xref: /openbmc/linux/net/core/bpf_sk_storage.c (revision 2f61c664)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2019 Facebook  */
3 #include <linux/rculist.h>
4 #include <linux/list.h>
5 #include <linux/hash.h>
6 #include <linux/types.h>
7 #include <linux/spinlock.h>
8 #include <linux/bpf.h>
9 #include <linux/btf.h>
10 #include <linux/btf_ids.h>
11 #include <linux/bpf_local_storage.h>
12 #include <net/bpf_sk_storage.h>
13 #include <net/sock.h>
14 #include <uapi/linux/sock_diag.h>
15 #include <uapi/linux/btf.h>
16 
17 DEFINE_BPF_STORAGE_CACHE(sk_cache);
18 
19 static struct bpf_local_storage_data *
20 bpf_sk_storage_lookup(struct sock *sk, struct bpf_map *map, bool cacheit_lockit)
21 {
22 	struct bpf_local_storage *sk_storage;
23 	struct bpf_local_storage_map *smap;
24 
25 	sk_storage = rcu_dereference(sk->sk_bpf_storage);
26 	if (!sk_storage)
27 		return NULL;
28 
29 	smap = (struct bpf_local_storage_map *)map;
30 	return bpf_local_storage_lookup(sk_storage, smap, cacheit_lockit);
31 }
32 
33 static int bpf_sk_storage_del(struct sock *sk, struct bpf_map *map)
34 {
35 	struct bpf_local_storage_data *sdata;
36 
37 	sdata = bpf_sk_storage_lookup(sk, map, false);
38 	if (!sdata)
39 		return -ENOENT;
40 
41 	bpf_selem_unlink(SELEM(sdata));
42 
43 	return 0;
44 }
45 
46 /* Called by __sk_destruct() & bpf_sk_storage_clone() */
47 void bpf_sk_storage_free(struct sock *sk)
48 {
49 	struct bpf_local_storage_elem *selem;
50 	struct bpf_local_storage *sk_storage;
51 	bool free_sk_storage = false;
52 	struct hlist_node *n;
53 
54 	rcu_read_lock();
55 	sk_storage = rcu_dereference(sk->sk_bpf_storage);
56 	if (!sk_storage) {
57 		rcu_read_unlock();
58 		return;
59 	}
60 
61 	/* Netiher the bpf_prog nor the bpf-map's syscall
62 	 * could be modifying the sk_storage->list now.
63 	 * Thus, no elem can be added-to or deleted-from the
64 	 * sk_storage->list by the bpf_prog or by the bpf-map's syscall.
65 	 *
66 	 * It is racing with bpf_local_storage_map_free() alone
67 	 * when unlinking elem from the sk_storage->list and
68 	 * the map's bucket->list.
69 	 */
70 	raw_spin_lock_bh(&sk_storage->lock);
71 	hlist_for_each_entry_safe(selem, n, &sk_storage->list, snode) {
72 		/* Always unlink from map before unlinking from
73 		 * sk_storage.
74 		 */
75 		bpf_selem_unlink_map(selem);
76 		free_sk_storage = bpf_selem_unlink_storage_nolock(sk_storage,
77 								  selem, true);
78 	}
79 	raw_spin_unlock_bh(&sk_storage->lock);
80 	rcu_read_unlock();
81 
82 	if (free_sk_storage)
83 		kfree_rcu(sk_storage, rcu);
84 }
85 
86 static void bpf_sk_storage_map_free(struct bpf_map *map)
87 {
88 	struct bpf_local_storage_map *smap;
89 
90 	smap = (struct bpf_local_storage_map *)map;
91 	bpf_local_storage_cache_idx_free(&sk_cache, smap->cache_idx);
92 	bpf_local_storage_map_free(smap, NULL);
93 }
94 
95 static struct bpf_map *bpf_sk_storage_map_alloc(union bpf_attr *attr)
96 {
97 	struct bpf_local_storage_map *smap;
98 
99 	smap = bpf_local_storage_map_alloc(attr);
100 	if (IS_ERR(smap))
101 		return ERR_CAST(smap);
102 
103 	smap->cache_idx = bpf_local_storage_cache_idx_get(&sk_cache);
104 	return &smap->map;
105 }
106 
107 static int notsupp_get_next_key(struct bpf_map *map, void *key,
108 				void *next_key)
109 {
110 	return -ENOTSUPP;
111 }
112 
113 static void *bpf_fd_sk_storage_lookup_elem(struct bpf_map *map, void *key)
114 {
115 	struct bpf_local_storage_data *sdata;
116 	struct socket *sock;
117 	int fd, err;
118 
119 	fd = *(int *)key;
120 	sock = sockfd_lookup(fd, &err);
121 	if (sock) {
122 		sdata = bpf_sk_storage_lookup(sock->sk, map, true);
123 		sockfd_put(sock);
124 		return sdata ? sdata->data : NULL;
125 	}
126 
127 	return ERR_PTR(err);
128 }
129 
130 static int bpf_fd_sk_storage_update_elem(struct bpf_map *map, void *key,
131 					 void *value, u64 map_flags)
132 {
133 	struct bpf_local_storage_data *sdata;
134 	struct socket *sock;
135 	int fd, err;
136 
137 	fd = *(int *)key;
138 	sock = sockfd_lookup(fd, &err);
139 	if (sock) {
140 		sdata = bpf_local_storage_update(
141 			sock->sk, (struct bpf_local_storage_map *)map, value,
142 			map_flags);
143 		sockfd_put(sock);
144 		return PTR_ERR_OR_ZERO(sdata);
145 	}
146 
147 	return err;
148 }
149 
150 static int bpf_fd_sk_storage_delete_elem(struct bpf_map *map, void *key)
151 {
152 	struct socket *sock;
153 	int fd, err;
154 
155 	fd = *(int *)key;
156 	sock = sockfd_lookup(fd, &err);
157 	if (sock) {
158 		err = bpf_sk_storage_del(sock->sk, map);
159 		sockfd_put(sock);
160 		return err;
161 	}
162 
163 	return err;
164 }
165 
166 static struct bpf_local_storage_elem *
167 bpf_sk_storage_clone_elem(struct sock *newsk,
168 			  struct bpf_local_storage_map *smap,
169 			  struct bpf_local_storage_elem *selem)
170 {
171 	struct bpf_local_storage_elem *copy_selem;
172 
173 	copy_selem = bpf_selem_alloc(smap, newsk, NULL, true);
174 	if (!copy_selem)
175 		return NULL;
176 
177 	if (map_value_has_spin_lock(&smap->map))
178 		copy_map_value_locked(&smap->map, SDATA(copy_selem)->data,
179 				      SDATA(selem)->data, true);
180 	else
181 		copy_map_value(&smap->map, SDATA(copy_selem)->data,
182 			       SDATA(selem)->data);
183 
184 	return copy_selem;
185 }
186 
187 int bpf_sk_storage_clone(const struct sock *sk, struct sock *newsk)
188 {
189 	struct bpf_local_storage *new_sk_storage = NULL;
190 	struct bpf_local_storage *sk_storage;
191 	struct bpf_local_storage_elem *selem;
192 	int ret = 0;
193 
194 	RCU_INIT_POINTER(newsk->sk_bpf_storage, NULL);
195 
196 	rcu_read_lock();
197 	sk_storage = rcu_dereference(sk->sk_bpf_storage);
198 
199 	if (!sk_storage || hlist_empty(&sk_storage->list))
200 		goto out;
201 
202 	hlist_for_each_entry_rcu(selem, &sk_storage->list, snode) {
203 		struct bpf_local_storage_elem *copy_selem;
204 		struct bpf_local_storage_map *smap;
205 		struct bpf_map *map;
206 
207 		smap = rcu_dereference(SDATA(selem)->smap);
208 		if (!(smap->map.map_flags & BPF_F_CLONE))
209 			continue;
210 
211 		/* Note that for lockless listeners adding new element
212 		 * here can race with cleanup in bpf_local_storage_map_free.
213 		 * Try to grab map refcnt to make sure that it's still
214 		 * alive and prevent concurrent removal.
215 		 */
216 		map = bpf_map_inc_not_zero(&smap->map);
217 		if (IS_ERR(map))
218 			continue;
219 
220 		copy_selem = bpf_sk_storage_clone_elem(newsk, smap, selem);
221 		if (!copy_selem) {
222 			ret = -ENOMEM;
223 			bpf_map_put(map);
224 			goto out;
225 		}
226 
227 		if (new_sk_storage) {
228 			bpf_selem_link_map(smap, copy_selem);
229 			bpf_selem_link_storage_nolock(new_sk_storage, copy_selem);
230 		} else {
231 			ret = bpf_local_storage_alloc(newsk, smap, copy_selem);
232 			if (ret) {
233 				kfree(copy_selem);
234 				atomic_sub(smap->elem_size,
235 					   &newsk->sk_omem_alloc);
236 				bpf_map_put(map);
237 				goto out;
238 			}
239 
240 			new_sk_storage =
241 				rcu_dereference(copy_selem->local_storage);
242 		}
243 		bpf_map_put(map);
244 	}
245 
246 out:
247 	rcu_read_unlock();
248 
249 	/* In case of an error, don't free anything explicitly here, the
250 	 * caller is responsible to call bpf_sk_storage_free.
251 	 */
252 
253 	return ret;
254 }
255 
256 BPF_CALL_4(bpf_sk_storage_get, struct bpf_map *, map, struct sock *, sk,
257 	   void *, value, u64, flags)
258 {
259 	struct bpf_local_storage_data *sdata;
260 
261 	if (!sk || !sk_fullsock(sk) || flags > BPF_SK_STORAGE_GET_F_CREATE)
262 		return (unsigned long)NULL;
263 
264 	sdata = bpf_sk_storage_lookup(sk, map, true);
265 	if (sdata)
266 		return (unsigned long)sdata->data;
267 
268 	if (flags == BPF_SK_STORAGE_GET_F_CREATE &&
269 	    /* Cannot add new elem to a going away sk.
270 	     * Otherwise, the new elem may become a leak
271 	     * (and also other memory issues during map
272 	     *  destruction).
273 	     */
274 	    refcount_inc_not_zero(&sk->sk_refcnt)) {
275 		sdata = bpf_local_storage_update(
276 			sk, (struct bpf_local_storage_map *)map, value,
277 			BPF_NOEXIST);
278 		/* sk must be a fullsock (guaranteed by verifier),
279 		 * so sock_gen_put() is unnecessary.
280 		 */
281 		sock_put(sk);
282 		return IS_ERR(sdata) ?
283 			(unsigned long)NULL : (unsigned long)sdata->data;
284 	}
285 
286 	return (unsigned long)NULL;
287 }
288 
289 BPF_CALL_2(bpf_sk_storage_delete, struct bpf_map *, map, struct sock *, sk)
290 {
291 	if (!sk || !sk_fullsock(sk))
292 		return -EINVAL;
293 
294 	if (refcount_inc_not_zero(&sk->sk_refcnt)) {
295 		int err;
296 
297 		err = bpf_sk_storage_del(sk, map);
298 		sock_put(sk);
299 		return err;
300 	}
301 
302 	return -ENOENT;
303 }
304 
305 static int bpf_sk_storage_charge(struct bpf_local_storage_map *smap,
306 				 void *owner, u32 size)
307 {
308 	struct sock *sk = (struct sock *)owner;
309 
310 	/* same check as in sock_kmalloc() */
311 	if (size <= sysctl_optmem_max &&
312 	    atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
313 		atomic_add(size, &sk->sk_omem_alloc);
314 		return 0;
315 	}
316 
317 	return -ENOMEM;
318 }
319 
320 static void bpf_sk_storage_uncharge(struct bpf_local_storage_map *smap,
321 				    void *owner, u32 size)
322 {
323 	struct sock *sk = owner;
324 
325 	atomic_sub(size, &sk->sk_omem_alloc);
326 }
327 
328 static struct bpf_local_storage __rcu **
329 bpf_sk_storage_ptr(void *owner)
330 {
331 	struct sock *sk = owner;
332 
333 	return &sk->sk_bpf_storage;
334 }
335 
336 static int sk_storage_map_btf_id;
337 const struct bpf_map_ops sk_storage_map_ops = {
338 	.map_meta_equal = bpf_map_meta_equal,
339 	.map_alloc_check = bpf_local_storage_map_alloc_check,
340 	.map_alloc = bpf_sk_storage_map_alloc,
341 	.map_free = bpf_sk_storage_map_free,
342 	.map_get_next_key = notsupp_get_next_key,
343 	.map_lookup_elem = bpf_fd_sk_storage_lookup_elem,
344 	.map_update_elem = bpf_fd_sk_storage_update_elem,
345 	.map_delete_elem = bpf_fd_sk_storage_delete_elem,
346 	.map_check_btf = bpf_local_storage_map_check_btf,
347 	.map_btf_name = "bpf_local_storage_map",
348 	.map_btf_id = &sk_storage_map_btf_id,
349 	.map_local_storage_charge = bpf_sk_storage_charge,
350 	.map_local_storage_uncharge = bpf_sk_storage_uncharge,
351 	.map_owner_storage_ptr = bpf_sk_storage_ptr,
352 };
353 
354 const struct bpf_func_proto bpf_sk_storage_get_proto = {
355 	.func		= bpf_sk_storage_get,
356 	.gpl_only	= false,
357 	.ret_type	= RET_PTR_TO_MAP_VALUE_OR_NULL,
358 	.arg1_type	= ARG_CONST_MAP_PTR,
359 	.arg2_type	= ARG_PTR_TO_BTF_ID_SOCK_COMMON,
360 	.arg3_type	= ARG_PTR_TO_MAP_VALUE_OR_NULL,
361 	.arg4_type	= ARG_ANYTHING,
362 };
363 
364 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto = {
365 	.func		= bpf_sk_storage_get,
366 	.gpl_only	= false,
367 	.ret_type	= RET_PTR_TO_MAP_VALUE_OR_NULL,
368 	.arg1_type	= ARG_CONST_MAP_PTR,
369 	.arg2_type	= ARG_PTR_TO_CTX, /* context is 'struct sock' */
370 	.arg3_type	= ARG_PTR_TO_MAP_VALUE_OR_NULL,
371 	.arg4_type	= ARG_ANYTHING,
372 };
373 
374 const struct bpf_func_proto bpf_sk_storage_delete_proto = {
375 	.func		= bpf_sk_storage_delete,
376 	.gpl_only	= false,
377 	.ret_type	= RET_INTEGER,
378 	.arg1_type	= ARG_CONST_MAP_PTR,
379 	.arg2_type	= ARG_PTR_TO_BTF_ID_SOCK_COMMON,
380 };
381 
382 static bool bpf_sk_storage_tracing_allowed(const struct bpf_prog *prog)
383 {
384 	const struct btf *btf_vmlinux;
385 	const struct btf_type *t;
386 	const char *tname;
387 	u32 btf_id;
388 
389 	if (prog->aux->dst_prog)
390 		return false;
391 
392 	/* Ensure the tracing program is not tracing
393 	 * any bpf_sk_storage*() function and also
394 	 * use the bpf_sk_storage_(get|delete) helper.
395 	 */
396 	switch (prog->expected_attach_type) {
397 	case BPF_TRACE_ITER:
398 	case BPF_TRACE_RAW_TP:
399 		/* bpf_sk_storage has no trace point */
400 		return true;
401 	case BPF_TRACE_FENTRY:
402 	case BPF_TRACE_FEXIT:
403 		btf_vmlinux = bpf_get_btf_vmlinux();
404 		btf_id = prog->aux->attach_btf_id;
405 		t = btf_type_by_id(btf_vmlinux, btf_id);
406 		tname = btf_name_by_offset(btf_vmlinux, t->name_off);
407 		return !!strncmp(tname, "bpf_sk_storage",
408 				 strlen("bpf_sk_storage"));
409 	default:
410 		return false;
411 	}
412 
413 	return false;
414 }
415 
416 BPF_CALL_4(bpf_sk_storage_get_tracing, struct bpf_map *, map, struct sock *, sk,
417 	   void *, value, u64, flags)
418 {
419 	if (in_irq() || in_nmi())
420 		return (unsigned long)NULL;
421 
422 	return (unsigned long)____bpf_sk_storage_get(map, sk, value, flags);
423 }
424 
425 BPF_CALL_2(bpf_sk_storage_delete_tracing, struct bpf_map *, map,
426 	   struct sock *, sk)
427 {
428 	if (in_irq() || in_nmi())
429 		return -EPERM;
430 
431 	return ____bpf_sk_storage_delete(map, sk);
432 }
433 
434 const struct bpf_func_proto bpf_sk_storage_get_tracing_proto = {
435 	.func		= bpf_sk_storage_get_tracing,
436 	.gpl_only	= false,
437 	.ret_type	= RET_PTR_TO_MAP_VALUE_OR_NULL,
438 	.arg1_type	= ARG_CONST_MAP_PTR,
439 	.arg2_type	= ARG_PTR_TO_BTF_ID,
440 	.arg2_btf_id	= &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON],
441 	.arg3_type	= ARG_PTR_TO_MAP_VALUE_OR_NULL,
442 	.arg4_type	= ARG_ANYTHING,
443 	.allowed	= bpf_sk_storage_tracing_allowed,
444 };
445 
446 const struct bpf_func_proto bpf_sk_storage_delete_tracing_proto = {
447 	.func		= bpf_sk_storage_delete_tracing,
448 	.gpl_only	= false,
449 	.ret_type	= RET_INTEGER,
450 	.arg1_type	= ARG_CONST_MAP_PTR,
451 	.arg2_type	= ARG_PTR_TO_BTF_ID,
452 	.arg2_btf_id	= &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON],
453 	.allowed	= bpf_sk_storage_tracing_allowed,
454 };
455 
456 struct bpf_sk_storage_diag {
457 	u32 nr_maps;
458 	struct bpf_map *maps[];
459 };
460 
461 /* The reply will be like:
462  * INET_DIAG_BPF_SK_STORAGES (nla_nest)
463  *	SK_DIAG_BPF_STORAGE (nla_nest)
464  *		SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32)
465  *		SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit)
466  *	SK_DIAG_BPF_STORAGE (nla_nest)
467  *		SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32)
468  *		SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit)
469  *	....
470  */
471 static int nla_value_size(u32 value_size)
472 {
473 	/* SK_DIAG_BPF_STORAGE (nla_nest)
474 	 *	SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32)
475 	 *	SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit)
476 	 */
477 	return nla_total_size(0) + nla_total_size(sizeof(u32)) +
478 		nla_total_size_64bit(value_size);
479 }
480 
481 void bpf_sk_storage_diag_free(struct bpf_sk_storage_diag *diag)
482 {
483 	u32 i;
484 
485 	if (!diag)
486 		return;
487 
488 	for (i = 0; i < diag->nr_maps; i++)
489 		bpf_map_put(diag->maps[i]);
490 
491 	kfree(diag);
492 }
493 EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_free);
494 
495 static bool diag_check_dup(const struct bpf_sk_storage_diag *diag,
496 			   const struct bpf_map *map)
497 {
498 	u32 i;
499 
500 	for (i = 0; i < diag->nr_maps; i++) {
501 		if (diag->maps[i] == map)
502 			return true;
503 	}
504 
505 	return false;
506 }
507 
508 struct bpf_sk_storage_diag *
509 bpf_sk_storage_diag_alloc(const struct nlattr *nla_stgs)
510 {
511 	struct bpf_sk_storage_diag *diag;
512 	struct nlattr *nla;
513 	u32 nr_maps = 0;
514 	int rem, err;
515 
516 	/* bpf_local_storage_map is currently limited to CAP_SYS_ADMIN as
517 	 * the map_alloc_check() side also does.
518 	 */
519 	if (!bpf_capable())
520 		return ERR_PTR(-EPERM);
521 
522 	nla_for_each_nested(nla, nla_stgs, rem) {
523 		if (nla_type(nla) == SK_DIAG_BPF_STORAGE_REQ_MAP_FD)
524 			nr_maps++;
525 	}
526 
527 	diag = kzalloc(sizeof(*diag) + sizeof(diag->maps[0]) * nr_maps,
528 		       GFP_KERNEL);
529 	if (!diag)
530 		return ERR_PTR(-ENOMEM);
531 
532 	nla_for_each_nested(nla, nla_stgs, rem) {
533 		struct bpf_map *map;
534 		int map_fd;
535 
536 		if (nla_type(nla) != SK_DIAG_BPF_STORAGE_REQ_MAP_FD)
537 			continue;
538 
539 		map_fd = nla_get_u32(nla);
540 		map = bpf_map_get(map_fd);
541 		if (IS_ERR(map)) {
542 			err = PTR_ERR(map);
543 			goto err_free;
544 		}
545 		if (map->map_type != BPF_MAP_TYPE_SK_STORAGE) {
546 			bpf_map_put(map);
547 			err = -EINVAL;
548 			goto err_free;
549 		}
550 		if (diag_check_dup(diag, map)) {
551 			bpf_map_put(map);
552 			err = -EEXIST;
553 			goto err_free;
554 		}
555 		diag->maps[diag->nr_maps++] = map;
556 	}
557 
558 	return diag;
559 
560 err_free:
561 	bpf_sk_storage_diag_free(diag);
562 	return ERR_PTR(err);
563 }
564 EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_alloc);
565 
566 static int diag_get(struct bpf_local_storage_data *sdata, struct sk_buff *skb)
567 {
568 	struct nlattr *nla_stg, *nla_value;
569 	struct bpf_local_storage_map *smap;
570 
571 	/* It cannot exceed max nlattr's payload */
572 	BUILD_BUG_ON(U16_MAX - NLA_HDRLEN < BPF_LOCAL_STORAGE_MAX_VALUE_SIZE);
573 
574 	nla_stg = nla_nest_start(skb, SK_DIAG_BPF_STORAGE);
575 	if (!nla_stg)
576 		return -EMSGSIZE;
577 
578 	smap = rcu_dereference(sdata->smap);
579 	if (nla_put_u32(skb, SK_DIAG_BPF_STORAGE_MAP_ID, smap->map.id))
580 		goto errout;
581 
582 	nla_value = nla_reserve_64bit(skb, SK_DIAG_BPF_STORAGE_MAP_VALUE,
583 				      smap->map.value_size,
584 				      SK_DIAG_BPF_STORAGE_PAD);
585 	if (!nla_value)
586 		goto errout;
587 
588 	if (map_value_has_spin_lock(&smap->map))
589 		copy_map_value_locked(&smap->map, nla_data(nla_value),
590 				      sdata->data, true);
591 	else
592 		copy_map_value(&smap->map, nla_data(nla_value), sdata->data);
593 
594 	nla_nest_end(skb, nla_stg);
595 	return 0;
596 
597 errout:
598 	nla_nest_cancel(skb, nla_stg);
599 	return -EMSGSIZE;
600 }
601 
602 static int bpf_sk_storage_diag_put_all(struct sock *sk, struct sk_buff *skb,
603 				       int stg_array_type,
604 				       unsigned int *res_diag_size)
605 {
606 	/* stg_array_type (e.g. INET_DIAG_BPF_SK_STORAGES) */
607 	unsigned int diag_size = nla_total_size(0);
608 	struct bpf_local_storage *sk_storage;
609 	struct bpf_local_storage_elem *selem;
610 	struct bpf_local_storage_map *smap;
611 	struct nlattr *nla_stgs;
612 	unsigned int saved_len;
613 	int err = 0;
614 
615 	rcu_read_lock();
616 
617 	sk_storage = rcu_dereference(sk->sk_bpf_storage);
618 	if (!sk_storage || hlist_empty(&sk_storage->list)) {
619 		rcu_read_unlock();
620 		return 0;
621 	}
622 
623 	nla_stgs = nla_nest_start(skb, stg_array_type);
624 	if (!nla_stgs)
625 		/* Continue to learn diag_size */
626 		err = -EMSGSIZE;
627 
628 	saved_len = skb->len;
629 	hlist_for_each_entry_rcu(selem, &sk_storage->list, snode) {
630 		smap = rcu_dereference(SDATA(selem)->smap);
631 		diag_size += nla_value_size(smap->map.value_size);
632 
633 		if (nla_stgs && diag_get(SDATA(selem), skb))
634 			/* Continue to learn diag_size */
635 			err = -EMSGSIZE;
636 	}
637 
638 	rcu_read_unlock();
639 
640 	if (nla_stgs) {
641 		if (saved_len == skb->len)
642 			nla_nest_cancel(skb, nla_stgs);
643 		else
644 			nla_nest_end(skb, nla_stgs);
645 	}
646 
647 	if (diag_size == nla_total_size(0)) {
648 		*res_diag_size = 0;
649 		return 0;
650 	}
651 
652 	*res_diag_size = diag_size;
653 	return err;
654 }
655 
656 int bpf_sk_storage_diag_put(struct bpf_sk_storage_diag *diag,
657 			    struct sock *sk, struct sk_buff *skb,
658 			    int stg_array_type,
659 			    unsigned int *res_diag_size)
660 {
661 	/* stg_array_type (e.g. INET_DIAG_BPF_SK_STORAGES) */
662 	unsigned int diag_size = nla_total_size(0);
663 	struct bpf_local_storage *sk_storage;
664 	struct bpf_local_storage_data *sdata;
665 	struct nlattr *nla_stgs;
666 	unsigned int saved_len;
667 	int err = 0;
668 	u32 i;
669 
670 	*res_diag_size = 0;
671 
672 	/* No map has been specified.  Dump all. */
673 	if (!diag->nr_maps)
674 		return bpf_sk_storage_diag_put_all(sk, skb, stg_array_type,
675 						   res_diag_size);
676 
677 	rcu_read_lock();
678 	sk_storage = rcu_dereference(sk->sk_bpf_storage);
679 	if (!sk_storage || hlist_empty(&sk_storage->list)) {
680 		rcu_read_unlock();
681 		return 0;
682 	}
683 
684 	nla_stgs = nla_nest_start(skb, stg_array_type);
685 	if (!nla_stgs)
686 		/* Continue to learn diag_size */
687 		err = -EMSGSIZE;
688 
689 	saved_len = skb->len;
690 	for (i = 0; i < diag->nr_maps; i++) {
691 		sdata = bpf_local_storage_lookup(sk_storage,
692 				(struct bpf_local_storage_map *)diag->maps[i],
693 				false);
694 
695 		if (!sdata)
696 			continue;
697 
698 		diag_size += nla_value_size(diag->maps[i]->value_size);
699 
700 		if (nla_stgs && diag_get(sdata, skb))
701 			/* Continue to learn diag_size */
702 			err = -EMSGSIZE;
703 	}
704 	rcu_read_unlock();
705 
706 	if (nla_stgs) {
707 		if (saved_len == skb->len)
708 			nla_nest_cancel(skb, nla_stgs);
709 		else
710 			nla_nest_end(skb, nla_stgs);
711 	}
712 
713 	if (diag_size == nla_total_size(0)) {
714 		*res_diag_size = 0;
715 		return 0;
716 	}
717 
718 	*res_diag_size = diag_size;
719 	return err;
720 }
721 EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_put);
722 
723 struct bpf_iter_seq_sk_storage_map_info {
724 	struct bpf_map *map;
725 	unsigned int bucket_id;
726 	unsigned skip_elems;
727 };
728 
729 static struct bpf_local_storage_elem *
730 bpf_sk_storage_map_seq_find_next(struct bpf_iter_seq_sk_storage_map_info *info,
731 				 struct bpf_local_storage_elem *prev_selem)
732 	__acquires(RCU) __releases(RCU)
733 {
734 	struct bpf_local_storage *sk_storage;
735 	struct bpf_local_storage_elem *selem;
736 	u32 skip_elems = info->skip_elems;
737 	struct bpf_local_storage_map *smap;
738 	u32 bucket_id = info->bucket_id;
739 	u32 i, count, n_buckets;
740 	struct bpf_local_storage_map_bucket *b;
741 
742 	smap = (struct bpf_local_storage_map *)info->map;
743 	n_buckets = 1U << smap->bucket_log;
744 	if (bucket_id >= n_buckets)
745 		return NULL;
746 
747 	/* try to find next selem in the same bucket */
748 	selem = prev_selem;
749 	count = 0;
750 	while (selem) {
751 		selem = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&selem->map_node)),
752 					 struct bpf_local_storage_elem, map_node);
753 		if (!selem) {
754 			/* not found, unlock and go to the next bucket */
755 			b = &smap->buckets[bucket_id++];
756 			rcu_read_unlock();
757 			skip_elems = 0;
758 			break;
759 		}
760 		sk_storage = rcu_dereference(selem->local_storage);
761 		if (sk_storage) {
762 			info->skip_elems = skip_elems + count;
763 			return selem;
764 		}
765 		count++;
766 	}
767 
768 	for (i = bucket_id; i < (1U << smap->bucket_log); i++) {
769 		b = &smap->buckets[i];
770 		rcu_read_lock();
771 		count = 0;
772 		hlist_for_each_entry_rcu(selem, &b->list, map_node) {
773 			sk_storage = rcu_dereference(selem->local_storage);
774 			if (sk_storage && count >= skip_elems) {
775 				info->bucket_id = i;
776 				info->skip_elems = count;
777 				return selem;
778 			}
779 			count++;
780 		}
781 		rcu_read_unlock();
782 		skip_elems = 0;
783 	}
784 
785 	info->bucket_id = i;
786 	info->skip_elems = 0;
787 	return NULL;
788 }
789 
790 static void *bpf_sk_storage_map_seq_start(struct seq_file *seq, loff_t *pos)
791 {
792 	struct bpf_local_storage_elem *selem;
793 
794 	selem = bpf_sk_storage_map_seq_find_next(seq->private, NULL);
795 	if (!selem)
796 		return NULL;
797 
798 	if (*pos == 0)
799 		++*pos;
800 	return selem;
801 }
802 
803 static void *bpf_sk_storage_map_seq_next(struct seq_file *seq, void *v,
804 					 loff_t *pos)
805 {
806 	struct bpf_iter_seq_sk_storage_map_info *info = seq->private;
807 
808 	++*pos;
809 	++info->skip_elems;
810 	return bpf_sk_storage_map_seq_find_next(seq->private, v);
811 }
812 
813 struct bpf_iter__bpf_sk_storage_map {
814 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
815 	__bpf_md_ptr(struct bpf_map *, map);
816 	__bpf_md_ptr(struct sock *, sk);
817 	__bpf_md_ptr(void *, value);
818 };
819 
820 DEFINE_BPF_ITER_FUNC(bpf_sk_storage_map, struct bpf_iter_meta *meta,
821 		     struct bpf_map *map, struct sock *sk,
822 		     void *value)
823 
824 static int __bpf_sk_storage_map_seq_show(struct seq_file *seq,
825 					 struct bpf_local_storage_elem *selem)
826 {
827 	struct bpf_iter_seq_sk_storage_map_info *info = seq->private;
828 	struct bpf_iter__bpf_sk_storage_map ctx = {};
829 	struct bpf_local_storage *sk_storage;
830 	struct bpf_iter_meta meta;
831 	struct bpf_prog *prog;
832 	int ret = 0;
833 
834 	meta.seq = seq;
835 	prog = bpf_iter_get_info(&meta, selem == NULL);
836 	if (prog) {
837 		ctx.meta = &meta;
838 		ctx.map = info->map;
839 		if (selem) {
840 			sk_storage = rcu_dereference(selem->local_storage);
841 			ctx.sk = sk_storage->owner;
842 			ctx.value = SDATA(selem)->data;
843 		}
844 		ret = bpf_iter_run_prog(prog, &ctx);
845 	}
846 
847 	return ret;
848 }
849 
850 static int bpf_sk_storage_map_seq_show(struct seq_file *seq, void *v)
851 {
852 	return __bpf_sk_storage_map_seq_show(seq, v);
853 }
854 
855 static void bpf_sk_storage_map_seq_stop(struct seq_file *seq, void *v)
856 	__releases(RCU)
857 {
858 	if (!v)
859 		(void)__bpf_sk_storage_map_seq_show(seq, v);
860 	else
861 		rcu_read_unlock();
862 }
863 
864 static int bpf_iter_init_sk_storage_map(void *priv_data,
865 					struct bpf_iter_aux_info *aux)
866 {
867 	struct bpf_iter_seq_sk_storage_map_info *seq_info = priv_data;
868 
869 	seq_info->map = aux->map;
870 	return 0;
871 }
872 
873 static int bpf_iter_attach_map(struct bpf_prog *prog,
874 			       union bpf_iter_link_info *linfo,
875 			       struct bpf_iter_aux_info *aux)
876 {
877 	struct bpf_map *map;
878 	int err = -EINVAL;
879 
880 	if (!linfo->map.map_fd)
881 		return -EBADF;
882 
883 	map = bpf_map_get_with_uref(linfo->map.map_fd);
884 	if (IS_ERR(map))
885 		return PTR_ERR(map);
886 
887 	if (map->map_type != BPF_MAP_TYPE_SK_STORAGE)
888 		goto put_map;
889 
890 	if (prog->aux->max_rdonly_access > map->value_size) {
891 		err = -EACCES;
892 		goto put_map;
893 	}
894 
895 	aux->map = map;
896 	return 0;
897 
898 put_map:
899 	bpf_map_put_with_uref(map);
900 	return err;
901 }
902 
903 static void bpf_iter_detach_map(struct bpf_iter_aux_info *aux)
904 {
905 	bpf_map_put_with_uref(aux->map);
906 }
907 
908 static const struct seq_operations bpf_sk_storage_map_seq_ops = {
909 	.start  = bpf_sk_storage_map_seq_start,
910 	.next   = bpf_sk_storage_map_seq_next,
911 	.stop   = bpf_sk_storage_map_seq_stop,
912 	.show   = bpf_sk_storage_map_seq_show,
913 };
914 
915 static const struct bpf_iter_seq_info iter_seq_info = {
916 	.seq_ops		= &bpf_sk_storage_map_seq_ops,
917 	.init_seq_private	= bpf_iter_init_sk_storage_map,
918 	.fini_seq_private	= NULL,
919 	.seq_priv_size		= sizeof(struct bpf_iter_seq_sk_storage_map_info),
920 };
921 
922 static struct bpf_iter_reg bpf_sk_storage_map_reg_info = {
923 	.target			= "bpf_sk_storage_map",
924 	.attach_target		= bpf_iter_attach_map,
925 	.detach_target		= bpf_iter_detach_map,
926 	.show_fdinfo		= bpf_iter_map_show_fdinfo,
927 	.fill_link_info		= bpf_iter_map_fill_link_info,
928 	.ctx_arg_info_size	= 2,
929 	.ctx_arg_info		= {
930 		{ offsetof(struct bpf_iter__bpf_sk_storage_map, sk),
931 		  PTR_TO_BTF_ID_OR_NULL },
932 		{ offsetof(struct bpf_iter__bpf_sk_storage_map, value),
933 		  PTR_TO_RDWR_BUF_OR_NULL },
934 	},
935 	.seq_info		= &iter_seq_info,
936 };
937 
938 static int __init bpf_sk_storage_map_iter_init(void)
939 {
940 	bpf_sk_storage_map_reg_info.ctx_arg_info[0].btf_id =
941 		btf_sock_ids[BTF_SOCK_TYPE_SOCK];
942 	return bpf_iter_reg_target(&bpf_sk_storage_map_reg_info);
943 }
944 late_initcall(bpf_sk_storage_map_iter_init);
945