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>
23 #include "trace_probe.h"
25 #define KPROBE_EVENT_SYSTEM "kprobes"
28 * Kprobe event core functions
31 struct list_head list
;
32 struct kretprobe rp
; /* Use rp.kp for kprobe use */
33 unsigned long __percpu
*nhit
;
34 const char *symbol
; /* symbol name */
35 struct trace_probe tp
;
38 #define SIZEOF_TRACE_KPROBE(n) \
39 (offsetof(struct trace_kprobe, tp.args) + \
40 (sizeof(struct probe_arg) * (n)))
43 static nokprobe_inline
bool trace_kprobe_is_return(struct trace_kprobe
*tk
)
45 return tk
->rp
.handler
!= NULL
;
48 static nokprobe_inline
const char *trace_kprobe_symbol(struct trace_kprobe
*tk
)
50 return tk
->symbol
? tk
->symbol
: "unknown";
53 static nokprobe_inline
unsigned long trace_kprobe_offset(struct trace_kprobe
*tk
)
55 return tk
->rp
.kp
.offset
;
58 static nokprobe_inline
bool trace_kprobe_has_gone(struct trace_kprobe
*tk
)
60 return !!(kprobe_gone(&tk
->rp
.kp
));
63 static nokprobe_inline
bool trace_kprobe_within_module(struct trace_kprobe
*tk
,
66 int len
= strlen(mod
->name
);
67 const char *name
= trace_kprobe_symbol(tk
);
68 return strncmp(mod
->name
, name
, len
) == 0 && name
[len
] == ':';
71 static nokprobe_inline
bool trace_kprobe_is_on_module(struct trace_kprobe
*tk
)
73 return !!strchr(trace_kprobe_symbol(tk
), ':');
76 static nokprobe_inline
unsigned long trace_kprobe_nhit(struct trace_kprobe
*tk
)
78 unsigned long nhit
= 0;
81 for_each_possible_cpu(cpu
)
82 nhit
+= *per_cpu_ptr(tk
->nhit
, cpu
);
87 static int register_kprobe_event(struct trace_kprobe
*tk
);
88 static int unregister_kprobe_event(struct trace_kprobe
*tk
);
90 static DEFINE_MUTEX(probe_lock
);
91 static LIST_HEAD(probe_list
);
93 static int kprobe_dispatcher(struct kprobe
*kp
, struct pt_regs
*regs
);
94 static int kretprobe_dispatcher(struct kretprobe_instance
*ri
,
95 struct pt_regs
*regs
);
97 /* Memory fetching by symbol */
104 unsigned long update_symbol_cache(struct symbol_cache
*sc
)
106 sc
->addr
= (unsigned long)kallsyms_lookup_name(sc
->symbol
);
109 sc
->addr
+= sc
->offset
;
114 void free_symbol_cache(struct symbol_cache
*sc
)
120 struct symbol_cache
*alloc_symbol_cache(const char *sym
, long offset
)
122 struct symbol_cache
*sc
;
124 if (!sym
|| strlen(sym
) == 0)
127 sc
= kzalloc(sizeof(struct symbol_cache
), GFP_KERNEL
);
131 sc
->symbol
= kstrdup(sym
, GFP_KERNEL
);
137 update_symbol_cache(sc
);
143 * Kprobes-specific fetch functions
145 #define DEFINE_FETCH_stack(type) \
146 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs, \
147 void *offset, void *dest) \
149 *(type *)dest = (type)regs_get_kernel_stack_nth(regs, \
150 (unsigned int)((unsigned long)offset)); \
152 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
154 DEFINE_BASIC_FETCH_FUNCS(stack
)
155 /* No string on the stack entry */
156 #define fetch_stack_string NULL
157 #define fetch_stack_string_size NULL
159 #define DEFINE_FETCH_memory(type) \
160 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs, \
161 void *addr, void *dest) \
164 if (probe_kernel_address(addr, retval)) \
167 *(type *)dest = retval; \
169 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
171 DEFINE_BASIC_FETCH_FUNCS(memory
)
173 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
174 * length and relative data location.
176 static void FETCH_FUNC_NAME(memory
, string
)(struct pt_regs
*regs
,
177 void *addr
, void *dest
)
179 int maxlen
= get_rloc_len(*(u32
*)dest
);
180 u8
*dst
= get_rloc_data(dest
);
187 * Try to get string again, since the string can be changed while
190 ret
= strncpy_from_unsafe(dst
, addr
, maxlen
);
192 if (ret
< 0) { /* Failed to fetch string */
194 *(u32
*)dest
= make_data_rloc(0, get_rloc_offs(*(u32
*)dest
));
196 *(u32
*)dest
= make_data_rloc(ret
, get_rloc_offs(*(u32
*)dest
));
199 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory
, string
));
201 /* Return the length of string -- including null terminal byte */
202 static void FETCH_FUNC_NAME(memory
, string_size
)(struct pt_regs
*regs
,
203 void *addr
, void *dest
)
214 ret
= __copy_from_user_inatomic(&c
, (u8
*)addr
+ len
, 1);
216 } while (c
&& ret
== 0 && len
< MAX_STRING_SIZE
);
221 if (ret
< 0) /* Failed to check the length */
226 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory
, string_size
));
228 #define DEFINE_FETCH_symbol(type) \
229 void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
231 struct symbol_cache *sc = data; \
233 fetch_memory_##type(regs, (void *)sc->addr, dest); \
237 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
239 DEFINE_BASIC_FETCH_FUNCS(symbol
)
240 DEFINE_FETCH_symbol(string
)
241 DEFINE_FETCH_symbol(string_size
)
243 /* kprobes don't support file_offset fetch methods */
244 #define fetch_file_offset_u8 NULL
245 #define fetch_file_offset_u16 NULL
246 #define fetch_file_offset_u32 NULL
247 #define fetch_file_offset_u64 NULL
248 #define fetch_file_offset_string NULL
249 #define fetch_file_offset_string_size NULL
251 /* Fetch type information table */
252 static const struct fetch_type kprobes_fetch_type_table
[] = {
254 [FETCH_TYPE_STRING
] = __ASSIGN_FETCH_TYPE("string", string
, string
,
255 sizeof(u32
), 1, "__data_loc char[]"),
256 [FETCH_TYPE_STRSIZE
] = __ASSIGN_FETCH_TYPE("string_size", u32
,
257 string_size
, sizeof(u32
), 0, "u32"),
259 ASSIGN_FETCH_TYPE(u8
, u8
, 0),
260 ASSIGN_FETCH_TYPE(u16
, u16
, 0),
261 ASSIGN_FETCH_TYPE(u32
, u32
, 0),
262 ASSIGN_FETCH_TYPE(u64
, u64
, 0),
263 ASSIGN_FETCH_TYPE(s8
, u8
, 1),
264 ASSIGN_FETCH_TYPE(s16
, u16
, 1),
265 ASSIGN_FETCH_TYPE(s32
, u32
, 1),
266 ASSIGN_FETCH_TYPE(s64
, u64
, 1),
267 ASSIGN_FETCH_TYPE_ALIAS(x8
, u8
, u8
, 0),
268 ASSIGN_FETCH_TYPE_ALIAS(x16
, u16
, u16
, 0),
269 ASSIGN_FETCH_TYPE_ALIAS(x32
, u32
, u32
, 0),
270 ASSIGN_FETCH_TYPE_ALIAS(x64
, u64
, u64
, 0),
272 ASSIGN_FETCH_TYPE_END
276 * Allocate new trace_probe and initialize it (including kprobes).
278 static struct trace_kprobe
*alloc_trace_kprobe(const char *group
,
283 int nargs
, bool is_return
)
285 struct trace_kprobe
*tk
;
288 tk
= kzalloc(SIZEOF_TRACE_KPROBE(nargs
), GFP_KERNEL
);
292 tk
->nhit
= alloc_percpu(unsigned long);
297 tk
->symbol
= kstrdup(symbol
, GFP_KERNEL
);
300 tk
->rp
.kp
.symbol_name
= tk
->symbol
;
301 tk
->rp
.kp
.offset
= offs
;
303 tk
->rp
.kp
.addr
= addr
;
306 tk
->rp
.handler
= kretprobe_dispatcher
;
308 tk
->rp
.kp
.pre_handler
= kprobe_dispatcher
;
310 if (!event
|| !is_good_name(event
)) {
315 tk
->tp
.call
.class = &tk
->tp
.class;
316 tk
->tp
.call
.name
= kstrdup(event
, GFP_KERNEL
);
317 if (!tk
->tp
.call
.name
)
320 if (!group
|| !is_good_name(group
)) {
325 tk
->tp
.class.system
= kstrdup(group
, GFP_KERNEL
);
326 if (!tk
->tp
.class.system
)
329 INIT_LIST_HEAD(&tk
->list
);
330 INIT_LIST_HEAD(&tk
->tp
.files
);
333 kfree(tk
->tp
.call
.name
);
335 free_percpu(tk
->nhit
);
340 static void free_trace_kprobe(struct trace_kprobe
*tk
)
344 for (i
= 0; i
< tk
->tp
.nr_args
; i
++)
345 traceprobe_free_probe_arg(&tk
->tp
.args
[i
]);
347 kfree(tk
->tp
.call
.class->system
);
348 kfree(tk
->tp
.call
.name
);
350 free_percpu(tk
->nhit
);
354 static struct trace_kprobe
*find_trace_kprobe(const char *event
,
357 struct trace_kprobe
*tk
;
359 list_for_each_entry(tk
, &probe_list
, list
)
360 if (strcmp(trace_event_name(&tk
->tp
.call
), event
) == 0 &&
361 strcmp(tk
->tp
.call
.class->system
, group
) == 0)
368 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
371 enable_trace_kprobe(struct trace_kprobe
*tk
, struct trace_event_file
*file
)
376 struct event_file_link
*link
;
378 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
385 list_add_tail_rcu(&link
->list
, &tk
->tp
.files
);
387 tk
->tp
.flags
|= TP_FLAG_TRACE
;
389 tk
->tp
.flags
|= TP_FLAG_PROFILE
;
391 if (trace_probe_is_registered(&tk
->tp
) && !trace_kprobe_has_gone(tk
)) {
392 if (trace_kprobe_is_return(tk
))
393 ret
= enable_kretprobe(&tk
->rp
);
395 ret
= enable_kprobe(&tk
->rp
.kp
);
402 * Disable trace_probe
403 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
406 disable_trace_kprobe(struct trace_kprobe
*tk
, struct trace_event_file
*file
)
408 struct event_file_link
*link
= NULL
;
413 link
= find_event_file_link(&tk
->tp
, file
);
419 list_del_rcu(&link
->list
);
421 if (!list_empty(&tk
->tp
.files
))
424 tk
->tp
.flags
&= ~TP_FLAG_TRACE
;
426 tk
->tp
.flags
&= ~TP_FLAG_PROFILE
;
428 if (!trace_probe_is_enabled(&tk
->tp
) && trace_probe_is_registered(&tk
->tp
)) {
429 if (trace_kprobe_is_return(tk
))
430 disable_kretprobe(&tk
->rp
);
432 disable_kprobe(&tk
->rp
.kp
);
438 * Synchronize with kprobe_trace_func/kretprobe_trace_func
439 * to ensure disabled (all running handlers are finished).
440 * This is not only for kfree(), but also the caller,
441 * trace_remove_event_call() supposes it for releasing
442 * event_call related objects, which will be accessed in
443 * the kprobe_trace_func/kretprobe_trace_func.
446 kfree(link
); /* Ignored if link == NULL */
452 /* Internal register function - just handle k*probes and flags */
453 static int __register_trace_kprobe(struct trace_kprobe
*tk
)
457 if (trace_probe_is_registered(&tk
->tp
))
460 for (i
= 0; i
< tk
->tp
.nr_args
; i
++)
461 traceprobe_update_arg(&tk
->tp
.args
[i
]);
463 /* Set/clear disabled flag according to tp->flag */
464 if (trace_probe_is_enabled(&tk
->tp
))
465 tk
->rp
.kp
.flags
&= ~KPROBE_FLAG_DISABLED
;
467 tk
->rp
.kp
.flags
|= KPROBE_FLAG_DISABLED
;
469 if (trace_kprobe_is_return(tk
))
470 ret
= register_kretprobe(&tk
->rp
);
472 ret
= register_kprobe(&tk
->rp
.kp
);
475 tk
->tp
.flags
|= TP_FLAG_REGISTERED
;
477 pr_warn("Could not insert probe at %s+%lu: %d\n",
478 trace_kprobe_symbol(tk
), trace_kprobe_offset(tk
), ret
);
479 if (ret
== -ENOENT
&& trace_kprobe_is_on_module(tk
)) {
480 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
482 } else if (ret
== -EILSEQ
) {
483 pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
492 /* Internal unregister function - just handle k*probes and flags */
493 static void __unregister_trace_kprobe(struct trace_kprobe
*tk
)
495 if (trace_probe_is_registered(&tk
->tp
)) {
496 if (trace_kprobe_is_return(tk
))
497 unregister_kretprobe(&tk
->rp
);
499 unregister_kprobe(&tk
->rp
.kp
);
500 tk
->tp
.flags
&= ~TP_FLAG_REGISTERED
;
501 /* Cleanup kprobe for reuse */
502 if (tk
->rp
.kp
.symbol_name
)
503 tk
->rp
.kp
.addr
= NULL
;
507 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
508 static int unregister_trace_kprobe(struct trace_kprobe
*tk
)
510 /* Enabled event can not be unregistered */
511 if (trace_probe_is_enabled(&tk
->tp
))
514 /* Will fail if probe is being used by ftrace or perf */
515 if (unregister_kprobe_event(tk
))
518 __unregister_trace_kprobe(tk
);
524 /* Register a trace_probe and probe_event */
525 static int register_trace_kprobe(struct trace_kprobe
*tk
)
527 struct trace_kprobe
*old_tk
;
530 mutex_lock(&probe_lock
);
532 /* Delete old (same name) event if exist */
533 old_tk
= find_trace_kprobe(trace_event_name(&tk
->tp
.call
),
534 tk
->tp
.call
.class->system
);
536 ret
= unregister_trace_kprobe(old_tk
);
539 free_trace_kprobe(old_tk
);
542 /* Register new event */
543 ret
= register_kprobe_event(tk
);
545 pr_warn("Failed to register probe event(%d)\n", ret
);
549 /* Register k*probe */
550 ret
= __register_trace_kprobe(tk
);
552 unregister_kprobe_event(tk
);
554 list_add_tail(&tk
->list
, &probe_list
);
557 mutex_unlock(&probe_lock
);
561 /* Module notifier call back, checking event on the module */
562 static int trace_kprobe_module_callback(struct notifier_block
*nb
,
563 unsigned long val
, void *data
)
565 struct module
*mod
= data
;
566 struct trace_kprobe
*tk
;
569 if (val
!= MODULE_STATE_COMING
)
572 /* Update probes on coming module */
573 mutex_lock(&probe_lock
);
574 list_for_each_entry(tk
, &probe_list
, list
) {
575 if (trace_kprobe_within_module(tk
, mod
)) {
576 /* Don't need to check busy - this should have gone. */
577 __unregister_trace_kprobe(tk
);
578 ret
= __register_trace_kprobe(tk
);
580 pr_warn("Failed to re-register probe %s on %s: %d\n",
581 trace_event_name(&tk
->tp
.call
),
585 mutex_unlock(&probe_lock
);
590 static struct notifier_block trace_kprobe_module_nb
= {
591 .notifier_call
= trace_kprobe_module_callback
,
592 .priority
= 1 /* Invoked after kprobe module callback */
595 static int create_trace_kprobe(int argc
, char **argv
)
599 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
600 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
602 * $retval : fetch return value
603 * $stack : fetch stack address
604 * $stackN : fetch Nth of stack (N:0-)
605 * $comm : fetch current task comm
606 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
607 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
608 * %REG : fetch register REG
609 * Dereferencing memory fetch:
610 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
611 * Alias name of args:
612 * NAME=FETCHARG : set NAME as alias of FETCHARG.
614 * FETCHARG:TYPE : use TYPE instead of unsigned long.
616 struct trace_kprobe
*tk
;
618 bool is_return
= false, is_delete
= false;
619 char *symbol
= NULL
, *event
= NULL
, *group
= NULL
;
621 unsigned long offset
= 0;
623 char buf
[MAX_EVENT_NAME_LEN
];
625 /* argc must be >= 1 */
626 if (argv
[0][0] == 'p')
628 else if (argv
[0][0] == 'r')
630 else if (argv
[0][0] == '-')
633 pr_info("Probe definition must be started with 'p', 'r' or"
638 if (argv
[0][1] == ':') {
640 if (strchr(event
, '/')) {
642 event
= strchr(group
, '/') + 1;
644 if (strlen(group
) == 0) {
645 pr_info("Group name is not specified\n");
649 if (strlen(event
) == 0) {
650 pr_info("Event name is not specified\n");
655 group
= KPROBE_EVENT_SYSTEM
;
659 pr_info("Delete command needs an event name.\n");
662 mutex_lock(&probe_lock
);
663 tk
= find_trace_kprobe(event
, group
);
665 mutex_unlock(&probe_lock
);
666 pr_info("Event %s/%s doesn't exist.\n", group
, event
);
669 /* delete an event */
670 ret
= unregister_trace_kprobe(tk
);
672 free_trace_kprobe(tk
);
673 mutex_unlock(&probe_lock
);
678 pr_info("Probe point is not specified.\n");
681 if (isdigit(argv
[1][0])) {
683 pr_info("Return probe point must be a symbol.\n");
686 /* an address specified */
687 ret
= kstrtoul(&argv
[1][0], 0, (unsigned long *)&addr
);
689 pr_info("Failed to parse address.\n");
693 /* a symbol specified */
695 /* TODO: support .init module functions */
696 ret
= traceprobe_split_symbol_offset(symbol
, &offset
);
698 pr_info("Failed to parse symbol.\n");
701 if (offset
&& is_return
) {
702 pr_info("Return probe must be used without offset.\n");
706 argc
-= 2; argv
+= 2;
710 /* Make a new event name */
712 snprintf(buf
, MAX_EVENT_NAME_LEN
, "%c_%s_%ld",
713 is_return
? 'r' : 'p', symbol
, offset
);
715 snprintf(buf
, MAX_EVENT_NAME_LEN
, "%c_0x%p",
716 is_return
? 'r' : 'p', addr
);
719 tk
= alloc_trace_kprobe(group
, event
, addr
, symbol
, offset
, argc
,
722 pr_info("Failed to allocate trace_probe.(%d)\n",
727 /* parse arguments */
729 for (i
= 0; i
< argc
&& i
< MAX_TRACE_ARGS
; i
++) {
730 struct probe_arg
*parg
= &tk
->tp
.args
[i
];
732 /* Increment count for freeing args in error case */
735 /* Parse argument name */
736 arg
= strchr(argv
[i
], '=');
739 parg
->name
= kstrdup(argv
[i
], GFP_KERNEL
);
742 /* If argument name is omitted, set "argN" */
743 snprintf(buf
, MAX_EVENT_NAME_LEN
, "arg%d", i
+ 1);
744 parg
->name
= kstrdup(buf
, GFP_KERNEL
);
748 pr_info("Failed to allocate argument[%d] name.\n", i
);
753 if (!is_good_name(parg
->name
)) {
754 pr_info("Invalid argument[%d] name: %s\n",
760 if (traceprobe_conflict_field_name(parg
->name
,
762 pr_info("Argument[%d] name '%s' conflicts with "
763 "another field.\n", i
, argv
[i
]);
768 /* Parse fetch argument */
769 ret
= traceprobe_parse_probe_arg(arg
, &tk
->tp
.size
, parg
,
771 kprobes_fetch_type_table
);
773 pr_info("Parse error at argument[%d]. (%d)\n", i
, ret
);
778 ret
= register_trace_kprobe(tk
);
784 free_trace_kprobe(tk
);
788 static int release_all_trace_kprobes(void)
790 struct trace_kprobe
*tk
;
793 mutex_lock(&probe_lock
);
794 /* Ensure no probe is in use. */
795 list_for_each_entry(tk
, &probe_list
, list
)
796 if (trace_probe_is_enabled(&tk
->tp
)) {
800 /* TODO: Use batch unregistration */
801 while (!list_empty(&probe_list
)) {
802 tk
= list_entry(probe_list
.next
, struct trace_kprobe
, list
);
803 ret
= unregister_trace_kprobe(tk
);
806 free_trace_kprobe(tk
);
810 mutex_unlock(&probe_lock
);
815 /* Probes listing interfaces */
816 static void *probes_seq_start(struct seq_file
*m
, loff_t
*pos
)
818 mutex_lock(&probe_lock
);
819 return seq_list_start(&probe_list
, *pos
);
822 static void *probes_seq_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
824 return seq_list_next(v
, &probe_list
, pos
);
827 static void probes_seq_stop(struct seq_file
*m
, void *v
)
829 mutex_unlock(&probe_lock
);
832 static int probes_seq_show(struct seq_file
*m
, void *v
)
834 struct trace_kprobe
*tk
= v
;
837 seq_putc(m
, trace_kprobe_is_return(tk
) ? 'r' : 'p');
838 seq_printf(m
, ":%s/%s", tk
->tp
.call
.class->system
,
839 trace_event_name(&tk
->tp
.call
));
842 seq_printf(m
, " 0x%p", tk
->rp
.kp
.addr
);
843 else if (tk
->rp
.kp
.offset
)
844 seq_printf(m
, " %s+%u", trace_kprobe_symbol(tk
),
847 seq_printf(m
, " %s", trace_kprobe_symbol(tk
));
849 for (i
= 0; i
< tk
->tp
.nr_args
; i
++)
850 seq_printf(m
, " %s=%s", tk
->tp
.args
[i
].name
, tk
->tp
.args
[i
].comm
);
856 static const struct seq_operations probes_seq_op
= {
857 .start
= probes_seq_start
,
858 .next
= probes_seq_next
,
859 .stop
= probes_seq_stop
,
860 .show
= probes_seq_show
863 static int probes_open(struct inode
*inode
, struct file
*file
)
867 if ((file
->f_mode
& FMODE_WRITE
) && (file
->f_flags
& O_TRUNC
)) {
868 ret
= release_all_trace_kprobes();
873 return seq_open(file
, &probes_seq_op
);
876 static ssize_t
probes_write(struct file
*file
, const char __user
*buffer
,
877 size_t count
, loff_t
*ppos
)
879 return traceprobe_probes_write(file
, buffer
, count
, ppos
,
880 create_trace_kprobe
);
883 static const struct file_operations kprobe_events_ops
= {
884 .owner
= THIS_MODULE
,
888 .release
= seq_release
,
889 .write
= probes_write
,
892 /* Probes profiling interfaces */
893 static int probes_profile_seq_show(struct seq_file
*m
, void *v
)
895 struct trace_kprobe
*tk
= v
;
897 seq_printf(m
, " %-44s %15lu %15lu\n",
898 trace_event_name(&tk
->tp
.call
),
899 trace_kprobe_nhit(tk
),
905 static const struct seq_operations profile_seq_op
= {
906 .start
= probes_seq_start
,
907 .next
= probes_seq_next
,
908 .stop
= probes_seq_stop
,
909 .show
= probes_profile_seq_show
912 static int profile_open(struct inode
*inode
, struct file
*file
)
914 return seq_open(file
, &profile_seq_op
);
917 static const struct file_operations kprobe_profile_ops
= {
918 .owner
= THIS_MODULE
,
919 .open
= profile_open
,
922 .release
= seq_release
,
926 static nokprobe_inline
void
927 __kprobe_trace_func(struct trace_kprobe
*tk
, struct pt_regs
*regs
,
928 struct trace_event_file
*trace_file
)
930 struct kprobe_trace_entry_head
*entry
;
931 struct ring_buffer_event
*event
;
932 struct ring_buffer
*buffer
;
934 unsigned long irq_flags
;
935 struct trace_event_call
*call
= &tk
->tp
.call
;
937 WARN_ON(call
!= trace_file
->event_call
);
939 if (trace_trigger_soft_disabled(trace_file
))
942 local_save_flags(irq_flags
);
943 pc
= preempt_count();
945 dsize
= __get_data_size(&tk
->tp
, regs
);
946 size
= sizeof(*entry
) + tk
->tp
.size
+ dsize
;
948 event
= trace_event_buffer_lock_reserve(&buffer
, trace_file
,
950 size
, irq_flags
, pc
);
954 entry
= ring_buffer_event_data(event
);
955 entry
->ip
= (unsigned long)tk
->rp
.kp
.addr
;
956 store_trace_args(sizeof(*entry
), &tk
->tp
, regs
, (u8
*)&entry
[1], dsize
);
958 event_trigger_unlock_commit_regs(trace_file
, buffer
, event
,
959 entry
, irq_flags
, pc
, regs
);
963 kprobe_trace_func(struct trace_kprobe
*tk
, struct pt_regs
*regs
)
965 struct event_file_link
*link
;
967 list_for_each_entry_rcu(link
, &tk
->tp
.files
, list
)
968 __kprobe_trace_func(tk
, regs
, link
->file
);
970 NOKPROBE_SYMBOL(kprobe_trace_func
);
972 /* Kretprobe handler */
973 static nokprobe_inline
void
974 __kretprobe_trace_func(struct trace_kprobe
*tk
, struct kretprobe_instance
*ri
,
975 struct pt_regs
*regs
,
976 struct trace_event_file
*trace_file
)
978 struct kretprobe_trace_entry_head
*entry
;
979 struct ring_buffer_event
*event
;
980 struct ring_buffer
*buffer
;
982 unsigned long irq_flags
;
983 struct trace_event_call
*call
= &tk
->tp
.call
;
985 WARN_ON(call
!= trace_file
->event_call
);
987 if (trace_trigger_soft_disabled(trace_file
))
990 local_save_flags(irq_flags
);
991 pc
= preempt_count();
993 dsize
= __get_data_size(&tk
->tp
, regs
);
994 size
= sizeof(*entry
) + tk
->tp
.size
+ dsize
;
996 event
= trace_event_buffer_lock_reserve(&buffer
, trace_file
,
998 size
, irq_flags
, pc
);
1002 entry
= ring_buffer_event_data(event
);
1003 entry
->func
= (unsigned long)tk
->rp
.kp
.addr
;
1004 entry
->ret_ip
= (unsigned long)ri
->ret_addr
;
1005 store_trace_args(sizeof(*entry
), &tk
->tp
, regs
, (u8
*)&entry
[1], dsize
);
1007 event_trigger_unlock_commit_regs(trace_file
, buffer
, event
,
1008 entry
, irq_flags
, pc
, regs
);
1012 kretprobe_trace_func(struct trace_kprobe
*tk
, struct kretprobe_instance
*ri
,
1013 struct pt_regs
*regs
)
1015 struct event_file_link
*link
;
1017 list_for_each_entry_rcu(link
, &tk
->tp
.files
, list
)
1018 __kretprobe_trace_func(tk
, ri
, regs
, link
->file
);
1020 NOKPROBE_SYMBOL(kretprobe_trace_func
);
1022 /* Event entry printers */
1023 static enum print_line_t
1024 print_kprobe_event(struct trace_iterator
*iter
, int flags
,
1025 struct trace_event
*event
)
1027 struct kprobe_trace_entry_head
*field
;
1028 struct trace_seq
*s
= &iter
->seq
;
1029 struct trace_probe
*tp
;
1033 field
= (struct kprobe_trace_entry_head
*)iter
->ent
;
1034 tp
= container_of(event
, struct trace_probe
, call
.event
);
1036 trace_seq_printf(s
, "%s: (", trace_event_name(&tp
->call
));
1038 if (!seq_print_ip_sym(s
, field
->ip
, flags
| TRACE_ITER_SYM_OFFSET
))
1041 trace_seq_putc(s
, ')');
1043 data
= (u8
*)&field
[1];
1044 for (i
= 0; i
< tp
->nr_args
; i
++)
1045 if (!tp
->args
[i
].type
->print(s
, tp
->args
[i
].name
,
1046 data
+ tp
->args
[i
].offset
, field
))
1049 trace_seq_putc(s
, '\n');
1051 return trace_handle_return(s
);
1054 static enum print_line_t
1055 print_kretprobe_event(struct trace_iterator
*iter
, int flags
,
1056 struct trace_event
*event
)
1058 struct kretprobe_trace_entry_head
*field
;
1059 struct trace_seq
*s
= &iter
->seq
;
1060 struct trace_probe
*tp
;
1064 field
= (struct kretprobe_trace_entry_head
*)iter
->ent
;
1065 tp
= container_of(event
, struct trace_probe
, call
.event
);
1067 trace_seq_printf(s
, "%s: (", trace_event_name(&tp
->call
));
1069 if (!seq_print_ip_sym(s
, field
->ret_ip
, flags
| TRACE_ITER_SYM_OFFSET
))
1072 trace_seq_puts(s
, " <- ");
1074 if (!seq_print_ip_sym(s
, field
->func
, flags
& ~TRACE_ITER_SYM_OFFSET
))
1077 trace_seq_putc(s
, ')');
1079 data
= (u8
*)&field
[1];
1080 for (i
= 0; i
< tp
->nr_args
; i
++)
1081 if (!tp
->args
[i
].type
->print(s
, tp
->args
[i
].name
,
1082 data
+ tp
->args
[i
].offset
, field
))
1085 trace_seq_putc(s
, '\n');
1088 return trace_handle_return(s
);
1092 static int kprobe_event_define_fields(struct trace_event_call
*event_call
)
1095 struct kprobe_trace_entry_head field
;
1096 struct trace_kprobe
*tk
= (struct trace_kprobe
*)event_call
->data
;
1098 DEFINE_FIELD(unsigned long, ip
, FIELD_STRING_IP
, 0);
1099 /* Set argument names as fields */
1100 for (i
= 0; i
< tk
->tp
.nr_args
; i
++) {
1101 struct probe_arg
*parg
= &tk
->tp
.args
[i
];
1103 ret
= trace_define_field(event_call
, parg
->type
->fmttype
,
1105 sizeof(field
) + parg
->offset
,
1107 parg
->type
->is_signed
,
1115 static int kretprobe_event_define_fields(struct trace_event_call
*event_call
)
1118 struct kretprobe_trace_entry_head field
;
1119 struct trace_kprobe
*tk
= (struct trace_kprobe
*)event_call
->data
;
1121 DEFINE_FIELD(unsigned long, func
, FIELD_STRING_FUNC
, 0);
1122 DEFINE_FIELD(unsigned long, ret_ip
, FIELD_STRING_RETIP
, 0);
1123 /* Set argument names as fields */
1124 for (i
= 0; i
< tk
->tp
.nr_args
; i
++) {
1125 struct probe_arg
*parg
= &tk
->tp
.args
[i
];
1127 ret
= trace_define_field(event_call
, parg
->type
->fmttype
,
1129 sizeof(field
) + parg
->offset
,
1131 parg
->type
->is_signed
,
1139 #ifdef CONFIG_PERF_EVENTS
1141 /* Kprobe profile handler */
1143 kprobe_perf_func(struct trace_kprobe
*tk
, struct pt_regs
*regs
)
1145 struct trace_event_call
*call
= &tk
->tp
.call
;
1146 struct bpf_prog
*prog
= call
->prog
;
1147 struct kprobe_trace_entry_head
*entry
;
1148 struct hlist_head
*head
;
1149 int size
, __size
, dsize
;
1152 if (prog
&& !trace_call_bpf(prog
, regs
))
1155 head
= this_cpu_ptr(call
->perf_events
);
1156 if (hlist_empty(head
))
1159 dsize
= __get_data_size(&tk
->tp
, regs
);
1160 __size
= sizeof(*entry
) + tk
->tp
.size
+ dsize
;
1161 size
= ALIGN(__size
+ sizeof(u32
), sizeof(u64
));
1162 size
-= sizeof(u32
);
1164 entry
= perf_trace_buf_alloc(size
, NULL
, &rctx
);
1168 entry
->ip
= (unsigned long)tk
->rp
.kp
.addr
;
1169 memset(&entry
[1], 0, dsize
);
1170 store_trace_args(sizeof(*entry
), &tk
->tp
, regs
, (u8
*)&entry
[1], dsize
);
1171 perf_trace_buf_submit(entry
, size
, rctx
, call
->event
.type
, 1, regs
,
1174 NOKPROBE_SYMBOL(kprobe_perf_func
);
1176 /* Kretprobe profile handler */
1178 kretprobe_perf_func(struct trace_kprobe
*tk
, struct kretprobe_instance
*ri
,
1179 struct pt_regs
*regs
)
1181 struct trace_event_call
*call
= &tk
->tp
.call
;
1182 struct bpf_prog
*prog
= call
->prog
;
1183 struct kretprobe_trace_entry_head
*entry
;
1184 struct hlist_head
*head
;
1185 int size
, __size
, dsize
;
1188 if (prog
&& !trace_call_bpf(prog
, regs
))
1191 head
= this_cpu_ptr(call
->perf_events
);
1192 if (hlist_empty(head
))
1195 dsize
= __get_data_size(&tk
->tp
, regs
);
1196 __size
= sizeof(*entry
) + tk
->tp
.size
+ dsize
;
1197 size
= ALIGN(__size
+ sizeof(u32
), sizeof(u64
));
1198 size
-= sizeof(u32
);
1200 entry
= perf_trace_buf_alloc(size
, NULL
, &rctx
);
1204 entry
->func
= (unsigned long)tk
->rp
.kp
.addr
;
1205 entry
->ret_ip
= (unsigned long)ri
->ret_addr
;
1206 store_trace_args(sizeof(*entry
), &tk
->tp
, regs
, (u8
*)&entry
[1], dsize
);
1207 perf_trace_buf_submit(entry
, size
, rctx
, call
->event
.type
, 1, regs
,
1210 NOKPROBE_SYMBOL(kretprobe_perf_func
);
1211 #endif /* CONFIG_PERF_EVENTS */
1214 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1216 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1217 * lockless, but we can't race with this __init function.
1219 static int kprobe_register(struct trace_event_call
*event
,
1220 enum trace_reg type
, void *data
)
1222 struct trace_kprobe
*tk
= (struct trace_kprobe
*)event
->data
;
1223 struct trace_event_file
*file
= data
;
1226 case TRACE_REG_REGISTER
:
1227 return enable_trace_kprobe(tk
, file
);
1228 case TRACE_REG_UNREGISTER
:
1229 return disable_trace_kprobe(tk
, file
);
1231 #ifdef CONFIG_PERF_EVENTS
1232 case TRACE_REG_PERF_REGISTER
:
1233 return enable_trace_kprobe(tk
, NULL
);
1234 case TRACE_REG_PERF_UNREGISTER
:
1235 return disable_trace_kprobe(tk
, NULL
);
1236 case TRACE_REG_PERF_OPEN
:
1237 case TRACE_REG_PERF_CLOSE
:
1238 case TRACE_REG_PERF_ADD
:
1239 case TRACE_REG_PERF_DEL
:
1246 static int kprobe_dispatcher(struct kprobe
*kp
, struct pt_regs
*regs
)
1248 struct trace_kprobe
*tk
= container_of(kp
, struct trace_kprobe
, rp
.kp
);
1250 raw_cpu_inc(*tk
->nhit
);
1252 if (tk
->tp
.flags
& TP_FLAG_TRACE
)
1253 kprobe_trace_func(tk
, regs
);
1254 #ifdef CONFIG_PERF_EVENTS
1255 if (tk
->tp
.flags
& TP_FLAG_PROFILE
)
1256 kprobe_perf_func(tk
, regs
);
1258 return 0; /* We don't tweek kernel, so just return 0 */
1260 NOKPROBE_SYMBOL(kprobe_dispatcher
);
1263 kretprobe_dispatcher(struct kretprobe_instance
*ri
, struct pt_regs
*regs
)
1265 struct trace_kprobe
*tk
= container_of(ri
->rp
, struct trace_kprobe
, rp
);
1267 raw_cpu_inc(*tk
->nhit
);
1269 if (tk
->tp
.flags
& TP_FLAG_TRACE
)
1270 kretprobe_trace_func(tk
, ri
, regs
);
1271 #ifdef CONFIG_PERF_EVENTS
1272 if (tk
->tp
.flags
& TP_FLAG_PROFILE
)
1273 kretprobe_perf_func(tk
, ri
, regs
);
1275 return 0; /* We don't tweek kernel, so just return 0 */
1277 NOKPROBE_SYMBOL(kretprobe_dispatcher
);
1279 static struct trace_event_functions kretprobe_funcs
= {
1280 .trace
= print_kretprobe_event
1283 static struct trace_event_functions kprobe_funcs
= {
1284 .trace
= print_kprobe_event
1287 static int register_kprobe_event(struct trace_kprobe
*tk
)
1289 struct trace_event_call
*call
= &tk
->tp
.call
;
1292 /* Initialize trace_event_call */
1293 INIT_LIST_HEAD(&call
->class->fields
);
1294 if (trace_kprobe_is_return(tk
)) {
1295 call
->event
.funcs
= &kretprobe_funcs
;
1296 call
->class->define_fields
= kretprobe_event_define_fields
;
1298 call
->event
.funcs
= &kprobe_funcs
;
1299 call
->class->define_fields
= kprobe_event_define_fields
;
1301 if (set_print_fmt(&tk
->tp
, trace_kprobe_is_return(tk
)) < 0)
1303 ret
= register_trace_event(&call
->event
);
1305 kfree(call
->print_fmt
);
1308 call
->flags
= TRACE_EVENT_FL_KPROBE
;
1309 call
->class->reg
= kprobe_register
;
1311 ret
= trace_add_event_call(call
);
1313 pr_info("Failed to register kprobe event: %s\n",
1314 trace_event_name(call
));
1315 kfree(call
->print_fmt
);
1316 unregister_trace_event(&call
->event
);
1321 static int unregister_kprobe_event(struct trace_kprobe
*tk
)
1325 /* tp->event is unregistered in trace_remove_event_call() */
1326 ret
= trace_remove_event_call(&tk
->tp
.call
);
1328 kfree(tk
->tp
.call
.print_fmt
);
1332 /* Make a tracefs interface for controlling probe points */
1333 static __init
int init_kprobe_trace(void)
1335 struct dentry
*d_tracer
;
1336 struct dentry
*entry
;
1338 if (register_module_notifier(&trace_kprobe_module_nb
))
1341 d_tracer
= tracing_init_dentry();
1342 if (IS_ERR(d_tracer
))
1345 entry
= tracefs_create_file("kprobe_events", 0644, d_tracer
,
1346 NULL
, &kprobe_events_ops
);
1348 /* Event list interface */
1350 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1352 /* Profile interface */
1353 entry
= tracefs_create_file("kprobe_profile", 0444, d_tracer
,
1354 NULL
, &kprobe_profile_ops
);
1357 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1360 fs_initcall(init_kprobe_trace
);
1363 #ifdef CONFIG_FTRACE_STARTUP_TEST
1365 * The "__used" keeps gcc from removing the function symbol
1366 * from the kallsyms table. 'noinline' makes sure that there
1367 * isn't an inlined version used by the test method below
1369 static __used __init noinline
int
1370 kprobe_trace_selftest_target(int a1
, int a2
, int a3
, int a4
, int a5
, int a6
)
1372 return a1
+ a2
+ a3
+ a4
+ a5
+ a6
;
1375 static struct __init trace_event_file
*
1376 find_trace_probe_file(struct trace_kprobe
*tk
, struct trace_array
*tr
)
1378 struct trace_event_file
*file
;
1380 list_for_each_entry(file
, &tr
->events
, list
)
1381 if (file
->event_call
== &tk
->tp
.call
)
1388 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1389 * stage, we can do this lockless.
1391 static __init
int kprobe_trace_self_tests_init(void)
1394 int (*target
)(int, int, int, int, int, int);
1395 struct trace_kprobe
*tk
;
1396 struct trace_event_file
*file
;
1398 if (tracing_is_disabled())
1401 target
= kprobe_trace_selftest_target
;
1403 pr_info("Testing kprobe tracing: ");
1405 ret
= traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1406 "$stack $stack0 +0($stack)",
1407 create_trace_kprobe
);
1408 if (WARN_ON_ONCE(ret
)) {
1409 pr_warn("error on probing function entry.\n");
1412 /* Enable trace point */
1413 tk
= find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM
);
1414 if (WARN_ON_ONCE(tk
== NULL
)) {
1415 pr_warn("error on getting new probe.\n");
1418 file
= find_trace_probe_file(tk
, top_trace_array());
1419 if (WARN_ON_ONCE(file
== NULL
)) {
1420 pr_warn("error on getting probe file.\n");
1423 enable_trace_kprobe(tk
, file
);
1427 ret
= traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1428 "$retval", create_trace_kprobe
);
1429 if (WARN_ON_ONCE(ret
)) {
1430 pr_warn("error on probing function return.\n");
1433 /* Enable trace point */
1434 tk
= find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM
);
1435 if (WARN_ON_ONCE(tk
== NULL
)) {
1436 pr_warn("error on getting 2nd new probe.\n");
1439 file
= find_trace_probe_file(tk
, top_trace_array());
1440 if (WARN_ON_ONCE(file
== NULL
)) {
1441 pr_warn("error on getting probe file.\n");
1444 enable_trace_kprobe(tk
, file
);
1451 ret
= target(1, 2, 3, 4, 5, 6);
1454 * Not expecting an error here, the check is only to prevent the
1455 * optimizer from removing the call to target() as otherwise there
1456 * are no side-effects and the call is never performed.
1461 /* Disable trace points before removing it */
1462 tk
= find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM
);
1463 if (WARN_ON_ONCE(tk
== NULL
)) {
1464 pr_warn("error on getting test probe.\n");
1467 if (trace_kprobe_nhit(tk
) != 1) {
1468 pr_warn("incorrect number of testprobe hits\n");
1472 file
= find_trace_probe_file(tk
, top_trace_array());
1473 if (WARN_ON_ONCE(file
== NULL
)) {
1474 pr_warn("error on getting probe file.\n");
1477 disable_trace_kprobe(tk
, file
);
1480 tk
= find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM
);
1481 if (WARN_ON_ONCE(tk
== NULL
)) {
1482 pr_warn("error on getting 2nd test probe.\n");
1485 if (trace_kprobe_nhit(tk
) != 1) {
1486 pr_warn("incorrect number of testprobe2 hits\n");
1490 file
= find_trace_probe_file(tk
, top_trace_array());
1491 if (WARN_ON_ONCE(file
== NULL
)) {
1492 pr_warn("error on getting probe file.\n");
1495 disable_trace_kprobe(tk
, file
);
1498 ret
= traceprobe_command("-:testprobe", create_trace_kprobe
);
1499 if (WARN_ON_ONCE(ret
)) {
1500 pr_warn("error on deleting a probe.\n");
1504 ret
= traceprobe_command("-:testprobe2", create_trace_kprobe
);
1505 if (WARN_ON_ONCE(ret
)) {
1506 pr_warn("error on deleting a probe.\n");
1511 release_all_trace_kprobes();
1513 pr_cont("NG: Some tests are failed. Please check them.\n");
1519 late_initcall(kprobe_trace_self_tests_init
);