1
2
3
4
5
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_ARRAY_FALSE(cgroup_bpf_enabled_key, MAX_CGROUP_BPF_ATTACH_TYPE);
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
31static void bpf_cgroup_storages_free(struct bpf_cgroup_storage *storages[])
32{
33 enum bpf_cgroup_storage_type stype;
34
35 for_each_cgroup_storage_type(stype)
36 bpf_cgroup_storage_free(storages[stype]);
37}
38
39static int bpf_cgroup_storages_alloc(struct bpf_cgroup_storage *storages[],
40 struct bpf_cgroup_storage *new_storages[],
41 enum bpf_attach_type type,
42 struct bpf_prog *prog,
43 struct cgroup *cgrp)
44{
45 enum bpf_cgroup_storage_type stype;
46 struct bpf_cgroup_storage_key key;
47 struct bpf_map *map;
48
49 key.cgroup_inode_id = cgroup_id(cgrp);
50 key.attach_type = type;
51
52 for_each_cgroup_storage_type(stype) {
53 map = prog->aux->cgroup_storage[stype];
54 if (!map)
55 continue;
56
57 storages[stype] = cgroup_storage_lookup((void *)map, &key, false);
58 if (storages[stype])
59 continue;
60
61 storages[stype] = bpf_cgroup_storage_alloc(prog, stype);
62 if (IS_ERR(storages[stype])) {
63 bpf_cgroup_storages_free(new_storages);
64 return -ENOMEM;
65 }
66
67 new_storages[stype] = storages[stype];
68 }
69
70 return 0;
71}
72
73static void bpf_cgroup_storages_assign(struct bpf_cgroup_storage *dst[],
74 struct bpf_cgroup_storage *src[])
75{
76 enum bpf_cgroup_storage_type stype;
77
78 for_each_cgroup_storage_type(stype)
79 dst[stype] = src[stype];
80}
81
82static void bpf_cgroup_storages_link(struct bpf_cgroup_storage *storages[],
83 struct cgroup *cgrp,
84 enum bpf_attach_type attach_type)
85{
86 enum bpf_cgroup_storage_type stype;
87
88 for_each_cgroup_storage_type(stype)
89 bpf_cgroup_storage_link(storages[stype], cgrp, attach_type);
90}
91
92
93
94
95
96
97static void bpf_cgroup_link_auto_detach(struct bpf_cgroup_link *link)
98{
99 cgroup_put(link->cgroup);
100 link->cgroup = NULL;
101}
102
103
104
105
106
107
108static void cgroup_bpf_release(struct work_struct *work)
109{
110 struct cgroup *p, *cgrp = container_of(work, struct cgroup,
111 bpf.release_work);
112 struct bpf_prog_array *old_array;
113 struct list_head *storages = &cgrp->bpf.storages;
114 struct bpf_cgroup_storage *storage, *stmp;
115
116 unsigned int atype;
117
118 mutex_lock(&cgroup_mutex);
119
120 for (atype = 0; atype < ARRAY_SIZE(cgrp->bpf.progs); atype++) {
121 struct list_head *progs = &cgrp->bpf.progs[atype];
122 struct bpf_prog_list *pl, *pltmp;
123
124 list_for_each_entry_safe(pl, pltmp, progs, node) {
125 list_del(&pl->node);
126 if (pl->prog)
127 bpf_prog_put(pl->prog);
128 if (pl->link)
129 bpf_cgroup_link_auto_detach(pl->link);
130 kfree(pl);
131 static_branch_dec(&cgroup_bpf_enabled_key[atype]);
132 }
133 old_array = rcu_dereference_protected(
134 cgrp->bpf.effective[atype],
135 lockdep_is_held(&cgroup_mutex));
136 bpf_prog_array_free(old_array);
137 }
138
139 list_for_each_entry_safe(storage, stmp, storages, list_cg) {
140 bpf_cgroup_storage_unlink(storage);
141 bpf_cgroup_storage_free(storage);
142 }
143
144 mutex_unlock(&cgroup_mutex);
145
146 for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p))
147 cgroup_bpf_put(p);
148
149 percpu_ref_exit(&cgrp->bpf.refcnt);
150 cgroup_put(cgrp);
151}
152
153
154
155
156
157
158static void cgroup_bpf_release_fn(struct percpu_ref *ref)
159{
160 struct cgroup *cgrp = container_of(ref, struct cgroup, bpf.refcnt);
161
162 INIT_WORK(&cgrp->bpf.release_work, cgroup_bpf_release);
163 queue_work(system_wq, &cgrp->bpf.release_work);
164}
165
166
167
168
169static struct bpf_prog *prog_list_prog(struct bpf_prog_list *pl)
170{
171 if (pl->prog)
172 return pl->prog;
173 if (pl->link)
174 return pl->link->link.prog;
175 return NULL;
176}
177
178
179
180
181static u32 prog_list_length(struct list_head *head)
182{
183 struct bpf_prog_list *pl;
184 u32 cnt = 0;
185
186 list_for_each_entry(pl, head, node) {
187 if (!prog_list_prog(pl))
188 continue;
189 cnt++;
190 }
191 return cnt;
192}
193
194
195
196
197
198static bool hierarchy_allows_attach(struct cgroup *cgrp,
199 enum cgroup_bpf_attach_type atype)
200{
201 struct cgroup *p;
202
203 p = cgroup_parent(cgrp);
204 if (!p)
205 return true;
206 do {
207 u32 flags = p->bpf.flags[atype];
208 u32 cnt;
209
210 if (flags & BPF_F_ALLOW_MULTI)
211 return true;
212 cnt = prog_list_length(&p->bpf.progs[atype]);
213 WARN_ON_ONCE(cnt > 1);
214 if (cnt == 1)
215 return !!(flags & BPF_F_ALLOW_OVERRIDE);
216 p = cgroup_parent(p);
217 } while (p);
218 return true;
219}
220
221
222
223
224
225
226
227static int compute_effective_progs(struct cgroup *cgrp,
228 enum cgroup_bpf_attach_type atype,
229 struct bpf_prog_array **array)
230{
231 struct bpf_prog_array_item *item;
232 struct bpf_prog_array *progs;
233 struct bpf_prog_list *pl;
234 struct cgroup *p = cgrp;
235 int cnt = 0;
236
237
238 do {
239 if (cnt == 0 || (p->bpf.flags[atype] & BPF_F_ALLOW_MULTI))
240 cnt += prog_list_length(&p->bpf.progs[atype]);
241 p = cgroup_parent(p);
242 } while (p);
243
244 progs = bpf_prog_array_alloc(cnt, GFP_KERNEL);
245 if (!progs)
246 return -ENOMEM;
247
248
249 cnt = 0;
250 p = cgrp;
251 do {
252 if (cnt > 0 && !(p->bpf.flags[atype] & BPF_F_ALLOW_MULTI))
253 continue;
254
255 list_for_each_entry(pl, &p->bpf.progs[atype], node) {
256 if (!prog_list_prog(pl))
257 continue;
258
259 item = &progs->items[cnt];
260 item->prog = prog_list_prog(pl);
261 bpf_cgroup_storages_assign(item->cgroup_storage,
262 pl->storage);
263 cnt++;
264 }
265 } while ((p = cgroup_parent(p)));
266
267 *array = progs;
268 return 0;
269}
270
271static void activate_effective_progs(struct cgroup *cgrp,
272 enum cgroup_bpf_attach_type atype,
273 struct bpf_prog_array *old_array)
274{
275 old_array = rcu_replace_pointer(cgrp->bpf.effective[atype], old_array,
276 lockdep_is_held(&cgroup_mutex));
277
278
279
280 bpf_prog_array_free(old_array);
281}
282
283
284
285
286
287int cgroup_bpf_inherit(struct cgroup *cgrp)
288{
289
290
291
292#define NR ARRAY_SIZE(cgrp->bpf.effective)
293 struct bpf_prog_array *arrays[NR] = {};
294 struct cgroup *p;
295 int ret, i;
296
297 ret = percpu_ref_init(&cgrp->bpf.refcnt, cgroup_bpf_release_fn, 0,
298 GFP_KERNEL);
299 if (ret)
300 return ret;
301
302 for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p))
303 cgroup_bpf_get(p);
304
305 for (i = 0; i < NR; i++)
306 INIT_LIST_HEAD(&cgrp->bpf.progs[i]);
307
308 INIT_LIST_HEAD(&cgrp->bpf.storages);
309
310 for (i = 0; i < NR; i++)
311 if (compute_effective_progs(cgrp, i, &arrays[i]))
312 goto cleanup;
313
314 for (i = 0; i < NR; i++)
315 activate_effective_progs(cgrp, i, arrays[i]);
316
317 return 0;
318cleanup:
319 for (i = 0; i < NR; i++)
320 bpf_prog_array_free(arrays[i]);
321
322 for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p))
323 cgroup_bpf_put(p);
324
325 percpu_ref_exit(&cgrp->bpf.refcnt);
326
327 return -ENOMEM;
328}
329
330static int update_effective_progs(struct cgroup *cgrp,
331 enum cgroup_bpf_attach_type atype)
332{
333 struct cgroup_subsys_state *css;
334 int err;
335
336
337 css_for_each_descendant_pre(css, &cgrp->self) {
338 struct cgroup *desc = container_of(css, struct cgroup, self);
339
340 if (percpu_ref_is_zero(&desc->bpf.refcnt))
341 continue;
342
343 err = compute_effective_progs(desc, atype, &desc->bpf.inactive);
344 if (err)
345 goto cleanup;
346 }
347
348
349 css_for_each_descendant_pre(css, &cgrp->self) {
350 struct cgroup *desc = container_of(css, struct cgroup, self);
351
352 if (percpu_ref_is_zero(&desc->bpf.refcnt)) {
353 if (unlikely(desc->bpf.inactive)) {
354 bpf_prog_array_free(desc->bpf.inactive);
355 desc->bpf.inactive = NULL;
356 }
357 continue;
358 }
359
360 activate_effective_progs(desc, atype, desc->bpf.inactive);
361 desc->bpf.inactive = NULL;
362 }
363
364 return 0;
365
366cleanup:
367
368
369
370 css_for_each_descendant_pre(css, &cgrp->self) {
371 struct cgroup *desc = container_of(css, struct cgroup, self);
372
373 bpf_prog_array_free(desc->bpf.inactive);
374 desc->bpf.inactive = NULL;
375 }
376
377 return err;
378}
379
380#define BPF_CGROUP_MAX_PROGS 64
381
382static struct bpf_prog_list *find_attach_entry(struct list_head *progs,
383 struct bpf_prog *prog,
384 struct bpf_cgroup_link *link,
385 struct bpf_prog *replace_prog,
386 bool allow_multi)
387{
388 struct bpf_prog_list *pl;
389
390
391 if (!allow_multi) {
392 if (list_empty(progs))
393 return NULL;
394 return list_first_entry(progs, typeof(*pl), node);
395 }
396
397 list_for_each_entry(pl, progs, node) {
398 if (prog && pl->prog == prog && prog != replace_prog)
399
400 return ERR_PTR(-EINVAL);
401 if (link && pl->link == link)
402
403 return ERR_PTR(-EINVAL);
404 }
405
406
407 if (replace_prog) {
408 list_for_each_entry(pl, progs, node) {
409 if (pl->prog == replace_prog)
410
411 return pl;
412 }
413
414 return ERR_PTR(-ENOENT);
415 }
416
417 return NULL;
418}
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433static int __cgroup_bpf_attach(struct cgroup *cgrp,
434 struct bpf_prog *prog, struct bpf_prog *replace_prog,
435 struct bpf_cgroup_link *link,
436 enum bpf_attach_type type, u32 flags)
437{
438 u32 saved_flags = (flags & (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI));
439 struct bpf_prog *old_prog = NULL;
440 struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE] = {};
441 struct bpf_cgroup_storage *new_storage[MAX_BPF_CGROUP_STORAGE_TYPE] = {};
442 enum cgroup_bpf_attach_type atype;
443 struct bpf_prog_list *pl;
444 struct list_head *progs;
445 int err;
446
447 if (((flags & BPF_F_ALLOW_OVERRIDE) && (flags & BPF_F_ALLOW_MULTI)) ||
448 ((flags & BPF_F_REPLACE) && !(flags & BPF_F_ALLOW_MULTI)))
449
450 return -EINVAL;
451 if (link && (prog || replace_prog))
452
453 return -EINVAL;
454 if (!!replace_prog != !!(flags & BPF_F_REPLACE))
455
456 return -EINVAL;
457
458 atype = to_cgroup_bpf_attach_type(type);
459 if (atype < 0)
460 return -EINVAL;
461
462 progs = &cgrp->bpf.progs[atype];
463
464 if (!hierarchy_allows_attach(cgrp, atype))
465 return -EPERM;
466
467 if (!list_empty(progs) && cgrp->bpf.flags[atype] != saved_flags)
468
469
470
471
472 return -EPERM;
473
474 if (prog_list_length(progs) >= BPF_CGROUP_MAX_PROGS)
475 return -E2BIG;
476
477 pl = find_attach_entry(progs, prog, link, replace_prog,
478 flags & BPF_F_ALLOW_MULTI);
479 if (IS_ERR(pl))
480 return PTR_ERR(pl);
481
482 if (bpf_cgroup_storages_alloc(storage, new_storage, type,
483 prog ? : link->link.prog, cgrp))
484 return -ENOMEM;
485
486 if (pl) {
487 old_prog = pl->prog;
488 } else {
489 pl = kmalloc(sizeof(*pl), GFP_KERNEL);
490 if (!pl) {
491 bpf_cgroup_storages_free(new_storage);
492 return -ENOMEM;
493 }
494 list_add_tail(&pl->node, progs);
495 }
496
497 pl->prog = prog;
498 pl->link = link;
499 bpf_cgroup_storages_assign(pl->storage, storage);
500 cgrp->bpf.flags[atype] = saved_flags;
501
502 err = update_effective_progs(cgrp, atype);
503 if (err)
504 goto cleanup;
505
506 if (old_prog)
507 bpf_prog_put(old_prog);
508 else
509 static_branch_inc(&cgroup_bpf_enabled_key[atype]);
510 bpf_cgroup_storages_link(new_storage, cgrp, type);
511 return 0;
512
513cleanup:
514 if (old_prog) {
515 pl->prog = old_prog;
516 pl->link = NULL;
517 }
518 bpf_cgroup_storages_free(new_storage);
519 if (!old_prog) {
520 list_del(&pl->node);
521 kfree(pl);
522 }
523 return err;
524}
525
526static int cgroup_bpf_attach(struct cgroup *cgrp,
527 struct bpf_prog *prog, struct bpf_prog *replace_prog,
528 struct bpf_cgroup_link *link,
529 enum bpf_attach_type type,
530 u32 flags)
531{
532 int ret;
533
534 mutex_lock(&cgroup_mutex);
535 ret = __cgroup_bpf_attach(cgrp, prog, replace_prog, link, type, flags);
536 mutex_unlock(&cgroup_mutex);
537 return ret;
538}
539
540
541
542
543static void replace_effective_prog(struct cgroup *cgrp,
544 enum cgroup_bpf_attach_type atype,
545 struct bpf_cgroup_link *link)
546{
547 struct bpf_prog_array_item *item;
548 struct cgroup_subsys_state *css;
549 struct bpf_prog_array *progs;
550 struct bpf_prog_list *pl;
551 struct list_head *head;
552 struct cgroup *cg;
553 int pos;
554
555 css_for_each_descendant_pre(css, &cgrp->self) {
556 struct cgroup *desc = container_of(css, struct cgroup, self);
557
558 if (percpu_ref_is_zero(&desc->bpf.refcnt))
559 continue;
560
561
562 for (pos = 0, cg = desc; cg; cg = cgroup_parent(cg)) {
563 if (pos && !(cg->bpf.flags[atype] & BPF_F_ALLOW_MULTI))
564 continue;
565
566 head = &cg->bpf.progs[atype];
567 list_for_each_entry(pl, head, node) {
568 if (!prog_list_prog(pl))
569 continue;
570 if (pl->link == link)
571 goto found;
572 pos++;
573 }
574 }
575found:
576 BUG_ON(!cg);
577 progs = rcu_dereference_protected(
578 desc->bpf.effective[atype],
579 lockdep_is_held(&cgroup_mutex));
580 item = &progs->items[pos];
581 WRITE_ONCE(item->prog, link->link.prog);
582 }
583}
584
585
586
587
588
589
590
591
592
593
594static int __cgroup_bpf_replace(struct cgroup *cgrp,
595 struct bpf_cgroup_link *link,
596 struct bpf_prog *new_prog)
597{
598 enum cgroup_bpf_attach_type atype;
599 struct bpf_prog *old_prog;
600 struct bpf_prog_list *pl;
601 struct list_head *progs;
602 bool found = false;
603
604 atype = to_cgroup_bpf_attach_type(link->type);
605 if (atype < 0)
606 return -EINVAL;
607
608 progs = &cgrp->bpf.progs[atype];
609
610 if (link->link.prog->type != new_prog->type)
611 return -EINVAL;
612
613 list_for_each_entry(pl, progs, node) {
614 if (pl->link == link) {
615 found = true;
616 break;
617 }
618 }
619 if (!found)
620 return -ENOENT;
621
622 old_prog = xchg(&link->link.prog, new_prog);
623 replace_effective_prog(cgrp, atype, link);
624 bpf_prog_put(old_prog);
625 return 0;
626}
627
628static int cgroup_bpf_replace(struct bpf_link *link, struct bpf_prog *new_prog,
629 struct bpf_prog *old_prog)
630{
631 struct bpf_cgroup_link *cg_link;
632 int ret;
633
634 cg_link = container_of(link, struct bpf_cgroup_link, link);
635
636 mutex_lock(&cgroup_mutex);
637
638 if (!cg_link->cgroup) {
639 ret = -ENOLINK;
640 goto out_unlock;
641 }
642 if (old_prog && link->prog != old_prog) {
643 ret = -EPERM;
644 goto out_unlock;
645 }
646 ret = __cgroup_bpf_replace(cg_link->cgroup, cg_link, new_prog);
647out_unlock:
648 mutex_unlock(&cgroup_mutex);
649 return ret;
650}
651
652static struct bpf_prog_list *find_detach_entry(struct list_head *progs,
653 struct bpf_prog *prog,
654 struct bpf_cgroup_link *link,
655 bool allow_multi)
656{
657 struct bpf_prog_list *pl;
658
659 if (!allow_multi) {
660 if (list_empty(progs))
661
662 return ERR_PTR(-ENOENT);
663
664
665
666
667 return list_first_entry(progs, typeof(*pl), node);
668 }
669
670 if (!prog && !link)
671
672
673
674 return ERR_PTR(-EINVAL);
675
676
677 list_for_each_entry(pl, progs, node) {
678 if (pl->prog == prog && pl->link == link)
679 return pl;
680 }
681 return ERR_PTR(-ENOENT);
682}
683
684
685
686
687
688
689
690
691
692
693
694
695static int __cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
696 struct bpf_cgroup_link *link, enum bpf_attach_type type)
697{
698 enum cgroup_bpf_attach_type atype;
699 struct bpf_prog *old_prog;
700 struct bpf_prog_list *pl;
701 struct list_head *progs;
702 u32 flags;
703 int err;
704
705 atype = to_cgroup_bpf_attach_type(type);
706 if (atype < 0)
707 return -EINVAL;
708
709 progs = &cgrp->bpf.progs[atype];
710 flags = cgrp->bpf.flags[atype];
711
712 if (prog && link)
713
714 return -EINVAL;
715
716 pl = find_detach_entry(progs, prog, link, flags & BPF_F_ALLOW_MULTI);
717 if (IS_ERR(pl))
718 return PTR_ERR(pl);
719
720
721 old_prog = pl->prog;
722 pl->prog = NULL;
723 pl->link = NULL;
724
725 err = update_effective_progs(cgrp, atype);
726 if (err)
727 goto cleanup;
728
729
730 list_del(&pl->node);
731 kfree(pl);
732 if (list_empty(progs))
733
734 cgrp->bpf.flags[atype] = 0;
735 if (old_prog)
736 bpf_prog_put(old_prog);
737 static_branch_dec(&cgroup_bpf_enabled_key[atype]);
738 return 0;
739
740cleanup:
741
742 pl->prog = old_prog;
743 pl->link = link;
744 return err;
745}
746
747static int cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
748 enum bpf_attach_type type)
749{
750 int ret;
751
752 mutex_lock(&cgroup_mutex);
753 ret = __cgroup_bpf_detach(cgrp, prog, NULL, type);
754 mutex_unlock(&cgroup_mutex);
755 return ret;
756}
757
758
759static int __cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
760 union bpf_attr __user *uattr)
761{
762 __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
763 enum bpf_attach_type type = attr->query.attach_type;
764 enum cgroup_bpf_attach_type atype;
765 struct bpf_prog_array *effective;
766 struct list_head *progs;
767 struct bpf_prog *prog;
768 int cnt, ret = 0, i;
769 u32 flags;
770
771 atype = to_cgroup_bpf_attach_type(type);
772 if (atype < 0)
773 return -EINVAL;
774
775 progs = &cgrp->bpf.progs[atype];
776 flags = cgrp->bpf.flags[atype];
777
778 effective = rcu_dereference_protected(cgrp->bpf.effective[atype],
779 lockdep_is_held(&cgroup_mutex));
780
781 if (attr->query.query_flags & BPF_F_QUERY_EFFECTIVE)
782 cnt = bpf_prog_array_length(effective);
783 else
784 cnt = prog_list_length(progs);
785
786 if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)))
787 return -EFAULT;
788 if (copy_to_user(&uattr->query.prog_cnt, &cnt, sizeof(cnt)))
789 return -EFAULT;
790 if (attr->query.prog_cnt == 0 || !prog_ids || !cnt)
791
792 return 0;
793 if (attr->query.prog_cnt < cnt) {
794 cnt = attr->query.prog_cnt;
795 ret = -ENOSPC;
796 }
797
798 if (attr->query.query_flags & BPF_F_QUERY_EFFECTIVE) {
799 return bpf_prog_array_copy_to_user(effective, prog_ids, cnt);
800 } else {
801 struct bpf_prog_list *pl;
802 u32 id;
803
804 i = 0;
805 list_for_each_entry(pl, progs, node) {
806 prog = prog_list_prog(pl);
807 id = prog->aux->id;
808 if (copy_to_user(prog_ids + i, &id, sizeof(id)))
809 return -EFAULT;
810 if (++i == cnt)
811 break;
812 }
813 }
814 return ret;
815}
816
817static int cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
818 union bpf_attr __user *uattr)
819{
820 int ret;
821
822 mutex_lock(&cgroup_mutex);
823 ret = __cgroup_bpf_query(cgrp, attr, uattr);
824 mutex_unlock(&cgroup_mutex);
825 return ret;
826}
827
828int cgroup_bpf_prog_attach(const union bpf_attr *attr,
829 enum bpf_prog_type ptype, struct bpf_prog *prog)
830{
831 struct bpf_prog *replace_prog = NULL;
832 struct cgroup *cgrp;
833 int ret;
834
835 cgrp = cgroup_get_from_fd(attr->target_fd);
836 if (IS_ERR(cgrp))
837 return PTR_ERR(cgrp);
838
839 if ((attr->attach_flags & BPF_F_ALLOW_MULTI) &&
840 (attr->attach_flags & BPF_F_REPLACE)) {
841 replace_prog = bpf_prog_get_type(attr->replace_bpf_fd, ptype);
842 if (IS_ERR(replace_prog)) {
843 cgroup_put(cgrp);
844 return PTR_ERR(replace_prog);
845 }
846 }
847
848 ret = cgroup_bpf_attach(cgrp, prog, replace_prog, NULL,
849 attr->attach_type, attr->attach_flags);
850
851 if (replace_prog)
852 bpf_prog_put(replace_prog);
853 cgroup_put(cgrp);
854 return ret;
855}
856
857int cgroup_bpf_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
858{
859 struct bpf_prog *prog;
860 struct cgroup *cgrp;
861 int ret;
862
863 cgrp = cgroup_get_from_fd(attr->target_fd);
864 if (IS_ERR(cgrp))
865 return PTR_ERR(cgrp);
866
867 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
868 if (IS_ERR(prog))
869 prog = NULL;
870
871 ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type);
872 if (prog)
873 bpf_prog_put(prog);
874
875 cgroup_put(cgrp);
876 return ret;
877}
878
879static void bpf_cgroup_link_release(struct bpf_link *link)
880{
881 struct bpf_cgroup_link *cg_link =
882 container_of(link, struct bpf_cgroup_link, link);
883 struct cgroup *cg;
884
885
886
887
888 if (!cg_link->cgroup)
889 return;
890
891 mutex_lock(&cgroup_mutex);
892
893
894 if (!cg_link->cgroup) {
895 mutex_unlock(&cgroup_mutex);
896 return;
897 }
898
899 WARN_ON(__cgroup_bpf_detach(cg_link->cgroup, NULL, cg_link,
900 cg_link->type));
901
902 cg = cg_link->cgroup;
903 cg_link->cgroup = NULL;
904
905 mutex_unlock(&cgroup_mutex);
906
907 cgroup_put(cg);
908}
909
910static void bpf_cgroup_link_dealloc(struct bpf_link *link)
911{
912 struct bpf_cgroup_link *cg_link =
913 container_of(link, struct bpf_cgroup_link, link);
914
915 kfree(cg_link);
916}
917
918static int bpf_cgroup_link_detach(struct bpf_link *link)
919{
920 bpf_cgroup_link_release(link);
921
922 return 0;
923}
924
925static void bpf_cgroup_link_show_fdinfo(const struct bpf_link *link,
926 struct seq_file *seq)
927{
928 struct bpf_cgroup_link *cg_link =
929 container_of(link, struct bpf_cgroup_link, link);
930 u64 cg_id = 0;
931
932 mutex_lock(&cgroup_mutex);
933 if (cg_link->cgroup)
934 cg_id = cgroup_id(cg_link->cgroup);
935 mutex_unlock(&cgroup_mutex);
936
937 seq_printf(seq,
938 "cgroup_id:\t%llu\n"
939 "attach_type:\t%d\n",
940 cg_id,
941 cg_link->type);
942}
943
944static int bpf_cgroup_link_fill_link_info(const struct bpf_link *link,
945 struct bpf_link_info *info)
946{
947 struct bpf_cgroup_link *cg_link =
948 container_of(link, struct bpf_cgroup_link, link);
949 u64 cg_id = 0;
950
951 mutex_lock(&cgroup_mutex);
952 if (cg_link->cgroup)
953 cg_id = cgroup_id(cg_link->cgroup);
954 mutex_unlock(&cgroup_mutex);
955
956 info->cgroup.cgroup_id = cg_id;
957 info->cgroup.attach_type = cg_link->type;
958 return 0;
959}
960
961static const struct bpf_link_ops bpf_cgroup_link_lops = {
962 .release = bpf_cgroup_link_release,
963 .dealloc = bpf_cgroup_link_dealloc,
964 .detach = bpf_cgroup_link_detach,
965 .update_prog = cgroup_bpf_replace,
966 .show_fdinfo = bpf_cgroup_link_show_fdinfo,
967 .fill_link_info = bpf_cgroup_link_fill_link_info,
968};
969
970int cgroup_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
971{
972 struct bpf_link_primer link_primer;
973 struct bpf_cgroup_link *link;
974 struct cgroup *cgrp;
975 int err;
976
977 if (attr->link_create.flags)
978 return -EINVAL;
979
980 cgrp = cgroup_get_from_fd(attr->link_create.target_fd);
981 if (IS_ERR(cgrp))
982 return PTR_ERR(cgrp);
983
984 link = kzalloc(sizeof(*link), GFP_USER);
985 if (!link) {
986 err = -ENOMEM;
987 goto out_put_cgroup;
988 }
989 bpf_link_init(&link->link, BPF_LINK_TYPE_CGROUP, &bpf_cgroup_link_lops,
990 prog);
991 link->cgroup = cgrp;
992 link->type = attr->link_create.attach_type;
993
994 err = bpf_link_prime(&link->link, &link_primer);
995 if (err) {
996 kfree(link);
997 goto out_put_cgroup;
998 }
999
1000 err = cgroup_bpf_attach(cgrp, NULL, NULL, link,
1001 link->type, BPF_F_ALLOW_MULTI);
1002 if (err) {
1003 bpf_link_cleanup(&link_primer);
1004 goto out_put_cgroup;
1005 }
1006
1007 return bpf_link_settle(&link_primer);
1008
1009out_put_cgroup:
1010 cgroup_put(cgrp);
1011 return err;
1012}
1013
1014int cgroup_bpf_prog_query(const union bpf_attr *attr,
1015 union bpf_attr __user *uattr)
1016{
1017 struct cgroup *cgrp;
1018 int ret;
1019
1020 cgrp = cgroup_get_from_fd(attr->query.target_fd);
1021 if (IS_ERR(cgrp))
1022 return PTR_ERR(cgrp);
1023
1024 ret = cgroup_bpf_query(cgrp, attr, uattr);
1025
1026 cgroup_put(cgrp);
1027 return ret;
1028}
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053int __cgroup_bpf_run_filter_skb(struct sock *sk,
1054 struct sk_buff *skb,
1055 enum cgroup_bpf_attach_type atype)
1056{
1057 unsigned int offset = skb->data - skb_network_header(skb);
1058 struct sock *save_sk;
1059 void *saved_data_end;
1060 struct cgroup *cgrp;
1061 int ret;
1062
1063 if (!sk || !sk_fullsock(sk))
1064 return 0;
1065
1066 if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)
1067 return 0;
1068
1069 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1070 save_sk = skb->sk;
1071 skb->sk = sk;
1072 __skb_push(skb, offset);
1073
1074
1075 bpf_compute_and_save_data_end(skb, &saved_data_end);
1076
1077 if (atype == CGROUP_INET_EGRESS) {
1078 ret = BPF_PROG_CGROUP_INET_EGRESS_RUN_ARRAY(
1079 cgrp->bpf.effective[atype], skb, __bpf_prog_run_save_cb);
1080 } else {
1081 ret = BPF_PROG_RUN_ARRAY_CG(cgrp->bpf.effective[atype], skb,
1082 __bpf_prog_run_save_cb);
1083 ret = (ret == 1 ? 0 : -EPERM);
1084 }
1085 bpf_restore_data_end(skb, saved_data_end);
1086 __skb_pull(skb, offset);
1087 skb->sk = save_sk;
1088
1089 return ret;
1090}
1091EXPORT_SYMBOL(__cgroup_bpf_run_filter_skb);
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106int __cgroup_bpf_run_filter_sk(struct sock *sk,
1107 enum cgroup_bpf_attach_type atype)
1108{
1109 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1110 int ret;
1111
1112 ret = BPF_PROG_RUN_ARRAY_CG(cgrp->bpf.effective[atype], sk, bpf_prog_run);
1113 return ret == 1 ? 0 : -EPERM;
1114}
1115EXPORT_SYMBOL(__cgroup_bpf_run_filter_sk);
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132int __cgroup_bpf_run_filter_sock_addr(struct sock *sk,
1133 struct sockaddr *uaddr,
1134 enum cgroup_bpf_attach_type atype,
1135 void *t_ctx,
1136 u32 *flags)
1137{
1138 struct bpf_sock_addr_kern ctx = {
1139 .sk = sk,
1140 .uaddr = uaddr,
1141 .t_ctx = t_ctx,
1142 };
1143 struct sockaddr_storage unspec;
1144 struct cgroup *cgrp;
1145 int ret;
1146
1147
1148
1149
1150 if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)
1151 return 0;
1152
1153 if (!ctx.uaddr) {
1154 memset(&unspec, 0, sizeof(unspec));
1155 ctx.uaddr = (struct sockaddr *)&unspec;
1156 }
1157
1158 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1159 ret = BPF_PROG_RUN_ARRAY_CG_FLAGS(cgrp->bpf.effective[atype], &ctx,
1160 bpf_prog_run, flags);
1161
1162 return ret == 1 ? 0 : -EPERM;
1163}
1164EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_addr);
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182int __cgroup_bpf_run_filter_sock_ops(struct sock *sk,
1183 struct bpf_sock_ops_kern *sock_ops,
1184 enum cgroup_bpf_attach_type atype)
1185{
1186 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1187 int ret;
1188
1189 ret = BPF_PROG_RUN_ARRAY_CG(cgrp->bpf.effective[atype], sock_ops,
1190 bpf_prog_run);
1191 return ret == 1 ? 0 : -EPERM;
1192}
1193EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_ops);
1194
1195int __cgroup_bpf_check_dev_permission(short dev_type, u32 major, u32 minor,
1196 short access, enum cgroup_bpf_attach_type atype)
1197{
1198 struct cgroup *cgrp;
1199 struct bpf_cgroup_dev_ctx ctx = {
1200 .access_type = (access << 16) | dev_type,
1201 .major = major,
1202 .minor = minor,
1203 };
1204 int allow;
1205
1206 rcu_read_lock();
1207 cgrp = task_dfl_cgroup(current);
1208 allow = BPF_PROG_RUN_ARRAY_CG(cgrp->bpf.effective[atype], &ctx,
1209 bpf_prog_run);
1210 rcu_read_unlock();
1211
1212 return !allow;
1213}
1214
1215static const struct bpf_func_proto *
1216cgroup_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1217{
1218 switch (func_id) {
1219 case BPF_FUNC_get_current_uid_gid:
1220 return &bpf_get_current_uid_gid_proto;
1221 case BPF_FUNC_get_local_storage:
1222 return &bpf_get_local_storage_proto;
1223 case BPF_FUNC_get_current_cgroup_id:
1224 return &bpf_get_current_cgroup_id_proto;
1225 case BPF_FUNC_perf_event_output:
1226 return &bpf_event_output_data_proto;
1227 default:
1228 return bpf_base_func_proto(func_id);
1229 }
1230}
1231
1232static const struct bpf_func_proto *
1233cgroup_dev_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1234{
1235 return cgroup_base_func_proto(func_id, prog);
1236}
1237
1238static bool cgroup_dev_is_valid_access(int off, int size,
1239 enum bpf_access_type type,
1240 const struct bpf_prog *prog,
1241 struct bpf_insn_access_aux *info)
1242{
1243 const int size_default = sizeof(__u32);
1244
1245 if (type == BPF_WRITE)
1246 return false;
1247
1248 if (off < 0 || off + size > sizeof(struct bpf_cgroup_dev_ctx))
1249 return false;
1250
1251 if (off % size != 0)
1252 return false;
1253
1254 switch (off) {
1255 case bpf_ctx_range(struct bpf_cgroup_dev_ctx, access_type):
1256 bpf_ctx_record_field_size(info, size_default);
1257 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
1258 return false;
1259 break;
1260 default:
1261 if (size != size_default)
1262 return false;
1263 }
1264
1265 return true;
1266}
1267
1268const struct bpf_prog_ops cg_dev_prog_ops = {
1269};
1270
1271const struct bpf_verifier_ops cg_dev_verifier_ops = {
1272 .get_func_proto = cgroup_dev_func_proto,
1273 .is_valid_access = cgroup_dev_is_valid_access,
1274};
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297int __cgroup_bpf_run_filter_sysctl(struct ctl_table_header *head,
1298 struct ctl_table *table, int write,
1299 char **buf, size_t *pcount, loff_t *ppos,
1300 enum cgroup_bpf_attach_type atype)
1301{
1302 struct bpf_sysctl_kern ctx = {
1303 .head = head,
1304 .table = table,
1305 .write = write,
1306 .ppos = ppos,
1307 .cur_val = NULL,
1308 .cur_len = PAGE_SIZE,
1309 .new_val = NULL,
1310 .new_len = 0,
1311 .new_updated = 0,
1312 };
1313 struct cgroup *cgrp;
1314 loff_t pos = 0;
1315 int ret;
1316
1317 ctx.cur_val = kmalloc_track_caller(ctx.cur_len, GFP_KERNEL);
1318 if (!ctx.cur_val ||
1319 table->proc_handler(table, 0, ctx.cur_val, &ctx.cur_len, &pos)) {
1320
1321 ctx.cur_len = 0;
1322 }
1323
1324 if (write && *buf && *pcount) {
1325
1326
1327
1328 ctx.new_val = kmalloc_track_caller(PAGE_SIZE, GFP_KERNEL);
1329 ctx.new_len = min_t(size_t, PAGE_SIZE, *pcount);
1330 if (ctx.new_val) {
1331 memcpy(ctx.new_val, *buf, ctx.new_len);
1332 } else {
1333
1334 ctx.new_len = 0;
1335 }
1336 }
1337
1338 rcu_read_lock();
1339 cgrp = task_dfl_cgroup(current);
1340 ret = BPF_PROG_RUN_ARRAY_CG(cgrp->bpf.effective[atype], &ctx, bpf_prog_run);
1341 rcu_read_unlock();
1342
1343 kfree(ctx.cur_val);
1344
1345 if (ret == 1 && ctx.new_updated) {
1346 kfree(*buf);
1347 *buf = ctx.new_val;
1348 *pcount = ctx.new_len;
1349 } else {
1350 kfree(ctx.new_val);
1351 }
1352
1353 return ret == 1 ? 0 : -EPERM;
1354}
1355
1356#ifdef CONFIG_NET
1357static bool __cgroup_bpf_prog_array_is_empty(struct cgroup *cgrp,
1358 enum cgroup_bpf_attach_type attach_type)
1359{
1360 struct bpf_prog_array *prog_array;
1361 bool empty;
1362
1363 rcu_read_lock();
1364 prog_array = rcu_dereference(cgrp->bpf.effective[attach_type]);
1365 empty = bpf_prog_array_is_empty(prog_array);
1366 rcu_read_unlock();
1367
1368 return empty;
1369}
1370
1371static int sockopt_alloc_buf(struct bpf_sockopt_kern *ctx, int max_optlen,
1372 struct bpf_sockopt_buf *buf)
1373{
1374 if (unlikely(max_optlen < 0))
1375 return -EINVAL;
1376
1377 if (unlikely(max_optlen > PAGE_SIZE)) {
1378
1379
1380
1381 max_optlen = PAGE_SIZE;
1382 }
1383
1384 if (max_optlen <= sizeof(buf->data)) {
1385
1386
1387
1388 ctx->optval = buf->data;
1389 ctx->optval_end = ctx->optval + max_optlen;
1390 return max_optlen;
1391 }
1392
1393 ctx->optval = kzalloc(max_optlen, GFP_USER);
1394 if (!ctx->optval)
1395 return -ENOMEM;
1396
1397 ctx->optval_end = ctx->optval + max_optlen;
1398
1399 return max_optlen;
1400}
1401
1402static void sockopt_free_buf(struct bpf_sockopt_kern *ctx,
1403 struct bpf_sockopt_buf *buf)
1404{
1405 if (ctx->optval == buf->data)
1406 return;
1407 kfree(ctx->optval);
1408}
1409
1410static bool sockopt_buf_allocated(struct bpf_sockopt_kern *ctx,
1411 struct bpf_sockopt_buf *buf)
1412{
1413 return ctx->optval != buf->data;
1414}
1415
1416int __cgroup_bpf_run_filter_setsockopt(struct sock *sk, int *level,
1417 int *optname, char __user *optval,
1418 int *optlen, char **kernel_optval)
1419{
1420 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1421 struct bpf_sockopt_buf buf = {};
1422 struct bpf_sockopt_kern ctx = {
1423 .sk = sk,
1424 .level = *level,
1425 .optname = *optname,
1426 };
1427 int ret, max_optlen;
1428
1429
1430
1431
1432
1433 if (__cgroup_bpf_prog_array_is_empty(cgrp, CGROUP_SETSOCKOPT))
1434 return 0;
1435
1436
1437
1438
1439
1440 max_optlen = max_t(int, 16, *optlen);
1441
1442 max_optlen = sockopt_alloc_buf(&ctx, max_optlen, &buf);
1443 if (max_optlen < 0)
1444 return max_optlen;
1445
1446 ctx.optlen = *optlen;
1447
1448 if (copy_from_user(ctx.optval, optval, min(*optlen, max_optlen)) != 0) {
1449 ret = -EFAULT;
1450 goto out;
1451 }
1452
1453 lock_sock(sk);
1454 ret = BPF_PROG_RUN_ARRAY_CG(cgrp->bpf.effective[CGROUP_SETSOCKOPT],
1455 &ctx, bpf_prog_run);
1456 release_sock(sk);
1457
1458 if (!ret) {
1459 ret = -EPERM;
1460 goto out;
1461 }
1462
1463 if (ctx.optlen == -1) {
1464
1465 ret = 1;
1466 } else if (ctx.optlen > max_optlen || ctx.optlen < -1) {
1467
1468 ret = -EFAULT;
1469 } else {
1470
1471 ret = 0;
1472
1473
1474 *level = ctx.level;
1475 *optname = ctx.optname;
1476
1477
1478
1479
1480 if (ctx.optlen != 0) {
1481 *optlen = ctx.optlen;
1482
1483
1484
1485
1486
1487
1488 if (!sockopt_buf_allocated(&ctx, &buf)) {
1489 void *p = kmalloc(ctx.optlen, GFP_USER);
1490
1491 if (!p) {
1492 ret = -ENOMEM;
1493 goto out;
1494 }
1495 memcpy(p, ctx.optval, ctx.optlen);
1496 *kernel_optval = p;
1497 } else {
1498 *kernel_optval = ctx.optval;
1499 }
1500
1501 return 0;
1502 }
1503 }
1504
1505out:
1506 sockopt_free_buf(&ctx, &buf);
1507 return ret;
1508}
1509
1510int __cgroup_bpf_run_filter_getsockopt(struct sock *sk, int level,
1511 int optname, char __user *optval,
1512 int __user *optlen, int max_optlen,
1513 int retval)
1514{
1515 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1516 struct bpf_sockopt_buf buf = {};
1517 struct bpf_sockopt_kern ctx = {
1518 .sk = sk,
1519 .level = level,
1520 .optname = optname,
1521 .retval = retval,
1522 };
1523 int ret;
1524
1525
1526
1527
1528
1529 if (__cgroup_bpf_prog_array_is_empty(cgrp, CGROUP_GETSOCKOPT))
1530 return retval;
1531
1532 ctx.optlen = max_optlen;
1533
1534 max_optlen = sockopt_alloc_buf(&ctx, max_optlen, &buf);
1535 if (max_optlen < 0)
1536 return max_optlen;
1537
1538 if (!retval) {
1539
1540
1541
1542
1543
1544
1545
1546 if (get_user(ctx.optlen, optlen)) {
1547 ret = -EFAULT;
1548 goto out;
1549 }
1550
1551 if (ctx.optlen < 0) {
1552 ret = -EFAULT;
1553 goto out;
1554 }
1555
1556 if (copy_from_user(ctx.optval, optval,
1557 min(ctx.optlen, max_optlen)) != 0) {
1558 ret = -EFAULT;
1559 goto out;
1560 }
1561 }
1562
1563 lock_sock(sk);
1564 ret = BPF_PROG_RUN_ARRAY_CG(cgrp->bpf.effective[CGROUP_GETSOCKOPT],
1565 &ctx, bpf_prog_run);
1566 release_sock(sk);
1567
1568 if (!ret) {
1569 ret = -EPERM;
1570 goto out;
1571 }
1572
1573 if (ctx.optlen > max_optlen || ctx.optlen < 0) {
1574 ret = -EFAULT;
1575 goto out;
1576 }
1577
1578
1579
1580
1581 if (ctx.retval != 0 && ctx.retval != retval) {
1582 ret = -EFAULT;
1583 goto out;
1584 }
1585
1586 if (ctx.optlen != 0) {
1587 if (copy_to_user(optval, ctx.optval, ctx.optlen) ||
1588 put_user(ctx.optlen, optlen)) {
1589 ret = -EFAULT;
1590 goto out;
1591 }
1592 }
1593
1594 ret = ctx.retval;
1595
1596out:
1597 sockopt_free_buf(&ctx, &buf);
1598 return ret;
1599}
1600
1601int __cgroup_bpf_run_filter_getsockopt_kern(struct sock *sk, int level,
1602 int optname, void *optval,
1603 int *optlen, int retval)
1604{
1605 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1606 struct bpf_sockopt_kern ctx = {
1607 .sk = sk,
1608 .level = level,
1609 .optname = optname,
1610 .retval = retval,
1611 .optlen = *optlen,
1612 .optval = optval,
1613 .optval_end = optval + *optlen,
1614 };
1615 int ret;
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625 ret = BPF_PROG_RUN_ARRAY_CG(cgrp->bpf.effective[CGROUP_GETSOCKOPT],
1626 &ctx, bpf_prog_run);
1627 if (!ret)
1628 return -EPERM;
1629
1630 if (ctx.optlen > *optlen)
1631 return -EFAULT;
1632
1633
1634
1635
1636 if (ctx.retval != 0 && ctx.retval != retval)
1637 return -EFAULT;
1638
1639
1640
1641 if (ctx.optlen != 0)
1642 *optlen = ctx.optlen;
1643
1644 return ctx.retval;
1645}
1646#endif
1647
1648static ssize_t sysctl_cpy_dir(const struct ctl_dir *dir, char **bufp,
1649 size_t *lenp)
1650{
1651 ssize_t tmp_ret = 0, ret;
1652
1653 if (dir->header.parent) {
1654 tmp_ret = sysctl_cpy_dir(dir->header.parent, bufp, lenp);
1655 if (tmp_ret < 0)
1656 return tmp_ret;
1657 }
1658
1659 ret = strscpy(*bufp, dir->header.ctl_table[0].procname, *lenp);
1660 if (ret < 0)
1661 return ret;
1662 *bufp += ret;
1663 *lenp -= ret;
1664 ret += tmp_ret;
1665
1666
1667 if (!ret)
1668 return ret;
1669
1670 tmp_ret = strscpy(*bufp, "/", *lenp);
1671 if (tmp_ret < 0)
1672 return tmp_ret;
1673 *bufp += tmp_ret;
1674 *lenp -= tmp_ret;
1675
1676 return ret + tmp_ret;
1677}
1678
1679BPF_CALL_4(bpf_sysctl_get_name, struct bpf_sysctl_kern *, ctx, char *, buf,
1680 size_t, buf_len, u64, flags)
1681{
1682 ssize_t tmp_ret = 0, ret;
1683
1684 if (!buf)
1685 return -EINVAL;
1686
1687 if (!(flags & BPF_F_SYSCTL_BASE_NAME)) {
1688 if (!ctx->head)
1689 return -EINVAL;
1690 tmp_ret = sysctl_cpy_dir(ctx->head->parent, &buf, &buf_len);
1691 if (tmp_ret < 0)
1692 return tmp_ret;
1693 }
1694
1695 ret = strscpy(buf, ctx->table->procname, buf_len);
1696
1697 return ret < 0 ? ret : tmp_ret + ret;
1698}
1699
1700static const struct bpf_func_proto bpf_sysctl_get_name_proto = {
1701 .func = bpf_sysctl_get_name,
1702 .gpl_only = false,
1703 .ret_type = RET_INTEGER,
1704 .arg1_type = ARG_PTR_TO_CTX,
1705 .arg2_type = ARG_PTR_TO_MEM,
1706 .arg3_type = ARG_CONST_SIZE,
1707 .arg4_type = ARG_ANYTHING,
1708};
1709
1710static int copy_sysctl_value(char *dst, size_t dst_len, char *src,
1711 size_t src_len)
1712{
1713 if (!dst)
1714 return -EINVAL;
1715
1716 if (!dst_len)
1717 return -E2BIG;
1718
1719 if (!src || !src_len) {
1720 memset(dst, 0, dst_len);
1721 return -EINVAL;
1722 }
1723
1724 memcpy(dst, src, min(dst_len, src_len));
1725
1726 if (dst_len > src_len) {
1727 memset(dst + src_len, '\0', dst_len - src_len);
1728 return src_len;
1729 }
1730
1731 dst[dst_len - 1] = '\0';
1732
1733 return -E2BIG;
1734}
1735
1736BPF_CALL_3(bpf_sysctl_get_current_value, struct bpf_sysctl_kern *, ctx,
1737 char *, buf, size_t, buf_len)
1738{
1739 return copy_sysctl_value(buf, buf_len, ctx->cur_val, ctx->cur_len);
1740}
1741
1742static const struct bpf_func_proto bpf_sysctl_get_current_value_proto = {
1743 .func = bpf_sysctl_get_current_value,
1744 .gpl_only = false,
1745 .ret_type = RET_INTEGER,
1746 .arg1_type = ARG_PTR_TO_CTX,
1747 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1748 .arg3_type = ARG_CONST_SIZE,
1749};
1750
1751BPF_CALL_3(bpf_sysctl_get_new_value, struct bpf_sysctl_kern *, ctx, char *, buf,
1752 size_t, buf_len)
1753{
1754 if (!ctx->write) {
1755 if (buf && buf_len)
1756 memset(buf, '\0', buf_len);
1757 return -EINVAL;
1758 }
1759 return copy_sysctl_value(buf, buf_len, ctx->new_val, ctx->new_len);
1760}
1761
1762static const struct bpf_func_proto bpf_sysctl_get_new_value_proto = {
1763 .func = bpf_sysctl_get_new_value,
1764 .gpl_only = false,
1765 .ret_type = RET_INTEGER,
1766 .arg1_type = ARG_PTR_TO_CTX,
1767 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1768 .arg3_type = ARG_CONST_SIZE,
1769};
1770
1771BPF_CALL_3(bpf_sysctl_set_new_value, struct bpf_sysctl_kern *, ctx,
1772 const char *, buf, size_t, buf_len)
1773{
1774 if (!ctx->write || !ctx->new_val || !ctx->new_len || !buf || !buf_len)
1775 return -EINVAL;
1776
1777 if (buf_len > PAGE_SIZE - 1)
1778 return -E2BIG;
1779
1780 memcpy(ctx->new_val, buf, buf_len);
1781 ctx->new_len = buf_len;
1782 ctx->new_updated = 1;
1783
1784 return 0;
1785}
1786
1787static const struct bpf_func_proto bpf_sysctl_set_new_value_proto = {
1788 .func = bpf_sysctl_set_new_value,
1789 .gpl_only = false,
1790 .ret_type = RET_INTEGER,
1791 .arg1_type = ARG_PTR_TO_CTX,
1792 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1793 .arg3_type = ARG_CONST_SIZE,
1794};
1795
1796static const struct bpf_func_proto *
1797sysctl_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1798{
1799 switch (func_id) {
1800 case BPF_FUNC_strtol:
1801 return &bpf_strtol_proto;
1802 case BPF_FUNC_strtoul:
1803 return &bpf_strtoul_proto;
1804 case BPF_FUNC_sysctl_get_name:
1805 return &bpf_sysctl_get_name_proto;
1806 case BPF_FUNC_sysctl_get_current_value:
1807 return &bpf_sysctl_get_current_value_proto;
1808 case BPF_FUNC_sysctl_get_new_value:
1809 return &bpf_sysctl_get_new_value_proto;
1810 case BPF_FUNC_sysctl_set_new_value:
1811 return &bpf_sysctl_set_new_value_proto;
1812 case BPF_FUNC_ktime_get_coarse_ns:
1813 return &bpf_ktime_get_coarse_ns_proto;
1814 default:
1815 return cgroup_base_func_proto(func_id, prog);
1816 }
1817}
1818
1819static bool sysctl_is_valid_access(int off, int size, enum bpf_access_type type,
1820 const struct bpf_prog *prog,
1821 struct bpf_insn_access_aux *info)
1822{
1823 const int size_default = sizeof(__u32);
1824
1825 if (off < 0 || off + size > sizeof(struct bpf_sysctl) || off % size)
1826 return false;
1827
1828 switch (off) {
1829 case bpf_ctx_range(struct bpf_sysctl, write):
1830 if (type != BPF_READ)
1831 return false;
1832 bpf_ctx_record_field_size(info, size_default);
1833 return bpf_ctx_narrow_access_ok(off, size, size_default);
1834 case bpf_ctx_range(struct bpf_sysctl, file_pos):
1835 if (type == BPF_READ) {
1836 bpf_ctx_record_field_size(info, size_default);
1837 return bpf_ctx_narrow_access_ok(off, size, size_default);
1838 } else {
1839 return size == size_default;
1840 }
1841 default:
1842 return false;
1843 }
1844}
1845
1846static u32 sysctl_convert_ctx_access(enum bpf_access_type type,
1847 const struct bpf_insn *si,
1848 struct bpf_insn *insn_buf,
1849 struct bpf_prog *prog, u32 *target_size)
1850{
1851 struct bpf_insn *insn = insn_buf;
1852 u32 read_size;
1853
1854 switch (si->off) {
1855 case offsetof(struct bpf_sysctl, write):
1856 *insn++ = BPF_LDX_MEM(
1857 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
1858 bpf_target_off(struct bpf_sysctl_kern, write,
1859 sizeof_field(struct bpf_sysctl_kern,
1860 write),
1861 target_size));
1862 break;
1863 case offsetof(struct bpf_sysctl, file_pos):
1864
1865
1866
1867
1868
1869 if (type == BPF_WRITE) {
1870 int treg = BPF_REG_9;
1871
1872 if (si->src_reg == treg || si->dst_reg == treg)
1873 --treg;
1874 if (si->src_reg == treg || si->dst_reg == treg)
1875 --treg;
1876 *insn++ = BPF_STX_MEM(
1877 BPF_DW, si->dst_reg, treg,
1878 offsetof(struct bpf_sysctl_kern, tmp_reg));
1879 *insn++ = BPF_LDX_MEM(
1880 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos),
1881 treg, si->dst_reg,
1882 offsetof(struct bpf_sysctl_kern, ppos));
1883 *insn++ = BPF_STX_MEM(
1884 BPF_SIZEOF(u32), treg, si->src_reg,
1885 bpf_ctx_narrow_access_offset(
1886 0, sizeof(u32), sizeof(loff_t)));
1887 *insn++ = BPF_LDX_MEM(
1888 BPF_DW, treg, si->dst_reg,
1889 offsetof(struct bpf_sysctl_kern, tmp_reg));
1890 } else {
1891 *insn++ = BPF_LDX_MEM(
1892 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos),
1893 si->dst_reg, si->src_reg,
1894 offsetof(struct bpf_sysctl_kern, ppos));
1895 read_size = bpf_size_to_bytes(BPF_SIZE(si->code));
1896 *insn++ = BPF_LDX_MEM(
1897 BPF_SIZE(si->code), si->dst_reg, si->dst_reg,
1898 bpf_ctx_narrow_access_offset(
1899 0, read_size, sizeof(loff_t)));
1900 }
1901 *target_size = sizeof(u32);
1902 break;
1903 }
1904
1905 return insn - insn_buf;
1906}
1907
1908const struct bpf_verifier_ops cg_sysctl_verifier_ops = {
1909 .get_func_proto = sysctl_func_proto,
1910 .is_valid_access = sysctl_is_valid_access,
1911 .convert_ctx_access = sysctl_convert_ctx_access,
1912};
1913
1914const struct bpf_prog_ops cg_sysctl_prog_ops = {
1915};
1916
1917#ifdef CONFIG_NET
1918BPF_CALL_1(bpf_get_netns_cookie_sockopt, struct bpf_sockopt_kern *, ctx)
1919{
1920 const struct net *net = ctx ? sock_net(ctx->sk) : &init_net;
1921
1922 return net->net_cookie;
1923}
1924
1925static const struct bpf_func_proto bpf_get_netns_cookie_sockopt_proto = {
1926 .func = bpf_get_netns_cookie_sockopt,
1927 .gpl_only = false,
1928 .ret_type = RET_INTEGER,
1929 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
1930};
1931#endif
1932
1933static const struct bpf_func_proto *
1934cg_sockopt_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1935{
1936 switch (func_id) {
1937#ifdef CONFIG_NET
1938 case BPF_FUNC_get_netns_cookie:
1939 return &bpf_get_netns_cookie_sockopt_proto;
1940 case BPF_FUNC_sk_storage_get:
1941 return &bpf_sk_storage_get_proto;
1942 case BPF_FUNC_sk_storage_delete:
1943 return &bpf_sk_storage_delete_proto;
1944 case BPF_FUNC_setsockopt:
1945 if (prog->expected_attach_type == BPF_CGROUP_SETSOCKOPT)
1946 return &bpf_sk_setsockopt_proto;
1947 return NULL;
1948 case BPF_FUNC_getsockopt:
1949 if (prog->expected_attach_type == BPF_CGROUP_SETSOCKOPT)
1950 return &bpf_sk_getsockopt_proto;
1951 return NULL;
1952#endif
1953#ifdef CONFIG_INET
1954 case BPF_FUNC_tcp_sock:
1955 return &bpf_tcp_sock_proto;
1956#endif
1957 default:
1958 return cgroup_base_func_proto(func_id, prog);
1959 }
1960}
1961
1962static bool cg_sockopt_is_valid_access(int off, int size,
1963 enum bpf_access_type type,
1964 const struct bpf_prog *prog,
1965 struct bpf_insn_access_aux *info)
1966{
1967 const int size_default = sizeof(__u32);
1968
1969 if (off < 0 || off >= sizeof(struct bpf_sockopt))
1970 return false;
1971
1972 if (off % size != 0)
1973 return false;
1974
1975 if (type == BPF_WRITE) {
1976 switch (off) {
1977 case offsetof(struct bpf_sockopt, retval):
1978 if (size != size_default)
1979 return false;
1980 return prog->expected_attach_type ==
1981 BPF_CGROUP_GETSOCKOPT;
1982 case offsetof(struct bpf_sockopt, optname):
1983 fallthrough;
1984 case offsetof(struct bpf_sockopt, level):
1985 if (size != size_default)
1986 return false;
1987 return prog->expected_attach_type ==
1988 BPF_CGROUP_SETSOCKOPT;
1989 case offsetof(struct bpf_sockopt, optlen):
1990 return size == size_default;
1991 default:
1992 return false;
1993 }
1994 }
1995
1996 switch (off) {
1997 case offsetof(struct bpf_sockopt, sk):
1998 if (size != sizeof(__u64))
1999 return false;
2000 info->reg_type = PTR_TO_SOCKET;
2001 break;
2002 case offsetof(struct bpf_sockopt, optval):
2003 if (size != sizeof(__u64))
2004 return false;
2005 info->reg_type = PTR_TO_PACKET;
2006 break;
2007 case offsetof(struct bpf_sockopt, optval_end):
2008 if (size != sizeof(__u64))
2009 return false;
2010 info->reg_type = PTR_TO_PACKET_END;
2011 break;
2012 case offsetof(struct bpf_sockopt, retval):
2013 if (size != size_default)
2014 return false;
2015 return prog->expected_attach_type == BPF_CGROUP_GETSOCKOPT;
2016 default:
2017 if (size != size_default)
2018 return false;
2019 break;
2020 }
2021 return true;
2022}
2023
2024#define CG_SOCKOPT_ACCESS_FIELD(T, F) \
2025 T(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, F), \
2026 si->dst_reg, si->src_reg, \
2027 offsetof(struct bpf_sockopt_kern, F))
2028
2029static u32 cg_sockopt_convert_ctx_access(enum bpf_access_type type,
2030 const struct bpf_insn *si,
2031 struct bpf_insn *insn_buf,
2032 struct bpf_prog *prog,
2033 u32 *target_size)
2034{
2035 struct bpf_insn *insn = insn_buf;
2036
2037 switch (si->off) {
2038 case offsetof(struct bpf_sockopt, sk):
2039 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, sk);
2040 break;
2041 case offsetof(struct bpf_sockopt, level):
2042 if (type == BPF_WRITE)
2043 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, level);
2044 else
2045 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, level);
2046 break;
2047 case offsetof(struct bpf_sockopt, optname):
2048 if (type == BPF_WRITE)
2049 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, optname);
2050 else
2051 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optname);
2052 break;
2053 case offsetof(struct bpf_sockopt, optlen):
2054 if (type == BPF_WRITE)
2055 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, optlen);
2056 else
2057 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optlen);
2058 break;
2059 case offsetof(struct bpf_sockopt, retval):
2060 if (type == BPF_WRITE)
2061 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, retval);
2062 else
2063 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, retval);
2064 break;
2065 case offsetof(struct bpf_sockopt, optval):
2066 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optval);
2067 break;
2068 case offsetof(struct bpf_sockopt, optval_end):
2069 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optval_end);
2070 break;
2071 }
2072
2073 return insn - insn_buf;
2074}
2075
2076static int cg_sockopt_get_prologue(struct bpf_insn *insn_buf,
2077 bool direct_write,
2078 const struct bpf_prog *prog)
2079{
2080
2081
2082 return 0;
2083}
2084
2085const struct bpf_verifier_ops cg_sockopt_verifier_ops = {
2086 .get_func_proto = cg_sockopt_func_proto,
2087 .is_valid_access = cg_sockopt_is_valid_access,
2088 .convert_ctx_access = cg_sockopt_convert_ctx_access,
2089 .gen_prologue = cg_sockopt_get_prologue,
2090};
2091
2092const struct bpf_prog_ops cg_sockopt_prog_ops = {
2093};
2094