2 * Kprobes-based tracing events
4 * Created by Masami Hiramatsu <mhiramat@redhat.com>
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.
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.
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
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>
34 #include "trace_output.h"
36 #define MAX_TRACE_ARGS 128
37 #define MAX_ARGSTR_LEN 63
38 #define MAX_EVENT_NAME_LEN 64
39 #define KPROBE_EVENT_SYSTEM "kprobes"
41 /* Reserved field names */
42 #define FIELD_STRING_IP "__probe_ip"
43 #define FIELD_STRING_NARGS "__probe_nargs"
44 #define FIELD_STRING_RETIP "__probe_ret_ip"
45 #define FIELD_STRING_FUNC "__probe_func"
47 const char *reserved_field_names
[] = {
50 "common_preempt_count",
61 unsigned long (*func
)(struct pt_regs
*, void *);
65 static __kprobes
unsigned long call_fetch(struct fetch_func
*f
,
68 return f
->func(regs
, f
->data
);
72 static __kprobes
unsigned long fetch_register(struct pt_regs
*regs
,
75 return regs_get_register(regs
, (unsigned int)((unsigned long)offset
));
78 static __kprobes
unsigned long fetch_stack(struct pt_regs
*regs
,
81 return regs_get_kernel_stack_nth(regs
,
82 (unsigned int)((unsigned long)num
));
85 static __kprobes
unsigned long fetch_memory(struct pt_regs
*regs
, void *addr
)
89 if (probe_kernel_address(addr
, retval
))
94 static __kprobes
unsigned long fetch_argument(struct pt_regs
*regs
, void *num
)
96 return regs_get_argument_nth(regs
, (unsigned int)((unsigned long)num
));
99 static __kprobes
unsigned long fetch_retvalue(struct pt_regs
*regs
,
102 return regs_return_value(regs
);
105 static __kprobes
unsigned long fetch_stack_address(struct pt_regs
*regs
,
108 return kernel_stack_pointer(regs
);
111 /* Memory fetching by symbol */
112 struct symbol_cache
{
118 static unsigned long update_symbol_cache(struct symbol_cache
*sc
)
120 sc
->addr
= (unsigned long)kallsyms_lookup_name(sc
->symbol
);
122 sc
->addr
+= sc
->offset
;
126 static void free_symbol_cache(struct symbol_cache
*sc
)
132 static struct symbol_cache
*alloc_symbol_cache(const char *sym
, long offset
)
134 struct symbol_cache
*sc
;
136 if (!sym
|| strlen(sym
) == 0)
138 sc
= kzalloc(sizeof(struct symbol_cache
), GFP_KERNEL
);
142 sc
->symbol
= kstrdup(sym
, GFP_KERNEL
);
149 update_symbol_cache(sc
);
153 static __kprobes
unsigned long fetch_symbol(struct pt_regs
*regs
, void *data
)
155 struct symbol_cache
*sc
= data
;
158 return fetch_memory(regs
, (void *)sc
->addr
);
163 /* Special indirect memory access interface */
164 struct indirect_fetch_data
{
165 struct fetch_func orig
;
169 static __kprobes
unsigned long fetch_indirect(struct pt_regs
*regs
, void *data
)
171 struct indirect_fetch_data
*ind
= data
;
174 addr
= call_fetch(&ind
->orig
, regs
);
177 return fetch_memory(regs
, (void *)addr
);
182 static __kprobes
void free_indirect_fetch_data(struct indirect_fetch_data
*data
)
184 if (data
->orig
.func
== fetch_indirect
)
185 free_indirect_fetch_data(data
->orig
.data
);
186 else if (data
->orig
.func
== fetch_symbol
)
187 free_symbol_cache(data
->orig
.data
);
192 * Kprobe event core functions
196 struct fetch_func fetch
;
200 /* Flags for trace_probe */
201 #define TP_FLAG_TRACE 1
202 #define TP_FLAG_PROFILE 2
205 struct list_head list
;
206 struct kretprobe rp
; /* Use rp.kp for kprobe use */
208 unsigned int flags
; /* For TP_FLAG_* */
209 const char *symbol
; /* symbol name */
210 struct ftrace_event_call call
;
211 struct trace_event event
;
212 unsigned int nr_args
;
213 struct probe_arg args
[];
216 #define SIZEOF_TRACE_PROBE(n) \
217 (offsetof(struct trace_probe, args) + \
218 (sizeof(struct probe_arg) * (n)))
220 static __kprobes
int probe_is_return(struct trace_probe
*tp
)
222 return tp
->rp
.handler
!= NULL
;
225 static __kprobes
const char *probe_symbol(struct trace_probe
*tp
)
227 return tp
->symbol
? tp
->symbol
: "unknown";
230 static int probe_arg_string(char *buf
, size_t n
, struct fetch_func
*ff
)
234 if (ff
->func
== fetch_argument
)
235 ret
= snprintf(buf
, n
, "$arg%lu", (unsigned long)ff
->data
);
236 else if (ff
->func
== fetch_register
) {
238 name
= regs_query_register_name((unsigned int)((long)ff
->data
));
239 ret
= snprintf(buf
, n
, "%%%s", name
);
240 } else if (ff
->func
== fetch_stack
)
241 ret
= snprintf(buf
, n
, "$stack%lu", (unsigned long)ff
->data
);
242 else if (ff
->func
== fetch_memory
)
243 ret
= snprintf(buf
, n
, "@0x%p", ff
->data
);
244 else if (ff
->func
== fetch_symbol
) {
245 struct symbol_cache
*sc
= ff
->data
;
246 ret
= snprintf(buf
, n
, "@%s%+ld", sc
->symbol
, sc
->offset
);
247 } else if (ff
->func
== fetch_retvalue
)
248 ret
= snprintf(buf
, n
, "$retval");
249 else if (ff
->func
== fetch_stack_address
)
250 ret
= snprintf(buf
, n
, "$stack");
251 else if (ff
->func
== fetch_indirect
) {
252 struct indirect_fetch_data
*id
= ff
->data
;
254 ret
= snprintf(buf
, n
, "%+ld(", id
->offset
);
258 ret
= probe_arg_string(buf
+ l
, n
- l
, &id
->orig
);
262 ret
= snprintf(buf
+ l
, n
- l
, ")");
271 static int register_probe_event(struct trace_probe
*tp
);
272 static void unregister_probe_event(struct trace_probe
*tp
);
274 static DEFINE_MUTEX(probe_lock
);
275 static LIST_HEAD(probe_list
);
277 static int kprobe_dispatcher(struct kprobe
*kp
, struct pt_regs
*regs
);
278 static int kretprobe_dispatcher(struct kretprobe_instance
*ri
,
279 struct pt_regs
*regs
);
282 * Allocate new trace_probe and initialize it (including kprobes).
284 static struct trace_probe
*alloc_trace_probe(const char *group
,
289 int nargs
, int is_return
)
291 struct trace_probe
*tp
;
293 tp
= kzalloc(SIZEOF_TRACE_PROBE(nargs
), GFP_KERNEL
);
295 return ERR_PTR(-ENOMEM
);
298 tp
->symbol
= kstrdup(symbol
, GFP_KERNEL
);
301 tp
->rp
.kp
.symbol_name
= tp
->symbol
;
302 tp
->rp
.kp
.offset
= offs
;
304 tp
->rp
.kp
.addr
= addr
;
307 tp
->rp
.handler
= kretprobe_dispatcher
;
309 tp
->rp
.kp
.pre_handler
= kprobe_dispatcher
;
313 tp
->call
.name
= kstrdup(event
, GFP_KERNEL
);
319 tp
->call
.system
= kstrdup(group
, GFP_KERNEL
);
320 if (!tp
->call
.system
)
323 INIT_LIST_HEAD(&tp
->list
);
326 kfree(tp
->call
.name
);
329 return ERR_PTR(-ENOMEM
);
332 static void free_probe_arg(struct probe_arg
*arg
)
334 if (arg
->fetch
.func
== fetch_symbol
)
335 free_symbol_cache(arg
->fetch
.data
);
336 else if (arg
->fetch
.func
== fetch_indirect
)
337 free_indirect_fetch_data(arg
->fetch
.data
);
341 static void free_trace_probe(struct trace_probe
*tp
)
345 for (i
= 0; i
< tp
->nr_args
; i
++)
346 free_probe_arg(&tp
->args
[i
]);
348 kfree(tp
->call
.system
);
349 kfree(tp
->call
.name
);
354 static struct trace_probe
*find_probe_event(const char *event
,
357 struct trace_probe
*tp
;
359 list_for_each_entry(tp
, &probe_list
, list
)
360 if (strcmp(tp
->call
.name
, event
) == 0 &&
361 strcmp(tp
->call
.system
, group
) == 0)
366 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
367 static void unregister_trace_probe(struct trace_probe
*tp
)
369 if (probe_is_return(tp
))
370 unregister_kretprobe(&tp
->rp
);
372 unregister_kprobe(&tp
->rp
.kp
);
374 unregister_probe_event(tp
);
377 /* Register a trace_probe and probe_event */
378 static int register_trace_probe(struct trace_probe
*tp
)
380 struct trace_probe
*old_tp
;
383 mutex_lock(&probe_lock
);
385 /* register as an event */
386 old_tp
= find_probe_event(tp
->call
.name
, tp
->call
.system
);
388 /* delete old event */
389 unregister_trace_probe(old_tp
);
390 free_trace_probe(old_tp
);
392 ret
= register_probe_event(tp
);
394 pr_warning("Faild to register probe event(%d)\n", ret
);
398 tp
->rp
.kp
.flags
|= KPROBE_FLAG_DISABLED
;
399 if (probe_is_return(tp
))
400 ret
= register_kretprobe(&tp
->rp
);
402 ret
= register_kprobe(&tp
->rp
.kp
);
405 pr_warning("Could not insert probe(%d)\n", ret
);
406 if (ret
== -EILSEQ
) {
407 pr_warning("Probing address(0x%p) is not an "
408 "instruction boundary.\n",
412 unregister_probe_event(tp
);
414 list_add_tail(&tp
->list
, &probe_list
);
416 mutex_unlock(&probe_lock
);
420 /* Split symbol and offset. */
421 static int split_symbol_offset(char *symbol
, unsigned long *offset
)
429 tmp
= strchr(symbol
, '+');
431 /* skip sign because strict_strtol doesn't accept '+' */
432 ret
= strict_strtoul(tmp
+ 1, 0, offset
);
441 #define PARAM_MAX_ARGS 16
442 #define PARAM_MAX_STACK (THREAD_SIZE / sizeof(unsigned long))
444 static int parse_probe_vars(char *arg
, struct fetch_func
*ff
, int is_return
)
449 if (strcmp(arg
, "retval") == 0) {
451 ff
->func
= fetch_retvalue
;
455 } else if (strncmp(arg
, "stack", 5) == 0) {
456 if (arg
[5] == '\0') {
457 ff
->func
= fetch_stack_address
;
459 } else if (isdigit(arg
[5])) {
460 ret
= strict_strtoul(arg
+ 5, 10, ¶m
);
461 if (ret
|| param
> PARAM_MAX_STACK
)
464 ff
->func
= fetch_stack
;
465 ff
->data
= (void *)param
;
469 } else if (strncmp(arg
, "arg", 3) == 0 && isdigit(arg
[3])) {
470 ret
= strict_strtoul(arg
+ 3, 10, ¶m
);
471 if (ret
|| param
> PARAM_MAX_ARGS
)
474 ff
->func
= fetch_argument
;
475 ff
->data
= (void *)param
;
482 static int parse_probe_arg(char *arg
, struct fetch_func
*ff
, int is_return
)
491 ret
= parse_probe_vars(arg
+ 1, ff
, is_return
);
493 case '%': /* named register */
494 ret
= regs_query_register_offset(arg
+ 1);
496 ff
->func
= fetch_register
;
497 ff
->data
= (void *)(unsigned long)ret
;
501 case '@': /* memory or symbol */
502 if (isdigit(arg
[1])) {
503 ret
= strict_strtoul(arg
+ 1, 0, ¶m
);
506 ff
->func
= fetch_memory
;
507 ff
->data
= (void *)param
;
509 ret
= split_symbol_offset(arg
+ 1, &offset
);
512 ff
->data
= alloc_symbol_cache(arg
+ 1, offset
);
514 ff
->func
= fetch_symbol
;
519 case '+': /* indirect memory */
521 tmp
= strchr(arg
, '(');
527 ret
= strict_strtol(arg
+ 1, 0, &offset
);
533 tmp
= strrchr(arg
, ')');
535 struct indirect_fetch_data
*id
;
537 id
= kzalloc(sizeof(struct indirect_fetch_data
),
542 ret
= parse_probe_arg(arg
, &id
->orig
, is_return
);
546 ff
->func
= fetch_indirect
;
547 ff
->data
= (void *)id
;
553 /* TODO: support custom handler */
559 /* Return 1 if name is reserved or already used by another argument */
560 static int conflict_field_name(const char *name
,
561 struct probe_arg
*args
, int narg
)
564 for (i
= 0; i
< ARRAY_SIZE(reserved_field_names
); i
++)
565 if (strcmp(reserved_field_names
[i
], name
) == 0)
567 for (i
= 0; i
< narg
; i
++)
568 if (strcmp(args
[i
].name
, name
) == 0)
573 static int create_trace_probe(int argc
, char **argv
)
577 * - Add kprobe: p[:[GRP/]EVENT] KSYM[+OFFS]|KADDR [FETCHARGS]
578 * - Add kretprobe: r[:[GRP/]EVENT] KSYM[+0] [FETCHARGS]
580 * $argN : fetch Nth of function argument. (N:0-)
581 * $retval : fetch return value
582 * $stack : fetch stack address
583 * $stackN : fetch Nth of stack (N:0-)
584 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
585 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
586 * %REG : fetch register REG
587 * Indirect memory fetch:
588 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
589 * Alias name of args:
590 * NAME=FETCHARG : set NAME as alias of FETCHARG.
592 struct trace_probe
*tp
;
595 char *symbol
= NULL
, *event
= NULL
, *arg
= NULL
, *group
= NULL
;
596 unsigned long offset
= 0;
598 char buf
[MAX_EVENT_NAME_LEN
];
601 pr_info("Probe point is not specified.\n");
605 if (argv
[0][0] == 'p')
607 else if (argv
[0][0] == 'r')
610 pr_info("Probe definition must be started with 'p' or 'r'.\n");
614 if (argv
[0][1] == ':') {
616 if (strchr(event
, '/')) {
618 event
= strchr(group
, '/') + 1;
620 if (strlen(group
) == 0) {
621 pr_info("Group name is not specifiled\n");
625 if (strlen(event
) == 0) {
626 pr_info("Event name is not specifiled\n");
631 if (isdigit(argv
[1][0])) {
633 pr_info("Return probe point must be a symbol.\n");
636 /* an address specified */
637 ret
= strict_strtoul(&argv
[0][2], 0, (unsigned long *)&addr
);
639 pr_info("Failed to parse address.\n");
643 /* a symbol specified */
645 /* TODO: support .init module functions */
646 ret
= split_symbol_offset(symbol
, &offset
);
648 pr_info("Failed to parse symbol.\n");
651 if (offset
&& is_return
) {
652 pr_info("Return probe must be used without offset.\n");
656 argc
-= 2; argv
+= 2;
660 group
= KPROBE_EVENT_SYSTEM
;
662 /* Make a new event name */
664 snprintf(buf
, MAX_EVENT_NAME_LEN
, "%c@%s%+ld",
665 is_return
? 'r' : 'p', symbol
, offset
);
667 snprintf(buf
, MAX_EVENT_NAME_LEN
, "%c@0x%p",
668 is_return
? 'r' : 'p', addr
);
671 tp
= alloc_trace_probe(group
, event
, addr
, symbol
, offset
, argc
,
674 pr_info("Failed to allocate trace_probe.(%d)\n",
679 /* parse arguments */
681 for (i
= 0; i
< argc
&& i
< MAX_TRACE_ARGS
; i
++) {
682 /* Parse argument name */
683 arg
= strchr(argv
[i
], '=');
689 if (conflict_field_name(argv
[i
], tp
->args
, i
)) {
690 pr_info("Argument%d name '%s' conflicts with "
691 "another field.\n", i
, argv
[i
]);
696 tp
->args
[i
].name
= kstrdup(argv
[i
], GFP_KERNEL
);
698 /* Parse fetch argument */
699 if (strlen(arg
) > MAX_ARGSTR_LEN
) {
700 pr_info("Argument%d(%s) is too long.\n", i
, arg
);
704 ret
= parse_probe_arg(arg
, &tp
->args
[i
].fetch
, is_return
);
706 pr_info("Parse error at argument%d. (%d)\n", i
, ret
);
707 kfree(tp
->args
[i
].name
);
714 ret
= register_trace_probe(tp
);
720 free_trace_probe(tp
);
724 static void cleanup_all_probes(void)
726 struct trace_probe
*tp
;
728 mutex_lock(&probe_lock
);
729 /* TODO: Use batch unregistration */
730 while (!list_empty(&probe_list
)) {
731 tp
= list_entry(probe_list
.next
, struct trace_probe
, list
);
732 unregister_trace_probe(tp
);
733 free_trace_probe(tp
);
735 mutex_unlock(&probe_lock
);
739 /* Probes listing interfaces */
740 static void *probes_seq_start(struct seq_file
*m
, loff_t
*pos
)
742 mutex_lock(&probe_lock
);
743 return seq_list_start(&probe_list
, *pos
);
746 static void *probes_seq_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
748 return seq_list_next(v
, &probe_list
, pos
);
751 static void probes_seq_stop(struct seq_file
*m
, void *v
)
753 mutex_unlock(&probe_lock
);
756 static int probes_seq_show(struct seq_file
*m
, void *v
)
758 struct trace_probe
*tp
= v
;
760 char buf
[MAX_ARGSTR_LEN
+ 1];
762 seq_printf(m
, "%c", probe_is_return(tp
) ? 'r' : 'p');
763 seq_printf(m
, ":%s", tp
->call
.name
);
766 seq_printf(m
, " %s+%u", probe_symbol(tp
), tp
->rp
.kp
.offset
);
768 seq_printf(m
, " 0x%p", tp
->rp
.kp
.addr
);
770 for (i
= 0; i
< tp
->nr_args
; i
++) {
771 ret
= probe_arg_string(buf
, MAX_ARGSTR_LEN
, &tp
->args
[i
].fetch
);
773 pr_warning("Argument%d decoding error(%d).\n", i
, ret
);
776 seq_printf(m
, " %s=%s", tp
->args
[i
].name
, buf
);
782 static const struct seq_operations probes_seq_op
= {
783 .start
= probes_seq_start
,
784 .next
= probes_seq_next
,
785 .stop
= probes_seq_stop
,
786 .show
= probes_seq_show
789 static int probes_open(struct inode
*inode
, struct file
*file
)
791 if ((file
->f_mode
& FMODE_WRITE
) &&
792 (file
->f_flags
& O_TRUNC
))
793 cleanup_all_probes();
795 return seq_open(file
, &probes_seq_op
);
798 static int command_trace_probe(const char *buf
)
801 int argc
= 0, ret
= 0;
803 argv
= argv_split(GFP_KERNEL
, buf
, &argc
);
808 ret
= create_trace_probe(argc
, argv
);
814 #define WRITE_BUFSIZE 128
816 static ssize_t
probes_write(struct file
*file
, const char __user
*buffer
,
817 size_t count
, loff_t
*ppos
)
824 kbuf
= kmalloc(WRITE_BUFSIZE
, GFP_KERNEL
);
829 while (done
< count
) {
831 if (size
>= WRITE_BUFSIZE
)
832 size
= WRITE_BUFSIZE
- 1;
833 if (copy_from_user(kbuf
, buffer
+ done
, size
)) {
838 tmp
= strchr(kbuf
, '\n');
841 size
= tmp
- kbuf
+ 1;
842 } else if (done
+ size
< count
) {
843 pr_warning("Line length is too long: "
844 "Should be less than %d.", WRITE_BUFSIZE
);
849 /* Remove comments */
850 tmp
= strchr(kbuf
, '#');
854 ret
= command_trace_probe(kbuf
);
864 static const struct file_operations kprobe_events_ops
= {
865 .owner
= THIS_MODULE
,
869 .release
= seq_release
,
870 .write
= probes_write
,
873 /* Probes profiling interfaces */
874 static int probes_profile_seq_show(struct seq_file
*m
, void *v
)
876 struct trace_probe
*tp
= v
;
878 seq_printf(m
, " %-44s %15lu %15lu\n", tp
->call
.name
, tp
->nhit
,
884 static const struct seq_operations profile_seq_op
= {
885 .start
= probes_seq_start
,
886 .next
= probes_seq_next
,
887 .stop
= probes_seq_stop
,
888 .show
= probes_profile_seq_show
891 static int profile_open(struct inode
*inode
, struct file
*file
)
893 return seq_open(file
, &profile_seq_op
);
896 static const struct file_operations kprobe_profile_ops
= {
897 .owner
= THIS_MODULE
,
898 .open
= profile_open
,
901 .release
= seq_release
,
905 static __kprobes
int kprobe_trace_func(struct kprobe
*kp
, struct pt_regs
*regs
)
907 struct trace_probe
*tp
= container_of(kp
, struct trace_probe
, rp
.kp
);
908 struct kprobe_trace_entry
*entry
;
909 struct ring_buffer_event
*event
;
910 struct ring_buffer
*buffer
;
912 unsigned long irq_flags
;
913 struct ftrace_event_call
*call
= &tp
->call
;
917 local_save_flags(irq_flags
);
918 pc
= preempt_count();
920 size
= SIZEOF_KPROBE_TRACE_ENTRY(tp
->nr_args
);
922 event
= trace_current_buffer_lock_reserve(&buffer
, call
->id
, size
,
927 entry
= ring_buffer_event_data(event
);
928 entry
->nargs
= tp
->nr_args
;
929 entry
->ip
= (unsigned long)kp
->addr
;
930 for (i
= 0; i
< tp
->nr_args
; i
++)
931 entry
->args
[i
] = call_fetch(&tp
->args
[i
].fetch
, regs
);
933 if (!filter_current_check_discard(buffer
, call
, entry
, event
))
934 trace_nowake_buffer_unlock_commit(buffer
, event
, irq_flags
, pc
);
938 /* Kretprobe handler */
939 static __kprobes
int kretprobe_trace_func(struct kretprobe_instance
*ri
,
940 struct pt_regs
*regs
)
942 struct trace_probe
*tp
= container_of(ri
->rp
, struct trace_probe
, rp
);
943 struct kretprobe_trace_entry
*entry
;
944 struct ring_buffer_event
*event
;
945 struct ring_buffer
*buffer
;
947 unsigned long irq_flags
;
948 struct ftrace_event_call
*call
= &tp
->call
;
950 local_save_flags(irq_flags
);
951 pc
= preempt_count();
953 size
= SIZEOF_KRETPROBE_TRACE_ENTRY(tp
->nr_args
);
955 event
= trace_current_buffer_lock_reserve(&buffer
, call
->id
, size
,
960 entry
= ring_buffer_event_data(event
);
961 entry
->nargs
= tp
->nr_args
;
962 entry
->func
= (unsigned long)tp
->rp
.kp
.addr
;
963 entry
->ret_ip
= (unsigned long)ri
->ret_addr
;
964 for (i
= 0; i
< tp
->nr_args
; i
++)
965 entry
->args
[i
] = call_fetch(&tp
->args
[i
].fetch
, regs
);
967 if (!filter_current_check_discard(buffer
, call
, entry
, event
))
968 trace_nowake_buffer_unlock_commit(buffer
, event
, irq_flags
, pc
);
973 /* Event entry printers */
975 print_kprobe_event(struct trace_iterator
*iter
, int flags
)
977 struct kprobe_trace_entry
*field
;
978 struct trace_seq
*s
= &iter
->seq
;
979 struct trace_event
*event
;
980 struct trace_probe
*tp
;
983 field
= (struct kprobe_trace_entry
*)iter
->ent
;
984 event
= ftrace_find_event(field
->ent
.type
);
985 tp
= container_of(event
, struct trace_probe
, event
);
987 if (!trace_seq_printf(s
, "%s: (", tp
->call
.name
))
990 if (!seq_print_ip_sym(s
, field
->ip
, flags
| TRACE_ITER_SYM_OFFSET
))
993 if (!trace_seq_puts(s
, ")"))
996 for (i
= 0; i
< field
->nargs
; i
++)
997 if (!trace_seq_printf(s
, " %s=%lx",
998 tp
->args
[i
].name
, field
->args
[i
]))
1001 if (!trace_seq_puts(s
, "\n"))
1004 return TRACE_TYPE_HANDLED
;
1006 return TRACE_TYPE_PARTIAL_LINE
;
1010 print_kretprobe_event(struct trace_iterator
*iter
, int flags
)
1012 struct kretprobe_trace_entry
*field
;
1013 struct trace_seq
*s
= &iter
->seq
;
1014 struct trace_event
*event
;
1015 struct trace_probe
*tp
;
1018 field
= (struct kretprobe_trace_entry
*)iter
->ent
;
1019 event
= ftrace_find_event(field
->ent
.type
);
1020 tp
= container_of(event
, struct trace_probe
, event
);
1022 if (!trace_seq_printf(s
, "%s: (", tp
->call
.name
))
1025 if (!seq_print_ip_sym(s
, field
->ret_ip
, flags
| TRACE_ITER_SYM_OFFSET
))
1028 if (!trace_seq_puts(s
, " <- "))
1031 if (!seq_print_ip_sym(s
, field
->func
, flags
& ~TRACE_ITER_SYM_OFFSET
))
1034 if (!trace_seq_puts(s
, ")"))
1037 for (i
= 0; i
< field
->nargs
; i
++)
1038 if (!trace_seq_printf(s
, " %s=%lx",
1039 tp
->args
[i
].name
, field
->args
[i
]))
1042 if (!trace_seq_puts(s
, "\n"))
1045 return TRACE_TYPE_HANDLED
;
1047 return TRACE_TYPE_PARTIAL_LINE
;
1050 static int probe_event_enable(struct ftrace_event_call
*call
)
1052 struct trace_probe
*tp
= (struct trace_probe
*)call
->data
;
1054 tp
->flags
|= TP_FLAG_TRACE
;
1055 if (probe_is_return(tp
))
1056 return enable_kretprobe(&tp
->rp
);
1058 return enable_kprobe(&tp
->rp
.kp
);
1061 static void probe_event_disable(struct ftrace_event_call
*call
)
1063 struct trace_probe
*tp
= (struct trace_probe
*)call
->data
;
1065 tp
->flags
&= ~TP_FLAG_TRACE
;
1066 if (!(tp
->flags
& (TP_FLAG_TRACE
| TP_FLAG_PROFILE
))) {
1067 if (probe_is_return(tp
))
1068 disable_kretprobe(&tp
->rp
);
1070 disable_kprobe(&tp
->rp
.kp
);
1074 static int probe_event_raw_init(struct ftrace_event_call
*event_call
)
1076 INIT_LIST_HEAD(&event_call
->fields
);
1082 #define DEFINE_FIELD(type, item, name, is_signed) \
1084 ret = trace_define_field(event_call, #type, name, \
1085 offsetof(typeof(field), item), \
1086 sizeof(field.item), is_signed, \
1092 static int kprobe_event_define_fields(struct ftrace_event_call
*event_call
)
1095 struct kprobe_trace_entry field
;
1096 struct trace_probe
*tp
= (struct trace_probe
*)event_call
->data
;
1098 ret
= trace_define_common_fields(event_call
);
1102 DEFINE_FIELD(unsigned long, ip
, FIELD_STRING_IP
, 0);
1103 DEFINE_FIELD(int, nargs
, FIELD_STRING_NARGS
, 1);
1104 /* Set argument names as fields */
1105 for (i
= 0; i
< tp
->nr_args
; i
++)
1106 DEFINE_FIELD(unsigned long, args
[i
], tp
->args
[i
].name
, 0);
1110 static int kretprobe_event_define_fields(struct ftrace_event_call
*event_call
)
1113 struct kretprobe_trace_entry field
;
1114 struct trace_probe
*tp
= (struct trace_probe
*)event_call
->data
;
1116 ret
= trace_define_common_fields(event_call
);
1120 DEFINE_FIELD(unsigned long, func
, FIELD_STRING_FUNC
, 0);
1121 DEFINE_FIELD(unsigned long, ret_ip
, FIELD_STRING_RETIP
, 0);
1122 DEFINE_FIELD(int, nargs
, FIELD_STRING_NARGS
, 1);
1123 /* Set argument names as fields */
1124 for (i
= 0; i
< tp
->nr_args
; i
++)
1125 DEFINE_FIELD(unsigned long, args
[i
], tp
->args
[i
].name
, 0);
1129 static int __probe_event_show_format(struct trace_seq
*s
,
1130 struct trace_probe
*tp
, const char *fmt
,
1136 if (!trace_seq_printf(s
, "\nprint fmt: \"%s", fmt
))
1139 for (i
= 0; i
< tp
->nr_args
; i
++)
1140 if (!trace_seq_printf(s
, " %s=%%lx", tp
->args
[i
].name
))
1143 if (!trace_seq_printf(s
, "\", %s", arg
))
1146 for (i
= 0; i
< tp
->nr_args
; i
++)
1147 if (!trace_seq_printf(s
, ", REC->%s", tp
->args
[i
].name
))
1150 return trace_seq_puts(s
, "\n");
1154 #define SHOW_FIELD(type, item, name) \
1156 ret = trace_seq_printf(s, "\tfield: " #type " %s;\t" \
1157 "offset:%u;\tsize:%u;\n", name, \
1158 (unsigned int)offsetof(typeof(field), item),\
1159 (unsigned int)sizeof(type)); \
1164 static int kprobe_event_show_format(struct ftrace_event_call
*call
,
1165 struct trace_seq
*s
)
1167 struct kprobe_trace_entry field
__attribute__((unused
));
1169 struct trace_probe
*tp
= (struct trace_probe
*)call
->data
;
1171 SHOW_FIELD(unsigned long, ip
, FIELD_STRING_IP
);
1172 SHOW_FIELD(int, nargs
, FIELD_STRING_NARGS
);
1175 for (i
= 0; i
< tp
->nr_args
; i
++)
1176 SHOW_FIELD(unsigned long, args
[i
], tp
->args
[i
].name
);
1177 trace_seq_puts(s
, "\n");
1179 return __probe_event_show_format(s
, tp
, "(%lx)",
1180 "REC->" FIELD_STRING_IP
);
1183 static int kretprobe_event_show_format(struct ftrace_event_call
*call
,
1184 struct trace_seq
*s
)
1186 struct kretprobe_trace_entry field
__attribute__((unused
));
1188 struct trace_probe
*tp
= (struct trace_probe
*)call
->data
;
1190 SHOW_FIELD(unsigned long, func
, FIELD_STRING_FUNC
);
1191 SHOW_FIELD(unsigned long, ret_ip
, FIELD_STRING_RETIP
);
1192 SHOW_FIELD(int, nargs
, FIELD_STRING_NARGS
);
1195 for (i
= 0; i
< tp
->nr_args
; i
++)
1196 SHOW_FIELD(unsigned long, args
[i
], tp
->args
[i
].name
);
1197 trace_seq_puts(s
, "\n");
1199 return __probe_event_show_format(s
, tp
, "(%lx <- %lx)",
1200 "REC->" FIELD_STRING_FUNC
1201 ", REC->" FIELD_STRING_RETIP
);
1204 #ifdef CONFIG_EVENT_PROFILE
1206 /* Kprobe profile handler */
1207 static __kprobes
int kprobe_profile_func(struct kprobe
*kp
,
1208 struct pt_regs
*regs
)
1210 struct trace_probe
*tp
= container_of(kp
, struct trace_probe
, rp
.kp
);
1211 struct ftrace_event_call
*call
= &tp
->call
;
1212 struct kprobe_trace_entry
*entry
;
1213 struct trace_entry
*ent
;
1214 int size
, __size
, i
, pc
, __cpu
;
1215 unsigned long irq_flags
;
1220 pc
= preempt_count();
1221 __size
= SIZEOF_KPROBE_TRACE_ENTRY(tp
->nr_args
);
1222 size
= ALIGN(__size
+ sizeof(u32
), sizeof(u64
));
1223 size
-= sizeof(u32
);
1224 if (WARN_ONCE(size
> FTRACE_MAX_PROFILE_SIZE
,
1225 "profile buffer not large enough"))
1229 * Protect the non nmi buffer
1230 * This also protects the rcu read side
1232 local_irq_save(irq_flags
);
1234 rctx
= perf_swevent_get_recursion_context();
1238 __cpu
= smp_processor_id();
1241 trace_buf
= rcu_dereference(perf_trace_buf_nmi
);
1243 trace_buf
= rcu_dereference(perf_trace_buf
);
1248 raw_data
= per_cpu_ptr(trace_buf
, __cpu
);
1250 /* Zero dead bytes from alignment to avoid buffer leak to userspace */
1251 *(u64
*)(&raw_data
[size
- sizeof(u64
)]) = 0ULL;
1252 entry
= (struct kprobe_trace_entry
*)raw_data
;
1255 tracing_generic_entry_update(ent
, irq_flags
, pc
);
1256 ent
->type
= call
->id
;
1257 entry
->nargs
= tp
->nr_args
;
1258 entry
->ip
= (unsigned long)kp
->addr
;
1259 for (i
= 0; i
< tp
->nr_args
; i
++)
1260 entry
->args
[i
] = call_fetch(&tp
->args
[i
].fetch
, regs
);
1261 perf_tp_event(call
->id
, entry
->ip
, 1, entry
, size
);
1264 perf_swevent_put_recursion_context(rctx
);
1266 local_irq_restore(irq_flags
);
1271 /* Kretprobe profile handler */
1272 static __kprobes
int kretprobe_profile_func(struct kretprobe_instance
*ri
,
1273 struct pt_regs
*regs
)
1275 struct trace_probe
*tp
= container_of(ri
->rp
, struct trace_probe
, rp
);
1276 struct ftrace_event_call
*call
= &tp
->call
;
1277 struct kretprobe_trace_entry
*entry
;
1278 struct trace_entry
*ent
;
1279 int size
, __size
, i
, pc
, __cpu
;
1280 unsigned long irq_flags
;
1285 pc
= preempt_count();
1286 __size
= SIZEOF_KRETPROBE_TRACE_ENTRY(tp
->nr_args
);
1287 size
= ALIGN(__size
+ sizeof(u32
), sizeof(u64
));
1288 size
-= sizeof(u32
);
1289 if (WARN_ONCE(size
> FTRACE_MAX_PROFILE_SIZE
,
1290 "profile buffer not large enough"))
1294 * Protect the non nmi buffer
1295 * This also protects the rcu read side
1297 local_irq_save(irq_flags
);
1299 rctx
= perf_swevent_get_recursion_context();
1303 __cpu
= smp_processor_id();
1306 trace_buf
= rcu_dereference(perf_trace_buf_nmi
);
1308 trace_buf
= rcu_dereference(perf_trace_buf
);
1313 raw_data
= per_cpu_ptr(trace_buf
, __cpu
);
1315 /* Zero dead bytes from alignment to avoid buffer leak to userspace */
1316 *(u64
*)(&raw_data
[size
- sizeof(u64
)]) = 0ULL;
1317 entry
= (struct kretprobe_trace_entry
*)raw_data
;
1320 tracing_generic_entry_update(ent
, irq_flags
, pc
);
1321 ent
->type
= call
->id
;
1322 entry
->nargs
= tp
->nr_args
;
1323 entry
->func
= (unsigned long)tp
->rp
.kp
.addr
;
1324 entry
->ret_ip
= (unsigned long)ri
->ret_addr
;
1325 for (i
= 0; i
< tp
->nr_args
; i
++)
1326 entry
->args
[i
] = call_fetch(&tp
->args
[i
].fetch
, regs
);
1327 perf_tp_event(call
->id
, entry
->ret_ip
, 1, entry
, size
);
1330 perf_swevent_put_recursion_context(rctx
);
1332 local_irq_restore(irq_flags
);
1337 static int probe_profile_enable(struct ftrace_event_call
*call
)
1339 struct trace_probe
*tp
= (struct trace_probe
*)call
->data
;
1341 tp
->flags
|= TP_FLAG_PROFILE
;
1343 if (probe_is_return(tp
))
1344 return enable_kretprobe(&tp
->rp
);
1346 return enable_kprobe(&tp
->rp
.kp
);
1349 static void probe_profile_disable(struct ftrace_event_call
*call
)
1351 struct trace_probe
*tp
= (struct trace_probe
*)call
->data
;
1353 tp
->flags
&= ~TP_FLAG_PROFILE
;
1355 if (!(tp
->flags
& TP_FLAG_TRACE
)) {
1356 if (probe_is_return(tp
))
1357 disable_kretprobe(&tp
->rp
);
1359 disable_kprobe(&tp
->rp
.kp
);
1362 #endif /* CONFIG_EVENT_PROFILE */
1366 int kprobe_dispatcher(struct kprobe
*kp
, struct pt_regs
*regs
)
1368 struct trace_probe
*tp
= container_of(kp
, struct trace_probe
, rp
.kp
);
1370 if (tp
->flags
& TP_FLAG_TRACE
)
1371 kprobe_trace_func(kp
, regs
);
1372 #ifdef CONFIG_EVENT_PROFILE
1373 if (tp
->flags
& TP_FLAG_PROFILE
)
1374 kprobe_profile_func(kp
, regs
);
1375 #endif /* CONFIG_EVENT_PROFILE */
1376 return 0; /* We don't tweek kernel, so just return 0 */
1380 int kretprobe_dispatcher(struct kretprobe_instance
*ri
, struct pt_regs
*regs
)
1382 struct trace_probe
*tp
= container_of(ri
->rp
, struct trace_probe
, rp
);
1384 if (tp
->flags
& TP_FLAG_TRACE
)
1385 kretprobe_trace_func(ri
, regs
);
1386 #ifdef CONFIG_EVENT_PROFILE
1387 if (tp
->flags
& TP_FLAG_PROFILE
)
1388 kretprobe_profile_func(ri
, regs
);
1389 #endif /* CONFIG_EVENT_PROFILE */
1390 return 0; /* We don't tweek kernel, so just return 0 */
1393 static int register_probe_event(struct trace_probe
*tp
)
1395 struct ftrace_event_call
*call
= &tp
->call
;
1398 /* Initialize ftrace_event_call */
1399 if (probe_is_return(tp
)) {
1400 tp
->event
.trace
= print_kretprobe_event
;
1401 call
->raw_init
= probe_event_raw_init
;
1402 call
->show_format
= kretprobe_event_show_format
;
1403 call
->define_fields
= kretprobe_event_define_fields
;
1405 tp
->event
.trace
= print_kprobe_event
;
1406 call
->raw_init
= probe_event_raw_init
;
1407 call
->show_format
= kprobe_event_show_format
;
1408 call
->define_fields
= kprobe_event_define_fields
;
1410 call
->event
= &tp
->event
;
1411 call
->id
= register_ftrace_event(&tp
->event
);
1415 call
->regfunc
= probe_event_enable
;
1416 call
->unregfunc
= probe_event_disable
;
1418 #ifdef CONFIG_EVENT_PROFILE
1419 atomic_set(&call
->profile_count
, -1);
1420 call
->profile_enable
= probe_profile_enable
;
1421 call
->profile_disable
= probe_profile_disable
;
1424 ret
= trace_add_event_call(call
);
1426 pr_info("Failed to register kprobe event: %s\n", call
->name
);
1427 unregister_ftrace_event(&tp
->event
);
1432 static void unregister_probe_event(struct trace_probe
*tp
)
1434 /* tp->event is unregistered in trace_remove_event_call() */
1435 trace_remove_event_call(&tp
->call
);
1438 /* Make a debugfs interface for controling probe points */
1439 static __init
int init_kprobe_trace(void)
1441 struct dentry
*d_tracer
;
1442 struct dentry
*entry
;
1444 d_tracer
= tracing_init_dentry();
1448 entry
= debugfs_create_file("kprobe_events", 0644, d_tracer
,
1449 NULL
, &kprobe_events_ops
);
1451 /* Event list interface */
1453 pr_warning("Could not create debugfs "
1454 "'kprobe_events' entry\n");
1456 /* Profile interface */
1457 entry
= debugfs_create_file("kprobe_profile", 0444, d_tracer
,
1458 NULL
, &kprobe_profile_ops
);
1461 pr_warning("Could not create debugfs "
1462 "'kprobe_profile' entry\n");
1465 fs_initcall(init_kprobe_trace
);
1468 #ifdef CONFIG_FTRACE_STARTUP_TEST
1470 static int kprobe_trace_selftest_target(int a1
, int a2
, int a3
,
1471 int a4
, int a5
, int a6
)
1473 return a1
+ a2
+ a3
+ a4
+ a5
+ a6
;
1476 static __init
int kprobe_trace_self_tests_init(void)
1479 int (*target
)(int, int, int, int, int, int);
1481 target
= kprobe_trace_selftest_target
;
1483 pr_info("Testing kprobe tracing: ");
1485 ret
= command_trace_probe("p:testprobe kprobe_trace_selftest_target "
1486 "$arg1 $arg2 $arg3 $arg4 $stack $stack0");
1487 if (WARN_ON_ONCE(ret
))
1488 pr_warning("error enabling function entry\n");
1490 ret
= command_trace_probe("r:testprobe2 kprobe_trace_selftest_target "
1492 if (WARN_ON_ONCE(ret
))
1493 pr_warning("error enabling function return\n");
1495 ret
= target(1, 2, 3, 4, 5, 6);
1497 cleanup_all_probes();
1503 late_initcall(kprobe_trace_self_tests_init
);