xref: /linux/kernel/trace/bpf_trace.c (revision a6cdeeb16bff89c8486324f53577db058cbe81ba)
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 <asm/tlb.h>
18 
19 #include "trace_probe.h"
20 #include "trace.h"
21 
22 #define bpf_event_rcu_dereference(p)					\
23 	rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
24 
25 #ifdef CONFIG_MODULES
26 struct bpf_trace_module {
27 	struct module *module;
28 	struct list_head list;
29 };
30 
31 static LIST_HEAD(bpf_trace_modules);
32 static DEFINE_MUTEX(bpf_module_mutex);
33 
34 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
35 {
36 	struct bpf_raw_event_map *btp, *ret = NULL;
37 	struct bpf_trace_module *btm;
38 	unsigned int i;
39 
40 	mutex_lock(&bpf_module_mutex);
41 	list_for_each_entry(btm, &bpf_trace_modules, list) {
42 		for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
43 			btp = &btm->module->bpf_raw_events[i];
44 			if (!strcmp(btp->tp->name, name)) {
45 				if (try_module_get(btm->module))
46 					ret = btp;
47 				goto out;
48 			}
49 		}
50 	}
51 out:
52 	mutex_unlock(&bpf_module_mutex);
53 	return ret;
54 }
55 #else
56 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
57 {
58 	return NULL;
59 }
60 #endif /* CONFIG_MODULES */
61 
62 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
63 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
64 
65 /**
66  * trace_call_bpf - invoke BPF program
67  * @call: tracepoint event
68  * @ctx: opaque context pointer
69  *
70  * kprobe handlers execute BPF programs via this helper.
71  * Can be used from static tracepoints in the future.
72  *
73  * Return: BPF programs always return an integer which is interpreted by
74  * kprobe handler as:
75  * 0 - return from kprobe (event is filtered out)
76  * 1 - store kprobe event into ring buffer
77  * Other values are reserved and currently alias to 1
78  */
79 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
80 {
81 	unsigned int ret;
82 
83 	if (in_nmi()) /* not supported yet */
84 		return 1;
85 
86 	preempt_disable();
87 
88 	if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
89 		/*
90 		 * since some bpf program is already running on this cpu,
91 		 * don't call into another bpf program (same or different)
92 		 * and don't send kprobe event into ring-buffer,
93 		 * so return zero here
94 		 */
95 		ret = 0;
96 		goto out;
97 	}
98 
99 	/*
100 	 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
101 	 * to all call sites, we did a bpf_prog_array_valid() there to check
102 	 * whether call->prog_array is empty or not, which is
103 	 * a heurisitc to speed up execution.
104 	 *
105 	 * If bpf_prog_array_valid() fetched prog_array was
106 	 * non-NULL, we go into trace_call_bpf() and do the actual
107 	 * proper rcu_dereference() under RCU lock.
108 	 * If it turns out that prog_array is NULL then, we bail out.
109 	 * For the opposite, if the bpf_prog_array_valid() fetched pointer
110 	 * was NULL, you'll skip the prog_array with the risk of missing
111 	 * out of events when it was updated in between this and the
112 	 * rcu_dereference() which is accepted risk.
113 	 */
114 	ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
115 
116  out:
117 	__this_cpu_dec(bpf_prog_active);
118 	preempt_enable();
119 
120 	return ret;
121 }
122 EXPORT_SYMBOL_GPL(trace_call_bpf);
123 
124 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
125 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
126 {
127 	regs_set_return_value(regs, rc);
128 	override_function_with_return(regs);
129 	return 0;
130 }
131 
132 static const struct bpf_func_proto bpf_override_return_proto = {
133 	.func		= bpf_override_return,
134 	.gpl_only	= true,
135 	.ret_type	= RET_INTEGER,
136 	.arg1_type	= ARG_PTR_TO_CTX,
137 	.arg2_type	= ARG_ANYTHING,
138 };
139 #endif
140 
141 BPF_CALL_3(bpf_probe_read, void *, dst, u32, size, const void *, unsafe_ptr)
142 {
143 	int ret;
144 
145 	ret = probe_kernel_read(dst, unsafe_ptr, size);
146 	if (unlikely(ret < 0))
147 		memset(dst, 0, size);
148 
149 	return ret;
150 }
151 
152 static const struct bpf_func_proto bpf_probe_read_proto = {
153 	.func		= bpf_probe_read,
154 	.gpl_only	= true,
155 	.ret_type	= RET_INTEGER,
156 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
157 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
158 	.arg3_type	= ARG_ANYTHING,
159 };
160 
161 BPF_CALL_3(bpf_probe_write_user, void *, unsafe_ptr, const void *, src,
162 	   u32, size)
163 {
164 	/*
165 	 * Ensure we're in user context which is safe for the helper to
166 	 * run. This helper has no business in a kthread.
167 	 *
168 	 * access_ok() should prevent writing to non-user memory, but in
169 	 * some situations (nommu, temporary switch, etc) access_ok() does
170 	 * not provide enough validation, hence the check on KERNEL_DS.
171 	 *
172 	 * nmi_uaccess_okay() ensures the probe is not run in an interim
173 	 * state, when the task or mm are switched. This is specifically
174 	 * required to prevent the use of temporary mm.
175 	 */
176 
177 	if (unlikely(in_interrupt() ||
178 		     current->flags & (PF_KTHREAD | PF_EXITING)))
179 		return -EPERM;
180 	if (unlikely(uaccess_kernel()))
181 		return -EPERM;
182 	if (unlikely(!nmi_uaccess_okay()))
183 		return -EPERM;
184 	if (!access_ok(unsafe_ptr, size))
185 		return -EPERM;
186 
187 	return probe_kernel_write(unsafe_ptr, src, size);
188 }
189 
190 static const struct bpf_func_proto bpf_probe_write_user_proto = {
191 	.func		= bpf_probe_write_user,
192 	.gpl_only	= true,
193 	.ret_type	= RET_INTEGER,
194 	.arg1_type	= ARG_ANYTHING,
195 	.arg2_type	= ARG_PTR_TO_MEM,
196 	.arg3_type	= ARG_CONST_SIZE,
197 };
198 
199 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
200 {
201 	pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
202 			    current->comm, task_pid_nr(current));
203 
204 	return &bpf_probe_write_user_proto;
205 }
206 
207 /*
208  * Only limited trace_printk() conversion specifiers allowed:
209  * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %s
210  */
211 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
212 	   u64, arg2, u64, arg3)
213 {
214 	bool str_seen = false;
215 	int mod[3] = {};
216 	int fmt_cnt = 0;
217 	u64 unsafe_addr;
218 	char buf[64];
219 	int i;
220 
221 	/*
222 	 * bpf_check()->check_func_arg()->check_stack_boundary()
223 	 * guarantees that fmt points to bpf program stack,
224 	 * fmt_size bytes of it were initialized and fmt_size > 0
225 	 */
226 	if (fmt[--fmt_size] != 0)
227 		return -EINVAL;
228 
229 	/* check format string for allowed specifiers */
230 	for (i = 0; i < fmt_size; i++) {
231 		if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
232 			return -EINVAL;
233 
234 		if (fmt[i] != '%')
235 			continue;
236 
237 		if (fmt_cnt >= 3)
238 			return -EINVAL;
239 
240 		/* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
241 		i++;
242 		if (fmt[i] == 'l') {
243 			mod[fmt_cnt]++;
244 			i++;
245 		} else if (fmt[i] == 'p' || fmt[i] == 's') {
246 			mod[fmt_cnt]++;
247 			/* disallow any further format extensions */
248 			if (fmt[i + 1] != 0 &&
249 			    !isspace(fmt[i + 1]) &&
250 			    !ispunct(fmt[i + 1]))
251 				return -EINVAL;
252 			fmt_cnt++;
253 			if (fmt[i] == 's') {
254 				if (str_seen)
255 					/* allow only one '%s' per fmt string */
256 					return -EINVAL;
257 				str_seen = true;
258 
259 				switch (fmt_cnt) {
260 				case 1:
261 					unsafe_addr = arg1;
262 					arg1 = (long) buf;
263 					break;
264 				case 2:
265 					unsafe_addr = arg2;
266 					arg2 = (long) buf;
267 					break;
268 				case 3:
269 					unsafe_addr = arg3;
270 					arg3 = (long) buf;
271 					break;
272 				}
273 				buf[0] = 0;
274 				strncpy_from_unsafe(buf,
275 						    (void *) (long) unsafe_addr,
276 						    sizeof(buf));
277 			}
278 			continue;
279 		}
280 
281 		if (fmt[i] == 'l') {
282 			mod[fmt_cnt]++;
283 			i++;
284 		}
285 
286 		if (fmt[i] != 'i' && fmt[i] != 'd' &&
287 		    fmt[i] != 'u' && fmt[i] != 'x')
288 			return -EINVAL;
289 		fmt_cnt++;
290 	}
291 
292 /* Horrid workaround for getting va_list handling working with different
293  * argument type combinations generically for 32 and 64 bit archs.
294  */
295 #define __BPF_TP_EMIT()	__BPF_ARG3_TP()
296 #define __BPF_TP(...)							\
297 	__trace_printk(0 /* Fake ip */,					\
298 		       fmt, ##__VA_ARGS__)
299 
300 #define __BPF_ARG1_TP(...)						\
301 	((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64))	\
302 	  ? __BPF_TP(arg1, ##__VA_ARGS__)				\
303 	  : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32))	\
304 	      ? __BPF_TP((long)arg1, ##__VA_ARGS__)			\
305 	      : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
306 
307 #define __BPF_ARG2_TP(...)						\
308 	((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64))	\
309 	  ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__)				\
310 	  : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32))	\
311 	      ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__)		\
312 	      : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
313 
314 #define __BPF_ARG3_TP(...)						\
315 	((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64))	\
316 	  ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__)				\
317 	  : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32))	\
318 	      ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__)		\
319 	      : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
320 
321 	return __BPF_TP_EMIT();
322 }
323 
324 static const struct bpf_func_proto bpf_trace_printk_proto = {
325 	.func		= bpf_trace_printk,
326 	.gpl_only	= true,
327 	.ret_type	= RET_INTEGER,
328 	.arg1_type	= ARG_PTR_TO_MEM,
329 	.arg2_type	= ARG_CONST_SIZE,
330 };
331 
332 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
333 {
334 	/*
335 	 * this program might be calling bpf_trace_printk,
336 	 * so allocate per-cpu printk buffers
337 	 */
338 	trace_printk_init_buffers();
339 
340 	return &bpf_trace_printk_proto;
341 }
342 
343 static __always_inline int
344 get_map_perf_counter(struct bpf_map *map, u64 flags,
345 		     u64 *value, u64 *enabled, u64 *running)
346 {
347 	struct bpf_array *array = container_of(map, struct bpf_array, map);
348 	unsigned int cpu = smp_processor_id();
349 	u64 index = flags & BPF_F_INDEX_MASK;
350 	struct bpf_event_entry *ee;
351 
352 	if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
353 		return -EINVAL;
354 	if (index == BPF_F_CURRENT_CPU)
355 		index = cpu;
356 	if (unlikely(index >= array->map.max_entries))
357 		return -E2BIG;
358 
359 	ee = READ_ONCE(array->ptrs[index]);
360 	if (!ee)
361 		return -ENOENT;
362 
363 	return perf_event_read_local(ee->event, value, enabled, running);
364 }
365 
366 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
367 {
368 	u64 value = 0;
369 	int err;
370 
371 	err = get_map_perf_counter(map, flags, &value, NULL, NULL);
372 	/*
373 	 * this api is ugly since we miss [-22..-2] range of valid
374 	 * counter values, but that's uapi
375 	 */
376 	if (err)
377 		return err;
378 	return value;
379 }
380 
381 static const struct bpf_func_proto bpf_perf_event_read_proto = {
382 	.func		= bpf_perf_event_read,
383 	.gpl_only	= true,
384 	.ret_type	= RET_INTEGER,
385 	.arg1_type	= ARG_CONST_MAP_PTR,
386 	.arg2_type	= ARG_ANYTHING,
387 };
388 
389 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
390 	   struct bpf_perf_event_value *, buf, u32, size)
391 {
392 	int err = -EINVAL;
393 
394 	if (unlikely(size != sizeof(struct bpf_perf_event_value)))
395 		goto clear;
396 	err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
397 				   &buf->running);
398 	if (unlikely(err))
399 		goto clear;
400 	return 0;
401 clear:
402 	memset(buf, 0, size);
403 	return err;
404 }
405 
406 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
407 	.func		= bpf_perf_event_read_value,
408 	.gpl_only	= true,
409 	.ret_type	= RET_INTEGER,
410 	.arg1_type	= ARG_CONST_MAP_PTR,
411 	.arg2_type	= ARG_ANYTHING,
412 	.arg3_type	= ARG_PTR_TO_UNINIT_MEM,
413 	.arg4_type	= ARG_CONST_SIZE,
414 };
415 
416 static DEFINE_PER_CPU(struct perf_sample_data, bpf_trace_sd);
417 
418 static __always_inline u64
419 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
420 			u64 flags, struct perf_sample_data *sd)
421 {
422 	struct bpf_array *array = container_of(map, struct bpf_array, map);
423 	unsigned int cpu = smp_processor_id();
424 	u64 index = flags & BPF_F_INDEX_MASK;
425 	struct bpf_event_entry *ee;
426 	struct perf_event *event;
427 
428 	if (index == BPF_F_CURRENT_CPU)
429 		index = cpu;
430 	if (unlikely(index >= array->map.max_entries))
431 		return -E2BIG;
432 
433 	ee = READ_ONCE(array->ptrs[index]);
434 	if (!ee)
435 		return -ENOENT;
436 
437 	event = ee->event;
438 	if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
439 		     event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
440 		return -EINVAL;
441 
442 	if (unlikely(event->oncpu != cpu))
443 		return -EOPNOTSUPP;
444 
445 	return perf_event_output(event, sd, regs);
446 }
447 
448 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
449 	   u64, flags, void *, data, u64, size)
450 {
451 	struct perf_sample_data *sd = this_cpu_ptr(&bpf_trace_sd);
452 	struct perf_raw_record raw = {
453 		.frag = {
454 			.size = size,
455 			.data = data,
456 		},
457 	};
458 
459 	if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
460 		return -EINVAL;
461 
462 	perf_sample_data_init(sd, 0, 0);
463 	sd->raw = &raw;
464 
465 	return __bpf_perf_event_output(regs, map, flags, sd);
466 }
467 
468 static const struct bpf_func_proto bpf_perf_event_output_proto = {
469 	.func		= bpf_perf_event_output,
470 	.gpl_only	= true,
471 	.ret_type	= RET_INTEGER,
472 	.arg1_type	= ARG_PTR_TO_CTX,
473 	.arg2_type	= ARG_CONST_MAP_PTR,
474 	.arg3_type	= ARG_ANYTHING,
475 	.arg4_type	= ARG_PTR_TO_MEM,
476 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
477 };
478 
479 static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs);
480 static DEFINE_PER_CPU(struct perf_sample_data, bpf_misc_sd);
481 
482 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
483 		     void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
484 {
485 	struct perf_sample_data *sd = this_cpu_ptr(&bpf_misc_sd);
486 	struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs);
487 	struct perf_raw_frag frag = {
488 		.copy		= ctx_copy,
489 		.size		= ctx_size,
490 		.data		= ctx,
491 	};
492 	struct perf_raw_record raw = {
493 		.frag = {
494 			{
495 				.next	= ctx_size ? &frag : NULL,
496 			},
497 			.size	= meta_size,
498 			.data	= meta,
499 		},
500 	};
501 
502 	perf_fetch_caller_regs(regs);
503 	perf_sample_data_init(sd, 0, 0);
504 	sd->raw = &raw;
505 
506 	return __bpf_perf_event_output(regs, map, flags, sd);
507 }
508 
509 BPF_CALL_0(bpf_get_current_task)
510 {
511 	return (long) current;
512 }
513 
514 static const struct bpf_func_proto bpf_get_current_task_proto = {
515 	.func		= bpf_get_current_task,
516 	.gpl_only	= true,
517 	.ret_type	= RET_INTEGER,
518 };
519 
520 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
521 {
522 	struct bpf_array *array = container_of(map, struct bpf_array, map);
523 	struct cgroup *cgrp;
524 
525 	if (unlikely(idx >= array->map.max_entries))
526 		return -E2BIG;
527 
528 	cgrp = READ_ONCE(array->ptrs[idx]);
529 	if (unlikely(!cgrp))
530 		return -EAGAIN;
531 
532 	return task_under_cgroup_hierarchy(current, cgrp);
533 }
534 
535 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
536 	.func           = bpf_current_task_under_cgroup,
537 	.gpl_only       = false,
538 	.ret_type       = RET_INTEGER,
539 	.arg1_type      = ARG_CONST_MAP_PTR,
540 	.arg2_type      = ARG_ANYTHING,
541 };
542 
543 BPF_CALL_3(bpf_probe_read_str, void *, dst, u32, size,
544 	   const void *, unsafe_ptr)
545 {
546 	int ret;
547 
548 	/*
549 	 * The strncpy_from_unsafe() call will likely not fill the entire
550 	 * buffer, but that's okay in this circumstance as we're probing
551 	 * arbitrary memory anyway similar to bpf_probe_read() and might
552 	 * as well probe the stack. Thus, memory is explicitly cleared
553 	 * only in error case, so that improper users ignoring return
554 	 * code altogether don't copy garbage; otherwise length of string
555 	 * is returned that can be used for bpf_perf_event_output() et al.
556 	 */
557 	ret = strncpy_from_unsafe(dst, unsafe_ptr, size);
558 	if (unlikely(ret < 0))
559 		memset(dst, 0, size);
560 
561 	return ret;
562 }
563 
564 static const struct bpf_func_proto bpf_probe_read_str_proto = {
565 	.func		= bpf_probe_read_str,
566 	.gpl_only	= true,
567 	.ret_type	= RET_INTEGER,
568 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
569 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
570 	.arg3_type	= ARG_ANYTHING,
571 };
572 
573 struct send_signal_irq_work {
574 	struct irq_work irq_work;
575 	struct task_struct *task;
576 	u32 sig;
577 };
578 
579 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
580 
581 static void do_bpf_send_signal(struct irq_work *entry)
582 {
583 	struct send_signal_irq_work *work;
584 
585 	work = container_of(entry, struct send_signal_irq_work, irq_work);
586 	group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, PIDTYPE_TGID);
587 }
588 
589 BPF_CALL_1(bpf_send_signal, u32, sig)
590 {
591 	struct send_signal_irq_work *work = NULL;
592 
593 	/* Similar to bpf_probe_write_user, task needs to be
594 	 * in a sound condition and kernel memory access be
595 	 * permitted in order to send signal to the current
596 	 * task.
597 	 */
598 	if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
599 		return -EPERM;
600 	if (unlikely(uaccess_kernel()))
601 		return -EPERM;
602 	if (unlikely(!nmi_uaccess_okay()))
603 		return -EPERM;
604 
605 	if (in_nmi()) {
606 		/* Do an early check on signal validity. Otherwise,
607 		 * the error is lost in deferred irq_work.
608 		 */
609 		if (unlikely(!valid_signal(sig)))
610 			return -EINVAL;
611 
612 		work = this_cpu_ptr(&send_signal_work);
613 		if (work->irq_work.flags & IRQ_WORK_BUSY)
614 			return -EBUSY;
615 
616 		/* Add the current task, which is the target of sending signal,
617 		 * to the irq_work. The current task may change when queued
618 		 * irq works get executed.
619 		 */
620 		work->task = current;
621 		work->sig = sig;
622 		irq_work_queue(&work->irq_work);
623 		return 0;
624 	}
625 
626 	return group_send_sig_info(sig, SEND_SIG_PRIV, current, PIDTYPE_TGID);
627 }
628 
629 static const struct bpf_func_proto bpf_send_signal_proto = {
630 	.func		= bpf_send_signal,
631 	.gpl_only	= false,
632 	.ret_type	= RET_INTEGER,
633 	.arg1_type	= ARG_ANYTHING,
634 };
635 
636 static const struct bpf_func_proto *
637 tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
638 {
639 	switch (func_id) {
640 	case BPF_FUNC_map_lookup_elem:
641 		return &bpf_map_lookup_elem_proto;
642 	case BPF_FUNC_map_update_elem:
643 		return &bpf_map_update_elem_proto;
644 	case BPF_FUNC_map_delete_elem:
645 		return &bpf_map_delete_elem_proto;
646 	case BPF_FUNC_map_push_elem:
647 		return &bpf_map_push_elem_proto;
648 	case BPF_FUNC_map_pop_elem:
649 		return &bpf_map_pop_elem_proto;
650 	case BPF_FUNC_map_peek_elem:
651 		return &bpf_map_peek_elem_proto;
652 	case BPF_FUNC_probe_read:
653 		return &bpf_probe_read_proto;
654 	case BPF_FUNC_ktime_get_ns:
655 		return &bpf_ktime_get_ns_proto;
656 	case BPF_FUNC_tail_call:
657 		return &bpf_tail_call_proto;
658 	case BPF_FUNC_get_current_pid_tgid:
659 		return &bpf_get_current_pid_tgid_proto;
660 	case BPF_FUNC_get_current_task:
661 		return &bpf_get_current_task_proto;
662 	case BPF_FUNC_get_current_uid_gid:
663 		return &bpf_get_current_uid_gid_proto;
664 	case BPF_FUNC_get_current_comm:
665 		return &bpf_get_current_comm_proto;
666 	case BPF_FUNC_trace_printk:
667 		return bpf_get_trace_printk_proto();
668 	case BPF_FUNC_get_smp_processor_id:
669 		return &bpf_get_smp_processor_id_proto;
670 	case BPF_FUNC_get_numa_node_id:
671 		return &bpf_get_numa_node_id_proto;
672 	case BPF_FUNC_perf_event_read:
673 		return &bpf_perf_event_read_proto;
674 	case BPF_FUNC_probe_write_user:
675 		return bpf_get_probe_write_proto();
676 	case BPF_FUNC_current_task_under_cgroup:
677 		return &bpf_current_task_under_cgroup_proto;
678 	case BPF_FUNC_get_prandom_u32:
679 		return &bpf_get_prandom_u32_proto;
680 	case BPF_FUNC_probe_read_str:
681 		return &bpf_probe_read_str_proto;
682 #ifdef CONFIG_CGROUPS
683 	case BPF_FUNC_get_current_cgroup_id:
684 		return &bpf_get_current_cgroup_id_proto;
685 #endif
686 	case BPF_FUNC_send_signal:
687 		return &bpf_send_signal_proto;
688 	default:
689 		return NULL;
690 	}
691 }
692 
693 static const struct bpf_func_proto *
694 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
695 {
696 	switch (func_id) {
697 	case BPF_FUNC_perf_event_output:
698 		return &bpf_perf_event_output_proto;
699 	case BPF_FUNC_get_stackid:
700 		return &bpf_get_stackid_proto;
701 	case BPF_FUNC_get_stack:
702 		return &bpf_get_stack_proto;
703 	case BPF_FUNC_perf_event_read_value:
704 		return &bpf_perf_event_read_value_proto;
705 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
706 	case BPF_FUNC_override_return:
707 		return &bpf_override_return_proto;
708 #endif
709 	default:
710 		return tracing_func_proto(func_id, prog);
711 	}
712 }
713 
714 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
715 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
716 					const struct bpf_prog *prog,
717 					struct bpf_insn_access_aux *info)
718 {
719 	if (off < 0 || off >= sizeof(struct pt_regs))
720 		return false;
721 	if (type != BPF_READ)
722 		return false;
723 	if (off % size != 0)
724 		return false;
725 	/*
726 	 * Assertion for 32 bit to make sure last 8 byte access
727 	 * (BPF_DW) to the last 4 byte member is disallowed.
728 	 */
729 	if (off + size > sizeof(struct pt_regs))
730 		return false;
731 
732 	return true;
733 }
734 
735 const struct bpf_verifier_ops kprobe_verifier_ops = {
736 	.get_func_proto  = kprobe_prog_func_proto,
737 	.is_valid_access = kprobe_prog_is_valid_access,
738 };
739 
740 const struct bpf_prog_ops kprobe_prog_ops = {
741 };
742 
743 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
744 	   u64, flags, void *, data, u64, size)
745 {
746 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
747 
748 	/*
749 	 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
750 	 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
751 	 * from there and call the same bpf_perf_event_output() helper inline.
752 	 */
753 	return ____bpf_perf_event_output(regs, map, flags, data, size);
754 }
755 
756 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
757 	.func		= bpf_perf_event_output_tp,
758 	.gpl_only	= true,
759 	.ret_type	= RET_INTEGER,
760 	.arg1_type	= ARG_PTR_TO_CTX,
761 	.arg2_type	= ARG_CONST_MAP_PTR,
762 	.arg3_type	= ARG_ANYTHING,
763 	.arg4_type	= ARG_PTR_TO_MEM,
764 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
765 };
766 
767 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
768 	   u64, flags)
769 {
770 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
771 
772 	/*
773 	 * Same comment as in bpf_perf_event_output_tp(), only that this time
774 	 * the other helper's function body cannot be inlined due to being
775 	 * external, thus we need to call raw helper function.
776 	 */
777 	return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
778 			       flags, 0, 0);
779 }
780 
781 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
782 	.func		= bpf_get_stackid_tp,
783 	.gpl_only	= true,
784 	.ret_type	= RET_INTEGER,
785 	.arg1_type	= ARG_PTR_TO_CTX,
786 	.arg2_type	= ARG_CONST_MAP_PTR,
787 	.arg3_type	= ARG_ANYTHING,
788 };
789 
790 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
791 	   u64, flags)
792 {
793 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
794 
795 	return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
796 			     (unsigned long) size, flags, 0);
797 }
798 
799 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
800 	.func		= bpf_get_stack_tp,
801 	.gpl_only	= true,
802 	.ret_type	= RET_INTEGER,
803 	.arg1_type	= ARG_PTR_TO_CTX,
804 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
805 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
806 	.arg4_type	= ARG_ANYTHING,
807 };
808 
809 static const struct bpf_func_proto *
810 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
811 {
812 	switch (func_id) {
813 	case BPF_FUNC_perf_event_output:
814 		return &bpf_perf_event_output_proto_tp;
815 	case BPF_FUNC_get_stackid:
816 		return &bpf_get_stackid_proto_tp;
817 	case BPF_FUNC_get_stack:
818 		return &bpf_get_stack_proto_tp;
819 	default:
820 		return tracing_func_proto(func_id, prog);
821 	}
822 }
823 
824 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
825 				    const struct bpf_prog *prog,
826 				    struct bpf_insn_access_aux *info)
827 {
828 	if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
829 		return false;
830 	if (type != BPF_READ)
831 		return false;
832 	if (off % size != 0)
833 		return false;
834 
835 	BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
836 	return true;
837 }
838 
839 const struct bpf_verifier_ops tracepoint_verifier_ops = {
840 	.get_func_proto  = tp_prog_func_proto,
841 	.is_valid_access = tp_prog_is_valid_access,
842 };
843 
844 const struct bpf_prog_ops tracepoint_prog_ops = {
845 };
846 
847 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
848 	   struct bpf_perf_event_value *, buf, u32, size)
849 {
850 	int err = -EINVAL;
851 
852 	if (unlikely(size != sizeof(struct bpf_perf_event_value)))
853 		goto clear;
854 	err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
855 				    &buf->running);
856 	if (unlikely(err))
857 		goto clear;
858 	return 0;
859 clear:
860 	memset(buf, 0, size);
861 	return err;
862 }
863 
864 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
865          .func           = bpf_perf_prog_read_value,
866          .gpl_only       = true,
867          .ret_type       = RET_INTEGER,
868          .arg1_type      = ARG_PTR_TO_CTX,
869          .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
870          .arg3_type      = ARG_CONST_SIZE,
871 };
872 
873 static const struct bpf_func_proto *
874 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
875 {
876 	switch (func_id) {
877 	case BPF_FUNC_perf_event_output:
878 		return &bpf_perf_event_output_proto_tp;
879 	case BPF_FUNC_get_stackid:
880 		return &bpf_get_stackid_proto_tp;
881 	case BPF_FUNC_get_stack:
882 		return &bpf_get_stack_proto_tp;
883 	case BPF_FUNC_perf_prog_read_value:
884 		return &bpf_perf_prog_read_value_proto;
885 	default:
886 		return tracing_func_proto(func_id, prog);
887 	}
888 }
889 
890 /*
891  * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
892  * to avoid potential recursive reuse issue when/if tracepoints are added
893  * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack
894  */
895 static DEFINE_PER_CPU(struct pt_regs, bpf_raw_tp_regs);
896 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
897 	   struct bpf_map *, map, u64, flags, void *, data, u64, size)
898 {
899 	struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
900 
901 	perf_fetch_caller_regs(regs);
902 	return ____bpf_perf_event_output(regs, map, flags, data, size);
903 }
904 
905 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
906 	.func		= bpf_perf_event_output_raw_tp,
907 	.gpl_only	= true,
908 	.ret_type	= RET_INTEGER,
909 	.arg1_type	= ARG_PTR_TO_CTX,
910 	.arg2_type	= ARG_CONST_MAP_PTR,
911 	.arg3_type	= ARG_ANYTHING,
912 	.arg4_type	= ARG_PTR_TO_MEM,
913 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
914 };
915 
916 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
917 	   struct bpf_map *, map, u64, flags)
918 {
919 	struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
920 
921 	perf_fetch_caller_regs(regs);
922 	/* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
923 	return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
924 			       flags, 0, 0);
925 }
926 
927 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
928 	.func		= bpf_get_stackid_raw_tp,
929 	.gpl_only	= true,
930 	.ret_type	= RET_INTEGER,
931 	.arg1_type	= ARG_PTR_TO_CTX,
932 	.arg2_type	= ARG_CONST_MAP_PTR,
933 	.arg3_type	= ARG_ANYTHING,
934 };
935 
936 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
937 	   void *, buf, u32, size, u64, flags)
938 {
939 	struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
940 
941 	perf_fetch_caller_regs(regs);
942 	return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
943 			     (unsigned long) size, flags, 0);
944 }
945 
946 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
947 	.func		= bpf_get_stack_raw_tp,
948 	.gpl_only	= true,
949 	.ret_type	= RET_INTEGER,
950 	.arg1_type	= ARG_PTR_TO_CTX,
951 	.arg2_type	= ARG_PTR_TO_MEM,
952 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
953 	.arg4_type	= ARG_ANYTHING,
954 };
955 
956 static const struct bpf_func_proto *
957 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
958 {
959 	switch (func_id) {
960 	case BPF_FUNC_perf_event_output:
961 		return &bpf_perf_event_output_proto_raw_tp;
962 	case BPF_FUNC_get_stackid:
963 		return &bpf_get_stackid_proto_raw_tp;
964 	case BPF_FUNC_get_stack:
965 		return &bpf_get_stack_proto_raw_tp;
966 	default:
967 		return tracing_func_proto(func_id, prog);
968 	}
969 }
970 
971 static bool raw_tp_prog_is_valid_access(int off, int size,
972 					enum bpf_access_type type,
973 					const struct bpf_prog *prog,
974 					struct bpf_insn_access_aux *info)
975 {
976 	/* largest tracepoint in the kernel has 12 args */
977 	if (off < 0 || off >= sizeof(__u64) * 12)
978 		return false;
979 	if (type != BPF_READ)
980 		return false;
981 	if (off % size != 0)
982 		return false;
983 	return true;
984 }
985 
986 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
987 	.get_func_proto  = raw_tp_prog_func_proto,
988 	.is_valid_access = raw_tp_prog_is_valid_access,
989 };
990 
991 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
992 };
993 
994 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
995 						 enum bpf_access_type type,
996 						 const struct bpf_prog *prog,
997 						 struct bpf_insn_access_aux *info)
998 {
999 	if (off == 0) {
1000 		if (size != sizeof(u64) || type != BPF_READ)
1001 			return false;
1002 		info->reg_type = PTR_TO_TP_BUFFER;
1003 	}
1004 	return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1005 }
1006 
1007 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1008 	.get_func_proto  = raw_tp_prog_func_proto,
1009 	.is_valid_access = raw_tp_writable_prog_is_valid_access,
1010 };
1011 
1012 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1013 };
1014 
1015 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1016 				    const struct bpf_prog *prog,
1017 				    struct bpf_insn_access_aux *info)
1018 {
1019 	const int size_u64 = sizeof(u64);
1020 
1021 	if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1022 		return false;
1023 	if (type != BPF_READ)
1024 		return false;
1025 	if (off % size != 0) {
1026 		if (sizeof(unsigned long) != 4)
1027 			return false;
1028 		if (size != 8)
1029 			return false;
1030 		if (off % size != 4)
1031 			return false;
1032 	}
1033 
1034 	switch (off) {
1035 	case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1036 		bpf_ctx_record_field_size(info, size_u64);
1037 		if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1038 			return false;
1039 		break;
1040 	case bpf_ctx_range(struct bpf_perf_event_data, addr):
1041 		bpf_ctx_record_field_size(info, size_u64);
1042 		if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1043 			return false;
1044 		break;
1045 	default:
1046 		if (size != sizeof(long))
1047 			return false;
1048 	}
1049 
1050 	return true;
1051 }
1052 
1053 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1054 				      const struct bpf_insn *si,
1055 				      struct bpf_insn *insn_buf,
1056 				      struct bpf_prog *prog, u32 *target_size)
1057 {
1058 	struct bpf_insn *insn = insn_buf;
1059 
1060 	switch (si->off) {
1061 	case offsetof(struct bpf_perf_event_data, sample_period):
1062 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1063 						       data), si->dst_reg, si->src_reg,
1064 				      offsetof(struct bpf_perf_event_data_kern, data));
1065 		*insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1066 				      bpf_target_off(struct perf_sample_data, period, 8,
1067 						     target_size));
1068 		break;
1069 	case offsetof(struct bpf_perf_event_data, addr):
1070 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1071 						       data), si->dst_reg, si->src_reg,
1072 				      offsetof(struct bpf_perf_event_data_kern, data));
1073 		*insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1074 				      bpf_target_off(struct perf_sample_data, addr, 8,
1075 						     target_size));
1076 		break;
1077 	default:
1078 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1079 						       regs), si->dst_reg, si->src_reg,
1080 				      offsetof(struct bpf_perf_event_data_kern, regs));
1081 		*insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1082 				      si->off);
1083 		break;
1084 	}
1085 
1086 	return insn - insn_buf;
1087 }
1088 
1089 const struct bpf_verifier_ops perf_event_verifier_ops = {
1090 	.get_func_proto		= pe_prog_func_proto,
1091 	.is_valid_access	= pe_prog_is_valid_access,
1092 	.convert_ctx_access	= pe_prog_convert_ctx_access,
1093 };
1094 
1095 const struct bpf_prog_ops perf_event_prog_ops = {
1096 };
1097 
1098 static DEFINE_MUTEX(bpf_event_mutex);
1099 
1100 #define BPF_TRACE_MAX_PROGS 64
1101 
1102 int perf_event_attach_bpf_prog(struct perf_event *event,
1103 			       struct bpf_prog *prog)
1104 {
1105 	struct bpf_prog_array *old_array;
1106 	struct bpf_prog_array *new_array;
1107 	int ret = -EEXIST;
1108 
1109 	/*
1110 	 * Kprobe override only works if they are on the function entry,
1111 	 * and only if they are on the opt-in list.
1112 	 */
1113 	if (prog->kprobe_override &&
1114 	    (!trace_kprobe_on_func_entry(event->tp_event) ||
1115 	     !trace_kprobe_error_injectable(event->tp_event)))
1116 		return -EINVAL;
1117 
1118 	mutex_lock(&bpf_event_mutex);
1119 
1120 	if (event->prog)
1121 		goto unlock;
1122 
1123 	old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1124 	if (old_array &&
1125 	    bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1126 		ret = -E2BIG;
1127 		goto unlock;
1128 	}
1129 
1130 	ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1131 	if (ret < 0)
1132 		goto unlock;
1133 
1134 	/* set the new array to event->tp_event and set event->prog */
1135 	event->prog = prog;
1136 	rcu_assign_pointer(event->tp_event->prog_array, new_array);
1137 	bpf_prog_array_free(old_array);
1138 
1139 unlock:
1140 	mutex_unlock(&bpf_event_mutex);
1141 	return ret;
1142 }
1143 
1144 void perf_event_detach_bpf_prog(struct perf_event *event)
1145 {
1146 	struct bpf_prog_array *old_array;
1147 	struct bpf_prog_array *new_array;
1148 	int ret;
1149 
1150 	mutex_lock(&bpf_event_mutex);
1151 
1152 	if (!event->prog)
1153 		goto unlock;
1154 
1155 	old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1156 	ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1157 	if (ret == -ENOENT)
1158 		goto unlock;
1159 	if (ret < 0) {
1160 		bpf_prog_array_delete_safe(old_array, event->prog);
1161 	} else {
1162 		rcu_assign_pointer(event->tp_event->prog_array, new_array);
1163 		bpf_prog_array_free(old_array);
1164 	}
1165 
1166 	bpf_prog_put(event->prog);
1167 	event->prog = NULL;
1168 
1169 unlock:
1170 	mutex_unlock(&bpf_event_mutex);
1171 }
1172 
1173 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1174 {
1175 	struct perf_event_query_bpf __user *uquery = info;
1176 	struct perf_event_query_bpf query = {};
1177 	struct bpf_prog_array *progs;
1178 	u32 *ids, prog_cnt, ids_len;
1179 	int ret;
1180 
1181 	if (!capable(CAP_SYS_ADMIN))
1182 		return -EPERM;
1183 	if (event->attr.type != PERF_TYPE_TRACEPOINT)
1184 		return -EINVAL;
1185 	if (copy_from_user(&query, uquery, sizeof(query)))
1186 		return -EFAULT;
1187 
1188 	ids_len = query.ids_len;
1189 	if (ids_len > BPF_TRACE_MAX_PROGS)
1190 		return -E2BIG;
1191 	ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1192 	if (!ids)
1193 		return -ENOMEM;
1194 	/*
1195 	 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1196 	 * is required when user only wants to check for uquery->prog_cnt.
1197 	 * There is no need to check for it since the case is handled
1198 	 * gracefully in bpf_prog_array_copy_info.
1199 	 */
1200 
1201 	mutex_lock(&bpf_event_mutex);
1202 	progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
1203 	ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
1204 	mutex_unlock(&bpf_event_mutex);
1205 
1206 	if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1207 	    copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1208 		ret = -EFAULT;
1209 
1210 	kfree(ids);
1211 	return ret;
1212 }
1213 
1214 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1215 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1216 
1217 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1218 {
1219 	struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1220 
1221 	for (; btp < __stop__bpf_raw_tp; btp++) {
1222 		if (!strcmp(btp->tp->name, name))
1223 			return btp;
1224 	}
1225 
1226 	return bpf_get_raw_tracepoint_module(name);
1227 }
1228 
1229 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1230 {
1231 	struct module *mod = __module_address((unsigned long)btp);
1232 
1233 	if (mod)
1234 		module_put(mod);
1235 }
1236 
1237 static __always_inline
1238 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1239 {
1240 	rcu_read_lock();
1241 	preempt_disable();
1242 	(void) BPF_PROG_RUN(prog, args);
1243 	preempt_enable();
1244 	rcu_read_unlock();
1245 }
1246 
1247 #define UNPACK(...)			__VA_ARGS__
1248 #define REPEAT_1(FN, DL, X, ...)	FN(X)
1249 #define REPEAT_2(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1250 #define REPEAT_3(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1251 #define REPEAT_4(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1252 #define REPEAT_5(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1253 #define REPEAT_6(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1254 #define REPEAT_7(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1255 #define REPEAT_8(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1256 #define REPEAT_9(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1257 #define REPEAT_10(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1258 #define REPEAT_11(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1259 #define REPEAT_12(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1260 #define REPEAT(X, FN, DL, ...)		REPEAT_##X(FN, DL, __VA_ARGS__)
1261 
1262 #define SARG(X)		u64 arg##X
1263 #define COPY(X)		args[X] = arg##X
1264 
1265 #define __DL_COM	(,)
1266 #define __DL_SEM	(;)
1267 
1268 #define __SEQ_0_11	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1269 
1270 #define BPF_TRACE_DEFN_x(x)						\
1271 	void bpf_trace_run##x(struct bpf_prog *prog,			\
1272 			      REPEAT(x, SARG, __DL_COM, __SEQ_0_11))	\
1273 	{								\
1274 		u64 args[x];						\
1275 		REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);			\
1276 		__bpf_trace_run(prog, args);				\
1277 	}								\
1278 	EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1279 BPF_TRACE_DEFN_x(1);
1280 BPF_TRACE_DEFN_x(2);
1281 BPF_TRACE_DEFN_x(3);
1282 BPF_TRACE_DEFN_x(4);
1283 BPF_TRACE_DEFN_x(5);
1284 BPF_TRACE_DEFN_x(6);
1285 BPF_TRACE_DEFN_x(7);
1286 BPF_TRACE_DEFN_x(8);
1287 BPF_TRACE_DEFN_x(9);
1288 BPF_TRACE_DEFN_x(10);
1289 BPF_TRACE_DEFN_x(11);
1290 BPF_TRACE_DEFN_x(12);
1291 
1292 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1293 {
1294 	struct tracepoint *tp = btp->tp;
1295 
1296 	/*
1297 	 * check that program doesn't access arguments beyond what's
1298 	 * available in this tracepoint
1299 	 */
1300 	if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1301 		return -EINVAL;
1302 
1303 	if (prog->aux->max_tp_access > btp->writable_size)
1304 		return -EINVAL;
1305 
1306 	return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
1307 }
1308 
1309 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1310 {
1311 	return __bpf_probe_register(btp, prog);
1312 }
1313 
1314 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1315 {
1316 	return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1317 }
1318 
1319 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
1320 			    u32 *fd_type, const char **buf,
1321 			    u64 *probe_offset, u64 *probe_addr)
1322 {
1323 	bool is_tracepoint, is_syscall_tp;
1324 	struct bpf_prog *prog;
1325 	int flags, err = 0;
1326 
1327 	prog = event->prog;
1328 	if (!prog)
1329 		return -ENOENT;
1330 
1331 	/* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
1332 	if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
1333 		return -EOPNOTSUPP;
1334 
1335 	*prog_id = prog->aux->id;
1336 	flags = event->tp_event->flags;
1337 	is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
1338 	is_syscall_tp = is_syscall_trace_event(event->tp_event);
1339 
1340 	if (is_tracepoint || is_syscall_tp) {
1341 		*buf = is_tracepoint ? event->tp_event->tp->name
1342 				     : event->tp_event->name;
1343 		*fd_type = BPF_FD_TYPE_TRACEPOINT;
1344 		*probe_offset = 0x0;
1345 		*probe_addr = 0x0;
1346 	} else {
1347 		/* kprobe/uprobe */
1348 		err = -EOPNOTSUPP;
1349 #ifdef CONFIG_KPROBE_EVENTS
1350 		if (flags & TRACE_EVENT_FL_KPROBE)
1351 			err = bpf_get_kprobe_info(event, fd_type, buf,
1352 						  probe_offset, probe_addr,
1353 						  event->attr.type == PERF_TYPE_TRACEPOINT);
1354 #endif
1355 #ifdef CONFIG_UPROBE_EVENTS
1356 		if (flags & TRACE_EVENT_FL_UPROBE)
1357 			err = bpf_get_uprobe_info(event, fd_type, buf,
1358 						  probe_offset,
1359 						  event->attr.type == PERF_TYPE_TRACEPOINT);
1360 #endif
1361 	}
1362 
1363 	return err;
1364 }
1365 
1366 #ifdef CONFIG_MODULES
1367 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
1368 			    void *module)
1369 {
1370 	struct bpf_trace_module *btm, *tmp;
1371 	struct module *mod = module;
1372 
1373 	if (mod->num_bpf_raw_events == 0 ||
1374 	    (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
1375 		return 0;
1376 
1377 	mutex_lock(&bpf_module_mutex);
1378 
1379 	switch (op) {
1380 	case MODULE_STATE_COMING:
1381 		btm = kzalloc(sizeof(*btm), GFP_KERNEL);
1382 		if (btm) {
1383 			btm->module = module;
1384 			list_add(&btm->list, &bpf_trace_modules);
1385 		}
1386 		break;
1387 	case MODULE_STATE_GOING:
1388 		list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
1389 			if (btm->module == module) {
1390 				list_del(&btm->list);
1391 				kfree(btm);
1392 				break;
1393 			}
1394 		}
1395 		break;
1396 	}
1397 
1398 	mutex_unlock(&bpf_module_mutex);
1399 
1400 	return 0;
1401 }
1402 
1403 static struct notifier_block bpf_module_nb = {
1404 	.notifier_call = bpf_event_notify,
1405 };
1406 
1407 static int __init bpf_event_init(void)
1408 {
1409 	register_module_notifier(&bpf_module_nb);
1410 	return 0;
1411 }
1412 
1413 static int __init send_signal_irq_work_init(void)
1414 {
1415 	int cpu;
1416 	struct send_signal_irq_work *work;
1417 
1418 	for_each_possible_cpu(cpu) {
1419 		work = per_cpu_ptr(&send_signal_work, cpu);
1420 		init_irq_work(&work->irq_work, do_bpf_send_signal);
1421 	}
1422 	return 0;
1423 }
1424 
1425 fs_initcall(bpf_event_init);
1426 subsys_initcall(send_signal_irq_work_init);
1427 #endif /* CONFIG_MODULES */
1428