xref: /openbmc/linux/kernel/trace/trace_syscalls.c (revision d87c25e8)
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 int trace_ctx;
302 	unsigned long args[6];
303 	int syscall_nr;
304 	int size;
305 
306 	syscall_nr = trace_get_syscall_nr(current, regs);
307 	if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
308 		return;
309 
310 	/* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE) */
311 	trace_file = rcu_dereference_sched(tr->enter_syscall_files[syscall_nr]);
312 	if (!trace_file)
313 		return;
314 
315 	if (trace_trigger_soft_disabled(trace_file))
316 		return;
317 
318 	sys_data = syscall_nr_to_meta(syscall_nr);
319 	if (!sys_data)
320 		return;
321 
322 	size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
323 
324 	trace_ctx = tracing_gen_ctx();
325 
326 	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
327 			sys_data->enter_event->event.type, size, trace_ctx);
328 	if (!event)
329 		return;
330 
331 	entry = ring_buffer_event_data(event);
332 	entry->nr = syscall_nr;
333 	syscall_get_arguments(current, regs, args);
334 	memcpy(entry->args, args, sizeof(unsigned long) * sys_data->nb_args);
335 
336 	event_trigger_unlock_commit(trace_file, buffer, event, entry,
337 				    trace_ctx);
338 }
339 
340 static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
341 {
342 	struct trace_array *tr = data;
343 	struct trace_event_file *trace_file;
344 	struct syscall_trace_exit *entry;
345 	struct syscall_metadata *sys_data;
346 	struct ring_buffer_event *event;
347 	struct trace_buffer *buffer;
348 	unsigned int trace_ctx;
349 	int syscall_nr;
350 
351 	syscall_nr = trace_get_syscall_nr(current, regs);
352 	if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
353 		return;
354 
355 	/* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE()) */
356 	trace_file = rcu_dereference_sched(tr->exit_syscall_files[syscall_nr]);
357 	if (!trace_file)
358 		return;
359 
360 	if (trace_trigger_soft_disabled(trace_file))
361 		return;
362 
363 	sys_data = syscall_nr_to_meta(syscall_nr);
364 	if (!sys_data)
365 		return;
366 
367 	trace_ctx = tracing_gen_ctx();
368 
369 	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
370 			sys_data->exit_event->event.type, sizeof(*entry),
371 			trace_ctx);
372 	if (!event)
373 		return;
374 
375 	entry = ring_buffer_event_data(event);
376 	entry->nr = syscall_nr;
377 	entry->ret = syscall_get_return_value(current, regs);
378 
379 	event_trigger_unlock_commit(trace_file, buffer, event, entry,
380 				    trace_ctx);
381 }
382 
383 static int reg_event_syscall_enter(struct trace_event_file *file,
384 				   struct trace_event_call *call)
385 {
386 	struct trace_array *tr = file->tr;
387 	int ret = 0;
388 	int num;
389 
390 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
391 	if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
392 		return -ENOSYS;
393 	mutex_lock(&syscall_trace_lock);
394 	if (!tr->sys_refcount_enter)
395 		ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
396 	if (!ret) {
397 		rcu_assign_pointer(tr->enter_syscall_files[num], file);
398 		tr->sys_refcount_enter++;
399 	}
400 	mutex_unlock(&syscall_trace_lock);
401 	return ret;
402 }
403 
404 static void unreg_event_syscall_enter(struct trace_event_file *file,
405 				      struct trace_event_call *call)
406 {
407 	struct trace_array *tr = file->tr;
408 	int num;
409 
410 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
411 	if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
412 		return;
413 	mutex_lock(&syscall_trace_lock);
414 	tr->sys_refcount_enter--;
415 	RCU_INIT_POINTER(tr->enter_syscall_files[num], NULL);
416 	if (!tr->sys_refcount_enter)
417 		unregister_trace_sys_enter(ftrace_syscall_enter, tr);
418 	mutex_unlock(&syscall_trace_lock);
419 }
420 
421 static int reg_event_syscall_exit(struct trace_event_file *file,
422 				  struct trace_event_call *call)
423 {
424 	struct trace_array *tr = file->tr;
425 	int ret = 0;
426 	int num;
427 
428 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
429 	if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
430 		return -ENOSYS;
431 	mutex_lock(&syscall_trace_lock);
432 	if (!tr->sys_refcount_exit)
433 		ret = register_trace_sys_exit(ftrace_syscall_exit, tr);
434 	if (!ret) {
435 		rcu_assign_pointer(tr->exit_syscall_files[num], file);
436 		tr->sys_refcount_exit++;
437 	}
438 	mutex_unlock(&syscall_trace_lock);
439 	return ret;
440 }
441 
442 static void unreg_event_syscall_exit(struct trace_event_file *file,
443 				     struct trace_event_call *call)
444 {
445 	struct trace_array *tr = file->tr;
446 	int num;
447 
448 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
449 	if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
450 		return;
451 	mutex_lock(&syscall_trace_lock);
452 	tr->sys_refcount_exit--;
453 	RCU_INIT_POINTER(tr->exit_syscall_files[num], NULL);
454 	if (!tr->sys_refcount_exit)
455 		unregister_trace_sys_exit(ftrace_syscall_exit, tr);
456 	mutex_unlock(&syscall_trace_lock);
457 }
458 
459 static int __init init_syscall_trace(struct trace_event_call *call)
460 {
461 	int id;
462 	int num;
463 
464 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
465 	if (num < 0 || num >= NR_syscalls) {
466 		pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
467 				((struct syscall_metadata *)call->data)->name);
468 		return -ENOSYS;
469 	}
470 
471 	if (set_syscall_print_fmt(call) < 0)
472 		return -ENOMEM;
473 
474 	id = trace_event_raw_init(call);
475 
476 	if (id < 0) {
477 		free_syscall_print_fmt(call);
478 		return id;
479 	}
480 
481 	return id;
482 }
483 
484 static struct trace_event_fields __refdata syscall_enter_fields_array[] = {
485 	SYSCALL_FIELD(int, __syscall_nr),
486 	{ .type = TRACE_FUNCTION_TYPE,
487 	  .define_fields = syscall_enter_define_fields },
488 	{}
489 };
490 
491 struct trace_event_functions enter_syscall_print_funcs = {
492 	.trace		= print_syscall_enter,
493 };
494 
495 struct trace_event_functions exit_syscall_print_funcs = {
496 	.trace		= print_syscall_exit,
497 };
498 
499 struct trace_event_class __refdata event_class_syscall_enter = {
500 	.system		= "syscalls",
501 	.reg		= syscall_enter_register,
502 	.fields_array	= syscall_enter_fields_array,
503 	.get_fields	= syscall_get_enter_fields,
504 	.raw_init	= init_syscall_trace,
505 };
506 
507 struct trace_event_class __refdata event_class_syscall_exit = {
508 	.system		= "syscalls",
509 	.reg		= syscall_exit_register,
510 	.fields_array	= (struct trace_event_fields[]){
511 		SYSCALL_FIELD(int, __syscall_nr),
512 		SYSCALL_FIELD(long, ret),
513 		{}
514 	},
515 	.fields		= LIST_HEAD_INIT(event_class_syscall_exit.fields),
516 	.raw_init	= init_syscall_trace,
517 };
518 
519 unsigned long __init __weak arch_syscall_addr(int nr)
520 {
521 	return (unsigned long)sys_call_table[nr];
522 }
523 
524 void __init init_ftrace_syscalls(void)
525 {
526 	struct syscall_metadata *meta;
527 	unsigned long addr;
528 	int i;
529 	void *ret;
530 
531 	if (!IS_ENABLED(CONFIG_HAVE_SPARSE_SYSCALL_NR)) {
532 		syscalls_metadata = kcalloc(NR_syscalls,
533 					sizeof(*syscalls_metadata),
534 					GFP_KERNEL);
535 		if (!syscalls_metadata) {
536 			WARN_ON(1);
537 			return;
538 		}
539 	}
540 
541 	for (i = 0; i < NR_syscalls; i++) {
542 		addr = arch_syscall_addr(i);
543 		meta = find_syscall_meta(addr);
544 		if (!meta)
545 			continue;
546 
547 		meta->syscall_nr = i;
548 
549 		if (!IS_ENABLED(CONFIG_HAVE_SPARSE_SYSCALL_NR)) {
550 			syscalls_metadata[i] = meta;
551 		} else {
552 			ret = xa_store(&syscalls_metadata_sparse, i, meta,
553 					GFP_KERNEL);
554 			WARN(xa_is_err(ret),
555 				"Syscall memory allocation failed\n");
556 		}
557 
558 	}
559 }
560 
561 #ifdef CONFIG_PERF_EVENTS
562 
563 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
564 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
565 static int sys_perf_refcount_enter;
566 static int sys_perf_refcount_exit;
567 
568 static int perf_call_bpf_enter(struct trace_event_call *call, struct pt_regs *regs,
569 			       struct syscall_metadata *sys_data,
570 			       struct syscall_trace_enter *rec)
571 {
572 	struct syscall_tp_t {
573 		unsigned long long regs;
574 		unsigned long syscall_nr;
575 		unsigned long args[SYSCALL_DEFINE_MAXARGS];
576 	} param;
577 	int i;
578 
579 	*(struct pt_regs **)&param = regs;
580 	param.syscall_nr = rec->nr;
581 	for (i = 0; i < sys_data->nb_args; i++)
582 		param.args[i] = rec->args[i];
583 	return trace_call_bpf(call, &param);
584 }
585 
586 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
587 {
588 	struct syscall_metadata *sys_data;
589 	struct syscall_trace_enter *rec;
590 	struct hlist_head *head;
591 	unsigned long args[6];
592 	bool valid_prog_array;
593 	int syscall_nr;
594 	int rctx;
595 	int size;
596 
597 	syscall_nr = trace_get_syscall_nr(current, regs);
598 	if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
599 		return;
600 	if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
601 		return;
602 
603 	sys_data = syscall_nr_to_meta(syscall_nr);
604 	if (!sys_data)
605 		return;
606 
607 	head = this_cpu_ptr(sys_data->enter_event->perf_events);
608 	valid_prog_array = bpf_prog_array_valid(sys_data->enter_event);
609 	if (!valid_prog_array && hlist_empty(head))
610 		return;
611 
612 	/* get the size after alignment with the u32 buffer size field */
613 	size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
614 	size = ALIGN(size + sizeof(u32), sizeof(u64));
615 	size -= sizeof(u32);
616 
617 	rec = perf_trace_buf_alloc(size, NULL, &rctx);
618 	if (!rec)
619 		return;
620 
621 	rec->nr = syscall_nr;
622 	syscall_get_arguments(current, regs, args);
623 	memcpy(&rec->args, args, sizeof(unsigned long) * sys_data->nb_args);
624 
625 	if ((valid_prog_array &&
626 	     !perf_call_bpf_enter(sys_data->enter_event, regs, sys_data, rec)) ||
627 	    hlist_empty(head)) {
628 		perf_swevent_put_recursion_context(rctx);
629 		return;
630 	}
631 
632 	perf_trace_buf_submit(rec, size, rctx,
633 			      sys_data->enter_event->event.type, 1, regs,
634 			      head, NULL);
635 }
636 
637 static int perf_sysenter_enable(struct trace_event_call *call)
638 {
639 	int ret = 0;
640 	int num;
641 
642 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
643 
644 	mutex_lock(&syscall_trace_lock);
645 	if (!sys_perf_refcount_enter)
646 		ret = register_trace_sys_enter(perf_syscall_enter, NULL);
647 	if (ret) {
648 		pr_info("event trace: Could not activate syscall entry trace point");
649 	} else {
650 		set_bit(num, enabled_perf_enter_syscalls);
651 		sys_perf_refcount_enter++;
652 	}
653 	mutex_unlock(&syscall_trace_lock);
654 	return ret;
655 }
656 
657 static void perf_sysenter_disable(struct trace_event_call *call)
658 {
659 	int num;
660 
661 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
662 
663 	mutex_lock(&syscall_trace_lock);
664 	sys_perf_refcount_enter--;
665 	clear_bit(num, enabled_perf_enter_syscalls);
666 	if (!sys_perf_refcount_enter)
667 		unregister_trace_sys_enter(perf_syscall_enter, NULL);
668 	mutex_unlock(&syscall_trace_lock);
669 }
670 
671 static int perf_call_bpf_exit(struct trace_event_call *call, struct pt_regs *regs,
672 			      struct syscall_trace_exit *rec)
673 {
674 	struct syscall_tp_t {
675 		unsigned long long regs;
676 		unsigned long syscall_nr;
677 		unsigned long ret;
678 	} param;
679 
680 	*(struct pt_regs **)&param = regs;
681 	param.syscall_nr = rec->nr;
682 	param.ret = rec->ret;
683 	return trace_call_bpf(call, &param);
684 }
685 
686 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
687 {
688 	struct syscall_metadata *sys_data;
689 	struct syscall_trace_exit *rec;
690 	struct hlist_head *head;
691 	bool valid_prog_array;
692 	int syscall_nr;
693 	int rctx;
694 	int size;
695 
696 	syscall_nr = trace_get_syscall_nr(current, regs);
697 	if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
698 		return;
699 	if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
700 		return;
701 
702 	sys_data = syscall_nr_to_meta(syscall_nr);
703 	if (!sys_data)
704 		return;
705 
706 	head = this_cpu_ptr(sys_data->exit_event->perf_events);
707 	valid_prog_array = bpf_prog_array_valid(sys_data->exit_event);
708 	if (!valid_prog_array && hlist_empty(head))
709 		return;
710 
711 	/* We can probably do that at build time */
712 	size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
713 	size -= sizeof(u32);
714 
715 	rec = perf_trace_buf_alloc(size, NULL, &rctx);
716 	if (!rec)
717 		return;
718 
719 	rec->nr = syscall_nr;
720 	rec->ret = syscall_get_return_value(current, regs);
721 
722 	if ((valid_prog_array &&
723 	     !perf_call_bpf_exit(sys_data->exit_event, regs, rec)) ||
724 	    hlist_empty(head)) {
725 		perf_swevent_put_recursion_context(rctx);
726 		return;
727 	}
728 
729 	perf_trace_buf_submit(rec, size, rctx, sys_data->exit_event->event.type,
730 			      1, regs, head, NULL);
731 }
732 
733 static int perf_sysexit_enable(struct trace_event_call *call)
734 {
735 	int ret = 0;
736 	int num;
737 
738 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
739 
740 	mutex_lock(&syscall_trace_lock);
741 	if (!sys_perf_refcount_exit)
742 		ret = register_trace_sys_exit(perf_syscall_exit, NULL);
743 	if (ret) {
744 		pr_info("event trace: Could not activate syscall exit trace point");
745 	} else {
746 		set_bit(num, enabled_perf_exit_syscalls);
747 		sys_perf_refcount_exit++;
748 	}
749 	mutex_unlock(&syscall_trace_lock);
750 	return ret;
751 }
752 
753 static void perf_sysexit_disable(struct trace_event_call *call)
754 {
755 	int num;
756 
757 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
758 
759 	mutex_lock(&syscall_trace_lock);
760 	sys_perf_refcount_exit--;
761 	clear_bit(num, enabled_perf_exit_syscalls);
762 	if (!sys_perf_refcount_exit)
763 		unregister_trace_sys_exit(perf_syscall_exit, NULL);
764 	mutex_unlock(&syscall_trace_lock);
765 }
766 
767 #endif /* CONFIG_PERF_EVENTS */
768 
769 static int syscall_enter_register(struct trace_event_call *event,
770 				 enum trace_reg type, void *data)
771 {
772 	struct trace_event_file *file = data;
773 
774 	switch (type) {
775 	case TRACE_REG_REGISTER:
776 		return reg_event_syscall_enter(file, event);
777 	case TRACE_REG_UNREGISTER:
778 		unreg_event_syscall_enter(file, event);
779 		return 0;
780 
781 #ifdef CONFIG_PERF_EVENTS
782 	case TRACE_REG_PERF_REGISTER:
783 		return perf_sysenter_enable(event);
784 	case TRACE_REG_PERF_UNREGISTER:
785 		perf_sysenter_disable(event);
786 		return 0;
787 	case TRACE_REG_PERF_OPEN:
788 	case TRACE_REG_PERF_CLOSE:
789 	case TRACE_REG_PERF_ADD:
790 	case TRACE_REG_PERF_DEL:
791 		return 0;
792 #endif
793 	}
794 	return 0;
795 }
796 
797 static int syscall_exit_register(struct trace_event_call *event,
798 				 enum trace_reg type, void *data)
799 {
800 	struct trace_event_file *file = data;
801 
802 	switch (type) {
803 	case TRACE_REG_REGISTER:
804 		return reg_event_syscall_exit(file, event);
805 	case TRACE_REG_UNREGISTER:
806 		unreg_event_syscall_exit(file, event);
807 		return 0;
808 
809 #ifdef CONFIG_PERF_EVENTS
810 	case TRACE_REG_PERF_REGISTER:
811 		return perf_sysexit_enable(event);
812 	case TRACE_REG_PERF_UNREGISTER:
813 		perf_sysexit_disable(event);
814 		return 0;
815 	case TRACE_REG_PERF_OPEN:
816 	case TRACE_REG_PERF_CLOSE:
817 	case TRACE_REG_PERF_ADD:
818 	case TRACE_REG_PERF_DEL:
819 		return 0;
820 #endif
821 	}
822 	return 0;
823 }
824