Loading...
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
50const 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 */
62typedef 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) \
69static __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} \
75static const char PRINT_TYPE_FMT_NAME(type)[] = fmt;
76
77DEFINE_BASIC_PRINT_TYPE_FUNC(u8, "%x", unsigned int)
78DEFINE_BASIC_PRINT_TYPE_FUNC(u16, "%x", unsigned int)
79DEFINE_BASIC_PRINT_TYPE_FUNC(u32, "%lx", unsigned long)
80DEFINE_BASIC_PRINT_TYPE_FUNC(u64, "%llx", unsigned long long)
81DEFINE_BASIC_PRINT_TYPE_FUNC(s8, "%d", int)
82DEFINE_BASIC_PRINT_TYPE_FUNC(s16, "%d", int)
83DEFINE_BASIC_PRINT_TYPE_FUNC(s32, "%ld", long)
84DEFINE_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
92static inline void *get_rloc_data(u32 *dl)
93{
94 return (u8 *)dl + get_rloc_offs(*dl);
95}
96
97/* For data_loc conversion */
98static 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 */
111typedef u32 string;
112typedef u32 string_size;
113
114/* Print type function for string type */
115static __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}
127static const char PRINT_TYPE_FMT_NAME(string)[] = "\\\"%s\\\"";
128
129/* Data fetch function type */
130typedef void (*fetch_func_t)(struct pt_regs *, void *, void *);
131
132struct fetch_param {
133 fetch_func_t fn;
134 void *data;
135};
136
137static __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) \
149DEFINE_FETCH_##method(u8) \
150DEFINE_FETCH_##method(u16) \
151DEFINE_FETCH_##method(u32) \
152DEFINE_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) \
165static __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}
171DEFINE_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) \
177static __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}
183DEFINE_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) \
189static __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}
194DEFINE_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) \
200static __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}
209DEFINE_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 */
214static __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 */
245static __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 */
268struct symbol_cache {
269 char *symbol;
270 long offset;
271 unsigned long addr;
272};
273
274static 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
282static void free_symbol_cache(struct symbol_cache *sc)
283{
284 kfree(sc->symbol);
285 kfree(sc);
286}
287
288static 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) \
310static __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}
319DEFINE_BASIC_FETCH_FUNCS(symbol)
320DEFINE_FETCH_symbol(string)
321DEFINE_FETCH_symbol(string_size)
322
323/* Dereference memory access function */
324struct deref_fetch_param {
325 struct fetch_param orig;
326 long offset;
327};
328
329#define DEFINE_FETCH_deref(type) \
330static __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}
342DEFINE_BASIC_FETCH_FUNCS(deref)
343DEFINE_FETCH_deref(string)
344DEFINE_FETCH_deref(string_size)
345
346static __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
354static __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 */
364struct 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) \
371static __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}
383DEFINE_BASIC_FETCH_FUNCS(bitfield)
384#define fetch_bitfield_string NULL
385#define fetch_bitfield_string_size NULL
386
387static __kprobes void
388update_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
400static __kprobes void
401free_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 */
421enum {
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 = { \
443ASSIGN_FETCH_FUNC(reg, ftype), \
444ASSIGN_FETCH_FUNC(stack, ftype), \
445ASSIGN_FETCH_FUNC(retval, ftype), \
446ASSIGN_FETCH_FUNC(memory, ftype), \
447ASSIGN_FETCH_FUNC(symbol, ftype), \
448ASSIGN_FETCH_FUNC(deref, ftype), \
449ASSIGN_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 */
460static 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
486static 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];
519fail:
520 return NULL;
521}
522
523/* Special function : only accept unsigned long */
524static __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
530static 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
549struct 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
563struct 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
581static __kprobes int trace_probe_is_return(struct trace_probe *tp)
582{
583 return tp->rp.handler != NULL;
584}
585
586static __kprobes const char *trace_probe_symbol(struct trace_probe *tp)
587{
588 return tp->symbol ? tp->symbol : "unknown";
589}
590
591static __kprobes unsigned long trace_probe_offset(struct trace_probe *tp)
592{
593 return tp->rp.kp.offset;
594}
595
596static __kprobes bool trace_probe_is_enabled(struct trace_probe *tp)
597{
598 return !!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE));
599}
600
601static __kprobes bool trace_probe_is_registered(struct trace_probe *tp)
602{
603 return !!(tp->flags & TP_FLAG_REGISTERED);
604}
605
606static __kprobes bool trace_probe_has_gone(struct trace_probe *tp)
607{
608 return !!(kprobe_gone(&tp->rp.kp));
609}
610
611static __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
619static __kprobes bool trace_probe_is_on_module(struct trace_probe *tp)
620{
621 return !!strchr(trace_probe_symbol(tp), ':');
622}
623
624static int register_probe_event(struct trace_probe *tp);
625static void unregister_probe_event(struct trace_probe *tp);
626
627static DEFINE_MUTEX(probe_lock);
628static LIST_HEAD(probe_list);
629
630static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
631static int kretprobe_dispatcher(struct kretprobe_instance *ri,
632 struct pt_regs *regs);
633
634/* Check the name is good for event/group/fields */
635static 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 */
649static 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;
698error:
699 kfree(tp->call.name);
700 kfree(tp->symbol);
701 kfree(tp);
702 return ERR_PTR(ret);
703}
704
705static 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
715static 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
727static 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
740static 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 */
753static 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 */
770static 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 */
782static 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 */
824static 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 */
839static void unregister_trace_probe(struct trace_probe *tp)
840{
841 __unregister_trace_probe(tp);
842 list_del(&tp->list);
843 unregister_probe_event(tp);
844}
845
846/* Register a trace_probe and probe_event */
847static int register_trace_probe(struct trace_probe *tp)
848{
849 struct trace_probe *old_tp;
850 int ret;
851
852 mutex_lock(&probe_lock);
853
854 /* Delete old (same name) event if exist */
855 old_tp = find_trace_probe(tp->call.name, tp->call.class->system);
856 if (old_tp) {
857 unregister_trace_probe(old_tp);
858 free_trace_probe(old_tp);
859 }
860
861 /* Register new event */
862 ret = register_probe_event(tp);
863 if (ret) {
864 pr_warning("Failed to register probe event(%d)\n", ret);
865 goto end;
866 }
867
868 /* Register k*probe */
869 ret = __register_trace_probe(tp);
870 if (ret < 0)
871 unregister_probe_event(tp);
872 else
873 list_add_tail(&tp->list, &probe_list);
874
875end:
876 mutex_unlock(&probe_lock);
877 return ret;
878}
879
880/* Module notifier call back, checking event on the module */
881static int trace_probe_module_callback(struct notifier_block *nb,
882 unsigned long val, void *data)
883{
884 struct module *mod = data;
885 struct trace_probe *tp;
886 int ret;
887
888 if (val != MODULE_STATE_COMING)
889 return NOTIFY_DONE;
890
891 /* Update probes on coming module */
892 mutex_lock(&probe_lock);
893 list_for_each_entry(tp, &probe_list, list) {
894 if (trace_probe_within_module(tp, mod)) {
895 __unregister_trace_probe(tp);
896 ret = __register_trace_probe(tp);
897 if (ret)
898 pr_warning("Failed to re-register probe %s on"
899 "%s: %d\n",
900 tp->call.name, mod->name, ret);
901 }
902 }
903 mutex_unlock(&probe_lock);
904
905 return NOTIFY_DONE;
906}
907
908static struct notifier_block trace_probe_module_nb = {
909 .notifier_call = trace_probe_module_callback,
910 .priority = 1 /* Invoked after kprobe module callback */
911};
912
913/* Split symbol and offset. */
914static int split_symbol_offset(char *symbol, unsigned long *offset)
915{
916 char *tmp;
917 int ret;
918
919 if (!offset)
920 return -EINVAL;
921
922 tmp = strchr(symbol, '+');
923 if (tmp) {
924 /* skip sign because strict_strtol doesn't accept '+' */
925 ret = strict_strtoul(tmp + 1, 0, offset);
926 if (ret)
927 return ret;
928 *tmp = '\0';
929 } else
930 *offset = 0;
931 return 0;
932}
933
934#define PARAM_MAX_ARGS 16
935#define PARAM_MAX_STACK (THREAD_SIZE / sizeof(unsigned long))
936
937static int parse_probe_vars(char *arg, const struct fetch_type *t,
938 struct fetch_param *f, int is_return)
939{
940 int ret = 0;
941 unsigned long param;
942
943 if (strcmp(arg, "retval") == 0) {
944 if (is_return)
945 f->fn = t->fetch[FETCH_MTD_retval];
946 else
947 ret = -EINVAL;
948 } else if (strncmp(arg, "stack", 5) == 0) {
949 if (arg[5] == '\0') {
950 if (strcmp(t->name, DEFAULT_FETCH_TYPE_STR) == 0)
951 f->fn = fetch_stack_address;
952 else
953 ret = -EINVAL;
954 } else if (isdigit(arg[5])) {
955 ret = strict_strtoul(arg + 5, 10, ¶m);
956 if (ret || param > PARAM_MAX_STACK)
957 ret = -EINVAL;
958 else {
959 f->fn = t->fetch[FETCH_MTD_stack];
960 f->data = (void *)param;
961 }
962 } else
963 ret = -EINVAL;
964 } else
965 ret = -EINVAL;
966 return ret;
967}
968
969/* Recursive argument parser */
970static int __parse_probe_arg(char *arg, const struct fetch_type *t,
971 struct fetch_param *f, int is_return)
972{
973 int ret = 0;
974 unsigned long param;
975 long offset;
976 char *tmp;
977
978 switch (arg[0]) {
979 case '$':
980 ret = parse_probe_vars(arg + 1, t, f, is_return);
981 break;
982 case '%': /* named register */
983 ret = regs_query_register_offset(arg + 1);
984 if (ret >= 0) {
985 f->fn = t->fetch[FETCH_MTD_reg];
986 f->data = (void *)(unsigned long)ret;
987 ret = 0;
988 }
989 break;
990 case '@': /* memory or symbol */
991 if (isdigit(arg[1])) {
992 ret = strict_strtoul(arg + 1, 0, ¶m);
993 if (ret)
994 break;
995 f->fn = t->fetch[FETCH_MTD_memory];
996 f->data = (void *)param;
997 } else {
998 ret = split_symbol_offset(arg + 1, &offset);
999 if (ret)
1000 break;
1001 f->data = alloc_symbol_cache(arg + 1, offset);
1002 if (f->data)
1003 f->fn = t->fetch[FETCH_MTD_symbol];
1004 }
1005 break;
1006 case '+': /* deref memory */
1007 arg++; /* Skip '+', because strict_strtol() rejects it. */
1008 case '-':
1009 tmp = strchr(arg, '(');
1010 if (!tmp)
1011 break;
1012 *tmp = '\0';
1013 ret = strict_strtol(arg, 0, &offset);
1014 if (ret)
1015 break;
1016 arg = tmp + 1;
1017 tmp = strrchr(arg, ')');
1018 if (tmp) {
1019 struct deref_fetch_param *dprm;
1020 const struct fetch_type *t2 = find_fetch_type(NULL);
1021 *tmp = '\0';
1022 dprm = kzalloc(sizeof(struct deref_fetch_param),
1023 GFP_KERNEL);
1024 if (!dprm)
1025 return -ENOMEM;
1026 dprm->offset = offset;
1027 ret = __parse_probe_arg(arg, t2, &dprm->orig,
1028 is_return);
1029 if (ret)
1030 kfree(dprm);
1031 else {
1032 f->fn = t->fetch[FETCH_MTD_deref];
1033 f->data = (void *)dprm;
1034 }
1035 }
1036 break;
1037 }
1038 if (!ret && !f->fn) { /* Parsed, but do not find fetch method */
1039 pr_info("%s type has no corresponding fetch method.\n",
1040 t->name);
1041 ret = -EINVAL;
1042 }
1043 return ret;
1044}
1045
1046#define BYTES_TO_BITS(nb) ((BITS_PER_LONG * (nb)) / sizeof(long))
1047
1048/* Bitfield type needs to be parsed into a fetch function */
1049static int __parse_bitfield_probe_arg(const char *bf,
1050 const struct fetch_type *t,
1051 struct fetch_param *f)
1052{
1053 struct bitfield_fetch_param *bprm;
1054 unsigned long bw, bo;
1055 char *tail;
1056
1057 if (*bf != 'b')
1058 return 0;
1059
1060 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1061 if (!bprm)
1062 return -ENOMEM;
1063 bprm->orig = *f;
1064 f->fn = t->fetch[FETCH_MTD_bitfield];
1065 f->data = (void *)bprm;
1066
1067 bw = simple_strtoul(bf + 1, &tail, 0); /* Use simple one */
1068 if (bw == 0 || *tail != '@')
1069 return -EINVAL;
1070
1071 bf = tail + 1;
1072 bo = simple_strtoul(bf, &tail, 0);
1073 if (tail == bf || *tail != '/')
1074 return -EINVAL;
1075
1076 bprm->hi_shift = BYTES_TO_BITS(t->size) - (bw + bo);
1077 bprm->low_shift = bprm->hi_shift + bo;
1078 return (BYTES_TO_BITS(t->size) < (bw + bo)) ? -EINVAL : 0;
1079}
1080
1081/* String length checking wrapper */
1082static int parse_probe_arg(char *arg, struct trace_probe *tp,
1083 struct probe_arg *parg, int is_return)
1084{
1085 const char *t;
1086 int ret;
1087
1088 if (strlen(arg) > MAX_ARGSTR_LEN) {
1089 pr_info("Argument is too long.: %s\n", arg);
1090 return -ENOSPC;
1091 }
1092 parg->comm = kstrdup(arg, GFP_KERNEL);
1093 if (!parg->comm) {
1094 pr_info("Failed to allocate memory for command '%s'.\n", arg);
1095 return -ENOMEM;
1096 }
1097 t = strchr(parg->comm, ':');
1098 if (t) {
1099 arg[t - parg->comm] = '\0';
1100 t++;
1101 }
1102 parg->type = find_fetch_type(t);
1103 if (!parg->type) {
1104 pr_info("Unsupported type: %s\n", t);
1105 return -EINVAL;
1106 }
1107 parg->offset = tp->size;
1108 tp->size += parg->type->size;
1109 ret = __parse_probe_arg(arg, parg->type, &parg->fetch, is_return);
1110 if (ret >= 0 && t != NULL)
1111 ret = __parse_bitfield_probe_arg(t, parg->type, &parg->fetch);
1112 if (ret >= 0) {
1113 parg->fetch_size.fn = get_fetch_size_function(parg->type,
1114 parg->fetch.fn);
1115 parg->fetch_size.data = parg->fetch.data;
1116 }
1117 return ret;
1118}
1119
1120/* Return 1 if name is reserved or already used by another argument */
1121static int conflict_field_name(const char *name,
1122 struct probe_arg *args, int narg)
1123{
1124 int i;
1125 for (i = 0; i < ARRAY_SIZE(reserved_field_names); i++)
1126 if (strcmp(reserved_field_names[i], name) == 0)
1127 return 1;
1128 for (i = 0; i < narg; i++)
1129 if (strcmp(args[i].name, name) == 0)
1130 return 1;
1131 return 0;
1132}
1133
1134static int create_trace_probe(int argc, char **argv)
1135{
1136 /*
1137 * Argument syntax:
1138 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
1139 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
1140 * Fetch args:
1141 * $retval : fetch return value
1142 * $stack : fetch stack address
1143 * $stackN : fetch Nth of stack (N:0-)
1144 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
1145 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
1146 * %REG : fetch register REG
1147 * Dereferencing memory fetch:
1148 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
1149 * Alias name of args:
1150 * NAME=FETCHARG : set NAME as alias of FETCHARG.
1151 * Type of args:
1152 * FETCHARG:TYPE : use TYPE instead of unsigned long.
1153 */
1154 struct trace_probe *tp;
1155 int i, ret = 0;
1156 int is_return = 0, is_delete = 0;
1157 char *symbol = NULL, *event = NULL, *group = NULL;
1158 char *arg;
1159 unsigned long offset = 0;
1160 void *addr = NULL;
1161 char buf[MAX_EVENT_NAME_LEN];
1162
1163 /* argc must be >= 1 */
1164 if (argv[0][0] == 'p')
1165 is_return = 0;
1166 else if (argv[0][0] == 'r')
1167 is_return = 1;
1168 else if (argv[0][0] == '-')
1169 is_delete = 1;
1170 else {
1171 pr_info("Probe definition must be started with 'p', 'r' or"
1172 " '-'.\n");
1173 return -EINVAL;
1174 }
1175
1176 if (argv[0][1] == ':') {
1177 event = &argv[0][2];
1178 if (strchr(event, '/')) {
1179 group = event;
1180 event = strchr(group, '/') + 1;
1181 event[-1] = '\0';
1182 if (strlen(group) == 0) {
1183 pr_info("Group name is not specified\n");
1184 return -EINVAL;
1185 }
1186 }
1187 if (strlen(event) == 0) {
1188 pr_info("Event name is not specified\n");
1189 return -EINVAL;
1190 }
1191 }
1192 if (!group)
1193 group = KPROBE_EVENT_SYSTEM;
1194
1195 if (is_delete) {
1196 if (!event) {
1197 pr_info("Delete command needs an event name.\n");
1198 return -EINVAL;
1199 }
1200 mutex_lock(&probe_lock);
1201 tp = find_trace_probe(event, group);
1202 if (!tp) {
1203 mutex_unlock(&probe_lock);
1204 pr_info("Event %s/%s doesn't exist.\n", group, event);
1205 return -ENOENT;
1206 }
1207 /* delete an event */
1208 unregister_trace_probe(tp);
1209 free_trace_probe(tp);
1210 mutex_unlock(&probe_lock);
1211 return 0;
1212 }
1213
1214 if (argc < 2) {
1215 pr_info("Probe point is not specified.\n");
1216 return -EINVAL;
1217 }
1218 if (isdigit(argv[1][0])) {
1219 if (is_return) {
1220 pr_info("Return probe point must be a symbol.\n");
1221 return -EINVAL;
1222 }
1223 /* an address specified */
1224 ret = strict_strtoul(&argv[1][0], 0, (unsigned long *)&addr);
1225 if (ret) {
1226 pr_info("Failed to parse address.\n");
1227 return ret;
1228 }
1229 } else {
1230 /* a symbol specified */
1231 symbol = argv[1];
1232 /* TODO: support .init module functions */
1233 ret = split_symbol_offset(symbol, &offset);
1234 if (ret) {
1235 pr_info("Failed to parse symbol.\n");
1236 return ret;
1237 }
1238 if (offset && is_return) {
1239 pr_info("Return probe must be used without offset.\n");
1240 return -EINVAL;
1241 }
1242 }
1243 argc -= 2; argv += 2;
1244
1245 /* setup a probe */
1246 if (!event) {
1247 /* Make a new event name */
1248 if (symbol)
1249 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
1250 is_return ? 'r' : 'p', symbol, offset);
1251 else
1252 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
1253 is_return ? 'r' : 'p', addr);
1254 event = buf;
1255 }
1256 tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
1257 is_return);
1258 if (IS_ERR(tp)) {
1259 pr_info("Failed to allocate trace_probe.(%d)\n",
1260 (int)PTR_ERR(tp));
1261 return PTR_ERR(tp);
1262 }
1263
1264 /* parse arguments */
1265 ret = 0;
1266 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
1267 /* Increment count for freeing args in error case */
1268 tp->nr_args++;
1269
1270 /* Parse argument name */
1271 arg = strchr(argv[i], '=');
1272 if (arg) {
1273 *arg++ = '\0';
1274 tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
1275 } else {
1276 arg = argv[i];
1277 /* If argument name is omitted, set "argN" */
1278 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
1279 tp->args[i].name = kstrdup(buf, GFP_KERNEL);
1280 }
1281
1282 if (!tp->args[i].name) {
1283 pr_info("Failed to allocate argument[%d] name.\n", i);
1284 ret = -ENOMEM;
1285 goto error;
1286 }
1287
1288 if (!is_good_name(tp->args[i].name)) {
1289 pr_info("Invalid argument[%d] name: %s\n",
1290 i, tp->args[i].name);
1291 ret = -EINVAL;
1292 goto error;
1293 }
1294
1295 if (conflict_field_name(tp->args[i].name, tp->args, i)) {
1296 pr_info("Argument[%d] name '%s' conflicts with "
1297 "another field.\n", i, argv[i]);
1298 ret = -EINVAL;
1299 goto error;
1300 }
1301
1302 /* Parse fetch argument */
1303 ret = parse_probe_arg(arg, tp, &tp->args[i], is_return);
1304 if (ret) {
1305 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
1306 goto error;
1307 }
1308 }
1309
1310 ret = register_trace_probe(tp);
1311 if (ret)
1312 goto error;
1313 return 0;
1314
1315error:
1316 free_trace_probe(tp);
1317 return ret;
1318}
1319
1320static void release_all_trace_probes(void)
1321{
1322 struct trace_probe *tp;
1323
1324 mutex_lock(&probe_lock);
1325 /* TODO: Use batch unregistration */
1326 while (!list_empty(&probe_list)) {
1327 tp = list_entry(probe_list.next, struct trace_probe, list);
1328 unregister_trace_probe(tp);
1329 free_trace_probe(tp);
1330 }
1331 mutex_unlock(&probe_lock);
1332}
1333
1334/* Probes listing interfaces */
1335static void *probes_seq_start(struct seq_file *m, loff_t *pos)
1336{
1337 mutex_lock(&probe_lock);
1338 return seq_list_start(&probe_list, *pos);
1339}
1340
1341static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
1342{
1343 return seq_list_next(v, &probe_list, pos);
1344}
1345
1346static void probes_seq_stop(struct seq_file *m, void *v)
1347{
1348 mutex_unlock(&probe_lock);
1349}
1350
1351static int probes_seq_show(struct seq_file *m, void *v)
1352{
1353 struct trace_probe *tp = v;
1354 int i;
1355
1356 seq_printf(m, "%c", trace_probe_is_return(tp) ? 'r' : 'p');
1357 seq_printf(m, ":%s/%s", tp->call.class->system, tp->call.name);
1358
1359 if (!tp->symbol)
1360 seq_printf(m, " 0x%p", tp->rp.kp.addr);
1361 else if (tp->rp.kp.offset)
1362 seq_printf(m, " %s+%u", trace_probe_symbol(tp),
1363 tp->rp.kp.offset);
1364 else
1365 seq_printf(m, " %s", trace_probe_symbol(tp));
1366
1367 for (i = 0; i < tp->nr_args; i++)
1368 seq_printf(m, " %s=%s", tp->args[i].name, tp->args[i].comm);
1369 seq_printf(m, "\n");
1370
1371 return 0;
1372}
1373
1374static const struct seq_operations probes_seq_op = {
1375 .start = probes_seq_start,
1376 .next = probes_seq_next,
1377 .stop = probes_seq_stop,
1378 .show = probes_seq_show
1379};
1380
1381static int probes_open(struct inode *inode, struct file *file)
1382{
1383 if ((file->f_mode & FMODE_WRITE) &&
1384 (file->f_flags & O_TRUNC))
1385 release_all_trace_probes();
1386
1387 return seq_open(file, &probes_seq_op);
1388}
1389
1390static int command_trace_probe(const char *buf)
1391{
1392 char **argv;
1393 int argc = 0, ret = 0;
1394
1395 argv = argv_split(GFP_KERNEL, buf, &argc);
1396 if (!argv)
1397 return -ENOMEM;
1398
1399 if (argc)
1400 ret = create_trace_probe(argc, argv);
1401
1402 argv_free(argv);
1403 return ret;
1404}
1405
1406#define WRITE_BUFSIZE 4096
1407
1408static ssize_t probes_write(struct file *file, const char __user *buffer,
1409 size_t count, loff_t *ppos)
1410{
1411 char *kbuf, *tmp;
1412 int ret;
1413 size_t done;
1414 size_t size;
1415
1416 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
1417 if (!kbuf)
1418 return -ENOMEM;
1419
1420 ret = done = 0;
1421 while (done < count) {
1422 size = count - done;
1423 if (size >= WRITE_BUFSIZE)
1424 size = WRITE_BUFSIZE - 1;
1425 if (copy_from_user(kbuf, buffer + done, size)) {
1426 ret = -EFAULT;
1427 goto out;
1428 }
1429 kbuf[size] = '\0';
1430 tmp = strchr(kbuf, '\n');
1431 if (tmp) {
1432 *tmp = '\0';
1433 size = tmp - kbuf + 1;
1434 } else if (done + size < count) {
1435 pr_warning("Line length is too long: "
1436 "Should be less than %d.", WRITE_BUFSIZE);
1437 ret = -EINVAL;
1438 goto out;
1439 }
1440 done += size;
1441 /* Remove comments */
1442 tmp = strchr(kbuf, '#');
1443 if (tmp)
1444 *tmp = '\0';
1445
1446 ret = command_trace_probe(kbuf);
1447 if (ret)
1448 goto out;
1449 }
1450 ret = done;
1451out:
1452 kfree(kbuf);
1453 return ret;
1454}
1455
1456static const struct file_operations kprobe_events_ops = {
1457 .owner = THIS_MODULE,
1458 .open = probes_open,
1459 .read = seq_read,
1460 .llseek = seq_lseek,
1461 .release = seq_release,
1462 .write = probes_write,
1463};
1464
1465/* Probes profiling interfaces */
1466static int probes_profile_seq_show(struct seq_file *m, void *v)
1467{
1468 struct trace_probe *tp = v;
1469
1470 seq_printf(m, " %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
1471 tp->rp.kp.nmissed);
1472
1473 return 0;
1474}
1475
1476static const struct seq_operations profile_seq_op = {
1477 .start = probes_seq_start,
1478 .next = probes_seq_next,
1479 .stop = probes_seq_stop,
1480 .show = probes_profile_seq_show
1481};
1482
1483static int profile_open(struct inode *inode, struct file *file)
1484{
1485 return seq_open(file, &profile_seq_op);
1486}
1487
1488static const struct file_operations kprobe_profile_ops = {
1489 .owner = THIS_MODULE,
1490 .open = profile_open,
1491 .read = seq_read,
1492 .llseek = seq_lseek,
1493 .release = seq_release,
1494};
1495
1496/* Sum up total data length for dynamic arraies (strings) */
1497static __kprobes int __get_data_size(struct trace_probe *tp,
1498 struct pt_regs *regs)
1499{
1500 int i, ret = 0;
1501 u32 len;
1502
1503 for (i = 0; i < tp->nr_args; i++)
1504 if (unlikely(tp->args[i].fetch_size.fn)) {
1505 call_fetch(&tp->args[i].fetch_size, regs, &len);
1506 ret += len;
1507 }
1508
1509 return ret;
1510}
1511
1512/* Store the value of each argument */
1513static __kprobes void store_trace_args(int ent_size, struct trace_probe *tp,
1514 struct pt_regs *regs,
1515 u8 *data, int maxlen)
1516{
1517 int i;
1518 u32 end = tp->size;
1519 u32 *dl; /* Data (relative) location */
1520
1521 for (i = 0; i < tp->nr_args; i++) {
1522 if (unlikely(tp->args[i].fetch_size.fn)) {
1523 /*
1524 * First, we set the relative location and
1525 * maximum data length to *dl
1526 */
1527 dl = (u32 *)(data + tp->args[i].offset);
1528 *dl = make_data_rloc(maxlen, end - tp->args[i].offset);
1529 /* Then try to fetch string or dynamic array data */
1530 call_fetch(&tp->args[i].fetch, regs, dl);
1531 /* Reduce maximum length */
1532 end += get_rloc_len(*dl);
1533 maxlen -= get_rloc_len(*dl);
1534 /* Trick here, convert data_rloc to data_loc */
1535 *dl = convert_rloc_to_loc(*dl,
1536 ent_size + tp->args[i].offset);
1537 } else
1538 /* Just fetching data normally */
1539 call_fetch(&tp->args[i].fetch, regs,
1540 data + tp->args[i].offset);
1541 }
1542}
1543
1544/* Kprobe handler */
1545static __kprobes void kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
1546{
1547 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1548 struct kprobe_trace_entry_head *entry;
1549 struct ring_buffer_event *event;
1550 struct ring_buffer *buffer;
1551 int size, dsize, pc;
1552 unsigned long irq_flags;
1553 struct ftrace_event_call *call = &tp->call;
1554
1555 tp->nhit++;
1556
1557 local_save_flags(irq_flags);
1558 pc = preempt_count();
1559
1560 dsize = __get_data_size(tp, regs);
1561 size = sizeof(*entry) + tp->size + dsize;
1562
1563 event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
1564 size, irq_flags, pc);
1565 if (!event)
1566 return;
1567
1568 entry = ring_buffer_event_data(event);
1569 entry->ip = (unsigned long)kp->addr;
1570 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1571
1572 if (!filter_current_check_discard(buffer, call, entry, event))
1573 trace_nowake_buffer_unlock_commit_regs(buffer, event,
1574 irq_flags, pc, regs);
1575}
1576
1577/* Kretprobe handler */
1578static __kprobes void kretprobe_trace_func(struct kretprobe_instance *ri,
1579 struct pt_regs *regs)
1580{
1581 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1582 struct kretprobe_trace_entry_head *entry;
1583 struct ring_buffer_event *event;
1584 struct ring_buffer *buffer;
1585 int size, pc, dsize;
1586 unsigned long irq_flags;
1587 struct ftrace_event_call *call = &tp->call;
1588
1589 local_save_flags(irq_flags);
1590 pc = preempt_count();
1591
1592 dsize = __get_data_size(tp, regs);
1593 size = sizeof(*entry) + tp->size + dsize;
1594
1595 event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
1596 size, irq_flags, pc);
1597 if (!event)
1598 return;
1599
1600 entry = ring_buffer_event_data(event);
1601 entry->func = (unsigned long)tp->rp.kp.addr;
1602 entry->ret_ip = (unsigned long)ri->ret_addr;
1603 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1604
1605 if (!filter_current_check_discard(buffer, call, entry, event))
1606 trace_nowake_buffer_unlock_commit_regs(buffer, event,
1607 irq_flags, pc, regs);
1608}
1609
1610/* Event entry printers */
1611enum print_line_t
1612print_kprobe_event(struct trace_iterator *iter, int flags,
1613 struct trace_event *event)
1614{
1615 struct kprobe_trace_entry_head *field;
1616 struct trace_seq *s = &iter->seq;
1617 struct trace_probe *tp;
1618 u8 *data;
1619 int i;
1620
1621 field = (struct kprobe_trace_entry_head *)iter->ent;
1622 tp = container_of(event, struct trace_probe, call.event);
1623
1624 if (!trace_seq_printf(s, "%s: (", tp->call.name))
1625 goto partial;
1626
1627 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1628 goto partial;
1629
1630 if (!trace_seq_puts(s, ")"))
1631 goto partial;
1632
1633 data = (u8 *)&field[1];
1634 for (i = 0; i < tp->nr_args; i++)
1635 if (!tp->args[i].type->print(s, tp->args[i].name,
1636 data + tp->args[i].offset, field))
1637 goto partial;
1638
1639 if (!trace_seq_puts(s, "\n"))
1640 goto partial;
1641
1642 return TRACE_TYPE_HANDLED;
1643partial:
1644 return TRACE_TYPE_PARTIAL_LINE;
1645}
1646
1647enum print_line_t
1648print_kretprobe_event(struct trace_iterator *iter, int flags,
1649 struct trace_event *event)
1650{
1651 struct kretprobe_trace_entry_head *field;
1652 struct trace_seq *s = &iter->seq;
1653 struct trace_probe *tp;
1654 u8 *data;
1655 int i;
1656
1657 field = (struct kretprobe_trace_entry_head *)iter->ent;
1658 tp = container_of(event, struct trace_probe, call.event);
1659
1660 if (!trace_seq_printf(s, "%s: (", tp->call.name))
1661 goto partial;
1662
1663 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1664 goto partial;
1665
1666 if (!trace_seq_puts(s, " <- "))
1667 goto partial;
1668
1669 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1670 goto partial;
1671
1672 if (!trace_seq_puts(s, ")"))
1673 goto partial;
1674
1675 data = (u8 *)&field[1];
1676 for (i = 0; i < tp->nr_args; i++)
1677 if (!tp->args[i].type->print(s, tp->args[i].name,
1678 data + tp->args[i].offset, field))
1679 goto partial;
1680
1681 if (!trace_seq_puts(s, "\n"))
1682 goto partial;
1683
1684 return TRACE_TYPE_HANDLED;
1685partial:
1686 return TRACE_TYPE_PARTIAL_LINE;
1687}
1688
1689#undef DEFINE_FIELD
1690#define DEFINE_FIELD(type, item, name, is_signed) \
1691 do { \
1692 ret = trace_define_field(event_call, #type, name, \
1693 offsetof(typeof(field), item), \
1694 sizeof(field.item), is_signed, \
1695 FILTER_OTHER); \
1696 if (ret) \
1697 return ret; \
1698 } while (0)
1699
1700static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
1701{
1702 int ret, i;
1703 struct kprobe_trace_entry_head field;
1704 struct trace_probe *tp = (struct trace_probe *)event_call->data;
1705
1706 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1707 /* Set argument names as fields */
1708 for (i = 0; i < tp->nr_args; i++) {
1709 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
1710 tp->args[i].name,
1711 sizeof(field) + tp->args[i].offset,
1712 tp->args[i].type->size,
1713 tp->args[i].type->is_signed,
1714 FILTER_OTHER);
1715 if (ret)
1716 return ret;
1717 }
1718 return 0;
1719}
1720
1721static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1722{
1723 int ret, i;
1724 struct kretprobe_trace_entry_head field;
1725 struct trace_probe *tp = (struct trace_probe *)event_call->data;
1726
1727 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1728 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1729 /* Set argument names as fields */
1730 for (i = 0; i < tp->nr_args; i++) {
1731 ret = trace_define_field(event_call, tp->args[i].type->fmttype,
1732 tp->args[i].name,
1733 sizeof(field) + tp->args[i].offset,
1734 tp->args[i].type->size,
1735 tp->args[i].type->is_signed,
1736 FILTER_OTHER);
1737 if (ret)
1738 return ret;
1739 }
1740 return 0;
1741}
1742
1743static int __set_print_fmt(struct trace_probe *tp, char *buf, int len)
1744{
1745 int i;
1746 int pos = 0;
1747
1748 const char *fmt, *arg;
1749
1750 if (!trace_probe_is_return(tp)) {
1751 fmt = "(%lx)";
1752 arg = "REC->" FIELD_STRING_IP;
1753 } else {
1754 fmt = "(%lx <- %lx)";
1755 arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
1756 }
1757
1758 /* When len=0, we just calculate the needed length */
1759#define LEN_OR_ZERO (len ? len - pos : 0)
1760
1761 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
1762
1763 for (i = 0; i < tp->nr_args; i++) {
1764 pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
1765 tp->args[i].name, tp->args[i].type->fmt);
1766 }
1767
1768 pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
1769
1770 for (i = 0; i < tp->nr_args; i++) {
1771 if (strcmp(tp->args[i].type->name, "string") == 0)
1772 pos += snprintf(buf + pos, LEN_OR_ZERO,
1773 ", __get_str(%s)",
1774 tp->args[i].name);
1775 else
1776 pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
1777 tp->args[i].name);
1778 }
1779
1780#undef LEN_OR_ZERO
1781
1782 /* return the length of print_fmt */
1783 return pos;
1784}
1785
1786static int set_print_fmt(struct trace_probe *tp)
1787{
1788 int len;
1789 char *print_fmt;
1790
1791 /* First: called with 0 length to calculate the needed length */
1792 len = __set_print_fmt(tp, NULL, 0);
1793 print_fmt = kmalloc(len + 1, GFP_KERNEL);
1794 if (!print_fmt)
1795 return -ENOMEM;
1796
1797 /* Second: actually write the @print_fmt */
1798 __set_print_fmt(tp, print_fmt, len + 1);
1799 tp->call.print_fmt = print_fmt;
1800
1801 return 0;
1802}
1803
1804#ifdef CONFIG_PERF_EVENTS
1805
1806/* Kprobe profile handler */
1807static __kprobes void kprobe_perf_func(struct kprobe *kp,
1808 struct pt_regs *regs)
1809{
1810 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1811 struct ftrace_event_call *call = &tp->call;
1812 struct kprobe_trace_entry_head *entry;
1813 struct hlist_head *head;
1814 int size, __size, dsize;
1815 int rctx;
1816
1817 dsize = __get_data_size(tp, regs);
1818 __size = sizeof(*entry) + tp->size + dsize;
1819 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1820 size -= sizeof(u32);
1821 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1822 "profile buffer not large enough"))
1823 return;
1824
1825 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1826 if (!entry)
1827 return;
1828
1829 entry->ip = (unsigned long)kp->addr;
1830 memset(&entry[1], 0, dsize);
1831 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1832
1833 head = this_cpu_ptr(call->perf_events);
1834 perf_trace_buf_submit(entry, size, rctx, entry->ip, 1, regs, head);
1835}
1836
1837/* Kretprobe profile handler */
1838static __kprobes void kretprobe_perf_func(struct kretprobe_instance *ri,
1839 struct pt_regs *regs)
1840{
1841 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1842 struct ftrace_event_call *call = &tp->call;
1843 struct kretprobe_trace_entry_head *entry;
1844 struct hlist_head *head;
1845 int size, __size, dsize;
1846 int rctx;
1847
1848 dsize = __get_data_size(tp, regs);
1849 __size = sizeof(*entry) + tp->size + dsize;
1850 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1851 size -= sizeof(u32);
1852 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1853 "profile buffer not large enough"))
1854 return;
1855
1856 entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1857 if (!entry)
1858 return;
1859
1860 entry->func = (unsigned long)tp->rp.kp.addr;
1861 entry->ret_ip = (unsigned long)ri->ret_addr;
1862 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1863
1864 head = this_cpu_ptr(call->perf_events);
1865 perf_trace_buf_submit(entry, size, rctx, entry->ret_ip, 1, regs, head);
1866}
1867#endif /* CONFIG_PERF_EVENTS */
1868
1869static __kprobes
1870int kprobe_register(struct ftrace_event_call *event, enum trace_reg type)
1871{
1872 struct trace_probe *tp = (struct trace_probe *)event->data;
1873
1874 switch (type) {
1875 case TRACE_REG_REGISTER:
1876 return enable_trace_probe(tp, TP_FLAG_TRACE);
1877 case TRACE_REG_UNREGISTER:
1878 disable_trace_probe(tp, TP_FLAG_TRACE);
1879 return 0;
1880
1881#ifdef CONFIG_PERF_EVENTS
1882 case TRACE_REG_PERF_REGISTER:
1883 return enable_trace_probe(tp, TP_FLAG_PROFILE);
1884 case TRACE_REG_PERF_UNREGISTER:
1885 disable_trace_probe(tp, TP_FLAG_PROFILE);
1886 return 0;
1887#endif
1888 }
1889 return 0;
1890}
1891
1892static __kprobes
1893int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1894{
1895 struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1896
1897 if (tp->flags & TP_FLAG_TRACE)
1898 kprobe_trace_func(kp, regs);
1899#ifdef CONFIG_PERF_EVENTS
1900 if (tp->flags & TP_FLAG_PROFILE)
1901 kprobe_perf_func(kp, regs);
1902#endif
1903 return 0; /* We don't tweek kernel, so just return 0 */
1904}
1905
1906static __kprobes
1907int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1908{
1909 struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1910
1911 if (tp->flags & TP_FLAG_TRACE)
1912 kretprobe_trace_func(ri, regs);
1913#ifdef CONFIG_PERF_EVENTS
1914 if (tp->flags & TP_FLAG_PROFILE)
1915 kretprobe_perf_func(ri, regs);
1916#endif
1917 return 0; /* We don't tweek kernel, so just return 0 */
1918}
1919
1920static struct trace_event_functions kretprobe_funcs = {
1921 .trace = print_kretprobe_event
1922};
1923
1924static struct trace_event_functions kprobe_funcs = {
1925 .trace = print_kprobe_event
1926};
1927
1928static int register_probe_event(struct trace_probe *tp)
1929{
1930 struct ftrace_event_call *call = &tp->call;
1931 int ret;
1932
1933 /* Initialize ftrace_event_call */
1934 INIT_LIST_HEAD(&call->class->fields);
1935 if (trace_probe_is_return(tp)) {
1936 call->event.funcs = &kretprobe_funcs;
1937 call->class->define_fields = kretprobe_event_define_fields;
1938 } else {
1939 call->event.funcs = &kprobe_funcs;
1940 call->class->define_fields = kprobe_event_define_fields;
1941 }
1942 if (set_print_fmt(tp) < 0)
1943 return -ENOMEM;
1944 ret = register_ftrace_event(&call->event);
1945 if (!ret) {
1946 kfree(call->print_fmt);
1947 return -ENODEV;
1948 }
1949 call->flags = 0;
1950 call->class->reg = kprobe_register;
1951 call->data = tp;
1952 ret = trace_add_event_call(call);
1953 if (ret) {
1954 pr_info("Failed to register kprobe event: %s\n", call->name);
1955 kfree(call->print_fmt);
1956 unregister_ftrace_event(&call->event);
1957 }
1958 return ret;
1959}
1960
1961static void unregister_probe_event(struct trace_probe *tp)
1962{
1963 /* tp->event is unregistered in trace_remove_event_call() */
1964 trace_remove_event_call(&tp->call);
1965 kfree(tp->call.print_fmt);
1966}
1967
1968/* Make a debugfs interface for controlling probe points */
1969static __init int init_kprobe_trace(void)
1970{
1971 struct dentry *d_tracer;
1972 struct dentry *entry;
1973
1974 if (register_module_notifier(&trace_probe_module_nb))
1975 return -EINVAL;
1976
1977 d_tracer = tracing_init_dentry();
1978 if (!d_tracer)
1979 return 0;
1980
1981 entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1982 NULL, &kprobe_events_ops);
1983
1984 /* Event list interface */
1985 if (!entry)
1986 pr_warning("Could not create debugfs "
1987 "'kprobe_events' entry\n");
1988
1989 /* Profile interface */
1990 entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1991 NULL, &kprobe_profile_ops);
1992
1993 if (!entry)
1994 pr_warning("Could not create debugfs "
1995 "'kprobe_profile' entry\n");
1996 return 0;
1997}
1998fs_initcall(init_kprobe_trace);
1999
2000
2001#ifdef CONFIG_FTRACE_STARTUP_TEST
2002
2003/*
2004 * The "__used" keeps gcc from removing the function symbol
2005 * from the kallsyms table.
2006 */
2007static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
2008 int a4, int a5, int a6)
2009{
2010 return a1 + a2 + a3 + a4 + a5 + a6;
2011}
2012
2013static __init int kprobe_trace_self_tests_init(void)
2014{
2015 int ret, warn = 0;
2016 int (*target)(int, int, int, int, int, int);
2017 struct trace_probe *tp;
2018
2019 target = kprobe_trace_selftest_target;
2020
2021 pr_info("Testing kprobe tracing: ");
2022
2023 ret = command_trace_probe("p:testprobe kprobe_trace_selftest_target "
2024 "$stack $stack0 +0($stack)");
2025 if (WARN_ON_ONCE(ret)) {
2026 pr_warning("error on probing function entry.\n");
2027 warn++;
2028 } else {
2029 /* Enable trace point */
2030 tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
2031 if (WARN_ON_ONCE(tp == NULL)) {
2032 pr_warning("error on getting new probe.\n");
2033 warn++;
2034 } else
2035 enable_trace_probe(tp, TP_FLAG_TRACE);
2036 }
2037
2038 ret = command_trace_probe("r:testprobe2 kprobe_trace_selftest_target "
2039 "$retval");
2040 if (WARN_ON_ONCE(ret)) {
2041 pr_warning("error on probing function return.\n");
2042 warn++;
2043 } else {
2044 /* Enable trace point */
2045 tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
2046 if (WARN_ON_ONCE(tp == NULL)) {
2047 pr_warning("error on getting new probe.\n");
2048 warn++;
2049 } else
2050 enable_trace_probe(tp, TP_FLAG_TRACE);
2051 }
2052
2053 if (warn)
2054 goto end;
2055
2056 ret = target(1, 2, 3, 4, 5, 6);
2057
2058 ret = command_trace_probe("-:testprobe");
2059 if (WARN_ON_ONCE(ret)) {
2060 pr_warning("error on deleting a probe.\n");
2061 warn++;
2062 }
2063
2064 ret = command_trace_probe("-:testprobe2");
2065 if (WARN_ON_ONCE(ret)) {
2066 pr_warning("error on deleting a probe.\n");
2067 warn++;
2068 }
2069
2070end:
2071 release_all_trace_probes();
2072 if (warn)
2073 pr_cont("NG: Some tests are failed. Please check them.\n");
2074 else
2075 pr_cont("OK\n");
2076 return 0;
2077}
2078
2079late_initcall(kprobe_trace_self_tests_init);
2080
2081#endif
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Kprobes-based tracing events
4 *
5 * Created by Masami Hiramatsu <mhiramat@redhat.com>
6 *
7 */
8#define pr_fmt(fmt) "trace_kprobe: " fmt
9
10#include <linux/bpf-cgroup.h>
11#include <linux/security.h>
12#include <linux/module.h>
13#include <linux/uaccess.h>
14#include <linux/rculist.h>
15#include <linux/error-injection.h>
16
17#include <asm/setup.h> /* for COMMAND_LINE_SIZE */
18
19#include "trace_dynevent.h"
20#include "trace_kprobe_selftest.h"
21#include "trace_probe.h"
22#include "trace_probe_tmpl.h"
23#include "trace_probe_kernel.h"
24
25#define KPROBE_EVENT_SYSTEM "kprobes"
26#define KRETPROBE_MAXACTIVE_MAX 4096
27
28/* Kprobe early definition from command line */
29static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
30
31static int __init set_kprobe_boot_events(char *str)
32{
33 strscpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
34 disable_tracing_selftest("running kprobe events");
35
36 return 1;
37}
38__setup("kprobe_event=", set_kprobe_boot_events);
39
40static int trace_kprobe_create(const char *raw_command);
41static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
42static int trace_kprobe_release(struct dyn_event *ev);
43static bool trace_kprobe_is_busy(struct dyn_event *ev);
44static bool trace_kprobe_match(const char *system, const char *event,
45 int argc, const char **argv, struct dyn_event *ev);
46
47static struct dyn_event_operations trace_kprobe_ops = {
48 .create = trace_kprobe_create,
49 .show = trace_kprobe_show,
50 .is_busy = trace_kprobe_is_busy,
51 .free = trace_kprobe_release,
52 .match = trace_kprobe_match,
53};
54
55/*
56 * Kprobe event core functions
57 */
58struct trace_kprobe {
59 struct dyn_event devent;
60 struct kretprobe rp; /* Use rp.kp for kprobe use */
61 unsigned long __percpu *nhit;
62 const char *symbol; /* symbol name */
63 struct trace_probe tp;
64};
65
66static bool is_trace_kprobe(struct dyn_event *ev)
67{
68 return ev->ops == &trace_kprobe_ops;
69}
70
71static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
72{
73 return container_of(ev, struct trace_kprobe, devent);
74}
75
76/**
77 * for_each_trace_kprobe - iterate over the trace_kprobe list
78 * @pos: the struct trace_kprobe * for each entry
79 * @dpos: the struct dyn_event * to use as a loop cursor
80 */
81#define for_each_trace_kprobe(pos, dpos) \
82 for_each_dyn_event(dpos) \
83 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
84
85static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
86{
87 return tk->rp.handler != NULL;
88}
89
90static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
91{
92 return tk->symbol ? tk->symbol : "unknown";
93}
94
95static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
96{
97 return tk->rp.kp.offset;
98}
99
100static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
101{
102 return kprobe_gone(&tk->rp.kp);
103}
104
105static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
106 struct module *mod)
107{
108 int len = strlen(module_name(mod));
109 const char *name = trace_kprobe_symbol(tk);
110
111 return strncmp(module_name(mod), name, len) == 0 && name[len] == ':';
112}
113
114static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
115{
116 char *p;
117 bool ret;
118
119 if (!tk->symbol)
120 return false;
121 p = strchr(tk->symbol, ':');
122 if (!p)
123 return true;
124 *p = '\0';
125 rcu_read_lock_sched();
126 ret = !!find_module(tk->symbol);
127 rcu_read_unlock_sched();
128 *p = ':';
129
130 return ret;
131}
132
133static bool trace_kprobe_is_busy(struct dyn_event *ev)
134{
135 struct trace_kprobe *tk = to_trace_kprobe(ev);
136
137 return trace_probe_is_enabled(&tk->tp);
138}
139
140static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
141 int argc, const char **argv)
142{
143 char buf[MAX_ARGSTR_LEN + 1];
144
145 if (!argc)
146 return true;
147
148 if (!tk->symbol)
149 snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
150 else if (tk->rp.kp.offset)
151 snprintf(buf, sizeof(buf), "%s+%u",
152 trace_kprobe_symbol(tk), tk->rp.kp.offset);
153 else
154 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
155 if (strcmp(buf, argv[0]))
156 return false;
157 argc--; argv++;
158
159 return trace_probe_match_command_args(&tk->tp, argc, argv);
160}
161
162static bool trace_kprobe_match(const char *system, const char *event,
163 int argc, const char **argv, struct dyn_event *ev)
164{
165 struct trace_kprobe *tk = to_trace_kprobe(ev);
166
167 return (event[0] == '\0' ||
168 strcmp(trace_probe_name(&tk->tp), event) == 0) &&
169 (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
170 trace_kprobe_match_command_head(tk, argc, argv);
171}
172
173static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
174{
175 unsigned long nhit = 0;
176 int cpu;
177
178 for_each_possible_cpu(cpu)
179 nhit += *per_cpu_ptr(tk->nhit, cpu);
180
181 return nhit;
182}
183
184static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
185{
186 return !(list_empty(&tk->rp.kp.list) &&
187 hlist_unhashed(&tk->rp.kp.hlist));
188}
189
190/* Return 0 if it fails to find the symbol address */
191static nokprobe_inline
192unsigned long trace_kprobe_address(struct trace_kprobe *tk)
193{
194 unsigned long addr;
195
196 if (tk->symbol) {
197 addr = (unsigned long)
198 kallsyms_lookup_name(trace_kprobe_symbol(tk));
199 if (addr)
200 addr += tk->rp.kp.offset;
201 } else {
202 addr = (unsigned long)tk->rp.kp.addr;
203 }
204 return addr;
205}
206
207static nokprobe_inline struct trace_kprobe *
208trace_kprobe_primary_from_call(struct trace_event_call *call)
209{
210 struct trace_probe *tp;
211
212 tp = trace_probe_primary_from_call(call);
213 if (WARN_ON_ONCE(!tp))
214 return NULL;
215
216 return container_of(tp, struct trace_kprobe, tp);
217}
218
219bool trace_kprobe_on_func_entry(struct trace_event_call *call)
220{
221 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
222
223 return tk ? (kprobe_on_func_entry(tk->rp.kp.addr,
224 tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
225 tk->rp.kp.addr ? 0 : tk->rp.kp.offset) == 0) : false;
226}
227
228bool trace_kprobe_error_injectable(struct trace_event_call *call)
229{
230 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
231
232 return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
233 false;
234}
235
236static int register_kprobe_event(struct trace_kprobe *tk);
237static int unregister_kprobe_event(struct trace_kprobe *tk);
238
239static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
240static int kretprobe_dispatcher(struct kretprobe_instance *ri,
241 struct pt_regs *regs);
242
243static void free_trace_kprobe(struct trace_kprobe *tk)
244{
245 if (tk) {
246 trace_probe_cleanup(&tk->tp);
247 kfree(tk->symbol);
248 free_percpu(tk->nhit);
249 kfree(tk);
250 }
251}
252
253/*
254 * Allocate new trace_probe and initialize it (including kprobes).
255 */
256static struct trace_kprobe *alloc_trace_kprobe(const char *group,
257 const char *event,
258 void *addr,
259 const char *symbol,
260 unsigned long offs,
261 int maxactive,
262 int nargs, bool is_return)
263{
264 struct trace_kprobe *tk;
265 int ret = -ENOMEM;
266
267 tk = kzalloc(struct_size(tk, tp.args, nargs), GFP_KERNEL);
268 if (!tk)
269 return ERR_PTR(ret);
270
271 tk->nhit = alloc_percpu(unsigned long);
272 if (!tk->nhit)
273 goto error;
274
275 if (symbol) {
276 tk->symbol = kstrdup(symbol, GFP_KERNEL);
277 if (!tk->symbol)
278 goto error;
279 tk->rp.kp.symbol_name = tk->symbol;
280 tk->rp.kp.offset = offs;
281 } else
282 tk->rp.kp.addr = addr;
283
284 if (is_return)
285 tk->rp.handler = kretprobe_dispatcher;
286 else
287 tk->rp.kp.pre_handler = kprobe_dispatcher;
288
289 tk->rp.maxactive = maxactive;
290 INIT_HLIST_NODE(&tk->rp.kp.hlist);
291 INIT_LIST_HEAD(&tk->rp.kp.list);
292
293 ret = trace_probe_init(&tk->tp, event, group, false);
294 if (ret < 0)
295 goto error;
296
297 dyn_event_init(&tk->devent, &trace_kprobe_ops);
298 return tk;
299error:
300 free_trace_kprobe(tk);
301 return ERR_PTR(ret);
302}
303
304static struct trace_kprobe *find_trace_kprobe(const char *event,
305 const char *group)
306{
307 struct dyn_event *pos;
308 struct trace_kprobe *tk;
309
310 for_each_trace_kprobe(tk, pos)
311 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
312 strcmp(trace_probe_group_name(&tk->tp), group) == 0)
313 return tk;
314 return NULL;
315}
316
317static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
318{
319 int ret = 0;
320
321 if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
322 if (trace_kprobe_is_return(tk))
323 ret = enable_kretprobe(&tk->rp);
324 else
325 ret = enable_kprobe(&tk->rp.kp);
326 }
327
328 return ret;
329}
330
331static void __disable_trace_kprobe(struct trace_probe *tp)
332{
333 struct trace_kprobe *tk;
334
335 list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
336 if (!trace_kprobe_is_registered(tk))
337 continue;
338 if (trace_kprobe_is_return(tk))
339 disable_kretprobe(&tk->rp);
340 else
341 disable_kprobe(&tk->rp.kp);
342 }
343}
344
345/*
346 * Enable trace_probe
347 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
348 */
349static int enable_trace_kprobe(struct trace_event_call *call,
350 struct trace_event_file *file)
351{
352 struct trace_probe *tp;
353 struct trace_kprobe *tk;
354 bool enabled;
355 int ret = 0;
356
357 tp = trace_probe_primary_from_call(call);
358 if (WARN_ON_ONCE(!tp))
359 return -ENODEV;
360 enabled = trace_probe_is_enabled(tp);
361
362 /* This also changes "enabled" state */
363 if (file) {
364 ret = trace_probe_add_file(tp, file);
365 if (ret)
366 return ret;
367 } else
368 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
369
370 if (enabled)
371 return 0;
372
373 list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
374 if (trace_kprobe_has_gone(tk))
375 continue;
376 ret = __enable_trace_kprobe(tk);
377 if (ret)
378 break;
379 enabled = true;
380 }
381
382 if (ret) {
383 /* Failed to enable one of them. Roll back all */
384 if (enabled)
385 __disable_trace_kprobe(tp);
386 if (file)
387 trace_probe_remove_file(tp, file);
388 else
389 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
390 }
391
392 return ret;
393}
394
395/*
396 * Disable trace_probe
397 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
398 */
399static int disable_trace_kprobe(struct trace_event_call *call,
400 struct trace_event_file *file)
401{
402 struct trace_probe *tp;
403
404 tp = trace_probe_primary_from_call(call);
405 if (WARN_ON_ONCE(!tp))
406 return -ENODEV;
407
408 if (file) {
409 if (!trace_probe_get_file_link(tp, file))
410 return -ENOENT;
411 if (!trace_probe_has_single_file(tp))
412 goto out;
413 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
414 } else
415 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
416
417 if (!trace_probe_is_enabled(tp))
418 __disable_trace_kprobe(tp);
419
420 out:
421 if (file)
422 /*
423 * Synchronization is done in below function. For perf event,
424 * file == NULL and perf_trace_event_unreg() calls
425 * tracepoint_synchronize_unregister() to ensure synchronize
426 * event. We don't need to care about it.
427 */
428 trace_probe_remove_file(tp, file);
429
430 return 0;
431}
432
433#if defined(CONFIG_DYNAMIC_FTRACE) && \
434 !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
435static bool __within_notrace_func(unsigned long addr)
436{
437 unsigned long offset, size;
438
439 if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
440 return false;
441
442 /* Get the entry address of the target function */
443 addr -= offset;
444
445 /*
446 * Since ftrace_location_range() does inclusive range check, we need
447 * to subtract 1 byte from the end address.
448 */
449 return !ftrace_location_range(addr, addr + size - 1);
450}
451
452static bool within_notrace_func(struct trace_kprobe *tk)
453{
454 unsigned long addr = trace_kprobe_address(tk);
455 char symname[KSYM_NAME_LEN], *p;
456
457 if (!__within_notrace_func(addr))
458 return false;
459
460 /* Check if the address is on a suffixed-symbol */
461 if (!lookup_symbol_name(addr, symname)) {
462 p = strchr(symname, '.');
463 if (!p)
464 return true;
465 *p = '\0';
466 addr = (unsigned long)kprobe_lookup_name(symname, 0);
467 if (addr)
468 return __within_notrace_func(addr);
469 }
470
471 return true;
472}
473#else
474#define within_notrace_func(tk) (false)
475#endif
476
477/* Internal register function - just handle k*probes and flags */
478static int __register_trace_kprobe(struct trace_kprobe *tk)
479{
480 int i, ret;
481
482 ret = security_locked_down(LOCKDOWN_KPROBES);
483 if (ret)
484 return ret;
485
486 if (trace_kprobe_is_registered(tk))
487 return -EINVAL;
488
489 if (within_notrace_func(tk)) {
490 pr_warn("Could not probe notrace function %ps\n",
491 (void *)trace_kprobe_address(tk));
492 return -EINVAL;
493 }
494
495 for (i = 0; i < tk->tp.nr_args; i++) {
496 ret = traceprobe_update_arg(&tk->tp.args[i]);
497 if (ret)
498 return ret;
499 }
500
501 /* Set/clear disabled flag according to tp->flag */
502 if (trace_probe_is_enabled(&tk->tp))
503 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
504 else
505 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
506
507 if (trace_kprobe_is_return(tk))
508 ret = register_kretprobe(&tk->rp);
509 else
510 ret = register_kprobe(&tk->rp.kp);
511
512 return ret;
513}
514
515/* Internal unregister function - just handle k*probes and flags */
516static void __unregister_trace_kprobe(struct trace_kprobe *tk)
517{
518 if (trace_kprobe_is_registered(tk)) {
519 if (trace_kprobe_is_return(tk))
520 unregister_kretprobe(&tk->rp);
521 else
522 unregister_kprobe(&tk->rp.kp);
523 /* Cleanup kprobe for reuse and mark it unregistered */
524 INIT_HLIST_NODE(&tk->rp.kp.hlist);
525 INIT_LIST_HEAD(&tk->rp.kp.list);
526 if (tk->rp.kp.symbol_name)
527 tk->rp.kp.addr = NULL;
528 }
529}
530
531/* Unregister a trace_probe and probe_event */
532static int unregister_trace_kprobe(struct trace_kprobe *tk)
533{
534 /* If other probes are on the event, just unregister kprobe */
535 if (trace_probe_has_sibling(&tk->tp))
536 goto unreg;
537
538 /* Enabled event can not be unregistered */
539 if (trace_probe_is_enabled(&tk->tp))
540 return -EBUSY;
541
542 /* If there's a reference to the dynamic event */
543 if (trace_event_dyn_busy(trace_probe_event_call(&tk->tp)))
544 return -EBUSY;
545
546 /* Will fail if probe is being used by ftrace or perf */
547 if (unregister_kprobe_event(tk))
548 return -EBUSY;
549
550unreg:
551 __unregister_trace_kprobe(tk);
552 dyn_event_remove(&tk->devent);
553 trace_probe_unlink(&tk->tp);
554
555 return 0;
556}
557
558static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
559 struct trace_kprobe *comp)
560{
561 struct trace_probe_event *tpe = orig->tp.event;
562 int i;
563
564 list_for_each_entry(orig, &tpe->probes, tp.list) {
565 if (strcmp(trace_kprobe_symbol(orig),
566 trace_kprobe_symbol(comp)) ||
567 trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
568 continue;
569
570 /*
571 * trace_probe_compare_arg_type() ensured that nr_args and
572 * each argument name and type are same. Let's compare comm.
573 */
574 for (i = 0; i < orig->tp.nr_args; i++) {
575 if (strcmp(orig->tp.args[i].comm,
576 comp->tp.args[i].comm))
577 break;
578 }
579
580 if (i == orig->tp.nr_args)
581 return true;
582 }
583
584 return false;
585}
586
587static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
588{
589 int ret;
590
591 ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
592 if (ret) {
593 /* Note that argument starts index = 2 */
594 trace_probe_log_set_index(ret + 1);
595 trace_probe_log_err(0, DIFF_ARG_TYPE);
596 return -EEXIST;
597 }
598 if (trace_kprobe_has_same_kprobe(to, tk)) {
599 trace_probe_log_set_index(0);
600 trace_probe_log_err(0, SAME_PROBE);
601 return -EEXIST;
602 }
603
604 /* Append to existing event */
605 ret = trace_probe_append(&tk->tp, &to->tp);
606 if (ret)
607 return ret;
608
609 /* Register k*probe */
610 ret = __register_trace_kprobe(tk);
611 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
612 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
613 ret = 0;
614 }
615
616 if (ret)
617 trace_probe_unlink(&tk->tp);
618 else
619 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
620
621 return ret;
622}
623
624/* Register a trace_probe and probe_event */
625static int register_trace_kprobe(struct trace_kprobe *tk)
626{
627 struct trace_kprobe *old_tk;
628 int ret;
629
630 mutex_lock(&event_mutex);
631
632 old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
633 trace_probe_group_name(&tk->tp));
634 if (old_tk) {
635 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
636 trace_probe_log_set_index(0);
637 trace_probe_log_err(0, DIFF_PROBE_TYPE);
638 ret = -EEXIST;
639 } else {
640 ret = append_trace_kprobe(tk, old_tk);
641 }
642 goto end;
643 }
644
645 /* Register new event */
646 ret = register_kprobe_event(tk);
647 if (ret) {
648 if (ret == -EEXIST) {
649 trace_probe_log_set_index(0);
650 trace_probe_log_err(0, EVENT_EXIST);
651 } else
652 pr_warn("Failed to register probe event(%d)\n", ret);
653 goto end;
654 }
655
656 /* Register k*probe */
657 ret = __register_trace_kprobe(tk);
658 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
659 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
660 ret = 0;
661 }
662
663 if (ret < 0)
664 unregister_kprobe_event(tk);
665 else
666 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
667
668end:
669 mutex_unlock(&event_mutex);
670 return ret;
671}
672
673/* Module notifier call back, checking event on the module */
674static int trace_kprobe_module_callback(struct notifier_block *nb,
675 unsigned long val, void *data)
676{
677 struct module *mod = data;
678 struct dyn_event *pos;
679 struct trace_kprobe *tk;
680 int ret;
681
682 if (val != MODULE_STATE_COMING)
683 return NOTIFY_DONE;
684
685 /* Update probes on coming module */
686 mutex_lock(&event_mutex);
687 for_each_trace_kprobe(tk, pos) {
688 if (trace_kprobe_within_module(tk, mod)) {
689 /* Don't need to check busy - this should have gone. */
690 __unregister_trace_kprobe(tk);
691 ret = __register_trace_kprobe(tk);
692 if (ret)
693 pr_warn("Failed to re-register probe %s on %s: %d\n",
694 trace_probe_name(&tk->tp),
695 module_name(mod), ret);
696 }
697 }
698 mutex_unlock(&event_mutex);
699
700 return NOTIFY_DONE;
701}
702
703static struct notifier_block trace_kprobe_module_nb = {
704 .notifier_call = trace_kprobe_module_callback,
705 .priority = 1 /* Invoked after kprobe module callback */
706};
707
708static int count_symbols(void *data, unsigned long unused)
709{
710 unsigned int *count = data;
711
712 (*count)++;
713
714 return 0;
715}
716
717struct sym_count_ctx {
718 unsigned int count;
719 const char *name;
720};
721
722static int count_mod_symbols(void *data, const char *name, unsigned long unused)
723{
724 struct sym_count_ctx *ctx = data;
725
726 if (strcmp(name, ctx->name) == 0)
727 ctx->count++;
728
729 return 0;
730}
731
732static unsigned int number_of_same_symbols(char *func_name)
733{
734 struct sym_count_ctx ctx = { .count = 0, .name = func_name };
735
736 kallsyms_on_each_match_symbol(count_symbols, func_name, &ctx.count);
737
738 module_kallsyms_on_each_symbol(NULL, count_mod_symbols, &ctx);
739
740 return ctx.count;
741}
742
743static int __trace_kprobe_create(int argc, const char *argv[])
744{
745 /*
746 * Argument syntax:
747 * - Add kprobe:
748 * p[:[GRP/][EVENT]] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
749 * - Add kretprobe:
750 * r[MAXACTIVE][:[GRP/][EVENT]] [MOD:]KSYM[+0] [FETCHARGS]
751 * Or
752 * p[:[GRP/][EVENT]] [MOD:]KSYM[+0]%return [FETCHARGS]
753 *
754 * Fetch args:
755 * $retval : fetch return value
756 * $stack : fetch stack address
757 * $stackN : fetch Nth of stack (N:0-)
758 * $comm : fetch current task comm
759 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
760 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
761 * %REG : fetch register REG
762 * Dereferencing memory fetch:
763 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
764 * Alias name of args:
765 * NAME=FETCHARG : set NAME as alias of FETCHARG.
766 * Type of args:
767 * FETCHARG:TYPE : use TYPE instead of unsigned long.
768 */
769 struct trace_kprobe *tk = NULL;
770 int i, len, new_argc = 0, ret = 0;
771 bool is_return = false;
772 char *symbol = NULL, *tmp = NULL;
773 const char **new_argv = NULL;
774 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
775 enum probe_print_type ptype;
776 int maxactive = 0;
777 long offset = 0;
778 void *addr = NULL;
779 char buf[MAX_EVENT_NAME_LEN];
780 char gbuf[MAX_EVENT_NAME_LEN];
781 char abuf[MAX_BTF_ARGS_LEN];
782 struct traceprobe_parse_context ctx = { .flags = TPARG_FL_KERNEL };
783
784 switch (argv[0][0]) {
785 case 'r':
786 is_return = true;
787 break;
788 case 'p':
789 break;
790 default:
791 return -ECANCELED;
792 }
793 if (argc < 2)
794 return -ECANCELED;
795
796 trace_probe_log_init("trace_kprobe", argc, argv);
797
798 event = strchr(&argv[0][1], ':');
799 if (event)
800 event++;
801
802 if (isdigit(argv[0][1])) {
803 if (!is_return) {
804 trace_probe_log_err(1, BAD_MAXACT_TYPE);
805 goto parse_error;
806 }
807 if (event)
808 len = event - &argv[0][1] - 1;
809 else
810 len = strlen(&argv[0][1]);
811 if (len > MAX_EVENT_NAME_LEN - 1) {
812 trace_probe_log_err(1, BAD_MAXACT);
813 goto parse_error;
814 }
815 memcpy(buf, &argv[0][1], len);
816 buf[len] = '\0';
817 ret = kstrtouint(buf, 0, &maxactive);
818 if (ret || !maxactive) {
819 trace_probe_log_err(1, BAD_MAXACT);
820 goto parse_error;
821 }
822 /* kretprobes instances are iterated over via a list. The
823 * maximum should stay reasonable.
824 */
825 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
826 trace_probe_log_err(1, MAXACT_TOO_BIG);
827 goto parse_error;
828 }
829 }
830
831 /* try to parse an address. if that fails, try to read the
832 * input as a symbol. */
833 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
834 trace_probe_log_set_index(1);
835 /* Check whether uprobe event specified */
836 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
837 ret = -ECANCELED;
838 goto error;
839 }
840 /* a symbol specified */
841 symbol = kstrdup(argv[1], GFP_KERNEL);
842 if (!symbol)
843 return -ENOMEM;
844
845 tmp = strchr(symbol, '%');
846 if (tmp) {
847 if (!strcmp(tmp, "%return")) {
848 *tmp = '\0';
849 is_return = true;
850 } else {
851 trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
852 goto parse_error;
853 }
854 }
855
856 /* TODO: support .init module functions */
857 ret = traceprobe_split_symbol_offset(symbol, &offset);
858 if (ret || offset < 0 || offset > UINT_MAX) {
859 trace_probe_log_err(0, BAD_PROBE_ADDR);
860 goto parse_error;
861 }
862 if (is_return)
863 ctx.flags |= TPARG_FL_RETURN;
864 ret = kprobe_on_func_entry(NULL, symbol, offset);
865 if (ret == 0 && !is_return)
866 ctx.flags |= TPARG_FL_FENTRY;
867 /* Defer the ENOENT case until register kprobe */
868 if (ret == -EINVAL && is_return) {
869 trace_probe_log_err(0, BAD_RETPROBE);
870 goto parse_error;
871 }
872 }
873
874 if (symbol && !strchr(symbol, ':')) {
875 unsigned int count;
876
877 count = number_of_same_symbols(symbol);
878 if (count > 1) {
879 /*
880 * Users should use ADDR to remove the ambiguity of
881 * using KSYM only.
882 */
883 trace_probe_log_err(0, NON_UNIQ_SYMBOL);
884 ret = -EADDRNOTAVAIL;
885
886 goto error;
887 } else if (count == 0) {
888 /*
889 * We can return ENOENT earlier than when register the
890 * kprobe.
891 */
892 trace_probe_log_err(0, BAD_PROBE_ADDR);
893 ret = -ENOENT;
894
895 goto error;
896 }
897 }
898
899 trace_probe_log_set_index(0);
900 if (event) {
901 ret = traceprobe_parse_event_name(&event, &group, gbuf,
902 event - argv[0]);
903 if (ret)
904 goto parse_error;
905 }
906
907 if (!event) {
908 /* Make a new event name */
909 if (symbol)
910 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
911 is_return ? 'r' : 'p', symbol, offset);
912 else
913 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
914 is_return ? 'r' : 'p', addr);
915 sanitize_event_name(buf);
916 event = buf;
917 }
918
919 argc -= 2; argv += 2;
920 ctx.funcname = symbol;
921 new_argv = traceprobe_expand_meta_args(argc, argv, &new_argc,
922 abuf, MAX_BTF_ARGS_LEN, &ctx);
923 if (IS_ERR(new_argv)) {
924 ret = PTR_ERR(new_argv);
925 new_argv = NULL;
926 goto out;
927 }
928 if (new_argv) {
929 argc = new_argc;
930 argv = new_argv;
931 }
932
933 /* setup a probe */
934 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
935 argc, is_return);
936 if (IS_ERR(tk)) {
937 ret = PTR_ERR(tk);
938 /* This must return -ENOMEM, else there is a bug */
939 WARN_ON_ONCE(ret != -ENOMEM);
940 goto out; /* We know tk is not allocated */
941 }
942
943 /* parse arguments */
944 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
945 trace_probe_log_set_index(i + 2);
946 ctx.offset = 0;
947 ret = traceprobe_parse_probe_arg(&tk->tp, i, argv[i], &ctx);
948 if (ret)
949 goto error; /* This can be -ENOMEM */
950 }
951
952 ptype = is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
953 ret = traceprobe_set_print_fmt(&tk->tp, ptype);
954 if (ret < 0)
955 goto error;
956
957 ret = register_trace_kprobe(tk);
958 if (ret) {
959 trace_probe_log_set_index(1);
960 if (ret == -EILSEQ)
961 trace_probe_log_err(0, BAD_INSN_BNDRY);
962 else if (ret == -ENOENT)
963 trace_probe_log_err(0, BAD_PROBE_ADDR);
964 else if (ret != -ENOMEM && ret != -EEXIST)
965 trace_probe_log_err(0, FAIL_REG_PROBE);
966 goto error;
967 }
968
969out:
970 traceprobe_finish_parse(&ctx);
971 trace_probe_log_clear();
972 kfree(new_argv);
973 kfree(symbol);
974 return ret;
975
976parse_error:
977 ret = -EINVAL;
978error:
979 free_trace_kprobe(tk);
980 goto out;
981}
982
983static int trace_kprobe_create(const char *raw_command)
984{
985 return trace_probe_create(raw_command, __trace_kprobe_create);
986}
987
988static int create_or_delete_trace_kprobe(const char *raw_command)
989{
990 int ret;
991
992 if (raw_command[0] == '-')
993 return dyn_event_release(raw_command, &trace_kprobe_ops);
994
995 ret = trace_kprobe_create(raw_command);
996 return ret == -ECANCELED ? -EINVAL : ret;
997}
998
999static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
1000{
1001 return create_or_delete_trace_kprobe(cmd->seq.buffer);
1002}
1003
1004/**
1005 * kprobe_event_cmd_init - Initialize a kprobe event command object
1006 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1007 * @buf: A pointer to the buffer used to build the command
1008 * @maxlen: The length of the buffer passed in @buf
1009 *
1010 * Initialize a synthetic event command object. Use this before
1011 * calling any of the other kprobe_event functions.
1012 */
1013void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
1014{
1015 dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
1016 trace_kprobe_run_command);
1017}
1018EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
1019
1020/**
1021 * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
1022 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1023 * @kretprobe: Is this a return probe?
1024 * @name: The name of the kprobe event
1025 * @loc: The location of the kprobe event
1026 * @...: Variable number of arg (pairs), one pair for each field
1027 *
1028 * NOTE: Users normally won't want to call this function directly, but
1029 * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
1030 * adds a NULL to the end of the arg list. If this function is used
1031 * directly, make sure the last arg in the variable arg list is NULL.
1032 *
1033 * Generate a kprobe event command to be executed by
1034 * kprobe_event_gen_cmd_end(). This function can be used to generate the
1035 * complete command or only the first part of it; in the latter case,
1036 * kprobe_event_add_fields() can be used to add more fields following this.
1037 *
1038 * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
1039 * returns -EINVAL if @loc == NULL.
1040 *
1041 * Return: 0 if successful, error otherwise.
1042 */
1043int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
1044 const char *name, const char *loc, ...)
1045{
1046 char buf[MAX_EVENT_NAME_LEN];
1047 struct dynevent_arg arg;
1048 va_list args;
1049 int ret;
1050
1051 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1052 return -EINVAL;
1053
1054 if (!loc)
1055 return -EINVAL;
1056
1057 if (kretprobe)
1058 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
1059 else
1060 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
1061
1062 ret = dynevent_str_add(cmd, buf);
1063 if (ret)
1064 return ret;
1065
1066 dynevent_arg_init(&arg, 0);
1067 arg.str = loc;
1068 ret = dynevent_arg_add(cmd, &arg, NULL);
1069 if (ret)
1070 return ret;
1071
1072 va_start(args, loc);
1073 for (;;) {
1074 const char *field;
1075
1076 field = va_arg(args, const char *);
1077 if (!field)
1078 break;
1079
1080 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1081 ret = -EINVAL;
1082 break;
1083 }
1084
1085 arg.str = field;
1086 ret = dynevent_arg_add(cmd, &arg, NULL);
1087 if (ret)
1088 break;
1089 }
1090 va_end(args);
1091
1092 return ret;
1093}
1094EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1095
1096/**
1097 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1098 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1099 * @...: Variable number of arg (pairs), one pair for each field
1100 *
1101 * NOTE: Users normally won't want to call this function directly, but
1102 * rather use the kprobe_event_add_fields() wrapper, which
1103 * automatically adds a NULL to the end of the arg list. If this
1104 * function is used directly, make sure the last arg in the variable
1105 * arg list is NULL.
1106 *
1107 * Add probe fields to an existing kprobe command using a variable
1108 * list of args. Fields are added in the same order they're listed.
1109 *
1110 * Return: 0 if successful, error otherwise.
1111 */
1112int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1113{
1114 struct dynevent_arg arg;
1115 va_list args;
1116 int ret = 0;
1117
1118 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1119 return -EINVAL;
1120
1121 dynevent_arg_init(&arg, 0);
1122
1123 va_start(args, cmd);
1124 for (;;) {
1125 const char *field;
1126
1127 field = va_arg(args, const char *);
1128 if (!field)
1129 break;
1130
1131 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1132 ret = -EINVAL;
1133 break;
1134 }
1135
1136 arg.str = field;
1137 ret = dynevent_arg_add(cmd, &arg, NULL);
1138 if (ret)
1139 break;
1140 }
1141 va_end(args);
1142
1143 return ret;
1144}
1145EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1146
1147/**
1148 * kprobe_event_delete - Delete a kprobe event
1149 * @name: The name of the kprobe event to delete
1150 *
1151 * Delete a kprobe event with the give @name from kernel code rather
1152 * than directly from the command line.
1153 *
1154 * Return: 0 if successful, error otherwise.
1155 */
1156int kprobe_event_delete(const char *name)
1157{
1158 char buf[MAX_EVENT_NAME_LEN];
1159
1160 snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1161
1162 return create_or_delete_trace_kprobe(buf);
1163}
1164EXPORT_SYMBOL_GPL(kprobe_event_delete);
1165
1166static int trace_kprobe_release(struct dyn_event *ev)
1167{
1168 struct trace_kprobe *tk = to_trace_kprobe(ev);
1169 int ret = unregister_trace_kprobe(tk);
1170
1171 if (!ret)
1172 free_trace_kprobe(tk);
1173 return ret;
1174}
1175
1176static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1177{
1178 struct trace_kprobe *tk = to_trace_kprobe(ev);
1179 int i;
1180
1181 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1182 if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1183 seq_printf(m, "%d", tk->rp.maxactive);
1184 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1185 trace_probe_name(&tk->tp));
1186
1187 if (!tk->symbol)
1188 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1189 else if (tk->rp.kp.offset)
1190 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1191 tk->rp.kp.offset);
1192 else
1193 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1194
1195 for (i = 0; i < tk->tp.nr_args; i++)
1196 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1197 seq_putc(m, '\n');
1198
1199 return 0;
1200}
1201
1202static int probes_seq_show(struct seq_file *m, void *v)
1203{
1204 struct dyn_event *ev = v;
1205
1206 if (!is_trace_kprobe(ev))
1207 return 0;
1208
1209 return trace_kprobe_show(m, ev);
1210}
1211
1212static const struct seq_operations probes_seq_op = {
1213 .start = dyn_event_seq_start,
1214 .next = dyn_event_seq_next,
1215 .stop = dyn_event_seq_stop,
1216 .show = probes_seq_show
1217};
1218
1219static int probes_open(struct inode *inode, struct file *file)
1220{
1221 int ret;
1222
1223 ret = security_locked_down(LOCKDOWN_TRACEFS);
1224 if (ret)
1225 return ret;
1226
1227 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1228 ret = dyn_events_release_all(&trace_kprobe_ops);
1229 if (ret < 0)
1230 return ret;
1231 }
1232
1233 return seq_open(file, &probes_seq_op);
1234}
1235
1236static ssize_t probes_write(struct file *file, const char __user *buffer,
1237 size_t count, loff_t *ppos)
1238{
1239 return trace_parse_run_command(file, buffer, count, ppos,
1240 create_or_delete_trace_kprobe);
1241}
1242
1243static const struct file_operations kprobe_events_ops = {
1244 .owner = THIS_MODULE,
1245 .open = probes_open,
1246 .read = seq_read,
1247 .llseek = seq_lseek,
1248 .release = seq_release,
1249 .write = probes_write,
1250};
1251
1252static unsigned long trace_kprobe_missed(struct trace_kprobe *tk)
1253{
1254 return trace_kprobe_is_return(tk) ?
1255 tk->rp.kp.nmissed + tk->rp.nmissed : tk->rp.kp.nmissed;
1256}
1257
1258/* Probes profiling interfaces */
1259static int probes_profile_seq_show(struct seq_file *m, void *v)
1260{
1261 struct dyn_event *ev = v;
1262 struct trace_kprobe *tk;
1263 unsigned long nmissed;
1264
1265 if (!is_trace_kprobe(ev))
1266 return 0;
1267
1268 tk = to_trace_kprobe(ev);
1269 nmissed = trace_kprobe_missed(tk);
1270 seq_printf(m, " %-44s %15lu %15lu\n",
1271 trace_probe_name(&tk->tp),
1272 trace_kprobe_nhit(tk),
1273 nmissed);
1274
1275 return 0;
1276}
1277
1278static const struct seq_operations profile_seq_op = {
1279 .start = dyn_event_seq_start,
1280 .next = dyn_event_seq_next,
1281 .stop = dyn_event_seq_stop,
1282 .show = probes_profile_seq_show
1283};
1284
1285static int profile_open(struct inode *inode, struct file *file)
1286{
1287 int ret;
1288
1289 ret = security_locked_down(LOCKDOWN_TRACEFS);
1290 if (ret)
1291 return ret;
1292
1293 return seq_open(file, &profile_seq_op);
1294}
1295
1296static const struct file_operations kprobe_profile_ops = {
1297 .owner = THIS_MODULE,
1298 .open = profile_open,
1299 .read = seq_read,
1300 .llseek = seq_lseek,
1301 .release = seq_release,
1302};
1303
1304/* Note that we don't verify it, since the code does not come from user space */
1305static int
1306process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
1307 void *base)
1308{
1309 struct pt_regs *regs = rec;
1310 unsigned long val;
1311 int ret;
1312
1313retry:
1314 /* 1st stage: get value from context */
1315 switch (code->op) {
1316 case FETCH_OP_REG:
1317 val = regs_get_register(regs, code->param);
1318 break;
1319 case FETCH_OP_STACK:
1320 val = regs_get_kernel_stack_nth(regs, code->param);
1321 break;
1322 case FETCH_OP_STACKP:
1323 val = kernel_stack_pointer(regs);
1324 break;
1325 case FETCH_OP_RETVAL:
1326 val = regs_return_value(regs);
1327 break;
1328#ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1329 case FETCH_OP_ARG:
1330 val = regs_get_kernel_argument(regs, code->param);
1331 break;
1332#endif
1333 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1334 code++;
1335 goto retry;
1336 default:
1337 ret = process_common_fetch_insn(code, &val);
1338 if (ret < 0)
1339 return ret;
1340 }
1341 code++;
1342
1343 return process_fetch_insn_bottom(code, val, dest, base);
1344}
1345NOKPROBE_SYMBOL(process_fetch_insn)
1346
1347/* Kprobe handler */
1348static nokprobe_inline void
1349__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1350 struct trace_event_file *trace_file)
1351{
1352 struct kprobe_trace_entry_head *entry;
1353 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1354 struct trace_event_buffer fbuffer;
1355 int dsize;
1356
1357 WARN_ON(call != trace_file->event_call);
1358
1359 if (trace_trigger_soft_disabled(trace_file))
1360 return;
1361
1362 dsize = __get_data_size(&tk->tp, regs);
1363
1364 entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1365 sizeof(*entry) + tk->tp.size + dsize);
1366 if (!entry)
1367 return;
1368
1369 fbuffer.regs = regs;
1370 entry->ip = (unsigned long)tk->rp.kp.addr;
1371 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1372
1373 trace_event_buffer_commit(&fbuffer);
1374}
1375
1376static void
1377kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1378{
1379 struct event_file_link *link;
1380
1381 trace_probe_for_each_link_rcu(link, &tk->tp)
1382 __kprobe_trace_func(tk, regs, link->file);
1383}
1384NOKPROBE_SYMBOL(kprobe_trace_func);
1385
1386/* Kretprobe handler */
1387static nokprobe_inline void
1388__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1389 struct pt_regs *regs,
1390 struct trace_event_file *trace_file)
1391{
1392 struct kretprobe_trace_entry_head *entry;
1393 struct trace_event_buffer fbuffer;
1394 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1395 int dsize;
1396
1397 WARN_ON(call != trace_file->event_call);
1398
1399 if (trace_trigger_soft_disabled(trace_file))
1400 return;
1401
1402 dsize = __get_data_size(&tk->tp, regs);
1403
1404 entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1405 sizeof(*entry) + tk->tp.size + dsize);
1406 if (!entry)
1407 return;
1408
1409 fbuffer.regs = regs;
1410 entry->func = (unsigned long)tk->rp.kp.addr;
1411 entry->ret_ip = get_kretprobe_retaddr(ri);
1412 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1413
1414 trace_event_buffer_commit(&fbuffer);
1415}
1416
1417static void
1418kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1419 struct pt_regs *regs)
1420{
1421 struct event_file_link *link;
1422
1423 trace_probe_for_each_link_rcu(link, &tk->tp)
1424 __kretprobe_trace_func(tk, ri, regs, link->file);
1425}
1426NOKPROBE_SYMBOL(kretprobe_trace_func);
1427
1428/* Event entry printers */
1429static enum print_line_t
1430print_kprobe_event(struct trace_iterator *iter, int flags,
1431 struct trace_event *event)
1432{
1433 struct kprobe_trace_entry_head *field;
1434 struct trace_seq *s = &iter->seq;
1435 struct trace_probe *tp;
1436
1437 field = (struct kprobe_trace_entry_head *)iter->ent;
1438 tp = trace_probe_primary_from_call(
1439 container_of(event, struct trace_event_call, event));
1440 if (WARN_ON_ONCE(!tp))
1441 goto out;
1442
1443 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1444
1445 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1446 goto out;
1447
1448 trace_seq_putc(s, ')');
1449
1450 if (trace_probe_print_args(s, tp->args, tp->nr_args,
1451 (u8 *)&field[1], field) < 0)
1452 goto out;
1453
1454 trace_seq_putc(s, '\n');
1455 out:
1456 return trace_handle_return(s);
1457}
1458
1459static enum print_line_t
1460print_kretprobe_event(struct trace_iterator *iter, int flags,
1461 struct trace_event *event)
1462{
1463 struct kretprobe_trace_entry_head *field;
1464 struct trace_seq *s = &iter->seq;
1465 struct trace_probe *tp;
1466
1467 field = (struct kretprobe_trace_entry_head *)iter->ent;
1468 tp = trace_probe_primary_from_call(
1469 container_of(event, struct trace_event_call, event));
1470 if (WARN_ON_ONCE(!tp))
1471 goto out;
1472
1473 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1474
1475 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1476 goto out;
1477
1478 trace_seq_puts(s, " <- ");
1479
1480 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1481 goto out;
1482
1483 trace_seq_putc(s, ')');
1484
1485 if (trace_probe_print_args(s, tp->args, tp->nr_args,
1486 (u8 *)&field[1], field) < 0)
1487 goto out;
1488
1489 trace_seq_putc(s, '\n');
1490
1491 out:
1492 return trace_handle_return(s);
1493}
1494
1495
1496static int kprobe_event_define_fields(struct trace_event_call *event_call)
1497{
1498 int ret;
1499 struct kprobe_trace_entry_head field;
1500 struct trace_probe *tp;
1501
1502 tp = trace_probe_primary_from_call(event_call);
1503 if (WARN_ON_ONCE(!tp))
1504 return -ENOENT;
1505
1506 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1507
1508 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1509}
1510
1511static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1512{
1513 int ret;
1514 struct kretprobe_trace_entry_head field;
1515 struct trace_probe *tp;
1516
1517 tp = trace_probe_primary_from_call(event_call);
1518 if (WARN_ON_ONCE(!tp))
1519 return -ENOENT;
1520
1521 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1522 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1523
1524 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1525}
1526
1527#ifdef CONFIG_PERF_EVENTS
1528
1529/* Kprobe profile handler */
1530static int
1531kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1532{
1533 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1534 struct kprobe_trace_entry_head *entry;
1535 struct hlist_head *head;
1536 int size, __size, dsize;
1537 int rctx;
1538
1539 if (bpf_prog_array_valid(call)) {
1540 unsigned long orig_ip = instruction_pointer(regs);
1541 int ret;
1542
1543 ret = trace_call_bpf(call, regs);
1544
1545 /*
1546 * We need to check and see if we modified the pc of the
1547 * pt_regs, and if so return 1 so that we don't do the
1548 * single stepping.
1549 */
1550 if (orig_ip != instruction_pointer(regs))
1551 return 1;
1552 if (!ret)
1553 return 0;
1554 }
1555
1556 head = this_cpu_ptr(call->perf_events);
1557 if (hlist_empty(head))
1558 return 0;
1559
1560 dsize = __get_data_size(&tk->tp, regs);
1561 __size = sizeof(*entry) + tk->tp.size + dsize;
1562 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1563 size -= sizeof(u32);
1564
1565 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1566 if (!entry)
1567 return 0;
1568
1569 entry->ip = (unsigned long)tk->rp.kp.addr;
1570 memset(&entry[1], 0, dsize);
1571 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1572 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1573 head, NULL);
1574 return 0;
1575}
1576NOKPROBE_SYMBOL(kprobe_perf_func);
1577
1578/* Kretprobe profile handler */
1579static void
1580kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1581 struct pt_regs *regs)
1582{
1583 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1584 struct kretprobe_trace_entry_head *entry;
1585 struct hlist_head *head;
1586 int size, __size, dsize;
1587 int rctx;
1588
1589 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1590 return;
1591
1592 head = this_cpu_ptr(call->perf_events);
1593 if (hlist_empty(head))
1594 return;
1595
1596 dsize = __get_data_size(&tk->tp, regs);
1597 __size = sizeof(*entry) + tk->tp.size + dsize;
1598 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1599 size -= sizeof(u32);
1600
1601 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1602 if (!entry)
1603 return;
1604
1605 entry->func = (unsigned long)tk->rp.kp.addr;
1606 entry->ret_ip = get_kretprobe_retaddr(ri);
1607 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1608 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1609 head, NULL);
1610}
1611NOKPROBE_SYMBOL(kretprobe_perf_func);
1612
1613int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1614 const char **symbol, u64 *probe_offset,
1615 u64 *probe_addr, unsigned long *missed,
1616 bool perf_type_tracepoint)
1617{
1618 const char *pevent = trace_event_name(event->tp_event);
1619 const char *group = event->tp_event->class->system;
1620 struct trace_kprobe *tk;
1621
1622 if (perf_type_tracepoint)
1623 tk = find_trace_kprobe(pevent, group);
1624 else
1625 tk = trace_kprobe_primary_from_call(event->tp_event);
1626 if (!tk)
1627 return -EINVAL;
1628
1629 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1630 : BPF_FD_TYPE_KPROBE;
1631 *probe_offset = tk->rp.kp.offset;
1632 *probe_addr = kallsyms_show_value(current_cred()) ?
1633 (unsigned long)tk->rp.kp.addr : 0;
1634 *symbol = tk->symbol;
1635 if (missed)
1636 *missed = trace_kprobe_missed(tk);
1637 return 0;
1638}
1639#endif /* CONFIG_PERF_EVENTS */
1640
1641/*
1642 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1643 *
1644 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1645 * lockless, but we can't race with this __init function.
1646 */
1647static int kprobe_register(struct trace_event_call *event,
1648 enum trace_reg type, void *data)
1649{
1650 struct trace_event_file *file = data;
1651
1652 switch (type) {
1653 case TRACE_REG_REGISTER:
1654 return enable_trace_kprobe(event, file);
1655 case TRACE_REG_UNREGISTER:
1656 return disable_trace_kprobe(event, file);
1657
1658#ifdef CONFIG_PERF_EVENTS
1659 case TRACE_REG_PERF_REGISTER:
1660 return enable_trace_kprobe(event, NULL);
1661 case TRACE_REG_PERF_UNREGISTER:
1662 return disable_trace_kprobe(event, NULL);
1663 case TRACE_REG_PERF_OPEN:
1664 case TRACE_REG_PERF_CLOSE:
1665 case TRACE_REG_PERF_ADD:
1666 case TRACE_REG_PERF_DEL:
1667 return 0;
1668#endif
1669 }
1670 return 0;
1671}
1672
1673static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1674{
1675 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1676 int ret = 0;
1677
1678 raw_cpu_inc(*tk->nhit);
1679
1680 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1681 kprobe_trace_func(tk, regs);
1682#ifdef CONFIG_PERF_EVENTS
1683 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1684 ret = kprobe_perf_func(tk, regs);
1685#endif
1686 return ret;
1687}
1688NOKPROBE_SYMBOL(kprobe_dispatcher);
1689
1690static int
1691kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1692{
1693 struct kretprobe *rp = get_kretprobe(ri);
1694 struct trace_kprobe *tk;
1695
1696 /*
1697 * There is a small chance that get_kretprobe(ri) returns NULL when
1698 * the kretprobe is unregister on another CPU between kretprobe's
1699 * trampoline_handler and this function.
1700 */
1701 if (unlikely(!rp))
1702 return 0;
1703
1704 tk = container_of(rp, struct trace_kprobe, rp);
1705 raw_cpu_inc(*tk->nhit);
1706
1707 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1708 kretprobe_trace_func(tk, ri, regs);
1709#ifdef CONFIG_PERF_EVENTS
1710 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1711 kretprobe_perf_func(tk, ri, regs);
1712#endif
1713 return 0; /* We don't tweak kernel, so just return 0 */
1714}
1715NOKPROBE_SYMBOL(kretprobe_dispatcher);
1716
1717static struct trace_event_functions kretprobe_funcs = {
1718 .trace = print_kretprobe_event
1719};
1720
1721static struct trace_event_functions kprobe_funcs = {
1722 .trace = print_kprobe_event
1723};
1724
1725static struct trace_event_fields kretprobe_fields_array[] = {
1726 { .type = TRACE_FUNCTION_TYPE,
1727 .define_fields = kretprobe_event_define_fields },
1728 {}
1729};
1730
1731static struct trace_event_fields kprobe_fields_array[] = {
1732 { .type = TRACE_FUNCTION_TYPE,
1733 .define_fields = kprobe_event_define_fields },
1734 {}
1735};
1736
1737static inline void init_trace_event_call(struct trace_kprobe *tk)
1738{
1739 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1740
1741 if (trace_kprobe_is_return(tk)) {
1742 call->event.funcs = &kretprobe_funcs;
1743 call->class->fields_array = kretprobe_fields_array;
1744 } else {
1745 call->event.funcs = &kprobe_funcs;
1746 call->class->fields_array = kprobe_fields_array;
1747 }
1748
1749 call->flags = TRACE_EVENT_FL_KPROBE;
1750 call->class->reg = kprobe_register;
1751}
1752
1753static int register_kprobe_event(struct trace_kprobe *tk)
1754{
1755 init_trace_event_call(tk);
1756
1757 return trace_probe_register_event_call(&tk->tp);
1758}
1759
1760static int unregister_kprobe_event(struct trace_kprobe *tk)
1761{
1762 return trace_probe_unregister_event_call(&tk->tp);
1763}
1764
1765#ifdef CONFIG_PERF_EVENTS
1766
1767/* create a trace_kprobe, but don't add it to global lists */
1768struct trace_event_call *
1769create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1770 bool is_return)
1771{
1772 enum probe_print_type ptype;
1773 struct trace_kprobe *tk;
1774 int ret;
1775 char *event;
1776
1777 if (func) {
1778 unsigned int count;
1779
1780 count = number_of_same_symbols(func);
1781 if (count > 1)
1782 /*
1783 * Users should use addr to remove the ambiguity of
1784 * using func only.
1785 */
1786 return ERR_PTR(-EADDRNOTAVAIL);
1787 else if (count == 0)
1788 /*
1789 * We can return ENOENT earlier than when register the
1790 * kprobe.
1791 */
1792 return ERR_PTR(-ENOENT);
1793 }
1794
1795 /*
1796 * local trace_kprobes are not added to dyn_event, so they are never
1797 * searched in find_trace_kprobe(). Therefore, there is no concern of
1798 * duplicated name here.
1799 */
1800 event = func ? func : "DUMMY_EVENT";
1801
1802 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1803 offs, 0 /* maxactive */, 0 /* nargs */,
1804 is_return);
1805
1806 if (IS_ERR(tk)) {
1807 pr_info("Failed to allocate trace_probe.(%d)\n",
1808 (int)PTR_ERR(tk));
1809 return ERR_CAST(tk);
1810 }
1811
1812 init_trace_event_call(tk);
1813
1814 ptype = trace_kprobe_is_return(tk) ?
1815 PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1816 if (traceprobe_set_print_fmt(&tk->tp, ptype) < 0) {
1817 ret = -ENOMEM;
1818 goto error;
1819 }
1820
1821 ret = __register_trace_kprobe(tk);
1822 if (ret < 0)
1823 goto error;
1824
1825 return trace_probe_event_call(&tk->tp);
1826error:
1827 free_trace_kprobe(tk);
1828 return ERR_PTR(ret);
1829}
1830
1831void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1832{
1833 struct trace_kprobe *tk;
1834
1835 tk = trace_kprobe_primary_from_call(event_call);
1836 if (unlikely(!tk))
1837 return;
1838
1839 if (trace_probe_is_enabled(&tk->tp)) {
1840 WARN_ON(1);
1841 return;
1842 }
1843
1844 __unregister_trace_kprobe(tk);
1845
1846 free_trace_kprobe(tk);
1847}
1848#endif /* CONFIG_PERF_EVENTS */
1849
1850static __init void enable_boot_kprobe_events(void)
1851{
1852 struct trace_array *tr = top_trace_array();
1853 struct trace_event_file *file;
1854 struct trace_kprobe *tk;
1855 struct dyn_event *pos;
1856
1857 mutex_lock(&event_mutex);
1858 for_each_trace_kprobe(tk, pos) {
1859 list_for_each_entry(file, &tr->events, list)
1860 if (file->event_call == trace_probe_event_call(&tk->tp))
1861 trace_event_enable_disable(file, 1, 0);
1862 }
1863 mutex_unlock(&event_mutex);
1864}
1865
1866static __init void setup_boot_kprobe_events(void)
1867{
1868 char *p, *cmd = kprobe_boot_events_buf;
1869 int ret;
1870
1871 strreplace(kprobe_boot_events_buf, ',', ' ');
1872
1873 while (cmd && *cmd != '\0') {
1874 p = strchr(cmd, ';');
1875 if (p)
1876 *p++ = '\0';
1877
1878 ret = create_or_delete_trace_kprobe(cmd);
1879 if (ret)
1880 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1881
1882 cmd = p;
1883 }
1884
1885 enable_boot_kprobe_events();
1886}
1887
1888/*
1889 * Register dynevent at core_initcall. This allows kernel to setup kprobe
1890 * events in postcore_initcall without tracefs.
1891 */
1892static __init int init_kprobe_trace_early(void)
1893{
1894 int ret;
1895
1896 ret = dyn_event_register(&trace_kprobe_ops);
1897 if (ret)
1898 return ret;
1899
1900 if (register_module_notifier(&trace_kprobe_module_nb))
1901 return -EINVAL;
1902
1903 return 0;
1904}
1905core_initcall(init_kprobe_trace_early);
1906
1907/* Make a tracefs interface for controlling probe points */
1908static __init int init_kprobe_trace(void)
1909{
1910 int ret;
1911
1912 ret = tracing_init_dentry();
1913 if (ret)
1914 return 0;
1915
1916 /* Event list interface */
1917 trace_create_file("kprobe_events", TRACE_MODE_WRITE,
1918 NULL, NULL, &kprobe_events_ops);
1919
1920 /* Profile interface */
1921 trace_create_file("kprobe_profile", TRACE_MODE_READ,
1922 NULL, NULL, &kprobe_profile_ops);
1923
1924 setup_boot_kprobe_events();
1925
1926 return 0;
1927}
1928fs_initcall(init_kprobe_trace);
1929
1930
1931#ifdef CONFIG_FTRACE_STARTUP_TEST
1932static __init struct trace_event_file *
1933find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1934{
1935 struct trace_event_file *file;
1936
1937 list_for_each_entry(file, &tr->events, list)
1938 if (file->event_call == trace_probe_event_call(&tk->tp))
1939 return file;
1940
1941 return NULL;
1942}
1943
1944/*
1945 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1946 * stage, we can do this lockless.
1947 */
1948static __init int kprobe_trace_self_tests_init(void)
1949{
1950 int ret, warn = 0;
1951 int (*target)(int, int, int, int, int, int);
1952 struct trace_kprobe *tk;
1953 struct trace_event_file *file;
1954
1955 if (tracing_is_disabled())
1956 return -ENODEV;
1957
1958 if (tracing_selftest_disabled)
1959 return 0;
1960
1961 target = kprobe_trace_selftest_target;
1962
1963 pr_info("Testing kprobe tracing: ");
1964
1965 ret = create_or_delete_trace_kprobe("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)");
1966 if (WARN_ON_ONCE(ret)) {
1967 pr_warn("error on probing function entry.\n");
1968 warn++;
1969 } else {
1970 /* Enable trace point */
1971 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1972 if (WARN_ON_ONCE(tk == NULL)) {
1973 pr_warn("error on getting new probe.\n");
1974 warn++;
1975 } else {
1976 file = find_trace_probe_file(tk, top_trace_array());
1977 if (WARN_ON_ONCE(file == NULL)) {
1978 pr_warn("error on getting probe file.\n");
1979 warn++;
1980 } else
1981 enable_trace_kprobe(
1982 trace_probe_event_call(&tk->tp), file);
1983 }
1984 }
1985
1986 ret = create_or_delete_trace_kprobe("r:testprobe2 kprobe_trace_selftest_target $retval");
1987 if (WARN_ON_ONCE(ret)) {
1988 pr_warn("error on probing function return.\n");
1989 warn++;
1990 } else {
1991 /* Enable trace point */
1992 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1993 if (WARN_ON_ONCE(tk == NULL)) {
1994 pr_warn("error on getting 2nd new probe.\n");
1995 warn++;
1996 } else {
1997 file = find_trace_probe_file(tk, top_trace_array());
1998 if (WARN_ON_ONCE(file == NULL)) {
1999 pr_warn("error on getting probe file.\n");
2000 warn++;
2001 } else
2002 enable_trace_kprobe(
2003 trace_probe_event_call(&tk->tp), file);
2004 }
2005 }
2006
2007 if (warn)
2008 goto end;
2009
2010 ret = target(1, 2, 3, 4, 5, 6);
2011
2012 /*
2013 * Not expecting an error here, the check is only to prevent the
2014 * optimizer from removing the call to target() as otherwise there
2015 * are no side-effects and the call is never performed.
2016 */
2017 if (ret != 21)
2018 warn++;
2019
2020 /* Disable trace points before removing it */
2021 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2022 if (WARN_ON_ONCE(tk == NULL)) {
2023 pr_warn("error on getting test probe.\n");
2024 warn++;
2025 } else {
2026 if (trace_kprobe_nhit(tk) != 1) {
2027 pr_warn("incorrect number of testprobe hits\n");
2028 warn++;
2029 }
2030
2031 file = find_trace_probe_file(tk, top_trace_array());
2032 if (WARN_ON_ONCE(file == NULL)) {
2033 pr_warn("error on getting probe file.\n");
2034 warn++;
2035 } else
2036 disable_trace_kprobe(
2037 trace_probe_event_call(&tk->tp), file);
2038 }
2039
2040 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2041 if (WARN_ON_ONCE(tk == NULL)) {
2042 pr_warn("error on getting 2nd test probe.\n");
2043 warn++;
2044 } else {
2045 if (trace_kprobe_nhit(tk) != 1) {
2046 pr_warn("incorrect number of testprobe2 hits\n");
2047 warn++;
2048 }
2049
2050 file = find_trace_probe_file(tk, top_trace_array());
2051 if (WARN_ON_ONCE(file == NULL)) {
2052 pr_warn("error on getting probe file.\n");
2053 warn++;
2054 } else
2055 disable_trace_kprobe(
2056 trace_probe_event_call(&tk->tp), file);
2057 }
2058
2059 ret = create_or_delete_trace_kprobe("-:testprobe");
2060 if (WARN_ON_ONCE(ret)) {
2061 pr_warn("error on deleting a probe.\n");
2062 warn++;
2063 }
2064
2065 ret = create_or_delete_trace_kprobe("-:testprobe2");
2066 if (WARN_ON_ONCE(ret)) {
2067 pr_warn("error on deleting a probe.\n");
2068 warn++;
2069 }
2070
2071end:
2072 ret = dyn_events_release_all(&trace_kprobe_ops);
2073 if (WARN_ON_ONCE(ret)) {
2074 pr_warn("error on cleaning up probes.\n");
2075 warn++;
2076 }
2077 /*
2078 * Wait for the optimizer work to finish. Otherwise it might fiddle
2079 * with probes in already freed __init text.
2080 */
2081 wait_for_kprobe_optimizer();
2082 if (warn)
2083 pr_cont("NG: Some tests are failed. Please check them.\n");
2084 else
2085 pr_cont("OK\n");
2086 return 0;
2087}
2088
2089late_initcall(kprobe_trace_self_tests_init);
2090
2091#endif