xref: /openbmc/linux/kernel/trace/trace_kprobe.c (revision 615c36f5)
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 int unregister_trace_probe(struct trace_probe *tp)
840 {
841 	/* Enabled event can not be unregistered */
842 	if (trace_probe_is_enabled(tp))
843 		return -EBUSY;
844 
845 	__unregister_trace_probe(tp);
846 	list_del(&tp->list);
847 	unregister_probe_event(tp);
848 
849 	return 0;
850 }
851 
852 /* Register a trace_probe and probe_event */
853 static int register_trace_probe(struct trace_probe *tp)
854 {
855 	struct trace_probe *old_tp;
856 	int ret;
857 
858 	mutex_lock(&probe_lock);
859 
860 	/* Delete old (same name) event if exist */
861 	old_tp = find_trace_probe(tp->call.name, tp->call.class->system);
862 	if (old_tp) {
863 		ret = unregister_trace_probe(old_tp);
864 		if (ret < 0)
865 			goto end;
866 		free_trace_probe(old_tp);
867 	}
868 
869 	/* Register new event */
870 	ret = register_probe_event(tp);
871 	if (ret) {
872 		pr_warning("Failed to register probe event(%d)\n", ret);
873 		goto end;
874 	}
875 
876 	/* Register k*probe */
877 	ret = __register_trace_probe(tp);
878 	if (ret < 0)
879 		unregister_probe_event(tp);
880 	else
881 		list_add_tail(&tp->list, &probe_list);
882 
883 end:
884 	mutex_unlock(&probe_lock);
885 	return ret;
886 }
887 
888 /* Module notifier call back, checking event on the module */
889 static int trace_probe_module_callback(struct notifier_block *nb,
890 				       unsigned long val, void *data)
891 {
892 	struct module *mod = data;
893 	struct trace_probe *tp;
894 	int ret;
895 
896 	if (val != MODULE_STATE_COMING)
897 		return NOTIFY_DONE;
898 
899 	/* Update probes on coming module */
900 	mutex_lock(&probe_lock);
901 	list_for_each_entry(tp, &probe_list, list) {
902 		if (trace_probe_within_module(tp, mod)) {
903 			/* Don't need to check busy - this should have gone. */
904 			__unregister_trace_probe(tp);
905 			ret = __register_trace_probe(tp);
906 			if (ret)
907 				pr_warning("Failed to re-register probe %s on"
908 					   "%s: %d\n",
909 					   tp->call.name, mod->name, ret);
910 		}
911 	}
912 	mutex_unlock(&probe_lock);
913 
914 	return NOTIFY_DONE;
915 }
916 
917 static struct notifier_block trace_probe_module_nb = {
918 	.notifier_call = trace_probe_module_callback,
919 	.priority = 1	/* Invoked after kprobe module callback */
920 };
921 
922 /* Split symbol and offset. */
923 static int split_symbol_offset(char *symbol, unsigned long *offset)
924 {
925 	char *tmp;
926 	int ret;
927 
928 	if (!offset)
929 		return -EINVAL;
930 
931 	tmp = strchr(symbol, '+');
932 	if (tmp) {
933 		/* skip sign because strict_strtol doesn't accept '+' */
934 		ret = strict_strtoul(tmp + 1, 0, offset);
935 		if (ret)
936 			return ret;
937 		*tmp = '\0';
938 	} else
939 		*offset = 0;
940 	return 0;
941 }
942 
943 #define PARAM_MAX_ARGS 16
944 #define PARAM_MAX_STACK (THREAD_SIZE / sizeof(unsigned long))
945 
946 static int parse_probe_vars(char *arg, const struct fetch_type *t,
947 			    struct fetch_param *f, int is_return)
948 {
949 	int ret = 0;
950 	unsigned long param;
951 
952 	if (strcmp(arg, "retval") == 0) {
953 		if (is_return)
954 			f->fn = t->fetch[FETCH_MTD_retval];
955 		else
956 			ret = -EINVAL;
957 	} else if (strncmp(arg, "stack", 5) == 0) {
958 		if (arg[5] == '\0') {
959 			if (strcmp(t->name, DEFAULT_FETCH_TYPE_STR) == 0)
960 				f->fn = fetch_stack_address;
961 			else
962 				ret = -EINVAL;
963 		} else if (isdigit(arg[5])) {
964 			ret = strict_strtoul(arg + 5, 10, &param);
965 			if (ret || param > PARAM_MAX_STACK)
966 				ret = -EINVAL;
967 			else {
968 				f->fn = t->fetch[FETCH_MTD_stack];
969 				f->data = (void *)param;
970 			}
971 		} else
972 			ret = -EINVAL;
973 	} else
974 		ret = -EINVAL;
975 	return ret;
976 }
977 
978 /* Recursive argument parser */
979 static int __parse_probe_arg(char *arg, const struct fetch_type *t,
980 			     struct fetch_param *f, int is_return)
981 {
982 	int ret = 0;
983 	unsigned long param;
984 	long offset;
985 	char *tmp;
986 
987 	switch (arg[0]) {
988 	case '$':
989 		ret = parse_probe_vars(arg + 1, t, f, is_return);
990 		break;
991 	case '%':	/* named register */
992 		ret = regs_query_register_offset(arg + 1);
993 		if (ret >= 0) {
994 			f->fn = t->fetch[FETCH_MTD_reg];
995 			f->data = (void *)(unsigned long)ret;
996 			ret = 0;
997 		}
998 		break;
999 	case '@':	/* memory or symbol */
1000 		if (isdigit(arg[1])) {
1001 			ret = strict_strtoul(arg + 1, 0, &param);
1002 			if (ret)
1003 				break;
1004 			f->fn = t->fetch[FETCH_MTD_memory];
1005 			f->data = (void *)param;
1006 		} else {
1007 			ret = split_symbol_offset(arg + 1, &offset);
1008 			if (ret)
1009 				break;
1010 			f->data = alloc_symbol_cache(arg + 1, offset);
1011 			if (f->data)
1012 				f->fn = t->fetch[FETCH_MTD_symbol];
1013 		}
1014 		break;
1015 	case '+':	/* deref memory */
1016 		arg++;	/* Skip '+', because strict_strtol() rejects it. */
1017 	case '-':
1018 		tmp = strchr(arg, '(');
1019 		if (!tmp)
1020 			break;
1021 		*tmp = '\0';
1022 		ret = strict_strtol(arg, 0, &offset);
1023 		if (ret)
1024 			break;
1025 		arg = tmp + 1;
1026 		tmp = strrchr(arg, ')');
1027 		if (tmp) {
1028 			struct deref_fetch_param *dprm;
1029 			const struct fetch_type *t2 = find_fetch_type(NULL);
1030 			*tmp = '\0';
1031 			dprm = kzalloc(sizeof(struct deref_fetch_param),
1032 				       GFP_KERNEL);
1033 			if (!dprm)
1034 				return -ENOMEM;
1035 			dprm->offset = offset;
1036 			ret = __parse_probe_arg(arg, t2, &dprm->orig,
1037 						is_return);
1038 			if (ret)
1039 				kfree(dprm);
1040 			else {
1041 				f->fn = t->fetch[FETCH_MTD_deref];
1042 				f->data = (void *)dprm;
1043 			}
1044 		}
1045 		break;
1046 	}
1047 	if (!ret && !f->fn) {	/* Parsed, but do not find fetch method */
1048 		pr_info("%s type has no corresponding fetch method.\n",
1049 			t->name);
1050 		ret = -EINVAL;
1051 	}
1052 	return ret;
1053 }
1054 
1055 #define BYTES_TO_BITS(nb)	((BITS_PER_LONG * (nb)) / sizeof(long))
1056 
1057 /* Bitfield type needs to be parsed into a fetch function */
1058 static int __parse_bitfield_probe_arg(const char *bf,
1059 				      const struct fetch_type *t,
1060 				      struct fetch_param *f)
1061 {
1062 	struct bitfield_fetch_param *bprm;
1063 	unsigned long bw, bo;
1064 	char *tail;
1065 
1066 	if (*bf != 'b')
1067 		return 0;
1068 
1069 	bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1070 	if (!bprm)
1071 		return -ENOMEM;
1072 	bprm->orig = *f;
1073 	f->fn = t->fetch[FETCH_MTD_bitfield];
1074 	f->data = (void *)bprm;
1075 
1076 	bw = simple_strtoul(bf + 1, &tail, 0);	/* Use simple one */
1077 	if (bw == 0 || *tail != '@')
1078 		return -EINVAL;
1079 
1080 	bf = tail + 1;
1081 	bo = simple_strtoul(bf, &tail, 0);
1082 	if (tail == bf || *tail != '/')
1083 		return -EINVAL;
1084 
1085 	bprm->hi_shift = BYTES_TO_BITS(t->size) - (bw + bo);
1086 	bprm->low_shift = bprm->hi_shift + bo;
1087 	return (BYTES_TO_BITS(t->size) < (bw + bo)) ? -EINVAL : 0;
1088 }
1089 
1090 /* String length checking wrapper */
1091 static int parse_probe_arg(char *arg, struct trace_probe *tp,
1092 			   struct probe_arg *parg, int is_return)
1093 {
1094 	const char *t;
1095 	int ret;
1096 
1097 	if (strlen(arg) > MAX_ARGSTR_LEN) {
1098 		pr_info("Argument is too long.: %s\n",  arg);
1099 		return -ENOSPC;
1100 	}
1101 	parg->comm = kstrdup(arg, GFP_KERNEL);
1102 	if (!parg->comm) {
1103 		pr_info("Failed to allocate memory for command '%s'.\n", arg);
1104 		return -ENOMEM;
1105 	}
1106 	t = strchr(parg->comm, ':');
1107 	if (t) {
1108 		arg[t - parg->comm] = '\0';
1109 		t++;
1110 	}
1111 	parg->type = find_fetch_type(t);
1112 	if (!parg->type) {
1113 		pr_info("Unsupported type: %s\n", t);
1114 		return -EINVAL;
1115 	}
1116 	parg->offset = tp->size;
1117 	tp->size += parg->type->size;
1118 	ret = __parse_probe_arg(arg, parg->type, &parg->fetch, is_return);
1119 	if (ret >= 0 && t != NULL)
1120 		ret = __parse_bitfield_probe_arg(t, parg->type, &parg->fetch);
1121 	if (ret >= 0) {
1122 		parg->fetch_size.fn = get_fetch_size_function(parg->type,
1123 							      parg->fetch.fn);
1124 		parg->fetch_size.data = parg->fetch.data;
1125 	}
1126 	return ret;
1127 }
1128 
1129 /* Return 1 if name is reserved or already used by another argument */
1130 static int conflict_field_name(const char *name,
1131 			       struct probe_arg *args, int narg)
1132 {
1133 	int i;
1134 	for (i = 0; i < ARRAY_SIZE(reserved_field_names); i++)
1135 		if (strcmp(reserved_field_names[i], name) == 0)
1136 			return 1;
1137 	for (i = 0; i < narg; i++)
1138 		if (strcmp(args[i].name, name) == 0)
1139 			return 1;
1140 	return 0;
1141 }
1142 
1143 static int create_trace_probe(int argc, char **argv)
1144 {
1145 	/*
1146 	 * Argument syntax:
1147 	 *  - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
1148 	 *  - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
1149 	 * Fetch args:
1150 	 *  $retval	: fetch return value
1151 	 *  $stack	: fetch stack address
1152 	 *  $stackN	: fetch Nth of stack (N:0-)
1153 	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
1154 	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
1155 	 *  %REG	: fetch register REG
1156 	 * Dereferencing memory fetch:
1157 	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
1158 	 * Alias name of args:
1159 	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
1160 	 * Type of args:
1161 	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
1162 	 */
1163 	struct trace_probe *tp;
1164 	int i, ret = 0;
1165 	int is_return = 0, is_delete = 0;
1166 	char *symbol = NULL, *event = NULL, *group = NULL;
1167 	char *arg;
1168 	unsigned long offset = 0;
1169 	void *addr = NULL;
1170 	char buf[MAX_EVENT_NAME_LEN];
1171 
1172 	/* argc must be >= 1 */
1173 	if (argv[0][0] == 'p')
1174 		is_return = 0;
1175 	else if (argv[0][0] == 'r')
1176 		is_return = 1;
1177 	else if (argv[0][0] == '-')
1178 		is_delete = 1;
1179 	else {
1180 		pr_info("Probe definition must be started with 'p', 'r' or"
1181 			" '-'.\n");
1182 		return -EINVAL;
1183 	}
1184 
1185 	if (argv[0][1] == ':') {
1186 		event = &argv[0][2];
1187 		if (strchr(event, '/')) {
1188 			group = event;
1189 			event = strchr(group, '/') + 1;
1190 			event[-1] = '\0';
1191 			if (strlen(group) == 0) {
1192 				pr_info("Group name is not specified\n");
1193 				return -EINVAL;
1194 			}
1195 		}
1196 		if (strlen(event) == 0) {
1197 			pr_info("Event name is not specified\n");
1198 			return -EINVAL;
1199 		}
1200 	}
1201 	if (!group)
1202 		group = KPROBE_EVENT_SYSTEM;
1203 
1204 	if (is_delete) {
1205 		if (!event) {
1206 			pr_info("Delete command needs an event name.\n");
1207 			return -EINVAL;
1208 		}
1209 		mutex_lock(&probe_lock);
1210 		tp = find_trace_probe(event, group);
1211 		if (!tp) {
1212 			mutex_unlock(&probe_lock);
1213 			pr_info("Event %s/%s doesn't exist.\n", group, event);
1214 			return -ENOENT;
1215 		}
1216 		/* delete an event */
1217 		ret = unregister_trace_probe(tp);
1218 		if (ret == 0)
1219 			free_trace_probe(tp);
1220 		mutex_unlock(&probe_lock);
1221 		return ret;
1222 	}
1223 
1224 	if (argc < 2) {
1225 		pr_info("Probe point is not specified.\n");
1226 		return -EINVAL;
1227 	}
1228 	if (isdigit(argv[1][0])) {
1229 		if (is_return) {
1230 			pr_info("Return probe point must be a symbol.\n");
1231 			return -EINVAL;
1232 		}
1233 		/* an address specified */
1234 		ret = strict_strtoul(&argv[1][0], 0, (unsigned long *)&addr);
1235 		if (ret) {
1236 			pr_info("Failed to parse address.\n");
1237 			return ret;
1238 		}
1239 	} else {
1240 		/* a symbol specified */
1241 		symbol = argv[1];
1242 		/* TODO: support .init module functions */
1243 		ret = split_symbol_offset(symbol, &offset);
1244 		if (ret) {
1245 			pr_info("Failed to parse symbol.\n");
1246 			return ret;
1247 		}
1248 		if (offset && is_return) {
1249 			pr_info("Return probe must be used without offset.\n");
1250 			return -EINVAL;
1251 		}
1252 	}
1253 	argc -= 2; argv += 2;
1254 
1255 	/* setup a probe */
1256 	if (!event) {
1257 		/* Make a new event name */
1258 		if (symbol)
1259 			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
1260 				 is_return ? 'r' : 'p', symbol, offset);
1261 		else
1262 			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
1263 				 is_return ? 'r' : 'p', addr);
1264 		event = buf;
1265 	}
1266 	tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
1267 			       is_return);
1268 	if (IS_ERR(tp)) {
1269 		pr_info("Failed to allocate trace_probe.(%d)\n",
1270 			(int)PTR_ERR(tp));
1271 		return PTR_ERR(tp);
1272 	}
1273 
1274 	/* parse arguments */
1275 	ret = 0;
1276 	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
1277 		/* Increment count for freeing args in error case */
1278 		tp->nr_args++;
1279 
1280 		/* Parse argument name */
1281 		arg = strchr(argv[i], '=');
1282 		if (arg) {
1283 			*arg++ = '\0';
1284 			tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
1285 		} else {
1286 			arg = argv[i];
1287 			/* If argument name is omitted, set "argN" */
1288 			snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
1289 			tp->args[i].name = kstrdup(buf, GFP_KERNEL);
1290 		}
1291 
1292 		if (!tp->args[i].name) {
1293 			pr_info("Failed to allocate argument[%d] name.\n", i);
1294 			ret = -ENOMEM;
1295 			goto error;
1296 		}
1297 
1298 		if (!is_good_name(tp->args[i].name)) {
1299 			pr_info("Invalid argument[%d] name: %s\n",
1300 				i, tp->args[i].name);
1301 			ret = -EINVAL;
1302 			goto error;
1303 		}
1304 
1305 		if (conflict_field_name(tp->args[i].name, tp->args, i)) {
1306 			pr_info("Argument[%d] name '%s' conflicts with "
1307 				"another field.\n", i, argv[i]);
1308 			ret = -EINVAL;
1309 			goto error;
1310 		}
1311 
1312 		/* Parse fetch argument */
1313 		ret = parse_probe_arg(arg, tp, &tp->args[i], is_return);
1314 		if (ret) {
1315 			pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
1316 			goto error;
1317 		}
1318 	}
1319 
1320 	ret = register_trace_probe(tp);
1321 	if (ret)
1322 		goto error;
1323 	return 0;
1324 
1325 error:
1326 	free_trace_probe(tp);
1327 	return ret;
1328 }
1329 
1330 static int release_all_trace_probes(void)
1331 {
1332 	struct trace_probe *tp;
1333 	int ret = 0;
1334 
1335 	mutex_lock(&probe_lock);
1336 	/* Ensure no probe is in use. */
1337 	list_for_each_entry(tp, &probe_list, list)
1338 		if (trace_probe_is_enabled(tp)) {
1339 			ret = -EBUSY;
1340 			goto end;
1341 		}
1342 	/* TODO: Use batch unregistration */
1343 	while (!list_empty(&probe_list)) {
1344 		tp = list_entry(probe_list.next, struct trace_probe, list);
1345 		unregister_trace_probe(tp);
1346 		free_trace_probe(tp);
1347 	}
1348 
1349 end:
1350 	mutex_unlock(&probe_lock);
1351 
1352 	return ret;
1353 }
1354 
1355 /* Probes listing interfaces */
1356 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
1357 {
1358 	mutex_lock(&probe_lock);
1359 	return seq_list_start(&probe_list, *pos);
1360 }
1361 
1362 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
1363 {
1364 	return seq_list_next(v, &probe_list, pos);
1365 }
1366 
1367 static void probes_seq_stop(struct seq_file *m, void *v)
1368 {
1369 	mutex_unlock(&probe_lock);
1370 }
1371 
1372 static int probes_seq_show(struct seq_file *m, void *v)
1373 {
1374 	struct trace_probe *tp = v;
1375 	int i;
1376 
1377 	seq_printf(m, "%c", trace_probe_is_return(tp) ? 'r' : 'p');
1378 	seq_printf(m, ":%s/%s", tp->call.class->system, tp->call.name);
1379 
1380 	if (!tp->symbol)
1381 		seq_printf(m, " 0x%p", tp->rp.kp.addr);
1382 	else if (tp->rp.kp.offset)
1383 		seq_printf(m, " %s+%u", trace_probe_symbol(tp),
1384 			   tp->rp.kp.offset);
1385 	else
1386 		seq_printf(m, " %s", trace_probe_symbol(tp));
1387 
1388 	for (i = 0; i < tp->nr_args; i++)
1389 		seq_printf(m, " %s=%s", tp->args[i].name, tp->args[i].comm);
1390 	seq_printf(m, "\n");
1391 
1392 	return 0;
1393 }
1394 
1395 static const struct seq_operations probes_seq_op = {
1396 	.start  = probes_seq_start,
1397 	.next   = probes_seq_next,
1398 	.stop   = probes_seq_stop,
1399 	.show   = probes_seq_show
1400 };
1401 
1402 static int probes_open(struct inode *inode, struct file *file)
1403 {
1404 	int ret;
1405 
1406 	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1407 		ret = release_all_trace_probes();
1408 		if (ret < 0)
1409 			return ret;
1410 	}
1411 
1412 	return seq_open(file, &probes_seq_op);
1413 }
1414 
1415 static int command_trace_probe(const char *buf)
1416 {
1417 	char **argv;
1418 	int argc = 0, ret = 0;
1419 
1420 	argv = argv_split(GFP_KERNEL, buf, &argc);
1421 	if (!argv)
1422 		return -ENOMEM;
1423 
1424 	if (argc)
1425 		ret = create_trace_probe(argc, argv);
1426 
1427 	argv_free(argv);
1428 	return ret;
1429 }
1430 
1431 #define WRITE_BUFSIZE 4096
1432 
1433 static ssize_t probes_write(struct file *file, const char __user *buffer,
1434 			    size_t count, loff_t *ppos)
1435 {
1436 	char *kbuf, *tmp;
1437 	int ret;
1438 	size_t done;
1439 	size_t size;
1440 
1441 	kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
1442 	if (!kbuf)
1443 		return -ENOMEM;
1444 
1445 	ret = done = 0;
1446 	while (done < count) {
1447 		size = count - done;
1448 		if (size >= WRITE_BUFSIZE)
1449 			size = WRITE_BUFSIZE - 1;
1450 		if (copy_from_user(kbuf, buffer + done, size)) {
1451 			ret = -EFAULT;
1452 			goto out;
1453 		}
1454 		kbuf[size] = '\0';
1455 		tmp = strchr(kbuf, '\n');
1456 		if (tmp) {
1457 			*tmp = '\0';
1458 			size = tmp - kbuf + 1;
1459 		} else if (done + size < count) {
1460 			pr_warning("Line length is too long: "
1461 				   "Should be less than %d.", WRITE_BUFSIZE);
1462 			ret = -EINVAL;
1463 			goto out;
1464 		}
1465 		done += size;
1466 		/* Remove comments */
1467 		tmp = strchr(kbuf, '#');
1468 		if (tmp)
1469 			*tmp = '\0';
1470 
1471 		ret = command_trace_probe(kbuf);
1472 		if (ret)
1473 			goto out;
1474 	}
1475 	ret = done;
1476 out:
1477 	kfree(kbuf);
1478 	return ret;
1479 }
1480 
1481 static const struct file_operations kprobe_events_ops = {
1482 	.owner          = THIS_MODULE,
1483 	.open           = probes_open,
1484 	.read           = seq_read,
1485 	.llseek         = seq_lseek,
1486 	.release        = seq_release,
1487 	.write		= probes_write,
1488 };
1489 
1490 /* Probes profiling interfaces */
1491 static int probes_profile_seq_show(struct seq_file *m, void *v)
1492 {
1493 	struct trace_probe *tp = v;
1494 
1495 	seq_printf(m, "  %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
1496 		   tp->rp.kp.nmissed);
1497 
1498 	return 0;
1499 }
1500 
1501 static const struct seq_operations profile_seq_op = {
1502 	.start  = probes_seq_start,
1503 	.next   = probes_seq_next,
1504 	.stop   = probes_seq_stop,
1505 	.show   = probes_profile_seq_show
1506 };
1507 
1508 static int profile_open(struct inode *inode, struct file *file)
1509 {
1510 	return seq_open(file, &profile_seq_op);
1511 }
1512 
1513 static const struct file_operations kprobe_profile_ops = {
1514 	.owner          = THIS_MODULE,
1515 	.open           = profile_open,
1516 	.read           = seq_read,
1517 	.llseek         = seq_lseek,
1518 	.release        = seq_release,
1519 };
1520 
1521 /* Sum up total data length for dynamic arraies (strings) */
1522 static __kprobes int __get_data_size(struct trace_probe *tp,
1523 				     struct pt_regs *regs)
1524 {
1525 	int i, ret = 0;
1526 	u32 len;
1527 
1528 	for (i = 0; i < tp->nr_args; i++)
1529 		if (unlikely(tp->args[i].fetch_size.fn)) {
1530 			call_fetch(&tp->args[i].fetch_size, regs, &len);
1531 			ret += len;
1532 		}
1533 
1534 	return ret;
1535 }
1536 
1537 /* Store the value of each argument */
1538 static __kprobes void store_trace_args(int ent_size, struct trace_probe *tp,
1539 				       struct pt_regs *regs,
1540 				       u8 *data, int maxlen)
1541 {
1542 	int i;
1543 	u32 end = tp->size;
1544 	u32 *dl;	/* Data (relative) location */
1545 
1546 	for (i = 0; i < tp->nr_args; i++) {
1547 		if (unlikely(tp->args[i].fetch_size.fn)) {
1548 			/*
1549 			 * First, we set the relative location and
1550 			 * maximum data length to *dl
1551 			 */
1552 			dl = (u32 *)(data + tp->args[i].offset);
1553 			*dl = make_data_rloc(maxlen, end - tp->args[i].offset);
1554 			/* Then try to fetch string or dynamic array data */
1555 			call_fetch(&tp->args[i].fetch, regs, dl);
1556 			/* Reduce maximum length */
1557 			end += get_rloc_len(*dl);
1558 			maxlen -= get_rloc_len(*dl);
1559 			/* Trick here, convert data_rloc to data_loc */
1560 			*dl = convert_rloc_to_loc(*dl,
1561 				 ent_size + tp->args[i].offset);
1562 		} else
1563 			/* Just fetching data normally */
1564 			call_fetch(&tp->args[i].fetch, regs,
1565 				   data + tp->args[i].offset);
1566 	}
1567 }
1568 
1569 /* Kprobe handler */
1570 static __kprobes void kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
1571 {
1572 	struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1573 	struct kprobe_trace_entry_head *entry;
1574 	struct ring_buffer_event *event;
1575 	struct ring_buffer *buffer;
1576 	int size, dsize, pc;
1577 	unsigned long irq_flags;
1578 	struct ftrace_event_call *call = &tp->call;
1579 
1580 	tp->nhit++;
1581 
1582 	local_save_flags(irq_flags);
1583 	pc = preempt_count();
1584 
1585 	dsize = __get_data_size(tp, regs);
1586 	size = sizeof(*entry) + tp->size + dsize;
1587 
1588 	event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
1589 						  size, irq_flags, pc);
1590 	if (!event)
1591 		return;
1592 
1593 	entry = ring_buffer_event_data(event);
1594 	entry->ip = (unsigned long)kp->addr;
1595 	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1596 
1597 	if (!filter_current_check_discard(buffer, call, entry, event))
1598 		trace_nowake_buffer_unlock_commit_regs(buffer, event,
1599 						       irq_flags, pc, regs);
1600 }
1601 
1602 /* Kretprobe handler */
1603 static __kprobes void kretprobe_trace_func(struct kretprobe_instance *ri,
1604 					  struct pt_regs *regs)
1605 {
1606 	struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1607 	struct kretprobe_trace_entry_head *entry;
1608 	struct ring_buffer_event *event;
1609 	struct ring_buffer *buffer;
1610 	int size, pc, dsize;
1611 	unsigned long irq_flags;
1612 	struct ftrace_event_call *call = &tp->call;
1613 
1614 	local_save_flags(irq_flags);
1615 	pc = preempt_count();
1616 
1617 	dsize = __get_data_size(tp, regs);
1618 	size = sizeof(*entry) + tp->size + dsize;
1619 
1620 	event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
1621 						  size, irq_flags, pc);
1622 	if (!event)
1623 		return;
1624 
1625 	entry = ring_buffer_event_data(event);
1626 	entry->func = (unsigned long)tp->rp.kp.addr;
1627 	entry->ret_ip = (unsigned long)ri->ret_addr;
1628 	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1629 
1630 	if (!filter_current_check_discard(buffer, call, entry, event))
1631 		trace_nowake_buffer_unlock_commit_regs(buffer, event,
1632 						       irq_flags, pc, regs);
1633 }
1634 
1635 /* Event entry printers */
1636 enum print_line_t
1637 print_kprobe_event(struct trace_iterator *iter, int flags,
1638 		   struct trace_event *event)
1639 {
1640 	struct kprobe_trace_entry_head *field;
1641 	struct trace_seq *s = &iter->seq;
1642 	struct trace_probe *tp;
1643 	u8 *data;
1644 	int i;
1645 
1646 	field = (struct kprobe_trace_entry_head *)iter->ent;
1647 	tp = container_of(event, struct trace_probe, call.event);
1648 
1649 	if (!trace_seq_printf(s, "%s: (", tp->call.name))
1650 		goto partial;
1651 
1652 	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1653 		goto partial;
1654 
1655 	if (!trace_seq_puts(s, ")"))
1656 		goto partial;
1657 
1658 	data = (u8 *)&field[1];
1659 	for (i = 0; i < tp->nr_args; i++)
1660 		if (!tp->args[i].type->print(s, tp->args[i].name,
1661 					     data + tp->args[i].offset, field))
1662 			goto partial;
1663 
1664 	if (!trace_seq_puts(s, "\n"))
1665 		goto partial;
1666 
1667 	return TRACE_TYPE_HANDLED;
1668 partial:
1669 	return TRACE_TYPE_PARTIAL_LINE;
1670 }
1671 
1672 enum print_line_t
1673 print_kretprobe_event(struct trace_iterator *iter, int flags,
1674 		      struct trace_event *event)
1675 {
1676 	struct kretprobe_trace_entry_head *field;
1677 	struct trace_seq *s = &iter->seq;
1678 	struct trace_probe *tp;
1679 	u8 *data;
1680 	int i;
1681 
1682 	field = (struct kretprobe_trace_entry_head *)iter->ent;
1683 	tp = container_of(event, struct trace_probe, call.event);
1684 
1685 	if (!trace_seq_printf(s, "%s: (", tp->call.name))
1686 		goto partial;
1687 
1688 	if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1689 		goto partial;
1690 
1691 	if (!trace_seq_puts(s, " <- "))
1692 		goto partial;
1693 
1694 	if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1695 		goto partial;
1696 
1697 	if (!trace_seq_puts(s, ")"))
1698 		goto partial;
1699 
1700 	data = (u8 *)&field[1];
1701 	for (i = 0; i < tp->nr_args; i++)
1702 		if (!tp->args[i].type->print(s, tp->args[i].name,
1703 					     data + tp->args[i].offset, field))
1704 			goto partial;
1705 
1706 	if (!trace_seq_puts(s, "\n"))
1707 		goto partial;
1708 
1709 	return TRACE_TYPE_HANDLED;
1710 partial:
1711 	return TRACE_TYPE_PARTIAL_LINE;
1712 }
1713 
1714 #undef DEFINE_FIELD
1715 #define DEFINE_FIELD(type, item, name, is_signed)			\
1716 	do {								\
1717 		ret = trace_define_field(event_call, #type, name,	\
1718 					 offsetof(typeof(field), item),	\
1719 					 sizeof(field.item), is_signed, \
1720 					 FILTER_OTHER);			\
1721 		if (ret)						\
1722 			return ret;					\
1723 	} while (0)
1724 
1725 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
1726 {
1727 	int ret, i;
1728 	struct kprobe_trace_entry_head field;
1729 	struct trace_probe *tp = (struct trace_probe *)event_call->data;
1730 
1731 	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1732 	/* Set argument names as fields */
1733 	for (i = 0; i < tp->nr_args; i++) {
1734 		ret = trace_define_field(event_call, tp->args[i].type->fmttype,
1735 					 tp->args[i].name,
1736 					 sizeof(field) + tp->args[i].offset,
1737 					 tp->args[i].type->size,
1738 					 tp->args[i].type->is_signed,
1739 					 FILTER_OTHER);
1740 		if (ret)
1741 			return ret;
1742 	}
1743 	return 0;
1744 }
1745 
1746 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1747 {
1748 	int ret, i;
1749 	struct kretprobe_trace_entry_head field;
1750 	struct trace_probe *tp = (struct trace_probe *)event_call->data;
1751 
1752 	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1753 	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1754 	/* Set argument names as fields */
1755 	for (i = 0; i < tp->nr_args; i++) {
1756 		ret = trace_define_field(event_call, tp->args[i].type->fmttype,
1757 					 tp->args[i].name,
1758 					 sizeof(field) + tp->args[i].offset,
1759 					 tp->args[i].type->size,
1760 					 tp->args[i].type->is_signed,
1761 					 FILTER_OTHER);
1762 		if (ret)
1763 			return ret;
1764 	}
1765 	return 0;
1766 }
1767 
1768 static int __set_print_fmt(struct trace_probe *tp, char *buf, int len)
1769 {
1770 	int i;
1771 	int pos = 0;
1772 
1773 	const char *fmt, *arg;
1774 
1775 	if (!trace_probe_is_return(tp)) {
1776 		fmt = "(%lx)";
1777 		arg = "REC->" FIELD_STRING_IP;
1778 	} else {
1779 		fmt = "(%lx <- %lx)";
1780 		arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
1781 	}
1782 
1783 	/* When len=0, we just calculate the needed length */
1784 #define LEN_OR_ZERO (len ? len - pos : 0)
1785 
1786 	pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
1787 
1788 	for (i = 0; i < tp->nr_args; i++) {
1789 		pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
1790 				tp->args[i].name, tp->args[i].type->fmt);
1791 	}
1792 
1793 	pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
1794 
1795 	for (i = 0; i < tp->nr_args; i++) {
1796 		if (strcmp(tp->args[i].type->name, "string") == 0)
1797 			pos += snprintf(buf + pos, LEN_OR_ZERO,
1798 					", __get_str(%s)",
1799 					tp->args[i].name);
1800 		else
1801 			pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
1802 					tp->args[i].name);
1803 	}
1804 
1805 #undef LEN_OR_ZERO
1806 
1807 	/* return the length of print_fmt */
1808 	return pos;
1809 }
1810 
1811 static int set_print_fmt(struct trace_probe *tp)
1812 {
1813 	int len;
1814 	char *print_fmt;
1815 
1816 	/* First: called with 0 length to calculate the needed length */
1817 	len = __set_print_fmt(tp, NULL, 0);
1818 	print_fmt = kmalloc(len + 1, GFP_KERNEL);
1819 	if (!print_fmt)
1820 		return -ENOMEM;
1821 
1822 	/* Second: actually write the @print_fmt */
1823 	__set_print_fmt(tp, print_fmt, len + 1);
1824 	tp->call.print_fmt = print_fmt;
1825 
1826 	return 0;
1827 }
1828 
1829 #ifdef CONFIG_PERF_EVENTS
1830 
1831 /* Kprobe profile handler */
1832 static __kprobes void kprobe_perf_func(struct kprobe *kp,
1833 					 struct pt_regs *regs)
1834 {
1835 	struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1836 	struct ftrace_event_call *call = &tp->call;
1837 	struct kprobe_trace_entry_head *entry;
1838 	struct hlist_head *head;
1839 	int size, __size, dsize;
1840 	int rctx;
1841 
1842 	dsize = __get_data_size(tp, regs);
1843 	__size = sizeof(*entry) + tp->size + dsize;
1844 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1845 	size -= sizeof(u32);
1846 	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1847 		     "profile buffer not large enough"))
1848 		return;
1849 
1850 	entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1851 	if (!entry)
1852 		return;
1853 
1854 	entry->ip = (unsigned long)kp->addr;
1855 	memset(&entry[1], 0, dsize);
1856 	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1857 
1858 	head = this_cpu_ptr(call->perf_events);
1859 	perf_trace_buf_submit(entry, size, rctx, entry->ip, 1, regs, head);
1860 }
1861 
1862 /* Kretprobe profile handler */
1863 static __kprobes void kretprobe_perf_func(struct kretprobe_instance *ri,
1864 					    struct pt_regs *regs)
1865 {
1866 	struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1867 	struct ftrace_event_call *call = &tp->call;
1868 	struct kretprobe_trace_entry_head *entry;
1869 	struct hlist_head *head;
1870 	int size, __size, dsize;
1871 	int rctx;
1872 
1873 	dsize = __get_data_size(tp, regs);
1874 	__size = sizeof(*entry) + tp->size + dsize;
1875 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1876 	size -= sizeof(u32);
1877 	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1878 		     "profile buffer not large enough"))
1879 		return;
1880 
1881 	entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1882 	if (!entry)
1883 		return;
1884 
1885 	entry->func = (unsigned long)tp->rp.kp.addr;
1886 	entry->ret_ip = (unsigned long)ri->ret_addr;
1887 	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1888 
1889 	head = this_cpu_ptr(call->perf_events);
1890 	perf_trace_buf_submit(entry, size, rctx, entry->ret_ip, 1, regs, head);
1891 }
1892 #endif	/* CONFIG_PERF_EVENTS */
1893 
1894 static __kprobes
1895 int kprobe_register(struct ftrace_event_call *event, enum trace_reg type)
1896 {
1897 	struct trace_probe *tp = (struct trace_probe *)event->data;
1898 
1899 	switch (type) {
1900 	case TRACE_REG_REGISTER:
1901 		return enable_trace_probe(tp, TP_FLAG_TRACE);
1902 	case TRACE_REG_UNREGISTER:
1903 		disable_trace_probe(tp, TP_FLAG_TRACE);
1904 		return 0;
1905 
1906 #ifdef CONFIG_PERF_EVENTS
1907 	case TRACE_REG_PERF_REGISTER:
1908 		return enable_trace_probe(tp, TP_FLAG_PROFILE);
1909 	case TRACE_REG_PERF_UNREGISTER:
1910 		disable_trace_probe(tp, TP_FLAG_PROFILE);
1911 		return 0;
1912 #endif
1913 	}
1914 	return 0;
1915 }
1916 
1917 static __kprobes
1918 int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1919 {
1920 	struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1921 
1922 	if (tp->flags & TP_FLAG_TRACE)
1923 		kprobe_trace_func(kp, regs);
1924 #ifdef CONFIG_PERF_EVENTS
1925 	if (tp->flags & TP_FLAG_PROFILE)
1926 		kprobe_perf_func(kp, regs);
1927 #endif
1928 	return 0;	/* We don't tweek kernel, so just return 0 */
1929 }
1930 
1931 static __kprobes
1932 int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1933 {
1934 	struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1935 
1936 	if (tp->flags & TP_FLAG_TRACE)
1937 		kretprobe_trace_func(ri, regs);
1938 #ifdef CONFIG_PERF_EVENTS
1939 	if (tp->flags & TP_FLAG_PROFILE)
1940 		kretprobe_perf_func(ri, regs);
1941 #endif
1942 	return 0;	/* We don't tweek kernel, so just return 0 */
1943 }
1944 
1945 static struct trace_event_functions kretprobe_funcs = {
1946 	.trace		= print_kretprobe_event
1947 };
1948 
1949 static struct trace_event_functions kprobe_funcs = {
1950 	.trace		= print_kprobe_event
1951 };
1952 
1953 static int register_probe_event(struct trace_probe *tp)
1954 {
1955 	struct ftrace_event_call *call = &tp->call;
1956 	int ret;
1957 
1958 	/* Initialize ftrace_event_call */
1959 	INIT_LIST_HEAD(&call->class->fields);
1960 	if (trace_probe_is_return(tp)) {
1961 		call->event.funcs = &kretprobe_funcs;
1962 		call->class->define_fields = kretprobe_event_define_fields;
1963 	} else {
1964 		call->event.funcs = &kprobe_funcs;
1965 		call->class->define_fields = kprobe_event_define_fields;
1966 	}
1967 	if (set_print_fmt(tp) < 0)
1968 		return -ENOMEM;
1969 	ret = register_ftrace_event(&call->event);
1970 	if (!ret) {
1971 		kfree(call->print_fmt);
1972 		return -ENODEV;
1973 	}
1974 	call->flags = 0;
1975 	call->class->reg = kprobe_register;
1976 	call->data = tp;
1977 	ret = trace_add_event_call(call);
1978 	if (ret) {
1979 		pr_info("Failed to register kprobe event: %s\n", call->name);
1980 		kfree(call->print_fmt);
1981 		unregister_ftrace_event(&call->event);
1982 	}
1983 	return ret;
1984 }
1985 
1986 static void unregister_probe_event(struct trace_probe *tp)
1987 {
1988 	/* tp->event is unregistered in trace_remove_event_call() */
1989 	trace_remove_event_call(&tp->call);
1990 	kfree(tp->call.print_fmt);
1991 }
1992 
1993 /* Make a debugfs interface for controlling probe points */
1994 static __init int init_kprobe_trace(void)
1995 {
1996 	struct dentry *d_tracer;
1997 	struct dentry *entry;
1998 
1999 	if (register_module_notifier(&trace_probe_module_nb))
2000 		return -EINVAL;
2001 
2002 	d_tracer = tracing_init_dentry();
2003 	if (!d_tracer)
2004 		return 0;
2005 
2006 	entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
2007 				    NULL, &kprobe_events_ops);
2008 
2009 	/* Event list interface */
2010 	if (!entry)
2011 		pr_warning("Could not create debugfs "
2012 			   "'kprobe_events' entry\n");
2013 
2014 	/* Profile interface */
2015 	entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
2016 				    NULL, &kprobe_profile_ops);
2017 
2018 	if (!entry)
2019 		pr_warning("Could not create debugfs "
2020 			   "'kprobe_profile' entry\n");
2021 	return 0;
2022 }
2023 fs_initcall(init_kprobe_trace);
2024 
2025 
2026 #ifdef CONFIG_FTRACE_STARTUP_TEST
2027 
2028 /*
2029  * The "__used" keeps gcc from removing the function symbol
2030  * from the kallsyms table.
2031  */
2032 static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
2033 					       int a4, int a5, int a6)
2034 {
2035 	return a1 + a2 + a3 + a4 + a5 + a6;
2036 }
2037 
2038 static __init int kprobe_trace_self_tests_init(void)
2039 {
2040 	int ret, warn = 0;
2041 	int (*target)(int, int, int, int, int, int);
2042 	struct trace_probe *tp;
2043 
2044 	target = kprobe_trace_selftest_target;
2045 
2046 	pr_info("Testing kprobe tracing: ");
2047 
2048 	ret = command_trace_probe("p:testprobe kprobe_trace_selftest_target "
2049 				  "$stack $stack0 +0($stack)");
2050 	if (WARN_ON_ONCE(ret)) {
2051 		pr_warning("error on probing function entry.\n");
2052 		warn++;
2053 	} else {
2054 		/* Enable trace point */
2055 		tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
2056 		if (WARN_ON_ONCE(tp == NULL)) {
2057 			pr_warning("error on getting new probe.\n");
2058 			warn++;
2059 		} else
2060 			enable_trace_probe(tp, TP_FLAG_TRACE);
2061 	}
2062 
2063 	ret = command_trace_probe("r:testprobe2 kprobe_trace_selftest_target "
2064 				  "$retval");
2065 	if (WARN_ON_ONCE(ret)) {
2066 		pr_warning("error on probing function return.\n");
2067 		warn++;
2068 	} else {
2069 		/* Enable trace point */
2070 		tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
2071 		if (WARN_ON_ONCE(tp == NULL)) {
2072 			pr_warning("error on getting new probe.\n");
2073 			warn++;
2074 		} else
2075 			enable_trace_probe(tp, TP_FLAG_TRACE);
2076 	}
2077 
2078 	if (warn)
2079 		goto end;
2080 
2081 	ret = target(1, 2, 3, 4, 5, 6);
2082 
2083 	/* Disable trace points before removing it */
2084 	tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
2085 	if (WARN_ON_ONCE(tp == NULL)) {
2086 		pr_warning("error on getting test probe.\n");
2087 		warn++;
2088 	} else
2089 		disable_trace_probe(tp, TP_FLAG_TRACE);
2090 
2091 	tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
2092 	if (WARN_ON_ONCE(tp == NULL)) {
2093 		pr_warning("error on getting 2nd test probe.\n");
2094 		warn++;
2095 	} else
2096 		disable_trace_probe(tp, TP_FLAG_TRACE);
2097 
2098 	ret = command_trace_probe("-:testprobe");
2099 	if (WARN_ON_ONCE(ret)) {
2100 		pr_warning("error on deleting a probe.\n");
2101 		warn++;
2102 	}
2103 
2104 	ret = command_trace_probe("-:testprobe2");
2105 	if (WARN_ON_ONCE(ret)) {
2106 		pr_warning("error on deleting a probe.\n");
2107 		warn++;
2108 	}
2109 
2110 end:
2111 	release_all_trace_probes();
2112 	if (warn)
2113 		pr_cont("NG: Some tests are failed. Please check them.\n");
2114 	else
2115 		pr_cont("OK\n");
2116 	return 0;
2117 }
2118 
2119 late_initcall(kprobe_trace_self_tests_init);
2120 
2121 #endif
2122