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 struct ftrace_event_class event_class_syscall_enter = {
25 .reg = syscall_enter_register
28 struct ftrace_event_class event_class_syscall_exit = {
30 .reg = syscall_exit_register
33 extern unsigned long __start_syscalls_metadata[];
34 extern unsigned long __stop_syscalls_metadata[];
36 static struct syscall_metadata **syscalls_metadata;
38 static struct syscall_metadata *find_syscall_meta(unsigned long syscall)
40 struct syscall_metadata *start;
41 struct syscall_metadata *stop;
42 char str[KSYM_SYMBOL_LEN];
45 start = (struct syscall_metadata *)__start_syscalls_metadata;
46 stop = (struct syscall_metadata *)__stop_syscalls_metadata;
47 kallsyms_lookup(syscall, NULL, NULL, NULL, str);
49 for ( ; start < stop; start++) {
51 * Only compare after the "sys" prefix. Archs that use
52 * syscall wrappers may have syscalls symbols aliases prefixed
53 * with "SyS" instead of "sys", leading to an unwanted
56 if (start->name && !strcmp(start->name + 3, str + 3))
62 static struct syscall_metadata *syscall_nr_to_meta(int nr)
64 if (!syscalls_metadata || nr >= NR_syscalls || nr < 0)
67 return syscalls_metadata[nr];
71 print_syscall_enter(struct trace_iterator *iter, int flags)
73 struct trace_seq *s = &iter->seq;
74 struct trace_entry *ent = iter->ent;
75 struct syscall_trace_enter *trace;
76 struct syscall_metadata *entry;
79 trace = (typeof(trace))ent;
81 entry = syscall_nr_to_meta(syscall);
86 if (entry->enter_event->id != ent->type) {
91 ret = trace_seq_printf(s, "%s(", entry->name);
93 return TRACE_TYPE_PARTIAL_LINE;
95 for (i = 0; i < entry->nb_args; i++) {
97 if (trace_flags & TRACE_ITER_VERBOSE) {
98 ret = trace_seq_printf(s, "%s ", entry->types[i]);
100 return TRACE_TYPE_PARTIAL_LINE;
102 /* parameter values */
103 ret = trace_seq_printf(s, "%s: %lx%s", entry->args[i],
105 i == entry->nb_args - 1 ? "" : ", ");
107 return TRACE_TYPE_PARTIAL_LINE;
110 ret = trace_seq_putc(s, ')');
112 return TRACE_TYPE_PARTIAL_LINE;
115 ret = trace_seq_putc(s, '\n');
117 return TRACE_TYPE_PARTIAL_LINE;
119 return TRACE_TYPE_HANDLED;
123 print_syscall_exit(struct trace_iterator *iter, int flags)
125 struct trace_seq *s = &iter->seq;
126 struct trace_entry *ent = iter->ent;
127 struct syscall_trace_exit *trace;
129 struct syscall_metadata *entry;
132 trace = (typeof(trace))ent;
134 entry = syscall_nr_to_meta(syscall);
137 trace_seq_printf(s, "\n");
138 return TRACE_TYPE_HANDLED;
141 if (entry->exit_event->id != ent->type) {
143 return TRACE_TYPE_UNHANDLED;
146 ret = trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
149 return TRACE_TYPE_PARTIAL_LINE;
151 return TRACE_TYPE_HANDLED;
154 extern char *__bad_type_size(void);
156 #define SYSCALL_FIELD(type, name) \
157 sizeof(type) != sizeof(trace.name) ? \
158 __bad_type_size() : \
159 #type, #name, offsetof(typeof(trace), name), \
160 sizeof(trace.name), is_signed_type(type)
163 int __set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len)
168 /* When len=0, we just calculate the needed length */
169 #define LEN_OR_ZERO (len ? len - pos : 0)
171 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
172 for (i = 0; i < entry->nb_args; i++) {
173 pos += snprintf(buf + pos, LEN_OR_ZERO, "%s: 0x%%0%zulx%s",
174 entry->args[i], sizeof(unsigned long),
175 i == entry->nb_args - 1 ? "" : ", ");
177 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
179 for (i = 0; i < entry->nb_args; i++) {
180 pos += snprintf(buf + pos, LEN_OR_ZERO,
181 ", ((unsigned long)(REC->%s))", entry->args[i]);
186 /* return the length of print_fmt */
190 static int set_syscall_print_fmt(struct ftrace_event_call *call)
194 struct syscall_metadata *entry = call->data;
196 if (entry->enter_event != call) {
197 call->print_fmt = "\"0x%lx\", REC->ret";
201 /* First: called with 0 length to calculate the needed length */
202 len = __set_enter_print_fmt(entry, NULL, 0);
204 print_fmt = kmalloc(len + 1, GFP_KERNEL);
208 /* Second: actually write the @print_fmt */
209 __set_enter_print_fmt(entry, print_fmt, len + 1);
210 call->print_fmt = print_fmt;
215 static void free_syscall_print_fmt(struct ftrace_event_call *call)
217 struct syscall_metadata *entry = call->data;
219 if (entry->enter_event == call)
220 kfree(call->print_fmt);
223 int syscall_enter_define_fields(struct ftrace_event_call *call)
225 struct syscall_trace_enter trace;
226 struct syscall_metadata *meta = call->data;
229 int offset = offsetof(typeof(trace), args);
231 ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
235 for (i = 0; i < meta->nb_args; i++) {
236 ret = trace_define_field(call, meta->types[i],
237 meta->args[i], offset,
238 sizeof(unsigned long), 0,
240 offset += sizeof(unsigned long);
246 int syscall_exit_define_fields(struct ftrace_event_call *call)
248 struct syscall_trace_exit trace;
251 ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
255 ret = trace_define_field(call, SYSCALL_FIELD(long, ret),
261 void ftrace_syscall_enter(void *ignore, struct pt_regs *regs, long id)
263 struct syscall_trace_enter *entry;
264 struct syscall_metadata *sys_data;
265 struct ring_buffer_event *event;
266 struct ring_buffer *buffer;
270 syscall_nr = syscall_get_nr(current, regs);
273 if (!test_bit(syscall_nr, enabled_enter_syscalls))
276 sys_data = syscall_nr_to_meta(syscall_nr);
280 size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
282 event = trace_current_buffer_lock_reserve(&buffer,
283 sys_data->enter_event->id, size, 0, 0);
287 entry = ring_buffer_event_data(event);
288 entry->nr = syscall_nr;
289 syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
291 if (!filter_current_check_discard(buffer, sys_data->enter_event,
293 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
296 void ftrace_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
298 struct syscall_trace_exit *entry;
299 struct syscall_metadata *sys_data;
300 struct ring_buffer_event *event;
301 struct ring_buffer *buffer;
304 syscall_nr = syscall_get_nr(current, regs);
307 if (!test_bit(syscall_nr, enabled_exit_syscalls))
310 sys_data = syscall_nr_to_meta(syscall_nr);
314 event = trace_current_buffer_lock_reserve(&buffer,
315 sys_data->exit_event->id, sizeof(*entry), 0, 0);
319 entry = ring_buffer_event_data(event);
320 entry->nr = syscall_nr;
321 entry->ret = syscall_get_return_value(current, regs);
323 if (!filter_current_check_discard(buffer, sys_data->exit_event,
325 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
328 int reg_event_syscall_enter(struct ftrace_event_call *call)
333 num = ((struct syscall_metadata *)call->data)->syscall_nr;
334 if (num < 0 || num >= NR_syscalls)
336 mutex_lock(&syscall_trace_lock);
337 if (!sys_refcount_enter)
338 ret = register_trace_sys_enter(ftrace_syscall_enter, NULL);
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, NULL);
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, NULL);
374 set_bit(num, enabled_exit_syscalls);
377 mutex_unlock(&syscall_trace_lock);
381 void unreg_event_syscall_exit(struct ftrace_event_call *call)
385 num = ((struct syscall_metadata *)call->data)->syscall_nr;
386 if (num < 0 || num >= NR_syscalls)
388 mutex_lock(&syscall_trace_lock);
390 clear_bit(num, enabled_exit_syscalls);
391 if (!sys_refcount_exit)
392 unregister_trace_sys_exit(ftrace_syscall_exit, NULL);
393 mutex_unlock(&syscall_trace_lock);
396 int init_syscall_trace(struct ftrace_event_call *call)
400 if (set_syscall_print_fmt(call) < 0)
403 id = trace_event_raw_init(call);
406 free_syscall_print_fmt(call);
413 unsigned long __init arch_syscall_addr(int nr)
415 return (unsigned long)sys_call_table[nr];
418 int __init init_ftrace_syscalls(void)
420 struct syscall_metadata *meta;
424 syscalls_metadata = kzalloc(sizeof(*syscalls_metadata) *
425 NR_syscalls, GFP_KERNEL);
426 if (!syscalls_metadata) {
431 for (i = 0; i < NR_syscalls; i++) {
432 addr = arch_syscall_addr(i);
433 meta = find_syscall_meta(addr);
437 meta->syscall_nr = i;
438 syscalls_metadata[i] = meta;
443 core_initcall(init_ftrace_syscalls);
445 #ifdef CONFIG_PERF_EVENTS
447 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
448 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
449 static int sys_perf_refcount_enter;
450 static int sys_perf_refcount_exit;
452 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
454 struct syscall_metadata *sys_data;
455 struct syscall_trace_enter *rec;
461 syscall_nr = syscall_get_nr(current, regs);
462 if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
465 sys_data = syscall_nr_to_meta(syscall_nr);
469 /* get the size after alignment with the u32 buffer size field */
470 size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
471 size = ALIGN(size + sizeof(u32), sizeof(u64));
474 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
475 "perf buffer not large enough"))
478 rec = (struct syscall_trace_enter *)perf_trace_buf_prepare(size,
479 sys_data->enter_event->id, &rctx, &flags);
483 rec->nr = syscall_nr;
484 syscall_get_arguments(current, regs, 0, sys_data->nb_args,
485 (unsigned long *)&rec->args);
486 perf_trace_buf_submit(rec, size, rctx, 0, 1, flags, regs);
489 int perf_sysenter_enable(struct ftrace_event_call *call)
494 num = ((struct syscall_metadata *)call->data)->syscall_nr;
496 mutex_lock(&syscall_trace_lock);
497 if (!sys_perf_refcount_enter)
498 ret = register_trace_sys_enter(perf_syscall_enter, NULL);
500 pr_info("event trace: Could not activate"
501 "syscall entry trace point");
503 set_bit(num, enabled_perf_enter_syscalls);
504 sys_perf_refcount_enter++;
506 mutex_unlock(&syscall_trace_lock);
510 void perf_sysenter_disable(struct ftrace_event_call *call)
514 num = ((struct syscall_metadata *)call->data)->syscall_nr;
516 mutex_lock(&syscall_trace_lock);
517 sys_perf_refcount_enter--;
518 clear_bit(num, enabled_perf_enter_syscalls);
519 if (!sys_perf_refcount_enter)
520 unregister_trace_sys_enter(perf_syscall_enter, NULL);
521 mutex_unlock(&syscall_trace_lock);
524 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
526 struct syscall_metadata *sys_data;
527 struct syscall_trace_exit *rec;
533 syscall_nr = syscall_get_nr(current, regs);
534 if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
537 sys_data = syscall_nr_to_meta(syscall_nr);
541 /* We can probably do that at build time */
542 size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
546 * Impossible, but be paranoid with the future
547 * How to put this check outside runtime?
549 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
550 "exit event has grown above perf buffer size"))
553 rec = (struct syscall_trace_exit *)perf_trace_buf_prepare(size,
554 sys_data->exit_event->id, &rctx, &flags);
558 rec->nr = syscall_nr;
559 rec->ret = syscall_get_return_value(current, regs);
561 perf_trace_buf_submit(rec, size, rctx, 0, 1, flags, regs);
564 int perf_sysexit_enable(struct ftrace_event_call *call)
569 num = ((struct syscall_metadata *)call->data)->syscall_nr;
571 mutex_lock(&syscall_trace_lock);
572 if (!sys_perf_refcount_exit)
573 ret = register_trace_sys_exit(perf_syscall_exit, NULL);
575 pr_info("event trace: Could not activate"
576 "syscall exit trace point");
578 set_bit(num, enabled_perf_exit_syscalls);
579 sys_perf_refcount_exit++;
581 mutex_unlock(&syscall_trace_lock);
585 void perf_sysexit_disable(struct ftrace_event_call *call)
589 num = ((struct syscall_metadata *)call->data)->syscall_nr;
591 mutex_lock(&syscall_trace_lock);
592 sys_perf_refcount_exit--;
593 clear_bit(num, enabled_perf_exit_syscalls);
594 if (!sys_perf_refcount_exit)
595 unregister_trace_sys_exit(perf_syscall_exit, NULL);
596 mutex_unlock(&syscall_trace_lock);
599 #endif /* CONFIG_PERF_EVENTS */
601 static int syscall_enter_register(struct ftrace_event_call *event,
605 case TRACE_REG_REGISTER:
606 return reg_event_syscall_enter(event);
607 case TRACE_REG_UNREGISTER:
608 unreg_event_syscall_enter(event);
611 #ifdef CONFIG_PERF_EVENTS
612 case TRACE_REG_PERF_REGISTER:
613 return perf_sysenter_enable(event);
614 case TRACE_REG_PERF_UNREGISTER:
615 perf_sysenter_disable(event);
622 static int syscall_exit_register(struct ftrace_event_call *event,
626 case TRACE_REG_REGISTER:
627 return reg_event_syscall_exit(event);
628 case TRACE_REG_UNREGISTER:
629 unreg_event_syscall_exit(event);
632 #ifdef CONFIG_PERF_EVENTS
633 case TRACE_REG_PERF_REGISTER:
634 return perf_sysexit_enable(event);
635 case TRACE_REG_PERF_UNREGISTER:
636 perf_sysexit_disable(event);