Loading...
Note: File does not exist in v3.15.
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Functions to manage eBPF programs attached to cgroups
4 *
5 * Copyright (c) 2016 Daniel Mack
6 */
7
8#include <linux/kernel.h>
9#include <linux/atomic.h>
10#include <linux/cgroup.h>
11#include <linux/filter.h>
12#include <linux/slab.h>
13#include <linux/sysctl.h>
14#include <linux/string.h>
15#include <linux/bpf.h>
16#include <linux/bpf-cgroup.h>
17#include <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
24DEFINE_STATIC_KEY_ARRAY_FALSE(cgroup_bpf_enabled_key, MAX_CGROUP_BPF_ATTACH_TYPE);
25EXPORT_SYMBOL(cgroup_bpf_enabled_key);
26
27/* __always_inline is necessary to prevent indirect call through run_prog
28 * function pointer.
29 */
30static __always_inline int
31bpf_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
66unsigned 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
88unsigned 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
110unsigned 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
130struct cgroup_lsm_atype {
131 u32 attach_btf_id;
132 int refcnt;
133};
134
135static struct cgroup_lsm_atype cgroup_lsm_atype[CGROUP_LSM_NUM];
136
137static enum cgroup_bpf_attach_type
138bpf_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
159void 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
172void bpf_cgroup_atype_put(int cgroup_atype)
173{
174 int i = cgroup_atype - CGROUP_LSM_START;
175
176 mutex_lock(&cgroup_mutex);
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 mutex_unlock(&cgroup_mutex);
181}
182#else
183static enum cgroup_bpf_attach_type
184bpf_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
192void cgroup_bpf_offline(struct cgroup *cgrp)
193{
194 cgroup_get(cgrp);
195 percpu_ref_kill(&cgrp->bpf.refcnt);
196}
197
198static 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
206static 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
240static 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
249static 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 */
264static 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 */
275static 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 mutex_lock(&cgroup_mutex);
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 mutex_unlock(&cgroup_mutex);
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 */
332static 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 */
343static 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 */
355static 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 */
372static 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 */
401static 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
445static 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 */
461int 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;
492cleanup:
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
504static 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
540cleanup:
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
556static 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 */
607static 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
707cleanup_trampoline:
708 if (type == BPF_LSM_CGROUP)
709 bpf_trampoline_unlink_cgroup_shim(new_prog);
710
711cleanup:
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
724static 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 mutex_lock(&cgroup_mutex);
733 ret = __cgroup_bpf_attach(cgrp, prog, replace_prog, link, type, flags);
734 mutex_unlock(&cgroup_mutex);
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 */
741static 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 }
773found:
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 * @type: Type of attach operation
789 *
790 * Must be called with cgroup_mutex held.
791 */
792static int __cgroup_bpf_replace(struct cgroup *cgrp,
793 struct bpf_cgroup_link *link,
794 struct bpf_prog *new_prog)
795{
796 enum cgroup_bpf_attach_type atype;
797 struct bpf_prog *old_prog;
798 struct bpf_prog_list *pl;
799 struct hlist_head *progs;
800 bool found = false;
801
802 atype = bpf_cgroup_atype_find(link->type, new_prog->aux->attach_btf_id);
803 if (atype < 0)
804 return -EINVAL;
805
806 progs = &cgrp->bpf.progs[atype];
807
808 if (link->link.prog->type != new_prog->type)
809 return -EINVAL;
810
811 hlist_for_each_entry(pl, progs, node) {
812 if (pl->link == link) {
813 found = true;
814 break;
815 }
816 }
817 if (!found)
818 return -ENOENT;
819
820 old_prog = xchg(&link->link.prog, new_prog);
821 replace_effective_prog(cgrp, atype, link);
822 bpf_prog_put(old_prog);
823 return 0;
824}
825
826static int cgroup_bpf_replace(struct bpf_link *link, struct bpf_prog *new_prog,
827 struct bpf_prog *old_prog)
828{
829 struct bpf_cgroup_link *cg_link;
830 int ret;
831
832 cg_link = container_of(link, struct bpf_cgroup_link, link);
833
834 mutex_lock(&cgroup_mutex);
835 /* link might have been auto-released by dying cgroup, so fail */
836 if (!cg_link->cgroup) {
837 ret = -ENOLINK;
838 goto out_unlock;
839 }
840 if (old_prog && link->prog != old_prog) {
841 ret = -EPERM;
842 goto out_unlock;
843 }
844 ret = __cgroup_bpf_replace(cg_link->cgroup, cg_link, new_prog);
845out_unlock:
846 mutex_unlock(&cgroup_mutex);
847 return ret;
848}
849
850static struct bpf_prog_list *find_detach_entry(struct hlist_head *progs,
851 struct bpf_prog *prog,
852 struct bpf_cgroup_link *link,
853 bool allow_multi)
854{
855 struct bpf_prog_list *pl;
856
857 if (!allow_multi) {
858 if (hlist_empty(progs))
859 /* report error when trying to detach and nothing is attached */
860 return ERR_PTR(-ENOENT);
861
862 /* to maintain backward compatibility NONE and OVERRIDE cgroups
863 * allow detaching with invalid FD (prog==NULL) in legacy mode
864 */
865 return hlist_entry(progs->first, typeof(*pl), node);
866 }
867
868 if (!prog && !link)
869 /* to detach MULTI prog the user has to specify valid FD
870 * of the program or link to be detached
871 */
872 return ERR_PTR(-EINVAL);
873
874 /* find the prog or link and detach it */
875 hlist_for_each_entry(pl, progs, node) {
876 if (pl->prog == prog && pl->link == link)
877 return pl;
878 }
879 return ERR_PTR(-ENOENT);
880}
881
882/**
883 * purge_effective_progs() - After compute_effective_progs fails to alloc new
884 * cgrp->bpf.inactive table we can recover by
885 * recomputing the array in place.
886 *
887 * @cgrp: The cgroup which descendants to travers
888 * @prog: A program to detach or NULL
889 * @link: A link to detach or NULL
890 * @atype: Type of detach operation
891 */
892static void purge_effective_progs(struct cgroup *cgrp, struct bpf_prog *prog,
893 struct bpf_cgroup_link *link,
894 enum cgroup_bpf_attach_type atype)
895{
896 struct cgroup_subsys_state *css;
897 struct bpf_prog_array *progs;
898 struct bpf_prog_list *pl;
899 struct hlist_head *head;
900 struct cgroup *cg;
901 int pos;
902
903 /* recompute effective prog array in place */
904 css_for_each_descendant_pre(css, &cgrp->self) {
905 struct cgroup *desc = container_of(css, struct cgroup, self);
906
907 if (percpu_ref_is_zero(&desc->bpf.refcnt))
908 continue;
909
910 /* find position of link or prog in effective progs array */
911 for (pos = 0, cg = desc; cg; cg = cgroup_parent(cg)) {
912 if (pos && !(cg->bpf.flags[atype] & BPF_F_ALLOW_MULTI))
913 continue;
914
915 head = &cg->bpf.progs[atype];
916 hlist_for_each_entry(pl, head, node) {
917 if (!prog_list_prog(pl))
918 continue;
919 if (pl->prog == prog && pl->link == link)
920 goto found;
921 pos++;
922 }
923 }
924
925 /* no link or prog match, skip the cgroup of this layer */
926 continue;
927found:
928 progs = rcu_dereference_protected(
929 desc->bpf.effective[atype],
930 lockdep_is_held(&cgroup_mutex));
931
932 /* Remove the program from the array */
933 WARN_ONCE(bpf_prog_array_delete_safe_at(progs, pos),
934 "Failed to purge a prog from array at index %d", pos);
935 }
936}
937
938/**
939 * __cgroup_bpf_detach() - Detach the program or link from a cgroup, and
940 * propagate the change to descendants
941 * @cgrp: The cgroup which descendants to traverse
942 * @prog: A program to detach or NULL
943 * @link: A link to detach or NULL
944 * @type: Type of detach operation
945 *
946 * At most one of @prog or @link can be non-NULL.
947 * Must be called with cgroup_mutex held.
948 */
949static int __cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
950 struct bpf_cgroup_link *link, enum bpf_attach_type type)
951{
952 enum cgroup_bpf_attach_type atype;
953 struct bpf_prog *old_prog;
954 struct bpf_prog_list *pl;
955 struct hlist_head *progs;
956 u32 attach_btf_id = 0;
957 u32 flags;
958
959 if (prog)
960 attach_btf_id = prog->aux->attach_btf_id;
961 if (link)
962 attach_btf_id = link->link.prog->aux->attach_btf_id;
963
964 atype = bpf_cgroup_atype_find(type, attach_btf_id);
965 if (atype < 0)
966 return -EINVAL;
967
968 progs = &cgrp->bpf.progs[atype];
969 flags = cgrp->bpf.flags[atype];
970
971 if (prog && link)
972 /* only one of prog or link can be specified */
973 return -EINVAL;
974
975 pl = find_detach_entry(progs, prog, link, flags & BPF_F_ALLOW_MULTI);
976 if (IS_ERR(pl))
977 return PTR_ERR(pl);
978
979 /* mark it deleted, so it's ignored while recomputing effective */
980 old_prog = pl->prog;
981 pl->prog = NULL;
982 pl->link = NULL;
983
984 if (update_effective_progs(cgrp, atype)) {
985 /* if update effective array failed replace the prog with a dummy prog*/
986 pl->prog = old_prog;
987 pl->link = link;
988 purge_effective_progs(cgrp, old_prog, link, atype);
989 }
990
991 /* now can actually delete it from this cgroup list */
992 hlist_del(&pl->node);
993
994 kfree(pl);
995 if (hlist_empty(progs))
996 /* last program was detached, reset flags to zero */
997 cgrp->bpf.flags[atype] = 0;
998 if (old_prog) {
999 if (type == BPF_LSM_CGROUP)
1000 bpf_trampoline_unlink_cgroup_shim(old_prog);
1001 bpf_prog_put(old_prog);
1002 }
1003 static_branch_dec(&cgroup_bpf_enabled_key[atype]);
1004 return 0;
1005}
1006
1007static int cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
1008 enum bpf_attach_type type)
1009{
1010 int ret;
1011
1012 mutex_lock(&cgroup_mutex);
1013 ret = __cgroup_bpf_detach(cgrp, prog, NULL, type);
1014 mutex_unlock(&cgroup_mutex);
1015 return ret;
1016}
1017
1018/* Must be called with cgroup_mutex held to avoid races. */
1019static int __cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
1020 union bpf_attr __user *uattr)
1021{
1022 __u32 __user *prog_attach_flags = u64_to_user_ptr(attr->query.prog_attach_flags);
1023 bool effective_query = attr->query.query_flags & BPF_F_QUERY_EFFECTIVE;
1024 __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
1025 enum bpf_attach_type type = attr->query.attach_type;
1026 enum cgroup_bpf_attach_type from_atype, to_atype;
1027 enum cgroup_bpf_attach_type atype;
1028 struct bpf_prog_array *effective;
1029 int cnt, ret = 0, i;
1030 int total_cnt = 0;
1031 u32 flags;
1032
1033 if (effective_query && prog_attach_flags)
1034 return -EINVAL;
1035
1036 if (type == BPF_LSM_CGROUP) {
1037 if (!effective_query && attr->query.prog_cnt &&
1038 prog_ids && !prog_attach_flags)
1039 return -EINVAL;
1040
1041 from_atype = CGROUP_LSM_START;
1042 to_atype = CGROUP_LSM_END;
1043 flags = 0;
1044 } else {
1045 from_atype = to_cgroup_bpf_attach_type(type);
1046 if (from_atype < 0)
1047 return -EINVAL;
1048 to_atype = from_atype;
1049 flags = cgrp->bpf.flags[from_atype];
1050 }
1051
1052 for (atype = from_atype; atype <= to_atype; atype++) {
1053 if (effective_query) {
1054 effective = rcu_dereference_protected(cgrp->bpf.effective[atype],
1055 lockdep_is_held(&cgroup_mutex));
1056 total_cnt += bpf_prog_array_length(effective);
1057 } else {
1058 total_cnt += prog_list_length(&cgrp->bpf.progs[atype]);
1059 }
1060 }
1061
1062 /* always output uattr->query.attach_flags as 0 during effective query */
1063 flags = effective_query ? 0 : flags;
1064 if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)))
1065 return -EFAULT;
1066 if (copy_to_user(&uattr->query.prog_cnt, &total_cnt, sizeof(total_cnt)))
1067 return -EFAULT;
1068 if (attr->query.prog_cnt == 0 || !prog_ids || !total_cnt)
1069 /* return early if user requested only program count + flags */
1070 return 0;
1071
1072 if (attr->query.prog_cnt < total_cnt) {
1073 total_cnt = attr->query.prog_cnt;
1074 ret = -ENOSPC;
1075 }
1076
1077 for (atype = from_atype; atype <= to_atype && total_cnt; atype++) {
1078 if (effective_query) {
1079 effective = rcu_dereference_protected(cgrp->bpf.effective[atype],
1080 lockdep_is_held(&cgroup_mutex));
1081 cnt = min_t(int, bpf_prog_array_length(effective), total_cnt);
1082 ret = bpf_prog_array_copy_to_user(effective, prog_ids, cnt);
1083 } else {
1084 struct hlist_head *progs;
1085 struct bpf_prog_list *pl;
1086 struct bpf_prog *prog;
1087 u32 id;
1088
1089 progs = &cgrp->bpf.progs[atype];
1090 cnt = min_t(int, prog_list_length(progs), total_cnt);
1091 i = 0;
1092 hlist_for_each_entry(pl, progs, node) {
1093 prog = prog_list_prog(pl);
1094 id = prog->aux->id;
1095 if (copy_to_user(prog_ids + i, &id, sizeof(id)))
1096 return -EFAULT;
1097 if (++i == cnt)
1098 break;
1099 }
1100
1101 if (prog_attach_flags) {
1102 flags = cgrp->bpf.flags[atype];
1103
1104 for (i = 0; i < cnt; i++)
1105 if (copy_to_user(prog_attach_flags + i,
1106 &flags, sizeof(flags)))
1107 return -EFAULT;
1108 prog_attach_flags += cnt;
1109 }
1110 }
1111
1112 prog_ids += cnt;
1113 total_cnt -= cnt;
1114 }
1115 return ret;
1116}
1117
1118static int cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
1119 union bpf_attr __user *uattr)
1120{
1121 int ret;
1122
1123 mutex_lock(&cgroup_mutex);
1124 ret = __cgroup_bpf_query(cgrp, attr, uattr);
1125 mutex_unlock(&cgroup_mutex);
1126 return ret;
1127}
1128
1129int cgroup_bpf_prog_attach(const union bpf_attr *attr,
1130 enum bpf_prog_type ptype, struct bpf_prog *prog)
1131{
1132 struct bpf_prog *replace_prog = NULL;
1133 struct cgroup *cgrp;
1134 int ret;
1135
1136 cgrp = cgroup_get_from_fd(attr->target_fd);
1137 if (IS_ERR(cgrp))
1138 return PTR_ERR(cgrp);
1139
1140 if ((attr->attach_flags & BPF_F_ALLOW_MULTI) &&
1141 (attr->attach_flags & BPF_F_REPLACE)) {
1142 replace_prog = bpf_prog_get_type(attr->replace_bpf_fd, ptype);
1143 if (IS_ERR(replace_prog)) {
1144 cgroup_put(cgrp);
1145 return PTR_ERR(replace_prog);
1146 }
1147 }
1148
1149 ret = cgroup_bpf_attach(cgrp, prog, replace_prog, NULL,
1150 attr->attach_type, attr->attach_flags);
1151
1152 if (replace_prog)
1153 bpf_prog_put(replace_prog);
1154 cgroup_put(cgrp);
1155 return ret;
1156}
1157
1158int cgroup_bpf_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
1159{
1160 struct bpf_prog *prog;
1161 struct cgroup *cgrp;
1162 int ret;
1163
1164 cgrp = cgroup_get_from_fd(attr->target_fd);
1165 if (IS_ERR(cgrp))
1166 return PTR_ERR(cgrp);
1167
1168 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1169 if (IS_ERR(prog))
1170 prog = NULL;
1171
1172 ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type);
1173 if (prog)
1174 bpf_prog_put(prog);
1175
1176 cgroup_put(cgrp);
1177 return ret;
1178}
1179
1180static void bpf_cgroup_link_release(struct bpf_link *link)
1181{
1182 struct bpf_cgroup_link *cg_link =
1183 container_of(link, struct bpf_cgroup_link, link);
1184 struct cgroup *cg;
1185
1186 /* link might have been auto-detached by dying cgroup already,
1187 * in that case our work is done here
1188 */
1189 if (!cg_link->cgroup)
1190 return;
1191
1192 mutex_lock(&cgroup_mutex);
1193
1194 /* re-check cgroup under lock again */
1195 if (!cg_link->cgroup) {
1196 mutex_unlock(&cgroup_mutex);
1197 return;
1198 }
1199
1200 WARN_ON(__cgroup_bpf_detach(cg_link->cgroup, NULL, cg_link,
1201 cg_link->type));
1202 if (cg_link->type == BPF_LSM_CGROUP)
1203 bpf_trampoline_unlink_cgroup_shim(cg_link->link.prog);
1204
1205 cg = cg_link->cgroup;
1206 cg_link->cgroup = NULL;
1207
1208 mutex_unlock(&cgroup_mutex);
1209
1210 cgroup_put(cg);
1211}
1212
1213static void bpf_cgroup_link_dealloc(struct bpf_link *link)
1214{
1215 struct bpf_cgroup_link *cg_link =
1216 container_of(link, struct bpf_cgroup_link, link);
1217
1218 kfree(cg_link);
1219}
1220
1221static int bpf_cgroup_link_detach(struct bpf_link *link)
1222{
1223 bpf_cgroup_link_release(link);
1224
1225 return 0;
1226}
1227
1228static void bpf_cgroup_link_show_fdinfo(const struct bpf_link *link,
1229 struct seq_file *seq)
1230{
1231 struct bpf_cgroup_link *cg_link =
1232 container_of(link, struct bpf_cgroup_link, link);
1233 u64 cg_id = 0;
1234
1235 mutex_lock(&cgroup_mutex);
1236 if (cg_link->cgroup)
1237 cg_id = cgroup_id(cg_link->cgroup);
1238 mutex_unlock(&cgroup_mutex);
1239
1240 seq_printf(seq,
1241 "cgroup_id:\t%llu\n"
1242 "attach_type:\t%d\n",
1243 cg_id,
1244 cg_link->type);
1245}
1246
1247static int bpf_cgroup_link_fill_link_info(const struct bpf_link *link,
1248 struct bpf_link_info *info)
1249{
1250 struct bpf_cgroup_link *cg_link =
1251 container_of(link, struct bpf_cgroup_link, link);
1252 u64 cg_id = 0;
1253
1254 mutex_lock(&cgroup_mutex);
1255 if (cg_link->cgroup)
1256 cg_id = cgroup_id(cg_link->cgroup);
1257 mutex_unlock(&cgroup_mutex);
1258
1259 info->cgroup.cgroup_id = cg_id;
1260 info->cgroup.attach_type = cg_link->type;
1261 return 0;
1262}
1263
1264static const struct bpf_link_ops bpf_cgroup_link_lops = {
1265 .release = bpf_cgroup_link_release,
1266 .dealloc = bpf_cgroup_link_dealloc,
1267 .detach = bpf_cgroup_link_detach,
1268 .update_prog = cgroup_bpf_replace,
1269 .show_fdinfo = bpf_cgroup_link_show_fdinfo,
1270 .fill_link_info = bpf_cgroup_link_fill_link_info,
1271};
1272
1273int cgroup_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
1274{
1275 struct bpf_link_primer link_primer;
1276 struct bpf_cgroup_link *link;
1277 struct cgroup *cgrp;
1278 int err;
1279
1280 if (attr->link_create.flags)
1281 return -EINVAL;
1282
1283 cgrp = cgroup_get_from_fd(attr->link_create.target_fd);
1284 if (IS_ERR(cgrp))
1285 return PTR_ERR(cgrp);
1286
1287 link = kzalloc(sizeof(*link), GFP_USER);
1288 if (!link) {
1289 err = -ENOMEM;
1290 goto out_put_cgroup;
1291 }
1292 bpf_link_init(&link->link, BPF_LINK_TYPE_CGROUP, &bpf_cgroup_link_lops,
1293 prog);
1294 link->cgroup = cgrp;
1295 link->type = attr->link_create.attach_type;
1296
1297 err = bpf_link_prime(&link->link, &link_primer);
1298 if (err) {
1299 kfree(link);
1300 goto out_put_cgroup;
1301 }
1302
1303 err = cgroup_bpf_attach(cgrp, NULL, NULL, link,
1304 link->type, BPF_F_ALLOW_MULTI);
1305 if (err) {
1306 bpf_link_cleanup(&link_primer);
1307 goto out_put_cgroup;
1308 }
1309
1310 return bpf_link_settle(&link_primer);
1311
1312out_put_cgroup:
1313 cgroup_put(cgrp);
1314 return err;
1315}
1316
1317int cgroup_bpf_prog_query(const union bpf_attr *attr,
1318 union bpf_attr __user *uattr)
1319{
1320 struct cgroup *cgrp;
1321 int ret;
1322
1323 cgrp = cgroup_get_from_fd(attr->query.target_fd);
1324 if (IS_ERR(cgrp))
1325 return PTR_ERR(cgrp);
1326
1327 ret = cgroup_bpf_query(cgrp, attr, uattr);
1328
1329 cgroup_put(cgrp);
1330 return ret;
1331}
1332
1333/**
1334 * __cgroup_bpf_run_filter_skb() - Run a program for packet filtering
1335 * @sk: The socket sending or receiving traffic
1336 * @skb: The skb that is being sent or received
1337 * @type: The type of program to be executed
1338 *
1339 * If no socket is passed, or the socket is not of type INET or INET6,
1340 * this function does nothing and returns 0.
1341 *
1342 * The program type passed in via @type must be suitable for network
1343 * filtering. No further check is performed to assert that.
1344 *
1345 * For egress packets, this function can return:
1346 * NET_XMIT_SUCCESS (0) - continue with packet output
1347 * NET_XMIT_DROP (1) - drop packet and notify TCP to call cwr
1348 * NET_XMIT_CN (2) - continue with packet output and notify TCP
1349 * to call cwr
1350 * -err - drop packet
1351 *
1352 * For ingress packets, this function will return -EPERM if any
1353 * attached program was found and if it returned != 1 during execution.
1354 * Otherwise 0 is returned.
1355 */
1356int __cgroup_bpf_run_filter_skb(struct sock *sk,
1357 struct sk_buff *skb,
1358 enum cgroup_bpf_attach_type atype)
1359{
1360 unsigned int offset = skb->data - skb_network_header(skb);
1361 struct sock *save_sk;
1362 void *saved_data_end;
1363 struct cgroup *cgrp;
1364 int ret;
1365
1366 if (!sk || !sk_fullsock(sk))
1367 return 0;
1368
1369 if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)
1370 return 0;
1371
1372 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1373 save_sk = skb->sk;
1374 skb->sk = sk;
1375 __skb_push(skb, offset);
1376
1377 /* compute pointers for the bpf prog */
1378 bpf_compute_and_save_data_end(skb, &saved_data_end);
1379
1380 if (atype == CGROUP_INET_EGRESS) {
1381 u32 flags = 0;
1382 bool cn;
1383
1384 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, skb,
1385 __bpf_prog_run_save_cb, 0, &flags);
1386
1387 /* Return values of CGROUP EGRESS BPF programs are:
1388 * 0: drop packet
1389 * 1: keep packet
1390 * 2: drop packet and cn
1391 * 3: keep packet and cn
1392 *
1393 * The returned value is then converted to one of the NET_XMIT
1394 * or an error code that is then interpreted as drop packet
1395 * (and no cn):
1396 * 0: NET_XMIT_SUCCESS skb should be transmitted
1397 * 1: NET_XMIT_DROP skb should be dropped and cn
1398 * 2: NET_XMIT_CN skb should be transmitted and cn
1399 * 3: -err skb should be dropped
1400 */
1401
1402 cn = flags & BPF_RET_SET_CN;
1403 if (ret && !IS_ERR_VALUE((long)ret))
1404 ret = -EFAULT;
1405 if (!ret)
1406 ret = (cn ? NET_XMIT_CN : NET_XMIT_SUCCESS);
1407 else
1408 ret = (cn ? NET_XMIT_DROP : ret);
1409 } else {
1410 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype,
1411 skb, __bpf_prog_run_save_cb, 0,
1412 NULL);
1413 if (ret && !IS_ERR_VALUE((long)ret))
1414 ret = -EFAULT;
1415 }
1416 bpf_restore_data_end(skb, saved_data_end);
1417 __skb_pull(skb, offset);
1418 skb->sk = save_sk;
1419
1420 return ret;
1421}
1422EXPORT_SYMBOL(__cgroup_bpf_run_filter_skb);
1423
1424/**
1425 * __cgroup_bpf_run_filter_sk() - Run a program on a sock
1426 * @sk: sock structure to manipulate
1427 * @type: The type of program to be executed
1428 *
1429 * socket is passed is expected to be of type INET or INET6.
1430 *
1431 * The program type passed in via @type must be suitable for sock
1432 * filtering. No further check is performed to assert that.
1433 *
1434 * This function will return %-EPERM if any if an attached program was found
1435 * and if it returned != 1 during execution. In all other cases, 0 is returned.
1436 */
1437int __cgroup_bpf_run_filter_sk(struct sock *sk,
1438 enum cgroup_bpf_attach_type atype)
1439{
1440 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1441
1442 return bpf_prog_run_array_cg(&cgrp->bpf, atype, sk, bpf_prog_run, 0,
1443 NULL);
1444}
1445EXPORT_SYMBOL(__cgroup_bpf_run_filter_sk);
1446
1447/**
1448 * __cgroup_bpf_run_filter_sock_addr() - Run a program on a sock and
1449 * provided by user sockaddr
1450 * @sk: sock struct that will use sockaddr
1451 * @uaddr: sockaddr struct provided by user
1452 * @type: The type of program to be executed
1453 * @t_ctx: Pointer to attach type specific context
1454 * @flags: Pointer to u32 which contains higher bits of BPF program
1455 * return value (OR'ed together).
1456 *
1457 * socket is expected to be of type INET or INET6.
1458 *
1459 * This function will return %-EPERM if an attached program is found and
1460 * returned value != 1 during execution. In all other cases, 0 is returned.
1461 */
1462int __cgroup_bpf_run_filter_sock_addr(struct sock *sk,
1463 struct sockaddr *uaddr,
1464 enum cgroup_bpf_attach_type atype,
1465 void *t_ctx,
1466 u32 *flags)
1467{
1468 struct bpf_sock_addr_kern ctx = {
1469 .sk = sk,
1470 .uaddr = uaddr,
1471 .t_ctx = t_ctx,
1472 };
1473 struct sockaddr_storage unspec;
1474 struct cgroup *cgrp;
1475
1476 /* Check socket family since not all sockets represent network
1477 * endpoint (e.g. AF_UNIX).
1478 */
1479 if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)
1480 return 0;
1481
1482 if (!ctx.uaddr) {
1483 memset(&unspec, 0, sizeof(unspec));
1484 ctx.uaddr = (struct sockaddr *)&unspec;
1485 }
1486
1487 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1488 return bpf_prog_run_array_cg(&cgrp->bpf, atype, &ctx, bpf_prog_run,
1489 0, flags);
1490}
1491EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_addr);
1492
1493/**
1494 * __cgroup_bpf_run_filter_sock_ops() - Run a program on a sock
1495 * @sk: socket to get cgroup from
1496 * @sock_ops: bpf_sock_ops_kern struct to pass to program. Contains
1497 * sk with connection information (IP addresses, etc.) May not contain
1498 * cgroup info if it is a req sock.
1499 * @type: The type of program to be executed
1500 *
1501 * socket passed is expected to be of type INET or INET6.
1502 *
1503 * The program type passed in via @type must be suitable for sock_ops
1504 * filtering. No further check is performed to assert that.
1505 *
1506 * This function will return %-EPERM if any if an attached program was found
1507 * and if it returned != 1 during execution. In all other cases, 0 is returned.
1508 */
1509int __cgroup_bpf_run_filter_sock_ops(struct sock *sk,
1510 struct bpf_sock_ops_kern *sock_ops,
1511 enum cgroup_bpf_attach_type atype)
1512{
1513 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1514
1515 return bpf_prog_run_array_cg(&cgrp->bpf, atype, sock_ops, bpf_prog_run,
1516 0, NULL);
1517}
1518EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_ops);
1519
1520int __cgroup_bpf_check_dev_permission(short dev_type, u32 major, u32 minor,
1521 short access, enum cgroup_bpf_attach_type atype)
1522{
1523 struct cgroup *cgrp;
1524 struct bpf_cgroup_dev_ctx ctx = {
1525 .access_type = (access << 16) | dev_type,
1526 .major = major,
1527 .minor = minor,
1528 };
1529 int ret;
1530
1531 rcu_read_lock();
1532 cgrp = task_dfl_cgroup(current);
1533 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, &ctx, bpf_prog_run, 0,
1534 NULL);
1535 rcu_read_unlock();
1536
1537 return ret;
1538}
1539
1540BPF_CALL_2(bpf_get_local_storage, struct bpf_map *, map, u64, flags)
1541{
1542 /* flags argument is not used now,
1543 * but provides an ability to extend the API.
1544 * verifier checks that its value is correct.
1545 */
1546 enum bpf_cgroup_storage_type stype = cgroup_storage_type(map);
1547 struct bpf_cgroup_storage *storage;
1548 struct bpf_cg_run_ctx *ctx;
1549 void *ptr;
1550
1551 /* get current cgroup storage from BPF run context */
1552 ctx = container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx);
1553 storage = ctx->prog_item->cgroup_storage[stype];
1554
1555 if (stype == BPF_CGROUP_STORAGE_SHARED)
1556 ptr = &READ_ONCE(storage->buf)->data[0];
1557 else
1558 ptr = this_cpu_ptr(storage->percpu_buf);
1559
1560 return (unsigned long)ptr;
1561}
1562
1563const struct bpf_func_proto bpf_get_local_storage_proto = {
1564 .func = bpf_get_local_storage,
1565 .gpl_only = false,
1566 .ret_type = RET_PTR_TO_MAP_VALUE,
1567 .arg1_type = ARG_CONST_MAP_PTR,
1568 .arg2_type = ARG_ANYTHING,
1569};
1570
1571BPF_CALL_0(bpf_get_retval)
1572{
1573 struct bpf_cg_run_ctx *ctx =
1574 container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx);
1575
1576 return ctx->retval;
1577}
1578
1579const struct bpf_func_proto bpf_get_retval_proto = {
1580 .func = bpf_get_retval,
1581 .gpl_only = false,
1582 .ret_type = RET_INTEGER,
1583};
1584
1585BPF_CALL_1(bpf_set_retval, int, retval)
1586{
1587 struct bpf_cg_run_ctx *ctx =
1588 container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx);
1589
1590 ctx->retval = retval;
1591 return 0;
1592}
1593
1594const struct bpf_func_proto bpf_set_retval_proto = {
1595 .func = bpf_set_retval,
1596 .gpl_only = false,
1597 .ret_type = RET_INTEGER,
1598 .arg1_type = ARG_ANYTHING,
1599};
1600
1601static const struct bpf_func_proto *
1602cgroup_dev_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1603{
1604 const struct bpf_func_proto *func_proto;
1605
1606 func_proto = cgroup_common_func_proto(func_id, prog);
1607 if (func_proto)
1608 return func_proto;
1609
1610 func_proto = cgroup_current_func_proto(func_id, prog);
1611 if (func_proto)
1612 return func_proto;
1613
1614 switch (func_id) {
1615 case BPF_FUNC_perf_event_output:
1616 return &bpf_event_output_data_proto;
1617 default:
1618 return bpf_base_func_proto(func_id);
1619 }
1620}
1621
1622static bool cgroup_dev_is_valid_access(int off, int size,
1623 enum bpf_access_type type,
1624 const struct bpf_prog *prog,
1625 struct bpf_insn_access_aux *info)
1626{
1627 const int size_default = sizeof(__u32);
1628
1629 if (type == BPF_WRITE)
1630 return false;
1631
1632 if (off < 0 || off + size > sizeof(struct bpf_cgroup_dev_ctx))
1633 return false;
1634 /* The verifier guarantees that size > 0. */
1635 if (off % size != 0)
1636 return false;
1637
1638 switch (off) {
1639 case bpf_ctx_range(struct bpf_cgroup_dev_ctx, access_type):
1640 bpf_ctx_record_field_size(info, size_default);
1641 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
1642 return false;
1643 break;
1644 default:
1645 if (size != size_default)
1646 return false;
1647 }
1648
1649 return true;
1650}
1651
1652const struct bpf_prog_ops cg_dev_prog_ops = {
1653};
1654
1655const struct bpf_verifier_ops cg_dev_verifier_ops = {
1656 .get_func_proto = cgroup_dev_func_proto,
1657 .is_valid_access = cgroup_dev_is_valid_access,
1658};
1659
1660/**
1661 * __cgroup_bpf_run_filter_sysctl - Run a program on sysctl
1662 *
1663 * @head: sysctl table header
1664 * @table: sysctl table
1665 * @write: sysctl is being read (= 0) or written (= 1)
1666 * @buf: pointer to buffer (in and out)
1667 * @pcount: value-result argument: value is size of buffer pointed to by @buf,
1668 * result is size of @new_buf if program set new value, initial value
1669 * otherwise
1670 * @ppos: value-result argument: value is position at which read from or write
1671 * to sysctl is happening, result is new position if program overrode it,
1672 * initial value otherwise
1673 * @type: type of program to be executed
1674 *
1675 * Program is run when sysctl is being accessed, either read or written, and
1676 * can allow or deny such access.
1677 *
1678 * This function will return %-EPERM if an attached program is found and
1679 * returned value != 1 during execution. In all other cases 0 is returned.
1680 */
1681int __cgroup_bpf_run_filter_sysctl(struct ctl_table_header *head,
1682 struct ctl_table *table, int write,
1683 char **buf, size_t *pcount, loff_t *ppos,
1684 enum cgroup_bpf_attach_type atype)
1685{
1686 struct bpf_sysctl_kern ctx = {
1687 .head = head,
1688 .table = table,
1689 .write = write,
1690 .ppos = ppos,
1691 .cur_val = NULL,
1692 .cur_len = PAGE_SIZE,
1693 .new_val = NULL,
1694 .new_len = 0,
1695 .new_updated = 0,
1696 };
1697 struct cgroup *cgrp;
1698 loff_t pos = 0;
1699 int ret;
1700
1701 ctx.cur_val = kmalloc_track_caller(ctx.cur_len, GFP_KERNEL);
1702 if (!ctx.cur_val ||
1703 table->proc_handler(table, 0, ctx.cur_val, &ctx.cur_len, &pos)) {
1704 /* Let BPF program decide how to proceed. */
1705 ctx.cur_len = 0;
1706 }
1707
1708 if (write && *buf && *pcount) {
1709 /* BPF program should be able to override new value with a
1710 * buffer bigger than provided by user.
1711 */
1712 ctx.new_val = kmalloc_track_caller(PAGE_SIZE, GFP_KERNEL);
1713 ctx.new_len = min_t(size_t, PAGE_SIZE, *pcount);
1714 if (ctx.new_val) {
1715 memcpy(ctx.new_val, *buf, ctx.new_len);
1716 } else {
1717 /* Let BPF program decide how to proceed. */
1718 ctx.new_len = 0;
1719 }
1720 }
1721
1722 rcu_read_lock();
1723 cgrp = task_dfl_cgroup(current);
1724 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, &ctx, bpf_prog_run, 0,
1725 NULL);
1726 rcu_read_unlock();
1727
1728 kfree(ctx.cur_val);
1729
1730 if (ret == 1 && ctx.new_updated) {
1731 kfree(*buf);
1732 *buf = ctx.new_val;
1733 *pcount = ctx.new_len;
1734 } else {
1735 kfree(ctx.new_val);
1736 }
1737
1738 return ret;
1739}
1740
1741#ifdef CONFIG_NET
1742static int sockopt_alloc_buf(struct bpf_sockopt_kern *ctx, int max_optlen,
1743 struct bpf_sockopt_buf *buf)
1744{
1745 if (unlikely(max_optlen < 0))
1746 return -EINVAL;
1747
1748 if (unlikely(max_optlen > PAGE_SIZE)) {
1749 /* We don't expose optvals that are greater than PAGE_SIZE
1750 * to the BPF program.
1751 */
1752 max_optlen = PAGE_SIZE;
1753 }
1754
1755 if (max_optlen <= sizeof(buf->data)) {
1756 /* When the optval fits into BPF_SOCKOPT_KERN_BUF_SIZE
1757 * bytes avoid the cost of kzalloc.
1758 */
1759 ctx->optval = buf->data;
1760 ctx->optval_end = ctx->optval + max_optlen;
1761 return max_optlen;
1762 }
1763
1764 ctx->optval = kzalloc(max_optlen, GFP_USER);
1765 if (!ctx->optval)
1766 return -ENOMEM;
1767
1768 ctx->optval_end = ctx->optval + max_optlen;
1769
1770 return max_optlen;
1771}
1772
1773static void sockopt_free_buf(struct bpf_sockopt_kern *ctx,
1774 struct bpf_sockopt_buf *buf)
1775{
1776 if (ctx->optval == buf->data)
1777 return;
1778 kfree(ctx->optval);
1779}
1780
1781static bool sockopt_buf_allocated(struct bpf_sockopt_kern *ctx,
1782 struct bpf_sockopt_buf *buf)
1783{
1784 return ctx->optval != buf->data;
1785}
1786
1787int __cgroup_bpf_run_filter_setsockopt(struct sock *sk, int *level,
1788 int *optname, char __user *optval,
1789 int *optlen, char **kernel_optval)
1790{
1791 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1792 struct bpf_sockopt_buf buf = {};
1793 struct bpf_sockopt_kern ctx = {
1794 .sk = sk,
1795 .level = *level,
1796 .optname = *optname,
1797 };
1798 int ret, max_optlen;
1799
1800 /* Allocate a bit more than the initial user buffer for
1801 * BPF program. The canonical use case is overriding
1802 * TCP_CONGESTION(nv) to TCP_CONGESTION(cubic).
1803 */
1804 max_optlen = max_t(int, 16, *optlen);
1805 max_optlen = sockopt_alloc_buf(&ctx, max_optlen, &buf);
1806 if (max_optlen < 0)
1807 return max_optlen;
1808
1809 ctx.optlen = *optlen;
1810
1811 if (copy_from_user(ctx.optval, optval, min(*optlen, max_optlen)) != 0) {
1812 ret = -EFAULT;
1813 goto out;
1814 }
1815
1816 lock_sock(sk);
1817 ret = bpf_prog_run_array_cg(&cgrp->bpf, CGROUP_SETSOCKOPT,
1818 &ctx, bpf_prog_run, 0, NULL);
1819 release_sock(sk);
1820
1821 if (ret)
1822 goto out;
1823
1824 if (ctx.optlen == -1) {
1825 /* optlen set to -1, bypass kernel */
1826 ret = 1;
1827 } else if (ctx.optlen > max_optlen || ctx.optlen < -1) {
1828 /* optlen is out of bounds */
1829 ret = -EFAULT;
1830 } else {
1831 /* optlen within bounds, run kernel handler */
1832 ret = 0;
1833
1834 /* export any potential modifications */
1835 *level = ctx.level;
1836 *optname = ctx.optname;
1837
1838 /* optlen == 0 from BPF indicates that we should
1839 * use original userspace data.
1840 */
1841 if (ctx.optlen != 0) {
1842 *optlen = ctx.optlen;
1843 /* We've used bpf_sockopt_kern->buf as an intermediary
1844 * storage, but the BPF program indicates that we need
1845 * to pass this data to the kernel setsockopt handler.
1846 * No way to export on-stack buf, have to allocate a
1847 * new buffer.
1848 */
1849 if (!sockopt_buf_allocated(&ctx, &buf)) {
1850 void *p = kmalloc(ctx.optlen, GFP_USER);
1851
1852 if (!p) {
1853 ret = -ENOMEM;
1854 goto out;
1855 }
1856 memcpy(p, ctx.optval, ctx.optlen);
1857 *kernel_optval = p;
1858 } else {
1859 *kernel_optval = ctx.optval;
1860 }
1861 /* export and don't free sockopt buf */
1862 return 0;
1863 }
1864 }
1865
1866out:
1867 sockopt_free_buf(&ctx, &buf);
1868 return ret;
1869}
1870
1871int __cgroup_bpf_run_filter_getsockopt(struct sock *sk, int level,
1872 int optname, char __user *optval,
1873 int __user *optlen, int max_optlen,
1874 int retval)
1875{
1876 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1877 struct bpf_sockopt_buf buf = {};
1878 struct bpf_sockopt_kern ctx = {
1879 .sk = sk,
1880 .level = level,
1881 .optname = optname,
1882 .current_task = current,
1883 };
1884 int ret;
1885
1886 ctx.optlen = max_optlen;
1887 max_optlen = sockopt_alloc_buf(&ctx, max_optlen, &buf);
1888 if (max_optlen < 0)
1889 return max_optlen;
1890
1891 if (!retval) {
1892 /* If kernel getsockopt finished successfully,
1893 * copy whatever was returned to the user back
1894 * into our temporary buffer. Set optlen to the
1895 * one that kernel returned as well to let
1896 * BPF programs inspect the value.
1897 */
1898
1899 if (get_user(ctx.optlen, optlen)) {
1900 ret = -EFAULT;
1901 goto out;
1902 }
1903
1904 if (ctx.optlen < 0) {
1905 ret = -EFAULT;
1906 goto out;
1907 }
1908
1909 if (copy_from_user(ctx.optval, optval,
1910 min(ctx.optlen, max_optlen)) != 0) {
1911 ret = -EFAULT;
1912 goto out;
1913 }
1914 }
1915
1916 lock_sock(sk);
1917 ret = bpf_prog_run_array_cg(&cgrp->bpf, CGROUP_GETSOCKOPT,
1918 &ctx, bpf_prog_run, retval, NULL);
1919 release_sock(sk);
1920
1921 if (ret < 0)
1922 goto out;
1923
1924 if (ctx.optlen > max_optlen || ctx.optlen < 0) {
1925 ret = -EFAULT;
1926 goto out;
1927 }
1928
1929 if (ctx.optlen != 0) {
1930 if (copy_to_user(optval, ctx.optval, ctx.optlen) ||
1931 put_user(ctx.optlen, optlen)) {
1932 ret = -EFAULT;
1933 goto out;
1934 }
1935 }
1936
1937out:
1938 sockopt_free_buf(&ctx, &buf);
1939 return ret;
1940}
1941
1942int __cgroup_bpf_run_filter_getsockopt_kern(struct sock *sk, int level,
1943 int optname, void *optval,
1944 int *optlen, int retval)
1945{
1946 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1947 struct bpf_sockopt_kern ctx = {
1948 .sk = sk,
1949 .level = level,
1950 .optname = optname,
1951 .optlen = *optlen,
1952 .optval = optval,
1953 .optval_end = optval + *optlen,
1954 .current_task = current,
1955 };
1956 int ret;
1957
1958 /* Note that __cgroup_bpf_run_filter_getsockopt doesn't copy
1959 * user data back into BPF buffer when reval != 0. This is
1960 * done as an optimization to avoid extra copy, assuming
1961 * kernel won't populate the data in case of an error.
1962 * Here we always pass the data and memset() should
1963 * be called if that data shouldn't be "exported".
1964 */
1965
1966 ret = bpf_prog_run_array_cg(&cgrp->bpf, CGROUP_GETSOCKOPT,
1967 &ctx, bpf_prog_run, retval, NULL);
1968 if (ret < 0)
1969 return ret;
1970
1971 if (ctx.optlen > *optlen)
1972 return -EFAULT;
1973
1974 /* BPF programs can shrink the buffer, export the modifications.
1975 */
1976 if (ctx.optlen != 0)
1977 *optlen = ctx.optlen;
1978
1979 return ret;
1980}
1981#endif
1982
1983static ssize_t sysctl_cpy_dir(const struct ctl_dir *dir, char **bufp,
1984 size_t *lenp)
1985{
1986 ssize_t tmp_ret = 0, ret;
1987
1988 if (dir->header.parent) {
1989 tmp_ret = sysctl_cpy_dir(dir->header.parent, bufp, lenp);
1990 if (tmp_ret < 0)
1991 return tmp_ret;
1992 }
1993
1994 ret = strscpy(*bufp, dir->header.ctl_table[0].procname, *lenp);
1995 if (ret < 0)
1996 return ret;
1997 *bufp += ret;
1998 *lenp -= ret;
1999 ret += tmp_ret;
2000
2001 /* Avoid leading slash. */
2002 if (!ret)
2003 return ret;
2004
2005 tmp_ret = strscpy(*bufp, "/", *lenp);
2006 if (tmp_ret < 0)
2007 return tmp_ret;
2008 *bufp += tmp_ret;
2009 *lenp -= tmp_ret;
2010
2011 return ret + tmp_ret;
2012}
2013
2014BPF_CALL_4(bpf_sysctl_get_name, struct bpf_sysctl_kern *, ctx, char *, buf,
2015 size_t, buf_len, u64, flags)
2016{
2017 ssize_t tmp_ret = 0, ret;
2018
2019 if (!buf)
2020 return -EINVAL;
2021
2022 if (!(flags & BPF_F_SYSCTL_BASE_NAME)) {
2023 if (!ctx->head)
2024 return -EINVAL;
2025 tmp_ret = sysctl_cpy_dir(ctx->head->parent, &buf, &buf_len);
2026 if (tmp_ret < 0)
2027 return tmp_ret;
2028 }
2029
2030 ret = strscpy(buf, ctx->table->procname, buf_len);
2031
2032 return ret < 0 ? ret : tmp_ret + ret;
2033}
2034
2035static const struct bpf_func_proto bpf_sysctl_get_name_proto = {
2036 .func = bpf_sysctl_get_name,
2037 .gpl_only = false,
2038 .ret_type = RET_INTEGER,
2039 .arg1_type = ARG_PTR_TO_CTX,
2040 .arg2_type = ARG_PTR_TO_MEM,
2041 .arg3_type = ARG_CONST_SIZE,
2042 .arg4_type = ARG_ANYTHING,
2043};
2044
2045static int copy_sysctl_value(char *dst, size_t dst_len, char *src,
2046 size_t src_len)
2047{
2048 if (!dst)
2049 return -EINVAL;
2050
2051 if (!dst_len)
2052 return -E2BIG;
2053
2054 if (!src || !src_len) {
2055 memset(dst, 0, dst_len);
2056 return -EINVAL;
2057 }
2058
2059 memcpy(dst, src, min(dst_len, src_len));
2060
2061 if (dst_len > src_len) {
2062 memset(dst + src_len, '\0', dst_len - src_len);
2063 return src_len;
2064 }
2065
2066 dst[dst_len - 1] = '\0';
2067
2068 return -E2BIG;
2069}
2070
2071BPF_CALL_3(bpf_sysctl_get_current_value, struct bpf_sysctl_kern *, ctx,
2072 char *, buf, size_t, buf_len)
2073{
2074 return copy_sysctl_value(buf, buf_len, ctx->cur_val, ctx->cur_len);
2075}
2076
2077static const struct bpf_func_proto bpf_sysctl_get_current_value_proto = {
2078 .func = bpf_sysctl_get_current_value,
2079 .gpl_only = false,
2080 .ret_type = RET_INTEGER,
2081 .arg1_type = ARG_PTR_TO_CTX,
2082 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
2083 .arg3_type = ARG_CONST_SIZE,
2084};
2085
2086BPF_CALL_3(bpf_sysctl_get_new_value, struct bpf_sysctl_kern *, ctx, char *, buf,
2087 size_t, buf_len)
2088{
2089 if (!ctx->write) {
2090 if (buf && buf_len)
2091 memset(buf, '\0', buf_len);
2092 return -EINVAL;
2093 }
2094 return copy_sysctl_value(buf, buf_len, ctx->new_val, ctx->new_len);
2095}
2096
2097static const struct bpf_func_proto bpf_sysctl_get_new_value_proto = {
2098 .func = bpf_sysctl_get_new_value,
2099 .gpl_only = false,
2100 .ret_type = RET_INTEGER,
2101 .arg1_type = ARG_PTR_TO_CTX,
2102 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
2103 .arg3_type = ARG_CONST_SIZE,
2104};
2105
2106BPF_CALL_3(bpf_sysctl_set_new_value, struct bpf_sysctl_kern *, ctx,
2107 const char *, buf, size_t, buf_len)
2108{
2109 if (!ctx->write || !ctx->new_val || !ctx->new_len || !buf || !buf_len)
2110 return -EINVAL;
2111
2112 if (buf_len > PAGE_SIZE - 1)
2113 return -E2BIG;
2114
2115 memcpy(ctx->new_val, buf, buf_len);
2116 ctx->new_len = buf_len;
2117 ctx->new_updated = 1;
2118
2119 return 0;
2120}
2121
2122static const struct bpf_func_proto bpf_sysctl_set_new_value_proto = {
2123 .func = bpf_sysctl_set_new_value,
2124 .gpl_only = false,
2125 .ret_type = RET_INTEGER,
2126 .arg1_type = ARG_PTR_TO_CTX,
2127 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
2128 .arg3_type = ARG_CONST_SIZE,
2129};
2130
2131static const struct bpf_func_proto *
2132sysctl_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
2133{
2134 const struct bpf_func_proto *func_proto;
2135
2136 func_proto = cgroup_common_func_proto(func_id, prog);
2137 if (func_proto)
2138 return func_proto;
2139
2140 func_proto = cgroup_current_func_proto(func_id, prog);
2141 if (func_proto)
2142 return func_proto;
2143
2144 switch (func_id) {
2145 case BPF_FUNC_sysctl_get_name:
2146 return &bpf_sysctl_get_name_proto;
2147 case BPF_FUNC_sysctl_get_current_value:
2148 return &bpf_sysctl_get_current_value_proto;
2149 case BPF_FUNC_sysctl_get_new_value:
2150 return &bpf_sysctl_get_new_value_proto;
2151 case BPF_FUNC_sysctl_set_new_value:
2152 return &bpf_sysctl_set_new_value_proto;
2153 case BPF_FUNC_ktime_get_coarse_ns:
2154 return &bpf_ktime_get_coarse_ns_proto;
2155 case BPF_FUNC_perf_event_output:
2156 return &bpf_event_output_data_proto;
2157 default:
2158 return bpf_base_func_proto(func_id);
2159 }
2160}
2161
2162static bool sysctl_is_valid_access(int off, int size, enum bpf_access_type type,
2163 const struct bpf_prog *prog,
2164 struct bpf_insn_access_aux *info)
2165{
2166 const int size_default = sizeof(__u32);
2167
2168 if (off < 0 || off + size > sizeof(struct bpf_sysctl) || off % size)
2169 return false;
2170
2171 switch (off) {
2172 case bpf_ctx_range(struct bpf_sysctl, write):
2173 if (type != BPF_READ)
2174 return false;
2175 bpf_ctx_record_field_size(info, size_default);
2176 return bpf_ctx_narrow_access_ok(off, size, size_default);
2177 case bpf_ctx_range(struct bpf_sysctl, file_pos):
2178 if (type == BPF_READ) {
2179 bpf_ctx_record_field_size(info, size_default);
2180 return bpf_ctx_narrow_access_ok(off, size, size_default);
2181 } else {
2182 return size == size_default;
2183 }
2184 default:
2185 return false;
2186 }
2187}
2188
2189static u32 sysctl_convert_ctx_access(enum bpf_access_type type,
2190 const struct bpf_insn *si,
2191 struct bpf_insn *insn_buf,
2192 struct bpf_prog *prog, u32 *target_size)
2193{
2194 struct bpf_insn *insn = insn_buf;
2195 u32 read_size;
2196
2197 switch (si->off) {
2198 case offsetof(struct bpf_sysctl, write):
2199 *insn++ = BPF_LDX_MEM(
2200 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
2201 bpf_target_off(struct bpf_sysctl_kern, write,
2202 sizeof_field(struct bpf_sysctl_kern,
2203 write),
2204 target_size));
2205 break;
2206 case offsetof(struct bpf_sysctl, file_pos):
2207 /* ppos is a pointer so it should be accessed via indirect
2208 * loads and stores. Also for stores additional temporary
2209 * register is used since neither src_reg nor dst_reg can be
2210 * overridden.
2211 */
2212 if (type == BPF_WRITE) {
2213 int treg = BPF_REG_9;
2214
2215 if (si->src_reg == treg || si->dst_reg == treg)
2216 --treg;
2217 if (si->src_reg == treg || si->dst_reg == treg)
2218 --treg;
2219 *insn++ = BPF_STX_MEM(
2220 BPF_DW, si->dst_reg, treg,
2221 offsetof(struct bpf_sysctl_kern, tmp_reg));
2222 *insn++ = BPF_LDX_MEM(
2223 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos),
2224 treg, si->dst_reg,
2225 offsetof(struct bpf_sysctl_kern, ppos));
2226 *insn++ = BPF_STX_MEM(
2227 BPF_SIZEOF(u32), treg, si->src_reg,
2228 bpf_ctx_narrow_access_offset(
2229 0, sizeof(u32), sizeof(loff_t)));
2230 *insn++ = BPF_LDX_MEM(
2231 BPF_DW, treg, si->dst_reg,
2232 offsetof(struct bpf_sysctl_kern, tmp_reg));
2233 } else {
2234 *insn++ = BPF_LDX_MEM(
2235 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos),
2236 si->dst_reg, si->src_reg,
2237 offsetof(struct bpf_sysctl_kern, ppos));
2238 read_size = bpf_size_to_bytes(BPF_SIZE(si->code));
2239 *insn++ = BPF_LDX_MEM(
2240 BPF_SIZE(si->code), si->dst_reg, si->dst_reg,
2241 bpf_ctx_narrow_access_offset(
2242 0, read_size, sizeof(loff_t)));
2243 }
2244 *target_size = sizeof(u32);
2245 break;
2246 }
2247
2248 return insn - insn_buf;
2249}
2250
2251const struct bpf_verifier_ops cg_sysctl_verifier_ops = {
2252 .get_func_proto = sysctl_func_proto,
2253 .is_valid_access = sysctl_is_valid_access,
2254 .convert_ctx_access = sysctl_convert_ctx_access,
2255};
2256
2257const struct bpf_prog_ops cg_sysctl_prog_ops = {
2258};
2259
2260#ifdef CONFIG_NET
2261BPF_CALL_1(bpf_get_netns_cookie_sockopt, struct bpf_sockopt_kern *, ctx)
2262{
2263 const struct net *net = ctx ? sock_net(ctx->sk) : &init_net;
2264
2265 return net->net_cookie;
2266}
2267
2268static const struct bpf_func_proto bpf_get_netns_cookie_sockopt_proto = {
2269 .func = bpf_get_netns_cookie_sockopt,
2270 .gpl_only = false,
2271 .ret_type = RET_INTEGER,
2272 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
2273};
2274#endif
2275
2276static const struct bpf_func_proto *
2277cg_sockopt_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
2278{
2279 const struct bpf_func_proto *func_proto;
2280
2281 func_proto = cgroup_common_func_proto(func_id, prog);
2282 if (func_proto)
2283 return func_proto;
2284
2285 func_proto = cgroup_current_func_proto(func_id, prog);
2286 if (func_proto)
2287 return func_proto;
2288
2289 switch (func_id) {
2290#ifdef CONFIG_NET
2291 case BPF_FUNC_get_netns_cookie:
2292 return &bpf_get_netns_cookie_sockopt_proto;
2293 case BPF_FUNC_sk_storage_get:
2294 return &bpf_sk_storage_get_proto;
2295 case BPF_FUNC_sk_storage_delete:
2296 return &bpf_sk_storage_delete_proto;
2297 case BPF_FUNC_setsockopt:
2298 if (prog->expected_attach_type == BPF_CGROUP_SETSOCKOPT)
2299 return &bpf_sk_setsockopt_proto;
2300 return NULL;
2301 case BPF_FUNC_getsockopt:
2302 if (prog->expected_attach_type == BPF_CGROUP_SETSOCKOPT)
2303 return &bpf_sk_getsockopt_proto;
2304 return NULL;
2305#endif
2306#ifdef CONFIG_INET
2307 case BPF_FUNC_tcp_sock:
2308 return &bpf_tcp_sock_proto;
2309#endif
2310 case BPF_FUNC_perf_event_output:
2311 return &bpf_event_output_data_proto;
2312 default:
2313 return bpf_base_func_proto(func_id);
2314 }
2315}
2316
2317static bool cg_sockopt_is_valid_access(int off, int size,
2318 enum bpf_access_type type,
2319 const struct bpf_prog *prog,
2320 struct bpf_insn_access_aux *info)
2321{
2322 const int size_default = sizeof(__u32);
2323
2324 if (off < 0 || off >= sizeof(struct bpf_sockopt))
2325 return false;
2326
2327 if (off % size != 0)
2328 return false;
2329
2330 if (type == BPF_WRITE) {
2331 switch (off) {
2332 case offsetof(struct bpf_sockopt, retval):
2333 if (size != size_default)
2334 return false;
2335 return prog->expected_attach_type ==
2336 BPF_CGROUP_GETSOCKOPT;
2337 case offsetof(struct bpf_sockopt, optname):
2338 fallthrough;
2339 case offsetof(struct bpf_sockopt, level):
2340 if (size != size_default)
2341 return false;
2342 return prog->expected_attach_type ==
2343 BPF_CGROUP_SETSOCKOPT;
2344 case offsetof(struct bpf_sockopt, optlen):
2345 return size == size_default;
2346 default:
2347 return false;
2348 }
2349 }
2350
2351 switch (off) {
2352 case offsetof(struct bpf_sockopt, sk):
2353 if (size != sizeof(__u64))
2354 return false;
2355 info->reg_type = PTR_TO_SOCKET;
2356 break;
2357 case offsetof(struct bpf_sockopt, optval):
2358 if (size != sizeof(__u64))
2359 return false;
2360 info->reg_type = PTR_TO_PACKET;
2361 break;
2362 case offsetof(struct bpf_sockopt, optval_end):
2363 if (size != sizeof(__u64))
2364 return false;
2365 info->reg_type = PTR_TO_PACKET_END;
2366 break;
2367 case offsetof(struct bpf_sockopt, retval):
2368 if (size != size_default)
2369 return false;
2370 return prog->expected_attach_type == BPF_CGROUP_GETSOCKOPT;
2371 default:
2372 if (size != size_default)
2373 return false;
2374 break;
2375 }
2376 return true;
2377}
2378
2379#define CG_SOCKOPT_ACCESS_FIELD(T, F) \
2380 T(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, F), \
2381 si->dst_reg, si->src_reg, \
2382 offsetof(struct bpf_sockopt_kern, F))
2383
2384static u32 cg_sockopt_convert_ctx_access(enum bpf_access_type type,
2385 const struct bpf_insn *si,
2386 struct bpf_insn *insn_buf,
2387 struct bpf_prog *prog,
2388 u32 *target_size)
2389{
2390 struct bpf_insn *insn = insn_buf;
2391
2392 switch (si->off) {
2393 case offsetof(struct bpf_sockopt, sk):
2394 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, sk);
2395 break;
2396 case offsetof(struct bpf_sockopt, level):
2397 if (type == BPF_WRITE)
2398 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, level);
2399 else
2400 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, level);
2401 break;
2402 case offsetof(struct bpf_sockopt, optname):
2403 if (type == BPF_WRITE)
2404 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, optname);
2405 else
2406 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optname);
2407 break;
2408 case offsetof(struct bpf_sockopt, optlen):
2409 if (type == BPF_WRITE)
2410 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_STX_MEM, optlen);
2411 else
2412 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optlen);
2413 break;
2414 case offsetof(struct bpf_sockopt, retval):
2415 BUILD_BUG_ON(offsetof(struct bpf_cg_run_ctx, run_ctx) != 0);
2416
2417 if (type == BPF_WRITE) {
2418 int treg = BPF_REG_9;
2419
2420 if (si->src_reg == treg || si->dst_reg == treg)
2421 --treg;
2422 if (si->src_reg == treg || si->dst_reg == treg)
2423 --treg;
2424 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, treg,
2425 offsetof(struct bpf_sockopt_kern, tmp_reg));
2426 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, current_task),
2427 treg, si->dst_reg,
2428 offsetof(struct bpf_sockopt_kern, current_task));
2429 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct task_struct, bpf_ctx),
2430 treg, treg,
2431 offsetof(struct task_struct, bpf_ctx));
2432 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(struct bpf_cg_run_ctx, retval),
2433 treg, si->src_reg,
2434 offsetof(struct bpf_cg_run_ctx, retval));
2435 *insn++ = BPF_LDX_MEM(BPF_DW, treg, si->dst_reg,
2436 offsetof(struct bpf_sockopt_kern, tmp_reg));
2437 } else {
2438 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, current_task),
2439 si->dst_reg, si->src_reg,
2440 offsetof(struct bpf_sockopt_kern, current_task));
2441 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct task_struct, bpf_ctx),
2442 si->dst_reg, si->dst_reg,
2443 offsetof(struct task_struct, bpf_ctx));
2444 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_cg_run_ctx, retval),
2445 si->dst_reg, si->dst_reg,
2446 offsetof(struct bpf_cg_run_ctx, retval));
2447 }
2448 break;
2449 case offsetof(struct bpf_sockopt, optval):
2450 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optval);
2451 break;
2452 case offsetof(struct bpf_sockopt, optval_end):
2453 *insn++ = CG_SOCKOPT_ACCESS_FIELD(BPF_LDX_MEM, optval_end);
2454 break;
2455 }
2456
2457 return insn - insn_buf;
2458}
2459
2460static int cg_sockopt_get_prologue(struct bpf_insn *insn_buf,
2461 bool direct_write,
2462 const struct bpf_prog *prog)
2463{
2464 /* Nothing to do for sockopt argument. The data is kzalloc'ated.
2465 */
2466 return 0;
2467}
2468
2469const struct bpf_verifier_ops cg_sockopt_verifier_ops = {
2470 .get_func_proto = cg_sockopt_func_proto,
2471 .is_valid_access = cg_sockopt_is_valid_access,
2472 .convert_ctx_access = cg_sockopt_convert_ctx_access,
2473 .gen_prologue = cg_sockopt_get_prologue,
2474};
2475
2476const struct bpf_prog_ops cg_sockopt_prog_ops = {
2477};
2478
2479/* Common helpers for cgroup hooks. */
2480const struct bpf_func_proto *
2481cgroup_common_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
2482{
2483 switch (func_id) {
2484 case BPF_FUNC_get_local_storage:
2485 return &bpf_get_local_storage_proto;
2486 case BPF_FUNC_get_retval:
2487 switch (prog->expected_attach_type) {
2488 case BPF_CGROUP_INET_INGRESS:
2489 case BPF_CGROUP_INET_EGRESS:
2490 case BPF_CGROUP_SOCK_OPS:
2491 case BPF_CGROUP_UDP4_RECVMSG:
2492 case BPF_CGROUP_UDP6_RECVMSG:
2493 case BPF_CGROUP_INET4_GETPEERNAME:
2494 case BPF_CGROUP_INET6_GETPEERNAME:
2495 case BPF_CGROUP_INET4_GETSOCKNAME:
2496 case BPF_CGROUP_INET6_GETSOCKNAME:
2497 return NULL;
2498 default:
2499 return &bpf_get_retval_proto;
2500 }
2501 case BPF_FUNC_set_retval:
2502 switch (prog->expected_attach_type) {
2503 case BPF_CGROUP_INET_INGRESS:
2504 case BPF_CGROUP_INET_EGRESS:
2505 case BPF_CGROUP_SOCK_OPS:
2506 case BPF_CGROUP_UDP4_RECVMSG:
2507 case BPF_CGROUP_UDP6_RECVMSG:
2508 case BPF_CGROUP_INET4_GETPEERNAME:
2509 case BPF_CGROUP_INET6_GETPEERNAME:
2510 case BPF_CGROUP_INET4_GETSOCKNAME:
2511 case BPF_CGROUP_INET6_GETSOCKNAME:
2512 return NULL;
2513 default:
2514 return &bpf_set_retval_proto;
2515 }
2516 default:
2517 return NULL;
2518 }
2519}
2520
2521/* Common helpers for cgroup hooks with valid process context. */
2522const struct bpf_func_proto *
2523cgroup_current_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
2524{
2525 switch (func_id) {
2526 case BPF_FUNC_get_current_uid_gid:
2527 return &bpf_get_current_uid_gid_proto;
2528 case BPF_FUNC_get_current_pid_tgid:
2529 return &bpf_get_current_pid_tgid_proto;
2530 case BPF_FUNC_get_current_comm:
2531 return &bpf_get_current_comm_proto;
2532 case BPF_FUNC_get_current_cgroup_id:
2533 return &bpf_get_current_cgroup_id_proto;
2534 case BPF_FUNC_get_current_ancestor_cgroup_id:
2535 return &bpf_get_current_ancestor_cgroup_id_proto;
2536#ifdef CONFIG_CGROUP_NET_CLASSID
2537 case BPF_FUNC_get_cgroup_classid:
2538 return &bpf_get_cgroup_classid_curr_proto;
2539#endif
2540 default:
2541 return NULL;
2542 }
2543}