xref: /openbmc/linux/kernel/trace/trace_kprobe.c (revision 82b5d164)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Kprobes-based tracing events
4  *
5  * Created by Masami Hiramatsu <mhiramat@redhat.com>
6  *
7  */
8 #define pr_fmt(fmt)	"trace_kprobe: " fmt
9 
10 #include <linux/module.h>
11 #include <linux/uaccess.h>
12 #include <linux/rculist.h>
13 #include <linux/error-injection.h>
14 #include <linux/security.h>
15 
16 #include <asm/setup.h>  /* for COMMAND_LINE_SIZE */
17 
18 #include "trace_dynevent.h"
19 #include "trace_kprobe_selftest.h"
20 #include "trace_probe.h"
21 #include "trace_probe_tmpl.h"
22 
23 #define KPROBE_EVENT_SYSTEM "kprobes"
24 #define KRETPROBE_MAXACTIVE_MAX 4096
25 #define MAX_KPROBE_CMDLINE_SIZE 1024
26 
27 /* Kprobe early definition from command line */
28 static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
29 static bool kprobe_boot_events_enabled __initdata;
30 
31 static int __init set_kprobe_boot_events(char *str)
32 {
33 	strlcpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
34 	return 0;
35 }
36 __setup("kprobe_event=", set_kprobe_boot_events);
37 
38 static int trace_kprobe_create(int argc, const char **argv);
39 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
40 static int trace_kprobe_release(struct dyn_event *ev);
41 static bool trace_kprobe_is_busy(struct dyn_event *ev);
42 static bool trace_kprobe_match(const char *system, const char *event,
43 			int argc, const char **argv, struct dyn_event *ev);
44 
45 static struct dyn_event_operations trace_kprobe_ops = {
46 	.create = trace_kprobe_create,
47 	.show = trace_kprobe_show,
48 	.is_busy = trace_kprobe_is_busy,
49 	.free = trace_kprobe_release,
50 	.match = trace_kprobe_match,
51 };
52 
53 /*
54  * Kprobe event core functions
55  */
56 struct trace_kprobe {
57 	struct dyn_event	devent;
58 	struct kretprobe	rp;	/* Use rp.kp for kprobe use */
59 	unsigned long __percpu *nhit;
60 	const char		*symbol;	/* symbol name */
61 	struct trace_probe	tp;
62 };
63 
64 static bool is_trace_kprobe(struct dyn_event *ev)
65 {
66 	return ev->ops == &trace_kprobe_ops;
67 }
68 
69 static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
70 {
71 	return container_of(ev, struct trace_kprobe, devent);
72 }
73 
74 /**
75  * for_each_trace_kprobe - iterate over the trace_kprobe list
76  * @pos:	the struct trace_kprobe * for each entry
77  * @dpos:	the struct dyn_event * to use as a loop cursor
78  */
79 #define for_each_trace_kprobe(pos, dpos)	\
80 	for_each_dyn_event(dpos)		\
81 		if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
82 
83 #define SIZEOF_TRACE_KPROBE(n)				\
84 	(offsetof(struct trace_kprobe, tp.args) +	\
85 	(sizeof(struct probe_arg) * (n)))
86 
87 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
88 {
89 	return tk->rp.handler != NULL;
90 }
91 
92 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
93 {
94 	return tk->symbol ? tk->symbol : "unknown";
95 }
96 
97 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
98 {
99 	return tk->rp.kp.offset;
100 }
101 
102 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
103 {
104 	return !!(kprobe_gone(&tk->rp.kp));
105 }
106 
107 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
108 						 struct module *mod)
109 {
110 	int len = strlen(mod->name);
111 	const char *name = trace_kprobe_symbol(tk);
112 	return strncmp(mod->name, name, len) == 0 && name[len] == ':';
113 }
114 
115 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
116 {
117 	char *p;
118 	bool ret;
119 
120 	if (!tk->symbol)
121 		return false;
122 	p = strchr(tk->symbol, ':');
123 	if (!p)
124 		return true;
125 	*p = '\0';
126 	mutex_lock(&module_mutex);
127 	ret = !!find_module(tk->symbol);
128 	mutex_unlock(&module_mutex);
129 	*p = ':';
130 
131 	return ret;
132 }
133 
134 static bool trace_kprobe_is_busy(struct dyn_event *ev)
135 {
136 	struct trace_kprobe *tk = to_trace_kprobe(ev);
137 
138 	return trace_probe_is_enabled(&tk->tp);
139 }
140 
141 static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
142 					    int argc, const char **argv)
143 {
144 	char buf[MAX_ARGSTR_LEN + 1];
145 
146 	if (!argc)
147 		return true;
148 
149 	if (!tk->symbol)
150 		snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
151 	else if (tk->rp.kp.offset)
152 		snprintf(buf, sizeof(buf), "%s+%u",
153 			 trace_kprobe_symbol(tk), tk->rp.kp.offset);
154 	else
155 		snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
156 	if (strcmp(buf, argv[0]))
157 		return false;
158 	argc--; argv++;
159 
160 	return trace_probe_match_command_args(&tk->tp, argc, argv);
161 }
162 
163 static bool trace_kprobe_match(const char *system, const char *event,
164 			int argc, const char **argv, struct dyn_event *ev)
165 {
166 	struct trace_kprobe *tk = to_trace_kprobe(ev);
167 
168 	return strcmp(trace_probe_name(&tk->tp), event) == 0 &&
169 	    (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
170 	    trace_kprobe_match_command_head(tk, argc, argv);
171 }
172 
173 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
174 {
175 	unsigned long nhit = 0;
176 	int cpu;
177 
178 	for_each_possible_cpu(cpu)
179 		nhit += *per_cpu_ptr(tk->nhit, cpu);
180 
181 	return nhit;
182 }
183 
184 static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
185 {
186 	return !(list_empty(&tk->rp.kp.list) &&
187 		 hlist_unhashed(&tk->rp.kp.hlist));
188 }
189 
190 /* Return 0 if it fails to find the symbol address */
191 static nokprobe_inline
192 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
193 {
194 	unsigned long addr;
195 
196 	if (tk->symbol) {
197 		addr = (unsigned long)
198 			kallsyms_lookup_name(trace_kprobe_symbol(tk));
199 		if (addr)
200 			addr += tk->rp.kp.offset;
201 	} else {
202 		addr = (unsigned long)tk->rp.kp.addr;
203 	}
204 	return addr;
205 }
206 
207 static nokprobe_inline struct trace_kprobe *
208 trace_kprobe_primary_from_call(struct trace_event_call *call)
209 {
210 	struct trace_probe *tp;
211 
212 	tp = trace_probe_primary_from_call(call);
213 	if (WARN_ON_ONCE(!tp))
214 		return NULL;
215 
216 	return container_of(tp, struct trace_kprobe, tp);
217 }
218 
219 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
220 {
221 	struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
222 
223 	return tk ? kprobe_on_func_entry(tk->rp.kp.addr,
224 			tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
225 			tk->rp.kp.addr ? 0 : tk->rp.kp.offset) : false;
226 }
227 
228 bool trace_kprobe_error_injectable(struct trace_event_call *call)
229 {
230 	struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
231 
232 	return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
233 	       false;
234 }
235 
236 static int register_kprobe_event(struct trace_kprobe *tk);
237 static int unregister_kprobe_event(struct trace_kprobe *tk);
238 
239 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
240 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
241 				struct pt_regs *regs);
242 
243 static void free_trace_kprobe(struct trace_kprobe *tk)
244 {
245 	if (tk) {
246 		trace_probe_cleanup(&tk->tp);
247 		kfree(tk->symbol);
248 		free_percpu(tk->nhit);
249 		kfree(tk);
250 	}
251 }
252 
253 /*
254  * Allocate new trace_probe and initialize it (including kprobes).
255  */
256 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
257 					     const char *event,
258 					     void *addr,
259 					     const char *symbol,
260 					     unsigned long offs,
261 					     int maxactive,
262 					     int nargs, bool is_return)
263 {
264 	struct trace_kprobe *tk;
265 	int ret = -ENOMEM;
266 
267 	tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
268 	if (!tk)
269 		return ERR_PTR(ret);
270 
271 	tk->nhit = alloc_percpu(unsigned long);
272 	if (!tk->nhit)
273 		goto error;
274 
275 	if (symbol) {
276 		tk->symbol = kstrdup(symbol, GFP_KERNEL);
277 		if (!tk->symbol)
278 			goto error;
279 		tk->rp.kp.symbol_name = tk->symbol;
280 		tk->rp.kp.offset = offs;
281 	} else
282 		tk->rp.kp.addr = addr;
283 
284 	if (is_return)
285 		tk->rp.handler = kretprobe_dispatcher;
286 	else
287 		tk->rp.kp.pre_handler = kprobe_dispatcher;
288 
289 	tk->rp.maxactive = maxactive;
290 	INIT_HLIST_NODE(&tk->rp.kp.hlist);
291 	INIT_LIST_HEAD(&tk->rp.kp.list);
292 
293 	ret = trace_probe_init(&tk->tp, event, group);
294 	if (ret < 0)
295 		goto error;
296 
297 	dyn_event_init(&tk->devent, &trace_kprobe_ops);
298 	return tk;
299 error:
300 	free_trace_kprobe(tk);
301 	return ERR_PTR(ret);
302 }
303 
304 static struct trace_kprobe *find_trace_kprobe(const char *event,
305 					      const char *group)
306 {
307 	struct dyn_event *pos;
308 	struct trace_kprobe *tk;
309 
310 	for_each_trace_kprobe(tk, pos)
311 		if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
312 		    strcmp(trace_probe_group_name(&tk->tp), group) == 0)
313 			return tk;
314 	return NULL;
315 }
316 
317 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
318 {
319 	int ret = 0;
320 
321 	if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
322 		if (trace_kprobe_is_return(tk))
323 			ret = enable_kretprobe(&tk->rp);
324 		else
325 			ret = enable_kprobe(&tk->rp.kp);
326 	}
327 
328 	return ret;
329 }
330 
331 static void __disable_trace_kprobe(struct trace_probe *tp)
332 {
333 	struct trace_probe *pos;
334 	struct trace_kprobe *tk;
335 
336 	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
337 		tk = container_of(pos, struct trace_kprobe, tp);
338 		if (!trace_kprobe_is_registered(tk))
339 			continue;
340 		if (trace_kprobe_is_return(tk))
341 			disable_kretprobe(&tk->rp);
342 		else
343 			disable_kprobe(&tk->rp.kp);
344 	}
345 }
346 
347 /*
348  * Enable trace_probe
349  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
350  */
351 static int enable_trace_kprobe(struct trace_event_call *call,
352 				struct trace_event_file *file)
353 {
354 	struct trace_probe *pos, *tp;
355 	struct trace_kprobe *tk;
356 	bool enabled;
357 	int ret = 0;
358 
359 	tp = trace_probe_primary_from_call(call);
360 	if (WARN_ON_ONCE(!tp))
361 		return -ENODEV;
362 	enabled = trace_probe_is_enabled(tp);
363 
364 	/* This also changes "enabled" state */
365 	if (file) {
366 		ret = trace_probe_add_file(tp, file);
367 		if (ret)
368 			return ret;
369 	} else
370 		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
371 
372 	if (enabled)
373 		return 0;
374 
375 	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
376 		tk = container_of(pos, struct trace_kprobe, tp);
377 		if (trace_kprobe_has_gone(tk))
378 			continue;
379 		ret = __enable_trace_kprobe(tk);
380 		if (ret)
381 			break;
382 		enabled = true;
383 	}
384 
385 	if (ret) {
386 		/* Failed to enable one of them. Roll back all */
387 		if (enabled)
388 			__disable_trace_kprobe(tp);
389 		if (file)
390 			trace_probe_remove_file(tp, file);
391 		else
392 			trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
393 	}
394 
395 	return ret;
396 }
397 
398 /*
399  * Disable trace_probe
400  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
401  */
402 static int disable_trace_kprobe(struct trace_event_call *call,
403 				struct trace_event_file *file)
404 {
405 	struct trace_probe *tp;
406 
407 	tp = trace_probe_primary_from_call(call);
408 	if (WARN_ON_ONCE(!tp))
409 		return -ENODEV;
410 
411 	if (file) {
412 		if (!trace_probe_get_file_link(tp, file))
413 			return -ENOENT;
414 		if (!trace_probe_has_single_file(tp))
415 			goto out;
416 		trace_probe_clear_flag(tp, TP_FLAG_TRACE);
417 	} else
418 		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
419 
420 	if (!trace_probe_is_enabled(tp))
421 		__disable_trace_kprobe(tp);
422 
423  out:
424 	if (file)
425 		/*
426 		 * Synchronization is done in below function. For perf event,
427 		 * file == NULL and perf_trace_event_unreg() calls
428 		 * tracepoint_synchronize_unregister() to ensure synchronize
429 		 * event. We don't need to care about it.
430 		 */
431 		trace_probe_remove_file(tp, file);
432 
433 	return 0;
434 }
435 
436 #if defined(CONFIG_KPROBES_ON_FTRACE) && \
437 	!defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
438 static bool within_notrace_func(struct trace_kprobe *tk)
439 {
440 	unsigned long offset, size, addr;
441 
442 	addr = trace_kprobe_address(tk);
443 	if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
444 		return false;
445 
446 	/* Get the entry address of the target function */
447 	addr -= offset;
448 
449 	/*
450 	 * Since ftrace_location_range() does inclusive range check, we need
451 	 * to subtract 1 byte from the end address.
452 	 */
453 	return !ftrace_location_range(addr, addr + size - 1);
454 }
455 #else
456 #define within_notrace_func(tk)	(false)
457 #endif
458 
459 /* Internal register function - just handle k*probes and flags */
460 static int __register_trace_kprobe(struct trace_kprobe *tk)
461 {
462 	int i, ret;
463 
464 	ret = security_locked_down(LOCKDOWN_KPROBES);
465 	if (ret)
466 		return ret;
467 
468 	if (trace_kprobe_is_registered(tk))
469 		return -EINVAL;
470 
471 	if (within_notrace_func(tk)) {
472 		pr_warn("Could not probe notrace function %s\n",
473 			trace_kprobe_symbol(tk));
474 		return -EINVAL;
475 	}
476 
477 	for (i = 0; i < tk->tp.nr_args; i++) {
478 		ret = traceprobe_update_arg(&tk->tp.args[i]);
479 		if (ret)
480 			return ret;
481 	}
482 
483 	/* Set/clear disabled flag according to tp->flag */
484 	if (trace_probe_is_enabled(&tk->tp))
485 		tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
486 	else
487 		tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
488 
489 	if (trace_kprobe_is_return(tk))
490 		ret = register_kretprobe(&tk->rp);
491 	else
492 		ret = register_kprobe(&tk->rp.kp);
493 
494 	return ret;
495 }
496 
497 /* Internal unregister function - just handle k*probes and flags */
498 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
499 {
500 	if (trace_kprobe_is_registered(tk)) {
501 		if (trace_kprobe_is_return(tk))
502 			unregister_kretprobe(&tk->rp);
503 		else
504 			unregister_kprobe(&tk->rp.kp);
505 		/* Cleanup kprobe for reuse and mark it unregistered */
506 		INIT_HLIST_NODE(&tk->rp.kp.hlist);
507 		INIT_LIST_HEAD(&tk->rp.kp.list);
508 		if (tk->rp.kp.symbol_name)
509 			tk->rp.kp.addr = NULL;
510 	}
511 }
512 
513 /* Unregister a trace_probe and probe_event */
514 static int unregister_trace_kprobe(struct trace_kprobe *tk)
515 {
516 	/* If other probes are on the event, just unregister kprobe */
517 	if (trace_probe_has_sibling(&tk->tp))
518 		goto unreg;
519 
520 	/* Enabled event can not be unregistered */
521 	if (trace_probe_is_enabled(&tk->tp))
522 		return -EBUSY;
523 
524 	/* Will fail if probe is being used by ftrace or perf */
525 	if (unregister_kprobe_event(tk))
526 		return -EBUSY;
527 
528 unreg:
529 	__unregister_trace_kprobe(tk);
530 	dyn_event_remove(&tk->devent);
531 	trace_probe_unlink(&tk->tp);
532 
533 	return 0;
534 }
535 
536 static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
537 					 struct trace_kprobe *comp)
538 {
539 	struct trace_probe_event *tpe = orig->tp.event;
540 	struct trace_probe *pos;
541 	int i;
542 
543 	list_for_each_entry(pos, &tpe->probes, list) {
544 		orig = container_of(pos, struct trace_kprobe, tp);
545 		if (strcmp(trace_kprobe_symbol(orig),
546 			   trace_kprobe_symbol(comp)) ||
547 		    trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
548 			continue;
549 
550 		/*
551 		 * trace_probe_compare_arg_type() ensured that nr_args and
552 		 * each argument name and type are same. Let's compare comm.
553 		 */
554 		for (i = 0; i < orig->tp.nr_args; i++) {
555 			if (strcmp(orig->tp.args[i].comm,
556 				   comp->tp.args[i].comm))
557 				break;
558 		}
559 
560 		if (i == orig->tp.nr_args)
561 			return true;
562 	}
563 
564 	return false;
565 }
566 
567 static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
568 {
569 	int ret;
570 
571 	ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
572 	if (ret) {
573 		/* Note that argument starts index = 2 */
574 		trace_probe_log_set_index(ret + 1);
575 		trace_probe_log_err(0, DIFF_ARG_TYPE);
576 		return -EEXIST;
577 	}
578 	if (trace_kprobe_has_same_kprobe(to, tk)) {
579 		trace_probe_log_set_index(0);
580 		trace_probe_log_err(0, SAME_PROBE);
581 		return -EEXIST;
582 	}
583 
584 	/* Append to existing event */
585 	ret = trace_probe_append(&tk->tp, &to->tp);
586 	if (ret)
587 		return ret;
588 
589 	/* Register k*probe */
590 	ret = __register_trace_kprobe(tk);
591 	if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
592 		pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
593 		ret = 0;
594 	}
595 
596 	if (ret)
597 		trace_probe_unlink(&tk->tp);
598 	else
599 		dyn_event_add(&tk->devent);
600 
601 	return ret;
602 }
603 
604 /* Register a trace_probe and probe_event */
605 static int register_trace_kprobe(struct trace_kprobe *tk)
606 {
607 	struct trace_kprobe *old_tk;
608 	int ret;
609 
610 	mutex_lock(&event_mutex);
611 
612 	old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
613 				   trace_probe_group_name(&tk->tp));
614 	if (old_tk) {
615 		if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
616 			trace_probe_log_set_index(0);
617 			trace_probe_log_err(0, DIFF_PROBE_TYPE);
618 			ret = -EEXIST;
619 		} else {
620 			ret = append_trace_kprobe(tk, old_tk);
621 		}
622 		goto end;
623 	}
624 
625 	/* Register new event */
626 	ret = register_kprobe_event(tk);
627 	if (ret) {
628 		pr_warn("Failed to register probe event(%d)\n", ret);
629 		goto end;
630 	}
631 
632 	/* Register k*probe */
633 	ret = __register_trace_kprobe(tk);
634 	if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
635 		pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
636 		ret = 0;
637 	}
638 
639 	if (ret < 0)
640 		unregister_kprobe_event(tk);
641 	else
642 		dyn_event_add(&tk->devent);
643 
644 end:
645 	mutex_unlock(&event_mutex);
646 	return ret;
647 }
648 
649 /* Module notifier call back, checking event on the module */
650 static int trace_kprobe_module_callback(struct notifier_block *nb,
651 				       unsigned long val, void *data)
652 {
653 	struct module *mod = data;
654 	struct dyn_event *pos;
655 	struct trace_kprobe *tk;
656 	int ret;
657 
658 	if (val != MODULE_STATE_COMING)
659 		return NOTIFY_DONE;
660 
661 	/* Update probes on coming module */
662 	mutex_lock(&event_mutex);
663 	for_each_trace_kprobe(tk, pos) {
664 		if (trace_kprobe_within_module(tk, mod)) {
665 			/* Don't need to check busy - this should have gone. */
666 			__unregister_trace_kprobe(tk);
667 			ret = __register_trace_kprobe(tk);
668 			if (ret)
669 				pr_warn("Failed to re-register probe %s on %s: %d\n",
670 					trace_probe_name(&tk->tp),
671 					mod->name, ret);
672 		}
673 	}
674 	mutex_unlock(&event_mutex);
675 
676 	return NOTIFY_DONE;
677 }
678 
679 static struct notifier_block trace_kprobe_module_nb = {
680 	.notifier_call = trace_kprobe_module_callback,
681 	.priority = 1	/* Invoked after kprobe module callback */
682 };
683 
684 /* Convert certain expected symbols into '_' when generating event names */
685 static inline void sanitize_event_name(char *name)
686 {
687 	while (*name++ != '\0')
688 		if (*name == ':' || *name == '.')
689 			*name = '_';
690 }
691 
692 static int trace_kprobe_create(int argc, const char *argv[])
693 {
694 	/*
695 	 * Argument syntax:
696 	 *  - Add kprobe:
697 	 *      p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
698 	 *  - Add kretprobe:
699 	 *      r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
700 	 * Fetch args:
701 	 *  $retval	: fetch return value
702 	 *  $stack	: fetch stack address
703 	 *  $stackN	: fetch Nth of stack (N:0-)
704 	 *  $comm       : fetch current task comm
705 	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
706 	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
707 	 *  %REG	: fetch register REG
708 	 * Dereferencing memory fetch:
709 	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
710 	 * Alias name of args:
711 	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
712 	 * Type of args:
713 	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
714 	 */
715 	struct trace_kprobe *tk = NULL;
716 	int i, len, ret = 0;
717 	bool is_return = false;
718 	char *symbol = NULL, *tmp = NULL;
719 	const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
720 	int maxactive = 0;
721 	long offset = 0;
722 	void *addr = NULL;
723 	char buf[MAX_EVENT_NAME_LEN];
724 	unsigned int flags = TPARG_FL_KERNEL;
725 
726 	switch (argv[0][0]) {
727 	case 'r':
728 		is_return = true;
729 		flags |= TPARG_FL_RETURN;
730 		break;
731 	case 'p':
732 		break;
733 	default:
734 		return -ECANCELED;
735 	}
736 	if (argc < 2)
737 		return -ECANCELED;
738 
739 	trace_probe_log_init("trace_kprobe", argc, argv);
740 
741 	event = strchr(&argv[0][1], ':');
742 	if (event)
743 		event++;
744 
745 	if (isdigit(argv[0][1])) {
746 		if (!is_return) {
747 			trace_probe_log_err(1, MAXACT_NO_KPROBE);
748 			goto parse_error;
749 		}
750 		if (event)
751 			len = event - &argv[0][1] - 1;
752 		else
753 			len = strlen(&argv[0][1]);
754 		if (len > MAX_EVENT_NAME_LEN - 1) {
755 			trace_probe_log_err(1, BAD_MAXACT);
756 			goto parse_error;
757 		}
758 		memcpy(buf, &argv[0][1], len);
759 		buf[len] = '\0';
760 		ret = kstrtouint(buf, 0, &maxactive);
761 		if (ret || !maxactive) {
762 			trace_probe_log_err(1, BAD_MAXACT);
763 			goto parse_error;
764 		}
765 		/* kretprobes instances are iterated over via a list. The
766 		 * maximum should stay reasonable.
767 		 */
768 		if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
769 			trace_probe_log_err(1, MAXACT_TOO_BIG);
770 			goto parse_error;
771 		}
772 	}
773 
774 	/* try to parse an address. if that fails, try to read the
775 	 * input as a symbol. */
776 	if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
777 		trace_probe_log_set_index(1);
778 		/* Check whether uprobe event specified */
779 		if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
780 			ret = -ECANCELED;
781 			goto error;
782 		}
783 		/* a symbol specified */
784 		symbol = kstrdup(argv[1], GFP_KERNEL);
785 		if (!symbol)
786 			return -ENOMEM;
787 		/* TODO: support .init module functions */
788 		ret = traceprobe_split_symbol_offset(symbol, &offset);
789 		if (ret || offset < 0 || offset > UINT_MAX) {
790 			trace_probe_log_err(0, BAD_PROBE_ADDR);
791 			goto parse_error;
792 		}
793 		if (kprobe_on_func_entry(NULL, symbol, offset))
794 			flags |= TPARG_FL_FENTRY;
795 		if (offset && is_return && !(flags & TPARG_FL_FENTRY)) {
796 			trace_probe_log_err(0, BAD_RETPROBE);
797 			goto parse_error;
798 		}
799 	}
800 
801 	trace_probe_log_set_index(0);
802 	if (event) {
803 		ret = traceprobe_parse_event_name(&event, &group, buf,
804 						  event - argv[0]);
805 		if (ret)
806 			goto parse_error;
807 	} else {
808 		/* Make a new event name */
809 		if (symbol)
810 			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
811 				 is_return ? 'r' : 'p', symbol, offset);
812 		else
813 			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
814 				 is_return ? 'r' : 'p', addr);
815 		sanitize_event_name(buf);
816 		event = buf;
817 	}
818 
819 	/* setup a probe */
820 	tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
821 			       argc - 2, is_return);
822 	if (IS_ERR(tk)) {
823 		ret = PTR_ERR(tk);
824 		/* This must return -ENOMEM, else there is a bug */
825 		WARN_ON_ONCE(ret != -ENOMEM);
826 		goto out;	/* We know tk is not allocated */
827 	}
828 	argc -= 2; argv += 2;
829 
830 	/* parse arguments */
831 	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
832 		tmp = kstrdup(argv[i], GFP_KERNEL);
833 		if (!tmp) {
834 			ret = -ENOMEM;
835 			goto error;
836 		}
837 
838 		trace_probe_log_set_index(i + 2);
839 		ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
840 		kfree(tmp);
841 		if (ret)
842 			goto error;	/* This can be -ENOMEM */
843 	}
844 
845 	ret = traceprobe_set_print_fmt(&tk->tp, is_return);
846 	if (ret < 0)
847 		goto error;
848 
849 	ret = register_trace_kprobe(tk);
850 	if (ret) {
851 		trace_probe_log_set_index(1);
852 		if (ret == -EILSEQ)
853 			trace_probe_log_err(0, BAD_INSN_BNDRY);
854 		else if (ret == -ENOENT)
855 			trace_probe_log_err(0, BAD_PROBE_ADDR);
856 		else if (ret != -ENOMEM && ret != -EEXIST)
857 			trace_probe_log_err(0, FAIL_REG_PROBE);
858 		goto error;
859 	}
860 
861 out:
862 	trace_probe_log_clear();
863 	kfree(symbol);
864 	return ret;
865 
866 parse_error:
867 	ret = -EINVAL;
868 error:
869 	free_trace_kprobe(tk);
870 	goto out;
871 }
872 
873 static int create_or_delete_trace_kprobe(int argc, char **argv)
874 {
875 	int ret;
876 
877 	if (argv[0][0] == '-')
878 		return dyn_event_release(argc, argv, &trace_kprobe_ops);
879 
880 	ret = trace_kprobe_create(argc, (const char **)argv);
881 	return ret == -ECANCELED ? -EINVAL : ret;
882 }
883 
884 static int trace_kprobe_release(struct dyn_event *ev)
885 {
886 	struct trace_kprobe *tk = to_trace_kprobe(ev);
887 	int ret = unregister_trace_kprobe(tk);
888 
889 	if (!ret)
890 		free_trace_kprobe(tk);
891 	return ret;
892 }
893 
894 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
895 {
896 	struct trace_kprobe *tk = to_trace_kprobe(ev);
897 	int i;
898 
899 	seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
900 	seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
901 				trace_probe_name(&tk->tp));
902 
903 	if (!tk->symbol)
904 		seq_printf(m, " 0x%p", tk->rp.kp.addr);
905 	else if (tk->rp.kp.offset)
906 		seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
907 			   tk->rp.kp.offset);
908 	else
909 		seq_printf(m, " %s", trace_kprobe_symbol(tk));
910 
911 	for (i = 0; i < tk->tp.nr_args; i++)
912 		seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
913 	seq_putc(m, '\n');
914 
915 	return 0;
916 }
917 
918 static int probes_seq_show(struct seq_file *m, void *v)
919 {
920 	struct dyn_event *ev = v;
921 
922 	if (!is_trace_kprobe(ev))
923 		return 0;
924 
925 	return trace_kprobe_show(m, ev);
926 }
927 
928 static const struct seq_operations probes_seq_op = {
929 	.start  = dyn_event_seq_start,
930 	.next   = dyn_event_seq_next,
931 	.stop   = dyn_event_seq_stop,
932 	.show   = probes_seq_show
933 };
934 
935 static int probes_open(struct inode *inode, struct file *file)
936 {
937 	int ret;
938 
939 	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
940 		ret = dyn_events_release_all(&trace_kprobe_ops);
941 		if (ret < 0)
942 			return ret;
943 	}
944 
945 	return seq_open(file, &probes_seq_op);
946 }
947 
948 static ssize_t probes_write(struct file *file, const char __user *buffer,
949 			    size_t count, loff_t *ppos)
950 {
951 	return trace_parse_run_command(file, buffer, count, ppos,
952 				       create_or_delete_trace_kprobe);
953 }
954 
955 static const struct file_operations kprobe_events_ops = {
956 	.owner          = THIS_MODULE,
957 	.open           = probes_open,
958 	.read           = seq_read,
959 	.llseek         = seq_lseek,
960 	.release        = seq_release,
961 	.write		= probes_write,
962 };
963 
964 /* Probes profiling interfaces */
965 static int probes_profile_seq_show(struct seq_file *m, void *v)
966 {
967 	struct dyn_event *ev = v;
968 	struct trace_kprobe *tk;
969 
970 	if (!is_trace_kprobe(ev))
971 		return 0;
972 
973 	tk = to_trace_kprobe(ev);
974 	seq_printf(m, "  %-44s %15lu %15lu\n",
975 		   trace_probe_name(&tk->tp),
976 		   trace_kprobe_nhit(tk),
977 		   tk->rp.kp.nmissed);
978 
979 	return 0;
980 }
981 
982 static const struct seq_operations profile_seq_op = {
983 	.start  = dyn_event_seq_start,
984 	.next   = dyn_event_seq_next,
985 	.stop   = dyn_event_seq_stop,
986 	.show   = probes_profile_seq_show
987 };
988 
989 static int profile_open(struct inode *inode, struct file *file)
990 {
991 	return seq_open(file, &profile_seq_op);
992 }
993 
994 static const struct file_operations kprobe_profile_ops = {
995 	.owner          = THIS_MODULE,
996 	.open           = profile_open,
997 	.read           = seq_read,
998 	.llseek         = seq_lseek,
999 	.release        = seq_release,
1000 };
1001 
1002 /* Kprobe specific fetch functions */
1003 
1004 /* Return the length of string -- including null terminal byte */
1005 static nokprobe_inline int
1006 fetch_store_strlen(unsigned long addr)
1007 {
1008 	int ret, len = 0;
1009 	u8 c;
1010 
1011 	do {
1012 		ret = probe_kernel_read(&c, (u8 *)addr + len, 1);
1013 		len++;
1014 	} while (c && ret == 0 && len < MAX_STRING_SIZE);
1015 
1016 	return (ret < 0) ? ret : len;
1017 }
1018 
1019 /* Return the length of string -- including null terminal byte */
1020 static nokprobe_inline int
1021 fetch_store_strlen_user(unsigned long addr)
1022 {
1023 	const void __user *uaddr =  (__force const void __user *)addr;
1024 
1025 	return strnlen_unsafe_user(uaddr, MAX_STRING_SIZE);
1026 }
1027 
1028 /*
1029  * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
1030  * length and relative data location.
1031  */
1032 static nokprobe_inline int
1033 fetch_store_string(unsigned long addr, void *dest, void *base)
1034 {
1035 	int maxlen = get_loc_len(*(u32 *)dest);
1036 	void *__dest;
1037 	long ret;
1038 
1039 	if (unlikely(!maxlen))
1040 		return -ENOMEM;
1041 
1042 	__dest = get_loc_data(dest, base);
1043 
1044 	/*
1045 	 * Try to get string again, since the string can be changed while
1046 	 * probing.
1047 	 */
1048 	ret = strncpy_from_unsafe(__dest, (void *)addr, maxlen);
1049 	if (ret >= 0)
1050 		*(u32 *)dest = make_data_loc(ret, __dest - base);
1051 
1052 	return ret;
1053 }
1054 
1055 /*
1056  * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
1057  * with max length and relative data location.
1058  */
1059 static nokprobe_inline int
1060 fetch_store_string_user(unsigned long addr, void *dest, void *base)
1061 {
1062 	const void __user *uaddr =  (__force const void __user *)addr;
1063 	int maxlen = get_loc_len(*(u32 *)dest);
1064 	void *__dest;
1065 	long ret;
1066 
1067 	if (unlikely(!maxlen))
1068 		return -ENOMEM;
1069 
1070 	__dest = get_loc_data(dest, base);
1071 
1072 	ret = strncpy_from_unsafe_user(__dest, uaddr, maxlen);
1073 	if (ret >= 0)
1074 		*(u32 *)dest = make_data_loc(ret, __dest - base);
1075 
1076 	return ret;
1077 }
1078 
1079 static nokprobe_inline int
1080 probe_mem_read(void *dest, void *src, size_t size)
1081 {
1082 	return probe_kernel_read(dest, src, size);
1083 }
1084 
1085 static nokprobe_inline int
1086 probe_mem_read_user(void *dest, void *src, size_t size)
1087 {
1088 	const void __user *uaddr =  (__force const void __user *)src;
1089 
1090 	return probe_user_read(dest, uaddr, size);
1091 }
1092 
1093 /* Note that we don't verify it, since the code does not come from user space */
1094 static int
1095 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
1096 		   void *base)
1097 {
1098 	unsigned long val;
1099 
1100 retry:
1101 	/* 1st stage: get value from context */
1102 	switch (code->op) {
1103 	case FETCH_OP_REG:
1104 		val = regs_get_register(regs, code->param);
1105 		break;
1106 	case FETCH_OP_STACK:
1107 		val = regs_get_kernel_stack_nth(regs, code->param);
1108 		break;
1109 	case FETCH_OP_STACKP:
1110 		val = kernel_stack_pointer(regs);
1111 		break;
1112 	case FETCH_OP_RETVAL:
1113 		val = regs_return_value(regs);
1114 		break;
1115 	case FETCH_OP_IMM:
1116 		val = code->immediate;
1117 		break;
1118 	case FETCH_OP_COMM:
1119 		val = (unsigned long)current->comm;
1120 		break;
1121 	case FETCH_OP_DATA:
1122 		val = (unsigned long)code->data;
1123 		break;
1124 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1125 	case FETCH_OP_ARG:
1126 		val = regs_get_kernel_argument(regs, code->param);
1127 		break;
1128 #endif
1129 	case FETCH_NOP_SYMBOL:	/* Ignore a place holder */
1130 		code++;
1131 		goto retry;
1132 	default:
1133 		return -EILSEQ;
1134 	}
1135 	code++;
1136 
1137 	return process_fetch_insn_bottom(code, val, dest, base);
1138 }
1139 NOKPROBE_SYMBOL(process_fetch_insn)
1140 
1141 /* Kprobe handler */
1142 static nokprobe_inline void
1143 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1144 		    struct trace_event_file *trace_file)
1145 {
1146 	struct kprobe_trace_entry_head *entry;
1147 	struct ring_buffer_event *event;
1148 	struct ring_buffer *buffer;
1149 	int size, dsize, pc;
1150 	unsigned long irq_flags;
1151 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1152 
1153 	WARN_ON(call != trace_file->event_call);
1154 
1155 	if (trace_trigger_soft_disabled(trace_file))
1156 		return;
1157 
1158 	local_save_flags(irq_flags);
1159 	pc = preempt_count();
1160 
1161 	dsize = __get_data_size(&tk->tp, regs);
1162 	size = sizeof(*entry) + tk->tp.size + dsize;
1163 
1164 	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1165 						call->event.type,
1166 						size, irq_flags, pc);
1167 	if (!event)
1168 		return;
1169 
1170 	entry = ring_buffer_event_data(event);
1171 	entry->ip = (unsigned long)tk->rp.kp.addr;
1172 	store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1173 
1174 	event_trigger_unlock_commit_regs(trace_file, buffer, event,
1175 					 entry, irq_flags, pc, regs);
1176 }
1177 
1178 static void
1179 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1180 {
1181 	struct event_file_link *link;
1182 
1183 	trace_probe_for_each_link_rcu(link, &tk->tp)
1184 		__kprobe_trace_func(tk, regs, link->file);
1185 }
1186 NOKPROBE_SYMBOL(kprobe_trace_func);
1187 
1188 /* Kretprobe handler */
1189 static nokprobe_inline void
1190 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1191 		       struct pt_regs *regs,
1192 		       struct trace_event_file *trace_file)
1193 {
1194 	struct kretprobe_trace_entry_head *entry;
1195 	struct ring_buffer_event *event;
1196 	struct ring_buffer *buffer;
1197 	int size, pc, dsize;
1198 	unsigned long irq_flags;
1199 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1200 
1201 	WARN_ON(call != trace_file->event_call);
1202 
1203 	if (trace_trigger_soft_disabled(trace_file))
1204 		return;
1205 
1206 	local_save_flags(irq_flags);
1207 	pc = preempt_count();
1208 
1209 	dsize = __get_data_size(&tk->tp, regs);
1210 	size = sizeof(*entry) + tk->tp.size + dsize;
1211 
1212 	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1213 						call->event.type,
1214 						size, irq_flags, pc);
1215 	if (!event)
1216 		return;
1217 
1218 	entry = ring_buffer_event_data(event);
1219 	entry->func = (unsigned long)tk->rp.kp.addr;
1220 	entry->ret_ip = (unsigned long)ri->ret_addr;
1221 	store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1222 
1223 	event_trigger_unlock_commit_regs(trace_file, buffer, event,
1224 					 entry, irq_flags, pc, regs);
1225 }
1226 
1227 static void
1228 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1229 		     struct pt_regs *regs)
1230 {
1231 	struct event_file_link *link;
1232 
1233 	trace_probe_for_each_link_rcu(link, &tk->tp)
1234 		__kretprobe_trace_func(tk, ri, regs, link->file);
1235 }
1236 NOKPROBE_SYMBOL(kretprobe_trace_func);
1237 
1238 /* Event entry printers */
1239 static enum print_line_t
1240 print_kprobe_event(struct trace_iterator *iter, int flags,
1241 		   struct trace_event *event)
1242 {
1243 	struct kprobe_trace_entry_head *field;
1244 	struct trace_seq *s = &iter->seq;
1245 	struct trace_probe *tp;
1246 
1247 	field = (struct kprobe_trace_entry_head *)iter->ent;
1248 	tp = trace_probe_primary_from_call(
1249 		container_of(event, struct trace_event_call, event));
1250 	if (WARN_ON_ONCE(!tp))
1251 		goto out;
1252 
1253 	trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1254 
1255 	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1256 		goto out;
1257 
1258 	trace_seq_putc(s, ')');
1259 
1260 	if (print_probe_args(s, tp->args, tp->nr_args,
1261 			     (u8 *)&field[1], field) < 0)
1262 		goto out;
1263 
1264 	trace_seq_putc(s, '\n');
1265  out:
1266 	return trace_handle_return(s);
1267 }
1268 
1269 static enum print_line_t
1270 print_kretprobe_event(struct trace_iterator *iter, int flags,
1271 		      struct trace_event *event)
1272 {
1273 	struct kretprobe_trace_entry_head *field;
1274 	struct trace_seq *s = &iter->seq;
1275 	struct trace_probe *tp;
1276 
1277 	field = (struct kretprobe_trace_entry_head *)iter->ent;
1278 	tp = trace_probe_primary_from_call(
1279 		container_of(event, struct trace_event_call, event));
1280 	if (WARN_ON_ONCE(!tp))
1281 		goto out;
1282 
1283 	trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1284 
1285 	if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1286 		goto out;
1287 
1288 	trace_seq_puts(s, " <- ");
1289 
1290 	if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1291 		goto out;
1292 
1293 	trace_seq_putc(s, ')');
1294 
1295 	if (print_probe_args(s, tp->args, tp->nr_args,
1296 			     (u8 *)&field[1], field) < 0)
1297 		goto out;
1298 
1299 	trace_seq_putc(s, '\n');
1300 
1301  out:
1302 	return trace_handle_return(s);
1303 }
1304 
1305 
1306 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1307 {
1308 	int ret;
1309 	struct kprobe_trace_entry_head field;
1310 	struct trace_probe *tp;
1311 
1312 	tp = trace_probe_primary_from_call(event_call);
1313 	if (WARN_ON_ONCE(!tp))
1314 		return -ENOENT;
1315 
1316 	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1317 
1318 	return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1319 }
1320 
1321 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1322 {
1323 	int ret;
1324 	struct kretprobe_trace_entry_head field;
1325 	struct trace_probe *tp;
1326 
1327 	tp = trace_probe_primary_from_call(event_call);
1328 	if (WARN_ON_ONCE(!tp))
1329 		return -ENOENT;
1330 
1331 	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1332 	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1333 
1334 	return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
1335 }
1336 
1337 #ifdef CONFIG_PERF_EVENTS
1338 
1339 /* Kprobe profile handler */
1340 static int
1341 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1342 {
1343 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1344 	struct kprobe_trace_entry_head *entry;
1345 	struct hlist_head *head;
1346 	int size, __size, dsize;
1347 	int rctx;
1348 
1349 	if (bpf_prog_array_valid(call)) {
1350 		unsigned long orig_ip = instruction_pointer(regs);
1351 		int ret;
1352 
1353 		ret = trace_call_bpf(call, regs);
1354 
1355 		/*
1356 		 * We need to check and see if we modified the pc of the
1357 		 * pt_regs, and if so return 1 so that we don't do the
1358 		 * single stepping.
1359 		 */
1360 		if (orig_ip != instruction_pointer(regs))
1361 			return 1;
1362 		if (!ret)
1363 			return 0;
1364 	}
1365 
1366 	head = this_cpu_ptr(call->perf_events);
1367 	if (hlist_empty(head))
1368 		return 0;
1369 
1370 	dsize = __get_data_size(&tk->tp, regs);
1371 	__size = sizeof(*entry) + tk->tp.size + dsize;
1372 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1373 	size -= sizeof(u32);
1374 
1375 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1376 	if (!entry)
1377 		return 0;
1378 
1379 	entry->ip = (unsigned long)tk->rp.kp.addr;
1380 	memset(&entry[1], 0, dsize);
1381 	store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1382 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1383 			      head, NULL);
1384 	return 0;
1385 }
1386 NOKPROBE_SYMBOL(kprobe_perf_func);
1387 
1388 /* Kretprobe profile handler */
1389 static void
1390 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1391 		    struct pt_regs *regs)
1392 {
1393 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1394 	struct kretprobe_trace_entry_head *entry;
1395 	struct hlist_head *head;
1396 	int size, __size, dsize;
1397 	int rctx;
1398 
1399 	if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1400 		return;
1401 
1402 	head = this_cpu_ptr(call->perf_events);
1403 	if (hlist_empty(head))
1404 		return;
1405 
1406 	dsize = __get_data_size(&tk->tp, regs);
1407 	__size = sizeof(*entry) + tk->tp.size + dsize;
1408 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1409 	size -= sizeof(u32);
1410 
1411 	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1412 	if (!entry)
1413 		return;
1414 
1415 	entry->func = (unsigned long)tk->rp.kp.addr;
1416 	entry->ret_ip = (unsigned long)ri->ret_addr;
1417 	store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1418 	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1419 			      head, NULL);
1420 }
1421 NOKPROBE_SYMBOL(kretprobe_perf_func);
1422 
1423 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1424 			const char **symbol, u64 *probe_offset,
1425 			u64 *probe_addr, bool perf_type_tracepoint)
1426 {
1427 	const char *pevent = trace_event_name(event->tp_event);
1428 	const char *group = event->tp_event->class->system;
1429 	struct trace_kprobe *tk;
1430 
1431 	if (perf_type_tracepoint)
1432 		tk = find_trace_kprobe(pevent, group);
1433 	else
1434 		tk = event->tp_event->data;
1435 	if (!tk)
1436 		return -EINVAL;
1437 
1438 	*fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1439 					      : BPF_FD_TYPE_KPROBE;
1440 	if (tk->symbol) {
1441 		*symbol = tk->symbol;
1442 		*probe_offset = tk->rp.kp.offset;
1443 		*probe_addr = 0;
1444 	} else {
1445 		*symbol = NULL;
1446 		*probe_offset = 0;
1447 		*probe_addr = (unsigned long)tk->rp.kp.addr;
1448 	}
1449 	return 0;
1450 }
1451 #endif	/* CONFIG_PERF_EVENTS */
1452 
1453 /*
1454  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1455  *
1456  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1457  * lockless, but we can't race with this __init function.
1458  */
1459 static int kprobe_register(struct trace_event_call *event,
1460 			   enum trace_reg type, void *data)
1461 {
1462 	struct trace_event_file *file = data;
1463 
1464 	switch (type) {
1465 	case TRACE_REG_REGISTER:
1466 		return enable_trace_kprobe(event, file);
1467 	case TRACE_REG_UNREGISTER:
1468 		return disable_trace_kprobe(event, file);
1469 
1470 #ifdef CONFIG_PERF_EVENTS
1471 	case TRACE_REG_PERF_REGISTER:
1472 		return enable_trace_kprobe(event, NULL);
1473 	case TRACE_REG_PERF_UNREGISTER:
1474 		return disable_trace_kprobe(event, NULL);
1475 	case TRACE_REG_PERF_OPEN:
1476 	case TRACE_REG_PERF_CLOSE:
1477 	case TRACE_REG_PERF_ADD:
1478 	case TRACE_REG_PERF_DEL:
1479 		return 0;
1480 #endif
1481 	}
1482 	return 0;
1483 }
1484 
1485 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1486 {
1487 	struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1488 	int ret = 0;
1489 
1490 	raw_cpu_inc(*tk->nhit);
1491 
1492 	if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1493 		kprobe_trace_func(tk, regs);
1494 #ifdef CONFIG_PERF_EVENTS
1495 	if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1496 		ret = kprobe_perf_func(tk, regs);
1497 #endif
1498 	return ret;
1499 }
1500 NOKPROBE_SYMBOL(kprobe_dispatcher);
1501 
1502 static int
1503 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1504 {
1505 	struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1506 
1507 	raw_cpu_inc(*tk->nhit);
1508 
1509 	if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1510 		kretprobe_trace_func(tk, ri, regs);
1511 #ifdef CONFIG_PERF_EVENTS
1512 	if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1513 		kretprobe_perf_func(tk, ri, regs);
1514 #endif
1515 	return 0;	/* We don't tweek kernel, so just return 0 */
1516 }
1517 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1518 
1519 static struct trace_event_functions kretprobe_funcs = {
1520 	.trace		= print_kretprobe_event
1521 };
1522 
1523 static struct trace_event_functions kprobe_funcs = {
1524 	.trace		= print_kprobe_event
1525 };
1526 
1527 static inline void init_trace_event_call(struct trace_kprobe *tk)
1528 {
1529 	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1530 
1531 	if (trace_kprobe_is_return(tk)) {
1532 		call->event.funcs = &kretprobe_funcs;
1533 		call->class->define_fields = kretprobe_event_define_fields;
1534 	} else {
1535 		call->event.funcs = &kprobe_funcs;
1536 		call->class->define_fields = kprobe_event_define_fields;
1537 	}
1538 
1539 	call->flags = TRACE_EVENT_FL_KPROBE;
1540 	call->class->reg = kprobe_register;
1541 }
1542 
1543 static int register_kprobe_event(struct trace_kprobe *tk)
1544 {
1545 	init_trace_event_call(tk);
1546 
1547 	return trace_probe_register_event_call(&tk->tp);
1548 }
1549 
1550 static int unregister_kprobe_event(struct trace_kprobe *tk)
1551 {
1552 	return trace_probe_unregister_event_call(&tk->tp);
1553 }
1554 
1555 #ifdef CONFIG_PERF_EVENTS
1556 /* create a trace_kprobe, but don't add it to global lists */
1557 struct trace_event_call *
1558 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1559 			  bool is_return)
1560 {
1561 	struct trace_kprobe *tk;
1562 	int ret;
1563 	char *event;
1564 
1565 	/*
1566 	 * local trace_kprobes are not added to dyn_event, so they are never
1567 	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1568 	 * duplicated name here.
1569 	 */
1570 	event = func ? func : "DUMMY_EVENT";
1571 
1572 	tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1573 				offs, 0 /* maxactive */, 0 /* nargs */,
1574 				is_return);
1575 
1576 	if (IS_ERR(tk)) {
1577 		pr_info("Failed to allocate trace_probe.(%d)\n",
1578 			(int)PTR_ERR(tk));
1579 		return ERR_CAST(tk);
1580 	}
1581 
1582 	init_trace_event_call(tk);
1583 
1584 	if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1585 		ret = -ENOMEM;
1586 		goto error;
1587 	}
1588 
1589 	ret = __register_trace_kprobe(tk);
1590 	if (ret < 0)
1591 		goto error;
1592 
1593 	return trace_probe_event_call(&tk->tp);
1594 error:
1595 	free_trace_kprobe(tk);
1596 	return ERR_PTR(ret);
1597 }
1598 
1599 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1600 {
1601 	struct trace_kprobe *tk;
1602 
1603 	tk = trace_kprobe_primary_from_call(event_call);
1604 	if (unlikely(!tk))
1605 		return;
1606 
1607 	if (trace_probe_is_enabled(&tk->tp)) {
1608 		WARN_ON(1);
1609 		return;
1610 	}
1611 
1612 	__unregister_trace_kprobe(tk);
1613 
1614 	free_trace_kprobe(tk);
1615 }
1616 #endif /* CONFIG_PERF_EVENTS */
1617 
1618 static __init void enable_boot_kprobe_events(void)
1619 {
1620 	struct trace_array *tr = top_trace_array();
1621 	struct trace_event_file *file;
1622 	struct trace_kprobe *tk;
1623 	struct dyn_event *pos;
1624 
1625 	mutex_lock(&event_mutex);
1626 	for_each_trace_kprobe(tk, pos) {
1627 		list_for_each_entry(file, &tr->events, list)
1628 			if (file->event_call == trace_probe_event_call(&tk->tp))
1629 				trace_event_enable_disable(file, 1, 0);
1630 	}
1631 	mutex_unlock(&event_mutex);
1632 }
1633 
1634 static __init void setup_boot_kprobe_events(void)
1635 {
1636 	char *p, *cmd = kprobe_boot_events_buf;
1637 	int ret;
1638 
1639 	strreplace(kprobe_boot_events_buf, ',', ' ');
1640 
1641 	while (cmd && *cmd != '\0') {
1642 		p = strchr(cmd, ';');
1643 		if (p)
1644 			*p++ = '\0';
1645 
1646 		ret = trace_run_command(cmd, create_or_delete_trace_kprobe);
1647 		if (ret)
1648 			pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1649 		else
1650 			kprobe_boot_events_enabled = true;
1651 
1652 		cmd = p;
1653 	}
1654 
1655 	enable_boot_kprobe_events();
1656 }
1657 
1658 /* Make a tracefs interface for controlling probe points */
1659 static __init int init_kprobe_trace(void)
1660 {
1661 	struct dentry *d_tracer;
1662 	struct dentry *entry;
1663 	int ret;
1664 
1665 	ret = dyn_event_register(&trace_kprobe_ops);
1666 	if (ret)
1667 		return ret;
1668 
1669 	if (register_module_notifier(&trace_kprobe_module_nb))
1670 		return -EINVAL;
1671 
1672 	d_tracer = tracing_init_dentry();
1673 	if (IS_ERR(d_tracer))
1674 		return 0;
1675 
1676 	entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1677 				    NULL, &kprobe_events_ops);
1678 
1679 	/* Event list interface */
1680 	if (!entry)
1681 		pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1682 
1683 	/* Profile interface */
1684 	entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1685 				    NULL, &kprobe_profile_ops);
1686 
1687 	if (!entry)
1688 		pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1689 
1690 	setup_boot_kprobe_events();
1691 
1692 	return 0;
1693 }
1694 fs_initcall(init_kprobe_trace);
1695 
1696 
1697 #ifdef CONFIG_FTRACE_STARTUP_TEST
1698 static __init struct trace_event_file *
1699 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1700 {
1701 	struct trace_event_file *file;
1702 
1703 	list_for_each_entry(file, &tr->events, list)
1704 		if (file->event_call == trace_probe_event_call(&tk->tp))
1705 			return file;
1706 
1707 	return NULL;
1708 }
1709 
1710 /*
1711  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1712  * stage, we can do this lockless.
1713  */
1714 static __init int kprobe_trace_self_tests_init(void)
1715 {
1716 	int ret, warn = 0;
1717 	int (*target)(int, int, int, int, int, int);
1718 	struct trace_kprobe *tk;
1719 	struct trace_event_file *file;
1720 
1721 	if (tracing_is_disabled())
1722 		return -ENODEV;
1723 
1724 	if (kprobe_boot_events_enabled) {
1725 		pr_info("Skipping kprobe tests due to kprobe_event on cmdline\n");
1726 		return 0;
1727 	}
1728 
1729 	target = kprobe_trace_selftest_target;
1730 
1731 	pr_info("Testing kprobe tracing: ");
1732 
1733 	ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
1734 				create_or_delete_trace_kprobe);
1735 	if (WARN_ON_ONCE(ret)) {
1736 		pr_warn("error on probing function entry.\n");
1737 		warn++;
1738 	} else {
1739 		/* Enable trace point */
1740 		tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1741 		if (WARN_ON_ONCE(tk == NULL)) {
1742 			pr_warn("error on getting new probe.\n");
1743 			warn++;
1744 		} else {
1745 			file = find_trace_probe_file(tk, top_trace_array());
1746 			if (WARN_ON_ONCE(file == NULL)) {
1747 				pr_warn("error on getting probe file.\n");
1748 				warn++;
1749 			} else
1750 				enable_trace_kprobe(
1751 					trace_probe_event_call(&tk->tp), file);
1752 		}
1753 	}
1754 
1755 	ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
1756 				create_or_delete_trace_kprobe);
1757 	if (WARN_ON_ONCE(ret)) {
1758 		pr_warn("error on probing function return.\n");
1759 		warn++;
1760 	} else {
1761 		/* Enable trace point */
1762 		tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1763 		if (WARN_ON_ONCE(tk == NULL)) {
1764 			pr_warn("error on getting 2nd new probe.\n");
1765 			warn++;
1766 		} else {
1767 			file = find_trace_probe_file(tk, top_trace_array());
1768 			if (WARN_ON_ONCE(file == NULL)) {
1769 				pr_warn("error on getting probe file.\n");
1770 				warn++;
1771 			} else
1772 				enable_trace_kprobe(
1773 					trace_probe_event_call(&tk->tp), file);
1774 		}
1775 	}
1776 
1777 	if (warn)
1778 		goto end;
1779 
1780 	ret = target(1, 2, 3, 4, 5, 6);
1781 
1782 	/*
1783 	 * Not expecting an error here, the check is only to prevent the
1784 	 * optimizer from removing the call to target() as otherwise there
1785 	 * are no side-effects and the call is never performed.
1786 	 */
1787 	if (ret != 21)
1788 		warn++;
1789 
1790 	/* Disable trace points before removing it */
1791 	tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1792 	if (WARN_ON_ONCE(tk == NULL)) {
1793 		pr_warn("error on getting test probe.\n");
1794 		warn++;
1795 	} else {
1796 		if (trace_kprobe_nhit(tk) != 1) {
1797 			pr_warn("incorrect number of testprobe hits\n");
1798 			warn++;
1799 		}
1800 
1801 		file = find_trace_probe_file(tk, top_trace_array());
1802 		if (WARN_ON_ONCE(file == NULL)) {
1803 			pr_warn("error on getting probe file.\n");
1804 			warn++;
1805 		} else
1806 			disable_trace_kprobe(
1807 				trace_probe_event_call(&tk->tp), file);
1808 	}
1809 
1810 	tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1811 	if (WARN_ON_ONCE(tk == NULL)) {
1812 		pr_warn("error on getting 2nd test probe.\n");
1813 		warn++;
1814 	} else {
1815 		if (trace_kprobe_nhit(tk) != 1) {
1816 			pr_warn("incorrect number of testprobe2 hits\n");
1817 			warn++;
1818 		}
1819 
1820 		file = find_trace_probe_file(tk, top_trace_array());
1821 		if (WARN_ON_ONCE(file == NULL)) {
1822 			pr_warn("error on getting probe file.\n");
1823 			warn++;
1824 		} else
1825 			disable_trace_kprobe(
1826 				trace_probe_event_call(&tk->tp), file);
1827 	}
1828 
1829 	ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
1830 	if (WARN_ON_ONCE(ret)) {
1831 		pr_warn("error on deleting a probe.\n");
1832 		warn++;
1833 	}
1834 
1835 	ret = trace_run_command("-:testprobe2", create_or_delete_trace_kprobe);
1836 	if (WARN_ON_ONCE(ret)) {
1837 		pr_warn("error on deleting a probe.\n");
1838 		warn++;
1839 	}
1840 
1841 end:
1842 	ret = dyn_events_release_all(&trace_kprobe_ops);
1843 	if (WARN_ON_ONCE(ret)) {
1844 		pr_warn("error on cleaning up probes.\n");
1845 		warn++;
1846 	}
1847 	/*
1848 	 * Wait for the optimizer work to finish. Otherwise it might fiddle
1849 	 * with probes in already freed __init text.
1850 	 */
1851 	wait_for_kprobe_optimizer();
1852 	if (warn)
1853 		pr_cont("NG: Some tests are failed. Please check them.\n");
1854 	else
1855 		pr_cont("OK\n");
1856 	return 0;
1857 }
1858 
1859 late_initcall(kprobe_trace_self_tests_init);
1860 
1861 #endif
1862