xref: /linux/kernel/trace/trace_kprobe.c (revision 6ebe6dbd6886af07b102aca42e44edbee94a22d9)
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 #define pr_fmt(fmt)	"trace_kprobe: " fmt
20 
21 #include <linux/module.h>
22 #include <linux/uaccess.h>
23 #include <linux/rculist.h>
24 
25 #include "trace_probe.h"
26 
27 #define KPROBE_EVENT_SYSTEM "kprobes"
28 #define KRETPROBE_MAXACTIVE_MAX 4096
29 
30 /**
31  * Kprobe event core functions
32  */
33 struct trace_kprobe {
34 	struct list_head	list;
35 	struct kretprobe	rp;	/* Use rp.kp for kprobe use */
36 	unsigned long __percpu *nhit;
37 	const char		*symbol;	/* symbol name */
38 	struct trace_probe	tp;
39 };
40 
41 #define SIZEOF_TRACE_KPROBE(n)				\
42 	(offsetof(struct trace_kprobe, tp.args) +	\
43 	(sizeof(struct probe_arg) * (n)))
44 
45 DEFINE_PER_CPU(int, bpf_kprobe_override);
46 
47 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
48 {
49 	return tk->rp.handler != NULL;
50 }
51 
52 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
53 {
54 	return tk->symbol ? tk->symbol : "unknown";
55 }
56 
57 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
58 {
59 	return tk->rp.kp.offset;
60 }
61 
62 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
63 {
64 	return !!(kprobe_gone(&tk->rp.kp));
65 }
66 
67 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
68 						 struct module *mod)
69 {
70 	int len = strlen(mod->name);
71 	const char *name = trace_kprobe_symbol(tk);
72 	return strncmp(mod->name, name, len) == 0 && name[len] == ':';
73 }
74 
75 static nokprobe_inline bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
76 {
77 	return !!strchr(trace_kprobe_symbol(tk), ':');
78 }
79 
80 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
81 {
82 	unsigned long nhit = 0;
83 	int cpu;
84 
85 	for_each_possible_cpu(cpu)
86 		nhit += *per_cpu_ptr(tk->nhit, cpu);
87 
88 	return nhit;
89 }
90 
91 int trace_kprobe_ftrace(struct trace_event_call *call)
92 {
93 	struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
94 	return kprobe_ftrace(&tk->rp.kp);
95 }
96 
97 int trace_kprobe_error_injectable(struct trace_event_call *call)
98 {
99 	struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
100 	unsigned long addr;
101 
102 	if (tk->symbol) {
103 		addr = (unsigned long)
104 			kallsyms_lookup_name(trace_kprobe_symbol(tk));
105 		addr += tk->rp.kp.offset;
106 	} else {
107 		addr = (unsigned long)tk->rp.kp.addr;
108 	}
109 	return within_kprobe_error_injection_list(addr);
110 }
111 
112 static int register_kprobe_event(struct trace_kprobe *tk);
113 static int unregister_kprobe_event(struct trace_kprobe *tk);
114 
115 static DEFINE_MUTEX(probe_lock);
116 static LIST_HEAD(probe_list);
117 
118 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
119 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
120 				struct pt_regs *regs);
121 
122 /* Memory fetching by symbol */
123 struct symbol_cache {
124 	char		*symbol;
125 	long		offset;
126 	unsigned long	addr;
127 };
128 
129 unsigned long update_symbol_cache(struct symbol_cache *sc)
130 {
131 	sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
132 
133 	if (sc->addr)
134 		sc->addr += sc->offset;
135 
136 	return sc->addr;
137 }
138 
139 void free_symbol_cache(struct symbol_cache *sc)
140 {
141 	kfree(sc->symbol);
142 	kfree(sc);
143 }
144 
145 struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
146 {
147 	struct symbol_cache *sc;
148 
149 	if (!sym || strlen(sym) == 0)
150 		return NULL;
151 
152 	sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
153 	if (!sc)
154 		return NULL;
155 
156 	sc->symbol = kstrdup(sym, GFP_KERNEL);
157 	if (!sc->symbol) {
158 		kfree(sc);
159 		return NULL;
160 	}
161 	sc->offset = offset;
162 	update_symbol_cache(sc);
163 
164 	return sc;
165 }
166 
167 /*
168  * Kprobes-specific fetch functions
169  */
170 #define DEFINE_FETCH_stack(type)					\
171 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,		\
172 					  void *offset, void *dest)	\
173 {									\
174 	*(type *)dest = (type)regs_get_kernel_stack_nth(regs,		\
175 				(unsigned int)((unsigned long)offset));	\
176 }									\
177 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
178 
179 DEFINE_BASIC_FETCH_FUNCS(stack)
180 /* No string on the stack entry */
181 #define fetch_stack_string	NULL
182 #define fetch_stack_string_size	NULL
183 
184 #define DEFINE_FETCH_memory(type)					\
185 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,		\
186 					  void *addr, void *dest)	\
187 {									\
188 	type retval;							\
189 	if (probe_kernel_address(addr, retval))				\
190 		*(type *)dest = 0;					\
191 	else								\
192 		*(type *)dest = retval;					\
193 }									\
194 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
195 
196 DEFINE_BASIC_FETCH_FUNCS(memory)
197 /*
198  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
199  * length and relative data location.
200  */
201 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
202 					    void *addr, void *dest)
203 {
204 	int maxlen = get_rloc_len(*(u32 *)dest);
205 	u8 *dst = get_rloc_data(dest);
206 	long ret;
207 
208 	if (!maxlen)
209 		return;
210 
211 	/*
212 	 * Try to get string again, since the string can be changed while
213 	 * probing.
214 	 */
215 	ret = strncpy_from_unsafe(dst, addr, maxlen);
216 
217 	if (ret < 0) {	/* Failed to fetch string */
218 		dst[0] = '\0';
219 		*(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
220 	} else {
221 		*(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
222 	}
223 }
224 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
225 
226 /* Return the length of string -- including null terminal byte */
227 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
228 						 void *addr, void *dest)
229 {
230 	mm_segment_t old_fs;
231 	int ret, len = 0;
232 	u8 c;
233 
234 	old_fs = get_fs();
235 	set_fs(KERNEL_DS);
236 	pagefault_disable();
237 
238 	do {
239 		ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
240 		len++;
241 	} while (c && ret == 0 && len < MAX_STRING_SIZE);
242 
243 	pagefault_enable();
244 	set_fs(old_fs);
245 
246 	if (ret < 0)	/* Failed to check the length */
247 		*(u32 *)dest = 0;
248 	else
249 		*(u32 *)dest = len;
250 }
251 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
252 
253 #define DEFINE_FETCH_symbol(type)					\
254 void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
255 {									\
256 	struct symbol_cache *sc = data;					\
257 	if (sc->addr)							\
258 		fetch_memory_##type(regs, (void *)sc->addr, dest);	\
259 	else								\
260 		*(type *)dest = 0;					\
261 }									\
262 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
263 
264 DEFINE_BASIC_FETCH_FUNCS(symbol)
265 DEFINE_FETCH_symbol(string)
266 DEFINE_FETCH_symbol(string_size)
267 
268 /* kprobes don't support file_offset fetch methods */
269 #define fetch_file_offset_u8		NULL
270 #define fetch_file_offset_u16		NULL
271 #define fetch_file_offset_u32		NULL
272 #define fetch_file_offset_u64		NULL
273 #define fetch_file_offset_string	NULL
274 #define fetch_file_offset_string_size	NULL
275 
276 /* Fetch type information table */
277 static const struct fetch_type kprobes_fetch_type_table[] = {
278 	/* Special types */
279 	[FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
280 					sizeof(u32), 1, "__data_loc char[]"),
281 	[FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
282 					string_size, sizeof(u32), 0, "u32"),
283 	/* Basic types */
284 	ASSIGN_FETCH_TYPE(u8,  u8,  0),
285 	ASSIGN_FETCH_TYPE(u16, u16, 0),
286 	ASSIGN_FETCH_TYPE(u32, u32, 0),
287 	ASSIGN_FETCH_TYPE(u64, u64, 0),
288 	ASSIGN_FETCH_TYPE(s8,  u8,  1),
289 	ASSIGN_FETCH_TYPE(s16, u16, 1),
290 	ASSIGN_FETCH_TYPE(s32, u32, 1),
291 	ASSIGN_FETCH_TYPE(s64, u64, 1),
292 	ASSIGN_FETCH_TYPE_ALIAS(x8,  u8,  u8,  0),
293 	ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
294 	ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
295 	ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
296 
297 	ASSIGN_FETCH_TYPE_END
298 };
299 
300 /*
301  * Allocate new trace_probe and initialize it (including kprobes).
302  */
303 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
304 					     const char *event,
305 					     void *addr,
306 					     const char *symbol,
307 					     unsigned long offs,
308 					     int maxactive,
309 					     int nargs, bool is_return)
310 {
311 	struct trace_kprobe *tk;
312 	int ret = -ENOMEM;
313 
314 	tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
315 	if (!tk)
316 		return ERR_PTR(ret);
317 
318 	tk->nhit = alloc_percpu(unsigned long);
319 	if (!tk->nhit)
320 		goto error;
321 
322 	if (symbol) {
323 		tk->symbol = kstrdup(symbol, GFP_KERNEL);
324 		if (!tk->symbol)
325 			goto error;
326 		tk->rp.kp.symbol_name = tk->symbol;
327 		tk->rp.kp.offset = offs;
328 	} else
329 		tk->rp.kp.addr = addr;
330 
331 	if (is_return)
332 		tk->rp.handler = kretprobe_dispatcher;
333 	else
334 		tk->rp.kp.pre_handler = kprobe_dispatcher;
335 
336 	tk->rp.maxactive = maxactive;
337 
338 	if (!event || !is_good_name(event)) {
339 		ret = -EINVAL;
340 		goto error;
341 	}
342 
343 	tk->tp.call.class = &tk->tp.class;
344 	tk->tp.call.name = kstrdup(event, GFP_KERNEL);
345 	if (!tk->tp.call.name)
346 		goto error;
347 
348 	if (!group || !is_good_name(group)) {
349 		ret = -EINVAL;
350 		goto error;
351 	}
352 
353 	tk->tp.class.system = kstrdup(group, GFP_KERNEL);
354 	if (!tk->tp.class.system)
355 		goto error;
356 
357 	INIT_LIST_HEAD(&tk->list);
358 	INIT_LIST_HEAD(&tk->tp.files);
359 	return tk;
360 error:
361 	kfree(tk->tp.call.name);
362 	kfree(tk->symbol);
363 	free_percpu(tk->nhit);
364 	kfree(tk);
365 	return ERR_PTR(ret);
366 }
367 
368 static void free_trace_kprobe(struct trace_kprobe *tk)
369 {
370 	int i;
371 
372 	for (i = 0; i < tk->tp.nr_args; i++)
373 		traceprobe_free_probe_arg(&tk->tp.args[i]);
374 
375 	kfree(tk->tp.call.class->system);
376 	kfree(tk->tp.call.name);
377 	kfree(tk->symbol);
378 	free_percpu(tk->nhit);
379 	kfree(tk);
380 }
381 
382 static struct trace_kprobe *find_trace_kprobe(const char *event,
383 					      const char *group)
384 {
385 	struct trace_kprobe *tk;
386 
387 	list_for_each_entry(tk, &probe_list, list)
388 		if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
389 		    strcmp(tk->tp.call.class->system, group) == 0)
390 			return tk;
391 	return NULL;
392 }
393 
394 /*
395  * Enable trace_probe
396  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
397  */
398 static int
399 enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
400 {
401 	int ret = 0;
402 
403 	if (file) {
404 		struct event_file_link *link;
405 
406 		link = kmalloc(sizeof(*link), GFP_KERNEL);
407 		if (!link) {
408 			ret = -ENOMEM;
409 			goto out;
410 		}
411 
412 		link->file = file;
413 		list_add_tail_rcu(&link->list, &tk->tp.files);
414 
415 		tk->tp.flags |= TP_FLAG_TRACE;
416 	} else
417 		tk->tp.flags |= TP_FLAG_PROFILE;
418 
419 	if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
420 		if (trace_kprobe_is_return(tk))
421 			ret = enable_kretprobe(&tk->rp);
422 		else
423 			ret = enable_kprobe(&tk->rp.kp);
424 	}
425  out:
426 	return ret;
427 }
428 
429 /*
430  * Disable trace_probe
431  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
432  */
433 static int
434 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
435 {
436 	struct event_file_link *link = NULL;
437 	int wait = 0;
438 	int ret = 0;
439 
440 	if (file) {
441 		link = find_event_file_link(&tk->tp, file);
442 		if (!link) {
443 			ret = -EINVAL;
444 			goto out;
445 		}
446 
447 		list_del_rcu(&link->list);
448 		wait = 1;
449 		if (!list_empty(&tk->tp.files))
450 			goto out;
451 
452 		tk->tp.flags &= ~TP_FLAG_TRACE;
453 	} else
454 		tk->tp.flags &= ~TP_FLAG_PROFILE;
455 
456 	if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
457 		if (trace_kprobe_is_return(tk))
458 			disable_kretprobe(&tk->rp);
459 		else
460 			disable_kprobe(&tk->rp.kp);
461 		wait = 1;
462 	}
463  out:
464 	if (wait) {
465 		/*
466 		 * Synchronize with kprobe_trace_func/kretprobe_trace_func
467 		 * to ensure disabled (all running handlers are finished).
468 		 * This is not only for kfree(), but also the caller,
469 		 * trace_remove_event_call() supposes it for releasing
470 		 * event_call related objects, which will be accessed in
471 		 * the kprobe_trace_func/kretprobe_trace_func.
472 		 */
473 		synchronize_sched();
474 		kfree(link);	/* Ignored if link == NULL */
475 	}
476 
477 	return ret;
478 }
479 
480 /* Internal register function - just handle k*probes and flags */
481 static int __register_trace_kprobe(struct trace_kprobe *tk)
482 {
483 	int i, ret;
484 
485 	if (trace_probe_is_registered(&tk->tp))
486 		return -EINVAL;
487 
488 	for (i = 0; i < tk->tp.nr_args; i++)
489 		traceprobe_update_arg(&tk->tp.args[i]);
490 
491 	/* Set/clear disabled flag according to tp->flag */
492 	if (trace_probe_is_enabled(&tk->tp))
493 		tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
494 	else
495 		tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
496 
497 	if (trace_kprobe_is_return(tk))
498 		ret = register_kretprobe(&tk->rp);
499 	else
500 		ret = register_kprobe(&tk->rp.kp);
501 
502 	if (ret == 0)
503 		tk->tp.flags |= TP_FLAG_REGISTERED;
504 	else {
505 		pr_warn("Could not insert probe at %s+%lu: %d\n",
506 			trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
507 		if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
508 			pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
509 			ret = 0;
510 		} else if (ret == -EILSEQ) {
511 			pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
512 				tk->rp.kp.addr);
513 			ret = -EINVAL;
514 		}
515 	}
516 
517 	return ret;
518 }
519 
520 /* Internal unregister function - just handle k*probes and flags */
521 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
522 {
523 	if (trace_probe_is_registered(&tk->tp)) {
524 		if (trace_kprobe_is_return(tk))
525 			unregister_kretprobe(&tk->rp);
526 		else
527 			unregister_kprobe(&tk->rp.kp);
528 		tk->tp.flags &= ~TP_FLAG_REGISTERED;
529 		/* Cleanup kprobe for reuse */
530 		if (tk->rp.kp.symbol_name)
531 			tk->rp.kp.addr = NULL;
532 	}
533 }
534 
535 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
536 static int unregister_trace_kprobe(struct trace_kprobe *tk)
537 {
538 	/* Enabled event can not be unregistered */
539 	if (trace_probe_is_enabled(&tk->tp))
540 		return -EBUSY;
541 
542 	/* Will fail if probe is being used by ftrace or perf */
543 	if (unregister_kprobe_event(tk))
544 		return -EBUSY;
545 
546 	__unregister_trace_kprobe(tk);
547 	list_del(&tk->list);
548 
549 	return 0;
550 }
551 
552 /* Register a trace_probe and probe_event */
553 static int register_trace_kprobe(struct trace_kprobe *tk)
554 {
555 	struct trace_kprobe *old_tk;
556 	int ret;
557 
558 	mutex_lock(&probe_lock);
559 
560 	/* Delete old (same name) event if exist */
561 	old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
562 			tk->tp.call.class->system);
563 	if (old_tk) {
564 		ret = unregister_trace_kprobe(old_tk);
565 		if (ret < 0)
566 			goto end;
567 		free_trace_kprobe(old_tk);
568 	}
569 
570 	/* Register new event */
571 	ret = register_kprobe_event(tk);
572 	if (ret) {
573 		pr_warn("Failed to register probe event(%d)\n", ret);
574 		goto end;
575 	}
576 
577 	/* Register k*probe */
578 	ret = __register_trace_kprobe(tk);
579 	if (ret < 0)
580 		unregister_kprobe_event(tk);
581 	else
582 		list_add_tail(&tk->list, &probe_list);
583 
584 end:
585 	mutex_unlock(&probe_lock);
586 	return ret;
587 }
588 
589 /* Module notifier call back, checking event on the module */
590 static int trace_kprobe_module_callback(struct notifier_block *nb,
591 				       unsigned long val, void *data)
592 {
593 	struct module *mod = data;
594 	struct trace_kprobe *tk;
595 	int ret;
596 
597 	if (val != MODULE_STATE_COMING)
598 		return NOTIFY_DONE;
599 
600 	/* Update probes on coming module */
601 	mutex_lock(&probe_lock);
602 	list_for_each_entry(tk, &probe_list, list) {
603 		if (trace_kprobe_within_module(tk, mod)) {
604 			/* Don't need to check busy - this should have gone. */
605 			__unregister_trace_kprobe(tk);
606 			ret = __register_trace_kprobe(tk);
607 			if (ret)
608 				pr_warn("Failed to re-register probe %s on %s: %d\n",
609 					trace_event_name(&tk->tp.call),
610 					mod->name, ret);
611 		}
612 	}
613 	mutex_unlock(&probe_lock);
614 
615 	return NOTIFY_DONE;
616 }
617 
618 static struct notifier_block trace_kprobe_module_nb = {
619 	.notifier_call = trace_kprobe_module_callback,
620 	.priority = 1	/* Invoked after kprobe module callback */
621 };
622 
623 /* Convert certain expected symbols into '_' when generating event names */
624 static inline void sanitize_event_name(char *name)
625 {
626 	while (*name++ != '\0')
627 		if (*name == ':' || *name == '.')
628 			*name = '_';
629 }
630 
631 static int create_trace_kprobe(int argc, char **argv)
632 {
633 	/*
634 	 * Argument syntax:
635 	 *  - Add kprobe:
636 	 *      p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
637 	 *  - Add kretprobe:
638 	 *      r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
639 	 * Fetch args:
640 	 *  $retval	: fetch return value
641 	 *  $stack	: fetch stack address
642 	 *  $stackN	: fetch Nth of stack (N:0-)
643 	 *  $comm       : fetch current task comm
644 	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
645 	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
646 	 *  %REG	: fetch register REG
647 	 * Dereferencing memory fetch:
648 	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
649 	 * Alias name of args:
650 	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
651 	 * Type of args:
652 	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
653 	 */
654 	struct trace_kprobe *tk;
655 	int i, ret = 0;
656 	bool is_return = false, is_delete = false;
657 	char *symbol = NULL, *event = NULL, *group = NULL;
658 	int maxactive = 0;
659 	char *arg;
660 	unsigned long offset = 0;
661 	void *addr = NULL;
662 	char buf[MAX_EVENT_NAME_LEN];
663 
664 	/* argc must be >= 1 */
665 	if (argv[0][0] == 'p')
666 		is_return = false;
667 	else if (argv[0][0] == 'r')
668 		is_return = true;
669 	else if (argv[0][0] == '-')
670 		is_delete = true;
671 	else {
672 		pr_info("Probe definition must be started with 'p', 'r' or"
673 			" '-'.\n");
674 		return -EINVAL;
675 	}
676 
677 	event = strchr(&argv[0][1], ':');
678 	if (event) {
679 		event[0] = '\0';
680 		event++;
681 	}
682 	if (is_return && isdigit(argv[0][1])) {
683 		ret = kstrtouint(&argv[0][1], 0, &maxactive);
684 		if (ret) {
685 			pr_info("Failed to parse maxactive.\n");
686 			return ret;
687 		}
688 		/* kretprobes instances are iterated over via a list. The
689 		 * maximum should stay reasonable.
690 		 */
691 		if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
692 			pr_info("Maxactive is too big (%d > %d).\n",
693 				maxactive, KRETPROBE_MAXACTIVE_MAX);
694 			return -E2BIG;
695 		}
696 	}
697 
698 	if (event) {
699 		if (strchr(event, '/')) {
700 			group = event;
701 			event = strchr(group, '/') + 1;
702 			event[-1] = '\0';
703 			if (strlen(group) == 0) {
704 				pr_info("Group name is not specified\n");
705 				return -EINVAL;
706 			}
707 		}
708 		if (strlen(event) == 0) {
709 			pr_info("Event name is not specified\n");
710 			return -EINVAL;
711 		}
712 	}
713 	if (!group)
714 		group = KPROBE_EVENT_SYSTEM;
715 
716 	if (is_delete) {
717 		if (!event) {
718 			pr_info("Delete command needs an event name.\n");
719 			return -EINVAL;
720 		}
721 		mutex_lock(&probe_lock);
722 		tk = find_trace_kprobe(event, group);
723 		if (!tk) {
724 			mutex_unlock(&probe_lock);
725 			pr_info("Event %s/%s doesn't exist.\n", group, event);
726 			return -ENOENT;
727 		}
728 		/* delete an event */
729 		ret = unregister_trace_kprobe(tk);
730 		if (ret == 0)
731 			free_trace_kprobe(tk);
732 		mutex_unlock(&probe_lock);
733 		return ret;
734 	}
735 
736 	if (argc < 2) {
737 		pr_info("Probe point is not specified.\n");
738 		return -EINVAL;
739 	}
740 
741 	/* try to parse an address. if that fails, try to read the
742 	 * input as a symbol. */
743 	if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
744 		/* a symbol specified */
745 		symbol = argv[1];
746 		/* TODO: support .init module functions */
747 		ret = traceprobe_split_symbol_offset(symbol, &offset);
748 		if (ret) {
749 			pr_info("Failed to parse either an address or a symbol.\n");
750 			return ret;
751 		}
752 		if (offset && is_return &&
753 		    !kprobe_on_func_entry(NULL, symbol, offset)) {
754 			pr_info("Given offset is not valid for return probe.\n");
755 			return -EINVAL;
756 		}
757 	}
758 	argc -= 2; argv += 2;
759 
760 	/* setup a probe */
761 	if (!event) {
762 		/* Make a new event name */
763 		if (symbol)
764 			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
765 				 is_return ? 'r' : 'p', symbol, offset);
766 		else
767 			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
768 				 is_return ? 'r' : 'p', addr);
769 		sanitize_event_name(buf);
770 		event = buf;
771 	}
772 	tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
773 			       argc, is_return);
774 	if (IS_ERR(tk)) {
775 		pr_info("Failed to allocate trace_probe.(%d)\n",
776 			(int)PTR_ERR(tk));
777 		return PTR_ERR(tk);
778 	}
779 
780 	/* parse arguments */
781 	ret = 0;
782 	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
783 		struct probe_arg *parg = &tk->tp.args[i];
784 
785 		/* Increment count for freeing args in error case */
786 		tk->tp.nr_args++;
787 
788 		/* Parse argument name */
789 		arg = strchr(argv[i], '=');
790 		if (arg) {
791 			*arg++ = '\0';
792 			parg->name = kstrdup(argv[i], GFP_KERNEL);
793 		} else {
794 			arg = argv[i];
795 			/* If argument name is omitted, set "argN" */
796 			snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
797 			parg->name = kstrdup(buf, GFP_KERNEL);
798 		}
799 
800 		if (!parg->name) {
801 			pr_info("Failed to allocate argument[%d] name.\n", i);
802 			ret = -ENOMEM;
803 			goto error;
804 		}
805 
806 		if (!is_good_name(parg->name)) {
807 			pr_info("Invalid argument[%d] name: %s\n",
808 				i, parg->name);
809 			ret = -EINVAL;
810 			goto error;
811 		}
812 
813 		if (traceprobe_conflict_field_name(parg->name,
814 							tk->tp.args, i)) {
815 			pr_info("Argument[%d] name '%s' conflicts with "
816 				"another field.\n", i, argv[i]);
817 			ret = -EINVAL;
818 			goto error;
819 		}
820 
821 		/* Parse fetch argument */
822 		ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
823 						is_return, true,
824 						kprobes_fetch_type_table);
825 		if (ret) {
826 			pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
827 			goto error;
828 		}
829 	}
830 
831 	ret = register_trace_kprobe(tk);
832 	if (ret)
833 		goto error;
834 	return 0;
835 
836 error:
837 	free_trace_kprobe(tk);
838 	return ret;
839 }
840 
841 static int release_all_trace_kprobes(void)
842 {
843 	struct trace_kprobe *tk;
844 	int ret = 0;
845 
846 	mutex_lock(&probe_lock);
847 	/* Ensure no probe is in use. */
848 	list_for_each_entry(tk, &probe_list, list)
849 		if (trace_probe_is_enabled(&tk->tp)) {
850 			ret = -EBUSY;
851 			goto end;
852 		}
853 	/* TODO: Use batch unregistration */
854 	while (!list_empty(&probe_list)) {
855 		tk = list_entry(probe_list.next, struct trace_kprobe, list);
856 		ret = unregister_trace_kprobe(tk);
857 		if (ret)
858 			goto end;
859 		free_trace_kprobe(tk);
860 	}
861 
862 end:
863 	mutex_unlock(&probe_lock);
864 
865 	return ret;
866 }
867 
868 /* Probes listing interfaces */
869 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
870 {
871 	mutex_lock(&probe_lock);
872 	return seq_list_start(&probe_list, *pos);
873 }
874 
875 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
876 {
877 	return seq_list_next(v, &probe_list, pos);
878 }
879 
880 static void probes_seq_stop(struct seq_file *m, void *v)
881 {
882 	mutex_unlock(&probe_lock);
883 }
884 
885 static int probes_seq_show(struct seq_file *m, void *v)
886 {
887 	struct trace_kprobe *tk = v;
888 	int i;
889 
890 	seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
891 	seq_printf(m, ":%s/%s", tk->tp.call.class->system,
892 			trace_event_name(&tk->tp.call));
893 
894 	if (!tk->symbol)
895 		seq_printf(m, " 0x%p", tk->rp.kp.addr);
896 	else if (tk->rp.kp.offset)
897 		seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
898 			   tk->rp.kp.offset);
899 	else
900 		seq_printf(m, " %s", trace_kprobe_symbol(tk));
901 
902 	for (i = 0; i < tk->tp.nr_args; i++)
903 		seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
904 	seq_putc(m, '\n');
905 
906 	return 0;
907 }
908 
909 static const struct seq_operations probes_seq_op = {
910 	.start  = probes_seq_start,
911 	.next   = probes_seq_next,
912 	.stop   = probes_seq_stop,
913 	.show   = probes_seq_show
914 };
915 
916 static int probes_open(struct inode *inode, struct file *file)
917 {
918 	int ret;
919 
920 	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
921 		ret = release_all_trace_kprobes();
922 		if (ret < 0)
923 			return ret;
924 	}
925 
926 	return seq_open(file, &probes_seq_op);
927 }
928 
929 static ssize_t probes_write(struct file *file, const char __user *buffer,
930 			    size_t count, loff_t *ppos)
931 {
932 	return trace_parse_run_command(file, buffer, count, ppos,
933 				       create_trace_kprobe);
934 }
935 
936 static const struct file_operations kprobe_events_ops = {
937 	.owner          = THIS_MODULE,
938 	.open           = probes_open,
939 	.read           = seq_read,
940 	.llseek         = seq_lseek,
941 	.release        = seq_release,
942 	.write		= probes_write,
943 };
944 
945 /* Probes profiling interfaces */
946 static int probes_profile_seq_show(struct seq_file *m, void *v)
947 {
948 	struct trace_kprobe *tk = v;
949 
950 	seq_printf(m, "  %-44s %15lu %15lu\n",
951 		   trace_event_name(&tk->tp.call),
952 		   trace_kprobe_nhit(tk),
953 		   tk->rp.kp.nmissed);
954 
955 	return 0;
956 }
957 
958 static const struct seq_operations profile_seq_op = {
959 	.start  = probes_seq_start,
960 	.next   = probes_seq_next,
961 	.stop   = probes_seq_stop,
962 	.show   = probes_profile_seq_show
963 };
964 
965 static int profile_open(struct inode *inode, struct file *file)
966 {
967 	return seq_open(file, &profile_seq_op);
968 }
969 
970 static const struct file_operations kprobe_profile_ops = {
971 	.owner          = THIS_MODULE,
972 	.open           = profile_open,
973 	.read           = seq_read,
974 	.llseek         = seq_lseek,
975 	.release        = seq_release,
976 };
977 
978 /* Kprobe handler */
979 static nokprobe_inline void
980 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
981 		    struct trace_event_file *trace_file)
982 {
983 	struct kprobe_trace_entry_head *entry;
984 	struct ring_buffer_event *event;
985 	struct ring_buffer *buffer;
986 	int size, dsize, pc;
987 	unsigned long irq_flags;
988 	struct trace_event_call *call = &tk->tp.call;
989 
990 	WARN_ON(call != trace_file->event_call);
991 
992 	if (trace_trigger_soft_disabled(trace_file))
993 		return;
994 
995 	local_save_flags(irq_flags);
996 	pc = preempt_count();
997 
998 	dsize = __get_data_size(&tk->tp, regs);
999 	size = sizeof(*entry) + tk->tp.size + dsize;
1000 
1001 	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1002 						call->event.type,
1003 						size, irq_flags, pc);
1004 	if (!event)
1005 		return;
1006 
1007 	entry = ring_buffer_event_data(event);
1008 	entry->ip = (unsigned long)tk->rp.kp.addr;
1009 	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1010 
1011 	event_trigger_unlock_commit_regs(trace_file, buffer, event,
1012 					 entry, irq_flags, pc, regs);
1013 }
1014 
1015 static void
1016 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1017 {
1018 	struct event_file_link *link;
1019 
1020 	list_for_each_entry_rcu(link, &tk->tp.files, list)
1021 		__kprobe_trace_func(tk, regs, link->file);
1022 }
1023 NOKPROBE_SYMBOL(kprobe_trace_func);
1024 
1025 /* Kretprobe handler */
1026 static nokprobe_inline void
1027 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1028 		       struct pt_regs *regs,
1029 		       struct trace_event_file *trace_file)
1030 {
1031 	struct kretprobe_trace_entry_head *entry;
1032 	struct ring_buffer_event *event;
1033 	struct ring_buffer *buffer;
1034 	int size, pc, dsize;
1035 	unsigned long irq_flags;
1036 	struct trace_event_call *call = &tk->tp.call;
1037 
1038 	WARN_ON(call != trace_file->event_call);
1039 
1040 	if (trace_trigger_soft_disabled(trace_file))
1041 		return;
1042 
1043 	local_save_flags(irq_flags);
1044 	pc = preempt_count();
1045 
1046 	dsize = __get_data_size(&tk->tp, regs);
1047 	size = sizeof(*entry) + tk->tp.size + dsize;
1048 
1049 	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1050 						call->event.type,
1051 						size, irq_flags, pc);
1052 	if (!event)
1053 		return;
1054 
1055 	entry = ring_buffer_event_data(event);
1056 	entry->func = (unsigned long)tk->rp.kp.addr;
1057 	entry->ret_ip = (unsigned long)ri->ret_addr;
1058 	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1059 
1060 	event_trigger_unlock_commit_regs(trace_file, buffer, event,
1061 					 entry, irq_flags, pc, regs);
1062 }
1063 
1064 static void
1065 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1066 		     struct pt_regs *regs)
1067 {
1068 	struct event_file_link *link;
1069 
1070 	list_for_each_entry_rcu(link, &tk->tp.files, list)
1071 		__kretprobe_trace_func(tk, ri, regs, link->file);
1072 }
1073 NOKPROBE_SYMBOL(kretprobe_trace_func);
1074 
1075 /* Event entry printers */
1076 static enum print_line_t
1077 print_kprobe_event(struct trace_iterator *iter, int flags,
1078 		   struct trace_event *event)
1079 {
1080 	struct kprobe_trace_entry_head *field;
1081 	struct trace_seq *s = &iter->seq;
1082 	struct trace_probe *tp;
1083 	u8 *data;
1084 	int i;
1085 
1086 	field = (struct kprobe_trace_entry_head *)iter->ent;
1087 	tp = container_of(event, struct trace_probe, call.event);
1088 
1089 	trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1090 
1091 	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1092 		goto out;
1093 
1094 	trace_seq_putc(s, ')');
1095 
1096 	data = (u8 *)&field[1];
1097 	for (i = 0; i < tp->nr_args; i++)
1098 		if (!tp->args[i].type->print(s, tp->args[i].name,
1099 					     data + tp->args[i].offset, field))
1100 			goto out;
1101 
1102 	trace_seq_putc(s, '\n');
1103  out:
1104 	return trace_handle_return(s);
1105 }
1106 
1107 static enum print_line_t
1108 print_kretprobe_event(struct trace_iterator *iter, int flags,
1109 		      struct trace_event *event)
1110 {
1111 	struct kretprobe_trace_entry_head *field;
1112 	struct trace_seq *s = &iter->seq;
1113 	struct trace_probe *tp;
1114 	u8 *data;
1115 	int i;
1116 
1117 	field = (struct kretprobe_trace_entry_head *)iter->ent;
1118 	tp = container_of(event, struct trace_probe, call.event);
1119 
1120 	trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1121 
1122 	if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1123 		goto out;
1124 
1125 	trace_seq_puts(s, " <- ");
1126 
1127 	if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1128 		goto out;
1129 
1130 	trace_seq_putc(s, ')');
1131 
1132 	data = (u8 *)&field[1];
1133 	for (i = 0; i < tp->nr_args; i++)
1134 		if (!tp->args[i].type->print(s, tp->args[i].name,
1135 					     data + tp->args[i].offset, field))
1136 			goto out;
1137 
1138 	trace_seq_putc(s, '\n');
1139 
1140  out:
1141 	return trace_handle_return(s);
1142 }
1143 
1144 
1145 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1146 {
1147 	int ret, i;
1148 	struct kprobe_trace_entry_head field;
1149 	struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1150 
1151 	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1152 	/* Set argument names as fields */
1153 	for (i = 0; i < tk->tp.nr_args; i++) {
1154 		struct probe_arg *parg = &tk->tp.args[i];
1155 
1156 		ret = trace_define_field(event_call, parg->type->fmttype,
1157 					 parg->name,
1158 					 sizeof(field) + parg->offset,
1159 					 parg->type->size,
1160 					 parg->type->is_signed,
1161 					 FILTER_OTHER);
1162 		if (ret)
1163 			return ret;
1164 	}
1165 	return 0;
1166 }
1167 
1168 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1169 {
1170 	int ret, i;
1171 	struct kretprobe_trace_entry_head field;
1172 	struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1173 
1174 	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1175 	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1176 	/* Set argument names as fields */
1177 	for (i = 0; i < tk->tp.nr_args; i++) {
1178 		struct probe_arg *parg = &tk->tp.args[i];
1179 
1180 		ret = trace_define_field(event_call, parg->type->fmttype,
1181 					 parg->name,
1182 					 sizeof(field) + parg->offset,
1183 					 parg->type->size,
1184 					 parg->type->is_signed,
1185 					 FILTER_OTHER);
1186 		if (ret)
1187 			return ret;
1188 	}
1189 	return 0;
1190 }
1191 
1192 #ifdef CONFIG_PERF_EVENTS
1193 
1194 /* Kprobe profile handler */
1195 static int
1196 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1197 {
1198 	struct trace_event_call *call = &tk->tp.call;
1199 	struct kprobe_trace_entry_head *entry;
1200 	struct hlist_head *head;
1201 	int size, __size, dsize;
1202 	int rctx;
1203 
1204 	if (bpf_prog_array_valid(call)) {
1205 		int ret;
1206 
1207 		ret = trace_call_bpf(call, regs);
1208 
1209 		/*
1210 		 * We need to check and see if we modified the pc of the
1211 		 * pt_regs, and if so clear the kprobe and return 1 so that we
1212 		 * don't do the instruction skipping.  Also reset our state so
1213 		 * we are clean the next pass through.
1214 		 */
1215 		if (__this_cpu_read(bpf_kprobe_override)) {
1216 			__this_cpu_write(bpf_kprobe_override, 0);
1217 			reset_current_kprobe();
1218 			return 1;
1219 		}
1220 		if (!ret)
1221 			return 0;
1222 	}
1223 
1224 	head = this_cpu_ptr(call->perf_events);
1225 	if (hlist_empty(head))
1226 		return 0;
1227 
1228 	dsize = __get_data_size(&tk->tp, regs);
1229 	__size = sizeof(*entry) + tk->tp.size + dsize;
1230 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1231 	size -= sizeof(u32);
1232 
1233 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1234 	if (!entry)
1235 		return 0;
1236 
1237 	entry->ip = (unsigned long)tk->rp.kp.addr;
1238 	memset(&entry[1], 0, dsize);
1239 	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1240 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1241 			      head, NULL);
1242 	return 0;
1243 }
1244 NOKPROBE_SYMBOL(kprobe_perf_func);
1245 
1246 /* Kretprobe profile handler */
1247 static void
1248 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1249 		    struct pt_regs *regs)
1250 {
1251 	struct trace_event_call *call = &tk->tp.call;
1252 	struct kretprobe_trace_entry_head *entry;
1253 	struct hlist_head *head;
1254 	int size, __size, dsize;
1255 	int rctx;
1256 
1257 	if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1258 		return;
1259 
1260 	head = this_cpu_ptr(call->perf_events);
1261 	if (hlist_empty(head))
1262 		return;
1263 
1264 	dsize = __get_data_size(&tk->tp, regs);
1265 	__size = sizeof(*entry) + tk->tp.size + dsize;
1266 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1267 	size -= sizeof(u32);
1268 
1269 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1270 	if (!entry)
1271 		return;
1272 
1273 	entry->func = (unsigned long)tk->rp.kp.addr;
1274 	entry->ret_ip = (unsigned long)ri->ret_addr;
1275 	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1276 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1277 			      head, NULL);
1278 }
1279 NOKPROBE_SYMBOL(kretprobe_perf_func);
1280 #endif	/* CONFIG_PERF_EVENTS */
1281 
1282 /*
1283  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1284  *
1285  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1286  * lockless, but we can't race with this __init function.
1287  */
1288 static int kprobe_register(struct trace_event_call *event,
1289 			   enum trace_reg type, void *data)
1290 {
1291 	struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1292 	struct trace_event_file *file = data;
1293 
1294 	switch (type) {
1295 	case TRACE_REG_REGISTER:
1296 		return enable_trace_kprobe(tk, file);
1297 	case TRACE_REG_UNREGISTER:
1298 		return disable_trace_kprobe(tk, file);
1299 
1300 #ifdef CONFIG_PERF_EVENTS
1301 	case TRACE_REG_PERF_REGISTER:
1302 		return enable_trace_kprobe(tk, NULL);
1303 	case TRACE_REG_PERF_UNREGISTER:
1304 		return disable_trace_kprobe(tk, NULL);
1305 	case TRACE_REG_PERF_OPEN:
1306 	case TRACE_REG_PERF_CLOSE:
1307 	case TRACE_REG_PERF_ADD:
1308 	case TRACE_REG_PERF_DEL:
1309 		return 0;
1310 #endif
1311 	}
1312 	return 0;
1313 }
1314 
1315 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1316 {
1317 	struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1318 	int ret = 0;
1319 
1320 	raw_cpu_inc(*tk->nhit);
1321 
1322 	if (tk->tp.flags & TP_FLAG_TRACE)
1323 		kprobe_trace_func(tk, regs);
1324 #ifdef CONFIG_PERF_EVENTS
1325 	if (tk->tp.flags & TP_FLAG_PROFILE) {
1326 		ret = kprobe_perf_func(tk, regs);
1327 		/*
1328 		 * The ftrace kprobe handler leaves it up to us to re-enable
1329 		 * preemption here before returning if we've modified the ip.
1330 		 */
1331 		if (ret)
1332 			preempt_enable_no_resched();
1333 	}
1334 #endif
1335 	return ret;
1336 }
1337 NOKPROBE_SYMBOL(kprobe_dispatcher);
1338 
1339 static int
1340 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1341 {
1342 	struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1343 
1344 	raw_cpu_inc(*tk->nhit);
1345 
1346 	if (tk->tp.flags & TP_FLAG_TRACE)
1347 		kretprobe_trace_func(tk, ri, regs);
1348 #ifdef CONFIG_PERF_EVENTS
1349 	if (tk->tp.flags & TP_FLAG_PROFILE)
1350 		kretprobe_perf_func(tk, ri, regs);
1351 #endif
1352 	return 0;	/* We don't tweek kernel, so just return 0 */
1353 }
1354 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1355 
1356 static struct trace_event_functions kretprobe_funcs = {
1357 	.trace		= print_kretprobe_event
1358 };
1359 
1360 static struct trace_event_functions kprobe_funcs = {
1361 	.trace		= print_kprobe_event
1362 };
1363 
1364 static int register_kprobe_event(struct trace_kprobe *tk)
1365 {
1366 	struct trace_event_call *call = &tk->tp.call;
1367 	int ret;
1368 
1369 	/* Initialize trace_event_call */
1370 	INIT_LIST_HEAD(&call->class->fields);
1371 	if (trace_kprobe_is_return(tk)) {
1372 		call->event.funcs = &kretprobe_funcs;
1373 		call->class->define_fields = kretprobe_event_define_fields;
1374 	} else {
1375 		call->event.funcs = &kprobe_funcs;
1376 		call->class->define_fields = kprobe_event_define_fields;
1377 	}
1378 	if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1379 		return -ENOMEM;
1380 	ret = register_trace_event(&call->event);
1381 	if (!ret) {
1382 		kfree(call->print_fmt);
1383 		return -ENODEV;
1384 	}
1385 	call->flags = TRACE_EVENT_FL_KPROBE;
1386 	call->class->reg = kprobe_register;
1387 	call->data = tk;
1388 	ret = trace_add_event_call(call);
1389 	if (ret) {
1390 		pr_info("Failed to register kprobe event: %s\n",
1391 			trace_event_name(call));
1392 		kfree(call->print_fmt);
1393 		unregister_trace_event(&call->event);
1394 	}
1395 	return ret;
1396 }
1397 
1398 static int unregister_kprobe_event(struct trace_kprobe *tk)
1399 {
1400 	int ret;
1401 
1402 	/* tp->event is unregistered in trace_remove_event_call() */
1403 	ret = trace_remove_event_call(&tk->tp.call);
1404 	if (!ret)
1405 		kfree(tk->tp.call.print_fmt);
1406 	return ret;
1407 }
1408 
1409 /* Make a tracefs interface for controlling probe points */
1410 static __init int init_kprobe_trace(void)
1411 {
1412 	struct dentry *d_tracer;
1413 	struct dentry *entry;
1414 
1415 	if (register_module_notifier(&trace_kprobe_module_nb))
1416 		return -EINVAL;
1417 
1418 	d_tracer = tracing_init_dentry();
1419 	if (IS_ERR(d_tracer))
1420 		return 0;
1421 
1422 	entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1423 				    NULL, &kprobe_events_ops);
1424 
1425 	/* Event list interface */
1426 	if (!entry)
1427 		pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1428 
1429 	/* Profile interface */
1430 	entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1431 				    NULL, &kprobe_profile_ops);
1432 
1433 	if (!entry)
1434 		pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1435 	return 0;
1436 }
1437 fs_initcall(init_kprobe_trace);
1438 
1439 
1440 #ifdef CONFIG_FTRACE_STARTUP_TEST
1441 /*
1442  * The "__used" keeps gcc from removing the function symbol
1443  * from the kallsyms table. 'noinline' makes sure that there
1444  * isn't an inlined version used by the test method below
1445  */
1446 static __used __init noinline int
1447 kprobe_trace_selftest_target(int a1, int a2, int a3, int a4, int a5, int a6)
1448 {
1449 	return a1 + a2 + a3 + a4 + a5 + a6;
1450 }
1451 
1452 static __init struct trace_event_file *
1453 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1454 {
1455 	struct trace_event_file *file;
1456 
1457 	list_for_each_entry(file, &tr->events, list)
1458 		if (file->event_call == &tk->tp.call)
1459 			return file;
1460 
1461 	return NULL;
1462 }
1463 
1464 /*
1465  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1466  * stage, we can do this lockless.
1467  */
1468 static __init int kprobe_trace_self_tests_init(void)
1469 {
1470 	int ret, warn = 0;
1471 	int (*target)(int, int, int, int, int, int);
1472 	struct trace_kprobe *tk;
1473 	struct trace_event_file *file;
1474 
1475 	if (tracing_is_disabled())
1476 		return -ENODEV;
1477 
1478 	target = kprobe_trace_selftest_target;
1479 
1480 	pr_info("Testing kprobe tracing: ");
1481 
1482 	ret = trace_run_command("p:testprobe kprobe_trace_selftest_target "
1483 				"$stack $stack0 +0($stack)",
1484 				create_trace_kprobe);
1485 	if (WARN_ON_ONCE(ret)) {
1486 		pr_warn("error on probing function entry.\n");
1487 		warn++;
1488 	} else {
1489 		/* Enable trace point */
1490 		tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1491 		if (WARN_ON_ONCE(tk == NULL)) {
1492 			pr_warn("error on getting new probe.\n");
1493 			warn++;
1494 		} else {
1495 			file = find_trace_probe_file(tk, top_trace_array());
1496 			if (WARN_ON_ONCE(file == NULL)) {
1497 				pr_warn("error on getting probe file.\n");
1498 				warn++;
1499 			} else
1500 				enable_trace_kprobe(tk, file);
1501 		}
1502 	}
1503 
1504 	ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target "
1505 				"$retval", create_trace_kprobe);
1506 	if (WARN_ON_ONCE(ret)) {
1507 		pr_warn("error on probing function return.\n");
1508 		warn++;
1509 	} else {
1510 		/* Enable trace point */
1511 		tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1512 		if (WARN_ON_ONCE(tk == NULL)) {
1513 			pr_warn("error on getting 2nd new probe.\n");
1514 			warn++;
1515 		} else {
1516 			file = find_trace_probe_file(tk, top_trace_array());
1517 			if (WARN_ON_ONCE(file == NULL)) {
1518 				pr_warn("error on getting probe file.\n");
1519 				warn++;
1520 			} else
1521 				enable_trace_kprobe(tk, file);
1522 		}
1523 	}
1524 
1525 	if (warn)
1526 		goto end;
1527 
1528 	ret = target(1, 2, 3, 4, 5, 6);
1529 
1530 	/*
1531 	 * Not expecting an error here, the check is only to prevent the
1532 	 * optimizer from removing the call to target() as otherwise there
1533 	 * are no side-effects and the call is never performed.
1534 	 */
1535 	if (ret != 21)
1536 		warn++;
1537 
1538 	/* Disable trace points before removing it */
1539 	tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1540 	if (WARN_ON_ONCE(tk == NULL)) {
1541 		pr_warn("error on getting test probe.\n");
1542 		warn++;
1543 	} else {
1544 		if (trace_kprobe_nhit(tk) != 1) {
1545 			pr_warn("incorrect number of testprobe hits\n");
1546 			warn++;
1547 		}
1548 
1549 		file = find_trace_probe_file(tk, top_trace_array());
1550 		if (WARN_ON_ONCE(file == NULL)) {
1551 			pr_warn("error on getting probe file.\n");
1552 			warn++;
1553 		} else
1554 			disable_trace_kprobe(tk, file);
1555 	}
1556 
1557 	tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1558 	if (WARN_ON_ONCE(tk == NULL)) {
1559 		pr_warn("error on getting 2nd test probe.\n");
1560 		warn++;
1561 	} else {
1562 		if (trace_kprobe_nhit(tk) != 1) {
1563 			pr_warn("incorrect number of testprobe2 hits\n");
1564 			warn++;
1565 		}
1566 
1567 		file = find_trace_probe_file(tk, top_trace_array());
1568 		if (WARN_ON_ONCE(file == NULL)) {
1569 			pr_warn("error on getting probe file.\n");
1570 			warn++;
1571 		} else
1572 			disable_trace_kprobe(tk, file);
1573 	}
1574 
1575 	ret = trace_run_command("-:testprobe", create_trace_kprobe);
1576 	if (WARN_ON_ONCE(ret)) {
1577 		pr_warn("error on deleting a probe.\n");
1578 		warn++;
1579 	}
1580 
1581 	ret = trace_run_command("-:testprobe2", create_trace_kprobe);
1582 	if (WARN_ON_ONCE(ret)) {
1583 		pr_warn("error on deleting a probe.\n");
1584 		warn++;
1585 	}
1586 
1587 end:
1588 	release_all_trace_kprobes();
1589 	/*
1590 	 * Wait for the optimizer work to finish. Otherwise it might fiddle
1591 	 * with probes in already freed __init text.
1592 	 */
1593 	wait_for_kprobe_optimizer();
1594 	if (warn)
1595 		pr_cont("NG: Some tests are failed. Please check them.\n");
1596 	else
1597 		pr_cont("OK\n");
1598 	return 0;
1599 }
1600 
1601 late_initcall(kprobe_trace_self_tests_init);
1602 
1603 #endif
1604