xref: /openbmc/linux/kernel/trace/trace_eprobe.c (revision 6d21fb7d)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * event probes
4  *
5  * Part of this code was copied from kernel/trace/trace_kprobe.c written by
6  * Masami Hiramatsu <mhiramat@kernel.org>
7  *
8  * Copyright (C) 2021, VMware Inc, Steven Rostedt <rostedt@goodmis.org>
9  * Copyright (C) 2021, VMware Inc, Tzvetomir Stoyanov tz.stoyanov@gmail.com>
10  *
11  */
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/ftrace.h>
15 
16 #include "trace_dynevent.h"
17 #include "trace_probe.h"
18 #include "trace_probe_tmpl.h"
19 #include "trace_probe_kernel.h"
20 
21 #define EPROBE_EVENT_SYSTEM "eprobes"
22 
23 struct trace_eprobe {
24 	/* tracepoint system */
25 	const char *event_system;
26 
27 	/* tracepoint event */
28 	const char *event_name;
29 
30 	/* filter string for the tracepoint */
31 	char *filter_str;
32 
33 	struct trace_event_call *event;
34 
35 	struct dyn_event	devent;
36 	struct trace_probe	tp;
37 };
38 
39 struct eprobe_data {
40 	struct trace_event_file	*file;
41 	struct trace_eprobe	*ep;
42 };
43 
44 
45 #define for_each_trace_eprobe_tp(ep, _tp) \
46 	list_for_each_entry(ep, trace_probe_probe_list(_tp), tp.list)
47 
48 static int __trace_eprobe_create(int argc, const char *argv[]);
49 
50 static void trace_event_probe_cleanup(struct trace_eprobe *ep)
51 {
52 	if (!ep)
53 		return;
54 	trace_probe_cleanup(&ep->tp);
55 	kfree(ep->event_name);
56 	kfree(ep->event_system);
57 	if (ep->event)
58 		trace_event_put_ref(ep->event);
59 	kfree(ep->filter_str);
60 	kfree(ep);
61 }
62 
63 static struct trace_eprobe *to_trace_eprobe(struct dyn_event *ev)
64 {
65 	return container_of(ev, struct trace_eprobe, devent);
66 }
67 
68 static int eprobe_dyn_event_create(const char *raw_command)
69 {
70 	return trace_probe_create(raw_command, __trace_eprobe_create);
71 }
72 
73 static int eprobe_dyn_event_show(struct seq_file *m, struct dyn_event *ev)
74 {
75 	struct trace_eprobe *ep = to_trace_eprobe(ev);
76 	int i;
77 
78 	seq_printf(m, "e:%s/%s", trace_probe_group_name(&ep->tp),
79 				trace_probe_name(&ep->tp));
80 	seq_printf(m, " %s.%s", ep->event_system, ep->event_name);
81 
82 	for (i = 0; i < ep->tp.nr_args; i++)
83 		seq_printf(m, " %s=%s", ep->tp.args[i].name, ep->tp.args[i].comm);
84 	seq_putc(m, '\n');
85 
86 	return 0;
87 }
88 
89 static int unregister_trace_eprobe(struct trace_eprobe *ep)
90 {
91 	/* If other probes are on the event, just unregister eprobe */
92 	if (trace_probe_has_sibling(&ep->tp))
93 		goto unreg;
94 
95 	/* Enabled event can not be unregistered */
96 	if (trace_probe_is_enabled(&ep->tp))
97 		return -EBUSY;
98 
99 	/* Will fail if probe is being used by ftrace or perf */
100 	if (trace_probe_unregister_event_call(&ep->tp))
101 		return -EBUSY;
102 
103 unreg:
104 	dyn_event_remove(&ep->devent);
105 	trace_probe_unlink(&ep->tp);
106 
107 	return 0;
108 }
109 
110 static int eprobe_dyn_event_release(struct dyn_event *ev)
111 {
112 	struct trace_eprobe *ep = to_trace_eprobe(ev);
113 	int ret = unregister_trace_eprobe(ep);
114 
115 	if (!ret)
116 		trace_event_probe_cleanup(ep);
117 	return ret;
118 }
119 
120 static bool eprobe_dyn_event_is_busy(struct dyn_event *ev)
121 {
122 	struct trace_eprobe *ep = to_trace_eprobe(ev);
123 
124 	return trace_probe_is_enabled(&ep->tp);
125 }
126 
127 static bool eprobe_dyn_event_match(const char *system, const char *event,
128 			int argc, const char **argv, struct dyn_event *ev)
129 {
130 	struct trace_eprobe *ep = to_trace_eprobe(ev);
131 	const char *slash;
132 
133 	/*
134 	 * We match the following:
135 	 *  event only			- match all eprobes with event name
136 	 *  system and event only	- match all system/event probes
137 	 *  system only			- match all system probes
138 	 *
139 	 * The below has the above satisfied with more arguments:
140 	 *
141 	 *  attached system/event	- If the arg has the system and event
142 	 *				  the probe is attached to, match
143 	 *				  probes with the attachment.
144 	 *
145 	 *  If any more args are given, then it requires a full match.
146 	 */
147 
148 	/*
149 	 * If system exists, but this probe is not part of that system
150 	 * do not match.
151 	 */
152 	if (system && strcmp(trace_probe_group_name(&ep->tp), system) != 0)
153 		return false;
154 
155 	/* Must match the event name */
156 	if (event[0] != '\0' && strcmp(trace_probe_name(&ep->tp), event) != 0)
157 		return false;
158 
159 	/* No arguments match all */
160 	if (argc < 1)
161 		return true;
162 
163 	/* First argument is the system/event the probe is attached to */
164 
165 	slash = strchr(argv[0], '/');
166 	if (!slash)
167 		slash = strchr(argv[0], '.');
168 	if (!slash)
169 		return false;
170 
171 	if (strncmp(ep->event_system, argv[0], slash - argv[0]))
172 		return false;
173 	if (strcmp(ep->event_name, slash + 1))
174 		return false;
175 
176 	argc--;
177 	argv++;
178 
179 	/* If there are no other args, then match */
180 	if (argc < 1)
181 		return true;
182 
183 	return trace_probe_match_command_args(&ep->tp, argc, argv);
184 }
185 
186 static struct dyn_event_operations eprobe_dyn_event_ops = {
187 	.create = eprobe_dyn_event_create,
188 	.show = eprobe_dyn_event_show,
189 	.is_busy = eprobe_dyn_event_is_busy,
190 	.free = eprobe_dyn_event_release,
191 	.match = eprobe_dyn_event_match,
192 };
193 
194 static struct trace_eprobe *alloc_event_probe(const char *group,
195 					      const char *this_event,
196 					      struct trace_event_call *event,
197 					      int nargs)
198 {
199 	struct trace_eprobe *ep;
200 	const char *event_name;
201 	const char *sys_name;
202 	int ret = -ENOMEM;
203 
204 	if (!event)
205 		return ERR_PTR(-ENODEV);
206 
207 	sys_name = event->class->system;
208 	event_name = trace_event_name(event);
209 
210 	ep = kzalloc(struct_size(ep, tp.args, nargs), GFP_KERNEL);
211 	if (!ep) {
212 		trace_event_put_ref(event);
213 		goto error;
214 	}
215 	ep->event = event;
216 	ep->event_name = kstrdup(event_name, GFP_KERNEL);
217 	if (!ep->event_name)
218 		goto error;
219 	ep->event_system = kstrdup(sys_name, GFP_KERNEL);
220 	if (!ep->event_system)
221 		goto error;
222 
223 	ret = trace_probe_init(&ep->tp, this_event, group, false);
224 	if (ret < 0)
225 		goto error;
226 
227 	dyn_event_init(&ep->devent, &eprobe_dyn_event_ops);
228 	return ep;
229 error:
230 	trace_event_probe_cleanup(ep);
231 	return ERR_PTR(ret);
232 }
233 
234 static int eprobe_event_define_fields(struct trace_event_call *event_call)
235 {
236 	struct eprobe_trace_entry_head field;
237 	struct trace_probe *tp;
238 
239 	tp = trace_probe_primary_from_call(event_call);
240 	if (WARN_ON_ONCE(!tp))
241 		return -ENOENT;
242 
243 	return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
244 }
245 
246 static struct trace_event_fields eprobe_fields_array[] = {
247 	{ .type = TRACE_FUNCTION_TYPE,
248 	  .define_fields = eprobe_event_define_fields },
249 	{}
250 };
251 
252 /* Event entry printers */
253 static enum print_line_t
254 print_eprobe_event(struct trace_iterator *iter, int flags,
255 		   struct trace_event *event)
256 {
257 	struct eprobe_trace_entry_head *field;
258 	struct trace_event_call *pevent;
259 	struct trace_event *probed_event;
260 	struct trace_seq *s = &iter->seq;
261 	struct trace_eprobe *ep;
262 	struct trace_probe *tp;
263 	unsigned int type;
264 
265 	field = (struct eprobe_trace_entry_head *)iter->ent;
266 	tp = trace_probe_primary_from_call(
267 		container_of(event, struct trace_event_call, event));
268 	if (WARN_ON_ONCE(!tp))
269 		goto out;
270 
271 	ep = container_of(tp, struct trace_eprobe, tp);
272 	type = ep->event->event.type;
273 
274 	trace_seq_printf(s, "%s: (", trace_probe_name(tp));
275 
276 	probed_event = ftrace_find_event(type);
277 	if (probed_event) {
278 		pevent = container_of(probed_event, struct trace_event_call, event);
279 		trace_seq_printf(s, "%s.%s", pevent->class->system,
280 				 trace_event_name(pevent));
281 	} else {
282 		trace_seq_printf(s, "%u", type);
283 	}
284 
285 	trace_seq_putc(s, ')');
286 
287 	if (trace_probe_print_args(s, tp->args, tp->nr_args,
288 			     (u8 *)&field[1], field) < 0)
289 		goto out;
290 
291 	trace_seq_putc(s, '\n');
292  out:
293 	return trace_handle_return(s);
294 }
295 
296 static nokprobe_inline unsigned long
297 get_event_field(struct fetch_insn *code, void *rec)
298 {
299 	struct ftrace_event_field *field = code->data;
300 	unsigned long val;
301 	void *addr;
302 
303 	addr = rec + field->offset;
304 
305 	if (is_string_field(field)) {
306 		switch (field->filter_type) {
307 		case FILTER_DYN_STRING:
308 			val = (unsigned long)(rec + (*(unsigned int *)addr & 0xffff));
309 			break;
310 		case FILTER_RDYN_STRING:
311 			val = (unsigned long)(addr + (*(unsigned int *)addr & 0xffff));
312 			break;
313 		case FILTER_STATIC_STRING:
314 			val = (unsigned long)addr;
315 			break;
316 		case FILTER_PTR_STRING:
317 			val = (unsigned long)(*(char *)addr);
318 			break;
319 		default:
320 			WARN_ON_ONCE(1);
321 			return 0;
322 		}
323 		return val;
324 	}
325 
326 	switch (field->size) {
327 	case 1:
328 		if (field->is_signed)
329 			val = *(char *)addr;
330 		else
331 			val = *(unsigned char *)addr;
332 		break;
333 	case 2:
334 		if (field->is_signed)
335 			val = *(short *)addr;
336 		else
337 			val = *(unsigned short *)addr;
338 		break;
339 	case 4:
340 		if (field->is_signed)
341 			val = *(int *)addr;
342 		else
343 			val = *(unsigned int *)addr;
344 		break;
345 	default:
346 		if (field->is_signed)
347 			val = *(long *)addr;
348 		else
349 			val = *(unsigned long *)addr;
350 		break;
351 	}
352 	return val;
353 }
354 
355 static int get_eprobe_size(struct trace_probe *tp, void *rec)
356 {
357 	struct fetch_insn *code;
358 	struct probe_arg *arg;
359 	int i, len, ret = 0;
360 
361 	for (i = 0; i < tp->nr_args; i++) {
362 		arg = tp->args + i;
363 		if (arg->dynamic) {
364 			unsigned long val;
365 
366 			code = arg->code;
367  retry:
368 			switch (code->op) {
369 			case FETCH_OP_TP_ARG:
370 				val = get_event_field(code, rec);
371 				break;
372 			case FETCH_NOP_SYMBOL:	/* Ignore a place holder */
373 				code++;
374 				goto retry;
375 			default:
376 				if (process_common_fetch_insn(code, &val) < 0)
377 					continue;
378 			}
379 			code++;
380 			len = process_fetch_insn_bottom(code, val, NULL, NULL);
381 			if (len > 0)
382 				ret += len;
383 		}
384 	}
385 
386 	return ret;
387 }
388 
389 /* Kprobe specific fetch functions */
390 
391 /* Note that we don't verify it, since the code does not come from user space */
392 static int
393 process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
394 		   void *base)
395 {
396 	unsigned long val;
397 	int ret;
398 
399  retry:
400 	switch (code->op) {
401 	case FETCH_OP_TP_ARG:
402 		val = get_event_field(code, rec);
403 		break;
404 	case FETCH_NOP_SYMBOL:	/* Ignore a place holder */
405 		code++;
406 		goto retry;
407 	default:
408 		ret = process_common_fetch_insn(code, &val);
409 		if (ret < 0)
410 			return ret;
411 	}
412 	code++;
413 	return process_fetch_insn_bottom(code, val, dest, base);
414 }
415 NOKPROBE_SYMBOL(process_fetch_insn)
416 
417 /* eprobe handler */
418 static inline void
419 __eprobe_trace_func(struct eprobe_data *edata, void *rec)
420 {
421 	struct eprobe_trace_entry_head *entry;
422 	struct trace_event_call *call = trace_probe_event_call(&edata->ep->tp);
423 	struct trace_event_buffer fbuffer;
424 	int dsize;
425 
426 	if (WARN_ON_ONCE(call != edata->file->event_call))
427 		return;
428 
429 	if (trace_trigger_soft_disabled(edata->file))
430 		return;
431 
432 	dsize = get_eprobe_size(&edata->ep->tp, rec);
433 
434 	entry = trace_event_buffer_reserve(&fbuffer, edata->file,
435 					   sizeof(*entry) + edata->ep->tp.size + dsize);
436 
437 	if (!entry)
438 		return;
439 
440 	entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
441 	store_trace_args(&entry[1], &edata->ep->tp, rec, sizeof(*entry), dsize);
442 
443 	trace_event_buffer_commit(&fbuffer);
444 }
445 
446 /*
447  * The event probe implementation uses event triggers to get access to
448  * the event it is attached to, but is not an actual trigger. The below
449  * functions are just stubs to fulfill what is needed to use the trigger
450  * infrastructure.
451  */
452 static int eprobe_trigger_init(struct event_trigger_data *data)
453 {
454 	return 0;
455 }
456 
457 static void eprobe_trigger_free(struct event_trigger_data *data)
458 {
459 
460 }
461 
462 static int eprobe_trigger_print(struct seq_file *m,
463 				struct event_trigger_data *data)
464 {
465 	/* Do not print eprobe event triggers */
466 	return 0;
467 }
468 
469 static void eprobe_trigger_func(struct event_trigger_data *data,
470 				struct trace_buffer *buffer, void *rec,
471 				struct ring_buffer_event *rbe)
472 {
473 	struct eprobe_data *edata = data->private_data;
474 
475 	if (unlikely(!rec))
476 		return;
477 
478 	__eprobe_trace_func(edata, rec);
479 }
480 
481 static struct event_trigger_ops eprobe_trigger_ops = {
482 	.trigger		= eprobe_trigger_func,
483 	.print			= eprobe_trigger_print,
484 	.init			= eprobe_trigger_init,
485 	.free			= eprobe_trigger_free,
486 };
487 
488 static int eprobe_trigger_cmd_parse(struct event_command *cmd_ops,
489 				    struct trace_event_file *file,
490 				    char *glob, char *cmd,
491 				    char *param_and_filter)
492 {
493 	return -1;
494 }
495 
496 static int eprobe_trigger_reg_func(char *glob,
497 				   struct event_trigger_data *data,
498 				   struct trace_event_file *file)
499 {
500 	return -1;
501 }
502 
503 static void eprobe_trigger_unreg_func(char *glob,
504 				      struct event_trigger_data *data,
505 				      struct trace_event_file *file)
506 {
507 
508 }
509 
510 static struct event_trigger_ops *eprobe_trigger_get_ops(char *cmd,
511 							char *param)
512 {
513 	return &eprobe_trigger_ops;
514 }
515 
516 static struct event_command event_trigger_cmd = {
517 	.name			= "eprobe",
518 	.trigger_type		= ETT_EVENT_EPROBE,
519 	.flags			= EVENT_CMD_FL_NEEDS_REC,
520 	.parse			= eprobe_trigger_cmd_parse,
521 	.reg			= eprobe_trigger_reg_func,
522 	.unreg			= eprobe_trigger_unreg_func,
523 	.unreg_all		= NULL,
524 	.get_trigger_ops	= eprobe_trigger_get_ops,
525 	.set_filter		= NULL,
526 };
527 
528 static struct event_trigger_data *
529 new_eprobe_trigger(struct trace_eprobe *ep, struct trace_event_file *file)
530 {
531 	struct event_trigger_data *trigger;
532 	struct event_filter *filter = NULL;
533 	struct eprobe_data *edata;
534 	int ret;
535 
536 	edata = kzalloc(sizeof(*edata), GFP_KERNEL);
537 	trigger = kzalloc(sizeof(*trigger), GFP_KERNEL);
538 	if (!trigger || !edata) {
539 		ret = -ENOMEM;
540 		goto error;
541 	}
542 
543 	trigger->flags = EVENT_TRIGGER_FL_PROBE;
544 	trigger->count = -1;
545 	trigger->ops = &eprobe_trigger_ops;
546 
547 	/*
548 	 * EVENT PROBE triggers are not registered as commands with
549 	 * register_event_command(), as they are not controlled by the user
550 	 * from the trigger file
551 	 */
552 	trigger->cmd_ops = &event_trigger_cmd;
553 
554 	INIT_LIST_HEAD(&trigger->list);
555 
556 	if (ep->filter_str) {
557 		ret = create_event_filter(file->tr, ep->event,
558 					ep->filter_str, false, &filter);
559 		if (ret)
560 			goto error;
561 	}
562 	RCU_INIT_POINTER(trigger->filter, filter);
563 
564 	edata->file = file;
565 	edata->ep = ep;
566 	trigger->private_data = edata;
567 
568 	return trigger;
569 error:
570 	free_event_filter(filter);
571 	kfree(edata);
572 	kfree(trigger);
573 	return ERR_PTR(ret);
574 }
575 
576 static int enable_eprobe(struct trace_eprobe *ep,
577 			 struct trace_event_file *eprobe_file)
578 {
579 	struct event_trigger_data *trigger;
580 	struct trace_event_file *file;
581 	struct trace_array *tr = eprobe_file->tr;
582 
583 	file = find_event_file(tr, ep->event_system, ep->event_name);
584 	if (!file)
585 		return -ENOENT;
586 	trigger = new_eprobe_trigger(ep, eprobe_file);
587 	if (IS_ERR(trigger))
588 		return PTR_ERR(trigger);
589 
590 	list_add_tail_rcu(&trigger->list, &file->triggers);
591 
592 	trace_event_trigger_enable_disable(file, 1);
593 	update_cond_flag(file);
594 
595 	return 0;
596 }
597 
598 static struct trace_event_functions eprobe_funcs = {
599 	.trace		= print_eprobe_event
600 };
601 
602 static int disable_eprobe(struct trace_eprobe *ep,
603 			  struct trace_array *tr)
604 {
605 	struct event_trigger_data *trigger = NULL, *iter;
606 	struct trace_event_file *file;
607 	struct event_filter *filter;
608 	struct eprobe_data *edata;
609 
610 	file = find_event_file(tr, ep->event_system, ep->event_name);
611 	if (!file)
612 		return -ENOENT;
613 
614 	list_for_each_entry(iter, &file->triggers, list) {
615 		if (!(iter->flags & EVENT_TRIGGER_FL_PROBE))
616 			continue;
617 		edata = iter->private_data;
618 		if (edata->ep == ep) {
619 			trigger = iter;
620 			break;
621 		}
622 	}
623 	if (!trigger)
624 		return -ENODEV;
625 
626 	list_del_rcu(&trigger->list);
627 
628 	trace_event_trigger_enable_disable(file, 0);
629 	update_cond_flag(file);
630 
631 	/* Make sure nothing is using the edata or trigger */
632 	tracepoint_synchronize_unregister();
633 
634 	filter = rcu_access_pointer(trigger->filter);
635 
636 	if (filter)
637 		free_event_filter(filter);
638 	kfree(edata);
639 	kfree(trigger);
640 
641 	return 0;
642 }
643 
644 static int enable_trace_eprobe(struct trace_event_call *call,
645 			       struct trace_event_file *file)
646 {
647 	struct trace_probe *tp;
648 	struct trace_eprobe *ep;
649 	bool enabled;
650 	int ret = 0;
651 	int cnt = 0;
652 
653 	tp = trace_probe_primary_from_call(call);
654 	if (WARN_ON_ONCE(!tp))
655 		return -ENODEV;
656 	enabled = trace_probe_is_enabled(tp);
657 
658 	/* This also changes "enabled" state */
659 	if (file) {
660 		ret = trace_probe_add_file(tp, file);
661 		if (ret)
662 			return ret;
663 	} else
664 		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
665 
666 	if (enabled)
667 		return 0;
668 
669 	for_each_trace_eprobe_tp(ep, tp) {
670 		ret = enable_eprobe(ep, file);
671 		if (ret)
672 			break;
673 		enabled = true;
674 		cnt++;
675 	}
676 
677 	if (ret) {
678 		/* Failed to enable one of them. Roll back all */
679 		if (enabled) {
680 			/*
681 			 * It's a bug if one failed for something other than memory
682 			 * not being available but another eprobe succeeded.
683 			 */
684 			WARN_ON_ONCE(ret != -ENOMEM);
685 
686 			for_each_trace_eprobe_tp(ep, tp) {
687 				disable_eprobe(ep, file->tr);
688 				if (!--cnt)
689 					break;
690 			}
691 		}
692 		if (file)
693 			trace_probe_remove_file(tp, file);
694 		else
695 			trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
696 	}
697 
698 	return ret;
699 }
700 
701 static int disable_trace_eprobe(struct trace_event_call *call,
702 				struct trace_event_file *file)
703 {
704 	struct trace_probe *tp;
705 	struct trace_eprobe *ep;
706 
707 	tp = trace_probe_primary_from_call(call);
708 	if (WARN_ON_ONCE(!tp))
709 		return -ENODEV;
710 
711 	if (file) {
712 		if (!trace_probe_get_file_link(tp, file))
713 			return -ENOENT;
714 		if (!trace_probe_has_single_file(tp))
715 			goto out;
716 		trace_probe_clear_flag(tp, TP_FLAG_TRACE);
717 	} else
718 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
719 
720 	if (!trace_probe_is_enabled(tp)) {
721 		for_each_trace_eprobe_tp(ep, tp)
722 			disable_eprobe(ep, file->tr);
723 	}
724 
725  out:
726 	if (file)
727 		/*
728 		 * Synchronization is done in below function. For perf event,
729 		 * file == NULL and perf_trace_event_unreg() calls
730 		 * tracepoint_synchronize_unregister() to ensure synchronize
731 		 * event. We don't need to care about it.
732 		 */
733 		trace_probe_remove_file(tp, file);
734 
735 	return 0;
736 }
737 
738 static int eprobe_register(struct trace_event_call *event,
739 			   enum trace_reg type, void *data)
740 {
741 	struct trace_event_file *file = data;
742 
743 	switch (type) {
744 	case TRACE_REG_REGISTER:
745 		return enable_trace_eprobe(event, file);
746 	case TRACE_REG_UNREGISTER:
747 		return disable_trace_eprobe(event, file);
748 #ifdef CONFIG_PERF_EVENTS
749 	case TRACE_REG_PERF_REGISTER:
750 	case TRACE_REG_PERF_UNREGISTER:
751 	case TRACE_REG_PERF_OPEN:
752 	case TRACE_REG_PERF_CLOSE:
753 	case TRACE_REG_PERF_ADD:
754 	case TRACE_REG_PERF_DEL:
755 		return 0;
756 #endif
757 	}
758 	return 0;
759 }
760 
761 static inline void init_trace_eprobe_call(struct trace_eprobe *ep)
762 {
763 	struct trace_event_call *call = trace_probe_event_call(&ep->tp);
764 
765 	call->flags = TRACE_EVENT_FL_EPROBE;
766 	call->event.funcs = &eprobe_funcs;
767 	call->class->fields_array = eprobe_fields_array;
768 	call->class->reg = eprobe_register;
769 }
770 
771 static struct trace_event_call *
772 find_and_get_event(const char *system, const char *event_name)
773 {
774 	struct trace_event_call *tp_event;
775 	const char *name;
776 
777 	list_for_each_entry(tp_event, &ftrace_events, list) {
778 		/* Skip other probes and ftrace events */
779 		if (tp_event->flags &
780 		    (TRACE_EVENT_FL_IGNORE_ENABLE |
781 		     TRACE_EVENT_FL_KPROBE |
782 		     TRACE_EVENT_FL_UPROBE |
783 		     TRACE_EVENT_FL_EPROBE))
784 			continue;
785 		if (!tp_event->class->system ||
786 		    strcmp(system, tp_event->class->system))
787 			continue;
788 		name = trace_event_name(tp_event);
789 		if (!name || strcmp(event_name, name))
790 			continue;
791 		if (!trace_event_try_get_ref(tp_event)) {
792 			return NULL;
793 			break;
794 		}
795 		return tp_event;
796 		break;
797 	}
798 	return NULL;
799 }
800 
801 static int trace_eprobe_tp_update_arg(struct trace_eprobe *ep, const char *argv[], int i)
802 {
803 	struct traceprobe_parse_context ctx = {
804 		.event = ep->event,
805 		.flags = TPARG_FL_KERNEL | TPARG_FL_TEVENT,
806 	};
807 	int ret;
808 
809 	ret = traceprobe_parse_probe_arg(&ep->tp, i, argv[i], &ctx);
810 	/* Handle symbols "@" */
811 	if (!ret)
812 		ret = traceprobe_update_arg(&ep->tp.args[i]);
813 
814 	traceprobe_finish_parse(&ctx);
815 	return ret;
816 }
817 
818 static int trace_eprobe_parse_filter(struct trace_eprobe *ep, int argc, const char *argv[])
819 {
820 	struct event_filter *dummy = NULL;
821 	int i, ret, len = 0;
822 	char *p;
823 
824 	if (argc == 0) {
825 		trace_probe_log_err(0, NO_EP_FILTER);
826 		return -EINVAL;
827 	}
828 
829 	/* Recover the filter string */
830 	for (i = 0; i < argc; i++)
831 		len += strlen(argv[i]) + 1;
832 
833 	ep->filter_str = kzalloc(len, GFP_KERNEL);
834 	if (!ep->filter_str)
835 		return -ENOMEM;
836 
837 	p = ep->filter_str;
838 	for (i = 0; i < argc; i++) {
839 		if (i)
840 			ret = snprintf(p, len, " %s", argv[i]);
841 		else
842 			ret = snprintf(p, len, "%s", argv[i]);
843 		p += ret;
844 		len -= ret;
845 	}
846 
847 	/*
848 	 * Ensure the filter string can be parsed correctly. Note, this
849 	 * filter string is for the original event, not for the eprobe.
850 	 */
851 	ret = create_event_filter(top_trace_array(), ep->event, ep->filter_str,
852 				  true, &dummy);
853 	free_event_filter(dummy);
854 	if (ret)
855 		goto error;
856 
857 	return 0;
858 error:
859 	kfree(ep->filter_str);
860 	ep->filter_str = NULL;
861 	return ret;
862 }
863 
864 static int __trace_eprobe_create(int argc, const char *argv[])
865 {
866 	/*
867 	 * Argument syntax:
868 	 *      e[:[GRP/][ENAME]] SYSTEM.EVENT [FETCHARGS] [if FILTER]
869 	 * Fetch args (no space):
870 	 *  <name>=$<field>[:TYPE]
871 	 */
872 	const char *event = NULL, *group = EPROBE_EVENT_SYSTEM;
873 	const char *sys_event = NULL, *sys_name = NULL;
874 	struct trace_event_call *event_call;
875 	struct trace_eprobe *ep = NULL;
876 	char buf1[MAX_EVENT_NAME_LEN];
877 	char buf2[MAX_EVENT_NAME_LEN];
878 	char gbuf[MAX_EVENT_NAME_LEN];
879 	int ret = 0, filter_idx = 0;
880 	int i, filter_cnt;
881 
882 	if (argc < 2 || argv[0][0] != 'e')
883 		return -ECANCELED;
884 
885 	trace_probe_log_init("event_probe", argc, argv);
886 
887 	event = strchr(&argv[0][1], ':');
888 	if (event) {
889 		event++;
890 		ret = traceprobe_parse_event_name(&event, &group, gbuf,
891 						  event - argv[0]);
892 		if (ret)
893 			goto parse_error;
894 	}
895 
896 	trace_probe_log_set_index(1);
897 	sys_event = argv[1];
898 	ret = traceprobe_parse_event_name(&sys_event, &sys_name, buf2, 0);
899 	if (ret || !sys_event || !sys_name) {
900 		trace_probe_log_err(0, NO_EVENT_INFO);
901 		goto parse_error;
902 	}
903 
904 	if (!event) {
905 		strscpy(buf1, sys_event, MAX_EVENT_NAME_LEN);
906 		event = buf1;
907 	}
908 
909 	for (i = 2; i < argc; i++) {
910 		if (!strcmp(argv[i], "if")) {
911 			filter_idx = i + 1;
912 			filter_cnt = argc - filter_idx;
913 			argc = i;
914 			break;
915 		}
916 	}
917 
918 	mutex_lock(&event_mutex);
919 	event_call = find_and_get_event(sys_name, sys_event);
920 	ep = alloc_event_probe(group, event, event_call, argc - 2);
921 	mutex_unlock(&event_mutex);
922 
923 	if (IS_ERR(ep)) {
924 		ret = PTR_ERR(ep);
925 		if (ret == -ENODEV)
926 			trace_probe_log_err(0, BAD_ATTACH_EVENT);
927 		/* This must return -ENOMEM or missing event, else there is a bug */
928 		WARN_ON_ONCE(ret != -ENOMEM && ret != -ENODEV);
929 		ep = NULL;
930 		goto error;
931 	}
932 
933 	if (filter_idx) {
934 		trace_probe_log_set_index(filter_idx);
935 		ret = trace_eprobe_parse_filter(ep, filter_cnt, argv + filter_idx);
936 		if (ret)
937 			goto parse_error;
938 	} else
939 		ep->filter_str = NULL;
940 
941 	argc -= 2; argv += 2;
942 	/* parse arguments */
943 	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
944 		trace_probe_log_set_index(i + 2);
945 		ret = trace_eprobe_tp_update_arg(ep, argv, i);
946 		if (ret)
947 			goto error;
948 	}
949 	ret = traceprobe_set_print_fmt(&ep->tp, PROBE_PRINT_EVENT);
950 	if (ret < 0)
951 		goto error;
952 	init_trace_eprobe_call(ep);
953 	mutex_lock(&event_mutex);
954 	ret = trace_probe_register_event_call(&ep->tp);
955 	if (ret) {
956 		if (ret == -EEXIST) {
957 			trace_probe_log_set_index(0);
958 			trace_probe_log_err(0, EVENT_EXIST);
959 		}
960 		mutex_unlock(&event_mutex);
961 		goto error;
962 	}
963 	ret = dyn_event_add(&ep->devent, &ep->tp.event->call);
964 	mutex_unlock(&event_mutex);
965 	return ret;
966 parse_error:
967 	ret = -EINVAL;
968 error:
969 	trace_event_probe_cleanup(ep);
970 	return ret;
971 }
972 
973 /*
974  * Register dynevent at core_initcall. This allows kernel to setup eprobe
975  * events in postcore_initcall without tracefs.
976  */
977 static __init int trace_events_eprobe_init_early(void)
978 {
979 	int err = 0;
980 
981 	err = dyn_event_register(&eprobe_dyn_event_ops);
982 	if (err)
983 		pr_warn("Could not register eprobe_dyn_event_ops\n");
984 
985 	return err;
986 }
987 core_initcall(trace_events_eprobe_init_early);
988