1 // SPDX-License-Identifier: GPL-2.0-only
3 * Functions to manage eBPF programs attached to cgroups
5 * Copyright (c) 2016 Daniel Mack
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>
20 #include <net/bpf_sk_storage.h>
22 #include "../cgroup/cgroup-internal.h"
24 DEFINE_STATIC_KEY_ARRAY_FALSE(cgroup_bpf_enabled_key, MAX_CGROUP_BPF_ATTACH_TYPE);
25 EXPORT_SYMBOL(cgroup_bpf_enabled_key);
27 /* __always_inline is necessary to prevent indirect call through run_prog
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)
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;
43 run_ctx.retval = retval;
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);
53 *(ret_flags) |= (func_ret >> 1);
56 if (!func_ret && !IS_ERR_VALUE((long)run_ctx.retval))
57 run_ctx.retval = -EPERM;
60 bpf_reset_run_ctx(old_run_ctx);
63 return run_ctx.retval;
66 unsigned int __cgroup_bpf_run_lsm_sock(const void *ctx,
67 const struct bpf_insn *insn)
69 const struct bpf_prog *shim_prog;
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));
80 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
82 ret = bpf_prog_run_array_cg(&cgrp->bpf,
83 shim_prog->aux->cgroup_atype,
84 ctx, bpf_prog_run, 0, NULL);
88 unsigned int __cgroup_bpf_run_lsm_socket(const void *ctx,
89 const struct bpf_insn *insn)
91 const struct bpf_prog *shim_prog;
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));
102 cgrp = sock_cgroup_ptr(&sock->sk->sk_cgrp_data);
104 ret = bpf_prog_run_array_cg(&cgrp->bpf,
105 shim_prog->aux->cgroup_atype,
106 ctx, bpf_prog_run, 0, NULL);
110 unsigned int __cgroup_bpf_run_lsm_current(const void *ctx,
111 const struct bpf_insn *insn)
113 const struct bpf_prog *shim_prog;
117 /*shim_prog = container_of(insn, struct bpf_prog, insnsi);*/
118 shim_prog = (const struct bpf_prog *)((void *)insn - offsetof(struct bpf_prog, insnsi));
120 /* We rely on trampoline's __bpf_prog_enter_lsm_cgroup to grab RCU read lock. */
121 cgrp = task_dfl_cgroup(current);
123 ret = bpf_prog_run_array_cg(&cgrp->bpf,
124 shim_prog->aux->cgroup_atype,
125 ctx, bpf_prog_run, 0, NULL);
129 #ifdef CONFIG_BPF_LSM
130 struct cgroup_lsm_atype {
135 static struct cgroup_lsm_atype cgroup_lsm_atype[CGROUP_LSM_NUM];
137 static enum cgroup_bpf_attach_type
138 bpf_cgroup_atype_find(enum bpf_attach_type attach_type, u32 attach_btf_id)
142 lockdep_assert_held(&cgroup_mutex);
144 if (attach_type != BPF_LSM_CGROUP)
145 return to_cgroup_bpf_attach_type(attach_type);
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;
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;
159 void bpf_cgroup_atype_get(u32 attach_btf_id, int cgroup_atype)
161 int i = cgroup_atype - CGROUP_LSM_START;
163 lockdep_assert_held(&cgroup_mutex);
165 WARN_ON_ONCE(cgroup_lsm_atype[i].attach_btf_id &&
166 cgroup_lsm_atype[i].attach_btf_id != attach_btf_id);
168 cgroup_lsm_atype[i].attach_btf_id = attach_btf_id;
169 cgroup_lsm_atype[i].refcnt++;
172 void bpf_cgroup_atype_put(int cgroup_atype)
174 int i = cgroup_atype - CGROUP_LSM_START;
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);
183 static enum cgroup_bpf_attach_type
184 bpf_cgroup_atype_find(enum bpf_attach_type attach_type, u32 attach_btf_id)
186 if (attach_type != BPF_LSM_CGROUP)
187 return to_cgroup_bpf_attach_type(attach_type);
190 #endif /* CONFIG_BPF_LSM */
192 void cgroup_bpf_offline(struct cgroup *cgrp)
195 percpu_ref_kill(&cgrp->bpf.refcnt);
198 static void bpf_cgroup_storages_free(struct bpf_cgroup_storage *storages[])
200 enum bpf_cgroup_storage_type stype;
202 for_each_cgroup_storage_type(stype)
203 bpf_cgroup_storage_free(storages[stype]);
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,
212 enum bpf_cgroup_storage_type stype;
213 struct bpf_cgroup_storage_key key;
216 key.cgroup_inode_id = cgroup_id(cgrp);
217 key.attach_type = type;
219 for_each_cgroup_storage_type(stype) {
220 map = prog->aux->cgroup_storage[stype];
224 storages[stype] = cgroup_storage_lookup((void *)map, &key, false);
228 storages[stype] = bpf_cgroup_storage_alloc(prog, stype);
229 if (IS_ERR(storages[stype])) {
230 bpf_cgroup_storages_free(new_storages);
234 new_storages[stype] = storages[stype];
240 static void bpf_cgroup_storages_assign(struct bpf_cgroup_storage *dst[],
241 struct bpf_cgroup_storage *src[])
243 enum bpf_cgroup_storage_type stype;
245 for_each_cgroup_storage_type(stype)
246 dst[stype] = src[stype];
249 static void bpf_cgroup_storages_link(struct bpf_cgroup_storage *storages[],
251 enum bpf_attach_type attach_type)
253 enum bpf_cgroup_storage_type stype;
255 for_each_cgroup_storage_type(stype)
256 bpf_cgroup_storage_link(storages[stype], cgrp, attach_type);
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.
264 static void bpf_cgroup_link_auto_detach(struct bpf_cgroup_link *link)
266 cgroup_put(link->cgroup);
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
275 static void cgroup_bpf_release(struct work_struct *work)
277 struct cgroup *p, *cgrp = container_of(work, struct cgroup,
279 struct bpf_prog_array *old_array;
280 struct list_head *storages = &cgrp->bpf.storages;
281 struct bpf_cgroup_storage *storage, *stmp;
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;
292 hlist_for_each_entry_safe(pl, pltmp, progs, node) {
293 hlist_del(&pl->node);
295 if (pl->prog->expected_attach_type == BPF_LSM_CGROUP)
296 bpf_trampoline_unlink_cgroup_shim(pl->prog);
297 bpf_prog_put(pl->prog);
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);
305 static_branch_dec(&cgroup_bpf_enabled_key[atype]);
307 old_array = rcu_dereference_protected(
308 cgrp->bpf.effective[atype],
309 lockdep_is_held(&cgroup_mutex));
310 bpf_prog_array_free(old_array);
313 list_for_each_entry_safe(storage, stmp, storages, list_cg) {
314 bpf_cgroup_storage_unlink(storage);
315 bpf_cgroup_storage_free(storage);
320 for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p))
323 percpu_ref_exit(&cgrp->bpf.refcnt);
328 * cgroup_bpf_release_fn() - callback used to schedule releasing
330 * @ref: percpu ref counter structure
332 static void cgroup_bpf_release_fn(struct percpu_ref *ref)
334 struct cgroup *cgrp = container_of(ref, struct cgroup, bpf.refcnt);
336 INIT_WORK(&cgrp->bpf.release_work, cgroup_bpf_release);
337 queue_work(system_wq, &cgrp->bpf.release_work);
340 /* Get underlying bpf_prog of bpf_prog_list entry, regardless if it's through
341 * link or direct prog.
343 static struct bpf_prog *prog_list_prog(struct bpf_prog_list *pl)
348 return pl->link->link.prog;
352 /* count number of elements in the list.
353 * it's slow but the list cannot be long
355 static u32 prog_list_length(struct hlist_head *head)
357 struct bpf_prog_list *pl;
360 hlist_for_each_entry(pl, head, node) {
361 if (!prog_list_prog(pl))
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
372 static bool hierarchy_allows_attach(struct cgroup *cgrp,
373 enum cgroup_bpf_attach_type atype)
377 p = cgroup_parent(cgrp);
381 u32 flags = p->bpf.flags[atype];
384 if (flags & BPF_F_ALLOW_MULTI)
386 cnt = prog_list_length(&p->bpf.progs[atype]);
387 WARN_ON_ONCE(cnt > 1);
389 return !!(flags & BPF_F_ALLOW_OVERRIDE);
390 p = cgroup_parent(p);
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
401 static int compute_effective_progs(struct cgroup *cgrp,
402 enum cgroup_bpf_attach_type atype,
403 struct bpf_prog_array **array)
405 struct bpf_prog_array_item *item;
406 struct bpf_prog_array *progs;
407 struct bpf_prog_list *pl;
408 struct cgroup *p = cgrp;
411 /* count number of effective programs by walking parents */
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);
418 progs = bpf_prog_array_alloc(cnt, GFP_KERNEL);
422 /* populate the array with effective progs */
426 if (cnt > 0 && !(p->bpf.flags[atype] & BPF_F_ALLOW_MULTI))
429 hlist_for_each_entry(pl, &p->bpf.progs[atype], node) {
430 if (!prog_list_prog(pl))
433 item = &progs->items[cnt];
434 item->prog = prog_list_prog(pl);
435 bpf_cgroup_storages_assign(item->cgroup_storage,
439 } while ((p = cgroup_parent(p)));
445 static void activate_effective_progs(struct cgroup *cgrp,
446 enum cgroup_bpf_attach_type atype,
447 struct bpf_prog_array *old_array)
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
454 bpf_prog_array_free(old_array);
458 * cgroup_bpf_inherit() - inherit effective programs from parent
459 * @cgrp: the cgroup to modify
461 int cgroup_bpf_inherit(struct cgroup *cgrp)
463 /* has to use marco instead of const int, since compiler thinks
464 * that array below is variable length
466 #define NR ARRAY_SIZE(cgrp->bpf.effective)
467 struct bpf_prog_array *arrays[NR] = {};
471 ret = percpu_ref_init(&cgrp->bpf.refcnt, cgroup_bpf_release_fn, 0,
476 for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p))
479 for (i = 0; i < NR; i++)
480 INIT_HLIST_HEAD(&cgrp->bpf.progs[i]);
482 INIT_LIST_HEAD(&cgrp->bpf.storages);
484 for (i = 0; i < NR; i++)
485 if (compute_effective_progs(cgrp, i, &arrays[i]))
488 for (i = 0; i < NR; i++)
489 activate_effective_progs(cgrp, i, arrays[i]);
493 for (i = 0; i < NR; i++)
494 bpf_prog_array_free(arrays[i]);
496 for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p))
499 percpu_ref_exit(&cgrp->bpf.refcnt);
504 static int update_effective_progs(struct cgroup *cgrp,
505 enum cgroup_bpf_attach_type atype)
507 struct cgroup_subsys_state *css;
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);
514 if (percpu_ref_is_zero(&desc->bpf.refcnt))
517 err = compute_effective_progs(desc, atype, &desc->bpf.inactive);
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);
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;
534 activate_effective_progs(desc, atype, desc->bpf.inactive);
535 desc->bpf.inactive = NULL;
541 /* oom while computing effective. Free all computed effective arrays
542 * since they were not activated
544 css_for_each_descendant_pre(css, &cgrp->self) {
545 struct cgroup *desc = container_of(css, struct cgroup, self);
547 bpf_prog_array_free(desc->bpf.inactive);
548 desc->bpf.inactive = NULL;
554 #define BPF_CGROUP_MAX_PROGS 64
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,
562 struct bpf_prog_list *pl;
564 /* single-attach case */
566 if (hlist_empty(progs))
568 return hlist_entry(progs->first, typeof(*pl), node);
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);
580 /* direct prog multi-attach w/ replacement case */
582 hlist_for_each_entry(pl, progs, node) {
583 if (pl->prog == replace_prog)
587 /* prog to replace not found for cgroup */
588 return ERR_PTR(-ENOENT);
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
604 * Exactly one of @prog or @link can be non-null.
605 * Must be called with cgroup_mutex held.
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)
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;
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 */
626 if (link && (prog || replace_prog))
627 /* only either link or prog/replace_prog can be specified */
629 if (!!replace_prog != !!(flags & BPF_F_REPLACE))
630 /* replace_prog implies BPF_F_REPLACE, and vice versa */
633 atype = bpf_cgroup_atype_find(type, new_prog->aux->attach_btf_id);
637 progs = &cgrp->bpf.progs[atype];
639 if (!hierarchy_allows_attach(cgrp, atype))
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
649 if (prog_list_length(progs) >= BPF_CGROUP_MAX_PROGS)
652 pl = find_attach_entry(progs, prog, link, replace_prog,
653 flags & BPF_F_ALLOW_MULTI);
657 if (bpf_cgroup_storages_alloc(storage, new_storage, type,
658 prog ? : link->link.prog, cgrp))
664 struct hlist_node *last = NULL;
666 pl = kmalloc(sizeof(*pl), GFP_KERNEL);
668 bpf_cgroup_storages_free(new_storage);
671 if (hlist_empty(progs))
672 hlist_add_head(&pl->node, progs);
674 hlist_for_each(last, progs) {
677 hlist_add_behind(&pl->node, last);
684 bpf_cgroup_storages_assign(pl->storage, storage);
685 cgrp->bpf.flags[atype] = saved_flags;
687 if (type == BPF_LSM_CGROUP) {
688 err = bpf_trampoline_link_cgroup_shim(new_prog, atype);
693 err = update_effective_progs(cgrp, atype);
695 goto cleanup_trampoline;
698 if (type == BPF_LSM_CGROUP)
699 bpf_trampoline_unlink_cgroup_shim(old_prog);
700 bpf_prog_put(old_prog);
702 static_branch_inc(&cgroup_bpf_enabled_key[atype]);
704 bpf_cgroup_storages_link(new_storage, cgrp, type);
708 if (type == BPF_LSM_CGROUP)
709 bpf_trampoline_unlink_cgroup_shim(new_prog);
716 bpf_cgroup_storages_free(new_storage);
718 hlist_del(&pl->node);
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,
733 ret = __cgroup_bpf_attach(cgrp, prog, replace_prog, link, type, flags);
738 /* Swap updated BPF program for given link in effective program arrays across
739 * all descendant cgroups. This function is guaranteed to succeed.
741 static void replace_effective_prog(struct cgroup *cgrp,
742 enum cgroup_bpf_attach_type atype,
743 struct bpf_cgroup_link *link)
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;
753 css_for_each_descendant_pre(css, &cgrp->self) {
754 struct cgroup *desc = container_of(css, struct cgroup, self);
756 if (percpu_ref_is_zero(&desc->bpf.refcnt))
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))
764 head = &cg->bpf.progs[atype];
765 hlist_for_each_entry(pl, head, node) {
766 if (!prog_list_prog(pl))
768 if (pl->link == link)
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);
784 * __cgroup_bpf_replace() - Replace link's program and propagate the change
786 * @cgrp: The cgroup which descendants to traverse
787 * @link: A link for which to replace BPF program
788 * @type: Type of attach operation
790 * Must be called with cgroup_mutex held.
792 static int __cgroup_bpf_replace(struct cgroup *cgrp,
793 struct bpf_cgroup_link *link,
794 struct bpf_prog *new_prog)
796 enum cgroup_bpf_attach_type atype;
797 struct bpf_prog *old_prog;
798 struct bpf_prog_list *pl;
799 struct hlist_head *progs;
802 atype = bpf_cgroup_atype_find(link->type, new_prog->aux->attach_btf_id);
806 progs = &cgrp->bpf.progs[atype];
808 if (link->link.prog->type != new_prog->type)
811 hlist_for_each_entry(pl, progs, node) {
812 if (pl->link == link) {
820 old_prog = xchg(&link->link.prog, new_prog);
821 replace_effective_prog(cgrp, atype, link);
822 bpf_prog_put(old_prog);
826 static int cgroup_bpf_replace(struct bpf_link *link, struct bpf_prog *new_prog,
827 struct bpf_prog *old_prog)
829 struct bpf_cgroup_link *cg_link;
832 cg_link = container_of(link, struct bpf_cgroup_link, link);
835 /* link might have been auto-released by dying cgroup, so fail */
836 if (!cg_link->cgroup) {
840 if (old_prog && link->prog != old_prog) {
844 ret = __cgroup_bpf_replace(cg_link->cgroup, cg_link, new_prog);
850 static struct bpf_prog_list *find_detach_entry(struct hlist_head *progs,
851 struct bpf_prog *prog,
852 struct bpf_cgroup_link *link,
855 struct bpf_prog_list *pl;
858 if (hlist_empty(progs))
859 /* report error when trying to detach and nothing is attached */
860 return ERR_PTR(-ENOENT);
862 /* to maintain backward compatibility NONE and OVERRIDE cgroups
863 * allow detaching with invalid FD (prog==NULL) in legacy mode
865 return hlist_entry(progs->first, typeof(*pl), node);
869 /* to detach MULTI prog the user has to specify valid FD
870 * of the program or link to be detached
872 return ERR_PTR(-EINVAL);
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)
879 return ERR_PTR(-ENOENT);
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.
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
892 static void purge_effective_progs(struct cgroup *cgrp, struct bpf_prog *prog,
893 struct bpf_cgroup_link *link,
894 enum cgroup_bpf_attach_type atype)
896 struct cgroup_subsys_state *css;
897 struct bpf_prog_array *progs;
898 struct bpf_prog_list *pl;
899 struct hlist_head *head;
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);
907 if (percpu_ref_is_zero(&desc->bpf.refcnt))
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))
915 head = &cg->bpf.progs[atype];
916 hlist_for_each_entry(pl, head, node) {
917 if (!prog_list_prog(pl))
919 if (pl->prog == prog && pl->link == link)
925 /* no link or prog match, skip the cgroup of this layer */
928 progs = rcu_dereference_protected(
929 desc->bpf.effective[atype],
930 lockdep_is_held(&cgroup_mutex));
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);
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
946 * At most one of @prog or @link can be non-NULL.
947 * Must be called with cgroup_mutex held.
949 static int __cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
950 struct bpf_cgroup_link *link, enum bpf_attach_type type)
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;
960 attach_btf_id = prog->aux->attach_btf_id;
962 attach_btf_id = link->link.prog->aux->attach_btf_id;
964 atype = bpf_cgroup_atype_find(type, attach_btf_id);
968 progs = &cgrp->bpf.progs[atype];
969 flags = cgrp->bpf.flags[atype];
972 /* only one of prog or link can be specified */
975 pl = find_detach_entry(progs, prog, link, flags & BPF_F_ALLOW_MULTI);
979 /* mark it deleted, so it's ignored while recomputing effective */
984 if (update_effective_progs(cgrp, atype)) {
985 /* if update effective array failed replace the prog with a dummy prog*/
988 purge_effective_progs(cgrp, old_prog, link, atype);
991 /* now can actually delete it from this cgroup list */
992 hlist_del(&pl->node);
995 if (hlist_empty(progs))
996 /* last program was detached, reset flags to zero */
997 cgrp->bpf.flags[atype] = 0;
999 if (type == BPF_LSM_CGROUP)
1000 bpf_trampoline_unlink_cgroup_shim(old_prog);
1001 bpf_prog_put(old_prog);
1003 static_branch_dec(&cgroup_bpf_enabled_key[atype]);
1007 static int cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
1008 enum bpf_attach_type type)
1013 ret = __cgroup_bpf_detach(cgrp, prog, NULL, type);
1018 /* Must be called with cgroup_mutex held to avoid races. */
1019 static int __cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
1020 union bpf_attr __user *uattr)
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;
1033 if (effective_query && prog_attach_flags)
1036 if (type == BPF_LSM_CGROUP) {
1037 if (!effective_query && attr->query.prog_cnt &&
1038 prog_ids && !prog_attach_flags)
1041 from_atype = CGROUP_LSM_START;
1042 to_atype = CGROUP_LSM_END;
1045 from_atype = to_cgroup_bpf_attach_type(type);
1048 to_atype = from_atype;
1049 flags = cgrp->bpf.flags[from_atype];
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);
1058 total_cnt += prog_list_length(&cgrp->bpf.progs[atype]);
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)))
1066 if (copy_to_user(&uattr->query.prog_cnt, &total_cnt, sizeof(total_cnt)))
1068 if (attr->query.prog_cnt == 0 || !prog_ids || !total_cnt)
1069 /* return early if user requested only program count + flags */
1072 if (attr->query.prog_cnt < total_cnt) {
1073 total_cnt = attr->query.prog_cnt;
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);
1084 struct hlist_head *progs;
1085 struct bpf_prog_list *pl;
1086 struct bpf_prog *prog;
1089 progs = &cgrp->bpf.progs[atype];
1090 cnt = min_t(int, prog_list_length(progs), total_cnt);
1092 hlist_for_each_entry(pl, progs, node) {
1093 prog = prog_list_prog(pl);
1095 if (copy_to_user(prog_ids + i, &id, sizeof(id)))
1101 if (prog_attach_flags) {
1102 flags = cgrp->bpf.flags[atype];
1104 for (i = 0; i < cnt; i++)
1105 if (copy_to_user(prog_attach_flags + i,
1106 &flags, sizeof(flags)))
1108 prog_attach_flags += cnt;
1118 static int cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
1119 union bpf_attr __user *uattr)
1124 ret = __cgroup_bpf_query(cgrp, attr, uattr);
1129 int cgroup_bpf_prog_attach(const union bpf_attr *attr,
1130 enum bpf_prog_type ptype, struct bpf_prog *prog)
1132 struct bpf_prog *replace_prog = NULL;
1133 struct cgroup *cgrp;
1136 cgrp = cgroup_get_from_fd(attr->target_fd);
1138 return PTR_ERR(cgrp);
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)) {
1145 return PTR_ERR(replace_prog);
1149 ret = cgroup_bpf_attach(cgrp, prog, replace_prog, NULL,
1150 attr->attach_type, attr->attach_flags);
1153 bpf_prog_put(replace_prog);
1158 int cgroup_bpf_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
1160 struct bpf_prog *prog;
1161 struct cgroup *cgrp;
1164 cgrp = cgroup_get_from_fd(attr->target_fd);
1166 return PTR_ERR(cgrp);
1168 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1172 ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type);
1180 static void bpf_cgroup_link_release(struct bpf_link *link)
1182 struct bpf_cgroup_link *cg_link =
1183 container_of(link, struct bpf_cgroup_link, link);
1186 /* link might have been auto-detached by dying cgroup already,
1187 * in that case our work is done here
1189 if (!cg_link->cgroup)
1194 /* re-check cgroup under lock again */
1195 if (!cg_link->cgroup) {
1200 WARN_ON(__cgroup_bpf_detach(cg_link->cgroup, NULL, cg_link,
1202 if (cg_link->type == BPF_LSM_CGROUP)
1203 bpf_trampoline_unlink_cgroup_shim(cg_link->link.prog);
1205 cg = cg_link->cgroup;
1206 cg_link->cgroup = NULL;
1213 static void bpf_cgroup_link_dealloc(struct bpf_link *link)
1215 struct bpf_cgroup_link *cg_link =
1216 container_of(link, struct bpf_cgroup_link, link);
1221 static int bpf_cgroup_link_detach(struct bpf_link *link)
1223 bpf_cgroup_link_release(link);
1228 static void bpf_cgroup_link_show_fdinfo(const struct bpf_link *link,
1229 struct seq_file *seq)
1231 struct bpf_cgroup_link *cg_link =
1232 container_of(link, struct bpf_cgroup_link, link);
1236 if (cg_link->cgroup)
1237 cg_id = cgroup_id(cg_link->cgroup);
1241 "cgroup_id:\t%llu\n"
1242 "attach_type:\t%d\n",
1247 static int bpf_cgroup_link_fill_link_info(const struct bpf_link *link,
1248 struct bpf_link_info *info)
1250 struct bpf_cgroup_link *cg_link =
1251 container_of(link, struct bpf_cgroup_link, link);
1255 if (cg_link->cgroup)
1256 cg_id = cgroup_id(cg_link->cgroup);
1259 info->cgroup.cgroup_id = cg_id;
1260 info->cgroup.attach_type = cg_link->type;
1264 static 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,
1273 int cgroup_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
1275 struct bpf_link_primer link_primer;
1276 struct bpf_cgroup_link *link;
1277 struct cgroup *cgrp;
1280 if (attr->link_create.flags)
1283 cgrp = cgroup_get_from_fd(attr->link_create.target_fd);
1285 return PTR_ERR(cgrp);
1287 link = kzalloc(sizeof(*link), GFP_USER);
1290 goto out_put_cgroup;
1292 bpf_link_init(&link->link, BPF_LINK_TYPE_CGROUP, &bpf_cgroup_link_lops,
1294 link->cgroup = cgrp;
1295 link->type = attr->link_create.attach_type;
1297 err = bpf_link_prime(&link->link, &link_primer);
1300 goto out_put_cgroup;
1303 err = cgroup_bpf_attach(cgrp, NULL, NULL, link,
1304 link->type, BPF_F_ALLOW_MULTI);
1306 bpf_link_cleanup(&link_primer);
1307 goto out_put_cgroup;
1310 return bpf_link_settle(&link_primer);
1317 int cgroup_bpf_prog_query(const union bpf_attr *attr,
1318 union bpf_attr __user *uattr)
1320 struct cgroup *cgrp;
1323 cgrp = cgroup_get_from_fd(attr->query.target_fd);
1325 return PTR_ERR(cgrp);
1327 ret = cgroup_bpf_query(cgrp, attr, uattr);
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
1339 * If no socket is passed, or the socket is not of type INET or INET6,
1340 * this function does nothing and returns 0.
1342 * The program type passed in via @type must be suitable for network
1343 * filtering. No further check is performed to assert that.
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
1350 * -err - drop packet
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.
1356 int __cgroup_bpf_run_filter_skb(struct sock *sk,
1357 struct sk_buff *skb,
1358 enum cgroup_bpf_attach_type atype)
1360 unsigned int offset = skb->data - skb_network_header(skb);
1361 struct sock *save_sk;
1362 void *saved_data_end;
1363 struct cgroup *cgrp;
1366 if (!sk || !sk_fullsock(sk))
1369 if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)
1372 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1375 __skb_push(skb, offset);
1377 /* compute pointers for the bpf prog */
1378 bpf_compute_and_save_data_end(skb, &saved_data_end);
1380 if (atype == CGROUP_INET_EGRESS) {
1384 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, skb,
1385 __bpf_prog_run_save_cb, 0, &flags);
1387 /* Return values of CGROUP EGRESS BPF programs are:
1390 * 2: drop packet and cn
1391 * 3: keep packet and cn
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
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
1402 cn = flags & BPF_RET_SET_CN;
1403 if (ret && !IS_ERR_VALUE((long)ret))
1406 ret = (cn ? NET_XMIT_CN : NET_XMIT_SUCCESS);
1408 ret = (cn ? NET_XMIT_DROP : ret);
1410 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype,
1411 skb, __bpf_prog_run_save_cb, 0,
1413 if (ret && !IS_ERR_VALUE((long)ret))
1416 bpf_restore_data_end(skb, saved_data_end);
1417 __skb_pull(skb, offset);
1422 EXPORT_SYMBOL(__cgroup_bpf_run_filter_skb);
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
1429 * socket is passed is expected to be of type INET or INET6.
1431 * The program type passed in via @type must be suitable for sock
1432 * filtering. No further check is performed to assert that.
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.
1437 int __cgroup_bpf_run_filter_sk(struct sock *sk,
1438 enum cgroup_bpf_attach_type atype)
1440 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1442 return bpf_prog_run_array_cg(&cgrp->bpf, atype, sk, bpf_prog_run, 0,
1445 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sk);
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).
1457 * socket is expected to be of type INET or INET6.
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.
1462 int __cgroup_bpf_run_filter_sock_addr(struct sock *sk,
1463 struct sockaddr *uaddr,
1464 enum cgroup_bpf_attach_type atype,
1468 struct bpf_sock_addr_kern ctx = {
1473 struct sockaddr_storage unspec;
1474 struct cgroup *cgrp;
1476 /* Check socket family since not all sockets represent network
1477 * endpoint (e.g. AF_UNIX).
1479 if (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)
1483 memset(&unspec, 0, sizeof(unspec));
1484 ctx.uaddr = (struct sockaddr *)&unspec;
1487 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1488 return bpf_prog_run_array_cg(&cgrp->bpf, atype, &ctx, bpf_prog_run,
1491 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_addr);
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
1501 * socket passed is expected to be of type INET or INET6.
1503 * The program type passed in via @type must be suitable for sock_ops
1504 * filtering. No further check is performed to assert that.
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.
1509 int __cgroup_bpf_run_filter_sock_ops(struct sock *sk,
1510 struct bpf_sock_ops_kern *sock_ops,
1511 enum cgroup_bpf_attach_type atype)
1513 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1515 return bpf_prog_run_array_cg(&cgrp->bpf, atype, sock_ops, bpf_prog_run,
1518 EXPORT_SYMBOL(__cgroup_bpf_run_filter_sock_ops);
1520 int __cgroup_bpf_check_dev_permission(short dev_type, u32 major, u32 minor,
1521 short access, enum cgroup_bpf_attach_type atype)
1523 struct cgroup *cgrp;
1524 struct bpf_cgroup_dev_ctx ctx = {
1525 .access_type = (access << 16) | dev_type,
1532 cgrp = task_dfl_cgroup(current);
1533 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, &ctx, bpf_prog_run, 0,
1540 BPF_CALL_2(bpf_get_local_storage, struct bpf_map *, map, u64, flags)
1542 /* flags argument is not used now,
1543 * but provides an ability to extend the API.
1544 * verifier checks that its value is correct.
1546 enum bpf_cgroup_storage_type stype = cgroup_storage_type(map);
1547 struct bpf_cgroup_storage *storage;
1548 struct bpf_cg_run_ctx *ctx;
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];
1555 if (stype == BPF_CGROUP_STORAGE_SHARED)
1556 ptr = &READ_ONCE(storage->buf)->data[0];
1558 ptr = this_cpu_ptr(storage->percpu_buf);
1560 return (unsigned long)ptr;
1563 const struct bpf_func_proto bpf_get_local_storage_proto = {
1564 .func = bpf_get_local_storage,
1566 .ret_type = RET_PTR_TO_MAP_VALUE,
1567 .arg1_type = ARG_CONST_MAP_PTR,
1568 .arg2_type = ARG_ANYTHING,
1571 BPF_CALL_0(bpf_get_retval)
1573 struct bpf_cg_run_ctx *ctx =
1574 container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx);
1579 const struct bpf_func_proto bpf_get_retval_proto = {
1580 .func = bpf_get_retval,
1582 .ret_type = RET_INTEGER,
1585 BPF_CALL_1(bpf_set_retval, int, retval)
1587 struct bpf_cg_run_ctx *ctx =
1588 container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx);
1590 ctx->retval = retval;
1594 const struct bpf_func_proto bpf_set_retval_proto = {
1595 .func = bpf_set_retval,
1597 .ret_type = RET_INTEGER,
1598 .arg1_type = ARG_ANYTHING,
1601 static const struct bpf_func_proto *
1602 cgroup_dev_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1604 const struct bpf_func_proto *func_proto;
1606 func_proto = cgroup_common_func_proto(func_id, prog);
1610 func_proto = cgroup_current_func_proto(func_id, prog);
1615 case BPF_FUNC_perf_event_output:
1616 return &bpf_event_output_data_proto;
1618 return bpf_base_func_proto(func_id);
1622 static 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)
1627 const int size_default = sizeof(__u32);
1629 if (type == BPF_WRITE)
1632 if (off < 0 || off + size > sizeof(struct bpf_cgroup_dev_ctx))
1634 /* The verifier guarantees that size > 0. */
1635 if (off % size != 0)
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))
1645 if (size != size_default)
1652 const struct bpf_prog_ops cg_dev_prog_ops = {
1655 const 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,
1661 * __cgroup_bpf_run_filter_sysctl - Run a program on sysctl
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
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
1675 * Program is run when sysctl is being accessed, either read or written, and
1676 * can allow or deny such access.
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.
1681 int __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)
1686 struct bpf_sysctl_kern ctx = {
1692 .cur_len = PAGE_SIZE,
1697 struct cgroup *cgrp;
1701 ctx.cur_val = kmalloc_track_caller(ctx.cur_len, GFP_KERNEL);
1703 table->proc_handler(table, 0, ctx.cur_val, &ctx.cur_len, &pos)) {
1704 /* Let BPF program decide how to proceed. */
1708 if (write && *buf && *pcount) {
1709 /* BPF program should be able to override new value with a
1710 * buffer bigger than provided by user.
1712 ctx.new_val = kmalloc_track_caller(PAGE_SIZE, GFP_KERNEL);
1713 ctx.new_len = min_t(size_t, PAGE_SIZE, *pcount);
1715 memcpy(ctx.new_val, *buf, ctx.new_len);
1717 /* Let BPF program decide how to proceed. */
1723 cgrp = task_dfl_cgroup(current);
1724 ret = bpf_prog_run_array_cg(&cgrp->bpf, atype, &ctx, bpf_prog_run, 0,
1730 if (ret == 1 && ctx.new_updated) {
1733 *pcount = ctx.new_len;
1742 static int sockopt_alloc_buf(struct bpf_sockopt_kern *ctx, int max_optlen,
1743 struct bpf_sockopt_buf *buf)
1745 if (unlikely(max_optlen < 0))
1748 if (unlikely(max_optlen > PAGE_SIZE)) {
1749 /* We don't expose optvals that are greater than PAGE_SIZE
1750 * to the BPF program.
1752 max_optlen = PAGE_SIZE;
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.
1759 ctx->optval = buf->data;
1760 ctx->optval_end = ctx->optval + max_optlen;
1764 ctx->optval = kzalloc(max_optlen, GFP_USER);
1768 ctx->optval_end = ctx->optval + max_optlen;
1773 static void sockopt_free_buf(struct bpf_sockopt_kern *ctx,
1774 struct bpf_sockopt_buf *buf)
1776 if (ctx->optval == buf->data)
1781 static bool sockopt_buf_allocated(struct bpf_sockopt_kern *ctx,
1782 struct bpf_sockopt_buf *buf)
1784 return ctx->optval != buf->data;
1787 int __cgroup_bpf_run_filter_setsockopt(struct sock *sk, int *level,
1788 int *optname, char __user *optval,
1789 int *optlen, char **kernel_optval)
1791 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1792 struct bpf_sockopt_buf buf = {};
1793 struct bpf_sockopt_kern ctx = {
1796 .optname = *optname,
1798 int ret, max_optlen;
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).
1804 max_optlen = max_t(int, 16, *optlen);
1805 max_optlen = sockopt_alloc_buf(&ctx, max_optlen, &buf);
1809 ctx.optlen = *optlen;
1811 if (copy_from_user(ctx.optval, optval, min(*optlen, max_optlen)) != 0) {
1817 ret = bpf_prog_run_array_cg(&cgrp->bpf, CGROUP_SETSOCKOPT,
1818 &ctx, bpf_prog_run, 0, NULL);
1824 if (ctx.optlen == -1) {
1825 /* optlen set to -1, bypass kernel */
1827 } else if (ctx.optlen > max_optlen || ctx.optlen < -1) {
1828 /* optlen is out of bounds */
1829 if (*optlen > PAGE_SIZE && ctx.optlen >= 0) {
1830 pr_info_once("bpf setsockopt: ignoring program buffer with optlen=%d (max_optlen=%d)\n",
1831 ctx.optlen, max_optlen);
1837 /* optlen within bounds, run kernel handler */
1840 /* export any potential modifications */
1842 *optname = ctx.optname;
1844 /* optlen == 0 from BPF indicates that we should
1845 * use original userspace data.
1847 if (ctx.optlen != 0) {
1848 *optlen = ctx.optlen;
1849 /* We've used bpf_sockopt_kern->buf as an intermediary
1850 * storage, but the BPF program indicates that we need
1851 * to pass this data to the kernel setsockopt handler.
1852 * No way to export on-stack buf, have to allocate a
1855 if (!sockopt_buf_allocated(&ctx, &buf)) {
1856 void *p = kmalloc(ctx.optlen, GFP_USER);
1862 memcpy(p, ctx.optval, ctx.optlen);
1865 *kernel_optval = ctx.optval;
1867 /* export and don't free sockopt buf */
1873 sockopt_free_buf(&ctx, &buf);
1877 int __cgroup_bpf_run_filter_getsockopt(struct sock *sk, int level,
1878 int optname, char __user *optval,
1879 int __user *optlen, int max_optlen,
1882 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1883 struct bpf_sockopt_buf buf = {};
1884 struct bpf_sockopt_kern ctx = {
1888 .current_task = current,
1893 orig_optlen = max_optlen;
1894 ctx.optlen = max_optlen;
1895 max_optlen = sockopt_alloc_buf(&ctx, max_optlen, &buf);
1900 /* If kernel getsockopt finished successfully,
1901 * copy whatever was returned to the user back
1902 * into our temporary buffer. Set optlen to the
1903 * one that kernel returned as well to let
1904 * BPF programs inspect the value.
1907 if (get_user(ctx.optlen, optlen)) {
1912 if (ctx.optlen < 0) {
1916 orig_optlen = ctx.optlen;
1918 if (copy_from_user(ctx.optval, optval,
1919 min(ctx.optlen, max_optlen)) != 0) {
1926 ret = bpf_prog_run_array_cg(&cgrp->bpf, CGROUP_GETSOCKOPT,
1927 &ctx, bpf_prog_run, retval, NULL);
1933 if (optval && (ctx.optlen > max_optlen || ctx.optlen < 0)) {
1934 if (orig_optlen > PAGE_SIZE && ctx.optlen >= 0) {
1935 pr_info_once("bpf getsockopt: ignoring program buffer with optlen=%d (max_optlen=%d)\n",
1936 ctx.optlen, max_optlen);
1944 if (ctx.optlen != 0) {
1945 if (optval && copy_to_user(optval, ctx.optval, ctx.optlen)) {
1949 if (put_user(ctx.optlen, optlen)) {
1956 sockopt_free_buf(&ctx, &buf);
1960 int __cgroup_bpf_run_filter_getsockopt_kern(struct sock *sk, int level,
1961 int optname, void *optval,
1962 int *optlen, int retval)
1964 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
1965 struct bpf_sockopt_kern ctx = {
1971 .optval_end = optval + *optlen,
1972 .current_task = current,
1976 /* Note that __cgroup_bpf_run_filter_getsockopt doesn't copy
1977 * user data back into BPF buffer when reval != 0. This is
1978 * done as an optimization to avoid extra copy, assuming
1979 * kernel won't populate the data in case of an error.
1980 * Here we always pass the data and memset() should
1981 * be called if that data shouldn't be "exported".
1984 ret = bpf_prog_run_array_cg(&cgrp->bpf, CGROUP_GETSOCKOPT,
1985 &ctx, bpf_prog_run, retval, NULL);
1989 if (ctx.optlen > *optlen)
1992 /* BPF programs can shrink the buffer, export the modifications.
1994 if (ctx.optlen != 0)
1995 *optlen = ctx.optlen;
2001 static ssize_t sysctl_cpy_dir(const struct ctl_dir *dir, char **bufp,
2004 ssize_t tmp_ret = 0, ret;
2006 if (dir->header.parent) {
2007 tmp_ret = sysctl_cpy_dir(dir->header.parent, bufp, lenp);
2012 ret = strscpy(*bufp, dir->header.ctl_table[0].procname, *lenp);
2019 /* Avoid leading slash. */
2023 tmp_ret = strscpy(*bufp, "/", *lenp);
2029 return ret + tmp_ret;
2032 BPF_CALL_4(bpf_sysctl_get_name, struct bpf_sysctl_kern *, ctx, char *, buf,
2033 size_t, buf_len, u64, flags)
2035 ssize_t tmp_ret = 0, ret;
2040 if (!(flags & BPF_F_SYSCTL_BASE_NAME)) {
2043 tmp_ret = sysctl_cpy_dir(ctx->head->parent, &buf, &buf_len);
2048 ret = strscpy(buf, ctx->table->procname, buf_len);
2050 return ret < 0 ? ret : tmp_ret + ret;
2053 static const struct bpf_func_proto bpf_sysctl_get_name_proto = {
2054 .func = bpf_sysctl_get_name,
2056 .ret_type = RET_INTEGER,
2057 .arg1_type = ARG_PTR_TO_CTX,
2058 .arg2_type = ARG_PTR_TO_MEM,
2059 .arg3_type = ARG_CONST_SIZE,
2060 .arg4_type = ARG_ANYTHING,
2063 static int copy_sysctl_value(char *dst, size_t dst_len, char *src,
2072 if (!src || !src_len) {
2073 memset(dst, 0, dst_len);
2077 memcpy(dst, src, min(dst_len, src_len));
2079 if (dst_len > src_len) {
2080 memset(dst + src_len, '\0', dst_len - src_len);
2084 dst[dst_len - 1] = '\0';
2089 BPF_CALL_3(bpf_sysctl_get_current_value, struct bpf_sysctl_kern *, ctx,
2090 char *, buf, size_t, buf_len)
2092 return copy_sysctl_value(buf, buf_len, ctx->cur_val, ctx->cur_len);
2095 static const struct bpf_func_proto bpf_sysctl_get_current_value_proto = {
2096 .func = bpf_sysctl_get_current_value,
2098 .ret_type = RET_INTEGER,
2099 .arg1_type = ARG_PTR_TO_CTX,
2100 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
2101 .arg3_type = ARG_CONST_SIZE,
2104 BPF_CALL_3(bpf_sysctl_get_new_value, struct bpf_sysctl_kern *, ctx, char *, buf,
2109 memset(buf, '\0', buf_len);
2112 return copy_sysctl_value(buf, buf_len, ctx->new_val, ctx->new_len);
2115 static const struct bpf_func_proto bpf_sysctl_get_new_value_proto = {
2116 .func = bpf_sysctl_get_new_value,
2118 .ret_type = RET_INTEGER,
2119 .arg1_type = ARG_PTR_TO_CTX,
2120 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
2121 .arg3_type = ARG_CONST_SIZE,
2124 BPF_CALL_3(bpf_sysctl_set_new_value, struct bpf_sysctl_kern *, ctx,
2125 const char *, buf, size_t, buf_len)
2127 if (!ctx->write || !ctx->new_val || !ctx->new_len || !buf || !buf_len)
2130 if (buf_len > PAGE_SIZE - 1)
2133 memcpy(ctx->new_val, buf, buf_len);
2134 ctx->new_len = buf_len;
2135 ctx->new_updated = 1;
2140 static const struct bpf_func_proto bpf_sysctl_set_new_value_proto = {
2141 .func = bpf_sysctl_set_new_value,
2143 .ret_type = RET_INTEGER,
2144 .arg1_type = ARG_PTR_TO_CTX,
2145 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
2146 .arg3_type = ARG_CONST_SIZE,
2149 static const struct bpf_func_proto *
2150 sysctl_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
2152 const struct bpf_func_proto *func_proto;
2154 func_proto = cgroup_common_func_proto(func_id, prog);
2158 func_proto = cgroup_current_func_proto(func_id, prog);
2163 case BPF_FUNC_sysctl_get_name:
2164 return &bpf_sysctl_get_name_proto;
2165 case BPF_FUNC_sysctl_get_current_value:
2166 return &bpf_sysctl_get_current_value_proto;
2167 case BPF_FUNC_sysctl_get_new_value:
2168 return &bpf_sysctl_get_new_value_proto;
2169 case BPF_FUNC_sysctl_set_new_value:
2170 return &bpf_sysctl_set_new_value_proto;
2171 case BPF_FUNC_ktime_get_coarse_ns:
2172 return &bpf_ktime_get_coarse_ns_proto;
2173 case BPF_FUNC_perf_event_output:
2174 return &bpf_event_output_data_proto;
2176 return bpf_base_func_proto(func_id);
2180 static bool sysctl_is_valid_access(int off, int size, enum bpf_access_type type,
2181 const struct bpf_prog *prog,
2182 struct bpf_insn_access_aux *info)
2184 const int size_default = sizeof(__u32);
2186 if (off < 0 || off + size > sizeof(struct bpf_sysctl) || off % size)
2190 case bpf_ctx_range(struct bpf_sysctl, write):
2191 if (type != BPF_READ)
2193 bpf_ctx_record_field_size(info, size_default);
2194 return bpf_ctx_narrow_access_ok(off, size, size_default);
2195 case bpf_ctx_range(struct bpf_sysctl, file_pos):
2196 if (type == BPF_READ) {
2197 bpf_ctx_record_field_size(info, size_default);
2198 return bpf_ctx_narrow_access_ok(off, size, size_default);
2200 return size == size_default;
2207 static u32 sysctl_convert_ctx_access(enum bpf_access_type type,
2208 const struct bpf_insn *si,
2209 struct bpf_insn *insn_buf,
2210 struct bpf_prog *prog, u32 *target_size)
2212 struct bpf_insn *insn = insn_buf;
2216 case offsetof(struct bpf_sysctl, write):
2217 *insn++ = BPF_LDX_MEM(
2218 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
2219 bpf_target_off(struct bpf_sysctl_kern, write,
2220 sizeof_field(struct bpf_sysctl_kern,
2224 case offsetof(struct bpf_sysctl, file_pos):
2225 /* ppos is a pointer so it should be accessed via indirect
2226 * loads and stores. Also for stores additional temporary
2227 * register is used since neither src_reg nor dst_reg can be
2230 if (type == BPF_WRITE) {
2231 int treg = BPF_REG_9;
2233 if (si->src_reg == treg || si->dst_reg == treg)
2235 if (si->src_reg == treg || si->dst_reg == treg)
2237 *insn++ = BPF_STX_MEM(
2238 BPF_DW, si->dst_reg, treg,
2239 offsetof(struct bpf_sysctl_kern, tmp_reg));
2240 *insn++ = BPF_LDX_MEM(
2241 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos),
2243 offsetof(struct bpf_sysctl_kern, ppos));
2244 *insn++ = BPF_RAW_INSN(
2245 BPF_CLASS(si->code) | BPF_MEM | BPF_SIZEOF(u32),
2247 bpf_ctx_narrow_access_offset(
2248 0, sizeof(u32), sizeof(loff_t)),
2250 *insn++ = BPF_LDX_MEM(
2251 BPF_DW, treg, si->dst_reg,
2252 offsetof(struct bpf_sysctl_kern, tmp_reg));
2254 *insn++ = BPF_LDX_MEM(
2255 BPF_FIELD_SIZEOF(struct bpf_sysctl_kern, ppos),
2256 si->dst_reg, si->src_reg,
2257 offsetof(struct bpf_sysctl_kern, ppos));
2258 read_size = bpf_size_to_bytes(BPF_SIZE(si->code));
2259 *insn++ = BPF_LDX_MEM(
2260 BPF_SIZE(si->code), si->dst_reg, si->dst_reg,
2261 bpf_ctx_narrow_access_offset(
2262 0, read_size, sizeof(loff_t)));
2264 *target_size = sizeof(u32);
2268 return insn - insn_buf;
2271 const struct bpf_verifier_ops cg_sysctl_verifier_ops = {
2272 .get_func_proto = sysctl_func_proto,
2273 .is_valid_access = sysctl_is_valid_access,
2274 .convert_ctx_access = sysctl_convert_ctx_access,
2277 const struct bpf_prog_ops cg_sysctl_prog_ops = {
2281 BPF_CALL_1(bpf_get_netns_cookie_sockopt, struct bpf_sockopt_kern *, ctx)
2283 const struct net *net = ctx ? sock_net(ctx->sk) : &init_net;
2285 return net->net_cookie;
2288 static const struct bpf_func_proto bpf_get_netns_cookie_sockopt_proto = {
2289 .func = bpf_get_netns_cookie_sockopt,
2291 .ret_type = RET_INTEGER,
2292 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
2296 static const struct bpf_func_proto *
2297 cg_sockopt_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
2299 const struct bpf_func_proto *func_proto;
2301 func_proto = cgroup_common_func_proto(func_id, prog);
2305 func_proto = cgroup_current_func_proto(func_id, prog);
2311 case BPF_FUNC_get_netns_cookie:
2312 return &bpf_get_netns_cookie_sockopt_proto;
2313 case BPF_FUNC_sk_storage_get:
2314 return &bpf_sk_storage_get_proto;
2315 case BPF_FUNC_sk_storage_delete:
2316 return &bpf_sk_storage_delete_proto;
2317 case BPF_FUNC_setsockopt:
2318 if (prog->expected_attach_type == BPF_CGROUP_SETSOCKOPT)
2319 return &bpf_sk_setsockopt_proto;
2321 case BPF_FUNC_getsockopt:
2322 if (prog->expected_attach_type == BPF_CGROUP_SETSOCKOPT)
2323 return &bpf_sk_getsockopt_proto;
2327 case BPF_FUNC_tcp_sock:
2328 return &bpf_tcp_sock_proto;
2330 case BPF_FUNC_perf_event_output:
2331 return &bpf_event_output_data_proto;
2333 return bpf_base_func_proto(func_id);
2337 static bool cg_sockopt_is_valid_access(int off, int size,
2338 enum bpf_access_type type,
2339 const struct bpf_prog *prog,
2340 struct bpf_insn_access_aux *info)
2342 const int size_default = sizeof(__u32);
2344 if (off < 0 || off >= sizeof(struct bpf_sockopt))
2347 if (off % size != 0)
2350 if (type == BPF_WRITE) {
2352 case offsetof(struct bpf_sockopt, retval):
2353 if (size != size_default)
2355 return prog->expected_attach_type ==
2356 BPF_CGROUP_GETSOCKOPT;
2357 case offsetof(struct bpf_sockopt, optname):
2359 case offsetof(struct bpf_sockopt, level):
2360 if (size != size_default)
2362 return prog->expected_attach_type ==
2363 BPF_CGROUP_SETSOCKOPT;
2364 case offsetof(struct bpf_sockopt, optlen):
2365 return size == size_default;
2372 case offsetof(struct bpf_sockopt, sk):
2373 if (size != sizeof(__u64))
2375 info->reg_type = PTR_TO_SOCKET;
2377 case offsetof(struct bpf_sockopt, optval):
2378 if (size != sizeof(__u64))
2380 info->reg_type = PTR_TO_PACKET;
2382 case offsetof(struct bpf_sockopt, optval_end):
2383 if (size != sizeof(__u64))
2385 info->reg_type = PTR_TO_PACKET_END;
2387 case offsetof(struct bpf_sockopt, retval):
2388 if (size != size_default)
2390 return prog->expected_attach_type == BPF_CGROUP_GETSOCKOPT;
2392 if (size != size_default)
2399 #define CG_SOCKOPT_READ_FIELD(F) \
2400 BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, F), \
2401 si->dst_reg, si->src_reg, \
2402 offsetof(struct bpf_sockopt_kern, F))
2404 #define CG_SOCKOPT_WRITE_FIELD(F) \
2405 BPF_RAW_INSN((BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, F) | \
2406 BPF_MEM | BPF_CLASS(si->code)), \
2407 si->dst_reg, si->src_reg, \
2408 offsetof(struct bpf_sockopt_kern, F), \
2411 static u32 cg_sockopt_convert_ctx_access(enum bpf_access_type type,
2412 const struct bpf_insn *si,
2413 struct bpf_insn *insn_buf,
2414 struct bpf_prog *prog,
2417 struct bpf_insn *insn = insn_buf;
2420 case offsetof(struct bpf_sockopt, sk):
2421 *insn++ = CG_SOCKOPT_READ_FIELD(sk);
2423 case offsetof(struct bpf_sockopt, level):
2424 if (type == BPF_WRITE)
2425 *insn++ = CG_SOCKOPT_WRITE_FIELD(level);
2427 *insn++ = CG_SOCKOPT_READ_FIELD(level);
2429 case offsetof(struct bpf_sockopt, optname):
2430 if (type == BPF_WRITE)
2431 *insn++ = CG_SOCKOPT_WRITE_FIELD(optname);
2433 *insn++ = CG_SOCKOPT_READ_FIELD(optname);
2435 case offsetof(struct bpf_sockopt, optlen):
2436 if (type == BPF_WRITE)
2437 *insn++ = CG_SOCKOPT_WRITE_FIELD(optlen);
2439 *insn++ = CG_SOCKOPT_READ_FIELD(optlen);
2441 case offsetof(struct bpf_sockopt, retval):
2442 BUILD_BUG_ON(offsetof(struct bpf_cg_run_ctx, run_ctx) != 0);
2444 if (type == BPF_WRITE) {
2445 int treg = BPF_REG_9;
2447 if (si->src_reg == treg || si->dst_reg == treg)
2449 if (si->src_reg == treg || si->dst_reg == treg)
2451 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, treg,
2452 offsetof(struct bpf_sockopt_kern, tmp_reg));
2453 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, current_task),
2455 offsetof(struct bpf_sockopt_kern, current_task));
2456 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct task_struct, bpf_ctx),
2458 offsetof(struct task_struct, bpf_ctx));
2459 *insn++ = BPF_RAW_INSN(BPF_CLASS(si->code) | BPF_MEM |
2460 BPF_FIELD_SIZEOF(struct bpf_cg_run_ctx, retval),
2462 offsetof(struct bpf_cg_run_ctx, retval),
2464 *insn++ = BPF_LDX_MEM(BPF_DW, treg, si->dst_reg,
2465 offsetof(struct bpf_sockopt_kern, tmp_reg));
2467 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sockopt_kern, current_task),
2468 si->dst_reg, si->src_reg,
2469 offsetof(struct bpf_sockopt_kern, current_task));
2470 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct task_struct, bpf_ctx),
2471 si->dst_reg, si->dst_reg,
2472 offsetof(struct task_struct, bpf_ctx));
2473 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_cg_run_ctx, retval),
2474 si->dst_reg, si->dst_reg,
2475 offsetof(struct bpf_cg_run_ctx, retval));
2478 case offsetof(struct bpf_sockopt, optval):
2479 *insn++ = CG_SOCKOPT_READ_FIELD(optval);
2481 case offsetof(struct bpf_sockopt, optval_end):
2482 *insn++ = CG_SOCKOPT_READ_FIELD(optval_end);
2486 return insn - insn_buf;
2489 static int cg_sockopt_get_prologue(struct bpf_insn *insn_buf,
2491 const struct bpf_prog *prog)
2493 /* Nothing to do for sockopt argument. The data is kzalloc'ated.
2498 const struct bpf_verifier_ops cg_sockopt_verifier_ops = {
2499 .get_func_proto = cg_sockopt_func_proto,
2500 .is_valid_access = cg_sockopt_is_valid_access,
2501 .convert_ctx_access = cg_sockopt_convert_ctx_access,
2502 .gen_prologue = cg_sockopt_get_prologue,
2505 const struct bpf_prog_ops cg_sockopt_prog_ops = {
2508 /* Common helpers for cgroup hooks. */
2509 const struct bpf_func_proto *
2510 cgroup_common_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
2513 case BPF_FUNC_get_local_storage:
2514 return &bpf_get_local_storage_proto;
2515 case BPF_FUNC_get_retval:
2516 switch (prog->expected_attach_type) {
2517 case BPF_CGROUP_INET_INGRESS:
2518 case BPF_CGROUP_INET_EGRESS:
2519 case BPF_CGROUP_SOCK_OPS:
2520 case BPF_CGROUP_UDP4_RECVMSG:
2521 case BPF_CGROUP_UDP6_RECVMSG:
2522 case BPF_CGROUP_INET4_GETPEERNAME:
2523 case BPF_CGROUP_INET6_GETPEERNAME:
2524 case BPF_CGROUP_INET4_GETSOCKNAME:
2525 case BPF_CGROUP_INET6_GETSOCKNAME:
2528 return &bpf_get_retval_proto;
2530 case BPF_FUNC_set_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_INET4_GETPEERNAME:
2538 case BPF_CGROUP_INET6_GETPEERNAME:
2539 case BPF_CGROUP_INET4_GETSOCKNAME:
2540 case BPF_CGROUP_INET6_GETSOCKNAME:
2543 return &bpf_set_retval_proto;
2550 /* Common helpers for cgroup hooks with valid process context. */
2551 const struct bpf_func_proto *
2552 cgroup_current_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
2555 case BPF_FUNC_get_current_uid_gid:
2556 return &bpf_get_current_uid_gid_proto;
2557 case BPF_FUNC_get_current_pid_tgid:
2558 return &bpf_get_current_pid_tgid_proto;
2559 case BPF_FUNC_get_current_comm:
2560 return &bpf_get_current_comm_proto;
2561 #ifdef CONFIG_CGROUP_NET_CLASSID
2562 case BPF_FUNC_get_cgroup_classid:
2563 return &bpf_get_cgroup_classid_curr_proto;