xref: /linux/kernel/trace/trace_kprobe.c (revision 23b0f90ba871f096474e1c27c3d14f455189d2d9)
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/cleanup.h>
12 #include <linux/error-injection.h>
13 #include <linux/module.h>
14 #include <linux/rculist.h>
15 #include <linux/security.h>
16 #include <linux/uaccess.h>
17 
18 #include <asm/setup.h>  /* for COMMAND_LINE_SIZE */
19 
20 #include "trace_dynevent.h"
21 #include "trace_kprobe_selftest.h"
22 #include "trace_probe.h"
23 #include "trace_probe_kernel.h"
24 #include "trace_probe_tmpl.h"
25 
26 #define KPROBE_EVENT_SYSTEM "kprobes"
27 #define KRETPROBE_MAXACTIVE_MAX 4096
28 
29 /* Kprobe early definition from command line */
30 static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
31 
32 static int __init set_kprobe_boot_events(char *str)
33 {
34 	strscpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
35 	disable_tracing_selftest("running kprobe events");
36 
37 	return 1;
38 }
39 __setup("kprobe_event=", set_kprobe_boot_events);
40 
41 static int trace_kprobe_create(const char *raw_command);
42 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
43 static int trace_kprobe_release(struct dyn_event *ev);
44 static bool trace_kprobe_is_busy(struct dyn_event *ev);
45 static bool trace_kprobe_match(const char *system, const char *event,
46 			int argc, const char **argv, struct dyn_event *ev);
47 
48 static struct dyn_event_operations trace_kprobe_ops = {
49 	.create = trace_kprobe_create,
50 	.show = trace_kprobe_show,
51 	.is_busy = trace_kprobe_is_busy,
52 	.free = trace_kprobe_release,
53 	.match = trace_kprobe_match,
54 };
55 
56 /*
57  * Kprobe event core functions
58  */
59 struct trace_kprobe {
60 	struct dyn_event	devent;
61 	struct kretprobe	rp;	/* Use rp.kp for kprobe use */
62 	unsigned long __percpu *nhit;
63 	const char		*symbol;	/* symbol name */
64 	struct trace_probe	tp;
65 };
66 
67 static bool is_trace_kprobe(struct dyn_event *ev)
68 {
69 	return ev->ops == &trace_kprobe_ops;
70 }
71 
72 static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
73 {
74 	return container_of(ev, struct trace_kprobe, devent);
75 }
76 
77 /**
78  * for_each_trace_kprobe - iterate over the trace_kprobe list
79  * @pos:	the struct trace_kprobe * for each entry
80  * @dpos:	the struct dyn_event * to use as a loop cursor
81  */
82 #define for_each_trace_kprobe(pos, dpos)	\
83 	for_each_dyn_event(dpos)		\
84 		if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
85 #define trace_kprobe_list_empty() list_empty(&dyn_event_list)
86 
87 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
88 {
89 	return tk->rp.handler != NULL;
90 }
91 
92 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
93 {
94 	return tk->symbol ? tk->symbol : "unknown";
95 }
96 
97 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
98 {
99 	return tk->rp.kp.offset;
100 }
101 
102 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
103 {
104 	return kprobe_gone(&tk->rp.kp);
105 }
106 
107 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
108 						 struct module *mod)
109 {
110 	int len = strlen(module_name(mod));
111 	const char *name = trace_kprobe_symbol(tk);
112 
113 	return strncmp(module_name(mod), name, len) == 0 && name[len] == ':';
114 }
115 
116 #ifdef CONFIG_MODULES
117 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
118 {
119 	char *p;
120 	bool ret;
121 
122 	if (!tk->symbol)
123 		return false;
124 	p = strchr(tk->symbol, ':');
125 	if (!p)
126 		return true;
127 	*p = '\0';
128 	scoped_guard(rcu)
129 		ret = !!find_module(tk->symbol);
130 	*p = ':';
131 
132 	return ret;
133 }
134 #else
135 static inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
136 {
137 	return false;
138 }
139 #endif
140 
141 static bool trace_kprobe_is_busy(struct dyn_event *ev)
142 {
143 	struct trace_kprobe *tk = to_trace_kprobe(ev);
144 
145 	return trace_probe_is_enabled(&tk->tp);
146 }
147 
148 static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
149 					    int argc, const char **argv)
150 {
151 	char buf[MAX_ARGSTR_LEN + 1];
152 
153 	if (!argc)
154 		return true;
155 
156 	if (!tk->symbol)
157 		snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
158 	else if (tk->rp.kp.offset)
159 		snprintf(buf, sizeof(buf), "%s+%u",
160 			 trace_kprobe_symbol(tk), tk->rp.kp.offset);
161 	else
162 		snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
163 	if (strcmp(buf, argv[0]))
164 		return false;
165 	argc--; argv++;
166 
167 	return trace_probe_match_command_args(&tk->tp, argc, argv);
168 }
169 
170 static bool trace_kprobe_match(const char *system, const char *event,
171 			int argc, const char **argv, struct dyn_event *ev)
172 {
173 	struct trace_kprobe *tk = to_trace_kprobe(ev);
174 
175 	return (event[0] == '\0' ||
176 		strcmp(trace_probe_name(&tk->tp), event) == 0) &&
177 	    (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
178 	    trace_kprobe_match_command_head(tk, argc, argv);
179 }
180 
181 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
182 {
183 	unsigned long nhit = 0;
184 	int cpu;
185 
186 	for_each_possible_cpu(cpu)
187 		nhit += *per_cpu_ptr(tk->nhit, cpu);
188 
189 	return nhit;
190 }
191 
192 static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
193 {
194 	return !(list_empty(&tk->rp.kp.list) &&
195 		 hlist_unhashed(&tk->rp.kp.hlist));
196 }
197 
198 /* Return 0 if it fails to find the symbol address */
199 static nokprobe_inline
200 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
201 {
202 	unsigned long addr;
203 
204 	if (tk->symbol) {
205 		addr = (unsigned long)
206 			kallsyms_lookup_name(trace_kprobe_symbol(tk));
207 		if (addr)
208 			addr += tk->rp.kp.offset;
209 	} else {
210 		addr = (unsigned long)tk->rp.kp.addr;
211 	}
212 	return addr;
213 }
214 
215 static nokprobe_inline struct trace_kprobe *
216 trace_kprobe_primary_from_call(struct trace_event_call *call)
217 {
218 	struct trace_probe *tp;
219 
220 	tp = trace_probe_primary_from_call(call);
221 	if (WARN_ON_ONCE(!tp))
222 		return NULL;
223 
224 	return container_of(tp, struct trace_kprobe, tp);
225 }
226 
227 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
228 {
229 	struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
230 
231 	return tk ? (kprobe_on_func_entry(tk->rp.kp.addr,
232 			tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
233 			tk->rp.kp.addr ? 0 : tk->rp.kp.offset) == 0) : false;
234 }
235 
236 bool trace_kprobe_error_injectable(struct trace_event_call *call)
237 {
238 	struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
239 
240 	return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
241 	       false;
242 }
243 
244 static int register_kprobe_event(struct trace_kprobe *tk);
245 static int unregister_kprobe_event(struct trace_kprobe *tk);
246 
247 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
248 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
249 				struct pt_regs *regs);
250 
251 static void free_trace_kprobe(struct trace_kprobe *tk)
252 {
253 	if (tk) {
254 		trace_probe_cleanup(&tk->tp);
255 		kfree(tk->symbol);
256 		free_percpu(tk->nhit);
257 		kfree(tk);
258 	}
259 }
260 
261 DEFINE_FREE(free_trace_kprobe, struct trace_kprobe *,
262 	if (!IS_ERR_OR_NULL(_T)) free_trace_kprobe(_T))
263 
264 /*
265  * Allocate new trace_probe and initialize it (including kprobes).
266  */
267 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
268 					     const char *event,
269 					     void *addr,
270 					     const char *symbol,
271 					     unsigned long offs,
272 					     int maxactive,
273 					     int nargs, bool is_return)
274 {
275 	struct trace_kprobe *tk __free(free_trace_kprobe) = NULL;
276 	int ret = -ENOMEM;
277 
278 	tk = kzalloc(struct_size(tk, tp.args, nargs), GFP_KERNEL);
279 	if (!tk)
280 		return ERR_PTR(ret);
281 
282 	tk->nhit = alloc_percpu(unsigned long);
283 	if (!tk->nhit)
284 		return ERR_PTR(ret);
285 
286 	if (symbol) {
287 		tk->symbol = kstrdup(symbol, GFP_KERNEL);
288 		if (!tk->symbol)
289 			return ERR_PTR(ret);
290 		tk->rp.kp.symbol_name = tk->symbol;
291 		tk->rp.kp.offset = offs;
292 	} else
293 		tk->rp.kp.addr = addr;
294 
295 	if (is_return)
296 		tk->rp.handler = kretprobe_dispatcher;
297 	else
298 		tk->rp.kp.pre_handler = kprobe_dispatcher;
299 
300 	tk->rp.maxactive = maxactive;
301 	INIT_HLIST_NODE(&tk->rp.kp.hlist);
302 	INIT_LIST_HEAD(&tk->rp.kp.list);
303 
304 	ret = trace_probe_init(&tk->tp, event, group, false, nargs);
305 	if (ret < 0)
306 		return ERR_PTR(ret);
307 
308 	dyn_event_init(&tk->devent, &trace_kprobe_ops);
309 	return_ptr(tk);
310 }
311 
312 static struct trace_kprobe *find_trace_kprobe(const char *event,
313 					      const char *group)
314 {
315 	struct dyn_event *pos;
316 	struct trace_kprobe *tk;
317 
318 	for_each_trace_kprobe(tk, pos)
319 		if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
320 		    strcmp(trace_probe_group_name(&tk->tp), group) == 0)
321 			return tk;
322 	return NULL;
323 }
324 
325 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
326 {
327 	int ret = 0;
328 
329 	if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
330 		if (trace_kprobe_is_return(tk))
331 			ret = enable_kretprobe(&tk->rp);
332 		else
333 			ret = enable_kprobe(&tk->rp.kp);
334 	}
335 
336 	return ret;
337 }
338 
339 static void __disable_trace_kprobe(struct trace_probe *tp)
340 {
341 	struct trace_kprobe *tk;
342 
343 	list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
344 		if (!trace_kprobe_is_registered(tk))
345 			continue;
346 		if (trace_kprobe_is_return(tk))
347 			disable_kretprobe(&tk->rp);
348 		else
349 			disable_kprobe(&tk->rp.kp);
350 	}
351 }
352 
353 /*
354  * Enable trace_probe
355  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
356  */
357 static int enable_trace_kprobe(struct trace_event_call *call,
358 				struct trace_event_file *file)
359 {
360 	struct trace_probe *tp;
361 	struct trace_kprobe *tk;
362 	bool enabled;
363 	int ret = 0;
364 
365 	tp = trace_probe_primary_from_call(call);
366 	if (WARN_ON_ONCE(!tp))
367 		return -ENODEV;
368 	enabled = trace_probe_is_enabled(tp);
369 
370 	/* This also changes "enabled" state */
371 	if (file) {
372 		ret = trace_probe_add_file(tp, file);
373 		if (ret)
374 			return ret;
375 	} else
376 		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
377 
378 	if (enabled)
379 		return 0;
380 
381 	list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
382 		if (trace_kprobe_has_gone(tk))
383 			continue;
384 		ret = __enable_trace_kprobe(tk);
385 		if (ret)
386 			break;
387 		enabled = true;
388 	}
389 
390 	if (ret) {
391 		/* Failed to enable one of them. Roll back all */
392 		if (enabled)
393 			__disable_trace_kprobe(tp);
394 		if (file)
395 			trace_probe_remove_file(tp, file);
396 		else
397 			trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
398 	}
399 
400 	return ret;
401 }
402 
403 /*
404  * Disable trace_probe
405  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
406  */
407 static int disable_trace_kprobe(struct trace_event_call *call,
408 				struct trace_event_file *file)
409 {
410 	struct trace_probe *tp;
411 
412 	tp = trace_probe_primary_from_call(call);
413 	if (WARN_ON_ONCE(!tp))
414 		return -ENODEV;
415 
416 	if (file) {
417 		if (!trace_probe_get_file_link(tp, file))
418 			return -ENOENT;
419 		if (!trace_probe_has_single_file(tp))
420 			goto out;
421 		trace_probe_clear_flag(tp, TP_FLAG_TRACE);
422 	} else
423 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
424 
425 	if (!trace_probe_is_enabled(tp))
426 		__disable_trace_kprobe(tp);
427 
428  out:
429 	if (file)
430 		/*
431 		 * Synchronization is done in below function. For perf event,
432 		 * file == NULL and perf_trace_event_unreg() calls
433 		 * tracepoint_synchronize_unregister() to ensure synchronize
434 		 * event. We don't need to care about it.
435 		 */
436 		trace_probe_remove_file(tp, file);
437 
438 	return 0;
439 }
440 
441 #if defined(CONFIG_DYNAMIC_FTRACE) && \
442 	!defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
443 static bool __within_notrace_func(unsigned long addr)
444 {
445 	unsigned long offset, size;
446 
447 	if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
448 		return false;
449 
450 	/* Get the entry address of the target function */
451 	addr -= offset;
452 
453 	/*
454 	 * Since ftrace_location_range() does inclusive range check, we need
455 	 * to subtract 1 byte from the end address.
456 	 */
457 	return !ftrace_location_range(addr, addr + size - 1);
458 }
459 
460 static bool within_notrace_func(struct trace_kprobe *tk)
461 {
462 	unsigned long addr = trace_kprobe_address(tk);
463 	char symname[KSYM_NAME_LEN], *p;
464 
465 	if (!__within_notrace_func(addr))
466 		return false;
467 
468 	/* Check if the address is on a suffixed-symbol */
469 	if (!lookup_symbol_name(addr, symname)) {
470 		p = strchr(symname, '.');
471 		if (!p)
472 			return true;
473 		*p = '\0';
474 		addr = (unsigned long)kprobe_lookup_name(symname, 0);
475 		if (addr)
476 			return __within_notrace_func(addr);
477 	}
478 
479 	return true;
480 }
481 #else
482 #define within_notrace_func(tk)	(false)
483 #endif
484 
485 /* Internal register function - just handle k*probes and flags */
486 static int __register_trace_kprobe(struct trace_kprobe *tk)
487 {
488 	int i, ret;
489 
490 	ret = security_locked_down(LOCKDOWN_KPROBES);
491 	if (ret)
492 		return ret;
493 
494 	if (trace_kprobe_is_registered(tk))
495 		return -EINVAL;
496 
497 	if (within_notrace_func(tk)) {
498 		pr_warn("Could not probe notrace function %ps\n",
499 			(void *)trace_kprobe_address(tk));
500 		return -EINVAL;
501 	}
502 
503 	for (i = 0; i < tk->tp.nr_args; i++) {
504 		ret = traceprobe_update_arg(&tk->tp.args[i]);
505 		if (ret)
506 			return ret;
507 	}
508 
509 	/* Set/clear disabled flag according to tp->flag */
510 	if (trace_probe_is_enabled(&tk->tp))
511 		tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
512 	else
513 		tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
514 
515 	if (trace_kprobe_is_return(tk))
516 		ret = register_kretprobe(&tk->rp);
517 	else
518 		ret = register_kprobe(&tk->rp.kp);
519 
520 	return ret;
521 }
522 
523 /* Internal unregister function - just handle k*probes and flags */
524 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
525 {
526 	if (trace_kprobe_is_registered(tk)) {
527 		if (trace_kprobe_is_return(tk))
528 			unregister_kretprobe(&tk->rp);
529 		else
530 			unregister_kprobe(&tk->rp.kp);
531 		/* Cleanup kprobe for reuse and mark it unregistered */
532 		INIT_HLIST_NODE(&tk->rp.kp.hlist);
533 		INIT_LIST_HEAD(&tk->rp.kp.list);
534 		if (tk->rp.kp.symbol_name)
535 			tk->rp.kp.addr = NULL;
536 	}
537 }
538 
539 /* Unregister a trace_probe and probe_event */
540 static int unregister_trace_kprobe(struct trace_kprobe *tk)
541 {
542 	/* If other probes are on the event, just unregister kprobe */
543 	if (trace_probe_has_sibling(&tk->tp))
544 		goto unreg;
545 
546 	/* Enabled event can not be unregistered */
547 	if (trace_probe_is_enabled(&tk->tp))
548 		return -EBUSY;
549 
550 	/* If there's a reference to the dynamic event */
551 	if (trace_event_dyn_busy(trace_probe_event_call(&tk->tp)))
552 		return -EBUSY;
553 
554 	/* Will fail if probe is being used by ftrace or perf */
555 	if (unregister_kprobe_event(tk))
556 		return -EBUSY;
557 
558 unreg:
559 	__unregister_trace_kprobe(tk);
560 	dyn_event_remove(&tk->devent);
561 	trace_probe_unlink(&tk->tp);
562 
563 	return 0;
564 }
565 
566 static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
567 					 struct trace_kprobe *comp)
568 {
569 	struct trace_probe_event *tpe = orig->tp.event;
570 	int i;
571 
572 	list_for_each_entry(orig, &tpe->probes, tp.list) {
573 		if (strcmp(trace_kprobe_symbol(orig),
574 			   trace_kprobe_symbol(comp)) ||
575 		    trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
576 			continue;
577 
578 		/*
579 		 * trace_probe_compare_arg_type() ensured that nr_args and
580 		 * each argument name and type are same. Let's compare comm.
581 		 */
582 		for (i = 0; i < orig->tp.nr_args; i++) {
583 			if (strcmp(orig->tp.args[i].comm,
584 				   comp->tp.args[i].comm))
585 				break;
586 		}
587 
588 		if (i == orig->tp.nr_args)
589 			return true;
590 	}
591 
592 	return false;
593 }
594 
595 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
596 {
597 	int ret;
598 
599 	ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
600 	if (ret) {
601 		/* Note that argument starts index = 2 */
602 		trace_probe_log_set_index(ret + 1);
603 		trace_probe_log_err(0, DIFF_ARG_TYPE);
604 		return -EEXIST;
605 	}
606 	if (trace_kprobe_has_same_kprobe(to, tk)) {
607 		trace_probe_log_set_index(0);
608 		trace_probe_log_err(0, SAME_PROBE);
609 		return -EEXIST;
610 	}
611 
612 	/* Append to existing event */
613 	ret = trace_probe_append(&tk->tp, &to->tp);
614 	if (ret)
615 		return ret;
616 
617 	/* Register k*probe */
618 	ret = __register_trace_kprobe(tk);
619 	if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
620 		pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
621 		ret = 0;
622 	}
623 
624 	if (ret)
625 		trace_probe_unlink(&tk->tp);
626 	else
627 		dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
628 
629 	return ret;
630 }
631 
632 /* Register a trace_probe and probe_event */
633 static int register_trace_kprobe(struct trace_kprobe *tk)
634 {
635 	struct trace_kprobe *old_tk;
636 	int ret;
637 
638 	guard(mutex)(&event_mutex);
639 
640 	old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
641 				   trace_probe_group_name(&tk->tp));
642 	if (old_tk) {
643 		if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
644 			trace_probe_log_set_index(0);
645 			trace_probe_log_err(0, DIFF_PROBE_TYPE);
646 			return -EEXIST;
647 		}
648 		return append_trace_kprobe(tk, old_tk);
649 	}
650 
651 	/* Register new event */
652 	ret = register_kprobe_event(tk);
653 	if (ret) {
654 		if (ret == -EEXIST) {
655 			trace_probe_log_set_index(0);
656 			trace_probe_log_err(0, EVENT_EXIST);
657 		} else
658 			pr_warn("Failed to register probe event(%d)\n", ret);
659 		return ret;
660 	}
661 
662 	/* Register k*probe */
663 	ret = __register_trace_kprobe(tk);
664 	if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
665 		pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
666 		ret = 0;
667 	}
668 
669 	if (ret < 0)
670 		unregister_kprobe_event(tk);
671 	else
672 		dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
673 
674 	return ret;
675 }
676 
677 #ifdef CONFIG_MODULES
678 static int validate_module_probe_symbol(const char *modname, const char *symbol);
679 
680 static int register_module_trace_kprobe(struct module *mod, struct trace_kprobe *tk)
681 {
682 	const char *p;
683 	int ret = 0;
684 
685 	p = strchr(trace_kprobe_symbol(tk), ':');
686 	if (p)
687 		ret = validate_module_probe_symbol(module_name(mod), p + 1);
688 	if (!ret)
689 		ret = __register_trace_kprobe(tk);
690 	return ret;
691 }
692 
693 /* Module notifier call back, checking event on the module */
694 static int trace_kprobe_module_callback(struct notifier_block *nb,
695 				       unsigned long val, void *data)
696 {
697 	struct module *mod = data;
698 	struct dyn_event *pos;
699 	struct trace_kprobe *tk;
700 	int ret;
701 
702 	if (val != MODULE_STATE_COMING)
703 		return NOTIFY_DONE;
704 
705 	/* Update probes on coming module */
706 	guard(mutex)(&event_mutex);
707 	for_each_trace_kprobe(tk, pos) {
708 		if (trace_kprobe_within_module(tk, mod)) {
709 			/* Don't need to check busy - this should have gone. */
710 			__unregister_trace_kprobe(tk);
711 			ret = register_module_trace_kprobe(mod, tk);
712 			if (ret)
713 				pr_warn("Failed to re-register probe %s on %s: %d\n",
714 					trace_probe_name(&tk->tp),
715 					module_name(mod), ret);
716 		}
717 	}
718 
719 	return NOTIFY_DONE;
720 }
721 
722 static struct notifier_block trace_kprobe_module_nb = {
723 	.notifier_call = trace_kprobe_module_callback,
724 	.priority = 2	/* Invoked after kprobe and jump_label module callback */
725 };
726 static int trace_kprobe_register_module_notifier(void)
727 {
728 	return register_module_notifier(&trace_kprobe_module_nb);
729 }
730 #else
731 static int trace_kprobe_register_module_notifier(void)
732 {
733 	return 0;
734 }
735 #endif /* CONFIG_MODULES */
736 
737 static int count_symbols(void *data, unsigned long unused)
738 {
739 	unsigned int *count = data;
740 
741 	(*count)++;
742 
743 	return 0;
744 }
745 
746 struct sym_count_ctx {
747 	unsigned int count;
748 	const char *name;
749 };
750 
751 static int count_mod_symbols(void *data, const char *name, unsigned long unused)
752 {
753 	struct sym_count_ctx *ctx = data;
754 
755 	if (strcmp(name, ctx->name) == 0)
756 		ctx->count++;
757 
758 	return 0;
759 }
760 
761 static unsigned int number_of_same_symbols(const char *mod, const char *func_name)
762 {
763 	struct sym_count_ctx ctx = { .count = 0, .name = func_name };
764 
765 	if (!mod)
766 		kallsyms_on_each_match_symbol(count_symbols, func_name, &ctx.count);
767 
768 	module_kallsyms_on_each_symbol(mod, count_mod_symbols, &ctx);
769 
770 	return ctx.count;
771 }
772 
773 static int validate_module_probe_symbol(const char *modname, const char *symbol)
774 {
775 	unsigned int count = number_of_same_symbols(modname, symbol);
776 
777 	if (count > 1) {
778 		/*
779 		 * Users should use ADDR to remove the ambiguity of
780 		 * using KSYM only.
781 		 */
782 		return -EADDRNOTAVAIL;
783 	} else if (count == 0) {
784 		/*
785 		 * We can return ENOENT earlier than when register the
786 		 * kprobe.
787 		 */
788 		return -ENOENT;
789 	}
790 	return 0;
791 }
792 
793 #ifdef CONFIG_MODULES
794 /* Return NULL if the module is not loaded or under unloading. */
795 static struct module *try_module_get_by_name(const char *name)
796 {
797 	struct module *mod;
798 
799 	guard(rcu)();
800 	mod = find_module(name);
801 	if (mod && !try_module_get(mod))
802 		mod = NULL;
803 	return mod;
804 }
805 #else
806 #define try_module_get_by_name(name)	(NULL)
807 #endif
808 
809 static int validate_probe_symbol(char *symbol)
810 {
811 	struct module *mod = NULL;
812 	char *modname = NULL, *p;
813 	int ret = 0;
814 
815 	p = strchr(symbol, ':');
816 	if (p) {
817 		modname = symbol;
818 		symbol = p + 1;
819 		*p = '\0';
820 		mod = try_module_get_by_name(modname);
821 		if (!mod)
822 			goto out;
823 	}
824 
825 	ret = validate_module_probe_symbol(modname, symbol);
826 out:
827 	if (p)
828 		*p = ':';
829 	if (mod)
830 		module_put(mod);
831 	return ret;
832 }
833 
834 static int trace_kprobe_entry_handler(struct kretprobe_instance *ri,
835 				      struct pt_regs *regs);
836 
837 static int trace_kprobe_create_internal(int argc, const char *argv[],
838 					struct traceprobe_parse_context *ctx)
839 {
840 	/*
841 	 * Argument syntax:
842 	 *  - Add kprobe:
843 	 *      p[:[GRP/][EVENT]] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
844 	 *  - Add kretprobe:
845 	 *      r[MAXACTIVE][:[GRP/][EVENT]] [MOD:]KSYM[+0] [FETCHARGS]
846 	 *    Or
847 	 *      p[:[GRP/][EVENT]] [MOD:]KSYM[+0]%return [FETCHARGS]
848 	 *
849 	 * Fetch args:
850 	 *  $retval	: fetch return value
851 	 *  $stack	: fetch stack address
852 	 *  $stackN	: fetch Nth of stack (N:0-)
853 	 *  $comm       : fetch current task comm
854 	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
855 	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
856 	 *  %REG	: fetch register REG
857 	 * Dereferencing memory fetch:
858 	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
859 	 * Alias name of args:
860 	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
861 	 * Type of args:
862 	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
863 	 */
864 	struct trace_kprobe *tk __free(free_trace_kprobe) = NULL;
865 	const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
866 	const char **new_argv __free(kfree) = NULL;
867 	int i, len, new_argc = 0, ret = 0;
868 	char *symbol __free(kfree) = NULL;
869 	char *ebuf __free(kfree) = NULL;
870 	char *gbuf __free(kfree) = NULL;
871 	char *abuf __free(kfree) = NULL;
872 	char *dbuf __free(kfree) = NULL;
873 	enum probe_print_type ptype;
874 	bool is_return = false;
875 	int maxactive = 0;
876 	void *addr = NULL;
877 	char *tmp = NULL;
878 	long offset = 0;
879 
880 	switch (argv[0][0]) {
881 	case 'r':
882 		is_return = true;
883 		break;
884 	case 'p':
885 		break;
886 	default:
887 		return -ECANCELED;
888 	}
889 	if (argc < 2)
890 		return -ECANCELED;
891 
892 	event = strchr(&argv[0][1], ':');
893 	if (event)
894 		event++;
895 
896 	if (isdigit(argv[0][1])) {
897 		char *buf __free(kfree) = NULL;
898 
899 		if (!is_return) {
900 			trace_probe_log_err(1, BAD_MAXACT_TYPE);
901 			return -EINVAL;
902 		}
903 		if (event)
904 			len = event - &argv[0][1] - 1;
905 		else
906 			len = strlen(&argv[0][1]);
907 		if (len > MAX_EVENT_NAME_LEN - 1) {
908 			trace_probe_log_err(1, BAD_MAXACT);
909 			return -EINVAL;
910 		}
911 		buf = kmemdup(&argv[0][1], len + 1, GFP_KERNEL);
912 		if (!buf)
913 			return -ENOMEM;
914 		buf[len] = '\0';
915 		ret = kstrtouint(buf, 0, &maxactive);
916 		if (ret || !maxactive) {
917 			trace_probe_log_err(1, BAD_MAXACT);
918 			return -EINVAL;
919 		}
920 		/* kretprobes instances are iterated over via a list. The
921 		 * maximum should stay reasonable.
922 		 */
923 		if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
924 			trace_probe_log_err(1, MAXACT_TOO_BIG);
925 			return -EINVAL;
926 		}
927 	}
928 
929 	/* try to parse an address. if that fails, try to read the
930 	 * input as a symbol. */
931 	if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
932 		trace_probe_log_set_index(1);
933 		/* Check whether uprobe event specified */
934 		if (strchr(argv[1], '/') && strchr(argv[1], ':'))
935 			return -ECANCELED;
936 
937 		/* a symbol specified */
938 		symbol = kstrdup(argv[1], GFP_KERNEL);
939 		if (!symbol)
940 			return -ENOMEM;
941 
942 		tmp = strchr(symbol, '%');
943 		if (tmp) {
944 			if (!strcmp(tmp, "%return")) {
945 				*tmp = '\0';
946 				is_return = true;
947 			} else {
948 				trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
949 				return -EINVAL;
950 			}
951 		}
952 
953 		/* TODO: support .init module functions */
954 		ret = traceprobe_split_symbol_offset(symbol, &offset);
955 		if (ret || offset < 0 || offset > UINT_MAX) {
956 			trace_probe_log_err(0, BAD_PROBE_ADDR);
957 			return -EINVAL;
958 		}
959 		ret = validate_probe_symbol(symbol);
960 		if (ret) {
961 			if (ret == -EADDRNOTAVAIL)
962 				trace_probe_log_err(0, NON_UNIQ_SYMBOL);
963 			else
964 				trace_probe_log_err(0, BAD_PROBE_ADDR);
965 			return -EINVAL;
966 		}
967 		if (is_return)
968 			ctx->flags |= TPARG_FL_RETURN;
969 		ret = kprobe_on_func_entry(NULL, symbol, offset);
970 		if (ret == 0 && !is_return)
971 			ctx->flags |= TPARG_FL_FENTRY;
972 		/* Defer the ENOENT case until register kprobe */
973 		if (ret == -EINVAL && is_return) {
974 			trace_probe_log_err(0, BAD_RETPROBE);
975 			return -EINVAL;
976 		}
977 	}
978 
979 	trace_probe_log_set_index(0);
980 	if (event) {
981 		gbuf = kmalloc(MAX_EVENT_NAME_LEN, GFP_KERNEL);
982 		if (!gbuf)
983 			return -ENOMEM;
984 		ret = traceprobe_parse_event_name(&event, &group, gbuf,
985 						  event - argv[0]);
986 		if (ret)
987 			return ret;
988 	}
989 
990 	if (!event) {
991 		/* Make a new event name */
992 		ebuf = kmalloc(MAX_EVENT_NAME_LEN, GFP_KERNEL);
993 		if (!ebuf)
994 			return -ENOMEM;
995 		if (symbol)
996 			snprintf(ebuf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
997 				 is_return ? 'r' : 'p', symbol, offset);
998 		else
999 			snprintf(ebuf, MAX_EVENT_NAME_LEN, "%c_0x%p",
1000 				 is_return ? 'r' : 'p', addr);
1001 		sanitize_event_name(ebuf);
1002 		event = ebuf;
1003 	}
1004 
1005 	abuf = kmalloc(MAX_BTF_ARGS_LEN, GFP_KERNEL);
1006 	if (!abuf)
1007 		return -ENOMEM;
1008 	argc -= 2; argv += 2;
1009 	ctx->funcname = symbol;
1010 	new_argv = traceprobe_expand_meta_args(argc, argv, &new_argc,
1011 					       abuf, MAX_BTF_ARGS_LEN, ctx);
1012 	if (IS_ERR(new_argv)) {
1013 		ret = PTR_ERR(new_argv);
1014 		new_argv = NULL;
1015 		return ret;
1016 	}
1017 	if (new_argv) {
1018 		argc = new_argc;
1019 		argv = new_argv;
1020 	}
1021 	if (argc > MAX_TRACE_ARGS) {
1022 		trace_probe_log_set_index(2);
1023 		trace_probe_log_err(0, TOO_MANY_ARGS);
1024 		return -E2BIG;
1025 	}
1026 
1027 	ret = traceprobe_expand_dentry_args(argc, argv, &dbuf);
1028 	if (ret)
1029 		return ret;
1030 
1031 	/* setup a probe */
1032 	tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
1033 				argc, is_return);
1034 	if (IS_ERR(tk)) {
1035 		ret = PTR_ERR(tk);
1036 		/* This must return -ENOMEM, else there is a bug */
1037 		WARN_ON_ONCE(ret != -ENOMEM);
1038 		return ret;	/* We know tk is not allocated */
1039 	}
1040 
1041 	/* parse arguments */
1042 	for (i = 0; i < argc; i++) {
1043 		trace_probe_log_set_index(i + 2);
1044 		ctx->offset = 0;
1045 		ret = traceprobe_parse_probe_arg(&tk->tp, i, argv[i], ctx);
1046 		if (ret)
1047 			return ret;	/* This can be -ENOMEM */
1048 	}
1049 	/* entry handler for kretprobe */
1050 	if (is_return && tk->tp.entry_arg) {
1051 		tk->rp.entry_handler = trace_kprobe_entry_handler;
1052 		tk->rp.data_size = traceprobe_get_entry_data_size(&tk->tp);
1053 	}
1054 
1055 	ptype = is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1056 	ret = traceprobe_set_print_fmt(&tk->tp, ptype);
1057 	if (ret < 0)
1058 		return ret;
1059 
1060 	ret = register_trace_kprobe(tk);
1061 	if (ret) {
1062 		trace_probe_log_set_index(1);
1063 		if (ret == -EILSEQ)
1064 			trace_probe_log_err(0, BAD_INSN_BNDRY);
1065 		else if (ret == -ENOENT)
1066 			trace_probe_log_err(0, BAD_PROBE_ADDR);
1067 		else if (ret != -ENOMEM && ret != -EEXIST)
1068 			trace_probe_log_err(0, FAIL_REG_PROBE);
1069 		return ret;
1070 	}
1071 	/*
1072 	 * Here, 'tk' has been registered to the list successfully,
1073 	 * so we don't need to free it.
1074 	 */
1075 	tk = NULL;
1076 
1077 	return 0;
1078 }
1079 
1080 static int trace_kprobe_create_cb(int argc, const char *argv[])
1081 {
1082 	struct traceprobe_parse_context *ctx __free(traceprobe_parse_context) = NULL;
1083 	int ret;
1084 
1085 	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1086 	if (!ctx)
1087 		return -ENOMEM;
1088 	ctx->flags = TPARG_FL_KERNEL;
1089 
1090 	trace_probe_log_init("trace_kprobe", argc, argv);
1091 
1092 	ret = trace_kprobe_create_internal(argc, argv, ctx);
1093 
1094 	trace_probe_log_clear();
1095 	return ret;
1096 }
1097 
1098 static int trace_kprobe_create(const char *raw_command)
1099 {
1100 	return trace_probe_create(raw_command, trace_kprobe_create_cb);
1101 }
1102 
1103 static int create_or_delete_trace_kprobe(const char *raw_command)
1104 {
1105 	int ret;
1106 
1107 	if (raw_command[0] == '-')
1108 		return dyn_event_release(raw_command, &trace_kprobe_ops);
1109 
1110 	ret = dyn_event_create(raw_command, &trace_kprobe_ops);
1111 	return ret == -ECANCELED ? -EINVAL : ret;
1112 }
1113 
1114 static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
1115 {
1116 	return create_or_delete_trace_kprobe(cmd->seq.buffer);
1117 }
1118 
1119 /**
1120  * kprobe_event_cmd_init - Initialize a kprobe event command object
1121  * @cmd: A pointer to the dynevent_cmd struct representing the new event
1122  * @buf: A pointer to the buffer used to build the command
1123  * @maxlen: The length of the buffer passed in @buf
1124  *
1125  * Initialize a synthetic event command object.  Use this before
1126  * calling any of the other kprobe_event functions.
1127  */
1128 void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
1129 {
1130 	dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
1131 			  trace_kprobe_run_command);
1132 }
1133 EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
1134 
1135 /**
1136  * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
1137  * @cmd: A pointer to the dynevent_cmd struct representing the new event
1138  * @kretprobe: Is this a return probe?
1139  * @name: The name of the kprobe event
1140  * @loc: The location of the kprobe event
1141  * @...: Variable number of arg (pairs), one pair for each field
1142  *
1143  * NOTE: Users normally won't want to call this function directly, but
1144  * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
1145  * adds a NULL to the end of the arg list.  If this function is used
1146  * directly, make sure the last arg in the variable arg list is NULL.
1147  *
1148  * Generate a kprobe event command to be executed by
1149  * kprobe_event_gen_cmd_end().  This function can be used to generate the
1150  * complete command or only the first part of it; in the latter case,
1151  * kprobe_event_add_fields() can be used to add more fields following this.
1152  *
1153  * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
1154  * returns -EINVAL if @loc == NULL.
1155  *
1156  * Return: 0 if successful, error otherwise.
1157  */
1158 int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
1159 				 const char *name, const char *loc, ...)
1160 {
1161 	char buf[MAX_EVENT_NAME_LEN];
1162 	struct dynevent_arg arg;
1163 	va_list args;
1164 	int ret;
1165 
1166 	if (cmd->type != DYNEVENT_TYPE_KPROBE)
1167 		return -EINVAL;
1168 
1169 	if (!loc)
1170 		return -EINVAL;
1171 
1172 	if (kretprobe)
1173 		snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
1174 	else
1175 		snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
1176 
1177 	ret = dynevent_str_add(cmd, buf);
1178 	if (ret)
1179 		return ret;
1180 
1181 	dynevent_arg_init(&arg, 0);
1182 	arg.str = loc;
1183 	ret = dynevent_arg_add(cmd, &arg, NULL);
1184 	if (ret)
1185 		return ret;
1186 
1187 	va_start(args, loc);
1188 	for (;;) {
1189 		const char *field;
1190 
1191 		field = va_arg(args, const char *);
1192 		if (!field)
1193 			break;
1194 
1195 		if (++cmd->n_fields > MAX_TRACE_ARGS) {
1196 			ret = -EINVAL;
1197 			break;
1198 		}
1199 
1200 		arg.str = field;
1201 		ret = dynevent_arg_add(cmd, &arg, NULL);
1202 		if (ret)
1203 			break;
1204 	}
1205 	va_end(args);
1206 
1207 	return ret;
1208 }
1209 EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1210 
1211 /**
1212  * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1213  * @cmd: A pointer to the dynevent_cmd struct representing the new event
1214  * @...: Variable number of arg (pairs), one pair for each field
1215  *
1216  * NOTE: Users normally won't want to call this function directly, but
1217  * rather use the kprobe_event_add_fields() wrapper, which
1218  * automatically adds a NULL to the end of the arg list.  If this
1219  * function is used directly, make sure the last arg in the variable
1220  * arg list is NULL.
1221  *
1222  * Add probe fields to an existing kprobe command using a variable
1223  * list of args.  Fields are added in the same order they're listed.
1224  *
1225  * Return: 0 if successful, error otherwise.
1226  */
1227 int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1228 {
1229 	struct dynevent_arg arg;
1230 	va_list args;
1231 	int ret = 0;
1232 
1233 	if (cmd->type != DYNEVENT_TYPE_KPROBE)
1234 		return -EINVAL;
1235 
1236 	dynevent_arg_init(&arg, 0);
1237 
1238 	va_start(args, cmd);
1239 	for (;;) {
1240 		const char *field;
1241 
1242 		field = va_arg(args, const char *);
1243 		if (!field)
1244 			break;
1245 
1246 		if (++cmd->n_fields > MAX_TRACE_ARGS) {
1247 			ret = -EINVAL;
1248 			break;
1249 		}
1250 
1251 		arg.str = field;
1252 		ret = dynevent_arg_add(cmd, &arg, NULL);
1253 		if (ret)
1254 			break;
1255 	}
1256 	va_end(args);
1257 
1258 	return ret;
1259 }
1260 EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1261 
1262 /**
1263  * kprobe_event_delete - Delete a kprobe event
1264  * @name: The name of the kprobe event to delete
1265  *
1266  * Delete a kprobe event with the give @name from kernel code rather
1267  * than directly from the command line.
1268  *
1269  * Return: 0 if successful, error otherwise.
1270  */
1271 int kprobe_event_delete(const char *name)
1272 {
1273 	char buf[MAX_EVENT_NAME_LEN];
1274 
1275 	snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1276 
1277 	return create_or_delete_trace_kprobe(buf);
1278 }
1279 EXPORT_SYMBOL_GPL(kprobe_event_delete);
1280 
1281 static int trace_kprobe_release(struct dyn_event *ev)
1282 {
1283 	struct trace_kprobe *tk = to_trace_kprobe(ev);
1284 	int ret = unregister_trace_kprobe(tk);
1285 
1286 	if (!ret)
1287 		free_trace_kprobe(tk);
1288 	return ret;
1289 }
1290 
1291 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1292 {
1293 	struct trace_kprobe *tk = to_trace_kprobe(ev);
1294 	int i;
1295 
1296 	seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1297 	if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1298 		seq_printf(m, "%d", tk->rp.maxactive);
1299 	seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1300 				trace_probe_name(&tk->tp));
1301 
1302 	if (!tk->symbol)
1303 		seq_printf(m, " 0x%p", tk->rp.kp.addr);
1304 	else if (tk->rp.kp.offset)
1305 		seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1306 			   tk->rp.kp.offset);
1307 	else
1308 		seq_printf(m, " %s", trace_kprobe_symbol(tk));
1309 
1310 	for (i = 0; i < tk->tp.nr_args; i++)
1311 		seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1312 	seq_putc(m, '\n');
1313 
1314 	return 0;
1315 }
1316 
1317 static int probes_seq_show(struct seq_file *m, void *v)
1318 {
1319 	struct dyn_event *ev = v;
1320 
1321 	if (!is_trace_kprobe(ev))
1322 		return 0;
1323 
1324 	return trace_kprobe_show(m, ev);
1325 }
1326 
1327 static const struct seq_operations probes_seq_op = {
1328 	.start  = dyn_event_seq_start,
1329 	.next   = dyn_event_seq_next,
1330 	.stop   = dyn_event_seq_stop,
1331 	.show   = probes_seq_show
1332 };
1333 
1334 static int probes_open(struct inode *inode, struct file *file)
1335 {
1336 	int ret;
1337 
1338 	ret = security_locked_down(LOCKDOWN_TRACEFS);
1339 	if (ret)
1340 		return ret;
1341 
1342 	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1343 		ret = dyn_events_release_all(&trace_kprobe_ops);
1344 		if (ret < 0)
1345 			return ret;
1346 	}
1347 
1348 	return seq_open(file, &probes_seq_op);
1349 }
1350 
1351 static ssize_t probes_write(struct file *file, const char __user *buffer,
1352 			    size_t count, loff_t *ppos)
1353 {
1354 	return trace_parse_run_command(file, buffer, count, ppos,
1355 				       create_or_delete_trace_kprobe);
1356 }
1357 
1358 static const struct file_operations kprobe_events_ops = {
1359 	.owner          = THIS_MODULE,
1360 	.open           = probes_open,
1361 	.read           = seq_read,
1362 	.llseek         = seq_lseek,
1363 	.release        = seq_release,
1364 	.write		= probes_write,
1365 };
1366 
1367 static unsigned long trace_kprobe_missed(struct trace_kprobe *tk)
1368 {
1369 	return trace_kprobe_is_return(tk) ?
1370 		tk->rp.kp.nmissed + tk->rp.nmissed : tk->rp.kp.nmissed;
1371 }
1372 
1373 /* Probes profiling interfaces */
1374 static int probes_profile_seq_show(struct seq_file *m, void *v)
1375 {
1376 	struct dyn_event *ev = v;
1377 	struct trace_kprobe *tk;
1378 	unsigned long nmissed;
1379 
1380 	if (!is_trace_kprobe(ev))
1381 		return 0;
1382 
1383 	tk = to_trace_kprobe(ev);
1384 	nmissed = trace_kprobe_missed(tk);
1385 	seq_printf(m, "  %-44s %15lu %15lu\n",
1386 		   trace_probe_name(&tk->tp),
1387 		   trace_kprobe_nhit(tk),
1388 		   nmissed);
1389 
1390 	return 0;
1391 }
1392 
1393 static const struct seq_operations profile_seq_op = {
1394 	.start  = dyn_event_seq_start,
1395 	.next   = dyn_event_seq_next,
1396 	.stop   = dyn_event_seq_stop,
1397 	.show   = probes_profile_seq_show
1398 };
1399 
1400 static int profile_open(struct inode *inode, struct file *file)
1401 {
1402 	int ret;
1403 
1404 	ret = security_locked_down(LOCKDOWN_TRACEFS);
1405 	if (ret)
1406 		return ret;
1407 
1408 	return seq_open(file, &profile_seq_op);
1409 }
1410 
1411 static const struct file_operations kprobe_profile_ops = {
1412 	.owner          = THIS_MODULE,
1413 	.open           = profile_open,
1414 	.read           = seq_read,
1415 	.llseek         = seq_lseek,
1416 	.release        = seq_release,
1417 };
1418 
1419 /* Note that we don't verify it, since the code does not come from user space */
1420 static int
1421 process_fetch_insn(struct fetch_insn *code, void *rec, void *edata,
1422 		   void *dest, void *base)
1423 {
1424 	struct pt_regs *regs = rec;
1425 	unsigned long val;
1426 	int ret;
1427 
1428 retry:
1429 	/* 1st stage: get value from context */
1430 	switch (code->op) {
1431 	case FETCH_OP_REG:
1432 		val = regs_get_register(regs, code->param);
1433 		break;
1434 	case FETCH_OP_STACK:
1435 		val = regs_get_kernel_stack_nth(regs, code->param);
1436 		break;
1437 	case FETCH_OP_STACKP:
1438 		val = kernel_stack_pointer(regs);
1439 		break;
1440 	case FETCH_OP_RETVAL:
1441 		val = regs_return_value(regs);
1442 		break;
1443 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1444 	case FETCH_OP_ARG:
1445 		val = regs_get_kernel_argument(regs, code->param);
1446 		break;
1447 	case FETCH_OP_EDATA:
1448 		val = *(unsigned long *)((unsigned long)edata + code->offset);
1449 		break;
1450 #endif
1451 	case FETCH_NOP_SYMBOL:	/* Ignore a place holder */
1452 		code++;
1453 		goto retry;
1454 	default:
1455 		ret = process_common_fetch_insn(code, &val);
1456 		if (ret < 0)
1457 			return ret;
1458 	}
1459 	code++;
1460 
1461 	return process_fetch_insn_bottom(code, val, dest, base);
1462 }
1463 NOKPROBE_SYMBOL(process_fetch_insn)
1464 
1465 /* Kprobe handler */
1466 static nokprobe_inline void
1467 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1468 		    struct trace_event_file *trace_file)
1469 {
1470 	struct kprobe_trace_entry_head *entry;
1471 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1472 	struct trace_event_buffer fbuffer;
1473 	int dsize;
1474 
1475 	WARN_ON(call != trace_file->event_call);
1476 
1477 	if (trace_trigger_soft_disabled(trace_file))
1478 		return;
1479 
1480 	dsize = __get_data_size(&tk->tp, regs, NULL);
1481 
1482 	entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1483 					   sizeof(*entry) + tk->tp.size + dsize);
1484 	if (!entry)
1485 		return;
1486 
1487 	fbuffer.regs = regs;
1488 	entry->ip = (unsigned long)tk->rp.kp.addr;
1489 	store_trace_args(&entry[1], &tk->tp, regs, NULL, sizeof(*entry), dsize);
1490 
1491 	trace_event_buffer_commit(&fbuffer);
1492 }
1493 
1494 static void
1495 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1496 {
1497 	struct event_file_link *link;
1498 
1499 	trace_probe_for_each_link_rcu(link, &tk->tp)
1500 		__kprobe_trace_func(tk, regs, link->file);
1501 }
1502 NOKPROBE_SYMBOL(kprobe_trace_func);
1503 
1504 /* Kretprobe handler */
1505 
1506 static int trace_kprobe_entry_handler(struct kretprobe_instance *ri,
1507 				      struct pt_regs *regs)
1508 {
1509 	struct kretprobe *rp = get_kretprobe(ri);
1510 	struct trace_kprobe *tk;
1511 
1512 	/*
1513 	 * There is a small chance that get_kretprobe(ri) returns NULL when
1514 	 * the kretprobe is unregister on another CPU between kretprobe's
1515 	 * trampoline_handler and this function.
1516 	 */
1517 	if (unlikely(!rp))
1518 		return -ENOENT;
1519 
1520 	tk = container_of(rp, struct trace_kprobe, rp);
1521 
1522 	/* store argument values into ri->data as entry data */
1523 	if (tk->tp.entry_arg)
1524 		store_trace_entry_data(ri->data, &tk->tp, regs);
1525 
1526 	return 0;
1527 }
1528 
1529 
1530 static nokprobe_inline void
1531 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1532 		       struct pt_regs *regs,
1533 		       struct trace_event_file *trace_file)
1534 {
1535 	struct kretprobe_trace_entry_head *entry;
1536 	struct trace_event_buffer fbuffer;
1537 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1538 	int dsize;
1539 
1540 	WARN_ON(call != trace_file->event_call);
1541 
1542 	if (trace_trigger_soft_disabled(trace_file))
1543 		return;
1544 
1545 	dsize = __get_data_size(&tk->tp, regs, ri->data);
1546 
1547 	entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1548 					   sizeof(*entry) + tk->tp.size + dsize);
1549 	if (!entry)
1550 		return;
1551 
1552 	fbuffer.regs = regs;
1553 	entry->func = (unsigned long)tk->rp.kp.addr;
1554 	entry->ret_ip = get_kretprobe_retaddr(ri);
1555 	store_trace_args(&entry[1], &tk->tp, regs, ri->data, sizeof(*entry), dsize);
1556 
1557 	trace_event_buffer_commit(&fbuffer);
1558 }
1559 
1560 static void
1561 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1562 		     struct pt_regs *regs)
1563 {
1564 	struct event_file_link *link;
1565 
1566 	trace_probe_for_each_link_rcu(link, &tk->tp)
1567 		__kretprobe_trace_func(tk, ri, regs, link->file);
1568 }
1569 NOKPROBE_SYMBOL(kretprobe_trace_func);
1570 
1571 /* Event entry printers */
1572 static enum print_line_t
1573 print_kprobe_event(struct trace_iterator *iter, int flags,
1574 		   struct trace_event *event)
1575 {
1576 	struct kprobe_trace_entry_head *field;
1577 	struct trace_seq *s = &iter->seq;
1578 	struct trace_probe *tp;
1579 
1580 	field = (struct kprobe_trace_entry_head *)iter->ent;
1581 	tp = trace_probe_primary_from_call(
1582 		container_of(event, struct trace_event_call, event));
1583 	if (WARN_ON_ONCE(!tp))
1584 		goto out;
1585 
1586 	trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1587 
1588 	if (!seq_print_ip_sym_offset(s, field->ip, flags))
1589 		goto out;
1590 
1591 	trace_seq_putc(s, ')');
1592 
1593 	if (trace_probe_print_args(s, tp->args, tp->nr_args,
1594 			     (u8 *)&field[1], field) < 0)
1595 		goto out;
1596 
1597 	trace_seq_putc(s, '\n');
1598  out:
1599 	return trace_handle_return(s);
1600 }
1601 
1602 static enum print_line_t
1603 print_kretprobe_event(struct trace_iterator *iter, int flags,
1604 		      struct trace_event *event)
1605 {
1606 	struct kretprobe_trace_entry_head *field;
1607 	struct trace_seq *s = &iter->seq;
1608 	struct trace_probe *tp;
1609 
1610 	field = (struct kretprobe_trace_entry_head *)iter->ent;
1611 	tp = trace_probe_primary_from_call(
1612 		container_of(event, struct trace_event_call, event));
1613 	if (WARN_ON_ONCE(!tp))
1614 		goto out;
1615 
1616 	trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1617 
1618 	if (!seq_print_ip_sym_offset(s, field->ret_ip, flags))
1619 		goto out;
1620 
1621 	trace_seq_puts(s, " <- ");
1622 
1623 	if (!seq_print_ip_sym_no_offset(s, field->func, flags))
1624 		goto out;
1625 
1626 	trace_seq_putc(s, ')');
1627 
1628 	if (trace_probe_print_args(s, tp->args, tp->nr_args,
1629 			     (u8 *)&field[1], field) < 0)
1630 		goto out;
1631 
1632 	trace_seq_putc(s, '\n');
1633 
1634  out:
1635 	return trace_handle_return(s);
1636 }
1637 
1638 
1639 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1640 {
1641 	int ret;
1642 	struct kprobe_trace_entry_head field;
1643 	struct trace_probe *tp;
1644 
1645 	tp = trace_probe_primary_from_call(event_call);
1646 	if (WARN_ON_ONCE(!tp))
1647 		return -ENOENT;
1648 
1649 	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1650 
1651 	return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1652 }
1653 
1654 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1655 {
1656 	int ret;
1657 	struct kretprobe_trace_entry_head field;
1658 	struct trace_probe *tp;
1659 
1660 	tp = trace_probe_primary_from_call(event_call);
1661 	if (WARN_ON_ONCE(!tp))
1662 		return -ENOENT;
1663 
1664 	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1665 	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1666 
1667 	return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1668 }
1669 
1670 #ifdef CONFIG_PERF_EVENTS
1671 
1672 /* Kprobe profile handler */
1673 static int
1674 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1675 {
1676 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1677 	struct kprobe_trace_entry_head *entry;
1678 	struct hlist_head *head;
1679 	int size, __size, dsize;
1680 	int rctx;
1681 
1682 	if (bpf_prog_array_valid(call)) {
1683 		unsigned long orig_ip = instruction_pointer(regs);
1684 		int ret;
1685 
1686 		ret = trace_call_bpf(call, regs);
1687 
1688 		/*
1689 		 * We need to check and see if we modified the pc of the
1690 		 * pt_regs, and if so return 1 so that we don't do the
1691 		 * single stepping.
1692 		 */
1693 		if (orig_ip != instruction_pointer(regs))
1694 			return 1;
1695 		if (!ret)
1696 			return 0;
1697 	}
1698 
1699 	head = this_cpu_ptr(call->perf_events);
1700 	if (hlist_empty(head))
1701 		return 0;
1702 
1703 	dsize = __get_data_size(&tk->tp, regs, NULL);
1704 	__size = sizeof(*entry) + tk->tp.size + dsize;
1705 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1706 	size -= sizeof(u32);
1707 
1708 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1709 	if (!entry)
1710 		return 0;
1711 
1712 	entry->ip = (unsigned long)tk->rp.kp.addr;
1713 	memset(&entry[1], 0, dsize);
1714 	store_trace_args(&entry[1], &tk->tp, regs, NULL, sizeof(*entry), dsize);
1715 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1716 			      head, NULL);
1717 	return 0;
1718 }
1719 NOKPROBE_SYMBOL(kprobe_perf_func);
1720 
1721 /* Kretprobe profile handler */
1722 static void
1723 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1724 		    struct pt_regs *regs)
1725 {
1726 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1727 	struct kretprobe_trace_entry_head *entry;
1728 	struct hlist_head *head;
1729 	int size, __size, dsize;
1730 	int rctx;
1731 
1732 	if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1733 		return;
1734 
1735 	head = this_cpu_ptr(call->perf_events);
1736 	if (hlist_empty(head))
1737 		return;
1738 
1739 	dsize = __get_data_size(&tk->tp, regs, ri->data);
1740 	__size = sizeof(*entry) + tk->tp.size + dsize;
1741 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1742 	size -= sizeof(u32);
1743 
1744 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1745 	if (!entry)
1746 		return;
1747 
1748 	entry->func = (unsigned long)tk->rp.kp.addr;
1749 	entry->ret_ip = get_kretprobe_retaddr(ri);
1750 	store_trace_args(&entry[1], &tk->tp, regs, ri->data, sizeof(*entry), dsize);
1751 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1752 			      head, NULL);
1753 }
1754 NOKPROBE_SYMBOL(kretprobe_perf_func);
1755 
1756 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1757 			const char **symbol, u64 *probe_offset,
1758 			u64 *probe_addr, unsigned long *missed,
1759 			bool perf_type_tracepoint)
1760 {
1761 	const char *pevent = trace_event_name(event->tp_event);
1762 	const char *group = event->tp_event->class->system;
1763 	struct trace_kprobe *tk;
1764 
1765 	if (perf_type_tracepoint)
1766 		tk = find_trace_kprobe(pevent, group);
1767 	else
1768 		tk = trace_kprobe_primary_from_call(event->tp_event);
1769 	if (!tk)
1770 		return -EINVAL;
1771 
1772 	*fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1773 					      : BPF_FD_TYPE_KPROBE;
1774 	*probe_offset = tk->rp.kp.offset;
1775 	*probe_addr = kallsyms_show_value(current_cred()) ?
1776 		      (unsigned long)tk->rp.kp.addr : 0;
1777 	*symbol = tk->symbol;
1778 	if (missed)
1779 		*missed = trace_kprobe_missed(tk);
1780 	return 0;
1781 }
1782 #endif	/* CONFIG_PERF_EVENTS */
1783 
1784 /*
1785  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1786  *
1787  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1788  * lockless, but we can't race with this __init function.
1789  */
1790 static int kprobe_register(struct trace_event_call *event,
1791 			   enum trace_reg type, void *data)
1792 {
1793 	struct trace_event_file *file = data;
1794 
1795 	switch (type) {
1796 	case TRACE_REG_REGISTER:
1797 		return enable_trace_kprobe(event, file);
1798 	case TRACE_REG_UNREGISTER:
1799 		return disable_trace_kprobe(event, file);
1800 
1801 #ifdef CONFIG_PERF_EVENTS
1802 	case TRACE_REG_PERF_REGISTER:
1803 		return enable_trace_kprobe(event, NULL);
1804 	case TRACE_REG_PERF_UNREGISTER:
1805 		return disable_trace_kprobe(event, NULL);
1806 	case TRACE_REG_PERF_OPEN:
1807 	case TRACE_REG_PERF_CLOSE:
1808 	case TRACE_REG_PERF_ADD:
1809 	case TRACE_REG_PERF_DEL:
1810 		return 0;
1811 #endif
1812 	}
1813 	return 0;
1814 }
1815 
1816 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1817 {
1818 	struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1819 	unsigned int flags = trace_probe_load_flag(&tk->tp);
1820 	int ret = 0;
1821 
1822 	raw_cpu_inc(*tk->nhit);
1823 
1824 	if (flags & TP_FLAG_TRACE)
1825 		kprobe_trace_func(tk, regs);
1826 #ifdef CONFIG_PERF_EVENTS
1827 	if (flags & TP_FLAG_PROFILE)
1828 		ret = kprobe_perf_func(tk, regs);
1829 #endif
1830 	return ret;
1831 }
1832 NOKPROBE_SYMBOL(kprobe_dispatcher);
1833 
1834 static int
1835 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1836 {
1837 	struct kretprobe *rp = get_kretprobe(ri);
1838 	struct trace_kprobe *tk;
1839 	unsigned int flags;
1840 
1841 	/*
1842 	 * There is a small chance that get_kretprobe(ri) returns NULL when
1843 	 * the kretprobe is unregister on another CPU between kretprobe's
1844 	 * trampoline_handler and this function.
1845 	 */
1846 	if (unlikely(!rp))
1847 		return 0;
1848 
1849 	tk = container_of(rp, struct trace_kprobe, rp);
1850 	raw_cpu_inc(*tk->nhit);
1851 
1852 	flags = trace_probe_load_flag(&tk->tp);
1853 	if (flags & TP_FLAG_TRACE)
1854 		kretprobe_trace_func(tk, ri, regs);
1855 #ifdef CONFIG_PERF_EVENTS
1856 	if (flags & TP_FLAG_PROFILE)
1857 		kretprobe_perf_func(tk, ri, regs);
1858 #endif
1859 	return 0;	/* We don't tweak kernel, so just return 0 */
1860 }
1861 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1862 
1863 static struct trace_event_functions kretprobe_funcs = {
1864 	.trace		= print_kretprobe_event
1865 };
1866 
1867 static struct trace_event_functions kprobe_funcs = {
1868 	.trace		= print_kprobe_event
1869 };
1870 
1871 static struct trace_event_fields kretprobe_fields_array[] = {
1872 	{ .type = TRACE_FUNCTION_TYPE,
1873 	  .define_fields = kretprobe_event_define_fields },
1874 	{}
1875 };
1876 
1877 static struct trace_event_fields kprobe_fields_array[] = {
1878 	{ .type = TRACE_FUNCTION_TYPE,
1879 	  .define_fields = kprobe_event_define_fields },
1880 	{}
1881 };
1882 
1883 static inline void init_trace_event_call(struct trace_kprobe *tk)
1884 {
1885 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1886 
1887 	if (trace_kprobe_is_return(tk)) {
1888 		call->event.funcs = &kretprobe_funcs;
1889 		call->class->fields_array = kretprobe_fields_array;
1890 	} else {
1891 		call->event.funcs = &kprobe_funcs;
1892 		call->class->fields_array = kprobe_fields_array;
1893 	}
1894 
1895 	call->flags = TRACE_EVENT_FL_KPROBE;
1896 	call->class->reg = kprobe_register;
1897 }
1898 
1899 static int register_kprobe_event(struct trace_kprobe *tk)
1900 {
1901 	init_trace_event_call(tk);
1902 
1903 	return trace_probe_register_event_call(&tk->tp);
1904 }
1905 
1906 static int unregister_kprobe_event(struct trace_kprobe *tk)
1907 {
1908 	return trace_probe_unregister_event_call(&tk->tp);
1909 }
1910 
1911 #ifdef CONFIG_PERF_EVENTS
1912 
1913 /* create a trace_kprobe, but don't add it to global lists */
1914 struct trace_event_call *
1915 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1916 			  bool is_return)
1917 {
1918 	enum probe_print_type ptype;
1919 	struct trace_kprobe *tk __free(free_trace_kprobe) = NULL;
1920 	int ret;
1921 	char *event;
1922 
1923 	if (func) {
1924 		ret = validate_probe_symbol(func);
1925 		if (ret)
1926 			return ERR_PTR(ret);
1927 	}
1928 
1929 	/*
1930 	 * local trace_kprobes are not added to dyn_event, so they are never
1931 	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1932 	 * duplicated name here.
1933 	 */
1934 	event = func ? func : "DUMMY_EVENT";
1935 
1936 	tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1937 				offs, 0 /* maxactive */, 0 /* nargs */,
1938 				is_return);
1939 
1940 	if (IS_ERR(tk)) {
1941 		pr_info("Failed to allocate trace_probe.(%d)\n",
1942 			(int)PTR_ERR(tk));
1943 		return ERR_CAST(tk);
1944 	}
1945 
1946 	init_trace_event_call(tk);
1947 
1948 	ptype = trace_kprobe_is_return(tk) ?
1949 		PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1950 	if (traceprobe_set_print_fmt(&tk->tp, ptype) < 0)
1951 		return ERR_PTR(-ENOMEM);
1952 
1953 	ret = __register_trace_kprobe(tk);
1954 	if (ret < 0)
1955 		return ERR_PTR(ret);
1956 
1957 	return trace_probe_event_call(&(no_free_ptr(tk)->tp));
1958 }
1959 
1960 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1961 {
1962 	struct trace_kprobe *tk;
1963 
1964 	tk = trace_kprobe_primary_from_call(event_call);
1965 	if (unlikely(!tk))
1966 		return;
1967 
1968 	if (trace_probe_is_enabled(&tk->tp)) {
1969 		WARN_ON(1);
1970 		return;
1971 	}
1972 
1973 	__unregister_trace_kprobe(tk);
1974 
1975 	free_trace_kprobe(tk);
1976 }
1977 #endif /* CONFIG_PERF_EVENTS */
1978 
1979 static __init void enable_boot_kprobe_events(void)
1980 {
1981 	struct trace_array *tr = top_trace_array();
1982 	struct trace_event_file *file;
1983 	struct trace_kprobe *tk;
1984 	struct dyn_event *pos;
1985 
1986 	if (trace_kprobe_list_empty())
1987 		return;
1988 
1989 	guard(mutex)(&event_mutex);
1990 	for_each_trace_kprobe(tk, pos) {
1991 		list_for_each_entry(file, &tr->events, list)
1992 			if (file->event_call == trace_probe_event_call(&tk->tp))
1993 				trace_event_enable_disable(file, 1, 0);
1994 	}
1995 }
1996 
1997 static __init void setup_boot_kprobe_events(void)
1998 {
1999 	char *p, *cmd = kprobe_boot_events_buf;
2000 	int ret;
2001 
2002 	strreplace(kprobe_boot_events_buf, ',', ' ');
2003 
2004 	while (cmd && *cmd != '\0') {
2005 		p = strchr(cmd, ';');
2006 		if (p)
2007 			*p++ = '\0';
2008 
2009 		ret = create_or_delete_trace_kprobe(cmd);
2010 		if (ret)
2011 			pr_warn("Failed to add event(%d): %s\n", ret, cmd);
2012 
2013 		cmd = p;
2014 	}
2015 
2016 	enable_boot_kprobe_events();
2017 }
2018 
2019 /*
2020  * Register dynevent at core_initcall. This allows kernel to setup kprobe
2021  * events in postcore_initcall without tracefs.
2022  */
2023 static __init int init_kprobe_trace_early(void)
2024 {
2025 	int ret;
2026 
2027 	ret = dyn_event_register(&trace_kprobe_ops);
2028 	if (ret)
2029 		return ret;
2030 
2031 	if (trace_kprobe_register_module_notifier())
2032 		return -EINVAL;
2033 
2034 	return 0;
2035 }
2036 core_initcall(init_kprobe_trace_early);
2037 
2038 /* Make a tracefs interface for controlling probe points */
2039 static __init int init_kprobe_trace(void)
2040 {
2041 	int ret;
2042 
2043 	ret = tracing_init_dentry();
2044 	if (ret)
2045 		return 0;
2046 
2047 	/* Event list interface */
2048 	trace_create_file("kprobe_events", TRACE_MODE_WRITE,
2049 			  NULL, NULL, &kprobe_events_ops);
2050 
2051 	/* Profile interface */
2052 	trace_create_file("kprobe_profile", TRACE_MODE_READ,
2053 			  NULL, NULL, &kprobe_profile_ops);
2054 
2055 	/* If no 'kprobe_event=' cmd is provided, return directly. */
2056 	if (kprobe_boot_events_buf[0] == '\0')
2057 		return 0;
2058 
2059 	setup_boot_kprobe_events();
2060 
2061 	return 0;
2062 }
2063 fs_initcall(init_kprobe_trace);
2064 
2065 
2066 #ifdef CONFIG_FTRACE_STARTUP_TEST
2067 static __init struct trace_event_file *
2068 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
2069 {
2070 	struct trace_event_file *file;
2071 
2072 	list_for_each_entry(file, &tr->events, list)
2073 		if (file->event_call == trace_probe_event_call(&tk->tp))
2074 			return file;
2075 
2076 	return NULL;
2077 }
2078 
2079 /*
2080  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
2081  * stage, we can do this lockless.
2082  */
2083 static __init int kprobe_trace_self_tests_init(void)
2084 {
2085 	int ret, warn = 0;
2086 	int (*target)(int, int, int, int, int, int);
2087 	struct trace_kprobe *tk;
2088 	struct trace_event_file *file;
2089 
2090 	if (unlikely(tracing_disabled))
2091 		return -ENODEV;
2092 
2093 	if (tracing_selftest_disabled)
2094 		return 0;
2095 
2096 	target = kprobe_trace_selftest_target;
2097 
2098 	pr_info("Testing kprobe tracing: ");
2099 
2100 	ret = create_or_delete_trace_kprobe("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)");
2101 	if (WARN_ONCE(ret, "error on probing function entry.")) {
2102 		warn++;
2103 	} else {
2104 		/* Enable trace point */
2105 		tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2106 		if (WARN_ONCE(tk == NULL, "error on probing function entry.")) {
2107 			warn++;
2108 		} else {
2109 			file = find_trace_probe_file(tk, top_trace_array());
2110 			if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2111 				warn++;
2112 			} else
2113 				enable_trace_kprobe(
2114 					trace_probe_event_call(&tk->tp), file);
2115 		}
2116 	}
2117 
2118 	ret = create_or_delete_trace_kprobe("r:testprobe2 kprobe_trace_selftest_target $retval");
2119 	if (WARN_ONCE(ret, "error on probing function return.")) {
2120 		warn++;
2121 	} else {
2122 		/* Enable trace point */
2123 		tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2124 		if (WARN_ONCE(tk == NULL, "error on getting 2nd new probe.")) {
2125 			warn++;
2126 		} else {
2127 			file = find_trace_probe_file(tk, top_trace_array());
2128 			if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2129 				warn++;
2130 			} else
2131 				enable_trace_kprobe(
2132 					trace_probe_event_call(&tk->tp), file);
2133 		}
2134 	}
2135 
2136 	if (warn)
2137 		goto end;
2138 
2139 	ret = target(1, 2, 3, 4, 5, 6);
2140 
2141 	/*
2142 	 * Not expecting an error here, the check is only to prevent the
2143 	 * optimizer from removing the call to target() as otherwise there
2144 	 * are no side-effects and the call is never performed.
2145 	 */
2146 	if (ret != 21)
2147 		warn++;
2148 
2149 	/* Disable trace points before removing it */
2150 	tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2151 	if (WARN_ONCE(tk == NULL, "error on getting test probe.")) {
2152 		warn++;
2153 	} else {
2154 		if (WARN_ONCE(trace_kprobe_nhit(tk) != 1,
2155 				 "incorrect number of testprobe hits."))
2156 			warn++;
2157 
2158 		file = find_trace_probe_file(tk, top_trace_array());
2159 		if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2160 			warn++;
2161 		} else
2162 			disable_trace_kprobe(
2163 				trace_probe_event_call(&tk->tp), file);
2164 	}
2165 
2166 	tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2167 	if (WARN_ONCE(tk == NULL, "error on getting 2nd test probe.")) {
2168 		warn++;
2169 	} else {
2170 		if (WARN_ONCE(trace_kprobe_nhit(tk) != 1,
2171 				 "incorrect number of testprobe2 hits."))
2172 			warn++;
2173 
2174 		file = find_trace_probe_file(tk, top_trace_array());
2175 		if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2176 			warn++;
2177 		} else
2178 			disable_trace_kprobe(
2179 				trace_probe_event_call(&tk->tp), file);
2180 	}
2181 
2182 	ret = create_or_delete_trace_kprobe("-:testprobe");
2183 	if (WARN_ONCE(ret, "error on deleting a probe."))
2184 		warn++;
2185 
2186 	ret = create_or_delete_trace_kprobe("-:testprobe2");
2187 	if (WARN_ONCE(ret, "error on deleting a probe."))
2188 		warn++;
2189 
2190 
2191 end:
2192 	/*
2193 	 * Wait for the optimizer work to finish. Otherwise it might fiddle
2194 	 * with probes in already freed __init text.
2195 	 */
2196 	wait_for_kprobe_optimizer();
2197 	if (warn)
2198 		pr_cont("NG: Some tests are failed. Please check them.\n");
2199 	else
2200 		pr_cont("OK\n");
2201 	return 0;
2202 }
2203 
2204 late_initcall(kprobe_trace_self_tests_init);
2205 
2206 #endif
2207