xref: /linux/kernel/trace/trace_eprobe.c (revision 4359a011e259a4608afc7fb3635370c9d4ba5943)
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 
20 #define EPROBE_EVENT_SYSTEM "eprobes"
21 
22 struct trace_eprobe {
23 	/* tracepoint system */
24 	const char *event_system;
25 
26 	/* tracepoint event */
27 	const char *event_name;
28 
29 	/* filter string for the tracepoint */
30 	char *filter_str;
31 
32 	struct trace_event_call *event;
33 
34 	struct dyn_event	devent;
35 	struct trace_probe	tp;
36 };
37 
38 struct eprobe_data {
39 	struct trace_event_file	*file;
40 	struct trace_eprobe	*ep;
41 };
42 
43 static int __trace_eprobe_create(int argc, const char *argv[]);
44 
45 static void trace_event_probe_cleanup(struct trace_eprobe *ep)
46 {
47 	if (!ep)
48 		return;
49 	trace_probe_cleanup(&ep->tp);
50 	kfree(ep->event_name);
51 	kfree(ep->event_system);
52 	if (ep->event)
53 		trace_event_put_ref(ep->event);
54 	kfree(ep);
55 }
56 
57 static struct trace_eprobe *to_trace_eprobe(struct dyn_event *ev)
58 {
59 	return container_of(ev, struct trace_eprobe, devent);
60 }
61 
62 static int eprobe_dyn_event_create(const char *raw_command)
63 {
64 	return trace_probe_create(raw_command, __trace_eprobe_create);
65 }
66 
67 static int eprobe_dyn_event_show(struct seq_file *m, struct dyn_event *ev)
68 {
69 	struct trace_eprobe *ep = to_trace_eprobe(ev);
70 	int i;
71 
72 	seq_printf(m, "e:%s/%s", trace_probe_group_name(&ep->tp),
73 				trace_probe_name(&ep->tp));
74 	seq_printf(m, " %s.%s", ep->event_system, ep->event_name);
75 
76 	for (i = 0; i < ep->tp.nr_args; i++)
77 		seq_printf(m, " %s=%s", ep->tp.args[i].name, ep->tp.args[i].comm);
78 	seq_putc(m, '\n');
79 
80 	return 0;
81 }
82 
83 static int unregister_trace_eprobe(struct trace_eprobe *ep)
84 {
85 	/* If other probes are on the event, just unregister eprobe */
86 	if (trace_probe_has_sibling(&ep->tp))
87 		goto unreg;
88 
89 	/* Enabled event can not be unregistered */
90 	if (trace_probe_is_enabled(&ep->tp))
91 		return -EBUSY;
92 
93 	/* Will fail if probe is being used by ftrace or perf */
94 	if (trace_probe_unregister_event_call(&ep->tp))
95 		return -EBUSY;
96 
97 unreg:
98 	dyn_event_remove(&ep->devent);
99 	trace_probe_unlink(&ep->tp);
100 
101 	return 0;
102 }
103 
104 static int eprobe_dyn_event_release(struct dyn_event *ev)
105 {
106 	struct trace_eprobe *ep = to_trace_eprobe(ev);
107 	int ret = unregister_trace_eprobe(ep);
108 
109 	if (!ret)
110 		trace_event_probe_cleanup(ep);
111 	return ret;
112 }
113 
114 static bool eprobe_dyn_event_is_busy(struct dyn_event *ev)
115 {
116 	struct trace_eprobe *ep = to_trace_eprobe(ev);
117 
118 	return trace_probe_is_enabled(&ep->tp);
119 }
120 
121 static bool eprobe_dyn_event_match(const char *system, const char *event,
122 			int argc, const char **argv, struct dyn_event *ev)
123 {
124 	struct trace_eprobe *ep = to_trace_eprobe(ev);
125 	const char *slash;
126 
127 	/*
128 	 * We match the following:
129 	 *  event only			- match all eprobes with event name
130 	 *  system and event only	- match all system/event probes
131 	 *  system only			- match all system probes
132 	 *
133 	 * The below has the above satisfied with more arguments:
134 	 *
135 	 *  attached system/event	- If the arg has the system and event
136 	 *				  the probe is attached to, match
137 	 *				  probes with the attachment.
138 	 *
139 	 *  If any more args are given, then it requires a full match.
140 	 */
141 
142 	/*
143 	 * If system exists, but this probe is not part of that system
144 	 * do not match.
145 	 */
146 	if (system && strcmp(trace_probe_group_name(&ep->tp), system) != 0)
147 		return false;
148 
149 	/* Must match the event name */
150 	if (event[0] != '\0' && strcmp(trace_probe_name(&ep->tp), event) != 0)
151 		return false;
152 
153 	/* No arguments match all */
154 	if (argc < 1)
155 		return true;
156 
157 	/* First argument is the system/event the probe is attached to */
158 
159 	slash = strchr(argv[0], '/');
160 	if (!slash)
161 		slash = strchr(argv[0], '.');
162 	if (!slash)
163 		return false;
164 
165 	if (strncmp(ep->event_system, argv[0], slash - argv[0]))
166 		return false;
167 	if (strcmp(ep->event_name, slash + 1))
168 		return false;
169 
170 	argc--;
171 	argv++;
172 
173 	/* If there are no other args, then match */
174 	if (argc < 1)
175 		return true;
176 
177 	return trace_probe_match_command_args(&ep->tp, argc, argv);
178 }
179 
180 static struct dyn_event_operations eprobe_dyn_event_ops = {
181 	.create = eprobe_dyn_event_create,
182 	.show = eprobe_dyn_event_show,
183 	.is_busy = eprobe_dyn_event_is_busy,
184 	.free = eprobe_dyn_event_release,
185 	.match = eprobe_dyn_event_match,
186 };
187 
188 static struct trace_eprobe *alloc_event_probe(const char *group,
189 					      const char *this_event,
190 					      struct trace_event_call *event,
191 					      int nargs)
192 {
193 	struct trace_eprobe *ep;
194 	const char *event_name;
195 	const char *sys_name;
196 	int ret = -ENOMEM;
197 
198 	if (!event)
199 		return ERR_PTR(-ENODEV);
200 
201 	sys_name = event->class->system;
202 	event_name = trace_event_name(event);
203 
204 	ep = kzalloc(struct_size(ep, tp.args, nargs), GFP_KERNEL);
205 	if (!ep) {
206 		trace_event_put_ref(event);
207 		goto error;
208 	}
209 	ep->event = event;
210 	ep->event_name = kstrdup(event_name, GFP_KERNEL);
211 	if (!ep->event_name)
212 		goto error;
213 	ep->event_system = kstrdup(sys_name, GFP_KERNEL);
214 	if (!ep->event_system)
215 		goto error;
216 
217 	ret = trace_probe_init(&ep->tp, this_event, group, false);
218 	if (ret < 0)
219 		goto error;
220 
221 	dyn_event_init(&ep->devent, &eprobe_dyn_event_ops);
222 	return ep;
223 error:
224 	trace_event_probe_cleanup(ep);
225 	return ERR_PTR(ret);
226 }
227 
228 static int trace_eprobe_tp_arg_update(struct trace_eprobe *ep, int i)
229 {
230 	struct probe_arg *parg = &ep->tp.args[i];
231 	struct ftrace_event_field *field;
232 	struct list_head *head;
233 	int ret = -ENOENT;
234 
235 	head = trace_get_fields(ep->event);
236 	list_for_each_entry(field, head, link) {
237 		if (!strcmp(parg->code->data, field->name)) {
238 			kfree(parg->code->data);
239 			parg->code->data = field;
240 			return 0;
241 		}
242 	}
243 
244 	/*
245 	 * Argument not found on event. But allow for comm and COMM
246 	 * to be used to get the current->comm.
247 	 */
248 	if (strcmp(parg->code->data, "COMM") == 0 ||
249 	    strcmp(parg->code->data, "comm") == 0) {
250 		parg->code->op = FETCH_OP_COMM;
251 		ret = 0;
252 	}
253 
254 	kfree(parg->code->data);
255 	parg->code->data = NULL;
256 	return ret;
257 }
258 
259 static int eprobe_event_define_fields(struct trace_event_call *event_call)
260 {
261 	struct eprobe_trace_entry_head field;
262 	struct trace_probe *tp;
263 
264 	tp = trace_probe_primary_from_call(event_call);
265 	if (WARN_ON_ONCE(!tp))
266 		return -ENOENT;
267 
268 	return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
269 }
270 
271 static struct trace_event_fields eprobe_fields_array[] = {
272 	{ .type = TRACE_FUNCTION_TYPE,
273 	  .define_fields = eprobe_event_define_fields },
274 	{}
275 };
276 
277 /* Event entry printers */
278 static enum print_line_t
279 print_eprobe_event(struct trace_iterator *iter, int flags,
280 		   struct trace_event *event)
281 {
282 	struct eprobe_trace_entry_head *field;
283 	struct trace_event_call *pevent;
284 	struct trace_event *probed_event;
285 	struct trace_seq *s = &iter->seq;
286 	struct trace_eprobe *ep;
287 	struct trace_probe *tp;
288 	unsigned int type;
289 
290 	field = (struct eprobe_trace_entry_head *)iter->ent;
291 	tp = trace_probe_primary_from_call(
292 		container_of(event, struct trace_event_call, event));
293 	if (WARN_ON_ONCE(!tp))
294 		goto out;
295 
296 	ep = container_of(tp, struct trace_eprobe, tp);
297 	type = ep->event->event.type;
298 
299 	trace_seq_printf(s, "%s: (", trace_probe_name(tp));
300 
301 	probed_event = ftrace_find_event(type);
302 	if (probed_event) {
303 		pevent = container_of(probed_event, struct trace_event_call, event);
304 		trace_seq_printf(s, "%s.%s", pevent->class->system,
305 				 trace_event_name(pevent));
306 	} else {
307 		trace_seq_printf(s, "%u", type);
308 	}
309 
310 	trace_seq_putc(s, ')');
311 
312 	if (print_probe_args(s, tp->args, tp->nr_args,
313 			     (u8 *)&field[1], field) < 0)
314 		goto out;
315 
316 	trace_seq_putc(s, '\n');
317  out:
318 	return trace_handle_return(s);
319 }
320 
321 static unsigned long get_event_field(struct fetch_insn *code, void *rec)
322 {
323 	struct ftrace_event_field *field = code->data;
324 	unsigned long val;
325 	void *addr;
326 
327 	addr = rec + field->offset;
328 
329 	if (is_string_field(field)) {
330 		switch (field->filter_type) {
331 		case FILTER_DYN_STRING:
332 			val = (unsigned long)(rec + (*(unsigned int *)addr & 0xffff));
333 			break;
334 		case FILTER_RDYN_STRING:
335 			val = (unsigned long)(addr + (*(unsigned int *)addr & 0xffff));
336 			break;
337 		case FILTER_STATIC_STRING:
338 			val = (unsigned long)addr;
339 			break;
340 		case FILTER_PTR_STRING:
341 			val = (unsigned long)(*(char *)addr);
342 			break;
343 		default:
344 			WARN_ON_ONCE(1);
345 			return 0;
346 		}
347 		return val;
348 	}
349 
350 	switch (field->size) {
351 	case 1:
352 		if (field->is_signed)
353 			val = *(char *)addr;
354 		else
355 			val = *(unsigned char *)addr;
356 		break;
357 	case 2:
358 		if (field->is_signed)
359 			val = *(short *)addr;
360 		else
361 			val = *(unsigned short *)addr;
362 		break;
363 	case 4:
364 		if (field->is_signed)
365 			val = *(int *)addr;
366 		else
367 			val = *(unsigned int *)addr;
368 		break;
369 	default:
370 		if (field->is_signed)
371 			val = *(long *)addr;
372 		else
373 			val = *(unsigned long *)addr;
374 		break;
375 	}
376 	return val;
377 }
378 
379 static int get_eprobe_size(struct trace_probe *tp, void *rec)
380 {
381 	struct fetch_insn *code;
382 	struct probe_arg *arg;
383 	int i, len, ret = 0;
384 
385 	for (i = 0; i < tp->nr_args; i++) {
386 		arg = tp->args + i;
387 		if (arg->dynamic) {
388 			unsigned long val;
389 
390 			code = arg->code;
391  retry:
392 			switch (code->op) {
393 			case FETCH_OP_TP_ARG:
394 				val = get_event_field(code, rec);
395 				break;
396 			case FETCH_OP_IMM:
397 				val = code->immediate;
398 				break;
399 			case FETCH_OP_COMM:
400 				val = (unsigned long)current->comm;
401 				break;
402 			case FETCH_OP_DATA:
403 				val = (unsigned long)code->data;
404 				break;
405 			case FETCH_NOP_SYMBOL:	/* Ignore a place holder */
406 				code++;
407 				goto retry;
408 			default:
409 				continue;
410 			}
411 			code++;
412 			len = process_fetch_insn_bottom(code, val, NULL, NULL);
413 			if (len > 0)
414 				ret += len;
415 		}
416 	}
417 
418 	return ret;
419 }
420 
421 /* Kprobe specific fetch functions */
422 
423 /* Note that we don't verify it, since the code does not come from user space */
424 static int
425 process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
426 		   void *base)
427 {
428 	unsigned long val;
429 
430  retry:
431 	switch (code->op) {
432 	case FETCH_OP_TP_ARG:
433 		val = get_event_field(code, rec);
434 		break;
435 	case FETCH_OP_IMM:
436 		val = code->immediate;
437 		break;
438 	case FETCH_OP_COMM:
439 		val = (unsigned long)current->comm;
440 		break;
441 	case FETCH_OP_DATA:
442 		val = (unsigned long)code->data;
443 		break;
444 	case FETCH_NOP_SYMBOL:	/* Ignore a place holder */
445 		code++;
446 		goto retry;
447 	default:
448 		return -EILSEQ;
449 	}
450 	code++;
451 	return process_fetch_insn_bottom(code, val, dest, base);
452 }
453 NOKPROBE_SYMBOL(process_fetch_insn)
454 
455 /* Return the length of string -- including null terminal byte */
456 static nokprobe_inline int
457 fetch_store_strlen_user(unsigned long addr)
458 {
459 	const void __user *uaddr =  (__force const void __user *)addr;
460 
461 	return strnlen_user_nofault(uaddr, MAX_STRING_SIZE);
462 }
463 
464 /* Return the length of string -- including null terminal byte */
465 static nokprobe_inline int
466 fetch_store_strlen(unsigned long addr)
467 {
468 	int ret, len = 0;
469 	u8 c;
470 
471 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
472 	if (addr < TASK_SIZE)
473 		return fetch_store_strlen_user(addr);
474 #endif
475 
476 	do {
477 		ret = copy_from_kernel_nofault(&c, (u8 *)addr + len, 1);
478 		len++;
479 	} while (c && ret == 0 && len < MAX_STRING_SIZE);
480 
481 	return (ret < 0) ? ret : len;
482 }
483 
484 /*
485  * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
486  * with max length and relative data location.
487  */
488 static nokprobe_inline int
489 fetch_store_string_user(unsigned long addr, void *dest, void *base)
490 {
491 	const void __user *uaddr =  (__force const void __user *)addr;
492 	int maxlen = get_loc_len(*(u32 *)dest);
493 	void *__dest;
494 	long ret;
495 
496 	if (unlikely(!maxlen))
497 		return -ENOMEM;
498 
499 	__dest = get_loc_data(dest, base);
500 
501 	ret = strncpy_from_user_nofault(__dest, uaddr, maxlen);
502 	if (ret >= 0)
503 		*(u32 *)dest = make_data_loc(ret, __dest - base);
504 
505 	return ret;
506 }
507 
508 /*
509  * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
510  * length and relative data location.
511  */
512 static nokprobe_inline int
513 fetch_store_string(unsigned long addr, void *dest, void *base)
514 {
515 	int maxlen = get_loc_len(*(u32 *)dest);
516 	void *__dest;
517 	long ret;
518 
519 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
520 	if ((unsigned long)addr < TASK_SIZE)
521 		return fetch_store_string_user(addr, dest, base);
522 #endif
523 
524 	if (unlikely(!maxlen))
525 		return -ENOMEM;
526 
527 	__dest = get_loc_data(dest, base);
528 
529 	/*
530 	 * Try to get string again, since the string can be changed while
531 	 * probing.
532 	 */
533 	ret = strncpy_from_kernel_nofault(__dest, (void *)addr, maxlen);
534 	if (ret >= 0)
535 		*(u32 *)dest = make_data_loc(ret, __dest - base);
536 
537 	return ret;
538 }
539 
540 static nokprobe_inline int
541 probe_mem_read_user(void *dest, void *src, size_t size)
542 {
543 	const void __user *uaddr =  (__force const void __user *)src;
544 
545 	return copy_from_user_nofault(dest, uaddr, size);
546 }
547 
548 static nokprobe_inline int
549 probe_mem_read(void *dest, void *src, size_t size)
550 {
551 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
552 	if ((unsigned long)src < TASK_SIZE)
553 		return probe_mem_read_user(dest, src, size);
554 #endif
555 	return copy_from_kernel_nofault(dest, src, size);
556 }
557 
558 /* eprobe handler */
559 static inline void
560 __eprobe_trace_func(struct eprobe_data *edata, void *rec)
561 {
562 	struct eprobe_trace_entry_head *entry;
563 	struct trace_event_call *call = trace_probe_event_call(&edata->ep->tp);
564 	struct trace_event_buffer fbuffer;
565 	int dsize;
566 
567 	if (WARN_ON_ONCE(call != edata->file->event_call))
568 		return;
569 
570 	if (trace_trigger_soft_disabled(edata->file))
571 		return;
572 
573 	dsize = get_eprobe_size(&edata->ep->tp, rec);
574 
575 	entry = trace_event_buffer_reserve(&fbuffer, edata->file,
576 					   sizeof(*entry) + edata->ep->tp.size + dsize);
577 
578 	if (!entry)
579 		return;
580 
581 	entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
582 	store_trace_args(&entry[1], &edata->ep->tp, rec, sizeof(*entry), dsize);
583 
584 	trace_event_buffer_commit(&fbuffer);
585 }
586 
587 /*
588  * The event probe implementation uses event triggers to get access to
589  * the event it is attached to, but is not an actual trigger. The below
590  * functions are just stubs to fulfill what is needed to use the trigger
591  * infrastructure.
592  */
593 static int eprobe_trigger_init(struct event_trigger_data *data)
594 {
595 	return 0;
596 }
597 
598 static void eprobe_trigger_free(struct event_trigger_data *data)
599 {
600 
601 }
602 
603 static int eprobe_trigger_print(struct seq_file *m,
604 				struct event_trigger_data *data)
605 {
606 	/* Do not print eprobe event triggers */
607 	return 0;
608 }
609 
610 static void eprobe_trigger_func(struct event_trigger_data *data,
611 				struct trace_buffer *buffer, void *rec,
612 				struct ring_buffer_event *rbe)
613 {
614 	struct eprobe_data *edata = data->private_data;
615 
616 	__eprobe_trace_func(edata, rec);
617 }
618 
619 static struct event_trigger_ops eprobe_trigger_ops = {
620 	.trigger		= eprobe_trigger_func,
621 	.print			= eprobe_trigger_print,
622 	.init			= eprobe_trigger_init,
623 	.free			= eprobe_trigger_free,
624 };
625 
626 static int eprobe_trigger_cmd_parse(struct event_command *cmd_ops,
627 				    struct trace_event_file *file,
628 				    char *glob, char *cmd,
629 				    char *param_and_filter)
630 {
631 	return -1;
632 }
633 
634 static int eprobe_trigger_reg_func(char *glob,
635 				   struct event_trigger_data *data,
636 				   struct trace_event_file *file)
637 {
638 	return -1;
639 }
640 
641 static void eprobe_trigger_unreg_func(char *glob,
642 				      struct event_trigger_data *data,
643 				      struct trace_event_file *file)
644 {
645 
646 }
647 
648 static struct event_trigger_ops *eprobe_trigger_get_ops(char *cmd,
649 							char *param)
650 {
651 	return &eprobe_trigger_ops;
652 }
653 
654 static struct event_command event_trigger_cmd = {
655 	.name			= "eprobe",
656 	.trigger_type		= ETT_EVENT_EPROBE,
657 	.flags			= EVENT_CMD_FL_NEEDS_REC,
658 	.parse			= eprobe_trigger_cmd_parse,
659 	.reg			= eprobe_trigger_reg_func,
660 	.unreg			= eprobe_trigger_unreg_func,
661 	.unreg_all		= NULL,
662 	.get_trigger_ops	= eprobe_trigger_get_ops,
663 	.set_filter		= NULL,
664 };
665 
666 static struct event_trigger_data *
667 new_eprobe_trigger(struct trace_eprobe *ep, struct trace_event_file *file)
668 {
669 	struct event_trigger_data *trigger;
670 	struct event_filter *filter = NULL;
671 	struct eprobe_data *edata;
672 	int ret;
673 
674 	edata = kzalloc(sizeof(*edata), GFP_KERNEL);
675 	trigger = kzalloc(sizeof(*trigger), GFP_KERNEL);
676 	if (!trigger || !edata) {
677 		ret = -ENOMEM;
678 		goto error;
679 	}
680 
681 	trigger->flags = EVENT_TRIGGER_FL_PROBE;
682 	trigger->count = -1;
683 	trigger->ops = &eprobe_trigger_ops;
684 
685 	/*
686 	 * EVENT PROBE triggers are not registered as commands with
687 	 * register_event_command(), as they are not controlled by the user
688 	 * from the trigger file
689 	 */
690 	trigger->cmd_ops = &event_trigger_cmd;
691 
692 	INIT_LIST_HEAD(&trigger->list);
693 
694 	if (ep->filter_str) {
695 		ret = create_event_filter(file->tr, file->event_call,
696 					ep->filter_str, false, &filter);
697 		if (ret)
698 			goto error;
699 	}
700 	RCU_INIT_POINTER(trigger->filter, filter);
701 
702 	edata->file = file;
703 	edata->ep = ep;
704 	trigger->private_data = edata;
705 
706 	return trigger;
707 error:
708 	free_event_filter(filter);
709 	kfree(edata);
710 	kfree(trigger);
711 	return ERR_PTR(ret);
712 }
713 
714 static int enable_eprobe(struct trace_eprobe *ep,
715 			 struct trace_event_file *eprobe_file)
716 {
717 	struct event_trigger_data *trigger;
718 	struct trace_event_file *file;
719 	struct trace_array *tr = eprobe_file->tr;
720 
721 	file = find_event_file(tr, ep->event_system, ep->event_name);
722 	if (!file)
723 		return -ENOENT;
724 	trigger = new_eprobe_trigger(ep, eprobe_file);
725 	if (IS_ERR(trigger))
726 		return PTR_ERR(trigger);
727 
728 	list_add_tail_rcu(&trigger->list, &file->triggers);
729 
730 	trace_event_trigger_enable_disable(file, 1);
731 	update_cond_flag(file);
732 
733 	return 0;
734 }
735 
736 static struct trace_event_functions eprobe_funcs = {
737 	.trace		= print_eprobe_event
738 };
739 
740 static int disable_eprobe(struct trace_eprobe *ep,
741 			  struct trace_array *tr)
742 {
743 	struct event_trigger_data *trigger = NULL, *iter;
744 	struct trace_event_file *file;
745 	struct event_filter *filter;
746 	struct eprobe_data *edata;
747 
748 	file = find_event_file(tr, ep->event_system, ep->event_name);
749 	if (!file)
750 		return -ENOENT;
751 
752 	list_for_each_entry(iter, &file->triggers, list) {
753 		if (!(iter->flags & EVENT_TRIGGER_FL_PROBE))
754 			continue;
755 		edata = iter->private_data;
756 		if (edata->ep == ep) {
757 			trigger = iter;
758 			break;
759 		}
760 	}
761 	if (!trigger)
762 		return -ENODEV;
763 
764 	list_del_rcu(&trigger->list);
765 
766 	trace_event_trigger_enable_disable(file, 0);
767 	update_cond_flag(file);
768 
769 	/* Make sure nothing is using the edata or trigger */
770 	tracepoint_synchronize_unregister();
771 
772 	filter = rcu_access_pointer(trigger->filter);
773 
774 	if (filter)
775 		free_event_filter(filter);
776 	kfree(edata);
777 	kfree(trigger);
778 
779 	return 0;
780 }
781 
782 static int enable_trace_eprobe(struct trace_event_call *call,
783 			       struct trace_event_file *file)
784 {
785 	struct trace_probe *pos, *tp;
786 	struct trace_eprobe *ep;
787 	bool enabled;
788 	int ret = 0;
789 
790 	tp = trace_probe_primary_from_call(call);
791 	if (WARN_ON_ONCE(!tp))
792 		return -ENODEV;
793 	enabled = trace_probe_is_enabled(tp);
794 
795 	/* This also changes "enabled" state */
796 	if (file) {
797 		ret = trace_probe_add_file(tp, file);
798 		if (ret)
799 			return ret;
800 	} else
801 		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
802 
803 	if (enabled)
804 		return 0;
805 
806 	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
807 		ep = container_of(pos, struct trace_eprobe, tp);
808 		ret = enable_eprobe(ep, file);
809 		if (ret)
810 			break;
811 		enabled = true;
812 	}
813 
814 	if (ret) {
815 		/* Failed to enable one of them. Roll back all */
816 		if (enabled)
817 			disable_eprobe(ep, file->tr);
818 		if (file)
819 			trace_probe_remove_file(tp, file);
820 		else
821 			trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
822 	}
823 
824 	return ret;
825 }
826 
827 static int disable_trace_eprobe(struct trace_event_call *call,
828 				struct trace_event_file *file)
829 {
830 	struct trace_probe *pos, *tp;
831 	struct trace_eprobe *ep;
832 
833 	tp = trace_probe_primary_from_call(call);
834 	if (WARN_ON_ONCE(!tp))
835 		return -ENODEV;
836 
837 	if (file) {
838 		if (!trace_probe_get_file_link(tp, file))
839 			return -ENOENT;
840 		if (!trace_probe_has_single_file(tp))
841 			goto out;
842 		trace_probe_clear_flag(tp, TP_FLAG_TRACE);
843 	} else
844 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
845 
846 	if (!trace_probe_is_enabled(tp)) {
847 		list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
848 			ep = container_of(pos, struct trace_eprobe, tp);
849 			disable_eprobe(ep, file->tr);
850 		}
851 	}
852 
853  out:
854 	if (file)
855 		/*
856 		 * Synchronization is done in below function. For perf event,
857 		 * file == NULL and perf_trace_event_unreg() calls
858 		 * tracepoint_synchronize_unregister() to ensure synchronize
859 		 * event. We don't need to care about it.
860 		 */
861 		trace_probe_remove_file(tp, file);
862 
863 	return 0;
864 }
865 
866 static int eprobe_register(struct trace_event_call *event,
867 			   enum trace_reg type, void *data)
868 {
869 	struct trace_event_file *file = data;
870 
871 	switch (type) {
872 	case TRACE_REG_REGISTER:
873 		return enable_trace_eprobe(event, file);
874 	case TRACE_REG_UNREGISTER:
875 		return disable_trace_eprobe(event, file);
876 #ifdef CONFIG_PERF_EVENTS
877 	case TRACE_REG_PERF_REGISTER:
878 	case TRACE_REG_PERF_UNREGISTER:
879 	case TRACE_REG_PERF_OPEN:
880 	case TRACE_REG_PERF_CLOSE:
881 	case TRACE_REG_PERF_ADD:
882 	case TRACE_REG_PERF_DEL:
883 		return 0;
884 #endif
885 	}
886 	return 0;
887 }
888 
889 static inline void init_trace_eprobe_call(struct trace_eprobe *ep)
890 {
891 	struct trace_event_call *call = trace_probe_event_call(&ep->tp);
892 
893 	call->flags = TRACE_EVENT_FL_EPROBE;
894 	call->event.funcs = &eprobe_funcs;
895 	call->class->fields_array = eprobe_fields_array;
896 	call->class->reg = eprobe_register;
897 }
898 
899 static struct trace_event_call *
900 find_and_get_event(const char *system, const char *event_name)
901 {
902 	struct trace_event_call *tp_event;
903 	const char *name;
904 
905 	list_for_each_entry(tp_event, &ftrace_events, list) {
906 		/* Skip other probes and ftrace events */
907 		if (tp_event->flags &
908 		    (TRACE_EVENT_FL_IGNORE_ENABLE |
909 		     TRACE_EVENT_FL_KPROBE |
910 		     TRACE_EVENT_FL_UPROBE |
911 		     TRACE_EVENT_FL_EPROBE))
912 			continue;
913 		if (!tp_event->class->system ||
914 		    strcmp(system, tp_event->class->system))
915 			continue;
916 		name = trace_event_name(tp_event);
917 		if (!name || strcmp(event_name, name))
918 			continue;
919 		if (!trace_event_try_get_ref(tp_event)) {
920 			return NULL;
921 			break;
922 		}
923 		return tp_event;
924 		break;
925 	}
926 	return NULL;
927 }
928 
929 static int trace_eprobe_tp_update_arg(struct trace_eprobe *ep, const char *argv[], int i)
930 {
931 	unsigned int flags = TPARG_FL_KERNEL | TPARG_FL_TPOINT;
932 	int ret;
933 
934 	ret = traceprobe_parse_probe_arg(&ep->tp, i, argv[i], flags);
935 	if (ret)
936 		return ret;
937 
938 	if (ep->tp.args[i].code->op == FETCH_OP_TP_ARG) {
939 		ret = trace_eprobe_tp_arg_update(ep, i);
940 		if (ret)
941 			trace_probe_log_err(0, BAD_ATTACH_ARG);
942 	}
943 
944 	/* Handle symbols "@" */
945 	if (!ret)
946 		ret = traceprobe_update_arg(&ep->tp.args[i]);
947 
948 	return ret;
949 }
950 
951 static int trace_eprobe_parse_filter(struct trace_eprobe *ep, int argc, const char *argv[])
952 {
953 	struct event_filter *dummy;
954 	int i, ret, len = 0;
955 	char *p;
956 
957 	if (argc == 0) {
958 		trace_probe_log_err(0, NO_EP_FILTER);
959 		return -EINVAL;
960 	}
961 
962 	/* Recover the filter string */
963 	for (i = 0; i < argc; i++)
964 		len += strlen(argv[i]) + 1;
965 
966 	ep->filter_str = kzalloc(len, GFP_KERNEL);
967 	if (!ep->filter_str)
968 		return -ENOMEM;
969 
970 	p = ep->filter_str;
971 	for (i = 0; i < argc; i++) {
972 		ret = snprintf(p, len, "%s ", argv[i]);
973 		if (ret < 0)
974 			goto error;
975 		if (ret > len) {
976 			ret = -E2BIG;
977 			goto error;
978 		}
979 		p += ret;
980 		len -= ret;
981 	}
982 	p[-1] = '\0';
983 
984 	/*
985 	 * Ensure the filter string can be parsed correctly. Note, this
986 	 * filter string is for the original event, not for the eprobe.
987 	 */
988 	ret = create_event_filter(top_trace_array(), ep->event, ep->filter_str,
989 				  true, &dummy);
990 	free_event_filter(dummy);
991 	if (ret)
992 		goto error;
993 
994 	return 0;
995 error:
996 	kfree(ep->filter_str);
997 	ep->filter_str = NULL;
998 	return ret;
999 }
1000 
1001 static int __trace_eprobe_create(int argc, const char *argv[])
1002 {
1003 	/*
1004 	 * Argument syntax:
1005 	 *      e[:[GRP/][ENAME]] SYSTEM.EVENT [FETCHARGS] [if FILTER]
1006 	 * Fetch args (no space):
1007 	 *  <name>=$<field>[:TYPE]
1008 	 */
1009 	const char *event = NULL, *group = EPROBE_EVENT_SYSTEM;
1010 	const char *sys_event = NULL, *sys_name = NULL;
1011 	struct trace_event_call *event_call;
1012 	struct trace_eprobe *ep = NULL;
1013 	char buf1[MAX_EVENT_NAME_LEN];
1014 	char buf2[MAX_EVENT_NAME_LEN];
1015 	char gbuf[MAX_EVENT_NAME_LEN];
1016 	int ret = 0, filter_idx = 0;
1017 	int i, filter_cnt;
1018 
1019 	if (argc < 2 || argv[0][0] != 'e')
1020 		return -ECANCELED;
1021 
1022 	trace_probe_log_init("event_probe", argc, argv);
1023 
1024 	event = strchr(&argv[0][1], ':');
1025 	if (event) {
1026 		event++;
1027 		ret = traceprobe_parse_event_name(&event, &group, gbuf,
1028 						  event - argv[0]);
1029 		if (ret)
1030 			goto parse_error;
1031 	}
1032 
1033 	trace_probe_log_set_index(1);
1034 	sys_event = argv[1];
1035 	ret = traceprobe_parse_event_name(&sys_event, &sys_name, buf2, 0);
1036 	if (ret || !sys_event || !sys_name) {
1037 		trace_probe_log_err(0, NO_EVENT_INFO);
1038 		goto parse_error;
1039 	}
1040 
1041 	if (!event) {
1042 		strscpy(buf1, argv[1], MAX_EVENT_NAME_LEN);
1043 		sanitize_event_name(buf1);
1044 		event = buf1;
1045 	}
1046 
1047 	for (i = 2; i < argc; i++) {
1048 		if (!strcmp(argv[i], "if")) {
1049 			filter_idx = i + 1;
1050 			filter_cnt = argc - filter_idx;
1051 			argc = i;
1052 			break;
1053 		}
1054 	}
1055 
1056 	mutex_lock(&event_mutex);
1057 	event_call = find_and_get_event(sys_name, sys_event);
1058 	ep = alloc_event_probe(group, event, event_call, argc - 2);
1059 	mutex_unlock(&event_mutex);
1060 
1061 	if (IS_ERR(ep)) {
1062 		ret = PTR_ERR(ep);
1063 		if (ret == -ENODEV)
1064 			trace_probe_log_err(0, BAD_ATTACH_EVENT);
1065 		/* This must return -ENOMEM or missing event, else there is a bug */
1066 		WARN_ON_ONCE(ret != -ENOMEM && ret != -ENODEV);
1067 		ep = NULL;
1068 		goto error;
1069 	}
1070 
1071 	if (filter_idx) {
1072 		trace_probe_log_set_index(filter_idx);
1073 		ret = trace_eprobe_parse_filter(ep, filter_cnt, argv + filter_idx);
1074 		if (ret)
1075 			goto parse_error;
1076 	} else
1077 		ep->filter_str = NULL;
1078 
1079 	argc -= 2; argv += 2;
1080 	/* parse arguments */
1081 	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
1082 		trace_probe_log_set_index(i + 2);
1083 		ret = trace_eprobe_tp_update_arg(ep, argv, i);
1084 		if (ret)
1085 			goto error;
1086 	}
1087 	ret = traceprobe_set_print_fmt(&ep->tp, PROBE_PRINT_EVENT);
1088 	if (ret < 0)
1089 		goto error;
1090 	init_trace_eprobe_call(ep);
1091 	mutex_lock(&event_mutex);
1092 	ret = trace_probe_register_event_call(&ep->tp);
1093 	if (ret) {
1094 		if (ret == -EEXIST) {
1095 			trace_probe_log_set_index(0);
1096 			trace_probe_log_err(0, EVENT_EXIST);
1097 		}
1098 		mutex_unlock(&event_mutex);
1099 		goto error;
1100 	}
1101 	ret = dyn_event_add(&ep->devent, &ep->tp.event->call);
1102 	mutex_unlock(&event_mutex);
1103 	return ret;
1104 parse_error:
1105 	ret = -EINVAL;
1106 error:
1107 	trace_event_probe_cleanup(ep);
1108 	return ret;
1109 }
1110 
1111 /*
1112  * Register dynevent at core_initcall. This allows kernel to setup eprobe
1113  * events in postcore_initcall without tracefs.
1114  */
1115 static __init int trace_events_eprobe_init_early(void)
1116 {
1117 	int err = 0;
1118 
1119 	err = dyn_event_register(&eprobe_dyn_event_ops);
1120 	if (err)
1121 		pr_warn("Could not register eprobe_dyn_event_ops\n");
1122 
1123 	return err;
1124 }
1125 core_initcall(trace_events_eprobe_init_early);
1126