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 trace_event_call *event,
17 enum trace_reg type, void *data);
18 static int syscall_exit_register(struct trace_event_call *event,
19 enum trace_reg type, void *data);
21 static struct list_head *
22 syscall_get_enter_fields(struct trace_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_array *tr = iter->tr;
114 struct trace_seq *s = &iter->seq;
115 struct trace_entry *ent = iter->ent;
116 struct syscall_trace_enter *trace;
117 struct syscall_metadata *entry;
120 trace = (typeof(trace))ent;
122 entry = syscall_nr_to_meta(syscall);
127 if (entry->enter_event->event.type != ent->type) {
132 trace_seq_printf(s, "%s(", entry->name);
134 for (i = 0; i < entry->nb_args; i++) {
136 if (trace_seq_has_overflowed(s))
139 /* parameter types */
140 if (tr->trace_flags & TRACE_ITER_VERBOSE)
141 trace_seq_printf(s, "%s ", entry->types[i]);
143 /* parameter values */
144 trace_seq_printf(s, "%s: %lx%s", entry->args[i],
146 i == entry->nb_args - 1 ? "" : ", ");
149 trace_seq_putc(s, ')');
151 trace_seq_putc(s, '\n');
153 return trace_handle_return(s);
156 static enum print_line_t
157 print_syscall_exit(struct trace_iterator *iter, int flags,
158 struct trace_event *event)
160 struct trace_seq *s = &iter->seq;
161 struct trace_entry *ent = iter->ent;
162 struct syscall_trace_exit *trace;
164 struct syscall_metadata *entry;
166 trace = (typeof(trace))ent;
168 entry = syscall_nr_to_meta(syscall);
171 trace_seq_putc(s, '\n');
175 if (entry->exit_event->event.type != ent->type) {
177 return TRACE_TYPE_UNHANDLED;
180 trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
184 return trace_handle_return(s);
187 extern char *__bad_type_size(void);
189 #define SYSCALL_FIELD(type, name) \
190 sizeof(type) != sizeof(trace.name) ? \
191 __bad_type_size() : \
192 #type, #name, offsetof(typeof(trace), name), \
193 sizeof(trace.name), is_signed_type(type)
196 __set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len)
201 /* When len=0, we just calculate the needed length */
202 #define LEN_OR_ZERO (len ? len - pos : 0)
204 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
205 for (i = 0; i < entry->nb_args; i++) {
206 pos += snprintf(buf + pos, LEN_OR_ZERO, "%s: 0x%%0%zulx%s",
207 entry->args[i], sizeof(unsigned long),
208 i == entry->nb_args - 1 ? "" : ", ");
210 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
212 for (i = 0; i < entry->nb_args; i++) {
213 pos += snprintf(buf + pos, LEN_OR_ZERO,
214 ", ((unsigned long)(REC->%s))", entry->args[i]);
219 /* return the length of print_fmt */
223 static int __init set_syscall_print_fmt(struct trace_event_call *call)
227 struct syscall_metadata *entry = call->data;
229 if (entry->enter_event != call) {
230 call->print_fmt = "\"0x%lx\", REC->ret";
234 /* First: called with 0 length to calculate the needed length */
235 len = __set_enter_print_fmt(entry, NULL, 0);
237 print_fmt = kmalloc(len + 1, GFP_KERNEL);
241 /* Second: actually write the @print_fmt */
242 __set_enter_print_fmt(entry, print_fmt, len + 1);
243 call->print_fmt = print_fmt;
248 static void __init free_syscall_print_fmt(struct trace_event_call *call)
250 struct syscall_metadata *entry = call->data;
252 if (entry->enter_event == call)
253 kfree(call->print_fmt);
256 static int __init syscall_enter_define_fields(struct trace_event_call *call)
258 struct syscall_trace_enter trace;
259 struct syscall_metadata *meta = call->data;
262 int offset = offsetof(typeof(trace), args);
264 ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
268 for (i = 0; i < meta->nb_args; i++) {
269 ret = trace_define_field(call, meta->types[i],
270 meta->args[i], offset,
271 sizeof(unsigned long), 0,
273 offset += sizeof(unsigned long);
279 static int __init syscall_exit_define_fields(struct trace_event_call *call)
281 struct syscall_trace_exit trace;
284 ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
288 ret = trace_define_field(call, SYSCALL_FIELD(long, ret),
294 static void ftrace_syscall_enter(void *data, struct pt_regs *regs, long id)
296 struct trace_array *tr = data;
297 struct trace_event_file *trace_file;
298 struct syscall_trace_enter *entry;
299 struct syscall_metadata *sys_data;
300 struct ring_buffer_event *event;
301 struct ring_buffer *buffer;
302 unsigned long irq_flags;
307 syscall_nr = trace_get_syscall_nr(current, regs);
308 if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
311 /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE) */
312 trace_file = rcu_dereference_sched(tr->enter_syscall_files[syscall_nr]);
316 if (trace_trigger_soft_disabled(trace_file))
319 sys_data = syscall_nr_to_meta(syscall_nr);
323 size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
325 local_save_flags(irq_flags);
326 pc = preempt_count();
328 buffer = tr->trace_buffer.buffer;
329 event = trace_buffer_lock_reserve(buffer,
330 sys_data->enter_event->event.type, size, irq_flags, pc);
334 entry = ring_buffer_event_data(event);
335 entry->nr = syscall_nr;
336 syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
338 event_trigger_unlock_commit(trace_file, buffer, event, entry,
342 static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
344 struct trace_array *tr = data;
345 struct trace_event_file *trace_file;
346 struct syscall_trace_exit *entry;
347 struct syscall_metadata *sys_data;
348 struct ring_buffer_event *event;
349 struct ring_buffer *buffer;
350 unsigned long irq_flags;
354 syscall_nr = trace_get_syscall_nr(current, regs);
355 if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
358 /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE()) */
359 trace_file = rcu_dereference_sched(tr->exit_syscall_files[syscall_nr]);
363 if (trace_trigger_soft_disabled(trace_file))
366 sys_data = syscall_nr_to_meta(syscall_nr);
370 local_save_flags(irq_flags);
371 pc = preempt_count();
373 buffer = tr->trace_buffer.buffer;
374 event = trace_buffer_lock_reserve(buffer,
375 sys_data->exit_event->event.type, sizeof(*entry),
380 entry = ring_buffer_event_data(event);
381 entry->nr = syscall_nr;
382 entry->ret = syscall_get_return_value(current, regs);
384 event_trigger_unlock_commit(trace_file, buffer, event, entry,
388 static int reg_event_syscall_enter(struct trace_event_file *file,
389 struct trace_event_call *call)
391 struct trace_array *tr = file->tr;
395 num = ((struct syscall_metadata *)call->data)->syscall_nr;
396 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
398 mutex_lock(&syscall_trace_lock);
399 if (!tr->sys_refcount_enter)
400 ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
402 rcu_assign_pointer(tr->enter_syscall_files[num], file);
403 tr->sys_refcount_enter++;
405 mutex_unlock(&syscall_trace_lock);
409 static void unreg_event_syscall_enter(struct trace_event_file *file,
410 struct trace_event_call *call)
412 struct trace_array *tr = file->tr;
415 num = ((struct syscall_metadata *)call->data)->syscall_nr;
416 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
418 mutex_lock(&syscall_trace_lock);
419 tr->sys_refcount_enter--;
420 RCU_INIT_POINTER(tr->enter_syscall_files[num], NULL);
421 if (!tr->sys_refcount_enter)
422 unregister_trace_sys_enter(ftrace_syscall_enter, tr);
423 mutex_unlock(&syscall_trace_lock);
426 static int reg_event_syscall_exit(struct trace_event_file *file,
427 struct trace_event_call *call)
429 struct trace_array *tr = file->tr;
433 num = ((struct syscall_metadata *)call->data)->syscall_nr;
434 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
436 mutex_lock(&syscall_trace_lock);
437 if (!tr->sys_refcount_exit)
438 ret = register_trace_sys_exit(ftrace_syscall_exit, tr);
440 rcu_assign_pointer(tr->exit_syscall_files[num], file);
441 tr->sys_refcount_exit++;
443 mutex_unlock(&syscall_trace_lock);
447 static void unreg_event_syscall_exit(struct trace_event_file *file,
448 struct trace_event_call *call)
450 struct trace_array *tr = file->tr;
453 num = ((struct syscall_metadata *)call->data)->syscall_nr;
454 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
456 mutex_lock(&syscall_trace_lock);
457 tr->sys_refcount_exit--;
458 RCU_INIT_POINTER(tr->exit_syscall_files[num], NULL);
459 if (!tr->sys_refcount_exit)
460 unregister_trace_sys_exit(ftrace_syscall_exit, tr);
461 mutex_unlock(&syscall_trace_lock);
464 static int __init init_syscall_trace(struct trace_event_call *call)
469 num = ((struct syscall_metadata *)call->data)->syscall_nr;
470 if (num < 0 || num >= NR_syscalls) {
471 pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
472 ((struct syscall_metadata *)call->data)->name);
476 if (set_syscall_print_fmt(call) < 0)
479 id = trace_event_raw_init(call);
482 free_syscall_print_fmt(call);
489 struct trace_event_functions enter_syscall_print_funcs = {
490 .trace = print_syscall_enter,
493 struct trace_event_functions exit_syscall_print_funcs = {
494 .trace = print_syscall_exit,
497 struct trace_event_class __refdata event_class_syscall_enter = {
498 .system = "syscalls",
499 .reg = syscall_enter_register,
500 .define_fields = syscall_enter_define_fields,
501 .get_fields = syscall_get_enter_fields,
502 .raw_init = init_syscall_trace,
505 struct trace_event_class __refdata event_class_syscall_exit = {
506 .system = "syscalls",
507 .reg = syscall_exit_register,
508 .define_fields = syscall_exit_define_fields,
509 .fields = LIST_HEAD_INIT(event_class_syscall_exit.fields),
510 .raw_init = init_syscall_trace,
513 unsigned long __init __weak arch_syscall_addr(int nr)
515 return (unsigned long)sys_call_table[nr];
518 void __init init_ftrace_syscalls(void)
520 struct syscall_metadata *meta;
524 syscalls_metadata = kcalloc(NR_syscalls, sizeof(*syscalls_metadata),
526 if (!syscalls_metadata) {
531 for (i = 0; i < NR_syscalls; i++) {
532 addr = arch_syscall_addr(i);
533 meta = find_syscall_meta(addr);
537 meta->syscall_nr = i;
538 syscalls_metadata[i] = meta;
542 #ifdef CONFIG_PERF_EVENTS
544 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
545 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
546 static int sys_perf_refcount_enter;
547 static int sys_perf_refcount_exit;
549 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
551 struct syscall_metadata *sys_data;
552 struct syscall_trace_enter *rec;
553 struct hlist_head *head;
558 syscall_nr = trace_get_syscall_nr(current, regs);
559 if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
561 if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
564 sys_data = syscall_nr_to_meta(syscall_nr);
568 head = this_cpu_ptr(sys_data->enter_event->perf_events);
569 if (hlist_empty(head))
572 /* get the size after alignment with the u32 buffer size field */
573 size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
574 size = ALIGN(size + sizeof(u32), sizeof(u64));
577 rec = (struct syscall_trace_enter *)perf_trace_buf_prepare(size,
578 sys_data->enter_event->event.type, NULL, &rctx);
582 rec->nr = syscall_nr;
583 syscall_get_arguments(current, regs, 0, sys_data->nb_args,
584 (unsigned long *)&rec->args);
585 perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head, NULL);
588 static int perf_sysenter_enable(struct trace_event_call *call)
593 num = ((struct syscall_metadata *)call->data)->syscall_nr;
595 mutex_lock(&syscall_trace_lock);
596 if (!sys_perf_refcount_enter)
597 ret = register_trace_sys_enter(perf_syscall_enter, NULL);
599 pr_info("event trace: Could not activate"
600 "syscall entry trace point");
602 set_bit(num, enabled_perf_enter_syscalls);
603 sys_perf_refcount_enter++;
605 mutex_unlock(&syscall_trace_lock);
609 static void perf_sysenter_disable(struct trace_event_call *call)
613 num = ((struct syscall_metadata *)call->data)->syscall_nr;
615 mutex_lock(&syscall_trace_lock);
616 sys_perf_refcount_enter--;
617 clear_bit(num, enabled_perf_enter_syscalls);
618 if (!sys_perf_refcount_enter)
619 unregister_trace_sys_enter(perf_syscall_enter, NULL);
620 mutex_unlock(&syscall_trace_lock);
623 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
625 struct syscall_metadata *sys_data;
626 struct syscall_trace_exit *rec;
627 struct hlist_head *head;
632 syscall_nr = trace_get_syscall_nr(current, regs);
633 if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
635 if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
638 sys_data = syscall_nr_to_meta(syscall_nr);
642 head = this_cpu_ptr(sys_data->exit_event->perf_events);
643 if (hlist_empty(head))
646 /* We can probably do that at build time */
647 size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
650 rec = (struct syscall_trace_exit *)perf_trace_buf_prepare(size,
651 sys_data->exit_event->event.type, NULL, &rctx);
655 rec->nr = syscall_nr;
656 rec->ret = syscall_get_return_value(current, regs);
657 perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head, NULL);
660 static int perf_sysexit_enable(struct trace_event_call *call)
665 num = ((struct syscall_metadata *)call->data)->syscall_nr;
667 mutex_lock(&syscall_trace_lock);
668 if (!sys_perf_refcount_exit)
669 ret = register_trace_sys_exit(perf_syscall_exit, NULL);
671 pr_info("event trace: Could not activate"
672 "syscall exit trace point");
674 set_bit(num, enabled_perf_exit_syscalls);
675 sys_perf_refcount_exit++;
677 mutex_unlock(&syscall_trace_lock);
681 static void perf_sysexit_disable(struct trace_event_call *call)
685 num = ((struct syscall_metadata *)call->data)->syscall_nr;
687 mutex_lock(&syscall_trace_lock);
688 sys_perf_refcount_exit--;
689 clear_bit(num, enabled_perf_exit_syscalls);
690 if (!sys_perf_refcount_exit)
691 unregister_trace_sys_exit(perf_syscall_exit, NULL);
692 mutex_unlock(&syscall_trace_lock);
695 #endif /* CONFIG_PERF_EVENTS */
697 static int syscall_enter_register(struct trace_event_call *event,
698 enum trace_reg type, void *data)
700 struct trace_event_file *file = data;
703 case TRACE_REG_REGISTER:
704 return reg_event_syscall_enter(file, event);
705 case TRACE_REG_UNREGISTER:
706 unreg_event_syscall_enter(file, event);
709 #ifdef CONFIG_PERF_EVENTS
710 case TRACE_REG_PERF_REGISTER:
711 return perf_sysenter_enable(event);
712 case TRACE_REG_PERF_UNREGISTER:
713 perf_sysenter_disable(event);
715 case TRACE_REG_PERF_OPEN:
716 case TRACE_REG_PERF_CLOSE:
717 case TRACE_REG_PERF_ADD:
718 case TRACE_REG_PERF_DEL:
725 static int syscall_exit_register(struct trace_event_call *event,
726 enum trace_reg type, void *data)
728 struct trace_event_file *file = data;
731 case TRACE_REG_REGISTER:
732 return reg_event_syscall_exit(file, event);
733 case TRACE_REG_UNREGISTER:
734 unreg_event_syscall_exit(file, event);
737 #ifdef CONFIG_PERF_EVENTS
738 case TRACE_REG_PERF_REGISTER:
739 return perf_sysexit_enable(event);
740 case TRACE_REG_PERF_UNREGISTER:
741 perf_sysexit_disable(event);
743 case TRACE_REG_PERF_OPEN:
744 case TRACE_REG_PERF_CLOSE:
745 case TRACE_REG_PERF_ADD:
746 case TRACE_REG_PERF_DEL: