xref: /linux/kernel/trace/bpf_trace.c (revision 2c956a5ad4de7376ee792e888809edf2b2b39b86)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
3  * Copyright (c) 2016 Facebook
4  */
5 #include <linux/kernel.h>
6 #include <linux/types.h>
7 #include <linux/slab.h>
8 #include <linux/bpf.h>
9 #include <linux/bpf_perf_event.h>
10 #include <linux/filter.h>
11 #include <linux/uaccess.h>
12 #include <linux/ctype.h>
13 #include <linux/kprobes.h>
14 #include <linux/spinlock.h>
15 #include <linux/syscalls.h>
16 #include <linux/error-injection.h>
17 #include <linux/btf_ids.h>
18 
19 #include <asm/tlb.h>
20 
21 #include "trace_probe.h"
22 #include "trace.h"
23 
24 #define CREATE_TRACE_POINTS
25 #include "bpf_trace.h"
26 
27 #define bpf_event_rcu_dereference(p)					\
28 	rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
29 
30 #ifdef CONFIG_MODULES
31 struct bpf_trace_module {
32 	struct module *module;
33 	struct list_head list;
34 };
35 
36 static LIST_HEAD(bpf_trace_modules);
37 static DEFINE_MUTEX(bpf_module_mutex);
38 
39 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
40 {
41 	struct bpf_raw_event_map *btp, *ret = NULL;
42 	struct bpf_trace_module *btm;
43 	unsigned int i;
44 
45 	mutex_lock(&bpf_module_mutex);
46 	list_for_each_entry(btm, &bpf_trace_modules, list) {
47 		for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
48 			btp = &btm->module->bpf_raw_events[i];
49 			if (!strcmp(btp->tp->name, name)) {
50 				if (try_module_get(btm->module))
51 					ret = btp;
52 				goto out;
53 			}
54 		}
55 	}
56 out:
57 	mutex_unlock(&bpf_module_mutex);
58 	return ret;
59 }
60 #else
61 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
62 {
63 	return NULL;
64 }
65 #endif /* CONFIG_MODULES */
66 
67 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
68 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
69 
70 /**
71  * trace_call_bpf - invoke BPF program
72  * @call: tracepoint event
73  * @ctx: opaque context pointer
74  *
75  * kprobe handlers execute BPF programs via this helper.
76  * Can be used from static tracepoints in the future.
77  *
78  * Return: BPF programs always return an integer which is interpreted by
79  * kprobe handler as:
80  * 0 - return from kprobe (event is filtered out)
81  * 1 - store kprobe event into ring buffer
82  * Other values are reserved and currently alias to 1
83  */
84 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
85 {
86 	unsigned int ret;
87 
88 	if (in_nmi()) /* not supported yet */
89 		return 1;
90 
91 	cant_sleep();
92 
93 	if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
94 		/*
95 		 * since some bpf program is already running on this cpu,
96 		 * don't call into another bpf program (same or different)
97 		 * and don't send kprobe event into ring-buffer,
98 		 * so return zero here
99 		 */
100 		ret = 0;
101 		goto out;
102 	}
103 
104 	/*
105 	 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
106 	 * to all call sites, we did a bpf_prog_array_valid() there to check
107 	 * whether call->prog_array is empty or not, which is
108 	 * a heurisitc to speed up execution.
109 	 *
110 	 * If bpf_prog_array_valid() fetched prog_array was
111 	 * non-NULL, we go into trace_call_bpf() and do the actual
112 	 * proper rcu_dereference() under RCU lock.
113 	 * If it turns out that prog_array is NULL then, we bail out.
114 	 * For the opposite, if the bpf_prog_array_valid() fetched pointer
115 	 * was NULL, you'll skip the prog_array with the risk of missing
116 	 * out of events when it was updated in between this and the
117 	 * rcu_dereference() which is accepted risk.
118 	 */
119 	ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
120 
121  out:
122 	__this_cpu_dec(bpf_prog_active);
123 
124 	return ret;
125 }
126 
127 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
128 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
129 {
130 	regs_set_return_value(regs, rc);
131 	override_function_with_return(regs);
132 	return 0;
133 }
134 
135 static const struct bpf_func_proto bpf_override_return_proto = {
136 	.func		= bpf_override_return,
137 	.gpl_only	= true,
138 	.ret_type	= RET_INTEGER,
139 	.arg1_type	= ARG_PTR_TO_CTX,
140 	.arg2_type	= ARG_ANYTHING,
141 };
142 #endif
143 
144 static __always_inline int
145 bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
146 {
147 	int ret;
148 
149 	ret = copy_from_user_nofault(dst, unsafe_ptr, size);
150 	if (unlikely(ret < 0))
151 		memset(dst, 0, size);
152 	return ret;
153 }
154 
155 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
156 	   const void __user *, unsafe_ptr)
157 {
158 	return bpf_probe_read_user_common(dst, size, unsafe_ptr);
159 }
160 
161 const struct bpf_func_proto bpf_probe_read_user_proto = {
162 	.func		= bpf_probe_read_user,
163 	.gpl_only	= true,
164 	.ret_type	= RET_INTEGER,
165 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
166 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
167 	.arg3_type	= ARG_ANYTHING,
168 };
169 
170 static __always_inline int
171 bpf_probe_read_user_str_common(void *dst, u32 size,
172 			       const void __user *unsafe_ptr)
173 {
174 	int ret;
175 
176 	ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
177 	if (unlikely(ret < 0))
178 		memset(dst, 0, size);
179 	return ret;
180 }
181 
182 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
183 	   const void __user *, unsafe_ptr)
184 {
185 	return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
186 }
187 
188 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
189 	.func		= bpf_probe_read_user_str,
190 	.gpl_only	= true,
191 	.ret_type	= RET_INTEGER,
192 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
193 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
194 	.arg3_type	= ARG_ANYTHING,
195 };
196 
197 static __always_inline int
198 bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr)
199 {
200 	int ret = security_locked_down(LOCKDOWN_BPF_READ);
201 
202 	if (unlikely(ret < 0))
203 		goto fail;
204 	ret = copy_from_kernel_nofault(dst, unsafe_ptr, size);
205 	if (unlikely(ret < 0))
206 		goto fail;
207 	return ret;
208 fail:
209 	memset(dst, 0, size);
210 	return ret;
211 }
212 
213 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
214 	   const void *, unsafe_ptr)
215 {
216 	return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
217 }
218 
219 const struct bpf_func_proto bpf_probe_read_kernel_proto = {
220 	.func		= bpf_probe_read_kernel,
221 	.gpl_only	= true,
222 	.ret_type	= RET_INTEGER,
223 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
224 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
225 	.arg3_type	= ARG_ANYTHING,
226 };
227 
228 static __always_inline int
229 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
230 {
231 	int ret = security_locked_down(LOCKDOWN_BPF_READ);
232 
233 	if (unlikely(ret < 0))
234 		goto fail;
235 
236 	/*
237 	 * The strncpy_from_kernel_nofault() call will likely not fill the
238 	 * entire buffer, but that's okay in this circumstance as we're probing
239 	 * arbitrary memory anyway similar to bpf_probe_read_*() and might
240 	 * as well probe the stack. Thus, memory is explicitly cleared
241 	 * only in error case, so that improper users ignoring return
242 	 * code altogether don't copy garbage; otherwise length of string
243 	 * is returned that can be used for bpf_perf_event_output() et al.
244 	 */
245 	ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
246 	if (unlikely(ret < 0))
247 		goto fail;
248 
249 	return ret;
250 fail:
251 	memset(dst, 0, size);
252 	return ret;
253 }
254 
255 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
256 	   const void *, unsafe_ptr)
257 {
258 	return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
259 }
260 
261 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
262 	.func		= bpf_probe_read_kernel_str,
263 	.gpl_only	= true,
264 	.ret_type	= RET_INTEGER,
265 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
266 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
267 	.arg3_type	= ARG_ANYTHING,
268 };
269 
270 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
271 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
272 	   const void *, unsafe_ptr)
273 {
274 	if ((unsigned long)unsafe_ptr < TASK_SIZE) {
275 		return bpf_probe_read_user_common(dst, size,
276 				(__force void __user *)unsafe_ptr);
277 	}
278 	return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
279 }
280 
281 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
282 	.func		= bpf_probe_read_compat,
283 	.gpl_only	= true,
284 	.ret_type	= RET_INTEGER,
285 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
286 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
287 	.arg3_type	= ARG_ANYTHING,
288 };
289 
290 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
291 	   const void *, unsafe_ptr)
292 {
293 	if ((unsigned long)unsafe_ptr < TASK_SIZE) {
294 		return bpf_probe_read_user_str_common(dst, size,
295 				(__force void __user *)unsafe_ptr);
296 	}
297 	return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
298 }
299 
300 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
301 	.func		= bpf_probe_read_compat_str,
302 	.gpl_only	= true,
303 	.ret_type	= RET_INTEGER,
304 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
305 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
306 	.arg3_type	= ARG_ANYTHING,
307 };
308 #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
309 
310 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
311 	   u32, size)
312 {
313 	/*
314 	 * Ensure we're in user context which is safe for the helper to
315 	 * run. This helper has no business in a kthread.
316 	 *
317 	 * access_ok() should prevent writing to non-user memory, but in
318 	 * some situations (nommu, temporary switch, etc) access_ok() does
319 	 * not provide enough validation, hence the check on KERNEL_DS.
320 	 *
321 	 * nmi_uaccess_okay() ensures the probe is not run in an interim
322 	 * state, when the task or mm are switched. This is specifically
323 	 * required to prevent the use of temporary mm.
324 	 */
325 
326 	if (unlikely(in_interrupt() ||
327 		     current->flags & (PF_KTHREAD | PF_EXITING)))
328 		return -EPERM;
329 	if (unlikely(uaccess_kernel()))
330 		return -EPERM;
331 	if (unlikely(!nmi_uaccess_okay()))
332 		return -EPERM;
333 
334 	return copy_to_user_nofault(unsafe_ptr, src, size);
335 }
336 
337 static const struct bpf_func_proto bpf_probe_write_user_proto = {
338 	.func		= bpf_probe_write_user,
339 	.gpl_only	= true,
340 	.ret_type	= RET_INTEGER,
341 	.arg1_type	= ARG_ANYTHING,
342 	.arg2_type	= ARG_PTR_TO_MEM,
343 	.arg3_type	= ARG_CONST_SIZE,
344 };
345 
346 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
347 {
348 	if (!capable(CAP_SYS_ADMIN))
349 		return NULL;
350 
351 	pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
352 			    current->comm, task_pid_nr(current));
353 
354 	return &bpf_probe_write_user_proto;
355 }
356 
357 static void bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
358 		size_t bufsz)
359 {
360 	void __user *user_ptr = (__force void __user *)unsafe_ptr;
361 
362 	buf[0] = 0;
363 
364 	switch (fmt_ptype) {
365 	case 's':
366 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
367 		if ((unsigned long)unsafe_ptr < TASK_SIZE) {
368 			strncpy_from_user_nofault(buf, user_ptr, bufsz);
369 			break;
370 		}
371 		fallthrough;
372 #endif
373 	case 'k':
374 		strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz);
375 		break;
376 	case 'u':
377 		strncpy_from_user_nofault(buf, user_ptr, bufsz);
378 		break;
379 	}
380 }
381 
382 static DEFINE_RAW_SPINLOCK(trace_printk_lock);
383 
384 #define BPF_TRACE_PRINTK_SIZE   1024
385 
386 static __printf(1, 0) int bpf_do_trace_printk(const char *fmt, ...)
387 {
388 	static char buf[BPF_TRACE_PRINTK_SIZE];
389 	unsigned long flags;
390 	va_list ap;
391 	int ret;
392 
393 	raw_spin_lock_irqsave(&trace_printk_lock, flags);
394 	va_start(ap, fmt);
395 	ret = vsnprintf(buf, sizeof(buf), fmt, ap);
396 	va_end(ap);
397 	/* vsnprintf() will not append null for zero-length strings */
398 	if (ret == 0)
399 		buf[0] = '\0';
400 	trace_bpf_trace_printk(buf);
401 	raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
402 
403 	return ret;
404 }
405 
406 /*
407  * Only limited trace_printk() conversion specifiers allowed:
408  * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %pB %pks %pus %s
409  */
410 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
411 	   u64, arg2, u64, arg3)
412 {
413 	int i, mod[3] = {}, fmt_cnt = 0;
414 	char buf[64], fmt_ptype;
415 	void *unsafe_ptr = NULL;
416 	bool str_seen = false;
417 
418 	/*
419 	 * bpf_check()->check_func_arg()->check_stack_boundary()
420 	 * guarantees that fmt points to bpf program stack,
421 	 * fmt_size bytes of it were initialized and fmt_size > 0
422 	 */
423 	if (fmt[--fmt_size] != 0)
424 		return -EINVAL;
425 
426 	/* check format string for allowed specifiers */
427 	for (i = 0; i < fmt_size; i++) {
428 		if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
429 			return -EINVAL;
430 
431 		if (fmt[i] != '%')
432 			continue;
433 
434 		if (fmt_cnt >= 3)
435 			return -EINVAL;
436 
437 		/* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
438 		i++;
439 		if (fmt[i] == 'l') {
440 			mod[fmt_cnt]++;
441 			i++;
442 		} else if (fmt[i] == 'p') {
443 			mod[fmt_cnt]++;
444 			if ((fmt[i + 1] == 'k' ||
445 			     fmt[i + 1] == 'u') &&
446 			    fmt[i + 2] == 's') {
447 				fmt_ptype = fmt[i + 1];
448 				i += 2;
449 				goto fmt_str;
450 			}
451 
452 			if (fmt[i + 1] == 'B') {
453 				i++;
454 				goto fmt_next;
455 			}
456 
457 			/* disallow any further format extensions */
458 			if (fmt[i + 1] != 0 &&
459 			    !isspace(fmt[i + 1]) &&
460 			    !ispunct(fmt[i + 1]))
461 				return -EINVAL;
462 
463 			goto fmt_next;
464 		} else if (fmt[i] == 's') {
465 			mod[fmt_cnt]++;
466 			fmt_ptype = fmt[i];
467 fmt_str:
468 			if (str_seen)
469 				/* allow only one '%s' per fmt string */
470 				return -EINVAL;
471 			str_seen = true;
472 
473 			if (fmt[i + 1] != 0 &&
474 			    !isspace(fmt[i + 1]) &&
475 			    !ispunct(fmt[i + 1]))
476 				return -EINVAL;
477 
478 			switch (fmt_cnt) {
479 			case 0:
480 				unsafe_ptr = (void *)(long)arg1;
481 				arg1 = (long)buf;
482 				break;
483 			case 1:
484 				unsafe_ptr = (void *)(long)arg2;
485 				arg2 = (long)buf;
486 				break;
487 			case 2:
488 				unsafe_ptr = (void *)(long)arg3;
489 				arg3 = (long)buf;
490 				break;
491 			}
492 
493 			bpf_trace_copy_string(buf, unsafe_ptr, fmt_ptype,
494 					sizeof(buf));
495 			goto fmt_next;
496 		}
497 
498 		if (fmt[i] == 'l') {
499 			mod[fmt_cnt]++;
500 			i++;
501 		}
502 
503 		if (fmt[i] != 'i' && fmt[i] != 'd' &&
504 		    fmt[i] != 'u' && fmt[i] != 'x')
505 			return -EINVAL;
506 fmt_next:
507 		fmt_cnt++;
508 	}
509 
510 /* Horrid workaround for getting va_list handling working with different
511  * argument type combinations generically for 32 and 64 bit archs.
512  */
513 #define __BPF_TP_EMIT()	__BPF_ARG3_TP()
514 #define __BPF_TP(...)							\
515 	bpf_do_trace_printk(fmt, ##__VA_ARGS__)
516 
517 #define __BPF_ARG1_TP(...)						\
518 	((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64))	\
519 	  ? __BPF_TP(arg1, ##__VA_ARGS__)				\
520 	  : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32))	\
521 	      ? __BPF_TP((long)arg1, ##__VA_ARGS__)			\
522 	      : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
523 
524 #define __BPF_ARG2_TP(...)						\
525 	((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64))	\
526 	  ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__)				\
527 	  : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32))	\
528 	      ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__)		\
529 	      : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
530 
531 #define __BPF_ARG3_TP(...)						\
532 	((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64))	\
533 	  ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__)				\
534 	  : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32))	\
535 	      ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__)		\
536 	      : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
537 
538 	return __BPF_TP_EMIT();
539 }
540 
541 static const struct bpf_func_proto bpf_trace_printk_proto = {
542 	.func		= bpf_trace_printk,
543 	.gpl_only	= true,
544 	.ret_type	= RET_INTEGER,
545 	.arg1_type	= ARG_PTR_TO_MEM,
546 	.arg2_type	= ARG_CONST_SIZE,
547 };
548 
549 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
550 {
551 	/*
552 	 * This program might be calling bpf_trace_printk,
553 	 * so enable the associated bpf_trace/bpf_trace_printk event.
554 	 * Repeat this each time as it is possible a user has
555 	 * disabled bpf_trace_printk events.  By loading a program
556 	 * calling bpf_trace_printk() however the user has expressed
557 	 * the intent to see such events.
558 	 */
559 	if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
560 		pr_warn_ratelimited("could not enable bpf_trace_printk events");
561 
562 	return &bpf_trace_printk_proto;
563 }
564 
565 #define MAX_SEQ_PRINTF_VARARGS		12
566 #define MAX_SEQ_PRINTF_MAX_MEMCPY	6
567 #define MAX_SEQ_PRINTF_STR_LEN		128
568 
569 struct bpf_seq_printf_buf {
570 	char buf[MAX_SEQ_PRINTF_MAX_MEMCPY][MAX_SEQ_PRINTF_STR_LEN];
571 };
572 static DEFINE_PER_CPU(struct bpf_seq_printf_buf, bpf_seq_printf_buf);
573 static DEFINE_PER_CPU(int, bpf_seq_printf_buf_used);
574 
575 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
576 	   const void *, data, u32, data_len)
577 {
578 	int err = -EINVAL, fmt_cnt = 0, memcpy_cnt = 0;
579 	int i, buf_used, copy_size, num_args;
580 	u64 params[MAX_SEQ_PRINTF_VARARGS];
581 	struct bpf_seq_printf_buf *bufs;
582 	const u64 *args = data;
583 
584 	buf_used = this_cpu_inc_return(bpf_seq_printf_buf_used);
585 	if (WARN_ON_ONCE(buf_used > 1)) {
586 		err = -EBUSY;
587 		goto out;
588 	}
589 
590 	bufs = this_cpu_ptr(&bpf_seq_printf_buf);
591 
592 	/*
593 	 * bpf_check()->check_func_arg()->check_stack_boundary()
594 	 * guarantees that fmt points to bpf program stack,
595 	 * fmt_size bytes of it were initialized and fmt_size > 0
596 	 */
597 	if (fmt[--fmt_size] != 0)
598 		goto out;
599 
600 	if (data_len & 7)
601 		goto out;
602 
603 	for (i = 0; i < fmt_size; i++) {
604 		if (fmt[i] == '%') {
605 			if (fmt[i + 1] == '%')
606 				i++;
607 			else if (!data || !data_len)
608 				goto out;
609 		}
610 	}
611 
612 	num_args = data_len / 8;
613 
614 	/* check format string for allowed specifiers */
615 	for (i = 0; i < fmt_size; i++) {
616 		/* only printable ascii for now. */
617 		if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
618 			err = -EINVAL;
619 			goto out;
620 		}
621 
622 		if (fmt[i] != '%')
623 			continue;
624 
625 		if (fmt[i + 1] == '%') {
626 			i++;
627 			continue;
628 		}
629 
630 		if (fmt_cnt >= MAX_SEQ_PRINTF_VARARGS) {
631 			err = -E2BIG;
632 			goto out;
633 		}
634 
635 		if (fmt_cnt >= num_args) {
636 			err = -EINVAL;
637 			goto out;
638 		}
639 
640 		/* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
641 		i++;
642 
643 		/* skip optional "[0 +-][num]" width formating field */
644 		while (fmt[i] == '0' || fmt[i] == '+'  || fmt[i] == '-' ||
645 		       fmt[i] == ' ')
646 			i++;
647 		if (fmt[i] >= '1' && fmt[i] <= '9') {
648 			i++;
649 			while (fmt[i] >= '0' && fmt[i] <= '9')
650 				i++;
651 		}
652 
653 		if (fmt[i] == 's') {
654 			void *unsafe_ptr;
655 
656 			/* try our best to copy */
657 			if (memcpy_cnt >= MAX_SEQ_PRINTF_MAX_MEMCPY) {
658 				err = -E2BIG;
659 				goto out;
660 			}
661 
662 			unsafe_ptr = (void *)(long)args[fmt_cnt];
663 			err = strncpy_from_kernel_nofault(bufs->buf[memcpy_cnt],
664 					unsafe_ptr, MAX_SEQ_PRINTF_STR_LEN);
665 			if (err < 0)
666 				bufs->buf[memcpy_cnt][0] = '\0';
667 			params[fmt_cnt] = (u64)(long)bufs->buf[memcpy_cnt];
668 
669 			fmt_cnt++;
670 			memcpy_cnt++;
671 			continue;
672 		}
673 
674 		if (fmt[i] == 'p') {
675 			if (fmt[i + 1] == 0 ||
676 			    fmt[i + 1] == 'K' ||
677 			    fmt[i + 1] == 'x' ||
678 			    fmt[i + 1] == 'B') {
679 				/* just kernel pointers */
680 				params[fmt_cnt] = args[fmt_cnt];
681 				fmt_cnt++;
682 				continue;
683 			}
684 
685 			/* only support "%pI4", "%pi4", "%pI6" and "%pi6". */
686 			if (fmt[i + 1] != 'i' && fmt[i + 1] != 'I') {
687 				err = -EINVAL;
688 				goto out;
689 			}
690 			if (fmt[i + 2] != '4' && fmt[i + 2] != '6') {
691 				err = -EINVAL;
692 				goto out;
693 			}
694 
695 			if (memcpy_cnt >= MAX_SEQ_PRINTF_MAX_MEMCPY) {
696 				err = -E2BIG;
697 				goto out;
698 			}
699 
700 
701 			copy_size = (fmt[i + 2] == '4') ? 4 : 16;
702 
703 			err = copy_from_kernel_nofault(bufs->buf[memcpy_cnt],
704 						(void *) (long) args[fmt_cnt],
705 						copy_size);
706 			if (err < 0)
707 				memset(bufs->buf[memcpy_cnt], 0, copy_size);
708 			params[fmt_cnt] = (u64)(long)bufs->buf[memcpy_cnt];
709 
710 			i += 2;
711 			fmt_cnt++;
712 			memcpy_cnt++;
713 			continue;
714 		}
715 
716 		if (fmt[i] == 'l') {
717 			i++;
718 			if (fmt[i] == 'l')
719 				i++;
720 		}
721 
722 		if (fmt[i] != 'i' && fmt[i] != 'd' &&
723 		    fmt[i] != 'u' && fmt[i] != 'x' &&
724 		    fmt[i] != 'X') {
725 			err = -EINVAL;
726 			goto out;
727 		}
728 
729 		params[fmt_cnt] = args[fmt_cnt];
730 		fmt_cnt++;
731 	}
732 
733 	/* Maximumly we can have MAX_SEQ_PRINTF_VARARGS parameter, just give
734 	 * all of them to seq_printf().
735 	 */
736 	seq_printf(m, fmt, params[0], params[1], params[2], params[3],
737 		   params[4], params[5], params[6], params[7], params[8],
738 		   params[9], params[10], params[11]);
739 
740 	err = seq_has_overflowed(m) ? -EOVERFLOW : 0;
741 out:
742 	this_cpu_dec(bpf_seq_printf_buf_used);
743 	return err;
744 }
745 
746 BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
747 
748 static const struct bpf_func_proto bpf_seq_printf_proto = {
749 	.func		= bpf_seq_printf,
750 	.gpl_only	= true,
751 	.ret_type	= RET_INTEGER,
752 	.arg1_type	= ARG_PTR_TO_BTF_ID,
753 	.arg1_btf_id	= &btf_seq_file_ids[0],
754 	.arg2_type	= ARG_PTR_TO_MEM,
755 	.arg3_type	= ARG_CONST_SIZE,
756 	.arg4_type      = ARG_PTR_TO_MEM_OR_NULL,
757 	.arg5_type      = ARG_CONST_SIZE_OR_ZERO,
758 };
759 
760 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
761 {
762 	return seq_write(m, data, len) ? -EOVERFLOW : 0;
763 }
764 
765 static const struct bpf_func_proto bpf_seq_write_proto = {
766 	.func		= bpf_seq_write,
767 	.gpl_only	= true,
768 	.ret_type	= RET_INTEGER,
769 	.arg1_type	= ARG_PTR_TO_BTF_ID,
770 	.arg1_btf_id	= &btf_seq_file_ids[0],
771 	.arg2_type	= ARG_PTR_TO_MEM,
772 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
773 };
774 
775 static __always_inline int
776 get_map_perf_counter(struct bpf_map *map, u64 flags,
777 		     u64 *value, u64 *enabled, u64 *running)
778 {
779 	struct bpf_array *array = container_of(map, struct bpf_array, map);
780 	unsigned int cpu = smp_processor_id();
781 	u64 index = flags & BPF_F_INDEX_MASK;
782 	struct bpf_event_entry *ee;
783 
784 	if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
785 		return -EINVAL;
786 	if (index == BPF_F_CURRENT_CPU)
787 		index = cpu;
788 	if (unlikely(index >= array->map.max_entries))
789 		return -E2BIG;
790 
791 	ee = READ_ONCE(array->ptrs[index]);
792 	if (!ee)
793 		return -ENOENT;
794 
795 	return perf_event_read_local(ee->event, value, enabled, running);
796 }
797 
798 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
799 {
800 	u64 value = 0;
801 	int err;
802 
803 	err = get_map_perf_counter(map, flags, &value, NULL, NULL);
804 	/*
805 	 * this api is ugly since we miss [-22..-2] range of valid
806 	 * counter values, but that's uapi
807 	 */
808 	if (err)
809 		return err;
810 	return value;
811 }
812 
813 static const struct bpf_func_proto bpf_perf_event_read_proto = {
814 	.func		= bpf_perf_event_read,
815 	.gpl_only	= true,
816 	.ret_type	= RET_INTEGER,
817 	.arg1_type	= ARG_CONST_MAP_PTR,
818 	.arg2_type	= ARG_ANYTHING,
819 };
820 
821 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
822 	   struct bpf_perf_event_value *, buf, u32, size)
823 {
824 	int err = -EINVAL;
825 
826 	if (unlikely(size != sizeof(struct bpf_perf_event_value)))
827 		goto clear;
828 	err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
829 				   &buf->running);
830 	if (unlikely(err))
831 		goto clear;
832 	return 0;
833 clear:
834 	memset(buf, 0, size);
835 	return err;
836 }
837 
838 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
839 	.func		= bpf_perf_event_read_value,
840 	.gpl_only	= true,
841 	.ret_type	= RET_INTEGER,
842 	.arg1_type	= ARG_CONST_MAP_PTR,
843 	.arg2_type	= ARG_ANYTHING,
844 	.arg3_type	= ARG_PTR_TO_UNINIT_MEM,
845 	.arg4_type	= ARG_CONST_SIZE,
846 };
847 
848 static __always_inline u64
849 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
850 			u64 flags, struct perf_sample_data *sd)
851 {
852 	struct bpf_array *array = container_of(map, struct bpf_array, map);
853 	unsigned int cpu = smp_processor_id();
854 	u64 index = flags & BPF_F_INDEX_MASK;
855 	struct bpf_event_entry *ee;
856 	struct perf_event *event;
857 
858 	if (index == BPF_F_CURRENT_CPU)
859 		index = cpu;
860 	if (unlikely(index >= array->map.max_entries))
861 		return -E2BIG;
862 
863 	ee = READ_ONCE(array->ptrs[index]);
864 	if (!ee)
865 		return -ENOENT;
866 
867 	event = ee->event;
868 	if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
869 		     event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
870 		return -EINVAL;
871 
872 	if (unlikely(event->oncpu != cpu))
873 		return -EOPNOTSUPP;
874 
875 	return perf_event_output(event, sd, regs);
876 }
877 
878 /*
879  * Support executing tracepoints in normal, irq, and nmi context that each call
880  * bpf_perf_event_output
881  */
882 struct bpf_trace_sample_data {
883 	struct perf_sample_data sds[3];
884 };
885 
886 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
887 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
888 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
889 	   u64, flags, void *, data, u64, size)
890 {
891 	struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
892 	int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
893 	struct perf_raw_record raw = {
894 		.frag = {
895 			.size = size,
896 			.data = data,
897 		},
898 	};
899 	struct perf_sample_data *sd;
900 	int err;
901 
902 	if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
903 		err = -EBUSY;
904 		goto out;
905 	}
906 
907 	sd = &sds->sds[nest_level - 1];
908 
909 	if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
910 		err = -EINVAL;
911 		goto out;
912 	}
913 
914 	perf_sample_data_init(sd, 0, 0);
915 	sd->raw = &raw;
916 
917 	err = __bpf_perf_event_output(regs, map, flags, sd);
918 
919 out:
920 	this_cpu_dec(bpf_trace_nest_level);
921 	return err;
922 }
923 
924 static const struct bpf_func_proto bpf_perf_event_output_proto = {
925 	.func		= bpf_perf_event_output,
926 	.gpl_only	= true,
927 	.ret_type	= RET_INTEGER,
928 	.arg1_type	= ARG_PTR_TO_CTX,
929 	.arg2_type	= ARG_CONST_MAP_PTR,
930 	.arg3_type	= ARG_ANYTHING,
931 	.arg4_type	= ARG_PTR_TO_MEM,
932 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
933 };
934 
935 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
936 struct bpf_nested_pt_regs {
937 	struct pt_regs regs[3];
938 };
939 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
940 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
941 
942 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
943 		     void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
944 {
945 	int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
946 	struct perf_raw_frag frag = {
947 		.copy		= ctx_copy,
948 		.size		= ctx_size,
949 		.data		= ctx,
950 	};
951 	struct perf_raw_record raw = {
952 		.frag = {
953 			{
954 				.next	= ctx_size ? &frag : NULL,
955 			},
956 			.size	= meta_size,
957 			.data	= meta,
958 		},
959 	};
960 	struct perf_sample_data *sd;
961 	struct pt_regs *regs;
962 	u64 ret;
963 
964 	if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
965 		ret = -EBUSY;
966 		goto out;
967 	}
968 	sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
969 	regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
970 
971 	perf_fetch_caller_regs(regs);
972 	perf_sample_data_init(sd, 0, 0);
973 	sd->raw = &raw;
974 
975 	ret = __bpf_perf_event_output(regs, map, flags, sd);
976 out:
977 	this_cpu_dec(bpf_event_output_nest_level);
978 	return ret;
979 }
980 
981 BPF_CALL_0(bpf_get_current_task)
982 {
983 	return (long) current;
984 }
985 
986 const struct bpf_func_proto bpf_get_current_task_proto = {
987 	.func		= bpf_get_current_task,
988 	.gpl_only	= true,
989 	.ret_type	= RET_INTEGER,
990 };
991 
992 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
993 {
994 	struct bpf_array *array = container_of(map, struct bpf_array, map);
995 	struct cgroup *cgrp;
996 
997 	if (unlikely(idx >= array->map.max_entries))
998 		return -E2BIG;
999 
1000 	cgrp = READ_ONCE(array->ptrs[idx]);
1001 	if (unlikely(!cgrp))
1002 		return -EAGAIN;
1003 
1004 	return task_under_cgroup_hierarchy(current, cgrp);
1005 }
1006 
1007 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
1008 	.func           = bpf_current_task_under_cgroup,
1009 	.gpl_only       = false,
1010 	.ret_type       = RET_INTEGER,
1011 	.arg1_type      = ARG_CONST_MAP_PTR,
1012 	.arg2_type      = ARG_ANYTHING,
1013 };
1014 
1015 struct send_signal_irq_work {
1016 	struct irq_work irq_work;
1017 	struct task_struct *task;
1018 	u32 sig;
1019 	enum pid_type type;
1020 };
1021 
1022 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
1023 
1024 static void do_bpf_send_signal(struct irq_work *entry)
1025 {
1026 	struct send_signal_irq_work *work;
1027 
1028 	work = container_of(entry, struct send_signal_irq_work, irq_work);
1029 	group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
1030 }
1031 
1032 static int bpf_send_signal_common(u32 sig, enum pid_type type)
1033 {
1034 	struct send_signal_irq_work *work = NULL;
1035 
1036 	/* Similar to bpf_probe_write_user, task needs to be
1037 	 * in a sound condition and kernel memory access be
1038 	 * permitted in order to send signal to the current
1039 	 * task.
1040 	 */
1041 	if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
1042 		return -EPERM;
1043 	if (unlikely(uaccess_kernel()))
1044 		return -EPERM;
1045 	if (unlikely(!nmi_uaccess_okay()))
1046 		return -EPERM;
1047 
1048 	if (irqs_disabled()) {
1049 		/* Do an early check on signal validity. Otherwise,
1050 		 * the error is lost in deferred irq_work.
1051 		 */
1052 		if (unlikely(!valid_signal(sig)))
1053 			return -EINVAL;
1054 
1055 		work = this_cpu_ptr(&send_signal_work);
1056 		if (atomic_read(&work->irq_work.flags) & IRQ_WORK_BUSY)
1057 			return -EBUSY;
1058 
1059 		/* Add the current task, which is the target of sending signal,
1060 		 * to the irq_work. The current task may change when queued
1061 		 * irq works get executed.
1062 		 */
1063 		work->task = current;
1064 		work->sig = sig;
1065 		work->type = type;
1066 		irq_work_queue(&work->irq_work);
1067 		return 0;
1068 	}
1069 
1070 	return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
1071 }
1072 
1073 BPF_CALL_1(bpf_send_signal, u32, sig)
1074 {
1075 	return bpf_send_signal_common(sig, PIDTYPE_TGID);
1076 }
1077 
1078 static const struct bpf_func_proto bpf_send_signal_proto = {
1079 	.func		= bpf_send_signal,
1080 	.gpl_only	= false,
1081 	.ret_type	= RET_INTEGER,
1082 	.arg1_type	= ARG_ANYTHING,
1083 };
1084 
1085 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
1086 {
1087 	return bpf_send_signal_common(sig, PIDTYPE_PID);
1088 }
1089 
1090 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
1091 	.func		= bpf_send_signal_thread,
1092 	.gpl_only	= false,
1093 	.ret_type	= RET_INTEGER,
1094 	.arg1_type	= ARG_ANYTHING,
1095 };
1096 
1097 BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
1098 {
1099 	long len;
1100 	char *p;
1101 
1102 	if (!sz)
1103 		return 0;
1104 
1105 	p = d_path(path, buf, sz);
1106 	if (IS_ERR(p)) {
1107 		len = PTR_ERR(p);
1108 	} else {
1109 		len = buf + sz - p;
1110 		memmove(buf, p, len);
1111 	}
1112 
1113 	return len;
1114 }
1115 
1116 BTF_SET_START(btf_allowlist_d_path)
1117 #ifdef CONFIG_SECURITY
1118 BTF_ID(func, security_file_permission)
1119 BTF_ID(func, security_inode_getattr)
1120 BTF_ID(func, security_file_open)
1121 #endif
1122 #ifdef CONFIG_SECURITY_PATH
1123 BTF_ID(func, security_path_truncate)
1124 #endif
1125 BTF_ID(func, vfs_truncate)
1126 BTF_ID(func, vfs_fallocate)
1127 BTF_ID(func, dentry_open)
1128 BTF_ID(func, vfs_getattr)
1129 BTF_ID(func, filp_close)
1130 BTF_SET_END(btf_allowlist_d_path)
1131 
1132 static bool bpf_d_path_allowed(const struct bpf_prog *prog)
1133 {
1134 	return btf_id_set_contains(&btf_allowlist_d_path, prog->aux->attach_btf_id);
1135 }
1136 
1137 BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
1138 
1139 static const struct bpf_func_proto bpf_d_path_proto = {
1140 	.func		= bpf_d_path,
1141 	.gpl_only	= false,
1142 	.ret_type	= RET_INTEGER,
1143 	.arg1_type	= ARG_PTR_TO_BTF_ID,
1144 	.arg1_btf_id	= &bpf_d_path_btf_ids[0],
1145 	.arg2_type	= ARG_PTR_TO_MEM,
1146 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
1147 	.allowed	= bpf_d_path_allowed,
1148 };
1149 
1150 const struct bpf_func_proto *
1151 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1152 {
1153 	switch (func_id) {
1154 	case BPF_FUNC_map_lookup_elem:
1155 		return &bpf_map_lookup_elem_proto;
1156 	case BPF_FUNC_map_update_elem:
1157 		return &bpf_map_update_elem_proto;
1158 	case BPF_FUNC_map_delete_elem:
1159 		return &bpf_map_delete_elem_proto;
1160 	case BPF_FUNC_map_push_elem:
1161 		return &bpf_map_push_elem_proto;
1162 	case BPF_FUNC_map_pop_elem:
1163 		return &bpf_map_pop_elem_proto;
1164 	case BPF_FUNC_map_peek_elem:
1165 		return &bpf_map_peek_elem_proto;
1166 	case BPF_FUNC_ktime_get_ns:
1167 		return &bpf_ktime_get_ns_proto;
1168 	case BPF_FUNC_ktime_get_boot_ns:
1169 		return &bpf_ktime_get_boot_ns_proto;
1170 	case BPF_FUNC_tail_call:
1171 		return &bpf_tail_call_proto;
1172 	case BPF_FUNC_get_current_pid_tgid:
1173 		return &bpf_get_current_pid_tgid_proto;
1174 	case BPF_FUNC_get_current_task:
1175 		return &bpf_get_current_task_proto;
1176 	case BPF_FUNC_get_current_uid_gid:
1177 		return &bpf_get_current_uid_gid_proto;
1178 	case BPF_FUNC_get_current_comm:
1179 		return &bpf_get_current_comm_proto;
1180 	case BPF_FUNC_trace_printk:
1181 		return bpf_get_trace_printk_proto();
1182 	case BPF_FUNC_get_smp_processor_id:
1183 		return &bpf_get_smp_processor_id_proto;
1184 	case BPF_FUNC_get_numa_node_id:
1185 		return &bpf_get_numa_node_id_proto;
1186 	case BPF_FUNC_perf_event_read:
1187 		return &bpf_perf_event_read_proto;
1188 	case BPF_FUNC_probe_write_user:
1189 		return bpf_get_probe_write_proto();
1190 	case BPF_FUNC_current_task_under_cgroup:
1191 		return &bpf_current_task_under_cgroup_proto;
1192 	case BPF_FUNC_get_prandom_u32:
1193 		return &bpf_get_prandom_u32_proto;
1194 	case BPF_FUNC_probe_read_user:
1195 		return &bpf_probe_read_user_proto;
1196 	case BPF_FUNC_probe_read_kernel:
1197 		return &bpf_probe_read_kernel_proto;
1198 	case BPF_FUNC_probe_read_user_str:
1199 		return &bpf_probe_read_user_str_proto;
1200 	case BPF_FUNC_probe_read_kernel_str:
1201 		return &bpf_probe_read_kernel_str_proto;
1202 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1203 	case BPF_FUNC_probe_read:
1204 		return &bpf_probe_read_compat_proto;
1205 	case BPF_FUNC_probe_read_str:
1206 		return &bpf_probe_read_compat_str_proto;
1207 #endif
1208 #ifdef CONFIG_CGROUPS
1209 	case BPF_FUNC_get_current_cgroup_id:
1210 		return &bpf_get_current_cgroup_id_proto;
1211 #endif
1212 	case BPF_FUNC_send_signal:
1213 		return &bpf_send_signal_proto;
1214 	case BPF_FUNC_send_signal_thread:
1215 		return &bpf_send_signal_thread_proto;
1216 	case BPF_FUNC_perf_event_read_value:
1217 		return &bpf_perf_event_read_value_proto;
1218 	case BPF_FUNC_get_ns_current_pid_tgid:
1219 		return &bpf_get_ns_current_pid_tgid_proto;
1220 	case BPF_FUNC_ringbuf_output:
1221 		return &bpf_ringbuf_output_proto;
1222 	case BPF_FUNC_ringbuf_reserve:
1223 		return &bpf_ringbuf_reserve_proto;
1224 	case BPF_FUNC_ringbuf_submit:
1225 		return &bpf_ringbuf_submit_proto;
1226 	case BPF_FUNC_ringbuf_discard:
1227 		return &bpf_ringbuf_discard_proto;
1228 	case BPF_FUNC_ringbuf_query:
1229 		return &bpf_ringbuf_query_proto;
1230 	case BPF_FUNC_jiffies64:
1231 		return &bpf_jiffies64_proto;
1232 	case BPF_FUNC_get_task_stack:
1233 		return &bpf_get_task_stack_proto;
1234 	case BPF_FUNC_copy_from_user:
1235 		return prog->aux->sleepable ? &bpf_copy_from_user_proto : NULL;
1236 	default:
1237 		return NULL;
1238 	}
1239 }
1240 
1241 static const struct bpf_func_proto *
1242 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1243 {
1244 	switch (func_id) {
1245 	case BPF_FUNC_perf_event_output:
1246 		return &bpf_perf_event_output_proto;
1247 	case BPF_FUNC_get_stackid:
1248 		return &bpf_get_stackid_proto;
1249 	case BPF_FUNC_get_stack:
1250 		return &bpf_get_stack_proto;
1251 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1252 	case BPF_FUNC_override_return:
1253 		return &bpf_override_return_proto;
1254 #endif
1255 	default:
1256 		return bpf_tracing_func_proto(func_id, prog);
1257 	}
1258 }
1259 
1260 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1261 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1262 					const struct bpf_prog *prog,
1263 					struct bpf_insn_access_aux *info)
1264 {
1265 	if (off < 0 || off >= sizeof(struct pt_regs))
1266 		return false;
1267 	if (type != BPF_READ)
1268 		return false;
1269 	if (off % size != 0)
1270 		return false;
1271 	/*
1272 	 * Assertion for 32 bit to make sure last 8 byte access
1273 	 * (BPF_DW) to the last 4 byte member is disallowed.
1274 	 */
1275 	if (off + size > sizeof(struct pt_regs))
1276 		return false;
1277 
1278 	return true;
1279 }
1280 
1281 const struct bpf_verifier_ops kprobe_verifier_ops = {
1282 	.get_func_proto  = kprobe_prog_func_proto,
1283 	.is_valid_access = kprobe_prog_is_valid_access,
1284 };
1285 
1286 const struct bpf_prog_ops kprobe_prog_ops = {
1287 };
1288 
1289 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1290 	   u64, flags, void *, data, u64, size)
1291 {
1292 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1293 
1294 	/*
1295 	 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1296 	 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1297 	 * from there and call the same bpf_perf_event_output() helper inline.
1298 	 */
1299 	return ____bpf_perf_event_output(regs, map, flags, data, size);
1300 }
1301 
1302 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1303 	.func		= bpf_perf_event_output_tp,
1304 	.gpl_only	= true,
1305 	.ret_type	= RET_INTEGER,
1306 	.arg1_type	= ARG_PTR_TO_CTX,
1307 	.arg2_type	= ARG_CONST_MAP_PTR,
1308 	.arg3_type	= ARG_ANYTHING,
1309 	.arg4_type	= ARG_PTR_TO_MEM,
1310 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
1311 };
1312 
1313 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1314 	   u64, flags)
1315 {
1316 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1317 
1318 	/*
1319 	 * Same comment as in bpf_perf_event_output_tp(), only that this time
1320 	 * the other helper's function body cannot be inlined due to being
1321 	 * external, thus we need to call raw helper function.
1322 	 */
1323 	return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1324 			       flags, 0, 0);
1325 }
1326 
1327 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1328 	.func		= bpf_get_stackid_tp,
1329 	.gpl_only	= true,
1330 	.ret_type	= RET_INTEGER,
1331 	.arg1_type	= ARG_PTR_TO_CTX,
1332 	.arg2_type	= ARG_CONST_MAP_PTR,
1333 	.arg3_type	= ARG_ANYTHING,
1334 };
1335 
1336 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1337 	   u64, flags)
1338 {
1339 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1340 
1341 	return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1342 			     (unsigned long) size, flags, 0);
1343 }
1344 
1345 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1346 	.func		= bpf_get_stack_tp,
1347 	.gpl_only	= true,
1348 	.ret_type	= RET_INTEGER,
1349 	.arg1_type	= ARG_PTR_TO_CTX,
1350 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
1351 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
1352 	.arg4_type	= ARG_ANYTHING,
1353 };
1354 
1355 static const struct bpf_func_proto *
1356 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1357 {
1358 	switch (func_id) {
1359 	case BPF_FUNC_perf_event_output:
1360 		return &bpf_perf_event_output_proto_tp;
1361 	case BPF_FUNC_get_stackid:
1362 		return &bpf_get_stackid_proto_tp;
1363 	case BPF_FUNC_get_stack:
1364 		return &bpf_get_stack_proto_tp;
1365 	default:
1366 		return bpf_tracing_func_proto(func_id, prog);
1367 	}
1368 }
1369 
1370 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1371 				    const struct bpf_prog *prog,
1372 				    struct bpf_insn_access_aux *info)
1373 {
1374 	if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1375 		return false;
1376 	if (type != BPF_READ)
1377 		return false;
1378 	if (off % size != 0)
1379 		return false;
1380 
1381 	BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1382 	return true;
1383 }
1384 
1385 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1386 	.get_func_proto  = tp_prog_func_proto,
1387 	.is_valid_access = tp_prog_is_valid_access,
1388 };
1389 
1390 const struct bpf_prog_ops tracepoint_prog_ops = {
1391 };
1392 
1393 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1394 	   struct bpf_perf_event_value *, buf, u32, size)
1395 {
1396 	int err = -EINVAL;
1397 
1398 	if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1399 		goto clear;
1400 	err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1401 				    &buf->running);
1402 	if (unlikely(err))
1403 		goto clear;
1404 	return 0;
1405 clear:
1406 	memset(buf, 0, size);
1407 	return err;
1408 }
1409 
1410 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1411          .func           = bpf_perf_prog_read_value,
1412          .gpl_only       = true,
1413          .ret_type       = RET_INTEGER,
1414          .arg1_type      = ARG_PTR_TO_CTX,
1415          .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1416          .arg3_type      = ARG_CONST_SIZE,
1417 };
1418 
1419 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1420 	   void *, buf, u32, size, u64, flags)
1421 {
1422 #ifndef CONFIG_X86
1423 	return -ENOENT;
1424 #else
1425 	static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1426 	struct perf_branch_stack *br_stack = ctx->data->br_stack;
1427 	u32 to_copy;
1428 
1429 	if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1430 		return -EINVAL;
1431 
1432 	if (unlikely(!br_stack))
1433 		return -EINVAL;
1434 
1435 	if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1436 		return br_stack->nr * br_entry_size;
1437 
1438 	if (!buf || (size % br_entry_size != 0))
1439 		return -EINVAL;
1440 
1441 	to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1442 	memcpy(buf, br_stack->entries, to_copy);
1443 
1444 	return to_copy;
1445 #endif
1446 }
1447 
1448 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1449 	.func           = bpf_read_branch_records,
1450 	.gpl_only       = true,
1451 	.ret_type       = RET_INTEGER,
1452 	.arg1_type      = ARG_PTR_TO_CTX,
1453 	.arg2_type      = ARG_PTR_TO_MEM_OR_NULL,
1454 	.arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1455 	.arg4_type      = ARG_ANYTHING,
1456 };
1457 
1458 static const struct bpf_func_proto *
1459 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1460 {
1461 	switch (func_id) {
1462 	case BPF_FUNC_perf_event_output:
1463 		return &bpf_perf_event_output_proto_tp;
1464 	case BPF_FUNC_get_stackid:
1465 		return &bpf_get_stackid_proto_pe;
1466 	case BPF_FUNC_get_stack:
1467 		return &bpf_get_stack_proto_pe;
1468 	case BPF_FUNC_perf_prog_read_value:
1469 		return &bpf_perf_prog_read_value_proto;
1470 	case BPF_FUNC_read_branch_records:
1471 		return &bpf_read_branch_records_proto;
1472 	default:
1473 		return bpf_tracing_func_proto(func_id, prog);
1474 	}
1475 }
1476 
1477 /*
1478  * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1479  * to avoid potential recursive reuse issue when/if tracepoints are added
1480  * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1481  *
1482  * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1483  * in normal, irq, and nmi context.
1484  */
1485 struct bpf_raw_tp_regs {
1486 	struct pt_regs regs[3];
1487 };
1488 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1489 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1490 static struct pt_regs *get_bpf_raw_tp_regs(void)
1491 {
1492 	struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1493 	int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1494 
1495 	if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1496 		this_cpu_dec(bpf_raw_tp_nest_level);
1497 		return ERR_PTR(-EBUSY);
1498 	}
1499 
1500 	return &tp_regs->regs[nest_level - 1];
1501 }
1502 
1503 static void put_bpf_raw_tp_regs(void)
1504 {
1505 	this_cpu_dec(bpf_raw_tp_nest_level);
1506 }
1507 
1508 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1509 	   struct bpf_map *, map, u64, flags, void *, data, u64, size)
1510 {
1511 	struct pt_regs *regs = get_bpf_raw_tp_regs();
1512 	int ret;
1513 
1514 	if (IS_ERR(regs))
1515 		return PTR_ERR(regs);
1516 
1517 	perf_fetch_caller_regs(regs);
1518 	ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1519 
1520 	put_bpf_raw_tp_regs();
1521 	return ret;
1522 }
1523 
1524 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1525 	.func		= bpf_perf_event_output_raw_tp,
1526 	.gpl_only	= true,
1527 	.ret_type	= RET_INTEGER,
1528 	.arg1_type	= ARG_PTR_TO_CTX,
1529 	.arg2_type	= ARG_CONST_MAP_PTR,
1530 	.arg3_type	= ARG_ANYTHING,
1531 	.arg4_type	= ARG_PTR_TO_MEM,
1532 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
1533 };
1534 
1535 extern const struct bpf_func_proto bpf_skb_output_proto;
1536 extern const struct bpf_func_proto bpf_xdp_output_proto;
1537 
1538 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1539 	   struct bpf_map *, map, u64, flags)
1540 {
1541 	struct pt_regs *regs = get_bpf_raw_tp_regs();
1542 	int ret;
1543 
1544 	if (IS_ERR(regs))
1545 		return PTR_ERR(regs);
1546 
1547 	perf_fetch_caller_regs(regs);
1548 	/* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1549 	ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1550 			      flags, 0, 0);
1551 	put_bpf_raw_tp_regs();
1552 	return ret;
1553 }
1554 
1555 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1556 	.func		= bpf_get_stackid_raw_tp,
1557 	.gpl_only	= true,
1558 	.ret_type	= RET_INTEGER,
1559 	.arg1_type	= ARG_PTR_TO_CTX,
1560 	.arg2_type	= ARG_CONST_MAP_PTR,
1561 	.arg3_type	= ARG_ANYTHING,
1562 };
1563 
1564 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1565 	   void *, buf, u32, size, u64, flags)
1566 {
1567 	struct pt_regs *regs = get_bpf_raw_tp_regs();
1568 	int ret;
1569 
1570 	if (IS_ERR(regs))
1571 		return PTR_ERR(regs);
1572 
1573 	perf_fetch_caller_regs(regs);
1574 	ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1575 			    (unsigned long) size, flags, 0);
1576 	put_bpf_raw_tp_regs();
1577 	return ret;
1578 }
1579 
1580 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1581 	.func		= bpf_get_stack_raw_tp,
1582 	.gpl_only	= true,
1583 	.ret_type	= RET_INTEGER,
1584 	.arg1_type	= ARG_PTR_TO_CTX,
1585 	.arg2_type	= ARG_PTR_TO_MEM,
1586 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
1587 	.arg4_type	= ARG_ANYTHING,
1588 };
1589 
1590 static const struct bpf_func_proto *
1591 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1592 {
1593 	switch (func_id) {
1594 	case BPF_FUNC_perf_event_output:
1595 		return &bpf_perf_event_output_proto_raw_tp;
1596 	case BPF_FUNC_get_stackid:
1597 		return &bpf_get_stackid_proto_raw_tp;
1598 	case BPF_FUNC_get_stack:
1599 		return &bpf_get_stack_proto_raw_tp;
1600 	default:
1601 		return bpf_tracing_func_proto(func_id, prog);
1602 	}
1603 }
1604 
1605 const struct bpf_func_proto *
1606 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1607 {
1608 	switch (func_id) {
1609 #ifdef CONFIG_NET
1610 	case BPF_FUNC_skb_output:
1611 		return &bpf_skb_output_proto;
1612 	case BPF_FUNC_xdp_output:
1613 		return &bpf_xdp_output_proto;
1614 	case BPF_FUNC_skc_to_tcp6_sock:
1615 		return &bpf_skc_to_tcp6_sock_proto;
1616 	case BPF_FUNC_skc_to_tcp_sock:
1617 		return &bpf_skc_to_tcp_sock_proto;
1618 	case BPF_FUNC_skc_to_tcp_timewait_sock:
1619 		return &bpf_skc_to_tcp_timewait_sock_proto;
1620 	case BPF_FUNC_skc_to_tcp_request_sock:
1621 		return &bpf_skc_to_tcp_request_sock_proto;
1622 	case BPF_FUNC_skc_to_udp6_sock:
1623 		return &bpf_skc_to_udp6_sock_proto;
1624 #endif
1625 	case BPF_FUNC_seq_printf:
1626 		return prog->expected_attach_type == BPF_TRACE_ITER ?
1627 		       &bpf_seq_printf_proto :
1628 		       NULL;
1629 	case BPF_FUNC_seq_write:
1630 		return prog->expected_attach_type == BPF_TRACE_ITER ?
1631 		       &bpf_seq_write_proto :
1632 		       NULL;
1633 	case BPF_FUNC_d_path:
1634 		return &bpf_d_path_proto;
1635 	default:
1636 		return raw_tp_prog_func_proto(func_id, prog);
1637 	}
1638 }
1639 
1640 static bool raw_tp_prog_is_valid_access(int off, int size,
1641 					enum bpf_access_type type,
1642 					const struct bpf_prog *prog,
1643 					struct bpf_insn_access_aux *info)
1644 {
1645 	if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1646 		return false;
1647 	if (type != BPF_READ)
1648 		return false;
1649 	if (off % size != 0)
1650 		return false;
1651 	return true;
1652 }
1653 
1654 static bool tracing_prog_is_valid_access(int off, int size,
1655 					 enum bpf_access_type type,
1656 					 const struct bpf_prog *prog,
1657 					 struct bpf_insn_access_aux *info)
1658 {
1659 	if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1660 		return false;
1661 	if (type != BPF_READ)
1662 		return false;
1663 	if (off % size != 0)
1664 		return false;
1665 	return btf_ctx_access(off, size, type, prog, info);
1666 }
1667 
1668 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1669 				     const union bpf_attr *kattr,
1670 				     union bpf_attr __user *uattr)
1671 {
1672 	return -ENOTSUPP;
1673 }
1674 
1675 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1676 	.get_func_proto  = raw_tp_prog_func_proto,
1677 	.is_valid_access = raw_tp_prog_is_valid_access,
1678 };
1679 
1680 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1681 };
1682 
1683 const struct bpf_verifier_ops tracing_verifier_ops = {
1684 	.get_func_proto  = tracing_prog_func_proto,
1685 	.is_valid_access = tracing_prog_is_valid_access,
1686 };
1687 
1688 const struct bpf_prog_ops tracing_prog_ops = {
1689 	.test_run = bpf_prog_test_run_tracing,
1690 };
1691 
1692 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1693 						 enum bpf_access_type type,
1694 						 const struct bpf_prog *prog,
1695 						 struct bpf_insn_access_aux *info)
1696 {
1697 	if (off == 0) {
1698 		if (size != sizeof(u64) || type != BPF_READ)
1699 			return false;
1700 		info->reg_type = PTR_TO_TP_BUFFER;
1701 	}
1702 	return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1703 }
1704 
1705 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1706 	.get_func_proto  = raw_tp_prog_func_proto,
1707 	.is_valid_access = raw_tp_writable_prog_is_valid_access,
1708 };
1709 
1710 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1711 };
1712 
1713 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1714 				    const struct bpf_prog *prog,
1715 				    struct bpf_insn_access_aux *info)
1716 {
1717 	const int size_u64 = sizeof(u64);
1718 
1719 	if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1720 		return false;
1721 	if (type != BPF_READ)
1722 		return false;
1723 	if (off % size != 0) {
1724 		if (sizeof(unsigned long) != 4)
1725 			return false;
1726 		if (size != 8)
1727 			return false;
1728 		if (off % size != 4)
1729 			return false;
1730 	}
1731 
1732 	switch (off) {
1733 	case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1734 		bpf_ctx_record_field_size(info, size_u64);
1735 		if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1736 			return false;
1737 		break;
1738 	case bpf_ctx_range(struct bpf_perf_event_data, addr):
1739 		bpf_ctx_record_field_size(info, size_u64);
1740 		if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1741 			return false;
1742 		break;
1743 	default:
1744 		if (size != sizeof(long))
1745 			return false;
1746 	}
1747 
1748 	return true;
1749 }
1750 
1751 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1752 				      const struct bpf_insn *si,
1753 				      struct bpf_insn *insn_buf,
1754 				      struct bpf_prog *prog, u32 *target_size)
1755 {
1756 	struct bpf_insn *insn = insn_buf;
1757 
1758 	switch (si->off) {
1759 	case offsetof(struct bpf_perf_event_data, sample_period):
1760 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1761 						       data), si->dst_reg, si->src_reg,
1762 				      offsetof(struct bpf_perf_event_data_kern, data));
1763 		*insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1764 				      bpf_target_off(struct perf_sample_data, period, 8,
1765 						     target_size));
1766 		break;
1767 	case offsetof(struct bpf_perf_event_data, addr):
1768 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1769 						       data), si->dst_reg, si->src_reg,
1770 				      offsetof(struct bpf_perf_event_data_kern, data));
1771 		*insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1772 				      bpf_target_off(struct perf_sample_data, addr, 8,
1773 						     target_size));
1774 		break;
1775 	default:
1776 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1777 						       regs), si->dst_reg, si->src_reg,
1778 				      offsetof(struct bpf_perf_event_data_kern, regs));
1779 		*insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1780 				      si->off);
1781 		break;
1782 	}
1783 
1784 	return insn - insn_buf;
1785 }
1786 
1787 const struct bpf_verifier_ops perf_event_verifier_ops = {
1788 	.get_func_proto		= pe_prog_func_proto,
1789 	.is_valid_access	= pe_prog_is_valid_access,
1790 	.convert_ctx_access	= pe_prog_convert_ctx_access,
1791 };
1792 
1793 const struct bpf_prog_ops perf_event_prog_ops = {
1794 };
1795 
1796 static DEFINE_MUTEX(bpf_event_mutex);
1797 
1798 #define BPF_TRACE_MAX_PROGS 64
1799 
1800 int perf_event_attach_bpf_prog(struct perf_event *event,
1801 			       struct bpf_prog *prog)
1802 {
1803 	struct bpf_prog_array *old_array;
1804 	struct bpf_prog_array *new_array;
1805 	int ret = -EEXIST;
1806 
1807 	/*
1808 	 * Kprobe override only works if they are on the function entry,
1809 	 * and only if they are on the opt-in list.
1810 	 */
1811 	if (prog->kprobe_override &&
1812 	    (!trace_kprobe_on_func_entry(event->tp_event) ||
1813 	     !trace_kprobe_error_injectable(event->tp_event)))
1814 		return -EINVAL;
1815 
1816 	mutex_lock(&bpf_event_mutex);
1817 
1818 	if (event->prog)
1819 		goto unlock;
1820 
1821 	old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1822 	if (old_array &&
1823 	    bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1824 		ret = -E2BIG;
1825 		goto unlock;
1826 	}
1827 
1828 	ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1829 	if (ret < 0)
1830 		goto unlock;
1831 
1832 	/* set the new array to event->tp_event and set event->prog */
1833 	event->prog = prog;
1834 	rcu_assign_pointer(event->tp_event->prog_array, new_array);
1835 	bpf_prog_array_free(old_array);
1836 
1837 unlock:
1838 	mutex_unlock(&bpf_event_mutex);
1839 	return ret;
1840 }
1841 
1842 void perf_event_detach_bpf_prog(struct perf_event *event)
1843 {
1844 	struct bpf_prog_array *old_array;
1845 	struct bpf_prog_array *new_array;
1846 	int ret;
1847 
1848 	mutex_lock(&bpf_event_mutex);
1849 
1850 	if (!event->prog)
1851 		goto unlock;
1852 
1853 	old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1854 	ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1855 	if (ret == -ENOENT)
1856 		goto unlock;
1857 	if (ret < 0) {
1858 		bpf_prog_array_delete_safe(old_array, event->prog);
1859 	} else {
1860 		rcu_assign_pointer(event->tp_event->prog_array, new_array);
1861 		bpf_prog_array_free(old_array);
1862 	}
1863 
1864 	bpf_prog_put(event->prog);
1865 	event->prog = NULL;
1866 
1867 unlock:
1868 	mutex_unlock(&bpf_event_mutex);
1869 }
1870 
1871 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1872 {
1873 	struct perf_event_query_bpf __user *uquery = info;
1874 	struct perf_event_query_bpf query = {};
1875 	struct bpf_prog_array *progs;
1876 	u32 *ids, prog_cnt, ids_len;
1877 	int ret;
1878 
1879 	if (!perfmon_capable())
1880 		return -EPERM;
1881 	if (event->attr.type != PERF_TYPE_TRACEPOINT)
1882 		return -EINVAL;
1883 	if (copy_from_user(&query, uquery, sizeof(query)))
1884 		return -EFAULT;
1885 
1886 	ids_len = query.ids_len;
1887 	if (ids_len > BPF_TRACE_MAX_PROGS)
1888 		return -E2BIG;
1889 	ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1890 	if (!ids)
1891 		return -ENOMEM;
1892 	/*
1893 	 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1894 	 * is required when user only wants to check for uquery->prog_cnt.
1895 	 * There is no need to check for it since the case is handled
1896 	 * gracefully in bpf_prog_array_copy_info.
1897 	 */
1898 
1899 	mutex_lock(&bpf_event_mutex);
1900 	progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
1901 	ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
1902 	mutex_unlock(&bpf_event_mutex);
1903 
1904 	if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1905 	    copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1906 		ret = -EFAULT;
1907 
1908 	kfree(ids);
1909 	return ret;
1910 }
1911 
1912 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1913 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1914 
1915 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1916 {
1917 	struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1918 
1919 	for (; btp < __stop__bpf_raw_tp; btp++) {
1920 		if (!strcmp(btp->tp->name, name))
1921 			return btp;
1922 	}
1923 
1924 	return bpf_get_raw_tracepoint_module(name);
1925 }
1926 
1927 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1928 {
1929 	struct module *mod = __module_address((unsigned long)btp);
1930 
1931 	if (mod)
1932 		module_put(mod);
1933 }
1934 
1935 static __always_inline
1936 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1937 {
1938 	cant_sleep();
1939 	rcu_read_lock();
1940 	(void) BPF_PROG_RUN(prog, args);
1941 	rcu_read_unlock();
1942 }
1943 
1944 #define UNPACK(...)			__VA_ARGS__
1945 #define REPEAT_1(FN, DL, X, ...)	FN(X)
1946 #define REPEAT_2(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1947 #define REPEAT_3(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1948 #define REPEAT_4(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1949 #define REPEAT_5(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1950 #define REPEAT_6(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1951 #define REPEAT_7(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1952 #define REPEAT_8(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1953 #define REPEAT_9(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1954 #define REPEAT_10(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1955 #define REPEAT_11(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1956 #define REPEAT_12(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1957 #define REPEAT(X, FN, DL, ...)		REPEAT_##X(FN, DL, __VA_ARGS__)
1958 
1959 #define SARG(X)		u64 arg##X
1960 #define COPY(X)		args[X] = arg##X
1961 
1962 #define __DL_COM	(,)
1963 #define __DL_SEM	(;)
1964 
1965 #define __SEQ_0_11	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1966 
1967 #define BPF_TRACE_DEFN_x(x)						\
1968 	void bpf_trace_run##x(struct bpf_prog *prog,			\
1969 			      REPEAT(x, SARG, __DL_COM, __SEQ_0_11))	\
1970 	{								\
1971 		u64 args[x];						\
1972 		REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);			\
1973 		__bpf_trace_run(prog, args);				\
1974 	}								\
1975 	EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1976 BPF_TRACE_DEFN_x(1);
1977 BPF_TRACE_DEFN_x(2);
1978 BPF_TRACE_DEFN_x(3);
1979 BPF_TRACE_DEFN_x(4);
1980 BPF_TRACE_DEFN_x(5);
1981 BPF_TRACE_DEFN_x(6);
1982 BPF_TRACE_DEFN_x(7);
1983 BPF_TRACE_DEFN_x(8);
1984 BPF_TRACE_DEFN_x(9);
1985 BPF_TRACE_DEFN_x(10);
1986 BPF_TRACE_DEFN_x(11);
1987 BPF_TRACE_DEFN_x(12);
1988 
1989 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1990 {
1991 	struct tracepoint *tp = btp->tp;
1992 
1993 	/*
1994 	 * check that program doesn't access arguments beyond what's
1995 	 * available in this tracepoint
1996 	 */
1997 	if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1998 		return -EINVAL;
1999 
2000 	if (prog->aux->max_tp_access > btp->writable_size)
2001 		return -EINVAL;
2002 
2003 	return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
2004 }
2005 
2006 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2007 {
2008 	return __bpf_probe_register(btp, prog);
2009 }
2010 
2011 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2012 {
2013 	return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
2014 }
2015 
2016 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
2017 			    u32 *fd_type, const char **buf,
2018 			    u64 *probe_offset, u64 *probe_addr)
2019 {
2020 	bool is_tracepoint, is_syscall_tp;
2021 	struct bpf_prog *prog;
2022 	int flags, err = 0;
2023 
2024 	prog = event->prog;
2025 	if (!prog)
2026 		return -ENOENT;
2027 
2028 	/* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
2029 	if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
2030 		return -EOPNOTSUPP;
2031 
2032 	*prog_id = prog->aux->id;
2033 	flags = event->tp_event->flags;
2034 	is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
2035 	is_syscall_tp = is_syscall_trace_event(event->tp_event);
2036 
2037 	if (is_tracepoint || is_syscall_tp) {
2038 		*buf = is_tracepoint ? event->tp_event->tp->name
2039 				     : event->tp_event->name;
2040 		*fd_type = BPF_FD_TYPE_TRACEPOINT;
2041 		*probe_offset = 0x0;
2042 		*probe_addr = 0x0;
2043 	} else {
2044 		/* kprobe/uprobe */
2045 		err = -EOPNOTSUPP;
2046 #ifdef CONFIG_KPROBE_EVENTS
2047 		if (flags & TRACE_EVENT_FL_KPROBE)
2048 			err = bpf_get_kprobe_info(event, fd_type, buf,
2049 						  probe_offset, probe_addr,
2050 						  event->attr.type == PERF_TYPE_TRACEPOINT);
2051 #endif
2052 #ifdef CONFIG_UPROBE_EVENTS
2053 		if (flags & TRACE_EVENT_FL_UPROBE)
2054 			err = bpf_get_uprobe_info(event, fd_type, buf,
2055 						  probe_offset,
2056 						  event->attr.type == PERF_TYPE_TRACEPOINT);
2057 #endif
2058 	}
2059 
2060 	return err;
2061 }
2062 
2063 static int __init send_signal_irq_work_init(void)
2064 {
2065 	int cpu;
2066 	struct send_signal_irq_work *work;
2067 
2068 	for_each_possible_cpu(cpu) {
2069 		work = per_cpu_ptr(&send_signal_work, cpu);
2070 		init_irq_work(&work->irq_work, do_bpf_send_signal);
2071 	}
2072 	return 0;
2073 }
2074 
2075 subsys_initcall(send_signal_irq_work_init);
2076 
2077 #ifdef CONFIG_MODULES
2078 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
2079 			    void *module)
2080 {
2081 	struct bpf_trace_module *btm, *tmp;
2082 	struct module *mod = module;
2083 
2084 	if (mod->num_bpf_raw_events == 0 ||
2085 	    (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
2086 		return 0;
2087 
2088 	mutex_lock(&bpf_module_mutex);
2089 
2090 	switch (op) {
2091 	case MODULE_STATE_COMING:
2092 		btm = kzalloc(sizeof(*btm), GFP_KERNEL);
2093 		if (btm) {
2094 			btm->module = module;
2095 			list_add(&btm->list, &bpf_trace_modules);
2096 		}
2097 		break;
2098 	case MODULE_STATE_GOING:
2099 		list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
2100 			if (btm->module == module) {
2101 				list_del(&btm->list);
2102 				kfree(btm);
2103 				break;
2104 			}
2105 		}
2106 		break;
2107 	}
2108 
2109 	mutex_unlock(&bpf_module_mutex);
2110 
2111 	return 0;
2112 }
2113 
2114 static struct notifier_block bpf_module_nb = {
2115 	.notifier_call = bpf_event_notify,
2116 };
2117 
2118 static int __init bpf_event_init(void)
2119 {
2120 	register_module_notifier(&bpf_module_nb);
2121 	return 0;
2122 }
2123 
2124 fs_initcall(bpf_event_init);
2125 #endif /* CONFIG_MODULES */
2126