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 <linux/bpf_lsm.h>
18 #include <linux/bpf_verifier.h>
19 #include <net/sock.h>
20 #include <net/bpf_sk_storage.h>
21
22 #include "../cgroup/cgroup-internal.h"
23
24 DEFINE_STATIC_KEY_ARRAY_FALSE(cgroup_bpf_enabled_key, MAX_CGROUP_BPF_ATTACH_TYPE);
25 EXPORT_SYMBOL(cgroup_bpf_enabled_key);
26
27 /* __always_inline is necessary to prevent indirect call through run_prog
28 * function pointer.
29 */
30 static __always_inline int
31 bpf_prog_run_array_cg(const struct cgroup_bpf *cgrp,
32 enum cgroup_bpf_attach_type atype,
33 const void *ctx, bpf_prog_run_fn run_prog,
34 int retval, u32 *ret_flags)
35 {
36 const struct bpf_prog_array_item *item;
37 const struct bpf_prog *prog;
38 const struct bpf_prog_array *array;
39 struct bpf_run_ctx *old_run_ctx;
40 struct bpf_cg_run_ctx run_ctx;
41 u32 func_ret;
42
43 run_ctx.retval = retval;
44 migrate_disable();
45 rcu_read_lock();
46 array = rcu_dereference(cgrp->effective[atype]);
47 item = &array->items[0];
48 old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
49 while ((prog = READ_ONCE(item->prog))) {
50 run_ctx.prog_item = item;
51 func_ret = run_prog(prog, ctx);
52 if (ret_flags) {
53 *(ret_flags) |= (func_ret >> 1);
54 func_ret &= 1;
55 }
56 if (!func_ret && !IS_ERR_VALUE((long)run_ctx.retval))
57 run_ctx.retval = -EPERM;
58 item++;
59 }
60 bpf_reset_run_ctx(old_run_ctx);
61 rcu_read_unlock();
62 migrate_enable();
63 return run_ctx.retval;
64 }
65
66 unsigned int __cgroup_bpf_run_lsm_sock(const void *ctx,
67 const struct bpf_insn *insn)
68 {
69 const struct bpf_prog *shim_prog;
70 struct sock *sk;
71 struct cgroup *cgrp;
72 int ret = 0;
73 u64 *args;
74
75 args = (u64 *)ctx;
76 sk = (void *)(unsigned long)args[0];
77 /*shim_prog = container_of(insn, struct bpf_prog, insnsi);*/
78 shim_prog = (const struct bpf_prog *)((void *)insn - offsetof(struct bpf_prog, insnsi));
79
80 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
81 if (likely(cgrp))
82 ret = bpf_prog_run_array_cg(&cgrp->bpf,
83 shim_prog->aux->cgroup_atype,
84 ctx, bpf_prog_run, 0, NULL);
85 return ret;
86 }
87
88 unsigned int __cgroup_bpf_run_lsm_socket(const void *ctx,
89 const struct bpf_insn *insn)
90 {
91 const struct bpf_prog *shim_prog;
92 struct socket *sock;
93 struct cgroup *cgrp;
94 int ret = 0;
95 u64 *args;
96
97 args = (u64 *)ctx;
98 sock = (void *)(unsigned long)args[0];
99 /*shim_prog = container_of(insn, struct bpf_prog, insnsi);*/
100 shim_prog = (const struct bpf_prog *)((void *)insn - offsetof(struct bpf_prog, insnsi));
101
102 cgrp = sock_cgroup_ptr(&sock->sk->sk_cgrp_data);
103 if (likely(cgrp))
104 ret = bpf_prog_run_array_cg(&cgrp->bpf,
105 shim_prog->aux->cgroup_atype,
106 ctx, bpf_prog_run, 0, NULL);
107 return ret;
108 }
109
110 unsigned int __cgroup_bpf_run_lsm_current(const void *ctx,
111 const struct bpf_insn *insn)
112 {
113 const struct bpf_prog *shim_prog;
114 struct cgroup *cgrp;
115 int ret = 0;
116
117 /*shim_prog = container_of(insn, struct bpf_prog, insnsi);*/
118 shim_prog = (const struct bpf_prog *)((void *)insn - offsetof(struct bpf_prog, insnsi));
119
120 /* We rely on trampoline's __bpf_prog_enter_lsm_cgroup to grab RCU read lock. */
121 cgrp = task_dfl_cgroup(current);
122 if (likely(cgrp))
123 ret = bpf_prog_run_array_cg(&cgrp->bpf,
124 shim_prog->aux->cgroup_atype,
125 ctx, bpf_prog_run, 0, NULL);
126 return ret;
127 }
128
129 #ifdef CONFIG_BPF_LSM
130 struct cgroup_lsm_atype {
131 u32 attach_btf_id;
132 int refcnt;
133 };
134
135 static struct cgroup_lsm_atype cgroup_lsm_atype[CGROUP_LSM_NUM];
136
137 static enum cgroup_bpf_attach_type
138 bpf_cgroup_atype_find(enum bpf_attach_type attach_type, u32 attach_btf_id)
139 {
140 int i;
141
142 lockdep_assert_held(&cgroup_mutex);
143
144 if (attach_type != BPF_LSM_CGROUP)
145 return to_cgroup_bpf_attach_type(attach_type);
146
147 for (i = 0; i < ARRAY_SIZE(cgroup_lsm_atype); i++)
148 if (cgroup_lsm_atype[i].attach_btf_id == attach_btf_id)
149 return CGROUP_LSM_START + i;
150
151 for (i = 0; i < ARRAY_SIZE(cgroup_lsm_atype); i++)
152 if (cgroup_lsm_atype[i].attach_btf_id == 0)
153 return CGROUP_LSM_START + i;
154
155 return -E2BIG;
156
157 }
158
159 void bpf_cgroup_atype_get(u32 attach_btf_id, int cgroup_atype)
160 {
161 int i = cgroup_atype - CGROUP_LSM_START;
162
163 lockdep_assert_held(&cgroup_mutex);
164
165 WARN_ON_ONCE(cgroup_lsm_atype[i].attach_btf_id &&
166 cgroup_lsm_atype[i].attach_btf_id != attach_btf_id);
167
168 cgroup_lsm_atype[i].attach_btf_id = attach_btf_id;
169 cgroup_lsm_atype[i].refcnt++;
170 }
171
172 void bpf_cgroup_atype_put(int cgroup_atype)
173 {
174 int i = cgroup_atype - CGROUP_LSM_START;
175
176 cgroup_lock();
177 if (--cgroup_lsm_atype[i].refcnt <= 0)
178 cgroup_lsm_atype[i].attach_btf_id = 0;
179 WARN_ON_ONCE(cgroup_lsm_atype[i].refcnt < 0);
180 cgroup_unlock();
181 }
182 #else
183 static enum cgroup_bpf_attach_type
184 bpf_cgroup_atype_find(enum bpf_attach_type attach_type, u32 attach_btf_id)
185 {
186 if (attach_type != BPF_LSM_CGROUP)
187 return to_cgroup_bpf_attach_type(attach_type);
188 return -EOPNOTSUPP;
189 }
190 #endif /* CONFIG_BPF_LSM */
191
192 void cgroup_bpf_offline(struct cgroup *cgrp)
193 {
194 cgroup_get(cgrp);
195 percpu_ref_kill(&cgrp->bpf.refcnt);
196 }
197
198 static void bpf_cgroup_storages_free(struct bpf_cgroup_storage *storages[])
199 {
200 enum bpf_cgroup_storage_type stype;
201
202 for_each_cgroup_storage_type(stype)
203 bpf_cgroup_storage_free(storages[stype]);
204 }
205
206 static int bpf_cgroup_storages_alloc(struct bpf_cgroup_storage *storages[],
207 struct bpf_cgroup_storage *new_storages[],
208 enum bpf_attach_type type,
209 struct bpf_prog *prog,
210 struct cgroup *cgrp)
211 {
212 enum bpf_cgroup_storage_type stype;
213 struct bpf_cgroup_storage_key key;
214 struct bpf_map *map;
215
216 key.cgroup_inode_id = cgroup_id(cgrp);
217 key.attach_type = type;
218
219 for_each_cgroup_storage_type(stype) {
220 map = prog->aux->cgroup_storage[stype];
221 if (!map)
222 continue;
223
224 storages[stype] = cgroup_storage_lookup((void *)map, &key, false);
225 if (storages[stype])
226 continue;
227
228 storages[stype] = bpf_cgroup_storage_alloc(prog, stype);
229 if (IS_ERR(storages[stype])) {
230 bpf_cgroup_storages_free(new_storages);
231 return -ENOMEM;
232 }
233
234 new_storages[stype] = storages[stype];
235 }
236
237 return 0;
238 }
239
240 static void bpf_cgroup_storages_assign(struct bpf_cgroup_storage *dst[],
241 struct bpf_cgroup_storage *src[])
242 {
243 enum bpf_cgroup_storage_type stype;
244
245 for_each_cgroup_storage_type(stype)
246 dst[stype] = src[stype];
247 }
248
249 static void bpf_cgroup_storages_link(struct bpf_cgroup_storage *storages[],
250 struct cgroup *cgrp,
251 enum bpf_attach_type attach_type)
252 {
253 enum bpf_cgroup_storage_type stype;
254
255 for_each_cgroup_storage_type(stype)
256 bpf_cgroup_storage_link(storages[stype], cgrp, attach_type);
257 }
258
259 /* Called when bpf_cgroup_link is auto-detached from dying cgroup.
260 * It drops cgroup and bpf_prog refcounts, and marks bpf_link as defunct. It
261 * doesn't free link memory, which will eventually be done by bpf_link's
262 * release() callback, when its last FD is closed.
263 */
264 static void bpf_cgroup_link_auto_detach(struct bpf_cgroup_link *link)
265 {
266 cgroup_put(link->cgroup);
267 link->cgroup = NULL;
268 }
269
270 /**
271 * cgroup_bpf_release() - put references of all bpf programs and
272 * release all cgroup bpf data
273 * @work: work structure embedded into the cgroup to modify
274 */
275 static void cgroup_bpf_release(struct work_struct *work)
276 {
277 struct cgroup *p, *cgrp = container_of(work, struct cgroup,
278 bpf.release_work);
279 struct bpf_prog_array *old_array;
280 struct list_head *storages = &cgrp->bpf.storages;
281 struct bpf_cgroup_storage *storage, *stmp;
282
283 unsigned int atype;
284
285 cgroup_lock();
286
287 for (atype = 0; atype < ARRAY_SIZE(cgrp->bpf.progs); atype++) {
288 struct hlist_head *progs = &cgrp->bpf.progs[atype];
289 struct bpf_prog_list *pl;
290 struct hlist_node *pltmp;
291
292 hlist_for_each_entry_safe(pl, pltmp, progs, node) {
293 hlist_del(&pl->node);
294 if (pl->prog) {
295 if (pl->prog->expected_attach_type == BPF_LSM_CGROUP)
296 bpf_trampoline_unlink_cgroup_shim(pl->prog);
297 bpf_prog_put(pl->prog);
298 }
299 if (pl->link) {
300 if (pl->link->link.prog->expected_attach_type == BPF_LSM_CGROUP)
301 bpf_trampoline_unlink_cgroup_shim(pl->link->link.prog);
302 bpf_cgroup_link_auto_detach(pl->link);
303 }
304 kfree(pl);
305 static_branch_dec(&cgroup_bpf_enabled_key[atype]);
306 }
307 old_array = rcu_dereference_protected(
308 cgrp->bpf.effective[atype],
309 lockdep_is_held(&cgroup_mutex));
310 bpf_prog_array_free(old_array);
311 }
312
313 list_for_each_entry_safe(storage, stmp, storages, list_cg) {
314 bpf_cgroup_storage_unlink(storage);
315 bpf_cgroup_storage_free(storage);
316 }
317
318 cgroup_unlock();
319
320 for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p))
321 cgroup_bpf_put(p);
322
323 percpu_ref_exit(&cgrp->bpf.refcnt);
324 cgroup_put(cgrp);
325 }
326
327 /**
328 * cgroup_bpf_release_fn() - callback used to schedule releasing
329 * of bpf cgroup data
330 * @ref: percpu ref counter structure
331 */
332 static void cgroup_bpf_release_fn(struct percpu_ref *ref)
333 {
334 struct cgroup *cgrp = container_of(ref, struct cgroup, bpf.refcnt);
335
336 INIT_WORK(&cgrp->bpf.release_work, cgroup_bpf_release);
337 queue_work(system_wq, &cgrp->bpf.release_work);
338 }
339
340 /* Get underlying bpf_prog of bpf_prog_list entry, regardless if it's through
341 * link or direct prog.
342 */
343 static struct bpf_prog *prog_list_prog(struct bpf_prog_list *pl)
344 {
345 if (pl->prog)
346 return pl->prog;
347 if (pl->link)
348 return pl->link->link.prog;
349 return NULL;
350 }
351
352 /* count number of elements in the list.
353 * it's slow but the list cannot be long
354 */
355 static u32 prog_list_length(struct hlist_head *head)
356 {
357 struct bpf_prog_list *pl;
358 u32 cnt = 0;
359
360 hlist_for_each_entry(pl, head, node) {
361 if (!prog_list_prog(pl))
362 continue;
363 cnt++;
364 }
365 return cnt;
366 }
367
368 /* if parent has non-overridable prog attached,
369 * disallow attaching new programs to the descendent cgroup.
370 * if parent has overridable or multi-prog, allow attaching
371 */
372 static bool hierarchy_allows_attach(struct cgroup *cgrp,
373 enum cgroup_bpf_attach_type atype)
374 {
375 struct cgroup *p;
376
377 p = cgroup_parent(cgrp);
378 if (!p)
379 return true;
380 do {
381 u32 flags = p->bpf.flags[atype];
382 u32 cnt;
383
384 if (flags & BPF_F_ALLOW_MULTI)
385 return true;
386 cnt = prog_list_length(&p->bpf.progs[atype]);
387 WARN_ON_ONCE(cnt > 1);
388 if (cnt == 1)
389 return !!(flags & BPF_F_ALLOW_OVERRIDE);
390 p = cgroup_parent(p);
391 } while (p);
392 return true;
393 }
394
395 /* compute a chain of effective programs for a given cgroup:
396 * start from the list of programs in this cgroup and add
397 * all parent programs.
398 * Note that parent's F_ALLOW_OVERRIDE-type program is yielding
399 * to programs in this cgroup
400 */
401 static int compute_effective_progs(struct cgroup *cgrp,
402 enum cgroup_bpf_attach_type atype,
403 struct bpf_prog_array **array)
404 {
405 struct bpf_prog_array_item *item;
406 struct bpf_prog_array *progs;
407 struct bpf_prog_list *pl;
408 struct cgroup *p = cgrp;
409 int cnt = 0;
410
411 /* count number of effective programs by walking parents */
412 do {
413 if (cnt == 0 || (p->bpf.flags[atype] & BPF_F_ALLOW_MULTI))
414 cnt += prog_list_length(&p->bpf.progs[atype]);
415 p = cgroup_parent(p);
416 } while (p);
417
418 progs = bpf_prog_array_alloc(cnt, GFP_KERNEL);
419 if (!progs)
420 return -ENOMEM;
421
422 /* populate the array with effective progs */
423 cnt = 0;
424 p = cgrp;
425 do {
426 if (cnt > 0 && !(p->bpf.flags[atype] & BPF_F_ALLOW_MULTI))
427 continue;
428
429 hlist_for_each_entry(pl, &p->bpf.progs[atype], node) {
430 if (!prog_list_prog(pl))
431 continue;
432
433 item = &progs->items[cnt];
434 item->prog = prog_list_prog(pl);
435 bpf_cgroup_storages_assign(item->cgroup_storage,
436 pl->storage);
437 cnt++;
438 }
439 } while ((p = cgroup_parent(p)));
440
441 *array = progs;
442 return 0;
443 }
444
445 static void activate_effective_progs(struct cgroup *cgrp,
446 enum cgroup_bpf_attach_type atype,
447 struct bpf_prog_array *old_array)
448 {
449 old_array = rcu_replace_pointer(cgrp->bpf.effective[atype], old_array,
450 lockdep_is_held(&cgroup_mutex));
451 /* free prog array after grace period, since __cgroup_bpf_run_*()
452 * might be still walking the array
453 */
454 bpf_prog_array_free(old_array);
455 }
456
457 /**
458 * cgroup_bpf_inherit() - inherit effective programs from parent
459 * @cgrp: the cgroup to modify
460 */
461 int cgroup_bpf_inherit(struct cgroup *cgrp)
462 {
463 /* has to use marco instead of const int, since compiler thinks
464 * that array below is variable length
465 */
466 #define NR ARRAY_SIZE(cgrp->bpf.effective)
467 struct bpf_prog_array *arrays[NR] = {};
468 struct cgroup *p;
469 int ret, i;
470
471 ret = percpu_ref_init(&cgrp->bpf.refcnt, cgroup_bpf_release_fn, 0,
472 GFP_KERNEL);
473 if (ret)
474 return ret;
475
476 for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p))
477 cgroup_bpf_get(p);
478
479 for (i = 0; i < NR; i++)
480 INIT_HLIST_HEAD(&cgrp->bpf.progs[i]);
481
482 INIT_LIST_HEAD(&cgrp->bpf.storages);
483
484 for (i = 0; i < NR; i++)
485 if (compute_effective_progs(cgrp, i, &arrays[i]))
486 goto cleanup;
487
488 for (i = 0; i < NR; i++)
489 activate_effective_progs(cgrp, i, arrays[i]);
490
491 return 0;
492 cleanup:
493 for (i = 0; i < NR; i++)
494 bpf_prog_array_free(arrays[i]);
495
496 for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p))
497 cgroup_bpf_put(p);
498
499 percpu_ref_exit(&cgrp->bpf.refcnt);
500
501 return -ENOMEM;
502 }
503
504 static int update_effective_progs(struct cgroup *cgrp,
505 enum cgroup_bpf_attach_type atype)
506 {
507 struct cgroup_subsys_state *css;
508 int err;
509
510 /* allocate and recompute effective prog arrays */
511 css_for_each_descendant_pre(css, &cgrp->self) {
512 struct cgroup *desc = container_of(css, struct cgroup, self);
513
514 if (percpu_ref_is_zero(&desc->bpf.refcnt))
515 continue;
516
517 err = compute_effective_progs(desc, atype, &desc->bpf.inactive);
518 if (err)
519 goto cleanup;
520 }
521
522 /* all allocations were successful. Activate all prog arrays */
523 css_for_each_descendant_pre(css, &cgrp->self) {
524 struct cgroup *desc = container_of(css, struct cgroup, self);
525
526 if (percpu_ref_is_zero(&desc->bpf.refcnt)) {
527 if (unlikely(desc->bpf.inactive)) {
528 bpf_prog_array_free(desc->bpf.inactive);
529 desc->bpf.inactive = NULL;
530 }
531 continue;
532 }
533
534 activate_effective_progs(desc, atype, desc->bpf.inactive);
535 desc->bpf.inactive = NULL;
536 }
537
538 return 0;
539
540 cleanup:
541 /* oom while computing effective. Free all computed effective arrays
542 * since they were not activated
543 */
544 css_for_each_descendant_pre(css, &cgrp->self) {
545 struct cgroup *desc = container_of(css, struct cgroup, self);
546
547 bpf_prog_array_free(desc->bpf.inactive);
548 desc->bpf.inactive = NULL;
549 }
550
551 return err;
552 }
553
554 #define BPF_CGROUP_MAX_PROGS 64
555
556 static struct bpf_prog_list *find_attach_entry(struct hlist_head *progs,
557 struct bpf_prog *prog,
558 struct bpf_cgroup_link *link,
559 struct bpf_prog *replace_prog,
560 bool allow_multi)
561 {
562 struct bpf_prog_list *pl;
563
564 /* single-attach case */
565 if (!allow_multi) {
566 if (hlist_empty(progs))
567 return NULL;
568 return hlist_entry(progs->first, typeof(*pl), node);
569 }
570
571 hlist_for_each_entry(pl, progs, node) {
572 if (prog && pl->prog == prog && prog != replace_prog)
573 /* disallow attaching the same prog twice */
574 return ERR_PTR(-EINVAL);
575 if (link && pl->link == link)
576 /* disallow attaching the same link twice */
577 return ERR_PTR(-EINVAL);
578 }
579
580 /* direct prog multi-attach w/ replacement case */
581 if (replace_prog) {
582 hlist_for_each_entry(pl, progs, node) {
583 if (pl->prog == replace_prog)
584 /* a match found */
585 return pl;
586 }
587 /* prog to replace not found for cgroup */
588 return ERR_PTR(-ENOENT);
589 }
590
591 return NULL;
592 }
593
594 /**
595 * __cgroup_bpf_attach() - Attach the program or the link to a cgroup, and
596 * propagate the change to descendants
597 * @cgrp: The cgroup which descendants to traverse
598 * @prog: A program to attach
599 * @link: A link to attach
600 * @replace_prog: Previously attached program to replace if BPF_F_REPLACE is set
601 * @type: Type of attach operation
602 * @flags: Option flags
603 *
604 * Exactly one of @prog or @link can be non-null.
605 * Must be called with cgroup_mutex held.
606 */
607 static int __cgroup_bpf_attach(struct cgroup *cgrp,
608 struct bpf_prog *prog, struct bpf_prog *replace_prog,
609 struct bpf_cgroup_link *link,
610 enum bpf_attach_type type, u32 flags)
611 {
612 u32 saved_flags = (flags & (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI));
613 struct bpf_prog *old_prog = NULL;
614 struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE] = {};
615 struct bpf_cgroup_storage *new_storage[MAX_BPF_CGROUP_STORAGE_TYPE] = {};
616 struct bpf_prog *new_prog = prog ? : link->link.prog;
617 enum cgroup_bpf_attach_type atype;
618 struct bpf_prog_list *pl;
619 struct hlist_head *progs;
620 int err;
621
622 if (((flags & BPF_F_ALLOW_OVERRIDE) && (flags & BPF_F_ALLOW_MULTI)) ||
623 ((flags & BPF_F_REPLACE) && !(flags & BPF_F_ALLOW_MULTI)))
624 /* invalid combination */
625 return -EINVAL;
626 if (link && (prog || replace_prog))
627 /* only either link or prog/replace_prog can be specified */
628 return -EINVAL;
629 if (!!replace_prog != !!(flags & BPF_F_REPLACE))
630 /* replace_prog implies BPF_F_REPLACE, and vice versa */
631 return -EINVAL;
632
633 atype = bpf_cgroup_atype_find(type, new_prog->aux->attach_btf_id);
634 if (atype < 0)
635 return -EINVAL;
636
637 progs = &cgrp->bpf.progs[atype];
638
639 if (!hierarchy_allows_attach(cgrp, atype))
640 return -EPERM;
641
642 if (!hlist_empty(progs) && cgrp->bpf.flags[atype] != saved_flags)
643 /* Disallow attaching non-overridable on top
644 * of existing overridable in this cgroup.
645 * Disallow attaching multi-prog if overridable or none
646 */
647 return -EPERM;
648
649 if (prog_list_length(progs) >= BPF_CGROUP_MAX_PROGS)
650 return -E2BIG;
651
652 pl = find_attach_entry(progs, prog, link, replace_prog,
653 flags & BPF_F_ALLOW_MULTI);
654 if (IS_ERR(pl))
655 return PTR_ERR(pl);
656
657 if (bpf_cgroup_storages_alloc(storage, new_storage, type,
658 prog ? : link->link.prog, cgrp))
659 return -ENOMEM;
660
661 if (pl) {
662 old_prog = pl->prog;
663 } else {
664 struct hlist_node *last = NULL;
665
666 pl = kmalloc(sizeof(*pl), GFP_KERNEL);
667 if (!pl) {
668 bpf_cgroup_storages_free(new_storage);
669 return -ENOMEM;
670 }
671 if (hlist_empty(progs))
672 hlist_add_head(&pl->node, progs);
673 else
674 hlist_for_each(last, progs) {
675 if (last->next)
676 continue;
677 hlist_add_behind(&pl->node, last);
678 break;
679 }
680 }
681
682 pl->prog = prog;
683 pl->link = link;
684 bpf_cgroup_storages_assign(pl->storage, storage);
685 cgrp->bpf.flags[atype] = saved_flags;
686
687 if (type == BPF_LSM_CGROUP) {
688 err = bpf_trampoline_link_cgroup_shim(new_prog, atype);
689 if (err)
690 goto cleanup;
691 }
692
693 err = update_effective_progs(cgrp, atype);
694 if (err)
695 goto cleanup_trampoline;
696
697 if (old_prog) {
698 if (type == BPF_LSM_CGROUP)
699 bpf_trampoline_unlink_cgroup_shim(old_prog);
700 bpf_prog_put(old_prog);
701 } else {
702 static_branch_inc(&cgroup_bpf_enabled_key[atype]);
703 }
704 bpf_cgroup_storages_link(new_storage, cgrp, type);
705 return 0;
706
707 cleanup_trampoline:
708 if (type == BPF_LSM_CGROUP)
709 bpf_trampoline_unlink_cgroup_shim(new_prog);
710
711 cleanup:
712 if (old_prog) {
713 pl->prog = old_prog;
714 pl->link = NULL;
715 }
716 bpf_cgroup_storages_free(new_storage);
717 if (!old_prog) {
718 hlist_del(&pl->node);
719 kfree(pl);
720 }
721 return err;
722 }
723
724 static int cgroup_bpf_attach(struct cgroup *cgrp,
725 struct bpf_prog *prog, struct bpf_prog *replace_prog,
726 struct bpf_cgroup_link *link,
727 enum bpf_attach_type type,
728 u32 flags)
729 {
730 int ret;
731
732 cgroup_lock();
733 ret = __cgroup_bpf_attach(cgrp, prog, replace_prog, link, type, flags);
734 cgroup_unlock();
735 return ret;
736 }
737
738 /* Swap updated BPF program for given link in effective program arrays across
739 * all descendant cgroups. This function is guaranteed to succeed.
740 */
741 static void replace_effective_prog(struct cgroup *cgrp,
742 enum cgroup_bpf_attach_type atype,
743 struct bpf_cgroup_link *link)
744 {
745 struct bpf_prog_array_item *item;
746 struct cgroup_subsys_state *css;
747 struct bpf_prog_array *progs;
748 struct bpf_prog_list *pl;
749 struct hlist_head *head;
750 struct cgroup *cg;
751 int pos;
752
753 css_for_each_descendant_pre(css, &cgrp->self) {
754 struct cgroup *desc = container_of(css, struct cgroup, self);
755
756 if (percpu_ref_is_zero(&desc->bpf.refcnt))
757 continue;
758
759 /* find position of link in effective progs array */
760 for (pos = 0, cg = desc; cg; cg = cgroup_parent(cg)) {
761 if (pos && !(cg->bpf.flags[atype] & BPF_F_ALLOW_MULTI))
762 continue;
763
764 head = &cg->bpf.progs[atype];
765 hlist_for_each_entry(pl, head, node) {
766 if (!prog_list_prog(pl))
767 continue;
768 if (pl->link == link)
769 goto found;
770 pos++;
771 }
772 }
773 found:
774 BUG_ON(!cg);
775 progs = rcu_dereference_protected(
776 desc->bpf.effective[atype],
777 lockdep_is_held(&cgroup_mutex));
778 item = &progs->items[pos];
779 WRITE_ONCE(item->prog, link->link.prog);
780 }
781 }
782
783 /**
784 * __cgroup_bpf_replace() - Replace link's program and propagate the change
785 * to descendants
786 * @cgrp: The cgroup which descendants to traverse
787 * @link: A link for which to replace BPF program
788 * @new_prog: &struct bpf_prog for the target BPF program with its refcnt
789 * incremented
790 *
791 * Must be called with cgroup_mutex held.
792 */
793 static int __cgroup_bpf_replace(struct cgroup *cgrp,
794 struct bpf_cgroup_link *link,
795 struct bpf_prog *new_prog)
796 {
797 enum cgroup_bpf_attach_type atype;
798 struct bpf_prog *old_prog;
799 struct bpf_prog_list *pl;
800 struct hlist_head *progs;
801 bool found = false;
802
803 atype = bpf_cgroup_atype_find(link->type, new_prog->aux->attach_btf_id);
804 if (atype < 0)
805 return -EINVAL;
806
807 progs = &cgrp->bpf.progs[atype];
808
809 if (link->link.prog->type != new_prog->type)
810 return -EINVAL;
811
812 hlist_for_each_entry(pl, progs, node) {
813 if (pl->link == link) {
814 found = true;
815 break;
816 }
817 }
818 if (!found)
819 return -ENOENT;
820
821 old_prog = xchg(&link->link.prog, new_prog);
822 replace_effective_prog(cgrp, atype, link);
823 bpf_prog_put(old_prog);
824 return 0;
825 }
826
827 static int cgroup_bpf_replace(struct bpf_link *link, struct bpf_prog *new_prog,
828 struct bpf_prog *old_prog)
829 {
830 struct bpf_cgroup_link *cg_link;
831 int ret;
832
833 cg_link = container_of(link, struct bpf_cgroup_link, link);
834
835 cgroup_lock();
836 /* link might have been auto-released by dying cgroup, so fail */
837 if (!cg_link->cgroup) {
838 ret = -ENOLINK;
839 goto out_unlock;
840 }
841 if (old_prog && link->prog != old_prog) {
842 ret = -EPERM;
843 goto out_unlock;
844 }
845 ret = __cgroup_bpf_replace(cg_link->cgroup, cg_link, new_prog);
846 out_unlock:
847 cgroup_unlock();
848 return ret;
849 }
850
851 static struct bpf_prog_list *find_detach_entry(struct hlist_head *progs,
852 struct bpf_prog *prog,
853 struct bpf_cgroup_link *link,
854 bool allow_multi)
855 {
856 struct bpf_prog_list *pl;
857
858 if (!allow_multi) {
859 if (hlist_empty(progs))
860 /* report error when trying to detach and nothing is attached */
861 return ERR_PTR(-ENOENT);
862
863 /* to maintain backward compatibility NONE and OVERRIDE cgroups
864 * allow detaching with invalid FD (prog==NULL) in legacy mode
865 */
866 return hlist_entry(progs->first, typeof(*pl), node);
867 }
868
869 if (!prog && !link)
870 /* to detach MULTI prog the user has to specify valid FD
871 * of the program or link to be detached
872 */
873 return ERR_PTR(-EINVAL);
874
875 /* find the prog or link and detach it */
876 hlist_for_each_entry(pl, progs, node) {
877 if (pl->prog == prog && pl->link == link)
878 return pl;
879 }
880 return ERR_PTR(-ENOENT);
881 }
882
883 /**
884 * purge_effective_progs() - After compute_effective_progs fails to alloc new
885 * cgrp->bpf.inactive table we can recover by
886 * recomputing the array in place.
887 *
888 * @cgrp: The cgroup which descendants to travers
889 * @prog: A program to detach or NULL
890 * @link: A link to detach or NULL
891 * @atype: Type of detach operation
892 */
893 static void purge_effective_progs(struct cgroup *cgrp, struct bpf_prog *prog,
894 struct bpf_cgroup_link *link,
895 enum cgroup_bpf_attach_type atype)
896 {
897 struct cgroup_subsys_state *css;
898 struct bpf_prog_array *progs;
899 struct bpf_prog_list *pl;
900 struct hlist_head *head;
901 struct cgroup *cg;
902 int pos;
903
904 /* recompute effective prog array in place */
905 css_for_each_descendant_pre(css, &cgrp->self) {
906 struct cgroup *desc = container_of(css, struct cgroup, self);
907
908 if (percpu_ref_is_zero(&desc->bpf.refcnt))
909 continue;
910
911 /* find position of link or prog in effective progs array */
912 for (pos = 0, cg = desc; cg; cg = cgroup_parent(cg)) {
913 if (pos && !(cg->bpf.flags[atype] & BPF_F_ALLOW_MULTI))
914 continue;
915
916 head = &cg->bpf.progs[atype];
917 hlist_for_each_entry(pl, head, node) {
918 if (!prog_list_prog(pl))
919 continue;
920 if (pl->prog == prog && pl->link == link)
921 goto found;
922 pos++;
923 }
924 }
925
926 /* no link or prog match, skip the cgroup of this layer */
927 continue;
928 found:
929 progs = rcu_dereference_protected(
930 desc->bpf.effective[atype],
931 lockdep_is_held(&cgroup_mutex));
932
933 /* Remove the program from the array */
934 WARN_ONCE(bpf_prog_array_delete_safe_at(progs, pos),
935 "Failed to purge a prog from array at index %d", pos);
936 }
937 }
938
939 /**
940 * __cgroup_bpf_detach() - Detach the program or link from a cgroup, and
941 * propagate the change to descendants
942 * @cgrp: The cgroup which descendants to traverse
943 * @prog: A program to detach or NULL
944 * @link: A link to detach or NULL
945 * @type: Type of detach operation
946 *
947 * At most one of @prog or @link can be non-NULL.
948 * Must be called with cgroup_mutex held.
949 */
950 static int __cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
951 struct bpf_cgroup_link *link, enum bpf_attach_type type)
952 {
953 enum cgroup_bpf_attach_type atype;
954 struct bpf_prog *old_prog;
955 struct bpf_prog_list *pl;
956 struct hlist_head *progs;
957 u32 attach_btf_id = 0;
958 u32 flags;
959
960 if (prog)
961 attach_btf_id = prog->aux->attach_btf_id;
962 if (link)
963 attach_btf_id = link->link.prog->aux->attach_btf_id;
964
965 atype = bpf_cgroup_atype_find(type, attach_btf_id);
966 if (atype < 0)
967 return -EINVAL;
968
969 progs = &cgrp->bpf.progs[atype];
970 flags = cgrp->bpf.flags[atype];
971
972 if (prog && link)
973 /* only one of prog or link can be specified */
974 return -EINVAL;
975
976 pl = find_detach_entry(progs, prog, link, flags & BPF_F_ALLOW_MULTI);
977 if (IS_ERR(pl))
978 return PTR_ERR(pl);
979
980 /* mark it deleted, so it's ignored while recomputing effective */
981 old_prog = pl->prog;
982 pl->prog = NULL;
983 pl->link = NULL;
984
985 if (update_effective_progs(cgrp, atype)) {
986 /* if update effective array failed replace the prog with a dummy prog*/
987 pl->prog = old_prog;
988 pl->link = link;
989 purge_effective_progs(cgrp, old_prog, link, atype);
990 }
991
992 /* now can actually delete it from this cgroup list */
993 hlist_del(&pl->node);
994
995 kfree(pl);
996 if (hlist_empty(progs))
997 /* last program was detached, reset flags to zero */
998 cgrp->bpf.flags[atype] = 0;
999 if (old_prog) {
1000 if (type == BPF_LSM_CGROUP)
1001 bpf_trampoline_unlink_cgroup_shim(old_prog);
1002 bpf_prog_put(old_prog);
1003 }
1004 static_branch_dec(&cgroup_bpf_enabled_key[atype]);
1005 return 0;
1006 }
1007
1008 static int cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
1009 enum bpf_attach_type type)
1010 {
1011 int ret;
1012
1013 cgroup_lock();
1014 ret = __cgroup_bpf_detach(cgrp, prog, NULL, type);
1015 cgroup_unlock();
1016 return ret;
1017 }
1018
1019 /* Must be called with cgroup_mutex held to avoid races. */
1020 static int __cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
1021 union bpf_attr __user *uattr)
1022 {
1023 __u32 __user *prog_attach_flags = u64_to_user_ptr(attr->query.prog_attach_flags);
1024 bool effective_query = attr->query.query_flags & BPF_F_QUERY_EFFECTIVE;
1025 __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
1026 enum bpf_attach_type type = attr->query.attach_type;
1027 enum cgroup_bpf_attach_type from_atype, to_atype;
1028 enum cgroup_bpf_attach_type atype;
1029 struct bpf_prog_array *effective;
1030 int cnt, ret = 0, i;
1031 int total_cnt = 0;
1032 u32 flags;
1033
1034 if (effective_query && prog_attach_flags)
1035 return -EINVAL;
1036
1037 if (type == BPF_LSM_CGROUP) {
1038 if (!effective_query && attr->query.prog_cnt &&
1039 prog_ids && !prog_attach_flags)
1040 return -EINVAL;
1041
1042 from_atype = CGROUP_LSM_START;
1043 to_atype = CGROUP_LSM_END;
1044 flags = 0;
1045 } else {
1046 from_atype = to_cgroup_bpf_attach_type(type);
1047 if (from_atype < 0)
1048 return -EINVAL;
1049 to_atype = from_atype;
1050 flags = cgrp->bpf.flags[from_atype];
1051 }
1052
1053 for (atype = from_atype; atype <= to_atype; atype++) {
1054 if (effective_query) {
1055 effective = rcu_dereference_protected(cgrp->bpf.effective[atype],
1056 lockdep_is_held(&cgroup_mutex));
1057 total_cnt += bpf_prog_array_length(effective);
1058 } else {
1059 total_cnt += prog_list_length(&cgrp->bpf.progs[atype]);
1060 }
1061 }
1062
1063 /* always output uattr->query.attach_flags as 0 during effective query */
1064 flags = effective_query ? 0 : flags;
1065 if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)))
1066 return -EFAULT;
1067 if (copy_to_user(&uattr->query.prog_cnt, &total_cnt, sizeof(total_cnt)))
1068 return -EFAULT;
1069 if (attr->query.prog_cnt == 0 || !prog_ids || !total_cnt)
1070 /* return early if user requested only program count + flags */
1071 return 0;
1072
1073 if (attr->query.prog_cnt < total_cnt) {
1074 total_cnt = attr->query.prog_cnt;
1075 ret = -ENOSPC;
1076 }
1077
1078 for (atype = from_atype; atype <= to_atype && total_cnt; atype++) {
1079 if (effective_query) {
1080 effective = rcu_dereference_protected(cgrp->bpf.effective[atype],
1081 lockdep_is_held(&cgroup_mutex));
1082 cnt = min_t(int, bpf_prog_array_length(effective), total_cnt);
1083 ret = bpf_prog_array_copy_to_user(effective, prog_ids, cnt);
1084 } else {
1085 struct hlist_head *progs;
1086 struct bpf_prog_list *pl;
1087 struct bpf_prog *prog;
1088 u32 id;
1089
1090 progs = &cgrp->bpf.progs[atype];
1091 cnt = min_t(int, prog_list_length(progs), total_cnt);
1092 i = 0;
1093 hlist_for_each_entry(pl, progs, node) {
1094 prog = prog_list_prog(pl);
1095 id = prog->aux->id;
1096 if (copy_to_user(prog_ids + i, &id, sizeof(id)))
1097 return -EFAULT;
1098 if (++i == cnt)
1099 break;
1100 }
1101
1102 if (prog_attach_flags) {
1103 flags = cgrp->bpf.flags[atype];
1104
1105 for (i = 0; i < cnt; i++)
1106 if (copy_to_user(prog_attach_flags + i,
1107 &flags, sizeof(flags)))
1108 return -EFAULT;
1109 prog_attach_flags += cnt;
1110 }
1111 }
1112
1113 prog_ids += cnt;
1114 total_cnt -= cnt;
1115 }
1116 return ret;
1117 }
1118
1119 static int cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
1120 union bpf_attr __user *uattr)
1121 {
1122 int ret;
1123
1124 cgroup_lock();
1125 ret = __cgroup_bpf_query(cgrp, attr, uattr);
1126 cgroup_unlock();
1127 return ret;
1128 }
1129
1130 int cgroup_bpf_prog_attach(const union bpf_attr *attr,
1131 enum bpf_prog_type ptype, struct bpf_prog *prog)
1132 {
1133 struct bpf_prog *replace_prog = NULL;
1134 struct cgroup *cgrp;
1135 int ret;
1136
1137 cgrp = cgroup_get_from_fd(attr->target_fd);
1138 if (IS_ERR(cgrp))
1139 return PTR_ERR(cgrp);
1140
1141 if ((attr->attach_flags & BPF_F_ALLOW_MULTI) &&
1142 (attr->attach_flags & BPF_F_REPLACE)) {
1143 replace_prog = bpf_prog_get_type(attr->replace_bpf_fd, ptype);
1144 if (IS_ERR(replace_prog)) {
1145 cgroup_put(cgrp);
1146 return PTR_ERR(replace_prog);
1147 }
1148 }
1149
1150 ret = cgroup_bpf_attach(cgrp, prog, replace_prog, NULL,
1151 attr->attach_type, attr->attach_flags);
1152
1153 if (replace_prog)
1154 bpf_prog_put(replace_prog);
1155 cgroup_put(cgrp);
1156 return ret;
1157 }
1158
1159 int cgroup_bpf_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
1160 {
1161 struct bpf_prog *prog;
1162 struct cgroup *cgrp;
1163 int ret;
1164
1165 cgrp = cgroup_get_from_fd(attr->target_fd);
1166 if (IS_ERR(cgrp))
1167 return PTR_ERR(cgrp);
1168
1169 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1170 if (IS_ERR(prog))
1171 prog = NULL;
1172
1173 ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type);
1174 if (prog)
1175 bpf_prog_put(prog);
1176
1177 cgroup_put(cgrp);
1178 return ret;
1179 }
1180
1181 static void bpf_cgroup_link_release(struct bpf_link *link)
1182 {
1183 struct bpf_cgroup_link *cg_link =
1184 container_of(link, struct bpf_cgroup_link, link);
1185 struct cgroup *cg;
1186
1187 /* link might have been auto-detached by dying cgroup already,
1188 * in that case our work is done here
1189 */
1190 if (!cg_link->cgroup)
1191 return;
1192
1193 cgroup_lock();
1194
1195 /* re-check cgroup under lock again */
1196 if (!cg_link->cgroup) {
1197 cgroup_unlock();
1198 return;
1199 }
1200
1201 WARN_ON(__cgroup_bpf_detach(cg_link->cgroup, NULL, cg_link,
1202 cg_link->type));
1203 if (cg_link->type == BPF_LSM_CGROUP)
1204 bpf_trampoline_unlink_cgroup_shim(cg_link->link.prog);
1205
1206 cg = cg_link->cgroup;
1207 cg_link->cgroup = NULL;
1208
1209 cgroup_unlock();
1210
1211 cgroup_put(cg);
1212 }
1213
1214 static void bpf_cgroup_link_dealloc(struct bpf_link *link)
1215 {
1216 struct bpf_cgroup_link *cg_link =
1217 container_of(link, struct bpf_cgroup_link, link);
1218
1219 kfree(cg_link);
1220 }
1221
1222 static int bpf_cgroup_link_detach(struct bpf_link *link)
1223 {
1224 bpf_cgroup_link_release(link);
1225
1226 return 0;
1227 }
1228
1229 static void bpf_cgroup_link_show_fdinfo(const struct bpf_link *link,
1230 struct seq_file *seq)
1231 {
1232 struct bpf_cgroup_link *cg_link =
1233 container_of(link, struct bpf_cgroup_link, link);
1234 u64 cg_id = 0;
1235
1236 cgroup_lock();
1237 if (cg_link->cgroup)
1238 cg_id = cgroup_id(cg_link->cgroup);
1239 cgroup_unlock();
1240
1241 seq_printf(seq,
1242 "cgroup_id:\t%llu\n"
1243 "attach_type:\t%d\n",
1244 cg_id,
1245 cg_link->type);
1246 }
1247
1248 static int bpf_cgroup_link_fill_link_info(const struct bpf_link *link,
1249 struct bpf_link_info *info)
1250 {
1251 struct bpf_cgroup_link *cg_link =
1252 container_of(link, struct bpf_cgroup_link, link);
1253 u64 cg_id = 0;
1254
1255 cgroup_lock();
1256 if (cg_link->cgroup)
1257 cg_id = cgroup_id(cg_link->cgroup);
1258 cgroup_unlock();
1259
1260 info->cgroup.cgroup_id = cg_id;
1261 info->cgroup.attach_type = cg_link->type;
1262 return 0;
1263 }
1264
1265 static const struct bpf_link_ops bpf_cgroup_link_lops = {
1266 .release = bpf_cgroup_link_release,
1267 .dealloc = bpf_cgroup_link_dealloc,
1268 .detach = bpf_cgroup_link_detach,
1269 .update_prog = cgroup_bpf_replace,
1270 .show_fdinfo = bpf_cgroup_link_show_fdinfo,
1271 .fill_link_info = bpf_cgroup_link_fill_link_info,
1272 };
1273
1274 int cgroup_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
1275 {
1276 struct bpf_link_primer link_primer;
1277 struct bpf_cgroup_link *link;
1278 struct cgroup *cgrp;
1279 int err;
1280
1281 if (attr->link_create.flags)
1282 return -EINVAL;
1283
1284 cgrp = cgroup_get_from_fd(attr->link_create.target_fd);
1285 if (IS_ERR(cgrp))
1286 return PTR_ERR(cgrp);
1287
1288 link = kzalloc(sizeof(*link), GFP_USER);
1289 if (!link) {
1290 err = -ENOMEM;
1291 goto out_put_cgroup;
1292 }
1293 bpf_link_init(&link->link, BPF_LINK_TYPE_CGROUP, &bpf_cgroup_link_lops,
1294 prog);
1295 link->cgroup = cgrp;
1296 link->type = attr->link_create.attach_type;
1297
1298 err = bpf_link_prime(&link->link, &link_primer);
1299 if (err) {
1300 kfree(link);
1301 goto out_put_cgroup;
1302 }
1303
1304 err = cgroup_bpf_attach(cgrp, NULL, NULL, link,
1305 link->type, BPF_F_ALLOW_MULTI);
1306 if (err) {
1307 bpf_link_cleanup(&link_primer);
1308 goto out_put_cgroup;
1309 }
1310
1311 return bpf_link_settle(&link_primer);
1312
1313 out_put_cgroup:
1314 cgroup_put(cgrp);
1315 return err;
1316 }
1317
1318 int cgroup_bpf_prog_query(const union bpf_attr *attr,
1319 union bpf_attr __user *uattr)
1320 {
1321 struct cgroup *cgrp;
1322 int ret;
1323
1324 cgrp = cgroup_get_from_fd(attr->query.target_fd);
1325 if (IS_ERR(cgrp))
1326 return PTR_ERR(cgrp);
1327
1328 ret = cgroup_bpf_query(cgrp, attr, uattr);
1329
1330 cgroup_put(cgrp);
1331 return ret;
1332 }
1333
1334 /**
1335 * __cgroup_bpf_run_filter_skb() - Run a program for packet filtering
1336 * @sk: The socket sending or receiving traffic
1337 * @skb: The skb that is being sent or received
1338 * @atype: The type of program to be executed
1339 *
1340 * If no socket is passed, or the socket is not of type INET or INET6,
1341 * this function does nothing and returns 0.
1342 *
1343 * The program type passed in via @type must be suitable for network
1344 * filtering. No further check is performed to assert that.
1345 *
1346 * For egress packets, this function can return:
1347 * NET_XMIT_SUCCESS (0) - continue with packet output
1348 * NET_XMIT_DROP (1) - drop packet and notify TCP to call cwr
1349 * NET_XMIT_CN (2) - continue with packet output and notify TCP
1350 * to call cwr
1351 * -err - drop packet
1352 *
1353 * For ingress packets, this function will return -EPERM if any
1354 * attached program was found and if it returned != 1 during execution.
1355 * Otherwise 0 is returned.
1356 */
1357 int __cgroup_bpf_run_filter_skb(struct sock *sk,
1358 struct sk_buff *skb,
1359 enum cgroup_bpf_attach_type atype)
1360 {
1361 unsigned int offset = -skb_network_offset(skb);
1362 struct sock *save_sk;
1363 void *saved_data_end;
1364 struct cgroup *cgrp;
1365 int ret;
1366
1367 if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)
1368 return 0;
1369
1370 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1371 save_sk = skb->sk;
1372 skb->sk = sk;
1373 __skb_push(skb, offset);
1374
1375 /* compute pointers for the bpf prog */
1376 bpf_compute_and_save_data_end(skb, &saved_data_end);
1377
1378 if (atype == CGROUP_INET_EGRESS) {
1379 u32 flags = 0;
1380 bool cn;
1381
1382 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, skb,
1383 __bpf_prog_run_save_cb, 0, &flags);
1384
1385 /* Return values of CGROUP EGRESS BPF programs are:
1386 * 0: drop packet
1387 * 1: keep packet
1388 * 2: drop packet and cn
1389 * 3: keep packet and cn
1390 *
1391 * The returned value is then converted to one of the NET_XMIT
1392 * or an error code that is then interpreted as drop packet
1393 * (and no cn):
1394 * 0: NET_XMIT_SUCCESS skb should be transmitted
1395 * 1: NET_XMIT_DROP skb should be dropped and cn
1396 * 2: NET_XMIT_CN skb should be transmitted and cn
1397 * 3: -err skb should be dropped
1398 */
1399
1400 cn = flags & BPF_RET_SET_CN;
1401 if (ret && !IS_ERR_VALUE((long)ret))
1402 ret = -EFAULT;
1403 if (!ret)
1404 ret = (cn ? NET_XMIT_CN : NET_XMIT_SUCCESS);
1405 else
1406 ret = (cn ? NET_XMIT_DROP : ret);
1407 } else {
1408 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype,
1409 skb, __bpf_prog_run_save_cb, 0,
1410 NULL);
1411 if (ret && !IS_ERR_VALUE((long)ret))
1412 ret = -EFAULT;
1413 }
1414 bpf_restore_data_end(skb, saved_data_end);
1415 __skb_pull(skb, offset);
1416 skb->sk = save_sk;
1417
1418 return ret;
1419 }
1420 EXPORT_SYMBOL(__cgroup_bpf_run_filter_skb);
1421
1422 /**
1423 * __cgroup_bpf_run_filter_sk() - Run a program on a sock
1424 * @sk: sock structure to manipulate
1425 * @atype: The type of program to be executed
1426 *
1427 * socket is passed is expected to be of type INET or INET6.
1428 *
1429 * The program type passed in via @type must be suitable for sock
1430 * filtering. No further check is performed to assert that.
1431 *
1432 * This function will return %-EPERM if any if an attached program was found
1433 * and if it returned != 1 during execution. In all other cases, 0 is returned.
1434 */
1435 int __cgroup_bpf_run_filter_sk(struct sock *sk,
1436 enum cgroup_bpf_attach_type atype)
1437 {
1438 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1439
1440 return bpf_prog_run_array_cg(&cgrp->bpf, atype, sk, bpf_prog_run, 0,
1441 NULL);
1442 }
1443 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sk);
1444
1445 /**
1446 * __cgroup_bpf_run_filter_sock_addr() - Run a program on a sock and
1447 * provided by user sockaddr
1448 * @sk: sock struct that will use sockaddr
1449 * @uaddr: sockaddr struct provided by user
1450 * @uaddrlen: Pointer to the size of the sockaddr struct provided by user. It is
1451 * read-only for AF_INET[6] uaddr but can be modified for AF_UNIX
1452 * uaddr.
1453 * @atype: The type of program to be executed
1454 * @t_ctx: Pointer to attach type specific context
1455 * @flags: Pointer to u32 which contains higher bits of BPF program
1456 * return value (OR'ed together).
1457 *
1458 * socket is expected to be of type INET, INET6 or UNIX.
1459 *
1460 * This function will return %-EPERM if an attached program is found and
1461 * returned value != 1 during execution. In all other cases, 0 is returned.
1462 */
1463 int __cgroup_bpf_run_filter_sock_addr(struct sock *sk,
1464 struct sockaddr *uaddr,
1465 int *uaddrlen,
1466 enum cgroup_bpf_attach_type atype,
1467 void *t_ctx,
1468 u32 *flags)
1469 {
1470 struct bpf_sock_addr_kern ctx = {
1471 .sk = sk,
1472 .uaddr = uaddr,
1473 .t_ctx = t_ctx,
1474 };
1475 struct sockaddr_storage unspec;
1476 struct cgroup *cgrp;
1477 int ret;
1478
1479 /* Check socket family since not all sockets represent network
1480 * endpoint (e.g. AF_UNIX).
1481 */
1482 if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6 &&
1483 sk->sk_family != AF_UNIX)
1484 return 0;
1485
1486 if (!ctx.uaddr) {
1487 memset(&unspec, 0, sizeof(unspec));
1488 ctx.uaddr = (struct sockaddr *)&unspec;
1489 ctx.uaddrlen = 0;
1490 } else {
1491 ctx.uaddrlen = *uaddrlen;
1492 }
1493
1494 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1495 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, &ctx, bpf_prog_run,
1496 0, flags);
1497
1498 if (!ret && uaddr)
1499 *uaddrlen = ctx.uaddrlen;
1500
1501 return ret;
1502 }
1503 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_addr);
1504
1505 /**
1506 * __cgroup_bpf_run_filter_sock_ops() - Run a program on a sock
1507 * @sk: socket to get cgroup from
1508 * @sock_ops: bpf_sock_ops_kern struct to pass to program. Contains
1509 * sk with connection information (IP addresses, etc.) May not contain
1510 * cgroup info if it is a req sock.
1511 * @atype: The type of program to be executed
1512 *
1513 * socket passed is expected to be of type INET or INET6.
1514 *
1515 * The program type passed in via @type must be suitable for sock_ops
1516 * filtering. No further check is performed to assert that.
1517 *
1518 * This function will return %-EPERM if any if an attached program was found
1519 * and if it returned != 1 during execution. In all other cases, 0 is returned.
1520 */
1521 int __cgroup_bpf_run_filter_sock_ops(struct sock *sk,
1522 struct bpf_sock_ops_kern *sock_ops,
1523 enum cgroup_bpf_attach_type atype)
1524 {
1525 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1526
1527 return bpf_prog_run_array_cg(&cgrp->bpf, atype, sock_ops, bpf_prog_run,
1528 0, NULL);
1529 }
1530 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_ops);
1531
1532 int __cgroup_bpf_check_dev_permission(short dev_type, u32 major, u32 minor,
1533 short access, enum cgroup_bpf_attach_type atype)
1534 {
1535 struct cgroup *cgrp;
1536 struct bpf_cgroup_dev_ctx ctx = {
1537 .access_type = (access << 16) | dev_type,
1538 .major = major,
1539 .minor = minor,
1540 };
1541 int ret;
1542
1543 rcu_read_lock();
1544 cgrp = task_dfl_cgroup(current);
1545 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, &ctx, bpf_prog_run, 0,
1546 NULL);
1547 rcu_read_unlock();
1548
1549 return ret;
1550 }
1551
1552 BPF_CALL_2(bpf_get_local_storage, struct bpf_map *, map, u64, flags)
1553 {
1554 /* flags argument is not used now,
1555 * but provides an ability to extend the API.
1556 * verifier checks that its value is correct.
1557 */
1558 enum bpf_cgroup_storage_type stype = cgroup_storage_type(map);
1559 struct bpf_cgroup_storage *storage;
1560 struct bpf_cg_run_ctx *ctx;
1561 void *ptr;
1562
1563 /* get current cgroup storage from BPF run context */
1564 ctx = container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx);
1565 storage = ctx->prog_item->cgroup_storage[stype];
1566
1567 if (stype == BPF_CGROUP_STORAGE_SHARED)
1568 ptr = &READ_ONCE(storage->buf)->data[0];
1569 else
1570 ptr = this_cpu_ptr(storage->percpu_buf);
1571
1572 return (unsigned long)ptr;
1573 }
1574
1575 const struct bpf_func_proto bpf_get_local_storage_proto = {
1576 .func = bpf_get_local_storage,
1577 .gpl_only = false,
1578 .ret_type = RET_PTR_TO_MAP_VALUE,
1579 .arg1_type = ARG_CONST_MAP_PTR,
1580 .arg2_type = ARG_ANYTHING,
1581 };
1582
1583 BPF_CALL_0(bpf_get_retval)
1584 {
1585 struct bpf_cg_run_ctx *ctx =
1586 container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx);
1587
1588 return ctx->retval;
1589 }
1590
1591 const struct bpf_func_proto bpf_get_retval_proto = {
1592 .func = bpf_get_retval,
1593 .gpl_only = false,
1594 .ret_type = RET_INTEGER,
1595 };
1596
1597 BPF_CALL_1(bpf_set_retval, int, retval)
1598 {
1599 struct bpf_cg_run_ctx *ctx =
1600 container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx);
1601
1602 ctx->retval = retval;
1603 return 0;
1604 }
1605
1606 const struct bpf_func_proto bpf_set_retval_proto = {
1607 .func = bpf_set_retval,
1608 .gpl_only = false,
1609 .ret_type = RET_INTEGER,
1610 .arg1_type = ARG_ANYTHING,
1611 };
1612
1613 static const struct bpf_func_proto *
1614 cgroup_dev_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1615 {
1616 const struct bpf_func_proto *func_proto;
1617
1618 func_proto = cgroup_common_func_proto(func_id, prog);
1619 if (func_proto)
1620 return func_proto;
1621
1622 func_proto = cgroup_current_func_proto(func_id, prog);
1623 if (func_proto)
1624 return func_proto;
1625
1626 switch (func_id) {
1627 case BPF_FUNC_perf_event_output:
1628 return &bpf_event_output_data_proto;
1629 default:
1630 return bpf_base_func_proto(func_id, prog);
1631 }
1632 }
1633
1634 static bool cgroup_dev_is_valid_access(int off, int size,
1635 enum bpf_access_type type,
1636 const struct bpf_prog *prog,
1637 struct bpf_insn_access_aux *info)
1638 {
1639 const int size_default = sizeof(__u32);
1640
1641 if (type == BPF_WRITE)
1642 return false;
1643
1644 if (off < 0 || off + size > sizeof(struct bpf_cgroup_dev_ctx))
1645 return false;
1646 /* The verifier guarantees that size > 0. */
1647 if (off % size != 0)
1648 return false;
1649
1650 switch (off) {
1651 case bpf_ctx_range(struct bpf_cgroup_dev_ctx, access_type):
1652 bpf_ctx_record_field_size(info, size_default);
1653 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
1654 return false;
1655 break;
1656 default:
1657 if (size != size_default)
1658 return false;
1659 }
1660
1661 return true;
1662 }
1663
1664 const struct bpf_prog_ops cg_dev_prog_ops = {
1665 };
1666
1667 const struct bpf_verifier_ops cg_dev_verifier_ops = {
1668 .get_func_proto = cgroup_dev_func_proto,
1669 .is_valid_access = cgroup_dev_is_valid_access,
1670 };
1671
1672 /**
1673 * __cgroup_bpf_run_filter_sysctl - Run a program on sysctl
1674 *
1675 * @head: sysctl table header
1676 * @table: sysctl table
1677 * @write: sysctl is being read (= 0) or written (= 1)
1678 * @buf: pointer to buffer (in and out)
1679 * @pcount: value-result argument: value is size of buffer pointed to by @buf,
1680 * result is size of @new_buf if program set new value, initial value
1681 * otherwise
1682 * @ppos: value-result argument: value is position at which read from or write
1683 * to sysctl is happening, result is new position if program overrode it,
1684 * initial value otherwise
1685 * @atype: type of program to be executed
1686 *
1687 * Program is run when sysctl is being accessed, either read or written, and
1688 * can allow or deny such access.
1689 *
1690 * This function will return %-EPERM if an attached program is found and
1691 * returned value != 1 during execution. In all other cases 0 is returned.
1692 */
1693 int __cgroup_bpf_run_filter_sysctl(struct ctl_table_header *head,
1694 struct ctl_table *table, int write,
1695 char **buf, size_t *pcount, loff_t *ppos,
1696 enum cgroup_bpf_attach_type atype)
1697 {
1698 struct bpf_sysctl_kern ctx = {
1699 .head = head,
1700 .table = table,
1701 .write = write,
1702 .ppos = ppos,
1703 .cur_val = NULL,
1704 .cur_len = PAGE_SIZE,
1705 .new_val = NULL,
1706 .new_len = 0,
1707 .new_updated = 0,
1708 };
1709 struct cgroup *cgrp;
1710 loff_t pos = 0;
1711 int ret;
1712
1713 ctx.cur_val = kmalloc_track_caller(ctx.cur_len, GFP_KERNEL);
1714 if (!ctx.cur_val ||
1715 table->proc_handler(table, 0, ctx.cur_val, &ctx.cur_len, &pos)) {
1716 /* Let BPF program decide how to proceed. */
1717 ctx.cur_len = 0;
1718 }
1719
1720 if (write && *buf && *pcount) {
1721 /* BPF program should be able to override new value with a
1722 * buffer bigger than provided by user.
1723 */
1724 ctx.new_val = kmalloc_track_caller(PAGE_SIZE, GFP_KERNEL);
1725 ctx.new_len = min_t(size_t, PAGE_SIZE, *pcount);
1726 if (ctx.new_val) {
1727 memcpy(ctx.new_val, *buf, ctx.new_len);
1728 } else {
1729 /* Let BPF program decide how to proceed. */
1730 ctx.new_len = 0;
1731 }
1732 }
1733
1734 rcu_read_lock();
1735 cgrp = task_dfl_cgroup(current);
1736 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, &ctx, bpf_prog_run, 0,
1737 NULL);
1738 rcu_read_unlock();
1739
1740 kfree(ctx.cur_val);
1741
1742 if (ret == 1 && ctx.new_updated) {
1743 kfree(*buf);
1744 *buf = ctx.new_val;
1745 *pcount = ctx.new_len;
1746 } else {
1747 kfree(ctx.new_val);
1748 }
1749
1750 return ret;
1751 }
1752
1753 #ifdef CONFIG_NET
1754 static int sockopt_alloc_buf(struct bpf_sockopt_kern *ctx, int max_optlen,
1755 struct bpf_sockopt_buf *buf)
1756 {
1757 if (unlikely(max_optlen < 0))
1758 return -EINVAL;
1759
1760 if (unlikely(max_optlen > PAGE_SIZE)) {
1761 /* We don't expose optvals that are greater than PAGE_SIZE
1762 * to the BPF program.
1763 */
1764 max_optlen = PAGE_SIZE;
1765 }
1766
1767 if (max_optlen <= sizeof(buf->data)) {
1768 /* When the optval fits into BPF_SOCKOPT_KERN_BUF_SIZE
1769 * bytes avoid the cost of kzalloc.
1770 */
1771 ctx->optval = buf->data;
1772 ctx->optval_end = ctx->optval + max_optlen;
1773 return max_optlen;
1774 }
1775
1776 ctx->optval = kzalloc(max_optlen, GFP_USER);
1777 if (!ctx->optval)
1778 return -ENOMEM;
1779
1780 ctx->optval_end = ctx->optval + max_optlen;
1781
1782 return max_optlen;
1783 }
1784
1785 static void sockopt_free_buf(struct bpf_sockopt_kern *ctx,
1786 struct bpf_sockopt_buf *buf)
1787 {
1788 if (ctx->optval == buf->data)
1789 return;
1790 kfree(ctx->optval);
1791 }
1792
1793 static bool sockopt_buf_allocated(struct bpf_sockopt_kern *ctx,
1794 struct bpf_sockopt_buf *buf)
1795 {
1796 return ctx->optval != buf->data;
1797 }
1798
1799 int __cgroup_bpf_run_filter_setsockopt(struct sock *sk, int *level,
1800 int *optname, sockptr_t optval,
1801 int *optlen, char **kernel_optval)
1802 {
1803 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1804 struct bpf_sockopt_buf buf = {};
1805 struct bpf_sockopt_kern ctx = {
1806 .sk = sk,
1807 .level = *level,
1808 .optname = *optname,
1809 };
1810 int ret, max_optlen;
1811
1812 /* Allocate a bit more than the initial user buffer for
1813 * BPF program. The canonical use case is overriding
1814 * TCP_CONGESTION(nv) to TCP_CONGESTION(cubic).
1815 */
1816 max_optlen = max_t(int, 16, *optlen);
1817 max_optlen = sockopt_alloc_buf(&ctx, max_optlen, &buf);
1818 if (max_optlen < 0)
1819 return max_optlen;
1820
1821 ctx.optlen = *optlen;
1822
1823 if (copy_from_sockptr(ctx.optval, optval,
1824 min(*optlen, max_optlen))) {
1825 ret = -EFAULT;
1826 goto out;
1827 }
1828
1829 lock_sock(sk);
1830 ret = bpf_prog_run_array_cg(&cgrp->bpf, CGROUP_SETSOCKOPT,
1831 &ctx, bpf_prog_run, 0, NULL);
1832 release_sock(sk);
1833
1834 if (ret)
1835 goto out;
1836
1837 if (ctx.optlen == -1) {
1838 /* optlen set to -1, bypass kernel */
1839 ret = 1;
1840 } else if (ctx.optlen > max_optlen || ctx.optlen < -1) {
1841 /* optlen is out of bounds */
1842 if (*optlen > PAGE_SIZE && ctx.optlen >= 0) {
1843 pr_info_once("bpf setsockopt: ignoring program buffer with optlen=%d (max_optlen=%d)\n",
1844 ctx.optlen, max_optlen);
1845 ret = 0;
1846 goto out;
1847 }
1848 ret = -EFAULT;
1849 } else {
1850 /* optlen within bounds, run kernel handler */
1851 ret = 0;
1852
1853 /* export any potential modifications */
1854 *level = ctx.level;
1855 *optname = ctx.optname;
1856
1857 /* optlen == 0 from BPF indicates that we should
1858 * use original userspace data.
1859 */
1860 if (ctx.optlen != 0) {
1861 *optlen = ctx.optlen;
1862 /* We've used bpf_sockopt_kern->buf as an intermediary
1863 * storage, but the BPF program indicates that we need
1864 * to pass this data to the kernel setsockopt handler.
1865 * No way to export on-stack buf, have to allocate a
1866 * new buffer.
1867 */
1868 if (!sockopt_buf_allocated(&ctx, &buf)) {
1869 void *p = kmalloc(ctx.optlen, GFP_USER);
1870
1871 if (!p) {
1872 ret = -ENOMEM;
1873 goto out;
1874 }
1875 memcpy(p, ctx.optval, ctx.optlen);
1876 *kernel_optval = p;
1877 } else {
1878 *kernel_optval = ctx.optval;
1879 }
1880 /* export and don't free sockopt buf */
1881 return 0;
1882 }
1883 }
1884
1885 out:
1886 sockopt_free_buf(&ctx, &buf);
1887 return ret;
1888 }
1889
1890 int __cgroup_bpf_run_filter_getsockopt(struct sock *sk, int level,
1891 int optname, sockptr_t optval,
1892 sockptr_t optlen, int max_optlen,
1893 int retval)
1894 {
1895 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1896 struct bpf_sockopt_buf buf = {};
1897 struct bpf_sockopt_kern ctx = {
1898 .sk = sk,
1899 .level = level,
1900 .optname = optname,
1901 .current_task = current,
1902 };
1903 int orig_optlen;
1904 int ret;
1905
1906 orig_optlen = max_optlen;
1907 ctx.optlen = max_optlen;
1908 max_optlen = sockopt_alloc_buf(&ctx, max_optlen, &buf);
1909 if (max_optlen < 0)
1910 return max_optlen;
1911
1912 if (!retval) {
1913 /* If kernel getsockopt finished successfully,
1914 * copy whatever was returned to the user back
1915 * into our temporary buffer. Set optlen to the
1916 * one that kernel returned as well to let
1917 * BPF programs inspect the value.
1918 */
1919 if (copy_from_sockptr(&ctx.optlen, optlen,
1920 sizeof(ctx.optlen))) {
1921 ret = -EFAULT;
1922 goto out;
1923 }
1924
1925 if (ctx.optlen < 0) {
1926 ret = -EFAULT;
1927 goto out;
1928 }
1929 orig_optlen = ctx.optlen;
1930
1931 if (copy_from_sockptr(ctx.optval, optval,
1932 min(ctx.optlen, max_optlen))) {
1933 ret = -EFAULT;
1934 goto out;
1935 }
1936 }
1937
1938 lock_sock(sk);
1939 ret = bpf_prog_run_array_cg(&cgrp->bpf, CGROUP_GETSOCKOPT,
1940 &ctx, bpf_prog_run, retval, NULL);
1941 release_sock(sk);
1942
1943 if (ret < 0)
1944 goto out;
1945
1946 if (!sockptr_is_null(optval) &&
1947 (ctx.optlen > max_optlen || ctx.optlen < 0)) {
1948 if (orig_optlen > PAGE_SIZE && ctx.optlen >= 0) {
1949 pr_info_once("bpf getsockopt: ignoring program buffer with optlen=%d (max_optlen=%d)\n",
1950 ctx.optlen, max_optlen);
1951 ret = retval;
1952 goto out;
1953 }
1954 ret = -EFAULT;
1955 goto out;
1956 }
1957
1958 if (ctx.optlen != 0) {
1959 if (!sockptr_is_null(optval) &&
1960 copy_to_sockptr(optval, ctx.optval, ctx.optlen)) {
1961 ret = -EFAULT;
1962 goto out;
1963 }
1964 if (copy_to_sockptr(optlen, &ctx.optlen, sizeof(ctx.optlen))) {
1965 ret = -EFAULT;
1966 goto out;
1967 }
1968 }
1969
1970 out:
1971 sockopt_free_buf(&ctx, &buf);
1972 return ret;
1973 }
1974
1975 int __cgroup_bpf_run_filter_getsockopt_kern(struct sock *sk, int level,
1976 int optname, void *optval,
1977 int *optlen, int retval)
1978 {
1979 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1980 struct bpf_sockopt_kern ctx = {
1981 .sk = sk,
1982 .level = level,
1983 .optname = optname,
1984 .optlen = *optlen,
1985 .optval = optval,
1986 .optval_end = optval + *optlen,
1987 .current_task = current,
1988 };
1989 int ret;
1990
1991 /* Note that __cgroup_bpf_run_filter_getsockopt doesn't copy
1992 * user data back into BPF buffer when reval != 0. This is
1993 * done as an optimization to avoid extra copy, assuming
1994 * kernel won't populate the data in case of an error.
1995 * Here we always pass the data and memset() should
1996 * be called if that data shouldn't be "exported".
1997 */
1998
1999 ret = bpf_prog_run_array_cg(&cgrp->bpf, CGROUP_GETSOCKOPT,
2000 &ctx, bpf_prog_run, retval, NULL);
2001 if (ret < 0)
2002 return ret;
2003
2004 if (ctx.optlen > *optlen)
2005 return -EFAULT;
2006
2007 /* BPF programs can shrink the buffer, export the modifications.
2008 */
2009 if (ctx.optlen != 0)
2010 *optlen = ctx.optlen;
2011
2012 return ret;
2013 }
2014 #endif
2015
2016 static ssize_t sysctl_cpy_dir(const struct ctl_dir *dir, char **bufp,
2017 size_t *lenp)
2018 {
2019 ssize_t tmp_ret = 0, ret;
2020
2021 if (dir->header.parent) {
2022 tmp_ret = sysctl_cpy_dir(dir->header.parent, bufp, lenp);
2023 if (tmp_ret < 0)
2024 return tmp_ret;
2025 }
2026
2027 ret = strscpy(*bufp, dir->header.ctl_table[0].procname, *lenp);
2028 if (ret < 0)
2029 return ret;
2030 *bufp += ret;
2031 *lenp -= ret;
2032 ret += tmp_ret;
2033
2034 /* Avoid leading slash. */
2035 if (!ret)
2036 return ret;
2037
2038 tmp_ret = strscpy(*bufp, "/", *lenp);
2039 if (tmp_ret < 0)
2040 return tmp_ret;
2041 *bufp += tmp_ret;
2042 *lenp -= tmp_ret;
2043
2044 return ret + tmp_ret;
2045 }
2046
2047 BPF_CALL_4(bpf_sysctl_get_name, struct bpf_sysctl_kern *, ctx, char *, buf,
2048 size_t, buf_len, u64, flags)
2049 {
2050 ssize_t tmp_ret = 0, ret;
2051
2052 if (!buf)
2053 return -EINVAL;
2054
2055 if (!(flags & BPF_F_SYSCTL_BASE_NAME)) {
2056 if (!ctx->head)
2057 return -EINVAL;
2058 tmp_ret = sysctl_cpy_dir(ctx->head->parent, &buf, &buf_len);
2059 if (tmp_ret < 0)
2060 return tmp_ret;
2061 }
2062
2063 ret = strscpy(buf, ctx->table->procname, buf_len);
2064
2065 return ret < 0 ? ret : tmp_ret + ret;
2066 }
2067
2068 static const struct bpf_func_proto bpf_sysctl_get_name_proto = {
2069 .func = bpf_sysctl_get_name,
2070 .gpl_only = false,
2071 .ret_type = RET_INTEGER,
2072 .arg1_type = ARG_PTR_TO_CTX,
2073 .arg2_type = ARG_PTR_TO_MEM,
2074 .arg3_type = ARG_CONST_SIZE,
2075 .arg4_type = ARG_ANYTHING,
2076 };
2077
2078 static int copy_sysctl_value(char *dst, size_t dst_len, char *src,
2079 size_t src_len)
2080 {
2081 if (!dst)
2082 return -EINVAL;
2083
2084 if (!dst_len)
2085 return -E2BIG;
2086
2087 if (!src || !src_len) {
2088 memset(dst, 0, dst_len);
2089 return -EINVAL;
2090 }
2091
2092 memcpy(dst, src, min(dst_len, src_len));
2093
2094 if (dst_len > src_len) {
2095 memset(dst + src_len, '\0', dst_len - src_len);
2096 return src_len;
2097 }
2098
2099 dst[dst_len - 1] = '\0';
2100
2101 return -E2BIG;
2102 }
2103
2104 BPF_CALL_3(bpf_sysctl_get_current_value, struct bpf_sysctl_kern *, ctx,
2105 char *, buf, size_t, buf_len)
2106 {
2107 return copy_sysctl_value(buf, buf_len, ctx->cur_val, ctx->cur_len);
2108 }
2109
2110 static const struct bpf_func_proto bpf_sysctl_get_current_value_proto = {
2111 .func = bpf_sysctl_get_current_value,
2112 .gpl_only = false,
2113 .ret_type = RET_INTEGER,
2114 .arg1_type = ARG_PTR_TO_CTX,
2115 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
2116 .arg3_type = ARG_CONST_SIZE,
2117 };
2118
2119 BPF_CALL_3(bpf_sysctl_get_new_value, struct bpf_sysctl_kern *, ctx, char *, buf,
2120 size_t, buf_len)
2121 {
2122 if (!ctx->write) {
2123 if (buf && buf_len)
2124 memset(buf, '\0', buf_len);
2125 return -EINVAL;
2126 }
2127 return copy_sysctl_value(buf, buf_len, ctx->new_val, ctx->new_len);
2128 }
2129
2130 static const struct bpf_func_proto bpf_sysctl_get_new_value_proto = {
2131 .func = bpf_sysctl_get_new_value,
2132 .gpl_only = false,
2133 .ret_type = RET_INTEGER,
2134 .arg1_type = ARG_PTR_TO_CTX,
2135 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
2136 .arg3_type = ARG_CONST_SIZE,
2137 };
2138
2139 BPF_CALL_3(bpf_sysctl_set_new_value, struct bpf_sysctl_kern *, ctx,
2140 const char *, buf, size_t, buf_len)
2141 {
2142 if (!ctx->write || !ctx->new_val || !ctx->new_len || !buf || !buf_len)
2143 return -EINVAL;
2144
2145 if (buf_len > PAGE_SIZE - 1)
2146 return -E2BIG;
2147
2148 memcpy(ctx->new_val, buf, buf_len);
2149 ctx->new_len = buf_len;
2150 ctx->new_updated = 1;
2151
2152 return 0;
2153 }
2154
2155 static const struct bpf_func_proto bpf_sysctl_set_new_value_proto = {
2156 .func = bpf_sysctl_set_new_value,
2157 .gpl_only = false,
2158 .ret_type = RET_INTEGER,
2159 .arg1_type = ARG_PTR_TO_CTX,
2160 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
2161 .arg3_type = ARG_CONST_SIZE,
2162 };
2163
2164 static const struct bpf_func_proto *
2165 sysctl_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
2166 {
2167 const struct bpf_func_proto *func_proto;
2168
2169 func_proto = cgroup_common_func_proto(func_id, prog);
2170 if (func_proto)
2171 return func_proto;
2172
2173 func_proto = cgroup_current_func_proto(func_id, prog);
2174 if (func_proto)
2175 return func_proto;
2176
2177 switch (func_id) {
2178 case BPF_FUNC_sysctl_get_name:
2179 return &bpf_sysctl_get_name_proto;
2180 case BPF_FUNC_sysctl_get_current_value:
2181 return &bpf_sysctl_get_current_value_proto;
2182 case BPF_FUNC_sysctl_get_new_value:
2183 return &bpf_sysctl_get_new_value_proto;
2184 case BPF_FUNC_sysctl_set_new_value:
2185 return &bpf_sysctl_set_new_value_proto;
2186 case BPF_FUNC_ktime_get_coarse_ns:
2187 return &bpf_ktime_get_coarse_ns_proto;
2188 case BPF_FUNC_perf_event_output:
2189 return &bpf_event_output_data_proto;
2190 default:
2191 return bpf_base_func_proto(func_id, prog);
2192 }
2193 }
2194
2195 static bool sysctl_is_valid_access(int off, int size, enum bpf_access_type type,
2196 const struct bpf_prog *prog,
2197 struct bpf_insn_access_aux *info)
2198 {
2199 const int size_default = sizeof(__u32);
2200
2201 if (off < 0 || off + size > sizeof(struct bpf_sysctl) || off % size)
2202 return false;
2203
2204 switch (off) {
2205 case bpf_ctx_range(struct bpf_sysctl, write):
2206 if (type != BPF_READ)
2207 return false;
2208 bpf_ctx_record_field_size(info, size_default);
2209 return bpf_ctx_narrow_access_ok(off, size, size_default);
2210 case bpf_ctx_range(struct bpf_sysctl, file_pos):
2211 if (type == BPF_READ) {
2212 bpf_ctx_record_field_size(info, size_default);
2213 return bpf_ctx_narrow_access_ok(off, size, size_default);
2214 } else {
2215 return size == size_default;
2216 }
2217 default:
2218 return false;
2219 }
2220 }
2221
2222 static u32 sysctl_convert_ctx_access(enum bpf_access_type type,
2223 const struct bpf_insn *si,
2224 struct bpf_insn *insn_buf,
2225 struct bpf_prog *prog, u32 *target_size)
2226 {
2227 struct bpf_insn *insn = insn_buf;
2228 u32 read_size;
2229
2230 switch (si->off) {
2231 case offsetof(struct bpf_sysctl, write):
2232 *insn++ = BPF_LDX_MEM(
2233 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
2234 bpf_target_off(struct bpf_sysctl_kern, write,
2235 sizeof_field(struct bpf_sysctl_kern,
2236 write),
2237 target_size));
2238 break;
2239 case offsetof(struct bpf_sysctl, file_pos):
2240 /* ppos is a pointer so it should be accessed via indirect
2241 * loads and stores. Also for stores additional temporary
2242 * register is used since neither src_reg nor dst_reg can be
2243 * overridden.
2244 */
2245 if (type == BPF_WRITE) {
2246 int treg = BPF_REG_9;
2247
2248 if (si->src_reg == treg || si->dst_reg == treg)
2249 --treg;
2250 if (si->src_reg == treg || si->dst_reg == treg)
2251 --treg;
2252 *insn++ = BPF_STX_MEM(
2253 BPF_DW, si->dst_reg, treg,
2254 offsetof(struct bpf_sysctl_kern, tmp_reg));
2255 *insn++ = BPF_LDX_MEM(
2256 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos),
2257 treg, si->dst_reg,
2258 offsetof(struct bpf_sysctl_kern, ppos));
2259 *insn++ = BPF_RAW_INSN(
2260 BPF_CLASS(si->code) | BPF_MEM | BPF_SIZEOF(u32),
2261 treg, si->src_reg,
2262 bpf_ctx_narrow_access_offset(
2263 0, sizeof(u32), sizeof(loff_t)),
2264 si->imm);
2265 *insn++ = BPF_LDX_MEM(
2266 BPF_DW, treg, si->dst_reg,
2267 offsetof(struct bpf_sysctl_kern, tmp_reg));
2268 } else {
2269 *insn++ = BPF_LDX_MEM(
2270 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos),
2271 si->dst_reg, si->src_reg,
2272 offsetof(struct bpf_sysctl_kern, ppos));
2273 read_size = bpf_size_to_bytes(BPF_SIZE(si->code));
2274 *insn++ = BPF_LDX_MEM(
2275 BPF_SIZE(si->code), si->dst_reg, si->dst_reg,
2276 bpf_ctx_narrow_access_offset(
2277 0, read_size, sizeof(loff_t)));
2278 }
2279 *target_size = sizeof(u32);
2280 break;
2281 }
2282
2283 return insn - insn_buf;
2284 }
2285
2286 const struct bpf_verifier_ops cg_sysctl_verifier_ops = {
2287 .get_func_proto = sysctl_func_proto,
2288 .is_valid_access = sysctl_is_valid_access,
2289 .convert_ctx_access = sysctl_convert_ctx_access,
2290 };
2291
2292 const struct bpf_prog_ops cg_sysctl_prog_ops = {
2293 };
2294
2295 #ifdef CONFIG_NET
2296 BPF_CALL_1(bpf_get_netns_cookie_sockopt, struct bpf_sockopt_kern *, ctx)
2297 {
2298 const struct net *net = ctx ? sock_net(ctx->sk) : &init_net;
2299
2300 return net->net_cookie;
2301 }
2302
2303 static const struct bpf_func_proto bpf_get_netns_cookie_sockopt_proto = {
2304 .func = bpf_get_netns_cookie_sockopt,
2305 .gpl_only = false,
2306 .ret_type = RET_INTEGER,
2307 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
2308 };
2309 #endif
2310
2311 static const struct bpf_func_proto *
2312 cg_sockopt_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
2313 {
2314 const struct bpf_func_proto *func_proto;
2315
2316 func_proto = cgroup_common_func_proto(func_id, prog);
2317 if (func_proto)
2318 return func_proto;
2319
2320 func_proto = cgroup_current_func_proto(func_id, prog);
2321 if (func_proto)
2322 return func_proto;
2323
2324 switch (func_id) {
2325 #ifdef CONFIG_NET
2326 case BPF_FUNC_get_netns_cookie:
2327 return &bpf_get_netns_cookie_sockopt_proto;
2328 case BPF_FUNC_sk_storage_get:
2329 return &bpf_sk_storage_get_proto;
2330 case BPF_FUNC_sk_storage_delete:
2331 return &bpf_sk_storage_delete_proto;
2332 case BPF_FUNC_setsockopt:
2333 if (prog->expected_attach_type == BPF_CGROUP_SETSOCKOPT)
2334 return &bpf_sk_setsockopt_proto;
2335 return NULL;
2336 case BPF_FUNC_getsockopt:
2337 if (prog->expected_attach_type == BPF_CGROUP_SETSOCKOPT)
2338 return &bpf_sk_getsockopt_proto;
2339 return NULL;
2340 #endif
2341 #ifdef CONFIG_INET
2342 case BPF_FUNC_tcp_sock:
2343 return &bpf_tcp_sock_proto;
2344 #endif
2345 case BPF_FUNC_perf_event_output:
2346 return &bpf_event_output_data_proto;
2347 default:
2348 return bpf_base_func_proto(func_id, prog);
2349 }
2350 }
2351
2352 static bool cg_sockopt_is_valid_access(int off, int size,
2353 enum bpf_access_type type,
2354 const struct bpf_prog *prog,
2355 struct bpf_insn_access_aux *info)
2356 {
2357 const int size_default = sizeof(__u32);
2358
2359 if (off < 0 || off >= sizeof(struct bpf_sockopt))
2360 return false;
2361
2362 if (off % size != 0)
2363 return false;
2364
2365 if (type == BPF_WRITE) {
2366 switch (off) {
2367 case offsetof(struct bpf_sockopt, retval):
2368 if (size != size_default)
2369 return false;
2370 return prog->expected_attach_type ==
2371 BPF_CGROUP_GETSOCKOPT;
2372 case offsetof(struct bpf_sockopt, optname):
2373 fallthrough;
2374 case offsetof(struct bpf_sockopt, level):
2375 if (size != size_default)
2376 return false;
2377 return prog->expected_attach_type ==
2378 BPF_CGROUP_SETSOCKOPT;
2379 case offsetof(struct bpf_sockopt, optlen):
2380 return size == size_default;
2381 default:
2382 return false;
2383 }
2384 }
2385
2386 switch (off) {
2387 case offsetof(struct bpf_sockopt, sk):
2388 if (size != sizeof(__u64))
2389 return false;
2390 info->reg_type = PTR_TO_SOCKET;
2391 break;
2392 case offsetof(struct bpf_sockopt, optval):
2393 if (size != sizeof(__u64))
2394 return false;
2395 info->reg_type = PTR_TO_PACKET;
2396 break;
2397 case offsetof(struct bpf_sockopt, optval_end):
2398 if (size != sizeof(__u64))
2399 return false;
2400 info->reg_type = PTR_TO_PACKET_END;
2401 break;
2402 case offsetof(struct bpf_sockopt, retval):
2403 if (size != size_default)
2404 return false;
2405 return prog->expected_attach_type == BPF_CGROUP_GETSOCKOPT;
2406 default:
2407 if (size != size_default)
2408 return false;
2409 break;
2410 }
2411 return true;
2412 }
2413
2414 #define CG_SOCKOPT_READ_FIELD(F) \
2415 BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, F), \
2416 si->dst_reg, si->src_reg, \
2417 offsetof(struct bpf_sockopt_kern, F))
2418
2419 #define CG_SOCKOPT_WRITE_FIELD(F) \
2420 BPF_RAW_INSN((BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, F) | \
2421 BPF_MEM | BPF_CLASS(si->code)), \
2422 si->dst_reg, si->src_reg, \
2423 offsetof(struct bpf_sockopt_kern, F), \
2424 si->imm)
2425
2426 static u32 cg_sockopt_convert_ctx_access(enum bpf_access_type type,
2427 const struct bpf_insn *si,
2428 struct bpf_insn *insn_buf,
2429 struct bpf_prog *prog,
2430 u32 *target_size)
2431 {
2432 struct bpf_insn *insn = insn_buf;
2433
2434 switch (si->off) {
2435 case offsetof(struct bpf_sockopt, sk):
2436 *insn++ = CG_SOCKOPT_READ_FIELD(sk);
2437 break;
2438 case offsetof(struct bpf_sockopt, level):
2439 if (type == BPF_WRITE)
2440 *insn++ = CG_SOCKOPT_WRITE_FIELD(level);
2441 else
2442 *insn++ = CG_SOCKOPT_READ_FIELD(level);
2443 break;
2444 case offsetof(struct bpf_sockopt, optname):
2445 if (type == BPF_WRITE)
2446 *insn++ = CG_SOCKOPT_WRITE_FIELD(optname);
2447 else
2448 *insn++ = CG_SOCKOPT_READ_FIELD(optname);
2449 break;
2450 case offsetof(struct bpf_sockopt, optlen):
2451 if (type == BPF_WRITE)
2452 *insn++ = CG_SOCKOPT_WRITE_FIELD(optlen);
2453 else
2454 *insn++ = CG_SOCKOPT_READ_FIELD(optlen);
2455 break;
2456 case offsetof(struct bpf_sockopt, retval):
2457 BUILD_BUG_ON(offsetof(struct bpf_cg_run_ctx, run_ctx) != 0);
2458
2459 if (type == BPF_WRITE) {
2460 int treg = BPF_REG_9;
2461
2462 if (si->src_reg == treg || si->dst_reg == treg)
2463 --treg;
2464 if (si->src_reg == treg || si->dst_reg == treg)
2465 --treg;
2466 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, treg,
2467 offsetof(struct bpf_sockopt_kern, tmp_reg));
2468 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, current_task),
2469 treg, si->dst_reg,
2470 offsetof(struct bpf_sockopt_kern, current_task));
2471 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct task_struct, bpf_ctx),
2472 treg, treg,
2473 offsetof(struct task_struct, bpf_ctx));
2474 *insn++ = BPF_RAW_INSN(BPF_CLASS(si->code) | BPF_MEM |
2475 BPF_FIELD_SIZEOF(struct bpf_cg_run_ctx, retval),
2476 treg, si->src_reg,
2477 offsetof(struct bpf_cg_run_ctx, retval),
2478 si->imm);
2479 *insn++ = BPF_LDX_MEM(BPF_DW, treg, si->dst_reg,
2480 offsetof(struct bpf_sockopt_kern, tmp_reg));
2481 } else {
2482 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, current_task),
2483 si->dst_reg, si->src_reg,
2484 offsetof(struct bpf_sockopt_kern, current_task));
2485 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct task_struct, bpf_ctx),
2486 si->dst_reg, si->dst_reg,
2487 offsetof(struct task_struct, bpf_ctx));
2488 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_cg_run_ctx, retval),
2489 si->dst_reg, si->dst_reg,
2490 offsetof(struct bpf_cg_run_ctx, retval));
2491 }
2492 break;
2493 case offsetof(struct bpf_sockopt, optval):
2494 *insn++ = CG_SOCKOPT_READ_FIELD(optval);
2495 break;
2496 case offsetof(struct bpf_sockopt, optval_end):
2497 *insn++ = CG_SOCKOPT_READ_FIELD(optval_end);
2498 break;
2499 }
2500
2501 return insn - insn_buf;
2502 }
2503
2504 static int cg_sockopt_get_prologue(struct bpf_insn *insn_buf,
2505 bool direct_write,
2506 const struct bpf_prog *prog)
2507 {
2508 /* Nothing to do for sockopt argument. The data is kzalloc'ated.
2509 */
2510 return 0;
2511 }
2512
2513 const struct bpf_verifier_ops cg_sockopt_verifier_ops = {
2514 .get_func_proto = cg_sockopt_func_proto,
2515 .is_valid_access = cg_sockopt_is_valid_access,
2516 .convert_ctx_access = cg_sockopt_convert_ctx_access,
2517 .gen_prologue = cg_sockopt_get_prologue,
2518 };
2519
2520 const struct bpf_prog_ops cg_sockopt_prog_ops = {
2521 };
2522
2523 /* Common helpers for cgroup hooks. */
2524 const struct bpf_func_proto *
2525 cgroup_common_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
2526 {
2527 switch (func_id) {
2528 case BPF_FUNC_get_local_storage:
2529 return &bpf_get_local_storage_proto;
2530 case BPF_FUNC_get_retval:
2531 switch (prog->expected_attach_type) {
2532 case BPF_CGROUP_INET_INGRESS:
2533 case BPF_CGROUP_INET_EGRESS:
2534 case BPF_CGROUP_SOCK_OPS:
2535 case BPF_CGROUP_UDP4_RECVMSG:
2536 case BPF_CGROUP_UDP6_RECVMSG:
2537 case BPF_CGROUP_UNIX_RECVMSG:
2538 case BPF_CGROUP_INET4_GETPEERNAME:
2539 case BPF_CGROUP_INET6_GETPEERNAME:
2540 case BPF_CGROUP_UNIX_GETPEERNAME:
2541 case BPF_CGROUP_INET4_GETSOCKNAME:
2542 case BPF_CGROUP_INET6_GETSOCKNAME:
2543 case BPF_CGROUP_UNIX_GETSOCKNAME:
2544 return NULL;
2545 default:
2546 return &bpf_get_retval_proto;
2547 }
2548 case BPF_FUNC_set_retval:
2549 switch (prog->expected_attach_type) {
2550 case BPF_CGROUP_INET_INGRESS:
2551 case BPF_CGROUP_INET_EGRESS:
2552 case BPF_CGROUP_SOCK_OPS:
2553 case BPF_CGROUP_UDP4_RECVMSG:
2554 case BPF_CGROUP_UDP6_RECVMSG:
2555 case BPF_CGROUP_UNIX_RECVMSG:
2556 case BPF_CGROUP_INET4_GETPEERNAME:
2557 case BPF_CGROUP_INET6_GETPEERNAME:
2558 case BPF_CGROUP_UNIX_GETPEERNAME:
2559 case BPF_CGROUP_INET4_GETSOCKNAME:
2560 case BPF_CGROUP_INET6_GETSOCKNAME:
2561 case BPF_CGROUP_UNIX_GETSOCKNAME:
2562 return NULL;
2563 default:
2564 return &bpf_set_retval_proto;
2565 }
2566 default:
2567 return NULL;
2568 }
2569 }
2570
2571 /* Common helpers for cgroup hooks with valid process context. */
2572 const struct bpf_func_proto *
2573 cgroup_current_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
2574 {
2575 switch (func_id) {
2576 case BPF_FUNC_get_current_uid_gid:
2577 return &bpf_get_current_uid_gid_proto;
2578 case BPF_FUNC_get_current_comm:
2579 return &bpf_get_current_comm_proto;
2580 #ifdef CONFIG_CGROUP_NET_CLASSID
2581 case BPF_FUNC_get_cgroup_classid:
2582 return &bpf_get_cgroup_classid_curr_proto;
2583 #endif
2584 default:
2585 return NULL;
2586 }
2587 }
2588
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.