drm/modes: Fix drm_mode_vrefres() docs
[drm/drm-misc.git] / kernel / trace / trace_kprobe.c
blob263fac44d3ca32ad0aee21c7e65ac744f1034ca7
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Kprobes-based tracing events
5 * Created by Masami Hiramatsu <mhiramat@redhat.com>
7 */
8 #define pr_fmt(fmt) "trace_kprobe: " fmt
10 #include <linux/bpf-cgroup.h>
11 #include <linux/security.h>
12 #include <linux/module.h>
13 #include <linux/uaccess.h>
14 #include <linux/rculist.h>
15 #include <linux/error-injection.h>
17 #include <asm/setup.h> /* for COMMAND_LINE_SIZE */
19 #include "trace_dynevent.h"
20 #include "trace_kprobe_selftest.h"
21 #include "trace_probe.h"
22 #include "trace_probe_tmpl.h"
23 #include "trace_probe_kernel.h"
25 #define KPROBE_EVENT_SYSTEM "kprobes"
26 #define KRETPROBE_MAXACTIVE_MAX 4096
28 /* Kprobe early definition from command line */
29 static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
31 static int __init set_kprobe_boot_events(char *str)
33 strscpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
34 disable_tracing_selftest("running kprobe events");
36 return 1;
38 __setup("kprobe_event=", set_kprobe_boot_events);
40 static int trace_kprobe_create(const char *raw_command);
41 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
42 static int trace_kprobe_release(struct dyn_event *ev);
43 static bool trace_kprobe_is_busy(struct dyn_event *ev);
44 static bool trace_kprobe_match(const char *system, const char *event,
45 int argc, const char **argv, struct dyn_event *ev);
47 static struct dyn_event_operations trace_kprobe_ops = {
48 .create = trace_kprobe_create,
49 .show = trace_kprobe_show,
50 .is_busy = trace_kprobe_is_busy,
51 .free = trace_kprobe_release,
52 .match = trace_kprobe_match,
56 * Kprobe event core functions
58 struct trace_kprobe {
59 struct dyn_event devent;
60 struct kretprobe rp; /* Use rp.kp for kprobe use */
61 unsigned long __percpu *nhit;
62 const char *symbol; /* symbol name */
63 struct trace_probe tp;
66 static bool is_trace_kprobe(struct dyn_event *ev)
68 return ev->ops == &trace_kprobe_ops;
71 static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
73 return container_of(ev, struct trace_kprobe, devent);
76 /**
77 * for_each_trace_kprobe - iterate over the trace_kprobe list
78 * @pos: the struct trace_kprobe * for each entry
79 * @dpos: the struct dyn_event * to use as a loop cursor
81 #define for_each_trace_kprobe(pos, dpos) \
82 for_each_dyn_event(dpos) \
83 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
85 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
87 return tk->rp.handler != NULL;
90 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
92 return tk->symbol ? tk->symbol : "unknown";
95 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
97 return tk->rp.kp.offset;
100 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
102 return kprobe_gone(&tk->rp.kp);
105 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
106 struct module *mod)
108 int len = strlen(module_name(mod));
109 const char *name = trace_kprobe_symbol(tk);
111 return strncmp(module_name(mod), name, len) == 0 && name[len] == ':';
114 #ifdef CONFIG_MODULES
115 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
117 char *p;
118 bool ret;
120 if (!tk->symbol)
121 return false;
122 p = strchr(tk->symbol, ':');
123 if (!p)
124 return true;
125 *p = '\0';
126 rcu_read_lock_sched();
127 ret = !!find_module(tk->symbol);
128 rcu_read_unlock_sched();
129 *p = ':';
131 return ret;
133 #else
134 static inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
136 return false;
138 #endif
140 static bool trace_kprobe_is_busy(struct dyn_event *ev)
142 struct trace_kprobe *tk = to_trace_kprobe(ev);
144 return trace_probe_is_enabled(&tk->tp);
147 static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
148 int argc, const char **argv)
150 char buf[MAX_ARGSTR_LEN + 1];
152 if (!argc)
153 return true;
155 if (!tk->symbol)
156 snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
157 else if (tk->rp.kp.offset)
158 snprintf(buf, sizeof(buf), "%s+%u",
159 trace_kprobe_symbol(tk), tk->rp.kp.offset);
160 else
161 snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
162 if (strcmp(buf, argv[0]))
163 return false;
164 argc--; argv++;
166 return trace_probe_match_command_args(&tk->tp, argc, argv);
169 static bool trace_kprobe_match(const char *system, const char *event,
170 int argc, const char **argv, struct dyn_event *ev)
172 struct trace_kprobe *tk = to_trace_kprobe(ev);
174 return (event[0] == '\0' ||
175 strcmp(trace_probe_name(&tk->tp), event) == 0) &&
176 (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
177 trace_kprobe_match_command_head(tk, argc, argv);
180 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
182 unsigned long nhit = 0;
183 int cpu;
185 for_each_possible_cpu(cpu)
186 nhit += *per_cpu_ptr(tk->nhit, cpu);
188 return nhit;
191 static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
193 return !(list_empty(&tk->rp.kp.list) &&
194 hlist_unhashed(&tk->rp.kp.hlist));
197 /* Return 0 if it fails to find the symbol address */
198 static nokprobe_inline
199 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
201 unsigned long addr;
203 if (tk->symbol) {
204 addr = (unsigned long)
205 kallsyms_lookup_name(trace_kprobe_symbol(tk));
206 if (addr)
207 addr += tk->rp.kp.offset;
208 } else {
209 addr = (unsigned long)tk->rp.kp.addr;
211 return addr;
214 static nokprobe_inline struct trace_kprobe *
215 trace_kprobe_primary_from_call(struct trace_event_call *call)
217 struct trace_probe *tp;
219 tp = trace_probe_primary_from_call(call);
220 if (WARN_ON_ONCE(!tp))
221 return NULL;
223 return container_of(tp, struct trace_kprobe, tp);
226 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
228 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
230 return tk ? (kprobe_on_func_entry(tk->rp.kp.addr,
231 tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
232 tk->rp.kp.addr ? 0 : tk->rp.kp.offset) == 0) : false;
235 bool trace_kprobe_error_injectable(struct trace_event_call *call)
237 struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
239 return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
240 false;
243 static int register_kprobe_event(struct trace_kprobe *tk);
244 static int unregister_kprobe_event(struct trace_kprobe *tk);
246 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
247 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
248 struct pt_regs *regs);
250 static void free_trace_kprobe(struct trace_kprobe *tk)
252 if (tk) {
253 trace_probe_cleanup(&tk->tp);
254 kfree(tk->symbol);
255 free_percpu(tk->nhit);
256 kfree(tk);
261 * Allocate new trace_probe and initialize it (including kprobes).
263 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
264 const char *event,
265 void *addr,
266 const char *symbol,
267 unsigned long offs,
268 int maxactive,
269 int nargs, bool is_return)
271 struct trace_kprobe *tk;
272 int ret = -ENOMEM;
274 tk = kzalloc(struct_size(tk, tp.args, nargs), GFP_KERNEL);
275 if (!tk)
276 return ERR_PTR(ret);
278 tk->nhit = alloc_percpu(unsigned long);
279 if (!tk->nhit)
280 goto error;
282 if (symbol) {
283 tk->symbol = kstrdup(symbol, GFP_KERNEL);
284 if (!tk->symbol)
285 goto error;
286 tk->rp.kp.symbol_name = tk->symbol;
287 tk->rp.kp.offset = offs;
288 } else
289 tk->rp.kp.addr = addr;
291 if (is_return)
292 tk->rp.handler = kretprobe_dispatcher;
293 else
294 tk->rp.kp.pre_handler = kprobe_dispatcher;
296 tk->rp.maxactive = maxactive;
297 INIT_HLIST_NODE(&tk->rp.kp.hlist);
298 INIT_LIST_HEAD(&tk->rp.kp.list);
300 ret = trace_probe_init(&tk->tp, event, group, false, nargs);
301 if (ret < 0)
302 goto error;
304 dyn_event_init(&tk->devent, &trace_kprobe_ops);
305 return tk;
306 error:
307 free_trace_kprobe(tk);
308 return ERR_PTR(ret);
311 static struct trace_kprobe *find_trace_kprobe(const char *event,
312 const char *group)
314 struct dyn_event *pos;
315 struct trace_kprobe *tk;
317 for_each_trace_kprobe(tk, pos)
318 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
319 strcmp(trace_probe_group_name(&tk->tp), group) == 0)
320 return tk;
321 return NULL;
324 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
326 int ret = 0;
328 if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
329 if (trace_kprobe_is_return(tk))
330 ret = enable_kretprobe(&tk->rp);
331 else
332 ret = enable_kprobe(&tk->rp.kp);
335 return ret;
338 static void __disable_trace_kprobe(struct trace_probe *tp)
340 struct trace_kprobe *tk;
342 list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
343 if (!trace_kprobe_is_registered(tk))
344 continue;
345 if (trace_kprobe_is_return(tk))
346 disable_kretprobe(&tk->rp);
347 else
348 disable_kprobe(&tk->rp.kp);
353 * Enable trace_probe
354 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
356 static int enable_trace_kprobe(struct trace_event_call *call,
357 struct trace_event_file *file)
359 struct trace_probe *tp;
360 struct trace_kprobe *tk;
361 bool enabled;
362 int ret = 0;
364 tp = trace_probe_primary_from_call(call);
365 if (WARN_ON_ONCE(!tp))
366 return -ENODEV;
367 enabled = trace_probe_is_enabled(tp);
369 /* This also changes "enabled" state */
370 if (file) {
371 ret = trace_probe_add_file(tp, file);
372 if (ret)
373 return ret;
374 } else
375 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
377 if (enabled)
378 return 0;
380 list_for_each_entry(tk, trace_probe_probe_list(tp), tp.list) {
381 if (trace_kprobe_has_gone(tk))
382 continue;
383 ret = __enable_trace_kprobe(tk);
384 if (ret)
385 break;
386 enabled = true;
389 if (ret) {
390 /* Failed to enable one of them. Roll back all */
391 if (enabled)
392 __disable_trace_kprobe(tp);
393 if (file)
394 trace_probe_remove_file(tp, file);
395 else
396 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
399 return ret;
403 * Disable trace_probe
404 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
406 static int disable_trace_kprobe(struct trace_event_call *call,
407 struct trace_event_file *file)
409 struct trace_probe *tp;
411 tp = trace_probe_primary_from_call(call);
412 if (WARN_ON_ONCE(!tp))
413 return -ENODEV;
415 if (file) {
416 if (!trace_probe_get_file_link(tp, file))
417 return -ENOENT;
418 if (!trace_probe_has_single_file(tp))
419 goto out;
420 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
421 } else
422 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
424 if (!trace_probe_is_enabled(tp))
425 __disable_trace_kprobe(tp);
427 out:
428 if (file)
430 * Synchronization is done in below function. For perf event,
431 * file == NULL and perf_trace_event_unreg() calls
432 * tracepoint_synchronize_unregister() to ensure synchronize
433 * event. We don't need to care about it.
435 trace_probe_remove_file(tp, file);
437 return 0;
440 #if defined(CONFIG_DYNAMIC_FTRACE) && \
441 !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
442 static bool __within_notrace_func(unsigned long addr)
444 unsigned long offset, size;
446 if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
447 return false;
449 /* Get the entry address of the target function */
450 addr -= offset;
453 * Since ftrace_location_range() does inclusive range check, we need
454 * to subtract 1 byte from the end address.
456 return !ftrace_location_range(addr, addr + size - 1);
459 static bool within_notrace_func(struct trace_kprobe *tk)
461 unsigned long addr = trace_kprobe_address(tk);
462 char symname[KSYM_NAME_LEN], *p;
464 if (!__within_notrace_func(addr))
465 return false;
467 /* Check if the address is on a suffixed-symbol */
468 if (!lookup_symbol_name(addr, symname)) {
469 p = strchr(symname, '.');
470 if (!p)
471 return true;
472 *p = '\0';
473 addr = (unsigned long)kprobe_lookup_name(symname, 0);
474 if (addr)
475 return __within_notrace_func(addr);
478 return true;
480 #else
481 #define within_notrace_func(tk) (false)
482 #endif
484 /* Internal register function - just handle k*probes and flags */
485 static int __register_trace_kprobe(struct trace_kprobe *tk)
487 int i, ret;
489 ret = security_locked_down(LOCKDOWN_KPROBES);
490 if (ret)
491 return ret;
493 if (trace_kprobe_is_registered(tk))
494 return -EINVAL;
496 if (within_notrace_func(tk)) {
497 pr_warn("Could not probe notrace function %ps\n",
498 (void *)trace_kprobe_address(tk));
499 return -EINVAL;
502 for (i = 0; i < tk->tp.nr_args; i++) {
503 ret = traceprobe_update_arg(&tk->tp.args[i]);
504 if (ret)
505 return ret;
508 /* Set/clear disabled flag according to tp->flag */
509 if (trace_probe_is_enabled(&tk->tp))
510 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
511 else
512 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
514 if (trace_kprobe_is_return(tk))
515 ret = register_kretprobe(&tk->rp);
516 else
517 ret = register_kprobe(&tk->rp.kp);
519 return ret;
522 /* Internal unregister function - just handle k*probes and flags */
523 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
525 if (trace_kprobe_is_registered(tk)) {
526 if (trace_kprobe_is_return(tk))
527 unregister_kretprobe(&tk->rp);
528 else
529 unregister_kprobe(&tk->rp.kp);
530 /* Cleanup kprobe for reuse and mark it unregistered */
531 INIT_HLIST_NODE(&tk->rp.kp.hlist);
532 INIT_LIST_HEAD(&tk->rp.kp.list);
533 if (tk->rp.kp.symbol_name)
534 tk->rp.kp.addr = NULL;
538 /* Unregister a trace_probe and probe_event */
539 static int unregister_trace_kprobe(struct trace_kprobe *tk)
541 /* If other probes are on the event, just unregister kprobe */
542 if (trace_probe_has_sibling(&tk->tp))
543 goto unreg;
545 /* Enabled event can not be unregistered */
546 if (trace_probe_is_enabled(&tk->tp))
547 return -EBUSY;
549 /* If there's a reference to the dynamic event */
550 if (trace_event_dyn_busy(trace_probe_event_call(&tk->tp)))
551 return -EBUSY;
553 /* Will fail if probe is being used by ftrace or perf */
554 if (unregister_kprobe_event(tk))
555 return -EBUSY;
557 unreg:
558 __unregister_trace_kprobe(tk);
559 dyn_event_remove(&tk->devent);
560 trace_probe_unlink(&tk->tp);
562 return 0;
565 static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
566 struct trace_kprobe *comp)
568 struct trace_probe_event *tpe = orig->tp.event;
569 int i;
571 list_for_each_entry(orig, &tpe->probes, tp.list) {
572 if (strcmp(trace_kprobe_symbol(orig),
573 trace_kprobe_symbol(comp)) ||
574 trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
575 continue;
578 * trace_probe_compare_arg_type() ensured that nr_args and
579 * each argument name and type are same. Let's compare comm.
581 for (i = 0; i < orig->tp.nr_args; i++) {
582 if (strcmp(orig->tp.args[i].comm,
583 comp->tp.args[i].comm))
584 break;
587 if (i == orig->tp.nr_args)
588 return true;
591 return false;
594 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
596 int ret;
598 ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
599 if (ret) {
600 /* Note that argument starts index = 2 */
601 trace_probe_log_set_index(ret + 1);
602 trace_probe_log_err(0, DIFF_ARG_TYPE);
603 return -EEXIST;
605 if (trace_kprobe_has_same_kprobe(to, tk)) {
606 trace_probe_log_set_index(0);
607 trace_probe_log_err(0, SAME_PROBE);
608 return -EEXIST;
611 /* Append to existing event */
612 ret = trace_probe_append(&tk->tp, &to->tp);
613 if (ret)
614 return ret;
616 /* Register k*probe */
617 ret = __register_trace_kprobe(tk);
618 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
619 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
620 ret = 0;
623 if (ret)
624 trace_probe_unlink(&tk->tp);
625 else
626 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
628 return ret;
631 /* Register a trace_probe and probe_event */
632 static int register_trace_kprobe(struct trace_kprobe *tk)
634 struct trace_kprobe *old_tk;
635 int ret;
637 mutex_lock(&event_mutex);
639 old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
640 trace_probe_group_name(&tk->tp));
641 if (old_tk) {
642 if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
643 trace_probe_log_set_index(0);
644 trace_probe_log_err(0, DIFF_PROBE_TYPE);
645 ret = -EEXIST;
646 } else {
647 ret = append_trace_kprobe(tk, old_tk);
649 goto end;
652 /* Register new event */
653 ret = register_kprobe_event(tk);
654 if (ret) {
655 if (ret == -EEXIST) {
656 trace_probe_log_set_index(0);
657 trace_probe_log_err(0, EVENT_EXIST);
658 } else
659 pr_warn("Failed to register probe event(%d)\n", ret);
660 goto end;
663 /* Register k*probe */
664 ret = __register_trace_kprobe(tk);
665 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
666 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
667 ret = 0;
670 if (ret < 0)
671 unregister_kprobe_event(tk);
672 else
673 dyn_event_add(&tk->devent, trace_probe_event_call(&tk->tp));
675 end:
676 mutex_unlock(&event_mutex);
677 return ret;
680 #ifdef CONFIG_MODULES
681 static int validate_module_probe_symbol(const char *modname, const char *symbol);
683 static int register_module_trace_kprobe(struct module *mod, struct trace_kprobe *tk)
685 const char *p;
686 int ret = 0;
688 p = strchr(trace_kprobe_symbol(tk), ':');
689 if (p)
690 ret = validate_module_probe_symbol(module_name(mod), p + 1);
691 if (!ret)
692 ret = __register_trace_kprobe(tk);
693 return ret;
696 /* Module notifier call back, checking event on the module */
697 static int trace_kprobe_module_callback(struct notifier_block *nb,
698 unsigned long val, void *data)
700 struct module *mod = data;
701 struct dyn_event *pos;
702 struct trace_kprobe *tk;
703 int ret;
705 if (val != MODULE_STATE_COMING)
706 return NOTIFY_DONE;
708 /* Update probes on coming module */
709 mutex_lock(&event_mutex);
710 for_each_trace_kprobe(tk, pos) {
711 if (trace_kprobe_within_module(tk, mod)) {
712 /* Don't need to check busy - this should have gone. */
713 __unregister_trace_kprobe(tk);
714 ret = register_module_trace_kprobe(mod, tk);
715 if (ret)
716 pr_warn("Failed to re-register probe %s on %s: %d\n",
717 trace_probe_name(&tk->tp),
718 module_name(mod), ret);
721 mutex_unlock(&event_mutex);
723 return NOTIFY_DONE;
726 static struct notifier_block trace_kprobe_module_nb = {
727 .notifier_call = trace_kprobe_module_callback,
728 .priority = 1 /* Invoked after kprobe module callback */
730 static int trace_kprobe_register_module_notifier(void)
732 return register_module_notifier(&trace_kprobe_module_nb);
734 #else
735 static int trace_kprobe_register_module_notifier(void)
737 return 0;
739 #endif /* CONFIG_MODULES */
741 static int count_symbols(void *data, unsigned long unused)
743 unsigned int *count = data;
745 (*count)++;
747 return 0;
750 struct sym_count_ctx {
751 unsigned int count;
752 const char *name;
755 static int count_mod_symbols(void *data, const char *name, unsigned long unused)
757 struct sym_count_ctx *ctx = data;
759 if (strcmp(name, ctx->name) == 0)
760 ctx->count++;
762 return 0;
765 static unsigned int number_of_same_symbols(const char *mod, const char *func_name)
767 struct sym_count_ctx ctx = { .count = 0, .name = func_name };
769 if (!mod)
770 kallsyms_on_each_match_symbol(count_symbols, func_name, &ctx.count);
772 module_kallsyms_on_each_symbol(mod, count_mod_symbols, &ctx);
774 return ctx.count;
777 static int validate_module_probe_symbol(const char *modname, const char *symbol)
779 unsigned int count = number_of_same_symbols(modname, symbol);
781 if (count > 1) {
783 * Users should use ADDR to remove the ambiguity of
784 * using KSYM only.
786 return -EADDRNOTAVAIL;
787 } else if (count == 0) {
789 * We can return ENOENT earlier than when register the
790 * kprobe.
792 return -ENOENT;
794 return 0;
797 #ifdef CONFIG_MODULES
798 /* Return NULL if the module is not loaded or under unloading. */
799 static struct module *try_module_get_by_name(const char *name)
801 struct module *mod;
803 rcu_read_lock_sched();
804 mod = find_module(name);
805 if (mod && !try_module_get(mod))
806 mod = NULL;
807 rcu_read_unlock_sched();
809 return mod;
811 #else
812 #define try_module_get_by_name(name) (NULL)
813 #endif
815 static int validate_probe_symbol(char *symbol)
817 struct module *mod = NULL;
818 char *modname = NULL, *p;
819 int ret = 0;
821 p = strchr(symbol, ':');
822 if (p) {
823 modname = symbol;
824 symbol = p + 1;
825 *p = '\0';
826 mod = try_module_get_by_name(modname);
827 if (!mod)
828 goto out;
831 ret = validate_module_probe_symbol(modname, symbol);
832 out:
833 if (p)
834 *p = ':';
835 if (mod)
836 module_put(mod);
837 return ret;
840 static int trace_kprobe_entry_handler(struct kretprobe_instance *ri,
841 struct pt_regs *regs);
843 static int __trace_kprobe_create(int argc, const char *argv[])
846 * Argument syntax:
847 * - Add kprobe:
848 * p[:[GRP/][EVENT]] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
849 * - Add kretprobe:
850 * r[MAXACTIVE][:[GRP/][EVENT]] [MOD:]KSYM[+0] [FETCHARGS]
851 * Or
852 * p[:[GRP/][EVENT]] [MOD:]KSYM[+0]%return [FETCHARGS]
854 * Fetch args:
855 * $retval : fetch return value
856 * $stack : fetch stack address
857 * $stackN : fetch Nth of stack (N:0-)
858 * $comm : fetch current task comm
859 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
860 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
861 * %REG : fetch register REG
862 * Dereferencing memory fetch:
863 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
864 * Alias name of args:
865 * NAME=FETCHARG : set NAME as alias of FETCHARG.
866 * Type of args:
867 * FETCHARG:TYPE : use TYPE instead of unsigned long.
869 struct trace_kprobe *tk = NULL;
870 int i, len, new_argc = 0, ret = 0;
871 bool is_return = false;
872 char *symbol = NULL, *tmp = NULL;
873 const char **new_argv = NULL;
874 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
875 enum probe_print_type ptype;
876 int maxactive = 0;
877 long offset = 0;
878 void *addr = NULL;
879 char buf[MAX_EVENT_NAME_LEN];
880 char gbuf[MAX_EVENT_NAME_LEN];
881 char abuf[MAX_BTF_ARGS_LEN];
882 char *dbuf = NULL;
883 struct traceprobe_parse_context ctx = { .flags = TPARG_FL_KERNEL };
885 switch (argv[0][0]) {
886 case 'r':
887 is_return = true;
888 break;
889 case 'p':
890 break;
891 default:
892 return -ECANCELED;
894 if (argc < 2)
895 return -ECANCELED;
897 trace_probe_log_init("trace_kprobe", argc, argv);
899 event = strchr(&argv[0][1], ':');
900 if (event)
901 event++;
903 if (isdigit(argv[0][1])) {
904 if (!is_return) {
905 trace_probe_log_err(1, BAD_MAXACT_TYPE);
906 goto parse_error;
908 if (event)
909 len = event - &argv[0][1] - 1;
910 else
911 len = strlen(&argv[0][1]);
912 if (len > MAX_EVENT_NAME_LEN - 1) {
913 trace_probe_log_err(1, BAD_MAXACT);
914 goto parse_error;
916 memcpy(buf, &argv[0][1], len);
917 buf[len] = '\0';
918 ret = kstrtouint(buf, 0, &maxactive);
919 if (ret || !maxactive) {
920 trace_probe_log_err(1, BAD_MAXACT);
921 goto parse_error;
923 /* kretprobes instances are iterated over via a list. The
924 * maximum should stay reasonable.
926 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
927 trace_probe_log_err(1, MAXACT_TOO_BIG);
928 goto parse_error;
932 /* try to parse an address. if that fails, try to read the
933 * input as a symbol. */
934 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
935 trace_probe_log_set_index(1);
936 /* Check whether uprobe event specified */
937 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
938 ret = -ECANCELED;
939 goto error;
941 /* a symbol specified */
942 symbol = kstrdup(argv[1], GFP_KERNEL);
943 if (!symbol)
944 return -ENOMEM;
946 tmp = strchr(symbol, '%');
947 if (tmp) {
948 if (!strcmp(tmp, "%return")) {
949 *tmp = '\0';
950 is_return = true;
951 } else {
952 trace_probe_log_err(tmp - symbol, BAD_ADDR_SUFFIX);
953 goto parse_error;
957 /* TODO: support .init module functions */
958 ret = traceprobe_split_symbol_offset(symbol, &offset);
959 if (ret || offset < 0 || offset > UINT_MAX) {
960 trace_probe_log_err(0, BAD_PROBE_ADDR);
961 goto parse_error;
963 ret = validate_probe_symbol(symbol);
964 if (ret) {
965 if (ret == -EADDRNOTAVAIL)
966 trace_probe_log_err(0, NON_UNIQ_SYMBOL);
967 else
968 trace_probe_log_err(0, BAD_PROBE_ADDR);
969 goto parse_error;
971 if (is_return)
972 ctx.flags |= TPARG_FL_RETURN;
973 ret = kprobe_on_func_entry(NULL, symbol, offset);
974 if (ret == 0 && !is_return)
975 ctx.flags |= TPARG_FL_FENTRY;
976 /* Defer the ENOENT case until register kprobe */
977 if (ret == -EINVAL && is_return) {
978 trace_probe_log_err(0, BAD_RETPROBE);
979 goto parse_error;
983 trace_probe_log_set_index(0);
984 if (event) {
985 ret = traceprobe_parse_event_name(&event, &group, gbuf,
986 event - argv[0]);
987 if (ret)
988 goto parse_error;
991 if (!event) {
992 /* Make a new event name */
993 if (symbol)
994 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
995 is_return ? 'r' : 'p', symbol, offset);
996 else
997 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
998 is_return ? 'r' : 'p', addr);
999 sanitize_event_name(buf);
1000 event = buf;
1003 argc -= 2; argv += 2;
1004 ctx.funcname = symbol;
1005 new_argv = traceprobe_expand_meta_args(argc, argv, &new_argc,
1006 abuf, MAX_BTF_ARGS_LEN, &ctx);
1007 if (IS_ERR(new_argv)) {
1008 ret = PTR_ERR(new_argv);
1009 new_argv = NULL;
1010 goto out;
1012 if (new_argv) {
1013 argc = new_argc;
1014 argv = new_argv;
1016 if (argc > MAX_TRACE_ARGS) {
1017 ret = -E2BIG;
1018 goto out;
1021 ret = traceprobe_expand_dentry_args(argc, argv, &dbuf);
1022 if (ret)
1023 goto out;
1025 /* setup a probe */
1026 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
1027 argc, is_return);
1028 if (IS_ERR(tk)) {
1029 ret = PTR_ERR(tk);
1030 /* This must return -ENOMEM, else there is a bug */
1031 WARN_ON_ONCE(ret != -ENOMEM);
1032 goto out; /* We know tk is not allocated */
1035 /* parse arguments */
1036 for (i = 0; i < argc; i++) {
1037 trace_probe_log_set_index(i + 2);
1038 ctx.offset = 0;
1039 ret = traceprobe_parse_probe_arg(&tk->tp, i, argv[i], &ctx);
1040 if (ret)
1041 goto error; /* This can be -ENOMEM */
1043 /* entry handler for kretprobe */
1044 if (is_return && tk->tp.entry_arg) {
1045 tk->rp.entry_handler = trace_kprobe_entry_handler;
1046 tk->rp.data_size = traceprobe_get_entry_data_size(&tk->tp);
1049 ptype = is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1050 ret = traceprobe_set_print_fmt(&tk->tp, ptype);
1051 if (ret < 0)
1052 goto error;
1054 ret = register_trace_kprobe(tk);
1055 if (ret) {
1056 trace_probe_log_set_index(1);
1057 if (ret == -EILSEQ)
1058 trace_probe_log_err(0, BAD_INSN_BNDRY);
1059 else if (ret == -ENOENT)
1060 trace_probe_log_err(0, BAD_PROBE_ADDR);
1061 else if (ret != -ENOMEM && ret != -EEXIST)
1062 trace_probe_log_err(0, FAIL_REG_PROBE);
1063 goto error;
1066 out:
1067 traceprobe_finish_parse(&ctx);
1068 trace_probe_log_clear();
1069 kfree(new_argv);
1070 kfree(symbol);
1071 kfree(dbuf);
1072 return ret;
1074 parse_error:
1075 ret = -EINVAL;
1076 error:
1077 free_trace_kprobe(tk);
1078 goto out;
1081 static int trace_kprobe_create(const char *raw_command)
1083 return trace_probe_create(raw_command, __trace_kprobe_create);
1086 static int create_or_delete_trace_kprobe(const char *raw_command)
1088 int ret;
1090 if (raw_command[0] == '-')
1091 return dyn_event_release(raw_command, &trace_kprobe_ops);
1093 ret = trace_kprobe_create(raw_command);
1094 return ret == -ECANCELED ? -EINVAL : ret;
1097 static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
1099 return create_or_delete_trace_kprobe(cmd->seq.buffer);
1103 * kprobe_event_cmd_init - Initialize a kprobe event command object
1104 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1105 * @buf: A pointer to the buffer used to build the command
1106 * @maxlen: The length of the buffer passed in @buf
1108 * Initialize a synthetic event command object. Use this before
1109 * calling any of the other kprobe_event functions.
1111 void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
1113 dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
1114 trace_kprobe_run_command);
1116 EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
1119 * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
1120 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1121 * @kretprobe: Is this a return probe?
1122 * @name: The name of the kprobe event
1123 * @loc: The location of the kprobe event
1124 * @...: Variable number of arg (pairs), one pair for each field
1126 * NOTE: Users normally won't want to call this function directly, but
1127 * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
1128 * adds a NULL to the end of the arg list. If this function is used
1129 * directly, make sure the last arg in the variable arg list is NULL.
1131 * Generate a kprobe event command to be executed by
1132 * kprobe_event_gen_cmd_end(). This function can be used to generate the
1133 * complete command or only the first part of it; in the latter case,
1134 * kprobe_event_add_fields() can be used to add more fields following this.
1136 * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
1137 * returns -EINVAL if @loc == NULL.
1139 * Return: 0 if successful, error otherwise.
1141 int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
1142 const char *name, const char *loc, ...)
1144 char buf[MAX_EVENT_NAME_LEN];
1145 struct dynevent_arg arg;
1146 va_list args;
1147 int ret;
1149 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1150 return -EINVAL;
1152 if (!loc)
1153 return -EINVAL;
1155 if (kretprobe)
1156 snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
1157 else
1158 snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
1160 ret = dynevent_str_add(cmd, buf);
1161 if (ret)
1162 return ret;
1164 dynevent_arg_init(&arg, 0);
1165 arg.str = loc;
1166 ret = dynevent_arg_add(cmd, &arg, NULL);
1167 if (ret)
1168 return ret;
1170 va_start(args, loc);
1171 for (;;) {
1172 const char *field;
1174 field = va_arg(args, const char *);
1175 if (!field)
1176 break;
1178 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1179 ret = -EINVAL;
1180 break;
1183 arg.str = field;
1184 ret = dynevent_arg_add(cmd, &arg, NULL);
1185 if (ret)
1186 break;
1188 va_end(args);
1190 return ret;
1192 EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1195 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1196 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1197 * @...: Variable number of arg (pairs), one pair for each field
1199 * NOTE: Users normally won't want to call this function directly, but
1200 * rather use the kprobe_event_add_fields() wrapper, which
1201 * automatically adds a NULL to the end of the arg list. If this
1202 * function is used directly, make sure the last arg in the variable
1203 * arg list is NULL.
1205 * Add probe fields to an existing kprobe command using a variable
1206 * list of args. Fields are added in the same order they're listed.
1208 * Return: 0 if successful, error otherwise.
1210 int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1212 struct dynevent_arg arg;
1213 va_list args;
1214 int ret = 0;
1216 if (cmd->type != DYNEVENT_TYPE_KPROBE)
1217 return -EINVAL;
1219 dynevent_arg_init(&arg, 0);
1221 va_start(args, cmd);
1222 for (;;) {
1223 const char *field;
1225 field = va_arg(args, const char *);
1226 if (!field)
1227 break;
1229 if (++cmd->n_fields > MAX_TRACE_ARGS) {
1230 ret = -EINVAL;
1231 break;
1234 arg.str = field;
1235 ret = dynevent_arg_add(cmd, &arg, NULL);
1236 if (ret)
1237 break;
1239 va_end(args);
1241 return ret;
1243 EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1246 * kprobe_event_delete - Delete a kprobe event
1247 * @name: The name of the kprobe event to delete
1249 * Delete a kprobe event with the give @name from kernel code rather
1250 * than directly from the command line.
1252 * Return: 0 if successful, error otherwise.
1254 int kprobe_event_delete(const char *name)
1256 char buf[MAX_EVENT_NAME_LEN];
1258 snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1260 return create_or_delete_trace_kprobe(buf);
1262 EXPORT_SYMBOL_GPL(kprobe_event_delete);
1264 static int trace_kprobe_release(struct dyn_event *ev)
1266 struct trace_kprobe *tk = to_trace_kprobe(ev);
1267 int ret = unregister_trace_kprobe(tk);
1269 if (!ret)
1270 free_trace_kprobe(tk);
1271 return ret;
1274 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1276 struct trace_kprobe *tk = to_trace_kprobe(ev);
1277 int i;
1279 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1280 if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1281 seq_printf(m, "%d", tk->rp.maxactive);
1282 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1283 trace_probe_name(&tk->tp));
1285 if (!tk->symbol)
1286 seq_printf(m, " 0x%p", tk->rp.kp.addr);
1287 else if (tk->rp.kp.offset)
1288 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1289 tk->rp.kp.offset);
1290 else
1291 seq_printf(m, " %s", trace_kprobe_symbol(tk));
1293 for (i = 0; i < tk->tp.nr_args; i++)
1294 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1295 seq_putc(m, '\n');
1297 return 0;
1300 static int probes_seq_show(struct seq_file *m, void *v)
1302 struct dyn_event *ev = v;
1304 if (!is_trace_kprobe(ev))
1305 return 0;
1307 return trace_kprobe_show(m, ev);
1310 static const struct seq_operations probes_seq_op = {
1311 .start = dyn_event_seq_start,
1312 .next = dyn_event_seq_next,
1313 .stop = dyn_event_seq_stop,
1314 .show = probes_seq_show
1317 static int probes_open(struct inode *inode, struct file *file)
1319 int ret;
1321 ret = security_locked_down(LOCKDOWN_TRACEFS);
1322 if (ret)
1323 return ret;
1325 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1326 ret = dyn_events_release_all(&trace_kprobe_ops);
1327 if (ret < 0)
1328 return ret;
1331 return seq_open(file, &probes_seq_op);
1334 static ssize_t probes_write(struct file *file, const char __user *buffer,
1335 size_t count, loff_t *ppos)
1337 return trace_parse_run_command(file, buffer, count, ppos,
1338 create_or_delete_trace_kprobe);
1341 static const struct file_operations kprobe_events_ops = {
1342 .owner = THIS_MODULE,
1343 .open = probes_open,
1344 .read = seq_read,
1345 .llseek = seq_lseek,
1346 .release = seq_release,
1347 .write = probes_write,
1350 static unsigned long trace_kprobe_missed(struct trace_kprobe *tk)
1352 return trace_kprobe_is_return(tk) ?
1353 tk->rp.kp.nmissed + tk->rp.nmissed : tk->rp.kp.nmissed;
1356 /* Probes profiling interfaces */
1357 static int probes_profile_seq_show(struct seq_file *m, void *v)
1359 struct dyn_event *ev = v;
1360 struct trace_kprobe *tk;
1361 unsigned long nmissed;
1363 if (!is_trace_kprobe(ev))
1364 return 0;
1366 tk = to_trace_kprobe(ev);
1367 nmissed = trace_kprobe_missed(tk);
1368 seq_printf(m, " %-44s %15lu %15lu\n",
1369 trace_probe_name(&tk->tp),
1370 trace_kprobe_nhit(tk),
1371 nmissed);
1373 return 0;
1376 static const struct seq_operations profile_seq_op = {
1377 .start = dyn_event_seq_start,
1378 .next = dyn_event_seq_next,
1379 .stop = dyn_event_seq_stop,
1380 .show = probes_profile_seq_show
1383 static int profile_open(struct inode *inode, struct file *file)
1385 int ret;
1387 ret = security_locked_down(LOCKDOWN_TRACEFS);
1388 if (ret)
1389 return ret;
1391 return seq_open(file, &profile_seq_op);
1394 static const struct file_operations kprobe_profile_ops = {
1395 .owner = THIS_MODULE,
1396 .open = profile_open,
1397 .read = seq_read,
1398 .llseek = seq_lseek,
1399 .release = seq_release,
1402 /* Note that we don't verify it, since the code does not come from user space */
1403 static int
1404 process_fetch_insn(struct fetch_insn *code, void *rec, void *edata,
1405 void *dest, void *base)
1407 struct pt_regs *regs = rec;
1408 unsigned long val;
1409 int ret;
1411 retry:
1412 /* 1st stage: get value from context */
1413 switch (code->op) {
1414 case FETCH_OP_REG:
1415 val = regs_get_register(regs, code->param);
1416 break;
1417 case FETCH_OP_STACK:
1418 val = regs_get_kernel_stack_nth(regs, code->param);
1419 break;
1420 case FETCH_OP_STACKP:
1421 val = kernel_stack_pointer(regs);
1422 break;
1423 case FETCH_OP_RETVAL:
1424 val = regs_return_value(regs);
1425 break;
1426 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1427 case FETCH_OP_ARG:
1428 val = regs_get_kernel_argument(regs, code->param);
1429 break;
1430 case FETCH_OP_EDATA:
1431 val = *(unsigned long *)((unsigned long)edata + code->offset);
1432 break;
1433 #endif
1434 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
1435 code++;
1436 goto retry;
1437 default:
1438 ret = process_common_fetch_insn(code, &val);
1439 if (ret < 0)
1440 return ret;
1442 code++;
1444 return process_fetch_insn_bottom(code, val, dest, base);
1446 NOKPROBE_SYMBOL(process_fetch_insn)
1448 /* Kprobe handler */
1449 static nokprobe_inline void
1450 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1451 struct trace_event_file *trace_file)
1453 struct kprobe_trace_entry_head *entry;
1454 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1455 struct trace_event_buffer fbuffer;
1456 int dsize;
1458 WARN_ON(call != trace_file->event_call);
1460 if (trace_trigger_soft_disabled(trace_file))
1461 return;
1463 dsize = __get_data_size(&tk->tp, regs, NULL);
1465 entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1466 sizeof(*entry) + tk->tp.size + dsize);
1467 if (!entry)
1468 return;
1470 fbuffer.regs = regs;
1471 entry->ip = (unsigned long)tk->rp.kp.addr;
1472 store_trace_args(&entry[1], &tk->tp, regs, NULL, sizeof(*entry), dsize);
1474 trace_event_buffer_commit(&fbuffer);
1477 static void
1478 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1480 struct event_file_link *link;
1482 trace_probe_for_each_link_rcu(link, &tk->tp)
1483 __kprobe_trace_func(tk, regs, link->file);
1485 NOKPROBE_SYMBOL(kprobe_trace_func);
1487 /* Kretprobe handler */
1489 static int trace_kprobe_entry_handler(struct kretprobe_instance *ri,
1490 struct pt_regs *regs)
1492 struct kretprobe *rp = get_kretprobe(ri);
1493 struct trace_kprobe *tk;
1496 * There is a small chance that get_kretprobe(ri) returns NULL when
1497 * the kretprobe is unregister on another CPU between kretprobe's
1498 * trampoline_handler and this function.
1500 if (unlikely(!rp))
1501 return -ENOENT;
1503 tk = container_of(rp, struct trace_kprobe, rp);
1505 /* store argument values into ri->data as entry data */
1506 if (tk->tp.entry_arg)
1507 store_trace_entry_data(ri->data, &tk->tp, regs);
1509 return 0;
1513 static nokprobe_inline void
1514 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1515 struct pt_regs *regs,
1516 struct trace_event_file *trace_file)
1518 struct kretprobe_trace_entry_head *entry;
1519 struct trace_event_buffer fbuffer;
1520 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1521 int dsize;
1523 WARN_ON(call != trace_file->event_call);
1525 if (trace_trigger_soft_disabled(trace_file))
1526 return;
1528 dsize = __get_data_size(&tk->tp, regs, ri->data);
1530 entry = trace_event_buffer_reserve(&fbuffer, trace_file,
1531 sizeof(*entry) + tk->tp.size + dsize);
1532 if (!entry)
1533 return;
1535 fbuffer.regs = regs;
1536 entry->func = (unsigned long)tk->rp.kp.addr;
1537 entry->ret_ip = get_kretprobe_retaddr(ri);
1538 store_trace_args(&entry[1], &tk->tp, regs, ri->data, sizeof(*entry), dsize);
1540 trace_event_buffer_commit(&fbuffer);
1543 static void
1544 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1545 struct pt_regs *regs)
1547 struct event_file_link *link;
1549 trace_probe_for_each_link_rcu(link, &tk->tp)
1550 __kretprobe_trace_func(tk, ri, regs, link->file);
1552 NOKPROBE_SYMBOL(kretprobe_trace_func);
1554 /* Event entry printers */
1555 static enum print_line_t
1556 print_kprobe_event(struct trace_iterator *iter, int flags,
1557 struct trace_event *event)
1559 struct kprobe_trace_entry_head *field;
1560 struct trace_seq *s = &iter->seq;
1561 struct trace_probe *tp;
1563 field = (struct kprobe_trace_entry_head *)iter->ent;
1564 tp = trace_probe_primary_from_call(
1565 container_of(event, struct trace_event_call, event));
1566 if (WARN_ON_ONCE(!tp))
1567 goto out;
1569 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1571 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1572 goto out;
1574 trace_seq_putc(s, ')');
1576 if (trace_probe_print_args(s, tp->args, tp->nr_args,
1577 (u8 *)&field[1], field) < 0)
1578 goto out;
1580 trace_seq_putc(s, '\n');
1581 out:
1582 return trace_handle_return(s);
1585 static enum print_line_t
1586 print_kretprobe_event(struct trace_iterator *iter, int flags,
1587 struct trace_event *event)
1589 struct kretprobe_trace_entry_head *field;
1590 struct trace_seq *s = &iter->seq;
1591 struct trace_probe *tp;
1593 field = (struct kretprobe_trace_entry_head *)iter->ent;
1594 tp = trace_probe_primary_from_call(
1595 container_of(event, struct trace_event_call, event));
1596 if (WARN_ON_ONCE(!tp))
1597 goto out;
1599 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1601 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1602 goto out;
1604 trace_seq_puts(s, " <- ");
1606 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1607 goto out;
1609 trace_seq_putc(s, ')');
1611 if (trace_probe_print_args(s, tp->args, tp->nr_args,
1612 (u8 *)&field[1], field) < 0)
1613 goto out;
1615 trace_seq_putc(s, '\n');
1617 out:
1618 return trace_handle_return(s);
1622 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1624 int ret;
1625 struct kprobe_trace_entry_head field;
1626 struct trace_probe *tp;
1628 tp = trace_probe_primary_from_call(event_call);
1629 if (WARN_ON_ONCE(!tp))
1630 return -ENOENT;
1632 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1634 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1637 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1639 int ret;
1640 struct kretprobe_trace_entry_head field;
1641 struct trace_probe *tp;
1643 tp = trace_probe_primary_from_call(event_call);
1644 if (WARN_ON_ONCE(!tp))
1645 return -ENOENT;
1647 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1648 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1650 return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1653 #ifdef CONFIG_PERF_EVENTS
1655 /* Kprobe profile handler */
1656 static int
1657 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1659 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1660 struct kprobe_trace_entry_head *entry;
1661 struct hlist_head *head;
1662 int size, __size, dsize;
1663 int rctx;
1665 if (bpf_prog_array_valid(call)) {
1666 unsigned long orig_ip = instruction_pointer(regs);
1667 int ret;
1669 ret = trace_call_bpf(call, regs);
1672 * We need to check and see if we modified the pc of the
1673 * pt_regs, and if so return 1 so that we don't do the
1674 * single stepping.
1676 if (orig_ip != instruction_pointer(regs))
1677 return 1;
1678 if (!ret)
1679 return 0;
1682 head = this_cpu_ptr(call->perf_events);
1683 if (hlist_empty(head))
1684 return 0;
1686 dsize = __get_data_size(&tk->tp, regs, NULL);
1687 __size = sizeof(*entry) + tk->tp.size + dsize;
1688 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1689 size -= sizeof(u32);
1691 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1692 if (!entry)
1693 return 0;
1695 entry->ip = (unsigned long)tk->rp.kp.addr;
1696 memset(&entry[1], 0, dsize);
1697 store_trace_args(&entry[1], &tk->tp, regs, NULL, sizeof(*entry), dsize);
1698 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1699 head, NULL);
1700 return 0;
1702 NOKPROBE_SYMBOL(kprobe_perf_func);
1704 /* Kretprobe profile handler */
1705 static void
1706 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1707 struct pt_regs *regs)
1709 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1710 struct kretprobe_trace_entry_head *entry;
1711 struct hlist_head *head;
1712 int size, __size, dsize;
1713 int rctx;
1715 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1716 return;
1718 head = this_cpu_ptr(call->perf_events);
1719 if (hlist_empty(head))
1720 return;
1722 dsize = __get_data_size(&tk->tp, regs, ri->data);
1723 __size = sizeof(*entry) + tk->tp.size + dsize;
1724 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1725 size -= sizeof(u32);
1727 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1728 if (!entry)
1729 return;
1731 entry->func = (unsigned long)tk->rp.kp.addr;
1732 entry->ret_ip = get_kretprobe_retaddr(ri);
1733 store_trace_args(&entry[1], &tk->tp, regs, ri->data, sizeof(*entry), dsize);
1734 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1735 head, NULL);
1737 NOKPROBE_SYMBOL(kretprobe_perf_func);
1739 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1740 const char **symbol, u64 *probe_offset,
1741 u64 *probe_addr, unsigned long *missed,
1742 bool perf_type_tracepoint)
1744 const char *pevent = trace_event_name(event->tp_event);
1745 const char *group = event->tp_event->class->system;
1746 struct trace_kprobe *tk;
1748 if (perf_type_tracepoint)
1749 tk = find_trace_kprobe(pevent, group);
1750 else
1751 tk = trace_kprobe_primary_from_call(event->tp_event);
1752 if (!tk)
1753 return -EINVAL;
1755 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1756 : BPF_FD_TYPE_KPROBE;
1757 *probe_offset = tk->rp.kp.offset;
1758 *probe_addr = kallsyms_show_value(current_cred()) ?
1759 (unsigned long)tk->rp.kp.addr : 0;
1760 *symbol = tk->symbol;
1761 if (missed)
1762 *missed = trace_kprobe_missed(tk);
1763 return 0;
1765 #endif /* CONFIG_PERF_EVENTS */
1768 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1770 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1771 * lockless, but we can't race with this __init function.
1773 static int kprobe_register(struct trace_event_call *event,
1774 enum trace_reg type, void *data)
1776 struct trace_event_file *file = data;
1778 switch (type) {
1779 case TRACE_REG_REGISTER:
1780 return enable_trace_kprobe(event, file);
1781 case TRACE_REG_UNREGISTER:
1782 return disable_trace_kprobe(event, file);
1784 #ifdef CONFIG_PERF_EVENTS
1785 case TRACE_REG_PERF_REGISTER:
1786 return enable_trace_kprobe(event, NULL);
1787 case TRACE_REG_PERF_UNREGISTER:
1788 return disable_trace_kprobe(event, NULL);
1789 case TRACE_REG_PERF_OPEN:
1790 case TRACE_REG_PERF_CLOSE:
1791 case TRACE_REG_PERF_ADD:
1792 case TRACE_REG_PERF_DEL:
1793 return 0;
1794 #endif
1796 return 0;
1799 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1801 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1802 int ret = 0;
1804 raw_cpu_inc(*tk->nhit);
1806 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1807 kprobe_trace_func(tk, regs);
1808 #ifdef CONFIG_PERF_EVENTS
1809 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1810 ret = kprobe_perf_func(tk, regs);
1811 #endif
1812 return ret;
1814 NOKPROBE_SYMBOL(kprobe_dispatcher);
1816 static int
1817 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1819 struct kretprobe *rp = get_kretprobe(ri);
1820 struct trace_kprobe *tk;
1823 * There is a small chance that get_kretprobe(ri) returns NULL when
1824 * the kretprobe is unregister on another CPU between kretprobe's
1825 * trampoline_handler and this function.
1827 if (unlikely(!rp))
1828 return 0;
1830 tk = container_of(rp, struct trace_kprobe, rp);
1831 raw_cpu_inc(*tk->nhit);
1833 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1834 kretprobe_trace_func(tk, ri, regs);
1835 #ifdef CONFIG_PERF_EVENTS
1836 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1837 kretprobe_perf_func(tk, ri, regs);
1838 #endif
1839 return 0; /* We don't tweak kernel, so just return 0 */
1841 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1843 static struct trace_event_functions kretprobe_funcs = {
1844 .trace = print_kretprobe_event
1847 static struct trace_event_functions kprobe_funcs = {
1848 .trace = print_kprobe_event
1851 static struct trace_event_fields kretprobe_fields_array[] = {
1852 { .type = TRACE_FUNCTION_TYPE,
1853 .define_fields = kretprobe_event_define_fields },
1857 static struct trace_event_fields kprobe_fields_array[] = {
1858 { .type = TRACE_FUNCTION_TYPE,
1859 .define_fields = kprobe_event_define_fields },
1863 static inline void init_trace_event_call(struct trace_kprobe *tk)
1865 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1867 if (trace_kprobe_is_return(tk)) {
1868 call->event.funcs = &kretprobe_funcs;
1869 call->class->fields_array = kretprobe_fields_array;
1870 } else {
1871 call->event.funcs = &kprobe_funcs;
1872 call->class->fields_array = kprobe_fields_array;
1875 call->flags = TRACE_EVENT_FL_KPROBE;
1876 call->class->reg = kprobe_register;
1879 static int register_kprobe_event(struct trace_kprobe *tk)
1881 init_trace_event_call(tk);
1883 return trace_probe_register_event_call(&tk->tp);
1886 static int unregister_kprobe_event(struct trace_kprobe *tk)
1888 return trace_probe_unregister_event_call(&tk->tp);
1891 #ifdef CONFIG_PERF_EVENTS
1893 /* create a trace_kprobe, but don't add it to global lists */
1894 struct trace_event_call *
1895 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1896 bool is_return)
1898 enum probe_print_type ptype;
1899 struct trace_kprobe *tk;
1900 int ret;
1901 char *event;
1903 if (func) {
1904 ret = validate_probe_symbol(func);
1905 if (ret)
1906 return ERR_PTR(ret);
1910 * local trace_kprobes are not added to dyn_event, so they are never
1911 * searched in find_trace_kprobe(). Therefore, there is no concern of
1912 * duplicated name here.
1914 event = func ? func : "DUMMY_EVENT";
1916 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1917 offs, 0 /* maxactive */, 0 /* nargs */,
1918 is_return);
1920 if (IS_ERR(tk)) {
1921 pr_info("Failed to allocate trace_probe.(%d)\n",
1922 (int)PTR_ERR(tk));
1923 return ERR_CAST(tk);
1926 init_trace_event_call(tk);
1928 ptype = trace_kprobe_is_return(tk) ?
1929 PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1930 if (traceprobe_set_print_fmt(&tk->tp, ptype) < 0) {
1931 ret = -ENOMEM;
1932 goto error;
1935 ret = __register_trace_kprobe(tk);
1936 if (ret < 0)
1937 goto error;
1939 return trace_probe_event_call(&tk->tp);
1940 error:
1941 free_trace_kprobe(tk);
1942 return ERR_PTR(ret);
1945 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1947 struct trace_kprobe *tk;
1949 tk = trace_kprobe_primary_from_call(event_call);
1950 if (unlikely(!tk))
1951 return;
1953 if (trace_probe_is_enabled(&tk->tp)) {
1954 WARN_ON(1);
1955 return;
1958 __unregister_trace_kprobe(tk);
1960 free_trace_kprobe(tk);
1962 #endif /* CONFIG_PERF_EVENTS */
1964 static __init void enable_boot_kprobe_events(void)
1966 struct trace_array *tr = top_trace_array();
1967 struct trace_event_file *file;
1968 struct trace_kprobe *tk;
1969 struct dyn_event *pos;
1971 mutex_lock(&event_mutex);
1972 for_each_trace_kprobe(tk, pos) {
1973 list_for_each_entry(file, &tr->events, list)
1974 if (file->event_call == trace_probe_event_call(&tk->tp))
1975 trace_event_enable_disable(file, 1, 0);
1977 mutex_unlock(&event_mutex);
1980 static __init void setup_boot_kprobe_events(void)
1982 char *p, *cmd = kprobe_boot_events_buf;
1983 int ret;
1985 strreplace(kprobe_boot_events_buf, ',', ' ');
1987 while (cmd && *cmd != '\0') {
1988 p = strchr(cmd, ';');
1989 if (p)
1990 *p++ = '\0';
1992 ret = create_or_delete_trace_kprobe(cmd);
1993 if (ret)
1994 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1996 cmd = p;
1999 enable_boot_kprobe_events();
2003 * Register dynevent at core_initcall. This allows kernel to setup kprobe
2004 * events in postcore_initcall without tracefs.
2006 static __init int init_kprobe_trace_early(void)
2008 int ret;
2010 ret = dyn_event_register(&trace_kprobe_ops);
2011 if (ret)
2012 return ret;
2014 if (trace_kprobe_register_module_notifier())
2015 return -EINVAL;
2017 return 0;
2019 core_initcall(init_kprobe_trace_early);
2021 /* Make a tracefs interface for controlling probe points */
2022 static __init int init_kprobe_trace(void)
2024 int ret;
2026 ret = tracing_init_dentry();
2027 if (ret)
2028 return 0;
2030 /* Event list interface */
2031 trace_create_file("kprobe_events", TRACE_MODE_WRITE,
2032 NULL, NULL, &kprobe_events_ops);
2034 /* Profile interface */
2035 trace_create_file("kprobe_profile", TRACE_MODE_READ,
2036 NULL, NULL, &kprobe_profile_ops);
2038 setup_boot_kprobe_events();
2040 return 0;
2042 fs_initcall(init_kprobe_trace);
2045 #ifdef CONFIG_FTRACE_STARTUP_TEST
2046 static __init struct trace_event_file *
2047 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
2049 struct trace_event_file *file;
2051 list_for_each_entry(file, &tr->events, list)
2052 if (file->event_call == trace_probe_event_call(&tk->tp))
2053 return file;
2055 return NULL;
2059 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
2060 * stage, we can do this lockless.
2062 static __init int kprobe_trace_self_tests_init(void)
2064 int ret, warn = 0;
2065 int (*target)(int, int, int, int, int, int);
2066 struct trace_kprobe *tk;
2067 struct trace_event_file *file;
2069 if (tracing_is_disabled())
2070 return -ENODEV;
2072 if (tracing_selftest_disabled)
2073 return 0;
2075 target = kprobe_trace_selftest_target;
2077 pr_info("Testing kprobe tracing: ");
2079 ret = create_or_delete_trace_kprobe("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)");
2080 if (WARN_ONCE(ret, "error on probing function entry.")) {
2081 warn++;
2082 } else {
2083 /* Enable trace point */
2084 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2085 if (WARN_ONCE(tk == NULL, "error on probing function entry.")) {
2086 warn++;
2087 } else {
2088 file = find_trace_probe_file(tk, top_trace_array());
2089 if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2090 warn++;
2091 } else
2092 enable_trace_kprobe(
2093 trace_probe_event_call(&tk->tp), file);
2097 ret = create_or_delete_trace_kprobe("r:testprobe2 kprobe_trace_selftest_target $retval");
2098 if (WARN_ONCE(ret, "error on probing function return.")) {
2099 warn++;
2100 } else {
2101 /* Enable trace point */
2102 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2103 if (WARN_ONCE(tk == NULL, "error on getting 2nd new probe.")) {
2104 warn++;
2105 } else {
2106 file = find_trace_probe_file(tk, top_trace_array());
2107 if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2108 warn++;
2109 } else
2110 enable_trace_kprobe(
2111 trace_probe_event_call(&tk->tp), file);
2115 if (warn)
2116 goto end;
2118 ret = target(1, 2, 3, 4, 5, 6);
2121 * Not expecting an error here, the check is only to prevent the
2122 * optimizer from removing the call to target() as otherwise there
2123 * are no side-effects and the call is never performed.
2125 if (ret != 21)
2126 warn++;
2128 /* Disable trace points before removing it */
2129 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2130 if (WARN_ONCE(tk == NULL, "error on getting test probe.")) {
2131 warn++;
2132 } else {
2133 if (WARN_ONCE(trace_kprobe_nhit(tk) != 1,
2134 "incorrect number of testprobe hits."))
2135 warn++;
2137 file = find_trace_probe_file(tk, top_trace_array());
2138 if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2139 warn++;
2140 } else
2141 disable_trace_kprobe(
2142 trace_probe_event_call(&tk->tp), file);
2145 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2146 if (WARN_ONCE(tk == NULL, "error on getting 2nd test probe.")) {
2147 warn++;
2148 } else {
2149 if (WARN_ONCE(trace_kprobe_nhit(tk) != 1,
2150 "incorrect number of testprobe2 hits."))
2151 warn++;
2153 file = find_trace_probe_file(tk, top_trace_array());
2154 if (WARN_ONCE(file == NULL, "error on getting probe file.")) {
2155 warn++;
2156 } else
2157 disable_trace_kprobe(
2158 trace_probe_event_call(&tk->tp), file);
2161 ret = create_or_delete_trace_kprobe("-:testprobe");
2162 if (WARN_ONCE(ret, "error on deleting a probe."))
2163 warn++;
2165 ret = create_or_delete_trace_kprobe("-:testprobe2");
2166 if (WARN_ONCE(ret, "error on deleting a probe."))
2167 warn++;
2170 end:
2172 * Wait for the optimizer work to finish. Otherwise it might fiddle
2173 * with probes in already freed __init text.
2175 wait_for_kprobe_optimizer();
2176 if (warn)
2177 pr_cont("NG: Some tests are failed. Please check them.\n");
2178 else
2179 pr_cont("OK\n");
2180 return 0;
2183 late_initcall(kprobe_trace_self_tests_init);
2185 #endif