1 #include <trace/syscall.h>
2 #include <trace/events/syscalls.h>
3 #include <linux/syscalls.h>
4 #include <linux/slab.h>
5 #include <linux/kernel.h>
6 #include <linux/module.h> /* for MODULE_NAME_LEN via KSYM_SYMBOL_LEN */
7 #include <linux/ftrace.h>
8 #include <linux/perf_event.h>
9 #include <asm/syscall.h>
11 #include "trace_output.h"
14 static DEFINE_MUTEX(syscall_trace_lock
);
16 static int syscall_enter_register(struct ftrace_event_call
*event
,
17 enum trace_reg type
, void *data
);
18 static int syscall_exit_register(struct ftrace_event_call
*event
,
19 enum trace_reg type
, void *data
);
21 static struct list_head
*
22 syscall_get_enter_fields(struct ftrace_event_call
*call
)
24 struct syscall_metadata
*entry
= call
->data
;
26 return &entry
->enter_fields
;
29 extern struct syscall_metadata
*__start_syscalls_metadata
[];
30 extern struct syscall_metadata
*__stop_syscalls_metadata
[];
32 static struct syscall_metadata
**syscalls_metadata
;
34 #ifndef ARCH_HAS_SYSCALL_MATCH_SYM_NAME
35 static inline bool arch_syscall_match_sym_name(const char *sym
, const char *name
)
38 * Only compare after the "sys" prefix. Archs that use
39 * syscall wrappers may have syscalls symbols aliases prefixed
40 * with ".SyS" or ".sys" instead of "sys", leading to an unwanted
43 return !strcmp(sym
+ 3, name
+ 3);
47 #ifdef ARCH_TRACE_IGNORE_COMPAT_SYSCALLS
49 * Some architectures that allow for 32bit applications
50 * to run on a 64bit kernel, do not map the syscalls for
51 * the 32bit tasks the same as they do for 64bit tasks.
55 * In such a case, instead of reporting the wrong syscalls,
58 * For an arch to ignore the compat syscalls it needs to
59 * define ARCH_TRACE_IGNORE_COMPAT_SYSCALLS as well as
60 * define the function arch_trace_is_compat_syscall() to let
61 * the tracing system know that it should ignore it.
64 trace_get_syscall_nr(struct task_struct
*task
, struct pt_regs
*regs
)
66 if (unlikely(arch_trace_is_compat_syscall(regs
)))
69 return syscall_get_nr(task
, regs
);
73 trace_get_syscall_nr(struct task_struct
*task
, struct pt_regs
*regs
)
75 return syscall_get_nr(task
, regs
);
77 #endif /* ARCH_TRACE_IGNORE_COMPAT_SYSCALLS */
79 static __init
struct syscall_metadata
*
80 find_syscall_meta(unsigned long syscall
)
82 struct syscall_metadata
**start
;
83 struct syscall_metadata
**stop
;
84 char str
[KSYM_SYMBOL_LEN
];
87 start
= __start_syscalls_metadata
;
88 stop
= __stop_syscalls_metadata
;
89 kallsyms_lookup(syscall
, NULL
, NULL
, NULL
, str
);
91 if (arch_syscall_match_sym_name(str
, "sys_ni_syscall"))
94 for ( ; start
< stop
; start
++) {
95 if ((*start
)->name
&& arch_syscall_match_sym_name(str
, (*start
)->name
))
101 static struct syscall_metadata
*syscall_nr_to_meta(int nr
)
103 if (!syscalls_metadata
|| nr
>= NR_syscalls
|| nr
< 0)
106 return syscalls_metadata
[nr
];
109 static enum print_line_t
110 print_syscall_enter(struct trace_iterator
*iter
, int flags
,
111 struct trace_event
*event
)
113 struct trace_seq
*s
= &iter
->seq
;
114 struct trace_entry
*ent
= iter
->ent
;
115 struct syscall_trace_enter
*trace
;
116 struct syscall_metadata
*entry
;
119 trace
= (typeof(trace
))ent
;
121 entry
= syscall_nr_to_meta(syscall
);
126 if (entry
->enter_event
->event
.type
!= ent
->type
) {
131 trace_seq_printf(s
, "%s(", entry
->name
);
133 for (i
= 0; i
< entry
->nb_args
; i
++) {
135 if (trace_seq_has_overflowed(s
))
138 /* parameter types */
139 if (trace_flags
& TRACE_ITER_VERBOSE
)
140 trace_seq_printf(s
, "%s ", entry
->types
[i
]);
142 /* parameter values */
143 trace_seq_printf(s
, "%s: %lx%s", entry
->args
[i
],
145 i
== entry
->nb_args
- 1 ? "" : ", ");
148 trace_seq_putc(s
, ')');
150 trace_seq_putc(s
, '\n');
152 return trace_handle_return(s
);
155 static enum print_line_t
156 print_syscall_exit(struct trace_iterator
*iter
, int flags
,
157 struct trace_event
*event
)
159 struct trace_seq
*s
= &iter
->seq
;
160 struct trace_entry
*ent
= iter
->ent
;
161 struct syscall_trace_exit
*trace
;
163 struct syscall_metadata
*entry
;
165 trace
= (typeof(trace
))ent
;
167 entry
= syscall_nr_to_meta(syscall
);
170 trace_seq_putc(s
, '\n');
174 if (entry
->exit_event
->event
.type
!= ent
->type
) {
176 return TRACE_TYPE_UNHANDLED
;
179 trace_seq_printf(s
, "%s -> 0x%lx\n", entry
->name
,
183 return trace_handle_return(s
);
186 extern char *__bad_type_size(void);
188 #define SYSCALL_FIELD(type, name) \
189 sizeof(type) != sizeof(trace.name) ? \
190 __bad_type_size() : \
191 #type, #name, offsetof(typeof(trace), name), \
192 sizeof(trace.name), is_signed_type(type)
195 __set_enter_print_fmt(struct syscall_metadata
*entry
, char *buf
, int len
)
200 /* When len=0, we just calculate the needed length */
201 #define LEN_OR_ZERO (len ? len - pos : 0)
203 pos
+= snprintf(buf
+ pos
, LEN_OR_ZERO
, "\"");
204 for (i
= 0; i
< entry
->nb_args
; i
++) {
205 pos
+= snprintf(buf
+ pos
, LEN_OR_ZERO
, "%s: 0x%%0%zulx%s",
206 entry
->args
[i
], sizeof(unsigned long),
207 i
== entry
->nb_args
- 1 ? "" : ", ");
209 pos
+= snprintf(buf
+ pos
, LEN_OR_ZERO
, "\"");
211 for (i
= 0; i
< entry
->nb_args
; i
++) {
212 pos
+= snprintf(buf
+ pos
, LEN_OR_ZERO
,
213 ", ((unsigned long)(REC->%s))", entry
->args
[i
]);
218 /* return the length of print_fmt */
222 static int __init
set_syscall_print_fmt(struct ftrace_event_call
*call
)
226 struct syscall_metadata
*entry
= call
->data
;
228 if (entry
->enter_event
!= call
) {
229 call
->print_fmt
= "\"0x%lx\", REC->ret";
233 /* First: called with 0 length to calculate the needed length */
234 len
= __set_enter_print_fmt(entry
, NULL
, 0);
236 print_fmt
= kmalloc(len
+ 1, GFP_KERNEL
);
240 /* Second: actually write the @print_fmt */
241 __set_enter_print_fmt(entry
, print_fmt
, len
+ 1);
242 call
->print_fmt
= print_fmt
;
247 static void __init
free_syscall_print_fmt(struct ftrace_event_call
*call
)
249 struct syscall_metadata
*entry
= call
->data
;
251 if (entry
->enter_event
== call
)
252 kfree(call
->print_fmt
);
255 static int __init
syscall_enter_define_fields(struct ftrace_event_call
*call
)
257 struct syscall_trace_enter trace
;
258 struct syscall_metadata
*meta
= call
->data
;
261 int offset
= offsetof(typeof(trace
), args
);
263 ret
= trace_define_field(call
, SYSCALL_FIELD(int, nr
), FILTER_OTHER
);
267 for (i
= 0; i
< meta
->nb_args
; i
++) {
268 ret
= trace_define_field(call
, meta
->types
[i
],
269 meta
->args
[i
], offset
,
270 sizeof(unsigned long), 0,
272 offset
+= sizeof(unsigned long);
278 static int __init
syscall_exit_define_fields(struct ftrace_event_call
*call
)
280 struct syscall_trace_exit trace
;
283 ret
= trace_define_field(call
, SYSCALL_FIELD(int, nr
), FILTER_OTHER
);
287 ret
= trace_define_field(call
, SYSCALL_FIELD(long, ret
),
293 static void ftrace_syscall_enter(void *data
, struct pt_regs
*regs
, long id
)
295 struct trace_array
*tr
= data
;
296 struct ftrace_event_file
*ftrace_file
;
297 struct syscall_trace_enter
*entry
;
298 struct syscall_metadata
*sys_data
;
299 struct ring_buffer_event
*event
;
300 struct ring_buffer
*buffer
;
301 unsigned long irq_flags
;
306 syscall_nr
= trace_get_syscall_nr(current
, regs
);
307 if (syscall_nr
< 0 || syscall_nr
>= NR_syscalls
)
310 /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE) */
311 ftrace_file
= rcu_dereference_sched(tr
->enter_syscall_files
[syscall_nr
]);
315 if (ftrace_trigger_soft_disabled(ftrace_file
))
318 sys_data
= syscall_nr_to_meta(syscall_nr
);
322 size
= sizeof(*entry
) + sizeof(unsigned long) * sys_data
->nb_args
;
324 local_save_flags(irq_flags
);
325 pc
= preempt_count();
327 buffer
= tr
->trace_buffer
.buffer
;
328 event
= trace_buffer_lock_reserve(buffer
,
329 sys_data
->enter_event
->event
.type
, size
, irq_flags
, pc
);
333 entry
= ring_buffer_event_data(event
);
334 entry
->nr
= syscall_nr
;
335 syscall_get_arguments(current
, regs
, 0, sys_data
->nb_args
, entry
->args
);
337 event_trigger_unlock_commit(ftrace_file
, buffer
, event
, entry
,
341 static void ftrace_syscall_exit(void *data
, struct pt_regs
*regs
, long ret
)
343 struct trace_array
*tr
= data
;
344 struct ftrace_event_file
*ftrace_file
;
345 struct syscall_trace_exit
*entry
;
346 struct syscall_metadata
*sys_data
;
347 struct ring_buffer_event
*event
;
348 struct ring_buffer
*buffer
;
349 unsigned long irq_flags
;
353 syscall_nr
= trace_get_syscall_nr(current
, regs
);
354 if (syscall_nr
< 0 || syscall_nr
>= NR_syscalls
)
357 /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE()) */
358 ftrace_file
= rcu_dereference_sched(tr
->exit_syscall_files
[syscall_nr
]);
362 if (ftrace_trigger_soft_disabled(ftrace_file
))
365 sys_data
= syscall_nr_to_meta(syscall_nr
);
369 local_save_flags(irq_flags
);
370 pc
= preempt_count();
372 buffer
= tr
->trace_buffer
.buffer
;
373 event
= trace_buffer_lock_reserve(buffer
,
374 sys_data
->exit_event
->event
.type
, sizeof(*entry
),
379 entry
= ring_buffer_event_data(event
);
380 entry
->nr
= syscall_nr
;
381 entry
->ret
= syscall_get_return_value(current
, regs
);
383 event_trigger_unlock_commit(ftrace_file
, buffer
, event
, entry
,
387 static int reg_event_syscall_enter(struct ftrace_event_file
*file
,
388 struct ftrace_event_call
*call
)
390 struct trace_array
*tr
= file
->tr
;
394 num
= ((struct syscall_metadata
*)call
->data
)->syscall_nr
;
395 if (WARN_ON_ONCE(num
< 0 || num
>= NR_syscalls
))
397 mutex_lock(&syscall_trace_lock
);
398 if (!tr
->sys_refcount_enter
)
399 ret
= register_trace_sys_enter(ftrace_syscall_enter
, tr
);
401 rcu_assign_pointer(tr
->enter_syscall_files
[num
], file
);
402 tr
->sys_refcount_enter
++;
404 mutex_unlock(&syscall_trace_lock
);
408 static void unreg_event_syscall_enter(struct ftrace_event_file
*file
,
409 struct ftrace_event_call
*call
)
411 struct trace_array
*tr
= file
->tr
;
414 num
= ((struct syscall_metadata
*)call
->data
)->syscall_nr
;
415 if (WARN_ON_ONCE(num
< 0 || num
>= NR_syscalls
))
417 mutex_lock(&syscall_trace_lock
);
418 tr
->sys_refcount_enter
--;
419 RCU_INIT_POINTER(tr
->enter_syscall_files
[num
], NULL
);
420 if (!tr
->sys_refcount_enter
)
421 unregister_trace_sys_enter(ftrace_syscall_enter
, tr
);
422 mutex_unlock(&syscall_trace_lock
);
425 static int reg_event_syscall_exit(struct ftrace_event_file
*file
,
426 struct ftrace_event_call
*call
)
428 struct trace_array
*tr
= file
->tr
;
432 num
= ((struct syscall_metadata
*)call
->data
)->syscall_nr
;
433 if (WARN_ON_ONCE(num
< 0 || num
>= NR_syscalls
))
435 mutex_lock(&syscall_trace_lock
);
436 if (!tr
->sys_refcount_exit
)
437 ret
= register_trace_sys_exit(ftrace_syscall_exit
, tr
);
439 rcu_assign_pointer(tr
->exit_syscall_files
[num
], file
);
440 tr
->sys_refcount_exit
++;
442 mutex_unlock(&syscall_trace_lock
);
446 static void unreg_event_syscall_exit(struct ftrace_event_file
*file
,
447 struct ftrace_event_call
*call
)
449 struct trace_array
*tr
= file
->tr
;
452 num
= ((struct syscall_metadata
*)call
->data
)->syscall_nr
;
453 if (WARN_ON_ONCE(num
< 0 || num
>= NR_syscalls
))
455 mutex_lock(&syscall_trace_lock
);
456 tr
->sys_refcount_exit
--;
457 RCU_INIT_POINTER(tr
->exit_syscall_files
[num
], NULL
);
458 if (!tr
->sys_refcount_exit
)
459 unregister_trace_sys_exit(ftrace_syscall_exit
, tr
);
460 mutex_unlock(&syscall_trace_lock
);
463 static int __init
init_syscall_trace(struct ftrace_event_call
*call
)
468 num
= ((struct syscall_metadata
*)call
->data
)->syscall_nr
;
469 if (num
< 0 || num
>= NR_syscalls
) {
470 pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
471 ((struct syscall_metadata
*)call
->data
)->name
);
475 if (set_syscall_print_fmt(call
) < 0)
478 id
= trace_event_raw_init(call
);
481 free_syscall_print_fmt(call
);
488 struct trace_event_functions enter_syscall_print_funcs
= {
489 .trace
= print_syscall_enter
,
492 struct trace_event_functions exit_syscall_print_funcs
= {
493 .trace
= print_syscall_exit
,
496 struct ftrace_event_class __refdata event_class_syscall_enter
= {
497 .system
= "syscalls",
498 .reg
= syscall_enter_register
,
499 .define_fields
= syscall_enter_define_fields
,
500 .get_fields
= syscall_get_enter_fields
,
501 .raw_init
= init_syscall_trace
,
504 struct ftrace_event_class __refdata event_class_syscall_exit
= {
505 .system
= "syscalls",
506 .reg
= syscall_exit_register
,
507 .define_fields
= syscall_exit_define_fields
,
508 .fields
= LIST_HEAD_INIT(event_class_syscall_exit
.fields
),
509 .raw_init
= init_syscall_trace
,
512 unsigned long __init __weak
arch_syscall_addr(int nr
)
514 return (unsigned long)sys_call_table
[nr
];
517 void __init
init_ftrace_syscalls(void)
519 struct syscall_metadata
*meta
;
523 syscalls_metadata
= kcalloc(NR_syscalls
, sizeof(*syscalls_metadata
),
525 if (!syscalls_metadata
) {
530 for (i
= 0; i
< NR_syscalls
; i
++) {
531 addr
= arch_syscall_addr(i
);
532 meta
= find_syscall_meta(addr
);
536 meta
->syscall_nr
= i
;
537 syscalls_metadata
[i
] = meta
;
541 #ifdef CONFIG_PERF_EVENTS
543 static DECLARE_BITMAP(enabled_perf_enter_syscalls
, NR_syscalls
);
544 static DECLARE_BITMAP(enabled_perf_exit_syscalls
, NR_syscalls
);
545 static int sys_perf_refcount_enter
;
546 static int sys_perf_refcount_exit
;
548 static void perf_syscall_enter(void *ignore
, struct pt_regs
*regs
, long id
)
550 struct syscall_metadata
*sys_data
;
551 struct syscall_trace_enter
*rec
;
552 struct hlist_head
*head
;
557 syscall_nr
= trace_get_syscall_nr(current
, regs
);
558 if (syscall_nr
< 0 || syscall_nr
>= NR_syscalls
)
560 if (!test_bit(syscall_nr
, enabled_perf_enter_syscalls
))
563 sys_data
= syscall_nr_to_meta(syscall_nr
);
567 head
= this_cpu_ptr(sys_data
->enter_event
->perf_events
);
568 if (hlist_empty(head
))
571 /* get the size after alignment with the u32 buffer size field */
572 size
= sizeof(unsigned long) * sys_data
->nb_args
+ sizeof(*rec
);
573 size
= ALIGN(size
+ sizeof(u32
), sizeof(u64
));
576 rec
= (struct syscall_trace_enter
*)perf_trace_buf_prepare(size
,
577 sys_data
->enter_event
->event
.type
, NULL
, &rctx
);
581 rec
->nr
= syscall_nr
;
582 syscall_get_arguments(current
, regs
, 0, sys_data
->nb_args
,
583 (unsigned long *)&rec
->args
);
584 perf_trace_buf_submit(rec
, size
, rctx
, 0, 1, regs
, head
, NULL
);
587 static int perf_sysenter_enable(struct ftrace_event_call
*call
)
592 num
= ((struct syscall_metadata
*)call
->data
)->syscall_nr
;
594 mutex_lock(&syscall_trace_lock
);
595 if (!sys_perf_refcount_enter
)
596 ret
= register_trace_sys_enter(perf_syscall_enter
, NULL
);
598 pr_info("event trace: Could not activate"
599 "syscall entry trace point");
601 set_bit(num
, enabled_perf_enter_syscalls
);
602 sys_perf_refcount_enter
++;
604 mutex_unlock(&syscall_trace_lock
);
608 static void perf_sysenter_disable(struct ftrace_event_call
*call
)
612 num
= ((struct syscall_metadata
*)call
->data
)->syscall_nr
;
614 mutex_lock(&syscall_trace_lock
);
615 sys_perf_refcount_enter
--;
616 clear_bit(num
, enabled_perf_enter_syscalls
);
617 if (!sys_perf_refcount_enter
)
618 unregister_trace_sys_enter(perf_syscall_enter
, NULL
);
619 mutex_unlock(&syscall_trace_lock
);
622 static void perf_syscall_exit(void *ignore
, struct pt_regs
*regs
, long ret
)
624 struct syscall_metadata
*sys_data
;
625 struct syscall_trace_exit
*rec
;
626 struct hlist_head
*head
;
631 syscall_nr
= trace_get_syscall_nr(current
, regs
);
632 if (syscall_nr
< 0 || syscall_nr
>= NR_syscalls
)
634 if (!test_bit(syscall_nr
, enabled_perf_exit_syscalls
))
637 sys_data
= syscall_nr_to_meta(syscall_nr
);
641 head
= this_cpu_ptr(sys_data
->exit_event
->perf_events
);
642 if (hlist_empty(head
))
645 /* We can probably do that at build time */
646 size
= ALIGN(sizeof(*rec
) + sizeof(u32
), sizeof(u64
));
649 rec
= (struct syscall_trace_exit
*)perf_trace_buf_prepare(size
,
650 sys_data
->exit_event
->event
.type
, NULL
, &rctx
);
654 rec
->nr
= syscall_nr
;
655 rec
->ret
= syscall_get_return_value(current
, regs
);
656 perf_trace_buf_submit(rec
, size
, rctx
, 0, 1, regs
, head
, NULL
);
659 static int perf_sysexit_enable(struct ftrace_event_call
*call
)
664 num
= ((struct syscall_metadata
*)call
->data
)->syscall_nr
;
666 mutex_lock(&syscall_trace_lock
);
667 if (!sys_perf_refcount_exit
)
668 ret
= register_trace_sys_exit(perf_syscall_exit
, NULL
);
670 pr_info("event trace: Could not activate"
671 "syscall exit trace point");
673 set_bit(num
, enabled_perf_exit_syscalls
);
674 sys_perf_refcount_exit
++;
676 mutex_unlock(&syscall_trace_lock
);
680 static void perf_sysexit_disable(struct ftrace_event_call
*call
)
684 num
= ((struct syscall_metadata
*)call
->data
)->syscall_nr
;
686 mutex_lock(&syscall_trace_lock
);
687 sys_perf_refcount_exit
--;
688 clear_bit(num
, enabled_perf_exit_syscalls
);
689 if (!sys_perf_refcount_exit
)
690 unregister_trace_sys_exit(perf_syscall_exit
, NULL
);
691 mutex_unlock(&syscall_trace_lock
);
694 #endif /* CONFIG_PERF_EVENTS */
696 static int syscall_enter_register(struct ftrace_event_call
*event
,
697 enum trace_reg type
, void *data
)
699 struct ftrace_event_file
*file
= data
;
702 case TRACE_REG_REGISTER
:
703 return reg_event_syscall_enter(file
, event
);
704 case TRACE_REG_UNREGISTER
:
705 unreg_event_syscall_enter(file
, event
);
708 #ifdef CONFIG_PERF_EVENTS
709 case TRACE_REG_PERF_REGISTER
:
710 return perf_sysenter_enable(event
);
711 case TRACE_REG_PERF_UNREGISTER
:
712 perf_sysenter_disable(event
);
714 case TRACE_REG_PERF_OPEN
:
715 case TRACE_REG_PERF_CLOSE
:
716 case TRACE_REG_PERF_ADD
:
717 case TRACE_REG_PERF_DEL
:
724 static int syscall_exit_register(struct ftrace_event_call
*event
,
725 enum trace_reg type
, void *data
)
727 struct ftrace_event_file
*file
= data
;
730 case TRACE_REG_REGISTER
:
731 return reg_event_syscall_exit(file
, event
);
732 case TRACE_REG_UNREGISTER
:
733 unreg_event_syscall_exit(file
, event
);
736 #ifdef CONFIG_PERF_EVENTS
737 case TRACE_REG_PERF_REGISTER
:
738 return perf_sysexit_enable(event
);
739 case TRACE_REG_PERF_UNREGISTER
:
740 perf_sysexit_disable(event
);
742 case TRACE_REG_PERF_OPEN
:
743 case TRACE_REG_PERF_CLOSE
:
744 case TRACE_REG_PERF_ADD
:
745 case TRACE_REG_PERF_DEL
: