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;
321 syscall_nr = trace_get_syscall_nr(current, regs);
322 if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
325 /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE) */
326 trace_file = rcu_dereference_sched(tr->enter_syscall_files[syscall_nr]);
330 if (trace_trigger_soft_disabled(trace_file))
333 sys_data = syscall_nr_to_meta(syscall_nr);
337 size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
339 local_save_flags(irq_flags);
340 pc = preempt_count();
342 buffer = tr->trace_buffer.buffer;
343 event = trace_buffer_lock_reserve(buffer,
344 sys_data->enter_event->event.type, size, irq_flags, pc);
348 entry = ring_buffer_event_data(event);
349 entry->nr = syscall_nr;
350 syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
352 event_trigger_unlock_commit(trace_file, buffer, event, entry,
356 static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
358 struct trace_array *tr = data;
359 struct trace_event_file *trace_file;
360 struct syscall_trace_exit *entry;
361 struct syscall_metadata *sys_data;
362 struct ring_buffer_event *event;
363 struct ring_buffer *buffer;
364 unsigned long irq_flags;
368 syscall_nr = trace_get_syscall_nr(current, regs);
369 if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
372 /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE()) */
373 trace_file = rcu_dereference_sched(tr->exit_syscall_files[syscall_nr]);
377 if (trace_trigger_soft_disabled(trace_file))
380 sys_data = syscall_nr_to_meta(syscall_nr);
384 local_save_flags(irq_flags);
385 pc = preempt_count();
387 buffer = tr->trace_buffer.buffer;
388 event = trace_buffer_lock_reserve(buffer,
389 sys_data->exit_event->event.type, sizeof(*entry),
394 entry = ring_buffer_event_data(event);
395 entry->nr = syscall_nr;
396 entry->ret = syscall_get_return_value(current, regs);
398 event_trigger_unlock_commit(trace_file, buffer, event, entry,
402 static int reg_event_syscall_enter(struct trace_event_file *file,
403 struct trace_event_call *call)
405 struct trace_array *tr = file->tr;
409 num = ((struct syscall_metadata *)call->data)->syscall_nr;
410 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
412 mutex_lock(&syscall_trace_lock);
413 if (!tr->sys_refcount_enter)
414 ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
416 rcu_assign_pointer(tr->enter_syscall_files[num], file);
417 tr->sys_refcount_enter++;
419 mutex_unlock(&syscall_trace_lock);
423 static void unreg_event_syscall_enter(struct trace_event_file *file,
424 struct trace_event_call *call)
426 struct trace_array *tr = file->tr;
429 num = ((struct syscall_metadata *)call->data)->syscall_nr;
430 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
432 mutex_lock(&syscall_trace_lock);
433 tr->sys_refcount_enter--;
434 RCU_INIT_POINTER(tr->enter_syscall_files[num], NULL);
435 if (!tr->sys_refcount_enter)
436 unregister_trace_sys_enter(ftrace_syscall_enter, tr);
437 mutex_unlock(&syscall_trace_lock);
440 static int reg_event_syscall_exit(struct trace_event_file *file,
441 struct trace_event_call *call)
443 struct trace_array *tr = file->tr;
447 num = ((struct syscall_metadata *)call->data)->syscall_nr;
448 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
450 mutex_lock(&syscall_trace_lock);
451 if (!tr->sys_refcount_exit)
452 ret = register_trace_sys_exit(ftrace_syscall_exit, tr);
454 rcu_assign_pointer(tr->exit_syscall_files[num], file);
455 tr->sys_refcount_exit++;
457 mutex_unlock(&syscall_trace_lock);
461 static void unreg_event_syscall_exit(struct trace_event_file *file,
462 struct trace_event_call *call)
464 struct trace_array *tr = file->tr;
467 num = ((struct syscall_metadata *)call->data)->syscall_nr;
468 if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
470 mutex_lock(&syscall_trace_lock);
471 tr->sys_refcount_exit--;
472 RCU_INIT_POINTER(tr->exit_syscall_files[num], NULL);
473 if (!tr->sys_refcount_exit)
474 unregister_trace_sys_exit(ftrace_syscall_exit, tr);
475 mutex_unlock(&syscall_trace_lock);
478 static int __init init_syscall_trace(struct trace_event_call *call)
483 num = ((struct syscall_metadata *)call->data)->syscall_nr;
484 if (num < 0 || num >= NR_syscalls) {
485 pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
486 ((struct syscall_metadata *)call->data)->name);
490 if (set_syscall_print_fmt(call) < 0)
493 id = trace_event_raw_init(call);
496 free_syscall_print_fmt(call);
503 struct trace_event_functions enter_syscall_print_funcs = {
504 .trace = print_syscall_enter,
507 struct trace_event_functions exit_syscall_print_funcs = {
508 .trace = print_syscall_exit,
511 struct trace_event_class __refdata event_class_syscall_enter = {
512 .system = "syscalls",
513 .reg = syscall_enter_register,
514 .define_fields = syscall_enter_define_fields,
515 .get_fields = syscall_get_enter_fields,
516 .raw_init = init_syscall_trace,
519 struct trace_event_class __refdata event_class_syscall_exit = {
520 .system = "syscalls",
521 .reg = syscall_exit_register,
522 .define_fields = syscall_exit_define_fields,
523 .fields = LIST_HEAD_INIT(event_class_syscall_exit.fields),
524 .raw_init = init_syscall_trace,
527 unsigned long __init __weak arch_syscall_addr(int nr)
529 return (unsigned long)sys_call_table[nr];
532 void __init init_ftrace_syscalls(void)
534 struct syscall_metadata *meta;
538 syscalls_metadata = kcalloc(NR_syscalls, sizeof(*syscalls_metadata),
540 if (!syscalls_metadata) {
545 for (i = 0; i < NR_syscalls; i++) {
546 addr = arch_syscall_addr(i);
547 meta = find_syscall_meta(addr);
551 meta->syscall_nr = i;
552 syscalls_metadata[i] = meta;
556 #ifdef CONFIG_PERF_EVENTS
558 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
559 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
560 static int sys_perf_refcount_enter;
561 static int sys_perf_refcount_exit;
563 static int perf_call_bpf_enter(struct bpf_prog *prog, struct pt_regs *regs,
564 struct syscall_metadata *sys_data,
565 struct syscall_trace_enter *rec) {
566 struct syscall_tp_t {
567 unsigned long long regs;
568 unsigned long syscall_nr;
569 unsigned long args[SYSCALL_DEFINE_MAXARGS];
573 *(struct pt_regs **)¶m = regs;
574 param.syscall_nr = rec->nr;
575 for (i = 0; i < sys_data->nb_args; i++)
576 param.args[i] = rec->args[i];
577 return trace_call_bpf(prog, ¶m);
580 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
582 struct syscall_metadata *sys_data;
583 struct syscall_trace_enter *rec;
584 struct hlist_head *head;
585 struct bpf_prog *prog;
590 syscall_nr = trace_get_syscall_nr(current, regs);
591 if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
593 if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
596 sys_data = syscall_nr_to_meta(syscall_nr);
600 prog = READ_ONCE(sys_data->enter_event->prog);
601 head = this_cpu_ptr(sys_data->enter_event->perf_events);
602 if (!prog && hlist_empty(head))
605 /* get the size after alignment with the u32 buffer size field */
606 size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
607 size = ALIGN(size + sizeof(u32), sizeof(u64));
610 rec = perf_trace_buf_alloc(size, NULL, &rctx);
614 rec->nr = syscall_nr;
615 syscall_get_arguments(current, regs, 0, sys_data->nb_args,
616 (unsigned long *)&rec->args);
618 if ((prog && !perf_call_bpf_enter(prog, regs, sys_data, rec)) ||
620 perf_swevent_put_recursion_context(rctx);
624 perf_trace_buf_submit(rec, size, rctx,
625 sys_data->enter_event->event.type, 1, regs,
629 static int perf_sysenter_enable(struct trace_event_call *call)
634 num = ((struct syscall_metadata *)call->data)->syscall_nr;
636 mutex_lock(&syscall_trace_lock);
637 if (!sys_perf_refcount_enter)
638 ret = register_trace_sys_enter(perf_syscall_enter, NULL);
640 pr_info("event trace: Could not activate syscall entry trace point");
642 set_bit(num, enabled_perf_enter_syscalls);
643 sys_perf_refcount_enter++;
645 mutex_unlock(&syscall_trace_lock);
649 static void perf_sysenter_disable(struct trace_event_call *call)
653 num = ((struct syscall_metadata *)call->data)->syscall_nr;
655 mutex_lock(&syscall_trace_lock);
656 sys_perf_refcount_enter--;
657 clear_bit(num, enabled_perf_enter_syscalls);
658 if (!sys_perf_refcount_enter)
659 unregister_trace_sys_enter(perf_syscall_enter, NULL);
660 mutex_unlock(&syscall_trace_lock);
663 static int perf_call_bpf_exit(struct bpf_prog *prog, struct pt_regs *regs,
664 struct syscall_trace_exit *rec) {
665 struct syscall_tp_t {
666 unsigned long long regs;
667 unsigned long syscall_nr;
671 *(struct pt_regs **)¶m = regs;
672 param.syscall_nr = rec->nr;
673 param.ret = rec->ret;
674 return trace_call_bpf(prog, ¶m);
677 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
679 struct syscall_metadata *sys_data;
680 struct syscall_trace_exit *rec;
681 struct hlist_head *head;
682 struct bpf_prog *prog;
687 syscall_nr = trace_get_syscall_nr(current, regs);
688 if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
690 if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
693 sys_data = syscall_nr_to_meta(syscall_nr);
697 prog = READ_ONCE(sys_data->exit_event->prog);
698 head = this_cpu_ptr(sys_data->exit_event->perf_events);
699 if (!prog && hlist_empty(head))
702 /* We can probably do that at build time */
703 size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
706 rec = perf_trace_buf_alloc(size, NULL, &rctx);
710 rec->nr = syscall_nr;
711 rec->ret = syscall_get_return_value(current, regs);
713 if ((prog && !perf_call_bpf_exit(prog, regs, rec)) ||
715 perf_swevent_put_recursion_context(rctx);
719 perf_trace_buf_submit(rec, size, rctx, sys_data->exit_event->event.type,
720 1, regs, head, NULL, NULL);
723 static int perf_sysexit_enable(struct trace_event_call *call)
728 num = ((struct syscall_metadata *)call->data)->syscall_nr;
730 mutex_lock(&syscall_trace_lock);
731 if (!sys_perf_refcount_exit)
732 ret = register_trace_sys_exit(perf_syscall_exit, NULL);
734 pr_info("event trace: Could not activate syscall exit trace point");
736 set_bit(num, enabled_perf_exit_syscalls);
737 sys_perf_refcount_exit++;
739 mutex_unlock(&syscall_trace_lock);
743 static void perf_sysexit_disable(struct trace_event_call *call)
747 num = ((struct syscall_metadata *)call->data)->syscall_nr;
749 mutex_lock(&syscall_trace_lock);
750 sys_perf_refcount_exit--;
751 clear_bit(num, enabled_perf_exit_syscalls);
752 if (!sys_perf_refcount_exit)
753 unregister_trace_sys_exit(perf_syscall_exit, NULL);
754 mutex_unlock(&syscall_trace_lock);
757 #endif /* CONFIG_PERF_EVENTS */
759 static int syscall_enter_register(struct trace_event_call *event,
760 enum trace_reg type, void *data)
762 struct trace_event_file *file = data;
765 case TRACE_REG_REGISTER:
766 return reg_event_syscall_enter(file, event);
767 case TRACE_REG_UNREGISTER:
768 unreg_event_syscall_enter(file, event);
771 #ifdef CONFIG_PERF_EVENTS
772 case TRACE_REG_PERF_REGISTER:
773 return perf_sysenter_enable(event);
774 case TRACE_REG_PERF_UNREGISTER:
775 perf_sysenter_disable(event);
777 case TRACE_REG_PERF_OPEN:
778 case TRACE_REG_PERF_CLOSE:
779 case TRACE_REG_PERF_ADD:
780 case TRACE_REG_PERF_DEL:
787 static int syscall_exit_register(struct trace_event_call *event,
788 enum trace_reg type, void *data)
790 struct trace_event_file *file = data;
793 case TRACE_REG_REGISTER:
794 return reg_event_syscall_exit(file, event);
795 case TRACE_REG_UNREGISTER:
796 unreg_event_syscall_exit(file, event);
799 #ifdef CONFIG_PERF_EVENTS
800 case TRACE_REG_PERF_REGISTER:
801 return perf_sysexit_enable(event);
802 case TRACE_REG_PERF_UNREGISTER:
803 perf_sysexit_disable(event);
805 case TRACE_REG_PERF_OPEN:
806 case TRACE_REG_PERF_CLOSE:
807 case TRACE_REG_PERF_ADD:
808 case TRACE_REG_PERF_DEL: