xref: /linux/kernel/trace/trace_kprobe.c (revision f3a8b6645dc2e60d11f20c1c23afd964ff4e55ae)
1 /*
2  * Kprobes-based tracing events
3  *
4  * Created by Masami Hiramatsu <mhiramat@redhat.com>
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write to the Free Software
17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
18  */
19 
20 #include <linux/module.h>
21 #include <linux/uaccess.h>
22 
23 #include "trace_probe.h"
24 
25 #define KPROBE_EVENT_SYSTEM "kprobes"
26 
27 /**
28  * Kprobe event core functions
29  */
30 struct trace_kprobe {
31 	struct list_head	list;
32 	struct kretprobe	rp;	/* Use rp.kp for kprobe use */
33 	unsigned long __percpu *nhit;
34 	const char		*symbol;	/* symbol name */
35 	struct trace_probe	tp;
36 };
37 
38 #define SIZEOF_TRACE_KPROBE(n)				\
39 	(offsetof(struct trace_kprobe, tp.args) +	\
40 	(sizeof(struct probe_arg) * (n)))
41 
42 
43 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
44 {
45 	return tk->rp.handler != NULL;
46 }
47 
48 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
49 {
50 	return tk->symbol ? tk->symbol : "unknown";
51 }
52 
53 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
54 {
55 	return tk->rp.kp.offset;
56 }
57 
58 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
59 {
60 	return !!(kprobe_gone(&tk->rp.kp));
61 }
62 
63 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
64 						 struct module *mod)
65 {
66 	int len = strlen(mod->name);
67 	const char *name = trace_kprobe_symbol(tk);
68 	return strncmp(mod->name, name, len) == 0 && name[len] == ':';
69 }
70 
71 static nokprobe_inline bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
72 {
73 	return !!strchr(trace_kprobe_symbol(tk), ':');
74 }
75 
76 static int register_kprobe_event(struct trace_kprobe *tk);
77 static int unregister_kprobe_event(struct trace_kprobe *tk);
78 
79 static DEFINE_MUTEX(probe_lock);
80 static LIST_HEAD(probe_list);
81 
82 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
83 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
84 				struct pt_regs *regs);
85 
86 /* Memory fetching by symbol */
87 struct symbol_cache {
88 	char		*symbol;
89 	long		offset;
90 	unsigned long	addr;
91 };
92 
93 unsigned long update_symbol_cache(struct symbol_cache *sc)
94 {
95 	sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
96 
97 	if (sc->addr)
98 		sc->addr += sc->offset;
99 
100 	return sc->addr;
101 }
102 
103 void free_symbol_cache(struct symbol_cache *sc)
104 {
105 	kfree(sc->symbol);
106 	kfree(sc);
107 }
108 
109 struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
110 {
111 	struct symbol_cache *sc;
112 
113 	if (!sym || strlen(sym) == 0)
114 		return NULL;
115 
116 	sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
117 	if (!sc)
118 		return NULL;
119 
120 	sc->symbol = kstrdup(sym, GFP_KERNEL);
121 	if (!sc->symbol) {
122 		kfree(sc);
123 		return NULL;
124 	}
125 	sc->offset = offset;
126 	update_symbol_cache(sc);
127 
128 	return sc;
129 }
130 
131 /*
132  * Kprobes-specific fetch functions
133  */
134 #define DEFINE_FETCH_stack(type)					\
135 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,		\
136 					  void *offset, void *dest)	\
137 {									\
138 	*(type *)dest = (type)regs_get_kernel_stack_nth(regs,		\
139 				(unsigned int)((unsigned long)offset));	\
140 }									\
141 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
142 
143 DEFINE_BASIC_FETCH_FUNCS(stack)
144 /* No string on the stack entry */
145 #define fetch_stack_string	NULL
146 #define fetch_stack_string_size	NULL
147 
148 #define DEFINE_FETCH_memory(type)					\
149 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,		\
150 					  void *addr, void *dest)	\
151 {									\
152 	type retval;							\
153 	if (probe_kernel_address(addr, retval))				\
154 		*(type *)dest = 0;					\
155 	else								\
156 		*(type *)dest = retval;					\
157 }									\
158 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
159 
160 DEFINE_BASIC_FETCH_FUNCS(memory)
161 /*
162  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
163  * length and relative data location.
164  */
165 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
166 					    void *addr, void *dest)
167 {
168 	int maxlen = get_rloc_len(*(u32 *)dest);
169 	u8 *dst = get_rloc_data(dest);
170 	long ret;
171 
172 	if (!maxlen)
173 		return;
174 
175 	/*
176 	 * Try to get string again, since the string can be changed while
177 	 * probing.
178 	 */
179 	ret = strncpy_from_unsafe(dst, addr, maxlen);
180 
181 	if (ret < 0) {	/* Failed to fetch string */
182 		dst[0] = '\0';
183 		*(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
184 	} else {
185 		*(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
186 	}
187 }
188 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
189 
190 /* Return the length of string -- including null terminal byte */
191 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
192 						 void *addr, void *dest)
193 {
194 	mm_segment_t old_fs;
195 	int ret, len = 0;
196 	u8 c;
197 
198 	old_fs = get_fs();
199 	set_fs(KERNEL_DS);
200 	pagefault_disable();
201 
202 	do {
203 		ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
204 		len++;
205 	} while (c && ret == 0 && len < MAX_STRING_SIZE);
206 
207 	pagefault_enable();
208 	set_fs(old_fs);
209 
210 	if (ret < 0)	/* Failed to check the length */
211 		*(u32 *)dest = 0;
212 	else
213 		*(u32 *)dest = len;
214 }
215 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
216 
217 #define DEFINE_FETCH_symbol(type)					\
218 void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
219 {									\
220 	struct symbol_cache *sc = data;					\
221 	if (sc->addr)							\
222 		fetch_memory_##type(regs, (void *)sc->addr, dest);	\
223 	else								\
224 		*(type *)dest = 0;					\
225 }									\
226 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
227 
228 DEFINE_BASIC_FETCH_FUNCS(symbol)
229 DEFINE_FETCH_symbol(string)
230 DEFINE_FETCH_symbol(string_size)
231 
232 /* kprobes don't support file_offset fetch methods */
233 #define fetch_file_offset_u8		NULL
234 #define fetch_file_offset_u16		NULL
235 #define fetch_file_offset_u32		NULL
236 #define fetch_file_offset_u64		NULL
237 #define fetch_file_offset_string	NULL
238 #define fetch_file_offset_string_size	NULL
239 
240 /* Fetch type information table */
241 static const struct fetch_type kprobes_fetch_type_table[] = {
242 	/* Special types */
243 	[FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
244 					sizeof(u32), 1, "__data_loc char[]"),
245 	[FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
246 					string_size, sizeof(u32), 0, "u32"),
247 	/* Basic types */
248 	ASSIGN_FETCH_TYPE(u8,  u8,  0),
249 	ASSIGN_FETCH_TYPE(u16, u16, 0),
250 	ASSIGN_FETCH_TYPE(u32, u32, 0),
251 	ASSIGN_FETCH_TYPE(u64, u64, 0),
252 	ASSIGN_FETCH_TYPE(s8,  u8,  1),
253 	ASSIGN_FETCH_TYPE(s16, u16, 1),
254 	ASSIGN_FETCH_TYPE(s32, u32, 1),
255 	ASSIGN_FETCH_TYPE(s64, u64, 1),
256 	ASSIGN_FETCH_TYPE_ALIAS(x8,  u8,  u8,  0),
257 	ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
258 	ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
259 	ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
260 
261 	ASSIGN_FETCH_TYPE_END
262 };
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 nargs, bool is_return)
273 {
274 	struct trace_kprobe *tk;
275 	int ret = -ENOMEM;
276 
277 	tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
278 	if (!tk)
279 		return ERR_PTR(ret);
280 
281 	tk->nhit = alloc_percpu(unsigned long);
282 	if (!tk->nhit)
283 		goto error;
284 
285 	if (symbol) {
286 		tk->symbol = kstrdup(symbol, GFP_KERNEL);
287 		if (!tk->symbol)
288 			goto error;
289 		tk->rp.kp.symbol_name = tk->symbol;
290 		tk->rp.kp.offset = offs;
291 	} else
292 		tk->rp.kp.addr = addr;
293 
294 	if (is_return)
295 		tk->rp.handler = kretprobe_dispatcher;
296 	else
297 		tk->rp.kp.pre_handler = kprobe_dispatcher;
298 
299 	if (!event || !is_good_name(event)) {
300 		ret = -EINVAL;
301 		goto error;
302 	}
303 
304 	tk->tp.call.class = &tk->tp.class;
305 	tk->tp.call.name = kstrdup(event, GFP_KERNEL);
306 	if (!tk->tp.call.name)
307 		goto error;
308 
309 	if (!group || !is_good_name(group)) {
310 		ret = -EINVAL;
311 		goto error;
312 	}
313 
314 	tk->tp.class.system = kstrdup(group, GFP_KERNEL);
315 	if (!tk->tp.class.system)
316 		goto error;
317 
318 	INIT_LIST_HEAD(&tk->list);
319 	INIT_LIST_HEAD(&tk->tp.files);
320 	return tk;
321 error:
322 	kfree(tk->tp.call.name);
323 	kfree(tk->symbol);
324 	free_percpu(tk->nhit);
325 	kfree(tk);
326 	return ERR_PTR(ret);
327 }
328 
329 static void free_trace_kprobe(struct trace_kprobe *tk)
330 {
331 	int i;
332 
333 	for (i = 0; i < tk->tp.nr_args; i++)
334 		traceprobe_free_probe_arg(&tk->tp.args[i]);
335 
336 	kfree(tk->tp.call.class->system);
337 	kfree(tk->tp.call.name);
338 	kfree(tk->symbol);
339 	free_percpu(tk->nhit);
340 	kfree(tk);
341 }
342 
343 static struct trace_kprobe *find_trace_kprobe(const char *event,
344 					      const char *group)
345 {
346 	struct trace_kprobe *tk;
347 
348 	list_for_each_entry(tk, &probe_list, list)
349 		if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
350 		    strcmp(tk->tp.call.class->system, group) == 0)
351 			return tk;
352 	return NULL;
353 }
354 
355 /*
356  * Enable trace_probe
357  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
358  */
359 static int
360 enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
361 {
362 	int ret = 0;
363 
364 	if (file) {
365 		struct event_file_link *link;
366 
367 		link = kmalloc(sizeof(*link), GFP_KERNEL);
368 		if (!link) {
369 			ret = -ENOMEM;
370 			goto out;
371 		}
372 
373 		link->file = file;
374 		list_add_tail_rcu(&link->list, &tk->tp.files);
375 
376 		tk->tp.flags |= TP_FLAG_TRACE;
377 	} else
378 		tk->tp.flags |= TP_FLAG_PROFILE;
379 
380 	if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
381 		if (trace_kprobe_is_return(tk))
382 			ret = enable_kretprobe(&tk->rp);
383 		else
384 			ret = enable_kprobe(&tk->rp.kp);
385 	}
386  out:
387 	return ret;
388 }
389 
390 /*
391  * Disable trace_probe
392  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
393  */
394 static int
395 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
396 {
397 	struct event_file_link *link = NULL;
398 	int wait = 0;
399 	int ret = 0;
400 
401 	if (file) {
402 		link = find_event_file_link(&tk->tp, file);
403 		if (!link) {
404 			ret = -EINVAL;
405 			goto out;
406 		}
407 
408 		list_del_rcu(&link->list);
409 		wait = 1;
410 		if (!list_empty(&tk->tp.files))
411 			goto out;
412 
413 		tk->tp.flags &= ~TP_FLAG_TRACE;
414 	} else
415 		tk->tp.flags &= ~TP_FLAG_PROFILE;
416 
417 	if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
418 		if (trace_kprobe_is_return(tk))
419 			disable_kretprobe(&tk->rp);
420 		else
421 			disable_kprobe(&tk->rp.kp);
422 		wait = 1;
423 	}
424  out:
425 	if (wait) {
426 		/*
427 		 * Synchronize with kprobe_trace_func/kretprobe_trace_func
428 		 * to ensure disabled (all running handlers are finished).
429 		 * This is not only for kfree(), but also the caller,
430 		 * trace_remove_event_call() supposes it for releasing
431 		 * event_call related objects, which will be accessed in
432 		 * the kprobe_trace_func/kretprobe_trace_func.
433 		 */
434 		synchronize_sched();
435 		kfree(link);	/* Ignored if link == NULL */
436 	}
437 
438 	return ret;
439 }
440 
441 /* Internal register function - just handle k*probes and flags */
442 static int __register_trace_kprobe(struct trace_kprobe *tk)
443 {
444 	int i, ret;
445 
446 	if (trace_probe_is_registered(&tk->tp))
447 		return -EINVAL;
448 
449 	for (i = 0; i < tk->tp.nr_args; i++)
450 		traceprobe_update_arg(&tk->tp.args[i]);
451 
452 	/* Set/clear disabled flag according to tp->flag */
453 	if (trace_probe_is_enabled(&tk->tp))
454 		tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
455 	else
456 		tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
457 
458 	if (trace_kprobe_is_return(tk))
459 		ret = register_kretprobe(&tk->rp);
460 	else
461 		ret = register_kprobe(&tk->rp.kp);
462 
463 	if (ret == 0)
464 		tk->tp.flags |= TP_FLAG_REGISTERED;
465 	else {
466 		pr_warn("Could not insert probe at %s+%lu: %d\n",
467 			trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
468 		if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
469 			pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
470 			ret = 0;
471 		} else if (ret == -EILSEQ) {
472 			pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
473 				tk->rp.kp.addr);
474 			ret = -EINVAL;
475 		}
476 	}
477 
478 	return ret;
479 }
480 
481 /* Internal unregister function - just handle k*probes and flags */
482 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
483 {
484 	if (trace_probe_is_registered(&tk->tp)) {
485 		if (trace_kprobe_is_return(tk))
486 			unregister_kretprobe(&tk->rp);
487 		else
488 			unregister_kprobe(&tk->rp.kp);
489 		tk->tp.flags &= ~TP_FLAG_REGISTERED;
490 		/* Cleanup kprobe for reuse */
491 		if (tk->rp.kp.symbol_name)
492 			tk->rp.kp.addr = NULL;
493 	}
494 }
495 
496 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
497 static int unregister_trace_kprobe(struct trace_kprobe *tk)
498 {
499 	/* Enabled event can not be unregistered */
500 	if (trace_probe_is_enabled(&tk->tp))
501 		return -EBUSY;
502 
503 	/* Will fail if probe is being used by ftrace or perf */
504 	if (unregister_kprobe_event(tk))
505 		return -EBUSY;
506 
507 	__unregister_trace_kprobe(tk);
508 	list_del(&tk->list);
509 
510 	return 0;
511 }
512 
513 /* Register a trace_probe and probe_event */
514 static int register_trace_kprobe(struct trace_kprobe *tk)
515 {
516 	struct trace_kprobe *old_tk;
517 	int ret;
518 
519 	mutex_lock(&probe_lock);
520 
521 	/* Delete old (same name) event if exist */
522 	old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
523 			tk->tp.call.class->system);
524 	if (old_tk) {
525 		ret = unregister_trace_kprobe(old_tk);
526 		if (ret < 0)
527 			goto end;
528 		free_trace_kprobe(old_tk);
529 	}
530 
531 	/* Register new event */
532 	ret = register_kprobe_event(tk);
533 	if (ret) {
534 		pr_warn("Failed to register probe event(%d)\n", ret);
535 		goto end;
536 	}
537 
538 	/* Register k*probe */
539 	ret = __register_trace_kprobe(tk);
540 	if (ret < 0)
541 		unregister_kprobe_event(tk);
542 	else
543 		list_add_tail(&tk->list, &probe_list);
544 
545 end:
546 	mutex_unlock(&probe_lock);
547 	return ret;
548 }
549 
550 /* Module notifier call back, checking event on the module */
551 static int trace_kprobe_module_callback(struct notifier_block *nb,
552 				       unsigned long val, void *data)
553 {
554 	struct module *mod = data;
555 	struct trace_kprobe *tk;
556 	int ret;
557 
558 	if (val != MODULE_STATE_COMING)
559 		return NOTIFY_DONE;
560 
561 	/* Update probes on coming module */
562 	mutex_lock(&probe_lock);
563 	list_for_each_entry(tk, &probe_list, list) {
564 		if (trace_kprobe_within_module(tk, mod)) {
565 			/* Don't need to check busy - this should have gone. */
566 			__unregister_trace_kprobe(tk);
567 			ret = __register_trace_kprobe(tk);
568 			if (ret)
569 				pr_warn("Failed to re-register probe %s on %s: %d\n",
570 					trace_event_name(&tk->tp.call),
571 					mod->name, ret);
572 		}
573 	}
574 	mutex_unlock(&probe_lock);
575 
576 	return NOTIFY_DONE;
577 }
578 
579 static struct notifier_block trace_kprobe_module_nb = {
580 	.notifier_call = trace_kprobe_module_callback,
581 	.priority = 1	/* Invoked after kprobe module callback */
582 };
583 
584 static int create_trace_kprobe(int argc, char **argv)
585 {
586 	/*
587 	 * Argument syntax:
588 	 *  - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
589 	 *  - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
590 	 * Fetch args:
591 	 *  $retval	: fetch return value
592 	 *  $stack	: fetch stack address
593 	 *  $stackN	: fetch Nth of stack (N:0-)
594 	 *  $comm       : fetch current task comm
595 	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
596 	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
597 	 *  %REG	: fetch register REG
598 	 * Dereferencing memory fetch:
599 	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
600 	 * Alias name of args:
601 	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
602 	 * Type of args:
603 	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
604 	 */
605 	struct trace_kprobe *tk;
606 	int i, ret = 0;
607 	bool is_return = false, is_delete = false;
608 	char *symbol = NULL, *event = NULL, *group = NULL;
609 	char *arg;
610 	unsigned long offset = 0;
611 	void *addr = NULL;
612 	char buf[MAX_EVENT_NAME_LEN];
613 
614 	/* argc must be >= 1 */
615 	if (argv[0][0] == 'p')
616 		is_return = false;
617 	else if (argv[0][0] == 'r')
618 		is_return = true;
619 	else if (argv[0][0] == '-')
620 		is_delete = true;
621 	else {
622 		pr_info("Probe definition must be started with 'p', 'r' or"
623 			" '-'.\n");
624 		return -EINVAL;
625 	}
626 
627 	if (argv[0][1] == ':') {
628 		event = &argv[0][2];
629 		if (strchr(event, '/')) {
630 			group = event;
631 			event = strchr(group, '/') + 1;
632 			event[-1] = '\0';
633 			if (strlen(group) == 0) {
634 				pr_info("Group name is not specified\n");
635 				return -EINVAL;
636 			}
637 		}
638 		if (strlen(event) == 0) {
639 			pr_info("Event name is not specified\n");
640 			return -EINVAL;
641 		}
642 	}
643 	if (!group)
644 		group = KPROBE_EVENT_SYSTEM;
645 
646 	if (is_delete) {
647 		if (!event) {
648 			pr_info("Delete command needs an event name.\n");
649 			return -EINVAL;
650 		}
651 		mutex_lock(&probe_lock);
652 		tk = find_trace_kprobe(event, group);
653 		if (!tk) {
654 			mutex_unlock(&probe_lock);
655 			pr_info("Event %s/%s doesn't exist.\n", group, event);
656 			return -ENOENT;
657 		}
658 		/* delete an event */
659 		ret = unregister_trace_kprobe(tk);
660 		if (ret == 0)
661 			free_trace_kprobe(tk);
662 		mutex_unlock(&probe_lock);
663 		return ret;
664 	}
665 
666 	if (argc < 2) {
667 		pr_info("Probe point is not specified.\n");
668 		return -EINVAL;
669 	}
670 	if (isdigit(argv[1][0])) {
671 		if (is_return) {
672 			pr_info("Return probe point must be a symbol.\n");
673 			return -EINVAL;
674 		}
675 		/* an address specified */
676 		ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
677 		if (ret) {
678 			pr_info("Failed to parse address.\n");
679 			return ret;
680 		}
681 	} else {
682 		/* a symbol specified */
683 		symbol = argv[1];
684 		/* TODO: support .init module functions */
685 		ret = traceprobe_split_symbol_offset(symbol, &offset);
686 		if (ret) {
687 			pr_info("Failed to parse symbol.\n");
688 			return ret;
689 		}
690 		if (offset && is_return) {
691 			pr_info("Return probe must be used without offset.\n");
692 			return -EINVAL;
693 		}
694 	}
695 	argc -= 2; argv += 2;
696 
697 	/* setup a probe */
698 	if (!event) {
699 		/* Make a new event name */
700 		if (symbol)
701 			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
702 				 is_return ? 'r' : 'p', symbol, offset);
703 		else
704 			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
705 				 is_return ? 'r' : 'p', addr);
706 		event = buf;
707 	}
708 	tk = alloc_trace_kprobe(group, event, addr, symbol, offset, argc,
709 			       is_return);
710 	if (IS_ERR(tk)) {
711 		pr_info("Failed to allocate trace_probe.(%d)\n",
712 			(int)PTR_ERR(tk));
713 		return PTR_ERR(tk);
714 	}
715 
716 	/* parse arguments */
717 	ret = 0;
718 	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
719 		struct probe_arg *parg = &tk->tp.args[i];
720 
721 		/* Increment count for freeing args in error case */
722 		tk->tp.nr_args++;
723 
724 		/* Parse argument name */
725 		arg = strchr(argv[i], '=');
726 		if (arg) {
727 			*arg++ = '\0';
728 			parg->name = kstrdup(argv[i], GFP_KERNEL);
729 		} else {
730 			arg = argv[i];
731 			/* If argument name is omitted, set "argN" */
732 			snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
733 			parg->name = kstrdup(buf, GFP_KERNEL);
734 		}
735 
736 		if (!parg->name) {
737 			pr_info("Failed to allocate argument[%d] name.\n", i);
738 			ret = -ENOMEM;
739 			goto error;
740 		}
741 
742 		if (!is_good_name(parg->name)) {
743 			pr_info("Invalid argument[%d] name: %s\n",
744 				i, parg->name);
745 			ret = -EINVAL;
746 			goto error;
747 		}
748 
749 		if (traceprobe_conflict_field_name(parg->name,
750 							tk->tp.args, i)) {
751 			pr_info("Argument[%d] name '%s' conflicts with "
752 				"another field.\n", i, argv[i]);
753 			ret = -EINVAL;
754 			goto error;
755 		}
756 
757 		/* Parse fetch argument */
758 		ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
759 						is_return, true,
760 						kprobes_fetch_type_table);
761 		if (ret) {
762 			pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
763 			goto error;
764 		}
765 	}
766 
767 	ret = register_trace_kprobe(tk);
768 	if (ret)
769 		goto error;
770 	return 0;
771 
772 error:
773 	free_trace_kprobe(tk);
774 	return ret;
775 }
776 
777 static int release_all_trace_kprobes(void)
778 {
779 	struct trace_kprobe *tk;
780 	int ret = 0;
781 
782 	mutex_lock(&probe_lock);
783 	/* Ensure no probe is in use. */
784 	list_for_each_entry(tk, &probe_list, list)
785 		if (trace_probe_is_enabled(&tk->tp)) {
786 			ret = -EBUSY;
787 			goto end;
788 		}
789 	/* TODO: Use batch unregistration */
790 	while (!list_empty(&probe_list)) {
791 		tk = list_entry(probe_list.next, struct trace_kprobe, list);
792 		ret = unregister_trace_kprobe(tk);
793 		if (ret)
794 			goto end;
795 		free_trace_kprobe(tk);
796 	}
797 
798 end:
799 	mutex_unlock(&probe_lock);
800 
801 	return ret;
802 }
803 
804 /* Probes listing interfaces */
805 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
806 {
807 	mutex_lock(&probe_lock);
808 	return seq_list_start(&probe_list, *pos);
809 }
810 
811 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
812 {
813 	return seq_list_next(v, &probe_list, pos);
814 }
815 
816 static void probes_seq_stop(struct seq_file *m, void *v)
817 {
818 	mutex_unlock(&probe_lock);
819 }
820 
821 static int probes_seq_show(struct seq_file *m, void *v)
822 {
823 	struct trace_kprobe *tk = v;
824 	int i;
825 
826 	seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
827 	seq_printf(m, ":%s/%s", tk->tp.call.class->system,
828 			trace_event_name(&tk->tp.call));
829 
830 	if (!tk->symbol)
831 		seq_printf(m, " 0x%p", tk->rp.kp.addr);
832 	else if (tk->rp.kp.offset)
833 		seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
834 			   tk->rp.kp.offset);
835 	else
836 		seq_printf(m, " %s", trace_kprobe_symbol(tk));
837 
838 	for (i = 0; i < tk->tp.nr_args; i++)
839 		seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
840 	seq_putc(m, '\n');
841 
842 	return 0;
843 }
844 
845 static const struct seq_operations probes_seq_op = {
846 	.start  = probes_seq_start,
847 	.next   = probes_seq_next,
848 	.stop   = probes_seq_stop,
849 	.show   = probes_seq_show
850 };
851 
852 static int probes_open(struct inode *inode, struct file *file)
853 {
854 	int ret;
855 
856 	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
857 		ret = release_all_trace_kprobes();
858 		if (ret < 0)
859 			return ret;
860 	}
861 
862 	return seq_open(file, &probes_seq_op);
863 }
864 
865 static ssize_t probes_write(struct file *file, const char __user *buffer,
866 			    size_t count, loff_t *ppos)
867 {
868 	return traceprobe_probes_write(file, buffer, count, ppos,
869 			create_trace_kprobe);
870 }
871 
872 static const struct file_operations kprobe_events_ops = {
873 	.owner          = THIS_MODULE,
874 	.open           = probes_open,
875 	.read           = seq_read,
876 	.llseek         = seq_lseek,
877 	.release        = seq_release,
878 	.write		= probes_write,
879 };
880 
881 /* Probes profiling interfaces */
882 static int probes_profile_seq_show(struct seq_file *m, void *v)
883 {
884 	struct trace_kprobe *tk = v;
885 	unsigned long nhit = 0;
886 	int cpu;
887 
888 	for_each_possible_cpu(cpu)
889 		nhit += *per_cpu_ptr(tk->nhit, cpu);
890 
891 	seq_printf(m, "  %-44s %15lu %15lu\n",
892 		   trace_event_name(&tk->tp.call), nhit,
893 		   tk->rp.kp.nmissed);
894 
895 	return 0;
896 }
897 
898 static const struct seq_operations profile_seq_op = {
899 	.start  = probes_seq_start,
900 	.next   = probes_seq_next,
901 	.stop   = probes_seq_stop,
902 	.show   = probes_profile_seq_show
903 };
904 
905 static int profile_open(struct inode *inode, struct file *file)
906 {
907 	return seq_open(file, &profile_seq_op);
908 }
909 
910 static const struct file_operations kprobe_profile_ops = {
911 	.owner          = THIS_MODULE,
912 	.open           = profile_open,
913 	.read           = seq_read,
914 	.llseek         = seq_lseek,
915 	.release        = seq_release,
916 };
917 
918 /* Kprobe handler */
919 static nokprobe_inline void
920 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
921 		    struct trace_event_file *trace_file)
922 {
923 	struct kprobe_trace_entry_head *entry;
924 	struct ring_buffer_event *event;
925 	struct ring_buffer *buffer;
926 	int size, dsize, pc;
927 	unsigned long irq_flags;
928 	struct trace_event_call *call = &tk->tp.call;
929 
930 	WARN_ON(call != trace_file->event_call);
931 
932 	if (trace_trigger_soft_disabled(trace_file))
933 		return;
934 
935 	local_save_flags(irq_flags);
936 	pc = preempt_count();
937 
938 	dsize = __get_data_size(&tk->tp, regs);
939 	size = sizeof(*entry) + tk->tp.size + dsize;
940 
941 	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
942 						call->event.type,
943 						size, irq_flags, pc);
944 	if (!event)
945 		return;
946 
947 	entry = ring_buffer_event_data(event);
948 	entry->ip = (unsigned long)tk->rp.kp.addr;
949 	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
950 
951 	event_trigger_unlock_commit_regs(trace_file, buffer, event,
952 					 entry, irq_flags, pc, regs);
953 }
954 
955 static void
956 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
957 {
958 	struct event_file_link *link;
959 
960 	list_for_each_entry_rcu(link, &tk->tp.files, list)
961 		__kprobe_trace_func(tk, regs, link->file);
962 }
963 NOKPROBE_SYMBOL(kprobe_trace_func);
964 
965 /* Kretprobe handler */
966 static nokprobe_inline void
967 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
968 		       struct pt_regs *regs,
969 		       struct trace_event_file *trace_file)
970 {
971 	struct kretprobe_trace_entry_head *entry;
972 	struct ring_buffer_event *event;
973 	struct ring_buffer *buffer;
974 	int size, pc, dsize;
975 	unsigned long irq_flags;
976 	struct trace_event_call *call = &tk->tp.call;
977 
978 	WARN_ON(call != trace_file->event_call);
979 
980 	if (trace_trigger_soft_disabled(trace_file))
981 		return;
982 
983 	local_save_flags(irq_flags);
984 	pc = preempt_count();
985 
986 	dsize = __get_data_size(&tk->tp, regs);
987 	size = sizeof(*entry) + tk->tp.size + dsize;
988 
989 	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
990 						call->event.type,
991 						size, irq_flags, pc);
992 	if (!event)
993 		return;
994 
995 	entry = ring_buffer_event_data(event);
996 	entry->func = (unsigned long)tk->rp.kp.addr;
997 	entry->ret_ip = (unsigned long)ri->ret_addr;
998 	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
999 
1000 	event_trigger_unlock_commit_regs(trace_file, buffer, event,
1001 					 entry, irq_flags, pc, regs);
1002 }
1003 
1004 static void
1005 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1006 		     struct pt_regs *regs)
1007 {
1008 	struct event_file_link *link;
1009 
1010 	list_for_each_entry_rcu(link, &tk->tp.files, list)
1011 		__kretprobe_trace_func(tk, ri, regs, link->file);
1012 }
1013 NOKPROBE_SYMBOL(kretprobe_trace_func);
1014 
1015 /* Event entry printers */
1016 static enum print_line_t
1017 print_kprobe_event(struct trace_iterator *iter, int flags,
1018 		   struct trace_event *event)
1019 {
1020 	struct kprobe_trace_entry_head *field;
1021 	struct trace_seq *s = &iter->seq;
1022 	struct trace_probe *tp;
1023 	u8 *data;
1024 	int i;
1025 
1026 	field = (struct kprobe_trace_entry_head *)iter->ent;
1027 	tp = container_of(event, struct trace_probe, call.event);
1028 
1029 	trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1030 
1031 	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1032 		goto out;
1033 
1034 	trace_seq_putc(s, ')');
1035 
1036 	data = (u8 *)&field[1];
1037 	for (i = 0; i < tp->nr_args; i++)
1038 		if (!tp->args[i].type->print(s, tp->args[i].name,
1039 					     data + tp->args[i].offset, field))
1040 			goto out;
1041 
1042 	trace_seq_putc(s, '\n');
1043  out:
1044 	return trace_handle_return(s);
1045 }
1046 
1047 static enum print_line_t
1048 print_kretprobe_event(struct trace_iterator *iter, int flags,
1049 		      struct trace_event *event)
1050 {
1051 	struct kretprobe_trace_entry_head *field;
1052 	struct trace_seq *s = &iter->seq;
1053 	struct trace_probe *tp;
1054 	u8 *data;
1055 	int i;
1056 
1057 	field = (struct kretprobe_trace_entry_head *)iter->ent;
1058 	tp = container_of(event, struct trace_probe, call.event);
1059 
1060 	trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1061 
1062 	if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1063 		goto out;
1064 
1065 	trace_seq_puts(s, " <- ");
1066 
1067 	if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1068 		goto out;
1069 
1070 	trace_seq_putc(s, ')');
1071 
1072 	data = (u8 *)&field[1];
1073 	for (i = 0; i < tp->nr_args; i++)
1074 		if (!tp->args[i].type->print(s, tp->args[i].name,
1075 					     data + tp->args[i].offset, field))
1076 			goto out;
1077 
1078 	trace_seq_putc(s, '\n');
1079 
1080  out:
1081 	return trace_handle_return(s);
1082 }
1083 
1084 
1085 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1086 {
1087 	int ret, i;
1088 	struct kprobe_trace_entry_head field;
1089 	struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1090 
1091 	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1092 	/* Set argument names as fields */
1093 	for (i = 0; i < tk->tp.nr_args; i++) {
1094 		struct probe_arg *parg = &tk->tp.args[i];
1095 
1096 		ret = trace_define_field(event_call, parg->type->fmttype,
1097 					 parg->name,
1098 					 sizeof(field) + parg->offset,
1099 					 parg->type->size,
1100 					 parg->type->is_signed,
1101 					 FILTER_OTHER);
1102 		if (ret)
1103 			return ret;
1104 	}
1105 	return 0;
1106 }
1107 
1108 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1109 {
1110 	int ret, i;
1111 	struct kretprobe_trace_entry_head field;
1112 	struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1113 
1114 	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1115 	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1116 	/* Set argument names as fields */
1117 	for (i = 0; i < tk->tp.nr_args; i++) {
1118 		struct probe_arg *parg = &tk->tp.args[i];
1119 
1120 		ret = trace_define_field(event_call, parg->type->fmttype,
1121 					 parg->name,
1122 					 sizeof(field) + parg->offset,
1123 					 parg->type->size,
1124 					 parg->type->is_signed,
1125 					 FILTER_OTHER);
1126 		if (ret)
1127 			return ret;
1128 	}
1129 	return 0;
1130 }
1131 
1132 #ifdef CONFIG_PERF_EVENTS
1133 
1134 /* Kprobe profile handler */
1135 static void
1136 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1137 {
1138 	struct trace_event_call *call = &tk->tp.call;
1139 	struct bpf_prog *prog = call->prog;
1140 	struct kprobe_trace_entry_head *entry;
1141 	struct hlist_head *head;
1142 	int size, __size, dsize;
1143 	int rctx;
1144 
1145 	if (prog && !trace_call_bpf(prog, regs))
1146 		return;
1147 
1148 	head = this_cpu_ptr(call->perf_events);
1149 	if (hlist_empty(head))
1150 		return;
1151 
1152 	dsize = __get_data_size(&tk->tp, regs);
1153 	__size = sizeof(*entry) + tk->tp.size + dsize;
1154 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1155 	size -= sizeof(u32);
1156 
1157 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1158 	if (!entry)
1159 		return;
1160 
1161 	entry->ip = (unsigned long)tk->rp.kp.addr;
1162 	memset(&entry[1], 0, dsize);
1163 	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1164 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1165 			      head, NULL);
1166 }
1167 NOKPROBE_SYMBOL(kprobe_perf_func);
1168 
1169 /* Kretprobe profile handler */
1170 static void
1171 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1172 		    struct pt_regs *regs)
1173 {
1174 	struct trace_event_call *call = &tk->tp.call;
1175 	struct bpf_prog *prog = call->prog;
1176 	struct kretprobe_trace_entry_head *entry;
1177 	struct hlist_head *head;
1178 	int size, __size, dsize;
1179 	int rctx;
1180 
1181 	if (prog && !trace_call_bpf(prog, regs))
1182 		return;
1183 
1184 	head = this_cpu_ptr(call->perf_events);
1185 	if (hlist_empty(head))
1186 		return;
1187 
1188 	dsize = __get_data_size(&tk->tp, regs);
1189 	__size = sizeof(*entry) + tk->tp.size + dsize;
1190 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1191 	size -= sizeof(u32);
1192 
1193 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1194 	if (!entry)
1195 		return;
1196 
1197 	entry->func = (unsigned long)tk->rp.kp.addr;
1198 	entry->ret_ip = (unsigned long)ri->ret_addr;
1199 	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1200 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1201 			      head, NULL);
1202 }
1203 NOKPROBE_SYMBOL(kretprobe_perf_func);
1204 #endif	/* CONFIG_PERF_EVENTS */
1205 
1206 /*
1207  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1208  *
1209  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1210  * lockless, but we can't race with this __init function.
1211  */
1212 static int kprobe_register(struct trace_event_call *event,
1213 			   enum trace_reg type, void *data)
1214 {
1215 	struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1216 	struct trace_event_file *file = data;
1217 
1218 	switch (type) {
1219 	case TRACE_REG_REGISTER:
1220 		return enable_trace_kprobe(tk, file);
1221 	case TRACE_REG_UNREGISTER:
1222 		return disable_trace_kprobe(tk, file);
1223 
1224 #ifdef CONFIG_PERF_EVENTS
1225 	case TRACE_REG_PERF_REGISTER:
1226 		return enable_trace_kprobe(tk, NULL);
1227 	case TRACE_REG_PERF_UNREGISTER:
1228 		return disable_trace_kprobe(tk, NULL);
1229 	case TRACE_REG_PERF_OPEN:
1230 	case TRACE_REG_PERF_CLOSE:
1231 	case TRACE_REG_PERF_ADD:
1232 	case TRACE_REG_PERF_DEL:
1233 		return 0;
1234 #endif
1235 	}
1236 	return 0;
1237 }
1238 
1239 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1240 {
1241 	struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1242 
1243 	raw_cpu_inc(*tk->nhit);
1244 
1245 	if (tk->tp.flags & TP_FLAG_TRACE)
1246 		kprobe_trace_func(tk, regs);
1247 #ifdef CONFIG_PERF_EVENTS
1248 	if (tk->tp.flags & TP_FLAG_PROFILE)
1249 		kprobe_perf_func(tk, regs);
1250 #endif
1251 	return 0;	/* We don't tweek kernel, so just return 0 */
1252 }
1253 NOKPROBE_SYMBOL(kprobe_dispatcher);
1254 
1255 static int
1256 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1257 {
1258 	struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1259 
1260 	raw_cpu_inc(*tk->nhit);
1261 
1262 	if (tk->tp.flags & TP_FLAG_TRACE)
1263 		kretprobe_trace_func(tk, ri, regs);
1264 #ifdef CONFIG_PERF_EVENTS
1265 	if (tk->tp.flags & TP_FLAG_PROFILE)
1266 		kretprobe_perf_func(tk, ri, regs);
1267 #endif
1268 	return 0;	/* We don't tweek kernel, so just return 0 */
1269 }
1270 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1271 
1272 static struct trace_event_functions kretprobe_funcs = {
1273 	.trace		= print_kretprobe_event
1274 };
1275 
1276 static struct trace_event_functions kprobe_funcs = {
1277 	.trace		= print_kprobe_event
1278 };
1279 
1280 static int register_kprobe_event(struct trace_kprobe *tk)
1281 {
1282 	struct trace_event_call *call = &tk->tp.call;
1283 	int ret;
1284 
1285 	/* Initialize trace_event_call */
1286 	INIT_LIST_HEAD(&call->class->fields);
1287 	if (trace_kprobe_is_return(tk)) {
1288 		call->event.funcs = &kretprobe_funcs;
1289 		call->class->define_fields = kretprobe_event_define_fields;
1290 	} else {
1291 		call->event.funcs = &kprobe_funcs;
1292 		call->class->define_fields = kprobe_event_define_fields;
1293 	}
1294 	if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1295 		return -ENOMEM;
1296 	ret = register_trace_event(&call->event);
1297 	if (!ret) {
1298 		kfree(call->print_fmt);
1299 		return -ENODEV;
1300 	}
1301 	call->flags = TRACE_EVENT_FL_KPROBE;
1302 	call->class->reg = kprobe_register;
1303 	call->data = tk;
1304 	ret = trace_add_event_call(call);
1305 	if (ret) {
1306 		pr_info("Failed to register kprobe event: %s\n",
1307 			trace_event_name(call));
1308 		kfree(call->print_fmt);
1309 		unregister_trace_event(&call->event);
1310 	}
1311 	return ret;
1312 }
1313 
1314 static int unregister_kprobe_event(struct trace_kprobe *tk)
1315 {
1316 	int ret;
1317 
1318 	/* tp->event is unregistered in trace_remove_event_call() */
1319 	ret = trace_remove_event_call(&tk->tp.call);
1320 	if (!ret)
1321 		kfree(tk->tp.call.print_fmt);
1322 	return ret;
1323 }
1324 
1325 /* Make a tracefs interface for controlling probe points */
1326 static __init int init_kprobe_trace(void)
1327 {
1328 	struct dentry *d_tracer;
1329 	struct dentry *entry;
1330 
1331 	if (register_module_notifier(&trace_kprobe_module_nb))
1332 		return -EINVAL;
1333 
1334 	d_tracer = tracing_init_dentry();
1335 	if (IS_ERR(d_tracer))
1336 		return 0;
1337 
1338 	entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1339 				    NULL, &kprobe_events_ops);
1340 
1341 	/* Event list interface */
1342 	if (!entry)
1343 		pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1344 
1345 	/* Profile interface */
1346 	entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1347 				    NULL, &kprobe_profile_ops);
1348 
1349 	if (!entry)
1350 		pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1351 	return 0;
1352 }
1353 fs_initcall(init_kprobe_trace);
1354 
1355 
1356 #ifdef CONFIG_FTRACE_STARTUP_TEST
1357 
1358 /*
1359  * The "__used" keeps gcc from removing the function symbol
1360  * from the kallsyms table.
1361  */
1362 static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1363 					       int a4, int a5, int a6)
1364 {
1365 	return a1 + a2 + a3 + a4 + a5 + a6;
1366 }
1367 
1368 static struct trace_event_file *
1369 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1370 {
1371 	struct trace_event_file *file;
1372 
1373 	list_for_each_entry(file, &tr->events, list)
1374 		if (file->event_call == &tk->tp.call)
1375 			return file;
1376 
1377 	return NULL;
1378 }
1379 
1380 /*
1381  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1382  * stage, we can do this lockless.
1383  */
1384 static __init int kprobe_trace_self_tests_init(void)
1385 {
1386 	int ret, warn = 0;
1387 	int (*target)(int, int, int, int, int, int);
1388 	struct trace_kprobe *tk;
1389 	struct trace_event_file *file;
1390 
1391 	if (tracing_is_disabled())
1392 		return -ENODEV;
1393 
1394 	target = kprobe_trace_selftest_target;
1395 
1396 	pr_info("Testing kprobe tracing: ");
1397 
1398 	ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1399 				  "$stack $stack0 +0($stack)",
1400 				  create_trace_kprobe);
1401 	if (WARN_ON_ONCE(ret)) {
1402 		pr_warn("error on probing function entry.\n");
1403 		warn++;
1404 	} else {
1405 		/* Enable trace point */
1406 		tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1407 		if (WARN_ON_ONCE(tk == NULL)) {
1408 			pr_warn("error on getting new probe.\n");
1409 			warn++;
1410 		} else {
1411 			file = find_trace_probe_file(tk, top_trace_array());
1412 			if (WARN_ON_ONCE(file == NULL)) {
1413 				pr_warn("error on getting probe file.\n");
1414 				warn++;
1415 			} else
1416 				enable_trace_kprobe(tk, file);
1417 		}
1418 	}
1419 
1420 	ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1421 				  "$retval", create_trace_kprobe);
1422 	if (WARN_ON_ONCE(ret)) {
1423 		pr_warn("error on probing function return.\n");
1424 		warn++;
1425 	} else {
1426 		/* Enable trace point */
1427 		tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1428 		if (WARN_ON_ONCE(tk == NULL)) {
1429 			pr_warn("error on getting 2nd new probe.\n");
1430 			warn++;
1431 		} else {
1432 			file = find_trace_probe_file(tk, top_trace_array());
1433 			if (WARN_ON_ONCE(file == NULL)) {
1434 				pr_warn("error on getting probe file.\n");
1435 				warn++;
1436 			} else
1437 				enable_trace_kprobe(tk, file);
1438 		}
1439 	}
1440 
1441 	if (warn)
1442 		goto end;
1443 
1444 	ret = target(1, 2, 3, 4, 5, 6);
1445 
1446 	/* Disable trace points before removing it */
1447 	tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1448 	if (WARN_ON_ONCE(tk == NULL)) {
1449 		pr_warn("error on getting test probe.\n");
1450 		warn++;
1451 	} else {
1452 		file = find_trace_probe_file(tk, top_trace_array());
1453 		if (WARN_ON_ONCE(file == NULL)) {
1454 			pr_warn("error on getting probe file.\n");
1455 			warn++;
1456 		} else
1457 			disable_trace_kprobe(tk, file);
1458 	}
1459 
1460 	tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1461 	if (WARN_ON_ONCE(tk == NULL)) {
1462 		pr_warn("error on getting 2nd test probe.\n");
1463 		warn++;
1464 	} else {
1465 		file = find_trace_probe_file(tk, top_trace_array());
1466 		if (WARN_ON_ONCE(file == NULL)) {
1467 			pr_warn("error on getting probe file.\n");
1468 			warn++;
1469 		} else
1470 			disable_trace_kprobe(tk, file);
1471 	}
1472 
1473 	ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1474 	if (WARN_ON_ONCE(ret)) {
1475 		pr_warn("error on deleting a probe.\n");
1476 		warn++;
1477 	}
1478 
1479 	ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1480 	if (WARN_ON_ONCE(ret)) {
1481 		pr_warn("error on deleting a probe.\n");
1482 		warn++;
1483 	}
1484 
1485 end:
1486 	release_all_trace_kprobes();
1487 	if (warn)
1488 		pr_cont("NG: Some tests are failed. Please check them.\n");
1489 	else
1490 		pr_cont("OK\n");
1491 	return 0;
1492 }
1493 
1494 late_initcall(kprobe_trace_self_tests_init);
1495 
1496 #endif
1497