xref: /linux/kernel/trace/trace_syscalls.c (revision 132db93572821ec2fdf81e354cc40f558faf7e4f)
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 
16 static DEFINE_MUTEX(syscall_trace_lock);
17 
18 static int syscall_enter_register(struct trace_event_call *event,
19 				 enum trace_reg type, void *data);
20 static int syscall_exit_register(struct trace_event_call *event,
21 				 enum trace_reg type, void *data);
22 
23 static struct list_head *
24 syscall_get_enter_fields(struct trace_event_call *call)
25 {
26 	struct syscall_metadata *entry = call->data;
27 
28 	return &entry->enter_fields;
29 }
30 
31 extern struct syscall_metadata *__start_syscalls_metadata[];
32 extern struct syscall_metadata *__stop_syscalls_metadata[];
33 
34 static DEFINE_XARRAY(syscalls_metadata_sparse);
35 static struct syscall_metadata **syscalls_metadata;
36 
37 #ifndef ARCH_HAS_SYSCALL_MATCH_SYM_NAME
38 static 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  */
66 static int
67 trace_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
75 static inline int
76 trace_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 
82 static __init struct syscall_metadata *
83 find_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 
104 static 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 
115 const 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 
126 static enum print_line_t
127 print_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->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, ')');
167 end:
168 	trace_seq_putc(s, '\n');
169 
170 	return trace_handle_return(s);
171 }
172 
173 static enum print_line_t
174 print_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 extern char *__bad_type_size(void);
205 
206 #define SYSCALL_FIELD(_type, _name) {					\
207 	.type = #_type, .name = #_name,					\
208 	.size = sizeof(_type), .align = __alignof__(_type),		\
209 	.is_signed = is_signed_type(_type), .filter_type = FILTER_OTHER }
210 
211 static int __init
212 __set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len)
213 {
214 	int i;
215 	int pos = 0;
216 
217 	/* When len=0, we just calculate the needed length */
218 #define LEN_OR_ZERO (len ? len - pos : 0)
219 
220 	pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
221 	for (i = 0; i < entry->nb_args; i++) {
222 		pos += snprintf(buf + pos, LEN_OR_ZERO, "%s: 0x%%0%zulx%s",
223 				entry->args[i], sizeof(unsigned long),
224 				i == entry->nb_args - 1 ? "" : ", ");
225 	}
226 	pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
227 
228 	for (i = 0; i < entry->nb_args; i++) {
229 		pos += snprintf(buf + pos, LEN_OR_ZERO,
230 				", ((unsigned long)(REC->%s))", entry->args[i]);
231 	}
232 
233 #undef LEN_OR_ZERO
234 
235 	/* return the length of print_fmt */
236 	return pos;
237 }
238 
239 static int __init set_syscall_print_fmt(struct trace_event_call *call)
240 {
241 	char *print_fmt;
242 	int len;
243 	struct syscall_metadata *entry = call->data;
244 
245 	if (entry->enter_event != call) {
246 		call->print_fmt = "\"0x%lx\", REC->ret";
247 		return 0;
248 	}
249 
250 	/* First: called with 0 length to calculate the needed length */
251 	len = __set_enter_print_fmt(entry, NULL, 0);
252 
253 	print_fmt = kmalloc(len + 1, GFP_KERNEL);
254 	if (!print_fmt)
255 		return -ENOMEM;
256 
257 	/* Second: actually write the @print_fmt */
258 	__set_enter_print_fmt(entry, print_fmt, len + 1);
259 	call->print_fmt = print_fmt;
260 
261 	return 0;
262 }
263 
264 static void __init free_syscall_print_fmt(struct trace_event_call *call)
265 {
266 	struct syscall_metadata *entry = call->data;
267 
268 	if (entry->enter_event == call)
269 		kfree(call->print_fmt);
270 }
271 
272 static int __init syscall_enter_define_fields(struct trace_event_call *call)
273 {
274 	struct syscall_trace_enter trace;
275 	struct syscall_metadata *meta = call->data;
276 	int offset = offsetof(typeof(trace), args);
277 	int ret = 0;
278 	int i;
279 
280 	for (i = 0; i < meta->nb_args; i++) {
281 		ret = trace_define_field(call, meta->types[i],
282 					 meta->args[i], offset,
283 					 sizeof(unsigned long), 0,
284 					 FILTER_OTHER);
285 		if (ret)
286 			break;
287 		offset += sizeof(unsigned long);
288 	}
289 
290 	return ret;
291 }
292 
293 static void ftrace_syscall_enter(void *data, struct pt_regs *regs, long id)
294 {
295 	struct trace_array *tr = data;
296 	struct trace_event_file *trace_file;
297 	struct syscall_trace_enter *entry;
298 	struct syscall_metadata *sys_data;
299 	struct ring_buffer_event *event;
300 	struct trace_buffer *buffer;
301 	unsigned long irq_flags;
302 	unsigned long args[6];
303 	int pc;
304 	int syscall_nr;
305 	int size;
306 
307 	syscall_nr = trace_get_syscall_nr(current, regs);
308 	if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
309 		return;
310 
311 	/* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE) */
312 	trace_file = rcu_dereference_sched(tr->enter_syscall_files[syscall_nr]);
313 	if (!trace_file)
314 		return;
315 
316 	if (trace_trigger_soft_disabled(trace_file))
317 		return;
318 
319 	sys_data = syscall_nr_to_meta(syscall_nr);
320 	if (!sys_data)
321 		return;
322 
323 	size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
324 
325 	local_save_flags(irq_flags);
326 	pc = preempt_count();
327 
328 	buffer = tr->array_buffer.buffer;
329 	event = trace_buffer_lock_reserve(buffer,
330 			sys_data->enter_event->event.type, size, irq_flags, pc);
331 	if (!event)
332 		return;
333 
334 	entry = ring_buffer_event_data(event);
335 	entry->nr = syscall_nr;
336 	syscall_get_arguments(current, regs, args);
337 	memcpy(entry->args, args, sizeof(unsigned long) * sys_data->nb_args);
338 
339 	event_trigger_unlock_commit(trace_file, buffer, event, entry,
340 				    irq_flags, pc);
341 }
342 
343 static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
344 {
345 	struct trace_array *tr = data;
346 	struct trace_event_file *trace_file;
347 	struct syscall_trace_exit *entry;
348 	struct syscall_metadata *sys_data;
349 	struct ring_buffer_event *event;
350 	struct trace_buffer *buffer;
351 	unsigned long irq_flags;
352 	int pc;
353 	int syscall_nr;
354 
355 	syscall_nr = trace_get_syscall_nr(current, regs);
356 	if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
357 		return;
358 
359 	/* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE()) */
360 	trace_file = rcu_dereference_sched(tr->exit_syscall_files[syscall_nr]);
361 	if (!trace_file)
362 		return;
363 
364 	if (trace_trigger_soft_disabled(trace_file))
365 		return;
366 
367 	sys_data = syscall_nr_to_meta(syscall_nr);
368 	if (!sys_data)
369 		return;
370 
371 	local_save_flags(irq_flags);
372 	pc = preempt_count();
373 
374 	buffer = tr->array_buffer.buffer;
375 	event = trace_buffer_lock_reserve(buffer,
376 			sys_data->exit_event->event.type, sizeof(*entry),
377 			irq_flags, pc);
378 	if (!event)
379 		return;
380 
381 	entry = ring_buffer_event_data(event);
382 	entry->nr = syscall_nr;
383 	entry->ret = syscall_get_return_value(current, regs);
384 
385 	event_trigger_unlock_commit(trace_file, buffer, event, entry,
386 				    irq_flags, pc);
387 }
388 
389 static int reg_event_syscall_enter(struct trace_event_file *file,
390 				   struct trace_event_call *call)
391 {
392 	struct trace_array *tr = file->tr;
393 	int ret = 0;
394 	int num;
395 
396 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
397 	if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
398 		return -ENOSYS;
399 	mutex_lock(&syscall_trace_lock);
400 	if (!tr->sys_refcount_enter)
401 		ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
402 	if (!ret) {
403 		rcu_assign_pointer(tr->enter_syscall_files[num], file);
404 		tr->sys_refcount_enter++;
405 	}
406 	mutex_unlock(&syscall_trace_lock);
407 	return ret;
408 }
409 
410 static void unreg_event_syscall_enter(struct trace_event_file *file,
411 				      struct trace_event_call *call)
412 {
413 	struct trace_array *tr = file->tr;
414 	int num;
415 
416 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
417 	if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
418 		return;
419 	mutex_lock(&syscall_trace_lock);
420 	tr->sys_refcount_enter--;
421 	RCU_INIT_POINTER(tr->enter_syscall_files[num], NULL);
422 	if (!tr->sys_refcount_enter)
423 		unregister_trace_sys_enter(ftrace_syscall_enter, tr);
424 	mutex_unlock(&syscall_trace_lock);
425 }
426 
427 static int reg_event_syscall_exit(struct trace_event_file *file,
428 				  struct trace_event_call *call)
429 {
430 	struct trace_array *tr = file->tr;
431 	int ret = 0;
432 	int num;
433 
434 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
435 	if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
436 		return -ENOSYS;
437 	mutex_lock(&syscall_trace_lock);
438 	if (!tr->sys_refcount_exit)
439 		ret = register_trace_sys_exit(ftrace_syscall_exit, tr);
440 	if (!ret) {
441 		rcu_assign_pointer(tr->exit_syscall_files[num], file);
442 		tr->sys_refcount_exit++;
443 	}
444 	mutex_unlock(&syscall_trace_lock);
445 	return ret;
446 }
447 
448 static void unreg_event_syscall_exit(struct trace_event_file *file,
449 				     struct trace_event_call *call)
450 {
451 	struct trace_array *tr = file->tr;
452 	int num;
453 
454 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
455 	if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
456 		return;
457 	mutex_lock(&syscall_trace_lock);
458 	tr->sys_refcount_exit--;
459 	RCU_INIT_POINTER(tr->exit_syscall_files[num], NULL);
460 	if (!tr->sys_refcount_exit)
461 		unregister_trace_sys_exit(ftrace_syscall_exit, tr);
462 	mutex_unlock(&syscall_trace_lock);
463 }
464 
465 static int __init init_syscall_trace(struct trace_event_call *call)
466 {
467 	int id;
468 	int num;
469 
470 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
471 	if (num < 0 || num >= NR_syscalls) {
472 		pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
473 				((struct syscall_metadata *)call->data)->name);
474 		return -ENOSYS;
475 	}
476 
477 	if (set_syscall_print_fmt(call) < 0)
478 		return -ENOMEM;
479 
480 	id = trace_event_raw_init(call);
481 
482 	if (id < 0) {
483 		free_syscall_print_fmt(call);
484 		return id;
485 	}
486 
487 	return id;
488 }
489 
490 static struct trace_event_fields __refdata syscall_enter_fields_array[] = {
491 	SYSCALL_FIELD(int, __syscall_nr),
492 	{ .type = TRACE_FUNCTION_TYPE,
493 	  .define_fields = syscall_enter_define_fields },
494 	{}
495 };
496 
497 struct trace_event_functions enter_syscall_print_funcs = {
498 	.trace		= print_syscall_enter,
499 };
500 
501 struct trace_event_functions exit_syscall_print_funcs = {
502 	.trace		= print_syscall_exit,
503 };
504 
505 struct trace_event_class __refdata event_class_syscall_enter = {
506 	.system		= "syscalls",
507 	.reg		= syscall_enter_register,
508 	.fields_array	= syscall_enter_fields_array,
509 	.get_fields	= syscall_get_enter_fields,
510 	.raw_init	= init_syscall_trace,
511 };
512 
513 struct trace_event_class __refdata event_class_syscall_exit = {
514 	.system		= "syscalls",
515 	.reg		= syscall_exit_register,
516 	.fields_array	= (struct trace_event_fields[]){
517 		SYSCALL_FIELD(int, __syscall_nr),
518 		SYSCALL_FIELD(long, ret),
519 		{}
520 	},
521 	.fields		= LIST_HEAD_INIT(event_class_syscall_exit.fields),
522 	.raw_init	= init_syscall_trace,
523 };
524 
525 unsigned long __init __weak arch_syscall_addr(int nr)
526 {
527 	return (unsigned long)sys_call_table[nr];
528 }
529 
530 void __init init_ftrace_syscalls(void)
531 {
532 	struct syscall_metadata *meta;
533 	unsigned long addr;
534 	int i;
535 	void *ret;
536 
537 	if (!IS_ENABLED(CONFIG_HAVE_SPARSE_SYSCALL_NR)) {
538 		syscalls_metadata = kcalloc(NR_syscalls,
539 					sizeof(*syscalls_metadata),
540 					GFP_KERNEL);
541 		if (!syscalls_metadata) {
542 			WARN_ON(1);
543 			return;
544 		}
545 	}
546 
547 	for (i = 0; i < NR_syscalls; i++) {
548 		addr = arch_syscall_addr(i);
549 		meta = find_syscall_meta(addr);
550 		if (!meta)
551 			continue;
552 
553 		meta->syscall_nr = i;
554 
555 		if (!IS_ENABLED(CONFIG_HAVE_SPARSE_SYSCALL_NR)) {
556 			syscalls_metadata[i] = meta;
557 		} else {
558 			ret = xa_store(&syscalls_metadata_sparse, i, meta,
559 					GFP_KERNEL);
560 			WARN(xa_is_err(ret),
561 				"Syscall memory allocation failed\n");
562 		}
563 
564 	}
565 }
566 
567 #ifdef CONFIG_PERF_EVENTS
568 
569 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
570 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
571 static int sys_perf_refcount_enter;
572 static int sys_perf_refcount_exit;
573 
574 static int perf_call_bpf_enter(struct trace_event_call *call, struct pt_regs *regs,
575 			       struct syscall_metadata *sys_data,
576 			       struct syscall_trace_enter *rec)
577 {
578 	struct syscall_tp_t {
579 		unsigned long long regs;
580 		unsigned long syscall_nr;
581 		unsigned long args[SYSCALL_DEFINE_MAXARGS];
582 	} param;
583 	int i;
584 
585 	*(struct pt_regs **)&param = regs;
586 	param.syscall_nr = rec->nr;
587 	for (i = 0; i < sys_data->nb_args; i++)
588 		param.args[i] = rec->args[i];
589 	return trace_call_bpf(call, &param);
590 }
591 
592 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
593 {
594 	struct syscall_metadata *sys_data;
595 	struct syscall_trace_enter *rec;
596 	struct hlist_head *head;
597 	unsigned long args[6];
598 	bool valid_prog_array;
599 	int syscall_nr;
600 	int rctx;
601 	int size;
602 
603 	syscall_nr = trace_get_syscall_nr(current, regs);
604 	if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
605 		return;
606 	if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
607 		return;
608 
609 	sys_data = syscall_nr_to_meta(syscall_nr);
610 	if (!sys_data)
611 		return;
612 
613 	head = this_cpu_ptr(sys_data->enter_event->perf_events);
614 	valid_prog_array = bpf_prog_array_valid(sys_data->enter_event);
615 	if (!valid_prog_array && hlist_empty(head))
616 		return;
617 
618 	/* get the size after alignment with the u32 buffer size field */
619 	size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
620 	size = ALIGN(size + sizeof(u32), sizeof(u64));
621 	size -= sizeof(u32);
622 
623 	rec = perf_trace_buf_alloc(size, NULL, &rctx);
624 	if (!rec)
625 		return;
626 
627 	rec->nr = syscall_nr;
628 	syscall_get_arguments(current, regs, args);
629 	memcpy(&rec->args, args, sizeof(unsigned long) * sys_data->nb_args);
630 
631 	if ((valid_prog_array &&
632 	     !perf_call_bpf_enter(sys_data->enter_event, regs, sys_data, rec)) ||
633 	    hlist_empty(head)) {
634 		perf_swevent_put_recursion_context(rctx);
635 		return;
636 	}
637 
638 	perf_trace_buf_submit(rec, size, rctx,
639 			      sys_data->enter_event->event.type, 1, regs,
640 			      head, NULL);
641 }
642 
643 static int perf_sysenter_enable(struct trace_event_call *call)
644 {
645 	int ret = 0;
646 	int num;
647 
648 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
649 
650 	mutex_lock(&syscall_trace_lock);
651 	if (!sys_perf_refcount_enter)
652 		ret = register_trace_sys_enter(perf_syscall_enter, NULL);
653 	if (ret) {
654 		pr_info("event trace: Could not activate syscall entry trace point");
655 	} else {
656 		set_bit(num, enabled_perf_enter_syscalls);
657 		sys_perf_refcount_enter++;
658 	}
659 	mutex_unlock(&syscall_trace_lock);
660 	return ret;
661 }
662 
663 static void perf_sysenter_disable(struct trace_event_call *call)
664 {
665 	int num;
666 
667 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
668 
669 	mutex_lock(&syscall_trace_lock);
670 	sys_perf_refcount_enter--;
671 	clear_bit(num, enabled_perf_enter_syscalls);
672 	if (!sys_perf_refcount_enter)
673 		unregister_trace_sys_enter(perf_syscall_enter, NULL);
674 	mutex_unlock(&syscall_trace_lock);
675 }
676 
677 static int perf_call_bpf_exit(struct trace_event_call *call, struct pt_regs *regs,
678 			      struct syscall_trace_exit *rec)
679 {
680 	struct syscall_tp_t {
681 		unsigned long long regs;
682 		unsigned long syscall_nr;
683 		unsigned long ret;
684 	} param;
685 
686 	*(struct pt_regs **)&param = regs;
687 	param.syscall_nr = rec->nr;
688 	param.ret = rec->ret;
689 	return trace_call_bpf(call, &param);
690 }
691 
692 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
693 {
694 	struct syscall_metadata *sys_data;
695 	struct syscall_trace_exit *rec;
696 	struct hlist_head *head;
697 	bool valid_prog_array;
698 	int syscall_nr;
699 	int rctx;
700 	int size;
701 
702 	syscall_nr = trace_get_syscall_nr(current, regs);
703 	if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
704 		return;
705 	if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
706 		return;
707 
708 	sys_data = syscall_nr_to_meta(syscall_nr);
709 	if (!sys_data)
710 		return;
711 
712 	head = this_cpu_ptr(sys_data->exit_event->perf_events);
713 	valid_prog_array = bpf_prog_array_valid(sys_data->exit_event);
714 	if (!valid_prog_array && hlist_empty(head))
715 		return;
716 
717 	/* We can probably do that at build time */
718 	size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
719 	size -= sizeof(u32);
720 
721 	rec = perf_trace_buf_alloc(size, NULL, &rctx);
722 	if (!rec)
723 		return;
724 
725 	rec->nr = syscall_nr;
726 	rec->ret = syscall_get_return_value(current, regs);
727 
728 	if ((valid_prog_array &&
729 	     !perf_call_bpf_exit(sys_data->exit_event, regs, rec)) ||
730 	    hlist_empty(head)) {
731 		perf_swevent_put_recursion_context(rctx);
732 		return;
733 	}
734 
735 	perf_trace_buf_submit(rec, size, rctx, sys_data->exit_event->event.type,
736 			      1, regs, head, NULL);
737 }
738 
739 static int perf_sysexit_enable(struct trace_event_call *call)
740 {
741 	int ret = 0;
742 	int num;
743 
744 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
745 
746 	mutex_lock(&syscall_trace_lock);
747 	if (!sys_perf_refcount_exit)
748 		ret = register_trace_sys_exit(perf_syscall_exit, NULL);
749 	if (ret) {
750 		pr_info("event trace: Could not activate syscall exit trace point");
751 	} else {
752 		set_bit(num, enabled_perf_exit_syscalls);
753 		sys_perf_refcount_exit++;
754 	}
755 	mutex_unlock(&syscall_trace_lock);
756 	return ret;
757 }
758 
759 static void perf_sysexit_disable(struct trace_event_call *call)
760 {
761 	int num;
762 
763 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
764 
765 	mutex_lock(&syscall_trace_lock);
766 	sys_perf_refcount_exit--;
767 	clear_bit(num, enabled_perf_exit_syscalls);
768 	if (!sys_perf_refcount_exit)
769 		unregister_trace_sys_exit(perf_syscall_exit, NULL);
770 	mutex_unlock(&syscall_trace_lock);
771 }
772 
773 #endif /* CONFIG_PERF_EVENTS */
774 
775 static int syscall_enter_register(struct trace_event_call *event,
776 				 enum trace_reg type, void *data)
777 {
778 	struct trace_event_file *file = data;
779 
780 	switch (type) {
781 	case TRACE_REG_REGISTER:
782 		return reg_event_syscall_enter(file, event);
783 	case TRACE_REG_UNREGISTER:
784 		unreg_event_syscall_enter(file, event);
785 		return 0;
786 
787 #ifdef CONFIG_PERF_EVENTS
788 	case TRACE_REG_PERF_REGISTER:
789 		return perf_sysenter_enable(event);
790 	case TRACE_REG_PERF_UNREGISTER:
791 		perf_sysenter_disable(event);
792 		return 0;
793 	case TRACE_REG_PERF_OPEN:
794 	case TRACE_REG_PERF_CLOSE:
795 	case TRACE_REG_PERF_ADD:
796 	case TRACE_REG_PERF_DEL:
797 		return 0;
798 #endif
799 	}
800 	return 0;
801 }
802 
803 static int syscall_exit_register(struct trace_event_call *event,
804 				 enum trace_reg type, void *data)
805 {
806 	struct trace_event_file *file = data;
807 
808 	switch (type) {
809 	case TRACE_REG_REGISTER:
810 		return reg_event_syscall_exit(file, event);
811 	case TRACE_REG_UNREGISTER:
812 		unreg_event_syscall_exit(file, event);
813 		return 0;
814 
815 #ifdef CONFIG_PERF_EVENTS
816 	case TRACE_REG_PERF_REGISTER:
817 		return perf_sysexit_enable(event);
818 	case TRACE_REG_PERF_UNREGISTER:
819 		perf_sysexit_disable(event);
820 		return 0;
821 	case TRACE_REG_PERF_OPEN:
822 	case TRACE_REG_PERF_CLOSE:
823 	case TRACE_REG_PERF_ADD:
824 	case TRACE_REG_PERF_DEL:
825 		return 0;
826 #endif
827 	}
828 	return 0;
829 }
830