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];
54 int syscall_name_to_nr(char *name)
58 if (!syscalls_metadata)
61 for (i = 0; i < NR_syscalls; i++) {
62 if (syscalls_metadata[i]) {
63 if (!strcmp(syscalls_metadata[i]->name, name))
70 void set_syscall_enter_id(int num, int id)
72 syscalls_metadata[num]->enter_id = id;
75 void set_syscall_exit_id(int num, int id)
77 syscalls_metadata[num]->exit_id = id;
81 print_syscall_enter(struct trace_iterator *iter, int flags)
83 struct trace_seq *s = &iter->seq;
84 struct trace_entry *ent = iter->ent;
85 struct syscall_trace_enter *trace;
86 struct syscall_metadata *entry;
89 trace = (typeof(trace))ent;
91 entry = syscall_nr_to_meta(syscall);
96 if (entry->enter_id != ent->type) {
101 ret = trace_seq_printf(s, "%s(", entry->name);
103 return TRACE_TYPE_PARTIAL_LINE;
105 for (i = 0; i < entry->nb_args; i++) {
106 /* parameter types */
107 if (trace_flags & TRACE_ITER_VERBOSE) {
108 ret = trace_seq_printf(s, "%s ", entry->types[i]);
110 return TRACE_TYPE_PARTIAL_LINE;
112 /* parameter values */
113 ret = trace_seq_printf(s, "%s: %lx%s", entry->args[i],
115 i == entry->nb_args - 1 ? "" : ", ");
117 return TRACE_TYPE_PARTIAL_LINE;
120 ret = trace_seq_putc(s, ')');
122 return TRACE_TYPE_PARTIAL_LINE;
125 ret = trace_seq_putc(s, '\n');
127 return TRACE_TYPE_PARTIAL_LINE;
129 return TRACE_TYPE_HANDLED;
133 print_syscall_exit(struct trace_iterator *iter, int flags)
135 struct trace_seq *s = &iter->seq;
136 struct trace_entry *ent = iter->ent;
137 struct syscall_trace_exit *trace;
139 struct syscall_metadata *entry;
142 trace = (typeof(trace))ent;
144 entry = syscall_nr_to_meta(syscall);
147 trace_seq_printf(s, "\n");
148 return TRACE_TYPE_HANDLED;
151 if (entry->exit_id != ent->type) {
153 return TRACE_TYPE_UNHANDLED;
156 ret = trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
159 return TRACE_TYPE_PARTIAL_LINE;
161 return TRACE_TYPE_HANDLED;
164 extern char *__bad_type_size(void);
166 #define SYSCALL_FIELD(type, name) \
167 sizeof(type) != sizeof(trace.name) ? \
168 __bad_type_size() : \
169 #type, #name, offsetof(typeof(trace), name), sizeof(trace.name)
171 int syscall_enter_format(struct ftrace_event_call *call, struct trace_seq *s)
176 struct syscall_metadata *entry;
177 struct syscall_trace_enter trace;
178 int offset = offsetof(struct syscall_trace_enter, args);
180 nr = syscall_name_to_nr(call->data);
181 entry = syscall_nr_to_meta(nr);
186 ret = trace_seq_printf(s, "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n",
187 SYSCALL_FIELD(int, nr));
191 for (i = 0; i < entry->nb_args; i++) {
192 ret = trace_seq_printf(s, "\tfield:%s %s;", entry->types[i],
196 ret = trace_seq_printf(s, "\toffset:%d;\tsize:%zu;\n", offset,
197 sizeof(unsigned long));
200 offset += sizeof(unsigned long);
203 trace_seq_puts(s, "\nprint fmt: \"");
204 for (i = 0; i < entry->nb_args; i++) {
205 ret = trace_seq_printf(s, "%s: 0x%%0%zulx%s", entry->args[i],
206 sizeof(unsigned long),
207 i == entry->nb_args - 1 ? "" : ", ");
211 trace_seq_putc(s, '"');
213 for (i = 0; i < entry->nb_args; i++) {
214 ret = trace_seq_printf(s, ", ((unsigned long)(REC->%s))",
220 return trace_seq_putc(s, '\n');
223 int syscall_exit_format(struct ftrace_event_call *call, struct trace_seq *s)
226 struct syscall_trace_exit trace;
228 ret = trace_seq_printf(s,
229 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
230 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n",
231 SYSCALL_FIELD(int, nr),
232 SYSCALL_FIELD(long, ret));
236 return trace_seq_printf(s, "\nprint fmt: \"0x%%lx\", REC->ret\n");
239 int syscall_enter_define_fields(struct ftrace_event_call *call)
241 struct syscall_trace_enter trace;
242 struct syscall_metadata *meta;
246 int offset = offsetof(typeof(trace), args);
248 nr = syscall_name_to_nr(call->data);
249 meta = syscall_nr_to_meta(nr);
254 ret = trace_define_common_fields(call);
258 for (i = 0; i < meta->nb_args; i++) {
259 ret = trace_define_field(call, meta->types[i],
260 meta->args[i], offset,
261 sizeof(unsigned long), 0,
263 offset += sizeof(unsigned long);
269 int syscall_exit_define_fields(struct ftrace_event_call *call)
271 struct syscall_trace_exit trace;
274 ret = trace_define_common_fields(call);
278 ret = trace_define_field(call, SYSCALL_FIELD(long, ret), 0,
284 void ftrace_syscall_enter(struct pt_regs *regs, long id)
286 struct syscall_trace_enter *entry;
287 struct syscall_metadata *sys_data;
288 struct ring_buffer_event *event;
289 struct ring_buffer *buffer;
293 syscall_nr = syscall_get_nr(current, regs);
296 if (!test_bit(syscall_nr, enabled_enter_syscalls))
299 sys_data = syscall_nr_to_meta(syscall_nr);
303 size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
305 event = trace_current_buffer_lock_reserve(&buffer, sys_data->enter_id,
310 entry = ring_buffer_event_data(event);
311 entry->nr = syscall_nr;
312 syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
314 if (!filter_current_check_discard(buffer, sys_data->enter_event,
316 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
319 void ftrace_syscall_exit(struct pt_regs *regs, long ret)
321 struct syscall_trace_exit *entry;
322 struct syscall_metadata *sys_data;
323 struct ring_buffer_event *event;
324 struct ring_buffer *buffer;
327 syscall_nr = syscall_get_nr(current, regs);
330 if (!test_bit(syscall_nr, enabled_exit_syscalls))
333 sys_data = syscall_nr_to_meta(syscall_nr);
337 event = trace_current_buffer_lock_reserve(&buffer, sys_data->exit_id,
338 sizeof(*entry), 0, 0);
342 entry = ring_buffer_event_data(event);
343 entry->nr = syscall_nr;
344 entry->ret = syscall_get_return_value(current, regs);
346 if (!filter_current_check_discard(buffer, sys_data->exit_event,
348 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
351 int reg_event_syscall_enter(void *ptr)
358 num = syscall_name_to_nr(name);
359 if (num < 0 || num >= NR_syscalls)
361 mutex_lock(&syscall_trace_lock);
362 if (!sys_refcount_enter)
363 ret = register_trace_sys_enter(ftrace_syscall_enter);
365 pr_info("event trace: Could not activate"
366 "syscall entry trace point");
368 set_bit(num, enabled_enter_syscalls);
369 sys_refcount_enter++;
371 mutex_unlock(&syscall_trace_lock);
375 void unreg_event_syscall_enter(void *ptr)
381 num = syscall_name_to_nr(name);
382 if (num < 0 || num >= NR_syscalls)
384 mutex_lock(&syscall_trace_lock);
385 sys_refcount_enter--;
386 clear_bit(num, enabled_enter_syscalls);
387 if (!sys_refcount_enter)
388 unregister_trace_sys_enter(ftrace_syscall_enter);
389 mutex_unlock(&syscall_trace_lock);
392 int reg_event_syscall_exit(void *ptr)
399 num = syscall_name_to_nr(name);
400 if (num < 0 || num >= NR_syscalls)
402 mutex_lock(&syscall_trace_lock);
403 if (!sys_refcount_exit)
404 ret = register_trace_sys_exit(ftrace_syscall_exit);
406 pr_info("event trace: Could not activate"
407 "syscall exit trace point");
409 set_bit(num, enabled_exit_syscalls);
412 mutex_unlock(&syscall_trace_lock);
416 void unreg_event_syscall_exit(void *ptr)
422 num = syscall_name_to_nr(name);
423 if (num < 0 || num >= NR_syscalls)
425 mutex_lock(&syscall_trace_lock);
427 clear_bit(num, enabled_exit_syscalls);
428 if (!sys_refcount_exit)
429 unregister_trace_sys_exit(ftrace_syscall_exit);
430 mutex_unlock(&syscall_trace_lock);
433 struct trace_event event_syscall_enter = {
434 .trace = print_syscall_enter,
437 struct trace_event event_syscall_exit = {
438 .trace = print_syscall_exit,
441 int __init init_ftrace_syscalls(void)
443 struct syscall_metadata *meta;
447 syscalls_metadata = kzalloc(sizeof(*syscalls_metadata) *
448 NR_syscalls, GFP_KERNEL);
449 if (!syscalls_metadata) {
454 for (i = 0; i < NR_syscalls; i++) {
455 addr = arch_syscall_addr(i);
456 meta = find_syscall_meta(addr);
457 syscalls_metadata[i] = meta;
462 core_initcall(init_ftrace_syscalls);
464 #ifdef CONFIG_EVENT_PROFILE
466 static DECLARE_BITMAP(enabled_prof_enter_syscalls, NR_syscalls);
467 static DECLARE_BITMAP(enabled_prof_exit_syscalls, NR_syscalls);
468 static int sys_prof_refcount_enter;
469 static int sys_prof_refcount_exit;
471 static void prof_syscall_enter(struct pt_regs *regs, long id)
473 struct syscall_metadata *sys_data;
474 struct syscall_trace_enter *rec;
481 syscall_nr = syscall_get_nr(current, regs);
482 if (!test_bit(syscall_nr, enabled_prof_enter_syscalls))
485 sys_data = syscall_nr_to_meta(syscall_nr);
489 /* get the size after alignment with the u32 buffer size field */
490 size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
491 size = ALIGN(size + sizeof(u32), sizeof(u64));
494 if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
495 "profile buffer not large enough"))
498 /* Protect the per cpu buffer, begin the rcu read side */
499 local_irq_save(flags);
501 cpu = smp_processor_id();
504 raw_data = rcu_dereference(trace_profile_buf_nmi);
506 raw_data = rcu_dereference(trace_profile_buf);
511 raw_data = per_cpu_ptr(raw_data, cpu);
513 /* zero the dead bytes from align to not leak stack to user */
514 *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
516 rec = (struct syscall_trace_enter *) raw_data;
517 tracing_generic_entry_update(&rec->ent, 0, 0);
518 rec->ent.type = sys_data->enter_id;
519 rec->nr = syscall_nr;
520 syscall_get_arguments(current, regs, 0, sys_data->nb_args,
521 (unsigned long *)&rec->args);
522 perf_tp_event(sys_data->enter_id, 0, 1, rec, size);
525 local_irq_restore(flags);
528 int reg_prof_syscall_enter(char *name)
533 num = syscall_name_to_nr(name);
534 if (num < 0 || num >= NR_syscalls)
537 mutex_lock(&syscall_trace_lock);
538 if (!sys_prof_refcount_enter)
539 ret = register_trace_sys_enter(prof_syscall_enter);
541 pr_info("event trace: Could not activate"
542 "syscall entry trace point");
544 set_bit(num, enabled_prof_enter_syscalls);
545 sys_prof_refcount_enter++;
547 mutex_unlock(&syscall_trace_lock);
551 void unreg_prof_syscall_enter(char *name)
555 num = syscall_name_to_nr(name);
556 if (num < 0 || num >= NR_syscalls)
559 mutex_lock(&syscall_trace_lock);
560 sys_prof_refcount_enter--;
561 clear_bit(num, enabled_prof_enter_syscalls);
562 if (!sys_prof_refcount_enter)
563 unregister_trace_sys_enter(prof_syscall_enter);
564 mutex_unlock(&syscall_trace_lock);
567 static void prof_syscall_exit(struct pt_regs *regs, long ret)
569 struct syscall_metadata *sys_data;
570 struct syscall_trace_exit *rec;
577 syscall_nr = syscall_get_nr(current, regs);
578 if (!test_bit(syscall_nr, enabled_prof_exit_syscalls))
581 sys_data = syscall_nr_to_meta(syscall_nr);
585 /* We can probably do that at build time */
586 size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
590 * Impossible, but be paranoid with the future
591 * How to put this check outside runtime?
593 if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
594 "exit event has grown above profile buffer size"))
597 /* Protect the per cpu buffer, begin the rcu read side */
598 local_irq_save(flags);
599 cpu = smp_processor_id();
602 raw_data = rcu_dereference(trace_profile_buf_nmi);
604 raw_data = rcu_dereference(trace_profile_buf);
609 raw_data = per_cpu_ptr(raw_data, cpu);
611 /* zero the dead bytes from align to not leak stack to user */
612 *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
614 rec = (struct syscall_trace_exit *)raw_data;
616 tracing_generic_entry_update(&rec->ent, 0, 0);
617 rec->ent.type = sys_data->exit_id;
618 rec->nr = syscall_nr;
619 rec->ret = syscall_get_return_value(current, regs);
621 perf_tp_event(sys_data->exit_id, 0, 1, rec, size);
624 local_irq_restore(flags);
627 int reg_prof_syscall_exit(char *name)
632 num = syscall_name_to_nr(name);
633 if (num < 0 || num >= NR_syscalls)
636 mutex_lock(&syscall_trace_lock);
637 if (!sys_prof_refcount_exit)
638 ret = register_trace_sys_exit(prof_syscall_exit);
640 pr_info("event trace: Could not activate"
641 "syscall entry trace point");
643 set_bit(num, enabled_prof_exit_syscalls);
644 sys_prof_refcount_exit++;
646 mutex_unlock(&syscall_trace_lock);
650 void unreg_prof_syscall_exit(char *name)
654 num = syscall_name_to_nr(name);
655 if (num < 0 || num >= NR_syscalls)
658 mutex_lock(&syscall_trace_lock);
659 sys_prof_refcount_exit--;
660 clear_bit(num, enabled_prof_exit_syscalls);
661 if (!sys_prof_refcount_exit)
662 unregister_trace_sys_exit(prof_syscall_exit);
663 mutex_unlock(&syscall_trace_lock);