1	// SPDX-License-Identifier: GPL-2.0-only
2	/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3	*/
4	#include <linux/bpf.h>
5	#include <linux/btf.h>
6	#include <linux/bpf-cgroup.h>
7	#include <linux/cgroup.h>
8	#include <linux/rcupdate.h>
9	#include <linux/random.h>
10	#include <linux/smp.h>
11	#include <linux/topology.h>
12	#include <linux/ktime.h>
13	#include <linux/sched.h>
14	#include <linux/uidgid.h>
15	#include <linux/filter.h>
16	#include <linux/ctype.h>
17	#include <linux/jiffies.h>
18	#include <linux/pid_namespace.h>
19	#include <linux/poison.h>
20	#include <linux/proc_ns.h>
21	#include <linux/sched/task.h>
22	#include <linux/security.h>
23	#include <linux/btf_ids.h>
24	#include <linux/bpf_mem_alloc.h>
25	#include <linux/kasan.h>
26
27	#include "../../lib/kstrtox.h"
28
29	/* If kernel subsystem is allowing eBPF programs to call this function,
30	* inside its own verifier_ops->get_func_proto() callback it should return
31	* bpf_map_lookup_elem_proto, so that verifier can properly check the arguments
32	*
33	* Different map implementations will rely on rcu in map methods
34	* lookup/update/delete, therefore eBPF programs must run under rcu lock
35	* if program is allowed to access maps, so check rcu_read_lock_held() or
36	* rcu_read_lock_trace_held() in all three functions.
37	*/
38	BPF_CALL_2(bpf_map_lookup_elem, struct bpf_map *, map, void *, key)
39	{
40	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
41	!rcu_read_lock_bh_held());
42	return (unsigned long) map->ops->map_lookup_elem(map, key);
43	}
44
45	const struct bpf_func_proto bpf_map_lookup_elem_proto = {
46	.func = bpf_map_lookup_elem,
47	.gpl_only = false,
48	.pkt_access = true,
49	.ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
50	.arg1_type = ARG_CONST_MAP_PTR,
51	.arg2_type = ARG_PTR_TO_MAP_KEY,
52	};
53
54	BPF_CALL_4(bpf_map_update_elem, struct bpf_map *, map, void *, key,
55	void *, value, u64, flags)
56	{
57	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
58	!rcu_read_lock_bh_held());
59	return map->ops->map_update_elem(map, key, value, flags);
60	}
61
62	const struct bpf_func_proto bpf_map_update_elem_proto = {
63	.func = bpf_map_update_elem,
64	.gpl_only = false,
65	.pkt_access = true,
66	.ret_type = RET_INTEGER,
67	.arg1_type = ARG_CONST_MAP_PTR,
68	.arg2_type = ARG_PTR_TO_MAP_KEY,
69	.arg3_type = ARG_PTR_TO_MAP_VALUE,
70	.arg4_type = ARG_ANYTHING,
71	};
72
73	BPF_CALL_2(bpf_map_delete_elem, struct bpf_map *, map, void *, key)
74	{
75	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
76	!rcu_read_lock_bh_held());
77	return map->ops->map_delete_elem(map, key);
78	}
79
80	const struct bpf_func_proto bpf_map_delete_elem_proto = {
81	.func = bpf_map_delete_elem,
82	.gpl_only = false,
83	.pkt_access = true,
84	.ret_type = RET_INTEGER,
85	.arg1_type = ARG_CONST_MAP_PTR,
86	.arg2_type = ARG_PTR_TO_MAP_KEY,
87	};
88
89	BPF_CALL_3(bpf_map_push_elem, struct bpf_map *, map, void *, value, u64, flags)
90	{
91	return map->ops->map_push_elem(map, value, flags);
92	}
93
94	const struct bpf_func_proto bpf_map_push_elem_proto = {
95	.func = bpf_map_push_elem,
96	.gpl_only = false,
97	.pkt_access = true,
98	.ret_type = RET_INTEGER,
99	.arg1_type = ARG_CONST_MAP_PTR,
100	.arg2_type = ARG_PTR_TO_MAP_VALUE,
101	.arg3_type = ARG_ANYTHING,
102	};
103
104	BPF_CALL_2(bpf_map_pop_elem, struct bpf_map *, map, void *, value)
105	{
106	return map->ops->map_pop_elem(map, value);
107	}
108
109	const struct bpf_func_proto bpf_map_pop_elem_proto = {
110	.func = bpf_map_pop_elem,
111	.gpl_only = false,
112	.ret_type = RET_INTEGER,
113	.arg1_type = ARG_CONST_MAP_PTR,
114	.arg2_type = ARG_PTR_TO_MAP_VALUE | MEM_UNINIT,
115	};
116
117	BPF_CALL_2(bpf_map_peek_elem, struct bpf_map *, map, void *, value)
118	{
119	return map->ops->map_peek_elem(map, value);
120	}
121
122	const struct bpf_func_proto bpf_map_peek_elem_proto = {
123	.func = bpf_map_peek_elem,
124	.gpl_only = false,
125	.ret_type = RET_INTEGER,
126	.arg1_type = ARG_CONST_MAP_PTR,
127	.arg2_type = ARG_PTR_TO_MAP_VALUE | MEM_UNINIT,
128	};
129
130	BPF_CALL_3(bpf_map_lookup_percpu_elem, struct bpf_map *, map, void *, key, u32, cpu)
131	{
132	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
133	return (unsigned long) map->ops->map_lookup_percpu_elem(map, key, cpu);
134	}
135
136	const struct bpf_func_proto bpf_map_lookup_percpu_elem_proto = {
137	.func = bpf_map_lookup_percpu_elem,
138	.gpl_only = false,
139	.pkt_access = true,
140	.ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
141	.arg1_type = ARG_CONST_MAP_PTR,
142	.arg2_type = ARG_PTR_TO_MAP_KEY,
143	.arg3_type = ARG_ANYTHING,
144	};
145
146	const struct bpf_func_proto bpf_get_prandom_u32_proto = {
147	.func = bpf_user_rnd_u32,
148	.gpl_only = false,
149	.ret_type = RET_INTEGER,
150	};
151
152	BPF_CALL_0(bpf_get_smp_processor_id)
153	{
154	return smp_processor_id();
155	}
156
157	const struct bpf_func_proto bpf_get_smp_processor_id_proto = {
158	.func = bpf_get_smp_processor_id,
159	.gpl_only = false,
160	.ret_type = RET_INTEGER,
161	};
162
163	BPF_CALL_0(bpf_get_numa_node_id)
164	{
165	return numa_node_id();
166	}
167
168	const struct bpf_func_proto bpf_get_numa_node_id_proto = {
169	.func = bpf_get_numa_node_id,
170	.gpl_only = false,
171	.ret_type = RET_INTEGER,
172	};
173
174	BPF_CALL_0(bpf_ktime_get_ns)
175	{
176	/* NMI safe access to clock monotonic */
177	return ktime_get_mono_fast_ns();
178	}
179
180	const struct bpf_func_proto bpf_ktime_get_ns_proto = {
181	.func = bpf_ktime_get_ns,
182	.gpl_only = false,
183	.ret_type = RET_INTEGER,
184	};
185
186	BPF_CALL_0(bpf_ktime_get_boot_ns)
187	{
188	/* NMI safe access to clock boottime */
189	return ktime_get_boot_fast_ns();
190	}
191
192	const struct bpf_func_proto bpf_ktime_get_boot_ns_proto = {
193	.func = bpf_ktime_get_boot_ns,
194	.gpl_only = false,
195	.ret_type = RET_INTEGER,
196	};
197
198	BPF_CALL_0(bpf_ktime_get_coarse_ns)
199	{
200	return ktime_get_coarse_ns();
201	}
202
203	const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto = {
204	.func = bpf_ktime_get_coarse_ns,
205	.gpl_only = false,
206	.ret_type = RET_INTEGER,
207	};
208
209	BPF_CALL_0(bpf_ktime_get_tai_ns)
210	{
211	/* NMI safe access to clock tai */
212	return ktime_get_tai_fast_ns();
213	}
214
215	const struct bpf_func_proto bpf_ktime_get_tai_ns_proto = {
216	.func = bpf_ktime_get_tai_ns,
217	.gpl_only = false,
218	.ret_type = RET_INTEGER,
219	};
220
221	BPF_CALL_0(bpf_get_current_pid_tgid)
222	{
223	struct task_struct *task = current;
224
225	if (unlikely(!task))
226	return -EINVAL;
227
228	return (u64) task->tgid << 32 | task->pid;
229	}
230
231	const struct bpf_func_proto bpf_get_current_pid_tgid_proto = {
232	.func = bpf_get_current_pid_tgid,
233	.gpl_only = false,
234	.ret_type = RET_INTEGER,
235	};
236
237	BPF_CALL_0(bpf_get_current_uid_gid)
238	{
239	struct task_struct *task = current;
240	kuid_t uid;
241	kgid_t gid;
242
243	if (unlikely(!task))
244	return -EINVAL;
245
246	current_uid_gid(&uid, &gid);
247	return (u64) from_kgid(to: &init_user_ns, gid) << 32 |
248	from_kuid(to: &init_user_ns, uid);
249	}
250
251	const struct bpf_func_proto bpf_get_current_uid_gid_proto = {
252	.func = bpf_get_current_uid_gid,
253	.gpl_only = false,
254	.ret_type = RET_INTEGER,
255	};
256
257	BPF_CALL_2(bpf_get_current_comm, char *, buf, u32, size)
258	{
259	struct task_struct *task = current;
260
261	if (unlikely(!task))
262	goto err_clear;
263
264	/* Verifier guarantees that size > 0 */
265	strscpy_pad(buf, task->comm, size);
266	return 0;
267	err_clear:
268	memset(buf, 0, size);
269	return -EINVAL;
270	}
271
272	const struct bpf_func_proto bpf_get_current_comm_proto = {
273	.func = bpf_get_current_comm,
274	.gpl_only = false,
275	.ret_type = RET_INTEGER,
276	.arg1_type = ARG_PTR_TO_UNINIT_MEM,
277	.arg2_type = ARG_CONST_SIZE,
278	};
279
280	#if defined(CONFIG_QUEUED_SPINLOCKS) || defined(CONFIG_BPF_ARCH_SPINLOCK)
281
282	static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
283	{
284	arch_spinlock_t *l = (void *)lock;
285	union {
286	__u32 val;
287	arch_spinlock_t lock;
288	} u = { .lock = __ARCH_SPIN_LOCK_UNLOCKED };
289
290	compiletime_assert(u.val == 0, "__ARCH_SPIN_LOCK_UNLOCKED not 0");
291	BUILD_BUG_ON(sizeof(*l) != sizeof(__u32));
292	BUILD_BUG_ON(sizeof(*lock) != sizeof(__u32));
293	preempt_disable();
294	arch_spin_lock(l);
295	}
296
297	static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
298	{
299	arch_spinlock_t *l = (void *)lock;
300
301	arch_spin_unlock(l);
302	preempt_enable();
303	}
304
305	#else
306
307	static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
308	{
309	atomic_t *l = (void *)lock;
310
311	BUILD_BUG_ON(sizeof(*l) != sizeof(*lock));
312	do {
313	atomic_cond_read_relaxed(l, !VAL);
314	} while (atomic_xchg(l, 1));
315	}
316
317	static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
318	{
319	atomic_t *l = (void *)lock;
320
321	atomic_set_release(l, 0);
322	}
323
324	#endif
325
326	static DEFINE_PER_CPU(unsigned long, irqsave_flags);
327
328	static inline void __bpf_spin_lock_irqsave(struct bpf_spin_lock *lock)
329	{
330	unsigned long flags;
331
332	local_irq_save(flags);
333	__bpf_spin_lock(lock);
334	__this_cpu_write(irqsave_flags, flags);
335	}
336
337	NOTRACE_BPF_CALL_1(bpf_spin_lock, struct bpf_spin_lock *, lock)
338	{
339	__bpf_spin_lock_irqsave(lock);
340	return 0;
341	}
342
343	const struct bpf_func_proto bpf_spin_lock_proto = {
344	.func = bpf_spin_lock,
345	.gpl_only = false,
346	.ret_type = RET_VOID,
347	.arg1_type = ARG_PTR_TO_SPIN_LOCK,
348	.arg1_btf_id = BPF_PTR_POISON,
349	};
350
351	static inline void __bpf_spin_unlock_irqrestore(struct bpf_spin_lock *lock)
352	{
353	unsigned long flags;
354
355	flags = __this_cpu_read(irqsave_flags);
356	__bpf_spin_unlock(lock);
357	local_irq_restore(flags);
358	}
359
360	NOTRACE_BPF_CALL_1(bpf_spin_unlock, struct bpf_spin_lock *, lock)
361	{
362	__bpf_spin_unlock_irqrestore(lock);
363	return 0;
364	}
365
366	const struct bpf_func_proto bpf_spin_unlock_proto = {
367	.func = bpf_spin_unlock,
368	.gpl_only = false,
369	.ret_type = RET_VOID,
370	.arg1_type = ARG_PTR_TO_SPIN_LOCK,
371	.arg1_btf_id = BPF_PTR_POISON,
372	};
373
374	void copy_map_value_locked(struct bpf_map *map, void *dst, void *src,
375	bool lock_src)
376	{
377	struct bpf_spin_lock *lock;
378
379	if (lock_src)
380	lock = src + map->record->spin_lock_off;
381	else
382	lock = dst + map->record->spin_lock_off;
383	preempt_disable();
384	__bpf_spin_lock_irqsave(lock);
385	copy_map_value(map, dst, src);
386	__bpf_spin_unlock_irqrestore(lock);
387	preempt_enable();
388	}
389
390	BPF_CALL_0(bpf_jiffies64)
391	{
392	return get_jiffies_64();
393	}
394
395	const struct bpf_func_proto bpf_jiffies64_proto = {
396	.func = bpf_jiffies64,
397	.gpl_only = false,
398	.ret_type = RET_INTEGER,
399	};
400
401	#ifdef CONFIG_CGROUPS
402	BPF_CALL_0(bpf_get_current_cgroup_id)
403	{
404	struct cgroup *cgrp;
405	u64 cgrp_id;
406
407	rcu_read_lock();
408	cgrp = task_dfl_cgroup(current);
409	cgrp_id = cgroup_id(cgrp);
410	rcu_read_unlock();
411
412	return cgrp_id;
413	}
414
415	const struct bpf_func_proto bpf_get_current_cgroup_id_proto = {
416	.func = bpf_get_current_cgroup_id,
417	.gpl_only = false,
418	.ret_type = RET_INTEGER,
419	};
420
421	BPF_CALL_1(bpf_get_current_ancestor_cgroup_id, int, ancestor_level)
422	{
423	struct cgroup *cgrp;
424	struct cgroup *ancestor;
425	u64 cgrp_id;
426
427	rcu_read_lock();
428	cgrp = task_dfl_cgroup(current);
429	ancestor = cgroup_ancestor(cgrp, ancestor_level);
430	cgrp_id = ancestor ? cgroup_id(cgrp: ancestor) : 0;
431	rcu_read_unlock();
432
433	return cgrp_id;
434	}
435
436	const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto = {
437	.func = bpf_get_current_ancestor_cgroup_id,
438	.gpl_only = false,
439	.ret_type = RET_INTEGER,
440	.arg1_type = ARG_ANYTHING,
441	};
442	#endif /* CONFIG_CGROUPS */
443
444	#define BPF_STRTOX_BASE_MASK 0x1F
445
446	static int __bpf_strtoull(const char *buf, size_t buf_len, u64 flags,
447	unsigned long long *res, bool *is_negative)
448	{
449	unsigned int base = flags & BPF_STRTOX_BASE_MASK;
450	const char *cur_buf = buf;
451	size_t cur_len = buf_len;
452	unsigned int consumed;
453	size_t val_len;
454	char str[64];
455
456	if (!buf || !buf_len || !res || !is_negative)
457	return -EINVAL;
458
459	if (base != 0 && base != 8 && base != 10 && base != 16)
460	return -EINVAL;
461
462	if (flags & ~BPF_STRTOX_BASE_MASK)
463	return -EINVAL;
464
465	while (cur_buf < buf + buf_len && isspace(*cur_buf))
466	++cur_buf;
467
468	*is_negative = (cur_buf < buf + buf_len && *cur_buf == '-');
469	if (*is_negative)
470	++cur_buf;
471
472	consumed = cur_buf - buf;
473	cur_len -= consumed;
474	if (!cur_len)
475	return -EINVAL;
476
477	cur_len = min(cur_len, sizeof(str) - 1);
478	memcpy(str, cur_buf, cur_len);
479	str[cur_len] = '\0';
480	cur_buf = str;
481
482	cur_buf = _parse_integer_fixup_radix(s: cur_buf, base: &base);
483	val_len = _parse_integer(s: cur_buf, base, res);
484
485	if (val_len & KSTRTOX_OVERFLOW)
486	return -ERANGE;
487
488	if (val_len == 0)
489	return -EINVAL;
490
491	cur_buf += val_len;
492	consumed += cur_buf - str;
493
494	return consumed;
495	}
496
497	static int __bpf_strtoll(const char *buf, size_t buf_len, u64 flags,
498	long long *res)
499	{
500	unsigned long long _res;
501	bool is_negative;
502	int err;
503
504	err = __bpf_strtoull(buf, buf_len, flags, res: &_res, is_negative: &is_negative);
505	if (err < 0)
506	return err;
507	if (is_negative) {
508	if ((long long)-_res > 0)
509	return -ERANGE;
510	*res = -_res;
511	} else {
512	if ((long long)_res < 0)
513	return -ERANGE;
514	*res = _res;
515	}
516	return err;
517	}
518
519	BPF_CALL_4(bpf_strtol, const char *, buf, size_t, buf_len, u64, flags,
520	long *, res)
521	{
522	long long _res;
523	int err;
524
525	err = __bpf_strtoll(buf, buf_len, flags, res: &_res);
526	if (err < 0)
527	return err;
528	if (_res != (long)_res)
529	return -ERANGE;
530	*res = _res;
531	return err;
532	}
533
534	const struct bpf_func_proto bpf_strtol_proto = {
535	.func = bpf_strtol,
536	.gpl_only = false,
537	.ret_type = RET_INTEGER,
538	.arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY,
539	.arg2_type = ARG_CONST_SIZE,
540	.arg3_type = ARG_ANYTHING,
541	.arg4_type = ARG_PTR_TO_LONG,
542	};
543
544	BPF_CALL_4(bpf_strtoul, const char *, buf, size_t, buf_len, u64, flags,
545	unsigned long *, res)
546	{
547	unsigned long long _res;
548	bool is_negative;
549	int err;
550
551	err = __bpf_strtoull(buf, buf_len, flags, res: &_res, is_negative: &is_negative);
552	if (err < 0)
553	return err;
554	if (is_negative)
555	return -EINVAL;
556	if (_res != (unsigned long)_res)
557	return -ERANGE;
558	*res = _res;
559	return err;
560	}
561
562	const struct bpf_func_proto bpf_strtoul_proto = {
563	.func = bpf_strtoul,
564	.gpl_only = false,
565	.ret_type = RET_INTEGER,
566	.arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY,
567	.arg2_type = ARG_CONST_SIZE,
568	.arg3_type = ARG_ANYTHING,
569	.arg4_type = ARG_PTR_TO_LONG,
570	};
571
572	BPF_CALL_3(bpf_strncmp, const char *, s1, u32, s1_sz, const char *, s2)
573	{
574	return strncmp(s1, s2, s1_sz);
575	}
576
577	static const struct bpf_func_proto bpf_strncmp_proto = {
578	.func = bpf_strncmp,
579	.gpl_only = false,
580	.ret_type = RET_INTEGER,
581	.arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY,
582	.arg2_type = ARG_CONST_SIZE,
583	.arg3_type = ARG_PTR_TO_CONST_STR,
584	};
585
586	BPF_CALL_4(bpf_get_ns_current_pid_tgid, u64, dev, u64, ino,
587	struct bpf_pidns_info *, nsdata, u32, size)
588	{
589	struct task_struct *task = current;
590	struct pid_namespace *pidns;
591	int err = -EINVAL;
592
593	if (unlikely(size != sizeof(struct bpf_pidns_info)))
594	goto clear;
595
596	if (unlikely((u64)(dev_t)dev != dev))
597	goto clear;
598
599	if (unlikely(!task))
600	goto clear;
601
602	pidns = task_active_pid_ns(tsk: task);
603	if (unlikely(!pidns)) {
604	err = -ENOENT;
605	goto clear;
606	}
607
608	if (!ns_match(ns: &pidns->ns, dev: (dev_t)dev, ino))
609	goto clear;
610
611	nsdata->pid = task_pid_nr_ns(tsk: task, ns: pidns);
612	nsdata->tgid = task_tgid_nr_ns(tsk: task, ns: pidns);
613	return 0;
614	clear:
615	memset((void *)nsdata, 0, (size_t) size);
616	return err;
617	}
618
619	const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto = {
620	.func = bpf_get_ns_current_pid_tgid,
621	.gpl_only = false,
622	.ret_type = RET_INTEGER,
623	.arg1_type = ARG_ANYTHING,
624	.arg2_type = ARG_ANYTHING,
625	.arg3_type = ARG_PTR_TO_UNINIT_MEM,
626	.arg4_type = ARG_CONST_SIZE,
627	};
628
629	static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
630	.func = bpf_get_raw_cpu_id,
631	.gpl_only = false,
632	.ret_type = RET_INTEGER,
633	};
634
635	BPF_CALL_5(bpf_event_output_data, void *, ctx, struct bpf_map *, map,
636	u64, flags, void *, data, u64, size)
637	{
638	if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
639	return -EINVAL;
640
641	return bpf_event_output(map, flags, meta: data, meta_size: size, NULL, ctx_size: 0, NULL);
642	}
643
644	const struct bpf_func_proto bpf_event_output_data_proto = {
645	.func = bpf_event_output_data,
646	.gpl_only = true,
647	.ret_type = RET_INTEGER,
648	.arg1_type = ARG_PTR_TO_CTX,
649	.arg2_type = ARG_CONST_MAP_PTR,
650	.arg3_type = ARG_ANYTHING,
651	.arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
652	.arg5_type = ARG_CONST_SIZE_OR_ZERO,
653	};
654
655	BPF_CALL_3(bpf_copy_from_user, void *, dst, u32, size,
656	const void __user *, user_ptr)
657	{
658	int ret = copy_from_user(to: dst, from: user_ptr, n: size);
659
660	if (unlikely(ret)) {
661	memset(dst, 0, size);
662	ret = -EFAULT;
663	}
664
665	return ret;
666	}
667
668	const struct bpf_func_proto bpf_copy_from_user_proto = {
669	.func = bpf_copy_from_user,
670	.gpl_only = false,
671	.might_sleep = true,
672	.ret_type = RET_INTEGER,
673	.arg1_type = ARG_PTR_TO_UNINIT_MEM,
674	.arg2_type = ARG_CONST_SIZE_OR_ZERO,
675	.arg3_type = ARG_ANYTHING,
676	};
677
678	BPF_CALL_5(bpf_copy_from_user_task, void *, dst, u32, size,
679	const void __user *, user_ptr, struct task_struct *, tsk, u64, flags)
680	{
681	int ret;
682
683	/* flags is not used yet */
684	if (unlikely(flags))
685	return -EINVAL;
686
687	if (unlikely(!size))
688	return 0;
689
690	ret = access_process_vm(tsk, addr: (unsigned long)user_ptr, buf: dst, len: size, gup_flags: 0);
691	if (ret == size)
692	return 0;
693
694	memset(dst, 0, size);
695	/* Return -EFAULT for partial read */
696	return ret < 0 ? ret : -EFAULT;
697	}
698
699	const struct bpf_func_proto bpf_copy_from_user_task_proto = {
700	.func = bpf_copy_from_user_task,
701	.gpl_only = true,
702	.might_sleep = true,
703	.ret_type = RET_INTEGER,
704	.arg1_type = ARG_PTR_TO_UNINIT_MEM,
705	.arg2_type = ARG_CONST_SIZE_OR_ZERO,
706	.arg3_type = ARG_ANYTHING,
707	.arg4_type = ARG_PTR_TO_BTF_ID,
708	.arg4_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
709	.arg5_type = ARG_ANYTHING
710	};
711
712	BPF_CALL_2(bpf_per_cpu_ptr, const void *, ptr, u32, cpu)
713	{
714	if (cpu >= nr_cpu_ids)
715	return (unsigned long)NULL;
716
717	return (unsigned long)per_cpu_ptr((const void __percpu *)ptr, cpu);
718	}
719
720	const struct bpf_func_proto bpf_per_cpu_ptr_proto = {
721	.func = bpf_per_cpu_ptr,
722	.gpl_only = false,
723	.ret_type = RET_PTR_TO_MEM_OR_BTF_ID | PTR_MAYBE_NULL | MEM_RDONLY,
724	.arg1_type = ARG_PTR_TO_PERCPU_BTF_ID,
725	.arg2_type = ARG_ANYTHING,
726	};
727
728	BPF_CALL_1(bpf_this_cpu_ptr, const void *, percpu_ptr)
729	{
730	return (unsigned long)this_cpu_ptr((const void __percpu *)percpu_ptr);
731	}
732
733	const struct bpf_func_proto bpf_this_cpu_ptr_proto = {
734	.func = bpf_this_cpu_ptr,
735	.gpl_only = false,
736	.ret_type = RET_PTR_TO_MEM_OR_BTF_ID | MEM_RDONLY,
737	.arg1_type = ARG_PTR_TO_PERCPU_BTF_ID,
738	};
739
740	static int bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
741	size_t bufsz)
742	{
743	void __user *user_ptr = (__force void __user *)unsafe_ptr;
744
745	buf[0] = 0;
746
747	switch (fmt_ptype) {
748	case 's':
749	#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
750	if ((unsigned long)unsafe_ptr < TASK_SIZE)
751	return strncpy_from_user_nofault(dst: buf, unsafe_addr: user_ptr, count: bufsz);
752	fallthrough;
753	#endif
754	case 'k':
755	return strncpy_from_kernel_nofault(dst: buf, unsafe_addr: unsafe_ptr, count: bufsz);
756	case 'u':
757	return strncpy_from_user_nofault(dst: buf, unsafe_addr: user_ptr, count: bufsz);
758	}
759
760	return -EINVAL;
761	}
762
763	/* Per-cpu temp buffers used by printf-like helpers to store the bprintf binary
764	* arguments representation.
765	*/
766	#define MAX_BPRINTF_BIN_ARGS 512
767
768	/* Support executing three nested bprintf helper calls on a given CPU */
769	#define MAX_BPRINTF_NEST_LEVEL 3
770	struct bpf_bprintf_buffers {
771	char bin_args[MAX_BPRINTF_BIN_ARGS];
772	char buf[MAX_BPRINTF_BUF];
773	};
774
775	static DEFINE_PER_CPU(struct bpf_bprintf_buffers[MAX_BPRINTF_NEST_LEVEL], bpf_bprintf_bufs);
776	static DEFINE_PER_CPU(int, bpf_bprintf_nest_level);
777
778	static int try_get_buffers(struct bpf_bprintf_buffers **bufs)
779	{
780	int nest_level;
781
782	preempt_disable();
783	nest_level = this_cpu_inc_return(bpf_bprintf_nest_level);
784	if (WARN_ON_ONCE(nest_level > MAX_BPRINTF_NEST_LEVEL)) {
785	this_cpu_dec(bpf_bprintf_nest_level);
786	preempt_enable();
787	return -EBUSY;
788	}
789	*bufs = this_cpu_ptr(&bpf_bprintf_bufs[nest_level - 1]);
790
791	return 0;
792	}
793
794	void bpf_bprintf_cleanup(struct bpf_bprintf_data *data)
795	{
796	if (!data->bin_args && !data->buf)
797	return;
798	if (WARN_ON_ONCE(this_cpu_read(bpf_bprintf_nest_level) == 0))
799	return;
800	this_cpu_dec(bpf_bprintf_nest_level);
801	preempt_enable();
802	}
803
804	/*
805	* bpf_bprintf_prepare - Generic pass on format strings for bprintf-like helpers
806	*
807	* Returns a negative value if fmt is an invalid format string or 0 otherwise.
808	*
809	* This can be used in two ways:
810	* - Format string verification only: when data->get_bin_args is false
811	* - Arguments preparation: in addition to the above verification, it writes in
812	* data->bin_args a binary representation of arguments usable by bstr_printf
813	* where pointers from BPF have been sanitized.
814	*
815	* In argument preparation mode, if 0 is returned, safe temporary buffers are
816	* allocated and bpf_bprintf_cleanup should be called to free them after use.
817	*/
818	int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args,
819	u32 num_args, struct bpf_bprintf_data *data)
820	{
821	bool get_buffers = (data->get_bin_args && num_args) || data->get_buf;
822	char *unsafe_ptr = NULL, *tmp_buf = NULL, *tmp_buf_end, *fmt_end;
823	struct bpf_bprintf_buffers *buffers = NULL;
824	size_t sizeof_cur_arg, sizeof_cur_ip;
825	int err, i, num_spec = 0;
826	u64 cur_arg;
827	char fmt_ptype, cur_ip[16], ip_spec[] = "%pXX";
828
829	fmt_end = strnchr(fmt, fmt_size, 0);
830	if (!fmt_end)
831	return -EINVAL;
832	fmt_size = fmt_end - fmt;
833
834	if (get_buffers && try_get_buffers(bufs: &buffers))
835	return -EBUSY;
836
837	if (data->get_bin_args) {
838	if (num_args)
839	tmp_buf = buffers->bin_args;
840	tmp_buf_end = tmp_buf + MAX_BPRINTF_BIN_ARGS;
841	data->bin_args = (u32 *)tmp_buf;
842	}
843
844	if (data->get_buf)
845	data->buf = buffers->buf;
846
847	for (i = 0; i < fmt_size; i++) {
848	if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
849	err = -EINVAL;
850	goto out;
851	}
852
853	if (fmt[i] != '%')
854	continue;
855
856	if (fmt[i + 1] == '%') {
857	i++;
858	continue;
859	}
860
861	if (num_spec >= num_args) {
862	err = -EINVAL;
863	goto out;
864	}
865
866	/* The string is zero-terminated so if fmt[i] != 0, we can
867	* always access fmt[i + 1], in the worst case it will be a 0
868	*/
869	i++;
870
871	/* skip optional "[0 +-][num]" width formatting field */
872	while (fmt[i] == '0' || fmt[i] == '+' || fmt[i] == '-' ||
873	fmt[i] == ' ')
874	i++;
875	if (fmt[i] >= '1' && fmt[i] <= '9') {
876	i++;
877	while (fmt[i] >= '0' && fmt[i] <= '9')
878	i++;
879	}
880
881	if (fmt[i] == 'p') {
882	sizeof_cur_arg = sizeof(long);
883
884	if ((fmt[i + 1] == 'k' || fmt[i + 1] == 'u') &&
885	fmt[i + 2] == 's') {
886	fmt_ptype = fmt[i + 1];
887	i += 2;
888	goto fmt_str;
889	}
890
891	if (fmt[i + 1] == 0 || isspace(fmt[i + 1]) ||
892	ispunct(fmt[i + 1]) || fmt[i + 1] == 'K' ||
893	fmt[i + 1] == 'x' || fmt[i + 1] == 's' ||
894	fmt[i + 1] == 'S') {
895	/* just kernel pointers */
896	if (tmp_buf)
897	cur_arg = raw_args[num_spec];
898	i++;
899	goto nocopy_fmt;
900	}
901
902	if (fmt[i + 1] == 'B') {
903	if (tmp_buf) {
904	err = snprintf(buf: tmp_buf,
905	size: (tmp_buf_end - tmp_buf),
906	fmt: "%pB",
907	(void *)(long)raw_args[num_spec]);
908	tmp_buf += (err + 1);
909	}
910
911	i++;
912	num_spec++;
913	continue;
914	}
915
916	/* only support "%pI4", "%pi4", "%pI6" and "%pi6". */
917	if ((fmt[i + 1] != 'i' && fmt[i + 1] != 'I') ||
918	(fmt[i + 2] != '4' && fmt[i + 2] != '6')) {
919	err = -EINVAL;
920	goto out;
921	}
922
923	i += 2;
924	if (!tmp_buf)
925	goto nocopy_fmt;
926
927	sizeof_cur_ip = (fmt[i] == '4') ? 4 : 16;
928	if (tmp_buf_end - tmp_buf < sizeof_cur_ip) {
929	err = -ENOSPC;
930	goto out;
931	}
932
933	unsafe_ptr = (char *)(long)raw_args[num_spec];
934	err = copy_from_kernel_nofault(dst: cur_ip, src: unsafe_ptr,
935	size: sizeof_cur_ip);
936	if (err < 0)
937	memset(cur_ip, 0, sizeof_cur_ip);
938
939	/* hack: bstr_printf expects IP addresses to be
940	* pre-formatted as strings, ironically, the easiest way
941	* to do that is to call snprintf.
942	*/
943	ip_spec[2] = fmt[i - 1];
944	ip_spec[3] = fmt[i];
945	err = snprintf(buf: tmp_buf, size: tmp_buf_end - tmp_buf,
946	fmt: ip_spec, &cur_ip);
947
948	tmp_buf += err + 1;
949	num_spec++;
950
951	continue;
952	} else if (fmt[i] == 's') {
953	fmt_ptype = fmt[i];
954	fmt_str:
955	if (fmt[i + 1] != 0 &&
956	!isspace(fmt[i + 1]) &&
957	!ispunct(fmt[i + 1])) {
958	err = -EINVAL;
959	goto out;
960	}
961
962	if (!tmp_buf)
963	goto nocopy_fmt;
964
965	if (tmp_buf_end == tmp_buf) {
966	err = -ENOSPC;
967	goto out;
968	}
969
970	unsafe_ptr = (char *)(long)raw_args[num_spec];
971	err = bpf_trace_copy_string(buf: tmp_buf, unsafe_ptr,
972	fmt_ptype,
973	bufsz: tmp_buf_end - tmp_buf);
974	if (err < 0) {
975	tmp_buf[0] = '\0';
976	err = 1;
977	}
978
979	tmp_buf += err;
980	num_spec++;
981
982	continue;
983	} else if (fmt[i] == 'c') {
984	if (!tmp_buf)
985	goto nocopy_fmt;
986
987	if (tmp_buf_end == tmp_buf) {
988	err = -ENOSPC;
989	goto out;
990	}
991
992	*tmp_buf = raw_args[num_spec];
993	tmp_buf++;
994	num_spec++;
995
996	continue;
997	}
998
999	sizeof_cur_arg = sizeof(int);
1000
1001	if (fmt[i] == 'l') {
1002	sizeof_cur_arg = sizeof(long);
1003	i++;
1004	}
1005	if (fmt[i] == 'l') {
1006	sizeof_cur_arg = sizeof(long long);
1007	i++;
1008	}
1009
1010	if (fmt[i] != 'i' && fmt[i] != 'd' && fmt[i] != 'u' &&
1011	fmt[i] != 'x' && fmt[i] != 'X') {
1012	err = -EINVAL;
1013	goto out;
1014	}
1015
1016	if (tmp_buf)
1017	cur_arg = raw_args[num_spec];
1018	nocopy_fmt:
1019	if (tmp_buf) {
1020	tmp_buf = PTR_ALIGN(tmp_buf, sizeof(u32));
1021	if (tmp_buf_end - tmp_buf < sizeof_cur_arg) {
1022	err = -ENOSPC;
1023	goto out;
1024	}
1025
1026	if (sizeof_cur_arg == 8) {
1027	*(u32 *)tmp_buf = *(u32 *)&cur_arg;
1028	*(u32 *)(tmp_buf + 4) = *((u32 *)&cur_arg + 1);
1029	} else {
1030	*(u32 *)tmp_buf = (u32)(long)cur_arg;
1031	}
1032	tmp_buf += sizeof_cur_arg;
1033	}
1034	num_spec++;
1035	}
1036
1037	err = 0;
1038	out:
1039	if (err)
1040	bpf_bprintf_cleanup(data);
1041	return err;
1042	}
1043
1044	BPF_CALL_5(bpf_snprintf, char *, str, u32, str_size, char *, fmt,
1045	const void *, args, u32, data_len)
1046	{
1047	struct bpf_bprintf_data data = {
1048	.get_bin_args = true,
1049	};
1050	int err, num_args;
1051
1052	if (data_len % 8 || data_len > MAX_BPRINTF_VARARGS * 8 ||
1053	(data_len && !args))
1054	return -EINVAL;
1055	num_args = data_len / 8;
1056
1057	/* ARG_PTR_TO_CONST_STR guarantees that fmt is zero-terminated so we
1058	* can safely give an unbounded size.
1059	*/
1060	err = bpf_bprintf_prepare(fmt, UINT_MAX, raw_args: args, num_args, data: &data);
1061	if (err < 0)
1062	return err;
1063
1064	err = bstr_printf(buf: str, size: str_size, fmt, bin_buf: data.bin_args);
1065
1066	bpf_bprintf_cleanup(data: &data);
1067
1068	return err + 1;
1069	}
1070
1071	const struct bpf_func_proto bpf_snprintf_proto = {
1072	.func = bpf_snprintf,
1073	.gpl_only = true,
1074	.ret_type = RET_INTEGER,
1075	.arg1_type = ARG_PTR_TO_MEM_OR_NULL,
1076	.arg2_type = ARG_CONST_SIZE_OR_ZERO,
1077	.arg3_type = ARG_PTR_TO_CONST_STR,
1078	.arg4_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
1079	.arg5_type = ARG_CONST_SIZE_OR_ZERO,
1080	};
1081
1082	/* BPF map elements can contain 'struct bpf_timer'.
1083	* Such map owns all of its BPF timers.
1084	* 'struct bpf_timer' is allocated as part of map element allocation
1085	* and it's zero initialized.
1086	* That space is used to keep 'struct bpf_timer_kern'.
1087	* bpf_timer_init() allocates 'struct bpf_hrtimer', inits hrtimer, and
1088	* remembers 'struct bpf_map *' pointer it's part of.
1089	* bpf_timer_set_callback() increments prog refcnt and assign bpf callback_fn.
1090	* bpf_timer_start() arms the timer.
1091	* If user space reference to a map goes to zero at this point
1092	* ops->map_release_uref callback is responsible for cancelling the timers,
1093	* freeing their memory, and decrementing prog's refcnts.
1094	* bpf_timer_cancel() cancels the timer and decrements prog's refcnt.
1095	* Inner maps can contain bpf timers as well. ops->map_release_uref is
1096	* freeing the timers when inner map is replaced or deleted by user space.
1097	*/
1098	struct bpf_hrtimer {
1099	struct hrtimer timer;
1100	struct bpf_map *map;
1101	struct bpf_prog *prog;
1102	void __rcu *callback_fn;
1103	void *value;
1104	struct rcu_head rcu;
1105	};
1106
1107	/* the actual struct hidden inside uapi struct bpf_timer */
1108	struct bpf_timer_kern {
1109	struct bpf_hrtimer *timer;
1110	/* bpf_spin_lock is used here instead of spinlock_t to make
1111	* sure that it always fits into space reserved by struct bpf_timer
1112	* regardless of LOCKDEP and spinlock debug flags.
1113	*/
1114	struct bpf_spin_lock lock;
1115	} __attribute__((aligned(8)));
1116
1117	static DEFINE_PER_CPU(struct bpf_hrtimer *, hrtimer_running);
1118
1119	static enum hrtimer_restart bpf_timer_cb(struct hrtimer *hrtimer)
1120	{
1121	struct bpf_hrtimer *t = container_of(hrtimer, struct bpf_hrtimer, timer);
1122	struct bpf_map *map = t->map;
1123	void *value = t->value;
1124	bpf_callback_t callback_fn;
1125	void *key;
1126	u32 idx;
1127
1128	BTF_TYPE_EMIT(struct bpf_timer);
1129	callback_fn = rcu_dereference_check(t->callback_fn, rcu_read_lock_bh_held());
1130	if (!callback_fn)
1131	goto out;
1132
1133	/* bpf_timer_cb() runs in hrtimer_run_softirq. It doesn't migrate and
1134	* cannot be preempted by another bpf_timer_cb() on the same cpu.
1135	* Remember the timer this callback is servicing to prevent
1136	* deadlock if callback_fn() calls bpf_timer_cancel() or
1137	* bpf_map_delete_elem() on the same timer.
1138	*/
1139	this_cpu_write(hrtimer_running, t);
1140	if (map->map_type == BPF_MAP_TYPE_ARRAY) {
1141	struct bpf_array *array = container_of(map, struct bpf_array, map);
1142
1143	/* compute the key */
1144	idx = ((char *)value - array->value) / array->elem_size;
1145	key = &idx;
1146	} else { /* hash or lru */
1147	key = value - round_up(map->key_size, 8);
1148	}
1149
1150	callback_fn((u64)(long)map, (u64)(long)key, (u64)(long)value, 0, 0);
1151	/* The verifier checked that return value is zero. */
1152
1153	this_cpu_write(hrtimer_running, NULL);
1154	out:
1155	return HRTIMER_NORESTART;
1156	}
1157
1158	BPF_CALL_3(bpf_timer_init, struct bpf_timer_kern *, timer, struct bpf_map *, map,
1159	u64, flags)
1160	{
1161	clockid_t clockid = flags & (MAX_CLOCKS - 1);
1162	struct bpf_hrtimer *t;
1163	int ret = 0;
1164
1165	BUILD_BUG_ON(MAX_CLOCKS != 16);
1166	BUILD_BUG_ON(sizeof(struct bpf_timer_kern) > sizeof(struct bpf_timer));
1167	BUILD_BUG_ON(__alignof__(struct bpf_timer_kern) != __alignof__(struct bpf_timer));
1168
1169	if (in_nmi())
1170	return -EOPNOTSUPP;
1171
1172	if (flags >= MAX_CLOCKS ||
1173	/* similar to timerfd except _ALARM variants are not supported */
1174	(clockid != CLOCK_MONOTONIC &&
1175	clockid != CLOCK_REALTIME &&
1176	clockid != CLOCK_BOOTTIME))
1177	return -EINVAL;
1178	__bpf_spin_lock_irqsave(lock: &timer->lock);
1179	t = timer->timer;
1180	if (t) {
1181	ret = -EBUSY;
1182	goto out;
1183	}
1184	/* allocate hrtimer via map_kmalloc to use memcg accounting */
1185	t = bpf_map_kmalloc_node(map, size: sizeof(*t), GFP_ATOMIC, node: map->numa_node);
1186	if (!t) {
1187	ret = -ENOMEM;
1188	goto out;
1189	}
1190	t->value = (void *)timer - map->record->timer_off;
1191	t->map = map;
1192	t->prog = NULL;
1193	rcu_assign_pointer(t->callback_fn, NULL);
1194	hrtimer_init(timer: &t->timer, which_clock: clockid, mode: HRTIMER_MODE_REL_SOFT);
1195	t->timer.function = bpf_timer_cb;
1196	WRITE_ONCE(timer->timer, t);
1197	/* Guarantee the order between timer->timer and map->usercnt. So
1198	* when there are concurrent uref release and bpf timer init, either
1199	* bpf_timer_cancel_and_free() called by uref release reads a no-NULL
1200	* timer or atomic64_read() below returns a zero usercnt.
1201	*/
1202	smp_mb();
1203	if (!atomic64_read(v: &map->usercnt)) {
1204	/* maps with timers must be either held by user space
1205	* or pinned in bpffs.
1206	*/
1207	WRITE_ONCE(timer->timer, NULL);
1208	kfree(objp: t);
1209	ret = -EPERM;
1210	}
1211	out:
1212	__bpf_spin_unlock_irqrestore(lock: &timer->lock);
1213	return ret;
1214	}
1215
1216	static const struct bpf_func_proto bpf_timer_init_proto = {
1217	.func = bpf_timer_init,
1218	.gpl_only = true,
1219	.ret_type = RET_INTEGER,
1220	.arg1_type = ARG_PTR_TO_TIMER,
1221	.arg2_type = ARG_CONST_MAP_PTR,
1222	.arg3_type = ARG_ANYTHING,
1223	};
1224
1225	BPF_CALL_3(bpf_timer_set_callback, struct bpf_timer_kern *, timer, void *, callback_fn,
1226	struct bpf_prog_aux *, aux)
1227	{
1228	struct bpf_prog *prev, *prog = aux->prog;
1229	struct bpf_hrtimer *t;
1230	int ret = 0;
1231
1232	if (in_nmi())
1233	return -EOPNOTSUPP;
1234	__bpf_spin_lock_irqsave(lock: &timer->lock);
1235	t = timer->timer;
1236	if (!t) {
1237	ret = -EINVAL;
1238	goto out;
1239	}
1240	if (!atomic64_read(v: &t->map->usercnt)) {
1241	/* maps with timers must be either held by user space
1242	* or pinned in bpffs. Otherwise timer might still be
1243	* running even when bpf prog is detached and user space
1244	* is gone, since map_release_uref won't ever be called.
1245	*/
1246	ret = -EPERM;
1247	goto out;
1248	}
1249	prev = t->prog;
1250	if (prev != prog) {
1251	/* Bump prog refcnt once. Every bpf_timer_set_callback()
1252	* can pick different callback_fn-s within the same prog.
1253	*/
1254	prog = bpf_prog_inc_not_zero(prog);
1255	if (IS_ERR(ptr: prog)) {
1256	ret = PTR_ERR(ptr: prog);
1257	goto out;
1258	}
1259	if (prev)
1260	/* Drop prev prog refcnt when swapping with new prog */
1261	bpf_prog_put(prog: prev);
1262	t->prog = prog;
1263	}
1264	rcu_assign_pointer(t->callback_fn, callback_fn);
1265	out:
1266	__bpf_spin_unlock_irqrestore(lock: &timer->lock);
1267	return ret;
1268	}
1269
1270	static const struct bpf_func_proto bpf_timer_set_callback_proto = {
1271	.func = bpf_timer_set_callback,
1272	.gpl_only = true,
1273	.ret_type = RET_INTEGER,
1274	.arg1_type = ARG_PTR_TO_TIMER,
1275	.arg2_type = ARG_PTR_TO_FUNC,
1276	};
1277
1278	BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, flags)
1279	{
1280	struct bpf_hrtimer *t;
1281	int ret = 0;
1282	enum hrtimer_mode mode;
1283
1284	if (in_nmi())
1285	return -EOPNOTSUPP;
1286	if (flags & ~(BPF_F_TIMER_ABS | BPF_F_TIMER_CPU_PIN))
1287	return -EINVAL;
1288	__bpf_spin_lock_irqsave(lock: &timer->lock);
1289	t = timer->timer;
1290	if (!t || !t->prog) {
1291	ret = -EINVAL;
1292	goto out;
1293	}
1294
1295	if (flags & BPF_F_TIMER_ABS)
1296	mode = HRTIMER_MODE_ABS_SOFT;
1297	else
1298	mode = HRTIMER_MODE_REL_SOFT;
1299
1300	if (flags & BPF_F_TIMER_CPU_PIN)
1301	mode |= HRTIMER_MODE_PINNED;
1302
1303	hrtimer_start(timer: &t->timer, tim: ns_to_ktime(ns: nsecs), mode);
1304	out:
1305	__bpf_spin_unlock_irqrestore(lock: &timer->lock);
1306	return ret;
1307	}
1308
1309	static const struct bpf_func_proto bpf_timer_start_proto = {
1310	.func = bpf_timer_start,
1311	.gpl_only = true,
1312	.ret_type = RET_INTEGER,
1313	.arg1_type = ARG_PTR_TO_TIMER,
1314	.arg2_type = ARG_ANYTHING,
1315	.arg3_type = ARG_ANYTHING,
1316	};
1317
1318	static void drop_prog_refcnt(struct bpf_hrtimer *t)
1319	{
1320	struct bpf_prog *prog = t->prog;
1321
1322	if (prog) {
1323	bpf_prog_put(prog);
1324	t->prog = NULL;
1325	rcu_assign_pointer(t->callback_fn, NULL);
1326	}
1327	}
1328
1329	BPF_CALL_1(bpf_timer_cancel, struct bpf_timer_kern *, timer)
1330	{
1331	struct bpf_hrtimer *t;
1332	int ret = 0;
1333
1334	if (in_nmi())
1335	return -EOPNOTSUPP;
1336	rcu_read_lock();
1337	__bpf_spin_lock_irqsave(lock: &timer->lock);
1338	t = timer->timer;
1339	if (!t) {
1340	ret = -EINVAL;
1341	goto out;
1342	}
1343	if (this_cpu_read(hrtimer_running) == t) {
1344	/* If bpf callback_fn is trying to bpf_timer_cancel()
1345	* its own timer the hrtimer_cancel() will deadlock
1346	* since it waits for callback_fn to finish
1347	*/
1348	ret = -EDEADLK;
1349	goto out;
1350	}
1351	drop_prog_refcnt(t);
1352	out:
1353	__bpf_spin_unlock_irqrestore(lock: &timer->lock);
1354	/* Cancel the timer and wait for associated callback to finish
1355	* if it was running.
1356	*/
1357	ret = ret ?: hrtimer_cancel(timer: &t->timer);
1358	rcu_read_unlock();
1359	return ret;
1360	}
1361
1362	static const struct bpf_func_proto bpf_timer_cancel_proto = {
1363	.func = bpf_timer_cancel,
1364	.gpl_only = true,
1365	.ret_type = RET_INTEGER,
1366	.arg1_type = ARG_PTR_TO_TIMER,
1367	};
1368
1369	/* This function is called by map_delete/update_elem for individual element and
1370	* by ops->map_release_uref when the user space reference to a map reaches zero.
1371	*/
1372	void bpf_timer_cancel_and_free(void *val)
1373	{
1374	struct bpf_timer_kern *timer = val;
1375	struct bpf_hrtimer *t;
1376
1377	/* Performance optimization: read timer->timer without lock first. */
1378	if (!READ_ONCE(timer->timer))
1379	return;
1380
1381	__bpf_spin_lock_irqsave(lock: &timer->lock);
1382	/* re-read it under lock */
1383	t = timer->timer;
1384	if (!t)
1385	goto out;
1386	drop_prog_refcnt(t);
1387	/* The subsequent bpf_timer_start/cancel() helpers won't be able to use
1388	* this timer, since it won't be initialized.
1389	*/
1390	WRITE_ONCE(timer->timer, NULL);
1391	out:
1392	__bpf_spin_unlock_irqrestore(lock: &timer->lock);
1393	if (!t)
1394	return;
1395	/* Cancel the timer and wait for callback to complete if it was running.
1396	* If hrtimer_cancel() can be safely called it's safe to call kfree(t)
1397	* right after for both preallocated and non-preallocated maps.
1398	* The timer->timer = NULL was already done and no code path can
1399	* see address 't' anymore.
1400	*
1401	* Check that bpf_map_delete/update_elem() wasn't called from timer
1402	* callback_fn. In such case don't call hrtimer_cancel() (since it will
1403	* deadlock) and don't call hrtimer_try_to_cancel() (since it will just
1404	* return -1). Though callback_fn is still running on this cpu it's
1405	* safe to do kfree(t) because bpf_timer_cb() read everything it needed
1406	* from 't'. The bpf subprog callback_fn won't be able to access 't',
1407	* since timer->timer = NULL was already done. The timer will be
1408	* effectively cancelled because bpf_timer_cb() will return
1409	* HRTIMER_NORESTART.
1410	*/
1411	if (this_cpu_read(hrtimer_running) != t)
1412	hrtimer_cancel(timer: &t->timer);
1413	kfree_rcu(t, rcu);
1414	}
1415
1416	BPF_CALL_2(bpf_kptr_xchg, void *, map_value, void *, ptr)
1417	{
1418	unsigned long *kptr = map_value;
1419
1420	/* This helper may be inlined by verifier. */
1421	return xchg(kptr, (unsigned long)ptr);
1422	}
1423
1424	/* Unlike other PTR_TO_BTF_ID helpers the btf_id in bpf_kptr_xchg()
1425	* helper is determined dynamically by the verifier. Use BPF_PTR_POISON to
1426	* denote type that verifier will determine.
1427	*/
1428	static const struct bpf_func_proto bpf_kptr_xchg_proto = {
1429	.func = bpf_kptr_xchg,
1430	.gpl_only = false,
1431	.ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
1432	.ret_btf_id = BPF_PTR_POISON,
1433	.arg1_type = ARG_PTR_TO_KPTR,
1434	.arg2_type = ARG_PTR_TO_BTF_ID_OR_NULL | OBJ_RELEASE,
1435	.arg2_btf_id = BPF_PTR_POISON,
1436	};
1437
1438	/* Since the upper 8 bits of dynptr->size is reserved, the
1439	* maximum supported size is 2^24 - 1.
1440	*/
1441	#define DYNPTR_MAX_SIZE ((1UL << 24) - 1)
1442	#define DYNPTR_TYPE_SHIFT 28
1443	#define DYNPTR_SIZE_MASK 0xFFFFFF
1444	#define DYNPTR_RDONLY_BIT BIT(31)
1445
1446	static bool __bpf_dynptr_is_rdonly(const struct bpf_dynptr_kern *ptr)
1447	{
1448	return ptr->size & DYNPTR_RDONLY_BIT;
1449	}
1450
1451	void bpf_dynptr_set_rdonly(struct bpf_dynptr_kern *ptr)
1452	{
1453	ptr->size |= DYNPTR_RDONLY_BIT;
1454	}
1455
1456	static void bpf_dynptr_set_type(struct bpf_dynptr_kern *ptr, enum bpf_dynptr_type type)
1457	{
1458	ptr->size |= type << DYNPTR_TYPE_SHIFT;
1459	}
1460
1461	static enum bpf_dynptr_type bpf_dynptr_get_type(const struct bpf_dynptr_kern *ptr)
1462	{
1463	return (ptr->size & ~(DYNPTR_RDONLY_BIT)) >> DYNPTR_TYPE_SHIFT;
1464	}
1465
1466	u32 __bpf_dynptr_size(const struct bpf_dynptr_kern *ptr)
1467	{
1468	return ptr->size & DYNPTR_SIZE_MASK;
1469	}
1470
1471	static void bpf_dynptr_set_size(struct bpf_dynptr_kern *ptr, u32 new_size)
1472	{
1473	u32 metadata = ptr->size & ~DYNPTR_SIZE_MASK;
1474
1475	ptr->size = new_size | metadata;
1476	}
1477
1478	int bpf_dynptr_check_size(u32 size)
1479	{
1480	return size > DYNPTR_MAX_SIZE ? -E2BIG : 0;
1481	}
1482
1483	void bpf_dynptr_init(struct bpf_dynptr_kern *ptr, void *data,
1484	enum bpf_dynptr_type type, u32 offset, u32 size)
1485	{
1486	ptr->data = data;
1487	ptr->offset = offset;
1488	ptr->size = size;
1489	bpf_dynptr_set_type(ptr, type);
1490	}
1491
1492	void bpf_dynptr_set_null(struct bpf_dynptr_kern *ptr)
1493	{
1494	memset(ptr, 0, sizeof(*ptr));
1495	}
1496
1497	static int bpf_dynptr_check_off_len(const struct bpf_dynptr_kern *ptr, u32 offset, u32 len)
1498	{
1499	u32 size = __bpf_dynptr_size(ptr);
1500
1501	if (len > size || offset > size - len)
1502	return -E2BIG;
1503
1504	return 0;
1505	}
1506
1507	BPF_CALL_4(bpf_dynptr_from_mem, void *, data, u32, size, u64, flags, struct bpf_dynptr_kern *, ptr)
1508	{
1509	int err;
1510
1511	BTF_TYPE_EMIT(struct bpf_dynptr);
1512
1513	err = bpf_dynptr_check_size(size);
1514	if (err)
1515	goto error;
1516
1517	/* flags is currently unsupported */
1518	if (flags) {
1519	err = -EINVAL;
1520	goto error;
1521	}
1522
1523	bpf_dynptr_init(ptr, data, type: BPF_DYNPTR_TYPE_LOCAL, offset: 0, size);
1524
1525	return 0;
1526
1527	error:
1528	bpf_dynptr_set_null(ptr);
1529	return err;
1530	}
1531
1532	static const struct bpf_func_proto bpf_dynptr_from_mem_proto = {
1533	.func = bpf_dynptr_from_mem,
1534	.gpl_only = false,
1535	.ret_type = RET_INTEGER,
1536	.arg1_type = ARG_PTR_TO_UNINIT_MEM,
1537	.arg2_type = ARG_CONST_SIZE_OR_ZERO,
1538	.arg3_type = ARG_ANYTHING,
1539	.arg4_type = ARG_PTR_TO_DYNPTR | DYNPTR_TYPE_LOCAL | MEM_UNINIT,
1540	};
1541
1542	BPF_CALL_5(bpf_dynptr_read, void *, dst, u32, len, const struct bpf_dynptr_kern *, src,
1543	u32, offset, u64, flags)
1544	{
1545	enum bpf_dynptr_type type;
1546	int err;
1547
1548	if (!src->data || flags)
1549	return -EINVAL;
1550
1551	err = bpf_dynptr_check_off_len(ptr: src, offset, len);
1552	if (err)
1553	return err;
1554
1555	type = bpf_dynptr_get_type(ptr: src);
1556
1557	switch (type) {
1558	case BPF_DYNPTR_TYPE_LOCAL:
1559	case BPF_DYNPTR_TYPE_RINGBUF:
1560	/* Source and destination may possibly overlap, hence use memmove to
1561	* copy the data. E.g. bpf_dynptr_from_mem may create two dynptr
1562	* pointing to overlapping PTR_TO_MAP_VALUE regions.
1563	*/
1564	memmove(dst, src->data + src->offset + offset, len);
1565	return 0;
1566	case BPF_DYNPTR_TYPE_SKB:
1567	return __bpf_skb_load_bytes(skb: src->data, offset: src->offset + offset, to: dst, len);
1568	case BPF_DYNPTR_TYPE_XDP:
1569	return __bpf_xdp_load_bytes(xdp: src->data, offset: src->offset + offset, buf: dst, len);
1570	default:
1571	WARN_ONCE(true, "bpf_dynptr_read: unknown dynptr type %d\n", type);
1572	return -EFAULT;
1573	}
1574	}
1575
1576	static const struct bpf_func_proto bpf_dynptr_read_proto = {
1577	.func = bpf_dynptr_read,
1578	.gpl_only = false,
1579	.ret_type = RET_INTEGER,
1580	.arg1_type = ARG_PTR_TO_UNINIT_MEM,
1581	.arg2_type = ARG_CONST_SIZE_OR_ZERO,
1582	.arg3_type = ARG_PTR_TO_DYNPTR | MEM_RDONLY,
1583	.arg4_type = ARG_ANYTHING,
1584	.arg5_type = ARG_ANYTHING,
1585	};
1586
1587	BPF_CALL_5(bpf_dynptr_write, const struct bpf_dynptr_kern *, dst, u32, offset, void *, src,
1588	u32, len, u64, flags)
1589	{
1590	enum bpf_dynptr_type type;
1591	int err;
1592
1593	if (!dst->data || __bpf_dynptr_is_rdonly(ptr: dst))
1594	return -EINVAL;
1595
1596	err = bpf_dynptr_check_off_len(ptr: dst, offset, len);
1597	if (err)
1598	return err;
1599
1600	type = bpf_dynptr_get_type(ptr: dst);
1601
1602	switch (type) {
1603	case BPF_DYNPTR_TYPE_LOCAL:
1604	case BPF_DYNPTR_TYPE_RINGBUF:
1605	if (flags)
1606	return -EINVAL;
1607	/* Source and destination may possibly overlap, hence use memmove to
1608	* copy the data. E.g. bpf_dynptr_from_mem may create two dynptr
1609	* pointing to overlapping PTR_TO_MAP_VALUE regions.
1610	*/
1611	memmove(dst->data + dst->offset + offset, src, len);
1612	return 0;
1613	case BPF_DYNPTR_TYPE_SKB:
1614	return __bpf_skb_store_bytes(skb: dst->data, offset: dst->offset + offset, from: src, len,
1615	flags);
1616	case BPF_DYNPTR_TYPE_XDP:
1617	if (flags)
1618	return -EINVAL;
1619	return __bpf_xdp_store_bytes(xdp: dst->data, offset: dst->offset + offset, buf: src, len);
1620	default:
1621	WARN_ONCE(true, "bpf_dynptr_write: unknown dynptr type %d\n", type);
1622	return -EFAULT;
1623	}
1624	}
1625
1626	static const struct bpf_func_proto bpf_dynptr_write_proto = {
1627	.func = bpf_dynptr_write,
1628	.gpl_only = false,
1629	.ret_type = RET_INTEGER,
1630	.arg1_type = ARG_PTR_TO_DYNPTR | MEM_RDONLY,
1631	.arg2_type = ARG_ANYTHING,
1632	.arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1633	.arg4_type = ARG_CONST_SIZE_OR_ZERO,
1634	.arg5_type = ARG_ANYTHING,
1635	};
1636
1637	BPF_CALL_3(bpf_dynptr_data, const struct bpf_dynptr_kern *, ptr, u32, offset, u32, len)
1638	{
1639	enum bpf_dynptr_type type;
1640	int err;
1641
1642	if (!ptr->data)
1643	return 0;
1644
1645	err = bpf_dynptr_check_off_len(ptr, offset, len);
1646	if (err)
1647	return 0;
1648
1649	if (__bpf_dynptr_is_rdonly(ptr))
1650	return 0;
1651
1652	type = bpf_dynptr_get_type(ptr);
1653
1654	switch (type) {
1655	case BPF_DYNPTR_TYPE_LOCAL:
1656	case BPF_DYNPTR_TYPE_RINGBUF:
1657	return (unsigned long)(ptr->data + ptr->offset + offset);
1658	case BPF_DYNPTR_TYPE_SKB:
1659	case BPF_DYNPTR_TYPE_XDP:
1660	/* skb and xdp dynptrs should use bpf_dynptr_slice / bpf_dynptr_slice_rdwr */
1661	return 0;
1662	default:
1663	WARN_ONCE(true, "bpf_dynptr_data: unknown dynptr type %d\n", type);
1664	return 0;
1665	}
1666	}
1667
1668	static const struct bpf_func_proto bpf_dynptr_data_proto = {
1669	.func = bpf_dynptr_data,
1670	.gpl_only = false,
1671	.ret_type = RET_PTR_TO_DYNPTR_MEM_OR_NULL,
1672	.arg1_type = ARG_PTR_TO_DYNPTR | MEM_RDONLY,
1673	.arg2_type = ARG_ANYTHING,
1674	.arg3_type = ARG_CONST_ALLOC_SIZE_OR_ZERO,
1675	};
1676
1677	const struct bpf_func_proto bpf_get_current_task_proto __weak;
1678	const struct bpf_func_proto bpf_get_current_task_btf_proto __weak;
1679	const struct bpf_func_proto bpf_probe_read_user_proto __weak;
1680	const struct bpf_func_proto bpf_probe_read_user_str_proto __weak;
1681	const struct bpf_func_proto bpf_probe_read_kernel_proto __weak;
1682	const struct bpf_func_proto bpf_probe_read_kernel_str_proto __weak;
1683	const struct bpf_func_proto bpf_task_pt_regs_proto __weak;
1684
1685	const struct bpf_func_proto *
1686	bpf_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1687	{
1688	switch (func_id) {
1689	case BPF_FUNC_map_lookup_elem:
1690	return &bpf_map_lookup_elem_proto;
1691	case BPF_FUNC_map_update_elem:
1692	return &bpf_map_update_elem_proto;
1693	case BPF_FUNC_map_delete_elem:
1694	return &bpf_map_delete_elem_proto;
1695	case BPF_FUNC_map_push_elem:
1696	return &bpf_map_push_elem_proto;
1697	case BPF_FUNC_map_pop_elem:
1698	return &bpf_map_pop_elem_proto;
1699	case BPF_FUNC_map_peek_elem:
1700	return &bpf_map_peek_elem_proto;
1701	case BPF_FUNC_map_lookup_percpu_elem:
1702	return &bpf_map_lookup_percpu_elem_proto;
1703	case BPF_FUNC_get_prandom_u32:
1704	return &bpf_get_prandom_u32_proto;
1705	case BPF_FUNC_get_smp_processor_id:
1706	return &bpf_get_raw_smp_processor_id_proto;
1707	case BPF_FUNC_get_numa_node_id:
1708	return &bpf_get_numa_node_id_proto;
1709	case BPF_FUNC_tail_call:
1710	return &bpf_tail_call_proto;
1711	case BPF_FUNC_ktime_get_ns:
1712	return &bpf_ktime_get_ns_proto;
1713	case BPF_FUNC_ktime_get_boot_ns:
1714	return &bpf_ktime_get_boot_ns_proto;
1715	case BPF_FUNC_ktime_get_tai_ns:
1716	return &bpf_ktime_get_tai_ns_proto;
1717	case BPF_FUNC_ringbuf_output:
1718	return &bpf_ringbuf_output_proto;
1719	case BPF_FUNC_ringbuf_reserve:
1720	return &bpf_ringbuf_reserve_proto;
1721	case BPF_FUNC_ringbuf_submit:
1722	return &bpf_ringbuf_submit_proto;
1723	case BPF_FUNC_ringbuf_discard:
1724	return &bpf_ringbuf_discard_proto;
1725	case BPF_FUNC_ringbuf_query:
1726	return &bpf_ringbuf_query_proto;
1727	case BPF_FUNC_strncmp:
1728	return &bpf_strncmp_proto;
1729	case BPF_FUNC_strtol:
1730	return &bpf_strtol_proto;
1731	case BPF_FUNC_strtoul:
1732	return &bpf_strtoul_proto;
1733	default:
1734	break;
1735	}
1736
1737	if (!bpf_token_capable(token: prog->aux->token, CAP_BPF))
1738	return NULL;
1739
1740	switch (func_id) {
1741	case BPF_FUNC_spin_lock:
1742	return &bpf_spin_lock_proto;
1743	case BPF_FUNC_spin_unlock:
1744	return &bpf_spin_unlock_proto;
1745	case BPF_FUNC_jiffies64:
1746	return &bpf_jiffies64_proto;
1747	case BPF_FUNC_per_cpu_ptr:
1748	return &bpf_per_cpu_ptr_proto;
1749	case BPF_FUNC_this_cpu_ptr:
1750	return &bpf_this_cpu_ptr_proto;
1751	case BPF_FUNC_timer_init:
1752	return &bpf_timer_init_proto;
1753	case BPF_FUNC_timer_set_callback:
1754	return &bpf_timer_set_callback_proto;
1755	case BPF_FUNC_timer_start:
1756	return &bpf_timer_start_proto;
1757	case BPF_FUNC_timer_cancel:
1758	return &bpf_timer_cancel_proto;
1759	case BPF_FUNC_kptr_xchg:
1760	return &bpf_kptr_xchg_proto;
1761	case BPF_FUNC_for_each_map_elem:
1762	return &bpf_for_each_map_elem_proto;
1763	case BPF_FUNC_loop:
1764	return &bpf_loop_proto;
1765	case BPF_FUNC_user_ringbuf_drain:
1766	return &bpf_user_ringbuf_drain_proto;
1767	case BPF_FUNC_ringbuf_reserve_dynptr:
1768	return &bpf_ringbuf_reserve_dynptr_proto;
1769	case BPF_FUNC_ringbuf_submit_dynptr:
1770	return &bpf_ringbuf_submit_dynptr_proto;
1771	case BPF_FUNC_ringbuf_discard_dynptr:
1772	return &bpf_ringbuf_discard_dynptr_proto;
1773	case BPF_FUNC_dynptr_from_mem:
1774	return &bpf_dynptr_from_mem_proto;
1775	case BPF_FUNC_dynptr_read:
1776	return &bpf_dynptr_read_proto;
1777	case BPF_FUNC_dynptr_write:
1778	return &bpf_dynptr_write_proto;
1779	case BPF_FUNC_dynptr_data:
1780	return &bpf_dynptr_data_proto;
1781	#ifdef CONFIG_CGROUPS
1782	case BPF_FUNC_cgrp_storage_get:
1783	return &bpf_cgrp_storage_get_proto;
1784	case BPF_FUNC_cgrp_storage_delete:
1785	return &bpf_cgrp_storage_delete_proto;
1786	case BPF_FUNC_get_current_cgroup_id:
1787	return &bpf_get_current_cgroup_id_proto;
1788	case BPF_FUNC_get_current_ancestor_cgroup_id:
1789	return &bpf_get_current_ancestor_cgroup_id_proto;
1790	#endif
1791	default:
1792	break;
1793	}
1794
1795	if (!bpf_token_capable(token: prog->aux->token, CAP_PERFMON))
1796	return NULL;
1797
1798	switch (func_id) {
1799	case BPF_FUNC_trace_printk:
1800	return bpf_get_trace_printk_proto();
1801	case BPF_FUNC_get_current_task:
1802	return &bpf_get_current_task_proto;
1803	case BPF_FUNC_get_current_task_btf:
1804	return &bpf_get_current_task_btf_proto;
1805	case BPF_FUNC_probe_read_user:
1806	return &bpf_probe_read_user_proto;
1807	case BPF_FUNC_probe_read_kernel:
1808	return security_locked_down(what: LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1809	NULL : &bpf_probe_read_kernel_proto;
1810	case BPF_FUNC_probe_read_user_str:
1811	return &bpf_probe_read_user_str_proto;
1812	case BPF_FUNC_probe_read_kernel_str:
1813	return security_locked_down(what: LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1814	NULL : &bpf_probe_read_kernel_str_proto;
1815	case BPF_FUNC_snprintf_btf:
1816	return &bpf_snprintf_btf_proto;
1817	case BPF_FUNC_snprintf:
1818	return &bpf_snprintf_proto;
1819	case BPF_FUNC_task_pt_regs:
1820	return &bpf_task_pt_regs_proto;
1821	case BPF_FUNC_trace_vprintk:
1822	return bpf_get_trace_vprintk_proto();
1823	default:
1824	return NULL;
1825	}
1826	}
1827
1828	void bpf_list_head_free(const struct btf_field *field, void *list_head,
1829	struct bpf_spin_lock *spin_lock)
1830	{
1831	struct list_head *head = list_head, *orig_head = list_head;
1832
1833	BUILD_BUG_ON(sizeof(struct list_head) > sizeof(struct bpf_list_head));
1834	BUILD_BUG_ON(__alignof__(struct list_head) > __alignof__(struct bpf_list_head));
1835
1836	/* Do the actual list draining outside the lock to not hold the lock for
1837	* too long, and also prevent deadlocks if tracing programs end up
1838	* executing on entry/exit of functions called inside the critical
1839	* section, and end up doing map ops that call bpf_list_head_free for
1840	* the same map value again.
1841	*/
1842	__bpf_spin_lock_irqsave(lock: spin_lock);
1843	if (!head->next || list_empty(head))
1844	goto unlock;
1845	head = head->next;
1846	unlock:
1847	INIT_LIST_HEAD(list: orig_head);
1848	__bpf_spin_unlock_irqrestore(lock: spin_lock);
1849
1850	while (head != orig_head) {
1851	void *obj = head;
1852
1853	obj -= field->graph_root.node_offset;
1854	head = head->next;
1855	/* The contained type can also have resources, including a
1856	* bpf_list_head which needs to be freed.
1857	*/
1858	migrate_disable();
1859	__bpf_obj_drop_impl(p: obj, rec: field->graph_root.value_rec, percpu: false);
1860	migrate_enable();
1861	}
1862	}
1863
1864	/* Like rbtree_postorder_for_each_entry_safe, but 'pos' and 'n' are
1865	* 'rb_node *', so field name of rb_node within containing struct is not
1866	* needed.
1867	*
1868	* Since bpf_rb_tree's node type has a corresponding struct btf_field with
1869	* graph_root.node_offset, it's not necessary to know field name
1870	* or type of node struct
1871	*/
1872	#define bpf_rbtree_postorder_for_each_entry_safe(pos, n, root) \
1873	for (pos = rb_first_postorder(root); \
1874	pos && ({ n = rb_next_postorder(pos); 1; }); \
1875	pos = n)
1876
1877	void bpf_rb_root_free(const struct btf_field *field, void *rb_root,
1878	struct bpf_spin_lock *spin_lock)
1879	{
1880	struct rb_root_cached orig_root, *root = rb_root;
1881	struct rb_node *pos, *n;
1882	void *obj;
1883
1884	BUILD_BUG_ON(sizeof(struct rb_root_cached) > sizeof(struct bpf_rb_root));
1885	BUILD_BUG_ON(__alignof__(struct rb_root_cached) > __alignof__(struct bpf_rb_root));
1886
1887	__bpf_spin_lock_irqsave(lock: spin_lock);
1888	orig_root = *root;
1889	*root = RB_ROOT_CACHED;
1890	__bpf_spin_unlock_irqrestore(lock: spin_lock);
1891
1892	bpf_rbtree_postorder_for_each_entry_safe(pos, n, &orig_root.rb_root) {
1893	obj = pos;
1894	obj -= field->graph_root.node_offset;
1895
1896
1897	migrate_disable();
1898	__bpf_obj_drop_impl(p: obj, rec: field->graph_root.value_rec, percpu: false);
1899	migrate_enable();
1900	}
1901	}
1902
1903	__bpf_kfunc_start_defs();
1904
1905	__bpf_kfunc void *bpf_obj_new_impl(u64 local_type_id__k, void *meta__ign)
1906	{
1907	struct btf_struct_meta *meta = meta__ign;
1908	u64 size = local_type_id__k;
1909	void *p;
1910
1911	p = bpf_mem_alloc(ma: &bpf_global_ma, size);
1912	if (!p)
1913	return NULL;
1914	if (meta)
1915	bpf_obj_init(rec: meta->record, obj: p);
1916	return p;
1917	}
1918
1919	__bpf_kfunc void *bpf_percpu_obj_new_impl(u64 local_type_id__k, void *meta__ign)
1920	{
1921	u64 size = local_type_id__k;
1922
1923	/* The verifier has ensured that meta__ign must be NULL */
1924	return bpf_mem_alloc(ma: &bpf_global_percpu_ma, size);
1925	}
1926
1927	/* Must be called under migrate_disable(), as required by bpf_mem_free */
1928	void __bpf_obj_drop_impl(void *p, const struct btf_record *rec, bool percpu)
1929	{
1930	struct bpf_mem_alloc *ma;
1931
1932	if (rec && rec->refcount_off >= 0 &&
1933	!refcount_dec_and_test(r: (refcount_t *)(p + rec->refcount_off))) {
1934	/* Object is refcounted and refcount_dec didn't result in 0
1935	* refcount. Return without freeing the object
1936	*/
1937	return;
1938	}
1939
1940	if (rec)
1941	bpf_obj_free_fields(rec, obj: p);
1942
1943	if (percpu)
1944	ma = &bpf_global_percpu_ma;
1945	else
1946	ma = &bpf_global_ma;
1947	bpf_mem_free_rcu(ma, ptr: p);
1948	}
1949
1950	__bpf_kfunc void bpf_obj_drop_impl(void *p__alloc, void *meta__ign)
1951	{
1952	struct btf_struct_meta *meta = meta__ign;
1953	void *p = p__alloc;
1954
1955	__bpf_obj_drop_impl(p, rec: meta ? meta->record : NULL, percpu: false);
1956	}
1957
1958	__bpf_kfunc void bpf_percpu_obj_drop_impl(void *p__alloc, void *meta__ign)
1959	{
1960	/* The verifier has ensured that meta__ign must be NULL */
1961	bpf_mem_free_rcu(ma: &bpf_global_percpu_ma, ptr: p__alloc);
1962	}
1963
1964	__bpf_kfunc void *bpf_refcount_acquire_impl(void *p__refcounted_kptr, void *meta__ign)
1965	{
1966	struct btf_struct_meta *meta = meta__ign;
1967	struct bpf_refcount *ref;
1968
1969	/* Could just cast directly to refcount_t *, but need some code using
1970	* bpf_refcount type so that it is emitted in vmlinux BTF
1971	*/
1972	ref = (struct bpf_refcount *)(p__refcounted_kptr + meta->record->refcount_off);
1973	if (!refcount_inc_not_zero(r: (refcount_t *)ref))
1974	return NULL;
1975
1976	/* Verifier strips KF_RET_NULL if input is owned ref, see is_kfunc_ret_null
1977	* in verifier.c
1978	*/
1979	return (void *)p__refcounted_kptr;
1980	}
1981
1982	static int __bpf_list_add(struct bpf_list_node_kern *node,
1983	struct bpf_list_head *head,
1984	bool tail, struct btf_record *rec, u64 off)
1985	{
1986	struct list_head *n = &node->list_head, *h = (void *)head;
1987
1988	/* If list_head was 0-initialized by map, bpf_obj_init_field wasn't
1989	* called on its fields, so init here
1990	*/
1991	if (unlikely(!h->next))
1992	INIT_LIST_HEAD(list: h);
1993
1994	/* node->owner != NULL implies !list_empty(n), no need to separately
1995	* check the latter
1996	*/
1997	if (cmpxchg(&node->owner, NULL, BPF_PTR_POISON)) {
1998	/* Only called from BPF prog, no need to migrate_disable */
1999	__bpf_obj_drop_impl(p: (void *)n - off, rec, percpu: false);
2000	return -EINVAL;
2001	}
2002
2003	tail ? list_add_tail(new: n, head: h) : list_add(new: n, head: h);
2004	WRITE_ONCE(node->owner, head);
2005
2006	return 0;
2007	}
2008
2009	__bpf_kfunc int bpf_list_push_front_impl(struct bpf_list_head *head,
2010	struct bpf_list_node *node,
2011	void *meta__ign, u64 off)
2012	{
2013	struct bpf_list_node_kern *n = (void *)node;
2014	struct btf_struct_meta *meta = meta__ign;
2015
2016	return __bpf_list_add(node: n, head, tail: false, rec: meta ? meta->record : NULL, off);
2017	}
2018
2019	__bpf_kfunc int bpf_list_push_back_impl(struct bpf_list_head *head,
2020	struct bpf_list_node *node,
2021	void *meta__ign, u64 off)
2022	{
2023	struct bpf_list_node_kern *n = (void *)node;
2024	struct btf_struct_meta *meta = meta__ign;
2025
2026	return __bpf_list_add(node: n, head, tail: true, rec: meta ? meta->record : NULL, off);
2027	}
2028
2029	static struct bpf_list_node *__bpf_list_del(struct bpf_list_head *head, bool tail)
2030	{
2031	struct list_head *n, *h = (void *)head;
2032	struct bpf_list_node_kern *node;
2033
2034	/* If list_head was 0-initialized by map, bpf_obj_init_field wasn't
2035	* called on its fields, so init here
2036	*/
2037	if (unlikely(!h->next))
2038	INIT_LIST_HEAD(list: h);
2039	if (list_empty(head: h))
2040	return NULL;
2041
2042	n = tail ? h->prev : h->next;
2043	node = container_of(n, struct bpf_list_node_kern, list_head);
2044	if (WARN_ON_ONCE(READ_ONCE(node->owner) != head))
2045	return NULL;
2046
2047	list_del_init(entry: n);
2048	WRITE_ONCE(node->owner, NULL);
2049	return (struct bpf_list_node *)n;
2050	}
2051
2052	__bpf_kfunc struct bpf_list_node *bpf_list_pop_front(struct bpf_list_head *head)
2053	{
2054	return __bpf_list_del(head, tail: false);
2055	}
2056
2057	__bpf_kfunc struct bpf_list_node *bpf_list_pop_back(struct bpf_list_head *head)
2058	{
2059	return __bpf_list_del(head, tail: true);
2060	}
2061
2062	__bpf_kfunc struct bpf_rb_node *bpf_rbtree_remove(struct bpf_rb_root *root,
2063	struct bpf_rb_node *node)
2064	{
2065	struct bpf_rb_node_kern *node_internal = (struct bpf_rb_node_kern *)node;
2066	struct rb_root_cached *r = (struct rb_root_cached *)root;
2067	struct rb_node *n = &node_internal->rb_node;
2068
2069	/* node_internal->owner != root implies either RB_EMPTY_NODE(n) or
2070	* n is owned by some other tree. No need to check RB_EMPTY_NODE(n)
2071	*/
2072	if (READ_ONCE(node_internal->owner) != root)
2073	return NULL;
2074
2075	rb_erase_cached(node: n, root: r);
2076	RB_CLEAR_NODE(n);
2077	WRITE_ONCE(node_internal->owner, NULL);
2078	return (struct bpf_rb_node *)n;
2079	}
2080
2081	/* Need to copy rbtree_add_cached's logic here because our 'less' is a BPF
2082	* program
2083	*/
2084	static int __bpf_rbtree_add(struct bpf_rb_root *root,
2085	struct bpf_rb_node_kern *node,
2086	void *less, struct btf_record *rec, u64 off)
2087	{
2088	struct rb_node **link = &((struct rb_root_cached *)root)->rb_root.rb_node;
2089	struct rb_node *parent = NULL, *n = &node->rb_node;
2090	bpf_callback_t cb = (bpf_callback_t)less;
2091	bool leftmost = true;
2092
2093	/* node->owner != NULL implies !RB_EMPTY_NODE(n), no need to separately
2094	* check the latter
2095	*/
2096	if (cmpxchg(&node->owner, NULL, BPF_PTR_POISON)) {
2097	/* Only called from BPF prog, no need to migrate_disable */
2098	__bpf_obj_drop_impl(p: (void *)n - off, rec, percpu: false);
2099	return -EINVAL;
2100	}
2101
2102	while (*link) {
2103	parent = *link;
2104	if (cb((uintptr_t)node, (uintptr_t)parent, 0, 0, 0)) {
2105	link = &parent->rb_left;
2106	} else {
2107	link = &parent->rb_right;
2108	leftmost = false;
2109	}
2110	}
2111
2112	rb_link_node(node: n, parent, rb_link: link);
2113	rb_insert_color_cached(node: n, root: (struct rb_root_cached *)root, leftmost);
2114	WRITE_ONCE(node->owner, root);
2115	return 0;
2116	}
2117
2118	__bpf_kfunc int bpf_rbtree_add_impl(struct bpf_rb_root *root, struct bpf_rb_node *node,
2119	bool (less)(struct bpf_rb_node *a, const struct bpf_rb_node *b),
2120	void *meta__ign, u64 off)
2121	{
2122	struct btf_struct_meta *meta = meta__ign;
2123	struct bpf_rb_node_kern *n = (void *)node;
2124
2125	return __bpf_rbtree_add(root, node: n, less: (void *)less, rec: meta ? meta->record : NULL, off);
2126	}
2127
2128	__bpf_kfunc struct bpf_rb_node *bpf_rbtree_first(struct bpf_rb_root *root)
2129	{
2130	struct rb_root_cached *r = (struct rb_root_cached *)root;
2131
2132	return (struct bpf_rb_node *)rb_first_cached(r);
2133	}
2134
2135	/**
2136	* bpf_task_acquire - Acquire a reference to a task. A task acquired by this
2137	* kfunc which is not stored in a map as a kptr, must be released by calling
2138	* bpf_task_release().
2139	* @p: The task on which a reference is being acquired.
2140	*/
2141	__bpf_kfunc struct task_struct *bpf_task_acquire(struct task_struct *p)
2142	{
2143	if (refcount_inc_not_zero(r: &p->rcu_users))
2144	return p;
2145	return NULL;
2146	}
2147
2148	/**
2149	* bpf_task_release - Release the reference acquired on a task.
2150	* @p: The task on which a reference is being released.
2151	*/
2152	__bpf_kfunc void bpf_task_release(struct task_struct *p)
2153	{
2154	put_task_struct_rcu_user(task: p);
2155	}
2156
2157	__bpf_kfunc void bpf_task_release_dtor(void *p)
2158	{
2159	put_task_struct_rcu_user(task: p);
2160	}
2161	CFI_NOSEAL(bpf_task_release_dtor);
2162
2163	#ifdef CONFIG_CGROUPS
2164	/**
2165	* bpf_cgroup_acquire - Acquire a reference to a cgroup. A cgroup acquired by
2166	* this kfunc which is not stored in a map as a kptr, must be released by
2167	* calling bpf_cgroup_release().
2168	* @cgrp: The cgroup on which a reference is being acquired.
2169	*/
2170	__bpf_kfunc struct cgroup *bpf_cgroup_acquire(struct cgroup *cgrp)
2171	{
2172	return cgroup_tryget(cgrp) ? cgrp : NULL;
2173	}
2174
2175	/**
2176	* bpf_cgroup_release - Release the reference acquired on a cgroup.
2177	* If this kfunc is invoked in an RCU read region, the cgroup is guaranteed to
2178	* not be freed until the current grace period has ended, even if its refcount
2179	* drops to 0.
2180	* @cgrp: The cgroup on which a reference is being released.
2181	*/
2182	__bpf_kfunc void bpf_cgroup_release(struct cgroup *cgrp)
2183	{
2184	cgroup_put(cgrp);
2185	}
2186
2187	__bpf_kfunc void bpf_cgroup_release_dtor(void *cgrp)
2188	{
2189	cgroup_put(cgrp);
2190	}
2191	CFI_NOSEAL(bpf_cgroup_release_dtor);
2192
2193	/**
2194	* bpf_cgroup_ancestor - Perform a lookup on an entry in a cgroup's ancestor
2195	* array. A cgroup returned by this kfunc which is not subsequently stored in a
2196	* map, must be released by calling bpf_cgroup_release().
2197	* @cgrp: The cgroup for which we're performing a lookup.
2198	* @level: The level of ancestor to look up.
2199	*/
2200	__bpf_kfunc struct cgroup *bpf_cgroup_ancestor(struct cgroup *cgrp, int level)
2201	{
2202	struct cgroup *ancestor;
2203
2204	if (level > cgrp->level || level < 0)
2205	return NULL;
2206
2207	/* cgrp's refcnt could be 0 here, but ancestors can still be accessed */
2208	ancestor = cgrp->ancestors[level];
2209	if (!cgroup_tryget(cgrp: ancestor))
2210	return NULL;
2211	return ancestor;
2212	}
2213
2214	/**
2215	* bpf_cgroup_from_id - Find a cgroup from its ID. A cgroup returned by this
2216	* kfunc which is not subsequently stored in a map, must be released by calling
2217	* bpf_cgroup_release().
2218	* @cgid: cgroup id.
2219	*/
2220	__bpf_kfunc struct cgroup *bpf_cgroup_from_id(u64 cgid)
2221	{
2222	struct cgroup *cgrp;
2223
2224	cgrp = cgroup_get_from_id(id: cgid);
2225	if (IS_ERR(ptr: cgrp))
2226	return NULL;
2227	return cgrp;
2228	}
2229
2230	/**
2231	* bpf_task_under_cgroup - wrap task_under_cgroup_hierarchy() as a kfunc, test
2232	* task's membership of cgroup ancestry.
2233	* @task: the task to be tested
2234	* @ancestor: possible ancestor of @task's cgroup
2235	*
2236	* Tests whether @task's default cgroup hierarchy is a descendant of @ancestor.
2237	* It follows all the same rules as cgroup_is_descendant, and only applies
2238	* to the default hierarchy.
2239	*/
2240	__bpf_kfunc long bpf_task_under_cgroup(struct task_struct *task,
2241	struct cgroup *ancestor)
2242	{
2243	long ret;
2244
2245	rcu_read_lock();
2246	ret = task_under_cgroup_hierarchy(task, ancestor);
2247	rcu_read_unlock();
2248	return ret;
2249	}
2250
2251	/**
2252	* bpf_task_get_cgroup1 - Acquires the associated cgroup of a task within a
2253	* specific cgroup1 hierarchy. The cgroup1 hierarchy is identified by its
2254	* hierarchy ID.
2255	* @task: The target task
2256	* @hierarchy_id: The ID of a cgroup1 hierarchy
2257	*
2258	* On success, the cgroup is returen. On failure, NULL is returned.
2259	*/
2260	__bpf_kfunc struct cgroup *
2261	bpf_task_get_cgroup1(struct task_struct *task, int hierarchy_id)
2262	{
2263	struct cgroup *cgrp = task_get_cgroup1(tsk: task, hierarchy_id);
2264
2265	if (IS_ERR(ptr: cgrp))
2266	return NULL;
2267	return cgrp;
2268	}
2269	#endif /* CONFIG_CGROUPS */
2270
2271	/**
2272	* bpf_task_from_pid - Find a struct task_struct from its pid by looking it up
2273	* in the root pid namespace idr. If a task is returned, it must either be
2274	* stored in a map, or released with bpf_task_release().
2275	* @pid: The pid of the task being looked up.
2276	*/
2277	__bpf_kfunc struct task_struct *bpf_task_from_pid(s32 pid)
2278	{
2279	struct task_struct *p;
2280
2281	rcu_read_lock();
2282	p = find_task_by_pid_ns(nr: pid, ns: &init_pid_ns);
2283	if (p)
2284	p = bpf_task_acquire(p);
2285	rcu_read_unlock();
2286
2287	return p;
2288	}
2289
2290	/**
2291	* bpf_dynptr_slice() - Obtain a read-only pointer to the dynptr data.
2292	* @ptr: The dynptr whose data slice to retrieve
2293	* @offset: Offset into the dynptr
2294	* @buffer__opt: User-provided buffer to copy contents into. May be NULL
2295	* @buffer__szk: Size (in bytes) of the buffer if present. This is the
2296	* length of the requested slice. This must be a constant.
2297	*
2298	* For non-skb and non-xdp type dynptrs, there is no difference between
2299	* bpf_dynptr_slice and bpf_dynptr_data.
2300	*
2301	* If buffer__opt is NULL, the call will fail if buffer_opt was needed.
2302	*
2303	* If the intention is to write to the data slice, please use
2304	* bpf_dynptr_slice_rdwr.
2305	*
2306	* The user must check that the returned pointer is not null before using it.
2307	*
2308	* Please note that in the case of skb and xdp dynptrs, bpf_dynptr_slice
2309	* does not change the underlying packet data pointers, so a call to
2310	* bpf_dynptr_slice will not invalidate any ctx->data/data_end pointers in
2311	* the bpf program.
2312	*
2313	* Return: NULL if the call failed (eg invalid dynptr), pointer to a read-only
2314	* data slice (can be either direct pointer to the data or a pointer to the user
2315	* provided buffer, with its contents containing the data, if unable to obtain
2316	* direct pointer)
2317	*/
2318	__bpf_kfunc void *bpf_dynptr_slice(const struct bpf_dynptr_kern *ptr, u32 offset,
2319	void *buffer__opt, u32 buffer__szk)
2320	{
2321	enum bpf_dynptr_type type;
2322	u32 len = buffer__szk;
2323	int err;
2324
2325	if (!ptr->data)
2326	return NULL;
2327
2328	err = bpf_dynptr_check_off_len(ptr, offset, len);
2329	if (err)
2330	return NULL;
2331
2332	type = bpf_dynptr_get_type(ptr);
2333
2334	switch (type) {
2335	case BPF_DYNPTR_TYPE_LOCAL:
2336	case BPF_DYNPTR_TYPE_RINGBUF:
2337	return ptr->data + ptr->offset + offset;
2338	case BPF_DYNPTR_TYPE_SKB:
2339	if (buffer__opt)
2340	return skb_header_pointer(skb: ptr->data, offset: ptr->offset + offset, len, buffer: buffer__opt);
2341	else
2342	return skb_pointer_if_linear(skb: ptr->data, offset: ptr->offset + offset, len);
2343	case BPF_DYNPTR_TYPE_XDP:
2344	{
2345	void *xdp_ptr = bpf_xdp_pointer(xdp: ptr->data, offset: ptr->offset + offset, len);
2346	if (!IS_ERR_OR_NULL(ptr: xdp_ptr))
2347	return xdp_ptr;
2348
2349	if (!buffer__opt)
2350	return NULL;
2351	bpf_xdp_copy_buf(xdp: ptr->data, off: ptr->offset + offset, buf: buffer__opt, len, flush: false);
2352	return buffer__opt;
2353	}
2354	default:
2355	WARN_ONCE(true, "unknown dynptr type %d\n", type);
2356	return NULL;
2357	}
2358	}
2359
2360	/**
2361	* bpf_dynptr_slice_rdwr() - Obtain a writable pointer to the dynptr data.
2362	* @ptr: The dynptr whose data slice to retrieve
2363	* @offset: Offset into the dynptr
2364	* @buffer__opt: User-provided buffer to copy contents into. May be NULL
2365	* @buffer__szk: Size (in bytes) of the buffer if present. This is the
2366	* length of the requested slice. This must be a constant.
2367	*
2368	* For non-skb and non-xdp type dynptrs, there is no difference between
2369	* bpf_dynptr_slice and bpf_dynptr_data.
2370	*
2371	* If buffer__opt is NULL, the call will fail if buffer_opt was needed.
2372	*
2373	* The returned pointer is writable and may point to either directly the dynptr
2374	* data at the requested offset or to the buffer if unable to obtain a direct
2375	* data pointer to (example: the requested slice is to the paged area of an skb
2376	* packet). In the case where the returned pointer is to the buffer, the user
2377	* is responsible for persisting writes through calling bpf_dynptr_write(). This
2378	* usually looks something like this pattern:
2379	*
2380	* struct eth_hdr *eth = bpf_dynptr_slice_rdwr(&dynptr, 0, buffer, sizeof(buffer));
2381	* if (!eth)
2382	* return TC_ACT_SHOT;
2383	*
2384	* // mutate eth header //
2385	*
2386	* if (eth == buffer)
2387	* bpf_dynptr_write(&ptr, 0, buffer, sizeof(buffer), 0);
2388	*
2389	* Please note that, as in the example above, the user must check that the
2390	* returned pointer is not null before using it.
2391	*
2392	* Please also note that in the case of skb and xdp dynptrs, bpf_dynptr_slice_rdwr
2393	* does not change the underlying packet data pointers, so a call to
2394	* bpf_dynptr_slice_rdwr will not invalidate any ctx->data/data_end pointers in
2395	* the bpf program.
2396	*
2397	* Return: NULL if the call failed (eg invalid dynptr), pointer to a
2398	* data slice (can be either direct pointer to the data or a pointer to the user
2399	* provided buffer, with its contents containing the data, if unable to obtain
2400	* direct pointer)
2401	*/
2402	__bpf_kfunc void *bpf_dynptr_slice_rdwr(const struct bpf_dynptr_kern *ptr, u32 offset,
2403	void *buffer__opt, u32 buffer__szk)
2404	{
2405	if (!ptr->data || __bpf_dynptr_is_rdonly(ptr))
2406	return NULL;
2407
2408	/* bpf_dynptr_slice_rdwr is the same logic as bpf_dynptr_slice.
2409	*
2410	* For skb-type dynptrs, it is safe to write into the returned pointer
2411	* if the bpf program allows skb data writes. There are two possiblities
2412	* that may occur when calling bpf_dynptr_slice_rdwr:
2413	*
2414	* 1) The requested slice is in the head of the skb. In this case, the
2415	* returned pointer is directly to skb data, and if the skb is cloned, the
2416	* verifier will have uncloned it (see bpf_unclone_prologue()) already.
2417	* The pointer can be directly written into.
2418	*
2419	* 2) Some portion of the requested slice is in the paged buffer area.
2420	* In this case, the requested data will be copied out into the buffer
2421	* and the returned pointer will be a pointer to the buffer. The skb
2422	* will not be pulled. To persist the write, the user will need to call
2423	* bpf_dynptr_write(), which will pull the skb and commit the write.
2424	*
2425	* Similarly for xdp programs, if the requested slice is not across xdp
2426	* fragments, then a direct pointer will be returned, otherwise the data
2427	* will be copied out into the buffer and the user will need to call
2428	* bpf_dynptr_write() to commit changes.
2429	*/
2430	return bpf_dynptr_slice(ptr, offset, buffer__opt, buffer__szk);
2431	}
2432
2433	__bpf_kfunc int bpf_dynptr_adjust(struct bpf_dynptr_kern *ptr, u32 start, u32 end)
2434	{
2435	u32 size;
2436
2437	if (!ptr->data || start > end)
2438	return -EINVAL;
2439
2440	size = __bpf_dynptr_size(ptr);
2441
2442	if (start > size || end > size)
2443	return -ERANGE;
2444
2445	ptr->offset += start;
2446	bpf_dynptr_set_size(ptr, new_size: end - start);
2447
2448	return 0;
2449	}
2450
2451	__bpf_kfunc bool bpf_dynptr_is_null(struct bpf_dynptr_kern *ptr)
2452	{
2453	return !ptr->data;
2454	}
2455
2456	__bpf_kfunc bool bpf_dynptr_is_rdonly(struct bpf_dynptr_kern *ptr)
2457	{
2458	if (!ptr->data)
2459	return false;
2460
2461	return __bpf_dynptr_is_rdonly(ptr);
2462	}
2463
2464	__bpf_kfunc __u32 bpf_dynptr_size(const struct bpf_dynptr_kern *ptr)
2465	{
2466	if (!ptr->data)
2467	return -EINVAL;
2468
2469	return __bpf_dynptr_size(ptr);
2470	}
2471
2472	__bpf_kfunc int bpf_dynptr_clone(struct bpf_dynptr_kern *ptr,
2473	struct bpf_dynptr_kern *clone__uninit)
2474	{
2475	if (!ptr->data) {
2476	bpf_dynptr_set_null(ptr: clone__uninit);
2477	return -EINVAL;
2478	}
2479
2480	*clone__uninit = *ptr;
2481
2482	return 0;
2483	}
2484
2485	__bpf_kfunc void *bpf_cast_to_kern_ctx(void *obj)
2486	{
2487	return obj;
2488	}
2489
2490	__bpf_kfunc void *bpf_rdonly_cast(const void *obj__ign, u32 btf_id__k)
2491	{
2492	return (void *)obj__ign;
2493	}
2494
2495	__bpf_kfunc void bpf_rcu_read_lock(void)
2496	{
2497	rcu_read_lock();
2498	}
2499
2500	__bpf_kfunc void bpf_rcu_read_unlock(void)
2501	{
2502	rcu_read_unlock();
2503	}
2504
2505	struct bpf_throw_ctx {
2506	struct bpf_prog_aux *aux;
2507	u64 sp;
2508	u64 bp;
2509	int cnt;
2510	};
2511
2512	static bool bpf_stack_walker(void *cookie, u64 ip, u64 sp, u64 bp)
2513	{
2514	struct bpf_throw_ctx *ctx = cookie;
2515	struct bpf_prog *prog;
2516
2517	if (!is_bpf_text_address(addr: ip))
2518	return !ctx->cnt;
2519	prog = bpf_prog_ksym_find(addr: ip);
2520	ctx->cnt++;
2521	if (bpf_is_subprog(prog))
2522	return true;
2523	ctx->aux = prog->aux;
2524	ctx->sp = sp;
2525	ctx->bp = bp;
2526	return false;
2527	}
2528
2529	__bpf_kfunc void bpf_throw(u64 cookie)
2530	{
2531	struct bpf_throw_ctx ctx = {};
2532
2533	arch_bpf_stack_walk(consume_fn: bpf_stack_walker, cookie: &ctx);
2534	WARN_ON_ONCE(!ctx.aux);
2535	if (ctx.aux)
2536	WARN_ON_ONCE(!ctx.aux->exception_boundary);
2537	WARN_ON_ONCE(!ctx.bp);
2538	WARN_ON_ONCE(!ctx.cnt);
2539	/* Prevent KASAN false positives for CONFIG_KASAN_STACK by unpoisoning
2540	* deeper stack depths than ctx.sp as we do not return from bpf_throw,
2541	* which skips compiler generated instrumentation to do the same.
2542	*/
2543	kasan_unpoison_task_stack_below(watermark: (void *)(long)ctx.sp);
2544	ctx.aux->bpf_exception_cb(cookie, ctx.sp, ctx.bp, 0, 0);
2545	WARN(1, "A call to BPF exception callback should never return\n");
2546	}
2547
2548	__bpf_kfunc_end_defs();
2549
2550	BTF_KFUNCS_START(generic_btf_ids)
2551	#ifdef CONFIG_CRASH_DUMP
2552	BTF_ID_FLAGS(func, crash_kexec, KF_DESTRUCTIVE)
2553	#endif
2554	BTF_ID_FLAGS(func, bpf_obj_new_impl, KF_ACQUIRE | KF_RET_NULL)
2555	BTF_ID_FLAGS(func, bpf_percpu_obj_new_impl, KF_ACQUIRE | KF_RET_NULL)
2556	BTF_ID_FLAGS(func, bpf_obj_drop_impl, KF_RELEASE)
2557	BTF_ID_FLAGS(func, bpf_percpu_obj_drop_impl, KF_RELEASE)
2558	BTF_ID_FLAGS(func, bpf_refcount_acquire_impl, KF_ACQUIRE | KF_RET_NULL | KF_RCU)
2559	BTF_ID_FLAGS(func, bpf_list_push_front_impl)
2560	BTF_ID_FLAGS(func, bpf_list_push_back_impl)
2561	BTF_ID_FLAGS(func, bpf_list_pop_front, KF_ACQUIRE | KF_RET_NULL)
2562	BTF_ID_FLAGS(func, bpf_list_pop_back, KF_ACQUIRE | KF_RET_NULL)
2563	BTF_ID_FLAGS(func, bpf_task_acquire, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
2564	BTF_ID_FLAGS(func, bpf_task_release, KF_RELEASE)
2565	BTF_ID_FLAGS(func, bpf_rbtree_remove, KF_ACQUIRE | KF_RET_NULL)
2566	BTF_ID_FLAGS(func, bpf_rbtree_add_impl)
2567	BTF_ID_FLAGS(func, bpf_rbtree_first, KF_RET_NULL)
2568
2569	#ifdef CONFIG_CGROUPS
2570	BTF_ID_FLAGS(func, bpf_cgroup_acquire, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
2571	BTF_ID_FLAGS(func, bpf_cgroup_release, KF_RELEASE)
2572	BTF_ID_FLAGS(func, bpf_cgroup_ancestor, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
2573	BTF_ID_FLAGS(func, bpf_cgroup_from_id, KF_ACQUIRE | KF_RET_NULL)
2574	BTF_ID_FLAGS(func, bpf_task_under_cgroup, KF_RCU)
2575	BTF_ID_FLAGS(func, bpf_task_get_cgroup1, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
2576	#endif
2577	BTF_ID_FLAGS(func, bpf_task_from_pid, KF_ACQUIRE | KF_RET_NULL)
2578	BTF_ID_FLAGS(func, bpf_throw)
2579	BTF_KFUNCS_END(generic_btf_ids)
2580
2581	static const struct btf_kfunc_id_set generic_kfunc_set = {
2582	.owner = THIS_MODULE,
2583	.set = &generic_btf_ids,
2584	};
2585
2586
2587	BTF_ID_LIST(generic_dtor_ids)
2588	BTF_ID(struct, task_struct)
2589	BTF_ID(func, bpf_task_release_dtor)
2590	#ifdef CONFIG_CGROUPS
2591	BTF_ID(struct, cgroup)
2592	BTF_ID(func, bpf_cgroup_release_dtor)
2593	#endif
2594
2595	BTF_KFUNCS_START(common_btf_ids)
2596	BTF_ID_FLAGS(func, bpf_cast_to_kern_ctx)
2597	BTF_ID_FLAGS(func, bpf_rdonly_cast)
2598	BTF_ID_FLAGS(func, bpf_rcu_read_lock)
2599	BTF_ID_FLAGS(func, bpf_rcu_read_unlock)
2600	BTF_ID_FLAGS(func, bpf_dynptr_slice, KF_RET_NULL)
2601	BTF_ID_FLAGS(func, bpf_dynptr_slice_rdwr, KF_RET_NULL)
2602	BTF_ID_FLAGS(func, bpf_iter_num_new, KF_ITER_NEW)
2603	BTF_ID_FLAGS(func, bpf_iter_num_next, KF_ITER_NEXT | KF_RET_NULL)
2604	BTF_ID_FLAGS(func, bpf_iter_num_destroy, KF_ITER_DESTROY)
2605	BTF_ID_FLAGS(func, bpf_iter_task_vma_new, KF_ITER_NEW | KF_RCU)
2606	BTF_ID_FLAGS(func, bpf_iter_task_vma_next, KF_ITER_NEXT | KF_RET_NULL)
2607	BTF_ID_FLAGS(func, bpf_iter_task_vma_destroy, KF_ITER_DESTROY)
2608	#ifdef CONFIG_CGROUPS
2609	BTF_ID_FLAGS(func, bpf_iter_css_task_new, KF_ITER_NEW | KF_TRUSTED_ARGS)
2610	BTF_ID_FLAGS(func, bpf_iter_css_task_next, KF_ITER_NEXT | KF_RET_NULL)
2611	BTF_ID_FLAGS(func, bpf_iter_css_task_destroy, KF_ITER_DESTROY)
2612	BTF_ID_FLAGS(func, bpf_iter_css_new, KF_ITER_NEW | KF_TRUSTED_ARGS | KF_RCU_PROTECTED)
2613	BTF_ID_FLAGS(func, bpf_iter_css_next, KF_ITER_NEXT | KF_RET_NULL)
2614	BTF_ID_FLAGS(func, bpf_iter_css_destroy, KF_ITER_DESTROY)
2615	#endif
2616	BTF_ID_FLAGS(func, bpf_iter_task_new, KF_ITER_NEW | KF_TRUSTED_ARGS | KF_RCU_PROTECTED)
2617	BTF_ID_FLAGS(func, bpf_iter_task_next, KF_ITER_NEXT | KF_RET_NULL)
2618	BTF_ID_FLAGS(func, bpf_iter_task_destroy, KF_ITER_DESTROY)
2619	BTF_ID_FLAGS(func, bpf_dynptr_adjust)
2620	BTF_ID_FLAGS(func, bpf_dynptr_is_null)
2621	BTF_ID_FLAGS(func, bpf_dynptr_is_rdonly)
2622	BTF_ID_FLAGS(func, bpf_dynptr_size)
2623	BTF_ID_FLAGS(func, bpf_dynptr_clone)
2624	BTF_KFUNCS_END(common_btf_ids)
2625
2626	static const struct btf_kfunc_id_set common_kfunc_set = {
2627	.owner = THIS_MODULE,
2628	.set = &common_btf_ids,
2629	};
2630
2631	static int __init kfunc_init(void)
2632	{
2633	int ret;
2634	const struct btf_id_dtor_kfunc generic_dtors[] = {
2635	{
2636	.btf_id = generic_dtor_ids[0],
2637	.kfunc_btf_id = generic_dtor_ids[1]
2638	},
2639	#ifdef CONFIG_CGROUPS
2640	{
2641	.btf_id = generic_dtor_ids[2],
2642	.kfunc_btf_id = generic_dtor_ids[3]
2643	},
2644	#endif
2645	};
2646
2647	ret = register_btf_kfunc_id_set(prog_type: BPF_PROG_TYPE_TRACING, s: &generic_kfunc_set);
2648	ret = ret ?: register_btf_kfunc_id_set(prog_type: BPF_PROG_TYPE_SCHED_CLS, s: &generic_kfunc_set);
2649	ret = ret ?: register_btf_kfunc_id_set(prog_type: BPF_PROG_TYPE_XDP, s: &generic_kfunc_set);
2650	ret = ret ?: register_btf_kfunc_id_set(prog_type: BPF_PROG_TYPE_STRUCT_OPS, s: &generic_kfunc_set);
2651	ret = ret ?: register_btf_id_dtor_kfuncs(dtors: generic_dtors,
2652	ARRAY_SIZE(generic_dtors),
2653	THIS_MODULE);
2654	return ret ?: register_btf_kfunc_id_set(prog_type: BPF_PROG_TYPE_UNSPEC, s: &common_kfunc_set);
2655	}
2656
2657	late_initcall(kfunc_init);
2658
2659	/* Get a pointer to dynptr data up to len bytes for read only access. If
2660	* the dynptr doesn't have continuous data up to len bytes, return NULL.
2661	*/
2662	const void *__bpf_dynptr_data(const struct bpf_dynptr_kern *ptr, u32 len)
2663	{
2664	return bpf_dynptr_slice(ptr, offset: 0, NULL, buffer__szk: len);
2665	}
2666
2667	/* Get a pointer to dynptr data up to len bytes for read write access. If
2668	* the dynptr doesn't have continuous data up to len bytes, or the dynptr
2669	* is read only, return NULL.
2670	*/
2671	void *__bpf_dynptr_data_rw(const struct bpf_dynptr_kern *ptr, u32 len)
2672	{
2673	if (__bpf_dynptr_is_rdonly(ptr))
2674	return NULL;
2675	return (void *)__bpf_dynptr_data(ptr, len);
2676	}
2677