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');
194 int syscall_exit_format(struct ftrace_event_call *call, struct trace_seq *s)
197 struct syscall_trace_exit trace;
199 ret = trace_seq_printf(s,
200 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;"
202 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;"
204 SYSCALL_FIELD(int, nr),
205 SYSCALL_FIELD(long, ret));
209 return trace_seq_printf(s, "\nprint fmt: \"0x%%lx\", REC->ret\n");
212 int syscall_enter_define_fields(struct ftrace_event_call *call)
214 struct syscall_trace_enter trace;
215 struct syscall_metadata *meta = call->data;
218 int offset = offsetof(typeof(trace), args);
220 ret = trace_define_common_fields(call);
224 ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
228 for (i = 0; i < meta->nb_args; i++) {
229 ret = trace_define_field(call, meta->types[i],
230 meta->args[i], offset,
231 sizeof(unsigned long), 0,
233 offset += sizeof(unsigned long);
239 int syscall_exit_define_fields(struct ftrace_event_call *call)
241 struct syscall_trace_exit trace;
244 ret = trace_define_common_fields(call);
248 ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
252 ret = trace_define_field(call, SYSCALL_FIELD(long, ret),
258 void ftrace_syscall_enter(struct pt_regs *regs, long id)
260 struct syscall_trace_enter *entry;
261 struct syscall_metadata *sys_data;
262 struct ring_buffer_event *event;
263 struct ring_buffer *buffer;
267 syscall_nr = syscall_get_nr(current, regs);
270 if (!test_bit(syscall_nr, enabled_enter_syscalls))
273 sys_data = syscall_nr_to_meta(syscall_nr);
277 size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
279 event = trace_current_buffer_lock_reserve(&buffer,
280 sys_data->enter_event->id, size, 0, 0);
284 entry = ring_buffer_event_data(event);
285 entry->nr = syscall_nr;
286 syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
288 if (!filter_current_check_discard(buffer, sys_data->enter_event,
290 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
293 void ftrace_syscall_exit(struct pt_regs *regs, long ret)
295 struct syscall_trace_exit *entry;
296 struct syscall_metadata *sys_data;
297 struct ring_buffer_event *event;
298 struct ring_buffer *buffer;
301 syscall_nr = syscall_get_nr(current, regs);
304 if (!test_bit(syscall_nr, enabled_exit_syscalls))
307 sys_data = syscall_nr_to_meta(syscall_nr);
311 event = trace_current_buffer_lock_reserve(&buffer,
312 sys_data->exit_event->id, sizeof(*entry), 0, 0);
316 entry = ring_buffer_event_data(event);
317 entry->nr = syscall_nr;
318 entry->ret = syscall_get_return_value(current, regs);
320 if (!filter_current_check_discard(buffer, sys_data->exit_event,
322 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
325 int reg_event_syscall_enter(struct ftrace_event_call *call)
330 num = ((struct syscall_metadata *)call->data)->syscall_nr;
331 if (num < 0 || num >= NR_syscalls)
333 mutex_lock(&syscall_trace_lock);
334 if (!sys_refcount_enter)
335 ret = register_trace_sys_enter(ftrace_syscall_enter);
337 pr_info("event trace: Could not activate"
338 "syscall entry trace point");
340 set_bit(num, enabled_enter_syscalls);
341 sys_refcount_enter++;
343 mutex_unlock(&syscall_trace_lock);
347 void unreg_event_syscall_enter(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 sys_refcount_enter--;
356 clear_bit(num, enabled_enter_syscalls);
357 if (!sys_refcount_enter)
358 unregister_trace_sys_enter(ftrace_syscall_enter);
359 mutex_unlock(&syscall_trace_lock);
362 int reg_event_syscall_exit(struct ftrace_event_call *call)
367 num = ((struct syscall_metadata *)call->data)->syscall_nr;
368 if (num < 0 || num >= NR_syscalls)
370 mutex_lock(&syscall_trace_lock);
371 if (!sys_refcount_exit)
372 ret = register_trace_sys_exit(ftrace_syscall_exit);
374 pr_info("event trace: Could not activate"
375 "syscall exit trace point");
377 set_bit(num, enabled_exit_syscalls);
380 mutex_unlock(&syscall_trace_lock);
384 void unreg_event_syscall_exit(struct ftrace_event_call *call)
388 num = ((struct syscall_metadata *)call->data)->syscall_nr;
389 if (num < 0 || num >= NR_syscalls)
391 mutex_lock(&syscall_trace_lock);
393 clear_bit(num, enabled_exit_syscalls);
394 if (!sys_refcount_exit)
395 unregister_trace_sys_exit(ftrace_syscall_exit);
396 mutex_unlock(&syscall_trace_lock);
399 int init_syscall_trace(struct ftrace_event_call *call)
403 id = register_ftrace_event(call->event);
407 INIT_LIST_HEAD(&call->fields);
411 int __init init_ftrace_syscalls(void)
413 struct syscall_metadata *meta;
417 syscalls_metadata = kzalloc(sizeof(*syscalls_metadata) *
418 NR_syscalls, GFP_KERNEL);
419 if (!syscalls_metadata) {
424 for (i = 0; i < NR_syscalls; i++) {
425 addr = arch_syscall_addr(i);
426 meta = find_syscall_meta(addr);
430 meta->syscall_nr = i;
431 syscalls_metadata[i] = meta;
436 core_initcall(init_ftrace_syscalls);
438 #ifdef CONFIG_EVENT_PROFILE
440 static DECLARE_BITMAP(enabled_prof_enter_syscalls, NR_syscalls);
441 static DECLARE_BITMAP(enabled_prof_exit_syscalls, NR_syscalls);
442 static int sys_prof_refcount_enter;
443 static int sys_prof_refcount_exit;
445 static void prof_syscall_enter(struct pt_regs *regs, long id)
447 struct syscall_metadata *sys_data;
448 struct syscall_trace_enter *rec;
457 syscall_nr = syscall_get_nr(current, regs);
458 if (!test_bit(syscall_nr, enabled_prof_enter_syscalls))
461 sys_data = syscall_nr_to_meta(syscall_nr);
465 /* get the size after alignment with the u32 buffer size field */
466 size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
467 size = ALIGN(size + sizeof(u32), sizeof(u64));
470 if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
471 "profile buffer not large enough"))
474 /* Protect the per cpu buffer, begin the rcu read side */
475 local_irq_save(flags);
477 rctx = perf_swevent_get_recursion_context();
481 cpu = smp_processor_id();
483 trace_buf = rcu_dereference(perf_trace_buf);
488 raw_data = per_cpu_ptr(trace_buf, cpu);
490 /* zero the dead bytes from align to not leak stack to user */
491 *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
493 rec = (struct syscall_trace_enter *) raw_data;
494 tracing_generic_entry_update(&rec->ent, 0, 0);
495 rec->ent.type = sys_data->enter_event->id;
496 rec->nr = syscall_nr;
497 syscall_get_arguments(current, regs, 0, sys_data->nb_args,
498 (unsigned long *)&rec->args);
499 perf_tp_event(sys_data->enter_event->id, 0, 1, rec, size);
502 perf_swevent_put_recursion_context(rctx);
504 local_irq_restore(flags);
507 int prof_sysenter_enable(struct ftrace_event_call *call)
512 num = ((struct syscall_metadata *)call->data)->syscall_nr;
514 mutex_lock(&syscall_trace_lock);
515 if (!sys_prof_refcount_enter)
516 ret = register_trace_sys_enter(prof_syscall_enter);
518 pr_info("event trace: Could not activate"
519 "syscall entry trace point");
521 set_bit(num, enabled_prof_enter_syscalls);
522 sys_prof_refcount_enter++;
524 mutex_unlock(&syscall_trace_lock);
528 void prof_sysenter_disable(struct ftrace_event_call *call)
532 num = ((struct syscall_metadata *)call->data)->syscall_nr;
534 mutex_lock(&syscall_trace_lock);
535 sys_prof_refcount_enter--;
536 clear_bit(num, enabled_prof_enter_syscalls);
537 if (!sys_prof_refcount_enter)
538 unregister_trace_sys_enter(prof_syscall_enter);
539 mutex_unlock(&syscall_trace_lock);
542 static void prof_syscall_exit(struct pt_regs *regs, long ret)
544 struct syscall_metadata *sys_data;
545 struct syscall_trace_exit *rec;
554 syscall_nr = syscall_get_nr(current, regs);
555 if (!test_bit(syscall_nr, enabled_prof_exit_syscalls))
558 sys_data = syscall_nr_to_meta(syscall_nr);
562 /* We can probably do that at build time */
563 size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
567 * Impossible, but be paranoid with the future
568 * How to put this check outside runtime?
570 if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
571 "exit event has grown above profile buffer size"))
574 /* Protect the per cpu buffer, begin the rcu read side */
575 local_irq_save(flags);
577 rctx = perf_swevent_get_recursion_context();
581 cpu = smp_processor_id();
583 trace_buf = rcu_dereference(perf_trace_buf);
588 raw_data = per_cpu_ptr(trace_buf, cpu);
590 /* zero the dead bytes from align to not leak stack to user */
591 *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
593 rec = (struct syscall_trace_exit *)raw_data;
595 tracing_generic_entry_update(&rec->ent, 0, 0);
596 rec->ent.type = sys_data->exit_event->id;
597 rec->nr = syscall_nr;
598 rec->ret = syscall_get_return_value(current, regs);
600 perf_tp_event(sys_data->exit_event->id, 0, 1, rec, size);
603 perf_swevent_put_recursion_context(rctx);
605 local_irq_restore(flags);
608 int prof_sysexit_enable(struct ftrace_event_call *call)
613 num = ((struct syscall_metadata *)call->data)->syscall_nr;
615 mutex_lock(&syscall_trace_lock);
616 if (!sys_prof_refcount_exit)
617 ret = register_trace_sys_exit(prof_syscall_exit);
619 pr_info("event trace: Could not activate"
620 "syscall entry trace point");
622 set_bit(num, enabled_prof_exit_syscalls);
623 sys_prof_refcount_exit++;
625 mutex_unlock(&syscall_trace_lock);
629 void prof_sysexit_disable(struct ftrace_event_call *call)
633 num = ((struct syscall_metadata *)call->data)->syscall_nr;
635 mutex_lock(&syscall_trace_lock);
636 sys_prof_refcount_exit--;
637 clear_bit(num, enabled_prof_exit_syscalls);
638 if (!sys_prof_refcount_exit)
639 unregister_trace_sys_exit(prof_syscall_exit);
640 mutex_unlock(&syscall_trace_lock);