xref: /linux/kernel/trace/bpf_trace.c (revision 8b4483658364f05b2e32845c8f445cdfd9452286)
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/syscalls.h>
15 #include <linux/error-injection.h>
16 
17 #include "trace_probe.h"
18 #include "trace.h"
19 
20 #ifdef CONFIG_MODULES
21 struct bpf_trace_module {
22 	struct module *module;
23 	struct list_head list;
24 };
25 
26 static LIST_HEAD(bpf_trace_modules);
27 static DEFINE_MUTEX(bpf_module_mutex);
28 
29 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
30 {
31 	struct bpf_raw_event_map *btp, *ret = NULL;
32 	struct bpf_trace_module *btm;
33 	unsigned int i;
34 
35 	mutex_lock(&bpf_module_mutex);
36 	list_for_each_entry(btm, &bpf_trace_modules, list) {
37 		for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
38 			btp = &btm->module->bpf_raw_events[i];
39 			if (!strcmp(btp->tp->name, name)) {
40 				if (try_module_get(btm->module))
41 					ret = btp;
42 				goto out;
43 			}
44 		}
45 	}
46 out:
47 	mutex_unlock(&bpf_module_mutex);
48 	return ret;
49 }
50 #else
51 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
52 {
53 	return NULL;
54 }
55 #endif /* CONFIG_MODULES */
56 
57 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
58 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
59 
60 /**
61  * trace_call_bpf - invoke BPF program
62  * @call: tracepoint event
63  * @ctx: opaque context pointer
64  *
65  * kprobe handlers execute BPF programs via this helper.
66  * Can be used from static tracepoints in the future.
67  *
68  * Return: BPF programs always return an integer which is interpreted by
69  * kprobe handler as:
70  * 0 - return from kprobe (event is filtered out)
71  * 1 - store kprobe event into ring buffer
72  * Other values are reserved and currently alias to 1
73  */
74 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
75 {
76 	unsigned int ret;
77 
78 	if (in_nmi()) /* not supported yet */
79 		return 1;
80 
81 	preempt_disable();
82 
83 	if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
84 		/*
85 		 * since some bpf program is already running on this cpu,
86 		 * don't call into another bpf program (same or different)
87 		 * and don't send kprobe event into ring-buffer,
88 		 * so return zero here
89 		 */
90 		ret = 0;
91 		goto out;
92 	}
93 
94 	/*
95 	 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
96 	 * to all call sites, we did a bpf_prog_array_valid() there to check
97 	 * whether call->prog_array is empty or not, which is
98 	 * a heurisitc to speed up execution.
99 	 *
100 	 * If bpf_prog_array_valid() fetched prog_array was
101 	 * non-NULL, we go into trace_call_bpf() and do the actual
102 	 * proper rcu_dereference() under RCU lock.
103 	 * If it turns out that prog_array is NULL then, we bail out.
104 	 * For the opposite, if the bpf_prog_array_valid() fetched pointer
105 	 * was NULL, you'll skip the prog_array with the risk of missing
106 	 * out of events when it was updated in between this and the
107 	 * rcu_dereference() which is accepted risk.
108 	 */
109 	ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
110 
111  out:
112 	__this_cpu_dec(bpf_prog_active);
113 	preempt_enable();
114 
115 	return ret;
116 }
117 EXPORT_SYMBOL_GPL(trace_call_bpf);
118 
119 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
120 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
121 {
122 	regs_set_return_value(regs, rc);
123 	override_function_with_return(regs);
124 	return 0;
125 }
126 
127 static const struct bpf_func_proto bpf_override_return_proto = {
128 	.func		= bpf_override_return,
129 	.gpl_only	= true,
130 	.ret_type	= RET_INTEGER,
131 	.arg1_type	= ARG_PTR_TO_CTX,
132 	.arg2_type	= ARG_ANYTHING,
133 };
134 #endif
135 
136 BPF_CALL_3(bpf_probe_read, void *, dst, u32, size, const void *, unsafe_ptr)
137 {
138 	int ret;
139 
140 	ret = probe_kernel_read(dst, unsafe_ptr, size);
141 	if (unlikely(ret < 0))
142 		memset(dst, 0, size);
143 
144 	return ret;
145 }
146 
147 static const struct bpf_func_proto bpf_probe_read_proto = {
148 	.func		= bpf_probe_read,
149 	.gpl_only	= true,
150 	.ret_type	= RET_INTEGER,
151 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
152 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
153 	.arg3_type	= ARG_ANYTHING,
154 };
155 
156 BPF_CALL_3(bpf_probe_write_user, void *, unsafe_ptr, const void *, src,
157 	   u32, size)
158 {
159 	/*
160 	 * Ensure we're in user context which is safe for the helper to
161 	 * run. This helper has no business in a kthread.
162 	 *
163 	 * access_ok() should prevent writing to non-user memory, but in
164 	 * some situations (nommu, temporary switch, etc) access_ok() does
165 	 * not provide enough validation, hence the check on KERNEL_DS.
166 	 */
167 
168 	if (unlikely(in_interrupt() ||
169 		     current->flags & (PF_KTHREAD | PF_EXITING)))
170 		return -EPERM;
171 	if (unlikely(uaccess_kernel()))
172 		return -EPERM;
173 	if (!access_ok(unsafe_ptr, size))
174 		return -EPERM;
175 
176 	return probe_kernel_write(unsafe_ptr, src, size);
177 }
178 
179 static const struct bpf_func_proto bpf_probe_write_user_proto = {
180 	.func		= bpf_probe_write_user,
181 	.gpl_only	= true,
182 	.ret_type	= RET_INTEGER,
183 	.arg1_type	= ARG_ANYTHING,
184 	.arg2_type	= ARG_PTR_TO_MEM,
185 	.arg3_type	= ARG_CONST_SIZE,
186 };
187 
188 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
189 {
190 	pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
191 			    current->comm, task_pid_nr(current));
192 
193 	return &bpf_probe_write_user_proto;
194 }
195 
196 /*
197  * Only limited trace_printk() conversion specifiers allowed:
198  * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %s
199  */
200 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
201 	   u64, arg2, u64, arg3)
202 {
203 	bool str_seen = false;
204 	int mod[3] = {};
205 	int fmt_cnt = 0;
206 	u64 unsafe_addr;
207 	char buf[64];
208 	int i;
209 
210 	/*
211 	 * bpf_check()->check_func_arg()->check_stack_boundary()
212 	 * guarantees that fmt points to bpf program stack,
213 	 * fmt_size bytes of it were initialized and fmt_size > 0
214 	 */
215 	if (fmt[--fmt_size] != 0)
216 		return -EINVAL;
217 
218 	/* check format string for allowed specifiers */
219 	for (i = 0; i < fmt_size; i++) {
220 		if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
221 			return -EINVAL;
222 
223 		if (fmt[i] != '%')
224 			continue;
225 
226 		if (fmt_cnt >= 3)
227 			return -EINVAL;
228 
229 		/* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
230 		i++;
231 		if (fmt[i] == 'l') {
232 			mod[fmt_cnt]++;
233 			i++;
234 		} else if (fmt[i] == 'p' || fmt[i] == 's') {
235 			mod[fmt_cnt]++;
236 			/* disallow any further format extensions */
237 			if (fmt[i + 1] != 0 &&
238 			    !isspace(fmt[i + 1]) &&
239 			    !ispunct(fmt[i + 1]))
240 				return -EINVAL;
241 			fmt_cnt++;
242 			if (fmt[i] == 's') {
243 				if (str_seen)
244 					/* allow only one '%s' per fmt string */
245 					return -EINVAL;
246 				str_seen = true;
247 
248 				switch (fmt_cnt) {
249 				case 1:
250 					unsafe_addr = arg1;
251 					arg1 = (long) buf;
252 					break;
253 				case 2:
254 					unsafe_addr = arg2;
255 					arg2 = (long) buf;
256 					break;
257 				case 3:
258 					unsafe_addr = arg3;
259 					arg3 = (long) buf;
260 					break;
261 				}
262 				buf[0] = 0;
263 				strncpy_from_unsafe(buf,
264 						    (void *) (long) unsafe_addr,
265 						    sizeof(buf));
266 			}
267 			continue;
268 		}
269 
270 		if (fmt[i] == 'l') {
271 			mod[fmt_cnt]++;
272 			i++;
273 		}
274 
275 		if (fmt[i] != 'i' && fmt[i] != 'd' &&
276 		    fmt[i] != 'u' && fmt[i] != 'x')
277 			return -EINVAL;
278 		fmt_cnt++;
279 	}
280 
281 /* Horrid workaround for getting va_list handling working with different
282  * argument type combinations generically for 32 and 64 bit archs.
283  */
284 #define __BPF_TP_EMIT()	__BPF_ARG3_TP()
285 #define __BPF_TP(...)							\
286 	__trace_printk(0 /* Fake ip */,					\
287 		       fmt, ##__VA_ARGS__)
288 
289 #define __BPF_ARG1_TP(...)						\
290 	((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64))	\
291 	  ? __BPF_TP(arg1, ##__VA_ARGS__)				\
292 	  : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32))	\
293 	      ? __BPF_TP((long)arg1, ##__VA_ARGS__)			\
294 	      : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
295 
296 #define __BPF_ARG2_TP(...)						\
297 	((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64))	\
298 	  ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__)				\
299 	  : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32))	\
300 	      ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__)		\
301 	      : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
302 
303 #define __BPF_ARG3_TP(...)						\
304 	((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64))	\
305 	  ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__)				\
306 	  : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32))	\
307 	      ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__)		\
308 	      : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
309 
310 	return __BPF_TP_EMIT();
311 }
312 
313 static const struct bpf_func_proto bpf_trace_printk_proto = {
314 	.func		= bpf_trace_printk,
315 	.gpl_only	= true,
316 	.ret_type	= RET_INTEGER,
317 	.arg1_type	= ARG_PTR_TO_MEM,
318 	.arg2_type	= ARG_CONST_SIZE,
319 };
320 
321 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
322 {
323 	/*
324 	 * this program might be calling bpf_trace_printk,
325 	 * so allocate per-cpu printk buffers
326 	 */
327 	trace_printk_init_buffers();
328 
329 	return &bpf_trace_printk_proto;
330 }
331 
332 static __always_inline int
333 get_map_perf_counter(struct bpf_map *map, u64 flags,
334 		     u64 *value, u64 *enabled, u64 *running)
335 {
336 	struct bpf_array *array = container_of(map, struct bpf_array, map);
337 	unsigned int cpu = smp_processor_id();
338 	u64 index = flags & BPF_F_INDEX_MASK;
339 	struct bpf_event_entry *ee;
340 
341 	if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
342 		return -EINVAL;
343 	if (index == BPF_F_CURRENT_CPU)
344 		index = cpu;
345 	if (unlikely(index >= array->map.max_entries))
346 		return -E2BIG;
347 
348 	ee = READ_ONCE(array->ptrs[index]);
349 	if (!ee)
350 		return -ENOENT;
351 
352 	return perf_event_read_local(ee->event, value, enabled, running);
353 }
354 
355 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
356 {
357 	u64 value = 0;
358 	int err;
359 
360 	err = get_map_perf_counter(map, flags, &value, NULL, NULL);
361 	/*
362 	 * this api is ugly since we miss [-22..-2] range of valid
363 	 * counter values, but that's uapi
364 	 */
365 	if (err)
366 		return err;
367 	return value;
368 }
369 
370 static const struct bpf_func_proto bpf_perf_event_read_proto = {
371 	.func		= bpf_perf_event_read,
372 	.gpl_only	= true,
373 	.ret_type	= RET_INTEGER,
374 	.arg1_type	= ARG_CONST_MAP_PTR,
375 	.arg2_type	= ARG_ANYTHING,
376 };
377 
378 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
379 	   struct bpf_perf_event_value *, buf, u32, size)
380 {
381 	int err = -EINVAL;
382 
383 	if (unlikely(size != sizeof(struct bpf_perf_event_value)))
384 		goto clear;
385 	err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
386 				   &buf->running);
387 	if (unlikely(err))
388 		goto clear;
389 	return 0;
390 clear:
391 	memset(buf, 0, size);
392 	return err;
393 }
394 
395 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
396 	.func		= bpf_perf_event_read_value,
397 	.gpl_only	= true,
398 	.ret_type	= RET_INTEGER,
399 	.arg1_type	= ARG_CONST_MAP_PTR,
400 	.arg2_type	= ARG_ANYTHING,
401 	.arg3_type	= ARG_PTR_TO_UNINIT_MEM,
402 	.arg4_type	= ARG_CONST_SIZE,
403 };
404 
405 static DEFINE_PER_CPU(struct perf_sample_data, bpf_trace_sd);
406 
407 static __always_inline u64
408 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
409 			u64 flags, struct perf_sample_data *sd)
410 {
411 	struct bpf_array *array = container_of(map, struct bpf_array, map);
412 	unsigned int cpu = smp_processor_id();
413 	u64 index = flags & BPF_F_INDEX_MASK;
414 	struct bpf_event_entry *ee;
415 	struct perf_event *event;
416 
417 	if (index == BPF_F_CURRENT_CPU)
418 		index = cpu;
419 	if (unlikely(index >= array->map.max_entries))
420 		return -E2BIG;
421 
422 	ee = READ_ONCE(array->ptrs[index]);
423 	if (!ee)
424 		return -ENOENT;
425 
426 	event = ee->event;
427 	if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
428 		     event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
429 		return -EINVAL;
430 
431 	if (unlikely(event->oncpu != cpu))
432 		return -EOPNOTSUPP;
433 
434 	return perf_event_output(event, sd, regs);
435 }
436 
437 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
438 	   u64, flags, void *, data, u64, size)
439 {
440 	struct perf_sample_data *sd = this_cpu_ptr(&bpf_trace_sd);
441 	struct perf_raw_record raw = {
442 		.frag = {
443 			.size = size,
444 			.data = data,
445 		},
446 	};
447 
448 	if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
449 		return -EINVAL;
450 
451 	perf_sample_data_init(sd, 0, 0);
452 	sd->raw = &raw;
453 
454 	return __bpf_perf_event_output(regs, map, flags, sd);
455 }
456 
457 static const struct bpf_func_proto bpf_perf_event_output_proto = {
458 	.func		= bpf_perf_event_output,
459 	.gpl_only	= true,
460 	.ret_type	= RET_INTEGER,
461 	.arg1_type	= ARG_PTR_TO_CTX,
462 	.arg2_type	= ARG_CONST_MAP_PTR,
463 	.arg3_type	= ARG_ANYTHING,
464 	.arg4_type	= ARG_PTR_TO_MEM,
465 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
466 };
467 
468 static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs);
469 static DEFINE_PER_CPU(struct perf_sample_data, bpf_misc_sd);
470 
471 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
472 		     void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
473 {
474 	struct perf_sample_data *sd = this_cpu_ptr(&bpf_misc_sd);
475 	struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs);
476 	struct perf_raw_frag frag = {
477 		.copy		= ctx_copy,
478 		.size		= ctx_size,
479 		.data		= ctx,
480 	};
481 	struct perf_raw_record raw = {
482 		.frag = {
483 			{
484 				.next	= ctx_size ? &frag : NULL,
485 			},
486 			.size	= meta_size,
487 			.data	= meta,
488 		},
489 	};
490 
491 	perf_fetch_caller_regs(regs);
492 	perf_sample_data_init(sd, 0, 0);
493 	sd->raw = &raw;
494 
495 	return __bpf_perf_event_output(regs, map, flags, sd);
496 }
497 
498 BPF_CALL_0(bpf_get_current_task)
499 {
500 	return (long) current;
501 }
502 
503 static const struct bpf_func_proto bpf_get_current_task_proto = {
504 	.func		= bpf_get_current_task,
505 	.gpl_only	= true,
506 	.ret_type	= RET_INTEGER,
507 };
508 
509 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
510 {
511 	struct bpf_array *array = container_of(map, struct bpf_array, map);
512 	struct cgroup *cgrp;
513 
514 	if (unlikely(idx >= array->map.max_entries))
515 		return -E2BIG;
516 
517 	cgrp = READ_ONCE(array->ptrs[idx]);
518 	if (unlikely(!cgrp))
519 		return -EAGAIN;
520 
521 	return task_under_cgroup_hierarchy(current, cgrp);
522 }
523 
524 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
525 	.func           = bpf_current_task_under_cgroup,
526 	.gpl_only       = false,
527 	.ret_type       = RET_INTEGER,
528 	.arg1_type      = ARG_CONST_MAP_PTR,
529 	.arg2_type      = ARG_ANYTHING,
530 };
531 
532 BPF_CALL_3(bpf_probe_read_str, void *, dst, u32, size,
533 	   const void *, unsafe_ptr)
534 {
535 	int ret;
536 
537 	/*
538 	 * The strncpy_from_unsafe() call will likely not fill the entire
539 	 * buffer, but that's okay in this circumstance as we're probing
540 	 * arbitrary memory anyway similar to bpf_probe_read() and might
541 	 * as well probe the stack. Thus, memory is explicitly cleared
542 	 * only in error case, so that improper users ignoring return
543 	 * code altogether don't copy garbage; otherwise length of string
544 	 * is returned that can be used for bpf_perf_event_output() et al.
545 	 */
546 	ret = strncpy_from_unsafe(dst, unsafe_ptr, size);
547 	if (unlikely(ret < 0))
548 		memset(dst, 0, size);
549 
550 	return ret;
551 }
552 
553 static const struct bpf_func_proto bpf_probe_read_str_proto = {
554 	.func		= bpf_probe_read_str,
555 	.gpl_only	= true,
556 	.ret_type	= RET_INTEGER,
557 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
558 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
559 	.arg3_type	= ARG_ANYTHING,
560 };
561 
562 static const struct bpf_func_proto *
563 tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
564 {
565 	switch (func_id) {
566 	case BPF_FUNC_map_lookup_elem:
567 		return &bpf_map_lookup_elem_proto;
568 	case BPF_FUNC_map_update_elem:
569 		return &bpf_map_update_elem_proto;
570 	case BPF_FUNC_map_delete_elem:
571 		return &bpf_map_delete_elem_proto;
572 	case BPF_FUNC_map_push_elem:
573 		return &bpf_map_push_elem_proto;
574 	case BPF_FUNC_map_pop_elem:
575 		return &bpf_map_pop_elem_proto;
576 	case BPF_FUNC_map_peek_elem:
577 		return &bpf_map_peek_elem_proto;
578 	case BPF_FUNC_probe_read:
579 		return &bpf_probe_read_proto;
580 	case BPF_FUNC_ktime_get_ns:
581 		return &bpf_ktime_get_ns_proto;
582 	case BPF_FUNC_tail_call:
583 		return &bpf_tail_call_proto;
584 	case BPF_FUNC_get_current_pid_tgid:
585 		return &bpf_get_current_pid_tgid_proto;
586 	case BPF_FUNC_get_current_task:
587 		return &bpf_get_current_task_proto;
588 	case BPF_FUNC_get_current_uid_gid:
589 		return &bpf_get_current_uid_gid_proto;
590 	case BPF_FUNC_get_current_comm:
591 		return &bpf_get_current_comm_proto;
592 	case BPF_FUNC_trace_printk:
593 		return bpf_get_trace_printk_proto();
594 	case BPF_FUNC_get_smp_processor_id:
595 		return &bpf_get_smp_processor_id_proto;
596 	case BPF_FUNC_get_numa_node_id:
597 		return &bpf_get_numa_node_id_proto;
598 	case BPF_FUNC_perf_event_read:
599 		return &bpf_perf_event_read_proto;
600 	case BPF_FUNC_probe_write_user:
601 		return bpf_get_probe_write_proto();
602 	case BPF_FUNC_current_task_under_cgroup:
603 		return &bpf_current_task_under_cgroup_proto;
604 	case BPF_FUNC_get_prandom_u32:
605 		return &bpf_get_prandom_u32_proto;
606 	case BPF_FUNC_probe_read_str:
607 		return &bpf_probe_read_str_proto;
608 #ifdef CONFIG_CGROUPS
609 	case BPF_FUNC_get_current_cgroup_id:
610 		return &bpf_get_current_cgroup_id_proto;
611 #endif
612 	default:
613 		return NULL;
614 	}
615 }
616 
617 static const struct bpf_func_proto *
618 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
619 {
620 	switch (func_id) {
621 	case BPF_FUNC_perf_event_output:
622 		return &bpf_perf_event_output_proto;
623 	case BPF_FUNC_get_stackid:
624 		return &bpf_get_stackid_proto;
625 	case BPF_FUNC_get_stack:
626 		return &bpf_get_stack_proto;
627 	case BPF_FUNC_perf_event_read_value:
628 		return &bpf_perf_event_read_value_proto;
629 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
630 	case BPF_FUNC_override_return:
631 		return &bpf_override_return_proto;
632 #endif
633 	default:
634 		return tracing_func_proto(func_id, prog);
635 	}
636 }
637 
638 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
639 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
640 					const struct bpf_prog *prog,
641 					struct bpf_insn_access_aux *info)
642 {
643 	if (off < 0 || off >= sizeof(struct pt_regs))
644 		return false;
645 	if (type != BPF_READ)
646 		return false;
647 	if (off % size != 0)
648 		return false;
649 	/*
650 	 * Assertion for 32 bit to make sure last 8 byte access
651 	 * (BPF_DW) to the last 4 byte member is disallowed.
652 	 */
653 	if (off + size > sizeof(struct pt_regs))
654 		return false;
655 
656 	return true;
657 }
658 
659 const struct bpf_verifier_ops kprobe_verifier_ops = {
660 	.get_func_proto  = kprobe_prog_func_proto,
661 	.is_valid_access = kprobe_prog_is_valid_access,
662 };
663 
664 const struct bpf_prog_ops kprobe_prog_ops = {
665 };
666 
667 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
668 	   u64, flags, void *, data, u64, size)
669 {
670 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
671 
672 	/*
673 	 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
674 	 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
675 	 * from there and call the same bpf_perf_event_output() helper inline.
676 	 */
677 	return ____bpf_perf_event_output(regs, map, flags, data, size);
678 }
679 
680 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
681 	.func		= bpf_perf_event_output_tp,
682 	.gpl_only	= true,
683 	.ret_type	= RET_INTEGER,
684 	.arg1_type	= ARG_PTR_TO_CTX,
685 	.arg2_type	= ARG_CONST_MAP_PTR,
686 	.arg3_type	= ARG_ANYTHING,
687 	.arg4_type	= ARG_PTR_TO_MEM,
688 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
689 };
690 
691 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
692 	   u64, flags)
693 {
694 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
695 
696 	/*
697 	 * Same comment as in bpf_perf_event_output_tp(), only that this time
698 	 * the other helper's function body cannot be inlined due to being
699 	 * external, thus we need to call raw helper function.
700 	 */
701 	return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
702 			       flags, 0, 0);
703 }
704 
705 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
706 	.func		= bpf_get_stackid_tp,
707 	.gpl_only	= true,
708 	.ret_type	= RET_INTEGER,
709 	.arg1_type	= ARG_PTR_TO_CTX,
710 	.arg2_type	= ARG_CONST_MAP_PTR,
711 	.arg3_type	= ARG_ANYTHING,
712 };
713 
714 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
715 	   u64, flags)
716 {
717 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
718 
719 	return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
720 			     (unsigned long) size, flags, 0);
721 }
722 
723 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
724 	.func		= bpf_get_stack_tp,
725 	.gpl_only	= true,
726 	.ret_type	= RET_INTEGER,
727 	.arg1_type	= ARG_PTR_TO_CTX,
728 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
729 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
730 	.arg4_type	= ARG_ANYTHING,
731 };
732 
733 static const struct bpf_func_proto *
734 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
735 {
736 	switch (func_id) {
737 	case BPF_FUNC_perf_event_output:
738 		return &bpf_perf_event_output_proto_tp;
739 	case BPF_FUNC_get_stackid:
740 		return &bpf_get_stackid_proto_tp;
741 	case BPF_FUNC_get_stack:
742 		return &bpf_get_stack_proto_tp;
743 	default:
744 		return tracing_func_proto(func_id, prog);
745 	}
746 }
747 
748 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
749 				    const struct bpf_prog *prog,
750 				    struct bpf_insn_access_aux *info)
751 {
752 	if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
753 		return false;
754 	if (type != BPF_READ)
755 		return false;
756 	if (off % size != 0)
757 		return false;
758 
759 	BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
760 	return true;
761 }
762 
763 const struct bpf_verifier_ops tracepoint_verifier_ops = {
764 	.get_func_proto  = tp_prog_func_proto,
765 	.is_valid_access = tp_prog_is_valid_access,
766 };
767 
768 const struct bpf_prog_ops tracepoint_prog_ops = {
769 };
770 
771 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
772 	   struct bpf_perf_event_value *, buf, u32, size)
773 {
774 	int err = -EINVAL;
775 
776 	if (unlikely(size != sizeof(struct bpf_perf_event_value)))
777 		goto clear;
778 	err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
779 				    &buf->running);
780 	if (unlikely(err))
781 		goto clear;
782 	return 0;
783 clear:
784 	memset(buf, 0, size);
785 	return err;
786 }
787 
788 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
789          .func           = bpf_perf_prog_read_value,
790          .gpl_only       = true,
791          .ret_type       = RET_INTEGER,
792          .arg1_type      = ARG_PTR_TO_CTX,
793          .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
794          .arg3_type      = ARG_CONST_SIZE,
795 };
796 
797 static const struct bpf_func_proto *
798 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
799 {
800 	switch (func_id) {
801 	case BPF_FUNC_perf_event_output:
802 		return &bpf_perf_event_output_proto_tp;
803 	case BPF_FUNC_get_stackid:
804 		return &bpf_get_stackid_proto_tp;
805 	case BPF_FUNC_get_stack:
806 		return &bpf_get_stack_proto_tp;
807 	case BPF_FUNC_perf_prog_read_value:
808 		return &bpf_perf_prog_read_value_proto;
809 	default:
810 		return tracing_func_proto(func_id, prog);
811 	}
812 }
813 
814 /*
815  * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
816  * to avoid potential recursive reuse issue when/if tracepoints are added
817  * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack
818  */
819 static DEFINE_PER_CPU(struct pt_regs, bpf_raw_tp_regs);
820 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
821 	   struct bpf_map *, map, u64, flags, void *, data, u64, size)
822 {
823 	struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
824 
825 	perf_fetch_caller_regs(regs);
826 	return ____bpf_perf_event_output(regs, map, flags, data, size);
827 }
828 
829 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
830 	.func		= bpf_perf_event_output_raw_tp,
831 	.gpl_only	= true,
832 	.ret_type	= RET_INTEGER,
833 	.arg1_type	= ARG_PTR_TO_CTX,
834 	.arg2_type	= ARG_CONST_MAP_PTR,
835 	.arg3_type	= ARG_ANYTHING,
836 	.arg4_type	= ARG_PTR_TO_MEM,
837 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
838 };
839 
840 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
841 	   struct bpf_map *, map, u64, flags)
842 {
843 	struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
844 
845 	perf_fetch_caller_regs(regs);
846 	/* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
847 	return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
848 			       flags, 0, 0);
849 }
850 
851 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
852 	.func		= bpf_get_stackid_raw_tp,
853 	.gpl_only	= true,
854 	.ret_type	= RET_INTEGER,
855 	.arg1_type	= ARG_PTR_TO_CTX,
856 	.arg2_type	= ARG_CONST_MAP_PTR,
857 	.arg3_type	= ARG_ANYTHING,
858 };
859 
860 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
861 	   void *, buf, u32, size, u64, flags)
862 {
863 	struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
864 
865 	perf_fetch_caller_regs(regs);
866 	return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
867 			     (unsigned long) size, flags, 0);
868 }
869 
870 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
871 	.func		= bpf_get_stack_raw_tp,
872 	.gpl_only	= true,
873 	.ret_type	= RET_INTEGER,
874 	.arg1_type	= ARG_PTR_TO_CTX,
875 	.arg2_type	= ARG_PTR_TO_MEM,
876 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
877 	.arg4_type	= ARG_ANYTHING,
878 };
879 
880 static const struct bpf_func_proto *
881 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
882 {
883 	switch (func_id) {
884 	case BPF_FUNC_perf_event_output:
885 		return &bpf_perf_event_output_proto_raw_tp;
886 	case BPF_FUNC_get_stackid:
887 		return &bpf_get_stackid_proto_raw_tp;
888 	case BPF_FUNC_get_stack:
889 		return &bpf_get_stack_proto_raw_tp;
890 	default:
891 		return tracing_func_proto(func_id, prog);
892 	}
893 }
894 
895 static bool raw_tp_prog_is_valid_access(int off, int size,
896 					enum bpf_access_type type,
897 					const struct bpf_prog *prog,
898 					struct bpf_insn_access_aux *info)
899 {
900 	/* largest tracepoint in the kernel has 12 args */
901 	if (off < 0 || off >= sizeof(__u64) * 12)
902 		return false;
903 	if (type != BPF_READ)
904 		return false;
905 	if (off % size != 0)
906 		return false;
907 	return true;
908 }
909 
910 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
911 	.get_func_proto  = raw_tp_prog_func_proto,
912 	.is_valid_access = raw_tp_prog_is_valid_access,
913 };
914 
915 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
916 };
917 
918 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
919 				    const struct bpf_prog *prog,
920 				    struct bpf_insn_access_aux *info)
921 {
922 	const int size_u64 = sizeof(u64);
923 
924 	if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
925 		return false;
926 	if (type != BPF_READ)
927 		return false;
928 	if (off % size != 0) {
929 		if (sizeof(unsigned long) != 4)
930 			return false;
931 		if (size != 8)
932 			return false;
933 		if (off % size != 4)
934 			return false;
935 	}
936 
937 	switch (off) {
938 	case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
939 		bpf_ctx_record_field_size(info, size_u64);
940 		if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
941 			return false;
942 		break;
943 	case bpf_ctx_range(struct bpf_perf_event_data, addr):
944 		bpf_ctx_record_field_size(info, size_u64);
945 		if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
946 			return false;
947 		break;
948 	default:
949 		if (size != sizeof(long))
950 			return false;
951 	}
952 
953 	return true;
954 }
955 
956 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
957 				      const struct bpf_insn *si,
958 				      struct bpf_insn *insn_buf,
959 				      struct bpf_prog *prog, u32 *target_size)
960 {
961 	struct bpf_insn *insn = insn_buf;
962 
963 	switch (si->off) {
964 	case offsetof(struct bpf_perf_event_data, sample_period):
965 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
966 						       data), si->dst_reg, si->src_reg,
967 				      offsetof(struct bpf_perf_event_data_kern, data));
968 		*insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
969 				      bpf_target_off(struct perf_sample_data, period, 8,
970 						     target_size));
971 		break;
972 	case offsetof(struct bpf_perf_event_data, addr):
973 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
974 						       data), si->dst_reg, si->src_reg,
975 				      offsetof(struct bpf_perf_event_data_kern, data));
976 		*insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
977 				      bpf_target_off(struct perf_sample_data, addr, 8,
978 						     target_size));
979 		break;
980 	default:
981 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
982 						       regs), si->dst_reg, si->src_reg,
983 				      offsetof(struct bpf_perf_event_data_kern, regs));
984 		*insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
985 				      si->off);
986 		break;
987 	}
988 
989 	return insn - insn_buf;
990 }
991 
992 const struct bpf_verifier_ops perf_event_verifier_ops = {
993 	.get_func_proto		= pe_prog_func_proto,
994 	.is_valid_access	= pe_prog_is_valid_access,
995 	.convert_ctx_access	= pe_prog_convert_ctx_access,
996 };
997 
998 const struct bpf_prog_ops perf_event_prog_ops = {
999 };
1000 
1001 static DEFINE_MUTEX(bpf_event_mutex);
1002 
1003 #define BPF_TRACE_MAX_PROGS 64
1004 
1005 int perf_event_attach_bpf_prog(struct perf_event *event,
1006 			       struct bpf_prog *prog)
1007 {
1008 	struct bpf_prog_array __rcu *old_array;
1009 	struct bpf_prog_array *new_array;
1010 	int ret = -EEXIST;
1011 
1012 	/*
1013 	 * Kprobe override only works if they are on the function entry,
1014 	 * and only if they are on the opt-in list.
1015 	 */
1016 	if (prog->kprobe_override &&
1017 	    (!trace_kprobe_on_func_entry(event->tp_event) ||
1018 	     !trace_kprobe_error_injectable(event->tp_event)))
1019 		return -EINVAL;
1020 
1021 	mutex_lock(&bpf_event_mutex);
1022 
1023 	if (event->prog)
1024 		goto unlock;
1025 
1026 	old_array = event->tp_event->prog_array;
1027 	if (old_array &&
1028 	    bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1029 		ret = -E2BIG;
1030 		goto unlock;
1031 	}
1032 
1033 	ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1034 	if (ret < 0)
1035 		goto unlock;
1036 
1037 	/* set the new array to event->tp_event and set event->prog */
1038 	event->prog = prog;
1039 	rcu_assign_pointer(event->tp_event->prog_array, new_array);
1040 	bpf_prog_array_free(old_array);
1041 
1042 unlock:
1043 	mutex_unlock(&bpf_event_mutex);
1044 	return ret;
1045 }
1046 
1047 void perf_event_detach_bpf_prog(struct perf_event *event)
1048 {
1049 	struct bpf_prog_array __rcu *old_array;
1050 	struct bpf_prog_array *new_array;
1051 	int ret;
1052 
1053 	mutex_lock(&bpf_event_mutex);
1054 
1055 	if (!event->prog)
1056 		goto unlock;
1057 
1058 	old_array = event->tp_event->prog_array;
1059 	ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1060 	if (ret == -ENOENT)
1061 		goto unlock;
1062 	if (ret < 0) {
1063 		bpf_prog_array_delete_safe(old_array, event->prog);
1064 	} else {
1065 		rcu_assign_pointer(event->tp_event->prog_array, new_array);
1066 		bpf_prog_array_free(old_array);
1067 	}
1068 
1069 	bpf_prog_put(event->prog);
1070 	event->prog = NULL;
1071 
1072 unlock:
1073 	mutex_unlock(&bpf_event_mutex);
1074 }
1075 
1076 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1077 {
1078 	struct perf_event_query_bpf __user *uquery = info;
1079 	struct perf_event_query_bpf query = {};
1080 	u32 *ids, prog_cnt, ids_len;
1081 	int ret;
1082 
1083 	if (!capable(CAP_SYS_ADMIN))
1084 		return -EPERM;
1085 	if (event->attr.type != PERF_TYPE_TRACEPOINT)
1086 		return -EINVAL;
1087 	if (copy_from_user(&query, uquery, sizeof(query)))
1088 		return -EFAULT;
1089 
1090 	ids_len = query.ids_len;
1091 	if (ids_len > BPF_TRACE_MAX_PROGS)
1092 		return -E2BIG;
1093 	ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1094 	if (!ids)
1095 		return -ENOMEM;
1096 	/*
1097 	 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1098 	 * is required when user only wants to check for uquery->prog_cnt.
1099 	 * There is no need to check for it since the case is handled
1100 	 * gracefully in bpf_prog_array_copy_info.
1101 	 */
1102 
1103 	mutex_lock(&bpf_event_mutex);
1104 	ret = bpf_prog_array_copy_info(event->tp_event->prog_array,
1105 				       ids,
1106 				       ids_len,
1107 				       &prog_cnt);
1108 	mutex_unlock(&bpf_event_mutex);
1109 
1110 	if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1111 	    copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1112 		ret = -EFAULT;
1113 
1114 	kfree(ids);
1115 	return ret;
1116 }
1117 
1118 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1119 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1120 
1121 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1122 {
1123 	struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1124 
1125 	for (; btp < __stop__bpf_raw_tp; btp++) {
1126 		if (!strcmp(btp->tp->name, name))
1127 			return btp;
1128 	}
1129 
1130 	return bpf_get_raw_tracepoint_module(name);
1131 }
1132 
1133 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1134 {
1135 	struct module *mod = __module_address((unsigned long)btp);
1136 
1137 	if (mod)
1138 		module_put(mod);
1139 }
1140 
1141 static __always_inline
1142 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1143 {
1144 	rcu_read_lock();
1145 	preempt_disable();
1146 	(void) BPF_PROG_RUN(prog, args);
1147 	preempt_enable();
1148 	rcu_read_unlock();
1149 }
1150 
1151 #define UNPACK(...)			__VA_ARGS__
1152 #define REPEAT_1(FN, DL, X, ...)	FN(X)
1153 #define REPEAT_2(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1154 #define REPEAT_3(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1155 #define REPEAT_4(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1156 #define REPEAT_5(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1157 #define REPEAT_6(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1158 #define REPEAT_7(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1159 #define REPEAT_8(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1160 #define REPEAT_9(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1161 #define REPEAT_10(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1162 #define REPEAT_11(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1163 #define REPEAT_12(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1164 #define REPEAT(X, FN, DL, ...)		REPEAT_##X(FN, DL, __VA_ARGS__)
1165 
1166 #define SARG(X)		u64 arg##X
1167 #define COPY(X)		args[X] = arg##X
1168 
1169 #define __DL_COM	(,)
1170 #define __DL_SEM	(;)
1171 
1172 #define __SEQ_0_11	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1173 
1174 #define BPF_TRACE_DEFN_x(x)						\
1175 	void bpf_trace_run##x(struct bpf_prog *prog,			\
1176 			      REPEAT(x, SARG, __DL_COM, __SEQ_0_11))	\
1177 	{								\
1178 		u64 args[x];						\
1179 		REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);			\
1180 		__bpf_trace_run(prog, args);				\
1181 	}								\
1182 	EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1183 BPF_TRACE_DEFN_x(1);
1184 BPF_TRACE_DEFN_x(2);
1185 BPF_TRACE_DEFN_x(3);
1186 BPF_TRACE_DEFN_x(4);
1187 BPF_TRACE_DEFN_x(5);
1188 BPF_TRACE_DEFN_x(6);
1189 BPF_TRACE_DEFN_x(7);
1190 BPF_TRACE_DEFN_x(8);
1191 BPF_TRACE_DEFN_x(9);
1192 BPF_TRACE_DEFN_x(10);
1193 BPF_TRACE_DEFN_x(11);
1194 BPF_TRACE_DEFN_x(12);
1195 
1196 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1197 {
1198 	struct tracepoint *tp = btp->tp;
1199 
1200 	/*
1201 	 * check that program doesn't access arguments beyond what's
1202 	 * available in this tracepoint
1203 	 */
1204 	if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1205 		return -EINVAL;
1206 
1207 	return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
1208 }
1209 
1210 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1211 {
1212 	return __bpf_probe_register(btp, prog);
1213 }
1214 
1215 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1216 {
1217 	return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1218 }
1219 
1220 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
1221 			    u32 *fd_type, const char **buf,
1222 			    u64 *probe_offset, u64 *probe_addr)
1223 {
1224 	bool is_tracepoint, is_syscall_tp;
1225 	struct bpf_prog *prog;
1226 	int flags, err = 0;
1227 
1228 	prog = event->prog;
1229 	if (!prog)
1230 		return -ENOENT;
1231 
1232 	/* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
1233 	if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
1234 		return -EOPNOTSUPP;
1235 
1236 	*prog_id = prog->aux->id;
1237 	flags = event->tp_event->flags;
1238 	is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
1239 	is_syscall_tp = is_syscall_trace_event(event->tp_event);
1240 
1241 	if (is_tracepoint || is_syscall_tp) {
1242 		*buf = is_tracepoint ? event->tp_event->tp->name
1243 				     : event->tp_event->name;
1244 		*fd_type = BPF_FD_TYPE_TRACEPOINT;
1245 		*probe_offset = 0x0;
1246 		*probe_addr = 0x0;
1247 	} else {
1248 		/* kprobe/uprobe */
1249 		err = -EOPNOTSUPP;
1250 #ifdef CONFIG_KPROBE_EVENTS
1251 		if (flags & TRACE_EVENT_FL_KPROBE)
1252 			err = bpf_get_kprobe_info(event, fd_type, buf,
1253 						  probe_offset, probe_addr,
1254 						  event->attr.type == PERF_TYPE_TRACEPOINT);
1255 #endif
1256 #ifdef CONFIG_UPROBE_EVENTS
1257 		if (flags & TRACE_EVENT_FL_UPROBE)
1258 			err = bpf_get_uprobe_info(event, fd_type, buf,
1259 						  probe_offset,
1260 						  event->attr.type == PERF_TYPE_TRACEPOINT);
1261 #endif
1262 	}
1263 
1264 	return err;
1265 }
1266 
1267 #ifdef CONFIG_MODULES
1268 int bpf_event_notify(struct notifier_block *nb, unsigned long op, void *module)
1269 {
1270 	struct bpf_trace_module *btm, *tmp;
1271 	struct module *mod = module;
1272 
1273 	if (mod->num_bpf_raw_events == 0 ||
1274 	    (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
1275 		return 0;
1276 
1277 	mutex_lock(&bpf_module_mutex);
1278 
1279 	switch (op) {
1280 	case MODULE_STATE_COMING:
1281 		btm = kzalloc(sizeof(*btm), GFP_KERNEL);
1282 		if (btm) {
1283 			btm->module = module;
1284 			list_add(&btm->list, &bpf_trace_modules);
1285 		}
1286 		break;
1287 	case MODULE_STATE_GOING:
1288 		list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
1289 			if (btm->module == module) {
1290 				list_del(&btm->list);
1291 				kfree(btm);
1292 				break;
1293 			}
1294 		}
1295 		break;
1296 	}
1297 
1298 	mutex_unlock(&bpf_module_mutex);
1299 
1300 	return 0;
1301 }
1302 
1303 static struct notifier_block bpf_module_nb = {
1304 	.notifier_call = bpf_event_notify,
1305 };
1306 
1307 int __init bpf_event_init(void)
1308 {
1309 	register_module_notifier(&bpf_module_nb);
1310 	return 0;
1311 }
1312 
1313 fs_initcall(bpf_event_init);
1314 #endif /* CONFIG_MODULES */
1315