xref: /linux/kernel/trace/trace_kprobe.c (revision 2c9b3512402ed192d1f43f4531fb5da947e72bd0)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Kprobes-based tracing events
4  *
5  * Created by Masami Hiramatsu <mhiramat@redhat.com>
6  *
7  */
8 #define pr_fmt(fmt)	"trace_kprobe: " fmt
9 
10 #include <linux/bpf-cgroup.h>
11 #include <linux/security.h>
12 #include <linux/module.h>
13 #include <linux/uaccess.h>
14 #include <linux/rculist.h>
15 #include <linux/error-injection.h>
16 
17 #include <asm/setup.h>  /* for COMMAND_LINE_SIZE */
18 
19 #include "trace_dynevent.h"
20 #include "trace_kprobe_selftest.h"
21 #include "trace_probe.h"
22 #include "trace_probe_tmpl.h"
23 #include "trace_probe_kernel.h"
24 
25 #define KPROBE_EVENT_SYSTEM "kprobes"
26 #define KRETPROBE_MAXACTIVE_MAX 4096
27 
28 /* Kprobe early definition from command line */
29 static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
30 
31 static int __init set_kprobe_boot_events(char *str)
32 {
33 	strscpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
34 	disable_tracing_selftest("running kprobe events");
35 
36 	return 1;
37 }
38 __setup("kprobe_event=", set_kprobe_boot_events);
39 
40 static int trace_kprobe_create(const char *raw_command);
41 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
42 static int trace_kprobe_release(struct dyn_event *ev);
43 static bool trace_kprobe_is_busy(struct dyn_event *ev);
44 static bool trace_kprobe_match(const char *system, const char *event,
45 			int argc, const char **argv, struct dyn_event *ev);
46 
47 static struct dyn_event_operations trace_kprobe_ops = {
48 	.create = trace_kprobe_create,
49 	.show = trace_kprobe_show,
50 	.is_busy = trace_kprobe_is_busy,
51 	.free = trace_kprobe_release,
52 	.match = trace_kprobe_match,
53 };
54 
55 /*
56  * Kprobe event core functions
57  */
58 struct trace_kprobe {
59 	struct dyn_event	devent;
60 	struct kretprobe	rp;	/* Use rp.kp for kprobe use */
61 	unsigned long __percpu *nhit;
62 	const char		*symbol;	/* symbol name */
63 	struct trace_probe	tp;
64 };
65 
66 static bool is_trace_kprobe(struct dyn_event *ev)
67 {
68 	return ev->ops == &trace_kprobe_ops;
69 }
70 
71 static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
72 {
73 	return container_of(ev, struct trace_kprobe, devent);
74 }
75 
76 /**
77  * for_each_trace_kprobe - iterate over the trace_kprobe list
78  * @pos:	the struct trace_kprobe * for each entry
79  * @dpos:	the struct dyn_event * to use as a loop cursor
80  */
81 #define for_each_trace_kprobe(pos, dpos)	\
82 	for_each_dyn_event(dpos)		\
83 		if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
84 
85 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
86 {
87 	return tk->rp.handler != NULL;
88 }
89 
90 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
91 {
92 	return tk->symbol ? tk->symbol : "unknown";
93 }
94 
95 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
96 {
97 	return tk->rp.kp.offset;
98 }
99 
100 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
101 {
102 	return kprobe_gone(&tk->rp.kp);
103 }
104 
105 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
106 						 struct module *mod)
107 {
108 	int len = strlen(module_name(mod));
109 	const char *name = trace_kprobe_symbol(tk);
110 
111 	return strncmp(module_name(mod), name, len) == 0 && name[len] == ':';
112 }
113 
114 #ifdef CONFIG_MODULES
115 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
116 {
117 	char *p;
118 	bool ret;
119 
120 	if (!tk->symbol)
121 		return false;
122 	p = strchr(tk->symbol, ':');
123 	if (!p)
124 		return true;
125 	*p = '\0';
126 	rcu_read_lock_sched();
127 	ret = !!find_module(tk->symbol);
128 	rcu_read_unlock_sched();
129 	*p = ':';
130 
131 	return ret;
132 }
133 #else
134 static inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
135 {
136 	return false;
137 }
138 #endif
139 
140 static bool trace_kprobe_is_busy(struct dyn_event *ev)
141 {
142 	struct trace_kprobe *tk = to_trace_kprobe(ev);
143 
144 	return trace_probe_is_enabled(&tk->tp);
145 }
146 
147 static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
148 					    int argc, const char **argv)
149 {
150 	char buf[MAX_ARGSTR_LEN + 1];
151 
152 	if (!argc)
153 		return true;
154 
155 	if (!tk->symbol)
156 		snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
157 	else if (tk->rp.kp.offset)
158 		snprintf(buf, sizeof(buf), "%s+%u",
159 			 trace_kprobe_symbol(tk), tk->rp.kp.offset);
160 	else
161 		snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
162 	if (strcmp(buf, argv[0]))
163 		return false;
164 	argc--; argv++;
165 
166 	return trace_probe_match_command_args(&tk->tp, argc, argv);
167 }
168 
169 static bool trace_kprobe_match(const char *system, const char *event,
170 			int argc, const char **argv, struct dyn_event *ev)
171 {
172 	struct trace_kprobe *tk = to_trace_kprobe(ev);
173 
174 	return (event[0] == '\0' ||
175 		strcmp(trace_probe_name(&tk->tp), event) == 0) &&
176 	    (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
177 	    trace_kprobe_match_command_head(tk, argc, argv);
178 }
179 
180 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
181 {
182 	unsigned long nhit = 0;
183 	int cpu;
184 
185 	for_each_possible_cpu(cpu)
186 		nhit += *per_cpu_ptr(tk->nhit, cpu);
187 
188 	return nhit;
189 }
190 
191 static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
192 {
193 	return !(list_empty(&tk->rp.kp.list) &&
194 		 hlist_unhashed(&tk->rp.kp.hlist));
195 }
196 
197 /* Return 0 if it fails to find the symbol address */
198 static nokprobe_inline
199 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
200 {
201 	unsigned long addr;
202 
203 	if (tk->symbol) {
204 		addr = (unsigned long)
205 			kallsyms_lookup_name(trace_kprobe_symbol(tk));
206 		if (addr)
207 			addr += tk->rp.kp.offset;
208 	} else {
209 		addr = (unsigned long)tk->rp.kp.addr;
210 	}
211 	return addr;
212 }
213 
214 static nokprobe_inline struct trace_kprobe *
215 trace_kprobe_primary_from_call(struct trace_event_call *call)
216 {
217 	struct trace_probe *tp;
218 
219 	tp = trace_probe_primary_from_call(call);
220 	if (WARN_ON_ONCE(!tp))
221 		return NULL;
222 
223 	return container_of(tp, struct trace_kprobe, tp);
224 }
225 
226 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
227 {
228 	struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
229 
230 	return tk ? (kprobe_on_func_entry(tk->rp.kp.addr,
231 			tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
232 			tk->rp.kp.addr ? 0 : tk->rp.kp.offset) == 0) : false;
233 }
234 
235 bool trace_kprobe_error_injectable(struct trace_event_call *call)
236 {
237 	struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
238 
239 	return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
240 	       false;
241 }
242 
243 static int register_kprobe_event(struct trace_kprobe *tk);
244 static int unregister_kprobe_event(struct trace_kprobe *tk);
245 
246 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
247 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
248 				struct pt_regs *regs);
249 
250 static void free_trace_kprobe(struct trace_kprobe *tk)
251 {
252 	if (tk) {
253 		trace_probe_cleanup(&tk->tp);
254 		kfree(tk->symbol);
255 		free_percpu(tk->nhit);
256 		kfree(tk);
257 	}
258 }
259 
260 /*
261  * Allocate new trace_probe and initialize it (including kprobes).
262  */
263 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
264 					     const char *event,
265 					     void *addr,
266 					     const char *symbol,
267 					     unsigned long offs,
268 					     int maxactive,
269 					     int nargs, bool is_return)
270 {
271 	struct trace_kprobe *tk;
272 	int ret = -ENOMEM;
273 
274 	tk = kzalloc(struct_size(tk, tp.args, nargs), GFP_KERNEL);
275 	if (!tk)
276 		return ERR_PTR(ret);
277 
278 	tk->nhit = alloc_percpu(unsigned long);
279 	if (!tk->nhit)
280 		goto error;
281 
282 	if (symbol) {
283 		tk->symbol = kstrdup(symbol, GFP_KERNEL);
284 		if (!tk->symbol)
285 			goto error;
286 		tk->rp.kp.symbol_name = tk->symbol;
287 		tk->rp.kp.offset = offs;
288 	} else
289 		tk->rp.kp.addr = addr;
290 
291 	if (is_return)
292 		tk->rp.handler = kretprobe_dispatcher;
293 	else
294 		tk->rp.kp.pre_handler = kprobe_dispatcher;
295 
296 	tk->rp.maxactive = maxactive;
297 	INIT_HLIST_NODE(&tk->rp.kp.hlist);
298 	INIT_LIST_HEAD(&tk->rp.kp.list);
299 
300 	ret = trace_probe_init(&tk->tp, event, group, false, nargs);
301 	if (ret < 0)
302 		goto error;
303 
304 	dyn_event_init(&tk->devent, &trace_kprobe_ops);
305 	return tk;
306 error:
307 	free_trace_kprobe(tk);
308 	return ERR_PTR(ret);
309 }
310 
311 static struct trace_kprobe *find_trace_kprobe(const char *event,
312 					      const char *group)
313 {
314 	struct dyn_event *pos;
315 	struct trace_kprobe *tk;
316 
317 	for_each_trace_kprobe(tk, pos)
318 		if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
319 		    strcmp(trace_probe_group_name(&tk->tp), group) == 0)
320 			return tk;
321 	return NULL;
322 }
323 
324 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
325 {
326 	int ret = 0;
327 
328 	if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
329 		if (trace_kprobe_is_return(tk))
330 			ret = enable_kretprobe(&tk->rp);
331 		else
332 			ret = enable_kprobe(&tk->rp.kp);
333 	}
334 
335 	return ret;
336 }
337 
338 static void __disable_trace_kprobe(struct trace_probe *tp)
339 {
340 	struct trace_kprobe *tk;
341 
342 	list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
343 		if (!trace_kprobe_is_registered(tk))
344 			continue;
345 		if (trace_kprobe_is_return(tk))
346 			disable_kretprobe(&tk->rp);
347 		else
348 			disable_kprobe(&tk->rp.kp);
349 	}
350 }
351 
352 /*
353  * Enable trace_probe
354  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
355  */
356 static int enable_trace_kprobe(struct trace_event_call *call,
357 				struct trace_event_file *file)
358 {
359 	struct trace_probe *tp;
360 	struct trace_kprobe *tk;
361 	bool enabled;
362 	int ret = 0;
363 
364 	tp = trace_probe_primary_from_call(call);
365 	if (WARN_ON_ONCE(!tp))
366 		return -ENODEV;
367 	enabled = trace_probe_is_enabled(tp);
368 
369 	/* This also changes "enabled" state */
370 	if (file) {
371 		ret = trace_probe_add_file(tp, file);
372 		if (ret)
373 			return ret;
374 	} else
375 		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
376 
377 	if (enabled)
378 		return 0;
379 
380 	list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
381 		if (trace_kprobe_has_gone(tk))
382 			continue;
383 		ret = __enable_trace_kprobe(tk);
384 		if (ret)
385 			break;
386 		enabled = true;
387 	}
388 
389 	if (ret) {
390 		/* Failed to enable one of them. Roll back all */
391 		if (enabled)
392 			__disable_trace_kprobe(tp);
393 		if (file)
394 			trace_probe_remove_file(tp, file);
395 		else
396 			trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
397 	}
398 
399 	return ret;
400 }
401 
402 /*
403  * Disable trace_probe
404  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
405  */
406 static int disable_trace_kprobe(struct trace_event_call *call,
407 				struct trace_event_file *file)
408 {
409 	struct trace_probe *tp;
410 
411 	tp = trace_probe_primary_from_call(call);
412 	if (WARN_ON_ONCE(!tp))
413 		return -ENODEV;
414 
415 	if (file) {
416 		if (!trace_probe_get_file_link(tp, file))
417 			return -ENOENT;
418 		if (!trace_probe_has_single_file(tp))
419 			goto out;
420 		trace_probe_clear_flag(tp, TP_FLAG_TRACE);
421 	} else
422 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
423 
424 	if (!trace_probe_is_enabled(tp))
425 		__disable_trace_kprobe(tp);
426 
427  out:
428 	if (file)
429 		/*
430 		 * Synchronization is done in below function. For perf event,
431 		 * file == NULL and perf_trace_event_unreg() calls
432 		 * tracepoint_synchronize_unregister() to ensure synchronize
433 		 * event. We don't need to care about it.
434 		 */
435 		trace_probe_remove_file(tp, file);
436 
437 	return 0;
438 }
439 
440 #if defined(CONFIG_DYNAMIC_FTRACE) && \
441 	!defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
442 static bool __within_notrace_func(unsigned long addr)
443 {
444 	unsigned long offset, size;
445 
446 	if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
447 		return false;
448 
449 	/* Get the entry address of the target function */
450 	addr -= offset;
451 
452 	/*
453 	 * Since ftrace_location_range() does inclusive range check, we need
454 	 * to subtract 1 byte from the end address.
455 	 */
456 	return !ftrace_location_range(addr, addr + size - 1);
457 }
458 
459 static bool within_notrace_func(struct trace_kprobe *tk)
460 {
461 	unsigned long addr = trace_kprobe_address(tk);
462 	char symname[KSYM_NAME_LEN], *p;
463 
464 	if (!__within_notrace_func(addr))
465 		return false;
466 
467 	/* Check if the address is on a suffixed-symbol */
468 	if (!lookup_symbol_name(addr, symname)) {
469 		p = strchr(symname, '.');
470 		if (!p)
471 			return true;
472 		*p = '\0';
473 		addr = (unsigned long)kprobe_lookup_name(symname, 0);
474 		if (addr)
475 			return __within_notrace_func(addr);
476 	}
477 
478 	return true;
479 }
480 #else
481 #define within_notrace_func(tk)	(false)
482 #endif
483 
484 /* Internal register function - just handle k*probes and flags */
485 static int __register_trace_kprobe(struct trace_kprobe *tk)
486 {
487 	int i, ret;
488 
489 	ret = security_locked_down(LOCKDOWN_KPROBES);
490 	if (ret)
491 		return ret;
492 
493 	if (trace_kprobe_is_registered(tk))
494 		return -EINVAL;
495 
496 	if (within_notrace_func(tk)) {
497 		pr_warn("Could not probe notrace function %ps\n",
498 			(void *)trace_kprobe_address(tk));
499 		return -EINVAL;
500 	}
501 
502 	for (i = 0; i < tk->tp.nr_args; i++) {
503 		ret = traceprobe_update_arg(&tk->tp.args[i]);
504 		if (ret)
505 			return ret;
506 	}
507 
508 	/* Set/clear disabled flag according to tp->flag */
509 	if (trace_probe_is_enabled(&tk->tp))
510 		tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
511 	else
512 		tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
513 
514 	if (trace_kprobe_is_return(tk))
515 		ret = register_kretprobe(&tk->rp);
516 	else
517 		ret = register_kprobe(&tk->rp.kp);
518 
519 	return ret;
520 }
521 
522 /* Internal unregister function - just handle k*probes and flags */
523 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
524 {
525 	if (trace_kprobe_is_registered(tk)) {
526 		if (trace_kprobe_is_return(tk))
527 			unregister_kretprobe(&tk->rp);
528 		else
529 			unregister_kprobe(&tk->rp.kp);
530 		/* Cleanup kprobe for reuse and mark it unregistered */
531 		INIT_HLIST_NODE(&tk->rp.kp.hlist);
532 		INIT_LIST_HEAD(&tk->rp.kp.list);
533 		if (tk->rp.kp.symbol_name)
534 			tk->rp.kp.addr = NULL;
535 	}
536 }
537 
538 /* Unregister a trace_probe and probe_event */
539 static int unregister_trace_kprobe(struct trace_kprobe *tk)
540 {
541 	/* If other probes are on the event, just unregister kprobe */
542 	if (trace_probe_has_sibling(&tk->tp))
543 		goto unreg;
544 
545 	/* Enabled event can not be unregistered */
546 	if (trace_probe_is_enabled(&tk->tp))
547 		return -EBUSY;
548 
549 	/* If there's a reference to the dynamic event */
550 	if (trace_event_dyn_busy(trace_probe_event_call(&tk->tp)))
551 		return -EBUSY;
552 
553 	/* Will fail if probe is being used by ftrace or perf */
554 	if (unregister_kprobe_event(tk))
555 		return -EBUSY;
556 
557 unreg:
558 	__unregister_trace_kprobe(tk);
559 	dyn_event_remove(&tk->devent);
560 	trace_probe_unlink(&tk->tp);
561 
562 	return 0;
563 }
564 
565 static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
566 					 struct trace_kprobe *comp)
567 {
568 	struct trace_probe_event *tpe = orig->tp.event;
569 	int i;
570 
571 	list_for_each_entry(orig, &tpe->probes, tp.list) {
572 		if (strcmp(trace_kprobe_symbol(orig),
573 			   trace_kprobe_symbol(comp)) ||
574 		    trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
575 			continue;
576 
577 		/*
578 		 * trace_probe_compare_arg_type() ensured that nr_args and
579 		 * each argument name and type are same. Let's compare comm.
580 		 */
581 		for (i = 0; i < orig->tp.nr_args; i++) {
582 			if (strcmp(orig->tp.args[i].comm,
583 				   comp->tp.args[i].comm))
584 				break;
585 		}
586 
587 		if (i == orig->tp.nr_args)
588 			return true;
589 	}
590 
591 	return false;
592 }
593 
594 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
595 {
596 	int ret;
597 
598 	ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
599 	if (ret) {
600 		/* Note that argument starts index = 2 */
601 		trace_probe_log_set_index(ret + 1);
602 		trace_probe_log_err(0, DIFF_ARG_TYPE);
603 		return -EEXIST;
604 	}
605 	if (trace_kprobe_has_same_kprobe(to, tk)) {
606 		trace_probe_log_set_index(0);
607 		trace_probe_log_err(0, SAME_PROBE);
608 		return -EEXIST;
609 	}
610 
611 	/* Append to existing event */
612 	ret = trace_probe_append(&tk->tp, &to->tp);
613 	if (ret)
614 		return ret;
615 
616 	/* Register k*probe */
617 	ret = __register_trace_kprobe(tk);
618 	if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
619 		pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
620 		ret = 0;
621 	}
622 
623 	if (ret)
624 		trace_probe_unlink(&tk->tp);
625 	else
626 		dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
627 
628 	return ret;
629 }
630 
631 /* Register a trace_probe and probe_event */
632 static int register_trace_kprobe(struct trace_kprobe *tk)
633 {
634 	struct trace_kprobe *old_tk;
635 	int ret;
636 
637 	mutex_lock(&event_mutex);
638 
639 	old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
640 				   trace_probe_group_name(&tk->tp));
641 	if (old_tk) {
642 		if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
643 			trace_probe_log_set_index(0);
644 			trace_probe_log_err(0, DIFF_PROBE_TYPE);
645 			ret = -EEXIST;
646 		} else {
647 			ret = append_trace_kprobe(tk, old_tk);
648 		}
649 		goto end;
650 	}
651 
652 	/* Register new event */
653 	ret = register_kprobe_event(tk);
654 	if (ret) {
655 		if (ret == -EEXIST) {
656 			trace_probe_log_set_index(0);
657 			trace_probe_log_err(0, EVENT_EXIST);
658 		} else
659 			pr_warn("Failed to register probe event(%d)\n", ret);
660 		goto end;
661 	}
662 
663 	/* Register k*probe */
664 	ret = __register_trace_kprobe(tk);
665 	if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
666 		pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
667 		ret = 0;
668 	}
669 
670 	if (ret < 0)
671 		unregister_kprobe_event(tk);
672 	else
673 		dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
674 
675 end:
676 	mutex_unlock(&event_mutex);
677 	return ret;
678 }
679 
680 #ifdef CONFIG_MODULES
681 static int validate_module_probe_symbol(const char *modname, const char *symbol);
682 
683 static int register_module_trace_kprobe(struct module *mod, struct trace_kprobe *tk)
684 {
685 	const char *p;
686 	int ret = 0;
687 
688 	p = strchr(trace_kprobe_symbol(tk), ':');
689 	if (p)
690 		ret = validate_module_probe_symbol(module_name(mod), p + 1);
691 	if (!ret)
692 		ret = __register_trace_kprobe(tk);
693 	return ret;
694 }
695 
696 /* Module notifier call back, checking event on the module */
697 static int trace_kprobe_module_callback(struct notifier_block *nb,
698 				       unsigned long val, void *data)
699 {
700 	struct module *mod = data;
701 	struct dyn_event *pos;
702 	struct trace_kprobe *tk;
703 	int ret;
704 
705 	if (val != MODULE_STATE_COMING)
706 		return NOTIFY_DONE;
707 
708 	/* Update probes on coming module */
709 	mutex_lock(&event_mutex);
710 	for_each_trace_kprobe(tk, pos) {
711 		if (trace_kprobe_within_module(tk, mod)) {
712 			/* Don't need to check busy - this should have gone. */
713 			__unregister_trace_kprobe(tk);
714 			ret = register_module_trace_kprobe(mod, tk);
715 			if (ret)
716 				pr_warn("Failed to re-register probe %s on %s: %d\n",
717 					trace_probe_name(&tk->tp),
718 					module_name(mod), ret);
719 		}
720 	}
721 	mutex_unlock(&event_mutex);
722 
723 	return NOTIFY_DONE;
724 }
725 
726 static struct notifier_block trace_kprobe_module_nb = {
727 	.notifier_call = trace_kprobe_module_callback,
728 	.priority = 1	/* Invoked after kprobe module callback */
729 };
730 static int trace_kprobe_register_module_notifier(void)
731 {
732 	return register_module_notifier(&trace_kprobe_module_nb);
733 }
734 #else
735 static int trace_kprobe_register_module_notifier(void)
736 {
737 	return 0;
738 }
739 #endif /* CONFIG_MODULES */
740 
741 static int count_symbols(void *data, unsigned long unused)
742 {
743 	unsigned int *count = data;
744 
745 	(*count)++;
746 
747 	return 0;
748 }
749 
750 struct sym_count_ctx {
751 	unsigned int count;
752 	const char *name;
753 };
754 
755 static int count_mod_symbols(void *data, const char *name, unsigned long unused)
756 {
757 	struct sym_count_ctx *ctx = data;
758 
759 	if (strcmp(name, ctx->name) == 0)
760 		ctx->count++;
761 
762 	return 0;
763 }
764 
765 static unsigned int number_of_same_symbols(const char *mod, const char *func_name)
766 {
767 	struct sym_count_ctx ctx = { .count = 0, .name = func_name };
768 
769 	if (!mod)
770 		kallsyms_on_each_match_symbol(count_symbols, func_name, &ctx.count);
771 
772 	module_kallsyms_on_each_symbol(mod, count_mod_symbols, &ctx);
773 
774 	return ctx.count;
775 }
776 
777 static int validate_module_probe_symbol(const char *modname, const char *symbol)
778 {
779 	unsigned int count = number_of_same_symbols(modname, symbol);
780 
781 	if (count > 1) {
782 		/*
783 		 * Users should use ADDR to remove the ambiguity of
784 		 * using KSYM only.
785 		 */
786 		return -EADDRNOTAVAIL;
787 	} else if (count == 0) {
788 		/*
789 		 * We can return ENOENT earlier than when register the
790 		 * kprobe.
791 		 */
792 		return -ENOENT;
793 	}
794 	return 0;
795 }
796 
797 #ifdef CONFIG_MODULES
798 /* Return NULL if the module is not loaded or under unloading. */
799 static struct module *try_module_get_by_name(const char *name)
800 {
801 	struct module *mod;
802 
803 	rcu_read_lock_sched();
804 	mod = find_module(name);
805 	if (mod && !try_module_get(mod))
806 		mod = NULL;
807 	rcu_read_unlock_sched();
808 
809 	return mod;
810 }
811 #else
812 #define try_module_get_by_name(name)	(NULL)
813 #endif
814 
815 static int validate_probe_symbol(char *symbol)
816 {
817 	struct module *mod = NULL;
818 	char *modname = NULL, *p;
819 	int ret = 0;
820 
821 	p = strchr(symbol, ':');
822 	if (p) {
823 		modname = symbol;
824 		symbol = p + 1;
825 		*p = '\0';
826 		mod = try_module_get_by_name(modname);
827 		if (!mod)
828 			goto out;
829 	}
830 
831 	ret = validate_module_probe_symbol(modname, symbol);
832 out:
833 	if (p)
834 		*p = ':';
835 	if (mod)
836 		module_put(mod);
837 	return ret;
838 }
839 
840 static int trace_kprobe_entry_handler(struct kretprobe_instance *ri,
841 				      struct pt_regs *regs);
842 
843 static int __trace_kprobe_create(int argc, const char *argv[])
844 {
845 	/*
846 	 * Argument syntax:
847 	 *  - Add kprobe:
848 	 *      p[:[GRP/][EVENT]] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
849 	 *  - Add kretprobe:
850 	 *      r[MAXACTIVE][:[GRP/][EVENT]] [MOD:]KSYM[+0] [FETCHARGS]
851 	 *    Or
852 	 *      p[:[GRP/][EVENT]] [MOD:]KSYM[+0]%return [FETCHARGS]
853 	 *
854 	 * Fetch args:
855 	 *  $retval	: fetch return value
856 	 *  $stack	: fetch stack address
857 	 *  $stackN	: fetch Nth of stack (N:0-)
858 	 *  $comm       : fetch current task comm
859 	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
860 	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
861 	 *  %REG	: fetch register REG
862 	 * Dereferencing memory fetch:
863 	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
864 	 * Alias name of args:
865 	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
866 	 * Type of args:
867 	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
868 	 */
869 	struct trace_kprobe *tk = NULL;
870 	int i, len, new_argc = 0, ret = 0;
871 	bool is_return = false;
872 	char *symbol = NULL, *tmp = NULL;
873 	const char **new_argv = NULL;
874 	const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
875 	enum probe_print_type ptype;
876 	int maxactive = 0;
877 	long offset = 0;
878 	void *addr = NULL;
879 	char buf[MAX_EVENT_NAME_LEN];
880 	char gbuf[MAX_EVENT_NAME_LEN];
881 	char abuf[MAX_BTF_ARGS_LEN];
882 	char *dbuf = NULL;
883 	struct traceprobe_parse_context ctx = { .flags = TPARG_FL_KERNEL };
884 
885 	switch (argv[0][0]) {
886 	case 'r':
887 		is_return = true;
888 		break;
889 	case 'p':
890 		break;
891 	default:
892 		return -ECANCELED;
893 	}
894 	if (argc < 2)
895 		return -ECANCELED;
896 
897 	trace_probe_log_init("trace_kprobe", argc, argv);
898 
899 	event = strchr(&argv[0][1], ':');
900 	if (event)
901 		event++;
902 
903 	if (isdigit(argv[0][1])) {
904 		if (!is_return) {
905 			trace_probe_log_err(1, BAD_MAXACT_TYPE);
906 			goto parse_error;
907 		}
908 		if (event)
909 			len = event - &argv[0][1] - 1;
910 		else
911 			len = strlen(&argv[0][1]);
912 		if (len > MAX_EVENT_NAME_LEN - 1) {
913 			trace_probe_log_err(1, BAD_MAXACT);
914 			goto parse_error;
915 		}
916 		memcpy(buf, &argv[0][1], len);
917 		buf[len] = '\0';
918 		ret = kstrtouint(buf, 0, &maxactive);
919 		if (ret || !maxactive) {
920 			trace_probe_log_err(1, BAD_MAXACT);
921 			goto parse_error;
922 		}
923 		/* kretprobes instances are iterated over via a list. The
924 		 * maximum should stay reasonable.
925 		 */
926 		if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
927 			trace_probe_log_err(1, MAXACT_TOO_BIG);
928 			goto parse_error;
929 		}
930 	}
931 
932 	/* try to parse an address. if that fails, try to read the
933 	 * input as a symbol. */
934 	if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
935 		trace_probe_log_set_index(1);
936 		/* Check whether uprobe event specified */
937 		if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
938 			ret = -ECANCELED;
939 			goto error;
940 		}
941 		/* a symbol specified */
942 		symbol = kstrdup(argv[1], GFP_KERNEL);
943 		if (!symbol)
944 			return -ENOMEM;
945 
946 		tmp = strchr(symbol, '%');
947 		if (tmp) {
948 			if (!strcmp(tmp, "%return")) {
949 				*tmp = '\0';
950 				is_return = true;
951 			} else {
952 				trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
953 				goto parse_error;
954 			}
955 		}
956 
957 		/* TODO: support .init module functions */
958 		ret = traceprobe_split_symbol_offset(symbol, &offset);
959 		if (ret || offset < 0 || offset > UINT_MAX) {
960 			trace_probe_log_err(0, BAD_PROBE_ADDR);
961 			goto parse_error;
962 		}
963 		ret = validate_probe_symbol(symbol);
964 		if (ret) {
965 			if (ret == -EADDRNOTAVAIL)
966 				trace_probe_log_err(0, NON_UNIQ_SYMBOL);
967 			else
968 				trace_probe_log_err(0, BAD_PROBE_ADDR);
969 			goto parse_error;
970 		}
971 		if (is_return)
972 			ctx.flags |= TPARG_FL_RETURN;
973 		ret = kprobe_on_func_entry(NULL, symbol, offset);
974 		if (ret == 0 && !is_return)
975 			ctx.flags |= TPARG_FL_FENTRY;
976 		/* Defer the ENOENT case until register kprobe */
977 		if (ret == -EINVAL && is_return) {
978 			trace_probe_log_err(0, BAD_RETPROBE);
979 			goto parse_error;
980 		}
981 	}
982 
983 	trace_probe_log_set_index(0);
984 	if (event) {
985 		ret = traceprobe_parse_event_name(&event, &group, gbuf,
986 						  event - argv[0]);
987 		if (ret)
988 			goto parse_error;
989 	}
990 
991 	if (!event) {
992 		/* Make a new event name */
993 		if (symbol)
994 			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
995 				 is_return ? 'r' : 'p', symbol, offset);
996 		else
997 			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
998 				 is_return ? 'r' : 'p', addr);
999 		sanitize_event_name(buf);
1000 		event = buf;
1001 	}
1002 
1003 	argc -= 2; argv += 2;
1004 	ctx.funcname = symbol;
1005 	new_argv = traceprobe_expand_meta_args(argc, argv, &new_argc,
1006 					       abuf, MAX_BTF_ARGS_LEN, &ctx);
1007 	if (IS_ERR(new_argv)) {
1008 		ret = PTR_ERR(new_argv);
1009 		new_argv = NULL;
1010 		goto out;
1011 	}
1012 	if (new_argv) {
1013 		argc = new_argc;
1014 		argv = new_argv;
1015 	}
1016 
1017 	ret = traceprobe_expand_dentry_args(argc, argv, &dbuf);
1018 	if (ret)
1019 		goto out;
1020 
1021 	/* setup a probe */
1022 	tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
1023 				argc, is_return);
1024 	if (IS_ERR(tk)) {
1025 		ret = PTR_ERR(tk);
1026 		/* This must return -ENOMEM, else there is a bug */
1027 		WARN_ON_ONCE(ret != -ENOMEM);
1028 		goto out;	/* We know tk is not allocated */
1029 	}
1030 
1031 	/* parse arguments */
1032 	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
1033 		trace_probe_log_set_index(i + 2);
1034 		ctx.offset = 0;
1035 		ret = traceprobe_parse_probe_arg(&tk->tp, i, argv[i], &ctx);
1036 		if (ret)
1037 			goto error;	/* This can be -ENOMEM */
1038 	}
1039 	/* entry handler for kretprobe */
1040 	if (is_return && tk->tp.entry_arg) {
1041 		tk->rp.entry_handler = trace_kprobe_entry_handler;
1042 		tk->rp.data_size = traceprobe_get_entry_data_size(&tk->tp);
1043 	}
1044 
1045 	ptype = is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1046 	ret = traceprobe_set_print_fmt(&tk->tp, ptype);
1047 	if (ret < 0)
1048 		goto error;
1049 
1050 	ret = register_trace_kprobe(tk);
1051 	if (ret) {
1052 		trace_probe_log_set_index(1);
1053 		if (ret == -EILSEQ)
1054 			trace_probe_log_err(0, BAD_INSN_BNDRY);
1055 		else if (ret == -ENOENT)
1056 			trace_probe_log_err(0, BAD_PROBE_ADDR);
1057 		else if (ret != -ENOMEM && ret != -EEXIST)
1058 			trace_probe_log_err(0, FAIL_REG_PROBE);
1059 		goto error;
1060 	}
1061 
1062 out:
1063 	traceprobe_finish_parse(&ctx);
1064 	trace_probe_log_clear();
1065 	kfree(new_argv);
1066 	kfree(symbol);
1067 	kfree(dbuf);
1068 	return ret;
1069 
1070 parse_error:
1071 	ret = -EINVAL;
1072 error:
1073 	free_trace_kprobe(tk);
1074 	goto out;
1075 }
1076 
1077 static int trace_kprobe_create(const char *raw_command)
1078 {
1079 	return trace_probe_create(raw_command, __trace_kprobe_create);
1080 }
1081 
1082 static int create_or_delete_trace_kprobe(const char *raw_command)
1083 {
1084 	int ret;
1085 
1086 	if (raw_command[0] == '-')
1087 		return dyn_event_release(raw_command, &trace_kprobe_ops);
1088 
1089 	ret = trace_kprobe_create(raw_command);
1090 	return ret == -ECANCELED ? -EINVAL : ret;
1091 }
1092 
1093 static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
1094 {
1095 	return create_or_delete_trace_kprobe(cmd->seq.buffer);
1096 }
1097 
1098 /**
1099  * kprobe_event_cmd_init - Initialize a kprobe event command object
1100  * @cmd: A pointer to the dynevent_cmd struct representing the new event
1101  * @buf: A pointer to the buffer used to build the command
1102  * @maxlen: The length of the buffer passed in @buf
1103  *
1104  * Initialize a synthetic event command object.  Use this before
1105  * calling any of the other kprobe_event functions.
1106  */
1107 void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
1108 {
1109 	dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
1110 			  trace_kprobe_run_command);
1111 }
1112 EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
1113 
1114 /**
1115  * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
1116  * @cmd: A pointer to the dynevent_cmd struct representing the new event
1117  * @kretprobe: Is this a return probe?
1118  * @name: The name of the kprobe event
1119  * @loc: The location of the kprobe event
1120  * @...: Variable number of arg (pairs), one pair for each field
1121  *
1122  * NOTE: Users normally won't want to call this function directly, but
1123  * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
1124  * adds a NULL to the end of the arg list.  If this function is used
1125  * directly, make sure the last arg in the variable arg list is NULL.
1126  *
1127  * Generate a kprobe event command to be executed by
1128  * kprobe_event_gen_cmd_end().  This function can be used to generate the
1129  * complete command or only the first part of it; in the latter case,
1130  * kprobe_event_add_fields() can be used to add more fields following this.
1131  *
1132  * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
1133  * returns -EINVAL if @loc == NULL.
1134  *
1135  * Return: 0 if successful, error otherwise.
1136  */
1137 int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
1138 				 const char *name, const char *loc, ...)
1139 {
1140 	char buf[MAX_EVENT_NAME_LEN];
1141 	struct dynevent_arg arg;
1142 	va_list args;
1143 	int ret;
1144 
1145 	if (cmd->type != DYNEVENT_TYPE_KPROBE)
1146 		return -EINVAL;
1147 
1148 	if (!loc)
1149 		return -EINVAL;
1150 
1151 	if (kretprobe)
1152 		snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
1153 	else
1154 		snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
1155 
1156 	ret = dynevent_str_add(cmd, buf);
1157 	if (ret)
1158 		return ret;
1159 
1160 	dynevent_arg_init(&arg, 0);
1161 	arg.str = loc;
1162 	ret = dynevent_arg_add(cmd, &arg, NULL);
1163 	if (ret)
1164 		return ret;
1165 
1166 	va_start(args, loc);
1167 	for (;;) {
1168 		const char *field;
1169 
1170 		field = va_arg(args, const char *);
1171 		if (!field)
1172 			break;
1173 
1174 		if (++cmd->n_fields > MAX_TRACE_ARGS) {
1175 			ret = -EINVAL;
1176 			break;
1177 		}
1178 
1179 		arg.str = field;
1180 		ret = dynevent_arg_add(cmd, &arg, NULL);
1181 		if (ret)
1182 			break;
1183 	}
1184 	va_end(args);
1185 
1186 	return ret;
1187 }
1188 EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1189 
1190 /**
1191  * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1192  * @cmd: A pointer to the dynevent_cmd struct representing the new event
1193  * @...: Variable number of arg (pairs), one pair for each field
1194  *
1195  * NOTE: Users normally won't want to call this function directly, but
1196  * rather use the kprobe_event_add_fields() wrapper, which
1197  * automatically adds a NULL to the end of the arg list.  If this
1198  * function is used directly, make sure the last arg in the variable
1199  * arg list is NULL.
1200  *
1201  * Add probe fields to an existing kprobe command using a variable
1202  * list of args.  Fields are added in the same order they're listed.
1203  *
1204  * Return: 0 if successful, error otherwise.
1205  */
1206 int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1207 {
1208 	struct dynevent_arg arg;
1209 	va_list args;
1210 	int ret = 0;
1211 
1212 	if (cmd->type != DYNEVENT_TYPE_KPROBE)
1213 		return -EINVAL;
1214 
1215 	dynevent_arg_init(&arg, 0);
1216 
1217 	va_start(args, cmd);
1218 	for (;;) {
1219 		const char *field;
1220 
1221 		field = va_arg(args, const char *);
1222 		if (!field)
1223 			break;
1224 
1225 		if (++cmd->n_fields > MAX_TRACE_ARGS) {
1226 			ret = -EINVAL;
1227 			break;
1228 		}
1229 
1230 		arg.str = field;
1231 		ret = dynevent_arg_add(cmd, &arg, NULL);
1232 		if (ret)
1233 			break;
1234 	}
1235 	va_end(args);
1236 
1237 	return ret;
1238 }
1239 EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1240 
1241 /**
1242  * kprobe_event_delete - Delete a kprobe event
1243  * @name: The name of the kprobe event to delete
1244  *
1245  * Delete a kprobe event with the give @name from kernel code rather
1246  * than directly from the command line.
1247  *
1248  * Return: 0 if successful, error otherwise.
1249  */
1250 int kprobe_event_delete(const char *name)
1251 {
1252 	char buf[MAX_EVENT_NAME_LEN];
1253 
1254 	snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1255 
1256 	return create_or_delete_trace_kprobe(buf);
1257 }
1258 EXPORT_SYMBOL_GPL(kprobe_event_delete);
1259 
1260 static int trace_kprobe_release(struct dyn_event *ev)
1261 {
1262 	struct trace_kprobe *tk = to_trace_kprobe(ev);
1263 	int ret = unregister_trace_kprobe(tk);
1264 
1265 	if (!ret)
1266 		free_trace_kprobe(tk);
1267 	return ret;
1268 }
1269 
1270 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1271 {
1272 	struct trace_kprobe *tk = to_trace_kprobe(ev);
1273 	int i;
1274 
1275 	seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1276 	if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1277 		seq_printf(m, "%d", tk->rp.maxactive);
1278 	seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1279 				trace_probe_name(&tk->tp));
1280 
1281 	if (!tk->symbol)
1282 		seq_printf(m, " 0x%p", tk->rp.kp.addr);
1283 	else if (tk->rp.kp.offset)
1284 		seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1285 			   tk->rp.kp.offset);
1286 	else
1287 		seq_printf(m, " %s", trace_kprobe_symbol(tk));
1288 
1289 	for (i = 0; i < tk->tp.nr_args; i++)
1290 		seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1291 	seq_putc(m, '\n');
1292 
1293 	return 0;
1294 }
1295 
1296 static int probes_seq_show(struct seq_file *m, void *v)
1297 {
1298 	struct dyn_event *ev = v;
1299 
1300 	if (!is_trace_kprobe(ev))
1301 		return 0;
1302 
1303 	return trace_kprobe_show(m, ev);
1304 }
1305 
1306 static const struct seq_operations probes_seq_op = {
1307 	.start  = dyn_event_seq_start,
1308 	.next   = dyn_event_seq_next,
1309 	.stop   = dyn_event_seq_stop,
1310 	.show   = probes_seq_show
1311 };
1312 
1313 static int probes_open(struct inode *inode, struct file *file)
1314 {
1315 	int ret;
1316 
1317 	ret = security_locked_down(LOCKDOWN_TRACEFS);
1318 	if (ret)
1319 		return ret;
1320 
1321 	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1322 		ret = dyn_events_release_all(&trace_kprobe_ops);
1323 		if (ret < 0)
1324 			return ret;
1325 	}
1326 
1327 	return seq_open(file, &probes_seq_op);
1328 }
1329 
1330 static ssize_t probes_write(struct file *file, const char __user *buffer,
1331 			    size_t count, loff_t *ppos)
1332 {
1333 	return trace_parse_run_command(file, buffer, count, ppos,
1334 				       create_or_delete_trace_kprobe);
1335 }
1336 
1337 static const struct file_operations kprobe_events_ops = {
1338 	.owner          = THIS_MODULE,
1339 	.open           = probes_open,
1340 	.read           = seq_read,
1341 	.llseek         = seq_lseek,
1342 	.release        = seq_release,
1343 	.write		= probes_write,
1344 };
1345 
1346 static unsigned long trace_kprobe_missed(struct trace_kprobe *tk)
1347 {
1348 	return trace_kprobe_is_return(tk) ?
1349 		tk->rp.kp.nmissed + tk->rp.nmissed : tk->rp.kp.nmissed;
1350 }
1351 
1352 /* Probes profiling interfaces */
1353 static int probes_profile_seq_show(struct seq_file *m, void *v)
1354 {
1355 	struct dyn_event *ev = v;
1356 	struct trace_kprobe *tk;
1357 	unsigned long nmissed;
1358 
1359 	if (!is_trace_kprobe(ev))
1360 		return 0;
1361 
1362 	tk = to_trace_kprobe(ev);
1363 	nmissed = trace_kprobe_missed(tk);
1364 	seq_printf(m, "  %-44s %15lu %15lu\n",
1365 		   trace_probe_name(&tk->tp),
1366 		   trace_kprobe_nhit(tk),
1367 		   nmissed);
1368 
1369 	return 0;
1370 }
1371 
1372 static const struct seq_operations profile_seq_op = {
1373 	.start  = dyn_event_seq_start,
1374 	.next   = dyn_event_seq_next,
1375 	.stop   = dyn_event_seq_stop,
1376 	.show   = probes_profile_seq_show
1377 };
1378 
1379 static int profile_open(struct inode *inode, struct file *file)
1380 {
1381 	int ret;
1382 
1383 	ret = security_locked_down(LOCKDOWN_TRACEFS);
1384 	if (ret)
1385 		return ret;
1386 
1387 	return seq_open(file, &profile_seq_op);
1388 }
1389 
1390 static const struct file_operations kprobe_profile_ops = {
1391 	.owner          = THIS_MODULE,
1392 	.open           = profile_open,
1393 	.read           = seq_read,
1394 	.llseek         = seq_lseek,
1395 	.release        = seq_release,
1396 };
1397 
1398 /* Note that we don't verify it, since the code does not come from user space */
1399 static int
1400 process_fetch_insn(struct fetch_insn *code, void *rec, void *edata,
1401 		   void *dest, void *base)
1402 {
1403 	struct pt_regs *regs = rec;
1404 	unsigned long val;
1405 	int ret;
1406 
1407 retry:
1408 	/* 1st stage: get value from context */
1409 	switch (code->op) {
1410 	case FETCH_OP_REG:
1411 		val = regs_get_register(regs, code->param);
1412 		break;
1413 	case FETCH_OP_STACK:
1414 		val = regs_get_kernel_stack_nth(regs, code->param);
1415 		break;
1416 	case FETCH_OP_STACKP:
1417 		val = kernel_stack_pointer(regs);
1418 		break;
1419 	case FETCH_OP_RETVAL:
1420 		val = regs_return_value(regs);
1421 		break;
1422 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1423 	case FETCH_OP_ARG:
1424 		val = regs_get_kernel_argument(regs, code->param);
1425 		break;
1426 	case FETCH_OP_EDATA:
1427 		val = *(unsigned long *)((unsigned long)edata + code->offset);
1428 		break;
1429 #endif
1430 	case FETCH_NOP_SYMBOL:	/* Ignore a place holder */
1431 		code++;
1432 		goto retry;
1433 	default:
1434 		ret = process_common_fetch_insn(code, &val);
1435 		if (ret < 0)
1436 			return ret;
1437 	}
1438 	code++;
1439 
1440 	return process_fetch_insn_bottom(code, val, dest, base);
1441 }
1442 NOKPROBE_SYMBOL(process_fetch_insn)
1443 
1444 /* Kprobe handler */
1445 static nokprobe_inline void
1446 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1447 		    struct trace_event_file *trace_file)
1448 {
1449 	struct kprobe_trace_entry_head *entry;
1450 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1451 	struct trace_event_buffer fbuffer;
1452 	int dsize;
1453 
1454 	WARN_ON(call != trace_file->event_call);
1455 
1456 	if (trace_trigger_soft_disabled(trace_file))
1457 		return;
1458 
1459 	dsize = __get_data_size(&tk->tp, regs, NULL);
1460 
1461 	entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1462 					   sizeof(*entry) + tk->tp.size + dsize);
1463 	if (!entry)
1464 		return;
1465 
1466 	fbuffer.regs = regs;
1467 	entry->ip = (unsigned long)tk->rp.kp.addr;
1468 	store_trace_args(&entry[1], &tk->tp, regs, NULL, sizeof(*entry), dsize);
1469 
1470 	trace_event_buffer_commit(&fbuffer);
1471 }
1472 
1473 static void
1474 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1475 {
1476 	struct event_file_link *link;
1477 
1478 	trace_probe_for_each_link_rcu(link, &tk->tp)
1479 		__kprobe_trace_func(tk, regs, link->file);
1480 }
1481 NOKPROBE_SYMBOL(kprobe_trace_func);
1482 
1483 /* Kretprobe handler */
1484 
1485 static int trace_kprobe_entry_handler(struct kretprobe_instance *ri,
1486 				      struct pt_regs *regs)
1487 {
1488 	struct kretprobe *rp = get_kretprobe(ri);
1489 	struct trace_kprobe *tk;
1490 
1491 	/*
1492 	 * There is a small chance that get_kretprobe(ri) returns NULL when
1493 	 * the kretprobe is unregister on another CPU between kretprobe's
1494 	 * trampoline_handler and this function.
1495 	 */
1496 	if (unlikely(!rp))
1497 		return -ENOENT;
1498 
1499 	tk = container_of(rp, struct trace_kprobe, rp);
1500 
1501 	/* store argument values into ri->data as entry data */
1502 	if (tk->tp.entry_arg)
1503 		store_trace_entry_data(ri->data, &tk->tp, regs);
1504 
1505 	return 0;
1506 }
1507 
1508 
1509 static nokprobe_inline void
1510 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1511 		       struct pt_regs *regs,
1512 		       struct trace_event_file *trace_file)
1513 {
1514 	struct kretprobe_trace_entry_head *entry;
1515 	struct trace_event_buffer fbuffer;
1516 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1517 	int dsize;
1518 
1519 	WARN_ON(call != trace_file->event_call);
1520 
1521 	if (trace_trigger_soft_disabled(trace_file))
1522 		return;
1523 
1524 	dsize = __get_data_size(&tk->tp, regs, ri->data);
1525 
1526 	entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1527 					   sizeof(*entry) + tk->tp.size + dsize);
1528 	if (!entry)
1529 		return;
1530 
1531 	fbuffer.regs = regs;
1532 	entry->func = (unsigned long)tk->rp.kp.addr;
1533 	entry->ret_ip = get_kretprobe_retaddr(ri);
1534 	store_trace_args(&entry[1], &tk->tp, regs, ri->data, sizeof(*entry), dsize);
1535 
1536 	trace_event_buffer_commit(&fbuffer);
1537 }
1538 
1539 static void
1540 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1541 		     struct pt_regs *regs)
1542 {
1543 	struct event_file_link *link;
1544 
1545 	trace_probe_for_each_link_rcu(link, &tk->tp)
1546 		__kretprobe_trace_func(tk, ri, regs, link->file);
1547 }
1548 NOKPROBE_SYMBOL(kretprobe_trace_func);
1549 
1550 /* Event entry printers */
1551 static enum print_line_t
1552 print_kprobe_event(struct trace_iterator *iter, int flags,
1553 		   struct trace_event *event)
1554 {
1555 	struct kprobe_trace_entry_head *field;
1556 	struct trace_seq *s = &iter->seq;
1557 	struct trace_probe *tp;
1558 
1559 	field = (struct kprobe_trace_entry_head *)iter->ent;
1560 	tp = trace_probe_primary_from_call(
1561 		container_of(event, struct trace_event_call, event));
1562 	if (WARN_ON_ONCE(!tp))
1563 		goto out;
1564 
1565 	trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1566 
1567 	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1568 		goto out;
1569 
1570 	trace_seq_putc(s, ')');
1571 
1572 	if (trace_probe_print_args(s, tp->args, tp->nr_args,
1573 			     (u8 *)&field[1], field) < 0)
1574 		goto out;
1575 
1576 	trace_seq_putc(s, '\n');
1577  out:
1578 	return trace_handle_return(s);
1579 }
1580 
1581 static enum print_line_t
1582 print_kretprobe_event(struct trace_iterator *iter, int flags,
1583 		      struct trace_event *event)
1584 {
1585 	struct kretprobe_trace_entry_head *field;
1586 	struct trace_seq *s = &iter->seq;
1587 	struct trace_probe *tp;
1588 
1589 	field = (struct kretprobe_trace_entry_head *)iter->ent;
1590 	tp = trace_probe_primary_from_call(
1591 		container_of(event, struct trace_event_call, event));
1592 	if (WARN_ON_ONCE(!tp))
1593 		goto out;
1594 
1595 	trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1596 
1597 	if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1598 		goto out;
1599 
1600 	trace_seq_puts(s, " <- ");
1601 
1602 	if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1603 		goto out;
1604 
1605 	trace_seq_putc(s, ')');
1606 
1607 	if (trace_probe_print_args(s, tp->args, tp->nr_args,
1608 			     (u8 *)&field[1], field) < 0)
1609 		goto out;
1610 
1611 	trace_seq_putc(s, '\n');
1612 
1613  out:
1614 	return trace_handle_return(s);
1615 }
1616 
1617 
1618 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1619 {
1620 	int ret;
1621 	struct kprobe_trace_entry_head field;
1622 	struct trace_probe *tp;
1623 
1624 	tp = trace_probe_primary_from_call(event_call);
1625 	if (WARN_ON_ONCE(!tp))
1626 		return -ENOENT;
1627 
1628 	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1629 
1630 	return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1631 }
1632 
1633 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1634 {
1635 	int ret;
1636 	struct kretprobe_trace_entry_head field;
1637 	struct trace_probe *tp;
1638 
1639 	tp = trace_probe_primary_from_call(event_call);
1640 	if (WARN_ON_ONCE(!tp))
1641 		return -ENOENT;
1642 
1643 	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1644 	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1645 
1646 	return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1647 }
1648 
1649 #ifdef CONFIG_PERF_EVENTS
1650 
1651 /* Kprobe profile handler */
1652 static int
1653 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1654 {
1655 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1656 	struct kprobe_trace_entry_head *entry;
1657 	struct hlist_head *head;
1658 	int size, __size, dsize;
1659 	int rctx;
1660 
1661 	if (bpf_prog_array_valid(call)) {
1662 		unsigned long orig_ip = instruction_pointer(regs);
1663 		int ret;
1664 
1665 		ret = trace_call_bpf(call, regs);
1666 
1667 		/*
1668 		 * We need to check and see if we modified the pc of the
1669 		 * pt_regs, and if so return 1 so that we don't do the
1670 		 * single stepping.
1671 		 */
1672 		if (orig_ip != instruction_pointer(regs))
1673 			return 1;
1674 		if (!ret)
1675 			return 0;
1676 	}
1677 
1678 	head = this_cpu_ptr(call->perf_events);
1679 	if (hlist_empty(head))
1680 		return 0;
1681 
1682 	dsize = __get_data_size(&tk->tp, regs, NULL);
1683 	__size = sizeof(*entry) + tk->tp.size + dsize;
1684 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1685 	size -= sizeof(u32);
1686 
1687 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1688 	if (!entry)
1689 		return 0;
1690 
1691 	entry->ip = (unsigned long)tk->rp.kp.addr;
1692 	memset(&entry[1], 0, dsize);
1693 	store_trace_args(&entry[1], &tk->tp, regs, NULL, sizeof(*entry), dsize);
1694 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1695 			      head, NULL);
1696 	return 0;
1697 }
1698 NOKPROBE_SYMBOL(kprobe_perf_func);
1699 
1700 /* Kretprobe profile handler */
1701 static void
1702 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1703 		    struct pt_regs *regs)
1704 {
1705 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1706 	struct kretprobe_trace_entry_head *entry;
1707 	struct hlist_head *head;
1708 	int size, __size, dsize;
1709 	int rctx;
1710 
1711 	if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1712 		return;
1713 
1714 	head = this_cpu_ptr(call->perf_events);
1715 	if (hlist_empty(head))
1716 		return;
1717 
1718 	dsize = __get_data_size(&tk->tp, regs, ri->data);
1719 	__size = sizeof(*entry) + tk->tp.size + dsize;
1720 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1721 	size -= sizeof(u32);
1722 
1723 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1724 	if (!entry)
1725 		return;
1726 
1727 	entry->func = (unsigned long)tk->rp.kp.addr;
1728 	entry->ret_ip = get_kretprobe_retaddr(ri);
1729 	store_trace_args(&entry[1], &tk->tp, regs, ri->data, sizeof(*entry), dsize);
1730 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1731 			      head, NULL);
1732 }
1733 NOKPROBE_SYMBOL(kretprobe_perf_func);
1734 
1735 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1736 			const char **symbol, u64 *probe_offset,
1737 			u64 *probe_addr, unsigned long *missed,
1738 			bool perf_type_tracepoint)
1739 {
1740 	const char *pevent = trace_event_name(event->tp_event);
1741 	const char *group = event->tp_event->class->system;
1742 	struct trace_kprobe *tk;
1743 
1744 	if (perf_type_tracepoint)
1745 		tk = find_trace_kprobe(pevent, group);
1746 	else
1747 		tk = trace_kprobe_primary_from_call(event->tp_event);
1748 	if (!tk)
1749 		return -EINVAL;
1750 
1751 	*fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1752 					      : BPF_FD_TYPE_KPROBE;
1753 	*probe_offset = tk->rp.kp.offset;
1754 	*probe_addr = kallsyms_show_value(current_cred()) ?
1755 		      (unsigned long)tk->rp.kp.addr : 0;
1756 	*symbol = tk->symbol;
1757 	if (missed)
1758 		*missed = trace_kprobe_missed(tk);
1759 	return 0;
1760 }
1761 #endif	/* CONFIG_PERF_EVENTS */
1762 
1763 /*
1764  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1765  *
1766  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1767  * lockless, but we can't race with this __init function.
1768  */
1769 static int kprobe_register(struct trace_event_call *event,
1770 			   enum trace_reg type, void *data)
1771 {
1772 	struct trace_event_file *file = data;
1773 
1774 	switch (type) {
1775 	case TRACE_REG_REGISTER:
1776 		return enable_trace_kprobe(event, file);
1777 	case TRACE_REG_UNREGISTER:
1778 		return disable_trace_kprobe(event, file);
1779 
1780 #ifdef CONFIG_PERF_EVENTS
1781 	case TRACE_REG_PERF_REGISTER:
1782 		return enable_trace_kprobe(event, NULL);
1783 	case TRACE_REG_PERF_UNREGISTER:
1784 		return disable_trace_kprobe(event, NULL);
1785 	case TRACE_REG_PERF_OPEN:
1786 	case TRACE_REG_PERF_CLOSE:
1787 	case TRACE_REG_PERF_ADD:
1788 	case TRACE_REG_PERF_DEL:
1789 		return 0;
1790 #endif
1791 	}
1792 	return 0;
1793 }
1794 
1795 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1796 {
1797 	struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1798 	int ret = 0;
1799 
1800 	raw_cpu_inc(*tk->nhit);
1801 
1802 	if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1803 		kprobe_trace_func(tk, regs);
1804 #ifdef CONFIG_PERF_EVENTS
1805 	if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1806 		ret = kprobe_perf_func(tk, regs);
1807 #endif
1808 	return ret;
1809 }
1810 NOKPROBE_SYMBOL(kprobe_dispatcher);
1811 
1812 static int
1813 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1814 {
1815 	struct kretprobe *rp = get_kretprobe(ri);
1816 	struct trace_kprobe *tk;
1817 
1818 	/*
1819 	 * There is a small chance that get_kretprobe(ri) returns NULL when
1820 	 * the kretprobe is unregister on another CPU between kretprobe's
1821 	 * trampoline_handler and this function.
1822 	 */
1823 	if (unlikely(!rp))
1824 		return 0;
1825 
1826 	tk = container_of(rp, struct trace_kprobe, rp);
1827 	raw_cpu_inc(*tk->nhit);
1828 
1829 	if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1830 		kretprobe_trace_func(tk, ri, regs);
1831 #ifdef CONFIG_PERF_EVENTS
1832 	if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1833 		kretprobe_perf_func(tk, ri, regs);
1834 #endif
1835 	return 0;	/* We don't tweak kernel, so just return 0 */
1836 }
1837 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1838 
1839 static struct trace_event_functions kretprobe_funcs = {
1840 	.trace		= print_kretprobe_event
1841 };
1842 
1843 static struct trace_event_functions kprobe_funcs = {
1844 	.trace		= print_kprobe_event
1845 };
1846 
1847 static struct trace_event_fields kretprobe_fields_array[] = {
1848 	{ .type = TRACE_FUNCTION_TYPE,
1849 	  .define_fields = kretprobe_event_define_fields },
1850 	{}
1851 };
1852 
1853 static struct trace_event_fields kprobe_fields_array[] = {
1854 	{ .type = TRACE_FUNCTION_TYPE,
1855 	  .define_fields = kprobe_event_define_fields },
1856 	{}
1857 };
1858 
1859 static inline void init_trace_event_call(struct trace_kprobe *tk)
1860 {
1861 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1862 
1863 	if (trace_kprobe_is_return(tk)) {
1864 		call->event.funcs = &kretprobe_funcs;
1865 		call->class->fields_array = kretprobe_fields_array;
1866 	} else {
1867 		call->event.funcs = &kprobe_funcs;
1868 		call->class->fields_array = kprobe_fields_array;
1869 	}
1870 
1871 	call->flags = TRACE_EVENT_FL_KPROBE;
1872 	call->class->reg = kprobe_register;
1873 }
1874 
1875 static int register_kprobe_event(struct trace_kprobe *tk)
1876 {
1877 	init_trace_event_call(tk);
1878 
1879 	return trace_probe_register_event_call(&tk->tp);
1880 }
1881 
1882 static int unregister_kprobe_event(struct trace_kprobe *tk)
1883 {
1884 	return trace_probe_unregister_event_call(&tk->tp);
1885 }
1886 
1887 #ifdef CONFIG_PERF_EVENTS
1888 
1889 /* create a trace_kprobe, but don't add it to global lists */
1890 struct trace_event_call *
1891 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1892 			  bool is_return)
1893 {
1894 	enum probe_print_type ptype;
1895 	struct trace_kprobe *tk;
1896 	int ret;
1897 	char *event;
1898 
1899 	if (func) {
1900 		ret = validate_probe_symbol(func);
1901 		if (ret)
1902 			return ERR_PTR(ret);
1903 	}
1904 
1905 	/*
1906 	 * local trace_kprobes are not added to dyn_event, so they are never
1907 	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1908 	 * duplicated name here.
1909 	 */
1910 	event = func ? func : "DUMMY_EVENT";
1911 
1912 	tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1913 				offs, 0 /* maxactive */, 0 /* nargs */,
1914 				is_return);
1915 
1916 	if (IS_ERR(tk)) {
1917 		pr_info("Failed to allocate trace_probe.(%d)\n",
1918 			(int)PTR_ERR(tk));
1919 		return ERR_CAST(tk);
1920 	}
1921 
1922 	init_trace_event_call(tk);
1923 
1924 	ptype = trace_kprobe_is_return(tk) ?
1925 		PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1926 	if (traceprobe_set_print_fmt(&tk->tp, ptype) < 0) {
1927 		ret = -ENOMEM;
1928 		goto error;
1929 	}
1930 
1931 	ret = __register_trace_kprobe(tk);
1932 	if (ret < 0)
1933 		goto error;
1934 
1935 	return trace_probe_event_call(&tk->tp);
1936 error:
1937 	free_trace_kprobe(tk);
1938 	return ERR_PTR(ret);
1939 }
1940 
1941 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1942 {
1943 	struct trace_kprobe *tk;
1944 
1945 	tk = trace_kprobe_primary_from_call(event_call);
1946 	if (unlikely(!tk))
1947 		return;
1948 
1949 	if (trace_probe_is_enabled(&tk->tp)) {
1950 		WARN_ON(1);
1951 		return;
1952 	}
1953 
1954 	__unregister_trace_kprobe(tk);
1955 
1956 	free_trace_kprobe(tk);
1957 }
1958 #endif /* CONFIG_PERF_EVENTS */
1959 
1960 static __init void enable_boot_kprobe_events(void)
1961 {
1962 	struct trace_array *tr = top_trace_array();
1963 	struct trace_event_file *file;
1964 	struct trace_kprobe *tk;
1965 	struct dyn_event *pos;
1966 
1967 	mutex_lock(&event_mutex);
1968 	for_each_trace_kprobe(tk, pos) {
1969 		list_for_each_entry(file, &tr->events, list)
1970 			if (file->event_call == trace_probe_event_call(&tk->tp))
1971 				trace_event_enable_disable(file, 1, 0);
1972 	}
1973 	mutex_unlock(&event_mutex);
1974 }
1975 
1976 static __init void setup_boot_kprobe_events(void)
1977 {
1978 	char *p, *cmd = kprobe_boot_events_buf;
1979 	int ret;
1980 
1981 	strreplace(kprobe_boot_events_buf, ',', ' ');
1982 
1983 	while (cmd && *cmd != '\0') {
1984 		p = strchr(cmd, ';');
1985 		if (p)
1986 			*p++ = '\0';
1987 
1988 		ret = create_or_delete_trace_kprobe(cmd);
1989 		if (ret)
1990 			pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1991 
1992 		cmd = p;
1993 	}
1994 
1995 	enable_boot_kprobe_events();
1996 }
1997 
1998 /*
1999  * Register dynevent at core_initcall. This allows kernel to setup kprobe
2000  * events in postcore_initcall without tracefs.
2001  */
2002 static __init int init_kprobe_trace_early(void)
2003 {
2004 	int ret;
2005 
2006 	ret = dyn_event_register(&trace_kprobe_ops);
2007 	if (ret)
2008 		return ret;
2009 
2010 	if (trace_kprobe_register_module_notifier())
2011 		return -EINVAL;
2012 
2013 	return 0;
2014 }
2015 core_initcall(init_kprobe_trace_early);
2016 
2017 /* Make a tracefs interface for controlling probe points */
2018 static __init int init_kprobe_trace(void)
2019 {
2020 	int ret;
2021 
2022 	ret = tracing_init_dentry();
2023 	if (ret)
2024 		return 0;
2025 
2026 	/* Event list interface */
2027 	trace_create_file("kprobe_events", TRACE_MODE_WRITE,
2028 			  NULL, NULL, &kprobe_events_ops);
2029 
2030 	/* Profile interface */
2031 	trace_create_file("kprobe_profile", TRACE_MODE_READ,
2032 			  NULL, NULL, &kprobe_profile_ops);
2033 
2034 	setup_boot_kprobe_events();
2035 
2036 	return 0;
2037 }
2038 fs_initcall(init_kprobe_trace);
2039 
2040 
2041 #ifdef CONFIG_FTRACE_STARTUP_TEST
2042 static __init struct trace_event_file *
2043 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
2044 {
2045 	struct trace_event_file *file;
2046 
2047 	list_for_each_entry(file, &tr->events, list)
2048 		if (file->event_call == trace_probe_event_call(&tk->tp))
2049 			return file;
2050 
2051 	return NULL;
2052 }
2053 
2054 /*
2055  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
2056  * stage, we can do this lockless.
2057  */
2058 static __init int kprobe_trace_self_tests_init(void)
2059 {
2060 	int ret, warn = 0;
2061 	int (*target)(int, int, int, int, int, int);
2062 	struct trace_kprobe *tk;
2063 	struct trace_event_file *file;
2064 
2065 	if (tracing_is_disabled())
2066 		return -ENODEV;
2067 
2068 	if (tracing_selftest_disabled)
2069 		return 0;
2070 
2071 	target = kprobe_trace_selftest_target;
2072 
2073 	pr_info("Testing kprobe tracing: ");
2074 
2075 	ret = create_or_delete_trace_kprobe("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)");
2076 	if (WARN_ONCE(ret, "error on probing function entry.")) {
2077 		warn++;
2078 	} else {
2079 		/* Enable trace point */
2080 		tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2081 		if (WARN_ONCE(tk == NULL, "error on probing function entry.")) {
2082 			warn++;
2083 		} else {
2084 			file = find_trace_probe_file(tk, top_trace_array());
2085 			if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2086 				warn++;
2087 			} else
2088 				enable_trace_kprobe(
2089 					trace_probe_event_call(&tk->tp), file);
2090 		}
2091 	}
2092 
2093 	ret = create_or_delete_trace_kprobe("r:testprobe2 kprobe_trace_selftest_target $retval");
2094 	if (WARN_ONCE(ret, "error on probing function return.")) {
2095 		warn++;
2096 	} else {
2097 		/* Enable trace point */
2098 		tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2099 		if (WARN_ONCE(tk == NULL, "error on getting 2nd new probe.")) {
2100 			warn++;
2101 		} else {
2102 			file = find_trace_probe_file(tk, top_trace_array());
2103 			if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2104 				warn++;
2105 			} else
2106 				enable_trace_kprobe(
2107 					trace_probe_event_call(&tk->tp), file);
2108 		}
2109 	}
2110 
2111 	if (warn)
2112 		goto end;
2113 
2114 	ret = target(1, 2, 3, 4, 5, 6);
2115 
2116 	/*
2117 	 * Not expecting an error here, the check is only to prevent the
2118 	 * optimizer from removing the call to target() as otherwise there
2119 	 * are no side-effects and the call is never performed.
2120 	 */
2121 	if (ret != 21)
2122 		warn++;
2123 
2124 	/* Disable trace points before removing it */
2125 	tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2126 	if (WARN_ONCE(tk == NULL, "error on getting test probe.")) {
2127 		warn++;
2128 	} else {
2129 		if (WARN_ONCE(trace_kprobe_nhit(tk) != 1,
2130 				 "incorrect number of testprobe hits."))
2131 			warn++;
2132 
2133 		file = find_trace_probe_file(tk, top_trace_array());
2134 		if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2135 			warn++;
2136 		} else
2137 			disable_trace_kprobe(
2138 				trace_probe_event_call(&tk->tp), file);
2139 	}
2140 
2141 	tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2142 	if (WARN_ONCE(tk == NULL, "error on getting 2nd test probe.")) {
2143 		warn++;
2144 	} else {
2145 		if (WARN_ONCE(trace_kprobe_nhit(tk) != 1,
2146 				 "incorrect number of testprobe2 hits."))
2147 			warn++;
2148 
2149 		file = find_trace_probe_file(tk, top_trace_array());
2150 		if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2151 			warn++;
2152 		} else
2153 			disable_trace_kprobe(
2154 				trace_probe_event_call(&tk->tp), file);
2155 	}
2156 
2157 	ret = create_or_delete_trace_kprobe("-:testprobe");
2158 	if (WARN_ONCE(ret, "error on deleting a probe."))
2159 		warn++;
2160 
2161 	ret = create_or_delete_trace_kprobe("-:testprobe2");
2162 	if (WARN_ONCE(ret, "error on deleting a probe."))
2163 		warn++;
2164 
2165 
2166 end:
2167 	/*
2168 	 * Wait for the optimizer work to finish. Otherwise it might fiddle
2169 	 * with probes in already freed __init text.
2170 	 */
2171 	wait_for_kprobe_optimizer();
2172 	if (warn)
2173 		pr_cont("NG: Some tests are failed. Please check them.\n");
2174 	else
2175 		pr_cont("OK\n");
2176 	return 0;
2177 }
2178 
2179 late_initcall(kprobe_trace_self_tests_init);
2180 
2181 #endif
2182