1 #include <trace/syscall.h>
2 #include <trace/events/syscalls.h>
3 #include <linux/kernel.h>
4 #include <linux/ftrace.h>
5 #include <linux/perf_event.h>
6 #include <asm/syscall.h>
8 #include "trace_output.h"
11 static DEFINE_MUTEX(syscall_trace_lock);
12 static int sys_refcount_enter;
13 static int sys_refcount_exit;
14 static DECLARE_BITMAP(enabled_enter_syscalls, NR_syscalls);
15 static DECLARE_BITMAP(enabled_exit_syscalls, NR_syscalls);
17 extern unsigned long __start_syscalls_metadata[];
18 extern unsigned long __stop_syscalls_metadata[];
20 static struct syscall_metadata **syscalls_metadata;
22 static struct syscall_metadata *find_syscall_meta(unsigned long syscall)
24 struct syscall_metadata *start;
25 struct syscall_metadata *stop;
26 char str[KSYM_SYMBOL_LEN];
29 start = (struct syscall_metadata *)__start_syscalls_metadata;
30 stop = (struct syscall_metadata *)__stop_syscalls_metadata;
31 kallsyms_lookup(syscall, NULL, NULL, NULL, str);
33 for ( ; start < stop; start++) {
35 * Only compare after the "sys" prefix. Archs that use
36 * syscall wrappers may have syscalls symbols aliases prefixed
37 * with "SyS" instead of "sys", leading to an unwanted
40 if (start->name && !strcmp(start->name + 3, str + 3))
46 static struct syscall_metadata *syscall_nr_to_meta(int nr)
48 if (!syscalls_metadata || nr >= NR_syscalls || nr < 0)
51 return syscalls_metadata[nr];
55 print_syscall_enter(struct trace_iterator *iter, int flags)
57 struct trace_seq *s = &iter->seq;
58 struct trace_entry *ent = iter->ent;
59 struct syscall_trace_enter *trace;
60 struct syscall_metadata *entry;
63 trace = (typeof(trace))ent;
65 entry = syscall_nr_to_meta(syscall);
70 if (entry->enter_event->id != ent->type) {
75 ret = trace_seq_printf(s, "%s(", entry->name);
77 return TRACE_TYPE_PARTIAL_LINE;
79 for (i = 0; i < entry->nb_args; i++) {
81 if (trace_flags & TRACE_ITER_VERBOSE) {
82 ret = trace_seq_printf(s, "%s ", entry->types[i]);
84 return TRACE_TYPE_PARTIAL_LINE;
86 /* parameter values */
87 ret = trace_seq_printf(s, "%s: %lx%s", entry->args[i],
89 i == entry->nb_args - 1 ? "" : ", ");
91 return TRACE_TYPE_PARTIAL_LINE;
94 ret = trace_seq_putc(s, ')');
96 return TRACE_TYPE_PARTIAL_LINE;
99 ret = trace_seq_putc(s, '\n');
101 return TRACE_TYPE_PARTIAL_LINE;
103 return TRACE_TYPE_HANDLED;
107 print_syscall_exit(struct trace_iterator *iter, int flags)
109 struct trace_seq *s = &iter->seq;
110 struct trace_entry *ent = iter->ent;
111 struct syscall_trace_exit *trace;
113 struct syscall_metadata *entry;
116 trace = (typeof(trace))ent;
118 entry = syscall_nr_to_meta(syscall);
121 trace_seq_printf(s, "\n");
122 return TRACE_TYPE_HANDLED;
125 if (entry->exit_event->id != ent->type) {
127 return TRACE_TYPE_UNHANDLED;
130 ret = trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
133 return TRACE_TYPE_PARTIAL_LINE;
135 return TRACE_TYPE_HANDLED;
138 extern char *__bad_type_size(void);
140 #define SYSCALL_FIELD(type, name) \
141 sizeof(type) != sizeof(trace.name) ? \
142 __bad_type_size() : \
143 #type, #name, offsetof(typeof(trace), name), \
144 sizeof(trace.name), is_signed_type(type)
146 int syscall_enter_format(struct ftrace_event_call *call, struct trace_seq *s)
150 struct syscall_metadata *entry = call->data;
151 struct syscall_trace_enter trace;
152 int offset = offsetof(struct syscall_trace_enter, args);
154 ret = trace_seq_printf(s, "\tfield:%s %s;\toffset:%zu;\tsize:%zu;"
156 SYSCALL_FIELD(int, nr));
160 for (i = 0; i < entry->nb_args; i++) {
161 ret = trace_seq_printf(s, "\tfield:%s %s;", entry->types[i],
165 ret = trace_seq_printf(s, "\toffset:%d;\tsize:%zu;"
166 "\tsigned:%u;\n", offset,
167 sizeof(unsigned long),
168 is_signed_type(unsigned long));
171 offset += sizeof(unsigned long);
174 trace_seq_puts(s, "\nprint fmt: \"");
175 for (i = 0; i < entry->nb_args; i++) {
176 ret = trace_seq_printf(s, "%s: 0x%%0%zulx%s", entry->args[i],
177 sizeof(unsigned long),
178 i == entry->nb_args - 1 ? "" : ", ");
182 trace_seq_putc(s, '"');
184 for (i = 0; i < entry->nb_args; i++) {
185 ret = trace_seq_printf(s, ", ((unsigned long)(REC->%s))",
191 return trace_seq_putc(s, '\n');
195 int __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 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 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 int syscall_exit_format(struct ftrace_event_call *call, struct trace_seq *s)
258 struct syscall_trace_exit trace;
260 ret = trace_seq_printf(s,
261 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;"
263 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;"
265 SYSCALL_FIELD(int, nr),
266 SYSCALL_FIELD(long, ret));
270 return trace_seq_printf(s, "\nprint fmt: \"0x%%lx\", REC->ret\n");
273 int syscall_enter_define_fields(struct ftrace_event_call *call)
275 struct syscall_trace_enter trace;
276 struct syscall_metadata *meta = call->data;
279 int offset = offsetof(typeof(trace), args);
281 ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
285 for (i = 0; i < meta->nb_args; i++) {
286 ret = trace_define_field(call, meta->types[i],
287 meta->args[i], offset,
288 sizeof(unsigned long), 0,
290 offset += sizeof(unsigned long);
296 int syscall_exit_define_fields(struct ftrace_event_call *call)
298 struct syscall_trace_exit trace;
301 ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
305 ret = trace_define_field(call, SYSCALL_FIELD(long, ret),
311 void ftrace_syscall_enter(struct pt_regs *regs, long id)
313 struct syscall_trace_enter *entry;
314 struct syscall_metadata *sys_data;
315 struct ring_buffer_event *event;
316 struct ring_buffer *buffer;
320 syscall_nr = syscall_get_nr(current, regs);
323 if (!test_bit(syscall_nr, enabled_enter_syscalls))
326 sys_data = syscall_nr_to_meta(syscall_nr);
330 size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
332 event = trace_current_buffer_lock_reserve(&buffer,
333 sys_data->enter_event->id, size, 0, 0);
337 entry = ring_buffer_event_data(event);
338 entry->nr = syscall_nr;
339 syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
341 if (!filter_current_check_discard(buffer, sys_data->enter_event,
343 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
346 void ftrace_syscall_exit(struct pt_regs *regs, long ret)
348 struct syscall_trace_exit *entry;
349 struct syscall_metadata *sys_data;
350 struct ring_buffer_event *event;
351 struct ring_buffer *buffer;
354 syscall_nr = syscall_get_nr(current, regs);
357 if (!test_bit(syscall_nr, enabled_exit_syscalls))
360 sys_data = syscall_nr_to_meta(syscall_nr);
364 event = trace_current_buffer_lock_reserve(&buffer,
365 sys_data->exit_event->id, sizeof(*entry), 0, 0);
369 entry = ring_buffer_event_data(event);
370 entry->nr = syscall_nr;
371 entry->ret = syscall_get_return_value(current, regs);
373 if (!filter_current_check_discard(buffer, sys_data->exit_event,
375 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
378 int reg_event_syscall_enter(struct ftrace_event_call *call)
383 num = ((struct syscall_metadata *)call->data)->syscall_nr;
384 if (num < 0 || num >= NR_syscalls)
386 mutex_lock(&syscall_trace_lock);
387 if (!sys_refcount_enter)
388 ret = register_trace_sys_enter(ftrace_syscall_enter);
390 set_bit(num, enabled_enter_syscalls);
391 sys_refcount_enter++;
393 mutex_unlock(&syscall_trace_lock);
397 void unreg_event_syscall_enter(struct ftrace_event_call *call)
401 num = ((struct syscall_metadata *)call->data)->syscall_nr;
402 if (num < 0 || num >= NR_syscalls)
404 mutex_lock(&syscall_trace_lock);
405 sys_refcount_enter--;
406 clear_bit(num, enabled_enter_syscalls);
407 if (!sys_refcount_enter)
408 unregister_trace_sys_enter(ftrace_syscall_enter);
409 mutex_unlock(&syscall_trace_lock);
412 int reg_event_syscall_exit(struct ftrace_event_call *call)
417 num = ((struct syscall_metadata *)call->data)->syscall_nr;
418 if (num < 0 || num >= NR_syscalls)
420 mutex_lock(&syscall_trace_lock);
421 if (!sys_refcount_exit)
422 ret = register_trace_sys_exit(ftrace_syscall_exit);
424 set_bit(num, enabled_exit_syscalls);
427 mutex_unlock(&syscall_trace_lock);
431 void unreg_event_syscall_exit(struct ftrace_event_call *call)
435 num = ((struct syscall_metadata *)call->data)->syscall_nr;
436 if (num < 0 || num >= NR_syscalls)
438 mutex_lock(&syscall_trace_lock);
440 clear_bit(num, enabled_exit_syscalls);
441 if (!sys_refcount_exit)
442 unregister_trace_sys_exit(ftrace_syscall_exit);
443 mutex_unlock(&syscall_trace_lock);
446 int init_syscall_trace(struct ftrace_event_call *call)
450 if (set_syscall_print_fmt(call) < 0)
453 id = register_ftrace_event(call->event);
455 free_syscall_print_fmt(call);
459 INIT_LIST_HEAD(&call->fields);
463 int __init init_ftrace_syscalls(void)
465 struct syscall_metadata *meta;
469 syscalls_metadata = kzalloc(sizeof(*syscalls_metadata) *
470 NR_syscalls, GFP_KERNEL);
471 if (!syscalls_metadata) {
476 for (i = 0; i < NR_syscalls; i++) {
477 addr = arch_syscall_addr(i);
478 meta = find_syscall_meta(addr);
482 meta->syscall_nr = i;
483 syscalls_metadata[i] = meta;
488 core_initcall(init_ftrace_syscalls);
490 #ifdef CONFIG_EVENT_PROFILE
492 static DECLARE_BITMAP(enabled_prof_enter_syscalls, NR_syscalls);
493 static DECLARE_BITMAP(enabled_prof_exit_syscalls, NR_syscalls);
494 static int sys_prof_refcount_enter;
495 static int sys_prof_refcount_exit;
497 static void prof_syscall_enter(struct pt_regs *regs, long id)
499 struct syscall_metadata *sys_data;
500 struct syscall_trace_enter *rec;
509 syscall_nr = syscall_get_nr(current, regs);
510 if (!test_bit(syscall_nr, enabled_prof_enter_syscalls))
513 sys_data = syscall_nr_to_meta(syscall_nr);
517 /* get the size after alignment with the u32 buffer size field */
518 size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
519 size = ALIGN(size + sizeof(u32), sizeof(u64));
522 if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
523 "profile buffer not large enough"))
526 /* Protect the per cpu buffer, begin the rcu read side */
527 local_irq_save(flags);
529 rctx = perf_swevent_get_recursion_context();
533 cpu = smp_processor_id();
535 trace_buf = rcu_dereference(perf_trace_buf);
540 raw_data = per_cpu_ptr(trace_buf, cpu);
542 /* zero the dead bytes from align to not leak stack to user */
543 *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
545 rec = (struct syscall_trace_enter *) raw_data;
546 tracing_generic_entry_update(&rec->ent, 0, 0);
547 rec->ent.type = sys_data->enter_event->id;
548 rec->nr = syscall_nr;
549 syscall_get_arguments(current, regs, 0, sys_data->nb_args,
550 (unsigned long *)&rec->args);
551 perf_tp_event(sys_data->enter_event->id, 0, 1, rec, size);
554 perf_swevent_put_recursion_context(rctx);
556 local_irq_restore(flags);
559 int prof_sysenter_enable(struct ftrace_event_call *call)
564 num = ((struct syscall_metadata *)call->data)->syscall_nr;
566 mutex_lock(&syscall_trace_lock);
567 if (!sys_prof_refcount_enter)
568 ret = register_trace_sys_enter(prof_syscall_enter);
570 pr_info("event trace: Could not activate"
571 "syscall entry trace point");
573 set_bit(num, enabled_prof_enter_syscalls);
574 sys_prof_refcount_enter++;
576 mutex_unlock(&syscall_trace_lock);
580 void prof_sysenter_disable(struct ftrace_event_call *call)
584 num = ((struct syscall_metadata *)call->data)->syscall_nr;
586 mutex_lock(&syscall_trace_lock);
587 sys_prof_refcount_enter--;
588 clear_bit(num, enabled_prof_enter_syscalls);
589 if (!sys_prof_refcount_enter)
590 unregister_trace_sys_enter(prof_syscall_enter);
591 mutex_unlock(&syscall_trace_lock);
594 static void prof_syscall_exit(struct pt_regs *regs, long ret)
596 struct syscall_metadata *sys_data;
597 struct syscall_trace_exit *rec;
606 syscall_nr = syscall_get_nr(current, regs);
607 if (!test_bit(syscall_nr, enabled_prof_exit_syscalls))
610 sys_data = syscall_nr_to_meta(syscall_nr);
614 /* We can probably do that at build time */
615 size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
619 * Impossible, but be paranoid with the future
620 * How to put this check outside runtime?
622 if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
623 "exit event has grown above profile buffer size"))
626 /* Protect the per cpu buffer, begin the rcu read side */
627 local_irq_save(flags);
629 rctx = perf_swevent_get_recursion_context();
633 cpu = smp_processor_id();
635 trace_buf = rcu_dereference(perf_trace_buf);
640 raw_data = per_cpu_ptr(trace_buf, cpu);
642 /* zero the dead bytes from align to not leak stack to user */
643 *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
645 rec = (struct syscall_trace_exit *)raw_data;
647 tracing_generic_entry_update(&rec->ent, 0, 0);
648 rec->ent.type = sys_data->exit_event->id;
649 rec->nr = syscall_nr;
650 rec->ret = syscall_get_return_value(current, regs);
652 perf_tp_event(sys_data->exit_event->id, 0, 1, rec, size);
655 perf_swevent_put_recursion_context(rctx);
657 local_irq_restore(flags);
660 int prof_sysexit_enable(struct ftrace_event_call *call)
665 num = ((struct syscall_metadata *)call->data)->syscall_nr;
667 mutex_lock(&syscall_trace_lock);
668 if (!sys_prof_refcount_exit)
669 ret = register_trace_sys_exit(prof_syscall_exit);
671 pr_info("event trace: Could not activate"
672 "syscall entry trace point");
674 set_bit(num, enabled_prof_exit_syscalls);
675 sys_prof_refcount_exit++;
677 mutex_unlock(&syscall_trace_lock);
681 void prof_sysexit_disable(struct ftrace_event_call *call)
685 num = ((struct syscall_metadata *)call->data)->syscall_nr;
687 mutex_lock(&syscall_trace_lock);
688 sys_prof_refcount_exit--;
689 clear_bit(num, enabled_prof_exit_syscalls);
690 if (!sys_prof_refcount_exit)
691 unregister_trace_sys_exit(prof_syscall_exit);
692 mutex_unlock(&syscall_trace_lock);