1 #include <trace/syscall.h>
2 #include <trace/events/syscalls.h>
3 #include <linux/slab.h>
4 #include <linux/kernel.h>
5 #include <linux/ftrace.h>
6 #include <linux/perf_event.h>
7 #include <asm/syscall.h>
9 #include "trace_output.h"
12 static DEFINE_MUTEX(syscall_trace_lock);
13 static int sys_refcount_enter;
14 static int sys_refcount_exit;
15 static DECLARE_BITMAP(enabled_enter_syscalls, NR_syscalls);
16 static DECLARE_BITMAP(enabled_exit_syscalls, NR_syscalls);
18 static int syscall_enter_register(struct ftrace_event_call *event,
20 static int syscall_exit_register(struct ftrace_event_call *event,
23 static int syscall_enter_define_fields(struct ftrace_event_call *call);
24 static int syscall_exit_define_fields(struct ftrace_event_call *call);
26 static struct list_head *
27 syscall_get_enter_fields(struct ftrace_event_call *call)
29 struct syscall_metadata *entry = call->data;
31 return &entry->enter_fields;
34 struct trace_event_functions enter_syscall_print_funcs = {
35 .trace = print_syscall_enter,
38 struct trace_event_functions exit_syscall_print_funcs = {
39 .trace = print_syscall_exit,
42 struct ftrace_event_class event_class_syscall_enter = {
44 .reg = syscall_enter_register,
45 .define_fields = syscall_enter_define_fields,
46 .get_fields = syscall_get_enter_fields,
47 .raw_init = init_syscall_trace,
50 struct ftrace_event_class event_class_syscall_exit = {
52 .reg = syscall_exit_register,
53 .define_fields = syscall_exit_define_fields,
54 .fields = LIST_HEAD_INIT(event_class_syscall_exit.fields),
55 .raw_init = init_syscall_trace,
58 extern struct syscall_metadata *__start_syscalls_metadata[];
59 extern struct syscall_metadata *__stop_syscalls_metadata[];
61 static struct syscall_metadata **syscalls_metadata;
63 static __init struct syscall_metadata *
64 find_syscall_meta(unsigned long syscall)
66 struct syscall_metadata **start;
67 struct syscall_metadata **stop;
68 char str[KSYM_SYMBOL_LEN];
71 start = __start_syscalls_metadata;
72 stop = __stop_syscalls_metadata;
73 kallsyms_lookup(syscall, NULL, NULL, NULL, str);
75 for ( ; start < stop; start++) {
77 * Only compare after the "sys" prefix. Archs that use
78 * syscall wrappers may have syscalls symbols aliases prefixed
79 * with "SyS" instead of "sys", leading to an unwanted
82 if ((*start)->name && !strcmp((*start)->name + 3, str + 3))
88 static struct syscall_metadata *syscall_nr_to_meta(int nr)
90 if (!syscalls_metadata || nr >= NR_syscalls || nr < 0)
93 return syscalls_metadata[nr];
97 print_syscall_enter(struct trace_iterator *iter, int flags,
98 struct trace_event *event)
100 struct trace_seq *s = &iter->seq;
101 struct trace_entry *ent = iter->ent;
102 struct syscall_trace_enter *trace;
103 struct syscall_metadata *entry;
106 trace = (typeof(trace))ent;
108 entry = syscall_nr_to_meta(syscall);
113 if (entry->enter_event->event.type != ent->type) {
118 ret = trace_seq_printf(s, "%s(", entry->name);
120 return TRACE_TYPE_PARTIAL_LINE;
122 for (i = 0; i < entry->nb_args; i++) {
123 /* parameter types */
124 if (trace_flags & TRACE_ITER_VERBOSE) {
125 ret = trace_seq_printf(s, "%s ", entry->types[i]);
127 return TRACE_TYPE_PARTIAL_LINE;
129 /* parameter values */
130 ret = trace_seq_printf(s, "%s: %lx%s", entry->args[i],
132 i == entry->nb_args - 1 ? "" : ", ");
134 return TRACE_TYPE_PARTIAL_LINE;
137 ret = trace_seq_putc(s, ')');
139 return TRACE_TYPE_PARTIAL_LINE;
142 ret = trace_seq_putc(s, '\n');
144 return TRACE_TYPE_PARTIAL_LINE;
146 return TRACE_TYPE_HANDLED;
150 print_syscall_exit(struct trace_iterator *iter, int flags,
151 struct trace_event *event)
153 struct trace_seq *s = &iter->seq;
154 struct trace_entry *ent = iter->ent;
155 struct syscall_trace_exit *trace;
157 struct syscall_metadata *entry;
160 trace = (typeof(trace))ent;
162 entry = syscall_nr_to_meta(syscall);
165 trace_seq_printf(s, "\n");
166 return TRACE_TYPE_HANDLED;
169 if (entry->exit_event->event.type != ent->type) {
171 return TRACE_TYPE_UNHANDLED;
174 ret = trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
177 return TRACE_TYPE_PARTIAL_LINE;
179 return TRACE_TYPE_HANDLED;
182 extern char *__bad_type_size(void);
184 #define SYSCALL_FIELD(type, name) \
185 sizeof(type) != sizeof(trace.name) ? \
186 __bad_type_size() : \
187 #type, #name, offsetof(typeof(trace), name), \
188 sizeof(trace.name), is_signed_type(type)
191 int __set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len)
196 /* When len=0, we just calculate the needed length */
197 #define LEN_OR_ZERO (len ? len - pos : 0)
199 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
200 for (i = 0; i < entry->nb_args; i++) {
201 pos += snprintf(buf + pos, LEN_OR_ZERO, "%s: 0x%%0%zulx%s",
202 entry->args[i], sizeof(unsigned long),
203 i == entry->nb_args - 1 ? "" : ", ");
205 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
207 for (i = 0; i < entry->nb_args; i++) {
208 pos += snprintf(buf + pos, LEN_OR_ZERO,
209 ", ((unsigned long)(REC->%s))", entry->args[i]);
214 /* return the length of print_fmt */
218 static int set_syscall_print_fmt(struct ftrace_event_call *call)
222 struct syscall_metadata *entry = call->data;
224 if (entry->enter_event != call) {
225 call->print_fmt = "\"0x%lx\", REC->ret";
229 /* First: called with 0 length to calculate the needed length */
230 len = __set_enter_print_fmt(entry, NULL, 0);
232 print_fmt = kmalloc(len + 1, GFP_KERNEL);
236 /* Second: actually write the @print_fmt */
237 __set_enter_print_fmt(entry, print_fmt, len + 1);
238 call->print_fmt = print_fmt;
243 static void free_syscall_print_fmt(struct ftrace_event_call *call)
245 struct syscall_metadata *entry = call->data;
247 if (entry->enter_event == call)
248 kfree(call->print_fmt);
251 static int syscall_enter_define_fields(struct ftrace_event_call *call)
253 struct syscall_trace_enter trace;
254 struct syscall_metadata *meta = call->data;
257 int offset = offsetof(typeof(trace), args);
259 ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
263 for (i = 0; i < meta->nb_args; i++) {
264 ret = trace_define_field(call, meta->types[i],
265 meta->args[i], offset,
266 sizeof(unsigned long), 0,
268 offset += sizeof(unsigned long);
274 static int syscall_exit_define_fields(struct ftrace_event_call *call)
276 struct syscall_trace_exit trace;
279 ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
283 ret = trace_define_field(call, SYSCALL_FIELD(long, ret),
289 void ftrace_syscall_enter(void *ignore, struct pt_regs *regs, long id)
291 struct syscall_trace_enter *entry;
292 struct syscall_metadata *sys_data;
293 struct ring_buffer_event *event;
294 struct ring_buffer *buffer;
298 syscall_nr = syscall_get_nr(current, regs);
301 if (!test_bit(syscall_nr, enabled_enter_syscalls))
304 sys_data = syscall_nr_to_meta(syscall_nr);
308 size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
310 event = trace_current_buffer_lock_reserve(&buffer,
311 sys_data->enter_event->event.type, size, 0, 0);
315 entry = ring_buffer_event_data(event);
316 entry->nr = syscall_nr;
317 syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
319 if (!filter_current_check_discard(buffer, sys_data->enter_event,
321 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
324 void ftrace_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
326 struct syscall_trace_exit *entry;
327 struct syscall_metadata *sys_data;
328 struct ring_buffer_event *event;
329 struct ring_buffer *buffer;
332 syscall_nr = syscall_get_nr(current, regs);
335 if (!test_bit(syscall_nr, enabled_exit_syscalls))
338 sys_data = syscall_nr_to_meta(syscall_nr);
342 event = trace_current_buffer_lock_reserve(&buffer,
343 sys_data->exit_event->event.type, sizeof(*entry), 0, 0);
347 entry = ring_buffer_event_data(event);
348 entry->nr = syscall_nr;
349 entry->ret = syscall_get_return_value(current, regs);
351 if (!filter_current_check_discard(buffer, sys_data->exit_event,
353 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
356 int reg_event_syscall_enter(struct ftrace_event_call *call)
361 num = ((struct syscall_metadata *)call->data)->syscall_nr;
362 if (num < 0 || num >= NR_syscalls)
364 mutex_lock(&syscall_trace_lock);
365 if (!sys_refcount_enter)
366 ret = register_trace_sys_enter(ftrace_syscall_enter, NULL);
368 set_bit(num, enabled_enter_syscalls);
369 sys_refcount_enter++;
371 mutex_unlock(&syscall_trace_lock);
375 void unreg_event_syscall_enter(struct ftrace_event_call *call)
379 num = ((struct syscall_metadata *)call->data)->syscall_nr;
380 if (num < 0 || num >= NR_syscalls)
382 mutex_lock(&syscall_trace_lock);
383 sys_refcount_enter--;
384 clear_bit(num, enabled_enter_syscalls);
385 if (!sys_refcount_enter)
386 unregister_trace_sys_enter(ftrace_syscall_enter, NULL);
387 mutex_unlock(&syscall_trace_lock);
390 int reg_event_syscall_exit(struct ftrace_event_call *call)
395 num = ((struct syscall_metadata *)call->data)->syscall_nr;
396 if (num < 0 || num >= NR_syscalls)
398 mutex_lock(&syscall_trace_lock);
399 if (!sys_refcount_exit)
400 ret = register_trace_sys_exit(ftrace_syscall_exit, NULL);
402 set_bit(num, enabled_exit_syscalls);
405 mutex_unlock(&syscall_trace_lock);
409 void unreg_event_syscall_exit(struct ftrace_event_call *call)
413 num = ((struct syscall_metadata *)call->data)->syscall_nr;
414 if (num < 0 || num >= NR_syscalls)
416 mutex_lock(&syscall_trace_lock);
418 clear_bit(num, enabled_exit_syscalls);
419 if (!sys_refcount_exit)
420 unregister_trace_sys_exit(ftrace_syscall_exit, NULL);
421 mutex_unlock(&syscall_trace_lock);
424 int init_syscall_trace(struct ftrace_event_call *call)
428 if (set_syscall_print_fmt(call) < 0)
431 id = trace_event_raw_init(call);
434 free_syscall_print_fmt(call);
441 unsigned long __init arch_syscall_addr(int nr)
443 return (unsigned long)sys_call_table[nr];
446 int __init init_ftrace_syscalls(void)
448 struct syscall_metadata *meta;
452 syscalls_metadata = kzalloc(sizeof(*syscalls_metadata) *
453 NR_syscalls, GFP_KERNEL);
454 if (!syscalls_metadata) {
459 for (i = 0; i < NR_syscalls; i++) {
460 addr = arch_syscall_addr(i);
461 meta = find_syscall_meta(addr);
465 meta->syscall_nr = i;
466 syscalls_metadata[i] = meta;
471 core_initcall(init_ftrace_syscalls);
473 #ifdef CONFIG_PERF_EVENTS
475 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
476 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
477 static int sys_perf_refcount_enter;
478 static int sys_perf_refcount_exit;
480 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
482 struct syscall_metadata *sys_data;
483 struct syscall_trace_enter *rec;
484 struct hlist_head *head;
489 syscall_nr = syscall_get_nr(current, regs);
490 if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
493 sys_data = syscall_nr_to_meta(syscall_nr);
497 /* get the size after alignment with the u32 buffer size field */
498 size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
499 size = ALIGN(size + sizeof(u32), sizeof(u64));
502 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
503 "perf buffer not large enough"))
506 rec = (struct syscall_trace_enter *)perf_trace_buf_prepare(size,
507 sys_data->enter_event->event.type, regs, &rctx);
511 rec->nr = syscall_nr;
512 syscall_get_arguments(current, regs, 0, sys_data->nb_args,
513 (unsigned long *)&rec->args);
515 head = this_cpu_ptr(sys_data->enter_event->perf_events);
516 perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head);
519 int perf_sysenter_enable(struct ftrace_event_call *call)
524 num = ((struct syscall_metadata *)call->data)->syscall_nr;
526 mutex_lock(&syscall_trace_lock);
527 if (!sys_perf_refcount_enter)
528 ret = register_trace_sys_enter(perf_syscall_enter, NULL);
530 pr_info("event trace: Could not activate"
531 "syscall entry trace point");
533 set_bit(num, enabled_perf_enter_syscalls);
534 sys_perf_refcount_enter++;
536 mutex_unlock(&syscall_trace_lock);
540 void perf_sysenter_disable(struct ftrace_event_call *call)
544 num = ((struct syscall_metadata *)call->data)->syscall_nr;
546 mutex_lock(&syscall_trace_lock);
547 sys_perf_refcount_enter--;
548 clear_bit(num, enabled_perf_enter_syscalls);
549 if (!sys_perf_refcount_enter)
550 unregister_trace_sys_enter(perf_syscall_enter, NULL);
551 mutex_unlock(&syscall_trace_lock);
554 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
556 struct syscall_metadata *sys_data;
557 struct syscall_trace_exit *rec;
558 struct hlist_head *head;
563 syscall_nr = syscall_get_nr(current, regs);
564 if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
567 sys_data = syscall_nr_to_meta(syscall_nr);
571 /* We can probably do that at build time */
572 size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
576 * Impossible, but be paranoid with the future
577 * How to put this check outside runtime?
579 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
580 "exit event has grown above perf buffer size"))
583 rec = (struct syscall_trace_exit *)perf_trace_buf_prepare(size,
584 sys_data->exit_event->event.type, regs, &rctx);
588 rec->nr = syscall_nr;
589 rec->ret = syscall_get_return_value(current, regs);
591 head = this_cpu_ptr(sys_data->exit_event->perf_events);
592 perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head);
595 int perf_sysexit_enable(struct ftrace_event_call *call)
600 num = ((struct syscall_metadata *)call->data)->syscall_nr;
602 mutex_lock(&syscall_trace_lock);
603 if (!sys_perf_refcount_exit)
604 ret = register_trace_sys_exit(perf_syscall_exit, NULL);
606 pr_info("event trace: Could not activate"
607 "syscall exit trace point");
609 set_bit(num, enabled_perf_exit_syscalls);
610 sys_perf_refcount_exit++;
612 mutex_unlock(&syscall_trace_lock);
616 void perf_sysexit_disable(struct ftrace_event_call *call)
620 num = ((struct syscall_metadata *)call->data)->syscall_nr;
622 mutex_lock(&syscall_trace_lock);
623 sys_perf_refcount_exit--;
624 clear_bit(num, enabled_perf_exit_syscalls);
625 if (!sys_perf_refcount_exit)
626 unregister_trace_sys_exit(perf_syscall_exit, NULL);
627 mutex_unlock(&syscall_trace_lock);
630 #endif /* CONFIG_PERF_EVENTS */
632 static int syscall_enter_register(struct ftrace_event_call *event,
636 case TRACE_REG_REGISTER:
637 return reg_event_syscall_enter(event);
638 case TRACE_REG_UNREGISTER:
639 unreg_event_syscall_enter(event);
642 #ifdef CONFIG_PERF_EVENTS
643 case TRACE_REG_PERF_REGISTER:
644 return perf_sysenter_enable(event);
645 case TRACE_REG_PERF_UNREGISTER:
646 perf_sysenter_disable(event);
653 static int syscall_exit_register(struct ftrace_event_call *event,
657 case TRACE_REG_REGISTER:
658 return reg_event_syscall_exit(event);
659 case TRACE_REG_UNREGISTER:
660 unreg_event_syscall_exit(event);
663 #ifdef CONFIG_PERF_EVENTS
664 case TRACE_REG_PERF_REGISTER:
665 return perf_sysexit_enable(event);
666 case TRACE_REG_PERF_UNREGISTER:
667 perf_sysexit_disable(event);