xref: /openbmc/linux/kernel/trace/trace_syscalls.c (revision 278002edb19bce2c628fafb0af936e77000f3a5b)
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 *
syscall_get_enter_fields(struct trace_event_call * call)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
arch_syscall_match_sym_name(const char * sym,const char * 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
trace_get_syscall_nr(struct task_struct * task,struct pt_regs * regs)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
trace_get_syscall_nr(struct task_struct * task,struct pt_regs * regs)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 *
find_syscall_meta(unsigned long syscall)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 
syscall_nr_to_meta(int nr)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 
get_syscall_name(int syscall)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
print_syscall_enter(struct trace_iterator * iter,int flags,struct trace_event * event)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
print_syscall_exit(struct trace_iterator * iter,int flags,struct trace_event * event)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 #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 
209 static int __init
__set_enter_print_fmt(struct syscall_metadata * entry,char * buf,int len)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 
set_syscall_print_fmt(struct trace_event_call * call)237 static 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 
free_syscall_print_fmt(struct trace_event_call * call)262 static 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 
syscall_enter_define_fields(struct trace_event_call * call)270 static 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 
ftrace_syscall_enter(void * data,struct pt_regs * regs,long id)291 static 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 	/*
303 	 * Syscall probe called with preemption enabled, but the ring
304 	 * buffer and per-cpu data require preemption to be disabled.
305 	 */
306 	guard(preempt_notrace)();
307 
308 	syscall_nr = trace_get_syscall_nr(current, regs);
309 	if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
310 		return;
311 
312 	/* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE) */
313 	trace_file = rcu_dereference_sched(tr->enter_syscall_files[syscall_nr]);
314 	if (!trace_file)
315 		return;
316 
317 	if (trace_trigger_soft_disabled(trace_file))
318 		return;
319 
320 	sys_data = syscall_nr_to_meta(syscall_nr);
321 	if (!sys_data)
322 		return;
323 
324 	size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
325 
326 	entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
327 	if (!entry)
328 		return;
329 
330 	entry = ring_buffer_event_data(fbuffer.event);
331 	entry->nr = syscall_nr;
332 	syscall_get_arguments(current, regs, args);
333 	memcpy(entry->args, args, sizeof(unsigned long) * sys_data->nb_args);
334 
335 	trace_event_buffer_commit(&fbuffer);
336 }
337 
ftrace_syscall_exit(void * data,struct pt_regs * regs,long ret)338 static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
339 {
340 	struct trace_array *tr = data;
341 	struct trace_event_file *trace_file;
342 	struct syscall_trace_exit *entry;
343 	struct syscall_metadata *sys_data;
344 	struct trace_event_buffer fbuffer;
345 	int syscall_nr;
346 
347 	/*
348 	 * Syscall probe called with preemption enabled, but the ring
349 	 * buffer and per-cpu data require preemption to be disabled.
350 	 */
351 	guard(preempt_notrace)();
352 
353 	syscall_nr = trace_get_syscall_nr(current, regs);
354 	if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
355 		return;
356 
357 	/* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE()) */
358 	trace_file = rcu_dereference_sched(tr->exit_syscall_files[syscall_nr]);
359 	if (!trace_file)
360 		return;
361 
362 	if (trace_trigger_soft_disabled(trace_file))
363 		return;
364 
365 	sys_data = syscall_nr_to_meta(syscall_nr);
366 	if (!sys_data)
367 		return;
368 
369 	entry = trace_event_buffer_reserve(&fbuffer, trace_file, sizeof(*entry));
370 	if (!entry)
371 		return;
372 
373 	entry = ring_buffer_event_data(fbuffer.event);
374 	entry->nr = syscall_nr;
375 	entry->ret = syscall_get_return_value(current, regs);
376 
377 	trace_event_buffer_commit(&fbuffer);
378 }
379 
reg_event_syscall_enter(struct trace_event_file * file,struct trace_event_call * call)380 static int reg_event_syscall_enter(struct trace_event_file *file,
381 				   struct trace_event_call *call)
382 {
383 	struct trace_array *tr = file->tr;
384 	int ret = 0;
385 	int num;
386 
387 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
388 	if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
389 		return -ENOSYS;
390 	mutex_lock(&syscall_trace_lock);
391 	if (!tr->sys_refcount_enter)
392 		ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
393 	if (!ret) {
394 		rcu_assign_pointer(tr->enter_syscall_files[num], file);
395 		tr->sys_refcount_enter++;
396 	}
397 	mutex_unlock(&syscall_trace_lock);
398 	return ret;
399 }
400 
unreg_event_syscall_enter(struct trace_event_file * file,struct trace_event_call * call)401 static void unreg_event_syscall_enter(struct trace_event_file *file,
402 				      struct trace_event_call *call)
403 {
404 	struct trace_array *tr = file->tr;
405 	int num;
406 
407 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
408 	if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
409 		return;
410 	mutex_lock(&syscall_trace_lock);
411 	tr->sys_refcount_enter--;
412 	RCU_INIT_POINTER(tr->enter_syscall_files[num], NULL);
413 	if (!tr->sys_refcount_enter)
414 		unregister_trace_sys_enter(ftrace_syscall_enter, tr);
415 	mutex_unlock(&syscall_trace_lock);
416 }
417 
reg_event_syscall_exit(struct trace_event_file * file,struct trace_event_call * call)418 static int reg_event_syscall_exit(struct trace_event_file *file,
419 				  struct trace_event_call *call)
420 {
421 	struct trace_array *tr = file->tr;
422 	int ret = 0;
423 	int num;
424 
425 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
426 	if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
427 		return -ENOSYS;
428 	mutex_lock(&syscall_trace_lock);
429 	if (!tr->sys_refcount_exit)
430 		ret = register_trace_sys_exit(ftrace_syscall_exit, tr);
431 	if (!ret) {
432 		rcu_assign_pointer(tr->exit_syscall_files[num], file);
433 		tr->sys_refcount_exit++;
434 	}
435 	mutex_unlock(&syscall_trace_lock);
436 	return ret;
437 }
438 
unreg_event_syscall_exit(struct trace_event_file * file,struct trace_event_call * call)439 static void unreg_event_syscall_exit(struct trace_event_file *file,
440 				     struct trace_event_call *call)
441 {
442 	struct trace_array *tr = file->tr;
443 	int num;
444 
445 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
446 	if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
447 		return;
448 	mutex_lock(&syscall_trace_lock);
449 	tr->sys_refcount_exit--;
450 	RCU_INIT_POINTER(tr->exit_syscall_files[num], NULL);
451 	if (!tr->sys_refcount_exit)
452 		unregister_trace_sys_exit(ftrace_syscall_exit, tr);
453 	mutex_unlock(&syscall_trace_lock);
454 }
455 
init_syscall_trace(struct trace_event_call * call)456 static int __init init_syscall_trace(struct trace_event_call *call)
457 {
458 	int id;
459 	int num;
460 
461 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
462 	if (num < 0 || num >= NR_syscalls) {
463 		pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
464 				((struct syscall_metadata *)call->data)->name);
465 		return -ENOSYS;
466 	}
467 
468 	if (set_syscall_print_fmt(call) < 0)
469 		return -ENOMEM;
470 
471 	id = trace_event_raw_init(call);
472 
473 	if (id < 0) {
474 		free_syscall_print_fmt(call);
475 		return id;
476 	}
477 
478 	return id;
479 }
480 
481 static struct trace_event_fields __refdata syscall_enter_fields_array[] = {
482 	SYSCALL_FIELD(int, __syscall_nr),
483 	{ .type = TRACE_FUNCTION_TYPE,
484 	  .define_fields = syscall_enter_define_fields },
485 	{}
486 };
487 
488 struct trace_event_functions enter_syscall_print_funcs = {
489 	.trace		= print_syscall_enter,
490 };
491 
492 struct trace_event_functions exit_syscall_print_funcs = {
493 	.trace		= print_syscall_exit,
494 };
495 
496 struct trace_event_class __refdata event_class_syscall_enter = {
497 	.system		= "syscalls",
498 	.reg		= syscall_enter_register,
499 	.fields_array	= syscall_enter_fields_array,
500 	.get_fields	= syscall_get_enter_fields,
501 	.raw_init	= init_syscall_trace,
502 };
503 
504 struct trace_event_class __refdata event_class_syscall_exit = {
505 	.system		= "syscalls",
506 	.reg		= syscall_exit_register,
507 	.fields_array	= (struct trace_event_fields[]){
508 		SYSCALL_FIELD(int, __syscall_nr),
509 		SYSCALL_FIELD(long, ret),
510 		{}
511 	},
512 	.fields		= LIST_HEAD_INIT(event_class_syscall_exit.fields),
513 	.raw_init	= init_syscall_trace,
514 };
515 
arch_syscall_addr(int nr)516 unsigned long __init __weak arch_syscall_addr(int nr)
517 {
518 	return (unsigned long)sys_call_table[nr];
519 }
520 
init_ftrace_syscalls(void)521 void __init init_ftrace_syscalls(void)
522 {
523 	struct syscall_metadata *meta;
524 	unsigned long addr;
525 	int i;
526 	void *ret;
527 
528 	if (!IS_ENABLED(CONFIG_HAVE_SPARSE_SYSCALL_NR)) {
529 		syscalls_metadata = kcalloc(NR_syscalls,
530 					sizeof(*syscalls_metadata),
531 					GFP_KERNEL);
532 		if (!syscalls_metadata) {
533 			WARN_ON(1);
534 			return;
535 		}
536 	}
537 
538 	for (i = 0; i < NR_syscalls; i++) {
539 		addr = arch_syscall_addr(i);
540 		meta = find_syscall_meta(addr);
541 		if (!meta)
542 			continue;
543 
544 		meta->syscall_nr = i;
545 
546 		if (!IS_ENABLED(CONFIG_HAVE_SPARSE_SYSCALL_NR)) {
547 			syscalls_metadata[i] = meta;
548 		} else {
549 			ret = xa_store(&syscalls_metadata_sparse, i, meta,
550 					GFP_KERNEL);
551 			WARN(xa_is_err(ret),
552 				"Syscall memory allocation failed\n");
553 		}
554 
555 	}
556 }
557 
558 #ifdef CONFIG_PERF_EVENTS
559 
560 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
561 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
562 static int sys_perf_refcount_enter;
563 static int sys_perf_refcount_exit;
564 
perf_call_bpf_enter(struct trace_event_call * call,struct pt_regs * regs,struct syscall_metadata * sys_data,struct syscall_trace_enter * rec)565 static int perf_call_bpf_enter(struct trace_event_call *call, struct pt_regs *regs,
566 			       struct syscall_metadata *sys_data,
567 			       struct syscall_trace_enter *rec)
568 {
569 	struct syscall_tp_t {
570 		struct trace_entry ent;
571 		unsigned long syscall_nr;
572 		unsigned long args[SYSCALL_DEFINE_MAXARGS];
573 	} __aligned(8) param;
574 	int i;
575 
576 	BUILD_BUG_ON(sizeof(param.ent) < sizeof(void *));
577 
578 	/* bpf prog requires 'regs' to be the first member in the ctx (a.k.a. &param) */
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 
perf_syscall_enter(void * ignore,struct pt_regs * regs,long id)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 
perf_sysenter_enable(struct trace_event_call * call)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 
perf_sysenter_disable(struct trace_event_call * call)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 
perf_call_bpf_exit(struct trace_event_call * call,struct pt_regs * regs,struct syscall_trace_exit * rec)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 		struct trace_entry ent;
676 		unsigned long syscall_nr;
677 		unsigned long ret;
678 	} __aligned(8) param;
679 
680 	/* bpf prog requires 'regs' to be the first member in the ctx (a.k.a. &param) */
681 	*(struct pt_regs **)&param = regs;
682 	param.syscall_nr = rec->nr;
683 	param.ret = rec->ret;
684 	return trace_call_bpf(call, &param);
685 }
686 
perf_syscall_exit(void * ignore,struct pt_regs * regs,long ret)687 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
688 {
689 	struct syscall_metadata *sys_data;
690 	struct syscall_trace_exit *rec;
691 	struct hlist_head *head;
692 	bool valid_prog_array;
693 	int syscall_nr;
694 	int rctx;
695 	int size;
696 
697 	syscall_nr = trace_get_syscall_nr(current, regs);
698 	if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
699 		return;
700 	if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
701 		return;
702 
703 	sys_data = syscall_nr_to_meta(syscall_nr);
704 	if (!sys_data)
705 		return;
706 
707 	head = this_cpu_ptr(sys_data->exit_event->perf_events);
708 	valid_prog_array = bpf_prog_array_valid(sys_data->exit_event);
709 	if (!valid_prog_array && hlist_empty(head))
710 		return;
711 
712 	/* We can probably do that at build time */
713 	size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
714 	size -= sizeof(u32);
715 
716 	rec = perf_trace_buf_alloc(size, NULL, &rctx);
717 	if (!rec)
718 		return;
719 
720 	rec->nr = syscall_nr;
721 	rec->ret = syscall_get_return_value(current, regs);
722 
723 	if ((valid_prog_array &&
724 	     !perf_call_bpf_exit(sys_data->exit_event, regs, rec)) ||
725 	    hlist_empty(head)) {
726 		perf_swevent_put_recursion_context(rctx);
727 		return;
728 	}
729 
730 	perf_trace_buf_submit(rec, size, rctx, sys_data->exit_event->event.type,
731 			      1, regs, head, NULL);
732 }
733 
perf_sysexit_enable(struct trace_event_call * call)734 static int perf_sysexit_enable(struct trace_event_call *call)
735 {
736 	int ret = 0;
737 	int num;
738 
739 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
740 
741 	mutex_lock(&syscall_trace_lock);
742 	if (!sys_perf_refcount_exit)
743 		ret = register_trace_sys_exit(perf_syscall_exit, NULL);
744 	if (ret) {
745 		pr_info("event trace: Could not activate syscall exit trace point");
746 	} else {
747 		set_bit(num, enabled_perf_exit_syscalls);
748 		sys_perf_refcount_exit++;
749 	}
750 	mutex_unlock(&syscall_trace_lock);
751 	return ret;
752 }
753 
perf_sysexit_disable(struct trace_event_call * call)754 static void perf_sysexit_disable(struct trace_event_call *call)
755 {
756 	int num;
757 
758 	num = ((struct syscall_metadata *)call->data)->syscall_nr;
759 
760 	mutex_lock(&syscall_trace_lock);
761 	sys_perf_refcount_exit--;
762 	clear_bit(num, enabled_perf_exit_syscalls);
763 	if (!sys_perf_refcount_exit)
764 		unregister_trace_sys_exit(perf_syscall_exit, NULL);
765 	mutex_unlock(&syscall_trace_lock);
766 }
767 
768 #endif /* CONFIG_PERF_EVENTS */
769 
syscall_enter_register(struct trace_event_call * event,enum trace_reg type,void * data)770 static int syscall_enter_register(struct trace_event_call *event,
771 				 enum trace_reg type, void *data)
772 {
773 	struct trace_event_file *file = data;
774 
775 	switch (type) {
776 	case TRACE_REG_REGISTER:
777 		return reg_event_syscall_enter(file, event);
778 	case TRACE_REG_UNREGISTER:
779 		unreg_event_syscall_enter(file, event);
780 		return 0;
781 
782 #ifdef CONFIG_PERF_EVENTS
783 	case TRACE_REG_PERF_REGISTER:
784 		return perf_sysenter_enable(event);
785 	case TRACE_REG_PERF_UNREGISTER:
786 		perf_sysenter_disable(event);
787 		return 0;
788 	case TRACE_REG_PERF_OPEN:
789 	case TRACE_REG_PERF_CLOSE:
790 	case TRACE_REG_PERF_ADD:
791 	case TRACE_REG_PERF_DEL:
792 		return 0;
793 #endif
794 	}
795 	return 0;
796 }
797 
syscall_exit_register(struct trace_event_call * event,enum trace_reg type,void * data)798 static int syscall_exit_register(struct trace_event_call *event,
799 				 enum trace_reg type, void *data)
800 {
801 	struct trace_event_file *file = data;
802 
803 	switch (type) {
804 	case TRACE_REG_REGISTER:
805 		return reg_event_syscall_exit(file, event);
806 	case TRACE_REG_UNREGISTER:
807 		unreg_event_syscall_exit(file, event);
808 		return 0;
809 
810 #ifdef CONFIG_PERF_EVENTS
811 	case TRACE_REG_PERF_REGISTER:
812 		return perf_sysexit_enable(event);
813 	case TRACE_REG_PERF_UNREGISTER:
814 		perf_sysexit_disable(event);
815 		return 0;
816 	case TRACE_REG_PERF_OPEN:
817 	case TRACE_REG_PERF_CLOSE:
818 	case TRACE_REG_PERF_ADD:
819 	case TRACE_REG_PERF_DEL:
820 		return 0;
821 #endif
822 	}
823 	return 0;
824 }
825