Loading...
Note: File does not exist in v3.15.
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Functions to manage eBPF programs attached to cgroups
4 *
5 * Copyright (c) 2016 Daniel Mack
6 */
7
8#include <linux/kernel.h>
9#include <linux/atomic.h>
10#include <linux/cgroup.h>
11#include <linux/filter.h>
12#include <linux/slab.h>
13#include <linux/sysctl.h>
14#include <linux/string.h>
15#include <linux/bpf.h>
16#include <linux/bpf-cgroup.h>
17#include <net/sock.h>
18#include <net/bpf_sk_storage.h>
19
20#include "../cgroup/cgroup-internal.h"
21
22DEFINE_STATIC_KEY_FALSE(cgroup_bpf_enabled_key);
23EXPORT_SYMBOL(cgroup_bpf_enabled_key);
24
25void cgroup_bpf_offline(struct cgroup *cgrp)
26{
27 cgroup_get(cgrp);
28 percpu_ref_kill(&cgrp->bpf.refcnt);
29}
30
31/**
32 * cgroup_bpf_release() - put references of all bpf programs and
33 * release all cgroup bpf data
34 * @work: work structure embedded into the cgroup to modify
35 */
36static void cgroup_bpf_release(struct work_struct *work)
37{
38 struct cgroup *cgrp = container_of(work, struct cgroup,
39 bpf.release_work);
40 enum bpf_cgroup_storage_type stype;
41 struct bpf_prog_array *old_array;
42 unsigned int type;
43
44 mutex_lock(&cgroup_mutex);
45
46 for (type = 0; type < ARRAY_SIZE(cgrp->bpf.progs); type++) {
47 struct list_head *progs = &cgrp->bpf.progs[type];
48 struct bpf_prog_list *pl, *tmp;
49
50 list_for_each_entry_safe(pl, tmp, progs, node) {
51 list_del(&pl->node);
52 bpf_prog_put(pl->prog);
53 for_each_cgroup_storage_type(stype) {
54 bpf_cgroup_storage_unlink(pl->storage[stype]);
55 bpf_cgroup_storage_free(pl->storage[stype]);
56 }
57 kfree(pl);
58 static_branch_dec(&cgroup_bpf_enabled_key);
59 }
60 old_array = rcu_dereference_protected(
61 cgrp->bpf.effective[type],
62 lockdep_is_held(&cgroup_mutex));
63 bpf_prog_array_free(old_array);
64 }
65
66 mutex_unlock(&cgroup_mutex);
67
68 percpu_ref_exit(&cgrp->bpf.refcnt);
69 cgroup_put(cgrp);
70}
71
72/**
73 * cgroup_bpf_release_fn() - callback used to schedule releasing
74 * of bpf cgroup data
75 * @ref: percpu ref counter structure
76 */
77static void cgroup_bpf_release_fn(struct percpu_ref *ref)
78{
79 struct cgroup *cgrp = container_of(ref, struct cgroup, bpf.refcnt);
80
81 INIT_WORK(&cgrp->bpf.release_work, cgroup_bpf_release);
82 queue_work(system_wq, &cgrp->bpf.release_work);
83}
84
85/* count number of elements in the list.
86 * it's slow but the list cannot be long
87 */
88static u32 prog_list_length(struct list_head *head)
89{
90 struct bpf_prog_list *pl;
91 u32 cnt = 0;
92
93 list_for_each_entry(pl, head, node) {
94 if (!pl->prog)
95 continue;
96 cnt++;
97 }
98 return cnt;
99}
100
101/* if parent has non-overridable prog attached,
102 * disallow attaching new programs to the descendent cgroup.
103 * if parent has overridable or multi-prog, allow attaching
104 */
105static bool hierarchy_allows_attach(struct cgroup *cgrp,
106 enum bpf_attach_type type,
107 u32 new_flags)
108{
109 struct cgroup *p;
110
111 p = cgroup_parent(cgrp);
112 if (!p)
113 return true;
114 do {
115 u32 flags = p->bpf.flags[type];
116 u32 cnt;
117
118 if (flags & BPF_F_ALLOW_MULTI)
119 return true;
120 cnt = prog_list_length(&p->bpf.progs[type]);
121 WARN_ON_ONCE(cnt > 1);
122 if (cnt == 1)
123 return !!(flags & BPF_F_ALLOW_OVERRIDE);
124 p = cgroup_parent(p);
125 } while (p);
126 return true;
127}
128
129/* compute a chain of effective programs for a given cgroup:
130 * start from the list of programs in this cgroup and add
131 * all parent programs.
132 * Note that parent's F_ALLOW_OVERRIDE-type program is yielding
133 * to programs in this cgroup
134 */
135static int compute_effective_progs(struct cgroup *cgrp,
136 enum bpf_attach_type type,
137 struct bpf_prog_array **array)
138{
139 enum bpf_cgroup_storage_type stype;
140 struct bpf_prog_array *progs;
141 struct bpf_prog_list *pl;
142 struct cgroup *p = cgrp;
143 int cnt = 0;
144
145 /* count number of effective programs by walking parents */
146 do {
147 if (cnt == 0 || (p->bpf.flags[type] & BPF_F_ALLOW_MULTI))
148 cnt += prog_list_length(&p->bpf.progs[type]);
149 p = cgroup_parent(p);
150 } while (p);
151
152 progs = bpf_prog_array_alloc(cnt, GFP_KERNEL);
153 if (!progs)
154 return -ENOMEM;
155
156 /* populate the array with effective progs */
157 cnt = 0;
158 p = cgrp;
159 do {
160 if (cnt > 0 && !(p->bpf.flags[type] & BPF_F_ALLOW_MULTI))
161 continue;
162
163 list_for_each_entry(pl, &p->bpf.progs[type], node) {
164 if (!pl->prog)
165 continue;
166
167 progs->items[cnt].prog = pl->prog;
168 for_each_cgroup_storage_type(stype)
169 progs->items[cnt].cgroup_storage[stype] =
170 pl->storage[stype];
171 cnt++;
172 }
173 } while ((p = cgroup_parent(p)));
174
175 *array = progs;
176 return 0;
177}
178
179static void activate_effective_progs(struct cgroup *cgrp,
180 enum bpf_attach_type type,
181 struct bpf_prog_array *old_array)
182{
183 rcu_swap_protected(cgrp->bpf.effective[type], old_array,
184 lockdep_is_held(&cgroup_mutex));
185 /* free prog array after grace period, since __cgroup_bpf_run_*()
186 * might be still walking the array
187 */
188 bpf_prog_array_free(old_array);
189}
190
191/**
192 * cgroup_bpf_inherit() - inherit effective programs from parent
193 * @cgrp: the cgroup to modify
194 */
195int cgroup_bpf_inherit(struct cgroup *cgrp)
196{
197/* has to use marco instead of const int, since compiler thinks
198 * that array below is variable length
199 */
200#define NR ARRAY_SIZE(cgrp->bpf.effective)
201 struct bpf_prog_array *arrays[NR] = {};
202 int ret, i;
203
204 ret = percpu_ref_init(&cgrp->bpf.refcnt, cgroup_bpf_release_fn, 0,
205 GFP_KERNEL);
206 if (ret)
207 return ret;
208
209 for (i = 0; i < NR; i++)
210 INIT_LIST_HEAD(&cgrp->bpf.progs[i]);
211
212 for (i = 0; i < NR; i++)
213 if (compute_effective_progs(cgrp, i, &arrays[i]))
214 goto cleanup;
215
216 for (i = 0; i < NR; i++)
217 activate_effective_progs(cgrp, i, arrays[i]);
218
219 return 0;
220cleanup:
221 for (i = 0; i < NR; i++)
222 bpf_prog_array_free(arrays[i]);
223
224 percpu_ref_exit(&cgrp->bpf.refcnt);
225
226 return -ENOMEM;
227}
228
229static int update_effective_progs(struct cgroup *cgrp,
230 enum bpf_attach_type type)
231{
232 struct cgroup_subsys_state *css;
233 int err;
234
235 /* allocate and recompute effective prog arrays */
236 css_for_each_descendant_pre(css, &cgrp->self) {
237 struct cgroup *desc = container_of(css, struct cgroup, self);
238
239 if (percpu_ref_is_zero(&desc->bpf.refcnt))
240 continue;
241
242 err = compute_effective_progs(desc, type, &desc->bpf.inactive);
243 if (err)
244 goto cleanup;
245 }
246
247 /* all allocations were successful. Activate all prog arrays */
248 css_for_each_descendant_pre(css, &cgrp->self) {
249 struct cgroup *desc = container_of(css, struct cgroup, self);
250
251 if (percpu_ref_is_zero(&desc->bpf.refcnt)) {
252 if (unlikely(desc->bpf.inactive)) {
253 bpf_prog_array_free(desc->bpf.inactive);
254 desc->bpf.inactive = NULL;
255 }
256 continue;
257 }
258
259 activate_effective_progs(desc, type, desc->bpf.inactive);
260 desc->bpf.inactive = NULL;
261 }
262
263 return 0;
264
265cleanup:
266 /* oom while computing effective. Free all computed effective arrays
267 * since they were not activated
268 */
269 css_for_each_descendant_pre(css, &cgrp->self) {
270 struct cgroup *desc = container_of(css, struct cgroup, self);
271
272 bpf_prog_array_free(desc->bpf.inactive);
273 desc->bpf.inactive = NULL;
274 }
275
276 return err;
277}
278
279#define BPF_CGROUP_MAX_PROGS 64
280
281/**
282 * __cgroup_bpf_attach() - Attach the program to a cgroup, and
283 * propagate the change to descendants
284 * @cgrp: The cgroup which descendants to traverse
285 * @prog: A program to attach
286 * @type: Type of attach operation
287 * @flags: Option flags
288 *
289 * Must be called with cgroup_mutex held.
290 */
291int __cgroup_bpf_attach(struct cgroup *cgrp, struct bpf_prog *prog,
292 enum bpf_attach_type type, u32 flags)
293{
294 struct list_head *progs = &cgrp->bpf.progs[type];
295 struct bpf_prog *old_prog = NULL;
296 struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE],
297 *old_storage[MAX_BPF_CGROUP_STORAGE_TYPE] = {NULL};
298 enum bpf_cgroup_storage_type stype;
299 struct bpf_prog_list *pl;
300 bool pl_was_allocated;
301 int err;
302
303 if ((flags & BPF_F_ALLOW_OVERRIDE) && (flags & BPF_F_ALLOW_MULTI))
304 /* invalid combination */
305 return -EINVAL;
306
307 if (!hierarchy_allows_attach(cgrp, type, flags))
308 return -EPERM;
309
310 if (!list_empty(progs) && cgrp->bpf.flags[type] != flags)
311 /* Disallow attaching non-overridable on top
312 * of existing overridable in this cgroup.
313 * Disallow attaching multi-prog if overridable or none
314 */
315 return -EPERM;
316
317 if (prog_list_length(progs) >= BPF_CGROUP_MAX_PROGS)
318 return -E2BIG;
319
320 for_each_cgroup_storage_type(stype) {
321 storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
322 if (IS_ERR(storage[stype])) {
323 storage[stype] = NULL;
324 for_each_cgroup_storage_type(stype)
325 bpf_cgroup_storage_free(storage[stype]);
326 return -ENOMEM;
327 }
328 }
329
330 if (flags & BPF_F_ALLOW_MULTI) {
331 list_for_each_entry(pl, progs, node) {
332 if (pl->prog == prog) {
333 /* disallow attaching the same prog twice */
334 for_each_cgroup_storage_type(stype)
335 bpf_cgroup_storage_free(storage[stype]);
336 return -EINVAL;
337 }
338 }
339
340 pl = kmalloc(sizeof(*pl), GFP_KERNEL);
341 if (!pl) {
342 for_each_cgroup_storage_type(stype)
343 bpf_cgroup_storage_free(storage[stype]);
344 return -ENOMEM;
345 }
346
347 pl_was_allocated = true;
348 pl->prog = prog;
349 for_each_cgroup_storage_type(stype)
350 pl->storage[stype] = storage[stype];
351 list_add_tail(&pl->node, progs);
352 } else {
353 if (list_empty(progs)) {
354 pl = kmalloc(sizeof(*pl), GFP_KERNEL);
355 if (!pl) {
356 for_each_cgroup_storage_type(stype)
357 bpf_cgroup_storage_free(storage[stype]);
358 return -ENOMEM;
359 }
360 pl_was_allocated = true;
361 list_add_tail(&pl->node, progs);
362 } else {
363 pl = list_first_entry(progs, typeof(*pl), node);
364 old_prog = pl->prog;
365 for_each_cgroup_storage_type(stype) {
366 old_storage[stype] = pl->storage[stype];
367 bpf_cgroup_storage_unlink(old_storage[stype]);
368 }
369 pl_was_allocated = false;
370 }
371 pl->prog = prog;
372 for_each_cgroup_storage_type(stype)
373 pl->storage[stype] = storage[stype];
374 }
375
376 cgrp->bpf.flags[type] = flags;
377
378 err = update_effective_progs(cgrp, type);
379 if (err)
380 goto cleanup;
381
382 static_branch_inc(&cgroup_bpf_enabled_key);
383 for_each_cgroup_storage_type(stype) {
384 if (!old_storage[stype])
385 continue;
386 bpf_cgroup_storage_free(old_storage[stype]);
387 }
388 if (old_prog) {
389 bpf_prog_put(old_prog);
390 static_branch_dec(&cgroup_bpf_enabled_key);
391 }
392 for_each_cgroup_storage_type(stype)
393 bpf_cgroup_storage_link(storage[stype], cgrp, type);
394 return 0;
395
396cleanup:
397 /* and cleanup the prog list */
398 pl->prog = old_prog;
399 for_each_cgroup_storage_type(stype) {
400 bpf_cgroup_storage_free(pl->storage[stype]);
401 pl->storage[stype] = old_storage[stype];
402 bpf_cgroup_storage_link(old_storage[stype], cgrp, type);
403 }
404 if (pl_was_allocated) {
405 list_del(&pl->node);
406 kfree(pl);
407 }
408 return err;
409}
410
411/**
412 * __cgroup_bpf_detach() - Detach the program from a cgroup, and
413 * propagate the change to descendants
414 * @cgrp: The cgroup which descendants to traverse
415 * @prog: A program to detach or NULL
416 * @type: Type of detach operation
417 *
418 * Must be called with cgroup_mutex held.
419 */
420int __cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
421 enum bpf_attach_type type)
422{
423 struct list_head *progs = &cgrp->bpf.progs[type];
424 enum bpf_cgroup_storage_type stype;
425 u32 flags = cgrp->bpf.flags[type];
426 struct bpf_prog *old_prog = NULL;
427 struct bpf_prog_list *pl;
428 int err;
429
430 if (flags & BPF_F_ALLOW_MULTI) {
431 if (!prog)
432 /* to detach MULTI prog the user has to specify valid FD
433 * of the program to be detached
434 */
435 return -EINVAL;
436 } else {
437 if (list_empty(progs))
438 /* report error when trying to detach and nothing is attached */
439 return -ENOENT;
440 }
441
442 if (flags & BPF_F_ALLOW_MULTI) {
443 /* find the prog and detach it */
444 list_for_each_entry(pl, progs, node) {
445 if (pl->prog != prog)
446 continue;
447 old_prog = prog;
448 /* mark it deleted, so it's ignored while
449 * recomputing effective
450 */
451 pl->prog = NULL;
452 break;
453 }
454 if (!old_prog)
455 return -ENOENT;
456 } else {
457 /* to maintain backward compatibility NONE and OVERRIDE cgroups
458 * allow detaching with invalid FD (prog==NULL)
459 */
460 pl = list_first_entry(progs, typeof(*pl), node);
461 old_prog = pl->prog;
462 pl->prog = NULL;
463 }
464
465 err = update_effective_progs(cgrp, type);
466 if (err)
467 goto cleanup;
468
469 /* now can actually delete it from this cgroup list */
470 list_del(&pl->node);
471 for_each_cgroup_storage_type(stype) {
472 bpf_cgroup_storage_unlink(pl->storage[stype]);
473 bpf_cgroup_storage_free(pl->storage[stype]);
474 }
475 kfree(pl);
476 if (list_empty(progs))
477 /* last program was detached, reset flags to zero */
478 cgrp->bpf.flags[type] = 0;
479
480 bpf_prog_put(old_prog);
481 static_branch_dec(&cgroup_bpf_enabled_key);
482 return 0;
483
484cleanup:
485 /* and restore back old_prog */
486 pl->prog = old_prog;
487 return err;
488}
489
490/* Must be called with cgroup_mutex held to avoid races. */
491int __cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
492 union bpf_attr __user *uattr)
493{
494 __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
495 enum bpf_attach_type type = attr->query.attach_type;
496 struct list_head *progs = &cgrp->bpf.progs[type];
497 u32 flags = cgrp->bpf.flags[type];
498 struct bpf_prog_array *effective;
499 int cnt, ret = 0, i;
500
501 effective = rcu_dereference_protected(cgrp->bpf.effective[type],
502 lockdep_is_held(&cgroup_mutex));
503
504 if (attr->query.query_flags & BPF_F_QUERY_EFFECTIVE)
505 cnt = bpf_prog_array_length(effective);
506 else
507 cnt = prog_list_length(progs);
508
509 if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)))
510 return -EFAULT;
511 if (copy_to_user(&uattr->query.prog_cnt, &cnt, sizeof(cnt)))
512 return -EFAULT;
513 if (attr->query.prog_cnt == 0 || !prog_ids || !cnt)
514 /* return early if user requested only program count + flags */
515 return 0;
516 if (attr->query.prog_cnt < cnt) {
517 cnt = attr->query.prog_cnt;
518 ret = -ENOSPC;
519 }
520
521 if (attr->query.query_flags & BPF_F_QUERY_EFFECTIVE) {
522 return bpf_prog_array_copy_to_user(effective, prog_ids, cnt);
523 } else {
524 struct bpf_prog_list *pl;
525 u32 id;
526
527 i = 0;
528 list_for_each_entry(pl, progs, node) {
529 id = pl->prog->aux->id;
530 if (copy_to_user(prog_ids + i, &id, sizeof(id)))
531 return -EFAULT;
532 if (++i == cnt)
533 break;
534 }
535 }
536 return ret;
537}
538
539int cgroup_bpf_prog_attach(const union bpf_attr *attr,
540 enum bpf_prog_type ptype, struct bpf_prog *prog)
541{
542 struct cgroup *cgrp;
543 int ret;
544
545 cgrp = cgroup_get_from_fd(attr->target_fd);
546 if (IS_ERR(cgrp))
547 return PTR_ERR(cgrp);
548
549 ret = cgroup_bpf_attach(cgrp, prog, attr->attach_type,
550 attr->attach_flags);
551 cgroup_put(cgrp);
552 return ret;
553}
554
555int cgroup_bpf_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
556{
557 struct bpf_prog *prog;
558 struct cgroup *cgrp;
559 int ret;
560
561 cgrp = cgroup_get_from_fd(attr->target_fd);
562 if (IS_ERR(cgrp))
563 return PTR_ERR(cgrp);
564
565 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
566 if (IS_ERR(prog))
567 prog = NULL;
568
569 ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type, 0);
570 if (prog)
571 bpf_prog_put(prog);
572
573 cgroup_put(cgrp);
574 return ret;
575}
576
577int cgroup_bpf_prog_query(const union bpf_attr *attr,
578 union bpf_attr __user *uattr)
579{
580 struct cgroup *cgrp;
581 int ret;
582
583 cgrp = cgroup_get_from_fd(attr->query.target_fd);
584 if (IS_ERR(cgrp))
585 return PTR_ERR(cgrp);
586
587 ret = cgroup_bpf_query(cgrp, attr, uattr);
588
589 cgroup_put(cgrp);
590 return ret;
591}
592
593/**
594 * __cgroup_bpf_run_filter_skb() - Run a program for packet filtering
595 * @sk: The socket sending or receiving traffic
596 * @skb: The skb that is being sent or received
597 * @type: The type of program to be exectuted
598 *
599 * If no socket is passed, or the socket is not of type INET or INET6,
600 * this function does nothing and returns 0.
601 *
602 * The program type passed in via @type must be suitable for network
603 * filtering. No further check is performed to assert that.
604 *
605 * For egress packets, this function can return:
606 * NET_XMIT_SUCCESS (0) - continue with packet output
607 * NET_XMIT_DROP (1) - drop packet and notify TCP to call cwr
608 * NET_XMIT_CN (2) - continue with packet output and notify TCP
609 * to call cwr
610 * -EPERM - drop packet
611 *
612 * For ingress packets, this function will return -EPERM if any
613 * attached program was found and if it returned != 1 during execution.
614 * Otherwise 0 is returned.
615 */
616int __cgroup_bpf_run_filter_skb(struct sock *sk,
617 struct sk_buff *skb,
618 enum bpf_attach_type type)
619{
620 unsigned int offset = skb->data - skb_network_header(skb);
621 struct sock *save_sk;
622 void *saved_data_end;
623 struct cgroup *cgrp;
624 int ret;
625
626 if (!sk || !sk_fullsock(sk))
627 return 0;
628
629 if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)
630 return 0;
631
632 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
633 save_sk = skb->sk;
634 skb->sk = sk;
635 __skb_push(skb, offset);
636
637 /* compute pointers for the bpf prog */
638 bpf_compute_and_save_data_end(skb, &saved_data_end);
639
640 if (type == BPF_CGROUP_INET_EGRESS) {
641 ret = BPF_PROG_CGROUP_INET_EGRESS_RUN_ARRAY(
642 cgrp->bpf.effective[type], skb, __bpf_prog_run_save_cb);
643 } else {
644 ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], skb,
645 __bpf_prog_run_save_cb);
646 ret = (ret == 1 ? 0 : -EPERM);
647 }
648 bpf_restore_data_end(skb, saved_data_end);
649 __skb_pull(skb, offset);
650 skb->sk = save_sk;
651
652 return ret;
653}
654EXPORT_SYMBOL(__cgroup_bpf_run_filter_skb);
655
656/**
657 * __cgroup_bpf_run_filter_sk() - Run a program on a sock
658 * @sk: sock structure to manipulate
659 * @type: The type of program to be exectuted
660 *
661 * socket is passed is expected to be of type INET or INET6.
662 *
663 * The program type passed in via @type must be suitable for sock
664 * filtering. No further check is performed to assert that.
665 *
666 * This function will return %-EPERM if any if an attached program was found
667 * and if it returned != 1 during execution. In all other cases, 0 is returned.
668 */
669int __cgroup_bpf_run_filter_sk(struct sock *sk,
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], sk, BPF_PROG_RUN);
676 return ret == 1 ? 0 : -EPERM;
677}
678EXPORT_SYMBOL(__cgroup_bpf_run_filter_sk);
679
680/**
681 * __cgroup_bpf_run_filter_sock_addr() - Run a program on a sock and
682 * provided by user sockaddr
683 * @sk: sock struct that will use sockaddr
684 * @uaddr: sockaddr struct provided by user
685 * @type: The type of program to be exectuted
686 * @t_ctx: Pointer to attach type specific context
687 *
688 * socket is expected to be of type INET or INET6.
689 *
690 * This function will return %-EPERM if an attached program is found and
691 * returned value != 1 during execution. In all other cases, 0 is returned.
692 */
693int __cgroup_bpf_run_filter_sock_addr(struct sock *sk,
694 struct sockaddr *uaddr,
695 enum bpf_attach_type type,
696 void *t_ctx)
697{
698 struct bpf_sock_addr_kern ctx = {
699 .sk = sk,
700 .uaddr = uaddr,
701 .t_ctx = t_ctx,
702 };
703 struct sockaddr_storage unspec;
704 struct cgroup *cgrp;
705 int ret;
706
707 /* Check socket family since not all sockets represent network
708 * endpoint (e.g. AF_UNIX).
709 */
710 if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)
711 return 0;
712
713 if (!ctx.uaddr) {
714 memset(&unspec, 0, sizeof(unspec));
715 ctx.uaddr = (struct sockaddr *)&unspec;
716 }
717
718 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
719 ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], &ctx, BPF_PROG_RUN);
720
721 return ret == 1 ? 0 : -EPERM;
722}
723EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_addr);
724
725/**
726 * __cgroup_bpf_run_filter_sock_ops() - Run a program on a sock
727 * @sk: socket to get cgroup from
728 * @sock_ops: bpf_sock_ops_kern struct to pass to program. Contains
729 * sk with connection information (IP addresses, etc.) May not contain
730 * cgroup info if it is a req sock.
731 * @type: The type of program to be exectuted
732 *
733 * socket passed is expected to be of type INET or INET6.
734 *
735 * The program type passed in via @type must be suitable for sock_ops
736 * filtering. No further check is performed to assert that.
737 *
738 * This function will return %-EPERM if any if an attached program was found
739 * and if it returned != 1 during execution. In all other cases, 0 is returned.
740 */
741int __cgroup_bpf_run_filter_sock_ops(struct sock *sk,
742 struct bpf_sock_ops_kern *sock_ops,
743 enum bpf_attach_type type)
744{
745 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
746 int ret;
747
748 ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], sock_ops,
749 BPF_PROG_RUN);
750 return ret == 1 ? 0 : -EPERM;
751}
752EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_ops);
753
754int __cgroup_bpf_check_dev_permission(short dev_type, u32 major, u32 minor,
755 short access, enum bpf_attach_type type)
756{
757 struct cgroup *cgrp;
758 struct bpf_cgroup_dev_ctx ctx = {
759 .access_type = (access << 16) | dev_type,
760 .major = major,
761 .minor = minor,
762 };
763 int allow = 1;
764
765 rcu_read_lock();
766 cgrp = task_dfl_cgroup(current);
767 allow = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], &ctx,
768 BPF_PROG_RUN);
769 rcu_read_unlock();
770
771 return !allow;
772}
773EXPORT_SYMBOL(__cgroup_bpf_check_dev_permission);
774
775static const struct bpf_func_proto *
776cgroup_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
777{
778 switch (func_id) {
779 case BPF_FUNC_map_lookup_elem:
780 return &bpf_map_lookup_elem_proto;
781 case BPF_FUNC_map_update_elem:
782 return &bpf_map_update_elem_proto;
783 case BPF_FUNC_map_delete_elem:
784 return &bpf_map_delete_elem_proto;
785 case BPF_FUNC_map_push_elem:
786 return &bpf_map_push_elem_proto;
787 case BPF_FUNC_map_pop_elem:
788 return &bpf_map_pop_elem_proto;
789 case BPF_FUNC_map_peek_elem:
790 return &bpf_map_peek_elem_proto;
791 case BPF_FUNC_get_current_uid_gid:
792 return &bpf_get_current_uid_gid_proto;
793 case BPF_FUNC_get_local_storage:
794 return &bpf_get_local_storage_proto;
795 case BPF_FUNC_get_current_cgroup_id:
796 return &bpf_get_current_cgroup_id_proto;
797 case BPF_FUNC_trace_printk:
798 if (capable(CAP_SYS_ADMIN))
799 return bpf_get_trace_printk_proto();
800 /* fall through */
801 default:
802 return NULL;
803 }
804}
805
806static const struct bpf_func_proto *
807cgroup_dev_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
808{
809 return cgroup_base_func_proto(func_id, prog);
810}
811
812static bool cgroup_dev_is_valid_access(int off, int size,
813 enum bpf_access_type type,
814 const struct bpf_prog *prog,
815 struct bpf_insn_access_aux *info)
816{
817 const int size_default = sizeof(__u32);
818
819 if (type == BPF_WRITE)
820 return false;
821
822 if (off < 0 || off + size > sizeof(struct bpf_cgroup_dev_ctx))
823 return false;
824 /* The verifier guarantees that size > 0. */
825 if (off % size != 0)
826 return false;
827
828 switch (off) {
829 case bpf_ctx_range(struct bpf_cgroup_dev_ctx, access_type):
830 bpf_ctx_record_field_size(info, size_default);
831 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
832 return false;
833 break;
834 default:
835 if (size != size_default)
836 return false;
837 }
838
839 return true;
840}
841
842const struct bpf_prog_ops cg_dev_prog_ops = {
843};
844
845const struct bpf_verifier_ops cg_dev_verifier_ops = {
846 .get_func_proto = cgroup_dev_func_proto,
847 .is_valid_access = cgroup_dev_is_valid_access,
848};
849
850/**
851 * __cgroup_bpf_run_filter_sysctl - Run a program on sysctl
852 *
853 * @head: sysctl table header
854 * @table: sysctl table
855 * @write: sysctl is being read (= 0) or written (= 1)
856 * @buf: pointer to buffer passed by user space
857 * @pcount: value-result argument: value is size of buffer pointed to by @buf,
858 * result is size of @new_buf if program set new value, initial value
859 * otherwise
860 * @ppos: value-result argument: value is position at which read from or write
861 * to sysctl is happening, result is new position if program overrode it,
862 * initial value otherwise
863 * @new_buf: pointer to pointer to new buffer that will be allocated if program
864 * overrides new value provided by user space on sysctl write
865 * NOTE: it's caller responsibility to free *new_buf if it was set
866 * @type: type of program to be executed
867 *
868 * Program is run when sysctl is being accessed, either read or written, and
869 * can allow or deny such access.
870 *
871 * This function will return %-EPERM if an attached program is found and
872 * returned value != 1 during execution. In all other cases 0 is returned.
873 */
874int __cgroup_bpf_run_filter_sysctl(struct ctl_table_header *head,
875 struct ctl_table *table, int write,
876 void __user *buf, size_t *pcount,
877 loff_t *ppos, void **new_buf,
878 enum bpf_attach_type type)
879{
880 struct bpf_sysctl_kern ctx = {
881 .head = head,
882 .table = table,
883 .write = write,
884 .ppos = ppos,
885 .cur_val = NULL,
886 .cur_len = PAGE_SIZE,
887 .new_val = NULL,
888 .new_len = 0,
889 .new_updated = 0,
890 };
891 struct cgroup *cgrp;
892 int ret;
893
894 ctx.cur_val = kmalloc_track_caller(ctx.cur_len, GFP_KERNEL);
895 if (ctx.cur_val) {
896 mm_segment_t old_fs;
897 loff_t pos = 0;
898
899 old_fs = get_fs();
900 set_fs(KERNEL_DS);
901 if (table->proc_handler(table, 0, (void __user *)ctx.cur_val,
902 &ctx.cur_len, &pos)) {
903 /* Let BPF program decide how to proceed. */
904 ctx.cur_len = 0;
905 }
906 set_fs(old_fs);
907 } else {
908 /* Let BPF program decide how to proceed. */
909 ctx.cur_len = 0;
910 }
911
912 if (write && buf && *pcount) {
913 /* BPF program should be able to override new value with a
914 * buffer bigger than provided by user.
915 */
916 ctx.new_val = kmalloc_track_caller(PAGE_SIZE, GFP_KERNEL);
917 ctx.new_len = min_t(size_t, PAGE_SIZE, *pcount);
918 if (!ctx.new_val ||
919 copy_from_user(ctx.new_val, buf, ctx.new_len))
920 /* Let BPF program decide how to proceed. */
921 ctx.new_len = 0;
922 }
923
924 rcu_read_lock();
925 cgrp = task_dfl_cgroup(current);
926 ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[type], &ctx, BPF_PROG_RUN);
927 rcu_read_unlock();
928
929 kfree(ctx.cur_val);
930
931 if (ret == 1 && ctx.new_updated) {
932 *new_buf = ctx.new_val;
933 *pcount = ctx.new_len;
934 } else {
935 kfree(ctx.new_val);
936 }
937
938 return ret == 1 ? 0 : -EPERM;
939}
940EXPORT_SYMBOL(__cgroup_bpf_run_filter_sysctl);
941
942#ifdef CONFIG_NET
943static bool __cgroup_bpf_prog_array_is_empty(struct cgroup *cgrp,
944 enum bpf_attach_type attach_type)
945{
946 struct bpf_prog_array *prog_array;
947 bool empty;
948
949 rcu_read_lock();
950 prog_array = rcu_dereference(cgrp->bpf.effective[attach_type]);
951 empty = bpf_prog_array_is_empty(prog_array);
952 rcu_read_unlock();
953
954 return empty;
955}
956
957static int sockopt_alloc_buf(struct bpf_sockopt_kern *ctx, int max_optlen)
958{
959 if (unlikely(max_optlen > PAGE_SIZE) || max_optlen < 0)
960 return -EINVAL;
961
962 ctx->optval = kzalloc(max_optlen, GFP_USER);
963 if (!ctx->optval)
964 return -ENOMEM;
965
966 ctx->optval_end = ctx->optval + max_optlen;
967
968 return 0;
969}
970
971static void sockopt_free_buf(struct bpf_sockopt_kern *ctx)
972{
973 kfree(ctx->optval);
974}
975
976int __cgroup_bpf_run_filter_setsockopt(struct sock *sk, int *level,
977 int *optname, char __user *optval,
978 int *optlen, char **kernel_optval)
979{
980 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
981 struct bpf_sockopt_kern ctx = {
982 .sk = sk,
983 .level = *level,
984 .optname = *optname,
985 };
986 int ret, max_optlen;
987
988 /* Opportunistic check to see whether we have any BPF program
989 * attached to the hook so we don't waste time allocating
990 * memory and locking the socket.
991 */
992 if (!cgroup_bpf_enabled ||
993 __cgroup_bpf_prog_array_is_empty(cgrp, BPF_CGROUP_SETSOCKOPT))
994 return 0;
995
996 /* Allocate a bit more than the initial user buffer for
997 * BPF program. The canonical use case is overriding
998 * TCP_CONGESTION(nv) to TCP_CONGESTION(cubic).
999 */
1000 max_optlen = max_t(int, 16, *optlen);
1001
1002 ret = sockopt_alloc_buf(&ctx, max_optlen);
1003 if (ret)
1004 return ret;
1005
1006 ctx.optlen = *optlen;
1007
1008 if (copy_from_user(ctx.optval, optval, *optlen) != 0) {
1009 ret = -EFAULT;
1010 goto out;
1011 }
1012
1013 lock_sock(sk);
1014 ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[BPF_CGROUP_SETSOCKOPT],
1015 &ctx, BPF_PROG_RUN);
1016 release_sock(sk);
1017
1018 if (!ret) {
1019 ret = -EPERM;
1020 goto out;
1021 }
1022
1023 if (ctx.optlen == -1) {
1024 /* optlen set to -1, bypass kernel */
1025 ret = 1;
1026 } else if (ctx.optlen > max_optlen || ctx.optlen < -1) {
1027 /* optlen is out of bounds */
1028 ret = -EFAULT;
1029 } else {
1030 /* optlen within bounds, run kernel handler */
1031 ret = 0;
1032
1033 /* export any potential modifications */
1034 *level = ctx.level;
1035 *optname = ctx.optname;
1036 *optlen = ctx.optlen;
1037 *kernel_optval = ctx.optval;
1038 }
1039
1040out:
1041 if (ret)
1042 sockopt_free_buf(&ctx);
1043 return ret;
1044}
1045EXPORT_SYMBOL(__cgroup_bpf_run_filter_setsockopt);
1046
1047int __cgroup_bpf_run_filter_getsockopt(struct sock *sk, int level,
1048 int optname, char __user *optval,
1049 int __user *optlen, int max_optlen,
1050 int retval)
1051{
1052 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1053 struct bpf_sockopt_kern ctx = {
1054 .sk = sk,
1055 .level = level,
1056 .optname = optname,
1057 .retval = retval,
1058 };
1059 int ret;
1060
1061 /* Opportunistic check to see whether we have any BPF program
1062 * attached to the hook so we don't waste time allocating
1063 * memory and locking the socket.
1064 */
1065 if (!cgroup_bpf_enabled ||
1066 __cgroup_bpf_prog_array_is_empty(cgrp, BPF_CGROUP_GETSOCKOPT))
1067 return retval;
1068
1069 ret = sockopt_alloc_buf(&ctx, max_optlen);
1070 if (ret)
1071 return ret;
1072
1073 ctx.optlen = max_optlen;
1074
1075 if (!retval) {
1076 /* If kernel getsockopt finished successfully,
1077 * copy whatever was returned to the user back
1078 * into our temporary buffer. Set optlen to the
1079 * one that kernel returned as well to let
1080 * BPF programs inspect the value.
1081 */
1082
1083 if (get_user(ctx.optlen, optlen)) {
1084 ret = -EFAULT;
1085 goto out;
1086 }
1087
1088 if (ctx.optlen > max_optlen)
1089 ctx.optlen = max_optlen;
1090
1091 if (copy_from_user(ctx.optval, optval, ctx.optlen) != 0) {
1092 ret = -EFAULT;
1093 goto out;
1094 }
1095 }
1096
1097 lock_sock(sk);
1098 ret = BPF_PROG_RUN_ARRAY(cgrp->bpf.effective[BPF_CGROUP_GETSOCKOPT],
1099 &ctx, BPF_PROG_RUN);
1100 release_sock(sk);
1101
1102 if (!ret) {
1103 ret = -EPERM;
1104 goto out;
1105 }
1106
1107 if (ctx.optlen > max_optlen) {
1108 ret = -EFAULT;
1109 goto out;
1110 }
1111
1112 /* BPF programs only allowed to set retval to 0, not some
1113 * arbitrary value.
1114 */
1115 if (ctx.retval != 0 && ctx.retval != retval) {
1116 ret = -EFAULT;
1117 goto out;
1118 }
1119
1120 if (copy_to_user(optval, ctx.optval, ctx.optlen) ||
1121 put_user(ctx.optlen, optlen)) {
1122 ret = -EFAULT;
1123 goto out;
1124 }
1125
1126 ret = ctx.retval;
1127
1128out:
1129 sockopt_free_buf(&ctx);
1130 return ret;
1131}
1132EXPORT_SYMBOL(__cgroup_bpf_run_filter_getsockopt);
1133#endif
1134
1135static ssize_t sysctl_cpy_dir(const struct ctl_dir *dir, char **bufp,
1136 size_t *lenp)
1137{
1138 ssize_t tmp_ret = 0, ret;
1139
1140 if (dir->header.parent) {
1141 tmp_ret = sysctl_cpy_dir(dir->header.parent, bufp, lenp);
1142 if (tmp_ret < 0)
1143 return tmp_ret;
1144 }
1145
1146 ret = strscpy(*bufp, dir->header.ctl_table[0].procname, *lenp);
1147 if (ret < 0)
1148 return ret;
1149 *bufp += ret;
1150 *lenp -= ret;
1151 ret += tmp_ret;
1152
1153 /* Avoid leading slash. */
1154 if (!ret)
1155 return ret;
1156
1157 tmp_ret = strscpy(*bufp, "/", *lenp);
1158 if (tmp_ret < 0)
1159 return tmp_ret;
1160 *bufp += tmp_ret;
1161 *lenp -= tmp_ret;
1162
1163 return ret + tmp_ret;
1164}
1165
1166BPF_CALL_4(bpf_sysctl_get_name, struct bpf_sysctl_kern *, ctx, char *, buf,
1167 size_t, buf_len, u64, flags)
1168{
1169 ssize_t tmp_ret = 0, ret;
1170
1171 if (!buf)
1172 return -EINVAL;
1173
1174 if (!(flags & BPF_F_SYSCTL_BASE_NAME)) {
1175 if (!ctx->head)
1176 return -EINVAL;
1177 tmp_ret = sysctl_cpy_dir(ctx->head->parent, &buf, &buf_len);
1178 if (tmp_ret < 0)
1179 return tmp_ret;
1180 }
1181
1182 ret = strscpy(buf, ctx->table->procname, buf_len);
1183
1184 return ret < 0 ? ret : tmp_ret + ret;
1185}
1186
1187static const struct bpf_func_proto bpf_sysctl_get_name_proto = {
1188 .func = bpf_sysctl_get_name,
1189 .gpl_only = false,
1190 .ret_type = RET_INTEGER,
1191 .arg1_type = ARG_PTR_TO_CTX,
1192 .arg2_type = ARG_PTR_TO_MEM,
1193 .arg3_type = ARG_CONST_SIZE,
1194 .arg4_type = ARG_ANYTHING,
1195};
1196
1197static int copy_sysctl_value(char *dst, size_t dst_len, char *src,
1198 size_t src_len)
1199{
1200 if (!dst)
1201 return -EINVAL;
1202
1203 if (!dst_len)
1204 return -E2BIG;
1205
1206 if (!src || !src_len) {
1207 memset(dst, 0, dst_len);
1208 return -EINVAL;
1209 }
1210
1211 memcpy(dst, src, min(dst_len, src_len));
1212
1213 if (dst_len > src_len) {
1214 memset(dst + src_len, '\0', dst_len - src_len);
1215 return src_len;
1216 }
1217
1218 dst[dst_len - 1] = '\0';
1219
1220 return -E2BIG;
1221}
1222
1223BPF_CALL_3(bpf_sysctl_get_current_value, struct bpf_sysctl_kern *, ctx,
1224 char *, buf, size_t, buf_len)
1225{
1226 return copy_sysctl_value(buf, buf_len, ctx->cur_val, ctx->cur_len);
1227}
1228
1229static const struct bpf_func_proto bpf_sysctl_get_current_value_proto = {
1230 .func = bpf_sysctl_get_current_value,
1231 .gpl_only = false,
1232 .ret_type = RET_INTEGER,
1233 .arg1_type = ARG_PTR_TO_CTX,
1234 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1235 .arg3_type = ARG_CONST_SIZE,
1236};
1237
1238BPF_CALL_3(bpf_sysctl_get_new_value, struct bpf_sysctl_kern *, ctx, char *, buf,
1239 size_t, buf_len)
1240{
1241 if (!ctx->write) {
1242 if (buf && buf_len)
1243 memset(buf, '\0', buf_len);
1244 return -EINVAL;
1245 }
1246 return copy_sysctl_value(buf, buf_len, ctx->new_val, ctx->new_len);
1247}
1248
1249static const struct bpf_func_proto bpf_sysctl_get_new_value_proto = {
1250 .func = bpf_sysctl_get_new_value,
1251 .gpl_only = false,
1252 .ret_type = RET_INTEGER,
1253 .arg1_type = ARG_PTR_TO_CTX,
1254 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1255 .arg3_type = ARG_CONST_SIZE,
1256};
1257
1258BPF_CALL_3(bpf_sysctl_set_new_value, struct bpf_sysctl_kern *, ctx,
1259 const char *, buf, size_t, buf_len)
1260{
1261 if (!ctx->write || !ctx->new_val || !ctx->new_len || !buf || !buf_len)
1262 return -EINVAL;
1263
1264 if (buf_len > PAGE_SIZE - 1)
1265 return -E2BIG;
1266
1267 memcpy(ctx->new_val, buf, buf_len);
1268 ctx->new_len = buf_len;
1269 ctx->new_updated = 1;
1270
1271 return 0;
1272}
1273
1274static const struct bpf_func_proto bpf_sysctl_set_new_value_proto = {
1275 .func = bpf_sysctl_set_new_value,
1276 .gpl_only = false,
1277 .ret_type = RET_INTEGER,
1278 .arg1_type = ARG_PTR_TO_CTX,
1279 .arg2_type = ARG_PTR_TO_MEM,
1280 .arg3_type = ARG_CONST_SIZE,
1281};
1282
1283static const struct bpf_func_proto *
1284sysctl_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1285{
1286 switch (func_id) {
1287 case BPF_FUNC_strtol:
1288 return &bpf_strtol_proto;
1289 case BPF_FUNC_strtoul:
1290 return &bpf_strtoul_proto;
1291 case BPF_FUNC_sysctl_get_name:
1292 return &bpf_sysctl_get_name_proto;
1293 case BPF_FUNC_sysctl_get_current_value:
1294 return &bpf_sysctl_get_current_value_proto;
1295 case BPF_FUNC_sysctl_get_new_value:
1296 return &bpf_sysctl_get_new_value_proto;
1297 case BPF_FUNC_sysctl_set_new_value:
1298 return &bpf_sysctl_set_new_value_proto;
1299 default:
1300 return cgroup_base_func_proto(func_id, prog);
1301 }
1302}
1303
1304static bool sysctl_is_valid_access(int off, int size, enum bpf_access_type type,
1305 const struct bpf_prog *prog,
1306 struct bpf_insn_access_aux *info)
1307{
1308 const int size_default = sizeof(__u32);
1309
1310 if (off < 0 || off + size > sizeof(struct bpf_sysctl) || off % size)
1311 return false;
1312
1313 switch (off) {
1314 case bpf_ctx_range(struct bpf_sysctl, write):
1315 if (type != BPF_READ)
1316 return false;
1317 bpf_ctx_record_field_size(info, size_default);
1318 return bpf_ctx_narrow_access_ok(off, size, size_default);
1319 case bpf_ctx_range(struct bpf_sysctl, file_pos):
1320 if (type == BPF_READ) {
1321 bpf_ctx_record_field_size(info, size_default);
1322 return bpf_ctx_narrow_access_ok(off, size, size_default);
1323 } else {
1324 return size == size_default;
1325 }
1326 default:
1327 return false;
1328 }
1329}
1330
1331static u32 sysctl_convert_ctx_access(enum bpf_access_type type,
1332 const struct bpf_insn *si,
1333 struct bpf_insn *insn_buf,
1334 struct bpf_prog *prog, u32 *target_size)
1335{
1336 struct bpf_insn *insn = insn_buf;
1337 u32 read_size;
1338
1339 switch (si->off) {
1340 case offsetof(struct bpf_sysctl, write):
1341 *insn++ = BPF_LDX_MEM(
1342 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
1343 bpf_target_off(struct bpf_sysctl_kern, write,
1344 FIELD_SIZEOF(struct bpf_sysctl_kern,
1345 write),
1346 target_size));
1347 break;
1348 case offsetof(struct bpf_sysctl, file_pos):
1349 /* ppos is a pointer so it should be accessed via indirect
1350 * loads and stores. Also for stores additional temporary
1351 * register is used since neither src_reg nor dst_reg can be
1352 * overridden.
1353 */
1354 if (type == BPF_WRITE) {
1355 int treg = BPF_REG_9;
1356
1357 if (si->src_reg == treg || si->dst_reg == treg)
1358 --treg;
1359 if (si->src_reg == treg || si->dst_reg == treg)
1360 --treg;
1361 *insn++ = BPF_STX_MEM(
1362 BPF_DW, si->dst_reg, treg,
1363 offsetof(struct bpf_sysctl_kern, tmp_reg));
1364 *insn++ = BPF_LDX_MEM(
1365 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos),
1366 treg, si->dst_reg,
1367 offsetof(struct bpf_sysctl_kern, ppos));
1368 *insn++ = BPF_STX_MEM(
1369 BPF_SIZEOF(u32), treg, si->src_reg,
1370 bpf_ctx_narrow_access_offset(
1371 0, sizeof(u32), sizeof(loff_t)));
1372 *insn++ = BPF_LDX_MEM(
1373 BPF_DW, treg, si->dst_reg,
1374 offsetof(struct bpf_sysctl_kern, tmp_reg));
1375 } else {
1376 *insn++ = BPF_LDX_MEM(
1377 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos),
1378 si->dst_reg, si->src_reg,
1379 offsetof(struct bpf_sysctl_kern, ppos));
1380 read_size = bpf_size_to_bytes(BPF_SIZE(si->code));
1381 *insn++ = BPF_LDX_MEM(
1382 BPF_SIZE(si->code), si->dst_reg, si->dst_reg,
1383 bpf_ctx_narrow_access_offset(
1384 0, read_size, sizeof(loff_t)));
1385 }
1386 *target_size = sizeof(u32);
1387 break;
1388 }
1389
1390 return insn - insn_buf;
1391}
1392
1393const struct bpf_verifier_ops cg_sysctl_verifier_ops = {
1394 .get_func_proto = sysctl_func_proto,
1395 .is_valid_access = sysctl_is_valid_access,
1396 .convert_ctx_access = sysctl_convert_ctx_access,
1397};
1398
1399const struct bpf_prog_ops cg_sysctl_prog_ops = {
1400};
1401
1402static const struct bpf_func_proto *
1403cg_sockopt_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1404{
1405 switch (func_id) {
1406#ifdef CONFIG_NET
1407 case BPF_FUNC_sk_storage_get:
1408 return &bpf_sk_storage_get_proto;
1409 case BPF_FUNC_sk_storage_delete:
1410 return &bpf_sk_storage_delete_proto;
1411#endif
1412#ifdef CONFIG_INET
1413 case BPF_FUNC_tcp_sock:
1414 return &bpf_tcp_sock_proto;
1415#endif
1416 default:
1417 return cgroup_base_func_proto(func_id, prog);
1418 }
1419}
1420
1421static bool cg_sockopt_is_valid_access(int off, int size,
1422 enum bpf_access_type type,
1423 const struct bpf_prog *prog,
1424 struct bpf_insn_access_aux *info)
1425{
1426 const int size_default = sizeof(__u32);
1427
1428 if (off < 0 || off >= sizeof(struct bpf_sockopt))
1429 return false;
1430
1431 if (off % size != 0)
1432 return false;
1433
1434 if (type == BPF_WRITE) {
1435 switch (off) {
1436 case offsetof(struct bpf_sockopt, retval):
1437 if (size != size_default)
1438 return false;
1439 return prog->expected_attach_type ==
1440 BPF_CGROUP_GETSOCKOPT;
1441 case offsetof(struct bpf_sockopt, optname):
1442 /* fallthrough */
1443 case offsetof(struct bpf_sockopt, level):
1444 if (size != size_default)
1445 return false;
1446 return prog->expected_attach_type ==
1447 BPF_CGROUP_SETSOCKOPT;
1448 case offsetof(struct bpf_sockopt, optlen):
1449 return size == size_default;
1450 default:
1451 return false;
1452 }
1453 }
1454
1455 switch (off) {
1456 case offsetof(struct bpf_sockopt, sk):
1457 if (size != sizeof(__u64))
1458 return false;
1459 info->reg_type = PTR_TO_SOCKET;
1460 break;
1461 case offsetof(struct bpf_sockopt, optval):
1462 if (size != sizeof(__u64))
1463 return false;
1464 info->reg_type = PTR_TO_PACKET;
1465 break;
1466 case offsetof(struct bpf_sockopt, optval_end):
1467 if (size != sizeof(__u64))
1468 return false;
1469 info->reg_type = PTR_TO_PACKET_END;
1470 break;
1471 case offsetof(struct bpf_sockopt, retval):
1472 if (size != size_default)
1473 return false;
1474 return prog->expected_attach_type == BPF_CGROUP_GETSOCKOPT;
1475 default:
1476 if (size != size_default)
1477 return false;
1478 break;
1479 }
1480 return true;
1481}
1482
1483#define CG_SOCKOPT_ACCESS_FIELD(T, F) \
1484 T(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, F), \
1485 si->dst_reg, si->src_reg, \
1486 offsetof(struct bpf_sockopt_kern, F))
1487
1488static u32 cg_sockopt_convert_ctx_access(enum bpf_access_type type,
1489 const struct bpf_insn *si,
1490 struct bpf_insn *insn_buf,
1491 struct bpf_prog *prog,
1492 u32 *target_size)
1493{
1494 struct bpf_insn *insn = insn_buf;
1495
1496 switch (si->off) {
1497 case offsetof(struct bpf_sockopt, sk):
1498 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, sk);
1499 break;
1500 case offsetof(struct bpf_sockopt, level):
1501 if (type == BPF_WRITE)
1502 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, level);
1503 else
1504 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, level);
1505 break;
1506 case offsetof(struct bpf_sockopt, optname):
1507 if (type == BPF_WRITE)
1508 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, optname);
1509 else
1510 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optname);
1511 break;
1512 case offsetof(struct bpf_sockopt, optlen):
1513 if (type == BPF_WRITE)
1514 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, optlen);
1515 else
1516 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optlen);
1517 break;
1518 case offsetof(struct bpf_sockopt, retval):
1519 if (type == BPF_WRITE)
1520 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, retval);
1521 else
1522 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, retval);
1523 break;
1524 case offsetof(struct bpf_sockopt, optval):
1525 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optval);
1526 break;
1527 case offsetof(struct bpf_sockopt, optval_end):
1528 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optval_end);
1529 break;
1530 }
1531
1532 return insn - insn_buf;
1533}
1534
1535static int cg_sockopt_get_prologue(struct bpf_insn *insn_buf,
1536 bool direct_write,
1537 const struct bpf_prog *prog)
1538{
1539 /* Nothing to do for sockopt argument. The data is kzalloc'ated.
1540 */
1541 return 0;
1542}
1543
1544const struct bpf_verifier_ops cg_sockopt_verifier_ops = {
1545 .get_func_proto = cg_sockopt_func_proto,
1546 .is_valid_access = cg_sockopt_is_valid_access,
1547 .convert_ctx_access = cg_sockopt_convert_ctx_access,
1548 .gen_prologue = cg_sockopt_get_prologue,
1549};
1550
1551const struct bpf_prog_ops cg_sockopt_prog_ops = {
1552};