xref: /linux/kernel/trace/trace_kprobe.c (revision 9e1e9d660255d7216067193d774f338d08d8528d)
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_flex(*tk, tp.args, nargs);
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 	/*
769 	 * If the symbol is found in vmlinux, use vmlinux resolution only.
770 	 * This prevents module symbols from shadowing vmlinux symbols
771 	 * and causing -EADDRNOTAVAIL for unqualified kprobe targets.
772 	 */
773 	if (!mod && ctx.count > 0)
774 		return ctx.count;
775 
776 	module_kallsyms_on_each_symbol(mod, count_mod_symbols, &ctx);
777 
778 	return ctx.count;
779 }
780 
781 static int validate_module_probe_symbol(const char *modname, const char *symbol)
782 {
783 	unsigned int count = number_of_same_symbols(modname, symbol);
784 
785 	if (count > 1) {
786 		/*
787 		 * Users should use ADDR to remove the ambiguity of
788 		 * using KSYM only.
789 		 */
790 		return -EADDRNOTAVAIL;
791 	} else if (count == 0) {
792 		/*
793 		 * We can return ENOENT earlier than when register the
794 		 * kprobe.
795 		 */
796 		return -ENOENT;
797 	}
798 	return 0;
799 }
800 
801 #ifdef CONFIG_MODULES
802 /* Return NULL if the module is not loaded or under unloading. */
803 static struct module *try_module_get_by_name(const char *name)
804 {
805 	struct module *mod;
806 
807 	guard(rcu)();
808 	mod = find_module(name);
809 	if (mod && !try_module_get(mod))
810 		mod = NULL;
811 	return mod;
812 }
813 #else
814 #define try_module_get_by_name(name)	(NULL)
815 #endif
816 
817 static int validate_probe_symbol(char *symbol)
818 {
819 	struct module *mod = NULL;
820 	char *modname = NULL, *p;
821 	int ret = 0;
822 
823 	p = strchr(symbol, ':');
824 	if (p) {
825 		modname = symbol;
826 		symbol = p + 1;
827 		*p = '\0';
828 		mod = try_module_get_by_name(modname);
829 		if (!mod)
830 			goto out;
831 	}
832 
833 	ret = validate_module_probe_symbol(modname, symbol);
834 out:
835 	if (p)
836 		*p = ':';
837 	if (mod)
838 		module_put(mod);
839 	return ret;
840 }
841 
842 static int trace_kprobe_entry_handler(struct kretprobe_instance *ri,
843 				      struct pt_regs *regs);
844 
845 static int trace_kprobe_create_internal(int argc, const char *argv[],
846 					struct traceprobe_parse_context *ctx)
847 {
848 	/*
849 	 * Argument syntax:
850 	 *  - Add kprobe:
851 	 *      p[:[GRP/][EVENT]] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
852 	 *  - Add kretprobe:
853 	 *      r[MAXACTIVE][:[GRP/][EVENT]] [MOD:]KSYM[+0] [FETCHARGS]
854 	 *    Or
855 	 *      p[:[GRP/][EVENT]] [MOD:]KSYM[+0]%return [FETCHARGS]
856 	 *
857 	 * Fetch args:
858 	 *  $retval	: fetch return value
859 	 *  $stack	: fetch stack address
860 	 *  $stackN	: fetch Nth of stack (N:0-)
861 	 *  $comm       : fetch current task comm
862 	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
863 	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
864 	 *  %REG	: fetch register REG
865 	 * Dereferencing memory fetch:
866 	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
867 	 * Alias name of args:
868 	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
869 	 * Type of args:
870 	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
871 	 */
872 	struct trace_kprobe *tk __free(free_trace_kprobe) = NULL;
873 	const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
874 	const char **new_argv __free(kfree) = NULL;
875 	int i, len, new_argc = 0, ret = 0;
876 	char *symbol __free(kfree) = NULL;
877 	char *ebuf __free(kfree) = NULL;
878 	char *gbuf __free(kfree) = NULL;
879 	char *abuf __free(kfree) = NULL;
880 	char *dbuf __free(kfree) = NULL;
881 	enum probe_print_type ptype;
882 	bool is_return = false;
883 	int maxactive = 0;
884 	void *addr = NULL;
885 	char *tmp = NULL;
886 	long offset = 0;
887 
888 	switch (argv[0][0]) {
889 	case 'r':
890 		is_return = true;
891 		break;
892 	case 'p':
893 		break;
894 	default:
895 		return -ECANCELED;
896 	}
897 	if (argc < 2)
898 		return -ECANCELED;
899 
900 	event = strchr(&argv[0][1], ':');
901 	if (event)
902 		event++;
903 
904 	if (isdigit(argv[0][1])) {
905 		char *buf __free(kfree) = NULL;
906 
907 		if (!is_return) {
908 			trace_probe_log_err(1, BAD_MAXACT_TYPE);
909 			return -EINVAL;
910 		}
911 		if (event)
912 			len = event - &argv[0][1] - 1;
913 		else
914 			len = strlen(&argv[0][1]);
915 		if (len > MAX_EVENT_NAME_LEN - 1) {
916 			trace_probe_log_err(1, BAD_MAXACT);
917 			return -EINVAL;
918 		}
919 		buf = kmemdup(&argv[0][1], len + 1, GFP_KERNEL);
920 		if (!buf)
921 			return -ENOMEM;
922 		buf[len] = '\0';
923 		ret = kstrtouint(buf, 0, &maxactive);
924 		if (ret || !maxactive) {
925 			trace_probe_log_err(1, BAD_MAXACT);
926 			return -EINVAL;
927 		}
928 		/* kretprobes instances are iterated over via a list. The
929 		 * maximum should stay reasonable.
930 		 */
931 		if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
932 			trace_probe_log_err(1, MAXACT_TOO_BIG);
933 			return -EINVAL;
934 		}
935 	}
936 
937 	/* try to parse an address. if that fails, try to read the
938 	 * input as a symbol. */
939 	if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
940 		trace_probe_log_set_index(1);
941 		/* Check whether uprobe event specified */
942 		if (strchr(argv[1], '/') && strchr(argv[1], ':'))
943 			return -ECANCELED;
944 
945 		/* a symbol specified */
946 		symbol = kstrdup(argv[1], GFP_KERNEL);
947 		if (!symbol)
948 			return -ENOMEM;
949 
950 		tmp = strchr(symbol, '%');
951 		if (tmp) {
952 			if (!strcmp(tmp, "%return")) {
953 				*tmp = '\0';
954 				is_return = true;
955 			} else {
956 				trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
957 				return -EINVAL;
958 			}
959 		}
960 
961 		/* TODO: support .init module functions */
962 		ret = traceprobe_split_symbol_offset(symbol, &offset);
963 		if (ret || offset < 0 || offset > UINT_MAX) {
964 			trace_probe_log_err(0, BAD_PROBE_ADDR);
965 			return -EINVAL;
966 		}
967 		ret = validate_probe_symbol(symbol);
968 		if (ret) {
969 			if (ret == -EADDRNOTAVAIL)
970 				trace_probe_log_err(0, NON_UNIQ_SYMBOL);
971 			else
972 				trace_probe_log_err(0, BAD_PROBE_ADDR);
973 			return -EINVAL;
974 		}
975 		if (is_return)
976 			ctx->flags |= TPARG_FL_RETURN;
977 		ret = kprobe_on_func_entry(NULL, symbol, offset);
978 		if (ret == 0 && !is_return)
979 			ctx->flags |= TPARG_FL_FENTRY;
980 		/* Defer the ENOENT case until register kprobe */
981 		if (ret == -EINVAL && is_return) {
982 			trace_probe_log_err(0, BAD_RETPROBE);
983 			return -EINVAL;
984 		}
985 	}
986 
987 	trace_probe_log_set_index(0);
988 	if (event) {
989 		gbuf = kmalloc(MAX_EVENT_NAME_LEN, GFP_KERNEL);
990 		if (!gbuf)
991 			return -ENOMEM;
992 		ret = traceprobe_parse_event_name(&event, &group, gbuf,
993 						  event - argv[0]);
994 		if (ret)
995 			return ret;
996 	}
997 
998 	if (!event) {
999 		/* Make a new event name */
1000 		ebuf = kmalloc(MAX_EVENT_NAME_LEN, GFP_KERNEL);
1001 		if (!ebuf)
1002 			return -ENOMEM;
1003 		if (symbol)
1004 			snprintf(ebuf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
1005 				 is_return ? 'r' : 'p', symbol, offset);
1006 		else
1007 			snprintf(ebuf, MAX_EVENT_NAME_LEN, "%c_0x%p",
1008 				 is_return ? 'r' : 'p', addr);
1009 		sanitize_event_name(ebuf);
1010 		event = ebuf;
1011 	}
1012 
1013 	abuf = kmalloc(MAX_BTF_ARGS_LEN, GFP_KERNEL);
1014 	if (!abuf)
1015 		return -ENOMEM;
1016 	argc -= 2; argv += 2;
1017 	ctx->funcname = symbol;
1018 	new_argv = traceprobe_expand_meta_args(argc, argv, &new_argc,
1019 					       abuf, MAX_BTF_ARGS_LEN, ctx);
1020 	if (IS_ERR(new_argv)) {
1021 		ret = PTR_ERR(new_argv);
1022 		new_argv = NULL;
1023 		return ret;
1024 	}
1025 	if (new_argv) {
1026 		argc = new_argc;
1027 		argv = new_argv;
1028 	}
1029 	if (argc > MAX_TRACE_ARGS) {
1030 		trace_probe_log_set_index(2);
1031 		trace_probe_log_err(0, TOO_MANY_ARGS);
1032 		return -E2BIG;
1033 	}
1034 
1035 	ret = traceprobe_expand_dentry_args(argc, argv, &dbuf);
1036 	if (ret)
1037 		return ret;
1038 
1039 	/* setup a probe */
1040 	tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
1041 				argc, is_return);
1042 	if (IS_ERR(tk)) {
1043 		ret = PTR_ERR(tk);
1044 		/* This must return -ENOMEM, else there is a bug */
1045 		WARN_ON_ONCE(ret != -ENOMEM);
1046 		return ret;	/* We know tk is not allocated */
1047 	}
1048 
1049 	/* parse arguments */
1050 	for (i = 0; i < argc; i++) {
1051 		trace_probe_log_set_index(i + 2);
1052 		ctx->offset = 0;
1053 		ret = traceprobe_parse_probe_arg(&tk->tp, i, argv[i], ctx);
1054 		if (ret)
1055 			return ret;	/* This can be -ENOMEM */
1056 	}
1057 	/* entry handler for kretprobe */
1058 	if (is_return && tk->tp.entry_arg) {
1059 		tk->rp.entry_handler = trace_kprobe_entry_handler;
1060 		tk->rp.data_size = traceprobe_get_entry_data_size(&tk->tp);
1061 	}
1062 
1063 	ptype = is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1064 	ret = traceprobe_set_print_fmt(&tk->tp, ptype);
1065 	if (ret < 0)
1066 		return ret;
1067 
1068 	ret = register_trace_kprobe(tk);
1069 	if (ret) {
1070 		trace_probe_log_set_index(1);
1071 		if (ret == -EILSEQ)
1072 			trace_probe_log_err(0, BAD_INSN_BNDRY);
1073 		else if (ret == -ENOENT)
1074 			trace_probe_log_err(0, BAD_PROBE_ADDR);
1075 		else if (ret != -ENOMEM && ret != -EEXIST)
1076 			trace_probe_log_err(0, FAIL_REG_PROBE);
1077 		return ret;
1078 	}
1079 	/*
1080 	 * Here, 'tk' has been registered to the list successfully,
1081 	 * so we don't need to free it.
1082 	 */
1083 	tk = NULL;
1084 
1085 	return 0;
1086 }
1087 
1088 static int trace_kprobe_create_cb(int argc, const char *argv[])
1089 {
1090 	struct traceprobe_parse_context *ctx __free(traceprobe_parse_context) = NULL;
1091 	int ret;
1092 
1093 	ctx = kzalloc_obj(*ctx);
1094 	if (!ctx)
1095 		return -ENOMEM;
1096 	ctx->flags = TPARG_FL_KERNEL;
1097 
1098 	trace_probe_log_init("trace_kprobe", argc, argv);
1099 
1100 	ret = trace_kprobe_create_internal(argc, argv, ctx);
1101 
1102 	trace_probe_log_clear();
1103 	return ret;
1104 }
1105 
1106 static int trace_kprobe_create(const char *raw_command)
1107 {
1108 	return trace_probe_create(raw_command, trace_kprobe_create_cb);
1109 }
1110 
1111 static int create_or_delete_trace_kprobe(const char *raw_command)
1112 {
1113 	int ret;
1114 
1115 	if (raw_command[0] == '-')
1116 		return dyn_event_release(raw_command, &trace_kprobe_ops);
1117 
1118 	ret = dyn_event_create(raw_command, &trace_kprobe_ops);
1119 	return ret == -ECANCELED ? -EINVAL : ret;
1120 }
1121 
1122 static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
1123 {
1124 	return create_or_delete_trace_kprobe(cmd->seq.buffer);
1125 }
1126 
1127 /**
1128  * kprobe_event_cmd_init - Initialize a kprobe event command object
1129  * @cmd: A pointer to the dynevent_cmd struct representing the new event
1130  * @buf: A pointer to the buffer used to build the command
1131  * @maxlen: The length of the buffer passed in @buf
1132  *
1133  * Initialize a synthetic event command object.  Use this before
1134  * calling any of the other kprobe_event functions.
1135  */
1136 void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
1137 {
1138 	dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
1139 			  trace_kprobe_run_command);
1140 }
1141 EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
1142 
1143 /**
1144  * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
1145  * @cmd: A pointer to the dynevent_cmd struct representing the new event
1146  * @kretprobe: Is this a return probe?
1147  * @name: The name of the kprobe event
1148  * @loc: The location of the kprobe event
1149  * @...: Variable number of arg (pairs), one pair for each field
1150  *
1151  * NOTE: Users normally won't want to call this function directly, but
1152  * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
1153  * adds a NULL to the end of the arg list.  If this function is used
1154  * directly, make sure the last arg in the variable arg list is NULL.
1155  *
1156  * Generate a kprobe event command to be executed by
1157  * kprobe_event_gen_cmd_end().  This function can be used to generate the
1158  * complete command or only the first part of it; in the latter case,
1159  * kprobe_event_add_fields() can be used to add more fields following this.
1160  *
1161  * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
1162  * returns -EINVAL if @loc == NULL.
1163  *
1164  * Return: 0 if successful, error otherwise.
1165  */
1166 int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
1167 				 const char *name, const char *loc, ...)
1168 {
1169 	char buf[MAX_EVENT_NAME_LEN];
1170 	struct dynevent_arg arg;
1171 	va_list args;
1172 	int ret;
1173 
1174 	if (cmd->type != DYNEVENT_TYPE_KPROBE)
1175 		return -EINVAL;
1176 
1177 	if (!loc)
1178 		return -EINVAL;
1179 
1180 	if (kretprobe)
1181 		snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
1182 	else
1183 		snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
1184 
1185 	ret = dynevent_str_add(cmd, buf);
1186 	if (ret)
1187 		return ret;
1188 
1189 	dynevent_arg_init(&arg, 0);
1190 	arg.str = loc;
1191 	ret = dynevent_arg_add(cmd, &arg, NULL);
1192 	if (ret)
1193 		return ret;
1194 
1195 	va_start(args, loc);
1196 	for (;;) {
1197 		const char *field;
1198 
1199 		field = va_arg(args, const char *);
1200 		if (!field)
1201 			break;
1202 
1203 		if (++cmd->n_fields > MAX_TRACE_ARGS) {
1204 			ret = -EINVAL;
1205 			break;
1206 		}
1207 
1208 		arg.str = field;
1209 		ret = dynevent_arg_add(cmd, &arg, NULL);
1210 		if (ret)
1211 			break;
1212 	}
1213 	va_end(args);
1214 
1215 	return ret;
1216 }
1217 EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1218 
1219 /**
1220  * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1221  * @cmd: A pointer to the dynevent_cmd struct representing the new event
1222  * @...: Variable number of arg (pairs), one pair for each field
1223  *
1224  * NOTE: Users normally won't want to call this function directly, but
1225  * rather use the kprobe_event_add_fields() wrapper, which
1226  * automatically adds a NULL to the end of the arg list.  If this
1227  * function is used directly, make sure the last arg in the variable
1228  * arg list is NULL.
1229  *
1230  * Add probe fields to an existing kprobe command using a variable
1231  * list of args.  Fields are added in the same order they're listed.
1232  *
1233  * Return: 0 if successful, error otherwise.
1234  */
1235 int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1236 {
1237 	struct dynevent_arg arg;
1238 	va_list args;
1239 	int ret = 0;
1240 
1241 	if (cmd->type != DYNEVENT_TYPE_KPROBE)
1242 		return -EINVAL;
1243 
1244 	dynevent_arg_init(&arg, 0);
1245 
1246 	va_start(args, cmd);
1247 	for (;;) {
1248 		const char *field;
1249 
1250 		field = va_arg(args, const char *);
1251 		if (!field)
1252 			break;
1253 
1254 		if (++cmd->n_fields > MAX_TRACE_ARGS) {
1255 			ret = -EINVAL;
1256 			break;
1257 		}
1258 
1259 		arg.str = field;
1260 		ret = dynevent_arg_add(cmd, &arg, NULL);
1261 		if (ret)
1262 			break;
1263 	}
1264 	va_end(args);
1265 
1266 	return ret;
1267 }
1268 EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1269 
1270 /**
1271  * kprobe_event_delete - Delete a kprobe event
1272  * @name: The name of the kprobe event to delete
1273  *
1274  * Delete a kprobe event with the give @name from kernel code rather
1275  * than directly from the command line.
1276  *
1277  * Return: 0 if successful, error otherwise.
1278  */
1279 int kprobe_event_delete(const char *name)
1280 {
1281 	char buf[MAX_EVENT_NAME_LEN];
1282 
1283 	snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1284 
1285 	return create_or_delete_trace_kprobe(buf);
1286 }
1287 EXPORT_SYMBOL_GPL(kprobe_event_delete);
1288 
1289 static int trace_kprobe_release(struct dyn_event *ev)
1290 {
1291 	struct trace_kprobe *tk = to_trace_kprobe(ev);
1292 	int ret = unregister_trace_kprobe(tk);
1293 
1294 	if (!ret)
1295 		free_trace_kprobe(tk);
1296 	return ret;
1297 }
1298 
1299 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1300 {
1301 	struct trace_kprobe *tk = to_trace_kprobe(ev);
1302 	int i;
1303 
1304 	seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1305 	if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1306 		seq_printf(m, "%d", tk->rp.maxactive);
1307 	seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1308 				trace_probe_name(&tk->tp));
1309 
1310 	if (!tk->symbol)
1311 		seq_printf(m, " 0x%p", tk->rp.kp.addr);
1312 	else if (tk->rp.kp.offset)
1313 		seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1314 			   tk->rp.kp.offset);
1315 	else
1316 		seq_printf(m, " %s", trace_kprobe_symbol(tk));
1317 
1318 	for (i = 0; i < tk->tp.nr_args; i++)
1319 		seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1320 	seq_putc(m, '\n');
1321 
1322 	return 0;
1323 }
1324 
1325 static int probes_seq_show(struct seq_file *m, void *v)
1326 {
1327 	struct dyn_event *ev = v;
1328 
1329 	if (!is_trace_kprobe(ev))
1330 		return 0;
1331 
1332 	return trace_kprobe_show(m, ev);
1333 }
1334 
1335 static const struct seq_operations probes_seq_op = {
1336 	.start  = dyn_event_seq_start,
1337 	.next   = dyn_event_seq_next,
1338 	.stop   = dyn_event_seq_stop,
1339 	.show   = probes_seq_show
1340 };
1341 
1342 static int probes_open(struct inode *inode, struct file *file)
1343 {
1344 	int ret;
1345 
1346 	ret = security_locked_down(LOCKDOWN_TRACEFS);
1347 	if (ret)
1348 		return ret;
1349 
1350 	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1351 		ret = dyn_events_release_all(&trace_kprobe_ops);
1352 		if (ret < 0)
1353 			return ret;
1354 	}
1355 
1356 	return seq_open(file, &probes_seq_op);
1357 }
1358 
1359 static ssize_t probes_write(struct file *file, const char __user *buffer,
1360 			    size_t count, loff_t *ppos)
1361 {
1362 	return trace_parse_run_command(file, buffer, count, ppos,
1363 				       create_or_delete_trace_kprobe);
1364 }
1365 
1366 static const struct file_operations kprobe_events_ops = {
1367 	.owner          = THIS_MODULE,
1368 	.open           = probes_open,
1369 	.read           = seq_read,
1370 	.llseek         = seq_lseek,
1371 	.release        = seq_release,
1372 	.write		= probes_write,
1373 };
1374 
1375 static unsigned long trace_kprobe_missed(struct trace_kprobe *tk)
1376 {
1377 	return trace_kprobe_is_return(tk) ?
1378 		tk->rp.kp.nmissed + tk->rp.nmissed : tk->rp.kp.nmissed;
1379 }
1380 
1381 /* Probes profiling interfaces */
1382 static int probes_profile_seq_show(struct seq_file *m, void *v)
1383 {
1384 	struct dyn_event *ev = v;
1385 	struct trace_kprobe *tk;
1386 	unsigned long nmissed;
1387 
1388 	if (!is_trace_kprobe(ev))
1389 		return 0;
1390 
1391 	tk = to_trace_kprobe(ev);
1392 	nmissed = trace_kprobe_missed(tk);
1393 	seq_printf(m, "  %-44s %15lu %15lu\n",
1394 		   trace_probe_name(&tk->tp),
1395 		   trace_kprobe_nhit(tk),
1396 		   nmissed);
1397 
1398 	return 0;
1399 }
1400 
1401 static const struct seq_operations profile_seq_op = {
1402 	.start  = dyn_event_seq_start,
1403 	.next   = dyn_event_seq_next,
1404 	.stop   = dyn_event_seq_stop,
1405 	.show   = probes_profile_seq_show
1406 };
1407 
1408 static int profile_open(struct inode *inode, struct file *file)
1409 {
1410 	int ret;
1411 
1412 	ret = security_locked_down(LOCKDOWN_TRACEFS);
1413 	if (ret)
1414 		return ret;
1415 
1416 	return seq_open(file, &profile_seq_op);
1417 }
1418 
1419 static const struct file_operations kprobe_profile_ops = {
1420 	.owner          = THIS_MODULE,
1421 	.open           = profile_open,
1422 	.read           = seq_read,
1423 	.llseek         = seq_lseek,
1424 	.release        = seq_release,
1425 };
1426 
1427 /* Note that we don't verify it, since the code does not come from user space */
1428 static int
1429 process_fetch_insn(struct fetch_insn *code, void *rec, void *edata,
1430 		   void *dest, void *base)
1431 {
1432 	struct pt_regs *regs = rec;
1433 	unsigned long val;
1434 	int ret;
1435 
1436 retry:
1437 	/* 1st stage: get value from context */
1438 	switch (code->op) {
1439 	case FETCH_OP_REG:
1440 		val = regs_get_register(regs, code->param);
1441 		break;
1442 	case FETCH_OP_STACK:
1443 		val = regs_get_kernel_stack_nth(regs, code->param);
1444 		break;
1445 	case FETCH_OP_STACKP:
1446 		val = kernel_stack_pointer(regs);
1447 		break;
1448 	case FETCH_OP_RETVAL:
1449 		val = regs_return_value(regs);
1450 		break;
1451 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1452 	case FETCH_OP_ARG:
1453 		val = regs_get_kernel_argument(regs, code->param);
1454 		break;
1455 	case FETCH_OP_EDATA:
1456 		val = *(unsigned long *)((unsigned long)edata + code->offset);
1457 		break;
1458 #endif
1459 	case FETCH_NOP_SYMBOL:	/* Ignore a place holder */
1460 		code++;
1461 		goto retry;
1462 	default:
1463 		ret = process_common_fetch_insn(code, &val);
1464 		if (ret < 0)
1465 			return ret;
1466 	}
1467 	code++;
1468 
1469 	return process_fetch_insn_bottom(code, val, dest, base);
1470 }
1471 NOKPROBE_SYMBOL(process_fetch_insn)
1472 
1473 /* Kprobe handler */
1474 static nokprobe_inline void
1475 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1476 		    struct trace_event_file *trace_file)
1477 {
1478 	struct kprobe_trace_entry_head *entry;
1479 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1480 	struct trace_event_buffer fbuffer;
1481 	int dsize;
1482 
1483 	WARN_ON(call != trace_file->event_call);
1484 
1485 	if (trace_trigger_soft_disabled(trace_file))
1486 		return;
1487 
1488 	dsize = __get_data_size(&tk->tp, regs, NULL);
1489 
1490 	entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1491 					   sizeof(*entry) + tk->tp.size + dsize);
1492 	if (!entry)
1493 		return;
1494 
1495 	fbuffer.regs = regs;
1496 	entry->ip = (unsigned long)tk->rp.kp.addr;
1497 	store_trace_args(&entry[1], &tk->tp, regs, NULL, sizeof(*entry), dsize);
1498 
1499 	trace_event_buffer_commit(&fbuffer);
1500 }
1501 
1502 static void
1503 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1504 {
1505 	struct event_file_link *link;
1506 
1507 	trace_probe_for_each_link_rcu(link, &tk->tp)
1508 		__kprobe_trace_func(tk, regs, link->file);
1509 }
1510 NOKPROBE_SYMBOL(kprobe_trace_func);
1511 
1512 /* Kretprobe handler */
1513 
1514 static int trace_kprobe_entry_handler(struct kretprobe_instance *ri,
1515 				      struct pt_regs *regs)
1516 {
1517 	struct kretprobe *rp = get_kretprobe(ri);
1518 	struct trace_kprobe *tk;
1519 
1520 	/*
1521 	 * There is a small chance that get_kretprobe(ri) returns NULL when
1522 	 * the kretprobe is unregister on another CPU between kretprobe's
1523 	 * trampoline_handler and this function.
1524 	 */
1525 	if (unlikely(!rp))
1526 		return -ENOENT;
1527 
1528 	tk = container_of(rp, struct trace_kprobe, rp);
1529 
1530 	/* store argument values into ri->data as entry data */
1531 	if (tk->tp.entry_arg)
1532 		store_trace_entry_data(ri->data, &tk->tp, regs);
1533 
1534 	return 0;
1535 }
1536 
1537 
1538 static nokprobe_inline void
1539 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1540 		       struct pt_regs *regs,
1541 		       struct trace_event_file *trace_file)
1542 {
1543 	struct kretprobe_trace_entry_head *entry;
1544 	struct trace_event_buffer fbuffer;
1545 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1546 	int dsize;
1547 
1548 	WARN_ON(call != trace_file->event_call);
1549 
1550 	if (trace_trigger_soft_disabled(trace_file))
1551 		return;
1552 
1553 	dsize = __get_data_size(&tk->tp, regs, ri->data);
1554 
1555 	entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1556 					   sizeof(*entry) + tk->tp.size + dsize);
1557 	if (!entry)
1558 		return;
1559 
1560 	fbuffer.regs = regs;
1561 	entry->func = (unsigned long)tk->rp.kp.addr;
1562 	entry->ret_ip = get_kretprobe_retaddr(ri);
1563 	store_trace_args(&entry[1], &tk->tp, regs, ri->data, sizeof(*entry), dsize);
1564 
1565 	trace_event_buffer_commit(&fbuffer);
1566 }
1567 
1568 static void
1569 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1570 		     struct pt_regs *regs)
1571 {
1572 	struct event_file_link *link;
1573 
1574 	trace_probe_for_each_link_rcu(link, &tk->tp)
1575 		__kretprobe_trace_func(tk, ri, regs, link->file);
1576 }
1577 NOKPROBE_SYMBOL(kretprobe_trace_func);
1578 
1579 /* Event entry printers */
1580 static enum print_line_t
1581 print_kprobe_event(struct trace_iterator *iter, int flags,
1582 		   struct trace_event *event)
1583 {
1584 	struct kprobe_trace_entry_head *field;
1585 	struct trace_seq *s = &iter->seq;
1586 	struct trace_probe *tp;
1587 
1588 	field = (struct kprobe_trace_entry_head *)iter->ent;
1589 	tp = trace_probe_primary_from_call(
1590 		container_of(event, struct trace_event_call, event));
1591 	if (WARN_ON_ONCE(!tp))
1592 		goto out;
1593 
1594 	trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1595 
1596 	if (!seq_print_ip_sym_offset(s, field->ip, flags))
1597 		goto out;
1598 
1599 	trace_seq_putc(s, ')');
1600 
1601 	if (trace_probe_print_args(s, tp->args, tp->nr_args,
1602 			     (u8 *)&field[1], field) < 0)
1603 		goto out;
1604 
1605 	trace_seq_putc(s, '\n');
1606  out:
1607 	return trace_handle_return(s);
1608 }
1609 
1610 static enum print_line_t
1611 print_kretprobe_event(struct trace_iterator *iter, int flags,
1612 		      struct trace_event *event)
1613 {
1614 	struct kretprobe_trace_entry_head *field;
1615 	struct trace_seq *s = &iter->seq;
1616 	struct trace_probe *tp;
1617 
1618 	field = (struct kretprobe_trace_entry_head *)iter->ent;
1619 	tp = trace_probe_primary_from_call(
1620 		container_of(event, struct trace_event_call, event));
1621 	if (WARN_ON_ONCE(!tp))
1622 		goto out;
1623 
1624 	trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1625 
1626 	if (!seq_print_ip_sym_offset(s, field->ret_ip, flags))
1627 		goto out;
1628 
1629 	trace_seq_puts(s, " <- ");
1630 
1631 	if (!seq_print_ip_sym_no_offset(s, field->func, flags))
1632 		goto out;
1633 
1634 	trace_seq_putc(s, ')');
1635 
1636 	if (trace_probe_print_args(s, tp->args, tp->nr_args,
1637 			     (u8 *)&field[1], field) < 0)
1638 		goto out;
1639 
1640 	trace_seq_putc(s, '\n');
1641 
1642  out:
1643 	return trace_handle_return(s);
1644 }
1645 
1646 
1647 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1648 {
1649 	int ret;
1650 	struct kprobe_trace_entry_head field;
1651 	struct trace_probe *tp;
1652 
1653 	tp = trace_probe_primary_from_call(event_call);
1654 	if (WARN_ON_ONCE(!tp))
1655 		return -ENOENT;
1656 
1657 	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1658 
1659 	return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1660 }
1661 
1662 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1663 {
1664 	int ret;
1665 	struct kretprobe_trace_entry_head field;
1666 	struct trace_probe *tp;
1667 
1668 	tp = trace_probe_primary_from_call(event_call);
1669 	if (WARN_ON_ONCE(!tp))
1670 		return -ENOENT;
1671 
1672 	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1673 	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1674 
1675 	return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1676 }
1677 
1678 #ifdef CONFIG_PERF_EVENTS
1679 
1680 /* Kprobe profile handler */
1681 static int
1682 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1683 {
1684 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1685 	struct kprobe_trace_entry_head *entry;
1686 	struct hlist_head *head;
1687 	int size, __size, dsize;
1688 	int rctx;
1689 
1690 	if (bpf_prog_array_valid(call)) {
1691 		unsigned long orig_ip = instruction_pointer(regs);
1692 		int ret;
1693 
1694 		ret = trace_call_bpf(call, regs);
1695 
1696 		/*
1697 		 * We need to check and see if we modified the pc of the
1698 		 * pt_regs, and if so return 1 so that we don't do the
1699 		 * single stepping.
1700 		 */
1701 		if (orig_ip != instruction_pointer(regs))
1702 			return 1;
1703 		if (!ret)
1704 			return 0;
1705 	}
1706 
1707 	head = this_cpu_ptr(call->perf_events);
1708 	if (hlist_empty(head))
1709 		return 0;
1710 
1711 	dsize = __get_data_size(&tk->tp, regs, NULL);
1712 	__size = sizeof(*entry) + tk->tp.size + dsize;
1713 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1714 	size -= sizeof(u32);
1715 
1716 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1717 	if (!entry)
1718 		return 0;
1719 
1720 	entry->ip = (unsigned long)tk->rp.kp.addr;
1721 	memset(&entry[1], 0, dsize);
1722 	store_trace_args(&entry[1], &tk->tp, regs, NULL, sizeof(*entry), dsize);
1723 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1724 			      head, NULL);
1725 	return 0;
1726 }
1727 NOKPROBE_SYMBOL(kprobe_perf_func);
1728 
1729 /* Kretprobe profile handler */
1730 static void
1731 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1732 		    struct pt_regs *regs)
1733 {
1734 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1735 	struct kretprobe_trace_entry_head *entry;
1736 	struct hlist_head *head;
1737 	int size, __size, dsize;
1738 	int rctx;
1739 
1740 	if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1741 		return;
1742 
1743 	head = this_cpu_ptr(call->perf_events);
1744 	if (hlist_empty(head))
1745 		return;
1746 
1747 	dsize = __get_data_size(&tk->tp, regs, ri->data);
1748 	__size = sizeof(*entry) + tk->tp.size + dsize;
1749 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1750 	size -= sizeof(u32);
1751 
1752 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1753 	if (!entry)
1754 		return;
1755 
1756 	entry->func = (unsigned long)tk->rp.kp.addr;
1757 	entry->ret_ip = get_kretprobe_retaddr(ri);
1758 	store_trace_args(&entry[1], &tk->tp, regs, ri->data, sizeof(*entry), dsize);
1759 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1760 			      head, NULL);
1761 }
1762 NOKPROBE_SYMBOL(kretprobe_perf_func);
1763 
1764 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1765 			const char **symbol, u64 *probe_offset,
1766 			u64 *probe_addr, unsigned long *missed,
1767 			bool perf_type_tracepoint)
1768 {
1769 	const char *pevent = trace_event_name(event->tp_event);
1770 	const char *group = event->tp_event->class->system;
1771 	struct trace_kprobe *tk;
1772 
1773 	if (perf_type_tracepoint)
1774 		tk = find_trace_kprobe(pevent, group);
1775 	else
1776 		tk = trace_kprobe_primary_from_call(event->tp_event);
1777 	if (!tk)
1778 		return -EINVAL;
1779 
1780 	*fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1781 					      : BPF_FD_TYPE_KPROBE;
1782 	*probe_offset = tk->rp.kp.offset;
1783 	*probe_addr = kallsyms_show_value(current_cred()) ?
1784 		      (unsigned long)tk->rp.kp.addr : 0;
1785 	*symbol = tk->symbol;
1786 	if (missed)
1787 		*missed = trace_kprobe_missed(tk);
1788 	return 0;
1789 }
1790 #endif	/* CONFIG_PERF_EVENTS */
1791 
1792 /*
1793  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1794  *
1795  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1796  * lockless, but we can't race with this __init function.
1797  */
1798 static int kprobe_register(struct trace_event_call *event,
1799 			   enum trace_reg type, void *data)
1800 {
1801 	struct trace_event_file *file = data;
1802 
1803 	switch (type) {
1804 	case TRACE_REG_REGISTER:
1805 		return enable_trace_kprobe(event, file);
1806 	case TRACE_REG_UNREGISTER:
1807 		return disable_trace_kprobe(event, file);
1808 
1809 #ifdef CONFIG_PERF_EVENTS
1810 	case TRACE_REG_PERF_REGISTER:
1811 		return enable_trace_kprobe(event, NULL);
1812 	case TRACE_REG_PERF_UNREGISTER:
1813 		return disable_trace_kprobe(event, NULL);
1814 	case TRACE_REG_PERF_OPEN:
1815 	case TRACE_REG_PERF_CLOSE:
1816 	case TRACE_REG_PERF_ADD:
1817 	case TRACE_REG_PERF_DEL:
1818 		return 0;
1819 #endif
1820 	}
1821 	return 0;
1822 }
1823 
1824 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1825 {
1826 	struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1827 	unsigned int flags = trace_probe_load_flag(&tk->tp);
1828 	int ret = 0;
1829 
1830 	raw_cpu_inc(*tk->nhit);
1831 
1832 	if (flags & TP_FLAG_TRACE)
1833 		kprobe_trace_func(tk, regs);
1834 #ifdef CONFIG_PERF_EVENTS
1835 	if (flags & TP_FLAG_PROFILE)
1836 		ret = kprobe_perf_func(tk, regs);
1837 #endif
1838 	return ret;
1839 }
1840 NOKPROBE_SYMBOL(kprobe_dispatcher);
1841 
1842 static int
1843 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1844 {
1845 	struct kretprobe *rp = get_kretprobe(ri);
1846 	struct trace_kprobe *tk;
1847 	unsigned int flags;
1848 
1849 	/*
1850 	 * There is a small chance that get_kretprobe(ri) returns NULL when
1851 	 * the kretprobe is unregister on another CPU between kretprobe's
1852 	 * trampoline_handler and this function.
1853 	 */
1854 	if (unlikely(!rp))
1855 		return 0;
1856 
1857 	tk = container_of(rp, struct trace_kprobe, rp);
1858 	raw_cpu_inc(*tk->nhit);
1859 
1860 	flags = trace_probe_load_flag(&tk->tp);
1861 	if (flags & TP_FLAG_TRACE)
1862 		kretprobe_trace_func(tk, ri, regs);
1863 #ifdef CONFIG_PERF_EVENTS
1864 	if (flags & TP_FLAG_PROFILE)
1865 		kretprobe_perf_func(tk, ri, regs);
1866 #endif
1867 	return 0;	/* We don't tweak kernel, so just return 0 */
1868 }
1869 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1870 
1871 static struct trace_event_functions kretprobe_funcs = {
1872 	.trace		= print_kretprobe_event
1873 };
1874 
1875 static struct trace_event_functions kprobe_funcs = {
1876 	.trace		= print_kprobe_event
1877 };
1878 
1879 static struct trace_event_fields kretprobe_fields_array[] = {
1880 	{ .type = TRACE_FUNCTION_TYPE,
1881 	  .define_fields = kretprobe_event_define_fields },
1882 	{}
1883 };
1884 
1885 static struct trace_event_fields kprobe_fields_array[] = {
1886 	{ .type = TRACE_FUNCTION_TYPE,
1887 	  .define_fields = kprobe_event_define_fields },
1888 	{}
1889 };
1890 
1891 static inline void init_trace_event_call(struct trace_kprobe *tk)
1892 {
1893 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1894 
1895 	if (trace_kprobe_is_return(tk)) {
1896 		call->event.funcs = &kretprobe_funcs;
1897 		call->class->fields_array = kretprobe_fields_array;
1898 	} else {
1899 		call->event.funcs = &kprobe_funcs;
1900 		call->class->fields_array = kprobe_fields_array;
1901 	}
1902 
1903 	call->flags = TRACE_EVENT_FL_KPROBE;
1904 	call->class->reg = kprobe_register;
1905 }
1906 
1907 static int register_kprobe_event(struct trace_kprobe *tk)
1908 {
1909 	init_trace_event_call(tk);
1910 
1911 	return trace_probe_register_event_call(&tk->tp);
1912 }
1913 
1914 static int unregister_kprobe_event(struct trace_kprobe *tk)
1915 {
1916 	return trace_probe_unregister_event_call(&tk->tp);
1917 }
1918 
1919 #ifdef CONFIG_PERF_EVENTS
1920 
1921 /* create a trace_kprobe, but don't add it to global lists */
1922 struct trace_event_call *
1923 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1924 			  bool is_return)
1925 {
1926 	enum probe_print_type ptype;
1927 	struct trace_kprobe *tk __free(free_trace_kprobe) = NULL;
1928 	int ret;
1929 	char *event;
1930 
1931 	if (func) {
1932 		ret = validate_probe_symbol(func);
1933 		if (ret)
1934 			return ERR_PTR(ret);
1935 	}
1936 
1937 	/*
1938 	 * local trace_kprobes are not added to dyn_event, so they are never
1939 	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1940 	 * duplicated name here.
1941 	 */
1942 	event = func ? func : "DUMMY_EVENT";
1943 
1944 	tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1945 				offs, 0 /* maxactive */, 0 /* nargs */,
1946 				is_return);
1947 
1948 	if (IS_ERR(tk)) {
1949 		pr_info("Failed to allocate trace_probe.(%d)\n",
1950 			(int)PTR_ERR(tk));
1951 		return ERR_CAST(tk);
1952 	}
1953 
1954 	init_trace_event_call(tk);
1955 
1956 	ptype = trace_kprobe_is_return(tk) ?
1957 		PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1958 	if (traceprobe_set_print_fmt(&tk->tp, ptype) < 0)
1959 		return ERR_PTR(-ENOMEM);
1960 
1961 	ret = __register_trace_kprobe(tk);
1962 	if (ret < 0)
1963 		return ERR_PTR(ret);
1964 
1965 	return trace_probe_event_call(&(no_free_ptr(tk)->tp));
1966 }
1967 
1968 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1969 {
1970 	struct trace_kprobe *tk;
1971 
1972 	tk = trace_kprobe_primary_from_call(event_call);
1973 	if (unlikely(!tk))
1974 		return;
1975 
1976 	if (trace_probe_is_enabled(&tk->tp)) {
1977 		WARN_ON(1);
1978 		return;
1979 	}
1980 
1981 	__unregister_trace_kprobe(tk);
1982 
1983 	free_trace_kprobe(tk);
1984 }
1985 #endif /* CONFIG_PERF_EVENTS */
1986 
1987 static __init void enable_boot_kprobe_events(void)
1988 {
1989 	struct trace_array *tr = top_trace_array();
1990 	struct trace_event_file *file;
1991 	struct trace_kprobe *tk;
1992 	struct dyn_event *pos;
1993 
1994 	if (trace_kprobe_list_empty())
1995 		return;
1996 
1997 	guard(mutex)(&event_mutex);
1998 	for_each_trace_kprobe(tk, pos) {
1999 		list_for_each_entry(file, &tr->events, list)
2000 			if (file->event_call == trace_probe_event_call(&tk->tp))
2001 				trace_event_enable_disable(file, 1, 0);
2002 	}
2003 }
2004 
2005 static __init void setup_boot_kprobe_events(void)
2006 {
2007 	char *p, *cmd = kprobe_boot_events_buf;
2008 	int ret;
2009 
2010 	strreplace(kprobe_boot_events_buf, ',', ' ');
2011 
2012 	while (cmd && *cmd != '\0') {
2013 		p = strchr(cmd, ';');
2014 		if (p)
2015 			*p++ = '\0';
2016 
2017 		ret = create_or_delete_trace_kprobe(cmd);
2018 		if (ret)
2019 			pr_warn("Failed to add event(%d): %s\n", ret, cmd);
2020 
2021 		cmd = p;
2022 	}
2023 
2024 	enable_boot_kprobe_events();
2025 }
2026 
2027 /*
2028  * Register dynevent at core_initcall. This allows kernel to setup kprobe
2029  * events in postcore_initcall without tracefs.
2030  */
2031 static __init int init_kprobe_trace_early(void)
2032 {
2033 	int ret;
2034 
2035 	ret = dyn_event_register(&trace_kprobe_ops);
2036 	if (ret)
2037 		return ret;
2038 
2039 	if (trace_kprobe_register_module_notifier())
2040 		return -EINVAL;
2041 
2042 	return 0;
2043 }
2044 core_initcall(init_kprobe_trace_early);
2045 
2046 /* Make a tracefs interface for controlling probe points */
2047 static __init int init_kprobe_trace(void)
2048 {
2049 	int ret;
2050 
2051 	ret = tracing_init_dentry();
2052 	if (ret)
2053 		return 0;
2054 
2055 	/* Event list interface */
2056 	trace_create_file("kprobe_events", TRACE_MODE_WRITE,
2057 			  NULL, NULL, &kprobe_events_ops);
2058 
2059 	/* Profile interface */
2060 	trace_create_file("kprobe_profile", TRACE_MODE_READ,
2061 			  NULL, NULL, &kprobe_profile_ops);
2062 
2063 	/* If no 'kprobe_event=' cmd is provided, return directly. */
2064 	if (kprobe_boot_events_buf[0] == '\0')
2065 		return 0;
2066 
2067 	setup_boot_kprobe_events();
2068 
2069 	return 0;
2070 }
2071 fs_initcall(init_kprobe_trace);
2072 
2073 
2074 #ifdef CONFIG_FTRACE_STARTUP_TEST
2075 static __init struct trace_event_file *
2076 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
2077 {
2078 	struct trace_event_file *file;
2079 
2080 	list_for_each_entry(file, &tr->events, list)
2081 		if (file->event_call == trace_probe_event_call(&tk->tp))
2082 			return file;
2083 
2084 	return NULL;
2085 }
2086 
2087 /*
2088  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
2089  * stage, we can do this lockless.
2090  */
2091 static __init int kprobe_trace_self_tests_init(void)
2092 {
2093 	int ret, warn = 0;
2094 	int (*target)(int, int, int, int, int, int);
2095 	struct trace_kprobe *tk;
2096 	struct trace_event_file *file;
2097 
2098 	if (unlikely(tracing_disabled))
2099 		return -ENODEV;
2100 
2101 	if (tracing_selftest_disabled)
2102 		return 0;
2103 
2104 	target = kprobe_trace_selftest_target;
2105 
2106 	pr_info("Testing kprobe tracing: ");
2107 
2108 	ret = create_or_delete_trace_kprobe("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)");
2109 	if (WARN_ONCE(ret, "error on probing function entry.")) {
2110 		warn++;
2111 	} else {
2112 		/* Enable trace point */
2113 		tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2114 		if (WARN_ONCE(tk == NULL, "error on probing function entry.")) {
2115 			warn++;
2116 		} else {
2117 			file = find_trace_probe_file(tk, top_trace_array());
2118 			if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2119 				warn++;
2120 			} else
2121 				enable_trace_kprobe(
2122 					trace_probe_event_call(&tk->tp), file);
2123 		}
2124 	}
2125 
2126 	ret = create_or_delete_trace_kprobe("r:testprobe2 kprobe_trace_selftest_target $retval");
2127 	if (WARN_ONCE(ret, "error on probing function return.")) {
2128 		warn++;
2129 	} else {
2130 		/* Enable trace point */
2131 		tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2132 		if (WARN_ONCE(tk == NULL, "error on getting 2nd new probe.")) {
2133 			warn++;
2134 		} else {
2135 			file = find_trace_probe_file(tk, top_trace_array());
2136 			if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2137 				warn++;
2138 			} else
2139 				enable_trace_kprobe(
2140 					trace_probe_event_call(&tk->tp), file);
2141 		}
2142 	}
2143 
2144 	if (warn)
2145 		goto end;
2146 
2147 	ret = target(1, 2, 3, 4, 5, 6);
2148 
2149 	/*
2150 	 * Not expecting an error here, the check is only to prevent the
2151 	 * optimizer from removing the call to target() as otherwise there
2152 	 * are no side-effects and the call is never performed.
2153 	 */
2154 	if (ret != 21)
2155 		warn++;
2156 
2157 	/* Disable trace points before removing it */
2158 	tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2159 	if (WARN_ONCE(tk == NULL, "error on getting test probe.")) {
2160 		warn++;
2161 	} else {
2162 		if (WARN_ONCE(trace_kprobe_nhit(tk) != 1,
2163 				 "incorrect number of testprobe hits."))
2164 			warn++;
2165 
2166 		file = find_trace_probe_file(tk, top_trace_array());
2167 		if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2168 			warn++;
2169 		} else
2170 			disable_trace_kprobe(
2171 				trace_probe_event_call(&tk->tp), file);
2172 	}
2173 
2174 	tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2175 	if (WARN_ONCE(tk == NULL, "error on getting 2nd test probe.")) {
2176 		warn++;
2177 	} else {
2178 		if (WARN_ONCE(trace_kprobe_nhit(tk) != 1,
2179 				 "incorrect number of testprobe2 hits."))
2180 			warn++;
2181 
2182 		file = find_trace_probe_file(tk, top_trace_array());
2183 		if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2184 			warn++;
2185 		} else
2186 			disable_trace_kprobe(
2187 				trace_probe_event_call(&tk->tp), file);
2188 	}
2189 
2190 	ret = create_or_delete_trace_kprobe("-:testprobe");
2191 	if (WARN_ONCE(ret, "error on deleting a probe."))
2192 		warn++;
2193 
2194 	ret = create_or_delete_trace_kprobe("-:testprobe2");
2195 	if (WARN_ONCE(ret, "error on deleting a probe."))
2196 		warn++;
2197 
2198 
2199 end:
2200 	/*
2201 	 * Wait for the optimizer work to finish. Otherwise it might fiddle
2202 	 * with probes in already freed __init text.
2203 	 */
2204 	wait_for_kprobe_optimizer();
2205 	if (warn)
2206 		pr_cont("NG: Some tests are failed. Please check them.\n");
2207 	else
2208 		pr_cont("OK\n");
2209 	return 0;
2210 }
2211 
2212 late_initcall(kprobe_trace_self_tests_init);
2213 
2214 #endif
2215