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
19 #define pr_fmt(fmt) "trace_kprobe: " fmt
21 #include <linux/module.h>
22 #include <linux/uaccess.h>
23 #include <linux/rculist.h>
24 #include <linux/error-injection.h>
26 #include "trace_probe.h"
28 #define KPROBE_EVENT_SYSTEM "kprobes"
29 #define KRETPROBE_MAXACTIVE_MAX 4096
32 * Kprobe event core functions
35 struct list_head list
;
36 struct kretprobe rp
; /* Use rp.kp for kprobe use */
37 unsigned long __percpu
*nhit
;
38 const char *symbol
; /* symbol name */
39 struct trace_probe tp
;
42 #define SIZEOF_TRACE_KPROBE(n) \
43 (offsetof(struct trace_kprobe, tp.args) + \
44 (sizeof(struct probe_arg) * (n)))
46 static nokprobe_inline
bool trace_kprobe_is_return(struct trace_kprobe
*tk
)
48 return tk
->rp
.handler
!= NULL
;
51 static nokprobe_inline
const char *trace_kprobe_symbol(struct trace_kprobe
*tk
)
53 return tk
->symbol
? tk
->symbol
: "unknown";
56 static nokprobe_inline
unsigned long trace_kprobe_offset(struct trace_kprobe
*tk
)
58 return tk
->rp
.kp
.offset
;
61 static nokprobe_inline
bool trace_kprobe_has_gone(struct trace_kprobe
*tk
)
63 return !!(kprobe_gone(&tk
->rp
.kp
));
66 static nokprobe_inline
bool trace_kprobe_within_module(struct trace_kprobe
*tk
,
69 int len
= strlen(mod
->name
);
70 const char *name
= trace_kprobe_symbol(tk
);
71 return strncmp(mod
->name
, name
, len
) == 0 && name
[len
] == ':';
74 static nokprobe_inline
bool trace_kprobe_is_on_module(struct trace_kprobe
*tk
)
76 return !!strchr(trace_kprobe_symbol(tk
), ':');
79 static nokprobe_inline
unsigned long trace_kprobe_nhit(struct trace_kprobe
*tk
)
81 unsigned long nhit
= 0;
84 for_each_possible_cpu(cpu
)
85 nhit
+= *per_cpu_ptr(tk
->nhit
, cpu
);
90 bool trace_kprobe_on_func_entry(struct trace_event_call
*call
)
92 struct trace_kprobe
*tk
= (struct trace_kprobe
*)call
->data
;
94 return kprobe_on_func_entry(tk
->rp
.kp
.addr
,
95 tk
->rp
.kp
.addr
? NULL
: tk
->rp
.kp
.symbol_name
,
96 tk
->rp
.kp
.addr
? 0 : tk
->rp
.kp
.offset
);
99 bool trace_kprobe_error_injectable(struct trace_event_call
*call
)
101 struct trace_kprobe
*tk
= (struct trace_kprobe
*)call
->data
;
105 addr
= (unsigned long)
106 kallsyms_lookup_name(trace_kprobe_symbol(tk
));
107 addr
+= tk
->rp
.kp
.offset
;
109 addr
= (unsigned long)tk
->rp
.kp
.addr
;
111 return within_error_injection_list(addr
);
114 static int register_kprobe_event(struct trace_kprobe
*tk
);
115 static int unregister_kprobe_event(struct trace_kprobe
*tk
);
117 static DEFINE_MUTEX(probe_lock
);
118 static LIST_HEAD(probe_list
);
120 static int kprobe_dispatcher(struct kprobe
*kp
, struct pt_regs
*regs
);
121 static int kretprobe_dispatcher(struct kretprobe_instance
*ri
,
122 struct pt_regs
*regs
);
124 /* Memory fetching by symbol */
125 struct symbol_cache
{
131 unsigned long update_symbol_cache(struct symbol_cache
*sc
)
133 sc
->addr
= (unsigned long)kallsyms_lookup_name(sc
->symbol
);
136 sc
->addr
+= sc
->offset
;
141 void free_symbol_cache(struct symbol_cache
*sc
)
147 struct symbol_cache
*alloc_symbol_cache(const char *sym
, long offset
)
149 struct symbol_cache
*sc
;
151 if (!sym
|| strlen(sym
) == 0)
154 sc
= kzalloc(sizeof(struct symbol_cache
), GFP_KERNEL
);
158 sc
->symbol
= kstrdup(sym
, GFP_KERNEL
);
164 update_symbol_cache(sc
);
170 * Kprobes-specific fetch functions
172 #define DEFINE_FETCH_stack(type) \
173 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs, \
174 void *offset, void *dest) \
176 *(type *)dest = (type)regs_get_kernel_stack_nth(regs, \
177 (unsigned int)((unsigned long)offset)); \
179 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
181 DEFINE_BASIC_FETCH_FUNCS(stack
)
182 /* No string on the stack entry */
183 #define fetch_stack_string NULL
184 #define fetch_stack_string_size NULL
186 #define DEFINE_FETCH_memory(type) \
187 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs, \
188 void *addr, void *dest) \
191 if (probe_kernel_address(addr, retval)) \
194 *(type *)dest = retval; \
196 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
198 DEFINE_BASIC_FETCH_FUNCS(memory
)
200 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
201 * length and relative data location.
203 static void FETCH_FUNC_NAME(memory
, string
)(struct pt_regs
*regs
,
204 void *addr
, void *dest
)
206 int maxlen
= get_rloc_len(*(u32
*)dest
);
207 u8
*dst
= get_rloc_data(dest
);
214 * Try to get string again, since the string can be changed while
217 ret
= strncpy_from_unsafe(dst
, addr
, maxlen
);
219 if (ret
< 0) { /* Failed to fetch string */
221 *(u32
*)dest
= make_data_rloc(0, get_rloc_offs(*(u32
*)dest
));
223 *(u32
*)dest
= make_data_rloc(ret
, get_rloc_offs(*(u32
*)dest
));
226 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory
, string
));
228 /* Return the length of string -- including null terminal byte */
229 static void FETCH_FUNC_NAME(memory
, string_size
)(struct pt_regs
*regs
,
230 void *addr
, void *dest
)
241 ret
= __copy_from_user_inatomic(&c
, (u8
*)addr
+ len
, 1);
243 } while (c
&& ret
== 0 && len
< MAX_STRING_SIZE
);
248 if (ret
< 0) /* Failed to check the length */
253 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory
, string_size
));
255 #define DEFINE_FETCH_symbol(type) \
256 void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
258 struct symbol_cache *sc = data; \
260 fetch_memory_##type(regs, (void *)sc->addr, dest); \
264 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
266 DEFINE_BASIC_FETCH_FUNCS(symbol
)
267 DEFINE_FETCH_symbol(string
)
268 DEFINE_FETCH_symbol(string_size
)
270 /* kprobes don't support file_offset fetch methods */
271 #define fetch_file_offset_u8 NULL
272 #define fetch_file_offset_u16 NULL
273 #define fetch_file_offset_u32 NULL
274 #define fetch_file_offset_u64 NULL
275 #define fetch_file_offset_string NULL
276 #define fetch_file_offset_string_size NULL
278 /* Fetch type information table */
279 static const struct fetch_type kprobes_fetch_type_table
[] = {
281 [FETCH_TYPE_STRING
] = __ASSIGN_FETCH_TYPE("string", string
, string
,
282 sizeof(u32
), 1, "__data_loc char[]"),
283 [FETCH_TYPE_STRSIZE
] = __ASSIGN_FETCH_TYPE("string_size", u32
,
284 string_size
, sizeof(u32
), 0, "u32"),
286 ASSIGN_FETCH_TYPE(u8
, u8
, 0),
287 ASSIGN_FETCH_TYPE(u16
, u16
, 0),
288 ASSIGN_FETCH_TYPE(u32
, u32
, 0),
289 ASSIGN_FETCH_TYPE(u64
, u64
, 0),
290 ASSIGN_FETCH_TYPE(s8
, u8
, 1),
291 ASSIGN_FETCH_TYPE(s16
, u16
, 1),
292 ASSIGN_FETCH_TYPE(s32
, u32
, 1),
293 ASSIGN_FETCH_TYPE(s64
, u64
, 1),
294 ASSIGN_FETCH_TYPE_ALIAS(x8
, u8
, u8
, 0),
295 ASSIGN_FETCH_TYPE_ALIAS(x16
, u16
, u16
, 0),
296 ASSIGN_FETCH_TYPE_ALIAS(x32
, u32
, u32
, 0),
297 ASSIGN_FETCH_TYPE_ALIAS(x64
, u64
, u64
, 0),
299 ASSIGN_FETCH_TYPE_END
303 * Allocate new trace_probe and initialize it (including kprobes).
305 static struct trace_kprobe
*alloc_trace_kprobe(const char *group
,
311 int nargs
, bool is_return
)
313 struct trace_kprobe
*tk
;
316 tk
= kzalloc(SIZEOF_TRACE_KPROBE(nargs
), GFP_KERNEL
);
320 tk
->nhit
= alloc_percpu(unsigned long);
325 tk
->symbol
= kstrdup(symbol
, GFP_KERNEL
);
328 tk
->rp
.kp
.symbol_name
= tk
->symbol
;
329 tk
->rp
.kp
.offset
= offs
;
331 tk
->rp
.kp
.addr
= addr
;
334 tk
->rp
.handler
= kretprobe_dispatcher
;
336 tk
->rp
.kp
.pre_handler
= kprobe_dispatcher
;
338 tk
->rp
.maxactive
= maxactive
;
340 if (!event
|| !is_good_name(event
)) {
345 tk
->tp
.call
.class = &tk
->tp
.class;
346 tk
->tp
.call
.name
= kstrdup(event
, GFP_KERNEL
);
347 if (!tk
->tp
.call
.name
)
350 if (!group
|| !is_good_name(group
)) {
355 tk
->tp
.class.system
= kstrdup(group
, GFP_KERNEL
);
356 if (!tk
->tp
.class.system
)
359 INIT_LIST_HEAD(&tk
->list
);
360 INIT_LIST_HEAD(&tk
->tp
.files
);
363 kfree(tk
->tp
.call
.name
);
365 free_percpu(tk
->nhit
);
370 static void free_trace_kprobe(struct trace_kprobe
*tk
)
374 for (i
= 0; i
< tk
->tp
.nr_args
; i
++)
375 traceprobe_free_probe_arg(&tk
->tp
.args
[i
]);
377 kfree(tk
->tp
.call
.class->system
);
378 kfree(tk
->tp
.call
.name
);
380 free_percpu(tk
->nhit
);
384 static struct trace_kprobe
*find_trace_kprobe(const char *event
,
387 struct trace_kprobe
*tk
;
389 list_for_each_entry(tk
, &probe_list
, list
)
390 if (strcmp(trace_event_name(&tk
->tp
.call
), event
) == 0 &&
391 strcmp(tk
->tp
.call
.class->system
, group
) == 0)
398 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
401 enable_trace_kprobe(struct trace_kprobe
*tk
, struct trace_event_file
*file
)
406 struct event_file_link
*link
;
408 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
415 list_add_tail_rcu(&link
->list
, &tk
->tp
.files
);
417 tk
->tp
.flags
|= TP_FLAG_TRACE
;
419 tk
->tp
.flags
|= TP_FLAG_PROFILE
;
421 if (trace_probe_is_registered(&tk
->tp
) && !trace_kprobe_has_gone(tk
)) {
422 if (trace_kprobe_is_return(tk
))
423 ret
= enable_kretprobe(&tk
->rp
);
425 ret
= enable_kprobe(&tk
->rp
.kp
);
432 * Disable trace_probe
433 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
436 disable_trace_kprobe(struct trace_kprobe
*tk
, struct trace_event_file
*file
)
438 struct event_file_link
*link
= NULL
;
443 link
= find_event_file_link(&tk
->tp
, file
);
449 list_del_rcu(&link
->list
);
451 if (!list_empty(&tk
->tp
.files
))
454 tk
->tp
.flags
&= ~TP_FLAG_TRACE
;
456 tk
->tp
.flags
&= ~TP_FLAG_PROFILE
;
458 if (!trace_probe_is_enabled(&tk
->tp
) && trace_probe_is_registered(&tk
->tp
)) {
459 if (trace_kprobe_is_return(tk
))
460 disable_kretprobe(&tk
->rp
);
462 disable_kprobe(&tk
->rp
.kp
);
468 * Synchronize with kprobe_trace_func/kretprobe_trace_func
469 * to ensure disabled (all running handlers are finished).
470 * This is not only for kfree(), but also the caller,
471 * trace_remove_event_call() supposes it for releasing
472 * event_call related objects, which will be accessed in
473 * the kprobe_trace_func/kretprobe_trace_func.
476 kfree(link
); /* Ignored if link == NULL */
482 /* Internal register function - just handle k*probes and flags */
483 static int __register_trace_kprobe(struct trace_kprobe
*tk
)
487 if (trace_probe_is_registered(&tk
->tp
))
490 for (i
= 0; i
< tk
->tp
.nr_args
; i
++)
491 traceprobe_update_arg(&tk
->tp
.args
[i
]);
493 /* Set/clear disabled flag according to tp->flag */
494 if (trace_probe_is_enabled(&tk
->tp
))
495 tk
->rp
.kp
.flags
&= ~KPROBE_FLAG_DISABLED
;
497 tk
->rp
.kp
.flags
|= KPROBE_FLAG_DISABLED
;
499 if (trace_kprobe_is_return(tk
))
500 ret
= register_kretprobe(&tk
->rp
);
502 ret
= register_kprobe(&tk
->rp
.kp
);
505 tk
->tp
.flags
|= TP_FLAG_REGISTERED
;
507 pr_warn("Could not insert probe at %s+%lu: %d\n",
508 trace_kprobe_symbol(tk
), trace_kprobe_offset(tk
), ret
);
509 if (ret
== -ENOENT
&& trace_kprobe_is_on_module(tk
)) {
510 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
512 } else if (ret
== -EILSEQ
) {
513 pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
522 /* Internal unregister function - just handle k*probes and flags */
523 static void __unregister_trace_kprobe(struct trace_kprobe
*tk
)
525 if (trace_probe_is_registered(&tk
->tp
)) {
526 if (trace_kprobe_is_return(tk
))
527 unregister_kretprobe(&tk
->rp
);
529 unregister_kprobe(&tk
->rp
.kp
);
530 tk
->tp
.flags
&= ~TP_FLAG_REGISTERED
;
531 /* Cleanup kprobe for reuse */
532 if (tk
->rp
.kp
.symbol_name
)
533 tk
->rp
.kp
.addr
= NULL
;
537 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
538 static int unregister_trace_kprobe(struct trace_kprobe
*tk
)
540 /* Enabled event can not be unregistered */
541 if (trace_probe_is_enabled(&tk
->tp
))
544 /* Will fail if probe is being used by ftrace or perf */
545 if (unregister_kprobe_event(tk
))
548 __unregister_trace_kprobe(tk
);
554 /* Register a trace_probe and probe_event */
555 static int register_trace_kprobe(struct trace_kprobe
*tk
)
557 struct trace_kprobe
*old_tk
;
560 mutex_lock(&probe_lock
);
562 /* Delete old (same name) event if exist */
563 old_tk
= find_trace_kprobe(trace_event_name(&tk
->tp
.call
),
564 tk
->tp
.call
.class->system
);
566 ret
= unregister_trace_kprobe(old_tk
);
569 free_trace_kprobe(old_tk
);
572 /* Register new event */
573 ret
= register_kprobe_event(tk
);
575 pr_warn("Failed to register probe event(%d)\n", ret
);
579 /* Register k*probe */
580 ret
= __register_trace_kprobe(tk
);
582 unregister_kprobe_event(tk
);
584 list_add_tail(&tk
->list
, &probe_list
);
587 mutex_unlock(&probe_lock
);
591 /* Module notifier call back, checking event on the module */
592 static int trace_kprobe_module_callback(struct notifier_block
*nb
,
593 unsigned long val
, void *data
)
595 struct module
*mod
= data
;
596 struct trace_kprobe
*tk
;
599 if (val
!= MODULE_STATE_COMING
)
602 /* Update probes on coming module */
603 mutex_lock(&probe_lock
);
604 list_for_each_entry(tk
, &probe_list
, list
) {
605 if (trace_kprobe_within_module(tk
, mod
)) {
606 /* Don't need to check busy - this should have gone. */
607 __unregister_trace_kprobe(tk
);
608 ret
= __register_trace_kprobe(tk
);
610 pr_warn("Failed to re-register probe %s on %s: %d\n",
611 trace_event_name(&tk
->tp
.call
),
615 mutex_unlock(&probe_lock
);
620 static struct notifier_block trace_kprobe_module_nb
= {
621 .notifier_call
= trace_kprobe_module_callback
,
622 .priority
= 1 /* Invoked after kprobe module callback */
625 /* Convert certain expected symbols into '_' when generating event names */
626 static inline void sanitize_event_name(char *name
)
628 while (*name
++ != '\0')
629 if (*name
== ':' || *name
== '.')
633 static int create_trace_kprobe(int argc
, char **argv
)
638 * p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
640 * r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
642 * $retval : fetch return value
643 * $stack : fetch stack address
644 * $stackN : fetch Nth of stack (N:0-)
645 * $comm : fetch current task comm
646 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
647 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
648 * %REG : fetch register REG
649 * Dereferencing memory fetch:
650 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
651 * Alias name of args:
652 * NAME=FETCHARG : set NAME as alias of FETCHARG.
654 * FETCHARG:TYPE : use TYPE instead of unsigned long.
656 struct trace_kprobe
*tk
;
658 bool is_return
= false, is_delete
= false;
659 char *symbol
= NULL
, *event
= NULL
, *group
= NULL
;
662 unsigned long offset
= 0;
664 char buf
[MAX_EVENT_NAME_LEN
];
666 /* argc must be >= 1 */
667 if (argv
[0][0] == 'p')
669 else if (argv
[0][0] == 'r')
671 else if (argv
[0][0] == '-')
674 pr_info("Probe definition must be started with 'p', 'r' or"
679 event
= strchr(&argv
[0][1], ':');
684 if (is_return
&& isdigit(argv
[0][1])) {
685 ret
= kstrtouint(&argv
[0][1], 0, &maxactive
);
687 pr_info("Failed to parse maxactive.\n");
690 /* kretprobes instances are iterated over via a list. The
691 * maximum should stay reasonable.
693 if (maxactive
> KRETPROBE_MAXACTIVE_MAX
) {
694 pr_info("Maxactive is too big (%d > %d).\n",
695 maxactive
, KRETPROBE_MAXACTIVE_MAX
);
701 if (strchr(event
, '/')) {
703 event
= strchr(group
, '/') + 1;
705 if (strlen(group
) == 0) {
706 pr_info("Group name is not specified\n");
710 if (strlen(event
) == 0) {
711 pr_info("Event name is not specified\n");
716 group
= KPROBE_EVENT_SYSTEM
;
720 pr_info("Delete command needs an event name.\n");
723 mutex_lock(&probe_lock
);
724 tk
= find_trace_kprobe(event
, group
);
726 mutex_unlock(&probe_lock
);
727 pr_info("Event %s/%s doesn't exist.\n", group
, event
);
730 /* delete an event */
731 ret
= unregister_trace_kprobe(tk
);
733 free_trace_kprobe(tk
);
734 mutex_unlock(&probe_lock
);
739 pr_info("Probe point is not specified.\n");
743 /* try to parse an address. if that fails, try to read the
744 * input as a symbol. */
745 if (kstrtoul(argv
[1], 0, (unsigned long *)&addr
)) {
746 /* a symbol specified */
748 /* TODO: support .init module functions */
749 ret
= traceprobe_split_symbol_offset(symbol
, &offset
);
751 pr_info("Failed to parse either an address or a symbol.\n");
754 if (offset
&& is_return
&&
755 !kprobe_on_func_entry(NULL
, symbol
, offset
)) {
756 pr_info("Given offset is not valid for return probe.\n");
760 argc
-= 2; argv
+= 2;
764 /* Make a new event name */
766 snprintf(buf
, MAX_EVENT_NAME_LEN
, "%c_%s_%ld",
767 is_return
? 'r' : 'p', symbol
, offset
);
769 snprintf(buf
, MAX_EVENT_NAME_LEN
, "%c_0x%p",
770 is_return
? 'r' : 'p', addr
);
771 sanitize_event_name(buf
);
774 tk
= alloc_trace_kprobe(group
, event
, addr
, symbol
, offset
, maxactive
,
777 pr_info("Failed to allocate trace_probe.(%d)\n",
782 /* parse arguments */
784 for (i
= 0; i
< argc
&& i
< MAX_TRACE_ARGS
; i
++) {
785 struct probe_arg
*parg
= &tk
->tp
.args
[i
];
787 /* Increment count for freeing args in error case */
790 /* Parse argument name */
791 arg
= strchr(argv
[i
], '=');
794 parg
->name
= kstrdup(argv
[i
], GFP_KERNEL
);
797 /* If argument name is omitted, set "argN" */
798 snprintf(buf
, MAX_EVENT_NAME_LEN
, "arg%d", i
+ 1);
799 parg
->name
= kstrdup(buf
, GFP_KERNEL
);
803 pr_info("Failed to allocate argument[%d] name.\n", i
);
808 if (!is_good_name(parg
->name
)) {
809 pr_info("Invalid argument[%d] name: %s\n",
815 if (traceprobe_conflict_field_name(parg
->name
,
817 pr_info("Argument[%d] name '%s' conflicts with "
818 "another field.\n", i
, argv
[i
]);
823 /* Parse fetch argument */
824 ret
= traceprobe_parse_probe_arg(arg
, &tk
->tp
.size
, parg
,
826 kprobes_fetch_type_table
);
828 pr_info("Parse error at argument[%d]. (%d)\n", i
, ret
);
833 ret
= register_trace_kprobe(tk
);
839 free_trace_kprobe(tk
);
843 static int release_all_trace_kprobes(void)
845 struct trace_kprobe
*tk
;
848 mutex_lock(&probe_lock
);
849 /* Ensure no probe is in use. */
850 list_for_each_entry(tk
, &probe_list
, list
)
851 if (trace_probe_is_enabled(&tk
->tp
)) {
855 /* TODO: Use batch unregistration */
856 while (!list_empty(&probe_list
)) {
857 tk
= list_entry(probe_list
.next
, struct trace_kprobe
, list
);
858 ret
= unregister_trace_kprobe(tk
);
861 free_trace_kprobe(tk
);
865 mutex_unlock(&probe_lock
);
870 /* Probes listing interfaces */
871 static void *probes_seq_start(struct seq_file
*m
, loff_t
*pos
)
873 mutex_lock(&probe_lock
);
874 return seq_list_start(&probe_list
, *pos
);
877 static void *probes_seq_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
879 return seq_list_next(v
, &probe_list
, pos
);
882 static void probes_seq_stop(struct seq_file
*m
, void *v
)
884 mutex_unlock(&probe_lock
);
887 static int probes_seq_show(struct seq_file
*m
, void *v
)
889 struct trace_kprobe
*tk
= v
;
892 seq_putc(m
, trace_kprobe_is_return(tk
) ? 'r' : 'p');
893 seq_printf(m
, ":%s/%s", tk
->tp
.call
.class->system
,
894 trace_event_name(&tk
->tp
.call
));
897 seq_printf(m
, " 0x%p", tk
->rp
.kp
.addr
);
898 else if (tk
->rp
.kp
.offset
)
899 seq_printf(m
, " %s+%u", trace_kprobe_symbol(tk
),
902 seq_printf(m
, " %s", trace_kprobe_symbol(tk
));
904 for (i
= 0; i
< tk
->tp
.nr_args
; i
++)
905 seq_printf(m
, " %s=%s", tk
->tp
.args
[i
].name
, tk
->tp
.args
[i
].comm
);
911 static const struct seq_operations probes_seq_op
= {
912 .start
= probes_seq_start
,
913 .next
= probes_seq_next
,
914 .stop
= probes_seq_stop
,
915 .show
= probes_seq_show
918 static int probes_open(struct inode
*inode
, struct file
*file
)
922 if ((file
->f_mode
& FMODE_WRITE
) && (file
->f_flags
& O_TRUNC
)) {
923 ret
= release_all_trace_kprobes();
928 return seq_open(file
, &probes_seq_op
);
931 static ssize_t
probes_write(struct file
*file
, const char __user
*buffer
,
932 size_t count
, loff_t
*ppos
)
934 return trace_parse_run_command(file
, buffer
, count
, ppos
,
935 create_trace_kprobe
);
938 static const struct file_operations kprobe_events_ops
= {
939 .owner
= THIS_MODULE
,
943 .release
= seq_release
,
944 .write
= probes_write
,
947 /* Probes profiling interfaces */
948 static int probes_profile_seq_show(struct seq_file
*m
, void *v
)
950 struct trace_kprobe
*tk
= v
;
952 seq_printf(m
, " %-44s %15lu %15lu\n",
953 trace_event_name(&tk
->tp
.call
),
954 trace_kprobe_nhit(tk
),
960 static const struct seq_operations profile_seq_op
= {
961 .start
= probes_seq_start
,
962 .next
= probes_seq_next
,
963 .stop
= probes_seq_stop
,
964 .show
= probes_profile_seq_show
967 static int profile_open(struct inode
*inode
, struct file
*file
)
969 return seq_open(file
, &profile_seq_op
);
972 static const struct file_operations kprobe_profile_ops
= {
973 .owner
= THIS_MODULE
,
974 .open
= profile_open
,
977 .release
= seq_release
,
981 static nokprobe_inline
void
982 __kprobe_trace_func(struct trace_kprobe
*tk
, struct pt_regs
*regs
,
983 struct trace_event_file
*trace_file
)
985 struct kprobe_trace_entry_head
*entry
;
986 struct ring_buffer_event
*event
;
987 struct ring_buffer
*buffer
;
989 unsigned long irq_flags
;
990 struct trace_event_call
*call
= &tk
->tp
.call
;
992 WARN_ON(call
!= trace_file
->event_call
);
994 if (trace_trigger_soft_disabled(trace_file
))
997 local_save_flags(irq_flags
);
998 pc
= preempt_count();
1000 dsize
= __get_data_size(&tk
->tp
, regs
);
1001 size
= sizeof(*entry
) + tk
->tp
.size
+ dsize
;
1003 event
= trace_event_buffer_lock_reserve(&buffer
, trace_file
,
1005 size
, irq_flags
, pc
);
1009 entry
= ring_buffer_event_data(event
);
1010 entry
->ip
= (unsigned long)tk
->rp
.kp
.addr
;
1011 store_trace_args(sizeof(*entry
), &tk
->tp
, regs
, (u8
*)&entry
[1], dsize
);
1013 event_trigger_unlock_commit_regs(trace_file
, buffer
, event
,
1014 entry
, irq_flags
, pc
, regs
);
1018 kprobe_trace_func(struct trace_kprobe
*tk
, struct pt_regs
*regs
)
1020 struct event_file_link
*link
;
1022 list_for_each_entry_rcu(link
, &tk
->tp
.files
, list
)
1023 __kprobe_trace_func(tk
, regs
, link
->file
);
1025 NOKPROBE_SYMBOL(kprobe_trace_func
);
1027 /* Kretprobe handler */
1028 static nokprobe_inline
void
1029 __kretprobe_trace_func(struct trace_kprobe
*tk
, struct kretprobe_instance
*ri
,
1030 struct pt_regs
*regs
,
1031 struct trace_event_file
*trace_file
)
1033 struct kretprobe_trace_entry_head
*entry
;
1034 struct ring_buffer_event
*event
;
1035 struct ring_buffer
*buffer
;
1036 int size
, pc
, dsize
;
1037 unsigned long irq_flags
;
1038 struct trace_event_call
*call
= &tk
->tp
.call
;
1040 WARN_ON(call
!= trace_file
->event_call
);
1042 if (trace_trigger_soft_disabled(trace_file
))
1045 local_save_flags(irq_flags
);
1046 pc
= preempt_count();
1048 dsize
= __get_data_size(&tk
->tp
, regs
);
1049 size
= sizeof(*entry
) + tk
->tp
.size
+ dsize
;
1051 event
= trace_event_buffer_lock_reserve(&buffer
, trace_file
,
1053 size
, irq_flags
, pc
);
1057 entry
= ring_buffer_event_data(event
);
1058 entry
->func
= (unsigned long)tk
->rp
.kp
.addr
;
1059 entry
->ret_ip
= (unsigned long)ri
->ret_addr
;
1060 store_trace_args(sizeof(*entry
), &tk
->tp
, regs
, (u8
*)&entry
[1], dsize
);
1062 event_trigger_unlock_commit_regs(trace_file
, buffer
, event
,
1063 entry
, irq_flags
, pc
, regs
);
1067 kretprobe_trace_func(struct trace_kprobe
*tk
, struct kretprobe_instance
*ri
,
1068 struct pt_regs
*regs
)
1070 struct event_file_link
*link
;
1072 list_for_each_entry_rcu(link
, &tk
->tp
.files
, list
)
1073 __kretprobe_trace_func(tk
, ri
, regs
, link
->file
);
1075 NOKPROBE_SYMBOL(kretprobe_trace_func
);
1077 /* Event entry printers */
1078 static enum print_line_t
1079 print_kprobe_event(struct trace_iterator
*iter
, int flags
,
1080 struct trace_event
*event
)
1082 struct kprobe_trace_entry_head
*field
;
1083 struct trace_seq
*s
= &iter
->seq
;
1084 struct trace_probe
*tp
;
1088 field
= (struct kprobe_trace_entry_head
*)iter
->ent
;
1089 tp
= container_of(event
, struct trace_probe
, call
.event
);
1091 trace_seq_printf(s
, "%s: (", trace_event_name(&tp
->call
));
1093 if (!seq_print_ip_sym(s
, field
->ip
, flags
| TRACE_ITER_SYM_OFFSET
))
1096 trace_seq_putc(s
, ')');
1098 data
= (u8
*)&field
[1];
1099 for (i
= 0; i
< tp
->nr_args
; i
++)
1100 if (!tp
->args
[i
].type
->print(s
, tp
->args
[i
].name
,
1101 data
+ tp
->args
[i
].offset
, field
))
1104 trace_seq_putc(s
, '\n');
1106 return trace_handle_return(s
);
1109 static enum print_line_t
1110 print_kretprobe_event(struct trace_iterator
*iter
, int flags
,
1111 struct trace_event
*event
)
1113 struct kretprobe_trace_entry_head
*field
;
1114 struct trace_seq
*s
= &iter
->seq
;
1115 struct trace_probe
*tp
;
1119 field
= (struct kretprobe_trace_entry_head
*)iter
->ent
;
1120 tp
= container_of(event
, struct trace_probe
, call
.event
);
1122 trace_seq_printf(s
, "%s: (", trace_event_name(&tp
->call
));
1124 if (!seq_print_ip_sym(s
, field
->ret_ip
, flags
| TRACE_ITER_SYM_OFFSET
))
1127 trace_seq_puts(s
, " <- ");
1129 if (!seq_print_ip_sym(s
, field
->func
, flags
& ~TRACE_ITER_SYM_OFFSET
))
1132 trace_seq_putc(s
, ')');
1134 data
= (u8
*)&field
[1];
1135 for (i
= 0; i
< tp
->nr_args
; i
++)
1136 if (!tp
->args
[i
].type
->print(s
, tp
->args
[i
].name
,
1137 data
+ tp
->args
[i
].offset
, field
))
1140 trace_seq_putc(s
, '\n');
1143 return trace_handle_return(s
);
1147 static int kprobe_event_define_fields(struct trace_event_call
*event_call
)
1150 struct kprobe_trace_entry_head field
;
1151 struct trace_kprobe
*tk
= (struct trace_kprobe
*)event_call
->data
;
1153 DEFINE_FIELD(unsigned long, ip
, FIELD_STRING_IP
, 0);
1154 /* Set argument names as fields */
1155 for (i
= 0; i
< tk
->tp
.nr_args
; i
++) {
1156 struct probe_arg
*parg
= &tk
->tp
.args
[i
];
1158 ret
= trace_define_field(event_call
, parg
->type
->fmttype
,
1160 sizeof(field
) + parg
->offset
,
1162 parg
->type
->is_signed
,
1170 static int kretprobe_event_define_fields(struct trace_event_call
*event_call
)
1173 struct kretprobe_trace_entry_head field
;
1174 struct trace_kprobe
*tk
= (struct trace_kprobe
*)event_call
->data
;
1176 DEFINE_FIELD(unsigned long, func
, FIELD_STRING_FUNC
, 0);
1177 DEFINE_FIELD(unsigned long, ret_ip
, FIELD_STRING_RETIP
, 0);
1178 /* Set argument names as fields */
1179 for (i
= 0; i
< tk
->tp
.nr_args
; i
++) {
1180 struct probe_arg
*parg
= &tk
->tp
.args
[i
];
1182 ret
= trace_define_field(event_call
, parg
->type
->fmttype
,
1184 sizeof(field
) + parg
->offset
,
1186 parg
->type
->is_signed
,
1194 #ifdef CONFIG_PERF_EVENTS
1196 /* Kprobe profile handler */
1198 kprobe_perf_func(struct trace_kprobe
*tk
, struct pt_regs
*regs
)
1200 struct trace_event_call
*call
= &tk
->tp
.call
;
1201 struct kprobe_trace_entry_head
*entry
;
1202 struct hlist_head
*head
;
1203 int size
, __size
, dsize
;
1206 if (bpf_prog_array_valid(call
)) {
1207 unsigned long orig_ip
= instruction_pointer(regs
);
1210 ret
= trace_call_bpf(call
, regs
);
1213 * We need to check and see if we modified the pc of the
1214 * pt_regs, and if so clear the kprobe and return 1 so that we
1215 * don't do the single stepping.
1216 * The ftrace kprobe handler leaves it up to us to re-enable
1217 * preemption here before returning if we've modified the ip.
1219 if (orig_ip
!= instruction_pointer(regs
)) {
1220 reset_current_kprobe();
1221 preempt_enable_no_resched();
1228 head
= this_cpu_ptr(call
->perf_events
);
1229 if (hlist_empty(head
))
1232 dsize
= __get_data_size(&tk
->tp
, regs
);
1233 __size
= sizeof(*entry
) + tk
->tp
.size
+ dsize
;
1234 size
= ALIGN(__size
+ sizeof(u32
), sizeof(u64
));
1235 size
-= sizeof(u32
);
1237 entry
= perf_trace_buf_alloc(size
, NULL
, &rctx
);
1241 entry
->ip
= (unsigned long)tk
->rp
.kp
.addr
;
1242 memset(&entry
[1], 0, dsize
);
1243 store_trace_args(sizeof(*entry
), &tk
->tp
, regs
, (u8
*)&entry
[1], dsize
);
1244 perf_trace_buf_submit(entry
, size
, rctx
, call
->event
.type
, 1, regs
,
1248 NOKPROBE_SYMBOL(kprobe_perf_func
);
1250 /* Kretprobe profile handler */
1252 kretprobe_perf_func(struct trace_kprobe
*tk
, struct kretprobe_instance
*ri
,
1253 struct pt_regs
*regs
)
1255 struct trace_event_call
*call
= &tk
->tp
.call
;
1256 struct kretprobe_trace_entry_head
*entry
;
1257 struct hlist_head
*head
;
1258 int size
, __size
, dsize
;
1261 if (bpf_prog_array_valid(call
) && !trace_call_bpf(call
, regs
))
1264 head
= this_cpu_ptr(call
->perf_events
);
1265 if (hlist_empty(head
))
1268 dsize
= __get_data_size(&tk
->tp
, regs
);
1269 __size
= sizeof(*entry
) + tk
->tp
.size
+ dsize
;
1270 size
= ALIGN(__size
+ sizeof(u32
), sizeof(u64
));
1271 size
-= sizeof(u32
);
1273 entry
= perf_trace_buf_alloc(size
, NULL
, &rctx
);
1277 entry
->func
= (unsigned long)tk
->rp
.kp
.addr
;
1278 entry
->ret_ip
= (unsigned long)ri
->ret_addr
;
1279 store_trace_args(sizeof(*entry
), &tk
->tp
, regs
, (u8
*)&entry
[1], dsize
);
1280 perf_trace_buf_submit(entry
, size
, rctx
, call
->event
.type
, 1, regs
,
1283 NOKPROBE_SYMBOL(kretprobe_perf_func
);
1284 #endif /* CONFIG_PERF_EVENTS */
1287 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1289 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1290 * lockless, but we can't race with this __init function.
1292 static int kprobe_register(struct trace_event_call
*event
,
1293 enum trace_reg type
, void *data
)
1295 struct trace_kprobe
*tk
= (struct trace_kprobe
*)event
->data
;
1296 struct trace_event_file
*file
= data
;
1299 case TRACE_REG_REGISTER
:
1300 return enable_trace_kprobe(tk
, file
);
1301 case TRACE_REG_UNREGISTER
:
1302 return disable_trace_kprobe(tk
, file
);
1304 #ifdef CONFIG_PERF_EVENTS
1305 case TRACE_REG_PERF_REGISTER
:
1306 return enable_trace_kprobe(tk
, NULL
);
1307 case TRACE_REG_PERF_UNREGISTER
:
1308 return disable_trace_kprobe(tk
, NULL
);
1309 case TRACE_REG_PERF_OPEN
:
1310 case TRACE_REG_PERF_CLOSE
:
1311 case TRACE_REG_PERF_ADD
:
1312 case TRACE_REG_PERF_DEL
:
1319 static int kprobe_dispatcher(struct kprobe
*kp
, struct pt_regs
*regs
)
1321 struct trace_kprobe
*tk
= container_of(kp
, struct trace_kprobe
, rp
.kp
);
1324 raw_cpu_inc(*tk
->nhit
);
1326 if (tk
->tp
.flags
& TP_FLAG_TRACE
)
1327 kprobe_trace_func(tk
, regs
);
1328 #ifdef CONFIG_PERF_EVENTS
1329 if (tk
->tp
.flags
& TP_FLAG_PROFILE
)
1330 ret
= kprobe_perf_func(tk
, regs
);
1334 NOKPROBE_SYMBOL(kprobe_dispatcher
);
1337 kretprobe_dispatcher(struct kretprobe_instance
*ri
, struct pt_regs
*regs
)
1339 struct trace_kprobe
*tk
= container_of(ri
->rp
, struct trace_kprobe
, rp
);
1341 raw_cpu_inc(*tk
->nhit
);
1343 if (tk
->tp
.flags
& TP_FLAG_TRACE
)
1344 kretprobe_trace_func(tk
, ri
, regs
);
1345 #ifdef CONFIG_PERF_EVENTS
1346 if (tk
->tp
.flags
& TP_FLAG_PROFILE
)
1347 kretprobe_perf_func(tk
, ri
, regs
);
1349 return 0; /* We don't tweek kernel, so just return 0 */
1351 NOKPROBE_SYMBOL(kretprobe_dispatcher
);
1353 static struct trace_event_functions kretprobe_funcs
= {
1354 .trace
= print_kretprobe_event
1357 static struct trace_event_functions kprobe_funcs
= {
1358 .trace
= print_kprobe_event
1361 static int register_kprobe_event(struct trace_kprobe
*tk
)
1363 struct trace_event_call
*call
= &tk
->tp
.call
;
1366 /* Initialize trace_event_call */
1367 INIT_LIST_HEAD(&call
->class->fields
);
1368 if (trace_kprobe_is_return(tk
)) {
1369 call
->event
.funcs
= &kretprobe_funcs
;
1370 call
->class->define_fields
= kretprobe_event_define_fields
;
1372 call
->event
.funcs
= &kprobe_funcs
;
1373 call
->class->define_fields
= kprobe_event_define_fields
;
1375 if (set_print_fmt(&tk
->tp
, trace_kprobe_is_return(tk
)) < 0)
1377 ret
= register_trace_event(&call
->event
);
1379 kfree(call
->print_fmt
);
1382 call
->flags
= TRACE_EVENT_FL_KPROBE
;
1383 call
->class->reg
= kprobe_register
;
1385 ret
= trace_add_event_call(call
);
1387 pr_info("Failed to register kprobe event: %s\n",
1388 trace_event_name(call
));
1389 kfree(call
->print_fmt
);
1390 unregister_trace_event(&call
->event
);
1395 static int unregister_kprobe_event(struct trace_kprobe
*tk
)
1399 /* tp->event is unregistered in trace_remove_event_call() */
1400 ret
= trace_remove_event_call(&tk
->tp
.call
);
1402 kfree(tk
->tp
.call
.print_fmt
);
1406 /* Make a tracefs interface for controlling probe points */
1407 static __init
int init_kprobe_trace(void)
1409 struct dentry
*d_tracer
;
1410 struct dentry
*entry
;
1412 if (register_module_notifier(&trace_kprobe_module_nb
))
1415 d_tracer
= tracing_init_dentry();
1416 if (IS_ERR(d_tracer
))
1419 entry
= tracefs_create_file("kprobe_events", 0644, d_tracer
,
1420 NULL
, &kprobe_events_ops
);
1422 /* Event list interface */
1424 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1426 /* Profile interface */
1427 entry
= tracefs_create_file("kprobe_profile", 0444, d_tracer
,
1428 NULL
, &kprobe_profile_ops
);
1431 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1434 fs_initcall(init_kprobe_trace
);
1437 #ifdef CONFIG_FTRACE_STARTUP_TEST
1439 * The "__used" keeps gcc from removing the function symbol
1440 * from the kallsyms table. 'noinline' makes sure that there
1441 * isn't an inlined version used by the test method below
1443 static __used __init noinline
int
1444 kprobe_trace_selftest_target(int a1
, int a2
, int a3
, int a4
, int a5
, int a6
)
1446 return a1
+ a2
+ a3
+ a4
+ a5
+ a6
;
1449 static __init
struct trace_event_file
*
1450 find_trace_probe_file(struct trace_kprobe
*tk
, struct trace_array
*tr
)
1452 struct trace_event_file
*file
;
1454 list_for_each_entry(file
, &tr
->events
, list
)
1455 if (file
->event_call
== &tk
->tp
.call
)
1462 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1463 * stage, we can do this lockless.
1465 static __init
int kprobe_trace_self_tests_init(void)
1468 int (*target
)(int, int, int, int, int, int);
1469 struct trace_kprobe
*tk
;
1470 struct trace_event_file
*file
;
1472 if (tracing_is_disabled())
1475 target
= kprobe_trace_selftest_target
;
1477 pr_info("Testing kprobe tracing: ");
1479 ret
= trace_run_command("p:testprobe kprobe_trace_selftest_target "
1480 "$stack $stack0 +0($stack)",
1481 create_trace_kprobe
);
1482 if (WARN_ON_ONCE(ret
)) {
1483 pr_warn("error on probing function entry.\n");
1486 /* Enable trace point */
1487 tk
= find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM
);
1488 if (WARN_ON_ONCE(tk
== NULL
)) {
1489 pr_warn("error on getting new probe.\n");
1492 file
= find_trace_probe_file(tk
, top_trace_array());
1493 if (WARN_ON_ONCE(file
== NULL
)) {
1494 pr_warn("error on getting probe file.\n");
1497 enable_trace_kprobe(tk
, file
);
1501 ret
= trace_run_command("r:testprobe2 kprobe_trace_selftest_target "
1502 "$retval", create_trace_kprobe
);
1503 if (WARN_ON_ONCE(ret
)) {
1504 pr_warn("error on probing function return.\n");
1507 /* Enable trace point */
1508 tk
= find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM
);
1509 if (WARN_ON_ONCE(tk
== NULL
)) {
1510 pr_warn("error on getting 2nd new probe.\n");
1513 file
= find_trace_probe_file(tk
, top_trace_array());
1514 if (WARN_ON_ONCE(file
== NULL
)) {
1515 pr_warn("error on getting probe file.\n");
1518 enable_trace_kprobe(tk
, file
);
1525 ret
= target(1, 2, 3, 4, 5, 6);
1528 * Not expecting an error here, the check is only to prevent the
1529 * optimizer from removing the call to target() as otherwise there
1530 * are no side-effects and the call is never performed.
1535 /* Disable trace points before removing it */
1536 tk
= find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM
);
1537 if (WARN_ON_ONCE(tk
== NULL
)) {
1538 pr_warn("error on getting test probe.\n");
1541 if (trace_kprobe_nhit(tk
) != 1) {
1542 pr_warn("incorrect number of testprobe hits\n");
1546 file
= find_trace_probe_file(tk
, top_trace_array());
1547 if (WARN_ON_ONCE(file
== NULL
)) {
1548 pr_warn("error on getting probe file.\n");
1551 disable_trace_kprobe(tk
, file
);
1554 tk
= find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM
);
1555 if (WARN_ON_ONCE(tk
== NULL
)) {
1556 pr_warn("error on getting 2nd test probe.\n");
1559 if (trace_kprobe_nhit(tk
) != 1) {
1560 pr_warn("incorrect number of testprobe2 hits\n");
1564 file
= find_trace_probe_file(tk
, top_trace_array());
1565 if (WARN_ON_ONCE(file
== NULL
)) {
1566 pr_warn("error on getting probe file.\n");
1569 disable_trace_kprobe(tk
, file
);
1572 ret
= trace_run_command("-:testprobe", create_trace_kprobe
);
1573 if (WARN_ON_ONCE(ret
)) {
1574 pr_warn("error on deleting a probe.\n");
1578 ret
= trace_run_command("-:testprobe2", create_trace_kprobe
);
1579 if (WARN_ON_ONCE(ret
)) {
1580 pr_warn("error on deleting a probe.\n");
1585 release_all_trace_kprobes();
1587 * Wait for the optimizer work to finish. Otherwise it might fiddle
1588 * with probes in already freed __init text.
1590 wait_for_kprobe_optimizer();
1592 pr_cont("NG: Some tests are failed. Please check them.\n");
1598 late_initcall(kprobe_trace_self_tests_init
);