xref: /linux/kernel/trace/trace_syscalls.c (revision dec1c62e91ba268ab2a6e339d4d7a59287d5eba1)
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 && 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 trace_event_buffer fbuffer;
300 	unsigned long args[6];
301 	int syscall_nr;
302 	int size;
303 
304 	syscall_nr = trace_get_syscall_nr(current, regs);
305 	if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
306 		return;
307 
308 	/* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE) */
309 	trace_file = rcu_dereference_sched(tr->enter_syscall_files[syscall_nr]);
310 	if (!trace_file)
311 		return;
312 
313 	if (trace_trigger_soft_disabled(trace_file))
314 		return;
315 
316 	sys_data = syscall_nr_to_meta(syscall_nr);
317 	if (!sys_data)
318 		return;
319 
320 	size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
321 
322 	entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
323 	if (!entry)
324 		return;
325 
326 	entry = ring_buffer_event_data(fbuffer.event);
327 	entry->nr = syscall_nr;
328 	syscall_get_arguments(current, regs, args);
329 	memcpy(entry->args, args, sizeof(unsigned long) * sys_data->nb_args);
330 
331 	trace_event_buffer_commit(&fbuffer);
332 }
333 
334 static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
335 {
336 	struct trace_array *tr = data;
337 	struct trace_event_file *trace_file;
338 	struct syscall_trace_exit *entry;
339 	struct syscall_metadata *sys_data;
340 	struct trace_event_buffer fbuffer;
341 	int syscall_nr;
342 
343 	syscall_nr = trace_get_syscall_nr(current, regs);
344 	if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
345 		return;
346 
347 	/* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE()) */
348 	trace_file = rcu_dereference_sched(tr->exit_syscall_files[syscall_nr]);
349 	if (!trace_file)
350 		return;
351 
352 	if (trace_trigger_soft_disabled(trace_file))
353 		return;
354 
355 	sys_data = syscall_nr_to_meta(syscall_nr);
356 	if (!sys_data)
357 		return;
358 
359 	entry = trace_event_buffer_reserve(&fbuffer, trace_file, sizeof(*entry));
360 	if (!entry)
361 		return;
362 
363 	entry = ring_buffer_event_data(fbuffer.event);
364 	entry->nr = syscall_nr;
365 	entry->ret = syscall_get_return_value(current, regs);
366 
367 	trace_event_buffer_commit(&fbuffer);
368 }
369 
370 static int reg_event_syscall_enter(struct trace_event_file *file,
371 				   struct trace_event_call *call)
372 {
373 	struct trace_array *tr = file->tr;
374 	int ret = 0;
375 	int num;
376 
377 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
378 	if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
379 		return -ENOSYS;
380 	mutex_lock(&syscall_trace_lock);
381 	if (!tr->sys_refcount_enter)
382 		ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
383 	if (!ret) {
384 		rcu_assign_pointer(tr->enter_syscall_files[num], file);
385 		tr->sys_refcount_enter++;
386 	}
387 	mutex_unlock(&syscall_trace_lock);
388 	return ret;
389 }
390 
391 static void unreg_event_syscall_enter(struct trace_event_file *file,
392 				      struct trace_event_call *call)
393 {
394 	struct trace_array *tr = file->tr;
395 	int num;
396 
397 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
398 	if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
399 		return;
400 	mutex_lock(&syscall_trace_lock);
401 	tr->sys_refcount_enter--;
402 	RCU_INIT_POINTER(tr->enter_syscall_files[num], NULL);
403 	if (!tr->sys_refcount_enter)
404 		unregister_trace_sys_enter(ftrace_syscall_enter, tr);
405 	mutex_unlock(&syscall_trace_lock);
406 }
407 
408 static int reg_event_syscall_exit(struct trace_event_file *file,
409 				  struct trace_event_call *call)
410 {
411 	struct trace_array *tr = file->tr;
412 	int ret = 0;
413 	int num;
414 
415 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
416 	if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
417 		return -ENOSYS;
418 	mutex_lock(&syscall_trace_lock);
419 	if (!tr->sys_refcount_exit)
420 		ret = register_trace_sys_exit(ftrace_syscall_exit, tr);
421 	if (!ret) {
422 		rcu_assign_pointer(tr->exit_syscall_files[num], file);
423 		tr->sys_refcount_exit++;
424 	}
425 	mutex_unlock(&syscall_trace_lock);
426 	return ret;
427 }
428 
429 static void unreg_event_syscall_exit(struct trace_event_file *file,
430 				     struct trace_event_call *call)
431 {
432 	struct trace_array *tr = file->tr;
433 	int num;
434 
435 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
436 	if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
437 		return;
438 	mutex_lock(&syscall_trace_lock);
439 	tr->sys_refcount_exit--;
440 	RCU_INIT_POINTER(tr->exit_syscall_files[num], NULL);
441 	if (!tr->sys_refcount_exit)
442 		unregister_trace_sys_exit(ftrace_syscall_exit, tr);
443 	mutex_unlock(&syscall_trace_lock);
444 }
445 
446 static int __init init_syscall_trace(struct trace_event_call *call)
447 {
448 	int id;
449 	int num;
450 
451 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
452 	if (num < 0 || num >= NR_syscalls) {
453 		pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
454 				((struct syscall_metadata *)call->data)->name);
455 		return -ENOSYS;
456 	}
457 
458 	if (set_syscall_print_fmt(call) < 0)
459 		return -ENOMEM;
460 
461 	id = trace_event_raw_init(call);
462 
463 	if (id < 0) {
464 		free_syscall_print_fmt(call);
465 		return id;
466 	}
467 
468 	return id;
469 }
470 
471 static struct trace_event_fields __refdata syscall_enter_fields_array[] = {
472 	SYSCALL_FIELD(int, __syscall_nr),
473 	{ .type = TRACE_FUNCTION_TYPE,
474 	  .define_fields = syscall_enter_define_fields },
475 	{}
476 };
477 
478 struct trace_event_functions enter_syscall_print_funcs = {
479 	.trace		= print_syscall_enter,
480 };
481 
482 struct trace_event_functions exit_syscall_print_funcs = {
483 	.trace		= print_syscall_exit,
484 };
485 
486 struct trace_event_class __refdata event_class_syscall_enter = {
487 	.system		= "syscalls",
488 	.reg		= syscall_enter_register,
489 	.fields_array	= syscall_enter_fields_array,
490 	.get_fields	= syscall_get_enter_fields,
491 	.raw_init	= init_syscall_trace,
492 };
493 
494 struct trace_event_class __refdata event_class_syscall_exit = {
495 	.system		= "syscalls",
496 	.reg		= syscall_exit_register,
497 	.fields_array	= (struct trace_event_fields[]){
498 		SYSCALL_FIELD(int, __syscall_nr),
499 		SYSCALL_FIELD(long, ret),
500 		{}
501 	},
502 	.fields		= LIST_HEAD_INIT(event_class_syscall_exit.fields),
503 	.raw_init	= init_syscall_trace,
504 };
505 
506 unsigned long __init __weak arch_syscall_addr(int nr)
507 {
508 	return (unsigned long)sys_call_table[nr];
509 }
510 
511 void __init init_ftrace_syscalls(void)
512 {
513 	struct syscall_metadata *meta;
514 	unsigned long addr;
515 	int i;
516 	void *ret;
517 
518 	if (!IS_ENABLED(CONFIG_HAVE_SPARSE_SYSCALL_NR)) {
519 		syscalls_metadata = kcalloc(NR_syscalls,
520 					sizeof(*syscalls_metadata),
521 					GFP_KERNEL);
522 		if (!syscalls_metadata) {
523 			WARN_ON(1);
524 			return;
525 		}
526 	}
527 
528 	for (i = 0; i < NR_syscalls; i++) {
529 		addr = arch_syscall_addr(i);
530 		meta = find_syscall_meta(addr);
531 		if (!meta)
532 			continue;
533 
534 		meta->syscall_nr = i;
535 
536 		if (!IS_ENABLED(CONFIG_HAVE_SPARSE_SYSCALL_NR)) {
537 			syscalls_metadata[i] = meta;
538 		} else {
539 			ret = xa_store(&syscalls_metadata_sparse, i, meta,
540 					GFP_KERNEL);
541 			WARN(xa_is_err(ret),
542 				"Syscall memory allocation failed\n");
543 		}
544 
545 	}
546 }
547 
548 #ifdef CONFIG_PERF_EVENTS
549 
550 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
551 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
552 static int sys_perf_refcount_enter;
553 static int sys_perf_refcount_exit;
554 
555 static int perf_call_bpf_enter(struct trace_event_call *call, struct pt_regs *regs,
556 			       struct syscall_metadata *sys_data,
557 			       struct syscall_trace_enter *rec)
558 {
559 	struct syscall_tp_t {
560 		unsigned long long regs;
561 		unsigned long syscall_nr;
562 		unsigned long args[SYSCALL_DEFINE_MAXARGS];
563 	} param;
564 	int i;
565 
566 	*(struct pt_regs **)&param = regs;
567 	param.syscall_nr = rec->nr;
568 	for (i = 0; i < sys_data->nb_args; i++)
569 		param.args[i] = rec->args[i];
570 	return trace_call_bpf(call, &param);
571 }
572 
573 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
574 {
575 	struct syscall_metadata *sys_data;
576 	struct syscall_trace_enter *rec;
577 	struct hlist_head *head;
578 	unsigned long args[6];
579 	bool valid_prog_array;
580 	int syscall_nr;
581 	int rctx;
582 	int size;
583 
584 	syscall_nr = trace_get_syscall_nr(current, regs);
585 	if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
586 		return;
587 	if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
588 		return;
589 
590 	sys_data = syscall_nr_to_meta(syscall_nr);
591 	if (!sys_data)
592 		return;
593 
594 	head = this_cpu_ptr(sys_data->enter_event->perf_events);
595 	valid_prog_array = bpf_prog_array_valid(sys_data->enter_event);
596 	if (!valid_prog_array && hlist_empty(head))
597 		return;
598 
599 	/* get the size after alignment with the u32 buffer size field */
600 	size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
601 	size = ALIGN(size + sizeof(u32), sizeof(u64));
602 	size -= sizeof(u32);
603 
604 	rec = perf_trace_buf_alloc(size, NULL, &rctx);
605 	if (!rec)
606 		return;
607 
608 	rec->nr = syscall_nr;
609 	syscall_get_arguments(current, regs, args);
610 	memcpy(&rec->args, args, sizeof(unsigned long) * sys_data->nb_args);
611 
612 	if ((valid_prog_array &&
613 	     !perf_call_bpf_enter(sys_data->enter_event, regs, sys_data, rec)) ||
614 	    hlist_empty(head)) {
615 		perf_swevent_put_recursion_context(rctx);
616 		return;
617 	}
618 
619 	perf_trace_buf_submit(rec, size, rctx,
620 			      sys_data->enter_event->event.type, 1, regs,
621 			      head, NULL);
622 }
623 
624 static int perf_sysenter_enable(struct trace_event_call *call)
625 {
626 	int ret = 0;
627 	int num;
628 
629 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
630 
631 	mutex_lock(&syscall_trace_lock);
632 	if (!sys_perf_refcount_enter)
633 		ret = register_trace_sys_enter(perf_syscall_enter, NULL);
634 	if (ret) {
635 		pr_info("event trace: Could not activate syscall entry trace point");
636 	} else {
637 		set_bit(num, enabled_perf_enter_syscalls);
638 		sys_perf_refcount_enter++;
639 	}
640 	mutex_unlock(&syscall_trace_lock);
641 	return ret;
642 }
643 
644 static void perf_sysenter_disable(struct trace_event_call *call)
645 {
646 	int num;
647 
648 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
649 
650 	mutex_lock(&syscall_trace_lock);
651 	sys_perf_refcount_enter--;
652 	clear_bit(num, enabled_perf_enter_syscalls);
653 	if (!sys_perf_refcount_enter)
654 		unregister_trace_sys_enter(perf_syscall_enter, NULL);
655 	mutex_unlock(&syscall_trace_lock);
656 }
657 
658 static int perf_call_bpf_exit(struct trace_event_call *call, struct pt_regs *regs,
659 			      struct syscall_trace_exit *rec)
660 {
661 	struct syscall_tp_t {
662 		unsigned long long regs;
663 		unsigned long syscall_nr;
664 		unsigned long ret;
665 	} param;
666 
667 	*(struct pt_regs **)&param = regs;
668 	param.syscall_nr = rec->nr;
669 	param.ret = rec->ret;
670 	return trace_call_bpf(call, &param);
671 }
672 
673 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
674 {
675 	struct syscall_metadata *sys_data;
676 	struct syscall_trace_exit *rec;
677 	struct hlist_head *head;
678 	bool valid_prog_array;
679 	int syscall_nr;
680 	int rctx;
681 	int size;
682 
683 	syscall_nr = trace_get_syscall_nr(current, regs);
684 	if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
685 		return;
686 	if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
687 		return;
688 
689 	sys_data = syscall_nr_to_meta(syscall_nr);
690 	if (!sys_data)
691 		return;
692 
693 	head = this_cpu_ptr(sys_data->exit_event->perf_events);
694 	valid_prog_array = bpf_prog_array_valid(sys_data->exit_event);
695 	if (!valid_prog_array && hlist_empty(head))
696 		return;
697 
698 	/* We can probably do that at build time */
699 	size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
700 	size -= sizeof(u32);
701 
702 	rec = perf_trace_buf_alloc(size, NULL, &rctx);
703 	if (!rec)
704 		return;
705 
706 	rec->nr = syscall_nr;
707 	rec->ret = syscall_get_return_value(current, regs);
708 
709 	if ((valid_prog_array &&
710 	     !perf_call_bpf_exit(sys_data->exit_event, regs, rec)) ||
711 	    hlist_empty(head)) {
712 		perf_swevent_put_recursion_context(rctx);
713 		return;
714 	}
715 
716 	perf_trace_buf_submit(rec, size, rctx, sys_data->exit_event->event.type,
717 			      1, regs, head, NULL);
718 }
719 
720 static int perf_sysexit_enable(struct trace_event_call *call)
721 {
722 	int ret = 0;
723 	int num;
724 
725 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
726 
727 	mutex_lock(&syscall_trace_lock);
728 	if (!sys_perf_refcount_exit)
729 		ret = register_trace_sys_exit(perf_syscall_exit, NULL);
730 	if (ret) {
731 		pr_info("event trace: Could not activate syscall exit trace point");
732 	} else {
733 		set_bit(num, enabled_perf_exit_syscalls);
734 		sys_perf_refcount_exit++;
735 	}
736 	mutex_unlock(&syscall_trace_lock);
737 	return ret;
738 }
739 
740 static void perf_sysexit_disable(struct trace_event_call *call)
741 {
742 	int num;
743 
744 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
745 
746 	mutex_lock(&syscall_trace_lock);
747 	sys_perf_refcount_exit--;
748 	clear_bit(num, enabled_perf_exit_syscalls);
749 	if (!sys_perf_refcount_exit)
750 		unregister_trace_sys_exit(perf_syscall_exit, NULL);
751 	mutex_unlock(&syscall_trace_lock);
752 }
753 
754 #endif /* CONFIG_PERF_EVENTS */
755 
756 static int syscall_enter_register(struct trace_event_call *event,
757 				 enum trace_reg type, void *data)
758 {
759 	struct trace_event_file *file = data;
760 
761 	switch (type) {
762 	case TRACE_REG_REGISTER:
763 		return reg_event_syscall_enter(file, event);
764 	case TRACE_REG_UNREGISTER:
765 		unreg_event_syscall_enter(file, event);
766 		return 0;
767 
768 #ifdef CONFIG_PERF_EVENTS
769 	case TRACE_REG_PERF_REGISTER:
770 		return perf_sysenter_enable(event);
771 	case TRACE_REG_PERF_UNREGISTER:
772 		perf_sysenter_disable(event);
773 		return 0;
774 	case TRACE_REG_PERF_OPEN:
775 	case TRACE_REG_PERF_CLOSE:
776 	case TRACE_REG_PERF_ADD:
777 	case TRACE_REG_PERF_DEL:
778 		return 0;
779 #endif
780 	}
781 	return 0;
782 }
783 
784 static int syscall_exit_register(struct trace_event_call *event,
785 				 enum trace_reg type, void *data)
786 {
787 	struct trace_event_file *file = data;
788 
789 	switch (type) {
790 	case TRACE_REG_REGISTER:
791 		return reg_event_syscall_exit(file, event);
792 	case TRACE_REG_UNREGISTER:
793 		unreg_event_syscall_exit(file, event);
794 		return 0;
795 
796 #ifdef CONFIG_PERF_EVENTS
797 	case TRACE_REG_PERF_REGISTER:
798 		return perf_sysexit_enable(event);
799 	case TRACE_REG_PERF_UNREGISTER:
800 		perf_sysexit_disable(event);
801 		return 0;
802 	case TRACE_REG_PERF_OPEN:
803 	case TRACE_REG_PERF_CLOSE:
804 	case TRACE_REG_PERF_ADD:
805 	case TRACE_REG_PERF_DEL:
806 		return 0;
807 #endif
808 	}
809 	return 0;
810 }
811