xref: /linux/kernel/trace/trace_kprobe.c (revision 8ec3b8432e4fe8d452f88f1ed9a3450e715bb797)
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 	"common_lock_depth",
57 	FIELD_STRING_IP,
58 	FIELD_STRING_RETIP,
59 	FIELD_STRING_FUNC,
60 };
61 
62 /* Printing function type */
63 typedef int (*print_type_func_t)(struct trace_seq *, const char *, void *,
64 				 void *);
65 #define PRINT_TYPE_FUNC_NAME(type)	print_type_##type
66 #define PRINT_TYPE_FMT_NAME(type)	print_type_format_##type
67 
68 /* Printing  in basic type function template */
69 #define DEFINE_BASIC_PRINT_TYPE_FUNC(type, fmt, cast)			\
70 static __kprobes int PRINT_TYPE_FUNC_NAME(type)(struct trace_seq *s,	\
71 						const char *name,	\
72 						void *data, void *ent)\
73 {									\
74 	return trace_seq_printf(s, " %s=" fmt, name, (cast)*(type *)data);\
75 }									\
76 static const char PRINT_TYPE_FMT_NAME(type)[] = fmt;
77 
78 DEFINE_BASIC_PRINT_TYPE_FUNC(u8, "%x", unsigned int)
79 DEFINE_BASIC_PRINT_TYPE_FUNC(u16, "%x", unsigned int)
80 DEFINE_BASIC_PRINT_TYPE_FUNC(u32, "%lx", unsigned long)
81 DEFINE_BASIC_PRINT_TYPE_FUNC(u64, "%llx", unsigned long long)
82 DEFINE_BASIC_PRINT_TYPE_FUNC(s8, "%d", int)
83 DEFINE_BASIC_PRINT_TYPE_FUNC(s16, "%d", int)
84 DEFINE_BASIC_PRINT_TYPE_FUNC(s32, "%ld", long)
85 DEFINE_BASIC_PRINT_TYPE_FUNC(s64, "%lld", long long)
86 
87 /* data_rloc: data relative location, compatible with u32 */
88 #define make_data_rloc(len, roffs)	\
89 	(((u32)(len) << 16) | ((u32)(roffs) & 0xffff))
90 #define get_rloc_len(dl)	((u32)(dl) >> 16)
91 #define get_rloc_offs(dl)	((u32)(dl) & 0xffff)
92 
93 static inline void *get_rloc_data(u32 *dl)
94 {
95 	return (u8 *)dl + get_rloc_offs(*dl);
96 }
97 
98 /* For data_loc conversion */
99 static inline void *get_loc_data(u32 *dl, void *ent)
100 {
101 	return (u8 *)ent + get_rloc_offs(*dl);
102 }
103 
104 /*
105  * Convert data_rloc to data_loc:
106  *  data_rloc stores the offset from data_rloc itself, but data_loc
107  *  stores the offset from event entry.
108  */
109 #define convert_rloc_to_loc(dl, offs)	((u32)(dl) + (offs))
110 
111 /* For defining macros, define string/string_size types */
112 typedef u32 string;
113 typedef u32 string_size;
114 
115 /* Print type function for string type */
116 static __kprobes int PRINT_TYPE_FUNC_NAME(string)(struct trace_seq *s,
117 						  const char *name,
118 						  void *data, void *ent)
119 {
120 	int len = *(u32 *)data >> 16;
121 
122 	if (!len)
123 		return trace_seq_printf(s, " %s=(fault)", name);
124 	else
125 		return trace_seq_printf(s, " %s=\"%s\"", name,
126 					(const char *)get_loc_data(data, ent));
127 }
128 static const char PRINT_TYPE_FMT_NAME(string)[] = "\\\"%s\\\"";
129 
130 /* Data fetch function type */
131 typedef	void (*fetch_func_t)(struct pt_regs *, void *, void *);
132 
133 struct fetch_param {
134 	fetch_func_t	fn;
135 	void *data;
136 };
137 
138 static __kprobes void call_fetch(struct fetch_param *fprm,
139 				 struct pt_regs *regs, void *dest)
140 {
141 	return fprm->fn(regs, fprm->data, dest);
142 }
143 
144 #define FETCH_FUNC_NAME(method, type)	fetch_##method##_##type
145 /*
146  * Define macro for basic types - we don't need to define s* types, because
147  * we have to care only about bitwidth at recording time.
148  */
149 #define DEFINE_BASIC_FETCH_FUNCS(method) \
150 DEFINE_FETCH_##method(u8)		\
151 DEFINE_FETCH_##method(u16)		\
152 DEFINE_FETCH_##method(u32)		\
153 DEFINE_FETCH_##method(u64)
154 
155 #define CHECK_FETCH_FUNCS(method, fn)			\
156 	(((FETCH_FUNC_NAME(method, u8) == fn) ||	\
157 	  (FETCH_FUNC_NAME(method, u16) == fn) ||	\
158 	  (FETCH_FUNC_NAME(method, u32) == fn) ||	\
159 	  (FETCH_FUNC_NAME(method, u64) == fn) ||	\
160 	  (FETCH_FUNC_NAME(method, string) == fn) ||	\
161 	  (FETCH_FUNC_NAME(method, string_size) == fn)) \
162 	 && (fn != NULL))
163 
164 /* Data fetch function templates */
165 #define DEFINE_FETCH_reg(type)						\
166 static __kprobes void FETCH_FUNC_NAME(reg, type)(struct pt_regs *regs,	\
167 					void *offset, void *dest)	\
168 {									\
169 	*(type *)dest = (type)regs_get_register(regs,			\
170 				(unsigned int)((unsigned long)offset));	\
171 }
172 DEFINE_BASIC_FETCH_FUNCS(reg)
173 /* No string on the register */
174 #define fetch_reg_string NULL
175 #define fetch_reg_string_size NULL
176 
177 #define DEFINE_FETCH_stack(type)					\
178 static __kprobes void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,\
179 					  void *offset, void *dest)	\
180 {									\
181 	*(type *)dest = (type)regs_get_kernel_stack_nth(regs,		\
182 				(unsigned int)((unsigned long)offset));	\
183 }
184 DEFINE_BASIC_FETCH_FUNCS(stack)
185 /* No string on the stack entry */
186 #define fetch_stack_string NULL
187 #define fetch_stack_string_size NULL
188 
189 #define DEFINE_FETCH_retval(type)					\
190 static __kprobes void FETCH_FUNC_NAME(retval, type)(struct pt_regs *regs,\
191 					  void *dummy, void *dest)	\
192 {									\
193 	*(type *)dest = (type)regs_return_value(regs);			\
194 }
195 DEFINE_BASIC_FETCH_FUNCS(retval)
196 /* No string on the retval */
197 #define fetch_retval_string NULL
198 #define fetch_retval_string_size NULL
199 
200 #define DEFINE_FETCH_memory(type)					\
201 static __kprobes void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,\
202 					  void *addr, void *dest)	\
203 {									\
204 	type retval;							\
205 	if (probe_kernel_address(addr, retval))				\
206 		*(type *)dest = 0;					\
207 	else								\
208 		*(type *)dest = retval;					\
209 }
210 DEFINE_BASIC_FETCH_FUNCS(memory)
211 /*
212  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
213  * length and relative data location.
214  */
215 static __kprobes void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
216 						      void *addr, void *dest)
217 {
218 	long ret;
219 	int maxlen = get_rloc_len(*(u32 *)dest);
220 	u8 *dst = get_rloc_data(dest);
221 	u8 *src = addr;
222 	mm_segment_t old_fs = get_fs();
223 	if (!maxlen)
224 		return;
225 	/*
226 	 * Try to get string again, since the string can be changed while
227 	 * probing.
228 	 */
229 	set_fs(KERNEL_DS);
230 	pagefault_disable();
231 	do
232 		ret = __copy_from_user_inatomic(dst++, src++, 1);
233 	while (dst[-1] && ret == 0 && src - (u8 *)addr < maxlen);
234 	dst[-1] = '\0';
235 	pagefault_enable();
236 	set_fs(old_fs);
237 
238 	if (ret < 0) {	/* Failed to fetch string */
239 		((u8 *)get_rloc_data(dest))[0] = '\0';
240 		*(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
241 	} else
242 		*(u32 *)dest = make_data_rloc(src - (u8 *)addr,
243 					      get_rloc_offs(*(u32 *)dest));
244 }
245 /* Return the length of string -- including null terminal byte */
246 static __kprobes void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
247 							void *addr, void *dest)
248 {
249 	int ret, len = 0;
250 	u8 c;
251 	mm_segment_t old_fs = get_fs();
252 
253 	set_fs(KERNEL_DS);
254 	pagefault_disable();
255 	do {
256 		ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
257 		len++;
258 	} while (c && ret == 0 && len < MAX_STRING_SIZE);
259 	pagefault_enable();
260 	set_fs(old_fs);
261 
262 	if (ret < 0)	/* Failed to check the length */
263 		*(u32 *)dest = 0;
264 	else
265 		*(u32 *)dest = len;
266 }
267 
268 /* Memory fetching by symbol */
269 struct symbol_cache {
270 	char *symbol;
271 	long offset;
272 	unsigned long addr;
273 };
274 
275 static unsigned long update_symbol_cache(struct symbol_cache *sc)
276 {
277 	sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
278 	if (sc->addr)
279 		sc->addr += sc->offset;
280 	return sc->addr;
281 }
282 
283 static void free_symbol_cache(struct symbol_cache *sc)
284 {
285 	kfree(sc->symbol);
286 	kfree(sc);
287 }
288 
289 static struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
290 {
291 	struct symbol_cache *sc;
292 
293 	if (!sym || strlen(sym) == 0)
294 		return NULL;
295 	sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
296 	if (!sc)
297 		return NULL;
298 
299 	sc->symbol = kstrdup(sym, GFP_KERNEL);
300 	if (!sc->symbol) {
301 		kfree(sc);
302 		return NULL;
303 	}
304 	sc->offset = offset;
305 
306 	update_symbol_cache(sc);
307 	return sc;
308 }
309 
310 #define DEFINE_FETCH_symbol(type)					\
311 static __kprobes void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs,\
312 					  void *data, void *dest)	\
313 {									\
314 	struct symbol_cache *sc = data;					\
315 	if (sc->addr)							\
316 		fetch_memory_##type(regs, (void *)sc->addr, dest);	\
317 	else								\
318 		*(type *)dest = 0;					\
319 }
320 DEFINE_BASIC_FETCH_FUNCS(symbol)
321 DEFINE_FETCH_symbol(string)
322 DEFINE_FETCH_symbol(string_size)
323 
324 /* Dereference memory access function */
325 struct deref_fetch_param {
326 	struct fetch_param orig;
327 	long offset;
328 };
329 
330 #define DEFINE_FETCH_deref(type)					\
331 static __kprobes void FETCH_FUNC_NAME(deref, type)(struct pt_regs *regs,\
332 					    void *data, void *dest)	\
333 {									\
334 	struct deref_fetch_param *dprm = data;				\
335 	unsigned long addr;						\
336 	call_fetch(&dprm->orig, regs, &addr);				\
337 	if (addr) {							\
338 		addr += dprm->offset;					\
339 		fetch_memory_##type(regs, (void *)addr, dest);		\
340 	} else								\
341 		*(type *)dest = 0;					\
342 }
343 DEFINE_BASIC_FETCH_FUNCS(deref)
344 DEFINE_FETCH_deref(string)
345 DEFINE_FETCH_deref(string_size)
346 
347 static __kprobes void free_deref_fetch_param(struct deref_fetch_param *data)
348 {
349 	if (CHECK_FETCH_FUNCS(deref, data->orig.fn))
350 		free_deref_fetch_param(data->orig.data);
351 	else if (CHECK_FETCH_FUNCS(symbol, data->orig.fn))
352 		free_symbol_cache(data->orig.data);
353 	kfree(data);
354 }
355 
356 /* Default (unsigned long) fetch type */
357 #define __DEFAULT_FETCH_TYPE(t) u##t
358 #define _DEFAULT_FETCH_TYPE(t) __DEFAULT_FETCH_TYPE(t)
359 #define DEFAULT_FETCH_TYPE _DEFAULT_FETCH_TYPE(BITS_PER_LONG)
360 #define DEFAULT_FETCH_TYPE_STR __stringify(DEFAULT_FETCH_TYPE)
361 
362 /* Fetch types */
363 enum {
364 	FETCH_MTD_reg = 0,
365 	FETCH_MTD_stack,
366 	FETCH_MTD_retval,
367 	FETCH_MTD_memory,
368 	FETCH_MTD_symbol,
369 	FETCH_MTD_deref,
370 	FETCH_MTD_END,
371 };
372 
373 #define ASSIGN_FETCH_FUNC(method, type)	\
374 	[FETCH_MTD_##method] = FETCH_FUNC_NAME(method, type)
375 
376 #define __ASSIGN_FETCH_TYPE(_name, ptype, ftype, _size, sign, _fmttype)	\
377 	{.name = _name,				\
378 	 .size = _size,					\
379 	 .is_signed = sign,				\
380 	 .print = PRINT_TYPE_FUNC_NAME(ptype),		\
381 	 .fmt = PRINT_TYPE_FMT_NAME(ptype),		\
382 	 .fmttype = _fmttype,				\
383 	 .fetch = {					\
384 ASSIGN_FETCH_FUNC(reg, ftype),				\
385 ASSIGN_FETCH_FUNC(stack, ftype),			\
386 ASSIGN_FETCH_FUNC(retval, ftype),			\
387 ASSIGN_FETCH_FUNC(memory, ftype),			\
388 ASSIGN_FETCH_FUNC(symbol, ftype),			\
389 ASSIGN_FETCH_FUNC(deref, ftype),			\
390 	  }						\
391 	}
392 
393 #define ASSIGN_FETCH_TYPE(ptype, ftype, sign)			\
394 	__ASSIGN_FETCH_TYPE(#ptype, ptype, ftype, sizeof(ftype), sign, #ptype)
395 
396 #define FETCH_TYPE_STRING 0
397 #define FETCH_TYPE_STRSIZE 1
398 
399 /* Fetch type information table */
400 static const struct fetch_type {
401 	const char	*name;		/* Name of type */
402 	size_t		size;		/* Byte size of type */
403 	int		is_signed;	/* Signed flag */
404 	print_type_func_t	print;	/* Print functions */
405 	const char	*fmt;		/* Fromat string */
406 	const char	*fmttype;	/* Name in format file */
407 	/* Fetch functions */
408 	fetch_func_t	fetch[FETCH_MTD_END];
409 } fetch_type_table[] = {
410 	/* Special types */
411 	[FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
412 					sizeof(u32), 1, "__data_loc char[]"),
413 	[FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
414 					string_size, sizeof(u32), 0, "u32"),
415 	/* Basic types */
416 	ASSIGN_FETCH_TYPE(u8,  u8,  0),
417 	ASSIGN_FETCH_TYPE(u16, u16, 0),
418 	ASSIGN_FETCH_TYPE(u32, u32, 0),
419 	ASSIGN_FETCH_TYPE(u64, u64, 0),
420 	ASSIGN_FETCH_TYPE(s8,  u8,  1),
421 	ASSIGN_FETCH_TYPE(s16, u16, 1),
422 	ASSIGN_FETCH_TYPE(s32, u32, 1),
423 	ASSIGN_FETCH_TYPE(s64, u64, 1),
424 };
425 
426 static const struct fetch_type *find_fetch_type(const char *type)
427 {
428 	int i;
429 
430 	if (!type)
431 		type = DEFAULT_FETCH_TYPE_STR;
432 
433 	for (i = 0; i < ARRAY_SIZE(fetch_type_table); i++)
434 		if (strcmp(type, fetch_type_table[i].name) == 0)
435 			return &fetch_type_table[i];
436 	return NULL;
437 }
438 
439 /* Special function : only accept unsigned long */
440 static __kprobes void fetch_stack_address(struct pt_regs *regs,
441 					  void *dummy, void *dest)
442 {
443 	*(unsigned long *)dest = kernel_stack_pointer(regs);
444 }
445 
446 static fetch_func_t get_fetch_size_function(const struct fetch_type *type,
447 					    fetch_func_t orig_fn)
448 {
449 	int i;
450 
451 	if (type != &fetch_type_table[FETCH_TYPE_STRING])
452 		return NULL;	/* Only string type needs size function */
453 	for (i = 0; i < FETCH_MTD_END; i++)
454 		if (type->fetch[i] == orig_fn)
455 			return fetch_type_table[FETCH_TYPE_STRSIZE].fetch[i];
456 
457 	WARN_ON(1);	/* This should not happen */
458 	return NULL;
459 }
460 
461 /**
462  * Kprobe event core functions
463  */
464 
465 struct probe_arg {
466 	struct fetch_param	fetch;
467 	struct fetch_param	fetch_size;
468 	unsigned int		offset;	/* Offset from argument entry */
469 	const char		*name;	/* Name of this argument */
470 	const char		*comm;	/* Command of this argument */
471 	const struct fetch_type	*type;	/* Type of this argument */
472 };
473 
474 /* Flags for trace_probe */
475 #define TP_FLAG_TRACE	1
476 #define TP_FLAG_PROFILE	2
477 
478 struct trace_probe {
479 	struct list_head	list;
480 	struct kretprobe	rp;	/* Use rp.kp for kprobe use */
481 	unsigned long 		nhit;
482 	unsigned int		flags;	/* For TP_FLAG_* */
483 	const char		*symbol;	/* symbol name */
484 	struct ftrace_event_class	class;
485 	struct ftrace_event_call	call;
486 	ssize_t			size;		/* trace entry size */
487 	unsigned int		nr_args;
488 	struct probe_arg	args[];
489 };
490 
491 #define SIZEOF_TRACE_PROBE(n)			\
492 	(offsetof(struct trace_probe, args) +	\
493 	(sizeof(struct probe_arg) * (n)))
494 
495 
496 static __kprobes int probe_is_return(struct trace_probe *tp)
497 {
498 	return tp->rp.handler != NULL;
499 }
500 
501 static __kprobes const char *probe_symbol(struct trace_probe *tp)
502 {
503 	return tp->symbol ? tp->symbol : "unknown";
504 }
505 
506 static int register_probe_event(struct trace_probe *tp);
507 static void unregister_probe_event(struct trace_probe *tp);
508 
509 static DEFINE_MUTEX(probe_lock);
510 static LIST_HEAD(probe_list);
511 
512 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
513 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
514 				struct pt_regs *regs);
515 
516 /* Check the name is good for event/group/fields */
517 static int is_good_name(const char *name)
518 {
519 	if (!isalpha(*name) && *name != '_')
520 		return 0;
521 	while (*++name != '\0') {
522 		if (!isalpha(*name) && !isdigit(*name) && *name != '_')
523 			return 0;
524 	}
525 	return 1;
526 }
527 
528 /*
529  * Allocate new trace_probe and initialize it (including kprobes).
530  */
531 static struct trace_probe *alloc_trace_probe(const char *group,
532 					     const char *event,
533 					     void *addr,
534 					     const char *symbol,
535 					     unsigned long offs,
536 					     int nargs, int is_return)
537 {
538 	struct trace_probe *tp;
539 	int ret = -ENOMEM;
540 
541 	tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
542 	if (!tp)
543 		return ERR_PTR(ret);
544 
545 	if (symbol) {
546 		tp->symbol = kstrdup(symbol, GFP_KERNEL);
547 		if (!tp->symbol)
548 			goto error;
549 		tp->rp.kp.symbol_name = tp->symbol;
550 		tp->rp.kp.offset = offs;
551 	} else
552 		tp->rp.kp.addr = addr;
553 
554 	if (is_return)
555 		tp->rp.handler = kretprobe_dispatcher;
556 	else
557 		tp->rp.kp.pre_handler = kprobe_dispatcher;
558 
559 	if (!event || !is_good_name(event)) {
560 		ret = -EINVAL;
561 		goto error;
562 	}
563 
564 	tp->call.class = &tp->class;
565 	tp->call.name = kstrdup(event, GFP_KERNEL);
566 	if (!tp->call.name)
567 		goto error;
568 
569 	if (!group || !is_good_name(group)) {
570 		ret = -EINVAL;
571 		goto error;
572 	}
573 
574 	tp->class.system = kstrdup(group, GFP_KERNEL);
575 	if (!tp->class.system)
576 		goto error;
577 
578 	INIT_LIST_HEAD(&tp->list);
579 	return tp;
580 error:
581 	kfree(tp->call.name);
582 	kfree(tp->symbol);
583 	kfree(tp);
584 	return ERR_PTR(ret);
585 }
586 
587 static void free_probe_arg(struct probe_arg *arg)
588 {
589 	if (CHECK_FETCH_FUNCS(deref, arg->fetch.fn))
590 		free_deref_fetch_param(arg->fetch.data);
591 	else if (CHECK_FETCH_FUNCS(symbol, arg->fetch.fn))
592 		free_symbol_cache(arg->fetch.data);
593 	kfree(arg->name);
594 	kfree(arg->comm);
595 }
596 
597 static void free_trace_probe(struct trace_probe *tp)
598 {
599 	int i;
600 
601 	for (i = 0; i < tp->nr_args; i++)
602 		free_probe_arg(&tp->args[i]);
603 
604 	kfree(tp->call.class->system);
605 	kfree(tp->call.name);
606 	kfree(tp->symbol);
607 	kfree(tp);
608 }
609 
610 static struct trace_probe *find_probe_event(const char *event,
611 					    const char *group)
612 {
613 	struct trace_probe *tp;
614 
615 	list_for_each_entry(tp, &probe_list, list)
616 		if (strcmp(tp->call.name, event) == 0 &&
617 		    strcmp(tp->call.class->system, group) == 0)
618 			return tp;
619 	return NULL;
620 }
621 
622 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
623 static void unregister_trace_probe(struct trace_probe *tp)
624 {
625 	if (probe_is_return(tp))
626 		unregister_kretprobe(&tp->rp);
627 	else
628 		unregister_kprobe(&tp->rp.kp);
629 	list_del(&tp->list);
630 	unregister_probe_event(tp);
631 }
632 
633 /* Register a trace_probe and probe_event */
634 static int register_trace_probe(struct trace_probe *tp)
635 {
636 	struct trace_probe *old_tp;
637 	int ret;
638 
639 	mutex_lock(&probe_lock);
640 
641 	/* register as an event */
642 	old_tp = find_probe_event(tp->call.name, tp->call.class->system);
643 	if (old_tp) {
644 		/* delete old event */
645 		unregister_trace_probe(old_tp);
646 		free_trace_probe(old_tp);
647 	}
648 	ret = register_probe_event(tp);
649 	if (ret) {
650 		pr_warning("Failed to register probe event(%d)\n", ret);
651 		goto end;
652 	}
653 
654 	tp->rp.kp.flags |= KPROBE_FLAG_DISABLED;
655 	if (probe_is_return(tp))
656 		ret = register_kretprobe(&tp->rp);
657 	else
658 		ret = register_kprobe(&tp->rp.kp);
659 
660 	if (ret) {
661 		pr_warning("Could not insert probe(%d)\n", ret);
662 		if (ret == -EILSEQ) {
663 			pr_warning("Probing address(0x%p) is not an "
664 				   "instruction boundary.\n",
665 				   tp->rp.kp.addr);
666 			ret = -EINVAL;
667 		}
668 		unregister_probe_event(tp);
669 	} else
670 		list_add_tail(&tp->list, &probe_list);
671 end:
672 	mutex_unlock(&probe_lock);
673 	return ret;
674 }
675 
676 /* Split symbol and offset. */
677 static int split_symbol_offset(char *symbol, unsigned long *offset)
678 {
679 	char *tmp;
680 	int ret;
681 
682 	if (!offset)
683 		return -EINVAL;
684 
685 	tmp = strchr(symbol, '+');
686 	if (tmp) {
687 		/* skip sign because strict_strtol doesn't accept '+' */
688 		ret = strict_strtoul(tmp + 1, 0, offset);
689 		if (ret)
690 			return ret;
691 		*tmp = '\0';
692 	} else
693 		*offset = 0;
694 	return 0;
695 }
696 
697 #define PARAM_MAX_ARGS 16
698 #define PARAM_MAX_STACK (THREAD_SIZE / sizeof(unsigned long))
699 
700 static int parse_probe_vars(char *arg, const struct fetch_type *t,
701 			    struct fetch_param *f, int is_return)
702 {
703 	int ret = 0;
704 	unsigned long param;
705 
706 	if (strcmp(arg, "retval") == 0) {
707 		if (is_return)
708 			f->fn = t->fetch[FETCH_MTD_retval];
709 		else
710 			ret = -EINVAL;
711 	} else if (strncmp(arg, "stack", 5) == 0) {
712 		if (arg[5] == '\0') {
713 			if (strcmp(t->name, DEFAULT_FETCH_TYPE_STR) == 0)
714 				f->fn = fetch_stack_address;
715 			else
716 				ret = -EINVAL;
717 		} else if (isdigit(arg[5])) {
718 			ret = strict_strtoul(arg + 5, 10, &param);
719 			if (ret || param > PARAM_MAX_STACK)
720 				ret = -EINVAL;
721 			else {
722 				f->fn = t->fetch[FETCH_MTD_stack];
723 				f->data = (void *)param;
724 			}
725 		} else
726 			ret = -EINVAL;
727 	} else
728 		ret = -EINVAL;
729 	return ret;
730 }
731 
732 /* Recursive argument parser */
733 static int __parse_probe_arg(char *arg, const struct fetch_type *t,
734 			     struct fetch_param *f, int is_return)
735 {
736 	int ret = 0;
737 	unsigned long param;
738 	long offset;
739 	char *tmp;
740 
741 	switch (arg[0]) {
742 	case '$':
743 		ret = parse_probe_vars(arg + 1, t, f, is_return);
744 		break;
745 	case '%':	/* named register */
746 		ret = regs_query_register_offset(arg + 1);
747 		if (ret >= 0) {
748 			f->fn = t->fetch[FETCH_MTD_reg];
749 			f->data = (void *)(unsigned long)ret;
750 			ret = 0;
751 		}
752 		break;
753 	case '@':	/* memory or symbol */
754 		if (isdigit(arg[1])) {
755 			ret = strict_strtoul(arg + 1, 0, &param);
756 			if (ret)
757 				break;
758 			f->fn = t->fetch[FETCH_MTD_memory];
759 			f->data = (void *)param;
760 		} else {
761 			ret = split_symbol_offset(arg + 1, &offset);
762 			if (ret)
763 				break;
764 			f->data = alloc_symbol_cache(arg + 1, offset);
765 			if (f->data)
766 				f->fn = t->fetch[FETCH_MTD_symbol];
767 		}
768 		break;
769 	case '+':	/* deref memory */
770 	case '-':
771 		tmp = strchr(arg, '(');
772 		if (!tmp)
773 			break;
774 		*tmp = '\0';
775 		ret = strict_strtol(arg + 1, 0, &offset);
776 		if (ret)
777 			break;
778 		if (arg[0] == '-')
779 			offset = -offset;
780 		arg = tmp + 1;
781 		tmp = strrchr(arg, ')');
782 		if (tmp) {
783 			struct deref_fetch_param *dprm;
784 			const struct fetch_type *t2 = find_fetch_type(NULL);
785 			*tmp = '\0';
786 			dprm = kzalloc(sizeof(struct deref_fetch_param),
787 				       GFP_KERNEL);
788 			if (!dprm)
789 				return -ENOMEM;
790 			dprm->offset = offset;
791 			ret = __parse_probe_arg(arg, t2, &dprm->orig,
792 						is_return);
793 			if (ret)
794 				kfree(dprm);
795 			else {
796 				f->fn = t->fetch[FETCH_MTD_deref];
797 				f->data = (void *)dprm;
798 			}
799 		}
800 		break;
801 	}
802 	if (!ret && !f->fn) {	/* Parsed, but do not find fetch method */
803 		pr_info("%s type has no corresponding fetch method.\n",
804 			t->name);
805 		ret = -EINVAL;
806 	}
807 	return ret;
808 }
809 
810 /* String length checking wrapper */
811 static int parse_probe_arg(char *arg, struct trace_probe *tp,
812 			   struct probe_arg *parg, int is_return)
813 {
814 	const char *t;
815 	int ret;
816 
817 	if (strlen(arg) > MAX_ARGSTR_LEN) {
818 		pr_info("Argument is too long.: %s\n",  arg);
819 		return -ENOSPC;
820 	}
821 	parg->comm = kstrdup(arg, GFP_KERNEL);
822 	if (!parg->comm) {
823 		pr_info("Failed to allocate memory for command '%s'.\n", arg);
824 		return -ENOMEM;
825 	}
826 	t = strchr(parg->comm, ':');
827 	if (t) {
828 		arg[t - parg->comm] = '\0';
829 		t++;
830 	}
831 	parg->type = find_fetch_type(t);
832 	if (!parg->type) {
833 		pr_info("Unsupported type: %s\n", t);
834 		return -EINVAL;
835 	}
836 	parg->offset = tp->size;
837 	tp->size += parg->type->size;
838 	ret = __parse_probe_arg(arg, parg->type, &parg->fetch, is_return);
839 	if (ret >= 0) {
840 		parg->fetch_size.fn = get_fetch_size_function(parg->type,
841 							      parg->fetch.fn);
842 		parg->fetch_size.data = parg->fetch.data;
843 	}
844 	return ret;
845 }
846 
847 /* Return 1 if name is reserved or already used by another argument */
848 static int conflict_field_name(const char *name,
849 			       struct probe_arg *args, int narg)
850 {
851 	int i;
852 	for (i = 0; i < ARRAY_SIZE(reserved_field_names); i++)
853 		if (strcmp(reserved_field_names[i], name) == 0)
854 			return 1;
855 	for (i = 0; i < narg; i++)
856 		if (strcmp(args[i].name, name) == 0)
857 			return 1;
858 	return 0;
859 }
860 
861 static int create_trace_probe(int argc, char **argv)
862 {
863 	/*
864 	 * Argument syntax:
865 	 *  - Add kprobe: p[:[GRP/]EVENT] KSYM[+OFFS]|KADDR [FETCHARGS]
866 	 *  - Add kretprobe: r[:[GRP/]EVENT] KSYM[+0] [FETCHARGS]
867 	 * Fetch args:
868 	 *  $retval	: fetch return value
869 	 *  $stack	: fetch stack address
870 	 *  $stackN	: fetch Nth of stack (N:0-)
871 	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
872 	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
873 	 *  %REG	: fetch register REG
874 	 * Dereferencing memory fetch:
875 	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
876 	 * Alias name of args:
877 	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
878 	 * Type of args:
879 	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
880 	 */
881 	struct trace_probe *tp;
882 	int i, ret = 0;
883 	int is_return = 0, is_delete = 0;
884 	char *symbol = NULL, *event = NULL, *group = NULL;
885 	char *arg;
886 	unsigned long offset = 0;
887 	void *addr = NULL;
888 	char buf[MAX_EVENT_NAME_LEN];
889 
890 	/* argc must be >= 1 */
891 	if (argv[0][0] == 'p')
892 		is_return = 0;
893 	else if (argv[0][0] == 'r')
894 		is_return = 1;
895 	else if (argv[0][0] == '-')
896 		is_delete = 1;
897 	else {
898 		pr_info("Probe definition must be started with 'p', 'r' or"
899 			" '-'.\n");
900 		return -EINVAL;
901 	}
902 
903 	if (argv[0][1] == ':') {
904 		event = &argv[0][2];
905 		if (strchr(event, '/')) {
906 			group = event;
907 			event = strchr(group, '/') + 1;
908 			event[-1] = '\0';
909 			if (strlen(group) == 0) {
910 				pr_info("Group name is not specified\n");
911 				return -EINVAL;
912 			}
913 		}
914 		if (strlen(event) == 0) {
915 			pr_info("Event name is not specified\n");
916 			return -EINVAL;
917 		}
918 	}
919 	if (!group)
920 		group = KPROBE_EVENT_SYSTEM;
921 
922 	if (is_delete) {
923 		if (!event) {
924 			pr_info("Delete command needs an event name.\n");
925 			return -EINVAL;
926 		}
927 		mutex_lock(&probe_lock);
928 		tp = find_probe_event(event, group);
929 		if (!tp) {
930 			mutex_unlock(&probe_lock);
931 			pr_info("Event %s/%s doesn't exist.\n", group, event);
932 			return -ENOENT;
933 		}
934 		/* delete an event */
935 		unregister_trace_probe(tp);
936 		free_trace_probe(tp);
937 		mutex_unlock(&probe_lock);
938 		return 0;
939 	}
940 
941 	if (argc < 2) {
942 		pr_info("Probe point is not specified.\n");
943 		return -EINVAL;
944 	}
945 	if (isdigit(argv[1][0])) {
946 		if (is_return) {
947 			pr_info("Return probe point must be a symbol.\n");
948 			return -EINVAL;
949 		}
950 		/* an address specified */
951 		ret = strict_strtoul(&argv[1][0], 0, (unsigned long *)&addr);
952 		if (ret) {
953 			pr_info("Failed to parse address.\n");
954 			return ret;
955 		}
956 	} else {
957 		/* a symbol specified */
958 		symbol = argv[1];
959 		/* TODO: support .init module functions */
960 		ret = split_symbol_offset(symbol, &offset);
961 		if (ret) {
962 			pr_info("Failed to parse symbol.\n");
963 			return ret;
964 		}
965 		if (offset && is_return) {
966 			pr_info("Return probe must be used without offset.\n");
967 			return -EINVAL;
968 		}
969 	}
970 	argc -= 2; argv += 2;
971 
972 	/* setup a probe */
973 	if (!event) {
974 		/* Make a new event name */
975 		if (symbol)
976 			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
977 				 is_return ? 'r' : 'p', symbol, offset);
978 		else
979 			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
980 				 is_return ? 'r' : 'p', addr);
981 		event = buf;
982 	}
983 	tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
984 			       is_return);
985 	if (IS_ERR(tp)) {
986 		pr_info("Failed to allocate trace_probe.(%d)\n",
987 			(int)PTR_ERR(tp));
988 		return PTR_ERR(tp);
989 	}
990 
991 	/* parse arguments */
992 	ret = 0;
993 	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
994 		/* Increment count for freeing args in error case */
995 		tp->nr_args++;
996 
997 		/* Parse argument name */
998 		arg = strchr(argv[i], '=');
999 		if (arg) {
1000 			*arg++ = '\0';
1001 			tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
1002 		} else {
1003 			arg = argv[i];
1004 			/* If argument name is omitted, set "argN" */
1005 			snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
1006 			tp->args[i].name = kstrdup(buf, GFP_KERNEL);
1007 		}
1008 
1009 		if (!tp->args[i].name) {
1010 			pr_info("Failed to allocate argument[%d] name.\n", i);
1011 			ret = -ENOMEM;
1012 			goto error;
1013 		}
1014 
1015 		if (!is_good_name(tp->args[i].name)) {
1016 			pr_info("Invalid argument[%d] name: %s\n",
1017 				i, tp->args[i].name);
1018 			ret = -EINVAL;
1019 			goto error;
1020 		}
1021 
1022 		if (conflict_field_name(tp->args[i].name, tp->args, i)) {
1023 			pr_info("Argument[%d] name '%s' conflicts with "
1024 				"another field.\n", i, argv[i]);
1025 			ret = -EINVAL;
1026 			goto error;
1027 		}
1028 
1029 		/* Parse fetch argument */
1030 		ret = parse_probe_arg(arg, tp, &tp->args[i], is_return);
1031 		if (ret) {
1032 			pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
1033 			goto error;
1034 		}
1035 	}
1036 
1037 	ret = register_trace_probe(tp);
1038 	if (ret)
1039 		goto error;
1040 	return 0;
1041 
1042 error:
1043 	free_trace_probe(tp);
1044 	return ret;
1045 }
1046 
1047 static void cleanup_all_probes(void)
1048 {
1049 	struct trace_probe *tp;
1050 
1051 	mutex_lock(&probe_lock);
1052 	/* TODO: Use batch unregistration */
1053 	while (!list_empty(&probe_list)) {
1054 		tp = list_entry(probe_list.next, struct trace_probe, list);
1055 		unregister_trace_probe(tp);
1056 		free_trace_probe(tp);
1057 	}
1058 	mutex_unlock(&probe_lock);
1059 }
1060 
1061 
1062 /* Probes listing interfaces */
1063 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
1064 {
1065 	mutex_lock(&probe_lock);
1066 	return seq_list_start(&probe_list, *pos);
1067 }
1068 
1069 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
1070 {
1071 	return seq_list_next(v, &probe_list, pos);
1072 }
1073 
1074 static void probes_seq_stop(struct seq_file *m, void *v)
1075 {
1076 	mutex_unlock(&probe_lock);
1077 }
1078 
1079 static int probes_seq_show(struct seq_file *m, void *v)
1080 {
1081 	struct trace_probe *tp = v;
1082 	int i;
1083 
1084 	seq_printf(m, "%c", probe_is_return(tp) ? 'r' : 'p');
1085 	seq_printf(m, ":%s/%s", tp->call.class->system, tp->call.name);
1086 
1087 	if (!tp->symbol)
1088 		seq_printf(m, " 0x%p", tp->rp.kp.addr);
1089 	else if (tp->rp.kp.offset)
1090 		seq_printf(m, " %s+%u", probe_symbol(tp), tp->rp.kp.offset);
1091 	else
1092 		seq_printf(m, " %s", probe_symbol(tp));
1093 
1094 	for (i = 0; i < tp->nr_args; i++)
1095 		seq_printf(m, " %s=%s", tp->args[i].name, tp->args[i].comm);
1096 	seq_printf(m, "\n");
1097 
1098 	return 0;
1099 }
1100 
1101 static const struct seq_operations probes_seq_op = {
1102 	.start  = probes_seq_start,
1103 	.next   = probes_seq_next,
1104 	.stop   = probes_seq_stop,
1105 	.show   = probes_seq_show
1106 };
1107 
1108 static int probes_open(struct inode *inode, struct file *file)
1109 {
1110 	if ((file->f_mode & FMODE_WRITE) &&
1111 	    (file->f_flags & O_TRUNC))
1112 		cleanup_all_probes();
1113 
1114 	return seq_open(file, &probes_seq_op);
1115 }
1116 
1117 static int command_trace_probe(const char *buf)
1118 {
1119 	char **argv;
1120 	int argc = 0, ret = 0;
1121 
1122 	argv = argv_split(GFP_KERNEL, buf, &argc);
1123 	if (!argv)
1124 		return -ENOMEM;
1125 
1126 	if (argc)
1127 		ret = create_trace_probe(argc, argv);
1128 
1129 	argv_free(argv);
1130 	return ret;
1131 }
1132 
1133 #define WRITE_BUFSIZE 128
1134 
1135 static ssize_t probes_write(struct file *file, const char __user *buffer,
1136 			    size_t count, loff_t *ppos)
1137 {
1138 	char *kbuf, *tmp;
1139 	int ret;
1140 	size_t done;
1141 	size_t size;
1142 
1143 	kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
1144 	if (!kbuf)
1145 		return -ENOMEM;
1146 
1147 	ret = done = 0;
1148 	while (done < count) {
1149 		size = count - done;
1150 		if (size >= WRITE_BUFSIZE)
1151 			size = WRITE_BUFSIZE - 1;
1152 		if (copy_from_user(kbuf, buffer + done, size)) {
1153 			ret = -EFAULT;
1154 			goto out;
1155 		}
1156 		kbuf[size] = '\0';
1157 		tmp = strchr(kbuf, '\n');
1158 		if (tmp) {
1159 			*tmp = '\0';
1160 			size = tmp - kbuf + 1;
1161 		} else if (done + size < count) {
1162 			pr_warning("Line length is too long: "
1163 				   "Should be less than %d.", WRITE_BUFSIZE);
1164 			ret = -EINVAL;
1165 			goto out;
1166 		}
1167 		done += size;
1168 		/* Remove comments */
1169 		tmp = strchr(kbuf, '#');
1170 		if (tmp)
1171 			*tmp = '\0';
1172 
1173 		ret = command_trace_probe(kbuf);
1174 		if (ret)
1175 			goto out;
1176 	}
1177 	ret = done;
1178 out:
1179 	kfree(kbuf);
1180 	return ret;
1181 }
1182 
1183 static const struct file_operations kprobe_events_ops = {
1184 	.owner          = THIS_MODULE,
1185 	.open           = probes_open,
1186 	.read           = seq_read,
1187 	.llseek         = seq_lseek,
1188 	.release        = seq_release,
1189 	.write		= probes_write,
1190 };
1191 
1192 /* Probes profiling interfaces */
1193 static int probes_profile_seq_show(struct seq_file *m, void *v)
1194 {
1195 	struct trace_probe *tp = v;
1196 
1197 	seq_printf(m, "  %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
1198 		   tp->rp.kp.nmissed);
1199 
1200 	return 0;
1201 }
1202 
1203 static const struct seq_operations profile_seq_op = {
1204 	.start  = probes_seq_start,
1205 	.next   = probes_seq_next,
1206 	.stop   = probes_seq_stop,
1207 	.show   = probes_profile_seq_show
1208 };
1209 
1210 static int profile_open(struct inode *inode, struct file *file)
1211 {
1212 	return seq_open(file, &profile_seq_op);
1213 }
1214 
1215 static const struct file_operations kprobe_profile_ops = {
1216 	.owner          = THIS_MODULE,
1217 	.open           = profile_open,
1218 	.read           = seq_read,
1219 	.llseek         = seq_lseek,
1220 	.release        = seq_release,
1221 };
1222 
1223 /* Sum up total data length for dynamic arraies (strings) */
1224 static __kprobes int __get_data_size(struct trace_probe *tp,
1225 				     struct pt_regs *regs)
1226 {
1227 	int i, ret = 0;
1228 	u32 len;
1229 
1230 	for (i = 0; i < tp->nr_args; i++)
1231 		if (unlikely(tp->args[i].fetch_size.fn)) {
1232 			call_fetch(&tp->args[i].fetch_size, regs, &len);
1233 			ret += len;
1234 		}
1235 
1236 	return ret;
1237 }
1238 
1239 /* Store the value of each argument */
1240 static __kprobes void store_trace_args(int ent_size, struct trace_probe *tp,
1241 				       struct pt_regs *regs,
1242 				       u8 *data, int maxlen)
1243 {
1244 	int i;
1245 	u32 end = tp->size;
1246 	u32 *dl;	/* Data (relative) location */
1247 
1248 	for (i = 0; i < tp->nr_args; i++) {
1249 		if (unlikely(tp->args[i].fetch_size.fn)) {
1250 			/*
1251 			 * First, we set the relative location and
1252 			 * maximum data length to *dl
1253 			 */
1254 			dl = (u32 *)(data + tp->args[i].offset);
1255 			*dl = make_data_rloc(maxlen, end - tp->args[i].offset);
1256 			/* Then try to fetch string or dynamic array data */
1257 			call_fetch(&tp->args[i].fetch, regs, dl);
1258 			/* Reduce maximum length */
1259 			end += get_rloc_len(*dl);
1260 			maxlen -= get_rloc_len(*dl);
1261 			/* Trick here, convert data_rloc to data_loc */
1262 			*dl = convert_rloc_to_loc(*dl,
1263 				 ent_size + tp->args[i].offset);
1264 		} else
1265 			/* Just fetching data normally */
1266 			call_fetch(&tp->args[i].fetch, regs,
1267 				   data + tp->args[i].offset);
1268 	}
1269 }
1270 
1271 /* Kprobe handler */
1272 static __kprobes void kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
1273 {
1274 	struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1275 	struct kprobe_trace_entry_head *entry;
1276 	struct ring_buffer_event *event;
1277 	struct ring_buffer *buffer;
1278 	int size, dsize, pc;
1279 	unsigned long irq_flags;
1280 	struct ftrace_event_call *call = &tp->call;
1281 
1282 	tp->nhit++;
1283 
1284 	local_save_flags(irq_flags);
1285 	pc = preempt_count();
1286 
1287 	dsize = __get_data_size(tp, regs);
1288 	size = sizeof(*entry) + tp->size + dsize;
1289 
1290 	event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
1291 						  size, irq_flags, pc);
1292 	if (!event)
1293 		return;
1294 
1295 	entry = ring_buffer_event_data(event);
1296 	entry->ip = (unsigned long)kp->addr;
1297 	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1298 
1299 	if (!filter_current_check_discard(buffer, call, entry, event))
1300 		trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
1301 }
1302 
1303 /* Kretprobe handler */
1304 static __kprobes void kretprobe_trace_func(struct kretprobe_instance *ri,
1305 					  struct pt_regs *regs)
1306 {
1307 	struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1308 	struct kretprobe_trace_entry_head *entry;
1309 	struct ring_buffer_event *event;
1310 	struct ring_buffer *buffer;
1311 	int size, pc, dsize;
1312 	unsigned long irq_flags;
1313 	struct ftrace_event_call *call = &tp->call;
1314 
1315 	local_save_flags(irq_flags);
1316 	pc = preempt_count();
1317 
1318 	dsize = __get_data_size(tp, regs);
1319 	size = sizeof(*entry) + tp->size + dsize;
1320 
1321 	event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
1322 						  size, irq_flags, pc);
1323 	if (!event)
1324 		return;
1325 
1326 	entry = ring_buffer_event_data(event);
1327 	entry->func = (unsigned long)tp->rp.kp.addr;
1328 	entry->ret_ip = (unsigned long)ri->ret_addr;
1329 	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1330 
1331 	if (!filter_current_check_discard(buffer, call, entry, event))
1332 		trace_nowake_buffer_unlock_commit(buffer, event, irq_flags, pc);
1333 }
1334 
1335 /* Event entry printers */
1336 enum print_line_t
1337 print_kprobe_event(struct trace_iterator *iter, int flags,
1338 		   struct trace_event *event)
1339 {
1340 	struct kprobe_trace_entry_head *field;
1341 	struct trace_seq *s = &iter->seq;
1342 	struct trace_probe *tp;
1343 	u8 *data;
1344 	int i;
1345 
1346 	field = (struct kprobe_trace_entry_head *)iter->ent;
1347 	tp = container_of(event, struct trace_probe, call.event);
1348 
1349 	if (!trace_seq_printf(s, "%s: (", tp->call.name))
1350 		goto partial;
1351 
1352 	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1353 		goto partial;
1354 
1355 	if (!trace_seq_puts(s, ")"))
1356 		goto partial;
1357 
1358 	data = (u8 *)&field[1];
1359 	for (i = 0; i < tp->nr_args; i++)
1360 		if (!tp->args[i].type->print(s, tp->args[i].name,
1361 					     data + tp->args[i].offset, field))
1362 			goto partial;
1363 
1364 	if (!trace_seq_puts(s, "\n"))
1365 		goto partial;
1366 
1367 	return TRACE_TYPE_HANDLED;
1368 partial:
1369 	return TRACE_TYPE_PARTIAL_LINE;
1370 }
1371 
1372 enum print_line_t
1373 print_kretprobe_event(struct trace_iterator *iter, int flags,
1374 		      struct trace_event *event)
1375 {
1376 	struct kretprobe_trace_entry_head *field;
1377 	struct trace_seq *s = &iter->seq;
1378 	struct trace_probe *tp;
1379 	u8 *data;
1380 	int i;
1381 
1382 	field = (struct kretprobe_trace_entry_head *)iter->ent;
1383 	tp = container_of(event, struct trace_probe, call.event);
1384 
1385 	if (!trace_seq_printf(s, "%s: (", tp->call.name))
1386 		goto partial;
1387 
1388 	if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1389 		goto partial;
1390 
1391 	if (!trace_seq_puts(s, " <- "))
1392 		goto partial;
1393 
1394 	if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1395 		goto partial;
1396 
1397 	if (!trace_seq_puts(s, ")"))
1398 		goto partial;
1399 
1400 	data = (u8 *)&field[1];
1401 	for (i = 0; i < tp->nr_args; i++)
1402 		if (!tp->args[i].type->print(s, tp->args[i].name,
1403 					     data + tp->args[i].offset, field))
1404 			goto partial;
1405 
1406 	if (!trace_seq_puts(s, "\n"))
1407 		goto partial;
1408 
1409 	return TRACE_TYPE_HANDLED;
1410 partial:
1411 	return TRACE_TYPE_PARTIAL_LINE;
1412 }
1413 
1414 static int probe_event_enable(struct ftrace_event_call *call)
1415 {
1416 	struct trace_probe *tp = (struct trace_probe *)call->data;
1417 
1418 	tp->flags |= TP_FLAG_TRACE;
1419 	if (probe_is_return(tp))
1420 		return enable_kretprobe(&tp->rp);
1421 	else
1422 		return enable_kprobe(&tp->rp.kp);
1423 }
1424 
1425 static void probe_event_disable(struct ftrace_event_call *call)
1426 {
1427 	struct trace_probe *tp = (struct trace_probe *)call->data;
1428 
1429 	tp->flags &= ~TP_FLAG_TRACE;
1430 	if (!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE))) {
1431 		if (probe_is_return(tp))
1432 			disable_kretprobe(&tp->rp);
1433 		else
1434 			disable_kprobe(&tp->rp.kp);
1435 	}
1436 }
1437 
1438 #undef DEFINE_FIELD
1439 #define DEFINE_FIELD(type, item, name, is_signed)			\
1440 	do {								\
1441 		ret = trace_define_field(event_call, #type, name,	\
1442 					 offsetof(typeof(field), item),	\
1443 					 sizeof(field.item), is_signed, \
1444 					 FILTER_OTHER);			\
1445 		if (ret)						\
1446 			return ret;					\
1447 	} while (0)
1448 
1449 static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
1450 {
1451 	int ret, i;
1452 	struct kprobe_trace_entry_head field;
1453 	struct trace_probe *tp = (struct trace_probe *)event_call->data;
1454 
1455 	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1456 	/* Set argument names as fields */
1457 	for (i = 0; i < tp->nr_args; i++) {
1458 		ret = trace_define_field(event_call, tp->args[i].type->fmttype,
1459 					 tp->args[i].name,
1460 					 sizeof(field) + tp->args[i].offset,
1461 					 tp->args[i].type->size,
1462 					 tp->args[i].type->is_signed,
1463 					 FILTER_OTHER);
1464 		if (ret)
1465 			return ret;
1466 	}
1467 	return 0;
1468 }
1469 
1470 static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1471 {
1472 	int ret, i;
1473 	struct kretprobe_trace_entry_head field;
1474 	struct trace_probe *tp = (struct trace_probe *)event_call->data;
1475 
1476 	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1477 	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1478 	/* Set argument names as fields */
1479 	for (i = 0; i < tp->nr_args; i++) {
1480 		ret = trace_define_field(event_call, tp->args[i].type->fmttype,
1481 					 tp->args[i].name,
1482 					 sizeof(field) + tp->args[i].offset,
1483 					 tp->args[i].type->size,
1484 					 tp->args[i].type->is_signed,
1485 					 FILTER_OTHER);
1486 		if (ret)
1487 			return ret;
1488 	}
1489 	return 0;
1490 }
1491 
1492 static int __set_print_fmt(struct trace_probe *tp, char *buf, int len)
1493 {
1494 	int i;
1495 	int pos = 0;
1496 
1497 	const char *fmt, *arg;
1498 
1499 	if (!probe_is_return(tp)) {
1500 		fmt = "(%lx)";
1501 		arg = "REC->" FIELD_STRING_IP;
1502 	} else {
1503 		fmt = "(%lx <- %lx)";
1504 		arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
1505 	}
1506 
1507 	/* When len=0, we just calculate the needed length */
1508 #define LEN_OR_ZERO (len ? len - pos : 0)
1509 
1510 	pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
1511 
1512 	for (i = 0; i < tp->nr_args; i++) {
1513 		pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
1514 				tp->args[i].name, tp->args[i].type->fmt);
1515 	}
1516 
1517 	pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
1518 
1519 	for (i = 0; i < tp->nr_args; i++) {
1520 		if (strcmp(tp->args[i].type->name, "string") == 0)
1521 			pos += snprintf(buf + pos, LEN_OR_ZERO,
1522 					", __get_str(%s)",
1523 					tp->args[i].name);
1524 		else
1525 			pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
1526 					tp->args[i].name);
1527 	}
1528 
1529 #undef LEN_OR_ZERO
1530 
1531 	/* return the length of print_fmt */
1532 	return pos;
1533 }
1534 
1535 static int set_print_fmt(struct trace_probe *tp)
1536 {
1537 	int len;
1538 	char *print_fmt;
1539 
1540 	/* First: called with 0 length to calculate the needed length */
1541 	len = __set_print_fmt(tp, NULL, 0);
1542 	print_fmt = kmalloc(len + 1, GFP_KERNEL);
1543 	if (!print_fmt)
1544 		return -ENOMEM;
1545 
1546 	/* Second: actually write the @print_fmt */
1547 	__set_print_fmt(tp, print_fmt, len + 1);
1548 	tp->call.print_fmt = print_fmt;
1549 
1550 	return 0;
1551 }
1552 
1553 #ifdef CONFIG_PERF_EVENTS
1554 
1555 /* Kprobe profile handler */
1556 static __kprobes void kprobe_perf_func(struct kprobe *kp,
1557 					 struct pt_regs *regs)
1558 {
1559 	struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1560 	struct ftrace_event_call *call = &tp->call;
1561 	struct kprobe_trace_entry_head *entry;
1562 	struct hlist_head *head;
1563 	int size, __size, dsize;
1564 	int rctx;
1565 
1566 	dsize = __get_data_size(tp, regs);
1567 	__size = sizeof(*entry) + tp->size + dsize;
1568 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1569 	size -= sizeof(u32);
1570 	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1571 		     "profile buffer not large enough"))
1572 		return;
1573 
1574 	entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1575 	if (!entry)
1576 		return;
1577 
1578 	entry->ip = (unsigned long)kp->addr;
1579 	memset(&entry[1], 0, dsize);
1580 	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1581 
1582 	head = this_cpu_ptr(call->perf_events);
1583 	perf_trace_buf_submit(entry, size, rctx, entry->ip, 1, regs, head);
1584 }
1585 
1586 /* Kretprobe profile handler */
1587 static __kprobes void kretprobe_perf_func(struct kretprobe_instance *ri,
1588 					    struct pt_regs *regs)
1589 {
1590 	struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1591 	struct ftrace_event_call *call = &tp->call;
1592 	struct kretprobe_trace_entry_head *entry;
1593 	struct hlist_head *head;
1594 	int size, __size, dsize;
1595 	int rctx;
1596 
1597 	dsize = __get_data_size(tp, regs);
1598 	__size = sizeof(*entry) + tp->size + dsize;
1599 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1600 	size -= sizeof(u32);
1601 	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1602 		     "profile buffer not large enough"))
1603 		return;
1604 
1605 	entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1606 	if (!entry)
1607 		return;
1608 
1609 	entry->func = (unsigned long)tp->rp.kp.addr;
1610 	entry->ret_ip = (unsigned long)ri->ret_addr;
1611 	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1612 
1613 	head = this_cpu_ptr(call->perf_events);
1614 	perf_trace_buf_submit(entry, size, rctx, entry->ret_ip, 1, regs, head);
1615 }
1616 
1617 static int probe_perf_enable(struct ftrace_event_call *call)
1618 {
1619 	struct trace_probe *tp = (struct trace_probe *)call->data;
1620 
1621 	tp->flags |= TP_FLAG_PROFILE;
1622 
1623 	if (probe_is_return(tp))
1624 		return enable_kretprobe(&tp->rp);
1625 	else
1626 		return enable_kprobe(&tp->rp.kp);
1627 }
1628 
1629 static void probe_perf_disable(struct ftrace_event_call *call)
1630 {
1631 	struct trace_probe *tp = (struct trace_probe *)call->data;
1632 
1633 	tp->flags &= ~TP_FLAG_PROFILE;
1634 
1635 	if (!(tp->flags & TP_FLAG_TRACE)) {
1636 		if (probe_is_return(tp))
1637 			disable_kretprobe(&tp->rp);
1638 		else
1639 			disable_kprobe(&tp->rp.kp);
1640 	}
1641 }
1642 #endif	/* CONFIG_PERF_EVENTS */
1643 
1644 static __kprobes
1645 int kprobe_register(struct ftrace_event_call *event, enum trace_reg type)
1646 {
1647 	switch (type) {
1648 	case TRACE_REG_REGISTER:
1649 		return probe_event_enable(event);
1650 	case TRACE_REG_UNREGISTER:
1651 		probe_event_disable(event);
1652 		return 0;
1653 
1654 #ifdef CONFIG_PERF_EVENTS
1655 	case TRACE_REG_PERF_REGISTER:
1656 		return probe_perf_enable(event);
1657 	case TRACE_REG_PERF_UNREGISTER:
1658 		probe_perf_disable(event);
1659 		return 0;
1660 #endif
1661 	}
1662 	return 0;
1663 }
1664 
1665 static __kprobes
1666 int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1667 {
1668 	struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1669 
1670 	if (tp->flags & TP_FLAG_TRACE)
1671 		kprobe_trace_func(kp, regs);
1672 #ifdef CONFIG_PERF_EVENTS
1673 	if (tp->flags & TP_FLAG_PROFILE)
1674 		kprobe_perf_func(kp, regs);
1675 #endif
1676 	return 0;	/* We don't tweek kernel, so just return 0 */
1677 }
1678 
1679 static __kprobes
1680 int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1681 {
1682 	struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1683 
1684 	if (tp->flags & TP_FLAG_TRACE)
1685 		kretprobe_trace_func(ri, regs);
1686 #ifdef CONFIG_PERF_EVENTS
1687 	if (tp->flags & TP_FLAG_PROFILE)
1688 		kretprobe_perf_func(ri, regs);
1689 #endif
1690 	return 0;	/* We don't tweek kernel, so just return 0 */
1691 }
1692 
1693 static struct trace_event_functions kretprobe_funcs = {
1694 	.trace		= print_kretprobe_event
1695 };
1696 
1697 static struct trace_event_functions kprobe_funcs = {
1698 	.trace		= print_kprobe_event
1699 };
1700 
1701 static int register_probe_event(struct trace_probe *tp)
1702 {
1703 	struct ftrace_event_call *call = &tp->call;
1704 	int ret;
1705 
1706 	/* Initialize ftrace_event_call */
1707 	INIT_LIST_HEAD(&call->class->fields);
1708 	if (probe_is_return(tp)) {
1709 		call->event.funcs = &kretprobe_funcs;
1710 		call->class->define_fields = kretprobe_event_define_fields;
1711 	} else {
1712 		call->event.funcs = &kprobe_funcs;
1713 		call->class->define_fields = kprobe_event_define_fields;
1714 	}
1715 	if (set_print_fmt(tp) < 0)
1716 		return -ENOMEM;
1717 	ret = register_ftrace_event(&call->event);
1718 	if (!ret) {
1719 		kfree(call->print_fmt);
1720 		return -ENODEV;
1721 	}
1722 	call->flags = 0;
1723 	call->class->reg = kprobe_register;
1724 	call->data = tp;
1725 	ret = trace_add_event_call(call);
1726 	if (ret) {
1727 		pr_info("Failed to register kprobe event: %s\n", call->name);
1728 		kfree(call->print_fmt);
1729 		unregister_ftrace_event(&call->event);
1730 	}
1731 	return ret;
1732 }
1733 
1734 static void unregister_probe_event(struct trace_probe *tp)
1735 {
1736 	/* tp->event is unregistered in trace_remove_event_call() */
1737 	trace_remove_event_call(&tp->call);
1738 	kfree(tp->call.print_fmt);
1739 }
1740 
1741 /* Make a debugfs interface for controling probe points */
1742 static __init int init_kprobe_trace(void)
1743 {
1744 	struct dentry *d_tracer;
1745 	struct dentry *entry;
1746 
1747 	d_tracer = tracing_init_dentry();
1748 	if (!d_tracer)
1749 		return 0;
1750 
1751 	entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1752 				    NULL, &kprobe_events_ops);
1753 
1754 	/* Event list interface */
1755 	if (!entry)
1756 		pr_warning("Could not create debugfs "
1757 			   "'kprobe_events' entry\n");
1758 
1759 	/* Profile interface */
1760 	entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1761 				    NULL, &kprobe_profile_ops);
1762 
1763 	if (!entry)
1764 		pr_warning("Could not create debugfs "
1765 			   "'kprobe_profile' entry\n");
1766 	return 0;
1767 }
1768 fs_initcall(init_kprobe_trace);
1769 
1770 
1771 #ifdef CONFIG_FTRACE_STARTUP_TEST
1772 
1773 static int kprobe_trace_selftest_target(int a1, int a2, int a3,
1774 					int a4, int a5, int a6)
1775 {
1776 	return a1 + a2 + a3 + a4 + a5 + a6;
1777 }
1778 
1779 static __init int kprobe_trace_self_tests_init(void)
1780 {
1781 	int ret, warn = 0;
1782 	int (*target)(int, int, int, int, int, int);
1783 	struct trace_probe *tp;
1784 
1785 	target = kprobe_trace_selftest_target;
1786 
1787 	pr_info("Testing kprobe tracing: ");
1788 
1789 	ret = command_trace_probe("p:testprobe kprobe_trace_selftest_target "
1790 				  "$stack $stack0 +0($stack)");
1791 	if (WARN_ON_ONCE(ret)) {
1792 		pr_warning("error on probing function entry.\n");
1793 		warn++;
1794 	} else {
1795 		/* Enable trace point */
1796 		tp = find_probe_event("testprobe", KPROBE_EVENT_SYSTEM);
1797 		if (WARN_ON_ONCE(tp == NULL)) {
1798 			pr_warning("error on getting new probe.\n");
1799 			warn++;
1800 		} else
1801 			probe_event_enable(&tp->call);
1802 	}
1803 
1804 	ret = command_trace_probe("r:testprobe2 kprobe_trace_selftest_target "
1805 				  "$retval");
1806 	if (WARN_ON_ONCE(ret)) {
1807 		pr_warning("error on probing function return.\n");
1808 		warn++;
1809 	} else {
1810 		/* Enable trace point */
1811 		tp = find_probe_event("testprobe2", KPROBE_EVENT_SYSTEM);
1812 		if (WARN_ON_ONCE(tp == NULL)) {
1813 			pr_warning("error on getting new probe.\n");
1814 			warn++;
1815 		} else
1816 			probe_event_enable(&tp->call);
1817 	}
1818 
1819 	if (warn)
1820 		goto end;
1821 
1822 	ret = target(1, 2, 3, 4, 5, 6);
1823 
1824 	ret = command_trace_probe("-:testprobe");
1825 	if (WARN_ON_ONCE(ret)) {
1826 		pr_warning("error on deleting a probe.\n");
1827 		warn++;
1828 	}
1829 
1830 	ret = command_trace_probe("-:testprobe2");
1831 	if (WARN_ON_ONCE(ret)) {
1832 		pr_warning("error on deleting a probe.\n");
1833 		warn++;
1834 	}
1835 
1836 end:
1837 	cleanup_all_probes();
1838 	if (warn)
1839 		pr_cont("NG: Some tests are failed. Please check them.\n");
1840 	else
1841 		pr_cont("OK\n");
1842 	return 0;
1843 }
1844 
1845 late_initcall(kprobe_trace_self_tests_init);
1846 
1847 #endif
1848