCKHASH:
032a6b35 1417 return &container_of(map, struct bpf_shtab, map)->progs;
604326b4
DB		 1418 default:
1419 break;
1420 }
1421
1422 return NULL;
1423}
1424
4675e234
CW		 1425static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1426 struct bpf_prog *old, u32 which)
604326b4
DB		 1427{
1428 struct sk_psock_progs *progs = sock_map_progs(map);
bb0de313 1429 struct bpf_prog **pprog;
604326b4
DB		 1430
1431 if (!progs)
1432 return -EOPNOTSUPP;
1433
1434 switch (which) {
1435 case BPF_SK_MSG_VERDICT:
bb0de313 1436 pprog = &progs->msg_parser;
604326b4 1437 break;
88759609 1438#if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
604326b4 1439 case BPF_SK_SKB_STREAM_PARSER:
ae8b8332 1440 pprog = &progs->stream_parser;
604326b4 1441 break;
88759609 1442#endif
604326b4 1443 case BPF_SK_SKB_STREAM_VERDICT:
a7ba4558
CW		 1444 if (progs->skb_verdict)
1445 return -EBUSY;
ae8b8332 1446 pprog = &progs->stream_verdict;
604326b4 1447 break;
a7ba4558
CW		 1448 case BPF_SK_SKB_VERDICT:
1449 if (progs->stream_verdict)
1450 return -EBUSY;
1451 pprog = &progs->skb_verdict;
1452 break;
604326b4
DB		 1453 default:
1454 return -EOPNOTSUPP;
1455 }
1456
bb0de313
LB		 1457 if (old)
1458 return psock_replace_prog(pprog, prog, old);
1459
1460 psock_set_prog(pprog, prog);
604326b4
DB		 1461 return 0;
1462}
1463
f747632b 1464static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
604326b4
DB		 1465{
1466 switch (link->map->map_type) {
1467 case BPF_MAP_TYPE_SOCKMAP:
1468 return sock_map_delete_from_link(link->map, sk,
1469 link->link_raw);
1470 case BPF_MAP_TYPE_SOCKHASH:
1471 return sock_hash_delete_from_link(link->map, sk,
1472 link->link_raw);
1473 default:
1474 break;
1475 }
1476}
f747632b
LB		 1477
1478static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1479{
1480 struct sk_psock_link *link;
1481
1482 while ((link = sk_psock_link_pop(psock))) {
1483 sock_map_unlink(sk, link);
1484 sk_psock_free_link(link);
1485 }
1486}
1487
1488void sock_map_unhash(struct sock *sk)
1489{
1490 void (*saved_unhash)(struct sock *sk);
1491 struct sk_psock *psock;
1492
1493 rcu_read_lock();
1494 psock = sk_psock(sk);
1495 if (unlikely(!psock)) {
1496 rcu_read_unlock();
1497 if (sk->sk_prot->unhash)
1498 sk->sk_prot->unhash(sk);
1499 return;
1500 }
1501
1502 saved_unhash = psock->saved_unhash;
1503 sock_map_remove_links(sk, psock);
1504 rcu_read_unlock();
1505 saved_unhash(sk);
1506}
94531cfc 1507EXPORT_SYMBOL_GPL(sock_map_unhash);
f747632b
LB		 1508
1509void sock_map_close(struct sock *sk, long timeout)
1510{
1511 void (*saved_close)(struct sock *sk, long timeout);
1512 struct sk_psock *psock;
1513
1514 lock_sock(sk);
1515 rcu_read_lock();
aadb2bb8 1516 psock = sk_psock_get(sk);
f747632b
LB		 1517 if (unlikely(!psock)) {
1518 rcu_read_unlock();
1519 release_sock(sk);
1520 return sk->sk_prot->close(sk, timeout);
1521 }
1522
1523 saved_close = psock->saved_close;
1524 sock_map_remove_links(sk, psock);
1525 rcu_read_unlock();
799aa7f9 1526 sk_psock_stop(psock, true);
aadb2bb8 1527 sk_psock_put(sk, psock);
f747632b
LB		 1528 release_sock(sk);
1529 saved_close(sk, timeout);
1530}
c6382918 1531EXPORT_SYMBOL_GPL(sock_map_close);
03653515
LB		 1532
1533static int sock_map_iter_attach_target(struct bpf_prog *prog,
1534 union bpf_iter_link_info *linfo,
1535 struct bpf_iter_aux_info *aux)
1536{
1537 struct bpf_map *map;
1538 int err = -EINVAL;
1539
1540 if (!linfo->map.map_fd)
1541 return -EBADF;
1542
1543 map = bpf_map_get_with_uref(linfo->map.map_fd);
1544 if (IS_ERR(map))
1545 return PTR_ERR(map);
1546
1547 if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1548 map->map_type != BPF_MAP_TYPE_SOCKHASH)
1549 goto put_map;
1550
1551 if (prog->aux->max_rdonly_access > map->key_size) {
1552 err = -EACCES;
1553 goto put_map;
1554 }
1555
1556 aux->map = map;
1557 return 0;
1558
1559put_map:
1560 bpf_map_put_with_uref(map);
1561 return err;
1562}
1563
1564static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1565{
1566 bpf_map_put_with_uref(aux->map);
1567}
1568
1569static struct bpf_iter_reg sock_map_iter_reg = {
1570 .target = "sockmap",
1571 .attach_target = sock_map_iter_attach_target,
1572 .detach_target = sock_map_iter_detach_target,
1573 .show_fdinfo = bpf_iter_map_show_fdinfo,
1574 .fill_link_info = bpf_iter_map_fill_link_info,
1575 .ctx_arg_info_size = 2,
1576 .ctx_arg_info = {
1577 { offsetof(struct bpf_iter__sockmap, key),
1578 PTR_TO_RDONLY_BUF_OR_NULL },
1579 { offsetof(struct bpf_iter__sockmap, sk),
1580 PTR_TO_BTF_ID_OR_NULL },
1581 },
1582};
1583
1584static int __init bpf_sockmap_iter_init(void)
1585{
1586 sock_map_iter_reg.ctx_arg_info[1].btf_id =
1587 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1588 return bpf_iter_reg_target(&sock_map_iter_reg);
1589}
1590late_initcall(bpf_sockmap_iter_init);
Ubuntu Jammy Linux kernel mirror