xref: /openbmc/linux/net/core/bpf_sk_storage.c (revision c699ce1a)
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), true);
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 *sk_storage;
52 	bool free_sk_storage = false;
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 	raw_spin_lock_bh(&sk_storage->lock);
62 	free_sk_storage = bpf_local_storage_unlink_nolock(sk_storage);
63 	raw_spin_unlock_bh(&sk_storage->lock);
64 	rcu_read_unlock();
65 
66 	if (free_sk_storage)
67 		kfree_rcu(sk_storage, rcu);
68 }
69 
70 static void bpf_sk_storage_map_free(struct bpf_map *map)
71 {
72 	bpf_local_storage_map_free(map, &sk_cache, NULL);
73 }
74 
75 static struct bpf_map *bpf_sk_storage_map_alloc(union bpf_attr *attr)
76 {
77 	return bpf_local_storage_map_alloc(attr, &sk_cache);
78 }
79 
80 static int notsupp_get_next_key(struct bpf_map *map, void *key,
81 				void *next_key)
82 {
83 	return -ENOTSUPP;
84 }
85 
86 static void *bpf_fd_sk_storage_lookup_elem(struct bpf_map *map, void *key)
87 {
88 	struct bpf_local_storage_data *sdata;
89 	struct socket *sock;
90 	int fd, err;
91 
92 	fd = *(int *)key;
93 	sock = sockfd_lookup(fd, &err);
94 	if (sock) {
95 		sdata = bpf_sk_storage_lookup(sock->sk, map, true);
96 		sockfd_put(sock);
97 		return sdata ? sdata->data : NULL;
98 	}
99 
100 	return ERR_PTR(err);
101 }
102 
103 static int bpf_fd_sk_storage_update_elem(struct bpf_map *map, void *key,
104 					 void *value, u64 map_flags)
105 {
106 	struct bpf_local_storage_data *sdata;
107 	struct socket *sock;
108 	int fd, err;
109 
110 	fd = *(int *)key;
111 	sock = sockfd_lookup(fd, &err);
112 	if (sock) {
113 		sdata = bpf_local_storage_update(
114 			sock->sk, (struct bpf_local_storage_map *)map, value,
115 			map_flags, GFP_ATOMIC);
116 		sockfd_put(sock);
117 		return PTR_ERR_OR_ZERO(sdata);
118 	}
119 
120 	return err;
121 }
122 
123 static int bpf_fd_sk_storage_delete_elem(struct bpf_map *map, void *key)
124 {
125 	struct socket *sock;
126 	int fd, err;
127 
128 	fd = *(int *)key;
129 	sock = sockfd_lookup(fd, &err);
130 	if (sock) {
131 		err = bpf_sk_storage_del(sock->sk, map);
132 		sockfd_put(sock);
133 		return err;
134 	}
135 
136 	return err;
137 }
138 
139 static struct bpf_local_storage_elem *
140 bpf_sk_storage_clone_elem(struct sock *newsk,
141 			  struct bpf_local_storage_map *smap,
142 			  struct bpf_local_storage_elem *selem)
143 {
144 	struct bpf_local_storage_elem *copy_selem;
145 
146 	copy_selem = bpf_selem_alloc(smap, newsk, NULL, true, GFP_ATOMIC);
147 	if (!copy_selem)
148 		return NULL;
149 
150 	if (btf_record_has_field(smap->map.record, BPF_SPIN_LOCK))
151 		copy_map_value_locked(&smap->map, SDATA(copy_selem)->data,
152 				      SDATA(selem)->data, true);
153 	else
154 		copy_map_value(&smap->map, SDATA(copy_selem)->data,
155 			       SDATA(selem)->data);
156 
157 	return copy_selem;
158 }
159 
160 int bpf_sk_storage_clone(const struct sock *sk, struct sock *newsk)
161 {
162 	struct bpf_local_storage *new_sk_storage = NULL;
163 	struct bpf_local_storage *sk_storage;
164 	struct bpf_local_storage_elem *selem;
165 	int ret = 0;
166 
167 	RCU_INIT_POINTER(newsk->sk_bpf_storage, NULL);
168 
169 	rcu_read_lock();
170 	sk_storage = rcu_dereference(sk->sk_bpf_storage);
171 
172 	if (!sk_storage || hlist_empty(&sk_storage->list))
173 		goto out;
174 
175 	hlist_for_each_entry_rcu(selem, &sk_storage->list, snode) {
176 		struct bpf_local_storage_elem *copy_selem;
177 		struct bpf_local_storage_map *smap;
178 		struct bpf_map *map;
179 
180 		smap = rcu_dereference(SDATA(selem)->smap);
181 		if (!(smap->map.map_flags & BPF_F_CLONE))
182 			continue;
183 
184 		/* Note that for lockless listeners adding new element
185 		 * here can race with cleanup in bpf_local_storage_map_free.
186 		 * Try to grab map refcnt to make sure that it's still
187 		 * alive and prevent concurrent removal.
188 		 */
189 		map = bpf_map_inc_not_zero(&smap->map);
190 		if (IS_ERR(map))
191 			continue;
192 
193 		copy_selem = bpf_sk_storage_clone_elem(newsk, smap, selem);
194 		if (!copy_selem) {
195 			ret = -ENOMEM;
196 			bpf_map_put(map);
197 			goto out;
198 		}
199 
200 		if (new_sk_storage) {
201 			bpf_selem_link_map(smap, copy_selem);
202 			bpf_selem_link_storage_nolock(new_sk_storage, copy_selem);
203 		} else {
204 			ret = bpf_local_storage_alloc(newsk, smap, copy_selem, GFP_ATOMIC);
205 			if (ret) {
206 				kfree(copy_selem);
207 				atomic_sub(smap->elem_size,
208 					   &newsk->sk_omem_alloc);
209 				bpf_map_put(map);
210 				goto out;
211 			}
212 
213 			new_sk_storage =
214 				rcu_dereference(copy_selem->local_storage);
215 		}
216 		bpf_map_put(map);
217 	}
218 
219 out:
220 	rcu_read_unlock();
221 
222 	/* In case of an error, don't free anything explicitly here, the
223 	 * caller is responsible to call bpf_sk_storage_free.
224 	 */
225 
226 	return ret;
227 }
228 
229 /* *gfp_flags* is a hidden argument provided by the verifier */
230 BPF_CALL_5(bpf_sk_storage_get, struct bpf_map *, map, struct sock *, sk,
231 	   void *, value, u64, flags, gfp_t, gfp_flags)
232 {
233 	struct bpf_local_storage_data *sdata;
234 
235 	WARN_ON_ONCE(!bpf_rcu_lock_held());
236 	if (!sk || !sk_fullsock(sk) || flags > BPF_SK_STORAGE_GET_F_CREATE)
237 		return (unsigned long)NULL;
238 
239 	sdata = bpf_sk_storage_lookup(sk, map, true);
240 	if (sdata)
241 		return (unsigned long)sdata->data;
242 
243 	if (flags == BPF_SK_STORAGE_GET_F_CREATE &&
244 	    /* Cannot add new elem to a going away sk.
245 	     * Otherwise, the new elem may become a leak
246 	     * (and also other memory issues during map
247 	     *  destruction).
248 	     */
249 	    refcount_inc_not_zero(&sk->sk_refcnt)) {
250 		sdata = bpf_local_storage_update(
251 			sk, (struct bpf_local_storage_map *)map, value,
252 			BPF_NOEXIST, gfp_flags);
253 		/* sk must be a fullsock (guaranteed by verifier),
254 		 * so sock_gen_put() is unnecessary.
255 		 */
256 		sock_put(sk);
257 		return IS_ERR(sdata) ?
258 			(unsigned long)NULL : (unsigned long)sdata->data;
259 	}
260 
261 	return (unsigned long)NULL;
262 }
263 
264 BPF_CALL_2(bpf_sk_storage_delete, struct bpf_map *, map, struct sock *, sk)
265 {
266 	WARN_ON_ONCE(!bpf_rcu_lock_held());
267 	if (!sk || !sk_fullsock(sk))
268 		return -EINVAL;
269 
270 	if (refcount_inc_not_zero(&sk->sk_refcnt)) {
271 		int err;
272 
273 		err = bpf_sk_storage_del(sk, map);
274 		sock_put(sk);
275 		return err;
276 	}
277 
278 	return -ENOENT;
279 }
280 
281 static int bpf_sk_storage_charge(struct bpf_local_storage_map *smap,
282 				 void *owner, u32 size)
283 {
284 	int optmem_max = READ_ONCE(sysctl_optmem_max);
285 	struct sock *sk = (struct sock *)owner;
286 
287 	/* same check as in sock_kmalloc() */
288 	if (size <= optmem_max &&
289 	    atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
290 		atomic_add(size, &sk->sk_omem_alloc);
291 		return 0;
292 	}
293 
294 	return -ENOMEM;
295 }
296 
297 static void bpf_sk_storage_uncharge(struct bpf_local_storage_map *smap,
298 				    void *owner, u32 size)
299 {
300 	struct sock *sk = owner;
301 
302 	atomic_sub(size, &sk->sk_omem_alloc);
303 }
304 
305 static struct bpf_local_storage __rcu **
306 bpf_sk_storage_ptr(void *owner)
307 {
308 	struct sock *sk = owner;
309 
310 	return &sk->sk_bpf_storage;
311 }
312 
313 const struct bpf_map_ops sk_storage_map_ops = {
314 	.map_meta_equal = bpf_map_meta_equal,
315 	.map_alloc_check = bpf_local_storage_map_alloc_check,
316 	.map_alloc = bpf_sk_storage_map_alloc,
317 	.map_free = bpf_sk_storage_map_free,
318 	.map_get_next_key = notsupp_get_next_key,
319 	.map_lookup_elem = bpf_fd_sk_storage_lookup_elem,
320 	.map_update_elem = bpf_fd_sk_storage_update_elem,
321 	.map_delete_elem = bpf_fd_sk_storage_delete_elem,
322 	.map_check_btf = bpf_local_storage_map_check_btf,
323 	.map_btf_id = &bpf_local_storage_map_btf_id[0],
324 	.map_local_storage_charge = bpf_sk_storage_charge,
325 	.map_local_storage_uncharge = bpf_sk_storage_uncharge,
326 	.map_owner_storage_ptr = bpf_sk_storage_ptr,
327 };
328 
329 const struct bpf_func_proto bpf_sk_storage_get_proto = {
330 	.func		= bpf_sk_storage_get,
331 	.gpl_only	= false,
332 	.ret_type	= RET_PTR_TO_MAP_VALUE_OR_NULL,
333 	.arg1_type	= ARG_CONST_MAP_PTR,
334 	.arg2_type	= ARG_PTR_TO_BTF_ID_SOCK_COMMON,
335 	.arg3_type	= ARG_PTR_TO_MAP_VALUE_OR_NULL,
336 	.arg4_type	= ARG_ANYTHING,
337 };
338 
339 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto = {
340 	.func		= bpf_sk_storage_get,
341 	.gpl_only	= false,
342 	.ret_type	= RET_PTR_TO_MAP_VALUE_OR_NULL,
343 	.arg1_type	= ARG_CONST_MAP_PTR,
344 	.arg2_type	= ARG_PTR_TO_CTX, /* context is 'struct sock' */
345 	.arg3_type	= ARG_PTR_TO_MAP_VALUE_OR_NULL,
346 	.arg4_type	= ARG_ANYTHING,
347 };
348 
349 const struct bpf_func_proto bpf_sk_storage_delete_proto = {
350 	.func		= bpf_sk_storage_delete,
351 	.gpl_only	= false,
352 	.ret_type	= RET_INTEGER,
353 	.arg1_type	= ARG_CONST_MAP_PTR,
354 	.arg2_type	= ARG_PTR_TO_BTF_ID_SOCK_COMMON,
355 };
356 
357 static bool bpf_sk_storage_tracing_allowed(const struct bpf_prog *prog)
358 {
359 	const struct btf *btf_vmlinux;
360 	const struct btf_type *t;
361 	const char *tname;
362 	u32 btf_id;
363 
364 	if (prog->aux->dst_prog)
365 		return false;
366 
367 	/* Ensure the tracing program is not tracing
368 	 * any bpf_sk_storage*() function and also
369 	 * use the bpf_sk_storage_(get|delete) helper.
370 	 */
371 	switch (prog->expected_attach_type) {
372 	case BPF_TRACE_ITER:
373 	case BPF_TRACE_RAW_TP:
374 		/* bpf_sk_storage has no trace point */
375 		return true;
376 	case BPF_TRACE_FENTRY:
377 	case BPF_TRACE_FEXIT:
378 		btf_vmlinux = bpf_get_btf_vmlinux();
379 		if (IS_ERR_OR_NULL(btf_vmlinux))
380 			return false;
381 		btf_id = prog->aux->attach_btf_id;
382 		t = btf_type_by_id(btf_vmlinux, btf_id);
383 		tname = btf_name_by_offset(btf_vmlinux, t->name_off);
384 		return !!strncmp(tname, "bpf_sk_storage",
385 				 strlen("bpf_sk_storage"));
386 	default:
387 		return false;
388 	}
389 
390 	return false;
391 }
392 
393 /* *gfp_flags* is a hidden argument provided by the verifier */
394 BPF_CALL_5(bpf_sk_storage_get_tracing, struct bpf_map *, map, struct sock *, sk,
395 	   void *, value, u64, flags, gfp_t, gfp_flags)
396 {
397 	WARN_ON_ONCE(!bpf_rcu_lock_held());
398 	if (in_hardirq() || in_nmi())
399 		return (unsigned long)NULL;
400 
401 	return (unsigned long)____bpf_sk_storage_get(map, sk, value, flags,
402 						     gfp_flags);
403 }
404 
405 BPF_CALL_2(bpf_sk_storage_delete_tracing, struct bpf_map *, map,
406 	   struct sock *, sk)
407 {
408 	WARN_ON_ONCE(!bpf_rcu_lock_held());
409 	if (in_hardirq() || in_nmi())
410 		return -EPERM;
411 
412 	return ____bpf_sk_storage_delete(map, sk);
413 }
414 
415 const struct bpf_func_proto bpf_sk_storage_get_tracing_proto = {
416 	.func		= bpf_sk_storage_get_tracing,
417 	.gpl_only	= false,
418 	.ret_type	= RET_PTR_TO_MAP_VALUE_OR_NULL,
419 	.arg1_type	= ARG_CONST_MAP_PTR,
420 	.arg2_type	= ARG_PTR_TO_BTF_ID,
421 	.arg2_btf_id	= &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON],
422 	.arg3_type	= ARG_PTR_TO_MAP_VALUE_OR_NULL,
423 	.arg4_type	= ARG_ANYTHING,
424 	.allowed	= bpf_sk_storage_tracing_allowed,
425 };
426 
427 const struct bpf_func_proto bpf_sk_storage_delete_tracing_proto = {
428 	.func		= bpf_sk_storage_delete_tracing,
429 	.gpl_only	= false,
430 	.ret_type	= RET_INTEGER,
431 	.arg1_type	= ARG_CONST_MAP_PTR,
432 	.arg2_type	= ARG_PTR_TO_BTF_ID,
433 	.arg2_btf_id	= &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON],
434 	.allowed	= bpf_sk_storage_tracing_allowed,
435 };
436 
437 struct bpf_sk_storage_diag {
438 	u32 nr_maps;
439 	struct bpf_map *maps[];
440 };
441 
442 /* The reply will be like:
443  * INET_DIAG_BPF_SK_STORAGES (nla_nest)
444  *	SK_DIAG_BPF_STORAGE (nla_nest)
445  *		SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32)
446  *		SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit)
447  *	SK_DIAG_BPF_STORAGE (nla_nest)
448  *		SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32)
449  *		SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit)
450  *	....
451  */
452 static int nla_value_size(u32 value_size)
453 {
454 	/* SK_DIAG_BPF_STORAGE (nla_nest)
455 	 *	SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32)
456 	 *	SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit)
457 	 */
458 	return nla_total_size(0) + nla_total_size(sizeof(u32)) +
459 		nla_total_size_64bit(value_size);
460 }
461 
462 void bpf_sk_storage_diag_free(struct bpf_sk_storage_diag *diag)
463 {
464 	u32 i;
465 
466 	if (!diag)
467 		return;
468 
469 	for (i = 0; i < diag->nr_maps; i++)
470 		bpf_map_put(diag->maps[i]);
471 
472 	kfree(diag);
473 }
474 EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_free);
475 
476 static bool diag_check_dup(const struct bpf_sk_storage_diag *diag,
477 			   const struct bpf_map *map)
478 {
479 	u32 i;
480 
481 	for (i = 0; i < diag->nr_maps; i++) {
482 		if (diag->maps[i] == map)
483 			return true;
484 	}
485 
486 	return false;
487 }
488 
489 struct bpf_sk_storage_diag *
490 bpf_sk_storage_diag_alloc(const struct nlattr *nla_stgs)
491 {
492 	struct bpf_sk_storage_diag *diag;
493 	struct nlattr *nla;
494 	u32 nr_maps = 0;
495 	int rem, err;
496 
497 	/* bpf_local_storage_map is currently limited to CAP_SYS_ADMIN as
498 	 * the map_alloc_check() side also does.
499 	 */
500 	if (!bpf_capable())
501 		return ERR_PTR(-EPERM);
502 
503 	nla_for_each_nested(nla, nla_stgs, rem) {
504 		if (nla_type(nla) == SK_DIAG_BPF_STORAGE_REQ_MAP_FD)
505 			nr_maps++;
506 	}
507 
508 	diag = kzalloc(struct_size(diag, maps, nr_maps), GFP_KERNEL);
509 	if (!diag)
510 		return ERR_PTR(-ENOMEM);
511 
512 	nla_for_each_nested(nla, nla_stgs, rem) {
513 		struct bpf_map *map;
514 		int map_fd;
515 
516 		if (nla_type(nla) != SK_DIAG_BPF_STORAGE_REQ_MAP_FD)
517 			continue;
518 
519 		map_fd = nla_get_u32(nla);
520 		map = bpf_map_get(map_fd);
521 		if (IS_ERR(map)) {
522 			err = PTR_ERR(map);
523 			goto err_free;
524 		}
525 		if (map->map_type != BPF_MAP_TYPE_SK_STORAGE) {
526 			bpf_map_put(map);
527 			err = -EINVAL;
528 			goto err_free;
529 		}
530 		if (diag_check_dup(diag, map)) {
531 			bpf_map_put(map);
532 			err = -EEXIST;
533 			goto err_free;
534 		}
535 		diag->maps[diag->nr_maps++] = map;
536 	}
537 
538 	return diag;
539 
540 err_free:
541 	bpf_sk_storage_diag_free(diag);
542 	return ERR_PTR(err);
543 }
544 EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_alloc);
545 
546 static int diag_get(struct bpf_local_storage_data *sdata, struct sk_buff *skb)
547 {
548 	struct nlattr *nla_stg, *nla_value;
549 	struct bpf_local_storage_map *smap;
550 
551 	/* It cannot exceed max nlattr's payload */
552 	BUILD_BUG_ON(U16_MAX - NLA_HDRLEN < BPF_LOCAL_STORAGE_MAX_VALUE_SIZE);
553 
554 	nla_stg = nla_nest_start(skb, SK_DIAG_BPF_STORAGE);
555 	if (!nla_stg)
556 		return -EMSGSIZE;
557 
558 	smap = rcu_dereference(sdata->smap);
559 	if (nla_put_u32(skb, SK_DIAG_BPF_STORAGE_MAP_ID, smap->map.id))
560 		goto errout;
561 
562 	nla_value = nla_reserve_64bit(skb, SK_DIAG_BPF_STORAGE_MAP_VALUE,
563 				      smap->map.value_size,
564 				      SK_DIAG_BPF_STORAGE_PAD);
565 	if (!nla_value)
566 		goto errout;
567 
568 	if (btf_record_has_field(smap->map.record, BPF_SPIN_LOCK))
569 		copy_map_value_locked(&smap->map, nla_data(nla_value),
570 				      sdata->data, true);
571 	else
572 		copy_map_value(&smap->map, nla_data(nla_value), sdata->data);
573 
574 	nla_nest_end(skb, nla_stg);
575 	return 0;
576 
577 errout:
578 	nla_nest_cancel(skb, nla_stg);
579 	return -EMSGSIZE;
580 }
581 
582 static int bpf_sk_storage_diag_put_all(struct sock *sk, struct sk_buff *skb,
583 				       int stg_array_type,
584 				       unsigned int *res_diag_size)
585 {
586 	/* stg_array_type (e.g. INET_DIAG_BPF_SK_STORAGES) */
587 	unsigned int diag_size = nla_total_size(0);
588 	struct bpf_local_storage *sk_storage;
589 	struct bpf_local_storage_elem *selem;
590 	struct bpf_local_storage_map *smap;
591 	struct nlattr *nla_stgs;
592 	unsigned int saved_len;
593 	int err = 0;
594 
595 	rcu_read_lock();
596 
597 	sk_storage = rcu_dereference(sk->sk_bpf_storage);
598 	if (!sk_storage || hlist_empty(&sk_storage->list)) {
599 		rcu_read_unlock();
600 		return 0;
601 	}
602 
603 	nla_stgs = nla_nest_start(skb, stg_array_type);
604 	if (!nla_stgs)
605 		/* Continue to learn diag_size */
606 		err = -EMSGSIZE;
607 
608 	saved_len = skb->len;
609 	hlist_for_each_entry_rcu(selem, &sk_storage->list, snode) {
610 		smap = rcu_dereference(SDATA(selem)->smap);
611 		diag_size += nla_value_size(smap->map.value_size);
612 
613 		if (nla_stgs && diag_get(SDATA(selem), skb))
614 			/* Continue to learn diag_size */
615 			err = -EMSGSIZE;
616 	}
617 
618 	rcu_read_unlock();
619 
620 	if (nla_stgs) {
621 		if (saved_len == skb->len)
622 			nla_nest_cancel(skb, nla_stgs);
623 		else
624 			nla_nest_end(skb, nla_stgs);
625 	}
626 
627 	if (diag_size == nla_total_size(0)) {
628 		*res_diag_size = 0;
629 		return 0;
630 	}
631 
632 	*res_diag_size = diag_size;
633 	return err;
634 }
635 
636 int bpf_sk_storage_diag_put(struct bpf_sk_storage_diag *diag,
637 			    struct sock *sk, struct sk_buff *skb,
638 			    int stg_array_type,
639 			    unsigned int *res_diag_size)
640 {
641 	/* stg_array_type (e.g. INET_DIAG_BPF_SK_STORAGES) */
642 	unsigned int diag_size = nla_total_size(0);
643 	struct bpf_local_storage *sk_storage;
644 	struct bpf_local_storage_data *sdata;
645 	struct nlattr *nla_stgs;
646 	unsigned int saved_len;
647 	int err = 0;
648 	u32 i;
649 
650 	*res_diag_size = 0;
651 
652 	/* No map has been specified.  Dump all. */
653 	if (!diag->nr_maps)
654 		return bpf_sk_storage_diag_put_all(sk, skb, stg_array_type,
655 						   res_diag_size);
656 
657 	rcu_read_lock();
658 	sk_storage = rcu_dereference(sk->sk_bpf_storage);
659 	if (!sk_storage || hlist_empty(&sk_storage->list)) {
660 		rcu_read_unlock();
661 		return 0;
662 	}
663 
664 	nla_stgs = nla_nest_start(skb, stg_array_type);
665 	if (!nla_stgs)
666 		/* Continue to learn diag_size */
667 		err = -EMSGSIZE;
668 
669 	saved_len = skb->len;
670 	for (i = 0; i < diag->nr_maps; i++) {
671 		sdata = bpf_local_storage_lookup(sk_storage,
672 				(struct bpf_local_storage_map *)diag->maps[i],
673 				false);
674 
675 		if (!sdata)
676 			continue;
677 
678 		diag_size += nla_value_size(diag->maps[i]->value_size);
679 
680 		if (nla_stgs && diag_get(sdata, skb))
681 			/* Continue to learn diag_size */
682 			err = -EMSGSIZE;
683 	}
684 	rcu_read_unlock();
685 
686 	if (nla_stgs) {
687 		if (saved_len == skb->len)
688 			nla_nest_cancel(skb, nla_stgs);
689 		else
690 			nla_nest_end(skb, nla_stgs);
691 	}
692 
693 	if (diag_size == nla_total_size(0)) {
694 		*res_diag_size = 0;
695 		return 0;
696 	}
697 
698 	*res_diag_size = diag_size;
699 	return err;
700 }
701 EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_put);
702 
703 struct bpf_iter_seq_sk_storage_map_info {
704 	struct bpf_map *map;
705 	unsigned int bucket_id;
706 	unsigned skip_elems;
707 };
708 
709 static struct bpf_local_storage_elem *
710 bpf_sk_storage_map_seq_find_next(struct bpf_iter_seq_sk_storage_map_info *info,
711 				 struct bpf_local_storage_elem *prev_selem)
712 	__acquires(RCU) __releases(RCU)
713 {
714 	struct bpf_local_storage *sk_storage;
715 	struct bpf_local_storage_elem *selem;
716 	u32 skip_elems = info->skip_elems;
717 	struct bpf_local_storage_map *smap;
718 	u32 bucket_id = info->bucket_id;
719 	u32 i, count, n_buckets;
720 	struct bpf_local_storage_map_bucket *b;
721 
722 	smap = (struct bpf_local_storage_map *)info->map;
723 	n_buckets = 1U << smap->bucket_log;
724 	if (bucket_id >= n_buckets)
725 		return NULL;
726 
727 	/* try to find next selem in the same bucket */
728 	selem = prev_selem;
729 	count = 0;
730 	while (selem) {
731 		selem = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&selem->map_node)),
732 					 struct bpf_local_storage_elem, map_node);
733 		if (!selem) {
734 			/* not found, unlock and go to the next bucket */
735 			b = &smap->buckets[bucket_id++];
736 			rcu_read_unlock();
737 			skip_elems = 0;
738 			break;
739 		}
740 		sk_storage = rcu_dereference(selem->local_storage);
741 		if (sk_storage) {
742 			info->skip_elems = skip_elems + count;
743 			return selem;
744 		}
745 		count++;
746 	}
747 
748 	for (i = bucket_id; i < (1U << smap->bucket_log); i++) {
749 		b = &smap->buckets[i];
750 		rcu_read_lock();
751 		count = 0;
752 		hlist_for_each_entry_rcu(selem, &b->list, map_node) {
753 			sk_storage = rcu_dereference(selem->local_storage);
754 			if (sk_storage && count >= skip_elems) {
755 				info->bucket_id = i;
756 				info->skip_elems = count;
757 				return selem;
758 			}
759 			count++;
760 		}
761 		rcu_read_unlock();
762 		skip_elems = 0;
763 	}
764 
765 	info->bucket_id = i;
766 	info->skip_elems = 0;
767 	return NULL;
768 }
769 
770 static void *bpf_sk_storage_map_seq_start(struct seq_file *seq, loff_t *pos)
771 {
772 	struct bpf_local_storage_elem *selem;
773 
774 	selem = bpf_sk_storage_map_seq_find_next(seq->private, NULL);
775 	if (!selem)
776 		return NULL;
777 
778 	if (*pos == 0)
779 		++*pos;
780 	return selem;
781 }
782 
783 static void *bpf_sk_storage_map_seq_next(struct seq_file *seq, void *v,
784 					 loff_t *pos)
785 {
786 	struct bpf_iter_seq_sk_storage_map_info *info = seq->private;
787 
788 	++*pos;
789 	++info->skip_elems;
790 	return bpf_sk_storage_map_seq_find_next(seq->private, v);
791 }
792 
793 struct bpf_iter__bpf_sk_storage_map {
794 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
795 	__bpf_md_ptr(struct bpf_map *, map);
796 	__bpf_md_ptr(struct sock *, sk);
797 	__bpf_md_ptr(void *, value);
798 };
799 
800 DEFINE_BPF_ITER_FUNC(bpf_sk_storage_map, struct bpf_iter_meta *meta,
801 		     struct bpf_map *map, struct sock *sk,
802 		     void *value)
803 
804 static int __bpf_sk_storage_map_seq_show(struct seq_file *seq,
805 					 struct bpf_local_storage_elem *selem)
806 {
807 	struct bpf_iter_seq_sk_storage_map_info *info = seq->private;
808 	struct bpf_iter__bpf_sk_storage_map ctx = {};
809 	struct bpf_local_storage *sk_storage;
810 	struct bpf_iter_meta meta;
811 	struct bpf_prog *prog;
812 	int ret = 0;
813 
814 	meta.seq = seq;
815 	prog = bpf_iter_get_info(&meta, selem == NULL);
816 	if (prog) {
817 		ctx.meta = &meta;
818 		ctx.map = info->map;
819 		if (selem) {
820 			sk_storage = rcu_dereference(selem->local_storage);
821 			ctx.sk = sk_storage->owner;
822 			ctx.value = SDATA(selem)->data;
823 		}
824 		ret = bpf_iter_run_prog(prog, &ctx);
825 	}
826 
827 	return ret;
828 }
829 
830 static int bpf_sk_storage_map_seq_show(struct seq_file *seq, void *v)
831 {
832 	return __bpf_sk_storage_map_seq_show(seq, v);
833 }
834 
835 static void bpf_sk_storage_map_seq_stop(struct seq_file *seq, void *v)
836 	__releases(RCU)
837 {
838 	if (!v)
839 		(void)__bpf_sk_storage_map_seq_show(seq, v);
840 	else
841 		rcu_read_unlock();
842 }
843 
844 static int bpf_iter_init_sk_storage_map(void *priv_data,
845 					struct bpf_iter_aux_info *aux)
846 {
847 	struct bpf_iter_seq_sk_storage_map_info *seq_info = priv_data;
848 
849 	bpf_map_inc_with_uref(aux->map);
850 	seq_info->map = aux->map;
851 	return 0;
852 }
853 
854 static void bpf_iter_fini_sk_storage_map(void *priv_data)
855 {
856 	struct bpf_iter_seq_sk_storage_map_info *seq_info = priv_data;
857 
858 	bpf_map_put_with_uref(seq_info->map);
859 }
860 
861 static int bpf_iter_attach_map(struct bpf_prog *prog,
862 			       union bpf_iter_link_info *linfo,
863 			       struct bpf_iter_aux_info *aux)
864 {
865 	struct bpf_map *map;
866 	int err = -EINVAL;
867 
868 	if (!linfo->map.map_fd)
869 		return -EBADF;
870 
871 	map = bpf_map_get_with_uref(linfo->map.map_fd);
872 	if (IS_ERR(map))
873 		return PTR_ERR(map);
874 
875 	if (map->map_type != BPF_MAP_TYPE_SK_STORAGE)
876 		goto put_map;
877 
878 	if (prog->aux->max_rdwr_access > map->value_size) {
879 		err = -EACCES;
880 		goto put_map;
881 	}
882 
883 	aux->map = map;
884 	return 0;
885 
886 put_map:
887 	bpf_map_put_with_uref(map);
888 	return err;
889 }
890 
891 static void bpf_iter_detach_map(struct bpf_iter_aux_info *aux)
892 {
893 	bpf_map_put_with_uref(aux->map);
894 }
895 
896 static const struct seq_operations bpf_sk_storage_map_seq_ops = {
897 	.start  = bpf_sk_storage_map_seq_start,
898 	.next   = bpf_sk_storage_map_seq_next,
899 	.stop   = bpf_sk_storage_map_seq_stop,
900 	.show   = bpf_sk_storage_map_seq_show,
901 };
902 
903 static const struct bpf_iter_seq_info iter_seq_info = {
904 	.seq_ops		= &bpf_sk_storage_map_seq_ops,
905 	.init_seq_private	= bpf_iter_init_sk_storage_map,
906 	.fini_seq_private	= bpf_iter_fini_sk_storage_map,
907 	.seq_priv_size		= sizeof(struct bpf_iter_seq_sk_storage_map_info),
908 };
909 
910 static struct bpf_iter_reg bpf_sk_storage_map_reg_info = {
911 	.target			= "bpf_sk_storage_map",
912 	.attach_target		= bpf_iter_attach_map,
913 	.detach_target		= bpf_iter_detach_map,
914 	.show_fdinfo		= bpf_iter_map_show_fdinfo,
915 	.fill_link_info		= bpf_iter_map_fill_link_info,
916 	.ctx_arg_info_size	= 2,
917 	.ctx_arg_info		= {
918 		{ offsetof(struct bpf_iter__bpf_sk_storage_map, sk),
919 		  PTR_TO_BTF_ID_OR_NULL },
920 		{ offsetof(struct bpf_iter__bpf_sk_storage_map, value),
921 		  PTR_TO_BUF | PTR_MAYBE_NULL },
922 	},
923 	.seq_info		= &iter_seq_info,
924 };
925 
926 static int __init bpf_sk_storage_map_iter_init(void)
927 {
928 	bpf_sk_storage_map_reg_info.ctx_arg_info[0].btf_id =
929 		btf_sock_ids[BTF_SOCK_TYPE_SOCK];
930 	return bpf_iter_reg_target(&bpf_sk_storage_map_reg_info);
931 }
932 late_initcall(bpf_sk_storage_map_iter_init);
933