xref: /openbmc/linux/kernel/trace/trace_kprobe.c (revision 81d67439)
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 <linux/limits.h>
34 #include <asm/bitsperlong.h>
35 
36 #include "trace.h"
37 #include "trace_output.h"
38 
39 #define MAX_TRACE_ARGS 128
40 #define MAX_ARGSTR_LEN 63
41 #define MAX_EVENT_NAME_LEN 64
42 #define MAX_STRING_SIZE PATH_MAX
43 #define KPROBE_EVENT_SYSTEM "kprobes"
44 
45 /* Reserved field names */
46 #define FIELD_STRING_IP "__probe_ip"
47 #define FIELD_STRING_RETIP "__probe_ret_ip"
48 #define FIELD_STRING_FUNC "__probe_func"
49 
50 const char *reserved_field_names[] = {
51 	"common_type",
52 	"common_flags",
53 	"common_preempt_count",
54 	"common_pid",
55 	"common_tgid",
56 	FIELD_STRING_IP,
57 	FIELD_STRING_RETIP,
58 	FIELD_STRING_FUNC,
59 };
60 
61 /* Printing function type */
62 typedef int (*print_type_func_t)(struct trace_seq *, const char *, void *,
63 				 void *);
64 #define PRINT_TYPE_FUNC_NAME(type)	print_type_##type
65 #define PRINT_TYPE_FMT_NAME(type)	print_type_format_##type
66 
67 /* Printing  in basic type function template */
68 #define DEFINE_BASIC_PRINT_TYPE_FUNC(type, fmt, cast)			\
69 static __kprobes int PRINT_TYPE_FUNC_NAME(type)(struct trace_seq *s,	\
70 						const char *name,	\
71 						void *data, void *ent)\
72 {									\
73 	return trace_seq_printf(s, " %s=" fmt, name, (cast)*(type *)data);\
74 }									\
75 static const char PRINT_TYPE_FMT_NAME(type)[] = fmt;
76 
77 DEFINE_BASIC_PRINT_TYPE_FUNC(u8, "%x", unsigned int)
78 DEFINE_BASIC_PRINT_TYPE_FUNC(u16, "%x", unsigned int)
79 DEFINE_BASIC_PRINT_TYPE_FUNC(u32, "%lx", unsigned long)
80 DEFINE_BASIC_PRINT_TYPE_FUNC(u64, "%llx", unsigned long long)
81 DEFINE_BASIC_PRINT_TYPE_FUNC(s8, "%d", int)
82 DEFINE_BASIC_PRINT_TYPE_FUNC(s16, "%d", int)
83 DEFINE_BASIC_PRINT_TYPE_FUNC(s32, "%ld", long)
84 DEFINE_BASIC_PRINT_TYPE_FUNC(s64, "%lld", long long)
85 
86 /* data_rloc: data relative location, compatible with u32 */
87 #define make_data_rloc(len, roffs)	\
88 	(((u32)(len) << 16) | ((u32)(roffs) & 0xffff))
89 #define get_rloc_len(dl)	((u32)(dl) >> 16)
90 #define get_rloc_offs(dl)	((u32)(dl) & 0xffff)
91 
92 static inline void *get_rloc_data(u32 *dl)
93 {
94 	return (u8 *)dl + get_rloc_offs(*dl);
95 }
96 
97 /* For data_loc conversion */
98 static inline void *get_loc_data(u32 *dl, void *ent)
99 {
100 	return (u8 *)ent + get_rloc_offs(*dl);
101 }
102 
103 /*
104  * Convert data_rloc to data_loc:
105  *  data_rloc stores the offset from data_rloc itself, but data_loc
106  *  stores the offset from event entry.
107  */
108 #define convert_rloc_to_loc(dl, offs)	((u32)(dl) + (offs))
109 
110 /* For defining macros, define string/string_size types */
111 typedef u32 string;
112 typedef u32 string_size;
113 
114 /* Print type function for string type */
115 static __kprobes int PRINT_TYPE_FUNC_NAME(string)(struct trace_seq *s,
116 						  const char *name,
117 						  void *data, void *ent)
118 {
119 	int len = *(u32 *)data >> 16;
120 
121 	if (!len)
122 		return trace_seq_printf(s, " %s=(fault)", name);
123 	else
124 		return trace_seq_printf(s, " %s=\"%s\"", name,
125 					(const char *)get_loc_data(data, ent));
126 }
127 static const char PRINT_TYPE_FMT_NAME(string)[] = "\\\"%s\\\"";
128 
129 /* Data fetch function type */
130 typedef	void (*fetch_func_t)(struct pt_regs *, void *, void *);
131 
132 struct fetch_param {
133 	fetch_func_t	fn;
134 	void *data;
135 };
136 
137 static __kprobes void call_fetch(struct fetch_param *fprm,
138 				 struct pt_regs *regs, void *dest)
139 {
140 	return fprm->fn(regs, fprm->data, dest);
141 }
142 
143 #define FETCH_FUNC_NAME(method, type)	fetch_##method##_##type
144 /*
145  * Define macro for basic types - we don't need to define s* types, because
146  * we have to care only about bitwidth at recording time.
147  */
148 #define DEFINE_BASIC_FETCH_FUNCS(method) \
149 DEFINE_FETCH_##method(u8)		\
150 DEFINE_FETCH_##method(u16)		\
151 DEFINE_FETCH_##method(u32)		\
152 DEFINE_FETCH_##method(u64)
153 
154 #define CHECK_FETCH_FUNCS(method, fn)			\
155 	(((FETCH_FUNC_NAME(method, u8) == fn) ||	\
156 	  (FETCH_FUNC_NAME(method, u16) == fn) ||	\
157 	  (FETCH_FUNC_NAME(method, u32) == fn) ||	\
158 	  (FETCH_FUNC_NAME(method, u64) == fn) ||	\
159 	  (FETCH_FUNC_NAME(method, string) == fn) ||	\
160 	  (FETCH_FUNC_NAME(method, string_size) == fn)) \
161 	 && (fn != NULL))
162 
163 /* Data fetch function templates */
164 #define DEFINE_FETCH_reg(type)						\
165 static __kprobes void FETCH_FUNC_NAME(reg, type)(struct pt_regs *regs,	\
166 					void *offset, void *dest)	\
167 {									\
168 	*(type *)dest = (type)regs_get_register(regs,			\
169 				(unsigned int)((unsigned long)offset));	\
170 }
171 DEFINE_BASIC_FETCH_FUNCS(reg)
172 /* No string on the register */
173 #define fetch_reg_string NULL
174 #define fetch_reg_string_size NULL
175 
176 #define DEFINE_FETCH_stack(type)					\
177 static __kprobes void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,\
178 					  void *offset, void *dest)	\
179 {									\
180 	*(type *)dest = (type)regs_get_kernel_stack_nth(regs,		\
181 				(unsigned int)((unsigned long)offset));	\
182 }
183 DEFINE_BASIC_FETCH_FUNCS(stack)
184 /* No string on the stack entry */
185 #define fetch_stack_string NULL
186 #define fetch_stack_string_size NULL
187 
188 #define DEFINE_FETCH_retval(type)					\
189 static __kprobes void FETCH_FUNC_NAME(retval, type)(struct pt_regs *regs,\
190 					  void *dummy, void *dest)	\
191 {									\
192 	*(type *)dest = (type)regs_return_value(regs);			\
193 }
194 DEFINE_BASIC_FETCH_FUNCS(retval)
195 /* No string on the retval */
196 #define fetch_retval_string NULL
197 #define fetch_retval_string_size NULL
198 
199 #define DEFINE_FETCH_memory(type)					\
200 static __kprobes void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,\
201 					  void *addr, void *dest)	\
202 {									\
203 	type retval;							\
204 	if (probe_kernel_address(addr, retval))				\
205 		*(type *)dest = 0;					\
206 	else								\
207 		*(type *)dest = retval;					\
208 }
209 DEFINE_BASIC_FETCH_FUNCS(memory)
210 /*
211  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
212  * length and relative data location.
213  */
214 static __kprobes void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
215 						      void *addr, void *dest)
216 {
217 	long ret;
218 	int maxlen = get_rloc_len(*(u32 *)dest);
219 	u8 *dst = get_rloc_data(dest);
220 	u8 *src = addr;
221 	mm_segment_t old_fs = get_fs();
222 	if (!maxlen)
223 		return;
224 	/*
225 	 * Try to get string again, since the string can be changed while
226 	 * probing.
227 	 */
228 	set_fs(KERNEL_DS);
229 	pagefault_disable();
230 	do
231 		ret = __copy_from_user_inatomic(dst++, src++, 1);
232 	while (dst[-1] && ret == 0 && src - (u8 *)addr < maxlen);
233 	dst[-1] = '\0';
234 	pagefault_enable();
235 	set_fs(old_fs);
236 
237 	if (ret < 0) {	/* Failed to fetch string */
238 		((u8 *)get_rloc_data(dest))[0] = '\0';
239 		*(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
240 	} else
241 		*(u32 *)dest = make_data_rloc(src - (u8 *)addr,
242 					      get_rloc_offs(*(u32 *)dest));
243 }
244 /* Return the length of string -- including null terminal byte */
245 static __kprobes void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
246 							void *addr, void *dest)
247 {
248 	int ret, len = 0;
249 	u8 c;
250 	mm_segment_t old_fs = get_fs();
251 
252 	set_fs(KERNEL_DS);
253 	pagefault_disable();
254 	do {
255 		ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
256 		len++;
257 	} while (c && ret == 0 && len < MAX_STRING_SIZE);
258 	pagefault_enable();
259 	set_fs(old_fs);
260 
261 	if (ret < 0)	/* Failed to check the length */
262 		*(u32 *)dest = 0;
263 	else
264 		*(u32 *)dest = len;
265 }
266 
267 /* Memory fetching by symbol */
268 struct symbol_cache {
269 	char *symbol;
270 	long offset;
271 	unsigned long addr;
272 };
273 
274 static unsigned long update_symbol_cache(struct symbol_cache *sc)
275 {
276 	sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
277 	if (sc->addr)
278 		sc->addr += sc->offset;
279 	return sc->addr;
280 }
281 
282 static void free_symbol_cache(struct symbol_cache *sc)
283 {
284 	kfree(sc->symbol);
285 	kfree(sc);
286 }
287 
288 static struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
289 {
290 	struct symbol_cache *sc;
291 
292 	if (!sym || strlen(sym) == 0)
293 		return NULL;
294 	sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
295 	if (!sc)
296 		return NULL;
297 
298 	sc->symbol = kstrdup(sym, GFP_KERNEL);
299 	if (!sc->symbol) {
300 		kfree(sc);
301 		return NULL;
302 	}
303 	sc->offset = offset;
304 
305 	update_symbol_cache(sc);
306 	return sc;
307 }
308 
309 #define DEFINE_FETCH_symbol(type)					\
310 static __kprobes void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs,\
311 					  void *data, void *dest)	\
312 {									\
313 	struct symbol_cache *sc = data;					\
314 	if (sc->addr)							\
315 		fetch_memory_##type(regs, (void *)sc->addr, dest);	\
316 	else								\
317 		*(type *)dest = 0;					\
318 }
319 DEFINE_BASIC_FETCH_FUNCS(symbol)
320 DEFINE_FETCH_symbol(string)
321 DEFINE_FETCH_symbol(string_size)
322 
323 /* Dereference memory access function */
324 struct deref_fetch_param {
325 	struct fetch_param orig;
326 	long offset;
327 };
328 
329 #define DEFINE_FETCH_deref(type)					\
330 static __kprobes void FETCH_FUNC_NAME(deref, type)(struct pt_regs *regs,\
331 					    void *data, void *dest)	\
332 {									\
333 	struct deref_fetch_param *dprm = data;				\
334 	unsigned long addr;						\
335 	call_fetch(&dprm->orig, regs, &addr);				\
336 	if (addr) {							\
337 		addr += dprm->offset;					\
338 		fetch_memory_##type(regs, (void *)addr, dest);		\
339 	} else								\
340 		*(type *)dest = 0;					\
341 }
342 DEFINE_BASIC_FETCH_FUNCS(deref)
343 DEFINE_FETCH_deref(string)
344 DEFINE_FETCH_deref(string_size)
345 
346 static __kprobes void update_deref_fetch_param(struct deref_fetch_param *data)
347 {
348 	if (CHECK_FETCH_FUNCS(deref, data->orig.fn))
349 		update_deref_fetch_param(data->orig.data);
350 	else if (CHECK_FETCH_FUNCS(symbol, data->orig.fn))
351 		update_symbol_cache(data->orig.data);
352 }
353 
354 static __kprobes void free_deref_fetch_param(struct deref_fetch_param *data)
355 {
356 	if (CHECK_FETCH_FUNCS(deref, data->orig.fn))
357 		free_deref_fetch_param(data->orig.data);
358 	else if (CHECK_FETCH_FUNCS(symbol, data->orig.fn))
359 		free_symbol_cache(data->orig.data);
360 	kfree(data);
361 }
362 
363 /* Bitfield fetch function */
364 struct bitfield_fetch_param {
365 	struct fetch_param orig;
366 	unsigned char hi_shift;
367 	unsigned char low_shift;
368 };
369 
370 #define DEFINE_FETCH_bitfield(type)					\
371 static __kprobes void FETCH_FUNC_NAME(bitfield, type)(struct pt_regs *regs,\
372 					    void *data, void *dest)	\
373 {									\
374 	struct bitfield_fetch_param *bprm = data;			\
375 	type buf = 0;							\
376 	call_fetch(&bprm->orig, regs, &buf);				\
377 	if (buf) {							\
378 		buf <<= bprm->hi_shift;					\
379 		buf >>= bprm->low_shift;				\
380 	}								\
381 	*(type *)dest = buf;						\
382 }
383 DEFINE_BASIC_FETCH_FUNCS(bitfield)
384 #define fetch_bitfield_string NULL
385 #define fetch_bitfield_string_size NULL
386 
387 static __kprobes void
388 update_bitfield_fetch_param(struct bitfield_fetch_param *data)
389 {
390 	/*
391 	 * Don't check the bitfield itself, because this must be the
392 	 * last fetch function.
393 	 */
394 	if (CHECK_FETCH_FUNCS(deref, data->orig.fn))
395 		update_deref_fetch_param(data->orig.data);
396 	else if (CHECK_FETCH_FUNCS(symbol, data->orig.fn))
397 		update_symbol_cache(data->orig.data);
398 }
399 
400 static __kprobes void
401 free_bitfield_fetch_param(struct bitfield_fetch_param *data)
402 {
403 	/*
404 	 * Don't check the bitfield itself, because this must be the
405 	 * last fetch function.
406 	 */
407 	if (CHECK_FETCH_FUNCS(deref, data->orig.fn))
408 		free_deref_fetch_param(data->orig.data);
409 	else if (CHECK_FETCH_FUNCS(symbol, data->orig.fn))
410 		free_symbol_cache(data->orig.data);
411 	kfree(data);
412 }
413 
414 /* Default (unsigned long) fetch type */
415 #define __DEFAULT_FETCH_TYPE(t) u##t
416 #define _DEFAULT_FETCH_TYPE(t) __DEFAULT_FETCH_TYPE(t)
417 #define DEFAULT_FETCH_TYPE _DEFAULT_FETCH_TYPE(BITS_PER_LONG)
418 #define DEFAULT_FETCH_TYPE_STR __stringify(DEFAULT_FETCH_TYPE)
419 
420 /* Fetch types */
421 enum {
422 	FETCH_MTD_reg = 0,
423 	FETCH_MTD_stack,
424 	FETCH_MTD_retval,
425 	FETCH_MTD_memory,
426 	FETCH_MTD_symbol,
427 	FETCH_MTD_deref,
428 	FETCH_MTD_bitfield,
429 	FETCH_MTD_END,
430 };
431 
432 #define ASSIGN_FETCH_FUNC(method, type)	\
433 	[FETCH_MTD_##method] = FETCH_FUNC_NAME(method, type)
434 
435 #define __ASSIGN_FETCH_TYPE(_name, ptype, ftype, _size, sign, _fmttype)	\
436 	{.name = _name,				\
437 	 .size = _size,					\
438 	 .is_signed = sign,				\
439 	 .print = PRINT_TYPE_FUNC_NAME(ptype),		\
440 	 .fmt = PRINT_TYPE_FMT_NAME(ptype),		\
441 	 .fmttype = _fmttype,				\
442 	 .fetch = {					\
443 ASSIGN_FETCH_FUNC(reg, ftype),				\
444 ASSIGN_FETCH_FUNC(stack, ftype),			\
445 ASSIGN_FETCH_FUNC(retval, ftype),			\
446 ASSIGN_FETCH_FUNC(memory, ftype),			\
447 ASSIGN_FETCH_FUNC(symbol, ftype),			\
448 ASSIGN_FETCH_FUNC(deref, ftype),			\
449 ASSIGN_FETCH_FUNC(bitfield, ftype),			\
450 	  }						\
451 	}
452 
453 #define ASSIGN_FETCH_TYPE(ptype, ftype, sign)			\
454 	__ASSIGN_FETCH_TYPE(#ptype, ptype, ftype, sizeof(ftype), sign, #ptype)
455 
456 #define FETCH_TYPE_STRING 0
457 #define FETCH_TYPE_STRSIZE 1
458 
459 /* Fetch type information table */
460 static const struct fetch_type {
461 	const char	*name;		/* Name of type */
462 	size_t		size;		/* Byte size of type */
463 	int		is_signed;	/* Signed flag */
464 	print_type_func_t	print;	/* Print functions */
465 	const char	*fmt;		/* Fromat string */
466 	const char	*fmttype;	/* Name in format file */
467 	/* Fetch functions */
468 	fetch_func_t	fetch[FETCH_MTD_END];
469 } fetch_type_table[] = {
470 	/* Special types */
471 	[FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
472 					sizeof(u32), 1, "__data_loc char[]"),
473 	[FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
474 					string_size, sizeof(u32), 0, "u32"),
475 	/* Basic types */
476 	ASSIGN_FETCH_TYPE(u8,  u8,  0),
477 	ASSIGN_FETCH_TYPE(u16, u16, 0),
478 	ASSIGN_FETCH_TYPE(u32, u32, 0),
479 	ASSIGN_FETCH_TYPE(u64, u64, 0),
480 	ASSIGN_FETCH_TYPE(s8,  u8,  1),
481 	ASSIGN_FETCH_TYPE(s16, u16, 1),
482 	ASSIGN_FETCH_TYPE(s32, u32, 1),
483 	ASSIGN_FETCH_TYPE(s64, u64, 1),
484 };
485 
486 static const struct fetch_type *find_fetch_type(const char *type)
487 {
488 	int i;
489 
490 	if (!type)
491 		type = DEFAULT_FETCH_TYPE_STR;
492 
493 	/* Special case: bitfield */
494 	if (*type == 'b') {
495 		unsigned long bs;
496 		type = strchr(type, '/');
497 		if (!type)
498 			goto fail;
499 		type++;
500 		if (strict_strtoul(type, 0, &bs))
501 			goto fail;
502 		switch (bs) {
503 		case 8:
504 			return find_fetch_type("u8");
505 		case 16:
506 			return find_fetch_type("u16");
507 		case 32:
508 			return find_fetch_type("u32");
509 		case 64:
510 			return find_fetch_type("u64");
511 		default:
512 			goto fail;
513 		}
514 	}
515 
516 	for (i = 0; i < ARRAY_SIZE(fetch_type_table); i++)
517 		if (strcmp(type, fetch_type_table[i].name) == 0)
518 			return &fetch_type_table[i];
519 fail:
520 	return NULL;
521 }
522 
523 /* Special function : only accept unsigned long */
524 static __kprobes void fetch_stack_address(struct pt_regs *regs,
525 					  void *dummy, void *dest)
526 {
527 	*(unsigned long *)dest = kernel_stack_pointer(regs);
528 }
529 
530 static fetch_func_t get_fetch_size_function(const struct fetch_type *type,
531 					    fetch_func_t orig_fn)
532 {
533 	int i;
534 
535 	if (type != &fetch_type_table[FETCH_TYPE_STRING])
536 		return NULL;	/* Only string type needs size function */
537 	for (i = 0; i < FETCH_MTD_END; i++)
538 		if (type->fetch[i] == orig_fn)
539 			return fetch_type_table[FETCH_TYPE_STRSIZE].fetch[i];
540 
541 	WARN_ON(1);	/* This should not happen */
542 	return NULL;
543 }
544 
545 /**
546  * Kprobe event core functions
547  */
548 
549 struct probe_arg {
550 	struct fetch_param	fetch;
551 	struct fetch_param	fetch_size;
552 	unsigned int		offset;	/* Offset from argument entry */
553 	const char		*name;	/* Name of this argument */
554 	const char		*comm;	/* Command of this argument */
555 	const struct fetch_type	*type;	/* Type of this argument */
556 };
557 
558 /* Flags for trace_probe */
559 #define TP_FLAG_TRACE	1
560 #define TP_FLAG_PROFILE	2
561 #define TP_FLAG_REGISTERED 4
562 
563 struct trace_probe {
564 	struct list_head	list;
565 	struct kretprobe	rp;	/* Use rp.kp for kprobe use */
566 	unsigned long 		nhit;
567 	unsigned int		flags;	/* For TP_FLAG_* */
568 	const char		*symbol;	/* symbol name */
569 	struct ftrace_event_class	class;
570 	struct ftrace_event_call	call;
571 	ssize_t			size;		/* trace entry size */
572 	unsigned int		nr_args;
573 	struct probe_arg	args[];
574 };
575 
576 #define SIZEOF_TRACE_PROBE(n)			\
577 	(offsetof(struct trace_probe, args) +	\
578 	(sizeof(struct probe_arg) * (n)))
579 
580 
581 static __kprobes int trace_probe_is_return(struct trace_probe *tp)
582 {
583 	return tp->rp.handler != NULL;
584 }
585 
586 static __kprobes const char *trace_probe_symbol(struct trace_probe *tp)
587 {
588 	return tp->symbol ? tp->symbol : "unknown";
589 }
590 
591 static __kprobes unsigned long trace_probe_offset(struct trace_probe *tp)
592 {
593 	return tp->rp.kp.offset;
594 }
595 
596 static __kprobes bool trace_probe_is_enabled(struct trace_probe *tp)
597 {
598 	return !!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE));
599 }
600 
601 static __kprobes bool trace_probe_is_registered(struct trace_probe *tp)
602 {
603 	return !!(tp->flags & TP_FLAG_REGISTERED);
604 }
605 
606 static __kprobes bool trace_probe_has_gone(struct trace_probe *tp)
607 {
608 	return !!(kprobe_gone(&tp->rp.kp));
609 }
610 
611 static __kprobes bool trace_probe_within_module(struct trace_probe *tp,
612 						struct module *mod)
613 {
614 	int len = strlen(mod->name);
615 	const char *name = trace_probe_symbol(tp);
616 	return strncmp(mod->name, name, len) == 0 && name[len] == ':';
617 }
618 
619 static __kprobes bool trace_probe_is_on_module(struct trace_probe *tp)
620 {
621 	return !!strchr(trace_probe_symbol(tp), ':');
622 }
623 
624 static int register_probe_event(struct trace_probe *tp);
625 static void unregister_probe_event(struct trace_probe *tp);
626 
627 static DEFINE_MUTEX(probe_lock);
628 static LIST_HEAD(probe_list);
629 
630 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
631 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
632 				struct pt_regs *regs);
633 
634 /* Check the name is good for event/group/fields */
635 static int is_good_name(const char *name)
636 {
637 	if (!isalpha(*name) && *name != '_')
638 		return 0;
639 	while (*++name != '\0') {
640 		if (!isalpha(*name) && !isdigit(*name) && *name != '_')
641 			return 0;
642 	}
643 	return 1;
644 }
645 
646 /*
647  * Allocate new trace_probe and initialize it (including kprobes).
648  */
649 static struct trace_probe *alloc_trace_probe(const char *group,
650 					     const char *event,
651 					     void *addr,
652 					     const char *symbol,
653 					     unsigned long offs,
654 					     int nargs, int is_return)
655 {
656 	struct trace_probe *tp;
657 	int ret = -ENOMEM;
658 
659 	tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
660 	if (!tp)
661 		return ERR_PTR(ret);
662 
663 	if (symbol) {
664 		tp->symbol = kstrdup(symbol, GFP_KERNEL);
665 		if (!tp->symbol)
666 			goto error;
667 		tp->rp.kp.symbol_name = tp->symbol;
668 		tp->rp.kp.offset = offs;
669 	} else
670 		tp->rp.kp.addr = addr;
671 
672 	if (is_return)
673 		tp->rp.handler = kretprobe_dispatcher;
674 	else
675 		tp->rp.kp.pre_handler = kprobe_dispatcher;
676 
677 	if (!event || !is_good_name(event)) {
678 		ret = -EINVAL;
679 		goto error;
680 	}
681 
682 	tp->call.class = &tp->class;
683 	tp->call.name = kstrdup(event, GFP_KERNEL);
684 	if (!tp->call.name)
685 		goto error;
686 
687 	if (!group || !is_good_name(group)) {
688 		ret = -EINVAL;
689 		goto error;
690 	}
691 
692 	tp->class.system = kstrdup(group, GFP_KERNEL);
693 	if (!tp->class.system)
694 		goto error;
695 
696 	INIT_LIST_HEAD(&tp->list);
697 	return tp;
698 error:
699 	kfree(tp->call.name);
700 	kfree(tp->symbol);
701 	kfree(tp);
702 	return ERR_PTR(ret);
703 }
704 
705 static void update_probe_arg(struct probe_arg *arg)
706 {
707 	if (CHECK_FETCH_FUNCS(bitfield, arg->fetch.fn))
708 		update_bitfield_fetch_param(arg->fetch.data);
709 	else if (CHECK_FETCH_FUNCS(deref, arg->fetch.fn))
710 		update_deref_fetch_param(arg->fetch.data);
711 	else if (CHECK_FETCH_FUNCS(symbol, arg->fetch.fn))
712 		update_symbol_cache(arg->fetch.data);
713 }
714 
715 static void free_probe_arg(struct probe_arg *arg)
716 {
717 	if (CHECK_FETCH_FUNCS(bitfield, arg->fetch.fn))
718 		free_bitfield_fetch_param(arg->fetch.data);
719 	else if (CHECK_FETCH_FUNCS(deref, arg->fetch.fn))
720 		free_deref_fetch_param(arg->fetch.data);
721 	else if (CHECK_FETCH_FUNCS(symbol, arg->fetch.fn))
722 		free_symbol_cache(arg->fetch.data);
723 	kfree(arg->name);
724 	kfree(arg->comm);
725 }
726 
727 static void free_trace_probe(struct trace_probe *tp)
728 {
729 	int i;
730 
731 	for (i = 0; i < tp->nr_args; i++)
732 		free_probe_arg(&tp->args[i]);
733 
734 	kfree(tp->call.class->system);
735 	kfree(tp->call.name);
736 	kfree(tp->symbol);
737 	kfree(tp);
738 }
739 
740 static struct trace_probe *find_trace_probe(const char *event,
741 					    const char *group)
742 {
743 	struct trace_probe *tp;
744 
745 	list_for_each_entry(tp, &probe_list, list)
746 		if (strcmp(tp->call.name, event) == 0 &&
747 		    strcmp(tp->call.class->system, group) == 0)
748 			return tp;
749 	return NULL;
750 }
751 
752 /* Enable trace_probe - @flag must be TP_FLAG_TRACE or TP_FLAG_PROFILE */
753 static int enable_trace_probe(struct trace_probe *tp, int flag)
754 {
755 	int ret = 0;
756 
757 	tp->flags |= flag;
758 	if (trace_probe_is_enabled(tp) && trace_probe_is_registered(tp) &&
759 	    !trace_probe_has_gone(tp)) {
760 		if (trace_probe_is_return(tp))
761 			ret = enable_kretprobe(&tp->rp);
762 		else
763 			ret = enable_kprobe(&tp->rp.kp);
764 	}
765 
766 	return ret;
767 }
768 
769 /* Disable trace_probe - @flag must be TP_FLAG_TRACE or TP_FLAG_PROFILE */
770 static void disable_trace_probe(struct trace_probe *tp, int flag)
771 {
772 	tp->flags &= ~flag;
773 	if (!trace_probe_is_enabled(tp) && trace_probe_is_registered(tp)) {
774 		if (trace_probe_is_return(tp))
775 			disable_kretprobe(&tp->rp);
776 		else
777 			disable_kprobe(&tp->rp.kp);
778 	}
779 }
780 
781 /* Internal register function - just handle k*probes and flags */
782 static int __register_trace_probe(struct trace_probe *tp)
783 {
784 	int i, ret;
785 
786 	if (trace_probe_is_registered(tp))
787 		return -EINVAL;
788 
789 	for (i = 0; i < tp->nr_args; i++)
790 		update_probe_arg(&tp->args[i]);
791 
792 	/* Set/clear disabled flag according to tp->flag */
793 	if (trace_probe_is_enabled(tp))
794 		tp->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
795 	else
796 		tp->rp.kp.flags |= KPROBE_FLAG_DISABLED;
797 
798 	if (trace_probe_is_return(tp))
799 		ret = register_kretprobe(&tp->rp);
800 	else
801 		ret = register_kprobe(&tp->rp.kp);
802 
803 	if (ret == 0)
804 		tp->flags |= TP_FLAG_REGISTERED;
805 	else {
806 		pr_warning("Could not insert probe at %s+%lu: %d\n",
807 			   trace_probe_symbol(tp), trace_probe_offset(tp), ret);
808 		if (ret == -ENOENT && trace_probe_is_on_module(tp)) {
809 			pr_warning("This probe might be able to register after"
810 				   "target module is loaded. Continue.\n");
811 			ret = 0;
812 		} else if (ret == -EILSEQ) {
813 			pr_warning("Probing address(0x%p) is not an "
814 				   "instruction boundary.\n",
815 				   tp->rp.kp.addr);
816 			ret = -EINVAL;
817 		}
818 	}
819 
820 	return ret;
821 }
822 
823 /* Internal unregister function - just handle k*probes and flags */
824 static void __unregister_trace_probe(struct trace_probe *tp)
825 {
826 	if (trace_probe_is_registered(tp)) {
827 		if (trace_probe_is_return(tp))
828 			unregister_kretprobe(&tp->rp);
829 		else
830 			unregister_kprobe(&tp->rp.kp);
831 		tp->flags &= ~TP_FLAG_REGISTERED;
832 		/* Cleanup kprobe for reuse */
833 		if (tp->rp.kp.symbol_name)
834 			tp->rp.kp.addr = NULL;
835 	}
836 }
837 
838 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
839 static void unregister_trace_probe(struct trace_probe *tp)
840 {
841 	__unregister_trace_probe(tp);
842 	list_del(&tp->list);
843 	unregister_probe_event(tp);
844 }
845 
846 /* Register a trace_probe and probe_event */
847 static int register_trace_probe(struct trace_probe *tp)
848 {
849 	struct trace_probe *old_tp;
850 	int ret;
851 
852 	mutex_lock(&probe_lock);
853 
854 	/* Delete old (same name) event if exist */
855 	old_tp = find_trace_probe(tp->call.name, tp->call.class->system);
856 	if (old_tp) {
857 		unregister_trace_probe(old_tp);
858 		free_trace_probe(old_tp);
859 	}
860 
861 	/* Register new event */
862 	ret = register_probe_event(tp);
863 	if (ret) {
864 		pr_warning("Failed to register probe event(%d)\n", ret);
865 		goto end;
866 	}
867 
868 	/* Register k*probe */
869 	ret = __register_trace_probe(tp);
870 	if (ret < 0)
871 		unregister_probe_event(tp);
872 	else
873 		list_add_tail(&tp->list, &probe_list);
874 
875 end:
876 	mutex_unlock(&probe_lock);
877 	return ret;
878 }
879 
880 /* Module notifier call back, checking event on the module */
881 static int trace_probe_module_callback(struct notifier_block *nb,
882 				       unsigned long val, void *data)
883 {
884 	struct module *mod = data;
885 	struct trace_probe *tp;
886 	int ret;
887 
888 	if (val != MODULE_STATE_COMING)
889 		return NOTIFY_DONE;
890 
891 	/* Update probes on coming module */
892 	mutex_lock(&probe_lock);
893 	list_for_each_entry(tp, &probe_list, list) {
894 		if (trace_probe_within_module(tp, mod)) {
895 			__unregister_trace_probe(tp);
896 			ret = __register_trace_probe(tp);
897 			if (ret)
898 				pr_warning("Failed to re-register probe %s on"
899 					   "%s: %d\n",
900 					   tp->call.name, mod->name, ret);
901 		}
902 	}
903 	mutex_unlock(&probe_lock);
904 
905 	return NOTIFY_DONE;
906 }
907 
908 static struct notifier_block trace_probe_module_nb = {
909 	.notifier_call = trace_probe_module_callback,
910 	.priority = 1	/* Invoked after kprobe module callback */
911 };
912 
913 /* Split symbol and offset. */
914 static int split_symbol_offset(char *symbol, unsigned long *offset)
915 {
916 	char *tmp;
917 	int ret;
918 
919 	if (!offset)
920 		return -EINVAL;
921 
922 	tmp = strchr(symbol, '+');
923 	if (tmp) {
924 		/* skip sign because strict_strtol doesn't accept '+' */
925 		ret = strict_strtoul(tmp + 1, 0, offset);
926 		if (ret)
927 			return ret;
928 		*tmp = '\0';
929 	} else
930 		*offset = 0;
931 	return 0;
932 }
933 
934 #define PARAM_MAX_ARGS 16
935 #define PARAM_MAX_STACK (THREAD_SIZE / sizeof(unsigned long))
936 
937 static int parse_probe_vars(char *arg, const struct fetch_type *t,
938 			    struct fetch_param *f, int is_return)
939 {
940 	int ret = 0;
941 	unsigned long param;
942 
943 	if (strcmp(arg, "retval") == 0) {
944 		if (is_return)
945 			f->fn = t->fetch[FETCH_MTD_retval];
946 		else
947 			ret = -EINVAL;
948 	} else if (strncmp(arg, "stack", 5) == 0) {
949 		if (arg[5] == '\0') {
950 			if (strcmp(t->name, DEFAULT_FETCH_TYPE_STR) == 0)
951 				f->fn = fetch_stack_address;
952 			else
953 				ret = -EINVAL;
954 		} else if (isdigit(arg[5])) {
955 			ret = strict_strtoul(arg + 5, 10, &param);
956 			if (ret || param > PARAM_MAX_STACK)
957 				ret = -EINVAL;
958 			else {
959 				f->fn = t->fetch[FETCH_MTD_stack];
960 				f->data = (void *)param;
961 			}
962 		} else
963 			ret = -EINVAL;
964 	} else
965 		ret = -EINVAL;
966 	return ret;
967 }
968 
969 /* Recursive argument parser */
970 static int __parse_probe_arg(char *arg, const struct fetch_type *t,
971 			     struct fetch_param *f, int is_return)
972 {
973 	int ret = 0;
974 	unsigned long param;
975 	long offset;
976 	char *tmp;
977 
978 	switch (arg[0]) {
979 	case '$':
980 		ret = parse_probe_vars(arg + 1, t, f, is_return);
981 		break;
982 	case '%':	/* named register */
983 		ret = regs_query_register_offset(arg + 1);
984 		if (ret >= 0) {
985 			f->fn = t->fetch[FETCH_MTD_reg];
986 			f->data = (void *)(unsigned long)ret;
987 			ret = 0;
988 		}
989 		break;
990 	case '@':	/* memory or symbol */
991 		if (isdigit(arg[1])) {
992 			ret = strict_strtoul(arg + 1, 0, &param);
993 			if (ret)
994 				break;
995 			f->fn = t->fetch[FETCH_MTD_memory];
996 			f->data = (void *)param;
997 		} else {
998 			ret = split_symbol_offset(arg + 1, &offset);
999 			if (ret)
1000 				break;
1001 			f->data = alloc_symbol_cache(arg + 1, offset);
1002 			if (f->data)
1003 				f->fn = t->fetch[FETCH_MTD_symbol];
1004 		}
1005 		break;
1006 	case '+':	/* deref memory */
1007 		arg++;	/* Skip '+', because strict_strtol() rejects it. */
1008 	case '-':
1009 		tmp = strchr(arg, '(');
1010 		if (!tmp)
1011 			break;
1012 		*tmp = '\0';
1013 		ret = strict_strtol(arg, 0, &offset);
1014 		if (ret)
1015 			break;
1016 		arg = tmp + 1;
1017 		tmp = strrchr(arg, ')');
1018 		if (tmp) {
1019 			struct deref_fetch_param *dprm;
1020 			const struct fetch_type *t2 = find_fetch_type(NULL);
1021 			*tmp = '\0';
1022 			dprm = kzalloc(sizeof(struct deref_fetch_param),
1023 				       GFP_KERNEL);
1024 			if (!dprm)
1025 				return -ENOMEM;
1026 			dprm->offset = offset;
1027 			ret = __parse_probe_arg(arg, t2, &dprm->orig,
1028 						is_return);
1029 			if (ret)
1030 				kfree(dprm);
1031 			else {
1032 				f->fn = t->fetch[FETCH_MTD_deref];
1033 				f->data = (void *)dprm;
1034 			}
1035 		}
1036 		break;
1037 	}
1038 	if (!ret && !f->fn) {	/* Parsed, but do not find fetch method */
1039 		pr_info("%s type has no corresponding fetch method.\n",
1040 			t->name);
1041 		ret = -EINVAL;
1042 	}
1043 	return ret;
1044 }
1045 
1046 #define BYTES_TO_BITS(nb)	((BITS_PER_LONG * (nb)) / sizeof(long))
1047 
1048 /* Bitfield type needs to be parsed into a fetch function */
1049 static int __parse_bitfield_probe_arg(const char *bf,
1050 				      const struct fetch_type *t,
1051 				      struct fetch_param *f)
1052 {
1053 	struct bitfield_fetch_param *bprm;
1054 	unsigned long bw, bo;
1055 	char *tail;
1056 
1057 	if (*bf != 'b')
1058 		return 0;
1059 
1060 	bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1061 	if (!bprm)
1062 		return -ENOMEM;
1063 	bprm->orig = *f;
1064 	f->fn = t->fetch[FETCH_MTD_bitfield];
1065 	f->data = (void *)bprm;
1066 
1067 	bw = simple_strtoul(bf + 1, &tail, 0);	/* Use simple one */
1068 	if (bw == 0 || *tail != '@')
1069 		return -EINVAL;
1070 
1071 	bf = tail + 1;
1072 	bo = simple_strtoul(bf, &tail, 0);
1073 	if (tail == bf || *tail != '/')
1074 		return -EINVAL;
1075 
1076 	bprm->hi_shift = BYTES_TO_BITS(t->size) - (bw + bo);
1077 	bprm->low_shift = bprm->hi_shift + bo;
1078 	return (BYTES_TO_BITS(t->size) < (bw + bo)) ? -EINVAL : 0;
1079 }
1080 
1081 /* String length checking wrapper */
1082 static int parse_probe_arg(char *arg, struct trace_probe *tp,
1083 			   struct probe_arg *parg, int is_return)
1084 {
1085 	const char *t;
1086 	int ret;
1087 
1088 	if (strlen(arg) > MAX_ARGSTR_LEN) {
1089 		pr_info("Argument is too long.: %s\n",  arg);
1090 		return -ENOSPC;
1091 	}
1092 	parg->comm = kstrdup(arg, GFP_KERNEL);
1093 	if (!parg->comm) {
1094 		pr_info("Failed to allocate memory for command '%s'.\n", arg);
1095 		return -ENOMEM;
1096 	}
1097 	t = strchr(parg->comm, ':');
1098 	if (t) {
1099 		arg[t - parg->comm] = '\0';
1100 		t++;
1101 	}
1102 	parg->type = find_fetch_type(t);
1103 	if (!parg->type) {
1104 		pr_info("Unsupported type: %s\n", t);
1105 		return -EINVAL;
1106 	}
1107 	parg->offset = tp->size;
1108 	tp->size += parg->type->size;
1109 	ret = __parse_probe_arg(arg, parg->type, &parg->fetch, is_return);
1110 	if (ret >= 0 && t != NULL)
1111 		ret = __parse_bitfield_probe_arg(t, parg->type, &parg->fetch);
1112 	if (ret >= 0) {
1113 		parg->fetch_size.fn = get_fetch_size_function(parg->type,
1114 							      parg->fetch.fn);
1115 		parg->fetch_size.data = parg->fetch.data;
1116 	}
1117 	return ret;
1118 }
1119 
1120 /* Return 1 if name is reserved or already used by another argument */
1121 static int conflict_field_name(const char *name,
1122 			       struct probe_arg *args, int narg)
1123 {
1124 	int i;
1125 	for (i = 0; i < ARRAY_SIZE(reserved_field_names); i++)
1126 		if (strcmp(reserved_field_names[i], name) == 0)
1127 			return 1;
1128 	for (i = 0; i < narg; i++)
1129 		if (strcmp(args[i].name, name) == 0)
1130 			return 1;
1131 	return 0;
1132 }
1133 
1134 static int create_trace_probe(int argc, char **argv)
1135 {
1136 	/*
1137 	 * Argument syntax:
1138 	 *  - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
1139 	 *  - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
1140 	 * Fetch args:
1141 	 *  $retval	: fetch return value
1142 	 *  $stack	: fetch stack address
1143 	 *  $stackN	: fetch Nth of stack (N:0-)
1144 	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
1145 	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
1146 	 *  %REG	: fetch register REG
1147 	 * Dereferencing memory fetch:
1148 	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
1149 	 * Alias name of args:
1150 	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
1151 	 * Type of args:
1152 	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
1153 	 */
1154 	struct trace_probe *tp;
1155 	int i, ret = 0;
1156 	int is_return = 0, is_delete = 0;
1157 	char *symbol = NULL, *event = NULL, *group = NULL;
1158 	char *arg;
1159 	unsigned long offset = 0;
1160 	void *addr = NULL;
1161 	char buf[MAX_EVENT_NAME_LEN];
1162 
1163 	/* argc must be >= 1 */
1164 	if (argv[0][0] == 'p')
1165 		is_return = 0;
1166 	else if (argv[0][0] == 'r')
1167 		is_return = 1;
1168 	else if (argv[0][0] == '-')
1169 		is_delete = 1;
1170 	else {
1171 		pr_info("Probe definition must be started with 'p', 'r' or"
1172 			" '-'.\n");
1173 		return -EINVAL;
1174 	}
1175 
1176 	if (argv[0][1] == ':') {
1177 		event = &argv[0][2];
1178 		if (strchr(event, '/')) {
1179 			group = event;
1180 			event = strchr(group, '/') + 1;
1181 			event[-1] = '\0';
1182 			if (strlen(group) == 0) {
1183 				pr_info("Group name is not specified\n");
1184 				return -EINVAL;
1185 			}
1186 		}
1187 		if (strlen(event) == 0) {
1188 			pr_info("Event name is not specified\n");
1189 			return -EINVAL;
1190 		}
1191 	}
1192 	if (!group)
1193 		group = KPROBE_EVENT_SYSTEM;
1194 
1195 	if (is_delete) {
1196 		if (!event) {
1197 			pr_info("Delete command needs an event name.\n");
1198 			return -EINVAL;
1199 		}
1200 		mutex_lock(&probe_lock);
1201 		tp = find_trace_probe(event, group);
1202 		if (!tp) {
1203 			mutex_unlock(&probe_lock);
1204 			pr_info("Event %s/%s doesn't exist.\n", group, event);
1205 			return -ENOENT;
1206 		}
1207 		/* delete an event */
1208 		unregister_trace_probe(tp);
1209 		free_trace_probe(tp);
1210 		mutex_unlock(&probe_lock);
1211 		return 0;
1212 	}
1213 
1214 	if (argc < 2) {
1215 		pr_info("Probe point is not specified.\n");
1216 		return -EINVAL;
1217 	}
1218 	if (isdigit(argv[1][0])) {
1219 		if (is_return) {
1220 			pr_info("Return probe point must be a symbol.\n");
1221 			return -EINVAL;
1222 		}
1223 		/* an address specified */
1224 		ret = strict_strtoul(&argv[1][0], 0, (unsigned long *)&addr);
1225 		if (ret) {
1226 			pr_info("Failed to parse address.\n");
1227 			return ret;
1228 		}
1229 	} else {
1230 		/* a symbol specified */
1231 		symbol = argv[1];
1232 		/* TODO: support .init module functions */
1233 		ret = split_symbol_offset(symbol, &offset);
1234 		if (ret) {
1235 			pr_info("Failed to parse symbol.\n");
1236 			return ret;
1237 		}
1238 		if (offset && is_return) {
1239 			pr_info("Return probe must be used without offset.\n");
1240 			return -EINVAL;
1241 		}
1242 	}
1243 	argc -= 2; argv += 2;
1244 
1245 	/* setup a probe */
1246 	if (!event) {
1247 		/* Make a new event name */
1248 		if (symbol)
1249 			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
1250 				 is_return ? 'r' : 'p', symbol, offset);
1251 		else
1252 			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
1253 				 is_return ? 'r' : 'p', addr);
1254 		event = buf;
1255 	}
1256 	tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
1257 			       is_return);
1258 	if (IS_ERR(tp)) {
1259 		pr_info("Failed to allocate trace_probe.(%d)\n",
1260 			(int)PTR_ERR(tp));
1261 		return PTR_ERR(tp);
1262 	}
1263 
1264 	/* parse arguments */
1265 	ret = 0;
1266 	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
1267 		/* Increment count for freeing args in error case */
1268 		tp->nr_args++;
1269 
1270 		/* Parse argument name */
1271 		arg = strchr(argv[i], '=');
1272 		if (arg) {
1273 			*arg++ = '\0';
1274 			tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
1275 		} else {
1276 			arg = argv[i];
1277 			/* If argument name is omitted, set "argN" */
1278 			snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
1279 			tp->args[i].name = kstrdup(buf, GFP_KERNEL);
1280 		}
1281 
1282 		if (!tp->args[i].name) {
1283 			pr_info("Failed to allocate argument[%d] name.\n", i);
1284 			ret = -ENOMEM;
1285 			goto error;
1286 		}
1287 
1288 		if (!is_good_name(tp->args[i].name)) {
1289 			pr_info("Invalid argument[%d] name: %s\n",
1290 				i, tp->args[i].name);
1291 			ret = -EINVAL;
1292 			goto error;
1293 		}
1294 
1295 		if (conflict_field_name(tp->args[i].name, tp->args, i)) {
1296 			pr_info("Argument[%d] name '%s' conflicts with "
1297 				"another field.\n", i, argv[i]);
1298 			ret = -EINVAL;
1299 			goto error;
1300 		}
1301 
1302 		/* Parse fetch argument */
1303 		ret = parse_probe_arg(arg, tp, &tp->args[i], is_return);
1304 		if (ret) {
1305 			pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
1306 			goto error;
1307 		}
1308 	}
1309 
1310 	ret = register_trace_probe(tp);
1311 	if (ret)
1312 		goto error;
1313 	return 0;
1314 
1315 error:
1316 	free_trace_probe(tp);
1317 	return ret;
1318 }
1319 
1320 static void release_all_trace_probes(void)
1321 {
1322 	struct trace_probe *tp;
1323 
1324 	mutex_lock(&probe_lock);
1325 	/* TODO: Use batch unregistration */
1326 	while (!list_empty(&probe_list)) {
1327 		tp = list_entry(probe_list.next, struct trace_probe, list);
1328 		unregister_trace_probe(tp);
1329 		free_trace_probe(tp);
1330 	}
1331 	mutex_unlock(&probe_lock);
1332 }
1333 
1334 /* Probes listing interfaces */
1335 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
1336 {
1337 	mutex_lock(&probe_lock);
1338 	return seq_list_start(&probe_list, *pos);
1339 }
1340 
1341 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
1342 {
1343 	return seq_list_next(v, &probe_list, pos);
1344 }
1345 
1346 static void probes_seq_stop(struct seq_file *m, void *v)
1347 {
1348 	mutex_unlock(&probe_lock);
1349 }
1350 
1351 static int probes_seq_show(struct seq_file *m, void *v)
1352 {
1353 	struct trace_probe *tp = v;
1354 	int i;
1355 
1356 	seq_printf(m, "%c", trace_probe_is_return(tp) ? 'r' : 'p');
1357 	seq_printf(m, ":%s/%s", tp->call.class->system, tp->call.name);
1358 
1359 	if (!tp->symbol)
1360 		seq_printf(m, " 0x%p", tp->rp.kp.addr);
1361 	else if (tp->rp.kp.offset)
1362 		seq_printf(m, " %s+%u", trace_probe_symbol(tp),
1363 			   tp->rp.kp.offset);
1364 	else
1365 		seq_printf(m, " %s", trace_probe_symbol(tp));
1366 
1367 	for (i = 0; i < tp->nr_args; i++)
1368 		seq_printf(m, " %s=%s", tp->args[i].name, tp->args[i].comm);
1369 	seq_printf(m, "\n");
1370 
1371 	return 0;
1372 }
1373 
1374 static const struct seq_operations probes_seq_op = {
1375 	.start  = probes_seq_start,
1376 	.next   = probes_seq_next,
1377 	.stop   = probes_seq_stop,
1378 	.show   = probes_seq_show
1379 };
1380 
1381 static int probes_open(struct inode *inode, struct file *file)
1382 {
1383 	if ((file->f_mode & FMODE_WRITE) &&
1384 	    (file->f_flags & O_TRUNC))
1385 		release_all_trace_probes();
1386 
1387 	return seq_open(file, &probes_seq_op);
1388 }
1389 
1390 static int command_trace_probe(const char *buf)
1391 {
1392 	char **argv;
1393 	int argc = 0, ret = 0;
1394 
1395 	argv = argv_split(GFP_KERNEL, buf, &argc);
1396 	if (!argv)
1397 		return -ENOMEM;
1398 
1399 	if (argc)
1400 		ret = create_trace_probe(argc, argv);
1401 
1402 	argv_free(argv);
1403 	return ret;
1404 }
1405 
1406 #define WRITE_BUFSIZE 4096
1407 
1408 static ssize_t probes_write(struct file *file, const char __user *buffer,
1409 			    size_t count, loff_t *ppos)
1410 {
1411 	char *kbuf, *tmp;
1412 	int ret;
1413 	size_t done;
1414 	size_t size;
1415 
1416 	kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
1417 	if (!kbuf)
1418 		return -ENOMEM;
1419 
1420 	ret = done = 0;
1421 	while (done < count) {
1422 		size = count - done;
1423 		if (size >= WRITE_BUFSIZE)
1424 			size = WRITE_BUFSIZE - 1;
1425 		if (copy_from_user(kbuf, buffer + done, size)) {
1426 			ret = -EFAULT;
1427 			goto out;
1428 		}
1429 		kbuf[size] = '\0';
1430 		tmp = strchr(kbuf, '\n');
1431 		if (tmp) {
1432 			*tmp = '\0';
1433 			size = tmp - kbuf + 1;
1434 		} else if (done + size < count) {
1435 			pr_warning("Line length is too long: "
1436 				   "Should be less than %d.", WRITE_BUFSIZE);
1437 			ret = -EINVAL;
1438 			goto out;
1439 		}
1440 		done += size;
1441 		/* Remove comments */
1442 		tmp = strchr(kbuf, '#');
1443 		if (tmp)
1444 			*tmp = '\0';
1445 
1446 		ret = command_trace_probe(kbuf);
1447 		if (ret)
1448 			goto out;
1449 	}
1450 	ret = done;
1451 out:
1452 	kfree(kbuf);
1453 	return ret;
1454 }
1455 
1456 static const struct file_operations kprobe_events_ops = {
1457 	.owner          = THIS_MODULE,
1458 	.open           = probes_open,
1459 	.read           = seq_read,
1460 	.llseek         = seq_lseek,
1461 	.release        = seq_release,
1462 	.write		= probes_write,
1463 };
1464 
1465 /* Probes profiling interfaces */
1466 static int probes_profile_seq_show(struct seq_file *m, void *v)
1467 {
1468 	struct trace_probe *tp = v;
1469 
1470 	seq_printf(m, "  %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
1471 		   tp->rp.kp.nmissed);
1472 
1473 	return 0;
1474 }
1475 
1476 static const struct seq_operations profile_seq_op = {
1477 	.start  = probes_seq_start,
1478 	.next   = probes_seq_next,
1479 	.stop   = probes_seq_stop,
1480 	.show   = probes_profile_seq_show
1481 };
1482 
1483 static int profile_open(struct inode *inode, struct file *file)
1484 {
1485 	return seq_open(file, &profile_seq_op);
1486 }
1487 
1488 static const struct file_operations kprobe_profile_ops = {
1489 	.owner          = THIS_MODULE,
1490 	.open           = profile_open,
1491 	.read           = seq_read,
1492 	.llseek         = seq_lseek,
1493 	.release        = seq_release,
1494 };
1495 
1496 /* Sum up total data length for dynamic arraies (strings) */
1497 static __kprobes int __get_data_size(struct trace_probe *tp,
1498 				     struct pt_regs *regs)
1499 {
1500 	int i, ret = 0;
1501 	u32 len;
1502 
1503 	for (i = 0; i < tp->nr_args; i++)
1504 		if (unlikely(tp->args[i].fetch_size.fn)) {
1505 			call_fetch(&tp->args[i].fetch_size, regs, &len);
1506 			ret += len;
1507 		}
1508 
1509 	return ret;
1510 }
1511 
1512 /* Store the value of each argument */
1513 static __kprobes void store_trace_args(int ent_size, struct trace_probe *tp,
1514 				       struct pt_regs *regs,
1515 				       u8 *data, int maxlen)
1516 {
1517 	int i;
1518 	u32 end = tp->size;
1519 	u32 *dl;	/* Data (relative) location */
1520 
1521 	for (i = 0; i < tp->nr_args; i++) {
1522 		if (unlikely(tp->args[i].fetch_size.fn)) {
1523 			/*
1524 			 * First, we set the relative location and
1525 			 * maximum data length to *dl
1526 			 */
1527 			dl = (u32 *)(data + tp->args[i].offset);
1528 			*dl = make_data_rloc(maxlen, end - tp->args[i].offset);
1529 			/* Then try to fetch string or dynamic array data */
1530 			call_fetch(&tp->args[i].fetch, regs, dl);
1531 			/* Reduce maximum length */
1532 			end += get_rloc_len(*dl);
1533 			maxlen -= get_rloc_len(*dl);
1534 			/* Trick here, convert data_rloc to data_loc */
1535 			*dl = convert_rloc_to_loc(*dl,
1536 				 ent_size + tp->args[i].offset);
1537 		} else
1538 			/* Just fetching data normally */
1539 			call_fetch(&tp->args[i].fetch, regs,
1540 				   data + tp->args[i].offset);
1541 	}
1542 }
1543 
1544 /* Kprobe handler */
1545 static __kprobes void kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
1546 {
1547 	struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1548 	struct kprobe_trace_entry_head *entry;
1549 	struct ring_buffer_event *event;
1550 	struct ring_buffer *buffer;
1551 	int size, dsize, pc;
1552 	unsigned long irq_flags;
1553 	struct ftrace_event_call *call = &tp->call;
1554 
1555 	tp->nhit++;
1556 
1557 	local_save_flags(irq_flags);
1558 	pc = preempt_count();
1559 
1560 	dsize = __get_data_size(tp, regs);
1561 	size = sizeof(*entry) + tp->size + dsize;
1562 
1563 	event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
1564 						  size, irq_flags, pc);
1565 	if (!event)
1566 		return;
1567 
1568 	entry = ring_buffer_event_data(event);
1569 	entry->ip = (unsigned long)kp->addr;
1570 	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1571 
1572 	if (!filter_current_check_discard(buffer, call, entry, event))
1573 		trace_nowake_buffer_unlock_commit_regs(buffer, event,
1574 						       irq_flags, pc, regs);
1575 }
1576 
1577 /* Kretprobe handler */
1578 static __kprobes void kretprobe_trace_func(struct kretprobe_instance *ri,
1579 					  struct pt_regs *regs)
1580 {
1581 	struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1582 	struct kretprobe_trace_entry_head *entry;
1583 	struct ring_buffer_event *event;
1584 	struct ring_buffer *buffer;
1585 	int size, pc, dsize;
1586 	unsigned long irq_flags;
1587 	struct ftrace_event_call *call = &tp->call;
1588 
1589 	local_save_flags(irq_flags);
1590 	pc = preempt_count();
1591 
1592 	dsize = __get_data_size(tp, regs);
1593 	size = sizeof(*entry) + tp->size + dsize;
1594 
1595 	event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
1596 						  size, irq_flags, pc);
1597 	if (!event)
1598 		return;
1599 
1600 	entry = ring_buffer_event_data(event);
1601 	entry->func = (unsigned long)tp->rp.kp.addr;
1602 	entry->ret_ip = (unsigned long)ri->ret_addr;
1603 	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1604 
1605 	if (!filter_current_check_discard(buffer, call, entry, event))
1606 		trace_nowake_buffer_unlock_commit_regs(buffer, event,
1607 						       irq_flags, pc, regs);
1608 }
1609 
1610 /* Event entry printers */
1611 enum print_line_t
1612 print_kprobe_event(struct trace_iterator *iter, int flags,
1613 		   struct trace_event *event)
1614 {
1615 	struct kprobe_trace_entry_head *field;
1616 	struct trace_seq *s = &iter->seq;
1617 	struct trace_probe *tp;
1618 	u8 *data;
1619 	int i;
1620 
1621 	field = (struct kprobe_trace_entry_head *)iter->ent;
1622 	tp = container_of(event, struct trace_probe, call.event);
1623 
1624 	if (!trace_seq_printf(s, "%s: (", tp->call.name))
1625 		goto partial;
1626 
1627 	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1628 		goto partial;
1629 
1630 	if (!trace_seq_puts(s, ")"))
1631 		goto partial;
1632 
1633 	data = (u8 *)&field[1];
1634 	for (i = 0; i < tp->nr_args; i++)
1635 		if (!tp->args[i].type->print(s, tp->args[i].name,
1636 					     data + tp->args[i].offset, field))
1637 			goto partial;
1638 
1639 	if (!trace_seq_puts(s, "\n"))
1640 		goto partial;
1641 
1642 	return TRACE_TYPE_HANDLED;
1643 partial:
1644 	return TRACE_TYPE_PARTIAL_LINE;
1645 }
1646 
1647 enum print_line_t
1648 print_kretprobe_event(struct trace_iterator *iter, int flags,
1649 		      struct trace_event *event)
1650 {
1651 	struct kretprobe_trace_entry_head *field;
1652 	struct trace_seq *s = &iter->seq;
1653 	struct trace_probe *tp;
1654 	u8 *data;
1655 	int i;
1656 
1657 	field = (struct kretprobe_trace_entry_head *)iter->ent;
1658 	tp = container_of(event, struct trace_probe, call.event);
1659 
1660 	if (!trace_seq_printf(s, "%s: (", tp->call.name))
1661 		goto partial;
1662 
1663 	if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1664 		goto partial;
1665 
1666 	if (!trace_seq_puts(s, " <- "))
1667 		goto partial;
1668 
1669 	if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1670 		goto partial;
1671 
1672 	if (!trace_seq_puts(s, ")"))
1673 		goto partial;
1674 
1675 	data = (u8 *)&field[1];
1676 	for (i = 0; i < tp->nr_args; i++)
1677 		if (!tp->args[i].type->print(s, tp->args[i].name,
1678 					     data + tp->args[i].offset, field))
1679 			goto partial;
1680 
1681 	if (!trace_seq_puts(s, "\n"))
1682 		goto partial;
1683 
1684 	return TRACE_TYPE_HANDLED;
1685 partial:
1686 	return TRACE_TYPE_PARTIAL_LINE;
1687 }
1688 
1689 #undef DEFINE_FIELD
1690 #define DEFINE_FIELD(type, item, name, is_signed)			\
1691 	do {								\
1692 		ret = trace_define_field(event_call, #type, name,	\
1693 					 offsetof(typeof(field), item),	\
1694 					 sizeof(field.item), is_signed, \
1695 					 FILTER_OTHER);			\
1696 		if (ret)						\
1697 			return ret;					\
1698 	} while (0)
1699 
1700 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
1701 {
1702 	int ret, i;
1703 	struct kprobe_trace_entry_head field;
1704 	struct trace_probe *tp = (struct trace_probe *)event_call->data;
1705 
1706 	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1707 	/* Set argument names as fields */
1708 	for (i = 0; i < tp->nr_args; i++) {
1709 		ret = trace_define_field(event_call, tp->args[i].type->fmttype,
1710 					 tp->args[i].name,
1711 					 sizeof(field) + tp->args[i].offset,
1712 					 tp->args[i].type->size,
1713 					 tp->args[i].type->is_signed,
1714 					 FILTER_OTHER);
1715 		if (ret)
1716 			return ret;
1717 	}
1718 	return 0;
1719 }
1720 
1721 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1722 {
1723 	int ret, i;
1724 	struct kretprobe_trace_entry_head field;
1725 	struct trace_probe *tp = (struct trace_probe *)event_call->data;
1726 
1727 	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1728 	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1729 	/* Set argument names as fields */
1730 	for (i = 0; i < tp->nr_args; i++) {
1731 		ret = trace_define_field(event_call, tp->args[i].type->fmttype,
1732 					 tp->args[i].name,
1733 					 sizeof(field) + tp->args[i].offset,
1734 					 tp->args[i].type->size,
1735 					 tp->args[i].type->is_signed,
1736 					 FILTER_OTHER);
1737 		if (ret)
1738 			return ret;
1739 	}
1740 	return 0;
1741 }
1742 
1743 static int __set_print_fmt(struct trace_probe *tp, char *buf, int len)
1744 {
1745 	int i;
1746 	int pos = 0;
1747 
1748 	const char *fmt, *arg;
1749 
1750 	if (!trace_probe_is_return(tp)) {
1751 		fmt = "(%lx)";
1752 		arg = "REC->" FIELD_STRING_IP;
1753 	} else {
1754 		fmt = "(%lx <- %lx)";
1755 		arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
1756 	}
1757 
1758 	/* When len=0, we just calculate the needed length */
1759 #define LEN_OR_ZERO (len ? len - pos : 0)
1760 
1761 	pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
1762 
1763 	for (i = 0; i < tp->nr_args; i++) {
1764 		pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
1765 				tp->args[i].name, tp->args[i].type->fmt);
1766 	}
1767 
1768 	pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
1769 
1770 	for (i = 0; i < tp->nr_args; i++) {
1771 		if (strcmp(tp->args[i].type->name, "string") == 0)
1772 			pos += snprintf(buf + pos, LEN_OR_ZERO,
1773 					", __get_str(%s)",
1774 					tp->args[i].name);
1775 		else
1776 			pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
1777 					tp->args[i].name);
1778 	}
1779 
1780 #undef LEN_OR_ZERO
1781 
1782 	/* return the length of print_fmt */
1783 	return pos;
1784 }
1785 
1786 static int set_print_fmt(struct trace_probe *tp)
1787 {
1788 	int len;
1789 	char *print_fmt;
1790 
1791 	/* First: called with 0 length to calculate the needed length */
1792 	len = __set_print_fmt(tp, NULL, 0);
1793 	print_fmt = kmalloc(len + 1, GFP_KERNEL);
1794 	if (!print_fmt)
1795 		return -ENOMEM;
1796 
1797 	/* Second: actually write the @print_fmt */
1798 	__set_print_fmt(tp, print_fmt, len + 1);
1799 	tp->call.print_fmt = print_fmt;
1800 
1801 	return 0;
1802 }
1803 
1804 #ifdef CONFIG_PERF_EVENTS
1805 
1806 /* Kprobe profile handler */
1807 static __kprobes void kprobe_perf_func(struct kprobe *kp,
1808 					 struct pt_regs *regs)
1809 {
1810 	struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1811 	struct ftrace_event_call *call = &tp->call;
1812 	struct kprobe_trace_entry_head *entry;
1813 	struct hlist_head *head;
1814 	int size, __size, dsize;
1815 	int rctx;
1816 
1817 	dsize = __get_data_size(tp, regs);
1818 	__size = sizeof(*entry) + tp->size + dsize;
1819 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1820 	size -= sizeof(u32);
1821 	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1822 		     "profile buffer not large enough"))
1823 		return;
1824 
1825 	entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1826 	if (!entry)
1827 		return;
1828 
1829 	entry->ip = (unsigned long)kp->addr;
1830 	memset(&entry[1], 0, dsize);
1831 	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1832 
1833 	head = this_cpu_ptr(call->perf_events);
1834 	perf_trace_buf_submit(entry, size, rctx, entry->ip, 1, regs, head);
1835 }
1836 
1837 /* Kretprobe profile handler */
1838 static __kprobes void kretprobe_perf_func(struct kretprobe_instance *ri,
1839 					    struct pt_regs *regs)
1840 {
1841 	struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1842 	struct ftrace_event_call *call = &tp->call;
1843 	struct kretprobe_trace_entry_head *entry;
1844 	struct hlist_head *head;
1845 	int size, __size, dsize;
1846 	int rctx;
1847 
1848 	dsize = __get_data_size(tp, regs);
1849 	__size = sizeof(*entry) + tp->size + dsize;
1850 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1851 	size -= sizeof(u32);
1852 	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1853 		     "profile buffer not large enough"))
1854 		return;
1855 
1856 	entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1857 	if (!entry)
1858 		return;
1859 
1860 	entry->func = (unsigned long)tp->rp.kp.addr;
1861 	entry->ret_ip = (unsigned long)ri->ret_addr;
1862 	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1863 
1864 	head = this_cpu_ptr(call->perf_events);
1865 	perf_trace_buf_submit(entry, size, rctx, entry->ret_ip, 1, regs, head);
1866 }
1867 #endif	/* CONFIG_PERF_EVENTS */
1868 
1869 static __kprobes
1870 int kprobe_register(struct ftrace_event_call *event, enum trace_reg type)
1871 {
1872 	struct trace_probe *tp = (struct trace_probe *)event->data;
1873 
1874 	switch (type) {
1875 	case TRACE_REG_REGISTER:
1876 		return enable_trace_probe(tp, TP_FLAG_TRACE);
1877 	case TRACE_REG_UNREGISTER:
1878 		disable_trace_probe(tp, TP_FLAG_TRACE);
1879 		return 0;
1880 
1881 #ifdef CONFIG_PERF_EVENTS
1882 	case TRACE_REG_PERF_REGISTER:
1883 		return enable_trace_probe(tp, TP_FLAG_PROFILE);
1884 	case TRACE_REG_PERF_UNREGISTER:
1885 		disable_trace_probe(tp, TP_FLAG_PROFILE);
1886 		return 0;
1887 #endif
1888 	}
1889 	return 0;
1890 }
1891 
1892 static __kprobes
1893 int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1894 {
1895 	struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1896 
1897 	if (tp->flags & TP_FLAG_TRACE)
1898 		kprobe_trace_func(kp, regs);
1899 #ifdef CONFIG_PERF_EVENTS
1900 	if (tp->flags & TP_FLAG_PROFILE)
1901 		kprobe_perf_func(kp, regs);
1902 #endif
1903 	return 0;	/* We don't tweek kernel, so just return 0 */
1904 }
1905 
1906 static __kprobes
1907 int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1908 {
1909 	struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1910 
1911 	if (tp->flags & TP_FLAG_TRACE)
1912 		kretprobe_trace_func(ri, regs);
1913 #ifdef CONFIG_PERF_EVENTS
1914 	if (tp->flags & TP_FLAG_PROFILE)
1915 		kretprobe_perf_func(ri, regs);
1916 #endif
1917 	return 0;	/* We don't tweek kernel, so just return 0 */
1918 }
1919 
1920 static struct trace_event_functions kretprobe_funcs = {
1921 	.trace		= print_kretprobe_event
1922 };
1923 
1924 static struct trace_event_functions kprobe_funcs = {
1925 	.trace		= print_kprobe_event
1926 };
1927 
1928 static int register_probe_event(struct trace_probe *tp)
1929 {
1930 	struct ftrace_event_call *call = &tp->call;
1931 	int ret;
1932 
1933 	/* Initialize ftrace_event_call */
1934 	INIT_LIST_HEAD(&call->class->fields);
1935 	if (trace_probe_is_return(tp)) {
1936 		call->event.funcs = &kretprobe_funcs;
1937 		call->class->define_fields = kretprobe_event_define_fields;
1938 	} else {
1939 		call->event.funcs = &kprobe_funcs;
1940 		call->class->define_fields = kprobe_event_define_fields;
1941 	}
1942 	if (set_print_fmt(tp) < 0)
1943 		return -ENOMEM;
1944 	ret = register_ftrace_event(&call->event);
1945 	if (!ret) {
1946 		kfree(call->print_fmt);
1947 		return -ENODEV;
1948 	}
1949 	call->flags = 0;
1950 	call->class->reg = kprobe_register;
1951 	call->data = tp;
1952 	ret = trace_add_event_call(call);
1953 	if (ret) {
1954 		pr_info("Failed to register kprobe event: %s\n", call->name);
1955 		kfree(call->print_fmt);
1956 		unregister_ftrace_event(&call->event);
1957 	}
1958 	return ret;
1959 }
1960 
1961 static void unregister_probe_event(struct trace_probe *tp)
1962 {
1963 	/* tp->event is unregistered in trace_remove_event_call() */
1964 	trace_remove_event_call(&tp->call);
1965 	kfree(tp->call.print_fmt);
1966 }
1967 
1968 /* Make a debugfs interface for controlling probe points */
1969 static __init int init_kprobe_trace(void)
1970 {
1971 	struct dentry *d_tracer;
1972 	struct dentry *entry;
1973 
1974 	if (register_module_notifier(&trace_probe_module_nb))
1975 		return -EINVAL;
1976 
1977 	d_tracer = tracing_init_dentry();
1978 	if (!d_tracer)
1979 		return 0;
1980 
1981 	entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1982 				    NULL, &kprobe_events_ops);
1983 
1984 	/* Event list interface */
1985 	if (!entry)
1986 		pr_warning("Could not create debugfs "
1987 			   "'kprobe_events' entry\n");
1988 
1989 	/* Profile interface */
1990 	entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1991 				    NULL, &kprobe_profile_ops);
1992 
1993 	if (!entry)
1994 		pr_warning("Could not create debugfs "
1995 			   "'kprobe_profile' entry\n");
1996 	return 0;
1997 }
1998 fs_initcall(init_kprobe_trace);
1999 
2000 
2001 #ifdef CONFIG_FTRACE_STARTUP_TEST
2002 
2003 /*
2004  * The "__used" keeps gcc from removing the function symbol
2005  * from the kallsyms table.
2006  */
2007 static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
2008 					       int a4, int a5, int a6)
2009 {
2010 	return a1 + a2 + a3 + a4 + a5 + a6;
2011 }
2012 
2013 static __init int kprobe_trace_self_tests_init(void)
2014 {
2015 	int ret, warn = 0;
2016 	int (*target)(int, int, int, int, int, int);
2017 	struct trace_probe *tp;
2018 
2019 	target = kprobe_trace_selftest_target;
2020 
2021 	pr_info("Testing kprobe tracing: ");
2022 
2023 	ret = command_trace_probe("p:testprobe kprobe_trace_selftest_target "
2024 				  "$stack $stack0 +0($stack)");
2025 	if (WARN_ON_ONCE(ret)) {
2026 		pr_warning("error on probing function entry.\n");
2027 		warn++;
2028 	} else {
2029 		/* Enable trace point */
2030 		tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
2031 		if (WARN_ON_ONCE(tp == NULL)) {
2032 			pr_warning("error on getting new probe.\n");
2033 			warn++;
2034 		} else
2035 			enable_trace_probe(tp, TP_FLAG_TRACE);
2036 	}
2037 
2038 	ret = command_trace_probe("r:testprobe2 kprobe_trace_selftest_target "
2039 				  "$retval");
2040 	if (WARN_ON_ONCE(ret)) {
2041 		pr_warning("error on probing function return.\n");
2042 		warn++;
2043 	} else {
2044 		/* Enable trace point */
2045 		tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
2046 		if (WARN_ON_ONCE(tp == NULL)) {
2047 			pr_warning("error on getting new probe.\n");
2048 			warn++;
2049 		} else
2050 			enable_trace_probe(tp, TP_FLAG_TRACE);
2051 	}
2052 
2053 	if (warn)
2054 		goto end;
2055 
2056 	ret = target(1, 2, 3, 4, 5, 6);
2057 
2058 	ret = command_trace_probe("-:testprobe");
2059 	if (WARN_ON_ONCE(ret)) {
2060 		pr_warning("error on deleting a probe.\n");
2061 		warn++;
2062 	}
2063 
2064 	ret = command_trace_probe("-:testprobe2");
2065 	if (WARN_ON_ONCE(ret)) {
2066 		pr_warning("error on deleting a probe.\n");
2067 		warn++;
2068 	}
2069 
2070 end:
2071 	release_all_trace_probes();
2072 	if (warn)
2073 		pr_cont("NG: Some tests are failed. Please check them.\n");
2074 	else
2075 		pr_cont("OK\n");
2076 	return 0;
2077 }
2078 
2079 late_initcall(kprobe_trace_self_tests_init);
2080 
2081 #endif
2082