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 <asm/syscall.h>
12 #include "trace_output.h"
15 static DEFINE_MUTEX(syscall_trace_lock);
17 static int syscall_enter_register(struct trace_event_call *event,
18 enum trace_reg type, void *data);
19 static int syscall_exit_register(struct trace_event_call *event,
20 enum trace_reg type, void *data);
22 static struct list_head *
23 syscall_get_enter_fields(struct trace_event_call *call)
25 struct syscall_metadata *entry = call->data;
27 return &entry->enter_fields;
30 extern struct syscall_metadata *__start_syscalls_metadata[];
31 extern struct syscall_metadata *__stop_syscalls_metadata[];
33 static struct syscall_metadata **syscalls_metadata;
35 #ifndef ARCH_HAS_SYSCALL_MATCH_SYM_NAME
36 static inline bool arch_syscall_match_sym_name(const char *sym, const char *name)
39 * Only compare after the "sys" prefix. Archs that use
40 * syscall wrappers may have syscalls symbols aliases prefixed
41 * with ".SyS" or ".sys" instead of "sys", leading to an unwanted
44 return !strcmp(sym + 3, name + 3);
48 #ifdef ARCH_TRACE_IGNORE_COMPAT_SYSCALLS
50 * Some architectures that allow for 32bit applications
51 * to run on a 64bit kernel, do not map the syscalls for
52 * the 32bit tasks the same as they do for 64bit tasks.
56 * In such a case, instead of reporting the wrong syscalls,
59 * For an arch to ignore the compat syscalls it needs to
60 * define ARCH_TRACE_IGNORE_COMPAT_SYSCALLS as well as
61 * define the function arch_trace_is_compat_syscall() to let
62 * the tracing system know that it should ignore it.
65 trace_get_syscall_nr(struct task_struct *task, struct pt_regs *regs)
67 if (unlikely(arch_trace_is_compat_syscall(regs)))
70 return syscall_get_nr(task, regs);
74 trace_get_syscall_nr(struct task_struct *task, struct pt_regs *regs)
76 return syscall_get_nr(task, regs);
78 #endif /* ARCH_TRACE_IGNORE_COMPAT_SYSCALLS */
80 static __init struct syscall_metadata *
81 find_syscall_meta(unsigned long syscall)
83 struct syscall_metadata **start;
84 struct syscall_metadata **stop;
85 char str[KSYM_SYMBOL_LEN];
88 start = __start_syscalls_metadata;
89 stop = __stop_syscalls_metadata;
90 kallsyms_lookup(syscall, NULL, NULL, NULL, str);
92 if (arch_syscall_match_sym_name(str, "sys_ni_syscall"))
95 for ( ; start < stop; start++) {
96 if ((*start)->name && arch_syscall_match_sym_name(str, (*start)->name))
102 static struct syscall_metadata *syscall_nr_to_meta(int nr)
104 if (!syscalls_metadata || nr >= NR_syscalls || nr < 0)
107 return syscalls_metadata[nr];
110 const char *get_syscall_name(int syscall)
112 struct syscall_metadata *entry;
114 entry = syscall_nr_to_meta(syscall);
121 static enum print_line_t
122 print_syscall_enter(struct trace_iterator *iter, int flags,
123 struct trace_event *event)
125 struct trace_array *tr = iter->tr;
126 struct trace_seq *s = &iter->seq;
127 struct trace_entry *ent = iter->ent;
128 struct syscall_trace_enter *trace;
129 struct syscall_metadata *entry;
132 trace = (typeof(trace))ent;
134 entry = syscall_nr_to_meta(syscall);
139 if (entry->enter_event->event.type != ent->type) {
144 trace_seq_printf(s, "%s(", entry->name);
146 for (i = 0; i < entry->nb_args; i++) {
148 if (trace_seq_has_overflowed(s))
151 /* parameter types */
152 if (tr->trace_flags & TRACE_ITER_VERBOSE)
153 trace_seq_printf(s, "%s ", entry->types[i]);
155 /* parameter values */
156 trace_seq_printf(s, "%s: %lx%s", entry->args[i],
158 i == entry->nb_args - 1 ? "" : ", ");
161 trace_seq_putc(s, ')');
163 trace_seq_putc(s, '\n');
165 return trace_handle_return(s);
168 static enum print_line_t
169 print_syscall_exit(struct trace_iterator *iter, int flags,
170 struct trace_event *event)
172 struct trace_seq *s = &iter->seq;
173 struct trace_entry *ent = iter->ent;
174 struct syscall_trace_exit *trace;
176 struct syscall_metadata *entry;
178 trace = (typeof(trace))ent;
180 entry = syscall_nr_to_meta(syscall);
183 trace_seq_putc(s, '\n');
187 if (entry->exit_event->event.type != ent->type) {
189 return TRACE_TYPE_UNHANDLED;
192 trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
196 return trace_handle_return(s);
199 extern char *__bad_type_size(void);
201 #define SYSCALL_FIELD(type, field, name) \
202 sizeof(type) != sizeof(trace.field) ? \
203 __bad_type_size() : \
204 #type, #name, offsetof(typeof(trace), field), \
205 sizeof(trace.field), is_signed_type(type)
208 __set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len)
213 /* When len=0, we just calculate the needed length */
214 #define LEN_OR_ZERO (len ? len - pos : 0)
216 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
217 for (i = 0; i < entry->nb_args; i++) {
218 pos += snprintf(buf + pos, LEN_OR_ZERO, "%s: 0x%%0%zulx%s",
219 entry->args[i], sizeof(unsigned long),
220 i == entry->nb_args - 1 ? "" : ", ");
222 pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
224 for (i = 0; i < entry->nb_args; i++) {
225 pos += snprintf(buf + pos, LEN_OR_ZERO,
226 ", ((unsigned long)(REC->%s))", entry->args[i]);
231 /* return the length of print_fmt */
235 static int __init set_syscall_print_fmt(struct trace_event_call *call)
239 struct syscall_metadata *entry = call->data;
241 if (entry->enter_event != call) {
242 call->print_fmt = "\"0x%lx\", REC->ret";
246 /* First: called with 0 length to calculate the needed length */
247 len = __set_enter_print_fmt(entry, NULL, 0);
249 print_fmt = kmalloc(len + 1, GFP_KERNEL);
253 /* Second: actually write the @print_fmt */
254 __set_enter_print_fmt(entry, print_fmt, len + 1);
255 call->print_fmt = print_fmt;
260 static void __init free_syscall_print_fmt(struct trace_event_call *call)
262 struct syscall_metadata *entry = call->data;
264 if (entry->enter_event == call)
265 kfree(call->print_fmt);
268 static int __init syscall_enter_define_fields(struct trace_event_call *call)
270 struct syscall_trace_enter trace;
271 struct syscall_metadata *meta = call->data;
274 int offset = offsetof(typeof(trace), args);
276 ret = trace_define_field(call, SYSCALL_FIELD(int, nr, __syscall_nr),
281 for (i = 0; i < meta->nb_args; i++) {
282 ret = trace_define_field(call, meta->types[i],
283 meta->args[i], offset,
284 sizeof(unsigned long), 0,
286 offset += sizeof(unsigned long);
292 static int __init syscall_exit_define_fields(struct trace_event_call *call)
294 struct syscall_trace_exit trace;
297 ret = trace_define_field(call, SYSCALL_FIELD(int, nr, __syscall_nr),
302 ret = trace_define_field(call, SYSCALL_FIELD(long, ret, ret),
308 static void ftrace_syscall_enter(void *data, struct pt_regs *regs, long id)
310 struct trace_array *tr = data;
311 struct trace_event_file *trace_file;
312 struct syscall_trace_enter *entry;
313 struct syscall_metadata *sys_data;
314 struct ring_buffer_event *event;
315 struct ring_buffer *buffer;
316 unsigned long irq_flags;
317 unsigned long args[6];
322 syscall_nr = trace_get_syscall_nr(current, regs);
323 if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
326 /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE) */
327 trace_file = rcu_dereference_sched(tr->enter_syscall_files[syscall_nr]);
331 if (trace_trigger_soft_disabled(trace_file))
334 sys_data = syscall_nr_to_meta(syscall_nr);
338 size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
340 local_save_flags(irq_flags);
341 pc = preempt_count();
343 buffer = tr->trace_buffer.buffer;
344 event = trace_buffer_lock_reserve(buffer,
345 sys_data->enter_event->event.type, size, irq_flags, pc);
349 entry = ring_buffer_event_data(event);
350 entry->nr = syscall_nr;
351 syscall_get_arguments(current, regs, args);
352 memcpy(entry->args, args, sizeof(unsigned long) * sys_data->nb_args);
354 event_trigger_unlock_commit(trace_file, buffer, event, entry,
358 static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
360 struct trace_array *tr = data;
361 struct trace_event_file *trace_file;
362 struct syscall_trace_exit *entry;
363 struct syscall_metadata *sys_data;
364 struct ring_buffer_event *event;
365 struct ring_buffer *buffer;
366 unsigned long irq_flags;
370 syscall_nr = trace_get_syscall_nr(current, regs);
371 if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
374 /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE()) */
375 trace_file = rcu_dereference_sched(tr->exit_syscall_files[syscall_nr]);
379 if (trace_trigger_soft_disabled(trace_file))
382 sys_data = syscall_nr_to_meta(syscall_nr);
386 local_save_flags(irq_flags);
387 pc = preempt_count();
389 buffer = tr->trace_buffer.buffer;
390 event = trace_buffer_lock_reserve(buffer,
391 sys_data->exit_event->event.type, sizeof(*entry),
396 entry = ring_buffer_event_data(event);
397 entry->nr = syscall_nr;
398 entry->ret = syscall_get_return_value(current, regs);
400 event_trigger_unlock_commit(trace_file, buffer, event, entry,
404 static int reg_event_syscall_enter(struct trace_event_file *file,
405 struct trace_event_call *call)
407 struct trace_array *tr = file->tr;
411 num = ((struct syscall_metadata *)call->data)->syscall_nr;
412 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
414 mutex_lock(&syscall_trace_lock);
415 if (!tr->sys_refcount_enter)
416 ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
418 rcu_assign_pointer(tr->enter_syscall_files[num], file);
419 tr->sys_refcount_enter++;
421 mutex_unlock(&syscall_trace_lock);
425 static void unreg_event_syscall_enter(struct trace_event_file *file,
426 struct trace_event_call *call)
428 struct trace_array *tr = file->tr;
431 num = ((struct syscall_metadata *)call->data)->syscall_nr;
432 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
434 mutex_lock(&syscall_trace_lock);
435 tr->sys_refcount_enter--;
436 RCU_INIT_POINTER(tr->enter_syscall_files[num], NULL);
437 if (!tr->sys_refcount_enter)
438 unregister_trace_sys_enter(ftrace_syscall_enter, tr);
439 mutex_unlock(&syscall_trace_lock);
442 static int reg_event_syscall_exit(struct trace_event_file *file,
443 struct trace_event_call *call)
445 struct trace_array *tr = file->tr;
449 num = ((struct syscall_metadata *)call->data)->syscall_nr;
450 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
452 mutex_lock(&syscall_trace_lock);
453 if (!tr->sys_refcount_exit)
454 ret = register_trace_sys_exit(ftrace_syscall_exit, tr);
456 rcu_assign_pointer(tr->exit_syscall_files[num], file);
457 tr->sys_refcount_exit++;
459 mutex_unlock(&syscall_trace_lock);
463 static void unreg_event_syscall_exit(struct trace_event_file *file,
464 struct trace_event_call *call)
466 struct trace_array *tr = file->tr;
469 num = ((struct syscall_metadata *)call->data)->syscall_nr;
470 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
472 mutex_lock(&syscall_trace_lock);
473 tr->sys_refcount_exit--;
474 RCU_INIT_POINTER(tr->exit_syscall_files[num], NULL);
475 if (!tr->sys_refcount_exit)
476 unregister_trace_sys_exit(ftrace_syscall_exit, tr);
477 mutex_unlock(&syscall_trace_lock);
480 static int __init init_syscall_trace(struct trace_event_call *call)
485 num = ((struct syscall_metadata *)call->data)->syscall_nr;
486 if (num < 0 || num >= NR_syscalls) {
487 pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
488 ((struct syscall_metadata *)call->data)->name);
492 if (set_syscall_print_fmt(call) < 0)
495 id = trace_event_raw_init(call);
498 free_syscall_print_fmt(call);
505 struct trace_event_functions enter_syscall_print_funcs = {
506 .trace = print_syscall_enter,
509 struct trace_event_functions exit_syscall_print_funcs = {
510 .trace = print_syscall_exit,
513 struct trace_event_class __refdata event_class_syscall_enter = {
514 .system = "syscalls",
515 .reg = syscall_enter_register,
516 .define_fields = syscall_enter_define_fields,
517 .get_fields = syscall_get_enter_fields,
518 .raw_init = init_syscall_trace,
521 struct trace_event_class __refdata event_class_syscall_exit = {
522 .system = "syscalls",
523 .reg = syscall_exit_register,
524 .define_fields = syscall_exit_define_fields,
525 .fields = LIST_HEAD_INIT(event_class_syscall_exit.fields),
526 .raw_init = init_syscall_trace,
529 unsigned long __init __weak arch_syscall_addr(int nr)
531 return (unsigned long)sys_call_table[nr];
534 void __init init_ftrace_syscalls(void)
536 struct syscall_metadata *meta;
540 syscalls_metadata = kcalloc(NR_syscalls, sizeof(*syscalls_metadata),
542 if (!syscalls_metadata) {
547 for (i = 0; i < NR_syscalls; i++) {
548 addr = arch_syscall_addr(i);
549 meta = find_syscall_meta(addr);
553 meta->syscall_nr = i;
554 syscalls_metadata[i] = meta;
558 #ifdef CONFIG_PERF_EVENTS
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;
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)
569 struct syscall_tp_t {
570 unsigned long long regs;
571 unsigned long syscall_nr;
572 unsigned long args[SYSCALL_DEFINE_MAXARGS];
576 *(struct pt_regs **)¶m = regs;
577 param.syscall_nr = rec->nr;
578 for (i = 0; i < sys_data->nb_args; i++)
579 param.args[i] = rec->args[i];
580 return trace_call_bpf(call, ¶m);
583 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
585 struct syscall_metadata *sys_data;
586 struct syscall_trace_enter *rec;
587 struct hlist_head *head;
588 unsigned long args[6];
589 bool valid_prog_array;
594 syscall_nr = trace_get_syscall_nr(current, regs);
595 if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
597 if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
600 sys_data = syscall_nr_to_meta(syscall_nr);
604 head = this_cpu_ptr(sys_data->enter_event->perf_events);
605 valid_prog_array = bpf_prog_array_valid(sys_data->enter_event);
606 if (!valid_prog_array && hlist_empty(head))
609 /* get the size after alignment with the u32 buffer size field */
610 size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
611 size = ALIGN(size + sizeof(u32), sizeof(u64));
614 rec = perf_trace_buf_alloc(size, NULL, &rctx);
618 rec->nr = syscall_nr;
619 syscall_get_arguments(current, regs, args);
620 memcpy(&rec->args, args, sizeof(unsigned long) * sys_data->nb_args);
622 if ((valid_prog_array &&
623 !perf_call_bpf_enter(sys_data->enter_event, regs, sys_data, rec)) ||
625 perf_swevent_put_recursion_context(rctx);
629 perf_trace_buf_submit(rec, size, rctx,
630 sys_data->enter_event->event.type, 1, regs,
634 static int perf_sysenter_enable(struct trace_event_call *call)
639 num = ((struct syscall_metadata *)call->data)->syscall_nr;
641 mutex_lock(&syscall_trace_lock);
642 if (!sys_perf_refcount_enter)
643 ret = register_trace_sys_enter(perf_syscall_enter, NULL);
645 pr_info("event trace: Could not activate syscall entry trace point");
647 set_bit(num, enabled_perf_enter_syscalls);
648 sys_perf_refcount_enter++;
650 mutex_unlock(&syscall_trace_lock);
654 static void perf_sysenter_disable(struct trace_event_call *call)
658 num = ((struct syscall_metadata *)call->data)->syscall_nr;
660 mutex_lock(&syscall_trace_lock);
661 sys_perf_refcount_enter--;
662 clear_bit(num, enabled_perf_enter_syscalls);
663 if (!sys_perf_refcount_enter)
664 unregister_trace_sys_enter(perf_syscall_enter, NULL);
665 mutex_unlock(&syscall_trace_lock);
668 static int perf_call_bpf_exit(struct trace_event_call *call, struct pt_regs *regs,
669 struct syscall_trace_exit *rec)
671 struct syscall_tp_t {
672 unsigned long long regs;
673 unsigned long syscall_nr;
677 *(struct pt_regs **)¶m = regs;
678 param.syscall_nr = rec->nr;
679 param.ret = rec->ret;
680 return trace_call_bpf(call, ¶m);
683 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
685 struct syscall_metadata *sys_data;
686 struct syscall_trace_exit *rec;
687 struct hlist_head *head;
688 bool valid_prog_array;
693 syscall_nr = trace_get_syscall_nr(current, regs);
694 if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
696 if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
699 sys_data = syscall_nr_to_meta(syscall_nr);
703 head = this_cpu_ptr(sys_data->exit_event->perf_events);
704 valid_prog_array = bpf_prog_array_valid(sys_data->exit_event);
705 if (!valid_prog_array && hlist_empty(head))
708 /* We can probably do that at build time */
709 size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
712 rec = perf_trace_buf_alloc(size, NULL, &rctx);
716 rec->nr = syscall_nr;
717 rec->ret = syscall_get_return_value(current, regs);
719 if ((valid_prog_array &&
720 !perf_call_bpf_exit(sys_data->exit_event, regs, rec)) ||
722 perf_swevent_put_recursion_context(rctx);
726 perf_trace_buf_submit(rec, size, rctx, sys_data->exit_event->event.type,
727 1, regs, head, NULL);
730 static int perf_sysexit_enable(struct trace_event_call *call)
735 num = ((struct syscall_metadata *)call->data)->syscall_nr;
737 mutex_lock(&syscall_trace_lock);
738 if (!sys_perf_refcount_exit)
739 ret = register_trace_sys_exit(perf_syscall_exit, NULL);
741 pr_info("event trace: Could not activate syscall exit trace point");
743 set_bit(num, enabled_perf_exit_syscalls);
744 sys_perf_refcount_exit++;
746 mutex_unlock(&syscall_trace_lock);
750 static void perf_sysexit_disable(struct trace_event_call *call)
754 num = ((struct syscall_metadata *)call->data)->syscall_nr;
756 mutex_lock(&syscall_trace_lock);
757 sys_perf_refcount_exit--;
758 clear_bit(num, enabled_perf_exit_syscalls);
759 if (!sys_perf_refcount_exit)
760 unregister_trace_sys_exit(perf_syscall_exit, NULL);
761 mutex_unlock(&syscall_trace_lock);
764 #endif /* CONFIG_PERF_EVENTS */
766 static int syscall_enter_register(struct trace_event_call *event,
767 enum trace_reg type, void *data)
769 struct trace_event_file *file = data;
772 case TRACE_REG_REGISTER:
773 return reg_event_syscall_enter(file, event);
774 case TRACE_REG_UNREGISTER:
775 unreg_event_syscall_enter(file, event);
778 #ifdef CONFIG_PERF_EVENTS
779 case TRACE_REG_PERF_REGISTER:
780 return perf_sysenter_enable(event);
781 case TRACE_REG_PERF_UNREGISTER:
782 perf_sysenter_disable(event);
784 case TRACE_REG_PERF_OPEN:
785 case TRACE_REG_PERF_CLOSE:
786 case TRACE_REG_PERF_ADD:
787 case TRACE_REG_PERF_DEL:
794 static int syscall_exit_register(struct trace_event_call *event,
795 enum trace_reg type, void *data)
797 struct trace_event_file *file = data;
800 case TRACE_REG_REGISTER:
801 return reg_event_syscall_exit(file, event);
802 case TRACE_REG_UNREGISTER:
803 unreg_event_syscall_exit(file, event);
806 #ifdef CONFIG_PERF_EVENTS
807 case TRACE_REG_PERF_REGISTER:
808 return perf_sysexit_enable(event);
809 case TRACE_REG_PERF_UNREGISTER:
810 perf_sysexit_disable(event);
812 case TRACE_REG_PERF_OPEN:
813 case TRACE_REG_PERF_CLOSE:
814 case TRACE_REG_PERF_ADD:
815 case TRACE_REG_PERF_DEL: