xref: /linux/kernel/bpf/helpers.c (revision f4db95b68ae68ebaf91d35cc0487ac1cbd04261e)
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 | MEM_WRITE,
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 | MEM_WRITE,
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 	.allow_fastcall	= true,
162 };
163 
164 BPF_CALL_0(bpf_get_numa_node_id)
165 {
166 	return numa_node_id();
167 }
168 
169 const struct bpf_func_proto bpf_get_numa_node_id_proto = {
170 	.func		= bpf_get_numa_node_id,
171 	.gpl_only	= false,
172 	.ret_type	= RET_INTEGER,
173 };
174 
175 BPF_CALL_0(bpf_ktime_get_ns)
176 {
177 	/* NMI safe access to clock monotonic */
178 	return ktime_get_mono_fast_ns();
179 }
180 
181 const struct bpf_func_proto bpf_ktime_get_ns_proto = {
182 	.func		= bpf_ktime_get_ns,
183 	.gpl_only	= false,
184 	.ret_type	= RET_INTEGER,
185 };
186 
187 BPF_CALL_0(bpf_ktime_get_boot_ns)
188 {
189 	/* NMI safe access to clock boottime */
190 	return ktime_get_boot_fast_ns();
191 }
192 
193 const struct bpf_func_proto bpf_ktime_get_boot_ns_proto = {
194 	.func		= bpf_ktime_get_boot_ns,
195 	.gpl_only	= false,
196 	.ret_type	= RET_INTEGER,
197 };
198 
199 BPF_CALL_0(bpf_ktime_get_coarse_ns)
200 {
201 	return ktime_get_coarse_ns();
202 }
203 
204 const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto = {
205 	.func		= bpf_ktime_get_coarse_ns,
206 	.gpl_only	= false,
207 	.ret_type	= RET_INTEGER,
208 };
209 
210 BPF_CALL_0(bpf_ktime_get_tai_ns)
211 {
212 	/* NMI safe access to clock tai */
213 	return ktime_get_tai_fast_ns();
214 }
215 
216 const struct bpf_func_proto bpf_ktime_get_tai_ns_proto = {
217 	.func		= bpf_ktime_get_tai_ns,
218 	.gpl_only	= false,
219 	.ret_type	= RET_INTEGER,
220 };
221 
222 BPF_CALL_0(bpf_get_current_pid_tgid)
223 {
224 	struct task_struct *task = current;
225 
226 	if (unlikely(!task))
227 		return -EINVAL;
228 
229 	return (u64) task->tgid << 32 | task->pid;
230 }
231 
232 const struct bpf_func_proto bpf_get_current_pid_tgid_proto = {
233 	.func		= bpf_get_current_pid_tgid,
234 	.gpl_only	= false,
235 	.ret_type	= RET_INTEGER,
236 };
237 
238 BPF_CALL_0(bpf_get_current_uid_gid)
239 {
240 	struct task_struct *task = current;
241 	kuid_t uid;
242 	kgid_t gid;
243 
244 	if (unlikely(!task))
245 		return -EINVAL;
246 
247 	current_uid_gid(&uid, &gid);
248 	return (u64) from_kgid(&init_user_ns, gid) << 32 |
249 		     from_kuid(&init_user_ns, uid);
250 }
251 
252 const struct bpf_func_proto bpf_get_current_uid_gid_proto = {
253 	.func		= bpf_get_current_uid_gid,
254 	.gpl_only	= false,
255 	.ret_type	= RET_INTEGER,
256 };
257 
258 BPF_CALL_2(bpf_get_current_comm, char *, buf, u32, size)
259 {
260 	struct task_struct *task = current;
261 
262 	if (unlikely(!task))
263 		goto err_clear;
264 
265 	/* Verifier guarantees that size > 0 */
266 	strscpy_pad(buf, task->comm, size);
267 	return 0;
268 err_clear:
269 	memset(buf, 0, size);
270 	return -EINVAL;
271 }
272 
273 const struct bpf_func_proto bpf_get_current_comm_proto = {
274 	.func		= bpf_get_current_comm,
275 	.gpl_only	= false,
276 	.ret_type	= RET_INTEGER,
277 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
278 	.arg2_type	= ARG_CONST_SIZE,
279 };
280 
281 #if defined(CONFIG_QUEUED_SPINLOCKS) || defined(CONFIG_BPF_ARCH_SPINLOCK)
282 
283 static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
284 {
285 	arch_spinlock_t *l = (void *)lock;
286 	union {
287 		__u32 val;
288 		arch_spinlock_t lock;
289 	} u = { .lock = __ARCH_SPIN_LOCK_UNLOCKED };
290 
291 	compiletime_assert(u.val == 0, "__ARCH_SPIN_LOCK_UNLOCKED not 0");
292 	BUILD_BUG_ON(sizeof(*l) != sizeof(__u32));
293 	BUILD_BUG_ON(sizeof(*lock) != sizeof(__u32));
294 	preempt_disable();
295 	arch_spin_lock(l);
296 }
297 
298 static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
299 {
300 	arch_spinlock_t *l = (void *)lock;
301 
302 	arch_spin_unlock(l);
303 	preempt_enable();
304 }
305 
306 #else
307 
308 static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
309 {
310 	atomic_t *l = (void *)lock;
311 
312 	BUILD_BUG_ON(sizeof(*l) != sizeof(*lock));
313 	do {
314 		atomic_cond_read_relaxed(l, !VAL);
315 	} while (atomic_xchg(l, 1));
316 }
317 
318 static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
319 {
320 	atomic_t *l = (void *)lock;
321 
322 	atomic_set_release(l, 0);
323 }
324 
325 #endif
326 
327 static DEFINE_PER_CPU(unsigned long, irqsave_flags);
328 
329 static inline void __bpf_spin_lock_irqsave(struct bpf_spin_lock *lock)
330 {
331 	unsigned long flags;
332 
333 	local_irq_save(flags);
334 	__bpf_spin_lock(lock);
335 	__this_cpu_write(irqsave_flags, flags);
336 }
337 
338 NOTRACE_BPF_CALL_1(bpf_spin_lock, struct bpf_spin_lock *, lock)
339 {
340 	__bpf_spin_lock_irqsave(lock);
341 	return 0;
342 }
343 
344 const struct bpf_func_proto bpf_spin_lock_proto = {
345 	.func		= bpf_spin_lock,
346 	.gpl_only	= false,
347 	.ret_type	= RET_VOID,
348 	.arg1_type	= ARG_PTR_TO_SPIN_LOCK,
349 	.arg1_btf_id    = BPF_PTR_POISON,
350 };
351 
352 static inline void __bpf_spin_unlock_irqrestore(struct bpf_spin_lock *lock)
353 {
354 	unsigned long flags;
355 
356 	flags = __this_cpu_read(irqsave_flags);
357 	__bpf_spin_unlock(lock);
358 	local_irq_restore(flags);
359 }
360 
361 NOTRACE_BPF_CALL_1(bpf_spin_unlock, struct bpf_spin_lock *, lock)
362 {
363 	__bpf_spin_unlock_irqrestore(lock);
364 	return 0;
365 }
366 
367 const struct bpf_func_proto bpf_spin_unlock_proto = {
368 	.func		= bpf_spin_unlock,
369 	.gpl_only	= false,
370 	.ret_type	= RET_VOID,
371 	.arg1_type	= ARG_PTR_TO_SPIN_LOCK,
372 	.arg1_btf_id    = BPF_PTR_POISON,
373 };
374 
375 void copy_map_value_locked(struct bpf_map *map, void *dst, void *src,
376 			   bool lock_src)
377 {
378 	struct bpf_spin_lock *lock;
379 
380 	if (lock_src)
381 		lock = src + map->record->spin_lock_off;
382 	else
383 		lock = dst + map->record->spin_lock_off;
384 	preempt_disable();
385 	__bpf_spin_lock_irqsave(lock);
386 	copy_map_value(map, dst, src);
387 	__bpf_spin_unlock_irqrestore(lock);
388 	preempt_enable();
389 }
390 
391 BPF_CALL_0(bpf_jiffies64)
392 {
393 	return get_jiffies_64();
394 }
395 
396 const struct bpf_func_proto bpf_jiffies64_proto = {
397 	.func		= bpf_jiffies64,
398 	.gpl_only	= false,
399 	.ret_type	= RET_INTEGER,
400 };
401 
402 #ifdef CONFIG_CGROUPS
403 BPF_CALL_0(bpf_get_current_cgroup_id)
404 {
405 	struct cgroup *cgrp;
406 	u64 cgrp_id;
407 
408 	rcu_read_lock();
409 	cgrp = task_dfl_cgroup(current);
410 	cgrp_id = cgroup_id(cgrp);
411 	rcu_read_unlock();
412 
413 	return cgrp_id;
414 }
415 
416 const struct bpf_func_proto bpf_get_current_cgroup_id_proto = {
417 	.func		= bpf_get_current_cgroup_id,
418 	.gpl_only	= false,
419 	.ret_type	= RET_INTEGER,
420 };
421 
422 BPF_CALL_1(bpf_get_current_ancestor_cgroup_id, int, ancestor_level)
423 {
424 	struct cgroup *cgrp;
425 	struct cgroup *ancestor;
426 	u64 cgrp_id;
427 
428 	rcu_read_lock();
429 	cgrp = task_dfl_cgroup(current);
430 	ancestor = cgroup_ancestor(cgrp, ancestor_level);
431 	cgrp_id = ancestor ? cgroup_id(ancestor) : 0;
432 	rcu_read_unlock();
433 
434 	return cgrp_id;
435 }
436 
437 const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto = {
438 	.func		= bpf_get_current_ancestor_cgroup_id,
439 	.gpl_only	= false,
440 	.ret_type	= RET_INTEGER,
441 	.arg1_type	= ARG_ANYTHING,
442 };
443 #endif /* CONFIG_CGROUPS */
444 
445 #define BPF_STRTOX_BASE_MASK 0x1F
446 
447 static int __bpf_strtoull(const char *buf, size_t buf_len, u64 flags,
448 			  unsigned long long *res, bool *is_negative)
449 {
450 	unsigned int base = flags & BPF_STRTOX_BASE_MASK;
451 	const char *cur_buf = buf;
452 	size_t cur_len = buf_len;
453 	unsigned int consumed;
454 	size_t val_len;
455 	char str[64];
456 
457 	if (!buf || !buf_len || !res || !is_negative)
458 		return -EINVAL;
459 
460 	if (base != 0 && base != 8 && base != 10 && base != 16)
461 		return -EINVAL;
462 
463 	if (flags & ~BPF_STRTOX_BASE_MASK)
464 		return -EINVAL;
465 
466 	while (cur_buf < buf + buf_len && isspace(*cur_buf))
467 		++cur_buf;
468 
469 	*is_negative = (cur_buf < buf + buf_len && *cur_buf == '-');
470 	if (*is_negative)
471 		++cur_buf;
472 
473 	consumed = cur_buf - buf;
474 	cur_len -= consumed;
475 	if (!cur_len)
476 		return -EINVAL;
477 
478 	cur_len = min(cur_len, sizeof(str) - 1);
479 	memcpy(str, cur_buf, cur_len);
480 	str[cur_len] = '\0';
481 	cur_buf = str;
482 
483 	cur_buf = _parse_integer_fixup_radix(cur_buf, &base);
484 	val_len = _parse_integer(cur_buf, base, res);
485 
486 	if (val_len & KSTRTOX_OVERFLOW)
487 		return -ERANGE;
488 
489 	if (val_len == 0)
490 		return -EINVAL;
491 
492 	cur_buf += val_len;
493 	consumed += cur_buf - str;
494 
495 	return consumed;
496 }
497 
498 static int __bpf_strtoll(const char *buf, size_t buf_len, u64 flags,
499 			 long long *res)
500 {
501 	unsigned long long _res;
502 	bool is_negative;
503 	int err;
504 
505 	err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
506 	if (err < 0)
507 		return err;
508 	if (is_negative) {
509 		if ((long long)-_res > 0)
510 			return -ERANGE;
511 		*res = -_res;
512 	} else {
513 		if ((long long)_res < 0)
514 			return -ERANGE;
515 		*res = _res;
516 	}
517 	return err;
518 }
519 
520 BPF_CALL_4(bpf_strtol, const char *, buf, size_t, buf_len, u64, flags,
521 	   s64 *, res)
522 {
523 	long long _res;
524 	int err;
525 
526 	*res = 0;
527 	err = __bpf_strtoll(buf, buf_len, flags, &_res);
528 	if (err < 0)
529 		return err;
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_FIXED_SIZE_MEM | MEM_UNINIT | MEM_WRITE | MEM_ALIGNED,
542 	.arg4_size	= sizeof(s64),
543 };
544 
545 BPF_CALL_4(bpf_strtoul, const char *, buf, size_t, buf_len, u64, flags,
546 	   u64 *, res)
547 {
548 	unsigned long long _res;
549 	bool is_negative;
550 	int err;
551 
552 	*res = 0;
553 	err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
554 	if (err < 0)
555 		return err;
556 	if (is_negative)
557 		return -EINVAL;
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_FIXED_SIZE_MEM | MEM_UNINIT | MEM_WRITE | MEM_ALIGNED,
570 	.arg4_size	= sizeof(u64),
571 };
572 
573 BPF_CALL_3(bpf_strncmp, const char *, s1, u32, s1_sz, const char *, s2)
574 {
575 	return strncmp(s1, s2, s1_sz);
576 }
577 
578 static const struct bpf_func_proto bpf_strncmp_proto = {
579 	.func		= bpf_strncmp,
580 	.gpl_only	= false,
581 	.ret_type	= RET_INTEGER,
582 	.arg1_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
583 	.arg2_type	= ARG_CONST_SIZE,
584 	.arg3_type	= ARG_PTR_TO_CONST_STR,
585 };
586 
587 BPF_CALL_4(bpf_get_ns_current_pid_tgid, u64, dev, u64, ino,
588 	   struct bpf_pidns_info *, nsdata, u32, size)
589 {
590 	struct task_struct *task = current;
591 	struct pid_namespace *pidns;
592 	int err = -EINVAL;
593 
594 	if (unlikely(size != sizeof(struct bpf_pidns_info)))
595 		goto clear;
596 
597 	if (unlikely((u64)(dev_t)dev != dev))
598 		goto clear;
599 
600 	if (unlikely(!task))
601 		goto clear;
602 
603 	pidns = task_active_pid_ns(task);
604 	if (unlikely(!pidns)) {
605 		err = -ENOENT;
606 		goto clear;
607 	}
608 
609 	if (!ns_match(&pidns->ns, (dev_t)dev, ino))
610 		goto clear;
611 
612 	nsdata->pid = task_pid_nr_ns(task, pidns);
613 	nsdata->tgid = task_tgid_nr_ns(task, pidns);
614 	return 0;
615 clear:
616 	memset((void *)nsdata, 0, (size_t) size);
617 	return err;
618 }
619 
620 const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto = {
621 	.func		= bpf_get_ns_current_pid_tgid,
622 	.gpl_only	= false,
623 	.ret_type	= RET_INTEGER,
624 	.arg1_type	= ARG_ANYTHING,
625 	.arg2_type	= ARG_ANYTHING,
626 	.arg3_type      = ARG_PTR_TO_UNINIT_MEM,
627 	.arg4_type      = ARG_CONST_SIZE,
628 };
629 
630 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
631 	.func		= bpf_get_raw_cpu_id,
632 	.gpl_only	= false,
633 	.ret_type	= RET_INTEGER,
634 };
635 
636 BPF_CALL_5(bpf_event_output_data, void *, ctx, struct bpf_map *, map,
637 	   u64, flags, void *, data, u64, size)
638 {
639 	if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
640 		return -EINVAL;
641 
642 	return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
643 }
644 
645 const struct bpf_func_proto bpf_event_output_data_proto =  {
646 	.func		= bpf_event_output_data,
647 	.gpl_only       = true,
648 	.ret_type       = RET_INTEGER,
649 	.arg1_type      = ARG_PTR_TO_CTX,
650 	.arg2_type      = ARG_CONST_MAP_PTR,
651 	.arg3_type      = ARG_ANYTHING,
652 	.arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
653 	.arg5_type      = ARG_CONST_SIZE_OR_ZERO,
654 };
655 
656 BPF_CALL_3(bpf_copy_from_user, void *, dst, u32, size,
657 	   const void __user *, user_ptr)
658 {
659 	int ret = copy_from_user(dst, user_ptr, size);
660 
661 	if (unlikely(ret)) {
662 		memset(dst, 0, size);
663 		ret = -EFAULT;
664 	}
665 
666 	return ret;
667 }
668 
669 const struct bpf_func_proto bpf_copy_from_user_proto = {
670 	.func		= bpf_copy_from_user,
671 	.gpl_only	= false,
672 	.might_sleep	= true,
673 	.ret_type	= RET_INTEGER,
674 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
675 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
676 	.arg3_type	= ARG_ANYTHING,
677 };
678 
679 BPF_CALL_5(bpf_copy_from_user_task, void *, dst, u32, size,
680 	   const void __user *, user_ptr, struct task_struct *, tsk, u64, flags)
681 {
682 	int ret;
683 
684 	/* flags is not used yet */
685 	if (unlikely(flags))
686 		return -EINVAL;
687 
688 	if (unlikely(!size))
689 		return 0;
690 
691 	ret = access_process_vm(tsk, (unsigned long)user_ptr, dst, size, 0);
692 	if (ret == size)
693 		return 0;
694 
695 	memset(dst, 0, size);
696 	/* Return -EFAULT for partial read */
697 	return ret < 0 ? ret : -EFAULT;
698 }
699 
700 const struct bpf_func_proto bpf_copy_from_user_task_proto = {
701 	.func		= bpf_copy_from_user_task,
702 	.gpl_only	= true,
703 	.might_sleep	= true,
704 	.ret_type	= RET_INTEGER,
705 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
706 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
707 	.arg3_type	= ARG_ANYTHING,
708 	.arg4_type	= ARG_PTR_TO_BTF_ID,
709 	.arg4_btf_id	= &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
710 	.arg5_type	= ARG_ANYTHING
711 };
712 
713 BPF_CALL_2(bpf_per_cpu_ptr, const void *, ptr, u32, cpu)
714 {
715 	if (cpu >= nr_cpu_ids)
716 		return (unsigned long)NULL;
717 
718 	return (unsigned long)per_cpu_ptr((const void __percpu *)(const uintptr_t)ptr, cpu);
719 }
720 
721 const struct bpf_func_proto bpf_per_cpu_ptr_proto = {
722 	.func		= bpf_per_cpu_ptr,
723 	.gpl_only	= false,
724 	.ret_type	= RET_PTR_TO_MEM_OR_BTF_ID | PTR_MAYBE_NULL | MEM_RDONLY,
725 	.arg1_type	= ARG_PTR_TO_PERCPU_BTF_ID,
726 	.arg2_type	= ARG_ANYTHING,
727 };
728 
729 BPF_CALL_1(bpf_this_cpu_ptr, const void *, percpu_ptr)
730 {
731 	return (unsigned long)this_cpu_ptr((const void __percpu *)(const uintptr_t)percpu_ptr);
732 }
733 
734 const struct bpf_func_proto bpf_this_cpu_ptr_proto = {
735 	.func		= bpf_this_cpu_ptr,
736 	.gpl_only	= false,
737 	.ret_type	= RET_PTR_TO_MEM_OR_BTF_ID | MEM_RDONLY,
738 	.arg1_type	= ARG_PTR_TO_PERCPU_BTF_ID,
739 };
740 
741 static int bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
742 		size_t bufsz)
743 {
744 	void __user *user_ptr = (__force void __user *)unsafe_ptr;
745 
746 	buf[0] = 0;
747 
748 	switch (fmt_ptype) {
749 	case 's':
750 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
751 		if ((unsigned long)unsafe_ptr < TASK_SIZE)
752 			return strncpy_from_user_nofault(buf, user_ptr, bufsz);
753 		fallthrough;
754 #endif
755 	case 'k':
756 		return strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz);
757 	case 'u':
758 		return strncpy_from_user_nofault(buf, user_ptr, bufsz);
759 	}
760 
761 	return -EINVAL;
762 }
763 
764 /* Per-cpu temp buffers used by printf-like helpers to store the bprintf binary
765  * arguments representation.
766  */
767 #define MAX_BPRINTF_BIN_ARGS	512
768 
769 /* Support executing three nested bprintf helper calls on a given CPU */
770 #define MAX_BPRINTF_NEST_LEVEL	3
771 struct bpf_bprintf_buffers {
772 	char bin_args[MAX_BPRINTF_BIN_ARGS];
773 	char buf[MAX_BPRINTF_BUF];
774 };
775 
776 static DEFINE_PER_CPU(struct bpf_bprintf_buffers[MAX_BPRINTF_NEST_LEVEL], bpf_bprintf_bufs);
777 static DEFINE_PER_CPU(int, bpf_bprintf_nest_level);
778 
779 static int try_get_buffers(struct bpf_bprintf_buffers **bufs)
780 {
781 	int nest_level;
782 
783 	preempt_disable();
784 	nest_level = this_cpu_inc_return(bpf_bprintf_nest_level);
785 	if (WARN_ON_ONCE(nest_level > MAX_BPRINTF_NEST_LEVEL)) {
786 		this_cpu_dec(bpf_bprintf_nest_level);
787 		preempt_enable();
788 		return -EBUSY;
789 	}
790 	*bufs = this_cpu_ptr(&bpf_bprintf_bufs[nest_level - 1]);
791 
792 	return 0;
793 }
794 
795 void bpf_bprintf_cleanup(struct bpf_bprintf_data *data)
796 {
797 	if (!data->bin_args && !data->buf)
798 		return;
799 	if (WARN_ON_ONCE(this_cpu_read(bpf_bprintf_nest_level) == 0))
800 		return;
801 	this_cpu_dec(bpf_bprintf_nest_level);
802 	preempt_enable();
803 }
804 
805 /*
806  * bpf_bprintf_prepare - Generic pass on format strings for bprintf-like helpers
807  *
808  * Returns a negative value if fmt is an invalid format string or 0 otherwise.
809  *
810  * This can be used in two ways:
811  * - Format string verification only: when data->get_bin_args is false
812  * - Arguments preparation: in addition to the above verification, it writes in
813  *   data->bin_args a binary representation of arguments usable by bstr_printf
814  *   where pointers from BPF have been sanitized.
815  *
816  * In argument preparation mode, if 0 is returned, safe temporary buffers are
817  * allocated and bpf_bprintf_cleanup should be called to free them after use.
818  */
819 int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args,
820 			u32 num_args, struct bpf_bprintf_data *data)
821 {
822 	bool get_buffers = (data->get_bin_args && num_args) || data->get_buf;
823 	char *unsafe_ptr = NULL, *tmp_buf = NULL, *tmp_buf_end, *fmt_end;
824 	struct bpf_bprintf_buffers *buffers = NULL;
825 	size_t sizeof_cur_arg, sizeof_cur_ip;
826 	int err, i, num_spec = 0;
827 	u64 cur_arg;
828 	char fmt_ptype, cur_ip[16], ip_spec[] = "%pXX";
829 
830 	fmt_end = strnchr(fmt, fmt_size, 0);
831 	if (!fmt_end)
832 		return -EINVAL;
833 	fmt_size = fmt_end - fmt;
834 
835 	if (get_buffers && try_get_buffers(&buffers))
836 		return -EBUSY;
837 
838 	if (data->get_bin_args) {
839 		if (num_args)
840 			tmp_buf = buffers->bin_args;
841 		tmp_buf_end = tmp_buf + MAX_BPRINTF_BIN_ARGS;
842 		data->bin_args = (u32 *)tmp_buf;
843 	}
844 
845 	if (data->get_buf)
846 		data->buf = buffers->buf;
847 
848 	for (i = 0; i < fmt_size; i++) {
849 		if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
850 			err = -EINVAL;
851 			goto out;
852 		}
853 
854 		if (fmt[i] != '%')
855 			continue;
856 
857 		if (fmt[i + 1] == '%') {
858 			i++;
859 			continue;
860 		}
861 
862 		if (num_spec >= num_args) {
863 			err = -EINVAL;
864 			goto out;
865 		}
866 
867 		/* The string is zero-terminated so if fmt[i] != 0, we can
868 		 * always access fmt[i + 1], in the worst case it will be a 0
869 		 */
870 		i++;
871 
872 		/* skip optional "[0 +-][num]" width formatting field */
873 		while (fmt[i] == '0' || fmt[i] == '+'  || fmt[i] == '-' ||
874 		       fmt[i] == ' ')
875 			i++;
876 		if (fmt[i] >= '1' && fmt[i] <= '9') {
877 			i++;
878 			while (fmt[i] >= '0' && fmt[i] <= '9')
879 				i++;
880 		}
881 
882 		if (fmt[i] == 'p') {
883 			sizeof_cur_arg = sizeof(long);
884 
885 			if ((fmt[i + 1] == 'k' || fmt[i + 1] == 'u') &&
886 			    fmt[i + 2] == 's') {
887 				fmt_ptype = fmt[i + 1];
888 				i += 2;
889 				goto fmt_str;
890 			}
891 
892 			if (fmt[i + 1] == 0 || isspace(fmt[i + 1]) ||
893 			    ispunct(fmt[i + 1]) || fmt[i + 1] == 'K' ||
894 			    fmt[i + 1] == 'x' || fmt[i + 1] == 's' ||
895 			    fmt[i + 1] == 'S') {
896 				/* just kernel pointers */
897 				if (tmp_buf)
898 					cur_arg = raw_args[num_spec];
899 				i++;
900 				goto nocopy_fmt;
901 			}
902 
903 			if (fmt[i + 1] == 'B') {
904 				if (tmp_buf)  {
905 					err = snprintf(tmp_buf,
906 						       (tmp_buf_end - tmp_buf),
907 						       "%pB",
908 						       (void *)(long)raw_args[num_spec]);
909 					tmp_buf += (err + 1);
910 				}
911 
912 				i++;
913 				num_spec++;
914 				continue;
915 			}
916 
917 			/* only support "%pI4", "%pi4", "%pI6" and "%pi6". */
918 			if ((fmt[i + 1] != 'i' && fmt[i + 1] != 'I') ||
919 			    (fmt[i + 2] != '4' && fmt[i + 2] != '6')) {
920 				err = -EINVAL;
921 				goto out;
922 			}
923 
924 			i += 2;
925 			if (!tmp_buf)
926 				goto nocopy_fmt;
927 
928 			sizeof_cur_ip = (fmt[i] == '4') ? 4 : 16;
929 			if (tmp_buf_end - tmp_buf < sizeof_cur_ip) {
930 				err = -ENOSPC;
931 				goto out;
932 			}
933 
934 			unsafe_ptr = (char *)(long)raw_args[num_spec];
935 			err = copy_from_kernel_nofault(cur_ip, unsafe_ptr,
936 						       sizeof_cur_ip);
937 			if (err < 0)
938 				memset(cur_ip, 0, sizeof_cur_ip);
939 
940 			/* hack: bstr_printf expects IP addresses to be
941 			 * pre-formatted as strings, ironically, the easiest way
942 			 * to do that is to call snprintf.
943 			 */
944 			ip_spec[2] = fmt[i - 1];
945 			ip_spec[3] = fmt[i];
946 			err = snprintf(tmp_buf, tmp_buf_end - tmp_buf,
947 				       ip_spec, &cur_ip);
948 
949 			tmp_buf += err + 1;
950 			num_spec++;
951 
952 			continue;
953 		} else if (fmt[i] == 's') {
954 			fmt_ptype = fmt[i];
955 fmt_str:
956 			if (fmt[i + 1] != 0 &&
957 			    !isspace(fmt[i + 1]) &&
958 			    !ispunct(fmt[i + 1])) {
959 				err = -EINVAL;
960 				goto out;
961 			}
962 
963 			if (!tmp_buf)
964 				goto nocopy_fmt;
965 
966 			if (tmp_buf_end == tmp_buf) {
967 				err = -ENOSPC;
968 				goto out;
969 			}
970 
971 			unsafe_ptr = (char *)(long)raw_args[num_spec];
972 			err = bpf_trace_copy_string(tmp_buf, unsafe_ptr,
973 						    fmt_ptype,
974 						    tmp_buf_end - tmp_buf);
975 			if (err < 0) {
976 				tmp_buf[0] = '\0';
977 				err = 1;
978 			}
979 
980 			tmp_buf += err;
981 			num_spec++;
982 
983 			continue;
984 		} else if (fmt[i] == 'c') {
985 			if (!tmp_buf)
986 				goto nocopy_fmt;
987 
988 			if (tmp_buf_end == tmp_buf) {
989 				err = -ENOSPC;
990 				goto out;
991 			}
992 
993 			*tmp_buf = raw_args[num_spec];
994 			tmp_buf++;
995 			num_spec++;
996 
997 			continue;
998 		}
999 
1000 		sizeof_cur_arg = sizeof(int);
1001 
1002 		if (fmt[i] == 'l') {
1003 			sizeof_cur_arg = sizeof(long);
1004 			i++;
1005 		}
1006 		if (fmt[i] == 'l') {
1007 			sizeof_cur_arg = sizeof(long long);
1008 			i++;
1009 		}
1010 
1011 		if (fmt[i] != 'i' && fmt[i] != 'd' && fmt[i] != 'u' &&
1012 		    fmt[i] != 'x' && fmt[i] != 'X') {
1013 			err = -EINVAL;
1014 			goto out;
1015 		}
1016 
1017 		if (tmp_buf)
1018 			cur_arg = raw_args[num_spec];
1019 nocopy_fmt:
1020 		if (tmp_buf) {
1021 			tmp_buf = PTR_ALIGN(tmp_buf, sizeof(u32));
1022 			if (tmp_buf_end - tmp_buf < sizeof_cur_arg) {
1023 				err = -ENOSPC;
1024 				goto out;
1025 			}
1026 
1027 			if (sizeof_cur_arg == 8) {
1028 				*(u32 *)tmp_buf = *(u32 *)&cur_arg;
1029 				*(u32 *)(tmp_buf + 4) = *((u32 *)&cur_arg + 1);
1030 			} else {
1031 				*(u32 *)tmp_buf = (u32)(long)cur_arg;
1032 			}
1033 			tmp_buf += sizeof_cur_arg;
1034 		}
1035 		num_spec++;
1036 	}
1037 
1038 	err = 0;
1039 out:
1040 	if (err)
1041 		bpf_bprintf_cleanup(data);
1042 	return err;
1043 }
1044 
1045 BPF_CALL_5(bpf_snprintf, char *, str, u32, str_size, char *, fmt,
1046 	   const void *, args, u32, data_len)
1047 {
1048 	struct bpf_bprintf_data data = {
1049 		.get_bin_args	= true,
1050 	};
1051 	int err, num_args;
1052 
1053 	if (data_len % 8 || data_len > MAX_BPRINTF_VARARGS * 8 ||
1054 	    (data_len && !args))
1055 		return -EINVAL;
1056 	num_args = data_len / 8;
1057 
1058 	/* ARG_PTR_TO_CONST_STR guarantees that fmt is zero-terminated so we
1059 	 * can safely give an unbounded size.
1060 	 */
1061 	err = bpf_bprintf_prepare(fmt, UINT_MAX, args, num_args, &data);
1062 	if (err < 0)
1063 		return err;
1064 
1065 	err = bstr_printf(str, str_size, fmt, data.bin_args);
1066 
1067 	bpf_bprintf_cleanup(&data);
1068 
1069 	return err + 1;
1070 }
1071 
1072 const struct bpf_func_proto bpf_snprintf_proto = {
1073 	.func		= bpf_snprintf,
1074 	.gpl_only	= true,
1075 	.ret_type	= RET_INTEGER,
1076 	.arg1_type	= ARG_PTR_TO_MEM_OR_NULL,
1077 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
1078 	.arg3_type	= ARG_PTR_TO_CONST_STR,
1079 	.arg4_type	= ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
1080 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
1081 };
1082 
1083 struct bpf_async_cb {
1084 	struct bpf_map *map;
1085 	struct bpf_prog *prog;
1086 	void __rcu *callback_fn;
1087 	void *value;
1088 	union {
1089 		struct rcu_head rcu;
1090 		struct work_struct delete_work;
1091 	};
1092 	u64 flags;
1093 };
1094 
1095 /* BPF map elements can contain 'struct bpf_timer'.
1096  * Such map owns all of its BPF timers.
1097  * 'struct bpf_timer' is allocated as part of map element allocation
1098  * and it's zero initialized.
1099  * That space is used to keep 'struct bpf_async_kern'.
1100  * bpf_timer_init() allocates 'struct bpf_hrtimer', inits hrtimer, and
1101  * remembers 'struct bpf_map *' pointer it's part of.
1102  * bpf_timer_set_callback() increments prog refcnt and assign bpf callback_fn.
1103  * bpf_timer_start() arms the timer.
1104  * If user space reference to a map goes to zero at this point
1105  * ops->map_release_uref callback is responsible for cancelling the timers,
1106  * freeing their memory, and decrementing prog's refcnts.
1107  * bpf_timer_cancel() cancels the timer and decrements prog's refcnt.
1108  * Inner maps can contain bpf timers as well. ops->map_release_uref is
1109  * freeing the timers when inner map is replaced or deleted by user space.
1110  */
1111 struct bpf_hrtimer {
1112 	struct bpf_async_cb cb;
1113 	struct hrtimer timer;
1114 	atomic_t cancelling;
1115 };
1116 
1117 struct bpf_work {
1118 	struct bpf_async_cb cb;
1119 	struct work_struct work;
1120 	struct work_struct delete_work;
1121 };
1122 
1123 /* the actual struct hidden inside uapi struct bpf_timer and bpf_wq */
1124 struct bpf_async_kern {
1125 	union {
1126 		struct bpf_async_cb *cb;
1127 		struct bpf_hrtimer *timer;
1128 		struct bpf_work *work;
1129 	};
1130 	/* bpf_spin_lock is used here instead of spinlock_t to make
1131 	 * sure that it always fits into space reserved by struct bpf_timer
1132 	 * regardless of LOCKDEP and spinlock debug flags.
1133 	 */
1134 	struct bpf_spin_lock lock;
1135 } __attribute__((aligned(8)));
1136 
1137 enum bpf_async_type {
1138 	BPF_ASYNC_TYPE_TIMER = 0,
1139 	BPF_ASYNC_TYPE_WQ,
1140 };
1141 
1142 static DEFINE_PER_CPU(struct bpf_hrtimer *, hrtimer_running);
1143 
1144 static enum hrtimer_restart bpf_timer_cb(struct hrtimer *hrtimer)
1145 {
1146 	struct bpf_hrtimer *t = container_of(hrtimer, struct bpf_hrtimer, timer);
1147 	struct bpf_map *map = t->cb.map;
1148 	void *value = t->cb.value;
1149 	bpf_callback_t callback_fn;
1150 	void *key;
1151 	u32 idx;
1152 
1153 	BTF_TYPE_EMIT(struct bpf_timer);
1154 	callback_fn = rcu_dereference_check(t->cb.callback_fn, rcu_read_lock_bh_held());
1155 	if (!callback_fn)
1156 		goto out;
1157 
1158 	/* bpf_timer_cb() runs in hrtimer_run_softirq. It doesn't migrate and
1159 	 * cannot be preempted by another bpf_timer_cb() on the same cpu.
1160 	 * Remember the timer this callback is servicing to prevent
1161 	 * deadlock if callback_fn() calls bpf_timer_cancel() or
1162 	 * bpf_map_delete_elem() on the same timer.
1163 	 */
1164 	this_cpu_write(hrtimer_running, t);
1165 	if (map->map_type == BPF_MAP_TYPE_ARRAY) {
1166 		struct bpf_array *array = container_of(map, struct bpf_array, map);
1167 
1168 		/* compute the key */
1169 		idx = ((char *)value - array->value) / array->elem_size;
1170 		key = &idx;
1171 	} else { /* hash or lru */
1172 		key = value - round_up(map->key_size, 8);
1173 	}
1174 
1175 	callback_fn((u64)(long)map, (u64)(long)key, (u64)(long)value, 0, 0);
1176 	/* The verifier checked that return value is zero. */
1177 
1178 	this_cpu_write(hrtimer_running, NULL);
1179 out:
1180 	return HRTIMER_NORESTART;
1181 }
1182 
1183 static void bpf_wq_work(struct work_struct *work)
1184 {
1185 	struct bpf_work *w = container_of(work, struct bpf_work, work);
1186 	struct bpf_async_cb *cb = &w->cb;
1187 	struct bpf_map *map = cb->map;
1188 	bpf_callback_t callback_fn;
1189 	void *value = cb->value;
1190 	void *key;
1191 	u32 idx;
1192 
1193 	BTF_TYPE_EMIT(struct bpf_wq);
1194 
1195 	callback_fn = READ_ONCE(cb->callback_fn);
1196 	if (!callback_fn)
1197 		return;
1198 
1199 	if (map->map_type == BPF_MAP_TYPE_ARRAY) {
1200 		struct bpf_array *array = container_of(map, struct bpf_array, map);
1201 
1202 		/* compute the key */
1203 		idx = ((char *)value - array->value) / array->elem_size;
1204 		key = &idx;
1205 	} else { /* hash or lru */
1206 		key = value - round_up(map->key_size, 8);
1207 	}
1208 
1209         rcu_read_lock_trace();
1210         migrate_disable();
1211 
1212 	callback_fn((u64)(long)map, (u64)(long)key, (u64)(long)value, 0, 0);
1213 
1214 	migrate_enable();
1215 	rcu_read_unlock_trace();
1216 }
1217 
1218 static void bpf_wq_delete_work(struct work_struct *work)
1219 {
1220 	struct bpf_work *w = container_of(work, struct bpf_work, delete_work);
1221 
1222 	cancel_work_sync(&w->work);
1223 
1224 	kfree_rcu(w, cb.rcu);
1225 }
1226 
1227 static void bpf_timer_delete_work(struct work_struct *work)
1228 {
1229 	struct bpf_hrtimer *t = container_of(work, struct bpf_hrtimer, cb.delete_work);
1230 
1231 	/* Cancel the timer and wait for callback to complete if it was running.
1232 	 * If hrtimer_cancel() can be safely called it's safe to call
1233 	 * kfree_rcu(t) right after for both preallocated and non-preallocated
1234 	 * maps.  The async->cb = NULL was already done and no code path can see
1235 	 * address 't' anymore. Timer if armed for existing bpf_hrtimer before
1236 	 * bpf_timer_cancel_and_free will have been cancelled.
1237 	 */
1238 	hrtimer_cancel(&t->timer);
1239 	kfree_rcu(t, cb.rcu);
1240 }
1241 
1242 static int __bpf_async_init(struct bpf_async_kern *async, struct bpf_map *map, u64 flags,
1243 			    enum bpf_async_type type)
1244 {
1245 	struct bpf_async_cb *cb;
1246 	struct bpf_hrtimer *t;
1247 	struct bpf_work *w;
1248 	clockid_t clockid;
1249 	size_t size;
1250 	int ret = 0;
1251 
1252 	if (in_nmi())
1253 		return -EOPNOTSUPP;
1254 
1255 	switch (type) {
1256 	case BPF_ASYNC_TYPE_TIMER:
1257 		size = sizeof(struct bpf_hrtimer);
1258 		break;
1259 	case BPF_ASYNC_TYPE_WQ:
1260 		size = sizeof(struct bpf_work);
1261 		break;
1262 	default:
1263 		return -EINVAL;
1264 	}
1265 
1266 	__bpf_spin_lock_irqsave(&async->lock);
1267 	t = async->timer;
1268 	if (t) {
1269 		ret = -EBUSY;
1270 		goto out;
1271 	}
1272 
1273 	/* allocate hrtimer via map_kmalloc to use memcg accounting */
1274 	cb = bpf_map_kmalloc_node(map, size, GFP_ATOMIC, map->numa_node);
1275 	if (!cb) {
1276 		ret = -ENOMEM;
1277 		goto out;
1278 	}
1279 
1280 	switch (type) {
1281 	case BPF_ASYNC_TYPE_TIMER:
1282 		clockid = flags & (MAX_CLOCKS - 1);
1283 		t = (struct bpf_hrtimer *)cb;
1284 
1285 		atomic_set(&t->cancelling, 0);
1286 		INIT_WORK(&t->cb.delete_work, bpf_timer_delete_work);
1287 		hrtimer_init(&t->timer, clockid, HRTIMER_MODE_REL_SOFT);
1288 		t->timer.function = bpf_timer_cb;
1289 		cb->value = (void *)async - map->record->timer_off;
1290 		break;
1291 	case BPF_ASYNC_TYPE_WQ:
1292 		w = (struct bpf_work *)cb;
1293 
1294 		INIT_WORK(&w->work, bpf_wq_work);
1295 		INIT_WORK(&w->delete_work, bpf_wq_delete_work);
1296 		cb->value = (void *)async - map->record->wq_off;
1297 		break;
1298 	}
1299 	cb->map = map;
1300 	cb->prog = NULL;
1301 	cb->flags = flags;
1302 	rcu_assign_pointer(cb->callback_fn, NULL);
1303 
1304 	WRITE_ONCE(async->cb, cb);
1305 	/* Guarantee the order between async->cb and map->usercnt. So
1306 	 * when there are concurrent uref release and bpf timer init, either
1307 	 * bpf_timer_cancel_and_free() called by uref release reads a no-NULL
1308 	 * timer or atomic64_read() below returns a zero usercnt.
1309 	 */
1310 	smp_mb();
1311 	if (!atomic64_read(&map->usercnt)) {
1312 		/* maps with timers must be either held by user space
1313 		 * or pinned in bpffs.
1314 		 */
1315 		WRITE_ONCE(async->cb, NULL);
1316 		kfree(cb);
1317 		ret = -EPERM;
1318 	}
1319 out:
1320 	__bpf_spin_unlock_irqrestore(&async->lock);
1321 	return ret;
1322 }
1323 
1324 BPF_CALL_3(bpf_timer_init, struct bpf_async_kern *, timer, struct bpf_map *, map,
1325 	   u64, flags)
1326 {
1327 	clock_t clockid = flags & (MAX_CLOCKS - 1);
1328 
1329 	BUILD_BUG_ON(MAX_CLOCKS != 16);
1330 	BUILD_BUG_ON(sizeof(struct bpf_async_kern) > sizeof(struct bpf_timer));
1331 	BUILD_BUG_ON(__alignof__(struct bpf_async_kern) != __alignof__(struct bpf_timer));
1332 
1333 	if (flags >= MAX_CLOCKS ||
1334 	    /* similar to timerfd except _ALARM variants are not supported */
1335 	    (clockid != CLOCK_MONOTONIC &&
1336 	     clockid != CLOCK_REALTIME &&
1337 	     clockid != CLOCK_BOOTTIME))
1338 		return -EINVAL;
1339 
1340 	return __bpf_async_init(timer, map, flags, BPF_ASYNC_TYPE_TIMER);
1341 }
1342 
1343 static const struct bpf_func_proto bpf_timer_init_proto = {
1344 	.func		= bpf_timer_init,
1345 	.gpl_only	= true,
1346 	.ret_type	= RET_INTEGER,
1347 	.arg1_type	= ARG_PTR_TO_TIMER,
1348 	.arg2_type	= ARG_CONST_MAP_PTR,
1349 	.arg3_type	= ARG_ANYTHING,
1350 };
1351 
1352 static int __bpf_async_set_callback(struct bpf_async_kern *async, void *callback_fn,
1353 				    struct bpf_prog_aux *aux, unsigned int flags,
1354 				    enum bpf_async_type type)
1355 {
1356 	struct bpf_prog *prev, *prog = aux->prog;
1357 	struct bpf_async_cb *cb;
1358 	int ret = 0;
1359 
1360 	if (in_nmi())
1361 		return -EOPNOTSUPP;
1362 	__bpf_spin_lock_irqsave(&async->lock);
1363 	cb = async->cb;
1364 	if (!cb) {
1365 		ret = -EINVAL;
1366 		goto out;
1367 	}
1368 	if (!atomic64_read(&cb->map->usercnt)) {
1369 		/* maps with timers must be either held by user space
1370 		 * or pinned in bpffs. Otherwise timer might still be
1371 		 * running even when bpf prog is detached and user space
1372 		 * is gone, since map_release_uref won't ever be called.
1373 		 */
1374 		ret = -EPERM;
1375 		goto out;
1376 	}
1377 	prev = cb->prog;
1378 	if (prev != prog) {
1379 		/* Bump prog refcnt once. Every bpf_timer_set_callback()
1380 		 * can pick different callback_fn-s within the same prog.
1381 		 */
1382 		prog = bpf_prog_inc_not_zero(prog);
1383 		if (IS_ERR(prog)) {
1384 			ret = PTR_ERR(prog);
1385 			goto out;
1386 		}
1387 		if (prev)
1388 			/* Drop prev prog refcnt when swapping with new prog */
1389 			bpf_prog_put(prev);
1390 		cb->prog = prog;
1391 	}
1392 	rcu_assign_pointer(cb->callback_fn, callback_fn);
1393 out:
1394 	__bpf_spin_unlock_irqrestore(&async->lock);
1395 	return ret;
1396 }
1397 
1398 BPF_CALL_3(bpf_timer_set_callback, struct bpf_async_kern *, timer, void *, callback_fn,
1399 	   struct bpf_prog_aux *, aux)
1400 {
1401 	return __bpf_async_set_callback(timer, callback_fn, aux, 0, BPF_ASYNC_TYPE_TIMER);
1402 }
1403 
1404 static const struct bpf_func_proto bpf_timer_set_callback_proto = {
1405 	.func		= bpf_timer_set_callback,
1406 	.gpl_only	= true,
1407 	.ret_type	= RET_INTEGER,
1408 	.arg1_type	= ARG_PTR_TO_TIMER,
1409 	.arg2_type	= ARG_PTR_TO_FUNC,
1410 };
1411 
1412 BPF_CALL_3(bpf_timer_start, struct bpf_async_kern *, timer, u64, nsecs, u64, flags)
1413 {
1414 	struct bpf_hrtimer *t;
1415 	int ret = 0;
1416 	enum hrtimer_mode mode;
1417 
1418 	if (in_nmi())
1419 		return -EOPNOTSUPP;
1420 	if (flags & ~(BPF_F_TIMER_ABS | BPF_F_TIMER_CPU_PIN))
1421 		return -EINVAL;
1422 	__bpf_spin_lock_irqsave(&timer->lock);
1423 	t = timer->timer;
1424 	if (!t || !t->cb.prog) {
1425 		ret = -EINVAL;
1426 		goto out;
1427 	}
1428 
1429 	if (flags & BPF_F_TIMER_ABS)
1430 		mode = HRTIMER_MODE_ABS_SOFT;
1431 	else
1432 		mode = HRTIMER_MODE_REL_SOFT;
1433 
1434 	if (flags & BPF_F_TIMER_CPU_PIN)
1435 		mode |= HRTIMER_MODE_PINNED;
1436 
1437 	hrtimer_start(&t->timer, ns_to_ktime(nsecs), mode);
1438 out:
1439 	__bpf_spin_unlock_irqrestore(&timer->lock);
1440 	return ret;
1441 }
1442 
1443 static const struct bpf_func_proto bpf_timer_start_proto = {
1444 	.func		= bpf_timer_start,
1445 	.gpl_only	= true,
1446 	.ret_type	= RET_INTEGER,
1447 	.arg1_type	= ARG_PTR_TO_TIMER,
1448 	.arg2_type	= ARG_ANYTHING,
1449 	.arg3_type	= ARG_ANYTHING,
1450 };
1451 
1452 static void drop_prog_refcnt(struct bpf_async_cb *async)
1453 {
1454 	struct bpf_prog *prog = async->prog;
1455 
1456 	if (prog) {
1457 		bpf_prog_put(prog);
1458 		async->prog = NULL;
1459 		rcu_assign_pointer(async->callback_fn, NULL);
1460 	}
1461 }
1462 
1463 BPF_CALL_1(bpf_timer_cancel, struct bpf_async_kern *, timer)
1464 {
1465 	struct bpf_hrtimer *t, *cur_t;
1466 	bool inc = false;
1467 	int ret = 0;
1468 
1469 	if (in_nmi())
1470 		return -EOPNOTSUPP;
1471 	rcu_read_lock();
1472 	__bpf_spin_lock_irqsave(&timer->lock);
1473 	t = timer->timer;
1474 	if (!t) {
1475 		ret = -EINVAL;
1476 		goto out;
1477 	}
1478 
1479 	cur_t = this_cpu_read(hrtimer_running);
1480 	if (cur_t == t) {
1481 		/* If bpf callback_fn is trying to bpf_timer_cancel()
1482 		 * its own timer the hrtimer_cancel() will deadlock
1483 		 * since it waits for callback_fn to finish.
1484 		 */
1485 		ret = -EDEADLK;
1486 		goto out;
1487 	}
1488 
1489 	/* Only account in-flight cancellations when invoked from a timer
1490 	 * callback, since we want to avoid waiting only if other _callbacks_
1491 	 * are waiting on us, to avoid introducing lockups. Non-callback paths
1492 	 * are ok, since nobody would synchronously wait for their completion.
1493 	 */
1494 	if (!cur_t)
1495 		goto drop;
1496 	atomic_inc(&t->cancelling);
1497 	/* Need full barrier after relaxed atomic_inc */
1498 	smp_mb__after_atomic();
1499 	inc = true;
1500 	if (atomic_read(&cur_t->cancelling)) {
1501 		/* We're cancelling timer t, while some other timer callback is
1502 		 * attempting to cancel us. In such a case, it might be possible
1503 		 * that timer t belongs to the other callback, or some other
1504 		 * callback waiting upon it (creating transitive dependencies
1505 		 * upon us), and we will enter a deadlock if we continue
1506 		 * cancelling and waiting for it synchronously, since it might
1507 		 * do the same. Bail!
1508 		 */
1509 		ret = -EDEADLK;
1510 		goto out;
1511 	}
1512 drop:
1513 	drop_prog_refcnt(&t->cb);
1514 out:
1515 	__bpf_spin_unlock_irqrestore(&timer->lock);
1516 	/* Cancel the timer and wait for associated callback to finish
1517 	 * if it was running.
1518 	 */
1519 	ret = ret ?: hrtimer_cancel(&t->timer);
1520 	if (inc)
1521 		atomic_dec(&t->cancelling);
1522 	rcu_read_unlock();
1523 	return ret;
1524 }
1525 
1526 static const struct bpf_func_proto bpf_timer_cancel_proto = {
1527 	.func		= bpf_timer_cancel,
1528 	.gpl_only	= true,
1529 	.ret_type	= RET_INTEGER,
1530 	.arg1_type	= ARG_PTR_TO_TIMER,
1531 };
1532 
1533 static struct bpf_async_cb *__bpf_async_cancel_and_free(struct bpf_async_kern *async)
1534 {
1535 	struct bpf_async_cb *cb;
1536 
1537 	/* Performance optimization: read async->cb without lock first. */
1538 	if (!READ_ONCE(async->cb))
1539 		return NULL;
1540 
1541 	__bpf_spin_lock_irqsave(&async->lock);
1542 	/* re-read it under lock */
1543 	cb = async->cb;
1544 	if (!cb)
1545 		goto out;
1546 	drop_prog_refcnt(cb);
1547 	/* The subsequent bpf_timer_start/cancel() helpers won't be able to use
1548 	 * this timer, since it won't be initialized.
1549 	 */
1550 	WRITE_ONCE(async->cb, NULL);
1551 out:
1552 	__bpf_spin_unlock_irqrestore(&async->lock);
1553 	return cb;
1554 }
1555 
1556 /* This function is called by map_delete/update_elem for individual element and
1557  * by ops->map_release_uref when the user space reference to a map reaches zero.
1558  */
1559 void bpf_timer_cancel_and_free(void *val)
1560 {
1561 	struct bpf_hrtimer *t;
1562 
1563 	t = (struct bpf_hrtimer *)__bpf_async_cancel_and_free(val);
1564 
1565 	if (!t)
1566 		return;
1567 	/* We check that bpf_map_delete/update_elem() was called from timer
1568 	 * callback_fn. In such case we don't call hrtimer_cancel() (since it
1569 	 * will deadlock) and don't call hrtimer_try_to_cancel() (since it will
1570 	 * just return -1). Though callback_fn is still running on this cpu it's
1571 	 * safe to do kfree(t) because bpf_timer_cb() read everything it needed
1572 	 * from 't'. The bpf subprog callback_fn won't be able to access 't',
1573 	 * since async->cb = NULL was already done. The timer will be
1574 	 * effectively cancelled because bpf_timer_cb() will return
1575 	 * HRTIMER_NORESTART.
1576 	 *
1577 	 * However, it is possible the timer callback_fn calling us armed the
1578 	 * timer _before_ calling us, such that failing to cancel it here will
1579 	 * cause it to possibly use struct hrtimer after freeing bpf_hrtimer.
1580 	 * Therefore, we _need_ to cancel any outstanding timers before we do
1581 	 * kfree_rcu, even though no more timers can be armed.
1582 	 *
1583 	 * Moreover, we need to schedule work even if timer does not belong to
1584 	 * the calling callback_fn, as on two different CPUs, we can end up in a
1585 	 * situation where both sides run in parallel, try to cancel one
1586 	 * another, and we end up waiting on both sides in hrtimer_cancel
1587 	 * without making forward progress, since timer1 depends on time2
1588 	 * callback to finish, and vice versa.
1589 	 *
1590 	 *  CPU 1 (timer1_cb)			CPU 2 (timer2_cb)
1591 	 *  bpf_timer_cancel_and_free(timer2)	bpf_timer_cancel_and_free(timer1)
1592 	 *
1593 	 * To avoid these issues, punt to workqueue context when we are in a
1594 	 * timer callback.
1595 	 */
1596 	if (this_cpu_read(hrtimer_running))
1597 		queue_work(system_unbound_wq, &t->cb.delete_work);
1598 	else
1599 		bpf_timer_delete_work(&t->cb.delete_work);
1600 }
1601 
1602 /* This function is called by map_delete/update_elem for individual element and
1603  * by ops->map_release_uref when the user space reference to a map reaches zero.
1604  */
1605 void bpf_wq_cancel_and_free(void *val)
1606 {
1607 	struct bpf_work *work;
1608 
1609 	BTF_TYPE_EMIT(struct bpf_wq);
1610 
1611 	work = (struct bpf_work *)__bpf_async_cancel_and_free(val);
1612 	if (!work)
1613 		return;
1614 	/* Trigger cancel of the sleepable work, but *do not* wait for
1615 	 * it to finish if it was running as we might not be in a
1616 	 * sleepable context.
1617 	 * kfree will be called once the work has finished.
1618 	 */
1619 	schedule_work(&work->delete_work);
1620 }
1621 
1622 BPF_CALL_2(bpf_kptr_xchg, void *, dst, void *, ptr)
1623 {
1624 	unsigned long *kptr = dst;
1625 
1626 	/* This helper may be inlined by verifier. */
1627 	return xchg(kptr, (unsigned long)ptr);
1628 }
1629 
1630 /* Unlike other PTR_TO_BTF_ID helpers the btf_id in bpf_kptr_xchg()
1631  * helper is determined dynamically by the verifier. Use BPF_PTR_POISON to
1632  * denote type that verifier will determine.
1633  */
1634 static const struct bpf_func_proto bpf_kptr_xchg_proto = {
1635 	.func         = bpf_kptr_xchg,
1636 	.gpl_only     = false,
1637 	.ret_type     = RET_PTR_TO_BTF_ID_OR_NULL,
1638 	.ret_btf_id   = BPF_PTR_POISON,
1639 	.arg1_type    = ARG_KPTR_XCHG_DEST,
1640 	.arg2_type    = ARG_PTR_TO_BTF_ID_OR_NULL | OBJ_RELEASE,
1641 	.arg2_btf_id  = BPF_PTR_POISON,
1642 };
1643 
1644 /* Since the upper 8 bits of dynptr->size is reserved, the
1645  * maximum supported size is 2^24 - 1.
1646  */
1647 #define DYNPTR_MAX_SIZE	((1UL << 24) - 1)
1648 #define DYNPTR_TYPE_SHIFT	28
1649 #define DYNPTR_SIZE_MASK	0xFFFFFF
1650 #define DYNPTR_RDONLY_BIT	BIT(31)
1651 
1652 bool __bpf_dynptr_is_rdonly(const struct bpf_dynptr_kern *ptr)
1653 {
1654 	return ptr->size & DYNPTR_RDONLY_BIT;
1655 }
1656 
1657 void bpf_dynptr_set_rdonly(struct bpf_dynptr_kern *ptr)
1658 {
1659 	ptr->size |= DYNPTR_RDONLY_BIT;
1660 }
1661 
1662 static void bpf_dynptr_set_type(struct bpf_dynptr_kern *ptr, enum bpf_dynptr_type type)
1663 {
1664 	ptr->size |= type << DYNPTR_TYPE_SHIFT;
1665 }
1666 
1667 static enum bpf_dynptr_type bpf_dynptr_get_type(const struct bpf_dynptr_kern *ptr)
1668 {
1669 	return (ptr->size & ~(DYNPTR_RDONLY_BIT)) >> DYNPTR_TYPE_SHIFT;
1670 }
1671 
1672 u32 __bpf_dynptr_size(const struct bpf_dynptr_kern *ptr)
1673 {
1674 	return ptr->size & DYNPTR_SIZE_MASK;
1675 }
1676 
1677 static void bpf_dynptr_set_size(struct bpf_dynptr_kern *ptr, u32 new_size)
1678 {
1679 	u32 metadata = ptr->size & ~DYNPTR_SIZE_MASK;
1680 
1681 	ptr->size = new_size | metadata;
1682 }
1683 
1684 int bpf_dynptr_check_size(u32 size)
1685 {
1686 	return size > DYNPTR_MAX_SIZE ? -E2BIG : 0;
1687 }
1688 
1689 void bpf_dynptr_init(struct bpf_dynptr_kern *ptr, void *data,
1690 		     enum bpf_dynptr_type type, u32 offset, u32 size)
1691 {
1692 	ptr->data = data;
1693 	ptr->offset = offset;
1694 	ptr->size = size;
1695 	bpf_dynptr_set_type(ptr, type);
1696 }
1697 
1698 void bpf_dynptr_set_null(struct bpf_dynptr_kern *ptr)
1699 {
1700 	memset(ptr, 0, sizeof(*ptr));
1701 }
1702 
1703 static int bpf_dynptr_check_off_len(const struct bpf_dynptr_kern *ptr, u32 offset, u32 len)
1704 {
1705 	u32 size = __bpf_dynptr_size(ptr);
1706 
1707 	if (len > size || offset > size - len)
1708 		return -E2BIG;
1709 
1710 	return 0;
1711 }
1712 
1713 BPF_CALL_4(bpf_dynptr_from_mem, void *, data, u32, size, u64, flags, struct bpf_dynptr_kern *, ptr)
1714 {
1715 	int err;
1716 
1717 	BTF_TYPE_EMIT(struct bpf_dynptr);
1718 
1719 	err = bpf_dynptr_check_size(size);
1720 	if (err)
1721 		goto error;
1722 
1723 	/* flags is currently unsupported */
1724 	if (flags) {
1725 		err = -EINVAL;
1726 		goto error;
1727 	}
1728 
1729 	bpf_dynptr_init(ptr, data, BPF_DYNPTR_TYPE_LOCAL, 0, size);
1730 
1731 	return 0;
1732 
1733 error:
1734 	bpf_dynptr_set_null(ptr);
1735 	return err;
1736 }
1737 
1738 static const struct bpf_func_proto bpf_dynptr_from_mem_proto = {
1739 	.func		= bpf_dynptr_from_mem,
1740 	.gpl_only	= false,
1741 	.ret_type	= RET_INTEGER,
1742 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
1743 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
1744 	.arg3_type	= ARG_ANYTHING,
1745 	.arg4_type	= ARG_PTR_TO_DYNPTR | DYNPTR_TYPE_LOCAL | MEM_UNINIT | MEM_WRITE,
1746 };
1747 
1748 BPF_CALL_5(bpf_dynptr_read, void *, dst, u32, len, const struct bpf_dynptr_kern *, src,
1749 	   u32, offset, u64, flags)
1750 {
1751 	enum bpf_dynptr_type type;
1752 	int err;
1753 
1754 	if (!src->data || flags)
1755 		return -EINVAL;
1756 
1757 	err = bpf_dynptr_check_off_len(src, offset, len);
1758 	if (err)
1759 		return err;
1760 
1761 	type = bpf_dynptr_get_type(src);
1762 
1763 	switch (type) {
1764 	case BPF_DYNPTR_TYPE_LOCAL:
1765 	case BPF_DYNPTR_TYPE_RINGBUF:
1766 		/* Source and destination may possibly overlap, hence use memmove to
1767 		 * copy the data. E.g. bpf_dynptr_from_mem may create two dynptr
1768 		 * pointing to overlapping PTR_TO_MAP_VALUE regions.
1769 		 */
1770 		memmove(dst, src->data + src->offset + offset, len);
1771 		return 0;
1772 	case BPF_DYNPTR_TYPE_SKB:
1773 		return __bpf_skb_load_bytes(src->data, src->offset + offset, dst, len);
1774 	case BPF_DYNPTR_TYPE_XDP:
1775 		return __bpf_xdp_load_bytes(src->data, src->offset + offset, dst, len);
1776 	default:
1777 		WARN_ONCE(true, "bpf_dynptr_read: unknown dynptr type %d\n", type);
1778 		return -EFAULT;
1779 	}
1780 }
1781 
1782 static const struct bpf_func_proto bpf_dynptr_read_proto = {
1783 	.func		= bpf_dynptr_read,
1784 	.gpl_only	= false,
1785 	.ret_type	= RET_INTEGER,
1786 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
1787 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
1788 	.arg3_type	= ARG_PTR_TO_DYNPTR | MEM_RDONLY,
1789 	.arg4_type	= ARG_ANYTHING,
1790 	.arg5_type	= ARG_ANYTHING,
1791 };
1792 
1793 BPF_CALL_5(bpf_dynptr_write, const struct bpf_dynptr_kern *, dst, u32, offset, void *, src,
1794 	   u32, len, u64, flags)
1795 {
1796 	enum bpf_dynptr_type type;
1797 	int err;
1798 
1799 	if (!dst->data || __bpf_dynptr_is_rdonly(dst))
1800 		return -EINVAL;
1801 
1802 	err = bpf_dynptr_check_off_len(dst, offset, len);
1803 	if (err)
1804 		return err;
1805 
1806 	type = bpf_dynptr_get_type(dst);
1807 
1808 	switch (type) {
1809 	case BPF_DYNPTR_TYPE_LOCAL:
1810 	case BPF_DYNPTR_TYPE_RINGBUF:
1811 		if (flags)
1812 			return -EINVAL;
1813 		/* Source and destination may possibly overlap, hence use memmove to
1814 		 * copy the data. E.g. bpf_dynptr_from_mem may create two dynptr
1815 		 * pointing to overlapping PTR_TO_MAP_VALUE regions.
1816 		 */
1817 		memmove(dst->data + dst->offset + offset, src, len);
1818 		return 0;
1819 	case BPF_DYNPTR_TYPE_SKB:
1820 		return __bpf_skb_store_bytes(dst->data, dst->offset + offset, src, len,
1821 					     flags);
1822 	case BPF_DYNPTR_TYPE_XDP:
1823 		if (flags)
1824 			return -EINVAL;
1825 		return __bpf_xdp_store_bytes(dst->data, dst->offset + offset, src, len);
1826 	default:
1827 		WARN_ONCE(true, "bpf_dynptr_write: unknown dynptr type %d\n", type);
1828 		return -EFAULT;
1829 	}
1830 }
1831 
1832 static const struct bpf_func_proto bpf_dynptr_write_proto = {
1833 	.func		= bpf_dynptr_write,
1834 	.gpl_only	= false,
1835 	.ret_type	= RET_INTEGER,
1836 	.arg1_type	= ARG_PTR_TO_DYNPTR | MEM_RDONLY,
1837 	.arg2_type	= ARG_ANYTHING,
1838 	.arg3_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
1839 	.arg4_type	= ARG_CONST_SIZE_OR_ZERO,
1840 	.arg5_type	= ARG_ANYTHING,
1841 };
1842 
1843 BPF_CALL_3(bpf_dynptr_data, const struct bpf_dynptr_kern *, ptr, u32, offset, u32, len)
1844 {
1845 	enum bpf_dynptr_type type;
1846 	int err;
1847 
1848 	if (!ptr->data)
1849 		return 0;
1850 
1851 	err = bpf_dynptr_check_off_len(ptr, offset, len);
1852 	if (err)
1853 		return 0;
1854 
1855 	if (__bpf_dynptr_is_rdonly(ptr))
1856 		return 0;
1857 
1858 	type = bpf_dynptr_get_type(ptr);
1859 
1860 	switch (type) {
1861 	case BPF_DYNPTR_TYPE_LOCAL:
1862 	case BPF_DYNPTR_TYPE_RINGBUF:
1863 		return (unsigned long)(ptr->data + ptr->offset + offset);
1864 	case BPF_DYNPTR_TYPE_SKB:
1865 	case BPF_DYNPTR_TYPE_XDP:
1866 		/* skb and xdp dynptrs should use bpf_dynptr_slice / bpf_dynptr_slice_rdwr */
1867 		return 0;
1868 	default:
1869 		WARN_ONCE(true, "bpf_dynptr_data: unknown dynptr type %d\n", type);
1870 		return 0;
1871 	}
1872 }
1873 
1874 static const struct bpf_func_proto bpf_dynptr_data_proto = {
1875 	.func		= bpf_dynptr_data,
1876 	.gpl_only	= false,
1877 	.ret_type	= RET_PTR_TO_DYNPTR_MEM_OR_NULL,
1878 	.arg1_type	= ARG_PTR_TO_DYNPTR | MEM_RDONLY,
1879 	.arg2_type	= ARG_ANYTHING,
1880 	.arg3_type	= ARG_CONST_ALLOC_SIZE_OR_ZERO,
1881 };
1882 
1883 const struct bpf_func_proto bpf_get_current_task_proto __weak;
1884 const struct bpf_func_proto bpf_get_current_task_btf_proto __weak;
1885 const struct bpf_func_proto bpf_probe_read_user_proto __weak;
1886 const struct bpf_func_proto bpf_probe_read_user_str_proto __weak;
1887 const struct bpf_func_proto bpf_probe_read_kernel_proto __weak;
1888 const struct bpf_func_proto bpf_probe_read_kernel_str_proto __weak;
1889 const struct bpf_func_proto bpf_task_pt_regs_proto __weak;
1890 
1891 const struct bpf_func_proto *
1892 bpf_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1893 {
1894 	switch (func_id) {
1895 	case BPF_FUNC_map_lookup_elem:
1896 		return &bpf_map_lookup_elem_proto;
1897 	case BPF_FUNC_map_update_elem:
1898 		return &bpf_map_update_elem_proto;
1899 	case BPF_FUNC_map_delete_elem:
1900 		return &bpf_map_delete_elem_proto;
1901 	case BPF_FUNC_map_push_elem:
1902 		return &bpf_map_push_elem_proto;
1903 	case BPF_FUNC_map_pop_elem:
1904 		return &bpf_map_pop_elem_proto;
1905 	case BPF_FUNC_map_peek_elem:
1906 		return &bpf_map_peek_elem_proto;
1907 	case BPF_FUNC_map_lookup_percpu_elem:
1908 		return &bpf_map_lookup_percpu_elem_proto;
1909 	case BPF_FUNC_get_prandom_u32:
1910 		return &bpf_get_prandom_u32_proto;
1911 	case BPF_FUNC_get_smp_processor_id:
1912 		return &bpf_get_raw_smp_processor_id_proto;
1913 	case BPF_FUNC_get_numa_node_id:
1914 		return &bpf_get_numa_node_id_proto;
1915 	case BPF_FUNC_tail_call:
1916 		return &bpf_tail_call_proto;
1917 	case BPF_FUNC_ktime_get_ns:
1918 		return &bpf_ktime_get_ns_proto;
1919 	case BPF_FUNC_ktime_get_boot_ns:
1920 		return &bpf_ktime_get_boot_ns_proto;
1921 	case BPF_FUNC_ktime_get_tai_ns:
1922 		return &bpf_ktime_get_tai_ns_proto;
1923 	case BPF_FUNC_ringbuf_output:
1924 		return &bpf_ringbuf_output_proto;
1925 	case BPF_FUNC_ringbuf_reserve:
1926 		return &bpf_ringbuf_reserve_proto;
1927 	case BPF_FUNC_ringbuf_submit:
1928 		return &bpf_ringbuf_submit_proto;
1929 	case BPF_FUNC_ringbuf_discard:
1930 		return &bpf_ringbuf_discard_proto;
1931 	case BPF_FUNC_ringbuf_query:
1932 		return &bpf_ringbuf_query_proto;
1933 	case BPF_FUNC_strncmp:
1934 		return &bpf_strncmp_proto;
1935 	case BPF_FUNC_strtol:
1936 		return &bpf_strtol_proto;
1937 	case BPF_FUNC_strtoul:
1938 		return &bpf_strtoul_proto;
1939 	case BPF_FUNC_get_current_pid_tgid:
1940 		return &bpf_get_current_pid_tgid_proto;
1941 	case BPF_FUNC_get_ns_current_pid_tgid:
1942 		return &bpf_get_ns_current_pid_tgid_proto;
1943 	default:
1944 		break;
1945 	}
1946 
1947 	if (!bpf_token_capable(prog->aux->token, CAP_BPF))
1948 		return NULL;
1949 
1950 	switch (func_id) {
1951 	case BPF_FUNC_spin_lock:
1952 		return &bpf_spin_lock_proto;
1953 	case BPF_FUNC_spin_unlock:
1954 		return &bpf_spin_unlock_proto;
1955 	case BPF_FUNC_jiffies64:
1956 		return &bpf_jiffies64_proto;
1957 	case BPF_FUNC_per_cpu_ptr:
1958 		return &bpf_per_cpu_ptr_proto;
1959 	case BPF_FUNC_this_cpu_ptr:
1960 		return &bpf_this_cpu_ptr_proto;
1961 	case BPF_FUNC_timer_init:
1962 		return &bpf_timer_init_proto;
1963 	case BPF_FUNC_timer_set_callback:
1964 		return &bpf_timer_set_callback_proto;
1965 	case BPF_FUNC_timer_start:
1966 		return &bpf_timer_start_proto;
1967 	case BPF_FUNC_timer_cancel:
1968 		return &bpf_timer_cancel_proto;
1969 	case BPF_FUNC_kptr_xchg:
1970 		return &bpf_kptr_xchg_proto;
1971 	case BPF_FUNC_for_each_map_elem:
1972 		return &bpf_for_each_map_elem_proto;
1973 	case BPF_FUNC_loop:
1974 		return &bpf_loop_proto;
1975 	case BPF_FUNC_user_ringbuf_drain:
1976 		return &bpf_user_ringbuf_drain_proto;
1977 	case BPF_FUNC_ringbuf_reserve_dynptr:
1978 		return &bpf_ringbuf_reserve_dynptr_proto;
1979 	case BPF_FUNC_ringbuf_submit_dynptr:
1980 		return &bpf_ringbuf_submit_dynptr_proto;
1981 	case BPF_FUNC_ringbuf_discard_dynptr:
1982 		return &bpf_ringbuf_discard_dynptr_proto;
1983 	case BPF_FUNC_dynptr_from_mem:
1984 		return &bpf_dynptr_from_mem_proto;
1985 	case BPF_FUNC_dynptr_read:
1986 		return &bpf_dynptr_read_proto;
1987 	case BPF_FUNC_dynptr_write:
1988 		return &bpf_dynptr_write_proto;
1989 	case BPF_FUNC_dynptr_data:
1990 		return &bpf_dynptr_data_proto;
1991 #ifdef CONFIG_CGROUPS
1992 	case BPF_FUNC_cgrp_storage_get:
1993 		return &bpf_cgrp_storage_get_proto;
1994 	case BPF_FUNC_cgrp_storage_delete:
1995 		return &bpf_cgrp_storage_delete_proto;
1996 	case BPF_FUNC_get_current_cgroup_id:
1997 		return &bpf_get_current_cgroup_id_proto;
1998 	case BPF_FUNC_get_current_ancestor_cgroup_id:
1999 		return &bpf_get_current_ancestor_cgroup_id_proto;
2000 #endif
2001 	default:
2002 		break;
2003 	}
2004 
2005 	if (!bpf_token_capable(prog->aux->token, CAP_PERFMON))
2006 		return NULL;
2007 
2008 	switch (func_id) {
2009 	case BPF_FUNC_trace_printk:
2010 		return bpf_get_trace_printk_proto();
2011 	case BPF_FUNC_get_current_task:
2012 		return &bpf_get_current_task_proto;
2013 	case BPF_FUNC_get_current_task_btf:
2014 		return &bpf_get_current_task_btf_proto;
2015 	case BPF_FUNC_probe_read_user:
2016 		return &bpf_probe_read_user_proto;
2017 	case BPF_FUNC_probe_read_kernel:
2018 		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
2019 		       NULL : &bpf_probe_read_kernel_proto;
2020 	case BPF_FUNC_probe_read_user_str:
2021 		return &bpf_probe_read_user_str_proto;
2022 	case BPF_FUNC_probe_read_kernel_str:
2023 		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
2024 		       NULL : &bpf_probe_read_kernel_str_proto;
2025 	case BPF_FUNC_snprintf_btf:
2026 		return &bpf_snprintf_btf_proto;
2027 	case BPF_FUNC_snprintf:
2028 		return &bpf_snprintf_proto;
2029 	case BPF_FUNC_task_pt_regs:
2030 		return &bpf_task_pt_regs_proto;
2031 	case BPF_FUNC_trace_vprintk:
2032 		return bpf_get_trace_vprintk_proto();
2033 	default:
2034 		return NULL;
2035 	}
2036 }
2037 EXPORT_SYMBOL_GPL(bpf_base_func_proto);
2038 
2039 void bpf_list_head_free(const struct btf_field *field, void *list_head,
2040 			struct bpf_spin_lock *spin_lock)
2041 {
2042 	struct list_head *head = list_head, *orig_head = list_head;
2043 
2044 	BUILD_BUG_ON(sizeof(struct list_head) > sizeof(struct bpf_list_head));
2045 	BUILD_BUG_ON(__alignof__(struct list_head) > __alignof__(struct bpf_list_head));
2046 
2047 	/* Do the actual list draining outside the lock to not hold the lock for
2048 	 * too long, and also prevent deadlocks if tracing programs end up
2049 	 * executing on entry/exit of functions called inside the critical
2050 	 * section, and end up doing map ops that call bpf_list_head_free for
2051 	 * the same map value again.
2052 	 */
2053 	__bpf_spin_lock_irqsave(spin_lock);
2054 	if (!head->next || list_empty(head))
2055 		goto unlock;
2056 	head = head->next;
2057 unlock:
2058 	INIT_LIST_HEAD(orig_head);
2059 	__bpf_spin_unlock_irqrestore(spin_lock);
2060 
2061 	while (head != orig_head) {
2062 		void *obj = head;
2063 
2064 		obj -= field->graph_root.node_offset;
2065 		head = head->next;
2066 		/* The contained type can also have resources, including a
2067 		 * bpf_list_head which needs to be freed.
2068 		 */
2069 		migrate_disable();
2070 		__bpf_obj_drop_impl(obj, field->graph_root.value_rec, false);
2071 		migrate_enable();
2072 	}
2073 }
2074 
2075 /* Like rbtree_postorder_for_each_entry_safe, but 'pos' and 'n' are
2076  * 'rb_node *', so field name of rb_node within containing struct is not
2077  * needed.
2078  *
2079  * Since bpf_rb_tree's node type has a corresponding struct btf_field with
2080  * graph_root.node_offset, it's not necessary to know field name
2081  * or type of node struct
2082  */
2083 #define bpf_rbtree_postorder_for_each_entry_safe(pos, n, root) \
2084 	for (pos = rb_first_postorder(root); \
2085 	    pos && ({ n = rb_next_postorder(pos); 1; }); \
2086 	    pos = n)
2087 
2088 void bpf_rb_root_free(const struct btf_field *field, void *rb_root,
2089 		      struct bpf_spin_lock *spin_lock)
2090 {
2091 	struct rb_root_cached orig_root, *root = rb_root;
2092 	struct rb_node *pos, *n;
2093 	void *obj;
2094 
2095 	BUILD_BUG_ON(sizeof(struct rb_root_cached) > sizeof(struct bpf_rb_root));
2096 	BUILD_BUG_ON(__alignof__(struct rb_root_cached) > __alignof__(struct bpf_rb_root));
2097 
2098 	__bpf_spin_lock_irqsave(spin_lock);
2099 	orig_root = *root;
2100 	*root = RB_ROOT_CACHED;
2101 	__bpf_spin_unlock_irqrestore(spin_lock);
2102 
2103 	bpf_rbtree_postorder_for_each_entry_safe(pos, n, &orig_root.rb_root) {
2104 		obj = pos;
2105 		obj -= field->graph_root.node_offset;
2106 
2107 
2108 		migrate_disable();
2109 		__bpf_obj_drop_impl(obj, field->graph_root.value_rec, false);
2110 		migrate_enable();
2111 	}
2112 }
2113 
2114 __bpf_kfunc_start_defs();
2115 
2116 __bpf_kfunc void *bpf_obj_new_impl(u64 local_type_id__k, void *meta__ign)
2117 {
2118 	struct btf_struct_meta *meta = meta__ign;
2119 	u64 size = local_type_id__k;
2120 	void *p;
2121 
2122 	p = bpf_mem_alloc(&bpf_global_ma, size);
2123 	if (!p)
2124 		return NULL;
2125 	if (meta)
2126 		bpf_obj_init(meta->record, p);
2127 	return p;
2128 }
2129 
2130 __bpf_kfunc void *bpf_percpu_obj_new_impl(u64 local_type_id__k, void *meta__ign)
2131 {
2132 	u64 size = local_type_id__k;
2133 
2134 	/* The verifier has ensured that meta__ign must be NULL */
2135 	return bpf_mem_alloc(&bpf_global_percpu_ma, size);
2136 }
2137 
2138 /* Must be called under migrate_disable(), as required by bpf_mem_free */
2139 void __bpf_obj_drop_impl(void *p, const struct btf_record *rec, bool percpu)
2140 {
2141 	struct bpf_mem_alloc *ma;
2142 
2143 	if (rec && rec->refcount_off >= 0 &&
2144 	    !refcount_dec_and_test((refcount_t *)(p + rec->refcount_off))) {
2145 		/* Object is refcounted and refcount_dec didn't result in 0
2146 		 * refcount. Return without freeing the object
2147 		 */
2148 		return;
2149 	}
2150 
2151 	if (rec)
2152 		bpf_obj_free_fields(rec, p);
2153 
2154 	if (percpu)
2155 		ma = &bpf_global_percpu_ma;
2156 	else
2157 		ma = &bpf_global_ma;
2158 	bpf_mem_free_rcu(ma, p);
2159 }
2160 
2161 __bpf_kfunc void bpf_obj_drop_impl(void *p__alloc, void *meta__ign)
2162 {
2163 	struct btf_struct_meta *meta = meta__ign;
2164 	void *p = p__alloc;
2165 
2166 	__bpf_obj_drop_impl(p, meta ? meta->record : NULL, false);
2167 }
2168 
2169 __bpf_kfunc void bpf_percpu_obj_drop_impl(void *p__alloc, void *meta__ign)
2170 {
2171 	/* The verifier has ensured that meta__ign must be NULL */
2172 	bpf_mem_free_rcu(&bpf_global_percpu_ma, p__alloc);
2173 }
2174 
2175 __bpf_kfunc void *bpf_refcount_acquire_impl(void *p__refcounted_kptr, void *meta__ign)
2176 {
2177 	struct btf_struct_meta *meta = meta__ign;
2178 	struct bpf_refcount *ref;
2179 
2180 	/* Could just cast directly to refcount_t *, but need some code using
2181 	 * bpf_refcount type so that it is emitted in vmlinux BTF
2182 	 */
2183 	ref = (struct bpf_refcount *)(p__refcounted_kptr + meta->record->refcount_off);
2184 	if (!refcount_inc_not_zero((refcount_t *)ref))
2185 		return NULL;
2186 
2187 	/* Verifier strips KF_RET_NULL if input is owned ref, see is_kfunc_ret_null
2188 	 * in verifier.c
2189 	 */
2190 	return (void *)p__refcounted_kptr;
2191 }
2192 
2193 static int __bpf_list_add(struct bpf_list_node_kern *node,
2194 			  struct bpf_list_head *head,
2195 			  bool tail, struct btf_record *rec, u64 off)
2196 {
2197 	struct list_head *n = &node->list_head, *h = (void *)head;
2198 
2199 	/* If list_head was 0-initialized by map, bpf_obj_init_field wasn't
2200 	 * called on its fields, so init here
2201 	 */
2202 	if (unlikely(!h->next))
2203 		INIT_LIST_HEAD(h);
2204 
2205 	/* node->owner != NULL implies !list_empty(n), no need to separately
2206 	 * check the latter
2207 	 */
2208 	if (cmpxchg(&node->owner, NULL, BPF_PTR_POISON)) {
2209 		/* Only called from BPF prog, no need to migrate_disable */
2210 		__bpf_obj_drop_impl((void *)n - off, rec, false);
2211 		return -EINVAL;
2212 	}
2213 
2214 	tail ? list_add_tail(n, h) : list_add(n, h);
2215 	WRITE_ONCE(node->owner, head);
2216 
2217 	return 0;
2218 }
2219 
2220 __bpf_kfunc int bpf_list_push_front_impl(struct bpf_list_head *head,
2221 					 struct bpf_list_node *node,
2222 					 void *meta__ign, u64 off)
2223 {
2224 	struct bpf_list_node_kern *n = (void *)node;
2225 	struct btf_struct_meta *meta = meta__ign;
2226 
2227 	return __bpf_list_add(n, head, false, meta ? meta->record : NULL, off);
2228 }
2229 
2230 __bpf_kfunc int bpf_list_push_back_impl(struct bpf_list_head *head,
2231 					struct bpf_list_node *node,
2232 					void *meta__ign, u64 off)
2233 {
2234 	struct bpf_list_node_kern *n = (void *)node;
2235 	struct btf_struct_meta *meta = meta__ign;
2236 
2237 	return __bpf_list_add(n, head, true, meta ? meta->record : NULL, off);
2238 }
2239 
2240 static struct bpf_list_node *__bpf_list_del(struct bpf_list_head *head, bool tail)
2241 {
2242 	struct list_head *n, *h = (void *)head;
2243 	struct bpf_list_node_kern *node;
2244 
2245 	/* If list_head was 0-initialized by map, bpf_obj_init_field wasn't
2246 	 * called on its fields, so init here
2247 	 */
2248 	if (unlikely(!h->next))
2249 		INIT_LIST_HEAD(h);
2250 	if (list_empty(h))
2251 		return NULL;
2252 
2253 	n = tail ? h->prev : h->next;
2254 	node = container_of(n, struct bpf_list_node_kern, list_head);
2255 	if (WARN_ON_ONCE(READ_ONCE(node->owner) != head))
2256 		return NULL;
2257 
2258 	list_del_init(n);
2259 	WRITE_ONCE(node->owner, NULL);
2260 	return (struct bpf_list_node *)n;
2261 }
2262 
2263 __bpf_kfunc struct bpf_list_node *bpf_list_pop_front(struct bpf_list_head *head)
2264 {
2265 	return __bpf_list_del(head, false);
2266 }
2267 
2268 __bpf_kfunc struct bpf_list_node *bpf_list_pop_back(struct bpf_list_head *head)
2269 {
2270 	return __bpf_list_del(head, true);
2271 }
2272 
2273 __bpf_kfunc struct bpf_rb_node *bpf_rbtree_remove(struct bpf_rb_root *root,
2274 						  struct bpf_rb_node *node)
2275 {
2276 	struct bpf_rb_node_kern *node_internal = (struct bpf_rb_node_kern *)node;
2277 	struct rb_root_cached *r = (struct rb_root_cached *)root;
2278 	struct rb_node *n = &node_internal->rb_node;
2279 
2280 	/* node_internal->owner != root implies either RB_EMPTY_NODE(n) or
2281 	 * n is owned by some other tree. No need to check RB_EMPTY_NODE(n)
2282 	 */
2283 	if (READ_ONCE(node_internal->owner) != root)
2284 		return NULL;
2285 
2286 	rb_erase_cached(n, r);
2287 	RB_CLEAR_NODE(n);
2288 	WRITE_ONCE(node_internal->owner, NULL);
2289 	return (struct bpf_rb_node *)n;
2290 }
2291 
2292 /* Need to copy rbtree_add_cached's logic here because our 'less' is a BPF
2293  * program
2294  */
2295 static int __bpf_rbtree_add(struct bpf_rb_root *root,
2296 			    struct bpf_rb_node_kern *node,
2297 			    void *less, struct btf_record *rec, u64 off)
2298 {
2299 	struct rb_node **link = &((struct rb_root_cached *)root)->rb_root.rb_node;
2300 	struct rb_node *parent = NULL, *n = &node->rb_node;
2301 	bpf_callback_t cb = (bpf_callback_t)less;
2302 	bool leftmost = true;
2303 
2304 	/* node->owner != NULL implies !RB_EMPTY_NODE(n), no need to separately
2305 	 * check the latter
2306 	 */
2307 	if (cmpxchg(&node->owner, NULL, BPF_PTR_POISON)) {
2308 		/* Only called from BPF prog, no need to migrate_disable */
2309 		__bpf_obj_drop_impl((void *)n - off, rec, false);
2310 		return -EINVAL;
2311 	}
2312 
2313 	while (*link) {
2314 		parent = *link;
2315 		if (cb((uintptr_t)node, (uintptr_t)parent, 0, 0, 0)) {
2316 			link = &parent->rb_left;
2317 		} else {
2318 			link = &parent->rb_right;
2319 			leftmost = false;
2320 		}
2321 	}
2322 
2323 	rb_link_node(n, parent, link);
2324 	rb_insert_color_cached(n, (struct rb_root_cached *)root, leftmost);
2325 	WRITE_ONCE(node->owner, root);
2326 	return 0;
2327 }
2328 
2329 __bpf_kfunc int bpf_rbtree_add_impl(struct bpf_rb_root *root, struct bpf_rb_node *node,
2330 				    bool (less)(struct bpf_rb_node *a, const struct bpf_rb_node *b),
2331 				    void *meta__ign, u64 off)
2332 {
2333 	struct btf_struct_meta *meta = meta__ign;
2334 	struct bpf_rb_node_kern *n = (void *)node;
2335 
2336 	return __bpf_rbtree_add(root, n, (void *)less, meta ? meta->record : NULL, off);
2337 }
2338 
2339 __bpf_kfunc struct bpf_rb_node *bpf_rbtree_first(struct bpf_rb_root *root)
2340 {
2341 	struct rb_root_cached *r = (struct rb_root_cached *)root;
2342 
2343 	return (struct bpf_rb_node *)rb_first_cached(r);
2344 }
2345 
2346 /**
2347  * bpf_task_acquire - Acquire a reference to a task. A task acquired by this
2348  * kfunc which is not stored in a map as a kptr, must be released by calling
2349  * bpf_task_release().
2350  * @p: The task on which a reference is being acquired.
2351  */
2352 __bpf_kfunc struct task_struct *bpf_task_acquire(struct task_struct *p)
2353 {
2354 	if (refcount_inc_not_zero(&p->rcu_users))
2355 		return p;
2356 	return NULL;
2357 }
2358 
2359 /**
2360  * bpf_task_release - Release the reference acquired on a task.
2361  * @p: The task on which a reference is being released.
2362  */
2363 __bpf_kfunc void bpf_task_release(struct task_struct *p)
2364 {
2365 	put_task_struct_rcu_user(p);
2366 }
2367 
2368 __bpf_kfunc void bpf_task_release_dtor(void *p)
2369 {
2370 	put_task_struct_rcu_user(p);
2371 }
2372 CFI_NOSEAL(bpf_task_release_dtor);
2373 
2374 #ifdef CONFIG_CGROUPS
2375 /**
2376  * bpf_cgroup_acquire - Acquire a reference to a cgroup. A cgroup acquired by
2377  * this kfunc which is not stored in a map as a kptr, must be released by
2378  * calling bpf_cgroup_release().
2379  * @cgrp: The cgroup on which a reference is being acquired.
2380  */
2381 __bpf_kfunc struct cgroup *bpf_cgroup_acquire(struct cgroup *cgrp)
2382 {
2383 	return cgroup_tryget(cgrp) ? cgrp : NULL;
2384 }
2385 
2386 /**
2387  * bpf_cgroup_release - Release the reference acquired on a cgroup.
2388  * If this kfunc is invoked in an RCU read region, the cgroup is guaranteed to
2389  * not be freed until the current grace period has ended, even if its refcount
2390  * drops to 0.
2391  * @cgrp: The cgroup on which a reference is being released.
2392  */
2393 __bpf_kfunc void bpf_cgroup_release(struct cgroup *cgrp)
2394 {
2395 	cgroup_put(cgrp);
2396 }
2397 
2398 __bpf_kfunc void bpf_cgroup_release_dtor(void *cgrp)
2399 {
2400 	cgroup_put(cgrp);
2401 }
2402 CFI_NOSEAL(bpf_cgroup_release_dtor);
2403 
2404 /**
2405  * bpf_cgroup_ancestor - Perform a lookup on an entry in a cgroup's ancestor
2406  * array. A cgroup returned by this kfunc which is not subsequently stored in a
2407  * map, must be released by calling bpf_cgroup_release().
2408  * @cgrp: The cgroup for which we're performing a lookup.
2409  * @level: The level of ancestor to look up.
2410  */
2411 __bpf_kfunc struct cgroup *bpf_cgroup_ancestor(struct cgroup *cgrp, int level)
2412 {
2413 	struct cgroup *ancestor;
2414 
2415 	if (level > cgrp->level || level < 0)
2416 		return NULL;
2417 
2418 	/* cgrp's refcnt could be 0 here, but ancestors can still be accessed */
2419 	ancestor = cgrp->ancestors[level];
2420 	if (!cgroup_tryget(ancestor))
2421 		return NULL;
2422 	return ancestor;
2423 }
2424 
2425 /**
2426  * bpf_cgroup_from_id - Find a cgroup from its ID. A cgroup returned by this
2427  * kfunc which is not subsequently stored in a map, must be released by calling
2428  * bpf_cgroup_release().
2429  * @cgid: cgroup id.
2430  */
2431 __bpf_kfunc struct cgroup *bpf_cgroup_from_id(u64 cgid)
2432 {
2433 	struct cgroup *cgrp;
2434 
2435 	cgrp = cgroup_get_from_id(cgid);
2436 	if (IS_ERR(cgrp))
2437 		return NULL;
2438 	return cgrp;
2439 }
2440 
2441 /**
2442  * bpf_task_under_cgroup - wrap task_under_cgroup_hierarchy() as a kfunc, test
2443  * task's membership of cgroup ancestry.
2444  * @task: the task to be tested
2445  * @ancestor: possible ancestor of @task's cgroup
2446  *
2447  * Tests whether @task's default cgroup hierarchy is a descendant of @ancestor.
2448  * It follows all the same rules as cgroup_is_descendant, and only applies
2449  * to the default hierarchy.
2450  */
2451 __bpf_kfunc long bpf_task_under_cgroup(struct task_struct *task,
2452 				       struct cgroup *ancestor)
2453 {
2454 	long ret;
2455 
2456 	rcu_read_lock();
2457 	ret = task_under_cgroup_hierarchy(task, ancestor);
2458 	rcu_read_unlock();
2459 	return ret;
2460 }
2461 
2462 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
2463 {
2464 	struct bpf_array *array = container_of(map, struct bpf_array, map);
2465 	struct cgroup *cgrp;
2466 
2467 	if (unlikely(idx >= array->map.max_entries))
2468 		return -E2BIG;
2469 
2470 	cgrp = READ_ONCE(array->ptrs[idx]);
2471 	if (unlikely(!cgrp))
2472 		return -EAGAIN;
2473 
2474 	return task_under_cgroup_hierarchy(current, cgrp);
2475 }
2476 
2477 const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
2478 	.func           = bpf_current_task_under_cgroup,
2479 	.gpl_only       = false,
2480 	.ret_type       = RET_INTEGER,
2481 	.arg1_type      = ARG_CONST_MAP_PTR,
2482 	.arg2_type      = ARG_ANYTHING,
2483 };
2484 
2485 /**
2486  * bpf_task_get_cgroup1 - Acquires the associated cgroup of a task within a
2487  * specific cgroup1 hierarchy. The cgroup1 hierarchy is identified by its
2488  * hierarchy ID.
2489  * @task: The target task
2490  * @hierarchy_id: The ID of a cgroup1 hierarchy
2491  *
2492  * On success, the cgroup is returen. On failure, NULL is returned.
2493  */
2494 __bpf_kfunc struct cgroup *
2495 bpf_task_get_cgroup1(struct task_struct *task, int hierarchy_id)
2496 {
2497 	struct cgroup *cgrp = task_get_cgroup1(task, hierarchy_id);
2498 
2499 	if (IS_ERR(cgrp))
2500 		return NULL;
2501 	return cgrp;
2502 }
2503 #endif /* CONFIG_CGROUPS */
2504 
2505 /**
2506  * bpf_task_from_pid - Find a struct task_struct from its pid by looking it up
2507  * in the root pid namespace idr. If a task is returned, it must either be
2508  * stored in a map, or released with bpf_task_release().
2509  * @pid: The pid of the task being looked up.
2510  */
2511 __bpf_kfunc struct task_struct *bpf_task_from_pid(s32 pid)
2512 {
2513 	struct task_struct *p;
2514 
2515 	rcu_read_lock();
2516 	p = find_task_by_pid_ns(pid, &init_pid_ns);
2517 	if (p)
2518 		p = bpf_task_acquire(p);
2519 	rcu_read_unlock();
2520 
2521 	return p;
2522 }
2523 
2524 /**
2525  * bpf_dynptr_slice() - Obtain a read-only pointer to the dynptr data.
2526  * @p: The dynptr whose data slice to retrieve
2527  * @offset: Offset into the dynptr
2528  * @buffer__opt: User-provided buffer to copy contents into.  May be NULL
2529  * @buffer__szk: Size (in bytes) of the buffer if present. This is the
2530  *               length of the requested slice. This must be a constant.
2531  *
2532  * For non-skb and non-xdp type dynptrs, there is no difference between
2533  * bpf_dynptr_slice and bpf_dynptr_data.
2534  *
2535  *  If buffer__opt is NULL, the call will fail if buffer_opt was needed.
2536  *
2537  * If the intention is to write to the data slice, please use
2538  * bpf_dynptr_slice_rdwr.
2539  *
2540  * The user must check that the returned pointer is not null before using it.
2541  *
2542  * Please note that in the case of skb and xdp dynptrs, bpf_dynptr_slice
2543  * does not change the underlying packet data pointers, so a call to
2544  * bpf_dynptr_slice will not invalidate any ctx->data/data_end pointers in
2545  * the bpf program.
2546  *
2547  * Return: NULL if the call failed (eg invalid dynptr), pointer to a read-only
2548  * data slice (can be either direct pointer to the data or a pointer to the user
2549  * provided buffer, with its contents containing the data, if unable to obtain
2550  * direct pointer)
2551  */
2552 __bpf_kfunc void *bpf_dynptr_slice(const struct bpf_dynptr *p, u32 offset,
2553 				   void *buffer__opt, u32 buffer__szk)
2554 {
2555 	const struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)p;
2556 	enum bpf_dynptr_type type;
2557 	u32 len = buffer__szk;
2558 	int err;
2559 
2560 	if (!ptr->data)
2561 		return NULL;
2562 
2563 	err = bpf_dynptr_check_off_len(ptr, offset, len);
2564 	if (err)
2565 		return NULL;
2566 
2567 	type = bpf_dynptr_get_type(ptr);
2568 
2569 	switch (type) {
2570 	case BPF_DYNPTR_TYPE_LOCAL:
2571 	case BPF_DYNPTR_TYPE_RINGBUF:
2572 		return ptr->data + ptr->offset + offset;
2573 	case BPF_DYNPTR_TYPE_SKB:
2574 		if (buffer__opt)
2575 			return skb_header_pointer(ptr->data, ptr->offset + offset, len, buffer__opt);
2576 		else
2577 			return skb_pointer_if_linear(ptr->data, ptr->offset + offset, len);
2578 	case BPF_DYNPTR_TYPE_XDP:
2579 	{
2580 		void *xdp_ptr = bpf_xdp_pointer(ptr->data, ptr->offset + offset, len);
2581 		if (!IS_ERR_OR_NULL(xdp_ptr))
2582 			return xdp_ptr;
2583 
2584 		if (!buffer__opt)
2585 			return NULL;
2586 		bpf_xdp_copy_buf(ptr->data, ptr->offset + offset, buffer__opt, len, false);
2587 		return buffer__opt;
2588 	}
2589 	default:
2590 		WARN_ONCE(true, "unknown dynptr type %d\n", type);
2591 		return NULL;
2592 	}
2593 }
2594 
2595 /**
2596  * bpf_dynptr_slice_rdwr() - Obtain a writable pointer to the dynptr data.
2597  * @p: The dynptr whose data slice to retrieve
2598  * @offset: Offset into the dynptr
2599  * @buffer__opt: User-provided buffer to copy contents into. May be NULL
2600  * @buffer__szk: Size (in bytes) of the buffer if present. This is the
2601  *               length of the requested slice. This must be a constant.
2602  *
2603  * For non-skb and non-xdp type dynptrs, there is no difference between
2604  * bpf_dynptr_slice and bpf_dynptr_data.
2605  *
2606  * If buffer__opt is NULL, the call will fail if buffer_opt was needed.
2607  *
2608  * The returned pointer is writable and may point to either directly the dynptr
2609  * data at the requested offset or to the buffer if unable to obtain a direct
2610  * data pointer to (example: the requested slice is to the paged area of an skb
2611  * packet). In the case where the returned pointer is to the buffer, the user
2612  * is responsible for persisting writes through calling bpf_dynptr_write(). This
2613  * usually looks something like this pattern:
2614  *
2615  * struct eth_hdr *eth = bpf_dynptr_slice_rdwr(&dynptr, 0, buffer, sizeof(buffer));
2616  * if (!eth)
2617  *	return TC_ACT_SHOT;
2618  *
2619  * // mutate eth header //
2620  *
2621  * if (eth == buffer)
2622  *	bpf_dynptr_write(&ptr, 0, buffer, sizeof(buffer), 0);
2623  *
2624  * Please note that, as in the example above, the user must check that the
2625  * returned pointer is not null before using it.
2626  *
2627  * Please also note that in the case of skb and xdp dynptrs, bpf_dynptr_slice_rdwr
2628  * does not change the underlying packet data pointers, so a call to
2629  * bpf_dynptr_slice_rdwr will not invalidate any ctx->data/data_end pointers in
2630  * the bpf program.
2631  *
2632  * Return: NULL if the call failed (eg invalid dynptr), pointer to a
2633  * data slice (can be either direct pointer to the data or a pointer to the user
2634  * provided buffer, with its contents containing the data, if unable to obtain
2635  * direct pointer)
2636  */
2637 __bpf_kfunc void *bpf_dynptr_slice_rdwr(const struct bpf_dynptr *p, u32 offset,
2638 					void *buffer__opt, u32 buffer__szk)
2639 {
2640 	const struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)p;
2641 
2642 	if (!ptr->data || __bpf_dynptr_is_rdonly(ptr))
2643 		return NULL;
2644 
2645 	/* bpf_dynptr_slice_rdwr is the same logic as bpf_dynptr_slice.
2646 	 *
2647 	 * For skb-type dynptrs, it is safe to write into the returned pointer
2648 	 * if the bpf program allows skb data writes. There are two possibilities
2649 	 * that may occur when calling bpf_dynptr_slice_rdwr:
2650 	 *
2651 	 * 1) The requested slice is in the head of the skb. In this case, the
2652 	 * returned pointer is directly to skb data, and if the skb is cloned, the
2653 	 * verifier will have uncloned it (see bpf_unclone_prologue()) already.
2654 	 * The pointer can be directly written into.
2655 	 *
2656 	 * 2) Some portion of the requested slice is in the paged buffer area.
2657 	 * In this case, the requested data will be copied out into the buffer
2658 	 * and the returned pointer will be a pointer to the buffer. The skb
2659 	 * will not be pulled. To persist the write, the user will need to call
2660 	 * bpf_dynptr_write(), which will pull the skb and commit the write.
2661 	 *
2662 	 * Similarly for xdp programs, if the requested slice is not across xdp
2663 	 * fragments, then a direct pointer will be returned, otherwise the data
2664 	 * will be copied out into the buffer and the user will need to call
2665 	 * bpf_dynptr_write() to commit changes.
2666 	 */
2667 	return bpf_dynptr_slice(p, offset, buffer__opt, buffer__szk);
2668 }
2669 
2670 __bpf_kfunc int bpf_dynptr_adjust(const struct bpf_dynptr *p, u32 start, u32 end)
2671 {
2672 	struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)p;
2673 	u32 size;
2674 
2675 	if (!ptr->data || start > end)
2676 		return -EINVAL;
2677 
2678 	size = __bpf_dynptr_size(ptr);
2679 
2680 	if (start > size || end > size)
2681 		return -ERANGE;
2682 
2683 	ptr->offset += start;
2684 	bpf_dynptr_set_size(ptr, end - start);
2685 
2686 	return 0;
2687 }
2688 
2689 __bpf_kfunc bool bpf_dynptr_is_null(const struct bpf_dynptr *p)
2690 {
2691 	struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)p;
2692 
2693 	return !ptr->data;
2694 }
2695 
2696 __bpf_kfunc bool bpf_dynptr_is_rdonly(const struct bpf_dynptr *p)
2697 {
2698 	struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)p;
2699 
2700 	if (!ptr->data)
2701 		return false;
2702 
2703 	return __bpf_dynptr_is_rdonly(ptr);
2704 }
2705 
2706 __bpf_kfunc __u32 bpf_dynptr_size(const struct bpf_dynptr *p)
2707 {
2708 	struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)p;
2709 
2710 	if (!ptr->data)
2711 		return -EINVAL;
2712 
2713 	return __bpf_dynptr_size(ptr);
2714 }
2715 
2716 __bpf_kfunc int bpf_dynptr_clone(const struct bpf_dynptr *p,
2717 				 struct bpf_dynptr *clone__uninit)
2718 {
2719 	struct bpf_dynptr_kern *clone = (struct bpf_dynptr_kern *)clone__uninit;
2720 	struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)p;
2721 
2722 	if (!ptr->data) {
2723 		bpf_dynptr_set_null(clone);
2724 		return -EINVAL;
2725 	}
2726 
2727 	*clone = *ptr;
2728 
2729 	return 0;
2730 }
2731 
2732 __bpf_kfunc void *bpf_cast_to_kern_ctx(void *obj)
2733 {
2734 	return obj;
2735 }
2736 
2737 __bpf_kfunc void *bpf_rdonly_cast(const void *obj__ign, u32 btf_id__k)
2738 {
2739 	return (void *)obj__ign;
2740 }
2741 
2742 __bpf_kfunc void bpf_rcu_read_lock(void)
2743 {
2744 	rcu_read_lock();
2745 }
2746 
2747 __bpf_kfunc void bpf_rcu_read_unlock(void)
2748 {
2749 	rcu_read_unlock();
2750 }
2751 
2752 struct bpf_throw_ctx {
2753 	struct bpf_prog_aux *aux;
2754 	u64 sp;
2755 	u64 bp;
2756 	int cnt;
2757 };
2758 
2759 static bool bpf_stack_walker(void *cookie, u64 ip, u64 sp, u64 bp)
2760 {
2761 	struct bpf_throw_ctx *ctx = cookie;
2762 	struct bpf_prog *prog;
2763 
2764 	if (!is_bpf_text_address(ip))
2765 		return !ctx->cnt;
2766 	prog = bpf_prog_ksym_find(ip);
2767 	ctx->cnt++;
2768 	if (bpf_is_subprog(prog))
2769 		return true;
2770 	ctx->aux = prog->aux;
2771 	ctx->sp = sp;
2772 	ctx->bp = bp;
2773 	return false;
2774 }
2775 
2776 __bpf_kfunc void bpf_throw(u64 cookie)
2777 {
2778 	struct bpf_throw_ctx ctx = {};
2779 
2780 	arch_bpf_stack_walk(bpf_stack_walker, &ctx);
2781 	WARN_ON_ONCE(!ctx.aux);
2782 	if (ctx.aux)
2783 		WARN_ON_ONCE(!ctx.aux->exception_boundary);
2784 	WARN_ON_ONCE(!ctx.bp);
2785 	WARN_ON_ONCE(!ctx.cnt);
2786 	/* Prevent KASAN false positives for CONFIG_KASAN_STACK by unpoisoning
2787 	 * deeper stack depths than ctx.sp as we do not return from bpf_throw,
2788 	 * which skips compiler generated instrumentation to do the same.
2789 	 */
2790 	kasan_unpoison_task_stack_below((void *)(long)ctx.sp);
2791 	ctx.aux->bpf_exception_cb(cookie, ctx.sp, ctx.bp, 0, 0);
2792 	WARN(1, "A call to BPF exception callback should never return\n");
2793 }
2794 
2795 __bpf_kfunc int bpf_wq_init(struct bpf_wq *wq, void *p__map, unsigned int flags)
2796 {
2797 	struct bpf_async_kern *async = (struct bpf_async_kern *)wq;
2798 	struct bpf_map *map = p__map;
2799 
2800 	BUILD_BUG_ON(sizeof(struct bpf_async_kern) > sizeof(struct bpf_wq));
2801 	BUILD_BUG_ON(__alignof__(struct bpf_async_kern) != __alignof__(struct bpf_wq));
2802 
2803 	if (flags)
2804 		return -EINVAL;
2805 
2806 	return __bpf_async_init(async, map, flags, BPF_ASYNC_TYPE_WQ);
2807 }
2808 
2809 __bpf_kfunc int bpf_wq_start(struct bpf_wq *wq, unsigned int flags)
2810 {
2811 	struct bpf_async_kern *async = (struct bpf_async_kern *)wq;
2812 	struct bpf_work *w;
2813 
2814 	if (in_nmi())
2815 		return -EOPNOTSUPP;
2816 	if (flags)
2817 		return -EINVAL;
2818 	w = READ_ONCE(async->work);
2819 	if (!w || !READ_ONCE(w->cb.prog))
2820 		return -EINVAL;
2821 
2822 	schedule_work(&w->work);
2823 	return 0;
2824 }
2825 
2826 __bpf_kfunc int bpf_wq_set_callback_impl(struct bpf_wq *wq,
2827 					 int (callback_fn)(void *map, int *key, void *value),
2828 					 unsigned int flags,
2829 					 void *aux__ign)
2830 {
2831 	struct bpf_prog_aux *aux = (struct bpf_prog_aux *)aux__ign;
2832 	struct bpf_async_kern *async = (struct bpf_async_kern *)wq;
2833 
2834 	if (flags)
2835 		return -EINVAL;
2836 
2837 	return __bpf_async_set_callback(async, callback_fn, aux, flags, BPF_ASYNC_TYPE_WQ);
2838 }
2839 
2840 __bpf_kfunc void bpf_preempt_disable(void)
2841 {
2842 	preempt_disable();
2843 }
2844 
2845 __bpf_kfunc void bpf_preempt_enable(void)
2846 {
2847 	preempt_enable();
2848 }
2849 
2850 struct bpf_iter_bits {
2851 	__u64 __opaque[2];
2852 } __aligned(8);
2853 
2854 #define BITS_ITER_NR_WORDS_MAX 511
2855 
2856 struct bpf_iter_bits_kern {
2857 	union {
2858 		__u64 *bits;
2859 		__u64 bits_copy;
2860 	};
2861 	int nr_bits;
2862 	int bit;
2863 } __aligned(8);
2864 
2865 /* On 64-bit hosts, unsigned long and u64 have the same size, so passing
2866  * a u64 pointer and an unsigned long pointer to find_next_bit() will
2867  * return the same result, as both point to the same 8-byte area.
2868  *
2869  * For 32-bit little-endian hosts, using a u64 pointer or unsigned long
2870  * pointer also makes no difference. This is because the first iterated
2871  * unsigned long is composed of bits 0-31 of the u64 and the second unsigned
2872  * long is composed of bits 32-63 of the u64.
2873  *
2874  * However, for 32-bit big-endian hosts, this is not the case. The first
2875  * iterated unsigned long will be bits 32-63 of the u64, so swap these two
2876  * ulong values within the u64.
2877  */
2878 static void swap_ulong_in_u64(u64 *bits, unsigned int nr)
2879 {
2880 #if (BITS_PER_LONG == 32) && defined(__BIG_ENDIAN)
2881 	unsigned int i;
2882 
2883 	for (i = 0; i < nr; i++)
2884 		bits[i] = (bits[i] >> 32) | ((u64)(u32)bits[i] << 32);
2885 #endif
2886 }
2887 
2888 /**
2889  * bpf_iter_bits_new() - Initialize a new bits iterator for a given memory area
2890  * @it: The new bpf_iter_bits to be created
2891  * @unsafe_ptr__ign: A pointer pointing to a memory area to be iterated over
2892  * @nr_words: The size of the specified memory area, measured in 8-byte units.
2893  * The maximum value of @nr_words is @BITS_ITER_NR_WORDS_MAX. This limit may be
2894  * further reduced by the BPF memory allocator implementation.
2895  *
2896  * This function initializes a new bpf_iter_bits structure for iterating over
2897  * a memory area which is specified by the @unsafe_ptr__ign and @nr_words. It
2898  * copies the data of the memory area to the newly created bpf_iter_bits @it for
2899  * subsequent iteration operations.
2900  *
2901  * On success, 0 is returned. On failure, ERR is returned.
2902  */
2903 __bpf_kfunc int
2904 bpf_iter_bits_new(struct bpf_iter_bits *it, const u64 *unsafe_ptr__ign, u32 nr_words)
2905 {
2906 	struct bpf_iter_bits_kern *kit = (void *)it;
2907 	u32 nr_bytes = nr_words * sizeof(u64);
2908 	u32 nr_bits = BYTES_TO_BITS(nr_bytes);
2909 	int err;
2910 
2911 	BUILD_BUG_ON(sizeof(struct bpf_iter_bits_kern) != sizeof(struct bpf_iter_bits));
2912 	BUILD_BUG_ON(__alignof__(struct bpf_iter_bits_kern) !=
2913 		     __alignof__(struct bpf_iter_bits));
2914 
2915 	kit->nr_bits = 0;
2916 	kit->bits_copy = 0;
2917 	kit->bit = -1;
2918 
2919 	if (!unsafe_ptr__ign || !nr_words)
2920 		return -EINVAL;
2921 	if (nr_words > BITS_ITER_NR_WORDS_MAX)
2922 		return -E2BIG;
2923 
2924 	/* Optimization for u64 mask */
2925 	if (nr_bits == 64) {
2926 		err = bpf_probe_read_kernel_common(&kit->bits_copy, nr_bytes, unsafe_ptr__ign);
2927 		if (err)
2928 			return -EFAULT;
2929 
2930 		swap_ulong_in_u64(&kit->bits_copy, nr_words);
2931 
2932 		kit->nr_bits = nr_bits;
2933 		return 0;
2934 	}
2935 
2936 	if (bpf_mem_alloc_check_size(false, nr_bytes))
2937 		return -E2BIG;
2938 
2939 	/* Fallback to memalloc */
2940 	kit->bits = bpf_mem_alloc(&bpf_global_ma, nr_bytes);
2941 	if (!kit->bits)
2942 		return -ENOMEM;
2943 
2944 	err = bpf_probe_read_kernel_common(kit->bits, nr_bytes, unsafe_ptr__ign);
2945 	if (err) {
2946 		bpf_mem_free(&bpf_global_ma, kit->bits);
2947 		return err;
2948 	}
2949 
2950 	swap_ulong_in_u64(kit->bits, nr_words);
2951 
2952 	kit->nr_bits = nr_bits;
2953 	return 0;
2954 }
2955 
2956 /**
2957  * bpf_iter_bits_next() - Get the next bit in a bpf_iter_bits
2958  * @it: The bpf_iter_bits to be checked
2959  *
2960  * This function returns a pointer to a number representing the value of the
2961  * next bit in the bits.
2962  *
2963  * If there are no further bits available, it returns NULL.
2964  */
2965 __bpf_kfunc int *bpf_iter_bits_next(struct bpf_iter_bits *it)
2966 {
2967 	struct bpf_iter_bits_kern *kit = (void *)it;
2968 	int bit = kit->bit, nr_bits = kit->nr_bits;
2969 	const void *bits;
2970 
2971 	if (!nr_bits || bit >= nr_bits)
2972 		return NULL;
2973 
2974 	bits = nr_bits == 64 ? &kit->bits_copy : kit->bits;
2975 	bit = find_next_bit(bits, nr_bits, bit + 1);
2976 	if (bit >= nr_bits) {
2977 		kit->bit = bit;
2978 		return NULL;
2979 	}
2980 
2981 	kit->bit = bit;
2982 	return &kit->bit;
2983 }
2984 
2985 /**
2986  * bpf_iter_bits_destroy() - Destroy a bpf_iter_bits
2987  * @it: The bpf_iter_bits to be destroyed
2988  *
2989  * Destroy the resource associated with the bpf_iter_bits.
2990  */
2991 __bpf_kfunc void bpf_iter_bits_destroy(struct bpf_iter_bits *it)
2992 {
2993 	struct bpf_iter_bits_kern *kit = (void *)it;
2994 
2995 	if (kit->nr_bits <= 64)
2996 		return;
2997 	bpf_mem_free(&bpf_global_ma, kit->bits);
2998 }
2999 
3000 /**
3001  * bpf_copy_from_user_str() - Copy a string from an unsafe user address
3002  * @dst:             Destination address, in kernel space.  This buffer must be
3003  *                   at least @dst__sz bytes long.
3004  * @dst__sz:         Maximum number of bytes to copy, includes the trailing NUL.
3005  * @unsafe_ptr__ign: Source address, in user space.
3006  * @flags:           The only supported flag is BPF_F_PAD_ZEROS
3007  *
3008  * Copies a NUL-terminated string from userspace to BPF space. If user string is
3009  * too long this will still ensure zero termination in the dst buffer unless
3010  * buffer size is 0.
3011  *
3012  * If BPF_F_PAD_ZEROS flag is set, memset the tail of @dst to 0 on success and
3013  * memset all of @dst on failure.
3014  */
3015 __bpf_kfunc int bpf_copy_from_user_str(void *dst, u32 dst__sz, const void __user *unsafe_ptr__ign, u64 flags)
3016 {
3017 	int ret;
3018 
3019 	if (unlikely(flags & ~BPF_F_PAD_ZEROS))
3020 		return -EINVAL;
3021 
3022 	if (unlikely(!dst__sz))
3023 		return 0;
3024 
3025 	ret = strncpy_from_user(dst, unsafe_ptr__ign, dst__sz - 1);
3026 	if (ret < 0) {
3027 		if (flags & BPF_F_PAD_ZEROS)
3028 			memset((char *)dst, 0, dst__sz);
3029 
3030 		return ret;
3031 	}
3032 
3033 	if (flags & BPF_F_PAD_ZEROS)
3034 		memset((char *)dst + ret, 0, dst__sz - ret);
3035 	else
3036 		((char *)dst)[ret] = '\0';
3037 
3038 	return ret + 1;
3039 }
3040 
3041 __bpf_kfunc_end_defs();
3042 
3043 BTF_KFUNCS_START(generic_btf_ids)
3044 #ifdef CONFIG_CRASH_DUMP
3045 BTF_ID_FLAGS(func, crash_kexec, KF_DESTRUCTIVE)
3046 #endif
3047 BTF_ID_FLAGS(func, bpf_obj_new_impl, KF_ACQUIRE | KF_RET_NULL)
3048 BTF_ID_FLAGS(func, bpf_percpu_obj_new_impl, KF_ACQUIRE | KF_RET_NULL)
3049 BTF_ID_FLAGS(func, bpf_obj_drop_impl, KF_RELEASE)
3050 BTF_ID_FLAGS(func, bpf_percpu_obj_drop_impl, KF_RELEASE)
3051 BTF_ID_FLAGS(func, bpf_refcount_acquire_impl, KF_ACQUIRE | KF_RET_NULL | KF_RCU)
3052 BTF_ID_FLAGS(func, bpf_list_push_front_impl)
3053 BTF_ID_FLAGS(func, bpf_list_push_back_impl)
3054 BTF_ID_FLAGS(func, bpf_list_pop_front, KF_ACQUIRE | KF_RET_NULL)
3055 BTF_ID_FLAGS(func, bpf_list_pop_back, KF_ACQUIRE | KF_RET_NULL)
3056 BTF_ID_FLAGS(func, bpf_task_acquire, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
3057 BTF_ID_FLAGS(func, bpf_task_release, KF_RELEASE)
3058 BTF_ID_FLAGS(func, bpf_rbtree_remove, KF_ACQUIRE | KF_RET_NULL)
3059 BTF_ID_FLAGS(func, bpf_rbtree_add_impl)
3060 BTF_ID_FLAGS(func, bpf_rbtree_first, KF_RET_NULL)
3061 
3062 #ifdef CONFIG_CGROUPS
3063 BTF_ID_FLAGS(func, bpf_cgroup_acquire, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
3064 BTF_ID_FLAGS(func, bpf_cgroup_release, KF_RELEASE)
3065 BTF_ID_FLAGS(func, bpf_cgroup_ancestor, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
3066 BTF_ID_FLAGS(func, bpf_cgroup_from_id, KF_ACQUIRE | KF_RET_NULL)
3067 BTF_ID_FLAGS(func, bpf_task_under_cgroup, KF_RCU)
3068 BTF_ID_FLAGS(func, bpf_task_get_cgroup1, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
3069 #endif
3070 BTF_ID_FLAGS(func, bpf_task_from_pid, KF_ACQUIRE | KF_RET_NULL)
3071 BTF_ID_FLAGS(func, bpf_throw)
3072 BTF_KFUNCS_END(generic_btf_ids)
3073 
3074 static const struct btf_kfunc_id_set generic_kfunc_set = {
3075 	.owner = THIS_MODULE,
3076 	.set   = &generic_btf_ids,
3077 };
3078 
3079 
3080 BTF_ID_LIST(generic_dtor_ids)
3081 BTF_ID(struct, task_struct)
3082 BTF_ID(func, bpf_task_release_dtor)
3083 #ifdef CONFIG_CGROUPS
3084 BTF_ID(struct, cgroup)
3085 BTF_ID(func, bpf_cgroup_release_dtor)
3086 #endif
3087 
3088 BTF_KFUNCS_START(common_btf_ids)
3089 BTF_ID_FLAGS(func, bpf_cast_to_kern_ctx)
3090 BTF_ID_FLAGS(func, bpf_rdonly_cast)
3091 BTF_ID_FLAGS(func, bpf_rcu_read_lock)
3092 BTF_ID_FLAGS(func, bpf_rcu_read_unlock)
3093 BTF_ID_FLAGS(func, bpf_dynptr_slice, KF_RET_NULL)
3094 BTF_ID_FLAGS(func, bpf_dynptr_slice_rdwr, KF_RET_NULL)
3095 BTF_ID_FLAGS(func, bpf_iter_num_new, KF_ITER_NEW)
3096 BTF_ID_FLAGS(func, bpf_iter_num_next, KF_ITER_NEXT | KF_RET_NULL)
3097 BTF_ID_FLAGS(func, bpf_iter_num_destroy, KF_ITER_DESTROY)
3098 BTF_ID_FLAGS(func, bpf_iter_task_vma_new, KF_ITER_NEW | KF_RCU)
3099 BTF_ID_FLAGS(func, bpf_iter_task_vma_next, KF_ITER_NEXT | KF_RET_NULL)
3100 BTF_ID_FLAGS(func, bpf_iter_task_vma_destroy, KF_ITER_DESTROY)
3101 #ifdef CONFIG_CGROUPS
3102 BTF_ID_FLAGS(func, bpf_iter_css_task_new, KF_ITER_NEW | KF_TRUSTED_ARGS)
3103 BTF_ID_FLAGS(func, bpf_iter_css_task_next, KF_ITER_NEXT | KF_RET_NULL)
3104 BTF_ID_FLAGS(func, bpf_iter_css_task_destroy, KF_ITER_DESTROY)
3105 BTF_ID_FLAGS(func, bpf_iter_css_new, KF_ITER_NEW | KF_TRUSTED_ARGS | KF_RCU_PROTECTED)
3106 BTF_ID_FLAGS(func, bpf_iter_css_next, KF_ITER_NEXT | KF_RET_NULL)
3107 BTF_ID_FLAGS(func, bpf_iter_css_destroy, KF_ITER_DESTROY)
3108 #endif
3109 BTF_ID_FLAGS(func, bpf_iter_task_new, KF_ITER_NEW | KF_TRUSTED_ARGS | KF_RCU_PROTECTED)
3110 BTF_ID_FLAGS(func, bpf_iter_task_next, KF_ITER_NEXT | KF_RET_NULL)
3111 BTF_ID_FLAGS(func, bpf_iter_task_destroy, KF_ITER_DESTROY)
3112 BTF_ID_FLAGS(func, bpf_dynptr_adjust)
3113 BTF_ID_FLAGS(func, bpf_dynptr_is_null)
3114 BTF_ID_FLAGS(func, bpf_dynptr_is_rdonly)
3115 BTF_ID_FLAGS(func, bpf_dynptr_size)
3116 BTF_ID_FLAGS(func, bpf_dynptr_clone)
3117 BTF_ID_FLAGS(func, bpf_modify_return_test_tp)
3118 BTF_ID_FLAGS(func, bpf_wq_init)
3119 BTF_ID_FLAGS(func, bpf_wq_set_callback_impl)
3120 BTF_ID_FLAGS(func, bpf_wq_start)
3121 BTF_ID_FLAGS(func, bpf_preempt_disable)
3122 BTF_ID_FLAGS(func, bpf_preempt_enable)
3123 BTF_ID_FLAGS(func, bpf_iter_bits_new, KF_ITER_NEW)
3124 BTF_ID_FLAGS(func, bpf_iter_bits_next, KF_ITER_NEXT | KF_RET_NULL)
3125 BTF_ID_FLAGS(func, bpf_iter_bits_destroy, KF_ITER_DESTROY)
3126 BTF_ID_FLAGS(func, bpf_copy_from_user_str, KF_SLEEPABLE)
3127 BTF_KFUNCS_END(common_btf_ids)
3128 
3129 static const struct btf_kfunc_id_set common_kfunc_set = {
3130 	.owner = THIS_MODULE,
3131 	.set   = &common_btf_ids,
3132 };
3133 
3134 static int __init kfunc_init(void)
3135 {
3136 	int ret;
3137 	const struct btf_id_dtor_kfunc generic_dtors[] = {
3138 		{
3139 			.btf_id       = generic_dtor_ids[0],
3140 			.kfunc_btf_id = generic_dtor_ids[1]
3141 		},
3142 #ifdef CONFIG_CGROUPS
3143 		{
3144 			.btf_id       = generic_dtor_ids[2],
3145 			.kfunc_btf_id = generic_dtor_ids[3]
3146 		},
3147 #endif
3148 	};
3149 
3150 	ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &generic_kfunc_set);
3151 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &generic_kfunc_set);
3152 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &generic_kfunc_set);
3153 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_STRUCT_OPS, &generic_kfunc_set);
3154 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SYSCALL, &generic_kfunc_set);
3155 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SKB, &generic_kfunc_set);
3156 	ret = ret ?: register_btf_id_dtor_kfuncs(generic_dtors,
3157 						  ARRAY_SIZE(generic_dtors),
3158 						  THIS_MODULE);
3159 	return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_UNSPEC, &common_kfunc_set);
3160 }
3161 
3162 late_initcall(kfunc_init);
3163 
3164 /* Get a pointer to dynptr data up to len bytes for read only access. If
3165  * the dynptr doesn't have continuous data up to len bytes, return NULL.
3166  */
3167 const void *__bpf_dynptr_data(const struct bpf_dynptr_kern *ptr, u32 len)
3168 {
3169 	const struct bpf_dynptr *p = (struct bpf_dynptr *)ptr;
3170 
3171 	return bpf_dynptr_slice(p, 0, NULL, len);
3172 }
3173 
3174 /* Get a pointer to dynptr data up to len bytes for read write access. If
3175  * the dynptr doesn't have continuous data up to len bytes, or the dynptr
3176  * is read only, return NULL.
3177  */
3178 void *__bpf_dynptr_data_rw(const struct bpf_dynptr_kern *ptr, u32 len)
3179 {
3180 	if (__bpf_dynptr_is_rdonly(ptr))
3181 		return NULL;
3182 	return (void *)__bpf_dynptr_data(ptr, len);
3183 }
3184