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)
147 int __set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len)
152 /* When len=0, we just calculate the needed length */
153 #define LEN_OR_ZERO (len ? len - pos : 0)
155 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
156 for (i = 0; i < entry->nb_args; i++) {
157 pos += snprintf(buf + pos, LEN_OR_ZERO, "%s: 0x%%0%zulx%s",
158 entry->args[i], sizeof(unsigned long),
159 i == entry->nb_args - 1 ? "" : ", ");
161 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
163 for (i = 0; i < entry->nb_args; i++) {
164 pos += snprintf(buf + pos, LEN_OR_ZERO,
165 ", ((unsigned long)(REC->%s))", entry->args[i]);
170 /* return the length of print_fmt */
174 static int set_syscall_print_fmt(struct ftrace_event_call *call)
178 struct syscall_metadata *entry = call->data;
180 if (entry->enter_event != call) {
181 call->print_fmt = "\"0x%lx\", REC->ret";
185 /* First: called with 0 length to calculate the needed length */
186 len = __set_enter_print_fmt(entry, NULL, 0);
188 print_fmt = kmalloc(len + 1, GFP_KERNEL);
192 /* Second: actually write the @print_fmt */
193 __set_enter_print_fmt(entry, print_fmt, len + 1);
194 call->print_fmt = print_fmt;
199 static void free_syscall_print_fmt(struct ftrace_event_call *call)
201 struct syscall_metadata *entry = call->data;
203 if (entry->enter_event == call)
204 kfree(call->print_fmt);
207 int syscall_enter_define_fields(struct ftrace_event_call *call)
209 struct syscall_trace_enter trace;
210 struct syscall_metadata *meta = call->data;
213 int offset = offsetof(typeof(trace), args);
215 ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
219 for (i = 0; i < meta->nb_args; i++) {
220 ret = trace_define_field(call, meta->types[i],
221 meta->args[i], offset,
222 sizeof(unsigned long), 0,
224 offset += sizeof(unsigned long);
230 int syscall_exit_define_fields(struct ftrace_event_call *call)
232 struct syscall_trace_exit trace;
235 ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
239 ret = trace_define_field(call, SYSCALL_FIELD(long, ret),
245 void ftrace_syscall_enter(struct pt_regs *regs, long id)
247 struct syscall_trace_enter *entry;
248 struct syscall_metadata *sys_data;
249 struct ring_buffer_event *event;
250 struct ring_buffer *buffer;
254 syscall_nr = syscall_get_nr(current, regs);
257 if (!test_bit(syscall_nr, enabled_enter_syscalls))
260 sys_data = syscall_nr_to_meta(syscall_nr);
264 size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
266 event = trace_current_buffer_lock_reserve(&buffer,
267 sys_data->enter_event->id, size, 0, 0);
271 entry = ring_buffer_event_data(event);
272 entry->nr = syscall_nr;
273 syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
275 if (!filter_current_check_discard(buffer, sys_data->enter_event,
277 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
280 void ftrace_syscall_exit(struct pt_regs *regs, long ret)
282 struct syscall_trace_exit *entry;
283 struct syscall_metadata *sys_data;
284 struct ring_buffer_event *event;
285 struct ring_buffer *buffer;
288 syscall_nr = syscall_get_nr(current, regs);
291 if (!test_bit(syscall_nr, enabled_exit_syscalls))
294 sys_data = syscall_nr_to_meta(syscall_nr);
298 event = trace_current_buffer_lock_reserve(&buffer,
299 sys_data->exit_event->id, sizeof(*entry), 0, 0);
303 entry = ring_buffer_event_data(event);
304 entry->nr = syscall_nr;
305 entry->ret = syscall_get_return_value(current, regs);
307 if (!filter_current_check_discard(buffer, sys_data->exit_event,
309 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
312 int reg_event_syscall_enter(struct ftrace_event_call *call)
317 num = ((struct syscall_metadata *)call->data)->syscall_nr;
318 if (num < 0 || num >= NR_syscalls)
320 mutex_lock(&syscall_trace_lock);
321 if (!sys_refcount_enter)
322 ret = register_trace_sys_enter(ftrace_syscall_enter);
324 set_bit(num, enabled_enter_syscalls);
325 sys_refcount_enter++;
327 mutex_unlock(&syscall_trace_lock);
331 void unreg_event_syscall_enter(struct ftrace_event_call *call)
335 num = ((struct syscall_metadata *)call->data)->syscall_nr;
336 if (num < 0 || num >= NR_syscalls)
338 mutex_lock(&syscall_trace_lock);
339 sys_refcount_enter--;
340 clear_bit(num, enabled_enter_syscalls);
341 if (!sys_refcount_enter)
342 unregister_trace_sys_enter(ftrace_syscall_enter);
343 mutex_unlock(&syscall_trace_lock);
346 int reg_event_syscall_exit(struct ftrace_event_call *call)
351 num = ((struct syscall_metadata *)call->data)->syscall_nr;
352 if (num < 0 || num >= NR_syscalls)
354 mutex_lock(&syscall_trace_lock);
355 if (!sys_refcount_exit)
356 ret = register_trace_sys_exit(ftrace_syscall_exit);
358 set_bit(num, enabled_exit_syscalls);
361 mutex_unlock(&syscall_trace_lock);
365 void unreg_event_syscall_exit(struct ftrace_event_call *call)
369 num = ((struct syscall_metadata *)call->data)->syscall_nr;
370 if (num < 0 || num >= NR_syscalls)
372 mutex_lock(&syscall_trace_lock);
374 clear_bit(num, enabled_exit_syscalls);
375 if (!sys_refcount_exit)
376 unregister_trace_sys_exit(ftrace_syscall_exit);
377 mutex_unlock(&syscall_trace_lock);
380 int init_syscall_trace(struct ftrace_event_call *call)
384 if (set_syscall_print_fmt(call) < 0)
387 id = trace_event_raw_init(call);
390 free_syscall_print_fmt(call);
397 unsigned long __init arch_syscall_addr(int nr)
399 return (unsigned long)sys_call_table[nr];
402 int __init init_ftrace_syscalls(void)
404 struct syscall_metadata *meta;
408 syscalls_metadata = kzalloc(sizeof(*syscalls_metadata) *
409 NR_syscalls, GFP_KERNEL);
410 if (!syscalls_metadata) {
415 for (i = 0; i < NR_syscalls; i++) {
416 addr = arch_syscall_addr(i);
417 meta = find_syscall_meta(addr);
421 meta->syscall_nr = i;
422 syscalls_metadata[i] = meta;
427 core_initcall(init_ftrace_syscalls);
429 #ifdef CONFIG_PERF_EVENTS
431 static DECLARE_BITMAP(enabled_prof_enter_syscalls, NR_syscalls);
432 static DECLARE_BITMAP(enabled_prof_exit_syscalls, NR_syscalls);
433 static int sys_prof_refcount_enter;
434 static int sys_prof_refcount_exit;
436 static void prof_syscall_enter(struct pt_regs *regs, long id)
438 struct syscall_metadata *sys_data;
439 struct syscall_trace_enter *rec;
445 syscall_nr = syscall_get_nr(current, regs);
446 if (!test_bit(syscall_nr, enabled_prof_enter_syscalls))
449 sys_data = syscall_nr_to_meta(syscall_nr);
453 /* get the size after alignment with the u32 buffer size field */
454 size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
455 size = ALIGN(size + sizeof(u32), sizeof(u64));
458 if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
459 "profile buffer not large enough"))
462 rec = (struct syscall_trace_enter *)ftrace_perf_buf_prepare(size,
463 sys_data->enter_event->id, &rctx, &flags);
467 rec->nr = syscall_nr;
468 syscall_get_arguments(current, regs, 0, sys_data->nb_args,
469 (unsigned long *)&rec->args);
470 ftrace_perf_buf_submit(rec, size, rctx, 0, 1, flags);
473 int prof_sysenter_enable(struct ftrace_event_call *call)
478 num = ((struct syscall_metadata *)call->data)->syscall_nr;
480 mutex_lock(&syscall_trace_lock);
481 if (!sys_prof_refcount_enter)
482 ret = register_trace_sys_enter(prof_syscall_enter);
484 pr_info("event trace: Could not activate"
485 "syscall entry trace point");
487 set_bit(num, enabled_prof_enter_syscalls);
488 sys_prof_refcount_enter++;
490 mutex_unlock(&syscall_trace_lock);
494 void prof_sysenter_disable(struct ftrace_event_call *call)
498 num = ((struct syscall_metadata *)call->data)->syscall_nr;
500 mutex_lock(&syscall_trace_lock);
501 sys_prof_refcount_enter--;
502 clear_bit(num, enabled_prof_enter_syscalls);
503 if (!sys_prof_refcount_enter)
504 unregister_trace_sys_enter(prof_syscall_enter);
505 mutex_unlock(&syscall_trace_lock);
508 static void prof_syscall_exit(struct pt_regs *regs, long ret)
510 struct syscall_metadata *sys_data;
511 struct syscall_trace_exit *rec;
517 syscall_nr = syscall_get_nr(current, regs);
518 if (!test_bit(syscall_nr, enabled_prof_exit_syscalls))
521 sys_data = syscall_nr_to_meta(syscall_nr);
525 /* We can probably do that at build time */
526 size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
530 * Impossible, but be paranoid with the future
531 * How to put this check outside runtime?
533 if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
534 "exit event has grown above profile buffer size"))
537 rec = (struct syscall_trace_exit *)ftrace_perf_buf_prepare(size,
538 sys_data->exit_event->id, &rctx, &flags);
542 rec->nr = syscall_nr;
543 rec->ret = syscall_get_return_value(current, regs);
545 ftrace_perf_buf_submit(rec, size, rctx, 0, 1, flags);
548 int prof_sysexit_enable(struct ftrace_event_call *call)
553 num = ((struct syscall_metadata *)call->data)->syscall_nr;
555 mutex_lock(&syscall_trace_lock);
556 if (!sys_prof_refcount_exit)
557 ret = register_trace_sys_exit(prof_syscall_exit);
559 pr_info("event trace: Could not activate"
560 "syscall exit trace point");
562 set_bit(num, enabled_prof_exit_syscalls);
563 sys_prof_refcount_exit++;
565 mutex_unlock(&syscall_trace_lock);
569 void prof_sysexit_disable(struct ftrace_event_call *call)
573 num = ((struct syscall_metadata *)call->data)->syscall_nr;
575 mutex_lock(&syscall_trace_lock);
576 sys_prof_refcount_exit--;
577 clear_bit(num, enabled_prof_exit_syscalls);
578 if (!sys_prof_refcount_exit)
579 unregister_trace_sys_exit(prof_syscall_exit);
580 mutex_unlock(&syscall_trace_lock);
583 #endif /* CONFIG_PERF_EVENTS */