xref: /openbmc/linux/kernel/trace/bpf_trace.c (revision df687341)
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 	perf_event_output(event, sd, regs);
435 	return 0;
436 }
437 
438 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
439 	   u64, flags, void *, data, u64, size)
440 {
441 	struct perf_sample_data *sd = this_cpu_ptr(&bpf_trace_sd);
442 	struct perf_raw_record raw = {
443 		.frag = {
444 			.size = size,
445 			.data = data,
446 		},
447 	};
448 
449 	if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
450 		return -EINVAL;
451 
452 	perf_sample_data_init(sd, 0, 0);
453 	sd->raw = &raw;
454 
455 	return __bpf_perf_event_output(regs, map, flags, sd);
456 }
457 
458 static const struct bpf_func_proto bpf_perf_event_output_proto = {
459 	.func		= bpf_perf_event_output,
460 	.gpl_only	= true,
461 	.ret_type	= RET_INTEGER,
462 	.arg1_type	= ARG_PTR_TO_CTX,
463 	.arg2_type	= ARG_CONST_MAP_PTR,
464 	.arg3_type	= ARG_ANYTHING,
465 	.arg4_type	= ARG_PTR_TO_MEM,
466 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
467 };
468 
469 static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs);
470 static DEFINE_PER_CPU(struct perf_sample_data, bpf_misc_sd);
471 
472 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
473 		     void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
474 {
475 	struct perf_sample_data *sd = this_cpu_ptr(&bpf_misc_sd);
476 	struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs);
477 	struct perf_raw_frag frag = {
478 		.copy		= ctx_copy,
479 		.size		= ctx_size,
480 		.data		= ctx,
481 	};
482 	struct perf_raw_record raw = {
483 		.frag = {
484 			{
485 				.next	= ctx_size ? &frag : NULL,
486 			},
487 			.size	= meta_size,
488 			.data	= meta,
489 		},
490 	};
491 
492 	perf_fetch_caller_regs(regs);
493 	perf_sample_data_init(sd, 0, 0);
494 	sd->raw = &raw;
495 
496 	return __bpf_perf_event_output(regs, map, flags, sd);
497 }
498 
499 BPF_CALL_0(bpf_get_current_task)
500 {
501 	return (long) current;
502 }
503 
504 static const struct bpf_func_proto bpf_get_current_task_proto = {
505 	.func		= bpf_get_current_task,
506 	.gpl_only	= true,
507 	.ret_type	= RET_INTEGER,
508 };
509 
510 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
511 {
512 	struct bpf_array *array = container_of(map, struct bpf_array, map);
513 	struct cgroup *cgrp;
514 
515 	if (unlikely(idx >= array->map.max_entries))
516 		return -E2BIG;
517 
518 	cgrp = READ_ONCE(array->ptrs[idx]);
519 	if (unlikely(!cgrp))
520 		return -EAGAIN;
521 
522 	return task_under_cgroup_hierarchy(current, cgrp);
523 }
524 
525 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
526 	.func           = bpf_current_task_under_cgroup,
527 	.gpl_only       = false,
528 	.ret_type       = RET_INTEGER,
529 	.arg1_type      = ARG_CONST_MAP_PTR,
530 	.arg2_type      = ARG_ANYTHING,
531 };
532 
533 BPF_CALL_3(bpf_probe_read_str, void *, dst, u32, size,
534 	   const void *, unsafe_ptr)
535 {
536 	int ret;
537 
538 	/*
539 	 * The strncpy_from_unsafe() call will likely not fill the entire
540 	 * buffer, but that's okay in this circumstance as we're probing
541 	 * arbitrary memory anyway similar to bpf_probe_read() and might
542 	 * as well probe the stack. Thus, memory is explicitly cleared
543 	 * only in error case, so that improper users ignoring return
544 	 * code altogether don't copy garbage; otherwise length of string
545 	 * is returned that can be used for bpf_perf_event_output() et al.
546 	 */
547 	ret = strncpy_from_unsafe(dst, unsafe_ptr, size);
548 	if (unlikely(ret < 0))
549 		memset(dst, 0, size);
550 
551 	return ret;
552 }
553 
554 static const struct bpf_func_proto bpf_probe_read_str_proto = {
555 	.func		= bpf_probe_read_str,
556 	.gpl_only	= true,
557 	.ret_type	= RET_INTEGER,
558 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
559 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
560 	.arg3_type	= ARG_ANYTHING,
561 };
562 
563 static const struct bpf_func_proto *
564 tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
565 {
566 	switch (func_id) {
567 	case BPF_FUNC_map_lookup_elem:
568 		return &bpf_map_lookup_elem_proto;
569 	case BPF_FUNC_map_update_elem:
570 		return &bpf_map_update_elem_proto;
571 	case BPF_FUNC_map_delete_elem:
572 		return &bpf_map_delete_elem_proto;
573 	case BPF_FUNC_probe_read:
574 		return &bpf_probe_read_proto;
575 	case BPF_FUNC_ktime_get_ns:
576 		return &bpf_ktime_get_ns_proto;
577 	case BPF_FUNC_tail_call:
578 		return &bpf_tail_call_proto;
579 	case BPF_FUNC_get_current_pid_tgid:
580 		return &bpf_get_current_pid_tgid_proto;
581 	case BPF_FUNC_get_current_task:
582 		return &bpf_get_current_task_proto;
583 	case BPF_FUNC_get_current_uid_gid:
584 		return &bpf_get_current_uid_gid_proto;
585 	case BPF_FUNC_get_current_comm:
586 		return &bpf_get_current_comm_proto;
587 	case BPF_FUNC_trace_printk:
588 		return bpf_get_trace_printk_proto();
589 	case BPF_FUNC_get_smp_processor_id:
590 		return &bpf_get_smp_processor_id_proto;
591 	case BPF_FUNC_get_numa_node_id:
592 		return &bpf_get_numa_node_id_proto;
593 	case BPF_FUNC_perf_event_read:
594 		return &bpf_perf_event_read_proto;
595 	case BPF_FUNC_probe_write_user:
596 		return bpf_get_probe_write_proto();
597 	case BPF_FUNC_current_task_under_cgroup:
598 		return &bpf_current_task_under_cgroup_proto;
599 	case BPF_FUNC_get_prandom_u32:
600 		return &bpf_get_prandom_u32_proto;
601 	case BPF_FUNC_probe_read_str:
602 		return &bpf_probe_read_str_proto;
603 #ifdef CONFIG_CGROUPS
604 	case BPF_FUNC_get_current_cgroup_id:
605 		return &bpf_get_current_cgroup_id_proto;
606 #endif
607 	default:
608 		return NULL;
609 	}
610 }
611 
612 static const struct bpf_func_proto *
613 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
614 {
615 	switch (func_id) {
616 	case BPF_FUNC_perf_event_output:
617 		return &bpf_perf_event_output_proto;
618 	case BPF_FUNC_get_stackid:
619 		return &bpf_get_stackid_proto;
620 	case BPF_FUNC_get_stack:
621 		return &bpf_get_stack_proto;
622 	case BPF_FUNC_perf_event_read_value:
623 		return &bpf_perf_event_read_value_proto;
624 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
625 	case BPF_FUNC_override_return:
626 		return &bpf_override_return_proto;
627 #endif
628 	default:
629 		return tracing_func_proto(func_id, prog);
630 	}
631 }
632 
633 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
634 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
635 					const struct bpf_prog *prog,
636 					struct bpf_insn_access_aux *info)
637 {
638 	if (off < 0 || off >= sizeof(struct pt_regs))
639 		return false;
640 	if (type != BPF_READ)
641 		return false;
642 	if (off % size != 0)
643 		return false;
644 	/*
645 	 * Assertion for 32 bit to make sure last 8 byte access
646 	 * (BPF_DW) to the last 4 byte member is disallowed.
647 	 */
648 	if (off + size > sizeof(struct pt_regs))
649 		return false;
650 
651 	return true;
652 }
653 
654 const struct bpf_verifier_ops kprobe_verifier_ops = {
655 	.get_func_proto  = kprobe_prog_func_proto,
656 	.is_valid_access = kprobe_prog_is_valid_access,
657 };
658 
659 const struct bpf_prog_ops kprobe_prog_ops = {
660 };
661 
662 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
663 	   u64, flags, void *, data, u64, size)
664 {
665 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
666 
667 	/*
668 	 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
669 	 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
670 	 * from there and call the same bpf_perf_event_output() helper inline.
671 	 */
672 	return ____bpf_perf_event_output(regs, map, flags, data, size);
673 }
674 
675 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
676 	.func		= bpf_perf_event_output_tp,
677 	.gpl_only	= true,
678 	.ret_type	= RET_INTEGER,
679 	.arg1_type	= ARG_PTR_TO_CTX,
680 	.arg2_type	= ARG_CONST_MAP_PTR,
681 	.arg3_type	= ARG_ANYTHING,
682 	.arg4_type	= ARG_PTR_TO_MEM,
683 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
684 };
685 
686 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
687 	   u64, flags)
688 {
689 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
690 
691 	/*
692 	 * Same comment as in bpf_perf_event_output_tp(), only that this time
693 	 * the other helper's function body cannot be inlined due to being
694 	 * external, thus we need to call raw helper function.
695 	 */
696 	return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
697 			       flags, 0, 0);
698 }
699 
700 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
701 	.func		= bpf_get_stackid_tp,
702 	.gpl_only	= true,
703 	.ret_type	= RET_INTEGER,
704 	.arg1_type	= ARG_PTR_TO_CTX,
705 	.arg2_type	= ARG_CONST_MAP_PTR,
706 	.arg3_type	= ARG_ANYTHING,
707 };
708 
709 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
710 	   u64, flags)
711 {
712 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
713 
714 	return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
715 			     (unsigned long) size, flags, 0);
716 }
717 
718 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
719 	.func		= bpf_get_stack_tp,
720 	.gpl_only	= true,
721 	.ret_type	= RET_INTEGER,
722 	.arg1_type	= ARG_PTR_TO_CTX,
723 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
724 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
725 	.arg4_type	= ARG_ANYTHING,
726 };
727 
728 static const struct bpf_func_proto *
729 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
730 {
731 	switch (func_id) {
732 	case BPF_FUNC_perf_event_output:
733 		return &bpf_perf_event_output_proto_tp;
734 	case BPF_FUNC_get_stackid:
735 		return &bpf_get_stackid_proto_tp;
736 	case BPF_FUNC_get_stack:
737 		return &bpf_get_stack_proto_tp;
738 	default:
739 		return tracing_func_proto(func_id, prog);
740 	}
741 }
742 
743 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
744 				    const struct bpf_prog *prog,
745 				    struct bpf_insn_access_aux *info)
746 {
747 	if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
748 		return false;
749 	if (type != BPF_READ)
750 		return false;
751 	if (off % size != 0)
752 		return false;
753 
754 	BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
755 	return true;
756 }
757 
758 const struct bpf_verifier_ops tracepoint_verifier_ops = {
759 	.get_func_proto  = tp_prog_func_proto,
760 	.is_valid_access = tp_prog_is_valid_access,
761 };
762 
763 const struct bpf_prog_ops tracepoint_prog_ops = {
764 };
765 
766 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
767 	   struct bpf_perf_event_value *, buf, u32, size)
768 {
769 	int err = -EINVAL;
770 
771 	if (unlikely(size != sizeof(struct bpf_perf_event_value)))
772 		goto clear;
773 	err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
774 				    &buf->running);
775 	if (unlikely(err))
776 		goto clear;
777 	return 0;
778 clear:
779 	memset(buf, 0, size);
780 	return err;
781 }
782 
783 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
784          .func           = bpf_perf_prog_read_value,
785          .gpl_only       = true,
786          .ret_type       = RET_INTEGER,
787          .arg1_type      = ARG_PTR_TO_CTX,
788          .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
789          .arg3_type      = ARG_CONST_SIZE,
790 };
791 
792 static const struct bpf_func_proto *
793 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
794 {
795 	switch (func_id) {
796 	case BPF_FUNC_perf_event_output:
797 		return &bpf_perf_event_output_proto_tp;
798 	case BPF_FUNC_get_stackid:
799 		return &bpf_get_stackid_proto_tp;
800 	case BPF_FUNC_get_stack:
801 		return &bpf_get_stack_proto_tp;
802 	case BPF_FUNC_perf_prog_read_value:
803 		return &bpf_perf_prog_read_value_proto;
804 	default:
805 		return tracing_func_proto(func_id, prog);
806 	}
807 }
808 
809 /*
810  * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
811  * to avoid potential recursive reuse issue when/if tracepoints are added
812  * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack
813  */
814 static DEFINE_PER_CPU(struct pt_regs, bpf_raw_tp_regs);
815 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
816 	   struct bpf_map *, map, u64, flags, void *, data, u64, size)
817 {
818 	struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
819 
820 	perf_fetch_caller_regs(regs);
821 	return ____bpf_perf_event_output(regs, map, flags, data, size);
822 }
823 
824 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
825 	.func		= bpf_perf_event_output_raw_tp,
826 	.gpl_only	= true,
827 	.ret_type	= RET_INTEGER,
828 	.arg1_type	= ARG_PTR_TO_CTX,
829 	.arg2_type	= ARG_CONST_MAP_PTR,
830 	.arg3_type	= ARG_ANYTHING,
831 	.arg4_type	= ARG_PTR_TO_MEM,
832 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
833 };
834 
835 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
836 	   struct bpf_map *, map, u64, flags)
837 {
838 	struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
839 
840 	perf_fetch_caller_regs(regs);
841 	/* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
842 	return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
843 			       flags, 0, 0);
844 }
845 
846 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
847 	.func		= bpf_get_stackid_raw_tp,
848 	.gpl_only	= true,
849 	.ret_type	= RET_INTEGER,
850 	.arg1_type	= ARG_PTR_TO_CTX,
851 	.arg2_type	= ARG_CONST_MAP_PTR,
852 	.arg3_type	= ARG_ANYTHING,
853 };
854 
855 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
856 	   void *, buf, u32, size, u64, flags)
857 {
858 	struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
859 
860 	perf_fetch_caller_regs(regs);
861 	return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
862 			     (unsigned long) size, flags, 0);
863 }
864 
865 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
866 	.func		= bpf_get_stack_raw_tp,
867 	.gpl_only	= true,
868 	.ret_type	= RET_INTEGER,
869 	.arg1_type	= ARG_PTR_TO_CTX,
870 	.arg2_type	= ARG_PTR_TO_MEM,
871 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
872 	.arg4_type	= ARG_ANYTHING,
873 };
874 
875 static const struct bpf_func_proto *
876 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
877 {
878 	switch (func_id) {
879 	case BPF_FUNC_perf_event_output:
880 		return &bpf_perf_event_output_proto_raw_tp;
881 	case BPF_FUNC_get_stackid:
882 		return &bpf_get_stackid_proto_raw_tp;
883 	case BPF_FUNC_get_stack:
884 		return &bpf_get_stack_proto_raw_tp;
885 	default:
886 		return tracing_func_proto(func_id, prog);
887 	}
888 }
889 
890 static bool raw_tp_prog_is_valid_access(int off, int size,
891 					enum bpf_access_type type,
892 					const struct bpf_prog *prog,
893 					struct bpf_insn_access_aux *info)
894 {
895 	/* largest tracepoint in the kernel has 12 args */
896 	if (off < 0 || off >= sizeof(__u64) * 12)
897 		return false;
898 	if (type != BPF_READ)
899 		return false;
900 	if (off % size != 0)
901 		return false;
902 	return true;
903 }
904 
905 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
906 	.get_func_proto  = raw_tp_prog_func_proto,
907 	.is_valid_access = raw_tp_prog_is_valid_access,
908 };
909 
910 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
911 };
912 
913 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
914 				    const struct bpf_prog *prog,
915 				    struct bpf_insn_access_aux *info)
916 {
917 	const int size_u64 = sizeof(u64);
918 
919 	if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
920 		return false;
921 	if (type != BPF_READ)
922 		return false;
923 	if (off % size != 0) {
924 		if (sizeof(unsigned long) != 4)
925 			return false;
926 		if (size != 8)
927 			return false;
928 		if (off % size != 4)
929 			return false;
930 	}
931 
932 	switch (off) {
933 	case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
934 		bpf_ctx_record_field_size(info, size_u64);
935 		if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
936 			return false;
937 		break;
938 	case bpf_ctx_range(struct bpf_perf_event_data, addr):
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 	default:
944 		if (size != sizeof(long))
945 			return false;
946 	}
947 
948 	return true;
949 }
950 
951 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
952 				      const struct bpf_insn *si,
953 				      struct bpf_insn *insn_buf,
954 				      struct bpf_prog *prog, u32 *target_size)
955 {
956 	struct bpf_insn *insn = insn_buf;
957 
958 	switch (si->off) {
959 	case offsetof(struct bpf_perf_event_data, sample_period):
960 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
961 						       data), si->dst_reg, si->src_reg,
962 				      offsetof(struct bpf_perf_event_data_kern, data));
963 		*insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
964 				      bpf_target_off(struct perf_sample_data, period, 8,
965 						     target_size));
966 		break;
967 	case offsetof(struct bpf_perf_event_data, addr):
968 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
969 						       data), si->dst_reg, si->src_reg,
970 				      offsetof(struct bpf_perf_event_data_kern, data));
971 		*insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
972 				      bpf_target_off(struct perf_sample_data, addr, 8,
973 						     target_size));
974 		break;
975 	default:
976 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
977 						       regs), si->dst_reg, si->src_reg,
978 				      offsetof(struct bpf_perf_event_data_kern, regs));
979 		*insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
980 				      si->off);
981 		break;
982 	}
983 
984 	return insn - insn_buf;
985 }
986 
987 const struct bpf_verifier_ops perf_event_verifier_ops = {
988 	.get_func_proto		= pe_prog_func_proto,
989 	.is_valid_access	= pe_prog_is_valid_access,
990 	.convert_ctx_access	= pe_prog_convert_ctx_access,
991 };
992 
993 const struct bpf_prog_ops perf_event_prog_ops = {
994 };
995 
996 static DEFINE_MUTEX(bpf_event_mutex);
997 
998 #define BPF_TRACE_MAX_PROGS 64
999 
1000 int perf_event_attach_bpf_prog(struct perf_event *event,
1001 			       struct bpf_prog *prog)
1002 {
1003 	struct bpf_prog_array __rcu *old_array;
1004 	struct bpf_prog_array *new_array;
1005 	int ret = -EEXIST;
1006 
1007 	/*
1008 	 * Kprobe override only works if they are on the function entry,
1009 	 * and only if they are on the opt-in list.
1010 	 */
1011 	if (prog->kprobe_override &&
1012 	    (!trace_kprobe_on_func_entry(event->tp_event) ||
1013 	     !trace_kprobe_error_injectable(event->tp_event)))
1014 		return -EINVAL;
1015 
1016 	mutex_lock(&bpf_event_mutex);
1017 
1018 	if (event->prog)
1019 		goto unlock;
1020 
1021 	old_array = event->tp_event->prog_array;
1022 	if (old_array &&
1023 	    bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1024 		ret = -E2BIG;
1025 		goto unlock;
1026 	}
1027 
1028 	ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1029 	if (ret < 0)
1030 		goto unlock;
1031 
1032 	/* set the new array to event->tp_event and set event->prog */
1033 	event->prog = prog;
1034 	rcu_assign_pointer(event->tp_event->prog_array, new_array);
1035 	bpf_prog_array_free(old_array);
1036 
1037 unlock:
1038 	mutex_unlock(&bpf_event_mutex);
1039 	return ret;
1040 }
1041 
1042 void perf_event_detach_bpf_prog(struct perf_event *event)
1043 {
1044 	struct bpf_prog_array __rcu *old_array;
1045 	struct bpf_prog_array *new_array;
1046 	int ret;
1047 
1048 	mutex_lock(&bpf_event_mutex);
1049 
1050 	if (!event->prog)
1051 		goto unlock;
1052 
1053 	old_array = event->tp_event->prog_array;
1054 	ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1055 	if (ret == -ENOENT)
1056 		goto unlock;
1057 	if (ret < 0) {
1058 		bpf_prog_array_delete_safe(old_array, event->prog);
1059 	} else {
1060 		rcu_assign_pointer(event->tp_event->prog_array, new_array);
1061 		bpf_prog_array_free(old_array);
1062 	}
1063 
1064 	bpf_prog_put(event->prog);
1065 	event->prog = NULL;
1066 
1067 unlock:
1068 	mutex_unlock(&bpf_event_mutex);
1069 }
1070 
1071 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1072 {
1073 	struct perf_event_query_bpf __user *uquery = info;
1074 	struct perf_event_query_bpf query = {};
1075 	u32 *ids, prog_cnt, ids_len;
1076 	int ret;
1077 
1078 	if (!capable(CAP_SYS_ADMIN))
1079 		return -EPERM;
1080 	if (event->attr.type != PERF_TYPE_TRACEPOINT)
1081 		return -EINVAL;
1082 	if (copy_from_user(&query, uquery, sizeof(query)))
1083 		return -EFAULT;
1084 
1085 	ids_len = query.ids_len;
1086 	if (ids_len > BPF_TRACE_MAX_PROGS)
1087 		return -E2BIG;
1088 	ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1089 	if (!ids)
1090 		return -ENOMEM;
1091 	/*
1092 	 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1093 	 * is required when user only wants to check for uquery->prog_cnt.
1094 	 * There is no need to check for it since the case is handled
1095 	 * gracefully in bpf_prog_array_copy_info.
1096 	 */
1097 
1098 	mutex_lock(&bpf_event_mutex);
1099 	ret = bpf_prog_array_copy_info(event->tp_event->prog_array,
1100 				       ids,
1101 				       ids_len,
1102 				       &prog_cnt);
1103 	mutex_unlock(&bpf_event_mutex);
1104 
1105 	if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1106 	    copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1107 		ret = -EFAULT;
1108 
1109 	kfree(ids);
1110 	return ret;
1111 }
1112 
1113 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1114 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1115 
1116 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1117 {
1118 	struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1119 
1120 	for (; btp < __stop__bpf_raw_tp; btp++) {
1121 		if (!strcmp(btp->tp->name, name))
1122 			return btp;
1123 	}
1124 
1125 	return bpf_get_raw_tracepoint_module(name);
1126 }
1127 
1128 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1129 {
1130 	struct module *mod = __module_address((unsigned long)btp);
1131 
1132 	if (mod)
1133 		module_put(mod);
1134 }
1135 
1136 static __always_inline
1137 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1138 {
1139 	rcu_read_lock();
1140 	preempt_disable();
1141 	(void) BPF_PROG_RUN(prog, args);
1142 	preempt_enable();
1143 	rcu_read_unlock();
1144 }
1145 
1146 #define UNPACK(...)			__VA_ARGS__
1147 #define REPEAT_1(FN, DL, X, ...)	FN(X)
1148 #define REPEAT_2(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1149 #define REPEAT_3(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1150 #define REPEAT_4(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1151 #define REPEAT_5(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1152 #define REPEAT_6(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1153 #define REPEAT_7(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1154 #define REPEAT_8(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1155 #define REPEAT_9(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1156 #define REPEAT_10(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1157 #define REPEAT_11(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1158 #define REPEAT_12(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1159 #define REPEAT(X, FN, DL, ...)		REPEAT_##X(FN, DL, __VA_ARGS__)
1160 
1161 #define SARG(X)		u64 arg##X
1162 #define COPY(X)		args[X] = arg##X
1163 
1164 #define __DL_COM	(,)
1165 #define __DL_SEM	(;)
1166 
1167 #define __SEQ_0_11	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1168 
1169 #define BPF_TRACE_DEFN_x(x)						\
1170 	void bpf_trace_run##x(struct bpf_prog *prog,			\
1171 			      REPEAT(x, SARG, __DL_COM, __SEQ_0_11))	\
1172 	{								\
1173 		u64 args[x];						\
1174 		REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);			\
1175 		__bpf_trace_run(prog, args);				\
1176 	}								\
1177 	EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1178 BPF_TRACE_DEFN_x(1);
1179 BPF_TRACE_DEFN_x(2);
1180 BPF_TRACE_DEFN_x(3);
1181 BPF_TRACE_DEFN_x(4);
1182 BPF_TRACE_DEFN_x(5);
1183 BPF_TRACE_DEFN_x(6);
1184 BPF_TRACE_DEFN_x(7);
1185 BPF_TRACE_DEFN_x(8);
1186 BPF_TRACE_DEFN_x(9);
1187 BPF_TRACE_DEFN_x(10);
1188 BPF_TRACE_DEFN_x(11);
1189 BPF_TRACE_DEFN_x(12);
1190 
1191 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1192 {
1193 	struct tracepoint *tp = btp->tp;
1194 
1195 	/*
1196 	 * check that program doesn't access arguments beyond what's
1197 	 * available in this tracepoint
1198 	 */
1199 	if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1200 		return -EINVAL;
1201 
1202 	return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
1203 }
1204 
1205 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1206 {
1207 	int err;
1208 
1209 	mutex_lock(&bpf_event_mutex);
1210 	err = __bpf_probe_register(btp, prog);
1211 	mutex_unlock(&bpf_event_mutex);
1212 	return err;
1213 }
1214 
1215 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1216 {
1217 	int err;
1218 
1219 	mutex_lock(&bpf_event_mutex);
1220 	err = tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1221 	mutex_unlock(&bpf_event_mutex);
1222 	return err;
1223 }
1224 
1225 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
1226 			    u32 *fd_type, const char **buf,
1227 			    u64 *probe_offset, u64 *probe_addr)
1228 {
1229 	bool is_tracepoint, is_syscall_tp;
1230 	struct bpf_prog *prog;
1231 	int flags, err = 0;
1232 
1233 	prog = event->prog;
1234 	if (!prog)
1235 		return -ENOENT;
1236 
1237 	/* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
1238 	if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
1239 		return -EOPNOTSUPP;
1240 
1241 	*prog_id = prog->aux->id;
1242 	flags = event->tp_event->flags;
1243 	is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
1244 	is_syscall_tp = is_syscall_trace_event(event->tp_event);
1245 
1246 	if (is_tracepoint || is_syscall_tp) {
1247 		*buf = is_tracepoint ? event->tp_event->tp->name
1248 				     : event->tp_event->name;
1249 		*fd_type = BPF_FD_TYPE_TRACEPOINT;
1250 		*probe_offset = 0x0;
1251 		*probe_addr = 0x0;
1252 	} else {
1253 		/* kprobe/uprobe */
1254 		err = -EOPNOTSUPP;
1255 #ifdef CONFIG_KPROBE_EVENTS
1256 		if (flags & TRACE_EVENT_FL_KPROBE)
1257 			err = bpf_get_kprobe_info(event, fd_type, buf,
1258 						  probe_offset, probe_addr,
1259 						  event->attr.type == PERF_TYPE_TRACEPOINT);
1260 #endif
1261 #ifdef CONFIG_UPROBE_EVENTS
1262 		if (flags & TRACE_EVENT_FL_UPROBE)
1263 			err = bpf_get_uprobe_info(event, fd_type, buf,
1264 						  probe_offset,
1265 						  event->attr.type == PERF_TYPE_TRACEPOINT);
1266 #endif
1267 	}
1268 
1269 	return err;
1270 }
1271 
1272 #ifdef CONFIG_MODULES
1273 int bpf_event_notify(struct notifier_block *nb, unsigned long op, void *module)
1274 {
1275 	struct bpf_trace_module *btm, *tmp;
1276 	struct module *mod = module;
1277 
1278 	if (mod->num_bpf_raw_events == 0 ||
1279 	    (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
1280 		return 0;
1281 
1282 	mutex_lock(&bpf_module_mutex);
1283 
1284 	switch (op) {
1285 	case MODULE_STATE_COMING:
1286 		btm = kzalloc(sizeof(*btm), GFP_KERNEL);
1287 		if (btm) {
1288 			btm->module = module;
1289 			list_add(&btm->list, &bpf_trace_modules);
1290 		}
1291 		break;
1292 	case MODULE_STATE_GOING:
1293 		list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
1294 			if (btm->module == module) {
1295 				list_del(&btm->list);
1296 				kfree(btm);
1297 				break;
1298 			}
1299 		}
1300 		break;
1301 	}
1302 
1303 	mutex_unlock(&bpf_module_mutex);
1304 
1305 	return 0;
1306 }
1307 
1308 static struct notifier_block bpf_module_nb = {
1309 	.notifier_call = bpf_event_notify,
1310 };
1311 
1312 int __init bpf_event_init(void)
1313 {
1314 	register_module_notifier(&bpf_module_nb);
1315 	return 0;
1316 }
1317 
1318 fs_initcall(bpf_event_init);
1319 #endif /* CONFIG_MODULES */
1320