xref: /linux/kernel/trace/trace_kprobe.c (revision 3a598264436e94c410c413088a7873fcad33616c)
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 #include <linux/kprobes.h>
23 #include <linux/seq_file.h>
24 #include <linux/slab.h>
25 #include <linux/smp.h>
26 #include <linux/debugfs.h>
27 #include <linux/types.h>
28 #include <linux/string.h>
29 #include <linux/ctype.h>
30 #include <linux/ptrace.h>
31 #include <linux/perf_event.h>
32 #include <linux/stringify.h>
33 #include <asm/bitsperlong.h>
34 
35 #include "trace.h"
36 #include "trace_output.h"
37 
38 #define MAX_TRACE_ARGS 128
39 #define MAX_ARGSTR_LEN 63
40 #define MAX_EVENT_NAME_LEN 64
41 #define KPROBE_EVENT_SYSTEM "kprobes"
42 
43 /* Reserved field names */
44 #define FIELD_STRING_IP "__probe_ip"
45 #define FIELD_STRING_RETIP "__probe_ret_ip"
46 #define FIELD_STRING_FUNC "__probe_func"
47 
48 const char *reserved_field_names[] = {
49 	"common_type",
50 	"common_flags",
51 	"common_preempt_count",
52 	"common_pid",
53 	"common_tgid",
54 	"common_lock_depth",
55 	FIELD_STRING_IP,
56 	FIELD_STRING_RETIP,
57 	FIELD_STRING_FUNC,
58 };
59 
60 /* Printing function type */
61 typedef int (*print_type_func_t)(struct trace_seq *, const char *, void *);
62 #define PRINT_TYPE_FUNC_NAME(type)	print_type_##type
63 #define PRINT_TYPE_FMT_NAME(type)	print_type_format_##type
64 
65 /* Printing  in basic type function template */
66 #define DEFINE_BASIC_PRINT_TYPE_FUNC(type, fmt, cast)			\
67 static __kprobes int PRINT_TYPE_FUNC_NAME(type)(struct trace_seq *s,	\
68 						const char *name, void *data)\
69 {									\
70 	return trace_seq_printf(s, " %s=" fmt, name, (cast)*(type *)data);\
71 }									\
72 static const char PRINT_TYPE_FMT_NAME(type)[] = fmt;
73 
74 DEFINE_BASIC_PRINT_TYPE_FUNC(u8, "%x", unsigned int)
75 DEFINE_BASIC_PRINT_TYPE_FUNC(u16, "%x", unsigned int)
76 DEFINE_BASIC_PRINT_TYPE_FUNC(u32, "%lx", unsigned long)
77 DEFINE_BASIC_PRINT_TYPE_FUNC(u64, "%llx", unsigned long long)
78 DEFINE_BASIC_PRINT_TYPE_FUNC(s8, "%d", int)
79 DEFINE_BASIC_PRINT_TYPE_FUNC(s16, "%d", int)
80 DEFINE_BASIC_PRINT_TYPE_FUNC(s32, "%ld", long)
81 DEFINE_BASIC_PRINT_TYPE_FUNC(s64, "%lld", long long)
82 
83 /* Data fetch function type */
84 typedef	void (*fetch_func_t)(struct pt_regs *, void *, void *);
85 
86 struct fetch_param {
87 	fetch_func_t	fn;
88 	void *data;
89 };
90 
91 static __kprobes void call_fetch(struct fetch_param *fprm,
92 				 struct pt_regs *regs, void *dest)
93 {
94 	return fprm->fn(regs, fprm->data, dest);
95 }
96 
97 #define FETCH_FUNC_NAME(kind, type)	fetch_##kind##_##type
98 /*
99  * Define macro for basic types - we don't need to define s* types, because
100  * we have to care only about bitwidth at recording time.
101  */
102 #define DEFINE_BASIC_FETCH_FUNCS(kind)  \
103 DEFINE_FETCH_##kind(u8)			\
104 DEFINE_FETCH_##kind(u16)		\
105 DEFINE_FETCH_##kind(u32)		\
106 DEFINE_FETCH_##kind(u64)
107 
108 #define CHECK_BASIC_FETCH_FUNCS(kind, fn)	\
109 	((FETCH_FUNC_NAME(kind, u8) == fn) ||	\
110 	 (FETCH_FUNC_NAME(kind, u16) == fn) ||	\
111 	 (FETCH_FUNC_NAME(kind, u32) == fn) ||	\
112 	 (FETCH_FUNC_NAME(kind, u64) == fn))
113 
114 /* Data fetch function templates */
115 #define DEFINE_FETCH_reg(type)						\
116 static __kprobes void FETCH_FUNC_NAME(reg, type)(struct pt_regs *regs,	\
117 					  void *offset, void *dest)	\
118 {									\
119 	*(type *)dest = (type)regs_get_register(regs,			\
120 				(unsigned int)((unsigned long)offset));	\
121 }
122 DEFINE_BASIC_FETCH_FUNCS(reg)
123 
124 #define DEFINE_FETCH_stack(type)					\
125 static __kprobes void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,\
126 					  void *offset, void *dest)	\
127 {									\
128 	*(type *)dest = (type)regs_get_kernel_stack_nth(regs,		\
129 				(unsigned int)((unsigned long)offset));	\
130 }
131 DEFINE_BASIC_FETCH_FUNCS(stack)
132 
133 #define DEFINE_FETCH_retval(type)					\
134 static __kprobes void FETCH_FUNC_NAME(retval, type)(struct pt_regs *regs,\
135 					  void *dummy, void *dest)	\
136 {									\
137 	*(type *)dest = (type)regs_return_value(regs);			\
138 }
139 DEFINE_BASIC_FETCH_FUNCS(retval)
140 
141 #define DEFINE_FETCH_memory(type)					\
142 static __kprobes void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,\
143 					  void *addr, void *dest)	\
144 {									\
145 	type retval;							\
146 	if (probe_kernel_address(addr, retval))				\
147 		*(type *)dest = 0;					\
148 	else								\
149 		*(type *)dest = retval;					\
150 }
151 DEFINE_BASIC_FETCH_FUNCS(memory)
152 
153 /* Memory fetching by symbol */
154 struct symbol_cache {
155 	char *symbol;
156 	long offset;
157 	unsigned long addr;
158 };
159 
160 static unsigned long update_symbol_cache(struct symbol_cache *sc)
161 {
162 	sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
163 	if (sc->addr)
164 		sc->addr += sc->offset;
165 	return sc->addr;
166 }
167 
168 static void free_symbol_cache(struct symbol_cache *sc)
169 {
170 	kfree(sc->symbol);
171 	kfree(sc);
172 }
173 
174 static struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
175 {
176 	struct symbol_cache *sc;
177 
178 	if (!sym || strlen(sym) == 0)
179 		return NULL;
180 	sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
181 	if (!sc)
182 		return NULL;
183 
184 	sc->symbol = kstrdup(sym, GFP_KERNEL);
185 	if (!sc->symbol) {
186 		kfree(sc);
187 		return NULL;
188 	}
189 	sc->offset = offset;
190 
191 	update_symbol_cache(sc);
192 	return sc;
193 }
194 
195 #define DEFINE_FETCH_symbol(type)					\
196 static __kprobes void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs,\
197 					  void *data, void *dest)	\
198 {									\
199 	struct symbol_cache *sc = data;					\
200 	if (sc->addr)							\
201 		fetch_memory_##type(regs, (void *)sc->addr, dest);	\
202 	else								\
203 		*(type *)dest = 0;					\
204 }
205 DEFINE_BASIC_FETCH_FUNCS(symbol)
206 
207 /* Dereference memory access function */
208 struct deref_fetch_param {
209 	struct fetch_param orig;
210 	long offset;
211 };
212 
213 #define DEFINE_FETCH_deref(type)					\
214 static __kprobes void FETCH_FUNC_NAME(deref, type)(struct pt_regs *regs,\
215 					    void *data, void *dest)	\
216 {									\
217 	struct deref_fetch_param *dprm = data;				\
218 	unsigned long addr;						\
219 	call_fetch(&dprm->orig, regs, &addr);				\
220 	if (addr) {							\
221 		addr += dprm->offset;					\
222 		fetch_memory_##type(regs, (void *)addr, dest);		\
223 	} else								\
224 		*(type *)dest = 0;					\
225 }
226 DEFINE_BASIC_FETCH_FUNCS(deref)
227 
228 static __kprobes void free_deref_fetch_param(struct deref_fetch_param *data)
229 {
230 	if (CHECK_BASIC_FETCH_FUNCS(deref, data->orig.fn))
231 		free_deref_fetch_param(data->orig.data);
232 	else if (CHECK_BASIC_FETCH_FUNCS(symbol, data->orig.fn))
233 		free_symbol_cache(data->orig.data);
234 	kfree(data);
235 }
236 
237 /* Default (unsigned long) fetch type */
238 #define __DEFAULT_FETCH_TYPE(t) u##t
239 #define _DEFAULT_FETCH_TYPE(t) __DEFAULT_FETCH_TYPE(t)
240 #define DEFAULT_FETCH_TYPE _DEFAULT_FETCH_TYPE(BITS_PER_LONG)
241 #define DEFAULT_FETCH_TYPE_STR __stringify(DEFAULT_FETCH_TYPE)
242 
243 #define ASSIGN_FETCH_FUNC(kind, type)	\
244 	.kind = FETCH_FUNC_NAME(kind, type)
245 
246 #define ASSIGN_FETCH_TYPE(ptype, ftype, sign)	\
247 	{.name = #ptype,			\
248 	 .size = sizeof(ftype),			\
249 	 .is_signed = sign,			\
250 	 .print = PRINT_TYPE_FUNC_NAME(ptype),	\
251 	 .fmt = PRINT_TYPE_FMT_NAME(ptype),	\
252 ASSIGN_FETCH_FUNC(reg, ftype),			\
253 ASSIGN_FETCH_FUNC(stack, ftype),		\
254 ASSIGN_FETCH_FUNC(retval, ftype),		\
255 ASSIGN_FETCH_FUNC(memory, ftype),		\
256 ASSIGN_FETCH_FUNC(symbol, ftype),		\
257 ASSIGN_FETCH_FUNC(deref, ftype),		\
258 	}
259 
260 /* Fetch type information table */
261 static const struct fetch_type {
262 	const char	*name;		/* Name of type */
263 	size_t		size;		/* Byte size of type */
264 	int		is_signed;	/* Signed flag */
265 	print_type_func_t	print;	/* Print functions */
266 	const char	*fmt;		/* Fromat string */
267 	/* Fetch functions */
268 	fetch_func_t	reg;
269 	fetch_func_t	stack;
270 	fetch_func_t	retval;
271 	fetch_func_t	memory;
272 	fetch_func_t	symbol;
273 	fetch_func_t	deref;
274 } fetch_type_table[] = {
275 	ASSIGN_FETCH_TYPE(u8,  u8,  0),
276 	ASSIGN_FETCH_TYPE(u16, u16, 0),
277 	ASSIGN_FETCH_TYPE(u32, u32, 0),
278 	ASSIGN_FETCH_TYPE(u64, u64, 0),
279 	ASSIGN_FETCH_TYPE(s8,  u8,  1),
280 	ASSIGN_FETCH_TYPE(s16, u16, 1),
281 	ASSIGN_FETCH_TYPE(s32, u32, 1),
282 	ASSIGN_FETCH_TYPE(s64, u64, 1),
283 };
284 
285 static const struct fetch_type *find_fetch_type(const char *type)
286 {
287 	int i;
288 
289 	if (!type)
290 		type = DEFAULT_FETCH_TYPE_STR;
291 
292 	for (i = 0; i < ARRAY_SIZE(fetch_type_table); i++)
293 		if (strcmp(type, fetch_type_table[i].name) == 0)
294 			return &fetch_type_table[i];
295 	return NULL;
296 }
297 
298 /* Special function : only accept unsigned long */
299 static __kprobes void fetch_stack_address(struct pt_regs *regs,
300 					  void *dummy, void *dest)
301 {
302 	*(unsigned long *)dest = kernel_stack_pointer(regs);
303 }
304 
305 /**
306  * Kprobe event core functions
307  */
308 
309 struct probe_arg {
310 	struct fetch_param	fetch;
311 	unsigned int		offset;	/* Offset from argument entry */
312 	const char		*name;	/* Name of this argument */
313 	const char		*comm;	/* Command of this argument */
314 	const struct fetch_type	*type;	/* Type of this argument */
315 };
316 
317 /* Flags for trace_probe */
318 #define TP_FLAG_TRACE	1
319 #define TP_FLAG_PROFILE	2
320 
321 struct trace_probe {
322 	struct list_head	list;
323 	struct kretprobe	rp;	/* Use rp.kp for kprobe use */
324 	unsigned long 		nhit;
325 	unsigned int		flags;	/* For TP_FLAG_* */
326 	const char		*symbol;	/* symbol name */
327 	struct ftrace_event_class	class;
328 	struct ftrace_event_call	call;
329 	ssize_t			size;		/* trace entry size */
330 	unsigned int		nr_args;
331 	struct probe_arg	args[];
332 };
333 
334 #define SIZEOF_TRACE_PROBE(n)			\
335 	(offsetof(struct trace_probe, args) +	\
336 	(sizeof(struct probe_arg) * (n)))
337 
338 
339 static __kprobes int probe_is_return(struct trace_probe *tp)
340 {
341 	return tp->rp.handler != NULL;
342 }
343 
344 static __kprobes const char *probe_symbol(struct trace_probe *tp)
345 {
346 	return tp->symbol ? tp->symbol : "unknown";
347 }
348 
349 static int register_probe_event(struct trace_probe *tp);
350 static void unregister_probe_event(struct trace_probe *tp);
351 
352 static DEFINE_MUTEX(probe_lock);
353 static LIST_HEAD(probe_list);
354 
355 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
356 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
357 				struct pt_regs *regs);
358 
359 /* Check the name is good for event/group */
360 static int check_event_name(const char *name)
361 {
362 	if (!isalpha(*name) && *name != '_')
363 		return 0;
364 	while (*++name != '\0') {
365 		if (!isalpha(*name) && !isdigit(*name) && *name != '_')
366 			return 0;
367 	}
368 	return 1;
369 }
370 
371 /*
372  * Allocate new trace_probe and initialize it (including kprobes).
373  */
374 static struct trace_probe *alloc_trace_probe(const char *group,
375 					     const char *event,
376 					     void *addr,
377 					     const char *symbol,
378 					     unsigned long offs,
379 					     int nargs, int is_return)
380 {
381 	struct trace_probe *tp;
382 	int ret = -ENOMEM;
383 
384 	tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
385 	if (!tp)
386 		return ERR_PTR(ret);
387 
388 	if (symbol) {
389 		tp->symbol = kstrdup(symbol, GFP_KERNEL);
390 		if (!tp->symbol)
391 			goto error;
392 		tp->rp.kp.symbol_name = tp->symbol;
393 		tp->rp.kp.offset = offs;
394 	} else
395 		tp->rp.kp.addr = addr;
396 
397 	if (is_return)
398 		tp->rp.handler = kretprobe_dispatcher;
399 	else
400 		tp->rp.kp.pre_handler = kprobe_dispatcher;
401 
402 	if (!event || !check_event_name(event)) {
403 		ret = -EINVAL;
404 		goto error;
405 	}
406 
407 	tp->call.class = &tp->class;
408 	tp->call.name = kstrdup(event, GFP_KERNEL);
409 	if (!tp->call.name)
410 		goto error;
411 
412 	if (!group || !check_event_name(group)) {
413 		ret = -EINVAL;
414 		goto error;
415 	}
416 
417 	tp->class.system = kstrdup(group, GFP_KERNEL);
418 	if (!tp->class.system)
419 		goto error;
420 
421 	INIT_LIST_HEAD(&tp->list);
422 	return tp;
423 error:
424 	kfree(tp->call.name);
425 	kfree(tp->symbol);
426 	kfree(tp);
427 	return ERR_PTR(ret);
428 }
429 
430 static void free_probe_arg(struct probe_arg *arg)
431 {
432 	if (CHECK_BASIC_FETCH_FUNCS(deref, arg->fetch.fn))
433 		free_deref_fetch_param(arg->fetch.data);
434 	else if (CHECK_BASIC_FETCH_FUNCS(symbol, arg->fetch.fn))
435 		free_symbol_cache(arg->fetch.data);
436 	kfree(arg->name);
437 	kfree(arg->comm);
438 }
439 
440 static void free_trace_probe(struct trace_probe *tp)
441 {
442 	int i;
443 
444 	for (i = 0; i < tp->nr_args; i++)
445 		free_probe_arg(&tp->args[i]);
446 
447 	kfree(tp->call.class->system);
448 	kfree(tp->call.name);
449 	kfree(tp->symbol);
450 	kfree(tp);
451 }
452 
453 static struct trace_probe *find_probe_event(const char *event,
454 					    const char *group)
455 {
456 	struct trace_probe *tp;
457 
458 	list_for_each_entry(tp, &probe_list, list)
459 		if (strcmp(tp->call.name, event) == 0 &&
460 		    strcmp(tp->call.class->system, group) == 0)
461 			return tp;
462 	return NULL;
463 }
464 
465 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
466 static void unregister_trace_probe(struct trace_probe *tp)
467 {
468 	if (probe_is_return(tp))
469 		unregister_kretprobe(&tp->rp);
470 	else
471 		unregister_kprobe(&tp->rp.kp);
472 	list_del(&tp->list);
473 	unregister_probe_event(tp);
474 }
475 
476 /* Register a trace_probe and probe_event */
477 static int register_trace_probe(struct trace_probe *tp)
478 {
479 	struct trace_probe *old_tp;
480 	int ret;
481 
482 	mutex_lock(&probe_lock);
483 
484 	/* register as an event */
485 	old_tp = find_probe_event(tp->call.name, tp->call.class->system);
486 	if (old_tp) {
487 		/* delete old event */
488 		unregister_trace_probe(old_tp);
489 		free_trace_probe(old_tp);
490 	}
491 	ret = register_probe_event(tp);
492 	if (ret) {
493 		pr_warning("Faild to register probe event(%d)\n", ret);
494 		goto end;
495 	}
496 
497 	tp->rp.kp.flags |= KPROBE_FLAG_DISABLED;
498 	if (probe_is_return(tp))
499 		ret = register_kretprobe(&tp->rp);
500 	else
501 		ret = register_kprobe(&tp->rp.kp);
502 
503 	if (ret) {
504 		pr_warning("Could not insert probe(%d)\n", ret);
505 		if (ret == -EILSEQ) {
506 			pr_warning("Probing address(0x%p) is not an "
507 				   "instruction boundary.\n",
508 				   tp->rp.kp.addr);
509 			ret = -EINVAL;
510 		}
511 		unregister_probe_event(tp);
512 	} else
513 		list_add_tail(&tp->list, &probe_list);
514 end:
515 	mutex_unlock(&probe_lock);
516 	return ret;
517 }
518 
519 /* Split symbol and offset. */
520 static int split_symbol_offset(char *symbol, unsigned long *offset)
521 {
522 	char *tmp;
523 	int ret;
524 
525 	if (!offset)
526 		return -EINVAL;
527 
528 	tmp = strchr(symbol, '+');
529 	if (tmp) {
530 		/* skip sign because strict_strtol doesn't accept '+' */
531 		ret = strict_strtoul(tmp + 1, 0, offset);
532 		if (ret)
533 			return ret;
534 		*tmp = '\0';
535 	} else
536 		*offset = 0;
537 	return 0;
538 }
539 
540 #define PARAM_MAX_ARGS 16
541 #define PARAM_MAX_STACK (THREAD_SIZE / sizeof(unsigned long))
542 
543 static int parse_probe_vars(char *arg, const struct fetch_type *t,
544 			    struct fetch_param *f, int is_return)
545 {
546 	int ret = 0;
547 	unsigned long param;
548 
549 	if (strcmp(arg, "retval") == 0) {
550 		if (is_return)
551 			f->fn = t->retval;
552 		else
553 			ret = -EINVAL;
554 	} else if (strncmp(arg, "stack", 5) == 0) {
555 		if (arg[5] == '\0') {
556 			if (strcmp(t->name, DEFAULT_FETCH_TYPE_STR) == 0)
557 				f->fn = fetch_stack_address;
558 			else
559 				ret = -EINVAL;
560 		} else if (isdigit(arg[5])) {
561 			ret = strict_strtoul(arg + 5, 10, &param);
562 			if (ret || param > PARAM_MAX_STACK)
563 				ret = -EINVAL;
564 			else {
565 				f->fn = t->stack;
566 				f->data = (void *)param;
567 			}
568 		} else
569 			ret = -EINVAL;
570 	} else
571 		ret = -EINVAL;
572 	return ret;
573 }
574 
575 /* Recursive argument parser */
576 static int __parse_probe_arg(char *arg, const struct fetch_type *t,
577 			     struct fetch_param *f, int is_return)
578 {
579 	int ret = 0;
580 	unsigned long param;
581 	long offset;
582 	char *tmp;
583 
584 	switch (arg[0]) {
585 	case '$':
586 		ret = parse_probe_vars(arg + 1, t, f, is_return);
587 		break;
588 	case '%':	/* named register */
589 		ret = regs_query_register_offset(arg + 1);
590 		if (ret >= 0) {
591 			f->fn = t->reg;
592 			f->data = (void *)(unsigned long)ret;
593 			ret = 0;
594 		}
595 		break;
596 	case '@':	/* memory or symbol */
597 		if (isdigit(arg[1])) {
598 			ret = strict_strtoul(arg + 1, 0, &param);
599 			if (ret)
600 				break;
601 			f->fn = t->memory;
602 			f->data = (void *)param;
603 		} else {
604 			ret = split_symbol_offset(arg + 1, &offset);
605 			if (ret)
606 				break;
607 			f->data = alloc_symbol_cache(arg + 1, offset);
608 			if (f->data)
609 				f->fn = t->symbol;
610 		}
611 		break;
612 	case '+':	/* deref memory */
613 	case '-':
614 		tmp = strchr(arg, '(');
615 		if (!tmp)
616 			break;
617 		*tmp = '\0';
618 		ret = strict_strtol(arg + 1, 0, &offset);
619 		if (ret)
620 			break;
621 		if (arg[0] == '-')
622 			offset = -offset;
623 		arg = tmp + 1;
624 		tmp = strrchr(arg, ')');
625 		if (tmp) {
626 			struct deref_fetch_param *dprm;
627 			const struct fetch_type *t2 = find_fetch_type(NULL);
628 			*tmp = '\0';
629 			dprm = kzalloc(sizeof(struct deref_fetch_param),
630 				       GFP_KERNEL);
631 			if (!dprm)
632 				return -ENOMEM;
633 			dprm->offset = offset;
634 			ret = __parse_probe_arg(arg, t2, &dprm->orig,
635 						is_return);
636 			if (ret)
637 				kfree(dprm);
638 			else {
639 				f->fn = t->deref;
640 				f->data = (void *)dprm;
641 			}
642 		}
643 		break;
644 	}
645 	if (!ret && !f->fn)
646 		ret = -EINVAL;
647 	return ret;
648 }
649 
650 /* String length checking wrapper */
651 static int parse_probe_arg(char *arg, struct trace_probe *tp,
652 			   struct probe_arg *parg, int is_return)
653 {
654 	const char *t;
655 
656 	if (strlen(arg) > MAX_ARGSTR_LEN) {
657 		pr_info("Argument is too long.: %s\n",  arg);
658 		return -ENOSPC;
659 	}
660 	parg->comm = kstrdup(arg, GFP_KERNEL);
661 	if (!parg->comm) {
662 		pr_info("Failed to allocate memory for command '%s'.\n", arg);
663 		return -ENOMEM;
664 	}
665 	t = strchr(parg->comm, ':');
666 	if (t) {
667 		arg[t - parg->comm] = '\0';
668 		t++;
669 	}
670 	parg->type = find_fetch_type(t);
671 	if (!parg->type) {
672 		pr_info("Unsupported type: %s\n", t);
673 		return -EINVAL;
674 	}
675 	parg->offset = tp->size;
676 	tp->size += parg->type->size;
677 	return __parse_probe_arg(arg, parg->type, &parg->fetch, is_return);
678 }
679 
680 /* Return 1 if name is reserved or already used by another argument */
681 static int conflict_field_name(const char *name,
682 			       struct probe_arg *args, int narg)
683 {
684 	int i;
685 	for (i = 0; i < ARRAY_SIZE(reserved_field_names); i++)
686 		if (strcmp(reserved_field_names[i], name) == 0)
687 			return 1;
688 	for (i = 0; i < narg; i++)
689 		if (strcmp(args[i].name, name) == 0)
690 			return 1;
691 	return 0;
692 }
693 
694 static int create_trace_probe(int argc, char **argv)
695 {
696 	/*
697 	 * Argument syntax:
698 	 *  - Add kprobe: p[:[GRP/]EVENT] KSYM[+OFFS]|KADDR [FETCHARGS]
699 	 *  - Add kretprobe: r[:[GRP/]EVENT] KSYM[+0] [FETCHARGS]
700 	 * Fetch args:
701 	 *  $retval	: fetch return value
702 	 *  $stack	: fetch stack address
703 	 *  $stackN	: fetch Nth of stack (N:0-)
704 	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
705 	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
706 	 *  %REG	: fetch register REG
707 	 * Dereferencing memory fetch:
708 	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
709 	 * Alias name of args:
710 	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
711 	 * Type of args:
712 	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
713 	 */
714 	struct trace_probe *tp;
715 	int i, ret = 0;
716 	int is_return = 0, is_delete = 0;
717 	char *symbol = NULL, *event = NULL, *group = NULL;
718 	char *arg, *tmp;
719 	unsigned long offset = 0;
720 	void *addr = NULL;
721 	char buf[MAX_EVENT_NAME_LEN];
722 
723 	/* argc must be >= 1 */
724 	if (argv[0][0] == 'p')
725 		is_return = 0;
726 	else if (argv[0][0] == 'r')
727 		is_return = 1;
728 	else if (argv[0][0] == '-')
729 		is_delete = 1;
730 	else {
731 		pr_info("Probe definition must be started with 'p', 'r' or"
732 			" '-'.\n");
733 		return -EINVAL;
734 	}
735 
736 	if (argv[0][1] == ':') {
737 		event = &argv[0][2];
738 		if (strchr(event, '/')) {
739 			group = event;
740 			event = strchr(group, '/') + 1;
741 			event[-1] = '\0';
742 			if (strlen(group) == 0) {
743 				pr_info("Group name is not specified\n");
744 				return -EINVAL;
745 			}
746 		}
747 		if (strlen(event) == 0) {
748 			pr_info("Event name is not specified\n");
749 			return -EINVAL;
750 		}
751 	}
752 	if (!group)
753 		group = KPROBE_EVENT_SYSTEM;
754 
755 	if (is_delete) {
756 		if (!event) {
757 			pr_info("Delete command needs an event name.\n");
758 			return -EINVAL;
759 		}
760 		tp = find_probe_event(event, group);
761 		if (!tp) {
762 			pr_info("Event %s/%s doesn't exist.\n", group, event);
763 			return -ENOENT;
764 		}
765 		/* delete an event */
766 		unregister_trace_probe(tp);
767 		free_trace_probe(tp);
768 		return 0;
769 	}
770 
771 	if (argc < 2) {
772 		pr_info("Probe point is not specified.\n");
773 		return -EINVAL;
774 	}
775 	if (isdigit(argv[1][0])) {
776 		if (is_return) {
777 			pr_info("Return probe point must be a symbol.\n");
778 			return -EINVAL;
779 		}
780 		/* an address specified */
781 		ret = strict_strtoul(&argv[1][0], 0, (unsigned long *)&addr);
782 		if (ret) {
783 			pr_info("Failed to parse address.\n");
784 			return ret;
785 		}
786 	} else {
787 		/* a symbol specified */
788 		symbol = argv[1];
789 		/* TODO: support .init module functions */
790 		ret = split_symbol_offset(symbol, &offset);
791 		if (ret) {
792 			pr_info("Failed to parse symbol.\n");
793 			return ret;
794 		}
795 		if (offset && is_return) {
796 			pr_info("Return probe must be used without offset.\n");
797 			return -EINVAL;
798 		}
799 	}
800 	argc -= 2; argv += 2;
801 
802 	/* setup a probe */
803 	if (!event) {
804 		/* Make a new event name */
805 		if (symbol)
806 			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
807 				 is_return ? 'r' : 'p', symbol, offset);
808 		else
809 			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
810 				 is_return ? 'r' : 'p', addr);
811 		event = buf;
812 	}
813 	tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
814 			       is_return);
815 	if (IS_ERR(tp)) {
816 		pr_info("Failed to allocate trace_probe.(%d)\n",
817 			(int)PTR_ERR(tp));
818 		return PTR_ERR(tp);
819 	}
820 
821 	/* parse arguments */
822 	ret = 0;
823 	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
824 		/* Parse argument name */
825 		arg = strchr(argv[i], '=');
826 		if (arg)
827 			*arg++ = '\0';
828 		else
829 			arg = argv[i];
830 
831 		tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
832 		if (!tp->args[i].name) {
833 			pr_info("Failed to allocate argument%d name '%s'.\n",
834 				i, argv[i]);
835 			ret = -ENOMEM;
836 			goto error;
837 		}
838 		tmp = strchr(tp->args[i].name, ':');
839 		if (tmp)
840 			*tmp = '_';	/* convert : to _ */
841 
842 		if (conflict_field_name(tp->args[i].name, tp->args, i)) {
843 			pr_info("Argument%d name '%s' conflicts with "
844 				"another field.\n", i, argv[i]);
845 			ret = -EINVAL;
846 			goto error;
847 		}
848 
849 		/* Parse fetch argument */
850 		ret = parse_probe_arg(arg, tp, &tp->args[i], is_return);
851 		if (ret) {
852 			pr_info("Parse error at argument%d. (%d)\n", i, ret);
853 			kfree(tp->args[i].name);
854 			goto error;
855 		}
856 
857 		tp->nr_args++;
858 	}
859 
860 	ret = register_trace_probe(tp);
861 	if (ret)
862 		goto error;
863 	return 0;
864 
865 error:
866 	free_trace_probe(tp);
867 	return ret;
868 }
869 
870 static void cleanup_all_probes(void)
871 {
872 	struct trace_probe *tp;
873 
874 	mutex_lock(&probe_lock);
875 	/* TODO: Use batch unregistration */
876 	while (!list_empty(&probe_list)) {
877 		tp = list_entry(probe_list.next, struct trace_probe, list);
878 		unregister_trace_probe(tp);
879 		free_trace_probe(tp);
880 	}
881 	mutex_unlock(&probe_lock);
882 }
883 
884 
885 /* Probes listing interfaces */
886 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
887 {
888 	mutex_lock(&probe_lock);
889 	return seq_list_start(&probe_list, *pos);
890 }
891 
892 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
893 {
894 	return seq_list_next(v, &probe_list, pos);
895 }
896 
897 static void probes_seq_stop(struct seq_file *m, void *v)
898 {
899 	mutex_unlock(&probe_lock);
900 }
901 
902 static int probes_seq_show(struct seq_file *m, void *v)
903 {
904 	struct trace_probe *tp = v;
905 	int i;
906 
907 	seq_printf(m, "%c", probe_is_return(tp) ? 'r' : 'p');
908 	seq_printf(m, ":%s/%s", tp->call.class->system, tp->call.name);
909 
910 	if (!tp->symbol)
911 		seq_printf(m, " 0x%p", tp->rp.kp.addr);
912 	else if (tp->rp.kp.offset)
913 		seq_printf(m, " %s+%u", probe_symbol(tp), tp->rp.kp.offset);
914 	else
915 		seq_printf(m, " %s", probe_symbol(tp));
916 
917 	for (i = 0; i < tp->nr_args; i++)
918 		seq_printf(m, " %s=%s", tp->args[i].name, tp->args[i].comm);
919 	seq_printf(m, "\n");
920 
921 	return 0;
922 }
923 
924 static const struct seq_operations probes_seq_op = {
925 	.start  = probes_seq_start,
926 	.next   = probes_seq_next,
927 	.stop   = probes_seq_stop,
928 	.show   = probes_seq_show
929 };
930 
931 static int probes_open(struct inode *inode, struct file *file)
932 {
933 	if ((file->f_mode & FMODE_WRITE) &&
934 	    (file->f_flags & O_TRUNC))
935 		cleanup_all_probes();
936 
937 	return seq_open(file, &probes_seq_op);
938 }
939 
940 static int command_trace_probe(const char *buf)
941 {
942 	char **argv;
943 	int argc = 0, ret = 0;
944 
945 	argv = argv_split(GFP_KERNEL, buf, &argc);
946 	if (!argv)
947 		return -ENOMEM;
948 
949 	if (argc)
950 		ret = create_trace_probe(argc, argv);
951 
952 	argv_free(argv);
953 	return ret;
954 }
955 
956 #define WRITE_BUFSIZE 128
957 
958 static ssize_t probes_write(struct file *file, const char __user *buffer,
959 			    size_t count, loff_t *ppos)
960 {
961 	char *kbuf, *tmp;
962 	int ret;
963 	size_t done;
964 	size_t size;
965 
966 	kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
967 	if (!kbuf)
968 		return -ENOMEM;
969 
970 	ret = done = 0;
971 	while (done < count) {
972 		size = count - done;
973 		if (size >= WRITE_BUFSIZE)
974 			size = WRITE_BUFSIZE - 1;
975 		if (copy_from_user(kbuf, buffer + done, size)) {
976 			ret = -EFAULT;
977 			goto out;
978 		}
979 		kbuf[size] = '\0';
980 		tmp = strchr(kbuf, '\n');
981 		if (tmp) {
982 			*tmp = '\0';
983 			size = tmp - kbuf + 1;
984 		} else if (done + size < count) {
985 			pr_warning("Line length is too long: "
986 				   "Should be less than %d.", WRITE_BUFSIZE);
987 			ret = -EINVAL;
988 			goto out;
989 		}
990 		done += size;
991 		/* Remove comments */
992 		tmp = strchr(kbuf, '#');
993 		if (tmp)
994 			*tmp = '\0';
995 
996 		ret = command_trace_probe(kbuf);
997 		if (ret)
998 			goto out;
999 	}
1000 	ret = done;
1001 out:
1002 	kfree(kbuf);
1003 	return ret;
1004 }
1005 
1006 static const struct file_operations kprobe_events_ops = {
1007 	.owner          = THIS_MODULE,
1008 	.open           = probes_open,
1009 	.read           = seq_read,
1010 	.llseek         = seq_lseek,
1011 	.release        = seq_release,
1012 	.write		= probes_write,
1013 };
1014 
1015 /* Probes profiling interfaces */
1016 static int probes_profile_seq_show(struct seq_file *m, void *v)
1017 {
1018 	struct trace_probe *tp = v;
1019 
1020 	seq_printf(m, "  %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
1021 		   tp->rp.kp.nmissed);
1022 
1023 	return 0;
1024 }
1025 
1026 static const struct seq_operations profile_seq_op = {
1027 	.start  = probes_seq_start,
1028 	.next   = probes_seq_next,
1029 	.stop   = probes_seq_stop,
1030 	.show   = probes_profile_seq_show
1031 };
1032 
1033 static int profile_open(struct inode *inode, struct file *file)
1034 {
1035 	return seq_open(file, &profile_seq_op);
1036 }
1037 
1038 static const struct file_operations kprobe_profile_ops = {
1039 	.owner          = THIS_MODULE,
1040 	.open           = profile_open,
1041 	.read           = seq_read,
1042 	.llseek         = seq_lseek,
1043 	.release        = seq_release,
1044 };
1045 
1046 /* Kprobe handler */
1047 static __kprobes void kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
1048 {
1049 	struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1050 	struct kprobe_trace_entry_head *entry;
1051 	struct ring_buffer_event *event;
1052 	struct ring_buffer *buffer;
1053 	u8 *data;
1054 	int size, i, pc;
1055 	unsigned long irq_flags;
1056 	struct ftrace_event_call *call = &tp->call;
1057 
1058 	tp->nhit++;
1059 
1060 	local_save_flags(irq_flags);
1061 	pc = preempt_count();
1062 
1063 	size = sizeof(*entry) + tp->size;
1064 
1065 	event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
1066 						  size, irq_flags, pc);
1067 	if (!event)
1068 		return;
1069 
1070 	entry = ring_buffer_event_data(event);
1071 	entry->ip = (unsigned long)kp->addr;
1072 	data = (u8 *)&entry[1];
1073 	for (i = 0; i < tp->nr_args; i++)
1074 		call_fetch(&tp->args[i].fetch, regs, data + tp->args[i].offset);
1075 
1076 	if (!filter_current_check_discard(buffer, call, entry, event))
1077 		trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
1078 }
1079 
1080 /* Kretprobe handler */
1081 static __kprobes void kretprobe_trace_func(struct kretprobe_instance *ri,
1082 					  struct pt_regs *regs)
1083 {
1084 	struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1085 	struct kretprobe_trace_entry_head *entry;
1086 	struct ring_buffer_event *event;
1087 	struct ring_buffer *buffer;
1088 	u8 *data;
1089 	int size, i, pc;
1090 	unsigned long irq_flags;
1091 	struct ftrace_event_call *call = &tp->call;
1092 
1093 	local_save_flags(irq_flags);
1094 	pc = preempt_count();
1095 
1096 	size = sizeof(*entry) + tp->size;
1097 
1098 	event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
1099 						  size, irq_flags, pc);
1100 	if (!event)
1101 		return;
1102 
1103 	entry = ring_buffer_event_data(event);
1104 	entry->func = (unsigned long)tp->rp.kp.addr;
1105 	entry->ret_ip = (unsigned long)ri->ret_addr;
1106 	data = (u8 *)&entry[1];
1107 	for (i = 0; i < tp->nr_args; i++)
1108 		call_fetch(&tp->args[i].fetch, regs, data + tp->args[i].offset);
1109 
1110 	if (!filter_current_check_discard(buffer, call, entry, event))
1111 		trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
1112 }
1113 
1114 /* Event entry printers */
1115 enum print_line_t
1116 print_kprobe_event(struct trace_iterator *iter, int flags,
1117 		   struct trace_event *event)
1118 {
1119 	struct kprobe_trace_entry_head *field;
1120 	struct trace_seq *s = &iter->seq;
1121 	struct trace_probe *tp;
1122 	u8 *data;
1123 	int i;
1124 
1125 	field = (struct kprobe_trace_entry_head *)iter->ent;
1126 	tp = container_of(event, struct trace_probe, call.event);
1127 
1128 	if (!trace_seq_printf(s, "%s: (", tp->call.name))
1129 		goto partial;
1130 
1131 	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1132 		goto partial;
1133 
1134 	if (!trace_seq_puts(s, ")"))
1135 		goto partial;
1136 
1137 	data = (u8 *)&field[1];
1138 	for (i = 0; i < tp->nr_args; i++)
1139 		if (!tp->args[i].type->print(s, tp->args[i].name,
1140 					     data + tp->args[i].offset))
1141 			goto partial;
1142 
1143 	if (!trace_seq_puts(s, "\n"))
1144 		goto partial;
1145 
1146 	return TRACE_TYPE_HANDLED;
1147 partial:
1148 	return TRACE_TYPE_PARTIAL_LINE;
1149 }
1150 
1151 enum print_line_t
1152 print_kretprobe_event(struct trace_iterator *iter, int flags,
1153 		      struct trace_event *event)
1154 {
1155 	struct kretprobe_trace_entry_head *field;
1156 	struct trace_seq *s = &iter->seq;
1157 	struct trace_probe *tp;
1158 	u8 *data;
1159 	int i;
1160 
1161 	field = (struct kretprobe_trace_entry_head *)iter->ent;
1162 	tp = container_of(event, struct trace_probe, call.event);
1163 
1164 	if (!trace_seq_printf(s, "%s: (", tp->call.name))
1165 		goto partial;
1166 
1167 	if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1168 		goto partial;
1169 
1170 	if (!trace_seq_puts(s, " <- "))
1171 		goto partial;
1172 
1173 	if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1174 		goto partial;
1175 
1176 	if (!trace_seq_puts(s, ")"))
1177 		goto partial;
1178 
1179 	data = (u8 *)&field[1];
1180 	for (i = 0; i < tp->nr_args; i++)
1181 		if (!tp->args[i].type->print(s, tp->args[i].name,
1182 					     data + tp->args[i].offset))
1183 			goto partial;
1184 
1185 	if (!trace_seq_puts(s, "\n"))
1186 		goto partial;
1187 
1188 	return TRACE_TYPE_HANDLED;
1189 partial:
1190 	return TRACE_TYPE_PARTIAL_LINE;
1191 }
1192 
1193 static int probe_event_enable(struct ftrace_event_call *call)
1194 {
1195 	struct trace_probe *tp = (struct trace_probe *)call->data;
1196 
1197 	tp->flags |= TP_FLAG_TRACE;
1198 	if (probe_is_return(tp))
1199 		return enable_kretprobe(&tp->rp);
1200 	else
1201 		return enable_kprobe(&tp->rp.kp);
1202 }
1203 
1204 static void probe_event_disable(struct ftrace_event_call *call)
1205 {
1206 	struct trace_probe *tp = (struct trace_probe *)call->data;
1207 
1208 	tp->flags &= ~TP_FLAG_TRACE;
1209 	if (!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE))) {
1210 		if (probe_is_return(tp))
1211 			disable_kretprobe(&tp->rp);
1212 		else
1213 			disable_kprobe(&tp->rp.kp);
1214 	}
1215 }
1216 
1217 static int probe_event_raw_init(struct ftrace_event_call *event_call)
1218 {
1219 	return 0;
1220 }
1221 
1222 #undef DEFINE_FIELD
1223 #define DEFINE_FIELD(type, item, name, is_signed)			\
1224 	do {								\
1225 		ret = trace_define_field(event_call, #type, name,	\
1226 					 offsetof(typeof(field), item),	\
1227 					 sizeof(field.item), is_signed, \
1228 					 FILTER_OTHER);			\
1229 		if (ret)						\
1230 			return ret;					\
1231 	} while (0)
1232 
1233 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
1234 {
1235 	int ret, i;
1236 	struct kprobe_trace_entry_head field;
1237 	struct trace_probe *tp = (struct trace_probe *)event_call->data;
1238 
1239 	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1240 	/* Set argument names as fields */
1241 	for (i = 0; i < tp->nr_args; i++) {
1242 		ret = trace_define_field(event_call, tp->args[i].type->name,
1243 					 tp->args[i].name,
1244 					 sizeof(field) + tp->args[i].offset,
1245 					 tp->args[i].type->size,
1246 					 tp->args[i].type->is_signed,
1247 					 FILTER_OTHER);
1248 		if (ret)
1249 			return ret;
1250 	}
1251 	return 0;
1252 }
1253 
1254 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1255 {
1256 	int ret, i;
1257 	struct kretprobe_trace_entry_head field;
1258 	struct trace_probe *tp = (struct trace_probe *)event_call->data;
1259 
1260 	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1261 	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1262 	/* Set argument names as fields */
1263 	for (i = 0; i < tp->nr_args; i++) {
1264 		ret = trace_define_field(event_call, tp->args[i].type->name,
1265 					 tp->args[i].name,
1266 					 sizeof(field) + tp->args[i].offset,
1267 					 tp->args[i].type->size,
1268 					 tp->args[i].type->is_signed,
1269 					 FILTER_OTHER);
1270 		if (ret)
1271 			return ret;
1272 	}
1273 	return 0;
1274 }
1275 
1276 static int __set_print_fmt(struct trace_probe *tp, char *buf, int len)
1277 {
1278 	int i;
1279 	int pos = 0;
1280 
1281 	const char *fmt, *arg;
1282 
1283 	if (!probe_is_return(tp)) {
1284 		fmt = "(%lx)";
1285 		arg = "REC->" FIELD_STRING_IP;
1286 	} else {
1287 		fmt = "(%lx <- %lx)";
1288 		arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
1289 	}
1290 
1291 	/* When len=0, we just calculate the needed length */
1292 #define LEN_OR_ZERO (len ? len - pos : 0)
1293 
1294 	pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
1295 
1296 	for (i = 0; i < tp->nr_args; i++) {
1297 		pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
1298 				tp->args[i].name, tp->args[i].type->fmt);
1299 	}
1300 
1301 	pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
1302 
1303 	for (i = 0; i < tp->nr_args; i++) {
1304 		pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
1305 				tp->args[i].name);
1306 	}
1307 
1308 #undef LEN_OR_ZERO
1309 
1310 	/* return the length of print_fmt */
1311 	return pos;
1312 }
1313 
1314 static int set_print_fmt(struct trace_probe *tp)
1315 {
1316 	int len;
1317 	char *print_fmt;
1318 
1319 	/* First: called with 0 length to calculate the needed length */
1320 	len = __set_print_fmt(tp, NULL, 0);
1321 	print_fmt = kmalloc(len + 1, GFP_KERNEL);
1322 	if (!print_fmt)
1323 		return -ENOMEM;
1324 
1325 	/* Second: actually write the @print_fmt */
1326 	__set_print_fmt(tp, print_fmt, len + 1);
1327 	tp->call.print_fmt = print_fmt;
1328 
1329 	return 0;
1330 }
1331 
1332 #ifdef CONFIG_PERF_EVENTS
1333 
1334 /* Kprobe profile handler */
1335 static __kprobes void kprobe_perf_func(struct kprobe *kp,
1336 					 struct pt_regs *regs)
1337 {
1338 	struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1339 	struct ftrace_event_call *call = &tp->call;
1340 	struct kprobe_trace_entry_head *entry;
1341 	struct hlist_head *head;
1342 	u8 *data;
1343 	int size, __size, i;
1344 	int rctx;
1345 
1346 	__size = sizeof(*entry) + tp->size;
1347 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1348 	size -= sizeof(u32);
1349 	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1350 		     "profile buffer not large enough"))
1351 		return;
1352 
1353 	entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1354 	if (!entry)
1355 		return;
1356 
1357 	entry->ip = (unsigned long)kp->addr;
1358 	data = (u8 *)&entry[1];
1359 	for (i = 0; i < tp->nr_args; i++)
1360 		call_fetch(&tp->args[i].fetch, regs, data + tp->args[i].offset);
1361 
1362 	head = this_cpu_ptr(call->perf_events);
1363 	perf_trace_buf_submit(entry, size, rctx, entry->ip, 1, regs, head);
1364 }
1365 
1366 /* Kretprobe profile handler */
1367 static __kprobes void kretprobe_perf_func(struct kretprobe_instance *ri,
1368 					    struct pt_regs *regs)
1369 {
1370 	struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1371 	struct ftrace_event_call *call = &tp->call;
1372 	struct kretprobe_trace_entry_head *entry;
1373 	struct hlist_head *head;
1374 	u8 *data;
1375 	int size, __size, i;
1376 	int rctx;
1377 
1378 	__size = sizeof(*entry) + tp->size;
1379 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1380 	size -= sizeof(u32);
1381 	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1382 		     "profile buffer not large enough"))
1383 		return;
1384 
1385 	entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1386 	if (!entry)
1387 		return;
1388 
1389 	entry->func = (unsigned long)tp->rp.kp.addr;
1390 	entry->ret_ip = (unsigned long)ri->ret_addr;
1391 	data = (u8 *)&entry[1];
1392 	for (i = 0; i < tp->nr_args; i++)
1393 		call_fetch(&tp->args[i].fetch, regs, data + tp->args[i].offset);
1394 
1395 	head = this_cpu_ptr(call->perf_events);
1396 	perf_trace_buf_submit(entry, size, rctx, entry->ret_ip, 1, regs, head);
1397 }
1398 
1399 static int probe_perf_enable(struct ftrace_event_call *call)
1400 {
1401 	struct trace_probe *tp = (struct trace_probe *)call->data;
1402 
1403 	tp->flags |= TP_FLAG_PROFILE;
1404 
1405 	if (probe_is_return(tp))
1406 		return enable_kretprobe(&tp->rp);
1407 	else
1408 		return enable_kprobe(&tp->rp.kp);
1409 }
1410 
1411 static void probe_perf_disable(struct ftrace_event_call *call)
1412 {
1413 	struct trace_probe *tp = (struct trace_probe *)call->data;
1414 
1415 	tp->flags &= ~TP_FLAG_PROFILE;
1416 
1417 	if (!(tp->flags & TP_FLAG_TRACE)) {
1418 		if (probe_is_return(tp))
1419 			disable_kretprobe(&tp->rp);
1420 		else
1421 			disable_kprobe(&tp->rp.kp);
1422 	}
1423 }
1424 #endif	/* CONFIG_PERF_EVENTS */
1425 
1426 static __kprobes
1427 int kprobe_register(struct ftrace_event_call *event, enum trace_reg type)
1428 {
1429 	switch (type) {
1430 	case TRACE_REG_REGISTER:
1431 		return probe_event_enable(event);
1432 	case TRACE_REG_UNREGISTER:
1433 		probe_event_disable(event);
1434 		return 0;
1435 
1436 #ifdef CONFIG_PERF_EVENTS
1437 	case TRACE_REG_PERF_REGISTER:
1438 		return probe_perf_enable(event);
1439 	case TRACE_REG_PERF_UNREGISTER:
1440 		probe_perf_disable(event);
1441 		return 0;
1442 #endif
1443 	}
1444 	return 0;
1445 }
1446 
1447 static __kprobes
1448 int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1449 {
1450 	struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1451 
1452 	if (tp->flags & TP_FLAG_TRACE)
1453 		kprobe_trace_func(kp, regs);
1454 #ifdef CONFIG_PERF_EVENTS
1455 	if (tp->flags & TP_FLAG_PROFILE)
1456 		kprobe_perf_func(kp, regs);
1457 #endif
1458 	return 0;	/* We don't tweek kernel, so just return 0 */
1459 }
1460 
1461 static __kprobes
1462 int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1463 {
1464 	struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1465 
1466 	if (tp->flags & TP_FLAG_TRACE)
1467 		kretprobe_trace_func(ri, regs);
1468 #ifdef CONFIG_PERF_EVENTS
1469 	if (tp->flags & TP_FLAG_PROFILE)
1470 		kretprobe_perf_func(ri, regs);
1471 #endif
1472 	return 0;	/* We don't tweek kernel, so just return 0 */
1473 }
1474 
1475 static struct trace_event_functions kretprobe_funcs = {
1476 	.trace		= print_kretprobe_event
1477 };
1478 
1479 static struct trace_event_functions kprobe_funcs = {
1480 	.trace		= print_kprobe_event
1481 };
1482 
1483 static int register_probe_event(struct trace_probe *tp)
1484 {
1485 	struct ftrace_event_call *call = &tp->call;
1486 	int ret;
1487 
1488 	/* Initialize ftrace_event_call */
1489 	if (probe_is_return(tp)) {
1490 		INIT_LIST_HEAD(&call->class->fields);
1491 		call->event.funcs = &kretprobe_funcs;
1492 		call->class->raw_init = probe_event_raw_init;
1493 		call->class->define_fields = kretprobe_event_define_fields;
1494 	} else {
1495 		INIT_LIST_HEAD(&call->class->fields);
1496 		call->event.funcs = &kprobe_funcs;
1497 		call->class->raw_init = probe_event_raw_init;
1498 		call->class->define_fields = kprobe_event_define_fields;
1499 	}
1500 	if (set_print_fmt(tp) < 0)
1501 		return -ENOMEM;
1502 	ret = register_ftrace_event(&call->event);
1503 	if (!ret) {
1504 		kfree(call->print_fmt);
1505 		return -ENODEV;
1506 	}
1507 	call->flags = 0;
1508 	call->class->reg = kprobe_register;
1509 	call->data = tp;
1510 	ret = trace_add_event_call(call);
1511 	if (ret) {
1512 		pr_info("Failed to register kprobe event: %s\n", call->name);
1513 		kfree(call->print_fmt);
1514 		unregister_ftrace_event(&call->event);
1515 	}
1516 	return ret;
1517 }
1518 
1519 static void unregister_probe_event(struct trace_probe *tp)
1520 {
1521 	/* tp->event is unregistered in trace_remove_event_call() */
1522 	trace_remove_event_call(&tp->call);
1523 	kfree(tp->call.print_fmt);
1524 }
1525 
1526 /* Make a debugfs interface for controling probe points */
1527 static __init int init_kprobe_trace(void)
1528 {
1529 	struct dentry *d_tracer;
1530 	struct dentry *entry;
1531 
1532 	d_tracer = tracing_init_dentry();
1533 	if (!d_tracer)
1534 		return 0;
1535 
1536 	entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1537 				    NULL, &kprobe_events_ops);
1538 
1539 	/* Event list interface */
1540 	if (!entry)
1541 		pr_warning("Could not create debugfs "
1542 			   "'kprobe_events' entry\n");
1543 
1544 	/* Profile interface */
1545 	entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1546 				    NULL, &kprobe_profile_ops);
1547 
1548 	if (!entry)
1549 		pr_warning("Could not create debugfs "
1550 			   "'kprobe_profile' entry\n");
1551 	return 0;
1552 }
1553 fs_initcall(init_kprobe_trace);
1554 
1555 
1556 #ifdef CONFIG_FTRACE_STARTUP_TEST
1557 
1558 static int kprobe_trace_selftest_target(int a1, int a2, int a3,
1559 					int a4, int a5, int a6)
1560 {
1561 	return a1 + a2 + a3 + a4 + a5 + a6;
1562 }
1563 
1564 static __init int kprobe_trace_self_tests_init(void)
1565 {
1566 	int ret, warn = 0;
1567 	int (*target)(int, int, int, int, int, int);
1568 	struct trace_probe *tp;
1569 
1570 	target = kprobe_trace_selftest_target;
1571 
1572 	pr_info("Testing kprobe tracing: ");
1573 
1574 	ret = command_trace_probe("p:testprobe kprobe_trace_selftest_target "
1575 				  "$stack $stack0 +0($stack)");
1576 	if (WARN_ON_ONCE(ret)) {
1577 		pr_warning("error on probing function entry.\n");
1578 		warn++;
1579 	} else {
1580 		/* Enable trace point */
1581 		tp = find_probe_event("testprobe", KPROBE_EVENT_SYSTEM);
1582 		if (WARN_ON_ONCE(tp == NULL)) {
1583 			pr_warning("error on getting new probe.\n");
1584 			warn++;
1585 		} else
1586 			probe_event_enable(&tp->call);
1587 	}
1588 
1589 	ret = command_trace_probe("r:testprobe2 kprobe_trace_selftest_target "
1590 				  "$retval");
1591 	if (WARN_ON_ONCE(ret)) {
1592 		pr_warning("error on probing function return.\n");
1593 		warn++;
1594 	} else {
1595 		/* Enable trace point */
1596 		tp = find_probe_event("testprobe2", KPROBE_EVENT_SYSTEM);
1597 		if (WARN_ON_ONCE(tp == NULL)) {
1598 			pr_warning("error on getting new probe.\n");
1599 			warn++;
1600 		} else
1601 			probe_event_enable(&tp->call);
1602 	}
1603 
1604 	if (warn)
1605 		goto end;
1606 
1607 	ret = target(1, 2, 3, 4, 5, 6);
1608 
1609 	ret = command_trace_probe("-:testprobe");
1610 	if (WARN_ON_ONCE(ret)) {
1611 		pr_warning("error on deleting a probe.\n");
1612 		warn++;
1613 	}
1614 
1615 	ret = command_trace_probe("-:testprobe2");
1616 	if (WARN_ON_ONCE(ret)) {
1617 		pr_warning("error on deleting a probe.\n");
1618 		warn++;
1619 	}
1620 
1621 end:
1622 	cleanup_all_probes();
1623 	if (warn)
1624 		pr_cont("NG: Some tests are failed. Please check them.\n");
1625 	else
1626 		pr_cont("OK\n");
1627 	return 0;
1628 }
1629 
1630 late_initcall(kprobe_trace_self_tests_init);
1631 
1632 #endif
1633