xref: /linux/kernel/trace/trace_kprobe.c (revision a94549dd87f5ea4ca50fee493df08a2dc6256b53)
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/module.h>
11 #include <linux/uaccess.h>
12 #include <linux/rculist.h>
13 #include <linux/error-injection.h>
14 #include <linux/security.h>
15 
16 #include "trace_dynevent.h"
17 #include "trace_kprobe_selftest.h"
18 #include "trace_probe.h"
19 #include "trace_probe_tmpl.h"
20 
21 #define KPROBE_EVENT_SYSTEM "kprobes"
22 #define KRETPROBE_MAXACTIVE_MAX 4096
23 
24 static int trace_kprobe_create(int argc, const char **argv);
25 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
26 static int trace_kprobe_release(struct dyn_event *ev);
27 static bool trace_kprobe_is_busy(struct dyn_event *ev);
28 static bool trace_kprobe_match(const char *system, const char *event,
29 			       struct dyn_event *ev);
30 
31 static struct dyn_event_operations trace_kprobe_ops = {
32 	.create = trace_kprobe_create,
33 	.show = trace_kprobe_show,
34 	.is_busy = trace_kprobe_is_busy,
35 	.free = trace_kprobe_release,
36 	.match = trace_kprobe_match,
37 };
38 
39 /*
40  * Kprobe event core functions
41  */
42 struct trace_kprobe {
43 	struct dyn_event	devent;
44 	struct kretprobe	rp;	/* Use rp.kp for kprobe use */
45 	unsigned long __percpu *nhit;
46 	const char		*symbol;	/* symbol name */
47 	struct trace_probe	tp;
48 };
49 
50 static bool is_trace_kprobe(struct dyn_event *ev)
51 {
52 	return ev->ops == &trace_kprobe_ops;
53 }
54 
55 static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
56 {
57 	return container_of(ev, struct trace_kprobe, devent);
58 }
59 
60 /**
61  * for_each_trace_kprobe - iterate over the trace_kprobe list
62  * @pos:	the struct trace_kprobe * for each entry
63  * @dpos:	the struct dyn_event * to use as a loop cursor
64  */
65 #define for_each_trace_kprobe(pos, dpos)	\
66 	for_each_dyn_event(dpos)		\
67 		if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
68 
69 #define SIZEOF_TRACE_KPROBE(n)				\
70 	(offsetof(struct trace_kprobe, tp.args) +	\
71 	(sizeof(struct probe_arg) * (n)))
72 
73 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
74 {
75 	return tk->rp.handler != NULL;
76 }
77 
78 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
79 {
80 	return tk->symbol ? tk->symbol : "unknown";
81 }
82 
83 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
84 {
85 	return tk->rp.kp.offset;
86 }
87 
88 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
89 {
90 	return !!(kprobe_gone(&tk->rp.kp));
91 }
92 
93 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
94 						 struct module *mod)
95 {
96 	int len = strlen(mod->name);
97 	const char *name = trace_kprobe_symbol(tk);
98 	return strncmp(mod->name, name, len) == 0 && name[len] == ':';
99 }
100 
101 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
102 {
103 	char *p;
104 	bool ret;
105 
106 	if (!tk->symbol)
107 		return false;
108 	p = strchr(tk->symbol, ':');
109 	if (!p)
110 		return true;
111 	*p = '\0';
112 	mutex_lock(&module_mutex);
113 	ret = !!find_module(tk->symbol);
114 	mutex_unlock(&module_mutex);
115 	*p = ':';
116 
117 	return ret;
118 }
119 
120 static bool trace_kprobe_is_busy(struct dyn_event *ev)
121 {
122 	struct trace_kprobe *tk = to_trace_kprobe(ev);
123 
124 	return trace_probe_is_enabled(&tk->tp);
125 }
126 
127 static bool trace_kprobe_match(const char *system, const char *event,
128 			       struct dyn_event *ev)
129 {
130 	struct trace_kprobe *tk = to_trace_kprobe(ev);
131 
132 	return strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
133 	    (!system || strcmp(tk->tp.call.class->system, system) == 0);
134 }
135 
136 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
137 {
138 	unsigned long nhit = 0;
139 	int cpu;
140 
141 	for_each_possible_cpu(cpu)
142 		nhit += *per_cpu_ptr(tk->nhit, cpu);
143 
144 	return nhit;
145 }
146 
147 /* Return 0 if it fails to find the symbol address */
148 static nokprobe_inline
149 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
150 {
151 	unsigned long addr;
152 
153 	if (tk->symbol) {
154 		addr = (unsigned long)
155 			kallsyms_lookup_name(trace_kprobe_symbol(tk));
156 		if (addr)
157 			addr += tk->rp.kp.offset;
158 	} else {
159 		addr = (unsigned long)tk->rp.kp.addr;
160 	}
161 	return addr;
162 }
163 
164 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
165 {
166 	struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
167 
168 	return kprobe_on_func_entry(tk->rp.kp.addr,
169 			tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
170 			tk->rp.kp.addr ? 0 : tk->rp.kp.offset);
171 }
172 
173 bool trace_kprobe_error_injectable(struct trace_event_call *call)
174 {
175 	struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
176 
177 	return within_error_injection_list(trace_kprobe_address(tk));
178 }
179 
180 static int register_kprobe_event(struct trace_kprobe *tk);
181 static int unregister_kprobe_event(struct trace_kprobe *tk);
182 
183 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
184 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
185 				struct pt_regs *regs);
186 
187 /*
188  * Allocate new trace_probe and initialize it (including kprobes).
189  */
190 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
191 					     const char *event,
192 					     void *addr,
193 					     const char *symbol,
194 					     unsigned long offs,
195 					     int maxactive,
196 					     int nargs, bool is_return)
197 {
198 	struct trace_kprobe *tk;
199 	int ret = -ENOMEM;
200 
201 	tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
202 	if (!tk)
203 		return ERR_PTR(ret);
204 
205 	tk->nhit = alloc_percpu(unsigned long);
206 	if (!tk->nhit)
207 		goto error;
208 
209 	if (symbol) {
210 		tk->symbol = kstrdup(symbol, GFP_KERNEL);
211 		if (!tk->symbol)
212 			goto error;
213 		tk->rp.kp.symbol_name = tk->symbol;
214 		tk->rp.kp.offset = offs;
215 	} else
216 		tk->rp.kp.addr = addr;
217 
218 	if (is_return)
219 		tk->rp.handler = kretprobe_dispatcher;
220 	else
221 		tk->rp.kp.pre_handler = kprobe_dispatcher;
222 
223 	tk->rp.maxactive = maxactive;
224 
225 	if (!event || !group) {
226 		ret = -EINVAL;
227 		goto error;
228 	}
229 
230 	tk->tp.call.class = &tk->tp.class;
231 	tk->tp.call.name = kstrdup(event, GFP_KERNEL);
232 	if (!tk->tp.call.name)
233 		goto error;
234 
235 	tk->tp.class.system = kstrdup(group, GFP_KERNEL);
236 	if (!tk->tp.class.system)
237 		goto error;
238 
239 	dyn_event_init(&tk->devent, &trace_kprobe_ops);
240 	INIT_LIST_HEAD(&tk->tp.files);
241 	return tk;
242 error:
243 	kfree(tk->tp.call.name);
244 	kfree(tk->symbol);
245 	free_percpu(tk->nhit);
246 	kfree(tk);
247 	return ERR_PTR(ret);
248 }
249 
250 static void free_trace_kprobe(struct trace_kprobe *tk)
251 {
252 	int i;
253 
254 	if (!tk)
255 		return;
256 
257 	for (i = 0; i < tk->tp.nr_args; i++)
258 		traceprobe_free_probe_arg(&tk->tp.args[i]);
259 
260 	kfree(tk->tp.call.class->system);
261 	kfree(tk->tp.call.name);
262 	kfree(tk->symbol);
263 	free_percpu(tk->nhit);
264 	kfree(tk);
265 }
266 
267 static struct trace_kprobe *find_trace_kprobe(const char *event,
268 					      const char *group)
269 {
270 	struct dyn_event *pos;
271 	struct trace_kprobe *tk;
272 
273 	for_each_trace_kprobe(tk, pos)
274 		if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
275 		    strcmp(tk->tp.call.class->system, group) == 0)
276 			return tk;
277 	return NULL;
278 }
279 
280 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
281 {
282 	int ret = 0;
283 
284 	if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
285 		if (trace_kprobe_is_return(tk))
286 			ret = enable_kretprobe(&tk->rp);
287 		else
288 			ret = enable_kprobe(&tk->rp.kp);
289 	}
290 
291 	return ret;
292 }
293 
294 /*
295  * Enable trace_probe
296  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
297  */
298 static int
299 enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
300 {
301 	struct event_file_link *link;
302 	int ret = 0;
303 
304 	if (file) {
305 		link = kmalloc(sizeof(*link), GFP_KERNEL);
306 		if (!link) {
307 			ret = -ENOMEM;
308 			goto out;
309 		}
310 
311 		link->file = file;
312 		list_add_tail_rcu(&link->list, &tk->tp.files);
313 
314 		tk->tp.flags |= TP_FLAG_TRACE;
315 		ret = __enable_trace_kprobe(tk);
316 		if (ret) {
317 			list_del_rcu(&link->list);
318 			kfree(link);
319 			tk->tp.flags &= ~TP_FLAG_TRACE;
320 		}
321 
322 	} else {
323 		tk->tp.flags |= TP_FLAG_PROFILE;
324 		ret = __enable_trace_kprobe(tk);
325 		if (ret)
326 			tk->tp.flags &= ~TP_FLAG_PROFILE;
327 	}
328  out:
329 	return ret;
330 }
331 
332 /*
333  * Disable trace_probe
334  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
335  */
336 static int
337 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
338 {
339 	struct event_file_link *link = NULL;
340 	int wait = 0;
341 	int ret = 0;
342 
343 	if (file) {
344 		link = find_event_file_link(&tk->tp, file);
345 		if (!link) {
346 			ret = -EINVAL;
347 			goto out;
348 		}
349 
350 		list_del_rcu(&link->list);
351 		wait = 1;
352 		if (!list_empty(&tk->tp.files))
353 			goto out;
354 
355 		tk->tp.flags &= ~TP_FLAG_TRACE;
356 	} else
357 		tk->tp.flags &= ~TP_FLAG_PROFILE;
358 
359 	if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
360 		if (trace_kprobe_is_return(tk))
361 			disable_kretprobe(&tk->rp);
362 		else
363 			disable_kprobe(&tk->rp.kp);
364 		wait = 1;
365 	}
366 
367 	/*
368 	 * if tk is not added to any list, it must be a local trace_kprobe
369 	 * created with perf_event_open. We don't need to wait for these
370 	 * trace_kprobes
371 	 */
372 	if (list_empty(&tk->devent.list))
373 		wait = 0;
374  out:
375 	if (wait) {
376 		/*
377 		 * Synchronize with kprobe_trace_func/kretprobe_trace_func
378 		 * to ensure disabled (all running handlers are finished).
379 		 * This is not only for kfree(), but also the caller,
380 		 * trace_remove_event_call() supposes it for releasing
381 		 * event_call related objects, which will be accessed in
382 		 * the kprobe_trace_func/kretprobe_trace_func.
383 		 */
384 		synchronize_rcu();
385 		kfree(link);	/* Ignored if link == NULL */
386 	}
387 
388 	return ret;
389 }
390 
391 #if defined(CONFIG_KPROBES_ON_FTRACE) && \
392 	!defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
393 static bool within_notrace_func(struct trace_kprobe *tk)
394 {
395 	unsigned long offset, size, addr;
396 
397 	addr = trace_kprobe_address(tk);
398 	if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
399 		return false;
400 
401 	/* Get the entry address of the target function */
402 	addr -= offset;
403 
404 	/*
405 	 * Since ftrace_location_range() does inclusive range check, we need
406 	 * to subtract 1 byte from the end address.
407 	 */
408 	return !ftrace_location_range(addr, addr + size - 1);
409 }
410 #else
411 #define within_notrace_func(tk)	(false)
412 #endif
413 
414 /* Internal register function - just handle k*probes and flags */
415 static int __register_trace_kprobe(struct trace_kprobe *tk)
416 {
417 	int i, ret;
418 
419 	ret = security_locked_down(LOCKDOWN_KPROBES);
420 	if (ret)
421 		return ret;
422 
423 	if (trace_probe_is_registered(&tk->tp))
424 		return -EINVAL;
425 
426 	if (within_notrace_func(tk)) {
427 		pr_warn("Could not probe notrace function %s\n",
428 			trace_kprobe_symbol(tk));
429 		return -EINVAL;
430 	}
431 
432 	for (i = 0; i < tk->tp.nr_args; i++) {
433 		ret = traceprobe_update_arg(&tk->tp.args[i]);
434 		if (ret)
435 			return ret;
436 	}
437 
438 	/* Set/clear disabled flag according to tp->flag */
439 	if (trace_probe_is_enabled(&tk->tp))
440 		tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
441 	else
442 		tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
443 
444 	if (trace_kprobe_is_return(tk))
445 		ret = register_kretprobe(&tk->rp);
446 	else
447 		ret = register_kprobe(&tk->rp.kp);
448 
449 	if (ret == 0)
450 		tk->tp.flags |= TP_FLAG_REGISTERED;
451 	return ret;
452 }
453 
454 /* Internal unregister function - just handle k*probes and flags */
455 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
456 {
457 	if (trace_probe_is_registered(&tk->tp)) {
458 		if (trace_kprobe_is_return(tk))
459 			unregister_kretprobe(&tk->rp);
460 		else
461 			unregister_kprobe(&tk->rp.kp);
462 		tk->tp.flags &= ~TP_FLAG_REGISTERED;
463 		/* Cleanup kprobe for reuse */
464 		if (tk->rp.kp.symbol_name)
465 			tk->rp.kp.addr = NULL;
466 	}
467 }
468 
469 /* Unregister a trace_probe and probe_event */
470 static int unregister_trace_kprobe(struct trace_kprobe *tk)
471 {
472 	/* Enabled event can not be unregistered */
473 	if (trace_probe_is_enabled(&tk->tp))
474 		return -EBUSY;
475 
476 	/* Will fail if probe is being used by ftrace or perf */
477 	if (unregister_kprobe_event(tk))
478 		return -EBUSY;
479 
480 	__unregister_trace_kprobe(tk);
481 	dyn_event_remove(&tk->devent);
482 
483 	return 0;
484 }
485 
486 /* Register a trace_probe and probe_event */
487 static int register_trace_kprobe(struct trace_kprobe *tk)
488 {
489 	struct trace_kprobe *old_tk;
490 	int ret;
491 
492 	mutex_lock(&event_mutex);
493 
494 	/* Delete old (same name) event if exist */
495 	old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
496 			tk->tp.call.class->system);
497 	if (old_tk) {
498 		ret = unregister_trace_kprobe(old_tk);
499 		if (ret < 0)
500 			goto end;
501 		free_trace_kprobe(old_tk);
502 	}
503 
504 	/* Register new event */
505 	ret = register_kprobe_event(tk);
506 	if (ret) {
507 		pr_warn("Failed to register probe event(%d)\n", ret);
508 		goto end;
509 	}
510 
511 	/* Register k*probe */
512 	ret = __register_trace_kprobe(tk);
513 	if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
514 		pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
515 		ret = 0;
516 	}
517 
518 	if (ret < 0)
519 		unregister_kprobe_event(tk);
520 	else
521 		dyn_event_add(&tk->devent);
522 
523 end:
524 	mutex_unlock(&event_mutex);
525 	return ret;
526 }
527 
528 /* Module notifier call back, checking event on the module */
529 static int trace_kprobe_module_callback(struct notifier_block *nb,
530 				       unsigned long val, void *data)
531 {
532 	struct module *mod = data;
533 	struct dyn_event *pos;
534 	struct trace_kprobe *tk;
535 	int ret;
536 
537 	if (val != MODULE_STATE_COMING)
538 		return NOTIFY_DONE;
539 
540 	/* Update probes on coming module */
541 	mutex_lock(&event_mutex);
542 	for_each_trace_kprobe(tk, pos) {
543 		if (trace_kprobe_within_module(tk, mod)) {
544 			/* Don't need to check busy - this should have gone. */
545 			__unregister_trace_kprobe(tk);
546 			ret = __register_trace_kprobe(tk);
547 			if (ret)
548 				pr_warn("Failed to re-register probe %s on %s: %d\n",
549 					trace_event_name(&tk->tp.call),
550 					mod->name, ret);
551 		}
552 	}
553 	mutex_unlock(&event_mutex);
554 
555 	return NOTIFY_DONE;
556 }
557 
558 static struct notifier_block trace_kprobe_module_nb = {
559 	.notifier_call = trace_kprobe_module_callback,
560 	.priority = 1	/* Invoked after kprobe module callback */
561 };
562 
563 /* Convert certain expected symbols into '_' when generating event names */
564 static inline void sanitize_event_name(char *name)
565 {
566 	while (*name++ != '\0')
567 		if (*name == ':' || *name == '.')
568 			*name = '_';
569 }
570 
571 static int trace_kprobe_create(int argc, const char *argv[])
572 {
573 	/*
574 	 * Argument syntax:
575 	 *  - Add kprobe:
576 	 *      p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
577 	 *  - Add kretprobe:
578 	 *      r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
579 	 * Fetch args:
580 	 *  $retval	: fetch return value
581 	 *  $stack	: fetch stack address
582 	 *  $stackN	: fetch Nth of stack (N:0-)
583 	 *  $comm       : fetch current task comm
584 	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
585 	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
586 	 *  %REG	: fetch register REG
587 	 * Dereferencing memory fetch:
588 	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
589 	 * Alias name of args:
590 	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
591 	 * Type of args:
592 	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
593 	 */
594 	struct trace_kprobe *tk = NULL;
595 	int i, len, ret = 0;
596 	bool is_return = false;
597 	char *symbol = NULL, *tmp = NULL;
598 	const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
599 	int maxactive = 0;
600 	long offset = 0;
601 	void *addr = NULL;
602 	char buf[MAX_EVENT_NAME_LEN];
603 	unsigned int flags = TPARG_FL_KERNEL;
604 
605 	switch (argv[0][0]) {
606 	case 'r':
607 		is_return = true;
608 		flags |= TPARG_FL_RETURN;
609 		break;
610 	case 'p':
611 		break;
612 	default:
613 		return -ECANCELED;
614 	}
615 	if (argc < 2)
616 		return -ECANCELED;
617 
618 	trace_probe_log_init("trace_kprobe", argc, argv);
619 
620 	event = strchr(&argv[0][1], ':');
621 	if (event)
622 		event++;
623 
624 	if (isdigit(argv[0][1])) {
625 		if (!is_return) {
626 			trace_probe_log_err(1, MAXACT_NO_KPROBE);
627 			goto parse_error;
628 		}
629 		if (event)
630 			len = event - &argv[0][1] - 1;
631 		else
632 			len = strlen(&argv[0][1]);
633 		if (len > MAX_EVENT_NAME_LEN - 1) {
634 			trace_probe_log_err(1, BAD_MAXACT);
635 			goto parse_error;
636 		}
637 		memcpy(buf, &argv[0][1], len);
638 		buf[len] = '\0';
639 		ret = kstrtouint(buf, 0, &maxactive);
640 		if (ret || !maxactive) {
641 			trace_probe_log_err(1, BAD_MAXACT);
642 			goto parse_error;
643 		}
644 		/* kretprobes instances are iterated over via a list. The
645 		 * maximum should stay reasonable.
646 		 */
647 		if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
648 			trace_probe_log_err(1, MAXACT_TOO_BIG);
649 			goto parse_error;
650 		}
651 	}
652 
653 	/* try to parse an address. if that fails, try to read the
654 	 * input as a symbol. */
655 	if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
656 		trace_probe_log_set_index(1);
657 		/* Check whether uprobe event specified */
658 		if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
659 			ret = -ECANCELED;
660 			goto error;
661 		}
662 		/* a symbol specified */
663 		symbol = kstrdup(argv[1], GFP_KERNEL);
664 		if (!symbol)
665 			return -ENOMEM;
666 		/* TODO: support .init module functions */
667 		ret = traceprobe_split_symbol_offset(symbol, &offset);
668 		if (ret || offset < 0 || offset > UINT_MAX) {
669 			trace_probe_log_err(0, BAD_PROBE_ADDR);
670 			goto parse_error;
671 		}
672 		if (kprobe_on_func_entry(NULL, symbol, offset))
673 			flags |= TPARG_FL_FENTRY;
674 		if (offset && is_return && !(flags & TPARG_FL_FENTRY)) {
675 			trace_probe_log_err(0, BAD_RETPROBE);
676 			goto parse_error;
677 		}
678 	}
679 
680 	trace_probe_log_set_index(0);
681 	if (event) {
682 		ret = traceprobe_parse_event_name(&event, &group, buf,
683 						  event - argv[0]);
684 		if (ret)
685 			goto parse_error;
686 	} else {
687 		/* Make a new event name */
688 		if (symbol)
689 			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
690 				 is_return ? 'r' : 'p', symbol, offset);
691 		else
692 			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
693 				 is_return ? 'r' : 'p', addr);
694 		sanitize_event_name(buf);
695 		event = buf;
696 	}
697 
698 	/* setup a probe */
699 	tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
700 			       argc - 2, is_return);
701 	if (IS_ERR(tk)) {
702 		ret = PTR_ERR(tk);
703 		/* This must return -ENOMEM, else there is a bug */
704 		WARN_ON_ONCE(ret != -ENOMEM);
705 		goto out;	/* We know tk is not allocated */
706 	}
707 	argc -= 2; argv += 2;
708 
709 	/* parse arguments */
710 	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
711 		tmp = kstrdup(argv[i], GFP_KERNEL);
712 		if (!tmp) {
713 			ret = -ENOMEM;
714 			goto error;
715 		}
716 
717 		trace_probe_log_set_index(i + 2);
718 		ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
719 		kfree(tmp);
720 		if (ret)
721 			goto error;	/* This can be -ENOMEM */
722 	}
723 
724 	ret = register_trace_kprobe(tk);
725 	if (ret) {
726 		trace_probe_log_set_index(1);
727 		if (ret == -EILSEQ)
728 			trace_probe_log_err(0, BAD_INSN_BNDRY);
729 		else if (ret == -ENOENT)
730 			trace_probe_log_err(0, BAD_PROBE_ADDR);
731 		else if (ret != -ENOMEM)
732 			trace_probe_log_err(0, FAIL_REG_PROBE);
733 		goto error;
734 	}
735 
736 out:
737 	trace_probe_log_clear();
738 	kfree(symbol);
739 	return ret;
740 
741 parse_error:
742 	ret = -EINVAL;
743 error:
744 	free_trace_kprobe(tk);
745 	goto out;
746 }
747 
748 static int create_or_delete_trace_kprobe(int argc, char **argv)
749 {
750 	int ret;
751 
752 	if (argv[0][0] == '-')
753 		return dyn_event_release(argc, argv, &trace_kprobe_ops);
754 
755 	ret = trace_kprobe_create(argc, (const char **)argv);
756 	return ret == -ECANCELED ? -EINVAL : ret;
757 }
758 
759 static int trace_kprobe_release(struct dyn_event *ev)
760 {
761 	struct trace_kprobe *tk = to_trace_kprobe(ev);
762 	int ret = unregister_trace_kprobe(tk);
763 
764 	if (!ret)
765 		free_trace_kprobe(tk);
766 	return ret;
767 }
768 
769 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
770 {
771 	struct trace_kprobe *tk = to_trace_kprobe(ev);
772 	int i;
773 
774 	seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
775 	seq_printf(m, ":%s/%s", tk->tp.call.class->system,
776 			trace_event_name(&tk->tp.call));
777 
778 	if (!tk->symbol)
779 		seq_printf(m, " 0x%p", tk->rp.kp.addr);
780 	else if (tk->rp.kp.offset)
781 		seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
782 			   tk->rp.kp.offset);
783 	else
784 		seq_printf(m, " %s", trace_kprobe_symbol(tk));
785 
786 	for (i = 0; i < tk->tp.nr_args; i++)
787 		seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
788 	seq_putc(m, '\n');
789 
790 	return 0;
791 }
792 
793 static int probes_seq_show(struct seq_file *m, void *v)
794 {
795 	struct dyn_event *ev = v;
796 
797 	if (!is_trace_kprobe(ev))
798 		return 0;
799 
800 	return trace_kprobe_show(m, ev);
801 }
802 
803 static const struct seq_operations probes_seq_op = {
804 	.start  = dyn_event_seq_start,
805 	.next   = dyn_event_seq_next,
806 	.stop   = dyn_event_seq_stop,
807 	.show   = probes_seq_show
808 };
809 
810 static int probes_open(struct inode *inode, struct file *file)
811 {
812 	int ret;
813 
814 	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
815 		ret = dyn_events_release_all(&trace_kprobe_ops);
816 		if (ret < 0)
817 			return ret;
818 	}
819 
820 	return seq_open(file, &probes_seq_op);
821 }
822 
823 static ssize_t probes_write(struct file *file, const char __user *buffer,
824 			    size_t count, loff_t *ppos)
825 {
826 	return trace_parse_run_command(file, buffer, count, ppos,
827 				       create_or_delete_trace_kprobe);
828 }
829 
830 static const struct file_operations kprobe_events_ops = {
831 	.owner          = THIS_MODULE,
832 	.open           = probes_open,
833 	.read           = seq_read,
834 	.llseek         = seq_lseek,
835 	.release        = seq_release,
836 	.write		= probes_write,
837 };
838 
839 /* Probes profiling interfaces */
840 static int probes_profile_seq_show(struct seq_file *m, void *v)
841 {
842 	struct dyn_event *ev = v;
843 	struct trace_kprobe *tk;
844 
845 	if (!is_trace_kprobe(ev))
846 		return 0;
847 
848 	tk = to_trace_kprobe(ev);
849 	seq_printf(m, "  %-44s %15lu %15lu\n",
850 		   trace_event_name(&tk->tp.call),
851 		   trace_kprobe_nhit(tk),
852 		   tk->rp.kp.nmissed);
853 
854 	return 0;
855 }
856 
857 static const struct seq_operations profile_seq_op = {
858 	.start  = dyn_event_seq_start,
859 	.next   = dyn_event_seq_next,
860 	.stop   = dyn_event_seq_stop,
861 	.show   = probes_profile_seq_show
862 };
863 
864 static int profile_open(struct inode *inode, struct file *file)
865 {
866 	return seq_open(file, &profile_seq_op);
867 }
868 
869 static const struct file_operations kprobe_profile_ops = {
870 	.owner          = THIS_MODULE,
871 	.open           = profile_open,
872 	.read           = seq_read,
873 	.llseek         = seq_lseek,
874 	.release        = seq_release,
875 };
876 
877 /* Kprobe specific fetch functions */
878 
879 /* Return the length of string -- including null terminal byte */
880 static nokprobe_inline int
881 fetch_store_strlen(unsigned long addr)
882 {
883 	int ret, len = 0;
884 	u8 c;
885 
886 	do {
887 		ret = probe_kernel_read(&c, (u8 *)addr + len, 1);
888 		len++;
889 	} while (c && ret == 0 && len < MAX_STRING_SIZE);
890 
891 	return (ret < 0) ? ret : len;
892 }
893 
894 /*
895  * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
896  * length and relative data location.
897  */
898 static nokprobe_inline int
899 fetch_store_string(unsigned long addr, void *dest, void *base)
900 {
901 	int maxlen = get_loc_len(*(u32 *)dest);
902 	u8 *dst = get_loc_data(dest, base);
903 	long ret;
904 
905 	if (unlikely(!maxlen))
906 		return -ENOMEM;
907 	/*
908 	 * Try to get string again, since the string can be changed while
909 	 * probing.
910 	 */
911 	ret = strncpy_from_unsafe(dst, (void *)addr, maxlen);
912 
913 	if (ret >= 0)
914 		*(u32 *)dest = make_data_loc(ret, (void *)dst - base);
915 	return ret;
916 }
917 
918 static nokprobe_inline int
919 probe_mem_read(void *dest, void *src, size_t size)
920 {
921 	return probe_kernel_read(dest, src, size);
922 }
923 
924 /* Note that we don't verify it, since the code does not come from user space */
925 static int
926 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
927 		   void *base)
928 {
929 	unsigned long val;
930 
931 retry:
932 	/* 1st stage: get value from context */
933 	switch (code->op) {
934 	case FETCH_OP_REG:
935 		val = regs_get_register(regs, code->param);
936 		break;
937 	case FETCH_OP_STACK:
938 		val = regs_get_kernel_stack_nth(regs, code->param);
939 		break;
940 	case FETCH_OP_STACKP:
941 		val = kernel_stack_pointer(regs);
942 		break;
943 	case FETCH_OP_RETVAL:
944 		val = regs_return_value(regs);
945 		break;
946 	case FETCH_OP_IMM:
947 		val = code->immediate;
948 		break;
949 	case FETCH_OP_COMM:
950 		val = (unsigned long)current->comm;
951 		break;
952 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
953 	case FETCH_OP_ARG:
954 		val = regs_get_kernel_argument(regs, code->param);
955 		break;
956 #endif
957 	case FETCH_NOP_SYMBOL:	/* Ignore a place holder */
958 		code++;
959 		goto retry;
960 	default:
961 		return -EILSEQ;
962 	}
963 	code++;
964 
965 	return process_fetch_insn_bottom(code, val, dest, base);
966 }
967 NOKPROBE_SYMBOL(process_fetch_insn)
968 
969 /* Kprobe handler */
970 static nokprobe_inline void
971 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
972 		    struct trace_event_file *trace_file)
973 {
974 	struct kprobe_trace_entry_head *entry;
975 	struct ring_buffer_event *event;
976 	struct ring_buffer *buffer;
977 	int size, dsize, pc;
978 	unsigned long irq_flags;
979 	struct trace_event_call *call = &tk->tp.call;
980 
981 	WARN_ON(call != trace_file->event_call);
982 
983 	if (trace_trigger_soft_disabled(trace_file))
984 		return;
985 
986 	local_save_flags(irq_flags);
987 	pc = preempt_count();
988 
989 	dsize = __get_data_size(&tk->tp, regs);
990 	size = sizeof(*entry) + tk->tp.size + dsize;
991 
992 	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
993 						call->event.type,
994 						size, irq_flags, pc);
995 	if (!event)
996 		return;
997 
998 	entry = ring_buffer_event_data(event);
999 	entry->ip = (unsigned long)tk->rp.kp.addr;
1000 	store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1001 
1002 	event_trigger_unlock_commit_regs(trace_file, buffer, event,
1003 					 entry, irq_flags, pc, regs);
1004 }
1005 
1006 static void
1007 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1008 {
1009 	struct event_file_link *link;
1010 
1011 	list_for_each_entry_rcu(link, &tk->tp.files, list)
1012 		__kprobe_trace_func(tk, regs, link->file);
1013 }
1014 NOKPROBE_SYMBOL(kprobe_trace_func);
1015 
1016 /* Kretprobe handler */
1017 static nokprobe_inline void
1018 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1019 		       struct pt_regs *regs,
1020 		       struct trace_event_file *trace_file)
1021 {
1022 	struct kretprobe_trace_entry_head *entry;
1023 	struct ring_buffer_event *event;
1024 	struct ring_buffer *buffer;
1025 	int size, pc, dsize;
1026 	unsigned long irq_flags;
1027 	struct trace_event_call *call = &tk->tp.call;
1028 
1029 	WARN_ON(call != trace_file->event_call);
1030 
1031 	if (trace_trigger_soft_disabled(trace_file))
1032 		return;
1033 
1034 	local_save_flags(irq_flags);
1035 	pc = preempt_count();
1036 
1037 	dsize = __get_data_size(&tk->tp, regs);
1038 	size = sizeof(*entry) + tk->tp.size + dsize;
1039 
1040 	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1041 						call->event.type,
1042 						size, irq_flags, pc);
1043 	if (!event)
1044 		return;
1045 
1046 	entry = ring_buffer_event_data(event);
1047 	entry->func = (unsigned long)tk->rp.kp.addr;
1048 	entry->ret_ip = (unsigned long)ri->ret_addr;
1049 	store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1050 
1051 	event_trigger_unlock_commit_regs(trace_file, buffer, event,
1052 					 entry, irq_flags, pc, regs);
1053 }
1054 
1055 static void
1056 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1057 		     struct pt_regs *regs)
1058 {
1059 	struct event_file_link *link;
1060 
1061 	list_for_each_entry_rcu(link, &tk->tp.files, list)
1062 		__kretprobe_trace_func(tk, ri, regs, link->file);
1063 }
1064 NOKPROBE_SYMBOL(kretprobe_trace_func);
1065 
1066 /* Event entry printers */
1067 static enum print_line_t
1068 print_kprobe_event(struct trace_iterator *iter, int flags,
1069 		   struct trace_event *event)
1070 {
1071 	struct kprobe_trace_entry_head *field;
1072 	struct trace_seq *s = &iter->seq;
1073 	struct trace_probe *tp;
1074 
1075 	field = (struct kprobe_trace_entry_head *)iter->ent;
1076 	tp = container_of(event, struct trace_probe, call.event);
1077 
1078 	trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1079 
1080 	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1081 		goto out;
1082 
1083 	trace_seq_putc(s, ')');
1084 
1085 	if (print_probe_args(s, tp->args, tp->nr_args,
1086 			     (u8 *)&field[1], field) < 0)
1087 		goto out;
1088 
1089 	trace_seq_putc(s, '\n');
1090  out:
1091 	return trace_handle_return(s);
1092 }
1093 
1094 static enum print_line_t
1095 print_kretprobe_event(struct trace_iterator *iter, int flags,
1096 		      struct trace_event *event)
1097 {
1098 	struct kretprobe_trace_entry_head *field;
1099 	struct trace_seq *s = &iter->seq;
1100 	struct trace_probe *tp;
1101 
1102 	field = (struct kretprobe_trace_entry_head *)iter->ent;
1103 	tp = container_of(event, struct trace_probe, call.event);
1104 
1105 	trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1106 
1107 	if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1108 		goto out;
1109 
1110 	trace_seq_puts(s, " <- ");
1111 
1112 	if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1113 		goto out;
1114 
1115 	trace_seq_putc(s, ')');
1116 
1117 	if (print_probe_args(s, tp->args, tp->nr_args,
1118 			     (u8 *)&field[1], field) < 0)
1119 		goto out;
1120 
1121 	trace_seq_putc(s, '\n');
1122 
1123  out:
1124 	return trace_handle_return(s);
1125 }
1126 
1127 
1128 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1129 {
1130 	int ret;
1131 	struct kprobe_trace_entry_head field;
1132 	struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1133 
1134 	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1135 
1136 	return traceprobe_define_arg_fields(event_call, sizeof(field), &tk->tp);
1137 }
1138 
1139 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1140 {
1141 	int ret;
1142 	struct kretprobe_trace_entry_head field;
1143 	struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1144 
1145 	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1146 	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1147 
1148 	return traceprobe_define_arg_fields(event_call, sizeof(field), &tk->tp);
1149 }
1150 
1151 #ifdef CONFIG_PERF_EVENTS
1152 
1153 /* Kprobe profile handler */
1154 static int
1155 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1156 {
1157 	struct trace_event_call *call = &tk->tp.call;
1158 	struct kprobe_trace_entry_head *entry;
1159 	struct hlist_head *head;
1160 	int size, __size, dsize;
1161 	int rctx;
1162 
1163 	if (bpf_prog_array_valid(call)) {
1164 		unsigned long orig_ip = instruction_pointer(regs);
1165 		int ret;
1166 
1167 		ret = trace_call_bpf(call, regs);
1168 
1169 		/*
1170 		 * We need to check and see if we modified the pc of the
1171 		 * pt_regs, and if so return 1 so that we don't do the
1172 		 * single stepping.
1173 		 */
1174 		if (orig_ip != instruction_pointer(regs))
1175 			return 1;
1176 		if (!ret)
1177 			return 0;
1178 	}
1179 
1180 	head = this_cpu_ptr(call->perf_events);
1181 	if (hlist_empty(head))
1182 		return 0;
1183 
1184 	dsize = __get_data_size(&tk->tp, regs);
1185 	__size = sizeof(*entry) + tk->tp.size + dsize;
1186 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1187 	size -= sizeof(u32);
1188 
1189 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1190 	if (!entry)
1191 		return 0;
1192 
1193 	entry->ip = (unsigned long)tk->rp.kp.addr;
1194 	memset(&entry[1], 0, dsize);
1195 	store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1196 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1197 			      head, NULL);
1198 	return 0;
1199 }
1200 NOKPROBE_SYMBOL(kprobe_perf_func);
1201 
1202 /* Kretprobe profile handler */
1203 static void
1204 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1205 		    struct pt_regs *regs)
1206 {
1207 	struct trace_event_call *call = &tk->tp.call;
1208 	struct kretprobe_trace_entry_head *entry;
1209 	struct hlist_head *head;
1210 	int size, __size, dsize;
1211 	int rctx;
1212 
1213 	if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1214 		return;
1215 
1216 	head = this_cpu_ptr(call->perf_events);
1217 	if (hlist_empty(head))
1218 		return;
1219 
1220 	dsize = __get_data_size(&tk->tp, regs);
1221 	__size = sizeof(*entry) + tk->tp.size + dsize;
1222 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1223 	size -= sizeof(u32);
1224 
1225 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1226 	if (!entry)
1227 		return;
1228 
1229 	entry->func = (unsigned long)tk->rp.kp.addr;
1230 	entry->ret_ip = (unsigned long)ri->ret_addr;
1231 	store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1232 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1233 			      head, NULL);
1234 }
1235 NOKPROBE_SYMBOL(kretprobe_perf_func);
1236 
1237 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1238 			const char **symbol, u64 *probe_offset,
1239 			u64 *probe_addr, bool perf_type_tracepoint)
1240 {
1241 	const char *pevent = trace_event_name(event->tp_event);
1242 	const char *group = event->tp_event->class->system;
1243 	struct trace_kprobe *tk;
1244 
1245 	if (perf_type_tracepoint)
1246 		tk = find_trace_kprobe(pevent, group);
1247 	else
1248 		tk = event->tp_event->data;
1249 	if (!tk)
1250 		return -EINVAL;
1251 
1252 	*fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1253 					      : BPF_FD_TYPE_KPROBE;
1254 	if (tk->symbol) {
1255 		*symbol = tk->symbol;
1256 		*probe_offset = tk->rp.kp.offset;
1257 		*probe_addr = 0;
1258 	} else {
1259 		*symbol = NULL;
1260 		*probe_offset = 0;
1261 		*probe_addr = (unsigned long)tk->rp.kp.addr;
1262 	}
1263 	return 0;
1264 }
1265 #endif	/* CONFIG_PERF_EVENTS */
1266 
1267 /*
1268  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1269  *
1270  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1271  * lockless, but we can't race with this __init function.
1272  */
1273 static int kprobe_register(struct trace_event_call *event,
1274 			   enum trace_reg type, void *data)
1275 {
1276 	struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1277 	struct trace_event_file *file = data;
1278 
1279 	switch (type) {
1280 	case TRACE_REG_REGISTER:
1281 		return enable_trace_kprobe(tk, file);
1282 	case TRACE_REG_UNREGISTER:
1283 		return disable_trace_kprobe(tk, file);
1284 
1285 #ifdef CONFIG_PERF_EVENTS
1286 	case TRACE_REG_PERF_REGISTER:
1287 		return enable_trace_kprobe(tk, NULL);
1288 	case TRACE_REG_PERF_UNREGISTER:
1289 		return disable_trace_kprobe(tk, NULL);
1290 	case TRACE_REG_PERF_OPEN:
1291 	case TRACE_REG_PERF_CLOSE:
1292 	case TRACE_REG_PERF_ADD:
1293 	case TRACE_REG_PERF_DEL:
1294 		return 0;
1295 #endif
1296 	}
1297 	return 0;
1298 }
1299 
1300 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1301 {
1302 	struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1303 	int ret = 0;
1304 
1305 	raw_cpu_inc(*tk->nhit);
1306 
1307 	if (tk->tp.flags & TP_FLAG_TRACE)
1308 		kprobe_trace_func(tk, regs);
1309 #ifdef CONFIG_PERF_EVENTS
1310 	if (tk->tp.flags & TP_FLAG_PROFILE)
1311 		ret = kprobe_perf_func(tk, regs);
1312 #endif
1313 	return ret;
1314 }
1315 NOKPROBE_SYMBOL(kprobe_dispatcher);
1316 
1317 static int
1318 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1319 {
1320 	struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1321 
1322 	raw_cpu_inc(*tk->nhit);
1323 
1324 	if (tk->tp.flags & TP_FLAG_TRACE)
1325 		kretprobe_trace_func(tk, ri, regs);
1326 #ifdef CONFIG_PERF_EVENTS
1327 	if (tk->tp.flags & TP_FLAG_PROFILE)
1328 		kretprobe_perf_func(tk, ri, regs);
1329 #endif
1330 	return 0;	/* We don't tweek kernel, so just return 0 */
1331 }
1332 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1333 
1334 static struct trace_event_functions kretprobe_funcs = {
1335 	.trace		= print_kretprobe_event
1336 };
1337 
1338 static struct trace_event_functions kprobe_funcs = {
1339 	.trace		= print_kprobe_event
1340 };
1341 
1342 static inline void init_trace_event_call(struct trace_kprobe *tk,
1343 					 struct trace_event_call *call)
1344 {
1345 	INIT_LIST_HEAD(&call->class->fields);
1346 	if (trace_kprobe_is_return(tk)) {
1347 		call->event.funcs = &kretprobe_funcs;
1348 		call->class->define_fields = kretprobe_event_define_fields;
1349 	} else {
1350 		call->event.funcs = &kprobe_funcs;
1351 		call->class->define_fields = kprobe_event_define_fields;
1352 	}
1353 
1354 	call->flags = TRACE_EVENT_FL_KPROBE;
1355 	call->class->reg = kprobe_register;
1356 	call->data = tk;
1357 }
1358 
1359 static int register_kprobe_event(struct trace_kprobe *tk)
1360 {
1361 	struct trace_event_call *call = &tk->tp.call;
1362 	int ret = 0;
1363 
1364 	init_trace_event_call(tk, call);
1365 
1366 	if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1367 		return -ENOMEM;
1368 	ret = register_trace_event(&call->event);
1369 	if (!ret) {
1370 		kfree(call->print_fmt);
1371 		return -ENODEV;
1372 	}
1373 	ret = trace_add_event_call(call);
1374 	if (ret) {
1375 		pr_info("Failed to register kprobe event: %s\n",
1376 			trace_event_name(call));
1377 		kfree(call->print_fmt);
1378 		unregister_trace_event(&call->event);
1379 	}
1380 	return ret;
1381 }
1382 
1383 static int unregister_kprobe_event(struct trace_kprobe *tk)
1384 {
1385 	int ret;
1386 
1387 	/* tp->event is unregistered in trace_remove_event_call() */
1388 	ret = trace_remove_event_call(&tk->tp.call);
1389 	if (!ret)
1390 		kfree(tk->tp.call.print_fmt);
1391 	return ret;
1392 }
1393 
1394 #ifdef CONFIG_PERF_EVENTS
1395 /* create a trace_kprobe, but don't add it to global lists */
1396 struct trace_event_call *
1397 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1398 			  bool is_return)
1399 {
1400 	struct trace_kprobe *tk;
1401 	int ret;
1402 	char *event;
1403 
1404 	/*
1405 	 * local trace_kprobes are not added to dyn_event, so they are never
1406 	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1407 	 * duplicated name here.
1408 	 */
1409 	event = func ? func : "DUMMY_EVENT";
1410 
1411 	tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1412 				offs, 0 /* maxactive */, 0 /* nargs */,
1413 				is_return);
1414 
1415 	if (IS_ERR(tk)) {
1416 		pr_info("Failed to allocate trace_probe.(%d)\n",
1417 			(int)PTR_ERR(tk));
1418 		return ERR_CAST(tk);
1419 	}
1420 
1421 	init_trace_event_call(tk, &tk->tp.call);
1422 
1423 	if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1424 		ret = -ENOMEM;
1425 		goto error;
1426 	}
1427 
1428 	ret = __register_trace_kprobe(tk);
1429 	if (ret < 0) {
1430 		kfree(tk->tp.call.print_fmt);
1431 		goto error;
1432 	}
1433 
1434 	return &tk->tp.call;
1435 error:
1436 	free_trace_kprobe(tk);
1437 	return ERR_PTR(ret);
1438 }
1439 
1440 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1441 {
1442 	struct trace_kprobe *tk;
1443 
1444 	tk = container_of(event_call, struct trace_kprobe, tp.call);
1445 
1446 	if (trace_probe_is_enabled(&tk->tp)) {
1447 		WARN_ON(1);
1448 		return;
1449 	}
1450 
1451 	__unregister_trace_kprobe(tk);
1452 
1453 	kfree(tk->tp.call.print_fmt);
1454 	free_trace_kprobe(tk);
1455 }
1456 #endif /* CONFIG_PERF_EVENTS */
1457 
1458 /* Make a tracefs interface for controlling probe points */
1459 static __init int init_kprobe_trace(void)
1460 {
1461 	struct dentry *d_tracer;
1462 	struct dentry *entry;
1463 	int ret;
1464 
1465 	ret = dyn_event_register(&trace_kprobe_ops);
1466 	if (ret)
1467 		return ret;
1468 
1469 	if (register_module_notifier(&trace_kprobe_module_nb))
1470 		return -EINVAL;
1471 
1472 	d_tracer = tracing_init_dentry();
1473 	if (IS_ERR(d_tracer))
1474 		return 0;
1475 
1476 	entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1477 				    NULL, &kprobe_events_ops);
1478 
1479 	/* Event list interface */
1480 	if (!entry)
1481 		pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1482 
1483 	/* Profile interface */
1484 	entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1485 				    NULL, &kprobe_profile_ops);
1486 
1487 	if (!entry)
1488 		pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1489 	return 0;
1490 }
1491 fs_initcall(init_kprobe_trace);
1492 
1493 
1494 #ifdef CONFIG_FTRACE_STARTUP_TEST
1495 static __init struct trace_event_file *
1496 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1497 {
1498 	struct trace_event_file *file;
1499 
1500 	list_for_each_entry(file, &tr->events, list)
1501 		if (file->event_call == &tk->tp.call)
1502 			return file;
1503 
1504 	return NULL;
1505 }
1506 
1507 /*
1508  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1509  * stage, we can do this lockless.
1510  */
1511 static __init int kprobe_trace_self_tests_init(void)
1512 {
1513 	int ret, warn = 0;
1514 	int (*target)(int, int, int, int, int, int);
1515 	struct trace_kprobe *tk;
1516 	struct trace_event_file *file;
1517 
1518 	if (tracing_is_disabled())
1519 		return -ENODEV;
1520 
1521 	target = kprobe_trace_selftest_target;
1522 
1523 	pr_info("Testing kprobe tracing: ");
1524 
1525 	ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
1526 				create_or_delete_trace_kprobe);
1527 	if (WARN_ON_ONCE(ret)) {
1528 		pr_warn("error on probing function entry.\n");
1529 		warn++;
1530 	} else {
1531 		/* Enable trace point */
1532 		tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1533 		if (WARN_ON_ONCE(tk == NULL)) {
1534 			pr_warn("error on getting new probe.\n");
1535 			warn++;
1536 		} else {
1537 			file = find_trace_probe_file(tk, top_trace_array());
1538 			if (WARN_ON_ONCE(file == NULL)) {
1539 				pr_warn("error on getting probe file.\n");
1540 				warn++;
1541 			} else
1542 				enable_trace_kprobe(tk, file);
1543 		}
1544 	}
1545 
1546 	ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
1547 				create_or_delete_trace_kprobe);
1548 	if (WARN_ON_ONCE(ret)) {
1549 		pr_warn("error on probing function return.\n");
1550 		warn++;
1551 	} else {
1552 		/* Enable trace point */
1553 		tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1554 		if (WARN_ON_ONCE(tk == NULL)) {
1555 			pr_warn("error on getting 2nd new probe.\n");
1556 			warn++;
1557 		} else {
1558 			file = find_trace_probe_file(tk, top_trace_array());
1559 			if (WARN_ON_ONCE(file == NULL)) {
1560 				pr_warn("error on getting probe file.\n");
1561 				warn++;
1562 			} else
1563 				enable_trace_kprobe(tk, file);
1564 		}
1565 	}
1566 
1567 	if (warn)
1568 		goto end;
1569 
1570 	ret = target(1, 2, 3, 4, 5, 6);
1571 
1572 	/*
1573 	 * Not expecting an error here, the check is only to prevent the
1574 	 * optimizer from removing the call to target() as otherwise there
1575 	 * are no side-effects and the call is never performed.
1576 	 */
1577 	if (ret != 21)
1578 		warn++;
1579 
1580 	/* Disable trace points before removing it */
1581 	tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1582 	if (WARN_ON_ONCE(tk == NULL)) {
1583 		pr_warn("error on getting test probe.\n");
1584 		warn++;
1585 	} else {
1586 		if (trace_kprobe_nhit(tk) != 1) {
1587 			pr_warn("incorrect number of testprobe hits\n");
1588 			warn++;
1589 		}
1590 
1591 		file = find_trace_probe_file(tk, top_trace_array());
1592 		if (WARN_ON_ONCE(file == NULL)) {
1593 			pr_warn("error on getting probe file.\n");
1594 			warn++;
1595 		} else
1596 			disable_trace_kprobe(tk, file);
1597 	}
1598 
1599 	tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1600 	if (WARN_ON_ONCE(tk == NULL)) {
1601 		pr_warn("error on getting 2nd test probe.\n");
1602 		warn++;
1603 	} else {
1604 		if (trace_kprobe_nhit(tk) != 1) {
1605 			pr_warn("incorrect number of testprobe2 hits\n");
1606 			warn++;
1607 		}
1608 
1609 		file = find_trace_probe_file(tk, top_trace_array());
1610 		if (WARN_ON_ONCE(file == NULL)) {
1611 			pr_warn("error on getting probe file.\n");
1612 			warn++;
1613 		} else
1614 			disable_trace_kprobe(tk, file);
1615 	}
1616 
1617 	ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
1618 	if (WARN_ON_ONCE(ret)) {
1619 		pr_warn("error on deleting a probe.\n");
1620 		warn++;
1621 	}
1622 
1623 	ret = trace_run_command("-:testprobe2", create_or_delete_trace_kprobe);
1624 	if (WARN_ON_ONCE(ret)) {
1625 		pr_warn("error on deleting a probe.\n");
1626 		warn++;
1627 	}
1628 
1629 end:
1630 	ret = dyn_events_release_all(&trace_kprobe_ops);
1631 	if (WARN_ON_ONCE(ret)) {
1632 		pr_warn("error on cleaning up probes.\n");
1633 		warn++;
1634 	}
1635 	/*
1636 	 * Wait for the optimizer work to finish. Otherwise it might fiddle
1637 	 * with probes in already freed __init text.
1638 	 */
1639 	wait_for_kprobe_optimizer();
1640 	if (warn)
1641 		pr_cont("NG: Some tests are failed. Please check them.\n");
1642 	else
1643 		pr_cont("OK\n");
1644 	return 0;
1645 }
1646 
1647 late_initcall(kprobe_trace_self_tests_init);
1648 
1649 #endif
1650