xref: /openbmc/linux/kernel/bpf/cgroup.c (revision b39b5f41)
1 /*
2  * Functions to manage eBPF programs attached to cgroups
3  *
4  * Copyright (c) 2016 Daniel Mack
5  *
6  * This file is subject to the terms and conditions of version 2 of the GNU
7  * General Public License.  See the file COPYING in the main directory of the
8  * Linux distribution for more details.
9  */
10 
11 #include <linux/kernel.h>
12 #include <linux/atomic.h>
13 #include <linux/cgroup.h>
14 #include <linux/slab.h>
15 #include <linux/bpf.h>
16 #include <linux/bpf-cgroup.h>
17 #include <net/sock.h>
18 
19 DEFINE_STATIC_KEY_FALSE(cgroup_bpf_enabled_key);
20 EXPORT_SYMBOL(cgroup_bpf_enabled_key);
21 
22 /**
23  * cgroup_bpf_put() - put references of all bpf programs
24  * @cgrp: the cgroup to modify
25  */
26 void cgroup_bpf_put(struct cgroup *cgrp)
27 {
28 	enum bpf_cgroup_storage_type stype;
29 	unsigned int type;
30 
31 	for (type = 0; type < ARRAY_SIZE(cgrp->bpf.progs); type++) {
32 		struct list_head *progs = &cgrp->bpf.progs[type];
33 		struct bpf_prog_list *pl, *tmp;
34 
35 		list_for_each_entry_safe(pl, tmp, progs, node) {
36 			list_del(&pl->node);
37 			bpf_prog_put(pl->prog);
38 			for_each_cgroup_storage_type(stype) {
39 				bpf_cgroup_storage_unlink(pl->storage[stype]);
40 				bpf_cgroup_storage_free(pl->storage[stype]);
41 			}
42 			kfree(pl);
43 			static_branch_dec(&cgroup_bpf_enabled_key);
44 		}
45 		bpf_prog_array_free(cgrp->bpf.effective[type]);
46 	}
47 }
48 
49 /* count number of elements in the list.
50  * it's slow but the list cannot be long
51  */
52 static u32 prog_list_length(struct list_head *head)
53 {
54 	struct bpf_prog_list *pl;
55 	u32 cnt = 0;
56 
57 	list_for_each_entry(pl, head, node) {
58 		if (!pl->prog)
59 			continue;
60 		cnt++;
61 	}
62 	return cnt;
63 }
64 
65 /* if parent has non-overridable prog attached,
66  * disallow attaching new programs to the descendent cgroup.
67  * if parent has overridable or multi-prog, allow attaching
68  */
69 static bool hierarchy_allows_attach(struct cgroup *cgrp,
70 				    enum bpf_attach_type type,
71 				    u32 new_flags)
72 {
73 	struct cgroup *p;
74 
75 	p = cgroup_parent(cgrp);
76 	if (!p)
77 		return true;
78 	do {
79 		u32 flags = p->bpf.flags[type];
80 		u32 cnt;
81 
82 		if (flags & BPF_F_ALLOW_MULTI)
83 			return true;
84 		cnt = prog_list_length(&p->bpf.progs[type]);
85 		WARN_ON_ONCE(cnt > 1);
86 		if (cnt == 1)
87 			return !!(flags & BPF_F_ALLOW_OVERRIDE);
88 		p = cgroup_parent(p);
89 	} while (p);
90 	return true;
91 }
92 
93 /* compute a chain of effective programs for a given cgroup:
94  * start from the list of programs in this cgroup and add
95  * all parent programs.
96  * Note that parent's F_ALLOW_OVERRIDE-type program is yielding
97  * to programs in this cgroup
98  */
99 static int compute_effective_progs(struct cgroup *cgrp,
100 				   enum bpf_attach_type type,
101 				   struct bpf_prog_array __rcu **array)
102 {
103 	enum bpf_cgroup_storage_type stype;
104 	struct bpf_prog_array *progs;
105 	struct bpf_prog_list *pl;
106 	struct cgroup *p = cgrp;
107 	int cnt = 0;
108 
109 	/* count number of effective programs by walking parents */
110 	do {
111 		if (cnt == 0 || (p->bpf.flags[type] & BPF_F_ALLOW_MULTI))
112 			cnt += prog_list_length(&p->bpf.progs[type]);
113 		p = cgroup_parent(p);
114 	} while (p);
115 
116 	progs = bpf_prog_array_alloc(cnt, GFP_KERNEL);
117 	if (!progs)
118 		return -ENOMEM;
119 
120 	/* populate the array with effective progs */
121 	cnt = 0;
122 	p = cgrp;
123 	do {
124 		if (cnt > 0 && !(p->bpf.flags[type] & BPF_F_ALLOW_MULTI))
125 			continue;
126 
127 		list_for_each_entry(pl, &p->bpf.progs[type], node) {
128 			if (!pl->prog)
129 				continue;
130 
131 			progs->items[cnt].prog = pl->prog;
132 			for_each_cgroup_storage_type(stype)
133 				progs->items[cnt].cgroup_storage[stype] =
134 					pl->storage[stype];
135 			cnt++;
136 		}
137 	} while ((p = cgroup_parent(p)));
138 
139 	rcu_assign_pointer(*array, progs);
140 	return 0;
141 }
142 
143 static void activate_effective_progs(struct cgroup *cgrp,
144 				     enum bpf_attach_type type,
145 				     struct bpf_prog_array __rcu *array)
146 {
147 	struct bpf_prog_array __rcu *old_array;
148 
149 	old_array = xchg(&cgrp->bpf.effective[type], array);
150 	/* free prog array after grace period, since __cgroup_bpf_run_*()
151 	 * might be still walking the array
152 	 */
153 	bpf_prog_array_free(old_array);
154 }
155 
156 /**
157  * cgroup_bpf_inherit() - inherit effective programs from parent
158  * @cgrp: the cgroup to modify
159  */
160 int cgroup_bpf_inherit(struct cgroup *cgrp)
161 {
162 /* has to use marco instead of const int, since compiler thinks
163  * that array below is variable length
164  */
165 #define	NR ARRAY_SIZE(cgrp->bpf.effective)
166 	struct bpf_prog_array __rcu *arrays[NR] = {};
167 	int i;
168 
169 	for (i = 0; i < NR; i++)
170 		INIT_LIST_HEAD(&cgrp->bpf.progs[i]);
171 
172 	for (i = 0; i < NR; i++)
173 		if (compute_effective_progs(cgrp, i, &arrays[i]))
174 			goto cleanup;
175 
176 	for (i = 0; i < NR; i++)
177 		activate_effective_progs(cgrp, i, arrays[i]);
178 
179 	return 0;
180 cleanup:
181 	for (i = 0; i < NR; i++)
182 		bpf_prog_array_free(arrays[i]);
183 	return -ENOMEM;
184 }
185 
186 static int update_effective_progs(struct cgroup *cgrp,
187 				  enum bpf_attach_type type)
188 {
189 	struct cgroup_subsys_state *css;
190 	int err;
191 
192 	/* allocate and recompute effective prog arrays */
193 	css_for_each_descendant_pre(css, &cgrp->self) {
194 		struct cgroup *desc = container_of(css, struct cgroup, self);
195 
196 		err = compute_effective_progs(desc, type, &desc->bpf.inactive);
197 		if (err)
198 			goto cleanup;
199 	}
200 
201 	/* all allocations were successful. Activate all prog arrays */
202 	css_for_each_descendant_pre(css, &cgrp->self) {
203 		struct cgroup *desc = container_of(css, struct cgroup, self);
204 
205 		activate_effective_progs(desc, type, desc->bpf.inactive);
206 		desc->bpf.inactive = NULL;
207 	}
208 
209 	return 0;
210 
211 cleanup:
212 	/* oom while computing effective. Free all computed effective arrays
213 	 * since they were not activated
214 	 */
215 	css_for_each_descendant_pre(css, &cgrp->self) {
216 		struct cgroup *desc = container_of(css, struct cgroup, self);
217 
218 		bpf_prog_array_free(desc->bpf.inactive);
219 		desc->bpf.inactive = NULL;
220 	}
221 
222 	return err;
223 }
224 
225 #define BPF_CGROUP_MAX_PROGS 64
226 
227 /**
228  * __cgroup_bpf_attach() - Attach the program to a cgroup, and
229  *                         propagate the change to descendants
230  * @cgrp: The cgroup which descendants to traverse
231  * @prog: A program to attach
232  * @type: Type of attach operation
233  *
234  * Must be called with cgroup_mutex held.
235  */
236 int __cgroup_bpf_attach(struct cgroup *cgrp, struct bpf_prog *prog,
237 			enum bpf_attach_type type, u32 flags)
238 {
239 	struct list_head *progs = &cgrp->bpf.progs[type];
240 	struct bpf_prog *old_prog = NULL;
241 	struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE],
242 		*old_storage[MAX_BPF_CGROUP_STORAGE_TYPE] = {NULL};
243 	enum bpf_cgroup_storage_type stype;
244 	struct bpf_prog_list *pl;
245 	bool pl_was_allocated;
246 	int err;
247 
248 	if ((flags & BPF_F_ALLOW_OVERRIDE) && (flags & BPF_F_ALLOW_MULTI))
249 		/* invalid combination */
250 		return -EINVAL;
251 
252 	if (!hierarchy_allows_attach(cgrp, type, flags))
253 		return -EPERM;
254 
255 	if (!list_empty(progs) && cgrp->bpf.flags[type] != flags)
256 		/* Disallow attaching non-overridable on top
257 		 * of existing overridable in this cgroup.
258 		 * Disallow attaching multi-prog if overridable or none
259 		 */
260 		return -EPERM;
261 
262 	if (prog_list_length(progs) >= BPF_CGROUP_MAX_PROGS)
263 		return -E2BIG;
264 
265 	for_each_cgroup_storage_type(stype) {
266 		storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
267 		if (IS_ERR(storage[stype])) {
268 			storage[stype] = NULL;
269 			for_each_cgroup_storage_type(stype)
270 				bpf_cgroup_storage_free(storage[stype]);
271 			return -ENOMEM;
272 		}
273 	}
274 
275 	if (flags & BPF_F_ALLOW_MULTI) {
276 		list_for_each_entry(pl, progs, node) {
277 			if (pl->prog == prog) {
278 				/* disallow attaching the same prog twice */
279 				for_each_cgroup_storage_type(stype)
280 					bpf_cgroup_storage_free(storage[stype]);
281 				return -EINVAL;
282 			}
283 		}
284 
285 		pl = kmalloc(sizeof(*pl), GFP_KERNEL);
286 		if (!pl) {
287 			for_each_cgroup_storage_type(stype)
288 				bpf_cgroup_storage_free(storage[stype]);
289 			return -ENOMEM;
290 		}
291 
292 		pl_was_allocated = true;
293 		pl->prog = prog;
294 		for_each_cgroup_storage_type(stype)
295 			pl->storage[stype] = storage[stype];
296 		list_add_tail(&pl->node, progs);
297 	} else {
298 		if (list_empty(progs)) {
299 			pl = kmalloc(sizeof(*pl), GFP_KERNEL);
300 			if (!pl) {
301 				for_each_cgroup_storage_type(stype)
302 					bpf_cgroup_storage_free(storage[stype]);
303 				return -ENOMEM;
304 			}
305 			pl_was_allocated = true;
306 			list_add_tail(&pl->node, progs);
307 		} else {
308 			pl = list_first_entry(progs, typeof(*pl), node);
309 			old_prog = pl->prog;
310 			for_each_cgroup_storage_type(stype) {
311 				old_storage[stype] = pl->storage[stype];
312 				bpf_cgroup_storage_unlink(old_storage[stype]);
313 			}
314 			pl_was_allocated = false;
315 		}
316 		pl->prog = prog;
317 		for_each_cgroup_storage_type(stype)
318 			pl->storage[stype] = storage[stype];
319 	}
320 
321 	cgrp->bpf.flags[type] = flags;
322 
323 	err = update_effective_progs(cgrp, type);
324 	if (err)
325 		goto cleanup;
326 
327 	static_branch_inc(&cgroup_bpf_enabled_key);
328 	for_each_cgroup_storage_type(stype) {
329 		if (!old_storage[stype])
330 			continue;
331 		bpf_cgroup_storage_free(old_storage[stype]);
332 	}
333 	if (old_prog) {
334 		bpf_prog_put(old_prog);
335 		static_branch_dec(&cgroup_bpf_enabled_key);
336 	}
337 	for_each_cgroup_storage_type(stype)
338 		bpf_cgroup_storage_link(storage[stype], cgrp, type);
339 	return 0;
340 
341 cleanup:
342 	/* and cleanup the prog list */
343 	pl->prog = old_prog;
344 	for_each_cgroup_storage_type(stype) {
345 		bpf_cgroup_storage_free(pl->storage[stype]);
346 		pl->storage[stype] = old_storage[stype];
347 		bpf_cgroup_storage_link(old_storage[stype], cgrp, type);
348 	}
349 	if (pl_was_allocated) {
350 		list_del(&pl->node);
351 		kfree(pl);
352 	}
353 	return err;
354 }
355 
356 /**
357  * __cgroup_bpf_detach() - Detach the program from a cgroup, and
358  *                         propagate the change to descendants
359  * @cgrp: The cgroup which descendants to traverse
360  * @prog: A program to detach or NULL
361  * @type: Type of detach operation
362  *
363  * Must be called with cgroup_mutex held.
364  */
365 int __cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
366 			enum bpf_attach_type type, u32 unused_flags)
367 {
368 	struct list_head *progs = &cgrp->bpf.progs[type];
369 	enum bpf_cgroup_storage_type stype;
370 	u32 flags = cgrp->bpf.flags[type];
371 	struct bpf_prog *old_prog = NULL;
372 	struct bpf_prog_list *pl;
373 	int err;
374 
375 	if (flags & BPF_F_ALLOW_MULTI) {
376 		if (!prog)
377 			/* to detach MULTI prog the user has to specify valid FD
378 			 * of the program to be detached
379 			 */
380 			return -EINVAL;
381 	} else {
382 		if (list_empty(progs))
383 			/* report error when trying to detach and nothing is attached */
384 			return -ENOENT;
385 	}
386 
387 	if (flags & BPF_F_ALLOW_MULTI) {
388 		/* find the prog and detach it */
389 		list_for_each_entry(pl, progs, node) {
390 			if (pl->prog != prog)
391 				continue;
392 			old_prog = prog;
393 			/* mark it deleted, so it's ignored while
394 			 * recomputing effective
395 			 */
396 			pl->prog = NULL;
397 			break;
398 		}
399 		if (!old_prog)
400 			return -ENOENT;
401 	} else {
402 		/* to maintain backward compatibility NONE and OVERRIDE cgroups
403 		 * allow detaching with invalid FD (prog==NULL)
404 		 */
405 		pl = list_first_entry(progs, typeof(*pl), node);
406 		old_prog = pl->prog;
407 		pl->prog = NULL;
408 	}
409 
410 	err = update_effective_progs(cgrp, type);
411 	if (err)
412 		goto cleanup;
413 
414 	/* now can actually delete it from this cgroup list */
415 	list_del(&pl->node);
416 	for_each_cgroup_storage_type(stype) {
417 		bpf_cgroup_storage_unlink(pl->storage[stype]);
418 		bpf_cgroup_storage_free(pl->storage[stype]);
419 	}
420 	kfree(pl);
421 	if (list_empty(progs))
422 		/* last program was detached, reset flags to zero */
423 		cgrp->bpf.flags[type] = 0;
424 
425 	bpf_prog_put(old_prog);
426 	static_branch_dec(&cgroup_bpf_enabled_key);
427 	return 0;
428 
429 cleanup:
430 	/* and restore back old_prog */
431 	pl->prog = old_prog;
432 	return err;
433 }
434 
435 /* Must be called with cgroup_mutex held to avoid races. */
436 int __cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
437 		       union bpf_attr __user *uattr)
438 {
439 	__u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
440 	enum bpf_attach_type type = attr->query.attach_type;
441 	struct list_head *progs = &cgrp->bpf.progs[type];
442 	u32 flags = cgrp->bpf.flags[type];
443 	int cnt, ret = 0, i;
444 
445 	if (attr->query.query_flags & BPF_F_QUERY_EFFECTIVE)
446 		cnt = bpf_prog_array_length(cgrp->bpf.effective[type]);
447 	else
448 		cnt = prog_list_length(progs);
449 
450 	if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)))
451 		return -EFAULT;
452 	if (copy_to_user(&uattr->query.prog_cnt, &cnt, sizeof(cnt)))
453 		return -EFAULT;
454 	if (attr->query.prog_cnt == 0 || !prog_ids || !cnt)
455 		/* return early if user requested only program count + flags */
456 		return 0;
457 	if (attr->query.prog_cnt < cnt) {
458 		cnt = attr->query.prog_cnt;
459 		ret = -ENOSPC;
460 	}
461 
462 	if (attr->query.query_flags & BPF_F_QUERY_EFFECTIVE) {
463 		return bpf_prog_array_copy_to_user(cgrp->bpf.effective[type],
464 						   prog_ids, cnt);
465 	} else {
466 		struct bpf_prog_list *pl;
467 		u32 id;
468 
469 		i = 0;
470 		list_for_each_entry(pl, progs, node) {
471 			id = pl->prog->aux->id;
472 			if (copy_to_user(prog_ids + i, &id, sizeof(id)))
473 				return -EFAULT;
474 			if (++i == cnt)
475 				break;
476 		}
477 	}
478 	return ret;
479 }
480 
481 int cgroup_bpf_prog_attach(const union bpf_attr *attr,
482 			   enum bpf_prog_type ptype, struct bpf_prog *prog)
483 {
484 	struct cgroup *cgrp;
485 	int ret;
486 
487 	cgrp = cgroup_get_from_fd(attr->target_fd);
488 	if (IS_ERR(cgrp))
489 		return PTR_ERR(cgrp);
490 
491 	ret = cgroup_bpf_attach(cgrp, prog, attr->attach_type,
492 				attr->attach_flags);
493 	cgroup_put(cgrp);
494 	return ret;
495 }
496 
497 int cgroup_bpf_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
498 {
499 	struct bpf_prog *prog;
500 	struct cgroup *cgrp;
501 	int ret;
502 
503 	cgrp = cgroup_get_from_fd(attr->target_fd);
504 	if (IS_ERR(cgrp))
505 		return PTR_ERR(cgrp);
506 
507 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
508 	if (IS_ERR(prog))
509 		prog = NULL;
510 
511 	ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type, 0);
512 	if (prog)
513 		bpf_prog_put(prog);
514 
515 	cgroup_put(cgrp);
516 	return ret;
517 }
518 
519 int cgroup_bpf_prog_query(const union bpf_attr *attr,
520 			  union bpf_attr __user *uattr)
521 {
522 	struct cgroup *cgrp;
523 	int ret;
524 
525 	cgrp = cgroup_get_from_fd(attr->query.target_fd);
526 	if (IS_ERR(cgrp))
527 		return PTR_ERR(cgrp);
528 
529 	ret = cgroup_bpf_query(cgrp, attr, uattr);
530 
531 	cgroup_put(cgrp);
532 	return ret;
533 }
534 
535 /**
536  * __cgroup_bpf_run_filter_skb() - Run a program for packet filtering
537  * @sk: The socket sending or receiving traffic
538  * @skb: The skb that is being sent or received
539  * @type: The type of program to be exectuted
540  *
541  * If no socket is passed, or the socket is not of type INET or INET6,
542  * this function does nothing and returns 0.
543  *
544  * The program type passed in via @type must be suitable for network
545  * filtering. No further check is performed to assert that.
546  *
547  * This function will return %-EPERM if any if an attached program was found
548  * and if it returned != 1 during execution. In all other cases, 0 is returned.
549  */
550 int __cgroup_bpf_run_filter_skb(struct sock *sk,
551 				struct sk_buff *skb,
552 				enum bpf_attach_type type)
553 {
554 	unsigned int offset = skb->data - skb_network_header(skb);
555 	struct sock *save_sk;
556 	void *saved_data_end;
557 	struct cgroup *cgrp;
558 	int ret;
559 
560 	if (!sk || !sk_fullsock(sk))
561 		return 0;
562 
563 	if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)
564 		return 0;
565 
566 	cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
567 	save_sk = skb->sk;
568 	skb->sk = sk;
569 	__skb_push(skb, offset);
570 
571 	/* compute pointers for the bpf prog */
572 	bpf_compute_and_save_data_end(skb, &saved_data_end);
573 
574 	ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], skb,
575 				 bpf_prog_run_save_cb);
576 	bpf_restore_data_end(skb, saved_data_end);
577 	__skb_pull(skb, offset);
578 	skb->sk = save_sk;
579 	return ret == 1 ? 0 : -EPERM;
580 }
581 EXPORT_SYMBOL(__cgroup_bpf_run_filter_skb);
582 
583 /**
584  * __cgroup_bpf_run_filter_sk() - Run a program on a sock
585  * @sk: sock structure to manipulate
586  * @type: The type of program to be exectuted
587  *
588  * socket is passed is expected to be of type INET or INET6.
589  *
590  * The program type passed in via @type must be suitable for sock
591  * filtering. No further check is performed to assert that.
592  *
593  * This function will return %-EPERM if any if an attached program was found
594  * and if it returned != 1 during execution. In all other cases, 0 is returned.
595  */
596 int __cgroup_bpf_run_filter_sk(struct sock *sk,
597 			       enum bpf_attach_type type)
598 {
599 	struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
600 	int ret;
601 
602 	ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], sk, BPF_PROG_RUN);
603 	return ret == 1 ? 0 : -EPERM;
604 }
605 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sk);
606 
607 /**
608  * __cgroup_bpf_run_filter_sock_addr() - Run a program on a sock and
609  *                                       provided by user sockaddr
610  * @sk: sock struct that will use sockaddr
611  * @uaddr: sockaddr struct provided by user
612  * @type: The type of program to be exectuted
613  * @t_ctx: Pointer to attach type specific context
614  *
615  * socket is expected to be of type INET or INET6.
616  *
617  * This function will return %-EPERM if an attached program is found and
618  * returned value != 1 during execution. In all other cases, 0 is returned.
619  */
620 int __cgroup_bpf_run_filter_sock_addr(struct sock *sk,
621 				      struct sockaddr *uaddr,
622 				      enum bpf_attach_type type,
623 				      void *t_ctx)
624 {
625 	struct bpf_sock_addr_kern ctx = {
626 		.sk = sk,
627 		.uaddr = uaddr,
628 		.t_ctx = t_ctx,
629 	};
630 	struct sockaddr_storage unspec;
631 	struct cgroup *cgrp;
632 	int ret;
633 
634 	/* Check socket family since not all sockets represent network
635 	 * endpoint (e.g. AF_UNIX).
636 	 */
637 	if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)
638 		return 0;
639 
640 	if (!ctx.uaddr) {
641 		memset(&unspec, 0, sizeof(unspec));
642 		ctx.uaddr = (struct sockaddr *)&unspec;
643 	}
644 
645 	cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
646 	ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], &ctx, BPF_PROG_RUN);
647 
648 	return ret == 1 ? 0 : -EPERM;
649 }
650 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_addr);
651 
652 /**
653  * __cgroup_bpf_run_filter_sock_ops() - Run a program on a sock
654  * @sk: socket to get cgroup from
655  * @sock_ops: bpf_sock_ops_kern struct to pass to program. Contains
656  * sk with connection information (IP addresses, etc.) May not contain
657  * cgroup info if it is a req sock.
658  * @type: The type of program to be exectuted
659  *
660  * socket passed is expected to be of type INET or INET6.
661  *
662  * The program type passed in via @type must be suitable for sock_ops
663  * filtering. No further check is performed to assert that.
664  *
665  * This function will return %-EPERM if any if an attached program was found
666  * and if it returned != 1 during execution. In all other cases, 0 is returned.
667  */
668 int __cgroup_bpf_run_filter_sock_ops(struct sock *sk,
669 				     struct bpf_sock_ops_kern *sock_ops,
670 				     enum bpf_attach_type type)
671 {
672 	struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
673 	int ret;
674 
675 	ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], sock_ops,
676 				 BPF_PROG_RUN);
677 	return ret == 1 ? 0 : -EPERM;
678 }
679 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_ops);
680 
681 int __cgroup_bpf_check_dev_permission(short dev_type, u32 major, u32 minor,
682 				      short access, enum bpf_attach_type type)
683 {
684 	struct cgroup *cgrp;
685 	struct bpf_cgroup_dev_ctx ctx = {
686 		.access_type = (access << 16) | dev_type,
687 		.major = major,
688 		.minor = minor,
689 	};
690 	int allow = 1;
691 
692 	rcu_read_lock();
693 	cgrp = task_dfl_cgroup(current);
694 	allow = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], &ctx,
695 				   BPF_PROG_RUN);
696 	rcu_read_unlock();
697 
698 	return !allow;
699 }
700 EXPORT_SYMBOL(__cgroup_bpf_check_dev_permission);
701 
702 static const struct bpf_func_proto *
703 cgroup_dev_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
704 {
705 	switch (func_id) {
706 	case BPF_FUNC_map_lookup_elem:
707 		return &bpf_map_lookup_elem_proto;
708 	case BPF_FUNC_map_update_elem:
709 		return &bpf_map_update_elem_proto;
710 	case BPF_FUNC_map_delete_elem:
711 		return &bpf_map_delete_elem_proto;
712 	case BPF_FUNC_get_current_uid_gid:
713 		return &bpf_get_current_uid_gid_proto;
714 	case BPF_FUNC_get_local_storage:
715 		return &bpf_get_local_storage_proto;
716 	case BPF_FUNC_get_current_cgroup_id:
717 		return &bpf_get_current_cgroup_id_proto;
718 	case BPF_FUNC_trace_printk:
719 		if (capable(CAP_SYS_ADMIN))
720 			return bpf_get_trace_printk_proto();
721 	default:
722 		return NULL;
723 	}
724 }
725 
726 static bool cgroup_dev_is_valid_access(int off, int size,
727 				       enum bpf_access_type type,
728 				       const struct bpf_prog *prog,
729 				       struct bpf_insn_access_aux *info)
730 {
731 	const int size_default = sizeof(__u32);
732 
733 	if (type == BPF_WRITE)
734 		return false;
735 
736 	if (off < 0 || off + size > sizeof(struct bpf_cgroup_dev_ctx))
737 		return false;
738 	/* The verifier guarantees that size > 0. */
739 	if (off % size != 0)
740 		return false;
741 
742 	switch (off) {
743 	case bpf_ctx_range(struct bpf_cgroup_dev_ctx, access_type):
744 		bpf_ctx_record_field_size(info, size_default);
745 		if (!bpf_ctx_narrow_access_ok(off, size, size_default))
746 			return false;
747 		break;
748 	default:
749 		if (size != size_default)
750 			return false;
751 	}
752 
753 	return true;
754 }
755 
756 const struct bpf_prog_ops cg_dev_prog_ops = {
757 };
758 
759 const struct bpf_verifier_ops cg_dev_verifier_ops = {
760 	.get_func_proto		= cgroup_dev_func_proto,
761 	.is_valid_access	= cgroup_dev_is_valid_access,
762 };
763