Loading...
1// SPDX-License-Identifier: GPL-2.0
2#include <trace/syscall.h>
3#include <trace/events/syscalls.h>
4#include <linux/syscalls.h>
5#include <linux/slab.h>
6#include <linux/kernel.h>
7#include <linux/module.h> /* for MODULE_NAME_LEN via KSYM_SYMBOL_LEN */
8#include <linux/ftrace.h>
9#include <linux/perf_event.h>
10#include <linux/xarray.h>
11#include <asm/syscall.h>
12
13#include "trace_output.h"
14#include "trace.h"
15
16static DEFINE_MUTEX(syscall_trace_lock);
17
18static int syscall_enter_register(struct trace_event_call *event,
19 enum trace_reg type, void *data);
20static int syscall_exit_register(struct trace_event_call *event,
21 enum trace_reg type, void *data);
22
23static struct list_head *
24syscall_get_enter_fields(struct trace_event_call *call)
25{
26 struct syscall_metadata *entry = call->data;
27
28 return &entry->enter_fields;
29}
30
31extern struct syscall_metadata *__start_syscalls_metadata[];
32extern struct syscall_metadata *__stop_syscalls_metadata[];
33
34static DEFINE_XARRAY(syscalls_metadata_sparse);
35static struct syscall_metadata **syscalls_metadata;
36
37#ifndef ARCH_HAS_SYSCALL_MATCH_SYM_NAME
38static inline bool arch_syscall_match_sym_name(const char *sym, const char *name)
39{
40 /*
41 * Only compare after the "sys" prefix. Archs that use
42 * syscall wrappers may have syscalls symbols aliases prefixed
43 * with ".SyS" or ".sys" instead of "sys", leading to an unwanted
44 * mismatch.
45 */
46 return !strcmp(sym + 3, name + 3);
47}
48#endif
49
50#ifdef ARCH_TRACE_IGNORE_COMPAT_SYSCALLS
51/*
52 * Some architectures that allow for 32bit applications
53 * to run on a 64bit kernel, do not map the syscalls for
54 * the 32bit tasks the same as they do for 64bit tasks.
55 *
56 * *cough*x86*cough*
57 *
58 * In such a case, instead of reporting the wrong syscalls,
59 * simply ignore them.
60 *
61 * For an arch to ignore the compat syscalls it needs to
62 * define ARCH_TRACE_IGNORE_COMPAT_SYSCALLS as well as
63 * define the function arch_trace_is_compat_syscall() to let
64 * the tracing system know that it should ignore it.
65 */
66static int
67trace_get_syscall_nr(struct task_struct *task, struct pt_regs *regs)
68{
69 if (unlikely(arch_trace_is_compat_syscall(regs)))
70 return -1;
71
72 return syscall_get_nr(task, regs);
73}
74#else
75static inline int
76trace_get_syscall_nr(struct task_struct *task, struct pt_regs *regs)
77{
78 return syscall_get_nr(task, regs);
79}
80#endif /* ARCH_TRACE_IGNORE_COMPAT_SYSCALLS */
81
82static __init struct syscall_metadata *
83find_syscall_meta(unsigned long syscall)
84{
85 struct syscall_metadata **start;
86 struct syscall_metadata **stop;
87 char str[KSYM_SYMBOL_LEN];
88
89
90 start = __start_syscalls_metadata;
91 stop = __stop_syscalls_metadata;
92 kallsyms_lookup(syscall, NULL, NULL, NULL, str);
93
94 if (arch_syscall_match_sym_name(str, "sys_ni_syscall"))
95 return NULL;
96
97 for ( ; start < stop; start++) {
98 if ((*start)->name && arch_syscall_match_sym_name(str, (*start)->name))
99 return *start;
100 }
101 return NULL;
102}
103
104static struct syscall_metadata *syscall_nr_to_meta(int nr)
105{
106 if (IS_ENABLED(CONFIG_HAVE_SPARSE_SYSCALL_NR))
107 return xa_load(&syscalls_metadata_sparse, (unsigned long)nr);
108
109 if (!syscalls_metadata || nr >= NR_syscalls || nr < 0)
110 return NULL;
111
112 return syscalls_metadata[nr];
113}
114
115const char *get_syscall_name(int syscall)
116{
117 struct syscall_metadata *entry;
118
119 entry = syscall_nr_to_meta(syscall);
120 if (!entry)
121 return NULL;
122
123 return entry->name;
124}
125
126static enum print_line_t
127print_syscall_enter(struct trace_iterator *iter, int flags,
128 struct trace_event *event)
129{
130 struct trace_array *tr = iter->tr;
131 struct trace_seq *s = &iter->seq;
132 struct trace_entry *ent = iter->ent;
133 struct syscall_trace_enter *trace;
134 struct syscall_metadata *entry;
135 int i, syscall;
136
137 trace = (typeof(trace))ent;
138 syscall = trace->nr;
139 entry = syscall_nr_to_meta(syscall);
140
141 if (!entry)
142 goto end;
143
144 if (entry->enter_event->event.type != ent->type) {
145 WARN_ON_ONCE(1);
146 goto end;
147 }
148
149 trace_seq_printf(s, "%s(", entry->name);
150
151 for (i = 0; i < entry->nb_args; i++) {
152
153 if (trace_seq_has_overflowed(s))
154 goto end;
155
156 /* parameter types */
157 if (tr && tr->trace_flags & TRACE_ITER_VERBOSE)
158 trace_seq_printf(s, "%s ", entry->types[i]);
159
160 /* parameter values */
161 trace_seq_printf(s, "%s: %lx%s", entry->args[i],
162 trace->args[i],
163 i == entry->nb_args - 1 ? "" : ", ");
164 }
165
166 trace_seq_putc(s, ')');
167end:
168 trace_seq_putc(s, '\n');
169
170 return trace_handle_return(s);
171}
172
173static enum print_line_t
174print_syscall_exit(struct trace_iterator *iter, int flags,
175 struct trace_event *event)
176{
177 struct trace_seq *s = &iter->seq;
178 struct trace_entry *ent = iter->ent;
179 struct syscall_trace_exit *trace;
180 int syscall;
181 struct syscall_metadata *entry;
182
183 trace = (typeof(trace))ent;
184 syscall = trace->nr;
185 entry = syscall_nr_to_meta(syscall);
186
187 if (!entry) {
188 trace_seq_putc(s, '\n');
189 goto out;
190 }
191
192 if (entry->exit_event->event.type != ent->type) {
193 WARN_ON_ONCE(1);
194 return TRACE_TYPE_UNHANDLED;
195 }
196
197 trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
198 trace->ret);
199
200 out:
201 return trace_handle_return(s);
202}
203
204#define SYSCALL_FIELD(_type, _name) { \
205 .type = #_type, .name = #_name, \
206 .size = sizeof(_type), .align = __alignof__(_type), \
207 .is_signed = is_signed_type(_type), .filter_type = FILTER_OTHER }
208
209static int __init
210__set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len)
211{
212 int i;
213 int pos = 0;
214
215 /* When len=0, we just calculate the needed length */
216#define LEN_OR_ZERO (len ? len - pos : 0)
217
218 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
219 for (i = 0; i < entry->nb_args; i++) {
220 pos += snprintf(buf + pos, LEN_OR_ZERO, "%s: 0x%%0%zulx%s",
221 entry->args[i], sizeof(unsigned long),
222 i == entry->nb_args - 1 ? "" : ", ");
223 }
224 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
225
226 for (i = 0; i < entry->nb_args; i++) {
227 pos += snprintf(buf + pos, LEN_OR_ZERO,
228 ", ((unsigned long)(REC->%s))", entry->args[i]);
229 }
230
231#undef LEN_OR_ZERO
232
233 /* return the length of print_fmt */
234 return pos;
235}
236
237static int __init set_syscall_print_fmt(struct trace_event_call *call)
238{
239 char *print_fmt;
240 int len;
241 struct syscall_metadata *entry = call->data;
242
243 if (entry->enter_event != call) {
244 call->print_fmt = "\"0x%lx\", REC->ret";
245 return 0;
246 }
247
248 /* First: called with 0 length to calculate the needed length */
249 len = __set_enter_print_fmt(entry, NULL, 0);
250
251 print_fmt = kmalloc(len + 1, GFP_KERNEL);
252 if (!print_fmt)
253 return -ENOMEM;
254
255 /* Second: actually write the @print_fmt */
256 __set_enter_print_fmt(entry, print_fmt, len + 1);
257 call->print_fmt = print_fmt;
258
259 return 0;
260}
261
262static void __init free_syscall_print_fmt(struct trace_event_call *call)
263{
264 struct syscall_metadata *entry = call->data;
265
266 if (entry->enter_event == call)
267 kfree(call->print_fmt);
268}
269
270static int __init syscall_enter_define_fields(struct trace_event_call *call)
271{
272 struct syscall_trace_enter trace;
273 struct syscall_metadata *meta = call->data;
274 int offset = offsetof(typeof(trace), args);
275 int ret = 0;
276 int i;
277
278 for (i = 0; i < meta->nb_args; i++) {
279 ret = trace_define_field(call, meta->types[i],
280 meta->args[i], offset,
281 sizeof(unsigned long), 0,
282 FILTER_OTHER);
283 if (ret)
284 break;
285 offset += sizeof(unsigned long);
286 }
287
288 return ret;
289}
290
291static void ftrace_syscall_enter(void *data, struct pt_regs *regs, long id)
292{
293 struct trace_array *tr = data;
294 struct trace_event_file *trace_file;
295 struct syscall_trace_enter *entry;
296 struct syscall_metadata *sys_data;
297 struct trace_event_buffer fbuffer;
298 unsigned long args[6];
299 int syscall_nr;
300 int size;
301
302 syscall_nr = trace_get_syscall_nr(current, regs);
303 if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
304 return;
305
306 /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE) */
307 trace_file = rcu_dereference_sched(tr->enter_syscall_files[syscall_nr]);
308 if (!trace_file)
309 return;
310
311 if (trace_trigger_soft_disabled(trace_file))
312 return;
313
314 sys_data = syscall_nr_to_meta(syscall_nr);
315 if (!sys_data)
316 return;
317
318 size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
319
320 entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
321 if (!entry)
322 return;
323
324 entry = ring_buffer_event_data(fbuffer.event);
325 entry->nr = syscall_nr;
326 syscall_get_arguments(current, regs, args);
327 memcpy(entry->args, args, sizeof(unsigned long) * sys_data->nb_args);
328
329 trace_event_buffer_commit(&fbuffer);
330}
331
332static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
333{
334 struct trace_array *tr = data;
335 struct trace_event_file *trace_file;
336 struct syscall_trace_exit *entry;
337 struct syscall_metadata *sys_data;
338 struct trace_event_buffer fbuffer;
339 int syscall_nr;
340
341 syscall_nr = trace_get_syscall_nr(current, regs);
342 if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
343 return;
344
345 /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE()) */
346 trace_file = rcu_dereference_sched(tr->exit_syscall_files[syscall_nr]);
347 if (!trace_file)
348 return;
349
350 if (trace_trigger_soft_disabled(trace_file))
351 return;
352
353 sys_data = syscall_nr_to_meta(syscall_nr);
354 if (!sys_data)
355 return;
356
357 entry = trace_event_buffer_reserve(&fbuffer, trace_file, sizeof(*entry));
358 if (!entry)
359 return;
360
361 entry = ring_buffer_event_data(fbuffer.event);
362 entry->nr = syscall_nr;
363 entry->ret = syscall_get_return_value(current, regs);
364
365 trace_event_buffer_commit(&fbuffer);
366}
367
368static int reg_event_syscall_enter(struct trace_event_file *file,
369 struct trace_event_call *call)
370{
371 struct trace_array *tr = file->tr;
372 int ret = 0;
373 int num;
374
375 num = ((struct syscall_metadata *)call->data)->syscall_nr;
376 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
377 return -ENOSYS;
378 mutex_lock(&syscall_trace_lock);
379 if (!tr->sys_refcount_enter)
380 ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
381 if (!ret) {
382 rcu_assign_pointer(tr->enter_syscall_files[num], file);
383 tr->sys_refcount_enter++;
384 }
385 mutex_unlock(&syscall_trace_lock);
386 return ret;
387}
388
389static void unreg_event_syscall_enter(struct trace_event_file *file,
390 struct trace_event_call *call)
391{
392 struct trace_array *tr = file->tr;
393 int num;
394
395 num = ((struct syscall_metadata *)call->data)->syscall_nr;
396 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
397 return;
398 mutex_lock(&syscall_trace_lock);
399 tr->sys_refcount_enter--;
400 RCU_INIT_POINTER(tr->enter_syscall_files[num], NULL);
401 if (!tr->sys_refcount_enter)
402 unregister_trace_sys_enter(ftrace_syscall_enter, tr);
403 mutex_unlock(&syscall_trace_lock);
404}
405
406static int reg_event_syscall_exit(struct trace_event_file *file,
407 struct trace_event_call *call)
408{
409 struct trace_array *tr = file->tr;
410 int ret = 0;
411 int num;
412
413 num = ((struct syscall_metadata *)call->data)->syscall_nr;
414 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
415 return -ENOSYS;
416 mutex_lock(&syscall_trace_lock);
417 if (!tr->sys_refcount_exit)
418 ret = register_trace_sys_exit(ftrace_syscall_exit, tr);
419 if (!ret) {
420 rcu_assign_pointer(tr->exit_syscall_files[num], file);
421 tr->sys_refcount_exit++;
422 }
423 mutex_unlock(&syscall_trace_lock);
424 return ret;
425}
426
427static void unreg_event_syscall_exit(struct trace_event_file *file,
428 struct trace_event_call *call)
429{
430 struct trace_array *tr = file->tr;
431 int num;
432
433 num = ((struct syscall_metadata *)call->data)->syscall_nr;
434 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
435 return;
436 mutex_lock(&syscall_trace_lock);
437 tr->sys_refcount_exit--;
438 RCU_INIT_POINTER(tr->exit_syscall_files[num], NULL);
439 if (!tr->sys_refcount_exit)
440 unregister_trace_sys_exit(ftrace_syscall_exit, tr);
441 mutex_unlock(&syscall_trace_lock);
442}
443
444static int __init init_syscall_trace(struct trace_event_call *call)
445{
446 int id;
447 int num;
448
449 num = ((struct syscall_metadata *)call->data)->syscall_nr;
450 if (num < 0 || num >= NR_syscalls) {
451 pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
452 ((struct syscall_metadata *)call->data)->name);
453 return -ENOSYS;
454 }
455
456 if (set_syscall_print_fmt(call) < 0)
457 return -ENOMEM;
458
459 id = trace_event_raw_init(call);
460
461 if (id < 0) {
462 free_syscall_print_fmt(call);
463 return id;
464 }
465
466 return id;
467}
468
469static struct trace_event_fields __refdata syscall_enter_fields_array[] = {
470 SYSCALL_FIELD(int, __syscall_nr),
471 { .type = TRACE_FUNCTION_TYPE,
472 .define_fields = syscall_enter_define_fields },
473 {}
474};
475
476struct trace_event_functions enter_syscall_print_funcs = {
477 .trace = print_syscall_enter,
478};
479
480struct trace_event_functions exit_syscall_print_funcs = {
481 .trace = print_syscall_exit,
482};
483
484struct trace_event_class __refdata event_class_syscall_enter = {
485 .system = "syscalls",
486 .reg = syscall_enter_register,
487 .fields_array = syscall_enter_fields_array,
488 .get_fields = syscall_get_enter_fields,
489 .raw_init = init_syscall_trace,
490};
491
492struct trace_event_class __refdata event_class_syscall_exit = {
493 .system = "syscalls",
494 .reg = syscall_exit_register,
495 .fields_array = (struct trace_event_fields[]){
496 SYSCALL_FIELD(int, __syscall_nr),
497 SYSCALL_FIELD(long, ret),
498 {}
499 },
500 .fields = LIST_HEAD_INIT(event_class_syscall_exit.fields),
501 .raw_init = init_syscall_trace,
502};
503
504unsigned long __init __weak arch_syscall_addr(int nr)
505{
506 return (unsigned long)sys_call_table[nr];
507}
508
509void __init init_ftrace_syscalls(void)
510{
511 struct syscall_metadata *meta;
512 unsigned long addr;
513 int i;
514 void *ret;
515
516 if (!IS_ENABLED(CONFIG_HAVE_SPARSE_SYSCALL_NR)) {
517 syscalls_metadata = kcalloc(NR_syscalls,
518 sizeof(*syscalls_metadata),
519 GFP_KERNEL);
520 if (!syscalls_metadata) {
521 WARN_ON(1);
522 return;
523 }
524 }
525
526 for (i = 0; i < NR_syscalls; i++) {
527 addr = arch_syscall_addr(i);
528 meta = find_syscall_meta(addr);
529 if (!meta)
530 continue;
531
532 meta->syscall_nr = i;
533
534 if (!IS_ENABLED(CONFIG_HAVE_SPARSE_SYSCALL_NR)) {
535 syscalls_metadata[i] = meta;
536 } else {
537 ret = xa_store(&syscalls_metadata_sparse, i, meta,
538 GFP_KERNEL);
539 WARN(xa_is_err(ret),
540 "Syscall memory allocation failed\n");
541 }
542
543 }
544}
545
546#ifdef CONFIG_PERF_EVENTS
547
548static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
549static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
550static int sys_perf_refcount_enter;
551static int sys_perf_refcount_exit;
552
553static int perf_call_bpf_enter(struct trace_event_call *call, struct pt_regs *regs,
554 struct syscall_metadata *sys_data,
555 struct syscall_trace_enter *rec)
556{
557 struct syscall_tp_t {
558 unsigned long long regs;
559 unsigned long syscall_nr;
560 unsigned long args[SYSCALL_DEFINE_MAXARGS];
561 } param;
562 int i;
563
564 *(struct pt_regs **)¶m = regs;
565 param.syscall_nr = rec->nr;
566 for (i = 0; i < sys_data->nb_args; i++)
567 param.args[i] = rec->args[i];
568 return trace_call_bpf(call, ¶m);
569}
570
571static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
572{
573 struct syscall_metadata *sys_data;
574 struct syscall_trace_enter *rec;
575 struct hlist_head *head;
576 unsigned long args[6];
577 bool valid_prog_array;
578 int syscall_nr;
579 int rctx;
580 int size;
581
582 syscall_nr = trace_get_syscall_nr(current, regs);
583 if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
584 return;
585 if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
586 return;
587
588 sys_data = syscall_nr_to_meta(syscall_nr);
589 if (!sys_data)
590 return;
591
592 head = this_cpu_ptr(sys_data->enter_event->perf_events);
593 valid_prog_array = bpf_prog_array_valid(sys_data->enter_event);
594 if (!valid_prog_array && hlist_empty(head))
595 return;
596
597 /* get the size after alignment with the u32 buffer size field */
598 size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
599 size = ALIGN(size + sizeof(u32), sizeof(u64));
600 size -= sizeof(u32);
601
602 rec = perf_trace_buf_alloc(size, NULL, &rctx);
603 if (!rec)
604 return;
605
606 rec->nr = syscall_nr;
607 syscall_get_arguments(current, regs, args);
608 memcpy(&rec->args, args, sizeof(unsigned long) * sys_data->nb_args);
609
610 if ((valid_prog_array &&
611 !perf_call_bpf_enter(sys_data->enter_event, regs, sys_data, rec)) ||
612 hlist_empty(head)) {
613 perf_swevent_put_recursion_context(rctx);
614 return;
615 }
616
617 perf_trace_buf_submit(rec, size, rctx,
618 sys_data->enter_event->event.type, 1, regs,
619 head, NULL);
620}
621
622static int perf_sysenter_enable(struct trace_event_call *call)
623{
624 int ret = 0;
625 int num;
626
627 num = ((struct syscall_metadata *)call->data)->syscall_nr;
628
629 mutex_lock(&syscall_trace_lock);
630 if (!sys_perf_refcount_enter)
631 ret = register_trace_sys_enter(perf_syscall_enter, NULL);
632 if (ret) {
633 pr_info("event trace: Could not activate syscall entry trace point");
634 } else {
635 set_bit(num, enabled_perf_enter_syscalls);
636 sys_perf_refcount_enter++;
637 }
638 mutex_unlock(&syscall_trace_lock);
639 return ret;
640}
641
642static void perf_sysenter_disable(struct trace_event_call *call)
643{
644 int num;
645
646 num = ((struct syscall_metadata *)call->data)->syscall_nr;
647
648 mutex_lock(&syscall_trace_lock);
649 sys_perf_refcount_enter--;
650 clear_bit(num, enabled_perf_enter_syscalls);
651 if (!sys_perf_refcount_enter)
652 unregister_trace_sys_enter(perf_syscall_enter, NULL);
653 mutex_unlock(&syscall_trace_lock);
654}
655
656static int perf_call_bpf_exit(struct trace_event_call *call, struct pt_regs *regs,
657 struct syscall_trace_exit *rec)
658{
659 struct syscall_tp_t {
660 unsigned long long regs;
661 unsigned long syscall_nr;
662 unsigned long ret;
663 } param;
664
665 *(struct pt_regs **)¶m = regs;
666 param.syscall_nr = rec->nr;
667 param.ret = rec->ret;
668 return trace_call_bpf(call, ¶m);
669}
670
671static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
672{
673 struct syscall_metadata *sys_data;
674 struct syscall_trace_exit *rec;
675 struct hlist_head *head;
676 bool valid_prog_array;
677 int syscall_nr;
678 int rctx;
679 int size;
680
681 syscall_nr = trace_get_syscall_nr(current, regs);
682 if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
683 return;
684 if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
685 return;
686
687 sys_data = syscall_nr_to_meta(syscall_nr);
688 if (!sys_data)
689 return;
690
691 head = this_cpu_ptr(sys_data->exit_event->perf_events);
692 valid_prog_array = bpf_prog_array_valid(sys_data->exit_event);
693 if (!valid_prog_array && hlist_empty(head))
694 return;
695
696 /* We can probably do that at build time */
697 size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
698 size -= sizeof(u32);
699
700 rec = perf_trace_buf_alloc(size, NULL, &rctx);
701 if (!rec)
702 return;
703
704 rec->nr = syscall_nr;
705 rec->ret = syscall_get_return_value(current, regs);
706
707 if ((valid_prog_array &&
708 !perf_call_bpf_exit(sys_data->exit_event, regs, rec)) ||
709 hlist_empty(head)) {
710 perf_swevent_put_recursion_context(rctx);
711 return;
712 }
713
714 perf_trace_buf_submit(rec, size, rctx, sys_data->exit_event->event.type,
715 1, regs, head, NULL);
716}
717
718static int perf_sysexit_enable(struct trace_event_call *call)
719{
720 int ret = 0;
721 int num;
722
723 num = ((struct syscall_metadata *)call->data)->syscall_nr;
724
725 mutex_lock(&syscall_trace_lock);
726 if (!sys_perf_refcount_exit)
727 ret = register_trace_sys_exit(perf_syscall_exit, NULL);
728 if (ret) {
729 pr_info("event trace: Could not activate syscall exit trace point");
730 } else {
731 set_bit(num, enabled_perf_exit_syscalls);
732 sys_perf_refcount_exit++;
733 }
734 mutex_unlock(&syscall_trace_lock);
735 return ret;
736}
737
738static void perf_sysexit_disable(struct trace_event_call *call)
739{
740 int num;
741
742 num = ((struct syscall_metadata *)call->data)->syscall_nr;
743
744 mutex_lock(&syscall_trace_lock);
745 sys_perf_refcount_exit--;
746 clear_bit(num, enabled_perf_exit_syscalls);
747 if (!sys_perf_refcount_exit)
748 unregister_trace_sys_exit(perf_syscall_exit, NULL);
749 mutex_unlock(&syscall_trace_lock);
750}
751
752#endif /* CONFIG_PERF_EVENTS */
753
754static int syscall_enter_register(struct trace_event_call *event,
755 enum trace_reg type, void *data)
756{
757 struct trace_event_file *file = data;
758
759 switch (type) {
760 case TRACE_REG_REGISTER:
761 return reg_event_syscall_enter(file, event);
762 case TRACE_REG_UNREGISTER:
763 unreg_event_syscall_enter(file, event);
764 return 0;
765
766#ifdef CONFIG_PERF_EVENTS
767 case TRACE_REG_PERF_REGISTER:
768 return perf_sysenter_enable(event);
769 case TRACE_REG_PERF_UNREGISTER:
770 perf_sysenter_disable(event);
771 return 0;
772 case TRACE_REG_PERF_OPEN:
773 case TRACE_REG_PERF_CLOSE:
774 case TRACE_REG_PERF_ADD:
775 case TRACE_REG_PERF_DEL:
776 return 0;
777#endif
778 }
779 return 0;
780}
781
782static int syscall_exit_register(struct trace_event_call *event,
783 enum trace_reg type, void *data)
784{
785 struct trace_event_file *file = data;
786
787 switch (type) {
788 case TRACE_REG_REGISTER:
789 return reg_event_syscall_exit(file, event);
790 case TRACE_REG_UNREGISTER:
791 unreg_event_syscall_exit(file, event);
792 return 0;
793
794#ifdef CONFIG_PERF_EVENTS
795 case TRACE_REG_PERF_REGISTER:
796 return perf_sysexit_enable(event);
797 case TRACE_REG_PERF_UNREGISTER:
798 perf_sysexit_disable(event);
799 return 0;
800 case TRACE_REG_PERF_OPEN:
801 case TRACE_REG_PERF_CLOSE:
802 case TRACE_REG_PERF_ADD:
803 case TRACE_REG_PERF_DEL:
804 return 0;
805#endif
806 }
807 return 0;
808}
1#include <trace/syscall.h>
2#include <trace/events/syscalls.h>
3#include <linux/syscalls.h>
4#include <linux/slab.h>
5#include <linux/kernel.h>
6#include <linux/module.h> /* for MODULE_NAME_LEN via KSYM_SYMBOL_LEN */
7#include <linux/ftrace.h>
8#include <linux/perf_event.h>
9#include <asm/syscall.h>
10
11#include "trace_output.h"
12#include "trace.h"
13
14static DEFINE_MUTEX(syscall_trace_lock);
15
16static int syscall_enter_register(struct ftrace_event_call *event,
17 enum trace_reg type, void *data);
18static int syscall_exit_register(struct ftrace_event_call *event,
19 enum trace_reg type, void *data);
20
21static struct list_head *
22syscall_get_enter_fields(struct ftrace_event_call *call)
23{
24 struct syscall_metadata *entry = call->data;
25
26 return &entry->enter_fields;
27}
28
29extern struct syscall_metadata *__start_syscalls_metadata[];
30extern struct syscall_metadata *__stop_syscalls_metadata[];
31
32static struct syscall_metadata **syscalls_metadata;
33
34#ifndef ARCH_HAS_SYSCALL_MATCH_SYM_NAME
35static inline bool arch_syscall_match_sym_name(const char *sym, const char *name)
36{
37 /*
38 * Only compare after the "sys" prefix. Archs that use
39 * syscall wrappers may have syscalls symbols aliases prefixed
40 * with ".SyS" or ".sys" instead of "sys", leading to an unwanted
41 * mismatch.
42 */
43 return !strcmp(sym + 3, name + 3);
44}
45#endif
46
47#ifdef ARCH_TRACE_IGNORE_COMPAT_SYSCALLS
48/*
49 * Some architectures that allow for 32bit applications
50 * to run on a 64bit kernel, do not map the syscalls for
51 * the 32bit tasks the same as they do for 64bit tasks.
52 *
53 * *cough*x86*cough*
54 *
55 * In such a case, instead of reporting the wrong syscalls,
56 * simply ignore them.
57 *
58 * For an arch to ignore the compat syscalls it needs to
59 * define ARCH_TRACE_IGNORE_COMPAT_SYSCALLS as well as
60 * define the function arch_trace_is_compat_syscall() to let
61 * the tracing system know that it should ignore it.
62 */
63static int
64trace_get_syscall_nr(struct task_struct *task, struct pt_regs *regs)
65{
66 if (unlikely(arch_trace_is_compat_syscall(regs)))
67 return -1;
68
69 return syscall_get_nr(task, regs);
70}
71#else
72static inline int
73trace_get_syscall_nr(struct task_struct *task, struct pt_regs *regs)
74{
75 return syscall_get_nr(task, regs);
76}
77#endif /* ARCH_TRACE_IGNORE_COMPAT_SYSCALLS */
78
79static __init struct syscall_metadata *
80find_syscall_meta(unsigned long syscall)
81{
82 struct syscall_metadata **start;
83 struct syscall_metadata **stop;
84 char str[KSYM_SYMBOL_LEN];
85
86
87 start = __start_syscalls_metadata;
88 stop = __stop_syscalls_metadata;
89 kallsyms_lookup(syscall, NULL, NULL, NULL, str);
90
91 if (arch_syscall_match_sym_name(str, "sys_ni_syscall"))
92 return NULL;
93
94 for ( ; start < stop; start++) {
95 if ((*start)->name && arch_syscall_match_sym_name(str, (*start)->name))
96 return *start;
97 }
98 return NULL;
99}
100
101static struct syscall_metadata *syscall_nr_to_meta(int nr)
102{
103 if (!syscalls_metadata || nr >= NR_syscalls || nr < 0)
104 return NULL;
105
106 return syscalls_metadata[nr];
107}
108
109static enum print_line_t
110print_syscall_enter(struct trace_iterator *iter, int flags,
111 struct trace_event *event)
112{
113 struct trace_seq *s = &iter->seq;
114 struct trace_entry *ent = iter->ent;
115 struct syscall_trace_enter *trace;
116 struct syscall_metadata *entry;
117 int i, ret, syscall;
118
119 trace = (typeof(trace))ent;
120 syscall = trace->nr;
121 entry = syscall_nr_to_meta(syscall);
122
123 if (!entry)
124 goto end;
125
126 if (entry->enter_event->event.type != ent->type) {
127 WARN_ON_ONCE(1);
128 goto end;
129 }
130
131 ret = trace_seq_printf(s, "%s(", entry->name);
132 if (!ret)
133 return TRACE_TYPE_PARTIAL_LINE;
134
135 for (i = 0; i < entry->nb_args; i++) {
136 /* parameter types */
137 if (trace_flags & TRACE_ITER_VERBOSE) {
138 ret = trace_seq_printf(s, "%s ", entry->types[i]);
139 if (!ret)
140 return TRACE_TYPE_PARTIAL_LINE;
141 }
142 /* parameter values */
143 ret = trace_seq_printf(s, "%s: %lx%s", entry->args[i],
144 trace->args[i],
145 i == entry->nb_args - 1 ? "" : ", ");
146 if (!ret)
147 return TRACE_TYPE_PARTIAL_LINE;
148 }
149
150 ret = trace_seq_putc(s, ')');
151 if (!ret)
152 return TRACE_TYPE_PARTIAL_LINE;
153
154end:
155 ret = trace_seq_putc(s, '\n');
156 if (!ret)
157 return TRACE_TYPE_PARTIAL_LINE;
158
159 return TRACE_TYPE_HANDLED;
160}
161
162static enum print_line_t
163print_syscall_exit(struct trace_iterator *iter, int flags,
164 struct trace_event *event)
165{
166 struct trace_seq *s = &iter->seq;
167 struct trace_entry *ent = iter->ent;
168 struct syscall_trace_exit *trace;
169 int syscall;
170 struct syscall_metadata *entry;
171 int ret;
172
173 trace = (typeof(trace))ent;
174 syscall = trace->nr;
175 entry = syscall_nr_to_meta(syscall);
176
177 if (!entry) {
178 trace_seq_putc(s, '\n');
179 return TRACE_TYPE_HANDLED;
180 }
181
182 if (entry->exit_event->event.type != ent->type) {
183 WARN_ON_ONCE(1);
184 return TRACE_TYPE_UNHANDLED;
185 }
186
187 ret = trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
188 trace->ret);
189 if (!ret)
190 return TRACE_TYPE_PARTIAL_LINE;
191
192 return TRACE_TYPE_HANDLED;
193}
194
195extern char *__bad_type_size(void);
196
197#define SYSCALL_FIELD(type, name) \
198 sizeof(type) != sizeof(trace.name) ? \
199 __bad_type_size() : \
200 #type, #name, offsetof(typeof(trace), name), \
201 sizeof(trace.name), is_signed_type(type)
202
203static int __init
204__set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len)
205{
206 int i;
207 int pos = 0;
208
209 /* When len=0, we just calculate the needed length */
210#define LEN_OR_ZERO (len ? len - pos : 0)
211
212 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
213 for (i = 0; i < entry->nb_args; i++) {
214 pos += snprintf(buf + pos, LEN_OR_ZERO, "%s: 0x%%0%zulx%s",
215 entry->args[i], sizeof(unsigned long),
216 i == entry->nb_args - 1 ? "" : ", ");
217 }
218 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
219
220 for (i = 0; i < entry->nb_args; i++) {
221 pos += snprintf(buf + pos, LEN_OR_ZERO,
222 ", ((unsigned long)(REC->%s))", entry->args[i]);
223 }
224
225#undef LEN_OR_ZERO
226
227 /* return the length of print_fmt */
228 return pos;
229}
230
231static int __init set_syscall_print_fmt(struct ftrace_event_call *call)
232{
233 char *print_fmt;
234 int len;
235 struct syscall_metadata *entry = call->data;
236
237 if (entry->enter_event != call) {
238 call->print_fmt = "\"0x%lx\", REC->ret";
239 return 0;
240 }
241
242 /* First: called with 0 length to calculate the needed length */
243 len = __set_enter_print_fmt(entry, NULL, 0);
244
245 print_fmt = kmalloc(len + 1, GFP_KERNEL);
246 if (!print_fmt)
247 return -ENOMEM;
248
249 /* Second: actually write the @print_fmt */
250 __set_enter_print_fmt(entry, print_fmt, len + 1);
251 call->print_fmt = print_fmt;
252
253 return 0;
254}
255
256static void __init free_syscall_print_fmt(struct ftrace_event_call *call)
257{
258 struct syscall_metadata *entry = call->data;
259
260 if (entry->enter_event == call)
261 kfree(call->print_fmt);
262}
263
264static int __init syscall_enter_define_fields(struct ftrace_event_call *call)
265{
266 struct syscall_trace_enter trace;
267 struct syscall_metadata *meta = call->data;
268 int ret;
269 int i;
270 int offset = offsetof(typeof(trace), args);
271
272 ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
273 if (ret)
274 return ret;
275
276 for (i = 0; i < meta->nb_args; i++) {
277 ret = trace_define_field(call, meta->types[i],
278 meta->args[i], offset,
279 sizeof(unsigned long), 0,
280 FILTER_OTHER);
281 offset += sizeof(unsigned long);
282 }
283
284 return ret;
285}
286
287static int __init syscall_exit_define_fields(struct ftrace_event_call *call)
288{
289 struct syscall_trace_exit trace;
290 int ret;
291
292 ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
293 if (ret)
294 return ret;
295
296 ret = trace_define_field(call, SYSCALL_FIELD(long, ret),
297 FILTER_OTHER);
298
299 return ret;
300}
301
302static void ftrace_syscall_enter(void *data, struct pt_regs *regs, long id)
303{
304 struct trace_array *tr = data;
305 struct ftrace_event_file *ftrace_file;
306 struct syscall_trace_enter *entry;
307 struct syscall_metadata *sys_data;
308 struct ring_buffer_event *event;
309 struct ring_buffer *buffer;
310 unsigned long irq_flags;
311 int pc;
312 int syscall_nr;
313 int size;
314
315 syscall_nr = trace_get_syscall_nr(current, regs);
316 if (syscall_nr < 0)
317 return;
318
319 /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE) */
320 ftrace_file = rcu_dereference_sched(tr->enter_syscall_files[syscall_nr]);
321 if (!ftrace_file)
322 return;
323
324 if (ftrace_trigger_soft_disabled(ftrace_file))
325 return;
326
327 sys_data = syscall_nr_to_meta(syscall_nr);
328 if (!sys_data)
329 return;
330
331 size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
332
333 local_save_flags(irq_flags);
334 pc = preempt_count();
335
336 buffer = tr->trace_buffer.buffer;
337 event = trace_buffer_lock_reserve(buffer,
338 sys_data->enter_event->event.type, size, irq_flags, pc);
339 if (!event)
340 return;
341
342 entry = ring_buffer_event_data(event);
343 entry->nr = syscall_nr;
344 syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
345
346 event_trigger_unlock_commit(ftrace_file, buffer, event, entry,
347 irq_flags, pc);
348}
349
350static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
351{
352 struct trace_array *tr = data;
353 struct ftrace_event_file *ftrace_file;
354 struct syscall_trace_exit *entry;
355 struct syscall_metadata *sys_data;
356 struct ring_buffer_event *event;
357 struct ring_buffer *buffer;
358 unsigned long irq_flags;
359 int pc;
360 int syscall_nr;
361
362 syscall_nr = trace_get_syscall_nr(current, regs);
363 if (syscall_nr < 0)
364 return;
365
366 /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE()) */
367 ftrace_file = rcu_dereference_sched(tr->exit_syscall_files[syscall_nr]);
368 if (!ftrace_file)
369 return;
370
371 if (ftrace_trigger_soft_disabled(ftrace_file))
372 return;
373
374 sys_data = syscall_nr_to_meta(syscall_nr);
375 if (!sys_data)
376 return;
377
378 local_save_flags(irq_flags);
379 pc = preempt_count();
380
381 buffer = tr->trace_buffer.buffer;
382 event = trace_buffer_lock_reserve(buffer,
383 sys_data->exit_event->event.type, sizeof(*entry),
384 irq_flags, pc);
385 if (!event)
386 return;
387
388 entry = ring_buffer_event_data(event);
389 entry->nr = syscall_nr;
390 entry->ret = syscall_get_return_value(current, regs);
391
392 event_trigger_unlock_commit(ftrace_file, buffer, event, entry,
393 irq_flags, pc);
394}
395
396static int reg_event_syscall_enter(struct ftrace_event_file *file,
397 struct ftrace_event_call *call)
398{
399 struct trace_array *tr = file->tr;
400 int ret = 0;
401 int num;
402
403 num = ((struct syscall_metadata *)call->data)->syscall_nr;
404 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
405 return -ENOSYS;
406 mutex_lock(&syscall_trace_lock);
407 if (!tr->sys_refcount_enter)
408 ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
409 if (!ret) {
410 rcu_assign_pointer(tr->enter_syscall_files[num], file);
411 tr->sys_refcount_enter++;
412 }
413 mutex_unlock(&syscall_trace_lock);
414 return ret;
415}
416
417static void unreg_event_syscall_enter(struct ftrace_event_file *file,
418 struct ftrace_event_call *call)
419{
420 struct trace_array *tr = file->tr;
421 int num;
422
423 num = ((struct syscall_metadata *)call->data)->syscall_nr;
424 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
425 return;
426 mutex_lock(&syscall_trace_lock);
427 tr->sys_refcount_enter--;
428 rcu_assign_pointer(tr->enter_syscall_files[num], NULL);
429 if (!tr->sys_refcount_enter)
430 unregister_trace_sys_enter(ftrace_syscall_enter, tr);
431 mutex_unlock(&syscall_trace_lock);
432}
433
434static int reg_event_syscall_exit(struct ftrace_event_file *file,
435 struct ftrace_event_call *call)
436{
437 struct trace_array *tr = file->tr;
438 int ret = 0;
439 int num;
440
441 num = ((struct syscall_metadata *)call->data)->syscall_nr;
442 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
443 return -ENOSYS;
444 mutex_lock(&syscall_trace_lock);
445 if (!tr->sys_refcount_exit)
446 ret = register_trace_sys_exit(ftrace_syscall_exit, tr);
447 if (!ret) {
448 rcu_assign_pointer(tr->exit_syscall_files[num], file);
449 tr->sys_refcount_exit++;
450 }
451 mutex_unlock(&syscall_trace_lock);
452 return ret;
453}
454
455static void unreg_event_syscall_exit(struct ftrace_event_file *file,
456 struct ftrace_event_call *call)
457{
458 struct trace_array *tr = file->tr;
459 int num;
460
461 num = ((struct syscall_metadata *)call->data)->syscall_nr;
462 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
463 return;
464 mutex_lock(&syscall_trace_lock);
465 tr->sys_refcount_exit--;
466 rcu_assign_pointer(tr->exit_syscall_files[num], NULL);
467 if (!tr->sys_refcount_exit)
468 unregister_trace_sys_exit(ftrace_syscall_exit, tr);
469 mutex_unlock(&syscall_trace_lock);
470}
471
472static int __init init_syscall_trace(struct ftrace_event_call *call)
473{
474 int id;
475 int num;
476
477 num = ((struct syscall_metadata *)call->data)->syscall_nr;
478 if (num < 0 || num >= NR_syscalls) {
479 pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
480 ((struct syscall_metadata *)call->data)->name);
481 return -ENOSYS;
482 }
483
484 if (set_syscall_print_fmt(call) < 0)
485 return -ENOMEM;
486
487 id = trace_event_raw_init(call);
488
489 if (id < 0) {
490 free_syscall_print_fmt(call);
491 return id;
492 }
493
494 return id;
495}
496
497struct trace_event_functions enter_syscall_print_funcs = {
498 .trace = print_syscall_enter,
499};
500
501struct trace_event_functions exit_syscall_print_funcs = {
502 .trace = print_syscall_exit,
503};
504
505struct ftrace_event_class __refdata event_class_syscall_enter = {
506 .system = "syscalls",
507 .reg = syscall_enter_register,
508 .define_fields = syscall_enter_define_fields,
509 .get_fields = syscall_get_enter_fields,
510 .raw_init = init_syscall_trace,
511};
512
513struct ftrace_event_class __refdata event_class_syscall_exit = {
514 .system = "syscalls",
515 .reg = syscall_exit_register,
516 .define_fields = syscall_exit_define_fields,
517 .fields = LIST_HEAD_INIT(event_class_syscall_exit.fields),
518 .raw_init = init_syscall_trace,
519};
520
521unsigned long __init __weak arch_syscall_addr(int nr)
522{
523 return (unsigned long)sys_call_table[nr];
524}
525
526static int __init init_ftrace_syscalls(void)
527{
528 struct syscall_metadata *meta;
529 unsigned long addr;
530 int i;
531
532 syscalls_metadata = kcalloc(NR_syscalls, sizeof(*syscalls_metadata),
533 GFP_KERNEL);
534 if (!syscalls_metadata) {
535 WARN_ON(1);
536 return -ENOMEM;
537 }
538
539 for (i = 0; i < NR_syscalls; i++) {
540 addr = arch_syscall_addr(i);
541 meta = find_syscall_meta(addr);
542 if (!meta)
543 continue;
544
545 meta->syscall_nr = i;
546 syscalls_metadata[i] = meta;
547 }
548
549 return 0;
550}
551early_initcall(init_ftrace_syscalls);
552
553#ifdef CONFIG_PERF_EVENTS
554
555static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
556static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
557static int sys_perf_refcount_enter;
558static int sys_perf_refcount_exit;
559
560static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
561{
562 struct syscall_metadata *sys_data;
563 struct syscall_trace_enter *rec;
564 struct hlist_head *head;
565 int syscall_nr;
566 int rctx;
567 int size;
568
569 syscall_nr = trace_get_syscall_nr(current, regs);
570 if (syscall_nr < 0)
571 return;
572 if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
573 return;
574
575 sys_data = syscall_nr_to_meta(syscall_nr);
576 if (!sys_data)
577 return;
578
579 head = this_cpu_ptr(sys_data->enter_event->perf_events);
580 if (hlist_empty(head))
581 return;
582
583 /* get the size after alignment with the u32 buffer size field */
584 size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
585 size = ALIGN(size + sizeof(u32), sizeof(u64));
586 size -= sizeof(u32);
587
588 rec = (struct syscall_trace_enter *)perf_trace_buf_prepare(size,
589 sys_data->enter_event->event.type, regs, &rctx);
590 if (!rec)
591 return;
592
593 rec->nr = syscall_nr;
594 syscall_get_arguments(current, regs, 0, sys_data->nb_args,
595 (unsigned long *)&rec->args);
596 perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head, NULL);
597}
598
599static int perf_sysenter_enable(struct ftrace_event_call *call)
600{
601 int ret = 0;
602 int num;
603
604 num = ((struct syscall_metadata *)call->data)->syscall_nr;
605
606 mutex_lock(&syscall_trace_lock);
607 if (!sys_perf_refcount_enter)
608 ret = register_trace_sys_enter(perf_syscall_enter, NULL);
609 if (ret) {
610 pr_info("event trace: Could not activate"
611 "syscall entry trace point");
612 } else {
613 set_bit(num, enabled_perf_enter_syscalls);
614 sys_perf_refcount_enter++;
615 }
616 mutex_unlock(&syscall_trace_lock);
617 return ret;
618}
619
620static void perf_sysenter_disable(struct ftrace_event_call *call)
621{
622 int num;
623
624 num = ((struct syscall_metadata *)call->data)->syscall_nr;
625
626 mutex_lock(&syscall_trace_lock);
627 sys_perf_refcount_enter--;
628 clear_bit(num, enabled_perf_enter_syscalls);
629 if (!sys_perf_refcount_enter)
630 unregister_trace_sys_enter(perf_syscall_enter, NULL);
631 mutex_unlock(&syscall_trace_lock);
632}
633
634static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
635{
636 struct syscall_metadata *sys_data;
637 struct syscall_trace_exit *rec;
638 struct hlist_head *head;
639 int syscall_nr;
640 int rctx;
641 int size;
642
643 syscall_nr = trace_get_syscall_nr(current, regs);
644 if (syscall_nr < 0)
645 return;
646 if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
647 return;
648
649 sys_data = syscall_nr_to_meta(syscall_nr);
650 if (!sys_data)
651 return;
652
653 head = this_cpu_ptr(sys_data->exit_event->perf_events);
654 if (hlist_empty(head))
655 return;
656
657 /* We can probably do that at build time */
658 size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
659 size -= sizeof(u32);
660
661 rec = (struct syscall_trace_exit *)perf_trace_buf_prepare(size,
662 sys_data->exit_event->event.type, regs, &rctx);
663 if (!rec)
664 return;
665
666 rec->nr = syscall_nr;
667 rec->ret = syscall_get_return_value(current, regs);
668 perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head, NULL);
669}
670
671static int perf_sysexit_enable(struct ftrace_event_call *call)
672{
673 int ret = 0;
674 int num;
675
676 num = ((struct syscall_metadata *)call->data)->syscall_nr;
677
678 mutex_lock(&syscall_trace_lock);
679 if (!sys_perf_refcount_exit)
680 ret = register_trace_sys_exit(perf_syscall_exit, NULL);
681 if (ret) {
682 pr_info("event trace: Could not activate"
683 "syscall exit trace point");
684 } else {
685 set_bit(num, enabled_perf_exit_syscalls);
686 sys_perf_refcount_exit++;
687 }
688 mutex_unlock(&syscall_trace_lock);
689 return ret;
690}
691
692static void perf_sysexit_disable(struct ftrace_event_call *call)
693{
694 int num;
695
696 num = ((struct syscall_metadata *)call->data)->syscall_nr;
697
698 mutex_lock(&syscall_trace_lock);
699 sys_perf_refcount_exit--;
700 clear_bit(num, enabled_perf_exit_syscalls);
701 if (!sys_perf_refcount_exit)
702 unregister_trace_sys_exit(perf_syscall_exit, NULL);
703 mutex_unlock(&syscall_trace_lock);
704}
705
706#endif /* CONFIG_PERF_EVENTS */
707
708static int syscall_enter_register(struct ftrace_event_call *event,
709 enum trace_reg type, void *data)
710{
711 struct ftrace_event_file *file = data;
712
713 switch (type) {
714 case TRACE_REG_REGISTER:
715 return reg_event_syscall_enter(file, event);
716 case TRACE_REG_UNREGISTER:
717 unreg_event_syscall_enter(file, event);
718 return 0;
719
720#ifdef CONFIG_PERF_EVENTS
721 case TRACE_REG_PERF_REGISTER:
722 return perf_sysenter_enable(event);
723 case TRACE_REG_PERF_UNREGISTER:
724 perf_sysenter_disable(event);
725 return 0;
726 case TRACE_REG_PERF_OPEN:
727 case TRACE_REG_PERF_CLOSE:
728 case TRACE_REG_PERF_ADD:
729 case TRACE_REG_PERF_DEL:
730 return 0;
731#endif
732 }
733 return 0;
734}
735
736static int syscall_exit_register(struct ftrace_event_call *event,
737 enum trace_reg type, void *data)
738{
739 struct ftrace_event_file *file = data;
740
741 switch (type) {
742 case TRACE_REG_REGISTER:
743 return reg_event_syscall_exit(file, event);
744 case TRACE_REG_UNREGISTER:
745 unreg_event_syscall_exit(file, event);
746 return 0;
747
748#ifdef CONFIG_PERF_EVENTS
749 case TRACE_REG_PERF_REGISTER:
750 return perf_sysexit_enable(event);
751 case TRACE_REG_PERF_UNREGISTER:
752 perf_sysexit_disable(event);
753 return 0;
754 case TRACE_REG_PERF_OPEN:
755 case TRACE_REG_PERF_CLOSE:
756 case TRACE_REG_PERF_ADD:
757 case TRACE_REG_PERF_DEL:
758 return 0;
759#endif
760 }
761 return 0;
762}