xref: /openbmc/linux/net/core/bpf_sk_storage.c (revision 4b33b5ff)
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 #include <linux/rcupdate_trace.h>
17 
18 DEFINE_BPF_STORAGE_CACHE(sk_cache);
19 
20 static struct bpf_local_storage_data *
21 bpf_sk_storage_lookup(struct sock *sk, struct bpf_map *map, bool cacheit_lockit)
22 {
23 	struct bpf_local_storage *sk_storage;
24 	struct bpf_local_storage_map *smap;
25 
26 	sk_storage =
27 		rcu_dereference_check(sk->sk_bpf_storage, bpf_rcu_lock_held());
28 	if (!sk_storage)
29 		return NULL;
30 
31 	smap = (struct bpf_local_storage_map *)map;
32 	return bpf_local_storage_lookup(sk_storage, smap, cacheit_lockit);
33 }
34 
35 static int bpf_sk_storage_del(struct sock *sk, struct bpf_map *map)
36 {
37 	struct bpf_local_storage_data *sdata;
38 
39 	sdata = bpf_sk_storage_lookup(sk, map, false);
40 	if (!sdata)
41 		return -ENOENT;
42 
43 	bpf_selem_unlink(SELEM(sdata));
44 
45 	return 0;
46 }
47 
48 /* Called by __sk_destruct() & bpf_sk_storage_clone() */
49 void bpf_sk_storage_free(struct sock *sk)
50 {
51 	struct bpf_local_storage_elem *selem;
52 	struct bpf_local_storage *sk_storage;
53 	bool free_sk_storage = false;
54 	struct hlist_node *n;
55 
56 	rcu_read_lock();
57 	sk_storage = rcu_dereference(sk->sk_bpf_storage);
58 	if (!sk_storage) {
59 		rcu_read_unlock();
60 		return;
61 	}
62 
63 	/* Netiher the bpf_prog nor the bpf-map's syscall
64 	 * could be modifying the sk_storage->list now.
65 	 * Thus, no elem can be added-to or deleted-from the
66 	 * sk_storage->list by the bpf_prog or by the bpf-map's syscall.
67 	 *
68 	 * It is racing with bpf_local_storage_map_free() alone
69 	 * when unlinking elem from the sk_storage->list and
70 	 * the map's bucket->list.
71 	 */
72 	raw_spin_lock_bh(&sk_storage->lock);
73 	hlist_for_each_entry_safe(selem, n, &sk_storage->list, snode) {
74 		/* Always unlink from map before unlinking from
75 		 * sk_storage.
76 		 */
77 		bpf_selem_unlink_map(selem);
78 		free_sk_storage = bpf_selem_unlink_storage_nolock(sk_storage,
79 								  selem, true);
80 	}
81 	raw_spin_unlock_bh(&sk_storage->lock);
82 	rcu_read_unlock();
83 
84 	if (free_sk_storage)
85 		kfree_rcu(sk_storage, rcu);
86 }
87 
88 static void bpf_sk_storage_map_free(struct bpf_map *map)
89 {
90 	struct bpf_local_storage_map *smap;
91 
92 	smap = (struct bpf_local_storage_map *)map;
93 	bpf_local_storage_cache_idx_free(&sk_cache, smap->cache_idx);
94 	bpf_local_storage_map_free(smap, NULL);
95 }
96 
97 static struct bpf_map *bpf_sk_storage_map_alloc(union bpf_attr *attr)
98 {
99 	struct bpf_local_storage_map *smap;
100 
101 	smap = bpf_local_storage_map_alloc(attr);
102 	if (IS_ERR(smap))
103 		return ERR_CAST(smap);
104 
105 	smap->cache_idx = bpf_local_storage_cache_idx_get(&sk_cache);
106 	return &smap->map;
107 }
108 
109 static int notsupp_get_next_key(struct bpf_map *map, void *key,
110 				void *next_key)
111 {
112 	return -ENOTSUPP;
113 }
114 
115 static void *bpf_fd_sk_storage_lookup_elem(struct bpf_map *map, void *key)
116 {
117 	struct bpf_local_storage_data *sdata;
118 	struct socket *sock;
119 	int fd, err;
120 
121 	fd = *(int *)key;
122 	sock = sockfd_lookup(fd, &err);
123 	if (sock) {
124 		sdata = bpf_sk_storage_lookup(sock->sk, map, true);
125 		sockfd_put(sock);
126 		return sdata ? sdata->data : NULL;
127 	}
128 
129 	return ERR_PTR(err);
130 }
131 
132 static int bpf_fd_sk_storage_update_elem(struct bpf_map *map, void *key,
133 					 void *value, u64 map_flags)
134 {
135 	struct bpf_local_storage_data *sdata;
136 	struct socket *sock;
137 	int fd, err;
138 
139 	fd = *(int *)key;
140 	sock = sockfd_lookup(fd, &err);
141 	if (sock) {
142 		sdata = bpf_local_storage_update(
143 			sock->sk, (struct bpf_local_storage_map *)map, value,
144 			map_flags, GFP_ATOMIC);
145 		sockfd_put(sock);
146 		return PTR_ERR_OR_ZERO(sdata);
147 	}
148 
149 	return err;
150 }
151 
152 static int bpf_fd_sk_storage_delete_elem(struct bpf_map *map, void *key)
153 {
154 	struct socket *sock;
155 	int fd, err;
156 
157 	fd = *(int *)key;
158 	sock = sockfd_lookup(fd, &err);
159 	if (sock) {
160 		err = bpf_sk_storage_del(sock->sk, map);
161 		sockfd_put(sock);
162 		return err;
163 	}
164 
165 	return err;
166 }
167 
168 static struct bpf_local_storage_elem *
169 bpf_sk_storage_clone_elem(struct sock *newsk,
170 			  struct bpf_local_storage_map *smap,
171 			  struct bpf_local_storage_elem *selem)
172 {
173 	struct bpf_local_storage_elem *copy_selem;
174 
175 	copy_selem = bpf_selem_alloc(smap, newsk, NULL, true, GFP_ATOMIC);
176 	if (!copy_selem)
177 		return NULL;
178 
179 	if (map_value_has_spin_lock(&smap->map))
180 		copy_map_value_locked(&smap->map, SDATA(copy_selem)->data,
181 				      SDATA(selem)->data, true);
182 	else
183 		copy_map_value(&smap->map, SDATA(copy_selem)->data,
184 			       SDATA(selem)->data);
185 
186 	return copy_selem;
187 }
188 
189 int bpf_sk_storage_clone(const struct sock *sk, struct sock *newsk)
190 {
191 	struct bpf_local_storage *new_sk_storage = NULL;
192 	struct bpf_local_storage *sk_storage;
193 	struct bpf_local_storage_elem *selem;
194 	int ret = 0;
195 
196 	RCU_INIT_POINTER(newsk->sk_bpf_storage, NULL);
197 
198 	rcu_read_lock();
199 	sk_storage = rcu_dereference(sk->sk_bpf_storage);
200 
201 	if (!sk_storage || hlist_empty(&sk_storage->list))
202 		goto out;
203 
204 	hlist_for_each_entry_rcu(selem, &sk_storage->list, snode) {
205 		struct bpf_local_storage_elem *copy_selem;
206 		struct bpf_local_storage_map *smap;
207 		struct bpf_map *map;
208 
209 		smap = rcu_dereference(SDATA(selem)->smap);
210 		if (!(smap->map.map_flags & BPF_F_CLONE))
211 			continue;
212 
213 		/* Note that for lockless listeners adding new element
214 		 * here can race with cleanup in bpf_local_storage_map_free.
215 		 * Try to grab map refcnt to make sure that it's still
216 		 * alive and prevent concurrent removal.
217 		 */
218 		map = bpf_map_inc_not_zero(&smap->map);
219 		if (IS_ERR(map))
220 			continue;
221 
222 		copy_selem = bpf_sk_storage_clone_elem(newsk, smap, selem);
223 		if (!copy_selem) {
224 			ret = -ENOMEM;
225 			bpf_map_put(map);
226 			goto out;
227 		}
228 
229 		if (new_sk_storage) {
230 			bpf_selem_link_map(smap, copy_selem);
231 			bpf_selem_link_storage_nolock(new_sk_storage, copy_selem);
232 		} else {
233 			ret = bpf_local_storage_alloc(newsk, smap, copy_selem, GFP_ATOMIC);
234 			if (ret) {
235 				kfree(copy_selem);
236 				atomic_sub(smap->elem_size,
237 					   &newsk->sk_omem_alloc);
238 				bpf_map_put(map);
239 				goto out;
240 			}
241 
242 			new_sk_storage =
243 				rcu_dereference(copy_selem->local_storage);
244 		}
245 		bpf_map_put(map);
246 	}
247 
248 out:
249 	rcu_read_unlock();
250 
251 	/* In case of an error, don't free anything explicitly here, the
252 	 * caller is responsible to call bpf_sk_storage_free.
253 	 */
254 
255 	return ret;
256 }
257 
258 /* *gfp_flags* is a hidden argument provided by the verifier */
259 BPF_CALL_5(bpf_sk_storage_get, struct bpf_map *, map, struct sock *, sk,
260 	   void *, value, u64, flags, gfp_t, gfp_flags)
261 {
262 	struct bpf_local_storage_data *sdata;
263 
264 	WARN_ON_ONCE(!bpf_rcu_lock_held());
265 	if (!sk || !sk_fullsock(sk) || flags > BPF_SK_STORAGE_GET_F_CREATE)
266 		return (unsigned long)NULL;
267 
268 	sdata = bpf_sk_storage_lookup(sk, map, true);
269 	if (sdata)
270 		return (unsigned long)sdata->data;
271 
272 	if (flags == BPF_SK_STORAGE_GET_F_CREATE &&
273 	    /* Cannot add new elem to a going away sk.
274 	     * Otherwise, the new elem may become a leak
275 	     * (and also other memory issues during map
276 	     *  destruction).
277 	     */
278 	    refcount_inc_not_zero(&sk->sk_refcnt)) {
279 		sdata = bpf_local_storage_update(
280 			sk, (struct bpf_local_storage_map *)map, value,
281 			BPF_NOEXIST, gfp_flags);
282 		/* sk must be a fullsock (guaranteed by verifier),
283 		 * so sock_gen_put() is unnecessary.
284 		 */
285 		sock_put(sk);
286 		return IS_ERR(sdata) ?
287 			(unsigned long)NULL : (unsigned long)sdata->data;
288 	}
289 
290 	return (unsigned long)NULL;
291 }
292 
293 BPF_CALL_2(bpf_sk_storage_delete, struct bpf_map *, map, struct sock *, sk)
294 {
295 	WARN_ON_ONCE(!bpf_rcu_lock_held());
296 	if (!sk || !sk_fullsock(sk))
297 		return -EINVAL;
298 
299 	if (refcount_inc_not_zero(&sk->sk_refcnt)) {
300 		int err;
301 
302 		err = bpf_sk_storage_del(sk, map);
303 		sock_put(sk);
304 		return err;
305 	}
306 
307 	return -ENOENT;
308 }
309 
310 static int bpf_sk_storage_charge(struct bpf_local_storage_map *smap,
311 				 void *owner, u32 size)
312 {
313 	struct sock *sk = (struct sock *)owner;
314 
315 	/* same check as in sock_kmalloc() */
316 	if (size <= sysctl_optmem_max &&
317 	    atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
318 		atomic_add(size, &sk->sk_omem_alloc);
319 		return 0;
320 	}
321 
322 	return -ENOMEM;
323 }
324 
325 static void bpf_sk_storage_uncharge(struct bpf_local_storage_map *smap,
326 				    void *owner, u32 size)
327 {
328 	struct sock *sk = owner;
329 
330 	atomic_sub(size, &sk->sk_omem_alloc);
331 }
332 
333 static struct bpf_local_storage __rcu **
334 bpf_sk_storage_ptr(void *owner)
335 {
336 	struct sock *sk = owner;
337 
338 	return &sk->sk_bpf_storage;
339 }
340 
341 static int sk_storage_map_btf_id;
342 const struct bpf_map_ops sk_storage_map_ops = {
343 	.map_meta_equal = bpf_map_meta_equal,
344 	.map_alloc_check = bpf_local_storage_map_alloc_check,
345 	.map_alloc = bpf_sk_storage_map_alloc,
346 	.map_free = bpf_sk_storage_map_free,
347 	.map_get_next_key = notsupp_get_next_key,
348 	.map_lookup_elem = bpf_fd_sk_storage_lookup_elem,
349 	.map_update_elem = bpf_fd_sk_storage_update_elem,
350 	.map_delete_elem = bpf_fd_sk_storage_delete_elem,
351 	.map_check_btf = bpf_local_storage_map_check_btf,
352 	.map_btf_name = "bpf_local_storage_map",
353 	.map_btf_id = &sk_storage_map_btf_id,
354 	.map_local_storage_charge = bpf_sk_storage_charge,
355 	.map_local_storage_uncharge = bpf_sk_storage_uncharge,
356 	.map_owner_storage_ptr = bpf_sk_storage_ptr,
357 };
358 
359 const struct bpf_func_proto bpf_sk_storage_get_proto = {
360 	.func		= bpf_sk_storage_get,
361 	.gpl_only	= false,
362 	.ret_type	= RET_PTR_TO_MAP_VALUE_OR_NULL,
363 	.arg1_type	= ARG_CONST_MAP_PTR,
364 	.arg2_type	= ARG_PTR_TO_BTF_ID_SOCK_COMMON,
365 	.arg3_type	= ARG_PTR_TO_MAP_VALUE_OR_NULL,
366 	.arg4_type	= ARG_ANYTHING,
367 };
368 
369 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto = {
370 	.func		= bpf_sk_storage_get,
371 	.gpl_only	= false,
372 	.ret_type	= RET_PTR_TO_MAP_VALUE_OR_NULL,
373 	.arg1_type	= ARG_CONST_MAP_PTR,
374 	.arg2_type	= ARG_PTR_TO_CTX, /* context is 'struct sock' */
375 	.arg3_type	= ARG_PTR_TO_MAP_VALUE_OR_NULL,
376 	.arg4_type	= ARG_ANYTHING,
377 };
378 
379 const struct bpf_func_proto bpf_sk_storage_delete_proto = {
380 	.func		= bpf_sk_storage_delete,
381 	.gpl_only	= false,
382 	.ret_type	= RET_INTEGER,
383 	.arg1_type	= ARG_CONST_MAP_PTR,
384 	.arg2_type	= ARG_PTR_TO_BTF_ID_SOCK_COMMON,
385 };
386 
387 static bool bpf_sk_storage_tracing_allowed(const struct bpf_prog *prog)
388 {
389 	const struct btf *btf_vmlinux;
390 	const struct btf_type *t;
391 	const char *tname;
392 	u32 btf_id;
393 
394 	if (prog->aux->dst_prog)
395 		return false;
396 
397 	/* Ensure the tracing program is not tracing
398 	 * any bpf_sk_storage*() function and also
399 	 * use the bpf_sk_storage_(get|delete) helper.
400 	 */
401 	switch (prog->expected_attach_type) {
402 	case BPF_TRACE_ITER:
403 	case BPF_TRACE_RAW_TP:
404 		/* bpf_sk_storage has no trace point */
405 		return true;
406 	case BPF_TRACE_FENTRY:
407 	case BPF_TRACE_FEXIT:
408 		btf_vmlinux = bpf_get_btf_vmlinux();
409 		if (IS_ERR_OR_NULL(btf_vmlinux))
410 			return false;
411 		btf_id = prog->aux->attach_btf_id;
412 		t = btf_type_by_id(btf_vmlinux, btf_id);
413 		tname = btf_name_by_offset(btf_vmlinux, t->name_off);
414 		return !!strncmp(tname, "bpf_sk_storage",
415 				 strlen("bpf_sk_storage"));
416 	default:
417 		return false;
418 	}
419 
420 	return false;
421 }
422 
423 /* *gfp_flags* is a hidden argument provided by the verifier */
424 BPF_CALL_5(bpf_sk_storage_get_tracing, struct bpf_map *, map, struct sock *, sk,
425 	   void *, value, u64, flags, gfp_t, gfp_flags)
426 {
427 	WARN_ON_ONCE(!bpf_rcu_lock_held());
428 	if (in_hardirq() || in_nmi())
429 		return (unsigned long)NULL;
430 
431 	return (unsigned long)____bpf_sk_storage_get(map, sk, value, flags,
432 						     gfp_flags);
433 }
434 
435 BPF_CALL_2(bpf_sk_storage_delete_tracing, struct bpf_map *, map,
436 	   struct sock *, sk)
437 {
438 	WARN_ON_ONCE(!bpf_rcu_lock_held());
439 	if (in_hardirq() || in_nmi())
440 		return -EPERM;
441 
442 	return ____bpf_sk_storage_delete(map, sk);
443 }
444 
445 const struct bpf_func_proto bpf_sk_storage_get_tracing_proto = {
446 	.func		= bpf_sk_storage_get_tracing,
447 	.gpl_only	= false,
448 	.ret_type	= RET_PTR_TO_MAP_VALUE_OR_NULL,
449 	.arg1_type	= ARG_CONST_MAP_PTR,
450 	.arg2_type	= ARG_PTR_TO_BTF_ID,
451 	.arg2_btf_id	= &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON],
452 	.arg3_type	= ARG_PTR_TO_MAP_VALUE_OR_NULL,
453 	.arg4_type	= ARG_ANYTHING,
454 	.allowed	= bpf_sk_storage_tracing_allowed,
455 };
456 
457 const struct bpf_func_proto bpf_sk_storage_delete_tracing_proto = {
458 	.func		= bpf_sk_storage_delete_tracing,
459 	.gpl_only	= false,
460 	.ret_type	= RET_INTEGER,
461 	.arg1_type	= ARG_CONST_MAP_PTR,
462 	.arg2_type	= ARG_PTR_TO_BTF_ID,
463 	.arg2_btf_id	= &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON],
464 	.allowed	= bpf_sk_storage_tracing_allowed,
465 };
466 
467 struct bpf_sk_storage_diag {
468 	u32 nr_maps;
469 	struct bpf_map *maps[];
470 };
471 
472 /* The reply will be like:
473  * INET_DIAG_BPF_SK_STORAGES (nla_nest)
474  *	SK_DIAG_BPF_STORAGE (nla_nest)
475  *		SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32)
476  *		SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit)
477  *	SK_DIAG_BPF_STORAGE (nla_nest)
478  *		SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32)
479  *		SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit)
480  *	....
481  */
482 static int nla_value_size(u32 value_size)
483 {
484 	/* SK_DIAG_BPF_STORAGE (nla_nest)
485 	 *	SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32)
486 	 *	SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit)
487 	 */
488 	return nla_total_size(0) + nla_total_size(sizeof(u32)) +
489 		nla_total_size_64bit(value_size);
490 }
491 
492 void bpf_sk_storage_diag_free(struct bpf_sk_storage_diag *diag)
493 {
494 	u32 i;
495 
496 	if (!diag)
497 		return;
498 
499 	for (i = 0; i < diag->nr_maps; i++)
500 		bpf_map_put(diag->maps[i]);
501 
502 	kfree(diag);
503 }
504 EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_free);
505 
506 static bool diag_check_dup(const struct bpf_sk_storage_diag *diag,
507 			   const struct bpf_map *map)
508 {
509 	u32 i;
510 
511 	for (i = 0; i < diag->nr_maps; i++) {
512 		if (diag->maps[i] == map)
513 			return true;
514 	}
515 
516 	return false;
517 }
518 
519 struct bpf_sk_storage_diag *
520 bpf_sk_storage_diag_alloc(const struct nlattr *nla_stgs)
521 {
522 	struct bpf_sk_storage_diag *diag;
523 	struct nlattr *nla;
524 	u32 nr_maps = 0;
525 	int rem, err;
526 
527 	/* bpf_local_storage_map is currently limited to CAP_SYS_ADMIN as
528 	 * the map_alloc_check() side also does.
529 	 */
530 	if (!bpf_capable())
531 		return ERR_PTR(-EPERM);
532 
533 	nla_for_each_nested(nla, nla_stgs, rem) {
534 		if (nla_type(nla) == SK_DIAG_BPF_STORAGE_REQ_MAP_FD)
535 			nr_maps++;
536 	}
537 
538 	diag = kzalloc(struct_size(diag, maps, nr_maps), GFP_KERNEL);
539 	if (!diag)
540 		return ERR_PTR(-ENOMEM);
541 
542 	nla_for_each_nested(nla, nla_stgs, rem) {
543 		struct bpf_map *map;
544 		int map_fd;
545 
546 		if (nla_type(nla) != SK_DIAG_BPF_STORAGE_REQ_MAP_FD)
547 			continue;
548 
549 		map_fd = nla_get_u32(nla);
550 		map = bpf_map_get(map_fd);
551 		if (IS_ERR(map)) {
552 			err = PTR_ERR(map);
553 			goto err_free;
554 		}
555 		if (map->map_type != BPF_MAP_TYPE_SK_STORAGE) {
556 			bpf_map_put(map);
557 			err = -EINVAL;
558 			goto err_free;
559 		}
560 		if (diag_check_dup(diag, map)) {
561 			bpf_map_put(map);
562 			err = -EEXIST;
563 			goto err_free;
564 		}
565 		diag->maps[diag->nr_maps++] = map;
566 	}
567 
568 	return diag;
569 
570 err_free:
571 	bpf_sk_storage_diag_free(diag);
572 	return ERR_PTR(err);
573 }
574 EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_alloc);
575 
576 static int diag_get(struct bpf_local_storage_data *sdata, struct sk_buff *skb)
577 {
578 	struct nlattr *nla_stg, *nla_value;
579 	struct bpf_local_storage_map *smap;
580 
581 	/* It cannot exceed max nlattr's payload */
582 	BUILD_BUG_ON(U16_MAX - NLA_HDRLEN < BPF_LOCAL_STORAGE_MAX_VALUE_SIZE);
583 
584 	nla_stg = nla_nest_start(skb, SK_DIAG_BPF_STORAGE);
585 	if (!nla_stg)
586 		return -EMSGSIZE;
587 
588 	smap = rcu_dereference(sdata->smap);
589 	if (nla_put_u32(skb, SK_DIAG_BPF_STORAGE_MAP_ID, smap->map.id))
590 		goto errout;
591 
592 	nla_value = nla_reserve_64bit(skb, SK_DIAG_BPF_STORAGE_MAP_VALUE,
593 				      smap->map.value_size,
594 				      SK_DIAG_BPF_STORAGE_PAD);
595 	if (!nla_value)
596 		goto errout;
597 
598 	if (map_value_has_spin_lock(&smap->map))
599 		copy_map_value_locked(&smap->map, nla_data(nla_value),
600 				      sdata->data, true);
601 	else
602 		copy_map_value(&smap->map, nla_data(nla_value), sdata->data);
603 
604 	nla_nest_end(skb, nla_stg);
605 	return 0;
606 
607 errout:
608 	nla_nest_cancel(skb, nla_stg);
609 	return -EMSGSIZE;
610 }
611 
612 static int bpf_sk_storage_diag_put_all(struct sock *sk, struct sk_buff *skb,
613 				       int stg_array_type,
614 				       unsigned int *res_diag_size)
615 {
616 	/* stg_array_type (e.g. INET_DIAG_BPF_SK_STORAGES) */
617 	unsigned int diag_size = nla_total_size(0);
618 	struct bpf_local_storage *sk_storage;
619 	struct bpf_local_storage_elem *selem;
620 	struct bpf_local_storage_map *smap;
621 	struct nlattr *nla_stgs;
622 	unsigned int saved_len;
623 	int err = 0;
624 
625 	rcu_read_lock();
626 
627 	sk_storage = rcu_dereference(sk->sk_bpf_storage);
628 	if (!sk_storage || hlist_empty(&sk_storage->list)) {
629 		rcu_read_unlock();
630 		return 0;
631 	}
632 
633 	nla_stgs = nla_nest_start(skb, stg_array_type);
634 	if (!nla_stgs)
635 		/* Continue to learn diag_size */
636 		err = -EMSGSIZE;
637 
638 	saved_len = skb->len;
639 	hlist_for_each_entry_rcu(selem, &sk_storage->list, snode) {
640 		smap = rcu_dereference(SDATA(selem)->smap);
641 		diag_size += nla_value_size(smap->map.value_size);
642 
643 		if (nla_stgs && diag_get(SDATA(selem), skb))
644 			/* Continue to learn diag_size */
645 			err = -EMSGSIZE;
646 	}
647 
648 	rcu_read_unlock();
649 
650 	if (nla_stgs) {
651 		if (saved_len == skb->len)
652 			nla_nest_cancel(skb, nla_stgs);
653 		else
654 			nla_nest_end(skb, nla_stgs);
655 	}
656 
657 	if (diag_size == nla_total_size(0)) {
658 		*res_diag_size = 0;
659 		return 0;
660 	}
661 
662 	*res_diag_size = diag_size;
663 	return err;
664 }
665 
666 int bpf_sk_storage_diag_put(struct bpf_sk_storage_diag *diag,
667 			    struct sock *sk, struct sk_buff *skb,
668 			    int stg_array_type,
669 			    unsigned int *res_diag_size)
670 {
671 	/* stg_array_type (e.g. INET_DIAG_BPF_SK_STORAGES) */
672 	unsigned int diag_size = nla_total_size(0);
673 	struct bpf_local_storage *sk_storage;
674 	struct bpf_local_storage_data *sdata;
675 	struct nlattr *nla_stgs;
676 	unsigned int saved_len;
677 	int err = 0;
678 	u32 i;
679 
680 	*res_diag_size = 0;
681 
682 	/* No map has been specified.  Dump all. */
683 	if (!diag->nr_maps)
684 		return bpf_sk_storage_diag_put_all(sk, skb, stg_array_type,
685 						   res_diag_size);
686 
687 	rcu_read_lock();
688 	sk_storage = rcu_dereference(sk->sk_bpf_storage);
689 	if (!sk_storage || hlist_empty(&sk_storage->list)) {
690 		rcu_read_unlock();
691 		return 0;
692 	}
693 
694 	nla_stgs = nla_nest_start(skb, stg_array_type);
695 	if (!nla_stgs)
696 		/* Continue to learn diag_size */
697 		err = -EMSGSIZE;
698 
699 	saved_len = skb->len;
700 	for (i = 0; i < diag->nr_maps; i++) {
701 		sdata = bpf_local_storage_lookup(sk_storage,
702 				(struct bpf_local_storage_map *)diag->maps[i],
703 				false);
704 
705 		if (!sdata)
706 			continue;
707 
708 		diag_size += nla_value_size(diag->maps[i]->value_size);
709 
710 		if (nla_stgs && diag_get(sdata, skb))
711 			/* Continue to learn diag_size */
712 			err = -EMSGSIZE;
713 	}
714 	rcu_read_unlock();
715 
716 	if (nla_stgs) {
717 		if (saved_len == skb->len)
718 			nla_nest_cancel(skb, nla_stgs);
719 		else
720 			nla_nest_end(skb, nla_stgs);
721 	}
722 
723 	if (diag_size == nla_total_size(0)) {
724 		*res_diag_size = 0;
725 		return 0;
726 	}
727 
728 	*res_diag_size = diag_size;
729 	return err;
730 }
731 EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_put);
732 
733 struct bpf_iter_seq_sk_storage_map_info {
734 	struct bpf_map *map;
735 	unsigned int bucket_id;
736 	unsigned skip_elems;
737 };
738 
739 static struct bpf_local_storage_elem *
740 bpf_sk_storage_map_seq_find_next(struct bpf_iter_seq_sk_storage_map_info *info,
741 				 struct bpf_local_storage_elem *prev_selem)
742 	__acquires(RCU) __releases(RCU)
743 {
744 	struct bpf_local_storage *sk_storage;
745 	struct bpf_local_storage_elem *selem;
746 	u32 skip_elems = info->skip_elems;
747 	struct bpf_local_storage_map *smap;
748 	u32 bucket_id = info->bucket_id;
749 	u32 i, count, n_buckets;
750 	struct bpf_local_storage_map_bucket *b;
751 
752 	smap = (struct bpf_local_storage_map *)info->map;
753 	n_buckets = 1U << smap->bucket_log;
754 	if (bucket_id >= n_buckets)
755 		return NULL;
756 
757 	/* try to find next selem in the same bucket */
758 	selem = prev_selem;
759 	count = 0;
760 	while (selem) {
761 		selem = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&selem->map_node)),
762 					 struct bpf_local_storage_elem, map_node);
763 		if (!selem) {
764 			/* not found, unlock and go to the next bucket */
765 			b = &smap->buckets[bucket_id++];
766 			rcu_read_unlock();
767 			skip_elems = 0;
768 			break;
769 		}
770 		sk_storage = rcu_dereference(selem->local_storage);
771 		if (sk_storage) {
772 			info->skip_elems = skip_elems + count;
773 			return selem;
774 		}
775 		count++;
776 	}
777 
778 	for (i = bucket_id; i < (1U << smap->bucket_log); i++) {
779 		b = &smap->buckets[i];
780 		rcu_read_lock();
781 		count = 0;
782 		hlist_for_each_entry_rcu(selem, &b->list, map_node) {
783 			sk_storage = rcu_dereference(selem->local_storage);
784 			if (sk_storage && count >= skip_elems) {
785 				info->bucket_id = i;
786 				info->skip_elems = count;
787 				return selem;
788 			}
789 			count++;
790 		}
791 		rcu_read_unlock();
792 		skip_elems = 0;
793 	}
794 
795 	info->bucket_id = i;
796 	info->skip_elems = 0;
797 	return NULL;
798 }
799 
800 static void *bpf_sk_storage_map_seq_start(struct seq_file *seq, loff_t *pos)
801 {
802 	struct bpf_local_storage_elem *selem;
803 
804 	selem = bpf_sk_storage_map_seq_find_next(seq->private, NULL);
805 	if (!selem)
806 		return NULL;
807 
808 	if (*pos == 0)
809 		++*pos;
810 	return selem;
811 }
812 
813 static void *bpf_sk_storage_map_seq_next(struct seq_file *seq, void *v,
814 					 loff_t *pos)
815 {
816 	struct bpf_iter_seq_sk_storage_map_info *info = seq->private;
817 
818 	++*pos;
819 	++info->skip_elems;
820 	return bpf_sk_storage_map_seq_find_next(seq->private, v);
821 }
822 
823 struct bpf_iter__bpf_sk_storage_map {
824 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
825 	__bpf_md_ptr(struct bpf_map *, map);
826 	__bpf_md_ptr(struct sock *, sk);
827 	__bpf_md_ptr(void *, value);
828 };
829 
830 DEFINE_BPF_ITER_FUNC(bpf_sk_storage_map, struct bpf_iter_meta *meta,
831 		     struct bpf_map *map, struct sock *sk,
832 		     void *value)
833 
834 static int __bpf_sk_storage_map_seq_show(struct seq_file *seq,
835 					 struct bpf_local_storage_elem *selem)
836 {
837 	struct bpf_iter_seq_sk_storage_map_info *info = seq->private;
838 	struct bpf_iter__bpf_sk_storage_map ctx = {};
839 	struct bpf_local_storage *sk_storage;
840 	struct bpf_iter_meta meta;
841 	struct bpf_prog *prog;
842 	int ret = 0;
843 
844 	meta.seq = seq;
845 	prog = bpf_iter_get_info(&meta, selem == NULL);
846 	if (prog) {
847 		ctx.meta = &meta;
848 		ctx.map = info->map;
849 		if (selem) {
850 			sk_storage = rcu_dereference(selem->local_storage);
851 			ctx.sk = sk_storage->owner;
852 			ctx.value = SDATA(selem)->data;
853 		}
854 		ret = bpf_iter_run_prog(prog, &ctx);
855 	}
856 
857 	return ret;
858 }
859 
860 static int bpf_sk_storage_map_seq_show(struct seq_file *seq, void *v)
861 {
862 	return __bpf_sk_storage_map_seq_show(seq, v);
863 }
864 
865 static void bpf_sk_storage_map_seq_stop(struct seq_file *seq, void *v)
866 	__releases(RCU)
867 {
868 	if (!v)
869 		(void)__bpf_sk_storage_map_seq_show(seq, v);
870 	else
871 		rcu_read_unlock();
872 }
873 
874 static int bpf_iter_init_sk_storage_map(void *priv_data,
875 					struct bpf_iter_aux_info *aux)
876 {
877 	struct bpf_iter_seq_sk_storage_map_info *seq_info = priv_data;
878 
879 	seq_info->map = aux->map;
880 	return 0;
881 }
882 
883 static int bpf_iter_attach_map(struct bpf_prog *prog,
884 			       union bpf_iter_link_info *linfo,
885 			       struct bpf_iter_aux_info *aux)
886 {
887 	struct bpf_map *map;
888 	int err = -EINVAL;
889 
890 	if (!linfo->map.map_fd)
891 		return -EBADF;
892 
893 	map = bpf_map_get_with_uref(linfo->map.map_fd);
894 	if (IS_ERR(map))
895 		return PTR_ERR(map);
896 
897 	if (map->map_type != BPF_MAP_TYPE_SK_STORAGE)
898 		goto put_map;
899 
900 	if (prog->aux->max_rdonly_access > map->value_size) {
901 		err = -EACCES;
902 		goto put_map;
903 	}
904 
905 	aux->map = map;
906 	return 0;
907 
908 put_map:
909 	bpf_map_put_with_uref(map);
910 	return err;
911 }
912 
913 static void bpf_iter_detach_map(struct bpf_iter_aux_info *aux)
914 {
915 	bpf_map_put_with_uref(aux->map);
916 }
917 
918 static const struct seq_operations bpf_sk_storage_map_seq_ops = {
919 	.start  = bpf_sk_storage_map_seq_start,
920 	.next   = bpf_sk_storage_map_seq_next,
921 	.stop   = bpf_sk_storage_map_seq_stop,
922 	.show   = bpf_sk_storage_map_seq_show,
923 };
924 
925 static const struct bpf_iter_seq_info iter_seq_info = {
926 	.seq_ops		= &bpf_sk_storage_map_seq_ops,
927 	.init_seq_private	= bpf_iter_init_sk_storage_map,
928 	.fini_seq_private	= NULL,
929 	.seq_priv_size		= sizeof(struct bpf_iter_seq_sk_storage_map_info),
930 };
931 
932 static struct bpf_iter_reg bpf_sk_storage_map_reg_info = {
933 	.target			= "bpf_sk_storage_map",
934 	.attach_target		= bpf_iter_attach_map,
935 	.detach_target		= bpf_iter_detach_map,
936 	.show_fdinfo		= bpf_iter_map_show_fdinfo,
937 	.fill_link_info		= bpf_iter_map_fill_link_info,
938 	.ctx_arg_info_size	= 2,
939 	.ctx_arg_info		= {
940 		{ offsetof(struct bpf_iter__bpf_sk_storage_map, sk),
941 		  PTR_TO_BTF_ID_OR_NULL },
942 		{ offsetof(struct bpf_iter__bpf_sk_storage_map, value),
943 		  PTR_TO_BUF | PTR_MAYBE_NULL },
944 	},
945 	.seq_info		= &iter_seq_info,
946 };
947 
948 static int __init bpf_sk_storage_map_iter_init(void)
949 {
950 	bpf_sk_storage_map_reg_info.ctx_arg_info[0].btf_id =
951 		btf_sock_ids[BTF_SOCK_TYPE_SOCK];
952 	return bpf_iter_reg_target(&bpf_sk_storage_map_reg_info);
953 }
954 late_initcall(bpf_sk_storage_map_iter_init);
955