xref: /linux/net/core/bpf_sk_storage.c (revision 3f41368fbfe1b3d5922d317fe1a0a0cab6846802)
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), false);
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 
53 	rcu_read_lock();
54 	sk_storage = rcu_dereference(sk->sk_bpf_storage);
55 	if (!sk_storage) {
56 		rcu_read_unlock();
57 		return;
58 	}
59 
60 	bpf_local_storage_destroy(sk_storage);
61 	rcu_read_unlock();
62 }
63 
64 static void bpf_sk_storage_map_free(struct bpf_map *map)
65 {
66 	bpf_local_storage_map_free(map, &sk_cache, NULL);
67 }
68 
69 static struct bpf_map *bpf_sk_storage_map_alloc(union bpf_attr *attr)
70 {
71 	return bpf_local_storage_map_alloc(attr, &sk_cache, false);
72 }
73 
74 static int notsupp_get_next_key(struct bpf_map *map, void *key,
75 				void *next_key)
76 {
77 	return -ENOTSUPP;
78 }
79 
80 static void *bpf_fd_sk_storage_lookup_elem(struct bpf_map *map, void *key)
81 {
82 	struct bpf_local_storage_data *sdata;
83 	struct socket *sock;
84 	int fd, err;
85 
86 	fd = *(int *)key;
87 	sock = sockfd_lookup(fd, &err);
88 	if (sock) {
89 		sdata = bpf_sk_storage_lookup(sock->sk, map, true);
90 		sockfd_put(sock);
91 		return sdata ? sdata->data : NULL;
92 	}
93 
94 	return ERR_PTR(err);
95 }
96 
97 static long bpf_fd_sk_storage_update_elem(struct bpf_map *map, void *key,
98 					  void *value, u64 map_flags)
99 {
100 	struct bpf_local_storage_data *sdata;
101 	struct socket *sock;
102 	int fd, err;
103 
104 	fd = *(int *)key;
105 	sock = sockfd_lookup(fd, &err);
106 	if (sock) {
107 		sdata = bpf_local_storage_update(
108 			sock->sk, (struct bpf_local_storage_map *)map, value,
109 			map_flags, GFP_ATOMIC);
110 		sockfd_put(sock);
111 		return PTR_ERR_OR_ZERO(sdata);
112 	}
113 
114 	return err;
115 }
116 
117 static long bpf_fd_sk_storage_delete_elem(struct bpf_map *map, void *key)
118 {
119 	struct socket *sock;
120 	int fd, err;
121 
122 	fd = *(int *)key;
123 	sock = sockfd_lookup(fd, &err);
124 	if (sock) {
125 		err = bpf_sk_storage_del(sock->sk, map);
126 		sockfd_put(sock);
127 		return err;
128 	}
129 
130 	return err;
131 }
132 
133 static struct bpf_local_storage_elem *
134 bpf_sk_storage_clone_elem(struct sock *newsk,
135 			  struct bpf_local_storage_map *smap,
136 			  struct bpf_local_storage_elem *selem)
137 {
138 	struct bpf_local_storage_elem *copy_selem;
139 
140 	copy_selem = bpf_selem_alloc(smap, newsk, NULL, true, GFP_ATOMIC);
141 	if (!copy_selem)
142 		return NULL;
143 
144 	if (btf_record_has_field(smap->map.record, BPF_SPIN_LOCK))
145 		copy_map_value_locked(&smap->map, SDATA(copy_selem)->data,
146 				      SDATA(selem)->data, true);
147 	else
148 		copy_map_value(&smap->map, SDATA(copy_selem)->data,
149 			       SDATA(selem)->data);
150 
151 	return copy_selem;
152 }
153 
154 int bpf_sk_storage_clone(const struct sock *sk, struct sock *newsk)
155 {
156 	struct bpf_local_storage *new_sk_storage = NULL;
157 	struct bpf_local_storage *sk_storage;
158 	struct bpf_local_storage_elem *selem;
159 	int ret = 0;
160 
161 	RCU_INIT_POINTER(newsk->sk_bpf_storage, NULL);
162 
163 	rcu_read_lock();
164 	sk_storage = rcu_dereference(sk->sk_bpf_storage);
165 
166 	if (!sk_storage || hlist_empty(&sk_storage->list))
167 		goto out;
168 
169 	hlist_for_each_entry_rcu(selem, &sk_storage->list, snode) {
170 		struct bpf_local_storage_elem *copy_selem;
171 		struct bpf_local_storage_map *smap;
172 		struct bpf_map *map;
173 
174 		smap = rcu_dereference(SDATA(selem)->smap);
175 		if (!(smap->map.map_flags & BPF_F_CLONE))
176 			continue;
177 
178 		/* Note that for lockless listeners adding new element
179 		 * here can race with cleanup in bpf_local_storage_map_free.
180 		 * Try to grab map refcnt to make sure that it's still
181 		 * alive and prevent concurrent removal.
182 		 */
183 		map = bpf_map_inc_not_zero(&smap->map);
184 		if (IS_ERR(map))
185 			continue;
186 
187 		copy_selem = bpf_sk_storage_clone_elem(newsk, smap, selem);
188 		if (!copy_selem) {
189 			ret = -ENOMEM;
190 			bpf_map_put(map);
191 			goto out;
192 		}
193 
194 		if (new_sk_storage) {
195 			bpf_selem_link_map(smap, copy_selem);
196 			bpf_selem_link_storage_nolock(new_sk_storage, copy_selem);
197 		} else {
198 			ret = bpf_local_storage_alloc(newsk, smap, copy_selem, GFP_ATOMIC);
199 			if (ret) {
200 				bpf_selem_free(copy_selem, smap, true);
201 				atomic_sub(smap->elem_size,
202 					   &newsk->sk_omem_alloc);
203 				bpf_map_put(map);
204 				goto out;
205 			}
206 
207 			new_sk_storage =
208 				rcu_dereference(copy_selem->local_storage);
209 		}
210 		bpf_map_put(map);
211 	}
212 
213 out:
214 	rcu_read_unlock();
215 
216 	/* In case of an error, don't free anything explicitly here, the
217 	 * caller is responsible to call bpf_sk_storage_free.
218 	 */
219 
220 	return ret;
221 }
222 
223 /* *gfp_flags* is a hidden argument provided by the verifier */
224 BPF_CALL_5(bpf_sk_storage_get, struct bpf_map *, map, struct sock *, sk,
225 	   void *, value, u64, flags, gfp_t, gfp_flags)
226 {
227 	struct bpf_local_storage_data *sdata;
228 
229 	WARN_ON_ONCE(!bpf_rcu_lock_held());
230 	if (!sk || !sk_fullsock(sk) || flags > BPF_SK_STORAGE_GET_F_CREATE)
231 		return (unsigned long)NULL;
232 
233 	sdata = bpf_sk_storage_lookup(sk, map, true);
234 	if (sdata)
235 		return (unsigned long)sdata->data;
236 
237 	if (flags == BPF_SK_STORAGE_GET_F_CREATE &&
238 	    /* Cannot add new elem to a going away sk.
239 	     * Otherwise, the new elem may become a leak
240 	     * (and also other memory issues during map
241 	     *  destruction).
242 	     */
243 	    refcount_inc_not_zero(&sk->sk_refcnt)) {
244 		sdata = bpf_local_storage_update(
245 			sk, (struct bpf_local_storage_map *)map, value,
246 			BPF_NOEXIST, gfp_flags);
247 		/* sk must be a fullsock (guaranteed by verifier),
248 		 * so sock_gen_put() is unnecessary.
249 		 */
250 		sock_put(sk);
251 		return IS_ERR(sdata) ?
252 			(unsigned long)NULL : (unsigned long)sdata->data;
253 	}
254 
255 	return (unsigned long)NULL;
256 }
257 
258 BPF_CALL_2(bpf_sk_storage_delete, struct bpf_map *, map, struct sock *, sk)
259 {
260 	WARN_ON_ONCE(!bpf_rcu_lock_held());
261 	if (!sk || !sk_fullsock(sk))
262 		return -EINVAL;
263 
264 	if (refcount_inc_not_zero(&sk->sk_refcnt)) {
265 		int err;
266 
267 		err = bpf_sk_storage_del(sk, map);
268 		sock_put(sk);
269 		return err;
270 	}
271 
272 	return -ENOENT;
273 }
274 
275 static int bpf_sk_storage_charge(struct bpf_local_storage_map *smap,
276 				 void *owner, u32 size)
277 {
278 	struct sock *sk = (struct sock *)owner;
279 	int optmem_max;
280 
281 	optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
282 	/* same check as in sock_kmalloc() */
283 	if (size <= optmem_max &&
284 	    atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
285 		atomic_add(size, &sk->sk_omem_alloc);
286 		return 0;
287 	}
288 
289 	return -ENOMEM;
290 }
291 
292 static void bpf_sk_storage_uncharge(struct bpf_local_storage_map *smap,
293 				    void *owner, u32 size)
294 {
295 	struct sock *sk = owner;
296 
297 	atomic_sub(size, &sk->sk_omem_alloc);
298 }
299 
300 static struct bpf_local_storage __rcu **
301 bpf_sk_storage_ptr(void *owner)
302 {
303 	struct sock *sk = owner;
304 
305 	return &sk->sk_bpf_storage;
306 }
307 
308 const struct bpf_map_ops sk_storage_map_ops = {
309 	.map_meta_equal = bpf_map_meta_equal,
310 	.map_alloc_check = bpf_local_storage_map_alloc_check,
311 	.map_alloc = bpf_sk_storage_map_alloc,
312 	.map_free = bpf_sk_storage_map_free,
313 	.map_get_next_key = notsupp_get_next_key,
314 	.map_lookup_elem = bpf_fd_sk_storage_lookup_elem,
315 	.map_update_elem = bpf_fd_sk_storage_update_elem,
316 	.map_delete_elem = bpf_fd_sk_storage_delete_elem,
317 	.map_check_btf = bpf_local_storage_map_check_btf,
318 	.map_btf_id = &bpf_local_storage_map_btf_id[0],
319 	.map_local_storage_charge = bpf_sk_storage_charge,
320 	.map_local_storage_uncharge = bpf_sk_storage_uncharge,
321 	.map_owner_storage_ptr = bpf_sk_storage_ptr,
322 	.map_mem_usage = bpf_local_storage_map_mem_usage,
323 };
324 
325 const struct bpf_func_proto bpf_sk_storage_get_proto = {
326 	.func		= bpf_sk_storage_get,
327 	.gpl_only	= false,
328 	.ret_type	= RET_PTR_TO_MAP_VALUE_OR_NULL,
329 	.arg1_type	= ARG_CONST_MAP_PTR,
330 	.arg2_type	= ARG_PTR_TO_BTF_ID_SOCK_COMMON,
331 	.arg3_type	= ARG_PTR_TO_MAP_VALUE_OR_NULL,
332 	.arg4_type	= ARG_ANYTHING,
333 };
334 
335 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto = {
336 	.func		= bpf_sk_storage_get,
337 	.gpl_only	= false,
338 	.ret_type	= RET_PTR_TO_MAP_VALUE_OR_NULL,
339 	.arg1_type	= ARG_CONST_MAP_PTR,
340 	.arg2_type	= ARG_PTR_TO_CTX, /* context is 'struct sock' */
341 	.arg3_type	= ARG_PTR_TO_MAP_VALUE_OR_NULL,
342 	.arg4_type	= ARG_ANYTHING,
343 };
344 
345 const struct bpf_func_proto bpf_sk_storage_delete_proto = {
346 	.func		= bpf_sk_storage_delete,
347 	.gpl_only	= false,
348 	.ret_type	= RET_INTEGER,
349 	.arg1_type	= ARG_CONST_MAP_PTR,
350 	.arg2_type	= ARG_PTR_TO_BTF_ID_SOCK_COMMON,
351 };
352 
353 static bool bpf_sk_storage_tracing_allowed(const struct bpf_prog *prog)
354 {
355 	const struct btf *btf_vmlinux;
356 	const struct btf_type *t;
357 	const char *tname;
358 	u32 btf_id;
359 
360 	if (prog->aux->dst_prog)
361 		return false;
362 
363 	/* Ensure the tracing program is not tracing
364 	 * any bpf_sk_storage*() function and also
365 	 * use the bpf_sk_storage_(get|delete) helper.
366 	 */
367 	switch (prog->expected_attach_type) {
368 	case BPF_TRACE_ITER:
369 	case BPF_TRACE_RAW_TP:
370 		/* bpf_sk_storage has no trace point */
371 		return true;
372 	case BPF_TRACE_FENTRY:
373 	case BPF_TRACE_FEXIT:
374 		btf_vmlinux = bpf_get_btf_vmlinux();
375 		if (IS_ERR_OR_NULL(btf_vmlinux))
376 			return false;
377 		btf_id = prog->aux->attach_btf_id;
378 		t = btf_type_by_id(btf_vmlinux, btf_id);
379 		tname = btf_name_by_offset(btf_vmlinux, t->name_off);
380 		return !!strncmp(tname, "bpf_sk_storage",
381 				 strlen("bpf_sk_storage"));
382 	default:
383 		return false;
384 	}
385 
386 	return false;
387 }
388 
389 /* *gfp_flags* is a hidden argument provided by the verifier */
390 BPF_CALL_5(bpf_sk_storage_get_tracing, struct bpf_map *, map, struct sock *, sk,
391 	   void *, value, u64, flags, gfp_t, gfp_flags)
392 {
393 	WARN_ON_ONCE(!bpf_rcu_lock_held());
394 	if (in_hardirq() || in_nmi())
395 		return (unsigned long)NULL;
396 
397 	return (unsigned long)____bpf_sk_storage_get(map, sk, value, flags,
398 						     gfp_flags);
399 }
400 
401 BPF_CALL_2(bpf_sk_storage_delete_tracing, struct bpf_map *, map,
402 	   struct sock *, sk)
403 {
404 	WARN_ON_ONCE(!bpf_rcu_lock_held());
405 	if (in_hardirq() || in_nmi())
406 		return -EPERM;
407 
408 	return ____bpf_sk_storage_delete(map, sk);
409 }
410 
411 const struct bpf_func_proto bpf_sk_storage_get_tracing_proto = {
412 	.func		= bpf_sk_storage_get_tracing,
413 	.gpl_only	= false,
414 	.ret_type	= RET_PTR_TO_MAP_VALUE_OR_NULL,
415 	.arg1_type	= ARG_CONST_MAP_PTR,
416 	.arg2_type	= ARG_PTR_TO_BTF_ID_OR_NULL,
417 	.arg2_btf_id	= &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON],
418 	.arg3_type	= ARG_PTR_TO_MAP_VALUE_OR_NULL,
419 	.arg4_type	= ARG_ANYTHING,
420 	.allowed	= bpf_sk_storage_tracing_allowed,
421 };
422 
423 const struct bpf_func_proto bpf_sk_storage_delete_tracing_proto = {
424 	.func		= bpf_sk_storage_delete_tracing,
425 	.gpl_only	= false,
426 	.ret_type	= RET_INTEGER,
427 	.arg1_type	= ARG_CONST_MAP_PTR,
428 	.arg2_type	= ARG_PTR_TO_BTF_ID_OR_NULL,
429 	.arg2_btf_id	= &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON],
430 	.allowed	= bpf_sk_storage_tracing_allowed,
431 };
432 
433 struct bpf_sk_storage_diag {
434 	u32 nr_maps;
435 	struct bpf_map *maps[];
436 };
437 
438 /* The reply will be like:
439  * INET_DIAG_BPF_SK_STORAGES (nla_nest)
440  *	SK_DIAG_BPF_STORAGE (nla_nest)
441  *		SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32)
442  *		SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit)
443  *	SK_DIAG_BPF_STORAGE (nla_nest)
444  *		SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32)
445  *		SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit)
446  *	....
447  */
448 static int nla_value_size(u32 value_size)
449 {
450 	/* SK_DIAG_BPF_STORAGE (nla_nest)
451 	 *	SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32)
452 	 *	SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit)
453 	 */
454 	return nla_total_size(0) + nla_total_size(sizeof(u32)) +
455 		nla_total_size_64bit(value_size);
456 }
457 
458 void bpf_sk_storage_diag_free(struct bpf_sk_storage_diag *diag)
459 {
460 	u32 i;
461 
462 	if (!diag)
463 		return;
464 
465 	for (i = 0; i < diag->nr_maps; i++)
466 		bpf_map_put(diag->maps[i]);
467 
468 	kfree(diag);
469 }
470 EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_free);
471 
472 static bool diag_check_dup(const struct bpf_sk_storage_diag *diag,
473 			   const struct bpf_map *map)
474 {
475 	u32 i;
476 
477 	for (i = 0; i < diag->nr_maps; i++) {
478 		if (diag->maps[i] == map)
479 			return true;
480 	}
481 
482 	return false;
483 }
484 
485 struct bpf_sk_storage_diag *
486 bpf_sk_storage_diag_alloc(const struct nlattr *nla_stgs)
487 {
488 	struct bpf_sk_storage_diag *diag;
489 	struct nlattr *nla;
490 	u32 nr_maps = 0;
491 	int rem, err;
492 
493 	/* bpf_local_storage_map is currently limited to CAP_SYS_ADMIN as
494 	 * the map_alloc_check() side also does.
495 	 */
496 	if (!bpf_capable())
497 		return ERR_PTR(-EPERM);
498 
499 	nla_for_each_nested_type(nla, SK_DIAG_BPF_STORAGE_REQ_MAP_FD,
500 				 nla_stgs, rem) {
501 		if (nla_len(nla) != sizeof(u32))
502 			return ERR_PTR(-EINVAL);
503 		nr_maps++;
504 	}
505 
506 	diag = kzalloc(struct_size(diag, maps, nr_maps), GFP_KERNEL);
507 	if (!diag)
508 		return ERR_PTR(-ENOMEM);
509 
510 	nla_for_each_nested_type(nla, SK_DIAG_BPF_STORAGE_REQ_MAP_FD,
511 				 nla_stgs, rem) {
512 		int map_fd = nla_get_u32(nla);
513 		struct bpf_map *map = bpf_map_get(map_fd);
514 
515 		if (IS_ERR(map)) {
516 			err = PTR_ERR(map);
517 			goto err_free;
518 		}
519 		if (map->map_type != BPF_MAP_TYPE_SK_STORAGE) {
520 			bpf_map_put(map);
521 			err = -EINVAL;
522 			goto err_free;
523 		}
524 		if (diag_check_dup(diag, map)) {
525 			bpf_map_put(map);
526 			err = -EEXIST;
527 			goto err_free;
528 		}
529 		diag->maps[diag->nr_maps++] = map;
530 	}
531 
532 	return diag;
533 
534 err_free:
535 	bpf_sk_storage_diag_free(diag);
536 	return ERR_PTR(err);
537 }
538 EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_alloc);
539 
540 static int diag_get(struct bpf_local_storage_data *sdata, struct sk_buff *skb)
541 {
542 	struct nlattr *nla_stg, *nla_value;
543 	struct bpf_local_storage_map *smap;
544 
545 	/* It cannot exceed max nlattr's payload */
546 	BUILD_BUG_ON(U16_MAX - NLA_HDRLEN < BPF_LOCAL_STORAGE_MAX_VALUE_SIZE);
547 
548 	nla_stg = nla_nest_start(skb, SK_DIAG_BPF_STORAGE);
549 	if (!nla_stg)
550 		return -EMSGSIZE;
551 
552 	smap = rcu_dereference(sdata->smap);
553 	if (nla_put_u32(skb, SK_DIAG_BPF_STORAGE_MAP_ID, smap->map.id))
554 		goto errout;
555 
556 	nla_value = nla_reserve_64bit(skb, SK_DIAG_BPF_STORAGE_MAP_VALUE,
557 				      smap->map.value_size,
558 				      SK_DIAG_BPF_STORAGE_PAD);
559 	if (!nla_value)
560 		goto errout;
561 
562 	if (btf_record_has_field(smap->map.record, BPF_SPIN_LOCK))
563 		copy_map_value_locked(&smap->map, nla_data(nla_value),
564 				      sdata->data, true);
565 	else
566 		copy_map_value(&smap->map, nla_data(nla_value), sdata->data);
567 
568 	nla_nest_end(skb, nla_stg);
569 	return 0;
570 
571 errout:
572 	nla_nest_cancel(skb, nla_stg);
573 	return -EMSGSIZE;
574 }
575 
576 static int bpf_sk_storage_diag_put_all(struct sock *sk, struct sk_buff *skb,
577 				       int stg_array_type,
578 				       unsigned int *res_diag_size)
579 {
580 	/* stg_array_type (e.g. INET_DIAG_BPF_SK_STORAGES) */
581 	unsigned int diag_size = nla_total_size(0);
582 	struct bpf_local_storage *sk_storage;
583 	struct bpf_local_storage_elem *selem;
584 	struct bpf_local_storage_map *smap;
585 	struct nlattr *nla_stgs;
586 	unsigned int saved_len;
587 	int err = 0;
588 
589 	rcu_read_lock();
590 
591 	sk_storage = rcu_dereference(sk->sk_bpf_storage);
592 	if (!sk_storage || hlist_empty(&sk_storage->list)) {
593 		rcu_read_unlock();
594 		return 0;
595 	}
596 
597 	nla_stgs = nla_nest_start(skb, stg_array_type);
598 	if (!nla_stgs)
599 		/* Continue to learn diag_size */
600 		err = -EMSGSIZE;
601 
602 	saved_len = skb->len;
603 	hlist_for_each_entry_rcu(selem, &sk_storage->list, snode) {
604 		smap = rcu_dereference(SDATA(selem)->smap);
605 		diag_size += nla_value_size(smap->map.value_size);
606 
607 		if (nla_stgs && diag_get(SDATA(selem), skb))
608 			/* Continue to learn diag_size */
609 			err = -EMSGSIZE;
610 	}
611 
612 	rcu_read_unlock();
613 
614 	if (nla_stgs) {
615 		if (saved_len == skb->len)
616 			nla_nest_cancel(skb, nla_stgs);
617 		else
618 			nla_nest_end(skb, nla_stgs);
619 	}
620 
621 	if (diag_size == nla_total_size(0)) {
622 		*res_diag_size = 0;
623 		return 0;
624 	}
625 
626 	*res_diag_size = diag_size;
627 	return err;
628 }
629 
630 int bpf_sk_storage_diag_put(struct bpf_sk_storage_diag *diag,
631 			    struct sock *sk, struct sk_buff *skb,
632 			    int stg_array_type,
633 			    unsigned int *res_diag_size)
634 {
635 	/* stg_array_type (e.g. INET_DIAG_BPF_SK_STORAGES) */
636 	unsigned int diag_size = nla_total_size(0);
637 	struct bpf_local_storage *sk_storage;
638 	struct bpf_local_storage_data *sdata;
639 	struct nlattr *nla_stgs;
640 	unsigned int saved_len;
641 	int err = 0;
642 	u32 i;
643 
644 	*res_diag_size = 0;
645 
646 	/* No map has been specified.  Dump all. */
647 	if (!diag->nr_maps)
648 		return bpf_sk_storage_diag_put_all(sk, skb, stg_array_type,
649 						   res_diag_size);
650 
651 	rcu_read_lock();
652 	sk_storage = rcu_dereference(sk->sk_bpf_storage);
653 	if (!sk_storage || hlist_empty(&sk_storage->list)) {
654 		rcu_read_unlock();
655 		return 0;
656 	}
657 
658 	nla_stgs = nla_nest_start(skb, stg_array_type);
659 	if (!nla_stgs)
660 		/* Continue to learn diag_size */
661 		err = -EMSGSIZE;
662 
663 	saved_len = skb->len;
664 	for (i = 0; i < diag->nr_maps; i++) {
665 		sdata = bpf_local_storage_lookup(sk_storage,
666 				(struct bpf_local_storage_map *)diag->maps[i],
667 				false);
668 
669 		if (!sdata)
670 			continue;
671 
672 		diag_size += nla_value_size(diag->maps[i]->value_size);
673 
674 		if (nla_stgs && diag_get(sdata, skb))
675 			/* Continue to learn diag_size */
676 			err = -EMSGSIZE;
677 	}
678 	rcu_read_unlock();
679 
680 	if (nla_stgs) {
681 		if (saved_len == skb->len)
682 			nla_nest_cancel(skb, nla_stgs);
683 		else
684 			nla_nest_end(skb, nla_stgs);
685 	}
686 
687 	if (diag_size == nla_total_size(0)) {
688 		*res_diag_size = 0;
689 		return 0;
690 	}
691 
692 	*res_diag_size = diag_size;
693 	return err;
694 }
695 EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_put);
696 
697 struct bpf_iter_seq_sk_storage_map_info {
698 	struct bpf_map *map;
699 	unsigned int bucket_id;
700 	unsigned skip_elems;
701 };
702 
703 static struct bpf_local_storage_elem *
704 bpf_sk_storage_map_seq_find_next(struct bpf_iter_seq_sk_storage_map_info *info,
705 				 struct bpf_local_storage_elem *prev_selem)
706 	__acquires(RCU) __releases(RCU)
707 {
708 	struct bpf_local_storage *sk_storage;
709 	struct bpf_local_storage_elem *selem;
710 	u32 skip_elems = info->skip_elems;
711 	struct bpf_local_storage_map *smap;
712 	u32 bucket_id = info->bucket_id;
713 	u32 i, count, n_buckets;
714 	struct bpf_local_storage_map_bucket *b;
715 
716 	smap = (struct bpf_local_storage_map *)info->map;
717 	n_buckets = 1U << smap->bucket_log;
718 	if (bucket_id >= n_buckets)
719 		return NULL;
720 
721 	/* try to find next selem in the same bucket */
722 	selem = prev_selem;
723 	count = 0;
724 	while (selem) {
725 		selem = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&selem->map_node)),
726 					 struct bpf_local_storage_elem, map_node);
727 		if (!selem) {
728 			/* not found, unlock and go to the next bucket */
729 			b = &smap->buckets[bucket_id++];
730 			rcu_read_unlock();
731 			skip_elems = 0;
732 			break;
733 		}
734 		sk_storage = rcu_dereference(selem->local_storage);
735 		if (sk_storage) {
736 			info->skip_elems = skip_elems + count;
737 			return selem;
738 		}
739 		count++;
740 	}
741 
742 	for (i = bucket_id; i < (1U << smap->bucket_log); i++) {
743 		b = &smap->buckets[i];
744 		rcu_read_lock();
745 		count = 0;
746 		hlist_for_each_entry_rcu(selem, &b->list, map_node) {
747 			sk_storage = rcu_dereference(selem->local_storage);
748 			if (sk_storage && count >= skip_elems) {
749 				info->bucket_id = i;
750 				info->skip_elems = count;
751 				return selem;
752 			}
753 			count++;
754 		}
755 		rcu_read_unlock();
756 		skip_elems = 0;
757 	}
758 
759 	info->bucket_id = i;
760 	info->skip_elems = 0;
761 	return NULL;
762 }
763 
764 static void *bpf_sk_storage_map_seq_start(struct seq_file *seq, loff_t *pos)
765 {
766 	struct bpf_local_storage_elem *selem;
767 
768 	selem = bpf_sk_storage_map_seq_find_next(seq->private, NULL);
769 	if (!selem)
770 		return NULL;
771 
772 	if (*pos == 0)
773 		++*pos;
774 	return selem;
775 }
776 
777 static void *bpf_sk_storage_map_seq_next(struct seq_file *seq, void *v,
778 					 loff_t *pos)
779 {
780 	struct bpf_iter_seq_sk_storage_map_info *info = seq->private;
781 
782 	++*pos;
783 	++info->skip_elems;
784 	return bpf_sk_storage_map_seq_find_next(seq->private, v);
785 }
786 
787 struct bpf_iter__bpf_sk_storage_map {
788 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
789 	__bpf_md_ptr(struct bpf_map *, map);
790 	__bpf_md_ptr(struct sock *, sk);
791 	__bpf_md_ptr(void *, value);
792 };
793 
794 DEFINE_BPF_ITER_FUNC(bpf_sk_storage_map, struct bpf_iter_meta *meta,
795 		     struct bpf_map *map, struct sock *sk,
796 		     void *value)
797 
798 static int __bpf_sk_storage_map_seq_show(struct seq_file *seq,
799 					 struct bpf_local_storage_elem *selem)
800 {
801 	struct bpf_iter_seq_sk_storage_map_info *info = seq->private;
802 	struct bpf_iter__bpf_sk_storage_map ctx = {};
803 	struct bpf_local_storage *sk_storage;
804 	struct bpf_iter_meta meta;
805 	struct bpf_prog *prog;
806 	int ret = 0;
807 
808 	meta.seq = seq;
809 	prog = bpf_iter_get_info(&meta, selem == NULL);
810 	if (prog) {
811 		ctx.meta = &meta;
812 		ctx.map = info->map;
813 		if (selem) {
814 			sk_storage = rcu_dereference(selem->local_storage);
815 			ctx.sk = sk_storage->owner;
816 			ctx.value = SDATA(selem)->data;
817 		}
818 		ret = bpf_iter_run_prog(prog, &ctx);
819 	}
820 
821 	return ret;
822 }
823 
824 static int bpf_sk_storage_map_seq_show(struct seq_file *seq, void *v)
825 {
826 	return __bpf_sk_storage_map_seq_show(seq, v);
827 }
828 
829 static void bpf_sk_storage_map_seq_stop(struct seq_file *seq, void *v)
830 	__releases(RCU)
831 {
832 	if (!v)
833 		(void)__bpf_sk_storage_map_seq_show(seq, v);
834 	else
835 		rcu_read_unlock();
836 }
837 
838 static int bpf_iter_init_sk_storage_map(void *priv_data,
839 					struct bpf_iter_aux_info *aux)
840 {
841 	struct bpf_iter_seq_sk_storage_map_info *seq_info = priv_data;
842 
843 	bpf_map_inc_with_uref(aux->map);
844 	seq_info->map = aux->map;
845 	return 0;
846 }
847 
848 static void bpf_iter_fini_sk_storage_map(void *priv_data)
849 {
850 	struct bpf_iter_seq_sk_storage_map_info *seq_info = priv_data;
851 
852 	bpf_map_put_with_uref(seq_info->map);
853 }
854 
855 static int bpf_iter_attach_map(struct bpf_prog *prog,
856 			       union bpf_iter_link_info *linfo,
857 			       struct bpf_iter_aux_info *aux)
858 {
859 	struct bpf_map *map;
860 	int err = -EINVAL;
861 
862 	if (!linfo->map.map_fd)
863 		return -EBADF;
864 
865 	map = bpf_map_get_with_uref(linfo->map.map_fd);
866 	if (IS_ERR(map))
867 		return PTR_ERR(map);
868 
869 	if (map->map_type != BPF_MAP_TYPE_SK_STORAGE)
870 		goto put_map;
871 
872 	if (prog->aux->max_rdwr_access > map->value_size) {
873 		err = -EACCES;
874 		goto put_map;
875 	}
876 
877 	aux->map = map;
878 	return 0;
879 
880 put_map:
881 	bpf_map_put_with_uref(map);
882 	return err;
883 }
884 
885 static void bpf_iter_detach_map(struct bpf_iter_aux_info *aux)
886 {
887 	bpf_map_put_with_uref(aux->map);
888 }
889 
890 static const struct seq_operations bpf_sk_storage_map_seq_ops = {
891 	.start  = bpf_sk_storage_map_seq_start,
892 	.next   = bpf_sk_storage_map_seq_next,
893 	.stop   = bpf_sk_storage_map_seq_stop,
894 	.show   = bpf_sk_storage_map_seq_show,
895 };
896 
897 static const struct bpf_iter_seq_info iter_seq_info = {
898 	.seq_ops		= &bpf_sk_storage_map_seq_ops,
899 	.init_seq_private	= bpf_iter_init_sk_storage_map,
900 	.fini_seq_private	= bpf_iter_fini_sk_storage_map,
901 	.seq_priv_size		= sizeof(struct bpf_iter_seq_sk_storage_map_info),
902 };
903 
904 static struct bpf_iter_reg bpf_sk_storage_map_reg_info = {
905 	.target			= "bpf_sk_storage_map",
906 	.attach_target		= bpf_iter_attach_map,
907 	.detach_target		= bpf_iter_detach_map,
908 	.show_fdinfo		= bpf_iter_map_show_fdinfo,
909 	.fill_link_info		= bpf_iter_map_fill_link_info,
910 	.ctx_arg_info_size	= 2,
911 	.ctx_arg_info		= {
912 		{ offsetof(struct bpf_iter__bpf_sk_storage_map, sk),
913 		  PTR_TO_BTF_ID_OR_NULL },
914 		{ offsetof(struct bpf_iter__bpf_sk_storage_map, value),
915 		  PTR_TO_BUF | PTR_MAYBE_NULL },
916 	},
917 	.seq_info		= &iter_seq_info,
918 };
919 
920 static int __init bpf_sk_storage_map_iter_init(void)
921 {
922 	bpf_sk_storage_map_reg_info.ctx_arg_info[0].btf_id =
923 		btf_sock_ids[BTF_SOCK_TYPE_SOCK];
924 	return bpf_iter_reg_target(&bpf_sk_storage_map_reg_info);
925 }
926 late_initcall(bpf_sk_storage_map_iter_init);
927