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 */
34 unsigned int flags
; /* For TP_FLAG_* */
35 const char *symbol
; /* symbol name */
36 struct ftrace_event_class
class;
37 struct ftrace_event_call call
;
38 struct ftrace_event_file
* __rcu
*files
;
39 ssize_t size
; /* trace entry size */
41 struct probe_arg args
[];
44 #define SIZEOF_TRACE_PROBE(n) \
45 (offsetof(struct trace_probe, args) + \
46 (sizeof(struct probe_arg) * (n)))
49 static __kprobes
bool trace_probe_is_return(struct trace_probe
*tp
)
51 return tp
->rp
.handler
!= NULL
;
54 static __kprobes
const char *trace_probe_symbol(struct trace_probe
*tp
)
56 return tp
->symbol
? tp
->symbol
: "unknown";
59 static __kprobes
unsigned long trace_probe_offset(struct trace_probe
*tp
)
61 return tp
->rp
.kp
.offset
;
64 static __kprobes
bool trace_probe_is_enabled(struct trace_probe
*tp
)
66 return !!(tp
->flags
& (TP_FLAG_TRACE
| TP_FLAG_PROFILE
));
69 static __kprobes
bool trace_probe_is_registered(struct trace_probe
*tp
)
71 return !!(tp
->flags
& TP_FLAG_REGISTERED
);
74 static __kprobes
bool trace_probe_has_gone(struct trace_probe
*tp
)
76 return !!(kprobe_gone(&tp
->rp
.kp
));
79 static __kprobes
bool trace_probe_within_module(struct trace_probe
*tp
,
82 int len
= strlen(mod
->name
);
83 const char *name
= trace_probe_symbol(tp
);
84 return strncmp(mod
->name
, name
, len
) == 0 && name
[len
] == ':';
87 static __kprobes
bool trace_probe_is_on_module(struct trace_probe
*tp
)
89 return !!strchr(trace_probe_symbol(tp
), ':');
92 static int register_probe_event(struct trace_probe
*tp
);
93 static int unregister_probe_event(struct trace_probe
*tp
);
95 static DEFINE_MUTEX(probe_lock
);
96 static LIST_HEAD(probe_list
);
98 static int kprobe_dispatcher(struct kprobe
*kp
, struct pt_regs
*regs
);
99 static int kretprobe_dispatcher(struct kretprobe_instance
*ri
,
100 struct pt_regs
*regs
);
103 * Allocate new trace_probe and initialize it (including kprobes).
105 static struct trace_probe
*alloc_trace_probe(const char *group
,
110 int nargs
, bool is_return
)
112 struct trace_probe
*tp
;
115 tp
= kzalloc(SIZEOF_TRACE_PROBE(nargs
), GFP_KERNEL
);
120 tp
->symbol
= kstrdup(symbol
, GFP_KERNEL
);
123 tp
->rp
.kp
.symbol_name
= tp
->symbol
;
124 tp
->rp
.kp
.offset
= offs
;
126 tp
->rp
.kp
.addr
= addr
;
129 tp
->rp
.handler
= kretprobe_dispatcher
;
131 tp
->rp
.kp
.pre_handler
= kprobe_dispatcher
;
133 if (!event
|| !is_good_name(event
)) {
138 tp
->call
.class = &tp
->class;
139 tp
->call
.name
= kstrdup(event
, GFP_KERNEL
);
143 if (!group
|| !is_good_name(group
)) {
148 tp
->class.system
= kstrdup(group
, GFP_KERNEL
);
149 if (!tp
->class.system
)
152 INIT_LIST_HEAD(&tp
->list
);
155 kfree(tp
->call
.name
);
161 static void free_trace_probe(struct trace_probe
*tp
)
165 for (i
= 0; i
< tp
->nr_args
; i
++)
166 traceprobe_free_probe_arg(&tp
->args
[i
]);
168 kfree(tp
->call
.class->system
);
169 kfree(tp
->call
.name
);
174 static struct trace_probe
*find_trace_probe(const char *event
,
177 struct trace_probe
*tp
;
179 list_for_each_entry(tp
, &probe_list
, list
)
180 if (strcmp(tp
->call
.name
, event
) == 0 &&
181 strcmp(tp
->call
.class->system
, group
) == 0)
186 static int trace_probe_nr_files(struct trace_probe
*tp
)
188 struct ftrace_event_file
**file
;
192 * Since all tp->files updater is protected by probe_enable_lock,
193 * we don't need to lock an rcu_read_lock.
195 file
= rcu_dereference_raw(tp
->files
);
203 static DEFINE_MUTEX(probe_enable_lock
);
207 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
210 enable_trace_probe(struct trace_probe
*tp
, struct ftrace_event_file
*file
)
214 mutex_lock(&probe_enable_lock
);
217 struct ftrace_event_file
**new, **old
;
218 int n
= trace_probe_nr_files(tp
);
220 old
= rcu_dereference_raw(tp
->files
);
221 /* 1 is for new one and 1 is for stopper */
222 new = kzalloc((n
+ 2) * sizeof(struct ftrace_event_file
*),
228 memcpy(new, old
, n
* sizeof(struct ftrace_event_file
*));
230 /* The last one keeps a NULL */
232 rcu_assign_pointer(tp
->files
, new);
233 tp
->flags
|= TP_FLAG_TRACE
;
236 /* Make sure the probe is done with old files */
241 tp
->flags
|= TP_FLAG_PROFILE
;
243 if (trace_probe_is_enabled(tp
) && trace_probe_is_registered(tp
) &&
244 !trace_probe_has_gone(tp
)) {
245 if (trace_probe_is_return(tp
))
246 ret
= enable_kretprobe(&tp
->rp
);
248 ret
= enable_kprobe(&tp
->rp
.kp
);
252 mutex_unlock(&probe_enable_lock
);
258 trace_probe_file_index(struct trace_probe
*tp
, struct ftrace_event_file
*file
)
260 struct ftrace_event_file
**files
;
264 * Since all tp->files updater is protected by probe_enable_lock,
265 * we don't need to lock an rcu_read_lock.
267 files
= rcu_dereference_raw(tp
->files
);
269 for (i
= 0; files
[i
]; i
++)
270 if (files
[i
] == file
)
278 * Disable trace_probe
279 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
282 disable_trace_probe(struct trace_probe
*tp
, struct ftrace_event_file
*file
)
284 struct ftrace_event_file
**old
= NULL
;
288 mutex_lock(&probe_enable_lock
);
291 struct ftrace_event_file
**new, **old
;
292 int n
= trace_probe_nr_files(tp
);
295 old
= rcu_dereference_raw(tp
->files
);
296 if (n
== 0 || trace_probe_file_index(tp
, file
) < 0) {
301 if (n
== 1) { /* Remove the last file */
302 tp
->flags
&= ~TP_FLAG_TRACE
;
305 new = kzalloc(n
* sizeof(struct ftrace_event_file
*),
312 /* This copy & check loop copies the NULL stopper too */
313 for (i
= 0, j
= 0; j
< n
&& i
< n
+ 1; i
++)
318 rcu_assign_pointer(tp
->files
, new);
321 tp
->flags
&= ~TP_FLAG_PROFILE
;
323 if (!trace_probe_is_enabled(tp
) && trace_probe_is_registered(tp
)) {
324 if (trace_probe_is_return(tp
))
325 disable_kretprobe(&tp
->rp
);
327 disable_kprobe(&tp
->rp
.kp
);
332 mutex_unlock(&probe_enable_lock
);
336 * Synchronize with kprobe_trace_func/kretprobe_trace_func
337 * to ensure disabled (all running handlers are finished).
338 * This is not only for kfree(), but also the caller,
339 * trace_remove_event_call() supposes it for releasing
340 * event_call related objects, which will be accessed in
341 * the kprobe_trace_func/kretprobe_trace_func.
344 kfree(old
); /* Ignored if link == NULL */
350 /* Internal register function - just handle k*probes and flags */
351 static int __register_trace_probe(struct trace_probe
*tp
)
355 if (trace_probe_is_registered(tp
))
358 for (i
= 0; i
< tp
->nr_args
; i
++)
359 traceprobe_update_arg(&tp
->args
[i
]);
361 /* Set/clear disabled flag according to tp->flag */
362 if (trace_probe_is_enabled(tp
))
363 tp
->rp
.kp
.flags
&= ~KPROBE_FLAG_DISABLED
;
365 tp
->rp
.kp
.flags
|= KPROBE_FLAG_DISABLED
;
367 if (trace_probe_is_return(tp
))
368 ret
= register_kretprobe(&tp
->rp
);
370 ret
= register_kprobe(&tp
->rp
.kp
);
373 tp
->flags
|= TP_FLAG_REGISTERED
;
375 pr_warning("Could not insert probe at %s+%lu: %d\n",
376 trace_probe_symbol(tp
), trace_probe_offset(tp
), ret
);
377 if (ret
== -ENOENT
&& trace_probe_is_on_module(tp
)) {
378 pr_warning("This probe might be able to register after"
379 "target module is loaded. Continue.\n");
381 } else if (ret
== -EILSEQ
) {
382 pr_warning("Probing address(0x%p) is not an "
383 "instruction boundary.\n",
392 /* Internal unregister function - just handle k*probes and flags */
393 static void __unregister_trace_probe(struct trace_probe
*tp
)
395 if (trace_probe_is_registered(tp
)) {
396 if (trace_probe_is_return(tp
))
397 unregister_kretprobe(&tp
->rp
);
399 unregister_kprobe(&tp
->rp
.kp
);
400 tp
->flags
&= ~TP_FLAG_REGISTERED
;
401 /* Cleanup kprobe for reuse */
402 if (tp
->rp
.kp
.symbol_name
)
403 tp
->rp
.kp
.addr
= NULL
;
407 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
408 static int unregister_trace_probe(struct trace_probe
*tp
)
410 /* Enabled event can not be unregistered */
411 if (trace_probe_is_enabled(tp
))
414 /* Will fail if probe is being used by ftrace or perf */
415 if (unregister_probe_event(tp
))
418 __unregister_trace_probe(tp
);
424 /* Register a trace_probe and probe_event */
425 static int register_trace_probe(struct trace_probe
*tp
)
427 struct trace_probe
*old_tp
;
430 mutex_lock(&probe_lock
);
432 /* Delete old (same name) event if exist */
433 old_tp
= find_trace_probe(tp
->call
.name
, tp
->call
.class->system
);
435 ret
= unregister_trace_probe(old_tp
);
438 free_trace_probe(old_tp
);
441 /* Register new event */
442 ret
= register_probe_event(tp
);
444 pr_warning("Failed to register probe event(%d)\n", ret
);
448 /* Register k*probe */
449 ret
= __register_trace_probe(tp
);
451 unregister_probe_event(tp
);
453 list_add_tail(&tp
->list
, &probe_list
);
456 mutex_unlock(&probe_lock
);
460 /* Module notifier call back, checking event on the module */
461 static int trace_probe_module_callback(struct notifier_block
*nb
,
462 unsigned long val
, void *data
)
464 struct module
*mod
= data
;
465 struct trace_probe
*tp
;
468 if (val
!= MODULE_STATE_COMING
)
471 /* Update probes on coming module */
472 mutex_lock(&probe_lock
);
473 list_for_each_entry(tp
, &probe_list
, list
) {
474 if (trace_probe_within_module(tp
, mod
)) {
475 /* Don't need to check busy - this should have gone. */
476 __unregister_trace_probe(tp
);
477 ret
= __register_trace_probe(tp
);
479 pr_warning("Failed to re-register probe %s on"
481 tp
->call
.name
, mod
->name
, ret
);
484 mutex_unlock(&probe_lock
);
489 static struct notifier_block trace_probe_module_nb
= {
490 .notifier_call
= trace_probe_module_callback
,
491 .priority
= 1 /* Invoked after kprobe module callback */
494 static int create_trace_probe(int argc
, char **argv
)
498 * - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
499 * - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
501 * $retval : fetch return value
502 * $stack : fetch stack address
503 * $stackN : fetch Nth of stack (N:0-)
504 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
505 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
506 * %REG : fetch register REG
507 * Dereferencing memory fetch:
508 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
509 * Alias name of args:
510 * NAME=FETCHARG : set NAME as alias of FETCHARG.
512 * FETCHARG:TYPE : use TYPE instead of unsigned long.
514 struct trace_probe
*tp
;
516 bool is_return
= false, is_delete
= false;
517 char *symbol
= NULL
, *event
= NULL
, *group
= NULL
;
519 unsigned long offset
= 0;
521 char buf
[MAX_EVENT_NAME_LEN
];
523 /* argc must be >= 1 */
524 if (argv
[0][0] == 'p')
526 else if (argv
[0][0] == 'r')
528 else if (argv
[0][0] == '-')
531 pr_info("Probe definition must be started with 'p', 'r' or"
536 if (argv
[0][1] == ':') {
538 if (strchr(event
, '/')) {
540 event
= strchr(group
, '/') + 1;
542 if (strlen(group
) == 0) {
543 pr_info("Group name is not specified\n");
547 if (strlen(event
) == 0) {
548 pr_info("Event name is not specified\n");
553 group
= KPROBE_EVENT_SYSTEM
;
557 pr_info("Delete command needs an event name.\n");
560 mutex_lock(&probe_lock
);
561 tp
= find_trace_probe(event
, group
);
563 mutex_unlock(&probe_lock
);
564 pr_info("Event %s/%s doesn't exist.\n", group
, event
);
567 /* delete an event */
568 ret
= unregister_trace_probe(tp
);
570 free_trace_probe(tp
);
571 mutex_unlock(&probe_lock
);
576 pr_info("Probe point is not specified.\n");
579 if (isdigit(argv
[1][0])) {
581 pr_info("Return probe point must be a symbol.\n");
584 /* an address specified */
585 ret
= kstrtoul(&argv
[1][0], 0, (unsigned long *)&addr
);
587 pr_info("Failed to parse address.\n");
591 /* a symbol specified */
593 /* TODO: support .init module functions */
594 ret
= traceprobe_split_symbol_offset(symbol
, &offset
);
596 pr_info("Failed to parse symbol.\n");
599 if (offset
&& is_return
) {
600 pr_info("Return probe must be used without offset.\n");
604 argc
-= 2; argv
+= 2;
608 /* Make a new event name */
610 snprintf(buf
, MAX_EVENT_NAME_LEN
, "%c_%s_%ld",
611 is_return
? 'r' : 'p', symbol
, offset
);
613 snprintf(buf
, MAX_EVENT_NAME_LEN
, "%c_0x%p",
614 is_return
? 'r' : 'p', addr
);
617 tp
= alloc_trace_probe(group
, event
, addr
, symbol
, offset
, argc
,
620 pr_info("Failed to allocate trace_probe.(%d)\n",
625 /* parse arguments */
627 for (i
= 0; i
< argc
&& i
< MAX_TRACE_ARGS
; i
++) {
628 /* Increment count for freeing args in error case */
631 /* Parse argument name */
632 arg
= strchr(argv
[i
], '=');
635 tp
->args
[i
].name
= kstrdup(argv
[i
], GFP_KERNEL
);
638 /* If argument name is omitted, set "argN" */
639 snprintf(buf
, MAX_EVENT_NAME_LEN
, "arg%d", i
+ 1);
640 tp
->args
[i
].name
= kstrdup(buf
, GFP_KERNEL
);
643 if (!tp
->args
[i
].name
) {
644 pr_info("Failed to allocate argument[%d] name.\n", i
);
649 if (!is_good_name(tp
->args
[i
].name
)) {
650 pr_info("Invalid argument[%d] name: %s\n",
651 i
, tp
->args
[i
].name
);
656 if (traceprobe_conflict_field_name(tp
->args
[i
].name
,
658 pr_info("Argument[%d] name '%s' conflicts with "
659 "another field.\n", i
, argv
[i
]);
664 /* Parse fetch argument */
665 ret
= traceprobe_parse_probe_arg(arg
, &tp
->size
, &tp
->args
[i
],
668 pr_info("Parse error at argument[%d]. (%d)\n", i
, ret
);
673 ret
= register_trace_probe(tp
);
679 free_trace_probe(tp
);
683 static int release_all_trace_probes(void)
685 struct trace_probe
*tp
;
688 mutex_lock(&probe_lock
);
689 /* Ensure no probe is in use. */
690 list_for_each_entry(tp
, &probe_list
, list
)
691 if (trace_probe_is_enabled(tp
)) {
695 /* TODO: Use batch unregistration */
696 while (!list_empty(&probe_list
)) {
697 tp
= list_entry(probe_list
.next
, struct trace_probe
, list
);
698 ret
= unregister_trace_probe(tp
);
701 free_trace_probe(tp
);
705 mutex_unlock(&probe_lock
);
710 /* Probes listing interfaces */
711 static void *probes_seq_start(struct seq_file
*m
, loff_t
*pos
)
713 mutex_lock(&probe_lock
);
714 return seq_list_start(&probe_list
, *pos
);
717 static void *probes_seq_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
719 return seq_list_next(v
, &probe_list
, pos
);
722 static void probes_seq_stop(struct seq_file
*m
, void *v
)
724 mutex_unlock(&probe_lock
);
727 static int probes_seq_show(struct seq_file
*m
, void *v
)
729 struct trace_probe
*tp
= v
;
732 seq_printf(m
, "%c", trace_probe_is_return(tp
) ? 'r' : 'p');
733 seq_printf(m
, ":%s/%s", tp
->call
.class->system
, tp
->call
.name
);
736 seq_printf(m
, " 0x%p", tp
->rp
.kp
.addr
);
737 else if (tp
->rp
.kp
.offset
)
738 seq_printf(m
, " %s+%u", trace_probe_symbol(tp
),
741 seq_printf(m
, " %s", trace_probe_symbol(tp
));
743 for (i
= 0; i
< tp
->nr_args
; i
++)
744 seq_printf(m
, " %s=%s", tp
->args
[i
].name
, tp
->args
[i
].comm
);
750 static const struct seq_operations probes_seq_op
= {
751 .start
= probes_seq_start
,
752 .next
= probes_seq_next
,
753 .stop
= probes_seq_stop
,
754 .show
= probes_seq_show
757 static int probes_open(struct inode
*inode
, struct file
*file
)
761 if ((file
->f_mode
& FMODE_WRITE
) && (file
->f_flags
& O_TRUNC
)) {
762 ret
= release_all_trace_probes();
767 return seq_open(file
, &probes_seq_op
);
770 static ssize_t
probes_write(struct file
*file
, const char __user
*buffer
,
771 size_t count
, loff_t
*ppos
)
773 return traceprobe_probes_write(file
, buffer
, count
, ppos
,
777 static const struct file_operations kprobe_events_ops
= {
778 .owner
= THIS_MODULE
,
782 .release
= seq_release
,
783 .write
= probes_write
,
786 /* Probes profiling interfaces */
787 static int probes_profile_seq_show(struct seq_file
*m
, void *v
)
789 struct trace_probe
*tp
= v
;
791 seq_printf(m
, " %-44s %15lu %15lu\n", tp
->call
.name
, tp
->nhit
,
797 static const struct seq_operations profile_seq_op
= {
798 .start
= probes_seq_start
,
799 .next
= probes_seq_next
,
800 .stop
= probes_seq_stop
,
801 .show
= probes_profile_seq_show
804 static int profile_open(struct inode
*inode
, struct file
*file
)
806 return seq_open(file
, &profile_seq_op
);
809 static const struct file_operations kprobe_profile_ops
= {
810 .owner
= THIS_MODULE
,
811 .open
= profile_open
,
814 .release
= seq_release
,
817 /* Sum up total data length for dynamic arraies (strings) */
818 static __kprobes
int __get_data_size(struct trace_probe
*tp
,
819 struct pt_regs
*regs
)
824 for (i
= 0; i
< tp
->nr_args
; i
++)
825 if (unlikely(tp
->args
[i
].fetch_size
.fn
)) {
826 call_fetch(&tp
->args
[i
].fetch_size
, regs
, &len
);
833 /* Store the value of each argument */
834 static __kprobes
void store_trace_args(int ent_size
, struct trace_probe
*tp
,
835 struct pt_regs
*regs
,
836 u8
*data
, int maxlen
)
840 u32
*dl
; /* Data (relative) location */
842 for (i
= 0; i
< tp
->nr_args
; i
++) {
843 if (unlikely(tp
->args
[i
].fetch_size
.fn
)) {
845 * First, we set the relative location and
846 * maximum data length to *dl
848 dl
= (u32
*)(data
+ tp
->args
[i
].offset
);
849 *dl
= make_data_rloc(maxlen
, end
- tp
->args
[i
].offset
);
850 /* Then try to fetch string or dynamic array data */
851 call_fetch(&tp
->args
[i
].fetch
, regs
, dl
);
852 /* Reduce maximum length */
853 end
+= get_rloc_len(*dl
);
854 maxlen
-= get_rloc_len(*dl
);
855 /* Trick here, convert data_rloc to data_loc */
856 *dl
= convert_rloc_to_loc(*dl
,
857 ent_size
+ tp
->args
[i
].offset
);
859 /* Just fetching data normally */
860 call_fetch(&tp
->args
[i
].fetch
, regs
,
861 data
+ tp
->args
[i
].offset
);
866 static __kprobes
void
867 __kprobe_trace_func(struct trace_probe
*tp
, struct pt_regs
*regs
,
868 struct ftrace_event_file
*ftrace_file
)
870 struct kprobe_trace_entry_head
*entry
;
871 struct ring_buffer_event
*event
;
872 struct ring_buffer
*buffer
;
874 unsigned long irq_flags
;
875 struct ftrace_event_call
*call
= &tp
->call
;
877 WARN_ON(call
!= ftrace_file
->event_call
);
879 if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT
, &ftrace_file
->flags
))
882 local_save_flags(irq_flags
);
883 pc
= preempt_count();
885 dsize
= __get_data_size(tp
, regs
);
886 size
= sizeof(*entry
) + tp
->size
+ dsize
;
888 event
= trace_event_buffer_lock_reserve(&buffer
, ftrace_file
,
890 size
, irq_flags
, pc
);
894 entry
= ring_buffer_event_data(event
);
895 entry
->ip
= (unsigned long)tp
->rp
.kp
.addr
;
896 store_trace_args(sizeof(*entry
), tp
, regs
, (u8
*)&entry
[1], dsize
);
898 if (!filter_current_check_discard(buffer
, call
, entry
, event
))
899 trace_buffer_unlock_commit_regs(buffer
, event
,
900 irq_flags
, pc
, regs
);
903 static __kprobes
void
904 kprobe_trace_func(struct trace_probe
*tp
, struct pt_regs
*regs
)
907 * Note: preempt is already disabled around the kprobe handler.
908 * However, we still need an smp_read_barrier_depends() corresponding
909 * to smp_wmb() in rcu_assign_pointer() to access the pointer.
911 struct ftrace_event_file
**file
= rcu_dereference_raw(tp
->files
);
917 __kprobe_trace_func(tp
, regs
, *file
);
922 /* Kretprobe handler */
923 static __kprobes
void
924 __kretprobe_trace_func(struct trace_probe
*tp
, struct kretprobe_instance
*ri
,
925 struct pt_regs
*regs
,
926 struct ftrace_event_file
*ftrace_file
)
928 struct kretprobe_trace_entry_head
*entry
;
929 struct ring_buffer_event
*event
;
930 struct ring_buffer
*buffer
;
932 unsigned long irq_flags
;
933 struct ftrace_event_call
*call
= &tp
->call
;
935 WARN_ON(call
!= ftrace_file
->event_call
);
937 if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT
, &ftrace_file
->flags
))
940 local_save_flags(irq_flags
);
941 pc
= preempt_count();
943 dsize
= __get_data_size(tp
, regs
);
944 size
= sizeof(*entry
) + tp
->size
+ dsize
;
946 event
= trace_event_buffer_lock_reserve(&buffer
, ftrace_file
,
948 size
, irq_flags
, pc
);
952 entry
= ring_buffer_event_data(event
);
953 entry
->func
= (unsigned long)tp
->rp
.kp
.addr
;
954 entry
->ret_ip
= (unsigned long)ri
->ret_addr
;
955 store_trace_args(sizeof(*entry
), tp
, regs
, (u8
*)&entry
[1], dsize
);
957 if (!filter_current_check_discard(buffer
, call
, entry
, event
))
958 trace_buffer_unlock_commit_regs(buffer
, event
,
959 irq_flags
, pc
, regs
);
962 static __kprobes
void
963 kretprobe_trace_func(struct trace_probe
*tp
, struct kretprobe_instance
*ri
,
964 struct pt_regs
*regs
)
967 * Note: preempt is already disabled around the kprobe handler.
968 * However, we still need an smp_read_barrier_depends() corresponding
969 * to smp_wmb() in rcu_assign_pointer() to access the pointer.
971 struct ftrace_event_file
**file
= rcu_dereference_raw(tp
->files
);
977 __kretprobe_trace_func(tp
, ri
, regs
, *file
);
982 /* Event entry printers */
983 static enum print_line_t
984 print_kprobe_event(struct trace_iterator
*iter
, int flags
,
985 struct trace_event
*event
)
987 struct kprobe_trace_entry_head
*field
;
988 struct trace_seq
*s
= &iter
->seq
;
989 struct trace_probe
*tp
;
993 field
= (struct kprobe_trace_entry_head
*)iter
->ent
;
994 tp
= container_of(event
, struct trace_probe
, call
.event
);
996 if (!trace_seq_printf(s
, "%s: (", tp
->call
.name
))
999 if (!seq_print_ip_sym(s
, field
->ip
, flags
| TRACE_ITER_SYM_OFFSET
))
1002 if (!trace_seq_puts(s
, ")"))
1005 data
= (u8
*)&field
[1];
1006 for (i
= 0; i
< tp
->nr_args
; i
++)
1007 if (!tp
->args
[i
].type
->print(s
, tp
->args
[i
].name
,
1008 data
+ tp
->args
[i
].offset
, field
))
1011 if (!trace_seq_puts(s
, "\n"))
1014 return TRACE_TYPE_HANDLED
;
1016 return TRACE_TYPE_PARTIAL_LINE
;
1019 static enum print_line_t
1020 print_kretprobe_event(struct trace_iterator
*iter
, int flags
,
1021 struct trace_event
*event
)
1023 struct kretprobe_trace_entry_head
*field
;
1024 struct trace_seq
*s
= &iter
->seq
;
1025 struct trace_probe
*tp
;
1029 field
= (struct kretprobe_trace_entry_head
*)iter
->ent
;
1030 tp
= container_of(event
, struct trace_probe
, call
.event
);
1032 if (!trace_seq_printf(s
, "%s: (", tp
->call
.name
))
1035 if (!seq_print_ip_sym(s
, field
->ret_ip
, flags
| TRACE_ITER_SYM_OFFSET
))
1038 if (!trace_seq_puts(s
, " <- "))
1041 if (!seq_print_ip_sym(s
, field
->func
, flags
& ~TRACE_ITER_SYM_OFFSET
))
1044 if (!trace_seq_puts(s
, ")"))
1047 data
= (u8
*)&field
[1];
1048 for (i
= 0; i
< tp
->nr_args
; i
++)
1049 if (!tp
->args
[i
].type
->print(s
, tp
->args
[i
].name
,
1050 data
+ tp
->args
[i
].offset
, field
))
1053 if (!trace_seq_puts(s
, "\n"))
1056 return TRACE_TYPE_HANDLED
;
1058 return TRACE_TYPE_PARTIAL_LINE
;
1062 static int kprobe_event_define_fields(struct ftrace_event_call
*event_call
)
1065 struct kprobe_trace_entry_head field
;
1066 struct trace_probe
*tp
= (struct trace_probe
*)event_call
->data
;
1068 DEFINE_FIELD(unsigned long, ip
, FIELD_STRING_IP
, 0);
1069 /* Set argument names as fields */
1070 for (i
= 0; i
< tp
->nr_args
; i
++) {
1071 ret
= trace_define_field(event_call
, tp
->args
[i
].type
->fmttype
,
1073 sizeof(field
) + tp
->args
[i
].offset
,
1074 tp
->args
[i
].type
->size
,
1075 tp
->args
[i
].type
->is_signed
,
1083 static int kretprobe_event_define_fields(struct ftrace_event_call
*event_call
)
1086 struct kretprobe_trace_entry_head field
;
1087 struct trace_probe
*tp
= (struct trace_probe
*)event_call
->data
;
1089 DEFINE_FIELD(unsigned long, func
, FIELD_STRING_FUNC
, 0);
1090 DEFINE_FIELD(unsigned long, ret_ip
, FIELD_STRING_RETIP
, 0);
1091 /* Set argument names as fields */
1092 for (i
= 0; i
< tp
->nr_args
; i
++) {
1093 ret
= trace_define_field(event_call
, tp
->args
[i
].type
->fmttype
,
1095 sizeof(field
) + tp
->args
[i
].offset
,
1096 tp
->args
[i
].type
->size
,
1097 tp
->args
[i
].type
->is_signed
,
1105 static int __set_print_fmt(struct trace_probe
*tp
, char *buf
, int len
)
1110 const char *fmt
, *arg
;
1112 if (!trace_probe_is_return(tp
)) {
1114 arg
= "REC->" FIELD_STRING_IP
;
1116 fmt
= "(%lx <- %lx)";
1117 arg
= "REC->" FIELD_STRING_FUNC
", REC->" FIELD_STRING_RETIP
;
1120 /* When len=0, we just calculate the needed length */
1121 #define LEN_OR_ZERO (len ? len - pos : 0)
1123 pos
+= snprintf(buf
+ pos
, LEN_OR_ZERO
, "\"%s", fmt
);
1125 for (i
= 0; i
< tp
->nr_args
; i
++) {
1126 pos
+= snprintf(buf
+ pos
, LEN_OR_ZERO
, " %s=%s",
1127 tp
->args
[i
].name
, tp
->args
[i
].type
->fmt
);
1130 pos
+= snprintf(buf
+ pos
, LEN_OR_ZERO
, "\", %s", arg
);
1132 for (i
= 0; i
< tp
->nr_args
; i
++) {
1133 if (strcmp(tp
->args
[i
].type
->name
, "string") == 0)
1134 pos
+= snprintf(buf
+ pos
, LEN_OR_ZERO
,
1138 pos
+= snprintf(buf
+ pos
, LEN_OR_ZERO
, ", REC->%s",
1144 /* return the length of print_fmt */
1148 static int set_print_fmt(struct trace_probe
*tp
)
1153 /* First: called with 0 length to calculate the needed length */
1154 len
= __set_print_fmt(tp
, NULL
, 0);
1155 print_fmt
= kmalloc(len
+ 1, GFP_KERNEL
);
1159 /* Second: actually write the @print_fmt */
1160 __set_print_fmt(tp
, print_fmt
, len
+ 1);
1161 tp
->call
.print_fmt
= print_fmt
;
1166 #ifdef CONFIG_PERF_EVENTS
1168 /* Kprobe profile handler */
1169 static __kprobes
void
1170 kprobe_perf_func(struct trace_probe
*tp
, struct pt_regs
*regs
)
1172 struct ftrace_event_call
*call
= &tp
->call
;
1173 struct kprobe_trace_entry_head
*entry
;
1174 struct hlist_head
*head
;
1175 int size
, __size
, dsize
;
1178 dsize
= __get_data_size(tp
, regs
);
1179 __size
= sizeof(*entry
) + tp
->size
+ dsize
;
1180 size
= ALIGN(__size
+ sizeof(u32
), sizeof(u64
));
1181 size
-= sizeof(u32
);
1182 if (WARN_ONCE(size
> PERF_MAX_TRACE_SIZE
,
1183 "profile buffer not large enough"))
1186 entry
= perf_trace_buf_prepare(size
, call
->event
.type
, regs
, &rctx
);
1190 entry
->ip
= (unsigned long)tp
->rp
.kp
.addr
;
1191 memset(&entry
[1], 0, dsize
);
1192 store_trace_args(sizeof(*entry
), tp
, regs
, (u8
*)&entry
[1], dsize
);
1194 head
= this_cpu_ptr(call
->perf_events
);
1195 perf_trace_buf_submit(entry
, size
, rctx
,
1196 entry
->ip
, 1, regs
, head
, NULL
);
1199 /* Kretprobe profile handler */
1200 static __kprobes
void
1201 kretprobe_perf_func(struct trace_probe
*tp
, struct kretprobe_instance
*ri
,
1202 struct pt_regs
*regs
)
1204 struct ftrace_event_call
*call
= &tp
->call
;
1205 struct kretprobe_trace_entry_head
*entry
;
1206 struct hlist_head
*head
;
1207 int size
, __size
, dsize
;
1210 dsize
= __get_data_size(tp
, regs
);
1211 __size
= sizeof(*entry
) + tp
->size
+ dsize
;
1212 size
= ALIGN(__size
+ sizeof(u32
), sizeof(u64
));
1213 size
-= sizeof(u32
);
1214 if (WARN_ONCE(size
> PERF_MAX_TRACE_SIZE
,
1215 "profile buffer not large enough"))
1218 entry
= perf_trace_buf_prepare(size
, call
->event
.type
, regs
, &rctx
);
1222 entry
->func
= (unsigned long)tp
->rp
.kp
.addr
;
1223 entry
->ret_ip
= (unsigned long)ri
->ret_addr
;
1224 store_trace_args(sizeof(*entry
), tp
, regs
, (u8
*)&entry
[1], dsize
);
1226 head
= this_cpu_ptr(call
->perf_events
);
1227 perf_trace_buf_submit(entry
, size
, rctx
,
1228 entry
->ret_ip
, 1, regs
, head
, NULL
);
1230 #endif /* CONFIG_PERF_EVENTS */
1233 int kprobe_register(struct ftrace_event_call
*event
,
1234 enum trace_reg type
, void *data
)
1236 struct trace_probe
*tp
= (struct trace_probe
*)event
->data
;
1237 struct ftrace_event_file
*file
= data
;
1240 case TRACE_REG_REGISTER
:
1241 return enable_trace_probe(tp
, file
);
1242 case TRACE_REG_UNREGISTER
:
1243 return disable_trace_probe(tp
, file
);
1245 #ifdef CONFIG_PERF_EVENTS
1246 case TRACE_REG_PERF_REGISTER
:
1247 return enable_trace_probe(tp
, NULL
);
1248 case TRACE_REG_PERF_UNREGISTER
:
1249 return disable_trace_probe(tp
, NULL
);
1250 case TRACE_REG_PERF_OPEN
:
1251 case TRACE_REG_PERF_CLOSE
:
1252 case TRACE_REG_PERF_ADD
:
1253 case TRACE_REG_PERF_DEL
:
1261 int kprobe_dispatcher(struct kprobe
*kp
, struct pt_regs
*regs
)
1263 struct trace_probe
*tp
= container_of(kp
, struct trace_probe
, rp
.kp
);
1267 if (tp
->flags
& TP_FLAG_TRACE
)
1268 kprobe_trace_func(tp
, regs
);
1269 #ifdef CONFIG_PERF_EVENTS
1270 if (tp
->flags
& TP_FLAG_PROFILE
)
1271 kprobe_perf_func(tp
, regs
);
1273 return 0; /* We don't tweek kernel, so just return 0 */
1277 int kretprobe_dispatcher(struct kretprobe_instance
*ri
, struct pt_regs
*regs
)
1279 struct trace_probe
*tp
= container_of(ri
->rp
, struct trace_probe
, rp
);
1283 if (tp
->flags
& TP_FLAG_TRACE
)
1284 kretprobe_trace_func(tp
, ri
, regs
);
1285 #ifdef CONFIG_PERF_EVENTS
1286 if (tp
->flags
& TP_FLAG_PROFILE
)
1287 kretprobe_perf_func(tp
, ri
, regs
);
1289 return 0; /* We don't tweek kernel, so just return 0 */
1292 static struct trace_event_functions kretprobe_funcs
= {
1293 .trace
= print_kretprobe_event
1296 static struct trace_event_functions kprobe_funcs
= {
1297 .trace
= print_kprobe_event
1300 static int register_probe_event(struct trace_probe
*tp
)
1302 struct ftrace_event_call
*call
= &tp
->call
;
1305 /* Initialize ftrace_event_call */
1306 INIT_LIST_HEAD(&call
->class->fields
);
1307 if (trace_probe_is_return(tp
)) {
1308 call
->event
.funcs
= &kretprobe_funcs
;
1309 call
->class->define_fields
= kretprobe_event_define_fields
;
1311 call
->event
.funcs
= &kprobe_funcs
;
1312 call
->class->define_fields
= kprobe_event_define_fields
;
1314 if (set_print_fmt(tp
) < 0)
1316 ret
= register_ftrace_event(&call
->event
);
1318 kfree(call
->print_fmt
);
1322 call
->class->reg
= kprobe_register
;
1324 ret
= trace_add_event_call(call
);
1326 pr_info("Failed to register kprobe event: %s\n", call
->name
);
1327 kfree(call
->print_fmt
);
1328 unregister_ftrace_event(&call
->event
);
1333 static int unregister_probe_event(struct trace_probe
*tp
)
1337 /* tp->event is unregistered in trace_remove_event_call() */
1338 ret
= trace_remove_event_call(&tp
->call
);
1340 kfree(tp
->call
.print_fmt
);
1344 /* Make a debugfs interface for controlling probe points */
1345 static __init
int init_kprobe_trace(void)
1347 struct dentry
*d_tracer
;
1348 struct dentry
*entry
;
1350 if (register_module_notifier(&trace_probe_module_nb
))
1353 d_tracer
= tracing_init_dentry();
1357 entry
= debugfs_create_file("kprobe_events", 0644, d_tracer
,
1358 NULL
, &kprobe_events_ops
);
1360 /* Event list interface */
1362 pr_warning("Could not create debugfs "
1363 "'kprobe_events' entry\n");
1365 /* Profile interface */
1366 entry
= debugfs_create_file("kprobe_profile", 0444, d_tracer
,
1367 NULL
, &kprobe_profile_ops
);
1370 pr_warning("Could not create debugfs "
1371 "'kprobe_profile' entry\n");
1374 fs_initcall(init_kprobe_trace
);
1377 #ifdef CONFIG_FTRACE_STARTUP_TEST
1380 * The "__used" keeps gcc from removing the function symbol
1381 * from the kallsyms table.
1383 static __used
int kprobe_trace_selftest_target(int a1
, int a2
, int a3
,
1384 int a4
, int a5
, int a6
)
1386 return a1
+ a2
+ a3
+ a4
+ a5
+ a6
;
1389 static struct ftrace_event_file
*
1390 find_trace_probe_file(struct trace_probe
*tp
, struct trace_array
*tr
)
1392 struct ftrace_event_file
*file
;
1394 list_for_each_entry(file
, &tr
->events
, list
)
1395 if (file
->event_call
== &tp
->call
)
1401 static __init
int kprobe_trace_self_tests_init(void)
1404 int (*target
)(int, int, int, int, int, int);
1405 struct trace_probe
*tp
;
1406 struct ftrace_event_file
*file
;
1408 target
= kprobe_trace_selftest_target
;
1410 pr_info("Testing kprobe tracing: ");
1412 ret
= traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1413 "$stack $stack0 +0($stack)",
1414 create_trace_probe
);
1415 if (WARN_ON_ONCE(ret
)) {
1416 pr_warn("error on probing function entry.\n");
1419 /* Enable trace point */
1420 tp
= find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM
);
1421 if (WARN_ON_ONCE(tp
== NULL
)) {
1422 pr_warn("error on getting new probe.\n");
1425 file
= find_trace_probe_file(tp
, top_trace_array());
1426 if (WARN_ON_ONCE(file
== NULL
)) {
1427 pr_warn("error on getting probe file.\n");
1430 enable_trace_probe(tp
, file
);
1434 ret
= traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1435 "$retval", create_trace_probe
);
1436 if (WARN_ON_ONCE(ret
)) {
1437 pr_warn("error on probing function return.\n");
1440 /* Enable trace point */
1441 tp
= find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM
);
1442 if (WARN_ON_ONCE(tp
== NULL
)) {
1443 pr_warn("error on getting 2nd new probe.\n");
1446 file
= find_trace_probe_file(tp
, top_trace_array());
1447 if (WARN_ON_ONCE(file
== NULL
)) {
1448 pr_warn("error on getting probe file.\n");
1451 enable_trace_probe(tp
, file
);
1458 ret
= target(1, 2, 3, 4, 5, 6);
1460 /* Disable trace points before removing it */
1461 tp
= find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM
);
1462 if (WARN_ON_ONCE(tp
== NULL
)) {
1463 pr_warn("error on getting test probe.\n");
1466 file
= find_trace_probe_file(tp
, top_trace_array());
1467 if (WARN_ON_ONCE(file
== NULL
)) {
1468 pr_warn("error on getting probe file.\n");
1471 disable_trace_probe(tp
, file
);
1474 tp
= find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM
);
1475 if (WARN_ON_ONCE(tp
== NULL
)) {
1476 pr_warn("error on getting 2nd test probe.\n");
1479 file
= find_trace_probe_file(tp
, top_trace_array());
1480 if (WARN_ON_ONCE(file
== NULL
)) {
1481 pr_warn("error on getting probe file.\n");
1484 disable_trace_probe(tp
, file
);
1487 ret
= traceprobe_command("-:testprobe", create_trace_probe
);
1488 if (WARN_ON_ONCE(ret
)) {
1489 pr_warn("error on deleting a probe.\n");
1493 ret
= traceprobe_command("-:testprobe2", create_trace_probe
);
1494 if (WARN_ON_ONCE(ret
)) {
1495 pr_warn("error on deleting a probe.\n");
1500 release_all_trace_probes();
1502 pr_cont("NG: Some tests are failed. Please check them.\n");
1508 late_initcall(kprobe_trace_self_tests_init
);