xref: /openbmc/linux/kernel/trace/trace_uprobe.c (revision 861e10be)
1 /*
2  * uprobes-based tracing events
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11  * GNU General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
16  *
17  * Copyright (C) IBM Corporation, 2010-2012
18  * Author:	Srikar Dronamraju <srikar@linux.vnet.ibm.com>
19  */
20 
21 #include <linux/module.h>
22 #include <linux/uaccess.h>
23 #include <linux/uprobes.h>
24 #include <linux/namei.h>
25 #include <linux/string.h>
26 
27 #include "trace_probe.h"
28 
29 #define UPROBE_EVENT_SYSTEM	"uprobes"
30 
31 struct trace_uprobe_filter {
32 	rwlock_t		rwlock;
33 	int			nr_systemwide;
34 	struct list_head	perf_events;
35 };
36 
37 /*
38  * uprobe event core functions
39  */
40 struct trace_uprobe {
41 	struct list_head		list;
42 	struct ftrace_event_class	class;
43 	struct ftrace_event_call	call;
44 	struct trace_uprobe_filter	filter;
45 	struct uprobe_consumer		consumer;
46 	struct inode			*inode;
47 	char				*filename;
48 	unsigned long			offset;
49 	unsigned long			nhit;
50 	unsigned int			flags;	/* For TP_FLAG_* */
51 	ssize_t				size;	/* trace entry size */
52 	unsigned int			nr_args;
53 	struct probe_arg		args[];
54 };
55 
56 #define SIZEOF_TRACE_UPROBE(n)			\
57 	(offsetof(struct trace_uprobe, args) +	\
58 	(sizeof(struct probe_arg) * (n)))
59 
60 static int register_uprobe_event(struct trace_uprobe *tu);
61 static void unregister_uprobe_event(struct trace_uprobe *tu);
62 
63 static DEFINE_MUTEX(uprobe_lock);
64 static LIST_HEAD(uprobe_list);
65 
66 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
67 
68 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
69 {
70 	rwlock_init(&filter->rwlock);
71 	filter->nr_systemwide = 0;
72 	INIT_LIST_HEAD(&filter->perf_events);
73 }
74 
75 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
76 {
77 	return !filter->nr_systemwide && list_empty(&filter->perf_events);
78 }
79 
80 /*
81  * Allocate new trace_uprobe and initialize it (including uprobes).
82  */
83 static struct trace_uprobe *
84 alloc_trace_uprobe(const char *group, const char *event, int nargs)
85 {
86 	struct trace_uprobe *tu;
87 
88 	if (!event || !is_good_name(event))
89 		return ERR_PTR(-EINVAL);
90 
91 	if (!group || !is_good_name(group))
92 		return ERR_PTR(-EINVAL);
93 
94 	tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
95 	if (!tu)
96 		return ERR_PTR(-ENOMEM);
97 
98 	tu->call.class = &tu->class;
99 	tu->call.name = kstrdup(event, GFP_KERNEL);
100 	if (!tu->call.name)
101 		goto error;
102 
103 	tu->class.system = kstrdup(group, GFP_KERNEL);
104 	if (!tu->class.system)
105 		goto error;
106 
107 	INIT_LIST_HEAD(&tu->list);
108 	tu->consumer.handler = uprobe_dispatcher;
109 	init_trace_uprobe_filter(&tu->filter);
110 	return tu;
111 
112 error:
113 	kfree(tu->call.name);
114 	kfree(tu);
115 
116 	return ERR_PTR(-ENOMEM);
117 }
118 
119 static void free_trace_uprobe(struct trace_uprobe *tu)
120 {
121 	int i;
122 
123 	for (i = 0; i < tu->nr_args; i++)
124 		traceprobe_free_probe_arg(&tu->args[i]);
125 
126 	iput(tu->inode);
127 	kfree(tu->call.class->system);
128 	kfree(tu->call.name);
129 	kfree(tu->filename);
130 	kfree(tu);
131 }
132 
133 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
134 {
135 	struct trace_uprobe *tu;
136 
137 	list_for_each_entry(tu, &uprobe_list, list)
138 		if (strcmp(tu->call.name, event) == 0 &&
139 		    strcmp(tu->call.class->system, group) == 0)
140 			return tu;
141 
142 	return NULL;
143 }
144 
145 /* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
146 static void unregister_trace_uprobe(struct trace_uprobe *tu)
147 {
148 	list_del(&tu->list);
149 	unregister_uprobe_event(tu);
150 	free_trace_uprobe(tu);
151 }
152 
153 /* Register a trace_uprobe and probe_event */
154 static int register_trace_uprobe(struct trace_uprobe *tu)
155 {
156 	struct trace_uprobe *old_tp;
157 	int ret;
158 
159 	mutex_lock(&uprobe_lock);
160 
161 	/* register as an event */
162 	old_tp = find_probe_event(tu->call.name, tu->call.class->system);
163 	if (old_tp)
164 		/* delete old event */
165 		unregister_trace_uprobe(old_tp);
166 
167 	ret = register_uprobe_event(tu);
168 	if (ret) {
169 		pr_warning("Failed to register probe event(%d)\n", ret);
170 		goto end;
171 	}
172 
173 	list_add_tail(&tu->list, &uprobe_list);
174 
175 end:
176 	mutex_unlock(&uprobe_lock);
177 
178 	return ret;
179 }
180 
181 /*
182  * Argument syntax:
183  *  - Add uprobe: p[:[GRP/]EVENT] PATH:SYMBOL[+offs] [FETCHARGS]
184  *
185  *  - Remove uprobe: -:[GRP/]EVENT
186  */
187 static int create_trace_uprobe(int argc, char **argv)
188 {
189 	struct trace_uprobe *tu;
190 	struct inode *inode;
191 	char *arg, *event, *group, *filename;
192 	char buf[MAX_EVENT_NAME_LEN];
193 	struct path path;
194 	unsigned long offset;
195 	bool is_delete;
196 	int i, ret;
197 
198 	inode = NULL;
199 	ret = 0;
200 	is_delete = false;
201 	event = NULL;
202 	group = NULL;
203 
204 	/* argc must be >= 1 */
205 	if (argv[0][0] == '-')
206 		is_delete = true;
207 	else if (argv[0][0] != 'p') {
208 		pr_info("Probe definition must be started with 'p' or '-'.\n");
209 		return -EINVAL;
210 	}
211 
212 	if (argv[0][1] == ':') {
213 		event = &argv[0][2];
214 		arg = strchr(event, '/');
215 
216 		if (arg) {
217 			group = event;
218 			event = arg + 1;
219 			event[-1] = '\0';
220 
221 			if (strlen(group) == 0) {
222 				pr_info("Group name is not specified\n");
223 				return -EINVAL;
224 			}
225 		}
226 		if (strlen(event) == 0) {
227 			pr_info("Event name is not specified\n");
228 			return -EINVAL;
229 		}
230 	}
231 	if (!group)
232 		group = UPROBE_EVENT_SYSTEM;
233 
234 	if (is_delete) {
235 		if (!event) {
236 			pr_info("Delete command needs an event name.\n");
237 			return -EINVAL;
238 		}
239 		mutex_lock(&uprobe_lock);
240 		tu = find_probe_event(event, group);
241 
242 		if (!tu) {
243 			mutex_unlock(&uprobe_lock);
244 			pr_info("Event %s/%s doesn't exist.\n", group, event);
245 			return -ENOENT;
246 		}
247 		/* delete an event */
248 		unregister_trace_uprobe(tu);
249 		mutex_unlock(&uprobe_lock);
250 		return 0;
251 	}
252 
253 	if (argc < 2) {
254 		pr_info("Probe point is not specified.\n");
255 		return -EINVAL;
256 	}
257 	if (isdigit(argv[1][0])) {
258 		pr_info("probe point must be have a filename.\n");
259 		return -EINVAL;
260 	}
261 	arg = strchr(argv[1], ':');
262 	if (!arg)
263 		goto fail_address_parse;
264 
265 	*arg++ = '\0';
266 	filename = argv[1];
267 	ret = kern_path(filename, LOOKUP_FOLLOW, &path);
268 	if (ret)
269 		goto fail_address_parse;
270 
271 	inode = igrab(path.dentry->d_inode);
272 	path_put(&path);
273 
274 	if (!inode || !S_ISREG(inode->i_mode)) {
275 		ret = -EINVAL;
276 		goto fail_address_parse;
277 	}
278 
279 	ret = kstrtoul(arg, 0, &offset);
280 	if (ret)
281 		goto fail_address_parse;
282 
283 	argc -= 2;
284 	argv += 2;
285 
286 	/* setup a probe */
287 	if (!event) {
288 		char *tail;
289 		char *ptr;
290 
291 		tail = kstrdup(kbasename(filename), GFP_KERNEL);
292 		if (!tail) {
293 			ret = -ENOMEM;
294 			goto fail_address_parse;
295 		}
296 
297 		ptr = strpbrk(tail, ".-_");
298 		if (ptr)
299 			*ptr = '\0';
300 
301 		snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
302 		event = buf;
303 		kfree(tail);
304 	}
305 
306 	tu = alloc_trace_uprobe(group, event, argc);
307 	if (IS_ERR(tu)) {
308 		pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
309 		ret = PTR_ERR(tu);
310 		goto fail_address_parse;
311 	}
312 	tu->offset = offset;
313 	tu->inode = inode;
314 	tu->filename = kstrdup(filename, GFP_KERNEL);
315 
316 	if (!tu->filename) {
317 		pr_info("Failed to allocate filename.\n");
318 		ret = -ENOMEM;
319 		goto error;
320 	}
321 
322 	/* parse arguments */
323 	ret = 0;
324 	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
325 		/* Increment count for freeing args in error case */
326 		tu->nr_args++;
327 
328 		/* Parse argument name */
329 		arg = strchr(argv[i], '=');
330 		if (arg) {
331 			*arg++ = '\0';
332 			tu->args[i].name = kstrdup(argv[i], GFP_KERNEL);
333 		} else {
334 			arg = argv[i];
335 			/* If argument name is omitted, set "argN" */
336 			snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
337 			tu->args[i].name = kstrdup(buf, GFP_KERNEL);
338 		}
339 
340 		if (!tu->args[i].name) {
341 			pr_info("Failed to allocate argument[%d] name.\n", i);
342 			ret = -ENOMEM;
343 			goto error;
344 		}
345 
346 		if (!is_good_name(tu->args[i].name)) {
347 			pr_info("Invalid argument[%d] name: %s\n", i, tu->args[i].name);
348 			ret = -EINVAL;
349 			goto error;
350 		}
351 
352 		if (traceprobe_conflict_field_name(tu->args[i].name, tu->args, i)) {
353 			pr_info("Argument[%d] name '%s' conflicts with "
354 				"another field.\n", i, argv[i]);
355 			ret = -EINVAL;
356 			goto error;
357 		}
358 
359 		/* Parse fetch argument */
360 		ret = traceprobe_parse_probe_arg(arg, &tu->size, &tu->args[i], false, false);
361 		if (ret) {
362 			pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
363 			goto error;
364 		}
365 	}
366 
367 	ret = register_trace_uprobe(tu);
368 	if (ret)
369 		goto error;
370 	return 0;
371 
372 error:
373 	free_trace_uprobe(tu);
374 	return ret;
375 
376 fail_address_parse:
377 	if (inode)
378 		iput(inode);
379 
380 	pr_info("Failed to parse address or file.\n");
381 
382 	return ret;
383 }
384 
385 static void cleanup_all_probes(void)
386 {
387 	struct trace_uprobe *tu;
388 
389 	mutex_lock(&uprobe_lock);
390 	while (!list_empty(&uprobe_list)) {
391 		tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
392 		unregister_trace_uprobe(tu);
393 	}
394 	mutex_unlock(&uprobe_lock);
395 }
396 
397 /* Probes listing interfaces */
398 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
399 {
400 	mutex_lock(&uprobe_lock);
401 	return seq_list_start(&uprobe_list, *pos);
402 }
403 
404 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
405 {
406 	return seq_list_next(v, &uprobe_list, pos);
407 }
408 
409 static void probes_seq_stop(struct seq_file *m, void *v)
410 {
411 	mutex_unlock(&uprobe_lock);
412 }
413 
414 static int probes_seq_show(struct seq_file *m, void *v)
415 {
416 	struct trace_uprobe *tu = v;
417 	int i;
418 
419 	seq_printf(m, "p:%s/%s", tu->call.class->system, tu->call.name);
420 	seq_printf(m, " %s:0x%p", tu->filename, (void *)tu->offset);
421 
422 	for (i = 0; i < tu->nr_args; i++)
423 		seq_printf(m, " %s=%s", tu->args[i].name, tu->args[i].comm);
424 
425 	seq_printf(m, "\n");
426 	return 0;
427 }
428 
429 static const struct seq_operations probes_seq_op = {
430 	.start	= probes_seq_start,
431 	.next	= probes_seq_next,
432 	.stop	= probes_seq_stop,
433 	.show	= probes_seq_show
434 };
435 
436 static int probes_open(struct inode *inode, struct file *file)
437 {
438 	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
439 		cleanup_all_probes();
440 
441 	return seq_open(file, &probes_seq_op);
442 }
443 
444 static ssize_t probes_write(struct file *file, const char __user *buffer,
445 			    size_t count, loff_t *ppos)
446 {
447 	return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
448 }
449 
450 static const struct file_operations uprobe_events_ops = {
451 	.owner		= THIS_MODULE,
452 	.open		= probes_open,
453 	.read		= seq_read,
454 	.llseek		= seq_lseek,
455 	.release	= seq_release,
456 	.write		= probes_write,
457 };
458 
459 /* Probes profiling interfaces */
460 static int probes_profile_seq_show(struct seq_file *m, void *v)
461 {
462 	struct trace_uprobe *tu = v;
463 
464 	seq_printf(m, "  %s %-44s %15lu\n", tu->filename, tu->call.name, tu->nhit);
465 	return 0;
466 }
467 
468 static const struct seq_operations profile_seq_op = {
469 	.start	= probes_seq_start,
470 	.next	= probes_seq_next,
471 	.stop	= probes_seq_stop,
472 	.show	= probes_profile_seq_show
473 };
474 
475 static int profile_open(struct inode *inode, struct file *file)
476 {
477 	return seq_open(file, &profile_seq_op);
478 }
479 
480 static const struct file_operations uprobe_profile_ops = {
481 	.owner		= THIS_MODULE,
482 	.open		= profile_open,
483 	.read		= seq_read,
484 	.llseek		= seq_lseek,
485 	.release	= seq_release,
486 };
487 
488 /* uprobe handler */
489 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs)
490 {
491 	struct uprobe_trace_entry_head *entry;
492 	struct ring_buffer_event *event;
493 	struct ring_buffer *buffer;
494 	u8 *data;
495 	int size, i, pc;
496 	unsigned long irq_flags;
497 	struct ftrace_event_call *call = &tu->call;
498 
499 	local_save_flags(irq_flags);
500 	pc = preempt_count();
501 
502 	size = sizeof(*entry) + tu->size;
503 
504 	event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
505 						  size, irq_flags, pc);
506 	if (!event)
507 		return 0;
508 
509 	entry = ring_buffer_event_data(event);
510 	entry->ip = instruction_pointer(task_pt_regs(current));
511 	data = (u8 *)&entry[1];
512 	for (i = 0; i < tu->nr_args; i++)
513 		call_fetch(&tu->args[i].fetch, regs, data + tu->args[i].offset);
514 
515 	if (!filter_current_check_discard(buffer, call, entry, event))
516 		trace_buffer_unlock_commit(buffer, event, irq_flags, pc);
517 
518 	return 0;
519 }
520 
521 /* Event entry printers */
522 static enum print_line_t
523 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
524 {
525 	struct uprobe_trace_entry_head *field;
526 	struct trace_seq *s = &iter->seq;
527 	struct trace_uprobe *tu;
528 	u8 *data;
529 	int i;
530 
531 	field = (struct uprobe_trace_entry_head *)iter->ent;
532 	tu = container_of(event, struct trace_uprobe, call.event);
533 
534 	if (!trace_seq_printf(s, "%s: (", tu->call.name))
535 		goto partial;
536 
537 	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
538 		goto partial;
539 
540 	if (!trace_seq_puts(s, ")"))
541 		goto partial;
542 
543 	data = (u8 *)&field[1];
544 	for (i = 0; i < tu->nr_args; i++) {
545 		if (!tu->args[i].type->print(s, tu->args[i].name,
546 					     data + tu->args[i].offset, field))
547 			goto partial;
548 	}
549 
550 	if (trace_seq_puts(s, "\n"))
551 		return TRACE_TYPE_HANDLED;
552 
553 partial:
554 	return TRACE_TYPE_PARTIAL_LINE;
555 }
556 
557 static inline bool is_trace_uprobe_enabled(struct trace_uprobe *tu)
558 {
559 	return tu->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE);
560 }
561 
562 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
563 				enum uprobe_filter_ctx ctx,
564 				struct mm_struct *mm);
565 
566 static int
567 probe_event_enable(struct trace_uprobe *tu, int flag, filter_func_t filter)
568 {
569 	int ret = 0;
570 
571 	if (is_trace_uprobe_enabled(tu))
572 		return -EINTR;
573 
574 	WARN_ON(!uprobe_filter_is_empty(&tu->filter));
575 
576 	tu->flags |= flag;
577 	tu->consumer.filter = filter;
578 	ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
579 	if (ret)
580 		tu->flags &= ~flag;
581 
582 	return ret;
583 }
584 
585 static void probe_event_disable(struct trace_uprobe *tu, int flag)
586 {
587 	if (!is_trace_uprobe_enabled(tu))
588 		return;
589 
590 	WARN_ON(!uprobe_filter_is_empty(&tu->filter));
591 
592 	uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
593 	tu->flags &= ~flag;
594 }
595 
596 static int uprobe_event_define_fields(struct ftrace_event_call *event_call)
597 {
598 	int ret, i;
599 	struct uprobe_trace_entry_head field;
600 	struct trace_uprobe *tu = (struct trace_uprobe *)event_call->data;
601 
602 	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
603 	/* Set argument names as fields */
604 	for (i = 0; i < tu->nr_args; i++) {
605 		ret = trace_define_field(event_call, tu->args[i].type->fmttype,
606 					 tu->args[i].name,
607 					 sizeof(field) + tu->args[i].offset,
608 					 tu->args[i].type->size,
609 					 tu->args[i].type->is_signed,
610 					 FILTER_OTHER);
611 
612 		if (ret)
613 			return ret;
614 	}
615 	return 0;
616 }
617 
618 #define LEN_OR_ZERO		(len ? len - pos : 0)
619 static int __set_print_fmt(struct trace_uprobe *tu, char *buf, int len)
620 {
621 	const char *fmt, *arg;
622 	int i;
623 	int pos = 0;
624 
625 	fmt = "(%lx)";
626 	arg = "REC->" FIELD_STRING_IP;
627 
628 	/* When len=0, we just calculate the needed length */
629 
630 	pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
631 
632 	for (i = 0; i < tu->nr_args; i++) {
633 		pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
634 				tu->args[i].name, tu->args[i].type->fmt);
635 	}
636 
637 	pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
638 
639 	for (i = 0; i < tu->nr_args; i++) {
640 		pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
641 				tu->args[i].name);
642 	}
643 
644 	return pos;	/* return the length of print_fmt */
645 }
646 #undef LEN_OR_ZERO
647 
648 static int set_print_fmt(struct trace_uprobe *tu)
649 {
650 	char *print_fmt;
651 	int len;
652 
653 	/* First: called with 0 length to calculate the needed length */
654 	len = __set_print_fmt(tu, NULL, 0);
655 	print_fmt = kmalloc(len + 1, GFP_KERNEL);
656 	if (!print_fmt)
657 		return -ENOMEM;
658 
659 	/* Second: actually write the @print_fmt */
660 	__set_print_fmt(tu, print_fmt, len + 1);
661 	tu->call.print_fmt = print_fmt;
662 
663 	return 0;
664 }
665 
666 #ifdef CONFIG_PERF_EVENTS
667 static bool
668 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
669 {
670 	struct perf_event *event;
671 
672 	if (filter->nr_systemwide)
673 		return true;
674 
675 	list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
676 		if (event->hw.tp_target->mm == mm)
677 			return true;
678 	}
679 
680 	return false;
681 }
682 
683 static inline bool
684 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
685 {
686 	return __uprobe_perf_filter(&tu->filter, event->hw.tp_target->mm);
687 }
688 
689 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
690 {
691 	bool done;
692 
693 	write_lock(&tu->filter.rwlock);
694 	if (event->hw.tp_target) {
695 		/*
696 		 * event->parent != NULL means copy_process(), we can avoid
697 		 * uprobe_apply(). current->mm must be probed and we can rely
698 		 * on dup_mmap() which preserves the already installed bp's.
699 		 *
700 		 * attr.enable_on_exec means that exec/mmap will install the
701 		 * breakpoints we need.
702 		 */
703 		done = tu->filter.nr_systemwide ||
704 			event->parent || event->attr.enable_on_exec ||
705 			uprobe_filter_event(tu, event);
706 		list_add(&event->hw.tp_list, &tu->filter.perf_events);
707 	} else {
708 		done = tu->filter.nr_systemwide;
709 		tu->filter.nr_systemwide++;
710 	}
711 	write_unlock(&tu->filter.rwlock);
712 
713 	if (!done)
714 		uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
715 
716 	return 0;
717 }
718 
719 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
720 {
721 	bool done;
722 
723 	write_lock(&tu->filter.rwlock);
724 	if (event->hw.tp_target) {
725 		list_del(&event->hw.tp_list);
726 		done = tu->filter.nr_systemwide ||
727 			(event->hw.tp_target->flags & PF_EXITING) ||
728 			uprobe_filter_event(tu, event);
729 	} else {
730 		tu->filter.nr_systemwide--;
731 		done = tu->filter.nr_systemwide;
732 	}
733 	write_unlock(&tu->filter.rwlock);
734 
735 	if (!done)
736 		uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
737 
738 	return 0;
739 }
740 
741 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
742 				enum uprobe_filter_ctx ctx, struct mm_struct *mm)
743 {
744 	struct trace_uprobe *tu;
745 	int ret;
746 
747 	tu = container_of(uc, struct trace_uprobe, consumer);
748 	read_lock(&tu->filter.rwlock);
749 	ret = __uprobe_perf_filter(&tu->filter, mm);
750 	read_unlock(&tu->filter.rwlock);
751 
752 	return ret;
753 }
754 
755 /* uprobe profile handler */
756 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs)
757 {
758 	struct ftrace_event_call *call = &tu->call;
759 	struct uprobe_trace_entry_head *entry;
760 	struct hlist_head *head;
761 	u8 *data;
762 	int size, __size, i;
763 	int rctx;
764 
765 	if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
766 		return UPROBE_HANDLER_REMOVE;
767 
768 	__size = sizeof(*entry) + tu->size;
769 	size = ALIGN(__size + sizeof(u32), sizeof(u64));
770 	size -= sizeof(u32);
771 	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
772 		return 0;
773 
774 	preempt_disable();
775 
776 	entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
777 	if (!entry)
778 		goto out;
779 
780 	entry->ip = instruction_pointer(task_pt_regs(current));
781 	data = (u8 *)&entry[1];
782 	for (i = 0; i < tu->nr_args; i++)
783 		call_fetch(&tu->args[i].fetch, regs, data + tu->args[i].offset);
784 
785 	head = this_cpu_ptr(call->perf_events);
786 	perf_trace_buf_submit(entry, size, rctx, entry->ip, 1, regs, head, NULL);
787 
788  out:
789 	preempt_enable();
790 	return 0;
791 }
792 #endif	/* CONFIG_PERF_EVENTS */
793 
794 static
795 int trace_uprobe_register(struct ftrace_event_call *event, enum trace_reg type, void *data)
796 {
797 	struct trace_uprobe *tu = (struct trace_uprobe *)event->data;
798 
799 	switch (type) {
800 	case TRACE_REG_REGISTER:
801 		return probe_event_enable(tu, TP_FLAG_TRACE, NULL);
802 
803 	case TRACE_REG_UNREGISTER:
804 		probe_event_disable(tu, TP_FLAG_TRACE);
805 		return 0;
806 
807 #ifdef CONFIG_PERF_EVENTS
808 	case TRACE_REG_PERF_REGISTER:
809 		return probe_event_enable(tu, TP_FLAG_PROFILE, uprobe_perf_filter);
810 
811 	case TRACE_REG_PERF_UNREGISTER:
812 		probe_event_disable(tu, TP_FLAG_PROFILE);
813 		return 0;
814 
815 	case TRACE_REG_PERF_OPEN:
816 		return uprobe_perf_open(tu, data);
817 
818 	case TRACE_REG_PERF_CLOSE:
819 		return uprobe_perf_close(tu, data);
820 
821 #endif
822 	default:
823 		return 0;
824 	}
825 	return 0;
826 }
827 
828 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
829 {
830 	struct trace_uprobe *tu;
831 	int ret = 0;
832 
833 	tu = container_of(con, struct trace_uprobe, consumer);
834 	tu->nhit++;
835 
836 	if (tu->flags & TP_FLAG_TRACE)
837 		ret |= uprobe_trace_func(tu, regs);
838 
839 #ifdef CONFIG_PERF_EVENTS
840 	if (tu->flags & TP_FLAG_PROFILE)
841 		ret |= uprobe_perf_func(tu, regs);
842 #endif
843 	return ret;
844 }
845 
846 static struct trace_event_functions uprobe_funcs = {
847 	.trace		= print_uprobe_event
848 };
849 
850 static int register_uprobe_event(struct trace_uprobe *tu)
851 {
852 	struct ftrace_event_call *call = &tu->call;
853 	int ret;
854 
855 	/* Initialize ftrace_event_call */
856 	INIT_LIST_HEAD(&call->class->fields);
857 	call->event.funcs = &uprobe_funcs;
858 	call->class->define_fields = uprobe_event_define_fields;
859 
860 	if (set_print_fmt(tu) < 0)
861 		return -ENOMEM;
862 
863 	ret = register_ftrace_event(&call->event);
864 	if (!ret) {
865 		kfree(call->print_fmt);
866 		return -ENODEV;
867 	}
868 	call->flags = 0;
869 	call->class->reg = trace_uprobe_register;
870 	call->data = tu;
871 	ret = trace_add_event_call(call);
872 
873 	if (ret) {
874 		pr_info("Failed to register uprobe event: %s\n", call->name);
875 		kfree(call->print_fmt);
876 		unregister_ftrace_event(&call->event);
877 	}
878 
879 	return ret;
880 }
881 
882 static void unregister_uprobe_event(struct trace_uprobe *tu)
883 {
884 	/* tu->event is unregistered in trace_remove_event_call() */
885 	trace_remove_event_call(&tu->call);
886 	kfree(tu->call.print_fmt);
887 	tu->call.print_fmt = NULL;
888 }
889 
890 /* Make a trace interface for controling probe points */
891 static __init int init_uprobe_trace(void)
892 {
893 	struct dentry *d_tracer;
894 
895 	d_tracer = tracing_init_dentry();
896 	if (!d_tracer)
897 		return 0;
898 
899 	trace_create_file("uprobe_events", 0644, d_tracer,
900 				    NULL, &uprobe_events_ops);
901 	/* Profile interface */
902 	trace_create_file("uprobe_profile", 0444, d_tracer,
903 				    NULL, &uprobe_profile_ops);
904 	return 0;
905 }
906 
907 fs_initcall(init_uprobe_trace);
908