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_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
224 for (i = 0; i < meta->nb_args; i++) {
225 ret = trace_define_field(call, meta->types[i],
226 meta->args[i], offset,
227 sizeof(unsigned long), 0,
229 offset += sizeof(unsigned long);
235 int syscall_exit_define_fields(struct ftrace_event_call *call)
237 struct syscall_trace_exit trace;
240 ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
244 ret = trace_define_field(call, SYSCALL_FIELD(long, ret),
250 void ftrace_syscall_enter(struct pt_regs *regs, long id)
252 struct syscall_trace_enter *entry;
253 struct syscall_metadata *sys_data;
254 struct ring_buffer_event *event;
255 struct ring_buffer *buffer;
259 syscall_nr = syscall_get_nr(current, regs);
262 if (!test_bit(syscall_nr, enabled_enter_syscalls))
265 sys_data = syscall_nr_to_meta(syscall_nr);
269 size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
271 event = trace_current_buffer_lock_reserve(&buffer,
272 sys_data->enter_event->id, size, 0, 0);
276 entry = ring_buffer_event_data(event);
277 entry->nr = syscall_nr;
278 syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
280 if (!filter_current_check_discard(buffer, sys_data->enter_event,
282 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
285 void ftrace_syscall_exit(struct pt_regs *regs, long ret)
287 struct syscall_trace_exit *entry;
288 struct syscall_metadata *sys_data;
289 struct ring_buffer_event *event;
290 struct ring_buffer *buffer;
293 syscall_nr = syscall_get_nr(current, regs);
296 if (!test_bit(syscall_nr, enabled_exit_syscalls))
299 sys_data = syscall_nr_to_meta(syscall_nr);
303 event = trace_current_buffer_lock_reserve(&buffer,
304 sys_data->exit_event->id, sizeof(*entry), 0, 0);
308 entry = ring_buffer_event_data(event);
309 entry->nr = syscall_nr;
310 entry->ret = syscall_get_return_value(current, regs);
312 if (!filter_current_check_discard(buffer, sys_data->exit_event,
314 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
317 int reg_event_syscall_enter(struct ftrace_event_call *call)
322 num = ((struct syscall_metadata *)call->data)->syscall_nr;
323 if (num < 0 || num >= NR_syscalls)
325 mutex_lock(&syscall_trace_lock);
326 if (!sys_refcount_enter)
327 ret = register_trace_sys_enter(ftrace_syscall_enter);
329 set_bit(num, enabled_enter_syscalls);
330 sys_refcount_enter++;
332 mutex_unlock(&syscall_trace_lock);
336 void unreg_event_syscall_enter(struct ftrace_event_call *call)
340 num = ((struct syscall_metadata *)call->data)->syscall_nr;
341 if (num < 0 || num >= NR_syscalls)
343 mutex_lock(&syscall_trace_lock);
344 sys_refcount_enter--;
345 clear_bit(num, enabled_enter_syscalls);
346 if (!sys_refcount_enter)
347 unregister_trace_sys_enter(ftrace_syscall_enter);
348 mutex_unlock(&syscall_trace_lock);
351 int reg_event_syscall_exit(struct ftrace_event_call *call)
356 num = ((struct syscall_metadata *)call->data)->syscall_nr;
357 if (num < 0 || num >= NR_syscalls)
359 mutex_lock(&syscall_trace_lock);
360 if (!sys_refcount_exit)
361 ret = register_trace_sys_exit(ftrace_syscall_exit);
363 set_bit(num, enabled_exit_syscalls);
366 mutex_unlock(&syscall_trace_lock);
370 void unreg_event_syscall_exit(struct ftrace_event_call *call)
374 num = ((struct syscall_metadata *)call->data)->syscall_nr;
375 if (num < 0 || num >= NR_syscalls)
377 mutex_lock(&syscall_trace_lock);
379 clear_bit(num, enabled_exit_syscalls);
380 if (!sys_refcount_exit)
381 unregister_trace_sys_exit(ftrace_syscall_exit);
382 mutex_unlock(&syscall_trace_lock);
385 int init_syscall_trace(struct ftrace_event_call *call)
389 id = register_ftrace_event(call->event);
393 INIT_LIST_HEAD(&call->fields);
397 int __init init_ftrace_syscalls(void)
399 struct syscall_metadata *meta;
403 syscalls_metadata = kzalloc(sizeof(*syscalls_metadata) *
404 NR_syscalls, GFP_KERNEL);
405 if (!syscalls_metadata) {
410 for (i = 0; i < NR_syscalls; i++) {
411 addr = arch_syscall_addr(i);
412 meta = find_syscall_meta(addr);
416 meta->syscall_nr = i;
417 syscalls_metadata[i] = meta;
422 core_initcall(init_ftrace_syscalls);
424 #ifdef CONFIG_EVENT_PROFILE
426 static DECLARE_BITMAP(enabled_prof_enter_syscalls, NR_syscalls);
427 static DECLARE_BITMAP(enabled_prof_exit_syscalls, NR_syscalls);
428 static int sys_prof_refcount_enter;
429 static int sys_prof_refcount_exit;
431 static void prof_syscall_enter(struct pt_regs *regs, long id)
433 struct syscall_metadata *sys_data;
434 struct syscall_trace_enter *rec;
443 syscall_nr = syscall_get_nr(current, regs);
444 if (!test_bit(syscall_nr, enabled_prof_enter_syscalls))
447 sys_data = syscall_nr_to_meta(syscall_nr);
451 /* get the size after alignment with the u32 buffer size field */
452 size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
453 size = ALIGN(size + sizeof(u32), sizeof(u64));
456 if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
457 "profile buffer not large enough"))
460 /* Protect the per cpu buffer, begin the rcu read side */
461 local_irq_save(flags);
463 rctx = perf_swevent_get_recursion_context();
467 cpu = smp_processor_id();
469 trace_buf = rcu_dereference(perf_trace_buf);
474 raw_data = per_cpu_ptr(trace_buf, cpu);
476 /* zero the dead bytes from align to not leak stack to user */
477 *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
479 rec = (struct syscall_trace_enter *) raw_data;
480 tracing_generic_entry_update(&rec->ent, 0, 0);
481 rec->ent.type = sys_data->enter_event->id;
482 rec->nr = syscall_nr;
483 syscall_get_arguments(current, regs, 0, sys_data->nb_args,
484 (unsigned long *)&rec->args);
485 perf_tp_event(sys_data->enter_event->id, 0, 1, rec, size);
488 perf_swevent_put_recursion_context(rctx);
490 local_irq_restore(flags);
493 int prof_sysenter_enable(struct ftrace_event_call *call)
498 num = ((struct syscall_metadata *)call->data)->syscall_nr;
500 mutex_lock(&syscall_trace_lock);
501 if (!sys_prof_refcount_enter)
502 ret = register_trace_sys_enter(prof_syscall_enter);
504 pr_info("event trace: Could not activate"
505 "syscall entry trace point");
507 set_bit(num, enabled_prof_enter_syscalls);
508 sys_prof_refcount_enter++;
510 mutex_unlock(&syscall_trace_lock);
514 void prof_sysenter_disable(struct ftrace_event_call *call)
518 num = ((struct syscall_metadata *)call->data)->syscall_nr;
520 mutex_lock(&syscall_trace_lock);
521 sys_prof_refcount_enter--;
522 clear_bit(num, enabled_prof_enter_syscalls);
523 if (!sys_prof_refcount_enter)
524 unregister_trace_sys_enter(prof_syscall_enter);
525 mutex_unlock(&syscall_trace_lock);
528 static void prof_syscall_exit(struct pt_regs *regs, long ret)
530 struct syscall_metadata *sys_data;
531 struct syscall_trace_exit *rec;
540 syscall_nr = syscall_get_nr(current, regs);
541 if (!test_bit(syscall_nr, enabled_prof_exit_syscalls))
544 sys_data = syscall_nr_to_meta(syscall_nr);
548 /* We can probably do that at build time */
549 size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
553 * Impossible, but be paranoid with the future
554 * How to put this check outside runtime?
556 if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
557 "exit event has grown above profile buffer size"))
560 /* Protect the per cpu buffer, begin the rcu read side */
561 local_irq_save(flags);
563 rctx = perf_swevent_get_recursion_context();
567 cpu = smp_processor_id();
569 trace_buf = rcu_dereference(perf_trace_buf);
574 raw_data = per_cpu_ptr(trace_buf, cpu);
576 /* zero the dead bytes from align to not leak stack to user */
577 *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
579 rec = (struct syscall_trace_exit *)raw_data;
581 tracing_generic_entry_update(&rec->ent, 0, 0);
582 rec->ent.type = sys_data->exit_event->id;
583 rec->nr = syscall_nr;
584 rec->ret = syscall_get_return_value(current, regs);
586 perf_tp_event(sys_data->exit_event->id, 0, 1, rec, size);
589 perf_swevent_put_recursion_context(rctx);
591 local_irq_restore(flags);
594 int prof_sysexit_enable(struct ftrace_event_call *call)
599 num = ((struct syscall_metadata *)call->data)->syscall_nr;
601 mutex_lock(&syscall_trace_lock);
602 if (!sys_prof_refcount_exit)
603 ret = register_trace_sys_exit(prof_syscall_exit);
605 pr_info("event trace: Could not activate"
606 "syscall entry trace point");
608 set_bit(num, enabled_prof_exit_syscalls);
609 sys_prof_refcount_exit++;
611 mutex_unlock(&syscall_trace_lock);
615 void prof_sysexit_disable(struct ftrace_event_call *call)
619 num = ((struct syscall_metadata *)call->data)->syscall_nr;
621 mutex_lock(&syscall_trace_lock);
622 sys_prof_refcount_exit--;
623 clear_bit(num, enabled_prof_exit_syscalls);
624 if (!sys_prof_refcount_exit)
625 unregister_trace_sys_exit(prof_syscall_exit);
626 mutex_unlock(&syscall_trace_lock);